2 * SPDX-License-Identifier: BSD-3-Clause
4 * Copyright (c) 1982, 1986, 1989, 1991, 1993
5 * The Regents of the University of California. All rights reserved.
6 * (c) UNIX System Laboratories, Inc.
7 * All or some portions of this file are derived from material licensed
8 * to the University of California by American Telephone and Telegraph
9 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
10 * the permission of UNIX System Laboratories, Inc.
12 * Redistribution and use in source and binary forms, with or without
13 * modification, are permitted provided that the following conditions
15 * 1. Redistributions of source code must retain the above copyright
16 * notice, this list of conditions and the following disclaimer.
17 * 2. Redistributions in binary form must reproduce the above copyright
18 * notice, this list of conditions and the following disclaimer in the
19 * documentation and/or other materials provided with the distribution.
20 * 3. Neither the name of the University nor the names of its contributors
21 * may be used to endorse or promote products derived from this software
22 * without specific prior written permission.
24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36 * @(#)kern_exit.c 8.7 (Berkeley) 2/12/94
39 #include <sys/cdefs.h>
40 __FBSDID("$FreeBSD$");
43 #include "opt_ktrace.h"
45 #include <sys/param.h>
46 #include <sys/systm.h>
47 #include <sys/sysproto.h>
48 #include <sys/capsicum.h>
49 #include <sys/eventhandler.h>
50 #include <sys/kernel.h>
52 #include <sys/malloc.h>
54 #include <sys/mutex.h>
56 #include <sys/procdesc.h>
60 #include <sys/vmmeter.h>
61 #include <sys/vnode.h>
62 #include <sys/racct.h>
63 #include <sys/resourcevar.h>
65 #include <sys/signalvar.h>
66 #include <sys/sched.h>
68 #include <sys/syscallsubr.h>
69 #include <sys/sysctl.h>
70 #include <sys/syslog.h>
71 #include <sys/ptrace.h>
72 #include <sys/acct.h> /* for acct_process() function prototype */
73 #include <sys/filedesc.h>
77 #include <sys/sysent.h>
78 #include <sys/timers.h>
79 #include <sys/umtxvar.h>
81 #include <sys/ktrace.h>
84 #include <security/audit/audit.h>
85 #include <security/mac/mac_framework.h>
88 #include <vm/vm_extern.h>
89 #include <vm/vm_param.h>
91 #include <vm/vm_map.h>
92 #include <vm/vm_page.h>
96 #include <sys/dtrace_bsd.h>
97 dtrace_execexit_func_t dtrace_fasttrap_exit;
100 SDT_PROVIDER_DECLARE(proc);
101 SDT_PROBE_DEFINE1(proc, , , exit, "int");
103 static int kern_kill_on_dbg_exit = 1;
104 SYSCTL_INT(_kern, OID_AUTO, kill_on_debugger_exit, CTLFLAG_RWTUN,
105 &kern_kill_on_dbg_exit, 0,
106 "Kill ptraced processes when debugger exits");
108 static bool kern_wait_dequeue_sigchld = 1;
109 SYSCTL_BOOL(_kern, OID_AUTO, wait_dequeue_sigchld, CTLFLAG_RWTUN,
110 &kern_wait_dequeue_sigchld, 0,
111 "Dequeue SIGCHLD on wait(2) for live process");
114 proc_realparent(struct proc *child)
116 struct proc *p, *parent;
118 sx_assert(&proctree_lock, SX_LOCKED);
119 if ((child->p_treeflag & P_TREE_ORPHANED) == 0)
120 return (child->p_pptr->p_pid == child->p_oppid ?
121 child->p_pptr : child->p_reaper);
122 for (p = child; (p->p_treeflag & P_TREE_FIRST_ORPHAN) == 0;) {
123 /* Cannot use LIST_PREV(), since the list head is not known. */
124 p = __containerof(p->p_orphan.le_prev, struct proc,
126 KASSERT((p->p_treeflag & P_TREE_ORPHANED) != 0,
127 ("missing P_ORPHAN %p", p));
129 parent = __containerof(p->p_orphan.le_prev, struct proc,
135 reaper_abandon_children(struct proc *p, bool exiting)
137 struct proc *p1, *p2, *ptmp;
139 sx_assert(&proctree_lock, SX_XLOCKED);
140 KASSERT(p != initproc, ("reaper_abandon_children for initproc"));
141 if ((p->p_treeflag & P_TREE_REAPER) == 0)
144 LIST_FOREACH_SAFE(p2, &p->p_reaplist, p_reapsibling, ptmp) {
145 LIST_REMOVE(p2, p_reapsibling);
147 p2->p_reapsubtree = p->p_reapsubtree;
148 LIST_INSERT_HEAD(&p1->p_reaplist, p2, p_reapsibling);
149 if (exiting && p2->p_pptr == p) {
151 proc_reparent(p2, p1, true);
155 KASSERT(LIST_EMPTY(&p->p_reaplist), ("p_reaplist not empty"));
156 p->p_treeflag &= ~P_TREE_REAPER;
160 reaper_clear(struct proc *p)
165 sx_assert(&proctree_lock, SX_LOCKED);
166 LIST_REMOVE(p, p_reapsibling);
167 if (p->p_reapsubtree == 1)
170 LIST_FOREACH(p1, &p->p_reaper->p_reaplist, p_reapsibling) {
171 if (p1->p_reapsubtree == p->p_reapsubtree) {
177 proc_id_clear(PROC_ID_REAP, p->p_reapsubtree);
181 proc_clear_orphan(struct proc *p)
185 sx_assert(&proctree_lock, SA_XLOCKED);
186 if ((p->p_treeflag & P_TREE_ORPHANED) == 0)
188 if ((p->p_treeflag & P_TREE_FIRST_ORPHAN) != 0) {
189 p1 = LIST_NEXT(p, p_orphan);
191 p1->p_treeflag |= P_TREE_FIRST_ORPHAN;
192 p->p_treeflag &= ~P_TREE_FIRST_ORPHAN;
194 LIST_REMOVE(p, p_orphan);
195 p->p_treeflag &= ~P_TREE_ORPHANED;
199 exit_onexit(struct proc *p)
201 MPASS(p->p_numthreads == 1);
202 umtx_thread_exit(FIRST_THREAD_IN_PROC(p));
206 * exit -- death of process.
209 sys_sys_exit(struct thread *td, struct sys_exit_args *uap)
212 exit1(td, uap->rval, 0);
217 proc_set_p2_wexit(struct proc *p)
219 PROC_LOCK_ASSERT(p, MA_OWNED);
220 p->p_flag2 |= P2_WEXIT;
224 * Exit: deallocate address space and other resources, change proc state to
225 * zombie, and unlink proc from allproc and parent's lists. Save exit status
226 * and rusage for wait(). Check for child processes and orphan them.
229 exit1(struct thread *td, int rval, int signo)
231 struct proc *p, *nq, *q, *t;
233 ksiginfo_t *ksi, *ksi1;
236 mtx_assert(&Giant, MA_NOTOWNED);
237 KASSERT(rval == 0 || signo == 0, ("exit1 rv %d sig %d", rval, signo));
238 TSPROCEXIT(td->td_proc->p_pid);
242 * XXX in case we're rebooting we just let init die in order to
243 * work around an unsolved stack overflow seen very late during
244 * shutdown on sparc64 when the gmirror worker process exists.
245 * XXX what to do now that sparc64 is gone... remove if?
247 if (p == initproc && rebooting == 0) {
248 printf("init died (signal %d, exit %d)\n", signo, rval);
249 panic("Going nowhere without my init!");
253 * Deref SU mp, since the thread does not return to userspace.
255 td_softdep_cleanup(td);
258 * MUST abort all other threads before proceeding past here.
261 proc_set_p2_wexit(p);
264 * First check if some other thread or external request got
265 * here before us. If so, act appropriately: exit or suspend.
266 * We must ensure that stop requests are handled before we set
269 thread_suspend_check(0);
270 while (p->p_flag & P_HADTHREADS) {
272 * Kill off the other threads. This requires
273 * some co-operation from other parts of the kernel
274 * so it may not be instantaneous. With this state set
275 * any thread attempting to interruptibly
276 * sleep will return immediately with EINTR or EWOULDBLOCK
277 * which will hopefully force them to back out to userland
278 * freeing resources as they go. Any thread attempting
279 * to return to userland will thread_exit() from ast().
280 * thread_exit() will unsuspend us when the last of the
281 * other threads exits.
282 * If there is already a thread singler after resumption,
283 * calling thread_single() will fail; in that case, we just
284 * re-check all suspension request, the thread should
285 * either be suspended there or exit.
287 if (!thread_single(p, SINGLE_EXIT))
289 * All other activity in this process is now
290 * stopped. Threading support has been turned
295 * Recheck for new stop or suspend requests which
296 * might appear while process lock was dropped in
299 thread_suspend_check(0);
301 KASSERT(p->p_numthreads == 1,
302 ("exit1: proc %p exiting with %d threads", p, p->p_numthreads));
303 racct_sub(p, RACCT_NTHR, 1);
305 /* Let event handler change exit status */
310 * Ignore any pending request to stop due to a stop signal.
311 * Once P_WEXIT is set, future requests will be ignored as
314 p->p_flag &= ~P_STOPPED_SIG;
315 KASSERT(!P_SHOULDSTOP(p), ("exiting process is stopped"));
317 /* Note that we are exiting. */
318 p->p_flag |= P_WEXIT;
321 * Wait for any processes that have a hold on our vmspace to
322 * release their reference.
324 while (p->p_lock > 0)
325 msleep(&p->p_lock, &p->p_mtx, PWAIT, "exithold", 0);
328 /* Drain the limit callout while we don't have the proc locked */
329 callout_drain(&p->p_limco);
333 * The Sun BSM exit token contains two components: an exit status as
334 * passed to exit(), and a return value to indicate what sort of exit
335 * it was. The exit status is WEXITSTATUS(rv), but it's not clear
336 * what the return value is.
338 AUDIT_ARG_EXIT(rval, 0);
339 AUDIT_SYSCALL_EXIT(0, td);
342 /* Are we a task leader with peers? */
343 if (p->p_peers != NULL && p == p->p_leader) {
344 mtx_lock(&ppeers_lock);
348 kern_psignal(q, SIGKILL);
352 while (p->p_peers != NULL)
353 msleep(p, &ppeers_lock, PWAIT, "exit1", 0);
354 mtx_unlock(&ppeers_lock);
360 * Check if any loadable modules need anything done at process exit.
361 * E.g. SYSV IPC stuff.
362 * Event handler could change exit status.
363 * XXX what if one of these generates an error?
365 EVENTHANDLER_DIRECT_INVOKE(process_exit, p);
368 * If parent is waiting for us to exit or exec,
369 * P_PPWAIT is set; we will wakeup the parent below.
376 * Stop the real interval timer. If the handler is currently
377 * executing, prevent it from rearming itself and let it finish.
379 if (timevalisset(&p->p_realtimer.it_value) &&
380 callout_stop(&p->p_itcallout) == 0) {
381 timevalclear(&p->p_realtimer.it_interval);
383 callout_drain(&p->p_itcallout);
388 if (p->p_sysent->sv_onexit != NULL)
389 p->p_sysent->sv_onexit(p);
393 * Reset any sigio structures pointing to us as a result of
394 * F_SETOWN with our pid. The P_WEXIT flag interlocks with fsetown().
396 funsetownlst(&p->p_sigiolst);
399 * Close open files and release open-file table.
406 * If this thread tickled GEOM, we need to wait for the giggling to
407 * stop before we return to userland
409 if (td->td_pflags & TDP_GEOM)
413 * Remove ourself from our leader's peer list and wake our leader.
415 if (p->p_leader->p_peers != NULL) {
416 mtx_lock(&ppeers_lock);
417 if (p->p_leader->p_peers != NULL) {
419 while (q->p_peers != p)
421 q->p_peers = p->p_peers;
424 mtx_unlock(&ppeers_lock);
427 exec_free_abi_mappings(p);
429 (void)acct_process(td);
435 * Release reference to text vnode etc
437 if (p->p_textvp != NULL) {
441 if (p->p_textdvp != NULL) {
445 if (p->p_binname != NULL) {
446 free(p->p_binname, M_PARGS);
451 * Release our limits structure.
453 lim_free(p->p_limit);
459 * Call machine-dependent code to release any
460 * machine-dependent resources other than the address space.
461 * The address space is released by "vmspace_exitfree(p)" in
466 WITNESS_WARN(WARN_PANIC, NULL, "process (pid %d) exiting", p->p_pid);
469 * Remove from allproc. It still sits in the hash.
471 sx_xlock(&allproc_lock);
472 LIST_REMOVE(p, p_list);
476 * Used by ddb's 'ps' command to find this process via the
479 p->p_list.le_prev = NULL;
481 sx_xunlock(&allproc_lock);
483 sx_xlock(&proctree_lock);
485 p->p_flag &= ~(P_TRACED | P_PPWAIT | P_PPTRACE);
489 * killjobc() might drop and re-acquire proctree_lock to
490 * revoke control tty if exiting process was a session leader.
495 * Reparent all children processes:
496 * - traced ones to the original parent (or init if we are that parent)
499 q = LIST_FIRST(&p->p_children);
500 if (q != NULL) /* only need this if any child is S_ZOMB */
502 for (; q != NULL; q = nq) {
503 nq = LIST_NEXT(q, p_sibling);
504 ksi = ksiginfo_alloc(M_WAITOK);
506 q->p_sigparent = SIGCHLD;
508 if ((q->p_flag & P_TRACED) == 0) {
509 proc_reparent(q, q->p_reaper, true);
510 if (q->p_state == PRS_ZOMBIE) {
512 * Inform reaper about the reparented
513 * zombie, since wait(2) has something
514 * new to report. Guarantee queueing
515 * of the SIGCHLD signal, similar to
516 * the _exit() behaviour, by providing
517 * our ksiginfo. Ksi is freed by the
520 if (q->p_ksi == NULL) {
523 ksiginfo_copy(q->p_ksi, ksi);
524 ksi->ksi_flags |= KSI_INS;
528 PROC_LOCK(q->p_reaper);
529 pksignal(q->p_reaper, SIGCHLD, ksi1);
530 PROC_UNLOCK(q->p_reaper);
531 } else if (q->p_pdeathsig > 0) {
533 * The child asked to received a signal
536 kern_psignal(q, q->p_pdeathsig);
540 * Traced processes are killed by default
541 * since their existence means someone is
544 t = proc_realparent(q);
546 proc_reparent(q, q->p_reaper, true);
549 proc_reparent(q, t, true);
553 * Since q was found on our children list, the
554 * proc_reparent() call moved q to the orphan
555 * list due to present P_TRACED flag. Clear
556 * orphan link for q now while q is locked.
558 proc_clear_orphan(q);
559 q->p_flag &= ~P_TRACED;
560 q->p_flag2 &= ~P2_PTRACE_FSTP;
563 FOREACH_THREAD_IN_PROC(q, tdt) {
564 tdt->td_dbgflags &= ~(TDB_SUSPEND | TDB_XSIG |
568 if (kern_kill_on_dbg_exit) {
569 q->p_flag &= ~P_STOPPED_TRACE;
570 kern_psignal(q, SIGKILL);
571 } else if ((q->p_flag & (P_STOPPED_TRACE |
572 P_STOPPED_SIG)) != 0) {
573 sigqueue_delete_proc(q, SIGTRAP);
583 * Also get rid of our orphans.
585 while ((q = LIST_FIRST(&p->p_orphans)) != NULL) {
587 KASSERT(q->p_oppid == p->p_pid,
588 ("orphan %p of %p has unexpected oppid %d", q, p,
590 q->p_oppid = q->p_reaper->p_pid;
593 * If we are the real parent of this process
594 * but it has been reparented to a debugger, then
595 * check if it asked for a signal when we exit.
597 if (q->p_pdeathsig > 0)
598 kern_psignal(q, q->p_pdeathsig);
599 CTR2(KTR_PTRACE, "exit: pid %d, clearing orphan %d", p->p_pid,
601 proc_clear_orphan(q);
606 if (SDT_PROBES_ENABLED()) {
607 int reason = CLD_EXITED;
608 if (WCOREDUMP(signo))
610 else if (WIFSIGNALED(signo))
612 SDT_PROBE1(proc, , , exit, reason);
616 /* Save exit status. */
620 if (p->p_sysent->sv_ontdexit != NULL)
621 p->p_sysent->sv_ontdexit(td);
625 * Tell the DTrace fasttrap provider about the exit if it
626 * has declared an interest.
628 if (dtrace_fasttrap_exit)
629 dtrace_fasttrap_exit(p);
633 * Notify interested parties of our demise.
635 KNOTE_LOCKED(p->p_klist, NOTE_EXIT);
638 * If this is a process with a descriptor, we may not need to deliver
639 * a signal to the parent. proctree_lock is held over
640 * procdesc_exit() to serialize concurrent calls to close() and
644 if (p->p_procdesc == NULL || procdesc_exit(p)) {
646 * Notify parent that we're gone. If parent has the
647 * PS_NOCLDWAIT flag set, or if the handler is set to SIG_IGN,
648 * notify process 1 instead (and hope it will handle this
651 PROC_LOCK(p->p_pptr);
652 mtx_lock(&p->p_pptr->p_sigacts->ps_mtx);
653 if (p->p_pptr->p_sigacts->ps_flag &
654 (PS_NOCLDWAIT | PS_CLDSIGIGN)) {
657 mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx);
660 proc_reparent(p, p->p_reaper, true);
661 p->p_sigparent = SIGCHLD;
662 PROC_LOCK(p->p_pptr);
665 * Notify parent, so in case he was wait(2)ing or
666 * executing waitpid(2) with our pid, he will
671 mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx);
673 if (p->p_pptr == p->p_reaper || p->p_pptr == initproc) {
675 } else if (p->p_sigparent != 0) {
676 if (p->p_sigparent == SIGCHLD) {
678 } else { /* LINUX thread */
683 PROC_LOCK(p->p_pptr);
684 sx_xunlock(&proctree_lock);
686 if (signal_parent == 1) {
688 } else if (signal_parent == 2) {
689 kern_psignal(p->p_pptr, p->p_sigparent);
692 /* Tell the prison that we are gone. */
693 prison_proc_free(p->p_ucred->cr_prison);
696 * The state PRS_ZOMBIE prevents other proesses from sending
697 * signal to the process, to avoid memory leak, we free memory
698 * for signal queue at the time when the state is set.
700 sigqueue_flush(&p->p_sigqueue);
701 sigqueue_flush(&td->td_sigqueue);
704 * We have to wait until after acquiring all locks before
705 * changing p_state. We need to avoid all possible context
706 * switches (including ones from blocking on a mutex) while
707 * marked as a zombie. We also have to set the zombie state
708 * before we release the parent process' proc lock to avoid
709 * a lost wakeup. So, we first call wakeup, then we grab the
710 * sched lock, update the state, and release the parent process'
714 cv_broadcast(&p->p_pwait);
715 sched_exit(p->p_pptr, td);
717 p->p_state = PRS_ZOMBIE;
718 PROC_UNLOCK(p->p_pptr);
721 * Save our children's rusage information in our exit rusage.
724 ruadd(&p->p_ru, &p->p_rux, &p->p_stats->p_cru, &p->p_crux);
728 * Make sure the scheduler takes this thread out of its tables etc.
729 * This will also release this thread's reference to the ucred.
730 * Other thread parts to release include pcb bits and such.
735 #ifndef _SYS_SYSPROTO_H_
744 sys_abort2(struct thread *td, struct abort2_args *uap)
746 struct proc *p = td->td_proc;
752 * Do it right now so we can log either proper call of abort2(), or
753 * note, that invalid argument was passed. 512 is big enough to
754 * handle 16 arguments' descriptions with additional comments.
756 sb = sbuf_new(NULL, NULL, 512, SBUF_FIXEDLEN);
758 sbuf_printf(sb, "%s(pid %d uid %d) aborted: ",
759 p->p_comm, p->p_pid, td->td_ucred->cr_uid);
761 * Since we can't return from abort2(), send SIGKILL in cases, where
762 * abort2() was called improperly
765 /* Prevent from DoSes from user-space. */
766 if (uap->nargs < 0 || uap->nargs > 16)
768 if (uap->nargs > 0) {
769 if (uap->args == NULL)
771 error = copyin(uap->args, uargs, uap->nargs * sizeof(void *));
776 * Limit size of 'reason' string to 128. Will fit even when
777 * maximal number of arguments was chosen to be logged.
779 if (uap->why != NULL) {
780 error = sbuf_copyin(sb, uap->why, 128);
784 sbuf_printf(sb, "(null)");
786 if (uap->nargs > 0) {
787 sbuf_printf(sb, "(");
788 for (i = 0;i < uap->nargs; i++)
789 sbuf_printf(sb, "%s%p", i == 0 ? "" : ", ", uargs[i]);
790 sbuf_printf(sb, ")");
793 * Final stage: arguments were proper, string has been
794 * successfully copied from userspace, and copying pointers
795 * from user-space succeed.
799 if (sig == SIGKILL) {
801 sbuf_printf(sb, " (Reason text inaccessible)");
805 log(LOG_INFO, "%s", sbuf_data(sb));
813 * The dirty work is handled by kern_wait().
816 owait(struct thread *td, struct owait_args *uap __unused)
820 error = kern_wait(td, WAIT_ANY, &status, 0, NULL);
822 td->td_retval[1] = status;
825 #endif /* COMPAT_43 */
828 * The dirty work is handled by kern_wait().
831 sys_wait4(struct thread *td, struct wait4_args *uap)
833 struct rusage ru, *rup;
836 if (uap->rusage != NULL)
840 error = kern_wait(td, uap->pid, &status, uap->options, rup);
841 if (uap->status != NULL && error == 0 && td->td_retval[0] != 0)
842 error = copyout(&status, uap->status, sizeof(status));
843 if (uap->rusage != NULL && error == 0 && td->td_retval[0] != 0)
844 error = copyout(&ru, uap->rusage, sizeof(struct rusage));
849 sys_wait6(struct thread *td, struct wait6_args *uap)
851 struct __wrusage wru, *wrup;
857 idtype = uap->idtype;
860 if (uap->wrusage != NULL)
865 if (uap->info != NULL) {
867 bzero(sip, sizeof(*sip));
872 * We expect all callers of wait6() to know about WEXITED and
875 error = kern_wait6(td, idtype, id, &status, uap->options, wrup, sip);
877 if (uap->status != NULL && error == 0 && td->td_retval[0] != 0)
878 error = copyout(&status, uap->status, sizeof(status));
879 if (uap->wrusage != NULL && error == 0 && td->td_retval[0] != 0)
880 error = copyout(&wru, uap->wrusage, sizeof(wru));
881 if (uap->info != NULL && error == 0)
882 error = copyout(&si, uap->info, sizeof(si));
887 * Reap the remains of a zombie process and optionally return status and
888 * rusage. Asserts and will release both the proctree_lock and the process
889 * lock as part of its work.
892 proc_reap(struct thread *td, struct proc *p, int *status, int options)
896 sx_assert(&proctree_lock, SA_XLOCKED);
897 PROC_LOCK_ASSERT(p, MA_OWNED);
898 KASSERT(p->p_state == PRS_ZOMBIE, ("proc_reap: !PRS_ZOMBIE"));
900 mtx_spin_wait_unlocked(&p->p_slock);
905 *status = KW_EXITCODE(p->p_xexit, p->p_xsig);
906 if (options & WNOWAIT) {
908 * Only poll, returning the status. Caller does not wish to
909 * release the proc struct just yet.
912 sx_xunlock(&proctree_lock);
917 sigqueue_take(p->p_ksi);
921 * If we got the child via a ptrace 'attach', we need to give it back
924 if (p->p_oppid != p->p_pptr->p_pid) {
926 t = proc_realparent(p);
930 "wait: traced child %d moved back to parent %d", p->p_pid,
932 proc_reparent(p, t, false);
934 pksignal(t, SIGCHLD, p->p_ksi);
936 cv_broadcast(&p->p_pwait);
938 sx_xunlock(&proctree_lock);
944 * Remove other references to this process to ensure we have an
945 * exclusive reference.
947 sx_xlock(PIDHASHLOCK(p->p_pid));
948 LIST_REMOVE(p, p_hash);
949 sx_xunlock(PIDHASHLOCK(p->p_pid));
950 LIST_REMOVE(p, p_sibling);
951 reaper_abandon_children(p, true);
954 proc_clear_orphan(p);
957 if (p->p_procdesc != NULL)
959 sx_xunlock(&proctree_lock);
961 proc_id_clear(PROC_ID_PID, p->p_pid);
964 knlist_detach(p->p_klist);
969 * Removal from allproc list and process group list paired with
970 * PROC_LOCK which was executed during that time should guarantee
971 * nothing can reach this process anymore. As such further locking
974 p->p_xexit = p->p_xsig = 0; /* XXX: why? */
977 ruadd(&q->p_stats->p_cru, &q->p_crux, &p->p_ru, &p->p_rux);
981 * Decrement the count of procs running with this uid.
983 (void)chgproccnt(p->p_ucred->cr_ruidinfo, -1, 0);
986 * Destroy resource accounting information associated with the process.
991 racct_sub(p, RACCT_NPROC, 1);
998 * Free credentials, arguments, and sigacts.
1001 pargs_drop(p->p_args);
1003 sigacts_free(p->p_sigacts);
1004 p->p_sigacts = NULL;
1007 * Do any thread-system specific cleanups.
1012 * Give vm and machine-dependent layer a chance to free anything that
1013 * cpu_exit couldn't release while still running in process context.
1017 mac_proc_destroy(p);
1020 KASSERT(FIRST_THREAD_IN_PROC(p),
1021 ("proc_reap: no residual thread!"));
1022 uma_zfree(proc_zone, p);
1023 atomic_add_int(&nprocs, -1);
1027 proc_to_reap(struct thread *td, struct proc *p, idtype_t idtype, id_t id,
1028 int *status, int options, struct __wrusage *wrusage, siginfo_t *siginfo,
1033 sx_assert(&proctree_lock, SA_XLOCKED);
1039 if (p->p_procdesc == NULL ||
1040 (p->p_pptr == td->td_proc &&
1041 (p->p_flag & P_TRACED) != 0)) {
1048 if (p->p_pid != (pid_t)id) {
1054 if (p->p_pgid != (pid_t)id) {
1060 if (p->p_session->s_sid != (pid_t)id) {
1066 if (p->p_ucred->cr_uid != (uid_t)id) {
1072 if (p->p_ucred->cr_gid != (gid_t)id) {
1078 if (p->p_ucred->cr_prison->pr_id != (int)id) {
1084 * It seems that the thread structures get zeroed out
1085 * at process exit. This makes it impossible to
1086 * support P_SETID, P_CID or P_CPUID.
1093 if (p_canwait(td, p)) {
1098 if (((options & WEXITED) == 0) && (p->p_state == PRS_ZOMBIE)) {
1104 * This special case handles a kthread spawned by linux_clone
1105 * (see linux_misc.c). The linux_wait4 and linux_waitpid
1106 * functions need to be able to distinguish between waiting
1107 * on a process and waiting on a thread. It is a thread if
1108 * p_sigparent is not SIGCHLD, and the WLINUXCLONE option
1109 * signifies we want to wait for threads and not processes.
1111 if ((p->p_sigparent != SIGCHLD) ^
1112 ((options & WLINUXCLONE) != 0)) {
1117 if (siginfo != NULL) {
1118 bzero(siginfo, sizeof(*siginfo));
1119 siginfo->si_errno = 0;
1122 * SUSv4 requires that the si_signo value is always
1123 * SIGCHLD. Obey it despite the rfork(2) interface
1124 * allows to request other signal for child exit
1127 siginfo->si_signo = SIGCHLD;
1130 * This is still a rough estimate. We will fix the
1131 * cases TRAPPED, STOPPED, and CONTINUED later.
1133 if (WCOREDUMP(p->p_xsig)) {
1134 siginfo->si_code = CLD_DUMPED;
1135 siginfo->si_status = WTERMSIG(p->p_xsig);
1136 } else if (WIFSIGNALED(p->p_xsig)) {
1137 siginfo->si_code = CLD_KILLED;
1138 siginfo->si_status = WTERMSIG(p->p_xsig);
1140 siginfo->si_code = CLD_EXITED;
1141 siginfo->si_status = p->p_xexit;
1144 siginfo->si_pid = p->p_pid;
1145 siginfo->si_uid = p->p_ucred->cr_uid;
1148 * The si_addr field would be useful additional
1149 * detail, but apparently the PC value may be lost
1150 * when we reach this point. bzero() above sets
1151 * siginfo->si_addr to NULL.
1156 * There should be no reason to limit resources usage info to
1157 * exited processes only. A snapshot about any resources used
1158 * by a stopped process may be exactly what is needed.
1160 if (wrusage != NULL) {
1161 rup = &wrusage->wru_self;
1164 calcru(p, &rup->ru_utime, &rup->ru_stime);
1167 rup = &wrusage->wru_children;
1168 *rup = p->p_stats->p_cru;
1169 calccru(p, &rup->ru_utime, &rup->ru_stime);
1172 if (p->p_state == PRS_ZOMBIE && !check_only) {
1173 proc_reap(td, p, status, options);
1180 kern_wait(struct thread *td, pid_t pid, int *status, int options,
1181 struct rusage *rusage)
1183 struct __wrusage wru, *wrup;
1189 * Translate the special pid values into the (idtype, pid)
1190 * pair for kern_wait6. The WAIT_MYPGRP case is handled by
1191 * kern_wait6() on its own.
1193 if (pid == WAIT_ANY) {
1196 } else if (pid < 0) {
1210 * For backward compatibility we implicitly add flags WEXITED
1211 * and WTRAPPED here.
1213 options |= WEXITED | WTRAPPED;
1214 ret = kern_wait6(td, idtype, id, status, options, wrup, NULL);
1216 *rusage = wru.wru_self;
1221 report_alive_proc(struct thread *td, struct proc *p, siginfo_t *siginfo,
1222 int *status, int options, int si_code)
1226 PROC_LOCK_ASSERT(p, MA_OWNED);
1227 sx_assert(&proctree_lock, SA_XLOCKED);
1228 MPASS(si_code == CLD_TRAPPED || si_code == CLD_STOPPED ||
1229 si_code == CLD_CONTINUED);
1231 cont = si_code == CLD_CONTINUED;
1232 if ((options & WNOWAIT) == 0) {
1234 p->p_flag &= ~P_CONTINUED;
1236 p->p_flag |= P_WAITED;
1237 if (kern_wait_dequeue_sigchld &&
1238 (td->td_proc->p_sysent->sv_flags & SV_SIG_WAITNDQ) == 0) {
1239 PROC_LOCK(td->td_proc);
1240 sigqueue_take(p->p_ksi);
1241 PROC_UNLOCK(td->td_proc);
1244 sx_xunlock(&proctree_lock);
1245 if (siginfo != NULL) {
1246 siginfo->si_code = si_code;
1247 siginfo->si_status = cont ? SIGCONT : p->p_xsig;
1250 *status = cont ? SIGCONT : W_STOPCODE(p->p_xsig);
1252 td->td_retval[0] = p->p_pid;
1256 kern_wait6(struct thread *td, idtype_t idtype, id_t id, int *status,
1257 int options, struct __wrusage *wrusage, siginfo_t *siginfo)
1261 int error, nfound, ret;
1264 AUDIT_ARG_VALUE((int)idtype); /* XXX - This is likely wrong! */
1265 AUDIT_ARG_PID((pid_t)id); /* XXX - This may be wrong! */
1266 AUDIT_ARG_VALUE(options);
1270 if ((pid_t)id == WAIT_MYPGRP && (idtype == P_PID || idtype == P_PGID)) {
1272 id = (id_t)q->p_pgid;
1277 /* If we don't know the option, just return. */
1278 if ((options & ~(WUNTRACED | WNOHANG | WCONTINUED | WNOWAIT |
1279 WEXITED | WTRAPPED | WLINUXCLONE)) != 0)
1281 if ((options & (WEXITED | WUNTRACED | WCONTINUED | WTRAPPED)) == 0) {
1283 * We will be unable to find any matching processes,
1284 * because there are no known events to look for.
1285 * Prefer to return error instead of blocking
1292 if (q->p_flag & P_STATCHILD) {
1294 q->p_flag &= ~P_STATCHILD;
1297 sx_xlock(&proctree_lock);
1300 LIST_FOREACH(p, &q->p_children, p_sibling) {
1302 ret = proc_to_reap(td, p, idtype, id, status, options,
1303 wrusage, siginfo, 0);
1306 else if (ret != 1) {
1307 td->td_retval[0] = pid;
1312 PROC_LOCK_ASSERT(p, MA_OWNED);
1314 if ((options & WTRAPPED) != 0 &&
1315 (p->p_flag & P_TRACED) != 0) {
1318 ((p->p_flag & (P_STOPPED_TRACE | P_STOPPED_SIG)) &&
1319 p->p_suspcount == p->p_numthreads &&
1320 (p->p_flag & P_WAITED) == 0);
1324 "wait: returning trapped pid %d status %#x "
1325 "(xstat %d) xthread %d",
1326 p->p_pid, W_STOPCODE(p->p_xsig), p->p_xsig,
1327 p->p_xthread != NULL ?
1328 p->p_xthread->td_tid : -1);
1329 report_alive_proc(td, p, siginfo, status,
1330 options, CLD_TRAPPED);
1334 if ((options & WUNTRACED) != 0 &&
1335 (p->p_flag & P_STOPPED_SIG) != 0) {
1337 report = (p->p_suspcount == p->p_numthreads &&
1338 ((p->p_flag & P_WAITED) == 0));
1341 report_alive_proc(td, p, siginfo, status,
1342 options, CLD_STOPPED);
1346 if ((options & WCONTINUED) != 0 &&
1347 (p->p_flag & P_CONTINUED) != 0) {
1348 report_alive_proc(td, p, siginfo, status, options,
1356 * Look in the orphans list too, to allow the parent to
1357 * collect it's child exit status even if child is being
1360 * Debugger detaches from the parent upon successful
1361 * switch-over from parent to child. At this point due to
1362 * re-parenting the parent loses the child to debugger and a
1363 * wait4(2) call would report that it has no children to wait
1364 * for. By maintaining a list of orphans we allow the parent
1365 * to successfully wait until the child becomes a zombie.
1368 LIST_FOREACH(p, &q->p_orphans, p_orphan) {
1369 ret = proc_to_reap(td, p, idtype, id, NULL, options,
1372 KASSERT(ret != -1, ("reaped an orphan (pid %d)",
1373 (int)td->td_retval[0]));
1381 sx_xunlock(&proctree_lock);
1384 if (options & WNOHANG) {
1385 sx_xunlock(&proctree_lock);
1386 td->td_retval[0] = 0;
1390 if (q->p_flag & P_STATCHILD) {
1391 q->p_flag &= ~P_STATCHILD;
1395 sx_xunlock(&proctree_lock);
1396 error = msleep(q, &q->p_mtx, PWAIT | PCATCH | PDROP, "wait", 0);
1403 proc_add_orphan(struct proc *child, struct proc *parent)
1406 sx_assert(&proctree_lock, SX_XLOCKED);
1407 KASSERT((child->p_flag & P_TRACED) != 0,
1408 ("proc_add_orphan: not traced"));
1410 if (LIST_EMPTY(&parent->p_orphans)) {
1411 child->p_treeflag |= P_TREE_FIRST_ORPHAN;
1412 LIST_INSERT_HEAD(&parent->p_orphans, child, p_orphan);
1414 LIST_INSERT_AFTER(LIST_FIRST(&parent->p_orphans),
1417 child->p_treeflag |= P_TREE_ORPHANED;
1421 * Make process 'parent' the new parent of process 'child'.
1422 * Must be called with an exclusive hold of proctree lock.
1425 proc_reparent(struct proc *child, struct proc *parent, bool set_oppid)
1428 sx_assert(&proctree_lock, SX_XLOCKED);
1429 PROC_LOCK_ASSERT(child, MA_OWNED);
1430 if (child->p_pptr == parent)
1433 PROC_LOCK(child->p_pptr);
1434 sigqueue_take(child->p_ksi);
1435 PROC_UNLOCK(child->p_pptr);
1436 LIST_REMOVE(child, p_sibling);
1437 LIST_INSERT_HEAD(&parent->p_children, child, p_sibling);
1439 proc_clear_orphan(child);
1440 if ((child->p_flag & P_TRACED) != 0) {
1441 proc_add_orphan(child, child->p_pptr);
1444 child->p_pptr = parent;
1446 child->p_oppid = parent->p_pid;