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>
76 #include <vm/vm_extern.h>
77 #include <vm/vm_param.h>
79 #include <vm/vm_map.h>
80 #include <vm/vm_page.h>
83 /* Required to be non-static for SysVR4 emulator */
84 MALLOC_DEFINE(M_ZOMBIE, "zombie", "zombie proc status");
86 /* Hook for NFS teardown procedure. */
87 void (*nlminfo_release_p)(struct proc *p);
96 sys_exit(struct thread *td, struct sys_exit_args *uap)
99 exit1(td, W_EXITCODE(uap->rval, 0));
104 * Exit: deallocate address space and other resources, change proc state
105 * to zombie, and unlink proc from allproc and parent's lists. Save exit
106 * status and rusage for wait(). Check for child processes and orphan them.
109 exit1(struct thread *td, int rv)
111 struct bintime new_switchtime;
112 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.
176 p->p_flag |= P_WEXIT;
178 PROC_LOCK(p->p_pptr);
179 sigqueue_take(p->p_ksi);
180 PROC_UNLOCK(p->p_pptr);
184 /* Are we a task leader? */
185 if (p == p->p_leader) {
186 mtx_lock(&ppeers_lock);
194 while (p->p_peers != NULL)
195 msleep(p, &ppeers_lock, PWAIT, "exit1", 0);
196 mtx_unlock(&ppeers_lock);
200 _STOPEVENT(p, S_EXIT, rv);
201 wakeup(&p->p_stype); /* Wakeup anyone in procfs' PIOCWAIT */
205 * Check if any loadable modules need anything done at process exit.
206 * E.g. SYSV IPC stuff
207 * XXX what if one of these generates an error?
209 EVENTHANDLER_INVOKE(process_exit, p);
211 MALLOC(p->p_ru, struct rusage *, sizeof(struct rusage),
214 * If parent is waiting for us to exit or exec,
215 * P_PPWAIT is set; we will wakeup the parent below.
219 p->p_flag &= ~(P_TRACED | P_PPWAIT);
222 * Stop the real interval timer. If the handler is currently
223 * executing, prevent it from rearming itself and let it finish.
225 if (timevalisset(&p->p_realtimer.it_value) &&
226 callout_stop(&p->p_itcallout) == 0) {
227 timevalclear(&p->p_realtimer.it_interval);
228 msleep(&p->p_itcallout, &p->p_mtx, PWAIT, "ritwait", 0);
229 KASSERT(!timevalisset(&p->p_realtimer.it_value),
230 ("realtime timer is still armed"));
232 sigqueue_flush(&p->p_sigqueue);
233 sigqueue_flush(&td->td_sigqueue);
237 * Reset any sigio structures pointing to us as a result of
238 * F_SETOWN with our pid.
240 mtx_lock(&Giant); /* XXX: not sure if needed */
241 funsetownlst(&p->p_sigiolst);
245 * If this process has an nlminfo data area (for lockd), release it
247 if (nlminfo_release_p != NULL && p->p_nlminfo != NULL)
248 (*nlminfo_release_p)(p);
251 * Close open files and release open-file table.
257 * If this thread tickled GEOM, we need to wait for the giggling to
258 * stop before we return to userland
260 if (td->td_pflags & TDP_GEOM)
264 * Remove ourself from our leader's peer list and wake our leader.
266 mtx_lock(&ppeers_lock);
267 if (p->p_leader->p_peers) {
269 while (q->p_peers != p)
271 q->p_peers = p->p_peers;
274 mtx_unlock(&ppeers_lock);
276 /* The next two chunks should probably be moved to vmspace_exit. */
279 * Release user portion of address space.
280 * This releases references to vnodes,
281 * which could cause I/O if the file has been unlinked.
282 * Need to do this early enough that we can still sleep.
283 * Can't free the entire vmspace as the kernel stack
284 * may be mapped within that space also.
286 * Processes sharing the same vmspace may exit in one order, and
287 * get cleaned up by vmspace_exit() in a different order. The
288 * last exiting process to reach this point releases as much of
289 * the environment as it can, and the last process cleaned up
290 * by vmspace_exit() (which decrements exitingcnt) cleans up the
293 atomic_add_int(&vm->vm_exitingcnt, 1);
295 refcnt = vm->vm_refcnt;
296 while (!atomic_cmpset_int(&vm->vm_refcnt, refcnt, refcnt - 1));
299 pmap_remove_pages(vmspace_pmap(vm), vm_map_min(&vm->vm_map),
300 vm_map_max(&vm->vm_map));
301 (void) vm_map_remove(&vm->vm_map, vm_map_min(&vm->vm_map),
302 vm_map_max(&vm->vm_map));
305 sx_xlock(&proctree_lock);
306 if (SESS_LEADER(p)) {
311 locked = VFS_LOCK_GIANT(sp->s_ttyvp->v_mount);
313 * Controlling process.
314 * Signal foreground pgrp,
315 * drain controlling terminal
316 * and revoke access to controlling terminal.
318 if (sp->s_ttyp && (sp->s_ttyp->t_session == sp)) {
320 if (sp->s_ttyp->t_pgrp) {
321 PGRP_LOCK(sp->s_ttyp->t_pgrp);
322 pgsignal(sp->s_ttyp->t_pgrp, SIGHUP, 1);
323 PGRP_UNLOCK(sp->s_ttyp->t_pgrp);
325 /* XXX tp should be locked. */
326 sx_xunlock(&proctree_lock);
328 sx_xlock(&proctree_lock);
330 * The tty could have been revoked
335 SESS_LOCK(p->p_session);
337 SESS_UNLOCK(p->p_session);
338 sx_xunlock(&proctree_lock);
339 VOP_LOCK(ttyvp, LK_EXCLUSIVE, td);
340 VOP_REVOKE(ttyvp, REVOKEALL);
342 sx_xlock(&proctree_lock);
347 SESS_LOCK(p->p_session);
349 SESS_UNLOCK(p->p_session);
353 * s_ttyp is not zero'd; we use this to indicate
354 * that the session once had a controlling terminal.
355 * (for logging and informational purposes)
357 VFS_UNLOCK_GIANT(locked);
359 SESS_LOCK(p->p_session);
361 SESS_UNLOCK(p->p_session);
363 fixjobc(p, p->p_pgrp, 0);
364 sx_xunlock(&proctree_lock);
365 (void)acct_process(td);
368 * Drain any pending records on the thread and release the trace
369 * file. It might be better if drain-and-clear were atomic.
373 mtx_lock(&ktrace_mtx);
374 p->p_traceflag = 0; /* don't trace the vrele() */
375 tracevp = p->p_tracevp;
377 tracecred = p->p_tracecred;
378 p->p_tracecred = NULL;
379 mtx_unlock(&ktrace_mtx);
381 if (tracevp != NULL) {
382 locked = VFS_LOCK_GIANT(tracevp->v_mount);
384 VFS_UNLOCK_GIANT(locked);
386 if (tracecred != NULL)
390 * Release reference to text vnode
392 if ((vtmp = p->p_textvp) != NULL) {
394 locked = VFS_LOCK_GIANT(vtmp->v_mount);
396 VFS_UNLOCK_GIANT(locked);
400 * Release our limits structure.
409 * Remove proc from allproc queue and pidhash chain.
410 * Place onto zombproc. Unlink from parent's child list.
412 sx_xlock(&allproc_lock);
413 LIST_REMOVE(p, p_list);
414 LIST_INSERT_HEAD(&zombproc, p, p_list);
415 LIST_REMOVE(p, p_hash);
416 sx_xunlock(&allproc_lock);
418 sx_xlock(&proctree_lock);
419 q = LIST_FIRST(&p->p_children);
420 if (q != NULL) /* only need this if any child is S_ZOMB */
422 for (; q != NULL; q = nq) {
423 nq = LIST_NEXT(q, p_sibling);
425 proc_reparent(q, initproc);
426 q->p_sigparent = SIGCHLD;
428 * Traced processes are killed
429 * since their existence means someone is screwing up.
431 if (q->p_flag & P_TRACED) {
432 q->p_flag &= ~(P_TRACED | P_STOPPED_TRACE);
439 * Save exit status and finalize rusage info except for times,
440 * adding in child rusage info.
445 p->p_stats->p_ru.ru_nvcsw++;
446 *p->p_ru = p->p_stats->p_ru;
447 ruadd(p->p_ru, &p->p_rux, &p->p_stats->p_cru, &p->p_crux);
450 * Notify interested parties of our demise.
452 KNOTE_LOCKED(&p->p_klist, NOTE_EXIT);
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);
478 * If this was the last child of our parent, notify
479 * parent, so in case he was wait(2)ing, he will
482 if (LIST_EMPTY(&pp->p_children))
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 PROC_UNLOCK(p->p_pptr);
498 * If this is a kthread, then wakeup anyone waiting for it to exit.
500 if (p->p_flag & P_KTHREAD)
505 * Finally, call machine-dependent code to release the remaining
506 * resources including address space.
507 * The address space is released by "vmspace_exitfree(p)" in
512 WITNESS_WARN(WARN_PANIC, &proctree_lock.sx_object,
513 "process (pid %d) exiting", p->p_pid);
516 PROC_LOCK(p->p_pptr);
517 sx_xunlock(&proctree_lock);
520 * We have to wait until after acquiring all locks before
521 * changing p_state. We need to avoid all possible context
522 * switches (including ones from blocking on a mutex) while
523 * marked as a zombie. We also have to set the zombie state
524 * before we release the parent process' proc lock to avoid
525 * a lost wakeup. So, we first call wakeup, then we grab the
526 * sched lock, update the state, and release the parent process'
530 mtx_lock_spin(&sched_lock);
531 p->p_state = PRS_ZOMBIE;
532 PROC_UNLOCK(p->p_pptr);
534 /* Do the same timestamp bookkeeping that mi_switch() would do. */
535 binuptime(&new_switchtime);
536 bintime_add(&p->p_rux.rux_runtime, &new_switchtime);
537 bintime_sub(&p->p_rux.rux_runtime, PCPU_PTR(switchtime));
538 PCPU_SET(switchtime, new_switchtime);
539 PCPU_SET(switchticks, ticks);
542 sched_exit(p->p_pptr, td);
545 * Hopefully no one will try to deliver a signal to the process this
548 knlist_destroy(&p->p_klist);
551 * Make sure the scheduler takes this thread out of its tables etc.
552 * This will also release this thread's reference to the ucred.
553 * Other thread parts to release include pcb bits and such.
559 #ifndef _SYS_SYSPROTO_H_
571 abort2(struct thread *td, struct abort2_args *uap)
573 struct proc *p = td->td_proc;
578 error = 0; /* satisfy compiler */
581 * Do it right now so we can log either proper call of abort2(), or
582 * note, that invalid argument was passed. 512 is big enough to
583 * handle 16 arguments' descriptions with additional comments.
585 sb = sbuf_new(NULL, NULL, 512, SBUF_FIXEDLEN);
587 sbuf_printf(sb, "%s(pid %d uid %d) aborted: ",
588 p->p_comm, p->p_pid, td->td_ucred->cr_uid);
590 * Since we can't return from abort2(), send SIGKILL in cases, where
591 * abort2() was called improperly
594 /* Prevent from DoSes from user-space. */
595 if (uap->nargs < 0 || uap->nargs > 16)
597 if (uap->args == NULL)
599 error = copyin(uap->args, uargs, uap->nargs * sizeof(void *));
603 * Limit size of 'reason' string to 128. Will fit even when
604 * maximal number of arguments was chosen to be logged.
606 if (uap->why != NULL) {
607 error = sbuf_copyin(sb, uap->why, 128);
611 sbuf_printf(sb, "(null)");
614 sbuf_printf(sb, "(");
615 for (i = 0;i < uap->nargs; i++)
616 sbuf_printf(sb, "%s%p", i == 0 ? "" : ", ", uargs[i]);
617 sbuf_printf(sb, ")");
620 * Final stage: arguments were proper, string has been
621 * successfully copied from userspace, and copying pointers
622 * from user-space succeed.
626 if (sig == SIGKILL) {
628 sbuf_printf(sb, " (Reason text inaccessible)");
632 log(LOG_INFO, "%s", sbuf_data(sb));
634 exit1(td, W_EXITCODE(0, sig));
641 * The dirty work is handled by kern_wait().
646 owait(struct thread *td, struct owait_args *uap __unused)
650 error = kern_wait(td, WAIT_ANY, &status, 0, NULL);
652 td->td_retval[1] = status;
655 #endif /* COMPAT_43 */
658 * The dirty work is handled by kern_wait().
663 wait4(struct thread *td, struct wait_args *uap)
665 struct rusage ru, *rup;
668 if (uap->rusage != NULL)
672 error = kern_wait(td, uap->pid, &status, uap->options, rup);
673 if (uap->status != NULL && error == 0)
674 error = copyout(&status, uap->status, sizeof(status));
675 if (uap->rusage != NULL && error == 0)
676 error = copyout(&ru, uap->rusage, sizeof(struct rusage));
681 kern_wait(struct thread *td, pid_t pid, int *status, int options,
682 struct rusage *rusage)
684 struct proc *p, *q, *t;
693 if (options &~ (WUNTRACED|WNOHANG|WCONTINUED|WLINUXCLONE))
696 if (q->p_flag & P_STATCHILD) {
698 q->p_flag &= ~P_STATCHILD;
702 sx_xlock(&proctree_lock);
703 LIST_FOREACH(p, &q->p_children, p_sibling) {
705 if (pid != WAIT_ANY &&
706 p->p_pid != pid && p->p_pgid != -pid) {
710 if (p_canwait(td, p)) {
716 * This special case handles a kthread spawned by linux_clone
717 * (see linux_misc.c). The linux_wait4 and linux_waitpid
718 * functions need to be able to distinguish between waiting
719 * on a process and waiting on a thread. It is a thread if
720 * p_sigparent is not SIGCHLD, and the WLINUXCLONE option
721 * signifies we want to wait for threads and not processes.
723 if ((p->p_sigparent != SIGCHLD) ^
724 ((options & WLINUXCLONE) != 0)) {
730 if (p->p_state == PRS_ZOMBIE) {
733 * It is possible that the last thread of this
734 * process is still running on another CPU
735 * in thread_exit() after having dropped the process
736 * lock via PROC_UNLOCK() but before it has completed
737 * cpu_throw(). In that case, the other thread must
738 * still hold sched_lock, so simply by acquiring
739 * sched_lock once we will wait long enough for the
740 * thread to exit in that case.
742 mtx_lock_spin(&sched_lock);
743 mtx_unlock_spin(&sched_lock);
745 td->td_retval[0] = p->p_pid;
747 *status = p->p_xstat; /* convert to int */
750 calcru(p, &rusage->ru_utime, &rusage->ru_stime);
754 sigqueue_take(p->p_ksi);
758 * If we got the child via a ptrace 'attach',
759 * we need to give it back to the old parent.
762 if (p->p_oppid && (t = pfind(p->p_oppid)) != NULL) {
767 tdsignal(t, NULL, SIGCHLD, p->p_ksi);
770 sx_xunlock(&proctree_lock);
775 * Remove other references to this process to ensure
776 * we have an exclusive reference.
778 sx_xlock(&allproc_lock);
779 LIST_REMOVE(p, p_list); /* off zombproc */
780 sx_xunlock(&allproc_lock);
781 LIST_REMOVE(p, p_sibling);
783 sx_xunlock(&proctree_lock);
786 * As a side effect of this lock, we know that
787 * all other writes to this proc are visible now, so
788 * no more locking is needed for p.
791 p->p_xstat = 0; /* XXX: why? */
794 ruadd(&q->p_stats->p_cru, &q->p_crux, p->p_ru,
797 FREE(p->p_ru, M_ZOMBIE);
801 * Decrement the count of procs running with this uid.
803 (void)chgproccnt(p->p_ucred->cr_ruidinfo, -1, 0);
806 * Free credentials, arguments, and sigacts.
810 pargs_drop(p->p_args);
812 sigacts_free(p->p_sigacts);
816 * Do any thread-system specific cleanups.
821 * Give vm and machine-dependent layer a chance
822 * to free anything that cpu_exit couldn't
823 * release while still running in process context.
829 KASSERT(FIRST_THREAD_IN_PROC(p),
830 ("kern_wait: no residual thread!"));
831 uma_zfree(proc_zone, p);
832 sx_xlock(&allproc_lock);
834 sx_xunlock(&allproc_lock);
837 mtx_lock_spin(&sched_lock);
838 if ((p->p_flag & P_STOPPED_SIG) &&
839 (p->p_suspcount == p->p_numthreads) &&
840 (p->p_flag & P_WAITED) == 0 &&
841 (p->p_flag & P_TRACED || options & WUNTRACED)) {
842 mtx_unlock_spin(&sched_lock);
843 p->p_flag |= P_WAITED;
844 sx_xunlock(&proctree_lock);
845 td->td_retval[0] = p->p_pid;
847 *status = W_STOPCODE(p->p_xstat);
851 sigqueue_take(p->p_ksi);
856 mtx_unlock_spin(&sched_lock);
857 if (options & WCONTINUED && (p->p_flag & P_CONTINUED)) {
858 sx_xunlock(&proctree_lock);
859 td->td_retval[0] = p->p_pid;
860 p->p_flag &= ~P_CONTINUED;
864 sigqueue_take(p->p_ksi);
874 sx_xunlock(&proctree_lock);
877 if (options & WNOHANG) {
878 sx_xunlock(&proctree_lock);
879 td->td_retval[0] = 0;
883 sx_xunlock(&proctree_lock);
884 if (q->p_flag & P_STATCHILD) {
885 q->p_flag &= ~P_STATCHILD;
888 error = msleep(q, &q->p_mtx, PWAIT | PCATCH, "wait", 0);
896 * Make process 'parent' the new parent of process 'child'.
897 * Must be called with an exclusive hold of proctree lock.
900 proc_reparent(struct proc *child, struct proc *parent)
903 sx_assert(&proctree_lock, SX_XLOCKED);
904 PROC_LOCK_ASSERT(child, MA_OWNED);
905 if (child->p_pptr == parent)
908 LIST_REMOVE(child, p_sibling);
909 LIST_INSERT_HEAD(&parent->p_children, child, p_sibling);
910 child->p_pptr = parent;