2 * Copyright (c) 1982, 1986, 1989, 1991, 1993
3 * The Regents of the University of California. All rights reserved.
4 * (c) UNIX System Laboratories, Inc.
5 * All or some portions of this file are derived from material licensed
6 * to the University of California by American Telephone and Telegraph
7 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8 * the permission of UNIX System Laboratories, Inc.
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 3. Neither the name of the University nor the names of its contributors
19 * may be used to endorse or promote products derived from this software
20 * without specific prior written permission.
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * @(#)kern_exit.c 8.7 (Berkeley) 2/12/94
35 * $FreeBSD: src/sys/kern/kern_exit.c,v 1.92.2.11 2003/01/13 22:51:16 dillon Exp $
38 #include "opt_ktrace.h"
40 #include <sys/param.h>
41 #include <sys/systm.h>
42 #include <sys/sysproto.h>
43 #include <sys/kernel.h>
44 #include <sys/malloc.h>
46 #include <sys/ktrace.h>
47 #include <sys/pioctl.h>
50 #include <sys/vnode.h>
51 #include <sys/resourcevar.h>
52 #include <sys/signalvar.h>
53 #include <sys/taskqueue.h>
54 #include <sys/ptrace.h>
55 #include <sys/acct.h> /* for acct_process() function prototype */
56 #include <sys/filedesc.h>
60 #include <sys/kern_syscall.h>
61 #include <sys/unistd.h>
62 #include <sys/eventhandler.h>
63 #include <sys/dsched.h>
66 #include <vm/vm_param.h>
69 #include <vm/vm_map.h>
70 #include <vm/vm_extern.h>
73 #include <sys/refcount.h>
74 #include <sys/thread2.h>
75 #include <sys/sysref2.h>
76 #include <sys/spinlock2.h>
77 #include <sys/mplock2.h>
79 #include <machine/vmm.h>
81 static void reaplwps(void *context
, int dummy
);
82 static void reaplwp(struct lwp
*lp
);
83 static void killlwps(struct lwp
*lp
);
85 static MALLOC_DEFINE(M_ATEXIT
, "atexit", "atexit callback");
88 * callout list for things to do at exit time
92 TAILQ_ENTRY(exitlist
) next
;
95 TAILQ_HEAD(exit_list_head
, exitlist
);
96 static struct exit_list_head exit_list
= TAILQ_HEAD_INITIALIZER(exit_list
);
101 static struct task
*deadlwp_task
[MAXCPU
];
102 static struct lwplist deadlwp_list
[MAXCPU
];
103 static struct lwkt_token deadlwp_token
[MAXCPU
];
109 * SYS_EXIT_ARGS(int rval)
112 sys_exit(struct exit_args
*uap
)
114 exit1(W_EXITCODE(uap
->rval
, 0));
120 * Death of a lwp or process with optional bells and whistles.
123 sys_extexit(struct extexit_args
*uap
)
125 struct proc
*p
= curproc
;
129 action
= EXTEXIT_ACTION(uap
->how
);
130 who
= EXTEXIT_WHO(uap
->how
);
132 /* Check parameters before we might perform some action */
145 error
= copyout(&uap
->status
, uap
->addr
, sizeof(uap
->status
));
153 lwkt_gettoken(&p
->p_token
);
158 * Be sure only to perform a simple lwp exit if there is at
159 * least one more lwp in the proc, which will call exit1()
160 * later, otherwise the proc will be an UNDEAD and not even a
163 if (p
->p_nthreads
> 1) {
164 lwp_exit(0, NULL
); /* called w/ p_token held */
167 /* else last lwp in proc: do the real thing */
169 default: /* to help gcc */
171 lwkt_reltoken(&p
->p_token
);
172 exit1(W_EXITCODE(uap
->status
, 0));
177 lwkt_reltoken(&p
->p_token
); /* safety */
181 * Kill all lwps associated with the current process except the
182 * current lwp. Return an error if we race another thread trying to
183 * do the same thing and lose the race.
185 * If forexec is non-zero the current thread and process flags are
186 * cleaned up so they can be reused.
188 * Caller must hold curproc->p_token
191 killalllwps(int forexec
)
193 struct lwp
*lp
= curthread
->td_lwp
;
194 struct proc
*p
= lp
->lwp_proc
;
198 * Interlock against P_WEXIT. Only one of the process's thread
199 * is allowed to do the master exit.
201 if (p
->p_flags
& P_WEXIT
)
203 p
->p_flags
|= P_WEXIT
;
206 * Set temporary stopped state in case we are racing a coredump.
207 * Otherwise the coredump may hang forever.
209 if (lp
->lwp_mpflags
& LWP_MP_WSTOP
) {
212 atomic_set_int(&lp
->lwp_mpflags
, LWP_MP_WSTOP
);
215 wakeup(&p
->p_nstopped
);
219 * Interlock with LWP_MP_WEXIT and kill any remaining LWPs
221 atomic_set_int(&lp
->lwp_mpflags
, LWP_MP_WEXIT
);
222 if (p
->p_nthreads
> 1)
226 * Undo temporary stopped state
229 atomic_clear_int(&lp
->lwp_mpflags
, LWP_MP_WSTOP
);
234 * If doing this for an exec, clean up the remaining thread
235 * (us) for continuing operation after all the other threads
239 atomic_clear_int(&lp
->lwp_mpflags
, LWP_MP_WEXIT
);
240 p
->p_flags
&= ~P_WEXIT
;
246 * Kill all LWPs except the current one. Do not try to signal
247 * LWPs which have exited on their own or have already been
251 killlwps(struct lwp
*lp
)
253 struct proc
*p
= lp
->lwp_proc
;
257 * Kill the remaining LWPs. We must send the signal before setting
258 * LWP_MP_WEXIT. The setting of WEXIT is optional but helps reduce
259 * races. tlp must be held across the call as it might block and
260 * allow the target lwp to rip itself out from under our loop.
262 FOREACH_LWP_IN_PROC(tlp
, p
) {
264 lwkt_gettoken(&tlp
->lwp_token
);
265 if ((tlp
->lwp_mpflags
& LWP_MP_WEXIT
) == 0) {
266 atomic_set_int(&tlp
->lwp_mpflags
, LWP_MP_WEXIT
);
267 lwpsignal(p
, tlp
, SIGKILL
);
269 lwkt_reltoken(&tlp
->lwp_token
);
274 * Wait for everything to clear out. Also make sure any tstop()s
275 * are signalled (we are holding p_token for the interlock).
278 while (p
->p_nthreads
> 1)
279 tsleep(&p
->p_nthreads
, 0, "killlwps", 0);
283 * Exit: deallocate address space and other resources, change proc state
284 * to zombie, and unlink proc from allproc and parent's lists. Save exit
285 * status and rusage for wait(). Check for child processes and orphan them.
290 struct thread
*td
= curthread
;
291 struct proc
*p
= td
->td_proc
;
292 struct lwp
*lp
= td
->td_lwp
;
296 struct sysreaper
*reap
;
302 lwkt_gettoken(&p
->p_token
);
305 kprintf("init died (signal %d, exit %d)\n",
306 WTERMSIG(rv
), WEXITSTATUS(rv
));
307 panic("Going nowhere without my init!");
309 varsymset_clean(&p
->p_varsymset
);
310 lockuninit(&p
->p_varsymset
.vx_lock
);
313 * Kill all lwps associated with the current process, return an
314 * error if we race another thread trying to do the same thing
317 error
= killalllwps(0);
323 /* are we a task leader? */
324 if (p
== p
->p_leader
) {
325 struct kill_args killArgs
;
326 killArgs
.signum
= SIGKILL
;
329 killArgs
.pid
= q
->p_pid
;
331 * The interface for kill is better
332 * than the internal signal
338 tsleep((caddr_t
)p
, 0, "exit1", 0);
344 STOPEVENT(p
, S_EXIT
, rv
);
345 p
->p_flags
|= P_POSTEXIT
; /* stop procfs stepping */
348 * Check if any loadable modules need anything done at process exit.
349 * e.g. SYSV IPC stuff
350 * XXX what if one of these generates an error?
355 * XXX: imho, the eventhandler stuff is much cleaner than this.
356 * Maybe we should move everything to use eventhandler.
358 TAILQ_FOREACH(ep
, &exit_list
, next
)
361 if (p
->p_flags
& P_PROFIL
)
364 SIGEMPTYSET(p
->p_siglist
);
365 SIGEMPTYSET(lp
->lwp_siglist
);
366 if (timevalisset(&p
->p_realtimer
.it_value
))
367 callout_stop_sync(&p
->p_ithandle
);
370 * Reset any sigio structures pointing to us as a result of
371 * F_SETOWN with our pid.
373 funsetownlst(&p
->p_sigiolst
);
376 * Close open files and release open-file table.
381 if (p
->p_leader
->p_peers
) {
383 while(q
->p_peers
!= p
)
385 q
->p_peers
= p
->p_peers
;
386 wakeup((caddr_t
)p
->p_leader
);
390 * XXX Shutdown SYSV semaphores
394 KKASSERT(p
->p_numposixlocks
== 0);
396 /* The next two chunks should probably be moved to vmspace_exit. */
400 * Clean up data related to virtual kernel operation. Clean up
401 * any vkernel context related to the current lwp now so we can
405 vkernel_lwp_exit(lp
);
410 * Release the user portion of address space. The exitbump prevents
411 * the vmspace from being completely eradicated (using holdcnt).
412 * This releases references to vnodes, which could cause I/O if the
413 * file has been unlinked. We need to do this early enough that
414 * we can still sleep.
416 * We can't free the entire vmspace as the kernel stack may be mapped
417 * within that space also.
419 * Processes sharing the same vmspace may exit in one order, and
420 * get cleaned up by vmspace_exit() in a different order. The
421 * last exiting process to reach this point releases as much of
422 * the environment as it can, and the last process cleaned up
423 * by vmspace_exit() (which decrements exitingcnt) cleans up the
426 * NOTE: Releasing p_token around this call is helpful if the
427 * vmspace had a huge RSS. Otherwise some other process
428 * trying to do an allproc or other scan (like 'ps') may
429 * stall for a long time.
431 lwkt_reltoken(&p
->p_token
);
433 lwkt_gettoken(&p
->p_token
);
435 if (SESS_LEADER(p
)) {
436 struct session
*sp
= p
->p_session
;
440 * We are the controlling process. Signal the
441 * foreground process group, drain the controlling
442 * terminal, and revoke access to the controlling
445 * NOTE: while waiting for the process group to exit
446 * it is possible that one of the processes in the
447 * group will revoke the tty, so the ttyclosesession()
448 * function will re-check sp->s_ttyvp.
450 if (sp
->s_ttyp
&& (sp
->s_ttyp
->t_session
== sp
)) {
451 if (sp
->s_ttyp
->t_pgrp
)
452 pgsignal(sp
->s_ttyp
->t_pgrp
, SIGHUP
, 1);
454 ttyclosesession(sp
, 1); /* also revoke */
457 * Release the tty. If someone has it open via
458 * /dev/tty then close it (since they no longer can
459 * once we've NULL'd it out).
461 ttyclosesession(sp
, 0);
464 * s_ttyp is not zero'd; we use this to indicate
465 * that the session once had a controlling terminal.
466 * (for logging and informational purposes)
471 fixjobc(p
, p
->p_pgrp
, 0);
472 (void)acct_process(p
);
478 ktrdestroy(&p
->p_tracenode
);
482 * Release reference to text vnode
484 if ((vtmp
= p
->p_textvp
) != NULL
) {
489 /* Release namecache handle to text file */
490 if (p
->p_textnch
.ncp
)
491 cache_drop(&p
->p_textnch
);
494 * We have to handle PPWAIT here or proc_move_allproc_zombie()
495 * will block on the PHOLD() the parent is doing.
497 * We are using the flag as an interlock so an atomic op is
498 * necessary to synchronize with the parent's cpu.
500 if (p
->p_flags
& P_PPWAIT
) {
501 if (p
->p_pptr
&& p
->p_pptr
->p_upmap
)
502 atomic_add_int(&p
->p_pptr
->p_upmap
->invfork
, -1);
503 atomic_clear_int(&p
->p_flags
, P_PPWAIT
);
508 * Move the process to the zombie list. This will block
509 * until the process p_lock count reaches 0. The process will
510 * not be reaped until TDF_EXITING is set by cpu_thread_exit(),
511 * which is called from cpu_proc_exit().
513 * Interlock against waiters using p_waitgen. We increment
514 * p_waitgen after completing the move of our process to the
517 * WARNING: pp becomes stale when we block, clear it now as a
520 proc_move_allproc_zombie(p
);
522 atomic_add_long(&pp
->p_waitgen
, 1);
526 * release controlled reaper for exit if we own it and return the
527 * remaining reaper (the one for us), which we will drop after we
530 reap
= reaper_exit(p
);
533 * Reparent all of this process's children to the init process or
534 * to the designated reaper. We must hold the reaper's p_token in
535 * order to safely mess with p_children.
537 * We already hold p->p_token (to remove the children from our list).
540 q
= LIST_FIRST(&p
->p_children
);
542 reproc
= reaper_get(reap
);
543 lwkt_gettoken(&reproc
->p_token
);
544 while ((q
= LIST_FIRST(&p
->p_children
)) != NULL
) {
546 lwkt_gettoken(&q
->p_token
);
547 if (q
!= LIST_FIRST(&p
->p_children
)) {
548 lwkt_reltoken(&q
->p_token
);
552 LIST_REMOVE(q
, p_sibling
);
553 LIST_INSERT_HEAD(&reproc
->p_children
, q
, p_sibling
);
555 q
->p_sigparent
= SIGCHLD
;
558 * Traced processes are killed
559 * since their existence means someone is screwing up.
561 if (q
->p_flags
& P_TRACED
) {
562 q
->p_flags
&= ~P_TRACED
;
565 lwkt_reltoken(&q
->p_token
);
568 lwkt_reltoken(&reproc
->p_token
);
573 * Save exit status and final rusage info, adding in child rusage
574 * info and self times.
576 calcru_proc(p
, &p
->p_ru
);
577 ruadd(&p
->p_ru
, &p
->p_cru
);
580 * notify interested parties of our demise.
582 KNOTE(&p
->p_klist
, NOTE_EXIT
);
585 * Notify parent that we're gone. If parent has the PS_NOCLDWAIT
586 * flag set, or if the handler is set to SIG_IGN, notify the reaper
587 * instead (it will handle this situation).
589 * NOTE: The reaper can still be the parent process.
593 if (p
->p_pptr
->p_sigacts
->ps_flag
& (PS_NOCLDWAIT
| PS_CLDSIGIGN
)) {
595 reproc
= reaper_get(reap
);
596 proc_reparent(p
, reproc
);
604 * Signal (possibly new) parent.
608 if (p
->p_sigparent
&& pp
!= initproc
) {
609 int sig
= p
->p_sigparent
;
611 if (sig
!= SIGUSR1
&& sig
!= SIGCHLD
)
615 ksignal(pp
, SIGCHLD
);
617 p
->p_flags
&= ~P_TRACED
;
621 * cpu_exit is responsible for clearing curproc, since
622 * it is heavily integrated with the thread/switching sequence.
624 * Other substructures are freed from wait().
629 * Finally, call machine-dependent code to release as many of the
630 * lwp's resources as we can and halt execution of this thread.
632 * pp is a wild pointer now but still the correct wakeup() target.
633 * lwp_exit() only uses it to send the wakeup() signal to the likely
634 * parent. Any reparenting race that occurs will get a signal
635 * automatically and not be an issue.
641 * Eventually called by every exiting LWP
643 * p->p_token must be held. mplock may be held and will be released.
646 lwp_exit(int masterexit
, void *waddr
)
648 struct thread
*td
= curthread
;
649 struct lwp
*lp
= td
->td_lwp
;
650 struct proc
*p
= lp
->lwp_proc
;
654 * Release the current user process designation on the process so
655 * the userland scheduler can work in someone else.
657 p
->p_usched
->release_curproc(lp
);
660 * lwp_exit() may be called without setting LWP_MP_WEXIT, so
661 * make sure it is set here.
663 ASSERT_LWKT_TOKEN_HELD(&p
->p_token
);
664 atomic_set_int(&lp
->lwp_mpflags
, LWP_MP_WEXIT
);
667 * Clean up any virtualization
670 vkernel_lwp_exit(lp
);
676 * Clean up select/poll support
678 kqueue_terminate(&lp
->lwp_kqueue
);
681 * Clean up any syscall-cached ucred
684 crfree(td
->td_ucred
);
689 * Nobody actually wakes us when the lock
690 * count reaches zero, so just wait one tick.
692 while (lp
->lwp_lock
> 0)
693 tsleep(lp
, 0, "lwpexit", 1);
695 /* Hand down resource usage to our proc */
696 ruadd(&p
->p_ru
, &lp
->lwp_ru
);
699 * If we don't hold the process until the LWP is reaped wait*()
700 * may try to dispose of its vmspace before all the LWPs have
701 * actually terminated.
706 * Do any remaining work that might block on us. We should be
707 * coded such that further blocking is ok after decrementing
708 * p_nthreads but don't take the chance.
710 dsched_exit_thread(td
);
711 biosched_done(curthread
);
714 * We have to use the reaper for all the LWPs except the one doing
715 * the master exit. The LWP doing the master exit can just be
716 * left on p_lwps and the process reaper will deal with it
717 * synchronously, which is much faster.
719 * Wakeup anyone waiting on p_nthreads to drop to 1 or 0.
721 * The process is left held until the reaper calls lwp_dispose() on
722 * the lp (after calling lwp_wait()).
724 if (masterexit
== 0) {
727 lwp_rb_tree_RB_REMOVE(&p
->p_lwp_tree
, lp
);
729 if ((p
->p_flags
& P_MAYBETHREADED
) && p
->p_nthreads
<= 1)
731 lwkt_gettoken(&deadlwp_token
[cpu
]);
732 LIST_INSERT_HEAD(&deadlwp_list
[cpu
], lp
, u
.lwp_reap_entry
);
733 taskqueue_enqueue(taskqueue_thread
[cpu
], deadlwp_task
[cpu
]);
734 lwkt_reltoken(&deadlwp_token
[cpu
]);
737 if ((p
->p_flags
& P_MAYBETHREADED
) && p
->p_nthreads
<= 1)
742 * We no longer need p_token.
744 * Tell the userland scheduler that we are going away
746 lwkt_reltoken(&p
->p_token
);
747 p
->p_usched
->heuristic_exiting(lp
, p
);
750 * Issue late wakeups after releasing our token to give us a chance
751 * to deschedule and switch away before another cpu in a wait*()
752 * reaps us. This is done as late as possible to reduce contention.
755 wakeup(&p
->p_nthreads
);
763 * Wait until a lwp is completely dead. The final interlock in this drama
764 * is when TDF_EXITING is set in cpu_thread_exit() just before the final
767 * At the point TDF_EXITING is set a complete exit is accomplished when
768 * TDF_RUNNING and TDF_PREEMPT_LOCK are both clear. td_mpflags has two
769 * post-switch interlock flags that can be used to wait for the TDF_
772 * Returns non-zero on success, and zero if the caller needs to retry
776 lwp_wait(struct lwp
*lp
)
778 struct thread
*td
= lp
->lwp_thread
;
781 KKASSERT(lwkt_preempted_proc() != lp
);
784 * This bit of code uses the thread destruction interlock
785 * managed by lwkt_switch_return() to wait for the lwp's
786 * thread to completely disengage.
788 * It is possible for us to race another cpu core so we
789 * have to do this correctly.
792 mpflags
= td
->td_mpflags
;
794 if (mpflags
& TDF_MP_EXITSIG
)
796 tsleep_interlock(td
, 0);
797 if (atomic_cmpset_int(&td
->td_mpflags
, mpflags
,
798 mpflags
| TDF_MP_EXITWAIT
)) {
799 tsleep(td
, PINTERLOCKED
, "lwpxt", 0);
804 * We've already waited for the core exit but there can still
805 * be other refs from e.g. process scans and such.
807 if (lp
->lwp_lock
> 0) {
808 tsleep(lp
, 0, "lwpwait1", 1);
812 tsleep(td
, 0, "lwpwait2", 1);
817 * Now that we have the thread destruction interlock these flags
818 * really should already be cleaned up, keep a check for safety.
820 * We can't rip its stack out from under it until TDF_EXITING is
821 * set and both TDF_RUNNING and TDF_PREEMPT_LOCK are clear.
822 * TDF_PREEMPT_LOCK must be checked because TDF_RUNNING
823 * will be cleared temporarily if a thread gets preempted.
825 while ((td
->td_flags
& (TDF_RUNNING
|
828 TDF_EXITING
)) != TDF_EXITING
) {
829 tsleep(lp
, 0, "lwpwait3", 1);
833 KASSERT((td
->td_flags
& (TDF_RUNQ
|TDF_TSLEEPQ
)) == 0,
834 ("lwp_wait: td %p (%s) still on run or sleep queue",
840 * Release the resources associated with a lwp.
841 * The lwp must be completely dead.
844 lwp_dispose(struct lwp
*lp
)
846 struct thread
*td
= lp
->lwp_thread
;
848 KKASSERT(lwkt_preempted_proc() != lp
);
849 KKASSERT(lp
->lwp_lock
== 0);
850 KKASSERT(td
->td_refs
== 0);
851 KKASSERT((td
->td_flags
& (TDF_RUNNING
|
854 TDF_EXITING
)) == TDF_EXITING
);
861 lp
->lwp_thread
= NULL
;
862 lwkt_free_thread(td
);
868 sys_wait4(struct wait_args
*uap
)
870 struct rusage rusage
;
873 error
= kern_wait(uap
->pid
, (uap
->status
? &status
: NULL
),
874 uap
->options
, (uap
->rusage
? &rusage
: NULL
),
875 &uap
->sysmsg_result
);
877 if (error
== 0 && uap
->status
)
878 error
= copyout(&status
, uap
->status
, sizeof(*uap
->status
));
879 if (error
== 0 && uap
->rusage
)
880 error
= copyout(&rusage
, uap
->rusage
, sizeof(*uap
->rusage
));
887 * wait_args(int pid, int *status, int options, struct rusage *rusage)
890 kern_wait(pid_t pid
, int *status
, int options
, struct rusage
*rusage
, int *res
)
892 struct thread
*td
= curthread
;
894 struct proc
*q
= td
->td_proc
;
904 if (options
&~ (WUNTRACED
|WNOHANG
|WCONTINUED
|WLINUXCLONE
))
908 * Protect the q->p_children list
910 lwkt_gettoken(&q
->p_token
);
913 * All sorts of things can change due to blocking so we have to loop
914 * all the way back up here.
916 * The problem is that if a process group is stopped and the parent
917 * is doing a wait*(..., WUNTRACED, ...), it will see the STOP
918 * of the child and then stop itself when it tries to return from the
919 * system call. When the process group is resumed the parent will
920 * then get the STOP status even though the child has now resumed
921 * (a followup wait*() will get the CONT status).
923 * Previously the CONT would overwrite the STOP because the tstop
924 * was handled within tsleep(), and the parent would only see
925 * the CONT when both are stopped and continued together. This little
926 * two-line hack restores this effect.
928 if (STOPLWP(q
, td
->td_lwp
))
936 * NOTE: We don't want to break q's p_token in the loop for the
937 * case where no children are found or we risk breaking the
938 * interlock between child and parent.
940 waitgen
= atomic_fetchadd_long(&q
->p_waitgen
, 0x80000000);
941 LIST_FOREACH(p
, &q
->p_children
, p_sibling
) {
942 if (pid
!= WAIT_ANY
&&
943 p
->p_pid
!= pid
&& p
->p_pgid
!= -pid
) {
948 * This special case handles a kthread spawned by linux_clone
949 * (see linux_misc.c). The linux_wait4 and linux_waitpid
950 * functions need to be able to distinguish between waiting
951 * on a process and waiting on a thread. It is a thread if
952 * p_sigparent is not SIGCHLD, and the WLINUXCLONE option
953 * signifies we want to wait for threads and not processes.
955 if ((p
->p_sigparent
!= SIGCHLD
) ^
956 ((options
& WLINUXCLONE
) != 0)) {
961 if (p
->p_stat
== SZOMB
) {
963 * We may go into SZOMB with threads still present.
964 * We must wait for them to exit before we can reap
965 * the master thread, otherwise we may race reaping
966 * non-master threads.
968 * Only this routine can remove a process from
969 * the zombie list and destroy it, use PACQUIREZOMB()
970 * to serialize us and loop if it blocks (interlocked
971 * by the parent's q->p_token).
973 * WARNING! (p) can be invalid when PHOLDZOMB(p)
974 * returns non-zero. Be sure not to
979 lwkt_gettoken(&p
->p_token
);
980 if (p
->p_pptr
!= q
) {
981 lwkt_reltoken(&p
->p_token
);
985 while (p
->p_nthreads
> 0) {
986 tsleep(&p
->p_nthreads
, 0, "lwpzomb", hz
);
990 * Reap any LWPs left in p->p_lwps. This is usually
991 * just the last LWP. This must be done before
992 * we loop on p_lock since the lwps hold a ref on
993 * it as a vmspace interlock.
995 * Once that is accomplished p_nthreads had better
998 while ((lp
= RB_ROOT(&p
->p_lwp_tree
)) != NULL
) {
1000 * Make sure no one is using this lwp, before
1001 * it is removed from the tree. If we didn't
1002 * wait it here, lwp tree iteration with
1003 * blocking operation would be broken.
1005 while (lp
->lwp_lock
> 0)
1006 tsleep(lp
, 0, "zomblwp", 1);
1007 lwp_rb_tree_RB_REMOVE(&p
->p_lwp_tree
, lp
);
1010 KKASSERT(p
->p_nthreads
== 0);
1013 * Don't do anything really bad until all references
1014 * to the process go away. This may include other
1015 * LWPs which are still in the process of being
1016 * reaped. We can't just pull the rug out from under
1017 * them because they may still be using the VM space.
1019 * Certain kernel facilities such as /proc will also
1020 * put a hold on the process for short periods of
1024 PSTALL(p
, "reap3", 0);
1026 /* Take care of our return values. */
1030 *status
= p
->p_xstat
;
1035 * If we got the child via a ptrace 'attach',
1036 * we need to give it back to the old parent.
1038 if (p
->p_oppid
&& (t
= pfind(p
->p_oppid
)) != NULL
) {
1041 proc_reparent(p
, t
);
1042 ksignal(t
, SIGCHLD
);
1046 lwkt_reltoken(&p
->p_token
);
1052 * Unlink the proc from its process group so that
1053 * the following operations won't lead to an
1054 * inconsistent state for processes running down
1057 proc_remove_zombie(p
);
1059 lwkt_reltoken(&p
->p_token
);
1063 ruadd(&q
->p_cru
, &p
->p_ru
);
1066 * Decrement the count of procs running with this uid.
1068 chgproccnt(p
->p_ucred
->cr_ruidinfo
, -1, 0);
1071 * Free up credentials. p_spin is required to
1072 * avoid races against allproc scans.
1074 spin_lock(&p
->p_spin
);
1077 spin_unlock(&p
->p_spin
);
1081 * Remove unused arguments
1085 if (pa
&& refcount_release(&pa
->ar_ref
)) {
1091 p
->p_sigacts
= NULL
;
1092 if (ps
&& refcount_release(&ps
->ps_refcnt
)) {
1093 kfree(ps
, M_SUBPROC
);
1098 * Our exitingcount was incremented when the process
1099 * became a zombie, now that the process has been
1100 * removed from (almost) all lists we should be able
1101 * to safely destroy its vmspace. Wait for any current
1102 * holders to go away (so the vmspace remains stable),
1105 * NOTE: Releasing the parent process (q) p_token
1106 * across the vmspace_exitfree() call is
1107 * important here to reduce stalls on
1108 * interactions with (q) (such as
1109 * fork/exec/wait or 'ps').
1111 PSTALL(p
, "reap4", 0);
1112 lwkt_reltoken(&q
->p_token
);
1113 vmspace_exitfree(p
);
1114 lwkt_gettoken(&q
->p_token
);
1115 PSTALL(p
, "reap5", 0);
1118 * NOTE: We have to officially release ZOMB in order
1119 * to ensure that a racing thread in kern_wait()
1120 * which blocked on ZOMB is woken up.
1125 atomic_add_int(&nprocs
, -1);
1129 if ((p
->p_stat
== SSTOP
|| p
->p_stat
== SCORE
) &&
1130 (p
->p_flags
& P_WAITED
) == 0 &&
1131 ((p
->p_flags
& P_TRACED
) || (options
& WUNTRACED
))) {
1133 lwkt_gettoken(&p
->p_token
);
1134 if (p
->p_pptr
!= q
) {
1135 lwkt_reltoken(&p
->p_token
);
1139 if ((p
->p_stat
!= SSTOP
&& p
->p_stat
!= SCORE
) ||
1140 (p
->p_flags
& P_WAITED
) != 0 ||
1141 ((p
->p_flags
& P_TRACED
) == 0 &&
1142 (options
& WUNTRACED
) == 0)) {
1143 lwkt_reltoken(&p
->p_token
);
1148 p
->p_flags
|= P_WAITED
;
1152 *status
= W_STOPCODE(p
->p_xstat
);
1153 /* Zero rusage so we get something consistent. */
1155 bzero(rusage
, sizeof(*rusage
));
1157 lwkt_reltoken(&p
->p_token
);
1161 if ((options
& WCONTINUED
) && (p
->p_flags
& P_CONTINUED
)) {
1163 lwkt_gettoken(&p
->p_token
);
1164 if (p
->p_pptr
!= q
) {
1165 lwkt_reltoken(&p
->p_token
);
1169 if ((p
->p_flags
& P_CONTINUED
) == 0) {
1170 lwkt_reltoken(&p
->p_token
);
1176 p
->p_flags
&= ~P_CONTINUED
;
1181 lwkt_reltoken(&p
->p_token
);
1190 if (options
& WNOHANG
) {
1197 * Wait for signal - interlocked using q->p_waitgen.
1200 while ((waitgen
& 0x7FFFFFFF) == (q
->p_waitgen
& 0x7FFFFFFF)) {
1201 tsleep_interlock(q
, PCATCH
);
1202 waitgen
= atomic_fetchadd_long(&q
->p_waitgen
, 0x80000000);
1203 if ((waitgen
& 0x7FFFFFFF) == (q
->p_waitgen
& 0x7FFFFFFF)) {
1204 error
= tsleep(q
, PCATCH
| PINTERLOCKED
, "wait", 0);
1210 lwkt_reltoken(&q
->p_token
);
1217 * Change child's parent process to parent.
1219 * p_children/p_sibling requires the parent's token, and
1220 * changing pptr requires the child's token, so we have to
1221 * get three tokens to do this operation. We also need to
1222 * hold pointers that might get ripped out from under us to
1223 * preserve structural integrity.
1225 * It is possible to race another reparent or disconnect or other
1226 * similar operation. We must retry when this situation occurs.
1227 * Once we successfully reparent the process we no longer care
1231 proc_reparent(struct proc
*child
, struct proc
*parent
)
1236 while ((opp
= child
->p_pptr
) != parent
) {
1238 lwkt_gettoken(&opp
->p_token
);
1239 lwkt_gettoken(&child
->p_token
);
1240 lwkt_gettoken(&parent
->p_token
);
1241 if (child
->p_pptr
!= opp
) {
1242 lwkt_reltoken(&parent
->p_token
);
1243 lwkt_reltoken(&child
->p_token
);
1244 lwkt_reltoken(&opp
->p_token
);
1248 LIST_REMOVE(child
, p_sibling
);
1249 LIST_INSERT_HEAD(&parent
->p_children
, child
, p_sibling
);
1250 child
->p_pptr
= parent
;
1251 lwkt_reltoken(&parent
->p_token
);
1252 lwkt_reltoken(&child
->p_token
);
1253 lwkt_reltoken(&opp
->p_token
);
1254 if (LIST_EMPTY(&opp
->p_children
))
1263 * The next two functions are to handle adding/deleting items on the
1267 * Take the arguments given and put them onto the exit callout list,
1268 * However first make sure that it's not already there.
1269 * returns 0 on success.
1273 at_exit(exitlist_fn function
)
1275 struct exitlist
*ep
;
1278 /* Be noisy if the programmer has lost track of things */
1279 if (rm_at_exit(function
))
1280 kprintf("WARNING: exit callout entry (%p) already present\n",
1283 ep
= kmalloc(sizeof(*ep
), M_ATEXIT
, M_NOWAIT
);
1286 ep
->function
= function
;
1287 TAILQ_INSERT_TAIL(&exit_list
, ep
, next
);
1292 * Scan the exit callout list for the given item and remove it.
1293 * Returns the number of items removed (0 or 1)
1296 rm_at_exit(exitlist_fn function
)
1298 struct exitlist
*ep
;
1300 TAILQ_FOREACH(ep
, &exit_list
, next
) {
1301 if (ep
->function
== function
) {
1302 TAILQ_REMOVE(&exit_list
, ep
, next
);
1303 kfree(ep
, M_ATEXIT
);
1311 * LWP reaper related code.
1314 reaplwps(void *context
, int dummy
)
1316 struct lwplist
*lwplist
= context
;
1320 lwkt_gettoken(&deadlwp_token
[cpu
]);
1321 while ((lp
= LIST_FIRST(lwplist
))) {
1322 LIST_REMOVE(lp
, u
.lwp_reap_entry
);
1325 lwkt_reltoken(&deadlwp_token
[cpu
]);
1329 reaplwp(struct lwp
*lp
)
1331 while (lwp_wait(lp
) == 0)
1341 for (cpu
= 0; cpu
< ncpus
; cpu
++) {
1342 lwkt_token_init(&deadlwp_token
[cpu
], "deadlwpl");
1343 LIST_INIT(&deadlwp_list
[cpu
]);
1344 deadlwp_task
[cpu
] = kmalloc(sizeof(*deadlwp_task
[cpu
]),
1345 M_DEVBUF
, M_WAITOK
);
1346 TASK_INIT(deadlwp_task
[cpu
], 0, reaplwps
, &deadlwp_list
[cpu
]);
1350 SYSINIT(deadlwpinit
, SI_SUB_CONFIGURE
, SI_ORDER_ANY
, deadlwp_init
, NULL
);