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/sysmsg.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>
75 #include <machine/vmm.h>
77 static void reaplwps(void *context, int dummy);
78 static void reaplwp(struct lwp *lp);
79 static void killlwps(struct lwp *lp);
81 static MALLOC_DEFINE(M_ATEXIT, "atexit", "atexit callback");
84 * callout list for things to do at exit time
88 TAILQ_ENTRY(exitlist) next;
91 TAILQ_HEAD(exit_list_head, exitlist);
92 static struct exit_list_head exit_list = TAILQ_HEAD_INITIALIZER(exit_list);
97 static struct task *deadlwp_task[MAXCPU];
98 static struct lwplist deadlwp_list[MAXCPU];
99 static struct lwkt_token deadlwp_token[MAXCPU];
101 void (*linux_task_drop_callback)(thread_t td);
102 void (*linux_proc_drop_callback)(struct proc *p);
108 * SYS_EXIT_ARGS(int rval)
111 sys_exit(struct sysmsg *sysmsg, const 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 sysmsg *sysmsg, const 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
229 if (fakestop && (lp->lwp_mpflags & LWP_MP_WSTOP)) {
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 sysmsg sysmsg;
331 sysmsg.extargs.kill.signum = SIGKILL;
334 sysmsg.extargs.kill.pid = q->p_pid;
336 * The interface for kill is better
337 * than the internal signal
339 sys_kill(&sysmsg, &sysmsg.extargs.kill);
343 tsleep((caddr_t)p, 0, "exit1", 0);
349 STOPEVENT(p, S_EXIT, rv);
350 p->p_flags |= P_POSTEXIT; /* stop procfs stepping */
353 * Check if any loadable modules need anything done at process exit.
354 * e.g. SYSV IPC stuff
355 * XXX what if one of these generates an error?
360 * XXX: imho, the eventhandler stuff is much cleaner than this.
361 * Maybe we should move everything to use eventhandler.
363 TAILQ_FOREACH(ep, &exit_list, next)
366 if (p->p_flags & P_PROFIL)
369 SIGEMPTYSET(p->p_siglist);
370 SIGEMPTYSET(lp->lwp_siglist);
371 if (timevalisset(&p->p_realtimer.it_value))
372 callout_terminate(&p->p_ithandle);
375 * Reset any sigio structures pointing to us as a result of
376 * F_SETOWN with our pid.
378 funsetownlst(&p->p_sigiolst);
381 * Close open files and release open-file table.
386 if (p->p_leader->p_peers) {
388 while(q->p_peers != p)
390 q->p_peers = p->p_peers;
391 wakeup((caddr_t)p->p_leader);
395 * XXX Shutdown SYSV semaphores
399 /* The next two chunks should probably be moved to vmspace_exit. */
403 * Clean up data related to virtual kernel operation. Clean up
404 * any vkernel context related to the current lwp now so we can
408 vkernel_lwp_exit(lp);
413 * Release the user portion of address space. The exitbump prevents
414 * the vmspace from being completely eradicated (using holdcnt).
415 * This releases references to vnodes, which could cause I/O if the
416 * file has been unlinked. We need to do this early enough that
417 * we can still sleep.
419 * We can't free the entire vmspace as the kernel stack may be mapped
420 * within that space also.
422 * Processes sharing the same vmspace may exit in one order, and
423 * get cleaned up by vmspace_exit() in a different order. The
424 * last exiting process to reach this point releases as much of
425 * the environment as it can, and the last process cleaned up
426 * by vmspace_exit() (which decrements exitingcnt) cleans up the
429 * NOTE: Releasing p_token around this call is helpful if the
430 * vmspace had a huge RSS. Otherwise some other process
431 * trying to do an allproc or other scan (like 'ps') may
432 * stall for a long time.
434 lwkt_reltoken(&p->p_token);
436 lwkt_gettoken(&p->p_token);
438 if (SESS_LEADER(p)) {
439 struct session *sp = p->p_session;
443 * We are the controlling process. Signal the
444 * foreground process group, drain the controlling
445 * terminal, and revoke access to the controlling
448 * NOTE: while waiting for the process group to exit
449 * it is possible that one of the processes in the
450 * group will revoke the tty, so the ttyclosesession()
451 * function will re-check sp->s_ttyvp.
453 if (sp->s_ttyp && (sp->s_ttyp->t_session == sp)) {
454 if (sp->s_ttyp->t_pgrp)
455 pgsignal(sp->s_ttyp->t_pgrp, SIGHUP, 1);
457 ttyclosesession(sp, 1); /* also revoke */
460 * Release the tty. If someone has it open via
461 * /dev/tty then close it (since they no longer can
462 * once we've NULL'd it out).
464 ttyclosesession(sp, 0);
467 * s_ttyp is not zero'd; we use this to indicate
468 * that the session once had a controlling terminal.
469 * (for logging and informational purposes)
474 fixjobc(p, p->p_pgrp, 0);
475 (void)acct_process(p);
481 ktrdestroy(&p->p_tracenode);
485 * Release reference to text vnode
487 if ((vtmp = p->p_textvp) != NULL) {
492 /* Release namecache handle to text file */
493 if (p->p_textnch.ncp)
494 cache_drop(&p->p_textnch);
497 * We have to handle PPWAIT here or proc_move_allproc_zombie()
498 * will block on the PHOLD() the parent is doing.
500 * We are using the flag as an interlock so an atomic op is
501 * necessary to synchronize with the parent's cpu.
503 if (p->p_flags & P_PPWAIT) {
504 if (p->p_pptr && p->p_pptr->p_upmap)
505 atomic_add_int(&p->p_pptr->p_upmap->invfork, -1);
506 atomic_clear_int(&p->p_flags, P_PPWAIT);
511 * Move the process to the zombie list. This will block
512 * until the process p_lock count reaches 0. The process will
513 * not be reaped until TDF_EXITING is set by cpu_thread_exit(),
514 * which is called from cpu_proc_exit().
516 * Interlock against waiters using p_waitgen. We increment
517 * p_waitgen after completing the move of our process to the
520 * WARNING: pp becomes stale when we block, clear it now as a
523 proc_move_allproc_zombie(p);
525 atomic_add_long(&pp->p_waitgen, 1);
529 * release controlled reaper for exit if we own it and return the
530 * remaining reaper (the one for us), which we will drop after we
533 reap = reaper_exit(p);
536 * Reparent all of this process's children to the init process or
537 * to the designated reaper. We must hold the reaper's p_token in
538 * order to safely mess with p_children.
540 * We already hold p->p_token (to remove the children from our list).
543 q = LIST_FIRST(&p->p_children);
545 reproc = reaper_get(reap);
546 lwkt_gettoken(&reproc->p_token);
547 while ((q = LIST_FIRST(&p->p_children)) != NULL) {
549 lwkt_gettoken(&q->p_token);
550 if (q != LIST_FIRST(&p->p_children)) {
551 lwkt_reltoken(&q->p_token);
555 LIST_REMOVE(q, p_sibling);
556 LIST_INSERT_HEAD(&reproc->p_children, q, p_sibling);
558 q->p_ppid = reproc->p_pid;
559 q->p_sigparent = SIGCHLD;
562 * Traced processes are killed
563 * since their existence means someone is screwing up.
565 if (q->p_flags & P_TRACED) {
566 q->p_flags &= ~P_TRACED;
569 lwkt_reltoken(&q->p_token);
572 lwkt_reltoken(&reproc->p_token);
577 * Save exit status and final rusage info. We no longer add
578 * child rusage info into self times, wait4() and kern_wait()
579 * handles it in order to properly support wait6().
581 calcru_proc(p, &p->p_ru);
582 /*ruadd(&p->p_ru, &p->p_cru); REMOVED */
585 * notify interested parties of our demise.
587 KNOTE(&p->p_klist, NOTE_EXIT);
590 * Notify parent that we're gone. If parent has the PS_NOCLDWAIT
591 * flag set, or if the handler is set to SIG_IGN, notify the reaper
592 * instead (it will handle this situation).
594 * NOTE: The reaper can still be the parent process.
598 if (p->p_pptr->p_sigacts->ps_flag & (PS_NOCLDWAIT | PS_CLDSIGIGN)) {
600 reproc = reaper_get(reap);
601 proc_reparent(p, reproc);
609 * Signal (possibly new) parent.
613 if (p->p_sigparent && pp != initproc) {
614 int sig = p->p_sigparent;
616 if (sig != SIGUSR1 && sig != SIGCHLD)
620 ksignal(pp, SIGCHLD);
622 p->p_flags &= ~P_TRACED;
626 * cpu_exit is responsible for clearing curproc, since
627 * it is heavily integrated with the thread/switching sequence.
629 * Other substructures are freed from wait().
632 struct plimit *rlimit;
640 * Finally, call machine-dependent code to release as many of the
641 * lwp's resources as we can and halt execution of this thread.
643 * pp is a wild pointer now but still the correct wakeup() target.
644 * lwp_exit() only uses it to send the wakeup() signal to the likely
645 * parent. Any reparenting race that occurs will get a signal
646 * automatically and not be an issue.
652 * Eventually called by every exiting LWP
654 * p->p_token must be held. mplock may be held and will be released.
657 lwp_exit(int masterexit, void *waddr)
659 struct thread *td = curthread;
660 struct lwp *lp = td->td_lwp;
661 struct proc *p = lp->lwp_proc;
665 * Release the current user process designation on the process so
666 * the userland scheduler can work in someone else.
668 p->p_usched->release_curproc(lp);
671 * Destroy the per-thread shared page and remove from any pmaps
677 * lwp_exit() may be called without setting LWP_MP_WEXIT, so
678 * make sure it is set here.
680 ASSERT_LWKT_TOKEN_HELD(&p->p_token);
681 atomic_set_int(&lp->lwp_mpflags, LWP_MP_WEXIT);
684 * Clean up any virtualization
687 vkernel_lwp_exit(lp);
693 * Clean up select/poll support
695 kqueue_terminate(&lp->lwp_kqueue);
697 if (td->td_linux_task)
698 linux_task_drop_callback(td);
699 if (masterexit && p->p_linux_mm)
700 linux_proc_drop_callback(p);
703 * Clean up any syscall-cached ucred or rlimit.
706 crfree(td->td_ucred);
710 struct plimit *rlimit;
712 rlimit = td->td_limit;
718 * Cleanup any cached descriptors for this thread
724 * Nobody actually wakes us when the lock
725 * count reaches zero, so just wait one tick.
727 while (lp->lwp_lock > 0)
728 tsleep(lp, 0, "lwpexit", 1);
730 /* Hand down resource usage to our proc */
731 ruadd(&p->p_ru, &lp->lwp_ru);
734 * If we don't hold the process until the LWP is reaped wait*()
735 * may try to dispose of its vmspace before all the LWPs have
736 * actually terminated.
741 * Do any remaining work that might block on us. We should be
742 * coded such that further blocking is ok after decrementing
743 * p_nthreads but don't take the chance.
745 dsched_exit_thread(td);
746 biosched_done(curthread);
749 * We have to use the reaper for all the LWPs except the one doing
750 * the master exit. The LWP doing the master exit can just be
751 * left on p_lwps and the process reaper will deal with it
752 * synchronously, which is much faster.
754 * Wakeup anyone waiting on p_nthreads to drop to 1 or 0.
756 * The process is left held until the reaper calls lwp_dispose() on
757 * the lp (after calling lwp_wait()).
759 if (masterexit == 0) {
762 lwp_rb_tree_RB_REMOVE(&p->p_lwp_tree, lp);
764 if ((p->p_flags & P_MAYBETHREADED) && p->p_nthreads <= 1)
766 lwkt_gettoken(&deadlwp_token[cpu]);
767 LIST_INSERT_HEAD(&deadlwp_list[cpu], lp, u.lwp_reap_entry);
768 taskqueue_enqueue(taskqueue_thread[cpu], deadlwp_task[cpu]);
769 lwkt_reltoken(&deadlwp_token[cpu]);
772 if ((p->p_flags & P_MAYBETHREADED) && p->p_nthreads <= 1)
777 * We no longer need p_token.
779 * Tell the userland scheduler that we are going away
781 lwkt_reltoken(&p->p_token);
782 p->p_usched->heuristic_exiting(lp, p);
785 * Issue late wakeups after releasing our token to give us a chance
786 * to deschedule and switch away before another cpu in a wait*()
787 * reaps us. This is done as late as possible to reduce contention.
790 wakeup(&p->p_nthreads);
798 * Wait until a lwp is completely dead. The final interlock in this drama
799 * is when TDF_EXITING is set in cpu_thread_exit() just before the final
802 * At the point TDF_EXITING is set a complete exit is accomplished when
803 * TDF_RUNNING and TDF_PREEMPT_LOCK are both clear. td_mpflags has two
804 * post-switch interlock flags that can be used to wait for the TDF_
807 * Returns non-zero on success, and zero if the caller needs to retry
811 lwp_wait(struct lwp *lp)
813 struct thread *td = lp->lwp_thread;
816 KKASSERT(lwkt_preempted_proc() != lp);
819 * This bit of code uses the thread destruction interlock
820 * managed by lwkt_switch_return() to wait for the lwp's
821 * thread to completely disengage.
823 * It is possible for us to race another cpu core so we
824 * have to do this correctly.
827 mpflags = td->td_mpflags;
829 if (mpflags & TDF_MP_EXITSIG)
831 tsleep_interlock(td, 0);
832 if (atomic_cmpset_int(&td->td_mpflags, mpflags,
833 mpflags | TDF_MP_EXITWAIT)) {
834 tsleep(td, PINTERLOCKED, "lwpxt", 0);
839 * We've already waited for the core exit but there can still
840 * be other refs from e.g. process scans and such.
842 if (lp->lwp_lock > 0) {
843 tsleep(lp, 0, "lwpwait1", 1);
847 tsleep(td, 0, "lwpwait2", 1);
852 * Now that we have the thread destruction interlock these flags
853 * really should already be cleaned up, keep a check for safety.
855 * We can't rip its stack out from under it until TDF_EXITING is
856 * set and both TDF_RUNNING and TDF_PREEMPT_LOCK are clear.
857 * TDF_PREEMPT_LOCK must be checked because TDF_RUNNING
858 * will be cleared temporarily if a thread gets preempted.
860 while ((td->td_flags & (TDF_RUNNING |
863 TDF_EXITING)) != TDF_EXITING) {
864 tsleep(lp, 0, "lwpwait3", 1);
868 KASSERT((td->td_flags & (TDF_RUNQ|TDF_TSLEEPQ)) == 0,
869 ("lwp_wait: td %p (%s) still on run or sleep queue",
875 * Release the resources associated with a lwp.
876 * The lwp must be completely dead.
879 lwp_dispose(struct lwp *lp)
881 struct thread *td = lp->lwp_thread;
883 KKASSERT(lwkt_preempted_proc() != lp);
884 KKASSERT(lp->lwp_lock == 0);
885 KKASSERT(td->td_refs == 0);
886 KKASSERT((td->td_flags & (TDF_RUNNING |
889 TDF_EXITING)) == TDF_EXITING);
896 lp->lwp_thread = NULL;
897 lwkt_free_thread(td);
903 sys_wait4(struct sysmsg *sysmsg, const struct wait_args *uap)
905 struct __wrusage wrusage;
912 options = uap->options | WEXITED | WTRAPPED;
915 if (id == WAIT_ANY) {
917 } else if (id == WAIT_MYPGRP) {
919 id = curproc->p_pgid;
927 error = kern_wait(idtype, id, &status, options, &wrusage,
928 NULL, &sysmsg->sysmsg_result);
930 if (error == 0 && uap->status)
931 error = copyout(&status, uap->status, sizeof(*uap->status));
932 if (error == 0 && uap->rusage) {
933 ruadd(&wrusage.wru_self, &wrusage.wru_children);
934 error = copyout(&wrusage.wru_self, uap->rusage, sizeof(*uap->rusage));
940 sys_wait6(struct sysmsg *sysmsg, const struct wait6_args *uap)
942 struct __wrusage wrusage;
952 * NOTE: wait6() requires WEXITED and WTRAPPED to be specified if
955 options = uap->options;
956 idtype = uap->idtype;
958 infop = uap->info ? &info : NULL;
963 if (id == WAIT_MYPGRP) {
965 id = curproc->p_pgid;
969 /* let kern_wait deal with the remainder */
973 error = kern_wait(idtype, id, &status, options,
974 &wrusage, infop, &sysmsg->sysmsg_result);
976 if (error == 0 && uap->status)
977 error = copyout(&status, uap->status, sizeof(*uap->status));
978 if (error == 0 && uap->wrusage)
979 error = copyout(&wrusage, uap->wrusage, sizeof(*uap->wrusage));
980 if (error == 0 && uap->info)
981 error = copyout(&info, uap->info, sizeof(*uap->info));
986 * kernel wait*() system call support
989 kern_wait(idtype_t idtype, id_t id, int *status, int options,
990 struct __wrusage *wrusage, siginfo_t *info, int *res)
992 struct thread *td = curthread;
994 struct proc *q = td->td_proc;
1003 * Must not have extraneous options. Must have at least one
1006 if (options &~ (WUNTRACED|WNOHANG|WCONTINUED|WLINUXCLONE|WSTOPPED|
1007 WEXITED|WTRAPPED|WNOWAIT)) {
1010 if ((options & (WEXITED | WUNTRACED | WCONTINUED | WTRAPPED)) == 0) {
1015 * Protect the q->p_children list
1017 lwkt_gettoken(&q->p_token);
1020 * All sorts of things can change due to blocking so we have to loop
1021 * all the way back up here.
1023 * The problem is that if a process group is stopped and the parent
1024 * is doing a wait*(..., WUNTRACED, ...), it will see the STOP
1025 * of the child and then stop itself when it tries to return from the
1026 * system call. When the process group is resumed the parent will
1027 * then get the STOP status even though the child has now resumed
1028 * (a followup wait*() will get the CONT status).
1030 * Previously the CONT would overwrite the STOP because the tstop
1031 * was handled within tsleep(), and the parent would only see
1032 * the CONT when both are stopped and continued together. This little
1033 * two-line hack restores this effect.
1035 * No locks are held so we can safely block the process here.
1037 if (STOPLWP(q, td->td_lwp))
1045 * NOTE: We don't want to break q's p_token in the loop for the
1046 * case where no children are found or we risk breaking the
1047 * interlock between child and parent.
1049 waitgen = atomic_fetchadd_long(&q->p_waitgen, 0x80000000);
1050 LIST_FOREACH(p, &q->p_children, p_sibling) {
1052 * Skip children that another thread is already uninterruptably
1055 if (PWAITRES_PENDING(p))
1059 * Filter, (p) will be held on fall-through. Try to optimize
1060 * this to avoid the atomic op until we are pretty sure we
1061 * want this process.
1068 if (p->p_pid != (pid_t)id)
1073 if (p->p_pgid != (pid_t)id)
1079 if (p->p_session && p->p_session->s_sid != (pid_t)id) {
1086 if (p->p_ucred->cr_uid != (uid_t)id) {
1093 if (p->p_ucred->cr_gid != (gid_t)id) {
1100 if (p->p_ucred->cr_prison &&
1101 p->p_ucred->cr_prison->pr_id != (int)id) {
1107 /* unsupported filter */
1110 /* (p) is held at this point */
1113 * This special case handles a kthread spawned by linux_clone
1114 * (see linux_misc.c). The linux_wait4 and linux_waitpid
1115 * functions need to be able to distinguish between waiting
1116 * on a process and waiting on a thread. It is a thread if
1117 * p_sigparent is not SIGCHLD, and the WLINUXCLONE option
1118 * signifies we want to wait for threads and not processes.
1120 if ((p->p_sigparent != SIGCHLD) ^
1121 ((options & WLINUXCLONE) != 0)) {
1127 if (p->p_stat == SZOMB && (options & WEXITED)) {
1129 * We may go into SZOMB with threads still present.
1130 * We must wait for them to exit before we can reap
1131 * the master thread, otherwise we may race reaping
1132 * non-master threads.
1134 * Only this routine can remove a process from
1135 * the zombie list and destroy it.
1137 * This function will fail after sleeping if another
1138 * thread owns the zombie lock. This function will
1139 * fail immediately or after sleeping if another
1140 * thread owns or obtains ownership of the reap via
1147 lwkt_gettoken(&p->p_token);
1148 if (p->p_pptr != q) {
1149 lwkt_reltoken(&p->p_token);
1156 * We are the reaper, from this point on the reap
1157 * cannot be aborted.
1160 while (p->p_nthreads > 0) {
1161 tsleep(&p->p_nthreads, 0, "lwpzomb", hz);
1165 * Reap any LWPs left in p->p_lwps. This is usually
1166 * just the last LWP. This must be done before
1167 * we loop on p_lock since the lwps hold a ref on
1168 * it as a vmspace interlock.
1170 * Once that is accomplished p_nthreads had better
1173 while ((lp = RB_ROOT(&p->p_lwp_tree)) != NULL) {
1175 * Make sure no one is using this lwp, before
1176 * it is removed from the tree. If we didn't
1177 * wait it here, lwp tree iteration with
1178 * blocking operation would be broken.
1180 while (lp->lwp_lock > 0)
1181 tsleep(lp, 0, "zomblwp", 1);
1182 lwp_rb_tree_RB_REMOVE(&p->p_lwp_tree, lp);
1185 KKASSERT(p->p_nthreads == 0);
1188 * Don't do anything really bad until all references
1189 * to the process go away. This may include other
1190 * LWPs which are still in the process of being
1191 * reaped. We can't just pull the rug out from under
1192 * them because they may still be using the VM space.
1194 * Certain kernel facilities such as /proc will also
1195 * put a hold on the process for short periods of
1198 PRELE(p); /* from top of loop */
1199 PSTALL(p, "reap3", 1); /* 1 ref (for PZOMBHOLD) */
1201 /* Take care of our return values. */
1204 *status = p->p_xstat;
1205 wrusage->wru_self = p->p_ru;
1206 wrusage->wru_children = p->p_cru;
1209 bzero(info, sizeof(*info));
1211 info->si_signo = SIGCHLD;
1212 if (WIFEXITED(p->p_xstat)) {
1213 info->si_code = CLD_EXITED;
1215 WEXITSTATUS(p->p_xstat);
1217 info->si_code = CLD_KILLED;
1218 info->si_status = WTERMSIG(p->p_xstat);
1220 info->si_pid = p->p_pid;
1221 info->si_uid = p->p_ucred->cr_uid;
1225 * WNOWAIT shortcuts to done here, leaving the
1226 * child on the zombie list.
1228 if (options & WNOWAIT) {
1229 lwkt_reltoken(&p->p_token);
1236 * If we got the child via a ptrace 'attach',
1237 * we need to give it back to the old parent.
1239 if (p->p_oppid && (t = pfind(p->p_oppid)) != NULL) {
1241 proc_reparent(p, t);
1242 ksignal(t, SIGCHLD);
1245 lwkt_reltoken(&p->p_token);
1252 * Unlink the proc from its process group so that
1253 * the following operations won't lead to an
1254 * inconsistent state for processes running down
1257 proc_remove_zombie(p);
1259 lwkt_reltoken(&p->p_token);
1263 ruadd(&q->p_cru, &p->p_ru);
1264 ruadd(&q->p_cru, &p->p_cru);
1267 * Decrement the count of procs running with this uid.
1269 chgproccnt(p->p_ucred->cr_ruidinfo, -1, 0);
1272 * Free up credentials. p_spin is required to
1273 * avoid races against allproc scans.
1275 spin_lock(&p->p_spin);
1278 spin_unlock(&p->p_spin);
1282 * Remove unused arguments
1286 if (pa && refcount_release(&pa->ar_ref)) {
1292 p->p_sigacts = NULL;
1293 if (ps && refcount_release(&ps->ps_refcnt)) {
1294 kfree(ps, M_SUBPROC);
1299 * Our exitingcount was incremented when the process
1300 * became a zombie, now that the process has been
1301 * removed from (almost) all lists we should be able
1302 * to safely destroy its vmspace. Wait for any current
1303 * holders to go away (so the vmspace remains stable),
1306 * NOTE: Releasing the parent process (q) p_token
1307 * across the vmspace_exitfree() call is
1308 * important here to reduce stalls on
1309 * interactions with (q) (such as
1310 * fork/exec/wait or 'ps').
1312 PSTALL(p, "reap4", 1);
1313 lwkt_reltoken(&q->p_token);
1314 vmspace_exitfree(p);
1315 lwkt_gettoken(&q->p_token);
1316 PSTALL(p, "reap5", 1);
1319 * NOTE: We have to officially release ZOMB in order
1320 * to ensure that a racing thread in kern_wait()
1321 * which blocked on ZOMB is woken up.
1324 kfree(p->p_uidpcpu, M_SUBPROC);
1326 atomic_add_int(&nprocs, -1);
1332 * Process has not yet exited
1334 if ((p->p_stat == SSTOP || p->p_stat == SCORE) &&
1335 (p->p_flags & P_WAITED) == 0 &&
1336 (((p->p_flags & P_TRACED) && (options & WTRAPPED)) ||
1337 (options & WSTOPPED))) {
1338 lwkt_gettoken(&p->p_token);
1339 if (p->p_pptr != q) {
1340 lwkt_reltoken(&p->p_token);
1344 if ((p->p_stat != SSTOP && p->p_stat != SCORE) ||
1345 (p->p_flags & P_WAITED) != 0 ||
1346 ((p->p_flags & P_TRACED) == 0 &&
1347 (options & WUNTRACED) == 0)) {
1348 lwkt_reltoken(&p->p_token);
1354 * Don't set P_WAITED if WNOWAIT specified, leaving
1355 * the process in a waitable state.
1357 if ((options & WNOWAIT) == 0)
1358 p->p_flags |= P_WAITED;
1361 *status = W_STOPCODE(p->p_xstat);
1362 /* Zero rusage so we get something consistent. */
1363 bzero(wrusage, sizeof(*wrusage));
1366 bzero(info, sizeof(*info));
1367 if (p->p_flags & P_TRACED)
1368 info->si_code = CLD_TRAPPED;
1370 info->si_code = CLD_STOPPED;
1371 info->si_status = WSTOPSIG(p->p_xstat);
1373 lwkt_reltoken(&p->p_token);
1377 if ((options & WCONTINUED) && (p->p_flags & P_CONTINUED)) {
1378 lwkt_gettoken(&p->p_token);
1379 if (p->p_pptr != q) {
1380 lwkt_reltoken(&p->p_token);
1384 if ((p->p_flags & P_CONTINUED) == 0) {
1385 lwkt_reltoken(&p->p_token);
1393 * Don't set P_WAITED if WNOWAIT specified, leaving
1394 * the process in a waitable state.
1396 if ((options & WNOWAIT) == 0)
1397 p->p_flags &= ~P_CONTINUED;
1402 bzero(info, sizeof(*info));
1403 info->si_code = CLD_CONTINUED;
1404 info->si_status = WSTOPSIG(p->p_xstat);
1406 lwkt_reltoken(&p->p_token);
1416 if (options & WNOHANG) {
1423 * Wait for signal - interlocked using q->p_waitgen.
1426 while ((waitgen & 0x7FFFFFFF) == (q->p_waitgen & 0x7FFFFFFF)) {
1427 tsleep_interlock(q, PCATCH);
1428 waitgen = atomic_fetchadd_long(&q->p_waitgen, 0x80000000);
1429 if ((waitgen & 0x7FFFFFFF) == (q->p_waitgen & 0x7FFFFFFF)) {
1430 error = tsleep(q, PCATCH | PINTERLOCKED, "wait", 0);
1436 lwkt_reltoken(&q->p_token);
1443 * Change child's parent process to parent.
1445 * p_children/p_sibling requires the parent's token, and
1446 * changing pptr requires the child's token, so we have to
1447 * get three tokens to do this operation. We also need to
1448 * hold pointers that might get ripped out from under us to
1449 * preserve structural integrity.
1451 * It is possible to race another reparent or disconnect or other
1452 * similar operation. We must retry when this situation occurs.
1453 * Once we successfully reparent the process we no longer care
1457 proc_reparent(struct proc *child, struct proc *parent)
1462 while ((opp = child->p_pptr) != parent) {
1464 lwkt_gettoken(&opp->p_token);
1465 lwkt_gettoken(&child->p_token);
1466 lwkt_gettoken(&parent->p_token);
1467 if (child->p_pptr != opp) {
1468 lwkt_reltoken(&parent->p_token);
1469 lwkt_reltoken(&child->p_token);
1470 lwkt_reltoken(&opp->p_token);
1474 LIST_REMOVE(child, p_sibling);
1475 LIST_INSERT_HEAD(&parent->p_children, child, p_sibling);
1476 child->p_pptr = parent;
1477 child->p_ppid = parent->p_pid;
1478 lwkt_reltoken(&parent->p_token);
1479 lwkt_reltoken(&child->p_token);
1480 lwkt_reltoken(&opp->p_token);
1481 if (LIST_EMPTY(&opp->p_children))
1490 * The next two functions are to handle adding/deleting items on the
1494 * Take the arguments given and put them onto the exit callout list,
1495 * However first make sure that it's not already there.
1496 * returns 0 on success.
1500 at_exit(exitlist_fn function)
1502 struct exitlist *ep;
1505 /* Be noisy if the programmer has lost track of things */
1506 if (rm_at_exit(function))
1507 kprintf("WARNING: exit callout entry (%p) already present\n",
1510 ep = kmalloc(sizeof(*ep), M_ATEXIT, M_NOWAIT);
1513 ep->function = function;
1514 TAILQ_INSERT_TAIL(&exit_list, ep, next);
1519 * Scan the exit callout list for the given item and remove it.
1520 * Returns the number of items removed (0 or 1)
1523 rm_at_exit(exitlist_fn function)
1525 struct exitlist *ep;
1527 TAILQ_FOREACH(ep, &exit_list, next) {
1528 if (ep->function == function) {
1529 TAILQ_REMOVE(&exit_list, ep, next);
1530 kfree(ep, M_ATEXIT);
1538 * LWP reaper related code.
1541 reaplwps(void *context, int dummy)
1543 struct lwplist *lwplist = context;
1547 lwkt_gettoken(&deadlwp_token[cpu]);
1548 while ((lp = LIST_FIRST(lwplist))) {
1549 LIST_REMOVE(lp, u.lwp_reap_entry);
1552 lwkt_reltoken(&deadlwp_token[cpu]);
1556 reaplwp(struct lwp *lp)
1558 while (lwp_wait(lp) == 0)
1568 for (cpu = 0; cpu < ncpus; cpu++) {
1569 lwkt_token_init(&deadlwp_token[cpu], "deadlwpl");
1570 LIST_INIT(&deadlwp_list[cpu]);
1571 deadlwp_task[cpu] = kmalloc(sizeof(*deadlwp_task[cpu]),
1572 M_DEVBUF, M_WAITOK);
1573 TASK_INIT(deadlwp_task[cpu], 0, reaplwps, &deadlwp_list[cpu]);
1577 SYSINIT(deadlwpinit, SI_SUB_CONFIGURE, SI_ORDER_ANY, deadlwp_init, NULL);