2005-01-07 01:45:51 +00:00
|
|
|
/*-
|
1997-02-10 02:22:35 +00:00
|
|
|
* Copyright (c) 1989, 1991, 1993, 1995
|
1994-05-24 10:09:53 +00:00
|
|
|
* The Regents of the University of California. All rights reserved.
|
|
|
|
|
*
|
|
|
|
|
* This code is derived from software contributed to Berkeley by
|
|
|
|
|
* Rick Macklem at The University of Guelph.
|
|
|
|
|
*
|
|
|
|
|
* Redistribution and use in source and binary forms, with or without
|
|
|
|
|
* modification, are permitted provided that the following conditions
|
|
|
|
|
* are met:
|
|
|
|
|
* 1. Redistributions of source code must retain the above copyright
|
|
|
|
|
* notice, this list of conditions and the following disclaimer.
|
|
|
|
|
* 2. 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.
|
|
|
|
|
* 4. Neither the name of the University 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 REGENTS 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 REGENTS 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.
|
|
|
|
|
*
|
1997-02-10 02:22:35 +00:00
|
|
|
* @(#)nfs_socket.c 8.5 (Berkeley) 3/30/95
|
1994-05-24 10:09:53 +00:00
|
|
|
*/
|
|
|
|
|
|
2001-09-18 23:32:09 +00:00
|
|
|
#include <sys/cdefs.h>
|
|
|
|
|
__FBSDID("$FreeBSD$");
|
|
|
|
|
|
1994-05-24 10:09:53 +00:00
|
|
|
/*
|
|
|
|
|
* Socket operations for use by nfs
|
|
|
|
|
*/
|
|
|
|
|
|
2006-01-28 19:24:40 +00:00
|
|
|
#include "opt_mac.h"
|
|
|
|
|
|
1994-05-24 10:09:53 +00:00
|
|
|
#include <sys/param.h>
|
|
|
|
|
#include <sys/systm.h>
|
2009-05-27 14:11:23 +00:00
|
|
|
#include <sys/jail.h>
|
1994-05-24 10:09:53 +00:00
|
|
|
#include <sys/kernel.h>
|
2001-05-01 08:13:21 +00:00
|
|
|
#include <sys/lock.h>
|
|
|
|
|
#include <sys/malloc.h>
|
1994-05-24 10:09:53 +00:00
|
|
|
#include <sys/mbuf.h>
|
2001-05-01 08:13:21 +00:00
|
|
|
#include <sys/mount.h>
|
|
|
|
|
#include <sys/mutex.h>
|
|
|
|
|
#include <sys/proc.h>
|
1994-05-24 10:09:53 +00:00
|
|
|
#include <sys/protosw.h>
|
2005-09-27 18:09:42 +00:00
|
|
|
#include <sys/refcount.h>
|
2001-05-01 08:13:21 +00:00
|
|
|
#include <sys/signalvar.h>
|
1994-05-24 10:09:53 +00:00
|
|
|
#include <sys/socket.h>
|
|
|
|
|
#include <sys/socketvar.h>
|
The VFS/BIO subsystem contained a number of hacks in order to optimize
piecemeal, middle-of-file writes for NFS. These hacks have caused no
end of trouble, especially when combined with mmap(). I've removed
them. Instead, NFS will issue a read-before-write to fully
instantiate the struct buf containing the write. NFS does, however,
optimize piecemeal appends to files. For most common file operations,
you will not notice the difference. The sole remaining fragment in
the VFS/BIO system is b_dirtyoff/end, which NFS uses to avoid cache
coherency issues with read-merge-write style operations. NFS also
optimizes the write-covers-entire-buffer case by avoiding the
read-before-write. There is quite a bit of room for further
optimization in these areas.
The VM system marks pages fully-valid (AKA vm_page_t->valid =
VM_PAGE_BITS_ALL) in several places, most noteably in vm_fault. This
is not correct operation. The vm_pager_get_pages() code is now
responsible for marking VM pages all-valid. A number of VM helper
routines have been added to aid in zeroing-out the invalid portions of
a VM page prior to the page being marked all-valid. This operation is
necessary to properly support mmap(). The zeroing occurs most often
when dealing with file-EOF situations. Several bugs have been fixed
in the NFS subsystem, including bits handling file and directory EOF
situations and buf->b_flags consistancy issues relating to clearing
B_ERROR & B_INVAL, and handling B_DONE.
getblk() and allocbuf() have been rewritten. B_CACHE operation is now
formally defined in comments and more straightforward in
implementation. B_CACHE for VMIO buffers is based on the validity of
the backing store. B_CACHE for non-VMIO buffers is based simply on
whether the buffer is B_INVAL or not (B_CACHE set if B_INVAL clear,
and vise-versa). biodone() is now responsible for setting B_CACHE
when a successful read completes. B_CACHE is also set when a bdwrite()
is initiated and when a bwrite() is initiated. VFS VOP_BWRITE
routines (there are only two - nfs_bwrite() and bwrite()) are now
expected to set B_CACHE. This means that bowrite() and bawrite() also
set B_CACHE indirectly.
There are a number of places in the code which were previously using
buf->b_bufsize (which is DEV_BSIZE aligned) when they should have
been using buf->b_bcount. These have been fixed. getblk() now clears
B_DONE on return because the rest of the system is so bad about
dealing with B_DONE.
Major fixes to NFS/TCP have been made. A server-side bug could cause
requests to be lost by the server due to nfs_realign() overwriting
other rpc's in the same TCP mbuf chain. The server's kernel must be
recompiled to get the benefit of the fixes.
Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
1999-05-02 23:57:16 +00:00
|
|
|
#include <sys/sysctl.h>
|
2001-05-01 08:13:21 +00:00
|
|
|
#include <sys/syslog.h>
|
|
|
|
|
#include <sys/vnode.h>
|
1994-05-24 10:09:53 +00:00
|
|
|
|
|
|
|
|
#include <netinet/in.h>
|
|
|
|
|
#include <netinet/tcp.h>
|
1994-10-17 17:47:45 +00:00
|
|
|
|
1994-05-24 10:09:53 +00:00
|
|
|
#include <nfs/rpcv2.h>
|
1995-06-27 11:07:30 +00:00
|
|
|
#include <nfs/nfsproto.h>
|
2001-09-18 23:32:09 +00:00
|
|
|
#include <nfsserver/nfs.h>
|
1994-05-24 10:09:53 +00:00
|
|
|
#include <nfs/xdr_subs.h>
|
2001-09-18 23:32:09 +00:00
|
|
|
#include <nfsserver/nfsm_subs.h>
|
1994-05-24 10:09:53 +00:00
|
|
|
|
2006-10-22 11:52:19 +00:00
|
|
|
#include <security/mac/mac_framework.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
|
|
|
#ifdef NFS_LEGACYRPC
|
|
|
|
|
|
1994-05-24 10:09:53 +00:00
|
|
|
#define TRUE 1
|
|
|
|
|
#define FALSE 0
|
|
|
|
|
|
The VFS/BIO subsystem contained a number of hacks in order to optimize
piecemeal, middle-of-file writes for NFS. These hacks have caused no
end of trouble, especially when combined with mmap(). I've removed
them. Instead, NFS will issue a read-before-write to fully
instantiate the struct buf containing the write. NFS does, however,
optimize piecemeal appends to files. For most common file operations,
you will not notice the difference. The sole remaining fragment in
the VFS/BIO system is b_dirtyoff/end, which NFS uses to avoid cache
coherency issues with read-merge-write style operations. NFS also
optimizes the write-covers-entire-buffer case by avoiding the
read-before-write. There is quite a bit of room for further
optimization in these areas.
The VM system marks pages fully-valid (AKA vm_page_t->valid =
VM_PAGE_BITS_ALL) in several places, most noteably in vm_fault. This
is not correct operation. The vm_pager_get_pages() code is now
responsible for marking VM pages all-valid. A number of VM helper
routines have been added to aid in zeroing-out the invalid portions of
a VM page prior to the page being marked all-valid. This operation is
necessary to properly support mmap(). The zeroing occurs most often
when dealing with file-EOF situations. Several bugs have been fixed
in the NFS subsystem, including bits handling file and directory EOF
situations and buf->b_flags consistancy issues relating to clearing
B_ERROR & B_INVAL, and handling B_DONE.
getblk() and allocbuf() have been rewritten. B_CACHE operation is now
formally defined in comments and more straightforward in
implementation. B_CACHE for VMIO buffers is based on the validity of
the backing store. B_CACHE for non-VMIO buffers is based simply on
whether the buffer is B_INVAL or not (B_CACHE set if B_INVAL clear,
and vise-versa). biodone() is now responsible for setting B_CACHE
when a successful read completes. B_CACHE is also set when a bdwrite()
is initiated and when a bwrite() is initiated. VFS VOP_BWRITE
routines (there are only two - nfs_bwrite() and bwrite()) are now
expected to set B_CACHE. This means that bowrite() and bawrite() also
set B_CACHE indirectly.
There are a number of places in the code which were previously using
buf->b_bufsize (which is DEV_BSIZE aligned) when they should have
been using buf->b_bcount. These have been fixed. getblk() now clears
B_DONE on return because the rest of the system is so bad about
dealing with B_DONE.
Major fixes to NFS/TCP have been made. A server-side bug could cause
requests to be lost by the server due to nfs_realign() overwriting
other rpc's in the same TCP mbuf chain. The server's kernel must be
recompiled to get the benefit of the fixes.
Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
1999-05-02 23:57:16 +00:00
|
|
|
static int nfs_realign_test;
|
|
|
|
|
static int nfs_realign_count;
|
|
|
|
|
|
2001-09-18 23:32:09 +00:00
|
|
|
SYSCTL_DECL(_vfs_nfsrv);
|
The VFS/BIO subsystem contained a number of hacks in order to optimize
piecemeal, middle-of-file writes for NFS. These hacks have caused no
end of trouble, especially when combined with mmap(). I've removed
them. Instead, NFS will issue a read-before-write to fully
instantiate the struct buf containing the write. NFS does, however,
optimize piecemeal appends to files. For most common file operations,
you will not notice the difference. The sole remaining fragment in
the VFS/BIO system is b_dirtyoff/end, which NFS uses to avoid cache
coherency issues with read-merge-write style operations. NFS also
optimizes the write-covers-entire-buffer case by avoiding the
read-before-write. There is quite a bit of room for further
optimization in these areas.
The VM system marks pages fully-valid (AKA vm_page_t->valid =
VM_PAGE_BITS_ALL) in several places, most noteably in vm_fault. This
is not correct operation. The vm_pager_get_pages() code is now
responsible for marking VM pages all-valid. A number of VM helper
routines have been added to aid in zeroing-out the invalid portions of
a VM page prior to the page being marked all-valid. This operation is
necessary to properly support mmap(). The zeroing occurs most often
when dealing with file-EOF situations. Several bugs have been fixed
in the NFS subsystem, including bits handling file and directory EOF
situations and buf->b_flags consistancy issues relating to clearing
B_ERROR & B_INVAL, and handling B_DONE.
getblk() and allocbuf() have been rewritten. B_CACHE operation is now
formally defined in comments and more straightforward in
implementation. B_CACHE for VMIO buffers is based on the validity of
the backing store. B_CACHE for non-VMIO buffers is based simply on
whether the buffer is B_INVAL or not (B_CACHE set if B_INVAL clear,
and vise-versa). biodone() is now responsible for setting B_CACHE
when a successful read completes. B_CACHE is also set when a bdwrite()
is initiated and when a bwrite() is initiated. VFS VOP_BWRITE
routines (there are only two - nfs_bwrite() and bwrite()) are now
expected to set B_CACHE. This means that bowrite() and bawrite() also
set B_CACHE indirectly.
There are a number of places in the code which were previously using
buf->b_bufsize (which is DEV_BSIZE aligned) when they should have
been using buf->b_bcount. These have been fixed. getblk() now clears
B_DONE on return because the rest of the system is so bad about
dealing with B_DONE.
Major fixes to NFS/TCP have been made. A server-side bug could cause
requests to be lost by the server due to nfs_realign() overwriting
other rpc's in the same TCP mbuf chain. The server's kernel must be
recompiled to get the benefit of the fixes.
Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
1999-05-02 23:57:16 +00:00
|
|
|
|
2001-09-18 23:32:09 +00:00
|
|
|
SYSCTL_INT(_vfs_nfsrv, OID_AUTO, realign_test, CTLFLAG_RW, &nfs_realign_test, 0, "");
|
|
|
|
|
SYSCTL_INT(_vfs_nfsrv, OID_AUTO, realign_count, CTLFLAG_RW, &nfs_realign_count, 0, "");
|
The VFS/BIO subsystem contained a number of hacks in order to optimize
piecemeal, middle-of-file writes for NFS. These hacks have caused no
end of trouble, especially when combined with mmap(). I've removed
them. Instead, NFS will issue a read-before-write to fully
instantiate the struct buf containing the write. NFS does, however,
optimize piecemeal appends to files. For most common file operations,
you will not notice the difference. The sole remaining fragment in
the VFS/BIO system is b_dirtyoff/end, which NFS uses to avoid cache
coherency issues with read-merge-write style operations. NFS also
optimizes the write-covers-entire-buffer case by avoiding the
read-before-write. There is quite a bit of room for further
optimization in these areas.
The VM system marks pages fully-valid (AKA vm_page_t->valid =
VM_PAGE_BITS_ALL) in several places, most noteably in vm_fault. This
is not correct operation. The vm_pager_get_pages() code is now
responsible for marking VM pages all-valid. A number of VM helper
routines have been added to aid in zeroing-out the invalid portions of
a VM page prior to the page being marked all-valid. This operation is
necessary to properly support mmap(). The zeroing occurs most often
when dealing with file-EOF situations. Several bugs have been fixed
in the NFS subsystem, including bits handling file and directory EOF
situations and buf->b_flags consistancy issues relating to clearing
B_ERROR & B_INVAL, and handling B_DONE.
getblk() and allocbuf() have been rewritten. B_CACHE operation is now
formally defined in comments and more straightforward in
implementation. B_CACHE for VMIO buffers is based on the validity of
the backing store. B_CACHE for non-VMIO buffers is based simply on
whether the buffer is B_INVAL or not (B_CACHE set if B_INVAL clear,
and vise-versa). biodone() is now responsible for setting B_CACHE
when a successful read completes. B_CACHE is also set when a bdwrite()
is initiated and when a bwrite() is initiated. VFS VOP_BWRITE
routines (there are only two - nfs_bwrite() and bwrite()) are now
expected to set B_CACHE. This means that bowrite() and bawrite() also
set B_CACHE indirectly.
There are a number of places in the code which were previously using
buf->b_bufsize (which is DEV_BSIZE aligned) when they should have
been using buf->b_bcount. These have been fixed. getblk() now clears
B_DONE on return because the rest of the system is so bad about
dealing with B_DONE.
Major fixes to NFS/TCP have been made. A server-side bug could cause
requests to be lost by the server due to nfs_realign() overwriting
other rpc's in the same TCP mbuf chain. The server's kernel must be
recompiled to get the benefit of the fixes.
Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
1999-05-02 23:57:16 +00:00
|
|
|
|
|
|
|
|
|
1994-05-24 10:09:53 +00:00
|
|
|
/*
|
|
|
|
|
* There is a congestion window for outstanding rpcs maintained per mount
|
|
|
|
|
* point. The cwnd size is adjusted in roughly the way that:
|
|
|
|
|
* Van Jacobson, Congestion avoidance and Control, In "Proceedings of
|
|
|
|
|
* SIGCOMM '88". ACM, August 1988.
|
|
|
|
|
* describes for TCP. The cwnd size is chopped in half on a retransmit timeout
|
|
|
|
|
* and incremented by 1/cwnd when each rpc reply is received and a full cwnd
|
|
|
|
|
* of rpcs is in progress.
|
|
|
|
|
* (The sent count and cwnd are scaled for integer arith.)
|
|
|
|
|
* Variants of "slow start" were tried and were found to be too much of a
|
|
|
|
|
* performance hit (ave. rtt 3 times larger),
|
|
|
|
|
* I suspect due to the large rtt that nfs rpcs have.
|
|
|
|
|
*/
|
|
|
|
|
#define NFS_CWNDSCALE 256
|
|
|
|
|
#define NFS_MAXCWND (NFS_CWNDSCALE * 32)
|
2004-03-07 16:28:31 +00:00
|
|
|
struct callout nfsrv_callout;
|
1994-05-24 10:09:53 +00:00
|
|
|
|
2001-09-18 23:32:09 +00:00
|
|
|
static void nfs_realign(struct mbuf **pm, int hsiz); /* XXX SHARED */
|
|
|
|
|
static int nfsrv_getstream(struct nfssvc_sock *, int);
|
1994-05-24 10:09:53 +00:00
|
|
|
|
The socket code upcalls into the NFS server using the so_upcall
mechanism so that early processing on mbufs can be performed before
a context switch to the NFS server threads. Because of this, if
the socket code is running without Giant, the NFS server also needs
to be able to run the upcall code without relying on the presence on
Giant. This change modifies the NFS server to run using a "giant
code lock" covering operation of the whole subsystem. Work is in
progress to move to data-based locking as part of the NFSv4 server
changes.
Introduce an NFS server subsystem lock, 'nfsd_mtx', and a set of
macros to operate on the lock:
NFSD_LOCK_ASSERT() Assert nfsd_mtx owned by current thread
NFSD_UNLOCK_ASSERT() Assert nfsd_mtx not owned by current thread
NFSD_LOCK_DONTCARE() Advisory: this function doesn't care
NFSD_LOCK() Lock nfsd_mtx
NFSD_UNLOCK() Unlock nfsd_mtx
Constify a number of global variables/structures in the NFS server
code, as they are not modified and contain constants only:
nfsrvv2_procid nfsrv_nfsv3_procid nonidempotent
nfsv2_repstat nfsv2_type nfsrv_nfsv3_procid
nfsrvv2_procid nfsrv_v2errmap nfsv3err_null
nfsv3err_getattr nfsv3err_setattr nfsv3err_lookup
nfsv3err_access nfsv3err_readlink nfsv3err_read
nfsv3err_write nfsv3err_create nfsv3err_mkdir
nfsv3err_symlink nfsv3err_mknod nfsv3err_remove
nfsv3err_rmdir nfsv3err_rename nfsv3err_link
nfsv3err_readdir nfsv3err_readdirplus nfsv3err_fsstat
nfsv3err_fsinfo nfsv3err_pathconf nfsv3err_commit
nfsrv_v3errmap
There are additional structures that should be constified but due
to their being passed into general purpose functions without const
arguments, I have not yet converted.
In general, acquire nfsd_mtx when accessing any of the global NFS
structures, including struct nfssvc_sock, struct nfsd, struct
nfsrv_descript.
Release nfsd_mtx whenever calling into VFS, and acquire Giant for
calls into VFS. Giant is not required for any part of the
operation of the NFS server with the exception of calls into VFS.
Giant will never by acquired in the upcall code path. However, it
may operate entirely covered by Giant, or not. If debug.mpsafenet
is set to 0, the system calls will acquire Giant across all
operations, and the upcall will assert Giant. As such, by default,
this enables locking and allows us to test assertions, but should not
cause any substantial new amount of code to be run without Giant.
Bugs should manifest in the form of lock assertion failures for now.
This approach is similar (but not identical) to modifications to the
BSD/OS NFS server code snapshot provided by BSDi as part of their
SMPng snapshot. The strategy is almost the same (single lock over
the NFS server), but differs in the following ways:
- Our NFS client and server code bases don't overlap, which means
both fewer bugs and easier locking (thanks Peter!). Also means
NFSD_*() as opposed to NFS_*().
- We make broad use of assertions, whereas the BSD/OS code does not.
- Made slightly different choices about how to handle macros building
packets but operating with side effects.
- We acquire Giant only when entering VFS from the NFS server daemon
threads.
- Serious bugs in BSD/OS implementation corrected -- the snapshot we
received was clearly a work in progress.
Based on ideas from: BSDi SMPng Snapshot
Reviewed by: rick@snowhite.cis.uoguelph.ca
Extensive testing by: kris
2004-05-24 04:06:14 +00:00
|
|
|
int32_t (*nfsrv3_procs[NFS_NPROCS])(struct nfsrv_descript *nd,
|
2001-09-18 23:32:09 +00:00
|
|
|
struct nfssvc_sock *slp,
|
|
|
|
|
struct mbuf **mreqp) = {
|
1994-05-24 10:09:53 +00:00
|
|
|
nfsrv_null,
|
|
|
|
|
nfsrv_getattr,
|
|
|
|
|
nfsrv_setattr,
|
|
|
|
|
nfsrv_lookup,
|
1995-06-27 11:07:30 +00:00
|
|
|
nfsrv3_access,
|
1994-05-24 10:09:53 +00:00
|
|
|
nfsrv_readlink,
|
|
|
|
|
nfsrv_read,
|
|
|
|
|
nfsrv_write,
|
|
|
|
|
nfsrv_create,
|
1995-06-27 11:07:30 +00:00
|
|
|
nfsrv_mkdir,
|
|
|
|
|
nfsrv_symlink,
|
|
|
|
|
nfsrv_mknod,
|
1994-05-24 10:09:53 +00:00
|
|
|
nfsrv_remove,
|
1995-06-27 11:07:30 +00:00
|
|
|
nfsrv_rmdir,
|
1994-05-24 10:09:53 +00:00
|
|
|
nfsrv_rename,
|
|
|
|
|
nfsrv_link,
|
|
|
|
|
nfsrv_readdir,
|
1995-06-27 11:07:30 +00:00
|
|
|
nfsrv_readdirplus,
|
1994-05-24 10:09:53 +00:00
|
|
|
nfsrv_statfs,
|
1995-06-27 11:07:30 +00:00
|
|
|
nfsrv_fsinfo,
|
|
|
|
|
nfsrv_pathconf,
|
|
|
|
|
nfsrv_commit,
|
|
|
|
|
nfsrv_noop
|
1994-05-24 10:09:53 +00:00
|
|
|
};
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
* Generate the rpc reply header
|
|
|
|
|
* siz arg. is used to decide if adding a cluster is worthwhile
|
|
|
|
|
*/
|
2001-09-28 04:37:08 +00:00
|
|
|
struct mbuf *
|
|
|
|
|
nfs_rephead(int siz, struct nfsrv_descript *nd, int err,
|
|
|
|
|
struct mbuf **mbp, caddr_t *bposp)
|
1994-05-24 10:09:53 +00:00
|
|
|
{
|
2001-09-18 23:32:09 +00:00
|
|
|
u_int32_t *tl;
|
|
|
|
|
struct mbuf *mreq;
|
1994-05-24 10:09:53 +00:00
|
|
|
caddr_t bpos;
|
2001-09-18 23:32:09 +00:00
|
|
|
struct mbuf *mb;
|
|
|
|
|
|
|
|
|
|
nd->nd_repstat = err;
|
|
|
|
|
if (err && (nd->nd_flag & ND_NFSV3) == 0) /* XXX recheck */
|
|
|
|
|
siz = 0;
|
2008-03-25 09:39:02 +00:00
|
|
|
MGETHDR(mreq, M_WAIT, MT_DATA);
|
1994-05-24 10:09:53 +00:00
|
|
|
mb = mreq;
|
|
|
|
|
/*
|
|
|
|
|
* If this is a big reply, use a cluster else
|
|
|
|
|
* try and leave leading space for the lower level headers.
|
|
|
|
|
*/
|
2002-07-15 19:40:23 +00:00
|
|
|
mreq->m_len = 6 * NFSX_UNSIGNED;
|
1994-05-24 10:09:53 +00:00
|
|
|
siz += RPC_REPLYSIZ;
|
2000-01-09 19:17:30 +00:00
|
|
|
if ((max_hdr + siz) >= MINCLSIZE) {
|
2008-03-25 09:39:02 +00:00
|
|
|
MCLGET(mreq, M_WAIT);
|
1994-05-24 10:09:53 +00:00
|
|
|
} else
|
2002-07-15 19:40:23 +00:00
|
|
|
mreq->m_data += min(max_hdr, M_TRAILINGSPACE(mreq));
|
1998-05-31 20:09:01 +00:00
|
|
|
tl = mtod(mreq, u_int32_t *);
|
1995-06-27 11:07:30 +00:00
|
|
|
bpos = ((caddr_t)tl) + mreq->m_len;
|
1994-10-17 17:47:45 +00:00
|
|
|
*tl++ = txdr_unsigned(nd->nd_retxid);
|
2002-01-08 19:41:06 +00:00
|
|
|
*tl++ = nfsrv_rpc_reply;
|
1995-06-27 11:07:30 +00:00
|
|
|
if (err == ERPCMISMATCH || (err & NFSERR_AUTHERR)) {
|
2002-01-08 19:41:06 +00:00
|
|
|
*tl++ = nfsrv_rpc_msgdenied;
|
1995-06-27 11:07:30 +00:00
|
|
|
if (err & NFSERR_AUTHERR) {
|
2002-01-08 19:41:06 +00:00
|
|
|
*tl++ = nfsrv_rpc_autherr;
|
1995-06-27 11:07:30 +00:00
|
|
|
*tl = txdr_unsigned(err & ~NFSERR_AUTHERR);
|
1994-05-24 10:09:53 +00:00
|
|
|
mreq->m_len -= NFSX_UNSIGNED;
|
|
|
|
|
bpos -= NFSX_UNSIGNED;
|
|
|
|
|
} else {
|
2002-01-08 19:41:06 +00:00
|
|
|
*tl++ = nfsrv_rpc_mismatch;
|
1995-06-27 11:07:30 +00:00
|
|
|
*tl++ = txdr_unsigned(RPC_VER2);
|
|
|
|
|
*tl = txdr_unsigned(RPC_VER2);
|
1994-05-24 10:09:53 +00:00
|
|
|
}
|
|
|
|
|
} else {
|
2002-01-08 19:41:06 +00:00
|
|
|
*tl++ = nfsrv_rpc_msgaccepted;
|
1995-06-27 11:07:30 +00:00
|
|
|
/*
|
2001-09-18 23:32:09 +00:00
|
|
|
* Send a RPCAUTH_NULL verifier - no Kerberos.
|
1995-06-27 11:07:30 +00:00
|
|
|
*/
|
2001-09-18 23:32:09 +00:00
|
|
|
*tl++ = 0;
|
|
|
|
|
*tl++ = 0;
|
1994-05-24 10:09:53 +00:00
|
|
|
switch (err) {
|
|
|
|
|
case EPROGUNAVAIL:
|
|
|
|
|
*tl = txdr_unsigned(RPC_PROGUNAVAIL);
|
|
|
|
|
break;
|
|
|
|
|
case EPROGMISMATCH:
|
|
|
|
|
*tl = txdr_unsigned(RPC_PROGMISMATCH);
|
2001-09-27 02:33:36 +00:00
|
|
|
tl = nfsm_build(u_int32_t *, 2 * NFSX_UNSIGNED);
|
2001-09-18 23:32:09 +00:00
|
|
|
*tl++ = txdr_unsigned(2);
|
|
|
|
|
*tl = txdr_unsigned(3);
|
1994-05-24 10:09:53 +00:00
|
|
|
break;
|
|
|
|
|
case EPROCUNAVAIL:
|
|
|
|
|
*tl = txdr_unsigned(RPC_PROCUNAVAIL);
|
|
|
|
|
break;
|
1995-06-27 11:07:30 +00:00
|
|
|
case EBADRPC:
|
|
|
|
|
*tl = txdr_unsigned(RPC_GARBAGE);
|
|
|
|
|
break;
|
1994-05-24 10:09:53 +00:00
|
|
|
default:
|
|
|
|
|
*tl = 0;
|
1995-06-27 11:07:30 +00:00
|
|
|
if (err != NFSERR_RETVOID) {
|
2001-09-27 02:33:36 +00:00
|
|
|
tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
|
1994-05-24 10:09:53 +00:00
|
|
|
if (err)
|
1995-06-27 11:07:30 +00:00
|
|
|
*tl = txdr_unsigned(nfsrv_errmap(nd, err));
|
1994-05-24 10:09:53 +00:00
|
|
|
else
|
1995-06-27 11:07:30 +00:00
|
|
|
*tl = 0;
|
1994-05-24 10:09:53 +00:00
|
|
|
}
|
|
|
|
|
break;
|
2001-09-28 04:37:08 +00:00
|
|
|
}
|
1994-05-24 10:09:53 +00:00
|
|
|
}
|
|
|
|
|
*mbp = mb;
|
|
|
|
|
*bposp = bpos;
|
1995-06-27 11:07:30 +00:00
|
|
|
if (err != 0 && err != NFSERR_RETVOID)
|
2001-09-18 23:32:09 +00:00
|
|
|
nfsrvstats.srvrpc_errs++;
|
2001-09-28 04:37:08 +00:00
|
|
|
return mreq;
|
1994-05-24 10:09:53 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
/*
|
The VFS/BIO subsystem contained a number of hacks in order to optimize
piecemeal, middle-of-file writes for NFS. These hacks have caused no
end of trouble, especially when combined with mmap(). I've removed
them. Instead, NFS will issue a read-before-write to fully
instantiate the struct buf containing the write. NFS does, however,
optimize piecemeal appends to files. For most common file operations,
you will not notice the difference. The sole remaining fragment in
the VFS/BIO system is b_dirtyoff/end, which NFS uses to avoid cache
coherency issues with read-merge-write style operations. NFS also
optimizes the write-covers-entire-buffer case by avoiding the
read-before-write. There is quite a bit of room for further
optimization in these areas.
The VM system marks pages fully-valid (AKA vm_page_t->valid =
VM_PAGE_BITS_ALL) in several places, most noteably in vm_fault. This
is not correct operation. The vm_pager_get_pages() code is now
responsible for marking VM pages all-valid. A number of VM helper
routines have been added to aid in zeroing-out the invalid portions of
a VM page prior to the page being marked all-valid. This operation is
necessary to properly support mmap(). The zeroing occurs most often
when dealing with file-EOF situations. Several bugs have been fixed
in the NFS subsystem, including bits handling file and directory EOF
situations and buf->b_flags consistancy issues relating to clearing
B_ERROR & B_INVAL, and handling B_DONE.
getblk() and allocbuf() have been rewritten. B_CACHE operation is now
formally defined in comments and more straightforward in
implementation. B_CACHE for VMIO buffers is based on the validity of
the backing store. B_CACHE for non-VMIO buffers is based simply on
whether the buffer is B_INVAL or not (B_CACHE set if B_INVAL clear,
and vise-versa). biodone() is now responsible for setting B_CACHE
when a successful read completes. B_CACHE is also set when a bdwrite()
is initiated and when a bwrite() is initiated. VFS VOP_BWRITE
routines (there are only two - nfs_bwrite() and bwrite()) are now
expected to set B_CACHE. This means that bowrite() and bawrite() also
set B_CACHE indirectly.
There are a number of places in the code which were previously using
buf->b_bufsize (which is DEV_BSIZE aligned) when they should have
been using buf->b_bcount. These have been fixed. getblk() now clears
B_DONE on return because the rest of the system is so bad about
dealing with B_DONE.
Major fixes to NFS/TCP have been made. A server-side bug could cause
requests to be lost by the server due to nfs_realign() overwriting
other rpc's in the same TCP mbuf chain. The server's kernel must be
recompiled to get the benefit of the fixes.
Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
1999-05-02 23:57:16 +00:00
|
|
|
* nfs_realign:
|
|
|
|
|
*
|
|
|
|
|
* Check for badly aligned mbuf data and realign by copying the unaligned
|
|
|
|
|
* portion of the data into a new mbuf chain and freeing the portions
|
|
|
|
|
* of the old chain that were replaced.
|
|
|
|
|
*
|
|
|
|
|
* We cannot simply realign the data within the existing mbuf chain
|
|
|
|
|
* because the underlying buffers may contain other rpc commands and
|
|
|
|
|
* we cannot afford to overwrite them.
|
|
|
|
|
*
|
|
|
|
|
* We would prefer to avoid this situation entirely. The situation does
|
|
|
|
|
* not occur with NFS/UDP and is supposed to only occassionally occur
|
|
|
|
|
* with TCP. Use vfs.nfs.realign_count and realign_test to check this.
|
1994-05-24 10:09:53 +00:00
|
|
|
*/
|
1995-12-17 21:14:36 +00:00
|
|
|
static void
|
2001-09-18 23:32:09 +00:00
|
|
|
nfs_realign(struct mbuf **pm, int hsiz) /* XXX COMMON */
|
1994-05-24 10:09:53 +00:00
|
|
|
{
|
The VFS/BIO subsystem contained a number of hacks in order to optimize
piecemeal, middle-of-file writes for NFS. These hacks have caused no
end of trouble, especially when combined with mmap(). I've removed
them. Instead, NFS will issue a read-before-write to fully
instantiate the struct buf containing the write. NFS does, however,
optimize piecemeal appends to files. For most common file operations,
you will not notice the difference. The sole remaining fragment in
the VFS/BIO system is b_dirtyoff/end, which NFS uses to avoid cache
coherency issues with read-merge-write style operations. NFS also
optimizes the write-covers-entire-buffer case by avoiding the
read-before-write. There is quite a bit of room for further
optimization in these areas.
The VM system marks pages fully-valid (AKA vm_page_t->valid =
VM_PAGE_BITS_ALL) in several places, most noteably in vm_fault. This
is not correct operation. The vm_pager_get_pages() code is now
responsible for marking VM pages all-valid. A number of VM helper
routines have been added to aid in zeroing-out the invalid portions of
a VM page prior to the page being marked all-valid. This operation is
necessary to properly support mmap(). The zeroing occurs most often
when dealing with file-EOF situations. Several bugs have been fixed
in the NFS subsystem, including bits handling file and directory EOF
situations and buf->b_flags consistancy issues relating to clearing
B_ERROR & B_INVAL, and handling B_DONE.
getblk() and allocbuf() have been rewritten. B_CACHE operation is now
formally defined in comments and more straightforward in
implementation. B_CACHE for VMIO buffers is based on the validity of
the backing store. B_CACHE for non-VMIO buffers is based simply on
whether the buffer is B_INVAL or not (B_CACHE set if B_INVAL clear,
and vise-versa). biodone() is now responsible for setting B_CACHE
when a successful read completes. B_CACHE is also set when a bdwrite()
is initiated and when a bwrite() is initiated. VFS VOP_BWRITE
routines (there are only two - nfs_bwrite() and bwrite()) are now
expected to set B_CACHE. This means that bowrite() and bawrite() also
set B_CACHE indirectly.
There are a number of places in the code which were previously using
buf->b_bufsize (which is DEV_BSIZE aligned) when they should have
been using buf->b_bcount. These have been fixed. getblk() now clears
B_DONE on return because the rest of the system is so bad about
dealing with B_DONE.
Major fixes to NFS/TCP have been made. A server-side bug could cause
requests to be lost by the server due to nfs_realign() overwriting
other rpc's in the same TCP mbuf chain. The server's kernel must be
recompiled to get the benefit of the fixes.
Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
1999-05-02 23:57:16 +00:00
|
|
|
struct mbuf *m;
|
|
|
|
|
struct mbuf *n = NULL;
|
|
|
|
|
int off = 0;
|
1994-05-24 10:09:53 +00:00
|
|
|
|
The VFS/BIO subsystem contained a number of hacks in order to optimize
piecemeal, middle-of-file writes for NFS. These hacks have caused no
end of trouble, especially when combined with mmap(). I've removed
them. Instead, NFS will issue a read-before-write to fully
instantiate the struct buf containing the write. NFS does, however,
optimize piecemeal appends to files. For most common file operations,
you will not notice the difference. The sole remaining fragment in
the VFS/BIO system is b_dirtyoff/end, which NFS uses to avoid cache
coherency issues with read-merge-write style operations. NFS also
optimizes the write-covers-entire-buffer case by avoiding the
read-before-write. There is quite a bit of room for further
optimization in these areas.
The VM system marks pages fully-valid (AKA vm_page_t->valid =
VM_PAGE_BITS_ALL) in several places, most noteably in vm_fault. This
is not correct operation. The vm_pager_get_pages() code is now
responsible for marking VM pages all-valid. A number of VM helper
routines have been added to aid in zeroing-out the invalid portions of
a VM page prior to the page being marked all-valid. This operation is
necessary to properly support mmap(). The zeroing occurs most often
when dealing with file-EOF situations. Several bugs have been fixed
in the NFS subsystem, including bits handling file and directory EOF
situations and buf->b_flags consistancy issues relating to clearing
B_ERROR & B_INVAL, and handling B_DONE.
getblk() and allocbuf() have been rewritten. B_CACHE operation is now
formally defined in comments and more straightforward in
implementation. B_CACHE for VMIO buffers is based on the validity of
the backing store. B_CACHE for non-VMIO buffers is based simply on
whether the buffer is B_INVAL or not (B_CACHE set if B_INVAL clear,
and vise-versa). biodone() is now responsible for setting B_CACHE
when a successful read completes. B_CACHE is also set when a bdwrite()
is initiated and when a bwrite() is initiated. VFS VOP_BWRITE
routines (there are only two - nfs_bwrite() and bwrite()) are now
expected to set B_CACHE. This means that bowrite() and bawrite() also
set B_CACHE indirectly.
There are a number of places in the code which were previously using
buf->b_bufsize (which is DEV_BSIZE aligned) when they should have
been using buf->b_bcount. These have been fixed. getblk() now clears
B_DONE on return because the rest of the system is so bad about
dealing with B_DONE.
Major fixes to NFS/TCP have been made. A server-side bug could cause
requests to be lost by the server due to nfs_realign() overwriting
other rpc's in the same TCP mbuf chain. The server's kernel must be
recompiled to get the benefit of the fixes.
Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
1999-05-02 23:57:16 +00:00
|
|
|
++nfs_realign_test;
|
|
|
|
|
while ((m = *pm) != NULL) {
|
|
|
|
|
if ((m->m_len & 0x3) || (mtod(m, intptr_t) & 0x3)) {
|
2008-03-25 09:39:02 +00:00
|
|
|
MGET(n, M_WAIT, MT_DATA);
|
The VFS/BIO subsystem contained a number of hacks in order to optimize
piecemeal, middle-of-file writes for NFS. These hacks have caused no
end of trouble, especially when combined with mmap(). I've removed
them. Instead, NFS will issue a read-before-write to fully
instantiate the struct buf containing the write. NFS does, however,
optimize piecemeal appends to files. For most common file operations,
you will not notice the difference. The sole remaining fragment in
the VFS/BIO system is b_dirtyoff/end, which NFS uses to avoid cache
coherency issues with read-merge-write style operations. NFS also
optimizes the write-covers-entire-buffer case by avoiding the
read-before-write. There is quite a bit of room for further
optimization in these areas.
The VM system marks pages fully-valid (AKA vm_page_t->valid =
VM_PAGE_BITS_ALL) in several places, most noteably in vm_fault. This
is not correct operation. The vm_pager_get_pages() code is now
responsible for marking VM pages all-valid. A number of VM helper
routines have been added to aid in zeroing-out the invalid portions of
a VM page prior to the page being marked all-valid. This operation is
necessary to properly support mmap(). The zeroing occurs most often
when dealing with file-EOF situations. Several bugs have been fixed
in the NFS subsystem, including bits handling file and directory EOF
situations and buf->b_flags consistancy issues relating to clearing
B_ERROR & B_INVAL, and handling B_DONE.
getblk() and allocbuf() have been rewritten. B_CACHE operation is now
formally defined in comments and more straightforward in
implementation. B_CACHE for VMIO buffers is based on the validity of
the backing store. B_CACHE for non-VMIO buffers is based simply on
whether the buffer is B_INVAL or not (B_CACHE set if B_INVAL clear,
and vise-versa). biodone() is now responsible for setting B_CACHE
when a successful read completes. B_CACHE is also set when a bdwrite()
is initiated and when a bwrite() is initiated. VFS VOP_BWRITE
routines (there are only two - nfs_bwrite() and bwrite()) are now
expected to set B_CACHE. This means that bowrite() and bawrite() also
set B_CACHE indirectly.
There are a number of places in the code which were previously using
buf->b_bufsize (which is DEV_BSIZE aligned) when they should have
been using buf->b_bcount. These have been fixed. getblk() now clears
B_DONE on return because the rest of the system is so bad about
dealing with B_DONE.
Major fixes to NFS/TCP have been made. A server-side bug could cause
requests to be lost by the server due to nfs_realign() overwriting
other rpc's in the same TCP mbuf chain. The server's kernel must be
recompiled to get the benefit of the fixes.
Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
1999-05-02 23:57:16 +00:00
|
|
|
if (m->m_len >= MINCLSIZE) {
|
2008-03-25 09:39:02 +00:00
|
|
|
MCLGET(n, M_WAIT);
|
1994-05-24 10:09:53 +00:00
|
|
|
}
|
The VFS/BIO subsystem contained a number of hacks in order to optimize
piecemeal, middle-of-file writes for NFS. These hacks have caused no
end of trouble, especially when combined with mmap(). I've removed
them. Instead, NFS will issue a read-before-write to fully
instantiate the struct buf containing the write. NFS does, however,
optimize piecemeal appends to files. For most common file operations,
you will not notice the difference. The sole remaining fragment in
the VFS/BIO system is b_dirtyoff/end, which NFS uses to avoid cache
coherency issues with read-merge-write style operations. NFS also
optimizes the write-covers-entire-buffer case by avoiding the
read-before-write. There is quite a bit of room for further
optimization in these areas.
The VM system marks pages fully-valid (AKA vm_page_t->valid =
VM_PAGE_BITS_ALL) in several places, most noteably in vm_fault. This
is not correct operation. The vm_pager_get_pages() code is now
responsible for marking VM pages all-valid. A number of VM helper
routines have been added to aid in zeroing-out the invalid portions of
a VM page prior to the page being marked all-valid. This operation is
necessary to properly support mmap(). The zeroing occurs most often
when dealing with file-EOF situations. Several bugs have been fixed
in the NFS subsystem, including bits handling file and directory EOF
situations and buf->b_flags consistancy issues relating to clearing
B_ERROR & B_INVAL, and handling B_DONE.
getblk() and allocbuf() have been rewritten. B_CACHE operation is now
formally defined in comments and more straightforward in
implementation. B_CACHE for VMIO buffers is based on the validity of
the backing store. B_CACHE for non-VMIO buffers is based simply on
whether the buffer is B_INVAL or not (B_CACHE set if B_INVAL clear,
and vise-versa). biodone() is now responsible for setting B_CACHE
when a successful read completes. B_CACHE is also set when a bdwrite()
is initiated and when a bwrite() is initiated. VFS VOP_BWRITE
routines (there are only two - nfs_bwrite() and bwrite()) are now
expected to set B_CACHE. This means that bowrite() and bawrite() also
set B_CACHE indirectly.
There are a number of places in the code which were previously using
buf->b_bufsize (which is DEV_BSIZE aligned) when they should have
been using buf->b_bcount. These have been fixed. getblk() now clears
B_DONE on return because the rest of the system is so bad about
dealing with B_DONE.
Major fixes to NFS/TCP have been made. A server-side bug could cause
requests to be lost by the server due to nfs_realign() overwriting
other rpc's in the same TCP mbuf chain. The server's kernel must be
recompiled to get the benefit of the fixes.
Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
1999-05-02 23:57:16 +00:00
|
|
|
n->m_len = 0;
|
|
|
|
|
break;
|
1994-05-24 10:09:53 +00:00
|
|
|
}
|
The VFS/BIO subsystem contained a number of hacks in order to optimize
piecemeal, middle-of-file writes for NFS. These hacks have caused no
end of trouble, especially when combined with mmap(). I've removed
them. Instead, NFS will issue a read-before-write to fully
instantiate the struct buf containing the write. NFS does, however,
optimize piecemeal appends to files. For most common file operations,
you will not notice the difference. The sole remaining fragment in
the VFS/BIO system is b_dirtyoff/end, which NFS uses to avoid cache
coherency issues with read-merge-write style operations. NFS also
optimizes the write-covers-entire-buffer case by avoiding the
read-before-write. There is quite a bit of room for further
optimization in these areas.
The VM system marks pages fully-valid (AKA vm_page_t->valid =
VM_PAGE_BITS_ALL) in several places, most noteably in vm_fault. This
is not correct operation. The vm_pager_get_pages() code is now
responsible for marking VM pages all-valid. A number of VM helper
routines have been added to aid in zeroing-out the invalid portions of
a VM page prior to the page being marked all-valid. This operation is
necessary to properly support mmap(). The zeroing occurs most often
when dealing with file-EOF situations. Several bugs have been fixed
in the NFS subsystem, including bits handling file and directory EOF
situations and buf->b_flags consistancy issues relating to clearing
B_ERROR & B_INVAL, and handling B_DONE.
getblk() and allocbuf() have been rewritten. B_CACHE operation is now
formally defined in comments and more straightforward in
implementation. B_CACHE for VMIO buffers is based on the validity of
the backing store. B_CACHE for non-VMIO buffers is based simply on
whether the buffer is B_INVAL or not (B_CACHE set if B_INVAL clear,
and vise-versa). biodone() is now responsible for setting B_CACHE
when a successful read completes. B_CACHE is also set when a bdwrite()
is initiated and when a bwrite() is initiated. VFS VOP_BWRITE
routines (there are only two - nfs_bwrite() and bwrite()) are now
expected to set B_CACHE. This means that bowrite() and bawrite() also
set B_CACHE indirectly.
There are a number of places in the code which were previously using
buf->b_bufsize (which is DEV_BSIZE aligned) when they should have
been using buf->b_bcount. These have been fixed. getblk() now clears
B_DONE on return because the rest of the system is so bad about
dealing with B_DONE.
Major fixes to NFS/TCP have been made. A server-side bug could cause
requests to be lost by the server due to nfs_realign() overwriting
other rpc's in the same TCP mbuf chain. The server's kernel must be
recompiled to get the benefit of the fixes.
Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
1999-05-02 23:57:16 +00:00
|
|
|
pm = &m->m_next;
|
|
|
|
|
}
|
1995-05-30 08:16:23 +00:00
|
|
|
|
The VFS/BIO subsystem contained a number of hacks in order to optimize
piecemeal, middle-of-file writes for NFS. These hacks have caused no
end of trouble, especially when combined with mmap(). I've removed
them. Instead, NFS will issue a read-before-write to fully
instantiate the struct buf containing the write. NFS does, however,
optimize piecemeal appends to files. For most common file operations,
you will not notice the difference. The sole remaining fragment in
the VFS/BIO system is b_dirtyoff/end, which NFS uses to avoid cache
coherency issues with read-merge-write style operations. NFS also
optimizes the write-covers-entire-buffer case by avoiding the
read-before-write. There is quite a bit of room for further
optimization in these areas.
The VM system marks pages fully-valid (AKA vm_page_t->valid =
VM_PAGE_BITS_ALL) in several places, most noteably in vm_fault. This
is not correct operation. The vm_pager_get_pages() code is now
responsible for marking VM pages all-valid. A number of VM helper
routines have been added to aid in zeroing-out the invalid portions of
a VM page prior to the page being marked all-valid. This operation is
necessary to properly support mmap(). The zeroing occurs most often
when dealing with file-EOF situations. Several bugs have been fixed
in the NFS subsystem, including bits handling file and directory EOF
situations and buf->b_flags consistancy issues relating to clearing
B_ERROR & B_INVAL, and handling B_DONE.
getblk() and allocbuf() have been rewritten. B_CACHE operation is now
formally defined in comments and more straightforward in
implementation. B_CACHE for VMIO buffers is based on the validity of
the backing store. B_CACHE for non-VMIO buffers is based simply on
whether the buffer is B_INVAL or not (B_CACHE set if B_INVAL clear,
and vise-versa). biodone() is now responsible for setting B_CACHE
when a successful read completes. B_CACHE is also set when a bdwrite()
is initiated and when a bwrite() is initiated. VFS VOP_BWRITE
routines (there are only two - nfs_bwrite() and bwrite()) are now
expected to set B_CACHE. This means that bowrite() and bawrite() also
set B_CACHE indirectly.
There are a number of places in the code which were previously using
buf->b_bufsize (which is DEV_BSIZE aligned) when they should have
been using buf->b_bcount. These have been fixed. getblk() now clears
B_DONE on return because the rest of the system is so bad about
dealing with B_DONE.
Major fixes to NFS/TCP have been made. A server-side bug could cause
requests to be lost by the server due to nfs_realign() overwriting
other rpc's in the same TCP mbuf chain. The server's kernel must be
recompiled to get the benefit of the fixes.
Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
1999-05-02 23:57:16 +00:00
|
|
|
/*
|
|
|
|
|
* If n is non-NULL, loop on m copying data, then replace the
|
|
|
|
|
* portion of the chain that had to be realigned.
|
|
|
|
|
*/
|
|
|
|
|
if (n != NULL) {
|
|
|
|
|
++nfs_realign_count;
|
|
|
|
|
while (m) {
|
|
|
|
|
m_copyback(n, off, m->m_len, mtod(m, caddr_t));
|
|
|
|
|
off += m->m_len;
|
|
|
|
|
m = m->m_next;
|
1994-05-24 10:09:53 +00:00
|
|
|
}
|
The VFS/BIO subsystem contained a number of hacks in order to optimize
piecemeal, middle-of-file writes for NFS. These hacks have caused no
end of trouble, especially when combined with mmap(). I've removed
them. Instead, NFS will issue a read-before-write to fully
instantiate the struct buf containing the write. NFS does, however,
optimize piecemeal appends to files. For most common file operations,
you will not notice the difference. The sole remaining fragment in
the VFS/BIO system is b_dirtyoff/end, which NFS uses to avoid cache
coherency issues with read-merge-write style operations. NFS also
optimizes the write-covers-entire-buffer case by avoiding the
read-before-write. There is quite a bit of room for further
optimization in these areas.
The VM system marks pages fully-valid (AKA vm_page_t->valid =
VM_PAGE_BITS_ALL) in several places, most noteably in vm_fault. This
is not correct operation. The vm_pager_get_pages() code is now
responsible for marking VM pages all-valid. A number of VM helper
routines have been added to aid in zeroing-out the invalid portions of
a VM page prior to the page being marked all-valid. This operation is
necessary to properly support mmap(). The zeroing occurs most often
when dealing with file-EOF situations. Several bugs have been fixed
in the NFS subsystem, including bits handling file and directory EOF
situations and buf->b_flags consistancy issues relating to clearing
B_ERROR & B_INVAL, and handling B_DONE.
getblk() and allocbuf() have been rewritten. B_CACHE operation is now
formally defined in comments and more straightforward in
implementation. B_CACHE for VMIO buffers is based on the validity of
the backing store. B_CACHE for non-VMIO buffers is based simply on
whether the buffer is B_INVAL or not (B_CACHE set if B_INVAL clear,
and vise-versa). biodone() is now responsible for setting B_CACHE
when a successful read completes. B_CACHE is also set when a bdwrite()
is initiated and when a bwrite() is initiated. VFS VOP_BWRITE
routines (there are only two - nfs_bwrite() and bwrite()) are now
expected to set B_CACHE. This means that bowrite() and bawrite() also
set B_CACHE indirectly.
There are a number of places in the code which were previously using
buf->b_bufsize (which is DEV_BSIZE aligned) when they should have
been using buf->b_bcount. These have been fixed. getblk() now clears
B_DONE on return because the rest of the system is so bad about
dealing with B_DONE.
Major fixes to NFS/TCP have been made. A server-side bug could cause
requests to be lost by the server due to nfs_realign() overwriting
other rpc's in the same TCP mbuf chain. The server's kernel must be
recompiled to get the benefit of the fixes.
Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
1999-05-02 23:57:16 +00:00
|
|
|
m_freem(*pm);
|
|
|
|
|
*pm = n;
|
1994-05-24 10:09:53 +00:00
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
* Parse an RPC request
|
|
|
|
|
* - verify it
|
|
|
|
|
* - fill in the cred struct.
|
|
|
|
|
*/
|
1994-05-25 09:21:21 +00:00
|
|
|
int
|
2001-09-18 23:32:09 +00:00
|
|
|
nfs_getreq(struct nfsrv_descript *nd, struct nfsd *nfsd, int has_header)
|
1994-05-24 10:09:53 +00:00
|
|
|
{
|
2001-09-18 23:32:09 +00:00
|
|
|
int len, i;
|
|
|
|
|
u_int32_t *tl;
|
|
|
|
|
caddr_t dpos;
|
1998-05-31 20:09:01 +00:00
|
|
|
u_int32_t nfsvers, auth_type;
|
2001-09-18 23:32:09 +00:00
|
|
|
int error = 0;
|
1994-05-24 10:09:53 +00:00
|
|
|
struct mbuf *mrep, *md;
|
|
|
|
|
|
The socket code upcalls into the NFS server using the so_upcall
mechanism so that early processing on mbufs can be performed before
a context switch to the NFS server threads. Because of this, if
the socket code is running without Giant, the NFS server also needs
to be able to run the upcall code without relying on the presence on
Giant. This change modifies the NFS server to run using a "giant
code lock" covering operation of the whole subsystem. Work is in
progress to move to data-based locking as part of the NFSv4 server
changes.
Introduce an NFS server subsystem lock, 'nfsd_mtx', and a set of
macros to operate on the lock:
NFSD_LOCK_ASSERT() Assert nfsd_mtx owned by current thread
NFSD_UNLOCK_ASSERT() Assert nfsd_mtx not owned by current thread
NFSD_LOCK_DONTCARE() Advisory: this function doesn't care
NFSD_LOCK() Lock nfsd_mtx
NFSD_UNLOCK() Unlock nfsd_mtx
Constify a number of global variables/structures in the NFS server
code, as they are not modified and contain constants only:
nfsrvv2_procid nfsrv_nfsv3_procid nonidempotent
nfsv2_repstat nfsv2_type nfsrv_nfsv3_procid
nfsrvv2_procid nfsrv_v2errmap nfsv3err_null
nfsv3err_getattr nfsv3err_setattr nfsv3err_lookup
nfsv3err_access nfsv3err_readlink nfsv3err_read
nfsv3err_write nfsv3err_create nfsv3err_mkdir
nfsv3err_symlink nfsv3err_mknod nfsv3err_remove
nfsv3err_rmdir nfsv3err_rename nfsv3err_link
nfsv3err_readdir nfsv3err_readdirplus nfsv3err_fsstat
nfsv3err_fsinfo nfsv3err_pathconf nfsv3err_commit
nfsrv_v3errmap
There are additional structures that should be constified but due
to their being passed into general purpose functions without const
arguments, I have not yet converted.
In general, acquire nfsd_mtx when accessing any of the global NFS
structures, including struct nfssvc_sock, struct nfsd, struct
nfsrv_descript.
Release nfsd_mtx whenever calling into VFS, and acquire Giant for
calls into VFS. Giant is not required for any part of the
operation of the NFS server with the exception of calls into VFS.
Giant will never by acquired in the upcall code path. However, it
may operate entirely covered by Giant, or not. If debug.mpsafenet
is set to 0, the system calls will acquire Giant across all
operations, and the upcall will assert Giant. As such, by default,
this enables locking and allows us to test assertions, but should not
cause any substantial new amount of code to be run without Giant.
Bugs should manifest in the form of lock assertion failures for now.
This approach is similar (but not identical) to modifications to the
BSD/OS NFS server code snapshot provided by BSDi as part of their
SMPng snapshot. The strategy is almost the same (single lock over
the NFS server), but differs in the following ways:
- Our NFS client and server code bases don't overlap, which means
both fewer bugs and easier locking (thanks Peter!). Also means
NFSD_*() as opposed to NFS_*().
- We make broad use of assertions, whereas the BSD/OS code does not.
- Made slightly different choices about how to handle macros building
packets but operating with side effects.
- We acquire Giant only when entering VFS from the NFS server daemon
threads.
- Serious bugs in BSD/OS implementation corrected -- the snapshot we
received was clearly a work in progress.
Based on ideas from: BSDi SMPng Snapshot
Reviewed by: rick@snowhite.cis.uoguelph.ca
Extensive testing by: kris
2004-05-24 04:06:14 +00:00
|
|
|
NFSD_LOCK_ASSERT();
|
|
|
|
|
|
1994-05-24 10:09:53 +00:00
|
|
|
mrep = nd->nd_mrep;
|
|
|
|
|
md = nd->nd_md;
|
|
|
|
|
dpos = nd->nd_dpos;
|
|
|
|
|
if (has_header) {
|
2005-01-19 22:53:40 +00:00
|
|
|
tl = nfsm_dissect_nonblock(u_int32_t *, 10 * NFSX_UNSIGNED);
|
1998-05-31 20:09:01 +00:00
|
|
|
nd->nd_retxid = fxdr_unsigned(u_int32_t, *tl++);
|
2002-01-08 19:41:06 +00:00
|
|
|
if (*tl++ != nfsrv_rpc_call) {
|
1994-05-24 10:09:53 +00:00
|
|
|
m_freem(mrep);
|
|
|
|
|
return (EBADRPC);
|
|
|
|
|
}
|
1995-06-27 11:07:30 +00:00
|
|
|
} else
|
2005-01-19 22:53:40 +00:00
|
|
|
tl = nfsm_dissect_nonblock(u_int32_t *, 8 * NFSX_UNSIGNED);
|
1994-05-24 10:09:53 +00:00
|
|
|
nd->nd_repstat = 0;
|
1995-06-27 11:07:30 +00:00
|
|
|
nd->nd_flag = 0;
|
2002-01-08 19:41:06 +00:00
|
|
|
if (*tl++ != nfsrv_rpc_vers) {
|
1994-05-24 10:09:53 +00:00
|
|
|
nd->nd_repstat = ERPCMISMATCH;
|
|
|
|
|
nd->nd_procnum = NFSPROC_NOOP;
|
|
|
|
|
return (0);
|
|
|
|
|
}
|
2002-01-08 19:41:06 +00:00
|
|
|
if (*tl != nfsrv_nfs_prog) {
|
2001-09-18 23:32:09 +00:00
|
|
|
nd->nd_repstat = EPROGUNAVAIL;
|
|
|
|
|
nd->nd_procnum = NFSPROC_NOOP;
|
|
|
|
|
return (0);
|
1994-05-24 10:09:53 +00:00
|
|
|
}
|
|
|
|
|
tl++;
|
1998-05-31 20:09:01 +00:00
|
|
|
nfsvers = fxdr_unsigned(u_int32_t, *tl++);
|
2001-09-18 23:32:09 +00:00
|
|
|
if (nfsvers < NFS_VER2 || nfsvers > NFS_VER3) {
|
1994-05-24 10:09:53 +00:00
|
|
|
nd->nd_repstat = EPROGMISMATCH;
|
|
|
|
|
nd->nd_procnum = NFSPROC_NOOP;
|
|
|
|
|
return (0);
|
|
|
|
|
}
|
1998-05-31 20:09:01 +00:00
|
|
|
nd->nd_procnum = fxdr_unsigned(u_int32_t, *tl++);
|
1994-05-24 10:09:53 +00:00
|
|
|
if (nd->nd_procnum == NFSPROC_NULL)
|
|
|
|
|
return (0);
|
2001-09-18 23:32:09 +00:00
|
|
|
if (nfsvers == NFS_VER3) {
|
|
|
|
|
nd->nd_flag = ND_NFSV3;
|
|
|
|
|
if (nd->nd_procnum >= NFS_NPROCS) {
|
|
|
|
|
nd->nd_repstat = EPROCUNAVAIL;
|
|
|
|
|
nd->nd_procnum = NFSPROC_NOOP;
|
|
|
|
|
return (0);
|
|
|
|
|
}
|
|
|
|
|
} else {
|
|
|
|
|
if (nd->nd_procnum > NFSV2PROC_STATFS) {
|
|
|
|
|
nd->nd_repstat = EPROCUNAVAIL;
|
|
|
|
|
nd->nd_procnum = NFSPROC_NOOP;
|
|
|
|
|
return (0);
|
|
|
|
|
}
|
|
|
|
|
/* Map the v2 procedure numbers into v3 ones */
|
2002-01-08 19:41:06 +00:00
|
|
|
nd->nd_procnum = nfsrv_nfsv3_procid[nd->nd_procnum];
|
2001-09-18 23:32:09 +00:00
|
|
|
}
|
1994-05-24 10:09:53 +00:00
|
|
|
auth_type = *tl++;
|
|
|
|
|
len = fxdr_unsigned(int, *tl++);
|
|
|
|
|
if (len < 0 || len > RPCAUTH_MAXSIZ) {
|
|
|
|
|
m_freem(mrep);
|
|
|
|
|
return (EBADRPC);
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
/*
|
2001-09-18 23:32:09 +00:00
|
|
|
* Handle auth_unix;
|
1994-05-24 10:09:53 +00:00
|
|
|
*/
|
2002-01-08 19:41:06 +00:00
|
|
|
if (auth_type == nfsrv_rpc_auth_unix) {
|
1994-05-24 10:09:53 +00:00
|
|
|
len = fxdr_unsigned(int, *++tl);
|
|
|
|
|
if (len < 0 || len > NFS_MAXNAMLEN) {
|
|
|
|
|
m_freem(mrep);
|
|
|
|
|
return (EBADRPC);
|
|
|
|
|
}
|
|
|
|
|
nfsm_adv(nfsm_rndup(len));
|
2005-01-19 22:53:40 +00:00
|
|
|
tl = nfsm_dissect_nonblock(u_int32_t *, 3 * NFSX_UNSIGNED);
|
2006-01-28 19:24:40 +00:00
|
|
|
nd->nd_cr->cr_uid = nd->nd_cr->cr_ruid =
|
|
|
|
|
nd->nd_cr->cr_svuid = fxdr_unsigned(uid_t, *tl++);
|
|
|
|
|
nd->nd_cr->cr_groups[0] = nd->nd_cr->cr_rgid =
|
|
|
|
|
nd->nd_cr->cr_svgid = fxdr_unsigned(gid_t, *tl++);
|
2006-04-06 23:33:11 +00:00
|
|
|
#ifdef MAC
|
2008-10-28 11:33:06 +00:00
|
|
|
mac_cred_associate_nfsd(nd->nd_cr);
|
2006-04-06 23:33:11 +00:00
|
|
|
#endif
|
1994-05-24 10:09:53 +00:00
|
|
|
len = fxdr_unsigned(int, *tl);
|
|
|
|
|
if (len < 0 || len > RPCAUTH_UNIXGIDS) {
|
|
|
|
|
m_freem(mrep);
|
|
|
|
|
return (EBADRPC);
|
|
|
|
|
}
|
2005-01-19 22:53:40 +00:00
|
|
|
tl = nfsm_dissect_nonblock(u_int32_t *, (len + 2) * NFSX_UNSIGNED);
|
1994-05-24 10:09:53 +00:00
|
|
|
for (i = 1; i <= len; i++)
|
1995-06-27 11:07:30 +00:00
|
|
|
if (i < NGROUPS)
|
2006-01-28 19:24:40 +00:00
|
|
|
nd->nd_cr->cr_groups[i] = fxdr_unsigned(gid_t, *tl++);
|
1995-06-27 11:07:30 +00:00
|
|
|
else
|
|
|
|
|
tl++;
|
2006-01-28 19:24:40 +00:00
|
|
|
nd->nd_cr->cr_ngroups = (len >= NGROUPS) ? NGROUPS : (len + 1);
|
|
|
|
|
if (nd->nd_cr->cr_ngroups > 1)
|
|
|
|
|
nfsrvw_sort(nd->nd_cr->cr_groups, nd->nd_cr->cr_ngroups);
|
1995-06-27 11:07:30 +00:00
|
|
|
len = fxdr_unsigned(int, *++tl);
|
|
|
|
|
if (len < 0 || len > RPCAUTH_MAXSIZ) {
|
1994-05-24 10:09:53 +00:00
|
|
|
m_freem(mrep);
|
|
|
|
|
return (EBADRPC);
|
|
|
|
|
}
|
1995-06-27 11:07:30 +00:00
|
|
|
if (len > 0)
|
|
|
|
|
nfsm_adv(nfsm_rndup(len));
|
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
|
|
|
nd->nd_credflavor = RPCAUTH_UNIX;
|
1998-05-31 17:27:58 +00:00
|
|
|
} else {
|
|
|
|
|
nd->nd_repstat = (NFSERR_AUTHERR | AUTH_REJECTCRED);
|
|
|
|
|
nd->nd_procnum = NFSPROC_NOOP;
|
|
|
|
|
return (0);
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
nd->nd_md = md;
|
|
|
|
|
nd->nd_dpos = dpos;
|
|
|
|
|
return (0);
|
|
|
|
|
nfsmout:
|
|
|
|
|
return (error);
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
* Socket upcall routine for the nfsd sockets.
|
|
|
|
|
* The caddr_t arg is a pointer to the "struct nfssvc_sock".
|
|
|
|
|
* Essentially do as much as possible non-blocking, else punt and it will
|
2008-03-25 09:39:02 +00:00
|
|
|
* be called with M_WAIT from an nfsd.
|
1998-05-31 17:27:58 +00:00
|
|
|
*/
|
2009-06-01 21:17:03 +00:00
|
|
|
int
|
2001-09-18 23:32:09 +00:00
|
|
|
nfsrv_rcv(struct socket *so, void *arg, int waitflag)
|
1998-05-31 17:27:58 +00:00
|
|
|
{
|
2001-09-18 23:32:09 +00:00
|
|
|
struct nfssvc_sock *slp = (struct nfssvc_sock *)arg;
|
|
|
|
|
struct mbuf *m;
|
1998-05-31 17:27:58 +00:00
|
|
|
struct mbuf *mp;
|
|
|
|
|
struct sockaddr *nam;
|
|
|
|
|
struct uio auio;
|
|
|
|
|
int flags, error;
|
|
|
|
|
|
The socket code upcalls into the NFS server using the so_upcall
mechanism so that early processing on mbufs can be performed before
a context switch to the NFS server threads. Because of this, if
the socket code is running without Giant, the NFS server also needs
to be able to run the upcall code without relying on the presence on
Giant. This change modifies the NFS server to run using a "giant
code lock" covering operation of the whole subsystem. Work is in
progress to move to data-based locking as part of the NFSv4 server
changes.
Introduce an NFS server subsystem lock, 'nfsd_mtx', and a set of
macros to operate on the lock:
NFSD_LOCK_ASSERT() Assert nfsd_mtx owned by current thread
NFSD_UNLOCK_ASSERT() Assert nfsd_mtx not owned by current thread
NFSD_LOCK_DONTCARE() Advisory: this function doesn't care
NFSD_LOCK() Lock nfsd_mtx
NFSD_UNLOCK() Unlock nfsd_mtx
Constify a number of global variables/structures in the NFS server
code, as they are not modified and contain constants only:
nfsrvv2_procid nfsrv_nfsv3_procid nonidempotent
nfsv2_repstat nfsv2_type nfsrv_nfsv3_procid
nfsrvv2_procid nfsrv_v2errmap nfsv3err_null
nfsv3err_getattr nfsv3err_setattr nfsv3err_lookup
nfsv3err_access nfsv3err_readlink nfsv3err_read
nfsv3err_write nfsv3err_create nfsv3err_mkdir
nfsv3err_symlink nfsv3err_mknod nfsv3err_remove
nfsv3err_rmdir nfsv3err_rename nfsv3err_link
nfsv3err_readdir nfsv3err_readdirplus nfsv3err_fsstat
nfsv3err_fsinfo nfsv3err_pathconf nfsv3err_commit
nfsrv_v3errmap
There are additional structures that should be constified but due
to their being passed into general purpose functions without const
arguments, I have not yet converted.
In general, acquire nfsd_mtx when accessing any of the global NFS
structures, including struct nfssvc_sock, struct nfsd, struct
nfsrv_descript.
Release nfsd_mtx whenever calling into VFS, and acquire Giant for
calls into VFS. Giant is not required for any part of the
operation of the NFS server with the exception of calls into VFS.
Giant will never by acquired in the upcall code path. However, it
may operate entirely covered by Giant, or not. If debug.mpsafenet
is set to 0, the system calls will acquire Giant across all
operations, and the upcall will assert Giant. As such, by default,
this enables locking and allows us to test assertions, but should not
cause any substantial new amount of code to be run without Giant.
Bugs should manifest in the form of lock assertion failures for now.
This approach is similar (but not identical) to modifications to the
BSD/OS NFS server code snapshot provided by BSDi as part of their
SMPng snapshot. The strategy is almost the same (single lock over
the NFS server), but differs in the following ways:
- Our NFS client and server code bases don't overlap, which means
both fewer bugs and easier locking (thanks Peter!). Also means
NFSD_*() as opposed to NFS_*().
- We make broad use of assertions, whereas the BSD/OS code does not.
- Made slightly different choices about how to handle macros building
packets but operating with side effects.
- We acquire Giant only when entering VFS from the NFS server daemon
threads.
- Serious bugs in BSD/OS implementation corrected -- the snapshot we
received was clearly a work in progress.
Based on ideas from: BSDi SMPng Snapshot
Reviewed by: rick@snowhite.cis.uoguelph.ca
Extensive testing by: kris
2004-05-24 04:06:14 +00:00
|
|
|
NFSD_UNLOCK_ASSERT();
|
|
|
|
|
|
|
|
|
|
/* XXXRW: Unlocked read. */
|
1998-05-31 17:27:58 +00:00
|
|
|
if ((slp->ns_flag & SLP_VALID) == 0)
|
2009-06-01 21:17:03 +00:00
|
|
|
return (SU_OK);
|
2004-07-13 07:05:38 +00:00
|
|
|
|
1998-05-31 17:27:58 +00:00
|
|
|
/*
|
2004-07-13 07:05:38 +00:00
|
|
|
* We can't do this in the context of a socket callback
|
|
|
|
|
* because we're called with locks held.
|
|
|
|
|
* XXX: SMP
|
1998-05-31 17:27:58 +00:00
|
|
|
*/
|
2003-02-19 05:47:46 +00:00
|
|
|
if (waitflag == M_DONTWAIT) {
|
The socket code upcalls into the NFS server using the so_upcall
mechanism so that early processing on mbufs can be performed before
a context switch to the NFS server threads. Because of this, if
the socket code is running without Giant, the NFS server also needs
to be able to run the upcall code without relying on the presence on
Giant. This change modifies the NFS server to run using a "giant
code lock" covering operation of the whole subsystem. Work is in
progress to move to data-based locking as part of the NFSv4 server
changes.
Introduce an NFS server subsystem lock, 'nfsd_mtx', and a set of
macros to operate on the lock:
NFSD_LOCK_ASSERT() Assert nfsd_mtx owned by current thread
NFSD_UNLOCK_ASSERT() Assert nfsd_mtx not owned by current thread
NFSD_LOCK_DONTCARE() Advisory: this function doesn't care
NFSD_LOCK() Lock nfsd_mtx
NFSD_UNLOCK() Unlock nfsd_mtx
Constify a number of global variables/structures in the NFS server
code, as they are not modified and contain constants only:
nfsrvv2_procid nfsrv_nfsv3_procid nonidempotent
nfsv2_repstat nfsv2_type nfsrv_nfsv3_procid
nfsrvv2_procid nfsrv_v2errmap nfsv3err_null
nfsv3err_getattr nfsv3err_setattr nfsv3err_lookup
nfsv3err_access nfsv3err_readlink nfsv3err_read
nfsv3err_write nfsv3err_create nfsv3err_mkdir
nfsv3err_symlink nfsv3err_mknod nfsv3err_remove
nfsv3err_rmdir nfsv3err_rename nfsv3err_link
nfsv3err_readdir nfsv3err_readdirplus nfsv3err_fsstat
nfsv3err_fsinfo nfsv3err_pathconf nfsv3err_commit
nfsrv_v3errmap
There are additional structures that should be constified but due
to their being passed into general purpose functions without const
arguments, I have not yet converted.
In general, acquire nfsd_mtx when accessing any of the global NFS
structures, including struct nfssvc_sock, struct nfsd, struct
nfsrv_descript.
Release nfsd_mtx whenever calling into VFS, and acquire Giant for
calls into VFS. Giant is not required for any part of the
operation of the NFS server with the exception of calls into VFS.
Giant will never by acquired in the upcall code path. However, it
may operate entirely covered by Giant, or not. If debug.mpsafenet
is set to 0, the system calls will acquire Giant across all
operations, and the upcall will assert Giant. As such, by default,
this enables locking and allows us to test assertions, but should not
cause any substantial new amount of code to be run without Giant.
Bugs should manifest in the form of lock assertion failures for now.
This approach is similar (but not identical) to modifications to the
BSD/OS NFS server code snapshot provided by BSDi as part of their
SMPng snapshot. The strategy is almost the same (single lock over
the NFS server), but differs in the following ways:
- Our NFS client and server code bases don't overlap, which means
both fewer bugs and easier locking (thanks Peter!). Also means
NFSD_*() as opposed to NFS_*().
- We make broad use of assertions, whereas the BSD/OS code does not.
- Made slightly different choices about how to handle macros building
packets but operating with side effects.
- We acquire Giant only when entering VFS from the NFS server daemon
threads.
- Serious bugs in BSD/OS implementation corrected -- the snapshot we
received was clearly a work in progress.
Based on ideas from: BSDi SMPng Snapshot
Reviewed by: rick@snowhite.cis.uoguelph.ca
Extensive testing by: kris
2004-05-24 04:06:14 +00:00
|
|
|
NFSD_LOCK();
|
2001-09-18 23:32:09 +00:00
|
|
|
slp->ns_flag |= SLP_NEEDQ;
|
|
|
|
|
goto dorecs;
|
1998-05-31 17:27:58 +00:00
|
|
|
}
|
2004-07-13 07:05:38 +00:00
|
|
|
|
2003-11-07 22:57:09 +00:00
|
|
|
|
The socket code upcalls into the NFS server using the so_upcall
mechanism so that early processing on mbufs can be performed before
a context switch to the NFS server threads. Because of this, if
the socket code is running without Giant, the NFS server also needs
to be able to run the upcall code without relying on the presence on
Giant. This change modifies the NFS server to run using a "giant
code lock" covering operation of the whole subsystem. Work is in
progress to move to data-based locking as part of the NFSv4 server
changes.
Introduce an NFS server subsystem lock, 'nfsd_mtx', and a set of
macros to operate on the lock:
NFSD_LOCK_ASSERT() Assert nfsd_mtx owned by current thread
NFSD_UNLOCK_ASSERT() Assert nfsd_mtx not owned by current thread
NFSD_LOCK_DONTCARE() Advisory: this function doesn't care
NFSD_LOCK() Lock nfsd_mtx
NFSD_UNLOCK() Unlock nfsd_mtx
Constify a number of global variables/structures in the NFS server
code, as they are not modified and contain constants only:
nfsrvv2_procid nfsrv_nfsv3_procid nonidempotent
nfsv2_repstat nfsv2_type nfsrv_nfsv3_procid
nfsrvv2_procid nfsrv_v2errmap nfsv3err_null
nfsv3err_getattr nfsv3err_setattr nfsv3err_lookup
nfsv3err_access nfsv3err_readlink nfsv3err_read
nfsv3err_write nfsv3err_create nfsv3err_mkdir
nfsv3err_symlink nfsv3err_mknod nfsv3err_remove
nfsv3err_rmdir nfsv3err_rename nfsv3err_link
nfsv3err_readdir nfsv3err_readdirplus nfsv3err_fsstat
nfsv3err_fsinfo nfsv3err_pathconf nfsv3err_commit
nfsrv_v3errmap
There are additional structures that should be constified but due
to their being passed into general purpose functions without const
arguments, I have not yet converted.
In general, acquire nfsd_mtx when accessing any of the global NFS
structures, including struct nfssvc_sock, struct nfsd, struct
nfsrv_descript.
Release nfsd_mtx whenever calling into VFS, and acquire Giant for
calls into VFS. Giant is not required for any part of the
operation of the NFS server with the exception of calls into VFS.
Giant will never by acquired in the upcall code path. However, it
may operate entirely covered by Giant, or not. If debug.mpsafenet
is set to 0, the system calls will acquire Giant across all
operations, and the upcall will assert Giant. As such, by default,
this enables locking and allows us to test assertions, but should not
cause any substantial new amount of code to be run without Giant.
Bugs should manifest in the form of lock assertion failures for now.
This approach is similar (but not identical) to modifications to the
BSD/OS NFS server code snapshot provided by BSDi as part of their
SMPng snapshot. The strategy is almost the same (single lock over
the NFS server), but differs in the following ways:
- Our NFS client and server code bases don't overlap, which means
both fewer bugs and easier locking (thanks Peter!). Also means
NFSD_*() as opposed to NFS_*().
- We make broad use of assertions, whereas the BSD/OS code does not.
- Made slightly different choices about how to handle macros building
packets but operating with side effects.
- We acquire Giant only when entering VFS from the NFS server daemon
threads.
- Serious bugs in BSD/OS implementation corrected -- the snapshot we
received was clearly a work in progress.
Based on ideas from: BSDi SMPng Snapshot
Reviewed by: rick@snowhite.cis.uoguelph.ca
Extensive testing by: kris
2004-05-24 04:06:14 +00:00
|
|
|
NFSD_LOCK();
|
2001-09-12 08:38:13 +00:00
|
|
|
auio.uio_td = NULL;
|
1998-05-31 17:27:58 +00:00
|
|
|
if (so->so_type == SOCK_STREAM) {
|
|
|
|
|
/*
|
|
|
|
|
* If there are already records on the queue, defer soreceive()
|
|
|
|
|
* to an nfsd so that there is feedback to the TCP layer that
|
|
|
|
|
* the nfs servers are heavily loaded.
|
|
|
|
|
*/
|
2004-04-04 19:13:35 +00:00
|
|
|
if (STAILQ_FIRST(&slp->ns_rec) != NULL &&
|
|
|
|
|
waitflag == M_DONTWAIT) {
|
1998-05-31 17:27:58 +00:00
|
|
|
slp->ns_flag |= SLP_NEEDQ;
|
|
|
|
|
goto dorecs;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
* Do soreceive().
|
|
|
|
|
*/
|
|
|
|
|
auio.uio_resid = 1000000000;
|
|
|
|
|
flags = MSG_DONTWAIT;
|
The socket code upcalls into the NFS server using the so_upcall
mechanism so that early processing on mbufs can be performed before
a context switch to the NFS server threads. Because of this, if
the socket code is running without Giant, the NFS server also needs
to be able to run the upcall code without relying on the presence on
Giant. This change modifies the NFS server to run using a "giant
code lock" covering operation of the whole subsystem. Work is in
progress to move to data-based locking as part of the NFSv4 server
changes.
Introduce an NFS server subsystem lock, 'nfsd_mtx', and a set of
macros to operate on the lock:
NFSD_LOCK_ASSERT() Assert nfsd_mtx owned by current thread
NFSD_UNLOCK_ASSERT() Assert nfsd_mtx not owned by current thread
NFSD_LOCK_DONTCARE() Advisory: this function doesn't care
NFSD_LOCK() Lock nfsd_mtx
NFSD_UNLOCK() Unlock nfsd_mtx
Constify a number of global variables/structures in the NFS server
code, as they are not modified and contain constants only:
nfsrvv2_procid nfsrv_nfsv3_procid nonidempotent
nfsv2_repstat nfsv2_type nfsrv_nfsv3_procid
nfsrvv2_procid nfsrv_v2errmap nfsv3err_null
nfsv3err_getattr nfsv3err_setattr nfsv3err_lookup
nfsv3err_access nfsv3err_readlink nfsv3err_read
nfsv3err_write nfsv3err_create nfsv3err_mkdir
nfsv3err_symlink nfsv3err_mknod nfsv3err_remove
nfsv3err_rmdir nfsv3err_rename nfsv3err_link
nfsv3err_readdir nfsv3err_readdirplus nfsv3err_fsstat
nfsv3err_fsinfo nfsv3err_pathconf nfsv3err_commit
nfsrv_v3errmap
There are additional structures that should be constified but due
to their being passed into general purpose functions without const
arguments, I have not yet converted.
In general, acquire nfsd_mtx when accessing any of the global NFS
structures, including struct nfssvc_sock, struct nfsd, struct
nfsrv_descript.
Release nfsd_mtx whenever calling into VFS, and acquire Giant for
calls into VFS. Giant is not required for any part of the
operation of the NFS server with the exception of calls into VFS.
Giant will never by acquired in the upcall code path. However, it
may operate entirely covered by Giant, or not. If debug.mpsafenet
is set to 0, the system calls will acquire Giant across all
operations, and the upcall will assert Giant. As such, by default,
this enables locking and allows us to test assertions, but should not
cause any substantial new amount of code to be run without Giant.
Bugs should manifest in the form of lock assertion failures for now.
This approach is similar (but not identical) to modifications to the
BSD/OS NFS server code snapshot provided by BSDi as part of their
SMPng snapshot. The strategy is almost the same (single lock over
the NFS server), but differs in the following ways:
- Our NFS client and server code bases don't overlap, which means
both fewer bugs and easier locking (thanks Peter!). Also means
NFSD_*() as opposed to NFS_*().
- We make broad use of assertions, whereas the BSD/OS code does not.
- Made slightly different choices about how to handle macros building
packets but operating with side effects.
- We acquire Giant only when entering VFS from the NFS server daemon
threads.
- Serious bugs in BSD/OS implementation corrected -- the snapshot we
received was clearly a work in progress.
Based on ideas from: BSDi SMPng Snapshot
Reviewed by: rick@snowhite.cis.uoguelph.ca
Extensive testing by: kris
2004-05-24 04:06:14 +00:00
|
|
|
NFSD_UNLOCK();
|
soreceive_generic(), and sopoll_generic(). Add new functions sosend(),
soreceive(), and sopoll(), which are wrappers for pru_sosend,
pru_soreceive, and pru_sopoll, and are now used univerally by socket
consumers rather than either directly invoking the old so*() functions
or directly invoking the protocol switch method (about an even split
prior to this commit).
This completes an architectural change that was begun in 1996 to permit
protocols to provide substitute implementations, as now used by UDP.
Consumers now uniformly invoke sosend(), soreceive(), and sopoll() to
perform these operations on sockets -- in particular, distributed file
systems and socket system calls.
Architectural head nod: sam, gnn, wollman
2006-07-24 15:20:08 +00:00
|
|
|
error = soreceive(so, &nam, &auio, &mp, NULL, &flags);
|
The socket code upcalls into the NFS server using the so_upcall
mechanism so that early processing on mbufs can be performed before
a context switch to the NFS server threads. Because of this, if
the socket code is running without Giant, the NFS server also needs
to be able to run the upcall code without relying on the presence on
Giant. This change modifies the NFS server to run using a "giant
code lock" covering operation of the whole subsystem. Work is in
progress to move to data-based locking as part of the NFSv4 server
changes.
Introduce an NFS server subsystem lock, 'nfsd_mtx', and a set of
macros to operate on the lock:
NFSD_LOCK_ASSERT() Assert nfsd_mtx owned by current thread
NFSD_UNLOCK_ASSERT() Assert nfsd_mtx not owned by current thread
NFSD_LOCK_DONTCARE() Advisory: this function doesn't care
NFSD_LOCK() Lock nfsd_mtx
NFSD_UNLOCK() Unlock nfsd_mtx
Constify a number of global variables/structures in the NFS server
code, as they are not modified and contain constants only:
nfsrvv2_procid nfsrv_nfsv3_procid nonidempotent
nfsv2_repstat nfsv2_type nfsrv_nfsv3_procid
nfsrvv2_procid nfsrv_v2errmap nfsv3err_null
nfsv3err_getattr nfsv3err_setattr nfsv3err_lookup
nfsv3err_access nfsv3err_readlink nfsv3err_read
nfsv3err_write nfsv3err_create nfsv3err_mkdir
nfsv3err_symlink nfsv3err_mknod nfsv3err_remove
nfsv3err_rmdir nfsv3err_rename nfsv3err_link
nfsv3err_readdir nfsv3err_readdirplus nfsv3err_fsstat
nfsv3err_fsinfo nfsv3err_pathconf nfsv3err_commit
nfsrv_v3errmap
There are additional structures that should be constified but due
to their being passed into general purpose functions without const
arguments, I have not yet converted.
In general, acquire nfsd_mtx when accessing any of the global NFS
structures, including struct nfssvc_sock, struct nfsd, struct
nfsrv_descript.
Release nfsd_mtx whenever calling into VFS, and acquire Giant for
calls into VFS. Giant is not required for any part of the
operation of the NFS server with the exception of calls into VFS.
Giant will never by acquired in the upcall code path. However, it
may operate entirely covered by Giant, or not. If debug.mpsafenet
is set to 0, the system calls will acquire Giant across all
operations, and the upcall will assert Giant. As such, by default,
this enables locking and allows us to test assertions, but should not
cause any substantial new amount of code to be run without Giant.
Bugs should manifest in the form of lock assertion failures for now.
This approach is similar (but not identical) to modifications to the
BSD/OS NFS server code snapshot provided by BSDi as part of their
SMPng snapshot. The strategy is almost the same (single lock over
the NFS server), but differs in the following ways:
- Our NFS client and server code bases don't overlap, which means
both fewer bugs and easier locking (thanks Peter!). Also means
NFSD_*() as opposed to NFS_*().
- We make broad use of assertions, whereas the BSD/OS code does not.
- Made slightly different choices about how to handle macros building
packets but operating with side effects.
- We acquire Giant only when entering VFS from the NFS server daemon
threads.
- Serious bugs in BSD/OS implementation corrected -- the snapshot we
received was clearly a work in progress.
Based on ideas from: BSDi SMPng Snapshot
Reviewed by: rick@snowhite.cis.uoguelph.ca
Extensive testing by: kris
2004-05-24 04:06:14 +00:00
|
|
|
NFSD_LOCK();
|
2002-07-11 17:54:58 +00:00
|
|
|
if (error || mp == NULL) {
|
1998-05-31 17:27:58 +00:00
|
|
|
if (error == EWOULDBLOCK)
|
|
|
|
|
slp->ns_flag |= SLP_NEEDQ;
|
|
|
|
|
else
|
|
|
|
|
slp->ns_flag |= SLP_DISCONN;
|
|
|
|
|
goto dorecs;
|
|
|
|
|
}
|
|
|
|
|
m = mp;
|
|
|
|
|
if (slp->ns_rawend) {
|
|
|
|
|
slp->ns_rawend->m_next = m;
|
|
|
|
|
slp->ns_cc += 1000000000 - auio.uio_resid;
|
|
|
|
|
} else {
|
|
|
|
|
slp->ns_raw = m;
|
|
|
|
|
slp->ns_cc = 1000000000 - auio.uio_resid;
|
|
|
|
|
}
|
|
|
|
|
while (m->m_next)
|
|
|
|
|
m = m->m_next;
|
|
|
|
|
slp->ns_rawend = m;
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
* Now try and parse record(s) out of the raw stream data.
|
|
|
|
|
*/
|
|
|
|
|
error = nfsrv_getstream(slp, waitflag);
|
|
|
|
|
if (error) {
|
|
|
|
|
if (error == EPERM)
|
|
|
|
|
slp->ns_flag |= SLP_DISCONN;
|
|
|
|
|
else
|
|
|
|
|
slp->ns_flag |= SLP_NEEDQ;
|
|
|
|
|
}
|
|
|
|
|
} else {
|
|
|
|
|
do {
|
|
|
|
|
auio.uio_resid = 1000000000;
|
|
|
|
|
flags = MSG_DONTWAIT;
|
The socket code upcalls into the NFS server using the so_upcall
mechanism so that early processing on mbufs can be performed before
a context switch to the NFS server threads. Because of this, if
the socket code is running without Giant, the NFS server also needs
to be able to run the upcall code without relying on the presence on
Giant. This change modifies the NFS server to run using a "giant
code lock" covering operation of the whole subsystem. Work is in
progress to move to data-based locking as part of the NFSv4 server
changes.
Introduce an NFS server subsystem lock, 'nfsd_mtx', and a set of
macros to operate on the lock:
NFSD_LOCK_ASSERT() Assert nfsd_mtx owned by current thread
NFSD_UNLOCK_ASSERT() Assert nfsd_mtx not owned by current thread
NFSD_LOCK_DONTCARE() Advisory: this function doesn't care
NFSD_LOCK() Lock nfsd_mtx
NFSD_UNLOCK() Unlock nfsd_mtx
Constify a number of global variables/structures in the NFS server
code, as they are not modified and contain constants only:
nfsrvv2_procid nfsrv_nfsv3_procid nonidempotent
nfsv2_repstat nfsv2_type nfsrv_nfsv3_procid
nfsrvv2_procid nfsrv_v2errmap nfsv3err_null
nfsv3err_getattr nfsv3err_setattr nfsv3err_lookup
nfsv3err_access nfsv3err_readlink nfsv3err_read
nfsv3err_write nfsv3err_create nfsv3err_mkdir
nfsv3err_symlink nfsv3err_mknod nfsv3err_remove
nfsv3err_rmdir nfsv3err_rename nfsv3err_link
nfsv3err_readdir nfsv3err_readdirplus nfsv3err_fsstat
nfsv3err_fsinfo nfsv3err_pathconf nfsv3err_commit
nfsrv_v3errmap
There are additional structures that should be constified but due
to their being passed into general purpose functions without const
arguments, I have not yet converted.
In general, acquire nfsd_mtx when accessing any of the global NFS
structures, including struct nfssvc_sock, struct nfsd, struct
nfsrv_descript.
Release nfsd_mtx whenever calling into VFS, and acquire Giant for
calls into VFS. Giant is not required for any part of the
operation of the NFS server with the exception of calls into VFS.
Giant will never by acquired in the upcall code path. However, it
may operate entirely covered by Giant, or not. If debug.mpsafenet
is set to 0, the system calls will acquire Giant across all
operations, and the upcall will assert Giant. As such, by default,
this enables locking and allows us to test assertions, but should not
cause any substantial new amount of code to be run without Giant.
Bugs should manifest in the form of lock assertion failures for now.
This approach is similar (but not identical) to modifications to the
BSD/OS NFS server code snapshot provided by BSDi as part of their
SMPng snapshot. The strategy is almost the same (single lock over
the NFS server), but differs in the following ways:
- Our NFS client and server code bases don't overlap, which means
both fewer bugs and easier locking (thanks Peter!). Also means
NFSD_*() as opposed to NFS_*().
- We make broad use of assertions, whereas the BSD/OS code does not.
- Made slightly different choices about how to handle macros building
packets but operating with side effects.
- We acquire Giant only when entering VFS from the NFS server daemon
threads.
- Serious bugs in BSD/OS implementation corrected -- the snapshot we
received was clearly a work in progress.
Based on ideas from: BSDi SMPng Snapshot
Reviewed by: rick@snowhite.cis.uoguelph.ca
Extensive testing by: kris
2004-05-24 04:06:14 +00:00
|
|
|
NFSD_UNLOCK();
|
soreceive_generic(), and sopoll_generic(). Add new functions sosend(),
soreceive(), and sopoll(), which are wrappers for pru_sosend,
pru_soreceive, and pru_sopoll, and are now used univerally by socket
consumers rather than either directly invoking the old so*() functions
or directly invoking the protocol switch method (about an even split
prior to this commit).
This completes an architectural change that was begun in 1996 to permit
protocols to provide substitute implementations, as now used by UDP.
Consumers now uniformly invoke sosend(), soreceive(), and sopoll() to
perform these operations on sockets -- in particular, distributed file
systems and socket system calls.
Architectural head nod: sam, gnn, wollman
2006-07-24 15:20:08 +00:00
|
|
|
error = soreceive(so, &nam, &auio, &mp, NULL, &flags);
|
1998-05-31 17:27:58 +00:00
|
|
|
if (mp) {
|
|
|
|
|
struct nfsrv_rec *rec;
|
|
|
|
|
rec = malloc(sizeof(struct nfsrv_rec),
|
2003-02-22 10:35:37 +00:00
|
|
|
M_NFSRVDESC,
|
|
|
|
|
waitflag == M_DONTWAIT ? M_NOWAIT : M_WAITOK);
|
1998-05-31 17:27:58 +00:00
|
|
|
if (!rec) {
|
|
|
|
|
if (nam)
|
2008-10-23 15:53:51 +00:00
|
|
|
free(nam, M_SONAME);
|
1998-05-31 17:27:58 +00:00
|
|
|
m_freem(mp);
|
The socket code upcalls into the NFS server using the so_upcall
mechanism so that early processing on mbufs can be performed before
a context switch to the NFS server threads. Because of this, if
the socket code is running without Giant, the NFS server also needs
to be able to run the upcall code without relying on the presence on
Giant. This change modifies the NFS server to run using a "giant
code lock" covering operation of the whole subsystem. Work is in
progress to move to data-based locking as part of the NFSv4 server
changes.
Introduce an NFS server subsystem lock, 'nfsd_mtx', and a set of
macros to operate on the lock:
NFSD_LOCK_ASSERT() Assert nfsd_mtx owned by current thread
NFSD_UNLOCK_ASSERT() Assert nfsd_mtx not owned by current thread
NFSD_LOCK_DONTCARE() Advisory: this function doesn't care
NFSD_LOCK() Lock nfsd_mtx
NFSD_UNLOCK() Unlock nfsd_mtx
Constify a number of global variables/structures in the NFS server
code, as they are not modified and contain constants only:
nfsrvv2_procid nfsrv_nfsv3_procid nonidempotent
nfsv2_repstat nfsv2_type nfsrv_nfsv3_procid
nfsrvv2_procid nfsrv_v2errmap nfsv3err_null
nfsv3err_getattr nfsv3err_setattr nfsv3err_lookup
nfsv3err_access nfsv3err_readlink nfsv3err_read
nfsv3err_write nfsv3err_create nfsv3err_mkdir
nfsv3err_symlink nfsv3err_mknod nfsv3err_remove
nfsv3err_rmdir nfsv3err_rename nfsv3err_link
nfsv3err_readdir nfsv3err_readdirplus nfsv3err_fsstat
nfsv3err_fsinfo nfsv3err_pathconf nfsv3err_commit
nfsrv_v3errmap
There are additional structures that should be constified but due
to their being passed into general purpose functions without const
arguments, I have not yet converted.
In general, acquire nfsd_mtx when accessing any of the global NFS
structures, including struct nfssvc_sock, struct nfsd, struct
nfsrv_descript.
Release nfsd_mtx whenever calling into VFS, and acquire Giant for
calls into VFS. Giant is not required for any part of the
operation of the NFS server with the exception of calls into VFS.
Giant will never by acquired in the upcall code path. However, it
may operate entirely covered by Giant, or not. If debug.mpsafenet
is set to 0, the system calls will acquire Giant across all
operations, and the upcall will assert Giant. As such, by default,
this enables locking and allows us to test assertions, but should not
cause any substantial new amount of code to be run without Giant.
Bugs should manifest in the form of lock assertion failures for now.
This approach is similar (but not identical) to modifications to the
BSD/OS NFS server code snapshot provided by BSDi as part of their
SMPng snapshot. The strategy is almost the same (single lock over
the NFS server), but differs in the following ways:
- Our NFS client and server code bases don't overlap, which means
both fewer bugs and easier locking (thanks Peter!). Also means
NFSD_*() as opposed to NFS_*().
- We make broad use of assertions, whereas the BSD/OS code does not.
- Made slightly different choices about how to handle macros building
packets but operating with side effects.
- We acquire Giant only when entering VFS from the NFS server daemon
threads.
- Serious bugs in BSD/OS implementation corrected -- the snapshot we
received was clearly a work in progress.
Based on ideas from: BSDi SMPng Snapshot
Reviewed by: rick@snowhite.cis.uoguelph.ca
Extensive testing by: kris
2004-05-24 04:06:14 +00:00
|
|
|
NFSD_LOCK();
|
1998-05-31 17:27:58 +00:00
|
|
|
continue;
|
|
|
|
|
}
|
The VFS/BIO subsystem contained a number of hacks in order to optimize
piecemeal, middle-of-file writes for NFS. These hacks have caused no
end of trouble, especially when combined with mmap(). I've removed
them. Instead, NFS will issue a read-before-write to fully
instantiate the struct buf containing the write. NFS does, however,
optimize piecemeal appends to files. For most common file operations,
you will not notice the difference. The sole remaining fragment in
the VFS/BIO system is b_dirtyoff/end, which NFS uses to avoid cache
coherency issues with read-merge-write style operations. NFS also
optimizes the write-covers-entire-buffer case by avoiding the
read-before-write. There is quite a bit of room for further
optimization in these areas.
The VM system marks pages fully-valid (AKA vm_page_t->valid =
VM_PAGE_BITS_ALL) in several places, most noteably in vm_fault. This
is not correct operation. The vm_pager_get_pages() code is now
responsible for marking VM pages all-valid. A number of VM helper
routines have been added to aid in zeroing-out the invalid portions of
a VM page prior to the page being marked all-valid. This operation is
necessary to properly support mmap(). The zeroing occurs most often
when dealing with file-EOF situations. Several bugs have been fixed
in the NFS subsystem, including bits handling file and directory EOF
situations and buf->b_flags consistancy issues relating to clearing
B_ERROR & B_INVAL, and handling B_DONE.
getblk() and allocbuf() have been rewritten. B_CACHE operation is now
formally defined in comments and more straightforward in
implementation. B_CACHE for VMIO buffers is based on the validity of
the backing store. B_CACHE for non-VMIO buffers is based simply on
whether the buffer is B_INVAL or not (B_CACHE set if B_INVAL clear,
and vise-versa). biodone() is now responsible for setting B_CACHE
when a successful read completes. B_CACHE is also set when a bdwrite()
is initiated and when a bwrite() is initiated. VFS VOP_BWRITE
routines (there are only two - nfs_bwrite() and bwrite()) are now
expected to set B_CACHE. This means that bowrite() and bawrite() also
set B_CACHE indirectly.
There are a number of places in the code which were previously using
buf->b_bufsize (which is DEV_BSIZE aligned) when they should have
been using buf->b_bcount. These have been fixed. getblk() now clears
B_DONE on return because the rest of the system is so bad about
dealing with B_DONE.
Major fixes to NFS/TCP have been made. A server-side bug could cause
requests to be lost by the server due to nfs_realign() overwriting
other rpc's in the same TCP mbuf chain. The server's kernel must be
recompiled to get the benefit of the fixes.
Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
1999-05-02 23:57:16 +00:00
|
|
|
nfs_realign(&mp, 10 * NFSX_UNSIGNED);
|
2007-04-15 15:50:50 +00:00
|
|
|
NFSD_LOCK();
|
1998-05-31 17:27:58 +00:00
|
|
|
rec->nr_address = nam;
|
|
|
|
|
rec->nr_packet = mp;
|
|
|
|
|
STAILQ_INSERT_TAIL(&slp->ns_rec, rec, nr_link);
|
The socket code upcalls into the NFS server using the so_upcall
mechanism so that early processing on mbufs can be performed before
a context switch to the NFS server threads. Because of this, if
the socket code is running without Giant, the NFS server also needs
to be able to run the upcall code without relying on the presence on
Giant. This change modifies the NFS server to run using a "giant
code lock" covering operation of the whole subsystem. Work is in
progress to move to data-based locking as part of the NFSv4 server
changes.
Introduce an NFS server subsystem lock, 'nfsd_mtx', and a set of
macros to operate on the lock:
NFSD_LOCK_ASSERT() Assert nfsd_mtx owned by current thread
NFSD_UNLOCK_ASSERT() Assert nfsd_mtx not owned by current thread
NFSD_LOCK_DONTCARE() Advisory: this function doesn't care
NFSD_LOCK() Lock nfsd_mtx
NFSD_UNLOCK() Unlock nfsd_mtx
Constify a number of global variables/structures in the NFS server
code, as they are not modified and contain constants only:
nfsrvv2_procid nfsrv_nfsv3_procid nonidempotent
nfsv2_repstat nfsv2_type nfsrv_nfsv3_procid
nfsrvv2_procid nfsrv_v2errmap nfsv3err_null
nfsv3err_getattr nfsv3err_setattr nfsv3err_lookup
nfsv3err_access nfsv3err_readlink nfsv3err_read
nfsv3err_write nfsv3err_create nfsv3err_mkdir
nfsv3err_symlink nfsv3err_mknod nfsv3err_remove
nfsv3err_rmdir nfsv3err_rename nfsv3err_link
nfsv3err_readdir nfsv3err_readdirplus nfsv3err_fsstat
nfsv3err_fsinfo nfsv3err_pathconf nfsv3err_commit
nfsrv_v3errmap
There are additional structures that should be constified but due
to their being passed into general purpose functions without const
arguments, I have not yet converted.
In general, acquire nfsd_mtx when accessing any of the global NFS
structures, including struct nfssvc_sock, struct nfsd, struct
nfsrv_descript.
Release nfsd_mtx whenever calling into VFS, and acquire Giant for
calls into VFS. Giant is not required for any part of the
operation of the NFS server with the exception of calls into VFS.
Giant will never by acquired in the upcall code path. However, it
may operate entirely covered by Giant, or not. If debug.mpsafenet
is set to 0, the system calls will acquire Giant across all
operations, and the upcall will assert Giant. As such, by default,
this enables locking and allows us to test assertions, but should not
cause any substantial new amount of code to be run without Giant.
Bugs should manifest in the form of lock assertion failures for now.
This approach is similar (but not identical) to modifications to the
BSD/OS NFS server code snapshot provided by BSDi as part of their
SMPng snapshot. The strategy is almost the same (single lock over
the NFS server), but differs in the following ways:
- Our NFS client and server code bases don't overlap, which means
both fewer bugs and easier locking (thanks Peter!). Also means
NFSD_*() as opposed to NFS_*().
- We make broad use of assertions, whereas the BSD/OS code does not.
- Made slightly different choices about how to handle macros building
packets but operating with side effects.
- We acquire Giant only when entering VFS from the NFS server daemon
threads.
- Serious bugs in BSD/OS implementation corrected -- the snapshot we
received was clearly a work in progress.
Based on ideas from: BSDi SMPng Snapshot
Reviewed by: rick@snowhite.cis.uoguelph.ca
Extensive testing by: kris
2004-05-24 04:06:14 +00:00
|
|
|
} else
|
|
|
|
|
NFSD_LOCK();
|
1998-05-31 17:27:58 +00:00
|
|
|
if (error) {
|
|
|
|
|
if ((so->so_proto->pr_flags & PR_CONNREQUIRED)
|
|
|
|
|
&& error != EWOULDBLOCK) {
|
|
|
|
|
slp->ns_flag |= SLP_DISCONN;
|
|
|
|
|
goto dorecs;
|
|
|
|
|
}
|
1995-06-27 11:07:30 +00:00
|
|
|
}
|
1998-05-31 17:27:58 +00:00
|
|
|
} while (mp);
|
|
|
|
|
}
|
1994-05-24 10:09:53 +00:00
|
|
|
|
1998-05-31 17:27:58 +00:00
|
|
|
/*
|
|
|
|
|
* Now try and process the request records, non-blocking.
|
|
|
|
|
*/
|
|
|
|
|
dorecs:
|
2003-02-19 05:47:46 +00:00
|
|
|
if (waitflag == M_DONTWAIT &&
|
2004-04-04 19:13:35 +00:00
|
|
|
(STAILQ_FIRST(&slp->ns_rec) != NULL ||
|
|
|
|
|
(slp->ns_flag & (SLP_NEEDQ | SLP_DISCONN))))
|
1998-05-31 17:27:58 +00:00
|
|
|
nfsrv_wakenfsd(slp);
|
The socket code upcalls into the NFS server using the so_upcall
mechanism so that early processing on mbufs can be performed before
a context switch to the NFS server threads. Because of this, if
the socket code is running without Giant, the NFS server also needs
to be able to run the upcall code without relying on the presence on
Giant. This change modifies the NFS server to run using a "giant
code lock" covering operation of the whole subsystem. Work is in
progress to move to data-based locking as part of the NFSv4 server
changes.
Introduce an NFS server subsystem lock, 'nfsd_mtx', and a set of
macros to operate on the lock:
NFSD_LOCK_ASSERT() Assert nfsd_mtx owned by current thread
NFSD_UNLOCK_ASSERT() Assert nfsd_mtx not owned by current thread
NFSD_LOCK_DONTCARE() Advisory: this function doesn't care
NFSD_LOCK() Lock nfsd_mtx
NFSD_UNLOCK() Unlock nfsd_mtx
Constify a number of global variables/structures in the NFS server
code, as they are not modified and contain constants only:
nfsrvv2_procid nfsrv_nfsv3_procid nonidempotent
nfsv2_repstat nfsv2_type nfsrv_nfsv3_procid
nfsrvv2_procid nfsrv_v2errmap nfsv3err_null
nfsv3err_getattr nfsv3err_setattr nfsv3err_lookup
nfsv3err_access nfsv3err_readlink nfsv3err_read
nfsv3err_write nfsv3err_create nfsv3err_mkdir
nfsv3err_symlink nfsv3err_mknod nfsv3err_remove
nfsv3err_rmdir nfsv3err_rename nfsv3err_link
nfsv3err_readdir nfsv3err_readdirplus nfsv3err_fsstat
nfsv3err_fsinfo nfsv3err_pathconf nfsv3err_commit
nfsrv_v3errmap
There are additional structures that should be constified but due
to their being passed into general purpose functions without const
arguments, I have not yet converted.
In general, acquire nfsd_mtx when accessing any of the global NFS
structures, including struct nfssvc_sock, struct nfsd, struct
nfsrv_descript.
Release nfsd_mtx whenever calling into VFS, and acquire Giant for
calls into VFS. Giant is not required for any part of the
operation of the NFS server with the exception of calls into VFS.
Giant will never by acquired in the upcall code path. However, it
may operate entirely covered by Giant, or not. If debug.mpsafenet
is set to 0, the system calls will acquire Giant across all
operations, and the upcall will assert Giant. As such, by default,
this enables locking and allows us to test assertions, but should not
cause any substantial new amount of code to be run without Giant.
Bugs should manifest in the form of lock assertion failures for now.
This approach is similar (but not identical) to modifications to the
BSD/OS NFS server code snapshot provided by BSDi as part of their
SMPng snapshot. The strategy is almost the same (single lock over
the NFS server), but differs in the following ways:
- Our NFS client and server code bases don't overlap, which means
both fewer bugs and easier locking (thanks Peter!). Also means
NFSD_*() as opposed to NFS_*().
- We make broad use of assertions, whereas the BSD/OS code does not.
- Made slightly different choices about how to handle macros building
packets but operating with side effects.
- We acquire Giant only when entering VFS from the NFS server daemon
threads.
- Serious bugs in BSD/OS implementation corrected -- the snapshot we
received was clearly a work in progress.
Based on ideas from: BSDi SMPng Snapshot
Reviewed by: rick@snowhite.cis.uoguelph.ca
Extensive testing by: kris
2004-05-24 04:06:14 +00:00
|
|
|
NFSD_UNLOCK();
|
1998-05-31 17:27:58 +00:00
|
|
|
}
|
1995-06-27 11:07:30 +00:00
|
|
|
|
1998-05-31 17:27:58 +00:00
|
|
|
/*
|
|
|
|
|
* Try and extract an RPC request from the mbuf data list received on a
|
|
|
|
|
* stream socket. The "waitflag" argument indicates whether or not it
|
|
|
|
|
* can sleep.
|
|
|
|
|
*/
|
|
|
|
|
static int
|
2001-09-18 23:32:09 +00:00
|
|
|
nfsrv_getstream(struct nfssvc_sock *slp, int waitflag)
|
1998-05-31 17:27:58 +00:00
|
|
|
{
|
2001-09-18 23:32:09 +00:00
|
|
|
struct mbuf *m, **mpp;
|
|
|
|
|
char *cp1, *cp2;
|
|
|
|
|
int len;
|
2002-07-17 01:07:08 +00:00
|
|
|
struct mbuf *om, *m2, *recm;
|
1998-05-31 20:09:01 +00:00
|
|
|
u_int32_t recmark;
|
1998-05-31 17:27:58 +00:00
|
|
|
|
The socket code upcalls into the NFS server using the so_upcall
mechanism so that early processing on mbufs can be performed before
a context switch to the NFS server threads. Because of this, if
the socket code is running without Giant, the NFS server also needs
to be able to run the upcall code without relying on the presence on
Giant. This change modifies the NFS server to run using a "giant
code lock" covering operation of the whole subsystem. Work is in
progress to move to data-based locking as part of the NFSv4 server
changes.
Introduce an NFS server subsystem lock, 'nfsd_mtx', and a set of
macros to operate on the lock:
NFSD_LOCK_ASSERT() Assert nfsd_mtx owned by current thread
NFSD_UNLOCK_ASSERT() Assert nfsd_mtx not owned by current thread
NFSD_LOCK_DONTCARE() Advisory: this function doesn't care
NFSD_LOCK() Lock nfsd_mtx
NFSD_UNLOCK() Unlock nfsd_mtx
Constify a number of global variables/structures in the NFS server
code, as they are not modified and contain constants only:
nfsrvv2_procid nfsrv_nfsv3_procid nonidempotent
nfsv2_repstat nfsv2_type nfsrv_nfsv3_procid
nfsrvv2_procid nfsrv_v2errmap nfsv3err_null
nfsv3err_getattr nfsv3err_setattr nfsv3err_lookup
nfsv3err_access nfsv3err_readlink nfsv3err_read
nfsv3err_write nfsv3err_create nfsv3err_mkdir
nfsv3err_symlink nfsv3err_mknod nfsv3err_remove
nfsv3err_rmdir nfsv3err_rename nfsv3err_link
nfsv3err_readdir nfsv3err_readdirplus nfsv3err_fsstat
nfsv3err_fsinfo nfsv3err_pathconf nfsv3err_commit
nfsrv_v3errmap
There are additional structures that should be constified but due
to their being passed into general purpose functions without const
arguments, I have not yet converted.
In general, acquire nfsd_mtx when accessing any of the global NFS
structures, including struct nfssvc_sock, struct nfsd, struct
nfsrv_descript.
Release nfsd_mtx whenever calling into VFS, and acquire Giant for
calls into VFS. Giant is not required for any part of the
operation of the NFS server with the exception of calls into VFS.
Giant will never by acquired in the upcall code path. However, it
may operate entirely covered by Giant, or not. If debug.mpsafenet
is set to 0, the system calls will acquire Giant across all
operations, and the upcall will assert Giant. As such, by default,
this enables locking and allows us to test assertions, but should not
cause any substantial new amount of code to be run without Giant.
Bugs should manifest in the form of lock assertion failures for now.
This approach is similar (but not identical) to modifications to the
BSD/OS NFS server code snapshot provided by BSDi as part of their
SMPng snapshot. The strategy is almost the same (single lock over
the NFS server), but differs in the following ways:
- Our NFS client and server code bases don't overlap, which means
both fewer bugs and easier locking (thanks Peter!). Also means
NFSD_*() as opposed to NFS_*().
- We make broad use of assertions, whereas the BSD/OS code does not.
- Made slightly different choices about how to handle macros building
packets but operating with side effects.
- We acquire Giant only when entering VFS from the NFS server daemon
threads.
- Serious bugs in BSD/OS implementation corrected -- the snapshot we
received was clearly a work in progress.
Based on ideas from: BSDi SMPng Snapshot
Reviewed by: rick@snowhite.cis.uoguelph.ca
Extensive testing by: kris
2004-05-24 04:06:14 +00:00
|
|
|
NFSD_LOCK_ASSERT();
|
|
|
|
|
|
1998-05-31 17:27:58 +00:00
|
|
|
if (slp->ns_flag & SLP_GETSTREAM)
|
|
|
|
|
panic("nfs getstream");
|
|
|
|
|
slp->ns_flag |= SLP_GETSTREAM;
|
|
|
|
|
for (;;) {
|
|
|
|
|
if (slp->ns_reclen == 0) {
|
|
|
|
|
if (slp->ns_cc < NFSX_UNSIGNED) {
|
|
|
|
|
slp->ns_flag &= ~SLP_GETSTREAM;
|
|
|
|
|
return (0);
|
|
|
|
|
}
|
|
|
|
|
m = slp->ns_raw;
|
|
|
|
|
if (m->m_len >= NFSX_UNSIGNED) {
|
|
|
|
|
bcopy(mtod(m, caddr_t), (caddr_t)&recmark, NFSX_UNSIGNED);
|
|
|
|
|
m->m_data += NFSX_UNSIGNED;
|
|
|
|
|
m->m_len -= NFSX_UNSIGNED;
|
|
|
|
|
} else {
|
|
|
|
|
cp1 = (caddr_t)&recmark;
|
|
|
|
|
cp2 = mtod(m, caddr_t);
|
|
|
|
|
while (cp1 < ((caddr_t)&recmark) + NFSX_UNSIGNED) {
|
|
|
|
|
while (m->m_len == 0) {
|
|
|
|
|
m = m->m_next;
|
|
|
|
|
cp2 = mtod(m, caddr_t);
|
|
|
|
|
}
|
|
|
|
|
*cp1++ = *cp2++;
|
|
|
|
|
m->m_data++;
|
|
|
|
|
m->m_len--;
|
1995-06-27 11:07:30 +00:00
|
|
|
}
|
1998-05-31 17:27:58 +00:00
|
|
|
}
|
|
|
|
|
slp->ns_cc -= NFSX_UNSIGNED;
|
|
|
|
|
recmark = ntohl(recmark);
|
|
|
|
|
slp->ns_reclen = recmark & ~0x80000000;
|
|
|
|
|
if (recmark & 0x80000000)
|
|
|
|
|
slp->ns_flag |= SLP_LASTFRAG;
|
|
|
|
|
else
|
|
|
|
|
slp->ns_flag &= ~SLP_LASTFRAG;
|
2006-03-01 14:17:32 +00:00
|
|
|
if (slp->ns_reclen > NFS_MAXPACKET || slp->ns_reclen <= 0) {
|
1998-05-31 17:27:58 +00:00
|
|
|
slp->ns_flag &= ~SLP_GETSTREAM;
|
|
|
|
|
return (EPERM);
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
* Now get the record part.
|
2002-07-17 01:07:08 +00:00
|
|
|
*
|
|
|
|
|
* Note that slp->ns_reclen may be 0. Linux sometimes
|
|
|
|
|
* generates 0-length RPCs.
|
1998-05-31 17:27:58 +00:00
|
|
|
*/
|
2002-07-17 01:07:08 +00:00
|
|
|
recm = NULL;
|
1998-05-31 17:27:58 +00:00
|
|
|
if (slp->ns_cc == slp->ns_reclen) {
|
|
|
|
|
recm = slp->ns_raw;
|
2002-07-11 17:54:58 +00:00
|
|
|
slp->ns_raw = slp->ns_rawend = NULL;
|
1998-05-31 17:27:58 +00:00
|
|
|
slp->ns_cc = slp->ns_reclen = 0;
|
|
|
|
|
} else if (slp->ns_cc > slp->ns_reclen) {
|
|
|
|
|
len = 0;
|
|
|
|
|
m = slp->ns_raw;
|
2002-07-11 17:54:58 +00:00
|
|
|
om = NULL;
|
2002-07-17 01:07:08 +00:00
|
|
|
|
1998-05-31 17:27:58 +00:00
|
|
|
while (len < slp->ns_reclen) {
|
|
|
|
|
if ((len + m->m_len) > slp->ns_reclen) {
|
The socket code upcalls into the NFS server using the so_upcall
mechanism so that early processing on mbufs can be performed before
a context switch to the NFS server threads. Because of this, if
the socket code is running without Giant, the NFS server also needs
to be able to run the upcall code without relying on the presence on
Giant. This change modifies the NFS server to run using a "giant
code lock" covering operation of the whole subsystem. Work is in
progress to move to data-based locking as part of the NFSv4 server
changes.
Introduce an NFS server subsystem lock, 'nfsd_mtx', and a set of
macros to operate on the lock:
NFSD_LOCK_ASSERT() Assert nfsd_mtx owned by current thread
NFSD_UNLOCK_ASSERT() Assert nfsd_mtx not owned by current thread
NFSD_LOCK_DONTCARE() Advisory: this function doesn't care
NFSD_LOCK() Lock nfsd_mtx
NFSD_UNLOCK() Unlock nfsd_mtx
Constify a number of global variables/structures in the NFS server
code, as they are not modified and contain constants only:
nfsrvv2_procid nfsrv_nfsv3_procid nonidempotent
nfsv2_repstat nfsv2_type nfsrv_nfsv3_procid
nfsrvv2_procid nfsrv_v2errmap nfsv3err_null
nfsv3err_getattr nfsv3err_setattr nfsv3err_lookup
nfsv3err_access nfsv3err_readlink nfsv3err_read
nfsv3err_write nfsv3err_create nfsv3err_mkdir
nfsv3err_symlink nfsv3err_mknod nfsv3err_remove
nfsv3err_rmdir nfsv3err_rename nfsv3err_link
nfsv3err_readdir nfsv3err_readdirplus nfsv3err_fsstat
nfsv3err_fsinfo nfsv3err_pathconf nfsv3err_commit
nfsrv_v3errmap
There are additional structures that should be constified but due
to their being passed into general purpose functions without const
arguments, I have not yet converted.
In general, acquire nfsd_mtx when accessing any of the global NFS
structures, including struct nfssvc_sock, struct nfsd, struct
nfsrv_descript.
Release nfsd_mtx whenever calling into VFS, and acquire Giant for
calls into VFS. Giant is not required for any part of the
operation of the NFS server with the exception of calls into VFS.
Giant will never by acquired in the upcall code path. However, it
may operate entirely covered by Giant, or not. If debug.mpsafenet
is set to 0, the system calls will acquire Giant across all
operations, and the upcall will assert Giant. As such, by default,
this enables locking and allows us to test assertions, but should not
cause any substantial new amount of code to be run without Giant.
Bugs should manifest in the form of lock assertion failures for now.
This approach is similar (but not identical) to modifications to the
BSD/OS NFS server code snapshot provided by BSDi as part of their
SMPng snapshot. The strategy is almost the same (single lock over
the NFS server), but differs in the following ways:
- Our NFS client and server code bases don't overlap, which means
both fewer bugs and easier locking (thanks Peter!). Also means
NFSD_*() as opposed to NFS_*().
- We make broad use of assertions, whereas the BSD/OS code does not.
- Made slightly different choices about how to handle macros building
packets but operating with side effects.
- We acquire Giant only when entering VFS from the NFS server daemon
threads.
- Serious bugs in BSD/OS implementation corrected -- the snapshot we
received was clearly a work in progress.
Based on ideas from: BSDi SMPng Snapshot
Reviewed by: rick@snowhite.cis.uoguelph.ca
Extensive testing by: kris
2004-05-24 04:06:14 +00:00
|
|
|
NFSD_UNLOCK();
|
1998-05-31 17:27:58 +00:00
|
|
|
m2 = m_copym(m, 0, slp->ns_reclen - len,
|
|
|
|
|
waitflag);
|
The socket code upcalls into the NFS server using the so_upcall
mechanism so that early processing on mbufs can be performed before
a context switch to the NFS server threads. Because of this, if
the socket code is running without Giant, the NFS server also needs
to be able to run the upcall code without relying on the presence on
Giant. This change modifies the NFS server to run using a "giant
code lock" covering operation of the whole subsystem. Work is in
progress to move to data-based locking as part of the NFSv4 server
changes.
Introduce an NFS server subsystem lock, 'nfsd_mtx', and a set of
macros to operate on the lock:
NFSD_LOCK_ASSERT() Assert nfsd_mtx owned by current thread
NFSD_UNLOCK_ASSERT() Assert nfsd_mtx not owned by current thread
NFSD_LOCK_DONTCARE() Advisory: this function doesn't care
NFSD_LOCK() Lock nfsd_mtx
NFSD_UNLOCK() Unlock nfsd_mtx
Constify a number of global variables/structures in the NFS server
code, as they are not modified and contain constants only:
nfsrvv2_procid nfsrv_nfsv3_procid nonidempotent
nfsv2_repstat nfsv2_type nfsrv_nfsv3_procid
nfsrvv2_procid nfsrv_v2errmap nfsv3err_null
nfsv3err_getattr nfsv3err_setattr nfsv3err_lookup
nfsv3err_access nfsv3err_readlink nfsv3err_read
nfsv3err_write nfsv3err_create nfsv3err_mkdir
nfsv3err_symlink nfsv3err_mknod nfsv3err_remove
nfsv3err_rmdir nfsv3err_rename nfsv3err_link
nfsv3err_readdir nfsv3err_readdirplus nfsv3err_fsstat
nfsv3err_fsinfo nfsv3err_pathconf nfsv3err_commit
nfsrv_v3errmap
There are additional structures that should be constified but due
to their being passed into general purpose functions without const
arguments, I have not yet converted.
In general, acquire nfsd_mtx when accessing any of the global NFS
structures, including struct nfssvc_sock, struct nfsd, struct
nfsrv_descript.
Release nfsd_mtx whenever calling into VFS, and acquire Giant for
calls into VFS. Giant is not required for any part of the
operation of the NFS server with the exception of calls into VFS.
Giant will never by acquired in the upcall code path. However, it
may operate entirely covered by Giant, or not. If debug.mpsafenet
is set to 0, the system calls will acquire Giant across all
operations, and the upcall will assert Giant. As such, by default,
this enables locking and allows us to test assertions, but should not
cause any substantial new amount of code to be run without Giant.
Bugs should manifest in the form of lock assertion failures for now.
This approach is similar (but not identical) to modifications to the
BSD/OS NFS server code snapshot provided by BSDi as part of their
SMPng snapshot. The strategy is almost the same (single lock over
the NFS server), but differs in the following ways:
- Our NFS client and server code bases don't overlap, which means
both fewer bugs and easier locking (thanks Peter!). Also means
NFSD_*() as opposed to NFS_*().
- We make broad use of assertions, whereas the BSD/OS code does not.
- Made slightly different choices about how to handle macros building
packets but operating with side effects.
- We acquire Giant only when entering VFS from the NFS server daemon
threads.
- Serious bugs in BSD/OS implementation corrected -- the snapshot we
received was clearly a work in progress.
Based on ideas from: BSDi SMPng Snapshot
Reviewed by: rick@snowhite.cis.uoguelph.ca
Extensive testing by: kris
2004-05-24 04:06:14 +00:00
|
|
|
NFSD_LOCK();
|
1998-05-31 17:27:58 +00:00
|
|
|
if (m2) {
|
|
|
|
|
if (om) {
|
|
|
|
|
om->m_next = m2;
|
|
|
|
|
recm = slp->ns_raw;
|
|
|
|
|
} else
|
|
|
|
|
recm = m2;
|
|
|
|
|
m->m_data += slp->ns_reclen - len;
|
|
|
|
|
m->m_len -= slp->ns_reclen - len;
|
|
|
|
|
len = slp->ns_reclen;
|
|
|
|
|
} else {
|
|
|
|
|
slp->ns_flag &= ~SLP_GETSTREAM;
|
|
|
|
|
return (EWOULDBLOCK);
|
|
|
|
|
}
|
|
|
|
|
} else if ((len + m->m_len) == slp->ns_reclen) {
|
|
|
|
|
om = m;
|
|
|
|
|
len += m->m_len;
|
|
|
|
|
m = m->m_next;
|
|
|
|
|
recm = slp->ns_raw;
|
2002-07-11 17:54:58 +00:00
|
|
|
om->m_next = NULL;
|
1998-05-31 17:27:58 +00:00
|
|
|
} else {
|
|
|
|
|
om = m;
|
|
|
|
|
len += m->m_len;
|
|
|
|
|
m = m->m_next;
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
slp->ns_raw = m;
|
|
|
|
|
slp->ns_cc -= len;
|
|
|
|
|
slp->ns_reclen = 0;
|
|
|
|
|
} else {
|
|
|
|
|
slp->ns_flag &= ~SLP_GETSTREAM;
|
1995-06-27 11:07:30 +00:00
|
|
|
return (0);
|
1998-05-31 17:27:58 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
* Accumulate the fragments into a record.
|
|
|
|
|
*/
|
|
|
|
|
mpp = &slp->ns_frag;
|
|
|
|
|
while (*mpp)
|
|
|
|
|
mpp = &((*mpp)->m_next);
|
|
|
|
|
*mpp = recm;
|
|
|
|
|
if (slp->ns_flag & SLP_LASTFRAG) {
|
|
|
|
|
struct nfsrv_rec *rec;
|
The socket code upcalls into the NFS server using the so_upcall
mechanism so that early processing on mbufs can be performed before
a context switch to the NFS server threads. Because of this, if
the socket code is running without Giant, the NFS server also needs
to be able to run the upcall code without relying on the presence on
Giant. This change modifies the NFS server to run using a "giant
code lock" covering operation of the whole subsystem. Work is in
progress to move to data-based locking as part of the NFSv4 server
changes.
Introduce an NFS server subsystem lock, 'nfsd_mtx', and a set of
macros to operate on the lock:
NFSD_LOCK_ASSERT() Assert nfsd_mtx owned by current thread
NFSD_UNLOCK_ASSERT() Assert nfsd_mtx not owned by current thread
NFSD_LOCK_DONTCARE() Advisory: this function doesn't care
NFSD_LOCK() Lock nfsd_mtx
NFSD_UNLOCK() Unlock nfsd_mtx
Constify a number of global variables/structures in the NFS server
code, as they are not modified and contain constants only:
nfsrvv2_procid nfsrv_nfsv3_procid nonidempotent
nfsv2_repstat nfsv2_type nfsrv_nfsv3_procid
nfsrvv2_procid nfsrv_v2errmap nfsv3err_null
nfsv3err_getattr nfsv3err_setattr nfsv3err_lookup
nfsv3err_access nfsv3err_readlink nfsv3err_read
nfsv3err_write nfsv3err_create nfsv3err_mkdir
nfsv3err_symlink nfsv3err_mknod nfsv3err_remove
nfsv3err_rmdir nfsv3err_rename nfsv3err_link
nfsv3err_readdir nfsv3err_readdirplus nfsv3err_fsstat
nfsv3err_fsinfo nfsv3err_pathconf nfsv3err_commit
nfsrv_v3errmap
There are additional structures that should be constified but due
to their being passed into general purpose functions without const
arguments, I have not yet converted.
In general, acquire nfsd_mtx when accessing any of the global NFS
structures, including struct nfssvc_sock, struct nfsd, struct
nfsrv_descript.
Release nfsd_mtx whenever calling into VFS, and acquire Giant for
calls into VFS. Giant is not required for any part of the
operation of the NFS server with the exception of calls into VFS.
Giant will never by acquired in the upcall code path. However, it
may operate entirely covered by Giant, or not. If debug.mpsafenet
is set to 0, the system calls will acquire Giant across all
operations, and the upcall will assert Giant. As such, by default,
this enables locking and allows us to test assertions, but should not
cause any substantial new amount of code to be run without Giant.
Bugs should manifest in the form of lock assertion failures for now.
This approach is similar (but not identical) to modifications to the
BSD/OS NFS server code snapshot provided by BSDi as part of their
SMPng snapshot. The strategy is almost the same (single lock over
the NFS server), but differs in the following ways:
- Our NFS client and server code bases don't overlap, which means
both fewer bugs and easier locking (thanks Peter!). Also means
NFSD_*() as opposed to NFS_*().
- We make broad use of assertions, whereas the BSD/OS code does not.
- Made slightly different choices about how to handle macros building
packets but operating with side effects.
- We acquire Giant only when entering VFS from the NFS server daemon
threads.
- Serious bugs in BSD/OS implementation corrected -- the snapshot we
received was clearly a work in progress.
Based on ideas from: BSDi SMPng Snapshot
Reviewed by: rick@snowhite.cis.uoguelph.ca
Extensive testing by: kris
2004-05-24 04:06:14 +00:00
|
|
|
NFSD_UNLOCK();
|
2003-02-22 10:35:37 +00:00
|
|
|
rec = malloc(sizeof(struct nfsrv_rec), M_NFSRVDESC,
|
|
|
|
|
waitflag == M_DONTWAIT ? M_NOWAIT : M_WAITOK);
|
2008-03-18 18:42:59 +00:00
|
|
|
if (rec) {
|
The VFS/BIO subsystem contained a number of hacks in order to optimize
piecemeal, middle-of-file writes for NFS. These hacks have caused no
end of trouble, especially when combined with mmap(). I've removed
them. Instead, NFS will issue a read-before-write to fully
instantiate the struct buf containing the write. NFS does, however,
optimize piecemeal appends to files. For most common file operations,
you will not notice the difference. The sole remaining fragment in
the VFS/BIO system is b_dirtyoff/end, which NFS uses to avoid cache
coherency issues with read-merge-write style operations. NFS also
optimizes the write-covers-entire-buffer case by avoiding the
read-before-write. There is quite a bit of room for further
optimization in these areas.
The VM system marks pages fully-valid (AKA vm_page_t->valid =
VM_PAGE_BITS_ALL) in several places, most noteably in vm_fault. This
is not correct operation. The vm_pager_get_pages() code is now
responsible for marking VM pages all-valid. A number of VM helper
routines have been added to aid in zeroing-out the invalid portions of
a VM page prior to the page being marked all-valid. This operation is
necessary to properly support mmap(). The zeroing occurs most often
when dealing with file-EOF situations. Several bugs have been fixed
in the NFS subsystem, including bits handling file and directory EOF
situations and buf->b_flags consistancy issues relating to clearing
B_ERROR & B_INVAL, and handling B_DONE.
getblk() and allocbuf() have been rewritten. B_CACHE operation is now
formally defined in comments and more straightforward in
implementation. B_CACHE for VMIO buffers is based on the validity of
the backing store. B_CACHE for non-VMIO buffers is based simply on
whether the buffer is B_INVAL or not (B_CACHE set if B_INVAL clear,
and vise-versa). biodone() is now responsible for setting B_CACHE
when a successful read completes. B_CACHE is also set when a bdwrite()
is initiated and when a bwrite() is initiated. VFS VOP_BWRITE
routines (there are only two - nfs_bwrite() and bwrite()) are now
expected to set B_CACHE. This means that bowrite() and bawrite() also
set B_CACHE indirectly.
There are a number of places in the code which were previously using
buf->b_bufsize (which is DEV_BSIZE aligned) when they should have
been using buf->b_bcount. These have been fixed. getblk() now clears
B_DONE on return because the rest of the system is so bad about
dealing with B_DONE.
Major fixes to NFS/TCP have been made. A server-side bug could cause
requests to be lost by the server due to nfs_realign() overwriting
other rpc's in the same TCP mbuf chain. The server's kernel must be
recompiled to get the benefit of the fixes.
Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
1999-05-02 23:57:16 +00:00
|
|
|
nfs_realign(&slp->ns_frag, 10 * NFSX_UNSIGNED);
|
2002-07-11 17:54:58 +00:00
|
|
|
rec->nr_address = NULL;
|
1998-05-31 17:27:58 +00:00
|
|
|
rec->nr_packet = slp->ns_frag;
|
2008-03-19 12:33:25 +00:00
|
|
|
NFSD_LOCK();
|
1998-05-31 17:27:58 +00:00
|
|
|
STAILQ_INSERT_TAIL(&slp->ns_rec, rec, nr_link);
|
2008-03-19 12:33:25 +00:00
|
|
|
} else {
|
|
|
|
|
NFSD_LOCK();
|
1998-05-31 17:27:58 +00:00
|
|
|
}
|
2008-03-18 18:42:59 +00:00
|
|
|
if (!rec) {
|
|
|
|
|
m_freem(slp->ns_frag);
|
|
|
|
|
}
|
2002-07-11 17:54:58 +00:00
|
|
|
slp->ns_frag = NULL;
|
1998-05-31 17:27:58 +00:00
|
|
|
}
|
1994-05-24 10:09:53 +00:00
|
|
|
}
|
1998-05-31 17:27:58 +00:00
|
|
|
}
|
1994-05-24 10:09:53 +00:00
|
|
|
|
1998-05-31 17:27:58 +00:00
|
|
|
/*
|
|
|
|
|
* Parse an RPC header.
|
|
|
|
|
*/
|
|
|
|
|
int
|
2001-09-18 23:32:09 +00:00
|
|
|
nfsrv_dorec(struct nfssvc_sock *slp, struct nfsd *nfsd,
|
|
|
|
|
struct nfsrv_descript **ndp)
|
1998-05-31 17:27:58 +00:00
|
|
|
{
|
|
|
|
|
struct nfsrv_rec *rec;
|
2001-09-18 23:32:09 +00:00
|
|
|
struct mbuf *m;
|
1998-05-31 17:27:58 +00:00
|
|
|
struct sockaddr *nam;
|
2001-09-18 23:32:09 +00:00
|
|
|
struct nfsrv_descript *nd;
|
1998-05-31 17:27:58 +00:00
|
|
|
int error;
|
|
|
|
|
|
The socket code upcalls into the NFS server using the so_upcall
mechanism so that early processing on mbufs can be performed before
a context switch to the NFS server threads. Because of this, if
the socket code is running without Giant, the NFS server also needs
to be able to run the upcall code without relying on the presence on
Giant. This change modifies the NFS server to run using a "giant
code lock" covering operation of the whole subsystem. Work is in
progress to move to data-based locking as part of the NFSv4 server
changes.
Introduce an NFS server subsystem lock, 'nfsd_mtx', and a set of
macros to operate on the lock:
NFSD_LOCK_ASSERT() Assert nfsd_mtx owned by current thread
NFSD_UNLOCK_ASSERT() Assert nfsd_mtx not owned by current thread
NFSD_LOCK_DONTCARE() Advisory: this function doesn't care
NFSD_LOCK() Lock nfsd_mtx
NFSD_UNLOCK() Unlock nfsd_mtx
Constify a number of global variables/structures in the NFS server
code, as they are not modified and contain constants only:
nfsrvv2_procid nfsrv_nfsv3_procid nonidempotent
nfsv2_repstat nfsv2_type nfsrv_nfsv3_procid
nfsrvv2_procid nfsrv_v2errmap nfsv3err_null
nfsv3err_getattr nfsv3err_setattr nfsv3err_lookup
nfsv3err_access nfsv3err_readlink nfsv3err_read
nfsv3err_write nfsv3err_create nfsv3err_mkdir
nfsv3err_symlink nfsv3err_mknod nfsv3err_remove
nfsv3err_rmdir nfsv3err_rename nfsv3err_link
nfsv3err_readdir nfsv3err_readdirplus nfsv3err_fsstat
nfsv3err_fsinfo nfsv3err_pathconf nfsv3err_commit
nfsrv_v3errmap
There are additional structures that should be constified but due
to their being passed into general purpose functions without const
arguments, I have not yet converted.
In general, acquire nfsd_mtx when accessing any of the global NFS
structures, including struct nfssvc_sock, struct nfsd, struct
nfsrv_descript.
Release nfsd_mtx whenever calling into VFS, and acquire Giant for
calls into VFS. Giant is not required for any part of the
operation of the NFS server with the exception of calls into VFS.
Giant will never by acquired in the upcall code path. However, it
may operate entirely covered by Giant, or not. If debug.mpsafenet
is set to 0, the system calls will acquire Giant across all
operations, and the upcall will assert Giant. As such, by default,
this enables locking and allows us to test assertions, but should not
cause any substantial new amount of code to be run without Giant.
Bugs should manifest in the form of lock assertion failures for now.
This approach is similar (but not identical) to modifications to the
BSD/OS NFS server code snapshot provided by BSDi as part of their
SMPng snapshot. The strategy is almost the same (single lock over
the NFS server), but differs in the following ways:
- Our NFS client and server code bases don't overlap, which means
both fewer bugs and easier locking (thanks Peter!). Also means
NFSD_*() as opposed to NFS_*().
- We make broad use of assertions, whereas the BSD/OS code does not.
- Made slightly different choices about how to handle macros building
packets but operating with side effects.
- We acquire Giant only when entering VFS from the NFS server daemon
threads.
- Serious bugs in BSD/OS implementation corrected -- the snapshot we
received was clearly a work in progress.
Based on ideas from: BSDi SMPng Snapshot
Reviewed by: rick@snowhite.cis.uoguelph.ca
Extensive testing by: kris
2004-05-24 04:06:14 +00:00
|
|
|
NFSD_LOCK_ASSERT();
|
|
|
|
|
|
1998-05-31 17:27:58 +00:00
|
|
|
*ndp = NULL;
|
2004-04-04 19:13:35 +00:00
|
|
|
if ((slp->ns_flag & SLP_VALID) == 0 ||
|
|
|
|
|
STAILQ_FIRST(&slp->ns_rec) == NULL)
|
1998-05-31 17:27:58 +00:00
|
|
|
return (ENOBUFS);
|
|
|
|
|
rec = STAILQ_FIRST(&slp->ns_rec);
|
2006-03-08 20:21:15 +00:00
|
|
|
KASSERT(rec->nr_packet != NULL, ("nfsrv_dorec: missing mbuf"));
|
1998-05-31 17:27:58 +00:00
|
|
|
STAILQ_REMOVE_HEAD(&slp->ns_rec, nr_link);
|
|
|
|
|
nam = rec->nr_address;
|
|
|
|
|
m = rec->nr_packet;
|
|
|
|
|
free(rec, M_NFSRVDESC);
|
The socket code upcalls into the NFS server using the so_upcall
mechanism so that early processing on mbufs can be performed before
a context switch to the NFS server threads. Because of this, if
the socket code is running without Giant, the NFS server also needs
to be able to run the upcall code without relying on the presence on
Giant. This change modifies the NFS server to run using a "giant
code lock" covering operation of the whole subsystem. Work is in
progress to move to data-based locking as part of the NFSv4 server
changes.
Introduce an NFS server subsystem lock, 'nfsd_mtx', and a set of
macros to operate on the lock:
NFSD_LOCK_ASSERT() Assert nfsd_mtx owned by current thread
NFSD_UNLOCK_ASSERT() Assert nfsd_mtx not owned by current thread
NFSD_LOCK_DONTCARE() Advisory: this function doesn't care
NFSD_LOCK() Lock nfsd_mtx
NFSD_UNLOCK() Unlock nfsd_mtx
Constify a number of global variables/structures in the NFS server
code, as they are not modified and contain constants only:
nfsrvv2_procid nfsrv_nfsv3_procid nonidempotent
nfsv2_repstat nfsv2_type nfsrv_nfsv3_procid
nfsrvv2_procid nfsrv_v2errmap nfsv3err_null
nfsv3err_getattr nfsv3err_setattr nfsv3err_lookup
nfsv3err_access nfsv3err_readlink nfsv3err_read
nfsv3err_write nfsv3err_create nfsv3err_mkdir
nfsv3err_symlink nfsv3err_mknod nfsv3err_remove
nfsv3err_rmdir nfsv3err_rename nfsv3err_link
nfsv3err_readdir nfsv3err_readdirplus nfsv3err_fsstat
nfsv3err_fsinfo nfsv3err_pathconf nfsv3err_commit
nfsrv_v3errmap
There are additional structures that should be constified but due
to their being passed into general purpose functions without const
arguments, I have not yet converted.
In general, acquire nfsd_mtx when accessing any of the global NFS
structures, including struct nfssvc_sock, struct nfsd, struct
nfsrv_descript.
Release nfsd_mtx whenever calling into VFS, and acquire Giant for
calls into VFS. Giant is not required for any part of the
operation of the NFS server with the exception of calls into VFS.
Giant will never by acquired in the upcall code path. However, it
may operate entirely covered by Giant, or not. If debug.mpsafenet
is set to 0, the system calls will acquire Giant across all
operations, and the upcall will assert Giant. As such, by default,
this enables locking and allows us to test assertions, but should not
cause any substantial new amount of code to be run without Giant.
Bugs should manifest in the form of lock assertion failures for now.
This approach is similar (but not identical) to modifications to the
BSD/OS NFS server code snapshot provided by BSDi as part of their
SMPng snapshot. The strategy is almost the same (single lock over
the NFS server), but differs in the following ways:
- Our NFS client and server code bases don't overlap, which means
both fewer bugs and easier locking (thanks Peter!). Also means
NFSD_*() as opposed to NFS_*().
- We make broad use of assertions, whereas the BSD/OS code does not.
- Made slightly different choices about how to handle macros building
packets but operating with side effects.
- We acquire Giant only when entering VFS from the NFS server daemon
threads.
- Serious bugs in BSD/OS implementation corrected -- the snapshot we
received was clearly a work in progress.
Based on ideas from: BSDi SMPng Snapshot
Reviewed by: rick@snowhite.cis.uoguelph.ca
Extensive testing by: kris
2004-05-24 04:06:14 +00:00
|
|
|
NFSD_UNLOCK();
|
2008-10-23 15:53:51 +00:00
|
|
|
nd = malloc(sizeof (struct nfsrv_descript),
|
2003-02-19 05:47:46 +00:00
|
|
|
M_NFSRVDESC, M_WAITOK);
|
2006-01-28 19:24:40 +00:00
|
|
|
nd->nd_cr = crget();
|
2009-05-27 14:11:23 +00:00
|
|
|
prison_hold(&prison0);
|
|
|
|
|
nd->nd_cr->cr_prison = &prison0;
|
The socket code upcalls into the NFS server using the so_upcall
mechanism so that early processing on mbufs can be performed before
a context switch to the NFS server threads. Because of this, if
the socket code is running without Giant, the NFS server also needs
to be able to run the upcall code without relying on the presence on
Giant. This change modifies the NFS server to run using a "giant
code lock" covering operation of the whole subsystem. Work is in
progress to move to data-based locking as part of the NFSv4 server
changes.
Introduce an NFS server subsystem lock, 'nfsd_mtx', and a set of
macros to operate on the lock:
NFSD_LOCK_ASSERT() Assert nfsd_mtx owned by current thread
NFSD_UNLOCK_ASSERT() Assert nfsd_mtx not owned by current thread
NFSD_LOCK_DONTCARE() Advisory: this function doesn't care
NFSD_LOCK() Lock nfsd_mtx
NFSD_UNLOCK() Unlock nfsd_mtx
Constify a number of global variables/structures in the NFS server
code, as they are not modified and contain constants only:
nfsrvv2_procid nfsrv_nfsv3_procid nonidempotent
nfsv2_repstat nfsv2_type nfsrv_nfsv3_procid
nfsrvv2_procid nfsrv_v2errmap nfsv3err_null
nfsv3err_getattr nfsv3err_setattr nfsv3err_lookup
nfsv3err_access nfsv3err_readlink nfsv3err_read
nfsv3err_write nfsv3err_create nfsv3err_mkdir
nfsv3err_symlink nfsv3err_mknod nfsv3err_remove
nfsv3err_rmdir nfsv3err_rename nfsv3err_link
nfsv3err_readdir nfsv3err_readdirplus nfsv3err_fsstat
nfsv3err_fsinfo nfsv3err_pathconf nfsv3err_commit
nfsrv_v3errmap
There are additional structures that should be constified but due
to their being passed into general purpose functions without const
arguments, I have not yet converted.
In general, acquire nfsd_mtx when accessing any of the global NFS
structures, including struct nfssvc_sock, struct nfsd, struct
nfsrv_descript.
Release nfsd_mtx whenever calling into VFS, and acquire Giant for
calls into VFS. Giant is not required for any part of the
operation of the NFS server with the exception of calls into VFS.
Giant will never by acquired in the upcall code path. However, it
may operate entirely covered by Giant, or not. If debug.mpsafenet
is set to 0, the system calls will acquire Giant across all
operations, and the upcall will assert Giant. As such, by default,
this enables locking and allows us to test assertions, but should not
cause any substantial new amount of code to be run without Giant.
Bugs should manifest in the form of lock assertion failures for now.
This approach is similar (but not identical) to modifications to the
BSD/OS NFS server code snapshot provided by BSDi as part of their
SMPng snapshot. The strategy is almost the same (single lock over
the NFS server), but differs in the following ways:
- Our NFS client and server code bases don't overlap, which means
both fewer bugs and easier locking (thanks Peter!). Also means
NFSD_*() as opposed to NFS_*().
- We make broad use of assertions, whereas the BSD/OS code does not.
- Made slightly different choices about how to handle macros building
packets but operating with side effects.
- We acquire Giant only when entering VFS from the NFS server daemon
threads.
- Serious bugs in BSD/OS implementation corrected -- the snapshot we
received was clearly a work in progress.
Based on ideas from: BSDi SMPng Snapshot
Reviewed by: rick@snowhite.cis.uoguelph.ca
Extensive testing by: kris
2004-05-24 04:06:14 +00:00
|
|
|
NFSD_LOCK();
|
1998-05-31 17:27:58 +00:00
|
|
|
nd->nd_md = nd->nd_mrep = m;
|
|
|
|
|
nd->nd_nam2 = nam;
|
|
|
|
|
nd->nd_dpos = mtod(m, caddr_t);
|
|
|
|
|
error = nfs_getreq(nd, nfsd, TRUE);
|
|
|
|
|
if (error) {
|
1998-11-13 09:44:12 +00:00
|
|
|
if (nam) {
|
2008-10-23 15:53:51 +00:00
|
|
|
free(nam, M_SONAME);
|
1998-11-13 09:44:12 +00:00
|
|
|
}
|
2006-01-28 19:24:40 +00:00
|
|
|
if (nd->nd_cr != NULL)
|
|
|
|
|
crfree(nd->nd_cr);
|
1998-05-31 17:27:58 +00:00
|
|
|
free((caddr_t)nd, M_NFSRVDESC);
|
|
|
|
|
return (error);
|
|
|
|
|
}
|
|
|
|
|
*ndp = nd;
|
|
|
|
|
nfsd->nfsd_nd = nd;
|
1994-05-24 10:09:53 +00:00
|
|
|
return (0);
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
* Search for a sleeping nfsd and wake it up.
|
|
|
|
|
* SIDE EFFECT: If none found, set NFSD_CHECKSLP flag, so that one of the
|
|
|
|
|
* running nfsds will go look for the work in the nfssvc_sock list.
|
|
|
|
|
*/
|
|
|
|
|
void
|
2001-09-18 23:32:09 +00:00
|
|
|
nfsrv_wakenfsd(struct nfssvc_sock *slp)
|
1994-05-24 10:09:53 +00:00
|
|
|
{
|
2001-09-18 23:32:09 +00:00
|
|
|
struct nfsd *nd;
|
1994-05-24 10:09:53 +00:00
|
|
|
|
The socket code upcalls into the NFS server using the so_upcall
mechanism so that early processing on mbufs can be performed before
a context switch to the NFS server threads. Because of this, if
the socket code is running without Giant, the NFS server also needs
to be able to run the upcall code without relying on the presence on
Giant. This change modifies the NFS server to run using a "giant
code lock" covering operation of the whole subsystem. Work is in
progress to move to data-based locking as part of the NFSv4 server
changes.
Introduce an NFS server subsystem lock, 'nfsd_mtx', and a set of
macros to operate on the lock:
NFSD_LOCK_ASSERT() Assert nfsd_mtx owned by current thread
NFSD_UNLOCK_ASSERT() Assert nfsd_mtx not owned by current thread
NFSD_LOCK_DONTCARE() Advisory: this function doesn't care
NFSD_LOCK() Lock nfsd_mtx
NFSD_UNLOCK() Unlock nfsd_mtx
Constify a number of global variables/structures in the NFS server
code, as they are not modified and contain constants only:
nfsrvv2_procid nfsrv_nfsv3_procid nonidempotent
nfsv2_repstat nfsv2_type nfsrv_nfsv3_procid
nfsrvv2_procid nfsrv_v2errmap nfsv3err_null
nfsv3err_getattr nfsv3err_setattr nfsv3err_lookup
nfsv3err_access nfsv3err_readlink nfsv3err_read
nfsv3err_write nfsv3err_create nfsv3err_mkdir
nfsv3err_symlink nfsv3err_mknod nfsv3err_remove
nfsv3err_rmdir nfsv3err_rename nfsv3err_link
nfsv3err_readdir nfsv3err_readdirplus nfsv3err_fsstat
nfsv3err_fsinfo nfsv3err_pathconf nfsv3err_commit
nfsrv_v3errmap
There are additional structures that should be constified but due
to their being passed into general purpose functions without const
arguments, I have not yet converted.
In general, acquire nfsd_mtx when accessing any of the global NFS
structures, including struct nfssvc_sock, struct nfsd, struct
nfsrv_descript.
Release nfsd_mtx whenever calling into VFS, and acquire Giant for
calls into VFS. Giant is not required for any part of the
operation of the NFS server with the exception of calls into VFS.
Giant will never by acquired in the upcall code path. However, it
may operate entirely covered by Giant, or not. If debug.mpsafenet
is set to 0, the system calls will acquire Giant across all
operations, and the upcall will assert Giant. As such, by default,
this enables locking and allows us to test assertions, but should not
cause any substantial new amount of code to be run without Giant.
Bugs should manifest in the form of lock assertion failures for now.
This approach is similar (but not identical) to modifications to the
BSD/OS NFS server code snapshot provided by BSDi as part of their
SMPng snapshot. The strategy is almost the same (single lock over
the NFS server), but differs in the following ways:
- Our NFS client and server code bases don't overlap, which means
both fewer bugs and easier locking (thanks Peter!). Also means
NFSD_*() as opposed to NFS_*().
- We make broad use of assertions, whereas the BSD/OS code does not.
- Made slightly different choices about how to handle macros building
packets but operating with side effects.
- We acquire Giant only when entering VFS from the NFS server daemon
threads.
- Serious bugs in BSD/OS implementation corrected -- the snapshot we
received was clearly a work in progress.
Based on ideas from: BSDi SMPng Snapshot
Reviewed by: rick@snowhite.cis.uoguelph.ca
Extensive testing by: kris
2004-05-24 04:06:14 +00:00
|
|
|
NFSD_LOCK_ASSERT();
|
|
|
|
|
|
1994-05-24 10:09:53 +00:00
|
|
|
if ((slp->ns_flag & SLP_VALID) == 0)
|
|
|
|
|
return;
|
2001-09-18 23:32:09 +00:00
|
|
|
TAILQ_FOREACH(nd, &nfsd_head, nfsd_chain) {
|
1995-06-27 11:07:30 +00:00
|
|
|
if (nd->nfsd_flag & NFSD_WAITING) {
|
|
|
|
|
nd->nfsd_flag &= ~NFSD_WAITING;
|
|
|
|
|
if (nd->nfsd_slp)
|
1994-05-24 10:09:53 +00:00
|
|
|
panic("nfsd wakeup");
|
|
|
|
|
slp->ns_sref++;
|
1995-06-27 11:07:30 +00:00
|
|
|
nd->nfsd_slp = slp;
|
2003-03-02 16:54:40 +00:00
|
|
|
wakeup(nd);
|
1994-05-24 10:09:53 +00:00
|
|
|
return;
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
slp->ns_flag |= SLP_DOREC;
|
1994-10-17 17:47:45 +00:00
|
|
|
nfsd_head_flag |= NFSD_CHECKSLP;
|
1994-05-24 10:09:53 +00:00
|
|
|
}
|
2001-09-18 23:32:09 +00:00
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
* This is the nfs send routine.
|
|
|
|
|
* For the server side:
|
|
|
|
|
* - return EINTR or ERESTART if interrupted by a signal
|
|
|
|
|
* - return EPIPE if a connection is lost for connection based sockets (TCP...)
|
|
|
|
|
* - do any cleanup required by recoverable socket errors (?)
|
|
|
|
|
*/
|
|
|
|
|
int
|
|
|
|
|
nfsrv_send(struct socket *so, struct sockaddr *nam, struct mbuf *top)
|
|
|
|
|
{
|
|
|
|
|
struct sockaddr *sendnam;
|
|
|
|
|
int error, soflags, flags;
|
|
|
|
|
|
The socket code upcalls into the NFS server using the so_upcall
mechanism so that early processing on mbufs can be performed before
a context switch to the NFS server threads. Because of this, if
the socket code is running without Giant, the NFS server also needs
to be able to run the upcall code without relying on the presence on
Giant. This change modifies the NFS server to run using a "giant
code lock" covering operation of the whole subsystem. Work is in
progress to move to data-based locking as part of the NFSv4 server
changes.
Introduce an NFS server subsystem lock, 'nfsd_mtx', and a set of
macros to operate on the lock:
NFSD_LOCK_ASSERT() Assert nfsd_mtx owned by current thread
NFSD_UNLOCK_ASSERT() Assert nfsd_mtx not owned by current thread
NFSD_LOCK_DONTCARE() Advisory: this function doesn't care
NFSD_LOCK() Lock nfsd_mtx
NFSD_UNLOCK() Unlock nfsd_mtx
Constify a number of global variables/structures in the NFS server
code, as they are not modified and contain constants only:
nfsrvv2_procid nfsrv_nfsv3_procid nonidempotent
nfsv2_repstat nfsv2_type nfsrv_nfsv3_procid
nfsrvv2_procid nfsrv_v2errmap nfsv3err_null
nfsv3err_getattr nfsv3err_setattr nfsv3err_lookup
nfsv3err_access nfsv3err_readlink nfsv3err_read
nfsv3err_write nfsv3err_create nfsv3err_mkdir
nfsv3err_symlink nfsv3err_mknod nfsv3err_remove
nfsv3err_rmdir nfsv3err_rename nfsv3err_link
nfsv3err_readdir nfsv3err_readdirplus nfsv3err_fsstat
nfsv3err_fsinfo nfsv3err_pathconf nfsv3err_commit
nfsrv_v3errmap
There are additional structures that should be constified but due
to their being passed into general purpose functions without const
arguments, I have not yet converted.
In general, acquire nfsd_mtx when accessing any of the global NFS
structures, including struct nfssvc_sock, struct nfsd, struct
nfsrv_descript.
Release nfsd_mtx whenever calling into VFS, and acquire Giant for
calls into VFS. Giant is not required for any part of the
operation of the NFS server with the exception of calls into VFS.
Giant will never by acquired in the upcall code path. However, it
may operate entirely covered by Giant, or not. If debug.mpsafenet
is set to 0, the system calls will acquire Giant across all
operations, and the upcall will assert Giant. As such, by default,
this enables locking and allows us to test assertions, but should not
cause any substantial new amount of code to be run without Giant.
Bugs should manifest in the form of lock assertion failures for now.
This approach is similar (but not identical) to modifications to the
BSD/OS NFS server code snapshot provided by BSDi as part of their
SMPng snapshot. The strategy is almost the same (single lock over
the NFS server), but differs in the following ways:
- Our NFS client and server code bases don't overlap, which means
both fewer bugs and easier locking (thanks Peter!). Also means
NFSD_*() as opposed to NFS_*().
- We make broad use of assertions, whereas the BSD/OS code does not.
- Made slightly different choices about how to handle macros building
packets but operating with side effects.
- We acquire Giant only when entering VFS from the NFS server daemon
threads.
- Serious bugs in BSD/OS implementation corrected -- the snapshot we
received was clearly a work in progress.
Based on ideas from: BSDi SMPng Snapshot
Reviewed by: rick@snowhite.cis.uoguelph.ca
Extensive testing by: kris
2004-05-24 04:06:14 +00:00
|
|
|
NFSD_UNLOCK_ASSERT();
|
2003-11-07 22:57:09 +00:00
|
|
|
|
2001-09-18 23:32:09 +00:00
|
|
|
soflags = so->so_proto->pr_flags;
|
2002-05-31 11:52:35 +00:00
|
|
|
if ((soflags & PR_CONNREQUIRED) || (so->so_state & SS_ISCONNECTED))
|
2002-07-11 17:54:58 +00:00
|
|
|
sendnam = NULL;
|
2002-05-31 11:52:35 +00:00
|
|
|
else
|
2001-09-18 23:32:09 +00:00
|
|
|
sendnam = nam;
|
|
|
|
|
if (so->so_type == SOCK_SEQPACKET)
|
|
|
|
|
flags = MSG_EOR;
|
|
|
|
|
else
|
|
|
|
|
flags = 0;
|
|
|
|
|
|
soreceive_generic(), and sopoll_generic(). Add new functions sosend(),
soreceive(), and sopoll(), which are wrappers for pru_sosend,
pru_soreceive, and pru_sopoll, and are now used univerally by socket
consumers rather than either directly invoking the old so*() functions
or directly invoking the protocol switch method (about an even split
prior to this commit).
This completes an architectural change that was begun in 1996 to permit
protocols to provide substitute implementations, as now used by UDP.
Consumers now uniformly invoke sosend(), soreceive(), and sopoll() to
perform these operations on sockets -- in particular, distributed file
systems and socket system calls.
Architectural head nod: sam, gnn, wollman
2006-07-24 15:20:08 +00:00
|
|
|
error = sosend(so, sendnam, 0, top, 0, flags, curthread/*XXX*/);
|
2001-09-18 23:32:09 +00:00
|
|
|
if (error == ENOBUFS && so->so_type == SOCK_DGRAM)
|
|
|
|
|
error = 0;
|
|
|
|
|
|
|
|
|
|
if (error) {
|
|
|
|
|
log(LOG_INFO, "nfsd send error %d\n", error);
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
* Handle any recoverable (soft) socket errors here. (?)
|
|
|
|
|
*/
|
|
|
|
|
if (error != EINTR && error != ERESTART &&
|
|
|
|
|
error != EWOULDBLOCK && error != EPIPE)
|
|
|
|
|
error = 0;
|
|
|
|
|
}
|
|
|
|
|
return (error);
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
/*
|
2001-11-15 23:49:02 +00:00
|
|
|
* NFS server timer routine.
|
2001-09-18 23:32:09 +00:00
|
|
|
*/
|
|
|
|
|
void
|
|
|
|
|
nfsrv_timer(void *arg)
|
|
|
|
|
{
|
|
|
|
|
struct nfssvc_sock *slp;
|
|
|
|
|
u_quad_t cur_usec;
|
|
|
|
|
|
The socket code upcalls into the NFS server using the so_upcall
mechanism so that early processing on mbufs can be performed before
a context switch to the NFS server threads. Because of this, if
the socket code is running without Giant, the NFS server also needs
to be able to run the upcall code without relying on the presence on
Giant. This change modifies the NFS server to run using a "giant
code lock" covering operation of the whole subsystem. Work is in
progress to move to data-based locking as part of the NFSv4 server
changes.
Introduce an NFS server subsystem lock, 'nfsd_mtx', and a set of
macros to operate on the lock:
NFSD_LOCK_ASSERT() Assert nfsd_mtx owned by current thread
NFSD_UNLOCK_ASSERT() Assert nfsd_mtx not owned by current thread
NFSD_LOCK_DONTCARE() Advisory: this function doesn't care
NFSD_LOCK() Lock nfsd_mtx
NFSD_UNLOCK() Unlock nfsd_mtx
Constify a number of global variables/structures in the NFS server
code, as they are not modified and contain constants only:
nfsrvv2_procid nfsrv_nfsv3_procid nonidempotent
nfsv2_repstat nfsv2_type nfsrv_nfsv3_procid
nfsrvv2_procid nfsrv_v2errmap nfsv3err_null
nfsv3err_getattr nfsv3err_setattr nfsv3err_lookup
nfsv3err_access nfsv3err_readlink nfsv3err_read
nfsv3err_write nfsv3err_create nfsv3err_mkdir
nfsv3err_symlink nfsv3err_mknod nfsv3err_remove
nfsv3err_rmdir nfsv3err_rename nfsv3err_link
nfsv3err_readdir nfsv3err_readdirplus nfsv3err_fsstat
nfsv3err_fsinfo nfsv3err_pathconf nfsv3err_commit
nfsrv_v3errmap
There are additional structures that should be constified but due
to their being passed into general purpose functions without const
arguments, I have not yet converted.
In general, acquire nfsd_mtx when accessing any of the global NFS
structures, including struct nfssvc_sock, struct nfsd, struct
nfsrv_descript.
Release nfsd_mtx whenever calling into VFS, and acquire Giant for
calls into VFS. Giant is not required for any part of the
operation of the NFS server with the exception of calls into VFS.
Giant will never by acquired in the upcall code path. However, it
may operate entirely covered by Giant, or not. If debug.mpsafenet
is set to 0, the system calls will acquire Giant across all
operations, and the upcall will assert Giant. As such, by default,
this enables locking and allows us to test assertions, but should not
cause any substantial new amount of code to be run without Giant.
Bugs should manifest in the form of lock assertion failures for now.
This approach is similar (but not identical) to modifications to the
BSD/OS NFS server code snapshot provided by BSDi as part of their
SMPng snapshot. The strategy is almost the same (single lock over
the NFS server), but differs in the following ways:
- Our NFS client and server code bases don't overlap, which means
both fewer bugs and easier locking (thanks Peter!). Also means
NFSD_*() as opposed to NFS_*().
- We make broad use of assertions, whereas the BSD/OS code does not.
- Made slightly different choices about how to handle macros building
packets but operating with side effects.
- We acquire Giant only when entering VFS from the NFS server daemon
threads.
- Serious bugs in BSD/OS implementation corrected -- the snapshot we
received was clearly a work in progress.
Based on ideas from: BSDi SMPng Snapshot
Reviewed by: rick@snowhite.cis.uoguelph.ca
Extensive testing by: kris
2004-05-24 04:06:14 +00:00
|
|
|
NFSD_LOCK();
|
2001-09-18 23:32:09 +00:00
|
|
|
/*
|
|
|
|
|
* Scan the write gathering queues for writes that need to be
|
|
|
|
|
* completed now.
|
|
|
|
|
*/
|
|
|
|
|
cur_usec = nfs_curusec();
|
|
|
|
|
TAILQ_FOREACH(slp, &nfssvc_sockhead, ns_chain) {
|
2001-11-15 23:49:02 +00:00
|
|
|
if (LIST_FIRST(&slp->ns_tq) &&
|
|
|
|
|
LIST_FIRST(&slp->ns_tq)->nd_time <= cur_usec)
|
|
|
|
|
nfsrv_wakenfsd(slp);
|
2001-09-18 23:32:09 +00:00
|
|
|
}
|
The socket code upcalls into the NFS server using the so_upcall
mechanism so that early processing on mbufs can be performed before
a context switch to the NFS server threads. Because of this, if
the socket code is running without Giant, the NFS server also needs
to be able to run the upcall code without relying on the presence on
Giant. This change modifies the NFS server to run using a "giant
code lock" covering operation of the whole subsystem. Work is in
progress to move to data-based locking as part of the NFSv4 server
changes.
Introduce an NFS server subsystem lock, 'nfsd_mtx', and a set of
macros to operate on the lock:
NFSD_LOCK_ASSERT() Assert nfsd_mtx owned by current thread
NFSD_UNLOCK_ASSERT() Assert nfsd_mtx not owned by current thread
NFSD_LOCK_DONTCARE() Advisory: this function doesn't care
NFSD_LOCK() Lock nfsd_mtx
NFSD_UNLOCK() Unlock nfsd_mtx
Constify a number of global variables/structures in the NFS server
code, as they are not modified and contain constants only:
nfsrvv2_procid nfsrv_nfsv3_procid nonidempotent
nfsv2_repstat nfsv2_type nfsrv_nfsv3_procid
nfsrvv2_procid nfsrv_v2errmap nfsv3err_null
nfsv3err_getattr nfsv3err_setattr nfsv3err_lookup
nfsv3err_access nfsv3err_readlink nfsv3err_read
nfsv3err_write nfsv3err_create nfsv3err_mkdir
nfsv3err_symlink nfsv3err_mknod nfsv3err_remove
nfsv3err_rmdir nfsv3err_rename nfsv3err_link
nfsv3err_readdir nfsv3err_readdirplus nfsv3err_fsstat
nfsv3err_fsinfo nfsv3err_pathconf nfsv3err_commit
nfsrv_v3errmap
There are additional structures that should be constified but due
to their being passed into general purpose functions without const
arguments, I have not yet converted.
In general, acquire nfsd_mtx when accessing any of the global NFS
structures, including struct nfssvc_sock, struct nfsd, struct
nfsrv_descript.
Release nfsd_mtx whenever calling into VFS, and acquire Giant for
calls into VFS. Giant is not required for any part of the
operation of the NFS server with the exception of calls into VFS.
Giant will never by acquired in the upcall code path. However, it
may operate entirely covered by Giant, or not. If debug.mpsafenet
is set to 0, the system calls will acquire Giant across all
operations, and the upcall will assert Giant. As such, by default,
this enables locking and allows us to test assertions, but should not
cause any substantial new amount of code to be run without Giant.
Bugs should manifest in the form of lock assertion failures for now.
This approach is similar (but not identical) to modifications to the
BSD/OS NFS server code snapshot provided by BSDi as part of their
SMPng snapshot. The strategy is almost the same (single lock over
the NFS server), but differs in the following ways:
- Our NFS client and server code bases don't overlap, which means
both fewer bugs and easier locking (thanks Peter!). Also means
NFSD_*() as opposed to NFS_*().
- We make broad use of assertions, whereas the BSD/OS code does not.
- Made slightly different choices about how to handle macros building
packets but operating with side effects.
- We acquire Giant only when entering VFS from the NFS server daemon
threads.
- Serious bugs in BSD/OS implementation corrected -- the snapshot we
received was clearly a work in progress.
Based on ideas from: BSDi SMPng Snapshot
Reviewed by: rick@snowhite.cis.uoguelph.ca
Extensive testing by: kris
2004-05-24 04:06:14 +00:00
|
|
|
NFSD_UNLOCK();
|
2004-03-07 16:28:31 +00:00
|
|
|
callout_reset(&nfsrv_callout, nfsrv_ticks, nfsrv_timer, NULL);
|
2001-09-18 23:32:09 +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
|
|
|
|
|
|
|
|
#endif /* NFS_LEGACYRPC */
|
