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 uint64_t new_switchtime;
114 struct proc *p, *nq, *q;
120 struct vnode *tracevp;
121 struct ucred *tracecred;
127 * Drop Giant if caller has it. Eventually we should warn about
128 * being called with Giant held.
130 while (mtx_owned(&Giant))
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 if (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.
178 p->p_flag |= P_WEXIT;
180 PROC_LOCK(p->p_pptr);
181 sigqueue_take(p->p_ksi);
182 PROC_UNLOCK(p->p_pptr);
188 * The Sun BSM exit token contains two components: an exit status as
189 * passed to exit(), and a return value to indicate what sort of exit
190 * it was. The exit status is WEXITSTATUS(rv), but it's not clear
191 * what the return value is.
193 AUDIT_ARG(exit, WEXITSTATUS(rv), 0);
194 AUDIT_SYSCALL_EXIT(0, td);
197 /* Are we a task leader? */
198 if (p == p->p_leader) {
199 mtx_lock(&ppeers_lock);
207 while (p->p_peers != NULL)
208 msleep(p, &ppeers_lock, PWAIT, "exit1", 0);
209 mtx_unlock(&ppeers_lock);
213 _STOPEVENT(p, S_EXIT, rv);
214 wakeup(&p->p_stype); /* Wakeup anyone in procfs' PIOCWAIT */
218 * Check if any loadable modules need anything done at process exit.
219 * E.g. SYSV IPC stuff
220 * XXX what if one of these generates an error?
222 EVENTHANDLER_INVOKE(process_exit, p);
224 MALLOC(p->p_ru, struct rusage *, sizeof(struct rusage),
227 * If parent is waiting for us to exit or exec,
228 * P_PPWAIT is set; we will wakeup the parent below.
232 p->p_flag &= ~(P_TRACED | P_PPWAIT);
235 * Stop the real interval timer. If the handler is currently
236 * executing, prevent it from rearming itself and let it finish.
238 if (timevalisset(&p->p_realtimer.it_value) &&
239 callout_stop(&p->p_itcallout) == 0) {
240 timevalclear(&p->p_realtimer.it_interval);
241 msleep(&p->p_itcallout, &p->p_mtx, PWAIT, "ritwait", 0);
242 KASSERT(!timevalisset(&p->p_realtimer.it_value),
243 ("realtime timer is still armed"));
245 sigqueue_flush(&p->p_sigqueue);
246 sigqueue_flush(&td->td_sigqueue);
250 * Reset any sigio structures pointing to us as a result of
251 * F_SETOWN with our pid.
253 mtx_lock(&Giant); /* XXX: not sure if needed */
254 funsetownlst(&p->p_sigiolst);
258 * If this process has an nlminfo data area (for lockd), release it
260 if (nlminfo_release_p != NULL && p->p_nlminfo != NULL)
261 (*nlminfo_release_p)(p);
264 * Close open files and release open-file table.
270 * If this thread tickled GEOM, we need to wait for the giggling to
271 * stop before we return to userland
273 if (td->td_pflags & TDP_GEOM)
277 * Remove ourself from our leader's peer list and wake our leader.
279 mtx_lock(&ppeers_lock);
280 if (p->p_leader->p_peers) {
282 while (q->p_peers != p)
284 q->p_peers = p->p_peers;
287 mtx_unlock(&ppeers_lock);
289 /* The next two chunks should probably be moved to vmspace_exit. */
292 * Release user portion of address space.
293 * This releases references to vnodes,
294 * which could cause I/O if the file has been unlinked.
295 * Need to do this early enough that we can still sleep.
296 * Can't free the entire vmspace as the kernel stack
297 * may be mapped within that space also.
299 * Processes sharing the same vmspace may exit in one order, and
300 * get cleaned up by vmspace_exit() in a different order. The
301 * last exiting process to reach this point releases as much of
302 * the environment as it can, and the last process cleaned up
303 * by vmspace_exit() (which decrements exitingcnt) cleans up the
306 atomic_add_int(&vm->vm_exitingcnt, 1);
308 refcnt = vm->vm_refcnt;
309 while (!atomic_cmpset_int(&vm->vm_refcnt, refcnt, refcnt - 1));
312 pmap_remove_pages(vmspace_pmap(vm), vm_map_min(&vm->vm_map),
313 vm_map_max(&vm->vm_map));
314 (void) vm_map_remove(&vm->vm_map, vm_map_min(&vm->vm_map),
315 vm_map_max(&vm->vm_map));
318 sx_xlock(&proctree_lock);
319 if (SESS_LEADER(p)) {
324 locked = VFS_LOCK_GIANT(sp->s_ttyvp->v_mount);
326 * Controlling process.
327 * Signal foreground pgrp,
328 * drain controlling terminal
329 * and revoke access to controlling terminal.
331 if (sp->s_ttyp && (sp->s_ttyp->t_session == sp)) {
333 if (sp->s_ttyp->t_pgrp) {
334 PGRP_LOCK(sp->s_ttyp->t_pgrp);
335 pgsignal(sp->s_ttyp->t_pgrp, SIGHUP, 1);
336 PGRP_UNLOCK(sp->s_ttyp->t_pgrp);
338 /* XXX tp should be locked. */
339 sx_xunlock(&proctree_lock);
341 sx_xlock(&proctree_lock);
343 * The tty could have been revoked
348 SESS_LOCK(p->p_session);
350 SESS_UNLOCK(p->p_session);
351 sx_xunlock(&proctree_lock);
352 VOP_LOCK(ttyvp, LK_EXCLUSIVE, td);
353 VOP_REVOKE(ttyvp, REVOKEALL);
355 sx_xlock(&proctree_lock);
360 SESS_LOCK(p->p_session);
362 SESS_UNLOCK(p->p_session);
366 * s_ttyp is not zero'd; we use this to indicate
367 * that the session once had a controlling terminal.
368 * (for logging and informational purposes)
370 VFS_UNLOCK_GIANT(locked);
372 SESS_LOCK(p->p_session);
374 SESS_UNLOCK(p->p_session);
376 fixjobc(p, p->p_pgrp, 0);
377 sx_xunlock(&proctree_lock);
378 (void)acct_process(td);
381 * Drain any pending records on the thread and release the trace
382 * file. It might be better if drain-and-clear were atomic.
386 mtx_lock(&ktrace_mtx);
387 p->p_traceflag = 0; /* don't trace the vrele() */
388 tracevp = p->p_tracevp;
390 tracecred = p->p_tracecred;
391 p->p_tracecred = NULL;
392 mtx_unlock(&ktrace_mtx);
394 if (tracevp != NULL) {
395 locked = VFS_LOCK_GIANT(tracevp->v_mount);
397 VFS_UNLOCK_GIANT(locked);
399 if (tracecred != NULL)
403 * Release reference to text vnode
405 if ((vtmp = p->p_textvp) != NULL) {
407 locked = VFS_LOCK_GIANT(vtmp->v_mount);
409 VFS_UNLOCK_GIANT(locked);
413 * Release our limits structure.
422 * Remove proc from allproc queue and pidhash chain.
423 * Place onto zombproc. Unlink from parent's child list.
425 sx_xlock(&allproc_lock);
426 LIST_REMOVE(p, p_list);
427 LIST_INSERT_HEAD(&zombproc, p, p_list);
428 LIST_REMOVE(p, p_hash);
429 sx_xunlock(&allproc_lock);
432 * Reparent all of our children to init.
434 sx_xlock(&proctree_lock);
435 q = LIST_FIRST(&p->p_children);
436 if (q != NULL) /* only need this if any child is S_ZOMB */
438 for (; q != NULL; q = nq) {
439 nq = LIST_NEXT(q, p_sibling);
441 proc_reparent(q, initproc);
442 q->p_sigparent = SIGCHLD;
444 * Traced processes are killed
445 * since their existence means someone is screwing up.
447 if (q->p_flag & P_TRACED) {
448 q->p_flag &= ~(P_TRACED | P_STOPPED_TRACE);
455 * Save exit status and finalize rusage info except for times,
456 * adding in child rusage info later when our time is locked.
461 p->p_stats->p_ru.ru_nvcsw++;
462 *p->p_ru = p->p_stats->p_ru;
465 * Notify interested parties of our demise.
467 KNOTE_LOCKED(&p->p_klist, NOTE_EXIT);
470 * Just delete all entries in the p_klist. At this point we won't
471 * report any more events, and there are nasty race conditions that
472 * can beat us if we don't.
474 knlist_clear(&p->p_klist, 1);
477 * Notify parent that we're gone. If parent has the PS_NOCLDWAIT
478 * flag set, or if the handler is set to SIG_IGN, notify process
479 * 1 instead (and hope it will handle this situation).
481 PROC_LOCK(p->p_pptr);
482 mtx_lock(&p->p_pptr->p_sigacts->ps_mtx);
483 if (p->p_pptr->p_sigacts->ps_flag & (PS_NOCLDWAIT | PS_CLDSIGIGN)) {
486 mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx);
489 proc_reparent(p, initproc);
490 p->p_sigparent = SIGCHLD;
491 PROC_LOCK(p->p_pptr);
493 * If this was the last child of our parent, notify
494 * parent, so in case he was wait(2)ing, he will
497 if (LIST_EMPTY(&pp->p_children))
500 mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx);
502 if (p->p_pptr == initproc)
503 psignal(p->p_pptr, SIGCHLD);
504 else if (p->p_sigparent != 0) {
505 if (p->p_sigparent == SIGCHLD)
507 else /* LINUX thread */
508 psignal(p->p_pptr, p->p_sigparent);
510 PROC_UNLOCK(p->p_pptr);
514 * Finally, call machine-dependent code to release the remaining
515 * resources including address space.
516 * The address space is released by "vmspace_exitfree(p)" in
521 WITNESS_WARN(WARN_PANIC, &proctree_lock.sx_object,
522 "process (pid %d) exiting", p->p_pid);
525 PROC_LOCK(p->p_pptr);
526 sx_xunlock(&proctree_lock);
529 * We have to wait until after acquiring all locks before
530 * changing p_state. We need to avoid all possible context
531 * switches (including ones from blocking on a mutex) while
532 * marked as a zombie. We also have to set the zombie state
533 * before we release the parent process' proc lock to avoid
534 * a lost wakeup. So, we first call wakeup, then we grab the
535 * sched lock, update the state, and release the parent process'
539 mtx_lock_spin(&sched_lock);
540 p->p_state = PRS_ZOMBIE;
541 PROC_UNLOCK(p->p_pptr);
543 ruadd(p->p_ru, &p->p_rux, &p->p_stats->p_cru, &p->p_crux);
545 /* Do the same timestamp bookkeeping that mi_switch() would do. */
546 new_switchtime = cpu_ticks();
547 p->p_rux.rux_runtime += (new_switchtime - PCPU_GET(switchtime));
548 p->p_rux.rux_uticks += td->td_uticks;
549 p->p_rux.rux_sticks += td->td_sticks;
550 p->p_rux.rux_iticks += td->td_iticks;
551 PCPU_SET(switchtime, new_switchtime);
552 PCPU_SET(switchticks, ticks);
555 sched_exit(p->p_pptr, td);
558 * Hopefully no one will try to deliver a signal to the process this
561 knlist_destroy(&p->p_klist);
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_
584 abort2(struct thread *td, struct abort2_args *uap)
586 struct proc *p = td->td_proc;
591 error = 0; /* satisfy compiler */
594 * Do it right now so we can log either proper call of abort2(), or
595 * note, that invalid argument was passed. 512 is big enough to
596 * handle 16 arguments' descriptions with additional comments.
598 sb = sbuf_new(NULL, NULL, 512, SBUF_FIXEDLEN);
600 sbuf_printf(sb, "%s(pid %d uid %d) aborted: ",
601 p->p_comm, p->p_pid, td->td_ucred->cr_uid);
603 * Since we can't return from abort2(), send SIGKILL in cases, where
604 * abort2() was called improperly
607 /* Prevent from DoSes from user-space. */
608 if (uap->nargs < 0 || uap->nargs > 16)
610 if (uap->args == NULL)
612 error = copyin(uap->args, uargs, uap->nargs * sizeof(void *));
616 * Limit size of 'reason' string to 128. Will fit even when
617 * maximal number of arguments was chosen to be logged.
619 if (uap->why != NULL) {
620 error = sbuf_copyin(sb, uap->why, 128);
624 sbuf_printf(sb, "(null)");
627 sbuf_printf(sb, "(");
628 for (i = 0;i < uap->nargs; i++)
629 sbuf_printf(sb, "%s%p", i == 0 ? "" : ", ", uargs[i]);
630 sbuf_printf(sb, ")");
633 * Final stage: arguments were proper, string has been
634 * successfully copied from userspace, and copying pointers
635 * from user-space succeed.
639 if (sig == SIGKILL) {
641 sbuf_printf(sb, " (Reason text inaccessible)");
645 log(LOG_INFO, "%s", sbuf_data(sb));
647 exit1(td, W_EXITCODE(0, sig));
654 * The dirty work is handled by kern_wait().
659 owait(struct thread *td, struct owait_args *uap __unused)
663 error = kern_wait(td, WAIT_ANY, &status, 0, NULL);
665 td->td_retval[1] = status;
668 #endif /* COMPAT_43 */
671 * The dirty work is handled by kern_wait().
676 wait4(struct thread *td, struct wait_args *uap)
678 struct rusage ru, *rup;
681 if (uap->rusage != NULL)
685 error = kern_wait(td, uap->pid, &status, uap->options, rup);
686 if (uap->status != NULL && error == 0)
687 error = copyout(&status, uap->status, sizeof(status));
688 if (uap->rusage != NULL && error == 0)
689 error = copyout(&ru, uap->rusage, sizeof(struct rusage));
694 kern_wait(struct thread *td, pid_t pid, int *status, int options,
695 struct rusage *rusage)
697 struct proc *p, *q, *t;
708 if (options &~ (WUNTRACED|WNOHANG|WCONTINUED|WLINUXCLONE))
711 if (q->p_flag & P_STATCHILD) {
713 q->p_flag &= ~P_STATCHILD;
717 sx_xlock(&proctree_lock);
718 LIST_FOREACH(p, &q->p_children, p_sibling) {
720 if (pid != WAIT_ANY &&
721 p->p_pid != pid && p->p_pgid != -pid) {
725 if (p_canwait(td, p)) {
731 * This special case handles a kthread spawned by linux_clone
732 * (see linux_misc.c). The linux_wait4 and linux_waitpid
733 * functions need to be able to distinguish between waiting
734 * on a process and waiting on a thread. It is a thread if
735 * p_sigparent is not SIGCHLD, and the WLINUXCLONE option
736 * signifies we want to wait for threads and not processes.
738 if ((p->p_sigparent != SIGCHLD) ^
739 ((options & WLINUXCLONE) != 0)) {
745 if (p->p_state == PRS_ZOMBIE) {
748 * It is possible that the last thread of this
749 * process is still running on another CPU
750 * in thread_exit() after having dropped the process
751 * lock via PROC_UNLOCK() but before it has completed
752 * cpu_throw(). In that case, the other thread must
753 * still hold sched_lock, so simply by acquiring
754 * sched_lock once we will wait long enough for the
755 * thread to exit in that case.
757 mtx_lock_spin(&sched_lock);
758 mtx_unlock_spin(&sched_lock);
760 td->td_retval[0] = p->p_pid;
762 *status = p->p_xstat; /* convert to int */
765 calcru(p, &rusage->ru_utime, &rusage->ru_stime);
769 sigqueue_take(p->p_ksi);
773 * If we got the child via a ptrace 'attach',
774 * we need to give it back to the old parent.
777 if (p->p_oppid && (t = pfind(p->p_oppid)) != NULL) {
782 tdsignal(t, NULL, SIGCHLD, p->p_ksi);
785 sx_xunlock(&proctree_lock);
790 * Remove other references to this process to ensure
791 * we have an exclusive reference.
793 sx_xlock(&allproc_lock);
794 LIST_REMOVE(p, p_list); /* off zombproc */
795 sx_xunlock(&allproc_lock);
796 LIST_REMOVE(p, p_sibling);
798 sx_xunlock(&proctree_lock);
801 * As a side effect of this lock, we know that
802 * all other writes to this proc are visible now, so
803 * no more locking is needed for p.
806 p->p_xstat = 0; /* XXX: why? */
809 ruadd(&q->p_stats->p_cru, &q->p_crux, p->p_ru,
812 FREE(p->p_ru, M_ZOMBIE);
816 * Decrement the count of procs running with this uid.
818 (void)chgproccnt(p->p_ucred->cr_ruidinfo, -1, 0);
821 * Free credentials, arguments, and sigacts.
825 pargs_drop(p->p_args);
827 sigacts_free(p->p_sigacts);
831 * Do any thread-system specific cleanups.
836 * Give vm and machine-dependent layer a chance
837 * to free anything that cpu_exit couldn't
838 * release while still running in process context.
847 KASSERT(FIRST_THREAD_IN_PROC(p),
848 ("kern_wait: no residual thread!"));
849 uma_zfree(proc_zone, p);
850 sx_xlock(&allproc_lock);
852 sx_xunlock(&allproc_lock);
855 mtx_lock_spin(&sched_lock);
856 if ((p->p_flag & P_STOPPED_SIG) &&
857 (p->p_suspcount == p->p_numthreads) &&
858 (p->p_flag & P_WAITED) == 0 &&
859 (p->p_flag & P_TRACED || options & WUNTRACED)) {
860 mtx_unlock_spin(&sched_lock);
861 p->p_flag |= P_WAITED;
862 sx_xunlock(&proctree_lock);
863 td->td_retval[0] = p->p_pid;
865 *status = W_STOPCODE(p->p_xstat);
869 sigqueue_take(p->p_ksi);
874 mtx_unlock_spin(&sched_lock);
875 if (options & WCONTINUED && (p->p_flag & P_CONTINUED)) {
876 sx_xunlock(&proctree_lock);
877 td->td_retval[0] = p->p_pid;
878 p->p_flag &= ~P_CONTINUED;
882 sigqueue_take(p->p_ksi);
892 sx_xunlock(&proctree_lock);
895 if (options & WNOHANG) {
896 sx_xunlock(&proctree_lock);
897 td->td_retval[0] = 0;
901 sx_xunlock(&proctree_lock);
902 if (q->p_flag & P_STATCHILD) {
903 q->p_flag &= ~P_STATCHILD;
906 error = msleep(q, &q->p_mtx, PWAIT | PCATCH, "wait", 0);
914 * Make process 'parent' the new parent of process 'child'.
915 * Must be called with an exclusive hold of proctree lock.
918 proc_reparent(struct proc *child, struct proc *parent)
921 sx_assert(&proctree_lock, SX_XLOCKED);
922 PROC_LOCK_ASSERT(child, MA_OWNED);
923 if (child->p_pptr == parent)
926 LIST_REMOVE(child, p_sibling);
927 LIST_INSERT_HEAD(&parent->p_children, child, p_sibling);
928 child->p_pptr = parent;