d/freebsd-nq
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
b6960f00fa
freebsd-nq/sys/kern/uipc_usrreq.c

2590 lines
64 KiB
C
Raw Normal View History

/* -> /*- for copyright notices, minor format tweaks as necessary
2005-01-06 23:35:40 +00:00
/*-
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
* Copyright (c) 1982, 1986, 1989, 1991, 1993
Canonicalize the UNIX domain socket copyright layout: original holders before more recent holders. MFC after: 3 days
2005-09-23 12:41:06 +00:00
* The Regents of the University of California.
Decompose the global UNIX domain sockets rwlock into two different locks: a global list/counter/generation counter protected by a new mutex unp_list_lock, and a global linkage rwlock, unp_global_rwlock, which protects the connections between UNIX domain sockets. This eliminates conditional lock acquisition that was previously a property of the global lock being held over sonewconn() leading to a call to uipc_attach(), which also required the global lock, but couldn't rely on it as other paths existed to uipc_attach() that didn't hold it: now uipc_attach() uses only the list lock, which follows the linkage lock in the lock order. It may also reduce contention on the global lock for some workloads. Add global UNIX domain socket locks to hard-coded witness lock order. MFC after: 1 week Discussed with: kris
2009-03-08 21:48:29 +00:00
* Copyright (c) 2004-2009 Robert N. M. Watson
Canonicalize the UNIX domain socket copyright layout: original holders before more recent holders. MFC after: 3 days
2005-09-23 12:41:06 +00:00
* All rights reserved.
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
*
* 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.
Renumber license clauses in sys/kern to avoid skipping #3
2016-09-15 13:16:20 +00:00
* 3. Neither the name of the University nor the names of its contributors
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
* 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.
*
Make networking domains drop-ins, through the magic of GNU ld. (Some day, there may even be LKMs.) Also, change the internal name of `unixdomain' to `localdomain' since AF_LOCAL is now the preferred name of this family. Declare netisr correctly and in the right place.
1995-05-11 00:13:26 +00:00
* From: @(#)uipc_usrreq.c 8.3 (Berkeley) 1/4/94
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
*/
Add additional comments to the top of the UNIX domain socket implementation providing some high level pointers regarding the implementation.
2006-07-23 20:06:45 +00:00
/*
* UNIX Domain (Local) Sockets
*
* This is an implementation of UNIX (local) domain sockets. Each socket has
* an associated struct unpcb (UNIX protocol control block). Stream sockets
* may be connected to 0 or 1 other socket. Datagram sockets may be
* connected to 0, 1, or many other sockets. Sockets may be created and
* connected in pairs (socketpair(2)), or bound/connected to using the file
* system name space. For most purposes, only the receive socket buffer is
* used, as sending on one socket delivers directly to the receive socket
Break introductory comment into two paragraphs to separate material on the garbage collection complications from general discussion of UNIX domain sockets. Staticize unp_addsockcred(). Remove XXX comment regarding Giant and v_socket -- v_socket is protected by the global UNIX domain socket lock.
2007-02-20 10:50:02 +00:00
* buffer of a second socket.
*
* The implementation is substantially complicated by the fact that
* "ancillary data", such as file descriptors or credentials, may be passed
* across UNIX domain sockets. The potential for passing UNIX domain sockets
* over other UNIX domain sockets requires the implementation of a simple
* garbage collector to find and tear down cycles of disconnected sockets.
Minor rearrangement of global variables, comments, etc, in UNIX domain sockets.
2007-02-14 15:05:40 +00:00
*
* TODO:
First cut at implementing SOCK_SEQPACKET support for UNIX (local) domain sockets. This allows for reliable bi-directional datagram communication over UNIX domain sockets, in contrast to SOCK_DGRAM (M:N, unreliable) or SOCK_STERAM (bi-directional bytestream). Largely, this reuses existing UNIX domain socket code. This allows applications requiring record- oriented semantics to do so reliably via local IPC. Some implementation notes (also present in XXX comments): - Currently we lack an sbappend variant able to do datagrams and control data without doing addresses, so we mark SOCK_SEQPACKET as PR_ADDR. Adding a new variant will solve this problem. - UNIX domain sockets on FreeBSD provide back-pressure/flow control notification for stream sockets by manipulating the send socket buffer's size during pru_send and pru_rcvd. This trick works less well for SOCK_SEQPACKET as sosend_generic() uses sb_hiwat not just to manage blocking, but also to determine maximum datagram size. Fixing this requires rethinking how back-pressure is done for SOCK_SEQPACKET; in the mean time, it's possible to get EMSGSIZE when buffers fill, instead of blocking. Discussed with: benl Reviewed by: bz, rpaulo MFC after: 3 months Sponsored by: Google
2009-10-05 14:49:16 +00:00
* RDM
Minor rearrangement of global variables, comments, etc, in UNIX domain sockets.
2007-02-14 15:05:40 +00:00
* rethink name space problems
* need a proper out-of-band
Add additional comments to the top of the UNIX domain socket implementation providing some high level pointers regarding the implementation.
2006-07-23 20:06:45 +00:00
*/
Use __FBSDID().
2003-06-11 00:56:59 +00:00
#include <sys/cdefs.h>
__FBSDID("$FreeBSD$");
Add DDB "show unpcb" command, allowing DDB to print out many pertinent details from UNIX domain socket protocol layer state.
2007-05-29 12:36:00 +00:00
#include "opt_ddb.h"
Introduce support for Mandatory Access Control and extensible kernel access control. Invoke the necessary MAC entry points to maintain labels on sockets. In particular, invoke entry points during socket allocation and destruction, as well as creation by a process or during an accept-scenario (sonewconn). For UNIX domain sockets, also assign a peer label. As the socket code isn't locked down yet, locking interactions are not yet clear. Various protocol stack socket operations (such as peer label assignment for IPv4) will follow. Obtained from: TrustedBSD Project Sponsored by: DARPA, NAI Labs
2002-07-31 03:03:22 +00:00
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
#include <sys/param.h>
Update kernel inclusions of capability.h to use capsicum.h instead; some further refinement is required as some device drivers intended to be portable over FreeBSD versions rely on __FreeBSD_version to decide whether to include capability.h. MFC after: 3 weeks
2014-03-16 10:55:57 +00:00
#include <sys/capsicum.h>
Undo part of the tangle of having sys/lock.h and sys/mutex.h included in other "system" header files. Also help the deprecation of lockmgr.h by making it a sub-include of sys/lock.h and removing sys/lockmgr.h form kernel .c files. Sort sys/*.h includes where possible in affected files. OK'ed by: bde (with reservations)
2001-05-01 08:13:21 +00:00
#include <sys/domain.h>
Revert the change of #includes in sys/filedesc.h and sys/socketvar.h. Requested by: bde Since locking sigio_lock is usually followed by calling pgsigio(), move the declaration of sigio_lock and the definitions of SIGIO_*() to sys/signalvar.h. While I am here, sort include files alphabetically, where possible.
2002-04-30 01:54:54 +00:00
#include <sys/fcntl.h>
Fixed duplicate definitions of M_FILE (one static).
1997-11-23 10:43:49 +00:00
#include <sys/malloc.h> /* XXX must be before <sys/file.h> */
Allow for nmbclusters and maxsockets to be increased via sysctl. An eventhandler is used to update all the various zones that depend on these values.
2006-04-21 09:25:40 +00:00
#include <sys/eventhandler.h>
Create a new branch of the kernel MIB, kern.ipc, to store all of the configurables and instrumentation related to inter-process communication mechanisms. Some variables, like mbuf statistics, are instrumented here for the first time. For mbuf statistics: also keep track of m_copym() and m_pullup() failures, and provide for the user's inspection the compiled-in values of MSIZE, MHLEN, MCLBYTES, and MINCLSIZE.
1997-02-24 20:30:58 +00:00
#include <sys/file.h>
Revert the change of #includes in sys/filedesc.h and sys/socketvar.h. Requested by: bde Since locking sigio_lock is usually followed by calling pgsigio(), move the declaration of sigio_lock and the definitions of SIGIO_*() to sys/signalvar.h. While I am here, sort include files alphabetically, where possible.
2002-04-30 01:54:54 +00:00
#include <sys/filedesc.h>
#include <sys/kernel.h>
#include <sys/lock.h>
Create a new branch of the kernel MIB, kern.ipc, to store all of the configurables and instrumentation related to inter-process communication mechanisms. Some variables, like mbuf statistics, are instrumented here for the first time. For mbuf statistics: also keep track of m_copym() and m_pullup() failures, and provide for the user's inspection the compiled-in values of MSIZE, MHLEN, MCLBYTES, and MINCLSIZE.
1997-02-24 20:30:58 +00:00
#include <sys/mbuf.h>
- Lock access to vrele() with VFS_LOCK_GIANT() rather than mtx_lock(&Giant). Sponsored by: Isilon Systems, Inc.
2006-01-30 08:19:01 +00:00
#include <sys/mount.h>
Revert the change of #includes in sys/filedesc.h and sys/socketvar.h. Requested by: bde Since locking sigio_lock is usually followed by calling pgsigio(), move the declaration of sigio_lock and the definitions of SIGIO_*() to sys/signalvar.h. While I am here, sort include files alphabetically, where possible.
2002-04-30 01:54:54 +00:00
#include <sys/mutex.h>
Create a new branch of the kernel MIB, kern.ipc, to store all of the configurables and instrumentation related to inter-process communication mechanisms. Some variables, like mbuf statistics, are instrumented here for the first time. For mbuf statistics: also keep track of m_copym() and m_pullup() failures, and provide for the user's inspection the compiled-in values of MSIZE, MHLEN, MCLBYTES, and MINCLSIZE.
1997-02-24 20:30:58 +00:00
#include <sys/namei.h>
#include <sys/proc.h>
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
#include <sys/protosw.h>
Reviewed by: jeff, rwatson MFC after: 1 week
2010-12-03 16:15:44 +00:00
#include <sys/queue.h>
Revert the change of #includes in sys/filedesc.h and sys/socketvar.h. Requested by: bde Since locking sigio_lock is usually followed by calling pgsigio(), move the declaration of sigio_lock and the definitions of SIGIO_*() to sys/signalvar.h. While I am here, sort include files alphabetically, where possible.
2002-04-30 01:54:54 +00:00
#include <sys/resourcevar.h>
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
#include <sys/rwlock.h>
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
#include <sys/socket.h>
#include <sys/socketvar.h>
Revert the change of #includes in sys/filedesc.h and sys/socketvar.h. Requested by: bde Since locking sigio_lock is usually followed by calling pgsigio(), move the declaration of sigio_lock and the definitions of SIGIO_*() to sys/signalvar.h. While I am here, sort include files alphabetically, where possible.
2002-04-30 01:54:54 +00:00
#include <sys/signalvar.h>
Create a new branch of the kernel MIB, kern.ipc, to store all of the configurables and instrumentation related to inter-process communication mechanisms. Some variables, like mbuf statistics, are instrumented here for the first time. For mbuf statistics: also keep track of m_copym() and m_pullup() failures, and provide for the user's inspection the compiled-in values of MSIZE, MHLEN, MCLBYTES, and MINCLSIZE.
1997-02-24 20:30:58 +00:00
#include <sys/stat.h>
Revert the change of #includes in sys/filedesc.h and sys/socketvar.h. Requested by: bde Since locking sigio_lock is usually followed by calling pgsigio(), move the declaration of sigio_lock and the definitions of SIGIO_*() to sys/signalvar.h. While I am here, sort include files alphabetically, where possible.
2002-04-30 01:54:54 +00:00
#include <sys/sx.h>
Create a new branch of the kernel MIB, kern.ipc, to store all of the configurables and instrumentation related to inter-process communication mechanisms. Some variables, like mbuf statistics, are instrumented here for the first time. For mbuf statistics: also keep track of m_copym() and m_pullup() failures, and provide for the user's inspection the compiled-in values of MSIZE, MHLEN, MCLBYTES, and MINCLSIZE.
1997-02-24 20:30:58 +00:00
#include <sys/sysctl.h>
Revert the change of #includes in sys/filedesc.h and sys/socketvar.h. Requested by: bde Since locking sigio_lock is usually followed by calling pgsigio(), move the declaration of sigio_lock and the definitions of SIGIO_*() to sys/signalvar.h. While I am here, sort include files alphabetically, where possible.
2002-04-30 01:54:54 +00:00
#include <sys/systm.h>
Correct a number of serious and closely related bugs in the UNIX domain socket file descriptor garbage collection code, which is intended to detect and clear cycles of orphaned file descriptors that are "in-flight" in a socket when that socket is closed before they are received. The algorithm present was both run at poor times (resulting in recursion and reentrance), and also buggy in the presence of parallelism. In order to fix these problems, make the following changes: - When there are in-flight sockets and a UNIX domain socket is destroyed, asynchronously schedule the garbage collector, rather than running it synchronously in the current context. This avoids lock order issues when the garbage collection code reenters the UNIX domain socket code, avoiding lock order reversals, deadlocks, etc. Run the code asynchronously in a task queue. - In the garbage collector, when skipping file descriptors that have entered a closing state (i.e., have f_count == 0), re-test the FDEFER flag, and decrement unp_defer. As file descriptors can now transition to a closed state, while the garbage collector is running, it is no longer the case that unp_defer will remain an accurate count of deferred sockets in the mark portion of the GC algorithm. Otherwise, the garbage collector will loop waiting waiting for unp_defer to reach zero, which it will never do as it is skipping file descriptors that were marked in an earlier pass, but now closed. - Acquire the UNIX domain socket subsystem lock in unp_discard() when modifying the unp_rights counter, or a read/write race is risked with other threads also manipulating the counter. While here: - Remove #if 0'd code regarding acquiring the socket buffer sleep lock in the garbage collector, this is not required as we are able to use the socket buffer receive lock to protect scanning the receive buffer for in-flight file descriptors on the socket buffer. - Annotate that the description of the garbage collector implementation is increasingly inaccurate and needs to be updated. - Add counters of the number of deferred garbage collections and recycled file descriptors. This will be removed and is here temporarily for debugging purposes. With these changes in place, the unp_passfd regression test now appears to be passed consistently on UP and SMP systems for extended runs, whereas before it hung quickly or panicked, depending on which bug was triggered. Reported by: Philip Kizer <pckizer at nostrum dot com> MFC after: 2 weeks
2005-11-10 16:06:04 +00:00
#include <sys/taskqueue.h>
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
#include <sys/un.h>
Convert socket structures to be type-stable and add a version number. Define a parameter which indicates the maximum number of sockets in a system, and use this to size the zone allocators used for sockets and for certain PCBs. Convert PF_LOCAL PCB structures to be type-stable and add a version number. Define an external format for infomation about socket structures and use it in several places. Define a mechanism to get all PF_LOCAL and PF_INET PCB lists through sysctl(3) without blocking network interrupts for an unreasonable length of time. This probably still has some bugs and/or race conditions, but it seems to work well enough on my machines. It is now possible for `netstat' to get almost all of its information via the sysctl(3) interface rather than reading kmem (changes to follow).
1998-05-15 20:11:40 +00:00
#include <sys/unpcb.h>
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
#include <sys/vnode.h>
Merge the remainder of kern_vimage.c and vimage.h into vnet.c and vnet.h, we now use jails (rather than vimages) as the abstraction for virtualization management, and what remained was specific to virtual network stacks. Minor cleanups are done in the process, and comments updated to reflect these changes. Reviewed by: bz Approved by: re (vimage blanket)
2009-08-01 19:26:27 +00:00
#include <net/vnet.h>
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
Add DDB "show unpcb" command, allowing DDB to print out many pertinent details from UNIX domain socket protocol layer state.
2007-05-29 12:36:00 +00:00
#ifdef DDB
#include <ddb/ddb.h>
#endif
Complete break-out of sys/sys/mac.h into sys/security/mac/mac_framework.h begun with a repo-copy of mac.h to mac_framework.h. sys/mac.h now contains the userspace and user<->kernel API and definitions, with all in-kernel interfaces moved to mac_framework.h, which is now included across most of the kernel instead. This change is the first step in a larger cleanup and sweep of MAC Framework interfaces in the kernel, and will not be MFC'd. Obtained from: TrustedBSD Project Sponsored by: SPARTA
2006-10-22 11:52:19 +00:00
#include <security/mac/mac_framework.h>
Remove references to vm_zone.h and switch over to the new uma API.
2002-03-20 04:11:52 +00:00
#include <vm/uma.h>
Convert socket structures to be type-stable and add a version number. Define a parameter which indicates the maximum number of sockets in a system, and use this to size the zone allocators used for sockets and for certain PCBs. Convert PF_LOCAL PCB structures to be type-stable and add a version number. Define an external format for infomation about socket structures and use it in several places. Define a mechanism to get all PF_LOCAL and PF_INET PCB lists through sysctl(3) without blocking network interrupts for an unreasonable length of time. This probably still has some bugs and/or race conditions, but it seems to work well enough on my machines. It is now possible for `netstat' to get almost all of its information via the sysctl(3) interface rather than reading kmem (changes to follow).
1998-05-15 20:11:40 +00:00
For some reason when I started to pass filedescent structures instead of pointers to the file structure receiving descriptors stopped to work when also at least few kilobytes of data is being send. In the kernel the soreceive_generic() function doesn't see control mbuf as the first mbuf and unp_externalize() is never called, first 6(?) kilobytes of data is missing as well on receiving end. This breaks for example tmux. I don't know yet why going from 8 bytes to sizeof(struct filedescent) per descriptor (or even to 16 bytes per descriptor) breaks things, but to work-around it for now use 8 bytes per file descriptor at the cost of memory allocation. Reported by: flo, Diane Bruce, Jan Beich <jbeich@tormail.org> Simple testcase provided by: mjg
2013-03-03 23:39:30 +00:00
MALLOC_DECLARE(M_FILECAPS);
Decompose the global UNIX domain sockets rwlock into two different locks: a global list/counter/generation counter protected by a new mutex unp_list_lock, and a global linkage rwlock, unp_global_rwlock, which protects the connections between UNIX domain sockets. This eliminates conditional lock acquisition that was previously a property of the global lock being held over sonewconn() leading to a call to uipc_attach(), which also required the global lock, but couldn't rely on it as other paths existed to uipc_attach() that didn't hold it: now uipc_attach() uses only the list lock, which follows the linkage lock in the lock order. It may also reduce contention on the global lock for some workloads. Add global UNIX domain socket locks to hard-coded witness lock order. MFC after: 1 week Discussed with: kris
2009-03-08 21:48:29 +00:00
/*
* Locking key:
* (l) Locked using list lock
* (g) Locked using linkage lock
*/
Minor rearrangement of global variables, comments, etc, in UNIX domain sockets.
2007-02-14 15:05:40 +00:00
static uma_zone_t unp_zone;
Decompose the global UNIX domain sockets rwlock into two different locks: a global list/counter/generation counter protected by a new mutex unp_list_lock, and a global linkage rwlock, unp_global_rwlock, which protects the connections between UNIX domain sockets. This eliminates conditional lock acquisition that was previously a property of the global lock being held over sonewconn() leading to a call to uipc_attach(), which also required the global lock, but couldn't rely on it as other paths existed to uipc_attach() that didn't hold it: now uipc_attach() uses only the list lock, which follows the linkage lock in the lock order. It may also reduce contention on the global lock for some workloads. Add global UNIX domain socket locks to hard-coded witness lock order. MFC after: 1 week Discussed with: kris
2009-03-08 21:48:29 +00:00
static unp_gen_t unp_gencnt; /* (l) */
static u_int unp_count; /* (l) Count of local sockets. */
Minor rearrangement of global variables, comments, etc, in UNIX domain sockets.
2007-02-14 15:05:40 +00:00
static ino_t unp_ino; /* Prototype for fake inode numbers. */
Decompose the global UNIX domain sockets rwlock into two different locks: a global list/counter/generation counter protected by a new mutex unp_list_lock, and a global linkage rwlock, unp_global_rwlock, which protects the connections between UNIX domain sockets. This eliminates conditional lock acquisition that was previously a property of the global lock being held over sonewconn() leading to a call to uipc_attach(), which also required the global lock, but couldn't rely on it as other paths existed to uipc_attach() that didn't hold it: now uipc_attach() uses only the list lock, which follows the linkage lock in the lock order. It may also reduce contention on the global lock for some workloads. Add global UNIX domain socket locks to hard-coded witness lock order. MFC after: 1 week Discussed with: kris
2009-03-08 21:48:29 +00:00
static int unp_rights; /* (g) File descriptors in flight. */
static struct unp_head unp_shead; /* (l) List of stream sockets. */
static struct unp_head unp_dhead; /* (l) List of datagram sockets. */
First cut at implementing SOCK_SEQPACKET support for UNIX (local) domain sockets. This allows for reliable bi-directional datagram communication over UNIX domain sockets, in contrast to SOCK_DGRAM (M:N, unreliable) or SOCK_STERAM (bi-directional bytestream). Largely, this reuses existing UNIX domain socket code. This allows applications requiring record- oriented semantics to do so reliably via local IPC. Some implementation notes (also present in XXX comments): - Currently we lack an sbappend variant able to do datagrams and control data without doing addresses, so we mark SOCK_SEQPACKET as PR_ADDR. Adding a new variant will solve this problem. - UNIX domain sockets on FreeBSD provide back-pressure/flow control notification for stream sockets by manipulating the send socket buffer's size during pru_send and pru_rcvd. This trick works less well for SOCK_SEQPACKET as sosend_generic() uses sb_hiwat not just to manage blocking, but also to determine maximum datagram size. Fixing this requires rethinking how back-pressure is done for SOCK_SEQPACKET; in the mean time, it's possible to get EMSGSIZE when buffers fill, instead of blocking. Discussed with: benl Reviewed by: bz, rpaulo MFC after: 3 months Sponsored by: Google
2009-10-05 14:49:16 +00:00
static struct unp_head unp_sphead; /* (l) List of seqpacket sockets. */
Convert socket structures to be type-stable and add a version number. Define a parameter which indicates the maximum number of sockets in a system, and use this to size the zone allocators used for sockets and for certain PCBs. Convert PF_LOCAL PCB structures to be type-stable and add a version number. Define an external format for infomation about socket structures and use it in several places. Define a mechanism to get all PF_LOCAL and PF_INET PCB lists through sysctl(3) without blocking network interrupts for an unreasonable length of time. This probably still has some bugs and/or race conditions, but it seems to work well enough on my machines. It is now possible for `netstat' to get almost all of its information via the sysctl(3) interface rather than reading kmem (changes to follow).
1998-05-15 20:11:40 +00:00
Reviewed by: jeff, rwatson MFC after: 1 week
2010-12-03 16:15:44 +00:00
struct unp_defer {
SLIST_ENTRY(unp_defer) ud_link;
struct file *ud_fp;
};
static SLIST_HEAD(, unp_defer) unp_defers;
static int unp_defers_count;
Minor rearrangement of global variables, comments, etc, in UNIX domain sockets.
2007-02-14 15:05:40 +00:00
static const struct sockaddr sun_noname = { sizeof(sun_noname), AF_LOCAL };
Convert socket structures to be type-stable and add a version number. Define a parameter which indicates the maximum number of sockets in a system, and use this to size the zone allocators used for sockets and for certain PCBs. Convert PF_LOCAL PCB structures to be type-stable and add a version number. Define an external format for infomation about socket structures and use it in several places. Define a mechanism to get all PF_LOCAL and PF_INET PCB lists through sysctl(3) without blocking network interrupts for an unreasonable length of time. This probably still has some bugs and/or race conditions, but it seems to work well enough on my machines. It is now possible for `netstat' to get almost all of its information via the sysctl(3) interface rather than reading kmem (changes to follow).
1998-05-15 20:11:40 +00:00
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
/*
Minor rearrangement of global variables, comments, etc, in UNIX domain sockets.
2007-02-14 15:05:40 +00:00
* Garbage collection of cyclic file descriptor/socket references occurs
* asynchronously in a taskqueue context in order to avoid recursion and
* reentrance in the UNIX domain socket, file descriptor, and socket layer
* code. See unp_gc() for a full description.
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
*/
Schedule garbage collection run for the in-flight rights passed over the unix domain sockets to the next tick, coalescing the serial calls until the collection fires. The thought is that more work for the collector could arise in the near time, allowing to clean more and not spend too much CPU on repeated collection when there is no garbage. Currently the collection task is fired immediately upon unix domain socket close if there are any rights in flight, which caused excessive CPU usage and too long blocking of the threads waiting for unp_list_lock and unp_link_rwlock in write mode. Robert noted that it would be nice if we could find some heuristic by which we decide whether to run GC a bit more quickly. E.g., if the number of UNIX domain sockets is close to its resource limit, but not quite. Reported and tested by: Markus Gebert <markus.gebert@hostpoint.ch> Reviewed by: rwatson MFC after: 2 weeks
2012-11-20 15:45:48 +00:00
static struct timeout_task unp_gc_task;
Another mega commit to staticize things.
1995-12-14 09:55:16 +00:00
Reviewed by: jeff, rwatson MFC after: 1 week
2010-12-03 16:15:44 +00:00
/*
* The close of unix domain sockets attached as SCM_RIGHTS is
* postponed to the taskqueue, to avoid arbitrary recursion depth.
* The attached sockets might have another sockets attached.
*/
static struct task unp_defer_task;
Move various UNIX socket global variables and sysctls from the middle of the file to the top.
2006-07-23 10:19:04 +00:00
/*
* Both send and receive buffers are allocated PIPSIZ bytes of buffering for
* stream sockets, although the total for sender and receiver is actually
* only PIPSIZ.
*
* Datagram sockets really use the sendspace as the maximum datagram size,
* and don't really want to reserve the sendspace. Their recvspace should be
* large enough for at least one max-size datagram plus address.
*/
#ifndef PIPSIZ
#define PIPSIZ 8192
#endif
static u_long unpst_sendspace = PIPSIZ;
static u_long unpst_recvspace = PIPSIZ;
static u_long unpdg_sendspace = 2*1024; /* really max datagram size */
static u_long unpdg_recvspace = 4*1024;
First cut at implementing SOCK_SEQPACKET support for UNIX (local) domain sockets. This allows for reliable bi-directional datagram communication over UNIX domain sockets, in contrast to SOCK_DGRAM (M:N, unreliable) or SOCK_STERAM (bi-directional bytestream). Largely, this reuses existing UNIX domain socket code. This allows applications requiring record- oriented semantics to do so reliably via local IPC. Some implementation notes (also present in XXX comments): - Currently we lack an sbappend variant able to do datagrams and control data without doing addresses, so we mark SOCK_SEQPACKET as PR_ADDR. Adding a new variant will solve this problem. - UNIX domain sockets on FreeBSD provide back-pressure/flow control notification for stream sockets by manipulating the send socket buffer's size during pru_send and pru_rcvd. This trick works less well for SOCK_SEQPACKET as sosend_generic() uses sb_hiwat not just to manage blocking, but also to determine maximum datagram size. Fixing this requires rethinking how back-pressure is done for SOCK_SEQPACKET; in the mean time, it's possible to get EMSGSIZE when buffers fill, instead of blocking. Discussed with: benl Reviewed by: bz, rpaulo MFC after: 3 months Sponsored by: Google
2009-10-05 14:49:16 +00:00
static u_long unpsp_sendspace = PIPSIZ; /* really max datagram size */
static u_long unpsp_recvspace = PIPSIZ;
Move various UNIX socket global variables and sysctls from the middle of the file to the top.
2006-07-23 10:19:04 +00:00
Mark all SYSCTL_NODEs static that have no corresponding SYSCTL_DECLs. The SYSCTL_NODE macro defines a list that stores all child-elements of that node. If there's no SYSCTL_DECL macro anywhere else, there's no reason why it shouldn't be static.
2011-11-07 15:43:11 +00:00
static SYSCTL_NODE(_net, PF_LOCAL, local, CTLFLAG_RW, 0, "Local domain");
static SYSCTL_NODE(_net_local, SOCK_STREAM, stream, CTLFLAG_RW, 0,
"SOCK_STREAM");
static SYSCTL_NODE(_net_local, SOCK_DGRAM, dgram, CTLFLAG_RW, 0, "SOCK_DGRAM");
static SYSCTL_NODE(_net_local, SOCK_SEQPACKET, seqpacket, CTLFLAG_RW, 0,
First cut at implementing SOCK_SEQPACKET support for UNIX (local) domain sockets. This allows for reliable bi-directional datagram communication over UNIX domain sockets, in contrast to SOCK_DGRAM (M:N, unreliable) or SOCK_STERAM (bi-directional bytestream). Largely, this reuses existing UNIX domain socket code. This allows applications requiring record- oriented semantics to do so reliably via local IPC. Some implementation notes (also present in XXX comments): - Currently we lack an sbappend variant able to do datagrams and control data without doing addresses, so we mark SOCK_SEQPACKET as PR_ADDR. Adding a new variant will solve this problem. - UNIX domain sockets on FreeBSD provide back-pressure/flow control notification for stream sockets by manipulating the send socket buffer's size during pru_send and pru_rcvd. This trick works less well for SOCK_SEQPACKET as sosend_generic() uses sb_hiwat not just to manage blocking, but also to determine maximum datagram size. Fixing this requires rethinking how back-pressure is done for SOCK_SEQPACKET; in the mean time, it's possible to get EMSGSIZE when buffers fill, instead of blocking. Discussed with: benl Reviewed by: bz, rpaulo MFC after: 3 months Sponsored by: Google
2009-10-05 14:49:16 +00:00
"SOCK_SEQPACKET");
Move definition of UNIX domain socket protosw and domain entries from uipc_proto.c to uipc_usrreq.c, making localdomain static. Remove uipc_proto.c as it's no longer used. With this change, UNIX domain sockets are entirely encapsulated in uipc_usrreq.c.
2006-08-07 12:02:43 +00:00
Move various UNIX socket global variables and sysctls from the middle of the file to the top.
2006-07-23 10:19:04 +00:00
SYSCTL_ULONG(_net_local_stream, OID_AUTO, sendspace, CTLFLAG_RW,
Fill in a few sysctl descriptions. Approved by: rwatson
2008-07-26 00:55:35 +00:00
&unpst_sendspace, 0, "Default stream send space.");
Move various UNIX socket global variables and sysctls from the middle of the file to the top.
2006-07-23 10:19:04 +00:00
SYSCTL_ULONG(_net_local_stream, OID_AUTO, recvspace, CTLFLAG_RW,
Fill in a few sysctl descriptions. Approved by: rwatson
2008-07-26 00:55:35 +00:00
&unpst_recvspace, 0, "Default stream receive space.");
Move various UNIX socket global variables and sysctls from the middle of the file to the top.
2006-07-23 10:19:04 +00:00
SYSCTL_ULONG(_net_local_dgram, OID_AUTO, maxdgram, CTLFLAG_RW,
Fill in a few sysctl descriptions. Approved by: rwatson
2008-07-26 00:55:35 +00:00
&unpdg_sendspace, 0, "Default datagram send space.");
Move various UNIX socket global variables and sysctls from the middle of the file to the top.
2006-07-23 10:19:04 +00:00
SYSCTL_ULONG(_net_local_dgram, OID_AUTO, recvspace, CTLFLAG_RW,
Fill in a few sysctl descriptions. Approved by: rwatson
2008-07-26 00:55:35 +00:00
&unpdg_recvspace, 0, "Default datagram receive space.");
First cut at implementing SOCK_SEQPACKET support for UNIX (local) domain sockets. This allows for reliable bi-directional datagram communication over UNIX domain sockets, in contrast to SOCK_DGRAM (M:N, unreliable) or SOCK_STERAM (bi-directional bytestream). Largely, this reuses existing UNIX domain socket code. This allows applications requiring record- oriented semantics to do so reliably via local IPC. Some implementation notes (also present in XXX comments): - Currently we lack an sbappend variant able to do datagrams and control data without doing addresses, so we mark SOCK_SEQPACKET as PR_ADDR. Adding a new variant will solve this problem. - UNIX domain sockets on FreeBSD provide back-pressure/flow control notification for stream sockets by manipulating the send socket buffer's size during pru_send and pru_rcvd. This trick works less well for SOCK_SEQPACKET as sosend_generic() uses sb_hiwat not just to manage blocking, but also to determine maximum datagram size. Fixing this requires rethinking how back-pressure is done for SOCK_SEQPACKET; in the mean time, it's possible to get EMSGSIZE when buffers fill, instead of blocking. Discussed with: benl Reviewed by: bz, rpaulo MFC after: 3 months Sponsored by: Google
2009-10-05 14:49:16 +00:00
SYSCTL_ULONG(_net_local_seqpacket, OID_AUTO, maxseqpacket, CTLFLAG_RW,
&unpsp_sendspace, 0, "Default seqpacket send space.");
SYSCTL_ULONG(_net_local_seqpacket, OID_AUTO, recvspace, CTLFLAG_RW,
&unpsp_recvspace, 0, "Default seqpacket receive space.");
Trim whitespaces at the end of lines. Use the commit to record proper log message for r216150. MFC after: 1 week If unix socket has a unix socket attached as the rights that has a unix socket attached as the rights that has a unix socket attached as the rights ... Kernel may overflow the stack on attempt to close such socket. Only close the rights file in the context of the current close if the file is not unix domain socket. Otherwise, postpone the work to taskqueue, preventing unlimited recursion. The pass of the unix domain sockets over the SCM_RIGHTS message control is not widely used, and more, the close of the socket with still attached rights is mostly an application failure. The change should not affect the performance of typical users of SCM_RIGHTS. Reviewed by: jeff, rwatson
2010-12-03 20:39:06 +00:00
SYSCTL_INT(_net_local, OID_AUTO, inflight, CTLFLAG_RD, &unp_rights, 0,
Fill in a few sysctl descriptions. Approved by: rwatson
2008-07-26 00:55:35 +00:00
"File descriptors in flight.");
Reviewed by: jeff, rwatson MFC after: 1 week
2010-12-03 16:15:44 +00:00
SYSCTL_INT(_net_local, OID_AUTO, deferred, CTLFLAG_RD,
Trim whitespaces at the end of lines. Use the commit to record proper log message for r216150. MFC after: 1 week If unix socket has a unix socket attached as the rights that has a unix socket attached as the rights that has a unix socket attached as the rights ... Kernel may overflow the stack on attempt to close such socket. Only close the rights file in the context of the current close if the file is not unix domain socket. Otherwise, postpone the work to taskqueue, preventing unlimited recursion. The pass of the unix domain sockets over the SCM_RIGHTS message control is not widely used, and more, the close of the socket with still attached rights is mostly an application failure. The change should not affect the performance of typical users of SCM_RIGHTS. Reviewed by: jeff, rwatson
2010-12-03 20:39:06 +00:00
&unp_defers_count, 0,
Reviewed by: jeff, rwatson MFC after: 1 week
2010-12-03 16:15:44 +00:00
"File descriptors deferred to taskqueue for close.");
Move various UNIX socket global variables and sysctls from the middle of the file to the top.
2006-07-23 10:19:04 +00:00
Remove spurious '/*-' marks and fix some other style problems. Submitted by: bde@
2010-07-22 05:42:29 +00:00
/*
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
* Locking and synchronization:
*
Listening sockets improvements. o Separate fields of struct socket that belong to listening from fields that belong to normal dataflow, and unionize them. This shrinks the structure a bit. - Take out selinfo's from the socket buffers into the socket. The first reason is to support braindamaged scenario when a socket is added to kevent(2) and then listen(2) is cast on it. The second reason is that there is future plan to make socket buffers pluggable, so that for a dataflow socket a socket buffer can be changed, and in this case we also want to keep same selinfos through the lifetime of a socket. - Remove struct struct so_accf. Since now listening stuff no longer affects struct socket size, just move its fields into listening part of the union. - Provide sol_upcall field and enforce that so_upcall_set() may be called only on a dataflow socket, which has buffers, and for listening sockets provide solisten_upcall_set(). o Remove ACCEPT_LOCK() global. - Add a mutex to socket, to be used instead of socket buffer lock to lock fields of struct socket that don't belong to a socket buffer. - Allow to acquire two socket locks, but the first one must belong to a listening socket. - Make soref()/sorele() to use atomic(9). This allows in some situations to do soref() without owning socket lock. There is place for improvement here, it is possible to make sorele() also to lock optionally. - Most protocols aren't touched by this change, except UNIX local sockets. See below for more information. o Reduce copy-and-paste in kernel modules that accept connections from listening sockets: provide function solisten_dequeue(), and use it in the following modules: ctl(4), iscsi(4), ng_btsocket(4), ng_ksocket(4), infiniband, rpc. o UNIX local sockets. - Removal of ACCEPT_LOCK() global uncovered several races in the UNIX local sockets. Most races exist around spawning a new socket, when we are connecting to a local listening socket. To cover them, we need to hold locks on both PCBs when spawning a third one. This means holding them across sonewconn(). This creates a LOR between pcb locks and unp_list_lock. - To fix the new LOR, abandon the global unp_list_lock in favor of global unp_link_lock. Indeed, separating these two locks didn't provide us any extra parralelism in the UNIX sockets. - Now call into uipc_attach() may happen with unp_link_lock hold if, we are accepting, or without unp_link_lock in case if we are just creating a socket. - Another problem in UNIX sockets is that uipc_close() basicly did nothing for a listening socket. The vnode remained opened for connections. This is fixed by removing vnode in uipc_close(). Maybe the right way would be to do it for all sockets (not only listening), simply move the vnode teardown from uipc_detach() to uipc_close()? Sponsored by: Netflix Differential Revision: https://reviews.freebsd.org/D9770
2017-06-08 21:30:34 +00:00
* Two types of locks exist in the local domain socket implementation: a
* a global linkage rwlock and per-unpcb mutexes. The linkage lock protects
* the socket count, global generation number, stream/datagram global lists and
Decompose the global UNIX domain sockets rwlock into two different locks: a global list/counter/generation counter protected by a new mutex unp_list_lock, and a global linkage rwlock, unp_global_rwlock, which protects the connections between UNIX domain sockets. This eliminates conditional lock acquisition that was previously a property of the global lock being held over sonewconn() leading to a call to uipc_attach(), which also required the global lock, but couldn't rely on it as other paths existed to uipc_attach() that didn't hold it: now uipc_attach() uses only the list lock, which follows the linkage lock in the lock order. It may also reduce contention on the global lock for some workloads. Add global UNIX domain socket locks to hard-coded witness lock order. MFC after: 1 week Discussed with: kris
2009-03-08 21:48:29 +00:00
* interconnection of unpcbs, the v_socket and unp_vnode pointers, and can be
* held exclusively over the acquisition of multiple unpcb locks to prevent
* deadlock.
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
*
* UNIX domain sockets each have an unpcb hung off of their so_pcb pointer,
* allocated in pru_attach() and freed in pru_detach(). The validity of that
* pointer is an invariant, so no lock is required to dereference the so_pcb
* pointer if a valid socket reference is held by the caller. In practice,
* this is always true during operations performed on a socket. Each unpcb
* has a back-pointer to its socket, unp_socket, which will be stable under
* the same circumstances.
*
* This pointer may only be safely dereferenced as long as a valid reference
* to the unpcb is held. Typically, this reference will be from the socket,
* or from another unpcb when the referring unpcb's lock is held (in order
* that the reference not be invalidated during use). For example, to follow
* unp->unp_conn->unp_socket, you need unlock the lock on unp, not unp_conn,
* as unp_socket remains valid as long as the reference to unp_conn is valid.
*
* Fields of unpcbss are locked using a per-unpcb lock, unp_mtx. Individual
* atomic reads without the lock may be performed "lockless", but more
* complex reads and read-modify-writes require the mutex to be held. No
* lock order is defined between unpcb locks -- multiple unpcb locks may be
Decompose the global UNIX domain sockets rwlock into two different locks: a global list/counter/generation counter protected by a new mutex unp_list_lock, and a global linkage rwlock, unp_global_rwlock, which protects the connections between UNIX domain sockets. This eliminates conditional lock acquisition that was previously a property of the global lock being held over sonewconn() leading to a call to uipc_attach(), which also required the global lock, but couldn't rely on it as other paths existed to uipc_attach() that didn't hold it: now uipc_attach() uses only the list lock, which follows the linkage lock in the lock order. It may also reduce contention on the global lock for some workloads. Add global UNIX domain socket locks to hard-coded witness lock order. MFC after: 1 week Discussed with: kris
2009-03-08 21:48:29 +00:00
* acquired at the same time only when holding the linkage rwlock
* exclusively, which prevents deadlocks.
Annotate the current UNIX domain socket locking strategies, order, strengths, and weaknesses in a comment. Assert a copyright over the changes made as part of the locking work.
2004-08-16 01:52:04 +00:00
*
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
* Blocking with UNIX domain sockets is a tricky issue: unlike most network
* protocols, bind() is a non-atomic operation, and connect() requires
* potential sleeping in the protocol, due to potentially waiting on local or
* distributed file systems. We try to separate "lookup" operations, which
* may sleep, and the IPC operations themselves, which typically can occur
* with relative atomicity as locks can be held over the entire operation.
Annotate the current UNIX domain socket locking strategies, order, strengths, and weaknesses in a comment. Assert a copyright over the changes made as part of the locking work.
2004-08-16 01:52:04 +00:00
*
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
* Another tricky issue is simultaneous multi-threaded or multi-process
* access to a single UNIX domain socket. These are handled by the flags
* UNP_CONNECTING and UNP_BINDING, which prevent concurrent connecting or
* binding, both of which involve dropping UNIX domain socket locks in order
* to perform namei() and other file system operations.
Annotate the current UNIX domain socket locking strategies, order, strengths, and weaknesses in a comment. Assert a copyright over the changes made as part of the locking work.
2004-08-16 01:52:04 +00:00
*/
Decompose the global UNIX domain sockets rwlock into two different locks: a global list/counter/generation counter protected by a new mutex unp_list_lock, and a global linkage rwlock, unp_global_rwlock, which protects the connections between UNIX domain sockets. This eliminates conditional lock acquisition that was previously a property of the global lock being held over sonewconn() leading to a call to uipc_attach(), which also required the global lock, but couldn't rely on it as other paths existed to uipc_attach() that didn't hold it: now uipc_attach() uses only the list lock, which follows the linkage lock in the lock order. It may also reduce contention on the global lock for some workloads. Add global UNIX domain socket locks to hard-coded witness lock order. MFC after: 1 week Discussed with: kris
2009-03-08 21:48:29 +00:00
static struct rwlock unp_link_rwlock;
Reviewed by: jeff, rwatson MFC after: 1 week
2010-12-03 16:15:44 +00:00
static struct mtx unp_defers_lock;
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
Decompose the global UNIX domain sockets rwlock into two different locks: a global list/counter/generation counter protected by a new mutex unp_list_lock, and a global linkage rwlock, unp_global_rwlock, which protects the connections between UNIX domain sockets. This eliminates conditional lock acquisition that was previously a property of the global lock being held over sonewconn() leading to a call to uipc_attach(), which also required the global lock, but couldn't rely on it as other paths existed to uipc_attach() that didn't hold it: now uipc_attach() uses only the list lock, which follows the linkage lock in the lock order. It may also reduce contention on the global lock for some workloads. Add global UNIX domain socket locks to hard-coded witness lock order. MFC after: 1 week Discussed with: kris
2009-03-08 21:48:29 +00:00
#define UNP_LINK_LOCK_INIT() rw_init(&unp_link_rwlock, \
"unp_link_rwlock")
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
Decompose the global UNIX domain sockets rwlock into two different locks: a global list/counter/generation counter protected by a new mutex unp_list_lock, and a global linkage rwlock, unp_global_rwlock, which protects the connections between UNIX domain sockets. This eliminates conditional lock acquisition that was previously a property of the global lock being held over sonewconn() leading to a call to uipc_attach(), which also required the global lock, but couldn't rely on it as other paths existed to uipc_attach() that didn't hold it: now uipc_attach() uses only the list lock, which follows the linkage lock in the lock order. It may also reduce contention on the global lock for some workloads. Add global UNIX domain socket locks to hard-coded witness lock order. MFC after: 1 week Discussed with: kris
2009-03-08 21:48:29 +00:00
#define UNP_LINK_LOCK_ASSERT() rw_assert(&unp_link_rwlock, \
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
RA_LOCKED)
Decompose the global UNIX domain sockets rwlock into two different locks: a global list/counter/generation counter protected by a new mutex unp_list_lock, and a global linkage rwlock, unp_global_rwlock, which protects the connections between UNIX domain sockets. This eliminates conditional lock acquisition that was previously a property of the global lock being held over sonewconn() leading to a call to uipc_attach(), which also required the global lock, but couldn't rely on it as other paths existed to uipc_attach() that didn't hold it: now uipc_attach() uses only the list lock, which follows the linkage lock in the lock order. It may also reduce contention on the global lock for some workloads. Add global UNIX domain socket locks to hard-coded witness lock order. MFC after: 1 week Discussed with: kris
2009-03-08 21:48:29 +00:00
#define UNP_LINK_UNLOCK_ASSERT() rw_assert(&unp_link_rwlock, \
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
RA_UNLOCKED)
Decompose the global UNIX domain sockets rwlock into two different locks: a global list/counter/generation counter protected by a new mutex unp_list_lock, and a global linkage rwlock, unp_global_rwlock, which protects the connections between UNIX domain sockets. This eliminates conditional lock acquisition that was previously a property of the global lock being held over sonewconn() leading to a call to uipc_attach(), which also required the global lock, but couldn't rely on it as other paths existed to uipc_attach() that didn't hold it: now uipc_attach() uses only the list lock, which follows the linkage lock in the lock order. It may also reduce contention on the global lock for some workloads. Add global UNIX domain socket locks to hard-coded witness lock order. MFC after: 1 week Discussed with: kris
2009-03-08 21:48:29 +00:00
#define UNP_LINK_RLOCK() rw_rlock(&unp_link_rwlock)
#define UNP_LINK_RUNLOCK() rw_runlock(&unp_link_rwlock)
#define UNP_LINK_WLOCK() rw_wlock(&unp_link_rwlock)
#define UNP_LINK_WUNLOCK() rw_wunlock(&unp_link_rwlock)
#define UNP_LINK_WLOCK_ASSERT() rw_assert(&unp_link_rwlock, \
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
RA_WLOCKED)
Listening sockets improvements. o Separate fields of struct socket that belong to listening from fields that belong to normal dataflow, and unionize them. This shrinks the structure a bit. - Take out selinfo's from the socket buffers into the socket. The first reason is to support braindamaged scenario when a socket is added to kevent(2) and then listen(2) is cast on it. The second reason is that there is future plan to make socket buffers pluggable, so that for a dataflow socket a socket buffer can be changed, and in this case we also want to keep same selinfos through the lifetime of a socket. - Remove struct struct so_accf. Since now listening stuff no longer affects struct socket size, just move its fields into listening part of the union. - Provide sol_upcall field and enforce that so_upcall_set() may be called only on a dataflow socket, which has buffers, and for listening sockets provide solisten_upcall_set(). o Remove ACCEPT_LOCK() global. - Add a mutex to socket, to be used instead of socket buffer lock to lock fields of struct socket that don't belong to a socket buffer. - Allow to acquire two socket locks, but the first one must belong to a listening socket. - Make soref()/sorele() to use atomic(9). This allows in some situations to do soref() without owning socket lock. There is place for improvement here, it is possible to make sorele() also to lock optionally. - Most protocols aren't touched by this change, except UNIX local sockets. See below for more information. o Reduce copy-and-paste in kernel modules that accept connections from listening sockets: provide function solisten_dequeue(), and use it in the following modules: ctl(4), iscsi(4), ng_btsocket(4), ng_ksocket(4), infiniband, rpc. o UNIX local sockets. - Removal of ACCEPT_LOCK() global uncovered several races in the UNIX local sockets. Most races exist around spawning a new socket, when we are connecting to a local listening socket. To cover them, we need to hold locks on both PCBs when spawning a third one. This means holding them across sonewconn(). This creates a LOR between pcb locks and unp_list_lock. - To fix the new LOR, abandon the global unp_list_lock in favor of global unp_link_lock. Indeed, separating these two locks didn't provide us any extra parralelism in the UNIX sockets. - Now call into uipc_attach() may happen with unp_link_lock hold if, we are accepting, or without unp_link_lock in case if we are just creating a socket. - Another problem in UNIX sockets is that uipc_close() basicly did nothing for a listening socket. The vnode remained opened for connections. This is fixed by removing vnode in uipc_close(). Maybe the right way would be to do it for all sockets (not only listening), simply move the vnode teardown from uipc_detach() to uipc_close()? Sponsored by: Netflix Differential Revision: https://reviews.freebsd.org/D9770
2017-06-08 21:30:34 +00:00
#define UNP_LINK_WOWNED() rw_wowned(&unp_link_rwlock)
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
Reviewed by: jeff, rwatson MFC after: 1 week
2010-12-03 16:15:44 +00:00
#define UNP_DEFERRED_LOCK_INIT() mtx_init(&unp_defers_lock, \
"unp_defer", NULL, MTX_DEF)
#define UNP_DEFERRED_LOCK() mtx_lock(&unp_defers_lock)
#define UNP_DEFERRED_UNLOCK() mtx_unlock(&unp_defers_lock)
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
#define UNP_PCB_LOCK_INIT(unp) mtx_init(&(unp)->unp_mtx, \
"unp_mtx", "unp_mtx", \
MTX_DUPOK|MTX_DEF|MTX_RECURSE)
#define UNP_PCB_LOCK_DESTROY(unp) mtx_destroy(&(unp)->unp_mtx)
#define UNP_PCB_LOCK(unp) mtx_lock(&(unp)->unp_mtx)
#define UNP_PCB_UNLOCK(unp) mtx_unlock(&(unp)->unp_mtx)
#define UNP_PCB_LOCK_ASSERT(unp) mtx_assert(&(unp)->unp_mtx, MA_OWNED)
Now that portalfs doesn't directly invoke uipc_connect2(), make it a static symbol. MFC after: 3 days
2008-10-06 18:43:11 +00:00
static int uipc_connect2(struct socket *, struct socket *);
Further minor cleanups to UNIX domain sockets: - Staticize and locally prototype functions uipc_ctloutput(), unp_dispose(), unp_init(), and unp_externalize(), none of which have been required outside of uipc_usrreq.c since uipc_proto.c was removed. - Remove stale prototype for uipc_usrreq(), which has not existed in the code since 1997 - Forward declare and staticize uipc_usrreqs structure in uipc_usrreq.c and not un.h. - Comment on why uipc_connect2() is still non-static -- it is used directly by fifofs. - Remove stale comments, tidy up whitespace. MFC after: 3 days (where applicable)
2008-10-03 13:01:56 +00:00
static int uipc_ctloutput(struct socket *, struct sockopt *);
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
static int unp_connect(struct socket *, struct sockaddr *,
Minor rearrangement of global variables, comments, etc, in UNIX domain sockets.
2007-02-14 15:05:40 +00:00
struct thread *);
- Implement two new system calls: int bindat(int fd, int s, const struct sockaddr *addr, socklen_t addrlen); int connectat(int fd, int s, const struct sockaddr *name, socklen_t namelen); which allow to bind and connect respectively to a UNIX domain socket with a path relative to the directory associated with the given file descriptor 'fd'. - Add manual pages for the new syscalls. - Make the new syscalls available for processes in capability mode sandbox. - Add capability rights CAP_BINDAT and CAP_CONNECTAT that has to be present on the directory descriptor for the syscalls to work. - Update audit(4) to support those two new syscalls and to handle path in sockaddr_un structure relative to the given directory descriptor. - Update procstat(1) to recognize the new capability rights. - Document the new capability rights in cap_rights_limit(2). Sponsored by: The FreeBSD Foundation Discussed with: rwatson, jilles, kib, des
2013-03-02 21:11:30 +00:00
static int unp_connectat(int, struct socket *, struct sockaddr *,
struct thread *);
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
static int unp_connect2(struct socket *so, struct socket *so2, int);
static void unp_disconnect(struct unpcb *unp, struct unpcb *unp2);
Rename unp_dispose_so() to unp_dispose(). It implements the dom_dispose method for local socket domain, so its name should match the method name.
2016-08-31 21:48:22 +00:00
static void unp_dispose(struct socket *so);
static void unp_dispose_mbuf(struct mbuf *);
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
static void unp_shutdown(struct unpcb *);
Remove the errno argument from unp_drop(). While there, add a comment to clarify that ECONNRESET should always be returned for POSIX conformance. Suggested by: Steven Hartland
2016-02-26 12:46:34 +00:00
static void unp_drop(struct unpcb *);
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
static void unp_gc(__unused void *, int);
Fix memory leak when one process send descriptor over UNIX domain socket, but the other process exited before receiving it.
2013-03-11 22:59:07 +00:00
static void unp_scan(struct mbuf *, void (*)(struct filedescent **, int));
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
static void unp_discard(struct file *);
For some reason when I started to pass filedescent structures instead of pointers to the file structure receiving descriptors stopped to work when also at least few kilobytes of data is being send. In the kernel the soreceive_generic() function doesn't see control mbuf as the first mbuf and unp_externalize() is never called, first 6(?) kilobytes of data is missing as well on receiving end. This breaks for example tmux. I don't know yet why going from 8 bytes to sizeof(struct filedescent) per descriptor (or even to 16 bytes per descriptor) breaks things, but to work-around it for now use 8 bytes per file descriptor at the cost of memory allocation. Reported by: flo, Diane Bruce, Jan Beich <jbeich@tormail.org> Simple testcase provided by: mjg
2013-03-03 23:39:30 +00:00
static void unp_freerights(struct filedescent **, int);
Further minor cleanups to UNIX domain sockets: - Staticize and locally prototype functions uipc_ctloutput(), unp_dispose(), unp_init(), and unp_externalize(), none of which have been required outside of uipc_usrreq.c since uipc_proto.c was removed. - Remove stale prototype for uipc_usrreq(), which has not existed in the code since 1997 - Forward declare and staticize uipc_usrreqs structure in uipc_usrreq.c and not un.h. - Comment on why uipc_connect2() is still non-static -- it is used directly by fifofs. - Remove stale comments, tidy up whitespace. MFC after: 3 days (where applicable)
2008-10-03 13:01:56 +00:00
static void unp_init(void);
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
static int unp_internalize(struct mbuf **, struct thread *);
Remove explicit locking of struct file. - Introduce a finit() which is used to initailize the fields of struct file in such a way that the ops vector is only valid after the data, type, and flags are valid. - Protect f_flag and f_count with atomic operations. - Remove the global list of all files and associated accounting. - Rewrite the unp garbage collection such that it no longer requires the global list of all files and instead uses a list of all unp sockets. - Mark sockets in the accept queue so we don't incorrectly gc them. Tested by: kris, pho
2007-12-30 01:42:15 +00:00
static void unp_internalize_fp(struct file *);
Implement SOCK_CLOEXEC, SOCK_NONBLOCK and MSG_CMSG_CLOEXEC. This change allows creating file descriptors with close-on-exec set in some situations. SOCK_CLOEXEC and SOCK_NONBLOCK can be OR'ed in socket() and socketpair()'s type parameter, and MSG_CMSG_CLOEXEC to recvmsg() makes file descriptors (SCM_RIGHTS) atomically close-on-exec. The numerical values for SOCK_CLOEXEC and SOCK_NONBLOCK are as in NetBSD. MSG_CMSG_CLOEXEC is the first free bit for MSG_*. The SOCK_* flags are not passed to MAC because this may cause incorrect failures and can be done later via fcntl() anyway. On the other hand, audit is expected to cope with the new flags. For MSG_CMSG_CLOEXEC, unp_externalize() is extended to take a flags argument. Reviewed by: kib
2013-03-19 20:58:17 +00:00
static int unp_externalize(struct mbuf *, struct mbuf **, int);
Reviewed by: jeff, rwatson MFC after: 1 week
2010-12-03 16:15:44 +00:00
static int unp_externalize_fp(struct file *);
Break introductory comment into two paragraphs to separate material on the garbage collection complications from general discussion of UNIX domain sockets. Staticize unp_addsockcred(). Remove XXX comment regarding Giant and v_socket -- v_socket is protected by the global UNIX domain socket lock.
2007-02-20 10:50:02 +00:00
static struct mbuf *unp_addsockcred(struct thread *, struct mbuf *);
Reviewed by: jeff, rwatson MFC after: 1 week
2010-12-03 16:15:44 +00:00
static void unp_process_defers(void * __unused, int);
Another mega commit to staticize things.
1995-12-14 09:55:16 +00:00
Move definition of UNIX domain socket protosw and domain entries from uipc_proto.c to uipc_usrreq.c, making localdomain static. Remove uipc_proto.c as it's no longer used. With this change, UNIX domain sockets are entirely encapsulated in uipc_usrreq.c.
2006-08-07 12:02:43 +00:00
/*
* Definitions of protocols supported in the LOCAL domain.
*/
static struct domain localdomain;
Differentiate pr_usrreqs for stream and datagram UNIX domain sockets, and employ soreceive_dgram for the datagram case. MFC after: 3 months
2008-10-08 06:19:49 +00:00
static struct pr_usrreqs uipc_usrreqs_dgram, uipc_usrreqs_stream;
First cut at implementing SOCK_SEQPACKET support for UNIX (local) domain sockets. This allows for reliable bi-directional datagram communication over UNIX domain sockets, in contrast to SOCK_DGRAM (M:N, unreliable) or SOCK_STERAM (bi-directional bytestream). Largely, this reuses existing UNIX domain socket code. This allows applications requiring record- oriented semantics to do so reliably via local IPC. Some implementation notes (also present in XXX comments): - Currently we lack an sbappend variant able to do datagrams and control data without doing addresses, so we mark SOCK_SEQPACKET as PR_ADDR. Adding a new variant will solve this problem. - UNIX domain sockets on FreeBSD provide back-pressure/flow control notification for stream sockets by manipulating the send socket buffer's size during pru_send and pru_rcvd. This trick works less well for SOCK_SEQPACKET as sosend_generic() uses sb_hiwat not just to manage blocking, but also to determine maximum datagram size. Fixing this requires rethinking how back-pressure is done for SOCK_SEQPACKET; in the mean time, it's possible to get EMSGSIZE when buffers fill, instead of blocking. Discussed with: benl Reviewed by: bz, rpaulo MFC after: 3 months Sponsored by: Google
2009-10-05 14:49:16 +00:00
static struct pr_usrreqs uipc_usrreqs_seqpacket;
Move definition of UNIX domain socket protosw and domain entries from uipc_proto.c to uipc_usrreq.c, making localdomain static. Remove uipc_proto.c as it's no longer used. With this change, UNIX domain sockets are entirely encapsulated in uipc_usrreq.c.
2006-08-07 12:02:43 +00:00
static struct protosw localsw[] = {
{
.pr_type = SOCK_STREAM,
.pr_domain = &localdomain,
.pr_flags = PR_CONNREQUIRED|PR_WANTRCVD|PR_RIGHTS,
.pr_ctloutput = &uipc_ctloutput,
Differentiate pr_usrreqs for stream and datagram UNIX domain sockets, and employ soreceive_dgram for the datagram case. MFC after: 3 months
2008-10-08 06:19:49 +00:00
.pr_usrreqs = &uipc_usrreqs_stream
Move definition of UNIX domain socket protosw and domain entries from uipc_proto.c to uipc_usrreq.c, making localdomain static. Remove uipc_proto.c as it's no longer used. With this change, UNIX domain sockets are entirely encapsulated in uipc_usrreq.c.
2006-08-07 12:02:43 +00:00
},
{
.pr_type = SOCK_DGRAM,
.pr_domain = &localdomain,
.pr_flags = PR_ATOMIC|PR_ADDR|PR_RIGHTS,
Supply the pr_ctloutput method for local datagram sockets, so that setsockopt() and getsockopt() work on them. This makes 'tools/regression/sockets/unix_cmsg -t dgram' more successful.
2012-09-07 21:06:54 +00:00
.pr_ctloutput = &uipc_ctloutput,
Differentiate pr_usrreqs for stream and datagram UNIX domain sockets, and employ soreceive_dgram for the datagram case. MFC after: 3 months
2008-10-08 06:19:49 +00:00
.pr_usrreqs = &uipc_usrreqs_dgram
Move definition of UNIX domain socket protosw and domain entries from uipc_proto.c to uipc_usrreq.c, making localdomain static. Remove uipc_proto.c as it's no longer used. With this change, UNIX domain sockets are entirely encapsulated in uipc_usrreq.c.
2006-08-07 12:02:43 +00:00
},
First cut at implementing SOCK_SEQPACKET support for UNIX (local) domain sockets. This allows for reliable bi-directional datagram communication over UNIX domain sockets, in contrast to SOCK_DGRAM (M:N, unreliable) or SOCK_STERAM (bi-directional bytestream). Largely, this reuses existing UNIX domain socket code. This allows applications requiring record- oriented semantics to do so reliably via local IPC. Some implementation notes (also present in XXX comments): - Currently we lack an sbappend variant able to do datagrams and control data without doing addresses, so we mark SOCK_SEQPACKET as PR_ADDR. Adding a new variant will solve this problem. - UNIX domain sockets on FreeBSD provide back-pressure/flow control notification for stream sockets by manipulating the send socket buffer's size during pru_send and pru_rcvd. This trick works less well for SOCK_SEQPACKET as sosend_generic() uses sb_hiwat not just to manage blocking, but also to determine maximum datagram size. Fixing this requires rethinking how back-pressure is done for SOCK_SEQPACKET; in the mean time, it's possible to get EMSGSIZE when buffers fill, instead of blocking. Discussed with: benl Reviewed by: bz, rpaulo MFC after: 3 months Sponsored by: Google
2009-10-05 14:49:16 +00:00
{
.pr_type = SOCK_SEQPACKET,
.pr_domain = &localdomain,
/*
* XXXRW: For now, PR_ADDR because soreceive will bump into them
* due to our use of sbappendaddr. A new sbappend variants is needed
* that supports both atomic record writes and control data.
*/
.pr_flags = PR_ADDR|PR_ATOMIC|PR_CONNREQUIRED|PR_WANTRCVD|
PR_RIGHTS,
Provide pr_ctloutput method for AF_LOCAL/SOCK_SEQPACKET sockets. This makes setsockopt() on them working. Reported by: Yuri <yuri rawbw.com> Approved by: re (kib)
2013-09-11 18:22:30 +00:00
.pr_ctloutput = &uipc_ctloutput,
First cut at implementing SOCK_SEQPACKET support for UNIX (local) domain sockets. This allows for reliable bi-directional datagram communication over UNIX domain sockets, in contrast to SOCK_DGRAM (M:N, unreliable) or SOCK_STERAM (bi-directional bytestream). Largely, this reuses existing UNIX domain socket code. This allows applications requiring record- oriented semantics to do so reliably via local IPC. Some implementation notes (also present in XXX comments): - Currently we lack an sbappend variant able to do datagrams and control data without doing addresses, so we mark SOCK_SEQPACKET as PR_ADDR. Adding a new variant will solve this problem. - UNIX domain sockets on FreeBSD provide back-pressure/flow control notification for stream sockets by manipulating the send socket buffer's size during pru_send and pru_rcvd. This trick works less well for SOCK_SEQPACKET as sosend_generic() uses sb_hiwat not just to manage blocking, but also to determine maximum datagram size. Fixing this requires rethinking how back-pressure is done for SOCK_SEQPACKET; in the mean time, it's possible to get EMSGSIZE when buffers fill, instead of blocking. Discussed with: benl Reviewed by: bz, rpaulo MFC after: 3 months Sponsored by: Google
2009-10-05 14:49:16 +00:00
.pr_usrreqs = &uipc_usrreqs_seqpacket,
},
Move definition of UNIX domain socket protosw and domain entries from uipc_proto.c to uipc_usrreq.c, making localdomain static. Remove uipc_proto.c as it's no longer used. With this change, UNIX domain sockets are entirely encapsulated in uipc_usrreq.c.
2006-08-07 12:02:43 +00:00
};
static struct domain localdomain = {
.dom_family = AF_LOCAL,
.dom_name = "local",
.dom_init = unp_init,
.dom_externalize = unp_externalize,
Rename unp_dispose_so() to unp_dispose(). It implements the dom_dispose method for local socket domain, so its name should match the method name.
2016-08-31 21:48:22 +00:00
.dom_dispose = unp_dispose,
Move definition of UNIX domain socket protosw and domain entries from uipc_proto.c to uipc_usrreq.c, making localdomain static. Remove uipc_proto.c as it's no longer used. With this change, UNIX domain sockets are entirely encapsulated in uipc_usrreq.c.
2006-08-07 12:02:43 +00:00
.dom_protosw = localsw,
kernel: use our nitems() macro when it is available through param.h. No functional change, only trivial cases are done in this sweep, Discussed in: freebsd-current
2016-04-19 23:48:27 +00:00
.dom_protoswNPROTOSW = &localsw[nitems(localsw)]
Move definition of UNIX domain socket protosw and domain entries from uipc_proto.c to uipc_usrreq.c, making localdomain static. Remove uipc_proto.c as it's no longer used. With this change, UNIX domain sockets are entirely encapsulated in uipc_usrreq.c.
2006-08-07 12:02:43 +00:00
};
DOMAIN_SET(local);
Change protocol switch pru_abort() API so that it returns void rather than an int, as an error here is not meaningful. Modify soabort() to unconditionally free the socket on the return of pru_abort(), and modify most protocols to no longer conditionally free the socket, since the caller will do this. This commit likely leaves parts of netinet and netinet6 in a situation where they may panic or leak memory, as they have not are not fully updated by this commit. This will be corrected shortly in followup commits to these components. MFC after: 3 months
2006-04-01 15:15:05 +00:00
static void
The long-awaited mega-massive-network-code- cleanup. Part I. This commit includes the following changes: 1) Old-style (pr_usrreq()) protocols are no longer supported, the compatibility glue for them is deleted, and the kernel will panic on boot if any are compiled in. 2) Certain protocol entry points are modified to take a process structure, so they they can easily tell whether or not it is possible to sleep, and also to access credentials. 3) SS_PRIV is no more, and with it goes the SO_PRIVSTATE setsockopt() call. Protocols should use the process pointer they are now passed. 4) The PF_LOCAL and PF_ROUTE families have been updated to use the new style, as has the `raw' skeleton family. 5) PF_LOCAL sockets now obey the process's umask when creating a socket in the filesystem. As a result, LINT is now broken. I'm hoping that some enterprising hacker with a bit more time will either make the broken bits work (should be easy for netipx) or dike them out.
1997-04-27 20:01:29 +00:00
uipc_abort(struct socket *so)
{
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
struct unpcb *unp, *unp2;
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
Always acquire the UNIX domain socket subsystem lock (UNP lock) before dereferencing sotounpcb() and checking its value, as so_pcb is protected by protocol locking, not subsystem locking. This prevents races during close() by one thread and use of ths socket in another. unp_bind() now assert the UNP lock, and uipc_bind() now acquires the lock around calls to unp_bind().
2004-08-16 04:41:03 +00:00
unp = sotounpcb(so);
Modify UNIX domain sockets to guarantee, and assume, that so_pcb is always defined for an in-use socket. This allows us to eliminate countless tests of whether so_pcb is non-NULL, eliminating dozens of error cases. For now, retain the call to sotryfree() in the uipc_abort() path, but this will eventually move to soabort(). These new assumptions should be largely correct, and will become more so as the socket/pcb reference model is fixed. Removing the notion that so_pcb can be non-NULL is a critical step towards further fine-graining of the UNIX domain socket locking, as the so_pcb reference no longer needs to be protected using locks, instead it is a property of the socket life cycle.
2006-03-17 13:52:57 +00:00
KASSERT(unp != NULL, ("uipc_abort: unp == NULL"));
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
Decompose the global UNIX domain sockets rwlock into two different locks: a global list/counter/generation counter protected by a new mutex unp_list_lock, and a global linkage rwlock, unp_global_rwlock, which protects the connections between UNIX domain sockets. This eliminates conditional lock acquisition that was previously a property of the global lock being held over sonewconn() leading to a call to uipc_attach(), which also required the global lock, but couldn't rely on it as other paths existed to uipc_attach() that didn't hold it: now uipc_attach() uses only the list lock, which follows the linkage lock in the lock order. It may also reduce contention on the global lock for some workloads. Add global UNIX domain socket locks to hard-coded witness lock order. MFC after: 1 week Discussed with: kris
2009-03-08 21:48:29 +00:00
UNP_LINK_WLOCK();
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
UNP_PCB_LOCK(unp);
unp2 = unp->unp_conn;
if (unp2 != NULL) {
UNP_PCB_LOCK(unp2);
Remove the errno argument from unp_drop(). While there, add a comment to clarify that ECONNRESET should always be returned for POSIX conformance. Suggested by: Steven Hartland
2016-02-26 12:46:34 +00:00
unp_drop(unp2);
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
UNP_PCB_UNLOCK(unp2);
}
UNP_PCB_UNLOCK(unp);
Decompose the global UNIX domain sockets rwlock into two different locks: a global list/counter/generation counter protected by a new mutex unp_list_lock, and a global linkage rwlock, unp_global_rwlock, which protects the connections between UNIX domain sockets. This eliminates conditional lock acquisition that was previously a property of the global lock being held over sonewconn() leading to a call to uipc_attach(), which also required the global lock, but couldn't rely on it as other paths existed to uipc_attach() that didn't hold it: now uipc_attach() uses only the list lock, which follows the linkage lock in the lock order. It may also reduce contention on the global lock for some workloads. Add global UNIX domain socket locks to hard-coded witness lock order. MFC after: 1 week Discussed with: kris
2009-03-08 21:48:29 +00:00
UNP_LINK_WUNLOCK();
The long-awaited mega-massive-network-code- cleanup. Part I. This commit includes the following changes: 1) Old-style (pr_usrreq()) protocols are no longer supported, the compatibility glue for them is deleted, and the kernel will panic on boot if any are compiled in. 2) Certain protocol entry points are modified to take a process structure, so they they can easily tell whether or not it is possible to sleep, and also to access credentials. 3) SS_PRIV is no more, and with it goes the SO_PRIVSTATE setsockopt() call. Protocols should use the process pointer they are now passed. 4) The PF_LOCAL and PF_ROUTE families have been updated to use the new style, as has the `raw' skeleton family. 5) PF_LOCAL sockets now obey the process's umask when creating a socket in the filesystem. As a result, LINT is now broken. I'm hoping that some enterprising hacker with a bit more time will either make the broken bits work (should be easy for netipx) or dike them out.
1997-04-27 20:01:29 +00:00
}
static int
Fix all areas of the system (or at least all those in LINT) to avoid storing socket addresses in mbufs. (Socket buffers are the one exception.) A number of kernel APIs needed to get fixed in order to make this happen. Also, fix three protocol families which kept PCBs in mbufs to not malloc them instead. Delete some old compatibility cruft while we're at it, and add some new routines in the in_cksum family.
1997-08-16 19:16:27 +00:00
uipc_accept(struct socket *so, struct sockaddr **nam)
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
{
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
struct unpcb *unp, *unp2;
Introduce a subsystem lock around UNIX domain sockets in order to protect global and allocated variables. This strategy is derived from work originally developed by BSDi for BSD/OS, and applied to FreeBSD by Sam Leffler: - Add unp_mtx, a global mutex which will protect all UNIX domain socket related variables, structures, etc. - Add UNP_LOCK(), UNP_UNLOCK(), UNP_LOCK_ASSERT() macros. - Acquire unp_mtx on entering most UNIX domain socket code, drop/re-acquire around calls into VFS, and release it on return. - Avoid performing sodupsockaddr() while holding the mutex, so in general move to allocating storage before acquiring the mutex to copy the data. - Make a stack copy of the xucred rather than copying out while holding unp_mtx. Copy the peer credential out after releasing the mutex. - Add additional assertions of vnode locks following VOP_CREATE(). A few notes: - Use of an sx lock for the file list mutex may cause problems with regard to unp_mtx when garbage collection passed file descriptors. - The locking in unp_pcblist() for sysctl monitoring is correct subject to the unpcb zone not returning memory for reuse by other subsystems (consistent with similar existing concerns). - Sam's version of this change, as with the BSD/OS version, made use of both a global lock and per-unpcb locks. However, in practice, the global lock covered all accesses, so I have simplified out the unpcb locks in the interest of getting this merged faster (reducing the overhead but not sacrificing granularity in most cases). We will want to explore possibilities for improving lock granularity in this code in the future. Submitted by: sam Sponsored by: FreeBSD Foundatiuon Obtained from: BSD/OS 5 snapshot provided by BSDi
2004-06-10 21:34:38 +00:00
const struct sockaddr *sa;
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
The long-awaited mega-massive-network-code- cleanup. Part I. This commit includes the following changes: 1) Old-style (pr_usrreq()) protocols are no longer supported, the compatibility glue for them is deleted, and the kernel will panic on boot if any are compiled in. 2) Certain protocol entry points are modified to take a process structure, so they they can easily tell whether or not it is possible to sleep, and also to access credentials. 3) SS_PRIV is no more, and with it goes the SO_PRIVSTATE setsockopt() call. Protocols should use the process pointer they are now passed. 4) The PF_LOCAL and PF_ROUTE families have been updated to use the new style, as has the `raw' skeleton family. 5) PF_LOCAL sockets now obey the process's umask when creating a socket in the filesystem. As a result, LINT is now broken. I'm hoping that some enterprising hacker with a bit more time will either make the broken bits work (should be easy for netipx) or dike them out.
1997-04-27 20:01:29 +00:00
/*
Re-wrap and other minor formatting and punctuation fixes for UNIX domain socket comments.
2006-07-22 17:24:55 +00:00
* Pass back name of connected socket, if it was bound and we are
* still connected (our peer may have closed already!).
The long-awaited mega-massive-network-code- cleanup. Part I. This commit includes the following changes: 1) Old-style (pr_usrreq()) protocols are no longer supported, the compatibility glue for them is deleted, and the kernel will panic on boot if any are compiled in. 2) Certain protocol entry points are modified to take a process structure, so they they can easily tell whether or not it is possible to sleep, and also to access credentials. 3) SS_PRIV is no more, and with it goes the SO_PRIVSTATE setsockopt() call. Protocols should use the process pointer they are now passed. 4) The PF_LOCAL and PF_ROUTE families have been updated to use the new style, as has the `raw' skeleton family. 5) PF_LOCAL sockets now obey the process's umask when creating a socket in the filesystem. As a result, LINT is now broken. I'm hoping that some enterprising hacker with a bit more time will either make the broken bits work (should be easy for netipx) or dike them out.
1997-04-27 20:01:29 +00:00
*/
Modify UNIX domain sockets to guarantee, and assume, that so_pcb is always defined for an in-use socket. This allows us to eliminate countless tests of whether so_pcb is non-NULL, eliminating dozens of error cases. For now, retain the call to sotryfree() in the uipc_abort() path, but this will eventually move to soabort(). These new assumptions should be largely correct, and will become more so as the socket/pcb reference model is fixed. Removing the notion that so_pcb can be non-NULL is a critical step towards further fine-graining of the UNIX domain socket locking, as the so_pcb reference no longer needs to be protected using locks, instead it is a property of the socket life cycle.
2006-03-17 13:52:57 +00:00
unp = sotounpcb(so);
KASSERT(unp != NULL, ("uipc_accept: unp == NULL"));
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
Introduce a subsystem lock around UNIX domain sockets in order to protect global and allocated variables. This strategy is derived from work originally developed by BSDi for BSD/OS, and applied to FreeBSD by Sam Leffler: - Add unp_mtx, a global mutex which will protect all UNIX domain socket related variables, structures, etc. - Add UNP_LOCK(), UNP_UNLOCK(), UNP_LOCK_ASSERT() macros. - Acquire unp_mtx on entering most UNIX domain socket code, drop/re-acquire around calls into VFS, and release it on return. - Avoid performing sodupsockaddr() while holding the mutex, so in general move to allocating storage before acquiring the mutex to copy the data. - Make a stack copy of the xucred rather than copying out while holding unp_mtx. Copy the peer credential out after releasing the mutex. - Add additional assertions of vnode locks following VOP_CREATE(). A few notes: - Use of an sx lock for the file list mutex may cause problems with regard to unp_mtx when garbage collection passed file descriptors. - The locking in unp_pcblist() for sysctl monitoring is correct subject to the unpcb zone not returning memory for reuse by other subsystems (consistent with similar existing concerns). - Sam's version of this change, as with the BSD/OS version, made use of both a global lock and per-unpcb locks. However, in practice, the global lock covered all accesses, so I have simplified out the unpcb locks in the interest of getting this merged faster (reducing the overhead but not sacrificing granularity in most cases). We will want to explore possibilities for improving lock granularity in this code in the future. Submitted by: sam Sponsored by: FreeBSD Foundatiuon Obtained from: BSD/OS 5 snapshot provided by BSDi
2004-06-10 21:34:38 +00:00
*nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
Decompose the global UNIX domain sockets rwlock into two different locks: a global list/counter/generation counter protected by a new mutex unp_list_lock, and a global linkage rwlock, unp_global_rwlock, which protects the connections between UNIX domain sockets. This eliminates conditional lock acquisition that was previously a property of the global lock being held over sonewconn() leading to a call to uipc_attach(), which also required the global lock, but couldn't rely on it as other paths existed to uipc_attach() that didn't hold it: now uipc_attach() uses only the list lock, which follows the linkage lock in the lock order. It may also reduce contention on the global lock for some workloads. Add global UNIX domain socket locks to hard-coded witness lock order. MFC after: 1 week Discussed with: kris
2009-03-08 21:48:29 +00:00
UNP_LINK_RLOCK();
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
unp2 = unp->unp_conn;
if (unp2 != NULL && unp2->unp_addr != NULL) {
UNP_PCB_LOCK(unp2);
sa = (struct sockaddr *) unp2->unp_addr;
bcopy(sa, *nam, sa->sa_len);
UNP_PCB_UNLOCK(unp2);
} else {
Introduce a subsystem lock around UNIX domain sockets in order to protect global and allocated variables. This strategy is derived from work originally developed by BSDi for BSD/OS, and applied to FreeBSD by Sam Leffler: - Add unp_mtx, a global mutex which will protect all UNIX domain socket related variables, structures, etc. - Add UNP_LOCK(), UNP_UNLOCK(), UNP_LOCK_ASSERT() macros. - Acquire unp_mtx on entering most UNIX domain socket code, drop/re-acquire around calls into VFS, and release it on return. - Avoid performing sodupsockaddr() while holding the mutex, so in general move to allocating storage before acquiring the mutex to copy the data. - Make a stack copy of the xucred rather than copying out while holding unp_mtx. Copy the peer credential out after releasing the mutex. - Add additional assertions of vnode locks following VOP_CREATE(). A few notes: - Use of an sx lock for the file list mutex may cause problems with regard to unp_mtx when garbage collection passed file descriptors. - The locking in unp_pcblist() for sysctl monitoring is correct subject to the unpcb zone not returning memory for reuse by other subsystems (consistent with similar existing concerns). - Sam's version of this change, as with the BSD/OS version, made use of both a global lock and per-unpcb locks. However, in practice, the global lock covered all accesses, so I have simplified out the unpcb locks in the interest of getting this merged faster (reducing the overhead but not sacrificing granularity in most cases). We will want to explore possibilities for improving lock granularity in this code in the future. Submitted by: sam Sponsored by: FreeBSD Foundatiuon Obtained from: BSD/OS 5 snapshot provided by BSDi
2004-06-10 21:34:38 +00:00
sa = &sun_noname;
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
bcopy(sa, *nam, sa->sa_len);
}
Decompose the global UNIX domain sockets rwlock into two different locks: a global list/counter/generation counter protected by a new mutex unp_list_lock, and a global linkage rwlock, unp_global_rwlock, which protects the connections between UNIX domain sockets. This eliminates conditional lock acquisition that was previously a property of the global lock being held over sonewconn() leading to a call to uipc_attach(), which also required the global lock, but couldn't rely on it as other paths existed to uipc_attach() that didn't hold it: now uipc_attach() uses only the list lock, which follows the linkage lock in the lock order. It may also reduce contention on the global lock for some workloads. Add global UNIX domain socket locks to hard-coded witness lock order. MFC after: 1 week Discussed with: kris
2009-03-08 21:48:29 +00:00
UNP_LINK_RUNLOCK();
Mechanical whitespace cleanup; parenthesize return values; other minor style nits.
2004-01-11 19:48:19 +00:00
return (0);
The long-awaited mega-massive-network-code- cleanup. Part I. This commit includes the following changes: 1) Old-style (pr_usrreq()) protocols are no longer supported, the compatibility glue for them is deleted, and the kernel will panic on boot if any are compiled in. 2) Certain protocol entry points are modified to take a process structure, so they they can easily tell whether or not it is possible to sleep, and also to access credentials. 3) SS_PRIV is no more, and with it goes the SO_PRIVSTATE setsockopt() call. Protocols should use the process pointer they are now passed. 4) The PF_LOCAL and PF_ROUTE families have been updated to use the new style, as has the `raw' skeleton family. 5) PF_LOCAL sockets now obey the process's umask when creating a socket in the filesystem. As a result, LINT is now broken. I'm hoping that some enterprising hacker with a bit more time will either make the broken bits work (should be easy for netipx) or dike them out.
1997-04-27 20:01:29 +00:00
}
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
The long-awaited mega-massive-network-code- cleanup. Part I. This commit includes the following changes: 1) Old-style (pr_usrreq()) protocols are no longer supported, the compatibility glue for them is deleted, and the kernel will panic on boot if any are compiled in. 2) Certain protocol entry points are modified to take a process structure, so they they can easily tell whether or not it is possible to sleep, and also to access credentials. 3) SS_PRIV is no more, and with it goes the SO_PRIVSTATE setsockopt() call. Protocols should use the process pointer they are now passed. 4) The PF_LOCAL and PF_ROUTE families have been updated to use the new style, as has the `raw' skeleton family. 5) PF_LOCAL sockets now obey the process's umask when creating a socket in the filesystem. As a result, LINT is now broken. I'm hoping that some enterprising hacker with a bit more time will either make the broken bits work (should be easy for netipx) or dike them out.
1997-04-27 20:01:29 +00:00
static int
KSE Milestone 2 Note ALL MODULES MUST BE RECOMPILED make the kernel aware that there are smaller units of scheduling than the process. (but only allow one thread per process at this time). This is functionally equivalent to teh previousl -current except that there is a thread associated with each process. Sorry john! (your next MFC will be a doosie!) Reviewed by: peter@freebsd.org, dillon@freebsd.org X-MFC after: ha ha ha ha
2001-09-12 08:38:13 +00:00
uipc_attach(struct socket *so, int proto, struct thread *td)
The long-awaited mega-massive-network-code- cleanup. Part I. This commit includes the following changes: 1) Old-style (pr_usrreq()) protocols are no longer supported, the compatibility glue for them is deleted, and the kernel will panic on boot if any are compiled in. 2) Certain protocol entry points are modified to take a process structure, so they they can easily tell whether or not it is possible to sleep, and also to access credentials. 3) SS_PRIV is no more, and with it goes the SO_PRIVSTATE setsockopt() call. Protocols should use the process pointer they are now passed. 4) The PF_LOCAL and PF_ROUTE families have been updated to use the new style, as has the `raw' skeleton family. 5) PF_LOCAL sockets now obey the process's umask when creating a socket in the filesystem. As a result, LINT is now broken. I'm hoping that some enterprising hacker with a bit more time will either make the broken bits work (should be easy for netipx) or dike them out.
1997-04-27 20:01:29 +00:00
{
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
u_long sendspace, recvspace;
Since unp_attach() and unp_detach() are now called only from uipc_attach() and uipc_detach(), merge them into their calling functions.
2006-07-23 10:25:28 +00:00
struct unpcb *unp;
Decompose the global UNIX domain sockets rwlock into two different locks: a global list/counter/generation counter protected by a new mutex unp_list_lock, and a global linkage rwlock, unp_global_rwlock, which protects the connections between UNIX domain sockets. This eliminates conditional lock acquisition that was previously a property of the global lock being held over sonewconn() leading to a call to uipc_attach(), which also required the global lock, but couldn't rely on it as other paths existed to uipc_attach() that didn't hold it: now uipc_attach() uses only the list lock, which follows the linkage lock in the lock order. It may also reduce contention on the global lock for some workloads. Add global UNIX domain socket locks to hard-coded witness lock order. MFC after: 1 week Discussed with: kris
2009-03-08 21:48:29 +00:00
int error;
Listening sockets improvements. o Separate fields of struct socket that belong to listening from fields that belong to normal dataflow, and unionize them. This shrinks the structure a bit. - Take out selinfo's from the socket buffers into the socket. The first reason is to support braindamaged scenario when a socket is added to kevent(2) and then listen(2) is cast on it. The second reason is that there is future plan to make socket buffers pluggable, so that for a dataflow socket a socket buffer can be changed, and in this case we also want to keep same selinfos through the lifetime of a socket. - Remove struct struct so_accf. Since now listening stuff no longer affects struct socket size, just move its fields into listening part of the union. - Provide sol_upcall field and enforce that so_upcall_set() may be called only on a dataflow socket, which has buffers, and for listening sockets provide solisten_upcall_set(). o Remove ACCEPT_LOCK() global. - Add a mutex to socket, to be used instead of socket buffer lock to lock fields of struct socket that don't belong to a socket buffer. - Allow to acquire two socket locks, but the first one must belong to a listening socket. - Make soref()/sorele() to use atomic(9). This allows in some situations to do soref() without owning socket lock. There is place for improvement here, it is possible to make sorele() also to lock optionally. - Most protocols aren't touched by this change, except UNIX local sockets. See below for more information. o Reduce copy-and-paste in kernel modules that accept connections from listening sockets: provide function solisten_dequeue(), and use it in the following modules: ctl(4), iscsi(4), ng_btsocket(4), ng_ksocket(4), infiniband, rpc. o UNIX local sockets. - Removal of ACCEPT_LOCK() global uncovered several races in the UNIX local sockets. Most races exist around spawning a new socket, when we are connecting to a local listening socket. To cover them, we need to hold locks on both PCBs when spawning a third one. This means holding them across sonewconn(). This creates a LOR between pcb locks and unp_list_lock. - To fix the new LOR, abandon the global unp_list_lock in favor of global unp_link_lock. Indeed, separating these two locks didn't provide us any extra parralelism in the UNIX sockets. - Now call into uipc_attach() may happen with unp_link_lock hold if, we are accepting, or without unp_link_lock in case if we are just creating a socket. - Another problem in UNIX sockets is that uipc_close() basicly did nothing for a listening socket. The vnode remained opened for connections. This is fixed by removing vnode in uipc_close(). Maybe the right way would be to do it for all sockets (not only listening), simply move the vnode teardown from uipc_detach() to uipc_close()? Sponsored by: Netflix Differential Revision: https://reviews.freebsd.org/D9770
2017-06-08 21:30:34 +00:00
bool locked;
Since unp_attach() and unp_detach() are now called only from uipc_attach() and uipc_detach(), merge them into their calling functions.
2006-07-23 10:25:28 +00:00
KASSERT(so->so_pcb == NULL, ("uipc_attach: so_pcb != NULL"));
if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
switch (so->so_type) {
case SOCK_STREAM:
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
sendspace = unpst_sendspace;
recvspace = unpst_recvspace;
Since unp_attach() and unp_detach() are now called only from uipc_attach() and uipc_detach(), merge them into their calling functions.
2006-07-23 10:25:28 +00:00
break;
case SOCK_DGRAM:
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
sendspace = unpdg_sendspace;
recvspace = unpdg_recvspace;
Since unp_attach() and unp_detach() are now called only from uipc_attach() and uipc_detach(), merge them into their calling functions.
2006-07-23 10:25:28 +00:00
break;
First cut at implementing SOCK_SEQPACKET support for UNIX (local) domain sockets. This allows for reliable bi-directional datagram communication over UNIX domain sockets, in contrast to SOCK_DGRAM (M:N, unreliable) or SOCK_STERAM (bi-directional bytestream). Largely, this reuses existing UNIX domain socket code. This allows applications requiring record- oriented semantics to do so reliably via local IPC. Some implementation notes (also present in XXX comments): - Currently we lack an sbappend variant able to do datagrams and control data without doing addresses, so we mark SOCK_SEQPACKET as PR_ADDR. Adding a new variant will solve this problem. - UNIX domain sockets on FreeBSD provide back-pressure/flow control notification for stream sockets by manipulating the send socket buffer's size during pru_send and pru_rcvd. This trick works less well for SOCK_SEQPACKET as sosend_generic() uses sb_hiwat not just to manage blocking, but also to determine maximum datagram size. Fixing this requires rethinking how back-pressure is done for SOCK_SEQPACKET; in the mean time, it's possible to get EMSGSIZE when buffers fill, instead of blocking. Discussed with: benl Reviewed by: bz, rpaulo MFC after: 3 months Sponsored by: Google
2009-10-05 14:49:16 +00:00
case SOCK_SEQPACKET:
sendspace = unpsp_sendspace;
recvspace = unpsp_recvspace;
break;
Since unp_attach() and unp_detach() are now called only from uipc_attach() and uipc_detach(), merge them into their calling functions.
2006-07-23 10:25:28 +00:00
default:
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
panic("uipc_attach");
Since unp_attach() and unp_detach() are now called only from uipc_attach() and uipc_detach(), merge them into their calling functions.
2006-07-23 10:25:28 +00:00
}
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
error = soreserve(so, sendspace, recvspace);
Since unp_attach() and unp_detach() are now called only from uipc_attach() and uipc_detach(), merge them into their calling functions.
2006-07-23 10:25:28 +00:00
if (error)
return (error);
}
Change unp_mtx to supporting recursion, and do not drop the unp_mtx over sonewconn() in unp_connect(). This avoids a race that occurs due to v_socket being an uncounted reference, as the lock was being released in order to call sonewconn(), which otherwise recurses into the UNIX domain socket code via pru_attach, as well as holding the lock over a sleeping memory allocation in uipc_attach(). Switch to a non-sleeping memory allocation during UNIX domain socket attach. This fix non-ideal in that it requires enabling recursion, but is a much smaller change than moving to using true references for v_socket. The reported panic occurs in unp_connect() following the return of sonewconn(). Update copyright year. Panic reported by: jhb
2007-02-14 12:22:11 +00:00
unp = uma_zalloc(unp_zone, M_NOWAIT | M_ZERO);
Since unp_attach() and unp_detach() are now called only from uipc_attach() and uipc_detach(), merge them into their calling functions.
2006-07-23 10:25:28 +00:00
if (unp == NULL)
return (ENOBUFS);
LIST_INIT(&unp->unp_refs);
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
UNP_PCB_LOCK_INIT(unp);
Since unp_attach() and unp_detach() are now called only from uipc_attach() and uipc_detach(), merge them into their calling functions.
2006-07-23 10:25:28 +00:00
unp->unp_socket = so;
so->so_pcb = unp;
- Close a race between enumerating UNIX domain socket pcb structures via sysctl and socket teardown by adding a reference count to the UNIX domain pcb object and fixing the sysctl that enumerates unpcbs to grab a reference on each unpcb while it builds the list to copy out to userland. - Close a race between UNIX domain pcb garbage collection (unp_gc()) and file descriptor teardown (fdrop()) by adding a new garbage collection flag FWAIT. unp_gc() sets FWAIT while it walks the message buffers in a UNIX domain socket looking for nested file descriptor references and clears the flag when it is finished. fdrop() checks to see if the flag is set on a file descriptor whose refcount just dropped to 0 and waits for unp_gc() to clear the flag before completely destroying the file descriptor. MFC after: 1 week Reviewed by: rwatson Submitted by: ups Hopefully makes the panics go away: mx1
2007-01-05 19:59:46 +00:00
unp->unp_refcount = 1;
Listening sockets improvements. o Separate fields of struct socket that belong to listening from fields that belong to normal dataflow, and unionize them. This shrinks the structure a bit. - Take out selinfo's from the socket buffers into the socket. The first reason is to support braindamaged scenario when a socket is added to kevent(2) and then listen(2) is cast on it. The second reason is that there is future plan to make socket buffers pluggable, so that for a dataflow socket a socket buffer can be changed, and in this case we also want to keep same selinfos through the lifetime of a socket. - Remove struct struct so_accf. Since now listening stuff no longer affects struct socket size, just move its fields into listening part of the union. - Provide sol_upcall field and enforce that so_upcall_set() may be called only on a dataflow socket, which has buffers, and for listening sockets provide solisten_upcall_set(). o Remove ACCEPT_LOCK() global. - Add a mutex to socket, to be used instead of socket buffer lock to lock fields of struct socket that don't belong to a socket buffer. - Allow to acquire two socket locks, but the first one must belong to a listening socket. - Make soref()/sorele() to use atomic(9). This allows in some situations to do soref() without owning socket lock. There is place for improvement here, it is possible to make sorele() also to lock optionally. - Most protocols aren't touched by this change, except UNIX local sockets. See below for more information. o Reduce copy-and-paste in kernel modules that accept connections from listening sockets: provide function solisten_dequeue(), and use it in the following modules: ctl(4), iscsi(4), ng_btsocket(4), ng_ksocket(4), infiniband, rpc. o UNIX local sockets. - Removal of ACCEPT_LOCK() global uncovered several races in the UNIX local sockets. Most races exist around spawning a new socket, when we are connecting to a local listening socket. To cover them, we need to hold locks on both PCBs when spawning a third one. This means holding them across sonewconn(). This creates a LOR between pcb locks and unp_list_lock. - To fix the new LOR, abandon the global unp_list_lock in favor of global unp_link_lock. Indeed, separating these two locks didn't provide us any extra parralelism in the UNIX sockets. - Now call into uipc_attach() may happen with unp_link_lock hold if, we are accepting, or without unp_link_lock in case if we are just creating a socket. - Another problem in UNIX sockets is that uipc_close() basicly did nothing for a listening socket. The vnode remained opened for connections. This is fixed by removing vnode in uipc_close(). Maybe the right way would be to do it for all sockets (not only listening), simply move the vnode teardown from uipc_detach() to uipc_close()? Sponsored by: Netflix Differential Revision: https://reviews.freebsd.org/D9770
2017-06-08 21:30:34 +00:00
if (so->so_listen != NULL)
Handle races with listening socket close when connecting a unix socket. If the listening socket is closed while sonewconn() is executing, the nascent child socket is aborted, which results in recursion on the unp_link lock when the child's pru_detach method is invoked. Fix this by using a flag to mark such sockets, and skip a part of the socket's teardown during detach. Reported by: Raviprakash Darbha <rdarbha@juniper.net> Tested by: pho MFC after: 2 weeks Differential Revision: https://reviews.freebsd.org/D7398
2016-08-08 20:25:04 +00:00
unp->unp_flags |= UNP_NASCENT;
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
Listening sockets improvements. o Separate fields of struct socket that belong to listening from fields that belong to normal dataflow, and unionize them. This shrinks the structure a bit. - Take out selinfo's from the socket buffers into the socket. The first reason is to support braindamaged scenario when a socket is added to kevent(2) and then listen(2) is cast on it. The second reason is that there is future plan to make socket buffers pluggable, so that for a dataflow socket a socket buffer can be changed, and in this case we also want to keep same selinfos through the lifetime of a socket. - Remove struct struct so_accf. Since now listening stuff no longer affects struct socket size, just move its fields into listening part of the union. - Provide sol_upcall field and enforce that so_upcall_set() may be called only on a dataflow socket, which has buffers, and for listening sockets provide solisten_upcall_set(). o Remove ACCEPT_LOCK() global. - Add a mutex to socket, to be used instead of socket buffer lock to lock fields of struct socket that don't belong to a socket buffer. - Allow to acquire two socket locks, but the first one must belong to a listening socket. - Make soref()/sorele() to use atomic(9). This allows in some situations to do soref() without owning socket lock. There is place for improvement here, it is possible to make sorele() also to lock optionally. - Most protocols aren't touched by this change, except UNIX local sockets. See below for more information. o Reduce copy-and-paste in kernel modules that accept connections from listening sockets: provide function solisten_dequeue(), and use it in the following modules: ctl(4), iscsi(4), ng_btsocket(4), ng_ksocket(4), infiniband, rpc. o UNIX local sockets. - Removal of ACCEPT_LOCK() global uncovered several races in the UNIX local sockets. Most races exist around spawning a new socket, when we are connecting to a local listening socket. To cover them, we need to hold locks on both PCBs when spawning a third one. This means holding them across sonewconn(). This creates a LOR between pcb locks and unp_list_lock. - To fix the new LOR, abandon the global unp_list_lock in favor of global unp_link_lock. Indeed, separating these two locks didn't provide us any extra parralelism in the UNIX sockets. - Now call into uipc_attach() may happen with unp_link_lock hold if, we are accepting, or without unp_link_lock in case if we are just creating a socket. - Another problem in UNIX sockets is that uipc_close() basicly did nothing for a listening socket. The vnode remained opened for connections. This is fixed by removing vnode in uipc_close(). Maybe the right way would be to do it for all sockets (not only listening), simply move the vnode teardown from uipc_detach() to uipc_close()? Sponsored by: Netflix Differential Revision: https://reviews.freebsd.org/D9770
2017-06-08 21:30:34 +00:00
if ((locked = UNP_LINK_WOWNED()) == false)
UNP_LINK_WLOCK();
Since unp_attach() and unp_detach() are now called only from uipc_attach() and uipc_detach(), merge them into their calling functions.
2006-07-23 10:25:28 +00:00
unp->unp_gencnt = ++unp_gencnt;
unp_count++;
First cut at implementing SOCK_SEQPACKET support for UNIX (local) domain sockets. This allows for reliable bi-directional datagram communication over UNIX domain sockets, in contrast to SOCK_DGRAM (M:N, unreliable) or SOCK_STERAM (bi-directional bytestream). Largely, this reuses existing UNIX domain socket code. This allows applications requiring record- oriented semantics to do so reliably via local IPC. Some implementation notes (also present in XXX comments): - Currently we lack an sbappend variant able to do datagrams and control data without doing addresses, so we mark SOCK_SEQPACKET as PR_ADDR. Adding a new variant will solve this problem. - UNIX domain sockets on FreeBSD provide back-pressure/flow control notification for stream sockets by manipulating the send socket buffer's size during pru_send and pru_rcvd. This trick works less well for SOCK_SEQPACKET as sosend_generic() uses sb_hiwat not just to manage blocking, but also to determine maximum datagram size. Fixing this requires rethinking how back-pressure is done for SOCK_SEQPACKET; in the mean time, it's possible to get EMSGSIZE when buffers fill, instead of blocking. Discussed with: benl Reviewed by: bz, rpaulo MFC after: 3 months Sponsored by: Google
2009-10-05 14:49:16 +00:00
switch (so->so_type) {
case SOCK_STREAM:
LIST_INSERT_HEAD(&unp_shead, unp, unp_link);
break;
case SOCK_DGRAM:
LIST_INSERT_HEAD(&unp_dhead, unp, unp_link);
break;
case SOCK_SEQPACKET:
LIST_INSERT_HEAD(&unp_sphead, unp, unp_link);
break;
default:
panic("uipc_attach");
}
Listening sockets improvements. o Separate fields of struct socket that belong to listening from fields that belong to normal dataflow, and unionize them. This shrinks the structure a bit. - Take out selinfo's from the socket buffers into the socket. The first reason is to support braindamaged scenario when a socket is added to kevent(2) and then listen(2) is cast on it. The second reason is that there is future plan to make socket buffers pluggable, so that for a dataflow socket a socket buffer can be changed, and in this case we also want to keep same selinfos through the lifetime of a socket. - Remove struct struct so_accf. Since now listening stuff no longer affects struct socket size, just move its fields into listening part of the union. - Provide sol_upcall field and enforce that so_upcall_set() may be called only on a dataflow socket, which has buffers, and for listening sockets provide solisten_upcall_set(). o Remove ACCEPT_LOCK() global. - Add a mutex to socket, to be used instead of socket buffer lock to lock fields of struct socket that don't belong to a socket buffer. - Allow to acquire two socket locks, but the first one must belong to a listening socket. - Make soref()/sorele() to use atomic(9). This allows in some situations to do soref() without owning socket lock. There is place for improvement here, it is possible to make sorele() also to lock optionally. - Most protocols aren't touched by this change, except UNIX local sockets. See below for more information. o Reduce copy-and-paste in kernel modules that accept connections from listening sockets: provide function solisten_dequeue(), and use it in the following modules: ctl(4), iscsi(4), ng_btsocket(4), ng_ksocket(4), infiniband, rpc. o UNIX local sockets. - Removal of ACCEPT_LOCK() global uncovered several races in the UNIX local sockets. Most races exist around spawning a new socket, when we are connecting to a local listening socket. To cover them, we need to hold locks on both PCBs when spawning a third one. This means holding them across sonewconn(). This creates a LOR between pcb locks and unp_list_lock. - To fix the new LOR, abandon the global unp_list_lock in favor of global unp_link_lock. Indeed, separating these two locks didn't provide us any extra parralelism in the UNIX sockets. - Now call into uipc_attach() may happen with unp_link_lock hold if, we are accepting, or without unp_link_lock in case if we are just creating a socket. - Another problem in UNIX sockets is that uipc_close() basicly did nothing for a listening socket. The vnode remained opened for connections. This is fixed by removing vnode in uipc_close(). Maybe the right way would be to do it for all sockets (not only listening), simply move the vnode teardown from uipc_detach() to uipc_close()? Sponsored by: Netflix Differential Revision: https://reviews.freebsd.org/D9770
2017-06-08 21:30:34 +00:00
if (locked == false)
UNP_LINK_WUNLOCK();
Since unp_attach() and unp_detach() are now called only from uipc_attach() and uipc_detach(), merge them into their calling functions.
2006-07-23 10:25:28 +00:00
return (0);
The long-awaited mega-massive-network-code- cleanup. Part I. This commit includes the following changes: 1) Old-style (pr_usrreq()) protocols are no longer supported, the compatibility glue for them is deleted, and the kernel will panic on boot if any are compiled in. 2) Certain protocol entry points are modified to take a process structure, so they they can easily tell whether or not it is possible to sleep, and also to access credentials. 3) SS_PRIV is no more, and with it goes the SO_PRIVSTATE setsockopt() call. Protocols should use the process pointer they are now passed. 4) The PF_LOCAL and PF_ROUTE families have been updated to use the new style, as has the `raw' skeleton family. 5) PF_LOCAL sockets now obey the process's umask when creating a socket in the filesystem. As a result, LINT is now broken. I'm hoping that some enterprising hacker with a bit more time will either make the broken bits work (should be easy for netipx) or dike them out.
1997-04-27 20:01:29 +00:00
}
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
The long-awaited mega-massive-network-code- cleanup. Part I. This commit includes the following changes: 1) Old-style (pr_usrreq()) protocols are no longer supported, the compatibility glue for them is deleted, and the kernel will panic on boot if any are compiled in. 2) Certain protocol entry points are modified to take a process structure, so they they can easily tell whether or not it is possible to sleep, and also to access credentials. 3) SS_PRIV is no more, and with it goes the SO_PRIVSTATE setsockopt() call. Protocols should use the process pointer they are now passed. 4) The PF_LOCAL and PF_ROUTE families have been updated to use the new style, as has the `raw' skeleton family. 5) PF_LOCAL sockets now obey the process's umask when creating a socket in the filesystem. As a result, LINT is now broken. I'm hoping that some enterprising hacker with a bit more time will either make the broken bits work (should be easy for netipx) or dike them out.
1997-04-27 20:01:29 +00:00
static int
- Implement two new system calls: int bindat(int fd, int s, const struct sockaddr *addr, socklen_t addrlen); int connectat(int fd, int s, const struct sockaddr *name, socklen_t namelen); which allow to bind and connect respectively to a UNIX domain socket with a path relative to the directory associated with the given file descriptor 'fd'. - Add manual pages for the new syscalls. - Make the new syscalls available for processes in capability mode sandbox. - Add capability rights CAP_BINDAT and CAP_CONNECTAT that has to be present on the directory descriptor for the syscalls to work. - Update audit(4) to support those two new syscalls and to handle path in sockaddr_un structure relative to the given directory descriptor. - Update procstat(1) to recognize the new capability rights. - Document the new capability rights in cap_rights_limit(2). Sponsored by: The FreeBSD Foundation Discussed with: rwatson, jilles, kib, des
2013-03-02 21:11:30 +00:00
uipc_bindat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td)
The long-awaited mega-massive-network-code- cleanup. Part I. This commit includes the following changes: 1) Old-style (pr_usrreq()) protocols are no longer supported, the compatibility glue for them is deleted, and the kernel will panic on boot if any are compiled in. 2) Certain protocol entry points are modified to take a process structure, so they they can easily tell whether or not it is possible to sleep, and also to access credentials. 3) SS_PRIV is no more, and with it goes the SO_PRIVSTATE setsockopt() call. Protocols should use the process pointer they are now passed. 4) The PF_LOCAL and PF_ROUTE families have been updated to use the new style, as has the `raw' skeleton family. 5) PF_LOCAL sockets now obey the process's umask when creating a socket in the filesystem. As a result, LINT is now broken. I'm hoping that some enterprising hacker with a bit more time will either make the broken bits work (should be easy for netipx) or dike them out.
1997-04-27 20:01:29 +00:00
{
Merge unp_bind() into uipc_bind(), as it is called only from uipc_bind().
2006-07-23 11:02:12 +00:00
struct sockaddr_un *soun = (struct sockaddr_un *)nam;
struct vattr vattr;
Remove the support for using non-mpsafe filesystem modules. In particular, do not lock Giant conditionally when calling into the filesystem module, remove the VFS_LOCK_GIANT() and related macros. Stop handling buffers belonging to non-mpsafe filesystems. The VFS_VERSION is bumped to indicate the interface change which does not result in the interface signatures changes. Conducted and reviewed by: attilio Tested by: pho
2012-10-22 17:50:54 +00:00
int error, namelen;
Merge unp_bind() into uipc_bind(), as it is called only from uipc_bind().
2006-07-23 11:02:12 +00:00
struct nameidata nd;
Always acquire the UNIX domain socket subsystem lock (UNP lock) before dereferencing sotounpcb() and checking its value, as so_pcb is protected by protocol locking, not subsystem locking. This prevents races during close() by one thread and use of ths socket in another. unp_bind() now assert the UNP lock, and uipc_bind() now acquires the lock around calls to unp_bind().
2004-08-16 04:41:03 +00:00
struct unpcb *unp;
Merge unp_bind() into uipc_bind(), as it is called only from uipc_bind().
2006-07-23 11:02:12 +00:00
struct vnode *vp;
struct mount *mp;
Change the cap_rights_t type from uint64_t to a structure that we can extend in the future in a backward compatible (API and ABI) way. The cap_rights_t represents capability rights. We used to use one bit to represent one right, but we are running out of spare bits. Currently the new structure provides place for 114 rights (so 50 more than the previous cap_rights_t), but it is possible to grow the structure to hold at least 285 rights, although we can make it even larger if 285 rights won't be enough. The structure definition looks like this: struct cap_rights { uint64_t cr_rights[CAP_RIGHTS_VERSION + 2]; }; The initial CAP_RIGHTS_VERSION is 0. The top two bits in the first element of the cr_rights[] array contain total number of elements in the array - 2. This means if those two bits are equal to 0, we have 2 array elements. The top two bits in all remaining array elements should be 0. The next five bits in all array elements contain array index. Only one bit is used and bit position in this five-bits range defines array index. This means there can be at most five array elements in the future. To define new right the CAPRIGHT() macro must be used. The macro takes two arguments - an array index and a bit to set, eg. #define CAP_PDKILL CAPRIGHT(1, 0x0000000000000800ULL) We still support aliases that combine few rights, but the rights have to belong to the same array element, eg: #define CAP_LOOKUP CAPRIGHT(0, 0x0000000000000400ULL) #define CAP_FCHMOD CAPRIGHT(0, 0x0000000000002000ULL) #define CAP_FCHMODAT (CAP_FCHMOD | CAP_LOOKUP) There is new API to manage the new cap_rights_t structure: cap_rights_t *cap_rights_init(cap_rights_t *rights, ...); void cap_rights_set(cap_rights_t *rights, ...); void cap_rights_clear(cap_rights_t *rights, ...); bool cap_rights_is_set(const cap_rights_t *rights, ...); bool cap_rights_is_valid(const cap_rights_t *rights); void cap_rights_merge(cap_rights_t *dst, const cap_rights_t *src); void cap_rights_remove(cap_rights_t *dst, const cap_rights_t *src); bool cap_rights_contains(const cap_rights_t *big, const cap_rights_t *little); Capability rights to the cap_rights_init(), cap_rights_set(), cap_rights_clear() and cap_rights_is_set() functions are provided by separating them with commas, eg: cap_rights_t rights; cap_rights_init(&rights, CAP_READ, CAP_WRITE, CAP_FSTAT); There is no need to terminate the list of rights, as those functions are actually macros that take care of the termination, eg: #define cap_rights_set(rights, ...) \ __cap_rights_set((rights), __VA_ARGS__, 0ULL) void __cap_rights_set(cap_rights_t *rights, ...); Thanks to using one bit as an array index we can assert in those functions that there are no two rights belonging to different array elements provided together. For example this is illegal and will be detected, because CAP_LOOKUP belongs to element 0 and CAP_PDKILL to element 1: cap_rights_init(&rights, CAP_LOOKUP | CAP_PDKILL); Providing several rights that belongs to the same array's element this way is correct, but is not advised. It should only be used for aliases definition. This commit also breaks compatibility with some existing Capsicum system calls, but I see no other way to do that. This should be fine as Capsicum is still experimental and this change is not going to 9.x. Sponsored by: The FreeBSD Foundation
2013-09-05 00:09:56 +00:00
cap_rights_t rights;
Merge unp_bind() into uipc_bind(), as it is called only from uipc_bind().
2006-07-23 11:02:12 +00:00
char *buf;
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
Make bind(2) and connect(2) return EAFNOSUPPORT for AF_UNIX on wrong address family. See https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=191586 for the original discussion. Reviewed by: terry
2014-07-14 06:00:01 +00:00
if (nam->sa_family != AF_UNIX)
return (EAFNOSUPPORT);
Always acquire the UNIX domain socket subsystem lock (UNP lock) before dereferencing sotounpcb() and checking its value, as so_pcb is protected by protocol locking, not subsystem locking. This prevents races during close() by one thread and use of ths socket in another. unp_bind() now assert the UNP lock, and uipc_bind() now acquires the lock around calls to unp_bind().
2004-08-16 04:41:03 +00:00
unp = sotounpcb(so);
Modify UNIX domain sockets to guarantee, and assume, that so_pcb is always defined for an in-use socket. This allows us to eliminate countless tests of whether so_pcb is non-NULL, eliminating dozens of error cases. For now, retain the call to sotryfree() in the uipc_abort() path, but this will eventually move to soabort(). These new assumptions should be largely correct, and will become more so as the socket/pcb reference model is fixed. Removing the notion that so_pcb can be non-NULL is a critical step towards further fine-graining of the UNIX domain socket locking, as the so_pcb reference no longer needs to be protected using locks, instead it is a property of the socket life cycle.
2006-03-17 13:52:57 +00:00
KASSERT(unp != NULL, ("uipc_bind: unp == NULL"));
Add two new unpcb flags, UNP_BINDING and UNP_CONNECTING, which will be used to mark UNIX domain sockets as being in the process of binding or connecting. Use these to prevent simultaneous bind or connect operations by multiple threads or processes on the same socket at the same time, which closes race conditions present in the UNIX domain socket implementation since inception.
2006-07-23 12:01:14 +00:00
Fix handling of corrupt compress(1)ed data. [11:04] Add missing length checks on unix socket addresses. [11:05] Approved by: so (cperciva) Approved by: re (kensmith) Security: FreeBSD-SA-11:04.compress Security: CVE-2011-2895 [11:04] Security: FreeBSD-SA-11:05.unix
2011-09-28 08:47:17 +00:00
if (soun->sun_len > sizeof(struct sockaddr_un))
return (EINVAL);
Add two new unpcb flags, UNP_BINDING and UNP_CONNECTING, which will be used to mark UNIX domain sockets as being in the process of binding or connecting. Use these to prevent simultaneous bind or connect operations by multiple threads or processes on the same socket at the same time, which closes race conditions present in the UNIX domain socket implementation since inception.
2006-07-23 12:01:14 +00:00
namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path);
if (namelen <= 0)
return (EINVAL);
Merge unp_bind() into uipc_bind(), as it is called only from uipc_bind().
2006-07-23 11:02:12 +00:00
/*
Add two new unpcb flags, UNP_BINDING and UNP_CONNECTING, which will be used to mark UNIX domain sockets as being in the process of binding or connecting. Use these to prevent simultaneous bind or connect operations by multiple threads or processes on the same socket at the same time, which closes race conditions present in the UNIX domain socket implementation since inception.
2006-07-23 12:01:14 +00:00
* We don't allow simultaneous bind() calls on a single UNIX domain
* socket, so flag in-progress operations, and return an error if an
* operation is already in progress.
*
* Historically, we have not allowed a socket to be rebound, so this
Clarify and update quite a few comments to reflect locking optimizations, the addition of unpcb refcounts, and bug fixes. Some of these fixes are appropriate for MFC. MFC after: 3 days
2007-05-11 12:10:45 +00:00
* also returns an error. Not allowing re-binding simplifies the
* implementation and avoids a great many possible failure modes.
Merge unp_bind() into uipc_bind(), as it is called only from uipc_bind().
2006-07-23 11:02:12 +00:00
*/
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
UNP_PCB_LOCK(unp);
Merge unp_bind() into uipc_bind(), as it is called only from uipc_bind().
2006-07-23 11:02:12 +00:00
if (unp->unp_vnode != NULL) {
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
UNP_PCB_UNLOCK(unp);
Merge unp_bind() into uipc_bind(), as it is called only from uipc_bind().
2006-07-23 11:02:12 +00:00
return (EINVAL);
}
Add two new unpcb flags, UNP_BINDING and UNP_CONNECTING, which will be used to mark UNIX domain sockets as being in the process of binding or connecting. Use these to prevent simultaneous bind or connect operations by multiple threads or processes on the same socket at the same time, which closes race conditions present in the UNIX domain socket implementation since inception.
2006-07-23 12:01:14 +00:00
if (unp->unp_flags & UNP_BINDING) {
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
UNP_PCB_UNLOCK(unp);
Add two new unpcb flags, UNP_BINDING and UNP_CONNECTING, which will be used to mark UNIX domain sockets as being in the process of binding or connecting. Use these to prevent simultaneous bind or connect operations by multiple threads or processes on the same socket at the same time, which closes race conditions present in the UNIX domain socket implementation since inception.
2006-07-23 12:01:14 +00:00
return (EALREADY);
Merge unp_bind() into uipc_bind(), as it is called only from uipc_bind().
2006-07-23 11:02:12 +00:00
}
Add two new unpcb flags, UNP_BINDING and UNP_CONNECTING, which will be used to mark UNIX domain sockets as being in the process of binding or connecting. Use these to prevent simultaneous bind or connect operations by multiple threads or processes on the same socket at the same time, which closes race conditions present in the UNIX domain socket implementation since inception.
2006-07-23 12:01:14 +00:00
unp->unp_flags |= UNP_BINDING;
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
UNP_PCB_UNLOCK(unp);
Merge unp_bind() into uipc_bind(), as it is called only from uipc_bind().
2006-07-23 11:02:12 +00:00
buf = malloc(namelen + 1, M_TEMP, M_WAITOK);
Use bcopy instead of strlcpy in uipc_bind and unp_connect, since soun->sun_path isn't a null-terminated string. As UNIX(4) states, "the terminating NUL is not part of the address." Since strlcpy has to return "the total length of the string [it] tried to create," it walks off the end of soun->sun_path looking for a \0. This reverts r105332. Reported by: Ryan Stone
2008-07-03 23:26:10 +00:00
bcopy(soun->sun_path, buf, namelen);
buf[namelen] = 0;
Merge unp_bind() into uipc_bind(), as it is called only from uipc_bind().
2006-07-23 11:02:12 +00:00
restart:
The VOP_LOOKUP() implementations for CREATE op do not put the name into namecache, to avoid cache trashing when doing large operations. E.g., tar archive extraction is not usually followed by access to many of the files created. Right now, each VOP_LOOKUP() implementation explicitely knowns about this quirk and tests for both MAKEENTRY flag presence and op != CREATE to make the call to cache_enter(). Centralize the handling of the quirk into VFS, by deciding to cache only by MAKEENTRY flag in VOP. VFS now sets NOCACHE flag for CREATE namei() calls. Note that the change in semantic is backward-compatible and could be merged to the stable branch, and is compatible with non-changed third-party filesystems which correctly handle MAKEENTRY. Suggested by: Chris Torek <torek@pi-coral.com> Reviewed by: mckusick Tested by: pho Sponsored by: The FreeBSD Foundation MFC after: 2 weeks
2014-12-18 10:01:12 +00:00
NDINIT_ATRIGHTS(&nd, CREATE, NOFOLLOW | LOCKPARENT | SAVENAME | NOCACHE,
Change the cap_rights_t type from uint64_t to a structure that we can extend in the future in a backward compatible (API and ABI) way. The cap_rights_t represents capability rights. We used to use one bit to represent one right, but we are running out of spare bits. Currently the new structure provides place for 114 rights (so 50 more than the previous cap_rights_t), but it is possible to grow the structure to hold at least 285 rights, although we can make it even larger if 285 rights won't be enough. The structure definition looks like this: struct cap_rights { uint64_t cr_rights[CAP_RIGHTS_VERSION + 2]; }; The initial CAP_RIGHTS_VERSION is 0. The top two bits in the first element of the cr_rights[] array contain total number of elements in the array - 2. This means if those two bits are equal to 0, we have 2 array elements. The top two bits in all remaining array elements should be 0. The next five bits in all array elements contain array index. Only one bit is used and bit position in this five-bits range defines array index. This means there can be at most five array elements in the future. To define new right the CAPRIGHT() macro must be used. The macro takes two arguments - an array index and a bit to set, eg. #define CAP_PDKILL CAPRIGHT(1, 0x0000000000000800ULL) We still support aliases that combine few rights, but the rights have to belong to the same array element, eg: #define CAP_LOOKUP CAPRIGHT(0, 0x0000000000000400ULL) #define CAP_FCHMOD CAPRIGHT(0, 0x0000000000002000ULL) #define CAP_FCHMODAT (CAP_FCHMOD | CAP_LOOKUP) There is new API to manage the new cap_rights_t structure: cap_rights_t *cap_rights_init(cap_rights_t *rights, ...); void cap_rights_set(cap_rights_t *rights, ...); void cap_rights_clear(cap_rights_t *rights, ...); bool cap_rights_is_set(const cap_rights_t *rights, ...); bool cap_rights_is_valid(const cap_rights_t *rights); void cap_rights_merge(cap_rights_t *dst, const cap_rights_t *src); void cap_rights_remove(cap_rights_t *dst, const cap_rights_t *src); bool cap_rights_contains(const cap_rights_t *big, const cap_rights_t *little); Capability rights to the cap_rights_init(), cap_rights_set(), cap_rights_clear() and cap_rights_is_set() functions are provided by separating them with commas, eg: cap_rights_t rights; cap_rights_init(&rights, CAP_READ, CAP_WRITE, CAP_FSTAT); There is no need to terminate the list of rights, as those functions are actually macros that take care of the termination, eg: #define cap_rights_set(rights, ...) \ __cap_rights_set((rights), __VA_ARGS__, 0ULL) void __cap_rights_set(cap_rights_t *rights, ...); Thanks to using one bit as an array index we can assert in those functions that there are no two rights belonging to different array elements provided together. For example this is illegal and will be detected, because CAP_LOOKUP belongs to element 0 and CAP_PDKILL to element 1: cap_rights_init(&rights, CAP_LOOKUP | CAP_PDKILL); Providing several rights that belongs to the same array's element this way is correct, but is not advised. It should only be used for aliases definition. This commit also breaks compatibility with some existing Capsicum system calls, but I see no other way to do that. This should be fine as Capsicum is still experimental and this change is not going to 9.x. Sponsored by: The FreeBSD Foundation
2013-09-05 00:09:56 +00:00
UIO_SYSSPACE, buf, fd, cap_rights_init(&rights, CAP_BINDAT), td);
Merge unp_bind() into uipc_bind(), as it is called only from uipc_bind().
2006-07-23 11:02:12 +00:00
/* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */
error = namei(&nd);
if (error)
Add two new unpcb flags, UNP_BINDING and UNP_CONNECTING, which will be used to mark UNIX domain sockets as being in the process of binding or connecting. Use these to prevent simultaneous bind or connect operations by multiple threads or processes on the same socket at the same time, which closes race conditions present in the UNIX domain socket implementation since inception.
2006-07-23 12:01:14 +00:00
goto error;
Merge unp_bind() into uipc_bind(), as it is called only from uipc_bind().
2006-07-23 11:02:12 +00:00
vp = nd.ni_vp;
if (vp != NULL || vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) {
NDFREE(&nd, NDF_ONLY_PNBUF);
if (nd.ni_dvp == vp)
vrele(nd.ni_dvp);
else
vput(nd.ni_dvp);
if (vp != NULL) {
vrele(vp);
error = EADDRINUSE;
Add two new unpcb flags, UNP_BINDING and UNP_CONNECTING, which will be used to mark UNIX domain sockets as being in the process of binding or connecting. Use these to prevent simultaneous bind or connect operations by multiple threads or processes on the same socket at the same time, which closes race conditions present in the UNIX domain socket implementation since inception.
2006-07-23 12:01:14 +00:00
goto error;
Merge unp_bind() into uipc_bind(), as it is called only from uipc_bind().
2006-07-23 11:02:12 +00:00
}
error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH);
if (error)
Add two new unpcb flags, UNP_BINDING and UNP_CONNECTING, which will be used to mark UNIX domain sockets as being in the process of binding or connecting. Use these to prevent simultaneous bind or connect operations by multiple threads or processes on the same socket at the same time, which closes race conditions present in the UNIX domain socket implementation since inception.
2006-07-23 12:01:14 +00:00
goto error;
Merge unp_bind() into uipc_bind(), as it is called only from uipc_bind().
2006-07-23 11:02:12 +00:00
goto restart;
}
VATTR_NULL(&vattr);
vattr.va_type = VSOCK;
vattr.va_mode = (ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask);
#ifdef MAC
Merge first in a series of TrustedBSD MAC Framework KPI changes from Mac OS X Leopard--rationalize naming for entry points to the following general forms: mac_<object>_<method/action> mac_<object>_check_<method/action> The previous naming scheme was inconsistent and mostly reversed from the new scheme. Also, make object types more consistent and remove spaces from object types that contain multiple parts ("posix_sem" -> "posixsem") to make mechanical parsing easier. Introduce a new "netinet" object type for certain IPv4/IPv6-related methods. Also simplify, slightly, some entry point names. All MAC policy modules will need to be recompiled, and modules not updates as part of this commit will need to be modified to conform to the new KPI. Sponsored by: SPARTA (original patches against Mac OS X) Obtained from: TrustedBSD Project, Apple Computer
2007-10-24 19:04:04 +00:00
error = mac_vnode_check_create(td->td_ucred, nd.ni_dvp, &nd.ni_cnd,
Merge unp_bind() into uipc_bind(), as it is called only from uipc_bind().
2006-07-23 11:02:12 +00:00
&vattr);
#endif
Remove VOP_LEASE and supporting functions. This hasn't been used since the removal of NQNFS, but was left in in case it was required for NFSv4. Since our new NFSv4 client and server can't use it for their requirements, GC the old mechanism, as well as other unused lease- related code and interfaces. Due to its impact on kernel programming and binary interfaces, this change should not be MFC'd. Proposed by: jeff Reviewed by: jeff Discussed with: rmacklem, zach loafman @ isilon
2009-04-10 10:52:19 +00:00
if (error == 0)
Merge unp_bind() into uipc_bind(), as it is called only from uipc_bind().
2006-07-23 11:02:12 +00:00
error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr);
NDFREE(&nd, NDF_ONLY_PNBUF);
vput(nd.ni_dvp);
if (error) {
vn_finished_write(mp);
Add two new unpcb flags, UNP_BINDING and UNP_CONNECTING, which will be used to mark UNIX domain sockets as being in the process of binding or connecting. Use these to prevent simultaneous bind or connect operations by multiple threads or processes on the same socket at the same time, which closes race conditions present in the UNIX domain socket implementation since inception.
2006-07-23 12:01:14 +00:00
goto error;
Merge unp_bind() into uipc_bind(), as it is called only from uipc_bind().
2006-07-23 11:02:12 +00:00
}
vp = nd.ni_vp;
When we do open, we should lock the vnode exclusively. This fixes few races: - fifo race, where two threads assign v_fifoinfo, - v_writecount modifications, - v_object modifications, - and probably more... Discussed with: kib, ups Approved by: re (rwatson)
2007-07-26 16:58:09 +00:00
ASSERT_VOP_ELOCKED(vp, "uipc_bind");
Merge unp_bind() into uipc_bind(), as it is called only from uipc_bind().
2006-07-23 11:02:12 +00:00
soun = (struct sockaddr_un *)sodupsockaddr(nam, M_WAITOK);
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
Decompose the global UNIX domain sockets rwlock into two different locks: a global list/counter/generation counter protected by a new mutex unp_list_lock, and a global linkage rwlock, unp_global_rwlock, which protects the connections between UNIX domain sockets. This eliminates conditional lock acquisition that was previously a property of the global lock being held over sonewconn() leading to a call to uipc_attach(), which also required the global lock, but couldn't rely on it as other paths existed to uipc_attach() that didn't hold it: now uipc_attach() uses only the list lock, which follows the linkage lock in the lock order. It may also reduce contention on the global lock for some workloads. Add global UNIX domain socket locks to hard-coded witness lock order. MFC after: 1 week Discussed with: kris
2009-03-08 21:48:29 +00:00
UNP_LINK_WLOCK();
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
UNP_PCB_LOCK(unp);
For UNIX sockets make vnode point not to the socket, but to the UNIX PCB, since the latter is the thing that links together VFS and sockets. While here, make the union in the struct vnode anonymous.
2017-06-02 17:31:25 +00:00
VOP_UNP_BIND(vp, unp);
Merge unp_bind() into uipc_bind(), as it is called only from uipc_bind().
2006-07-23 11:02:12 +00:00
unp->unp_vnode = vp;
unp->unp_addr = soun;
Add two new unpcb flags, UNP_BINDING and UNP_CONNECTING, which will be used to mark UNIX domain sockets as being in the process of binding or connecting. Use these to prevent simultaneous bind or connect operations by multiple threads or processes on the same socket at the same time, which closes race conditions present in the UNIX domain socket implementation since inception.
2006-07-23 12:01:14 +00:00
unp->unp_flags &= ~UNP_BINDING;
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
UNP_PCB_UNLOCK(unp);
Decompose the global UNIX domain sockets rwlock into two different locks: a global list/counter/generation counter protected by a new mutex unp_list_lock, and a global linkage rwlock, unp_global_rwlock, which protects the connections between UNIX domain sockets. This eliminates conditional lock acquisition that was previously a property of the global lock being held over sonewconn() leading to a call to uipc_attach(), which also required the global lock, but couldn't rely on it as other paths existed to uipc_attach() that didn't hold it: now uipc_attach() uses only the list lock, which follows the linkage lock in the lock order. It may also reduce contention on the global lock for some workloads. Add global UNIX domain socket locks to hard-coded witness lock order. MFC after: 1 week Discussed with: kris
2009-03-08 21:48:29 +00:00
UNP_LINK_WUNLOCK();
VOP_LOCK1() (and so VOP_LOCK()) and VOP_UNLOCK() are only used in conjuction with 'thread' argument passing which is always curthread. Remove the unuseful extra-argument and pass explicitly curthread to lower layer functions, when necessary. KPI results broken by this change, which should affect several ports, so version bumping and manpage update will be further committed. Tested by: kris, pho, Diego Sardina <siarodx at gmail dot com>
2008-01-13 14:44:15 +00:00
VOP_UNLOCK(vp, 0);
Merge unp_bind() into uipc_bind(), as it is called only from uipc_bind().
2006-07-23 11:02:12 +00:00
vn_finished_write(mp);
Add two new unpcb flags, UNP_BINDING and UNP_CONNECTING, which will be used to mark UNIX domain sockets as being in the process of binding or connecting. Use these to prevent simultaneous bind or connect operations by multiple threads or processes on the same socket at the same time, which closes race conditions present in the UNIX domain socket implementation since inception.
2006-07-23 12:01:14 +00:00
free(buf, M_TEMP);
return (0);
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
Add two new unpcb flags, UNP_BINDING and UNP_CONNECTING, which will be used to mark UNIX domain sockets as being in the process of binding or connecting. Use these to prevent simultaneous bind or connect operations by multiple threads or processes on the same socket at the same time, which closes race conditions present in the UNIX domain socket implementation since inception.
2006-07-23 12:01:14 +00:00
error:
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
UNP_PCB_LOCK(unp);
Add two new unpcb flags, UNP_BINDING and UNP_CONNECTING, which will be used to mark UNIX domain sockets as being in the process of binding or connecting. Use these to prevent simultaneous bind or connect operations by multiple threads or processes on the same socket at the same time, which closes race conditions present in the UNIX domain socket implementation since inception.
2006-07-23 12:01:14 +00:00
unp->unp_flags &= ~UNP_BINDING;
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
UNP_PCB_UNLOCK(unp);
Merge unp_bind() into uipc_bind(), as it is called only from uipc_bind().
2006-07-23 11:02:12 +00:00
free(buf, M_TEMP);
Always acquire the UNIX domain socket subsystem lock (UNP lock) before dereferencing sotounpcb() and checking its value, as so_pcb is protected by protocol locking, not subsystem locking. This prevents races during close() by one thread and use of ths socket in another. unp_bind() now assert the UNP lock, and uipc_bind() now acquires the lock around calls to unp_bind().
2004-08-16 04:41:03 +00:00
return (error);
The long-awaited mega-massive-network-code- cleanup. Part I. This commit includes the following changes: 1) Old-style (pr_usrreq()) protocols are no longer supported, the compatibility glue for them is deleted, and the kernel will panic on boot if any are compiled in. 2) Certain protocol entry points are modified to take a process structure, so they they can easily tell whether or not it is possible to sleep, and also to access credentials. 3) SS_PRIV is no more, and with it goes the SO_PRIVSTATE setsockopt() call. Protocols should use the process pointer they are now passed. 4) The PF_LOCAL and PF_ROUTE families have been updated to use the new style, as has the `raw' skeleton family. 5) PF_LOCAL sockets now obey the process's umask when creating a socket in the filesystem. As a result, LINT is now broken. I'm hoping that some enterprising hacker with a bit more time will either make the broken bits work (should be easy for netipx) or dike them out.
1997-04-27 20:01:29 +00:00
}
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
- Implement two new system calls: int bindat(int fd, int s, const struct sockaddr *addr, socklen_t addrlen); int connectat(int fd, int s, const struct sockaddr *name, socklen_t namelen); which allow to bind and connect respectively to a UNIX domain socket with a path relative to the directory associated with the given file descriptor 'fd'. - Add manual pages for the new syscalls. - Make the new syscalls available for processes in capability mode sandbox. - Add capability rights CAP_BINDAT and CAP_CONNECTAT that has to be present on the directory descriptor for the syscalls to work. - Update audit(4) to support those two new syscalls and to handle path in sockaddr_un structure relative to the given directory descriptor. - Update procstat(1) to recognize the new capability rights. - Document the new capability rights in cap_rights_limit(2). Sponsored by: The FreeBSD Foundation Discussed with: rwatson, jilles, kib, des
2013-03-02 21:11:30 +00:00
static int
uipc_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
{
return (uipc_bindat(AT_FDCWD, so, nam, td));
}
The long-awaited mega-massive-network-code- cleanup. Part I. This commit includes the following changes: 1) Old-style (pr_usrreq()) protocols are no longer supported, the compatibility glue for them is deleted, and the kernel will panic on boot if any are compiled in. 2) Certain protocol entry points are modified to take a process structure, so they they can easily tell whether or not it is possible to sleep, and also to access credentials. 3) SS_PRIV is no more, and with it goes the SO_PRIVSTATE setsockopt() call. Protocols should use the process pointer they are now passed. 4) The PF_LOCAL and PF_ROUTE families have been updated to use the new style, as has the `raw' skeleton family. 5) PF_LOCAL sockets now obey the process's umask when creating a socket in the filesystem. As a result, LINT is now broken. I'm hoping that some enterprising hacker with a bit more time will either make the broken bits work (should be easy for netipx) or dike them out.
1997-04-27 20:01:29 +00:00
static int
KSE Milestone 2 Note ALL MODULES MUST BE RECOMPILED make the kernel aware that there are smaller units of scheduling than the process. (but only allow one thread per process at this time). This is functionally equivalent to teh previousl -current except that there is a thread associated with each process. Sorry john! (your next MFC will be a doosie!) Reviewed by: peter@freebsd.org, dillon@freebsd.org X-MFC after: ha ha ha ha
2001-09-12 08:38:13 +00:00
uipc_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
The long-awaited mega-massive-network-code- cleanup. Part I. This commit includes the following changes: 1) Old-style (pr_usrreq()) protocols are no longer supported, the compatibility glue for them is deleted, and the kernel will panic on boot if any are compiled in. 2) Certain protocol entry points are modified to take a process structure, so they they can easily tell whether or not it is possible to sleep, and also to access credentials. 3) SS_PRIV is no more, and with it goes the SO_PRIVSTATE setsockopt() call. Protocols should use the process pointer they are now passed. 4) The PF_LOCAL and PF_ROUTE families have been updated to use the new style, as has the `raw' skeleton family. 5) PF_LOCAL sockets now obey the process's umask when creating a socket in the filesystem. As a result, LINT is now broken. I'm hoping that some enterprising hacker with a bit more time will either make the broken bits work (should be easy for netipx) or dike them out.
1997-04-27 20:01:29 +00:00
{
Introduce a subsystem lock around UNIX domain sockets in order to protect global and allocated variables. This strategy is derived from work originally developed by BSDi for BSD/OS, and applied to FreeBSD by Sam Leffler: - Add unp_mtx, a global mutex which will protect all UNIX domain socket related variables, structures, etc. - Add UNP_LOCK(), UNP_UNLOCK(), UNP_LOCK_ASSERT() macros. - Acquire unp_mtx on entering most UNIX domain socket code, drop/re-acquire around calls into VFS, and release it on return. - Avoid performing sodupsockaddr() while holding the mutex, so in general move to allocating storage before acquiring the mutex to copy the data. - Make a stack copy of the xucred rather than copying out while holding unp_mtx. Copy the peer credential out after releasing the mutex. - Add additional assertions of vnode locks following VOP_CREATE(). A few notes: - Use of an sx lock for the file list mutex may cause problems with regard to unp_mtx when garbage collection passed file descriptors. - The locking in unp_pcblist() for sysctl monitoring is correct subject to the unpcb zone not returning memory for reuse by other subsystems (consistent with similar existing concerns). - Sam's version of this change, as with the BSD/OS version, made use of both a global lock and per-unpcb locks. However, in practice, the global lock covered all accesses, so I have simplified out the unpcb locks in the interest of getting this merged faster (reducing the overhead but not sacrificing granularity in most cases). We will want to explore possibilities for improving lock granularity in this code in the future. Submitted by: sam Sponsored by: FreeBSD Foundatiuon Obtained from: BSD/OS 5 snapshot provided by BSDi
2004-06-10 21:34:38 +00:00
int error;
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
In uipc_connect(), assert that the passed thread is curthread, and pass td into unp_connect() instead of reading curthread.
2004-07-25 23:30:43 +00:00
KASSERT(td == curthread, ("uipc_connect: td != curthread"));
Decompose the global UNIX domain sockets rwlock into two different locks: a global list/counter/generation counter protected by a new mutex unp_list_lock, and a global linkage rwlock, unp_global_rwlock, which protects the connections between UNIX domain sockets. This eliminates conditional lock acquisition that was previously a property of the global lock being held over sonewconn() leading to a call to uipc_attach(), which also required the global lock, but couldn't rely on it as other paths existed to uipc_attach() that didn't hold it: now uipc_attach() uses only the list lock, which follows the linkage lock in the lock order. It may also reduce contention on the global lock for some workloads. Add global UNIX domain socket locks to hard-coded witness lock order. MFC after: 1 week Discussed with: kris
2009-03-08 21:48:29 +00:00
UNP_LINK_WLOCK();
In uipc_connect(), assert that the passed thread is curthread, and pass td into unp_connect() instead of reading curthread.
2004-07-25 23:30:43 +00:00
error = unp_connect(so, nam, td);
Decompose the global UNIX domain sockets rwlock into two different locks: a global list/counter/generation counter protected by a new mutex unp_list_lock, and a global linkage rwlock, unp_global_rwlock, which protects the connections between UNIX domain sockets. This eliminates conditional lock acquisition that was previously a property of the global lock being held over sonewconn() leading to a call to uipc_attach(), which also required the global lock, but couldn't rely on it as other paths existed to uipc_attach() that didn't hold it: now uipc_attach() uses only the list lock, which follows the linkage lock in the lock order. It may also reduce contention on the global lock for some workloads. Add global UNIX domain socket locks to hard-coded witness lock order. MFC after: 1 week Discussed with: kris
2009-03-08 21:48:29 +00:00
UNP_LINK_WUNLOCK();
Introduce a subsystem lock around UNIX domain sockets in order to protect global and allocated variables. This strategy is derived from work originally developed by BSDi for BSD/OS, and applied to FreeBSD by Sam Leffler: - Add unp_mtx, a global mutex which will protect all UNIX domain socket related variables, structures, etc. - Add UNP_LOCK(), UNP_UNLOCK(), UNP_LOCK_ASSERT() macros. - Acquire unp_mtx on entering most UNIX domain socket code, drop/re-acquire around calls into VFS, and release it on return. - Avoid performing sodupsockaddr() while holding the mutex, so in general move to allocating storage before acquiring the mutex to copy the data. - Make a stack copy of the xucred rather than copying out while holding unp_mtx. Copy the peer credential out after releasing the mutex. - Add additional assertions of vnode locks following VOP_CREATE(). A few notes: - Use of an sx lock for the file list mutex may cause problems with regard to unp_mtx when garbage collection passed file descriptors. - The locking in unp_pcblist() for sysctl monitoring is correct subject to the unpcb zone not returning memory for reuse by other subsystems (consistent with similar existing concerns). - Sam's version of this change, as with the BSD/OS version, made use of both a global lock and per-unpcb locks. However, in practice, the global lock covered all accesses, so I have simplified out the unpcb locks in the interest of getting this merged faster (reducing the overhead but not sacrificing granularity in most cases). We will want to explore possibilities for improving lock granularity in this code in the future. Submitted by: sam Sponsored by: FreeBSD Foundatiuon Obtained from: BSD/OS 5 snapshot provided by BSDi
2004-06-10 21:34:38 +00:00
return (error);
The long-awaited mega-massive-network-code- cleanup. Part I. This commit includes the following changes: 1) Old-style (pr_usrreq()) protocols are no longer supported, the compatibility glue for them is deleted, and the kernel will panic on boot if any are compiled in. 2) Certain protocol entry points are modified to take a process structure, so they they can easily tell whether or not it is possible to sleep, and also to access credentials. 3) SS_PRIV is no more, and with it goes the SO_PRIVSTATE setsockopt() call. Protocols should use the process pointer they are now passed. 4) The PF_LOCAL and PF_ROUTE families have been updated to use the new style, as has the `raw' skeleton family. 5) PF_LOCAL sockets now obey the process's umask when creating a socket in the filesystem. As a result, LINT is now broken. I'm hoping that some enterprising hacker with a bit more time will either make the broken bits work (should be easy for netipx) or dike them out.
1997-04-27 20:01:29 +00:00
}
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
- Implement two new system calls: int bindat(int fd, int s, const struct sockaddr *addr, socklen_t addrlen); int connectat(int fd, int s, const struct sockaddr *name, socklen_t namelen); which allow to bind and connect respectively to a UNIX domain socket with a path relative to the directory associated with the given file descriptor 'fd'. - Add manual pages for the new syscalls. - Make the new syscalls available for processes in capability mode sandbox. - Add capability rights CAP_BINDAT and CAP_CONNECTAT that has to be present on the directory descriptor for the syscalls to work. - Update audit(4) to support those two new syscalls and to handle path in sockaddr_un structure relative to the given directory descriptor. - Update procstat(1) to recognize the new capability rights. - Document the new capability rights in cap_rights_limit(2). Sponsored by: The FreeBSD Foundation Discussed with: rwatson, jilles, kib, des
2013-03-02 21:11:30 +00:00
static int
uipc_connectat(int fd, struct socket *so, struct sockaddr *nam,
struct thread *td)
{
int error;
KASSERT(td == curthread, ("uipc_connectat: td != curthread"));
UNP_LINK_WLOCK();
error = unp_connectat(fd, so, nam, td);
UNP_LINK_WUNLOCK();
return (error);
}
Change semantics of socket close and detach. Add a new protocol switch function, pru_close, to notify protocols that the file descriptor or other consumer of a socket is closing the socket. pru_abort is now a notification of close also, and no longer detaches. pru_detach is no longer used to notify of close, and will be called during socket tear-down by sofree() when all references to a socket evaporate after an earlier call to abort or close the socket. This means detach is now an unconditional teardown of a socket, whereas previously sockets could persist after detach of the protocol retained a reference. This faciliates sharing mutexes between layers of the network stack as the mutex is required during the checking and removal of references at the head of sofree(). With this change, pru_detach can now assume that the mutex will no longer be required by the socket layer after completion, whereas before this was not necessarily true. Reviewed by: gnn
2006-07-21 17:11:15 +00:00
static void
uipc_close(struct socket *so)
{
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
struct unpcb *unp, *unp2;
Listening sockets improvements. o Separate fields of struct socket that belong to listening from fields that belong to normal dataflow, and unionize them. This shrinks the structure a bit. - Take out selinfo's from the socket buffers into the socket. The first reason is to support braindamaged scenario when a socket is added to kevent(2) and then listen(2) is cast on it. The second reason is that there is future plan to make socket buffers pluggable, so that for a dataflow socket a socket buffer can be changed, and in this case we also want to keep same selinfos through the lifetime of a socket. - Remove struct struct so_accf. Since now listening stuff no longer affects struct socket size, just move its fields into listening part of the union. - Provide sol_upcall field and enforce that so_upcall_set() may be called only on a dataflow socket, which has buffers, and for listening sockets provide solisten_upcall_set(). o Remove ACCEPT_LOCK() global. - Add a mutex to socket, to be used instead of socket buffer lock to lock fields of struct socket that don't belong to a socket buffer. - Allow to acquire two socket locks, but the first one must belong to a listening socket. - Make soref()/sorele() to use atomic(9). This allows in some situations to do soref() without owning socket lock. There is place for improvement here, it is possible to make sorele() also to lock optionally. - Most protocols aren't touched by this change, except UNIX local sockets. See below for more information. o Reduce copy-and-paste in kernel modules that accept connections from listening sockets: provide function solisten_dequeue(), and use it in the following modules: ctl(4), iscsi(4), ng_btsocket(4), ng_ksocket(4), infiniband, rpc. o UNIX local sockets. - Removal of ACCEPT_LOCK() global uncovered several races in the UNIX local sockets. Most races exist around spawning a new socket, when we are connecting to a local listening socket. To cover them, we need to hold locks on both PCBs when spawning a third one. This means holding them across sonewconn(). This creates a LOR between pcb locks and unp_list_lock. - To fix the new LOR, abandon the global unp_list_lock in favor of global unp_link_lock. Indeed, separating these two locks didn't provide us any extra parralelism in the UNIX sockets. - Now call into uipc_attach() may happen with unp_link_lock hold if, we are accepting, or without unp_link_lock in case if we are just creating a socket. - Another problem in UNIX sockets is that uipc_close() basicly did nothing for a listening socket. The vnode remained opened for connections. This is fixed by removing vnode in uipc_close(). Maybe the right way would be to do it for all sockets (not only listening), simply move the vnode teardown from uipc_detach() to uipc_close()? Sponsored by: Netflix Differential Revision: https://reviews.freebsd.org/D9770
2017-06-08 21:30:34 +00:00
struct vnode *vp = NULL;
Change semantics of socket close and detach. Add a new protocol switch function, pru_close, to notify protocols that the file descriptor or other consumer of a socket is closing the socket. pru_abort is now a notification of close also, and no longer detaches. pru_detach is no longer used to notify of close, and will be called during socket tear-down by sofree() when all references to a socket evaporate after an earlier call to abort or close the socket. This means detach is now an unconditional teardown of a socket, whereas previously sockets could persist after detach of the protocol retained a reference. This faciliates sharing mutexes between layers of the network stack as the mutex is required during the checking and removal of references at the head of sofree(). With this change, pru_detach can now assume that the mutex will no longer be required by the socket layer after completion, whereas before this was not necessarily true. Reviewed by: gnn
2006-07-21 17:11:15 +00:00
unp = sotounpcb(so);
KASSERT(unp != NULL, ("uipc_close: unp == NULL"));
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
Decompose the global UNIX domain sockets rwlock into two different locks: a global list/counter/generation counter protected by a new mutex unp_list_lock, and a global linkage rwlock, unp_global_rwlock, which protects the connections between UNIX domain sockets. This eliminates conditional lock acquisition that was previously a property of the global lock being held over sonewconn() leading to a call to uipc_attach(), which also required the global lock, but couldn't rely on it as other paths existed to uipc_attach() that didn't hold it: now uipc_attach() uses only the list lock, which follows the linkage lock in the lock order. It may also reduce contention on the global lock for some workloads. Add global UNIX domain socket locks to hard-coded witness lock order. MFC after: 1 week Discussed with: kris
2009-03-08 21:48:29 +00:00
UNP_LINK_WLOCK();
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
UNP_PCB_LOCK(unp);
unp2 = unp->unp_conn;
if (unp2 != NULL) {
UNP_PCB_LOCK(unp2);
unp_disconnect(unp, unp2);
UNP_PCB_UNLOCK(unp2);
}
Listening sockets improvements. o Separate fields of struct socket that belong to listening from fields that belong to normal dataflow, and unionize them. This shrinks the structure a bit. - Take out selinfo's from the socket buffers into the socket. The first reason is to support braindamaged scenario when a socket is added to kevent(2) and then listen(2) is cast on it. The second reason is that there is future plan to make socket buffers pluggable, so that for a dataflow socket a socket buffer can be changed, and in this case we also want to keep same selinfos through the lifetime of a socket. - Remove struct struct so_accf. Since now listening stuff no longer affects struct socket size, just move its fields into listening part of the union. - Provide sol_upcall field and enforce that so_upcall_set() may be called only on a dataflow socket, which has buffers, and for listening sockets provide solisten_upcall_set(). o Remove ACCEPT_LOCK() global. - Add a mutex to socket, to be used instead of socket buffer lock to lock fields of struct socket that don't belong to a socket buffer. - Allow to acquire two socket locks, but the first one must belong to a listening socket. - Make soref()/sorele() to use atomic(9). This allows in some situations to do soref() without owning socket lock. There is place for improvement here, it is possible to make sorele() also to lock optionally. - Most protocols aren't touched by this change, except UNIX local sockets. See below for more information. o Reduce copy-and-paste in kernel modules that accept connections from listening sockets: provide function solisten_dequeue(), and use it in the following modules: ctl(4), iscsi(4), ng_btsocket(4), ng_ksocket(4), infiniband, rpc. o UNIX local sockets. - Removal of ACCEPT_LOCK() global uncovered several races in the UNIX local sockets. Most races exist around spawning a new socket, when we are connecting to a local listening socket. To cover them, we need to hold locks on both PCBs when spawning a third one. This means holding them across sonewconn(). This creates a LOR between pcb locks and unp_list_lock. - To fix the new LOR, abandon the global unp_list_lock in favor of global unp_link_lock. Indeed, separating these two locks didn't provide us any extra parralelism in the UNIX sockets. - Now call into uipc_attach() may happen with unp_link_lock hold if, we are accepting, or without unp_link_lock in case if we are just creating a socket. - Another problem in UNIX sockets is that uipc_close() basicly did nothing for a listening socket. The vnode remained opened for connections. This is fixed by removing vnode in uipc_close(). Maybe the right way would be to do it for all sockets (not only listening), simply move the vnode teardown from uipc_detach() to uipc_close()? Sponsored by: Netflix Differential Revision: https://reviews.freebsd.org/D9770
2017-06-08 21:30:34 +00:00
if (SOLISTENING(so) && ((vp = unp->unp_vnode) != NULL)) {
VOP_UNP_DETACH(vp);
unp->unp_vnode = NULL;
}
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
UNP_PCB_UNLOCK(unp);
Decompose the global UNIX domain sockets rwlock into two different locks: a global list/counter/generation counter protected by a new mutex unp_list_lock, and a global linkage rwlock, unp_global_rwlock, which protects the connections between UNIX domain sockets. This eliminates conditional lock acquisition that was previously a property of the global lock being held over sonewconn() leading to a call to uipc_attach(), which also required the global lock, but couldn't rely on it as other paths existed to uipc_attach() that didn't hold it: now uipc_attach() uses only the list lock, which follows the linkage lock in the lock order. It may also reduce contention on the global lock for some workloads. Add global UNIX domain socket locks to hard-coded witness lock order. MFC after: 1 week Discussed with: kris
2009-03-08 21:48:29 +00:00
UNP_LINK_WUNLOCK();
Listening sockets improvements. o Separate fields of struct socket that belong to listening from fields that belong to normal dataflow, and unionize them. This shrinks the structure a bit. - Take out selinfo's from the socket buffers into the socket. The first reason is to support braindamaged scenario when a socket is added to kevent(2) and then listen(2) is cast on it. The second reason is that there is future plan to make socket buffers pluggable, so that for a dataflow socket a socket buffer can be changed, and in this case we also want to keep same selinfos through the lifetime of a socket. - Remove struct struct so_accf. Since now listening stuff no longer affects struct socket size, just move its fields into listening part of the union. - Provide sol_upcall field and enforce that so_upcall_set() may be called only on a dataflow socket, which has buffers, and for listening sockets provide solisten_upcall_set(). o Remove ACCEPT_LOCK() global. - Add a mutex to socket, to be used instead of socket buffer lock to lock fields of struct socket that don't belong to a socket buffer. - Allow to acquire two socket locks, but the first one must belong to a listening socket. - Make soref()/sorele() to use atomic(9). This allows in some situations to do soref() without owning socket lock. There is place for improvement here, it is possible to make sorele() also to lock optionally. - Most protocols aren't touched by this change, except UNIX local sockets. See below for more information. o Reduce copy-and-paste in kernel modules that accept connections from listening sockets: provide function solisten_dequeue(), and use it in the following modules: ctl(4), iscsi(4), ng_btsocket(4), ng_ksocket(4), infiniband, rpc. o UNIX local sockets. - Removal of ACCEPT_LOCK() global uncovered several races in the UNIX local sockets. Most races exist around spawning a new socket, when we are connecting to a local listening socket. To cover them, we need to hold locks on both PCBs when spawning a third one. This means holding them across sonewconn(). This creates a LOR between pcb locks and unp_list_lock. - To fix the new LOR, abandon the global unp_list_lock in favor of global unp_link_lock. Indeed, separating these two locks didn't provide us any extra parralelism in the UNIX sockets. - Now call into uipc_attach() may happen with unp_link_lock hold if, we are accepting, or without unp_link_lock in case if we are just creating a socket. - Another problem in UNIX sockets is that uipc_close() basicly did nothing for a listening socket. The vnode remained opened for connections. This is fixed by removing vnode in uipc_close(). Maybe the right way would be to do it for all sockets (not only listening), simply move the vnode teardown from uipc_detach() to uipc_close()? Sponsored by: Netflix Differential Revision: https://reviews.freebsd.org/D9770
2017-06-08 21:30:34 +00:00
if (vp)
vrele(vp);
Change semantics of socket close and detach. Add a new protocol switch function, pru_close, to notify protocols that the file descriptor or other consumer of a socket is closing the socket. pru_abort is now a notification of close also, and no longer detaches. pru_detach is no longer used to notify of close, and will be called during socket tear-down by sofree() when all references to a socket evaporate after an earlier call to abort or close the socket. This means detach is now an unconditional teardown of a socket, whereas previously sockets could persist after detach of the protocol retained a reference. This faciliates sharing mutexes between layers of the network stack as the mutex is required during the checking and removal of references at the head of sofree(). With this change, pru_detach can now assume that the mutex will no longer be required by the socket layer after completion, whereas before this was not necessarily true. Reviewed by: gnn
2006-07-21 17:11:15 +00:00
}
Now that portalfs doesn't directly invoke uipc_connect2(), make it a static symbol. MFC after: 3 days
2008-10-06 18:43:11 +00:00
static int
The long-awaited mega-massive-network-code- cleanup. Part I. This commit includes the following changes: 1) Old-style (pr_usrreq()) protocols are no longer supported, the compatibility glue for them is deleted, and the kernel will panic on boot if any are compiled in. 2) Certain protocol entry points are modified to take a process structure, so they they can easily tell whether or not it is possible to sleep, and also to access credentials. 3) SS_PRIV is no more, and with it goes the SO_PRIVSTATE setsockopt() call. Protocols should use the process pointer they are now passed. 4) The PF_LOCAL and PF_ROUTE families have been updated to use the new style, as has the `raw' skeleton family. 5) PF_LOCAL sockets now obey the process's umask when creating a socket in the filesystem. As a result, LINT is now broken. I'm hoping that some enterprising hacker with a bit more time will either make the broken bits work (should be easy for netipx) or dike them out.
1997-04-27 20:01:29 +00:00
uipc_connect2(struct socket *so1, struct socket *so2)
{
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
struct unpcb *unp, *unp2;
Introduce a subsystem lock around UNIX domain sockets in order to protect global and allocated variables. This strategy is derived from work originally developed by BSDi for BSD/OS, and applied to FreeBSD by Sam Leffler: - Add unp_mtx, a global mutex which will protect all UNIX domain socket related variables, structures, etc. - Add UNP_LOCK(), UNP_UNLOCK(), UNP_LOCK_ASSERT() macros. - Acquire unp_mtx on entering most UNIX domain socket code, drop/re-acquire around calls into VFS, and release it on return. - Avoid performing sodupsockaddr() while holding the mutex, so in general move to allocating storage before acquiring the mutex to copy the data. - Make a stack copy of the xucred rather than copying out while holding unp_mtx. Copy the peer credential out after releasing the mutex. - Add additional assertions of vnode locks following VOP_CREATE(). A few notes: - Use of an sx lock for the file list mutex may cause problems with regard to unp_mtx when garbage collection passed file descriptors. - The locking in unp_pcblist() for sysctl monitoring is correct subject to the unpcb zone not returning memory for reuse by other subsystems (consistent with similar existing concerns). - Sam's version of this change, as with the BSD/OS version, made use of both a global lock and per-unpcb locks. However, in practice, the global lock covered all accesses, so I have simplified out the unpcb locks in the interest of getting this merged faster (reducing the overhead but not sacrificing granularity in most cases). We will want to explore possibilities for improving lock granularity in this code in the future. Submitted by: sam Sponsored by: FreeBSD Foundatiuon Obtained from: BSD/OS 5 snapshot provided by BSDi
2004-06-10 21:34:38 +00:00
int error;
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
Decompose the global UNIX domain sockets rwlock into two different locks: a global list/counter/generation counter protected by a new mutex unp_list_lock, and a global linkage rwlock, unp_global_rwlock, which protects the connections between UNIX domain sockets. This eliminates conditional lock acquisition that was previously a property of the global lock being held over sonewconn() leading to a call to uipc_attach(), which also required the global lock, but couldn't rely on it as other paths existed to uipc_attach() that didn't hold it: now uipc_attach() uses only the list lock, which follows the linkage lock in the lock order. It may also reduce contention on the global lock for some workloads. Add global UNIX domain socket locks to hard-coded witness lock order. MFC after: 1 week Discussed with: kris
2009-03-08 21:48:29 +00:00
UNP_LINK_WLOCK();
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
unp = so1->so_pcb;
Modify UNIX domain sockets to guarantee, and assume, that so_pcb is always defined for an in-use socket. This allows us to eliminate countless tests of whether so_pcb is non-NULL, eliminating dozens of error cases. For now, retain the call to sotryfree() in the uipc_abort() path, but this will eventually move to soabort(). These new assumptions should be largely correct, and will become more so as the socket/pcb reference model is fixed. Removing the notion that so_pcb can be non-NULL is a critical step towards further fine-graining of the UNIX domain socket locking, as the so_pcb reference no longer needs to be protected using locks, instead it is a property of the socket life cycle.
2006-03-17 13:52:57 +00:00
KASSERT(unp != NULL, ("uipc_connect2: unp == NULL"));
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
UNP_PCB_LOCK(unp);
unp2 = so2->so_pcb;
KASSERT(unp2 != NULL, ("uipc_connect2: unp2 == NULL"));
UNP_PCB_LOCK(unp2);
Implement unix(4) socket options LOCAL_CREDS and LOCAL_CONNWAIT. - Add unp_addsockcred() (for LOCAL_CREDS). - Add an argument to unp_connect2() to differentiate between PRU_CONNECT and PRU_CONNECT2. (for LOCAL_CONNWAIT) Obtained from: NetBSD (with some changes)
2005-04-13 00:01:46 +00:00
error = unp_connect2(so1, so2, PRU_CONNECT2);
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
UNP_PCB_UNLOCK(unp2);
UNP_PCB_UNLOCK(unp);
Decompose the global UNIX domain sockets rwlock into two different locks: a global list/counter/generation counter protected by a new mutex unp_list_lock, and a global linkage rwlock, unp_global_rwlock, which protects the connections between UNIX domain sockets. This eliminates conditional lock acquisition that was previously a property of the global lock being held over sonewconn() leading to a call to uipc_attach(), which also required the global lock, but couldn't rely on it as other paths existed to uipc_attach() that didn't hold it: now uipc_attach() uses only the list lock, which follows the linkage lock in the lock order. It may also reduce contention on the global lock for some workloads. Add global UNIX domain socket locks to hard-coded witness lock order. MFC after: 1 week Discussed with: kris
2009-03-08 21:48:29 +00:00
UNP_LINK_WUNLOCK();
Introduce a subsystem lock around UNIX domain sockets in order to protect global and allocated variables. This strategy is derived from work originally developed by BSDi for BSD/OS, and applied to FreeBSD by Sam Leffler: - Add unp_mtx, a global mutex which will protect all UNIX domain socket related variables, structures, etc. - Add UNP_LOCK(), UNP_UNLOCK(), UNP_LOCK_ASSERT() macros. - Acquire unp_mtx on entering most UNIX domain socket code, drop/re-acquire around calls into VFS, and release it on return. - Avoid performing sodupsockaddr() while holding the mutex, so in general move to allocating storage before acquiring the mutex to copy the data. - Make a stack copy of the xucred rather than copying out while holding unp_mtx. Copy the peer credential out after releasing the mutex. - Add additional assertions of vnode locks following VOP_CREATE(). A few notes: - Use of an sx lock for the file list mutex may cause problems with regard to unp_mtx when garbage collection passed file descriptors. - The locking in unp_pcblist() for sysctl monitoring is correct subject to the unpcb zone not returning memory for reuse by other subsystems (consistent with similar existing concerns). - Sam's version of this change, as with the BSD/OS version, made use of both a global lock and per-unpcb locks. However, in practice, the global lock covered all accesses, so I have simplified out the unpcb locks in the interest of getting this merged faster (reducing the overhead but not sacrificing granularity in most cases). We will want to explore possibilities for improving lock granularity in this code in the future. Submitted by: sam Sponsored by: FreeBSD Foundatiuon Obtained from: BSD/OS 5 snapshot provided by BSDi
2004-06-10 21:34:38 +00:00
return (error);
The long-awaited mega-massive-network-code- cleanup. Part I. This commit includes the following changes: 1) Old-style (pr_usrreq()) protocols are no longer supported, the compatibility glue for them is deleted, and the kernel will panic on boot if any are compiled in. 2) Certain protocol entry points are modified to take a process structure, so they they can easily tell whether or not it is possible to sleep, and also to access credentials. 3) SS_PRIV is no more, and with it goes the SO_PRIVSTATE setsockopt() call. Protocols should use the process pointer they are now passed. 4) The PF_LOCAL and PF_ROUTE families have been updated to use the new style, as has the `raw' skeleton family. 5) PF_LOCAL sockets now obey the process's umask when creating a socket in the filesystem. As a result, LINT is now broken. I'm hoping that some enterprising hacker with a bit more time will either make the broken bits work (should be easy for netipx) or dike them out.
1997-04-27 20:01:29 +00:00
}
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
Chance protocol switch method pru_detach() so that it returns void rather than an error. Detaches do not "fail", they other occur or the protocol flags SS_PROTOREF to take ownership of the socket. soclose() no longer looks at so_pcb to see if it's NULL, relying entirely on the protocol to decide whether it's time to free the socket or not using SS_PROTOREF. so_pcb is now entirely owned and managed by the protocol code. Likewise, no longer test so_pcb in other socket functions, such as soreceive(), which have no business digging into protocol internals. Protocol detach routines no longer try to free the socket on detach, this is performed in the socket code if the protocol permits it. In rts_detach(), no longer test for rp != NULL in detach, and likewise in other protocols that don't permit a NULL so_pcb, reduce the incidence of testing for it during detach. netinet and netinet6 are not fully updated to this change, which will be in an upcoming commit. In their current state they may leak memory or panic. MFC after: 3 months
2006-04-01 15:42:02 +00:00
static void
The long-awaited mega-massive-network-code- cleanup. Part I. This commit includes the following changes: 1) Old-style (pr_usrreq()) protocols are no longer supported, the compatibility glue for them is deleted, and the kernel will panic on boot if any are compiled in. 2) Certain protocol entry points are modified to take a process structure, so they they can easily tell whether or not it is possible to sleep, and also to access credentials. 3) SS_PRIV is no more, and with it goes the SO_PRIVSTATE setsockopt() call. Protocols should use the process pointer they are now passed. 4) The PF_LOCAL and PF_ROUTE families have been updated to use the new style, as has the `raw' skeleton family. 5) PF_LOCAL sockets now obey the process's umask when creating a socket in the filesystem. As a result, LINT is now broken. I'm hoping that some enterprising hacker with a bit more time will either make the broken bits work (should be easy for netipx) or dike them out.
1997-04-27 20:01:29 +00:00
uipc_detach(struct socket *so)
{
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
struct unpcb *unp, *unp2;
- Close a race between enumerating UNIX domain socket pcb structures via sysctl and socket teardown by adding a reference count to the UNIX domain pcb object and fixing the sysctl that enumerates unpcbs to grab a reference on each unpcb while it builds the list to copy out to userland. - Close a race between UNIX domain pcb garbage collection (unp_gc()) and file descriptor teardown (fdrop()) by adding a new garbage collection flag FWAIT. unp_gc() sets FWAIT while it walks the message buffers in a UNIX domain socket looking for nested file descriptor references and clears the flag when it is finished. fdrop() checks to see if the flag is set on a file descriptor whose refcount just dropped to 0 and waits for unp_gc() to clear the flag before completely destroying the file descriptor. MFC after: 1 week Reviewed by: rwatson Submitted by: ups Hopefully makes the panics go away: mx1
2007-01-05 19:59:46 +00:00
struct sockaddr_un *saved_unp_addr;
Since unp_attach() and unp_detach() are now called only from uipc_attach() and uipc_detach(), merge them into their calling functions.
2006-07-23 10:25:28 +00:00
struct vnode *vp;
- Close a race between enumerating UNIX domain socket pcb structures via sysctl and socket teardown by adding a reference count to the UNIX domain pcb object and fixing the sysctl that enumerates unpcbs to grab a reference on each unpcb while it builds the list to copy out to userland. - Close a race between UNIX domain pcb garbage collection (unp_gc()) and file descriptor teardown (fdrop()) by adding a new garbage collection flag FWAIT. unp_gc() sets FWAIT while it walks the message buffers in a UNIX domain socket looking for nested file descriptor references and clears the flag when it is finished. fdrop() checks to see if the flag is set on a file descriptor whose refcount just dropped to 0 and waits for unp_gc() to clear the flag before completely destroying the file descriptor. MFC after: 1 week Reviewed by: rwatson Submitted by: ups Hopefully makes the panics go away: mx1
2007-01-05 19:59:46 +00:00
int freeunp, local_unp_rights;
The long-awaited mega-massive-network-code- cleanup. Part I. This commit includes the following changes: 1) Old-style (pr_usrreq()) protocols are no longer supported, the compatibility glue for them is deleted, and the kernel will panic on boot if any are compiled in. 2) Certain protocol entry points are modified to take a process structure, so they they can easily tell whether or not it is possible to sleep, and also to access credentials. 3) SS_PRIV is no more, and with it goes the SO_PRIVSTATE setsockopt() call. Protocols should use the process pointer they are now passed. 4) The PF_LOCAL and PF_ROUTE families have been updated to use the new style, as has the `raw' skeleton family. 5) PF_LOCAL sockets now obey the process's umask when creating a socket in the filesystem. As a result, LINT is now broken. I'm hoping that some enterprising hacker with a bit more time will either make the broken bits work (should be easy for netipx) or dike them out.
1997-04-27 20:01:29 +00:00
Always acquire the UNIX domain socket subsystem lock (UNP lock) before dereferencing sotounpcb() and checking its value, as so_pcb is protected by protocol locking, not subsystem locking. This prevents races during close() by one thread and use of ths socket in another. unp_bind() now assert the UNP lock, and uipc_bind() now acquires the lock around calls to unp_bind().
2004-08-16 04:41:03 +00:00
unp = sotounpcb(so);
Modify UNIX domain sockets to guarantee, and assume, that so_pcb is always defined for an in-use socket. This allows us to eliminate countless tests of whether so_pcb is non-NULL, eliminating dozens of error cases. For now, retain the call to sotryfree() in the uipc_abort() path, but this will eventually move to soabort(). These new assumptions should be largely correct, and will become more so as the socket/pcb reference model is fixed. Removing the notion that so_pcb can be non-NULL is a critical step towards further fine-graining of the UNIX domain socket locking, as the so_pcb reference no longer needs to be protected using locks, instead it is a property of the socket life cycle.
2006-03-17 13:52:57 +00:00
KASSERT(unp != NULL, ("uipc_detach: unp == NULL"));
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
Handle races with listening socket close when connecting a unix socket. If the listening socket is closed while sonewconn() is executing, the nascent child socket is aborted, which results in recursion on the unp_link lock when the child's pru_detach method is invoked. Fix this by using a flag to mark such sockets, and skip a part of the socket's teardown during detach. Reported by: Raviprakash Darbha <rdarbha@juniper.net> Tested by: pho MFC after: 2 weeks Differential Revision: https://reviews.freebsd.org/D7398
2016-08-08 20:25:04 +00:00
vp = NULL;
local_unp_rights = 0;
Listening sockets improvements. o Separate fields of struct socket that belong to listening from fields that belong to normal dataflow, and unionize them. This shrinks the structure a bit. - Take out selinfo's from the socket buffers into the socket. The first reason is to support braindamaged scenario when a socket is added to kevent(2) and then listen(2) is cast on it. The second reason is that there is future plan to make socket buffers pluggable, so that for a dataflow socket a socket buffer can be changed, and in this case we also want to keep same selinfos through the lifetime of a socket. - Remove struct struct so_accf. Since now listening stuff no longer affects struct socket size, just move its fields into listening part of the union. - Provide sol_upcall field and enforce that so_upcall_set() may be called only on a dataflow socket, which has buffers, and for listening sockets provide solisten_upcall_set(). o Remove ACCEPT_LOCK() global. - Add a mutex to socket, to be used instead of socket buffer lock to lock fields of struct socket that don't belong to a socket buffer. - Allow to acquire two socket locks, but the first one must belong to a listening socket. - Make soref()/sorele() to use atomic(9). This allows in some situations to do soref() without owning socket lock. There is place for improvement here, it is possible to make sorele() also to lock optionally. - Most protocols aren't touched by this change, except UNIX local sockets. See below for more information. o Reduce copy-and-paste in kernel modules that accept connections from listening sockets: provide function solisten_dequeue(), and use it in the following modules: ctl(4), iscsi(4), ng_btsocket(4), ng_ksocket(4), infiniband, rpc. o UNIX local sockets. - Removal of ACCEPT_LOCK() global uncovered several races in the UNIX local sockets. Most races exist around spawning a new socket, when we are connecting to a local listening socket. To cover them, we need to hold locks on both PCBs when spawning a third one. This means holding them across sonewconn(). This creates a LOR between pcb locks and unp_list_lock. - To fix the new LOR, abandon the global unp_list_lock in favor of global unp_link_lock. Indeed, separating these two locks didn't provide us any extra parralelism in the UNIX sockets. - Now call into uipc_attach() may happen with unp_link_lock hold if, we are accepting, or without unp_link_lock in case if we are just creating a socket. - Another problem in UNIX sockets is that uipc_close() basicly did nothing for a listening socket. The vnode remained opened for connections. This is fixed by removing vnode in uipc_close(). Maybe the right way would be to do it for all sockets (not only listening), simply move the vnode teardown from uipc_detach() to uipc_close()? Sponsored by: Netflix Differential Revision: https://reviews.freebsd.org/D9770
2017-06-08 21:30:34 +00:00
UNP_LINK_WLOCK();
Since unp_attach() and unp_detach() are now called only from uipc_attach() and uipc_detach(), merge them into their calling functions.
2006-07-23 10:25:28 +00:00
LIST_REMOVE(unp, unp_link);
unp->unp_gencnt = ++unp_gencnt;
--unp_count;
Handle races with listening socket close when connecting a unix socket. If the listening socket is closed while sonewconn() is executing, the nascent child socket is aborted, which results in recursion on the unp_link lock when the child's pru_detach method is invoked. Fix this by using a flag to mark such sockets, and skip a part of the socket's teardown during detach. Reported by: Raviprakash Darbha <rdarbha@juniper.net> Tested by: pho MFC after: 2 weeks Differential Revision: https://reviews.freebsd.org/D7398
2016-08-08 20:25:04 +00:00
UNP_PCB_LOCK(unp);
Listening sockets improvements. o Separate fields of struct socket that belong to listening from fields that belong to normal dataflow, and unionize them. This shrinks the structure a bit. - Take out selinfo's from the socket buffers into the socket. The first reason is to support braindamaged scenario when a socket is added to kevent(2) and then listen(2) is cast on it. The second reason is that there is future plan to make socket buffers pluggable, so that for a dataflow socket a socket buffer can be changed, and in this case we also want to keep same selinfos through the lifetime of a socket. - Remove struct struct so_accf. Since now listening stuff no longer affects struct socket size, just move its fields into listening part of the union. - Provide sol_upcall field and enforce that so_upcall_set() may be called only on a dataflow socket, which has buffers, and for listening sockets provide solisten_upcall_set(). o Remove ACCEPT_LOCK() global. - Add a mutex to socket, to be used instead of socket buffer lock to lock fields of struct socket that don't belong to a socket buffer. - Allow to acquire two socket locks, but the first one must belong to a listening socket. - Make soref()/sorele() to use atomic(9). This allows in some situations to do soref() without owning socket lock. There is place for improvement here, it is possible to make sorele() also to lock optionally. - Most protocols aren't touched by this change, except UNIX local sockets. See below for more information. o Reduce copy-and-paste in kernel modules that accept connections from listening sockets: provide function solisten_dequeue(), and use it in the following modules: ctl(4), iscsi(4), ng_btsocket(4), ng_ksocket(4), infiniband, rpc. o UNIX local sockets. - Removal of ACCEPT_LOCK() global uncovered several races in the UNIX local sockets. Most races exist around spawning a new socket, when we are connecting to a local listening socket. To cover them, we need to hold locks on both PCBs when spawning a third one. This means holding them across sonewconn(). This creates a LOR between pcb locks and unp_list_lock. - To fix the new LOR, abandon the global unp_list_lock in favor of global unp_link_lock. Indeed, separating these two locks didn't provide us any extra parralelism in the UNIX sockets. - Now call into uipc_attach() may happen with unp_link_lock hold if, we are accepting, or without unp_link_lock in case if we are just creating a socket. - Another problem in UNIX sockets is that uipc_close() basicly did nothing for a listening socket. The vnode remained opened for connections. This is fixed by removing vnode in uipc_close(). Maybe the right way would be to do it for all sockets (not only listening), simply move the vnode teardown from uipc_detach() to uipc_close()? Sponsored by: Netflix Differential Revision: https://reviews.freebsd.org/D9770
2017-06-08 21:30:34 +00:00
if ((unp->unp_flags & UNP_NASCENT) != 0)
goto teardown;
Handle races with listening socket close when connecting a unix socket. If the listening socket is closed while sonewconn() is executing, the nascent child socket is aborted, which results in recursion on the unp_link lock when the child's pru_detach method is invoked. Fix this by using a flag to mark such sockets, and skip a part of the socket's teardown during detach. Reported by: Raviprakash Darbha <rdarbha@juniper.net> Tested by: pho MFC after: 2 weeks Differential Revision: https://reviews.freebsd.org/D7398
2016-08-08 20:25:04 +00:00
Since unp_attach() and unp_detach() are now called only from uipc_attach() and uipc_detach(), merge them into their calling functions.
2006-07-23 10:25:28 +00:00
if ((vp = unp->unp_vnode) != NULL) {
Introduce VOP_UNP_BIND(), VOP_UNP_CONNECT(), and VOP_UNP_DETACH() operations for setting and accessing vnode's v_socket field. The operations are necessary to implement proper unix socket handling on layered file systems like nullfs(5). This change fixes the long standing issue with nullfs(5) being in that unix sockets did not work between lower and upper layers: if we bound to a socket on the lower layer we could connect only to the lower path; if we bound to the upper layer we could connect only to the upper path. The new behavior is one can connect to both the lower and the upper paths regardless what layer path one binds to. PR: kern/51583, kern/159663 Suggested by: kib Reviewed by: arch MFC after: 2 weeks
2012-02-29 21:38:31 +00:00
VOP_UNP_DETACH(vp);
Since unp_attach() and unp_detach() are now called only from uipc_attach() and uipc_detach(), merge them into their calling functions.
2006-07-23 10:25:28 +00:00
unp->unp_vnode = NULL;
}
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
unp2 = unp->unp_conn;
if (unp2 != NULL) {
UNP_PCB_LOCK(unp2);
unp_disconnect(unp, unp2);
UNP_PCB_UNLOCK(unp2);
}
/*
Decompose the global UNIX domain sockets rwlock into two different locks: a global list/counter/generation counter protected by a new mutex unp_list_lock, and a global linkage rwlock, unp_global_rwlock, which protects the connections between UNIX domain sockets. This eliminates conditional lock acquisition that was previously a property of the global lock being held over sonewconn() leading to a call to uipc_attach(), which also required the global lock, but couldn't rely on it as other paths existed to uipc_attach() that didn't hold it: now uipc_attach() uses only the list lock, which follows the linkage lock in the lock order. It may also reduce contention on the global lock for some workloads. Add global UNIX domain socket locks to hard-coded witness lock order. MFC after: 1 week Discussed with: kris
2009-03-08 21:48:29 +00:00
* We hold the linkage lock exclusively, so it's OK to acquire
Remove or update several stale comments. A bit of whitespace/style cleanup. Update copyright. MFC after: 3 days (applicable changes)
2008-10-03 09:01:55 +00:00
* multiple pcb locks at a time.
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
*/
Since unp_attach() and unp_detach() are now called only from uipc_attach() and uipc_detach(), merge them into their calling functions.
2006-07-23 10:25:28 +00:00
while (!LIST_EMPTY(&unp->unp_refs)) {
struct unpcb *ref = LIST_FIRST(&unp->unp_refs);
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
UNP_PCB_LOCK(ref);
Remove the errno argument from unp_drop(). While there, add a comment to clarify that ECONNRESET should always be returned for POSIX conformance. Suggested by: Steven Hartland
2016-02-26 12:46:34 +00:00
unp_drop(ref);
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
UNP_PCB_UNLOCK(ref);
Since unp_attach() and unp_detach() are now called only from uipc_attach() and uipc_detach(), merge them into their calling functions.
2006-07-23 10:25:28 +00:00
}
Remove explicit locking of struct file. - Introduce a finit() which is used to initailize the fields of struct file in such a way that the ops vector is only valid after the data, type, and flags are valid. - Protect f_flag and f_count with atomic operations. - Remove the global list of all files and associated accounting. - Rewrite the unp garbage collection such that it no longer requires the global list of all files and instead uses a list of all unp sockets. - Mark sockets in the accept queue so we don't incorrectly gc them. Tested by: kris, pho
2007-12-30 01:42:15 +00:00
local_unp_rights = unp_rights;
Handle races with listening socket close when connecting a unix socket. If the listening socket is closed while sonewconn() is executing, the nascent child socket is aborted, which results in recursion on the unp_link lock when the child's pru_detach method is invoked. Fix this by using a flag to mark such sockets, and skip a part of the socket's teardown during detach. Reported by: Raviprakash Darbha <rdarbha@juniper.net> Tested by: pho MFC after: 2 weeks Differential Revision: https://reviews.freebsd.org/D7398
2016-08-08 20:25:04 +00:00
teardown:
Listening sockets improvements. o Separate fields of struct socket that belong to listening from fields that belong to normal dataflow, and unionize them. This shrinks the structure a bit. - Take out selinfo's from the socket buffers into the socket. The first reason is to support braindamaged scenario when a socket is added to kevent(2) and then listen(2) is cast on it. The second reason is that there is future plan to make socket buffers pluggable, so that for a dataflow socket a socket buffer can be changed, and in this case we also want to keep same selinfos through the lifetime of a socket. - Remove struct struct so_accf. Since now listening stuff no longer affects struct socket size, just move its fields into listening part of the union. - Provide sol_upcall field and enforce that so_upcall_set() may be called only on a dataflow socket, which has buffers, and for listening sockets provide solisten_upcall_set(). o Remove ACCEPT_LOCK() global. - Add a mutex to socket, to be used instead of socket buffer lock to lock fields of struct socket that don't belong to a socket buffer. - Allow to acquire two socket locks, but the first one must belong to a listening socket. - Make soref()/sorele() to use atomic(9). This allows in some situations to do soref() without owning socket lock. There is place for improvement here, it is possible to make sorele() also to lock optionally. - Most protocols aren't touched by this change, except UNIX local sockets. See below for more information. o Reduce copy-and-paste in kernel modules that accept connections from listening sockets: provide function solisten_dequeue(), and use it in the following modules: ctl(4), iscsi(4), ng_btsocket(4), ng_ksocket(4), infiniband, rpc. o UNIX local sockets. - Removal of ACCEPT_LOCK() global uncovered several races in the UNIX local sockets. Most races exist around spawning a new socket, when we are connecting to a local listening socket. To cover them, we need to hold locks on both PCBs when spawning a third one. This means holding them across sonewconn(). This creates a LOR between pcb locks and unp_list_lock. - To fix the new LOR, abandon the global unp_list_lock in favor of global unp_link_lock. Indeed, separating these two locks didn't provide us any extra parralelism in the UNIX sockets. - Now call into uipc_attach() may happen with unp_link_lock hold if, we are accepting, or without unp_link_lock in case if we are just creating a socket. - Another problem in UNIX sockets is that uipc_close() basicly did nothing for a listening socket. The vnode remained opened for connections. This is fixed by removing vnode in uipc_close(). Maybe the right way would be to do it for all sockets (not only listening), simply move the vnode teardown from uipc_detach() to uipc_close()? Sponsored by: Netflix Differential Revision: https://reviews.freebsd.org/D9770
2017-06-08 21:30:34 +00:00
UNP_LINK_WUNLOCK();
Since unp_attach() and unp_detach() are now called only from uipc_attach() and uipc_detach(), merge them into their calling functions.
2006-07-23 10:25:28 +00:00
unp->unp_socket->so_pcb = NULL;
- Close a race between enumerating UNIX domain socket pcb structures via sysctl and socket teardown by adding a reference count to the UNIX domain pcb object and fixing the sysctl that enumerates unpcbs to grab a reference on each unpcb while it builds the list to copy out to userland. - Close a race between UNIX domain pcb garbage collection (unp_gc()) and file descriptor teardown (fdrop()) by adding a new garbage collection flag FWAIT. unp_gc() sets FWAIT while it walks the message buffers in a UNIX domain socket looking for nested file descriptor references and clears the flag when it is finished. fdrop() checks to see if the flag is set on a file descriptor whose refcount just dropped to 0 and waits for unp_gc() to clear the flag before completely destroying the file descriptor. MFC after: 1 week Reviewed by: rwatson Submitted by: ups Hopefully makes the panics go away: mx1
2007-01-05 19:59:46 +00:00
saved_unp_addr = unp->unp_addr;
unp->unp_addr = NULL;
unp->unp_refcount--;
freeunp = (unp->unp_refcount == 0);
if (saved_unp_addr != NULL)
Retire the MALLOC and FREE macros. They are an abomination unto style(9). MFC after: 3 months
2008-10-23 15:53:51 +00:00
free(saved_unp_addr, M_SONAME);
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
if (freeunp) {
UNP_PCB_LOCK_DESTROY(unp);
- Close a race between enumerating UNIX domain socket pcb structures via sysctl and socket teardown by adding a reference count to the UNIX domain pcb object and fixing the sysctl that enumerates unpcbs to grab a reference on each unpcb while it builds the list to copy out to userland. - Close a race between UNIX domain pcb garbage collection (unp_gc()) and file descriptor teardown (fdrop()) by adding a new garbage collection flag FWAIT. unp_gc() sets FWAIT while it walks the message buffers in a UNIX domain socket looking for nested file descriptor references and clears the flag when it is finished. fdrop() checks to see if the flag is set on a file descriptor whose refcount just dropped to 0 and waits for unp_gc() to clear the flag before completely destroying the file descriptor. MFC after: 1 week Reviewed by: rwatson Submitted by: ups Hopefully makes the panics go away: mx1
2007-01-05 19:59:46 +00:00
uma_zfree(unp_zone, unp);
In uipc_close(), we no longer always free the unpcb, as the last reference may be dropped later. In this case, always unlock the unpcb so as not to leak the lock. Found by: kris (BugMagnet)
2007-03-12 14:52:00 +00:00
} else
UNP_PCB_UNLOCK(unp);
Remove the support for using non-mpsafe filesystem modules. In particular, do not lock Giant conditionally when calling into the filesystem module, remove the VFS_LOCK_GIANT() and related macros. Stop handling buffers belonging to non-mpsafe filesystems. The VFS_VERSION is bumped to indicate the interface change which does not result in the interface signatures changes. Conducted and reviewed by: attilio Tested by: pho
2012-10-22 17:50:54 +00:00
if (vp)
Since unp_attach() and unp_detach() are now called only from uipc_attach() and uipc_detach(), merge them into their calling functions.
2006-07-23 10:25:28 +00:00
vrele(vp);
if (local_unp_rights)
Schedule garbage collection run for the in-flight rights passed over the unix domain sockets to the next tick, coalescing the serial calls until the collection fires. The thought is that more work for the collector could arise in the near time, allowing to clean more and not spend too much CPU on repeated collection when there is no garbage. Currently the collection task is fired immediately upon unix domain socket close if there are any rights in flight, which caused excessive CPU usage and too long blocking of the threads waiting for unp_list_lock and unp_link_rwlock in write mode. Robert noted that it would be nice if we could find some heuristic by which we decide whether to run GC a bit more quickly. E.g., if the number of UNIX domain sockets is close to its resource limit, but not quite. Reported and tested by: Markus Gebert <markus.gebert@hostpoint.ch> Reviewed by: rwatson MFC after: 2 weeks
2012-11-20 15:45:48 +00:00
taskqueue_enqueue_timeout(taskqueue_thread, &unp_gc_task, -1);
The long-awaited mega-massive-network-code- cleanup. Part I. This commit includes the following changes: 1) Old-style (pr_usrreq()) protocols are no longer supported, the compatibility glue for them is deleted, and the kernel will panic on boot if any are compiled in. 2) Certain protocol entry points are modified to take a process structure, so they they can easily tell whether or not it is possible to sleep, and also to access credentials. 3) SS_PRIV is no more, and with it goes the SO_PRIVSTATE setsockopt() call. Protocols should use the process pointer they are now passed. 4) The PF_LOCAL and PF_ROUTE families have been updated to use the new style, as has the `raw' skeleton family. 5) PF_LOCAL sockets now obey the process's umask when creating a socket in the filesystem. As a result, LINT is now broken. I'm hoping that some enterprising hacker with a bit more time will either make the broken bits work (should be easy for netipx) or dike them out.
1997-04-27 20:01:29 +00:00
}
static int
uipc_disconnect(struct socket *so)
{
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
struct unpcb *unp, *unp2;
The long-awaited mega-massive-network-code- cleanup. Part I. This commit includes the following changes: 1) Old-style (pr_usrreq()) protocols are no longer supported, the compatibility glue for them is deleted, and the kernel will panic on boot if any are compiled in. 2) Certain protocol entry points are modified to take a process structure, so they they can easily tell whether or not it is possible to sleep, and also to access credentials. 3) SS_PRIV is no more, and with it goes the SO_PRIVSTATE setsockopt() call. Protocols should use the process pointer they are now passed. 4) The PF_LOCAL and PF_ROUTE families have been updated to use the new style, as has the `raw' skeleton family. 5) PF_LOCAL sockets now obey the process's umask when creating a socket in the filesystem. As a result, LINT is now broken. I'm hoping that some enterprising hacker with a bit more time will either make the broken bits work (should be easy for netipx) or dike them out.
1997-04-27 20:01:29 +00:00
Always acquire the UNIX domain socket subsystem lock (UNP lock) before dereferencing sotounpcb() and checking its value, as so_pcb is protected by protocol locking, not subsystem locking. This prevents races during close() by one thread and use of ths socket in another. unp_bind() now assert the UNP lock, and uipc_bind() now acquires the lock around calls to unp_bind().
2004-08-16 04:41:03 +00:00
unp = sotounpcb(so);
Modify UNIX domain sockets to guarantee, and assume, that so_pcb is always defined for an in-use socket. This allows us to eliminate countless tests of whether so_pcb is non-NULL, eliminating dozens of error cases. For now, retain the call to sotryfree() in the uipc_abort() path, but this will eventually move to soabort(). These new assumptions should be largely correct, and will become more so as the socket/pcb reference model is fixed. Removing the notion that so_pcb can be non-NULL is a critical step towards further fine-graining of the UNIX domain socket locking, as the so_pcb reference no longer needs to be protected using locks, instead it is a property of the socket life cycle.
2006-03-17 13:52:57 +00:00
KASSERT(unp != NULL, ("uipc_disconnect: unp == NULL"));
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
Decompose the global UNIX domain sockets rwlock into two different locks: a global list/counter/generation counter protected by a new mutex unp_list_lock, and a global linkage rwlock, unp_global_rwlock, which protects the connections between UNIX domain sockets. This eliminates conditional lock acquisition that was previously a property of the global lock being held over sonewconn() leading to a call to uipc_attach(), which also required the global lock, but couldn't rely on it as other paths existed to uipc_attach() that didn't hold it: now uipc_attach() uses only the list lock, which follows the linkage lock in the lock order. It may also reduce contention on the global lock for some workloads. Add global UNIX domain socket locks to hard-coded witness lock order. MFC after: 1 week Discussed with: kris
2009-03-08 21:48:29 +00:00
UNP_LINK_WLOCK();
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
UNP_PCB_LOCK(unp);
unp2 = unp->unp_conn;
if (unp2 != NULL) {
UNP_PCB_LOCK(unp2);
unp_disconnect(unp, unp2);
UNP_PCB_UNLOCK(unp2);
}
UNP_PCB_UNLOCK(unp);
Decompose the global UNIX domain sockets rwlock into two different locks: a global list/counter/generation counter protected by a new mutex unp_list_lock, and a global linkage rwlock, unp_global_rwlock, which protects the connections between UNIX domain sockets. This eliminates conditional lock acquisition that was previously a property of the global lock being held over sonewconn() leading to a call to uipc_attach(), which also required the global lock, but couldn't rely on it as other paths existed to uipc_attach() that didn't hold it: now uipc_attach() uses only the list lock, which follows the linkage lock in the lock order. It may also reduce contention on the global lock for some workloads. Add global UNIX domain socket locks to hard-coded witness lock order. MFC after: 1 week Discussed with: kris
2009-03-08 21:48:29 +00:00
UNP_LINK_WUNLOCK();
Mechanical whitespace cleanup; parenthesize return values; other minor style nits.
2004-01-11 19:48:19 +00:00
return (0);
The long-awaited mega-massive-network-code- cleanup. Part I. This commit includes the following changes: 1) Old-style (pr_usrreq()) protocols are no longer supported, the compatibility glue for them is deleted, and the kernel will panic on boot if any are compiled in. 2) Certain protocol entry points are modified to take a process structure, so they they can easily tell whether or not it is possible to sleep, and also to access credentials. 3) SS_PRIV is no more, and with it goes the SO_PRIVSTATE setsockopt() call. Protocols should use the process pointer they are now passed. 4) The PF_LOCAL and PF_ROUTE families have been updated to use the new style, as has the `raw' skeleton family. 5) PF_LOCAL sockets now obey the process's umask when creating a socket in the filesystem. As a result, LINT is now broken. I'm hoping that some enterprising hacker with a bit more time will either make the broken bits work (should be easy for netipx) or dike them out.
1997-04-27 20:01:29 +00:00
}
static int
Push the assignment of a new or updated so_qlimit from solisten() following the protocol pru_listen() call to solisten_proto(), so that it occurs under the socket lock acquisition that also sets SO_ACCEPTCONN. This requires passing the new backlog parameter to the protocol, which also allows the protocol to be aware of changes in queue limit should it wish to do something about the new queue limit. This continues a move towards the socket layer acting as a library for the protocol. Bump __FreeBSD_version due to a change in the in-kernel protocol interface. This change has been tested with IPv4 and UNIX domain sockets, but not other protocols.
2005-10-30 19:44:40 +00:00
uipc_listen(struct socket *so, int backlog, struct thread *td)
The long-awaited mega-massive-network-code- cleanup. Part I. This commit includes the following changes: 1) Old-style (pr_usrreq()) protocols are no longer supported, the compatibility glue for them is deleted, and the kernel will panic on boot if any are compiled in. 2) Certain protocol entry points are modified to take a process structure, so they they can easily tell whether or not it is possible to sleep, and also to access credentials. 3) SS_PRIV is no more, and with it goes the SO_PRIVSTATE setsockopt() call. Protocols should use the process pointer they are now passed. 4) The PF_LOCAL and PF_ROUTE families have been updated to use the new style, as has the `raw' skeleton family. 5) PF_LOCAL sockets now obey the process's umask when creating a socket in the filesystem. As a result, LINT is now broken. I'm hoping that some enterprising hacker with a bit more time will either make the broken bits work (should be easy for netipx) or dike them out.
1997-04-27 20:01:29 +00:00
{
Always acquire the UNIX domain socket subsystem lock (UNP lock) before dereferencing sotounpcb() and checking its value, as so_pcb is protected by protocol locking, not subsystem locking. This prevents races during close() by one thread and use of ths socket in another. unp_bind() now assert the UNP lock, and uipc_bind() now acquires the lock around calls to unp_bind().
2004-08-16 04:41:03 +00:00
struct unpcb *unp;
Introduce a subsystem lock around UNIX domain sockets in order to protect global and allocated variables. This strategy is derived from work originally developed by BSDi for BSD/OS, and applied to FreeBSD by Sam Leffler: - Add unp_mtx, a global mutex which will protect all UNIX domain socket related variables, structures, etc. - Add UNP_LOCK(), UNP_UNLOCK(), UNP_LOCK_ASSERT() macros. - Acquire unp_mtx on entering most UNIX domain socket code, drop/re-acquire around calls into VFS, and release it on return. - Avoid performing sodupsockaddr() while holding the mutex, so in general move to allocating storage before acquiring the mutex to copy the data. - Make a stack copy of the xucred rather than copying out while holding unp_mtx. Copy the peer credential out after releasing the mutex. - Add additional assertions of vnode locks following VOP_CREATE(). A few notes: - Use of an sx lock for the file list mutex may cause problems with regard to unp_mtx when garbage collection passed file descriptors. - The locking in unp_pcblist() for sysctl monitoring is correct subject to the unpcb zone not returning memory for reuse by other subsystems (consistent with similar existing concerns). - Sam's version of this change, as with the BSD/OS version, made use of both a global lock and per-unpcb locks. However, in practice, the global lock covered all accesses, so I have simplified out the unpcb locks in the interest of getting this merged faster (reducing the overhead but not sacrificing granularity in most cases). We will want to explore possibilities for improving lock granularity in this code in the future. Submitted by: sam Sponsored by: FreeBSD Foundatiuon Obtained from: BSD/OS 5 snapshot provided by BSDi
2004-06-10 21:34:38 +00:00
int error;
The long-awaited mega-massive-network-code- cleanup. Part I. This commit includes the following changes: 1) Old-style (pr_usrreq()) protocols are no longer supported, the compatibility glue for them is deleted, and the kernel will panic on boot if any are compiled in. 2) Certain protocol entry points are modified to take a process structure, so they they can easily tell whether or not it is possible to sleep, and also to access credentials. 3) SS_PRIV is no more, and with it goes the SO_PRIVSTATE setsockopt() call. Protocols should use the process pointer they are now passed. 4) The PF_LOCAL and PF_ROUTE families have been updated to use the new style, as has the `raw' skeleton family. 5) PF_LOCAL sockets now obey the process's umask when creating a socket in the filesystem. As a result, LINT is now broken. I'm hoping that some enterprising hacker with a bit more time will either make the broken bits work (should be easy for netipx) or dike them out.
1997-04-27 20:01:29 +00:00
For non-listening AF_UNIX sockets return error code EOPNOTSUPP to match documentation and SUS.
2017-01-25 22:26:45 +00:00
if (so->so_type != SOCK_STREAM && so->so_type != SOCK_SEQPACKET)
return (EOPNOTSUPP);
Always acquire the UNIX domain socket subsystem lock (UNP lock) before dereferencing sotounpcb() and checking its value, as so_pcb is protected by protocol locking, not subsystem locking. This prevents races during close() by one thread and use of ths socket in another. unp_bind() now assert the UNP lock, and uipc_bind() now acquires the lock around calls to unp_bind().
2004-08-16 04:41:03 +00:00
unp = sotounpcb(so);
Modify UNIX domain sockets to guarantee, and assume, that so_pcb is always defined for an in-use socket. This allows us to eliminate countless tests of whether so_pcb is non-NULL, eliminating dozens of error cases. For now, retain the call to sotryfree() in the uipc_abort() path, but this will eventually move to soabort(). These new assumptions should be largely correct, and will become more so as the socket/pcb reference model is fixed. Removing the notion that so_pcb can be non-NULL is a critical step towards further fine-graining of the UNIX domain socket locking, as the so_pcb reference no longer needs to be protected using locks, instead it is a property of the socket life cycle.
2006-03-17 13:52:57 +00:00
KASSERT(unp != NULL, ("uipc_listen: unp == NULL"));
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
UNP_PCB_LOCK(unp);
Modify UNIX domain sockets to guarantee, and assume, that so_pcb is always defined for an in-use socket. This allows us to eliminate countless tests of whether so_pcb is non-NULL, eliminating dozens of error cases. For now, retain the call to sotryfree() in the uipc_abort() path, but this will eventually move to soabort(). These new assumptions should be largely correct, and will become more so as the socket/pcb reference model is fixed. Removing the notion that so_pcb can be non-NULL is a critical step towards further fine-graining of the UNIX domain socket locking, as the so_pcb reference no longer needs to be protected using locks, instead it is a property of the socket life cycle.
2006-03-17 13:52:57 +00:00
if (unp->unp_vnode == NULL) {
Let listen() return EDESTADDRREQ when not bound. We currently return EINVAL when calling listen() on a UNIX socket that has not been bound to a pathname. If my interpretation of POSIX is correct, we should return EDESTADDRREQ: "The socket is not bound to a local address, and the protocol does not support listening on an unbound socket." Return EDESTADDRREQ instead when not bound and not connected. Differential Revision: https://reviews.freebsd.org/D3038 Reviewed by: gnn, network
2015-07-10 06:47:14 +00:00
/* Already connected or not bound to an address. */
error = unp->unp_conn != NULL ? EINVAL : EDESTADDRREQ;
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
UNP_PCB_UNLOCK(unp);
Let listen() return EDESTADDRREQ when not bound. We currently return EINVAL when calling listen() on a UNIX socket that has not been bound to a pathname. If my interpretation of POSIX is correct, we should return EDESTADDRREQ: "The socket is not bound to a local address, and the protocol does not support listening on an unbound socket." Return EDESTADDRREQ instead when not bound and not connected. Differential Revision: https://reviews.freebsd.org/D3038 Reviewed by: gnn, network
2015-07-10 06:47:14 +00:00
return (error);
Always acquire the UNIX domain socket subsystem lock (UNP lock) before dereferencing sotounpcb() and checking its value, as so_pcb is protected by protocol locking, not subsystem locking. This prevents races during close() by one thread and use of ths socket in another. unp_bind() now assert the UNP lock, and uipc_bind() now acquires the lock around calls to unp_bind().
2004-08-16 04:41:03 +00:00
}
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
SOCK_LOCK(so);
error = solisten_proto_check(so);
if (error == 0) {
cru2x(td->td_ucred, &unp->unp_peercred);
solisten_proto(so, backlog);
}
SOCK_UNLOCK(so);
UNP_PCB_UNLOCK(unp);
Introduce a subsystem lock around UNIX domain sockets in order to protect global and allocated variables. This strategy is derived from work originally developed by BSDi for BSD/OS, and applied to FreeBSD by Sam Leffler: - Add unp_mtx, a global mutex which will protect all UNIX domain socket related variables, structures, etc. - Add UNP_LOCK(), UNP_UNLOCK(), UNP_LOCK_ASSERT() macros. - Acquire unp_mtx on entering most UNIX domain socket code, drop/re-acquire around calls into VFS, and release it on return. - Avoid performing sodupsockaddr() while holding the mutex, so in general move to allocating storage before acquiring the mutex to copy the data. - Make a stack copy of the xucred rather than copying out while holding unp_mtx. Copy the peer credential out after releasing the mutex. - Add additional assertions of vnode locks following VOP_CREATE(). A few notes: - Use of an sx lock for the file list mutex may cause problems with regard to unp_mtx when garbage collection passed file descriptors. - The locking in unp_pcblist() for sysctl monitoring is correct subject to the unpcb zone not returning memory for reuse by other subsystems (consistent with similar existing concerns). - Sam's version of this change, as with the BSD/OS version, made use of both a global lock and per-unpcb locks. However, in practice, the global lock covered all accesses, so I have simplified out the unpcb locks in the interest of getting this merged faster (reducing the overhead but not sacrificing granularity in most cases). We will want to explore possibilities for improving lock granularity in this code in the future. Submitted by: sam Sponsored by: FreeBSD Foundatiuon Obtained from: BSD/OS 5 snapshot provided by BSDi
2004-06-10 21:34:38 +00:00
return (error);
The long-awaited mega-massive-network-code- cleanup. Part I. This commit includes the following changes: 1) Old-style (pr_usrreq()) protocols are no longer supported, the compatibility glue for them is deleted, and the kernel will panic on boot if any are compiled in. 2) Certain protocol entry points are modified to take a process structure, so they they can easily tell whether or not it is possible to sleep, and also to access credentials. 3) SS_PRIV is no more, and with it goes the SO_PRIVSTATE setsockopt() call. Protocols should use the process pointer they are now passed. 4) The PF_LOCAL and PF_ROUTE families have been updated to use the new style, as has the `raw' skeleton family. 5) PF_LOCAL sockets now obey the process's umask when creating a socket in the filesystem. As a result, LINT is now broken. I'm hoping that some enterprising hacker with a bit more time will either make the broken bits work (should be easy for netipx) or dike them out.
1997-04-27 20:01:29 +00:00
}
static int
Fix all areas of the system (or at least all those in LINT) to avoid storing socket addresses in mbufs. (Socket buffers are the one exception.) A number of kernel APIs needed to get fixed in order to make this happen. Also, fix three protocol families which kept PCBs in mbufs to not malloc them instead. Delete some old compatibility cruft while we're at it, and add some new routines in the in_cksum family.
1997-08-16 19:16:27 +00:00
uipc_peeraddr(struct socket *so, struct sockaddr **nam)
The long-awaited mega-massive-network-code- cleanup. Part I. This commit includes the following changes: 1) Old-style (pr_usrreq()) protocols are no longer supported, the compatibility glue for them is deleted, and the kernel will panic on boot if any are compiled in. 2) Certain protocol entry points are modified to take a process structure, so they they can easily tell whether or not it is possible to sleep, and also to access credentials. 3) SS_PRIV is no more, and with it goes the SO_PRIVSTATE setsockopt() call. Protocols should use the process pointer they are now passed. 4) The PF_LOCAL and PF_ROUTE families have been updated to use the new style, as has the `raw' skeleton family. 5) PF_LOCAL sockets now obey the process's umask when creating a socket in the filesystem. As a result, LINT is now broken. I'm hoping that some enterprising hacker with a bit more time will either make the broken bits work (should be easy for netipx) or dike them out.
1997-04-27 20:01:29 +00:00
{
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
struct unpcb *unp, *unp2;
Introduce a subsystem lock around UNIX domain sockets in order to protect global and allocated variables. This strategy is derived from work originally developed by BSDi for BSD/OS, and applied to FreeBSD by Sam Leffler: - Add unp_mtx, a global mutex which will protect all UNIX domain socket related variables, structures, etc. - Add UNP_LOCK(), UNP_UNLOCK(), UNP_LOCK_ASSERT() macros. - Acquire unp_mtx on entering most UNIX domain socket code, drop/re-acquire around calls into VFS, and release it on return. - Avoid performing sodupsockaddr() while holding the mutex, so in general move to allocating storage before acquiring the mutex to copy the data. - Make a stack copy of the xucred rather than copying out while holding unp_mtx. Copy the peer credential out after releasing the mutex. - Add additional assertions of vnode locks following VOP_CREATE(). A few notes: - Use of an sx lock for the file list mutex may cause problems with regard to unp_mtx when garbage collection passed file descriptors. - The locking in unp_pcblist() for sysctl monitoring is correct subject to the unpcb zone not returning memory for reuse by other subsystems (consistent with similar existing concerns). - Sam's version of this change, as with the BSD/OS version, made use of both a global lock and per-unpcb locks. However, in practice, the global lock covered all accesses, so I have simplified out the unpcb locks in the interest of getting this merged faster (reducing the overhead but not sacrificing granularity in most cases). We will want to explore possibilities for improving lock granularity in this code in the future. Submitted by: sam Sponsored by: FreeBSD Foundatiuon Obtained from: BSD/OS 5 snapshot provided by BSDi
2004-06-10 21:34:38 +00:00
const struct sockaddr *sa;
The long-awaited mega-massive-network-code- cleanup. Part I. This commit includes the following changes: 1) Old-style (pr_usrreq()) protocols are no longer supported, the compatibility glue for them is deleted, and the kernel will panic on boot if any are compiled in. 2) Certain protocol entry points are modified to take a process structure, so they they can easily tell whether or not it is possible to sleep, and also to access credentials. 3) SS_PRIV is no more, and with it goes the SO_PRIVSTATE setsockopt() call. Protocols should use the process pointer they are now passed. 4) The PF_LOCAL and PF_ROUTE families have been updated to use the new style, as has the `raw' skeleton family. 5) PF_LOCAL sockets now obey the process's umask when creating a socket in the filesystem. As a result, LINT is now broken. I'm hoping that some enterprising hacker with a bit more time will either make the broken bits work (should be easy for netipx) or dike them out.
1997-04-27 20:01:29 +00:00
Modify UNIX domain sockets to guarantee, and assume, that so_pcb is always defined for an in-use socket. This allows us to eliminate countless tests of whether so_pcb is non-NULL, eliminating dozens of error cases. For now, retain the call to sotryfree() in the uipc_abort() path, but this will eventually move to soabort(). These new assumptions should be largely correct, and will become more so as the socket/pcb reference model is fixed. Removing the notion that so_pcb can be non-NULL is a critical step towards further fine-graining of the UNIX domain socket locking, as the so_pcb reference no longer needs to be protected using locks, instead it is a property of the socket life cycle.
2006-03-17 13:52:57 +00:00
unp = sotounpcb(so);
KASSERT(unp != NULL, ("uipc_peeraddr: unp == NULL"));
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
Introduce a subsystem lock around UNIX domain sockets in order to protect global and allocated variables. This strategy is derived from work originally developed by BSDi for BSD/OS, and applied to FreeBSD by Sam Leffler: - Add unp_mtx, a global mutex which will protect all UNIX domain socket related variables, structures, etc. - Add UNP_LOCK(), UNP_UNLOCK(), UNP_LOCK_ASSERT() macros. - Acquire unp_mtx on entering most UNIX domain socket code, drop/re-acquire around calls into VFS, and release it on return. - Avoid performing sodupsockaddr() while holding the mutex, so in general move to allocating storage before acquiring the mutex to copy the data. - Make a stack copy of the xucred rather than copying out while holding unp_mtx. Copy the peer credential out after releasing the mutex. - Add additional assertions of vnode locks following VOP_CREATE(). A few notes: - Use of an sx lock for the file list mutex may cause problems with regard to unp_mtx when garbage collection passed file descriptors. - The locking in unp_pcblist() for sysctl monitoring is correct subject to the unpcb zone not returning memory for reuse by other subsystems (consistent with similar existing concerns). - Sam's version of this change, as with the BSD/OS version, made use of both a global lock and per-unpcb locks. However, in practice, the global lock covered all accesses, so I have simplified out the unpcb locks in the interest of getting this merged faster (reducing the overhead but not sacrificing granularity in most cases). We will want to explore possibilities for improving lock granularity in this code in the future. Submitted by: sam Sponsored by: FreeBSD Foundatiuon Obtained from: BSD/OS 5 snapshot provided by BSDi
2004-06-10 21:34:38 +00:00
*nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
Fix a deadlock in the getpeername() method for UNIX domain sockets. Instead of locking the local unp followed by the remote unp, use the same locking model as accept() and read lock the global link lock followed by the remote unp while fetching the remote sockaddr. Reported by: Mel Flynn mel.flynn of mailing.thruhere.net Reviewed by: rwatson MFC after: 1 week
2009-06-18 20:56:22 +00:00
UNP_LINK_RLOCK();
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
/*
* XXX: It seems that this test always fails even when connection is
* established. So, this else clause is added as workaround to
* return PF_LOCAL sockaddr.
*/
unp2 = unp->unp_conn;
if (unp2 != NULL) {
UNP_PCB_LOCK(unp2);
if (unp2->unp_addr != NULL)
Fix a deadlock in the getpeername() method for UNIX domain sockets. Instead of locking the local unp followed by the remote unp, use the same locking model as accept() and read lock the global link lock followed by the remote unp while fetching the remote sockaddr. Reported by: Mel Flynn mel.flynn of mailing.thruhere.net Reviewed by: rwatson MFC after: 1 week
2009-06-18 20:56:22 +00:00
sa = (struct sockaddr *) unp2->unp_addr;
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
else
sa = &sun_noname;
bcopy(sa, *nam, sa->sa_len);
UNP_PCB_UNLOCK(unp2);
} else {
Introduce a subsystem lock around UNIX domain sockets in order to protect global and allocated variables. This strategy is derived from work originally developed by BSDi for BSD/OS, and applied to FreeBSD by Sam Leffler: - Add unp_mtx, a global mutex which will protect all UNIX domain socket related variables, structures, etc. - Add UNP_LOCK(), UNP_UNLOCK(), UNP_LOCK_ASSERT() macros. - Acquire unp_mtx on entering most UNIX domain socket code, drop/re-acquire around calls into VFS, and release it on return. - Avoid performing sodupsockaddr() while holding the mutex, so in general move to allocating storage before acquiring the mutex to copy the data. - Make a stack copy of the xucred rather than copying out while holding unp_mtx. Copy the peer credential out after releasing the mutex. - Add additional assertions of vnode locks following VOP_CREATE(). A few notes: - Use of an sx lock for the file list mutex may cause problems with regard to unp_mtx when garbage collection passed file descriptors. - The locking in unp_pcblist() for sysctl monitoring is correct subject to the unpcb zone not returning memory for reuse by other subsystems (consistent with similar existing concerns). - Sam's version of this change, as with the BSD/OS version, made use of both a global lock and per-unpcb locks. However, in practice, the global lock covered all accesses, so I have simplified out the unpcb locks in the interest of getting this merged faster (reducing the overhead but not sacrificing granularity in most cases). We will want to explore possibilities for improving lock granularity in this code in the future. Submitted by: sam Sponsored by: FreeBSD Foundatiuon Obtained from: BSD/OS 5 snapshot provided by BSDi
2004-06-10 21:34:38 +00:00
sa = &sun_noname;
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
bcopy(sa, *nam, sa->sa_len);
Added comment why this workaround is required. Suggested by: sam MFC after: 1 week
2003-01-22 18:03:06 +00:00
}
Fix a deadlock in the getpeername() method for UNIX domain sockets. Instead of locking the local unp followed by the remote unp, use the same locking model as accept() and read lock the global link lock followed by the remote unp while fetching the remote sockaddr. Reported by: Mel Flynn mel.flynn of mailing.thruhere.net Reviewed by: rwatson MFC after: 1 week
2009-06-18 20:56:22 +00:00
UNP_LINK_RUNLOCK();
Mechanical whitespace cleanup; parenthesize return values; other minor style nits.
2004-01-11 19:48:19 +00:00
return (0);
The long-awaited mega-massive-network-code- cleanup. Part I. This commit includes the following changes: 1) Old-style (pr_usrreq()) protocols are no longer supported, the compatibility glue for them is deleted, and the kernel will panic on boot if any are compiled in. 2) Certain protocol entry points are modified to take a process structure, so they they can easily tell whether or not it is possible to sleep, and also to access credentials. 3) SS_PRIV is no more, and with it goes the SO_PRIVSTATE setsockopt() call. Protocols should use the process pointer they are now passed. 4) The PF_LOCAL and PF_ROUTE families have been updated to use the new style, as has the `raw' skeleton family. 5) PF_LOCAL sockets now obey the process's umask when creating a socket in the filesystem. As a result, LINT is now broken. I'm hoping that some enterprising hacker with a bit more time will either make the broken bits work (should be easy for netipx) or dike them out.
1997-04-27 20:01:29 +00:00
}
static int
uipc_rcvd(struct socket *so, int flags)
{
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
struct unpcb *unp, *unp2;
The long-awaited mega-massive-network-code- cleanup. Part I. This commit includes the following changes: 1) Old-style (pr_usrreq()) protocols are no longer supported, the compatibility glue for them is deleted, and the kernel will panic on boot if any are compiled in. 2) Certain protocol entry points are modified to take a process structure, so they they can easily tell whether or not it is possible to sleep, and also to access credentials. 3) SS_PRIV is no more, and with it goes the SO_PRIVSTATE setsockopt() call. Protocols should use the process pointer they are now passed. 4) The PF_LOCAL and PF_ROUTE families have been updated to use the new style, as has the `raw' skeleton family. 5) PF_LOCAL sockets now obey the process's umask when creating a socket in the filesystem. As a result, LINT is now broken. I'm hoping that some enterprising hacker with a bit more time will either make the broken bits work (should be easy for netipx) or dike them out.
1997-04-27 20:01:29 +00:00
struct socket *so2;
Reduce periods of simultaneous acquisition of various socket buffer locks and the unplock during uipc_rcvd() and uipc_send() by caching certain values from one structure while its locks are held, and applying them to a second structure while its locks are held. If done carefully, this should be correct, and will reduce the amount of work done with the global unp lock held. Tested by: kris (earlier version)
2006-07-11 21:49:54 +00:00
u_int mbcnt, sbcc;
The long-awaited mega-massive-network-code- cleanup. Part I. This commit includes the following changes: 1) Old-style (pr_usrreq()) protocols are no longer supported, the compatibility glue for them is deleted, and the kernel will panic on boot if any are compiled in. 2) Certain protocol entry points are modified to take a process structure, so they they can easily tell whether or not it is possible to sleep, and also to access credentials. 3) SS_PRIV is no more, and with it goes the SO_PRIVSTATE setsockopt() call. Protocols should use the process pointer they are now passed. 4) The PF_LOCAL and PF_ROUTE families have been updated to use the new style, as has the `raw' skeleton family. 5) PF_LOCAL sockets now obey the process's umask when creating a socket in the filesystem. As a result, LINT is now broken. I'm hoping that some enterprising hacker with a bit more time will either make the broken bits work (should be easy for netipx) or dike them out.
1997-04-27 20:01:29 +00:00
Always acquire the UNIX domain socket subsystem lock (UNP lock) before dereferencing sotounpcb() and checking its value, as so_pcb is protected by protocol locking, not subsystem locking. This prevents races during close() by one thread and use of ths socket in another. unp_bind() now assert the UNP lock, and uipc_bind() now acquires the lock around calls to unp_bind().
2004-08-16 04:41:03 +00:00
unp = sotounpcb(so);
Merge from projects/sendfile: - Use KASSERT()s instead of panic(). - Use sbavail() instead of sb_cc. Sponsored by: Nginx, Inc. Sponsored by: Netflix
2014-11-12 10:17:46 +00:00
KASSERT(unp != NULL, ("%s: unp == NULL", __func__));
KASSERT(so->so_type == SOCK_STREAM || so->so_type == SOCK_SEQPACKET,
("%s: socktype %d", __func__, so->so_type));
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
/*
* Adjust backpressure on sender and wakeup any waiting to write.
*
Clarify and update quite a few comments to reflect locking optimizations, the addition of unpcb refcounts, and bug fixes. Some of these fixes are appropriate for MFC. MFC after: 3 days
2007-05-11 12:10:45 +00:00
* The unp lock is acquired to maintain the validity of the unp_conn
* pointer; no lock on unp2 is required as unp2->unp_socket will be
* static as long as we don't permit unp2 to disconnect from unp,
* which is prevented by the lock on unp. We cache values from
* so_rcv to avoid holding the so_rcv lock over the entire
* transaction on the remote so_snd.
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
*/
SOCKBUF_LOCK(&so->so_rcv);
mbcnt = so->so_rcv.sb_mbcnt;
Merge from projects/sendfile: - Use KASSERT()s instead of panic(). - Use sbavail() instead of sb_cc. Sponsored by: Nginx, Inc. Sponsored by: Netflix
2014-11-12 10:17:46 +00:00
sbcc = sbavail(&so->so_rcv);
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
SOCKBUF_UNLOCK(&so->so_rcv);
Replace 4.4BSD Lite's unix domain socket backpressure hack with a cleaner mechanism, based on the new SB_STOP sockbuf flag. The old hack dynamically changed the sending sockbuf's high water mark whenever adding or removing data from the receiving sockbuf. It worked for stream sockets, but it never worked for SOCK_SEQPACKET sockets because of their atomic nature. If the sockbuf was partially full, it might return EMSGSIZE instead of blocking. The new solution is based on DragonFlyBSD's fix from commit 3a6117bbe0ed6a87605c1e43e12a1438d8844380 on 2008-05-27. It adds an SB_STOP flag to sockbufs. Whenever uipc_send surpasses the socket's size limit, it sets SB_STOP on the sending sockbuf. sbspace() will then return 0 for that sockbuf, causing sosend_generic and friends to block. uipc_rcvd will likewise clear SB_STOP. There are two fringe benefits: uipc_{send,rcvd} no longer need to call chgsbsize() on every send and receive because they don't change the sockbuf's high water mark. Also, uipc_sense no longer needs to acquire the UIPC linkage lock, because it's simpler to compute the st_blksizes. There is one drawback: since sbspace() will only ever return 0 or the maximum, sosend_generic will allow the sockbuf to exceed its nominal maximum size by at most one packet of size less than the max. I don't think that's a serious problem. In fact, I'm not even positive that FreeBSD guarantees a socket will always stay within its nominal size limit. sys/sys/sockbuf.h Add the SB_STOP flag and adjust sbspace() sys/sys/unpcb.h Delete the obsolete unp_cc and unp_mbcnt fields from struct unpcb. sys/kern/uipc_usrreq.c Adjust uipc_rcvd, uipc_send, and uipc_sense to use the SB_STOP backpressure mechanism. Removing obsolete unpcb fields from db_show_unpcb. tests/sys/kern/unix_seqpacket_test.c Clear expected failures from ATF. Obtained from: DragonFly BSD PR: kern/185812 Reviewed by: silence from freebsd-net@ and rwatson@ MFC after: 3 weeks Sponsored by: Spectra Logic Corporation
2014-03-13 18:42:12 +00:00
/*
* There is a benign race condition at this point. If we're planning to
* clear SB_STOP, but uipc_send is called on the connected socket at
* this instant, it might add data to the sockbuf and set SB_STOP. Then
* we would erroneously clear SB_STOP below, even though the sockbuf is
* full. The race is benign because the only ill effect is to allow the
* sockbuf to exceed its size limit, and the size limits are not
* strictly guaranteed anyway.
*/
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
UNP_PCB_LOCK(unp);
unp2 = unp->unp_conn;
if (unp2 == NULL) {
UNP_PCB_UNLOCK(unp);
return (0);
The long-awaited mega-massive-network-code- cleanup. Part I. This commit includes the following changes: 1) Old-style (pr_usrreq()) protocols are no longer supported, the compatibility glue for them is deleted, and the kernel will panic on boot if any are compiled in. 2) Certain protocol entry points are modified to take a process structure, so they they can easily tell whether or not it is possible to sleep, and also to access credentials. 3) SS_PRIV is no more, and with it goes the SO_PRIVSTATE setsockopt() call. Protocols should use the process pointer they are now passed. 4) The PF_LOCAL and PF_ROUTE families have been updated to use the new style, as has the `raw' skeleton family. 5) PF_LOCAL sockets now obey the process's umask when creating a socket in the filesystem. As a result, LINT is now broken. I'm hoping that some enterprising hacker with a bit more time will either make the broken bits work (should be easy for netipx) or dike them out.
1997-04-27 20:01:29 +00:00
}
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
so2 = unp2->unp_socket;
SOCKBUF_LOCK(&so2->so_snd);
Replace 4.4BSD Lite's unix domain socket backpressure hack with a cleaner mechanism, based on the new SB_STOP sockbuf flag. The old hack dynamically changed the sending sockbuf's high water mark whenever adding or removing data from the receiving sockbuf. It worked for stream sockets, but it never worked for SOCK_SEQPACKET sockets because of their atomic nature. If the sockbuf was partially full, it might return EMSGSIZE instead of blocking. The new solution is based on DragonFlyBSD's fix from commit 3a6117bbe0ed6a87605c1e43e12a1438d8844380 on 2008-05-27. It adds an SB_STOP flag to sockbufs. Whenever uipc_send surpasses the socket's size limit, it sets SB_STOP on the sending sockbuf. sbspace() will then return 0 for that sockbuf, causing sosend_generic and friends to block. uipc_rcvd will likewise clear SB_STOP. There are two fringe benefits: uipc_{send,rcvd} no longer need to call chgsbsize() on every send and receive because they don't change the sockbuf's high water mark. Also, uipc_sense no longer needs to acquire the UIPC linkage lock, because it's simpler to compute the st_blksizes. There is one drawback: since sbspace() will only ever return 0 or the maximum, sosend_generic will allow the sockbuf to exceed its nominal maximum size by at most one packet of size less than the max. I don't think that's a serious problem. In fact, I'm not even positive that FreeBSD guarantees a socket will always stay within its nominal size limit. sys/sys/sockbuf.h Add the SB_STOP flag and adjust sbspace() sys/sys/unpcb.h Delete the obsolete unp_cc and unp_mbcnt fields from struct unpcb. sys/kern/uipc_usrreq.c Adjust uipc_rcvd, uipc_send, and uipc_sense to use the SB_STOP backpressure mechanism. Removing obsolete unpcb fields from db_show_unpcb. tests/sys/kern/unix_seqpacket_test.c Clear expected failures from ATF. Obtained from: DragonFly BSD PR: kern/185812 Reviewed by: silence from freebsd-net@ and rwatson@ MFC after: 3 weeks Sponsored by: Spectra Logic Corporation
2014-03-13 18:42:12 +00:00
if (sbcc < so2->so_snd.sb_hiwat && mbcnt < so2->so_snd.sb_mbmax)
so2->so_snd.sb_flags &= ~SB_STOP;
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
sowwakeup_locked(so2);
UNP_PCB_UNLOCK(unp);
Mechanical whitespace cleanup; parenthesize return values; other minor style nits.
2004-01-11 19:48:19 +00:00
return (0);
The long-awaited mega-massive-network-code- cleanup. Part I. This commit includes the following changes: 1) Old-style (pr_usrreq()) protocols are no longer supported, the compatibility glue for them is deleted, and the kernel will panic on boot if any are compiled in. 2) Certain protocol entry points are modified to take a process structure, so they they can easily tell whether or not it is possible to sleep, and also to access credentials. 3) SS_PRIV is no more, and with it goes the SO_PRIVSTATE setsockopt() call. Protocols should use the process pointer they are now passed. 4) The PF_LOCAL and PF_ROUTE families have been updated to use the new style, as has the `raw' skeleton family. 5) PF_LOCAL sockets now obey the process's umask when creating a socket in the filesystem. As a result, LINT is now broken. I'm hoping that some enterprising hacker with a bit more time will either make the broken bits work (should be easy for netipx) or dike them out.
1997-04-27 20:01:29 +00:00
}
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
The long-awaited mega-massive-network-code- cleanup. Part I. This commit includes the following changes: 1) Old-style (pr_usrreq()) protocols are no longer supported, the compatibility glue for them is deleted, and the kernel will panic on boot if any are compiled in. 2) Certain protocol entry points are modified to take a process structure, so they they can easily tell whether or not it is possible to sleep, and also to access credentials. 3) SS_PRIV is no more, and with it goes the SO_PRIVSTATE setsockopt() call. Protocols should use the process pointer they are now passed. 4) The PF_LOCAL and PF_ROUTE families have been updated to use the new style, as has the `raw' skeleton family. 5) PF_LOCAL sockets now obey the process's umask when creating a socket in the filesystem. As a result, LINT is now broken. I'm hoping that some enterprising hacker with a bit more time will either make the broken bits work (should be easy for netipx) or dike them out.
1997-04-27 20:01:29 +00:00
static int
Fix all areas of the system (or at least all those in LINT) to avoid storing socket addresses in mbufs. (Socket buffers are the one exception.) A number of kernel APIs needed to get fixed in order to make this happen. Also, fix three protocol families which kept PCBs in mbufs to not malloc them instead. Delete some old compatibility cruft while we're at it, and add some new routines in the in_cksum family.
1997-08-16 19:16:27 +00:00
uipc_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
style(9)-ize function headers, remove use of 'register'. MFC after: 3 days
2005-02-20 23:22:13 +00:00
struct mbuf *control, struct thread *td)
The long-awaited mega-massive-network-code- cleanup. Part I. This commit includes the following changes: 1) Old-style (pr_usrreq()) protocols are no longer supported, the compatibility glue for them is deleted, and the kernel will panic on boot if any are compiled in. 2) Certain protocol entry points are modified to take a process structure, so they they can easily tell whether or not it is possible to sleep, and also to access credentials. 3) SS_PRIV is no more, and with it goes the SO_PRIVSTATE setsockopt() call. Protocols should use the process pointer they are now passed. 4) The PF_LOCAL and PF_ROUTE families have been updated to use the new style, as has the `raw' skeleton family. 5) PF_LOCAL sockets now obey the process's umask when creating a socket in the filesystem. As a result, LINT is now broken. I'm hoping that some enterprising hacker with a bit more time will either make the broken bits work (should be easy for netipx) or dike them out.
1997-04-27 20:01:29 +00:00
{
In uipc_send() and uipc_rcvd(), store unp->unp_conn pointer in unp2 while working with the second unpcb to make the code more clear.
2006-07-22 18:41:42 +00:00
struct unpcb *unp, *unp2;
The long-awaited mega-massive-network-code- cleanup. Part I. This commit includes the following changes: 1) Old-style (pr_usrreq()) protocols are no longer supported, the compatibility glue for them is deleted, and the kernel will panic on boot if any are compiled in. 2) Certain protocol entry points are modified to take a process structure, so they they can easily tell whether or not it is possible to sleep, and also to access credentials. 3) SS_PRIV is no more, and with it goes the SO_PRIVSTATE setsockopt() call. Protocols should use the process pointer they are now passed. 4) The PF_LOCAL and PF_ROUTE families have been updated to use the new style, as has the `raw' skeleton family. 5) PF_LOCAL sockets now obey the process's umask when creating a socket in the filesystem. As a result, LINT is now broken. I'm hoping that some enterprising hacker with a bit more time will either make the broken bits work (should be easy for netipx) or dike them out.
1997-04-27 20:01:29 +00:00
struct socket *so2;
Replace 4.4BSD Lite's unix domain socket backpressure hack with a cleaner mechanism, based on the new SB_STOP sockbuf flag. The old hack dynamically changed the sending sockbuf's high water mark whenever adding or removing data from the receiving sockbuf. It worked for stream sockets, but it never worked for SOCK_SEQPACKET sockets because of their atomic nature. If the sockbuf was partially full, it might return EMSGSIZE instead of blocking. The new solution is based on DragonFlyBSD's fix from commit 3a6117bbe0ed6a87605c1e43e12a1438d8844380 on 2008-05-27. It adds an SB_STOP flag to sockbufs. Whenever uipc_send surpasses the socket's size limit, it sets SB_STOP on the sending sockbuf. sbspace() will then return 0 for that sockbuf, causing sosend_generic and friends to block. uipc_rcvd will likewise clear SB_STOP. There are two fringe benefits: uipc_{send,rcvd} no longer need to call chgsbsize() on every send and receive because they don't change the sockbuf's high water mark. Also, uipc_sense no longer needs to acquire the UIPC linkage lock, because it's simpler to compute the st_blksizes. There is one drawback: since sbspace() will only ever return 0 or the maximum, sosend_generic will allow the sockbuf to exceed its nominal maximum size by at most one packet of size less than the max. I don't think that's a serious problem. In fact, I'm not even positive that FreeBSD guarantees a socket will always stay within its nominal size limit. sys/sys/sockbuf.h Add the SB_STOP flag and adjust sbspace() sys/sys/unpcb.h Delete the obsolete unp_cc and unp_mbcnt fields from struct unpcb. sys/kern/uipc_usrreq.c Adjust uipc_rcvd, uipc_send, and uipc_sense to use the SB_STOP backpressure mechanism. Removing obsolete unpcb fields from db_show_unpcb. tests/sys/kern/unix_seqpacket_test.c Clear expected failures from ATF. Obtained from: DragonFly BSD PR: kern/185812 Reviewed by: silence from freebsd-net@ and rwatson@ MFC after: 3 weeks Sponsored by: Spectra Logic Corporation
2014-03-13 18:42:12 +00:00
u_int mbcnt, sbcc;
In uipc_send() and uipc_rcvd(), store unp->unp_conn pointer in unp2 while working with the second unpcb to make the code more clear.
2006-07-22 18:41:42 +00:00
int error = 0;
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
Always acquire the UNIX domain socket subsystem lock (UNP lock) before dereferencing sotounpcb() and checking its value, as so_pcb is protected by protocol locking, not subsystem locking. This prevents races during close() by one thread and use of ths socket in another. unp_bind() now assert the UNP lock, and uipc_bind() now acquires the lock around calls to unp_bind().
2004-08-16 04:41:03 +00:00
unp = sotounpcb(so);
Merge from projects/sendfile: - Use KASSERT()s instead of panic(). - Use sbavail() instead of sb_cc. Sponsored by: Nginx, Inc. Sponsored by: Netflix
2014-11-12 10:17:46 +00:00
KASSERT(unp != NULL, ("%s: unp == NULL", __func__));
KASSERT(so->so_type == SOCK_STREAM || so->so_type == SOCK_DGRAM ||
so->so_type == SOCK_SEQPACKET,
("%s: socktype %d", __func__, so->so_type));
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
The long-awaited mega-massive-network-code- cleanup. Part I. This commit includes the following changes: 1) Old-style (pr_usrreq()) protocols are no longer supported, the compatibility glue for them is deleted, and the kernel will panic on boot if any are compiled in. 2) Certain protocol entry points are modified to take a process structure, so they they can easily tell whether or not it is possible to sleep, and also to access credentials. 3) SS_PRIV is no more, and with it goes the SO_PRIVSTATE setsockopt() call. Protocols should use the process pointer they are now passed. 4) The PF_LOCAL and PF_ROUTE families have been updated to use the new style, as has the `raw' skeleton family. 5) PF_LOCAL sockets now obey the process's umask when creating a socket in the filesystem. As a result, LINT is now broken. I'm hoping that some enterprising hacker with a bit more time will either make the broken bits work (should be easy for netipx) or dike them out.
1997-04-27 20:01:29 +00:00
if (flags & PRUS_OOB) {
error = EOPNOTSUPP;
goto release;
}
Prefer NULL to 0 when testing and assigning pointer values.
2004-03-30 02:16:25 +00:00
if (control != NULL && (error = unp_internalize(&control, td)))
The long-awaited mega-massive-network-code- cleanup. Part I. This commit includes the following changes: 1) Old-style (pr_usrreq()) protocols are no longer supported, the compatibility glue for them is deleted, and the kernel will panic on boot if any are compiled in. 2) Certain protocol entry points are modified to take a process structure, so they they can easily tell whether or not it is possible to sleep, and also to access credentials. 3) SS_PRIV is no more, and with it goes the SO_PRIVSTATE setsockopt() call. Protocols should use the process pointer they are now passed. 4) The PF_LOCAL and PF_ROUTE families have been updated to use the new style, as has the `raw' skeleton family. 5) PF_LOCAL sockets now obey the process's umask when creating a socket in the filesystem. As a result, LINT is now broken. I'm hoping that some enterprising hacker with a bit more time will either make the broken bits work (should be easy for netipx) or dike them out.
1997-04-27 20:01:29 +00:00
goto release;
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
if ((nam != NULL) || (flags & PRUS_EOF))
Decompose the global UNIX domain sockets rwlock into two different locks: a global list/counter/generation counter protected by a new mutex unp_list_lock, and a global linkage rwlock, unp_global_rwlock, which protects the connections between UNIX domain sockets. This eliminates conditional lock acquisition that was previously a property of the global lock being held over sonewconn() leading to a call to uipc_attach(), which also required the global lock, but couldn't rely on it as other paths existed to uipc_attach() that didn't hold it: now uipc_attach() uses only the list lock, which follows the linkage lock in the lock order. It may also reduce contention on the global lock for some workloads. Add global UNIX domain socket locks to hard-coded witness lock order. MFC after: 1 week Discussed with: kris
2009-03-08 21:48:29 +00:00
UNP_LINK_WLOCK();
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
else
Decompose the global UNIX domain sockets rwlock into two different locks: a global list/counter/generation counter protected by a new mutex unp_list_lock, and a global linkage rwlock, unp_global_rwlock, which protects the connections between UNIX domain sockets. This eliminates conditional lock acquisition that was previously a property of the global lock being held over sonewconn() leading to a call to uipc_attach(), which also required the global lock, but couldn't rely on it as other paths existed to uipc_attach() that didn't hold it: now uipc_attach() uses only the list lock, which follows the linkage lock in the lock order. It may also reduce contention on the global lock for some workloads. Add global UNIX domain socket locks to hard-coded witness lock order. MFC after: 1 week Discussed with: kris
2009-03-08 21:48:29 +00:00
UNP_LINK_RLOCK();
The long-awaited mega-massive-network-code- cleanup. Part I. This commit includes the following changes: 1) Old-style (pr_usrreq()) protocols are no longer supported, the compatibility glue for them is deleted, and the kernel will panic on boot if any are compiled in. 2) Certain protocol entry points are modified to take a process structure, so they they can easily tell whether or not it is possible to sleep, and also to access credentials. 3) SS_PRIV is no more, and with it goes the SO_PRIVSTATE setsockopt() call. Protocols should use the process pointer they are now passed. 4) The PF_LOCAL and PF_ROUTE families have been updated to use the new style, as has the `raw' skeleton family. 5) PF_LOCAL sockets now obey the process's umask when creating a socket in the filesystem. As a result, LINT is now broken. I'm hoping that some enterprising hacker with a bit more time will either make the broken bits work (should be easy for netipx) or dike them out.
1997-04-27 20:01:29 +00:00
switch (so->so_type) {
Mechanical whitespace cleanup; parenthesize return values; other minor style nits.
2004-01-11 19:48:19 +00:00
case SOCK_DGRAM:
The long-awaited mega-massive-network-code- cleanup. Part I. This commit includes the following changes: 1) Old-style (pr_usrreq()) protocols are no longer supported, the compatibility glue for them is deleted, and the kernel will panic on boot if any are compiled in. 2) Certain protocol entry points are modified to take a process structure, so they they can easily tell whether or not it is possible to sleep, and also to access credentials. 3) SS_PRIV is no more, and with it goes the SO_PRIVSTATE setsockopt() call. Protocols should use the process pointer they are now passed. 4) The PF_LOCAL and PF_ROUTE families have been updated to use the new style, as has the `raw' skeleton family. 5) PF_LOCAL sockets now obey the process's umask when creating a socket in the filesystem. As a result, LINT is now broken. I'm hoping that some enterprising hacker with a bit more time will either make the broken bits work (should be easy for netipx) or dike them out.
1997-04-27 20:01:29 +00:00
{
Mark sun_noname as const since it's immutable. Update definitions of functions that potentially accept &sun_noname (sbappendaddr(), et al) to accept a const sockaddr pointer.
2004-06-04 04:07:08 +00:00
const struct sockaddr *from;
Remove trailing whitespace.
1995-05-30 08:16:23 +00:00
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
unp2 = unp->unp_conn;
Prefer NULL to 0 when testing and assigning pointer values.
2004-03-30 02:16:25 +00:00
if (nam != NULL) {
Decompose the global UNIX domain sockets rwlock into two different locks: a global list/counter/generation counter protected by a new mutex unp_list_lock, and a global linkage rwlock, unp_global_rwlock, which protects the connections between UNIX domain sockets. This eliminates conditional lock acquisition that was previously a property of the global lock being held over sonewconn() leading to a call to uipc_attach(), which also required the global lock, but couldn't rely on it as other paths existed to uipc_attach() that didn't hold it: now uipc_attach() uses only the list lock, which follows the linkage lock in the lock order. It may also reduce contention on the global lock for some workloads. Add global UNIX domain socket locks to hard-coded witness lock order. MFC after: 1 week Discussed with: kris
2009-03-08 21:48:29 +00:00
UNP_LINK_WLOCK_ASSERT();
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
if (unp2 != NULL) {
The long-awaited mega-massive-network-code- cleanup. Part I. This commit includes the following changes: 1) Old-style (pr_usrreq()) protocols are no longer supported, the compatibility glue for them is deleted, and the kernel will panic on boot if any are compiled in. 2) Certain protocol entry points are modified to take a process structure, so they they can easily tell whether or not it is possible to sleep, and also to access credentials. 3) SS_PRIV is no more, and with it goes the SO_PRIVSTATE setsockopt() call. Protocols should use the process pointer they are now passed. 4) The PF_LOCAL and PF_ROUTE families have been updated to use the new style, as has the `raw' skeleton family. 5) PF_LOCAL sockets now obey the process's umask when creating a socket in the filesystem. As a result, LINT is now broken. I'm hoping that some enterprising hacker with a bit more time will either make the broken bits work (should be easy for netipx) or dike them out.
1997-04-27 20:01:29 +00:00
error = EISCONN;
break;
}
KSE Milestone 2 Note ALL MODULES MUST BE RECOMPILED make the kernel aware that there are smaller units of scheduling than the process. (but only allow one thread per process at this time). This is functionally equivalent to teh previousl -current except that there is a thread associated with each process. Sorry john! (your next MFC will be a doosie!) Reviewed by: peter@freebsd.org, dillon@freebsd.org X-MFC after: ha ha ha ha
2001-09-12 08:38:13 +00:00
error = unp_connect(so, nam, td);
The long-awaited mega-massive-network-code- cleanup. Part I. This commit includes the following changes: 1) Old-style (pr_usrreq()) protocols are no longer supported, the compatibility glue for them is deleted, and the kernel will panic on boot if any are compiled in. 2) Certain protocol entry points are modified to take a process structure, so they they can easily tell whether or not it is possible to sleep, and also to access credentials. 3) SS_PRIV is no more, and with it goes the SO_PRIVSTATE setsockopt() call. Protocols should use the process pointer they are now passed. 4) The PF_LOCAL and PF_ROUTE families have been updated to use the new style, as has the `raw' skeleton family. 5) PF_LOCAL sockets now obey the process's umask when creating a socket in the filesystem. As a result, LINT is now broken. I'm hoping that some enterprising hacker with a bit more time will either make the broken bits work (should be easy for netipx) or dike them out.
1997-04-27 20:01:29 +00:00
if (error)
break;
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
unp2 = unp->unp_conn;
The long-awaited mega-massive-network-code- cleanup. Part I. This commit includes the following changes: 1) Old-style (pr_usrreq()) protocols are no longer supported, the compatibility glue for them is deleted, and the kernel will panic on boot if any are compiled in. 2) Certain protocol entry points are modified to take a process structure, so they they can easily tell whether or not it is possible to sleep, and also to access credentials. 3) SS_PRIV is no more, and with it goes the SO_PRIVSTATE setsockopt() call. Protocols should use the process pointer they are now passed. 4) The PF_LOCAL and PF_ROUTE families have been updated to use the new style, as has the `raw' skeleton family. 5) PF_LOCAL sockets now obey the process's umask when creating a socket in the filesystem. As a result, LINT is now broken. I'm hoping that some enterprising hacker with a bit more time will either make the broken bits work (should be easy for netipx) or dike them out.
1997-04-27 20:01:29 +00:00
}
Remove or update several stale comments. A bit of whitespace/style cleanup. Update copyright. MFC after: 3 days (applicable changes)
2008-10-03 09:01:55 +00:00
Close a race that occurs when using sendto() to connect and send on a UNIX domain socket at the same time as the remote host is closing the new connections as quickly as they open. Since the connect() and send() paths are non-atomic with respect to another, it is possible for the second thread's close() call to disconnect the two sockets as connect() returns, leading to the consumer (which plans to send()) with a NULL kernel pointer to its proposed peer. As a result, after acquiring the UNIX domain socket subsystem lock, we need to revalidate the connection pointers even though connect() has technically succeed, and reurn an error to say that there's no connection on which to perform the send. We might want to rethink the specific errno number, perhaps ECONNRESET would be better. PR: 100940 Reported by: Young Hyun <youngh at caida dot org> MFC after: 2 weeks MFC note: Some adaptation will be required
2006-07-31 23:00:05 +00:00
/*
* Because connect() and send() are non-atomic in a sendto()
* with a target address, it's possible that the socket will
* have disconnected before the send() can run. In that case
* return the slightly counter-intuitive but otherwise
* correct error that the socket is not connected.
*/
if (unp2 == NULL) {
error = ENOTCONN;
break;
}
Remove two simultaneous acquisitions of multiple unpcb locks from uipc_send in cases where only a global read lock is held by breaking them out and avoiding the unpcb lock acquire in the common case. This avoids deadlocks which manifested with X11, and should also marginally further improve performance. Reported by: sepotvin, brooks
2007-03-01 09:00:42 +00:00
/* Lockless read. */
if (unp2->unp_flags & UNP_WANTCRED)
control = unp_addsockcred(td, control);
UNP_PCB_LOCK(unp);
Prefer NULL to 0 when testing and assigning pointer values.
2004-03-30 02:16:25 +00:00
if (unp->unp_addr != NULL)
Fix all areas of the system (or at least all those in LINT) to avoid storing socket addresses in mbufs. (Socket buffers are the one exception.) A number of kernel APIs needed to get fixed in order to make this happen. Also, fix three protocol families which kept PCBs in mbufs to not malloc them instead. Delete some old compatibility cruft while we're at it, and add some new routines in the in_cksum family.
1997-08-16 19:16:27 +00:00
from = (struct sockaddr *)unp->unp_addr;
The long-awaited mega-massive-network-code- cleanup. Part I. This commit includes the following changes: 1) Old-style (pr_usrreq()) protocols are no longer supported, the compatibility glue for them is deleted, and the kernel will panic on boot if any are compiled in. 2) Certain protocol entry points are modified to take a process structure, so they they can easily tell whether or not it is possible to sleep, and also to access credentials. 3) SS_PRIV is no more, and with it goes the SO_PRIVSTATE setsockopt() call. Protocols should use the process pointer they are now passed. 4) The PF_LOCAL and PF_ROUTE families have been updated to use the new style, as has the `raw' skeleton family. 5) PF_LOCAL sockets now obey the process's umask when creating a socket in the filesystem. As a result, LINT is now broken. I'm hoping that some enterprising hacker with a bit more time will either make the broken bits work (should be easy for netipx) or dike them out.
1997-04-27 20:01:29 +00:00
else
from = &sun_noname;
Remove two simultaneous acquisitions of multiple unpcb locks from uipc_send in cases where only a global read lock is held by breaking them out and avoiding the unpcb lock acquire in the common case. This avoids deadlocks which manifested with X11, and should also marginally further improve performance. Reported by: sepotvin, brooks
2007-03-01 09:00:42 +00:00
so2 = unp2->unp_socket;
Merge next step in socket buffer locking: - sowakeup() now asserts the socket buffer lock on entry. Move the call to KNOTE higher in sowakeup() so that it is made with the socket buffer lock held for consistency with other calls. Release the socket buffer lock prior to calling into pgsigio(), so_upcall(), or aio_swake(). Locking for this event management will need revisiting in the future, but this model avoids lock order reversals when upcalls into other subsystems result in socket/socket buffer operations. Assert that the socket buffer lock is not held at the end of the function. - Wrapper macros for sowakeup(), sorwakeup() and sowwakeup(), now have _locked versions which assert the socket buffer lock on entry. If a wakeup is required by sb_notify(), invoke sowakeup(); otherwise, unconditionally release the socket buffer lock. This results in the socket buffer lock being released whether a wakeup is required or not. - Break out socantsendmore() into socantsendmore_locked() that asserts the socket buffer lock. socantsendmore() unconditionally locks the socket buffer before calling socantsendmore_locked(). Note that both functions return with the socket buffer unlocked as socantsendmore_locked() calls sowwakeup_locked() which has the same properties. Assert that the socket buffer is unlocked on return. - Break out socantrcvmore() into socantrcvmore_locked() that asserts the socket buffer lock. socantrcvmore() unconditionally locks the socket buffer before calling socantrcvmore_locked(). Note that both functions return with the socket buffer unlocked as socantrcvmore_locked() calls sorwakeup_locked() which has similar properties. Assert that the socket buffer is unlocked on return. - Break out sbrelease() into a sbrelease_locked() that asserts the socket buffer lock. sbrelease() unconditionally locks the socket buffer before calling sbrelease_locked(). sbrelease_locked() now invokes sbflush_locked() instead of sbflush(). - Assert the socket buffer lock in socket buffer sanity check functions sblastrecordchk(), sblastmbufchk(). - Assert the socket buffer lock in SBLINKRECORD(). - Break out various sbappend() functions into sbappend_locked() (and variations on that name) that assert the socket buffer lock. The !_locked() variations unconditionally lock the socket buffer before calling their _locked counterparts. Internally, make sure to call _locked() support routines, etc, if already holding the socket buffer lock. - Break out sbinsertoob() into sbinsertoob_locked() that asserts the socket buffer lock. sbinsertoob() unconditionally locks the socket buffer before calling sbinsertoob_locked(). - Break out sbflush() into sbflush_locked() that asserts the socket buffer lock. sbflush() unconditionally locks the socket buffer before calling sbflush_locked(). Update panic strings for new function names. - Break out sbdrop() into sbdrop_locked() that asserts the socket buffer lock. sbdrop() unconditionally locks the socket buffer before calling sbdrop_locked(). - Break out sbdroprecord() into sbdroprecord_locked() that asserts the socket buffer lock. sbdroprecord() unconditionally locks the socket buffer before calling sbdroprecord_locked(). - sofree() now calls socantsendmore_locked() and re-acquires the socket buffer lock on return. It also now calls sbrelease_locked(). - sorflush() now calls socantrcvmore_locked() and re-acquires the socket buffer lock on return. Clean up/mess up other behavior in sorflush() relating to the temporary stack copy of the socket buffer used with dom_dispose by more properly initializing the temporary copy, and selectively bzeroing/copying more carefully to prevent WITNESS from getting confused by improperly initialized mutexes. Annotate why that's necessary, or at least, needed. - soisconnected() now calls sbdrop_locked() before unlocking the socket buffer to avoid locking overhead. Some parts of this change were: Submitted by: sam Sponsored by: FreeBSD Foundation Obtained from: BSD/OS
2004-06-21 00:20:43 +00:00
SOCKBUF_LOCK(&so2->so_rcv);
Partial revert of r262867. r262867 was described as fixing socket buffer checks for SOCK_SEQPACKET, but also changed one of the SOCK_DGRAM code paths to use the new sbappendaddr_nospacecheck_locked() function. This lead to SOCK_DGRAM bypassing socket buffer limits.
2014-08-03 22:37:21 +00:00
if (sbappendaddr_locked(&so2->so_rcv, from, m,
Fix PR kern/185813 "SOCK_SEQPACKET AF_UNIX sockets with asymmetrical buffers drop packets". It was caused by a check for the space available in a sockbuf, but it was checking the wrong sockbuf. sys/sys/sockbuf.h sys/kern/uipc_sockbuf.c Add sbappendaddr_nospacecheck_locked(), which is just like sbappendaddr_locked but doesn't validate the receiving socket's space. Factor out common code into sbappendaddr_locked_internal(). We shouldn't simply make sbappendaddr_locked check the space and then call sbappendaddr_nospacecheck_locked, because that would cause the O(n) function m_length to be called twice. sys/kern/uipc_usrreq.c Use sbappendaddr_nospacecheck_locked for SOCK_SEQPACKET sockets, because the receiving sockbuf's size limit is irrelevant. tests/sys/kern/unix_seqpacket_test.c Now that 185813 is fixed, pipe_128k_8k fails intermittently due to 185812. Make it fail every time by adding a usleep after starting the writer thread and before starting the reader thread in test_pipe. That gives the writer time to fill up its send buffer. Also, clear the expected failure message due to 185813. It actually said "185812", but that was a typo. PR: kern/185813 Reviewed by: silence from freebsd-net@ and rwatson@ MFC after: 3 weeks Sponsored by: Spectra Logic Corporation
2014-03-06 20:24:15 +00:00
control)) {
Reduce the number of unnecessary unlock-relocks on socket buffer mutexes associated with performing a wakeup on the socket buffer: - When performing an sbappend*() followed by a so[rw]wakeup(), explicitly acquire the socket buffer lock and use the _locked() variants of both calls. Note that the _locked() sowakeup() versions unlock the mutex on return. This is done in uipc_send(), divert_packet(), mroute socket_send(), raw_append(), tcp_reass(), tcp_input(), and udp_append(). - When the socket buffer lock is dropped before a sowakeup(), remove the explicit unlock and use the _locked() sowakeup() variant. This is done in soisdisconnecting(), soisdisconnected() when setting the can't send/ receive flags and dropping data, and in uipc_rcvd() which adjusting back-pressure on the sockets. For UNIX domain sockets running mpsafe with a contention-intensive SMP mysql benchmark, this results in a 1.6% query rate improvement due to reduce mutex costs.
2004-06-26 19:10:39 +00:00
sorwakeup_locked(so2);
Prefer NULL to 0 when testing and assigning pointer values.
2004-03-30 02:16:25 +00:00
m = NULL;
control = NULL;
Mechanical whitespace cleanup; parenthesize return values; other minor style nits.
2004-01-11 19:48:19 +00:00
} else {
Merge next step in socket buffer locking: - sowakeup() now asserts the socket buffer lock on entry. Move the call to KNOTE higher in sowakeup() so that it is made with the socket buffer lock held for consistency with other calls. Release the socket buffer lock prior to calling into pgsigio(), so_upcall(), or aio_swake(). Locking for this event management will need revisiting in the future, but this model avoids lock order reversals when upcalls into other subsystems result in socket/socket buffer operations. Assert that the socket buffer lock is not held at the end of the function. - Wrapper macros for sowakeup(), sorwakeup() and sowwakeup(), now have _locked versions which assert the socket buffer lock on entry. If a wakeup is required by sb_notify(), invoke sowakeup(); otherwise, unconditionally release the socket buffer lock. This results in the socket buffer lock being released whether a wakeup is required or not. - Break out socantsendmore() into socantsendmore_locked() that asserts the socket buffer lock. socantsendmore() unconditionally locks the socket buffer before calling socantsendmore_locked(). Note that both functions return with the socket buffer unlocked as socantsendmore_locked() calls sowwakeup_locked() which has the same properties. Assert that the socket buffer is unlocked on return. - Break out socantrcvmore() into socantrcvmore_locked() that asserts the socket buffer lock. socantrcvmore() unconditionally locks the socket buffer before calling socantrcvmore_locked(). Note that both functions return with the socket buffer unlocked as socantrcvmore_locked() calls sorwakeup_locked() which has similar properties. Assert that the socket buffer is unlocked on return. - Break out sbrelease() into a sbrelease_locked() that asserts the socket buffer lock. sbrelease() unconditionally locks the socket buffer before calling sbrelease_locked(). sbrelease_locked() now invokes sbflush_locked() instead of sbflush(). - Assert the socket buffer lock in socket buffer sanity check functions sblastrecordchk(), sblastmbufchk(). - Assert the socket buffer lock in SBLINKRECORD(). - Break out various sbappend() functions into sbappend_locked() (and variations on that name) that assert the socket buffer lock. The !_locked() variations unconditionally lock the socket buffer before calling their _locked counterparts. Internally, make sure to call _locked() support routines, etc, if already holding the socket buffer lock. - Break out sbinsertoob() into sbinsertoob_locked() that asserts the socket buffer lock. sbinsertoob() unconditionally locks the socket buffer before calling sbinsertoob_locked(). - Break out sbflush() into sbflush_locked() that asserts the socket buffer lock. sbflush() unconditionally locks the socket buffer before calling sbflush_locked(). Update panic strings for new function names. - Break out sbdrop() into sbdrop_locked() that asserts the socket buffer lock. sbdrop() unconditionally locks the socket buffer before calling sbdrop_locked(). - Break out sbdroprecord() into sbdroprecord_locked() that asserts the socket buffer lock. sbdroprecord() unconditionally locks the socket buffer before calling sbdroprecord_locked(). - sofree() now calls socantsendmore_locked() and re-acquires the socket buffer lock on return. It also now calls sbrelease_locked(). - sorflush() now calls socantrcvmore_locked() and re-acquires the socket buffer lock on return. Clean up/mess up other behavior in sorflush() relating to the temporary stack copy of the socket buffer used with dom_dispose by more properly initializing the temporary copy, and selectively bzeroing/copying more carefully to prevent WITNESS from getting confused by improperly initialized mutexes. Annotate why that's necessary, or at least, needed. - soisconnected() now calls sbdrop_locked() before unlocking the socket buffer to avoid locking overhead. Some parts of this change were: Submitted by: sam Sponsored by: FreeBSD Foundation Obtained from: BSD/OS
2004-06-21 00:20:43 +00:00
SOCKBUF_UNLOCK(&so2->so_rcv);
The long-awaited mega-massive-network-code- cleanup. Part I. This commit includes the following changes: 1) Old-style (pr_usrreq()) protocols are no longer supported, the compatibility glue for them is deleted, and the kernel will panic on boot if any are compiled in. 2) Certain protocol entry points are modified to take a process structure, so they they can easily tell whether or not it is possible to sleep, and also to access credentials. 3) SS_PRIV is no more, and with it goes the SO_PRIVSTATE setsockopt() call. Protocols should use the process pointer they are now passed. 4) The PF_LOCAL and PF_ROUTE families have been updated to use the new style, as has the `raw' skeleton family. 5) PF_LOCAL sockets now obey the process's umask when creating a socket in the filesystem. As a result, LINT is now broken. I'm hoping that some enterprising hacker with a bit more time will either make the broken bits work (should be easy for netipx) or dike them out.
1997-04-27 20:01:29 +00:00
error = ENOBUFS;
Mechanical whitespace cleanup; parenthesize return values; other minor style nits.
2004-01-11 19:48:19 +00:00
}
Remove two simultaneous acquisitions of multiple unpcb locks from uipc_send in cases where only a global read lock is held by breaking them out and avoiding the unpcb lock acquire in the common case. This avoids deadlocks which manifested with X11, and should also marginally further improve performance. Reported by: sepotvin, brooks
2007-03-01 09:00:42 +00:00
if (nam != NULL) {
Decompose the global UNIX domain sockets rwlock into two different locks: a global list/counter/generation counter protected by a new mutex unp_list_lock, and a global linkage rwlock, unp_global_rwlock, which protects the connections between UNIX domain sockets. This eliminates conditional lock acquisition that was previously a property of the global lock being held over sonewconn() leading to a call to uipc_attach(), which also required the global lock, but couldn't rely on it as other paths existed to uipc_attach() that didn't hold it: now uipc_attach() uses only the list lock, which follows the linkage lock in the lock order. It may also reduce contention on the global lock for some workloads. Add global UNIX domain socket locks to hard-coded witness lock order. MFC after: 1 week Discussed with: kris
2009-03-08 21:48:29 +00:00
UNP_LINK_WLOCK_ASSERT();
Remove two simultaneous acquisitions of multiple unpcb locks from uipc_send in cases where only a global read lock is held by breaking them out and avoiding the unpcb lock acquire in the common case. This avoids deadlocks which manifested with X11, and should also marginally further improve performance. Reported by: sepotvin, brooks
2007-03-01 09:00:42 +00:00
UNP_PCB_LOCK(unp2);
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
unp_disconnect(unp, unp2);
Remove two simultaneous acquisitions of multiple unpcb locks from uipc_send in cases where only a global read lock is held by breaking them out and avoiding the unpcb lock acquire in the common case. This avoids deadlocks which manifested with X11, and should also marginally further improve performance. Reported by: sepotvin, brooks
2007-03-01 09:00:42 +00:00
UNP_PCB_UNLOCK(unp2);
}
UNP_PCB_UNLOCK(unp);
The long-awaited mega-massive-network-code- cleanup. Part I. This commit includes the following changes: 1) Old-style (pr_usrreq()) protocols are no longer supported, the compatibility glue for them is deleted, and the kernel will panic on boot if any are compiled in. 2) Certain protocol entry points are modified to take a process structure, so they they can easily tell whether or not it is possible to sleep, and also to access credentials. 3) SS_PRIV is no more, and with it goes the SO_PRIVSTATE setsockopt() call. Protocols should use the process pointer they are now passed. 4) The PF_LOCAL and PF_ROUTE families have been updated to use the new style, as has the `raw' skeleton family. 5) PF_LOCAL sockets now obey the process's umask when creating a socket in the filesystem. As a result, LINT is now broken. I'm hoping that some enterprising hacker with a bit more time will either make the broken bits work (should be easy for netipx) or dike them out.
1997-04-27 20:01:29 +00:00
break;
}
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
First cut at implementing SOCK_SEQPACKET support for UNIX (local) domain sockets. This allows for reliable bi-directional datagram communication over UNIX domain sockets, in contrast to SOCK_DGRAM (M:N, unreliable) or SOCK_STERAM (bi-directional bytestream). Largely, this reuses existing UNIX domain socket code. This allows applications requiring record- oriented semantics to do so reliably via local IPC. Some implementation notes (also present in XXX comments): - Currently we lack an sbappend variant able to do datagrams and control data without doing addresses, so we mark SOCK_SEQPACKET as PR_ADDR. Adding a new variant will solve this problem. - UNIX domain sockets on FreeBSD provide back-pressure/flow control notification for stream sockets by manipulating the send socket buffer's size during pru_send and pru_rcvd. This trick works less well for SOCK_SEQPACKET as sosend_generic() uses sb_hiwat not just to manage blocking, but also to determine maximum datagram size. Fixing this requires rethinking how back-pressure is done for SOCK_SEQPACKET; in the mean time, it's possible to get EMSGSIZE when buffers fill, instead of blocking. Discussed with: benl Reviewed by: bz, rpaulo MFC after: 3 months Sponsored by: Google
2009-10-05 14:49:16 +00:00
case SOCK_SEQPACKET:
The long-awaited mega-massive-network-code- cleanup. Part I. This commit includes the following changes: 1) Old-style (pr_usrreq()) protocols are no longer supported, the compatibility glue for them is deleted, and the kernel will panic on boot if any are compiled in. 2) Certain protocol entry points are modified to take a process structure, so they they can easily tell whether or not it is possible to sleep, and also to access credentials. 3) SS_PRIV is no more, and with it goes the SO_PRIVSTATE setsockopt() call. Protocols should use the process pointer they are now passed. 4) The PF_LOCAL and PF_ROUTE families have been updated to use the new style, as has the `raw' skeleton family. 5) PF_LOCAL sockets now obey the process's umask when creating a socket in the filesystem. As a result, LINT is now broken. I'm hoping that some enterprising hacker with a bit more time will either make the broken bits work (should be easy for netipx) or dike them out.
1997-04-27 20:01:29 +00:00
case SOCK_STREAM:
if ((so->so_state & SS_ISCONNECTED) == 0) {
Prefer NULL to 0 when testing and assigning pointer values.
2004-03-30 02:16:25 +00:00
if (nam != NULL) {
Decompose the global UNIX domain sockets rwlock into two different locks: a global list/counter/generation counter protected by a new mutex unp_list_lock, and a global linkage rwlock, unp_global_rwlock, which protects the connections between UNIX domain sockets. This eliminates conditional lock acquisition that was previously a property of the global lock being held over sonewconn() leading to a call to uipc_attach(), which also required the global lock, but couldn't rely on it as other paths existed to uipc_attach() that didn't hold it: now uipc_attach() uses only the list lock, which follows the linkage lock in the lock order. It may also reduce contention on the global lock for some workloads. Add global UNIX domain socket locks to hard-coded witness lock order. MFC after: 1 week Discussed with: kris
2009-03-08 21:48:29 +00:00
UNP_LINK_WLOCK_ASSERT();
KSE Milestone 2 Note ALL MODULES MUST BE RECOMPILED make the kernel aware that there are smaller units of scheduling than the process. (but only allow one thread per process at this time). This is functionally equivalent to teh previousl -current except that there is a thread associated with each process. Sorry john! (your next MFC will be a doosie!) Reviewed by: peter@freebsd.org, dillon@freebsd.org X-MFC after: ha ha ha ha
2001-09-12 08:38:13 +00:00
error = unp_connect(so, nam, td);
The long-awaited mega-massive-network-code- cleanup. Part I. This commit includes the following changes: 1) Old-style (pr_usrreq()) protocols are no longer supported, the compatibility glue for them is deleted, and the kernel will panic on boot if any are compiled in. 2) Certain protocol entry points are modified to take a process structure, so they they can easily tell whether or not it is possible to sleep, and also to access credentials. 3) SS_PRIV is no more, and with it goes the SO_PRIVSTATE setsockopt() call. Protocols should use the process pointer they are now passed. 4) The PF_LOCAL and PF_ROUTE families have been updated to use the new style, as has the `raw' skeleton family. 5) PF_LOCAL sockets now obey the process's umask when creating a socket in the filesystem. As a result, LINT is now broken. I'm hoping that some enterprising hacker with a bit more time will either make the broken bits work (should be easy for netipx) or dike them out.
1997-04-27 20:01:29 +00:00
if (error)
break; /* XXX */
} else {
error = ENOTCONN;
break;
}
Remove two simultaneous acquisitions of multiple unpcb locks from uipc_send in cases where only a global read lock is held by breaking them out and avoiding the unpcb lock acquire in the common case. This avoids deadlocks which manifested with X11, and should also marginally further improve performance. Reported by: sepotvin, brooks
2007-03-01 09:00:42 +00:00
}
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
Reduce periods of simultaneous acquisition of various socket buffer locks and the unplock during uipc_rcvd() and uipc_send() by caching certain values from one structure while its locks are held, and applying them to a second structure while its locks are held. If done carefully, this should be correct, and will reduce the amount of work done with the global unp lock held. Tested by: kris (earlier version)
2006-07-11 21:49:54 +00:00
/* Lockless read. */
The socket field so_state is used to hold a variety of socket related flags relating to several aspects of socket functionality. This change breaks out several bits relating to send and receive operation into a new per-socket buffer field, sb_state, in order to facilitate locking. This is required because, in order to provide more granular locking of sockets, different state fields have different locking properties. The following fields are moved to sb_state: SS_CANTRCVMORE (so_state) SS_CANTSENDMORE (so_state) SS_RCVATMARK (so_state) Rename respectively to: SBS_CANTRCVMORE (so_rcv.sb_state) SBS_CANTSENDMORE (so_snd.sb_state) SBS_RCVATMARK (so_rcv.sb_state) This facilitates locking by isolating fields to be located with other identically locked fields, and permits greater granularity in socket locking by avoiding storing fields with different locking semantics in the same short (avoiding locking conflicts). In the future, we may wish to coallesce sb_state and sb_flags; for the time being I leave them separate and there is no additional memory overhead due to the packing/alignment of shorts in the socket buffer structure.
2004-06-14 18:16:22 +00:00
if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
The long-awaited mega-massive-network-code- cleanup. Part I. This commit includes the following changes: 1) Old-style (pr_usrreq()) protocols are no longer supported, the compatibility glue for them is deleted, and the kernel will panic on boot if any are compiled in. 2) Certain protocol entry points are modified to take a process structure, so they they can easily tell whether or not it is possible to sleep, and also to access credentials. 3) SS_PRIV is no more, and with it goes the SO_PRIVSTATE setsockopt() call. Protocols should use the process pointer they are now passed. 4) The PF_LOCAL and PF_ROUTE families have been updated to use the new style, as has the `raw' skeleton family. 5) PF_LOCAL sockets now obey the process's umask when creating a socket in the filesystem. As a result, LINT is now broken. I'm hoping that some enterprising hacker with a bit more time will either make the broken bits work (should be easy for netipx) or dike them out.
1997-04-27 20:01:29 +00:00
error = EPIPE;
break;
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
}
Remove or update several stale comments. A bit of whitespace/style cleanup. Update copyright. MFC after: 3 days (applicable changes)
2008-10-03 09:01:55 +00:00
Close a race that occurs when using sendto() to connect and send on a UNIX domain socket at the same time as the remote host is closing the new connections as quickly as they open. Since the connect() and send() paths are non-atomic with respect to another, it is possible for the second thread's close() call to disconnect the two sockets as connect() returns, leading to the consumer (which plans to send()) with a NULL kernel pointer to its proposed peer. As a result, after acquiring the UNIX domain socket subsystem lock, we need to revalidate the connection pointers even though connect() has technically succeed, and reurn an error to say that there's no connection on which to perform the send. We might want to rethink the specific errno number, perhaps ECONNRESET would be better. PR: 100940 Reported by: Young Hyun <youngh at caida dot org> MFC after: 2 weeks MFC note: Some adaptation will be required
2006-07-31 23:00:05 +00:00
/*
* Because connect() and send() are non-atomic in a sendto()
* with a target address, it's possible that the socket will
* have disconnected before the send() can run. In that case
* return the slightly counter-intuitive but otherwise
* correct error that the socket is not connected.
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
*
Fix a deadlock in the getpeername() method for UNIX domain sockets. Instead of locking the local unp followed by the remote unp, use the same locking model as accept() and read lock the global link lock followed by the remote unp while fetching the remote sockaddr. Reported by: Mel Flynn mel.flynn of mailing.thruhere.net Reviewed by: rwatson MFC after: 1 week
2009-06-18 20:56:22 +00:00
* Locking here must be done carefully: the linkage lock
Clarify and update quite a few comments to reflect locking optimizations, the addition of unpcb refcounts, and bug fixes. Some of these fixes are appropriate for MFC. MFC after: 3 days
2007-05-11 12:10:45 +00:00
* prevents interconnections between unpcbs from changing, so
* we can traverse from unp to unp2 without acquiring unp's
* lock. Socket buffer locks follow unpcb locks, so we can
* acquire both remote and lock socket buffer locks.
Close a race that occurs when using sendto() to connect and send on a UNIX domain socket at the same time as the remote host is closing the new connections as quickly as they open. Since the connect() and send() paths are non-atomic with respect to another, it is possible for the second thread's close() call to disconnect the two sockets as connect() returns, leading to the consumer (which plans to send()) with a NULL kernel pointer to its proposed peer. As a result, after acquiring the UNIX domain socket subsystem lock, we need to revalidate the connection pointers even though connect() has technically succeed, and reurn an error to say that there's no connection on which to perform the send. We might want to rethink the specific errno number, perhaps ECONNRESET would be better. PR: 100940 Reported by: Young Hyun <youngh at caida dot org> MFC after: 2 weeks MFC note: Some adaptation will be required
2006-07-31 23:00:05 +00:00
*/
In uipc_send() and uipc_rcvd(), store unp->unp_conn pointer in unp2 while working with the second unpcb to make the code more clear.
2006-07-22 18:41:42 +00:00
unp2 = unp->unp_conn;
Close a race that occurs when using sendto() to connect and send on a UNIX domain socket at the same time as the remote host is closing the new connections as quickly as they open. Since the connect() and send() paths are non-atomic with respect to another, it is possible for the second thread's close() call to disconnect the two sockets as connect() returns, leading to the consumer (which plans to send()) with a NULL kernel pointer to its proposed peer. As a result, after acquiring the UNIX domain socket subsystem lock, we need to revalidate the connection pointers even though connect() has technically succeed, and reurn an error to say that there's no connection on which to perform the send. We might want to rethink the specific errno number, perhaps ECONNRESET would be better. PR: 100940 Reported by: Young Hyun <youngh at caida dot org> MFC after: 2 weeks MFC note: Some adaptation will be required
2006-07-31 23:00:05 +00:00
if (unp2 == NULL) {
error = ENOTCONN;
break;
}
In uipc_send() and uipc_rcvd(), store unp->unp_conn pointer in unp2 while working with the second unpcb to make the code more clear.
2006-07-22 18:41:42 +00:00
so2 = unp2->unp_socket;
Remove two simultaneous acquisitions of multiple unpcb locks from uipc_send in cases where only a global read lock is held by breaking them out and avoiding the unpcb lock acquire in the common case. This avoids deadlocks which manifested with X11, and should also marginally further improve performance. Reported by: sepotvin, brooks
2007-03-01 09:00:42 +00:00
UNP_PCB_LOCK(unp2);
Merge next step in socket buffer locking: - sowakeup() now asserts the socket buffer lock on entry. Move the call to KNOTE higher in sowakeup() so that it is made with the socket buffer lock held for consistency with other calls. Release the socket buffer lock prior to calling into pgsigio(), so_upcall(), or aio_swake(). Locking for this event management will need revisiting in the future, but this model avoids lock order reversals when upcalls into other subsystems result in socket/socket buffer operations. Assert that the socket buffer lock is not held at the end of the function. - Wrapper macros for sowakeup(), sorwakeup() and sowwakeup(), now have _locked versions which assert the socket buffer lock on entry. If a wakeup is required by sb_notify(), invoke sowakeup(); otherwise, unconditionally release the socket buffer lock. This results in the socket buffer lock being released whether a wakeup is required or not. - Break out socantsendmore() into socantsendmore_locked() that asserts the socket buffer lock. socantsendmore() unconditionally locks the socket buffer before calling socantsendmore_locked(). Note that both functions return with the socket buffer unlocked as socantsendmore_locked() calls sowwakeup_locked() which has the same properties. Assert that the socket buffer is unlocked on return. - Break out socantrcvmore() into socantrcvmore_locked() that asserts the socket buffer lock. socantrcvmore() unconditionally locks the socket buffer before calling socantrcvmore_locked(). Note that both functions return with the socket buffer unlocked as socantrcvmore_locked() calls sorwakeup_locked() which has similar properties. Assert that the socket buffer is unlocked on return. - Break out sbrelease() into a sbrelease_locked() that asserts the socket buffer lock. sbrelease() unconditionally locks the socket buffer before calling sbrelease_locked(). sbrelease_locked() now invokes sbflush_locked() instead of sbflush(). - Assert the socket buffer lock in socket buffer sanity check functions sblastrecordchk(), sblastmbufchk(). - Assert the socket buffer lock in SBLINKRECORD(). - Break out various sbappend() functions into sbappend_locked() (and variations on that name) that assert the socket buffer lock. The !_locked() variations unconditionally lock the socket buffer before calling their _locked counterparts. Internally, make sure to call _locked() support routines, etc, if already holding the socket buffer lock. - Break out sbinsertoob() into sbinsertoob_locked() that asserts the socket buffer lock. sbinsertoob() unconditionally locks the socket buffer before calling sbinsertoob_locked(). - Break out sbflush() into sbflush_locked() that asserts the socket buffer lock. sbflush() unconditionally locks the socket buffer before calling sbflush_locked(). Update panic strings for new function names. - Break out sbdrop() into sbdrop_locked() that asserts the socket buffer lock. sbdrop() unconditionally locks the socket buffer before calling sbdrop_locked(). - Break out sbdroprecord() into sbdroprecord_locked() that asserts the socket buffer lock. sbdroprecord() unconditionally locks the socket buffer before calling sbdroprecord_locked(). - sofree() now calls socantsendmore_locked() and re-acquires the socket buffer lock on return. It also now calls sbrelease_locked(). - sorflush() now calls socantrcvmore_locked() and re-acquires the socket buffer lock on return. Clean up/mess up other behavior in sorflush() relating to the temporary stack copy of the socket buffer used with dom_dispose by more properly initializing the temporary copy, and selectively bzeroing/copying more carefully to prevent WITNESS from getting confused by improperly initialized mutexes. Annotate why that's necessary, or at least, needed. - soisconnected() now calls sbdrop_locked() before unlocking the socket buffer to avoid locking overhead. Some parts of this change were: Submitted by: sam Sponsored by: FreeBSD Foundation Obtained from: BSD/OS
2004-06-21 00:20:43 +00:00
SOCKBUF_LOCK(&so2->so_rcv);
In uipc_send() and uipc_rcvd(), store unp->unp_conn pointer in unp2 while working with the second unpcb to make the code more clear.
2006-07-22 18:41:42 +00:00
if (unp2->unp_flags & UNP_WANTCRED) {
Implement unix(4) socket options LOCAL_CREDS and LOCAL_CONNWAIT. - Add unp_addsockcred() (for LOCAL_CREDS). - Add an argument to unp_connect2() to differentiate between PRU_CONNECT and PRU_CONNECT2. (for LOCAL_CONNWAIT) Obtained from: NetBSD (with some changes)
2005-04-13 00:01:46 +00:00
/*
Update comment.
2012-11-16 14:00:54 +00:00
* Credentials are passed only once on SOCK_STREAM
* and SOCK_SEQPACKET.
Implement unix(4) socket options LOCAL_CREDS and LOCAL_CONNWAIT. - Add unp_addsockcred() (for LOCAL_CREDS). - Add an argument to unp_connect2() to differentiate between PRU_CONNECT and PRU_CONNECT2. (for LOCAL_CONNWAIT) Obtained from: NetBSD (with some changes)
2005-04-13 00:01:46 +00:00
*/
In uipc_send() and uipc_rcvd(), store unp->unp_conn pointer in unp2 while working with the second unpcb to make the code more clear.
2006-07-22 18:41:42 +00:00
unp2->unp_flags &= ~UNP_WANTCRED;
Implement unix(4) socket options LOCAL_CREDS and LOCAL_CONNWAIT. - Add unp_addsockcred() (for LOCAL_CREDS). - Add an argument to unp_connect2() to differentiate between PRU_CONNECT and PRU_CONNECT2. (for LOCAL_CONNWAIT) Obtained from: NetBSD (with some changes)
2005-04-13 00:01:46 +00:00
control = unp_addsockcred(td, control);
}
The long-awaited mega-massive-network-code- cleanup. Part I. This commit includes the following changes: 1) Old-style (pr_usrreq()) protocols are no longer supported, the compatibility glue for them is deleted, and the kernel will panic on boot if any are compiled in. 2) Certain protocol entry points are modified to take a process structure, so they they can easily tell whether or not it is possible to sleep, and also to access credentials. 3) SS_PRIV is no more, and with it goes the SO_PRIVSTATE setsockopt() call. Protocols should use the process pointer they are now passed. 4) The PF_LOCAL and PF_ROUTE families have been updated to use the new style, as has the `raw' skeleton family. 5) PF_LOCAL sockets now obey the process's umask when creating a socket in the filesystem. As a result, LINT is now broken. I'm hoping that some enterprising hacker with a bit more time will either make the broken bits work (should be easy for netipx) or dike them out.
1997-04-27 20:01:29 +00:00
/*
Re-wrap and other minor formatting and punctuation fixes for UNIX domain socket comments.
2006-07-22 17:24:55 +00:00
* Send to paired receive port, and then reduce send buffer
* hiwater marks to maintain backpressure. Wake up readers.
The long-awaited mega-massive-network-code- cleanup. Part I. This commit includes the following changes: 1) Old-style (pr_usrreq()) protocols are no longer supported, the compatibility glue for them is deleted, and the kernel will panic on boot if any are compiled in. 2) Certain protocol entry points are modified to take a process structure, so they they can easily tell whether or not it is possible to sleep, and also to access credentials. 3) SS_PRIV is no more, and with it goes the SO_PRIVSTATE setsockopt() call. Protocols should use the process pointer they are now passed. 4) The PF_LOCAL and PF_ROUTE families have been updated to use the new style, as has the `raw' skeleton family. 5) PF_LOCAL sockets now obey the process's umask when creating a socket in the filesystem. As a result, LINT is now broken. I'm hoping that some enterprising hacker with a bit more time will either make the broken bits work (should be easy for netipx) or dike them out.
1997-04-27 20:01:29 +00:00
*/
First cut at implementing SOCK_SEQPACKET support for UNIX (local) domain sockets. This allows for reliable bi-directional datagram communication over UNIX domain sockets, in contrast to SOCK_DGRAM (M:N, unreliable) or SOCK_STERAM (bi-directional bytestream). Largely, this reuses existing UNIX domain socket code. This allows applications requiring record- oriented semantics to do so reliably via local IPC. Some implementation notes (also present in XXX comments): - Currently we lack an sbappend variant able to do datagrams and control data without doing addresses, so we mark SOCK_SEQPACKET as PR_ADDR. Adding a new variant will solve this problem. - UNIX domain sockets on FreeBSD provide back-pressure/flow control notification for stream sockets by manipulating the send socket buffer's size during pru_send and pru_rcvd. This trick works less well for SOCK_SEQPACKET as sosend_generic() uses sb_hiwat not just to manage blocking, but also to determine maximum datagram size. Fixing this requires rethinking how back-pressure is done for SOCK_SEQPACKET; in the mean time, it's possible to get EMSGSIZE when buffers fill, instead of blocking. Discussed with: benl Reviewed by: bz, rpaulo MFC after: 3 months Sponsored by: Google
2009-10-05 14:49:16 +00:00
switch (so->so_type) {
case SOCK_STREAM:
if (control != NULL) {
if (sbappendcontrol_locked(&so2->so_rcv, m,
control))
control = NULL;
} else
Make it possible for sbappend() to preserve M_NOTREADY on mbufs, just like sbappendstream() does. Although, M_NOTREADY may appear only on SOCK_STREAM sockets, due to sendfile(2) supporting only the latter, there is a corner case of AF_UNIX/SOCK_STREAM socket, that still uses records for the sake of control data, albeit being stream socket. Provide private version of m_clrprotoflags(), which understands PRUS_NOTREADY, similar to m_demote().
2016-01-08 19:03:20 +00:00
sbappend_locked(&so2->so_rcv, m, flags);
First cut at implementing SOCK_SEQPACKET support for UNIX (local) domain sockets. This allows for reliable bi-directional datagram communication over UNIX domain sockets, in contrast to SOCK_DGRAM (M:N, unreliable) or SOCK_STERAM (bi-directional bytestream). Largely, this reuses existing UNIX domain socket code. This allows applications requiring record- oriented semantics to do so reliably via local IPC. Some implementation notes (also present in XXX comments): - Currently we lack an sbappend variant able to do datagrams and control data without doing addresses, so we mark SOCK_SEQPACKET as PR_ADDR. Adding a new variant will solve this problem. - UNIX domain sockets on FreeBSD provide back-pressure/flow control notification for stream sockets by manipulating the send socket buffer's size during pru_send and pru_rcvd. This trick works less well for SOCK_SEQPACKET as sosend_generic() uses sb_hiwat not just to manage blocking, but also to determine maximum datagram size. Fixing this requires rethinking how back-pressure is done for SOCK_SEQPACKET; in the mean time, it's possible to get EMSGSIZE when buffers fill, instead of blocking. Discussed with: benl Reviewed by: bz, rpaulo MFC after: 3 months Sponsored by: Google
2009-10-05 14:49:16 +00:00
break;
case SOCK_SEQPACKET: {
const struct sockaddr *from;
from = &sun_noname;
Fix PR kern/185813 "SOCK_SEQPACKET AF_UNIX sockets with asymmetrical buffers drop packets". It was caused by a check for the space available in a sockbuf, but it was checking the wrong sockbuf. sys/sys/sockbuf.h sys/kern/uipc_sockbuf.c Add sbappendaddr_nospacecheck_locked(), which is just like sbappendaddr_locked but doesn't validate the receiving socket's space. Factor out common code into sbappendaddr_locked_internal(). We shouldn't simply make sbappendaddr_locked check the space and then call sbappendaddr_nospacecheck_locked, because that would cause the O(n) function m_length to be called twice. sys/kern/uipc_usrreq.c Use sbappendaddr_nospacecheck_locked for SOCK_SEQPACKET sockets, because the receiving sockbuf's size limit is irrelevant. tests/sys/kern/unix_seqpacket_test.c Now that 185813 is fixed, pipe_128k_8k fails intermittently due to 185812. Make it fail every time by adding a usleep after starting the writer thread and before starting the reader thread in test_pipe. That gives the writer time to fill up its send buffer. Also, clear the expected failure message due to 185813. It actually said "185812", but that was a typo. PR: kern/185813 Reviewed by: silence from freebsd-net@ and rwatson@ MFC after: 3 weeks Sponsored by: Spectra Logic Corporation
2014-03-06 20:24:15 +00:00
/*
* Don't check for space available in so2->so_rcv.
* Unix domain sockets only check for space in the
* sending sockbuf, and that check is performed one
* level up the stack.
*/
if (sbappendaddr_nospacecheck_locked(&so2->so_rcv,
from, m, control))
Prefer NULL to 0 when testing and assigning pointer values.
2004-03-30 02:16:25 +00:00
control = NULL;
First cut at implementing SOCK_SEQPACKET support for UNIX (local) domain sockets. This allows for reliable bi-directional datagram communication over UNIX domain sockets, in contrast to SOCK_DGRAM (M:N, unreliable) or SOCK_STERAM (bi-directional bytestream). Largely, this reuses existing UNIX domain socket code. This allows applications requiring record- oriented semantics to do so reliably via local IPC. Some implementation notes (also present in XXX comments): - Currently we lack an sbappend variant able to do datagrams and control data without doing addresses, so we mark SOCK_SEQPACKET as PR_ADDR. Adding a new variant will solve this problem. - UNIX domain sockets on FreeBSD provide back-pressure/flow control notification for stream sockets by manipulating the send socket buffer's size during pru_send and pru_rcvd. This trick works less well for SOCK_SEQPACKET as sosend_generic() uses sb_hiwat not just to manage blocking, but also to determine maximum datagram size. Fixing this requires rethinking how back-pressure is done for SOCK_SEQPACKET; in the mean time, it's possible to get EMSGSIZE when buffers fill, instead of blocking. Discussed with: benl Reviewed by: bz, rpaulo MFC after: 3 months Sponsored by: Google
2009-10-05 14:49:16 +00:00
break;
}
}
Replace 4.4BSD Lite's unix domain socket backpressure hack with a cleaner mechanism, based on the new SB_STOP sockbuf flag. The old hack dynamically changed the sending sockbuf's high water mark whenever adding or removing data from the receiving sockbuf. It worked for stream sockets, but it never worked for SOCK_SEQPACKET sockets because of their atomic nature. If the sockbuf was partially full, it might return EMSGSIZE instead of blocking. The new solution is based on DragonFlyBSD's fix from commit 3a6117bbe0ed6a87605c1e43e12a1438d8844380 on 2008-05-27. It adds an SB_STOP flag to sockbufs. Whenever uipc_send surpasses the socket's size limit, it sets SB_STOP on the sending sockbuf. sbspace() will then return 0 for that sockbuf, causing sosend_generic and friends to block. uipc_rcvd will likewise clear SB_STOP. There are two fringe benefits: uipc_{send,rcvd} no longer need to call chgsbsize() on every send and receive because they don't change the sockbuf's high water mark. Also, uipc_sense no longer needs to acquire the UIPC linkage lock, because it's simpler to compute the st_blksizes. There is one drawback: since sbspace() will only ever return 0 or the maximum, sosend_generic will allow the sockbuf to exceed its nominal maximum size by at most one packet of size less than the max. I don't think that's a serious problem. In fact, I'm not even positive that FreeBSD guarantees a socket will always stay within its nominal size limit. sys/sys/sockbuf.h Add the SB_STOP flag and adjust sbspace() sys/sys/unpcb.h Delete the obsolete unp_cc and unp_mbcnt fields from struct unpcb. sys/kern/uipc_usrreq.c Adjust uipc_rcvd, uipc_send, and uipc_sense to use the SB_STOP backpressure mechanism. Removing obsolete unpcb fields from db_show_unpcb. tests/sys/kern/unix_seqpacket_test.c Clear expected failures from ATF. Obtained from: DragonFly BSD PR: kern/185812 Reviewed by: silence from freebsd-net@ and rwatson@ MFC after: 3 weeks Sponsored by: Spectra Logic Corporation
2014-03-13 18:42:12 +00:00
mbcnt = so2->so_rcv.sb_mbcnt;
Merge from projects/sendfile: - Use KASSERT()s instead of panic(). - Use sbavail() instead of sb_cc. Sponsored by: Nginx, Inc. Sponsored by: Netflix
2014-11-12 10:17:46 +00:00
sbcc = sbavail(&so2->so_rcv);
if (sbcc)
sorwakeup_locked(so2);
else
SOCKBUF_UNLOCK(&so2->so_rcv);
Reduce periods of simultaneous acquisition of various socket buffer locks and the unplock during uipc_rcvd() and uipc_send() by caching certain values from one structure while its locks are held, and applying them to a second structure while its locks are held. If done carefully, this should be correct, and will reduce the amount of work done with the global unp lock held. Tested by: kris (earlier version)
2006-07-11 21:49:54 +00:00
Replace 4.4BSD Lite's unix domain socket backpressure hack with a cleaner mechanism, based on the new SB_STOP sockbuf flag. The old hack dynamically changed the sending sockbuf's high water mark whenever adding or removing data from the receiving sockbuf. It worked for stream sockets, but it never worked for SOCK_SEQPACKET sockets because of their atomic nature. If the sockbuf was partially full, it might return EMSGSIZE instead of blocking. The new solution is based on DragonFlyBSD's fix from commit 3a6117bbe0ed6a87605c1e43e12a1438d8844380 on 2008-05-27. It adds an SB_STOP flag to sockbufs. Whenever uipc_send surpasses the socket's size limit, it sets SB_STOP on the sending sockbuf. sbspace() will then return 0 for that sockbuf, causing sosend_generic and friends to block. uipc_rcvd will likewise clear SB_STOP. There are two fringe benefits: uipc_{send,rcvd} no longer need to call chgsbsize() on every send and receive because they don't change the sockbuf's high water mark. Also, uipc_sense no longer needs to acquire the UIPC linkage lock, because it's simpler to compute the st_blksizes. There is one drawback: since sbspace() will only ever return 0 or the maximum, sosend_generic will allow the sockbuf to exceed its nominal maximum size by at most one packet of size less than the max. I don't think that's a serious problem. In fact, I'm not even positive that FreeBSD guarantees a socket will always stay within its nominal size limit. sys/sys/sockbuf.h Add the SB_STOP flag and adjust sbspace() sys/sys/unpcb.h Delete the obsolete unp_cc and unp_mbcnt fields from struct unpcb. sys/kern/uipc_usrreq.c Adjust uipc_rcvd, uipc_send, and uipc_sense to use the SB_STOP backpressure mechanism. Removing obsolete unpcb fields from db_show_unpcb. tests/sys/kern/unix_seqpacket_test.c Clear expected failures from ATF. Obtained from: DragonFly BSD PR: kern/185812 Reviewed by: silence from freebsd-net@ and rwatson@ MFC after: 3 weeks Sponsored by: Spectra Logic Corporation
2014-03-13 18:42:12 +00:00
/*
* The PCB lock on unp2 protects the SB_STOP flag. Without it,
* it would be possible for uipc_rcvd to be called at this
* point, drain the receiving sockbuf, clear SB_STOP, and then
* we would set SB_STOP below. That could lead to an empty
* sockbuf having SB_STOP set
*/
Reduce periods of simultaneous acquisition of various socket buffer locks and the unplock during uipc_rcvd() and uipc_send() by caching certain values from one structure while its locks are held, and applying them to a second structure while its locks are held. If done carefully, this should be correct, and will reduce the amount of work done with the global unp lock held. Tested by: kris (earlier version)
2006-07-11 21:49:54 +00:00
SOCKBUF_LOCK(&so->so_snd);
Replace 4.4BSD Lite's unix domain socket backpressure hack with a cleaner mechanism, based on the new SB_STOP sockbuf flag. The old hack dynamically changed the sending sockbuf's high water mark whenever adding or removing data from the receiving sockbuf. It worked for stream sockets, but it never worked for SOCK_SEQPACKET sockets because of their atomic nature. If the sockbuf was partially full, it might return EMSGSIZE instead of blocking. The new solution is based on DragonFlyBSD's fix from commit 3a6117bbe0ed6a87605c1e43e12a1438d8844380 on 2008-05-27. It adds an SB_STOP flag to sockbufs. Whenever uipc_send surpasses the socket's size limit, it sets SB_STOP on the sending sockbuf. sbspace() will then return 0 for that sockbuf, causing sosend_generic and friends to block. uipc_rcvd will likewise clear SB_STOP. There are two fringe benefits: uipc_{send,rcvd} no longer need to call chgsbsize() on every send and receive because they don't change the sockbuf's high water mark. Also, uipc_sense no longer needs to acquire the UIPC linkage lock, because it's simpler to compute the st_blksizes. There is one drawback: since sbspace() will only ever return 0 or the maximum, sosend_generic will allow the sockbuf to exceed its nominal maximum size by at most one packet of size less than the max. I don't think that's a serious problem. In fact, I'm not even positive that FreeBSD guarantees a socket will always stay within its nominal size limit. sys/sys/sockbuf.h Add the SB_STOP flag and adjust sbspace() sys/sys/unpcb.h Delete the obsolete unp_cc and unp_mbcnt fields from struct unpcb. sys/kern/uipc_usrreq.c Adjust uipc_rcvd, uipc_send, and uipc_sense to use the SB_STOP backpressure mechanism. Removing obsolete unpcb fields from db_show_unpcb. tests/sys/kern/unix_seqpacket_test.c Clear expected failures from ATF. Obtained from: DragonFly BSD PR: kern/185812 Reviewed by: silence from freebsd-net@ and rwatson@ MFC after: 3 weeks Sponsored by: Spectra Logic Corporation
2014-03-13 18:42:12 +00:00
if (sbcc >= so->so_snd.sb_hiwat || mbcnt >= so->so_snd.sb_mbmax)
so->so_snd.sb_flags |= SB_STOP;
Add send buffer locking to uipc_send(). Without this locking a race can occur between a reader and a writer that results in a panic upon close, e.g., "panic: sbflush_locked: cc 4 || mb 0xffffff0052afa400 || mbcnt 0" Reviewed by: rwatson@ MFC after: 2 weeks
2004-12-22 20:28:46 +00:00
SOCKBUF_UNLOCK(&so->so_snd);
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
UNP_PCB_UNLOCK(unp2);
Prefer NULL to 0 when testing and assigning pointer values.
2004-03-30 02:16:25 +00:00
m = NULL;
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
break;
}
The long-awaited mega-massive-network-code- cleanup. Part I. This commit includes the following changes: 1) Old-style (pr_usrreq()) protocols are no longer supported, the compatibility glue for them is deleted, and the kernel will panic on boot if any are compiled in. 2) Certain protocol entry points are modified to take a process structure, so they they can easily tell whether or not it is possible to sleep, and also to access credentials. 3) SS_PRIV is no more, and with it goes the SO_PRIVSTATE setsockopt() call. Protocols should use the process pointer they are now passed. 4) The PF_LOCAL and PF_ROUTE families have been updated to use the new style, as has the `raw' skeleton family. 5) PF_LOCAL sockets now obey the process's umask when creating a socket in the filesystem. As a result, LINT is now broken. I'm hoping that some enterprising hacker with a bit more time will either make the broken bits work (should be easy for netipx) or dike them out.
1997-04-27 20:01:29 +00:00
/*
Remove or update several stale comments. A bit of whitespace/style cleanup. Update copyright. MFC after: 3 days (applicable changes)
2008-10-03 09:01:55 +00:00
* PRUS_EOF is equivalent to pru_send followed by pru_shutdown.
The long-awaited mega-massive-network-code- cleanup. Part I. This commit includes the following changes: 1) Old-style (pr_usrreq()) protocols are no longer supported, the compatibility glue for them is deleted, and the kernel will panic on boot if any are compiled in. 2) Certain protocol entry points are modified to take a process structure, so they they can easily tell whether or not it is possible to sleep, and also to access credentials. 3) SS_PRIV is no more, and with it goes the SO_PRIVSTATE setsockopt() call. Protocols should use the process pointer they are now passed. 4) The PF_LOCAL and PF_ROUTE families have been updated to use the new style, as has the `raw' skeleton family. 5) PF_LOCAL sockets now obey the process's umask when creating a socket in the filesystem. As a result, LINT is now broken. I'm hoping that some enterprising hacker with a bit more time will either make the broken bits work (should be easy for netipx) or dike them out.
1997-04-27 20:01:29 +00:00
*/
if (flags & PRUS_EOF) {
Remove two simultaneous acquisitions of multiple unpcb locks from uipc_send in cases where only a global read lock is held by breaking them out and avoiding the unpcb lock acquire in the common case. This avoids deadlocks which manifested with X11, and should also marginally further improve performance. Reported by: sepotvin, brooks
2007-03-01 09:00:42 +00:00
UNP_PCB_LOCK(unp);
The long-awaited mega-massive-network-code- cleanup. Part I. This commit includes the following changes: 1) Old-style (pr_usrreq()) protocols are no longer supported, the compatibility glue for them is deleted, and the kernel will panic on boot if any are compiled in. 2) Certain protocol entry points are modified to take a process structure, so they they can easily tell whether or not it is possible to sleep, and also to access credentials. 3) SS_PRIV is no more, and with it goes the SO_PRIVSTATE setsockopt() call. Protocols should use the process pointer they are now passed. 4) The PF_LOCAL and PF_ROUTE families have been updated to use the new style, as has the `raw' skeleton family. 5) PF_LOCAL sockets now obey the process's umask when creating a socket in the filesystem. As a result, LINT is now broken. I'm hoping that some enterprising hacker with a bit more time will either make the broken bits work (should be easy for netipx) or dike them out.
1997-04-27 20:01:29 +00:00
socantsendmore(so);
unp_shutdown(unp);
Remove two simultaneous acquisitions of multiple unpcb locks from uipc_send in cases where only a global read lock is held by breaking them out and avoiding the unpcb lock acquire in the common case. This avoids deadlocks which manifested with X11, and should also marginally further improve performance. Reported by: sepotvin, brooks
2007-03-01 09:00:42 +00:00
UNP_PCB_UNLOCK(unp);
The long-awaited mega-massive-network-code- cleanup. Part I. This commit includes the following changes: 1) Old-style (pr_usrreq()) protocols are no longer supported, the compatibility glue for them is deleted, and the kernel will panic on boot if any are compiled in. 2) Certain protocol entry points are modified to take a process structure, so they they can easily tell whether or not it is possible to sleep, and also to access credentials. 3) SS_PRIV is no more, and with it goes the SO_PRIVSTATE setsockopt() call. Protocols should use the process pointer they are now passed. 4) The PF_LOCAL and PF_ROUTE families have been updated to use the new style, as has the `raw' skeleton family. 5) PF_LOCAL sockets now obey the process's umask when creating a socket in the filesystem. As a result, LINT is now broken. I'm hoping that some enterprising hacker with a bit more time will either make the broken bits work (should be easy for netipx) or dike them out.
1997-04-27 20:01:29 +00:00
}
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
if ((nam != NULL) || (flags & PRUS_EOF))
Decompose the global UNIX domain sockets rwlock into two different locks: a global list/counter/generation counter protected by a new mutex unp_list_lock, and a global linkage rwlock, unp_global_rwlock, which protects the connections between UNIX domain sockets. This eliminates conditional lock acquisition that was previously a property of the global lock being held over sonewconn() leading to a call to uipc_attach(), which also required the global lock, but couldn't rely on it as other paths existed to uipc_attach() that didn't hold it: now uipc_attach() uses only the list lock, which follows the linkage lock in the lock order. It may also reduce contention on the global lock for some workloads. Add global UNIX domain socket locks to hard-coded witness lock order. MFC after: 1 week Discussed with: kris
2009-03-08 21:48:29 +00:00
UNP_LINK_WUNLOCK();
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
else
Decompose the global UNIX domain sockets rwlock into two different locks: a global list/counter/generation counter protected by a new mutex unp_list_lock, and a global linkage rwlock, unp_global_rwlock, which protects the connections between UNIX domain sockets. This eliminates conditional lock acquisition that was previously a property of the global lock being held over sonewconn() leading to a call to uipc_attach(), which also required the global lock, but couldn't rely on it as other paths existed to uipc_attach() that didn't hold it: now uipc_attach() uses only the list lock, which follows the linkage lock in the lock order. It may also reduce contention on the global lock for some workloads. Add global UNIX domain socket locks to hard-coded witness lock order. MFC after: 1 week Discussed with: kris
2009-03-08 21:48:29 +00:00
UNP_LINK_RUNLOCK();
The long-awaited mega-massive-network-code- cleanup. Part I. This commit includes the following changes: 1) Old-style (pr_usrreq()) protocols are no longer supported, the compatibility glue for them is deleted, and the kernel will panic on boot if any are compiled in. 2) Certain protocol entry points are modified to take a process structure, so they they can easily tell whether or not it is possible to sleep, and also to access credentials. 3) SS_PRIV is no more, and with it goes the SO_PRIVSTATE setsockopt() call. Protocols should use the process pointer they are now passed. 4) The PF_LOCAL and PF_ROUTE families have been updated to use the new style, as has the `raw' skeleton family. 5) PF_LOCAL sockets now obey the process's umask when creating a socket in the filesystem. As a result, LINT is now broken. I'm hoping that some enterprising hacker with a bit more time will either make the broken bits work (should be easy for netipx) or dike them out.
1997-04-27 20:01:29 +00:00
Prefer NULL to 0 when testing and assigning pointer values.
2004-03-30 02:16:25 +00:00
if (control != NULL && error != 0)
Rename unp_dispose_so() to unp_dispose(). It implements the dom_dispose method for local socket domain, so its name should match the method name.
2016-08-31 21:48:22 +00:00
unp_dispose_mbuf(control);
Fix descriptor leak provoked by KKIS.05051999.003b exploit code. unp_internalize() takes a reference to the descriptor. If the send fails after unp_internalize(), the control mbuf would be freed ophaning the reference. Tested in -CURRENT by: Pierre Beyssac <beyssac@enst.fr>
1999-05-10 18:09:39 +00:00
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
release:
Prefer NULL to 0 when testing and assigning pointer values.
2004-03-30 02:16:25 +00:00
if (control != NULL)
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
m_freem(control);
Fix two issues with not ready data in sockets (read: sendfile) in UNIX sockets. o Check that socket is still connected in uipc_ready(). If not we are responsible to free mbufs. o In uipc_send() if socket appears to be disconnected, but we are sending data with pending I/Os, don't free mbufs. Reported by: Kevin Bowling <kbowling llnw.com> Tested by: Kevin Bowling <kbowling llnw.com> PR: 222259 Reported by: Mark Martinec <Mark.Martinec ijs.si> MFC after: 3 days
2017-09-13 16:47:23 +00:00
/*
* In case of PRUS_NOTREADY, uipc_ready() is responsible
* for freeing memory.
*/
if (m != NULL && (flags & PRUS_NOTREADY) == 0)
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
m_freem(m);
Mechanical whitespace cleanup; parenthesize return values; other minor style nits.
2004-01-11 19:48:19 +00:00
return (error);
The long-awaited mega-massive-network-code- cleanup. Part I. This commit includes the following changes: 1) Old-style (pr_usrreq()) protocols are no longer supported, the compatibility glue for them is deleted, and the kernel will panic on boot if any are compiled in. 2) Certain protocol entry points are modified to take a process structure, so they they can easily tell whether or not it is possible to sleep, and also to access credentials. 3) SS_PRIV is no more, and with it goes the SO_PRIVSTATE setsockopt() call. Protocols should use the process pointer they are now passed. 4) The PF_LOCAL and PF_ROUTE families have been updated to use the new style, as has the `raw' skeleton family. 5) PF_LOCAL sockets now obey the process's umask when creating a socket in the filesystem. As a result, LINT is now broken. I'm hoping that some enterprising hacker with a bit more time will either make the broken bits work (should be easy for netipx) or dike them out.
1997-04-27 20:01:29 +00:00
}
Merge from projects/sendfile: Provide pru_ready for AF_LOCAL sockets. Local sockets sendsdata directly to the receive buffer of the peer, thus pru_ready also works on the peer socket. Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-11-30 13:40:58 +00:00
static int
uipc_ready(struct socket *so, struct mbuf *m, int count)
{
struct unpcb *unp, *unp2;
struct socket *so2;
int error;
unp = sotounpcb(so);
UNP_LINK_RLOCK();
Fix two issues with not ready data in sockets (read: sendfile) in UNIX sockets. o Check that socket is still connected in uipc_ready(). If not we are responsible to free mbufs. o In uipc_send() if socket appears to be disconnected, but we are sending data with pending I/Os, don't free mbufs. Reported by: Kevin Bowling <kbowling llnw.com> Tested by: Kevin Bowling <kbowling llnw.com> PR: 222259 Reported by: Mark Martinec <Mark.Martinec ijs.si> MFC after: 3 days
2017-09-13 16:47:23 +00:00
if ((unp2 = unp->unp_conn) == NULL) {
UNP_LINK_RUNLOCK();
for (int i = 0; i < count; i++)
m = m_free(m);
return (ECONNRESET);
}
Merge from projects/sendfile: Provide pru_ready for AF_LOCAL sockets. Local sockets sendsdata directly to the receive buffer of the peer, thus pru_ready also works on the peer socket. Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-11-30 13:40:58 +00:00
UNP_PCB_LOCK(unp2);
so2 = unp2->unp_socket;
SOCKBUF_LOCK(&so2->so_rcv);
if ((error = sbready(&so2->so_rcv, m, count)) == 0)
sorwakeup_locked(so2);
else
SOCKBUF_UNLOCK(&so2->so_rcv);
UNP_PCB_UNLOCK(unp2);
UNP_LINK_RUNLOCK();
return (error);
}
The long-awaited mega-massive-network-code- cleanup. Part I. This commit includes the following changes: 1) Old-style (pr_usrreq()) protocols are no longer supported, the compatibility glue for them is deleted, and the kernel will panic on boot if any are compiled in. 2) Certain protocol entry points are modified to take a process structure, so they they can easily tell whether or not it is possible to sleep, and also to access credentials. 3) SS_PRIV is no more, and with it goes the SO_PRIVSTATE setsockopt() call. Protocols should use the process pointer they are now passed. 4) The PF_LOCAL and PF_ROUTE families have been updated to use the new style, as has the `raw' skeleton family. 5) PF_LOCAL sockets now obey the process's umask when creating a socket in the filesystem. As a result, LINT is now broken. I'm hoping that some enterprising hacker with a bit more time will either make the broken bits work (should be easy for netipx) or dike them out.
1997-04-27 20:01:29 +00:00
static int
uipc_sense(struct socket *so, struct stat *sb)
{
Replace 4.4BSD Lite's unix domain socket backpressure hack with a cleaner mechanism, based on the new SB_STOP sockbuf flag. The old hack dynamically changed the sending sockbuf's high water mark whenever adding or removing data from the receiving sockbuf. It worked for stream sockets, but it never worked for SOCK_SEQPACKET sockets because of their atomic nature. If the sockbuf was partially full, it might return EMSGSIZE instead of blocking. The new solution is based on DragonFlyBSD's fix from commit 3a6117bbe0ed6a87605c1e43e12a1438d8844380 on 2008-05-27. It adds an SB_STOP flag to sockbufs. Whenever uipc_send surpasses the socket's size limit, it sets SB_STOP on the sending sockbuf. sbspace() will then return 0 for that sockbuf, causing sosend_generic and friends to block. uipc_rcvd will likewise clear SB_STOP. There are two fringe benefits: uipc_{send,rcvd} no longer need to call chgsbsize() on every send and receive because they don't change the sockbuf's high water mark. Also, uipc_sense no longer needs to acquire the UIPC linkage lock, because it's simpler to compute the st_blksizes. There is one drawback: since sbspace() will only ever return 0 or the maximum, sosend_generic will allow the sockbuf to exceed its nominal maximum size by at most one packet of size less than the max. I don't think that's a serious problem. In fact, I'm not even positive that FreeBSD guarantees a socket will always stay within its nominal size limit. sys/sys/sockbuf.h Add the SB_STOP flag and adjust sbspace() sys/sys/unpcb.h Delete the obsolete unp_cc and unp_mbcnt fields from struct unpcb. sys/kern/uipc_usrreq.c Adjust uipc_rcvd, uipc_send, and uipc_sense to use the SB_STOP backpressure mechanism. Removing obsolete unpcb fields from db_show_unpcb. tests/sys/kern/unix_seqpacket_test.c Clear expected failures from ATF. Obtained from: DragonFly BSD PR: kern/185812 Reviewed by: silence from freebsd-net@ and rwatson@ MFC after: 3 weeks Sponsored by: Spectra Logic Corporation
2014-03-13 18:42:12 +00:00
struct unpcb *unp;
The long-awaited mega-massive-network-code- cleanup. Part I. This commit includes the following changes: 1) Old-style (pr_usrreq()) protocols are no longer supported, the compatibility glue for them is deleted, and the kernel will panic on boot if any are compiled in. 2) Certain protocol entry points are modified to take a process structure, so they they can easily tell whether or not it is possible to sleep, and also to access credentials. 3) SS_PRIV is no more, and with it goes the SO_PRIVSTATE setsockopt() call. Protocols should use the process pointer they are now passed. 4) The PF_LOCAL and PF_ROUTE families have been updated to use the new style, as has the `raw' skeleton family. 5) PF_LOCAL sockets now obey the process's umask when creating a socket in the filesystem. As a result, LINT is now broken. I'm hoping that some enterprising hacker with a bit more time will either make the broken bits work (should be easy for netipx) or dike them out.
1997-04-27 20:01:29 +00:00
Always acquire the UNIX domain socket subsystem lock (UNP lock) before dereferencing sotounpcb() and checking its value, as so_pcb is protected by protocol locking, not subsystem locking. This prevents races during close() by one thread and use of ths socket in another. unp_bind() now assert the UNP lock, and uipc_bind() now acquires the lock around calls to unp_bind().
2004-08-16 04:41:03 +00:00
unp = sotounpcb(so);
Modify UNIX domain sockets to guarantee, and assume, that so_pcb is always defined for an in-use socket. This allows us to eliminate countless tests of whether so_pcb is non-NULL, eliminating dozens of error cases. For now, retain the call to sotryfree() in the uipc_abort() path, but this will eventually move to soabort(). These new assumptions should be largely correct, and will become more so as the socket/pcb reference model is fixed. Removing the notion that so_pcb can be non-NULL is a critical step towards further fine-graining of the UNIX domain socket locking, as the so_pcb reference no longer needs to be protected using locks, instead it is a property of the socket life cycle.
2006-03-17 13:52:57 +00:00
KASSERT(unp != NULL, ("uipc_sense: unp == NULL"));
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
The long-awaited mega-massive-network-code- cleanup. Part I. This commit includes the following changes: 1) Old-style (pr_usrreq()) protocols are no longer supported, the compatibility glue for them is deleted, and the kernel will panic on boot if any are compiled in. 2) Certain protocol entry points are modified to take a process structure, so they they can easily tell whether or not it is possible to sleep, and also to access credentials. 3) SS_PRIV is no more, and with it goes the SO_PRIVSTATE setsockopt() call. Protocols should use the process pointer they are now passed. 4) The PF_LOCAL and PF_ROUTE families have been updated to use the new style, as has the `raw' skeleton family. 5) PF_LOCAL sockets now obey the process's umask when creating a socket in the filesystem. As a result, LINT is now broken. I'm hoping that some enterprising hacker with a bit more time will either make the broken bits work (should be easy for netipx) or dike them out.
1997-04-27 20:01:29 +00:00
sb->st_blksize = so->so_snd.sb_hiwat;
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
UNP_PCB_LOCK(unp);
Second half of the dev_t cleanup. The big lines are: NODEV -> NULL NOUDEV -> NODEV udev_t -> dev_t udev2dev() -> findcdev() Various minor adjustments including handling of userland access to kernel space struct cdev etc.
2004-06-17 17:16:53 +00:00
sb->st_dev = NODEV;
The long-awaited mega-massive-network-code- cleanup. Part I. This commit includes the following changes: 1) Old-style (pr_usrreq()) protocols are no longer supported, the compatibility glue for them is deleted, and the kernel will panic on boot if any are compiled in. 2) Certain protocol entry points are modified to take a process structure, so they they can easily tell whether or not it is possible to sleep, and also to access credentials. 3) SS_PRIV is no more, and with it goes the SO_PRIVSTATE setsockopt() call. Protocols should use the process pointer they are now passed. 4) The PF_LOCAL and PF_ROUTE families have been updated to use the new style, as has the `raw' skeleton family. 5) PF_LOCAL sockets now obey the process's umask when creating a socket in the filesystem. As a result, LINT is now broken. I'm hoping that some enterprising hacker with a bit more time will either make the broken bits work (should be easy for netipx) or dike them out.
1997-04-27 20:01:29 +00:00
if (unp->unp_ino == 0)
Ensure that the made-up inode number for a Unix domain socket is persistent.
2002-12-25 07:59:39 +00:00
unp->unp_ino = (++unp_ino == 0) ? ++unp_ino : unp_ino;
The long-awaited mega-massive-network-code- cleanup. Part I. This commit includes the following changes: 1) Old-style (pr_usrreq()) protocols are no longer supported, the compatibility glue for them is deleted, and the kernel will panic on boot if any are compiled in. 2) Certain protocol entry points are modified to take a process structure, so they they can easily tell whether or not it is possible to sleep, and also to access credentials. 3) SS_PRIV is no more, and with it goes the SO_PRIVSTATE setsockopt() call. Protocols should use the process pointer they are now passed. 4) The PF_LOCAL and PF_ROUTE families have been updated to use the new style, as has the `raw' skeleton family. 5) PF_LOCAL sockets now obey the process's umask when creating a socket in the filesystem. As a result, LINT is now broken. I'm hoping that some enterprising hacker with a bit more time will either make the broken bits work (should be easy for netipx) or dike them out.
1997-04-27 20:01:29 +00:00
sb->st_ino = unp->unp_ino;
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
UNP_PCB_UNLOCK(unp);
The long-awaited mega-massive-network-code- cleanup. Part I. This commit includes the following changes: 1) Old-style (pr_usrreq()) protocols are no longer supported, the compatibility glue for them is deleted, and the kernel will panic on boot if any are compiled in. 2) Certain protocol entry points are modified to take a process structure, so they they can easily tell whether or not it is possible to sleep, and also to access credentials. 3) SS_PRIV is no more, and with it goes the SO_PRIVSTATE setsockopt() call. Protocols should use the process pointer they are now passed. 4) The PF_LOCAL and PF_ROUTE families have been updated to use the new style, as has the `raw' skeleton family. 5) PF_LOCAL sockets now obey the process's umask when creating a socket in the filesystem. As a result, LINT is now broken. I'm hoping that some enterprising hacker with a bit more time will either make the broken bits work (should be easy for netipx) or dike them out.
1997-04-27 20:01:29 +00:00
return (0);
}
static int
uipc_shutdown(struct socket *so)
{
Always acquire the UNIX domain socket subsystem lock (UNP lock) before dereferencing sotounpcb() and checking its value, as so_pcb is protected by protocol locking, not subsystem locking. This prevents races during close() by one thread and use of ths socket in another. unp_bind() now assert the UNP lock, and uipc_bind() now acquires the lock around calls to unp_bind().
2004-08-16 04:41:03 +00:00
struct unpcb *unp;
The long-awaited mega-massive-network-code- cleanup. Part I. This commit includes the following changes: 1) Old-style (pr_usrreq()) protocols are no longer supported, the compatibility glue for them is deleted, and the kernel will panic on boot if any are compiled in. 2) Certain protocol entry points are modified to take a process structure, so they they can easily tell whether or not it is possible to sleep, and also to access credentials. 3) SS_PRIV is no more, and with it goes the SO_PRIVSTATE setsockopt() call. Protocols should use the process pointer they are now passed. 4) The PF_LOCAL and PF_ROUTE families have been updated to use the new style, as has the `raw' skeleton family. 5) PF_LOCAL sockets now obey the process's umask when creating a socket in the filesystem. As a result, LINT is now broken. I'm hoping that some enterprising hacker with a bit more time will either make the broken bits work (should be easy for netipx) or dike them out.
1997-04-27 20:01:29 +00:00
Always acquire the UNIX domain socket subsystem lock (UNP lock) before dereferencing sotounpcb() and checking its value, as so_pcb is protected by protocol locking, not subsystem locking. This prevents races during close() by one thread and use of ths socket in another. unp_bind() now assert the UNP lock, and uipc_bind() now acquires the lock around calls to unp_bind().
2004-08-16 04:41:03 +00:00
unp = sotounpcb(so);
Modify UNIX domain sockets to guarantee, and assume, that so_pcb is always defined for an in-use socket. This allows us to eliminate countless tests of whether so_pcb is non-NULL, eliminating dozens of error cases. For now, retain the call to sotryfree() in the uipc_abort() path, but this will eventually move to soabort(). These new assumptions should be largely correct, and will become more so as the socket/pcb reference model is fixed. Removing the notion that so_pcb can be non-NULL is a critical step towards further fine-graining of the UNIX domain socket locking, as the so_pcb reference no longer needs to be protected using locks, instead it is a property of the socket life cycle.
2006-03-17 13:52:57 +00:00
KASSERT(unp != NULL, ("uipc_shutdown: unp == NULL"));
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
Decompose the global UNIX domain sockets rwlock into two different locks: a global list/counter/generation counter protected by a new mutex unp_list_lock, and a global linkage rwlock, unp_global_rwlock, which protects the connections between UNIX domain sockets. This eliminates conditional lock acquisition that was previously a property of the global lock being held over sonewconn() leading to a call to uipc_attach(), which also required the global lock, but couldn't rely on it as other paths existed to uipc_attach() that didn't hold it: now uipc_attach() uses only the list lock, which follows the linkage lock in the lock order. It may also reduce contention on the global lock for some workloads. Add global UNIX domain socket locks to hard-coded witness lock order. MFC after: 1 week Discussed with: kris
2009-03-08 21:48:29 +00:00
UNP_LINK_WLOCK();
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
UNP_PCB_LOCK(unp);
The long-awaited mega-massive-network-code- cleanup. Part I. This commit includes the following changes: 1) Old-style (pr_usrreq()) protocols are no longer supported, the compatibility glue for them is deleted, and the kernel will panic on boot if any are compiled in. 2) Certain protocol entry points are modified to take a process structure, so they they can easily tell whether or not it is possible to sleep, and also to access credentials. 3) SS_PRIV is no more, and with it goes the SO_PRIVSTATE setsockopt() call. Protocols should use the process pointer they are now passed. 4) The PF_LOCAL and PF_ROUTE families have been updated to use the new style, as has the `raw' skeleton family. 5) PF_LOCAL sockets now obey the process's umask when creating a socket in the filesystem. As a result, LINT is now broken. I'm hoping that some enterprising hacker with a bit more time will either make the broken bits work (should be easy for netipx) or dike them out.
1997-04-27 20:01:29 +00:00
socantsendmore(so);
unp_shutdown(unp);
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
UNP_PCB_UNLOCK(unp);
Decompose the global UNIX domain sockets rwlock into two different locks: a global list/counter/generation counter protected by a new mutex unp_list_lock, and a global linkage rwlock, unp_global_rwlock, which protects the connections between UNIX domain sockets. This eliminates conditional lock acquisition that was previously a property of the global lock being held over sonewconn() leading to a call to uipc_attach(), which also required the global lock, but couldn't rely on it as other paths existed to uipc_attach() that didn't hold it: now uipc_attach() uses only the list lock, which follows the linkage lock in the lock order. It may also reduce contention on the global lock for some workloads. Add global UNIX domain socket locks to hard-coded witness lock order. MFC after: 1 week Discussed with: kris
2009-03-08 21:48:29 +00:00
UNP_LINK_WUNLOCK();
Mechanical whitespace cleanup; parenthesize return values; other minor style nits.
2004-01-11 19:48:19 +00:00
return (0);
The long-awaited mega-massive-network-code- cleanup. Part I. This commit includes the following changes: 1) Old-style (pr_usrreq()) protocols are no longer supported, the compatibility glue for them is deleted, and the kernel will panic on boot if any are compiled in. 2) Certain protocol entry points are modified to take a process structure, so they they can easily tell whether or not it is possible to sleep, and also to access credentials. 3) SS_PRIV is no more, and with it goes the SO_PRIVSTATE setsockopt() call. Protocols should use the process pointer they are now passed. 4) The PF_LOCAL and PF_ROUTE families have been updated to use the new style, as has the `raw' skeleton family. 5) PF_LOCAL sockets now obey the process's umask when creating a socket in the filesystem. As a result, LINT is now broken. I'm hoping that some enterprising hacker with a bit more time will either make the broken bits work (should be easy for netipx) or dike them out.
1997-04-27 20:01:29 +00:00
}
static int
Fix all areas of the system (or at least all those in LINT) to avoid storing socket addresses in mbufs. (Socket buffers are the one exception.) A number of kernel APIs needed to get fixed in order to make this happen. Also, fix three protocol families which kept PCBs in mbufs to not malloc them instead. Delete some old compatibility cruft while we're at it, and add some new routines in the in_cksum family.
1997-08-16 19:16:27 +00:00
uipc_sockaddr(struct socket *so, struct sockaddr **nam)
The long-awaited mega-massive-network-code- cleanup. Part I. This commit includes the following changes: 1) Old-style (pr_usrreq()) protocols are no longer supported, the compatibility glue for them is deleted, and the kernel will panic on boot if any are compiled in. 2) Certain protocol entry points are modified to take a process structure, so they they can easily tell whether or not it is possible to sleep, and also to access credentials. 3) SS_PRIV is no more, and with it goes the SO_PRIVSTATE setsockopt() call. Protocols should use the process pointer they are now passed. 4) The PF_LOCAL and PF_ROUTE families have been updated to use the new style, as has the `raw' skeleton family. 5) PF_LOCAL sockets now obey the process's umask when creating a socket in the filesystem. As a result, LINT is now broken. I'm hoping that some enterprising hacker with a bit more time will either make the broken bits work (should be easy for netipx) or dike them out.
1997-04-27 20:01:29 +00:00
{
Always acquire the UNIX domain socket subsystem lock (UNP lock) before dereferencing sotounpcb() and checking its value, as so_pcb is protected by protocol locking, not subsystem locking. This prevents races during close() by one thread and use of ths socket in another. unp_bind() now assert the UNP lock, and uipc_bind() now acquires the lock around calls to unp_bind().
2004-08-16 04:41:03 +00:00
struct unpcb *unp;
Introduce a subsystem lock around UNIX domain sockets in order to protect global and allocated variables. This strategy is derived from work originally developed by BSDi for BSD/OS, and applied to FreeBSD by Sam Leffler: - Add unp_mtx, a global mutex which will protect all UNIX domain socket related variables, structures, etc. - Add UNP_LOCK(), UNP_UNLOCK(), UNP_LOCK_ASSERT() macros. - Acquire unp_mtx on entering most UNIX domain socket code, drop/re-acquire around calls into VFS, and release it on return. - Avoid performing sodupsockaddr() while holding the mutex, so in general move to allocating storage before acquiring the mutex to copy the data. - Make a stack copy of the xucred rather than copying out while holding unp_mtx. Copy the peer credential out after releasing the mutex. - Add additional assertions of vnode locks following VOP_CREATE(). A few notes: - Use of an sx lock for the file list mutex may cause problems with regard to unp_mtx when garbage collection passed file descriptors. - The locking in unp_pcblist() for sysctl monitoring is correct subject to the unpcb zone not returning memory for reuse by other subsystems (consistent with similar existing concerns). - Sam's version of this change, as with the BSD/OS version, made use of both a global lock and per-unpcb locks. However, in practice, the global lock covered all accesses, so I have simplified out the unpcb locks in the interest of getting this merged faster (reducing the overhead but not sacrificing granularity in most cases). We will want to explore possibilities for improving lock granularity in this code in the future. Submitted by: sam Sponsored by: FreeBSD Foundatiuon Obtained from: BSD/OS 5 snapshot provided by BSDi
2004-06-10 21:34:38 +00:00
const struct sockaddr *sa;
The long-awaited mega-massive-network-code- cleanup. Part I. This commit includes the following changes: 1) Old-style (pr_usrreq()) protocols are no longer supported, the compatibility glue for them is deleted, and the kernel will panic on boot if any are compiled in. 2) Certain protocol entry points are modified to take a process structure, so they they can easily tell whether or not it is possible to sleep, and also to access credentials. 3) SS_PRIV is no more, and with it goes the SO_PRIVSTATE setsockopt() call. Protocols should use the process pointer they are now passed. 4) The PF_LOCAL and PF_ROUTE families have been updated to use the new style, as has the `raw' skeleton family. 5) PF_LOCAL sockets now obey the process's umask when creating a socket in the filesystem. As a result, LINT is now broken. I'm hoping that some enterprising hacker with a bit more time will either make the broken bits work (should be easy for netipx) or dike them out.
1997-04-27 20:01:29 +00:00
Modify UNIX domain sockets to guarantee, and assume, that so_pcb is always defined for an in-use socket. This allows us to eliminate countless tests of whether so_pcb is non-NULL, eliminating dozens of error cases. For now, retain the call to sotryfree() in the uipc_abort() path, but this will eventually move to soabort(). These new assumptions should be largely correct, and will become more so as the socket/pcb reference model is fixed. Removing the notion that so_pcb can be non-NULL is a critical step towards further fine-graining of the UNIX domain socket locking, as the so_pcb reference no longer needs to be protected using locks, instead it is a property of the socket life cycle.
2006-03-17 13:52:57 +00:00
unp = sotounpcb(so);
KASSERT(unp != NULL, ("uipc_sockaddr: unp == NULL"));
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
Introduce a subsystem lock around UNIX domain sockets in order to protect global and allocated variables. This strategy is derived from work originally developed by BSDi for BSD/OS, and applied to FreeBSD by Sam Leffler: - Add unp_mtx, a global mutex which will protect all UNIX domain socket related variables, structures, etc. - Add UNP_LOCK(), UNP_UNLOCK(), UNP_LOCK_ASSERT() macros. - Acquire unp_mtx on entering most UNIX domain socket code, drop/re-acquire around calls into VFS, and release it on return. - Avoid performing sodupsockaddr() while holding the mutex, so in general move to allocating storage before acquiring the mutex to copy the data. - Make a stack copy of the xucred rather than copying out while holding unp_mtx. Copy the peer credential out after releasing the mutex. - Add additional assertions of vnode locks following VOP_CREATE(). A few notes: - Use of an sx lock for the file list mutex may cause problems with regard to unp_mtx when garbage collection passed file descriptors. - The locking in unp_pcblist() for sysctl monitoring is correct subject to the unpcb zone not returning memory for reuse by other subsystems (consistent with similar existing concerns). - Sam's version of this change, as with the BSD/OS version, made use of both a global lock and per-unpcb locks. However, in practice, the global lock covered all accesses, so I have simplified out the unpcb locks in the interest of getting this merged faster (reducing the overhead but not sacrificing granularity in most cases). We will want to explore possibilities for improving lock granularity in this code in the future. Submitted by: sam Sponsored by: FreeBSD Foundatiuon Obtained from: BSD/OS 5 snapshot provided by BSDi
2004-06-10 21:34:38 +00:00
*nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
UNP_PCB_LOCK(unp);
Prefer NULL to 0 when testing and assigning pointer values.
2004-03-30 02:16:25 +00:00
if (unp->unp_addr != NULL)
Introduce a subsystem lock around UNIX domain sockets in order to protect global and allocated variables. This strategy is derived from work originally developed by BSDi for BSD/OS, and applied to FreeBSD by Sam Leffler: - Add unp_mtx, a global mutex which will protect all UNIX domain socket related variables, structures, etc. - Add UNP_LOCK(), UNP_UNLOCK(), UNP_LOCK_ASSERT() macros. - Acquire unp_mtx on entering most UNIX domain socket code, drop/re-acquire around calls into VFS, and release it on return. - Avoid performing sodupsockaddr() while holding the mutex, so in general move to allocating storage before acquiring the mutex to copy the data. - Make a stack copy of the xucred rather than copying out while holding unp_mtx. Copy the peer credential out after releasing the mutex. - Add additional assertions of vnode locks following VOP_CREATE(). A few notes: - Use of an sx lock for the file list mutex may cause problems with regard to unp_mtx when garbage collection passed file descriptors. - The locking in unp_pcblist() for sysctl monitoring is correct subject to the unpcb zone not returning memory for reuse by other subsystems (consistent with similar existing concerns). - Sam's version of this change, as with the BSD/OS version, made use of both a global lock and per-unpcb locks. However, in practice, the global lock covered all accesses, so I have simplified out the unpcb locks in the interest of getting this merged faster (reducing the overhead but not sacrificing granularity in most cases). We will want to explore possibilities for improving lock granularity in this code in the future. Submitted by: sam Sponsored by: FreeBSD Foundatiuon Obtained from: BSD/OS 5 snapshot provided by BSDi
2004-06-10 21:34:38 +00:00
sa = (struct sockaddr *) unp->unp_addr;
Change uipc_sockaddr so that a sockaddr_un without a path is returned nam for an unbound socket instead of leaving nam untouched in that case. This way, the getsockname() output can be used to determine the address family of such sockets (AF_LOCAL). Reviewed by: iedowse Approved by: rwatson
2001-04-24 19:09:23 +00:00
else
Introduce a subsystem lock around UNIX domain sockets in order to protect global and allocated variables. This strategy is derived from work originally developed by BSDi for BSD/OS, and applied to FreeBSD by Sam Leffler: - Add unp_mtx, a global mutex which will protect all UNIX domain socket related variables, structures, etc. - Add UNP_LOCK(), UNP_UNLOCK(), UNP_LOCK_ASSERT() macros. - Acquire unp_mtx on entering most UNIX domain socket code, drop/re-acquire around calls into VFS, and release it on return. - Avoid performing sodupsockaddr() while holding the mutex, so in general move to allocating storage before acquiring the mutex to copy the data. - Make a stack copy of the xucred rather than copying out while holding unp_mtx. Copy the peer credential out after releasing the mutex. - Add additional assertions of vnode locks following VOP_CREATE(). A few notes: - Use of an sx lock for the file list mutex may cause problems with regard to unp_mtx when garbage collection passed file descriptors. - The locking in unp_pcblist() for sysctl monitoring is correct subject to the unpcb zone not returning memory for reuse by other subsystems (consistent with similar existing concerns). - Sam's version of this change, as with the BSD/OS version, made use of both a global lock and per-unpcb locks. However, in practice, the global lock covered all accesses, so I have simplified out the unpcb locks in the interest of getting this merged faster (reducing the overhead but not sacrificing granularity in most cases). We will want to explore possibilities for improving lock granularity in this code in the future. Submitted by: sam Sponsored by: FreeBSD Foundatiuon Obtained from: BSD/OS 5 snapshot provided by BSDi
2004-06-10 21:34:38 +00:00
sa = &sun_noname;
bcopy(sa, *nam, sa->sa_len);
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
UNP_PCB_UNLOCK(unp);
Mechanical whitespace cleanup; parenthesize return values; other minor style nits.
2004-01-11 19:48:19 +00:00
return (0);
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
}
Differentiate pr_usrreqs for stream and datagram UNIX domain sockets, and employ soreceive_dgram for the datagram case. MFC after: 3 months
2008-10-08 06:19:49 +00:00
static struct pr_usrreqs uipc_usrreqs_dgram = {
Initialize struct pr_userreqs in new/sparse style and fill in common default elements in net_init_domain(). This makes it possible to grep these structures and see any bogosities.
2004-11-08 14:44:54 +00:00
.pru_abort = uipc_abort,
.pru_accept = uipc_accept,
.pru_attach = uipc_attach,
.pru_bind = uipc_bind,
- Implement two new system calls: int bindat(int fd, int s, const struct sockaddr *addr, socklen_t addrlen); int connectat(int fd, int s, const struct sockaddr *name, socklen_t namelen); which allow to bind and connect respectively to a UNIX domain socket with a path relative to the directory associated with the given file descriptor 'fd'. - Add manual pages for the new syscalls. - Make the new syscalls available for processes in capability mode sandbox. - Add capability rights CAP_BINDAT and CAP_CONNECTAT that has to be present on the directory descriptor for the syscalls to work. - Update audit(4) to support those two new syscalls and to handle path in sockaddr_un structure relative to the given directory descriptor. - Update procstat(1) to recognize the new capability rights. - Document the new capability rights in cap_rights_limit(2). Sponsored by: The FreeBSD Foundation Discussed with: rwatson, jilles, kib, des
2013-03-02 21:11:30 +00:00
.pru_bindat = uipc_bindat,
Initialize struct pr_userreqs in new/sparse style and fill in common default elements in net_init_domain(). This makes it possible to grep these structures and see any bogosities.
2004-11-08 14:44:54 +00:00
.pru_connect = uipc_connect,
- Implement two new system calls: int bindat(int fd, int s, const struct sockaddr *addr, socklen_t addrlen); int connectat(int fd, int s, const struct sockaddr *name, socklen_t namelen); which allow to bind and connect respectively to a UNIX domain socket with a path relative to the directory associated with the given file descriptor 'fd'. - Add manual pages for the new syscalls. - Make the new syscalls available for processes in capability mode sandbox. - Add capability rights CAP_BINDAT and CAP_CONNECTAT that has to be present on the directory descriptor for the syscalls to work. - Update audit(4) to support those two new syscalls and to handle path in sockaddr_un structure relative to the given directory descriptor. - Update procstat(1) to recognize the new capability rights. - Document the new capability rights in cap_rights_limit(2). Sponsored by: The FreeBSD Foundation Discussed with: rwatson, jilles, kib, des
2013-03-02 21:11:30 +00:00
.pru_connectat = uipc_connectat,
Initialize struct pr_userreqs in new/sparse style and fill in common default elements in net_init_domain(). This makes it possible to grep these structures and see any bogosities.
2004-11-08 14:44:54 +00:00
.pru_connect2 = uipc_connect2,
.pru_detach = uipc_detach,
.pru_disconnect = uipc_disconnect,
.pru_listen = uipc_listen,
.pru_peeraddr = uipc_peeraddr,
.pru_rcvd = uipc_rcvd,
.pru_send = uipc_send,
.pru_sense = uipc_sense,
.pru_shutdown = uipc_shutdown,
.pru_sockaddr = uipc_sockaddr,
Differentiate pr_usrreqs for stream and datagram UNIX domain sockets, and employ soreceive_dgram for the datagram case. MFC after: 3 months
2008-10-08 06:19:49 +00:00
.pru_soreceive = soreceive_dgram,
.pru_close = uipc_close,
};
First cut at implementing SOCK_SEQPACKET support for UNIX (local) domain sockets. This allows for reliable bi-directional datagram communication over UNIX domain sockets, in contrast to SOCK_DGRAM (M:N, unreliable) or SOCK_STERAM (bi-directional bytestream). Largely, this reuses existing UNIX domain socket code. This allows applications requiring record- oriented semantics to do so reliably via local IPC. Some implementation notes (also present in XXX comments): - Currently we lack an sbappend variant able to do datagrams and control data without doing addresses, so we mark SOCK_SEQPACKET as PR_ADDR. Adding a new variant will solve this problem. - UNIX domain sockets on FreeBSD provide back-pressure/flow control notification for stream sockets by manipulating the send socket buffer's size during pru_send and pru_rcvd. This trick works less well for SOCK_SEQPACKET as sosend_generic() uses sb_hiwat not just to manage blocking, but also to determine maximum datagram size. Fixing this requires rethinking how back-pressure is done for SOCK_SEQPACKET; in the mean time, it's possible to get EMSGSIZE when buffers fill, instead of blocking. Discussed with: benl Reviewed by: bz, rpaulo MFC after: 3 months Sponsored by: Google
2009-10-05 14:49:16 +00:00
static struct pr_usrreqs uipc_usrreqs_seqpacket = {
.pru_abort = uipc_abort,
.pru_accept = uipc_accept,
.pru_attach = uipc_attach,
.pru_bind = uipc_bind,
- Implement two new system calls: int bindat(int fd, int s, const struct sockaddr *addr, socklen_t addrlen); int connectat(int fd, int s, const struct sockaddr *name, socklen_t namelen); which allow to bind and connect respectively to a UNIX domain socket with a path relative to the directory associated with the given file descriptor 'fd'. - Add manual pages for the new syscalls. - Make the new syscalls available for processes in capability mode sandbox. - Add capability rights CAP_BINDAT and CAP_CONNECTAT that has to be present on the directory descriptor for the syscalls to work. - Update audit(4) to support those two new syscalls and to handle path in sockaddr_un structure relative to the given directory descriptor. - Update procstat(1) to recognize the new capability rights. - Document the new capability rights in cap_rights_limit(2). Sponsored by: The FreeBSD Foundation Discussed with: rwatson, jilles, kib, des
2013-03-02 21:11:30 +00:00
.pru_bindat = uipc_bindat,
First cut at implementing SOCK_SEQPACKET support for UNIX (local) domain sockets. This allows for reliable bi-directional datagram communication over UNIX domain sockets, in contrast to SOCK_DGRAM (M:N, unreliable) or SOCK_STERAM (bi-directional bytestream). Largely, this reuses existing UNIX domain socket code. This allows applications requiring record- oriented semantics to do so reliably via local IPC. Some implementation notes (also present in XXX comments): - Currently we lack an sbappend variant able to do datagrams and control data without doing addresses, so we mark SOCK_SEQPACKET as PR_ADDR. Adding a new variant will solve this problem. - UNIX domain sockets on FreeBSD provide back-pressure/flow control notification for stream sockets by manipulating the send socket buffer's size during pru_send and pru_rcvd. This trick works less well for SOCK_SEQPACKET as sosend_generic() uses sb_hiwat not just to manage blocking, but also to determine maximum datagram size. Fixing this requires rethinking how back-pressure is done for SOCK_SEQPACKET; in the mean time, it's possible to get EMSGSIZE when buffers fill, instead of blocking. Discussed with: benl Reviewed by: bz, rpaulo MFC after: 3 months Sponsored by: Google
2009-10-05 14:49:16 +00:00
.pru_connect = uipc_connect,
- Implement two new system calls: int bindat(int fd, int s, const struct sockaddr *addr, socklen_t addrlen); int connectat(int fd, int s, const struct sockaddr *name, socklen_t namelen); which allow to bind and connect respectively to a UNIX domain socket with a path relative to the directory associated with the given file descriptor 'fd'. - Add manual pages for the new syscalls. - Make the new syscalls available for processes in capability mode sandbox. - Add capability rights CAP_BINDAT and CAP_CONNECTAT that has to be present on the directory descriptor for the syscalls to work. - Update audit(4) to support those two new syscalls and to handle path in sockaddr_un structure relative to the given directory descriptor. - Update procstat(1) to recognize the new capability rights. - Document the new capability rights in cap_rights_limit(2). Sponsored by: The FreeBSD Foundation Discussed with: rwatson, jilles, kib, des
2013-03-02 21:11:30 +00:00
.pru_connectat = uipc_connectat,
First cut at implementing SOCK_SEQPACKET support for UNIX (local) domain sockets. This allows for reliable bi-directional datagram communication over UNIX domain sockets, in contrast to SOCK_DGRAM (M:N, unreliable) or SOCK_STERAM (bi-directional bytestream). Largely, this reuses existing UNIX domain socket code. This allows applications requiring record- oriented semantics to do so reliably via local IPC. Some implementation notes (also present in XXX comments): - Currently we lack an sbappend variant able to do datagrams and control data without doing addresses, so we mark SOCK_SEQPACKET as PR_ADDR. Adding a new variant will solve this problem. - UNIX domain sockets on FreeBSD provide back-pressure/flow control notification for stream sockets by manipulating the send socket buffer's size during pru_send and pru_rcvd. This trick works less well for SOCK_SEQPACKET as sosend_generic() uses sb_hiwat not just to manage blocking, but also to determine maximum datagram size. Fixing this requires rethinking how back-pressure is done for SOCK_SEQPACKET; in the mean time, it's possible to get EMSGSIZE when buffers fill, instead of blocking. Discussed with: benl Reviewed by: bz, rpaulo MFC after: 3 months Sponsored by: Google
2009-10-05 14:49:16 +00:00
.pru_connect2 = uipc_connect2,
.pru_detach = uipc_detach,
.pru_disconnect = uipc_disconnect,
.pru_listen = uipc_listen,
.pru_peeraddr = uipc_peeraddr,
.pru_rcvd = uipc_rcvd,
.pru_send = uipc_send,
.pru_sense = uipc_sense,
.pru_shutdown = uipc_shutdown,
.pru_sockaddr = uipc_sockaddr,
.pru_soreceive = soreceive_generic, /* XXX: or...? */
.pru_close = uipc_close,
};
Differentiate pr_usrreqs for stream and datagram UNIX domain sockets, and employ soreceive_dgram for the datagram case. MFC after: 3 months
2008-10-08 06:19:49 +00:00
static struct pr_usrreqs uipc_usrreqs_stream = {
.pru_abort = uipc_abort,
.pru_accept = uipc_accept,
.pru_attach = uipc_attach,
.pru_bind = uipc_bind,
- Implement two new system calls: int bindat(int fd, int s, const struct sockaddr *addr, socklen_t addrlen); int connectat(int fd, int s, const struct sockaddr *name, socklen_t namelen); which allow to bind and connect respectively to a UNIX domain socket with a path relative to the directory associated with the given file descriptor 'fd'. - Add manual pages for the new syscalls. - Make the new syscalls available for processes in capability mode sandbox. - Add capability rights CAP_BINDAT and CAP_CONNECTAT that has to be present on the directory descriptor for the syscalls to work. - Update audit(4) to support those two new syscalls and to handle path in sockaddr_un structure relative to the given directory descriptor. - Update procstat(1) to recognize the new capability rights. - Document the new capability rights in cap_rights_limit(2). Sponsored by: The FreeBSD Foundation Discussed with: rwatson, jilles, kib, des
2013-03-02 21:11:30 +00:00
.pru_bindat = uipc_bindat,
Differentiate pr_usrreqs for stream and datagram UNIX domain sockets, and employ soreceive_dgram for the datagram case. MFC after: 3 months
2008-10-08 06:19:49 +00:00
.pru_connect = uipc_connect,
- Implement two new system calls: int bindat(int fd, int s, const struct sockaddr *addr, socklen_t addrlen); int connectat(int fd, int s, const struct sockaddr *name, socklen_t namelen); which allow to bind and connect respectively to a UNIX domain socket with a path relative to the directory associated with the given file descriptor 'fd'. - Add manual pages for the new syscalls. - Make the new syscalls available for processes in capability mode sandbox. - Add capability rights CAP_BINDAT and CAP_CONNECTAT that has to be present on the directory descriptor for the syscalls to work. - Update audit(4) to support those two new syscalls and to handle path in sockaddr_un structure relative to the given directory descriptor. - Update procstat(1) to recognize the new capability rights. - Document the new capability rights in cap_rights_limit(2). Sponsored by: The FreeBSD Foundation Discussed with: rwatson, jilles, kib, des
2013-03-02 21:11:30 +00:00
.pru_connectat = uipc_connectat,
Differentiate pr_usrreqs for stream and datagram UNIX domain sockets, and employ soreceive_dgram for the datagram case. MFC after: 3 months
2008-10-08 06:19:49 +00:00
.pru_connect2 = uipc_connect2,
.pru_detach = uipc_detach,
.pru_disconnect = uipc_disconnect,
.pru_listen = uipc_listen,
.pru_peeraddr = uipc_peeraddr,
.pru_rcvd = uipc_rcvd,
.pru_send = uipc_send,
Merge from projects/sendfile: Provide pru_ready for AF_LOCAL sockets. Local sockets sendsdata directly to the receive buffer of the peer, thus pru_ready also works on the peer socket. Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-11-30 13:40:58 +00:00
.pru_ready = uipc_ready,
Differentiate pr_usrreqs for stream and datagram UNIX domain sockets, and employ soreceive_dgram for the datagram case. MFC after: 3 months
2008-10-08 06:19:49 +00:00
.pru_sense = uipc_sense,
.pru_shutdown = uipc_shutdown,
.pru_sockaddr = uipc_sockaddr,
.pru_soreceive = soreceive_generic,
Change semantics of socket close and detach. Add a new protocol switch function, pru_close, to notify protocols that the file descriptor or other consumer of a socket is closing the socket. pru_abort is now a notification of close also, and no longer detaches. pru_detach is no longer used to notify of close, and will be called during socket tear-down by sofree() when all references to a socket evaporate after an earlier call to abort or close the socket. This means detach is now an unconditional teardown of a socket, whereas previously sockets could persist after detach of the protocol retained a reference. This faciliates sharing mutexes between layers of the network stack as the mutex is required during the checking and removal of references at the head of sofree(). With this change, pru_detach can now assume that the mutex will no longer be required by the socket layer after completion, whereas before this was not necessarily true. Reviewed by: gnn
2006-07-21 17:11:15 +00:00
.pru_close = uipc_close,
The long-awaited mega-massive-network-code- cleanup. Part I. This commit includes the following changes: 1) Old-style (pr_usrreq()) protocols are no longer supported, the compatibility glue for them is deleted, and the kernel will panic on boot if any are compiled in. 2) Certain protocol entry points are modified to take a process structure, so they they can easily tell whether or not it is possible to sleep, and also to access credentials. 3) SS_PRIV is no more, and with it goes the SO_PRIVSTATE setsockopt() call. Protocols should use the process pointer they are now passed. 4) The PF_LOCAL and PF_ROUTE families have been updated to use the new style, as has the `raw' skeleton family. 5) PF_LOCAL sockets now obey the process's umask when creating a socket in the filesystem. As a result, LINT is now broken. I'm hoping that some enterprising hacker with a bit more time will either make the broken bits work (should be easy for netipx) or dike them out.
1997-04-27 20:01:29 +00:00
};
Implement a LOCAL_PEERCRED socket option which returns a `struct xucred` with the credentials of the connected peer. Obviously this only works (and makes sense) on SOCK_STREAM sockets. This works for both the connect(2) and listen(2) callers. There is precise documentation of the semantics in unix(4). Reviewed by: dwmalone (eyeballed)
2001-08-17 22:01:18 +00:00
Further minor cleanups to UNIX domain sockets: - Staticize and locally prototype functions uipc_ctloutput(), unp_dispose(), unp_init(), and unp_externalize(), none of which have been required outside of uipc_usrreq.c since uipc_proto.c was removed. - Remove stale prototype for uipc_usrreq(), which has not existed in the code since 1997 - Forward declare and staticize uipc_usrreqs structure in uipc_usrreq.c and not un.h. - Comment on why uipc_connect2() is still non-static -- it is used directly by fifofs. - Remove stale comments, tidy up whitespace. MFC after: 3 days (where applicable)
2008-10-03 13:01:56 +00:00
static int
style(9)-ize function headers, remove use of 'register'. MFC after: 3 days
2005-02-20 23:22:13 +00:00
uipc_ctloutput(struct socket *so, struct sockopt *sopt)
Implement a LOCAL_PEERCRED socket option which returns a `struct xucred` with the credentials of the connected peer. Obviously this only works (and makes sense) on SOCK_STREAM sockets. This works for both the connect(2) and listen(2) callers. There is precise documentation of the semantics in unix(4). Reviewed by: dwmalone (eyeballed)
2001-08-17 22:01:18 +00:00
{
Always acquire the UNIX domain socket subsystem lock (UNP lock) before dereferencing sotounpcb() and checking its value, as so_pcb is protected by protocol locking, not subsystem locking. This prevents races during close() by one thread and use of ths socket in another. unp_bind() now assert the UNP lock, and uipc_bind() now acquires the lock around calls to unp_bind().
2004-08-16 04:41:03 +00:00
struct unpcb *unp;
Introduce a subsystem lock around UNIX domain sockets in order to protect global and allocated variables. This strategy is derived from work originally developed by BSDi for BSD/OS, and applied to FreeBSD by Sam Leffler: - Add unp_mtx, a global mutex which will protect all UNIX domain socket related variables, structures, etc. - Add UNP_LOCK(), UNP_UNLOCK(), UNP_LOCK_ASSERT() macros. - Acquire unp_mtx on entering most UNIX domain socket code, drop/re-acquire around calls into VFS, and release it on return. - Avoid performing sodupsockaddr() while holding the mutex, so in general move to allocating storage before acquiring the mutex to copy the data. - Make a stack copy of the xucred rather than copying out while holding unp_mtx. Copy the peer credential out after releasing the mutex. - Add additional assertions of vnode locks following VOP_CREATE(). A few notes: - Use of an sx lock for the file list mutex may cause problems with regard to unp_mtx when garbage collection passed file descriptors. - The locking in unp_pcblist() for sysctl monitoring is correct subject to the unpcb zone not returning memory for reuse by other subsystems (consistent with similar existing concerns). - Sam's version of this change, as with the BSD/OS version, made use of both a global lock and per-unpcb locks. However, in practice, the global lock covered all accesses, so I have simplified out the unpcb locks in the interest of getting this merged faster (reducing the overhead but not sacrificing granularity in most cases). We will want to explore possibilities for improving lock granularity in this code in the future. Submitted by: sam Sponsored by: FreeBSD Foundatiuon Obtained from: BSD/OS 5 snapshot provided by BSDi
2004-06-10 21:34:38 +00:00
struct xucred xu;
Implement unix(4) socket options LOCAL_CREDS and LOCAL_CONNWAIT. - Add unp_addsockcred() (for LOCAL_CREDS). - Add an argument to unp_connect2() to differentiate between PRU_CONNECT and PRU_CONNECT2. (for LOCAL_CONNWAIT) Obtained from: NetBSD (with some changes)
2005-04-13 00:01:46 +00:00
int error, optval;
Check sopt_level in uipc_ctloutput() and return early if it is non-zero. This prevents unintended consequnces when an application calls things like setsockopt(x, SOL_SOCKET, SO_REUSEADDR, ...) on a Unix domain socket.
2005-04-20 02:57:56 +00:00
if (sopt->sopt_level != 0)
return (EINVAL);
Implement unix(4) socket options LOCAL_CREDS and LOCAL_CONNWAIT. - Add unp_addsockcred() (for LOCAL_CREDS). - Add an argument to unp_connect2() to differentiate between PRU_CONNECT and PRU_CONNECT2. (for LOCAL_CONNWAIT) Obtained from: NetBSD (with some changes)
2005-04-13 00:01:46 +00:00
unp = sotounpcb(so);
Modify UNIX domain sockets to guarantee, and assume, that so_pcb is always defined for an in-use socket. This allows us to eliminate countless tests of whether so_pcb is non-NULL, eliminating dozens of error cases. For now, retain the call to sotryfree() in the uipc_abort() path, but this will eventually move to soabort(). These new assumptions should be largely correct, and will become more so as the socket/pcb reference model is fixed. Removing the notion that so_pcb can be non-NULL is a critical step towards further fine-graining of the UNIX domain socket locking, as the so_pcb reference no longer needs to be protected using locks, instead it is a property of the socket life cycle.
2006-03-17 13:52:57 +00:00
KASSERT(unp != NULL, ("uipc_ctloutput: unp == NULL"));
Implement unix(4) socket options LOCAL_CREDS and LOCAL_CONNWAIT. - Add unp_addsockcred() (for LOCAL_CREDS). - Add an argument to unp_connect2() to differentiate between PRU_CONNECT and PRU_CONNECT2. (for LOCAL_CONNWAIT) Obtained from: NetBSD (with some changes)
2005-04-13 00:01:46 +00:00
error = 0;
Implement a LOCAL_PEERCRED socket option which returns a `struct xucred` with the credentials of the connected peer. Obviously this only works (and makes sense) on SOCK_STREAM sockets. This works for both the connect(2) and listen(2) callers. There is precise documentation of the semantics in unix(4). Reviewed by: dwmalone (eyeballed)
2001-08-17 22:01:18 +00:00
switch (sopt->sopt_dir) {
case SOPT_GET:
switch (sopt->sopt_name) {
case LOCAL_PEERCRED:
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
UNP_PCB_LOCK(unp);
Implement a LOCAL_PEERCRED socket option which returns a `struct xucred` with the credentials of the connected peer. Obviously this only works (and makes sense) on SOCK_STREAM sockets. This works for both the connect(2) and listen(2) callers. There is precise documentation of the semantics in unix(4). Reviewed by: dwmalone (eyeballed)
2001-08-17 22:01:18 +00:00
if (unp->unp_flags & UNP_HAVEPC)
Introduce a subsystem lock around UNIX domain sockets in order to protect global and allocated variables. This strategy is derived from work originally developed by BSDi for BSD/OS, and applied to FreeBSD by Sam Leffler: - Add unp_mtx, a global mutex which will protect all UNIX domain socket related variables, structures, etc. - Add UNP_LOCK(), UNP_UNLOCK(), UNP_LOCK_ASSERT() macros. - Acquire unp_mtx on entering most UNIX domain socket code, drop/re-acquire around calls into VFS, and release it on return. - Avoid performing sodupsockaddr() while holding the mutex, so in general move to allocating storage before acquiring the mutex to copy the data. - Make a stack copy of the xucred rather than copying out while holding unp_mtx. Copy the peer credential out after releasing the mutex. - Add additional assertions of vnode locks following VOP_CREATE(). A few notes: - Use of an sx lock for the file list mutex may cause problems with regard to unp_mtx when garbage collection passed file descriptors. - The locking in unp_pcblist() for sysctl monitoring is correct subject to the unpcb zone not returning memory for reuse by other subsystems (consistent with similar existing concerns). - Sam's version of this change, as with the BSD/OS version, made use of both a global lock and per-unpcb locks. However, in practice, the global lock covered all accesses, so I have simplified out the unpcb locks in the interest of getting this merged faster (reducing the overhead but not sacrificing granularity in most cases). We will want to explore possibilities for improving lock granularity in this code in the future. Submitted by: sam Sponsored by: FreeBSD Foundatiuon Obtained from: BSD/OS 5 snapshot provided by BSDi
2004-06-10 21:34:38 +00:00
xu = unp->unp_peercred;
Implement a LOCAL_PEERCRED socket option which returns a `struct xucred` with the credentials of the connected peer. Obviously this only works (and makes sense) on SOCK_STREAM sockets. This works for both the connect(2) and listen(2) callers. There is precise documentation of the semantics in unix(4). Reviewed by: dwmalone (eyeballed)
2001-08-17 22:01:18 +00:00
else {
if (so->so_type == SOCK_STREAM)
error = ENOTCONN;
else
error = EINVAL;
}
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
UNP_PCB_UNLOCK(unp);
Introduce a subsystem lock around UNIX domain sockets in order to protect global and allocated variables. This strategy is derived from work originally developed by BSDi for BSD/OS, and applied to FreeBSD by Sam Leffler: - Add unp_mtx, a global mutex which will protect all UNIX domain socket related variables, structures, etc. - Add UNP_LOCK(), UNP_UNLOCK(), UNP_LOCK_ASSERT() macros. - Acquire unp_mtx on entering most UNIX domain socket code, drop/re-acquire around calls into VFS, and release it on return. - Avoid performing sodupsockaddr() while holding the mutex, so in general move to allocating storage before acquiring the mutex to copy the data. - Make a stack copy of the xucred rather than copying out while holding unp_mtx. Copy the peer credential out after releasing the mutex. - Add additional assertions of vnode locks following VOP_CREATE(). A few notes: - Use of an sx lock for the file list mutex may cause problems with regard to unp_mtx when garbage collection passed file descriptors. - The locking in unp_pcblist() for sysctl monitoring is correct subject to the unpcb zone not returning memory for reuse by other subsystems (consistent with similar existing concerns). - Sam's version of this change, as with the BSD/OS version, made use of both a global lock and per-unpcb locks. However, in practice, the global lock covered all accesses, so I have simplified out the unpcb locks in the interest of getting this merged faster (reducing the overhead but not sacrificing granularity in most cases). We will want to explore possibilities for improving lock granularity in this code in the future. Submitted by: sam Sponsored by: FreeBSD Foundatiuon Obtained from: BSD/OS 5 snapshot provided by BSDi
2004-06-10 21:34:38 +00:00
if (error == 0)
error = sooptcopyout(sopt, &xu, sizeof(xu));
Implement a LOCAL_PEERCRED socket option which returns a `struct xucred` with the credentials of the connected peer. Obviously this only works (and makes sense) on SOCK_STREAM sockets. This works for both the connect(2) and listen(2) callers. There is precise documentation of the semantics in unix(4). Reviewed by: dwmalone (eyeballed)
2001-08-17 22:01:18 +00:00
break;
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
Implement unix(4) socket options LOCAL_CREDS and LOCAL_CONNWAIT. - Add unp_addsockcred() (for LOCAL_CREDS). - Add an argument to unp_connect2() to differentiate between PRU_CONNECT and PRU_CONNECT2. (for LOCAL_CONNWAIT) Obtained from: NetBSD (with some changes)
2005-04-13 00:01:46 +00:00
case LOCAL_CREDS:
Correct typos in comments. MFC after: 3 weeks
2008-01-10 12:29:12 +00:00
/* Unlocked read. */
Implement unix(4) socket options LOCAL_CREDS and LOCAL_CONNWAIT. - Add unp_addsockcred() (for LOCAL_CREDS). - Add an argument to unp_connect2() to differentiate between PRU_CONNECT and PRU_CONNECT2. (for LOCAL_CONNWAIT) Obtained from: NetBSD (with some changes)
2005-04-13 00:01:46 +00:00
optval = unp->unp_flags & UNP_WANTCRED ? 1 : 0;
error = sooptcopyout(sopt, &optval, sizeof(optval));
break;
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
Implement unix(4) socket options LOCAL_CREDS and LOCAL_CONNWAIT. - Add unp_addsockcred() (for LOCAL_CREDS). - Add an argument to unp_connect2() to differentiate between PRU_CONNECT and PRU_CONNECT2. (for LOCAL_CONNWAIT) Obtained from: NetBSD (with some changes)
2005-04-13 00:01:46 +00:00
case LOCAL_CONNWAIT:
Correct typos in comments. MFC after: 3 weeks
2008-01-10 12:29:12 +00:00
/* Unlocked read. */
Implement unix(4) socket options LOCAL_CREDS and LOCAL_CONNWAIT. - Add unp_addsockcred() (for LOCAL_CREDS). - Add an argument to unp_connect2() to differentiate between PRU_CONNECT and PRU_CONNECT2. (for LOCAL_CONNWAIT) Obtained from: NetBSD (with some changes)
2005-04-13 00:01:46 +00:00
optval = unp->unp_flags & UNP_CONNWAIT ? 1 : 0;
error = sooptcopyout(sopt, &optval, sizeof(optval));
break;
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
Implement a LOCAL_PEERCRED socket option which returns a `struct xucred` with the credentials of the connected peer. Obviously this only works (and makes sense) on SOCK_STREAM sockets. This works for both the connect(2) and listen(2) callers. There is precise documentation of the semantics in unix(4). Reviewed by: dwmalone (eyeballed)
2001-08-17 22:01:18 +00:00
default:
error = EOPNOTSUPP;
break;
}
break;
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
Implement a LOCAL_PEERCRED socket option which returns a `struct xucred` with the credentials of the connected peer. Obviously this only works (and makes sense) on SOCK_STREAM sockets. This works for both the connect(2) and listen(2) callers. There is precise documentation of the semantics in unix(4). Reviewed by: dwmalone (eyeballed)
2001-08-17 22:01:18 +00:00
case SOPT_SET:
Implement unix(4) socket options LOCAL_CREDS and LOCAL_CONNWAIT. - Add unp_addsockcred() (for LOCAL_CREDS). - Add an argument to unp_connect2() to differentiate between PRU_CONNECT and PRU_CONNECT2. (for LOCAL_CONNWAIT) Obtained from: NetBSD (with some changes)
2005-04-13 00:01:46 +00:00
switch (sopt->sopt_name) {
case LOCAL_CREDS:
case LOCAL_CONNWAIT:
error = sooptcopyin(sopt, &optval, sizeof(optval),
sizeof(optval));
if (error)
break;
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
#define OPTSET(bit) do { \
UNP_PCB_LOCK(unp); \
if (optval) \
unp->unp_flags |= bit; \
else \
unp->unp_flags &= ~bit; \
UNP_PCB_UNLOCK(unp); \
} while (0)
Implement unix(4) socket options LOCAL_CREDS and LOCAL_CONNWAIT. - Add unp_addsockcred() (for LOCAL_CREDS). - Add an argument to unp_connect2() to differentiate between PRU_CONNECT and PRU_CONNECT2. (for LOCAL_CONNWAIT) Obtained from: NetBSD (with some changes)
2005-04-13 00:01:46 +00:00
switch (sopt->sopt_name) {
case LOCAL_CREDS:
OPTSET(UNP_WANTCRED);
break;
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
Implement unix(4) socket options LOCAL_CREDS and LOCAL_CONNWAIT. - Add unp_addsockcred() (for LOCAL_CREDS). - Add an argument to unp_connect2() to differentiate between PRU_CONNECT and PRU_CONNECT2. (for LOCAL_CONNWAIT) Obtained from: NetBSD (with some changes)
2005-04-13 00:01:46 +00:00
case LOCAL_CONNWAIT:
OPTSET(UNP_CONNWAIT);
break;
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
Implement unix(4) socket options LOCAL_CREDS and LOCAL_CONNWAIT. - Add unp_addsockcred() (for LOCAL_CREDS). - Add an argument to unp_connect2() to differentiate between PRU_CONNECT and PRU_CONNECT2. (for LOCAL_CONNWAIT) Obtained from: NetBSD (with some changes)
2005-04-13 00:01:46 +00:00
default:
break;
}
break;
#undef OPTSET
default:
error = ENOPROTOOPT;
break;
}
Add missing break. Found by: marcus
2005-04-25 00:48:04 +00:00
break;
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
Implement a LOCAL_PEERCRED socket option which returns a `struct xucred` with the credentials of the connected peer. Obviously this only works (and makes sense) on SOCK_STREAM sockets. This works for both the connect(2) and listen(2) callers. There is precise documentation of the semantics in unix(4). Reviewed by: dwmalone (eyeballed)
2001-08-17 22:01:18 +00:00
default:
error = EOPNOTSUPP;
break;
}
return (error);
}
Mechanical whitespace cleanup; parenthesize return values; other minor style nits.
2004-01-11 19:48:19 +00:00
Another mega commit to staticize things.
1995-12-14 09:55:16 +00:00
static int
style(9)-ize function headers, remove use of 'register'. MFC after: 3 days
2005-02-20 23:22:13 +00:00
unp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
- Implement two new system calls: int bindat(int fd, int s, const struct sockaddr *addr, socklen_t addrlen); int connectat(int fd, int s, const struct sockaddr *name, socklen_t namelen); which allow to bind and connect respectively to a UNIX domain socket with a path relative to the directory associated with the given file descriptor 'fd'. - Add manual pages for the new syscalls. - Make the new syscalls available for processes in capability mode sandbox. - Add capability rights CAP_BINDAT and CAP_CONNECTAT that has to be present on the directory descriptor for the syscalls to work. - Update audit(4) to support those two new syscalls and to handle path in sockaddr_un structure relative to the given directory descriptor. - Update procstat(1) to recognize the new capability rights. - Document the new capability rights in cap_rights_limit(2). Sponsored by: The FreeBSD Foundation Discussed with: rwatson, jilles, kib, des
2013-03-02 21:11:30 +00:00
{
return (unp_connectat(AT_FDCWD, so, nam, td));
}
static int
unp_connectat(int fd, struct socket *so, struct sockaddr *nam,
struct thread *td)
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
{
style(9)-ize function headers, remove use of 'register'. MFC after: 3 days
2005-02-20 23:22:13 +00:00
struct sockaddr_un *soun = (struct sockaddr_un *)nam;
struct vnode *vp;
Listening sockets improvements. o Separate fields of struct socket that belong to listening from fields that belong to normal dataflow, and unionize them. This shrinks the structure a bit. - Take out selinfo's from the socket buffers into the socket. The first reason is to support braindamaged scenario when a socket is added to kevent(2) and then listen(2) is cast on it. The second reason is that there is future plan to make socket buffers pluggable, so that for a dataflow socket a socket buffer can be changed, and in this case we also want to keep same selinfos through the lifetime of a socket. - Remove struct struct so_accf. Since now listening stuff no longer affects struct socket size, just move its fields into listening part of the union. - Provide sol_upcall field and enforce that so_upcall_set() may be called only on a dataflow socket, which has buffers, and for listening sockets provide solisten_upcall_set(). o Remove ACCEPT_LOCK() global. - Add a mutex to socket, to be used instead of socket buffer lock to lock fields of struct socket that don't belong to a socket buffer. - Allow to acquire two socket locks, but the first one must belong to a listening socket. - Make soref()/sorele() to use atomic(9). This allows in some situations to do soref() without owning socket lock. There is place for improvement here, it is possible to make sorele() also to lock optionally. - Most protocols aren't touched by this change, except UNIX local sockets. See below for more information. o Reduce copy-and-paste in kernel modules that accept connections from listening sockets: provide function solisten_dequeue(), and use it in the following modules: ctl(4), iscsi(4), ng_btsocket(4), ng_ksocket(4), infiniband, rpc. o UNIX local sockets. - Removal of ACCEPT_LOCK() global uncovered several races in the UNIX local sockets. Most races exist around spawning a new socket, when we are connecting to a local listening socket. To cover them, we need to hold locks on both PCBs when spawning a third one. This means holding them across sonewconn(). This creates a LOR between pcb locks and unp_list_lock. - To fix the new LOR, abandon the global unp_list_lock in favor of global unp_link_lock. Indeed, separating these two locks didn't provide us any extra parralelism in the UNIX sockets. - Now call into uipc_attach() may happen with unp_link_lock hold if, we are accepting, or without unp_link_lock in case if we are just creating a socket. - Another problem in UNIX sockets is that uipc_close() basicly did nothing for a listening socket. The vnode remained opened for connections. This is fixed by removing vnode in uipc_close(). Maybe the right way would be to do it for all sockets (not only listening), simply move the vnode teardown from uipc_detach() to uipc_close()? Sponsored by: Netflix Differential Revision: https://reviews.freebsd.org/D9770
2017-06-08 21:30:34 +00:00
struct socket *so2;
After completing a name lookup for a target UNIX domain socket to connect to, re-check that the local UNIX domain socket hasn't been closed while we slept, and if so, return EINVAL. This affects the system running both with and without Giant over the network stack, and recent ULE changes appear to cause it to trigger more frequently than previously under load. While here, improve catching of possibly closed UNIX domain sockets in one or two additional circumstances. I have a much larger set of related changes in Perforce, but they require more testing before they can be merged. One debugging printf is left in place to indicate when such a race takes place: this is typically triggered by a buggy application that simultaenously connect()'s and close()'s a UNIX domain socket file descriptor. I'll remove this at some point in the future, but am interested in seeing how frequently this is reported. In the case of Martin's reported problem, it appears to be a result of a non-thread safe syslog() implementation in the C library, which does not synchronize access to its logging file descriptor. Reported by: mbr
2004-08-14 03:43:49 +00:00
struct unpcb *unp, *unp2, *unp3;
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
struct nameidata nd;
Fix all areas of the system (or at least all those in LINT) to avoid storing socket addresses in mbufs. (Socket buffers are the one exception.) A number of kernel APIs needed to get fixed in order to make this happen. Also, fix three protocol families which kept PCBs in mbufs to not malloc them instead. Delete some old compatibility cruft while we're at it, and add some new routines in the in_cksum family.
1997-08-16 19:16:27 +00:00
char buf[SOCK_MAXADDRLEN];
Introduce a subsystem lock around UNIX domain sockets in order to protect global and allocated variables. This strategy is derived from work originally developed by BSDi for BSD/OS, and applied to FreeBSD by Sam Leffler: - Add unp_mtx, a global mutex which will protect all UNIX domain socket related variables, structures, etc. - Add UNP_LOCK(), UNP_UNLOCK(), UNP_LOCK_ASSERT() macros. - Acquire unp_mtx on entering most UNIX domain socket code, drop/re-acquire around calls into VFS, and release it on return. - Avoid performing sodupsockaddr() while holding the mutex, so in general move to allocating storage before acquiring the mutex to copy the data. - Make a stack copy of the xucred rather than copying out while holding unp_mtx. Copy the peer credential out after releasing the mutex. - Add additional assertions of vnode locks following VOP_CREATE(). A few notes: - Use of an sx lock for the file list mutex may cause problems with regard to unp_mtx when garbage collection passed file descriptors. - The locking in unp_pcblist() for sysctl monitoring is correct subject to the unpcb zone not returning memory for reuse by other subsystems (consistent with similar existing concerns). - Sam's version of this change, as with the BSD/OS version, made use of both a global lock and per-unpcb locks. However, in practice, the global lock covered all accesses, so I have simplified out the unpcb locks in the interest of getting this merged faster (reducing the overhead but not sacrificing granularity in most cases). We will want to explore possibilities for improving lock granularity in this code in the future. Submitted by: sam Sponsored by: FreeBSD Foundatiuon Obtained from: BSD/OS 5 snapshot provided by BSDi
2004-06-10 21:34:38 +00:00
struct sockaddr *sa;
Change the cap_rights_t type from uint64_t to a structure that we can extend in the future in a backward compatible (API and ABI) way. The cap_rights_t represents capability rights. We used to use one bit to represent one right, but we are running out of spare bits. Currently the new structure provides place for 114 rights (so 50 more than the previous cap_rights_t), but it is possible to grow the structure to hold at least 285 rights, although we can make it even larger if 285 rights won't be enough. The structure definition looks like this: struct cap_rights { uint64_t cr_rights[CAP_RIGHTS_VERSION + 2]; }; The initial CAP_RIGHTS_VERSION is 0. The top two bits in the first element of the cr_rights[] array contain total number of elements in the array - 2. This means if those two bits are equal to 0, we have 2 array elements. The top two bits in all remaining array elements should be 0. The next five bits in all array elements contain array index. Only one bit is used and bit position in this five-bits range defines array index. This means there can be at most five array elements in the future. To define new right the CAPRIGHT() macro must be used. The macro takes two arguments - an array index and a bit to set, eg. #define CAP_PDKILL CAPRIGHT(1, 0x0000000000000800ULL) We still support aliases that combine few rights, but the rights have to belong to the same array element, eg: #define CAP_LOOKUP CAPRIGHT(0, 0x0000000000000400ULL) #define CAP_FCHMOD CAPRIGHT(0, 0x0000000000002000ULL) #define CAP_FCHMODAT (CAP_FCHMOD | CAP_LOOKUP) There is new API to manage the new cap_rights_t structure: cap_rights_t *cap_rights_init(cap_rights_t *rights, ...); void cap_rights_set(cap_rights_t *rights, ...); void cap_rights_clear(cap_rights_t *rights, ...); bool cap_rights_is_set(const cap_rights_t *rights, ...); bool cap_rights_is_valid(const cap_rights_t *rights); void cap_rights_merge(cap_rights_t *dst, const cap_rights_t *src); void cap_rights_remove(cap_rights_t *dst, const cap_rights_t *src); bool cap_rights_contains(const cap_rights_t *big, const cap_rights_t *little); Capability rights to the cap_rights_init(), cap_rights_set(), cap_rights_clear() and cap_rights_is_set() functions are provided by separating them with commas, eg: cap_rights_t rights; cap_rights_init(&rights, CAP_READ, CAP_WRITE, CAP_FSTAT); There is no need to terminate the list of rights, as those functions are actually macros that take care of the termination, eg: #define cap_rights_set(rights, ...) \ __cap_rights_set((rights), __VA_ARGS__, 0ULL) void __cap_rights_set(cap_rights_t *rights, ...); Thanks to using one bit as an array index we can assert in those functions that there are no two rights belonging to different array elements provided together. For example this is illegal and will be detected, because CAP_LOOKUP belongs to element 0 and CAP_PDKILL to element 1: cap_rights_init(&rights, CAP_LOOKUP | CAP_PDKILL); Providing several rights that belongs to the same array's element this way is correct, but is not advised. It should only be used for aliases definition. This commit also breaks compatibility with some existing Capsicum system calls, but I see no other way to do that. This should be fine as Capsicum is still experimental and this change is not going to 9.x. Sponsored by: The FreeBSD Foundation
2013-09-05 00:09:56 +00:00
cap_rights_t rights;
int error, len;
Introduce a subsystem lock around UNIX domain sockets in order to protect global and allocated variables. This strategy is derived from work originally developed by BSDi for BSD/OS, and applied to FreeBSD by Sam Leffler: - Add unp_mtx, a global mutex which will protect all UNIX domain socket related variables, structures, etc. - Add UNP_LOCK(), UNP_UNLOCK(), UNP_LOCK_ASSERT() macros. - Acquire unp_mtx on entering most UNIX domain socket code, drop/re-acquire around calls into VFS, and release it on return. - Avoid performing sodupsockaddr() while holding the mutex, so in general move to allocating storage before acquiring the mutex to copy the data. - Make a stack copy of the xucred rather than copying out while holding unp_mtx. Copy the peer credential out after releasing the mutex. - Add additional assertions of vnode locks following VOP_CREATE(). A few notes: - Use of an sx lock for the file list mutex may cause problems with regard to unp_mtx when garbage collection passed file descriptors. - The locking in unp_pcblist() for sysctl monitoring is correct subject to the unpcb zone not returning memory for reuse by other subsystems (consistent with similar existing concerns). - Sam's version of this change, as with the BSD/OS version, made use of both a global lock and per-unpcb locks. However, in practice, the global lock covered all accesses, so I have simplified out the unpcb locks in the interest of getting this merged faster (reducing the overhead but not sacrificing granularity in most cases). We will want to explore possibilities for improving lock granularity in this code in the future. Submitted by: sam Sponsored by: FreeBSD Foundatiuon Obtained from: BSD/OS 5 snapshot provided by BSDi
2004-06-10 21:34:38 +00:00
Make bind(2) and connect(2) return EAFNOSUPPORT for AF_UNIX on wrong address family. See https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=191586 for the original discussion. Reviewed by: terry
2014-07-14 06:00:01 +00:00
if (nam->sa_family != AF_UNIX)
return (EAFNOSUPPORT);
Decompose the global UNIX domain sockets rwlock into two different locks: a global list/counter/generation counter protected by a new mutex unp_list_lock, and a global linkage rwlock, unp_global_rwlock, which protects the connections between UNIX domain sockets. This eliminates conditional lock acquisition that was previously a property of the global lock being held over sonewconn() leading to a call to uipc_attach(), which also required the global lock, but couldn't rely on it as other paths existed to uipc_attach() that didn't hold it: now uipc_attach() uses only the list lock, which follows the linkage lock in the lock order. It may also reduce contention on the global lock for some workloads. Add global UNIX domain socket locks to hard-coded witness lock order. MFC after: 1 week Discussed with: kris
2009-03-08 21:48:29 +00:00
UNP_LINK_WLOCK_ASSERT();
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
Modify UNIX domain sockets to guarantee, and assume, that so_pcb is always defined for an in-use socket. This allows us to eliminate countless tests of whether so_pcb is non-NULL, eliminating dozens of error cases. For now, retain the call to sotryfree() in the uipc_abort() path, but this will eventually move to soabort(). These new assumptions should be largely correct, and will become more so as the socket/pcb reference model is fixed. Removing the notion that so_pcb can be non-NULL is a critical step towards further fine-graining of the UNIX domain socket locking, as the so_pcb reference no longer needs to be protected using locks, instead it is a property of the socket life cycle.
2006-03-17 13:52:57 +00:00
unp = sotounpcb(so);
KASSERT(unp != NULL, ("unp_connect: unp == NULL"));
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
Fix handling of corrupt compress(1)ed data. [11:04] Add missing length checks on unix socket addresses. [11:05] Approved by: so (cperciva) Approved by: re (kensmith) Security: FreeBSD-SA-11:04.compress Security: CVE-2011-2895 [11:04] Security: FreeBSD-SA-11:05.unix
2011-09-28 08:47:17 +00:00
if (nam->sa_len > sizeof(struct sockaddr_un))
return (EINVAL);
Fix all areas of the system (or at least all those in LINT) to avoid storing socket addresses in mbufs. (Socket buffers are the one exception.) A number of kernel APIs needed to get fixed in order to make this happen. Also, fix three protocol families which kept PCBs in mbufs to not malloc them instead. Delete some old compatibility cruft while we're at it, and add some new routines in the in_cksum family.
1997-08-16 19:16:27 +00:00
len = nam->sa_len - offsetof(struct sockaddr_un, sun_path);
if (len <= 0)
Mechanical whitespace cleanup; parenthesize return values; other minor style nits.
2004-01-11 19:48:19 +00:00
return (EINVAL);
Use bcopy instead of strlcpy in uipc_bind and unp_connect, since soun->sun_path isn't a null-terminated string. As UNIX(4) states, "the terminating NUL is not part of the address." Since strlcpy has to return "the total length of the string [it] tried to create," it walks off the end of soun->sun_path looking for a \0. This reverts r105332. Reported by: Ryan Stone
2008-07-03 23:26:10 +00:00
bcopy(soun->sun_path, buf, len);
buf[len] = 0;
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
UNP_PCB_LOCK(unp);
Add two new unpcb flags, UNP_BINDING and UNP_CONNECTING, which will be used to mark UNIX domain sockets as being in the process of binding or connecting. Use these to prevent simultaneous bind or connect operations by multiple threads or processes on the same socket at the same time, which closes race conditions present in the UNIX domain socket implementation since inception.
2006-07-23 12:01:14 +00:00
if (unp->unp_flags & UNP_CONNECTING) {
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
UNP_PCB_UNLOCK(unp);
Add two new unpcb flags, UNP_BINDING and UNP_CONNECTING, which will be used to mark UNIX domain sockets as being in the process of binding or connecting. Use these to prevent simultaneous bind or connect operations by multiple threads or processes on the same socket at the same time, which closes race conditions present in the UNIX domain socket implementation since inception.
2006-07-23 12:01:14 +00:00
return (EALREADY);
}
Decompose the global UNIX domain sockets rwlock into two different locks: a global list/counter/generation counter protected by a new mutex unp_list_lock, and a global linkage rwlock, unp_global_rwlock, which protects the connections between UNIX domain sockets. This eliminates conditional lock acquisition that was previously a property of the global lock being held over sonewconn() leading to a call to uipc_attach(), which also required the global lock, but couldn't rely on it as other paths existed to uipc_attach() that didn't hold it: now uipc_attach() uses only the list lock, which follows the linkage lock in the lock order. It may also reduce contention on the global lock for some workloads. Add global UNIX domain socket locks to hard-coded witness lock order. MFC after: 1 week Discussed with: kris
2009-03-08 21:48:29 +00:00
UNP_LINK_WUNLOCK();
Set UNP_CONNECTING when committing to moving ahead in unp_connect(). This logic was lost when merging the remainder of these changes in 1.178.
2007-02-13 21:00:57 +00:00
unp->unp_flags |= UNP_CONNECTING;
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
UNP_PCB_UNLOCK(unp);
Introduce a subsystem lock around UNIX domain sockets in order to protect global and allocated variables. This strategy is derived from work originally developed by BSDi for BSD/OS, and applied to FreeBSD by Sam Leffler: - Add unp_mtx, a global mutex which will protect all UNIX domain socket related variables, structures, etc. - Add UNP_LOCK(), UNP_UNLOCK(), UNP_LOCK_ASSERT() macros. - Acquire unp_mtx on entering most UNIX domain socket code, drop/re-acquire around calls into VFS, and release it on return. - Avoid performing sodupsockaddr() while holding the mutex, so in general move to allocating storage before acquiring the mutex to copy the data. - Make a stack copy of the xucred rather than copying out while holding unp_mtx. Copy the peer credential out after releasing the mutex. - Add additional assertions of vnode locks following VOP_CREATE(). A few notes: - Use of an sx lock for the file list mutex may cause problems with regard to unp_mtx when garbage collection passed file descriptors. - The locking in unp_pcblist() for sysctl monitoring is correct subject to the unpcb zone not returning memory for reuse by other subsystems (consistent with similar existing concerns). - Sam's version of this change, as with the BSD/OS version, made use of both a global lock and per-unpcb locks. However, in practice, the global lock covered all accesses, so I have simplified out the unpcb locks in the interest of getting this merged faster (reducing the overhead but not sacrificing granularity in most cases). We will want to explore possibilities for improving lock granularity in this code in the future. Submitted by: sam Sponsored by: FreeBSD Foundatiuon Obtained from: BSD/OS 5 snapshot provided by BSDi
2004-06-10 21:34:38 +00:00
sa = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
- Implement two new system calls: int bindat(int fd, int s, const struct sockaddr *addr, socklen_t addrlen); int connectat(int fd, int s, const struct sockaddr *name, socklen_t namelen); which allow to bind and connect respectively to a UNIX domain socket with a path relative to the directory associated with the given file descriptor 'fd'. - Add manual pages for the new syscalls. - Make the new syscalls available for processes in capability mode sandbox. - Add capability rights CAP_BINDAT and CAP_CONNECTAT that has to be present on the directory descriptor for the syscalls to work. - Update audit(4) to support those two new syscalls and to handle path in sockaddr_un structure relative to the given directory descriptor. - Update procstat(1) to recognize the new capability rights. - Document the new capability rights in cap_rights_limit(2). Sponsored by: The FreeBSD Foundation Discussed with: rwatson, jilles, kib, des
2013-03-02 21:11:30 +00:00
NDINIT_ATRIGHTS(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF,
Change the cap_rights_t type from uint64_t to a structure that we can extend in the future in a backward compatible (API and ABI) way. The cap_rights_t represents capability rights. We used to use one bit to represent one right, but we are running out of spare bits. Currently the new structure provides place for 114 rights (so 50 more than the previous cap_rights_t), but it is possible to grow the structure to hold at least 285 rights, although we can make it even larger if 285 rights won't be enough. The structure definition looks like this: struct cap_rights { uint64_t cr_rights[CAP_RIGHTS_VERSION + 2]; }; The initial CAP_RIGHTS_VERSION is 0. The top two bits in the first element of the cr_rights[] array contain total number of elements in the array - 2. This means if those two bits are equal to 0, we have 2 array elements. The top two bits in all remaining array elements should be 0. The next five bits in all array elements contain array index. Only one bit is used and bit position in this five-bits range defines array index. This means there can be at most five array elements in the future. To define new right the CAPRIGHT() macro must be used. The macro takes two arguments - an array index and a bit to set, eg. #define CAP_PDKILL CAPRIGHT(1, 0x0000000000000800ULL) We still support aliases that combine few rights, but the rights have to belong to the same array element, eg: #define CAP_LOOKUP CAPRIGHT(0, 0x0000000000000400ULL) #define CAP_FCHMOD CAPRIGHT(0, 0x0000000000002000ULL) #define CAP_FCHMODAT (CAP_FCHMOD | CAP_LOOKUP) There is new API to manage the new cap_rights_t structure: cap_rights_t *cap_rights_init(cap_rights_t *rights, ...); void cap_rights_set(cap_rights_t *rights, ...); void cap_rights_clear(cap_rights_t *rights, ...); bool cap_rights_is_set(const cap_rights_t *rights, ...); bool cap_rights_is_valid(const cap_rights_t *rights); void cap_rights_merge(cap_rights_t *dst, const cap_rights_t *src); void cap_rights_remove(cap_rights_t *dst, const cap_rights_t *src); bool cap_rights_contains(const cap_rights_t *big, const cap_rights_t *little); Capability rights to the cap_rights_init(), cap_rights_set(), cap_rights_clear() and cap_rights_is_set() functions are provided by separating them with commas, eg: cap_rights_t rights; cap_rights_init(&rights, CAP_READ, CAP_WRITE, CAP_FSTAT); There is no need to terminate the list of rights, as those functions are actually macros that take care of the termination, eg: #define cap_rights_set(rights, ...) \ __cap_rights_set((rights), __VA_ARGS__, 0ULL) void __cap_rights_set(cap_rights_t *rights, ...); Thanks to using one bit as an array index we can assert in those functions that there are no two rights belonging to different array elements provided together. For example this is illegal and will be detected, because CAP_LOOKUP belongs to element 0 and CAP_PDKILL to element 1: cap_rights_init(&rights, CAP_LOOKUP | CAP_PDKILL); Providing several rights that belongs to the same array's element this way is correct, but is not advised. It should only be used for aliases definition. This commit also breaks compatibility with some existing Capsicum system calls, but I see no other way to do that. This should be fine as Capsicum is still experimental and this change is not going to 9.x. Sponsored by: The FreeBSD Foundation
2013-09-05 00:09:56 +00:00
UIO_SYSSPACE, buf, fd, cap_rights_init(&rights, CAP_CONNECTAT), td);
All of this is cosmetic. prototypes, #includes, printfs and so on. Makes GCC a lot more silent.
1994-10-02 17:35:40 +00:00
error = namei(&nd);
if (error)
Introduce a subsystem lock around UNIX domain sockets in order to protect global and allocated variables. This strategy is derived from work originally developed by BSDi for BSD/OS, and applied to FreeBSD by Sam Leffler: - Add unp_mtx, a global mutex which will protect all UNIX domain socket related variables, structures, etc. - Add UNP_LOCK(), UNP_UNLOCK(), UNP_LOCK_ASSERT() macros. - Acquire unp_mtx on entering most UNIX domain socket code, drop/re-acquire around calls into VFS, and release it on return. - Avoid performing sodupsockaddr() while holding the mutex, so in general move to allocating storage before acquiring the mutex to copy the data. - Make a stack copy of the xucred rather than copying out while holding unp_mtx. Copy the peer credential out after releasing the mutex. - Add additional assertions of vnode locks following VOP_CREATE(). A few notes: - Use of an sx lock for the file list mutex may cause problems with regard to unp_mtx when garbage collection passed file descriptors. - The locking in unp_pcblist() for sysctl monitoring is correct subject to the unpcb zone not returning memory for reuse by other subsystems (consistent with similar existing concerns). - Sam's version of this change, as with the BSD/OS version, made use of both a global lock and per-unpcb locks. However, in practice, the global lock covered all accesses, so I have simplified out the unpcb locks in the interest of getting this merged faster (reducing the overhead but not sacrificing granularity in most cases). We will want to explore possibilities for improving lock granularity in this code in the future. Submitted by: sam Sponsored by: FreeBSD Foundatiuon Obtained from: BSD/OS 5 snapshot provided by BSDi
2004-06-10 21:34:38 +00:00
vp = NULL;
else
vp = nd.ni_vp;
ASSERT_VOP_LOCKED(vp, "unp_connect");
Introduce NDFREE (and remove VOP_ABORTOP)
1999-12-15 23:02:35 +00:00
NDFREE(&nd, NDF_ONLY_PNBUF);
Introduce a subsystem lock around UNIX domain sockets in order to protect global and allocated variables. This strategy is derived from work originally developed by BSDi for BSD/OS, and applied to FreeBSD by Sam Leffler: - Add unp_mtx, a global mutex which will protect all UNIX domain socket related variables, structures, etc. - Add UNP_LOCK(), UNP_UNLOCK(), UNP_LOCK_ASSERT() macros. - Acquire unp_mtx on entering most UNIX domain socket code, drop/re-acquire around calls into VFS, and release it on return. - Avoid performing sodupsockaddr() while holding the mutex, so in general move to allocating storage before acquiring the mutex to copy the data. - Make a stack copy of the xucred rather than copying out while holding unp_mtx. Copy the peer credential out after releasing the mutex. - Add additional assertions of vnode locks following VOP_CREATE(). A few notes: - Use of an sx lock for the file list mutex may cause problems with regard to unp_mtx when garbage collection passed file descriptors. - The locking in unp_pcblist() for sysctl monitoring is correct subject to the unpcb zone not returning memory for reuse by other subsystems (consistent with similar existing concerns). - Sam's version of this change, as with the BSD/OS version, made use of both a global lock and per-unpcb locks. However, in practice, the global lock covered all accesses, so I have simplified out the unpcb locks in the interest of getting this merged faster (reducing the overhead but not sacrificing granularity in most cases). We will want to explore possibilities for improving lock granularity in this code in the future. Submitted by: sam Sponsored by: FreeBSD Foundatiuon Obtained from: BSD/OS 5 snapshot provided by BSDi
2004-06-10 21:34:38 +00:00
if (error)
goto bad;
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
if (vp->v_type != VSOCK) {
error = ENOTSOCK;
goto bad;
}
Add an additional MAC check to the UNIX domain socket connect path: check that the subject has read/write access to the vnode using the vnode MAC check. MFC after: 3 weeks Submitted by: Spencer Minear <spencer_minear at securecomputing dot com> Obtained from: TrustedBSD Project
2007-02-22 09:37:44 +00:00
#ifdef MAC
Merge first in a series of TrustedBSD MAC Framework KPI changes from Mac OS X Leopard--rationalize naming for entry points to the following general forms: mac_<object>_<method/action> mac_<object>_check_<method/action> The previous naming scheme was inconsistent and mostly reversed from the new scheme. Also, make object types more consistent and remove spaces from object types that contain multiple parts ("posix_sem" -> "posixsem") to make mechanical parsing easier. Introduce a new "netinet" object type for certain IPv4/IPv6-related methods. Also simplify, slightly, some entry point names. All MAC policy modules will need to be recompiled, and modules not updates as part of this commit will need to be modified to conform to the new KPI. Sponsored by: SPARTA (original patches against Mac OS X) Obtained from: TrustedBSD Project, Apple Computer
2007-10-24 19:04:04 +00:00
error = mac_vnode_check_open(td->td_ucred, vp, VWRITE | VREAD);
Add an additional MAC check to the UNIX domain socket connect path: check that the subject has read/write access to the vnode using the vnode MAC check. MFC after: 3 weeks Submitted by: Spencer Minear <spencer_minear at securecomputing dot com> Obtained from: TrustedBSD Project
2007-02-22 09:37:44 +00:00
if (error)
goto bad;
#endif
Simple p_ucred -> td_ucred changes to start using the per-thread ucred reference.
2002-02-27 18:32:23 +00:00
error = VOP_ACCESS(vp, VWRITE, td->td_ucred, td);
All of this is cosmetic. prototypes, #includes, printfs and so on. Makes GCC a lot more silent.
1994-10-02 17:35:40 +00:00
if (error)
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
goto bad;
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
After completing a name lookup for a target UNIX domain socket to connect to, re-check that the local UNIX domain socket hasn't been closed while we slept, and if so, return EINVAL. This affects the system running both with and without Giant over the network stack, and recent ULE changes appear to cause it to trigger more frequently than previously under load. While here, improve catching of possibly closed UNIX domain sockets in one or two additional circumstances. I have a much larger set of related changes in Perforce, but they require more testing before they can be merged. One debugging printf is left in place to indicate when such a race takes place: this is typically triggered by a buggy application that simultaenously connect()'s and close()'s a UNIX domain socket file descriptor. I'll remove this at some point in the future, but am interested in seeing how frequently this is reported. In the case of Martin's reported problem, it appears to be a result of a non-thread safe syslog() implementation in the C library, which does not synchronize access to its logging file descriptor. Reported by: mbr
2004-08-14 03:43:49 +00:00
unp = sotounpcb(so);
Modify UNIX domain sockets to guarantee, and assume, that so_pcb is always defined for an in-use socket. This allows us to eliminate countless tests of whether so_pcb is non-NULL, eliminating dozens of error cases. For now, retain the call to sotryfree() in the uipc_abort() path, but this will eventually move to soabort(). These new assumptions should be largely correct, and will become more so as the socket/pcb reference model is fixed. Removing the notion that so_pcb can be non-NULL is a critical step towards further fine-graining of the UNIX domain socket locking, as the so_pcb reference no longer needs to be protected using locks, instead it is a property of the socket life cycle.
2006-03-17 13:52:57 +00:00
KASSERT(unp != NULL, ("unp_connect: unp == NULL"));
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
/*
Decompose the global UNIX domain sockets rwlock into two different locks: a global list/counter/generation counter protected by a new mutex unp_list_lock, and a global linkage rwlock, unp_global_rwlock, which protects the connections between UNIX domain sockets. This eliminates conditional lock acquisition that was previously a property of the global lock being held over sonewconn() leading to a call to uipc_attach(), which also required the global lock, but couldn't rely on it as other paths existed to uipc_attach() that didn't hold it: now uipc_attach() uses only the list lock, which follows the linkage lock in the lock order. It may also reduce contention on the global lock for some workloads. Add global UNIX domain socket locks to hard-coded witness lock order. MFC after: 1 week Discussed with: kris
2009-03-08 21:48:29 +00:00
* Lock linkage lock for two reasons: make sure v_socket is stable,
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
* and to protect simultaneous locking of multiple pcbs.
*/
Decompose the global UNIX domain sockets rwlock into two different locks: a global list/counter/generation counter protected by a new mutex unp_list_lock, and a global linkage rwlock, unp_global_rwlock, which protects the connections between UNIX domain sockets. This eliminates conditional lock acquisition that was previously a property of the global lock being held over sonewconn() leading to a call to uipc_attach(), which also required the global lock, but couldn't rely on it as other paths existed to uipc_attach() that didn't hold it: now uipc_attach() uses only the list lock, which follows the linkage lock in the lock order. It may also reduce contention on the global lock for some workloads. Add global UNIX domain socket locks to hard-coded witness lock order. MFC after: 1 week Discussed with: kris
2009-03-08 21:48:29 +00:00
UNP_LINK_WLOCK();
For UNIX sockets make vnode point not to the socket, but to the UNIX PCB, since the latter is the thing that links together VFS and sockets. While here, make the union in the struct vnode anonymous.
2017-06-02 17:31:25 +00:00
VOP_UNP_CONNECT(vp, &unp2);
if (unp2 == NULL) {
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
error = ECONNREFUSED;
Drop Giant and acquire the UNIX domain socket subsystem lock a bit earlier in unp_connect() so that vp->v_socket can't change between our copying its value to a local variable and later use of that variable. This may have been responsible for a panic during shutdown that I experienced where simultaneous closing of a listen socket by rpcbind and a new connection being made to rpcbind by mountd.
2004-07-18 01:29:43 +00:00
goto bad2;
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
}
For UNIX sockets make vnode point not to the socket, but to the UNIX PCB, since the latter is the thing that links together VFS and sockets. While here, make the union in the struct vnode anonymous.
2017-06-02 17:31:25 +00:00
so2 = unp2->unp_socket;
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
if (so->so_type != so2->so_type) {
error = EPROTOTYPE;
Drop Giant and acquire the UNIX domain socket subsystem lock a bit earlier in unp_connect() so that vp->v_socket can't change between our copying its value to a local variable and later use of that variable. This may have been responsible for a panic during shutdown that I experienced where simultaneous closing of a listen socket by rpcbind and a new connection being made to rpcbind by mountd.
2004-07-18 01:29:43 +00:00
goto bad2;
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
}
Listening sockets improvements. o Separate fields of struct socket that belong to listening from fields that belong to normal dataflow, and unionize them. This shrinks the structure a bit. - Take out selinfo's from the socket buffers into the socket. The first reason is to support braindamaged scenario when a socket is added to kevent(2) and then listen(2) is cast on it. The second reason is that there is future plan to make socket buffers pluggable, so that for a dataflow socket a socket buffer can be changed, and in this case we also want to keep same selinfos through the lifetime of a socket. - Remove struct struct so_accf. Since now listening stuff no longer affects struct socket size, just move its fields into listening part of the union. - Provide sol_upcall field and enforce that so_upcall_set() may be called only on a dataflow socket, which has buffers, and for listening sockets provide solisten_upcall_set(). o Remove ACCEPT_LOCK() global. - Add a mutex to socket, to be used instead of socket buffer lock to lock fields of struct socket that don't belong to a socket buffer. - Allow to acquire two socket locks, but the first one must belong to a listening socket. - Make soref()/sorele() to use atomic(9). This allows in some situations to do soref() without owning socket lock. There is place for improvement here, it is possible to make sorele() also to lock optionally. - Most protocols aren't touched by this change, except UNIX local sockets. See below for more information. o Reduce copy-and-paste in kernel modules that accept connections from listening sockets: provide function solisten_dequeue(), and use it in the following modules: ctl(4), iscsi(4), ng_btsocket(4), ng_ksocket(4), infiniband, rpc. o UNIX local sockets. - Removal of ACCEPT_LOCK() global uncovered several races in the UNIX local sockets. Most races exist around spawning a new socket, when we are connecting to a local listening socket. To cover them, we need to hold locks on both PCBs when spawning a third one. This means holding them across sonewconn(). This creates a LOR between pcb locks and unp_list_lock. - To fix the new LOR, abandon the global unp_list_lock in favor of global unp_link_lock. Indeed, separating these two locks didn't provide us any extra parralelism in the UNIX sockets. - Now call into uipc_attach() may happen with unp_link_lock hold if, we are accepting, or without unp_link_lock in case if we are just creating a socket. - Another problem in UNIX sockets is that uipc_close() basicly did nothing for a listening socket. The vnode remained opened for connections. This is fixed by removing vnode in uipc_close(). Maybe the right way would be to do it for all sockets (not only listening), simply move the vnode teardown from uipc_detach() to uipc_close()? Sponsored by: Netflix Differential Revision: https://reviews.freebsd.org/D9770
2017-06-08 21:30:34 +00:00
UNP_PCB_LOCK(unp);
UNP_PCB_LOCK(unp2);
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
if (so2->so_options & SO_ACCEPTCONN) {
Mfp4 CH=177274,177280,177284-177285,177297,177324-177325 VNET socket push back: try to minimize the number of places where we have to switch vnets and narrow down the time we stay switched. Add assertions to the socket code to catch possibly unset vnets as seen in r204147. While this reduces the number of vnet recursion in some places like NFS, POSIX local sockets and some netgraph, .. recursions are impossible to fix. The current expectations are documented at the beginning of uipc_socket.c along with the other information there. Sponsored by: The FreeBSD Foundation Sponsored by: CK Software GmbH Reviewed by: jhb Tested by: zec Tested by: Mikolaj Golub (to.my.trociny gmail.com) MFC after: 2 weeks
2011-02-16 21:29:13 +00:00
CURVNET_SET(so2->so_vnet);
Listening sockets improvements. o Separate fields of struct socket that belong to listening from fields that belong to normal dataflow, and unionize them. This shrinks the structure a bit. - Take out selinfo's from the socket buffers into the socket. The first reason is to support braindamaged scenario when a socket is added to kevent(2) and then listen(2) is cast on it. The second reason is that there is future plan to make socket buffers pluggable, so that for a dataflow socket a socket buffer can be changed, and in this case we also want to keep same selinfos through the lifetime of a socket. - Remove struct struct so_accf. Since now listening stuff no longer affects struct socket size, just move its fields into listening part of the union. - Provide sol_upcall field and enforce that so_upcall_set() may be called only on a dataflow socket, which has buffers, and for listening sockets provide solisten_upcall_set(). o Remove ACCEPT_LOCK() global. - Add a mutex to socket, to be used instead of socket buffer lock to lock fields of struct socket that don't belong to a socket buffer. - Allow to acquire two socket locks, but the first one must belong to a listening socket. - Make soref()/sorele() to use atomic(9). This allows in some situations to do soref() without owning socket lock. There is place for improvement here, it is possible to make sorele() also to lock optionally. - Most protocols aren't touched by this change, except UNIX local sockets. See below for more information. o Reduce copy-and-paste in kernel modules that accept connections from listening sockets: provide function solisten_dequeue(), and use it in the following modules: ctl(4), iscsi(4), ng_btsocket(4), ng_ksocket(4), infiniband, rpc. o UNIX local sockets. - Removal of ACCEPT_LOCK() global uncovered several races in the UNIX local sockets. Most races exist around spawning a new socket, when we are connecting to a local listening socket. To cover them, we need to hold locks on both PCBs when spawning a third one. This means holding them across sonewconn(). This creates a LOR between pcb locks and unp_list_lock. - To fix the new LOR, abandon the global unp_list_lock in favor of global unp_link_lock. Indeed, separating these two locks didn't provide us any extra parralelism in the UNIX sockets. - Now call into uipc_attach() may happen with unp_link_lock hold if, we are accepting, or without unp_link_lock in case if we are just creating a socket. - Another problem in UNIX sockets is that uipc_close() basicly did nothing for a listening socket. The vnode remained opened for connections. This is fixed by removing vnode in uipc_close(). Maybe the right way would be to do it for all sockets (not only listening), simply move the vnode teardown from uipc_detach() to uipc_close()? Sponsored by: Netflix Differential Revision: https://reviews.freebsd.org/D9770
2017-06-08 21:30:34 +00:00
so2 = sonewconn(so2, 0);
Mfp4 CH=177274,177280,177284-177285,177297,177324-177325 VNET socket push back: try to minimize the number of places where we have to switch vnets and narrow down the time we stay switched. Add assertions to the socket code to catch possibly unset vnets as seen in r204147. While this reduces the number of vnet recursion in some places like NFS, POSIX local sockets and some netgraph, .. recursions are impossible to fix. The current expectations are documented at the beginning of uipc_socket.c along with the other information there. Sponsored by: The FreeBSD Foundation Sponsored by: CK Software GmbH Reviewed by: jhb Tested by: zec Tested by: Mikolaj Golub (to.my.trociny gmail.com) MFC after: 2 weeks
2011-02-16 21:29:13 +00:00
CURVNET_RESTORE();
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
} else
Listening sockets improvements. o Separate fields of struct socket that belong to listening from fields that belong to normal dataflow, and unionize them. This shrinks the structure a bit. - Take out selinfo's from the socket buffers into the socket. The first reason is to support braindamaged scenario when a socket is added to kevent(2) and then listen(2) is cast on it. The second reason is that there is future plan to make socket buffers pluggable, so that for a dataflow socket a socket buffer can be changed, and in this case we also want to keep same selinfos through the lifetime of a socket. - Remove struct struct so_accf. Since now listening stuff no longer affects struct socket size, just move its fields into listening part of the union. - Provide sol_upcall field and enforce that so_upcall_set() may be called only on a dataflow socket, which has buffers, and for listening sockets provide solisten_upcall_set(). o Remove ACCEPT_LOCK() global. - Add a mutex to socket, to be used instead of socket buffer lock to lock fields of struct socket that don't belong to a socket buffer. - Allow to acquire two socket locks, but the first one must belong to a listening socket. - Make soref()/sorele() to use atomic(9). This allows in some situations to do soref() without owning socket lock. There is place for improvement here, it is possible to make sorele() also to lock optionally. - Most protocols aren't touched by this change, except UNIX local sockets. See below for more information. o Reduce copy-and-paste in kernel modules that accept connections from listening sockets: provide function solisten_dequeue(), and use it in the following modules: ctl(4), iscsi(4), ng_btsocket(4), ng_ksocket(4), infiniband, rpc. o UNIX local sockets. - Removal of ACCEPT_LOCK() global uncovered several races in the UNIX local sockets. Most races exist around spawning a new socket, when we are connecting to a local listening socket. To cover them, we need to hold locks on both PCBs when spawning a third one. This means holding them across sonewconn(). This creates a LOR between pcb locks and unp_list_lock. - To fix the new LOR, abandon the global unp_list_lock in favor of global unp_link_lock. Indeed, separating these two locks didn't provide us any extra parralelism in the UNIX sockets. - Now call into uipc_attach() may happen with unp_link_lock hold if, we are accepting, or without unp_link_lock in case if we are just creating a socket. - Another problem in UNIX sockets is that uipc_close() basicly did nothing for a listening socket. The vnode remained opened for connections. This is fixed by removing vnode in uipc_close(). Maybe the right way would be to do it for all sockets (not only listening), simply move the vnode teardown from uipc_detach() to uipc_close()? Sponsored by: Netflix Differential Revision: https://reviews.freebsd.org/D9770
2017-06-08 21:30:34 +00:00
so2 = NULL;
if (so2 == NULL) {
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
error = ECONNREFUSED;
Listening sockets improvements. o Separate fields of struct socket that belong to listening from fields that belong to normal dataflow, and unionize them. This shrinks the structure a bit. - Take out selinfo's from the socket buffers into the socket. The first reason is to support braindamaged scenario when a socket is added to kevent(2) and then listen(2) is cast on it. The second reason is that there is future plan to make socket buffers pluggable, so that for a dataflow socket a socket buffer can be changed, and in this case we also want to keep same selinfos through the lifetime of a socket. - Remove struct struct so_accf. Since now listening stuff no longer affects struct socket size, just move its fields into listening part of the union. - Provide sol_upcall field and enforce that so_upcall_set() may be called only on a dataflow socket, which has buffers, and for listening sockets provide solisten_upcall_set(). o Remove ACCEPT_LOCK() global. - Add a mutex to socket, to be used instead of socket buffer lock to lock fields of struct socket that don't belong to a socket buffer. - Allow to acquire two socket locks, but the first one must belong to a listening socket. - Make soref()/sorele() to use atomic(9). This allows in some situations to do soref() without owning socket lock. There is place for improvement here, it is possible to make sorele() also to lock optionally. - Most protocols aren't touched by this change, except UNIX local sockets. See below for more information. o Reduce copy-and-paste in kernel modules that accept connections from listening sockets: provide function solisten_dequeue(), and use it in the following modules: ctl(4), iscsi(4), ng_btsocket(4), ng_ksocket(4), infiniband, rpc. o UNIX local sockets. - Removal of ACCEPT_LOCK() global uncovered several races in the UNIX local sockets. Most races exist around spawning a new socket, when we are connecting to a local listening socket. To cover them, we need to hold locks on both PCBs when spawning a third one. This means holding them across sonewconn(). This creates a LOR between pcb locks and unp_list_lock. - To fix the new LOR, abandon the global unp_list_lock in favor of global unp_link_lock. Indeed, separating these two locks didn't provide us any extra parralelism in the UNIX sockets. - Now call into uipc_attach() may happen with unp_link_lock hold if, we are accepting, or without unp_link_lock in case if we are just creating a socket. - Another problem in UNIX sockets is that uipc_close() basicly did nothing for a listening socket. The vnode remained opened for connections. This is fixed by removing vnode in uipc_close(). Maybe the right way would be to do it for all sockets (not only listening), simply move the vnode teardown from uipc_detach() to uipc_close()? Sponsored by: Netflix Differential Revision: https://reviews.freebsd.org/D9770
2017-06-08 21:30:34 +00:00
goto bad3;
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
}
Listening sockets improvements. o Separate fields of struct socket that belong to listening from fields that belong to normal dataflow, and unionize them. This shrinks the structure a bit. - Take out selinfo's from the socket buffers into the socket. The first reason is to support braindamaged scenario when a socket is added to kevent(2) and then listen(2) is cast on it. The second reason is that there is future plan to make socket buffers pluggable, so that for a dataflow socket a socket buffer can be changed, and in this case we also want to keep same selinfos through the lifetime of a socket. - Remove struct struct so_accf. Since now listening stuff no longer affects struct socket size, just move its fields into listening part of the union. - Provide sol_upcall field and enforce that so_upcall_set() may be called only on a dataflow socket, which has buffers, and for listening sockets provide solisten_upcall_set(). o Remove ACCEPT_LOCK() global. - Add a mutex to socket, to be used instead of socket buffer lock to lock fields of struct socket that don't belong to a socket buffer. - Allow to acquire two socket locks, but the first one must belong to a listening socket. - Make soref()/sorele() to use atomic(9). This allows in some situations to do soref() without owning socket lock. There is place for improvement here, it is possible to make sorele() also to lock optionally. - Most protocols aren't touched by this change, except UNIX local sockets. See below for more information. o Reduce copy-and-paste in kernel modules that accept connections from listening sockets: provide function solisten_dequeue(), and use it in the following modules: ctl(4), iscsi(4), ng_btsocket(4), ng_ksocket(4), infiniband, rpc. o UNIX local sockets. - Removal of ACCEPT_LOCK() global uncovered several races in the UNIX local sockets. Most races exist around spawning a new socket, when we are connecting to a local listening socket. To cover them, we need to hold locks on both PCBs when spawning a third one. This means holding them across sonewconn(). This creates a LOR between pcb locks and unp_list_lock. - To fix the new LOR, abandon the global unp_list_lock in favor of global unp_link_lock. Indeed, separating these two locks didn't provide us any extra parralelism in the UNIX sockets. - Now call into uipc_attach() may happen with unp_link_lock hold if, we are accepting, or without unp_link_lock in case if we are just creating a socket. - Another problem in UNIX sockets is that uipc_close() basicly did nothing for a listening socket. The vnode remained opened for connections. This is fixed by removing vnode in uipc_close(). Maybe the right way would be to do it for all sockets (not only listening), simply move the vnode teardown from uipc_detach() to uipc_close()? Sponsored by: Netflix Differential Revision: https://reviews.freebsd.org/D9770
2017-06-08 21:30:34 +00:00
unp3 = sotounpcb(so2);
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
UNP_PCB_LOCK(unp3);
Introduce a subsystem lock around UNIX domain sockets in order to protect global and allocated variables. This strategy is derived from work originally developed by BSDi for BSD/OS, and applied to FreeBSD by Sam Leffler: - Add unp_mtx, a global mutex which will protect all UNIX domain socket related variables, structures, etc. - Add UNP_LOCK(), UNP_UNLOCK(), UNP_LOCK_ASSERT() macros. - Acquire unp_mtx on entering most UNIX domain socket code, drop/re-acquire around calls into VFS, and release it on return. - Avoid performing sodupsockaddr() while holding the mutex, so in general move to allocating storage before acquiring the mutex to copy the data. - Make a stack copy of the xucred rather than copying out while holding unp_mtx. Copy the peer credential out after releasing the mutex. - Add additional assertions of vnode locks following VOP_CREATE(). A few notes: - Use of an sx lock for the file list mutex may cause problems with regard to unp_mtx when garbage collection passed file descriptors. - The locking in unp_pcblist() for sysctl monitoring is correct subject to the unpcb zone not returning memory for reuse by other subsystems (consistent with similar existing concerns). - Sam's version of this change, as with the BSD/OS version, made use of both a global lock and per-unpcb locks. However, in practice, the global lock covered all accesses, so I have simplified out the unpcb locks in the interest of getting this merged faster (reducing the overhead but not sacrificing granularity in most cases). We will want to explore possibilities for improving lock granularity in this code in the future. Submitted by: sam Sponsored by: FreeBSD Foundatiuon Obtained from: BSD/OS 5 snapshot provided by BSDi
2004-06-10 21:34:38 +00:00
if (unp2->unp_addr != NULL) {
bcopy(unp2->unp_addr, sa, unp2->unp_addr->sun_len);
unp3->unp_addr = (struct sockaddr_un *) sa;
sa = NULL;
}
White space and comment tweaks. MFC after: 3 weeks
2009-01-01 20:03:22 +00:00
Implement a LOCAL_PEERCRED socket option which returns a `struct xucred` with the credentials of the connected peer. Obviously this only works (and makes sense) on SOCK_STREAM sockets. This works for both the connect(2) and listen(2) callers. There is precise documentation of the semantics in unix(4). Reviewed by: dwmalone (eyeballed)
2001-08-17 22:01:18 +00:00
/*
Fxi a bunch of typos. PR: misc/174625 Submitted by: Jeremy Chadwick <jdc@koitsu.org>
2013-05-10 16:41:26 +00:00
* The connector's (client's) credentials are copied from its
Re-wrap and other minor formatting and punctuation fixes for UNIX domain socket comments.
2006-07-22 17:24:55 +00:00
* process structure at the time of connect() (which is now).
Implement a LOCAL_PEERCRED socket option which returns a `struct xucred` with the credentials of the connected peer. Obviously this only works (and makes sense) on SOCK_STREAM sockets. This works for both the connect(2) and listen(2) callers. There is precise documentation of the semantics in unix(4). Reviewed by: dwmalone (eyeballed)
2001-08-17 22:01:18 +00:00
*/
Simple p_ucred -> td_ucred changes to start using the per-thread ucred reference.
2002-02-27 18:32:23 +00:00
cru2x(td->td_ucred, &unp3->unp_peercred);
Implement a LOCAL_PEERCRED socket option which returns a `struct xucred` with the credentials of the connected peer. Obviously this only works (and makes sense) on SOCK_STREAM sockets. This works for both the connect(2) and listen(2) callers. There is precise documentation of the semantics in unix(4). Reviewed by: dwmalone (eyeballed)
2001-08-17 22:01:18 +00:00
unp3->unp_flags |= UNP_HAVEPC;
White space and comment tweaks. MFC after: 3 weeks
2009-01-01 20:03:22 +00:00
Implement a LOCAL_PEERCRED socket option which returns a `struct xucred` with the credentials of the connected peer. Obviously this only works (and makes sense) on SOCK_STREAM sockets. This works for both the connect(2) and listen(2) callers. There is precise documentation of the semantics in unix(4). Reviewed by: dwmalone (eyeballed)
2001-08-17 22:01:18 +00:00
/*
Re-wrap and other minor formatting and punctuation fixes for UNIX domain socket comments.
2006-07-22 17:24:55 +00:00
* The receiver's (server's) credentials are copied from the
* unp_peercred member of socket on which the former called
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
* listen(); uipc_listen() cached that process's credentials
Re-wrap and other minor formatting and punctuation fixes for UNIX domain socket comments.
2006-07-22 17:24:55 +00:00
* at that time so we can use them now.
Implement a LOCAL_PEERCRED socket option which returns a `struct xucred` with the credentials of the connected peer. Obviously this only works (and makes sense) on SOCK_STREAM sockets. This works for both the connect(2) and listen(2) callers. There is precise documentation of the semantics in unix(4). Reviewed by: dwmalone (eyeballed)
2001-08-17 22:01:18 +00:00
*/
memcpy(&unp->unp_peercred, &unp2->unp_peercred,
sizeof(unp->unp_peercred));
unp->unp_flags |= UNP_HAVEPC;
Inherit LOCAL_CREDS option from listen socket for sockets returned by accept(2). PR: kern/90644 Submitted by: Andrey Simonenko OK'ed by: mdodd Tested by: NetBSD regress/sys/kern/unfdpass/unfdpass.c MFC after: 1 month
2006-04-24 19:09:33 +00:00
if (unp2->unp_flags & UNP_WANTCRED)
unp3->unp_flags |= UNP_WANTCRED;
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
UNP_PCB_UNLOCK(unp2);
Listening sockets improvements. o Separate fields of struct socket that belong to listening from fields that belong to normal dataflow, and unionize them. This shrinks the structure a bit. - Take out selinfo's from the socket buffers into the socket. The first reason is to support braindamaged scenario when a socket is added to kevent(2) and then listen(2) is cast on it. The second reason is that there is future plan to make socket buffers pluggable, so that for a dataflow socket a socket buffer can be changed, and in this case we also want to keep same selinfos through the lifetime of a socket. - Remove struct struct so_accf. Since now listening stuff no longer affects struct socket size, just move its fields into listening part of the union. - Provide sol_upcall field and enforce that so_upcall_set() may be called only on a dataflow socket, which has buffers, and for listening sockets provide solisten_upcall_set(). o Remove ACCEPT_LOCK() global. - Add a mutex to socket, to be used instead of socket buffer lock to lock fields of struct socket that don't belong to a socket buffer. - Allow to acquire two socket locks, but the first one must belong to a listening socket. - Make soref()/sorele() to use atomic(9). This allows in some situations to do soref() without owning socket lock. There is place for improvement here, it is possible to make sorele() also to lock optionally. - Most protocols aren't touched by this change, except UNIX local sockets. See below for more information. o Reduce copy-and-paste in kernel modules that accept connections from listening sockets: provide function solisten_dequeue(), and use it in the following modules: ctl(4), iscsi(4), ng_btsocket(4), ng_ksocket(4), infiniband, rpc. o UNIX local sockets. - Removal of ACCEPT_LOCK() global uncovered several races in the UNIX local sockets. Most races exist around spawning a new socket, when we are connecting to a local listening socket. To cover them, we need to hold locks on both PCBs when spawning a third one. This means holding them across sonewconn(). This creates a LOR between pcb locks and unp_list_lock. - To fix the new LOR, abandon the global unp_list_lock in favor of global unp_link_lock. Indeed, separating these two locks didn't provide us any extra parralelism in the UNIX sockets. - Now call into uipc_attach() may happen with unp_link_lock hold if, we are accepting, or without unp_link_lock in case if we are just creating a socket. - Another problem in UNIX sockets is that uipc_close() basicly did nothing for a listening socket. The vnode remained opened for connections. This is fixed by removing vnode in uipc_close(). Maybe the right way would be to do it for all sockets (not only listening), simply move the vnode teardown from uipc_detach() to uipc_close()? Sponsored by: Netflix Differential Revision: https://reviews.freebsd.org/D9770
2017-06-08 21:30:34 +00:00
unp2 = unp3;
Introduce support for Mandatory Access Control and extensible kernel access control. Invoke the necessary MAC entry points to maintain labels on sockets. In particular, invoke entry points during socket allocation and destruction, as well as creation by a process or during an accept-scenario (sonewconn). For UNIX domain sockets, also assign a peer label. As the socket code isn't locked down yet, locking interactions are not yet clear. Various protocol stack socket operations (such as peer label assignment for IPv4) will follow. Obtained from: TrustedBSD Project Sponsored by: DARPA, NAI Labs
2002-07-31 03:03:22 +00:00
#ifdef MAC
Listening sockets improvements. o Separate fields of struct socket that belong to listening from fields that belong to normal dataflow, and unionize them. This shrinks the structure a bit. - Take out selinfo's from the socket buffers into the socket. The first reason is to support braindamaged scenario when a socket is added to kevent(2) and then listen(2) is cast on it. The second reason is that there is future plan to make socket buffers pluggable, so that for a dataflow socket a socket buffer can be changed, and in this case we also want to keep same selinfos through the lifetime of a socket. - Remove struct struct so_accf. Since now listening stuff no longer affects struct socket size, just move its fields into listening part of the union. - Provide sol_upcall field and enforce that so_upcall_set() may be called only on a dataflow socket, which has buffers, and for listening sockets provide solisten_upcall_set(). o Remove ACCEPT_LOCK() global. - Add a mutex to socket, to be used instead of socket buffer lock to lock fields of struct socket that don't belong to a socket buffer. - Allow to acquire two socket locks, but the first one must belong to a listening socket. - Make soref()/sorele() to use atomic(9). This allows in some situations to do soref() without owning socket lock. There is place for improvement here, it is possible to make sorele() also to lock optionally. - Most protocols aren't touched by this change, except UNIX local sockets. See below for more information. o Reduce copy-and-paste in kernel modules that accept connections from listening sockets: provide function solisten_dequeue(), and use it in the following modules: ctl(4), iscsi(4), ng_btsocket(4), ng_ksocket(4), infiniband, rpc. o UNIX local sockets. - Removal of ACCEPT_LOCK() global uncovered several races in the UNIX local sockets. Most races exist around spawning a new socket, when we are connecting to a local listening socket. To cover them, we need to hold locks on both PCBs when spawning a third one. This means holding them across sonewconn(). This creates a LOR between pcb locks and unp_list_lock. - To fix the new LOR, abandon the global unp_list_lock in favor of global unp_link_lock. Indeed, separating these two locks didn't provide us any extra parralelism in the UNIX sockets. - Now call into uipc_attach() may happen with unp_link_lock hold if, we are accepting, or without unp_link_lock in case if we are just creating a socket. - Another problem in UNIX sockets is that uipc_close() basicly did nothing for a listening socket. The vnode remained opened for connections. This is fixed by removing vnode in uipc_close(). Maybe the right way would be to do it for all sockets (not only listening), simply move the vnode teardown from uipc_detach() to uipc_close()? Sponsored by: Netflix Differential Revision: https://reviews.freebsd.org/D9770
2017-06-08 21:30:34 +00:00
mac_socketpeer_set_from_socket(so, so2);
mac_socketpeer_set_from_socket(so2, so);
Introduce support for Mandatory Access Control and extensible kernel access control. Invoke the necessary MAC entry points to maintain labels on sockets. In particular, invoke entry points during socket allocation and destruction, as well as creation by a process or during an accept-scenario (sonewconn). For UNIX domain sockets, also assign a peer label. As the socket code isn't locked down yet, locking interactions are not yet clear. Various protocol stack socket operations (such as peer label assignment for IPv4) will follow. Obtained from: TrustedBSD Project Sponsored by: DARPA, NAI Labs
2002-07-31 03:03:22 +00:00
#endif
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
}
Listening sockets improvements. o Separate fields of struct socket that belong to listening from fields that belong to normal dataflow, and unionize them. This shrinks the structure a bit. - Take out selinfo's from the socket buffers into the socket. The first reason is to support braindamaged scenario when a socket is added to kevent(2) and then listen(2) is cast on it. The second reason is that there is future plan to make socket buffers pluggable, so that for a dataflow socket a socket buffer can be changed, and in this case we also want to keep same selinfos through the lifetime of a socket. - Remove struct struct so_accf. Since now listening stuff no longer affects struct socket size, just move its fields into listening part of the union. - Provide sol_upcall field and enforce that so_upcall_set() may be called only on a dataflow socket, which has buffers, and for listening sockets provide solisten_upcall_set(). o Remove ACCEPT_LOCK() global. - Add a mutex to socket, to be used instead of socket buffer lock to lock fields of struct socket that don't belong to a socket buffer. - Allow to acquire two socket locks, but the first one must belong to a listening socket. - Make soref()/sorele() to use atomic(9). This allows in some situations to do soref() without owning socket lock. There is place for improvement here, it is possible to make sorele() also to lock optionally. - Most protocols aren't touched by this change, except UNIX local sockets. See below for more information. o Reduce copy-and-paste in kernel modules that accept connections from listening sockets: provide function solisten_dequeue(), and use it in the following modules: ctl(4), iscsi(4), ng_btsocket(4), ng_ksocket(4), infiniband, rpc. o UNIX local sockets. - Removal of ACCEPT_LOCK() global uncovered several races in the UNIX local sockets. Most races exist around spawning a new socket, when we are connecting to a local listening socket. To cover them, we need to hold locks on both PCBs when spawning a third one. This means holding them across sonewconn(). This creates a LOR between pcb locks and unp_list_lock. - To fix the new LOR, abandon the global unp_list_lock in favor of global unp_link_lock. Indeed, separating these two locks didn't provide us any extra parralelism in the UNIX sockets. - Now call into uipc_attach() may happen with unp_link_lock hold if, we are accepting, or without unp_link_lock in case if we are just creating a socket. - Another problem in UNIX sockets is that uipc_close() basicly did nothing for a listening socket. The vnode remained opened for connections. This is fixed by removing vnode in uipc_close(). Maybe the right way would be to do it for all sockets (not only listening), simply move the vnode teardown from uipc_detach() to uipc_close()? Sponsored by: Netflix Differential Revision: https://reviews.freebsd.org/D9770
2017-06-08 21:30:34 +00:00
KASSERT(unp2 != NULL && so2 != NULL && unp2->unp_socket == so2 &&
sotounpcb(so2) == unp2,
("%s: unp2 %p so2 %p", __func__, unp2, so2));
Implement unix(4) socket options LOCAL_CREDS and LOCAL_CONNWAIT. - Add unp_addsockcred() (for LOCAL_CREDS). - Add an argument to unp_connect2() to differentiate between PRU_CONNECT and PRU_CONNECT2. (for LOCAL_CONNWAIT) Obtained from: NetBSD (with some changes)
2005-04-13 00:01:46 +00:00
error = unp_connect2(so, so2, PRU_CONNECT);
Listening sockets improvements. o Separate fields of struct socket that belong to listening from fields that belong to normal dataflow, and unionize them. This shrinks the structure a bit. - Take out selinfo's from the socket buffers into the socket. The first reason is to support braindamaged scenario when a socket is added to kevent(2) and then listen(2) is cast on it. The second reason is that there is future plan to make socket buffers pluggable, so that for a dataflow socket a socket buffer can be changed, and in this case we also want to keep same selinfos through the lifetime of a socket. - Remove struct struct so_accf. Since now listening stuff no longer affects struct socket size, just move its fields into listening part of the union. - Provide sol_upcall field and enforce that so_upcall_set() may be called only on a dataflow socket, which has buffers, and for listening sockets provide solisten_upcall_set(). o Remove ACCEPT_LOCK() global. - Add a mutex to socket, to be used instead of socket buffer lock to lock fields of struct socket that don't belong to a socket buffer. - Allow to acquire two socket locks, but the first one must belong to a listening socket. - Make soref()/sorele() to use atomic(9). This allows in some situations to do soref() without owning socket lock. There is place for improvement here, it is possible to make sorele() also to lock optionally. - Most protocols aren't touched by this change, except UNIX local sockets. See below for more information. o Reduce copy-and-paste in kernel modules that accept connections from listening sockets: provide function solisten_dequeue(), and use it in the following modules: ctl(4), iscsi(4), ng_btsocket(4), ng_ksocket(4), infiniband, rpc. o UNIX local sockets. - Removal of ACCEPT_LOCK() global uncovered several races in the UNIX local sockets. Most races exist around spawning a new socket, when we are connecting to a local listening socket. To cover them, we need to hold locks on both PCBs when spawning a third one. This means holding them across sonewconn(). This creates a LOR between pcb locks and unp_list_lock. - To fix the new LOR, abandon the global unp_list_lock in favor of global unp_link_lock. Indeed, separating these two locks didn't provide us any extra parralelism in the UNIX sockets. - Now call into uipc_attach() may happen with unp_link_lock hold if, we are accepting, or without unp_link_lock in case if we are just creating a socket. - Another problem in UNIX sockets is that uipc_close() basicly did nothing for a listening socket. The vnode remained opened for connections. This is fixed by removing vnode in uipc_close(). Maybe the right way would be to do it for all sockets (not only listening), simply move the vnode teardown from uipc_detach() to uipc_close()? Sponsored by: Netflix Differential Revision: https://reviews.freebsd.org/D9770
2017-06-08 21:30:34 +00:00
bad3:
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
UNP_PCB_UNLOCK(unp2);
UNP_PCB_UNLOCK(unp);
Introduce a subsystem lock around UNIX domain sockets in order to protect global and allocated variables. This strategy is derived from work originally developed by BSDi for BSD/OS, and applied to FreeBSD by Sam Leffler: - Add unp_mtx, a global mutex which will protect all UNIX domain socket related variables, structures, etc. - Add UNP_LOCK(), UNP_UNLOCK(), UNP_LOCK_ASSERT() macros. - Acquire unp_mtx on entering most UNIX domain socket code, drop/re-acquire around calls into VFS, and release it on return. - Avoid performing sodupsockaddr() while holding the mutex, so in general move to allocating storage before acquiring the mutex to copy the data. - Make a stack copy of the xucred rather than copying out while holding unp_mtx. Copy the peer credential out after releasing the mutex. - Add additional assertions of vnode locks following VOP_CREATE(). A few notes: - Use of an sx lock for the file list mutex may cause problems with regard to unp_mtx when garbage collection passed file descriptors. - The locking in unp_pcblist() for sysctl monitoring is correct subject to the unpcb zone not returning memory for reuse by other subsystems (consistent with similar existing concerns). - Sam's version of this change, as with the BSD/OS version, made use of both a global lock and per-unpcb locks. However, in practice, the global lock covered all accesses, so I have simplified out the unpcb locks in the interest of getting this merged faster (reducing the overhead but not sacrificing granularity in most cases). We will want to explore possibilities for improving lock granularity in this code in the future. Submitted by: sam Sponsored by: FreeBSD Foundatiuon Obtained from: BSD/OS 5 snapshot provided by BSDi
2004-06-10 21:34:38 +00:00
bad2:
Decompose the global UNIX domain sockets rwlock into two different locks: a global list/counter/generation counter protected by a new mutex unp_list_lock, and a global linkage rwlock, unp_global_rwlock, which protects the connections between UNIX domain sockets. This eliminates conditional lock acquisition that was previously a property of the global lock being held over sonewconn() leading to a call to uipc_attach(), which also required the global lock, but couldn't rely on it as other paths existed to uipc_attach() that didn't hold it: now uipc_attach() uses only the list lock, which follows the linkage lock in the lock order. It may also reduce contention on the global lock for some workloads. Add global UNIX domain socket locks to hard-coded witness lock order. MFC after: 1 week Discussed with: kris
2009-03-08 21:48:29 +00:00
UNP_LINK_WUNLOCK();
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
bad:
Introduce a subsystem lock around UNIX domain sockets in order to protect global and allocated variables. This strategy is derived from work originally developed by BSDi for BSD/OS, and applied to FreeBSD by Sam Leffler: - Add unp_mtx, a global mutex which will protect all UNIX domain socket related variables, structures, etc. - Add UNP_LOCK(), UNP_UNLOCK(), UNP_LOCK_ASSERT() macros. - Acquire unp_mtx on entering most UNIX domain socket code, drop/re-acquire around calls into VFS, and release it on return. - Avoid performing sodupsockaddr() while holding the mutex, so in general move to allocating storage before acquiring the mutex to copy the data. - Make a stack copy of the xucred rather than copying out while holding unp_mtx. Copy the peer credential out after releasing the mutex. - Add additional assertions of vnode locks following VOP_CREATE(). A few notes: - Use of an sx lock for the file list mutex may cause problems with regard to unp_mtx when garbage collection passed file descriptors. - The locking in unp_pcblist() for sysctl monitoring is correct subject to the unpcb zone not returning memory for reuse by other subsystems (consistent with similar existing concerns). - Sam's version of this change, as with the BSD/OS version, made use of both a global lock and per-unpcb locks. However, in practice, the global lock covered all accesses, so I have simplified out the unpcb locks in the interest of getting this merged faster (reducing the overhead but not sacrificing granularity in most cases). We will want to explore possibilities for improving lock granularity in this code in the future. Submitted by: sam Sponsored by: FreeBSD Foundatiuon Obtained from: BSD/OS 5 snapshot provided by BSDi
2004-06-10 21:34:38 +00:00
if (vp != NULL)
vput(vp);
free(sa, M_SONAME);
Decompose the global UNIX domain sockets rwlock into two different locks: a global list/counter/generation counter protected by a new mutex unp_list_lock, and a global linkage rwlock, unp_global_rwlock, which protects the connections between UNIX domain sockets. This eliminates conditional lock acquisition that was previously a property of the global lock being held over sonewconn() leading to a call to uipc_attach(), which also required the global lock, but couldn't rely on it as other paths existed to uipc_attach() that didn't hold it: now uipc_attach() uses only the list lock, which follows the linkage lock in the lock order. It may also reduce contention on the global lock for some workloads. Add global UNIX domain socket locks to hard-coded witness lock order. MFC after: 1 week Discussed with: kris
2009-03-08 21:48:29 +00:00
UNP_LINK_WLOCK();
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
UNP_PCB_LOCK(unp);
Add two new unpcb flags, UNP_BINDING and UNP_CONNECTING, which will be used to mark UNIX domain sockets as being in the process of binding or connecting. Use these to prevent simultaneous bind or connect operations by multiple threads or processes on the same socket at the same time, which closes race conditions present in the UNIX domain socket implementation since inception.
2006-07-23 12:01:14 +00:00
unp->unp_flags &= ~UNP_CONNECTING;
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
UNP_PCB_UNLOCK(unp);
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
return (error);
}
Export uipc_connect2() from uipc_usrreq.c instead of unp_connect2(), and consume that interface in portalfs and fifofs instead. In the new world order, unp_connect2() assumes that the unpcb mutex is held, whereas uipc_connect2() validates that the passed sockets are UNIX domain sockets, then grabs the mutex. NB: the portalfs and fifofs code gets down and dirty with UNIX domain sockets. Maybe this is a bad thing.
2004-03-31 01:41:30 +00:00
static int
Implement unix(4) socket options LOCAL_CREDS and LOCAL_CONNWAIT. - Add unp_addsockcred() (for LOCAL_CREDS). - Add an argument to unp_connect2() to differentiate between PRU_CONNECT and PRU_CONNECT2. (for LOCAL_CONNWAIT) Obtained from: NetBSD (with some changes)
2005-04-13 00:01:46 +00:00
unp_connect2(struct socket *so, struct socket *so2, int req)
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
{
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
struct unpcb *unp;
style(9)-ize function headers, remove use of 'register'. MFC after: 3 days
2005-02-20 23:22:13 +00:00
struct unpcb *unp2;
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
unp = sotounpcb(so);
KASSERT(unp != NULL, ("unp_connect2: unp == NULL"));
unp2 = sotounpcb(so2);
KASSERT(unp2 != NULL, ("unp_connect2: unp2 == NULL"));
Decompose the global UNIX domain sockets rwlock into two different locks: a global list/counter/generation counter protected by a new mutex unp_list_lock, and a global linkage rwlock, unp_global_rwlock, which protects the connections between UNIX domain sockets. This eliminates conditional lock acquisition that was previously a property of the global lock being held over sonewconn() leading to a call to uipc_attach(), which also required the global lock, but couldn't rely on it as other paths existed to uipc_attach() that didn't hold it: now uipc_attach() uses only the list lock, which follows the linkage lock in the lock order. It may also reduce contention on the global lock for some workloads. Add global UNIX domain socket locks to hard-coded witness lock order. MFC after: 1 week Discussed with: kris
2009-03-08 21:48:29 +00:00
UNP_LINK_WLOCK_ASSERT();
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
UNP_PCB_LOCK_ASSERT(unp);
UNP_PCB_LOCK_ASSERT(unp2);
Introduce a subsystem lock around UNIX domain sockets in order to protect global and allocated variables. This strategy is derived from work originally developed by BSDi for BSD/OS, and applied to FreeBSD by Sam Leffler: - Add unp_mtx, a global mutex which will protect all UNIX domain socket related variables, structures, etc. - Add UNP_LOCK(), UNP_UNLOCK(), UNP_LOCK_ASSERT() macros. - Acquire unp_mtx on entering most UNIX domain socket code, drop/re-acquire around calls into VFS, and release it on return. - Avoid performing sodupsockaddr() while holding the mutex, so in general move to allocating storage before acquiring the mutex to copy the data. - Make a stack copy of the xucred rather than copying out while holding unp_mtx. Copy the peer credential out after releasing the mutex. - Add additional assertions of vnode locks following VOP_CREATE(). A few notes: - Use of an sx lock for the file list mutex may cause problems with regard to unp_mtx when garbage collection passed file descriptors. - The locking in unp_pcblist() for sysctl monitoring is correct subject to the unpcb zone not returning memory for reuse by other subsystems (consistent with similar existing concerns). - Sam's version of this change, as with the BSD/OS version, made use of both a global lock and per-unpcb locks. However, in practice, the global lock covered all accesses, so I have simplified out the unpcb locks in the interest of getting this merged faster (reducing the overhead but not sacrificing granularity in most cases). We will want to explore possibilities for improving lock granularity in this code in the future. Submitted by: sam Sponsored by: FreeBSD Foundatiuon Obtained from: BSD/OS 5 snapshot provided by BSDi
2004-06-10 21:34:38 +00:00
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
if (so2->so_type != so->so_type)
return (EPROTOTYPE);
Handle races with listening socket close when connecting a unix socket. If the listening socket is closed while sonewconn() is executing, the nascent child socket is aborted, which results in recursion on the unp_link lock when the child's pru_detach method is invoked. Fix this by using a flag to mark such sockets, and skip a part of the socket's teardown during detach. Reported by: Raviprakash Darbha <rdarbha@juniper.net> Tested by: pho MFC after: 2 weeks Differential Revision: https://reviews.freebsd.org/D7398
2016-08-08 20:25:04 +00:00
unp2->unp_flags &= ~UNP_NASCENT;
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
unp->unp_conn = unp2;
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
switch (so->so_type) {
case SOCK_DGRAM:
Convert socket structures to be type-stable and add a version number. Define a parameter which indicates the maximum number of sockets in a system, and use this to size the zone allocators used for sockets and for certain PCBs. Convert PF_LOCAL PCB structures to be type-stable and add a version number. Define an external format for infomation about socket structures and use it in several places. Define a mechanism to get all PF_LOCAL and PF_INET PCB lists through sysctl(3) without blocking network interrupts for an unreasonable length of time. This probably still has some bugs and/or race conditions, but it seems to work well enough on my machines. It is now possible for `netstat' to get almost all of its information via the sysctl(3) interface rather than reading kmem (changes to follow).
1998-05-15 20:11:40 +00:00
LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink);
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
soisconnected(so);
break;
case SOCK_STREAM:
First cut at implementing SOCK_SEQPACKET support for UNIX (local) domain sockets. This allows for reliable bi-directional datagram communication over UNIX domain sockets, in contrast to SOCK_DGRAM (M:N, unreliable) or SOCK_STERAM (bi-directional bytestream). Largely, this reuses existing UNIX domain socket code. This allows applications requiring record- oriented semantics to do so reliably via local IPC. Some implementation notes (also present in XXX comments): - Currently we lack an sbappend variant able to do datagrams and control data without doing addresses, so we mark SOCK_SEQPACKET as PR_ADDR. Adding a new variant will solve this problem. - UNIX domain sockets on FreeBSD provide back-pressure/flow control notification for stream sockets by manipulating the send socket buffer's size during pru_send and pru_rcvd. This trick works less well for SOCK_SEQPACKET as sosend_generic() uses sb_hiwat not just to manage blocking, but also to determine maximum datagram size. Fixing this requires rethinking how back-pressure is done for SOCK_SEQPACKET; in the mean time, it's possible to get EMSGSIZE when buffers fill, instead of blocking. Discussed with: benl Reviewed by: bz, rpaulo MFC after: 3 months Sponsored by: Google
2009-10-05 14:49:16 +00:00
case SOCK_SEQPACKET:
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
unp2->unp_conn = unp;
Implement unix(4) socket options LOCAL_CREDS and LOCAL_CONNWAIT. - Add unp_addsockcred() (for LOCAL_CREDS). - Add an argument to unp_connect2() to differentiate between PRU_CONNECT and PRU_CONNECT2. (for LOCAL_CONNWAIT) Obtained from: NetBSD (with some changes)
2005-04-13 00:01:46 +00:00
if (req == PRU_CONNECT &&
((unp->unp_flags | unp2->unp_flags) & UNP_CONNWAIT))
soisconnecting(so);
else
soisconnected(so);
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
soisconnected(so2);
break;
default:
panic("unp_connect2");
}
return (0);
}
Another mega commit to staticize things.
1995-12-14 09:55:16 +00:00
static void
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
unp_disconnect(struct unpcb *unp, struct unpcb *unp2)
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
{
Hold SOCK_LOCK(so) when frobbing so_state when disconnecting a connected UNIX domain datagram socket.
2004-06-20 21:29:56 +00:00
struct socket *so;
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
KASSERT(unp2 != NULL, ("unp_disconnect: unp2 == NULL"));
Decompose the global UNIX domain sockets rwlock into two different locks: a global list/counter/generation counter protected by a new mutex unp_list_lock, and a global linkage rwlock, unp_global_rwlock, which protects the connections between UNIX domain sockets. This eliminates conditional lock acquisition that was previously a property of the global lock being held over sonewconn() leading to a call to uipc_attach(), which also required the global lock, but couldn't rely on it as other paths existed to uipc_attach() that didn't hold it: now uipc_attach() uses only the list lock, which follows the linkage lock in the lock order. It may also reduce contention on the global lock for some workloads. Add global UNIX domain socket locks to hard-coded witness lock order. MFC after: 1 week Discussed with: kris
2009-03-08 21:48:29 +00:00
UNP_LINK_WLOCK_ASSERT();
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
UNP_PCB_LOCK_ASSERT(unp);
UNP_PCB_LOCK_ASSERT(unp2);
Introduce a subsystem lock around UNIX domain sockets in order to protect global and allocated variables. This strategy is derived from work originally developed by BSDi for BSD/OS, and applied to FreeBSD by Sam Leffler: - Add unp_mtx, a global mutex which will protect all UNIX domain socket related variables, structures, etc. - Add UNP_LOCK(), UNP_UNLOCK(), UNP_LOCK_ASSERT() macros. - Acquire unp_mtx on entering most UNIX domain socket code, drop/re-acquire around calls into VFS, and release it on return. - Avoid performing sodupsockaddr() while holding the mutex, so in general move to allocating storage before acquiring the mutex to copy the data. - Make a stack copy of the xucred rather than copying out while holding unp_mtx. Copy the peer credential out after releasing the mutex. - Add additional assertions of vnode locks following VOP_CREATE(). A few notes: - Use of an sx lock for the file list mutex may cause problems with regard to unp_mtx when garbage collection passed file descriptors. - The locking in unp_pcblist() for sysctl monitoring is correct subject to the unpcb zone not returning memory for reuse by other subsystems (consistent with similar existing concerns). - Sam's version of this change, as with the BSD/OS version, made use of both a global lock and per-unpcb locks. However, in practice, the global lock covered all accesses, so I have simplified out the unpcb locks in the interest of getting this merged faster (reducing the overhead but not sacrificing granularity in most cases). We will want to explore possibilities for improving lock granularity in this code in the future. Submitted by: sam Sponsored by: FreeBSD Foundatiuon Obtained from: BSD/OS 5 snapshot provided by BSDi
2004-06-10 21:34:38 +00:00
Prefer NULL to 0 when testing and assigning pointer values.
2004-03-30 02:16:25 +00:00
unp->unp_conn = NULL;
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
switch (unp->unp_socket->so_type) {
case SOCK_DGRAM:
Convert socket structures to be type-stable and add a version number. Define a parameter which indicates the maximum number of sockets in a system, and use this to size the zone allocators used for sockets and for certain PCBs. Convert PF_LOCAL PCB structures to be type-stable and add a version number. Define an external format for infomation about socket structures and use it in several places. Define a mechanism to get all PF_LOCAL and PF_INET PCB lists through sysctl(3) without blocking network interrupts for an unreasonable length of time. This probably still has some bugs and/or race conditions, but it seems to work well enough on my machines. It is now possible for `netstat' to get almost all of its information via the sysctl(3) interface rather than reading kmem (changes to follow).
1998-05-15 20:11:40 +00:00
LIST_REMOVE(unp, unp_reflink);
Hold SOCK_LOCK(so) when frobbing so_state when disconnecting a connected UNIX domain datagram socket.
2004-06-20 21:29:56 +00:00
so = unp->unp_socket;
SOCK_LOCK(so);
so->so_state &= ~SS_ISCONNECTED;
SOCK_UNLOCK(so);
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
break;
case SOCK_STREAM:
First cut at implementing SOCK_SEQPACKET support for UNIX (local) domain sockets. This allows for reliable bi-directional datagram communication over UNIX domain sockets, in contrast to SOCK_DGRAM (M:N, unreliable) or SOCK_STERAM (bi-directional bytestream). Largely, this reuses existing UNIX domain socket code. This allows applications requiring record- oriented semantics to do so reliably via local IPC. Some implementation notes (also present in XXX comments): - Currently we lack an sbappend variant able to do datagrams and control data without doing addresses, so we mark SOCK_SEQPACKET as PR_ADDR. Adding a new variant will solve this problem. - UNIX domain sockets on FreeBSD provide back-pressure/flow control notification for stream sockets by manipulating the send socket buffer's size during pru_send and pru_rcvd. This trick works less well for SOCK_SEQPACKET as sosend_generic() uses sb_hiwat not just to manage blocking, but also to determine maximum datagram size. Fixing this requires rethinking how back-pressure is done for SOCK_SEQPACKET; in the mean time, it's possible to get EMSGSIZE when buffers fill, instead of blocking. Discussed with: benl Reviewed by: bz, rpaulo MFC after: 3 months Sponsored by: Google
2009-10-05 14:49:16 +00:00
case SOCK_SEQPACKET:
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
soisdisconnected(unp->unp_socket);
Prefer NULL to 0 when testing and assigning pointer values.
2004-03-30 02:16:25 +00:00
unp2->unp_conn = NULL;
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
soisdisconnected(unp2->unp_socket);
break;
}
}
Introduce a subsystem lock around UNIX domain sockets in order to protect global and allocated variables. This strategy is derived from work originally developed by BSDi for BSD/OS, and applied to FreeBSD by Sam Leffler: - Add unp_mtx, a global mutex which will protect all UNIX domain socket related variables, structures, etc. - Add UNP_LOCK(), UNP_UNLOCK(), UNP_LOCK_ASSERT() macros. - Acquire unp_mtx on entering most UNIX domain socket code, drop/re-acquire around calls into VFS, and release it on return. - Avoid performing sodupsockaddr() while holding the mutex, so in general move to allocating storage before acquiring the mutex to copy the data. - Make a stack copy of the xucred rather than copying out while holding unp_mtx. Copy the peer credential out after releasing the mutex. - Add additional assertions of vnode locks following VOP_CREATE(). A few notes: - Use of an sx lock for the file list mutex may cause problems with regard to unp_mtx when garbage collection passed file descriptors. - The locking in unp_pcblist() for sysctl monitoring is correct subject to the unpcb zone not returning memory for reuse by other subsystems (consistent with similar existing concerns). - Sam's version of this change, as with the BSD/OS version, made use of both a global lock and per-unpcb locks. However, in practice, the global lock covered all accesses, so I have simplified out the unpcb locks in the interest of getting this merged faster (reducing the overhead but not sacrificing granularity in most cases). We will want to explore possibilities for improving lock granularity in this code in the future. Submitted by: sam Sponsored by: FreeBSD Foundatiuon Obtained from: BSD/OS 5 snapshot provided by BSDi
2004-06-10 21:34:38 +00:00
/*
Clarify and update quite a few comments to reflect locking optimizations, the addition of unpcb refcounts, and bug fixes. Some of these fixes are appropriate for MFC. MFC after: 3 days
2007-05-11 12:10:45 +00:00
* unp_pcblist() walks the global list of struct unpcb's to generate a
* pointer list, bumping the refcount on each unpcb. It then copies them out
* sequentially, validating the generation number on each to see if it has
* been detached. All of this is necessary because copyout() may sleep on
* disk I/O.
Introduce a subsystem lock around UNIX domain sockets in order to protect global and allocated variables. This strategy is derived from work originally developed by BSDi for BSD/OS, and applied to FreeBSD by Sam Leffler: - Add unp_mtx, a global mutex which will protect all UNIX domain socket related variables, structures, etc. - Add UNP_LOCK(), UNP_UNLOCK(), UNP_LOCK_ASSERT() macros. - Acquire unp_mtx on entering most UNIX domain socket code, drop/re-acquire around calls into VFS, and release it on return. - Avoid performing sodupsockaddr() while holding the mutex, so in general move to allocating storage before acquiring the mutex to copy the data. - Make a stack copy of the xucred rather than copying out while holding unp_mtx. Copy the peer credential out after releasing the mutex. - Add additional assertions of vnode locks following VOP_CREATE(). A few notes: - Use of an sx lock for the file list mutex may cause problems with regard to unp_mtx when garbage collection passed file descriptors. - The locking in unp_pcblist() for sysctl monitoring is correct subject to the unpcb zone not returning memory for reuse by other subsystems (consistent with similar existing concerns). - Sam's version of this change, as with the BSD/OS version, made use of both a global lock and per-unpcb locks. However, in practice, the global lock covered all accesses, so I have simplified out the unpcb locks in the interest of getting this merged faster (reducing the overhead but not sacrificing granularity in most cases). We will want to explore possibilities for improving lock granularity in this code in the future. Submitted by: sam Sponsored by: FreeBSD Foundatiuon Obtained from: BSD/OS 5 snapshot provided by BSDi
2004-06-10 21:34:38 +00:00
*/
Convert socket structures to be type-stable and add a version number. Define a parameter which indicates the maximum number of sockets in a system, and use this to size the zone allocators used for sockets and for certain PCBs. Convert PF_LOCAL PCB structures to be type-stable and add a version number. Define an external format for infomation about socket structures and use it in several places. Define a mechanism to get all PF_LOCAL and PF_INET PCB lists through sysctl(3) without blocking network interrupts for an unreasonable length of time. This probably still has some bugs and/or race conditions, but it seems to work well enough on my machines. It is now possible for `netstat' to get almost all of its information via the sysctl(3) interface rather than reading kmem (changes to follow).
1998-05-15 20:11:40 +00:00
static int
Previous commit changing SYSCTL_HANDLER_ARGS violated KNF. Pointed out by: bde
2000-07-04 11:25:35 +00:00
unp_pcblist(SYSCTL_HANDLER_ARGS)
Convert socket structures to be type-stable and add a version number. Define a parameter which indicates the maximum number of sockets in a system, and use this to size the zone allocators used for sockets and for certain PCBs. Convert PF_LOCAL PCB structures to be type-stable and add a version number. Define an external format for infomation about socket structures and use it in several places. Define a mechanism to get all PF_LOCAL and PF_INET PCB lists through sysctl(3) without blocking network interrupts for an unreasonable length of time. This probably still has some bugs and/or race conditions, but it seems to work well enough on my machines. It is now possible for `netstat' to get almost all of its information via the sysctl(3) interface rather than reading kmem (changes to follow).
1998-05-15 20:11:40 +00:00
{
Nitpicking and dusting performed on a train. Removes trivial warnings about unused variables, labels and other lint.
1998-10-25 17:44:59 +00:00
int error, i, n;
- Close a race between enumerating UNIX domain socket pcb structures via sysctl and socket teardown by adding a reference count to the UNIX domain pcb object and fixing the sysctl that enumerates unpcbs to grab a reference on each unpcb while it builds the list to copy out to userland. - Close a race between UNIX domain pcb garbage collection (unp_gc()) and file descriptor teardown (fdrop()) by adding a new garbage collection flag FWAIT. unp_gc() sets FWAIT while it walks the message buffers in a UNIX domain socket looking for nested file descriptor references and clears the flag when it is finished. fdrop() checks to see if the flag is set on a file descriptor whose refcount just dropped to 0 and waits for unp_gc() to clear the flag before completely destroying the file descriptor. MFC after: 1 week Reviewed by: rwatson Submitted by: ups Hopefully makes the panics go away: mx1
2007-01-05 19:59:46 +00:00
int freeunp;
Convert socket structures to be type-stable and add a version number. Define a parameter which indicates the maximum number of sockets in a system, and use this to size the zone allocators used for sockets and for certain PCBs. Convert PF_LOCAL PCB structures to be type-stable and add a version number. Define an external format for infomation about socket structures and use it in several places. Define a mechanism to get all PF_LOCAL and PF_INET PCB lists through sysctl(3) without blocking network interrupts for an unreasonable length of time. This probably still has some bugs and/or race conditions, but it seems to work well enough on my machines. It is now possible for `netstat' to get almost all of its information via the sysctl(3) interface rather than reading kmem (changes to follow).
1998-05-15 20:11:40 +00:00
struct unpcb *unp, **unp_list;
unp_gen_t gencnt;
Don't alocate a 400 byte buffer on the stack, Nor 800 bytes of structures.. MFC after: 2 weeks
2001-08-18 02:53:50 +00:00
struct xunpgen *xug;
Convert socket structures to be type-stable and add a version number. Define a parameter which indicates the maximum number of sockets in a system, and use this to size the zone allocators used for sockets and for certain PCBs. Convert PF_LOCAL PCB structures to be type-stable and add a version number. Define an external format for infomation about socket structures and use it in several places. Define a mechanism to get all PF_LOCAL and PF_INET PCB lists through sysctl(3) without blocking network interrupts for an unreasonable length of time. This probably still has some bugs and/or race conditions, but it seems to work well enough on my machines. It is now possible for `netstat' to get almost all of its information via the sysctl(3) interface rather than reading kmem (changes to follow).
1998-05-15 20:11:40 +00:00
struct unp_head *head;
Don't alocate a 400 byte buffer on the stack, Nor 800 bytes of structures.. MFC after: 2 weeks
2001-08-18 02:53:50 +00:00
struct xunpcb *xu;
Convert socket structures to be type-stable and add a version number. Define a parameter which indicates the maximum number of sockets in a system, and use this to size the zone allocators used for sockets and for certain PCBs. Convert PF_LOCAL PCB structures to be type-stable and add a version number. Define an external format for infomation about socket structures and use it in several places. Define a mechanism to get all PF_LOCAL and PF_INET PCB lists through sysctl(3) without blocking network interrupts for an unreasonable length of time. This probably still has some bugs and/or race conditions, but it seems to work well enough on my machines. It is now possible for `netstat' to get almost all of its information via the sysctl(3) interface rather than reading kmem (changes to follow).
1998-05-15 20:11:40 +00:00
First cut at implementing SOCK_SEQPACKET support for UNIX (local) domain sockets. This allows for reliable bi-directional datagram communication over UNIX domain sockets, in contrast to SOCK_DGRAM (M:N, unreliable) or SOCK_STERAM (bi-directional bytestream). Largely, this reuses existing UNIX domain socket code. This allows applications requiring record- oriented semantics to do so reliably via local IPC. Some implementation notes (also present in XXX comments): - Currently we lack an sbappend variant able to do datagrams and control data without doing addresses, so we mark SOCK_SEQPACKET as PR_ADDR. Adding a new variant will solve this problem. - UNIX domain sockets on FreeBSD provide back-pressure/flow control notification for stream sockets by manipulating the send socket buffer's size during pru_send and pru_rcvd. This trick works less well for SOCK_SEQPACKET as sosend_generic() uses sb_hiwat not just to manage blocking, but also to determine maximum datagram size. Fixing this requires rethinking how back-pressure is done for SOCK_SEQPACKET; in the mean time, it's possible to get EMSGSIZE when buffers fill, instead of blocking. Discussed with: benl Reviewed by: bz, rpaulo MFC after: 3 months Sponsored by: Google
2009-10-05 14:49:16 +00:00
switch ((intptr_t)arg1) {
case SOCK_STREAM:
head = &unp_shead;
break;
case SOCK_DGRAM:
head = &unp_dhead;
break;
case SOCK_SEQPACKET:
head = &unp_sphead;
break;
default:
Fix build on amd64, where sysctl arg1 is a pointer. Reported by: Mr Tinderbox MFC after: 3 months
2009-10-05 22:23:12 +00:00
panic("unp_pcblist: arg1 %d", (int)(intptr_t)arg1);
First cut at implementing SOCK_SEQPACKET support for UNIX (local) domain sockets. This allows for reliable bi-directional datagram communication over UNIX domain sockets, in contrast to SOCK_DGRAM (M:N, unreliable) or SOCK_STERAM (bi-directional bytestream). Largely, this reuses existing UNIX domain socket code. This allows applications requiring record- oriented semantics to do so reliably via local IPC. Some implementation notes (also present in XXX comments): - Currently we lack an sbappend variant able to do datagrams and control data without doing addresses, so we mark SOCK_SEQPACKET as PR_ADDR. Adding a new variant will solve this problem. - UNIX domain sockets on FreeBSD provide back-pressure/flow control notification for stream sockets by manipulating the send socket buffer's size during pru_send and pru_rcvd. This trick works less well for SOCK_SEQPACKET as sosend_generic() uses sb_hiwat not just to manage blocking, but also to determine maximum datagram size. Fixing this requires rethinking how back-pressure is done for SOCK_SEQPACKET; in the mean time, it's possible to get EMSGSIZE when buffers fill, instead of blocking. Discussed with: benl Reviewed by: bz, rpaulo MFC after: 3 months Sponsored by: Google
2009-10-05 14:49:16 +00:00
}
Convert socket structures to be type-stable and add a version number. Define a parameter which indicates the maximum number of sockets in a system, and use this to size the zone allocators used for sockets and for certain PCBs. Convert PF_LOCAL PCB structures to be type-stable and add a version number. Define an external format for infomation about socket structures and use it in several places. Define a mechanism to get all PF_LOCAL and PF_INET PCB lists through sysctl(3) without blocking network interrupts for an unreasonable length of time. This probably still has some bugs and/or race conditions, but it seems to work well enough on my machines. It is now possible for `netstat' to get almost all of its information via the sysctl(3) interface rather than reading kmem (changes to follow).
1998-05-15 20:11:40 +00:00
/*
* The process of preparing the PCB list is too time-consuming and
* resource-intensive to repeat twice on every request.
*/
Prefer NULL to 0 when testing and assigning pointer values.
2004-03-30 02:16:25 +00:00
if (req->oldptr == NULL) {
Convert socket structures to be type-stable and add a version number. Define a parameter which indicates the maximum number of sockets in a system, and use this to size the zone allocators used for sockets and for certain PCBs. Convert PF_LOCAL PCB structures to be type-stable and add a version number. Define an external format for infomation about socket structures and use it in several places. Define a mechanism to get all PF_LOCAL and PF_INET PCB lists through sysctl(3) without blocking network interrupts for an unreasonable length of time. This probably still has some bugs and/or race conditions, but it seems to work well enough on my machines. It is now possible for `netstat' to get almost all of its information via the sysctl(3) interface rather than reading kmem (changes to follow).
1998-05-15 20:11:40 +00:00
n = unp_count;
Don't alocate a 400 byte buffer on the stack, Nor 800 bytes of structures.. MFC after: 2 weeks
2001-08-18 02:53:50 +00:00
req->oldidx = 2 * (sizeof *xug)
Convert socket structures to be type-stable and add a version number. Define a parameter which indicates the maximum number of sockets in a system, and use this to size the zone allocators used for sockets and for certain PCBs. Convert PF_LOCAL PCB structures to be type-stable and add a version number. Define an external format for infomation about socket structures and use it in several places. Define a mechanism to get all PF_LOCAL and PF_INET PCB lists through sysctl(3) without blocking network interrupts for an unreasonable length of time. This probably still has some bugs and/or race conditions, but it seems to work well enough on my machines. It is now possible for `netstat' to get almost all of its information via the sysctl(3) interface rather than reading kmem (changes to follow).
1998-05-15 20:11:40 +00:00
+ (n + n/8) * sizeof(struct xunpcb);
Mechanical whitespace cleanup; parenthesize return values; other minor style nits.
2004-01-11 19:48:19 +00:00
return (0);
Convert socket structures to be type-stable and add a version number. Define a parameter which indicates the maximum number of sockets in a system, and use this to size the zone allocators used for sockets and for certain PCBs. Convert PF_LOCAL PCB structures to be type-stable and add a version number. Define an external format for infomation about socket structures and use it in several places. Define a mechanism to get all PF_LOCAL and PF_INET PCB lists through sysctl(3) without blocking network interrupts for an unreasonable length of time. This probably still has some bugs and/or race conditions, but it seems to work well enough on my machines. It is now possible for `netstat' to get almost all of its information via the sysctl(3) interface rather than reading kmem (changes to follow).
1998-05-15 20:11:40 +00:00
}
Prefer NULL to 0 when testing and assigning pointer values.
2004-03-30 02:16:25 +00:00
if (req->newptr != NULL)
Mechanical whitespace cleanup; parenthesize return values; other minor style nits.
2004-01-11 19:48:19 +00:00
return (EPERM);
Convert socket structures to be type-stable and add a version number. Define a parameter which indicates the maximum number of sockets in a system, and use this to size the zone allocators used for sockets and for certain PCBs. Convert PF_LOCAL PCB structures to be type-stable and add a version number. Define an external format for infomation about socket structures and use it in several places. Define a mechanism to get all PF_LOCAL and PF_INET PCB lists through sysctl(3) without blocking network interrupts for an unreasonable length of time. This probably still has some bugs and/or race conditions, but it seems to work well enough on my machines. It is now possible for `netstat' to get almost all of its information via the sysctl(3) interface rather than reading kmem (changes to follow).
1998-05-15 20:11:40 +00:00
/*
* OK, now we're committed to doing something.
*/
Back out M_* changes, per decision of the TRB. Approved by: trb
2003-02-19 05:47:46 +00:00
xug = malloc(sizeof(*xug), M_TEMP, M_WAITOK);
Listening sockets improvements. o Separate fields of struct socket that belong to listening from fields that belong to normal dataflow, and unionize them. This shrinks the structure a bit. - Take out selinfo's from the socket buffers into the socket. The first reason is to support braindamaged scenario when a socket is added to kevent(2) and then listen(2) is cast on it. The second reason is that there is future plan to make socket buffers pluggable, so that for a dataflow socket a socket buffer can be changed, and in this case we also want to keep same selinfos through the lifetime of a socket. - Remove struct struct so_accf. Since now listening stuff no longer affects struct socket size, just move its fields into listening part of the union. - Provide sol_upcall field and enforce that so_upcall_set() may be called only on a dataflow socket, which has buffers, and for listening sockets provide solisten_upcall_set(). o Remove ACCEPT_LOCK() global. - Add a mutex to socket, to be used instead of socket buffer lock to lock fields of struct socket that don't belong to a socket buffer. - Allow to acquire two socket locks, but the first one must belong to a listening socket. - Make soref()/sorele() to use atomic(9). This allows in some situations to do soref() without owning socket lock. There is place for improvement here, it is possible to make sorele() also to lock optionally. - Most protocols aren't touched by this change, except UNIX local sockets. See below for more information. o Reduce copy-and-paste in kernel modules that accept connections from listening sockets: provide function solisten_dequeue(), and use it in the following modules: ctl(4), iscsi(4), ng_btsocket(4), ng_ksocket(4), infiniband, rpc. o UNIX local sockets. - Removal of ACCEPT_LOCK() global uncovered several races in the UNIX local sockets. Most races exist around spawning a new socket, when we are connecting to a local listening socket. To cover them, we need to hold locks on both PCBs when spawning a third one. This means holding them across sonewconn(). This creates a LOR between pcb locks and unp_list_lock. - To fix the new LOR, abandon the global unp_list_lock in favor of global unp_link_lock. Indeed, separating these two locks didn't provide us any extra parralelism in the UNIX sockets. - Now call into uipc_attach() may happen with unp_link_lock hold if, we are accepting, or without unp_link_lock in case if we are just creating a socket. - Another problem in UNIX sockets is that uipc_close() basicly did nothing for a listening socket. The vnode remained opened for connections. This is fixed by removing vnode in uipc_close(). Maybe the right way would be to do it for all sockets (not only listening), simply move the vnode teardown from uipc_detach() to uipc_close()? Sponsored by: Netflix Differential Revision: https://reviews.freebsd.org/D9770
2017-06-08 21:30:34 +00:00
UNP_LINK_RLOCK();
Convert socket structures to be type-stable and add a version number. Define a parameter which indicates the maximum number of sockets in a system, and use this to size the zone allocators used for sockets and for certain PCBs. Convert PF_LOCAL PCB structures to be type-stable and add a version number. Define an external format for infomation about socket structures and use it in several places. Define a mechanism to get all PF_LOCAL and PF_INET PCB lists through sysctl(3) without blocking network interrupts for an unreasonable length of time. This probably still has some bugs and/or race conditions, but it seems to work well enough on my machines. It is now possible for `netstat' to get almost all of its information via the sysctl(3) interface rather than reading kmem (changes to follow).
1998-05-15 20:11:40 +00:00
gencnt = unp_gencnt;
n = unp_count;
Listening sockets improvements. o Separate fields of struct socket that belong to listening from fields that belong to normal dataflow, and unionize them. This shrinks the structure a bit. - Take out selinfo's from the socket buffers into the socket. The first reason is to support braindamaged scenario when a socket is added to kevent(2) and then listen(2) is cast on it. The second reason is that there is future plan to make socket buffers pluggable, so that for a dataflow socket a socket buffer can be changed, and in this case we also want to keep same selinfos through the lifetime of a socket. - Remove struct struct so_accf. Since now listening stuff no longer affects struct socket size, just move its fields into listening part of the union. - Provide sol_upcall field and enforce that so_upcall_set() may be called only on a dataflow socket, which has buffers, and for listening sockets provide solisten_upcall_set(). o Remove ACCEPT_LOCK() global. - Add a mutex to socket, to be used instead of socket buffer lock to lock fields of struct socket that don't belong to a socket buffer. - Allow to acquire two socket locks, but the first one must belong to a listening socket. - Make soref()/sorele() to use atomic(9). This allows in some situations to do soref() without owning socket lock. There is place for improvement here, it is possible to make sorele() also to lock optionally. - Most protocols aren't touched by this change, except UNIX local sockets. See below for more information. o Reduce copy-and-paste in kernel modules that accept connections from listening sockets: provide function solisten_dequeue(), and use it in the following modules: ctl(4), iscsi(4), ng_btsocket(4), ng_ksocket(4), infiniband, rpc. o UNIX local sockets. - Removal of ACCEPT_LOCK() global uncovered several races in the UNIX local sockets. Most races exist around spawning a new socket, when we are connecting to a local listening socket. To cover them, we need to hold locks on both PCBs when spawning a third one. This means holding them across sonewconn(). This creates a LOR between pcb locks and unp_list_lock. - To fix the new LOR, abandon the global unp_list_lock in favor of global unp_link_lock. Indeed, separating these two locks didn't provide us any extra parralelism in the UNIX sockets. - Now call into uipc_attach() may happen with unp_link_lock hold if, we are accepting, or without unp_link_lock in case if we are just creating a socket. - Another problem in UNIX sockets is that uipc_close() basicly did nothing for a listening socket. The vnode remained opened for connections. This is fixed by removing vnode in uipc_close(). Maybe the right way would be to do it for all sockets (not only listening), simply move the vnode teardown from uipc_detach() to uipc_close()? Sponsored by: Netflix Differential Revision: https://reviews.freebsd.org/D9770
2017-06-08 21:30:34 +00:00
UNP_LINK_RUNLOCK();
Convert socket structures to be type-stable and add a version number. Define a parameter which indicates the maximum number of sockets in a system, and use this to size the zone allocators used for sockets and for certain PCBs. Convert PF_LOCAL PCB structures to be type-stable and add a version number. Define an external format for infomation about socket structures and use it in several places. Define a mechanism to get all PF_LOCAL and PF_INET PCB lists through sysctl(3) without blocking network interrupts for an unreasonable length of time. This probably still has some bugs and/or race conditions, but it seems to work well enough on my machines. It is now possible for `netstat' to get almost all of its information via the sysctl(3) interface rather than reading kmem (changes to follow).
1998-05-15 20:11:40 +00:00
Don't alocate a 400 byte buffer on the stack, Nor 800 bytes of structures.. MFC after: 2 weeks
2001-08-18 02:53:50 +00:00
xug->xug_len = sizeof *xug;
xug->xug_count = n;
xug->xug_gen = gencnt;
xug->xug_sogen = so_gencnt;
error = SYSCTL_OUT(req, xug, sizeof *xug);
if (error) {
free(xug, M_TEMP);
Mechanical whitespace cleanup; parenthesize return values; other minor style nits.
2004-01-11 19:48:19 +00:00
return (error);
Don't alocate a 400 byte buffer on the stack, Nor 800 bytes of structures.. MFC after: 2 weeks
2001-08-18 02:53:50 +00:00
}
Convert socket structures to be type-stable and add a version number. Define a parameter which indicates the maximum number of sockets in a system, and use this to size the zone allocators used for sockets and for certain PCBs. Convert PF_LOCAL PCB structures to be type-stable and add a version number. Define an external format for infomation about socket structures and use it in several places. Define a mechanism to get all PF_LOCAL and PF_INET PCB lists through sysctl(3) without blocking network interrupts for an unreasonable length of time. This probably still has some bugs and/or race conditions, but it seems to work well enough on my machines. It is now possible for `netstat' to get almost all of its information via the sysctl(3) interface rather than reading kmem (changes to follow).
1998-05-15 20:11:40 +00:00
Back out M_* changes, per decision of the TRB. Approved by: trb
2003-02-19 05:47:46 +00:00
unp_list = malloc(n * sizeof *unp_list, M_TEMP, M_WAITOK);
Mechanical whitespace cleanup; parenthesize return values; other minor style nits.
2004-01-11 19:48:19 +00:00
Listening sockets improvements. o Separate fields of struct socket that belong to listening from fields that belong to normal dataflow, and unionize them. This shrinks the structure a bit. - Take out selinfo's from the socket buffers into the socket. The first reason is to support braindamaged scenario when a socket is added to kevent(2) and then listen(2) is cast on it. The second reason is that there is future plan to make socket buffers pluggable, so that for a dataflow socket a socket buffer can be changed, and in this case we also want to keep same selinfos through the lifetime of a socket. - Remove struct struct so_accf. Since now listening stuff no longer affects struct socket size, just move its fields into listening part of the union. - Provide sol_upcall field and enforce that so_upcall_set() may be called only on a dataflow socket, which has buffers, and for listening sockets provide solisten_upcall_set(). o Remove ACCEPT_LOCK() global. - Add a mutex to socket, to be used instead of socket buffer lock to lock fields of struct socket that don't belong to a socket buffer. - Allow to acquire two socket locks, but the first one must belong to a listening socket. - Make soref()/sorele() to use atomic(9). This allows in some situations to do soref() without owning socket lock. There is place for improvement here, it is possible to make sorele() also to lock optionally. - Most protocols aren't touched by this change, except UNIX local sockets. See below for more information. o Reduce copy-and-paste in kernel modules that accept connections from listening sockets: provide function solisten_dequeue(), and use it in the following modules: ctl(4), iscsi(4), ng_btsocket(4), ng_ksocket(4), infiniband, rpc. o UNIX local sockets. - Removal of ACCEPT_LOCK() global uncovered several races in the UNIX local sockets. Most races exist around spawning a new socket, when we are connecting to a local listening socket. To cover them, we need to hold locks on both PCBs when spawning a third one. This means holding them across sonewconn(). This creates a LOR between pcb locks and unp_list_lock. - To fix the new LOR, abandon the global unp_list_lock in favor of global unp_link_lock. Indeed, separating these two locks didn't provide us any extra parralelism in the UNIX sockets. - Now call into uipc_attach() may happen with unp_link_lock hold if, we are accepting, or without unp_link_lock in case if we are just creating a socket. - Another problem in UNIX sockets is that uipc_close() basicly did nothing for a listening socket. The vnode remained opened for connections. This is fixed by removing vnode in uipc_close(). Maybe the right way would be to do it for all sockets (not only listening), simply move the vnode teardown from uipc_detach() to uipc_close()? Sponsored by: Netflix Differential Revision: https://reviews.freebsd.org/D9770
2017-06-08 21:30:34 +00:00
UNP_LINK_RLOCK();
This is a partial commit of the patch from PR 14914: Alot of the code in sys/kern directly accesses the *Q_HEAD and *Q_ENTRY structures for list operations. This patch makes all list operations in sys/kern use the queue(3) macros, rather than directly accessing the *Q_{HEAD,ENTRY} structures. This batch of changes compile to the same object files. Reviewed by: phk Submitted by: Jake Burkholder <jake@checker.org> PR: 14914
1999-11-16 10:56:05 +00:00
for (unp = LIST_FIRST(head), i = 0; unp && i < n;
unp = LIST_NEXT(unp, unp_link)) {
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
UNP_PCB_LOCK(unp);
- Combine kern.ps_showallprocs and kern.ipc.showallsockets into a single kern.security.seeotheruids_permitted, describes as: "Unprivileged processes may see subjects/objects with different real uid" NOTE: kern.ps_showallprocs exists in -STABLE, and therefore there is an API change. kern.ipc.showallsockets does not. - Check kern.security.seeotheruids_permitted in cr_cansee(). - Replace visibility calls to socheckuid() with cr_cansee() (retain the change to socheckuid() in ipfw, where it is used for rule-matching). - Remove prison_unpcb() and make use of cr_cansee() against the UNIX domain socket credential instead of comparing root vnodes for the UDS and the process. This allows multiple jails to share the same chroot() and not see each others UNIX domain sockets. - Remove unused socheckproc(). Now that cr_cansee() is used universally for socket visibility, a variety of policies are more consistently enforced, including uid-based restrictions and jail-based restrictions. This also better-supports the introduction of additional MAC models. Reviewed by: ps, billf Obtained from: TrustedBSD Project
2001-10-09 21:40:30 +00:00
if (unp->unp_gencnt <= gencnt) {
Simple p_ucred -> td_ucred changes to start using the per-thread ucred reference.
2002-02-27 18:32:23 +00:00
if (cr_cansee(req->td->td_ucred,
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
unp->unp_socket->so_cred)) {
UNP_PCB_UNLOCK(unp);
Only allow users to see their own socket connections if kern.ipc.showallsockets is set to 0. Submitted by: billf (with modifications by me) Inspired by: Dave McKay (aka pm aka Packet Magnet) Reviewed by: peter MFC after: 2 weeks
2001-10-05 07:06:32 +00:00
continue;
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
}
Convert socket structures to be type-stable and add a version number. Define a parameter which indicates the maximum number of sockets in a system, and use this to size the zone allocators used for sockets and for certain PCBs. Convert PF_LOCAL PCB structures to be type-stable and add a version number. Define an external format for infomation about socket structures and use it in several places. Define a mechanism to get all PF_LOCAL and PF_INET PCB lists through sysctl(3) without blocking network interrupts for an unreasonable length of time. This probably still has some bugs and/or race conditions, but it seems to work well enough on my machines. It is now possible for `netstat' to get almost all of its information via the sysctl(3) interface rather than reading kmem (changes to follow).
1998-05-15 20:11:40 +00:00
unp_list[i++] = unp;
- Close a race between enumerating UNIX domain socket pcb structures via sysctl and socket teardown by adding a reference count to the UNIX domain pcb object and fixing the sysctl that enumerates unpcbs to grab a reference on each unpcb while it builds the list to copy out to userland. - Close a race between UNIX domain pcb garbage collection (unp_gc()) and file descriptor teardown (fdrop()) by adding a new garbage collection flag FWAIT. unp_gc() sets FWAIT while it walks the message buffers in a UNIX domain socket looking for nested file descriptor references and clears the flag when it is finished. fdrop() checks to see if the flag is set on a file descriptor whose refcount just dropped to 0 and waits for unp_gc() to clear the flag before completely destroying the file descriptor. MFC after: 1 week Reviewed by: rwatson Submitted by: ups Hopefully makes the panics go away: mx1
2007-01-05 19:59:46 +00:00
unp->unp_refcount++;
Only allow users to see their own socket connections if kern.ipc.showallsockets is set to 0. Submitted by: billf (with modifications by me) Inspired by: Dave McKay (aka pm aka Packet Magnet) Reviewed by: peter MFC after: 2 weeks
2001-10-05 07:06:32 +00:00
}
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
UNP_PCB_UNLOCK(unp);
Convert socket structures to be type-stable and add a version number. Define a parameter which indicates the maximum number of sockets in a system, and use this to size the zone allocators used for sockets and for certain PCBs. Convert PF_LOCAL PCB structures to be type-stable and add a version number. Define an external format for infomation about socket structures and use it in several places. Define a mechanism to get all PF_LOCAL and PF_INET PCB lists through sysctl(3) without blocking network interrupts for an unreasonable length of time. This probably still has some bugs and/or race conditions, but it seems to work well enough on my machines. It is now possible for `netstat' to get almost all of its information via the sysctl(3) interface rather than reading kmem (changes to follow).
1998-05-15 20:11:40 +00:00
}
Listening sockets improvements. o Separate fields of struct socket that belong to listening from fields that belong to normal dataflow, and unionize them. This shrinks the structure a bit. - Take out selinfo's from the socket buffers into the socket. The first reason is to support braindamaged scenario when a socket is added to kevent(2) and then listen(2) is cast on it. The second reason is that there is future plan to make socket buffers pluggable, so that for a dataflow socket a socket buffer can be changed, and in this case we also want to keep same selinfos through the lifetime of a socket. - Remove struct struct so_accf. Since now listening stuff no longer affects struct socket size, just move its fields into listening part of the union. - Provide sol_upcall field and enforce that so_upcall_set() may be called only on a dataflow socket, which has buffers, and for listening sockets provide solisten_upcall_set(). o Remove ACCEPT_LOCK() global. - Add a mutex to socket, to be used instead of socket buffer lock to lock fields of struct socket that don't belong to a socket buffer. - Allow to acquire two socket locks, but the first one must belong to a listening socket. - Make soref()/sorele() to use atomic(9). This allows in some situations to do soref() without owning socket lock. There is place for improvement here, it is possible to make sorele() also to lock optionally. - Most protocols aren't touched by this change, except UNIX local sockets. See below for more information. o Reduce copy-and-paste in kernel modules that accept connections from listening sockets: provide function solisten_dequeue(), and use it in the following modules: ctl(4), iscsi(4), ng_btsocket(4), ng_ksocket(4), infiniband, rpc. o UNIX local sockets. - Removal of ACCEPT_LOCK() global uncovered several races in the UNIX local sockets. Most races exist around spawning a new socket, when we are connecting to a local listening socket. To cover them, we need to hold locks on both PCBs when spawning a third one. This means holding them across sonewconn(). This creates a LOR between pcb locks and unp_list_lock. - To fix the new LOR, abandon the global unp_list_lock in favor of global unp_link_lock. Indeed, separating these two locks didn't provide us any extra parralelism in the UNIX sockets. - Now call into uipc_attach() may happen with unp_link_lock hold if, we are accepting, or without unp_link_lock in case if we are just creating a socket. - Another problem in UNIX sockets is that uipc_close() basicly did nothing for a listening socket. The vnode remained opened for connections. This is fixed by removing vnode in uipc_close(). Maybe the right way would be to do it for all sockets (not only listening), simply move the vnode teardown from uipc_detach() to uipc_close()? Sponsored by: Netflix Differential Revision: https://reviews.freebsd.org/D9770
2017-06-08 21:30:34 +00:00
UNP_LINK_RUNLOCK();
Re-wrap and other minor formatting and punctuation fixes for UNIX domain socket comments.
2006-07-22 17:24:55 +00:00
n = i; /* In case we lost some during malloc. */
Convert socket structures to be type-stable and add a version number. Define a parameter which indicates the maximum number of sockets in a system, and use this to size the zone allocators used for sockets and for certain PCBs. Convert PF_LOCAL PCB structures to be type-stable and add a version number. Define an external format for infomation about socket structures and use it in several places. Define a mechanism to get all PF_LOCAL and PF_INET PCB lists through sysctl(3) without blocking network interrupts for an unreasonable length of time. This probably still has some bugs and/or race conditions, but it seems to work well enough on my machines. It is now possible for `netstat' to get almost all of its information via the sysctl(3) interface rather than reading kmem (changes to follow).
1998-05-15 20:11:40 +00:00
error = 0;
Fix two issues which were missed in FreeBSD-SA-05:08.kmem. Reported by: Uwe Doering
2005-05-07 00:41:36 +00:00
xu = malloc(sizeof(*xu), M_TEMP, M_WAITOK | M_ZERO);
Convert socket structures to be type-stable and add a version number. Define a parameter which indicates the maximum number of sockets in a system, and use this to size the zone allocators used for sockets and for certain PCBs. Convert PF_LOCAL PCB structures to be type-stable and add a version number. Define an external format for infomation about socket structures and use it in several places. Define a mechanism to get all PF_LOCAL and PF_INET PCB lists through sysctl(3) without blocking network interrupts for an unreasonable length of time. This probably still has some bugs and/or race conditions, but it seems to work well enough on my machines. It is now possible for `netstat' to get almost all of its information via the sysctl(3) interface rather than reading kmem (changes to follow).
1998-05-15 20:11:40 +00:00
for (i = 0; i < n; i++) {
unp = unp_list[i];
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
UNP_PCB_LOCK(unp);
- Close a race between enumerating UNIX domain socket pcb structures via sysctl and socket teardown by adding a reference count to the UNIX domain pcb object and fixing the sysctl that enumerates unpcbs to grab a reference on each unpcb while it builds the list to copy out to userland. - Close a race between UNIX domain pcb garbage collection (unp_gc()) and file descriptor teardown (fdrop()) by adding a new garbage collection flag FWAIT. unp_gc() sets FWAIT while it walks the message buffers in a UNIX domain socket looking for nested file descriptor references and clears the flag when it is finished. fdrop() checks to see if the flag is set on a file descriptor whose refcount just dropped to 0 and waits for unp_gc() to clear the flag before completely destroying the file descriptor. MFC after: 1 week Reviewed by: rwatson Submitted by: ups Hopefully makes the panics go away: mx1
2007-01-05 19:59:46 +00:00
unp->unp_refcount--;
if (unp->unp_refcount != 0 && unp->unp_gencnt <= gencnt) {
Don't alocate a 400 byte buffer on the stack, Nor 800 bytes of structures.. MFC after: 2 weeks
2001-08-18 02:53:50 +00:00
xu->xu_len = sizeof *xu;
xu->xu_unpp = unp;
Convert socket structures to be type-stable and add a version number. Define a parameter which indicates the maximum number of sockets in a system, and use this to size the zone allocators used for sockets and for certain PCBs. Convert PF_LOCAL PCB structures to be type-stable and add a version number. Define an external format for infomation about socket structures and use it in several places. Define a mechanism to get all PF_LOCAL and PF_INET PCB lists through sysctl(3) without blocking network interrupts for an unreasonable length of time. This probably still has some bugs and/or race conditions, but it seems to work well enough on my machines. It is now possible for `netstat' to get almost all of its information via the sysctl(3) interface rather than reading kmem (changes to follow).
1998-05-15 20:11:40 +00:00
/*
* XXX - need more locking here to protect against
* connect/disconnect races for SMP.
*/
Prefer NULL to 0 when testing and assigning pointer values.
2004-03-30 02:16:25 +00:00
if (unp->unp_addr != NULL)
Mechanical whitespace cleanup; parenthesize return values; other minor style nits.
2004-01-11 19:48:19 +00:00
bcopy(unp->unp_addr, &xu->xu_addr,
Convert socket structures to be type-stable and add a version number. Define a parameter which indicates the maximum number of sockets in a system, and use this to size the zone allocators used for sockets and for certain PCBs. Convert PF_LOCAL PCB structures to be type-stable and add a version number. Define an external format for infomation about socket structures and use it in several places. Define a mechanism to get all PF_LOCAL and PF_INET PCB lists through sysctl(3) without blocking network interrupts for an unreasonable length of time. This probably still has some bugs and/or race conditions, but it seems to work well enough on my machines. It is now possible for `netstat' to get almost all of its information via the sysctl(3) interface rather than reading kmem (changes to follow).
1998-05-15 20:11:40 +00:00
unp->unp_addr->sun_len);
Prefer NULL to 0 when testing and assigning pointer values.
2004-03-30 02:16:25 +00:00
if (unp->unp_conn != NULL &&
unp->unp_conn->unp_addr != NULL)
Convert socket structures to be type-stable and add a version number. Define a parameter which indicates the maximum number of sockets in a system, and use this to size the zone allocators used for sockets and for certain PCBs. Convert PF_LOCAL PCB structures to be type-stable and add a version number. Define an external format for infomation about socket structures and use it in several places. Define a mechanism to get all PF_LOCAL and PF_INET PCB lists through sysctl(3) without blocking network interrupts for an unreasonable length of time. This probably still has some bugs and/or race conditions, but it seems to work well enough on my machines. It is now possible for `netstat' to get almost all of its information via the sysctl(3) interface rather than reading kmem (changes to follow).
1998-05-15 20:11:40 +00:00
bcopy(unp->unp_conn->unp_addr,
Don't alocate a 400 byte buffer on the stack, Nor 800 bytes of structures.. MFC after: 2 weeks
2001-08-18 02:53:50 +00:00
&xu->xu_caddr,
Convert socket structures to be type-stable and add a version number. Define a parameter which indicates the maximum number of sockets in a system, and use this to size the zone allocators used for sockets and for certain PCBs. Convert PF_LOCAL PCB structures to be type-stable and add a version number. Define an external format for infomation about socket structures and use it in several places. Define a mechanism to get all PF_LOCAL and PF_INET PCB lists through sysctl(3) without blocking network interrupts for an unreasonable length of time. This probably still has some bugs and/or race conditions, but it seems to work well enough on my machines. It is now possible for `netstat' to get almost all of its information via the sysctl(3) interface rather than reading kmem (changes to follow).
1998-05-15 20:11:40 +00:00
unp->unp_conn->unp_addr->sun_len);
Don't alocate a 400 byte buffer on the stack, Nor 800 bytes of structures.. MFC after: 2 weeks
2001-08-18 02:53:50 +00:00
bcopy(unp, &xu->xu_unp, sizeof *unp);
sotoxsocket(unp->unp_socket, &xu->xu_socket);
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
UNP_PCB_UNLOCK(unp);
Don't alocate a 400 byte buffer on the stack, Nor 800 bytes of structures.. MFC after: 2 weeks
2001-08-18 02:53:50 +00:00
error = SYSCTL_OUT(req, xu, sizeof *xu);
- Close a race between enumerating UNIX domain socket pcb structures via sysctl and socket teardown by adding a reference count to the UNIX domain pcb object and fixing the sysctl that enumerates unpcbs to grab a reference on each unpcb while it builds the list to copy out to userland. - Close a race between UNIX domain pcb garbage collection (unp_gc()) and file descriptor teardown (fdrop()) by adding a new garbage collection flag FWAIT. unp_gc() sets FWAIT while it walks the message buffers in a UNIX domain socket looking for nested file descriptor references and clears the flag when it is finished. fdrop() checks to see if the flag is set on a file descriptor whose refcount just dropped to 0 and waits for unp_gc() to clear the flag before completely destroying the file descriptor. MFC after: 1 week Reviewed by: rwatson Submitted by: ups Hopefully makes the panics go away: mx1
2007-01-05 19:59:46 +00:00
} else {
freeunp = (unp->unp_refcount == 0);
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
UNP_PCB_UNLOCK(unp);
if (freeunp) {
UNP_PCB_LOCK_DESTROY(unp);
- Close a race between enumerating UNIX domain socket pcb structures via sysctl and socket teardown by adding a reference count to the UNIX domain pcb object and fixing the sysctl that enumerates unpcbs to grab a reference on each unpcb while it builds the list to copy out to userland. - Close a race between UNIX domain pcb garbage collection (unp_gc()) and file descriptor teardown (fdrop()) by adding a new garbage collection flag FWAIT. unp_gc() sets FWAIT while it walks the message buffers in a UNIX domain socket looking for nested file descriptor references and clears the flag when it is finished. fdrop() checks to see if the flag is set on a file descriptor whose refcount just dropped to 0 and waits for unp_gc() to clear the flag before completely destroying the file descriptor. MFC after: 1 week Reviewed by: rwatson Submitted by: ups Hopefully makes the panics go away: mx1
2007-01-05 19:59:46 +00:00
uma_zfree(unp_zone, unp);
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
}
Convert socket structures to be type-stable and add a version number. Define a parameter which indicates the maximum number of sockets in a system, and use this to size the zone allocators used for sockets and for certain PCBs. Convert PF_LOCAL PCB structures to be type-stable and add a version number. Define an external format for infomation about socket structures and use it in several places. Define a mechanism to get all PF_LOCAL and PF_INET PCB lists through sysctl(3) without blocking network interrupts for an unreasonable length of time. This probably still has some bugs and/or race conditions, but it seems to work well enough on my machines. It is now possible for `netstat' to get almost all of its information via the sysctl(3) interface rather than reading kmem (changes to follow).
1998-05-15 20:11:40 +00:00
}
}
Don't alocate a 400 byte buffer on the stack, Nor 800 bytes of structures.. MFC after: 2 weeks
2001-08-18 02:53:50 +00:00
free(xu, M_TEMP);
Convert socket structures to be type-stable and add a version number. Define a parameter which indicates the maximum number of sockets in a system, and use this to size the zone allocators used for sockets and for certain PCBs. Convert PF_LOCAL PCB structures to be type-stable and add a version number. Define an external format for infomation about socket structures and use it in several places. Define a mechanism to get all PF_LOCAL and PF_INET PCB lists through sysctl(3) without blocking network interrupts for an unreasonable length of time. This probably still has some bugs and/or race conditions, but it seems to work well enough on my machines. It is now possible for `netstat' to get almost all of its information via the sysctl(3) interface rather than reading kmem (changes to follow).
1998-05-15 20:11:40 +00:00
if (!error) {
/*
Re-wrap and other minor formatting and punctuation fixes for UNIX domain socket comments.
2006-07-22 17:24:55 +00:00
* Give the user an updated idea of our state. If the
* generation differs from what we told her before, she knows
* that something happened while we were processing this
* request, and it might be necessary to retry.
Convert socket structures to be type-stable and add a version number. Define a parameter which indicates the maximum number of sockets in a system, and use this to size the zone allocators used for sockets and for certain PCBs. Convert PF_LOCAL PCB structures to be type-stable and add a version number. Define an external format for infomation about socket structures and use it in several places. Define a mechanism to get all PF_LOCAL and PF_INET PCB lists through sysctl(3) without blocking network interrupts for an unreasonable length of time. This probably still has some bugs and/or race conditions, but it seems to work well enough on my machines. It is now possible for `netstat' to get almost all of its information via the sysctl(3) interface rather than reading kmem (changes to follow).
1998-05-15 20:11:40 +00:00
*/
Don't alocate a 400 byte buffer on the stack, Nor 800 bytes of structures.. MFC after: 2 weeks
2001-08-18 02:53:50 +00:00
xug->xug_gen = unp_gencnt;
xug->xug_sogen = so_gencnt;
xug->xug_count = unp_count;
error = SYSCTL_OUT(req, xug, sizeof *xug);
Convert socket structures to be type-stable and add a version number. Define a parameter which indicates the maximum number of sockets in a system, and use this to size the zone allocators used for sockets and for certain PCBs. Convert PF_LOCAL PCB structures to be type-stable and add a version number. Define an external format for infomation about socket structures and use it in several places. Define a mechanism to get all PF_LOCAL and PF_INET PCB lists through sysctl(3) without blocking network interrupts for an unreasonable length of time. This probably still has some bugs and/or race conditions, but it seems to work well enough on my machines. It is now possible for `netstat' to get almost all of its information via the sysctl(3) interface rather than reading kmem (changes to follow).
1998-05-15 20:11:40 +00:00
}
free(unp_list, M_TEMP);
Don't alocate a 400 byte buffer on the stack, Nor 800 bytes of structures.. MFC after: 2 weeks
2001-08-18 02:53:50 +00:00
free(xug, M_TEMP);
Mechanical whitespace cleanup; parenthesize return values; other minor style nits.
2004-01-11 19:48:19 +00:00
return (error);
Convert socket structures to be type-stable and add a version number. Define a parameter which indicates the maximum number of sockets in a system, and use this to size the zone allocators used for sockets and for certain PCBs. Convert PF_LOCAL PCB structures to be type-stable and add a version number. Define an external format for infomation about socket structures and use it in several places. Define a mechanism to get all PF_LOCAL and PF_INET PCB lists through sysctl(3) without blocking network interrupts for an unreasonable length of time. This probably still has some bugs and/or race conditions, but it seems to work well enough on my machines. It is now possible for `netstat' to get almost all of its information via the sysctl(3) interface rather than reading kmem (changes to follow).
1998-05-15 20:11:40 +00:00
}
Specify a CTLTYPE_FOO so that a future sysctl(8) change does not need to rely on the format string.
2011-01-18 21:14:18 +00:00
SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist, CTLTYPE_OPAQUE | CTLFLAG_RD,
(void *)(intptr_t)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb",
"List of active local datagram sockets");
SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist, CTLTYPE_OPAQUE | CTLFLAG_RD,
(void *)(intptr_t)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb",
"List of active local stream sockets");
SYSCTL_PROC(_net_local_seqpacket, OID_AUTO, pcblist,
CTLTYPE_OPAQUE | CTLFLAG_RD,
(void *)(intptr_t)SOCK_SEQPACKET, 0, unp_pcblist, "S,xunpcb",
"List of active local seqpacket sockets");
Convert socket structures to be type-stable and add a version number. Define a parameter which indicates the maximum number of sockets in a system, and use this to size the zone allocators used for sockets and for certain PCBs. Convert PF_LOCAL PCB structures to be type-stable and add a version number. Define an external format for infomation about socket structures and use it in several places. Define a mechanism to get all PF_LOCAL and PF_INET PCB lists through sysctl(3) without blocking network interrupts for an unreasonable length of time. This probably still has some bugs and/or race conditions, but it seems to work well enough on my machines. It is now possible for `netstat' to get almost all of its information via the sysctl(3) interface rather than reading kmem (changes to follow).
1998-05-15 20:11:40 +00:00
Another mega commit to staticize things.
1995-12-14 09:55:16 +00:00
static void
style(9)-ize function headers, remove use of 'register'. MFC after: 3 days
2005-02-20 23:22:13 +00:00
unp_shutdown(struct unpcb *unp)
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
{
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
struct unpcb *unp2;
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
struct socket *so;
Decompose the global UNIX domain sockets rwlock into two different locks: a global list/counter/generation counter protected by a new mutex unp_list_lock, and a global linkage rwlock, unp_global_rwlock, which protects the connections between UNIX domain sockets. This eliminates conditional lock acquisition that was previously a property of the global lock being held over sonewconn() leading to a call to uipc_attach(), which also required the global lock, but couldn't rely on it as other paths existed to uipc_attach() that didn't hold it: now uipc_attach() uses only the list lock, which follows the linkage lock in the lock order. It may also reduce contention on the global lock for some workloads. Add global UNIX domain socket locks to hard-coded witness lock order. MFC after: 1 week Discussed with: kris
2009-03-08 21:48:29 +00:00
UNP_LINK_WLOCK_ASSERT();
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
UNP_PCB_LOCK_ASSERT(unp);
Introduce a subsystem lock around UNIX domain sockets in order to protect global and allocated variables. This strategy is derived from work originally developed by BSDi for BSD/OS, and applied to FreeBSD by Sam Leffler: - Add unp_mtx, a global mutex which will protect all UNIX domain socket related variables, structures, etc. - Add UNP_LOCK(), UNP_UNLOCK(), UNP_LOCK_ASSERT() macros. - Acquire unp_mtx on entering most UNIX domain socket code, drop/re-acquire around calls into VFS, and release it on return. - Avoid performing sodupsockaddr() while holding the mutex, so in general move to allocating storage before acquiring the mutex to copy the data. - Make a stack copy of the xucred rather than copying out while holding unp_mtx. Copy the peer credential out after releasing the mutex. - Add additional assertions of vnode locks following VOP_CREATE(). A few notes: - Use of an sx lock for the file list mutex may cause problems with regard to unp_mtx when garbage collection passed file descriptors. - The locking in unp_pcblist() for sysctl monitoring is correct subject to the unpcb zone not returning memory for reuse by other subsystems (consistent with similar existing concerns). - Sam's version of this change, as with the BSD/OS version, made use of both a global lock and per-unpcb locks. However, in practice, the global lock covered all accesses, so I have simplified out the unpcb locks in the interest of getting this merged faster (reducing the overhead but not sacrificing granularity in most cases). We will want to explore possibilities for improving lock granularity in this code in the future. Submitted by: sam Sponsored by: FreeBSD Foundatiuon Obtained from: BSD/OS 5 snapshot provided by BSDi
2004-06-10 21:34:38 +00:00
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
unp2 = unp->unp_conn;
First cut at implementing SOCK_SEQPACKET support for UNIX (local) domain sockets. This allows for reliable bi-directional datagram communication over UNIX domain sockets, in contrast to SOCK_DGRAM (M:N, unreliable) or SOCK_STERAM (bi-directional bytestream). Largely, this reuses existing UNIX domain socket code. This allows applications requiring record- oriented semantics to do so reliably via local IPC. Some implementation notes (also present in XXX comments): - Currently we lack an sbappend variant able to do datagrams and control data without doing addresses, so we mark SOCK_SEQPACKET as PR_ADDR. Adding a new variant will solve this problem. - UNIX domain sockets on FreeBSD provide back-pressure/flow control notification for stream sockets by manipulating the send socket buffer's size during pru_send and pru_rcvd. This trick works less well for SOCK_SEQPACKET as sosend_generic() uses sb_hiwat not just to manage blocking, but also to determine maximum datagram size. Fixing this requires rethinking how back-pressure is done for SOCK_SEQPACKET; in the mean time, it's possible to get EMSGSIZE when buffers fill, instead of blocking. Discussed with: benl Reviewed by: bz, rpaulo MFC after: 3 months Sponsored by: Google
2009-10-05 14:49:16 +00:00
if ((unp->unp_socket->so_type == SOCK_STREAM ||
(unp->unp_socket->so_type == SOCK_SEQPACKET)) && unp2 != NULL) {
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
so = unp2->unp_socket;
if (so != NULL)
socantrcvmore(so);
}
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
}
Another mega commit to staticize things.
1995-12-14 09:55:16 +00:00
static void
Remove the errno argument from unp_drop(). While there, add a comment to clarify that ECONNRESET should always be returned for POSIX conformance. Suggested by: Steven Hartland
2016-02-26 12:46:34 +00:00
unp_drop(struct unpcb *unp)
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
{
struct socket *so = unp->unp_socket;
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
struct unpcb *unp2;
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
Decompose the global UNIX domain sockets rwlock into two different locks: a global list/counter/generation counter protected by a new mutex unp_list_lock, and a global linkage rwlock, unp_global_rwlock, which protects the connections between UNIX domain sockets. This eliminates conditional lock acquisition that was previously a property of the global lock being held over sonewconn() leading to a call to uipc_attach(), which also required the global lock, but couldn't rely on it as other paths existed to uipc_attach() that didn't hold it: now uipc_attach() uses only the list lock, which follows the linkage lock in the lock order. It may also reduce contention on the global lock for some workloads. Add global UNIX domain socket locks to hard-coded witness lock order. MFC after: 1 week Discussed with: kris
2009-03-08 21:48:29 +00:00
UNP_LINK_WLOCK_ASSERT();
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
UNP_PCB_LOCK_ASSERT(unp);
Introduce a subsystem lock around UNIX domain sockets in order to protect global and allocated variables. This strategy is derived from work originally developed by BSDi for BSD/OS, and applied to FreeBSD by Sam Leffler: - Add unp_mtx, a global mutex which will protect all UNIX domain socket related variables, structures, etc. - Add UNP_LOCK(), UNP_UNLOCK(), UNP_LOCK_ASSERT() macros. - Acquire unp_mtx on entering most UNIX domain socket code, drop/re-acquire around calls into VFS, and release it on return. - Avoid performing sodupsockaddr() while holding the mutex, so in general move to allocating storage before acquiring the mutex to copy the data. - Make a stack copy of the xucred rather than copying out while holding unp_mtx. Copy the peer credential out after releasing the mutex. - Add additional assertions of vnode locks following VOP_CREATE(). A few notes: - Use of an sx lock for the file list mutex may cause problems with regard to unp_mtx when garbage collection passed file descriptors. - The locking in unp_pcblist() for sysctl monitoring is correct subject to the unpcb zone not returning memory for reuse by other subsystems (consistent with similar existing concerns). - Sam's version of this change, as with the BSD/OS version, made use of both a global lock and per-unpcb locks. However, in practice, the global lock covered all accesses, so I have simplified out the unpcb locks in the interest of getting this merged faster (reducing the overhead but not sacrificing granularity in most cases). We will want to explore possibilities for improving lock granularity in this code in the future. Submitted by: sam Sponsored by: FreeBSD Foundatiuon Obtained from: BSD/OS 5 snapshot provided by BSDi
2004-06-10 21:34:38 +00:00
Remove the errno argument from unp_drop(). While there, add a comment to clarify that ECONNRESET should always be returned for POSIX conformance. Suggested by: Steven Hartland
2016-02-26 12:46:34 +00:00
/*
* Regardless of whether the socket's peer dropped the connection
* with this socket by aborting or disconnecting, POSIX requires
* that ECONNRESET is returned.
*/
so->so_error = ECONNRESET;
Revise locking strategy used for UNIX domain sockets in order to improve concurrency: - Add per-unpcb mutexes protecting unpcb connection state, fields, etc. - Replace global UNP mutex with a global UNP rwlock, which will protect the UNIX domain socket connection topology, v_socket, and be acquired exclusively before acquiring more than per-unpcb at a time in order to avoid lock order issues. In performance measurements involving MySQL, this change has little or no overhead on UP (+/- 1%), but leads to a significant (5%-30%) improvement in multi-processor measurements using the sysbench and supersmack benchmarks. Much testing by: kris Approved by: re (kensmith)
2007-02-26 20:47:52 +00:00
unp2 = unp->unp_conn;
if (unp2 == NULL)
return;
UNP_PCB_LOCK(unp2);
unp_disconnect(unp, unp2);
UNP_PCB_UNLOCK(unp2);
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
}
Hopefully improve control message passing over Unix domain sockets. 1) Allow the sending of more than one control message at a time over a unix domain socket. This should cover the PR 29499. 2) This requires that unp_{ex,in}ternalize and unp_scan understand mbufs with more than one control message at a time. 3) Internalize and externalize used to work on the mbuf in-place. This made life quite complicated and the code for sizeof(int) < sizeof(file *) could end up doing the wrong thing. The patch always create a new mbuf/cluster now. This resulted in the change of the prototype for the domain externalise function. 4) You can now send SCM_TIMESTAMP messages. 5) Always use CMSG_DATA(cm) to determine the start where the data in unp_{ex,in}ternalize. It was using ((struct cmsghdr *)cm + 1) in some places, which gives the wrong alignment on the alpha. (NetBSD made this fix some time ago). This results in an ABI change for discriptor passing and creds passing on the alpha. (Probably on the IA64 and Spare ports too). 6) Fix userland programs to use CMSG_* macros too. 7) Be more careful about freeing mbufs containing (file *)s. This is made possible by the prototype change of externalise. PR: 29499 MFC after: 6 weeks
2001-10-04 13:11:48 +00:00
static void
For some reason when I started to pass filedescent structures instead of pointers to the file structure receiving descriptors stopped to work when also at least few kilobytes of data is being send. In the kernel the soreceive_generic() function doesn't see control mbuf as the first mbuf and unp_externalize() is never called, first 6(?) kilobytes of data is missing as well on receiving end. This breaks for example tmux. I don't know yet why going from 8 bytes to sizeof(struct filedescent) per descriptor (or even to 16 bytes per descriptor) breaks things, but to work-around it for now use 8 bytes per file descriptor at the cost of memory allocation. Reported by: flo, Diane Bruce, Jan Beich <jbeich@tormail.org> Simple testcase provided by: mjg
2013-03-03 23:39:30 +00:00
unp_freerights(struct filedescent **fdep, int fdcount)
Hopefully improve control message passing over Unix domain sockets. 1) Allow the sending of more than one control message at a time over a unix domain socket. This should cover the PR 29499. 2) This requires that unp_{ex,in}ternalize and unp_scan understand mbufs with more than one control message at a time. 3) Internalize and externalize used to work on the mbuf in-place. This made life quite complicated and the code for sizeof(int) < sizeof(file *) could end up doing the wrong thing. The patch always create a new mbuf/cluster now. This resulted in the change of the prototype for the domain externalise function. 4) You can now send SCM_TIMESTAMP messages. 5) Always use CMSG_DATA(cm) to determine the start where the data in unp_{ex,in}ternalize. It was using ((struct cmsghdr *)cm + 1) in some places, which gives the wrong alignment on the alpha. (NetBSD made this fix some time ago). This results in an ABI change for discriptor passing and creds passing on the alpha. (Probably on the IA64 and Spare ports too). 6) Fix userland programs to use CMSG_* macros too. 7) Be more careful about freeing mbufs containing (file *)s. This is made possible by the prototype change of externalise. PR: 29499 MFC after: 6 weeks
2001-10-04 13:11:48 +00:00
{
struct file *fp;
Merge Capsicum overhaul: - Capability is no longer separate descriptor type. Now every descriptor has set of its own capability rights. - The cap_new(2) system call is left, but it is no longer documented and should not be used in new code. - The new syscall cap_rights_limit(2) should be used instead of cap_new(2), which limits capability rights of the given descriptor without creating a new one. - The cap_getrights(2) syscall is renamed to cap_rights_get(2). - If CAP_IOCTL capability right is present we can further reduce allowed ioctls list with the new cap_ioctls_limit(2) syscall. List of allowed ioctls can be retrived with cap_ioctls_get(2) syscall. - If CAP_FCNTL capability right is present we can further reduce fcntls that can be used with the new cap_fcntls_limit(2) syscall and retrive them with cap_fcntls_get(2). - To support ioctl and fcntl white-listing the filedesc structure was heavly modified. - The audit subsystem, kdump and procstat tools were updated to recognize new syscalls. - Capability rights were revised and eventhough I tried hard to provide backward API and ABI compatibility there are some incompatible changes that are described in detail below: CAP_CREATE old behaviour: - Allow for openat(2)+O_CREAT. - Allow for linkat(2). - Allow for symlinkat(2). CAP_CREATE new behaviour: - Allow for openat(2)+O_CREAT. Added CAP_LINKAT: - Allow for linkat(2). ABI: Reuses CAP_RMDIR bit. - Allow to be target for renameat(2). Added CAP_SYMLINKAT: - Allow for symlinkat(2). Removed CAP_DELETE. Old behaviour: - Allow for unlinkat(2) when removing non-directory object. - Allow to be source for renameat(2). Removed CAP_RMDIR. Old behaviour: - Allow for unlinkat(2) when removing directory. Added CAP_RENAMEAT: - Required for source directory for the renameat(2) syscall. Added CAP_UNLINKAT (effectively it replaces CAP_DELETE and CAP_RMDIR): - Allow for unlinkat(2) on any object. - Required if target of renameat(2) exists and will be removed by this call. Removed CAP_MAPEXEC. CAP_MMAP old behaviour: - Allow for mmap(2) with any combination of PROT_NONE, PROT_READ and PROT_WRITE. CAP_MMAP new behaviour: - Allow for mmap(2)+PROT_NONE. Added CAP_MMAP_R: - Allow for mmap(PROT_READ). Added CAP_MMAP_W: - Allow for mmap(PROT_WRITE). Added CAP_MMAP_X: - Allow for mmap(PROT_EXEC). Added CAP_MMAP_RW: - Allow for mmap(PROT_READ | PROT_WRITE). Added CAP_MMAP_RX: - Allow for mmap(PROT_READ | PROT_EXEC). Added CAP_MMAP_WX: - Allow for mmap(PROT_WRITE | PROT_EXEC). Added CAP_MMAP_RWX: - Allow for mmap(PROT_READ | PROT_WRITE | PROT_EXEC). Renamed CAP_MKDIR to CAP_MKDIRAT. Renamed CAP_MKFIFO to CAP_MKFIFOAT. Renamed CAP_MKNODE to CAP_MKNODEAT. CAP_READ old behaviour: - Allow pread(2). - Disallow read(2), readv(2) (if there is no CAP_SEEK). CAP_READ new behaviour: - Allow read(2), readv(2). - Disallow pread(2) (CAP_SEEK was also required). CAP_WRITE old behaviour: - Allow pwrite(2). - Disallow write(2), writev(2) (if there is no CAP_SEEK). CAP_WRITE new behaviour: - Allow write(2), writev(2). - Disallow pwrite(2) (CAP_SEEK was also required). Added convinient defines: #define CAP_PREAD (CAP_SEEK | CAP_READ) #define CAP_PWRITE (CAP_SEEK | CAP_WRITE) #define CAP_MMAP_R (CAP_MMAP | CAP_SEEK | CAP_READ) #define CAP_MMAP_W (CAP_MMAP | CAP_SEEK | CAP_WRITE) #define CAP_MMAP_X (CAP_MMAP | CAP_SEEK | 0x0000000000000008ULL) #define CAP_MMAP_RW (CAP_MMAP_R | CAP_MMAP_W) #define CAP_MMAP_RX (CAP_MMAP_R | CAP_MMAP_X) #define CAP_MMAP_WX (CAP_MMAP_W | CAP_MMAP_X) #define CAP_MMAP_RWX (CAP_MMAP_R | CAP_MMAP_W | CAP_MMAP_X) #define CAP_RECV CAP_READ #define CAP_SEND CAP_WRITE #define CAP_SOCK_CLIENT \ (CAP_CONNECT | CAP_GETPEERNAME | CAP_GETSOCKNAME | CAP_GETSOCKOPT | \ CAP_PEELOFF | CAP_RECV | CAP_SEND | CAP_SETSOCKOPT | CAP_SHUTDOWN) #define CAP_SOCK_SERVER \ (CAP_ACCEPT | CAP_BIND | CAP_GETPEERNAME | CAP_GETSOCKNAME | \ CAP_GETSOCKOPT | CAP_LISTEN | CAP_PEELOFF | CAP_RECV | CAP_SEND | \ CAP_SETSOCKOPT | CAP_SHUTDOWN) Added defines for backward API compatibility: #define CAP_MAPEXEC CAP_MMAP_X #define CAP_DELETE CAP_UNLINKAT #define CAP_MKDIR CAP_MKDIRAT #define CAP_RMDIR CAP_UNLINKAT #define CAP_MKFIFO CAP_MKFIFOAT #define CAP_MKNOD CAP_MKNODAT #define CAP_SOCK_ALL (CAP_SOCK_CLIENT | CAP_SOCK_SERVER) Sponsored by: The FreeBSD Foundation Reviewed by: Christoph Mallon <christoph.mallon@gmx.de> Many aspects discussed with: rwatson, benl, jonathan ABI compatibility discussed with: kib
2013-03-02 00:53:12 +00:00
int i;
Hopefully improve control message passing over Unix domain sockets. 1) Allow the sending of more than one control message at a time over a unix domain socket. This should cover the PR 29499. 2) This requires that unp_{ex,in}ternalize and unp_scan understand mbufs with more than one control message at a time. 3) Internalize and externalize used to work on the mbuf in-place. This made life quite complicated and the code for sizeof(int) < sizeof(file *) could end up doing the wrong thing. The patch always create a new mbuf/cluster now. This resulted in the change of the prototype for the domain externalise function. 4) You can now send SCM_TIMESTAMP messages. 5) Always use CMSG_DATA(cm) to determine the start where the data in unp_{ex,in}ternalize. It was using ((struct cmsghdr *)cm + 1) in some places, which gives the wrong alignment on the alpha. (NetBSD made this fix some time ago). This results in an ABI change for discriptor passing and creds passing on the alpha. (Probably on the IA64 and Spare ports too). 6) Fix userland programs to use CMSG_* macros too. 7) Be more careful about freeing mbufs containing (file *)s. This is made possible by the prototype change of externalise. PR: 29499 MFC after: 6 weeks
2001-10-04 13:11:48 +00:00
Improve r250890, so that we stop processing of a message with zero descriptors as early as possible, and assert that number of descriptors is positive in unp_freerights(). Reviewed by: mjg, pjd, jilles
2013-06-04 11:19:08 +00:00
KASSERT(fdcount > 0, ("%s: fdcount %d", __func__, fdcount));
For some reason when I started to pass filedescent structures instead of pointers to the file structure receiving descriptors stopped to work when also at least few kilobytes of data is being send. In the kernel the soreceive_generic() function doesn't see control mbuf as the first mbuf and unp_externalize() is never called, first 6(?) kilobytes of data is missing as well on receiving end. This breaks for example tmux. I don't know yet why going from 8 bytes to sizeof(struct filedescent) per descriptor (or even to 16 bytes per descriptor) breaks things, but to work-around it for now use 8 bytes per file descriptor at the cost of memory allocation. Reported by: flo, Diane Bruce, Jan Beich <jbeich@tormail.org> Simple testcase provided by: mjg
2013-03-03 23:39:30 +00:00
for (i = 0; i < fdcount; i++) {
fp = fdep[i]->fde_file;
filecaps_free(&fdep[i]->fde_caps);
Hopefully improve control message passing over Unix domain sockets. 1) Allow the sending of more than one control message at a time over a unix domain socket. This should cover the PR 29499. 2) This requires that unp_{ex,in}ternalize and unp_scan understand mbufs with more than one control message at a time. 3) Internalize and externalize used to work on the mbuf in-place. This made life quite complicated and the code for sizeof(int) < sizeof(file *) could end up doing the wrong thing. The patch always create a new mbuf/cluster now. This resulted in the change of the prototype for the domain externalise function. 4) You can now send SCM_TIMESTAMP messages. 5) Always use CMSG_DATA(cm) to determine the start where the data in unp_{ex,in}ternalize. It was using ((struct cmsghdr *)cm + 1) in some places, which gives the wrong alignment on the alpha. (NetBSD made this fix some time ago). This results in an ABI change for discriptor passing and creds passing on the alpha. (Probably on the IA64 and Spare ports too). 6) Fix userland programs to use CMSG_* macros too. 7) Be more careful about freeing mbufs containing (file *)s. This is made possible by the prototype change of externalise. PR: 29499 MFC after: 6 weeks
2001-10-04 13:11:48 +00:00
unp_discard(fp);
}
For some reason when I started to pass filedescent structures instead of pointers to the file structure receiving descriptors stopped to work when also at least few kilobytes of data is being send. In the kernel the soreceive_generic() function doesn't see control mbuf as the first mbuf and unp_externalize() is never called, first 6(?) kilobytes of data is missing as well on receiving end. This breaks for example tmux. I don't know yet why going from 8 bytes to sizeof(struct filedescent) per descriptor (or even to 16 bytes per descriptor) breaks things, but to work-around it for now use 8 bytes per file descriptor at the cost of memory allocation. Reported by: flo, Diane Bruce, Jan Beich <jbeich@tormail.org> Simple testcase provided by: mjg
2013-03-03 23:39:30 +00:00
free(fdep[0], M_FILECAPS);
Hopefully improve control message passing over Unix domain sockets. 1) Allow the sending of more than one control message at a time over a unix domain socket. This should cover the PR 29499. 2) This requires that unp_{ex,in}ternalize and unp_scan understand mbufs with more than one control message at a time. 3) Internalize and externalize used to work on the mbuf in-place. This made life quite complicated and the code for sizeof(int) < sizeof(file *) could end up doing the wrong thing. The patch always create a new mbuf/cluster now. This resulted in the change of the prototype for the domain externalise function. 4) You can now send SCM_TIMESTAMP messages. 5) Always use CMSG_DATA(cm) to determine the start where the data in unp_{ex,in}ternalize. It was using ((struct cmsghdr *)cm + 1) in some places, which gives the wrong alignment on the alpha. (NetBSD made this fix some time ago). This results in an ABI change for discriptor passing and creds passing on the alpha. (Probably on the IA64 and Spare ports too). 6) Fix userland programs to use CMSG_* macros too. 7) Be more careful about freeing mbufs containing (file *)s. This is made possible by the prototype change of externalise. PR: 29499 MFC after: 6 weeks
2001-10-04 13:11:48 +00:00
}
Further minor cleanups to UNIX domain sockets: - Staticize and locally prototype functions uipc_ctloutput(), unp_dispose(), unp_init(), and unp_externalize(), none of which have been required outside of uipc_usrreq.c since uipc_proto.c was removed. - Remove stale prototype for uipc_usrreq(), which has not existed in the code since 1997 - Forward declare and staticize uipc_usrreqs structure in uipc_usrreq.c and not un.h. - Comment on why uipc_connect2() is still non-static -- it is used directly by fifofs. - Remove stale comments, tidy up whitespace. MFC after: 3 days (where applicable)
2008-10-03 13:01:56 +00:00
static int
Implement SOCK_CLOEXEC, SOCK_NONBLOCK and MSG_CMSG_CLOEXEC. This change allows creating file descriptors with close-on-exec set in some situations. SOCK_CLOEXEC and SOCK_NONBLOCK can be OR'ed in socket() and socketpair()'s type parameter, and MSG_CMSG_CLOEXEC to recvmsg() makes file descriptors (SCM_RIGHTS) atomically close-on-exec. The numerical values for SOCK_CLOEXEC and SOCK_NONBLOCK are as in NetBSD. MSG_CMSG_CLOEXEC is the first free bit for MSG_*. The SOCK_* flags are not passed to MAC because this may cause incorrect failures and can be done later via fcntl() anyway. On the other hand, audit is expected to cope with the new flags. For MSG_CMSG_CLOEXEC, unp_externalize() is extended to take a flags argument. Reviewed by: kib
2013-03-19 20:58:17 +00:00
unp_externalize(struct mbuf *control, struct mbuf **controlp, int flags)
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
{
KSE Milestone 2 Note ALL MODULES MUST BE RECOMPILED make the kernel aware that there are smaller units of scheduling than the process. (but only allow one thread per process at this time). This is functionally equivalent to teh previousl -current except that there is a thread associated with each process. Sorry john! (your next MFC will be a doosie!) Reviewed by: peter@freebsd.org, dillon@freebsd.org X-MFC after: ha ha ha ha
2001-09-12 08:38:13 +00:00
struct thread *td = curthread; /* XXX */
Hopefully improve control message passing over Unix domain sockets. 1) Allow the sending of more than one control message at a time over a unix domain socket. This should cover the PR 29499. 2) This requires that unp_{ex,in}ternalize and unp_scan understand mbufs with more than one control message at a time. 3) Internalize and externalize used to work on the mbuf in-place. This made life quite complicated and the code for sizeof(int) < sizeof(file *) could end up doing the wrong thing. The patch always create a new mbuf/cluster now. This resulted in the change of the prototype for the domain externalise function. 4) You can now send SCM_TIMESTAMP messages. 5) Always use CMSG_DATA(cm) to determine the start where the data in unp_{ex,in}ternalize. It was using ((struct cmsghdr *)cm + 1) in some places, which gives the wrong alignment on the alpha. (NetBSD made this fix some time ago). This results in an ABI change for discriptor passing and creds passing on the alpha. (Probably on the IA64 and Spare ports too). 6) Fix userland programs to use CMSG_* macros too. 7) Be more careful about freeing mbufs containing (file *)s. This is made possible by the prototype change of externalise. PR: 29499 MFC after: 6 weeks
2001-10-04 13:11:48 +00:00
struct cmsghdr *cm = mtod(control, struct cmsghdr *);
int i;
int *fdp;
Merge Capsicum overhaul: - Capability is no longer separate descriptor type. Now every descriptor has set of its own capability rights. - The cap_new(2) system call is left, but it is no longer documented and should not be used in new code. - The new syscall cap_rights_limit(2) should be used instead of cap_new(2), which limits capability rights of the given descriptor without creating a new one. - The cap_getrights(2) syscall is renamed to cap_rights_get(2). - If CAP_IOCTL capability right is present we can further reduce allowed ioctls list with the new cap_ioctls_limit(2) syscall. List of allowed ioctls can be retrived with cap_ioctls_get(2) syscall. - If CAP_FCNTL capability right is present we can further reduce fcntls that can be used with the new cap_fcntls_limit(2) syscall and retrive them with cap_fcntls_get(2). - To support ioctl and fcntl white-listing the filedesc structure was heavly modified. - The audit subsystem, kdump and procstat tools were updated to recognize new syscalls. - Capability rights were revised and eventhough I tried hard to provide backward API and ABI compatibility there are some incompatible changes that are described in detail below: CAP_CREATE old behaviour: - Allow for openat(2)+O_CREAT. - Allow for linkat(2). - Allow for symlinkat(2). CAP_CREATE new behaviour: - Allow for openat(2)+O_CREAT. Added CAP_LINKAT: - Allow for linkat(2). ABI: Reuses CAP_RMDIR bit. - Allow to be target for renameat(2). Added CAP_SYMLINKAT: - Allow for symlinkat(2). Removed CAP_DELETE. Old behaviour: - Allow for unlinkat(2) when removing non-directory object. - Allow to be source for renameat(2). Removed CAP_RMDIR. Old behaviour: - Allow for unlinkat(2) when removing directory. Added CAP_RENAMEAT: - Required for source directory for the renameat(2) syscall. Added CAP_UNLINKAT (effectively it replaces CAP_DELETE and CAP_RMDIR): - Allow for unlinkat(2) on any object. - Required if target of renameat(2) exists and will be removed by this call. Removed CAP_MAPEXEC. CAP_MMAP old behaviour: - Allow for mmap(2) with any combination of PROT_NONE, PROT_READ and PROT_WRITE. CAP_MMAP new behaviour: - Allow for mmap(2)+PROT_NONE. Added CAP_MMAP_R: - Allow for mmap(PROT_READ). Added CAP_MMAP_W: - Allow for mmap(PROT_WRITE). Added CAP_MMAP_X: - Allow for mmap(PROT_EXEC). Added CAP_MMAP_RW: - Allow for mmap(PROT_READ | PROT_WRITE). Added CAP_MMAP_RX: - Allow for mmap(PROT_READ | PROT_EXEC). Added CAP_MMAP_WX: - Allow for mmap(PROT_WRITE | PROT_EXEC). Added CAP_MMAP_RWX: - Allow for mmap(PROT_READ | PROT_WRITE | PROT_EXEC). Renamed CAP_MKDIR to CAP_MKDIRAT. Renamed CAP_MKFIFO to CAP_MKFIFOAT. Renamed CAP_MKNODE to CAP_MKNODEAT. CAP_READ old behaviour: - Allow pread(2). - Disallow read(2), readv(2) (if there is no CAP_SEEK). CAP_READ new behaviour: - Allow read(2), readv(2). - Disallow pread(2) (CAP_SEEK was also required). CAP_WRITE old behaviour: - Allow pwrite(2). - Disallow write(2), writev(2) (if there is no CAP_SEEK). CAP_WRITE new behaviour: - Allow write(2), writev(2). - Disallow pwrite(2) (CAP_SEEK was also required). Added convinient defines: #define CAP_PREAD (CAP_SEEK | CAP_READ) #define CAP_PWRITE (CAP_SEEK | CAP_WRITE) #define CAP_MMAP_R (CAP_MMAP | CAP_SEEK | CAP_READ) #define CAP_MMAP_W (CAP_MMAP | CAP_SEEK | CAP_WRITE) #define CAP_MMAP_X (CAP_MMAP | CAP_SEEK | 0x0000000000000008ULL) #define CAP_MMAP_RW (CAP_MMAP_R | CAP_MMAP_W) #define CAP_MMAP_RX (CAP_MMAP_R | CAP_MMAP_X) #define CAP_MMAP_WX (CAP_MMAP_W | CAP_MMAP_X) #define CAP_MMAP_RWX (CAP_MMAP_R | CAP_MMAP_W | CAP_MMAP_X) #define CAP_RECV CAP_READ #define CAP_SEND CAP_WRITE #define CAP_SOCK_CLIENT \ (CAP_CONNECT | CAP_GETPEERNAME | CAP_GETSOCKNAME | CAP_GETSOCKOPT | \ CAP_PEELOFF | CAP_RECV | CAP_SEND | CAP_SETSOCKOPT | CAP_SHUTDOWN) #define CAP_SOCK_SERVER \ (CAP_ACCEPT | CAP_BIND | CAP_GETPEERNAME | CAP_GETSOCKNAME | \ CAP_GETSOCKOPT | CAP_LISTEN | CAP_PEELOFF | CAP_RECV | CAP_SEND | \ CAP_SETSOCKOPT | CAP_SHUTDOWN) Added defines for backward API compatibility: #define CAP_MAPEXEC CAP_MMAP_X #define CAP_DELETE CAP_UNLINKAT #define CAP_MKDIR CAP_MKDIRAT #define CAP_RMDIR CAP_UNLINKAT #define CAP_MKFIFO CAP_MKFIFOAT #define CAP_MKNOD CAP_MKNODAT #define CAP_SOCK_ALL (CAP_SOCK_CLIENT | CAP_SOCK_SERVER) Sponsored by: The FreeBSD Foundation Reviewed by: Christoph Mallon <christoph.mallon@gmx.de> Many aspects discussed with: rwatson, benl, jonathan ABI compatibility discussed with: kib
2013-03-02 00:53:12 +00:00
struct filedesc *fdesc = td->td_proc->p_fd;
fd: move out actual fp installation to _finstall Use it in fd passing functions as the first step towards fd code cleanup.
2015-06-14 14:08:52 +00:00
struct filedescent **fdep;
Hopefully improve control message passing over Unix domain sockets. 1) Allow the sending of more than one control message at a time over a unix domain socket. This should cover the PR 29499. 2) This requires that unp_{ex,in}ternalize and unp_scan understand mbufs with more than one control message at a time. 3) Internalize and externalize used to work on the mbuf in-place. This made life quite complicated and the code for sizeof(int) < sizeof(file *) could end up doing the wrong thing. The patch always create a new mbuf/cluster now. This resulted in the change of the prototype for the domain externalise function. 4) You can now send SCM_TIMESTAMP messages. 5) Always use CMSG_DATA(cm) to determine the start where the data in unp_{ex,in}ternalize. It was using ((struct cmsghdr *)cm + 1) in some places, which gives the wrong alignment on the alpha. (NetBSD made this fix some time ago). This results in an ABI change for discriptor passing and creds passing on the alpha. (Probably on the IA64 and Spare ports too). 6) Fix userland programs to use CMSG_* macros too. 7) Be more careful about freeing mbufs containing (file *)s. This is made possible by the prototype change of externalise. PR: 29499 MFC after: 6 weeks
2001-10-04 13:11:48 +00:00
void *data;
socklen_t clen = control->m_len, datalen;
int error, newfds;
u_int newlen;
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
Decompose the global UNIX domain sockets rwlock into two different locks: a global list/counter/generation counter protected by a new mutex unp_list_lock, and a global linkage rwlock, unp_global_rwlock, which protects the connections between UNIX domain sockets. This eliminates conditional lock acquisition that was previously a property of the global lock being held over sonewconn() leading to a call to uipc_attach(), which also required the global lock, but couldn't rely on it as other paths existed to uipc_attach() that didn't hold it: now uipc_attach() uses only the list lock, which follows the linkage lock in the lock order. It may also reduce contention on the global lock for some workloads. Add global UNIX domain socket locks to hard-coded witness lock order. MFC after: 1 week Discussed with: kris
2009-03-08 21:48:29 +00:00
UNP_LINK_UNLOCK_ASSERT();
Add UNP_UNLOCK_ASSERT() to asser that the UNIX domain socket subsystem lock is not held. Rather than annotating that the lock is released after calls to unp_detach() with a comment, annotate with an assertion. Assert that the UNIX domain socket subsystem lock is not held when unp_externalize() and unp_internalize() are called.
2004-08-19 01:45:16 +00:00
Hopefully improve control message passing over Unix domain sockets. 1) Allow the sending of more than one control message at a time over a unix domain socket. This should cover the PR 29499. 2) This requires that unp_{ex,in}ternalize and unp_scan understand mbufs with more than one control message at a time. 3) Internalize and externalize used to work on the mbuf in-place. This made life quite complicated and the code for sizeof(int) < sizeof(file *) could end up doing the wrong thing. The patch always create a new mbuf/cluster now. This resulted in the change of the prototype for the domain externalise function. 4) You can now send SCM_TIMESTAMP messages. 5) Always use CMSG_DATA(cm) to determine the start where the data in unp_{ex,in}ternalize. It was using ((struct cmsghdr *)cm + 1) in some places, which gives the wrong alignment on the alpha. (NetBSD made this fix some time ago). This results in an ABI change for discriptor passing and creds passing on the alpha. (Probably on the IA64 and Spare ports too). 6) Fix userland programs to use CMSG_* macros too. 7) Be more careful about freeing mbufs containing (file *)s. This is made possible by the prototype change of externalise. PR: 29499 MFC after: 6 weeks
2001-10-04 13:11:48 +00:00
error = 0;
if (controlp != NULL) /* controlp == NULL => free control messages */
*controlp = NULL;
while (cm != NULL) {
if (sizeof(*cm) > clen || cm->cmsg_len > clen) {
error = EINVAL;
break;
}
data = CMSG_DATA(cm);
datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data;
if (cm->cmsg_level == SOL_SOCKET
&& cm->cmsg_type == SCM_RIGHTS) {
Merge Capsicum overhaul: - Capability is no longer separate descriptor type. Now every descriptor has set of its own capability rights. - The cap_new(2) system call is left, but it is no longer documented and should not be used in new code. - The new syscall cap_rights_limit(2) should be used instead of cap_new(2), which limits capability rights of the given descriptor without creating a new one. - The cap_getrights(2) syscall is renamed to cap_rights_get(2). - If CAP_IOCTL capability right is present we can further reduce allowed ioctls list with the new cap_ioctls_limit(2) syscall. List of allowed ioctls can be retrived with cap_ioctls_get(2) syscall. - If CAP_FCNTL capability right is present we can further reduce fcntls that can be used with the new cap_fcntls_limit(2) syscall and retrive them with cap_fcntls_get(2). - To support ioctl and fcntl white-listing the filedesc structure was heavly modified. - The audit subsystem, kdump and procstat tools were updated to recognize new syscalls. - Capability rights were revised and eventhough I tried hard to provide backward API and ABI compatibility there are some incompatible changes that are described in detail below: CAP_CREATE old behaviour: - Allow for openat(2)+O_CREAT. - Allow for linkat(2). - Allow for symlinkat(2). CAP_CREATE new behaviour: - Allow for openat(2)+O_CREAT. Added CAP_LINKAT: - Allow for linkat(2). ABI: Reuses CAP_RMDIR bit. - Allow to be target for renameat(2). Added CAP_SYMLINKAT: - Allow for symlinkat(2). Removed CAP_DELETE. Old behaviour: - Allow for unlinkat(2) when removing non-directory object. - Allow to be source for renameat(2). Removed CAP_RMDIR. Old behaviour: - Allow for unlinkat(2) when removing directory. Added CAP_RENAMEAT: - Required for source directory for the renameat(2) syscall. Added CAP_UNLINKAT (effectively it replaces CAP_DELETE and CAP_RMDIR): - Allow for unlinkat(2) on any object. - Required if target of renameat(2) exists and will be removed by this call. Removed CAP_MAPEXEC. CAP_MMAP old behaviour: - Allow for mmap(2) with any combination of PROT_NONE, PROT_READ and PROT_WRITE. CAP_MMAP new behaviour: - Allow for mmap(2)+PROT_NONE. Added CAP_MMAP_R: - Allow for mmap(PROT_READ). Added CAP_MMAP_W: - Allow for mmap(PROT_WRITE). Added CAP_MMAP_X: - Allow for mmap(PROT_EXEC). Added CAP_MMAP_RW: - Allow for mmap(PROT_READ | PROT_WRITE). Added CAP_MMAP_RX: - Allow for mmap(PROT_READ | PROT_EXEC). Added CAP_MMAP_WX: - Allow for mmap(PROT_WRITE | PROT_EXEC). Added CAP_MMAP_RWX: - Allow for mmap(PROT_READ | PROT_WRITE | PROT_EXEC). Renamed CAP_MKDIR to CAP_MKDIRAT. Renamed CAP_MKFIFO to CAP_MKFIFOAT. Renamed CAP_MKNODE to CAP_MKNODEAT. CAP_READ old behaviour: - Allow pread(2). - Disallow read(2), readv(2) (if there is no CAP_SEEK). CAP_READ new behaviour: - Allow read(2), readv(2). - Disallow pread(2) (CAP_SEEK was also required). CAP_WRITE old behaviour: - Allow pwrite(2). - Disallow write(2), writev(2) (if there is no CAP_SEEK). CAP_WRITE new behaviour: - Allow write(2), writev(2). - Disallow pwrite(2) (CAP_SEEK was also required). Added convinient defines: #define CAP_PREAD (CAP_SEEK | CAP_READ) #define CAP_PWRITE (CAP_SEEK | CAP_WRITE) #define CAP_MMAP_R (CAP_MMAP | CAP_SEEK | CAP_READ) #define CAP_MMAP_W (CAP_MMAP | CAP_SEEK | CAP_WRITE) #define CAP_MMAP_X (CAP_MMAP | CAP_SEEK | 0x0000000000000008ULL) #define CAP_MMAP_RW (CAP_MMAP_R | CAP_MMAP_W) #define CAP_MMAP_RX (CAP_MMAP_R | CAP_MMAP_X) #define CAP_MMAP_WX (CAP_MMAP_W | CAP_MMAP_X) #define CAP_MMAP_RWX (CAP_MMAP_R | CAP_MMAP_W | CAP_MMAP_X) #define CAP_RECV CAP_READ #define CAP_SEND CAP_WRITE #define CAP_SOCK_CLIENT \ (CAP_CONNECT | CAP_GETPEERNAME | CAP_GETSOCKNAME | CAP_GETSOCKOPT | \ CAP_PEELOFF | CAP_RECV | CAP_SEND | CAP_SETSOCKOPT | CAP_SHUTDOWN) #define CAP_SOCK_SERVER \ (CAP_ACCEPT | CAP_BIND | CAP_GETPEERNAME | CAP_GETSOCKNAME | \ CAP_GETSOCKOPT | CAP_LISTEN | CAP_PEELOFF | CAP_RECV | CAP_SEND | \ CAP_SETSOCKOPT | CAP_SHUTDOWN) Added defines for backward API compatibility: #define CAP_MAPEXEC CAP_MMAP_X #define CAP_DELETE CAP_UNLINKAT #define CAP_MKDIR CAP_MKDIRAT #define CAP_RMDIR CAP_UNLINKAT #define CAP_MKFIFO CAP_MKFIFOAT #define CAP_MKNOD CAP_MKNODAT #define CAP_SOCK_ALL (CAP_SOCK_CLIENT | CAP_SOCK_SERVER) Sponsored by: The FreeBSD Foundation Reviewed by: Christoph Mallon <christoph.mallon@gmx.de> Many aspects discussed with: rwatson, benl, jonathan ABI compatibility discussed with: kib
2013-03-02 00:53:12 +00:00
newfds = datalen / sizeof(*fdep);
Improve r250890, so that we stop processing of a message with zero descriptors as early as possible, and assert that number of descriptors is positive in unp_freerights(). Reviewed by: mjg, pjd, jilles
2013-06-04 11:19:08 +00:00
if (newfds == 0)
goto next;
Merge Capsicum overhaul: - Capability is no longer separate descriptor type. Now every descriptor has set of its own capability rights. - The cap_new(2) system call is left, but it is no longer documented and should not be used in new code. - The new syscall cap_rights_limit(2) should be used instead of cap_new(2), which limits capability rights of the given descriptor without creating a new one. - The cap_getrights(2) syscall is renamed to cap_rights_get(2). - If CAP_IOCTL capability right is present we can further reduce allowed ioctls list with the new cap_ioctls_limit(2) syscall. List of allowed ioctls can be retrived with cap_ioctls_get(2) syscall. - If CAP_FCNTL capability right is present we can further reduce fcntls that can be used with the new cap_fcntls_limit(2) syscall and retrive them with cap_fcntls_get(2). - To support ioctl and fcntl white-listing the filedesc structure was heavly modified. - The audit subsystem, kdump and procstat tools were updated to recognize new syscalls. - Capability rights were revised and eventhough I tried hard to provide backward API and ABI compatibility there are some incompatible changes that are described in detail below: CAP_CREATE old behaviour: - Allow for openat(2)+O_CREAT. - Allow for linkat(2). - Allow for symlinkat(2). CAP_CREATE new behaviour: - Allow for openat(2)+O_CREAT. Added CAP_LINKAT: - Allow for linkat(2). ABI: Reuses CAP_RMDIR bit. - Allow to be target for renameat(2). Added CAP_SYMLINKAT: - Allow for symlinkat(2). Removed CAP_DELETE. Old behaviour: - Allow for unlinkat(2) when removing non-directory object. - Allow to be source for renameat(2). Removed CAP_RMDIR. Old behaviour: - Allow for unlinkat(2) when removing directory. Added CAP_RENAMEAT: - Required for source directory for the renameat(2) syscall. Added CAP_UNLINKAT (effectively it replaces CAP_DELETE and CAP_RMDIR): - Allow for unlinkat(2) on any object. - Required if target of renameat(2) exists and will be removed by this call. Removed CAP_MAPEXEC. CAP_MMAP old behaviour: - Allow for mmap(2) with any combination of PROT_NONE, PROT_READ and PROT_WRITE. CAP_MMAP new behaviour: - Allow for mmap(2)+PROT_NONE. Added CAP_MMAP_R: - Allow for mmap(PROT_READ). Added CAP_MMAP_W: - Allow for mmap(PROT_WRITE). Added CAP_MMAP_X: - Allow for mmap(PROT_EXEC). Added CAP_MMAP_RW: - Allow for mmap(PROT_READ | PROT_WRITE). Added CAP_MMAP_RX: - Allow for mmap(PROT_READ | PROT_EXEC). Added CAP_MMAP_WX: - Allow for mmap(PROT_WRITE | PROT_EXEC). Added CAP_MMAP_RWX: - Allow for mmap(PROT_READ | PROT_WRITE | PROT_EXEC). Renamed CAP_MKDIR to CAP_MKDIRAT. Renamed CAP_MKFIFO to CAP_MKFIFOAT. Renamed CAP_MKNODE to CAP_MKNODEAT. CAP_READ old behaviour: - Allow pread(2). - Disallow read(2), readv(2) (if there is no CAP_SEEK). CAP_READ new behaviour: - Allow read(2), readv(2). - Disallow pread(2) (CAP_SEEK was also required). CAP_WRITE old behaviour: - Allow pwrite(2). - Disallow write(2), writev(2) (if there is no CAP_SEEK). CAP_WRITE new behaviour: - Allow write(2), writev(2). - Disallow pwrite(2) (CAP_SEEK was also required). Added convinient defines: #define CAP_PREAD (CAP_SEEK | CAP_READ) #define CAP_PWRITE (CAP_SEEK | CAP_WRITE) #define CAP_MMAP_R (CAP_MMAP | CAP_SEEK | CAP_READ) #define CAP_MMAP_W (CAP_MMAP | CAP_SEEK | CAP_WRITE) #define CAP_MMAP_X (CAP_MMAP | CAP_SEEK | 0x0000000000000008ULL) #define CAP_MMAP_RW (CAP_MMAP_R | CAP_MMAP_W) #define CAP_MMAP_RX (CAP_MMAP_R | CAP_MMAP_X) #define CAP_MMAP_WX (CAP_MMAP_W | CAP_MMAP_X) #define CAP_MMAP_RWX (CAP_MMAP_R | CAP_MMAP_W | CAP_MMAP_X) #define CAP_RECV CAP_READ #define CAP_SEND CAP_WRITE #define CAP_SOCK_CLIENT \ (CAP_CONNECT | CAP_GETPEERNAME | CAP_GETSOCKNAME | CAP_GETSOCKOPT | \ CAP_PEELOFF | CAP_RECV | CAP_SEND | CAP_SETSOCKOPT | CAP_SHUTDOWN) #define CAP_SOCK_SERVER \ (CAP_ACCEPT | CAP_BIND | CAP_GETPEERNAME | CAP_GETSOCKNAME | \ CAP_GETSOCKOPT | CAP_LISTEN | CAP_PEELOFF | CAP_RECV | CAP_SEND | \ CAP_SETSOCKOPT | CAP_SHUTDOWN) Added defines for backward API compatibility: #define CAP_MAPEXEC CAP_MMAP_X #define CAP_DELETE CAP_UNLINKAT #define CAP_MKDIR CAP_MKDIRAT #define CAP_RMDIR CAP_UNLINKAT #define CAP_MKFIFO CAP_MKFIFOAT #define CAP_MKNOD CAP_MKNODAT #define CAP_SOCK_ALL (CAP_SOCK_CLIENT | CAP_SOCK_SERVER) Sponsored by: The FreeBSD Foundation Reviewed by: Christoph Mallon <christoph.mallon@gmx.de> Many aspects discussed with: rwatson, benl, jonathan ABI compatibility discussed with: kib
2013-03-02 00:53:12 +00:00
fdep = data;
Hopefully improve control message passing over Unix domain sockets. 1) Allow the sending of more than one control message at a time over a unix domain socket. This should cover the PR 29499. 2) This requires that unp_{ex,in}ternalize and unp_scan understand mbufs with more than one control message at a time. 3) Internalize and externalize used to work on the mbuf in-place. This made life quite complicated and the code for sizeof(int) < sizeof(file *) could end up doing the wrong thing. The patch always create a new mbuf/cluster now. This resulted in the change of the prototype for the domain externalise function. 4) You can now send SCM_TIMESTAMP messages. 5) Always use CMSG_DATA(cm) to determine the start where the data in unp_{ex,in}ternalize. It was using ((struct cmsghdr *)cm + 1) in some places, which gives the wrong alignment on the alpha. (NetBSD made this fix some time ago). This results in an ABI change for discriptor passing and creds passing on the alpha. (Probably on the IA64 and Spare ports too). 6) Fix userland programs to use CMSG_* macros too. 7) Be more careful about freeing mbufs containing (file *)s. This is made possible by the prototype change of externalise. PR: 29499 MFC after: 6 weeks
2001-10-04 13:11:48 +00:00
s/discriptors/descriptors/
2003-03-23 19:41:34 +00:00
/* If we're not outputting the descriptors free them. */
Hopefully improve control message passing over Unix domain sockets. 1) Allow the sending of more than one control message at a time over a unix domain socket. This should cover the PR 29499. 2) This requires that unp_{ex,in}ternalize and unp_scan understand mbufs with more than one control message at a time. 3) Internalize and externalize used to work on the mbuf in-place. This made life quite complicated and the code for sizeof(int) < sizeof(file *) could end up doing the wrong thing. The patch always create a new mbuf/cluster now. This resulted in the change of the prototype for the domain externalise function. 4) You can now send SCM_TIMESTAMP messages. 5) Always use CMSG_DATA(cm) to determine the start where the data in unp_{ex,in}ternalize. It was using ((struct cmsghdr *)cm + 1) in some places, which gives the wrong alignment on the alpha. (NetBSD made this fix some time ago). This results in an ABI change for discriptor passing and creds passing on the alpha. (Probably on the IA64 and Spare ports too). 6) Fix userland programs to use CMSG_* macros too. 7) Be more careful about freeing mbufs containing (file *)s. This is made possible by the prototype change of externalise. PR: 29499 MFC after: 6 weeks
2001-10-04 13:11:48 +00:00
if (error || controlp == NULL) {
Merge Capsicum overhaul: - Capability is no longer separate descriptor type. Now every descriptor has set of its own capability rights. - The cap_new(2) system call is left, but it is no longer documented and should not be used in new code. - The new syscall cap_rights_limit(2) should be used instead of cap_new(2), which limits capability rights of the given descriptor without creating a new one. - The cap_getrights(2) syscall is renamed to cap_rights_get(2). - If CAP_IOCTL capability right is present we can further reduce allowed ioctls list with the new cap_ioctls_limit(2) syscall. List of allowed ioctls can be retrived with cap_ioctls_get(2) syscall. - If CAP_FCNTL capability right is present we can further reduce fcntls that can be used with the new cap_fcntls_limit(2) syscall and retrive them with cap_fcntls_get(2). - To support ioctl and fcntl white-listing the filedesc structure was heavly modified. - The audit subsystem, kdump and procstat tools were updated to recognize new syscalls. - Capability rights were revised and eventhough I tried hard to provide backward API and ABI compatibility there are some incompatible changes that are described in detail below: CAP_CREATE old behaviour: - Allow for openat(2)+O_CREAT. - Allow for linkat(2). - Allow for symlinkat(2). CAP_CREATE new behaviour: - Allow for openat(2)+O_CREAT. Added CAP_LINKAT: - Allow for linkat(2). ABI: Reuses CAP_RMDIR bit. - Allow to be target for renameat(2). Added CAP_SYMLINKAT: - Allow for symlinkat(2). Removed CAP_DELETE. Old behaviour: - Allow for unlinkat(2) when removing non-directory object. - Allow to be source for renameat(2). Removed CAP_RMDIR. Old behaviour: - Allow for unlinkat(2) when removing directory. Added CAP_RENAMEAT: - Required for source directory for the renameat(2) syscall. Added CAP_UNLINKAT (effectively it replaces CAP_DELETE and CAP_RMDIR): - Allow for unlinkat(2) on any object. - Required if target of renameat(2) exists and will be removed by this call. Removed CAP_MAPEXEC. CAP_MMAP old behaviour: - Allow for mmap(2) with any combination of PROT_NONE, PROT_READ and PROT_WRITE. CAP_MMAP new behaviour: - Allow for mmap(2)+PROT_NONE. Added CAP_MMAP_R: - Allow for mmap(PROT_READ). Added CAP_MMAP_W: - Allow for mmap(PROT_WRITE). Added CAP_MMAP_X: - Allow for mmap(PROT_EXEC). Added CAP_MMAP_RW: - Allow for mmap(PROT_READ | PROT_WRITE). Added CAP_MMAP_RX: - Allow for mmap(PROT_READ | PROT_EXEC). Added CAP_MMAP_WX: - Allow for mmap(PROT_WRITE | PROT_EXEC). Added CAP_MMAP_RWX: - Allow for mmap(PROT_READ | PROT_WRITE | PROT_EXEC). Renamed CAP_MKDIR to CAP_MKDIRAT. Renamed CAP_MKFIFO to CAP_MKFIFOAT. Renamed CAP_MKNODE to CAP_MKNODEAT. CAP_READ old behaviour: - Allow pread(2). - Disallow read(2), readv(2) (if there is no CAP_SEEK). CAP_READ new behaviour: - Allow read(2), readv(2). - Disallow pread(2) (CAP_SEEK was also required). CAP_WRITE old behaviour: - Allow pwrite(2). - Disallow write(2), writev(2) (if there is no CAP_SEEK). CAP_WRITE new behaviour: - Allow write(2), writev(2). - Disallow pwrite(2) (CAP_SEEK was also required). Added convinient defines: #define CAP_PREAD (CAP_SEEK | CAP_READ) #define CAP_PWRITE (CAP_SEEK | CAP_WRITE) #define CAP_MMAP_R (CAP_MMAP | CAP_SEEK | CAP_READ) #define CAP_MMAP_W (CAP_MMAP | CAP_SEEK | CAP_WRITE) #define CAP_MMAP_X (CAP_MMAP | CAP_SEEK | 0x0000000000000008ULL) #define CAP_MMAP_RW (CAP_MMAP_R | CAP_MMAP_W) #define CAP_MMAP_RX (CAP_MMAP_R | CAP_MMAP_X) #define CAP_MMAP_WX (CAP_MMAP_W | CAP_MMAP_X) #define CAP_MMAP_RWX (CAP_MMAP_R | CAP_MMAP_W | CAP_MMAP_X) #define CAP_RECV CAP_READ #define CAP_SEND CAP_WRITE #define CAP_SOCK_CLIENT \ (CAP_CONNECT | CAP_GETPEERNAME | CAP_GETSOCKNAME | CAP_GETSOCKOPT | \ CAP_PEELOFF | CAP_RECV | CAP_SEND | CAP_SETSOCKOPT | CAP_SHUTDOWN) #define CAP_SOCK_SERVER \ (CAP_ACCEPT | CAP_BIND | CAP_GETPEERNAME | CAP_GETSOCKNAME | \ CAP_GETSOCKOPT | CAP_LISTEN | CAP_PEELOFF | CAP_RECV | CAP_SEND | \ CAP_SETSOCKOPT | CAP_SHUTDOWN) Added defines for backward API compatibility: #define CAP_MAPEXEC CAP_MMAP_X #define CAP_DELETE CAP_UNLINKAT #define CAP_MKDIR CAP_MKDIRAT #define CAP_RMDIR CAP_UNLINKAT #define CAP_MKFIFO CAP_MKFIFOAT #define CAP_MKNOD CAP_MKNODAT #define CAP_SOCK_ALL (CAP_SOCK_CLIENT | CAP_SOCK_SERVER) Sponsored by: The FreeBSD Foundation Reviewed by: Christoph Mallon <christoph.mallon@gmx.de> Many aspects discussed with: rwatson, benl, jonathan ABI compatibility discussed with: kib
2013-03-02 00:53:12 +00:00
unp_freerights(fdep, newfds);
Hopefully improve control message passing over Unix domain sockets. 1) Allow the sending of more than one control message at a time over a unix domain socket. This should cover the PR 29499. 2) This requires that unp_{ex,in}ternalize and unp_scan understand mbufs with more than one control message at a time. 3) Internalize and externalize used to work on the mbuf in-place. This made life quite complicated and the code for sizeof(int) < sizeof(file *) could end up doing the wrong thing. The patch always create a new mbuf/cluster now. This resulted in the change of the prototype for the domain externalise function. 4) You can now send SCM_TIMESTAMP messages. 5) Always use CMSG_DATA(cm) to determine the start where the data in unp_{ex,in}ternalize. It was using ((struct cmsghdr *)cm + 1) in some places, which gives the wrong alignment on the alpha. (NetBSD made this fix some time ago). This results in an ABI change for discriptor passing and creds passing on the alpha. (Probably on the IA64 and Spare ports too). 6) Fix userland programs to use CMSG_* macros too. 7) Be more careful about freeing mbufs containing (file *)s. This is made possible by the prototype change of externalise. PR: 29499 MFC after: 6 weeks
2001-10-04 13:11:48 +00:00
goto next;
}
Merge Capsicum overhaul: - Capability is no longer separate descriptor type. Now every descriptor has set of its own capability rights. - The cap_new(2) system call is left, but it is no longer documented and should not be used in new code. - The new syscall cap_rights_limit(2) should be used instead of cap_new(2), which limits capability rights of the given descriptor without creating a new one. - The cap_getrights(2) syscall is renamed to cap_rights_get(2). - If CAP_IOCTL capability right is present we can further reduce allowed ioctls list with the new cap_ioctls_limit(2) syscall. List of allowed ioctls can be retrived with cap_ioctls_get(2) syscall. - If CAP_FCNTL capability right is present we can further reduce fcntls that can be used with the new cap_fcntls_limit(2) syscall and retrive them with cap_fcntls_get(2). - To support ioctl and fcntl white-listing the filedesc structure was heavly modified. - The audit subsystem, kdump and procstat tools were updated to recognize new syscalls. - Capability rights were revised and eventhough I tried hard to provide backward API and ABI compatibility there are some incompatible changes that are described in detail below: CAP_CREATE old behaviour: - Allow for openat(2)+O_CREAT. - Allow for linkat(2). - Allow for symlinkat(2). CAP_CREATE new behaviour: - Allow for openat(2)+O_CREAT. Added CAP_LINKAT: - Allow for linkat(2). ABI: Reuses CAP_RMDIR bit. - Allow to be target for renameat(2). Added CAP_SYMLINKAT: - Allow for symlinkat(2). Removed CAP_DELETE. Old behaviour: - Allow for unlinkat(2) when removing non-directory object. - Allow to be source for renameat(2). Removed CAP_RMDIR. Old behaviour: - Allow for unlinkat(2) when removing directory. Added CAP_RENAMEAT: - Required for source directory for the renameat(2) syscall. Added CAP_UNLINKAT (effectively it replaces CAP_DELETE and CAP_RMDIR): - Allow for unlinkat(2) on any object. - Required if target of renameat(2) exists and will be removed by this call. Removed CAP_MAPEXEC. CAP_MMAP old behaviour: - Allow for mmap(2) with any combination of PROT_NONE, PROT_READ and PROT_WRITE. CAP_MMAP new behaviour: - Allow for mmap(2)+PROT_NONE. Added CAP_MMAP_R: - Allow for mmap(PROT_READ). Added CAP_MMAP_W: - Allow for mmap(PROT_WRITE). Added CAP_MMAP_X: - Allow for mmap(PROT_EXEC). Added CAP_MMAP_RW: - Allow for mmap(PROT_READ | PROT_WRITE). Added CAP_MMAP_RX: - Allow for mmap(PROT_READ | PROT_EXEC). Added CAP_MMAP_WX: - Allow for mmap(PROT_WRITE | PROT_EXEC). Added CAP_MMAP_RWX: - Allow for mmap(PROT_READ | PROT_WRITE | PROT_EXEC). Renamed CAP_MKDIR to CAP_MKDIRAT. Renamed CAP_MKFIFO to CAP_MKFIFOAT. Renamed CAP_MKNODE to CAP_MKNODEAT. CAP_READ old behaviour: - Allow pread(2). - Disallow read(2), readv(2) (if there is no CAP_SEEK). CAP_READ new behaviour: - Allow read(2), readv(2). - Disallow pread(2) (CAP_SEEK was also required). CAP_WRITE old behaviour: - Allow pwrite(2). - Disallow write(2), writev(2) (if there is no CAP_SEEK). CAP_WRITE new behaviour: - Allow write(2), writev(2). - Disallow pwrite(2) (CAP_SEEK was also required). Added convinient defines: #define CAP_PREAD (CAP_SEEK | CAP_READ) #define CAP_PWRITE (CAP_SEEK | CAP_WRITE) #define CAP_MMAP_R (CAP_MMAP | CAP_SEEK | CAP_READ) #define CAP_MMAP_W (CAP_MMAP | CAP_SEEK | CAP_WRITE) #define CAP_MMAP_X (CAP_MMAP | CAP_SEEK | 0x0000000000000008ULL) #define CAP_MMAP_RW (CAP_MMAP_R | CAP_MMAP_W) #define CAP_MMAP_RX (CAP_MMAP_R | CAP_MMAP_X) #define CAP_MMAP_WX (CAP_MMAP_W | CAP_MMAP_X) #define CAP_MMAP_RWX (CAP_MMAP_R | CAP_MMAP_W | CAP_MMAP_X) #define CAP_RECV CAP_READ #define CAP_SEND CAP_WRITE #define CAP_SOCK_CLIENT \ (CAP_CONNECT | CAP_GETPEERNAME | CAP_GETSOCKNAME | CAP_GETSOCKOPT | \ CAP_PEELOFF | CAP_RECV | CAP_SEND | CAP_SETSOCKOPT | CAP_SHUTDOWN) #define CAP_SOCK_SERVER \ (CAP_ACCEPT | CAP_BIND | CAP_GETPEERNAME | CAP_GETSOCKNAME | \ CAP_GETSOCKOPT | CAP_LISTEN | CAP_PEELOFF | CAP_RECV | CAP_SEND | \ CAP_SETSOCKOPT | CAP_SHUTDOWN) Added defines for backward API compatibility: #define CAP_MAPEXEC CAP_MMAP_X #define CAP_DELETE CAP_UNLINKAT #define CAP_MKDIR CAP_MKDIRAT #define CAP_RMDIR CAP_UNLINKAT #define CAP_MKFIFO CAP_MKFIFOAT #define CAP_MKNOD CAP_MKNODAT #define CAP_SOCK_ALL (CAP_SOCK_CLIENT | CAP_SOCK_SERVER) Sponsored by: The FreeBSD Foundation Reviewed by: Christoph Mallon <christoph.mallon@gmx.de> Many aspects discussed with: rwatson, benl, jonathan ABI compatibility discussed with: kib
2013-03-02 00:53:12 +00:00
FILEDESC_XLOCK(fdesc);
Remove or update several stale comments. A bit of whitespace/style cleanup. Update copyright. MFC after: 3 days (applicable changes)
2008-10-03 09:01:55 +00:00
Enable SCM_RIGHTS on alpha. Allocate necessary buffer as conversion between int and struct file *. Approved by: jkh Submitted by: brian Reviewed by: bde, brian, peter
2000-03-09 15:15:27 +00:00
/*
Re-wrap and other minor formatting and punctuation fixes for UNIX domain socket comments.
2006-07-22 17:24:55 +00:00
* Now change each pointer to an fd in the global
* table to an integer that is the index to the local
* fd table entry that we set up to point to the
* global one we are transferring.
Enable SCM_RIGHTS on alpha. Allocate necessary buffer as conversion between int and struct file *. Approved by: jkh Submitted by: brian Reviewed by: bde, brian, peter
2000-03-09 15:15:27 +00:00
*/
Hopefully improve control message passing over Unix domain sockets. 1) Allow the sending of more than one control message at a time over a unix domain socket. This should cover the PR 29499. 2) This requires that unp_{ex,in}ternalize and unp_scan understand mbufs with more than one control message at a time. 3) Internalize and externalize used to work on the mbuf in-place. This made life quite complicated and the code for sizeof(int) < sizeof(file *) could end up doing the wrong thing. The patch always create a new mbuf/cluster now. This resulted in the change of the prototype for the domain externalise function. 4) You can now send SCM_TIMESTAMP messages. 5) Always use CMSG_DATA(cm) to determine the start where the data in unp_{ex,in}ternalize. It was using ((struct cmsghdr *)cm + 1) in some places, which gives the wrong alignment on the alpha. (NetBSD made this fix some time ago). This results in an ABI change for discriptor passing and creds passing on the alpha. (Probably on the IA64 and Spare ports too). 6) Fix userland programs to use CMSG_* macros too. 7) Be more careful about freeing mbufs containing (file *)s. This is made possible by the prototype change of externalise. PR: 29499 MFC after: 6 weeks
2001-10-04 13:11:48 +00:00
newlen = newfds * sizeof(int);
*controlp = sbcreatecontrol(NULL, newlen,
SCM_RIGHTS, SOL_SOCKET);
if (*controlp == NULL) {
Merge Capsicum overhaul: - Capability is no longer separate descriptor type. Now every descriptor has set of its own capability rights. - The cap_new(2) system call is left, but it is no longer documented and should not be used in new code. - The new syscall cap_rights_limit(2) should be used instead of cap_new(2), which limits capability rights of the given descriptor without creating a new one. - The cap_getrights(2) syscall is renamed to cap_rights_get(2). - If CAP_IOCTL capability right is present we can further reduce allowed ioctls list with the new cap_ioctls_limit(2) syscall. List of allowed ioctls can be retrived with cap_ioctls_get(2) syscall. - If CAP_FCNTL capability right is present we can further reduce fcntls that can be used with the new cap_fcntls_limit(2) syscall and retrive them with cap_fcntls_get(2). - To support ioctl and fcntl white-listing the filedesc structure was heavly modified. - The audit subsystem, kdump and procstat tools were updated to recognize new syscalls. - Capability rights were revised and eventhough I tried hard to provide backward API and ABI compatibility there are some incompatible changes that are described in detail below: CAP_CREATE old behaviour: - Allow for openat(2)+O_CREAT. - Allow for linkat(2). - Allow for symlinkat(2). CAP_CREATE new behaviour: - Allow for openat(2)+O_CREAT. Added CAP_LINKAT: - Allow for linkat(2). ABI: Reuses CAP_RMDIR bit. - Allow to be target for renameat(2). Added CAP_SYMLINKAT: - Allow for symlinkat(2). Removed CAP_DELETE. Old behaviour: - Allow for unlinkat(2) when removing non-directory object. - Allow to be source for renameat(2). Removed CAP_RMDIR. Old behaviour: - Allow for unlinkat(2) when removing directory. Added CAP_RENAMEAT: - Required for source directory for the renameat(2) syscall. Added CAP_UNLINKAT (effectively it replaces CAP_DELETE and CAP_RMDIR): - Allow for unlinkat(2) on any object. - Required if target of renameat(2) exists and will be removed by this call. Removed CAP_MAPEXEC. CAP_MMAP old behaviour: - Allow for mmap(2) with any combination of PROT_NONE, PROT_READ and PROT_WRITE. CAP_MMAP new behaviour: - Allow for mmap(2)+PROT_NONE. Added CAP_MMAP_R: - Allow for mmap(PROT_READ). Added CAP_MMAP_W: - Allow for mmap(PROT_WRITE). Added CAP_MMAP_X: - Allow for mmap(PROT_EXEC). Added CAP_MMAP_RW: - Allow for mmap(PROT_READ | PROT_WRITE). Added CAP_MMAP_RX: - Allow for mmap(PROT_READ | PROT_EXEC). Added CAP_MMAP_WX: - Allow for mmap(PROT_WRITE | PROT_EXEC). Added CAP_MMAP_RWX: - Allow for mmap(PROT_READ | PROT_WRITE | PROT_EXEC). Renamed CAP_MKDIR to CAP_MKDIRAT. Renamed CAP_MKFIFO to CAP_MKFIFOAT. Renamed CAP_MKNODE to CAP_MKNODEAT. CAP_READ old behaviour: - Allow pread(2). - Disallow read(2), readv(2) (if there is no CAP_SEEK). CAP_READ new behaviour: - Allow read(2), readv(2). - Disallow pread(2) (CAP_SEEK was also required). CAP_WRITE old behaviour: - Allow pwrite(2). - Disallow write(2), writev(2) (if there is no CAP_SEEK). CAP_WRITE new behaviour: - Allow write(2), writev(2). - Disallow pwrite(2) (CAP_SEEK was also required). Added convinient defines: #define CAP_PREAD (CAP_SEEK | CAP_READ) #define CAP_PWRITE (CAP_SEEK | CAP_WRITE) #define CAP_MMAP_R (CAP_MMAP | CAP_SEEK | CAP_READ) #define CAP_MMAP_W (CAP_MMAP | CAP_SEEK | CAP_WRITE) #define CAP_MMAP_X (CAP_MMAP | CAP_SEEK | 0x0000000000000008ULL) #define CAP_MMAP_RW (CAP_MMAP_R | CAP_MMAP_W) #define CAP_MMAP_RX (CAP_MMAP_R | CAP_MMAP_X) #define CAP_MMAP_WX (CAP_MMAP_W | CAP_MMAP_X) #define CAP_MMAP_RWX (CAP_MMAP_R | CAP_MMAP_W | CAP_MMAP_X) #define CAP_RECV CAP_READ #define CAP_SEND CAP_WRITE #define CAP_SOCK_CLIENT \ (CAP_CONNECT | CAP_GETPEERNAME | CAP_GETSOCKNAME | CAP_GETSOCKOPT | \ CAP_PEELOFF | CAP_RECV | CAP_SEND | CAP_SETSOCKOPT | CAP_SHUTDOWN) #define CAP_SOCK_SERVER \ (CAP_ACCEPT | CAP_BIND | CAP_GETPEERNAME | CAP_GETSOCKNAME | \ CAP_GETSOCKOPT | CAP_LISTEN | CAP_PEELOFF | CAP_RECV | CAP_SEND | \ CAP_SETSOCKOPT | CAP_SHUTDOWN) Added defines for backward API compatibility: #define CAP_MAPEXEC CAP_MMAP_X #define CAP_DELETE CAP_UNLINKAT #define CAP_MKDIR CAP_MKDIRAT #define CAP_RMDIR CAP_UNLINKAT #define CAP_MKFIFO CAP_MKFIFOAT #define CAP_MKNOD CAP_MKNODAT #define CAP_SOCK_ALL (CAP_SOCK_CLIENT | CAP_SOCK_SERVER) Sponsored by: The FreeBSD Foundation Reviewed by: Christoph Mallon <christoph.mallon@gmx.de> Many aspects discussed with: rwatson, benl, jonathan ABI compatibility discussed with: kib
2013-03-02 00:53:12 +00:00
FILEDESC_XUNLOCK(fdesc);
Hopefully improve control message passing over Unix domain sockets. 1) Allow the sending of more than one control message at a time over a unix domain socket. This should cover the PR 29499. 2) This requires that unp_{ex,in}ternalize and unp_scan understand mbufs with more than one control message at a time. 3) Internalize and externalize used to work on the mbuf in-place. This made life quite complicated and the code for sizeof(int) < sizeof(file *) could end up doing the wrong thing. The patch always create a new mbuf/cluster now. This resulted in the change of the prototype for the domain externalise function. 4) You can now send SCM_TIMESTAMP messages. 5) Always use CMSG_DATA(cm) to determine the start where the data in unp_{ex,in}ternalize. It was using ((struct cmsghdr *)cm + 1) in some places, which gives the wrong alignment on the alpha. (NetBSD made this fix some time ago). This results in an ABI change for discriptor passing and creds passing on the alpha. (Probably on the IA64 and Spare ports too). 6) Fix userland programs to use CMSG_* macros too. 7) Be more careful about freeing mbufs containing (file *)s. This is made possible by the prototype change of externalise. PR: 29499 MFC after: 6 weeks
2001-10-04 13:11:48 +00:00
error = E2BIG;
Merge Capsicum overhaul: - Capability is no longer separate descriptor type. Now every descriptor has set of its own capability rights. - The cap_new(2) system call is left, but it is no longer documented and should not be used in new code. - The new syscall cap_rights_limit(2) should be used instead of cap_new(2), which limits capability rights of the given descriptor without creating a new one. - The cap_getrights(2) syscall is renamed to cap_rights_get(2). - If CAP_IOCTL capability right is present we can further reduce allowed ioctls list with the new cap_ioctls_limit(2) syscall. List of allowed ioctls can be retrived with cap_ioctls_get(2) syscall. - If CAP_FCNTL capability right is present we can further reduce fcntls that can be used with the new cap_fcntls_limit(2) syscall and retrive them with cap_fcntls_get(2). - To support ioctl and fcntl white-listing the filedesc structure was heavly modified. - The audit subsystem, kdump and procstat tools were updated to recognize new syscalls. - Capability rights were revised and eventhough I tried hard to provide backward API and ABI compatibility there are some incompatible changes that are described in detail below: CAP_CREATE old behaviour: - Allow for openat(2)+O_CREAT. - Allow for linkat(2). - Allow for symlinkat(2). CAP_CREATE new behaviour: - Allow for openat(2)+O_CREAT. Added CAP_LINKAT: - Allow for linkat(2). ABI: Reuses CAP_RMDIR bit. - Allow to be target for renameat(2). Added CAP_SYMLINKAT: - Allow for symlinkat(2). Removed CAP_DELETE. Old behaviour: - Allow for unlinkat(2) when removing non-directory object. - Allow to be source for renameat(2). Removed CAP_RMDIR. Old behaviour: - Allow for unlinkat(2) when removing directory. Added CAP_RENAMEAT: - Required for source directory for the renameat(2) syscall. Added CAP_UNLINKAT (effectively it replaces CAP_DELETE and CAP_RMDIR): - Allow for unlinkat(2) on any object. - Required if target of renameat(2) exists and will be removed by this call. Removed CAP_MAPEXEC. CAP_MMAP old behaviour: - Allow for mmap(2) with any combination of PROT_NONE, PROT_READ and PROT_WRITE. CAP_MMAP new behaviour: - Allow for mmap(2)+PROT_NONE. Added CAP_MMAP_R: - Allow for mmap(PROT_READ). Added CAP_MMAP_W: - Allow for mmap(PROT_WRITE). Added CAP_MMAP_X: - Allow for mmap(PROT_EXEC). Added CAP_MMAP_RW: - Allow for mmap(PROT_READ | PROT_WRITE). Added CAP_MMAP_RX: - Allow for mmap(PROT_READ | PROT_EXEC). Added CAP_MMAP_WX: - Allow for mmap(PROT_WRITE | PROT_EXEC). Added CAP_MMAP_RWX: - Allow for mmap(PROT_READ | PROT_WRITE | PROT_EXEC). Renamed CAP_MKDIR to CAP_MKDIRAT. Renamed CAP_MKFIFO to CAP_MKFIFOAT. Renamed CAP_MKNODE to CAP_MKNODEAT. CAP_READ old behaviour: - Allow pread(2). - Disallow read(2), readv(2) (if there is no CAP_SEEK). CAP_READ new behaviour: - Allow read(2), readv(2). - Disallow pread(2) (CAP_SEEK was also required). CAP_WRITE old behaviour: - Allow pwrite(2). - Disallow write(2), writev(2) (if there is no CAP_SEEK). CAP_WRITE new behaviour: - Allow write(2), writev(2). - Disallow pwrite(2) (CAP_SEEK was also required). Added convinient defines: #define CAP_PREAD (CAP_SEEK | CAP_READ) #define CAP_PWRITE (CAP_SEEK | CAP_WRITE) #define CAP_MMAP_R (CAP_MMAP | CAP_SEEK | CAP_READ) #define CAP_MMAP_W (CAP_MMAP | CAP_SEEK | CAP_WRITE) #define CAP_MMAP_X (CAP_MMAP | CAP_SEEK | 0x0000000000000008ULL) #define CAP_MMAP_RW (CAP_MMAP_R | CAP_MMAP_W) #define CAP_MMAP_RX (CAP_MMAP_R | CAP_MMAP_X) #define CAP_MMAP_WX (CAP_MMAP_W | CAP_MMAP_X) #define CAP_MMAP_RWX (CAP_MMAP_R | CAP_MMAP_W | CAP_MMAP_X) #define CAP_RECV CAP_READ #define CAP_SEND CAP_WRITE #define CAP_SOCK_CLIENT \ (CAP_CONNECT | CAP_GETPEERNAME | CAP_GETSOCKNAME | CAP_GETSOCKOPT | \ CAP_PEELOFF | CAP_RECV | CAP_SEND | CAP_SETSOCKOPT | CAP_SHUTDOWN) #define CAP_SOCK_SERVER \ (CAP_ACCEPT | CAP_BIND | CAP_GETPEERNAME | CAP_GETSOCKNAME | \ CAP_GETSOCKOPT | CAP_LISTEN | CAP_PEELOFF | CAP_RECV | CAP_SEND | \ CAP_SETSOCKOPT | CAP_SHUTDOWN) Added defines for backward API compatibility: #define CAP_MAPEXEC CAP_MMAP_X #define CAP_DELETE CAP_UNLINKAT #define CAP_MKDIR CAP_MKDIRAT #define CAP_RMDIR CAP_UNLINKAT #define CAP_MKFIFO CAP_MKFIFOAT #define CAP_MKNOD CAP_MKNODAT #define CAP_SOCK_ALL (CAP_SOCK_CLIENT | CAP_SOCK_SERVER) Sponsored by: The FreeBSD Foundation Reviewed by: Christoph Mallon <christoph.mallon@gmx.de> Many aspects discussed with: rwatson, benl, jonathan ABI compatibility discussed with: kib
2013-03-02 00:53:12 +00:00
unp_freerights(fdep, newfds);
Hopefully improve control message passing over Unix domain sockets. 1) Allow the sending of more than one control message at a time over a unix domain socket. This should cover the PR 29499. 2) This requires that unp_{ex,in}ternalize and unp_scan understand mbufs with more than one control message at a time. 3) Internalize and externalize used to work on the mbuf in-place. This made life quite complicated and the code for sizeof(int) < sizeof(file *) could end up doing the wrong thing. The patch always create a new mbuf/cluster now. This resulted in the change of the prototype for the domain externalise function. 4) You can now send SCM_TIMESTAMP messages. 5) Always use CMSG_DATA(cm) to determine the start where the data in unp_{ex,in}ternalize. It was using ((struct cmsghdr *)cm + 1) in some places, which gives the wrong alignment on the alpha. (NetBSD made this fix some time ago). This results in an ABI change for discriptor passing and creds passing on the alpha. (Probably on the IA64 and Spare ports too). 6) Fix userland programs to use CMSG_* macros too. 7) Be more careful about freeing mbufs containing (file *)s. This is made possible by the prototype change of externalise. PR: 29499 MFC after: 6 weeks
2001-10-04 13:11:48 +00:00
goto next;
}
fdp = (int *)
CMSG_DATA(mtod(*controlp, struct cmsghdr *));
Add fdallocn function and use it when passing fds over unix socket. This gets rid of "unp_externalize fdalloc failed" panic. Reviewed by: pjd MFC after: 1 week
2013-04-14 17:08:34 +00:00
if (fdallocn(td, 0, fdp, newfds) != 0) {
ussreq: use saved fdp pointer insted of td->td_proc->p_fd No functional changes.
2015-06-12 06:28:22 +00:00
FILEDESC_XUNLOCK(fdesc);
Add fdallocn function and use it when passing fds over unix socket. This gets rid of "unp_externalize fdalloc failed" panic. Reviewed by: pjd MFC after: 1 week
2013-04-14 17:08:34 +00:00
error = EMSGSIZE;
unp_freerights(fdep, newfds);
m_freem(*controlp);
*controlp = NULL;
goto next;
}
For some reason when I started to pass filedescent structures instead of pointers to the file structure receiving descriptors stopped to work when also at least few kilobytes of data is being send. In the kernel the soreceive_generic() function doesn't see control mbuf as the first mbuf and unp_externalize() is never called, first 6(?) kilobytes of data is missing as well on receiving end. This breaks for example tmux. I don't know yet why going from 8 bytes to sizeof(struct filedescent) per descriptor (or even to 16 bytes per descriptor) breaks things, but to work-around it for now use 8 bytes per file descriptor at the cost of memory allocation. Reported by: flo, Diane Bruce, Jan Beich <jbeich@tormail.org> Simple testcase provided by: mjg
2013-03-03 23:39:30 +00:00
for (i = 0; i < newfds; i++, fdp++) {
fd: move out actual fp installation to _finstall Use it in fd passing functions as the first step towards fd code cleanup.
2015-06-14 14:08:52 +00:00
_finstall(fdesc, fdep[i]->fde_file, *fdp,
(flags & MSG_CMSG_CLOEXEC) != 0 ? UF_EXCLOSE : 0,
&fdep[i]->fde_caps);
unp_externalize_fp(fdep[i]->fde_file);
Hopefully improve control message passing over Unix domain sockets. 1) Allow the sending of more than one control message at a time over a unix domain socket. This should cover the PR 29499. 2) This requires that unp_{ex,in}ternalize and unp_scan understand mbufs with more than one control message at a time. 3) Internalize and externalize used to work on the mbuf in-place. This made life quite complicated and the code for sizeof(int) < sizeof(file *) could end up doing the wrong thing. The patch always create a new mbuf/cluster now. This resulted in the change of the prototype for the domain externalise function. 4) You can now send SCM_TIMESTAMP messages. 5) Always use CMSG_DATA(cm) to determine the start where the data in unp_{ex,in}ternalize. It was using ((struct cmsghdr *)cm + 1) in some places, which gives the wrong alignment on the alpha. (NetBSD made this fix some time ago). This results in an ABI change for discriptor passing and creds passing on the alpha. (Probably on the IA64 and Spare ports too). 6) Fix userland programs to use CMSG_* macros too. 7) Be more careful about freeing mbufs containing (file *)s. This is made possible by the prototype change of externalise. PR: 29499 MFC after: 6 weeks
2001-10-04 13:11:48 +00:00
}
Merge Capsicum overhaul: - Capability is no longer separate descriptor type. Now every descriptor has set of its own capability rights. - The cap_new(2) system call is left, but it is no longer documented and should not be used in new code. - The new syscall cap_rights_limit(2) should be used instead of cap_new(2), which limits capability rights of the given descriptor without creating a new one. - The cap_getrights(2) syscall is renamed to cap_rights_get(2). - If CAP_IOCTL capability right is present we can further reduce allowed ioctls list with the new cap_ioctls_limit(2) syscall. List of allowed ioctls can be retrived with cap_ioctls_get(2) syscall. - If CAP_FCNTL capability right is present we can further reduce fcntls that can be used with the new cap_fcntls_limit(2) syscall and retrive them with cap_fcntls_get(2). - To support ioctl and fcntl white-listing the filedesc structure was heavly modified. - The audit subsystem, kdump and procstat tools were updated to recognize new syscalls. - Capability rights were revised and eventhough I tried hard to provide backward API and ABI compatibility there are some incompatible changes that are described in detail below: CAP_CREATE old behaviour: - Allow for openat(2)+O_CREAT. - Allow for linkat(2). - Allow for symlinkat(2). CAP_CREATE new behaviour: - Allow for openat(2)+O_CREAT. Added CAP_LINKAT: - Allow for linkat(2). ABI: Reuses CAP_RMDIR bit. - Allow to be target for renameat(2). Added CAP_SYMLINKAT: - Allow for symlinkat(2). Removed CAP_DELETE. Old behaviour: - Allow for unlinkat(2) when removing non-directory object. - Allow to be source for renameat(2). Removed CAP_RMDIR. Old behaviour: - Allow for unlinkat(2) when removing directory. Added CAP_RENAMEAT: - Required for source directory for the renameat(2) syscall. Added CAP_UNLINKAT (effectively it replaces CAP_DELETE and CAP_RMDIR): - Allow for unlinkat(2) on any object. - Required if target of renameat(2) exists and will be removed by this call. Removed CAP_MAPEXEC. CAP_MMAP old behaviour: - Allow for mmap(2) with any combination of PROT_NONE, PROT_READ and PROT_WRITE. CAP_MMAP new behaviour: - Allow for mmap(2)+PROT_NONE. Added CAP_MMAP_R: - Allow for mmap(PROT_READ). Added CAP_MMAP_W: - Allow for mmap(PROT_WRITE). Added CAP_MMAP_X: - Allow for mmap(PROT_EXEC). Added CAP_MMAP_RW: - Allow for mmap(PROT_READ | PROT_WRITE). Added CAP_MMAP_RX: - Allow for mmap(PROT_READ | PROT_EXEC). Added CAP_MMAP_WX: - Allow for mmap(PROT_WRITE | PROT_EXEC). Added CAP_MMAP_RWX: - Allow for mmap(PROT_READ | PROT_WRITE | PROT_EXEC). Renamed CAP_MKDIR to CAP_MKDIRAT. Renamed CAP_MKFIFO to CAP_MKFIFOAT. Renamed CAP_MKNODE to CAP_MKNODEAT. CAP_READ old behaviour: - Allow pread(2). - Disallow read(2), readv(2) (if there is no CAP_SEEK). CAP_READ new behaviour: - Allow read(2), readv(2). - Disallow pread(2) (CAP_SEEK was also required). CAP_WRITE old behaviour: - Allow pwrite(2). - Disallow write(2), writev(2) (if there is no CAP_SEEK). CAP_WRITE new behaviour: - Allow write(2), writev(2). - Disallow pwrite(2) (CAP_SEEK was also required). Added convinient defines: #define CAP_PREAD (CAP_SEEK | CAP_READ) #define CAP_PWRITE (CAP_SEEK | CAP_WRITE) #define CAP_MMAP_R (CAP_MMAP | CAP_SEEK | CAP_READ) #define CAP_MMAP_W (CAP_MMAP | CAP_SEEK | CAP_WRITE) #define CAP_MMAP_X (CAP_MMAP | CAP_SEEK | 0x0000000000000008ULL) #define CAP_MMAP_RW (CAP_MMAP_R | CAP_MMAP_W) #define CAP_MMAP_RX (CAP_MMAP_R | CAP_MMAP_X) #define CAP_MMAP_WX (CAP_MMAP_W | CAP_MMAP_X) #define CAP_MMAP_RWX (CAP_MMAP_R | CAP_MMAP_W | CAP_MMAP_X) #define CAP_RECV CAP_READ #define CAP_SEND CAP_WRITE #define CAP_SOCK_CLIENT \ (CAP_CONNECT | CAP_GETPEERNAME | CAP_GETSOCKNAME | CAP_GETSOCKOPT | \ CAP_PEELOFF | CAP_RECV | CAP_SEND | CAP_SETSOCKOPT | CAP_SHUTDOWN) #define CAP_SOCK_SERVER \ (CAP_ACCEPT | CAP_BIND | CAP_GETPEERNAME | CAP_GETSOCKNAME | \ CAP_GETSOCKOPT | CAP_LISTEN | CAP_PEELOFF | CAP_RECV | CAP_SEND | \ CAP_SETSOCKOPT | CAP_SHUTDOWN) Added defines for backward API compatibility: #define CAP_MAPEXEC CAP_MMAP_X #define CAP_DELETE CAP_UNLINKAT #define CAP_MKDIR CAP_MKDIRAT #define CAP_RMDIR CAP_UNLINKAT #define CAP_MKFIFO CAP_MKFIFOAT #define CAP_MKNOD CAP_MKNODAT #define CAP_SOCK_ALL (CAP_SOCK_CLIENT | CAP_SOCK_SERVER) Sponsored by: The FreeBSD Foundation Reviewed by: Christoph Mallon <christoph.mallon@gmx.de> Many aspects discussed with: rwatson, benl, jonathan ABI compatibility discussed with: kib
2013-03-02 00:53:12 +00:00
FILEDESC_XUNLOCK(fdesc);
Improve r250890, so that we stop processing of a message with zero descriptors as early as possible, and assert that number of descriptors is positive in unp_freerights(). Reviewed by: mjg, pjd, jilles
2013-06-04 11:19:08 +00:00
free(fdep[0], M_FILECAPS);
Re-wrap and other minor formatting and punctuation fixes for UNIX domain socket comments.
2006-07-22 17:24:55 +00:00
} else {
/* We can just copy anything else across. */
Hopefully improve control message passing over Unix domain sockets. 1) Allow the sending of more than one control message at a time over a unix domain socket. This should cover the PR 29499. 2) This requires that unp_{ex,in}ternalize and unp_scan understand mbufs with more than one control message at a time. 3) Internalize and externalize used to work on the mbuf in-place. This made life quite complicated and the code for sizeof(int) < sizeof(file *) could end up doing the wrong thing. The patch always create a new mbuf/cluster now. This resulted in the change of the prototype for the domain externalise function. 4) You can now send SCM_TIMESTAMP messages. 5) Always use CMSG_DATA(cm) to determine the start where the data in unp_{ex,in}ternalize. It was using ((struct cmsghdr *)cm + 1) in some places, which gives the wrong alignment on the alpha. (NetBSD made this fix some time ago). This results in an ABI change for discriptor passing and creds passing on the alpha. (Probably on the IA64 and Spare ports too). 6) Fix userland programs to use CMSG_* macros too. 7) Be more careful about freeing mbufs containing (file *)s. This is made possible by the prototype change of externalise. PR: 29499 MFC after: 6 weeks
2001-10-04 13:11:48 +00:00
if (error || controlp == NULL)
goto next;
*controlp = sbcreatecontrol(NULL, datalen,
cm->cmsg_type, cm->cmsg_level);
if (*controlp == NULL) {
error = ENOBUFS;
goto next;
}
bcopy(data,
CMSG_DATA(mtod(*controlp, struct cmsghdr *)),
datalen);
Enable SCM_RIGHTS on alpha. Allocate necessary buffer as conversion between int and struct file *. Approved by: jkh Submitted by: brian Reviewed by: bde, brian, peter
2000-03-09 15:15:27 +00:00
}
Hopefully improve control message passing over Unix domain sockets. 1) Allow the sending of more than one control message at a time over a unix domain socket. This should cover the PR 29499. 2) This requires that unp_{ex,in}ternalize and unp_scan understand mbufs with more than one control message at a time. 3) Internalize and externalize used to work on the mbuf in-place. This made life quite complicated and the code for sizeof(int) < sizeof(file *) could end up doing the wrong thing. The patch always create a new mbuf/cluster now. This resulted in the change of the prototype for the domain externalise function. 4) You can now send SCM_TIMESTAMP messages. 5) Always use CMSG_DATA(cm) to determine the start where the data in unp_{ex,in}ternalize. It was using ((struct cmsghdr *)cm + 1) in some places, which gives the wrong alignment on the alpha. (NetBSD made this fix some time ago). This results in an ABI change for discriptor passing and creds passing on the alpha. (Probably on the IA64 and Spare ports too). 6) Fix userland programs to use CMSG_* macros too. 7) Be more careful about freeing mbufs containing (file *)s. This is made possible by the prototype change of externalise. PR: 29499 MFC after: 6 weeks
2001-10-04 13:11:48 +00:00
controlp = &(*controlp)->m_next;
next:
if (CMSG_SPACE(datalen) < clen) {
clen -= CMSG_SPACE(datalen);
cm = (struct cmsghdr *)
((caddr_t)cm + CMSG_SPACE(datalen));
} else {
clen = 0;
cm = NULL;
Enable SCM_RIGHTS on alpha. Allocate necessary buffer as conversion between int and struct file *. Approved by: jkh Submitted by: brian Reviewed by: bde, brian, peter
2000-03-09 15:15:27 +00:00
}
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
}
Enable SCM_RIGHTS on alpha. Allocate necessary buffer as conversion between int and struct file *. Approved by: jkh Submitted by: brian Reviewed by: bde, brian, peter
2000-03-09 15:15:27 +00:00
Hopefully improve control message passing over Unix domain sockets. 1) Allow the sending of more than one control message at a time over a unix domain socket. This should cover the PR 29499. 2) This requires that unp_{ex,in}ternalize and unp_scan understand mbufs with more than one control message at a time. 3) Internalize and externalize used to work on the mbuf in-place. This made life quite complicated and the code for sizeof(int) < sizeof(file *) could end up doing the wrong thing. The patch always create a new mbuf/cluster now. This resulted in the change of the prototype for the domain externalise function. 4) You can now send SCM_TIMESTAMP messages. 5) Always use CMSG_DATA(cm) to determine the start where the data in unp_{ex,in}ternalize. It was using ((struct cmsghdr *)cm + 1) in some places, which gives the wrong alignment on the alpha. (NetBSD made this fix some time ago). This results in an ABI change for discriptor passing and creds passing on the alpha. (Probably on the IA64 and Spare ports too). 6) Fix userland programs to use CMSG_* macros too. 7) Be more careful about freeing mbufs containing (file *)s. This is made possible by the prototype change of externalise. PR: 29499 MFC after: 6 weeks
2001-10-04 13:11:48 +00:00
m_freem(control);
return (error);
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
}
Allow for nmbclusters and maxsockets to be increased via sysctl. An eventhandler is used to update all the various zones that depend on these values.
2006-04-21 09:25:40 +00:00
static void
unp_zone_change(void *tag)
{
uma_zone_set_max(unp_zone, maxsockets);
}
Further minor cleanups to UNIX domain sockets: - Staticize and locally prototype functions uipc_ctloutput(), unp_dispose(), unp_init(), and unp_externalize(), none of which have been required outside of uipc_usrreq.c since uipc_proto.c was removed. - Remove stale prototype for uipc_usrreq(), which has not existed in the code since 1997 - Forward declare and staticize uipc_usrreqs structure in uipc_usrreq.c and not un.h. - Comment on why uipc_connect2() is still non-static -- it is used directly by fifofs. - Remove stale comments, tidy up whitespace. MFC after: 3 days (where applicable)
2008-10-03 13:01:56 +00:00
static void
Convert socket structures to be type-stable and add a version number. Define a parameter which indicates the maximum number of sockets in a system, and use this to size the zone allocators used for sockets and for certain PCBs. Convert PF_LOCAL PCB structures to be type-stable and add a version number. Define an external format for infomation about socket structures and use it in several places. Define a mechanism to get all PF_LOCAL and PF_INET PCB lists through sysctl(3) without blocking network interrupts for an unreasonable length of time. This probably still has some bugs and/or race conditions, but it seems to work well enough on my machines. It is now possible for `netstat' to get almost all of its information via the sysctl(3) interface rather than reading kmem (changes to follow).
1998-05-15 20:11:40 +00:00
unp_init(void)
{
Re-wrap and other minor formatting and punctuation fixes for UNIX domain socket comments.
2006-07-22 17:24:55 +00:00
Change the curvnet variable from a global const struct vnet *, previously always pointing to the default vnet context, to a dynamically changing thread-local one. The currvnet context should be set on entry to networking code via CURVNET_SET() macros, and reverted to previous state via CURVNET_RESTORE(). Recursions on curvnet are permitted, though strongly discuouraged. This change should have no functional impact on nooptions VIMAGE kernel builds, where CURVNET_* macros expand to whitespace. The curthread->td_vnet (aka curvnet) variable's purpose is to be an indicator of the vnet context in which the current network-related operation takes place, in case we cannot deduce the current vnet context from any other source, such as by looking at mbuf's m->m_pkthdr.rcvif->if_vnet, sockets's so->so_vnet etc. Moreover, so far curvnet has turned out to be an invaluable consistency checking aid: it helps to catch cases when sockets, ifnets or any other vnet-aware structures may have leaked from one vnet to another. The exact placement of the CURVNET_SET() / CURVNET_RESTORE() macros was a result of an empirical iterative process, whith an aim to reduce recursions on CURVNET_SET() to a minimum, while still reducing the scope of CURVNET_SET() to networking only operations - the alternative would be calling CURVNET_SET() on each system call entry. In general, curvnet has to be set in three typicall cases: when processing socket-related requests from userspace or from within the kernel; when processing inbound traffic flowing from device drivers to upper layers of the networking stack, and when executing timer-driven networking functions. This change also introduces a DDB subcommand to show the list of all vnet instances. Approved by: julian (mentor)
2009-05-05 10:56:12 +00:00
#ifdef VIMAGE
if (!IS_DEFAULT_VNET(curvnet))
return;
#endif
Remove references to vm_zone.h and switch over to the new uma API.
2002-03-20 04:11:52 +00:00
unp_zone = uma_zcreate("unpcb", sizeof(struct unpcb), NULL, NULL,
- Close a race between enumerating UNIX domain socket pcb structures via sysctl and socket teardown by adding a reference count to the UNIX domain pcb object and fixing the sysctl that enumerates unpcbs to grab a reference on each unpcb while it builds the list to copy out to userland. - Close a race between UNIX domain pcb garbage collection (unp_gc()) and file descriptor teardown (fdrop()) by adding a new garbage collection flag FWAIT. unp_gc() sets FWAIT while it walks the message buffers in a UNIX domain socket looking for nested file descriptor references and clears the flag when it is finished. fdrop() checks to see if the flag is set on a file descriptor whose refcount just dropped to 0 and waits for unp_gc() to clear the flag before completely destroying the file descriptor. MFC after: 1 week Reviewed by: rwatson Submitted by: ups Hopefully makes the panics go away: mx1
2007-01-05 19:59:46 +00:00
NULL, NULL, UMA_ALIGN_PTR, 0);
Prefer NULL to 0 when testing and assigning pointer values.
2004-03-30 02:16:25 +00:00
if (unp_zone == NULL)
Convert socket structures to be type-stable and add a version number. Define a parameter which indicates the maximum number of sockets in a system, and use this to size the zone allocators used for sockets and for certain PCBs. Convert PF_LOCAL PCB structures to be type-stable and add a version number. Define an external format for infomation about socket structures and use it in several places. Define a mechanism to get all PF_LOCAL and PF_INET PCB lists through sysctl(3) without blocking network interrupts for an unreasonable length of time. This probably still has some bugs and/or race conditions, but it seems to work well enough on my machines. It is now possible for `netstat' to get almost all of its information via the sysctl(3) interface rather than reading kmem (changes to follow).
1998-05-15 20:11:40 +00:00
panic("unp_init");
Allow for nmbclusters and maxsockets to be increased via sysctl. An eventhandler is used to update all the various zones that depend on these values.
2006-04-21 09:25:40 +00:00
uma_zone_set_max(unp_zone, maxsockets);
Configure UMA warnings for the following zones: - unp_zone: kern.ipc.maxsockets limit reached - socket_zone: kern.ipc.maxsockets limit reached - zone_mbuf: kern.ipc.nmbufs limit reached - zone_clust: kern.ipc.nmbclusters limit reached - zone_jumbop: kern.ipc.nmbjumbop limit reached - zone_jumbo9: kern.ipc.nmbjumbo9 limit reached - zone_jumbo16: kern.ipc.nmbjumbo16 limit reached Note that those warnings are printed not often than every five minutes and can be globally turned off by setting sysctl/tunable vm.zone_warnings to 0. Discussed on: arch Obtained from: WHEEL Systems MFC after: 2 weeks
2012-12-07 22:30:30 +00:00
uma_zone_set_warning(unp_zone, "kern.ipc.maxsockets limit reached");
Allow for nmbclusters and maxsockets to be increased via sysctl. An eventhandler is used to update all the various zones that depend on these values.
2006-04-21 09:25:40 +00:00
EVENTHANDLER_REGISTER(maxsockets_change, unp_zone_change,
NULL, EVENTHANDLER_PRI_ANY);
Convert socket structures to be type-stable and add a version number. Define a parameter which indicates the maximum number of sockets in a system, and use this to size the zone allocators used for sockets and for certain PCBs. Convert PF_LOCAL PCB structures to be type-stable and add a version number. Define an external format for infomation about socket structures and use it in several places. Define a mechanism to get all PF_LOCAL and PF_INET PCB lists through sysctl(3) without blocking network interrupts for an unreasonable length of time. This probably still has some bugs and/or race conditions, but it seems to work well enough on my machines. It is now possible for `netstat' to get almost all of its information via the sysctl(3) interface rather than reading kmem (changes to follow).
1998-05-15 20:11:40 +00:00
LIST_INIT(&unp_dhead);
LIST_INIT(&unp_shead);
First cut at implementing SOCK_SEQPACKET support for UNIX (local) domain sockets. This allows for reliable bi-directional datagram communication over UNIX domain sockets, in contrast to SOCK_DGRAM (M:N, unreliable) or SOCK_STERAM (bi-directional bytestream). Largely, this reuses existing UNIX domain socket code. This allows applications requiring record- oriented semantics to do so reliably via local IPC. Some implementation notes (also present in XXX comments): - Currently we lack an sbappend variant able to do datagrams and control data without doing addresses, so we mark SOCK_SEQPACKET as PR_ADDR. Adding a new variant will solve this problem. - UNIX domain sockets on FreeBSD provide back-pressure/flow control notification for stream sockets by manipulating the send socket buffer's size during pru_send and pru_rcvd. This trick works less well for SOCK_SEQPACKET as sosend_generic() uses sb_hiwat not just to manage blocking, but also to determine maximum datagram size. Fixing this requires rethinking how back-pressure is done for SOCK_SEQPACKET; in the mean time, it's possible to get EMSGSIZE when buffers fill, instead of blocking. Discussed with: benl Reviewed by: bz, rpaulo MFC after: 3 months Sponsored by: Google
2009-10-05 14:49:16 +00:00
LIST_INIT(&unp_sphead);
Reviewed by: jeff, rwatson MFC after: 1 week
2010-12-03 16:15:44 +00:00
SLIST_INIT(&unp_defers);
Schedule garbage collection run for the in-flight rights passed over the unix domain sockets to the next tick, coalescing the serial calls until the collection fires. The thought is that more work for the collector could arise in the near time, allowing to clean more and not spend too much CPU on repeated collection when there is no garbage. Currently the collection task is fired immediately upon unix domain socket close if there are any rights in flight, which caused excessive CPU usage and too long blocking of the threads waiting for unp_list_lock and unp_link_rwlock in write mode. Robert noted that it would be nice if we could find some heuristic by which we decide whether to run GC a bit more quickly. E.g., if the number of UNIX domain sockets is close to its resource limit, but not quite. Reported and tested by: Markus Gebert <markus.gebert@hostpoint.ch> Reviewed by: rwatson MFC after: 2 weeks
2012-11-20 15:45:48 +00:00
TIMEOUT_TASK_INIT(taskqueue_thread, &unp_gc_task, 0, unp_gc, NULL);
Reviewed by: jeff, rwatson MFC after: 1 week
2010-12-03 16:15:44 +00:00
TASK_INIT(&unp_defer_task, 0, unp_process_defers, NULL);
Decompose the global UNIX domain sockets rwlock into two different locks: a global list/counter/generation counter protected by a new mutex unp_list_lock, and a global linkage rwlock, unp_global_rwlock, which protects the connections between UNIX domain sockets. This eliminates conditional lock acquisition that was previously a property of the global lock being held over sonewconn() leading to a call to uipc_attach(), which also required the global lock, but couldn't rely on it as other paths existed to uipc_attach() that didn't hold it: now uipc_attach() uses only the list lock, which follows the linkage lock in the lock order. It may also reduce contention on the global lock for some workloads. Add global UNIX domain socket locks to hard-coded witness lock order. MFC after: 1 week Discussed with: kris
2009-03-08 21:48:29 +00:00
UNP_LINK_LOCK_INIT();
Reviewed by: jeff, rwatson MFC after: 1 week
2010-12-03 16:15:44 +00:00
UNP_DEFERRED_LOCK_INIT();
Convert socket structures to be type-stable and add a version number. Define a parameter which indicates the maximum number of sockets in a system, and use this to size the zone allocators used for sockets and for certain PCBs. Convert PF_LOCAL PCB structures to be type-stable and add a version number. Define an external format for infomation about socket structures and use it in several places. Define a mechanism to get all PF_LOCAL and PF_INET PCB lists through sysctl(3) without blocking network interrupts for an unreasonable length of time. This probably still has some bugs and/or race conditions, but it seems to work well enough on my machines. It is now possible for `netstat' to get almost all of its information via the sysctl(3) interface rather than reading kmem (changes to follow).
1998-05-15 20:11:40 +00:00
}
Another mega commit to staticize things.
1995-12-14 09:55:16 +00:00
static int
style(9)-ize function headers, remove use of 'register'. MFC after: 3 days
2005-02-20 23:22:13 +00:00
unp_internalize(struct mbuf **controlp, struct thread *td)
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
{
Hopefully improve control message passing over Unix domain sockets. 1) Allow the sending of more than one control message at a time over a unix domain socket. This should cover the PR 29499. 2) This requires that unp_{ex,in}ternalize and unp_scan understand mbufs with more than one control message at a time. 3) Internalize and externalize used to work on the mbuf in-place. This made life quite complicated and the code for sizeof(int) < sizeof(file *) could end up doing the wrong thing. The patch always create a new mbuf/cluster now. This resulted in the change of the prototype for the domain externalise function. 4) You can now send SCM_TIMESTAMP messages. 5) Always use CMSG_DATA(cm) to determine the start where the data in unp_{ex,in}ternalize. It was using ((struct cmsghdr *)cm + 1) in some places, which gives the wrong alignment on the alpha. (NetBSD made this fix some time ago). This results in an ABI change for discriptor passing and creds passing on the alpha. (Probably on the IA64 and Spare ports too). 6) Fix userland programs to use CMSG_* macros too. 7) Be more careful about freeing mbufs containing (file *)s. This is made possible by the prototype change of externalise. PR: 29499 MFC after: 6 weeks
2001-10-04 13:11:48 +00:00
struct mbuf *control = *controlp;
KSE Milestone 2 Note ALL MODULES MUST BE RECOMPILED make the kernel aware that there are smaller units of scheduling than the process. (but only allow one thread per process at this time). This is functionally equivalent to teh previousl -current except that there is a thread associated with each process. Sorry john! (your next MFC will be a doosie!) Reviewed by: peter@freebsd.org, dillon@freebsd.org X-MFC after: ha ha ha ha
2001-09-12 08:38:13 +00:00
struct proc *p = td->td_proc;
Merge Capsicum overhaul: - Capability is no longer separate descriptor type. Now every descriptor has set of its own capability rights. - The cap_new(2) system call is left, but it is no longer documented and should not be used in new code. - The new syscall cap_rights_limit(2) should be used instead of cap_new(2), which limits capability rights of the given descriptor without creating a new one. - The cap_getrights(2) syscall is renamed to cap_rights_get(2). - If CAP_IOCTL capability right is present we can further reduce allowed ioctls list with the new cap_ioctls_limit(2) syscall. List of allowed ioctls can be retrived with cap_ioctls_get(2) syscall. - If CAP_FCNTL capability right is present we can further reduce fcntls that can be used with the new cap_fcntls_limit(2) syscall and retrive them with cap_fcntls_get(2). - To support ioctl and fcntl white-listing the filedesc structure was heavly modified. - The audit subsystem, kdump and procstat tools were updated to recognize new syscalls. - Capability rights were revised and eventhough I tried hard to provide backward API and ABI compatibility there are some incompatible changes that are described in detail below: CAP_CREATE old behaviour: - Allow for openat(2)+O_CREAT. - Allow for linkat(2). - Allow for symlinkat(2). CAP_CREATE new behaviour: - Allow for openat(2)+O_CREAT. Added CAP_LINKAT: - Allow for linkat(2). ABI: Reuses CAP_RMDIR bit. - Allow to be target for renameat(2). Added CAP_SYMLINKAT: - Allow for symlinkat(2). Removed CAP_DELETE. Old behaviour: - Allow for unlinkat(2) when removing non-directory object. - Allow to be source for renameat(2). Removed CAP_RMDIR. Old behaviour: - Allow for unlinkat(2) when removing directory. Added CAP_RENAMEAT: - Required for source directory for the renameat(2) syscall. Added CAP_UNLINKAT (effectively it replaces CAP_DELETE and CAP_RMDIR): - Allow for unlinkat(2) on any object. - Required if target of renameat(2) exists and will be removed by this call. Removed CAP_MAPEXEC. CAP_MMAP old behaviour: - Allow for mmap(2) with any combination of PROT_NONE, PROT_READ and PROT_WRITE. CAP_MMAP new behaviour: - Allow for mmap(2)+PROT_NONE. Added CAP_MMAP_R: - Allow for mmap(PROT_READ). Added CAP_MMAP_W: - Allow for mmap(PROT_WRITE). Added CAP_MMAP_X: - Allow for mmap(PROT_EXEC). Added CAP_MMAP_RW: - Allow for mmap(PROT_READ | PROT_WRITE). Added CAP_MMAP_RX: - Allow for mmap(PROT_READ | PROT_EXEC). Added CAP_MMAP_WX: - Allow for mmap(PROT_WRITE | PROT_EXEC). Added CAP_MMAP_RWX: - Allow for mmap(PROT_READ | PROT_WRITE | PROT_EXEC). Renamed CAP_MKDIR to CAP_MKDIRAT. Renamed CAP_MKFIFO to CAP_MKFIFOAT. Renamed CAP_MKNODE to CAP_MKNODEAT. CAP_READ old behaviour: - Allow pread(2). - Disallow read(2), readv(2) (if there is no CAP_SEEK). CAP_READ new behaviour: - Allow read(2), readv(2). - Disallow pread(2) (CAP_SEEK was also required). CAP_WRITE old behaviour: - Allow pwrite(2). - Disallow write(2), writev(2) (if there is no CAP_SEEK). CAP_WRITE new behaviour: - Allow write(2), writev(2). - Disallow pwrite(2) (CAP_SEEK was also required). Added convinient defines: #define CAP_PREAD (CAP_SEEK | CAP_READ) #define CAP_PWRITE (CAP_SEEK | CAP_WRITE) #define CAP_MMAP_R (CAP_MMAP | CAP_SEEK | CAP_READ) #define CAP_MMAP_W (CAP_MMAP | CAP_SEEK | CAP_WRITE) #define CAP_MMAP_X (CAP_MMAP | CAP_SEEK | 0x0000000000000008ULL) #define CAP_MMAP_RW (CAP_MMAP_R | CAP_MMAP_W) #define CAP_MMAP_RX (CAP_MMAP_R | CAP_MMAP_X) #define CAP_MMAP_WX (CAP_MMAP_W | CAP_MMAP_X) #define CAP_MMAP_RWX (CAP_MMAP_R | CAP_MMAP_W | CAP_MMAP_X) #define CAP_RECV CAP_READ #define CAP_SEND CAP_WRITE #define CAP_SOCK_CLIENT \ (CAP_CONNECT | CAP_GETPEERNAME | CAP_GETSOCKNAME | CAP_GETSOCKOPT | \ CAP_PEELOFF | CAP_RECV | CAP_SEND | CAP_SETSOCKOPT | CAP_SHUTDOWN) #define CAP_SOCK_SERVER \ (CAP_ACCEPT | CAP_BIND | CAP_GETPEERNAME | CAP_GETSOCKNAME | \ CAP_GETSOCKOPT | CAP_LISTEN | CAP_PEELOFF | CAP_RECV | CAP_SEND | \ CAP_SETSOCKOPT | CAP_SHUTDOWN) Added defines for backward API compatibility: #define CAP_MAPEXEC CAP_MMAP_X #define CAP_DELETE CAP_UNLINKAT #define CAP_MKDIR CAP_MKDIRAT #define CAP_RMDIR CAP_UNLINKAT #define CAP_MKFIFO CAP_MKFIFOAT #define CAP_MKNOD CAP_MKNODAT #define CAP_SOCK_ALL (CAP_SOCK_CLIENT | CAP_SOCK_SERVER) Sponsored by: The FreeBSD Foundation Reviewed by: Christoph Mallon <christoph.mallon@gmx.de> Many aspects discussed with: rwatson, benl, jonathan ABI compatibility discussed with: kib
2013-03-02 00:53:12 +00:00
struct filedesc *fdesc = p->p_fd;
Add support of passing SCM_BINTIME ancillary data object for PF_LOCAL sockets. PR: kern/175883 Submitted by: Andrey Simonenko <simon@comsys.ntu-kpi.kiev.ua> Discussed with: glebius, phk MFC after: 2 weeks
2013-02-15 13:00:20 +00:00
struct bintime *bt;
Hopefully improve control message passing over Unix domain sockets. 1) Allow the sending of more than one control message at a time over a unix domain socket. This should cover the PR 29499. 2) This requires that unp_{ex,in}ternalize and unp_scan understand mbufs with more than one control message at a time. 3) Internalize and externalize used to work on the mbuf in-place. This made life quite complicated and the code for sizeof(int) < sizeof(file *) could end up doing the wrong thing. The patch always create a new mbuf/cluster now. This resulted in the change of the prototype for the domain externalise function. 4) You can now send SCM_TIMESTAMP messages. 5) Always use CMSG_DATA(cm) to determine the start where the data in unp_{ex,in}ternalize. It was using ((struct cmsghdr *)cm + 1) in some places, which gives the wrong alignment on the alpha. (NetBSD made this fix some time ago). This results in an ABI change for discriptor passing and creds passing on the alpha. (Probably on the IA64 and Spare ports too). 6) Fix userland programs to use CMSG_* macros too. 7) Be more careful about freeing mbufs containing (file *)s. This is made possible by the prototype change of externalise. PR: 29499 MFC after: 6 weeks
2001-10-04 13:11:48 +00:00
struct cmsghdr *cm = mtod(control, struct cmsghdr *);
struct cmsgcred *cmcred;
For some reason when I started to pass filedescent structures instead of pointers to the file structure receiving descriptors stopped to work when also at least few kilobytes of data is being send. In the kernel the soreceive_generic() function doesn't see control mbuf as the first mbuf and unp_externalize() is never called, first 6(?) kilobytes of data is missing as well on receiving end. This breaks for example tmux. I don't know yet why going from 8 bytes to sizeof(struct filedescent) per descriptor (or even to 16 bytes per descriptor) breaks things, but to work-around it for now use 8 bytes per file descriptor at the cost of memory allocation. Reported by: flo, Diane Bruce, Jan Beich <jbeich@tormail.org> Simple testcase provided by: mjg
2013-03-03 23:39:30 +00:00
struct filedescent *fde, **fdep, *fdev;
Hopefully improve control message passing over Unix domain sockets. 1) Allow the sending of more than one control message at a time over a unix domain socket. This should cover the PR 29499. 2) This requires that unp_{ex,in}ternalize and unp_scan understand mbufs with more than one control message at a time. 3) Internalize and externalize used to work on the mbuf in-place. This made life quite complicated and the code for sizeof(int) < sizeof(file *) could end up doing the wrong thing. The patch always create a new mbuf/cluster now. This resulted in the change of the prototype for the domain externalise function. 4) You can now send SCM_TIMESTAMP messages. 5) Always use CMSG_DATA(cm) to determine the start where the data in unp_{ex,in}ternalize. It was using ((struct cmsghdr *)cm + 1) in some places, which gives the wrong alignment on the alpha. (NetBSD made this fix some time ago). This results in an ABI change for discriptor passing and creds passing on the alpha. (Probably on the IA64 and Spare ports too). 6) Fix userland programs to use CMSG_* macros too. 7) Be more careful about freeing mbufs containing (file *)s. This is made possible by the prototype change of externalise. PR: 29499 MFC after: 6 weeks
2001-10-04 13:11:48 +00:00
struct file *fp;
struct timeval *tv;
Add a new socket option SO_TS_CLOCK to pick from several different clock sources to return timestamps when SO_TIMESTAMP is enabled. Two additional clock sources are: o nanosecond resolution realtime clock (equivalent of CLOCK_REALTIME); o nanosecond resolution monotonic clock (equivalent of CLOCK_MONOTONIC). In addition to this, this option provides unified interface to get bintime (equivalent of using SO_BINTIME), except it also supported with IPv6 where SO_BINTIME has never been supported. The long term plan is to depreciate SO_BINTIME and move everything to using SO_TS_CLOCK. Idea for this enhancement has been briefly discussed on the Net session during dev summit in Ottawa last June and the general input was positive. This change is believed to benefit network benchmarks/profiling as well as other scenarios where precise time of arrival measurement is necessary. There are two regression test cases as part of this commit: one extends unix domain test code (unix_cmsg) to test new SCM_XXX types and another one implementis totally new test case which exchanges UDP packets between two processes using both conventional methods (i.e. calling clock_gettime(2) before recv(2) and after send(2)), as well as using setsockopt()+recv() in receive path. The resulting delays are checked for sanity for all supported clock types. Reviewed by: adrian, gnn Differential Revision: https://reviews.freebsd.org/D9171
2017-01-16 17:46:38 +00:00
struct timespec *ts;
Cosmetic changes to unp_internalize Don't throw away the result of fget_unlocked. Move fdp increment to for loop to make it consistent with similar code elsewhere. MFC after: 1 week
2014-07-23 18:04:52 +00:00
int i, *fdp;
Hopefully improve control message passing over Unix domain sockets. 1) Allow the sending of more than one control message at a time over a unix domain socket. This should cover the PR 29499. 2) This requires that unp_{ex,in}ternalize and unp_scan understand mbufs with more than one control message at a time. 3) Internalize and externalize used to work on the mbuf in-place. This made life quite complicated and the code for sizeof(int) < sizeof(file *) could end up doing the wrong thing. The patch always create a new mbuf/cluster now. This resulted in the change of the prototype for the domain externalise function. 4) You can now send SCM_TIMESTAMP messages. 5) Always use CMSG_DATA(cm) to determine the start where the data in unp_{ex,in}ternalize. It was using ((struct cmsghdr *)cm + 1) in some places, which gives the wrong alignment on the alpha. (NetBSD made this fix some time ago). This results in an ABI change for discriptor passing and creds passing on the alpha. (Probably on the IA64 and Spare ports too). 6) Fix userland programs to use CMSG_* macros too. 7) Be more careful about freeing mbufs containing (file *)s. This is made possible by the prototype change of externalise. PR: 29499 MFC after: 6 weeks
2001-10-04 13:11:48 +00:00
void *data;
socklen_t clen = control->m_len, datalen;
int error, oldfds;
Enable SCM_RIGHTS on alpha. Allocate necessary buffer as conversion between int and struct file *. Approved by: jkh Submitted by: brian Reviewed by: bde, brian, peter
2000-03-09 15:15:27 +00:00
u_int newlen;
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
Decompose the global UNIX domain sockets rwlock into two different locks: a global list/counter/generation counter protected by a new mutex unp_list_lock, and a global linkage rwlock, unp_global_rwlock, which protects the connections between UNIX domain sockets. This eliminates conditional lock acquisition that was previously a property of the global lock being held over sonewconn() leading to a call to uipc_attach(), which also required the global lock, but couldn't rely on it as other paths existed to uipc_attach() that didn't hold it: now uipc_attach() uses only the list lock, which follows the linkage lock in the lock order. It may also reduce contention on the global lock for some workloads. Add global UNIX domain socket locks to hard-coded witness lock order. MFC after: 1 week Discussed with: kris
2009-03-08 21:48:29 +00:00
UNP_LINK_UNLOCK_ASSERT();
Add UNP_UNLOCK_ASSERT() to asser that the UNIX domain socket subsystem lock is not held. Rather than annotating that the lock is released after calls to unp_detach() with a comment, annotate with an assertion. Assert that the UNIX domain socket subsystem lock is not held when unp_externalize() and unp_internalize() are called.
2004-08-19 01:45:16 +00:00
Hopefully improve control message passing over Unix domain sockets. 1) Allow the sending of more than one control message at a time over a unix domain socket. This should cover the PR 29499. 2) This requires that unp_{ex,in}ternalize and unp_scan understand mbufs with more than one control message at a time. 3) Internalize and externalize used to work on the mbuf in-place. This made life quite complicated and the code for sizeof(int) < sizeof(file *) could end up doing the wrong thing. The patch always create a new mbuf/cluster now. This resulted in the change of the prototype for the domain externalise function. 4) You can now send SCM_TIMESTAMP messages. 5) Always use CMSG_DATA(cm) to determine the start where the data in unp_{ex,in}ternalize. It was using ((struct cmsghdr *)cm + 1) in some places, which gives the wrong alignment on the alpha. (NetBSD made this fix some time ago). This results in an ABI change for discriptor passing and creds passing on the alpha. (Probably on the IA64 and Spare ports too). 6) Fix userland programs to use CMSG_* macros too. 7) Be more careful about freeing mbufs containing (file *)s. This is made possible by the prototype change of externalise. PR: 29499 MFC after: 6 weeks
2001-10-04 13:11:48 +00:00
error = 0;
*controlp = NULL;
while (cm != NULL) {
if (sizeof(*cm) > clen || cm->cmsg_level != SOL_SOCKET
Check lower bound of cmsg_len. If passed cm->cmsg_len was below cmsghdr size the experssion: datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data; would give negative result. However, in practice it would not result in a crash because the kernel would try to obtain garbage fds for given process and would error out with EBADF. PR: 124908 Submitted by: campbell mumble.net (modified a little) MFC after: 1 week
2014-06-27 05:04:36 +00:00
|| cm->cmsg_len > clen || cm->cmsg_len < sizeof(*cm)) {
Hopefully improve control message passing over Unix domain sockets. 1) Allow the sending of more than one control message at a time over a unix domain socket. This should cover the PR 29499. 2) This requires that unp_{ex,in}ternalize and unp_scan understand mbufs with more than one control message at a time. 3) Internalize and externalize used to work on the mbuf in-place. This made life quite complicated and the code for sizeof(int) < sizeof(file *) could end up doing the wrong thing. The patch always create a new mbuf/cluster now. This resulted in the change of the prototype for the domain externalise function. 4) You can now send SCM_TIMESTAMP messages. 5) Always use CMSG_DATA(cm) to determine the start where the data in unp_{ex,in}ternalize. It was using ((struct cmsghdr *)cm + 1) in some places, which gives the wrong alignment on the alpha. (NetBSD made this fix some time ago). This results in an ABI change for discriptor passing and creds passing on the alpha. (Probably on the IA64 and Spare ports too). 6) Fix userland programs to use CMSG_* macros too. 7) Be more careful about freeing mbufs containing (file *)s. This is made possible by the prototype change of externalise. PR: 29499 MFC after: 6 weeks
2001-10-04 13:11:48 +00:00
error = EINVAL;
goto out;
}
data = CMSG_DATA(cm);
datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data;
switch (cm->cmsg_type) {
/*
* Fill in credential information.
*/
case SCM_CREDS:
*controlp = sbcreatecontrol(NULL, sizeof(*cmcred),
SCM_CREDS, SOL_SOCKET);
if (*controlp == NULL) {
error = ENOBUFS;
goto out;
}
cmcred = (struct cmsgcred *)
CMSG_DATA(mtod(*controlp, struct cmsghdr *));
cmcred->cmcred_pid = p->p_pid;
Simple p_ucred -> td_ucred changes to start using the per-thread ucred reference.
2002-02-27 18:32:23 +00:00
cmcred->cmcred_uid = td->td_ucred->cr_ruid;
cmcred->cmcred_gid = td->td_ucred->cr_rgid;
cmcred->cmcred_euid = td->td_ucred->cr_uid;
cmcred->cmcred_ngroups = MIN(td->td_ucred->cr_ngroups,
Further minor cleanups to UNIX domain sockets: - Staticize and locally prototype functions uipc_ctloutput(), unp_dispose(), unp_init(), and unp_externalize(), none of which have been required outside of uipc_usrreq.c since uipc_proto.c was removed. - Remove stale prototype for uipc_usrreq(), which has not existed in the code since 1997 - Forward declare and staticize uipc_usrreqs structure in uipc_usrreq.c and not un.h. - Comment on why uipc_connect2() is still non-static -- it is used directly by fifofs. - Remove stale comments, tidy up whitespace. MFC after: 3 days (where applicable)
2008-10-03 13:01:56 +00:00
CMGROUP_MAX);
Hopefully improve control message passing over Unix domain sockets. 1) Allow the sending of more than one control message at a time over a unix domain socket. This should cover the PR 29499. 2) This requires that unp_{ex,in}ternalize and unp_scan understand mbufs with more than one control message at a time. 3) Internalize and externalize used to work on the mbuf in-place. This made life quite complicated and the code for sizeof(int) < sizeof(file *) could end up doing the wrong thing. The patch always create a new mbuf/cluster now. This resulted in the change of the prototype for the domain externalise function. 4) You can now send SCM_TIMESTAMP messages. 5) Always use CMSG_DATA(cm) to determine the start where the data in unp_{ex,in}ternalize. It was using ((struct cmsghdr *)cm + 1) in some places, which gives the wrong alignment on the alpha. (NetBSD made this fix some time ago). This results in an ABI change for discriptor passing and creds passing on the alpha. (Probably on the IA64 and Spare ports too). 6) Fix userland programs to use CMSG_* macros too. 7) Be more careful about freeing mbufs containing (file *)s. This is made possible by the prototype change of externalise. PR: 29499 MFC after: 6 weeks
2001-10-04 13:11:48 +00:00
for (i = 0; i < cmcred->cmcred_ngroups; i++)
cmcred->cmcred_groups[i] =
Simple p_ucred -> td_ucred changes to start using the per-thread ucred reference.
2002-02-27 18:32:23 +00:00
td->td_ucred->cr_groups[i];
Hopefully improve control message passing over Unix domain sockets. 1) Allow the sending of more than one control message at a time over a unix domain socket. This should cover the PR 29499. 2) This requires that unp_{ex,in}ternalize and unp_scan understand mbufs with more than one control message at a time. 3) Internalize and externalize used to work on the mbuf in-place. This made life quite complicated and the code for sizeof(int) < sizeof(file *) could end up doing the wrong thing. The patch always create a new mbuf/cluster now. This resulted in the change of the prototype for the domain externalise function. 4) You can now send SCM_TIMESTAMP messages. 5) Always use CMSG_DATA(cm) to determine the start where the data in unp_{ex,in}ternalize. It was using ((struct cmsghdr *)cm + 1) in some places, which gives the wrong alignment on the alpha. (NetBSD made this fix some time ago). This results in an ABI change for discriptor passing and creds passing on the alpha. (Probably on the IA64 and Spare ports too). 6) Fix userland programs to use CMSG_* macros too. 7) Be more careful about freeing mbufs containing (file *)s. This is made possible by the prototype change of externalise. PR: 29499 MFC after: 6 weeks
2001-10-04 13:11:48 +00:00
break;
Add support to sendmsg()/recvmsg() for passing credentials between processes using AF_LOCAL sockets. This hack is going to be used with Secure RPC to duplicate a feature of STREAMS which has no real counterpart in sockets (with STREAMS/TLI, you can apparently use t_getinfo() to learn UID of a local process on the other side of a transport endpoint). What happens is this: the client sets up a sendmsg() call with ancillary data using the SCM_CREDS socket-level control message type. It does not need to fill in the structure. When the kernel notices the data, unp_internalize() fills in the cmesgcred structure with the sending process' credentials (UID, EUID, GID, and ancillary groups). This data is later delivered to the receiving process. The receiver can then perform the follwing tests: - Did the client send ancillary data? o Yes, proceed. o No, refuse to authenticate the client. - The the client send data of type SCM_CREDS? o Yes, proceed. o No, refuse to authenticate the client. - Is the cmsgcred structure the right size? o Yes, proceed. o No, signal a possible error. The receiver can now inspect the credential information and use it to authenticate the client.
1997-03-21 16:12:32 +00:00
Hopefully improve control message passing over Unix domain sockets. 1) Allow the sending of more than one control message at a time over a unix domain socket. This should cover the PR 29499. 2) This requires that unp_{ex,in}ternalize and unp_scan understand mbufs with more than one control message at a time. 3) Internalize and externalize used to work on the mbuf in-place. This made life quite complicated and the code for sizeof(int) < sizeof(file *) could end up doing the wrong thing. The patch always create a new mbuf/cluster now. This resulted in the change of the prototype for the domain externalise function. 4) You can now send SCM_TIMESTAMP messages. 5) Always use CMSG_DATA(cm) to determine the start where the data in unp_{ex,in}ternalize. It was using ((struct cmsghdr *)cm + 1) in some places, which gives the wrong alignment on the alpha. (NetBSD made this fix some time ago). This results in an ABI change for discriptor passing and creds passing on the alpha. (Probably on the IA64 and Spare ports too). 6) Fix userland programs to use CMSG_* macros too. 7) Be more careful about freeing mbufs containing (file *)s. This is made possible by the prototype change of externalise. PR: 29499 MFC after: 6 weeks
2001-10-04 13:11:48 +00:00
case SCM_RIGHTS:
oldfds = datalen / sizeof (int);
Improve r250890, so that we stop processing of a message with zero descriptors as early as possible, and assert that number of descriptors is positive in unp_freerights(). Reviewed by: mjg, pjd, jilles
2013-06-04 11:19:08 +00:00
if (oldfds == 0)
break;
Hopefully improve control message passing over Unix domain sockets. 1) Allow the sending of more than one control message at a time over a unix domain socket. This should cover the PR 29499. 2) This requires that unp_{ex,in}ternalize and unp_scan understand mbufs with more than one control message at a time. 3) Internalize and externalize used to work on the mbuf in-place. This made life quite complicated and the code for sizeof(int) < sizeof(file *) could end up doing the wrong thing. The patch always create a new mbuf/cluster now. This resulted in the change of the prototype for the domain externalise function. 4) You can now send SCM_TIMESTAMP messages. 5) Always use CMSG_DATA(cm) to determine the start where the data in unp_{ex,in}ternalize. It was using ((struct cmsghdr *)cm + 1) in some places, which gives the wrong alignment on the alpha. (NetBSD made this fix some time ago). This results in an ABI change for discriptor passing and creds passing on the alpha. (Probably on the IA64 and Spare ports too). 6) Fix userland programs to use CMSG_* macros too. 7) Be more careful about freeing mbufs containing (file *)s. This is made possible by the prototype change of externalise. PR: 29499 MFC after: 6 weeks
2001-10-04 13:11:48 +00:00
/*
Re-wrap and other minor formatting and punctuation fixes for UNIX domain socket comments.
2006-07-22 17:24:55 +00:00
* Check that all the FDs passed in refer to legal
* files. If not, reject the entire operation.
Hopefully improve control message passing over Unix domain sockets. 1) Allow the sending of more than one control message at a time over a unix domain socket. This should cover the PR 29499. 2) This requires that unp_{ex,in}ternalize and unp_scan understand mbufs with more than one control message at a time. 3) Internalize and externalize used to work on the mbuf in-place. This made life quite complicated and the code for sizeof(int) < sizeof(file *) could end up doing the wrong thing. The patch always create a new mbuf/cluster now. This resulted in the change of the prototype for the domain externalise function. 4) You can now send SCM_TIMESTAMP messages. 5) Always use CMSG_DATA(cm) to determine the start where the data in unp_{ex,in}ternalize. It was using ((struct cmsghdr *)cm + 1) in some places, which gives the wrong alignment on the alpha. (NetBSD made this fix some time ago). This results in an ABI change for discriptor passing and creds passing on the alpha. (Probably on the IA64 and Spare ports too). 6) Fix userland programs to use CMSG_* macros too. 7) Be more careful about freeing mbufs containing (file *)s. This is made possible by the prototype change of externalise. PR: 29499 MFC after: 6 weeks
2001-10-04 13:11:48 +00:00
*/
fdp = data;
Merge Capsicum overhaul: - Capability is no longer separate descriptor type. Now every descriptor has set of its own capability rights. - The cap_new(2) system call is left, but it is no longer documented and should not be used in new code. - The new syscall cap_rights_limit(2) should be used instead of cap_new(2), which limits capability rights of the given descriptor without creating a new one. - The cap_getrights(2) syscall is renamed to cap_rights_get(2). - If CAP_IOCTL capability right is present we can further reduce allowed ioctls list with the new cap_ioctls_limit(2) syscall. List of allowed ioctls can be retrived with cap_ioctls_get(2) syscall. - If CAP_FCNTL capability right is present we can further reduce fcntls that can be used with the new cap_fcntls_limit(2) syscall and retrive them with cap_fcntls_get(2). - To support ioctl and fcntl white-listing the filedesc structure was heavly modified. - The audit subsystem, kdump and procstat tools were updated to recognize new syscalls. - Capability rights were revised and eventhough I tried hard to provide backward API and ABI compatibility there are some incompatible changes that are described in detail below: CAP_CREATE old behaviour: - Allow for openat(2)+O_CREAT. - Allow for linkat(2). - Allow for symlinkat(2). CAP_CREATE new behaviour: - Allow for openat(2)+O_CREAT. Added CAP_LINKAT: - Allow for linkat(2). ABI: Reuses CAP_RMDIR bit. - Allow to be target for renameat(2). Added CAP_SYMLINKAT: - Allow for symlinkat(2). Removed CAP_DELETE. Old behaviour: - Allow for unlinkat(2) when removing non-directory object. - Allow to be source for renameat(2). Removed CAP_RMDIR. Old behaviour: - Allow for unlinkat(2) when removing directory. Added CAP_RENAMEAT: - Required for source directory for the renameat(2) syscall. Added CAP_UNLINKAT (effectively it replaces CAP_DELETE and CAP_RMDIR): - Allow for unlinkat(2) on any object. - Required if target of renameat(2) exists and will be removed by this call. Removed CAP_MAPEXEC. CAP_MMAP old behaviour: - Allow for mmap(2) with any combination of PROT_NONE, PROT_READ and PROT_WRITE. CAP_MMAP new behaviour: - Allow for mmap(2)+PROT_NONE. Added CAP_MMAP_R: - Allow for mmap(PROT_READ). Added CAP_MMAP_W: - Allow for mmap(PROT_WRITE). Added CAP_MMAP_X: - Allow for mmap(PROT_EXEC). Added CAP_MMAP_RW: - Allow for mmap(PROT_READ | PROT_WRITE). Added CAP_MMAP_RX: - Allow for mmap(PROT_READ | PROT_EXEC). Added CAP_MMAP_WX: - Allow for mmap(PROT_WRITE | PROT_EXEC). Added CAP_MMAP_RWX: - Allow for mmap(PROT_READ | PROT_WRITE | PROT_EXEC). Renamed CAP_MKDIR to CAP_MKDIRAT. Renamed CAP_MKFIFO to CAP_MKFIFOAT. Renamed CAP_MKNODE to CAP_MKNODEAT. CAP_READ old behaviour: - Allow pread(2). - Disallow read(2), readv(2) (if there is no CAP_SEEK). CAP_READ new behaviour: - Allow read(2), readv(2). - Disallow pread(2) (CAP_SEEK was also required). CAP_WRITE old behaviour: - Allow pwrite(2). - Disallow write(2), writev(2) (if there is no CAP_SEEK). CAP_WRITE new behaviour: - Allow write(2), writev(2). - Disallow pwrite(2) (CAP_SEEK was also required). Added convinient defines: #define CAP_PREAD (CAP_SEEK | CAP_READ) #define CAP_PWRITE (CAP_SEEK | CAP_WRITE) #define CAP_MMAP_R (CAP_MMAP | CAP_SEEK | CAP_READ) #define CAP_MMAP_W (CAP_MMAP | CAP_SEEK | CAP_WRITE) #define CAP_MMAP_X (CAP_MMAP | CAP_SEEK | 0x0000000000000008ULL) #define CAP_MMAP_RW (CAP_MMAP_R | CAP_MMAP_W) #define CAP_MMAP_RX (CAP_MMAP_R | CAP_MMAP_X) #define CAP_MMAP_WX (CAP_MMAP_W | CAP_MMAP_X) #define CAP_MMAP_RWX (CAP_MMAP_R | CAP_MMAP_W | CAP_MMAP_X) #define CAP_RECV CAP_READ #define CAP_SEND CAP_WRITE #define CAP_SOCK_CLIENT \ (CAP_CONNECT | CAP_GETPEERNAME | CAP_GETSOCKNAME | CAP_GETSOCKOPT | \ CAP_PEELOFF | CAP_RECV | CAP_SEND | CAP_SETSOCKOPT | CAP_SHUTDOWN) #define CAP_SOCK_SERVER \ (CAP_ACCEPT | CAP_BIND | CAP_GETPEERNAME | CAP_GETSOCKNAME | \ CAP_GETSOCKOPT | CAP_LISTEN | CAP_PEELOFF | CAP_RECV | CAP_SEND | \ CAP_SETSOCKOPT | CAP_SHUTDOWN) Added defines for backward API compatibility: #define CAP_MAPEXEC CAP_MMAP_X #define CAP_DELETE CAP_UNLINKAT #define CAP_MKDIR CAP_MKDIRAT #define CAP_RMDIR CAP_UNLINKAT #define CAP_MKFIFO CAP_MKFIFOAT #define CAP_MKNOD CAP_MKNODAT #define CAP_SOCK_ALL (CAP_SOCK_CLIENT | CAP_SOCK_SERVER) Sponsored by: The FreeBSD Foundation Reviewed by: Christoph Mallon <christoph.mallon@gmx.de> Many aspects discussed with: rwatson, benl, jonathan ABI compatibility discussed with: kib
2013-03-02 00:53:12 +00:00
FILEDESC_SLOCK(fdesc);
Cosmetic changes to unp_internalize Don't throw away the result of fget_unlocked. Move fdp increment to for loop to make it consistent with similar code elsewhere. MFC after: 1 week
2014-07-23 18:04:52 +00:00
for (i = 0; i < oldfds; i++, fdp++) {
fp = fget_locked(fdesc, *fdp);
if (fp == NULL) {
Merge Capsicum overhaul: - Capability is no longer separate descriptor type. Now every descriptor has set of its own capability rights. - The cap_new(2) system call is left, but it is no longer documented and should not be used in new code. - The new syscall cap_rights_limit(2) should be used instead of cap_new(2), which limits capability rights of the given descriptor without creating a new one. - The cap_getrights(2) syscall is renamed to cap_rights_get(2). - If CAP_IOCTL capability right is present we can further reduce allowed ioctls list with the new cap_ioctls_limit(2) syscall. List of allowed ioctls can be retrived with cap_ioctls_get(2) syscall. - If CAP_FCNTL capability right is present we can further reduce fcntls that can be used with the new cap_fcntls_limit(2) syscall and retrive them with cap_fcntls_get(2). - To support ioctl and fcntl white-listing the filedesc structure was heavly modified. - The audit subsystem, kdump and procstat tools were updated to recognize new syscalls. - Capability rights were revised and eventhough I tried hard to provide backward API and ABI compatibility there are some incompatible changes that are described in detail below: CAP_CREATE old behaviour: - Allow for openat(2)+O_CREAT. - Allow for linkat(2). - Allow for symlinkat(2). CAP_CREATE new behaviour: - Allow for openat(2)+O_CREAT. Added CAP_LINKAT: - Allow for linkat(2). ABI: Reuses CAP_RMDIR bit. - Allow to be target for renameat(2). Added CAP_SYMLINKAT: - Allow for symlinkat(2). Removed CAP_DELETE. Old behaviour: - Allow for unlinkat(2) when removing non-directory object. - Allow to be source for renameat(2). Removed CAP_RMDIR. Old behaviour: - Allow for unlinkat(2) when removing directory. Added CAP_RENAMEAT: - Required for source directory for the renameat(2) syscall. Added CAP_UNLINKAT (effectively it replaces CAP_DELETE and CAP_RMDIR): - Allow for unlinkat(2) on any object. - Required if target of renameat(2) exists and will be removed by this call. Removed CAP_MAPEXEC. CAP_MMAP old behaviour: - Allow for mmap(2) with any combination of PROT_NONE, PROT_READ and PROT_WRITE. CAP_MMAP new behaviour: - Allow for mmap(2)+PROT_NONE. Added CAP_MMAP_R: - Allow for mmap(PROT_READ). Added CAP_MMAP_W: - Allow for mmap(PROT_WRITE). Added CAP_MMAP_X: - Allow for mmap(PROT_EXEC). Added CAP_MMAP_RW: - Allow for mmap(PROT_READ | PROT_WRITE). Added CAP_MMAP_RX: - Allow for mmap(PROT_READ | PROT_EXEC). Added CAP_MMAP_WX: - Allow for mmap(PROT_WRITE | PROT_EXEC). Added CAP_MMAP_RWX: - Allow for mmap(PROT_READ | PROT_WRITE | PROT_EXEC). Renamed CAP_MKDIR to CAP_MKDIRAT. Renamed CAP_MKFIFO to CAP_MKFIFOAT. Renamed CAP_MKNODE to CAP_MKNODEAT. CAP_READ old behaviour: - Allow pread(2). - Disallow read(2), readv(2) (if there is no CAP_SEEK). CAP_READ new behaviour: - Allow read(2), readv(2). - Disallow pread(2) (CAP_SEEK was also required). CAP_WRITE old behaviour: - Allow pwrite(2). - Disallow write(2), writev(2) (if there is no CAP_SEEK). CAP_WRITE new behaviour: - Allow write(2), writev(2). - Disallow pwrite(2) (CAP_SEEK was also required). Added convinient defines: #define CAP_PREAD (CAP_SEEK | CAP_READ) #define CAP_PWRITE (CAP_SEEK | CAP_WRITE) #define CAP_MMAP_R (CAP_MMAP | CAP_SEEK | CAP_READ) #define CAP_MMAP_W (CAP_MMAP | CAP_SEEK | CAP_WRITE) #define CAP_MMAP_X (CAP_MMAP | CAP_SEEK | 0x0000000000000008ULL) #define CAP_MMAP_RW (CAP_MMAP_R | CAP_MMAP_W) #define CAP_MMAP_RX (CAP_MMAP_R | CAP_MMAP_X) #define CAP_MMAP_WX (CAP_MMAP_W | CAP_MMAP_X) #define CAP_MMAP_RWX (CAP_MMAP_R | CAP_MMAP_W | CAP_MMAP_X) #define CAP_RECV CAP_READ #define CAP_SEND CAP_WRITE #define CAP_SOCK_CLIENT \ (CAP_CONNECT | CAP_GETPEERNAME | CAP_GETSOCKNAME | CAP_GETSOCKOPT | \ CAP_PEELOFF | CAP_RECV | CAP_SEND | CAP_SETSOCKOPT | CAP_SHUTDOWN) #define CAP_SOCK_SERVER \ (CAP_ACCEPT | CAP_BIND | CAP_GETPEERNAME | CAP_GETSOCKNAME | \ CAP_GETSOCKOPT | CAP_LISTEN | CAP_PEELOFF | CAP_RECV | CAP_SEND | \ CAP_SETSOCKOPT | CAP_SHUTDOWN) Added defines for backward API compatibility: #define CAP_MAPEXEC CAP_MMAP_X #define CAP_DELETE CAP_UNLINKAT #define CAP_MKDIR CAP_MKDIRAT #define CAP_RMDIR CAP_UNLINKAT #define CAP_MKFIFO CAP_MKFIFOAT #define CAP_MKNOD CAP_MKNODAT #define CAP_SOCK_ALL (CAP_SOCK_CLIENT | CAP_SOCK_SERVER) Sponsored by: The FreeBSD Foundation Reviewed by: Christoph Mallon <christoph.mallon@gmx.de> Many aspects discussed with: rwatson, benl, jonathan ABI compatibility discussed with: kib
2013-03-02 00:53:12 +00:00
FILEDESC_SUNLOCK(fdesc);
Hopefully improve control message passing over Unix domain sockets. 1) Allow the sending of more than one control message at a time over a unix domain socket. This should cover the PR 29499. 2) This requires that unp_{ex,in}ternalize and unp_scan understand mbufs with more than one control message at a time. 3) Internalize and externalize used to work on the mbuf in-place. This made life quite complicated and the code for sizeof(int) < sizeof(file *) could end up doing the wrong thing. The patch always create a new mbuf/cluster now. This resulted in the change of the prototype for the domain externalise function. 4) You can now send SCM_TIMESTAMP messages. 5) Always use CMSG_DATA(cm) to determine the start where the data in unp_{ex,in}ternalize. It was using ((struct cmsghdr *)cm + 1) in some places, which gives the wrong alignment on the alpha. (NetBSD made this fix some time ago). This results in an ABI change for discriptor passing and creds passing on the alpha. (Probably on the IA64 and Spare ports too). 6) Fix userland programs to use CMSG_* macros too. 7) Be more careful about freeing mbufs containing (file *)s. This is made possible by the prototype change of externalise. PR: 29499 MFC after: 6 weeks
2001-10-04 13:11:48 +00:00
error = EBADF;
goto out;
}
Do not allow kqueues to be passed via unix domain sockets.
2003-02-15 06:04:55 +00:00
if (!(fp->f_ops->fo_flags & DFLAG_PASSABLE)) {
Merge Capsicum overhaul: - Capability is no longer separate descriptor type. Now every descriptor has set of its own capability rights. - The cap_new(2) system call is left, but it is no longer documented and should not be used in new code. - The new syscall cap_rights_limit(2) should be used instead of cap_new(2), which limits capability rights of the given descriptor without creating a new one. - The cap_getrights(2) syscall is renamed to cap_rights_get(2). - If CAP_IOCTL capability right is present we can further reduce allowed ioctls list with the new cap_ioctls_limit(2) syscall. List of allowed ioctls can be retrived with cap_ioctls_get(2) syscall. - If CAP_FCNTL capability right is present we can further reduce fcntls that can be used with the new cap_fcntls_limit(2) syscall and retrive them with cap_fcntls_get(2). - To support ioctl and fcntl white-listing the filedesc structure was heavly modified. - The audit subsystem, kdump and procstat tools were updated to recognize new syscalls. - Capability rights were revised and eventhough I tried hard to provide backward API and ABI compatibility there are some incompatible changes that are described in detail below: CAP_CREATE old behaviour: - Allow for openat(2)+O_CREAT. - Allow for linkat(2). - Allow for symlinkat(2). CAP_CREATE new behaviour: - Allow for openat(2)+O_CREAT. Added CAP_LINKAT: - Allow for linkat(2). ABI: Reuses CAP_RMDIR bit. - Allow to be target for renameat(2). Added CAP_SYMLINKAT: - Allow for symlinkat(2). Removed CAP_DELETE. Old behaviour: - Allow for unlinkat(2) when removing non-directory object. - Allow to be source for renameat(2). Removed CAP_RMDIR. Old behaviour: - Allow for unlinkat(2) when removing directory. Added CAP_RENAMEAT: - Required for source directory for the renameat(2) syscall. Added CAP_UNLINKAT (effectively it replaces CAP_DELETE and CAP_RMDIR): - Allow for unlinkat(2) on any object. - Required if target of renameat(2) exists and will be removed by this call. Removed CAP_MAPEXEC. CAP_MMAP old behaviour: - Allow for mmap(2) with any combination of PROT_NONE, PROT_READ and PROT_WRITE. CAP_MMAP new behaviour: - Allow for mmap(2)+PROT_NONE. Added CAP_MMAP_R: - Allow for mmap(PROT_READ). Added CAP_MMAP_W: - Allow for mmap(PROT_WRITE). Added CAP_MMAP_X: - Allow for mmap(PROT_EXEC). Added CAP_MMAP_RW: - Allow for mmap(PROT_READ | PROT_WRITE). Added CAP_MMAP_RX: - Allow for mmap(PROT_READ | PROT_EXEC). Added CAP_MMAP_WX: - Allow for mmap(PROT_WRITE | PROT_EXEC). Added CAP_MMAP_RWX: - Allow for mmap(PROT_READ | PROT_WRITE | PROT_EXEC). Renamed CAP_MKDIR to CAP_MKDIRAT. Renamed CAP_MKFIFO to CAP_MKFIFOAT. Renamed CAP_MKNODE to CAP_MKNODEAT. CAP_READ old behaviour: - Allow pread(2). - Disallow read(2), readv(2) (if there is no CAP_SEEK). CAP_READ new behaviour: - Allow read(2), readv(2). - Disallow pread(2) (CAP_SEEK was also required). CAP_WRITE old behaviour: - Allow pwrite(2). - Disallow write(2), writev(2) (if there is no CAP_SEEK). CAP_WRITE new behaviour: - Allow write(2), writev(2). - Disallow pwrite(2) (CAP_SEEK was also required). Added convinient defines: #define CAP_PREAD (CAP_SEEK | CAP_READ) #define CAP_PWRITE (CAP_SEEK | CAP_WRITE) #define CAP_MMAP_R (CAP_MMAP | CAP_SEEK | CAP_READ) #define CAP_MMAP_W (CAP_MMAP | CAP_SEEK | CAP_WRITE) #define CAP_MMAP_X (CAP_MMAP | CAP_SEEK | 0x0000000000000008ULL) #define CAP_MMAP_RW (CAP_MMAP_R | CAP_MMAP_W) #define CAP_MMAP_RX (CAP_MMAP_R | CAP_MMAP_X) #define CAP_MMAP_WX (CAP_MMAP_W | CAP_MMAP_X) #define CAP_MMAP_RWX (CAP_MMAP_R | CAP_MMAP_W | CAP_MMAP_X) #define CAP_RECV CAP_READ #define CAP_SEND CAP_WRITE #define CAP_SOCK_CLIENT \ (CAP_CONNECT | CAP_GETPEERNAME | CAP_GETSOCKNAME | CAP_GETSOCKOPT | \ CAP_PEELOFF | CAP_RECV | CAP_SEND | CAP_SETSOCKOPT | CAP_SHUTDOWN) #define CAP_SOCK_SERVER \ (CAP_ACCEPT | CAP_BIND | CAP_GETPEERNAME | CAP_GETSOCKNAME | \ CAP_GETSOCKOPT | CAP_LISTEN | CAP_PEELOFF | CAP_RECV | CAP_SEND | \ CAP_SETSOCKOPT | CAP_SHUTDOWN) Added defines for backward API compatibility: #define CAP_MAPEXEC CAP_MMAP_X #define CAP_DELETE CAP_UNLINKAT #define CAP_MKDIR CAP_MKDIRAT #define CAP_RMDIR CAP_UNLINKAT #define CAP_MKFIFO CAP_MKFIFOAT #define CAP_MKNOD CAP_MKNODAT #define CAP_SOCK_ALL (CAP_SOCK_CLIENT | CAP_SOCK_SERVER) Sponsored by: The FreeBSD Foundation Reviewed by: Christoph Mallon <christoph.mallon@gmx.de> Many aspects discussed with: rwatson, benl, jonathan ABI compatibility discussed with: kib
2013-03-02 00:53:12 +00:00
FILEDESC_SUNLOCK(fdesc);
Do not allow kqueues to be passed via unix domain sockets.
2003-02-15 06:04:55 +00:00
error = EOPNOTSUPP;
goto out;
}
Hopefully improve control message passing over Unix domain sockets. 1) Allow the sending of more than one control message at a time over a unix domain socket. This should cover the PR 29499. 2) This requires that unp_{ex,in}ternalize and unp_scan understand mbufs with more than one control message at a time. 3) Internalize and externalize used to work on the mbuf in-place. This made life quite complicated and the code for sizeof(int) < sizeof(file *) could end up doing the wrong thing. The patch always create a new mbuf/cluster now. This resulted in the change of the prototype for the domain externalise function. 4) You can now send SCM_TIMESTAMP messages. 5) Always use CMSG_DATA(cm) to determine the start where the data in unp_{ex,in}ternalize. It was using ((struct cmsghdr *)cm + 1) in some places, which gives the wrong alignment on the alpha. (NetBSD made this fix some time ago). This results in an ABI change for discriptor passing and creds passing on the alpha. (Probably on the IA64 and Spare ports too). 6) Fix userland programs to use CMSG_* macros too. 7) Be more careful about freeing mbufs containing (file *)s. This is made possible by the prototype change of externalise. PR: 29499 MFC after: 6 weeks
2001-10-04 13:11:48 +00:00
}
Replace custom file descriptor array sleep lock constructed using a mutex and flags with an sxlock. This leads to a significant and measurable performance improvement as a result of access to shared locking for frequent lookup operations, reduced general overhead, and reduced overhead in the event of contention. All of these are imported for threaded applications where simultaneous access to a shared file descriptor array occurs frequently. Kris has reported 2x-4x transaction rate improvements on 8-core MySQL benchmarks; smaller improvements can be expected for many workloads as a result of reduced overhead. - Generally eliminate the distinction between "fast" and regular acquisisition of the filedesc lock; the plan is that they will now all be fast. Change all locking instances to either shared or exclusive locks. - Correct a bug (pointed out by kib) in fdfree() where previously msleep() was called without the mutex held; sx_sleep() is now always called with the sxlock held exclusively. - Universally hold the struct file lock over changes to struct file, rather than the filedesc lock or no lock. Always update the f_ops field last. A further memory barrier is required here in the future (discussed with jhb). - Improve locking and reference management in linux_at(), which fails to properly acquire vnode references before using vnode pointers. Annotate improper use of vn_fullpath(), which will be replaced at a future date. In fcntl(), we conservatively acquire an exclusive lock, even though in some cases a shared lock may be sufficient, which should be revisited. The dropping of the filedesc lock in fdgrowtable() is no longer required as the sxlock can be held over the sleep operation; we should consider removing that (pointed out by attilio). Tested by: kris Discussed with: jhb, kris, attilio, jeff
2007-04-04 09:11:34 +00:00
Hopefully improve control message passing over Unix domain sockets. 1) Allow the sending of more than one control message at a time over a unix domain socket. This should cover the PR 29499. 2) This requires that unp_{ex,in}ternalize and unp_scan understand mbufs with more than one control message at a time. 3) Internalize and externalize used to work on the mbuf in-place. This made life quite complicated and the code for sizeof(int) < sizeof(file *) could end up doing the wrong thing. The patch always create a new mbuf/cluster now. This resulted in the change of the prototype for the domain externalise function. 4) You can now send SCM_TIMESTAMP messages. 5) Always use CMSG_DATA(cm) to determine the start where the data in unp_{ex,in}ternalize. It was using ((struct cmsghdr *)cm + 1) in some places, which gives the wrong alignment on the alpha. (NetBSD made this fix some time ago). This results in an ABI change for discriptor passing and creds passing on the alpha. (Probably on the IA64 and Spare ports too). 6) Fix userland programs to use CMSG_* macros too. 7) Be more careful about freeing mbufs containing (file *)s. This is made possible by the prototype change of externalise. PR: 29499 MFC after: 6 weeks
2001-10-04 13:11:48 +00:00
/*
Further minor cleanups to UNIX domain sockets: - Staticize and locally prototype functions uipc_ctloutput(), unp_dispose(), unp_init(), and unp_externalize(), none of which have been required outside of uipc_usrreq.c since uipc_proto.c was removed. - Remove stale prototype for uipc_usrreq(), which has not existed in the code since 1997 - Forward declare and staticize uipc_usrreqs structure in uipc_usrreq.c and not un.h. - Comment on why uipc_connect2() is still non-static -- it is used directly by fifofs. - Remove stale comments, tidy up whitespace. MFC after: 3 days (where applicable)
2008-10-03 13:01:56 +00:00
* Now replace the integer FDs with pointers to the
Merge Capsicum overhaul: - Capability is no longer separate descriptor type. Now every descriptor has set of its own capability rights. - The cap_new(2) system call is left, but it is no longer documented and should not be used in new code. - The new syscall cap_rights_limit(2) should be used instead of cap_new(2), which limits capability rights of the given descriptor without creating a new one. - The cap_getrights(2) syscall is renamed to cap_rights_get(2). - If CAP_IOCTL capability right is present we can further reduce allowed ioctls list with the new cap_ioctls_limit(2) syscall. List of allowed ioctls can be retrived with cap_ioctls_get(2) syscall. - If CAP_FCNTL capability right is present we can further reduce fcntls that can be used with the new cap_fcntls_limit(2) syscall and retrive them with cap_fcntls_get(2). - To support ioctl and fcntl white-listing the filedesc structure was heavly modified. - The audit subsystem, kdump and procstat tools were updated to recognize new syscalls. - Capability rights were revised and eventhough I tried hard to provide backward API and ABI compatibility there are some incompatible changes that are described in detail below: CAP_CREATE old behaviour: - Allow for openat(2)+O_CREAT. - Allow for linkat(2). - Allow for symlinkat(2). CAP_CREATE new behaviour: - Allow for openat(2)+O_CREAT. Added CAP_LINKAT: - Allow for linkat(2). ABI: Reuses CAP_RMDIR bit. - Allow to be target for renameat(2). Added CAP_SYMLINKAT: - Allow for symlinkat(2). Removed CAP_DELETE. Old behaviour: - Allow for unlinkat(2) when removing non-directory object. - Allow to be source for renameat(2). Removed CAP_RMDIR. Old behaviour: - Allow for unlinkat(2) when removing directory. Added CAP_RENAMEAT: - Required for source directory for the renameat(2) syscall. Added CAP_UNLINKAT (effectively it replaces CAP_DELETE and CAP_RMDIR): - Allow for unlinkat(2) on any object. - Required if target of renameat(2) exists and will be removed by this call. Removed CAP_MAPEXEC. CAP_MMAP old behaviour: - Allow for mmap(2) with any combination of PROT_NONE, PROT_READ and PROT_WRITE. CAP_MMAP new behaviour: - Allow for mmap(2)+PROT_NONE. Added CAP_MMAP_R: - Allow for mmap(PROT_READ). Added CAP_MMAP_W: - Allow for mmap(PROT_WRITE). Added CAP_MMAP_X: - Allow for mmap(PROT_EXEC). Added CAP_MMAP_RW: - Allow for mmap(PROT_READ | PROT_WRITE). Added CAP_MMAP_RX: - Allow for mmap(PROT_READ | PROT_EXEC). Added CAP_MMAP_WX: - Allow for mmap(PROT_WRITE | PROT_EXEC). Added CAP_MMAP_RWX: - Allow for mmap(PROT_READ | PROT_WRITE | PROT_EXEC). Renamed CAP_MKDIR to CAP_MKDIRAT. Renamed CAP_MKFIFO to CAP_MKFIFOAT. Renamed CAP_MKNODE to CAP_MKNODEAT. CAP_READ old behaviour: - Allow pread(2). - Disallow read(2), readv(2) (if there is no CAP_SEEK). CAP_READ new behaviour: - Allow read(2), readv(2). - Disallow pread(2) (CAP_SEEK was also required). CAP_WRITE old behaviour: - Allow pwrite(2). - Disallow write(2), writev(2) (if there is no CAP_SEEK). CAP_WRITE new behaviour: - Allow write(2), writev(2). - Disallow pwrite(2) (CAP_SEEK was also required). Added convinient defines: #define CAP_PREAD (CAP_SEEK | CAP_READ) #define CAP_PWRITE (CAP_SEEK | CAP_WRITE) #define CAP_MMAP_R (CAP_MMAP | CAP_SEEK | CAP_READ) #define CAP_MMAP_W (CAP_MMAP | CAP_SEEK | CAP_WRITE) #define CAP_MMAP_X (CAP_MMAP | CAP_SEEK | 0x0000000000000008ULL) #define CAP_MMAP_RW (CAP_MMAP_R | CAP_MMAP_W) #define CAP_MMAP_RX (CAP_MMAP_R | CAP_MMAP_X) #define CAP_MMAP_WX (CAP_MMAP_W | CAP_MMAP_X) #define CAP_MMAP_RWX (CAP_MMAP_R | CAP_MMAP_W | CAP_MMAP_X) #define CAP_RECV CAP_READ #define CAP_SEND CAP_WRITE #define CAP_SOCK_CLIENT \ (CAP_CONNECT | CAP_GETPEERNAME | CAP_GETSOCKNAME | CAP_GETSOCKOPT | \ CAP_PEELOFF | CAP_RECV | CAP_SEND | CAP_SETSOCKOPT | CAP_SHUTDOWN) #define CAP_SOCK_SERVER \ (CAP_ACCEPT | CAP_BIND | CAP_GETPEERNAME | CAP_GETSOCKNAME | \ CAP_GETSOCKOPT | CAP_LISTEN | CAP_PEELOFF | CAP_RECV | CAP_SEND | \ CAP_SETSOCKOPT | CAP_SHUTDOWN) Added defines for backward API compatibility: #define CAP_MAPEXEC CAP_MMAP_X #define CAP_DELETE CAP_UNLINKAT #define CAP_MKDIR CAP_MKDIRAT #define CAP_RMDIR CAP_UNLINKAT #define CAP_MKFIFO CAP_MKFIFOAT #define CAP_MKNOD CAP_MKNODAT #define CAP_SOCK_ALL (CAP_SOCK_CLIENT | CAP_SOCK_SERVER) Sponsored by: The FreeBSD Foundation Reviewed by: Christoph Mallon <christoph.mallon@gmx.de> Many aspects discussed with: rwatson, benl, jonathan ABI compatibility discussed with: kib
2013-03-02 00:53:12 +00:00
* file structure and capability rights.
Hopefully improve control message passing over Unix domain sockets. 1) Allow the sending of more than one control message at a time over a unix domain socket. This should cover the PR 29499. 2) This requires that unp_{ex,in}ternalize and unp_scan understand mbufs with more than one control message at a time. 3) Internalize and externalize used to work on the mbuf in-place. This made life quite complicated and the code for sizeof(int) < sizeof(file *) could end up doing the wrong thing. The patch always create a new mbuf/cluster now. This resulted in the change of the prototype for the domain externalise function. 4) You can now send SCM_TIMESTAMP messages. 5) Always use CMSG_DATA(cm) to determine the start where the data in unp_{ex,in}ternalize. It was using ((struct cmsghdr *)cm + 1) in some places, which gives the wrong alignment on the alpha. (NetBSD made this fix some time ago). This results in an ABI change for discriptor passing and creds passing on the alpha. (Probably on the IA64 and Spare ports too). 6) Fix userland programs to use CMSG_* macros too. 7) Be more careful about freeing mbufs containing (file *)s. This is made possible by the prototype change of externalise. PR: 29499 MFC after: 6 weeks
2001-10-04 13:11:48 +00:00
*/
For some reason when I started to pass filedescent structures instead of pointers to the file structure receiving descriptors stopped to work when also at least few kilobytes of data is being send. In the kernel the soreceive_generic() function doesn't see control mbuf as the first mbuf and unp_externalize() is never called, first 6(?) kilobytes of data is missing as well on receiving end. This breaks for example tmux. I don't know yet why going from 8 bytes to sizeof(struct filedescent) per descriptor (or even to 16 bytes per descriptor) breaks things, but to work-around it for now use 8 bytes per file descriptor at the cost of memory allocation. Reported by: flo, Diane Bruce, Jan Beich <jbeich@tormail.org> Simple testcase provided by: mjg
2013-03-03 23:39:30 +00:00
newlen = oldfds * sizeof(fdep[0]);
Hopefully improve control message passing over Unix domain sockets. 1) Allow the sending of more than one control message at a time over a unix domain socket. This should cover the PR 29499. 2) This requires that unp_{ex,in}ternalize and unp_scan understand mbufs with more than one control message at a time. 3) Internalize and externalize used to work on the mbuf in-place. This made life quite complicated and the code for sizeof(int) < sizeof(file *) could end up doing the wrong thing. The patch always create a new mbuf/cluster now. This resulted in the change of the prototype for the domain externalise function. 4) You can now send SCM_TIMESTAMP messages. 5) Always use CMSG_DATA(cm) to determine the start where the data in unp_{ex,in}ternalize. It was using ((struct cmsghdr *)cm + 1) in some places, which gives the wrong alignment on the alpha. (NetBSD made this fix some time ago). This results in an ABI change for discriptor passing and creds passing on the alpha. (Probably on the IA64 and Spare ports too). 6) Fix userland programs to use CMSG_* macros too. 7) Be more careful about freeing mbufs containing (file *)s. This is made possible by the prototype change of externalise. PR: 29499 MFC after: 6 weeks
2001-10-04 13:11:48 +00:00
*controlp = sbcreatecontrol(NULL, newlen,
SCM_RIGHTS, SOL_SOCKET);
if (*controlp == NULL) {
Merge Capsicum overhaul: - Capability is no longer separate descriptor type. Now every descriptor has set of its own capability rights. - The cap_new(2) system call is left, but it is no longer documented and should not be used in new code. - The new syscall cap_rights_limit(2) should be used instead of cap_new(2), which limits capability rights of the given descriptor without creating a new one. - The cap_getrights(2) syscall is renamed to cap_rights_get(2). - If CAP_IOCTL capability right is present we can further reduce allowed ioctls list with the new cap_ioctls_limit(2) syscall. List of allowed ioctls can be retrived with cap_ioctls_get(2) syscall. - If CAP_FCNTL capability right is present we can further reduce fcntls that can be used with the new cap_fcntls_limit(2) syscall and retrive them with cap_fcntls_get(2). - To support ioctl and fcntl white-listing the filedesc structure was heavly modified. - The audit subsystem, kdump and procstat tools were updated to recognize new syscalls. - Capability rights were revised and eventhough I tried hard to provide backward API and ABI compatibility there are some incompatible changes that are described in detail below: CAP_CREATE old behaviour: - Allow for openat(2)+O_CREAT. - Allow for linkat(2). - Allow for symlinkat(2). CAP_CREATE new behaviour: - Allow for openat(2)+O_CREAT. Added CAP_LINKAT: - Allow for linkat(2). ABI: Reuses CAP_RMDIR bit. - Allow to be target for renameat(2). Added CAP_SYMLINKAT: - Allow for symlinkat(2). Removed CAP_DELETE. Old behaviour: - Allow for unlinkat(2) when removing non-directory object. - Allow to be source for renameat(2). Removed CAP_RMDIR. Old behaviour: - Allow for unlinkat(2) when removing directory. Added CAP_RENAMEAT: - Required for source directory for the renameat(2) syscall. Added CAP_UNLINKAT (effectively it replaces CAP_DELETE and CAP_RMDIR): - Allow for unlinkat(2) on any object. - Required if target of renameat(2) exists and will be removed by this call. Removed CAP_MAPEXEC. CAP_MMAP old behaviour: - Allow for mmap(2) with any combination of PROT_NONE, PROT_READ and PROT_WRITE. CAP_MMAP new behaviour: - Allow for mmap(2)+PROT_NONE. Added CAP_MMAP_R: - Allow for mmap(PROT_READ). Added CAP_MMAP_W: - Allow for mmap(PROT_WRITE). Added CAP_MMAP_X: - Allow for mmap(PROT_EXEC). Added CAP_MMAP_RW: - Allow for mmap(PROT_READ | PROT_WRITE). Added CAP_MMAP_RX: - Allow for mmap(PROT_READ | PROT_EXEC). Added CAP_MMAP_WX: - Allow for mmap(PROT_WRITE | PROT_EXEC). Added CAP_MMAP_RWX: - Allow for mmap(PROT_READ | PROT_WRITE | PROT_EXEC). Renamed CAP_MKDIR to CAP_MKDIRAT. Renamed CAP_MKFIFO to CAP_MKFIFOAT. Renamed CAP_MKNODE to CAP_MKNODEAT. CAP_READ old behaviour: - Allow pread(2). - Disallow read(2), readv(2) (if there is no CAP_SEEK). CAP_READ new behaviour: - Allow read(2), readv(2). - Disallow pread(2) (CAP_SEEK was also required). CAP_WRITE old behaviour: - Allow pwrite(2). - Disallow write(2), writev(2) (if there is no CAP_SEEK). CAP_WRITE new behaviour: - Allow write(2), writev(2). - Disallow pwrite(2) (CAP_SEEK was also required). Added convinient defines: #define CAP_PREAD (CAP_SEEK | CAP_READ) #define CAP_PWRITE (CAP_SEEK | CAP_WRITE) #define CAP_MMAP_R (CAP_MMAP | CAP_SEEK | CAP_READ) #define CAP_MMAP_W (CAP_MMAP | CAP_SEEK | CAP_WRITE) #define CAP_MMAP_X (CAP_MMAP | CAP_SEEK | 0x0000000000000008ULL) #define CAP_MMAP_RW (CAP_MMAP_R | CAP_MMAP_W) #define CAP_MMAP_RX (CAP_MMAP_R | CAP_MMAP_X) #define CAP_MMAP_WX (CAP_MMAP_W | CAP_MMAP_X) #define CAP_MMAP_RWX (CAP_MMAP_R | CAP_MMAP_W | CAP_MMAP_X) #define CAP_RECV CAP_READ #define CAP_SEND CAP_WRITE #define CAP_SOCK_CLIENT \ (CAP_CONNECT | CAP_GETPEERNAME | CAP_GETSOCKNAME | CAP_GETSOCKOPT | \ CAP_PEELOFF | CAP_RECV | CAP_SEND | CAP_SETSOCKOPT | CAP_SHUTDOWN) #define CAP_SOCK_SERVER \ (CAP_ACCEPT | CAP_BIND | CAP_GETPEERNAME | CAP_GETSOCKNAME | \ CAP_GETSOCKOPT | CAP_LISTEN | CAP_PEELOFF | CAP_RECV | CAP_SEND | \ CAP_SETSOCKOPT | CAP_SHUTDOWN) Added defines for backward API compatibility: #define CAP_MAPEXEC CAP_MMAP_X #define CAP_DELETE CAP_UNLINKAT #define CAP_MKDIR CAP_MKDIRAT #define CAP_RMDIR CAP_UNLINKAT #define CAP_MKFIFO CAP_MKFIFOAT #define CAP_MKNOD CAP_MKNODAT #define CAP_SOCK_ALL (CAP_SOCK_CLIENT | CAP_SOCK_SERVER) Sponsored by: The FreeBSD Foundation Reviewed by: Christoph Mallon <christoph.mallon@gmx.de> Many aspects discussed with: rwatson, benl, jonathan ABI compatibility discussed with: kib
2013-03-02 00:53:12 +00:00
FILEDESC_SUNLOCK(fdesc);
Hopefully improve control message passing over Unix domain sockets. 1) Allow the sending of more than one control message at a time over a unix domain socket. This should cover the PR 29499. 2) This requires that unp_{ex,in}ternalize and unp_scan understand mbufs with more than one control message at a time. 3) Internalize and externalize used to work on the mbuf in-place. This made life quite complicated and the code for sizeof(int) < sizeof(file *) could end up doing the wrong thing. The patch always create a new mbuf/cluster now. This resulted in the change of the prototype for the domain externalise function. 4) You can now send SCM_TIMESTAMP messages. 5) Always use CMSG_DATA(cm) to determine the start where the data in unp_{ex,in}ternalize. It was using ((struct cmsghdr *)cm + 1) in some places, which gives the wrong alignment on the alpha. (NetBSD made this fix some time ago). This results in an ABI change for discriptor passing and creds passing on the alpha. (Probably on the IA64 and Spare ports too). 6) Fix userland programs to use CMSG_* macros too. 7) Be more careful about freeing mbufs containing (file *)s. This is made possible by the prototype change of externalise. PR: 29499 MFC after: 6 weeks
2001-10-04 13:11:48 +00:00
error = E2BIG;
goto out;
}
fdp = data;
For some reason when I started to pass filedescent structures instead of pointers to the file structure receiving descriptors stopped to work when also at least few kilobytes of data is being send. In the kernel the soreceive_generic() function doesn't see control mbuf as the first mbuf and unp_externalize() is never called, first 6(?) kilobytes of data is missing as well on receiving end. This breaks for example tmux. I don't know yet why going from 8 bytes to sizeof(struct filedescent) per descriptor (or even to 16 bytes per descriptor) breaks things, but to work-around it for now use 8 bytes per file descriptor at the cost of memory allocation. Reported by: flo, Diane Bruce, Jan Beich <jbeich@tormail.org> Simple testcase provided by: mjg
2013-03-03 23:39:30 +00:00
fdep = (struct filedescent **)
Hopefully improve control message passing over Unix domain sockets. 1) Allow the sending of more than one control message at a time over a unix domain socket. This should cover the PR 29499. 2) This requires that unp_{ex,in}ternalize and unp_scan understand mbufs with more than one control message at a time. 3) Internalize and externalize used to work on the mbuf in-place. This made life quite complicated and the code for sizeof(int) < sizeof(file *) could end up doing the wrong thing. The patch always create a new mbuf/cluster now. This resulted in the change of the prototype for the domain externalise function. 4) You can now send SCM_TIMESTAMP messages. 5) Always use CMSG_DATA(cm) to determine the start where the data in unp_{ex,in}ternalize. It was using ((struct cmsghdr *)cm + 1) in some places, which gives the wrong alignment on the alpha. (NetBSD made this fix some time ago). This results in an ABI change for discriptor passing and creds passing on the alpha. (Probably on the IA64 and Spare ports too). 6) Fix userland programs to use CMSG_* macros too. 7) Be more careful about freeing mbufs containing (file *)s. This is made possible by the prototype change of externalise. PR: 29499 MFC after: 6 weeks
2001-10-04 13:11:48 +00:00
CMSG_DATA(mtod(*controlp, struct cmsghdr *));
For some reason when I started to pass filedescent structures instead of pointers to the file structure receiving descriptors stopped to work when also at least few kilobytes of data is being send. In the kernel the soreceive_generic() function doesn't see control mbuf as the first mbuf and unp_externalize() is never called, first 6(?) kilobytes of data is missing as well on receiving end. This breaks for example tmux. I don't know yet why going from 8 bytes to sizeof(struct filedescent) per descriptor (or even to 16 bytes per descriptor) breaks things, but to work-around it for now use 8 bytes per file descriptor at the cost of memory allocation. Reported by: flo, Diane Bruce, Jan Beich <jbeich@tormail.org> Simple testcase provided by: mjg
2013-03-03 23:39:30 +00:00
fdev = malloc(sizeof(*fdev) * oldfds, M_FILECAPS,
M_WAITOK);
for (i = 0; i < oldfds; i++, fdev++, fdp++) {
Merge Capsicum overhaul: - Capability is no longer separate descriptor type. Now every descriptor has set of its own capability rights. - The cap_new(2) system call is left, but it is no longer documented and should not be used in new code. - The new syscall cap_rights_limit(2) should be used instead of cap_new(2), which limits capability rights of the given descriptor without creating a new one. - The cap_getrights(2) syscall is renamed to cap_rights_get(2). - If CAP_IOCTL capability right is present we can further reduce allowed ioctls list with the new cap_ioctls_limit(2) syscall. List of allowed ioctls can be retrived with cap_ioctls_get(2) syscall. - If CAP_FCNTL capability right is present we can further reduce fcntls that can be used with the new cap_fcntls_limit(2) syscall and retrive them with cap_fcntls_get(2). - To support ioctl and fcntl white-listing the filedesc structure was heavly modified. - The audit subsystem, kdump and procstat tools were updated to recognize new syscalls. - Capability rights were revised and eventhough I tried hard to provide backward API and ABI compatibility there are some incompatible changes that are described in detail below: CAP_CREATE old behaviour: - Allow for openat(2)+O_CREAT. - Allow for linkat(2). - Allow for symlinkat(2). CAP_CREATE new behaviour: - Allow for openat(2)+O_CREAT. Added CAP_LINKAT: - Allow for linkat(2). ABI: Reuses CAP_RMDIR bit. - Allow to be target for renameat(2). Added CAP_SYMLINKAT: - Allow for symlinkat(2). Removed CAP_DELETE. Old behaviour: - Allow for unlinkat(2) when removing non-directory object. - Allow to be source for renameat(2). Removed CAP_RMDIR. Old behaviour: - Allow for unlinkat(2) when removing directory. Added CAP_RENAMEAT: - Required for source directory for the renameat(2) syscall. Added CAP_UNLINKAT (effectively it replaces CAP_DELETE and CAP_RMDIR): - Allow for unlinkat(2) on any object. - Required if target of renameat(2) exists and will be removed by this call. Removed CAP_MAPEXEC. CAP_MMAP old behaviour: - Allow for mmap(2) with any combination of PROT_NONE, PROT_READ and PROT_WRITE. CAP_MMAP new behaviour: - Allow for mmap(2)+PROT_NONE. Added CAP_MMAP_R: - Allow for mmap(PROT_READ). Added CAP_MMAP_W: - Allow for mmap(PROT_WRITE). Added CAP_MMAP_X: - Allow for mmap(PROT_EXEC). Added CAP_MMAP_RW: - Allow for mmap(PROT_READ | PROT_WRITE). Added CAP_MMAP_RX: - Allow for mmap(PROT_READ | PROT_EXEC). Added CAP_MMAP_WX: - Allow for mmap(PROT_WRITE | PROT_EXEC). Added CAP_MMAP_RWX: - Allow for mmap(PROT_READ | PROT_WRITE | PROT_EXEC). Renamed CAP_MKDIR to CAP_MKDIRAT. Renamed CAP_MKFIFO to CAP_MKFIFOAT. Renamed CAP_MKNODE to CAP_MKNODEAT. CAP_READ old behaviour: - Allow pread(2). - Disallow read(2), readv(2) (if there is no CAP_SEEK). CAP_READ new behaviour: - Allow read(2), readv(2). - Disallow pread(2) (CAP_SEEK was also required). CAP_WRITE old behaviour: - Allow pwrite(2). - Disallow write(2), writev(2) (if there is no CAP_SEEK). CAP_WRITE new behaviour: - Allow write(2), writev(2). - Disallow pwrite(2) (CAP_SEEK was also required). Added convinient defines: #define CAP_PREAD (CAP_SEEK | CAP_READ) #define CAP_PWRITE (CAP_SEEK | CAP_WRITE) #define CAP_MMAP_R (CAP_MMAP | CAP_SEEK | CAP_READ) #define CAP_MMAP_W (CAP_MMAP | CAP_SEEK | CAP_WRITE) #define CAP_MMAP_X (CAP_MMAP | CAP_SEEK | 0x0000000000000008ULL) #define CAP_MMAP_RW (CAP_MMAP_R | CAP_MMAP_W) #define CAP_MMAP_RX (CAP_MMAP_R | CAP_MMAP_X) #define CAP_MMAP_WX (CAP_MMAP_W | CAP_MMAP_X) #define CAP_MMAP_RWX (CAP_MMAP_R | CAP_MMAP_W | CAP_MMAP_X) #define CAP_RECV CAP_READ #define CAP_SEND CAP_WRITE #define CAP_SOCK_CLIENT \ (CAP_CONNECT | CAP_GETPEERNAME | CAP_GETSOCKNAME | CAP_GETSOCKOPT | \ CAP_PEELOFF | CAP_RECV | CAP_SEND | CAP_SETSOCKOPT | CAP_SHUTDOWN) #define CAP_SOCK_SERVER \ (CAP_ACCEPT | CAP_BIND | CAP_GETPEERNAME | CAP_GETSOCKNAME | \ CAP_GETSOCKOPT | CAP_LISTEN | CAP_PEELOFF | CAP_RECV | CAP_SEND | \ CAP_SETSOCKOPT | CAP_SHUTDOWN) Added defines for backward API compatibility: #define CAP_MAPEXEC CAP_MMAP_X #define CAP_DELETE CAP_UNLINKAT #define CAP_MKDIR CAP_MKDIRAT #define CAP_RMDIR CAP_UNLINKAT #define CAP_MKFIFO CAP_MKFIFOAT #define CAP_MKNOD CAP_MKNODAT #define CAP_SOCK_ALL (CAP_SOCK_CLIENT | CAP_SOCK_SERVER) Sponsored by: The FreeBSD Foundation Reviewed by: Christoph Mallon <christoph.mallon@gmx.de> Many aspects discussed with: rwatson, benl, jonathan ABI compatibility discussed with: kib
2013-03-02 00:53:12 +00:00
fde = &fdesc->fd_ofiles[*fdp];
For some reason when I started to pass filedescent structures instead of pointers to the file structure receiving descriptors stopped to work when also at least few kilobytes of data is being send. In the kernel the soreceive_generic() function doesn't see control mbuf as the first mbuf and unp_externalize() is never called, first 6(?) kilobytes of data is missing as well on receiving end. This breaks for example tmux. I don't know yet why going from 8 bytes to sizeof(struct filedescent) per descriptor (or even to 16 bytes per descriptor) breaks things, but to work-around it for now use 8 bytes per file descriptor at the cost of memory allocation. Reported by: flo, Diane Bruce, Jan Beich <jbeich@tormail.org> Simple testcase provided by: mjg
2013-03-03 23:39:30 +00:00
fdep[i] = fdev;
fdep[i]->fde_file = fde->fde_file;
filecaps_copy(&fde->fde_caps,
fd: make the common case in filecaps_copy work lockless The filedesc lock is only needed if ioctls caps are present, which is a rare situation. This is a step towards reducing the scope of the filedesc lock.
2015-09-07 20:02:56 +00:00
&fdep[i]->fde_caps, true);
For some reason when I started to pass filedescent structures instead of pointers to the file structure receiving descriptors stopped to work when also at least few kilobytes of data is being send. In the kernel the soreceive_generic() function doesn't see control mbuf as the first mbuf and unp_externalize() is never called, first 6(?) kilobytes of data is missing as well on receiving end. This breaks for example tmux. I don't know yet why going from 8 bytes to sizeof(struct filedescent) per descriptor (or even to 16 bytes per descriptor) breaks things, but to work-around it for now use 8 bytes per file descriptor at the cost of memory allocation. Reported by: flo, Diane Bruce, Jan Beich <jbeich@tormail.org> Simple testcase provided by: mjg
2013-03-03 23:39:30 +00:00
unp_internalize_fp(fdep[i]->fde_file);
Hopefully improve control message passing over Unix domain sockets. 1) Allow the sending of more than one control message at a time over a unix domain socket. This should cover the PR 29499. 2) This requires that unp_{ex,in}ternalize and unp_scan understand mbufs with more than one control message at a time. 3) Internalize and externalize used to work on the mbuf in-place. This made life quite complicated and the code for sizeof(int) < sizeof(file *) could end up doing the wrong thing. The patch always create a new mbuf/cluster now. This resulted in the change of the prototype for the domain externalise function. 4) You can now send SCM_TIMESTAMP messages. 5) Always use CMSG_DATA(cm) to determine the start where the data in unp_{ex,in}ternalize. It was using ((struct cmsghdr *)cm + 1) in some places, which gives the wrong alignment on the alpha. (NetBSD made this fix some time ago). This results in an ABI change for discriptor passing and creds passing on the alpha. (Probably on the IA64 and Spare ports too). 6) Fix userland programs to use CMSG_* macros too. 7) Be more careful about freeing mbufs containing (file *)s. This is made possible by the prototype change of externalise. PR: 29499 MFC after: 6 weeks
2001-10-04 13:11:48 +00:00
}
Merge Capsicum overhaul: - Capability is no longer separate descriptor type. Now every descriptor has set of its own capability rights. - The cap_new(2) system call is left, but it is no longer documented and should not be used in new code. - The new syscall cap_rights_limit(2) should be used instead of cap_new(2), which limits capability rights of the given descriptor without creating a new one. - The cap_getrights(2) syscall is renamed to cap_rights_get(2). - If CAP_IOCTL capability right is present we can further reduce allowed ioctls list with the new cap_ioctls_limit(2) syscall. List of allowed ioctls can be retrived with cap_ioctls_get(2) syscall. - If CAP_FCNTL capability right is present we can further reduce fcntls that can be used with the new cap_fcntls_limit(2) syscall and retrive them with cap_fcntls_get(2). - To support ioctl and fcntl white-listing the filedesc structure was heavly modified. - The audit subsystem, kdump and procstat tools were updated to recognize new syscalls. - Capability rights were revised and eventhough I tried hard to provide backward API and ABI compatibility there are some incompatible changes that are described in detail below: CAP_CREATE old behaviour: - Allow for openat(2)+O_CREAT. - Allow for linkat(2). - Allow for symlinkat(2). CAP_CREATE new behaviour: - Allow for openat(2)+O_CREAT. Added CAP_LINKAT: - Allow for linkat(2). ABI: Reuses CAP_RMDIR bit. - Allow to be target for renameat(2). Added CAP_SYMLINKAT: - Allow for symlinkat(2). Removed CAP_DELETE. Old behaviour: - Allow for unlinkat(2) when removing non-directory object. - Allow to be source for renameat(2). Removed CAP_RMDIR. Old behaviour: - Allow for unlinkat(2) when removing directory. Added CAP_RENAMEAT: - Required for source directory for the renameat(2) syscall. Added CAP_UNLINKAT (effectively it replaces CAP_DELETE and CAP_RMDIR): - Allow for unlinkat(2) on any object. - Required if target of renameat(2) exists and will be removed by this call. Removed CAP_MAPEXEC. CAP_MMAP old behaviour: - Allow for mmap(2) with any combination of PROT_NONE, PROT_READ and PROT_WRITE. CAP_MMAP new behaviour: - Allow for mmap(2)+PROT_NONE. Added CAP_MMAP_R: - Allow for mmap(PROT_READ). Added CAP_MMAP_W: - Allow for mmap(PROT_WRITE). Added CAP_MMAP_X: - Allow for mmap(PROT_EXEC). Added CAP_MMAP_RW: - Allow for mmap(PROT_READ | PROT_WRITE). Added CAP_MMAP_RX: - Allow for mmap(PROT_READ | PROT_EXEC). Added CAP_MMAP_WX: - Allow for mmap(PROT_WRITE | PROT_EXEC). Added CAP_MMAP_RWX: - Allow for mmap(PROT_READ | PROT_WRITE | PROT_EXEC). Renamed CAP_MKDIR to CAP_MKDIRAT. Renamed CAP_MKFIFO to CAP_MKFIFOAT. Renamed CAP_MKNODE to CAP_MKNODEAT. CAP_READ old behaviour: - Allow pread(2). - Disallow read(2), readv(2) (if there is no CAP_SEEK). CAP_READ new behaviour: - Allow read(2), readv(2). - Disallow pread(2) (CAP_SEEK was also required). CAP_WRITE old behaviour: - Allow pwrite(2). - Disallow write(2), writev(2) (if there is no CAP_SEEK). CAP_WRITE new behaviour: - Allow write(2), writev(2). - Disallow pwrite(2) (CAP_SEEK was also required). Added convinient defines: #define CAP_PREAD (CAP_SEEK | CAP_READ) #define CAP_PWRITE (CAP_SEEK | CAP_WRITE) #define CAP_MMAP_R (CAP_MMAP | CAP_SEEK | CAP_READ) #define CAP_MMAP_W (CAP_MMAP | CAP_SEEK | CAP_WRITE) #define CAP_MMAP_X (CAP_MMAP | CAP_SEEK | 0x0000000000000008ULL) #define CAP_MMAP_RW (CAP_MMAP_R | CAP_MMAP_W) #define CAP_MMAP_RX (CAP_MMAP_R | CAP_MMAP_X) #define CAP_MMAP_WX (CAP_MMAP_W | CAP_MMAP_X) #define CAP_MMAP_RWX (CAP_MMAP_R | CAP_MMAP_W | CAP_MMAP_X) #define CAP_RECV CAP_READ #define CAP_SEND CAP_WRITE #define CAP_SOCK_CLIENT \ (CAP_CONNECT | CAP_GETPEERNAME | CAP_GETSOCKNAME | CAP_GETSOCKOPT | \ CAP_PEELOFF | CAP_RECV | CAP_SEND | CAP_SETSOCKOPT | CAP_SHUTDOWN) #define CAP_SOCK_SERVER \ (CAP_ACCEPT | CAP_BIND | CAP_GETPEERNAME | CAP_GETSOCKNAME | \ CAP_GETSOCKOPT | CAP_LISTEN | CAP_PEELOFF | CAP_RECV | CAP_SEND | \ CAP_SETSOCKOPT | CAP_SHUTDOWN) Added defines for backward API compatibility: #define CAP_MAPEXEC CAP_MMAP_X #define CAP_DELETE CAP_UNLINKAT #define CAP_MKDIR CAP_MKDIRAT #define CAP_RMDIR CAP_UNLINKAT #define CAP_MKFIFO CAP_MKFIFOAT #define CAP_MKNOD CAP_MKNODAT #define CAP_SOCK_ALL (CAP_SOCK_CLIENT | CAP_SOCK_SERVER) Sponsored by: The FreeBSD Foundation Reviewed by: Christoph Mallon <christoph.mallon@gmx.de> Many aspects discussed with: rwatson, benl, jonathan ABI compatibility discussed with: kib
2013-03-02 00:53:12 +00:00
FILEDESC_SUNLOCK(fdesc);
Hopefully improve control message passing over Unix domain sockets. 1) Allow the sending of more than one control message at a time over a unix domain socket. This should cover the PR 29499. 2) This requires that unp_{ex,in}ternalize and unp_scan understand mbufs with more than one control message at a time. 3) Internalize and externalize used to work on the mbuf in-place. This made life quite complicated and the code for sizeof(int) < sizeof(file *) could end up doing the wrong thing. The patch always create a new mbuf/cluster now. This resulted in the change of the prototype for the domain externalise function. 4) You can now send SCM_TIMESTAMP messages. 5) Always use CMSG_DATA(cm) to determine the start where the data in unp_{ex,in}ternalize. It was using ((struct cmsghdr *)cm + 1) in some places, which gives the wrong alignment on the alpha. (NetBSD made this fix some time ago). This results in an ABI change for discriptor passing and creds passing on the alpha. (Probably on the IA64 and Spare ports too). 6) Fix userland programs to use CMSG_* macros too. 7) Be more careful about freeing mbufs containing (file *)s. This is made possible by the prototype change of externalise. PR: 29499 MFC after: 6 weeks
2001-10-04 13:11:48 +00:00
break;
Enable SCM_RIGHTS on alpha. Allocate necessary buffer as conversion between int and struct file *. Approved by: jkh Submitted by: brian Reviewed by: bde, brian, peter
2000-03-09 15:15:27 +00:00
Hopefully improve control message passing over Unix domain sockets. 1) Allow the sending of more than one control message at a time over a unix domain socket. This should cover the PR 29499. 2) This requires that unp_{ex,in}ternalize and unp_scan understand mbufs with more than one control message at a time. 3) Internalize and externalize used to work on the mbuf in-place. This made life quite complicated and the code for sizeof(int) < sizeof(file *) could end up doing the wrong thing. The patch always create a new mbuf/cluster now. This resulted in the change of the prototype for the domain externalise function. 4) You can now send SCM_TIMESTAMP messages. 5) Always use CMSG_DATA(cm) to determine the start where the data in unp_{ex,in}ternalize. It was using ((struct cmsghdr *)cm + 1) in some places, which gives the wrong alignment on the alpha. (NetBSD made this fix some time ago). This results in an ABI change for discriptor passing and creds passing on the alpha. (Probably on the IA64 and Spare ports too). 6) Fix userland programs to use CMSG_* macros too. 7) Be more careful about freeing mbufs containing (file *)s. This is made possible by the prototype change of externalise. PR: 29499 MFC after: 6 weeks
2001-10-04 13:11:48 +00:00
case SCM_TIMESTAMP:
*controlp = sbcreatecontrol(NULL, sizeof(*tv),
SCM_TIMESTAMP, SOL_SOCKET);
if (*controlp == NULL) {
error = ENOBUFS;
goto out;
}
tv = (struct timeval *)
CMSG_DATA(mtod(*controlp, struct cmsghdr *));
microtime(tv);
break;
Add support of passing SCM_BINTIME ancillary data object for PF_LOCAL sockets. PR: kern/175883 Submitted by: Andrey Simonenko <simon@comsys.ntu-kpi.kiev.ua> Discussed with: glebius, phk MFC after: 2 weeks
2013-02-15 13:00:20 +00:00
case SCM_BINTIME:
*controlp = sbcreatecontrol(NULL, sizeof(*bt),
SCM_BINTIME, SOL_SOCKET);
if (*controlp == NULL) {
error = ENOBUFS;
goto out;
}
bt = (struct bintime *)
CMSG_DATA(mtod(*controlp, struct cmsghdr *));
bintime(bt);
break;
Add a new socket option SO_TS_CLOCK to pick from several different clock sources to return timestamps when SO_TIMESTAMP is enabled. Two additional clock sources are: o nanosecond resolution realtime clock (equivalent of CLOCK_REALTIME); o nanosecond resolution monotonic clock (equivalent of CLOCK_MONOTONIC). In addition to this, this option provides unified interface to get bintime (equivalent of using SO_BINTIME), except it also supported with IPv6 where SO_BINTIME has never been supported. The long term plan is to depreciate SO_BINTIME and move everything to using SO_TS_CLOCK. Idea for this enhancement has been briefly discussed on the Net session during dev summit in Ottawa last June and the general input was positive. This change is believed to benefit network benchmarks/profiling as well as other scenarios where precise time of arrival measurement is necessary. There are two regression test cases as part of this commit: one extends unix domain test code (unix_cmsg) to test new SCM_XXX types and another one implementis totally new test case which exchanges UDP packets between two processes using both conventional methods (i.e. calling clock_gettime(2) before recv(2) and after send(2)), as well as using setsockopt()+recv() in receive path. The resulting delays are checked for sanity for all supported clock types. Reviewed by: adrian, gnn Differential Revision: https://reviews.freebsd.org/D9171
2017-01-16 17:46:38 +00:00
case SCM_REALTIME:
*controlp = sbcreatecontrol(NULL, sizeof(*ts),
SCM_REALTIME, SOL_SOCKET);
if (*controlp == NULL) {
error = ENOBUFS;
goto out;
}
ts = (struct timespec *)
CMSG_DATA(mtod(*controlp, struct cmsghdr *));
nanotime(ts);
break;
case SCM_MONOTONIC:
*controlp = sbcreatecontrol(NULL, sizeof(*ts),
SCM_MONOTONIC, SOL_SOCKET);
if (*controlp == NULL) {
error = ENOBUFS;
goto out;
}
ts = (struct timespec *)
CMSG_DATA(mtod(*controlp, struct cmsghdr *));
nanouptime(ts);
break;
Hopefully improve control message passing over Unix domain sockets. 1) Allow the sending of more than one control message at a time over a unix domain socket. This should cover the PR 29499. 2) This requires that unp_{ex,in}ternalize and unp_scan understand mbufs with more than one control message at a time. 3) Internalize and externalize used to work on the mbuf in-place. This made life quite complicated and the code for sizeof(int) < sizeof(file *) could end up doing the wrong thing. The patch always create a new mbuf/cluster now. This resulted in the change of the prototype for the domain externalise function. 4) You can now send SCM_TIMESTAMP messages. 5) Always use CMSG_DATA(cm) to determine the start where the data in unp_{ex,in}ternalize. It was using ((struct cmsghdr *)cm + 1) in some places, which gives the wrong alignment on the alpha. (NetBSD made this fix some time ago). This results in an ABI change for discriptor passing and creds passing on the alpha. (Probably on the IA64 and Spare ports too). 6) Fix userland programs to use CMSG_* macros too. 7) Be more careful about freeing mbufs containing (file *)s. This is made possible by the prototype change of externalise. PR: 29499 MFC after: 6 weeks
2001-10-04 13:11:48 +00:00
default:
error = EINVAL;
goto out;
Enable SCM_RIGHTS on alpha. Allocate necessary buffer as conversion between int and struct file *. Approved by: jkh Submitted by: brian Reviewed by: bde, brian, peter
2000-03-09 15:15:27 +00:00
}
Hopefully improve control message passing over Unix domain sockets. 1) Allow the sending of more than one control message at a time over a unix domain socket. This should cover the PR 29499. 2) This requires that unp_{ex,in}ternalize and unp_scan understand mbufs with more than one control message at a time. 3) Internalize and externalize used to work on the mbuf in-place. This made life quite complicated and the code for sizeof(int) < sizeof(file *) could end up doing the wrong thing. The patch always create a new mbuf/cluster now. This resulted in the change of the prototype for the domain externalise function. 4) You can now send SCM_TIMESTAMP messages. 5) Always use CMSG_DATA(cm) to determine the start where the data in unp_{ex,in}ternalize. It was using ((struct cmsghdr *)cm + 1) in some places, which gives the wrong alignment on the alpha. (NetBSD made this fix some time ago). This results in an ABI change for discriptor passing and creds passing on the alpha. (Probably on the IA64 and Spare ports too). 6) Fix userland programs to use CMSG_* macros too. 7) Be more careful about freeing mbufs containing (file *)s. This is made possible by the prototype change of externalise. PR: 29499 MFC after: 6 weeks
2001-10-04 13:11:48 +00:00
controlp = &(*controlp)->m_next;
if (CMSG_SPACE(datalen) < clen) {
clen -= CMSG_SPACE(datalen);
cm = (struct cmsghdr *)
((caddr_t)cm + CMSG_SPACE(datalen));
} else {
clen = 0;
cm = NULL;
Enable SCM_RIGHTS on alpha. Allocate necessary buffer as conversion between int and struct file *. Approved by: jkh Submitted by: brian Reviewed by: bde, brian, peter
2000-03-09 15:15:27 +00:00
}
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
}
Hopefully improve control message passing over Unix domain sockets. 1) Allow the sending of more than one control message at a time over a unix domain socket. This should cover the PR 29499. 2) This requires that unp_{ex,in}ternalize and unp_scan understand mbufs with more than one control message at a time. 3) Internalize and externalize used to work on the mbuf in-place. This made life quite complicated and the code for sizeof(int) < sizeof(file *) could end up doing the wrong thing. The patch always create a new mbuf/cluster now. This resulted in the change of the prototype for the domain externalise function. 4) You can now send SCM_TIMESTAMP messages. 5) Always use CMSG_DATA(cm) to determine the start where the data in unp_{ex,in}ternalize. It was using ((struct cmsghdr *)cm + 1) in some places, which gives the wrong alignment on the alpha. (NetBSD made this fix some time ago). This results in an ABI change for discriptor passing and creds passing on the alpha. (Probably on the IA64 and Spare ports too). 6) Fix userland programs to use CMSG_* macros too. 7) Be more careful about freeing mbufs containing (file *)s. This is made possible by the prototype change of externalise. PR: 29499 MFC after: 6 weeks
2001-10-04 13:11:48 +00:00
out:
m_freem(control);
return (error);
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
}
Break introductory comment into two paragraphs to separate material on the garbage collection complications from general discussion of UNIX domain sockets. Staticize unp_addsockcred(). Remove XXX comment regarding Giant and v_socket -- v_socket is protected by the global UNIX domain socket lock.
2007-02-20 10:50:02 +00:00
static struct mbuf *
Implement unix(4) socket options LOCAL_CREDS and LOCAL_CONNWAIT. - Add unp_addsockcred() (for LOCAL_CREDS). - Add an argument to unp_connect2() to differentiate between PRU_CONNECT and PRU_CONNECT2. (for LOCAL_CONNWAIT) Obtained from: NetBSD (with some changes)
2005-04-13 00:01:46 +00:00
unp_addsockcred(struct thread *td, struct mbuf *control)
{
o There are two methods to get a process credentials over the unix sockets: 1) A sender sends SCM_CREDS message to a reciever, struct cmsgcred; 2) A reciever sets LOCAL_CREDS socket option and gets sender credentials in control message, struct sockcred. Both methods use the same control message type SCM_CREDS with the same control message level SOL_SOCKET, so they are indistinguishable for the receiver. A difference in struct cmsgcred and struct sockcred layouts may lead to unwanted effects. Now for sockets with LOCAL_CREDS option remove all previous linked SCM_CREDS control messages and then add a control message with struct sockcred so the process specifically asked for the peer credentials by LOCAL_CREDS option always gets struct sockcred. PR: kern/90800 Submitted by: Andrey Simonenko Regres. tests: tools/regression/sockets/unix_cmsg/ MFC after: 1 month
2006-06-13 14:33:35 +00:00
struct mbuf *m, *n, *n_prev;
Implement unix(4) socket options LOCAL_CREDS and LOCAL_CONNWAIT. - Add unp_addsockcred() (for LOCAL_CREDS). - Add an argument to unp_connect2() to differentiate between PRU_CONNECT and PRU_CONNECT2. (for LOCAL_CONNWAIT) Obtained from: NetBSD (with some changes)
2005-04-13 00:01:46 +00:00
struct sockcred *sc;
o There are two methods to get a process credentials over the unix sockets: 1) A sender sends SCM_CREDS message to a reciever, struct cmsgcred; 2) A reciever sets LOCAL_CREDS socket option and gets sender credentials in control message, struct sockcred. Both methods use the same control message type SCM_CREDS with the same control message level SOL_SOCKET, so they are indistinguishable for the receiver. A difference in struct cmsgcred and struct sockcred layouts may lead to unwanted effects. Now for sockets with LOCAL_CREDS option remove all previous linked SCM_CREDS control messages and then add a control message with struct sockcred so the process specifically asked for the peer credentials by LOCAL_CREDS option always gets struct sockcred. PR: kern/90800 Submitted by: Andrey Simonenko Regres. tests: tools/regression/sockets/unix_cmsg/ MFC after: 1 month
2006-06-13 14:33:35 +00:00
const struct cmsghdr *cm;
Implement unix(4) socket options LOCAL_CREDS and LOCAL_CONNWAIT. - Add unp_addsockcred() (for LOCAL_CREDS). - Add an argument to unp_connect2() to differentiate between PRU_CONNECT and PRU_CONNECT2. (for LOCAL_CONNWAIT) Obtained from: NetBSD (with some changes)
2005-04-13 00:01:46 +00:00
int ngroups;
int i;
ngroups = MIN(td->td_ucred->cr_ngroups, CMGROUP_MAX);
m = sbcreatecontrol(NULL, SOCKCREDSIZE(ngroups), SCM_CREDS, SOL_SOCKET);
if (m == NULL)
return (control);
sc = (struct sockcred *) CMSG_DATA(mtod(m, struct cmsghdr *));
sc->sc_uid = td->td_ucred->cr_ruid;
sc->sc_euid = td->td_ucred->cr_uid;
sc->sc_gid = td->td_ucred->cr_rgid;
sc->sc_egid = td->td_ucred->cr_gid;
sc->sc_ngroups = ngroups;
for (i = 0; i < sc->sc_ngroups; i++)
sc->sc_groups[i] = td->td_ucred->cr_groups[i];
/*
Re-wrap and other minor formatting and punctuation fixes for UNIX domain socket comments.
2006-07-22 17:24:55 +00:00
* Unlink SCM_CREDS control messages (struct cmsgcred), since just
* created SCM_CREDS control message (struct sockcred) has another
* format.
Implement unix(4) socket options LOCAL_CREDS and LOCAL_CONNWAIT. - Add unp_addsockcred() (for LOCAL_CREDS). - Add an argument to unp_connect2() to differentiate between PRU_CONNECT and PRU_CONNECT2. (for LOCAL_CONNWAIT) Obtained from: NetBSD (with some changes)
2005-04-13 00:01:46 +00:00
*/
o There are two methods to get a process credentials over the unix sockets: 1) A sender sends SCM_CREDS message to a reciever, struct cmsgcred; 2) A reciever sets LOCAL_CREDS socket option and gets sender credentials in control message, struct sockcred. Both methods use the same control message type SCM_CREDS with the same control message level SOL_SOCKET, so they are indistinguishable for the receiver. A difference in struct cmsgcred and struct sockcred layouts may lead to unwanted effects. Now for sockets with LOCAL_CREDS option remove all previous linked SCM_CREDS control messages and then add a control message with struct sockcred so the process specifically asked for the peer credentials by LOCAL_CREDS option always gets struct sockcred. PR: kern/90800 Submitted by: Andrey Simonenko Regres. tests: tools/regression/sockets/unix_cmsg/ MFC after: 1 month
2006-06-13 14:33:35 +00:00
if (control != NULL)
for (n = control, n_prev = NULL; n != NULL;) {
cm = mtod(n, struct cmsghdr *);
if (cm->cmsg_level == SOL_SOCKET &&
cm->cmsg_type == SCM_CREDS) {
if (n_prev == NULL)
control = n->m_next;
else
n_prev->m_next = n->m_next;
n = m_free(n);
} else {
n_prev = n;
n = n->m_next;
}
}
/* Prepend it to the head. */
m->m_next = control;
return (m);
Implement unix(4) socket options LOCAL_CREDS and LOCAL_CONNWAIT. - Add unp_addsockcred() (for LOCAL_CREDS). - Add an argument to unp_connect2() to differentiate between PRU_CONNECT and PRU_CONNECT2. (for LOCAL_CONNWAIT) Obtained from: NetBSD (with some changes)
2005-04-13 00:01:46 +00:00
}
Remove explicit locking of struct file. - Introduce a finit() which is used to initailize the fields of struct file in such a way that the ops vector is only valid after the data, type, and flags are valid. - Protect f_flag and f_count with atomic operations. - Remove the global list of all files and associated accounting. - Rewrite the unp garbage collection such that it no longer requires the global list of all files and instead uses a list of all unp sockets. - Mark sockets in the accept queue so we don't incorrectly gc them. Tested by: kris, pho
2007-12-30 01:42:15 +00:00
static struct unpcb *
fptounp(struct file *fp)
{
struct socket *so;
if (fp->f_type != DTYPE_SOCKET)
return (NULL);
if ((so = fp->f_data) == NULL)
return (NULL);
if (so->so_proto->pr_domain != &localdomain)
return (NULL);
return sotounpcb(so);
}
static void
unp_discard(struct file *fp)
{
Reviewed by: jeff, rwatson MFC after: 1 week
2010-12-03 16:15:44 +00:00
struct unp_defer *dr;
if (unp_externalize_fp(fp)) {
dr = malloc(sizeof(*dr), M_TEMP, M_WAITOK);
dr->ud_fp = fp;
UNP_DEFERRED_LOCK();
SLIST_INSERT_HEAD(&unp_defers, dr, ud_link);
UNP_DEFERRED_UNLOCK();
atomic_add_int(&unp_defers_count, 1);
taskqueue_enqueue(taskqueue_thread, &unp_defer_task);
} else
(void) closef(fp, (struct thread *)NULL);
}
Remove explicit locking of struct file. - Introduce a finit() which is used to initailize the fields of struct file in such a way that the ops vector is only valid after the data, type, and flags are valid. - Protect f_flag and f_count with atomic operations. - Remove the global list of all files and associated accounting. - Rewrite the unp garbage collection such that it no longer requires the global list of all files and instead uses a list of all unp sockets. - Mark sockets in the accept queue so we don't incorrectly gc them. Tested by: kris, pho
2007-12-30 01:42:15 +00:00
Reviewed by: jeff, rwatson MFC after: 1 week
2010-12-03 16:15:44 +00:00
static void
unp_process_defers(void *arg __unused, int pending)
{
struct unp_defer *dr;
SLIST_HEAD(, unp_defer) drl;
int count;
SLIST_INIT(&drl);
for (;;) {
UNP_DEFERRED_LOCK();
if (SLIST_FIRST(&unp_defers) == NULL) {
UNP_DEFERRED_UNLOCK();
break;
}
SLIST_SWAP(&unp_defers, &drl, unp_defer);
UNP_DEFERRED_UNLOCK();
count = 0;
while ((dr = SLIST_FIRST(&drl)) != NULL) {
SLIST_REMOVE_HEAD(&drl, ud_link);
closef(dr->ud_fp, NULL);
free(dr, M_TEMP);
count++;
}
atomic_add_int(&unp_defers_count, -count);
}
Remove explicit locking of struct file. - Introduce a finit() which is used to initailize the fields of struct file in such a way that the ops vector is only valid after the data, type, and flags are valid. - Protect f_flag and f_count with atomic operations. - Remove the global list of all files and associated accounting. - Rewrite the unp garbage collection such that it no longer requires the global list of all files and instead uses a list of all unp sockets. - Mark sockets in the accept queue so we don't incorrectly gc them. Tested by: kris, pho
2007-12-30 01:42:15 +00:00
}
static void
unp_internalize_fp(struct file *fp)
{
struct unpcb *unp;
Decompose the global UNIX domain sockets rwlock into two different locks: a global list/counter/generation counter protected by a new mutex unp_list_lock, and a global linkage rwlock, unp_global_rwlock, which protects the connections between UNIX domain sockets. This eliminates conditional lock acquisition that was previously a property of the global lock being held over sonewconn() leading to a call to uipc_attach(), which also required the global lock, but couldn't rely on it as other paths existed to uipc_attach() that didn't hold it: now uipc_attach() uses only the list lock, which follows the linkage lock in the lock order. It may also reduce contention on the global lock for some workloads. Add global UNIX domain socket locks to hard-coded witness lock order. MFC after: 1 week Discussed with: kris
2009-03-08 21:48:29 +00:00
UNP_LINK_WLOCK();
Remove explicit locking of struct file. - Introduce a finit() which is used to initailize the fields of struct file in such a way that the ops vector is only valid after the data, type, and flags are valid. - Protect f_flag and f_count with atomic operations. - Remove the global list of all files and associated accounting. - Rewrite the unp garbage collection such that it no longer requires the global list of all files and instead uses a list of all unp sockets. - Mark sockets in the accept queue so we don't incorrectly gc them. Tested by: kris, pho
2007-12-30 01:42:15 +00:00
if ((unp = fptounp(fp)) != NULL) {
unp->unp_file = fp;
unp->unp_msgcount++;
}
- Place the fhold() in unp_internalize_fp to be more consistent with refs. - Clear all of the gc flags before doing a run. Stale flags were causing us to skip some descriptors. - If a unp socket has been marked REF in a gc pass it can't be dead. Found by: rwatson's test tool.
2008-01-01 01:46:42 +00:00
fhold(fp);
Remove explicit locking of struct file. - Introduce a finit() which is used to initailize the fields of struct file in such a way that the ops vector is only valid after the data, type, and flags are valid. - Protect f_flag and f_count with atomic operations. - Remove the global list of all files and associated accounting. - Rewrite the unp garbage collection such that it no longer requires the global list of all files and instead uses a list of all unp sockets. - Mark sockets in the accept queue so we don't incorrectly gc them. Tested by: kris, pho
2007-12-30 01:42:15 +00:00
unp_rights++;
Decompose the global UNIX domain sockets rwlock into two different locks: a global list/counter/generation counter protected by a new mutex unp_list_lock, and a global linkage rwlock, unp_global_rwlock, which protects the connections between UNIX domain sockets. This eliminates conditional lock acquisition that was previously a property of the global lock being held over sonewconn() leading to a call to uipc_attach(), which also required the global lock, but couldn't rely on it as other paths existed to uipc_attach() that didn't hold it: now uipc_attach() uses only the list lock, which follows the linkage lock in the lock order. It may also reduce contention on the global lock for some workloads. Add global UNIX domain socket locks to hard-coded witness lock order. MFC after: 1 week Discussed with: kris
2009-03-08 21:48:29 +00:00
UNP_LINK_WUNLOCK();
Remove explicit locking of struct file. - Introduce a finit() which is used to initailize the fields of struct file in such a way that the ops vector is only valid after the data, type, and flags are valid. - Protect f_flag and f_count with atomic operations. - Remove the global list of all files and associated accounting. - Rewrite the unp garbage collection such that it no longer requires the global list of all files and instead uses a list of all unp sockets. - Mark sockets in the accept queue so we don't incorrectly gc them. Tested by: kris, pho
2007-12-30 01:42:15 +00:00
}
Reviewed by: jeff, rwatson MFC after: 1 week
2010-12-03 16:15:44 +00:00
static int
Remove explicit locking of struct file. - Introduce a finit() which is used to initailize the fields of struct file in such a way that the ops vector is only valid after the data, type, and flags are valid. - Protect f_flag and f_count with atomic operations. - Remove the global list of all files and associated accounting. - Rewrite the unp garbage collection such that it no longer requires the global list of all files and instead uses a list of all unp sockets. - Mark sockets in the accept queue so we don't incorrectly gc them. Tested by: kris, pho
2007-12-30 01:42:15 +00:00
unp_externalize_fp(struct file *fp)
{
struct unpcb *unp;
Reviewed by: jeff, rwatson MFC after: 1 week
2010-12-03 16:15:44 +00:00
int ret;
Remove explicit locking of struct file. - Introduce a finit() which is used to initailize the fields of struct file in such a way that the ops vector is only valid after the data, type, and flags are valid. - Protect f_flag and f_count with atomic operations. - Remove the global list of all files and associated accounting. - Rewrite the unp garbage collection such that it no longer requires the global list of all files and instead uses a list of all unp sockets. - Mark sockets in the accept queue so we don't incorrectly gc them. Tested by: kris, pho
2007-12-30 01:42:15 +00:00
Decompose the global UNIX domain sockets rwlock into two different locks: a global list/counter/generation counter protected by a new mutex unp_list_lock, and a global linkage rwlock, unp_global_rwlock, which protects the connections between UNIX domain sockets. This eliminates conditional lock acquisition that was previously a property of the global lock being held over sonewconn() leading to a call to uipc_attach(), which also required the global lock, but couldn't rely on it as other paths existed to uipc_attach() that didn't hold it: now uipc_attach() uses only the list lock, which follows the linkage lock in the lock order. It may also reduce contention on the global lock for some workloads. Add global UNIX domain socket locks to hard-coded witness lock order. MFC after: 1 week Discussed with: kris
2009-03-08 21:48:29 +00:00
UNP_LINK_WLOCK();
Reviewed by: jeff, rwatson MFC after: 1 week
2010-12-03 16:15:44 +00:00
if ((unp = fptounp(fp)) != NULL) {
Remove explicit locking of struct file. - Introduce a finit() which is used to initailize the fields of struct file in such a way that the ops vector is only valid after the data, type, and flags are valid. - Protect f_flag and f_count with atomic operations. - Remove the global list of all files and associated accounting. - Rewrite the unp garbage collection such that it no longer requires the global list of all files and instead uses a list of all unp sockets. - Mark sockets in the accept queue so we don't incorrectly gc them. Tested by: kris, pho
2007-12-30 01:42:15 +00:00
unp->unp_msgcount--;
Reviewed by: jeff, rwatson MFC after: 1 week
2010-12-03 16:15:44 +00:00
ret = 1;
} else
ret = 0;
Remove explicit locking of struct file. - Introduce a finit() which is used to initailize the fields of struct file in such a way that the ops vector is only valid after the data, type, and flags are valid. - Protect f_flag and f_count with atomic operations. - Remove the global list of all files and associated accounting. - Rewrite the unp garbage collection such that it no longer requires the global list of all files and instead uses a list of all unp sockets. - Mark sockets in the accept queue so we don't incorrectly gc them. Tested by: kris, pho
2007-12-30 01:42:15 +00:00
unp_rights--;
Decompose the global UNIX domain sockets rwlock into two different locks: a global list/counter/generation counter protected by a new mutex unp_list_lock, and a global linkage rwlock, unp_global_rwlock, which protects the connections between UNIX domain sockets. This eliminates conditional lock acquisition that was previously a property of the global lock being held over sonewconn() leading to a call to uipc_attach(), which also required the global lock, but couldn't rely on it as other paths existed to uipc_attach() that didn't hold it: now uipc_attach() uses only the list lock, which follows the linkage lock in the lock order. It may also reduce contention on the global lock for some workloads. Add global UNIX domain socket locks to hard-coded witness lock order. MFC after: 1 week Discussed with: kris
2009-03-08 21:48:29 +00:00
UNP_LINK_WUNLOCK();
Reviewed by: jeff, rwatson MFC after: 1 week
2010-12-03 16:15:44 +00:00
return (ret);
Remove explicit locking of struct file. - Introduce a finit() which is used to initailize the fields of struct file in such a way that the ops vector is only valid after the data, type, and flags are valid. - Protect f_flag and f_count with atomic operations. - Remove the global list of all files and associated accounting. - Rewrite the unp garbage collection such that it no longer requires the global list of all files and instead uses a list of all unp sockets. - Mark sockets in the accept queue so we don't incorrectly gc them. Tested by: kris, pho
2007-12-30 01:42:15 +00:00
}
Don't hold the UNIX domain socket subsystem lock over the body of the UNIX domain socket garbage collection implementation, as that risks holding the mutex over potentially sleeping operations (as well as introducing some nasty lock order issues, etc). unp_gc() will hold the lock long enough to do necessary deferal checks and set that it's running, but then release it until it needs to reset the gc state. RELENG_5 candidate. Discussed with: alfred
2004-08-25 21:24:36 +00:00
/*
Correct a number of serious and closely related bugs in the UNIX domain socket file descriptor garbage collection code, which is intended to detect and clear cycles of orphaned file descriptors that are "in-flight" in a socket when that socket is closed before they are received. The algorithm present was both run at poor times (resulting in recursion and reentrance), and also buggy in the presence of parallelism. In order to fix these problems, make the following changes: - When there are in-flight sockets and a UNIX domain socket is destroyed, asynchronously schedule the garbage collector, rather than running it synchronously in the current context. This avoids lock order issues when the garbage collection code reenters the UNIX domain socket code, avoiding lock order reversals, deadlocks, etc. Run the code asynchronously in a task queue. - In the garbage collector, when skipping file descriptors that have entered a closing state (i.e., have f_count == 0), re-test the FDEFER flag, and decrement unp_defer. As file descriptors can now transition to a closed state, while the garbage collector is running, it is no longer the case that unp_defer will remain an accurate count of deferred sockets in the mark portion of the GC algorithm. Otherwise, the garbage collector will loop waiting waiting for unp_defer to reach zero, which it will never do as it is skipping file descriptors that were marked in an earlier pass, but now closed. - Acquire the UNIX domain socket subsystem lock in unp_discard() when modifying the unp_rights counter, or a read/write race is risked with other threads also manipulating the counter. While here: - Remove #if 0'd code regarding acquiring the socket buffer sleep lock in the garbage collector, this is not required as we are able to use the socket buffer receive lock to protect scanning the receive buffer for in-flight file descriptors on the socket buffer. - Annotate that the description of the garbage collector implementation is increasingly inaccurate and needs to be updated. - Add counters of the number of deferred garbage collections and recycled file descriptors. This will be removed and is here temporarily for debugging purposes. With these changes in place, the unp_passfd regression test now appears to be passed consistently on UP and SMP systems for extended runs, whereas before it hung quickly or panicked, depending on which bug was triggered. Reported by: Philip Kizer <pckizer at nostrum dot com> MFC after: 2 weeks
2005-11-10 16:06:04 +00:00
* unp_defer indicates whether additional work has been defered for a future
* pass through unp_gc(). It is thread local and does not require explicit
* synchronization.
Don't hold the UNIX domain socket subsystem lock over the body of the UNIX domain socket garbage collection implementation, as that risks holding the mutex over potentially sleeping operations (as well as introducing some nasty lock order issues, etc). unp_gc() will hold the lock long enough to do necessary deferal checks and set that it's running, but then release it until it needs to reset the gc state. RELENG_5 candidate. Discussed with: alfred
2004-08-25 21:24:36 +00:00
*/
Remove explicit locking of struct file. - Introduce a finit() which is used to initailize the fields of struct file in such a way that the ops vector is only valid after the data, type, and flags are valid. - Protect f_flag and f_count with atomic operations. - Remove the global list of all files and associated accounting. - Rewrite the unp garbage collection such that it no longer requires the global list of all files and instead uses a list of all unp sockets. - Mark sockets in the accept queue so we don't incorrectly gc them. Tested by: kris, pho
2007-12-30 01:42:15 +00:00
static int unp_marked;
static int unp_unreachable;
Correct a number of serious and closely related bugs in the UNIX domain socket file descriptor garbage collection code, which is intended to detect and clear cycles of orphaned file descriptors that are "in-flight" in a socket when that socket is closed before they are received. The algorithm present was both run at poor times (resulting in recursion and reentrance), and also buggy in the presence of parallelism. In order to fix these problems, make the following changes: - When there are in-flight sockets and a UNIX domain socket is destroyed, asynchronously schedule the garbage collector, rather than running it synchronously in the current context. This avoids lock order issues when the garbage collection code reenters the UNIX domain socket code, avoiding lock order reversals, deadlocks, etc. Run the code asynchronously in a task queue. - In the garbage collector, when skipping file descriptors that have entered a closing state (i.e., have f_count == 0), re-test the FDEFER flag, and decrement unp_defer. As file descriptors can now transition to a closed state, while the garbage collector is running, it is no longer the case that unp_defer will remain an accurate count of deferred sockets in the mark portion of the GC algorithm. Otherwise, the garbage collector will loop waiting waiting for unp_defer to reach zero, which it will never do as it is skipping file descriptors that were marked in an earlier pass, but now closed. - Acquire the UNIX domain socket subsystem lock in unp_discard() when modifying the unp_rights counter, or a read/write race is risked with other threads also manipulating the counter. While here: - Remove #if 0'd code regarding acquiring the socket buffer sleep lock in the garbage collector, this is not required as we are able to use the socket buffer receive lock to protect scanning the receive buffer for in-flight file descriptors on the socket buffer. - Annotate that the description of the garbage collector implementation is increasingly inaccurate and needs to be updated. - Add counters of the number of deferred garbage collections and recycled file descriptors. This will be removed and is here temporarily for debugging purposes. With these changes in place, the unp_passfd regression test now appears to be passed consistently on UP and SMP systems for extended runs, whereas before it hung quickly or panicked, depending on which bug was triggered. Reported by: Philip Kizer <pckizer at nostrum dot com> MFC after: 2 weeks
2005-11-10 16:06:04 +00:00
Remove explicit locking of struct file. - Introduce a finit() which is used to initailize the fields of struct file in such a way that the ops vector is only valid after the data, type, and flags are valid. - Protect f_flag and f_count with atomic operations. - Remove the global list of all files and associated accounting. - Rewrite the unp garbage collection such that it no longer requires the global list of all files and instead uses a list of all unp sockets. - Mark sockets in the accept queue so we don't incorrectly gc them. Tested by: kris, pho
2007-12-30 01:42:15 +00:00
static void
Fix memory leak when one process send descriptor over UNIX domain socket, but the other process exited before receiving it.
2013-03-11 22:59:07 +00:00
unp_accessable(struct filedescent **fdep, int fdcount)
Remove explicit locking of struct file. - Introduce a finit() which is used to initailize the fields of struct file in such a way that the ops vector is only valid after the data, type, and flags are valid. - Protect f_flag and f_count with atomic operations. - Remove the global list of all files and associated accounting. - Rewrite the unp garbage collection such that it no longer requires the global list of all files and instead uses a list of all unp sockets. - Mark sockets in the accept queue so we don't incorrectly gc them. Tested by: kris, pho
2007-12-30 01:42:15 +00:00
{
struct unpcb *unp;
Fix memory leak when one process send descriptor over UNIX domain socket, but the other process exited before receiving it.
2013-03-11 22:59:07 +00:00
struct file *fp;
int i;
Remove explicit locking of struct file. - Introduce a finit() which is used to initailize the fields of struct file in such a way that the ops vector is only valid after the data, type, and flags are valid. - Protect f_flag and f_count with atomic operations. - Remove the global list of all files and associated accounting. - Rewrite the unp garbage collection such that it no longer requires the global list of all files and instead uses a list of all unp sockets. - Mark sockets in the accept queue so we don't incorrectly gc them. Tested by: kris, pho
2007-12-30 01:42:15 +00:00
Fix memory leak when one process send descriptor over UNIX domain socket, but the other process exited before receiving it.
2013-03-11 22:59:07 +00:00
for (i = 0; i < fdcount; i++) {
fp = fdep[i]->fde_file;
if ((unp = fptounp(fp)) == NULL)
continue;
if (unp->unp_gcflag & UNPGC_REF)
continue;
unp->unp_gcflag &= ~UNPGC_DEAD;
unp->unp_gcflag |= UNPGC_REF;
unp_marked++;
}
Remove explicit locking of struct file. - Introduce a finit() which is used to initailize the fields of struct file in such a way that the ops vector is only valid after the data, type, and flags are valid. - Protect f_flag and f_count with atomic operations. - Remove the global list of all files and associated accounting. - Rewrite the unp garbage collection such that it no longer requires the global list of all files and instead uses a list of all unp sockets. - Mark sockets in the accept queue so we don't incorrectly gc them. Tested by: kris, pho
2007-12-30 01:42:15 +00:00
}
static void
unp_gc_process(struct unpcb *unp)
{
Listening sockets improvements. o Separate fields of struct socket that belong to listening from fields that belong to normal dataflow, and unionize them. This shrinks the structure a bit. - Take out selinfo's from the socket buffers into the socket. The first reason is to support braindamaged scenario when a socket is added to kevent(2) and then listen(2) is cast on it. The second reason is that there is future plan to make socket buffers pluggable, so that for a dataflow socket a socket buffer can be changed, and in this case we also want to keep same selinfos through the lifetime of a socket. - Remove struct struct so_accf. Since now listening stuff no longer affects struct socket size, just move its fields into listening part of the union. - Provide sol_upcall field and enforce that so_upcall_set() may be called only on a dataflow socket, which has buffers, and for listening sockets provide solisten_upcall_set(). o Remove ACCEPT_LOCK() global. - Add a mutex to socket, to be used instead of socket buffer lock to lock fields of struct socket that don't belong to a socket buffer. - Allow to acquire two socket locks, but the first one must belong to a listening socket. - Make soref()/sorele() to use atomic(9). This allows in some situations to do soref() without owning socket lock. There is place for improvement here, it is possible to make sorele() also to lock optionally. - Most protocols aren't touched by this change, except UNIX local sockets. See below for more information. o Reduce copy-and-paste in kernel modules that accept connections from listening sockets: provide function solisten_dequeue(), and use it in the following modules: ctl(4), iscsi(4), ng_btsocket(4), ng_ksocket(4), infiniband, rpc. o UNIX local sockets. - Removal of ACCEPT_LOCK() global uncovered several races in the UNIX local sockets. Most races exist around spawning a new socket, when we are connecting to a local listening socket. To cover them, we need to hold locks on both PCBs when spawning a third one. This means holding them across sonewconn(). This creates a LOR between pcb locks and unp_list_lock. - To fix the new LOR, abandon the global unp_list_lock in favor of global unp_link_lock. Indeed, separating these two locks didn't provide us any extra parralelism in the UNIX sockets. - Now call into uipc_attach() may happen with unp_link_lock hold if, we are accepting, or without unp_link_lock in case if we are just creating a socket. - Another problem in UNIX sockets is that uipc_close() basicly did nothing for a listening socket. The vnode remained opened for connections. This is fixed by removing vnode in uipc_close(). Maybe the right way would be to do it for all sockets (not only listening), simply move the vnode teardown from uipc_detach() to uipc_close()? Sponsored by: Netflix Differential Revision: https://reviews.freebsd.org/D9770
2017-06-08 21:30:34 +00:00
struct socket *so, *soa;
Remove explicit locking of struct file. - Introduce a finit() which is used to initailize the fields of struct file in such a way that the ops vector is only valid after the data, type, and flags are valid. - Protect f_flag and f_count with atomic operations. - Remove the global list of all files and associated accounting. - Rewrite the unp garbage collection such that it no longer requires the global list of all files and instead uses a list of all unp sockets. - Mark sockets in the accept queue so we don't incorrectly gc them. Tested by: kris, pho
2007-12-30 01:42:15 +00:00
struct file *fp;
/* Already processed. */
if (unp->unp_gcflag & UNPGC_SCANNED)
return;
fp = unp->unp_file;
Remove or update several stale comments. A bit of whitespace/style cleanup. Update copyright. MFC after: 3 days (applicable changes)
2008-10-03 09:01:55 +00:00
Remove explicit locking of struct file. - Introduce a finit() which is used to initailize the fields of struct file in such a way that the ops vector is only valid after the data, type, and flags are valid. - Protect f_flag and f_count with atomic operations. - Remove the global list of all files and associated accounting. - Rewrite the unp garbage collection such that it no longer requires the global list of all files and instead uses a list of all unp sockets. - Mark sockets in the accept queue so we don't incorrectly gc them. Tested by: kris, pho
2007-12-30 01:42:15 +00:00
/*
* Check for a socket potentially in a cycle. It must be in a
* queue as indicated by msgcount, and this must equal the file
* reference count. Note that when msgcount is 0 the file is NULL.
*/
- Place the fhold() in unp_internalize_fp to be more consistent with refs. - Clear all of the gc flags before doing a run. Stale flags were causing us to skip some descriptors. - If a unp socket has been marked REF in a gc pass it can't be dead. Found by: rwatson's test tool.
2008-01-01 01:46:42 +00:00
if ((unp->unp_gcflag & UNPGC_REF) == 0 && fp &&
unp->unp_msgcount != 0 && fp->f_count == unp->unp_msgcount) {
Remove explicit locking of struct file. - Introduce a finit() which is used to initailize the fields of struct file in such a way that the ops vector is only valid after the data, type, and flags are valid. - Protect f_flag and f_count with atomic operations. - Remove the global list of all files and associated accounting. - Rewrite the unp garbage collection such that it no longer requires the global list of all files and instead uses a list of all unp sockets. - Mark sockets in the accept queue so we don't incorrectly gc them. Tested by: kris, pho
2007-12-30 01:42:15 +00:00
unp->unp_gcflag |= UNPGC_DEAD;
unp_unreachable++;
return;
}
Remove or update several stale comments. A bit of whitespace/style cleanup. Update copyright. MFC after: 3 days (applicable changes)
2008-10-03 09:01:55 +00:00
Remove explicit locking of struct file. - Introduce a finit() which is used to initailize the fields of struct file in such a way that the ops vector is only valid after the data, type, and flags are valid. - Protect f_flag and f_count with atomic operations. - Remove the global list of all files and associated accounting. - Rewrite the unp garbage collection such that it no longer requires the global list of all files and instead uses a list of all unp sockets. - Mark sockets in the accept queue so we don't incorrectly gc them. Tested by: kris, pho
2007-12-30 01:42:15 +00:00
so = unp->unp_socket;
Listening sockets improvements. o Separate fields of struct socket that belong to listening from fields that belong to normal dataflow, and unionize them. This shrinks the structure a bit. - Take out selinfo's from the socket buffers into the socket. The first reason is to support braindamaged scenario when a socket is added to kevent(2) and then listen(2) is cast on it. The second reason is that there is future plan to make socket buffers pluggable, so that for a dataflow socket a socket buffer can be changed, and in this case we also want to keep same selinfos through the lifetime of a socket. - Remove struct struct so_accf. Since now listening stuff no longer affects struct socket size, just move its fields into listening part of the union. - Provide sol_upcall field and enforce that so_upcall_set() may be called only on a dataflow socket, which has buffers, and for listening sockets provide solisten_upcall_set(). o Remove ACCEPT_LOCK() global. - Add a mutex to socket, to be used instead of socket buffer lock to lock fields of struct socket that don't belong to a socket buffer. - Allow to acquire two socket locks, but the first one must belong to a listening socket. - Make soref()/sorele() to use atomic(9). This allows in some situations to do soref() without owning socket lock. There is place for improvement here, it is possible to make sorele() also to lock optionally. - Most protocols aren't touched by this change, except UNIX local sockets. See below for more information. o Reduce copy-and-paste in kernel modules that accept connections from listening sockets: provide function solisten_dequeue(), and use it in the following modules: ctl(4), iscsi(4), ng_btsocket(4), ng_ksocket(4), infiniband, rpc. o UNIX local sockets. - Removal of ACCEPT_LOCK() global uncovered several races in the UNIX local sockets. Most races exist around spawning a new socket, when we are connecting to a local listening socket. To cover them, we need to hold locks on both PCBs when spawning a third one. This means holding them across sonewconn(). This creates a LOR between pcb locks and unp_list_lock. - To fix the new LOR, abandon the global unp_list_lock in favor of global unp_link_lock. Indeed, separating these two locks didn't provide us any extra parralelism in the UNIX sockets. - Now call into uipc_attach() may happen with unp_link_lock hold if, we are accepting, or without unp_link_lock in case if we are just creating a socket. - Another problem in UNIX sockets is that uipc_close() basicly did nothing for a listening socket. The vnode remained opened for connections. This is fixed by removing vnode in uipc_close(). Maybe the right way would be to do it for all sockets (not only listening), simply move the vnode teardown from uipc_detach() to uipc_close()? Sponsored by: Netflix Differential Revision: https://reviews.freebsd.org/D9770
2017-06-08 21:30:34 +00:00
SOCK_LOCK(so);
if (SOLISTENING(so)) {
/*
* Mark all sockets in our accept queue.
*/
TAILQ_FOREACH(soa, &so->sol_comp, so_list) {
if (sotounpcb(soa)->unp_gcflag & UNPGC_IGNORE_RIGHTS)
continue;
SOCKBUF_LOCK(&soa->so_rcv);
unp_scan(soa->so_rcv.sb_mb, unp_accessable);
SOCKBUF_UNLOCK(&soa->so_rcv);
}
} else {
/*
* Mark all sockets we reference with RIGHTS.
*/
if ((unp->unp_gcflag & UNPGC_IGNORE_RIGHTS) == 0) {
SOCKBUF_LOCK(&so->so_rcv);
unp_scan(so->so_rcv.sb_mb, unp_accessable);
SOCKBUF_UNLOCK(&so->so_rcv);
}
Remove explicit locking of struct file. - Introduce a finit() which is used to initailize the fields of struct file in such a way that the ops vector is only valid after the data, type, and flags are valid. - Protect f_flag and f_count with atomic operations. - Remove the global list of all files and associated accounting. - Rewrite the unp garbage collection such that it no longer requires the global list of all files and instead uses a list of all unp sockets. - Mark sockets in the accept queue so we don't incorrectly gc them. Tested by: kris, pho
2007-12-30 01:42:15 +00:00
}
Listening sockets improvements. o Separate fields of struct socket that belong to listening from fields that belong to normal dataflow, and unionize them. This shrinks the structure a bit. - Take out selinfo's from the socket buffers into the socket. The first reason is to support braindamaged scenario when a socket is added to kevent(2) and then listen(2) is cast on it. The second reason is that there is future plan to make socket buffers pluggable, so that for a dataflow socket a socket buffer can be changed, and in this case we also want to keep same selinfos through the lifetime of a socket. - Remove struct struct so_accf. Since now listening stuff no longer affects struct socket size, just move its fields into listening part of the union. - Provide sol_upcall field and enforce that so_upcall_set() may be called only on a dataflow socket, which has buffers, and for listening sockets provide solisten_upcall_set(). o Remove ACCEPT_LOCK() global. - Add a mutex to socket, to be used instead of socket buffer lock to lock fields of struct socket that don't belong to a socket buffer. - Allow to acquire two socket locks, but the first one must belong to a listening socket. - Make soref()/sorele() to use atomic(9). This allows in some situations to do soref() without owning socket lock. There is place for improvement here, it is possible to make sorele() also to lock optionally. - Most protocols aren't touched by this change, except UNIX local sockets. See below for more information. o Reduce copy-and-paste in kernel modules that accept connections from listening sockets: provide function solisten_dequeue(), and use it in the following modules: ctl(4), iscsi(4), ng_btsocket(4), ng_ksocket(4), infiniband, rpc. o UNIX local sockets. - Removal of ACCEPT_LOCK() global uncovered several races in the UNIX local sockets. Most races exist around spawning a new socket, when we are connecting to a local listening socket. To cover them, we need to hold locks on both PCBs when spawning a third one. This means holding them across sonewconn(). This creates a LOR between pcb locks and unp_list_lock. - To fix the new LOR, abandon the global unp_list_lock in favor of global unp_link_lock. Indeed, separating these two locks didn't provide us any extra parralelism in the UNIX sockets. - Now call into uipc_attach() may happen with unp_link_lock hold if, we are accepting, or without unp_link_lock in case if we are just creating a socket. - Another problem in UNIX sockets is that uipc_close() basicly did nothing for a listening socket. The vnode remained opened for connections. This is fixed by removing vnode in uipc_close(). Maybe the right way would be to do it for all sockets (not only listening), simply move the vnode teardown from uipc_detach() to uipc_close()? Sponsored by: Netflix Differential Revision: https://reviews.freebsd.org/D9770
2017-06-08 21:30:34 +00:00
SOCK_UNLOCK(so);
Remove explicit locking of struct file. - Introduce a finit() which is used to initailize the fields of struct file in such a way that the ops vector is only valid after the data, type, and flags are valid. - Protect f_flag and f_count with atomic operations. - Remove the global list of all files and associated accounting. - Rewrite the unp garbage collection such that it no longer requires the global list of all files and instead uses a list of all unp sockets. - Mark sockets in the accept queue so we don't incorrectly gc them. Tested by: kris, pho
2007-12-30 01:42:15 +00:00
unp->unp_gcflag |= UNPGC_SCANNED;
}
Correct a number of serious and closely related bugs in the UNIX domain socket file descriptor garbage collection code, which is intended to detect and clear cycles of orphaned file descriptors that are "in-flight" in a socket when that socket is closed before they are received. The algorithm present was both run at poor times (resulting in recursion and reentrance), and also buggy in the presence of parallelism. In order to fix these problems, make the following changes: - When there are in-flight sockets and a UNIX domain socket is destroyed, asynchronously schedule the garbage collector, rather than running it synchronously in the current context. This avoids lock order issues when the garbage collection code reenters the UNIX domain socket code, avoiding lock order reversals, deadlocks, etc. Run the code asynchronously in a task queue. - In the garbage collector, when skipping file descriptors that have entered a closing state (i.e., have f_count == 0), re-test the FDEFER flag, and decrement unp_defer. As file descriptors can now transition to a closed state, while the garbage collector is running, it is no longer the case that unp_defer will remain an accurate count of deferred sockets in the mark portion of the GC algorithm. Otherwise, the garbage collector will loop waiting waiting for unp_defer to reach zero, which it will never do as it is skipping file descriptors that were marked in an earlier pass, but now closed. - Acquire the UNIX domain socket subsystem lock in unp_discard() when modifying the unp_rights counter, or a read/write race is risked with other threads also manipulating the counter. While here: - Remove #if 0'd code regarding acquiring the socket buffer sleep lock in the garbage collector, this is not required as we are able to use the socket buffer receive lock to protect scanning the receive buffer for in-flight file descriptors on the socket buffer. - Annotate that the description of the garbage collector implementation is increasingly inaccurate and needs to be updated. - Add counters of the number of deferred garbage collections and recycled file descriptors. This will be removed and is here temporarily for debugging purposes. With these changes in place, the unp_passfd regression test now appears to be passed consistently on UP and SMP systems for extended runs, whereas before it hung quickly or panicked, depending on which bug was triggered. Reported by: Philip Kizer <pckizer at nostrum dot com> MFC after: 2 weeks
2005-11-10 16:06:04 +00:00
static int unp_recycled;
Fill in a few sysctl descriptions. Approved by: rwatson
2008-07-26 00:55:35 +00:00
SYSCTL_INT(_net_local, OID_AUTO, recycled, CTLFLAG_RD, &unp_recycled, 0,
"Number of unreachable sockets claimed by the garbage collector.");
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
Remove explicit locking of struct file. - Introduce a finit() which is used to initailize the fields of struct file in such a way that the ops vector is only valid after the data, type, and flags are valid. - Protect f_flag and f_count with atomic operations. - Remove the global list of all files and associated accounting. - Rewrite the unp garbage collection such that it no longer requires the global list of all files and instead uses a list of all unp sockets. - Mark sockets in the accept queue so we don't incorrectly gc them. Tested by: kris, pho
2007-12-30 01:42:15 +00:00
static int unp_taskcount;
Fill in a few sysctl descriptions. Approved by: rwatson
2008-07-26 00:55:35 +00:00
SYSCTL_INT(_net_local, OID_AUTO, taskcount, CTLFLAG_RD, &unp_taskcount, 0,
"Number of times the garbage collector has run.");
Remove explicit locking of struct file. - Introduce a finit() which is used to initailize the fields of struct file in such a way that the ops vector is only valid after the data, type, and flags are valid. - Protect f_flag and f_count with atomic operations. - Remove the global list of all files and associated accounting. - Rewrite the unp garbage collection such that it no longer requires the global list of all files and instead uses a list of all unp sockets. - Mark sockets in the accept queue so we don't incorrectly gc them. Tested by: kris, pho
2007-12-30 01:42:15 +00:00
Another mega commit to staticize things.
1995-12-14 09:55:16 +00:00
static void
Correct a number of serious and closely related bugs in the UNIX domain socket file descriptor garbage collection code, which is intended to detect and clear cycles of orphaned file descriptors that are "in-flight" in a socket when that socket is closed before they are received. The algorithm present was both run at poor times (resulting in recursion and reentrance), and also buggy in the presence of parallelism. In order to fix these problems, make the following changes: - When there are in-flight sockets and a UNIX domain socket is destroyed, asynchronously schedule the garbage collector, rather than running it synchronously in the current context. This avoids lock order issues when the garbage collection code reenters the UNIX domain socket code, avoiding lock order reversals, deadlocks, etc. Run the code asynchronously in a task queue. - In the garbage collector, when skipping file descriptors that have entered a closing state (i.e., have f_count == 0), re-test the FDEFER flag, and decrement unp_defer. As file descriptors can now transition to a closed state, while the garbage collector is running, it is no longer the case that unp_defer will remain an accurate count of deferred sockets in the mark portion of the GC algorithm. Otherwise, the garbage collector will loop waiting waiting for unp_defer to reach zero, which it will never do as it is skipping file descriptors that were marked in an earlier pass, but now closed. - Acquire the UNIX domain socket subsystem lock in unp_discard() when modifying the unp_rights counter, or a read/write race is risked with other threads also manipulating the counter. While here: - Remove #if 0'd code regarding acquiring the socket buffer sleep lock in the garbage collector, this is not required as we are able to use the socket buffer receive lock to protect scanning the receive buffer for in-flight file descriptors on the socket buffer. - Annotate that the description of the garbage collector implementation is increasingly inaccurate and needs to be updated. - Add counters of the number of deferred garbage collections and recycled file descriptors. This will be removed and is here temporarily for debugging purposes. With these changes in place, the unp_passfd regression test now appears to be passed consistently on UP and SMP systems for extended runs, whereas before it hung quickly or panicked, depending on which bug was triggered. Reported by: Philip Kizer <pckizer at nostrum dot com> MFC after: 2 weeks
2005-11-10 16:06:04 +00:00
unp_gc(__unused void *arg, int pending)
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
{
First cut at implementing SOCK_SEQPACKET support for UNIX (local) domain sockets. This allows for reliable bi-directional datagram communication over UNIX domain sockets, in contrast to SOCK_DGRAM (M:N, unreliable) or SOCK_STERAM (bi-directional bytestream). Largely, this reuses existing UNIX domain socket code. This allows applications requiring record- oriented semantics to do so reliably via local IPC. Some implementation notes (also present in XXX comments): - Currently we lack an sbappend variant able to do datagrams and control data without doing addresses, so we mark SOCK_SEQPACKET as PR_ADDR. Adding a new variant will solve this problem. - UNIX domain sockets on FreeBSD provide back-pressure/flow control notification for stream sockets by manipulating the send socket buffer's size during pru_send and pru_rcvd. This trick works less well for SOCK_SEQPACKET as sosend_generic() uses sb_hiwat not just to manage blocking, but also to determine maximum datagram size. Fixing this requires rethinking how back-pressure is done for SOCK_SEQPACKET; in the mean time, it's possible to get EMSGSIZE when buffers fill, instead of blocking. Discussed with: benl Reviewed by: bz, rpaulo MFC after: 3 months Sponsored by: Google
2009-10-05 14:49:16 +00:00
struct unp_head *heads[] = { &unp_dhead, &unp_shead, &unp_sphead,
NULL };
Remove explicit locking of struct file. - Introduce a finit() which is used to initailize the fields of struct file in such a way that the ops vector is only valid after the data, type, and flags are valid. - Protect f_flag and f_count with atomic operations. - Remove the global list of all files and associated accounting. - Rewrite the unp garbage collection such that it no longer requires the global list of all files and instead uses a list of all unp sockets. - Mark sockets in the accept queue so we don't incorrectly gc them. Tested by: kris, pho
2007-12-30 01:42:15 +00:00
struct unp_head **head;
The unp_gc() function drops and reaquires lock between scan and collect phases. The unp_discard() function executes unp_externalize_fp(), which might make the socket eligible for gc-ing, and then, later, taskqueue will close the socket. Since unp_gc() dropped the list lock to do the malloc, close might happen after the mark step but before the collection step, causing collection to not find the socket and miss one array element. I believe that the race was there before r216158, but the stated revision made the window much wider by postponing the close to taskqueue sometimes. Only process as much array elements as we find the sockets during second phase of gc [1]. Take linkage lock and recheck the eligibility of the socket for gc, as well as call fhold() under the linkage lock. Reported and tested by: jmallett Submitted by: jmallett [1] Reviewed by: rwatson, jeff (possibly) MFC after: 1 week
2011-02-01 13:33:49 +00:00
struct file *f, **unref;
Remove explicit locking of struct file. - Introduce a finit() which is used to initailize the fields of struct file in such a way that the ops vector is only valid after the data, type, and flags are valid. - Protect f_flag and f_count with atomic operations. - Remove the global list of all files and associated accounting. - Rewrite the unp garbage collection such that it no longer requires the global list of all files and instead uses a list of all unp sockets. - Mark sockets in the accept queue so we don't incorrectly gc them. Tested by: kris, pho
2007-12-30 01:42:15 +00:00
struct unpcb *unp;
The unp_gc() function drops and reaquires lock between scan and collect phases. The unp_discard() function executes unp_externalize_fp(), which might make the socket eligible for gc-ing, and then, later, taskqueue will close the socket. Since unp_gc() dropped the list lock to do the malloc, close might happen after the mark step but before the collection step, causing collection to not find the socket and miss one array element. I believe that the race was there before r216158, but the stated revision made the window much wider by postponing the close to taskqueue sometimes. Only process as much array elements as we find the sockets during second phase of gc [1]. Take linkage lock and recheck the eligibility of the socket for gc, as well as call fhold() under the linkage lock. Reported and tested by: jmallett Submitted by: jmallett [1] Reviewed by: rwatson, jeff (possibly) MFC after: 1 week
2011-02-01 13:33:49 +00:00
int i, total;
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
Correct a number of serious and closely related bugs in the UNIX domain socket file descriptor garbage collection code, which is intended to detect and clear cycles of orphaned file descriptors that are "in-flight" in a socket when that socket is closed before they are received. The algorithm present was both run at poor times (resulting in recursion and reentrance), and also buggy in the presence of parallelism. In order to fix these problems, make the following changes: - When there are in-flight sockets and a UNIX domain socket is destroyed, asynchronously schedule the garbage collector, rather than running it synchronously in the current context. This avoids lock order issues when the garbage collection code reenters the UNIX domain socket code, avoiding lock order reversals, deadlocks, etc. Run the code asynchronously in a task queue. - In the garbage collector, when skipping file descriptors that have entered a closing state (i.e., have f_count == 0), re-test the FDEFER flag, and decrement unp_defer. As file descriptors can now transition to a closed state, while the garbage collector is running, it is no longer the case that unp_defer will remain an accurate count of deferred sockets in the mark portion of the GC algorithm. Otherwise, the garbage collector will loop waiting waiting for unp_defer to reach zero, which it will never do as it is skipping file descriptors that were marked in an earlier pass, but now closed. - Acquire the UNIX domain socket subsystem lock in unp_discard() when modifying the unp_rights counter, or a read/write race is risked with other threads also manipulating the counter. While here: - Remove #if 0'd code regarding acquiring the socket buffer sleep lock in the garbage collector, this is not required as we are able to use the socket buffer receive lock to protect scanning the receive buffer for in-flight file descriptors on the socket buffer. - Annotate that the description of the garbage collector implementation is increasingly inaccurate and needs to be updated. - Add counters of the number of deferred garbage collections and recycled file descriptors. This will be removed and is here temporarily for debugging purposes. With these changes in place, the unp_passfd regression test now appears to be passed consistently on UP and SMP systems for extended runs, whereas before it hung quickly or panicked, depending on which bug was triggered. Reported by: Philip Kizer <pckizer at nostrum dot com> MFC after: 2 weeks
2005-11-10 16:06:04 +00:00
unp_taskcount++;
Listening sockets improvements. o Separate fields of struct socket that belong to listening from fields that belong to normal dataflow, and unionize them. This shrinks the structure a bit. - Take out selinfo's from the socket buffers into the socket. The first reason is to support braindamaged scenario when a socket is added to kevent(2) and then listen(2) is cast on it. The second reason is that there is future plan to make socket buffers pluggable, so that for a dataflow socket a socket buffer can be changed, and in this case we also want to keep same selinfos through the lifetime of a socket. - Remove struct struct so_accf. Since now listening stuff no longer affects struct socket size, just move its fields into listening part of the union. - Provide sol_upcall field and enforce that so_upcall_set() may be called only on a dataflow socket, which has buffers, and for listening sockets provide solisten_upcall_set(). o Remove ACCEPT_LOCK() global. - Add a mutex to socket, to be used instead of socket buffer lock to lock fields of struct socket that don't belong to a socket buffer. - Allow to acquire two socket locks, but the first one must belong to a listening socket. - Make soref()/sorele() to use atomic(9). This allows in some situations to do soref() without owning socket lock. There is place for improvement here, it is possible to make sorele() also to lock optionally. - Most protocols aren't touched by this change, except UNIX local sockets. See below for more information. o Reduce copy-and-paste in kernel modules that accept connections from listening sockets: provide function solisten_dequeue(), and use it in the following modules: ctl(4), iscsi(4), ng_btsocket(4), ng_ksocket(4), infiniband, rpc. o UNIX local sockets. - Removal of ACCEPT_LOCK() global uncovered several races in the UNIX local sockets. Most races exist around spawning a new socket, when we are connecting to a local listening socket. To cover them, we need to hold locks on both PCBs when spawning a third one. This means holding them across sonewconn(). This creates a LOR between pcb locks and unp_list_lock. - To fix the new LOR, abandon the global unp_list_lock in favor of global unp_link_lock. Indeed, separating these two locks didn't provide us any extra parralelism in the UNIX sockets. - Now call into uipc_attach() may happen with unp_link_lock hold if, we are accepting, or without unp_link_lock in case if we are just creating a socket. - Another problem in UNIX sockets is that uipc_close() basicly did nothing for a listening socket. The vnode remained opened for connections. This is fixed by removing vnode in uipc_close(). Maybe the right way would be to do it for all sockets (not only listening), simply move the vnode teardown from uipc_detach() to uipc_close()? Sponsored by: Netflix Differential Revision: https://reviews.freebsd.org/D9770
2017-06-08 21:30:34 +00:00
UNP_LINK_RLOCK();
Remove explicit locking of struct file. - Introduce a finit() which is used to initailize the fields of struct file in such a way that the ops vector is only valid after the data, type, and flags are valid. - Protect f_flag and f_count with atomic operations. - Remove the global list of all files and associated accounting. - Rewrite the unp garbage collection such that it no longer requires the global list of all files and instead uses a list of all unp sockets. - Mark sockets in the accept queue so we don't incorrectly gc them. Tested by: kris, pho
2007-12-30 01:42:15 +00:00
/*
Fix cleanup race between unp_dispose and unp_gc unp_dispose and unp_gc could race to teardown the same mbuf chains, which can lead to dereferencing freed filedesc pointers. This patch adds an IGNORE_RIGHTS flag on unpcbs marking the unpcb's RIGHTS as invalid/freed. The flag is protected by UNP_LIST_LOCK. To serialize against unp_gc, unp_dispose needs the socket object. Change the dom_dispose() KPI to take a socket object instead of an mbuf chain directly. PR: 194264 Differential Revision: https://reviews.freebsd.org/D3044 Reviewed by: mjg (earlier version) Approved by: markj (mentor) Obtained from: mjg MFC after: 1 month Sponsored by: EMC / Isilon Storage Division
2015-07-14 02:00:50 +00:00
* First clear all gc flags from previous runs, apart from
* UNPGC_IGNORE_RIGHTS.
Remove explicit locking of struct file. - Introduce a finit() which is used to initailize the fields of struct file in such a way that the ops vector is only valid after the data, type, and flags are valid. - Protect f_flag and f_count with atomic operations. - Remove the global list of all files and associated accounting. - Rewrite the unp garbage collection such that it no longer requires the global list of all files and instead uses a list of all unp sockets. - Mark sockets in the accept queue so we don't incorrectly gc them. Tested by: kris, pho
2007-12-30 01:42:15 +00:00
*/
for (head = heads; *head != NULL; head++)
LIST_FOREACH(unp, *head, unp_link)
Fix cleanup race between unp_dispose and unp_gc unp_dispose and unp_gc could race to teardown the same mbuf chains, which can lead to dereferencing freed filedesc pointers. This patch adds an IGNORE_RIGHTS flag on unpcbs marking the unpcb's RIGHTS as invalid/freed. The flag is protected by UNP_LIST_LOCK. To serialize against unp_gc, unp_dispose needs the socket object. Change the dom_dispose() KPI to take a socket object instead of an mbuf chain directly. PR: 194264 Differential Revision: https://reviews.freebsd.org/D3044 Reviewed by: mjg (earlier version) Approved by: markj (mentor) Obtained from: mjg MFC after: 1 month Sponsored by: EMC / Isilon Storage Division
2015-07-14 02:00:50 +00:00
unp->unp_gcflag =
(unp->unp_gcflag & UNPGC_IGNORE_RIGHTS);
Remove or update several stale comments. A bit of whitespace/style cleanup. Update copyright. MFC after: 3 days (applicable changes)
2008-10-03 09:01:55 +00:00
Mechanical whitespace cleanup; parenthesize return values; other minor style nits.
2004-01-11 19:48:19 +00:00
/*
Remove explicit locking of struct file. - Introduce a finit() which is used to initailize the fields of struct file in such a way that the ops vector is only valid after the data, type, and flags are valid. - Protect f_flag and f_count with atomic operations. - Remove the global list of all files and associated accounting. - Rewrite the unp garbage collection such that it no longer requires the global list of all files and instead uses a list of all unp sockets. - Mark sockets in the accept queue so we don't incorrectly gc them. Tested by: kris, pho
2007-12-30 01:42:15 +00:00
* Scan marking all reachable sockets with UNPGC_REF. Once a socket
* is reachable all of the sockets it references are reachable.
* Stop the scan once we do a complete loop without discovering
* a new reachable socket.
Add comments to hard-to-follow File descriptor handling code
1996-12-05 22:41:13 +00:00
*/
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
do {
Remove explicit locking of struct file. - Introduce a finit() which is used to initailize the fields of struct file in such a way that the ops vector is only valid after the data, type, and flags are valid. - Protect f_flag and f_count with atomic operations. - Remove the global list of all files and associated accounting. - Rewrite the unp garbage collection such that it no longer requires the global list of all files and instead uses a list of all unp sockets. - Mark sockets in the accept queue so we don't incorrectly gc them. Tested by: kris, pho
2007-12-30 01:42:15 +00:00
unp_unreachable = 0;
unp_marked = 0;
for (head = heads; *head != NULL; head++)
LIST_FOREACH(unp, *head, unp_link)
unp_gc_process(unp);
} while (unp_marked);
Listening sockets improvements. o Separate fields of struct socket that belong to listening from fields that belong to normal dataflow, and unionize them. This shrinks the structure a bit. - Take out selinfo's from the socket buffers into the socket. The first reason is to support braindamaged scenario when a socket is added to kevent(2) and then listen(2) is cast on it. The second reason is that there is future plan to make socket buffers pluggable, so that for a dataflow socket a socket buffer can be changed, and in this case we also want to keep same selinfos through the lifetime of a socket. - Remove struct struct so_accf. Since now listening stuff no longer affects struct socket size, just move its fields into listening part of the union. - Provide sol_upcall field and enforce that so_upcall_set() may be called only on a dataflow socket, which has buffers, and for listening sockets provide solisten_upcall_set(). o Remove ACCEPT_LOCK() global. - Add a mutex to socket, to be used instead of socket buffer lock to lock fields of struct socket that don't belong to a socket buffer. - Allow to acquire two socket locks, but the first one must belong to a listening socket. - Make soref()/sorele() to use atomic(9). This allows in some situations to do soref() without owning socket lock. There is place for improvement here, it is possible to make sorele() also to lock optionally. - Most protocols aren't touched by this change, except UNIX local sockets. See below for more information. o Reduce copy-and-paste in kernel modules that accept connections from listening sockets: provide function solisten_dequeue(), and use it in the following modules: ctl(4), iscsi(4), ng_btsocket(4), ng_ksocket(4), infiniband, rpc. o UNIX local sockets. - Removal of ACCEPT_LOCK() global uncovered several races in the UNIX local sockets. Most races exist around spawning a new socket, when we are connecting to a local listening socket. To cover them, we need to hold locks on both PCBs when spawning a third one. This means holding them across sonewconn(). This creates a LOR between pcb locks and unp_list_lock. - To fix the new LOR, abandon the global unp_list_lock in favor of global unp_link_lock. Indeed, separating these two locks didn't provide us any extra parralelism in the UNIX sockets. - Now call into uipc_attach() may happen with unp_link_lock hold if, we are accepting, or without unp_link_lock in case if we are just creating a socket. - Another problem in UNIX sockets is that uipc_close() basicly did nothing for a listening socket. The vnode remained opened for connections. This is fixed by removing vnode in uipc_close(). Maybe the right way would be to do it for all sockets (not only listening), simply move the vnode teardown from uipc_detach() to uipc_close()? Sponsored by: Netflix Differential Revision: https://reviews.freebsd.org/D9770
2017-06-08 21:30:34 +00:00
UNP_LINK_RUNLOCK();
Remove explicit locking of struct file. - Introduce a finit() which is used to initailize the fields of struct file in such a way that the ops vector is only valid after the data, type, and flags are valid. - Protect f_flag and f_count with atomic operations. - Remove the global list of all files and associated accounting. - Rewrite the unp garbage collection such that it no longer requires the global list of all files and instead uses a list of all unp sockets. - Mark sockets in the accept queue so we don't incorrectly gc them. Tested by: kris, pho
2007-12-30 01:42:15 +00:00
if (unp_unreachable == 0)
return;
Remove or update several stale comments. A bit of whitespace/style cleanup. Update copyright. MFC after: 3 days (applicable changes)
2008-10-03 09:01:55 +00:00
Remove explicit locking of struct file. - Introduce a finit() which is used to initailize the fields of struct file in such a way that the ops vector is only valid after the data, type, and flags are valid. - Protect f_flag and f_count with atomic operations. - Remove the global list of all files and associated accounting. - Rewrite the unp garbage collection such that it no longer requires the global list of all files and instead uses a list of all unp sockets. - Mark sockets in the accept queue so we don't incorrectly gc them. Tested by: kris, pho
2007-12-30 01:42:15 +00:00
/*
* Allocate space for a local list of dead unpcbs.
*/
unref = malloc(unp_unreachable * sizeof(struct file *),
M_TEMP, M_WAITOK);
Remove or update several stale comments. A bit of whitespace/style cleanup. Update copyright. MFC after: 3 days (applicable changes)
2008-10-03 09:01:55 +00:00
Remove explicit locking of struct file. - Introduce a finit() which is used to initailize the fields of struct file in such a way that the ops vector is only valid after the data, type, and flags are valid. - Protect f_flag and f_count with atomic operations. - Remove the global list of all files and associated accounting. - Rewrite the unp garbage collection such that it no longer requires the global list of all files and instead uses a list of all unp sockets. - Mark sockets in the accept queue so we don't incorrectly gc them. Tested by: kris, pho
2007-12-30 01:42:15 +00:00
/*
* Iterate looking for sockets which have been specifically marked
* as as unreachable and store them locally.
*/
The unp_gc() function drops and reaquires lock between scan and collect phases. The unp_discard() function executes unp_externalize_fp(), which might make the socket eligible for gc-ing, and then, later, taskqueue will close the socket. Since unp_gc() dropped the list lock to do the malloc, close might happen after the mark step but before the collection step, causing collection to not find the socket and miss one array element. I believe that the race was there before r216158, but the stated revision made the window much wider by postponing the close to taskqueue sometimes. Only process as much array elements as we find the sockets during second phase of gc [1]. Take linkage lock and recheck the eligibility of the socket for gc, as well as call fhold() under the linkage lock. Reported and tested by: jmallett Submitted by: jmallett [1] Reviewed by: rwatson, jeff (possibly) MFC after: 1 week
2011-02-01 13:33:49 +00:00
UNP_LINK_RLOCK();
for (total = 0, head = heads; *head != NULL; head++)
Remove explicit locking of struct file. - Introduce a finit() which is used to initailize the fields of struct file in such a way that the ops vector is only valid after the data, type, and flags are valid. - Protect f_flag and f_count with atomic operations. - Remove the global list of all files and associated accounting. - Rewrite the unp garbage collection such that it no longer requires the global list of all files and instead uses a list of all unp sockets. - Mark sockets in the accept queue so we don't incorrectly gc them. Tested by: kris, pho
2007-12-30 01:42:15 +00:00
LIST_FOREACH(unp, *head, unp_link)
The unp_gc() function drops and reaquires lock between scan and collect phases. The unp_discard() function executes unp_externalize_fp(), which might make the socket eligible for gc-ing, and then, later, taskqueue will close the socket. Since unp_gc() dropped the list lock to do the malloc, close might happen after the mark step but before the collection step, causing collection to not find the socket and miss one array element. I believe that the race was there before r216158, but the stated revision made the window much wider by postponing the close to taskqueue sometimes. Only process as much array elements as we find the sockets during second phase of gc [1]. Take linkage lock and recheck the eligibility of the socket for gc, as well as call fhold() under the linkage lock. Reported and tested by: jmallett Submitted by: jmallett [1] Reviewed by: rwatson, jeff (possibly) MFC after: 1 week
2011-02-01 13:33:49 +00:00
if ((unp->unp_gcflag & UNPGC_DEAD) != 0) {
f = unp->unp_file;
if (unp->unp_msgcount == 0 || f == NULL ||
f->f_count != unp->unp_msgcount)
continue;
unref[total++] = f;
fhold(f);
KASSERT(total <= unp_unreachable,
Remove explicit locking of struct file. - Introduce a finit() which is used to initailize the fields of struct file in such a way that the ops vector is only valid after the data, type, and flags are valid. - Protect f_flag and f_count with atomic operations. - Remove the global list of all files and associated accounting. - Rewrite the unp garbage collection such that it no longer requires the global list of all files and instead uses a list of all unp sockets. - Mark sockets in the accept queue so we don't incorrectly gc them. Tested by: kris, pho
2007-12-30 01:42:15 +00:00
("unp_gc: incorrect unreachable count."));
- Close a race between enumerating UNIX domain socket pcb structures via sysctl and socket teardown by adding a reference count to the UNIX domain pcb object and fixing the sysctl that enumerates unpcbs to grab a reference on each unpcb while it builds the list to copy out to userland. - Close a race between UNIX domain pcb garbage collection (unp_gc()) and file descriptor teardown (fdrop()) by adding a new garbage collection flag FWAIT. unp_gc() sets FWAIT while it walks the message buffers in a UNIX domain socket looking for nested file descriptor references and clears the flag when it is finished. fdrop() checks to see if the flag is set on a file descriptor whose refcount just dropped to 0 and waits for unp_gc() to clear the flag before completely destroying the file descriptor. MFC after: 1 week Reviewed by: rwatson Submitted by: ups Hopefully makes the panics go away: mx1
2007-01-05 19:59:46 +00:00
}
The unp_gc() function drops and reaquires lock between scan and collect phases. The unp_discard() function executes unp_externalize_fp(), which might make the socket eligible for gc-ing, and then, later, taskqueue will close the socket. Since unp_gc() dropped the list lock to do the malloc, close might happen after the mark step but before the collection step, causing collection to not find the socket and miss one array element. I believe that the race was there before r216158, but the stated revision made the window much wider by postponing the close to taskqueue sometimes. Only process as much array elements as we find the sockets during second phase of gc [1]. Take linkage lock and recheck the eligibility of the socket for gc, as well as call fhold() under the linkage lock. Reported and tested by: jmallett Submitted by: jmallett [1] Reviewed by: rwatson, jeff (possibly) MFC after: 1 week
2011-02-01 13:33:49 +00:00
UNP_LINK_RUNLOCK();
Remove or update several stale comments. A bit of whitespace/style cleanup. Update copyright. MFC after: 3 days (applicable changes)
2008-10-03 09:01:55 +00:00
Mechanical whitespace cleanup; parenthesize return values; other minor style nits.
2004-01-11 19:48:19 +00:00
/*
Remove explicit locking of struct file. - Introduce a finit() which is used to initailize the fields of struct file in such a way that the ops vector is only valid after the data, type, and flags are valid. - Protect f_flag and f_count with atomic operations. - Remove the global list of all files and associated accounting. - Rewrite the unp garbage collection such that it no longer requires the global list of all files and instead uses a list of all unp sockets. - Mark sockets in the accept queue so we don't incorrectly gc them. Tested by: kris, pho
2007-12-30 01:42:15 +00:00
* Now flush all sockets, free'ing rights. This will free the
* struct files associated with these sockets but leave each socket
* with one remaining ref.
Add comments to hard-to-follow File descriptor handling code
1996-12-05 22:41:13 +00:00
*/
Mfp4 CH=177274,177280,177284-177285,177297,177324-177325 VNET socket push back: try to minimize the number of places where we have to switch vnets and narrow down the time we stay switched. Add assertions to the socket code to catch possibly unset vnets as seen in r204147. While this reduces the number of vnet recursion in some places like NFS, POSIX local sockets and some netgraph, .. recursions are impossible to fix. The current expectations are documented at the beginning of uipc_socket.c along with the other information there. Sponsored by: The FreeBSD Foundation Sponsored by: CK Software GmbH Reviewed by: jhb Tested by: zec Tested by: Mikolaj Golub (to.my.trociny gmail.com) MFC after: 2 weeks
2011-02-16 21:29:13 +00:00
for (i = 0; i < total; i++) {
struct socket *so;
so = unref[i]->f_data;
CURVNET_SET(so->so_vnet);
sorflush(so);
CURVNET_RESTORE();
}
Remove or update several stale comments. A bit of whitespace/style cleanup. Update copyright. MFC after: 3 days (applicable changes)
2008-10-03 09:01:55 +00:00
Remove explicit locking of struct file. - Introduce a finit() which is used to initailize the fields of struct file in such a way that the ops vector is only valid after the data, type, and flags are valid. - Protect f_flag and f_count with atomic operations. - Remove the global list of all files and associated accounting. - Rewrite the unp garbage collection such that it no longer requires the global list of all files and instead uses a list of all unp sockets. - Mark sockets in the accept queue so we don't incorrectly gc them. Tested by: kris, pho
2007-12-30 01:42:15 +00:00
/*
* And finally release the sockets so they can be reclaimed.
*/
The unp_gc() function drops and reaquires lock between scan and collect phases. The unp_discard() function executes unp_externalize_fp(), which might make the socket eligible for gc-ing, and then, later, taskqueue will close the socket. Since unp_gc() dropped the list lock to do the malloc, close might happen after the mark step but before the collection step, causing collection to not find the socket and miss one array element. I believe that the race was there before r216158, but the stated revision made the window much wider by postponing the close to taskqueue sometimes. Only process as much array elements as we find the sockets during second phase of gc [1]. Take linkage lock and recheck the eligibility of the socket for gc, as well as call fhold() under the linkage lock. Reported and tested by: jmallett Submitted by: jmallett [1] Reviewed by: rwatson, jeff (possibly) MFC after: 1 week
2011-02-01 13:33:49 +00:00
for (i = 0; i < total; i++)
Remove explicit locking of struct file. - Introduce a finit() which is used to initailize the fields of struct file in such a way that the ops vector is only valid after the data, type, and flags are valid. - Protect f_flag and f_count with atomic operations. - Remove the global list of all files and associated accounting. - Rewrite the unp garbage collection such that it no longer requires the global list of all files and instead uses a list of all unp sockets. - Mark sockets in the accept queue so we don't incorrectly gc them. Tested by: kris, pho
2007-12-30 01:42:15 +00:00
fdrop(unref[i], NULL);
The unp_gc() function drops and reaquires lock between scan and collect phases. The unp_discard() function executes unp_externalize_fp(), which might make the socket eligible for gc-ing, and then, later, taskqueue will close the socket. Since unp_gc() dropped the list lock to do the malloc, close might happen after the mark step but before the collection step, causing collection to not find the socket and miss one array element. I believe that the race was there before r216158, but the stated revision made the window much wider by postponing the close to taskqueue sometimes. Only process as much array elements as we find the sockets during second phase of gc [1]. Take linkage lock and recheck the eligibility of the socket for gc, as well as call fhold() under the linkage lock. Reported and tested by: jmallett Submitted by: jmallett [1] Reviewed by: rwatson, jeff (possibly) MFC after: 1 week
2011-02-01 13:33:49 +00:00
unp_recycled += total;
Remove explicit locking of struct file. - Introduce a finit() which is used to initailize the fields of struct file in such a way that the ops vector is only valid after the data, type, and flags are valid. - Protect f_flag and f_count with atomic operations. - Remove the global list of all files and associated accounting. - Rewrite the unp garbage collection such that it no longer requires the global list of all files and instead uses a list of all unp sockets. - Mark sockets in the accept queue so we don't incorrectly gc them. Tested by: kris, pho
2007-12-30 01:42:15 +00:00
free(unref, M_TEMP);
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
}
Further minor cleanups to UNIX domain sockets: - Staticize and locally prototype functions uipc_ctloutput(), unp_dispose(), unp_init(), and unp_externalize(), none of which have been required outside of uipc_usrreq.c since uipc_proto.c was removed. - Remove stale prototype for uipc_usrreq(), which has not existed in the code since 1997 - Forward declare and staticize uipc_usrreqs structure in uipc_usrreq.c and not un.h. - Comment on why uipc_connect2() is still non-static -- it is used directly by fifofs. - Remove stale comments, tidy up whitespace. MFC after: 3 days (where applicable)
2008-10-03 13:01:56 +00:00
static void
Rename unp_dispose_so() to unp_dispose(). It implements the dom_dispose method for local socket domain, so its name should match the method name.
2016-08-31 21:48:22 +00:00
unp_dispose_mbuf(struct mbuf *m)
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
{
This is the kernel Lite/2 commit. There are some requisite userland changes, so don't expect to be able to run the kernel as-is (very well) without the appropriate Lite/2 userland changes. The system boots and can mount UFS filesystems. Untested: ext2fs, msdosfs, NFS Known problems: Incorrect Berkeley ID strings in some files. Mount_std mounts will not work until the getfsent library routine is changed. Reviewed by: various people Submitted by: Jeffery Hsu <hsu@freebsd.org>
1997-02-10 02:22:35 +00:00
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
if (m)
Fix memory leak when one process send descriptor over UNIX domain socket, but the other process exited before receiving it.
2013-03-11 22:59:07 +00:00
unp_scan(m, unp_freerights);
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
}
Fix cleanup race between unp_dispose and unp_gc unp_dispose and unp_gc could race to teardown the same mbuf chains, which can lead to dereferencing freed filedesc pointers. This patch adds an IGNORE_RIGHTS flag on unpcbs marking the unpcb's RIGHTS as invalid/freed. The flag is protected by UNP_LIST_LOCK. To serialize against unp_gc, unp_dispose needs the socket object. Change the dom_dispose() KPI to take a socket object instead of an mbuf chain directly. PR: 194264 Differential Revision: https://reviews.freebsd.org/D3044 Reviewed by: mjg (earlier version) Approved by: markj (mentor) Obtained from: mjg MFC after: 1 month Sponsored by: EMC / Isilon Storage Division
2015-07-14 02:00:50 +00:00
/*
* Synchronize against unp_gc, which can trip over data as we are freeing it.
*/
static void
Rename unp_dispose_so() to unp_dispose(). It implements the dom_dispose method for local socket domain, so its name should match the method name.
2016-08-31 21:48:22 +00:00
unp_dispose(struct socket *so)
Fix cleanup race between unp_dispose and unp_gc unp_dispose and unp_gc could race to teardown the same mbuf chains, which can lead to dereferencing freed filedesc pointers. This patch adds an IGNORE_RIGHTS flag on unpcbs marking the unpcb's RIGHTS as invalid/freed. The flag is protected by UNP_LIST_LOCK. To serialize against unp_gc, unp_dispose needs the socket object. Change the dom_dispose() KPI to take a socket object instead of an mbuf chain directly. PR: 194264 Differential Revision: https://reviews.freebsd.org/D3044 Reviewed by: mjg (earlier version) Approved by: markj (mentor) Obtained from: mjg MFC after: 1 month Sponsored by: EMC / Isilon Storage Division
2015-07-14 02:00:50 +00:00
{
struct unpcb *unp;
unp = sotounpcb(so);
Listening sockets improvements. o Separate fields of struct socket that belong to listening from fields that belong to normal dataflow, and unionize them. This shrinks the structure a bit. - Take out selinfo's from the socket buffers into the socket. The first reason is to support braindamaged scenario when a socket is added to kevent(2) and then listen(2) is cast on it. The second reason is that there is future plan to make socket buffers pluggable, so that for a dataflow socket a socket buffer can be changed, and in this case we also want to keep same selinfos through the lifetime of a socket. - Remove struct struct so_accf. Since now listening stuff no longer affects struct socket size, just move its fields into listening part of the union. - Provide sol_upcall field and enforce that so_upcall_set() may be called only on a dataflow socket, which has buffers, and for listening sockets provide solisten_upcall_set(). o Remove ACCEPT_LOCK() global. - Add a mutex to socket, to be used instead of socket buffer lock to lock fields of struct socket that don't belong to a socket buffer. - Allow to acquire two socket locks, but the first one must belong to a listening socket. - Make soref()/sorele() to use atomic(9). This allows in some situations to do soref() without owning socket lock. There is place for improvement here, it is possible to make sorele() also to lock optionally. - Most protocols aren't touched by this change, except UNIX local sockets. See below for more information. o Reduce copy-and-paste in kernel modules that accept connections from listening sockets: provide function solisten_dequeue(), and use it in the following modules: ctl(4), iscsi(4), ng_btsocket(4), ng_ksocket(4), infiniband, rpc. o UNIX local sockets. - Removal of ACCEPT_LOCK() global uncovered several races in the UNIX local sockets. Most races exist around spawning a new socket, when we are connecting to a local listening socket. To cover them, we need to hold locks on both PCBs when spawning a third one. This means holding them across sonewconn(). This creates a LOR between pcb locks and unp_list_lock. - To fix the new LOR, abandon the global unp_list_lock in favor of global unp_link_lock. Indeed, separating these two locks didn't provide us any extra parralelism in the UNIX sockets. - Now call into uipc_attach() may happen with unp_link_lock hold if, we are accepting, or without unp_link_lock in case if we are just creating a socket. - Another problem in UNIX sockets is that uipc_close() basicly did nothing for a listening socket. The vnode remained opened for connections. This is fixed by removing vnode in uipc_close(). Maybe the right way would be to do it for all sockets (not only listening), simply move the vnode teardown from uipc_detach() to uipc_close()? Sponsored by: Netflix Differential Revision: https://reviews.freebsd.org/D9770
2017-06-08 21:30:34 +00:00
UNP_LINK_WLOCK();
Fix cleanup race between unp_dispose and unp_gc unp_dispose and unp_gc could race to teardown the same mbuf chains, which can lead to dereferencing freed filedesc pointers. This patch adds an IGNORE_RIGHTS flag on unpcbs marking the unpcb's RIGHTS as invalid/freed. The flag is protected by UNP_LIST_LOCK. To serialize against unp_gc, unp_dispose needs the socket object. Change the dom_dispose() KPI to take a socket object instead of an mbuf chain directly. PR: 194264 Differential Revision: https://reviews.freebsd.org/D3044 Reviewed by: mjg (earlier version) Approved by: markj (mentor) Obtained from: mjg MFC after: 1 month Sponsored by: EMC / Isilon Storage Division
2015-07-14 02:00:50 +00:00
unp->unp_gcflag |= UNPGC_IGNORE_RIGHTS;
Listening sockets improvements. o Separate fields of struct socket that belong to listening from fields that belong to normal dataflow, and unionize them. This shrinks the structure a bit. - Take out selinfo's from the socket buffers into the socket. The first reason is to support braindamaged scenario when a socket is added to kevent(2) and then listen(2) is cast on it. The second reason is that there is future plan to make socket buffers pluggable, so that for a dataflow socket a socket buffer can be changed, and in this case we also want to keep same selinfos through the lifetime of a socket. - Remove struct struct so_accf. Since now listening stuff no longer affects struct socket size, just move its fields into listening part of the union. - Provide sol_upcall field and enforce that so_upcall_set() may be called only on a dataflow socket, which has buffers, and for listening sockets provide solisten_upcall_set(). o Remove ACCEPT_LOCK() global. - Add a mutex to socket, to be used instead of socket buffer lock to lock fields of struct socket that don't belong to a socket buffer. - Allow to acquire two socket locks, but the first one must belong to a listening socket. - Make soref()/sorele() to use atomic(9). This allows in some situations to do soref() without owning socket lock. There is place for improvement here, it is possible to make sorele() also to lock optionally. - Most protocols aren't touched by this change, except UNIX local sockets. See below for more information. o Reduce copy-and-paste in kernel modules that accept connections from listening sockets: provide function solisten_dequeue(), and use it in the following modules: ctl(4), iscsi(4), ng_btsocket(4), ng_ksocket(4), infiniband, rpc. o UNIX local sockets. - Removal of ACCEPT_LOCK() global uncovered several races in the UNIX local sockets. Most races exist around spawning a new socket, when we are connecting to a local listening socket. To cover them, we need to hold locks on both PCBs when spawning a third one. This means holding them across sonewconn(). This creates a LOR between pcb locks and unp_list_lock. - To fix the new LOR, abandon the global unp_list_lock in favor of global unp_link_lock. Indeed, separating these two locks didn't provide us any extra parralelism in the UNIX sockets. - Now call into uipc_attach() may happen with unp_link_lock hold if, we are accepting, or without unp_link_lock in case if we are just creating a socket. - Another problem in UNIX sockets is that uipc_close() basicly did nothing for a listening socket. The vnode remained opened for connections. This is fixed by removing vnode in uipc_close(). Maybe the right way would be to do it for all sockets (not only listening), simply move the vnode teardown from uipc_detach() to uipc_close()? Sponsored by: Netflix Differential Revision: https://reviews.freebsd.org/D9770
2017-06-08 21:30:34 +00:00
UNP_LINK_WUNLOCK();
if (!SOLISTENING(so))
unp_dispose_mbuf(so->so_rcv.sb_mb);
Fix cleanup race between unp_dispose and unp_gc unp_dispose and unp_gc could race to teardown the same mbuf chains, which can lead to dereferencing freed filedesc pointers. This patch adds an IGNORE_RIGHTS flag on unpcbs marking the unpcb's RIGHTS as invalid/freed. The flag is protected by UNP_LIST_LOCK. To serialize against unp_gc, unp_dispose needs the socket object. Change the dom_dispose() KPI to take a socket object instead of an mbuf chain directly. PR: 194264 Differential Revision: https://reviews.freebsd.org/D3044 Reviewed by: mjg (earlier version) Approved by: markj (mentor) Obtained from: mjg MFC after: 1 month Sponsored by: EMC / Isilon Storage Division
2015-07-14 02:00:50 +00:00
}
Another mega commit to staticize things.
1995-12-14 09:55:16 +00:00
static void
Fix memory leak when one process send descriptor over UNIX domain socket, but the other process exited before receiving it.
2013-03-11 22:59:07 +00:00
unp_scan(struct mbuf *m0, void (*op)(struct filedescent **, int))
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
{
Hopefully improve control message passing over Unix domain sockets. 1) Allow the sending of more than one control message at a time over a unix domain socket. This should cover the PR 29499. 2) This requires that unp_{ex,in}ternalize and unp_scan understand mbufs with more than one control message at a time. 3) Internalize and externalize used to work on the mbuf in-place. This made life quite complicated and the code for sizeof(int) < sizeof(file *) could end up doing the wrong thing. The patch always create a new mbuf/cluster now. This resulted in the change of the prototype for the domain externalise function. 4) You can now send SCM_TIMESTAMP messages. 5) Always use CMSG_DATA(cm) to determine the start where the data in unp_{ex,in}ternalize. It was using ((struct cmsghdr *)cm + 1) in some places, which gives the wrong alignment on the alpha. (NetBSD made this fix some time ago). This results in an ABI change for discriptor passing and creds passing on the alpha. (Probably on the IA64 and Spare ports too). 6) Fix userland programs to use CMSG_* macros too. 7) Be more careful about freeing mbufs containing (file *)s. This is made possible by the prototype change of externalise. PR: 29499 MFC after: 6 weeks
2001-10-04 13:11:48 +00:00
struct mbuf *m;
struct cmsghdr *cm;
void *data;
socklen_t clen, datalen;
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
Prefer NULL to 0 when testing and assigning pointer values.
2004-03-30 02:16:25 +00:00
while (m0 != NULL) {
Hopefully improve control message passing over Unix domain sockets. 1) Allow the sending of more than one control message at a time over a unix domain socket. This should cover the PR 29499. 2) This requires that unp_{ex,in}ternalize and unp_scan understand mbufs with more than one control message at a time. 3) Internalize and externalize used to work on the mbuf in-place. This made life quite complicated and the code for sizeof(int) < sizeof(file *) could end up doing the wrong thing. The patch always create a new mbuf/cluster now. This resulted in the change of the prototype for the domain externalise function. 4) You can now send SCM_TIMESTAMP messages. 5) Always use CMSG_DATA(cm) to determine the start where the data in unp_{ex,in}ternalize. It was using ((struct cmsghdr *)cm + 1) in some places, which gives the wrong alignment on the alpha. (NetBSD made this fix some time ago). This results in an ABI change for discriptor passing and creds passing on the alpha. (Probably on the IA64 and Spare ports too). 6) Fix userland programs to use CMSG_* macros too. 7) Be more careful about freeing mbufs containing (file *)s. This is made possible by the prototype change of externalise. PR: 29499 MFC after: 6 weeks
2001-10-04 13:11:48 +00:00
for (m = m0; m; m = m->m_next) {
When scanning for control messages, don't process the data mbufs. This could cause hangs if a unix domain socket was closed with data still to be read from it. Tested by: Andrea Campi <andrea@webcom.it>
2001-10-29 20:04:03 +00:00
if (m->m_type != MT_CONTROL)
Hopefully improve control message passing over Unix domain sockets. 1) Allow the sending of more than one control message at a time over a unix domain socket. This should cover the PR 29499. 2) This requires that unp_{ex,in}ternalize and unp_scan understand mbufs with more than one control message at a time. 3) Internalize and externalize used to work on the mbuf in-place. This made life quite complicated and the code for sizeof(int) < sizeof(file *) could end up doing the wrong thing. The patch always create a new mbuf/cluster now. This resulted in the change of the prototype for the domain externalise function. 4) You can now send SCM_TIMESTAMP messages. 5) Always use CMSG_DATA(cm) to determine the start where the data in unp_{ex,in}ternalize. It was using ((struct cmsghdr *)cm + 1) in some places, which gives the wrong alignment on the alpha. (NetBSD made this fix some time ago). This results in an ABI change for discriptor passing and creds passing on the alpha. (Probably on the IA64 and Spare ports too). 6) Fix userland programs to use CMSG_* macros too. 7) Be more careful about freeing mbufs containing (file *)s. This is made possible by the prototype change of externalise. PR: 29499 MFC after: 6 weeks
2001-10-04 13:11:48 +00:00
continue;
cm = mtod(m, struct cmsghdr *);
clen = m->m_len;
while (cm != NULL) {
if (sizeof(*cm) > clen || cm->cmsg_len > clen)
break;
data = CMSG_DATA(cm);
datalen = (caddr_t)cm + cm->cmsg_len
- (caddr_t)data;
if (cm->cmsg_level == SOL_SOCKET &&
cm->cmsg_type == SCM_RIGHTS) {
Fix memory leak when one process send descriptor over UNIX domain socket, but the other process exited before receiving it.
2013-03-11 22:59:07 +00:00
(*op)(data, datalen /
sizeof(struct filedescent *));
Hopefully improve control message passing over Unix domain sockets. 1) Allow the sending of more than one control message at a time over a unix domain socket. This should cover the PR 29499. 2) This requires that unp_{ex,in}ternalize and unp_scan understand mbufs with more than one control message at a time. 3) Internalize and externalize used to work on the mbuf in-place. This made life quite complicated and the code for sizeof(int) < sizeof(file *) could end up doing the wrong thing. The patch always create a new mbuf/cluster now. This resulted in the change of the prototype for the domain externalise function. 4) You can now send SCM_TIMESTAMP messages. 5) Always use CMSG_DATA(cm) to determine the start where the data in unp_{ex,in}ternalize. It was using ((struct cmsghdr *)cm + 1) in some places, which gives the wrong alignment on the alpha. (NetBSD made this fix some time ago). This results in an ABI change for discriptor passing and creds passing on the alpha. (Probably on the IA64 and Spare ports too). 6) Fix userland programs to use CMSG_* macros too. 7) Be more careful about freeing mbufs containing (file *)s. This is made possible by the prototype change of externalise. PR: 29499 MFC after: 6 weeks
2001-10-04 13:11:48 +00:00
}
if (CMSG_SPACE(datalen) < clen) {
clen -= CMSG_SPACE(datalen);
cm = (struct cmsghdr *)
((caddr_t)cm + CMSG_SPACE(datalen));
} else {
clen = 0;
cm = NULL;
}
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
}
Hopefully improve control message passing over Unix domain sockets. 1) Allow the sending of more than one control message at a time over a unix domain socket. This should cover the PR 29499. 2) This requires that unp_{ex,in}ternalize and unp_scan understand mbufs with more than one control message at a time. 3) Internalize and externalize used to work on the mbuf in-place. This made life quite complicated and the code for sizeof(int) < sizeof(file *) could end up doing the wrong thing. The patch always create a new mbuf/cluster now. This resulted in the change of the prototype for the domain externalise function. 4) You can now send SCM_TIMESTAMP messages. 5) Always use CMSG_DATA(cm) to determine the start where the data in unp_{ex,in}ternalize. It was using ((struct cmsghdr *)cm + 1) in some places, which gives the wrong alignment on the alpha. (NetBSD made this fix some time ago). This results in an ABI change for discriptor passing and creds passing on the alpha. (Probably on the IA64 and Spare ports too). 6) Fix userland programs to use CMSG_* macros too. 7) Be more careful about freeing mbufs containing (file *)s. This is made possible by the prototype change of externalise. PR: 29499 MFC after: 6 weeks
2001-10-04 13:11:48 +00:00
}
Deprecate m_act. Use m_nextpkt always.
2014-07-17 05:21:16 +00:00
m0 = m0->m_nextpkt;
BSD 4.4 Lite Kernel Sources
1994-05-24 10:09:53 +00:00
}
}
When detaching an unix domain socket, uipc_detach() checks unp->unp_vnode pointer to detect if there is a vnode associated with (binded to) this socket and does necessary cleanup if there is. The issue is that after forced unmount this check may be too late as the unp_vnode is reclaimed and the reference is stale. To fix this provide a helper function that is called on a socket vnode reclamation to do necessary cleanup. Pointed by: kib Reviewed by: kib MFC after: 2 weeks
2012-02-25 10:15:41 +00:00
/*
* A helper function called by VFS before socket-type vnode reclamation.
* For an active vnode it clears unp_vnode pointer and decrements unp_vnode
* use count.
*/
void
vfs_unp_reclaim(struct vnode *vp)
{
struct unpcb *unp;
int active;
ASSERT_VOP_ELOCKED(vp, "vfs_unp_reclaim");
KASSERT(vp->v_type == VSOCK,
("vfs_unp_reclaim: vp->v_type != VSOCK"));
active = 0;
UNP_LINK_WLOCK();
For UNIX sockets make vnode point not to the socket, but to the UNIX PCB, since the latter is the thing that links together VFS and sockets. While here, make the union in the struct vnode anonymous.
2017-06-02 17:31:25 +00:00
VOP_UNP_CONNECT(vp, &unp);
When detaching an unix domain socket, uipc_detach() checks unp->unp_vnode pointer to detect if there is a vnode associated with (binded to) this socket and does necessary cleanup if there is. The issue is that after forced unmount this check may be too late as the unp_vnode is reclaimed and the reference is stale. To fix this provide a helper function that is called on a socket vnode reclamation to do necessary cleanup. Pointed by: kib Reviewed by: kib MFC after: 2 weeks
2012-02-25 10:15:41 +00:00
if (unp == NULL)
goto done;
UNP_PCB_LOCK(unp);
Introduce VOP_UNP_BIND(), VOP_UNP_CONNECT(), and VOP_UNP_DETACH() operations for setting and accessing vnode's v_socket field. The operations are necessary to implement proper unix socket handling on layered file systems like nullfs(5). This change fixes the long standing issue with nullfs(5) being in that unix sockets did not work between lower and upper layers: if we bound to a socket on the lower layer we could connect only to the lower path; if we bound to the upper layer we could connect only to the upper path. The new behavior is one can connect to both the lower and the upper paths regardless what layer path one binds to. PR: kern/51583, kern/159663 Suggested by: kib Reviewed by: arch MFC after: 2 weeks
2012-02-29 21:38:31 +00:00
if (unp->unp_vnode == vp) {
VOP_UNP_DETACH(vp);
When detaching an unix domain socket, uipc_detach() checks unp->unp_vnode pointer to detect if there is a vnode associated with (binded to) this socket and does necessary cleanup if there is. The issue is that after forced unmount this check may be too late as the unp_vnode is reclaimed and the reference is stale. To fix this provide a helper function that is called on a socket vnode reclamation to do necessary cleanup. Pointed by: kib Reviewed by: kib MFC after: 2 weeks
2012-02-25 10:15:41 +00:00
unp->unp_vnode = NULL;
active = 1;
}
UNP_PCB_UNLOCK(unp);
done:
UNP_LINK_WUNLOCK();
if (active)
vunref(vp);
}
Add DDB "show unpcb" command, allowing DDB to print out many pertinent details from UNIX domain socket protocol layer state.
2007-05-29 12:36:00 +00:00
#ifdef DDB
static void
db_print_indent(int indent)
{
int i;
for (i = 0; i < indent; i++)
db_printf(" ");
}
static void
db_print_unpflags(int unp_flags)
{
int comma;
comma = 0;
if (unp_flags & UNP_HAVEPC) {
db_printf("%sUNP_HAVEPC", comma ? ", " : "");
comma = 1;
}
if (unp_flags & UNP_WANTCRED) {
db_printf("%sUNP_WANTCRED", comma ? ", " : "");
comma = 1;
}
if (unp_flags & UNP_CONNWAIT) {
db_printf("%sUNP_CONNWAIT", comma ? ", " : "");
comma = 1;
}
if (unp_flags & UNP_CONNECTING) {
db_printf("%sUNP_CONNECTING", comma ? ", " : "");
comma = 1;
}
if (unp_flags & UNP_BINDING) {
db_printf("%sUNP_BINDING", comma ? ", " : "");
comma = 1;
}
}
static void
db_print_xucred(int indent, struct xucred *xu)
{
int comma, i;
db_print_indent(indent);
db_printf("cr_version: %u cr_uid: %u cr_ngroups: %d\n",
xu->cr_version, xu->cr_uid, xu->cr_ngroups);
db_print_indent(indent);
db_printf("cr_groups: ");
comma = 0;
for (i = 0; i < xu->cr_ngroups; i++) {
db_printf("%s%u", comma ? ", " : "", xu->cr_groups[i]);
comma = 1;
}
db_printf("\n");
}
static void
db_print_unprefs(int indent, struct unp_head *uh)
{
struct unpcb *unp;
int counter;
counter = 0;
LIST_FOREACH(unp, uh, unp_reflink) {
if (counter % 4 == 0)
db_print_indent(indent);
db_printf("%p ", unp);
if (counter % 4 == 3)
db_printf("\n");
counter++;
}
if (counter != 0 && counter % 4 != 0)
db_printf("\n");
}
DB_SHOW_COMMAND(unpcb, db_show_unpcb)
{
struct unpcb *unp;
if (!have_addr) {
db_printf("usage: show unpcb <addr>\n");
return;
}
unp = (struct unpcb *)addr;
db_printf("unp_socket: %p unp_vnode: %p\n", unp->unp_socket,
unp->unp_vnode);
Fix up kernel sources to be ready for a 64-bit ino_t. Original code by: Gleb Kurtsou
2012-09-27 23:30:49 +00:00
db_printf("unp_ino: %ju unp_conn: %p\n", (uintmax_t)unp->unp_ino,
Add DDB "show unpcb" command, allowing DDB to print out many pertinent details from UNIX domain socket protocol layer state.
2007-05-29 12:36:00 +00:00
unp->unp_conn);
db_printf("unp_refs:\n");
db_print_unprefs(2, &unp->unp_refs);
/* XXXRW: Would be nice to print the full address, if any. */
db_printf("unp_addr: %p\n", unp->unp_addr);
Replace 4.4BSD Lite's unix domain socket backpressure hack with a cleaner mechanism, based on the new SB_STOP sockbuf flag. The old hack dynamically changed the sending sockbuf's high water mark whenever adding or removing data from the receiving sockbuf. It worked for stream sockets, but it never worked for SOCK_SEQPACKET sockets because of their atomic nature. If the sockbuf was partially full, it might return EMSGSIZE instead of blocking. The new solution is based on DragonFlyBSD's fix from commit 3a6117bbe0ed6a87605c1e43e12a1438d8844380 on 2008-05-27. It adds an SB_STOP flag to sockbufs. Whenever uipc_send surpasses the socket's size limit, it sets SB_STOP on the sending sockbuf. sbspace() will then return 0 for that sockbuf, causing sosend_generic and friends to block. uipc_rcvd will likewise clear SB_STOP. There are two fringe benefits: uipc_{send,rcvd} no longer need to call chgsbsize() on every send and receive because they don't change the sockbuf's high water mark. Also, uipc_sense no longer needs to acquire the UIPC linkage lock, because it's simpler to compute the st_blksizes. There is one drawback: since sbspace() will only ever return 0 or the maximum, sosend_generic will allow the sockbuf to exceed its nominal maximum size by at most one packet of size less than the max. I don't think that's a serious problem. In fact, I'm not even positive that FreeBSD guarantees a socket will always stay within its nominal size limit. sys/sys/sockbuf.h Add the SB_STOP flag and adjust sbspace() sys/sys/unpcb.h Delete the obsolete unp_cc and unp_mbcnt fields from struct unpcb. sys/kern/uipc_usrreq.c Adjust uipc_rcvd, uipc_send, and uipc_sense to use the SB_STOP backpressure mechanism. Removing obsolete unpcb fields from db_show_unpcb. tests/sys/kern/unix_seqpacket_test.c Clear expected failures from ATF. Obtained from: DragonFly BSD PR: kern/185812 Reviewed by: silence from freebsd-net@ and rwatson@ MFC after: 3 weeks Sponsored by: Spectra Logic Corporation
2014-03-13 18:42:12 +00:00
db_printf("unp_gencnt: %llu\n",
Add DDB "show unpcb" command, allowing DDB to print out many pertinent details from UNIX domain socket protocol layer state.
2007-05-29 12:36:00 +00:00
(unsigned long long)unp->unp_gencnt);
db_printf("unp_flags: %x (", unp->unp_flags);
db_print_unpflags(unp->unp_flags);
db_printf(")\n");
db_printf("unp_peercred:\n");
db_print_xucred(2, &unp->unp_peercred);
db_printf("unp_refcount: %u\n", unp->unp_refcount);
}
#endif
Reference in New Issue Copy Permalink