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_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>
56 #include <sys/vmmeter.h>
57 #include <sys/vnode.h>
58 #include <sys/resourcevar.h>
60 #include <sys/signalvar.h>
61 #include <sys/sched.h>
63 #include <sys/syscallsubr.h>
64 #include <sys/syslog.h>
65 #include <sys/ptrace.h>
66 #include <sys/acct.h> /* for acct_process() function prototype */
67 #include <sys/filedesc.h>
72 #include <sys/ktrace.h>
75 #include <security/audit/audit.h>
78 #include <vm/vm_extern.h>
79 #include <vm/vm_param.h>
81 #include <vm/vm_map.h>
82 #include <vm/vm_page.h>
85 /* Required to be non-static for SysVR4 emulator */
86 MALLOC_DEFINE(M_ZOMBIE, "zombie", "zombie proc status");
88 /* Hook for NFS teardown procedure. */
89 void (*nlminfo_release_p)(struct proc *p);
98 sys_exit(struct thread *td, struct sys_exit_args *uap)
101 exit1(td, W_EXITCODE(uap->rval, 0));
106 * Exit: deallocate address space and other resources, change proc state
107 * to zombie, and unlink proc from allproc and parent's lists. Save exit
108 * status and rusage for wait(). Check for child processes and orphan them.
111 exit1(struct thread *td, int rv)
113 struct proc *p, *nq, *q;
118 struct vnode *tracevp;
119 struct ucred *tracecred;
125 * Drop Giant if caller has it. Eventually we should warn about
126 * being called with Giant held.
128 while (mtx_owned(&Giant))
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 if (p->p_flag & P_HADTHREADS) {
145 * First check if some other thread got here before us..
146 * if so, act apropriatly, (exit or suspend);
148 thread_suspend_check(0);
151 * Kill off the other threads. This requires
152 * some co-operation from other parts of the kernel
153 * so it may not be instantaneous. With this state set
154 * any thread entering the kernel from userspace will
155 * thread_exit() in trap(). Any thread attempting to
156 * sleep will return immediately with EINTR or EWOULDBLOCK
157 * which will hopefully force them to back out to userland
158 * freeing resources as they go. Any thread attempting
159 * to return to userland will thread_exit() from userret().
160 * thread_exit() will unsuspend us when the last of the
161 * other threads exits.
162 * If there is already a thread singler after resumption,
163 * calling thread_single will fail; in that case, we just
164 * re-check all suspension request, the thread should
165 * either be suspended there or exit.
167 if (thread_single(SINGLE_EXIT))
171 * All other activity in this process is now stopped.
172 * Threading support has been turned off.
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 PROC_LOCK(p->p_pptr);
201 sigqueue_take(p->p_ksi);
202 PROC_UNLOCK(p->p_pptr);
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);
239 MALLOC(p->p_ru, struct rusage *, sizeof(struct rusage),
242 * If parent is waiting for us to exit or exec,
243 * P_PPWAIT is set; we will wakeup the parent below.
247 p->p_flag &= ~(P_TRACED | P_PPWAIT);
250 * Stop the real interval timer. If the handler is currently
251 * executing, prevent it from rearming itself and let it finish.
253 if (timevalisset(&p->p_realtimer.it_value) &&
254 callout_stop(&p->p_itcallout) == 0) {
255 timevalclear(&p->p_realtimer.it_interval);
256 msleep(&p->p_itcallout, &p->p_mtx, PWAIT, "ritwait", 0);
257 KASSERT(!timevalisset(&p->p_realtimer.it_value),
258 ("realtime timer is still armed"));
260 sigqueue_flush(&p->p_sigqueue);
261 sigqueue_flush(&td->td_sigqueue);
265 * Reset any sigio structures pointing to us as a result of
266 * F_SETOWN with our pid.
268 funsetownlst(&p->p_sigiolst);
271 * If this process has an nlminfo data area (for lockd), release it
273 if (nlminfo_release_p != NULL && p->p_nlminfo != NULL)
274 (*nlminfo_release_p)(p);
277 * Close open files and release open-file table.
283 * If this thread tickled GEOM, we need to wait for the giggling to
284 * stop before we return to userland
286 if (td->td_pflags & TDP_GEOM)
290 * Remove ourself from our leader's peer list and wake our leader.
292 mtx_lock(&ppeers_lock);
293 if (p->p_leader->p_peers) {
295 while (q->p_peers != p)
297 q->p_peers = p->p_peers;
300 mtx_unlock(&ppeers_lock);
304 mtx_lock(&Giant); /* XXX TTY */
305 sx_xlock(&proctree_lock);
306 if (SESS_LEADER(p)) {
312 * Controlling process.
313 * Signal foreground pgrp,
314 * drain controlling terminal
315 * and revoke access to controlling terminal.
317 if (sp->s_ttyp && (sp->s_ttyp->t_session == sp)) {
319 if (sp->s_ttyp->t_pgrp) {
320 PGRP_LOCK(sp->s_ttyp->t_pgrp);
321 pgsignal(sp->s_ttyp->t_pgrp, SIGHUP, 1);
322 PGRP_UNLOCK(sp->s_ttyp->t_pgrp);
324 /* XXX tp should be locked. */
325 sx_xunlock(&proctree_lock);
327 sx_xlock(&proctree_lock);
329 * The tty could have been revoked
334 SESS_LOCK(p->p_session);
336 SESS_UNLOCK(p->p_session);
337 sx_xunlock(&proctree_lock);
338 VOP_LOCK(ttyvp, LK_EXCLUSIVE, td);
339 VOP_REVOKE(ttyvp, REVOKEALL);
341 sx_xlock(&proctree_lock);
346 SESS_LOCK(p->p_session);
348 SESS_UNLOCK(p->p_session);
352 * s_ttyp is not zero'd; we use this to indicate
353 * that the session once had a controlling terminal.
354 * (for logging and informational purposes)
357 SESS_LOCK(p->p_session);
359 SESS_UNLOCK(p->p_session);
361 fixjobc(p, p->p_pgrp, 0);
362 sx_xunlock(&proctree_lock);
363 (void)acct_process(td);
367 * Drain any pending records on the thread and release the trace
368 * file. It might be better if drain-and-clear were atomic.
372 mtx_lock(&ktrace_mtx);
373 p->p_traceflag = 0; /* don't trace the vrele() */
374 tracevp = p->p_tracevp;
376 tracecred = p->p_tracecred;
377 p->p_tracecred = NULL;
378 mtx_unlock(&ktrace_mtx);
380 if (tracevp != NULL) {
381 locked = VFS_LOCK_GIANT(tracevp->v_mount);
383 VFS_UNLOCK_GIANT(locked);
385 if (tracecred != NULL)
389 * Release reference to text vnode
391 if ((vtmp = p->p_textvp) != NULL) {
393 locked = VFS_LOCK_GIANT(vtmp->v_mount);
395 VFS_UNLOCK_GIANT(locked);
399 * Release our limits structure.
408 * Remove proc from allproc queue and pidhash chain.
409 * Place onto zombproc. Unlink from parent's child list.
411 sx_xlock(&allproc_lock);
412 LIST_REMOVE(p, p_list);
413 LIST_INSERT_HEAD(&zombproc, p, p_list);
414 LIST_REMOVE(p, p_hash);
415 sx_xunlock(&allproc_lock);
418 * Reparent all of our children to init.
420 sx_xlock(&proctree_lock);
421 q = LIST_FIRST(&p->p_children);
422 if (q != NULL) /* only need this if any child is S_ZOMB */
424 for (; q != NULL; q = nq) {
425 nq = LIST_NEXT(q, p_sibling);
427 proc_reparent(q, initproc);
428 q->p_sigparent = SIGCHLD;
430 * Traced processes are killed
431 * since their existence means someone is screwing up.
433 if (q->p_flag & P_TRACED) {
434 q->p_flag &= ~(P_TRACED | P_STOPPED_TRACE);
441 * Save exit status and finalize rusage info except for times,
442 * adding in child rusage info later when our time is locked.
447 p->p_stats->p_ru.ru_nvcsw++;
448 *p->p_ru = p->p_stats->p_ru;
451 * Notify interested parties of our demise.
453 KNOTE_LOCKED(&p->p_klist, NOTE_EXIT);
456 * Just delete all entries in the p_klist. At this point we won't
457 * report any more events, and there are nasty race conditions that
458 * can beat us if we don't.
460 knlist_clear(&p->p_klist, 1);
463 * Notify parent that we're gone. If parent has the PS_NOCLDWAIT
464 * flag set, or if the handler is set to SIG_IGN, notify process
465 * 1 instead (and hope it will handle this situation).
467 PROC_LOCK(p->p_pptr);
468 mtx_lock(&p->p_pptr->p_sigacts->ps_mtx);
469 if (p->p_pptr->p_sigacts->ps_flag & (PS_NOCLDWAIT | PS_CLDSIGIGN)) {
472 mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx);
475 proc_reparent(p, initproc);
476 p->p_sigparent = SIGCHLD;
477 PROC_LOCK(p->p_pptr);
479 * If this was the last child of our parent, notify
480 * parent, so in case he was wait(2)ing, he will
483 if (LIST_EMPTY(&pp->p_children))
486 mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx);
488 if (p->p_pptr == initproc)
489 psignal(p->p_pptr, SIGCHLD);
490 else if (p->p_sigparent != 0) {
491 if (p->p_sigparent == SIGCHLD)
493 else /* LINUX thread */
494 psignal(p->p_pptr, p->p_sigparent);
496 PROC_UNLOCK(p->p_pptr);
500 * Finally, call machine-dependent code to release the remaining
501 * resources including address space.
502 * The address space is released by "vmspace_exitfree(p)" in
507 WITNESS_WARN(WARN_PANIC, &proctree_lock.sx_object,
508 "process (pid %d) exiting", p->p_pid);
511 PROC_LOCK(p->p_pptr);
512 sx_xunlock(&proctree_lock);
515 * We have to wait until after acquiring all locks before
516 * changing p_state. We need to avoid all possible context
517 * switches (including ones from blocking on a mutex) while
518 * marked as a zombie. We also have to set the zombie state
519 * before we release the parent process' proc lock to avoid
520 * a lost wakeup. So, we first call wakeup, then we grab the
521 * sched lock, update the state, and release the parent process'
525 mtx_lock_spin(&sched_lock);
526 p->p_state = PRS_ZOMBIE;
527 PROC_UNLOCK(p->p_pptr);
529 sched_exit(p->p_pptr, td);
532 * Hopefully no one will try to deliver a signal to the process this
535 knlist_destroy(&p->p_klist);
538 * Make sure the scheduler takes this thread out of its tables etc.
539 * This will also release this thread's reference to the ucred.
540 * Other thread parts to release include pcb bits and such.
546 #ifndef _SYS_SYSPROTO_H_
558 abort2(struct thread *td, struct abort2_args *uap)
560 struct proc *p = td->td_proc;
565 error = 0; /* satisfy compiler */
568 * Do it right now so we can log either proper call of abort2(), or
569 * note, that invalid argument was passed. 512 is big enough to
570 * handle 16 arguments' descriptions with additional comments.
572 sb = sbuf_new(NULL, NULL, 512, SBUF_FIXEDLEN);
574 sbuf_printf(sb, "%s(pid %d uid %d) aborted: ",
575 p->p_comm, p->p_pid, td->td_ucred->cr_uid);
577 * Since we can't return from abort2(), send SIGKILL in cases, where
578 * abort2() was called improperly
581 /* Prevent from DoSes from user-space. */
582 if (uap->nargs < 0 || uap->nargs > 16)
584 if (uap->args == NULL)
586 error = copyin(uap->args, uargs, uap->nargs * sizeof(void *));
590 * Limit size of 'reason' string to 128. Will fit even when
591 * maximal number of arguments was chosen to be logged.
593 if (uap->why != NULL) {
594 error = sbuf_copyin(sb, uap->why, 128);
598 sbuf_printf(sb, "(null)");
601 sbuf_printf(sb, "(");
602 for (i = 0;i < uap->nargs; i++)
603 sbuf_printf(sb, "%s%p", i == 0 ? "" : ", ", uargs[i]);
604 sbuf_printf(sb, ")");
607 * Final stage: arguments were proper, string has been
608 * successfully copied from userspace, and copying pointers
609 * from user-space succeed.
613 if (sig == SIGKILL) {
615 sbuf_printf(sb, " (Reason text inaccessible)");
619 log(LOG_INFO, "%s", sbuf_data(sb));
621 exit1(td, W_EXITCODE(0, sig));
628 * The dirty work is handled by kern_wait().
633 owait(struct thread *td, struct owait_args *uap __unused)
637 error = kern_wait(td, WAIT_ANY, &status, 0, NULL);
639 td->td_retval[1] = status;
642 #endif /* COMPAT_43 */
645 * The dirty work is handled by kern_wait().
650 wait4(struct thread *td, struct wait_args *uap)
652 struct rusage ru, *rup;
655 if (uap->rusage != NULL)
659 error = kern_wait(td, uap->pid, &status, uap->options, rup);
660 if (uap->status != NULL && error == 0)
661 error = copyout(&status, uap->status, sizeof(status));
662 if (uap->rusage != NULL && error == 0)
663 error = copyout(&ru, uap->rusage, sizeof(struct rusage));
668 kern_wait(struct thread *td, pid_t pid, int *status, int options,
669 struct rusage *rusage)
671 struct proc *p, *q, *t;
682 if (options &~ (WUNTRACED|WNOHANG|WCONTINUED|WLINUXCLONE))
685 if (q->p_flag & P_STATCHILD) {
687 q->p_flag &= ~P_STATCHILD;
691 sx_xlock(&proctree_lock);
692 LIST_FOREACH(p, &q->p_children, p_sibling) {
694 if (pid != WAIT_ANY &&
695 p->p_pid != pid && p->p_pgid != -pid) {
699 if (p_canwait(td, p)) {
705 * This special case handles a kthread spawned by linux_clone
706 * (see linux_misc.c). The linux_wait4 and linux_waitpid
707 * functions need to be able to distinguish between waiting
708 * on a process and waiting on a thread. It is a thread if
709 * p_sigparent is not SIGCHLD, and the WLINUXCLONE option
710 * signifies we want to wait for threads and not processes.
712 if ((p->p_sigparent != SIGCHLD) ^
713 ((options & WLINUXCLONE) != 0)) {
719 if (p->p_state == PRS_ZOMBIE) {
722 * It is possible that the last thread of this
723 * process is still running on another CPU
724 * in thread_exit() after having dropped the process
725 * lock via PROC_UNLOCK() but before it has completed
726 * cpu_throw(). In that case, the other thread must
727 * still hold sched_lock, so simply by acquiring
728 * sched_lock once we will wait long enough for the
729 * thread to exit in that case.
731 mtx_lock_spin(&sched_lock);
732 mtx_unlock_spin(&sched_lock);
734 td->td_retval[0] = p->p_pid;
736 *status = p->p_xstat; /* convert to int */
739 calcru(p, &rusage->ru_utime, &rusage->ru_stime);
743 sigqueue_take(p->p_ksi);
747 * If we got the child via a ptrace 'attach',
748 * we need to give it back to the old parent.
751 if (p->p_oppid && (t = pfind(p->p_oppid)) != NULL) {
756 tdsignal(t, NULL, SIGCHLD, p->p_ksi);
759 sx_xunlock(&proctree_lock);
764 * Remove other references to this process to ensure
765 * we have an exclusive reference.
767 sx_xlock(&allproc_lock);
768 LIST_REMOVE(p, p_list); /* off zombproc */
769 sx_xunlock(&allproc_lock);
770 LIST_REMOVE(p, p_sibling);
772 sx_xunlock(&proctree_lock);
775 * As a side effect of this lock, we know that
776 * all other writes to this proc are visible now, so
777 * no more locking is needed for p.
780 p->p_xstat = 0; /* XXX: why? */
783 ruadd(&q->p_stats->p_cru, &q->p_crux, p->p_ru,
786 FREE(p->p_ru, M_ZOMBIE);
790 * Decrement the count of procs running with this uid.
792 (void)chgproccnt(p->p_ucred->cr_ruidinfo, -1, 0);
795 * Free credentials, arguments, and sigacts.
799 pargs_drop(p->p_args);
801 sigacts_free(p->p_sigacts);
805 * Do any thread-system specific cleanups.
810 * Give vm and machine-dependent layer a chance
811 * to free anything that cpu_exit couldn't
812 * release while still running in process context.
821 KASSERT(FIRST_THREAD_IN_PROC(p),
822 ("kern_wait: no residual thread!"));
823 uma_zfree(proc_zone, p);
824 sx_xlock(&allproc_lock);
826 sx_xunlock(&allproc_lock);
829 mtx_lock_spin(&sched_lock);
830 if ((p->p_flag & P_STOPPED_SIG) &&
831 (p->p_suspcount == p->p_numthreads) &&
832 (p->p_flag & P_WAITED) == 0 &&
833 (p->p_flag & P_TRACED || options & WUNTRACED)) {
834 mtx_unlock_spin(&sched_lock);
835 p->p_flag |= P_WAITED;
836 sx_xunlock(&proctree_lock);
837 td->td_retval[0] = p->p_pid;
839 *status = W_STOPCODE(p->p_xstat);
843 sigqueue_take(p->p_ksi);
848 mtx_unlock_spin(&sched_lock);
849 if (options & WCONTINUED && (p->p_flag & P_CONTINUED)) {
850 sx_xunlock(&proctree_lock);
851 td->td_retval[0] = p->p_pid;
852 p->p_flag &= ~P_CONTINUED;
856 sigqueue_take(p->p_ksi);
866 sx_xunlock(&proctree_lock);
869 if (options & WNOHANG) {
870 sx_xunlock(&proctree_lock);
871 td->td_retval[0] = 0;
875 sx_xunlock(&proctree_lock);
876 if (q->p_flag & P_STATCHILD) {
877 q->p_flag &= ~P_STATCHILD;
880 error = msleep(q, &q->p_mtx, PWAIT | PCATCH, "wait", 0);
888 * Make process 'parent' the new parent of process 'child'.
889 * Must be called with an exclusive hold of proctree lock.
892 proc_reparent(struct proc *child, struct proc *parent)
895 sx_assert(&proctree_lock, SX_XLOCKED);
896 PROC_LOCK_ASSERT(child, MA_OWNED);
897 if (child->p_pptr == parent)
900 LIST_REMOVE(child, p_sibling);
901 LIST_INSERT_HEAD(&parent->p_children, child, p_sibling);
902 child->p_pptr = parent;