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);
94 SDT_PROBE_ARGTYPE(proc, kernel, , exit, 0, "int");
96 /* Required to be non-static for SysVR4 emulator */
97 MALLOC_DEFINE(M_ZOMBIE, "zombie", "zombie proc status");
99 /* Hook for NFS teardown procedure. */
100 void (*nlminfo_release_p)(struct proc *p);
103 * exit -- death of process.
106 sys_exit(struct thread *td, struct sys_exit_args *uap)
109 exit1(td, W_EXITCODE(uap->rval, 0));
114 * Exit: deallocate address space and other resources, change proc state to
115 * zombie, and unlink proc from allproc and parent's lists. Save exit status
116 * and rusage for wait(). Check for child processes and orphan them.
119 exit1(struct thread *td, int rv)
121 struct proc *p, *nq, *q;
123 struct vnode *ttyvp = NULL;
125 struct vnode *tracevp;
126 struct ucred *tracecred;
131 mtx_assert(&Giant, MA_NOTOWNED);
135 printf("init died (signal %d, exit %d)\n",
136 WTERMSIG(rv), WEXITSTATUS(rv));
137 panic("Going nowhere without my init!");
141 * MUST abort all other threads before proceeding past here.
144 while (p->p_flag & P_HADTHREADS) {
146 * First check if some other thread got here before us..
147 * if so, act apropriatly, (exit or suspend);
149 thread_suspend_check(0);
152 * Kill off the other threads. This requires
153 * some co-operation from other parts of the kernel
154 * so it may not be instantaneous. With this state set
155 * any thread entering the kernel from userspace will
156 * thread_exit() in trap(). Any thread attempting to
157 * sleep will return immediately with EINTR or EWOULDBLOCK
158 * which will hopefully force them to back out to userland
159 * freeing resources as they go. Any thread attempting
160 * to return to userland will thread_exit() from userret().
161 * thread_exit() will unsuspend us when the last of the
162 * other threads exits.
163 * If there is already a thread singler after resumption,
164 * calling thread_single will fail; in that case, we just
165 * re-check all suspension request, the thread should
166 * either be suspended there or exit.
168 if (! thread_single(SINGLE_EXIT))
172 * All other activity in this process is now stopped.
173 * Threading support has been turned off.
176 KASSERT(p->p_numthreads == 1,
177 ("exit1: proc %p exiting with %d threads", p, p->p_numthreads));
179 * Wakeup anyone in procfs' PIOCWAIT. They should have a hold
180 * on our vmspace, so we should block below until they have
181 * released their reference to us. Note that if they have
182 * requested S_EXIT stops we will block here until they ack
185 _STOPEVENT(p, S_EXIT, rv);
188 * Note that we are exiting and do another wakeup of anyone in
189 * PIOCWAIT in case they aren't listening for S_EXIT stops or
190 * decided to wait again after we told them we are exiting.
192 p->p_flag |= P_WEXIT;
196 * Wait for any processes that have a hold on our vmspace to
197 * release their reference.
199 while (p->p_lock > 0)
200 msleep(&p->p_lock, &p->p_mtx, PWAIT, "exithold", 0);
203 /* Drain the limit callout while we don't have the proc locked */
204 callout_drain(&p->p_limco);
208 * The Sun BSM exit token contains two components: an exit status as
209 * passed to exit(), and a return value to indicate what sort of exit
210 * it was. The exit status is WEXITSTATUS(rv), but it's not clear
211 * what the return value is.
213 AUDIT_ARG_EXIT(WEXITSTATUS(rv), 0);
214 AUDIT_SYSCALL_EXIT(0, td);
217 /* Are we a task leader? */
218 if (p == p->p_leader) {
219 mtx_lock(&ppeers_lock);
227 while (p->p_peers != NULL)
228 msleep(p, &ppeers_lock, PWAIT, "exit1", 0);
229 mtx_unlock(&ppeers_lock);
233 * Check if any loadable modules need anything done at process exit.
234 * E.g. SYSV IPC stuff
235 * XXX what if one of these generates an error?
237 EVENTHANDLER_INVOKE(process_exit, p);
240 * If parent is waiting for us to exit or exec,
241 * P_PPWAIT is set; we will wakeup the parent below.
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 vn_lock(ttyvp, LK_EXCLUSIVE | LK_RETRY);
339 if (ttyvp->v_type != VBAD)
340 VOP_REVOKE(ttyvp, REVOKEALL);
341 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. */
354 * Disable tracing, then drain any pending records and release
357 if (p->p_traceflag != 0) {
359 mtx_lock(&ktrace_mtx);
361 mtx_unlock(&ktrace_mtx);
365 mtx_lock(&ktrace_mtx);
366 tracevp = p->p_tracevp;
368 tracecred = p->p_tracecred;
369 p->p_tracecred = NULL;
370 mtx_unlock(&ktrace_mtx);
372 if (tracevp != NULL) {
373 locked = VFS_LOCK_GIANT(tracevp->v_mount);
375 VFS_UNLOCK_GIANT(locked);
377 if (tracecred != NULL)
382 * Release reference to text vnode
384 if ((vtmp = p->p_textvp) != NULL) {
386 locked = VFS_LOCK_GIANT(vtmp->v_mount);
388 VFS_UNLOCK_GIANT(locked);
392 * Release our limits structure.
401 * Remove proc from allproc queue and pidhash chain.
402 * Place onto zombproc. Unlink from parent's child list.
404 sx_xlock(&allproc_lock);
405 LIST_REMOVE(p, p_list);
406 LIST_INSERT_HEAD(&zombproc, p, p_list);
407 LIST_REMOVE(p, p_hash);
408 sx_xunlock(&allproc_lock);
411 * Call machine-dependent code to release any
412 * machine-dependent resources other than the address space.
413 * The address space is released by "vmspace_exitfree(p)" in
418 WITNESS_WARN(WARN_PANIC, NULL, "process (pid %d) exiting", p->p_pid);
421 * Reparent all of our children to init.
423 sx_xlock(&proctree_lock);
424 q = LIST_FIRST(&p->p_children);
425 if (q != NULL) /* only need this if any child is S_ZOMB */
427 for (; q != NULL; q = nq) {
428 nq = LIST_NEXT(q, p_sibling);
430 proc_reparent(q, initproc);
431 q->p_sigparent = SIGCHLD;
433 * Traced processes are killed
434 * since their existence means someone is screwing up.
436 if (q->p_flag & P_TRACED) {
439 q->p_flag &= ~(P_TRACED | P_STOPPED_TRACE);
440 FOREACH_THREAD_IN_PROC(q, temp)
441 temp->td_dbgflags &= ~TDB_SUSPEND;
447 /* Save exit status. */
452 /* Tell the prison that we are gone. */
453 prison_proc_free(p->p_ucred->cr_prison);
457 * Tell the DTrace fasttrap provider about the exit if it
458 * has declared an interest.
460 if (dtrace_fasttrap_exit)
461 dtrace_fasttrap_exit(p);
465 * Notify interested parties of our demise.
467 KNOTE_LOCKED(&p->p_klist, NOTE_EXIT);
470 int reason = CLD_EXITED;
473 else if (WIFSIGNALED(rv))
475 SDT_PROBE(proc, kernel, , exit, reason, 0, 0, 0, 0);
479 * Just delete all entries in the p_klist. At this point we won't
480 * report any more events, and there are nasty race conditions that
481 * can beat us if we don't.
483 knlist_clear(&p->p_klist, 1);
486 * Notify parent that we're gone. If parent has the PS_NOCLDWAIT
487 * flag set, or if the handler is set to SIG_IGN, notify process
488 * 1 instead (and hope it will handle this situation).
490 PROC_LOCK(p->p_pptr);
491 mtx_lock(&p->p_pptr->p_sigacts->ps_mtx);
492 if (p->p_pptr->p_sigacts->ps_flag & (PS_NOCLDWAIT | PS_CLDSIGIGN)) {
495 mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx);
498 proc_reparent(p, initproc);
499 p->p_sigparent = SIGCHLD;
500 PROC_LOCK(p->p_pptr);
503 * Notify parent, so in case he was wait(2)ing or
504 * executing waitpid(2) with our pid, he will
509 mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx);
511 if (p->p_pptr == initproc)
512 psignal(p->p_pptr, SIGCHLD);
513 else if (p->p_sigparent != 0) {
514 if (p->p_sigparent == SIGCHLD)
516 else /* LINUX thread */
517 psignal(p->p_pptr, p->p_sigparent);
519 sx_xunlock(&proctree_lock);
522 * The state PRS_ZOMBIE prevents other proesses from sending
523 * signal to the process, to avoid memory leak, we free memory
524 * for signal queue at the time when the state is set.
526 sigqueue_flush(&p->p_sigqueue);
527 sigqueue_flush(&td->td_sigqueue);
530 * We have to wait until after acquiring all locks before
531 * changing p_state. We need to avoid all possible context
532 * switches (including ones from blocking on a mutex) while
533 * marked as a zombie. We also have to set the zombie state
534 * before we release the parent process' proc lock to avoid
535 * a lost wakeup. So, we first call wakeup, then we grab the
536 * sched lock, update the state, and release the parent process'
540 cv_broadcast(&p->p_pwait);
541 sched_exit(p->p_pptr, td);
543 p->p_state = PRS_ZOMBIE;
544 PROC_UNLOCK(p->p_pptr);
547 * Hopefully no one will try to deliver a signal to the process this
550 knlist_destroy(&p->p_klist);
553 * Save our children's rusage information in our exit rusage.
555 ruadd(&p->p_ru, &p->p_rux, &p->p_stats->p_cru, &p->p_crux);
558 * Make sure the scheduler takes this thread out of its tables etc.
559 * This will also release this thread's reference to the ucred.
560 * Other thread parts to release include pcb bits and such.
566 #ifndef _SYS_SYSPROTO_H_
575 abort2(struct thread *td, struct abort2_args *uap)
577 struct proc *p = td->td_proc;
583 * Do it right now so we can log either proper call of abort2(), or
584 * note, that invalid argument was passed. 512 is big enough to
585 * handle 16 arguments' descriptions with additional comments.
587 sb = sbuf_new(NULL, NULL, 512, SBUF_FIXEDLEN);
589 sbuf_printf(sb, "%s(pid %d uid %d) aborted: ",
590 p->p_comm, p->p_pid, td->td_ucred->cr_uid);
592 * Since we can't return from abort2(), send SIGKILL in cases, where
593 * abort2() was called improperly
596 /* Prevent from DoSes from user-space. */
597 if (uap->nargs < 0 || uap->nargs > 16)
599 if (uap->nargs > 0) {
600 if (uap->args == NULL)
602 error = copyin(uap->args, uargs, uap->nargs * sizeof(void *));
607 * Limit size of 'reason' string to 128. Will fit even when
608 * maximal number of arguments was chosen to be logged.
610 if (uap->why != NULL) {
611 error = sbuf_copyin(sb, uap->why, 128);
615 sbuf_printf(sb, "(null)");
617 if (uap->nargs > 0) {
618 sbuf_printf(sb, "(");
619 for (i = 0;i < uap->nargs; i++)
620 sbuf_printf(sb, "%s%p", i == 0 ? "" : ", ", uargs[i]);
621 sbuf_printf(sb, ")");
624 * Final stage: arguments were proper, string has been
625 * successfully copied from userspace, and copying pointers
626 * from user-space succeed.
630 if (sig == SIGKILL) {
632 sbuf_printf(sb, " (Reason text inaccessible)");
636 log(LOG_INFO, "%s", sbuf_data(sb));
638 exit1(td, W_EXITCODE(0, sig));
645 * The dirty work is handled by kern_wait().
648 owait(struct thread *td, struct owait_args *uap __unused)
652 error = kern_wait(td, WAIT_ANY, &status, 0, NULL);
654 td->td_retval[1] = status;
657 #endif /* COMPAT_43 */
660 * The dirty work is handled by kern_wait().
663 wait4(struct thread *td, struct wait_args *uap)
665 struct rusage ru, *rup;
668 if (uap->rusage != NULL)
672 error = kern_wait(td, uap->pid, &status, uap->options, rup);
673 if (uap->status != NULL && error == 0)
674 error = copyout(&status, uap->status, sizeof(status));
675 if (uap->rusage != NULL && error == 0)
676 error = copyout(&ru, uap->rusage, sizeof(struct rusage));
681 * Reap the remains of a zombie process and optionally return status and
682 * rusage. Asserts and will release both the proctree_lock and the process
683 * lock as part of its work.
686 proc_reap(struct thread *td, struct proc *p, int *status, int options,
687 struct rusage *rusage)
691 sx_assert(&proctree_lock, SA_XLOCKED);
692 PROC_LOCK_ASSERT(p, MA_OWNED);
693 PROC_SLOCK_ASSERT(p, MA_OWNED);
694 KASSERT(p->p_state == PRS_ZOMBIE, ("proc_reap: !PRS_ZOMBIE"));
699 calcru(p, &rusage->ru_utime, &rusage->ru_stime);
702 td->td_retval[0] = p->p_pid;
704 *status = p->p_xstat; /* convert to int */
705 if (options & WNOWAIT) {
707 * Only poll, returning the status. Caller does not wish to
708 * release the proc struct just yet.
711 sx_xunlock(&proctree_lock);
716 sigqueue_take(p->p_ksi);
721 * If we got the child via a ptrace 'attach', we need to give it back
724 if (p->p_oppid && (t = pfind(p->p_oppid)) != NULL) {
729 tdsignal(t, NULL, SIGCHLD, p->p_ksi);
731 cv_broadcast(&p->p_pwait);
733 sx_xunlock(&proctree_lock);
738 * Remove other references to this process to ensure we have an
739 * exclusive reference.
741 sx_xlock(&allproc_lock);
742 LIST_REMOVE(p, p_list); /* off zombproc */
743 sx_xunlock(&allproc_lock);
744 LIST_REMOVE(p, p_sibling);
746 sx_xunlock(&proctree_lock);
749 * As a side effect of this lock, we know that all other writes to
750 * this proc are visible now, so no more locking is needed for p.
753 p->p_xstat = 0; /* XXX: why? */
756 ruadd(&q->p_stats->p_cru, &q->p_crux, &p->p_ru, &p->p_rux);
760 * Decrement the count of procs running with this uid.
762 (void)chgproccnt(p->p_ucred->cr_ruidinfo, -1, 0);
765 * Free credentials, arguments, and sigacts.
769 pargs_drop(p->p_args);
771 sigacts_free(p->p_sigacts);
775 * Do any thread-system specific cleanups.
780 * Give vm and machine-dependent layer a chance to free anything that
781 * cpu_exit couldn't release while still running in process context.
787 KASSERT(FIRST_THREAD_IN_PROC(p),
788 ("proc_reap: no residual thread!"));
789 uma_zfree(proc_zone, p);
790 sx_xlock(&allproc_lock);
792 sx_xunlock(&allproc_lock);
796 kern_wait(struct thread *td, pid_t pid, int *status, int options,
797 struct rusage *rusage)
803 AUDIT_ARG_VALUE(options);
811 if (options &~ (WUNTRACED|WNOHANG|WCONTINUED|WNOWAIT|WLINUXCLONE))
814 if (q->p_flag & P_STATCHILD) {
816 q->p_flag &= ~P_STATCHILD;
820 sx_xlock(&proctree_lock);
821 LIST_FOREACH(p, &q->p_children, p_sibling) {
823 if (pid != WAIT_ANY &&
824 p->p_pid != pid && p->p_pgid != -pid) {
828 if (p_canwait(td, p)) {
834 * This special case handles a kthread spawned by linux_clone
835 * (see linux_misc.c). The linux_wait4 and linux_waitpid
836 * functions need to be able to distinguish between waiting
837 * on a process and waiting on a thread. It is a thread if
838 * p_sigparent is not SIGCHLD, and the WLINUXCLONE option
839 * signifies we want to wait for threads and not processes.
841 if ((p->p_sigparent != SIGCHLD) ^
842 ((options & WLINUXCLONE) != 0)) {
849 if (p->p_state == PRS_ZOMBIE) {
850 proc_reap(td, p, status, options, rusage);
853 if ((p->p_flag & P_STOPPED_SIG) &&
854 (p->p_suspcount == p->p_numthreads) &&
855 (p->p_flag & P_WAITED) == 0 &&
856 (p->p_flag & P_TRACED || options & WUNTRACED)) {
858 p->p_flag |= P_WAITED;
859 sx_xunlock(&proctree_lock);
860 td->td_retval[0] = p->p_pid;
862 *status = W_STOPCODE(p->p_xstat);
865 sigqueue_take(p->p_ksi);
872 if (options & WCONTINUED && (p->p_flag & P_CONTINUED)) {
873 sx_xunlock(&proctree_lock);
874 td->td_retval[0] = p->p_pid;
875 p->p_flag &= ~P_CONTINUED;
878 sigqueue_take(p->p_ksi);
889 sx_xunlock(&proctree_lock);
892 if (options & WNOHANG) {
893 sx_xunlock(&proctree_lock);
894 td->td_retval[0] = 0;
898 sx_xunlock(&proctree_lock);
899 if (q->p_flag & P_STATCHILD) {
900 q->p_flag &= ~P_STATCHILD;
903 error = msleep(q, &q->p_mtx, PWAIT | PCATCH, "wait", 0);
911 * Make process 'parent' the new parent of process 'child'.
912 * Must be called with an exclusive hold of proctree lock.
915 proc_reparent(struct proc *child, struct proc *parent)
918 sx_assert(&proctree_lock, SX_XLOCKED);
919 PROC_LOCK_ASSERT(child, MA_OWNED);
920 if (child->p_pptr == parent)
923 PROC_LOCK(child->p_pptr);
924 sigqueue_take(child->p_ksi);
925 PROC_UNLOCK(child->p_pptr);
926 LIST_REMOVE(child, p_sibling);
927 LIST_INSERT_HEAD(&parent->p_children, child, p_sibling);
928 child->p_pptr = parent;