freebsd-dev/sys/kern/kern_fork.c

823 lines
20 KiB
C
Raw Normal View History

/*-
1994-05-24 10:09:53 +00:00
* Copyright (c) 1982, 1986, 1989, 1991, 1993
* The Regents of the University of California. All rights reserved.
* (c) UNIX System Laboratories, Inc.
* All or some portions of this file are derived from material licensed
* to the University of California by American Telephone and Telegraph
* Co. or Unix System Laboratories, Inc. and are reproduced herein with
* the permission of UNIX System Laboratories, Inc.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
* 4. Neither the name of the University nor the names of its contributors
* may be used to endorse or promote products derived from this software
* without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
* SUCH DAMAGE.
*
* @(#)kern_fork.c 8.6 (Berkeley) 4/8/94
*/
2003-06-11 00:56:59 +00:00
#include <sys/cdefs.h>
__FBSDID("$FreeBSD$");
#include "opt_ktrace.h"
#include "opt_mac.h"
1994-05-24 10:09:53 +00:00
#include <sys/param.h>
#include <sys/systm.h>
#include <sys/sysproto.h>
#include <sys/eventhandler.h>
1994-05-24 10:09:53 +00:00
#include <sys/filedesc.h>
#include <sys/kernel.h>
#include <sys/kthread.h>
#include <sys/sysctl.h>
Rework the witness code to work with sx locks as well as mutexes. - Introduce lock classes and lock objects. Each lock class specifies a name and set of flags (or properties) shared by all locks of a given type. Currently there are three lock classes: spin mutexes, sleep mutexes, and sx locks. A lock object specifies properties of an additional lock along with a lock name and all of the extra stuff needed to make witness work with a given lock. This abstract lock stuff is defined in sys/lock.h. The lockmgr constants, types, and prototypes have been moved to sys/lockmgr.h. For temporary backwards compatability, sys/lock.h includes sys/lockmgr.h. - Replace proc->p_spinlocks with a per-CPU list, PCPU(spinlocks), of spin locks held. By making this per-cpu, we do not have to jump through magic hoops to deal with sched_lock changing ownership during context switches. - Replace proc->p_heldmtx, formerly a list of held sleep mutexes, with proc->p_sleeplocks, which is a list of held sleep locks including sleep mutexes and sx locks. - Add helper macros for logging lock events via the KTR_LOCK KTR logging level so that the log messages are consistent. - Add some new flags that can be passed to mtx_init(): - MTX_NOWITNESS - specifies that this lock should be ignored by witness. This is used for the mutex that blocks a sx lock for example. - MTX_QUIET - this is not new, but you can pass this to mtx_init() now and no events will be logged for this lock, so that one doesn't have to change all the individual mtx_lock/unlock() operations. - All lock objects maintain an initialized flag. Use this flag to export a mtx_initialized() macro that can be safely called from drivers. Also, we on longer walk the all_mtx list if MUTEX_DEBUG is defined as witness performs the corresponding checks using the initialized flag. - The lock order reversal messages have been improved to output slightly more accurate file and line numbers.
2001-03-28 09:03:24 +00:00
#include <sys/lock.h>
1994-05-24 10:09:53 +00:00
#include <sys/malloc.h>
#include <sys/mutex.h>
#include <sys/priv.h>
1994-05-24 10:09:53 +00:00
#include <sys/proc.h>
#include <sys/pioctl.h>
1994-05-24 10:09:53 +00:00
#include <sys/resourcevar.h>
#include <sys/sched.h>
#include <sys/syscall.h>
#include <sys/vmmeter.h>
1994-05-24 10:09:53 +00:00
#include <sys/vnode.h>
#include <sys/acct.h>
#include <sys/ktr.h>
1994-05-24 10:09:53 +00:00
#include <sys/ktrace.h>
#include <sys/unistd.h>
#include <sys/sx.h>
#include <sys/signalvar.h>
1994-05-24 10:09:53 +00:00
#include <security/audit/audit.h>
#include <security/mac/mac_framework.h>
#include <vm/vm.h>
#include <vm/pmap.h>
#include <vm/vm_map.h>
#include <vm/vm_extern.h>
#include <vm/uma.h>
#ifndef _SYS_SYSPROTO_H_
struct fork_args {
int dummy;
};
#endif
1994-05-24 10:09:53 +00:00
/* ARGSUSED */
int
fork(td, uap)
struct thread *td;
1994-05-24 10:09:53 +00:00
struct fork_args *uap;
{
int error;
struct proc *p2;
error = fork1(td, RFFDG | RFPROC, 0, &p2);
if (error == 0) {
td->td_retval[0] = p2->p_pid;
td->td_retval[1] = 0;
}
return (error);
1994-05-24 10:09:53 +00:00
}
/* ARGSUSED */
int
vfork(td, uap)
struct thread *td;
struct vfork_args *uap;
1994-05-24 10:09:53 +00:00
{
int error;
struct proc *p2;
error = fork1(td, RFFDG | RFPROC | RFPPWAIT | RFMEM, 0, &p2);
if (error == 0) {
td->td_retval[0] = p2->p_pid;
td->td_retval[1] = 0;
}
return (error);
}
1994-05-24 10:09:53 +00:00
int
rfork(td, uap)
struct thread *td;
struct rfork_args *uap;
{
struct proc *p2;
int error;
/* Don't allow kernel-only flags. */
if ((uap->flags & RFKERNELONLY) != 0)
return (EINVAL);
AUDIT_ARG(fflags, uap->flags);
error = fork1(td, uap->flags, 0, &p2);
if (error == 0) {
td->td_retval[0] = p2 ? p2->p_pid : 0;
td->td_retval[1] = 0;
}
return (error);
1994-05-24 10:09:53 +00:00
}
int nprocs = 1; /* process 0 */
int lastpid = 0;
SYSCTL_INT(_kern, OID_AUTO, lastpid, CTLFLAG_RD, &lastpid, 0,
"Last used PID");
1994-05-24 10:09:53 +00:00
/*
* Random component to lastpid generation. We mix in a random factor to make
* it a little harder to predict. We sanity check the modulus value to avoid
* doing it in critical paths. Don't let it be too small or we pointlessly
* waste randomness entropy, and don't let it be impossibly large. Using a
* modulus that is too big causes a LOT more process table scans and slows
* down fork processing as the pidchecked caching is defeated.
*/
static int randompid = 0;
static int
sysctl_kern_randompid(SYSCTL_HANDLER_ARGS)
{
int error, pid;
error = sysctl_wire_old_buffer(req, sizeof(int));
if (error != 0)
return(error);
sx_xlock(&allproc_lock);
pid = randompid;
error = sysctl_handle_int(oidp, &pid, 0, req);
if (error == 0 && req->newptr != NULL) {
if (pid < 0 || pid > PID_MAX - 100) /* out of range */
pid = PID_MAX - 100;
else if (pid < 2) /* NOP */
pid = 0;
else if (pid < 100) /* Make it reasonable */
pid = 100;
randompid = pid;
}
sx_xunlock(&allproc_lock);
return (error);
}
SYSCTL_PROC(_kern, OID_AUTO, randompid, CTLTYPE_INT|CTLFLAG_RW,
0, 0, sysctl_kern_randompid, "I", "Random PID modulus");
int
fork1(td, flags, pages, procp)
struct thread *td;
int flags;
int pages;
struct proc **procp;
1994-05-24 10:09:53 +00:00
{
struct proc *p1, *p2, *pptr;
1994-05-24 10:09:53 +00:00
struct proc *newproc;
int ok, trypid;
static int curfail, pidchecked = 0;
static struct timeval lastfail;
struct filedesc *fd;
struct filedesc_to_leader *fdtol;
struct thread *td2;
struct sigacts *newsigacts;
struct vmspace *vm2;
int error;
/* Can't copy and clear. */
if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG))
return (EINVAL);
1994-05-24 10:09:53 +00:00
p1 = td->td_proc;
/*
* Here we don't create a new process, but we divorce
* certain parts of a process from itself.
*/
if ((flags & RFPROC) == 0) {
Take out the single-threading code in fork. After discussions with jeff, alc, (various Ironport people), david Xu, and mostly Alfred (who found the problem) it has been demonstrated that this is not needed for our implementations of threads and represents a real (as in we've seen it happen a lot) deadlock danger. Several points: Since forking multiple threads is not allowed, and posix states that any mutexes owned by othre threads wilol be owned in the child by phantom threads, and therads shouldn't ba accessing shared structures without protection, It can be proved that if this leads to the child process accessing inconsistent data, it's a programming error. The mode of thread_single() being used in fork() is the wrong one. It is using SINGLE_NO_EXIT when it should be using SINGLE_BOUNDARY. Even if this we used, System processes have no need to do it as they have no userland to get inconsistent. This commmit first fixes the above bugs to get tehm correct in CVS. then removes them with #ifdef. This is so that history contains the corrected version should it be needed in the future. This code may be needed if we implement the forkall() syscall from Solaris. It may be needed for other non-posix thread libraries at some time in the future, so let the code sit for a short while while I do some work on it anyhow. This removes a reproducible lockup in NFS. It may be argued that maybe doing a fork while holding a vnode lock may not be the best idea in th efirst place but it shouldn't cause a deadlock. The removal has been running under soak test for several days now. This removal should be seriously considered for 7.0 and RELENG_6. Note. There is code in the core-dumping code that may have a similar problem with coredumping threaded processes MFC After: 4 days
2007-10-23 17:54:15 +00:00
if (((p1->p_flag & (P_HADTHREADS|P_SYSTEM)) == P_HADTHREADS) &&
(flags & (RFCFDG | RFFDG))) {
PROC_LOCK(p1);
if (thread_single(SINGLE_BOUNDARY)) {
PROC_UNLOCK(p1);
return (ERESTART);
}
PROC_UNLOCK(p1);
}
error = vm_forkproc(td, NULL, NULL, NULL, flags);
if (error)
goto norfproc_fail;
/*
* Close all file descriptors.
*/
if (flags & RFCFDG) {
struct filedesc *fdtmp;
fdtmp = fdinit(td->td_proc->p_fd);
fdfree(td);
p1->p_fd = fdtmp;
}
/*
* Unshare file descriptors (from parent).
*/
if (flags & RFFDG)
fdunshare(p1, td);
norfproc_fail:
Take out the single-threading code in fork. After discussions with jeff, alc, (various Ironport people), david Xu, and mostly Alfred (who found the problem) it has been demonstrated that this is not needed for our implementations of threads and represents a real (as in we've seen it happen a lot) deadlock danger. Several points: Since forking multiple threads is not allowed, and posix states that any mutexes owned by othre threads wilol be owned in the child by phantom threads, and therads shouldn't ba accessing shared structures without protection, It can be proved that if this leads to the child process accessing inconsistent data, it's a programming error. The mode of thread_single() being used in fork() is the wrong one. It is using SINGLE_NO_EXIT when it should be using SINGLE_BOUNDARY. Even if this we used, System processes have no need to do it as they have no userland to get inconsistent. This commmit first fixes the above bugs to get tehm correct in CVS. then removes them with #ifdef. This is so that history contains the corrected version should it be needed in the future. This code may be needed if we implement the forkall() syscall from Solaris. It may be needed for other non-posix thread libraries at some time in the future, so let the code sit for a short while while I do some work on it anyhow. This removes a reproducible lockup in NFS. It may be argued that maybe doing a fork while holding a vnode lock may not be the best idea in th efirst place but it shouldn't cause a deadlock. The removal has been running under soak test for several days now. This removal should be seriously considered for 7.0 and RELENG_6. Note. There is code in the core-dumping code that may have a similar problem with coredumping threaded processes MFC After: 4 days
2007-10-23 17:54:15 +00:00
if (((p1->p_flag & (P_HADTHREADS|P_SYSTEM)) == P_HADTHREADS) &&
(flags & (RFCFDG | RFFDG))) {
PROC_LOCK(p1);
thread_single_end();
PROC_UNLOCK(p1);
}
*procp = NULL;
return (error);
}
/*
* XXX
* We did have single-threading code here
* however it proved un-needed and caused problems
*/
/* Allocate new proc. */
newproc = uma_zalloc(proc_zone, M_WAITOK);
if (TAILQ_EMPTY(&newproc->p_threads)) {
td2 = thread_alloc();
if (td2 == NULL) {
error = ENOMEM;
goto fail1;
}
proc_linkup(newproc, td2);
sched_newproc(newproc, td2);
} else
td2 = FIRST_THREAD_IN_PROC(newproc);
/* Allocate and switch to an alternate kstack if specified. */
if (pages != 0) {
if (!vm_thread_new_altkstack(td2, pages)) {
error = ENOMEM;
goto fail1;
}
}
if ((flags & RFMEM) == 0) {
vm2 = vmspace_fork(p1->p_vmspace);
if (vm2 == NULL) {
error = ENOMEM;
goto fail1;
}
} else
vm2 = NULL;
#ifdef MAC
mac_proc_init(newproc);
#endif
knlist_init(&newproc->p_klist, &newproc->p_mtx, NULL, NULL, NULL);
Moderate rewrite of kernel ktrace code to attempt to generally improve reliability when tracing fast-moving processes or writing traces to slow file systems by avoiding unbounded queueuing and dropped records. Record loss was previously possible when the global pool of records become depleted as a result of record generation outstripping record commit, which occurred quickly in many common situations. These changes partially restore the 4.x model of committing ktrace records at the point of trace generation (synchronous), but maintain the 5.x deferred record commit behavior (asynchronous) for situations where entering VFS and sleeping is not possible (i.e., in the scheduler). Records are now queued per-process as opposed to globally, with processes responsible for committing records from their own context as required. - Eliminate the ktrace worker thread and global record queue, as they are no longer used. Keep the global free record list, as records are still used. - Add a per-process record queue, which will hold any asynchronously generated records, such as from context switches. This replaces the global queue as the place to submit asynchronous records to. - When a record is committed asynchronously, simply queue it to the process. - When a record is committed synchronously, first drain any pending per-process records in order to maintain ordering as best we can. Currently ordering between competing threads is provided via a global ktrace_sx, but a per-process flag or lock may be desirable in the future. - When a process returns to user space following a system call, trap, signal delivery, etc, flush any pending records. - When a process exits, flush any pending records. - Assert on process tear-down that there are no pending records. - Slightly abstract the notion of being "in ktrace", which is used to prevent the recursive generation of records, as well as generating traces for ktrace events. Future work here might look at changing the set of events marked for synchronous and asynchronous record generation, re-balancing queue depth, timeliness of commit to disk, and so on. I.e., performing a drain every (n) records. MFC after: 1 month Discussed with: jhb Requested by: Marc Olzheim <marcolz at stack dot nl>
2005-11-13 13:27:44 +00:00
STAILQ_INIT(&newproc->p_ktr);
/* We have to lock the process tree while we look for a pid. */
sx_slock(&proctree_lock);
1994-05-24 10:09:53 +00:00
/*
* Although process entries are dynamically created, we still keep
* a global limit on the maximum number we will create. Don't allow
* a nonprivileged user to use the last ten processes; don't let root
1994-05-24 10:09:53 +00:00
* exceed the limit. The variable nprocs is the current number of
* processes, maxproc is the limit.
*/
sx_xlock(&allproc_lock);
if ((nprocs >= maxproc - 10 && priv_check_cred(td->td_ucred,
PRIV_MAXPROC, 0) != 0) || nprocs >= maxproc) {
error = EAGAIN;
goto fail;
1994-05-24 10:09:53 +00:00
}
1994-05-24 10:09:53 +00:00
/*
* Increment the count of procs running with this uid. Don't allow
* a nonprivileged user to exceed their current limit.
*
* XXXRW: Can we avoid privilege here if it's not needed?
1994-05-24 10:09:53 +00:00
*/
error = priv_check_cred(td->td_ucred, PRIV_PROC_LIMIT, 0);
if (error == 0)
ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1, 0);
else {
PROC_LOCK(p1);
ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1,
lim_cur(p1, RLIMIT_NPROC));
PROC_UNLOCK(p1);
}
if (!ok) {
error = EAGAIN;
goto fail;
1994-05-24 10:09:53 +00:00
}
/*
* Increment the nprocs resource before blocking can occur. There
* are hard-limits as to the number of processes that can run.
*/
nprocs++;
1994-05-24 10:09:53 +00:00
/*
* Find an unused process ID. We remember a range of unused IDs
* ready to use (from lastpid+1 through pidchecked-1).
*
* If RFHIGHPID is set (used during system boot), do not allocate
* low-numbered pids.
1994-05-24 10:09:53 +00:00
*/
trypid = lastpid + 1;
if (flags & RFHIGHPID) {
if (trypid < 10)
trypid = 10;
} else {
if (randompid)
trypid += arc4random() % randompid;
}
1994-05-24 10:09:53 +00:00
retry:
/*
* If the process ID prototype has wrapped around,
* restart somewhat above 0, as the low-numbered procs
* tend to include daemons that don't exit.
*/
if (trypid >= PID_MAX) {
trypid = trypid % PID_MAX;
if (trypid < 100)
trypid += 100;
1994-05-24 10:09:53 +00:00
pidchecked = 0;
}
if (trypid >= pidchecked) {
1994-05-24 10:09:53 +00:00
int doingzomb = 0;
pidchecked = PID_MAX;
/*
* Scan the active and zombie procs to check whether this pid
* is in use. Remember the lowest pid that's greater
* than trypid, so we can avoid checking for a while.
1994-05-24 10:09:53 +00:00
*/
p2 = LIST_FIRST(&allproc);
1994-05-24 10:09:53 +00:00
again:
for (; p2 != NULL; p2 = LIST_NEXT(p2, p_list)) {
while (p2->p_pid == trypid ||
(p2->p_pgrp != NULL &&
(p2->p_pgrp->pg_id == trypid ||
(p2->p_session != NULL &&
p2->p_session->s_sid == trypid)))) {
trypid++;
if (trypid >= pidchecked)
1994-05-24 10:09:53 +00:00
goto retry;
}
if (p2->p_pid > trypid && pidchecked > p2->p_pid)
1994-05-24 10:09:53 +00:00
pidchecked = p2->p_pid;
if (p2->p_pgrp != NULL) {
if (p2->p_pgrp->pg_id > trypid &&
pidchecked > p2->p_pgrp->pg_id)
pidchecked = p2->p_pgrp->pg_id;
if (p2->p_session != NULL &&
p2->p_session->s_sid > trypid &&
pidchecked > p2->p_session->s_sid)
pidchecked = p2->p_session->s_sid;
}
1994-05-24 10:09:53 +00:00
}
if (!doingzomb) {
doingzomb = 1;
p2 = LIST_FIRST(&zombproc);
1994-05-24 10:09:53 +00:00
goto again;
}
}
sx_sunlock(&proctree_lock);
1994-05-24 10:09:53 +00:00
/*
* RFHIGHPID does not mess with the lastpid counter during boot.
*/
if (flags & RFHIGHPID)
pidchecked = 0;
else
lastpid = trypid;
p2 = newproc;
p2->p_state = PRS_NEW; /* protect against others */
p2->p_pid = trypid;
/*
* Allow the scheduler to initialize the child.
*/
thread_lock(td);
sched_fork(td, td2);
thread_unlock(td);
AUDIT_ARG(pid, p2->p_pid);
LIST_INSERT_HEAD(&allproc, p2, p_list);
LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash);
Close race conditions between fork() and [sg]etpriority()'s PRIO_USER case, possibly also other places that deferences p_ucred. In the past, we insert a new process into the allproc list right after PID allocation, and release the allproc_lock sx. Because most content in new proc's structure is not yet initialized, this could lead to undefined result if we do not handle PRS_NEW with care. The problem with PRS_NEW state is that it does not provide fine grained information about how much initialization is done for a new process. By defination, after PRIO_USER setpriority(), all processes that belongs to given user should have their nice value set to the specified value. Therefore, if p_{start,end}copy section was done for a PRS_NEW process, we can not safely ignore it because p_nice is in this area. On the other hand, we should be careful on PRS_NEW processes because we do not allow non-root users to lower their nice values, and without a successful copy of the copy section, we can get stale values that is inherted from the uninitialized area of the process structure. This commit tries to close the race condition by grabbing proc mutex *before* we release allproc_lock xlock, and do copy as well as zero immediately after the allproc_lock xunlock. This guarantees that the new process would have its p_copy and p_zero sections, as well as user credential informaion initialized. In getpriority() case, instead of grabbing PROC_LOCK for a PRS_NEW process, we just skip the process in question, because it does not affect the final result of the call, as the p_nice value would be copied from its parent, and we will see it during allproc traverse. Other potential solutions are still under evaluation. Discussed with: davidxu, jhb, rwatson PR: kern/108071 MFC after: 2 weeks
2007-02-26 03:38:09 +00:00
PROC_LOCK(p2);
PROC_LOCK(p1);
sx_xunlock(&allproc_lock);
Close race conditions between fork() and [sg]etpriority()'s PRIO_USER case, possibly also other places that deferences p_ucred. In the past, we insert a new process into the allproc list right after PID allocation, and release the allproc_lock sx. Because most content in new proc's structure is not yet initialized, this could lead to undefined result if we do not handle PRS_NEW with care. The problem with PRS_NEW state is that it does not provide fine grained information about how much initialization is done for a new process. By defination, after PRIO_USER setpriority(), all processes that belongs to given user should have their nice value set to the specified value. Therefore, if p_{start,end}copy section was done for a PRS_NEW process, we can not safely ignore it because p_nice is in this area. On the other hand, we should be careful on PRS_NEW processes because we do not allow non-root users to lower their nice values, and without a successful copy of the copy section, we can get stale values that is inherted from the uninitialized area of the process structure. This commit tries to close the race condition by grabbing proc mutex *before* we release allproc_lock xlock, and do copy as well as zero immediately after the allproc_lock xunlock. This guarantees that the new process would have its p_copy and p_zero sections, as well as user credential informaion initialized. In getpriority() case, instead of grabbing PROC_LOCK for a PRS_NEW process, we just skip the process in question, because it does not affect the final result of the call, as the p_nice value would be copied from its parent, and we will see it during allproc traverse. Other potential solutions are still under evaluation. Discussed with: davidxu, jhb, rwatson PR: kern/108071 MFC after: 2 weeks
2007-02-26 03:38:09 +00:00
bcopy(&p1->p_startcopy, &p2->p_startcopy,
__rangeof(struct proc, p_startcopy, p_endcopy));
PROC_UNLOCK(p1);
bzero(&p2->p_startzero,
__rangeof(struct proc, p_startzero, p_endzero));
p2->p_ucred = crhold(td->td_ucred);
PROC_UNLOCK(p2);
/*
* Malloc things while we don't hold any locks.
*/
if (flags & RFSIGSHARE)
newsigacts = NULL;
else
newsigacts = sigacts_alloc();
/*
* Copy filedesc.
*/
if (flags & RFCFDG) {
fd = fdinit(p1->p_fd);
fdtol = NULL;
} else if (flags & RFFDG) {
fd = fdcopy(p1->p_fd);
fdtol = NULL;
} else {
fd = fdshare(p1->p_fd);
if (p1->p_fdtol == NULL)
p1->p_fdtol =
filedesc_to_leader_alloc(NULL,
NULL,
p1->p_leader);
if ((flags & RFTHREAD) != 0) {
/*
* Shared file descriptor table and
* shared process leaders.
*/
fdtol = p1->p_fdtol;
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
FILEDESC_XLOCK(p1->p_fd);
fdtol->fdl_refcount++;
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
FILEDESC_XUNLOCK(p1->p_fd);
} else {
/*
* Shared file descriptor table, and
* different process leaders
*/
fdtol = filedesc_to_leader_alloc(p1->p_fdtol,
p1->p_fd,
p2);
}
}
1994-05-24 10:09:53 +00:00
/*
* Make a proc table entry for the new process.
* Start by zeroing the section of proc that is zero-initialized,
* then copy the section that is copied directly from the parent.
*/
PROC_LOCK(p2);
PROC_LOCK(p1);
bzero(&td2->td_startzero,
__rangeof(struct thread, td_startzero, td_endzero));
bcopy(&td->td_startcopy, &td2->td_startcopy,
__rangeof(struct thread, td_startcopy, td_endcopy));
1994-05-24 10:09:53 +00:00
bcopy(&p2->p_comm, &td2->td_name, sizeof(td2->td_name));
td2->td_sigstk = td->td_sigstk;
td2->td_sigmask = td->td_sigmask;
td2->td_flags = TDF_INMEM;
1994-05-24 10:09:53 +00:00
/*
* Duplicate sub-structures as needed.
* Increase reference counts on shared objects.
*/
p2->p_flag = P_INMEM;
p2->p_swtick = ticks;
if (p1->p_flag & P_PROFIL)
startprofclock(p2);
td2->td_ucred = crhold(p2->p_ucred);
pargs_hold(p2->p_args);
if (flags & RFSIGSHARE) {
p2->p_sigacts = sigacts_hold(p1->p_sigacts);
} else {
sigacts_copy(newsigacts, p1->p_sigacts);
p2->p_sigacts = newsigacts;
}
if (flags & RFLINUXTHPN)
p2->p_sigparent = SIGUSR1;
else
p2->p_sigparent = SIGCHLD;
1994-05-24 10:09:53 +00:00
p2->p_textvp = p1->p_textvp;
p2->p_fd = fd;
p2->p_fdtol = fdtol;
1994-05-24 10:09:53 +00:00
/*
* p_limit is copy-on-write. Bump its refcount.
1994-05-24 10:09:53 +00:00
*/
lim_fork(p1, p2);
pstats_fork(p1->p_stats, p2->p_stats);
PROC_UNLOCK(p1);
PROC_UNLOCK(p2);
1994-05-24 10:09:53 +00:00
/* Bump references to the text vnode (for procfs) */
if (p2->p_textvp)
vref(p2->p_textvp);
/*
* Set up linkage for kernel based threading.
*/
if ((flags & RFTHREAD) != 0) {
mtx_lock(&ppeers_lock);
p2->p_peers = p1->p_peers;
p1->p_peers = p2;
p2->p_leader = p1->p_leader;
mtx_unlock(&ppeers_lock);
PROC_LOCK(p1->p_leader);
if ((p1->p_leader->p_flag & P_WEXIT) != 0) {
PROC_UNLOCK(p1->p_leader);
/*
* The task leader is exiting, so process p1 is
* going to be killed shortly. Since p1 obviously
* isn't dead yet, we know that the leader is either
* sending SIGKILL's to all the processes in this
* task or is sleeping waiting for all the peers to
* exit. We let p1 complete the fork, but we need
* to go ahead and kill the new process p2 since
* the task leader may not get a chance to send
* SIGKILL to it. We leave it on the list so that
* the task leader will wait for this new process
* to commit suicide.
*/
PROC_LOCK(p2);
psignal(p2, SIGKILL);
PROC_UNLOCK(p2);
} else
PROC_UNLOCK(p1->p_leader);
} else {
p2->p_peers = NULL;
p2->p_leader = p2;
}
sx_xlock(&proctree_lock);
PGRP_LOCK(p1->p_pgrp);
PROC_LOCK(p2);
PROC_LOCK(p1);
/*
* Preserve some more flags in subprocess. P_PROFIL has already
* been preserved.
*/
p2->p_flag |= p1->p_flag & P_SUGID;
td2->td_pflags |= td->td_pflags & TDP_ALTSTACK;
SESS_LOCK(p1->p_session);
1994-05-24 10:09:53 +00:00
if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT)
p2->p_flag |= P_CONTROLT;
SESS_UNLOCK(p1->p_session);
if (flags & RFPPWAIT)
1994-05-24 10:09:53 +00:00
p2->p_flag |= P_PPWAIT;
p2->p_pgrp = p1->p_pgrp;
LIST_INSERT_AFTER(p1, p2, p_pglist);
2002-04-02 17:12:06 +00:00
PGRP_UNLOCK(p1->p_pgrp);
LIST_INIT(&p2->p_children);
callout_init(&p2->p_itcallout, CALLOUT_MPSAFE);
1994-05-24 10:09:53 +00:00
#ifdef KTRACE
/*
* Copy traceflag and tracefile if enabled.
1994-05-24 10:09:53 +00:00
*/
mtx_lock(&ktrace_mtx);
KASSERT(p2->p_tracevp == NULL, ("new process has a ktrace vnode"));
if (p1->p_traceflag & KTRFAC_INHERIT) {
1994-05-24 10:09:53 +00:00
p2->p_traceflag = p1->p_traceflag;
if ((p2->p_tracevp = p1->p_tracevp) != NULL) {
VREF(p2->p_tracevp);
KASSERT(p1->p_tracecred != NULL,
("ktrace vnode with no cred"));
p2->p_tracecred = crhold(p1->p_tracecred);
}
1994-05-24 10:09:53 +00:00
}
mtx_unlock(&ktrace_mtx);
1994-05-24 10:09:53 +00:00
#endif
/*
* If PF_FORK is set, the child process inherits the
* procfs ioctl flags from its parent.
*/
if (p1->p_pfsflags & PF_FORK) {
p2->p_stops = p1->p_stops;
p2->p_pfsflags = p1->p_pfsflags;
}
1994-05-24 10:09:53 +00:00
/*
* This begins the section where we must prevent the parent
* from being swapped.
1994-05-24 10:09:53 +00:00
*/
_PHOLD(p1);
PROC_UNLOCK(p1);
/*
* Attach the new process to its parent.
*
* If RFNOWAIT is set, the newly created process becomes a child
* of init. This effectively disassociates the child from the
* parent.
*/
if (flags & RFNOWAIT)
pptr = initproc;
else
pptr = p1;
p2->p_pptr = pptr;
LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling);
sx_xunlock(&proctree_lock);
/* Inform accounting that we have forked. */
p2->p_acflag = AFORK;
PROC_UNLOCK(p2);
/*
The biggie: Get rid of the UPAGES from the top of the per-process address space. (!) Have each process use the kernel stack and pcb in the kvm space. Since the stacks are at a different address, we cannot copy the stack at fork() and allow the child to return up through the function call tree to return to user mode - create a new execution context and have the new process begin executing from cpu_switch() and go to user mode directly. In theory this should speed up fork a bit. Context switch the tss_esp0 pointer in the common tss. This is a lot simpler since than swithching the gdt[GPROC0_SEL].sd.sd_base pointer to each process's tss since the esp0 pointer is a 32 bit pointer, and the sd_base setting is split into three different bit sections at non-aligned boundaries and requires a lot of twiddling to reset. The 8K of memory at the top of the process space is now empty, and unmapped (and unmappable, it's higher than VM_MAXUSER_ADDRESS). Simplity the pmap code to manage process contexts, we no longer have to double map the UPAGES, this simplifies and should measuably speed up fork(). The following parts came from John Dyson: Set PG_G on the UPAGES that are now in kernel context, and invalidate them when swapping them out. Move the upages object (upobj) from the vmspace to the proc structure. Now that the UPAGES (pcb and kernel stack) are out of user space, make rfork(..RFMEM..) do what was intended by sharing the vmspace entirely via reference counting rather than simply inheriting the mappings.
1997-04-07 07:16:06 +00:00
* Finish creating the child process. It will return via a different
* execution path later. (ie: directly into user mode)
*/
vm_forkproc(td, p2, td2, vm2, flags);
1994-05-24 10:09:53 +00:00
if (flags == (RFFDG | RFPROC)) {
PCPU_INC(cnt.v_forks);
PCPU_ADD(cnt.v_forkpages, p2->p_vmspace->vm_dsize +
p2->p_vmspace->vm_ssize);
} else if (flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM)) {
PCPU_INC(cnt.v_vforks);
PCPU_ADD(cnt.v_vforkpages, p2->p_vmspace->vm_dsize +
p2->p_vmspace->vm_ssize);
} else if (p1 == &proc0) {
PCPU_INC(cnt.v_kthreads);
PCPU_ADD(cnt.v_kthreadpages, p2->p_vmspace->vm_dsize +
p2->p_vmspace->vm_ssize);
} else {
PCPU_INC(cnt.v_rforks);
PCPU_ADD(cnt.v_rforkpages, p2->p_vmspace->vm_dsize +
p2->p_vmspace->vm_ssize);
}
/*
* Both processes are set up, now check if any loadable modules want
* to adjust anything.
* What if they have an error? XXX
*/
EVENTHANDLER_INVOKE(process_fork, p1, p2, flags);
/*
* Set the child start time and mark the process as being complete.
*/
microuptime(&p2->p_stats->p_start);
PROC_SLOCK(p2);
p2->p_state = PRS_NORMAL;
PROC_SUNLOCK(p2);
1994-05-24 10:09:53 +00:00
/*
* If RFSTOPPED not requested, make child runnable and add to
* run queue.
1994-05-24 10:09:53 +00:00
*/
if ((flags & RFSTOPPED) == 0) {
thread_lock(td2);
TD_SET_CAN_RUN(td2);
sched_add(td2, SRQ_BORING);
thread_unlock(td2);
}
1994-05-24 10:09:53 +00:00
/*
* Now can be swapped.
*/
PROC_LOCK(p1);
_PRELE(p1);
1994-05-24 10:09:53 +00:00
/*
* Tell any interested parties about the new process.
*/
KNOTE_LOCKED(&p1->p_klist, NOTE_FORK | p2->p_pid);
PROC_UNLOCK(p1);
1994-05-24 10:09:53 +00:00
/*
* Preserve synchronization semantics of vfork. If waiting for
* child to exec or exit, set P_PPWAIT on child, and sleep on our
* proc (in case of exit).
*/
PROC_LOCK(p2);
while (p2->p_flag & P_PPWAIT)
msleep(p1, &p2->p_mtx, PWAIT, "ppwait", 0);
PROC_UNLOCK(p2);
1994-05-24 10:09:53 +00:00
/*
* Return child proc pointer to parent.
1994-05-24 10:09:53 +00:00
*/
*procp = p2;
1994-05-24 10:09:53 +00:00
return (0);
fail:
sx_sunlock(&proctree_lock);
if (ppsratecheck(&lastfail, &curfail, 1))
printf("maxproc limit exceeded by uid %i, please see tuning(7) and login.conf(5).\n",
td->td_ucred->cr_ruid);
sx_xunlock(&allproc_lock);
#ifdef MAC
mac_proc_destroy(newproc);
#endif
fail1:
uma_zfree(proc_zone, newproc);
pause("fork", hz / 2);
return (error);
1994-05-24 10:09:53 +00:00
}
/*
* Handle the return of a child process from fork1(). This function
* is called from the MD fork_trampoline() entry point.
*/
void
fork_exit(callout, arg, frame)
void (*callout)(void *, struct trapframe *);
void *arg;
struct trapframe *frame;
{
struct proc *p;
struct thread *td;
struct thread *dtd;
td = curthread;
p = td->td_proc;
KASSERT(p->p_state == PRS_NORMAL, ("executing process is still new"));
CTR4(KTR_PROC, "fork_exit: new thread %p (kse %p, pid %d, %s)",
td, td->td_sched, p->p_pid, td->td_name);
sched_fork_exit(td);
/*
* Processes normally resume in mi_switch() after being
* cpu_switch()'ed to, but when children start up they arrive here
* instead, so we must do much the same things as mi_switch() would.
*/
if ((dtd = PCPU_GET(deadthread))) {
PCPU_SET(deadthread, NULL);
thread_stash(dtd);
}
thread_unlock(td);
/*
* cpu_set_fork_handler intercepts this function call to
* have this call a non-return function to stay in kernel mode.
* initproc has its own fork handler, but it does return.
*/
KASSERT(callout != NULL, ("NULL callout in fork_exit"));
callout(arg, frame);
/*
* Check if a kernel thread misbehaved and returned from its main
* function.
*/
if (p->p_flag & P_KTHREAD) {
printf("Kernel thread \"%s\" (pid %d) exited prematurely.\n",
td->td_name, p->p_pid);
kproc_exit(0);
}
mtx_assert(&Giant, MA_NOTOWNED);
EVENTHANDLER_INVOKE(schedtail, p);
}
/*
* Simplified back end of syscall(), used when returning from fork()
* directly into user mode. Giant is not held on entry, and must not
* be held on return. This function is passed in to fork_exit() as the
* first parameter and is called when returning to a new userland process.
*/
void
fork_return(td, frame)
struct thread *td;
struct trapframe *frame;
{
userret(td, frame);
#ifdef KTRACE
if (KTRPOINT(td, KTR_SYSRET))
ktrsysret(SYS_fork, 0, 0);
#endif
mtx_assert(&Giant, MA_NOTOWNED);
}