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. All advertising materials mentioning features or use of this software
19 * must display the following acknowledgement:
20 * This product includes software developed by the University of
21 * California, Berkeley and its contributors.
22 * 4. Neither the name of the University nor the names of its contributors
23 * may be used to endorse or promote products derived from this software
24 * without specific prior written permission.
26 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
38 * @(#)kern_exit.c 8.7 (Berkeley) 2/12/94
39 * $FreeBSD: src/sys/kern/kern_exit.c,v 1.92.2.11 2003/01/13 22:51:16 dillon Exp $
40 * $DragonFly: src/sys/kern/kern_exit.c,v 1.84 2007/07/12 21:56:22 dillon Exp $
43 #include "opt_compat.h"
44 #include "opt_ktrace.h"
46 #include <sys/param.h>
47 #include <sys/systm.h>
48 #include <sys/sysproto.h>
49 #include <sys/kernel.h>
50 #include <sys/malloc.h>
52 #include <sys/ktrace.h>
53 #include <sys/pioctl.h>
56 #include <sys/vnode.h>
57 #include <sys/resourcevar.h>
58 #include <sys/signalvar.h>
59 #include <sys/taskqueue.h>
60 #include <sys/ptrace.h>
61 #include <sys/acct.h> /* for acct_process() function prototype */
62 #include <sys/filedesc.h>
67 #include <sys/kern_syscall.h>
68 #include <sys/upcall.h>
70 #include <sys/unistd.h>
73 #include <vm/vm_param.h>
76 #include <vm/vm_map.h>
77 #include <vm/vm_zone.h>
78 #include <vm/vm_extern.h>
81 #include <sys/thread2.h>
82 #include <sys/sysref2.h>
84 static void reaplwps(void *context, int dummy);
85 static void killlwps(struct lwp *lp);
87 static MALLOC_DEFINE(M_ATEXIT, "atexit", "atexit callback");
88 static MALLOC_DEFINE(M_ZOMBIE, "zombie", "zombie proc status");
91 * callout list for things to do at exit time
95 TAILQ_ENTRY(exitlist) next;
98 TAILQ_HEAD(exit_list_head, exitlist);
99 static struct exit_list_head exit_list = TAILQ_HEAD_INITIALIZER(exit_list);
104 struct task *deadlwp_task[MAXCPU];
105 struct lwplist deadlwp_list[MAXCPU];
111 * SYS_EXIT_ARGS(int rval)
114 sys_exit(struct exit_args *uap)
116 exit1(W_EXITCODE(uap->rval, 0));
122 * Death of a lwp or process with optional bells and whistles.
125 sys_extexit(struct extexit_args *uap)
130 action = EXTEXIT_ACTION(uap->how);
131 who = EXTEXIT_WHO(uap->how);
133 /* Check parameters before we might perform some action */
148 error = copyout(&uap->status, uap->addr, sizeof(uap->status));
160 * Be sure only to perform a simple lwp exit if there is at
161 * least one more lwp in the proc, which will call exit1()
162 * later, otherwise the proc will be an UNDEAD and not even a
165 if (curproc->p_nthreads > 1) {
169 /* else last lwp in proc: do the real thing */
172 default: /* to help gcc */
174 exit1(W_EXITCODE(uap->status, 0));
182 * Kill all lwps associated with the current process except the
183 * current lwp. Return an error if we race another thread trying to
184 * do the same thing and lose the race.
186 * If forexec is non-zero the current thread and process flags are
187 * cleaned up so they can be reused.
190 killalllwps(int forexec)
192 struct lwp *lp = curthread->td_lwp;
193 struct proc *p = lp->lwp_proc;
196 * Interlock against P_WEXIT. Only one of the process's thread
197 * is allowed to do the master exit.
199 if (p->p_flag & P_WEXIT)
201 p->p_flag |= P_WEXIT;
204 * Interlock with LWP_WEXIT and kill any remaining LWPs
206 lp->lwp_flag |= LWP_WEXIT;
207 if (p->p_nthreads > 1)
211 * If doing this for an exec, clean up the remaining thread
212 * (us) for continuing operation after all the other threads
216 lp->lwp_flag &= ~LWP_WEXIT;
217 p->p_flag &= ~P_WEXIT;
223 * Kill all LWPs except the current one. Do not try to signal
224 * LWPs which have exited on their own or have already been
228 killlwps(struct lwp *lp)
230 struct proc *p = lp->lwp_proc;
234 * Kill the remaining LWPs, interlock with LWP_WEXIT.
236 FOREACH_LWP_IN_PROC(tlp, p) {
237 if ((tlp->lwp_flag & LWP_WEXIT) == 0) {
238 lwpsignal(p, tlp, SIGKILL);
239 tlp->lwp_flag |= LWP_WEXIT;
244 * Wait for everything to clear out.
246 while (p->p_nthreads > 1) {
248 kprintf("killlwps: waiting for %d lwps of pid "
250 p->p_nthreads - 1, p->p_pid);
251 tsleep(&p->p_nthreads, 0, "killlwps", hz);
256 * Exit: deallocate address space and other resources, change proc state
257 * to zombie, and unlink proc from allproc and parent's lists. Save exit
258 * status and rusage for wait(). Check for child processes and orphan them.
263 struct thread *td = curthread;
264 struct proc *p = td->td_proc;
265 struct lwp *lp = td->td_lwp;
273 kprintf("init died (signal %d, exit %d)\n",
274 WTERMSIG(rv), WEXITSTATUS(rv));
275 panic("Going nowhere without my init!");
279 * Kill all lwps associated with the current process, return an
280 * error if we race another thread trying to do the same thing
283 error = killalllwps(0);
289 caps_exit(lp->lwp_thread);
292 /* are we a task leader? */
293 if (p == p->p_leader) {
294 struct kill_args killArgs;
295 killArgs.signum = SIGKILL;
298 killArgs.pid = q->p_pid;
300 * The interface for kill is better
301 * than the internal signal
308 tsleep((caddr_t)p, 0, "exit1", 0);
314 STOPEVENT(p, S_EXIT, rv);
315 wakeup(&p->p_stype); /* Wakeup anyone in procfs' PIOCWAIT */
318 * Check if any loadable modules need anything done at process exit.
319 * e.g. SYSV IPC stuff
320 * XXX what if one of these generates an error?
322 TAILQ_FOREACH(ep, &exit_list, next)
325 if (p->p_flag & P_PROFIL)
328 * If parent is waiting for us to exit or exec,
329 * P_PPWAIT is set; we will wakeup the parent below.
331 p->p_flag &= ~(P_TRACED | P_PPWAIT);
332 SIGEMPTYSET(p->p_siglist);
333 SIGEMPTYSET(lp->lwp_siglist);
334 if (timevalisset(&p->p_realtimer.it_value))
335 callout_stop(&p->p_ithandle);
338 * Reset any sigio structures pointing to us as a result of
339 * F_SETOWN with our pid.
341 funsetownlst(&p->p_sigiolst);
344 * Close open files and release open-file table.
350 if(p->p_leader->p_peers) {
352 while(q->p_peers != p)
354 q->p_peers = p->p_peers;
355 wakeup((caddr_t)p->p_leader);
359 * XXX Shutdown SYSV semaphores
363 KKASSERT(p->p_numposixlocks == 0);
365 /* The next two chunks should probably be moved to vmspace_exit. */
369 * Release upcalls associated with this process
375 * Clean up data related to virtual kernel operation. Clean up
376 * any vkernel context related to the current lwp now so we can
380 vkernel_lwp_exit(lp);
385 * Release user portion of address space.
386 * This releases references to vnodes,
387 * which could cause I/O if the file has been unlinked.
388 * Need to do this early enough that we can still sleep.
389 * Can't free the entire vmspace as the kernel stack
390 * may be mapped within that space also.
392 * Processes sharing the same vmspace may exit in one order, and
393 * get cleaned up by vmspace_exit() in a different order. The
394 * last exiting process to reach this point releases as much of
395 * the environment as it can, and the last process cleaned up
396 * by vmspace_exit() (which decrements exitingcnt) cleans up the
400 sysref_put(&vm->vm_sysref);
402 if (SESS_LEADER(p)) {
403 struct session *sp = p->p_session;
407 * We are the controlling process. Signal the
408 * foreground process group, drain the controlling
409 * terminal, and revoke access to the controlling
412 * NOTE: while waiting for the process group to exit
413 * it is possible that one of the processes in the
414 * group will revoke the tty, so the ttyclosesession()
415 * function will re-check sp->s_ttyvp.
417 if (sp->s_ttyp && (sp->s_ttyp->t_session == sp)) {
418 if (sp->s_ttyp->t_pgrp)
419 pgsignal(sp->s_ttyp->t_pgrp, SIGHUP, 1);
421 ttyclosesession(sp, 1); /* also revoke */
424 * Release the tty. If someone has it open via
425 * /dev/tty then close it (since they no longer can
426 * once we've NULL'd it out).
428 ttyclosesession(sp, 0);
431 * s_ttyp is not zero'd; we use this to indicate
432 * that the session once had a controlling terminal.
433 * (for logging and informational purposes)
438 fixjobc(p, p->p_pgrp, 0);
439 (void)acct_process(p);
445 ktrdestroy(&p->p_tracenode);
449 * Release reference to text vnode
451 if ((vtmp = p->p_textvp) != NULL) {
457 * Move the process to the zombie list. This will block
458 * until the process p_lock count reaches 0. The process will
459 * not be reaped until TDF_EXITING is set by cpu_thread_exit(),
460 * which is called from cpu_proc_exit().
462 proc_move_allproc_zombie(p);
464 q = LIST_FIRST(&p->p_children);
465 if (q) /* only need this if any child is S_ZOMB */
466 wakeup((caddr_t) initproc);
467 for (; q != 0; q = nq) {
468 nq = LIST_NEXT(q, p_sibling);
469 LIST_REMOVE(q, p_sibling);
470 LIST_INSERT_HEAD(&initproc->p_children, q, p_sibling);
471 q->p_pptr = initproc;
472 q->p_sigparent = SIGCHLD;
474 * Traced processes are killed
475 * since their existence means someone is screwing up.
477 if (q->p_flag & P_TRACED) {
478 q->p_flag &= ~P_TRACED;
484 * Save exit status and final rusage info, adding in child rusage
485 * info and self times.
488 calcru_proc(p, &p->p_ru);
489 ruadd(&p->p_ru, &p->p_cru);
492 * notify interested parties of our demise.
494 KNOTE(&p->p_klist, NOTE_EXIT);
497 * Notify parent that we're gone. If parent has the PS_NOCLDWAIT
498 * flag set, notify process 1 instead (and hope it will handle
501 if (p->p_pptr->p_sigacts->ps_flag & PS_NOCLDWAIT) {
502 struct proc *pp = p->p_pptr;
503 proc_reparent(p, initproc);
505 * If this was the last child of our parent, notify
506 * parent, so in case he was wait(2)ing, he will
509 if (LIST_EMPTY(&pp->p_children))
513 if (p->p_sigparent && p->p_pptr != initproc) {
514 ksignal(p->p_pptr, p->p_sigparent);
516 ksignal(p->p_pptr, SIGCHLD);
519 wakeup((caddr_t)p->p_pptr);
521 * cpu_exit is responsible for clearing curproc, since
522 * it is heavily integrated with the thread/switching sequence.
524 * Other substructures are freed from wait().
526 plimit_free(&p->p_limit);
529 * Release the current user process designation on the process so
530 * the userland scheduler can work in someone else.
532 p->p_usched->release_curproc(lp);
535 * Finally, call machine-dependent code to release as many of the
536 * lwp's resources as we can and halt execution of this thread.
542 lwp_exit(int masterexit)
544 struct lwp *lp = curthread->td_lwp;
545 struct proc *p = lp->lwp_proc;
548 * lwp_exit() may be called without setting LWP_WEXIT, so
549 * make sure it is set here.
551 lp->lwp_flag |= LWP_WEXIT;
554 * Clean up any virtualization
557 vkernel_lwp_exit(lp);
560 * Nobody actually wakes us when the lock
561 * count reaches zero, so just wait one tick.
563 while (lp->lwp_lock > 0)
564 tsleep(lp, 0, "lwpexit", 1);
566 /* Hand down resource usage to our proc */
567 ruadd(&p->p_ru, &lp->lwp_ru);
570 * If we don't hold the process until the LWP is reaped wait*()
571 * may try to dispose of its vmspace before all the LWPs have
572 * actually terminated.
577 * We have to use the reaper for all the LWPs except the one doing
578 * the master exit. The LWP doing the master exit can just be
579 * left on p_lwps and the process reaper will deal with it
580 * synchronously, which is much faster.
582 if (masterexit == 0) {
583 LIST_REMOVE(lp, lwp_list);
585 wakeup(&p->p_nthreads);
586 LIST_INSERT_HEAD(&deadlwp_list[mycpuid], lp, lwp_list);
587 taskqueue_enqueue(taskqueue_thread[mycpuid], deadlwp_task[mycpuid]);
595 * Wait until a lwp is completely dead.
597 * If the thread is still executing, which can't be waited upon,
598 * return failure. The caller is responsible of waiting a little
599 * bit and checking again.
602 * while (!lwp_wait(lp))
603 * tsleep(lp, 0, "lwpwait", 1);
606 lwp_wait(struct lwp *lp)
608 struct thread *td = lp->lwp_thread;;
610 KKASSERT(lwkt_preempted_proc() != lp);
612 while (lp->lwp_lock > 0)
613 tsleep(lp, 0, "lwpwait1", 1);
618 * The lwp's thread may still be in the middle
619 * of switching away, we can't rip its stack out from
620 * under it until TDF_EXITING is set and both
621 * TDF_RUNNING and TDF_PREEMPT_LOCK are clear.
622 * TDF_PREEMPT_LOCK must be checked because TDF_RUNNING
623 * will be cleared temporarily if a thread gets
626 * YYY no wakeup occurs, so we simply return failure
627 * and let the caller deal with sleeping and calling
630 if ((td->td_flags & (TDF_RUNNING|TDF_PREEMPT_LOCK|TDF_EXITING)) !=
638 * Release the resources associated with a lwp.
639 * The lwp must be completely dead.
642 lwp_dispose(struct lwp *lp)
644 struct thread *td = lp->lwp_thread;;
646 KKASSERT(lwkt_preempted_proc() != lp);
647 KKASSERT(td->td_refs == 0);
648 KKASSERT((td->td_flags & (TDF_RUNNING|TDF_PREEMPT_LOCK|TDF_EXITING)) ==
656 lp->lwp_thread = NULL;
657 lwkt_free_thread(td);
663 sys_wait4(struct wait_args *uap)
665 struct rusage rusage;
668 error = kern_wait(uap->pid, uap->status ? &status : NULL,
669 uap->options, uap->rusage ? &rusage : NULL, &uap->sysmsg_fds[0]);
671 if (error == 0 && uap->status)
672 error = copyout(&status, uap->status, sizeof(*uap->status));
673 if (error == 0 && uap->rusage)
674 error = copyout(&rusage, uap->rusage, sizeof(*uap->rusage));
681 * wait_args(int pid, int *status, int options, struct rusage *rusage)
684 kern_wait(pid_t pid, int *status, int options, struct rusage *rusage, int *res)
686 struct thread *td = curthread;
687 struct proc *q = td->td_proc;
693 if (options &~ (WUNTRACED|WNOHANG|WLINUXCLONE))
697 * Hack for backwards compatibility with badly written user code.
698 * Or perhaps we have to do this anyway, it is unclear. XXX
700 * The problem is that if a process group is stopped and the parent
701 * is doing a wait*(..., WUNTRACED, ...), it will see the STOP
702 * of the child and then stop itself when it tries to return from the
703 * system call. When the process group is resumed the parent will
704 * then get the STOP status even though the child has now resumed
705 * (a followup wait*() will get the CONT status).
707 * Previously the CONT would overwrite the STOP because the tstop
708 * was handled within tsleep(), and the parent would only see
709 * the CONT when both are stopped and continued together. This litte
710 * two-line hack restores this effect.
712 while (q->p_stat == SSTOP)
716 LIST_FOREACH(p, &q->p_children, p_sibling) {
717 if (pid != WAIT_ANY &&
718 p->p_pid != pid && p->p_pgid != -pid)
721 /* This special case handles a kthread spawned by linux_clone
722 * (see linux_misc.c). The linux_wait4 and linux_waitpid
723 * functions need to be able to distinguish between waiting
724 * on a process and waiting on a thread. It is a thread if
725 * p_sigparent is not SIGCHLD, and the WLINUXCLONE option
726 * signifies we want to wait for threads and not processes.
728 if ((p->p_sigparent != SIGCHLD) ^
729 ((options & WLINUXCLONE) != 0)) {
734 if (p->p_stat == SZOMB) {
736 * Reap any LWPs left in p->p_lwps. This is usually
737 * just the last LWP. This must be done before
738 * we loop on p_lock since the lwps hold a ref on
739 * it as a vmspace interlock.
741 * Once that is accomplished p_nthreads had better
744 reaplwps(&p->p_lwps, 0);
745 KKASSERT(p->p_nthreads == 0);
748 * Don't do anything really bad until all references
749 * to the process go away. This may include other
750 * LWPs which are still in the process of being
751 * reaped. We can't just pull the rug out from under
752 * them because they may still be using the VM space.
754 * Certain kernel facilities such as /proc will also
755 * put a hold on the process for short periods of
759 tsleep(p, 0, "reap3", hz);
761 /* scheduling hook for heuristic */
762 /* XXX no lwp available, we need a different heuristic */
764 p->p_usched->heuristic_exiting(td->td_lwp, deadlp);
767 /* Take care of our return values. */
770 *status = p->p_xstat;
774 * If we got the child via a ptrace 'attach',
775 * we need to give it back to the old parent.
777 if (p->p_oppid && (t = pfind(p->p_oppid))) {
785 ruadd(&q->p_cru, &p->p_ru);
788 * Decrement the count of procs running with this uid.
790 chgproccnt(p->p_ucred->cr_ruidinfo, -1, 0);
793 * Free up credentials.
799 * Remove unused arguments
801 if (p->p_args && --p->p_args->ar_ref == 0)
802 FREE(p->p_args, M_PARGS);
805 * Finally finished with old proc entry.
806 * Unlink it from its process group and free it.
808 proc_remove_zombie(p);
811 if (--p->p_sigacts->ps_refcnt == 0) {
812 kfree(p->p_sigacts, M_SUBPROC);
821 if (p->p_stat == SSTOP && (p->p_flag & P_WAITED) == 0 &&
822 (p->p_flag & P_TRACED || options & WUNTRACED)) {
823 p->p_flag |= P_WAITED;
827 *status = W_STOPCODE(p->p_xstat);
828 /* Zero rusage so we get something consistent. */
830 bzero(rusage, sizeof(rusage));
836 if (options & WNOHANG) {
840 error = tsleep((caddr_t)q, PCATCH, "wait", 0);
847 * make process 'parent' the new parent of process 'child'.
850 proc_reparent(struct proc *child, struct proc *parent)
853 if (child->p_pptr == parent)
856 LIST_REMOVE(child, p_sibling);
857 LIST_INSERT_HEAD(&parent->p_children, child, p_sibling);
858 child->p_pptr = parent;
862 * The next two functions are to handle adding/deleting items on the
866 * Take the arguments given and put them onto the exit callout list,
867 * However first make sure that it's not already there.
868 * returns 0 on success.
872 at_exit(exitlist_fn function)
877 /* Be noisy if the programmer has lost track of things */
878 if (rm_at_exit(function))
879 kprintf("WARNING: exit callout entry (%p) already present\n",
882 ep = kmalloc(sizeof(*ep), M_ATEXIT, M_NOWAIT);
885 ep->function = function;
886 TAILQ_INSERT_TAIL(&exit_list, ep, next);
891 * Scan the exit callout list for the given item and remove it.
892 * Returns the number of items removed (0 or 1)
895 rm_at_exit(exitlist_fn function)
899 TAILQ_FOREACH(ep, &exit_list, next) {
900 if (ep->function == function) {
901 TAILQ_REMOVE(&exit_list, ep, next);
910 * LWP reaper related code.
913 reaplwps(void *context, int dummy)
915 struct lwplist *lwplist = context;
918 while ((lp = LIST_FIRST(lwplist))) {
919 LIST_REMOVE(lp, lwp_list);
920 while (lwp_wait(lp) == 0)
921 tsleep(lp, 0, "lwpreap", 1);
931 for (cpu = 0; cpu < ncpus; cpu++) {
932 LIST_INIT(&deadlwp_list[cpu]);
933 deadlwp_task[cpu] = kmalloc(sizeof(*deadlwp_task[cpu]), M_DEVBUF, M_WAITOK);
934 TASK_INIT(deadlwp_task[cpu], 0, reaplwps, &deadlwp_list[cpu]);
938 SYSINIT(deadlwpinit, SI_SUB_CONFIGURE, SI_ORDER_ANY, deadlwp_init, NULL);