d/freebsd-skq
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
755bab6d55
freebsd-skq/sys/rpc/svc.c

1483 lines
36 KiB
C
Raw Normal View History

Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
/* $NetBSD: svc.c,v 1.21 2000/07/06 03:10:35 christos Exp $ */
Replace Sun RPC license in TI-RPC library with a 3-clause BSD license, with the explicit permission of Sun Microsystems in 2009.
2013-11-25 19:04:36 +00:00
/*-
sys: further adoption of SPDX licensing ID tags. Mainly focus on files that use BSD 3-Clause license. The Software Package Data Exchange (SPDX) group provides a specification to make it easier for automated tools to detect and summarize well known opensource licenses. We are gradually adopting the specification, noting that the tags are considered only advisory and do not, in any way, superceed or replace the license texts. Special thanks to Wind River for providing access to "The Duke of Highlander" tool: an older (2014) run over FreeBSD tree was useful as a starting point.
2017-11-20 19:43:44 +00:00
* SPDX-License-Identifier: BSD-3-Clause
*
Replace Sun RPC license in TI-RPC library with a 3-clause BSD license, with the explicit permission of Sun Microsystems in 2009.
2013-11-25 19:04:36 +00:00
* Copyright (c) 2009, Sun Microsystems, Inc.
* All rights reserved.
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
*
Replace Sun RPC license in TI-RPC library with a 3-clause BSD license, with the explicit permission of Sun Microsystems in 2009.
2013-11-25 19:04:36 +00:00
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions are met:
* - Redistributions of source code must retain the above copyright notice,
* this list of conditions and the following disclaimer.
* - Redistributions in binary form must reproduce the above copyright notice,
* this list of conditions and the following disclaimer in the documentation
* and/or other materials provided with the distribution.
* - Neither the name of Sun Microsystems, Inc. nor the names of its
* contributors may be used to endorse or promote products derived
* from this software without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
* AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE
* LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
* CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
* SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
* INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
* CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
* ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
* POSSIBILITY OF SUCH DAMAGE.
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
*/
#if defined(LIBC_SCCS) && !defined(lint)
static char *sccsid2 = "@(#)svc.c 1.44 88/02/08 Copyr 1984 Sun Micro";
static char *sccsid = "@(#)svc.c 2.4 88/08/11 4.0 RPCSRC";
#endif
#include <sys/cdefs.h>
__FBSDID("$FreeBSD$");
/*
* svc.c, Server-side remote procedure call interface.
*
* There are two sets of procedures here. The xprt routines are
* for handling transport handles. The svc routines handle the
* list of service routines.
*
* Copyright (C) 1984, Sun Microsystems, Inc.
*/
#include <sys/param.h>
#include <sys/lock.h>
#include <sys/kernel.h>
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
#include <sys/kthread.h>
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
#include <sys/malloc.h>
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
#include <sys/mbuf.h>
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
#include <sys/mutex.h>
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
#include <sys/proc.h>
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
#include <sys/queue.h>
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
#include <sys/socketvar.h>
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
#include <sys/systm.h>
Split RPC pool threads into number of smaller semi-isolated groups. Old design with unified thread pool was good from the point of thread utilization. But single pool-wide mutex became huge congestion point for systems with many CPUs. To reduce the congestion create several thread groups within a pool (one group for every 6 CPUs and 12 threads), each group with own mutex. Each connection during its registration is assigned to one of the groups in round-robin fashion. File affinify code may still move requests between the groups, but otherwise groups are self-contained. MFC after: 2 weeks Sponsored by: iXsystems, Inc.
2014-06-08 11:19:32 +00:00
#include <sys/smp.h>
Rework NFS Duplicate Request Cache cleanup logic. - Introduce additional hash to group requests by hash of sockref. This allows to process TCP acknowledgements without looping though all the cache, and as result allows to do it every time. - Indroduce additional callbacks to notify application layer about sockets disconnection. Without this last few requests processed just before socket disconnection never processed their ACKs and stuck in cache for many hours. - Implement transport-specific method for tracking reply acknowledgements. New implementation does not cross multiple stack layers to get the data and does not have race conditions that previously made some requests stuck in cache. This could be done more efficiently at sockbuf layer, but that would broke some KBIs, while I don't know other consumers for it aside NFS. - Instead of traversing all DRC twice per request, run cleaning only once per request, and except in some conditions traverse only single hash slot at a time. Together this limits NFS DRC growth only to situations of real connectivity problems. If network is working well, and so all replies are acknowledged, cache remains almost empty even after hours of heavy load. Without this change on the same test cache was growing to many thousand requests even with perfectly working local network. As another result this reduces CPU time spent on the DRC handling during SPEC NFS benchmark from about 10% to 0.5%. Sponsored by: iXsystems, Inc.
2014-01-03 15:09:59 +00:00
#include <sys/sx.h>
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
#include <sys/ucred.h>
#include <rpc/rpc.h>
#include <rpc/rpcb_clnt.h>
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
#include <rpc/replay.h>
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
Minor changes to improve compatibility with older FreeBSD releases.
2008-03-28 09:50:32 +00:00
#include <rpc/rpc_com.h>
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
#define SVC_VERSQUIET 0x0001 /* keep quiet about vers mismatch */
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
#define version_keepquiet(xp) (SVC_EXT(xp)->xp_flags & SVC_VERSQUIET)
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
static struct svc_callout *svc_find(SVCPOOL *pool, rpcprog_t, rpcvers_t,
char *);
Split RPC pool threads into number of smaller semi-isolated groups. Old design with unified thread pool was good from the point of thread utilization. But single pool-wide mutex became huge congestion point for systems with many CPUs. To reduce the congestion create several thread groups within a pool (one group for every 6 CPUs and 12 threads), each group with own mutex. Each connection during its registration is assigned to one of the groups in round-robin fashion. File affinify code may still move requests between the groups, but otherwise groups are self-contained. MFC after: 2 weeks Sponsored by: iXsystems, Inc.
2014-06-08 11:19:32 +00:00
static void svc_new_thread(SVCGROUP *grp);
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
static void xprt_unregister_locked(SVCXPRT *xprt);
Fix overflow bugs in and remove obsolete limit from kernel RPC implementation. The kernel RPC code, which is responsible for the low-level scheduling of incoming NFS requests, contains a throttling mechanism that prevents too much kernel memory from being tied up by NFS requests that are being serviced. When the throttle is engaged, the RPC layer stops servicing incoming NFS sockets, resulting ultimately in backpressure on the clients (if they're using TCP). However, this is a very heavy-handed mechanism as it prevents all clients from making any requests, regardless of how heavy or light they are. (Thus, when engaged, the throttle often prevents clients from even mounting the filesystem.) The throttle mechanism applies specifically to requests that have been received by the RPC layer (from a TCP or UDP socket) and are queued waiting to be serviced by one of the nfsd threads; it does not limit the amount of backlog in the socket buffers. The original implementation limited the total bytes of queued requests to the minimum of a quarter of (nmbclusters * MCLBYTES) and 45 MiB. The former limit seems reasonable, since requests queued in the socket buffers and replies being constructed to the requests in progress will all require some amount of network memory, but the 45 MiB limit is plainly ridiculous for modern memory sizes: when running 256 service threads on a busy server, 45 MiB would result in just a single maximum-sized NFS3PROC_WRITE queued per thread before throttling. Removing this limit exposed integer-overflow bugs in the original computation, and related bugs in the routines that actually account for the amount of traffic enqueued for service threads. The old implementation also attempted to reduce accounting overhead by batching updates until each queue is fully drained, but this is prone to livelock, resulting in repeated accumulate-throttle-drain cycles on a busy server. Various data types are changed to long or unsigned long; explicit 64-bit types are not used due to the unavailability of 64-bit atomics on many 32-bit platforms, but those platforms also cannot support nmbclusters large enough to cause overflow. This code (in a 10.1 kernel) is presently running on production NFS servers at CSAIL. Summary of this revision: * Removes 45 MiB limit on requests queued for nfsd service threads * Fixes integer-overflow and signedness bugs * Avoids unnecessary throttling by not deferring accounting for completed requests Differential Revision: https://reviews.freebsd.org/D2165 Reviewed by: rmacklem, mav MFC after: 30 days Relnotes: yes Sponsored by: MIT Computer Science & Artificial Intelligence Laboratory
2015-04-01 00:45:47 +00:00
static void svc_change_space_used(SVCPOOL *pool, long delta);
Move most of NFS file handle affinity code out of the heavily congested global RPC thread pool lock and protect it with own set of locks. On synthetic benchmarks this improves peak NFS request rate by 40%.
2013-12-30 20:23:15 +00:00
static bool_t svc_request_space_available(SVCPOOL *pool);
add svcpool_close to handle killed nfsd threads This patch adds a new function to the server krpc called svcpool_close(). It is similar to svcpool_destroy(), but does not free the data structures, so that the pool can be used again. This function is then used instead of svcpool_destroy(), svcpool_create() when the nfsd threads are killed. PR: 204340 Reported by: Panzura Approved by: rmacklem Obtained from: rmacklem MFC after: 1 week
2017-02-14 17:49:08 +00:00
static void svcpool_cleanup(SVCPOOL *pool);
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
/* *************** SVCXPRT related stuff **************** */
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
static int svcpool_minthread_sysctl(SYSCTL_HANDLER_ARGS);
static int svcpool_maxthread_sysctl(SYSCTL_HANDLER_ARGS);
Split RPC pool threads into number of smaller semi-isolated groups. Old design with unified thread pool was good from the point of thread utilization. But single pool-wide mutex became huge congestion point for systems with many CPUs. To reduce the congestion create several thread groups within a pool (one group for every 6 CPUs and 12 threads), each group with own mutex. Each connection during its registration is assigned to one of the groups in round-robin fashion. File affinify code may still move requests between the groups, but otherwise groups are self-contained. MFC after: 2 weeks Sponsored by: iXsystems, Inc.
2014-06-08 11:19:32 +00:00
static int svcpool_threads_sysctl(SYSCTL_HANDLER_ARGS);
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
SVCPOOL*
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
svcpool_create(const char *name, struct sysctl_oid_list *sysctl_base)
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
{
SVCPOOL *pool;
Split RPC pool threads into number of smaller semi-isolated groups. Old design with unified thread pool was good from the point of thread utilization. But single pool-wide mutex became huge congestion point for systems with many CPUs. To reduce the congestion create several thread groups within a pool (one group for every 6 CPUs and 12 threads), each group with own mutex. Each connection during its registration is assigned to one of the groups in round-robin fashion. File affinify code may still move requests between the groups, but otherwise groups are self-contained. MFC after: 2 weeks Sponsored by: iXsystems, Inc.
2014-06-08 11:19:32 +00:00
SVCGROUP *grp;
int g;
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
pool = malloc(sizeof(SVCPOOL), M_RPC, M_WAITOK|M_ZERO);
mtx_init(&pool->sp_lock, "sp_lock", NULL, MTX_DEF);
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
pool->sp_name = name;
pool->sp_state = SVCPOOL_INIT;
pool->sp_proc = NULL;
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
TAILQ_INIT(&pool->sp_callouts);
Rework NFS Duplicate Request Cache cleanup logic. - Introduce additional hash to group requests by hash of sockref. This allows to process TCP acknowledgements without looping though all the cache, and as result allows to do it every time. - Indroduce additional callbacks to notify application layer about sockets disconnection. Without this last few requests processed just before socket disconnection never processed their ACKs and stuck in cache for many hours. - Implement transport-specific method for tracking reply acknowledgements. New implementation does not cross multiple stack layers to get the data and does not have race conditions that previously made some requests stuck in cache. This could be done more efficiently at sockbuf layer, but that would broke some KBIs, while I don't know other consumers for it aside NFS. - Instead of traversing all DRC twice per request, run cleaning only once per request, and except in some conditions traverse only single hash slot at a time. Together this limits NFS DRC growth only to situations of real connectivity problems. If network is working well, and so all replies are acknowledged, cache remains almost empty even after hours of heavy load. Without this change on the same test cache was growing to many thousand requests even with perfectly working local network. As another result this reduces CPU time spent on the DRC handling during SPEC NFS benchmark from about 10% to 0.5%. Sponsored by: iXsystems, Inc.
2014-01-03 15:09:59 +00:00
TAILQ_INIT(&pool->sp_lcallouts);
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
pool->sp_minthreads = 1;
pool->sp_maxthreads = 1;
Split RPC pool threads into number of smaller semi-isolated groups. Old design with unified thread pool was good from the point of thread utilization. But single pool-wide mutex became huge congestion point for systems with many CPUs. To reduce the congestion create several thread groups within a pool (one group for every 6 CPUs and 12 threads), each group with own mutex. Each connection during its registration is assigned to one of the groups in round-robin fashion. File affinify code may still move requests between the groups, but otherwise groups are self-contained. MFC after: 2 weeks Sponsored by: iXsystems, Inc.
2014-06-08 11:19:32 +00:00
pool->sp_groupcount = 1;
for (g = 0; g < SVC_MAXGROUPS; g++) {
grp = &pool->sp_groups[g];
mtx_init(&grp->sg_lock, "sg_lock", NULL, MTX_DEF);
grp->sg_pool = pool;
grp->sg_state = SVCPOOL_ACTIVE;
TAILQ_INIT(&grp->sg_xlist);
TAILQ_INIT(&grp->sg_active);
LIST_INIT(&grp->sg_idlethreads);
grp->sg_minthreads = 1;
grp->sg_maxthreads = 1;
}
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
/*
Fix overflow bugs in and remove obsolete limit from kernel RPC implementation. The kernel RPC code, which is responsible for the low-level scheduling of incoming NFS requests, contains a throttling mechanism that prevents too much kernel memory from being tied up by NFS requests that are being serviced. When the throttle is engaged, the RPC layer stops servicing incoming NFS sockets, resulting ultimately in backpressure on the clients (if they're using TCP). However, this is a very heavy-handed mechanism as it prevents all clients from making any requests, regardless of how heavy or light they are. (Thus, when engaged, the throttle often prevents clients from even mounting the filesystem.) The throttle mechanism applies specifically to requests that have been received by the RPC layer (from a TCP or UDP socket) and are queued waiting to be serviced by one of the nfsd threads; it does not limit the amount of backlog in the socket buffers. The original implementation limited the total bytes of queued requests to the minimum of a quarter of (nmbclusters * MCLBYTES) and 45 MiB. The former limit seems reasonable, since requests queued in the socket buffers and replies being constructed to the requests in progress will all require some amount of network memory, but the 45 MiB limit is plainly ridiculous for modern memory sizes: when running 256 service threads on a busy server, 45 MiB would result in just a single maximum-sized NFS3PROC_WRITE queued per thread before throttling. Removing this limit exposed integer-overflow bugs in the original computation, and related bugs in the routines that actually account for the amount of traffic enqueued for service threads. The old implementation also attempted to reduce accounting overhead by batching updates until each queue is fully drained, but this is prone to livelock, resulting in repeated accumulate-throttle-drain cycles on a busy server. Various data types are changed to long or unsigned long; explicit 64-bit types are not used due to the unavailability of 64-bit atomics on many 32-bit platforms, but those platforms also cannot support nmbclusters large enough to cause overflow. This code (in a 10.1 kernel) is presently running on production NFS servers at CSAIL. Summary of this revision: * Removes 45 MiB limit on requests queued for nfsd service threads * Fixes integer-overflow and signedness bugs * Avoids unnecessary throttling by not deferring accounting for completed requests Differential Revision: https://reviews.freebsd.org/D2165 Reviewed by: rmacklem, mav MFC after: 30 days Relnotes: yes Sponsored by: MIT Computer Science & Artificial Intelligence Laboratory
2015-04-01 00:45:47 +00:00
* Don't use more than a quarter of mbuf clusters. Nota bene:
* nmbclusters is an int, but nmbclusters*MCLBYTES may overflow
* on LP64 architectures, so cast to u_long to avoid undefined
* behavior. (ILP32 architectures cannot have nmbclusters
* large enough to overflow for other reasons.)
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
*/
Fix overflow bugs in and remove obsolete limit from kernel RPC implementation. The kernel RPC code, which is responsible for the low-level scheduling of incoming NFS requests, contains a throttling mechanism that prevents too much kernel memory from being tied up by NFS requests that are being serviced. When the throttle is engaged, the RPC layer stops servicing incoming NFS sockets, resulting ultimately in backpressure on the clients (if they're using TCP). However, this is a very heavy-handed mechanism as it prevents all clients from making any requests, regardless of how heavy or light they are. (Thus, when engaged, the throttle often prevents clients from even mounting the filesystem.) The throttle mechanism applies specifically to requests that have been received by the RPC layer (from a TCP or UDP socket) and are queued waiting to be serviced by one of the nfsd threads; it does not limit the amount of backlog in the socket buffers. The original implementation limited the total bytes of queued requests to the minimum of a quarter of (nmbclusters * MCLBYTES) and 45 MiB. The former limit seems reasonable, since requests queued in the socket buffers and replies being constructed to the requests in progress will all require some amount of network memory, but the 45 MiB limit is plainly ridiculous for modern memory sizes: when running 256 service threads on a busy server, 45 MiB would result in just a single maximum-sized NFS3PROC_WRITE queued per thread before throttling. Removing this limit exposed integer-overflow bugs in the original computation, and related bugs in the routines that actually account for the amount of traffic enqueued for service threads. The old implementation also attempted to reduce accounting overhead by batching updates until each queue is fully drained, but this is prone to livelock, resulting in repeated accumulate-throttle-drain cycles on a busy server. Various data types are changed to long or unsigned long; explicit 64-bit types are not used due to the unavailability of 64-bit atomics on many 32-bit platforms, but those platforms also cannot support nmbclusters large enough to cause overflow. This code (in a 10.1 kernel) is presently running on production NFS servers at CSAIL. Summary of this revision: * Removes 45 MiB limit on requests queued for nfsd service threads * Fixes integer-overflow and signedness bugs * Avoids unnecessary throttling by not deferring accounting for completed requests Differential Revision: https://reviews.freebsd.org/D2165 Reviewed by: rmacklem, mav MFC after: 30 days Relnotes: yes Sponsored by: MIT Computer Science & Artificial Intelligence Laboratory
2015-04-01 00:45:47 +00:00
pool->sp_space_high = (u_long)nmbclusters * MCLBYTES / 4;
pool->sp_space_low = (pool->sp_space_high / 3) * 2;
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
sysctl_ctx_init(&pool->sp_sysctl);
if (sysctl_base) {
SYSCTL_ADD_PROC(&pool->sp_sysctl, sysctl_base, OID_AUTO,
Mark more nodes as CTLFLAG_MPSAFE or CTLFLAG_NEEDGIANT (17 of many) r357614 added CTLFLAG_NEEDGIANT to make it easier to find nodes that are still not MPSAFE (or already are but aren’t properly marked). Use it in preparation for a general review of all nodes. This is non-functional change that adds annotations to SYSCTL_NODE and SYSCTL_PROC nodes using one of the soon-to-be-required flags. Mark all obvious cases as MPSAFE. All entries that haven't been marked as MPSAFE before are by default marked as NEEDGIANT Approved by: kib (mentor, blanket) Commented by: kib, gallatin, melifaro Differential Revision: https://reviews.freebsd.org/D23718
2020-02-26 14:26:36 +00:00
"minthreads", CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
Split RPC pool threads into number of smaller semi-isolated groups. Old design with unified thread pool was good from the point of thread utilization. But single pool-wide mutex became huge congestion point for systems with many CPUs. To reduce the congestion create several thread groups within a pool (one group for every 6 CPUs and 12 threads), each group with own mutex. Each connection during its registration is assigned to one of the groups in round-robin fashion. File affinify code may still move requests between the groups, but otherwise groups are self-contained. MFC after: 2 weeks Sponsored by: iXsystems, Inc.
2014-06-08 11:19:32 +00:00
pool, 0, svcpool_minthread_sysctl, "I",
"Minimal number of threads");
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
SYSCTL_ADD_PROC(&pool->sp_sysctl, sysctl_base, OID_AUTO,
Mark more nodes as CTLFLAG_MPSAFE or CTLFLAG_NEEDGIANT (17 of many) r357614 added CTLFLAG_NEEDGIANT to make it easier to find nodes that are still not MPSAFE (or already are but aren’t properly marked). Use it in preparation for a general review of all nodes. This is non-functional change that adds annotations to SYSCTL_NODE and SYSCTL_PROC nodes using one of the soon-to-be-required flags. Mark all obvious cases as MPSAFE. All entries that haven't been marked as MPSAFE before are by default marked as NEEDGIANT Approved by: kib (mentor, blanket) Commented by: kib, gallatin, melifaro Differential Revision: https://reviews.freebsd.org/D23718
2020-02-26 14:26:36 +00:00
"maxthreads", CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
Split RPC pool threads into number of smaller semi-isolated groups. Old design with unified thread pool was good from the point of thread utilization. But single pool-wide mutex became huge congestion point for systems with many CPUs. To reduce the congestion create several thread groups within a pool (one group for every 6 CPUs and 12 threads), each group with own mutex. Each connection during its registration is assigned to one of the groups in round-robin fashion. File affinify code may still move requests between the groups, but otherwise groups are self-contained. MFC after: 2 weeks Sponsored by: iXsystems, Inc.
2014-06-08 11:19:32 +00:00
pool, 0, svcpool_maxthread_sysctl, "I",
"Maximal number of threads");
SYSCTL_ADD_PROC(&pool->sp_sysctl, sysctl_base, OID_AUTO,
Mark more nodes as CTLFLAG_MPSAFE or CTLFLAG_NEEDGIANT (17 of many) r357614 added CTLFLAG_NEEDGIANT to make it easier to find nodes that are still not MPSAFE (or already are but aren’t properly marked). Use it in preparation for a general review of all nodes. This is non-functional change that adds annotations to SYSCTL_NODE and SYSCTL_PROC nodes using one of the soon-to-be-required flags. Mark all obvious cases as MPSAFE. All entries that haven't been marked as MPSAFE before are by default marked as NEEDGIANT Approved by: kib (mentor, blanket) Commented by: kib, gallatin, melifaro Differential Revision: https://reviews.freebsd.org/D23718
2020-02-26 14:26:36 +00:00
"threads", CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE,
Split RPC pool threads into number of smaller semi-isolated groups. Old design with unified thread pool was good from the point of thread utilization. But single pool-wide mutex became huge congestion point for systems with many CPUs. To reduce the congestion create several thread groups within a pool (one group for every 6 CPUs and 12 threads), each group with own mutex. Each connection during its registration is assigned to one of the groups in round-robin fashion. File affinify code may still move requests between the groups, but otherwise groups are self-contained. MFC after: 2 weeks Sponsored by: iXsystems, Inc.
2014-06-08 11:19:32 +00:00
pool, 0, svcpool_threads_sysctl, "I",
"Current number of threads");
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
SYSCTL_ADD_INT(&pool->sp_sysctl, sysctl_base, OID_AUTO,
Split RPC pool threads into number of smaller semi-isolated groups. Old design with unified thread pool was good from the point of thread utilization. But single pool-wide mutex became huge congestion point for systems with many CPUs. To reduce the congestion create several thread groups within a pool (one group for every 6 CPUs and 12 threads), each group with own mutex. Each connection during its registration is assigned to one of the groups in round-robin fashion. File affinify code may still move requests between the groups, but otherwise groups are self-contained. MFC after: 2 weeks Sponsored by: iXsystems, Inc.
2014-06-08 11:19:32 +00:00
"groups", CTLFLAG_RD, &pool->sp_groupcount, 0,
"Number of thread groups");
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
Fix overflow bugs in and remove obsolete limit from kernel RPC implementation. The kernel RPC code, which is responsible for the low-level scheduling of incoming NFS requests, contains a throttling mechanism that prevents too much kernel memory from being tied up by NFS requests that are being serviced. When the throttle is engaged, the RPC layer stops servicing incoming NFS sockets, resulting ultimately in backpressure on the clients (if they're using TCP). However, this is a very heavy-handed mechanism as it prevents all clients from making any requests, regardless of how heavy or light they are. (Thus, when engaged, the throttle often prevents clients from even mounting the filesystem.) The throttle mechanism applies specifically to requests that have been received by the RPC layer (from a TCP or UDP socket) and are queued waiting to be serviced by one of the nfsd threads; it does not limit the amount of backlog in the socket buffers. The original implementation limited the total bytes of queued requests to the minimum of a quarter of (nmbclusters * MCLBYTES) and 45 MiB. The former limit seems reasonable, since requests queued in the socket buffers and replies being constructed to the requests in progress will all require some amount of network memory, but the 45 MiB limit is plainly ridiculous for modern memory sizes: when running 256 service threads on a busy server, 45 MiB would result in just a single maximum-sized NFS3PROC_WRITE queued per thread before throttling. Removing this limit exposed integer-overflow bugs in the original computation, and related bugs in the routines that actually account for the amount of traffic enqueued for service threads. The old implementation also attempted to reduce accounting overhead by batching updates until each queue is fully drained, but this is prone to livelock, resulting in repeated accumulate-throttle-drain cycles on a busy server. Various data types are changed to long or unsigned long; explicit 64-bit types are not used due to the unavailability of 64-bit atomics on many 32-bit platforms, but those platforms also cannot support nmbclusters large enough to cause overflow. This code (in a 10.1 kernel) is presently running on production NFS servers at CSAIL. Summary of this revision: * Removes 45 MiB limit on requests queued for nfsd service threads * Fixes integer-overflow and signedness bugs * Avoids unnecessary throttling by not deferring accounting for completed requests Differential Revision: https://reviews.freebsd.org/D2165 Reviewed by: rmacklem, mav MFC after: 30 days Relnotes: yes Sponsored by: MIT Computer Science & Artificial Intelligence Laboratory
2015-04-01 00:45:47 +00:00
SYSCTL_ADD_ULONG(&pool->sp_sysctl, sysctl_base, OID_AUTO,
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
"request_space_used", CTLFLAG_RD,
Fix overflow bugs in and remove obsolete limit from kernel RPC implementation. The kernel RPC code, which is responsible for the low-level scheduling of incoming NFS requests, contains a throttling mechanism that prevents too much kernel memory from being tied up by NFS requests that are being serviced. When the throttle is engaged, the RPC layer stops servicing incoming NFS sockets, resulting ultimately in backpressure on the clients (if they're using TCP). However, this is a very heavy-handed mechanism as it prevents all clients from making any requests, regardless of how heavy or light they are. (Thus, when engaged, the throttle often prevents clients from even mounting the filesystem.) The throttle mechanism applies specifically to requests that have been received by the RPC layer (from a TCP or UDP socket) and are queued waiting to be serviced by one of the nfsd threads; it does not limit the amount of backlog in the socket buffers. The original implementation limited the total bytes of queued requests to the minimum of a quarter of (nmbclusters * MCLBYTES) and 45 MiB. The former limit seems reasonable, since requests queued in the socket buffers and replies being constructed to the requests in progress will all require some amount of network memory, but the 45 MiB limit is plainly ridiculous for modern memory sizes: when running 256 service threads on a busy server, 45 MiB would result in just a single maximum-sized NFS3PROC_WRITE queued per thread before throttling. Removing this limit exposed integer-overflow bugs in the original computation, and related bugs in the routines that actually account for the amount of traffic enqueued for service threads. The old implementation also attempted to reduce accounting overhead by batching updates until each queue is fully drained, but this is prone to livelock, resulting in repeated accumulate-throttle-drain cycles on a busy server. Various data types are changed to long or unsigned long; explicit 64-bit types are not used due to the unavailability of 64-bit atomics on many 32-bit platforms, but those platforms also cannot support nmbclusters large enough to cause overflow. This code (in a 10.1 kernel) is presently running on production NFS servers at CSAIL. Summary of this revision: * Removes 45 MiB limit on requests queued for nfsd service threads * Fixes integer-overflow and signedness bugs * Avoids unnecessary throttling by not deferring accounting for completed requests Differential Revision: https://reviews.freebsd.org/D2165 Reviewed by: rmacklem, mav MFC after: 30 days Relnotes: yes Sponsored by: MIT Computer Science & Artificial Intelligence Laboratory
2015-04-01 00:45:47 +00:00
&pool->sp_space_used,
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
"Space in parsed but not handled requests.");
Fix overflow bugs in and remove obsolete limit from kernel RPC implementation. The kernel RPC code, which is responsible for the low-level scheduling of incoming NFS requests, contains a throttling mechanism that prevents too much kernel memory from being tied up by NFS requests that are being serviced. When the throttle is engaged, the RPC layer stops servicing incoming NFS sockets, resulting ultimately in backpressure on the clients (if they're using TCP). However, this is a very heavy-handed mechanism as it prevents all clients from making any requests, regardless of how heavy or light they are. (Thus, when engaged, the throttle often prevents clients from even mounting the filesystem.) The throttle mechanism applies specifically to requests that have been received by the RPC layer (from a TCP or UDP socket) and are queued waiting to be serviced by one of the nfsd threads; it does not limit the amount of backlog in the socket buffers. The original implementation limited the total bytes of queued requests to the minimum of a quarter of (nmbclusters * MCLBYTES) and 45 MiB. The former limit seems reasonable, since requests queued in the socket buffers and replies being constructed to the requests in progress will all require some amount of network memory, but the 45 MiB limit is plainly ridiculous for modern memory sizes: when running 256 service threads on a busy server, 45 MiB would result in just a single maximum-sized NFS3PROC_WRITE queued per thread before throttling. Removing this limit exposed integer-overflow bugs in the original computation, and related bugs in the routines that actually account for the amount of traffic enqueued for service threads. The old implementation also attempted to reduce accounting overhead by batching updates until each queue is fully drained, but this is prone to livelock, resulting in repeated accumulate-throttle-drain cycles on a busy server. Various data types are changed to long or unsigned long; explicit 64-bit types are not used due to the unavailability of 64-bit atomics on many 32-bit platforms, but those platforms also cannot support nmbclusters large enough to cause overflow. This code (in a 10.1 kernel) is presently running on production NFS servers at CSAIL. Summary of this revision: * Removes 45 MiB limit on requests queued for nfsd service threads * Fixes integer-overflow and signedness bugs * Avoids unnecessary throttling by not deferring accounting for completed requests Differential Revision: https://reviews.freebsd.org/D2165 Reviewed by: rmacklem, mav MFC after: 30 days Relnotes: yes Sponsored by: MIT Computer Science & Artificial Intelligence Laboratory
2015-04-01 00:45:47 +00:00
SYSCTL_ADD_ULONG(&pool->sp_sysctl, sysctl_base, OID_AUTO,
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
"request_space_used_highest", CTLFLAG_RD,
Fix overflow bugs in and remove obsolete limit from kernel RPC implementation. The kernel RPC code, which is responsible for the low-level scheduling of incoming NFS requests, contains a throttling mechanism that prevents too much kernel memory from being tied up by NFS requests that are being serviced. When the throttle is engaged, the RPC layer stops servicing incoming NFS sockets, resulting ultimately in backpressure on the clients (if they're using TCP). However, this is a very heavy-handed mechanism as it prevents all clients from making any requests, regardless of how heavy or light they are. (Thus, when engaged, the throttle often prevents clients from even mounting the filesystem.) The throttle mechanism applies specifically to requests that have been received by the RPC layer (from a TCP or UDP socket) and are queued waiting to be serviced by one of the nfsd threads; it does not limit the amount of backlog in the socket buffers. The original implementation limited the total bytes of queued requests to the minimum of a quarter of (nmbclusters * MCLBYTES) and 45 MiB. The former limit seems reasonable, since requests queued in the socket buffers and replies being constructed to the requests in progress will all require some amount of network memory, but the 45 MiB limit is plainly ridiculous for modern memory sizes: when running 256 service threads on a busy server, 45 MiB would result in just a single maximum-sized NFS3PROC_WRITE queued per thread before throttling. Removing this limit exposed integer-overflow bugs in the original computation, and related bugs in the routines that actually account for the amount of traffic enqueued for service threads. The old implementation also attempted to reduce accounting overhead by batching updates until each queue is fully drained, but this is prone to livelock, resulting in repeated accumulate-throttle-drain cycles on a busy server. Various data types are changed to long or unsigned long; explicit 64-bit types are not used due to the unavailability of 64-bit atomics on many 32-bit platforms, but those platforms also cannot support nmbclusters large enough to cause overflow. This code (in a 10.1 kernel) is presently running on production NFS servers at CSAIL. Summary of this revision: * Removes 45 MiB limit on requests queued for nfsd service threads * Fixes integer-overflow and signedness bugs * Avoids unnecessary throttling by not deferring accounting for completed requests Differential Revision: https://reviews.freebsd.org/D2165 Reviewed by: rmacklem, mav MFC after: 30 days Relnotes: yes Sponsored by: MIT Computer Science & Artificial Intelligence Laboratory
2015-04-01 00:45:47 +00:00
&pool->sp_space_used_highest,
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
"Highest space used since reboot.");
Fix overflow bugs in and remove obsolete limit from kernel RPC implementation. The kernel RPC code, which is responsible for the low-level scheduling of incoming NFS requests, contains a throttling mechanism that prevents too much kernel memory from being tied up by NFS requests that are being serviced. When the throttle is engaged, the RPC layer stops servicing incoming NFS sockets, resulting ultimately in backpressure on the clients (if they're using TCP). However, this is a very heavy-handed mechanism as it prevents all clients from making any requests, regardless of how heavy or light they are. (Thus, when engaged, the throttle often prevents clients from even mounting the filesystem.) The throttle mechanism applies specifically to requests that have been received by the RPC layer (from a TCP or UDP socket) and are queued waiting to be serviced by one of the nfsd threads; it does not limit the amount of backlog in the socket buffers. The original implementation limited the total bytes of queued requests to the minimum of a quarter of (nmbclusters * MCLBYTES) and 45 MiB. The former limit seems reasonable, since requests queued in the socket buffers and replies being constructed to the requests in progress will all require some amount of network memory, but the 45 MiB limit is plainly ridiculous for modern memory sizes: when running 256 service threads on a busy server, 45 MiB would result in just a single maximum-sized NFS3PROC_WRITE queued per thread before throttling. Removing this limit exposed integer-overflow bugs in the original computation, and related bugs in the routines that actually account for the amount of traffic enqueued for service threads. The old implementation also attempted to reduce accounting overhead by batching updates until each queue is fully drained, but this is prone to livelock, resulting in repeated accumulate-throttle-drain cycles on a busy server. Various data types are changed to long or unsigned long; explicit 64-bit types are not used due to the unavailability of 64-bit atomics on many 32-bit platforms, but those platforms also cannot support nmbclusters large enough to cause overflow. This code (in a 10.1 kernel) is presently running on production NFS servers at CSAIL. Summary of this revision: * Removes 45 MiB limit on requests queued for nfsd service threads * Fixes integer-overflow and signedness bugs * Avoids unnecessary throttling by not deferring accounting for completed requests Differential Revision: https://reviews.freebsd.org/D2165 Reviewed by: rmacklem, mav MFC after: 30 days Relnotes: yes Sponsored by: MIT Computer Science & Artificial Intelligence Laboratory
2015-04-01 00:45:47 +00:00
SYSCTL_ADD_ULONG(&pool->sp_sysctl, sysctl_base, OID_AUTO,
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
"request_space_high", CTLFLAG_RW,
Fix overflow bugs in and remove obsolete limit from kernel RPC implementation. The kernel RPC code, which is responsible for the low-level scheduling of incoming NFS requests, contains a throttling mechanism that prevents too much kernel memory from being tied up by NFS requests that are being serviced. When the throttle is engaged, the RPC layer stops servicing incoming NFS sockets, resulting ultimately in backpressure on the clients (if they're using TCP). However, this is a very heavy-handed mechanism as it prevents all clients from making any requests, regardless of how heavy or light they are. (Thus, when engaged, the throttle often prevents clients from even mounting the filesystem.) The throttle mechanism applies specifically to requests that have been received by the RPC layer (from a TCP or UDP socket) and are queued waiting to be serviced by one of the nfsd threads; it does not limit the amount of backlog in the socket buffers. The original implementation limited the total bytes of queued requests to the minimum of a quarter of (nmbclusters * MCLBYTES) and 45 MiB. The former limit seems reasonable, since requests queued in the socket buffers and replies being constructed to the requests in progress will all require some amount of network memory, but the 45 MiB limit is plainly ridiculous for modern memory sizes: when running 256 service threads on a busy server, 45 MiB would result in just a single maximum-sized NFS3PROC_WRITE queued per thread before throttling. Removing this limit exposed integer-overflow bugs in the original computation, and related bugs in the routines that actually account for the amount of traffic enqueued for service threads. The old implementation also attempted to reduce accounting overhead by batching updates until each queue is fully drained, but this is prone to livelock, resulting in repeated accumulate-throttle-drain cycles on a busy server. Various data types are changed to long or unsigned long; explicit 64-bit types are not used due to the unavailability of 64-bit atomics on many 32-bit platforms, but those platforms also cannot support nmbclusters large enough to cause overflow. This code (in a 10.1 kernel) is presently running on production NFS servers at CSAIL. Summary of this revision: * Removes 45 MiB limit on requests queued for nfsd service threads * Fixes integer-overflow and signedness bugs * Avoids unnecessary throttling by not deferring accounting for completed requests Differential Revision: https://reviews.freebsd.org/D2165 Reviewed by: rmacklem, mav MFC after: 30 days Relnotes: yes Sponsored by: MIT Computer Science & Artificial Intelligence Laboratory
2015-04-01 00:45:47 +00:00
&pool->sp_space_high,
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
"Maximum space in parsed but not handled requests.");
Fix overflow bugs in and remove obsolete limit from kernel RPC implementation. The kernel RPC code, which is responsible for the low-level scheduling of incoming NFS requests, contains a throttling mechanism that prevents too much kernel memory from being tied up by NFS requests that are being serviced. When the throttle is engaged, the RPC layer stops servicing incoming NFS sockets, resulting ultimately in backpressure on the clients (if they're using TCP). However, this is a very heavy-handed mechanism as it prevents all clients from making any requests, regardless of how heavy or light they are. (Thus, when engaged, the throttle often prevents clients from even mounting the filesystem.) The throttle mechanism applies specifically to requests that have been received by the RPC layer (from a TCP or UDP socket) and are queued waiting to be serviced by one of the nfsd threads; it does not limit the amount of backlog in the socket buffers. The original implementation limited the total bytes of queued requests to the minimum of a quarter of (nmbclusters * MCLBYTES) and 45 MiB. The former limit seems reasonable, since requests queued in the socket buffers and replies being constructed to the requests in progress will all require some amount of network memory, but the 45 MiB limit is plainly ridiculous for modern memory sizes: when running 256 service threads on a busy server, 45 MiB would result in just a single maximum-sized NFS3PROC_WRITE queued per thread before throttling. Removing this limit exposed integer-overflow bugs in the original computation, and related bugs in the routines that actually account for the amount of traffic enqueued for service threads. The old implementation also attempted to reduce accounting overhead by batching updates until each queue is fully drained, but this is prone to livelock, resulting in repeated accumulate-throttle-drain cycles on a busy server. Various data types are changed to long or unsigned long; explicit 64-bit types are not used due to the unavailability of 64-bit atomics on many 32-bit platforms, but those platforms also cannot support nmbclusters large enough to cause overflow. This code (in a 10.1 kernel) is presently running on production NFS servers at CSAIL. Summary of this revision: * Removes 45 MiB limit on requests queued for nfsd service threads * Fixes integer-overflow and signedness bugs * Avoids unnecessary throttling by not deferring accounting for completed requests Differential Revision: https://reviews.freebsd.org/D2165 Reviewed by: rmacklem, mav MFC after: 30 days Relnotes: yes Sponsored by: MIT Computer Science & Artificial Intelligence Laboratory
2015-04-01 00:45:47 +00:00
SYSCTL_ADD_ULONG(&pool->sp_sysctl, sysctl_base, OID_AUTO,
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
"request_space_low", CTLFLAG_RW,
Fix overflow bugs in and remove obsolete limit from kernel RPC implementation. The kernel RPC code, which is responsible for the low-level scheduling of incoming NFS requests, contains a throttling mechanism that prevents too much kernel memory from being tied up by NFS requests that are being serviced. When the throttle is engaged, the RPC layer stops servicing incoming NFS sockets, resulting ultimately in backpressure on the clients (if they're using TCP). However, this is a very heavy-handed mechanism as it prevents all clients from making any requests, regardless of how heavy or light they are. (Thus, when engaged, the throttle often prevents clients from even mounting the filesystem.) The throttle mechanism applies specifically to requests that have been received by the RPC layer (from a TCP or UDP socket) and are queued waiting to be serviced by one of the nfsd threads; it does not limit the amount of backlog in the socket buffers. The original implementation limited the total bytes of queued requests to the minimum of a quarter of (nmbclusters * MCLBYTES) and 45 MiB. The former limit seems reasonable, since requests queued in the socket buffers and replies being constructed to the requests in progress will all require some amount of network memory, but the 45 MiB limit is plainly ridiculous for modern memory sizes: when running 256 service threads on a busy server, 45 MiB would result in just a single maximum-sized NFS3PROC_WRITE queued per thread before throttling. Removing this limit exposed integer-overflow bugs in the original computation, and related bugs in the routines that actually account for the amount of traffic enqueued for service threads. The old implementation also attempted to reduce accounting overhead by batching updates until each queue is fully drained, but this is prone to livelock, resulting in repeated accumulate-throttle-drain cycles on a busy server. Various data types are changed to long or unsigned long; explicit 64-bit types are not used due to the unavailability of 64-bit atomics on many 32-bit platforms, but those platforms also cannot support nmbclusters large enough to cause overflow. This code (in a 10.1 kernel) is presently running on production NFS servers at CSAIL. Summary of this revision: * Removes 45 MiB limit on requests queued for nfsd service threads * Fixes integer-overflow and signedness bugs * Avoids unnecessary throttling by not deferring accounting for completed requests Differential Revision: https://reviews.freebsd.org/D2165 Reviewed by: rmacklem, mav MFC after: 30 days Relnotes: yes Sponsored by: MIT Computer Science & Artificial Intelligence Laboratory
2015-04-01 00:45:47 +00:00
&pool->sp_space_low,
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
"Low water mark for request space.");
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the kernel changes.
2011-01-12 19:54:19 +00:00
SYSCTL_ADD_INT(&pool->sp_sysctl, sysctl_base, OID_AUTO,
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
"request_space_throttled", CTLFLAG_RD,
&pool->sp_space_throttled, 0,
"Whether nfs requests are currently throttled");
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the kernel changes.
2011-01-12 19:54:19 +00:00
SYSCTL_ADD_INT(&pool->sp_sysctl, sysctl_base, OID_AUTO,
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
"request_space_throttle_count", CTLFLAG_RD,
&pool->sp_space_throttle_count, 0,
"Count of times throttling based on request space has occurred");
}
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
return pool;
}
add svcpool_close to handle killed nfsd threads This patch adds a new function to the server krpc called svcpool_close(). It is similar to svcpool_destroy(), but does not free the data structures, so that the pool can be used again. This function is then used instead of svcpool_destroy(), svcpool_create() when the nfsd threads are killed. PR: 204340 Reported by: Panzura Approved by: rmacklem Obtained from: rmacklem MFC after: 1 week
2017-02-14 17:49:08 +00:00
/*
* Code common to svcpool_destroy() and svcpool_close(), which cleans up
* the pool data structures.
*/
static void
svcpool_cleanup(SVCPOOL *pool)
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
{
Split RPC pool threads into number of smaller semi-isolated groups. Old design with unified thread pool was good from the point of thread utilization. But single pool-wide mutex became huge congestion point for systems with many CPUs. To reduce the congestion create several thread groups within a pool (one group for every 6 CPUs and 12 threads), each group with own mutex. Each connection during its registration is assigned to one of the groups in round-robin fashion. File affinify code may still move requests between the groups, but otherwise groups are self-contained. MFC after: 2 weeks Sponsored by: iXsystems, Inc.
2014-06-08 11:19:32 +00:00
SVCGROUP *grp;
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
SVCXPRT *xprt, *nxprt;
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
struct svc_callout *s;
Rework NFS Duplicate Request Cache cleanup logic. - Introduce additional hash to group requests by hash of sockref. This allows to process TCP acknowledgements without looping though all the cache, and as result allows to do it every time. - Indroduce additional callbacks to notify application layer about sockets disconnection. Without this last few requests processed just before socket disconnection never processed their ACKs and stuck in cache for many hours. - Implement transport-specific method for tracking reply acknowledgements. New implementation does not cross multiple stack layers to get the data and does not have race conditions that previously made some requests stuck in cache. This could be done more efficiently at sockbuf layer, but that would broke some KBIs, while I don't know other consumers for it aside NFS. - Instead of traversing all DRC twice per request, run cleaning only once per request, and except in some conditions traverse only single hash slot at a time. Together this limits NFS DRC growth only to situations of real connectivity problems. If network is working well, and so all replies are acknowledged, cache remains almost empty even after hours of heavy load. Without this change on the same test cache was growing to many thousand requests even with perfectly working local network. As another result this reduces CPU time spent on the DRC handling during SPEC NFS benchmark from about 10% to 0.5%. Sponsored by: iXsystems, Inc.
2014-01-03 15:09:59 +00:00
struct svc_loss_callout *sl;
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
struct svcxprt_list cleanup;
Split RPC pool threads into number of smaller semi-isolated groups. Old design with unified thread pool was good from the point of thread utilization. But single pool-wide mutex became huge congestion point for systems with many CPUs. To reduce the congestion create several thread groups within a pool (one group for every 6 CPUs and 12 threads), each group with own mutex. Each connection during its registration is assigned to one of the groups in round-robin fashion. File affinify code may still move requests between the groups, but otherwise groups are self-contained. MFC after: 2 weeks Sponsored by: iXsystems, Inc.
2014-06-08 11:19:32 +00:00
int g;
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
TAILQ_INIT(&cleanup);
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
Split RPC pool threads into number of smaller semi-isolated groups. Old design with unified thread pool was good from the point of thread utilization. But single pool-wide mutex became huge congestion point for systems with many CPUs. To reduce the congestion create several thread groups within a pool (one group for every 6 CPUs and 12 threads), each group with own mutex. Each connection during its registration is assigned to one of the groups in round-robin fashion. File affinify code may still move requests between the groups, but otherwise groups are self-contained. MFC after: 2 weeks Sponsored by: iXsystems, Inc.
2014-06-08 11:19:32 +00:00
for (g = 0; g < SVC_MAXGROUPS; g++) {
grp = &pool->sp_groups[g];
mtx_lock(&grp->sg_lock);
while ((xprt = TAILQ_FIRST(&grp->sg_xlist)) != NULL) {
xprt_unregister_locked(xprt);
TAILQ_INSERT_TAIL(&cleanup, xprt, xp_link);
}
mtx_unlock(&grp->sg_lock);
}
TAILQ_FOREACH_SAFE(xprt, &cleanup, xp_link, nxprt) {
nfsd: fix a NFSv4.1 Linux client mount stuck in CLOSE_WAIT It was reported that a NFSv4.1 Linux client mount against a FreeBSD12 server was hung, with the TCP connection in CLOSE_WAIT state on the server. When a NFSv4.1/4.2 mount is done and the back channel is bound to the TCP connection, the soclose() is delayed until a new TCP connection is bound to the back channel, due to a reference count being held on the SVCXPRT structure in the krpc for the socket. Without the soclose() call, the socket will remain in CLOSE_WAIT and this somehow caused the Linux client to hang. This patch adds calls to soshutdown(.., SHUT_WR) that are performed when the server side krpc sees that the socket is no longer usable. Since this can be done before the back channel is bound to a new TCP connection, it allows the TCP connection to proceed to CLOSED state. PR: 254590 Reported by: jbreitman@tildenparkcapital.com Reviewed by: tuexen Comments by: kevans MFC after: 2 weeks Differential Revision: https://reviews.freebsd.org/D29526
2021-04-27 22:32:35 +00:00
if (xprt->xp_socket != NULL)
soshutdown(xprt->xp_socket, SHUT_WR);
Split RPC pool threads into number of smaller semi-isolated groups. Old design with unified thread pool was good from the point of thread utilization. But single pool-wide mutex became huge congestion point for systems with many CPUs. To reduce the congestion create several thread groups within a pool (one group for every 6 CPUs and 12 threads), each group with own mutex. Each connection during its registration is assigned to one of the groups in round-robin fashion. File affinify code may still move requests between the groups, but otherwise groups are self-contained. MFC after: 2 weeks Sponsored by: iXsystems, Inc.
2014-06-08 11:19:32 +00:00
SVC_RELEASE(xprt);
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
}
Split RPC pool threads into number of smaller semi-isolated groups. Old design with unified thread pool was good from the point of thread utilization. But single pool-wide mutex became huge congestion point for systems with many CPUs. To reduce the congestion create several thread groups within a pool (one group for every 6 CPUs and 12 threads), each group with own mutex. Each connection during its registration is assigned to one of the groups in round-robin fashion. File affinify code may still move requests between the groups, but otherwise groups are self-contained. MFC after: 2 weeks Sponsored by: iXsystems, Inc.
2014-06-08 11:19:32 +00:00
mtx_lock(&pool->sp_lock);
Rework NFS Duplicate Request Cache cleanup logic. - Introduce additional hash to group requests by hash of sockref. This allows to process TCP acknowledgements without looping though all the cache, and as result allows to do it every time. - Indroduce additional callbacks to notify application layer about sockets disconnection. Without this last few requests processed just before socket disconnection never processed their ACKs and stuck in cache for many hours. - Implement transport-specific method for tracking reply acknowledgements. New implementation does not cross multiple stack layers to get the data and does not have race conditions that previously made some requests stuck in cache. This could be done more efficiently at sockbuf layer, but that would broke some KBIs, while I don't know other consumers for it aside NFS. - Instead of traversing all DRC twice per request, run cleaning only once per request, and except in some conditions traverse only single hash slot at a time. Together this limits NFS DRC growth only to situations of real connectivity problems. If network is working well, and so all replies are acknowledged, cache remains almost empty even after hours of heavy load. Without this change on the same test cache was growing to many thousand requests even with perfectly working local network. As another result this reduces CPU time spent on the DRC handling during SPEC NFS benchmark from about 10% to 0.5%. Sponsored by: iXsystems, Inc.
2014-01-03 15:09:59 +00:00
while ((s = TAILQ_FIRST(&pool->sp_callouts)) != NULL) {
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
mtx_unlock(&pool->sp_lock);
svc_unreg(pool, s->sc_prog, s->sc_vers);
mtx_lock(&pool->sp_lock);
}
Rework NFS Duplicate Request Cache cleanup logic. - Introduce additional hash to group requests by hash of sockref. This allows to process TCP acknowledgements without looping though all the cache, and as result allows to do it every time. - Indroduce additional callbacks to notify application layer about sockets disconnection. Without this last few requests processed just before socket disconnection never processed their ACKs and stuck in cache for many hours. - Implement transport-specific method for tracking reply acknowledgements. New implementation does not cross multiple stack layers to get the data and does not have race conditions that previously made some requests stuck in cache. This could be done more efficiently at sockbuf layer, but that would broke some KBIs, while I don't know other consumers for it aside NFS. - Instead of traversing all DRC twice per request, run cleaning only once per request, and except in some conditions traverse only single hash slot at a time. Together this limits NFS DRC growth only to situations of real connectivity problems. If network is working well, and so all replies are acknowledged, cache remains almost empty even after hours of heavy load. Without this change on the same test cache was growing to many thousand requests even with perfectly working local network. As another result this reduces CPU time spent on the DRC handling during SPEC NFS benchmark from about 10% to 0.5%. Sponsored by: iXsystems, Inc.
2014-01-03 15:09:59 +00:00
while ((sl = TAILQ_FIRST(&pool->sp_lcallouts)) != NULL) {
mtx_unlock(&pool->sp_lock);
svc_loss_unreg(pool, sl->slc_dispatch);
mtx_lock(&pool->sp_lock);
}
Fix a lockorder reversal I introduced in r193436 when I moved the mtx_destroy() of the pool mutex to after SVC_RELEASE(), because the pool mutex was still locked when soclose() was called by svc_dg_destroy(). To fix this, an mtx_unlock() was added where mtx_destroy() was before r193436. Reviewed by: jhb Tested by: pho Approved by: rwatson (mentor)
2009-06-07 01:06:56 +00:00
mtx_unlock(&pool->sp_lock);
add svcpool_close to handle killed nfsd threads This patch adds a new function to the server krpc called svcpool_close(). It is similar to svcpool_destroy(), but does not free the data structures, so that the pool can be used again. This function is then used instead of svcpool_destroy(), svcpool_create() when the nfsd threads are killed. PR: 204340 Reported by: Panzura Approved by: rmacklem Obtained from: rmacklem MFC after: 1 week
2017-02-14 17:49:08 +00:00
}
void
svcpool_destroy(SVCPOOL *pool)
{
SVCGROUP *grp;
int g;
svcpool_cleanup(pool);
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
Split RPC pool threads into number of smaller semi-isolated groups. Old design with unified thread pool was good from the point of thread utilization. But single pool-wide mutex became huge congestion point for systems with many CPUs. To reduce the congestion create several thread groups within a pool (one group for every 6 CPUs and 12 threads), each group with own mutex. Each connection during its registration is assigned to one of the groups in round-robin fashion. File affinify code may still move requests between the groups, but otherwise groups are self-contained. MFC after: 2 weeks Sponsored by: iXsystems, Inc.
2014-06-08 11:19:32 +00:00
for (g = 0; g < SVC_MAXGROUPS; g++) {
grp = &pool->sp_groups[g];
mtx_destroy(&grp->sg_lock);
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
}
Fix two races in the server side krpc w.r.t upcalls: Add a flag so that soupcall_clear() is only called once to cancel an upcall. Move the test for xprt_registered in the upcall down to after the mtx_lock() of the pool mutex, to catch the case where it is unregistered while the upcall is waiting for the mutex. Also, move the mtx_destroy() of the pool mutex to after SVC_RELEASE(), so that it isn't destroyed before the upcalls are disabled. Reviewed by: dfr, jhb Tested by: pho Approved by: kib (mentor)
2009-06-04 14:13:06 +00:00
mtx_destroy(&pool->sp_lock);
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
if (pool->sp_rcache)
replay_freecache(pool->sp_rcache);
sysctl_ctx_free(&pool->sp_sysctl);
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
free(pool, M_RPC);
}
add svcpool_close to handle killed nfsd threads This patch adds a new function to the server krpc called svcpool_close(). It is similar to svcpool_destroy(), but does not free the data structures, so that the pool can be used again. This function is then used instead of svcpool_destroy(), svcpool_create() when the nfsd threads are killed. PR: 204340 Reported by: Panzura Approved by: rmacklem Obtained from: rmacklem MFC after: 1 week
2017-02-14 17:49:08 +00:00
/*
* Similar to svcpool_destroy(), except that it does not destroy the actual
* data structures. As such, "pool" may be used again.
*/
void
svcpool_close(SVCPOOL *pool)
{
SVCGROUP *grp;
int g;
svcpool_cleanup(pool);
/* Now, initialize the pool's state for a fresh svc_run() call. */
mtx_lock(&pool->sp_lock);
pool->sp_state = SVCPOOL_INIT;
mtx_unlock(&pool->sp_lock);
for (g = 0; g < SVC_MAXGROUPS; g++) {
grp = &pool->sp_groups[g];
mtx_lock(&grp->sg_lock);
grp->sg_state = SVCPOOL_ACTIVE;
mtx_unlock(&grp->sg_lock);
}
}
Split RPC pool threads into number of smaller semi-isolated groups. Old design with unified thread pool was good from the point of thread utilization. But single pool-wide mutex became huge congestion point for systems with many CPUs. To reduce the congestion create several thread groups within a pool (one group for every 6 CPUs and 12 threads), each group with own mutex. Each connection during its registration is assigned to one of the groups in round-robin fashion. File affinify code may still move requests between the groups, but otherwise groups are self-contained. MFC after: 2 weeks Sponsored by: iXsystems, Inc.
2014-06-08 11:19:32 +00:00
/*
* Sysctl handler to get the present thread count on a pool
*/
static int
svcpool_threads_sysctl(SYSCTL_HANDLER_ARGS)
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
{
Split RPC pool threads into number of smaller semi-isolated groups. Old design with unified thread pool was good from the point of thread utilization. But single pool-wide mutex became huge congestion point for systems with many CPUs. To reduce the congestion create several thread groups within a pool (one group for every 6 CPUs and 12 threads), each group with own mutex. Each connection during its registration is assigned to one of the groups in round-robin fashion. File affinify code may still move requests between the groups, but otherwise groups are self-contained. MFC after: 2 weeks Sponsored by: iXsystems, Inc.
2014-06-08 11:19:32 +00:00
SVCPOOL *pool;
int threads, error, g;
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
Split RPC pool threads into number of smaller semi-isolated groups. Old design with unified thread pool was good from the point of thread utilization. But single pool-wide mutex became huge congestion point for systems with many CPUs. To reduce the congestion create several thread groups within a pool (one group for every 6 CPUs and 12 threads), each group with own mutex. Each connection during its registration is assigned to one of the groups in round-robin fashion. File affinify code may still move requests between the groups, but otherwise groups are self-contained. MFC after: 2 weeks Sponsored by: iXsystems, Inc.
2014-06-08 11:19:32 +00:00
pool = oidp->oid_arg1;
threads = 0;
mtx_lock(&pool->sp_lock);
for (g = 0; g < pool->sp_groupcount; g++)
threads += pool->sp_groups[g].sg_threadcount;
mtx_unlock(&pool->sp_lock);
error = sysctl_handle_int(oidp, &threads, 0, req);
return (error);
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
}
/*
* Sysctl handler to set the minimum thread count on a pool
*/
static int
svcpool_minthread_sysctl(SYSCTL_HANDLER_ARGS)
{
SVCPOOL *pool;
Split RPC pool threads into number of smaller semi-isolated groups. Old design with unified thread pool was good from the point of thread utilization. But single pool-wide mutex became huge congestion point for systems with many CPUs. To reduce the congestion create several thread groups within a pool (one group for every 6 CPUs and 12 threads), each group with own mutex. Each connection during its registration is assigned to one of the groups in round-robin fashion. File affinify code may still move requests between the groups, but otherwise groups are self-contained. MFC after: 2 weeks Sponsored by: iXsystems, Inc.
2014-06-08 11:19:32 +00:00
int newminthreads, error, g;
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
pool = oidp->oid_arg1;
newminthreads = pool->sp_minthreads;
error = sysctl_handle_int(oidp, &newminthreads, 0, req);
if (error == 0 && newminthreads != pool->sp_minthreads) {
if (newminthreads > pool->sp_maxthreads)
return (EINVAL);
mtx_lock(&pool->sp_lock);
pool->sp_minthreads = newminthreads;
Split RPC pool threads into number of smaller semi-isolated groups. Old design with unified thread pool was good from the point of thread utilization. But single pool-wide mutex became huge congestion point for systems with many CPUs. To reduce the congestion create several thread groups within a pool (one group for every 6 CPUs and 12 threads), each group with own mutex. Each connection during its registration is assigned to one of the groups in round-robin fashion. File affinify code may still move requests between the groups, but otherwise groups are self-contained. MFC after: 2 weeks Sponsored by: iXsystems, Inc.
2014-06-08 11:19:32 +00:00
for (g = 0; g < pool->sp_groupcount; g++) {
pool->sp_groups[g].sg_minthreads = max(1,
pool->sp_minthreads / pool->sp_groupcount);
}
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
mtx_unlock(&pool->sp_lock);
}
return (error);
}
/*
* Sysctl handler to set the maximum thread count on a pool
*/
static int
svcpool_maxthread_sysctl(SYSCTL_HANDLER_ARGS)
{
SVCPOOL *pool;
Split RPC pool threads into number of smaller semi-isolated groups. Old design with unified thread pool was good from the point of thread utilization. But single pool-wide mutex became huge congestion point for systems with many CPUs. To reduce the congestion create several thread groups within a pool (one group for every 6 CPUs and 12 threads), each group with own mutex. Each connection during its registration is assigned to one of the groups in round-robin fashion. File affinify code may still move requests between the groups, but otherwise groups are self-contained. MFC after: 2 weeks Sponsored by: iXsystems, Inc.
2014-06-08 11:19:32 +00:00
int newmaxthreads, error, g;
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
pool = oidp->oid_arg1;
newmaxthreads = pool->sp_maxthreads;
error = sysctl_handle_int(oidp, &newmaxthreads, 0, req);
if (error == 0 && newmaxthreads != pool->sp_maxthreads) {
if (newmaxthreads < pool->sp_minthreads)
return (EINVAL);
mtx_lock(&pool->sp_lock);
pool->sp_maxthreads = newmaxthreads;
Split RPC pool threads into number of smaller semi-isolated groups. Old design with unified thread pool was good from the point of thread utilization. But single pool-wide mutex became huge congestion point for systems with many CPUs. To reduce the congestion create several thread groups within a pool (one group for every 6 CPUs and 12 threads), each group with own mutex. Each connection during its registration is assigned to one of the groups in round-robin fashion. File affinify code may still move requests between the groups, but otherwise groups are self-contained. MFC after: 2 weeks Sponsored by: iXsystems, Inc.
2014-06-08 11:19:32 +00:00
for (g = 0; g < pool->sp_groupcount; g++) {
pool->sp_groups[g].sg_maxthreads = max(1,
pool->sp_maxthreads / pool->sp_groupcount);
}
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
mtx_unlock(&pool->sp_lock);
}
return (error);
}
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
/*
* Activate a transport handle.
*/
void
xprt_register(SVCXPRT *xprt)
{
SVCPOOL *pool = xprt->xp_pool;
Split RPC pool threads into number of smaller semi-isolated groups. Old design with unified thread pool was good from the point of thread utilization. But single pool-wide mutex became huge congestion point for systems with many CPUs. To reduce the congestion create several thread groups within a pool (one group for every 6 CPUs and 12 threads), each group with own mutex. Each connection during its registration is assigned to one of the groups in round-robin fashion. File affinify code may still move requests between the groups, but otherwise groups are self-contained. MFC after: 2 weeks Sponsored by: iXsystems, Inc.
2014-06-08 11:19:32 +00:00
SVCGROUP *grp;
int g;
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
Since svc_[dg|vc|tli|tp]_create() did not hold a reference count on the SVCXPTR structure returned by them, it was possible for the structure to be free'd before svc_reg() had been completed using the structure. This patch acquires a reference count on the newly created structure that is returned by svc_[dg|vc|tli|tp]_create(). It also adds the appropriate SVC_RELEASE() calls to the callers, except the experimental nfs subsystem. The latter will be committed separately. Submitted by: dfr Tested by: pho Approved by: kib (mentor)
2009-06-17 22:50:26 +00:00
SVC_ACQUIRE(xprt);
Split RPC pool threads into number of smaller semi-isolated groups. Old design with unified thread pool was good from the point of thread utilization. But single pool-wide mutex became huge congestion point for systems with many CPUs. To reduce the congestion create several thread groups within a pool (one group for every 6 CPUs and 12 threads), each group with own mutex. Each connection during its registration is assigned to one of the groups in round-robin fashion. File affinify code may still move requests between the groups, but otherwise groups are self-contained. MFC after: 2 weeks Sponsored by: iXsystems, Inc.
2014-06-08 11:19:32 +00:00
g = atomic_fetchadd_int(&pool->sp_nextgroup, 1) % pool->sp_groupcount;
xprt->xp_group = grp = &pool->sp_groups[g];
mtx_lock(&grp->sg_lock);
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
xprt->xp_registered = TRUE;
xprt->xp_active = FALSE;
Split RPC pool threads into number of smaller semi-isolated groups. Old design with unified thread pool was good from the point of thread utilization. But single pool-wide mutex became huge congestion point for systems with many CPUs. To reduce the congestion create several thread groups within a pool (one group for every 6 CPUs and 12 threads), each group with own mutex. Each connection during its registration is assigned to one of the groups in round-robin fashion. File affinify code may still move requests between the groups, but otherwise groups are self-contained. MFC after: 2 weeks Sponsored by: iXsystems, Inc.
2014-06-08 11:19:32 +00:00
TAILQ_INSERT_TAIL(&grp->sg_xlist, xprt, xp_link);
mtx_unlock(&grp->sg_lock);
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
}
/*
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
* De-activate a transport handle. Note: the locked version doesn't
* release the transport - caller must do that after dropping the pool
* lock.
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
*/
static void
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
xprt_unregister_locked(SVCXPRT *xprt)
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
{
Split RPC pool threads into number of smaller semi-isolated groups. Old design with unified thread pool was good from the point of thread utilization. But single pool-wide mutex became huge congestion point for systems with many CPUs. To reduce the congestion create several thread groups within a pool (one group for every 6 CPUs and 12 threads), each group with own mutex. Each connection during its registration is assigned to one of the groups in round-robin fashion. File affinify code may still move requests between the groups, but otherwise groups are self-contained. MFC after: 2 weeks Sponsored by: iXsystems, Inc.
2014-06-08 11:19:32 +00:00
SVCGROUP *grp = xprt->xp_group;
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
Split RPC pool threads into number of smaller semi-isolated groups. Old design with unified thread pool was good from the point of thread utilization. But single pool-wide mutex became huge congestion point for systems with many CPUs. To reduce the congestion create several thread groups within a pool (one group for every 6 CPUs and 12 threads), each group with own mutex. Each connection during its registration is assigned to one of the groups in round-robin fashion. File affinify code may still move requests between the groups, but otherwise groups are self-contained. MFC after: 2 weeks Sponsored by: iXsystems, Inc.
2014-06-08 11:19:32 +00:00
mtx_assert(&grp->sg_lock, MA_OWNED);
Add a check to xprt_unregister() to catch the case where another thread has already unregistered the structure. Also add a KASSERT() to xprt_unregister_locked() to check that the structure hasn't already been unregistered. Reviewed by: jhb Tested by: pho Approved by: kib (mentor)
2009-06-07 20:38:41 +00:00
KASSERT(xprt->xp_registered == TRUE,
("xprt_unregister_locked: not registered"));
Remove several linear list traversals per request from RPC server code. Do not insert active ports into pool->sp_active list if they are success- fully assigned to some thread. This makes that list include only ports that really require attention, and so traversal can be reduced to simple taking the first one. Remove idle thread from pool->sp_idlethreads list when assigning some work (port of requests) to it. That again makes possible to replace list traversals with simple taking the first element.
2013-12-20 17:39:07 +00:00
xprt_inactive_locked(xprt);
Split RPC pool threads into number of smaller semi-isolated groups. Old design with unified thread pool was good from the point of thread utilization. But single pool-wide mutex became huge congestion point for systems with many CPUs. To reduce the congestion create several thread groups within a pool (one group for every 6 CPUs and 12 threads), each group with own mutex. Each connection during its registration is assigned to one of the groups in round-robin fashion. File affinify code may still move requests between the groups, but otherwise groups are self-contained. MFC after: 2 weeks Sponsored by: iXsystems, Inc.
2014-06-08 11:19:32 +00:00
TAILQ_REMOVE(&grp->sg_xlist, xprt, xp_link);
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
xprt->xp_registered = FALSE;
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
}
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
void
xprt_unregister(SVCXPRT *xprt)
{
Split RPC pool threads into number of smaller semi-isolated groups. Old design with unified thread pool was good from the point of thread utilization. But single pool-wide mutex became huge congestion point for systems with many CPUs. To reduce the congestion create several thread groups within a pool (one group for every 6 CPUs and 12 threads), each group with own mutex. Each connection during its registration is assigned to one of the groups in round-robin fashion. File affinify code may still move requests between the groups, but otherwise groups are self-contained. MFC after: 2 weeks Sponsored by: iXsystems, Inc.
2014-06-08 11:19:32 +00:00
SVCGROUP *grp = xprt->xp_group;
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
Split RPC pool threads into number of smaller semi-isolated groups. Old design with unified thread pool was good from the point of thread utilization. But single pool-wide mutex became huge congestion point for systems with many CPUs. To reduce the congestion create several thread groups within a pool (one group for every 6 CPUs and 12 threads), each group with own mutex. Each connection during its registration is assigned to one of the groups in round-robin fashion. File affinify code may still move requests between the groups, but otherwise groups are self-contained. MFC after: 2 weeks Sponsored by: iXsystems, Inc.
2014-06-08 11:19:32 +00:00
mtx_lock(&grp->sg_lock);
Add a check to xprt_unregister() to catch the case where another thread has already unregistered the structure. Also add a KASSERT() to xprt_unregister_locked() to check that the structure hasn't already been unregistered. Reviewed by: jhb Tested by: pho Approved by: kib (mentor)
2009-06-07 20:38:41 +00:00
if (xprt->xp_registered == FALSE) {
/* Already unregistered by another thread */
Split RPC pool threads into number of smaller semi-isolated groups. Old design with unified thread pool was good from the point of thread utilization. But single pool-wide mutex became huge congestion point for systems with many CPUs. To reduce the congestion create several thread groups within a pool (one group for every 6 CPUs and 12 threads), each group with own mutex. Each connection during its registration is assigned to one of the groups in round-robin fashion. File affinify code may still move requests between the groups, but otherwise groups are self-contained. MFC after: 2 weeks Sponsored by: iXsystems, Inc.
2014-06-08 11:19:32 +00:00
mtx_unlock(&grp->sg_lock);
Add a check to xprt_unregister() to catch the case where another thread has already unregistered the structure. Also add a KASSERT() to xprt_unregister_locked() to check that the structure hasn't already been unregistered. Reviewed by: jhb Tested by: pho Approved by: kib (mentor)
2009-06-07 20:38:41 +00:00
return;
}
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
xprt_unregister_locked(xprt);
Split RPC pool threads into number of smaller semi-isolated groups. Old design with unified thread pool was good from the point of thread utilization. But single pool-wide mutex became huge congestion point for systems with many CPUs. To reduce the congestion create several thread groups within a pool (one group for every 6 CPUs and 12 threads), each group with own mutex. Each connection during its registration is assigned to one of the groups in round-robin fashion. File affinify code may still move requests between the groups, but otherwise groups are self-contained. MFC after: 2 weeks Sponsored by: iXsystems, Inc.
2014-06-08 11:19:32 +00:00
mtx_unlock(&grp->sg_lock);
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
nfsd: fix a NFSv4.1 Linux client mount stuck in CLOSE_WAIT It was reported that a NFSv4.1 Linux client mount against a FreeBSD12 server was hung, with the TCP connection in CLOSE_WAIT state on the server. When a NFSv4.1/4.2 mount is done and the back channel is bound to the TCP connection, the soclose() is delayed until a new TCP connection is bound to the back channel, due to a reference count being held on the SVCXPRT structure in the krpc for the socket. Without the soclose() call, the socket will remain in CLOSE_WAIT and this somehow caused the Linux client to hang. This patch adds calls to soshutdown(.., SHUT_WR) that are performed when the server side krpc sees that the socket is no longer usable. Since this can be done before the back channel is bound to a new TCP connection, it allows the TCP connection to proceed to CLOSED state. PR: 254590 Reported by: jbreitman@tildenparkcapital.com Reviewed by: tuexen Comments by: kevans MFC after: 2 weeks Differential Revision: https://reviews.freebsd.org/D29526
2021-04-27 22:32:35 +00:00
if (xprt->xp_socket != NULL)
soshutdown(xprt->xp_socket, SHUT_WR);
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
SVC_RELEASE(xprt);
}
Remove several linear list traversals per request from RPC server code. Do not insert active ports into pool->sp_active list if they are success- fully assigned to some thread. This makes that list include only ports that really require attention, and so traversal can be reduced to simple taking the first one. Remove idle thread from pool->sp_idlethreads list when assigning some work (port of requests) to it. That again makes possible to replace list traversals with simple taking the first element.
2013-12-20 17:39:07 +00:00
/*
* Attempt to assign a service thread to this transport.
*/
static int
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
xprt_assignthread(SVCXPRT *xprt)
{
Split RPC pool threads into number of smaller semi-isolated groups. Old design with unified thread pool was good from the point of thread utilization. But single pool-wide mutex became huge congestion point for systems with many CPUs. To reduce the congestion create several thread groups within a pool (one group for every 6 CPUs and 12 threads), each group with own mutex. Each connection during its registration is assigned to one of the groups in round-robin fashion. File affinify code may still move requests between the groups, but otherwise groups are self-contained. MFC after: 2 weeks Sponsored by: iXsystems, Inc.
2014-06-08 11:19:32 +00:00
SVCGROUP *grp = xprt->xp_group;
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
SVCTHREAD *st;
Split RPC pool threads into number of smaller semi-isolated groups. Old design with unified thread pool was good from the point of thread utilization. But single pool-wide mutex became huge congestion point for systems with many CPUs. To reduce the congestion create several thread groups within a pool (one group for every 6 CPUs and 12 threads), each group with own mutex. Each connection during its registration is assigned to one of the groups in round-robin fashion. File affinify code may still move requests between the groups, but otherwise groups are self-contained. MFC after: 2 weeks Sponsored by: iXsystems, Inc.
2014-06-08 11:19:32 +00:00
mtx_assert(&grp->sg_lock, MA_OWNED);
st = LIST_FIRST(&grp->sg_idlethreads);
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
if (st) {
Remove several linear list traversals per request from RPC server code. Do not insert active ports into pool->sp_active list if they are success- fully assigned to some thread. This makes that list include only ports that really require attention, and so traversal can be reduced to simple taking the first one. Remove idle thread from pool->sp_idlethreads list when assigning some work (port of requests) to it. That again makes possible to replace list traversals with simple taking the first element.
2013-12-20 17:39:07 +00:00
LIST_REMOVE(st, st_ilink);
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
SVC_ACQUIRE(xprt);
xprt->xp_thread = st;
st->st_xprt = xprt;
cv_signal(&st->st_cond);
Remove several linear list traversals per request from RPC server code. Do not insert active ports into pool->sp_active list if they are success- fully assigned to some thread. This makes that list include only ports that really require attention, and so traversal can be reduced to simple taking the first one. Remove idle thread from pool->sp_idlethreads list when assigning some work (port of requests) to it. That again makes possible to replace list traversals with simple taking the first element.
2013-12-20 17:39:07 +00:00
return (TRUE);
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
} else {
/*
* See if we can create a new thread. The
* actual thread creation happens in
* svc_run_internal because our locking state
* is poorly defined (we are typically called
* from a socket upcall). Don't create more
* than one thread per second.
*/
Split RPC pool threads into number of smaller semi-isolated groups. Old design with unified thread pool was good from the point of thread utilization. But single pool-wide mutex became huge congestion point for systems with many CPUs. To reduce the congestion create several thread groups within a pool (one group for every 6 CPUs and 12 threads), each group with own mutex. Each connection during its registration is assigned to one of the groups in round-robin fashion. File affinify code may still move requests between the groups, but otherwise groups are self-contained. MFC after: 2 weeks Sponsored by: iXsystems, Inc.
2014-06-08 11:19:32 +00:00
if (grp->sg_state == SVCPOOL_ACTIVE
&& grp->sg_lastcreatetime < time_uptime
&& grp->sg_threadcount < grp->sg_maxthreads) {
grp->sg_state = SVCPOOL_THREADWANTED;
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
}
}
Remove several linear list traversals per request from RPC server code. Do not insert active ports into pool->sp_active list if they are success- fully assigned to some thread. This makes that list include only ports that really require attention, and so traversal can be reduced to simple taking the first one. Remove idle thread from pool->sp_idlethreads list when assigning some work (port of requests) to it. That again makes possible to replace list traversals with simple taking the first element.
2013-12-20 17:39:07 +00:00
return (FALSE);
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
}
void
xprt_active(SVCXPRT *xprt)
{
Split RPC pool threads into number of smaller semi-isolated groups. Old design with unified thread pool was good from the point of thread utilization. But single pool-wide mutex became huge congestion point for systems with many CPUs. To reduce the congestion create several thread groups within a pool (one group for every 6 CPUs and 12 threads), each group with own mutex. Each connection during its registration is assigned to one of the groups in round-robin fashion. File affinify code may still move requests between the groups, but otherwise groups are self-contained. MFC after: 2 weeks Sponsored by: iXsystems, Inc.
2014-06-08 11:19:32 +00:00
SVCGROUP *grp = xprt->xp_group;
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
Split RPC pool threads into number of smaller semi-isolated groups. Old design with unified thread pool was good from the point of thread utilization. But single pool-wide mutex became huge congestion point for systems with many CPUs. To reduce the congestion create several thread groups within a pool (one group for every 6 CPUs and 12 threads), each group with own mutex. Each connection during its registration is assigned to one of the groups in round-robin fashion. File affinify code may still move requests between the groups, but otherwise groups are self-contained. MFC after: 2 weeks Sponsored by: iXsystems, Inc.
2014-06-08 11:19:32 +00:00
mtx_lock(&grp->sg_lock);
Fix two races in the server side krpc w.r.t upcalls: Add a flag so that soupcall_clear() is only called once to cancel an upcall. Move the test for xprt_registered in the upcall down to after the mtx_lock() of the pool mutex, to catch the case where it is unregistered while the upcall is waiting for the mutex. Also, move the mtx_destroy() of the pool mutex to after SVC_RELEASE(), so that it isn't destroyed before the upcalls are disabled. Reviewed by: dfr, jhb Tested by: pho Approved by: kib (mentor)
2009-06-04 14:13:06 +00:00
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
if (!xprt->xp_registered) {
/*
* Race with xprt_unregister - we lose.
*/
Split RPC pool threads into number of smaller semi-isolated groups. Old design with unified thread pool was good from the point of thread utilization. But single pool-wide mutex became huge congestion point for systems with many CPUs. To reduce the congestion create several thread groups within a pool (one group for every 6 CPUs and 12 threads), each group with own mutex. Each connection during its registration is assigned to one of the groups in round-robin fashion. File affinify code may still move requests between the groups, but otherwise groups are self-contained. MFC after: 2 weeks Sponsored by: iXsystems, Inc.
2014-06-08 11:19:32 +00:00
mtx_unlock(&grp->sg_lock);
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
return;
}
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
if (!xprt->xp_active) {
xprt->xp_active = TRUE;
Remove several linear list traversals per request from RPC server code. Do not insert active ports into pool->sp_active list if they are success- fully assigned to some thread. This makes that list include only ports that really require attention, and so traversal can be reduced to simple taking the first one. Remove idle thread from pool->sp_idlethreads list when assigning some work (port of requests) to it. That again makes possible to replace list traversals with simple taking the first element.
2013-12-20 17:39:07 +00:00
if (xprt->xp_thread == NULL) {
Split RPC pool threads into number of smaller semi-isolated groups. Old design with unified thread pool was good from the point of thread utilization. But single pool-wide mutex became huge congestion point for systems with many CPUs. To reduce the congestion create several thread groups within a pool (one group for every 6 CPUs and 12 threads), each group with own mutex. Each connection during its registration is assigned to one of the groups in round-robin fashion. File affinify code may still move requests between the groups, but otherwise groups are self-contained. MFC after: 2 weeks Sponsored by: iXsystems, Inc.
2014-06-08 11:19:32 +00:00
if (!svc_request_space_available(xprt->xp_pool) ||
Move most of NFS file handle affinity code out of the heavily congested global RPC thread pool lock and protect it with own set of locks. On synthetic benchmarks this improves peak NFS request rate by 40%.
2013-12-30 20:23:15 +00:00
!xprt_assignthread(xprt))
Split RPC pool threads into number of smaller semi-isolated groups. Old design with unified thread pool was good from the point of thread utilization. But single pool-wide mutex became huge congestion point for systems with many CPUs. To reduce the congestion create several thread groups within a pool (one group for every 6 CPUs and 12 threads), each group with own mutex. Each connection during its registration is assigned to one of the groups in round-robin fashion. File affinify code may still move requests between the groups, but otherwise groups are self-contained. MFC after: 2 weeks Sponsored by: iXsystems, Inc.
2014-06-08 11:19:32 +00:00
TAILQ_INSERT_TAIL(&grp->sg_active, xprt,
Remove several linear list traversals per request from RPC server code. Do not insert active ports into pool->sp_active list if they are success- fully assigned to some thread. This makes that list include only ports that really require attention, and so traversal can be reduced to simple taking the first one. Remove idle thread from pool->sp_idlethreads list when assigning some work (port of requests) to it. That again makes possible to replace list traversals with simple taking the first element.
2013-12-20 17:39:07 +00:00
xp_alink);
}
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
}
Split RPC pool threads into number of smaller semi-isolated groups. Old design with unified thread pool was good from the point of thread utilization. But single pool-wide mutex became huge congestion point for systems with many CPUs. To reduce the congestion create several thread groups within a pool (one group for every 6 CPUs and 12 threads), each group with own mutex. Each connection during its registration is assigned to one of the groups in round-robin fashion. File affinify code may still move requests between the groups, but otherwise groups are self-contained. MFC after: 2 weeks Sponsored by: iXsystems, Inc.
2014-06-08 11:19:32 +00:00
mtx_unlock(&grp->sg_lock);
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
}
void
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
xprt_inactive_locked(SVCXPRT *xprt)
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
{
Split RPC pool threads into number of smaller semi-isolated groups. Old design with unified thread pool was good from the point of thread utilization. But single pool-wide mutex became huge congestion point for systems with many CPUs. To reduce the congestion create several thread groups within a pool (one group for every 6 CPUs and 12 threads), each group with own mutex. Each connection during its registration is assigned to one of the groups in round-robin fashion. File affinify code may still move requests between the groups, but otherwise groups are self-contained. MFC after: 2 weeks Sponsored by: iXsystems, Inc.
2014-06-08 11:19:32 +00:00
SVCGROUP *grp = xprt->xp_group;
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
Split RPC pool threads into number of smaller semi-isolated groups. Old design with unified thread pool was good from the point of thread utilization. But single pool-wide mutex became huge congestion point for systems with many CPUs. To reduce the congestion create several thread groups within a pool (one group for every 6 CPUs and 12 threads), each group with own mutex. Each connection during its registration is assigned to one of the groups in round-robin fashion. File affinify code may still move requests between the groups, but otherwise groups are self-contained. MFC after: 2 weeks Sponsored by: iXsystems, Inc.
2014-06-08 11:19:32 +00:00
mtx_assert(&grp->sg_lock, MA_OWNED);
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
if (xprt->xp_active) {
Remove several linear list traversals per request from RPC server code. Do not insert active ports into pool->sp_active list if they are success- fully assigned to some thread. This makes that list include only ports that really require attention, and so traversal can be reduced to simple taking the first one. Remove idle thread from pool->sp_idlethreads list when assigning some work (port of requests) to it. That again makes possible to replace list traversals with simple taking the first element.
2013-12-20 17:39:07 +00:00
if (xprt->xp_thread == NULL)
Split RPC pool threads into number of smaller semi-isolated groups. Old design with unified thread pool was good from the point of thread utilization. But single pool-wide mutex became huge congestion point for systems with many CPUs. To reduce the congestion create several thread groups within a pool (one group for every 6 CPUs and 12 threads), each group with own mutex. Each connection during its registration is assigned to one of the groups in round-robin fashion. File affinify code may still move requests between the groups, but otherwise groups are self-contained. MFC after: 2 weeks Sponsored by: iXsystems, Inc.
2014-06-08 11:19:32 +00:00
TAILQ_REMOVE(&grp->sg_active, xprt, xp_alink);
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
xprt->xp_active = FALSE;
}
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
}
void
xprt_inactive(SVCXPRT *xprt)
{
Split RPC pool threads into number of smaller semi-isolated groups. Old design with unified thread pool was good from the point of thread utilization. But single pool-wide mutex became huge congestion point for systems with many CPUs. To reduce the congestion create several thread groups within a pool (one group for every 6 CPUs and 12 threads), each group with own mutex. Each connection during its registration is assigned to one of the groups in round-robin fashion. File affinify code may still move requests between the groups, but otherwise groups are self-contained. MFC after: 2 weeks Sponsored by: iXsystems, Inc.
2014-06-08 11:19:32 +00:00
SVCGROUP *grp = xprt->xp_group;
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
Split RPC pool threads into number of smaller semi-isolated groups. Old design with unified thread pool was good from the point of thread utilization. But single pool-wide mutex became huge congestion point for systems with many CPUs. To reduce the congestion create several thread groups within a pool (one group for every 6 CPUs and 12 threads), each group with own mutex. Each connection during its registration is assigned to one of the groups in round-robin fashion. File affinify code may still move requests between the groups, but otherwise groups are self-contained. MFC after: 2 weeks Sponsored by: iXsystems, Inc.
2014-06-08 11:19:32 +00:00
mtx_lock(&grp->sg_lock);
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
xprt_inactive_locked(xprt);
Split RPC pool threads into number of smaller semi-isolated groups. Old design with unified thread pool was good from the point of thread utilization. But single pool-wide mutex became huge congestion point for systems with many CPUs. To reduce the congestion create several thread groups within a pool (one group for every 6 CPUs and 12 threads), each group with own mutex. Each connection during its registration is assigned to one of the groups in round-robin fashion. File affinify code may still move requests between the groups, but otherwise groups are self-contained. MFC after: 2 weeks Sponsored by: iXsystems, Inc.
2014-06-08 11:19:32 +00:00
mtx_unlock(&grp->sg_lock);
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
}
Introduce xprt_inactive_self() -- variant for use when sure that port is assigned to thread. For example, withing receive handlers. In that case the function reduces to single assignment and can avoid locking.
2013-12-29 11:19:09 +00:00
/*
* Variant of xprt_inactive() for use only when sure that port is
sys/rpc: minor spelling fixes. No functional change.
2016-05-06 01:49:46 +00:00
* assigned to thread. For example, within receive handlers.
Introduce xprt_inactive_self() -- variant for use when sure that port is assigned to thread. For example, withing receive handlers. In that case the function reduces to single assignment and can avoid locking.
2013-12-29 11:19:09 +00:00
*/
void
xprt_inactive_self(SVCXPRT *xprt)
{
KASSERT(xprt->xp_thread != NULL,
("xprt_inactive_self(%p) with NULL xp_thread", xprt));
xprt->xp_active = FALSE;
}
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
/*
* Add a service program to the callout list.
* The dispatch routine will be called when a rpc request for this
* program number comes in.
*/
bool_t
svc_reg(SVCXPRT *xprt, const rpcprog_t prog, const rpcvers_t vers,
void (*dispatch)(struct svc_req *, SVCXPRT *),
const struct netconfig *nconf)
{
SVCPOOL *pool = xprt->xp_pool;
struct svc_callout *s;
char *netid = NULL;
int flag = 0;
/* VARIABLES PROTECTED BY svc_lock: s, svc_head */
if (xprt->xp_netid) {
netid = strdup(xprt->xp_netid, M_RPC);
flag = 1;
} else if (nconf && nconf->nc_netid) {
netid = strdup(nconf->nc_netid, M_RPC);
flag = 1;
} /* must have been created with svc_raw_create */
if ((netid == NULL) && (flag == 1)) {
return (FALSE);
}
mtx_lock(&pool->sp_lock);
if ((s = svc_find(pool, prog, vers, netid)) != NULL) {
if (netid)
free(netid, M_RPC);
if (s->sc_dispatch == dispatch)
goto rpcb_it; /* he is registering another xptr */
mtx_unlock(&pool->sp_lock);
return (FALSE);
}
s = malloc(sizeof (struct svc_callout), M_RPC, M_NOWAIT);
if (s == NULL) {
if (netid)
free(netid, M_RPC);
mtx_unlock(&pool->sp_lock);
return (FALSE);
}
s->sc_prog = prog;
s->sc_vers = vers;
s->sc_dispatch = dispatch;
s->sc_netid = netid;
TAILQ_INSERT_TAIL(&pool->sp_callouts, s, sc_link);
if ((xprt->xp_netid == NULL) && (flag == 1) && netid)
((SVCXPRT *) xprt)->xp_netid = strdup(netid, M_RPC);
rpcb_it:
mtx_unlock(&pool->sp_lock);
/* now register the information with the local binder service */
if (nconf) {
bool_t dummy;
struct netconfig tnc;
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
struct netbuf nb;
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
tnc = *nconf;
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
nb.buf = &xprt->xp_ltaddr;
nb.len = xprt->xp_ltaddr.ss_len;
dummy = rpcb_set(prog, vers, &tnc, &nb);
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
return (dummy);
}
return (TRUE);
}
/*
* Remove a service program from the callout list.
*/
void
svc_unreg(SVCPOOL *pool, const rpcprog_t prog, const rpcvers_t vers)
{
struct svc_callout *s;
/* unregister the information anyway */
(void) rpcb_unset(prog, vers, NULL);
mtx_lock(&pool->sp_lock);
while ((s = svc_find(pool, prog, vers, NULL)) != NULL) {
TAILQ_REMOVE(&pool->sp_callouts, s, sc_link);
if (s->sc_netid)
mem_free(s->sc_netid, sizeof (s->sc_netid) + 1);
mem_free(s, sizeof (struct svc_callout));
}
mtx_unlock(&pool->sp_lock);
}
Rework NFS Duplicate Request Cache cleanup logic. - Introduce additional hash to group requests by hash of sockref. This allows to process TCP acknowledgements without looping though all the cache, and as result allows to do it every time. - Indroduce additional callbacks to notify application layer about sockets disconnection. Without this last few requests processed just before socket disconnection never processed their ACKs and stuck in cache for many hours. - Implement transport-specific method for tracking reply acknowledgements. New implementation does not cross multiple stack layers to get the data and does not have race conditions that previously made some requests stuck in cache. This could be done more efficiently at sockbuf layer, but that would broke some KBIs, while I don't know other consumers for it aside NFS. - Instead of traversing all DRC twice per request, run cleaning only once per request, and except in some conditions traverse only single hash slot at a time. Together this limits NFS DRC growth only to situations of real connectivity problems. If network is working well, and so all replies are acknowledged, cache remains almost empty even after hours of heavy load. Without this change on the same test cache was growing to many thousand requests even with perfectly working local network. As another result this reduces CPU time spent on the DRC handling during SPEC NFS benchmark from about 10% to 0.5%. Sponsored by: iXsystems, Inc.
2014-01-03 15:09:59 +00:00
/*
* Add a service connection loss program to the callout list.
* The dispatch routine will be called when some port in ths pool die.
*/
bool_t
svc_loss_reg(SVCXPRT *xprt, void (*dispatch)(SVCXPRT *))
{
SVCPOOL *pool = xprt->xp_pool;
struct svc_loss_callout *s;
mtx_lock(&pool->sp_lock);
TAILQ_FOREACH(s, &pool->sp_lcallouts, slc_link) {
if (s->slc_dispatch == dispatch)
break;
}
if (s != NULL) {
mtx_unlock(&pool->sp_lock);
return (TRUE);
}
Fix incorrect (fortunately bigger) malloc size. Submitted by: pfg MFC after: 1 week
2016-03-19 11:48:06 +00:00
s = malloc(sizeof(struct svc_loss_callout), M_RPC, M_NOWAIT);
Rework NFS Duplicate Request Cache cleanup logic. - Introduce additional hash to group requests by hash of sockref. This allows to process TCP acknowledgements without looping though all the cache, and as result allows to do it every time. - Indroduce additional callbacks to notify application layer about sockets disconnection. Without this last few requests processed just before socket disconnection never processed their ACKs and stuck in cache for many hours. - Implement transport-specific method for tracking reply acknowledgements. New implementation does not cross multiple stack layers to get the data and does not have race conditions that previously made some requests stuck in cache. This could be done more efficiently at sockbuf layer, but that would broke some KBIs, while I don't know other consumers for it aside NFS. - Instead of traversing all DRC twice per request, run cleaning only once per request, and except in some conditions traverse only single hash slot at a time. Together this limits NFS DRC growth only to situations of real connectivity problems. If network is working well, and so all replies are acknowledged, cache remains almost empty even after hours of heavy load. Without this change on the same test cache was growing to many thousand requests even with perfectly working local network. As another result this reduces CPU time spent on the DRC handling during SPEC NFS benchmark from about 10% to 0.5%. Sponsored by: iXsystems, Inc.
2014-01-03 15:09:59 +00:00
if (s == NULL) {
mtx_unlock(&pool->sp_lock);
return (FALSE);
}
s->slc_dispatch = dispatch;
TAILQ_INSERT_TAIL(&pool->sp_lcallouts, s, slc_link);
mtx_unlock(&pool->sp_lock);
return (TRUE);
}
/*
* Remove a service connection loss program from the callout list.
*/
void
svc_loss_unreg(SVCPOOL *pool, void (*dispatch)(SVCXPRT *))
{
struct svc_loss_callout *s;
mtx_lock(&pool->sp_lock);
TAILQ_FOREACH(s, &pool->sp_lcallouts, slc_link) {
if (s->slc_dispatch == dispatch) {
TAILQ_REMOVE(&pool->sp_lcallouts, s, slc_link);
free(s, M_RPC);
break;
}
}
mtx_unlock(&pool->sp_lock);
}
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
/* ********************** CALLOUT list related stuff ************* */
/*
* Search the callout list for a program number, return the callout
* struct.
*/
static struct svc_callout *
svc_find(SVCPOOL *pool, rpcprog_t prog, rpcvers_t vers, char *netid)
{
struct svc_callout *s;
mtx_assert(&pool->sp_lock, MA_OWNED);
TAILQ_FOREACH(s, &pool->sp_callouts, sc_link) {
if (s->sc_prog == prog && s->sc_vers == vers
&& (netid == NULL || s->sc_netid == NULL ||
strcmp(netid, s->sc_netid) == 0))
break;
}
return (s);
}
/* ******************* REPLY GENERATION ROUTINES ************ */
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
static bool_t
svc_sendreply_common(struct svc_req *rqstp, struct rpc_msg *rply,
struct mbuf *body)
{
SVCXPRT *xprt = rqstp->rq_xprt;
bool_t ok;
if (rqstp->rq_args) {
m_freem(rqstp->rq_args);
rqstp->rq_args = NULL;
}
if (xprt->xp_pool->sp_rcache)
replay_setreply(xprt->xp_pool->sp_rcache,
rply, svc_getrpccaller(rqstp), body);
if (!SVCAUTH_WRAP(&rqstp->rq_auth, &body))
return (FALSE);
Rework NFS Duplicate Request Cache cleanup logic. - Introduce additional hash to group requests by hash of sockref. This allows to process TCP acknowledgements without looping though all the cache, and as result allows to do it every time. - Indroduce additional callbacks to notify application layer about sockets disconnection. Without this last few requests processed just before socket disconnection never processed their ACKs and stuck in cache for many hours. - Implement transport-specific method for tracking reply acknowledgements. New implementation does not cross multiple stack layers to get the data and does not have race conditions that previously made some requests stuck in cache. This could be done more efficiently at sockbuf layer, but that would broke some KBIs, while I don't know other consumers for it aside NFS. - Instead of traversing all DRC twice per request, run cleaning only once per request, and except in some conditions traverse only single hash slot at a time. Together this limits NFS DRC growth only to situations of real connectivity problems. If network is working well, and so all replies are acknowledged, cache remains almost empty even after hours of heavy load. Without this change on the same test cache was growing to many thousand requests even with perfectly working local network. As another result this reduces CPU time spent on the DRC handling during SPEC NFS benchmark from about 10% to 0.5%. Sponsored by: iXsystems, Inc.
2014-01-03 15:09:59 +00:00
ok = SVC_REPLY(xprt, rply, rqstp->rq_addr, body, &rqstp->rq_reply_seq);
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
if (rqstp->rq_addr) {
free(rqstp->rq_addr, M_SONAME);
rqstp->rq_addr = NULL;
}
return (ok);
}
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
/*
* Send a reply to an rpc request
*/
bool_t
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
svc_sendreply(struct svc_req *rqstp, xdrproc_t xdr_results, void * xdr_location)
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
{
struct rpc_msg rply;
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
struct mbuf *m;
XDR xdrs;
bool_t ok;
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
rply.rm_xid = rqstp->rq_xid;
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
rply.rm_direction = REPLY;
rply.rm_reply.rp_stat = MSG_ACCEPTED;
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
rply.acpted_rply.ar_verf = rqstp->rq_verf;
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
rply.acpted_rply.ar_stat = SUCCESS;
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
rply.acpted_rply.ar_results.where = NULL;
rply.acpted_rply.ar_results.proc = (xdrproc_t) xdr_void;
Use m_get(), m_gethdr() and m_getcl() instead of historic macros. Sponsored by: Nginx, Inc.
2013-03-12 12:17:19 +00:00
m = m_getcl(M_WAITOK, MT_DATA, 0);
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
xdrmbuf_create(&xdrs, m, XDR_ENCODE);
ok = xdr_results(&xdrs, xdr_location);
XDR_DESTROY(&xdrs);
if (ok) {
return (svc_sendreply_common(rqstp, &rply, m));
} else {
m_freem(m);
return (FALSE);
}
}
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
bool_t
svc_sendreply_mbuf(struct svc_req *rqstp, struct mbuf *m)
{
struct rpc_msg rply;
rply.rm_xid = rqstp->rq_xid;
rply.rm_direction = REPLY;
rply.rm_reply.rp_stat = MSG_ACCEPTED;
rply.acpted_rply.ar_verf = rqstp->rq_verf;
rply.acpted_rply.ar_stat = SUCCESS;
rply.acpted_rply.ar_results.where = NULL;
rply.acpted_rply.ar_results.proc = (xdrproc_t) xdr_void;
return (svc_sendreply_common(rqstp, &rply, m));
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
}
/*
* No procedure error reply
*/
void
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
svcerr_noproc(struct svc_req *rqstp)
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
{
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
SVCXPRT *xprt = rqstp->rq_xprt;
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
struct rpc_msg rply;
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
rply.rm_xid = rqstp->rq_xid;
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
rply.rm_direction = REPLY;
rply.rm_reply.rp_stat = MSG_ACCEPTED;
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
rply.acpted_rply.ar_verf = rqstp->rq_verf;
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
rply.acpted_rply.ar_stat = PROC_UNAVAIL;
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
if (xprt->xp_pool->sp_rcache)
replay_setreply(xprt->xp_pool->sp_rcache,
&rply, svc_getrpccaller(rqstp), NULL);
svc_sendreply_common(rqstp, &rply, NULL);
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
}
/*
* Can't decode args error reply
*/
void
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
svcerr_decode(struct svc_req *rqstp)
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
{
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
SVCXPRT *xprt = rqstp->rq_xprt;
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
struct rpc_msg rply;
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
rply.rm_xid = rqstp->rq_xid;
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
rply.rm_direction = REPLY;
rply.rm_reply.rp_stat = MSG_ACCEPTED;
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
rply.acpted_rply.ar_verf = rqstp->rq_verf;
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
rply.acpted_rply.ar_stat = GARBAGE_ARGS;
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
if (xprt->xp_pool->sp_rcache)
replay_setreply(xprt->xp_pool->sp_rcache,
&rply, (struct sockaddr *) &xprt->xp_rtaddr, NULL);
svc_sendreply_common(rqstp, &rply, NULL);
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
}
/*
* Some system error
*/
void
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
svcerr_systemerr(struct svc_req *rqstp)
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
{
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
SVCXPRT *xprt = rqstp->rq_xprt;
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
struct rpc_msg rply;
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
rply.rm_xid = rqstp->rq_xid;
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
rply.rm_direction = REPLY;
rply.rm_reply.rp_stat = MSG_ACCEPTED;
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
rply.acpted_rply.ar_verf = rqstp->rq_verf;
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
rply.acpted_rply.ar_stat = SYSTEM_ERR;
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
if (xprt->xp_pool->sp_rcache)
replay_setreply(xprt->xp_pool->sp_rcache,
&rply, svc_getrpccaller(rqstp), NULL);
svc_sendreply_common(rqstp, &rply, NULL);
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
}
/*
* Authentication error reply
*/
void
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
svcerr_auth(struct svc_req *rqstp, enum auth_stat why)
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
{
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
SVCXPRT *xprt = rqstp->rq_xprt;
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
struct rpc_msg rply;
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
rply.rm_xid = rqstp->rq_xid;
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
rply.rm_direction = REPLY;
rply.rm_reply.rp_stat = MSG_DENIED;
rply.rjcted_rply.rj_stat = AUTH_ERROR;
rply.rjcted_rply.rj_why = why;
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
if (xprt->xp_pool->sp_rcache)
replay_setreply(xprt->xp_pool->sp_rcache,
&rply, svc_getrpccaller(rqstp), NULL);
svc_sendreply_common(rqstp, &rply, NULL);
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
}
/*
* Auth too weak error reply
*/
void
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
svcerr_weakauth(struct svc_req *rqstp)
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
{
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
svcerr_auth(rqstp, AUTH_TOOWEAK);
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
}
/*
* Program unavailable error reply
*/
void
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
svcerr_noprog(struct svc_req *rqstp)
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
{
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
SVCXPRT *xprt = rqstp->rq_xprt;
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
struct rpc_msg rply;
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
rply.rm_xid = rqstp->rq_xid;
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
rply.rm_direction = REPLY;
rply.rm_reply.rp_stat = MSG_ACCEPTED;
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
rply.acpted_rply.ar_verf = rqstp->rq_verf;
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
rply.acpted_rply.ar_stat = PROG_UNAVAIL;
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
if (xprt->xp_pool->sp_rcache)
replay_setreply(xprt->xp_pool->sp_rcache,
&rply, svc_getrpccaller(rqstp), NULL);
svc_sendreply_common(rqstp, &rply, NULL);
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
}
/*
* Program version mismatch error reply
*/
void
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
svcerr_progvers(struct svc_req *rqstp, rpcvers_t low_vers, rpcvers_t high_vers)
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
{
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
SVCXPRT *xprt = rqstp->rq_xprt;
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
struct rpc_msg rply;
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
rply.rm_xid = rqstp->rq_xid;
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
rply.rm_direction = REPLY;
rply.rm_reply.rp_stat = MSG_ACCEPTED;
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
rply.acpted_rply.ar_verf = rqstp->rq_verf;
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
rply.acpted_rply.ar_stat = PROG_MISMATCH;
rply.acpted_rply.ar_vers.low = (uint32_t)low_vers;
rply.acpted_rply.ar_vers.high = (uint32_t)high_vers;
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
if (xprt->xp_pool->sp_rcache)
replay_setreply(xprt->xp_pool->sp_rcache,
&rply, svc_getrpccaller(rqstp), NULL);
svc_sendreply_common(rqstp, &rply, NULL);
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
}
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
/*
* Allocate a new server transport structure. All fields are
* initialized to zero and xp_p3 is initialized to point at an
* extension structure to hold various flags and authentication
* parameters.
*/
SVCXPRT *
Convert `svc_xprt_alloc(..)` and `svc_xprt_free(..)`'s prototypes to ANSI C style prototypes MFC after: 1 week Sponsored by: EMC / Isilon Storage Division
2016-07-11 07:17:52 +00:00
svc_xprt_alloc(void)
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
{
SVCXPRT *xprt;
SVCXPRT_EXT *ext;
xprt = mem_alloc(sizeof(SVCXPRT));
ext = mem_alloc(sizeof(SVCXPRT_EXT));
xprt->xp_p3 = ext;
refcount_init(&xprt->xp_refs, 1);
return (xprt);
}
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
/*
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
* Free a server transport structure.
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
*/
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
void
Convert `svc_xprt_alloc(..)` and `svc_xprt_free(..)`'s prototypes to ANSI C style prototypes MFC after: 1 week Sponsored by: EMC / Isilon Storage Division
2016-07-11 07:17:52 +00:00
svc_xprt_free(SVCXPRT *xprt)
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
{
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
mem_free(xprt->xp_p3, sizeof(SVCXPRT_EXT));
Fix a potential memory leak in the NFS over TLS handling code. For the TLS case where there is a "user@domain" name specified in the X.509 v3 certificate presented by the client in the otherName component of subjectAltName, a gid list is allocated via mem_alloc(). This needs to be free'd. Otherwise xp_gidp == NULL and free() handles that. (The size argument to mem_free() is not used by FreeBSD, so it can be 0.) This leak would not have occurred for any other case than NFS over TLS with the "user@domain" in the client's certificate.
2020-09-05 00:50:52 +00:00
/* The size argument is ignored, so 0 is ok. */
mem_free(xprt->xp_gidp, 0);
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
mem_free(xprt, sizeof(SVCXPRT));
}
/* ******************* SERVER INPUT STUFF ******************* */
/*
* Read RPC requests from a transport and queue them to be
* executed. We handle authentication and replay cache replies here.
* Actually dispatching the RPC is deferred till svc_executereq.
*/
static enum xprt_stat
svc_getreq(SVCXPRT *xprt, struct svc_req **rqstp_ret)
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
{
SVCPOOL *pool = xprt->xp_pool;
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
struct svc_req *r;
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
struct rpc_msg msg;
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
struct mbuf *args;
Rework NFS Duplicate Request Cache cleanup logic. - Introduce additional hash to group requests by hash of sockref. This allows to process TCP acknowledgements without looping though all the cache, and as result allows to do it every time. - Indroduce additional callbacks to notify application layer about sockets disconnection. Without this last few requests processed just before socket disconnection never processed their ACKs and stuck in cache for many hours. - Implement transport-specific method for tracking reply acknowledgements. New implementation does not cross multiple stack layers to get the data and does not have race conditions that previously made some requests stuck in cache. This could be done more efficiently at sockbuf layer, but that would broke some KBIs, while I don't know other consumers for it aside NFS. - Instead of traversing all DRC twice per request, run cleaning only once per request, and except in some conditions traverse only single hash slot at a time. Together this limits NFS DRC growth only to situations of real connectivity problems. If network is working well, and so all replies are acknowledged, cache remains almost empty even after hours of heavy load. Without this change on the same test cache was growing to many thousand requests even with perfectly working local network. As another result this reduces CPU time spent on the DRC handling during SPEC NFS benchmark from about 10% to 0.5%. Sponsored by: iXsystems, Inc.
2014-01-03 15:09:59 +00:00
struct svc_loss_callout *s;
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
enum xprt_stat stat;
/* now receive msgs from xprtprt (support batch calls) */
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
r = malloc(sizeof(*r), M_RPC, M_WAITOK|M_ZERO);
msg.rm_call.cb_cred.oa_base = r->rq_credarea;
msg.rm_call.cb_verf.oa_base = &r->rq_credarea[MAX_AUTH_BYTES];
r->rq_clntcred = &r->rq_credarea[2*MAX_AUTH_BYTES];
if (SVC_RECV(xprt, &msg, &r->rq_addr, &args)) {
enum auth_stat why;
/*
* Handle replays and authenticate before queuing the
* request to be executed.
*/
SVC_ACQUIRE(xprt);
r->rq_xprt = xprt;
if (pool->sp_rcache) {
struct rpc_msg repmsg;
struct mbuf *repbody;
enum replay_state rs;
rs = replay_find(pool->sp_rcache, &msg,
svc_getrpccaller(r), &repmsg, &repbody);
switch (rs) {
case RS_NEW:
break;
case RS_DONE:
SVC_REPLY(xprt, &repmsg, r->rq_addr,
Rework NFS Duplicate Request Cache cleanup logic. - Introduce additional hash to group requests by hash of sockref. This allows to process TCP acknowledgements without looping though all the cache, and as result allows to do it every time. - Indroduce additional callbacks to notify application layer about sockets disconnection. Without this last few requests processed just before socket disconnection never processed their ACKs and stuck in cache for many hours. - Implement transport-specific method for tracking reply acknowledgements. New implementation does not cross multiple stack layers to get the data and does not have race conditions that previously made some requests stuck in cache. This could be done more efficiently at sockbuf layer, but that would broke some KBIs, while I don't know other consumers for it aside NFS. - Instead of traversing all DRC twice per request, run cleaning only once per request, and except in some conditions traverse only single hash slot at a time. Together this limits NFS DRC growth only to situations of real connectivity problems. If network is working well, and so all replies are acknowledged, cache remains almost empty even after hours of heavy load. Without this change on the same test cache was growing to many thousand requests even with perfectly working local network. As another result this reduces CPU time spent on the DRC handling during SPEC NFS benchmark from about 10% to 0.5%. Sponsored by: iXsystems, Inc.
2014-01-03 15:09:59 +00:00
repbody, &r->rq_reply_seq);
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
if (r->rq_addr) {
free(r->rq_addr, M_SONAME);
r->rq_addr = NULL;
}
When the regular NFS server replied to a UDP client out of the replay cache, it did not free the request argument mbuf list, resulting in a leak. This patch fixes that leak. Tested by: danny AT cs.huji.ac.il PR: kern/144330 Submitted by: to.my.trociny AT gmail.com (earlier version) Reviewed by: dfr MFC after: 2 weeks
2010-03-23 23:03:30 +00:00
m_freem(args);
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
goto call_done;
default:
When the regular NFS server replied to a UDP client out of the replay cache, it did not free the request argument mbuf list, resulting in a leak. This patch fixes that leak. Tested by: danny AT cs.huji.ac.il PR: kern/144330 Submitted by: to.my.trociny AT gmail.com (earlier version) Reviewed by: dfr MFC after: 2 weeks
2010-03-23 23:03:30 +00:00
m_freem(args);
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
goto call_done;
}
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
}
r->rq_xid = msg.rm_xid;
r->rq_prog = msg.rm_call.cb_prog;
r->rq_vers = msg.rm_call.cb_vers;
r->rq_proc = msg.rm_call.cb_proc;
r->rq_size = sizeof(*r) + m_length(args, NULL);
r->rq_args = args;
if ((why = _authenticate(r, &msg)) != AUTH_OK) {
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
/*
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
* RPCSEC_GSS uses this return code
* for requests that form part of its
* context establishment protocol and
* should not be dispatched to the
* application.
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
*/
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
if (why != RPCSEC_GSS_NODISPATCH)
svcerr_auth(r, why);
goto call_done;
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
}
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
if (!SVCAUTH_UNWRAP(&r->rq_auth, &r->rq_args)) {
svcerr_decode(r);
goto call_done;
}
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
/*
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
* Everything checks out, return request to caller.
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
*/
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
*rqstp_ret = r;
r = NULL;
}
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
call_done:
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
if (r) {
svc_freereq(r);
r = NULL;
}
if ((stat = SVC_STAT(xprt)) == XPRT_DIED) {
Rework NFS Duplicate Request Cache cleanup logic. - Introduce additional hash to group requests by hash of sockref. This allows to process TCP acknowledgements without looping though all the cache, and as result allows to do it every time. - Indroduce additional callbacks to notify application layer about sockets disconnection. Without this last few requests processed just before socket disconnection never processed their ACKs and stuck in cache for many hours. - Implement transport-specific method for tracking reply acknowledgements. New implementation does not cross multiple stack layers to get the data and does not have race conditions that previously made some requests stuck in cache. This could be done more efficiently at sockbuf layer, but that would broke some KBIs, while I don't know other consumers for it aside NFS. - Instead of traversing all DRC twice per request, run cleaning only once per request, and except in some conditions traverse only single hash slot at a time. Together this limits NFS DRC growth only to situations of real connectivity problems. If network is working well, and so all replies are acknowledged, cache remains almost empty even after hours of heavy load. Without this change on the same test cache was growing to many thousand requests even with perfectly working local network. As another result this reduces CPU time spent on the DRC handling during SPEC NFS benchmark from about 10% to 0.5%. Sponsored by: iXsystems, Inc.
2014-01-03 15:09:59 +00:00
TAILQ_FOREACH(s, &pool->sp_lcallouts, slc_link)
(*s->slc_dispatch)(xprt);
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
xprt_unregister(xprt);
}
return (stat);
}
static void
svc_executereq(struct svc_req *rqstp)
{
SVCXPRT *xprt = rqstp->rq_xprt;
SVCPOOL *pool = xprt->xp_pool;
int prog_found;
rpcvers_t low_vers;
rpcvers_t high_vers;
struct svc_callout *s;
/* now match message with a registered service*/
prog_found = FALSE;
low_vers = (rpcvers_t) -1L;
high_vers = (rpcvers_t) 0L;
TAILQ_FOREACH(s, &pool->sp_callouts, sc_link) {
if (s->sc_prog == rqstp->rq_prog) {
if (s->sc_vers == rqstp->rq_vers) {
/*
* We hand ownership of r to the
* dispatch method - they must call
* svc_freereq.
*/
(*s->sc_dispatch)(rqstp, xprt);
return;
} /* found correct version */
prog_found = TRUE;
if (s->sc_vers < low_vers)
low_vers = s->sc_vers;
if (s->sc_vers > high_vers)
high_vers = s->sc_vers;
} /* found correct program */
}
/*
* if we got here, the program or version
* is not served ...
*/
if (prog_found)
svcerr_progvers(rqstp, low_vers, high_vers);
else
svcerr_noprog(rqstp);
svc_freereq(rqstp);
}
static void
Split RPC pool threads into number of smaller semi-isolated groups. Old design with unified thread pool was good from the point of thread utilization. But single pool-wide mutex became huge congestion point for systems with many CPUs. To reduce the congestion create several thread groups within a pool (one group for every 6 CPUs and 12 threads), each group with own mutex. Each connection during its registration is assigned to one of the groups in round-robin fashion. File affinify code may still move requests between the groups, but otherwise groups are self-contained. MFC after: 2 weeks Sponsored by: iXsystems, Inc.
2014-06-08 11:19:32 +00:00
svc_checkidle(SVCGROUP *grp)
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
{
SVCXPRT *xprt, *nxprt;
time_t timo;
struct svcxprt_list cleanup;
TAILQ_INIT(&cleanup);
Split RPC pool threads into number of smaller semi-isolated groups. Old design with unified thread pool was good from the point of thread utilization. But single pool-wide mutex became huge congestion point for systems with many CPUs. To reduce the congestion create several thread groups within a pool (one group for every 6 CPUs and 12 threads), each group with own mutex. Each connection during its registration is assigned to one of the groups in round-robin fashion. File affinify code may still move requests between the groups, but otherwise groups are self-contained. MFC after: 2 weeks Sponsored by: iXsystems, Inc.
2014-06-08 11:19:32 +00:00
TAILQ_FOREACH_SAFE(xprt, &grp->sg_xlist, xp_link, nxprt) {
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
/*
* Only some transports have idle timers. Don't time
* something out which is just waking up.
*/
if (!xprt->xp_idletimeout || xprt->xp_thread)
continue;
timo = xprt->xp_lastactive + xprt->xp_idletimeout;
if (time_uptime > timo) {
xprt_unregister_locked(xprt);
TAILQ_INSERT_TAIL(&cleanup, xprt, xp_link);
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
}
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
}
Split RPC pool threads into number of smaller semi-isolated groups. Old design with unified thread pool was good from the point of thread utilization. But single pool-wide mutex became huge congestion point for systems with many CPUs. To reduce the congestion create several thread groups within a pool (one group for every 6 CPUs and 12 threads), each group with own mutex. Each connection during its registration is assigned to one of the groups in round-robin fashion. File affinify code may still move requests between the groups, but otherwise groups are self-contained. MFC after: 2 weeks Sponsored by: iXsystems, Inc.
2014-06-08 11:19:32 +00:00
mtx_unlock(&grp->sg_lock);
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
TAILQ_FOREACH_SAFE(xprt, &cleanup, xp_link, nxprt) {
nfsd: fix a NFSv4.1 Linux client mount stuck in CLOSE_WAIT It was reported that a NFSv4.1 Linux client mount against a FreeBSD12 server was hung, with the TCP connection in CLOSE_WAIT state on the server. When a NFSv4.1/4.2 mount is done and the back channel is bound to the TCP connection, the soclose() is delayed until a new TCP connection is bound to the back channel, due to a reference count being held on the SVCXPRT structure in the krpc for the socket. Without the soclose() call, the socket will remain in CLOSE_WAIT and this somehow caused the Linux client to hang. This patch adds calls to soshutdown(.., SHUT_WR) that are performed when the server side krpc sees that the socket is no longer usable. Since this can be done before the back channel is bound to a new TCP connection, it allows the TCP connection to proceed to CLOSED state. PR: 254590 Reported by: jbreitman@tildenparkcapital.com Reviewed by: tuexen Comments by: kevans MFC after: 2 weeks Differential Revision: https://reviews.freebsd.org/D29526
2021-04-27 22:32:35 +00:00
soshutdown(xprt->xp_socket, SHUT_WR);
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
SVC_RELEASE(xprt);
}
Split RPC pool threads into number of smaller semi-isolated groups. Old design with unified thread pool was good from the point of thread utilization. But single pool-wide mutex became huge congestion point for systems with many CPUs. To reduce the congestion create several thread groups within a pool (one group for every 6 CPUs and 12 threads), each group with own mutex. Each connection during its registration is assigned to one of the groups in round-robin fashion. File affinify code may still move requests between the groups, but otherwise groups are self-contained. MFC after: 2 weeks Sponsored by: iXsystems, Inc.
2014-06-08 11:19:32 +00:00
mtx_lock(&grp->sg_lock);
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
}
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
static void
svc_assign_waiting_sockets(SVCPOOL *pool)
{
Split RPC pool threads into number of smaller semi-isolated groups. Old design with unified thread pool was good from the point of thread utilization. But single pool-wide mutex became huge congestion point for systems with many CPUs. To reduce the congestion create several thread groups within a pool (one group for every 6 CPUs and 12 threads), each group with own mutex. Each connection during its registration is assigned to one of the groups in round-robin fashion. File affinify code may still move requests between the groups, but otherwise groups are self-contained. MFC after: 2 weeks Sponsored by: iXsystems, Inc.
2014-06-08 11:19:32 +00:00
SVCGROUP *grp;
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
SVCXPRT *xprt;
Split RPC pool threads into number of smaller semi-isolated groups. Old design with unified thread pool was good from the point of thread utilization. But single pool-wide mutex became huge congestion point for systems with many CPUs. To reduce the congestion create several thread groups within a pool (one group for every 6 CPUs and 12 threads), each group with own mutex. Each connection during its registration is assigned to one of the groups in round-robin fashion. File affinify code may still move requests between the groups, but otherwise groups are self-contained. MFC after: 2 weeks Sponsored by: iXsystems, Inc.
2014-06-08 11:19:32 +00:00
int g;
for (g = 0; g < pool->sp_groupcount; g++) {
grp = &pool->sp_groups[g];
mtx_lock(&grp->sg_lock);
while ((xprt = TAILQ_FIRST(&grp->sg_active)) != NULL) {
if (xprt_assignthread(xprt))
TAILQ_REMOVE(&grp->sg_active, xprt, xp_alink);
else
break;
}
mtx_unlock(&grp->sg_lock);
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
}
}
Move most of NFS file handle affinity code out of the heavily congested global RPC thread pool lock and protect it with own set of locks. On synthetic benchmarks this improves peak NFS request rate by 40%.
2013-12-30 20:23:15 +00:00
static void
Fix overflow bugs in and remove obsolete limit from kernel RPC implementation. The kernel RPC code, which is responsible for the low-level scheduling of incoming NFS requests, contains a throttling mechanism that prevents too much kernel memory from being tied up by NFS requests that are being serviced. When the throttle is engaged, the RPC layer stops servicing incoming NFS sockets, resulting ultimately in backpressure on the clients (if they're using TCP). However, this is a very heavy-handed mechanism as it prevents all clients from making any requests, regardless of how heavy or light they are. (Thus, when engaged, the throttle often prevents clients from even mounting the filesystem.) The throttle mechanism applies specifically to requests that have been received by the RPC layer (from a TCP or UDP socket) and are queued waiting to be serviced by one of the nfsd threads; it does not limit the amount of backlog in the socket buffers. The original implementation limited the total bytes of queued requests to the minimum of a quarter of (nmbclusters * MCLBYTES) and 45 MiB. The former limit seems reasonable, since requests queued in the socket buffers and replies being constructed to the requests in progress will all require some amount of network memory, but the 45 MiB limit is plainly ridiculous for modern memory sizes: when running 256 service threads on a busy server, 45 MiB would result in just a single maximum-sized NFS3PROC_WRITE queued per thread before throttling. Removing this limit exposed integer-overflow bugs in the original computation, and related bugs in the routines that actually account for the amount of traffic enqueued for service threads. The old implementation also attempted to reduce accounting overhead by batching updates until each queue is fully drained, but this is prone to livelock, resulting in repeated accumulate-throttle-drain cycles on a busy server. Various data types are changed to long or unsigned long; explicit 64-bit types are not used due to the unavailability of 64-bit atomics on many 32-bit platforms, but those platforms also cannot support nmbclusters large enough to cause overflow. This code (in a 10.1 kernel) is presently running on production NFS servers at CSAIL. Summary of this revision: * Removes 45 MiB limit on requests queued for nfsd service threads * Fixes integer-overflow and signedness bugs * Avoids unnecessary throttling by not deferring accounting for completed requests Differential Revision: https://reviews.freebsd.org/D2165 Reviewed by: rmacklem, mav MFC after: 30 days Relnotes: yes Sponsored by: MIT Computer Science & Artificial Intelligence Laboratory
2015-04-01 00:45:47 +00:00
svc_change_space_used(SVCPOOL *pool, long delta)
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
{
Fix overflow bugs in and remove obsolete limit from kernel RPC implementation. The kernel RPC code, which is responsible for the low-level scheduling of incoming NFS requests, contains a throttling mechanism that prevents too much kernel memory from being tied up by NFS requests that are being serviced. When the throttle is engaged, the RPC layer stops servicing incoming NFS sockets, resulting ultimately in backpressure on the clients (if they're using TCP). However, this is a very heavy-handed mechanism as it prevents all clients from making any requests, regardless of how heavy or light they are. (Thus, when engaged, the throttle often prevents clients from even mounting the filesystem.) The throttle mechanism applies specifically to requests that have been received by the RPC layer (from a TCP or UDP socket) and are queued waiting to be serviced by one of the nfsd threads; it does not limit the amount of backlog in the socket buffers. The original implementation limited the total bytes of queued requests to the minimum of a quarter of (nmbclusters * MCLBYTES) and 45 MiB. The former limit seems reasonable, since requests queued in the socket buffers and replies being constructed to the requests in progress will all require some amount of network memory, but the 45 MiB limit is plainly ridiculous for modern memory sizes: when running 256 service threads on a busy server, 45 MiB would result in just a single maximum-sized NFS3PROC_WRITE queued per thread before throttling. Removing this limit exposed integer-overflow bugs in the original computation, and related bugs in the routines that actually account for the amount of traffic enqueued for service threads. The old implementation also attempted to reduce accounting overhead by batching updates until each queue is fully drained, but this is prone to livelock, resulting in repeated accumulate-throttle-drain cycles on a busy server. Various data types are changed to long or unsigned long; explicit 64-bit types are not used due to the unavailability of 64-bit atomics on many 32-bit platforms, but those platforms also cannot support nmbclusters large enough to cause overflow. This code (in a 10.1 kernel) is presently running on production NFS servers at CSAIL. Summary of this revision: * Removes 45 MiB limit on requests queued for nfsd service threads * Fixes integer-overflow and signedness bugs * Avoids unnecessary throttling by not deferring accounting for completed requests Differential Revision: https://reviews.freebsd.org/D2165 Reviewed by: rmacklem, mav MFC after: 30 days Relnotes: yes Sponsored by: MIT Computer Science & Artificial Intelligence Laboratory
2015-04-01 00:45:47 +00:00
unsigned long value;
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
Fix overflow bugs in and remove obsolete limit from kernel RPC implementation. The kernel RPC code, which is responsible for the low-level scheduling of incoming NFS requests, contains a throttling mechanism that prevents too much kernel memory from being tied up by NFS requests that are being serviced. When the throttle is engaged, the RPC layer stops servicing incoming NFS sockets, resulting ultimately in backpressure on the clients (if they're using TCP). However, this is a very heavy-handed mechanism as it prevents all clients from making any requests, regardless of how heavy or light they are. (Thus, when engaged, the throttle often prevents clients from even mounting the filesystem.) The throttle mechanism applies specifically to requests that have been received by the RPC layer (from a TCP or UDP socket) and are queued waiting to be serviced by one of the nfsd threads; it does not limit the amount of backlog in the socket buffers. The original implementation limited the total bytes of queued requests to the minimum of a quarter of (nmbclusters * MCLBYTES) and 45 MiB. The former limit seems reasonable, since requests queued in the socket buffers and replies being constructed to the requests in progress will all require some amount of network memory, but the 45 MiB limit is plainly ridiculous for modern memory sizes: when running 256 service threads on a busy server, 45 MiB would result in just a single maximum-sized NFS3PROC_WRITE queued per thread before throttling. Removing this limit exposed integer-overflow bugs in the original computation, and related bugs in the routines that actually account for the amount of traffic enqueued for service threads. The old implementation also attempted to reduce accounting overhead by batching updates until each queue is fully drained, but this is prone to livelock, resulting in repeated accumulate-throttle-drain cycles on a busy server. Various data types are changed to long or unsigned long; explicit 64-bit types are not used due to the unavailability of 64-bit atomics on many 32-bit platforms, but those platforms also cannot support nmbclusters large enough to cause overflow. This code (in a 10.1 kernel) is presently running on production NFS servers at CSAIL. Summary of this revision: * Removes 45 MiB limit on requests queued for nfsd service threads * Fixes integer-overflow and signedness bugs * Avoids unnecessary throttling by not deferring accounting for completed requests Differential Revision: https://reviews.freebsd.org/D2165 Reviewed by: rmacklem, mav MFC after: 30 days Relnotes: yes Sponsored by: MIT Computer Science & Artificial Intelligence Laboratory
2015-04-01 00:45:47 +00:00
value = atomic_fetchadd_long(&pool->sp_space_used, delta) + delta;
Move most of NFS file handle affinity code out of the heavily congested global RPC thread pool lock and protect it with own set of locks. On synthetic benchmarks this improves peak NFS request rate by 40%.
2013-12-30 20:23:15 +00:00
if (delta > 0) {
if (value >= pool->sp_space_high && !pool->sp_space_throttled) {
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
pool->sp_space_throttled = TRUE;
pool->sp_space_throttle_count++;
}
Move most of NFS file handle affinity code out of the heavily congested global RPC thread pool lock and protect it with own set of locks. On synthetic benchmarks this improves peak NFS request rate by 40%.
2013-12-30 20:23:15 +00:00
if (value > pool->sp_space_used_highest)
pool->sp_space_used_highest = value;
} else {
if (value < pool->sp_space_low && pool->sp_space_throttled) {
pool->sp_space_throttled = FALSE;
svc_assign_waiting_sockets(pool);
}
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
}
}
Move most of NFS file handle affinity code out of the heavily congested global RPC thread pool lock and protect it with own set of locks. On synthetic benchmarks this improves peak NFS request rate by 40%.
2013-12-30 20:23:15 +00:00
static bool_t
svc_request_space_available(SVCPOOL *pool)
{
if (pool->sp_space_throttled)
return (FALSE);
return (TRUE);
}
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
static void
Split RPC pool threads into number of smaller semi-isolated groups. Old design with unified thread pool was good from the point of thread utilization. But single pool-wide mutex became huge congestion point for systems with many CPUs. To reduce the congestion create several thread groups within a pool (one group for every 6 CPUs and 12 threads), each group with own mutex. Each connection during its registration is assigned to one of the groups in round-robin fashion. File affinify code may still move requests between the groups, but otherwise groups are self-contained. MFC after: 2 weeks Sponsored by: iXsystems, Inc.
2014-06-08 11:19:32 +00:00
svc_run_internal(SVCGROUP *grp, bool_t ismaster)
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
{
Split RPC pool threads into number of smaller semi-isolated groups. Old design with unified thread pool was good from the point of thread utilization. But single pool-wide mutex became huge congestion point for systems with many CPUs. To reduce the congestion create several thread groups within a pool (one group for every 6 CPUs and 12 threads), each group with own mutex. Each connection during its registration is assigned to one of the groups in round-robin fashion. File affinify code may still move requests between the groups, but otherwise groups are self-contained. MFC after: 2 weeks Sponsored by: iXsystems, Inc.
2014-06-08 11:19:32 +00:00
SVCPOOL *pool = grp->sg_pool;
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
SVCTHREAD *st, *stpref;
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
SVCXPRT *xprt;
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
enum xprt_stat stat;
struct svc_req *rqstp;
Current reaction of the nfsd worker threads to any signal is exit. This is not correct at least for the stop requests. Check for stop conditions and suspend threads if requested. Reported and tested by: pho Sponsored by: The FreeBSD Foundation MFC after: 1 week
2014-12-08 16:33:18 +00:00
struct proc *p;
Fix overflow bugs in and remove obsolete limit from kernel RPC implementation. The kernel RPC code, which is responsible for the low-level scheduling of incoming NFS requests, contains a throttling mechanism that prevents too much kernel memory from being tied up by NFS requests that are being serviced. When the throttle is engaged, the RPC layer stops servicing incoming NFS sockets, resulting ultimately in backpressure on the clients (if they're using TCP). However, this is a very heavy-handed mechanism as it prevents all clients from making any requests, regardless of how heavy or light they are. (Thus, when engaged, the throttle often prevents clients from even mounting the filesystem.) The throttle mechanism applies specifically to requests that have been received by the RPC layer (from a TCP or UDP socket) and are queued waiting to be serviced by one of the nfsd threads; it does not limit the amount of backlog in the socket buffers. The original implementation limited the total bytes of queued requests to the minimum of a quarter of (nmbclusters * MCLBYTES) and 45 MiB. The former limit seems reasonable, since requests queued in the socket buffers and replies being constructed to the requests in progress will all require some amount of network memory, but the 45 MiB limit is plainly ridiculous for modern memory sizes: when running 256 service threads on a busy server, 45 MiB would result in just a single maximum-sized NFS3PROC_WRITE queued per thread before throttling. Removing this limit exposed integer-overflow bugs in the original computation, and related bugs in the routines that actually account for the amount of traffic enqueued for service threads. The old implementation also attempted to reduce accounting overhead by batching updates until each queue is fully drained, but this is prone to livelock, resulting in repeated accumulate-throttle-drain cycles on a busy server. Various data types are changed to long or unsigned long; explicit 64-bit types are not used due to the unavailability of 64-bit atomics on many 32-bit platforms, but those platforms also cannot support nmbclusters large enough to cause overflow. This code (in a 10.1 kernel) is presently running on production NFS servers at CSAIL. Summary of this revision: * Removes 45 MiB limit on requests queued for nfsd service threads * Fixes integer-overflow and signedness bugs * Avoids unnecessary throttling by not deferring accounting for completed requests Differential Revision: https://reviews.freebsd.org/D2165 Reviewed by: rmacklem, mav MFC after: 30 days Relnotes: yes Sponsored by: MIT Computer Science & Artificial Intelligence Laboratory
2015-04-01 00:45:47 +00:00
long sz;
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
int error;
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
st = mem_alloc(sizeof(*st));
Introduce new per-thread lock to protect the list of requests. This allows to slightly simplify svc_run_internal() code: if we processed all the requests in a queue, then we know that new one will not appear. MFC after: 2 weeks
2014-06-08 09:40:26 +00:00
mtx_init(&st->st_lock, "st_lock", NULL, MTX_DEF);
Move most of NFS file handle affinity code out of the heavily congested global RPC thread pool lock and protect it with own set of locks. On synthetic benchmarks this improves peak NFS request rate by 40%.
2013-12-30 20:23:15 +00:00
st->st_pool = pool;
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
st->st_xprt = NULL;
STAILQ_INIT(&st->st_reqs);
cv_init(&st->st_cond, "rpcsvc");
Split RPC pool threads into number of smaller semi-isolated groups. Old design with unified thread pool was good from the point of thread utilization. But single pool-wide mutex became huge congestion point for systems with many CPUs. To reduce the congestion create several thread groups within a pool (one group for every 6 CPUs and 12 threads), each group with own mutex. Each connection during its registration is assigned to one of the groups in round-robin fashion. File affinify code may still move requests between the groups, but otherwise groups are self-contained. MFC after: 2 weeks Sponsored by: iXsystems, Inc.
2014-06-08 11:19:32 +00:00
mtx_lock(&grp->sg_lock);
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
/*
* If we are a new thread which was spawned to cope with
* increased load, set the state back to SVCPOOL_ACTIVE.
*/
Split RPC pool threads into number of smaller semi-isolated groups. Old design with unified thread pool was good from the point of thread utilization. But single pool-wide mutex became huge congestion point for systems with many CPUs. To reduce the congestion create several thread groups within a pool (one group for every 6 CPUs and 12 threads), each group with own mutex. Each connection during its registration is assigned to one of the groups in round-robin fashion. File affinify code may still move requests between the groups, but otherwise groups are self-contained. MFC after: 2 weeks Sponsored by: iXsystems, Inc.
2014-06-08 11:19:32 +00:00
if (grp->sg_state == SVCPOOL_THREADSTARTING)
grp->sg_state = SVCPOOL_ACTIVE;
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
Split RPC pool threads into number of smaller semi-isolated groups. Old design with unified thread pool was good from the point of thread utilization. But single pool-wide mutex became huge congestion point for systems with many CPUs. To reduce the congestion create several thread groups within a pool (one group for every 6 CPUs and 12 threads), each group with own mutex. Each connection during its registration is assigned to one of the groups in round-robin fashion. File affinify code may still move requests between the groups, but otherwise groups are self-contained. MFC after: 2 weeks Sponsored by: iXsystems, Inc.
2014-06-08 11:19:32 +00:00
while (grp->sg_state != SVCPOOL_CLOSING) {
Some minor tuning to rpc/svc.c: - close cosmetic race in svc_exit(); - do not set wait timeout for idle threads if we have no use for wakeups; - create new requested thread sooner, not only after some another thread wakeup, that may happen later under constant load.
2013-11-14 13:51:53 +00:00
/*
* Create new thread if requested.
*/
Split RPC pool threads into number of smaller semi-isolated groups. Old design with unified thread pool was good from the point of thread utilization. But single pool-wide mutex became huge congestion point for systems with many CPUs. To reduce the congestion create several thread groups within a pool (one group for every 6 CPUs and 12 threads), each group with own mutex. Each connection during its registration is assigned to one of the groups in round-robin fashion. File affinify code may still move requests between the groups, but otherwise groups are self-contained. MFC after: 2 weeks Sponsored by: iXsystems, Inc.
2014-06-08 11:19:32 +00:00
if (grp->sg_state == SVCPOOL_THREADWANTED) {
grp->sg_state = SVCPOOL_THREADSTARTING;
grp->sg_lastcreatetime = time_uptime;
mtx_unlock(&grp->sg_lock);
svc_new_thread(grp);
mtx_lock(&grp->sg_lock);
Some minor tuning to rpc/svc.c: - close cosmetic race in svc_exit(); - do not set wait timeout for idle threads if we have no use for wakeups; - create new requested thread sooner, not only after some another thread wakeup, that may happen later under constant load.
2013-11-14 13:51:53 +00:00
continue;
}
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
/*
* Check for idle transports once per second.
*/
Split RPC pool threads into number of smaller semi-isolated groups. Old design with unified thread pool was good from the point of thread utilization. But single pool-wide mutex became huge congestion point for systems with many CPUs. To reduce the congestion create several thread groups within a pool (one group for every 6 CPUs and 12 threads), each group with own mutex. Each connection during its registration is assigned to one of the groups in round-robin fashion. File affinify code may still move requests between the groups, but otherwise groups are self-contained. MFC after: 2 weeks Sponsored by: iXsystems, Inc.
2014-06-08 11:19:32 +00:00
if (time_uptime > grp->sg_lastidlecheck) {
grp->sg_lastidlecheck = time_uptime;
svc_checkidle(grp);
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
}
xprt = st->st_xprt;
Introduce new per-thread lock to protect the list of requests. This allows to slightly simplify svc_run_internal() code: if we processed all the requests in a queue, then we know that new one will not appear. MFC after: 2 weeks
2014-06-08 09:40:26 +00:00
if (!xprt) {
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
/*
* Enforce maxthreads count.
*/
Fix the server side krpc so that the kernel nfsd threads terminate. Occationally the kernel nfsd threads would not terminate when a SIGKILL was posted for the kernel process (called nfsd (slave)). When this occurred, the thread associated with the process (called "ismaster") had returned from svc_run_internal() and was sleeping waiting for the other threads to terminate. The other threads (created by kthread_start()) were still in svc_run_internal() handling NFS RPCs. The only way this could occur is for the "ismaster" thread to return from svc_run_internal() without having called svc_exit(). There was only one place in the code where this could happen and this patch stops that from happening. Since the problem is intermittent, I cannot be sure if this has fixed the problem, but I have not seen an occurrence of the problem with this patch applied. Reviewed by: kib MFC after: 2 weeks Differential Revision: https://reviews.freebsd.org/D16087
2018-07-02 17:50:46 +00:00
if (!ismaster && grp->sg_threadcount >
grp->sg_maxthreads)
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
break;
/*
* Before sleeping, see if we can find an
* active transport which isn't being serviced
* by a thread.
*/
Remove several linear list traversals per request from RPC server code. Do not insert active ports into pool->sp_active list if they are success- fully assigned to some thread. This makes that list include only ports that really require attention, and so traversal can be reduced to simple taking the first one. Remove idle thread from pool->sp_idlethreads list when assigning some work (port of requests) to it. That again makes possible to replace list traversals with simple taking the first element.
2013-12-20 17:39:07 +00:00
if (svc_request_space_available(pool) &&
Split RPC pool threads into number of smaller semi-isolated groups. Old design with unified thread pool was good from the point of thread utilization. But single pool-wide mutex became huge congestion point for systems with many CPUs. To reduce the congestion create several thread groups within a pool (one group for every 6 CPUs and 12 threads), each group with own mutex. Each connection during its registration is assigned to one of the groups in round-robin fashion. File affinify code may still move requests between the groups, but otherwise groups are self-contained. MFC after: 2 weeks Sponsored by: iXsystems, Inc.
2014-06-08 11:19:32 +00:00
(xprt = TAILQ_FIRST(&grp->sg_active)) != NULL) {
TAILQ_REMOVE(&grp->sg_active, xprt, xp_alink);
Remove several linear list traversals per request from RPC server code. Do not insert active ports into pool->sp_active list if they are success- fully assigned to some thread. This makes that list include only ports that really require attention, and so traversal can be reduced to simple taking the first one. Remove idle thread from pool->sp_idlethreads list when assigning some work (port of requests) to it. That again makes possible to replace list traversals with simple taking the first element.
2013-12-20 17:39:07 +00:00
SVC_ACQUIRE(xprt);
xprt->xp_thread = st;
st->st_xprt = xprt;
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
continue;
Remove several linear list traversals per request from RPC server code. Do not insert active ports into pool->sp_active list if they are success- fully assigned to some thread. This makes that list include only ports that really require attention, and so traversal can be reduced to simple taking the first one. Remove idle thread from pool->sp_idlethreads list when assigning some work (port of requests) to it. That again makes possible to replace list traversals with simple taking the first element.
2013-12-20 17:39:07 +00:00
}
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
Split RPC pool threads into number of smaller semi-isolated groups. Old design with unified thread pool was good from the point of thread utilization. But single pool-wide mutex became huge congestion point for systems with many CPUs. To reduce the congestion create several thread groups within a pool (one group for every 6 CPUs and 12 threads), each group with own mutex. Each connection during its registration is assigned to one of the groups in round-robin fashion. File affinify code may still move requests between the groups, but otherwise groups are self-contained. MFC after: 2 weeks Sponsored by: iXsystems, Inc.
2014-06-08 11:19:32 +00:00
LIST_INSERT_HEAD(&grp->sg_idlethreads, st, st_ilink);
Some minor tuning to rpc/svc.c: - close cosmetic race in svc_exit(); - do not set wait timeout for idle threads if we have no use for wakeups; - create new requested thread sooner, not only after some another thread wakeup, that may happen later under constant load.
2013-11-14 13:51:53 +00:00
if (ismaster || (!ismaster &&
Split RPC pool threads into number of smaller semi-isolated groups. Old design with unified thread pool was good from the point of thread utilization. But single pool-wide mutex became huge congestion point for systems with many CPUs. To reduce the congestion create several thread groups within a pool (one group for every 6 CPUs and 12 threads), each group with own mutex. Each connection during its registration is assigned to one of the groups in round-robin fashion. File affinify code may still move requests between the groups, but otherwise groups are self-contained. MFC after: 2 weeks Sponsored by: iXsystems, Inc.
2014-06-08 11:19:32 +00:00
grp->sg_threadcount > grp->sg_minthreads))
Some minor tuning to rpc/svc.c: - close cosmetic race in svc_exit(); - do not set wait timeout for idle threads if we have no use for wakeups; - create new requested thread sooner, not only after some another thread wakeup, that may happen later under constant load.
2013-11-14 13:51:53 +00:00
error = cv_timedwait_sig(&st->st_cond,
Split RPC pool threads into number of smaller semi-isolated groups. Old design with unified thread pool was good from the point of thread utilization. But single pool-wide mutex became huge congestion point for systems with many CPUs. To reduce the congestion create several thread groups within a pool (one group for every 6 CPUs and 12 threads), each group with own mutex. Each connection during its registration is assigned to one of the groups in round-robin fashion. File affinify code may still move requests between the groups, but otherwise groups are self-contained. MFC after: 2 weeks Sponsored by: iXsystems, Inc.
2014-06-08 11:19:32 +00:00
&grp->sg_lock, 5 * hz);
Some minor tuning to rpc/svc.c: - close cosmetic race in svc_exit(); - do not set wait timeout for idle threads if we have no use for wakeups; - create new requested thread sooner, not only after some another thread wakeup, that may happen later under constant load.
2013-11-14 13:51:53 +00:00
else
error = cv_wait_sig(&st->st_cond,
Split RPC pool threads into number of smaller semi-isolated groups. Old design with unified thread pool was good from the point of thread utilization. But single pool-wide mutex became huge congestion point for systems with many CPUs. To reduce the congestion create several thread groups within a pool (one group for every 6 CPUs and 12 threads), each group with own mutex. Each connection during its registration is assigned to one of the groups in round-robin fashion. File affinify code may still move requests between the groups, but otherwise groups are self-contained. MFC after: 2 weeks Sponsored by: iXsystems, Inc.
2014-06-08 11:19:32 +00:00
&grp->sg_lock);
Remove st_idle variable, duplicating st_xprt. MFC after: 2 weeks
2014-06-08 10:18:22 +00:00
if (st->st_xprt == NULL)
Remove several linear list traversals per request from RPC server code. Do not insert active ports into pool->sp_active list if they are success- fully assigned to some thread. This makes that list include only ports that really require attention, and so traversal can be reduced to simple taking the first one. Remove idle thread from pool->sp_idlethreads list when assigning some work (port of requests) to it. That again makes possible to replace list traversals with simple taking the first element.
2013-12-20 17:39:07 +00:00
LIST_REMOVE(st, st_ilink);
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
/*
* Reduce worker thread count when idle.
*/
if (error == EWOULDBLOCK) {
if (!ismaster
Split RPC pool threads into number of smaller semi-isolated groups. Old design with unified thread pool was good from the point of thread utilization. But single pool-wide mutex became huge congestion point for systems with many CPUs. To reduce the congestion create several thread groups within a pool (one group for every 6 CPUs and 12 threads), each group with own mutex. Each connection during its registration is assigned to one of the groups in round-robin fashion. File affinify code may still move requests between the groups, but otherwise groups are self-contained. MFC after: 2 weeks Sponsored by: iXsystems, Inc.
2014-06-08 11:19:32 +00:00
&& (grp->sg_threadcount
> grp->sg_minthreads)
Introduce new per-thread lock to protect the list of requests. This allows to slightly simplify svc_run_internal() code: if we processed all the requests in a queue, then we know that new one will not appear. MFC after: 2 weeks
2014-06-08 09:40:26 +00:00
&& !st->st_xprt)
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
break;
Current reaction of the nfsd worker threads to any signal is exit. This is not correct at least for the stop requests. Check for stop conditions and suspend threads if requested. Reported and tested by: pho Sponsored by: The FreeBSD Foundation MFC after: 1 week
2014-12-08 16:33:18 +00:00
} else if (error != 0) {
KASSERT(error == EINTR || error == ERESTART,
("non-signal error %d", error));
Split RPC pool threads into number of smaller semi-isolated groups. Old design with unified thread pool was good from the point of thread utilization. But single pool-wide mutex became huge congestion point for systems with many CPUs. To reduce the congestion create several thread groups within a pool (one group for every 6 CPUs and 12 threads), each group with own mutex. Each connection during its registration is assigned to one of the groups in round-robin fashion. File affinify code may still move requests between the groups, but otherwise groups are self-contained. MFC after: 2 weeks Sponsored by: iXsystems, Inc.
2014-06-08 11:19:32 +00:00
mtx_unlock(&grp->sg_lock);
Current reaction of the nfsd worker threads to any signal is exit. This is not correct at least for the stop requests. Check for stop conditions and suspend threads if requested. Reported and tested by: pho Sponsored by: The FreeBSD Foundation MFC after: 1 week
2014-12-08 16:33:18 +00:00
p = curproc;
PROC_LOCK(p);
Add facility to stop all userspace processes. The supposed use of the feature is to quisce the system before suspend. Stop is implemented by reusing the thread_single(9) with the special mode SINGLE_ALLPROC. SINGLE_ALLPROC differs from the existing single-threading modes by allowing (requiring) caller to operate on other process. Interruptible sleeps for !TDF_SBDRY threads are suspended like SIGSTOP does it, instead of aborting the sleep, like SINGLE_NO_EXIT, to avoid spurious EINTRs on resume. Provide debugging sysctl debug.stop_all_proc, which causes total stop and suspends syncer, while waiting for variable reset for resume. It is used for debugging; should be removed after the real use of the interface is added. In collaboration with: pho Discussed with: avg Sponsored by: The FreeBSD Foundation MFC after: 2 weeks
2014-12-13 16:18:29 +00:00
if (P_SHOULDSTOP(p) ||
(p->p_flag & P_TOTAL_STOP) != 0) {
Current reaction of the nfsd worker threads to any signal is exit. This is not correct at least for the stop requests. Check for stop conditions and suspend threads if requested. Reported and tested by: pho Sponsored by: The FreeBSD Foundation MFC after: 1 week
2014-12-08 16:33:18 +00:00
thread_suspend_check(0);
PROC_UNLOCK(p);
mtx_lock(&grp->sg_lock);
} else {
PROC_UNLOCK(p);
svc_exit(pool);
mtx_lock(&grp->sg_lock);
break;
}
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
}
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
continue;
}
Split RPC pool threads into number of smaller semi-isolated groups. Old design with unified thread pool was good from the point of thread utilization. But single pool-wide mutex became huge congestion point for systems with many CPUs. To reduce the congestion create several thread groups within a pool (one group for every 6 CPUs and 12 threads), each group with own mutex. Each connection during its registration is assigned to one of the groups in round-robin fashion. File affinify code may still move requests between the groups, but otherwise groups are self-contained. MFC after: 2 weeks Sponsored by: iXsystems, Inc.
2014-06-08 11:19:32 +00:00
mtx_unlock(&grp->sg_lock);
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
Introduce new per-thread lock to protect the list of requests. This allows to slightly simplify svc_run_internal() code: if we processed all the requests in a queue, then we know that new one will not appear. MFC after: 2 weeks
2014-06-08 09:40:26 +00:00
/*
* Drain the transport socket and queue up any RPCs.
*/
xprt->xp_lastactive = time_uptime;
do {
if (!svc_request_space_available(pool))
break;
rqstp = NULL;
stat = svc_getreq(xprt, &rqstp);
if (rqstp) {
svc_change_space_used(pool, rqstp->rq_size);
/*
* See if the application has a preference
* for some other thread.
*/
if (pool->sp_assign) {
stpref = pool->sp_assign(st, rqstp);
Fix race in r267221. MFC after: 2 weeks
2014-06-09 15:00:43 +00:00
rqstp->rq_thread = stpref;
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
STAILQ_INSERT_TAIL(&stpref->st_reqs,
rqstp, rq_link);
Introduce new per-thread lock to protect the list of requests. This allows to slightly simplify svc_run_internal() code: if we processed all the requests in a queue, then we know that new one will not appear. MFC after: 2 weeks
2014-06-08 09:40:26 +00:00
mtx_unlock(&stpref->st_lock);
if (stpref != st)
rqstp = NULL;
Fix race in r267221. MFC after: 2 weeks
2014-06-09 15:00:43 +00:00
} else {
rqstp->rq_thread = st;
Introduce new per-thread lock to protect the list of requests. This allows to slightly simplify svc_run_internal() code: if we processed all the requests in a queue, then we know that new one will not appear. MFC after: 2 weeks
2014-06-08 09:40:26 +00:00
STAILQ_INSERT_TAIL(&st->st_reqs,
rqstp, rq_link);
Fix race in r267221. MFC after: 2 weeks
2014-06-09 15:00:43 +00:00
}
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
}
Introduce new per-thread lock to protect the list of requests. This allows to slightly simplify svc_run_internal() code: if we processed all the requests in a queue, then we know that new one will not appear. MFC after: 2 weeks
2014-06-08 09:40:26 +00:00
} while (rqstp == NULL && stat == XPRT_MOREREQS
Split RPC pool threads into number of smaller semi-isolated groups. Old design with unified thread pool was good from the point of thread utilization. But single pool-wide mutex became huge congestion point for systems with many CPUs. To reduce the congestion create several thread groups within a pool (one group for every 6 CPUs and 12 threads), each group with own mutex. Each connection during its registration is assigned to one of the groups in round-robin fashion. File affinify code may still move requests between the groups, but otherwise groups are self-contained. MFC after: 2 weeks Sponsored by: iXsystems, Inc.
2014-06-08 11:19:32 +00:00
&& grp->sg_state != SVCPOOL_CLOSING);
Introduce new per-thread lock to protect the list of requests. This allows to slightly simplify svc_run_internal() code: if we processed all the requests in a queue, then we know that new one will not appear. MFC after: 2 weeks
2014-06-08 09:40:26 +00:00
/*
* Move this transport to the end of the active list to
* ensure fairness when multiple transports are active.
* If this was the last queued request, svc_getreq will end
* up calling xprt_inactive to remove from the active list.
*/
Split RPC pool threads into number of smaller semi-isolated groups. Old design with unified thread pool was good from the point of thread utilization. But single pool-wide mutex became huge congestion point for systems with many CPUs. To reduce the congestion create several thread groups within a pool (one group for every 6 CPUs and 12 threads), each group with own mutex. Each connection during its registration is assigned to one of the groups in round-robin fashion. File affinify code may still move requests between the groups, but otherwise groups are self-contained. MFC after: 2 weeks Sponsored by: iXsystems, Inc.
2014-06-08 11:19:32 +00:00
mtx_lock(&grp->sg_lock);
Introduce new per-thread lock to protect the list of requests. This allows to slightly simplify svc_run_internal() code: if we processed all the requests in a queue, then we know that new one will not appear. MFC after: 2 weeks
2014-06-08 09:40:26 +00:00
xprt->xp_thread = NULL;
st->st_xprt = NULL;
if (xprt->xp_active) {
if (!svc_request_space_available(pool) ||
!xprt_assignthread(xprt))
Split RPC pool threads into number of smaller semi-isolated groups. Old design with unified thread pool was good from the point of thread utilization. But single pool-wide mutex became huge congestion point for systems with many CPUs. To reduce the congestion create several thread groups within a pool (one group for every 6 CPUs and 12 threads), each group with own mutex. Each connection during its registration is assigned to one of the groups in round-robin fashion. File affinify code may still move requests between the groups, but otherwise groups are self-contained. MFC after: 2 weeks Sponsored by: iXsystems, Inc.
2014-06-08 11:19:32 +00:00
TAILQ_INSERT_TAIL(&grp->sg_active,
Introduce new per-thread lock to protect the list of requests. This allows to slightly simplify svc_run_internal() code: if we processed all the requests in a queue, then we know that new one will not appear. MFC after: 2 weeks
2014-06-08 09:40:26 +00:00
xprt, xp_alink);
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
}
Split RPC pool threads into number of smaller semi-isolated groups. Old design with unified thread pool was good from the point of thread utilization. But single pool-wide mutex became huge congestion point for systems with many CPUs. To reduce the congestion create several thread groups within a pool (one group for every 6 CPUs and 12 threads), each group with own mutex. Each connection during its registration is assigned to one of the groups in round-robin fashion. File affinify code may still move requests between the groups, but otherwise groups are self-contained. MFC after: 2 weeks Sponsored by: iXsystems, Inc.
2014-06-08 11:19:32 +00:00
mtx_unlock(&grp->sg_lock);
Introduce new per-thread lock to protect the list of requests. This allows to slightly simplify svc_run_internal() code: if we processed all the requests in a queue, then we know that new one will not appear. MFC after: 2 weeks
2014-06-08 09:40:26 +00:00
SVC_RELEASE(xprt);
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
/*
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
* Execute what we have queued.
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
*/
Introduce new per-thread lock to protect the list of requests. This allows to slightly simplify svc_run_internal() code: if we processed all the requests in a queue, then we know that new one will not appear. MFC after: 2 weeks
2014-06-08 09:40:26 +00:00
mtx_lock(&st->st_lock);
while ((rqstp = STAILQ_FIRST(&st->st_reqs)) != NULL) {
STAILQ_REMOVE_HEAD(&st->st_reqs, rq_link);
mtx_unlock(&st->st_lock);
Fix overflow bugs in and remove obsolete limit from kernel RPC implementation. The kernel RPC code, which is responsible for the low-level scheduling of incoming NFS requests, contains a throttling mechanism that prevents too much kernel memory from being tied up by NFS requests that are being serviced. When the throttle is engaged, the RPC layer stops servicing incoming NFS sockets, resulting ultimately in backpressure on the clients (if they're using TCP). However, this is a very heavy-handed mechanism as it prevents all clients from making any requests, regardless of how heavy or light they are. (Thus, when engaged, the throttle often prevents clients from even mounting the filesystem.) The throttle mechanism applies specifically to requests that have been received by the RPC layer (from a TCP or UDP socket) and are queued waiting to be serviced by one of the nfsd threads; it does not limit the amount of backlog in the socket buffers. The original implementation limited the total bytes of queued requests to the minimum of a quarter of (nmbclusters * MCLBYTES) and 45 MiB. The former limit seems reasonable, since requests queued in the socket buffers and replies being constructed to the requests in progress will all require some amount of network memory, but the 45 MiB limit is plainly ridiculous for modern memory sizes: when running 256 service threads on a busy server, 45 MiB would result in just a single maximum-sized NFS3PROC_WRITE queued per thread before throttling. Removing this limit exposed integer-overflow bugs in the original computation, and related bugs in the routines that actually account for the amount of traffic enqueued for service threads. The old implementation also attempted to reduce accounting overhead by batching updates until each queue is fully drained, but this is prone to livelock, resulting in repeated accumulate-throttle-drain cycles on a busy server. Various data types are changed to long or unsigned long; explicit 64-bit types are not used due to the unavailability of 64-bit atomics on many 32-bit platforms, but those platforms also cannot support nmbclusters large enough to cause overflow. This code (in a 10.1 kernel) is presently running on production NFS servers at CSAIL. Summary of this revision: * Removes 45 MiB limit on requests queued for nfsd service threads * Fixes integer-overflow and signedness bugs * Avoids unnecessary throttling by not deferring accounting for completed requests Differential Revision: https://reviews.freebsd.org/D2165 Reviewed by: rmacklem, mav MFC after: 30 days Relnotes: yes Sponsored by: MIT Computer Science & Artificial Intelligence Laboratory
2015-04-01 00:45:47 +00:00
sz = (long)rqstp->rq_size;
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
svc_executereq(rqstp);
Fix overflow bugs in and remove obsolete limit from kernel RPC implementation. The kernel RPC code, which is responsible for the low-level scheduling of incoming NFS requests, contains a throttling mechanism that prevents too much kernel memory from being tied up by NFS requests that are being serviced. When the throttle is engaged, the RPC layer stops servicing incoming NFS sockets, resulting ultimately in backpressure on the clients (if they're using TCP). However, this is a very heavy-handed mechanism as it prevents all clients from making any requests, regardless of how heavy or light they are. (Thus, when engaged, the throttle often prevents clients from even mounting the filesystem.) The throttle mechanism applies specifically to requests that have been received by the RPC layer (from a TCP or UDP socket) and are queued waiting to be serviced by one of the nfsd threads; it does not limit the amount of backlog in the socket buffers. The original implementation limited the total bytes of queued requests to the minimum of a quarter of (nmbclusters * MCLBYTES) and 45 MiB. The former limit seems reasonable, since requests queued in the socket buffers and replies being constructed to the requests in progress will all require some amount of network memory, but the 45 MiB limit is plainly ridiculous for modern memory sizes: when running 256 service threads on a busy server, 45 MiB would result in just a single maximum-sized NFS3PROC_WRITE queued per thread before throttling. Removing this limit exposed integer-overflow bugs in the original computation, and related bugs in the routines that actually account for the amount of traffic enqueued for service threads. The old implementation also attempted to reduce accounting overhead by batching updates until each queue is fully drained, but this is prone to livelock, resulting in repeated accumulate-throttle-drain cycles on a busy server. Various data types are changed to long or unsigned long; explicit 64-bit types are not used due to the unavailability of 64-bit atomics on many 32-bit platforms, but those platforms also cannot support nmbclusters large enough to cause overflow. This code (in a 10.1 kernel) is presently running on production NFS servers at CSAIL. Summary of this revision: * Removes 45 MiB limit on requests queued for nfsd service threads * Fixes integer-overflow and signedness bugs * Avoids unnecessary throttling by not deferring accounting for completed requests Differential Revision: https://reviews.freebsd.org/D2165 Reviewed by: rmacklem, mav MFC after: 30 days Relnotes: yes Sponsored by: MIT Computer Science & Artificial Intelligence Laboratory
2015-04-01 00:45:47 +00:00
svc_change_space_used(pool, -sz);
Introduce new per-thread lock to protect the list of requests. This allows to slightly simplify svc_run_internal() code: if we processed all the requests in a queue, then we know that new one will not appear. MFC after: 2 weeks
2014-06-08 09:40:26 +00:00
mtx_lock(&st->st_lock);
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
}
Introduce new per-thread lock to protect the list of requests. This allows to slightly simplify svc_run_internal() code: if we processed all the requests in a queue, then we know that new one will not appear. MFC after: 2 weeks
2014-06-08 09:40:26 +00:00
mtx_unlock(&st->st_lock);
Split RPC pool threads into number of smaller semi-isolated groups. Old design with unified thread pool was good from the point of thread utilization. But single pool-wide mutex became huge congestion point for systems with many CPUs. To reduce the congestion create several thread groups within a pool (one group for every 6 CPUs and 12 threads), each group with own mutex. Each connection during its registration is assigned to one of the groups in round-robin fashion. File affinify code may still move requests between the groups, but otherwise groups are self-contained. MFC after: 2 weeks Sponsored by: iXsystems, Inc.
2014-06-08 11:19:32 +00:00
mtx_lock(&grp->sg_lock);
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
}
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
if (st->st_xprt) {
xprt = st->st_xprt;
st->st_xprt = NULL;
SVC_RELEASE(xprt);
}
KASSERT(STAILQ_EMPTY(&st->st_reqs), ("stray reqs on exit"));
Introduce new per-thread lock to protect the list of requests. This allows to slightly simplify svc_run_internal() code: if we processed all the requests in a queue, then we know that new one will not appear. MFC after: 2 weeks
2014-06-08 09:40:26 +00:00
mtx_destroy(&st->st_lock);
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
cv_destroy(&st->st_cond);
mem_free(st, sizeof(*st));
Split RPC pool threads into number of smaller semi-isolated groups. Old design with unified thread pool was good from the point of thread utilization. But single pool-wide mutex became huge congestion point for systems with many CPUs. To reduce the congestion create several thread groups within a pool (one group for every 6 CPUs and 12 threads), each group with own mutex. Each connection during its registration is assigned to one of the groups in round-robin fashion. File affinify code may still move requests between the groups, but otherwise groups are self-contained. MFC after: 2 weeks Sponsored by: iXsystems, Inc.
2014-06-08 11:19:32 +00:00
grp->sg_threadcount--;
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
if (!ismaster)
Split RPC pool threads into number of smaller semi-isolated groups. Old design with unified thread pool was good from the point of thread utilization. But single pool-wide mutex became huge congestion point for systems with many CPUs. To reduce the congestion create several thread groups within a pool (one group for every 6 CPUs and 12 threads), each group with own mutex. Each connection during its registration is assigned to one of the groups in round-robin fashion. File affinify code may still move requests between the groups, but otherwise groups are self-contained. MFC after: 2 weeks Sponsored by: iXsystems, Inc.
2014-06-08 11:19:32 +00:00
wakeup(grp);
mtx_unlock(&grp->sg_lock);
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
}
static void
svc_thread_start(void *arg)
{
Split RPC pool threads into number of smaller semi-isolated groups. Old design with unified thread pool was good from the point of thread utilization. But single pool-wide mutex became huge congestion point for systems with many CPUs. To reduce the congestion create several thread groups within a pool (one group for every 6 CPUs and 12 threads), each group with own mutex. Each connection during its registration is assigned to one of the groups in round-robin fashion. File affinify code may still move requests between the groups, but otherwise groups are self-contained. MFC after: 2 weeks Sponsored by: iXsystems, Inc.
2014-06-08 11:19:32 +00:00
svc_run_internal((SVCGROUP *) arg, FALSE);
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
kthread_exit();
}
static void
Split RPC pool threads into number of smaller semi-isolated groups. Old design with unified thread pool was good from the point of thread utilization. But single pool-wide mutex became huge congestion point for systems with many CPUs. To reduce the congestion create several thread groups within a pool (one group for every 6 CPUs and 12 threads), each group with own mutex. Each connection during its registration is assigned to one of the groups in round-robin fashion. File affinify code may still move requests between the groups, but otherwise groups are self-contained. MFC after: 2 weeks Sponsored by: iXsystems, Inc.
2014-06-08 11:19:32 +00:00
svc_new_thread(SVCGROUP *grp)
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
{
Split RPC pool threads into number of smaller semi-isolated groups. Old design with unified thread pool was good from the point of thread utilization. But single pool-wide mutex became huge congestion point for systems with many CPUs. To reduce the congestion create several thread groups within a pool (one group for every 6 CPUs and 12 threads), each group with own mutex. Each connection during its registration is assigned to one of the groups in round-robin fashion. File affinify code may still move requests between the groups, but otherwise groups are self-contained. MFC after: 2 weeks Sponsored by: iXsystems, Inc.
2014-06-08 11:19:32 +00:00
SVCPOOL *pool = grp->sg_pool;
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
struct thread *td;
Improve locking of sg_threadcount. MFC after: 1 week
2015-11-19 08:04:05 +00:00
mtx_lock(&grp->sg_lock);
Split RPC pool threads into number of smaller semi-isolated groups. Old design with unified thread pool was good from the point of thread utilization. But single pool-wide mutex became huge congestion point for systems with many CPUs. To reduce the congestion create several thread groups within a pool (one group for every 6 CPUs and 12 threads), each group with own mutex. Each connection during its registration is assigned to one of the groups in round-robin fashion. File affinify code may still move requests between the groups, but otherwise groups are self-contained. MFC after: 2 weeks Sponsored by: iXsystems, Inc.
2014-06-08 11:19:32 +00:00
grp->sg_threadcount++;
Improve locking of sg_threadcount. MFC after: 1 week
2015-11-19 08:04:05 +00:00
mtx_unlock(&grp->sg_lock);
Split RPC pool threads into number of smaller semi-isolated groups. Old design with unified thread pool was good from the point of thread utilization. But single pool-wide mutex became huge congestion point for systems with many CPUs. To reduce the congestion create several thread groups within a pool (one group for every 6 CPUs and 12 threads), each group with own mutex. Each connection during its registration is assigned to one of the groups in round-robin fashion. File affinify code may still move requests between the groups, but otherwise groups are self-contained. MFC after: 2 weeks Sponsored by: iXsystems, Inc.
2014-06-08 11:19:32 +00:00
kthread_add(svc_thread_start, grp, pool->sp_proc, &td, 0, 0,
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
"%s: service", pool->sp_name);
}
void
svc_run(SVCPOOL *pool)
{
Split RPC pool threads into number of smaller semi-isolated groups. Old design with unified thread pool was good from the point of thread utilization. But single pool-wide mutex became huge congestion point for systems with many CPUs. To reduce the congestion create several thread groups within a pool (one group for every 6 CPUs and 12 threads), each group with own mutex. Each connection during its registration is assigned to one of the groups in round-robin fashion. File affinify code may still move requests between the groups, but otherwise groups are self-contained. MFC after: 2 weeks Sponsored by: iXsystems, Inc.
2014-06-08 11:19:32 +00:00
int g, i;
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
struct proc *p;
struct thread *td;
Split RPC pool threads into number of smaller semi-isolated groups. Old design with unified thread pool was good from the point of thread utilization. But single pool-wide mutex became huge congestion point for systems with many CPUs. To reduce the congestion create several thread groups within a pool (one group for every 6 CPUs and 12 threads), each group with own mutex. Each connection during its registration is assigned to one of the groups in round-robin fashion. File affinify code may still move requests between the groups, but otherwise groups are self-contained. MFC after: 2 weeks Sponsored by: iXsystems, Inc.
2014-06-08 11:19:32 +00:00
SVCGROUP *grp;
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
p = curproc;
td = curthread;
snprintf(td->td_name, sizeof(td->td_name),
"%s: master", pool->sp_name);
pool->sp_state = SVCPOOL_ACTIVE;
pool->sp_proc = p;
Split RPC pool threads into number of smaller semi-isolated groups. Old design with unified thread pool was good from the point of thread utilization. But single pool-wide mutex became huge congestion point for systems with many CPUs. To reduce the congestion create several thread groups within a pool (one group for every 6 CPUs and 12 threads), each group with own mutex. Each connection during its registration is assigned to one of the groups in round-robin fashion. File affinify code may still move requests between the groups, but otherwise groups are self-contained. MFC after: 2 weeks Sponsored by: iXsystems, Inc.
2014-06-08 11:19:32 +00:00
/* Choose group count based on number of threads and CPUs. */
pool->sp_groupcount = max(1, min(SVC_MAXGROUPS,
min(pool->sp_maxthreads / 2, mp_ncpus) / 6));
for (g = 0; g < pool->sp_groupcount; g++) {
grp = &pool->sp_groups[g];
grp->sg_minthreads = max(1,
pool->sp_minthreads / pool->sp_groupcount);
grp->sg_maxthreads = max(1,
pool->sp_maxthreads / pool->sp_groupcount);
grp->sg_lastcreatetime = time_uptime;
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
}
Split RPC pool threads into number of smaller semi-isolated groups. Old design with unified thread pool was good from the point of thread utilization. But single pool-wide mutex became huge congestion point for systems with many CPUs. To reduce the congestion create several thread groups within a pool (one group for every 6 CPUs and 12 threads), each group with own mutex. Each connection during its registration is assigned to one of the groups in round-robin fashion. File affinify code may still move requests between the groups, but otherwise groups are self-contained. MFC after: 2 weeks Sponsored by: iXsystems, Inc.
2014-06-08 11:19:32 +00:00
/* Starting threads */
Improve locking of sg_threadcount. MFC after: 1 week
2015-11-19 08:04:05 +00:00
pool->sp_groups[0].sg_threadcount++;
Split RPC pool threads into number of smaller semi-isolated groups. Old design with unified thread pool was good from the point of thread utilization. But single pool-wide mutex became huge congestion point for systems with many CPUs. To reduce the congestion create several thread groups within a pool (one group for every 6 CPUs and 12 threads), each group with own mutex. Each connection during its registration is assigned to one of the groups in round-robin fashion. File affinify code may still move requests between the groups, but otherwise groups are self-contained. MFC after: 2 weeks Sponsored by: iXsystems, Inc.
2014-06-08 11:19:32 +00:00
for (g = 0; g < pool->sp_groupcount; g++) {
grp = &pool->sp_groups[g];
for (i = ((g == 0) ? 1 : 0); i < grp->sg_minthreads; i++)
svc_new_thread(grp);
}
svc_run_internal(&pool->sp_groups[0], TRUE);
/* Waiting for threads to stop. */
for (g = 0; g < pool->sp_groupcount; g++) {
grp = &pool->sp_groups[g];
mtx_lock(&grp->sg_lock);
while (grp->sg_threadcount > 0)
msleep(grp, &grp->sg_lock, 0, "svcexit", 0);
mtx_unlock(&grp->sg_lock);
}
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
}
void
svc_exit(SVCPOOL *pool)
{
Split RPC pool threads into number of smaller semi-isolated groups. Old design with unified thread pool was good from the point of thread utilization. But single pool-wide mutex became huge congestion point for systems with many CPUs. To reduce the congestion create several thread groups within a pool (one group for every 6 CPUs and 12 threads), each group with own mutex. Each connection during its registration is assigned to one of the groups in round-robin fashion. File affinify code may still move requests between the groups, but otherwise groups are self-contained. MFC after: 2 weeks Sponsored by: iXsystems, Inc.
2014-06-08 11:19:32 +00:00
SVCGROUP *grp;
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
SVCTHREAD *st;
Split RPC pool threads into number of smaller semi-isolated groups. Old design with unified thread pool was good from the point of thread utilization. But single pool-wide mutex became huge congestion point for systems with many CPUs. To reduce the congestion create several thread groups within a pool (one group for every 6 CPUs and 12 threads), each group with own mutex. Each connection during its registration is assigned to one of the groups in round-robin fashion. File affinify code may still move requests between the groups, but otherwise groups are self-contained. MFC after: 2 weeks Sponsored by: iXsystems, Inc.
2014-06-08 11:19:32 +00:00
int g;
pool->sp_state = SVCPOOL_CLOSING;
for (g = 0; g < pool->sp_groupcount; g++) {
grp = &pool->sp_groups[g];
mtx_lock(&grp->sg_lock);
if (grp->sg_state != SVCPOOL_CLOSING) {
grp->sg_state = SVCPOOL_CLOSING;
LIST_FOREACH(st, &grp->sg_idlethreads, st_ilink)
cv_signal(&st->st_cond);
}
mtx_unlock(&grp->sg_lock);
Some minor tuning to rpc/svc.c: - close cosmetic race in svc_exit(); - do not set wait timeout for idle threads if we have no use for wakeups; - create new requested thread sooner, not only after some another thread wakeup, that may happen later under constant load.
2013-11-14 13:51:53 +00:00
}
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
}
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
bool_t
svc_getargs(struct svc_req *rqstp, xdrproc_t xargs, void *args)
{
struct mbuf *m;
XDR xdrs;
bool_t stat;
m = rqstp->rq_args;
rqstp->rq_args = NULL;
xdrmbuf_create(&xdrs, m, XDR_DECODE);
stat = xargs(&xdrs, args);
XDR_DESTROY(&xdrs);
return (stat);
}
bool_t
svc_freeargs(struct svc_req *rqstp, xdrproc_t xargs, void *args)
{
XDR xdrs;
if (rqstp->rq_addr) {
free(rqstp->rq_addr, M_SONAME);
rqstp->rq_addr = NULL;
}
xdrs.x_op = XDR_FREE;
return (xargs(&xdrs, args));
}
void
svc_freereq(struct svc_req *rqstp)
{
SVCTHREAD *st;
SVCPOOL *pool;
st = rqstp->rq_thread;
if (st) {
Move most of NFS file handle affinity code out of the heavily congested global RPC thread pool lock and protect it with own set of locks. On synthetic benchmarks this improves peak NFS request rate by 40%.
2013-12-30 20:23:15 +00:00
pool = st->st_pool;
Implement support for RPCSEC_GSS authentication to both the NFS client and server. This replaces the RPC implementation of the NFS client and server with the newer RPC implementation originally developed (actually ported from the userland sunrpc code) to support the NFS Lock Manager. I have tested this code extensively and I believe it is stable and that performance is at least equal to the legacy RPC implementation. The NFS code currently contains support for both the new RPC implementation and the older legacy implementation inherited from the original NFS codebase. The default is to use the new implementation - add the NFS_LEGACYRPC option to fall back to the old code. When I merge this support back to RELENG_7, I will probably change this so that users have to 'opt in' to get the new code. To use RPCSEC_GSS on either client or server, you must build a kernel which includes the KGSSAPI option and the crypto device. On the userland side, you must build at least a new libc, mountd, mount_nfs and gssd. You must install new versions of /etc/rc.d/gssd and /etc/rc.d/nfsd and add 'gssd_enable=YES' to /etc/rc.conf. As long as gssd is running, you should be able to mount an NFS filesystem from a server that requires RPCSEC_GSS authentication. The mount itself can happen without any kerberos credentials but all access to the filesystem will be denied unless the accessing user has a valid ticket file in the standard place (/tmp/krb5cc_<uid>). There is currently no support for situations where the ticket file is in a different place, such as when the user logged in via SSH and has delegated credentials from that login. This restriction is also present in Solaris and Linux. In theory, we could improve this in future, possibly using Brooks Davis' implementation of variant symlinks. Supporting RPCSEC_GSS on a server is nearly as simple. You must create service creds for the server in the form 'nfs/<fqdn>@<REALM>' and install them in /etc/krb5.keytab. The standard heimdal utility ktutil makes this fairly easy. After the service creds have been created, you can add a '-sec=krb5' option to /etc/exports and restart both mountd and nfsd. The only other difference an administrator should notice is that nfsd doesn't fork to create service threads any more. In normal operation, there will be two nfsd processes, one in userland waiting for TCP connections and one in the kernel handling requests. The latter process will create as many kthreads as required - these should be visible via 'top -H'. The code has some support for varying the number of service threads according to load but initially at least, nfsd uses a fixed number of threads according to the value supplied to its '-n' option. Sponsored by: Isilon Systems MFC after: 1 month
2008-11-03 10:38:00 +00:00
if (pool->sp_done)
pool->sp_done(st, rqstp);
}
if (rqstp->rq_auth.svc_ah_ops)
SVCAUTH_RELEASE(&rqstp->rq_auth);
if (rqstp->rq_xprt) {
SVC_RELEASE(rqstp->rq_xprt);
}
if (rqstp->rq_addr)
free(rqstp->rq_addr, M_SONAME);
if (rqstp->rq_args)
m_freem(rqstp->rq_args);
free(rqstp, M_RPC);
}
Reference in New Issue Copy Permalink