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>
71 #include <sys/ktrace.h>
74 #include <security/audit/audit.h>
75 #include <security/mac/mac_framework.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);
92 * exit -- death of process.
95 sys_exit(struct thread *td, struct sys_exit_args *uap)
98 exit1(td, W_EXITCODE(uap->rval, 0));
103 * Exit: deallocate address space and other resources, change proc state to
104 * zombie, and unlink proc from allproc and parent's lists. Save exit status
105 * and rusage for wait(). Check for child processes and orphan them.
108 exit1(struct thread *td, int rv)
110 struct proc *p, *nq, *q;
115 struct vnode *tracevp;
116 struct ucred *tracecred;
122 * Drop Giant if caller has it. Eventually we should warn about
123 * being called with Giant held.
125 while (mtx_owned(&Giant))
130 printf("init died (signal %d, exit %d)\n",
131 WTERMSIG(rv), WEXITSTATUS(rv));
132 panic("Going nowhere without my init!");
136 * MUST abort all other threads before proceeding past here.
139 if (p->p_flag & P_HADTHREADS) {
142 * First check if some other thread got here before us..
143 * if so, act apropriatly, (exit or suspend);
145 thread_suspend_check(0);
148 * Kill off the other threads. This requires
149 * some co-operation from other parts of the kernel
150 * so it may not be instantaneous. With this state set
151 * any thread entering the kernel from userspace will
152 * thread_exit() in trap(). Any thread attempting to
153 * sleep will return immediately with EINTR or EWOULDBLOCK
154 * which will hopefully force them to back out to userland
155 * freeing resources as they go. Any thread attempting
156 * to return to userland will thread_exit() from userret().
157 * thread_exit() will unsuspend us when the last of the
158 * other threads exits.
159 * If there is already a thread singler after resumption,
160 * calling thread_single will fail; in that case, we just
161 * re-check all suspension request, the thread should
162 * either be suspended there or exit.
164 if (thread_single(SINGLE_EXIT))
168 * All other activity in this process is now stopped.
169 * Threading support has been turned off.
172 KASSERT(p->p_numthreads == 1,
173 ("exit1: proc %p exiting with %d threads", p, p->p_numthreads));
175 * Wakeup anyone in procfs' PIOCWAIT. They should have a hold
176 * on our vmspace, so we should block below until they have
177 * released their reference to us. Note that if they have
178 * requested S_EXIT stops we will block here until they ack
181 _STOPEVENT(p, S_EXIT, rv);
184 * Note that we are exiting and do another wakeup of anyone in
185 * PIOCWAIT in case they aren't listening for S_EXIT stops or
186 * decided to wait again after we told them we are exiting.
188 p->p_flag |= P_WEXIT;
192 * Wait for any processes that have a hold on our vmspace to
193 * release their reference.
195 while (p->p_lock > 0)
196 msleep(&p->p_lock, &p->p_mtx, PWAIT, "exithold", 0);
199 /* Drain the limit callout while we don't have the proc locked */
200 callout_drain(&p->p_limco);
204 * The Sun BSM exit token contains two components: an exit status as
205 * passed to exit(), and a return value to indicate what sort of exit
206 * it was. The exit status is WEXITSTATUS(rv), but it's not clear
207 * what the return value is.
209 AUDIT_ARG(exit, WEXITSTATUS(rv), 0);
210 AUDIT_SYSCALL_EXIT(0, td);
213 /* Are we a task leader? */
214 if (p == p->p_leader) {
215 mtx_lock(&ppeers_lock);
223 while (p->p_peers != NULL)
224 msleep(p, &ppeers_lock, PWAIT, "exit1", 0);
225 mtx_unlock(&ppeers_lock);
229 * Check if any loadable modules need anything done at process exit.
230 * E.g. SYSV IPC stuff
231 * XXX what if one of these generates an error?
233 EVENTHANDLER_INVOKE(process_exit, p);
236 * If parent is waiting for us to exit or exec,
237 * P_PPWAIT is set; we will wakeup the parent below.
241 p->p_flag &= ~(P_TRACED | P_PPWAIT);
244 * Stop the real interval timer. If the handler is currently
245 * executing, prevent it from rearming itself and let it finish.
247 if (timevalisset(&p->p_realtimer.it_value) &&
248 callout_stop(&p->p_itcallout) == 0) {
249 timevalclear(&p->p_realtimer.it_interval);
250 msleep(&p->p_itcallout, &p->p_mtx, PWAIT, "ritwait", 0);
251 KASSERT(!timevalisset(&p->p_realtimer.it_value),
252 ("realtime timer is still armed"));
257 * Reset any sigio structures pointing to us as a result of
258 * F_SETOWN with our pid.
260 funsetownlst(&p->p_sigiolst);
263 * If this process has an nlminfo data area (for lockd), release it
265 if (nlminfo_release_p != NULL && p->p_nlminfo != NULL)
266 (*nlminfo_release_p)(p);
269 * Close open files and release open-file table.
275 * If this thread tickled GEOM, we need to wait for the giggling to
276 * stop before we return to userland
278 if (td->td_pflags & TDP_GEOM)
282 * Remove ourself from our leader's peer list and wake our leader.
284 mtx_lock(&ppeers_lock);
285 if (p->p_leader->p_peers) {
287 while (q->p_peers != p)
289 q->p_peers = p->p_peers;
292 mtx_unlock(&ppeers_lock);
296 mtx_lock(&Giant); /* XXX TTY */
297 sx_xlock(&proctree_lock);
298 if (SESS_LEADER(p)) {
304 * Controlling process.
305 * Signal foreground pgrp,
306 * drain controlling terminal
307 * and revoke access to controlling terminal.
309 if (sp->s_ttyp && (sp->s_ttyp->t_session == sp)) {
311 if (sp->s_ttyp->t_pgrp) {
312 PGRP_LOCK(sp->s_ttyp->t_pgrp);
313 pgsignal(sp->s_ttyp->t_pgrp, SIGHUP, 1);
314 PGRP_UNLOCK(sp->s_ttyp->t_pgrp);
316 /* XXX tp should be locked. */
317 sx_xunlock(&proctree_lock);
319 sx_xlock(&proctree_lock);
321 * The tty could have been revoked
326 SESS_LOCK(p->p_session);
328 SESS_UNLOCK(p->p_session);
329 sx_xunlock(&proctree_lock);
330 VOP_LOCK(ttyvp, LK_EXCLUSIVE, td);
331 VOP_REVOKE(ttyvp, REVOKEALL);
333 sx_xlock(&proctree_lock);
338 SESS_LOCK(p->p_session);
340 SESS_UNLOCK(p->p_session);
344 * s_ttyp is not zero'd; we use this to indicate
345 * that the session once had a controlling terminal.
346 * (for logging and informational purposes)
349 SESS_LOCK(p->p_session);
351 SESS_UNLOCK(p->p_session);
353 fixjobc(p, p->p_pgrp, 0);
354 sx_xunlock(&proctree_lock);
355 (void)acct_process(td);
359 * Disable tracing, then drain any pending records and release
362 if (p->p_traceflag != 0) {
364 mtx_lock(&ktrace_mtx);
366 mtx_unlock(&ktrace_mtx);
370 mtx_lock(&ktrace_mtx);
371 tracevp = p->p_tracevp;
373 tracecred = p->p_tracecred;
374 p->p_tracecred = NULL;
375 mtx_unlock(&ktrace_mtx);
377 if (tracevp != NULL) {
378 locked = VFS_LOCK_GIANT(tracevp->v_mount);
380 VFS_UNLOCK_GIANT(locked);
382 if (tracecred != NULL)
387 * Release reference to text vnode
389 if ((vtmp = p->p_textvp) != NULL) {
391 locked = VFS_LOCK_GIANT(vtmp->v_mount);
393 VFS_UNLOCK_GIANT(locked);
397 * Release our limits structure.
406 * Remove proc from allproc queue and pidhash chain.
407 * Place onto zombproc. Unlink from parent's child list.
409 sx_xlock(&allproc_lock);
410 LIST_REMOVE(p, p_list);
411 LIST_INSERT_HEAD(&zombproc, p, p_list);
412 LIST_REMOVE(p, p_hash);
413 sx_xunlock(&allproc_lock);
416 * Call machine-dependent code to release any
417 * machine-dependent resources other than the address space.
418 * The address space is released by "vmspace_exitfree(p)" in
423 WITNESS_WARN(WARN_PANIC, NULL, "process (pid %d) exiting", p->p_pid);
426 * Reparent all of our children to init.
428 sx_xlock(&proctree_lock);
429 q = LIST_FIRST(&p->p_children);
430 if (q != NULL) /* only need this if any child is S_ZOMB */
432 for (; q != NULL; q = nq) {
433 nq = LIST_NEXT(q, p_sibling);
435 proc_reparent(q, initproc);
436 q->p_sigparent = SIGCHLD;
438 * Traced processes are killed
439 * since their existence means someone is screwing up.
441 if (q->p_flag & P_TRACED) {
442 q->p_flag &= ~(P_TRACED | P_STOPPED_TRACE);
448 /* Save exit status. */
453 * Notify interested parties of our demise.
455 KNOTE_LOCKED(&p->p_klist, NOTE_EXIT);
458 * Just delete all entries in the p_klist. At this point we won't
459 * report any more events, and there are nasty race conditions that
460 * can beat us if we don't.
462 knlist_clear(&p->p_klist, 1);
465 * Notify parent that we're gone. If parent has the PS_NOCLDWAIT
466 * flag set, or if the handler is set to SIG_IGN, notify process
467 * 1 instead (and hope it will handle this situation).
469 PROC_LOCK(p->p_pptr);
470 mtx_lock(&p->p_pptr->p_sigacts->ps_mtx);
471 if (p->p_pptr->p_sigacts->ps_flag & (PS_NOCLDWAIT | PS_CLDSIGIGN)) {
474 mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx);
477 proc_reparent(p, initproc);
478 p->p_sigparent = SIGCHLD;
479 PROC_LOCK(p->p_pptr);
481 * If this was the last child of our parent, notify
482 * parent, so in case he was wait(2)ing, he will
485 if (LIST_EMPTY(&pp->p_children))
488 mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx);
490 if (p->p_pptr == initproc)
491 psignal(p->p_pptr, SIGCHLD);
492 else if (p->p_sigparent != 0) {
493 if (p->p_sigparent == SIGCHLD)
495 else /* LINUX thread */
496 psignal(p->p_pptr, p->p_sigparent);
498 sx_xunlock(&proctree_lock);
501 * The state PRS_ZOMBIE prevents other proesses from sending
502 * signal to the process, to avoid memory leak, we free memory
503 * for signal queue at the time when the state is set.
505 sigqueue_flush(&p->p_sigqueue);
506 sigqueue_flush(&td->td_sigqueue);
509 * We have to wait until after acquiring all locks before
510 * changing p_state. We need to avoid all possible context
511 * switches (including ones from blocking on a mutex) while
512 * marked as a zombie. We also have to set the zombie state
513 * before we release the parent process' proc lock to avoid
514 * a lost wakeup. So, we first call wakeup, then we grab the
515 * sched lock, update the state, and release the parent process'
519 PROC_SLOCK(p->p_pptr);
520 sched_exit(p->p_pptr, td);
521 PROC_SUNLOCK(p->p_pptr);
523 p->p_state = PRS_ZOMBIE;
524 PROC_UNLOCK(p->p_pptr);
527 * Hopefully no one will try to deliver a signal to the process this
530 knlist_destroy(&p->p_klist);
533 * Save our children's rusage information in our exit rusage.
535 ruadd(&p->p_ru, &p->p_rux, &p->p_stats->p_cru, &p->p_crux);
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_
555 abort2(struct thread *td, struct abort2_args *uap)
557 struct proc *p = td->td_proc;
562 error = 0; /* satisfy compiler */
565 * Do it right now so we can log either proper call of abort2(), or
566 * note, that invalid argument was passed. 512 is big enough to
567 * handle 16 arguments' descriptions with additional comments.
569 sb = sbuf_new(NULL, NULL, 512, SBUF_FIXEDLEN);
571 sbuf_printf(sb, "%s(pid %d uid %d) aborted: ",
572 p->p_comm, p->p_pid, td->td_ucred->cr_uid);
574 * Since we can't return from abort2(), send SIGKILL in cases, where
575 * abort2() was called improperly
578 /* Prevent from DoSes from user-space. */
579 if (uap->nargs < 0 || uap->nargs > 16)
581 if (uap->args == NULL)
583 error = copyin(uap->args, uargs, uap->nargs * sizeof(void *));
587 * Limit size of 'reason' string to 128. Will fit even when
588 * maximal number of arguments was chosen to be logged.
590 if (uap->why != NULL) {
591 error = sbuf_copyin(sb, uap->why, 128);
595 sbuf_printf(sb, "(null)");
598 sbuf_printf(sb, "(");
599 for (i = 0;i < uap->nargs; i++)
600 sbuf_printf(sb, "%s%p", i == 0 ? "" : ", ", uargs[i]);
601 sbuf_printf(sb, ")");
604 * Final stage: arguments were proper, string has been
605 * successfully copied from userspace, and copying pointers
606 * from user-space succeed.
610 if (sig == SIGKILL) {
612 sbuf_printf(sb, " (Reason text inaccessible)");
616 log(LOG_INFO, "%s", sbuf_data(sb));
618 exit1(td, W_EXITCODE(0, sig));
625 * The dirty work is handled by kern_wait().
628 owait(struct thread *td, struct owait_args *uap __unused)
632 error = kern_wait(td, WAIT_ANY, &status, 0, NULL);
634 td->td_retval[1] = status;
637 #endif /* COMPAT_43 */
640 * The dirty work is handled by kern_wait().
643 wait4(struct thread *td, struct wait_args *uap)
645 struct rusage ru, *rup;
648 if (uap->rusage != NULL)
652 error = kern_wait(td, uap->pid, &status, uap->options, rup);
653 if (uap->status != NULL && error == 0)
654 error = copyout(&status, uap->status, sizeof(status));
655 if (uap->rusage != NULL && error == 0)
656 error = copyout(&ru, uap->rusage, sizeof(struct rusage));
661 kern_wait(struct thread *td, pid_t pid, int *status, int options,
662 struct rusage *rusage)
664 struct proc *p, *q, *t;
675 if (options &~ (WUNTRACED|WNOHANG|WCONTINUED|WLINUXCLONE))
678 if (q->p_flag & P_STATCHILD) {
680 q->p_flag &= ~P_STATCHILD;
684 sx_xlock(&proctree_lock);
685 LIST_FOREACH(p, &q->p_children, p_sibling) {
687 if (pid != WAIT_ANY &&
688 p->p_pid != pid && p->p_pgid != -pid) {
692 if (p_canwait(td, p)) {
698 * This special case handles a kthread spawned by linux_clone
699 * (see linux_misc.c). The linux_wait4 and linux_waitpid
700 * functions need to be able to distinguish between waiting
701 * on a process and waiting on a thread. It is a thread if
702 * p_sigparent is not SIGCHLD, and the WLINUXCLONE option
703 * signifies we want to wait for threads and not processes.
705 if ((p->p_sigparent != SIGCHLD) ^
706 ((options & WLINUXCLONE) != 0)) {
713 if (p->p_state == PRS_ZOMBIE) {
716 calcru(p, &rusage->ru_utime, &rusage->ru_stime);
719 td->td_retval[0] = p->p_pid;
721 *status = p->p_xstat; /* convert to int */
723 sigqueue_take(p->p_ksi);
727 * If we got the child via a ptrace 'attach',
728 * we need to give it back to the old parent.
731 if (p->p_oppid && (t = pfind(p->p_oppid)) != NULL) {
736 tdsignal(t, NULL, SIGCHLD, p->p_ksi);
739 sx_xunlock(&proctree_lock);
744 * Remove other references to this process to ensure
745 * we have an 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
756 * all other writes to this proc are visible now, so
757 * no more locking is needed for p.
760 p->p_xstat = 0; /* XXX: why? */
763 ruadd(&q->p_stats->p_cru, &q->p_crux, &p->p_ru,
768 * Decrement the count of procs running with this uid.
770 (void)chgproccnt(p->p_ucred->cr_ruidinfo, -1, 0);
773 * Free credentials, arguments, and sigacts.
777 pargs_drop(p->p_args);
779 sigacts_free(p->p_sigacts);
783 * Do any thread-system specific cleanups.
788 * Give vm and machine-dependent layer a chance
789 * to free anything that cpu_exit couldn't
790 * release while still running in process context.
796 KASSERT(FIRST_THREAD_IN_PROC(p),
797 ("kern_wait: no residual thread!"));
798 uma_zfree(proc_zone, p);
799 sx_xlock(&allproc_lock);
801 sx_xunlock(&allproc_lock);
804 if ((p->p_flag & P_STOPPED_SIG) &&
805 (p->p_suspcount == p->p_numthreads) &&
806 (p->p_flag & P_WAITED) == 0 &&
807 (p->p_flag & P_TRACED || options & WUNTRACED)) {
809 p->p_flag |= P_WAITED;
810 sx_xunlock(&proctree_lock);
811 td->td_retval[0] = p->p_pid;
813 *status = W_STOPCODE(p->p_xstat);
816 sigqueue_take(p->p_ksi);
823 if (options & WCONTINUED && (p->p_flag & P_CONTINUED)) {
824 sx_xunlock(&proctree_lock);
825 td->td_retval[0] = p->p_pid;
826 p->p_flag &= ~P_CONTINUED;
829 sigqueue_take(p->p_ksi);
840 sx_xunlock(&proctree_lock);
843 if (options & WNOHANG) {
844 sx_xunlock(&proctree_lock);
845 td->td_retval[0] = 0;
849 sx_xunlock(&proctree_lock);
850 if (q->p_flag & P_STATCHILD) {
851 q->p_flag &= ~P_STATCHILD;
854 error = msleep(q, &q->p_mtx, PWAIT | PCATCH, "wait", 0);
862 * Make process 'parent' the new parent of process 'child'.
863 * Must be called with an exclusive hold of proctree lock.
866 proc_reparent(struct proc *child, struct proc *parent)
869 sx_assert(&proctree_lock, SX_XLOCKED);
870 PROC_LOCK_ASSERT(child, MA_OWNED);
871 if (child->p_pptr == parent)
874 PROC_LOCK(child->p_pptr);
875 sigqueue_take(child->p_ksi);
876 PROC_UNLOCK(child->p_pptr);
877 LIST_REMOVE(child, p_sibling);
878 LIST_INSERT_HEAD(&parent->p_children, child, p_sibling);
879 child->p_pptr = parent;