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>
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_LOCKED);
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 * Exit: deallocate address space and other resources, change proc state to
218 * zombie, and unlink proc from allproc and parent's lists. Save exit status
219 * and rusage for wait(). Check for child processes and orphan them.
222 exit1(struct thread *td, int rval, int signo)
224 struct proc *p, *nq, *q, *t;
226 ksiginfo_t *ksi, *ksi1;
229 mtx_assert(&Giant, MA_NOTOWNED);
230 KASSERT(rval == 0 || signo == 0, ("exit1 rv %d sig %d", rval, signo));
234 * XXX in case we're rebooting we just let init die in order to
235 * work around an unsolved stack overflow seen very late during
236 * shutdown on sparc64 when the gmirror worker process exists.
237 * XXX what to do now that sparc64 is gone... remove if?
239 if (p == initproc && rebooting == 0) {
240 printf("init died (signal %d, exit %d)\n", signo, rval);
241 panic("Going nowhere without my init!");
245 * Deref SU mp, since the thread does not return to userspace.
247 td_softdep_cleanup(td);
250 * MUST abort all other threads before proceeding past here.
254 * First check if some other thread or external request got
255 * here before us. If so, act appropriately: exit or suspend.
256 * We must ensure that stop requests are handled before we set
259 thread_suspend_check(0);
260 while (p->p_flag & P_HADTHREADS) {
262 * Kill off the other threads. This requires
263 * some co-operation from other parts of the kernel
264 * so it may not be instantaneous. With this state set
265 * any thread entering the kernel from userspace will
266 * thread_exit() in trap(). Any thread attempting to
267 * sleep will return immediately with EINTR or EWOULDBLOCK
268 * which will hopefully force them to back out to userland
269 * freeing resources as they go. Any thread attempting
270 * to return to userland will thread_exit() from userret().
271 * thread_exit() will unsuspend us when the last of the
272 * other threads exits.
273 * If there is already a thread singler after resumption,
274 * calling thread_single will fail; in that case, we just
275 * re-check all suspension request, the thread should
276 * either be suspended there or exit.
278 if (!thread_single(p, SINGLE_EXIT))
280 * All other activity in this process is now
281 * stopped. Threading support has been turned
286 * Recheck for new stop or suspend requests which
287 * might appear while process lock was dropped in
290 thread_suspend_check(0);
292 KASSERT(p->p_numthreads == 1,
293 ("exit1: proc %p exiting with %d threads", p, p->p_numthreads));
294 racct_sub(p, RACCT_NTHR, 1);
296 /* Let event handler change exit status */
301 * Ignore any pending request to stop due to a stop signal.
302 * Once P_WEXIT is set, future requests will be ignored as
305 p->p_flag &= ~P_STOPPED_SIG;
306 KASSERT(!P_SHOULDSTOP(p), ("exiting process is stopped"));
308 /* Note that we are exiting. */
309 p->p_flag |= P_WEXIT;
312 * Wait for any processes that have a hold on our vmspace to
313 * release their reference.
315 while (p->p_lock > 0)
316 msleep(&p->p_lock, &p->p_mtx, PWAIT, "exithold", 0);
319 /* Drain the limit callout while we don't have the proc locked */
320 callout_drain(&p->p_limco);
324 * The Sun BSM exit token contains two components: an exit status as
325 * passed to exit(), and a return value to indicate what sort of exit
326 * it was. The exit status is WEXITSTATUS(rv), but it's not clear
327 * what the return value is.
329 AUDIT_ARG_EXIT(rval, 0);
330 AUDIT_SYSCALL_EXIT(0, td);
333 /* Are we a task leader with peers? */
334 if (p->p_peers != NULL && p == p->p_leader) {
335 mtx_lock(&ppeers_lock);
339 kern_psignal(q, SIGKILL);
343 while (p->p_peers != NULL)
344 msleep(p, &ppeers_lock, PWAIT, "exit1", 0);
345 mtx_unlock(&ppeers_lock);
351 * Check if any loadable modules need anything done at process exit.
352 * E.g. SYSV IPC stuff.
353 * Event handler could change exit status.
354 * XXX what if one of these generates an error?
356 EVENTHANDLER_DIRECT_INVOKE(process_exit, p);
359 * If parent is waiting for us to exit or exec,
360 * P_PPWAIT is set; we will wakeup the parent below.
367 * Stop the real interval timer. If the handler is currently
368 * executing, prevent it from rearming itself and let it finish.
370 if (timevalisset(&p->p_realtimer.it_value) &&
371 _callout_stop_safe(&p->p_itcallout, CS_EXECUTING, NULL) == 0) {
372 timevalclear(&p->p_realtimer.it_interval);
373 msleep(&p->p_itcallout, &p->p_mtx, PWAIT, "ritwait", 0);
374 KASSERT(!timevalisset(&p->p_realtimer.it_value),
375 ("realtime timer is still armed"));
380 if (p->p_sysent->sv_onexit != NULL)
381 p->p_sysent->sv_onexit(p);
385 * Reset any sigio structures pointing to us as a result of
386 * F_SETOWN with our pid. The P_WEXIT flag interlocks with fsetown().
388 funsetownlst(&p->p_sigiolst);
391 * Close open files and release open-file table.
398 * If this thread tickled GEOM, we need to wait for the giggling to
399 * stop before we return to userland
401 if (td->td_pflags & TDP_GEOM)
405 * Remove ourself from our leader's peer list and wake our leader.
407 if (p->p_leader->p_peers != NULL) {
408 mtx_lock(&ppeers_lock);
409 if (p->p_leader->p_peers != NULL) {
411 while (q->p_peers != p)
413 q->p_peers = p->p_peers;
416 mtx_unlock(&ppeers_lock);
419 exec_free_abi_mappings(p);
421 (void)acct_process(td);
427 * Release reference to text vnode etc
429 if (p->p_textvp != NULL) {
433 if (p->p_textdvp != NULL) {
437 if (p->p_binname != NULL) {
438 free(p->p_binname, M_PARGS);
443 * Release our limits structure.
445 lim_free(p->p_limit);
451 * Call machine-dependent code to release any
452 * machine-dependent resources other than the address space.
453 * The address space is released by "vmspace_exitfree(p)" in
458 WITNESS_WARN(WARN_PANIC, NULL, "process (pid %d) exiting", p->p_pid);
461 * Remove from allproc. It still sits in the hash.
463 sx_xlock(&allproc_lock);
464 LIST_REMOVE(p, p_list);
468 * Used by ddb's 'ps' command to find this process via the
471 p->p_list.le_prev = NULL;
473 sx_xunlock(&allproc_lock);
475 sx_xlock(&proctree_lock);
477 p->p_flag &= ~(P_TRACED | P_PPWAIT | P_PPTRACE);
481 * killjobc() might drop and re-acquire proctree_lock to
482 * revoke control tty if exiting process was a session leader.
487 * Reparent all children processes:
488 * - traced ones to the original parent (or init if we are that parent)
491 q = LIST_FIRST(&p->p_children);
492 if (q != NULL) /* only need this if any child is S_ZOMB */
494 for (; q != NULL; q = nq) {
495 nq = LIST_NEXT(q, p_sibling);
496 ksi = ksiginfo_alloc(TRUE);
498 q->p_sigparent = SIGCHLD;
500 if ((q->p_flag & P_TRACED) == 0) {
501 proc_reparent(q, q->p_reaper, true);
502 if (q->p_state == PRS_ZOMBIE) {
504 * Inform reaper about the reparented
505 * zombie, since wait(2) has something
506 * new to report. Guarantee queueing
507 * of the SIGCHLD signal, similar to
508 * the _exit() behaviour, by providing
509 * our ksiginfo. Ksi is freed by the
512 if (q->p_ksi == NULL) {
515 ksiginfo_copy(q->p_ksi, ksi);
516 ksi->ksi_flags |= KSI_INS;
520 PROC_LOCK(q->p_reaper);
521 pksignal(q->p_reaper, SIGCHLD, ksi1);
522 PROC_UNLOCK(q->p_reaper);
523 } else if (q->p_pdeathsig > 0) {
525 * The child asked to received a signal
528 kern_psignal(q, q->p_pdeathsig);
532 * Traced processes are killed by default
533 * since their existence means someone is
536 t = proc_realparent(q);
538 proc_reparent(q, q->p_reaper, true);
541 proc_reparent(q, t, true);
545 * Since q was found on our children list, the
546 * proc_reparent() call moved q to the orphan
547 * list due to present P_TRACED flag. Clear
548 * orphan link for q now while q is locked.
550 proc_clear_orphan(q);
551 q->p_flag &= ~P_TRACED;
552 q->p_flag2 &= ~P2_PTRACE_FSTP;
555 FOREACH_THREAD_IN_PROC(q, tdt) {
556 tdt->td_dbgflags &= ~(TDB_SUSPEND | TDB_XSIG |
560 if (kern_kill_on_dbg_exit) {
561 q->p_flag &= ~P_STOPPED_TRACE;
562 kern_psignal(q, SIGKILL);
563 } else if ((q->p_flag & (P_STOPPED_TRACE |
564 P_STOPPED_SIG)) != 0) {
565 sigqueue_delete_proc(q, SIGTRAP);
575 * Also get rid of our orphans.
577 while ((q = LIST_FIRST(&p->p_orphans)) != NULL) {
579 KASSERT(q->p_oppid == p->p_pid,
580 ("orphan %p of %p has unexpected oppid %d", q, p,
582 q->p_oppid = q->p_reaper->p_pid;
585 * If we are the real parent of this process
586 * but it has been reparented to a debugger, then
587 * check if it asked for a signal when we exit.
589 if (q->p_pdeathsig > 0)
590 kern_psignal(q, q->p_pdeathsig);
591 CTR2(KTR_PTRACE, "exit: pid %d, clearing orphan %d", p->p_pid,
593 proc_clear_orphan(q);
598 if (SDT_PROBES_ENABLED()) {
599 int reason = CLD_EXITED;
600 if (WCOREDUMP(signo))
602 else if (WIFSIGNALED(signo))
604 SDT_PROBE1(proc, , , exit, reason);
608 /* Save exit status. */
612 if (p->p_sysent->sv_ontdexit != NULL)
613 p->p_sysent->sv_ontdexit(td);
617 * Tell the DTrace fasttrap provider about the exit if it
618 * has declared an interest.
620 if (dtrace_fasttrap_exit)
621 dtrace_fasttrap_exit(p);
625 * Notify interested parties of our demise.
627 KNOTE_LOCKED(p->p_klist, NOTE_EXIT);
630 * If this is a process with a descriptor, we may not need to deliver
631 * a signal to the parent. proctree_lock is held over
632 * procdesc_exit() to serialize concurrent calls to close() and
636 if (p->p_procdesc == NULL || procdesc_exit(p)) {
638 * Notify parent that we're gone. If parent has the
639 * PS_NOCLDWAIT flag set, or if the handler is set to SIG_IGN,
640 * notify process 1 instead (and hope it will handle this
643 PROC_LOCK(p->p_pptr);
644 mtx_lock(&p->p_pptr->p_sigacts->ps_mtx);
645 if (p->p_pptr->p_sigacts->ps_flag &
646 (PS_NOCLDWAIT | PS_CLDSIGIGN)) {
649 mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx);
652 proc_reparent(p, p->p_reaper, true);
653 p->p_sigparent = SIGCHLD;
654 PROC_LOCK(p->p_pptr);
657 * Notify parent, so in case he was wait(2)ing or
658 * executing waitpid(2) with our pid, he will
663 mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx);
665 if (p->p_pptr == p->p_reaper || p->p_pptr == initproc) {
667 } else if (p->p_sigparent != 0) {
668 if (p->p_sigparent == SIGCHLD) {
670 } else { /* LINUX thread */
675 PROC_LOCK(p->p_pptr);
676 sx_xunlock(&proctree_lock);
678 if (signal_parent == 1) {
680 } else if (signal_parent == 2) {
681 kern_psignal(p->p_pptr, p->p_sigparent);
684 /* Tell the prison that we are gone. */
685 prison_proc_free(p->p_ucred->cr_prison);
688 * The state PRS_ZOMBIE prevents other proesses from sending
689 * signal to the process, to avoid memory leak, we free memory
690 * for signal queue at the time when the state is set.
692 sigqueue_flush(&p->p_sigqueue);
693 sigqueue_flush(&td->td_sigqueue);
696 * We have to wait until after acquiring all locks before
697 * changing p_state. We need to avoid all possible context
698 * switches (including ones from blocking on a mutex) while
699 * marked as a zombie. We also have to set the zombie state
700 * before we release the parent process' proc lock to avoid
701 * a lost wakeup. So, we first call wakeup, then we grab the
702 * sched lock, update the state, and release the parent process'
706 cv_broadcast(&p->p_pwait);
707 sched_exit(p->p_pptr, td);
709 p->p_state = PRS_ZOMBIE;
710 PROC_UNLOCK(p->p_pptr);
713 * Save our children's rusage information in our exit rusage.
716 ruadd(&p->p_ru, &p->p_rux, &p->p_stats->p_cru, &p->p_crux);
720 * Make sure the scheduler takes this thread out of its tables etc.
721 * This will also release this thread's reference to the ucred.
722 * Other thread parts to release include pcb bits and such.
727 #ifndef _SYS_SYSPROTO_H_
736 sys_abort2(struct thread *td, struct abort2_args *uap)
738 struct proc *p = td->td_proc;
744 * Do it right now so we can log either proper call of abort2(), or
745 * note, that invalid argument was passed. 512 is big enough to
746 * handle 16 arguments' descriptions with additional comments.
748 sb = sbuf_new(NULL, NULL, 512, SBUF_FIXEDLEN);
750 sbuf_printf(sb, "%s(pid %d uid %d) aborted: ",
751 p->p_comm, p->p_pid, td->td_ucred->cr_uid);
753 * Since we can't return from abort2(), send SIGKILL in cases, where
754 * abort2() was called improperly
757 /* Prevent from DoSes from user-space. */
758 if (uap->nargs < 0 || uap->nargs > 16)
760 if (uap->nargs > 0) {
761 if (uap->args == NULL)
763 error = copyin(uap->args, uargs, uap->nargs * sizeof(void *));
768 * Limit size of 'reason' string to 128. Will fit even when
769 * maximal number of arguments was chosen to be logged.
771 if (uap->why != NULL) {
772 error = sbuf_copyin(sb, uap->why, 128);
776 sbuf_printf(sb, "(null)");
778 if (uap->nargs > 0) {
779 sbuf_printf(sb, "(");
780 for (i = 0;i < uap->nargs; i++)
781 sbuf_printf(sb, "%s%p", i == 0 ? "" : ", ", uargs[i]);
782 sbuf_printf(sb, ")");
785 * Final stage: arguments were proper, string has been
786 * successfully copied from userspace, and copying pointers
787 * from user-space succeed.
791 if (sig == SIGKILL) {
793 sbuf_printf(sb, " (Reason text inaccessible)");
797 log(LOG_INFO, "%s", sbuf_data(sb));
805 * The dirty work is handled by kern_wait().
808 owait(struct thread *td, struct owait_args *uap __unused)
812 error = kern_wait(td, WAIT_ANY, &status, 0, NULL);
814 td->td_retval[1] = status;
817 #endif /* COMPAT_43 */
820 * The dirty work is handled by kern_wait().
823 sys_wait4(struct thread *td, struct wait4_args *uap)
825 struct rusage ru, *rup;
828 if (uap->rusage != NULL)
832 error = kern_wait(td, uap->pid, &status, uap->options, rup);
833 if (uap->status != NULL && error == 0 && td->td_retval[0] != 0)
834 error = copyout(&status, uap->status, sizeof(status));
835 if (uap->rusage != NULL && error == 0 && td->td_retval[0] != 0)
836 error = copyout(&ru, uap->rusage, sizeof(struct rusage));
841 sys_wait6(struct thread *td, struct wait6_args *uap)
843 struct __wrusage wru, *wrup;
849 idtype = uap->idtype;
852 if (uap->wrusage != NULL)
857 if (uap->info != NULL) {
859 bzero(sip, sizeof(*sip));
864 * We expect all callers of wait6() to know about WEXITED and
867 error = kern_wait6(td, idtype, id, &status, uap->options, wrup, sip);
869 if (uap->status != NULL && error == 0 && td->td_retval[0] != 0)
870 error = copyout(&status, uap->status, sizeof(status));
871 if (uap->wrusage != NULL && error == 0 && td->td_retval[0] != 0)
872 error = copyout(&wru, uap->wrusage, sizeof(wru));
873 if (uap->info != NULL && error == 0)
874 error = copyout(&si, uap->info, sizeof(si));
879 * Reap the remains of a zombie process and optionally return status and
880 * rusage. Asserts and will release both the proctree_lock and the process
881 * lock as part of its work.
884 proc_reap(struct thread *td, struct proc *p, int *status, int options)
888 sx_assert(&proctree_lock, SA_XLOCKED);
889 PROC_LOCK_ASSERT(p, MA_OWNED);
890 KASSERT(p->p_state == PRS_ZOMBIE, ("proc_reap: !PRS_ZOMBIE"));
892 mtx_spin_wait_unlocked(&p->p_slock);
897 *status = KW_EXITCODE(p->p_xexit, p->p_xsig);
898 if (options & WNOWAIT) {
900 * Only poll, returning the status. Caller does not wish to
901 * release the proc struct just yet.
904 sx_xunlock(&proctree_lock);
909 sigqueue_take(p->p_ksi);
913 * If we got the child via a ptrace 'attach', we need to give it back
916 if (p->p_oppid != p->p_pptr->p_pid) {
918 t = proc_realparent(p);
922 "wait: traced child %d moved back to parent %d", p->p_pid,
924 proc_reparent(p, t, false);
926 pksignal(t, SIGCHLD, p->p_ksi);
928 cv_broadcast(&p->p_pwait);
930 sx_xunlock(&proctree_lock);
936 * Remove other references to this process to ensure we have an
937 * exclusive reference.
939 sx_xlock(PIDHASHLOCK(p->p_pid));
940 LIST_REMOVE(p, p_hash);
941 sx_xunlock(PIDHASHLOCK(p->p_pid));
942 LIST_REMOVE(p, p_sibling);
943 reaper_abandon_children(p, true);
946 proc_clear_orphan(p);
949 if (p->p_procdesc != NULL)
951 sx_xunlock(&proctree_lock);
953 proc_id_clear(PROC_ID_PID, p->p_pid);
956 knlist_detach(p->p_klist);
961 * Removal from allproc list and process group list paired with
962 * PROC_LOCK which was executed during that time should guarantee
963 * nothing can reach this process anymore. As such further locking
966 p->p_xexit = p->p_xsig = 0; /* XXX: why? */
969 ruadd(&q->p_stats->p_cru, &q->p_crux, &p->p_ru, &p->p_rux);
973 * Decrement the count of procs running with this uid.
975 (void)chgproccnt(p->p_ucred->cr_ruidinfo, -1, 0);
978 * Destroy resource accounting information associated with the process.
983 racct_sub(p, RACCT_NPROC, 1);
990 * Free credentials, arguments, and sigacts.
993 pargs_drop(p->p_args);
995 sigacts_free(p->p_sigacts);
999 * Do any thread-system specific cleanups.
1004 * Give vm and machine-dependent layer a chance to free anything that
1005 * cpu_exit couldn't release while still running in process context.
1009 mac_proc_destroy(p);
1012 KASSERT(FIRST_THREAD_IN_PROC(p),
1013 ("proc_reap: no residual thread!"));
1014 uma_zfree(proc_zone, p);
1015 atomic_add_int(&nprocs, -1);
1019 proc_to_reap(struct thread *td, struct proc *p, idtype_t idtype, id_t id,
1020 int *status, int options, struct __wrusage *wrusage, siginfo_t *siginfo,
1025 sx_assert(&proctree_lock, SA_XLOCKED);
1031 if (p->p_procdesc == NULL ||
1032 (p->p_pptr == td->td_proc &&
1033 (p->p_flag & P_TRACED) != 0)) {
1040 if (p->p_pid != (pid_t)id) {
1046 if (p->p_pgid != (pid_t)id) {
1052 if (p->p_session->s_sid != (pid_t)id) {
1058 if (p->p_ucred->cr_uid != (uid_t)id) {
1064 if (p->p_ucred->cr_gid != (gid_t)id) {
1070 if (p->p_ucred->cr_prison->pr_id != (int)id) {
1076 * It seems that the thread structures get zeroed out
1077 * at process exit. This makes it impossible to
1078 * support P_SETID, P_CID or P_CPUID.
1085 if (p_canwait(td, p)) {
1090 if (((options & WEXITED) == 0) && (p->p_state == PRS_ZOMBIE)) {
1096 * This special case handles a kthread spawned by linux_clone
1097 * (see linux_misc.c). The linux_wait4 and linux_waitpid
1098 * functions need to be able to distinguish between waiting
1099 * on a process and waiting on a thread. It is a thread if
1100 * p_sigparent is not SIGCHLD, and the WLINUXCLONE option
1101 * signifies we want to wait for threads and not processes.
1103 if ((p->p_sigparent != SIGCHLD) ^
1104 ((options & WLINUXCLONE) != 0)) {
1109 if (siginfo != NULL) {
1110 bzero(siginfo, sizeof(*siginfo));
1111 siginfo->si_errno = 0;
1114 * SUSv4 requires that the si_signo value is always
1115 * SIGCHLD. Obey it despite the rfork(2) interface
1116 * allows to request other signal for child exit
1119 siginfo->si_signo = SIGCHLD;
1122 * This is still a rough estimate. We will fix the
1123 * cases TRAPPED, STOPPED, and CONTINUED later.
1125 if (WCOREDUMP(p->p_xsig)) {
1126 siginfo->si_code = CLD_DUMPED;
1127 siginfo->si_status = WTERMSIG(p->p_xsig);
1128 } else if (WIFSIGNALED(p->p_xsig)) {
1129 siginfo->si_code = CLD_KILLED;
1130 siginfo->si_status = WTERMSIG(p->p_xsig);
1132 siginfo->si_code = CLD_EXITED;
1133 siginfo->si_status = p->p_xexit;
1136 siginfo->si_pid = p->p_pid;
1137 siginfo->si_uid = p->p_ucred->cr_uid;
1140 * The si_addr field would be useful additional
1141 * detail, but apparently the PC value may be lost
1142 * when we reach this point. bzero() above sets
1143 * siginfo->si_addr to NULL.
1148 * There should be no reason to limit resources usage info to
1149 * exited processes only. A snapshot about any resources used
1150 * by a stopped process may be exactly what is needed.
1152 if (wrusage != NULL) {
1153 rup = &wrusage->wru_self;
1156 calcru(p, &rup->ru_utime, &rup->ru_stime);
1159 rup = &wrusage->wru_children;
1160 *rup = p->p_stats->p_cru;
1161 calccru(p, &rup->ru_utime, &rup->ru_stime);
1164 if (p->p_state == PRS_ZOMBIE && !check_only) {
1165 proc_reap(td, p, status, options);
1172 kern_wait(struct thread *td, pid_t pid, int *status, int options,
1173 struct rusage *rusage)
1175 struct __wrusage wru, *wrup;
1181 * Translate the special pid values into the (idtype, pid)
1182 * pair for kern_wait6. The WAIT_MYPGRP case is handled by
1183 * kern_wait6() on its own.
1185 if (pid == WAIT_ANY) {
1188 } else if (pid < 0) {
1202 * For backward compatibility we implicitly add flags WEXITED
1203 * and WTRAPPED here.
1205 options |= WEXITED | WTRAPPED;
1206 ret = kern_wait6(td, idtype, id, status, options, wrup, NULL);
1208 *rusage = wru.wru_self;
1213 report_alive_proc(struct thread *td, struct proc *p, siginfo_t *siginfo,
1214 int *status, int options, int si_code)
1218 PROC_LOCK_ASSERT(p, MA_OWNED);
1219 sx_assert(&proctree_lock, SA_XLOCKED);
1220 MPASS(si_code == CLD_TRAPPED || si_code == CLD_STOPPED ||
1221 si_code == CLD_CONTINUED);
1223 cont = si_code == CLD_CONTINUED;
1224 if ((options & WNOWAIT) == 0) {
1226 p->p_flag &= ~P_CONTINUED;
1228 p->p_flag |= P_WAITED;
1229 if (kern_wait_dequeue_sigchld &&
1230 (td->td_proc->p_sysent->sv_flags & SV_SIG_WAITNDQ) == 0) {
1231 PROC_LOCK(td->td_proc);
1232 sigqueue_take(p->p_ksi);
1233 PROC_UNLOCK(td->td_proc);
1236 sx_xunlock(&proctree_lock);
1237 if (siginfo != NULL) {
1238 siginfo->si_code = si_code;
1239 siginfo->si_status = cont ? SIGCONT : p->p_xsig;
1242 *status = cont ? SIGCONT : W_STOPCODE(p->p_xsig);
1244 td->td_retval[0] = p->p_pid;
1248 kern_wait6(struct thread *td, idtype_t idtype, id_t id, int *status,
1249 int options, struct __wrusage *wrusage, siginfo_t *siginfo)
1253 int error, nfound, ret;
1256 AUDIT_ARG_VALUE((int)idtype); /* XXX - This is likely wrong! */
1257 AUDIT_ARG_PID((pid_t)id); /* XXX - This may be wrong! */
1258 AUDIT_ARG_VALUE(options);
1262 if ((pid_t)id == WAIT_MYPGRP && (idtype == P_PID || idtype == P_PGID)) {
1264 id = (id_t)q->p_pgid;
1269 /* If we don't know the option, just return. */
1270 if ((options & ~(WUNTRACED | WNOHANG | WCONTINUED | WNOWAIT |
1271 WEXITED | WTRAPPED | WLINUXCLONE)) != 0)
1273 if ((options & (WEXITED | WUNTRACED | WCONTINUED | WTRAPPED)) == 0) {
1275 * We will be unable to find any matching processes,
1276 * because there are no known events to look for.
1277 * Prefer to return error instead of blocking
1284 if (q->p_flag & P_STATCHILD) {
1286 q->p_flag &= ~P_STATCHILD;
1289 sx_xlock(&proctree_lock);
1292 LIST_FOREACH(p, &q->p_children, p_sibling) {
1294 ret = proc_to_reap(td, p, idtype, id, status, options,
1295 wrusage, siginfo, 0);
1298 else if (ret != 1) {
1299 td->td_retval[0] = pid;
1304 PROC_LOCK_ASSERT(p, MA_OWNED);
1306 if ((options & WTRAPPED) != 0 &&
1307 (p->p_flag & P_TRACED) != 0) {
1310 ((p->p_flag & (P_STOPPED_TRACE | P_STOPPED_SIG)) &&
1311 p->p_suspcount == p->p_numthreads &&
1312 (p->p_flag & P_WAITED) == 0);
1316 "wait: returning trapped pid %d status %#x "
1317 "(xstat %d) xthread %d",
1318 p->p_pid, W_STOPCODE(p->p_xsig), p->p_xsig,
1319 p->p_xthread != NULL ?
1320 p->p_xthread->td_tid : -1);
1321 report_alive_proc(td, p, siginfo, status,
1322 options, CLD_TRAPPED);
1326 if ((options & WUNTRACED) != 0 &&
1327 (p->p_flag & P_STOPPED_SIG) != 0) {
1329 report = (p->p_suspcount == p->p_numthreads &&
1330 ((p->p_flag & P_WAITED) == 0));
1333 report_alive_proc(td, p, siginfo, status,
1334 options, CLD_STOPPED);
1338 if ((options & WCONTINUED) != 0 &&
1339 (p->p_flag & P_CONTINUED) != 0) {
1340 report_alive_proc(td, p, siginfo, status, options,
1348 * Look in the orphans list too, to allow the parent to
1349 * collect it's child exit status even if child is being
1352 * Debugger detaches from the parent upon successful
1353 * switch-over from parent to child. At this point due to
1354 * re-parenting the parent loses the child to debugger and a
1355 * wait4(2) call would report that it has no children to wait
1356 * for. By maintaining a list of orphans we allow the parent
1357 * to successfully wait until the child becomes a zombie.
1360 LIST_FOREACH(p, &q->p_orphans, p_orphan) {
1361 ret = proc_to_reap(td, p, idtype, id, NULL, options,
1364 KASSERT(ret != -1, ("reaped an orphan (pid %d)",
1365 (int)td->td_retval[0]));
1373 sx_xunlock(&proctree_lock);
1376 if (options & WNOHANG) {
1377 sx_xunlock(&proctree_lock);
1378 td->td_retval[0] = 0;
1382 if (q->p_flag & P_STATCHILD) {
1383 q->p_flag &= ~P_STATCHILD;
1387 sx_xunlock(&proctree_lock);
1388 error = msleep(q, &q->p_mtx, PWAIT | PCATCH | PDROP, "wait", 0);
1395 proc_add_orphan(struct proc *child, struct proc *parent)
1398 sx_assert(&proctree_lock, SX_XLOCKED);
1399 KASSERT((child->p_flag & P_TRACED) != 0,
1400 ("proc_add_orphan: not traced"));
1402 if (LIST_EMPTY(&parent->p_orphans)) {
1403 child->p_treeflag |= P_TREE_FIRST_ORPHAN;
1404 LIST_INSERT_HEAD(&parent->p_orphans, child, p_orphan);
1406 LIST_INSERT_AFTER(LIST_FIRST(&parent->p_orphans),
1409 child->p_treeflag |= P_TREE_ORPHANED;
1413 * Make process 'parent' the new parent of process 'child'.
1414 * Must be called with an exclusive hold of proctree lock.
1417 proc_reparent(struct proc *child, struct proc *parent, bool set_oppid)
1420 sx_assert(&proctree_lock, SX_XLOCKED);
1421 PROC_LOCK_ASSERT(child, MA_OWNED);
1422 if (child->p_pptr == parent)
1425 PROC_LOCK(child->p_pptr);
1426 sigqueue_take(child->p_ksi);
1427 PROC_UNLOCK(child->p_pptr);
1428 LIST_REMOVE(child, p_sibling);
1429 LIST_INSERT_HEAD(&parent->p_children, child, p_sibling);
1431 proc_clear_orphan(child);
1432 if ((child->p_flag & P_TRACED) != 0) {
1433 proc_add_orphan(child, child->p_pptr);
1436 child->p_pptr = parent;
1438 child->p_oppid = parent->p_pid;