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>
73 #include <sys/refcount.h>
74 #include <sys/thread2.h>
75 #include <sys/spinlock2.h>
76 #include <sys/mplock2.h>
78 #include <machine/vmm.h>
80 static void reaplwps(void *context, int dummy);
81 static void reaplwp(struct lwp *lp);
82 static void killlwps(struct lwp *lp);
84 static MALLOC_DEFINE(M_ATEXIT, "atexit", "atexit callback");
87 * callout list for things to do at exit time
91 TAILQ_ENTRY(exitlist) next;
94 TAILQ_HEAD(exit_list_head, exitlist);
95 static struct exit_list_head exit_list = TAILQ_HEAD_INITIALIZER(exit_list);
100 static struct task *deadlwp_task[MAXCPU];
101 static struct lwplist deadlwp_list[MAXCPU];
102 static struct lwkt_token deadlwp_token[MAXCPU];
108 * SYS_EXIT_ARGS(int rval)
111 sys_exit(struct exit_args *uap)
113 exit1(W_EXITCODE(uap->rval, 0));
119 * Death of a lwp or process with optional bells and whistles.
122 sys_extexit(struct extexit_args *uap)
124 struct proc *p = curproc;
128 action = EXTEXIT_ACTION(uap->how);
129 who = EXTEXIT_WHO(uap->how);
131 /* Check parameters before we might perform some action */
144 error = copyout(&uap->status, uap->addr, sizeof(uap->status));
152 lwkt_gettoken(&p->p_token);
157 * Be sure only to perform a simple lwp exit if there is at
158 * least one more lwp in the proc, which will call exit1()
159 * later, otherwise the proc will be an UNDEAD and not even a
162 if (p->p_nthreads > 1) {
163 lwp_exit(0, NULL); /* called w/ p_token held */
166 /* else last lwp in proc: do the real thing */
168 default: /* to help gcc */
170 lwkt_reltoken(&p->p_token);
171 exit1(W_EXITCODE(uap->status, 0));
176 lwkt_reltoken(&p->p_token); /* safety */
180 * Kill all lwps associated with the current process except the
181 * current lwp. Return an error if we race another thread trying to
182 * do the same thing and lose the race.
184 * If forexec is non-zero the current thread and process flags are
185 * cleaned up so they can be reused.
188 killalllwps(int forexec)
190 struct lwp *lp = curthread->td_lwp;
191 struct proc *p = lp->lwp_proc;
195 * Interlock against P_WEXIT. Only one of the process's thread
196 * is allowed to do the master exit.
198 lwkt_gettoken(&p->p_token);
199 if (p->p_flags & P_WEXIT) {
200 lwkt_reltoken(&p->p_token);
203 p->p_flags |= P_WEXIT;
204 lwkt_gettoken(&lp->lwp_token);
207 * Set temporary stopped state in case we are racing a coredump.
208 * Otherwise the coredump may hang forever.
210 if (lp->lwp_mpflags & LWP_MP_WSTOP) {
213 atomic_set_int(&lp->lwp_mpflags, LWP_MP_WSTOP);
216 wakeup(&p->p_nstopped);
220 * Interlock with LWP_MP_WEXIT and kill any remaining LWPs
222 atomic_set_int(&lp->lwp_mpflags, LWP_MP_WEXIT);
223 if (p->p_nthreads > 1)
227 * Undo temporary stopped state
230 atomic_clear_int(&lp->lwp_mpflags, LWP_MP_WSTOP);
235 * If doing this for an exec, clean up the remaining thread
236 * (us) for continuing operation after all the other threads
240 atomic_clear_int(&lp->lwp_mpflags, LWP_MP_WEXIT);
241 p->p_flags &= ~P_WEXIT;
243 lwkt_reltoken(&lp->lwp_token);
244 lwkt_reltoken(&p->p_token);
250 * Kill all LWPs except the current one. Do not try to signal
251 * LWPs which have exited on their own or have already been
255 killlwps(struct lwp *lp)
257 struct proc *p = lp->lwp_proc;
261 * Kill the remaining LWPs. We must send the signal before setting
262 * LWP_MP_WEXIT. The setting of WEXIT is optional but helps reduce
263 * races. tlp must be held across the call as it might block and
264 * allow the target lwp to rip itself out from under our loop.
266 FOREACH_LWP_IN_PROC(tlp, p) {
268 lwkt_gettoken(&tlp->lwp_token);
269 if ((tlp->lwp_mpflags & LWP_MP_WEXIT) == 0) {
270 atomic_set_int(&tlp->lwp_mpflags, LWP_MP_WEXIT);
271 lwpsignal(p, tlp, SIGKILL);
273 lwkt_reltoken(&tlp->lwp_token);
278 * Wait for everything to clear out. Also make sure any tstop()s
279 * are signalled (we are holding p_token for the interlock).
282 while (p->p_nthreads > 1)
283 tsleep(&p->p_nthreads, 0, "killlwps", 0);
287 * Exit: deallocate address space and other resources, change proc state
288 * to zombie, and unlink proc from allproc and parent's lists. Save exit
289 * status and rusage for wait(). Check for child processes and orphan them.
294 struct thread *td = curthread;
295 struct proc *p = td->td_proc;
296 struct lwp *lp = td->td_lwp;
300 struct sysreaper *reap;
306 lwkt_gettoken(&p->p_token);
309 kprintf("init died (signal %d, exit %d)\n",
310 WTERMSIG(rv), WEXITSTATUS(rv));
311 panic("Going nowhere without my init!");
313 varsymset_clean(&p->p_varsymset);
314 lockuninit(&p->p_varsymset.vx_lock);
317 * Kill all lwps associated with the current process, return an
318 * error if we race another thread trying to do the same thing
321 error = killalllwps(0);
327 /* are we a task leader? */
328 if (p == p->p_leader) {
329 struct kill_args killArgs;
330 killArgs.signum = SIGKILL;
333 killArgs.pid = q->p_pid;
335 * The interface for kill is better
336 * than the internal signal
342 tsleep((caddr_t)p, 0, "exit1", 0);
348 STOPEVENT(p, S_EXIT, rv);
349 p->p_flags |= P_POSTEXIT; /* stop procfs stepping */
352 * Check if any loadable modules need anything done at process exit.
353 * e.g. SYSV IPC stuff
354 * XXX what if one of these generates an error?
359 * XXX: imho, the eventhandler stuff is much cleaner than this.
360 * Maybe we should move everything to use eventhandler.
362 TAILQ_FOREACH(ep, &exit_list, next)
365 if (p->p_flags & P_PROFIL)
368 SIGEMPTYSET(p->p_siglist);
369 SIGEMPTYSET(lp->lwp_siglist);
370 if (timevalisset(&p->p_realtimer.it_value))
371 callout_stop_sync(&p->p_ithandle);
374 * Reset any sigio structures pointing to us as a result of
375 * F_SETOWN with our pid.
377 funsetownlst(&p->p_sigiolst);
380 * Close open files and release open-file table.
385 if (p->p_leader->p_peers) {
387 while(q->p_peers != p)
389 q->p_peers = p->p_peers;
390 wakeup((caddr_t)p->p_leader);
394 * XXX Shutdown SYSV semaphores
398 /* The next two chunks should probably be moved to vmspace_exit. */
402 * Clean up data related to virtual kernel operation. Clean up
403 * any vkernel context related to the current lwp now so we can
407 vkernel_lwp_exit(lp);
412 * Release the user portion of address space. The exitbump prevents
413 * the vmspace from being completely eradicated (using holdcnt).
414 * This releases references to vnodes, which could cause I/O if the
415 * file has been unlinked. We need to do this early enough that
416 * we can still sleep.
418 * We can't free the entire vmspace as the kernel stack may be mapped
419 * within that space also.
421 * Processes sharing the same vmspace may exit in one order, and
422 * get cleaned up by vmspace_exit() in a different order. The
423 * last exiting process to reach this point releases as much of
424 * the environment as it can, and the last process cleaned up
425 * by vmspace_exit() (which decrements exitingcnt) cleans up the
428 * NOTE: Releasing p_token around this call is helpful if the
429 * vmspace had a huge RSS. Otherwise some other process
430 * trying to do an allproc or other scan (like 'ps') may
431 * stall for a long time.
433 lwkt_reltoken(&p->p_token);
435 lwkt_gettoken(&p->p_token);
437 if (SESS_LEADER(p)) {
438 struct session *sp = p->p_session;
442 * We are the controlling process. Signal the
443 * foreground process group, drain the controlling
444 * terminal, and revoke access to the controlling
447 * NOTE: while waiting for the process group to exit
448 * it is possible that one of the processes in the
449 * group will revoke the tty, so the ttyclosesession()
450 * function will re-check sp->s_ttyvp.
452 if (sp->s_ttyp && (sp->s_ttyp->t_session == sp)) {
453 if (sp->s_ttyp->t_pgrp)
454 pgsignal(sp->s_ttyp->t_pgrp, SIGHUP, 1);
456 ttyclosesession(sp, 1); /* also revoke */
459 * Release the tty. If someone has it open via
460 * /dev/tty then close it (since they no longer can
461 * once we've NULL'd it out).
463 ttyclosesession(sp, 0);
466 * s_ttyp is not zero'd; we use this to indicate
467 * that the session once had a controlling terminal.
468 * (for logging and informational purposes)
473 fixjobc(p, p->p_pgrp, 0);
474 (void)acct_process(p);
480 ktrdestroy(&p->p_tracenode);
484 * Release reference to text vnode
486 if ((vtmp = p->p_textvp) != NULL) {
491 /* Release namecache handle to text file */
492 if (p->p_textnch.ncp)
493 cache_drop(&p->p_textnch);
496 * We have to handle PPWAIT here or proc_move_allproc_zombie()
497 * will block on the PHOLD() the parent is doing.
499 * We are using the flag as an interlock so an atomic op is
500 * necessary to synchronize with the parent's cpu.
502 if (p->p_flags & P_PPWAIT) {
503 if (p->p_pptr && p->p_pptr->p_upmap)
504 atomic_add_int(&p->p_pptr->p_upmap->invfork, -1);
505 atomic_clear_int(&p->p_flags, P_PPWAIT);
510 * Move the process to the zombie list. This will block
511 * until the process p_lock count reaches 0. The process will
512 * not be reaped until TDF_EXITING is set by cpu_thread_exit(),
513 * which is called from cpu_proc_exit().
515 * Interlock against waiters using p_waitgen. We increment
516 * p_waitgen after completing the move of our process to the
519 * WARNING: pp becomes stale when we block, clear it now as a
522 proc_move_allproc_zombie(p);
524 atomic_add_long(&pp->p_waitgen, 1);
528 * release controlled reaper for exit if we own it and return the
529 * remaining reaper (the one for us), which we will drop after we
532 reap = reaper_exit(p);
535 * Reparent all of this process's children to the init process or
536 * to the designated reaper. We must hold the reaper's p_token in
537 * order to safely mess with p_children.
539 * We already hold p->p_token (to remove the children from our list).
542 q = LIST_FIRST(&p->p_children);
544 reproc = reaper_get(reap);
545 lwkt_gettoken(&reproc->p_token);
546 while ((q = LIST_FIRST(&p->p_children)) != NULL) {
548 lwkt_gettoken(&q->p_token);
549 if (q != LIST_FIRST(&p->p_children)) {
550 lwkt_reltoken(&q->p_token);
554 LIST_REMOVE(q, p_sibling);
555 LIST_INSERT_HEAD(&reproc->p_children, q, p_sibling);
557 q->p_ppid = reproc->p_pid;
558 q->p_sigparent = SIGCHLD;
561 * Traced processes are killed
562 * since their existence means someone is screwing up.
564 if (q->p_flags & P_TRACED) {
565 q->p_flags &= ~P_TRACED;
568 lwkt_reltoken(&q->p_token);
571 lwkt_reltoken(&reproc->p_token);
576 * Save exit status and final rusage info. We no longer add
577 * child rusage info into self times, wait4() and kern_wait()
578 * handles it in order to properly support wait6().
580 calcru_proc(p, &p->p_ru);
581 /*ruadd(&p->p_ru, &p->p_cru); REMOVED */
584 * notify interested parties of our demise.
586 KNOTE(&p->p_klist, NOTE_EXIT);
589 * Notify parent that we're gone. If parent has the PS_NOCLDWAIT
590 * flag set, or if the handler is set to SIG_IGN, notify the reaper
591 * instead (it will handle this situation).
593 * NOTE: The reaper can still be the parent process.
597 if (p->p_pptr->p_sigacts->ps_flag & (PS_NOCLDWAIT | PS_CLDSIGIGN)) {
599 reproc = reaper_get(reap);
600 proc_reparent(p, reproc);
608 * Signal (possibly new) parent.
612 if (p->p_sigparent && pp != initproc) {
613 int sig = p->p_sigparent;
615 if (sig != SIGUSR1 && sig != SIGCHLD)
619 ksignal(pp, SIGCHLD);
621 p->p_flags &= ~P_TRACED;
625 * cpu_exit is responsible for clearing curproc, since
626 * it is heavily integrated with the thread/switching sequence.
628 * Other substructures are freed from wait().
631 struct plimit *rlimit;
639 * Finally, call machine-dependent code to release as many of the
640 * lwp's resources as we can and halt execution of this thread.
642 * pp is a wild pointer now but still the correct wakeup() target.
643 * lwp_exit() only uses it to send the wakeup() signal to the likely
644 * parent. Any reparenting race that occurs will get a signal
645 * automatically and not be an issue.
651 * Eventually called by every exiting LWP
653 * p->p_token must be held. mplock may be held and will be released.
656 lwp_exit(int masterexit, void *waddr)
658 struct thread *td = curthread;
659 struct lwp *lp = td->td_lwp;
660 struct proc *p = lp->lwp_proc;
664 * Release the current user process designation on the process so
665 * the userland scheduler can work in someone else.
667 p->p_usched->release_curproc(lp);
670 * lwp_exit() may be called without setting LWP_MP_WEXIT, so
671 * make sure it is set here.
673 ASSERT_LWKT_TOKEN_HELD(&p->p_token);
674 atomic_set_int(&lp->lwp_mpflags, LWP_MP_WEXIT);
677 * Clean up any virtualization
680 vkernel_lwp_exit(lp);
686 * Clean up select/poll support
688 kqueue_terminate(&lp->lwp_kqueue);
691 * Clean up any syscall-cached ucred or rlimit.
694 crfree(td->td_ucred);
698 struct plimit *rlimit;
700 rlimit = td->td_limit;
706 * Cleanup any cached descriptors for this thread
712 * Nobody actually wakes us when the lock
713 * count reaches zero, so just wait one tick.
715 while (lp->lwp_lock > 0)
716 tsleep(lp, 0, "lwpexit", 1);
718 /* Hand down resource usage to our proc */
719 ruadd(&p->p_ru, &lp->lwp_ru);
722 * If we don't hold the process until the LWP is reaped wait*()
723 * may try to dispose of its vmspace before all the LWPs have
724 * actually terminated.
729 * Do any remaining work that might block on us. We should be
730 * coded such that further blocking is ok after decrementing
731 * p_nthreads but don't take the chance.
733 dsched_exit_thread(td);
734 biosched_done(curthread);
737 * We have to use the reaper for all the LWPs except the one doing
738 * the master exit. The LWP doing the master exit can just be
739 * left on p_lwps and the process reaper will deal with it
740 * synchronously, which is much faster.
742 * Wakeup anyone waiting on p_nthreads to drop to 1 or 0.
744 * The process is left held until the reaper calls lwp_dispose() on
745 * the lp (after calling lwp_wait()).
747 if (masterexit == 0) {
750 lwp_rb_tree_RB_REMOVE(&p->p_lwp_tree, lp);
752 if ((p->p_flags & P_MAYBETHREADED) && p->p_nthreads <= 1)
754 lwkt_gettoken(&deadlwp_token[cpu]);
755 LIST_INSERT_HEAD(&deadlwp_list[cpu], lp, u.lwp_reap_entry);
756 taskqueue_enqueue(taskqueue_thread[cpu], deadlwp_task[cpu]);
757 lwkt_reltoken(&deadlwp_token[cpu]);
760 if ((p->p_flags & P_MAYBETHREADED) && p->p_nthreads <= 1)
765 * We no longer need p_token.
767 * Tell the userland scheduler that we are going away
769 lwkt_reltoken(&p->p_token);
770 p->p_usched->heuristic_exiting(lp, p);
773 * Issue late wakeups after releasing our token to give us a chance
774 * to deschedule and switch away before another cpu in a wait*()
775 * reaps us. This is done as late as possible to reduce contention.
778 wakeup(&p->p_nthreads);
786 * Wait until a lwp is completely dead. The final interlock in this drama
787 * is when TDF_EXITING is set in cpu_thread_exit() just before the final
790 * At the point TDF_EXITING is set a complete exit is accomplished when
791 * TDF_RUNNING and TDF_PREEMPT_LOCK are both clear. td_mpflags has two
792 * post-switch interlock flags that can be used to wait for the TDF_
795 * Returns non-zero on success, and zero if the caller needs to retry
799 lwp_wait(struct lwp *lp)
801 struct thread *td = lp->lwp_thread;
804 KKASSERT(lwkt_preempted_proc() != lp);
807 * This bit of code uses the thread destruction interlock
808 * managed by lwkt_switch_return() to wait for the lwp's
809 * thread to completely disengage.
811 * It is possible for us to race another cpu core so we
812 * have to do this correctly.
815 mpflags = td->td_mpflags;
817 if (mpflags & TDF_MP_EXITSIG)
819 tsleep_interlock(td, 0);
820 if (atomic_cmpset_int(&td->td_mpflags, mpflags,
821 mpflags | TDF_MP_EXITWAIT)) {
822 tsleep(td, PINTERLOCKED, "lwpxt", 0);
827 * We've already waited for the core exit but there can still
828 * be other refs from e.g. process scans and such.
830 if (lp->lwp_lock > 0) {
831 tsleep(lp, 0, "lwpwait1", 1);
835 tsleep(td, 0, "lwpwait2", 1);
840 * Now that we have the thread destruction interlock these flags
841 * really should already be cleaned up, keep a check for safety.
843 * We can't rip its stack out from under it until TDF_EXITING is
844 * set and both TDF_RUNNING and TDF_PREEMPT_LOCK are clear.
845 * TDF_PREEMPT_LOCK must be checked because TDF_RUNNING
846 * will be cleared temporarily if a thread gets preempted.
848 while ((td->td_flags & (TDF_RUNNING |
851 TDF_EXITING)) != TDF_EXITING) {
852 tsleep(lp, 0, "lwpwait3", 1);
856 KASSERT((td->td_flags & (TDF_RUNQ|TDF_TSLEEPQ)) == 0,
857 ("lwp_wait: td %p (%s) still on run or sleep queue",
863 * Release the resources associated with a lwp.
864 * The lwp must be completely dead.
867 lwp_dispose(struct lwp *lp)
869 struct thread *td = lp->lwp_thread;
871 KKASSERT(lwkt_preempted_proc() != lp);
872 KKASSERT(lp->lwp_lock == 0);
873 KKASSERT(td->td_refs == 0);
874 KKASSERT((td->td_flags & (TDF_RUNNING |
877 TDF_EXITING)) == TDF_EXITING);
884 lp->lwp_thread = NULL;
885 lwkt_free_thread(td);
891 sys_wait4(struct wait_args *uap)
893 struct __wrusage wrusage;
900 options = uap->options | WEXITED | WTRAPPED;
903 if (id == WAIT_ANY) {
905 } else if (id == WAIT_MYPGRP) {
907 id = curproc->p_pgid;
915 error = kern_wait(idtype, id, &status, options, &wrusage,
916 NULL, &uap->sysmsg_result);
918 if (error == 0 && uap->status)
919 error = copyout(&status, uap->status, sizeof(*uap->status));
920 if (error == 0 && uap->rusage) {
921 ruadd(&wrusage.wru_self, &wrusage.wru_children);
922 error = copyout(&wrusage.wru_self, uap->rusage, sizeof(*uap->rusage));
928 sys_wait6(struct wait6_args *uap)
930 struct __wrusage wrusage;
931 struct __siginfo info;
932 struct __siginfo *infop;
940 * NOTE: wait6() requires WEXITED and WTRAPPED to be specified if
943 options = uap->options;
944 idtype = uap->idtype;
946 infop = uap->info ? &info : NULL;
951 if (id == WAIT_MYPGRP) {
953 id = curproc->p_pgid;
957 /* let kern_wait deal with the remainder */
961 error = kern_wait(idtype, id, &status, options,
962 &wrusage, infop, &uap->sysmsg_result);
964 if (error == 0 && uap->status)
965 error = copyout(&status, uap->status, sizeof(*uap->status));
966 if (error == 0 && uap->wrusage)
967 error = copyout(&wrusage, uap->wrusage, sizeof(*uap->wrusage));
968 if (error == 0 && uap->info)
969 error = copyout(&info, uap->info, sizeof(*uap->info));
974 * kernel wait*() system call support
977 kern_wait(idtype_t idtype, id_t id, int *status, int options,
978 struct __wrusage *wrusage, struct __siginfo *info, int *res)
980 struct thread *td = curthread;
982 struct proc *q = td->td_proc;
991 * Must not have extraneous options. Must have at least one
994 if (options &~ (WUNTRACED|WNOHANG|WCONTINUED|WLINUXCLONE|WSTOPPED|
995 WEXITED|WTRAPPED|WNOWAIT)) {
998 if ((options & (WEXITED | WUNTRACED | WCONTINUED | WTRAPPED)) == 0) {
1003 * Protect the q->p_children list
1005 lwkt_gettoken(&q->p_token);
1008 * All sorts of things can change due to blocking so we have to loop
1009 * all the way back up here.
1011 * The problem is that if a process group is stopped and the parent
1012 * is doing a wait*(..., WUNTRACED, ...), it will see the STOP
1013 * of the child and then stop itself when it tries to return from the
1014 * system call. When the process group is resumed the parent will
1015 * then get the STOP status even though the child has now resumed
1016 * (a followup wait*() will get the CONT status).
1018 * Previously the CONT would overwrite the STOP because the tstop
1019 * was handled within tsleep(), and the parent would only see
1020 * the CONT when both are stopped and continued together. This little
1021 * two-line hack restores this effect.
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;
1180 info->si_code = CLD_KILLED;
1182 info->si_code = CLD_EXITED;
1183 info->si_status = p->p_xstat;
1184 info->si_pid = p->p_pid;
1185 info->si_uid = p->p_ucred->cr_uid;
1189 * WNOWAIT shortcuts to done here, leaving the
1190 * child on the zombie list.
1192 if (options & WNOWAIT) {
1193 lwkt_reltoken(&p->p_token);
1200 * If we got the child via a ptrace 'attach',
1201 * we need to give it back to the old parent.
1203 if (p->p_oppid && (t = pfind(p->p_oppid)) != NULL) {
1205 proc_reparent(p, t);
1206 ksignal(t, SIGCHLD);
1209 lwkt_reltoken(&p->p_token);
1216 * Unlink the proc from its process group so that
1217 * the following operations won't lead to an
1218 * inconsistent state for processes running down
1221 proc_remove_zombie(p);
1223 lwkt_reltoken(&p->p_token);
1227 ruadd(&q->p_cru, &p->p_ru);
1228 ruadd(&q->p_cru, &p->p_cru);
1231 * Decrement the count of procs running with this uid.
1233 chgproccnt(p->p_ucred->cr_ruidinfo, -1, 0);
1236 * Free up credentials. p_spin is required to
1237 * avoid races against allproc scans.
1239 spin_lock(&p->p_spin);
1242 spin_unlock(&p->p_spin);
1246 * Remove unused arguments
1250 if (pa && refcount_release(&pa->ar_ref)) {
1256 p->p_sigacts = NULL;
1257 if (ps && refcount_release(&ps->ps_refcnt)) {
1258 kfree(ps, M_SUBPROC);
1263 * Our exitingcount was incremented when the process
1264 * became a zombie, now that the process has been
1265 * removed from (almost) all lists we should be able
1266 * to safely destroy its vmspace. Wait for any current
1267 * holders to go away (so the vmspace remains stable),
1270 * NOTE: Releasing the parent process (q) p_token
1271 * across the vmspace_exitfree() call is
1272 * important here to reduce stalls on
1273 * interactions with (q) (such as
1274 * fork/exec/wait or 'ps').
1276 PSTALL(p, "reap4", 1);
1277 lwkt_reltoken(&q->p_token);
1278 vmspace_exitfree(p);
1279 lwkt_gettoken(&q->p_token);
1280 PSTALL(p, "reap5", 1);
1283 * NOTE: We have to officially release ZOMB in order
1284 * to ensure that a racing thread in kern_wait()
1285 * which blocked on ZOMB is woken up.
1288 kfree(p->p_uidpcpu, M_SUBPROC);
1290 atomic_add_int(&nprocs, -1);
1296 * Process has not yet exited
1298 if ((p->p_stat == SSTOP || p->p_stat == SCORE) &&
1299 (p->p_flags & P_WAITED) == 0 &&
1300 (((p->p_flags & P_TRACED) && (options & WTRAPPED)) ||
1301 (options & WSTOPPED))) {
1302 lwkt_gettoken(&p->p_token);
1303 if (p->p_pptr != q) {
1304 lwkt_reltoken(&p->p_token);
1308 if ((p->p_stat != SSTOP && p->p_stat != SCORE) ||
1309 (p->p_flags & P_WAITED) != 0 ||
1310 ((p->p_flags & P_TRACED) == 0 &&
1311 (options & WUNTRACED) == 0)) {
1312 lwkt_reltoken(&p->p_token);
1318 * Don't set P_WAITED if WNOWAIT specified, leaving
1319 * the process in a waitable state.
1321 if ((options & WNOWAIT) == 0)
1322 p->p_flags |= P_WAITED;
1325 *status = W_STOPCODE(p->p_xstat);
1326 /* Zero rusage so we get something consistent. */
1327 bzero(wrusage, sizeof(*wrusage));
1330 bzero(info, sizeof(*info));
1331 if (p->p_flags & P_TRACED)
1332 info->si_code = CLD_TRAPPED;
1334 info->si_code = CLD_STOPPED;
1335 info->si_status = p->p_xstat;
1337 lwkt_reltoken(&p->p_token);
1341 if ((options & WCONTINUED) && (p->p_flags & P_CONTINUED)) {
1342 lwkt_gettoken(&p->p_token);
1343 if (p->p_pptr != q) {
1344 lwkt_reltoken(&p->p_token);
1348 if ((p->p_flags & P_CONTINUED) == 0) {
1349 lwkt_reltoken(&p->p_token);
1357 * Don't set P_WAITED if WNOWAIT specified, leaving
1358 * the process in a waitable state.
1360 if ((options & WNOWAIT) == 0)
1361 p->p_flags &= ~P_CONTINUED;
1366 bzero(info, sizeof(*info));
1367 info->si_code = CLD_CONTINUED;
1368 info->si_status = p->p_xstat;
1370 lwkt_reltoken(&p->p_token);
1380 if (options & WNOHANG) {
1387 * Wait for signal - interlocked using q->p_waitgen.
1390 while ((waitgen & 0x7FFFFFFF) == (q->p_waitgen & 0x7FFFFFFF)) {
1391 tsleep_interlock(q, PCATCH);
1392 waitgen = atomic_fetchadd_long(&q->p_waitgen, 0x80000000);
1393 if ((waitgen & 0x7FFFFFFF) == (q->p_waitgen & 0x7FFFFFFF)) {
1394 error = tsleep(q, PCATCH | PINTERLOCKED, "wait", 0);
1400 lwkt_reltoken(&q->p_token);
1407 * Change child's parent process to parent.
1409 * p_children/p_sibling requires the parent's token, and
1410 * changing pptr requires the child's token, so we have to
1411 * get three tokens to do this operation. We also need to
1412 * hold pointers that might get ripped out from under us to
1413 * preserve structural integrity.
1415 * It is possible to race another reparent or disconnect or other
1416 * similar operation. We must retry when this situation occurs.
1417 * Once we successfully reparent the process we no longer care
1421 proc_reparent(struct proc *child, struct proc *parent)
1426 while ((opp = child->p_pptr) != parent) {
1428 lwkt_gettoken(&opp->p_token);
1429 lwkt_gettoken(&child->p_token);
1430 lwkt_gettoken(&parent->p_token);
1431 if (child->p_pptr != opp) {
1432 lwkt_reltoken(&parent->p_token);
1433 lwkt_reltoken(&child->p_token);
1434 lwkt_reltoken(&opp->p_token);
1438 LIST_REMOVE(child, p_sibling);
1439 LIST_INSERT_HEAD(&parent->p_children, child, p_sibling);
1440 child->p_pptr = parent;
1441 child->p_ppid = parent->p_pid;
1442 lwkt_reltoken(&parent->p_token);
1443 lwkt_reltoken(&child->p_token);
1444 lwkt_reltoken(&opp->p_token);
1445 if (LIST_EMPTY(&opp->p_children))
1454 * The next two functions are to handle adding/deleting items on the
1458 * Take the arguments given and put them onto the exit callout list,
1459 * However first make sure that it's not already there.
1460 * returns 0 on success.
1464 at_exit(exitlist_fn function)
1466 struct exitlist *ep;
1469 /* Be noisy if the programmer has lost track of things */
1470 if (rm_at_exit(function))
1471 kprintf("WARNING: exit callout entry (%p) already present\n",
1474 ep = kmalloc(sizeof(*ep), M_ATEXIT, M_NOWAIT);
1477 ep->function = function;
1478 TAILQ_INSERT_TAIL(&exit_list, ep, next);
1483 * Scan the exit callout list for the given item and remove it.
1484 * Returns the number of items removed (0 or 1)
1487 rm_at_exit(exitlist_fn function)
1489 struct exitlist *ep;
1491 TAILQ_FOREACH(ep, &exit_list, next) {
1492 if (ep->function == function) {
1493 TAILQ_REMOVE(&exit_list, ep, next);
1494 kfree(ep, M_ATEXIT);
1502 * LWP reaper related code.
1505 reaplwps(void *context, int dummy)
1507 struct lwplist *lwplist = context;
1511 lwkt_gettoken(&deadlwp_token[cpu]);
1512 while ((lp = LIST_FIRST(lwplist))) {
1513 LIST_REMOVE(lp, u.lwp_reap_entry);
1516 lwkt_reltoken(&deadlwp_token[cpu]);
1520 reaplwp(struct lwp *lp)
1522 while (lwp_wait(lp) == 0)
1532 for (cpu = 0; cpu < ncpus; cpu++) {
1533 lwkt_token_init(&deadlwp_token[cpu], "deadlwpl");
1534 LIST_INIT(&deadlwp_list[cpu]);
1535 deadlwp_task[cpu] = kmalloc(sizeof(*deadlwp_task[cpu]),
1536 M_DEVBUF, M_WAITOK);
1537 TASK_INIT(deadlwp_task[cpu], 0, reaplwps, &deadlwp_list[cpu]);
1541 SYSINIT(deadlwpinit, SI_SUB_CONFIGURE, SI_ORDER_ANY, deadlwp_init, NULL);