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 /* 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 * XXX in case we're rebooting we just let init die in order to
136 * work around an unsolved stack overflow seen very late during
137 * shutdown on sparc64 when the gmirror worker process exists.
139 if (p == initproc && rebooting == 0) {
140 printf("init died (signal %d, exit %d)\n",
141 WTERMSIG(rv), WEXITSTATUS(rv));
142 panic("Going nowhere without my init!");
146 * MUST abort all other threads before proceeding past here.
149 while (p->p_flag & P_HADTHREADS) {
151 * First check if some other thread got here before us..
152 * if so, act apropriatly, (exit or suspend);
154 thread_suspend_check(0);
157 * Kill off the other threads. This requires
158 * some co-operation from other parts of the kernel
159 * so it may not be instantaneous. With this state set
160 * any thread entering the kernel from userspace will
161 * thread_exit() in trap(). Any thread attempting to
162 * sleep will return immediately with EINTR or EWOULDBLOCK
163 * which will hopefully force them to back out to userland
164 * freeing resources as they go. Any thread attempting
165 * to return to userland will thread_exit() from userret().
166 * thread_exit() will unsuspend us when the last of the
167 * other threads exits.
168 * If there is already a thread singler after resumption,
169 * calling thread_single will fail; in that case, we just
170 * re-check all suspension request, the thread should
171 * either be suspended there or exit.
173 if (! thread_single(SINGLE_EXIT))
177 * All other activity in this process is now stopped.
178 * Threading support has been turned off.
181 KASSERT(p->p_numthreads == 1,
182 ("exit1: proc %p exiting with %d threads", p, p->p_numthreads));
184 * Wakeup anyone in procfs' PIOCWAIT. They should have a hold
185 * on our vmspace, so we should block below until they have
186 * released their reference to us. Note that if they have
187 * requested S_EXIT stops we will block here until they ack
190 _STOPEVENT(p, S_EXIT, rv);
193 * Note that we are exiting and do another wakeup of anyone in
194 * PIOCWAIT in case they aren't listening for S_EXIT stops or
195 * decided to wait again after we told them we are exiting.
197 p->p_flag |= P_WEXIT;
201 * Wait for any processes that have a hold on our vmspace to
202 * release their reference.
204 while (p->p_lock > 0)
205 msleep(&p->p_lock, &p->p_mtx, PWAIT, "exithold", 0);
208 /* Drain the limit callout while we don't have the proc locked */
209 callout_drain(&p->p_limco);
213 * The Sun BSM exit token contains two components: an exit status as
214 * passed to exit(), and a return value to indicate what sort of exit
215 * it was. The exit status is WEXITSTATUS(rv), but it's not clear
216 * what the return value is.
218 AUDIT_ARG_EXIT(WEXITSTATUS(rv), 0);
219 AUDIT_SYSCALL_EXIT(0, td);
222 /* Are we a task leader? */
223 if (p == p->p_leader) {
224 mtx_lock(&ppeers_lock);
232 while (p->p_peers != NULL)
233 msleep(p, &ppeers_lock, PWAIT, "exit1", 0);
234 mtx_unlock(&ppeers_lock);
238 * Check if any loadable modules need anything done at process exit.
239 * E.g. SYSV IPC stuff
240 * XXX what if one of these generates an error?
242 EVENTHANDLER_INVOKE(process_exit, p);
245 * If parent is waiting for us to exit or exec,
246 * P_PPWAIT is set; we will wakeup the parent below.
250 p->p_flag &= ~(P_TRACED | P_PPWAIT);
253 * Stop the real interval timer. If the handler is currently
254 * executing, prevent it from rearming itself and let it finish.
256 if (timevalisset(&p->p_realtimer.it_value) &&
257 callout_stop(&p->p_itcallout) == 0) {
258 timevalclear(&p->p_realtimer.it_interval);
259 msleep(&p->p_itcallout, &p->p_mtx, PWAIT, "ritwait", 0);
260 KASSERT(!timevalisset(&p->p_realtimer.it_value),
261 ("realtime timer is still armed"));
266 * Reset any sigio structures pointing to us as a result of
267 * F_SETOWN with our pid.
269 funsetownlst(&p->p_sigiolst);
272 * If this process has an nlminfo data area (for lockd), release it
274 if (nlminfo_release_p != NULL && p->p_nlminfo != NULL)
275 (*nlminfo_release_p)(p);
278 * Close open files and release open-file table.
284 * If this thread tickled GEOM, we need to wait for the giggling to
285 * stop before we return to userland
287 if (td->td_pflags & TDP_GEOM)
291 * Remove ourself from our leader's peer list and wake our leader.
293 mtx_lock(&ppeers_lock);
294 if (p->p_leader->p_peers) {
296 while (q->p_peers != p)
298 q->p_peers = p->p_peers;
301 mtx_unlock(&ppeers_lock);
305 sx_xlock(&proctree_lock);
306 if (SESS_LEADER(p)) {
307 struct session *sp = p->p_session;
311 * s_ttyp is not zero'd; we use this to indicate that
312 * the session once had a controlling terminal. (for
313 * logging and informational purposes)
324 * Signal foreground pgrp and revoke access to
325 * controlling terminal if it has not been revoked
328 * Because the TTY may have been revoked in the mean
329 * time and could already have a new session associated
330 * with it, make sure we don't send a SIGHUP to a
331 * foreground process group that does not belong to this
337 if (tp->t_session == sp)
338 tty_signal_pgrp(tp, SIGHUP);
343 sx_xunlock(&proctree_lock);
344 if (vn_lock(ttyvp, LK_EXCLUSIVE) == 0) {
345 VOP_REVOKE(ttyvp, REVOKEALL);
346 VOP_UNLOCK(ttyvp, 0);
348 sx_xlock(&proctree_lock);
351 fixjobc(p, p->p_pgrp, 0);
352 sx_xunlock(&proctree_lock);
353 (void)acct_process(td);
355 /* Release the TTY now we've unlocked everything. */
360 * Disable tracing, then drain any pending records and release
363 if (p->p_traceflag != 0) {
365 mtx_lock(&ktrace_mtx);
367 mtx_unlock(&ktrace_mtx);
371 mtx_lock(&ktrace_mtx);
372 tracevp = p->p_tracevp;
374 tracecred = p->p_tracecred;
375 p->p_tracecred = NULL;
376 mtx_unlock(&ktrace_mtx);
378 if (tracevp != NULL) {
379 locked = VFS_LOCK_GIANT(tracevp->v_mount);
381 VFS_UNLOCK_GIANT(locked);
383 if (tracecred != NULL)
388 * Release reference to text vnode
390 if ((vtmp = p->p_textvp) != NULL) {
392 locked = VFS_LOCK_GIANT(vtmp->v_mount);
394 VFS_UNLOCK_GIANT(locked);
398 * Release our limits structure.
407 * Remove proc from allproc queue and pidhash chain.
408 * Place onto zombproc. Unlink from parent's child list.
410 sx_xlock(&allproc_lock);
411 LIST_REMOVE(p, p_list);
412 LIST_INSERT_HEAD(&zombproc, p, p_list);
413 LIST_REMOVE(p, p_hash);
414 sx_xunlock(&allproc_lock);
417 * Call machine-dependent code to release any
418 * machine-dependent resources other than the address space.
419 * The address space is released by "vmspace_exitfree(p)" in
424 WITNESS_WARN(WARN_PANIC, NULL, "process (pid %d) exiting", p->p_pid);
427 * Reparent all of our children to init.
429 sx_xlock(&proctree_lock);
430 q = LIST_FIRST(&p->p_children);
431 if (q != NULL) /* only need this if any child is S_ZOMB */
433 for (; q != NULL; q = nq) {
434 nq = LIST_NEXT(q, p_sibling);
436 proc_reparent(q, initproc);
437 q->p_sigparent = SIGCHLD;
439 * Traced processes are killed
440 * since their existence means someone is screwing up.
442 if (q->p_flag & P_TRACED) {
445 q->p_flag &= ~(P_TRACED | P_STOPPED_TRACE);
446 FOREACH_THREAD_IN_PROC(q, temp)
447 temp->td_dbgflags &= ~TDB_SUSPEND;
453 /* Save exit status. */
458 /* Tell the prison that we are gone. */
459 prison_proc_free(p->p_ucred->cr_prison);
463 * Tell the DTrace fasttrap provider about the exit if it
464 * has declared an interest.
466 if (dtrace_fasttrap_exit)
467 dtrace_fasttrap_exit(p);
471 * Notify interested parties of our demise.
473 KNOTE_LOCKED(&p->p_klist, NOTE_EXIT);
476 int reason = CLD_EXITED;
479 else if (WIFSIGNALED(rv))
481 SDT_PROBE(proc, kernel, , exit, reason, 0, 0, 0, 0);
485 * Just delete all entries in the p_klist. At this point we won't
486 * report any more events, and there are nasty race conditions that
487 * can beat us if we don't.
489 knlist_clear(&p->p_klist, 1);
492 * Notify parent that we're gone. If parent has the PS_NOCLDWAIT
493 * flag set, or if the handler is set to SIG_IGN, notify process
494 * 1 instead (and hope it will handle this situation).
496 PROC_LOCK(p->p_pptr);
497 mtx_lock(&p->p_pptr->p_sigacts->ps_mtx);
498 if (p->p_pptr->p_sigacts->ps_flag & (PS_NOCLDWAIT | PS_CLDSIGIGN)) {
501 mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx);
504 proc_reparent(p, initproc);
505 p->p_sigparent = SIGCHLD;
506 PROC_LOCK(p->p_pptr);
509 * Notify parent, so in case he was wait(2)ing or
510 * executing waitpid(2) with our pid, he will
515 mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx);
517 if (p->p_pptr == initproc)
518 psignal(p->p_pptr, SIGCHLD);
519 else if (p->p_sigparent != 0) {
520 if (p->p_sigparent == SIGCHLD)
522 else /* LINUX thread */
523 psignal(p->p_pptr, p->p_sigparent);
525 sx_xunlock(&proctree_lock);
528 * The state PRS_ZOMBIE prevents other proesses from sending
529 * signal to the process, to avoid memory leak, we free memory
530 * for signal queue at the time when the state is set.
532 sigqueue_flush(&p->p_sigqueue);
533 sigqueue_flush(&td->td_sigqueue);
536 * We have to wait until after acquiring all locks before
537 * changing p_state. We need to avoid all possible context
538 * switches (including ones from blocking on a mutex) while
539 * marked as a zombie. We also have to set the zombie state
540 * before we release the parent process' proc lock to avoid
541 * a lost wakeup. So, we first call wakeup, then we grab the
542 * sched lock, update the state, and release the parent process'
546 cv_broadcast(&p->p_pwait);
547 sched_exit(p->p_pptr, td);
549 p->p_state = PRS_ZOMBIE;
550 PROC_UNLOCK(p->p_pptr);
553 * Hopefully no one will try to deliver a signal to the process this
556 knlist_destroy(&p->p_klist);
559 * Save our children's rusage information in our exit rusage.
561 ruadd(&p->p_ru, &p->p_rux, &p->p_stats->p_cru, &p->p_crux);
564 * Make sure the scheduler takes this thread out of its tables etc.
565 * This will also release this thread's reference to the ucred.
566 * Other thread parts to release include pcb bits and such.
572 #ifndef _SYS_SYSPROTO_H_
581 abort2(struct thread *td, struct abort2_args *uap)
583 struct proc *p = td->td_proc;
589 * Do it right now so we can log either proper call of abort2(), or
590 * note, that invalid argument was passed. 512 is big enough to
591 * handle 16 arguments' descriptions with additional comments.
593 sb = sbuf_new(NULL, NULL, 512, SBUF_FIXEDLEN);
595 sbuf_printf(sb, "%s(pid %d uid %d) aborted: ",
596 p->p_comm, p->p_pid, td->td_ucred->cr_uid);
598 * Since we can't return from abort2(), send SIGKILL in cases, where
599 * abort2() was called improperly
602 /* Prevent from DoSes from user-space. */
603 if (uap->nargs < 0 || uap->nargs > 16)
605 if (uap->nargs > 0) {
606 if (uap->args == NULL)
608 error = copyin(uap->args, uargs, uap->nargs * sizeof(void *));
613 * Limit size of 'reason' string to 128. Will fit even when
614 * maximal number of arguments was chosen to be logged.
616 if (uap->why != NULL) {
617 error = sbuf_copyin(sb, uap->why, 128);
621 sbuf_printf(sb, "(null)");
623 if (uap->nargs > 0) {
624 sbuf_printf(sb, "(");
625 for (i = 0;i < uap->nargs; i++)
626 sbuf_printf(sb, "%s%p", i == 0 ? "" : ", ", uargs[i]);
627 sbuf_printf(sb, ")");
630 * Final stage: arguments were proper, string has been
631 * successfully copied from userspace, and copying pointers
632 * from user-space succeed.
636 if (sig == SIGKILL) {
638 sbuf_printf(sb, " (Reason text inaccessible)");
642 log(LOG_INFO, "%s", sbuf_data(sb));
644 exit1(td, W_EXITCODE(0, sig));
651 * The dirty work is handled by kern_wait().
654 owait(struct thread *td, struct owait_args *uap __unused)
658 error = kern_wait(td, WAIT_ANY, &status, 0, NULL);
660 td->td_retval[1] = status;
663 #endif /* COMPAT_43 */
666 * The dirty work is handled by kern_wait().
669 wait4(struct thread *td, struct wait_args *uap)
671 struct rusage ru, *rup;
674 if (uap->rusage != NULL)
678 error = kern_wait(td, uap->pid, &status, uap->options, rup);
679 if (uap->status != NULL && error == 0)
680 error = copyout(&status, uap->status, sizeof(status));
681 if (uap->rusage != NULL && error == 0)
682 error = copyout(&ru, uap->rusage, sizeof(struct rusage));
687 * Reap the remains of a zombie process and optionally return status and
688 * rusage. Asserts and will release both the proctree_lock and the process
689 * lock as part of its work.
692 proc_reap(struct thread *td, struct proc *p, int *status, int options,
693 struct rusage *rusage)
697 sx_assert(&proctree_lock, SA_XLOCKED);
698 PROC_LOCK_ASSERT(p, MA_OWNED);
699 PROC_SLOCK_ASSERT(p, MA_OWNED);
700 KASSERT(p->p_state == PRS_ZOMBIE, ("proc_reap: !PRS_ZOMBIE"));
705 calcru(p, &rusage->ru_utime, &rusage->ru_stime);
708 td->td_retval[0] = p->p_pid;
710 *status = p->p_xstat; /* convert to int */
711 if (options & WNOWAIT) {
713 * Only poll, returning the status. Caller does not wish to
714 * release the proc struct just yet.
717 sx_xunlock(&proctree_lock);
722 sigqueue_take(p->p_ksi);
727 * If we got the child via a ptrace 'attach', we need to give it back
730 if (p->p_oppid && (t = pfind(p->p_oppid)) != NULL) {
735 pksignal(t, SIGCHLD, p->p_ksi);
737 cv_broadcast(&p->p_pwait);
739 sx_xunlock(&proctree_lock);
744 * Remove other references to this process to ensure we have an
745 * exclusive reference.
747 sx_xlock(&allproc_lock);
748 LIST_REMOVE(p, p_list); /* off zombproc */
749 sx_xunlock(&allproc_lock);
750 LIST_REMOVE(p, p_sibling);
752 sx_xunlock(&proctree_lock);
755 * As a side effect of this lock, we know that all other writes to
756 * this proc are visible now, so no more locking is needed for p.
759 p->p_xstat = 0; /* XXX: why? */
762 ruadd(&q->p_stats->p_cru, &q->p_crux, &p->p_ru, &p->p_rux);
766 * Decrement the count of procs running with this uid.
768 (void)chgproccnt(p->p_ucred->cr_ruidinfo, -1, 0);
771 * Free credentials, arguments, and sigacts.
775 pargs_drop(p->p_args);
777 sigacts_free(p->p_sigacts);
781 * Do any thread-system specific cleanups.
786 * Give vm and machine-dependent layer a chance to free anything that
787 * cpu_exit couldn't release while still running in process context.
793 KASSERT(FIRST_THREAD_IN_PROC(p),
794 ("proc_reap: no residual thread!"));
795 uma_zfree(proc_zone, p);
796 sx_xlock(&allproc_lock);
798 sx_xunlock(&allproc_lock);
802 kern_wait(struct thread *td, pid_t pid, int *status, int options,
803 struct rusage *rusage)
809 AUDIT_ARG_VALUE(options);
817 if (options &~ (WUNTRACED|WNOHANG|WCONTINUED|WNOWAIT|WLINUXCLONE))
820 if (q->p_flag & P_STATCHILD) {
822 q->p_flag &= ~P_STATCHILD;
826 sx_xlock(&proctree_lock);
827 LIST_FOREACH(p, &q->p_children, p_sibling) {
829 if (pid != WAIT_ANY &&
830 p->p_pid != pid && p->p_pgid != -pid) {
834 if (p_canwait(td, p)) {
840 * This special case handles a kthread spawned by linux_clone
841 * (see linux_misc.c). The linux_wait4 and linux_waitpid
842 * functions need to be able to distinguish between waiting
843 * on a process and waiting on a thread. It is a thread if
844 * p_sigparent is not SIGCHLD, and the WLINUXCLONE option
845 * signifies we want to wait for threads and not processes.
847 if ((p->p_sigparent != SIGCHLD) ^
848 ((options & WLINUXCLONE) != 0)) {
855 if (p->p_state == PRS_ZOMBIE) {
856 proc_reap(td, p, status, options, rusage);
859 if ((p->p_flag & P_STOPPED_SIG) &&
860 (p->p_suspcount == p->p_numthreads) &&
861 (p->p_flag & P_WAITED) == 0 &&
862 (p->p_flag & P_TRACED || options & WUNTRACED)) {
864 p->p_flag |= P_WAITED;
865 sx_xunlock(&proctree_lock);
866 td->td_retval[0] = p->p_pid;
868 *status = W_STOPCODE(p->p_xstat);
871 sigqueue_take(p->p_ksi);
878 if (options & WCONTINUED && (p->p_flag & P_CONTINUED)) {
879 sx_xunlock(&proctree_lock);
880 td->td_retval[0] = p->p_pid;
881 p->p_flag &= ~P_CONTINUED;
884 sigqueue_take(p->p_ksi);
895 sx_xunlock(&proctree_lock);
898 if (options & WNOHANG) {
899 sx_xunlock(&proctree_lock);
900 td->td_retval[0] = 0;
904 sx_xunlock(&proctree_lock);
905 if (q->p_flag & P_STATCHILD) {
906 q->p_flag &= ~P_STATCHILD;
909 error = msleep(q, &q->p_mtx, PWAIT | PCATCH, "wait", 0);
917 * Make process 'parent' the new parent of process 'child'.
918 * Must be called with an exclusive hold of proctree lock.
921 proc_reparent(struct proc *child, struct proc *parent)
924 sx_assert(&proctree_lock, SX_XLOCKED);
925 PROC_LOCK_ASSERT(child, MA_OWNED);
926 if (child->p_pptr == parent)
929 PROC_LOCK(child->p_pptr);
930 sigqueue_take(child->p_ksi);
931 PROC_UNLOCK(child->p_pptr);
932 LIST_REMOVE(child, p_sibling);
933 LIST_INSERT_HEAD(&parent->p_children, child, p_sibling);
934 child->p_pptr = parent;