2 * Copyright (c) 1982, 1986, 1989, 1991, 1993
3 * The Regents of the University of California. All rights reserved.
4 * (c) UNIX System Laboratories, Inc.
5 * All or some portions of this file are derived from material licensed
6 * to the University of California by American Telephone and Telegraph
7 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8 * the permission of UNIX System Laboratories, Inc.
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 3. Neither the name of the University nor the names of its contributors
19 * may be used to endorse or promote products derived from this software
20 * without specific prior written permission.
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * @(#)kern_exit.c 8.7 (Berkeley) 2/12/94
35 * $FreeBSD: src/sys/kern/kern_exit.c,v 1.92.2.11 2003/01/13 22:51:16 dillon Exp $
38 #include "opt_ktrace.h"
40 #include <sys/param.h>
41 #include <sys/systm.h>
42 #include <sys/sysproto.h>
43 #include <sys/kernel.h>
44 #include <sys/malloc.h>
46 #include <sys/ktrace.h>
47 #include <sys/pioctl.h>
50 #include <sys/vnode.h>
51 #include <sys/resourcevar.h>
52 #include <sys/signalvar.h>
53 #include <sys/taskqueue.h>
54 #include <sys/ptrace.h>
55 #include <sys/acct.h> /* for acct_process() function prototype */
56 #include <sys/filedesc.h>
60 #include <sys/kern_syscall.h>
61 #include <sys/unistd.h>
62 #include <sys/eventhandler.h>
63 #include <sys/dsched.h>
66 #include <vm/vm_param.h>
69 #include <vm/vm_map.h>
70 #include <vm/vm_extern.h>
72 #include <sys/refcount.h>
73 #include <sys/spinlock2.h>
74 #include <sys/mplock2.h>
76 #include <machine/vmm.h>
78 static void reaplwps(void *context, int dummy);
79 static void reaplwp(struct lwp *lp);
80 static void killlwps(struct lwp *lp);
82 static MALLOC_DEFINE(M_ATEXIT, "atexit", "atexit callback");
85 * callout list for things to do at exit time
89 TAILQ_ENTRY(exitlist) next;
92 TAILQ_HEAD(exit_list_head, exitlist);
93 static struct exit_list_head exit_list = TAILQ_HEAD_INITIALIZER(exit_list);
98 static struct task *deadlwp_task[MAXCPU];
99 static struct lwplist deadlwp_list[MAXCPU];
100 static struct lwkt_token deadlwp_token[MAXCPU];
106 * SYS_EXIT_ARGS(int rval)
109 sys_exit(struct exit_args *uap)
111 exit1(W_EXITCODE(uap->rval, 0));
117 * Death of a lwp or process with optional bells and whistles.
120 sys_extexit(struct extexit_args *uap)
122 struct proc *p = curproc;
126 action = EXTEXIT_ACTION(uap->how);
127 who = EXTEXIT_WHO(uap->how);
129 /* Check parameters before we might perform some action */
142 error = copyout(&uap->status, uap->addr, sizeof(uap->status));
150 lwkt_gettoken(&p->p_token);
155 * Be sure only to perform a simple lwp exit if there is at
156 * least one more lwp in the proc, which will call exit1()
157 * later, otherwise the proc will be an UNDEAD and not even a
160 if (p->p_nthreads > 1) {
161 lwp_exit(0, NULL); /* called w/ p_token held */
164 /* else last lwp in proc: do the real thing */
166 default: /* to help gcc */
168 lwkt_reltoken(&p->p_token);
169 exit1(W_EXITCODE(uap->status, 0));
174 lwkt_reltoken(&p->p_token); /* safety */
178 * Kill all lwps associated with the current process except the
179 * current lwp. Return an error if we race another thread trying to
180 * do the same thing and lose the race.
182 * If forexec is non-zero the current thread and process flags are
183 * cleaned up so they can be reused.
186 killalllwps(int forexec)
188 struct lwp *lp = curthread->td_lwp;
189 struct proc *p = lp->lwp_proc;
193 * Interlock against P_WEXIT. Only one of the process's thread
194 * is allowed to do the master exit.
196 lwkt_gettoken(&p->p_token);
197 if (p->p_flags & P_WEXIT) {
198 lwkt_reltoken(&p->p_token);
201 p->p_flags |= P_WEXIT;
202 lwkt_gettoken(&lp->lwp_token);
205 * Set temporary stopped state in case we are racing a coredump.
206 * Otherwise the coredump may hang forever.
208 if (lp->lwp_mpflags & LWP_MP_WSTOP) {
211 atomic_set_int(&lp->lwp_mpflags, LWP_MP_WSTOP);
214 wakeup(&p->p_nstopped);
218 * Interlock with LWP_MP_WEXIT and kill any remaining LWPs
220 atomic_set_int(&lp->lwp_mpflags, LWP_MP_WEXIT);
221 if (p->p_nthreads > 1)
225 * Undo temporary stopped state
227 if (fakestop && (lp->lwp_mpflags & LWP_MP_WSTOP)) {
228 atomic_clear_int(&lp->lwp_mpflags, LWP_MP_WSTOP);
233 * If doing this for an exec, clean up the remaining thread
234 * (us) for continuing operation after all the other threads
238 atomic_clear_int(&lp->lwp_mpflags, LWP_MP_WEXIT);
239 p->p_flags &= ~P_WEXIT;
241 lwkt_reltoken(&lp->lwp_token);
242 lwkt_reltoken(&p->p_token);
248 * Kill all LWPs except the current one. Do not try to signal
249 * LWPs which have exited on their own or have already been
253 killlwps(struct lwp *lp)
255 struct proc *p = lp->lwp_proc;
259 * Kill the remaining LWPs. We must send the signal before setting
260 * LWP_MP_WEXIT. The setting of WEXIT is optional but helps reduce
261 * races. tlp must be held across the call as it might block and
262 * allow the target lwp to rip itself out from under our loop.
264 FOREACH_LWP_IN_PROC(tlp, p) {
266 lwkt_gettoken(&tlp->lwp_token);
267 if ((tlp->lwp_mpflags & LWP_MP_WEXIT) == 0) {
268 atomic_set_int(&tlp->lwp_mpflags, LWP_MP_WEXIT);
269 lwpsignal(p, tlp, SIGKILL);
271 lwkt_reltoken(&tlp->lwp_token);
276 * Wait for everything to clear out. Also make sure any tstop()s
277 * are signalled (we are holding p_token for the interlock).
280 while (p->p_nthreads > 1)
281 tsleep(&p->p_nthreads, 0, "killlwps", 0);
285 * Exit: deallocate address space and other resources, change proc state
286 * to zombie, and unlink proc from allproc and parent's lists. Save exit
287 * status and rusage for wait(). Check for child processes and orphan them.
292 struct thread *td = curthread;
293 struct proc *p = td->td_proc;
294 struct lwp *lp = td->td_lwp;
298 struct sysreaper *reap;
304 lwkt_gettoken(&p->p_token);
307 kprintf("init died (signal %d, exit %d)\n",
308 WTERMSIG(rv), WEXITSTATUS(rv));
309 panic("Going nowhere without my init!");
311 varsymset_clean(&p->p_varsymset);
312 lockuninit(&p->p_varsymset.vx_lock);
315 * Kill all lwps associated with the current process, return an
316 * error if we race another thread trying to do the same thing
319 error = killalllwps(0);
325 /* are we a task leader? */
326 if (p == p->p_leader) {
327 struct kill_args killArgs;
328 killArgs.signum = SIGKILL;
331 killArgs.pid = q->p_pid;
333 * The interface for kill is better
334 * than the internal signal
340 tsleep((caddr_t)p, 0, "exit1", 0);
346 STOPEVENT(p, S_EXIT, rv);
347 p->p_flags |= P_POSTEXIT; /* stop procfs stepping */
350 * Check if any loadable modules need anything done at process exit.
351 * e.g. SYSV IPC stuff
352 * XXX what if one of these generates an error?
357 * XXX: imho, the eventhandler stuff is much cleaner than this.
358 * Maybe we should move everything to use eventhandler.
360 TAILQ_FOREACH(ep, &exit_list, next)
363 if (p->p_flags & P_PROFIL)
366 SIGEMPTYSET(p->p_siglist);
367 SIGEMPTYSET(lp->lwp_siglist);
368 if (timevalisset(&p->p_realtimer.it_value))
369 callout_terminate(&p->p_ithandle);
372 * Reset any sigio structures pointing to us as a result of
373 * F_SETOWN with our pid.
375 funsetownlst(&p->p_sigiolst);
378 * Close open files and release open-file table.
383 if (p->p_leader->p_peers) {
385 while(q->p_peers != p)
387 q->p_peers = p->p_peers;
388 wakeup((caddr_t)p->p_leader);
392 * XXX Shutdown SYSV semaphores
396 /* The next two chunks should probably be moved to vmspace_exit. */
400 * Clean up data related to virtual kernel operation. Clean up
401 * any vkernel context related to the current lwp now so we can
405 vkernel_lwp_exit(lp);
410 * Release the user portion of address space. The exitbump prevents
411 * the vmspace from being completely eradicated (using holdcnt).
412 * This releases references to vnodes, which could cause I/O if the
413 * file has been unlinked. We need to do this early enough that
414 * we can still sleep.
416 * We can't free the entire vmspace as the kernel stack may be mapped
417 * within that space also.
419 * Processes sharing the same vmspace may exit in one order, and
420 * get cleaned up by vmspace_exit() in a different order. The
421 * last exiting process to reach this point releases as much of
422 * the environment as it can, and the last process cleaned up
423 * by vmspace_exit() (which decrements exitingcnt) cleans up the
426 * NOTE: Releasing p_token around this call is helpful if the
427 * vmspace had a huge RSS. Otherwise some other process
428 * trying to do an allproc or other scan (like 'ps') may
429 * stall for a long time.
431 lwkt_reltoken(&p->p_token);
433 lwkt_gettoken(&p->p_token);
435 if (SESS_LEADER(p)) {
436 struct session *sp = p->p_session;
440 * We are the controlling process. Signal the
441 * foreground process group, drain the controlling
442 * terminal, and revoke access to the controlling
445 * NOTE: while waiting for the process group to exit
446 * it is possible that one of the processes in the
447 * group will revoke the tty, so the ttyclosesession()
448 * function will re-check sp->s_ttyvp.
450 if (sp->s_ttyp && (sp->s_ttyp->t_session == sp)) {
451 if (sp->s_ttyp->t_pgrp)
452 pgsignal(sp->s_ttyp->t_pgrp, SIGHUP, 1);
454 ttyclosesession(sp, 1); /* also revoke */
457 * Release the tty. If someone has it open via
458 * /dev/tty then close it (since they no longer can
459 * once we've NULL'd it out).
461 ttyclosesession(sp, 0);
464 * s_ttyp is not zero'd; we use this to indicate
465 * that the session once had a controlling terminal.
466 * (for logging and informational purposes)
471 fixjobc(p, p->p_pgrp, 0);
472 (void)acct_process(p);
478 ktrdestroy(&p->p_tracenode);
482 * Release reference to text vnode
484 if ((vtmp = p->p_textvp) != NULL) {
489 /* Release namecache handle to text file */
490 if (p->p_textnch.ncp)
491 cache_drop(&p->p_textnch);
494 * We have to handle PPWAIT here or proc_move_allproc_zombie()
495 * will block on the PHOLD() the parent is doing.
497 * We are using the flag as an interlock so an atomic op is
498 * necessary to synchronize with the parent's cpu.
500 if (p->p_flags & P_PPWAIT) {
501 if (p->p_pptr && p->p_pptr->p_upmap)
502 atomic_add_int(&p->p_pptr->p_upmap->invfork, -1);
503 atomic_clear_int(&p->p_flags, P_PPWAIT);
508 * Move the process to the zombie list. This will block
509 * until the process p_lock count reaches 0. The process will
510 * not be reaped until TDF_EXITING is set by cpu_thread_exit(),
511 * which is called from cpu_proc_exit().
513 * Interlock against waiters using p_waitgen. We increment
514 * p_waitgen after completing the move of our process to the
517 * WARNING: pp becomes stale when we block, clear it now as a
520 proc_move_allproc_zombie(p);
522 atomic_add_long(&pp->p_waitgen, 1);
526 * release controlled reaper for exit if we own it and return the
527 * remaining reaper (the one for us), which we will drop after we
530 reap = reaper_exit(p);
533 * Reparent all of this process's children to the init process or
534 * to the designated reaper. We must hold the reaper's p_token in
535 * order to safely mess with p_children.
537 * We already hold p->p_token (to remove the children from our list).
540 q = LIST_FIRST(&p->p_children);
542 reproc = reaper_get(reap);
543 lwkt_gettoken(&reproc->p_token);
544 while ((q = LIST_FIRST(&p->p_children)) != NULL) {
546 lwkt_gettoken(&q->p_token);
547 if (q != LIST_FIRST(&p->p_children)) {
548 lwkt_reltoken(&q->p_token);
552 LIST_REMOVE(q, p_sibling);
553 LIST_INSERT_HEAD(&reproc->p_children, q, p_sibling);
555 q->p_ppid = reproc->p_pid;
556 q->p_sigparent = SIGCHLD;
559 * Traced processes are killed
560 * since their existence means someone is screwing up.
562 if (q->p_flags & P_TRACED) {
563 q->p_flags &= ~P_TRACED;
566 lwkt_reltoken(&q->p_token);
569 lwkt_reltoken(&reproc->p_token);
574 * Save exit status and final rusage info. We no longer add
575 * child rusage info into self times, wait4() and kern_wait()
576 * handles it in order to properly support wait6().
578 calcru_proc(p, &p->p_ru);
579 /*ruadd(&p->p_ru, &p->p_cru); REMOVED */
582 * notify interested parties of our demise.
584 KNOTE(&p->p_klist, NOTE_EXIT);
587 * Notify parent that we're gone. If parent has the PS_NOCLDWAIT
588 * flag set, or if the handler is set to SIG_IGN, notify the reaper
589 * instead (it will handle this situation).
591 * NOTE: The reaper can still be the parent process.
595 if (p->p_pptr->p_sigacts->ps_flag & (PS_NOCLDWAIT | PS_CLDSIGIGN)) {
597 reproc = reaper_get(reap);
598 proc_reparent(p, reproc);
606 * Signal (possibly new) parent.
610 if (p->p_sigparent && pp != initproc) {
611 int sig = p->p_sigparent;
613 if (sig != SIGUSR1 && sig != SIGCHLD)
617 ksignal(pp, SIGCHLD);
619 p->p_flags &= ~P_TRACED;
623 * cpu_exit is responsible for clearing curproc, since
624 * it is heavily integrated with the thread/switching sequence.
626 * Other substructures are freed from wait().
629 struct plimit *rlimit;
637 * Finally, call machine-dependent code to release as many of the
638 * lwp's resources as we can and halt execution of this thread.
640 * pp is a wild pointer now but still the correct wakeup() target.
641 * lwp_exit() only uses it to send the wakeup() signal to the likely
642 * parent. Any reparenting race that occurs will get a signal
643 * automatically and not be an issue.
649 * Eventually called by every exiting LWP
651 * p->p_token must be held. mplock may be held and will be released.
654 lwp_exit(int masterexit, void *waddr)
656 struct thread *td = curthread;
657 struct lwp *lp = td->td_lwp;
658 struct proc *p = lp->lwp_proc;
662 * Release the current user process designation on the process so
663 * the userland scheduler can work in someone else.
665 p->p_usched->release_curproc(lp);
668 * lwp_exit() may be called without setting LWP_MP_WEXIT, so
669 * make sure it is set here.
671 ASSERT_LWKT_TOKEN_HELD(&p->p_token);
672 atomic_set_int(&lp->lwp_mpflags, LWP_MP_WEXIT);
675 * Clean up any virtualization
678 vkernel_lwp_exit(lp);
684 * Clean up select/poll support
686 kqueue_terminate(&lp->lwp_kqueue);
689 * Clean up any syscall-cached ucred or rlimit.
692 crfree(td->td_ucred);
696 struct plimit *rlimit;
698 rlimit = td->td_limit;
704 * Cleanup any cached descriptors for this thread
710 * Nobody actually wakes us when the lock
711 * count reaches zero, so just wait one tick.
713 while (lp->lwp_lock > 0)
714 tsleep(lp, 0, "lwpexit", 1);
716 /* Hand down resource usage to our proc */
717 ruadd(&p->p_ru, &lp->lwp_ru);
720 * If we don't hold the process until the LWP is reaped wait*()
721 * may try to dispose of its vmspace before all the LWPs have
722 * actually terminated.
727 * Do any remaining work that might block on us. We should be
728 * coded such that further blocking is ok after decrementing
729 * p_nthreads but don't take the chance.
731 dsched_exit_thread(td);
732 biosched_done(curthread);
735 * We have to use the reaper for all the LWPs except the one doing
736 * the master exit. The LWP doing the master exit can just be
737 * left on p_lwps and the process reaper will deal with it
738 * synchronously, which is much faster.
740 * Wakeup anyone waiting on p_nthreads to drop to 1 or 0.
742 * The process is left held until the reaper calls lwp_dispose() on
743 * the lp (after calling lwp_wait()).
745 if (masterexit == 0) {
748 lwp_rb_tree_RB_REMOVE(&p->p_lwp_tree, lp);
750 if ((p->p_flags & P_MAYBETHREADED) && p->p_nthreads <= 1)
752 lwkt_gettoken(&deadlwp_token[cpu]);
753 LIST_INSERT_HEAD(&deadlwp_list[cpu], lp, u.lwp_reap_entry);
754 taskqueue_enqueue(taskqueue_thread[cpu], deadlwp_task[cpu]);
755 lwkt_reltoken(&deadlwp_token[cpu]);
758 if ((p->p_flags & P_MAYBETHREADED) && p->p_nthreads <= 1)
763 * We no longer need p_token.
765 * Tell the userland scheduler that we are going away
767 lwkt_reltoken(&p->p_token);
768 p->p_usched->heuristic_exiting(lp, p);
771 * Issue late wakeups after releasing our token to give us a chance
772 * to deschedule and switch away before another cpu in a wait*()
773 * reaps us. This is done as late as possible to reduce contention.
776 wakeup(&p->p_nthreads);
784 * Wait until a lwp is completely dead. The final interlock in this drama
785 * is when TDF_EXITING is set in cpu_thread_exit() just before the final
788 * At the point TDF_EXITING is set a complete exit is accomplished when
789 * TDF_RUNNING and TDF_PREEMPT_LOCK are both clear. td_mpflags has two
790 * post-switch interlock flags that can be used to wait for the TDF_
793 * Returns non-zero on success, and zero if the caller needs to retry
797 lwp_wait(struct lwp *lp)
799 struct thread *td = lp->lwp_thread;
802 KKASSERT(lwkt_preempted_proc() != lp);
805 * This bit of code uses the thread destruction interlock
806 * managed by lwkt_switch_return() to wait for the lwp's
807 * thread to completely disengage.
809 * It is possible for us to race another cpu core so we
810 * have to do this correctly.
813 mpflags = td->td_mpflags;
815 if (mpflags & TDF_MP_EXITSIG)
817 tsleep_interlock(td, 0);
818 if (atomic_cmpset_int(&td->td_mpflags, mpflags,
819 mpflags | TDF_MP_EXITWAIT)) {
820 tsleep(td, PINTERLOCKED, "lwpxt", 0);
825 * We've already waited for the core exit but there can still
826 * be other refs from e.g. process scans and such.
828 if (lp->lwp_lock > 0) {
829 tsleep(lp, 0, "lwpwait1", 1);
833 tsleep(td, 0, "lwpwait2", 1);
838 * Now that we have the thread destruction interlock these flags
839 * really should already be cleaned up, keep a check for safety.
841 * We can't rip its stack out from under it until TDF_EXITING is
842 * set and both TDF_RUNNING and TDF_PREEMPT_LOCK are clear.
843 * TDF_PREEMPT_LOCK must be checked because TDF_RUNNING
844 * will be cleared temporarily if a thread gets preempted.
846 while ((td->td_flags & (TDF_RUNNING |
849 TDF_EXITING)) != TDF_EXITING) {
850 tsleep(lp, 0, "lwpwait3", 1);
854 KASSERT((td->td_flags & (TDF_RUNQ|TDF_TSLEEPQ)) == 0,
855 ("lwp_wait: td %p (%s) still on run or sleep queue",
861 * Release the resources associated with a lwp.
862 * The lwp must be completely dead.
865 lwp_dispose(struct lwp *lp)
867 struct thread *td = lp->lwp_thread;
869 KKASSERT(lwkt_preempted_proc() != lp);
870 KKASSERT(lp->lwp_lock == 0);
871 KKASSERT(td->td_refs == 0);
872 KKASSERT((td->td_flags & (TDF_RUNNING |
875 TDF_EXITING)) == TDF_EXITING);
882 lp->lwp_thread = NULL;
883 lwkt_free_thread(td);
889 sys_wait4(struct wait_args *uap)
891 struct __wrusage wrusage;
898 options = uap->options | WEXITED | WTRAPPED;
901 if (id == WAIT_ANY) {
903 } else if (id == WAIT_MYPGRP) {
905 id = curproc->p_pgid;
913 error = kern_wait(idtype, id, &status, options, &wrusage,
914 NULL, &uap->sysmsg_result);
916 if (error == 0 && uap->status)
917 error = copyout(&status, uap->status, sizeof(*uap->status));
918 if (error == 0 && uap->rusage) {
919 ruadd(&wrusage.wru_self, &wrusage.wru_children);
920 error = copyout(&wrusage.wru_self, uap->rusage, sizeof(*uap->rusage));
926 sys_wait6(struct wait6_args *uap)
928 struct __wrusage wrusage;
938 * NOTE: wait6() requires WEXITED and WTRAPPED to be specified if
941 options = uap->options;
942 idtype = uap->idtype;
944 infop = uap->info ? &info : NULL;
949 if (id == WAIT_MYPGRP) {
951 id = curproc->p_pgid;
955 /* let kern_wait deal with the remainder */
959 error = kern_wait(idtype, id, &status, options,
960 &wrusage, infop, &uap->sysmsg_result);
962 if (error == 0 && uap->status)
963 error = copyout(&status, uap->status, sizeof(*uap->status));
964 if (error == 0 && uap->wrusage)
965 error = copyout(&wrusage, uap->wrusage, sizeof(*uap->wrusage));
966 if (error == 0 && uap->info)
967 error = copyout(&info, uap->info, sizeof(*uap->info));
972 * kernel wait*() system call support
975 kern_wait(idtype_t idtype, id_t id, int *status, int options,
976 struct __wrusage *wrusage, siginfo_t *info, int *res)
978 struct thread *td = curthread;
980 struct proc *q = td->td_proc;
989 * Must not have extraneous options. Must have at least one
992 if (options &~ (WUNTRACED|WNOHANG|WCONTINUED|WLINUXCLONE|WSTOPPED|
993 WEXITED|WTRAPPED|WNOWAIT)) {
996 if ((options & (WEXITED | WUNTRACED | WCONTINUED | WTRAPPED)) == 0) {
1001 * Protect the q->p_children list
1003 lwkt_gettoken(&q->p_token);
1006 * All sorts of things can change due to blocking so we have to loop
1007 * all the way back up here.
1009 * The problem is that if a process group is stopped and the parent
1010 * is doing a wait*(..., WUNTRACED, ...), it will see the STOP
1011 * of the child and then stop itself when it tries to return from the
1012 * system call. When the process group is resumed the parent will
1013 * then get the STOP status even though the child has now resumed
1014 * (a followup wait*() will get the CONT status).
1016 * Previously the CONT would overwrite the STOP because the tstop
1017 * was handled within tsleep(), and the parent would only see
1018 * the CONT when both are stopped and continued together. This little
1019 * two-line hack restores this effect.
1021 * No locks are held so we can safely block the process here.
1023 if (STOPLWP(q, td->td_lwp))
1031 * NOTE: We don't want to break q's p_token in the loop for the
1032 * case where no children are found or we risk breaking the
1033 * interlock between child and parent.
1035 waitgen = atomic_fetchadd_long(&q->p_waitgen, 0x80000000);
1036 LIST_FOREACH(p, &q->p_children, p_sibling) {
1038 * Filter, (p) will be held on fall-through. Try to optimize
1039 * this to avoid the atomic op until we are pretty sure we
1040 * want this process.
1047 if (p->p_pid != (pid_t)id)
1052 if (p->p_pgid != (pid_t)id)
1058 if (p->p_session && p->p_session->s_sid != (pid_t)id) {
1065 if (p->p_ucred->cr_uid != (uid_t)id) {
1072 if (p->p_ucred->cr_gid != (gid_t)id) {
1079 if (p->p_ucred->cr_prison &&
1080 p->p_ucred->cr_prison->pr_id != (int)id) {
1086 /* unsupported filter */
1089 /* (p) is held at this point */
1092 * This special case handles a kthread spawned by linux_clone
1093 * (see linux_misc.c). The linux_wait4 and linux_waitpid
1094 * functions need to be able to distinguish between waiting
1095 * on a process and waiting on a thread. It is a thread if
1096 * p_sigparent is not SIGCHLD, and the WLINUXCLONE option
1097 * signifies we want to wait for threads and not processes.
1099 if ((p->p_sigparent != SIGCHLD) ^
1100 ((options & WLINUXCLONE) != 0)) {
1106 if (p->p_stat == SZOMB && (options & WEXITED)) {
1108 * We may go into SZOMB with threads still present.
1109 * We must wait for them to exit before we can reap
1110 * the master thread, otherwise we may race reaping
1111 * non-master threads.
1113 * Only this routine can remove a process from
1114 * the zombie list and destroy it.
1120 lwkt_gettoken(&p->p_token);
1121 if (p->p_pptr != q) {
1122 lwkt_reltoken(&p->p_token);
1127 while (p->p_nthreads > 0) {
1128 tsleep(&p->p_nthreads, 0, "lwpzomb", hz);
1132 * Reap any LWPs left in p->p_lwps. This is usually
1133 * just the last LWP. This must be done before
1134 * we loop on p_lock since the lwps hold a ref on
1135 * it as a vmspace interlock.
1137 * Once that is accomplished p_nthreads had better
1140 while ((lp = RB_ROOT(&p->p_lwp_tree)) != NULL) {
1142 * Make sure no one is using this lwp, before
1143 * it is removed from the tree. If we didn't
1144 * wait it here, lwp tree iteration with
1145 * blocking operation would be broken.
1147 while (lp->lwp_lock > 0)
1148 tsleep(lp, 0, "zomblwp", 1);
1149 lwp_rb_tree_RB_REMOVE(&p->p_lwp_tree, lp);
1152 KKASSERT(p->p_nthreads == 0);
1155 * Don't do anything really bad until all references
1156 * to the process go away. This may include other
1157 * LWPs which are still in the process of being
1158 * reaped. We can't just pull the rug out from under
1159 * them because they may still be using the VM space.
1161 * Certain kernel facilities such as /proc will also
1162 * put a hold on the process for short periods of
1165 PRELE(p); /* from top of loop */
1166 PSTALL(p, "reap3", 1); /* 1 ref (for PZOMBHOLD) */
1168 /* Take care of our return values. */
1171 *status = p->p_xstat;
1172 wrusage->wru_self = p->p_ru;
1173 wrusage->wru_children = p->p_cru;
1176 bzero(info, sizeof(*info));
1178 info->si_signo = SIGCHLD;
1179 if (WIFEXITED(p->p_xstat)) {
1180 info->si_code = CLD_EXITED;
1182 WEXITSTATUS(p->p_xstat);
1184 info->si_code = CLD_KILLED;
1185 info->si_status = WTERMSIG(p->p_xstat);
1187 info->si_pid = p->p_pid;
1188 info->si_uid = p->p_ucred->cr_uid;
1192 * WNOWAIT shortcuts to done here, leaving the
1193 * child on the zombie list.
1195 if (options & WNOWAIT) {
1196 lwkt_reltoken(&p->p_token);
1203 * If we got the child via a ptrace 'attach',
1204 * we need to give it back to the old parent.
1206 if (p->p_oppid && (t = pfind(p->p_oppid)) != NULL) {
1208 proc_reparent(p, t);
1209 ksignal(t, SIGCHLD);
1212 lwkt_reltoken(&p->p_token);
1219 * Unlink the proc from its process group so that
1220 * the following operations won't lead to an
1221 * inconsistent state for processes running down
1224 proc_remove_zombie(p);
1226 lwkt_reltoken(&p->p_token);
1230 ruadd(&q->p_cru, &p->p_ru);
1231 ruadd(&q->p_cru, &p->p_cru);
1234 * Decrement the count of procs running with this uid.
1236 chgproccnt(p->p_ucred->cr_ruidinfo, -1, 0);
1239 * Free up credentials. p_spin is required to
1240 * avoid races against allproc scans.
1242 spin_lock(&p->p_spin);
1245 spin_unlock(&p->p_spin);
1249 * Remove unused arguments
1253 if (pa && refcount_release(&pa->ar_ref)) {
1259 p->p_sigacts = NULL;
1260 if (ps && refcount_release(&ps->ps_refcnt)) {
1261 kfree(ps, M_SUBPROC);
1266 * Our exitingcount was incremented when the process
1267 * became a zombie, now that the process has been
1268 * removed from (almost) all lists we should be able
1269 * to safely destroy its vmspace. Wait for any current
1270 * holders to go away (so the vmspace remains stable),
1273 * NOTE: Releasing the parent process (q) p_token
1274 * across the vmspace_exitfree() call is
1275 * important here to reduce stalls on
1276 * interactions with (q) (such as
1277 * fork/exec/wait or 'ps').
1279 PSTALL(p, "reap4", 1);
1280 lwkt_reltoken(&q->p_token);
1281 vmspace_exitfree(p);
1282 lwkt_gettoken(&q->p_token);
1283 PSTALL(p, "reap5", 1);
1286 * NOTE: We have to officially release ZOMB in order
1287 * to ensure that a racing thread in kern_wait()
1288 * which blocked on ZOMB is woken up.
1291 kfree(p->p_uidpcpu, M_SUBPROC);
1293 atomic_add_int(&nprocs, -1);
1299 * Process has not yet exited
1301 if ((p->p_stat == SSTOP || p->p_stat == SCORE) &&
1302 (p->p_flags & P_WAITED) == 0 &&
1303 (((p->p_flags & P_TRACED) && (options & WTRAPPED)) ||
1304 (options & WSTOPPED))) {
1305 lwkt_gettoken(&p->p_token);
1306 if (p->p_pptr != q) {
1307 lwkt_reltoken(&p->p_token);
1311 if ((p->p_stat != SSTOP && p->p_stat != SCORE) ||
1312 (p->p_flags & P_WAITED) != 0 ||
1313 ((p->p_flags & P_TRACED) == 0 &&
1314 (options & WUNTRACED) == 0)) {
1315 lwkt_reltoken(&p->p_token);
1321 * Don't set P_WAITED if WNOWAIT specified, leaving
1322 * the process in a waitable state.
1324 if ((options & WNOWAIT) == 0)
1325 p->p_flags |= P_WAITED;
1328 *status = W_STOPCODE(p->p_xstat);
1329 /* Zero rusage so we get something consistent. */
1330 bzero(wrusage, sizeof(*wrusage));
1333 bzero(info, sizeof(*info));
1334 if (p->p_flags & P_TRACED)
1335 info->si_code = CLD_TRAPPED;
1337 info->si_code = CLD_STOPPED;
1338 info->si_status = WSTOPSIG(p->p_xstat);
1340 lwkt_reltoken(&p->p_token);
1344 if ((options & WCONTINUED) && (p->p_flags & P_CONTINUED)) {
1345 lwkt_gettoken(&p->p_token);
1346 if (p->p_pptr != q) {
1347 lwkt_reltoken(&p->p_token);
1351 if ((p->p_flags & P_CONTINUED) == 0) {
1352 lwkt_reltoken(&p->p_token);
1360 * Don't set P_WAITED if WNOWAIT specified, leaving
1361 * the process in a waitable state.
1363 if ((options & WNOWAIT) == 0)
1364 p->p_flags &= ~P_CONTINUED;
1369 bzero(info, sizeof(*info));
1370 info->si_code = CLD_CONTINUED;
1371 info->si_status = WSTOPSIG(p->p_xstat);
1373 lwkt_reltoken(&p->p_token);
1383 if (options & WNOHANG) {
1390 * Wait for signal - interlocked using q->p_waitgen.
1393 while ((waitgen & 0x7FFFFFFF) == (q->p_waitgen & 0x7FFFFFFF)) {
1394 tsleep_interlock(q, PCATCH);
1395 waitgen = atomic_fetchadd_long(&q->p_waitgen, 0x80000000);
1396 if ((waitgen & 0x7FFFFFFF) == (q->p_waitgen & 0x7FFFFFFF)) {
1397 error = tsleep(q, PCATCH | PINTERLOCKED, "wait", 0);
1403 lwkt_reltoken(&q->p_token);
1410 * Change child's parent process to parent.
1412 * p_children/p_sibling requires the parent's token, and
1413 * changing pptr requires the child's token, so we have to
1414 * get three tokens to do this operation. We also need to
1415 * hold pointers that might get ripped out from under us to
1416 * preserve structural integrity.
1418 * It is possible to race another reparent or disconnect or other
1419 * similar operation. We must retry when this situation occurs.
1420 * Once we successfully reparent the process we no longer care
1424 proc_reparent(struct proc *child, struct proc *parent)
1429 while ((opp = child->p_pptr) != parent) {
1431 lwkt_gettoken(&opp->p_token);
1432 lwkt_gettoken(&child->p_token);
1433 lwkt_gettoken(&parent->p_token);
1434 if (child->p_pptr != opp) {
1435 lwkt_reltoken(&parent->p_token);
1436 lwkt_reltoken(&child->p_token);
1437 lwkt_reltoken(&opp->p_token);
1441 LIST_REMOVE(child, p_sibling);
1442 LIST_INSERT_HEAD(&parent->p_children, child, p_sibling);
1443 child->p_pptr = parent;
1444 child->p_ppid = parent->p_pid;
1445 lwkt_reltoken(&parent->p_token);
1446 lwkt_reltoken(&child->p_token);
1447 lwkt_reltoken(&opp->p_token);
1448 if (LIST_EMPTY(&opp->p_children))
1457 * The next two functions are to handle adding/deleting items on the
1461 * Take the arguments given and put them onto the exit callout list,
1462 * However first make sure that it's not already there.
1463 * returns 0 on success.
1467 at_exit(exitlist_fn function)
1469 struct exitlist *ep;
1472 /* Be noisy if the programmer has lost track of things */
1473 if (rm_at_exit(function))
1474 kprintf("WARNING: exit callout entry (%p) already present\n",
1477 ep = kmalloc(sizeof(*ep), M_ATEXIT, M_NOWAIT);
1480 ep->function = function;
1481 TAILQ_INSERT_TAIL(&exit_list, ep, next);
1486 * Scan the exit callout list for the given item and remove it.
1487 * Returns the number of items removed (0 or 1)
1490 rm_at_exit(exitlist_fn function)
1492 struct exitlist *ep;
1494 TAILQ_FOREACH(ep, &exit_list, next) {
1495 if (ep->function == function) {
1496 TAILQ_REMOVE(&exit_list, ep, next);
1497 kfree(ep, M_ATEXIT);
1505 * LWP reaper related code.
1508 reaplwps(void *context, int dummy)
1510 struct lwplist *lwplist = context;
1514 lwkt_gettoken(&deadlwp_token[cpu]);
1515 while ((lp = LIST_FIRST(lwplist))) {
1516 LIST_REMOVE(lp, u.lwp_reap_entry);
1519 lwkt_reltoken(&deadlwp_token[cpu]);
1523 reaplwp(struct lwp *lp)
1525 while (lwp_wait(lp) == 0)
1535 for (cpu = 0; cpu < ncpus; cpu++) {
1536 lwkt_token_init(&deadlwp_token[cpu], "deadlwpl");
1537 LIST_INIT(&deadlwp_list[cpu]);
1538 deadlwp_task[cpu] = kmalloc(sizeof(*deadlwp_task[cpu]),
1539 M_DEVBUF, M_WAITOK);
1540 TASK_INIT(deadlwp_task[cpu], 0, reaplwps, &deadlwp_list[cpu]);
1544 SYSINIT(deadlwpinit, SI_SUB_CONFIGURE, SI_ORDER_ANY, deadlwp_init, NULL);