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_compat.h"
39 #include "opt_ktrace.h"
41 #include <sys/param.h>
42 #include <sys/systm.h>
43 #include <sys/sysproto.h>
44 #include <sys/kernel.h>
45 #include <sys/malloc.h>
47 #include <sys/ktrace.h>
48 #include <sys/pioctl.h>
51 #include <sys/vnode.h>
52 #include <sys/resourcevar.h>
53 #include <sys/signalvar.h>
54 #include <sys/taskqueue.h>
55 #include <sys/ptrace.h>
56 #include <sys/acct.h> /* for acct_process() function prototype */
57 #include <sys/filedesc.h>
61 #include <sys/kern_syscall.h>
62 #include <sys/unistd.h>
63 #include <sys/eventhandler.h>
64 #include <sys/dsched.h>
67 #include <vm/vm_param.h>
70 #include <vm/vm_map.h>
71 #include <vm/vm_extern.h>
74 #include <sys/refcount.h>
75 #include <sys/thread2.h>
76 #include <sys/sysref2.h>
77 #include <sys/mplock2.h>
79 #include <machine/vmm.h>
81 static void reaplwps(void *context, int dummy);
82 static void reaplwp(struct lwp *lp);
83 static void killlwps(struct lwp *lp);
85 static MALLOC_DEFINE(M_ATEXIT, "atexit", "atexit callback");
88 * callout list for things to do at exit time
92 TAILQ_ENTRY(exitlist) next;
95 TAILQ_HEAD(exit_list_head, exitlist);
96 static struct exit_list_head exit_list = TAILQ_HEAD_INITIALIZER(exit_list);
101 static struct task *deadlwp_task[MAXCPU];
102 static struct lwplist deadlwp_list[MAXCPU];
103 static struct lwkt_token deadlwp_token[MAXCPU];
109 * SYS_EXIT_ARGS(int rval)
112 sys_exit(struct exit_args *uap)
114 exit1(W_EXITCODE(uap->rval, 0));
120 * Death of a lwp or process with optional bells and whistles.
123 sys_extexit(struct extexit_args *uap)
125 struct proc *p = curproc;
129 action = EXTEXIT_ACTION(uap->how);
130 who = EXTEXIT_WHO(uap->how);
132 /* Check parameters before we might perform some action */
145 error = copyout(&uap->status, uap->addr, sizeof(uap->status));
153 lwkt_gettoken(&p->p_token);
158 * Be sure only to perform a simple lwp exit if there is at
159 * least one more lwp in the proc, which will call exit1()
160 * later, otherwise the proc will be an UNDEAD and not even a
163 if (p->p_nthreads > 1) {
164 lwp_exit(0, NULL); /* called w/ p_token held */
167 /* else last lwp in proc: do the real thing */
169 default: /* to help gcc */
171 lwkt_reltoken(&p->p_token);
172 exit1(W_EXITCODE(uap->status, 0));
177 lwkt_reltoken(&p->p_token); /* safety */
181 * Kill all lwps associated with the current process except the
182 * current lwp. Return an error if we race another thread trying to
183 * do the same thing and lose the race.
185 * If forexec is non-zero the current thread and process flags are
186 * cleaned up so they can be reused.
188 * Caller must hold curproc->p_token
191 killalllwps(int forexec)
193 struct lwp *lp = curthread->td_lwp;
194 struct proc *p = lp->lwp_proc;
198 * Interlock against P_WEXIT. Only one of the process's thread
199 * is allowed to do the master exit.
201 if (p->p_flags & P_WEXIT)
203 p->p_flags |= P_WEXIT;
206 * Set temporary stopped state in case we are racing a coredump.
207 * Otherwise the coredump may hang forever.
209 if (lp->lwp_mpflags & LWP_MP_WSTOP) {
212 atomic_set_int(&lp->lwp_mpflags, LWP_MP_WSTOP);
215 wakeup(&p->p_nstopped);
219 * Interlock with LWP_MP_WEXIT and kill any remaining LWPs
221 atomic_set_int(&lp->lwp_mpflags, LWP_MP_WEXIT);
222 if (p->p_nthreads > 1)
226 * Undo temporary stopped state
229 atomic_clear_int(&lp->lwp_mpflags, LWP_MP_WSTOP);
234 * If doing this for an exec, clean up the remaining thread
235 * (us) for continuing operation after all the other threads
239 atomic_clear_int(&lp->lwp_mpflags, LWP_MP_WEXIT);
240 p->p_flags &= ~P_WEXIT;
246 * Kill all LWPs except the current one. Do not try to signal
247 * LWPs which have exited on their own or have already been
251 killlwps(struct lwp *lp)
253 struct proc *p = lp->lwp_proc;
257 * Kill the remaining LWPs. We must send the signal before setting
258 * LWP_MP_WEXIT. The setting of WEXIT is optional but helps reduce
259 * races. tlp must be held across the call as it might block and
260 * allow the target lwp to rip itself out from under our loop.
262 FOREACH_LWP_IN_PROC(tlp, p) {
264 lwkt_gettoken(&tlp->lwp_token);
265 if ((tlp->lwp_mpflags & LWP_MP_WEXIT) == 0) {
266 lwpsignal(p, tlp, SIGKILL);
267 atomic_set_int(&tlp->lwp_mpflags, LWP_MP_WEXIT);
269 lwkt_reltoken(&tlp->lwp_token);
274 * Wait for everything to clear out. Also make sure any tstop()s
275 * are signalled (we are holding p_token for the interlock).
278 while (p->p_nthreads > 1)
279 tsleep(&p->p_nthreads, 0, "killlwps", 0);
283 * Exit: deallocate address space and other resources, change proc state
284 * to zombie, and unlink proc from allproc and parent's lists. Save exit
285 * status and rusage for wait(). Check for child processes and orphan them.
290 struct thread *td = curthread;
291 struct proc *p = td->td_proc;
292 struct lwp *lp = td->td_lwp;
296 struct sysreaper *reap;
302 lwkt_gettoken(&p->p_token);
305 kprintf("init died (signal %d, exit %d)\n",
306 WTERMSIG(rv), WEXITSTATUS(rv));
307 panic("Going nowhere without my init!");
309 varsymset_clean(&p->p_varsymset);
310 lockuninit(&p->p_varsymset.vx_lock);
313 * Kill all lwps associated with the current process, return an
314 * error if we race another thread trying to do the same thing
317 error = killalllwps(0);
323 /* are we a task leader? */
324 if (p == p->p_leader) {
325 struct kill_args killArgs;
326 killArgs.signum = SIGKILL;
329 killArgs.pid = q->p_pid;
331 * The interface for kill is better
332 * than the internal signal
338 tsleep((caddr_t)p, 0, "exit1", 0);
344 STOPEVENT(p, S_EXIT, rv);
345 p->p_flags |= P_POSTEXIT; /* stop procfs stepping */
348 * Check if any loadable modules need anything done at process exit.
349 * e.g. SYSV IPC stuff
350 * XXX what if one of these generates an error?
353 EVENTHANDLER_INVOKE(process_exit, p);
356 * XXX: imho, the eventhandler stuff is much cleaner than this.
357 * Maybe we should move everything to use eventhandler.
359 TAILQ_FOREACH(ep, &exit_list, next)
362 if (p->p_flags & P_PROFIL)
365 SIGEMPTYSET(p->p_siglist);
366 SIGEMPTYSET(lp->lwp_siglist);
367 if (timevalisset(&p->p_realtimer.it_value))
368 callout_stop_sync(&p->p_ithandle);
371 * Reset any sigio structures pointing to us as a result of
372 * F_SETOWN with our pid.
374 funsetownlst(&p->p_sigiolst);
377 * Close open files and release open-file table.
382 if (p->p_leader->p_peers) {
384 while(q->p_peers != p)
386 q->p_peers = p->p_peers;
387 wakeup((caddr_t)p->p_leader);
391 * XXX Shutdown SYSV semaphores
395 KKASSERT(p->p_numposixlocks == 0);
397 /* The next two chunks should probably be moved to vmspace_exit. */
401 * Clean up data related to virtual kernel operation. Clean up
402 * any vkernel context related to the current lwp now so we can
406 vkernel_lwp_exit(lp);
411 * Release the user portion of address space. The exitbump prevents
412 * the vmspace from being completely eradicated (using holdcnt).
413 * This releases references to vnodes, which could cause I/O if the
414 * file has been unlinked. We need to do this early enough that
415 * we can still sleep.
417 * We can't free the entire vmspace as the kernel stack may be mapped
418 * within that space also.
420 * Processes sharing the same vmspace may exit in one order, and
421 * get cleaned up by vmspace_exit() in a different order. The
422 * last exiting process to reach this point releases as much of
423 * the environment as it can, and the last process cleaned up
424 * by vmspace_exit() (which decrements exitingcnt) cleans up the
427 * NOTE: Releasing p_token around this call is helpful if the
428 * vmspace had a huge RSS. Otherwise some other process
429 * trying to do an allproc or other scan (like 'ps') may
430 * stall for a long time.
432 lwkt_reltoken(&p->p_token);
434 lwkt_gettoken(&p->p_token);
436 if (SESS_LEADER(p)) {
437 struct session *sp = p->p_session;
441 * We are the controlling process. Signal the
442 * foreground process group, drain the controlling
443 * terminal, and revoke access to the controlling
446 * NOTE: while waiting for the process group to exit
447 * it is possible that one of the processes in the
448 * group will revoke the tty, so the ttyclosesession()
449 * function will re-check sp->s_ttyvp.
451 if (sp->s_ttyp && (sp->s_ttyp->t_session == sp)) {
452 if (sp->s_ttyp->t_pgrp)
453 pgsignal(sp->s_ttyp->t_pgrp, SIGHUP, 1);
455 ttyclosesession(sp, 1); /* also revoke */
458 * Release the tty. If someone has it open via
459 * /dev/tty then close it (since they no longer can
460 * once we've NULL'd it out).
462 ttyclosesession(sp, 0);
465 * s_ttyp is not zero'd; we use this to indicate
466 * that the session once had a controlling terminal.
467 * (for logging and informational purposes)
472 fixjobc(p, p->p_pgrp, 0);
473 (void)acct_process(p);
479 ktrdestroy(&p->p_tracenode);
483 * Release reference to text vnode
485 if ((vtmp = p->p_textvp) != NULL) {
490 /* Release namecache handle to text file */
491 if (p->p_textnch.ncp)
492 cache_drop(&p->p_textnch);
495 * We have to handle PPWAIT here or proc_move_allproc_zombie()
496 * will block on the PHOLD() the parent is doing.
498 * We are using the flag as an interlock so an atomic op is
499 * necessary to synchronize with the parent's cpu.
501 if (p->p_flags & P_PPWAIT) {
502 if (p->p_pptr && p->p_pptr->p_upmap)
503 atomic_add_int(&p->p_pptr->p_upmap->invfork, -1);
504 atomic_clear_int(&p->p_flags, P_PPWAIT);
509 * Move the process to the zombie list. This will block
510 * until the process p_lock count reaches 0. The process will
511 * not be reaped until TDF_EXITING is set by cpu_thread_exit(),
512 * which is called from cpu_proc_exit().
514 * Interlock against waiters using p_waitgen. We increment
515 * p_waitgen after completing the move of our process to the
518 * WARNING: pp becomes stale when we block, clear it now as a
521 proc_move_allproc_zombie(p);
523 atomic_add_long(&pp->p_waitgen, 1);
527 * release controlled reaper for exit if we own it and return the
528 * remaining reaper (the one for us), which we will drop after we
531 reap = reaper_exit(p);
534 * Reparent all of this process's children to the init process or
535 * to the designated reaper. We must hold the reaper's p_token in
536 * order to safely mess with p_children.
538 * We already hold p->p_token (to remove the children from our list).
541 q = LIST_FIRST(&p->p_children);
543 reproc = reaper_get(reap);
544 lwkt_gettoken(&reproc->p_token);
545 while ((q = LIST_FIRST(&p->p_children)) != NULL) {
547 lwkt_gettoken(&q->p_token);
548 if (q != LIST_FIRST(&p->p_children)) {
549 lwkt_reltoken(&q->p_token);
553 LIST_REMOVE(q, p_sibling);
554 LIST_INSERT_HEAD(&reproc->p_children, q, p_sibling);
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, adding in child rusage
575 * info and self times.
577 calcru_proc(p, &p->p_ru);
578 ruadd(&p->p_ru, &p->p_cru);
581 * notify interested parties of our demise.
583 KNOTE(&p->p_klist, NOTE_EXIT);
586 * Notify parent that we're gone. If parent has the PS_NOCLDWAIT
587 * flag set, or if the handler is set to SIG_IGN, notify the reaper
588 * instead (it will handle this situation).
590 * NOTE: The reaper can still be the parent process.
594 if (p->p_pptr->p_sigacts->ps_flag & (PS_NOCLDWAIT | PS_CLDSIGIGN)) {
596 reproc = reaper_get(reap);
597 proc_reparent(p, reproc);
605 * Signal (possibly new) parent.
609 if (p->p_sigparent && pp != initproc) {
610 int sig = p->p_sigparent;
612 if (sig != SIGUSR1 && sig != SIGCHLD)
616 ksignal(pp, SIGCHLD);
618 p->p_flags &= ~P_TRACED;
622 * cpu_exit is responsible for clearing curproc, since
623 * it is heavily integrated with the thread/switching sequence.
625 * Other substructures are freed from wait().
630 * Finally, call machine-dependent code to release as many of the
631 * lwp's resources as we can and halt execution of this thread.
633 * pp is a wild pointer now but still the correct wakeup() target.
634 * lwp_exit() only uses it to send the wakeup() signal to the likely
635 * parent. Any reparenting race that occurs will get a signal
636 * automatically and not be an issue.
642 * Eventually called by every exiting LWP
644 * p->p_token must be held. mplock may be held and will be released.
647 lwp_exit(int masterexit, void *waddr)
649 struct thread *td = curthread;
650 struct lwp *lp = td->td_lwp;
651 struct proc *p = lp->lwp_proc;
655 * Release the current user process designation on the process so
656 * the userland scheduler can work in someone else.
658 p->p_usched->release_curproc(lp);
661 * lwp_exit() may be called without setting LWP_MP_WEXIT, so
662 * make sure it is set here.
664 ASSERT_LWKT_TOKEN_HELD(&p->p_token);
665 atomic_set_int(&lp->lwp_mpflags, LWP_MP_WEXIT);
668 * Clean up any virtualization
671 vkernel_lwp_exit(lp);
677 * Clean up select/poll support
679 kqueue_terminate(&lp->lwp_kqueue);
682 * Clean up any syscall-cached ucred
685 crfree(td->td_ucred);
690 * Nobody actually wakes us when the lock
691 * count reaches zero, so just wait one tick.
693 while (lp->lwp_lock > 0)
694 tsleep(lp, 0, "lwpexit", 1);
696 /* Hand down resource usage to our proc */
697 ruadd(&p->p_ru, &lp->lwp_ru);
700 * If we don't hold the process until the LWP is reaped wait*()
701 * may try to dispose of its vmspace before all the LWPs have
702 * actually terminated.
707 * Do any remaining work that might block on us. We should be
708 * coded such that further blocking is ok after decrementing
709 * p_nthreads but don't take the chance.
711 dsched_exit_thread(td);
712 biosched_done(curthread);
715 * We have to use the reaper for all the LWPs except the one doing
716 * the master exit. The LWP doing the master exit can just be
717 * left on p_lwps and the process reaper will deal with it
718 * synchronously, which is much faster.
720 * Wakeup anyone waiting on p_nthreads to drop to 1 or 0.
722 * The process is left held until the reaper calls lwp_dispose() on
723 * the lp (after calling lwp_wait()).
725 if (masterexit == 0) {
728 lwp_rb_tree_RB_REMOVE(&p->p_lwp_tree, lp);
730 if ((p->p_flags & P_MAYBETHREADED) && p->p_nthreads <= 1)
732 lwkt_gettoken(&deadlwp_token[cpu]);
733 LIST_INSERT_HEAD(&deadlwp_list[cpu], lp, u.lwp_reap_entry);
734 taskqueue_enqueue(taskqueue_thread[cpu], deadlwp_task[cpu]);
735 lwkt_reltoken(&deadlwp_token[cpu]);
738 if ((p->p_flags & P_MAYBETHREADED) && p->p_nthreads <= 1)
743 * We no longer need p_token.
745 * Tell the userland scheduler that we are going away
747 lwkt_reltoken(&p->p_token);
748 p->p_usched->heuristic_exiting(lp, p);
751 * Issue late wakeups after releasing our token to give us a chance
752 * to deschedule and switch away before another cpu in a wait*()
753 * reaps us. This is done as late as possible to reduce contention.
756 wakeup(&p->p_nthreads);
764 * Wait until a lwp is completely dead. The final interlock in this drama
765 * is when TDF_EXITING is set in cpu_thread_exit() just before the final
768 * At the point TDF_EXITING is set a complete exit is accomplished when
769 * TDF_RUNNING and TDF_PREEMPT_LOCK are both clear. td_mpflags has two
770 * post-switch interlock flags that can be used to wait for the TDF_
773 * Returns non-zero on success, and zero if the caller needs to retry
777 lwp_wait(struct lwp *lp)
779 struct thread *td = lp->lwp_thread;
782 KKASSERT(lwkt_preempted_proc() != lp);
785 * This bit of code uses the thread destruction interlock
786 * managed by lwkt_switch_return() to wait for the lwp's
787 * thread to completely disengage.
789 * It is possible for us to race another cpu core so we
790 * have to do this correctly.
793 mpflags = td->td_mpflags;
795 if (mpflags & TDF_MP_EXITSIG)
797 tsleep_interlock(td, 0);
798 if (atomic_cmpset_int(&td->td_mpflags, mpflags,
799 mpflags | TDF_MP_EXITWAIT)) {
800 tsleep(td, PINTERLOCKED, "lwpxt", 0);
805 * We've already waited for the core exit but there can still
806 * be other refs from e.g. process scans and such.
808 if (lp->lwp_lock > 0) {
809 tsleep(lp, 0, "lwpwait1", 1);
813 tsleep(td, 0, "lwpwait2", 1);
818 * Now that we have the thread destruction interlock these flags
819 * really should already be cleaned up, keep a check for safety.
821 * We can't rip its stack out from under it until TDF_EXITING is
822 * set and both TDF_RUNNING and TDF_PREEMPT_LOCK are clear.
823 * TDF_PREEMPT_LOCK must be checked because TDF_RUNNING
824 * will be cleared temporarily if a thread gets preempted.
826 while ((td->td_flags & (TDF_RUNNING |
829 TDF_EXITING)) != TDF_EXITING) {
830 tsleep(lp, 0, "lwpwait3", 1);
834 KASSERT((td->td_flags & (TDF_RUNQ|TDF_TSLEEPQ)) == 0,
835 ("lwp_wait: td %p (%s) still on run or sleep queue",
841 * Release the resources associated with a lwp.
842 * The lwp must be completely dead.
845 lwp_dispose(struct lwp *lp)
847 struct thread *td = lp->lwp_thread;
849 KKASSERT(lwkt_preempted_proc() != lp);
850 KKASSERT(lp->lwp_lock == 0);
851 KKASSERT(td->td_refs == 0);
852 KKASSERT((td->td_flags & (TDF_RUNNING |
855 TDF_EXITING)) == TDF_EXITING);
862 lp->lwp_thread = NULL;
863 lwkt_free_thread(td);
869 sys_wait4(struct wait_args *uap)
871 struct rusage rusage;
874 error = kern_wait(uap->pid, (uap->status ? &status : NULL),
875 uap->options, (uap->rusage ? &rusage : NULL),
876 &uap->sysmsg_result);
878 if (error == 0 && uap->status)
879 error = copyout(&status, uap->status, sizeof(*uap->status));
880 if (error == 0 && uap->rusage)
881 error = copyout(&rusage, uap->rusage, sizeof(*uap->rusage));
888 * wait_args(int pid, int *status, int options, struct rusage *rusage)
891 kern_wait(pid_t pid, int *status, int options, struct rusage *rusage, int *res)
893 struct thread *td = curthread;
895 struct proc *q = td->td_proc;
904 if (options &~ (WUNTRACED|WNOHANG|WCONTINUED|WLINUXCLONE))
908 * Protect the q->p_children list
910 lwkt_gettoken(&q->p_token);
913 * All sorts of things can change due to blocking so we have to loop
914 * all the way back up here.
916 * The problem is that if a process group is stopped and the parent
917 * is doing a wait*(..., WUNTRACED, ...), it will see the STOP
918 * of the child and then stop itself when it tries to return from the
919 * system call. When the process group is resumed the parent will
920 * then get the STOP status even though the child has now resumed
921 * (a followup wait*() will get the CONT status).
923 * Previously the CONT would overwrite the STOP because the tstop
924 * was handled within tsleep(), and the parent would only see
925 * the CONT when both are stopped and continued together. This little
926 * two-line hack restores this effect.
928 while (q->p_stat == SSTOP || q->p_stat == SCORE)
936 * NOTE: We don't want to break q's p_token in the loop for the
937 * case where no children are found or we risk breaking the
938 * interlock between child and parent.
940 waitgen = atomic_fetchadd_long(&q->p_waitgen, 0x80000000);
941 LIST_FOREACH(p, &q->p_children, p_sibling) {
942 if (pid != WAIT_ANY &&
943 p->p_pid != pid && p->p_pgid != -pid) {
948 * This special case handles a kthread spawned by linux_clone
949 * (see linux_misc.c). The linux_wait4 and linux_waitpid
950 * functions need to be able to distinguish between waiting
951 * on a process and waiting on a thread. It is a thread if
952 * p_sigparent is not SIGCHLD, and the WLINUXCLONE option
953 * signifies we want to wait for threads and not processes.
955 if ((p->p_sigparent != SIGCHLD) ^
956 ((options & WLINUXCLONE) != 0)) {
961 if (p->p_stat == SZOMB) {
963 * We may go into SZOMB with threads still present.
964 * We must wait for them to exit before we can reap
965 * the master thread, otherwise we may race reaping
966 * non-master threads.
968 * Only this routine can remove a process from
969 * the zombie list and destroy it, use PACQUIREZOMB()
970 * to serialize us and loop if it blocks (interlocked
971 * by the parent's q->p_token).
973 * WARNING! (p) can be invalid when PHOLDZOMB(p)
974 * returns non-zero. Be sure not to
979 lwkt_gettoken(&p->p_token);
980 if (p->p_pptr != q) {
981 lwkt_reltoken(&p->p_token);
985 while (p->p_nthreads > 0) {
986 tsleep(&p->p_nthreads, 0, "lwpzomb", hz);
990 * Reap any LWPs left in p->p_lwps. This is usually
991 * just the last LWP. This must be done before
992 * we loop on p_lock since the lwps hold a ref on
993 * it as a vmspace interlock.
995 * Once that is accomplished p_nthreads had better
998 while ((lp = RB_ROOT(&p->p_lwp_tree)) != NULL) {
1000 * Make sure no one is using this lwp, before
1001 * it is removed from the tree. If we didn't
1002 * wait it here, lwp tree iteration with
1003 * blocking operation would be broken.
1005 while (lp->lwp_lock > 0)
1006 tsleep(lp, 0, "zomblwp", 1);
1007 lwp_rb_tree_RB_REMOVE(&p->p_lwp_tree, lp);
1010 KKASSERT(p->p_nthreads == 0);
1013 * Don't do anything really bad until all references
1014 * to the process go away. This may include other
1015 * LWPs which are still in the process of being
1016 * reaped. We can't just pull the rug out from under
1017 * them because they may still be using the VM space.
1019 * Certain kernel facilities such as /proc will also
1020 * put a hold on the process for short periods of
1024 PSTALL(p, "reap3", 0);
1026 /* Take care of our return values. */
1030 *status = p->p_xstat;
1035 * If we got the child via a ptrace 'attach',
1036 * we need to give it back to the old parent.
1038 if (p->p_oppid && (t = pfind(p->p_oppid)) != NULL) {
1041 proc_reparent(p, t);
1042 ksignal(t, SIGCHLD);
1046 lwkt_reltoken(&p->p_token);
1052 * Unlink the proc from its process group so that
1053 * the following operations won't lead to an
1054 * inconsistent state for processes running down
1057 proc_remove_zombie(p);
1059 lwkt_reltoken(&p->p_token);
1063 ruadd(&q->p_cru, &p->p_ru);
1066 * Decrement the count of procs running with this uid.
1068 chgproccnt(p->p_ucred->cr_ruidinfo, -1, 0);
1071 * Free up credentials.
1077 * Remove unused arguments
1081 if (pa && refcount_release(&pa->ar_ref)) {
1087 p->p_sigacts = NULL;
1088 if (ps && refcount_release(&ps->ps_refcnt)) {
1089 kfree(ps, M_SUBPROC);
1094 * Our exitingcount was incremented when the process
1095 * became a zombie, now that the process has been
1096 * removed from (almost) all lists we should be able
1097 * to safely destroy its vmspace. Wait for any current
1098 * holders to go away (so the vmspace remains stable),
1101 * NOTE: Releasing the parent process (q) p_token
1102 * across the vmspace_exitfree() call is
1103 * important here to reduce stalls on
1104 * interactions with (q) (such as
1105 * fork/exec/wait or 'ps').
1107 PSTALL(p, "reap4", 0);
1108 lwkt_reltoken(&q->p_token);
1109 vmspace_exitfree(p);
1110 lwkt_gettoken(&q->p_token);
1111 PSTALL(p, "reap5", 0);
1114 * NOTE: We have to officially release ZOMB in order
1115 * to ensure that a racing thread in kern_wait()
1116 * which blocked on ZOMB is woken up.
1121 atomic_add_int(&nprocs, -1);
1125 if ((p->p_stat == SSTOP || p->p_stat == SCORE) &&
1126 (p->p_flags & P_WAITED) == 0 &&
1127 ((p->p_flags & P_TRACED) || (options & WUNTRACED))) {
1129 lwkt_gettoken(&p->p_token);
1130 if (p->p_pptr != q) {
1131 lwkt_reltoken(&p->p_token);
1135 if ((p->p_stat != SSTOP && p->p_stat != SCORE) ||
1136 (p->p_flags & P_WAITED) != 0 ||
1137 ((p->p_flags & P_TRACED) == 0 &&
1138 (options & WUNTRACED) == 0)) {
1139 lwkt_reltoken(&p->p_token);
1144 p->p_flags |= P_WAITED;
1148 *status = W_STOPCODE(p->p_xstat);
1149 /* Zero rusage so we get something consistent. */
1151 bzero(rusage, sizeof(*rusage));
1153 lwkt_reltoken(&p->p_token);
1157 if ((options & WCONTINUED) && (p->p_flags & P_CONTINUED)) {
1159 lwkt_gettoken(&p->p_token);
1160 if (p->p_pptr != q) {
1161 lwkt_reltoken(&p->p_token);
1165 if ((p->p_flags & P_CONTINUED) == 0) {
1166 lwkt_reltoken(&p->p_token);
1172 p->p_flags &= ~P_CONTINUED;
1177 lwkt_reltoken(&p->p_token);
1186 if (options & WNOHANG) {
1193 * Wait for signal - interlocked using q->p_waitgen.
1196 while ((waitgen & 0x7FFFFFFF) == (q->p_waitgen & 0x7FFFFFFF)) {
1197 tsleep_interlock(q, PCATCH);
1198 waitgen = atomic_fetchadd_long(&q->p_waitgen, 0x80000000);
1199 if ((waitgen & 0x7FFFFFFF) == (q->p_waitgen & 0x7FFFFFFF)) {
1200 error = tsleep(q, PCATCH | PINTERLOCKED, "wait", 0);
1206 lwkt_reltoken(&q->p_token);
1213 * Change child's parent process to parent.
1215 * p_children/p_sibling requires the parent's token, and
1216 * changing pptr requires the child's token, so we have to
1217 * get three tokens to do this operation. We also need to
1218 * hold pointers that might get ripped out from under us to
1219 * preserve structural integrity.
1221 * It is possible to race another reparent or disconnect or other
1222 * similar operation. We must retry when this situation occurs.
1223 * Once we successfully reparent the process we no longer care
1227 proc_reparent(struct proc *child, struct proc *parent)
1232 while ((opp = child->p_pptr) != parent) {
1234 lwkt_gettoken(&opp->p_token);
1235 lwkt_gettoken(&child->p_token);
1236 lwkt_gettoken(&parent->p_token);
1237 if (child->p_pptr != opp) {
1238 lwkt_reltoken(&parent->p_token);
1239 lwkt_reltoken(&child->p_token);
1240 lwkt_reltoken(&opp->p_token);
1244 LIST_REMOVE(child, p_sibling);
1245 LIST_INSERT_HEAD(&parent->p_children, child, p_sibling);
1246 child->p_pptr = parent;
1247 lwkt_reltoken(&parent->p_token);
1248 lwkt_reltoken(&child->p_token);
1249 lwkt_reltoken(&opp->p_token);
1250 if (LIST_EMPTY(&opp->p_children))
1259 * The next two functions are to handle adding/deleting items on the
1263 * Take the arguments given and put them onto the exit callout list,
1264 * However first make sure that it's not already there.
1265 * returns 0 on success.
1269 at_exit(exitlist_fn function)
1271 struct exitlist *ep;
1274 /* Be noisy if the programmer has lost track of things */
1275 if (rm_at_exit(function))
1276 kprintf("WARNING: exit callout entry (%p) already present\n",
1279 ep = kmalloc(sizeof(*ep), M_ATEXIT, M_NOWAIT);
1282 ep->function = function;
1283 TAILQ_INSERT_TAIL(&exit_list, ep, next);
1288 * Scan the exit callout list for the given item and remove it.
1289 * Returns the number of items removed (0 or 1)
1292 rm_at_exit(exitlist_fn function)
1294 struct exitlist *ep;
1296 TAILQ_FOREACH(ep, &exit_list, next) {
1297 if (ep->function == function) {
1298 TAILQ_REMOVE(&exit_list, ep, next);
1299 kfree(ep, M_ATEXIT);
1307 * LWP reaper related code.
1310 reaplwps(void *context, int dummy)
1312 struct lwplist *lwplist = context;
1316 lwkt_gettoken(&deadlwp_token[cpu]);
1317 while ((lp = LIST_FIRST(lwplist))) {
1318 LIST_REMOVE(lp, u.lwp_reap_entry);
1321 lwkt_reltoken(&deadlwp_token[cpu]);
1325 reaplwp(struct lwp *lp)
1327 while (lwp_wait(lp) == 0)
1337 for (cpu = 0; cpu < ncpus; cpu++) {
1338 lwkt_token_init(&deadlwp_token[cpu], "deadlwpl");
1339 LIST_INIT(&deadlwp_list[cpu]);
1340 deadlwp_task[cpu] = kmalloc(sizeof(*deadlwp_task[cpu]),
1341 M_DEVBUF, M_WAITOK);
1342 TASK_INIT(deadlwp_task[cpu], 0, reaplwps, &deadlwp_list[cpu]);
1346 SYSINIT(deadlwpinit, SI_SUB_CONFIGURE, SI_ORDER_ANY, deadlwp_init, NULL);