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/racct.h>
60 #include <sys/resourcevar.h>
62 #include <sys/signalvar.h>
63 #include <sys/sched.h>
65 #include <sys/syscallsubr.h>
66 #include <sys/syslog.h>
67 #include <sys/ptrace.h>
68 #include <sys/acct.h> /* for acct_process() function prototype */
69 #include <sys/filedesc.h>
74 #include <sys/ktrace.h>
77 #include <security/audit/audit.h>
78 #include <security/mac/mac_framework.h>
81 #include <vm/vm_extern.h>
82 #include <vm/vm_param.h>
84 #include <vm/vm_map.h>
85 #include <vm/vm_page.h>
89 #include <sys/dtrace_bsd.h>
90 dtrace_execexit_func_t dtrace_fasttrap_exit;
93 SDT_PROVIDER_DECLARE(proc);
94 SDT_PROBE_DEFINE(proc, kernel, , exit, exit);
95 SDT_PROBE_ARGTYPE(proc, kernel, , exit, 0, "int");
97 /* Required to be non-static for SysVR4 emulator */
98 MALLOC_DEFINE(M_ZOMBIE, "zombie", "zombie proc status");
100 /* Hook for NFS teardown procedure. */
101 void (*nlminfo_release_p)(struct proc *p);
104 * exit -- death of process.
107 sys_exit(struct thread *td, struct sys_exit_args *uap)
110 exit1(td, W_EXITCODE(uap->rval, 0));
115 * Exit: deallocate address space and other resources, change proc state to
116 * zombie, and unlink proc from allproc and parent's lists. Save exit status
117 * and rusage for wait(). Check for child processes and orphan them.
120 exit1(struct thread *td, int rv)
122 struct proc *p, *nq, *q;
124 struct vnode *ttyvp = NULL;
128 mtx_assert(&Giant, MA_NOTOWNED);
132 * XXX in case we're rebooting we just let init die in order to
133 * work around an unsolved stack overflow seen very late during
134 * shutdown on sparc64 when the gmirror worker process exists.
136 if (p == initproc && rebooting == 0) {
137 printf("init died (signal %d, exit %d)\n",
138 WTERMSIG(rv), WEXITSTATUS(rv));
139 panic("Going nowhere without my init!");
143 * MUST abort all other threads before proceeding past here.
146 while (p->p_flag & P_HADTHREADS) {
148 * First check if some other thread got here before us..
149 * if so, act apropriatly, (exit or suspend);
151 thread_suspend_check(0);
154 * Kill off the other threads. This requires
155 * some co-operation from other parts of the kernel
156 * so it may not be instantaneous. With this state set
157 * any thread entering the kernel from userspace will
158 * thread_exit() in trap(). Any thread attempting to
159 * sleep will return immediately with EINTR or EWOULDBLOCK
160 * which will hopefully force them to back out to userland
161 * freeing resources as they go. Any thread attempting
162 * to return to userland will thread_exit() from userret().
163 * thread_exit() will unsuspend us when the last of the
164 * other threads exits.
165 * If there is already a thread singler after resumption,
166 * calling thread_single will fail; in that case, we just
167 * re-check all suspension request, the thread should
168 * either be suspended there or exit.
170 if (! thread_single(SINGLE_EXIT))
174 * All other activity in this process is now stopped.
175 * Threading support has been turned off.
178 KASSERT(p->p_numthreads == 1,
179 ("exit1: proc %p exiting with %d threads", p, p->p_numthreads));
180 racct_sub(p, RACCT_NTHR, 1);
182 * Wakeup anyone in procfs' PIOCWAIT. They should have a hold
183 * on our vmspace, so we should block below until they have
184 * released their reference to us. Note that if they have
185 * requested S_EXIT stops we will block here until they ack
188 _STOPEVENT(p, S_EXIT, rv);
191 * Note that we are exiting and do another wakeup of anyone in
192 * PIOCWAIT in case they aren't listening for S_EXIT stops or
193 * decided to wait again after we told them we are exiting.
195 p->p_flag |= P_WEXIT;
199 * Wait for any processes that have a hold on our vmspace to
200 * release their reference.
202 while (p->p_lock > 0)
203 msleep(&p->p_lock, &p->p_mtx, PWAIT, "exithold", 0);
205 p->p_xstat = rv; /* Let event handler change exit status */
207 /* Drain the limit callout while we don't have the proc locked */
208 callout_drain(&p->p_limco);
212 * The Sun BSM exit token contains two components: an exit status as
213 * passed to exit(), and a return value to indicate what sort of exit
214 * it was. The exit status is WEXITSTATUS(rv), but it's not clear
215 * what the return value is.
217 AUDIT_ARG_EXIT(WEXITSTATUS(rv), 0);
218 AUDIT_SYSCALL_EXIT(0, td);
221 /* Are we a task leader? */
222 if (p == p->p_leader) {
223 mtx_lock(&ppeers_lock);
231 while (p->p_peers != NULL)
232 msleep(p, &ppeers_lock, PWAIT, "exit1", 0);
233 mtx_unlock(&ppeers_lock);
237 * Check if any loadable modules need anything done at process exit.
238 * E.g. SYSV IPC stuff
239 * XXX what if one of these generates an error?
241 EVENTHANDLER_INVOKE(process_exit, p);
244 * If parent is waiting for us to exit or exec,
245 * P_PPWAIT is set; we will wakeup the parent below.
248 rv = p->p_xstat; /* Event handler could change exit status */
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. */
362 * Release reference to text vnode
364 if ((vtmp = p->p_textvp) != NULL) {
366 locked = VFS_LOCK_GIANT(vtmp->v_mount);
368 VFS_UNLOCK_GIANT(locked);
372 * Release our limits structure.
383 * Remove proc from allproc queue and pidhash chain.
384 * Place onto zombproc. Unlink from parent's child list.
386 sx_xlock(&allproc_lock);
387 LIST_REMOVE(p, p_list);
388 LIST_INSERT_HEAD(&zombproc, p, p_list);
389 LIST_REMOVE(p, p_hash);
390 sx_xunlock(&allproc_lock);
393 * Call machine-dependent code to release any
394 * machine-dependent resources other than the address space.
395 * The address space is released by "vmspace_exitfree(p)" in
400 WITNESS_WARN(WARN_PANIC, NULL, "process (pid %d) exiting", p->p_pid);
403 * Reparent all of our children to init.
405 sx_xlock(&proctree_lock);
406 q = LIST_FIRST(&p->p_children);
407 if (q != NULL) /* only need this if any child is S_ZOMB */
409 for (; q != NULL; q = nq) {
410 nq = LIST_NEXT(q, p_sibling);
412 proc_reparent(q, initproc);
413 q->p_sigparent = SIGCHLD;
415 * Traced processes are killed
416 * since their existence means someone is screwing up.
418 if (q->p_flag & P_TRACED) {
421 q->p_flag &= ~(P_TRACED | P_STOPPED_TRACE);
422 FOREACH_THREAD_IN_PROC(q, temp)
423 temp->td_dbgflags &= ~TDB_SUSPEND;
429 /* Save exit status. */
433 /* Tell the prison that we are gone. */
434 prison_proc_free(p->p_ucred->cr_prison);
438 * Tell the DTrace fasttrap provider about the exit if it
439 * has declared an interest.
441 if (dtrace_fasttrap_exit)
442 dtrace_fasttrap_exit(p);
446 * Notify interested parties of our demise.
448 KNOTE_LOCKED(&p->p_klist, NOTE_EXIT);
451 int reason = CLD_EXITED;
454 else if (WIFSIGNALED(rv))
456 SDT_PROBE(proc, kernel, , exit, reason, 0, 0, 0, 0);
460 * Just delete all entries in the p_klist. At this point we won't
461 * report any more events, and there are nasty race conditions that
462 * can beat us if we don't.
464 knlist_clear(&p->p_klist, 1);
467 * Notify parent that we're gone. If parent has the PS_NOCLDWAIT
468 * flag set, or if the handler is set to SIG_IGN, notify process
469 * 1 instead (and hope it will handle this situation).
471 PROC_LOCK(p->p_pptr);
472 mtx_lock(&p->p_pptr->p_sigacts->ps_mtx);
473 if (p->p_pptr->p_sigacts->ps_flag & (PS_NOCLDWAIT | PS_CLDSIGIGN)) {
476 mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx);
479 proc_reparent(p, initproc);
480 p->p_sigparent = SIGCHLD;
481 PROC_LOCK(p->p_pptr);
484 * Notify parent, so in case he was wait(2)ing or
485 * executing waitpid(2) with our pid, he will
490 mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx);
492 if (p->p_pptr == initproc)
493 psignal(p->p_pptr, SIGCHLD);
494 else if (p->p_sigparent != 0) {
495 if (p->p_sigparent == SIGCHLD)
497 else /* LINUX thread */
498 psignal(p->p_pptr, p->p_sigparent);
500 sx_xunlock(&proctree_lock);
503 * The state PRS_ZOMBIE prevents other proesses from sending
504 * signal to the process, to avoid memory leak, we free memory
505 * for signal queue at the time when the state is set.
507 sigqueue_flush(&p->p_sigqueue);
508 sigqueue_flush(&td->td_sigqueue);
511 * We have to wait until after acquiring all locks before
512 * changing p_state. We need to avoid all possible context
513 * switches (including ones from blocking on a mutex) while
514 * marked as a zombie. We also have to set the zombie state
515 * before we release the parent process' proc lock to avoid
516 * a lost wakeup. So, we first call wakeup, then we grab the
517 * sched lock, update the state, and release the parent process'
521 cv_broadcast(&p->p_pwait);
522 sched_exit(p->p_pptr, td);
524 p->p_state = PRS_ZOMBIE;
525 PROC_UNLOCK(p->p_pptr);
528 * Hopefully no one will try to deliver a signal to the process this
531 knlist_destroy(&p->p_klist);
534 * Save our children's rusage information in our exit rusage.
536 ruadd(&p->p_ru, &p->p_rux, &p->p_stats->p_cru, &p->p_crux);
539 * Make sure the scheduler takes this thread out of its tables etc.
540 * This will also release this thread's reference to the ucred.
541 * Other thread parts to release include pcb bits and such.
547 #ifndef _SYS_SYSPROTO_H_
556 abort2(struct thread *td, struct abort2_args *uap)
558 struct proc *p = td->td_proc;
564 * Do it right now so we can log either proper call of abort2(), or
565 * note, that invalid argument was passed. 512 is big enough to
566 * handle 16 arguments' descriptions with additional comments.
568 sb = sbuf_new(NULL, NULL, 512, SBUF_FIXEDLEN);
570 sbuf_printf(sb, "%s(pid %d uid %d) aborted: ",
571 p->p_comm, p->p_pid, td->td_ucred->cr_uid);
573 * Since we can't return from abort2(), send SIGKILL in cases, where
574 * abort2() was called improperly
577 /* Prevent from DoSes from user-space. */
578 if (uap->nargs < 0 || uap->nargs > 16)
580 if (uap->nargs > 0) {
581 if (uap->args == NULL)
583 error = copyin(uap->args, uargs, uap->nargs * sizeof(void *));
588 * Limit size of 'reason' string to 128. Will fit even when
589 * maximal number of arguments was chosen to be logged.
591 if (uap->why != NULL) {
592 error = sbuf_copyin(sb, uap->why, 128);
596 sbuf_printf(sb, "(null)");
598 if (uap->nargs > 0) {
599 sbuf_printf(sb, "(");
600 for (i = 0;i < uap->nargs; i++)
601 sbuf_printf(sb, "%s%p", i == 0 ? "" : ", ", uargs[i]);
602 sbuf_printf(sb, ")");
605 * Final stage: arguments were proper, string has been
606 * successfully copied from userspace, and copying pointers
607 * from user-space succeed.
611 if (sig == SIGKILL) {
613 sbuf_printf(sb, " (Reason text inaccessible)");
617 log(LOG_INFO, "%s", sbuf_data(sb));
619 exit1(td, W_EXITCODE(0, sig));
626 * The dirty work is handled by kern_wait().
629 owait(struct thread *td, struct owait_args *uap __unused)
633 error = kern_wait(td, WAIT_ANY, &status, 0, NULL);
635 td->td_retval[1] = status;
638 #endif /* COMPAT_43 */
641 * The dirty work is handled by kern_wait().
644 wait4(struct thread *td, struct wait_args *uap)
646 struct rusage ru, *rup;
649 if (uap->rusage != NULL)
653 error = kern_wait(td, uap->pid, &status, uap->options, rup);
654 if (uap->status != NULL && error == 0)
655 error = copyout(&status, uap->status, sizeof(status));
656 if (uap->rusage != NULL && error == 0)
657 error = copyout(&ru, uap->rusage, sizeof(struct rusage));
662 * Reap the remains of a zombie process and optionally return status and
663 * rusage. Asserts and will release both the proctree_lock and the process
664 * lock as part of its work.
667 proc_reap(struct thread *td, struct proc *p, int *status, int options,
668 struct rusage *rusage)
672 sx_assert(&proctree_lock, SA_XLOCKED);
673 PROC_LOCK_ASSERT(p, MA_OWNED);
674 PROC_SLOCK_ASSERT(p, MA_OWNED);
675 KASSERT(p->p_state == PRS_ZOMBIE, ("proc_reap: !PRS_ZOMBIE"));
680 calcru(p, &rusage->ru_utime, &rusage->ru_stime);
683 td->td_retval[0] = p->p_pid;
685 *status = p->p_xstat; /* convert to int */
686 if (options & WNOWAIT) {
688 * Only poll, returning the status. Caller does not wish to
689 * release the proc struct just yet.
692 sx_xunlock(&proctree_lock);
697 sigqueue_take(p->p_ksi);
702 * If we got the child via a ptrace 'attach', we need to give it back
705 if (p->p_oppid && (t = pfind(p->p_oppid)) != NULL) {
710 pksignal(t, SIGCHLD, p->p_ksi);
712 cv_broadcast(&p->p_pwait);
714 sx_xunlock(&proctree_lock);
719 * Remove other references to this process to ensure we have an
720 * exclusive reference.
722 sx_xlock(&allproc_lock);
723 LIST_REMOVE(p, p_list); /* off zombproc */
724 sx_xunlock(&allproc_lock);
725 LIST_REMOVE(p, p_sibling);
727 sx_xunlock(&proctree_lock);
730 * As a side effect of this lock, we know that all other writes to
731 * this proc are visible now, so no more locking is needed for p.
734 p->p_xstat = 0; /* XXX: why? */
737 ruadd(&q->p_stats->p_cru, &q->p_crux, &p->p_ru, &p->p_rux);
741 * Decrement the count of procs running with this uid.
743 (void)chgproccnt(p->p_ucred->cr_ruidinfo, -1, 0);
746 * Destroy resource accounting information associated with the process.
749 PROC_LOCK(p->p_pptr);
750 racct_sub(p->p_pptr, RACCT_NPROC, 1);
751 PROC_UNLOCK(p->p_pptr);
754 * Free credentials, arguments, and sigacts.
758 pargs_drop(p->p_args);
760 sigacts_free(p->p_sigacts);
764 * Do any thread-system specific cleanups.
769 * Give vm and machine-dependent layer a chance to free anything that
770 * cpu_exit couldn't release while still running in process context.
776 KASSERT(FIRST_THREAD_IN_PROC(p),
777 ("proc_reap: no residual thread!"));
778 uma_zfree(proc_zone, p);
779 sx_xlock(&allproc_lock);
781 sx_xunlock(&allproc_lock);
785 kern_wait(struct thread *td, pid_t pid, int *status, int options,
786 struct rusage *rusage)
792 AUDIT_ARG_VALUE(options);
800 if (options &~ (WUNTRACED|WNOHANG|WCONTINUED|WNOWAIT|WLINUXCLONE))
803 if (q->p_flag & P_STATCHILD) {
805 q->p_flag &= ~P_STATCHILD;
809 sx_xlock(&proctree_lock);
810 LIST_FOREACH(p, &q->p_children, p_sibling) {
812 if (pid != WAIT_ANY &&
813 p->p_pid != pid && p->p_pgid != -pid) {
817 if (p_canwait(td, p)) {
823 * This special case handles a kthread spawned by linux_clone
824 * (see linux_misc.c). The linux_wait4 and linux_waitpid
825 * functions need to be able to distinguish between waiting
826 * on a process and waiting on a thread. It is a thread if
827 * p_sigparent is not SIGCHLD, and the WLINUXCLONE option
828 * signifies we want to wait for threads and not processes.
830 if ((p->p_sigparent != SIGCHLD) ^
831 ((options & WLINUXCLONE) != 0)) {
838 if (p->p_state == PRS_ZOMBIE) {
839 proc_reap(td, p, status, options, rusage);
842 if ((p->p_flag & P_STOPPED_SIG) &&
843 (p->p_suspcount == p->p_numthreads) &&
844 (p->p_flag & P_WAITED) == 0 &&
845 (p->p_flag & P_TRACED || options & WUNTRACED)) {
847 p->p_flag |= P_WAITED;
848 sx_xunlock(&proctree_lock);
849 td->td_retval[0] = p->p_pid;
851 *status = W_STOPCODE(p->p_xstat);
854 sigqueue_take(p->p_ksi);
861 if (options & WCONTINUED && (p->p_flag & P_CONTINUED)) {
862 sx_xunlock(&proctree_lock);
863 td->td_retval[0] = p->p_pid;
864 p->p_flag &= ~P_CONTINUED;
867 sigqueue_take(p->p_ksi);
878 sx_xunlock(&proctree_lock);
881 if (options & WNOHANG) {
882 sx_xunlock(&proctree_lock);
883 td->td_retval[0] = 0;
887 sx_xunlock(&proctree_lock);
888 if (q->p_flag & P_STATCHILD) {
889 q->p_flag &= ~P_STATCHILD;
892 error = msleep(q, &q->p_mtx, PWAIT | PCATCH, "wait", 0);
900 * Make process 'parent' the new parent of process 'child'.
901 * Must be called with an exclusive hold of proctree lock.
904 proc_reparent(struct proc *child, struct proc *parent)
908 sx_assert(&proctree_lock, SX_XLOCKED);
909 PROC_LOCK_ASSERT(child, MA_OWNED);
910 if (child->p_pptr == parent)
913 locked = PROC_LOCKED(parent);
916 racct_add_force(parent, RACCT_NPROC, 1);
919 PROC_LOCK(child->p_pptr);
920 racct_sub(child->p_pptr, RACCT_NPROC, 1);
921 sigqueue_take(child->p_ksi);
922 PROC_UNLOCK(child->p_pptr);
923 LIST_REMOVE(child, p_sibling);
924 LIST_INSERT_HEAD(&parent->p_children, child, p_sibling);
925 child->p_pptr = parent;