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 * 4. 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
37 #include <sys/cdefs.h>
38 __FBSDID("$FreeBSD$");
40 #include "opt_compat.h"
41 #include "opt_kdtrace.h"
42 #include "opt_ktrace.h"
44 #include <sys/param.h>
45 #include <sys/systm.h>
46 #include <sys/sysproto.h>
47 #include <sys/eventhandler.h>
48 #include <sys/kernel.h>
49 #include <sys/malloc.h>
51 #include <sys/mutex.h>
53 #include <sys/pioctl.h>
57 #include <sys/vmmeter.h>
58 #include <sys/vnode.h>
59 #include <sys/resourcevar.h>
61 #include <sys/signalvar.h>
62 #include <sys/sched.h>
64 #include <sys/syscallsubr.h>
65 #include <sys/syslog.h>
66 #include <sys/ptrace.h>
67 #include <sys/acct.h> /* for acct_process() function prototype */
68 #include <sys/filedesc.h>
73 #include <sys/ktrace.h>
76 #include <security/audit/audit.h>
77 #include <security/mac/mac_framework.h>
80 #include <vm/vm_extern.h>
81 #include <vm/vm_param.h>
83 #include <vm/vm_map.h>
84 #include <vm/vm_page.h>
88 #include <sys/dtrace_bsd.h>
89 dtrace_execexit_func_t dtrace_fasttrap_exit;
92 SDT_PROVIDER_DECLARE(proc);
93 SDT_PROBE_DEFINE(proc, kernel, , exit, exit);
94 SDT_PROBE_ARGTYPE(proc, kernel, , exit, 0, "int");
96 /* Hook for NFS teardown procedure. */
97 void (*nlminfo_release_p)(struct proc *p);
100 * exit -- death of process.
103 sys_exit(struct thread *td, struct sys_exit_args *uap)
106 exit1(td, W_EXITCODE(uap->rval, 0));
111 * Exit: deallocate address space and other resources, change proc state to
112 * zombie, and unlink proc from allproc and parent's lists. Save exit status
113 * and rusage for wait(). Check for child processes and orphan them.
116 exit1(struct thread *td, int rv)
118 struct proc *p, *nq, *q;
120 struct vnode *ttyvp = NULL;
124 mtx_assert(&Giant, MA_NOTOWNED);
128 * XXX in case we're rebooting we just let init die in order to
129 * work around an unsolved stack overflow seen very late during
130 * shutdown on sparc64 when the gmirror worker process exists.
132 if (p == initproc && rebooting == 0) {
133 printf("init died (signal %d, exit %d)\n",
134 WTERMSIG(rv), WEXITSTATUS(rv));
135 panic("Going nowhere without my init!");
139 * MUST abort all other threads before proceeding past here.
142 while (p->p_flag & P_HADTHREADS) {
144 * First check if some other thread got here before us..
145 * if so, act apropriatly, (exit or suspend);
147 thread_suspend_check(0);
150 * Kill off the other threads. This requires
151 * some co-operation from other parts of the kernel
152 * so it may not be instantaneous. With this state set
153 * any thread entering the kernel from userspace will
154 * thread_exit() in trap(). Any thread attempting to
155 * sleep will return immediately with EINTR or EWOULDBLOCK
156 * which will hopefully force them to back out to userland
157 * freeing resources as they go. Any thread attempting
158 * to return to userland will thread_exit() from userret().
159 * thread_exit() will unsuspend us when the last of the
160 * other threads exits.
161 * If there is already a thread singler after resumption,
162 * calling thread_single will fail; in that case, we just
163 * re-check all suspension request, the thread should
164 * either be suspended there or exit.
166 if (! thread_single(SINGLE_EXIT))
170 * All other activity in this process is now stopped.
171 * Threading support has been turned off.
174 KASSERT(p->p_numthreads == 1,
175 ("exit1: proc %p exiting with %d threads", p, p->p_numthreads));
177 * Wakeup anyone in procfs' PIOCWAIT. They should have a hold
178 * on our vmspace, so we should block below until they have
179 * released their reference to us. Note that if they have
180 * requested S_EXIT stops we will block here until they ack
183 _STOPEVENT(p, S_EXIT, rv);
186 * Note that we are exiting and do another wakeup of anyone in
187 * PIOCWAIT in case they aren't listening for S_EXIT stops or
188 * decided to wait again after we told them we are exiting.
190 p->p_flag |= P_WEXIT;
194 * Wait for any processes that have a hold on our vmspace to
195 * release their reference.
197 while (p->p_lock > 0)
198 msleep(&p->p_lock, &p->p_mtx, PWAIT, "exithold", 0);
200 p->p_xstat = rv; /* Let event handler change exit status */
202 /* Drain the limit callout while we don't have the proc locked */
203 callout_drain(&p->p_limco);
207 * The Sun BSM exit token contains two components: an exit status as
208 * passed to exit(), and a return value to indicate what sort of exit
209 * it was. The exit status is WEXITSTATUS(rv), but it's not clear
210 * what the return value is.
212 AUDIT_ARG_EXIT(WEXITSTATUS(rv), 0);
213 AUDIT_SYSCALL_EXIT(0, td);
216 /* Are we a task leader? */
217 if (p == p->p_leader) {
218 mtx_lock(&ppeers_lock);
226 while (p->p_peers != NULL)
227 msleep(p, &ppeers_lock, PWAIT, "exit1", 0);
228 mtx_unlock(&ppeers_lock);
232 * Check if any loadable modules need anything done at process exit.
233 * E.g. SYSV IPC stuff
234 * XXX what if one of these generates an error?
236 EVENTHANDLER_INVOKE(process_exit, p);
239 * If parent is waiting for us to exit or exec,
240 * P_PPWAIT is set; we will wakeup the parent below.
243 rv = p->p_xstat; /* Event handler could change exit status */
245 p->p_flag &= ~(P_TRACED | P_PPWAIT);
248 * Stop the real interval timer. If the handler is currently
249 * executing, prevent it from rearming itself and let it finish.
251 if (timevalisset(&p->p_realtimer.it_value) &&
252 callout_stop(&p->p_itcallout) == 0) {
253 timevalclear(&p->p_realtimer.it_interval);
254 msleep(&p->p_itcallout, &p->p_mtx, PWAIT, "ritwait", 0);
255 KASSERT(!timevalisset(&p->p_realtimer.it_value),
256 ("realtime timer is still armed"));
261 * Reset any sigio structures pointing to us as a result of
262 * F_SETOWN with our pid.
264 funsetownlst(&p->p_sigiolst);
267 * If this process has an nlminfo data area (for lockd), release it
269 if (nlminfo_release_p != NULL && p->p_nlminfo != NULL)
270 (*nlminfo_release_p)(p);
273 * Close open files and release open-file table.
279 * If this thread tickled GEOM, we need to wait for the giggling to
280 * stop before we return to userland
282 if (td->td_pflags & TDP_GEOM)
286 * Remove ourself from our leader's peer list and wake our leader.
288 mtx_lock(&ppeers_lock);
289 if (p->p_leader->p_peers) {
291 while (q->p_peers != p)
293 q->p_peers = p->p_peers;
296 mtx_unlock(&ppeers_lock);
300 sx_xlock(&proctree_lock);
301 if (SESS_LEADER(p)) {
302 struct session *sp = p->p_session;
306 * s_ttyp is not zero'd; we use this to indicate that
307 * the session once had a controlling terminal. (for
308 * logging and informational purposes)
318 * Signal foreground pgrp and revoke access to
319 * controlling terminal if it has not been revoked
322 * Because the TTY may have been revoked in the mean
323 * time and could already have a new session associated
324 * with it, make sure we don't send a SIGHUP to a
325 * foreground process group that does not belong to this
331 if (tp->t_session == sp)
332 tty_signal_pgrp(tp, SIGHUP);
337 sx_xunlock(&proctree_lock);
338 if (vn_lock(ttyvp, LK_EXCLUSIVE) == 0) {
339 VOP_REVOKE(ttyvp, REVOKEALL);
340 VOP_UNLOCK(ttyvp, 0);
342 sx_xlock(&proctree_lock);
345 fixjobc(p, p->p_pgrp, 0);
346 sx_xunlock(&proctree_lock);
347 (void)acct_process(td);
349 /* Release the TTY now we've unlocked everything. */
356 * Release reference to text vnode
358 if ((vtmp = p->p_textvp) != NULL) {
360 locked = VFS_LOCK_GIANT(vtmp->v_mount);
362 VFS_UNLOCK_GIANT(locked);
366 * Release our limits structure.
375 * Remove proc from allproc queue and pidhash chain.
376 * Place onto zombproc. Unlink from parent's child list.
378 sx_xlock(&allproc_lock);
379 LIST_REMOVE(p, p_list);
380 LIST_INSERT_HEAD(&zombproc, p, p_list);
381 LIST_REMOVE(p, p_hash);
382 sx_xunlock(&allproc_lock);
385 * Call machine-dependent code to release any
386 * machine-dependent resources other than the address space.
387 * The address space is released by "vmspace_exitfree(p)" in
392 WITNESS_WARN(WARN_PANIC, NULL, "process (pid %d) exiting", p->p_pid);
395 * Reparent all of our children to init.
397 sx_xlock(&proctree_lock);
398 q = LIST_FIRST(&p->p_children);
399 if (q != NULL) /* only need this if any child is S_ZOMB */
401 for (; q != NULL; q = nq) {
402 nq = LIST_NEXT(q, p_sibling);
404 proc_reparent(q, initproc);
405 q->p_sigparent = SIGCHLD;
407 * Traced processes are killed
408 * since their existence means someone is screwing up.
410 if (q->p_flag & P_TRACED) {
413 q->p_flag &= ~(P_TRACED | P_STOPPED_TRACE);
414 FOREACH_THREAD_IN_PROC(q, temp)
415 temp->td_dbgflags &= ~TDB_SUSPEND;
421 /* Save exit status. */
425 /* Tell the prison that we are gone. */
426 prison_proc_free(p->p_ucred->cr_prison);
430 * Tell the DTrace fasttrap provider about the exit if it
431 * has declared an interest.
433 if (dtrace_fasttrap_exit)
434 dtrace_fasttrap_exit(p);
438 * Notify interested parties of our demise.
440 KNOTE_LOCKED(&p->p_klist, NOTE_EXIT);
443 int reason = CLD_EXITED;
446 else if (WIFSIGNALED(rv))
448 SDT_PROBE(proc, kernel, , exit, reason, 0, 0, 0, 0);
452 * Just delete all entries in the p_klist. At this point we won't
453 * report any more events, and there are nasty race conditions that
454 * can beat us if we don't.
456 knlist_clear(&p->p_klist, 1);
459 * Notify parent that we're gone. If parent has the PS_NOCLDWAIT
460 * flag set, or if the handler is set to SIG_IGN, notify process
461 * 1 instead (and hope it will handle this situation).
463 PROC_LOCK(p->p_pptr);
464 mtx_lock(&p->p_pptr->p_sigacts->ps_mtx);
465 if (p->p_pptr->p_sigacts->ps_flag & (PS_NOCLDWAIT | PS_CLDSIGIGN)) {
468 mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx);
471 proc_reparent(p, initproc);
472 p->p_sigparent = SIGCHLD;
473 PROC_LOCK(p->p_pptr);
476 * Notify parent, so in case he was wait(2)ing or
477 * executing waitpid(2) with our pid, he will
482 mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx);
484 if (p->p_pptr == initproc)
485 psignal(p->p_pptr, SIGCHLD);
486 else if (p->p_sigparent != 0) {
487 if (p->p_sigparent == SIGCHLD)
489 else /* LINUX thread */
490 psignal(p->p_pptr, p->p_sigparent);
492 sx_xunlock(&proctree_lock);
495 * The state PRS_ZOMBIE prevents other proesses from sending
496 * signal to the process, to avoid memory leak, we free memory
497 * for signal queue at the time when the state is set.
499 sigqueue_flush(&p->p_sigqueue);
500 sigqueue_flush(&td->td_sigqueue);
503 * We have to wait until after acquiring all locks before
504 * changing p_state. We need to avoid all possible context
505 * switches (including ones from blocking on a mutex) while
506 * marked as a zombie. We also have to set the zombie state
507 * before we release the parent process' proc lock to avoid
508 * a lost wakeup. So, we first call wakeup, then we grab the
509 * sched lock, update the state, and release the parent process'
513 cv_broadcast(&p->p_pwait);
514 sched_exit(p->p_pptr, td);
516 p->p_state = PRS_ZOMBIE;
517 PROC_UNLOCK(p->p_pptr);
520 * Hopefully no one will try to deliver a signal to the process this
523 knlist_destroy(&p->p_klist);
526 * Save our children's rusage information in our exit rusage.
528 ruadd(&p->p_ru, &p->p_rux, &p->p_stats->p_cru, &p->p_crux);
531 * Make sure the scheduler takes this thread out of its tables etc.
532 * This will also release this thread's reference to the ucred.
533 * Other thread parts to release include pcb bits and such.
539 #ifndef _SYS_SYSPROTO_H_
548 abort2(struct thread *td, struct abort2_args *uap)
550 struct proc *p = td->td_proc;
556 * Do it right now so we can log either proper call of abort2(), or
557 * note, that invalid argument was passed. 512 is big enough to
558 * handle 16 arguments' descriptions with additional comments.
560 sb = sbuf_new(NULL, NULL, 512, SBUF_FIXEDLEN);
562 sbuf_printf(sb, "%s(pid %d uid %d) aborted: ",
563 p->p_comm, p->p_pid, td->td_ucred->cr_uid);
565 * Since we can't return from abort2(), send SIGKILL in cases, where
566 * abort2() was called improperly
569 /* Prevent from DoSes from user-space. */
570 if (uap->nargs < 0 || uap->nargs > 16)
572 if (uap->nargs > 0) {
573 if (uap->args == NULL)
575 error = copyin(uap->args, uargs, uap->nargs * sizeof(void *));
580 * Limit size of 'reason' string to 128. Will fit even when
581 * maximal number of arguments was chosen to be logged.
583 if (uap->why != NULL) {
584 error = sbuf_copyin(sb, uap->why, 128);
588 sbuf_printf(sb, "(null)");
590 if (uap->nargs > 0) {
591 sbuf_printf(sb, "(");
592 for (i = 0;i < uap->nargs; i++)
593 sbuf_printf(sb, "%s%p", i == 0 ? "" : ", ", uargs[i]);
594 sbuf_printf(sb, ")");
597 * Final stage: arguments were proper, string has been
598 * successfully copied from userspace, and copying pointers
599 * from user-space succeed.
603 if (sig == SIGKILL) {
605 sbuf_printf(sb, " (Reason text inaccessible)");
609 log(LOG_INFO, "%s", sbuf_data(sb));
611 exit1(td, W_EXITCODE(0, sig));
618 * The dirty work is handled by kern_wait().
621 owait(struct thread *td, struct owait_args *uap __unused)
625 error = kern_wait(td, WAIT_ANY, &status, 0, NULL);
627 td->td_retval[1] = status;
630 #endif /* COMPAT_43 */
633 * The dirty work is handled by kern_wait().
636 wait4(struct thread *td, struct wait_args *uap)
638 struct rusage ru, *rup;
641 if (uap->rusage != NULL)
645 error = kern_wait(td, uap->pid, &status, uap->options, rup);
646 if (uap->status != NULL && error == 0)
647 error = copyout(&status, uap->status, sizeof(status));
648 if (uap->rusage != NULL && error == 0)
649 error = copyout(&ru, uap->rusage, sizeof(struct rusage));
654 * Reap the remains of a zombie process and optionally return status and
655 * rusage. Asserts and will release both the proctree_lock and the process
656 * lock as part of its work.
659 proc_reap(struct thread *td, struct proc *p, int *status, int options,
660 struct rusage *rusage)
664 sx_assert(&proctree_lock, SA_XLOCKED);
665 PROC_LOCK_ASSERT(p, MA_OWNED);
666 PROC_SLOCK_ASSERT(p, MA_OWNED);
667 KASSERT(p->p_state == PRS_ZOMBIE, ("proc_reap: !PRS_ZOMBIE"));
672 calcru(p, &rusage->ru_utime, &rusage->ru_stime);
675 td->td_retval[0] = p->p_pid;
677 *status = p->p_xstat; /* convert to int */
678 if (options & WNOWAIT) {
680 * Only poll, returning the status. Caller does not wish to
681 * release the proc struct just yet.
684 sx_xunlock(&proctree_lock);
689 sigqueue_take(p->p_ksi);
694 * If we got the child via a ptrace 'attach', we need to give it back
697 if (p->p_oppid && (t = pfind(p->p_oppid)) != NULL) {
702 tdsignal(t, NULL, SIGCHLD, p->p_ksi);
704 cv_broadcast(&p->p_pwait);
706 sx_xunlock(&proctree_lock);
711 * Remove other references to this process to ensure we have an
712 * exclusive reference.
714 sx_xlock(&allproc_lock);
715 LIST_REMOVE(p, p_list); /* off zombproc */
716 sx_xunlock(&allproc_lock);
717 LIST_REMOVE(p, p_sibling);
719 sx_xunlock(&proctree_lock);
722 * As a side effect of this lock, we know that all other writes to
723 * this proc are visible now, so no more locking is needed for p.
726 p->p_xstat = 0; /* XXX: why? */
729 ruadd(&q->p_stats->p_cru, &q->p_crux, &p->p_ru, &p->p_rux);
733 * Decrement the count of procs running with this uid.
735 (void)chgproccnt(p->p_ucred->cr_ruidinfo, -1, 0);
738 * Free credentials, arguments, and sigacts.
742 pargs_drop(p->p_args);
744 sigacts_free(p->p_sigacts);
748 * Do any thread-system specific cleanups.
753 * Give vm and machine-dependent layer a chance to free anything that
754 * cpu_exit couldn't release while still running in process context.
760 KASSERT(FIRST_THREAD_IN_PROC(p),
761 ("proc_reap: no residual thread!"));
762 uma_zfree(proc_zone, p);
763 sx_xlock(&allproc_lock);
765 sx_xunlock(&allproc_lock);
769 kern_wait(struct thread *td, pid_t pid, int *status, int options,
770 struct rusage *rusage)
776 AUDIT_ARG_VALUE(options);
784 if (options &~ (WUNTRACED|WNOHANG|WCONTINUED|WNOWAIT|WLINUXCLONE))
787 if (q->p_flag & P_STATCHILD) {
789 q->p_flag &= ~P_STATCHILD;
793 sx_xlock(&proctree_lock);
794 LIST_FOREACH(p, &q->p_children, p_sibling) {
796 if (pid != WAIT_ANY &&
797 p->p_pid != pid && p->p_pgid != -pid) {
801 if (p_canwait(td, p)) {
807 * This special case handles a kthread spawned by linux_clone
808 * (see linux_misc.c). The linux_wait4 and linux_waitpid
809 * functions need to be able to distinguish between waiting
810 * on a process and waiting on a thread. It is a thread if
811 * p_sigparent is not SIGCHLD, and the WLINUXCLONE option
812 * signifies we want to wait for threads and not processes.
814 if ((p->p_sigparent != SIGCHLD) ^
815 ((options & WLINUXCLONE) != 0)) {
822 if (p->p_state == PRS_ZOMBIE) {
823 proc_reap(td, p, status, options, rusage);
826 if ((p->p_flag & P_STOPPED_SIG) &&
827 (p->p_suspcount == p->p_numthreads) &&
828 (p->p_flag & P_WAITED) == 0 &&
829 (p->p_flag & P_TRACED || options & WUNTRACED)) {
831 p->p_flag |= P_WAITED;
832 sx_xunlock(&proctree_lock);
833 td->td_retval[0] = p->p_pid;
835 *status = W_STOPCODE(p->p_xstat);
838 sigqueue_take(p->p_ksi);
845 if (options & WCONTINUED && (p->p_flag & P_CONTINUED)) {
846 sx_xunlock(&proctree_lock);
847 td->td_retval[0] = p->p_pid;
848 p->p_flag &= ~P_CONTINUED;
851 sigqueue_take(p->p_ksi);
862 sx_xunlock(&proctree_lock);
865 if (options & WNOHANG) {
866 sx_xunlock(&proctree_lock);
867 td->td_retval[0] = 0;
871 sx_xunlock(&proctree_lock);
872 if (q->p_flag & P_STATCHILD) {
873 q->p_flag &= ~P_STATCHILD;
876 error = msleep(q, &q->p_mtx, PWAIT | PCATCH, "wait", 0);
884 * Make process 'parent' the new parent of process 'child'.
885 * Must be called with an exclusive hold of proctree lock.
888 proc_reparent(struct proc *child, struct proc *parent)
891 sx_assert(&proctree_lock, SX_XLOCKED);
892 PROC_LOCK_ASSERT(child, MA_OWNED);
893 if (child->p_pptr == parent)
896 PROC_LOCK(child->p_pptr);
897 sigqueue_take(child->p_ksi);
898 PROC_UNLOCK(child->p_pptr);
899 LIST_REMOVE(child, p_sibling);
900 LIST_INSERT_HEAD(&parent->p_children, child, p_sibling);
901 child->p_pptr = parent;