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>
41 #include "opt_ktrace.h"
43 #include <sys/param.h>
44 #include <sys/systm.h>
45 #include <sys/sysproto.h>
46 #include <sys/capsicum.h>
47 #include <sys/eventhandler.h>
48 #include <sys/kernel.h>
50 #include <sys/malloc.h>
52 #include <sys/mutex.h>
54 #include <sys/procdesc.h>
58 #include <sys/vmmeter.h>
59 #include <sys/vnode.h>
60 #include <sys/racct.h>
61 #include <sys/resourcevar.h>
63 #include <sys/signalvar.h>
64 #include <sys/sched.h>
66 #include <sys/syscallsubr.h>
67 #include <sys/sysctl.h>
68 #include <sys/syslog.h>
69 #include <sys/ptrace.h>
70 #include <sys/acct.h> /* for acct_process() function prototype */
71 #include <sys/filedesc.h>
75 #include <sys/sysent.h>
76 #include <sys/timers.h>
77 #include <sys/umtxvar.h>
79 #include <sys/ktrace.h>
82 #include <security/audit/audit.h>
83 #include <security/mac/mac_framework.h>
86 #include <vm/vm_extern.h>
87 #include <vm/vm_param.h>
89 #include <vm/vm_map.h>
90 #include <vm/vm_page.h>
94 #include <sys/dtrace_bsd.h>
95 dtrace_execexit_func_t dtrace_fasttrap_exit;
98 SDT_PROVIDER_DECLARE(proc);
99 SDT_PROBE_DEFINE1(proc, , , exit, "int");
101 static int kern_kill_on_dbg_exit = 1;
102 SYSCTL_INT(_kern, OID_AUTO, kill_on_debugger_exit, CTLFLAG_RWTUN,
103 &kern_kill_on_dbg_exit, 0,
104 "Kill ptraced processes when debugger exits");
106 static bool kern_wait_dequeue_sigchld = 1;
107 SYSCTL_BOOL(_kern, OID_AUTO, wait_dequeue_sigchld, CTLFLAG_RWTUN,
108 &kern_wait_dequeue_sigchld, 0,
109 "Dequeue SIGCHLD on wait(2) for live process");
112 proc_realparent(struct proc *child)
114 struct proc *p, *parent;
116 sx_assert(&proctree_lock, SX_LOCKED);
117 if ((child->p_treeflag & P_TREE_ORPHANED) == 0)
118 return (child->p_pptr->p_pid == child->p_oppid ?
119 child->p_pptr : child->p_reaper);
120 for (p = child; (p->p_treeflag & P_TREE_FIRST_ORPHAN) == 0;) {
121 /* Cannot use LIST_PREV(), since the list head is not known. */
122 p = __containerof(p->p_orphan.le_prev, struct proc,
124 KASSERT((p->p_treeflag & P_TREE_ORPHANED) != 0,
125 ("missing P_ORPHAN %p", p));
127 parent = __containerof(p->p_orphan.le_prev, struct proc,
133 reaper_abandon_children(struct proc *p, bool exiting)
135 struct proc *p1, *p2, *ptmp;
137 sx_assert(&proctree_lock, SX_XLOCKED);
138 KASSERT(p != initproc, ("reaper_abandon_children for initproc"));
139 if ((p->p_treeflag & P_TREE_REAPER) == 0)
142 LIST_FOREACH_SAFE(p2, &p->p_reaplist, p_reapsibling, ptmp) {
143 LIST_REMOVE(p2, p_reapsibling);
145 p2->p_reapsubtree = p->p_reapsubtree;
146 LIST_INSERT_HEAD(&p1->p_reaplist, p2, p_reapsibling);
147 if (exiting && p2->p_pptr == p) {
149 proc_reparent(p2, p1, true);
153 KASSERT(LIST_EMPTY(&p->p_reaplist), ("p_reaplist not empty"));
154 p->p_treeflag &= ~P_TREE_REAPER;
158 reaper_clear(struct proc *p)
163 sx_assert(&proctree_lock, SX_LOCKED);
164 LIST_REMOVE(p, p_reapsibling);
165 if (p->p_reapsubtree == 1)
168 LIST_FOREACH(p1, &p->p_reaper->p_reaplist, p_reapsibling) {
169 if (p1->p_reapsubtree == p->p_reapsubtree) {
175 proc_id_clear(PROC_ID_REAP, p->p_reapsubtree);
179 proc_clear_orphan(struct proc *p)
183 sx_assert(&proctree_lock, SA_XLOCKED);
184 if ((p->p_treeflag & P_TREE_ORPHANED) == 0)
186 if ((p->p_treeflag & P_TREE_FIRST_ORPHAN) != 0) {
187 p1 = LIST_NEXT(p, p_orphan);
189 p1->p_treeflag |= P_TREE_FIRST_ORPHAN;
190 p->p_treeflag &= ~P_TREE_FIRST_ORPHAN;
192 LIST_REMOVE(p, p_orphan);
193 p->p_treeflag &= ~P_TREE_ORPHANED;
197 exit_onexit(struct proc *p)
199 MPASS(p->p_numthreads == 1);
200 umtx_thread_exit(FIRST_THREAD_IN_PROC(p));
204 * exit -- death of process.
207 sys_sys_exit(struct thread *td, struct sys_exit_args *uap)
210 exit1(td, uap->rval, 0);
215 proc_set_p2_wexit(struct proc *p)
217 PROC_LOCK_ASSERT(p, MA_OWNED);
218 p->p_flag2 |= P2_WEXIT;
222 * Exit: deallocate address space and other resources, change proc state to
223 * zombie, and unlink proc from allproc and parent's lists. Save exit status
224 * and rusage for wait(). Check for child processes and orphan them.
227 exit1(struct thread *td, int rval, int signo)
229 struct proc *p, *nq, *q, *t;
231 ksiginfo_t *ksi, *ksi1;
234 mtx_assert(&Giant, MA_NOTOWNED);
235 KASSERT(rval == 0 || signo == 0, ("exit1 rv %d sig %d", rval, signo));
236 TSPROCEXIT(td->td_proc->p_pid);
240 * In case we're rebooting we just let init die in order to
241 * work around an issues where pid 1 might get a fatal signal.
242 * For instance, if network interface serving NFS root is
243 * going down due to reboot, page-in requests for text are
246 if (p == initproc && rebooting == 0) {
247 printf("init died (signal %d, exit %d)\n", signo, rval);
248 panic("Going nowhere without my init!");
252 * Deref SU mp, since the thread does not return to userspace.
254 td_softdep_cleanup(td);
257 * MUST abort all other threads before proceeding past here.
260 proc_set_p2_wexit(p);
263 * First check if some other thread or external request got
264 * here before us. If so, act appropriately: exit or suspend.
265 * We must ensure that stop requests are handled before we set
268 thread_suspend_check(0);
269 while (p->p_flag & P_HADTHREADS) {
271 * Kill off the other threads. This requires
272 * some co-operation from other parts of the kernel
273 * so it may not be instantaneous. With this state set
274 * any thread attempting to interruptibly
275 * sleep will return immediately with EINTR or EWOULDBLOCK
276 * which will hopefully force them to back out to userland
277 * freeing resources as they go. Any thread attempting
278 * to return to userland will thread_exit() from ast().
279 * thread_exit() will unsuspend us when the last of the
280 * other threads exits.
281 * If there is already a thread singler after resumption,
282 * calling thread_single() will fail; in that case, we just
283 * re-check all suspension request, the thread should
284 * either be suspended there or exit.
286 if (!thread_single(p, SINGLE_EXIT))
288 * All other activity in this process is now
289 * stopped. Threading support has been turned
294 * Recheck for new stop or suspend requests which
295 * might appear while process lock was dropped in
298 thread_suspend_check(0);
300 KASSERT(p->p_numthreads == 1,
301 ("exit1: proc %p exiting with %d threads", p, p->p_numthreads));
302 racct_sub(p, RACCT_NTHR, 1);
304 /* Let event handler change exit status */
309 * Ignore any pending request to stop due to a stop signal.
310 * Once P_WEXIT is set, future requests will be ignored as
313 p->p_flag &= ~P_STOPPED_SIG;
314 KASSERT(!P_SHOULDSTOP(p), ("exiting process is stopped"));
316 /* Note that we are exiting. */
317 p->p_flag |= P_WEXIT;
320 * Wait for any processes that have a hold on our vmspace to
321 * release their reference.
323 while (p->p_lock > 0)
324 msleep(&p->p_lock, &p->p_mtx, PWAIT, "exithold", 0);
327 /* Drain the limit callout while we don't have the proc locked */
328 callout_drain(&p->p_limco);
332 * The Sun BSM exit token contains two components: an exit status as
333 * passed to exit(), and a return value to indicate what sort of exit
334 * it was. The exit status is WEXITSTATUS(rv), but it's not clear
335 * what the return value is.
337 AUDIT_ARG_EXIT(rval, 0);
338 AUDIT_SYSCALL_EXIT(0, td);
341 /* Are we a task leader with peers? */
342 if (p->p_peers != NULL && p == p->p_leader) {
343 mtx_lock(&ppeers_lock);
347 kern_psignal(q, SIGKILL);
351 while (p->p_peers != NULL)
352 msleep(p, &ppeers_lock, PWAIT, "exit1", 0);
353 mtx_unlock(&ppeers_lock);
359 * Check if any loadable modules need anything done at process exit.
360 * E.g. SYSV IPC stuff.
361 * Event handler could change exit status.
362 * XXX what if one of these generates an error?
364 EVENTHANDLER_DIRECT_INVOKE(process_exit, p);
367 * If parent is waiting for us to exit or exec,
368 * P_PPWAIT is set; we will wakeup the parent below.
375 * Stop the real interval timer. If the handler is currently
376 * executing, prevent it from rearming itself and let it finish.
378 if (timevalisset(&p->p_realtimer.it_value) &&
379 callout_stop(&p->p_itcallout) == 0) {
380 timevalclear(&p->p_realtimer.it_interval);
382 callout_drain(&p->p_itcallout);
387 if (p->p_sysent->sv_onexit != NULL)
388 p->p_sysent->sv_onexit(p);
392 * Reset any sigio structures pointing to us as a result of
393 * F_SETOWN with our pid. The P_WEXIT flag interlocks with fsetown().
395 funsetownlst(&p->p_sigiolst);
398 * Close open files and release open-file table.
405 * If this thread tickled GEOM, we need to wait for the giggling to
406 * stop before we return to userland
408 if (td->td_pflags & TDP_GEOM)
412 * Remove ourself from our leader's peer list and wake our leader.
414 if (p->p_leader->p_peers != NULL) {
415 mtx_lock(&ppeers_lock);
416 if (p->p_leader->p_peers != NULL) {
418 while (q->p_peers != p)
420 q->p_peers = p->p_peers;
423 mtx_unlock(&ppeers_lock);
426 exec_free_abi_mappings(p);
428 (void)acct_process(td);
434 * Release reference to text vnode etc
436 if (p->p_textvp != NULL) {
440 if (p->p_textdvp != NULL) {
444 if (p->p_binname != NULL) {
445 free(p->p_binname, M_PARGS);
450 * Release our limits structure.
452 lim_free(p->p_limit);
458 * Call machine-dependent code to release any
459 * machine-dependent resources other than the address space.
460 * The address space is released by "vmspace_exitfree(p)" in
465 WITNESS_WARN(WARN_PANIC, NULL, "process (pid %d) exiting", p->p_pid);
468 * Remove from allproc. It still sits in the hash.
470 sx_xlock(&allproc_lock);
471 LIST_REMOVE(p, p_list);
475 * Used by ddb's 'ps' command to find this process via the
478 p->p_list.le_prev = NULL;
480 sx_xunlock(&allproc_lock);
482 sx_xlock(&proctree_lock);
484 p->p_flag &= ~(P_TRACED | P_PPWAIT | P_PPTRACE);
488 * killjobc() might drop and re-acquire proctree_lock to
489 * revoke control tty if exiting process was a session leader.
494 * Reparent all children processes:
495 * - traced ones to the original parent (or init if we are that parent)
498 q = LIST_FIRST(&p->p_children);
499 if (q != NULL) /* only need this if any child is S_ZOMB */
501 for (; q != NULL; q = nq) {
502 nq = LIST_NEXT(q, p_sibling);
503 ksi = ksiginfo_alloc(M_WAITOK);
505 q->p_sigparent = SIGCHLD;
507 if ((q->p_flag & P_TRACED) == 0) {
508 proc_reparent(q, q->p_reaper, true);
509 if (q->p_state == PRS_ZOMBIE) {
511 * Inform reaper about the reparented
512 * zombie, since wait(2) has something
513 * new to report. Guarantee queueing
514 * of the SIGCHLD signal, similar to
515 * the _exit() behaviour, by providing
516 * our ksiginfo. Ksi is freed by the
519 if (q->p_ksi == NULL) {
522 ksiginfo_copy(q->p_ksi, ksi);
523 ksi->ksi_flags |= KSI_INS;
527 PROC_LOCK(q->p_reaper);
528 pksignal(q->p_reaper, SIGCHLD, ksi1);
529 PROC_UNLOCK(q->p_reaper);
530 } else if (q->p_pdeathsig > 0) {
532 * The child asked to received a signal
535 kern_psignal(q, q->p_pdeathsig);
539 * Traced processes are killed by default
540 * since their existence means someone is
543 t = proc_realparent(q);
545 proc_reparent(q, q->p_reaper, true);
548 proc_reparent(q, t, true);
552 * Since q was found on our children list, the
553 * proc_reparent() call moved q to the orphan
554 * list due to present P_TRACED flag. Clear
555 * orphan link for q now while q is locked.
557 proc_clear_orphan(q);
558 q->p_flag &= ~P_TRACED;
559 q->p_flag2 &= ~P2_PTRACE_FSTP;
562 FOREACH_THREAD_IN_PROC(q, tdt) {
563 tdt->td_dbgflags &= ~(TDB_SUSPEND | TDB_XSIG |
567 if (kern_kill_on_dbg_exit) {
568 q->p_flag &= ~P_STOPPED_TRACE;
569 kern_psignal(q, SIGKILL);
570 } else if ((q->p_flag & (P_STOPPED_TRACE |
571 P_STOPPED_SIG)) != 0) {
572 sigqueue_delete_proc(q, SIGTRAP);
582 * Also get rid of our orphans.
584 while ((q = LIST_FIRST(&p->p_orphans)) != NULL) {
586 KASSERT(q->p_oppid == p->p_pid,
587 ("orphan %p of %p has unexpected oppid %d", q, p,
589 q->p_oppid = q->p_reaper->p_pid;
592 * If we are the real parent of this process
593 * but it has been reparented to a debugger, then
594 * check if it asked for a signal when we exit.
596 if (q->p_pdeathsig > 0)
597 kern_psignal(q, q->p_pdeathsig);
598 CTR2(KTR_PTRACE, "exit: pid %d, clearing orphan %d", p->p_pid,
600 proc_clear_orphan(q);
605 if (SDT_PROBES_ENABLED()) {
606 int reason = CLD_EXITED;
607 if (WCOREDUMP(signo))
609 else if (WIFSIGNALED(signo))
611 SDT_PROBE1(proc, , , exit, reason);
615 /* Save exit status. */
619 if (p->p_sysent->sv_ontdexit != NULL)
620 p->p_sysent->sv_ontdexit(td);
624 * Tell the DTrace fasttrap provider about the exit if it
625 * has declared an interest.
627 if (dtrace_fasttrap_exit)
628 dtrace_fasttrap_exit(p);
632 * Notify interested parties of our demise.
634 KNOTE_LOCKED(p->p_klist, NOTE_EXIT);
637 * If this is a process with a descriptor, we may not need to deliver
638 * a signal to the parent. proctree_lock is held over
639 * procdesc_exit() to serialize concurrent calls to close() and
643 if (p->p_procdesc == NULL || procdesc_exit(p)) {
645 * Notify parent that we're gone. If parent has the
646 * PS_NOCLDWAIT flag set, or if the handler is set to SIG_IGN,
647 * notify process 1 instead (and hope it will handle this
650 PROC_LOCK(p->p_pptr);
651 mtx_lock(&p->p_pptr->p_sigacts->ps_mtx);
652 if (p->p_pptr->p_sigacts->ps_flag &
653 (PS_NOCLDWAIT | PS_CLDSIGIGN)) {
656 mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx);
659 proc_reparent(p, p->p_reaper, true);
660 p->p_sigparent = SIGCHLD;
661 PROC_LOCK(p->p_pptr);
664 * Notify parent, so in case he was wait(2)ing or
665 * executing waitpid(2) with our pid, he will
670 mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx);
672 if (p->p_pptr == p->p_reaper || p->p_pptr == initproc) {
674 } else if (p->p_sigparent != 0) {
675 if (p->p_sigparent == SIGCHLD) {
677 } else { /* LINUX thread */
682 PROC_LOCK(p->p_pptr);
683 sx_xunlock(&proctree_lock);
685 if (signal_parent == 1) {
687 } else if (signal_parent == 2) {
688 kern_psignal(p->p_pptr, p->p_sigparent);
691 /* Tell the prison that we are gone. */
692 prison_proc_free(p->p_ucred->cr_prison);
695 * The state PRS_ZOMBIE prevents other proesses from sending
696 * signal to the process, to avoid memory leak, we free memory
697 * for signal queue at the time when the state is set.
699 sigqueue_flush(&p->p_sigqueue);
700 sigqueue_flush(&td->td_sigqueue);
703 * We have to wait until after acquiring all locks before
704 * changing p_state. We need to avoid all possible context
705 * switches (including ones from blocking on a mutex) while
706 * marked as a zombie. We also have to set the zombie state
707 * before we release the parent process' proc lock to avoid
708 * a lost wakeup. So, we first call wakeup, then we grab the
709 * sched lock, update the state, and release the parent process'
713 cv_broadcast(&p->p_pwait);
714 sched_exit(p->p_pptr, td);
716 p->p_state = PRS_ZOMBIE;
717 PROC_UNLOCK(p->p_pptr);
720 * Save our children's rusage information in our exit rusage.
723 ruadd(&p->p_ru, &p->p_rux, &p->p_stats->p_cru, &p->p_crux);
727 * Make sure the scheduler takes this thread out of its tables etc.
728 * This will also release this thread's reference to the ucred.
729 * Other thread parts to release include pcb bits and such.
734 #ifndef _SYS_SYSPROTO_H_
743 sys_abort2(struct thread *td, struct abort2_args *uap)
745 struct proc *p = td->td_proc;
751 * Do it right now so we can log either proper call of abort2(), or
752 * note, that invalid argument was passed. 512 is big enough to
753 * handle 16 arguments' descriptions with additional comments.
755 sb = sbuf_new(NULL, NULL, 512, SBUF_FIXEDLEN);
757 sbuf_printf(sb, "%s(pid %d uid %d) aborted: ",
758 p->p_comm, p->p_pid, td->td_ucred->cr_uid);
760 * Since we can't return from abort2(), send SIGKILL in cases, where
761 * abort2() was called improperly
764 /* Prevent from DoSes from user-space. */
765 if (uap->nargs < 0 || uap->nargs > 16)
767 if (uap->nargs > 0) {
768 if (uap->args == NULL)
770 error = copyin(uap->args, uargs, uap->nargs * sizeof(void *));
775 * Limit size of 'reason' string to 128. Will fit even when
776 * maximal number of arguments was chosen to be logged.
778 if (uap->why != NULL) {
779 error = sbuf_copyin(sb, uap->why, 128);
783 sbuf_printf(sb, "(null)");
785 if (uap->nargs > 0) {
786 sbuf_printf(sb, "(");
787 for (i = 0;i < uap->nargs; i++)
788 sbuf_printf(sb, "%s%p", i == 0 ? "" : ", ", uargs[i]);
789 sbuf_printf(sb, ")");
792 * Final stage: arguments were proper, string has been
793 * successfully copied from userspace, and copying pointers
794 * from user-space succeed.
798 if (sig == SIGKILL) {
800 sbuf_printf(sb, " (Reason text inaccessible)");
804 log(LOG_INFO, "%s", sbuf_data(sb));
812 * The dirty work is handled by kern_wait().
815 owait(struct thread *td, struct owait_args *uap __unused)
819 error = kern_wait(td, WAIT_ANY, &status, 0, NULL);
821 td->td_retval[1] = status;
824 #endif /* COMPAT_43 */
827 * The dirty work is handled by kern_wait().
830 sys_wait4(struct thread *td, struct wait4_args *uap)
832 struct rusage ru, *rup;
835 if (uap->rusage != NULL)
839 error = kern_wait(td, uap->pid, &status, uap->options, rup);
840 if (uap->status != NULL && error == 0 && td->td_retval[0] != 0)
841 error = copyout(&status, uap->status, sizeof(status));
842 if (uap->rusage != NULL && error == 0 && td->td_retval[0] != 0)
843 error = copyout(&ru, uap->rusage, sizeof(struct rusage));
848 sys_wait6(struct thread *td, struct wait6_args *uap)
850 struct __wrusage wru, *wrup;
856 idtype = uap->idtype;
859 if (uap->wrusage != NULL)
864 if (uap->info != NULL) {
866 bzero(sip, sizeof(*sip));
871 * We expect all callers of wait6() to know about WEXITED and
874 error = kern_wait6(td, idtype, id, &status, uap->options, wrup, sip);
876 if (uap->status != NULL && error == 0 && td->td_retval[0] != 0)
877 error = copyout(&status, uap->status, sizeof(status));
878 if (uap->wrusage != NULL && error == 0 && td->td_retval[0] != 0)
879 error = copyout(&wru, uap->wrusage, sizeof(wru));
880 if (uap->info != NULL && error == 0)
881 error = copyout(&si, uap->info, sizeof(si));
886 * Reap the remains of a zombie process and optionally return status and
887 * rusage. Asserts and will release both the proctree_lock and the process
888 * lock as part of its work.
891 proc_reap(struct thread *td, struct proc *p, int *status, int options)
895 sx_assert(&proctree_lock, SA_XLOCKED);
896 PROC_LOCK_ASSERT(p, MA_OWNED);
897 KASSERT(p->p_state == PRS_ZOMBIE, ("proc_reap: !PRS_ZOMBIE"));
899 mtx_spin_wait_unlocked(&p->p_slock);
904 *status = KW_EXITCODE(p->p_xexit, p->p_xsig);
905 if (options & WNOWAIT) {
907 * Only poll, returning the status. Caller does not wish to
908 * release the proc struct just yet.
911 sx_xunlock(&proctree_lock);
916 sigqueue_take(p->p_ksi);
920 * If we got the child via a ptrace 'attach', we need to give it back
923 if (p->p_oppid != p->p_pptr->p_pid) {
925 t = proc_realparent(p);
929 "wait: traced child %d moved back to parent %d", p->p_pid,
931 proc_reparent(p, t, false);
933 pksignal(t, SIGCHLD, p->p_ksi);
935 cv_broadcast(&p->p_pwait);
937 sx_xunlock(&proctree_lock);
943 * Remove other references to this process to ensure we have an
944 * exclusive reference.
946 sx_xlock(PIDHASHLOCK(p->p_pid));
947 LIST_REMOVE(p, p_hash);
948 sx_xunlock(PIDHASHLOCK(p->p_pid));
949 LIST_REMOVE(p, p_sibling);
950 reaper_abandon_children(p, true);
953 proc_clear_orphan(p);
956 if (p->p_procdesc != NULL)
958 sx_xunlock(&proctree_lock);
960 proc_id_clear(PROC_ID_PID, p->p_pid);
963 knlist_detach(p->p_klist);
968 * Removal from allproc list and process group list paired with
969 * PROC_LOCK which was executed during that time should guarantee
970 * nothing can reach this process anymore. As such further locking
973 p->p_xexit = p->p_xsig = 0; /* XXX: why? */
976 ruadd(&q->p_stats->p_cru, &q->p_crux, &p->p_ru, &p->p_rux);
980 * Decrement the count of procs running with this uid.
982 (void)chgproccnt(p->p_ucred->cr_ruidinfo, -1, 0);
985 * Destroy resource accounting information associated with the process.
990 racct_sub(p, RACCT_NPROC, 1);
997 * Free credentials, arguments, and sigacts.
1000 pargs_drop(p->p_args);
1002 sigacts_free(p->p_sigacts);
1003 p->p_sigacts = NULL;
1006 * Do any thread-system specific cleanups.
1011 * Give vm and machine-dependent layer a chance to free anything that
1012 * cpu_exit couldn't release while still running in process context.
1016 mac_proc_destroy(p);
1019 KASSERT(FIRST_THREAD_IN_PROC(p),
1020 ("proc_reap: no residual thread!"));
1021 uma_zfree(proc_zone, p);
1022 atomic_add_int(&nprocs, -1);
1026 proc_to_reap(struct thread *td, struct proc *p, idtype_t idtype, id_t id,
1027 int *status, int options, struct __wrusage *wrusage, siginfo_t *siginfo,
1032 sx_assert(&proctree_lock, SA_XLOCKED);
1038 if (p->p_procdesc == NULL ||
1039 (p->p_pptr == td->td_proc &&
1040 (p->p_flag & P_TRACED) != 0)) {
1047 if (p->p_pid != (pid_t)id) {
1053 if (p->p_pgid != (pid_t)id) {
1059 if (p->p_session->s_sid != (pid_t)id) {
1065 if (p->p_ucred->cr_uid != (uid_t)id) {
1071 if (p->p_ucred->cr_gid != (gid_t)id) {
1077 if (p->p_ucred->cr_prison->pr_id != (int)id) {
1083 * It seems that the thread structures get zeroed out
1084 * at process exit. This makes it impossible to
1085 * support P_SETID, P_CID or P_CPUID.
1092 if (p_canwait(td, p)) {
1097 if (((options & WEXITED) == 0) && (p->p_state == PRS_ZOMBIE)) {
1103 * This special case handles a kthread spawned by linux_clone
1104 * (see linux_misc.c). The linux_wait4 and linux_waitpid
1105 * functions need to be able to distinguish between waiting
1106 * on a process and waiting on a thread. It is a thread if
1107 * p_sigparent is not SIGCHLD, and the WLINUXCLONE option
1108 * signifies we want to wait for threads and not processes.
1110 if ((p->p_sigparent != SIGCHLD) ^
1111 ((options & WLINUXCLONE) != 0)) {
1116 if (siginfo != NULL) {
1117 bzero(siginfo, sizeof(*siginfo));
1118 siginfo->si_errno = 0;
1121 * SUSv4 requires that the si_signo value is always
1122 * SIGCHLD. Obey it despite the rfork(2) interface
1123 * allows to request other signal for child exit
1126 siginfo->si_signo = SIGCHLD;
1129 * This is still a rough estimate. We will fix the
1130 * cases TRAPPED, STOPPED, and CONTINUED later.
1132 if (WCOREDUMP(p->p_xsig)) {
1133 siginfo->si_code = CLD_DUMPED;
1134 siginfo->si_status = WTERMSIG(p->p_xsig);
1135 } else if (WIFSIGNALED(p->p_xsig)) {
1136 siginfo->si_code = CLD_KILLED;
1137 siginfo->si_status = WTERMSIG(p->p_xsig);
1139 siginfo->si_code = CLD_EXITED;
1140 siginfo->si_status = p->p_xexit;
1143 siginfo->si_pid = p->p_pid;
1144 siginfo->si_uid = p->p_ucred->cr_uid;
1147 * The si_addr field would be useful additional
1148 * detail, but apparently the PC value may be lost
1149 * when we reach this point. bzero() above sets
1150 * siginfo->si_addr to NULL.
1155 * There should be no reason to limit resources usage info to
1156 * exited processes only. A snapshot about any resources used
1157 * by a stopped process may be exactly what is needed.
1159 if (wrusage != NULL) {
1160 rup = &wrusage->wru_self;
1163 calcru(p, &rup->ru_utime, &rup->ru_stime);
1166 rup = &wrusage->wru_children;
1167 *rup = p->p_stats->p_cru;
1168 calccru(p, &rup->ru_utime, &rup->ru_stime);
1171 if (p->p_state == PRS_ZOMBIE && !check_only) {
1172 proc_reap(td, p, status, options);
1179 kern_wait(struct thread *td, pid_t pid, int *status, int options,
1180 struct rusage *rusage)
1182 struct __wrusage wru, *wrup;
1188 * Translate the special pid values into the (idtype, pid)
1189 * pair for kern_wait6. The WAIT_MYPGRP case is handled by
1190 * kern_wait6() on its own.
1192 if (pid == WAIT_ANY) {
1195 } else if (pid < 0) {
1209 * For backward compatibility we implicitly add flags WEXITED
1210 * and WTRAPPED here.
1212 options |= WEXITED | WTRAPPED;
1213 ret = kern_wait6(td, idtype, id, status, options, wrup, NULL);
1215 *rusage = wru.wru_self;
1220 report_alive_proc(struct thread *td, struct proc *p, siginfo_t *siginfo,
1221 int *status, int options, int si_code)
1225 PROC_LOCK_ASSERT(p, MA_OWNED);
1226 sx_assert(&proctree_lock, SA_XLOCKED);
1227 MPASS(si_code == CLD_TRAPPED || si_code == CLD_STOPPED ||
1228 si_code == CLD_CONTINUED);
1230 cont = si_code == CLD_CONTINUED;
1231 if ((options & WNOWAIT) == 0) {
1233 p->p_flag &= ~P_CONTINUED;
1235 p->p_flag |= P_WAITED;
1236 if (kern_wait_dequeue_sigchld &&
1237 (td->td_proc->p_sysent->sv_flags & SV_SIG_WAITNDQ) == 0) {
1238 PROC_LOCK(td->td_proc);
1239 sigqueue_take(p->p_ksi);
1240 PROC_UNLOCK(td->td_proc);
1243 sx_xunlock(&proctree_lock);
1244 if (siginfo != NULL) {
1245 siginfo->si_code = si_code;
1246 siginfo->si_status = cont ? SIGCONT : p->p_xsig;
1249 *status = cont ? SIGCONT : W_STOPCODE(p->p_xsig);
1251 td->td_retval[0] = p->p_pid;
1255 kern_wait6(struct thread *td, idtype_t idtype, id_t id, int *status,
1256 int options, struct __wrusage *wrusage, siginfo_t *siginfo)
1260 int error, nfound, ret;
1263 AUDIT_ARG_VALUE((int)idtype); /* XXX - This is likely wrong! */
1264 AUDIT_ARG_PID((pid_t)id); /* XXX - This may be wrong! */
1265 AUDIT_ARG_VALUE(options);
1269 if ((pid_t)id == WAIT_MYPGRP && (idtype == P_PID || idtype == P_PGID)) {
1271 id = (id_t)q->p_pgid;
1276 /* If we don't know the option, just return. */
1277 if ((options & ~(WUNTRACED | WNOHANG | WCONTINUED | WNOWAIT |
1278 WEXITED | WTRAPPED | WLINUXCLONE)) != 0)
1280 if ((options & (WEXITED | WUNTRACED | WCONTINUED | WTRAPPED)) == 0) {
1282 * We will be unable to find any matching processes,
1283 * because there are no known events to look for.
1284 * Prefer to return error instead of blocking
1291 if (q->p_flag & P_STATCHILD) {
1293 q->p_flag &= ~P_STATCHILD;
1296 sx_xlock(&proctree_lock);
1299 LIST_FOREACH(p, &q->p_children, p_sibling) {
1301 ret = proc_to_reap(td, p, idtype, id, status, options,
1302 wrusage, siginfo, 0);
1305 else if (ret != 1) {
1306 td->td_retval[0] = pid;
1311 PROC_LOCK_ASSERT(p, MA_OWNED);
1313 if ((options & WTRAPPED) != 0 &&
1314 (p->p_flag & P_TRACED) != 0) {
1317 ((p->p_flag & (P_STOPPED_TRACE | P_STOPPED_SIG)) &&
1318 p->p_suspcount == p->p_numthreads &&
1319 (p->p_flag & P_WAITED) == 0);
1323 "wait: returning trapped pid %d status %#x "
1324 "(xstat %d) xthread %d",
1325 p->p_pid, W_STOPCODE(p->p_xsig), p->p_xsig,
1326 p->p_xthread != NULL ?
1327 p->p_xthread->td_tid : -1);
1328 report_alive_proc(td, p, siginfo, status,
1329 options, CLD_TRAPPED);
1333 if ((options & WUNTRACED) != 0 &&
1334 (p->p_flag & P_STOPPED_SIG) != 0) {
1336 report = (p->p_suspcount == p->p_numthreads &&
1337 ((p->p_flag & P_WAITED) == 0));
1340 report_alive_proc(td, p, siginfo, status,
1341 options, CLD_STOPPED);
1345 if ((options & WCONTINUED) != 0 &&
1346 (p->p_flag & P_CONTINUED) != 0) {
1347 report_alive_proc(td, p, siginfo, status, options,
1355 * Look in the orphans list too, to allow the parent to
1356 * collect it's child exit status even if child is being
1359 * Debugger detaches from the parent upon successful
1360 * switch-over from parent to child. At this point due to
1361 * re-parenting the parent loses the child to debugger and a
1362 * wait4(2) call would report that it has no children to wait
1363 * for. By maintaining a list of orphans we allow the parent
1364 * to successfully wait until the child becomes a zombie.
1367 LIST_FOREACH(p, &q->p_orphans, p_orphan) {
1368 ret = proc_to_reap(td, p, idtype, id, NULL, options,
1371 KASSERT(ret != -1, ("reaped an orphan (pid %d)",
1372 (int)td->td_retval[0]));
1380 sx_xunlock(&proctree_lock);
1383 if (options & WNOHANG) {
1384 sx_xunlock(&proctree_lock);
1385 td->td_retval[0] = 0;
1389 if (q->p_flag & P_STATCHILD) {
1390 q->p_flag &= ~P_STATCHILD;
1394 sx_xunlock(&proctree_lock);
1395 error = msleep(q, &q->p_mtx, PWAIT | PCATCH | PDROP, "wait", 0);
1402 proc_add_orphan(struct proc *child, struct proc *parent)
1405 sx_assert(&proctree_lock, SX_XLOCKED);
1406 KASSERT((child->p_flag & P_TRACED) != 0,
1407 ("proc_add_orphan: not traced"));
1409 if (LIST_EMPTY(&parent->p_orphans)) {
1410 child->p_treeflag |= P_TREE_FIRST_ORPHAN;
1411 LIST_INSERT_HEAD(&parent->p_orphans, child, p_orphan);
1413 LIST_INSERT_AFTER(LIST_FIRST(&parent->p_orphans),
1416 child->p_treeflag |= P_TREE_ORPHANED;
1420 * Make process 'parent' the new parent of process 'child'.
1421 * Must be called with an exclusive hold of proctree lock.
1424 proc_reparent(struct proc *child, struct proc *parent, bool set_oppid)
1427 sx_assert(&proctree_lock, SX_XLOCKED);
1428 PROC_LOCK_ASSERT(child, MA_OWNED);
1429 if (child->p_pptr == parent)
1432 PROC_LOCK(child->p_pptr);
1433 sigqueue_take(child->p_ksi);
1434 PROC_UNLOCK(child->p_pptr);
1435 LIST_REMOVE(child, p_sibling);
1436 LIST_INSERT_HEAD(&parent->p_children, child, p_sibling);
1438 proc_clear_orphan(child);
1439 if ((child->p_flag & P_TRACED) != 0) {
1440 proc_add_orphan(child, child->p_pptr);
1443 child->p_pptr = parent;
1445 child->p_oppid = parent->p_pid;