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;
127 mtx_assert(&Giant, MA_NOTOWNED);
131 * XXX in case we're rebooting we just let init die in order to
132 * work around an unsolved stack overflow seen very late during
133 * shutdown on sparc64 when the gmirror worker process exists.
135 if (p == initproc && rebooting == 0) {
136 printf("init died (signal %d, exit %d)\n",
137 WTERMSIG(rv), WEXITSTATUS(rv));
138 panic("Going nowhere without my init!");
142 * MUST abort all other threads before proceeding past here.
145 while (p->p_flag & P_HADTHREADS) {
147 * First check if some other thread got here before us..
148 * if so, act apropriatly, (exit or suspend);
150 thread_suspend_check(0);
153 * Kill off the other threads. This requires
154 * some co-operation from other parts of the kernel
155 * so it may not be instantaneous. With this state set
156 * any thread entering the kernel from userspace will
157 * thread_exit() in trap(). Any thread attempting to
158 * sleep will return immediately with EINTR or EWOULDBLOCK
159 * which will hopefully force them to back out to userland
160 * freeing resources as they go. Any thread attempting
161 * to return to userland will thread_exit() from userret().
162 * thread_exit() will unsuspend us when the last of the
163 * other threads exits.
164 * If there is already a thread singler after resumption,
165 * calling thread_single will fail; in that case, we just
166 * re-check all suspension request, the thread should
167 * either be suspended there or exit.
169 if (! thread_single(SINGLE_EXIT))
173 * All other activity in this process is now stopped.
174 * Threading support has been turned off.
177 KASSERT(p->p_numthreads == 1,
178 ("exit1: proc %p exiting with %d threads", p, p->p_numthreads));
180 * Wakeup anyone in procfs' PIOCWAIT. They should have a hold
181 * on our vmspace, so we should block below until they have
182 * released their reference to us. Note that if they have
183 * requested S_EXIT stops we will block here until they ack
186 _STOPEVENT(p, S_EXIT, rv);
189 * Note that we are exiting and do another wakeup of anyone in
190 * PIOCWAIT in case they aren't listening for S_EXIT stops or
191 * decided to wait again after we told them we are exiting.
193 p->p_flag |= P_WEXIT;
197 * Wait for any processes that have a hold on our vmspace to
198 * release their reference.
200 while (p->p_lock > 0)
201 msleep(&p->p_lock, &p->p_mtx, PWAIT, "exithold", 0);
203 p->p_xstat = rv; /* Let event handler change exit status */
205 /* Drain the limit callout while we don't have the proc locked */
206 callout_drain(&p->p_limco);
210 * The Sun BSM exit token contains two components: an exit status as
211 * passed to exit(), and a return value to indicate what sort of exit
212 * it was. The exit status is WEXITSTATUS(rv), but it's not clear
213 * what the return value is.
215 AUDIT_ARG_EXIT(WEXITSTATUS(rv), 0);
216 AUDIT_SYSCALL_EXIT(0, td);
219 /* Are we a task leader? */
220 if (p == p->p_leader) {
221 mtx_lock(&ppeers_lock);
229 while (p->p_peers != NULL)
230 msleep(p, &ppeers_lock, PWAIT, "exit1", 0);
231 mtx_unlock(&ppeers_lock);
235 * Check if any loadable modules need anything done at process exit.
236 * E.g. SYSV IPC stuff
237 * XXX what if one of these generates an error?
239 EVENTHANDLER_INVOKE(process_exit, p);
242 * If parent is waiting for us to exit or exec,
243 * P_PPWAIT is set; we will wakeup the parent below.
246 rv = p->p_xstat; /* Event handler could change exit status */
248 p->p_flag &= ~(P_TRACED | P_PPWAIT);
251 * Stop the real interval timer. If the handler is currently
252 * executing, prevent it from rearming itself and let it finish.
254 if (timevalisset(&p->p_realtimer.it_value) &&
255 callout_stop(&p->p_itcallout) == 0) {
256 timevalclear(&p->p_realtimer.it_interval);
257 msleep(&p->p_itcallout, &p->p_mtx, PWAIT, "ritwait", 0);
258 KASSERT(!timevalisset(&p->p_realtimer.it_value),
259 ("realtime timer is still armed"));
264 * Reset any sigio structures pointing to us as a result of
265 * F_SETOWN with our pid.
267 funsetownlst(&p->p_sigiolst);
270 * If this process has an nlminfo data area (for lockd), release it
272 if (nlminfo_release_p != NULL && p->p_nlminfo != NULL)
273 (*nlminfo_release_p)(p);
276 * Close open files and release open-file table.
282 * If this thread tickled GEOM, we need to wait for the giggling to
283 * stop before we return to userland
285 if (td->td_pflags & TDP_GEOM)
289 * Remove ourself from our leader's peer list and wake our leader.
291 mtx_lock(&ppeers_lock);
292 if (p->p_leader->p_peers) {
294 while (q->p_peers != p)
296 q->p_peers = p->p_peers;
299 mtx_unlock(&ppeers_lock);
303 sx_xlock(&proctree_lock);
304 if (SESS_LEADER(p)) {
305 struct session *sp = p->p_session;
309 * s_ttyp is not zero'd; we use this to indicate that
310 * the session once had a controlling terminal. (for
311 * logging and informational purposes)
321 * Signal foreground pgrp and revoke access to
322 * controlling terminal if it has not been revoked
325 * Because the TTY may have been revoked in the mean
326 * time and could already have a new session associated
327 * with it, make sure we don't send a SIGHUP to a
328 * foreground process group that does not belong to this
334 if (tp->t_session == sp)
335 tty_signal_pgrp(tp, SIGHUP);
340 sx_xunlock(&proctree_lock);
341 if (vn_lock(ttyvp, LK_EXCLUSIVE) == 0) {
342 VOP_REVOKE(ttyvp, REVOKEALL);
343 VOP_UNLOCK(ttyvp, 0);
345 sx_xlock(&proctree_lock);
348 fixjobc(p, p->p_pgrp, 0);
349 sx_xunlock(&proctree_lock);
350 (void)acct_process(td);
352 /* Release the TTY now we've unlocked everything. */
359 * Release reference to text vnode
361 if ((vtmp = p->p_textvp) != NULL) {
363 locked = VFS_LOCK_GIANT(vtmp->v_mount);
365 VFS_UNLOCK_GIANT(locked);
369 * Release our limits structure.
378 * Remove proc from allproc queue and pidhash chain.
379 * Place onto zombproc. Unlink from parent's child list.
381 sx_xlock(&allproc_lock);
382 LIST_REMOVE(p, p_list);
383 LIST_INSERT_HEAD(&zombproc, p, p_list);
384 LIST_REMOVE(p, p_hash);
385 sx_xunlock(&allproc_lock);
388 * Call machine-dependent code to release any
389 * machine-dependent resources other than the address space.
390 * The address space is released by "vmspace_exitfree(p)" in
395 WITNESS_WARN(WARN_PANIC, NULL, "process (pid %d) exiting", p->p_pid);
398 * Reparent all of our children to init.
400 sx_xlock(&proctree_lock);
401 q = LIST_FIRST(&p->p_children);
402 if (q != NULL) /* only need this if any child is S_ZOMB */
404 for (; q != NULL; q = nq) {
405 nq = LIST_NEXT(q, p_sibling);
407 proc_reparent(q, initproc);
408 q->p_sigparent = SIGCHLD;
410 * Traced processes are killed
411 * since their existence means someone is screwing up.
413 if (q->p_flag & P_TRACED) {
416 q->p_flag &= ~(P_TRACED | P_STOPPED_TRACE);
417 FOREACH_THREAD_IN_PROC(q, temp)
418 temp->td_dbgflags &= ~TDB_SUSPEND;
424 /* Save exit status. */
428 /* Tell the prison that we are gone. */
429 prison_proc_free(p->p_ucred->cr_prison);
433 * Tell the DTrace fasttrap provider about the exit if it
434 * has declared an interest.
436 if (dtrace_fasttrap_exit)
437 dtrace_fasttrap_exit(p);
441 * Notify interested parties of our demise.
443 KNOTE_LOCKED(&p->p_klist, NOTE_EXIT);
446 int reason = CLD_EXITED;
449 else if (WIFSIGNALED(rv))
451 SDT_PROBE(proc, kernel, , exit, reason, 0, 0, 0, 0);
455 * Just delete all entries in the p_klist. At this point we won't
456 * report any more events, and there are nasty race conditions that
457 * can beat us if we don't.
459 knlist_clear(&p->p_klist, 1);
462 * Notify parent that we're gone. If parent has the PS_NOCLDWAIT
463 * flag set, or if the handler is set to SIG_IGN, notify process
464 * 1 instead (and hope it will handle this situation).
466 PROC_LOCK(p->p_pptr);
467 mtx_lock(&p->p_pptr->p_sigacts->ps_mtx);
468 if (p->p_pptr->p_sigacts->ps_flag & (PS_NOCLDWAIT | PS_CLDSIGIGN)) {
471 mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx);
474 proc_reparent(p, initproc);
475 p->p_sigparent = SIGCHLD;
476 PROC_LOCK(p->p_pptr);
479 * Notify parent, so in case he was wait(2)ing or
480 * executing waitpid(2) with our pid, he will
485 mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx);
487 if (p->p_pptr == initproc)
488 psignal(p->p_pptr, SIGCHLD);
489 else if (p->p_sigparent != 0) {
490 if (p->p_sigparent == SIGCHLD)
492 else /* LINUX thread */
493 psignal(p->p_pptr, p->p_sigparent);
495 sx_xunlock(&proctree_lock);
498 * The state PRS_ZOMBIE prevents other proesses from sending
499 * signal to the process, to avoid memory leak, we free memory
500 * for signal queue at the time when the state is set.
502 sigqueue_flush(&p->p_sigqueue);
503 sigqueue_flush(&td->td_sigqueue);
506 * We have to wait until after acquiring all locks before
507 * changing p_state. We need to avoid all possible context
508 * switches (including ones from blocking on a mutex) while
509 * marked as a zombie. We also have to set the zombie state
510 * before we release the parent process' proc lock to avoid
511 * a lost wakeup. So, we first call wakeup, then we grab the
512 * sched lock, update the state, and release the parent process'
516 cv_broadcast(&p->p_pwait);
517 sched_exit(p->p_pptr, td);
519 p->p_state = PRS_ZOMBIE;
520 PROC_UNLOCK(p->p_pptr);
523 * Hopefully no one will try to deliver a signal to the process this
526 knlist_destroy(&p->p_klist);
529 * Save our children's rusage information in our exit rusage.
531 ruadd(&p->p_ru, &p->p_rux, &p->p_stats->p_cru, &p->p_crux);
534 * Make sure the scheduler takes this thread out of its tables etc.
535 * This will also release this thread's reference to the ucred.
536 * Other thread parts to release include pcb bits and such.
542 #ifndef _SYS_SYSPROTO_H_
551 abort2(struct thread *td, struct abort2_args *uap)
553 struct proc *p = td->td_proc;
559 * Do it right now so we can log either proper call of abort2(), or
560 * note, that invalid argument was passed. 512 is big enough to
561 * handle 16 arguments' descriptions with additional comments.
563 sb = sbuf_new(NULL, NULL, 512, SBUF_FIXEDLEN);
565 sbuf_printf(sb, "%s(pid %d uid %d) aborted: ",
566 p->p_comm, p->p_pid, td->td_ucred->cr_uid);
568 * Since we can't return from abort2(), send SIGKILL in cases, where
569 * abort2() was called improperly
572 /* Prevent from DoSes from user-space. */
573 if (uap->nargs < 0 || uap->nargs > 16)
575 if (uap->nargs > 0) {
576 if (uap->args == NULL)
578 error = copyin(uap->args, uargs, uap->nargs * sizeof(void *));
583 * Limit size of 'reason' string to 128. Will fit even when
584 * maximal number of arguments was chosen to be logged.
586 if (uap->why != NULL) {
587 error = sbuf_copyin(sb, uap->why, 128);
591 sbuf_printf(sb, "(null)");
593 if (uap->nargs > 0) {
594 sbuf_printf(sb, "(");
595 for (i = 0;i < uap->nargs; i++)
596 sbuf_printf(sb, "%s%p", i == 0 ? "" : ", ", uargs[i]);
597 sbuf_printf(sb, ")");
600 * Final stage: arguments were proper, string has been
601 * successfully copied from userspace, and copying pointers
602 * from user-space succeed.
606 if (sig == SIGKILL) {
608 sbuf_printf(sb, " (Reason text inaccessible)");
612 log(LOG_INFO, "%s", sbuf_data(sb));
614 exit1(td, W_EXITCODE(0, sig));
621 * The dirty work is handled by kern_wait().
624 owait(struct thread *td, struct owait_args *uap __unused)
628 error = kern_wait(td, WAIT_ANY, &status, 0, NULL);
630 td->td_retval[1] = status;
633 #endif /* COMPAT_43 */
636 * The dirty work is handled by kern_wait().
639 wait4(struct thread *td, struct wait_args *uap)
641 struct rusage ru, *rup;
644 if (uap->rusage != NULL)
648 error = kern_wait(td, uap->pid, &status, uap->options, rup);
649 if (uap->status != NULL && error == 0)
650 error = copyout(&status, uap->status, sizeof(status));
651 if (uap->rusage != NULL && error == 0)
652 error = copyout(&ru, uap->rusage, sizeof(struct rusage));
657 * Reap the remains of a zombie process and optionally return status and
658 * rusage. Asserts and will release both the proctree_lock and the process
659 * lock as part of its work.
662 proc_reap(struct thread *td, struct proc *p, int *status, int options,
663 struct rusage *rusage)
667 sx_assert(&proctree_lock, SA_XLOCKED);
668 PROC_LOCK_ASSERT(p, MA_OWNED);
669 PROC_SLOCK_ASSERT(p, MA_OWNED);
670 KASSERT(p->p_state == PRS_ZOMBIE, ("proc_reap: !PRS_ZOMBIE"));
675 calcru(p, &rusage->ru_utime, &rusage->ru_stime);
678 td->td_retval[0] = p->p_pid;
680 *status = p->p_xstat; /* convert to int */
681 if (options & WNOWAIT) {
683 * Only poll, returning the status. Caller does not wish to
684 * release the proc struct just yet.
687 sx_xunlock(&proctree_lock);
692 sigqueue_take(p->p_ksi);
697 * If we got the child via a ptrace 'attach', we need to give it back
700 if (p->p_oppid && (t = pfind(p->p_oppid)) != NULL) {
705 tdsignal(t, NULL, SIGCHLD, p->p_ksi);
707 cv_broadcast(&p->p_pwait);
709 sx_xunlock(&proctree_lock);
714 * Remove other references to this process to ensure we have an
715 * exclusive reference.
717 sx_xlock(&allproc_lock);
718 LIST_REMOVE(p, p_list); /* off zombproc */
719 sx_xunlock(&allproc_lock);
720 LIST_REMOVE(p, p_sibling);
722 sx_xunlock(&proctree_lock);
725 * As a side effect of this lock, we know that all other writes to
726 * this proc are visible now, so no more locking is needed for p.
729 p->p_xstat = 0; /* XXX: why? */
732 ruadd(&q->p_stats->p_cru, &q->p_crux, &p->p_ru, &p->p_rux);
736 * Decrement the count of procs running with this uid.
738 (void)chgproccnt(p->p_ucred->cr_ruidinfo, -1, 0);
741 * Free credentials, arguments, and sigacts.
745 pargs_drop(p->p_args);
747 sigacts_free(p->p_sigacts);
751 * Do any thread-system specific cleanups.
756 * Give vm and machine-dependent layer a chance to free anything that
757 * cpu_exit couldn't release while still running in process context.
763 KASSERT(FIRST_THREAD_IN_PROC(p),
764 ("proc_reap: no residual thread!"));
765 uma_zfree(proc_zone, p);
766 sx_xlock(&allproc_lock);
768 sx_xunlock(&allproc_lock);
772 kern_wait(struct thread *td, pid_t pid, int *status, int options,
773 struct rusage *rusage)
779 AUDIT_ARG_VALUE(options);
787 if (options &~ (WUNTRACED|WNOHANG|WCONTINUED|WNOWAIT|WLINUXCLONE))
790 if (q->p_flag & P_STATCHILD) {
792 q->p_flag &= ~P_STATCHILD;
796 sx_xlock(&proctree_lock);
797 LIST_FOREACH(p, &q->p_children, p_sibling) {
799 if (pid != WAIT_ANY &&
800 p->p_pid != pid && p->p_pgid != -pid) {
804 if (p_canwait(td, p)) {
810 * This special case handles a kthread spawned by linux_clone
811 * (see linux_misc.c). The linux_wait4 and linux_waitpid
812 * functions need to be able to distinguish between waiting
813 * on a process and waiting on a thread. It is a thread if
814 * p_sigparent is not SIGCHLD, and the WLINUXCLONE option
815 * signifies we want to wait for threads and not processes.
817 if ((p->p_sigparent != SIGCHLD) ^
818 ((options & WLINUXCLONE) != 0)) {
825 if (p->p_state == PRS_ZOMBIE) {
826 proc_reap(td, p, status, options, rusage);
829 if ((p->p_flag & P_STOPPED_SIG) &&
830 (p->p_suspcount == p->p_numthreads) &&
831 (p->p_flag & P_WAITED) == 0 &&
832 (p->p_flag & P_TRACED || options & WUNTRACED)) {
834 p->p_flag |= P_WAITED;
835 sx_xunlock(&proctree_lock);
836 td->td_retval[0] = p->p_pid;
838 *status = W_STOPCODE(p->p_xstat);
841 sigqueue_take(p->p_ksi);
848 if (options & WCONTINUED && (p->p_flag & P_CONTINUED)) {
849 sx_xunlock(&proctree_lock);
850 td->td_retval[0] = p->p_pid;
851 p->p_flag &= ~P_CONTINUED;
854 sigqueue_take(p->p_ksi);
865 sx_xunlock(&proctree_lock);
868 if (options & WNOHANG) {
869 sx_xunlock(&proctree_lock);
870 td->td_retval[0] = 0;
874 sx_xunlock(&proctree_lock);
875 if (q->p_flag & P_STATCHILD) {
876 q->p_flag &= ~P_STATCHILD;
879 error = msleep(q, &q->p_mtx, PWAIT | PCATCH, "wait", 0);
887 * Make process 'parent' the new parent of process 'child'.
888 * Must be called with an exclusive hold of proctree lock.
891 proc_reparent(struct proc *child, struct proc *parent)
894 sx_assert(&proctree_lock, SX_XLOCKED);
895 PROC_LOCK_ASSERT(child, MA_OWNED);
896 if (child->p_pptr == parent)
899 PROC_LOCK(child->p_pptr);
900 sigqueue_take(child->p_ksi);
901 PROC_UNLOCK(child->p_pptr);
902 LIST_REMOVE(child, p_sibling);
903 LIST_INSERT_HEAD(&parent->p_children, child, p_sibling);
904 child->p_pptr = parent;