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 1.83 2007/07/03 17:22:14 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);
86 static MALLOC_DEFINE(M_ATEXIT, "atexit", "atexit callback");
87 static MALLOC_DEFINE(M_ZOMBIE, "zombie", "zombie proc status");
90 * callout list for things to do at exit time
94 TAILQ_ENTRY(exitlist) next;
97 TAILQ_HEAD(exit_list_head, exitlist);
98 static struct exit_list_head exit_list = TAILQ_HEAD_INITIALIZER(exit_list);
103 struct task *deadlwp_task[MAXCPU];
104 struct lwplist deadlwp_list[MAXCPU];
110 * SYS_EXIT_ARGS(int rval)
113 sys_exit(struct exit_args *uap)
115 exit1(W_EXITCODE(uap->rval, 0));
121 * Death of a lwp or process with optional bells and whistles.
124 sys_extexit(struct extexit_args *uap)
129 action = EXTEXIT_ACTION(uap->how);
130 who = EXTEXIT_WHO(uap->how);
132 /* Check parameters before we might perform some action */
147 error = copyout(&uap->status, uap->addr, sizeof(uap->status));
159 * Be sure only to perform a simple lwp exit if there is at
160 * least one more lwp in the proc, which will call exit1()
161 * later, otherwise the proc will be an UNDEAD and not even a
164 if (curproc->p_nthreads > 1) {
168 /* else last lwp in proc: do the real thing */
171 default: /* to help gcc */
173 exit1(W_EXITCODE(uap->status, 0));
181 * Kill all LWPs except the current one. Do not try to signal
182 * LWPs which have exited on their own or have already been
186 killlwps(struct lwp *lp)
188 struct proc *p = lp->lwp_proc;
192 * Kill the remaining LWPs, interlock with LWP_WEXIT.
194 FOREACH_LWP_IN_PROC(tlp, p) {
195 if ((tlp->lwp_flag & LWP_WEXIT) == 0) {
196 lwpsignal(p, tlp, SIGKILL);
197 tlp->lwp_flag |= LWP_WEXIT;
202 * Wait for everything to clear out.
204 while (p->p_nthreads > 1) {
206 kprintf("killlwps: waiting for %d lwps of pid "
208 p->p_nthreads - 1, p->p_pid);
209 tsleep(&p->p_nthreads, 0, "killlwps", hz);
214 * Exit: deallocate address space and other resources, change proc state
215 * to zombie, and unlink proc from allproc and parent's lists. Save exit
216 * status and rusage for wait(). Check for child processes and orphan them.
221 struct thread *td = curthread;
222 struct proc *p = td->td_proc;
223 struct lwp *lp = td->td_lwp;
230 kprintf("init died (signal %d, exit %d)\n",
231 WTERMSIG(rv), WEXITSTATUS(rv));
232 panic("Going nowhere without my init!");
236 * Interlock against P_WEXIT. Only one of the process's thread
237 * is allowed to do the master exit.
239 if (p->p_flag & P_WEXIT) {
243 p->p_flag |= P_WEXIT;
246 * Interlock with LWP_WEXIT and kill any remaining LWPs
248 lp->lwp_flag |= LWP_WEXIT;
249 if (p->p_nthreads > 1)
252 caps_exit(lp->lwp_thread);
255 /* are we a task leader? */
256 if (p == p->p_leader) {
257 struct kill_args killArgs;
258 killArgs.signum = SIGKILL;
261 killArgs.pid = q->p_pid;
263 * The interface for kill is better
264 * than the internal signal
271 tsleep((caddr_t)p, 0, "exit1", 0);
277 STOPEVENT(p, S_EXIT, rv);
278 wakeup(&p->p_stype); /* Wakeup anyone in procfs' PIOCWAIT */
281 * Check if any loadable modules need anything done at process exit.
282 * e.g. SYSV IPC stuff
283 * XXX what if one of these generates an error?
285 TAILQ_FOREACH(ep, &exit_list, next)
288 if (p->p_flag & P_PROFIL)
291 * If parent is waiting for us to exit or exec,
292 * P_PPWAIT is set; we will wakeup the parent below.
294 p->p_flag &= ~(P_TRACED | P_PPWAIT);
295 SIGEMPTYSET(p->p_siglist);
296 SIGEMPTYSET(lp->lwp_siglist);
297 if (timevalisset(&p->p_realtimer.it_value))
298 callout_stop(&p->p_ithandle);
301 * Reset any sigio structures pointing to us as a result of
302 * F_SETOWN with our pid.
304 funsetownlst(&p->p_sigiolst);
307 * Close open files and release open-file table.
313 if(p->p_leader->p_peers) {
315 while(q->p_peers != p)
317 q->p_peers = p->p_peers;
318 wakeup((caddr_t)p->p_leader);
322 * XXX Shutdown SYSV semaphores
326 KKASSERT(p->p_numposixlocks == 0);
328 /* The next two chunks should probably be moved to vmspace_exit. */
332 * Release upcalls associated with this process
338 * Clean up data related to virtual kernel operation. Clean up
339 * any vkernel context related to the current lwp now so we can
343 vkernel_lwp_exit(lp);
348 * Release user portion of address space.
349 * This releases references to vnodes,
350 * which could cause I/O if the file has been unlinked.
351 * Need to do this early enough that we can still sleep.
352 * Can't free the entire vmspace as the kernel stack
353 * may be mapped within that space also.
355 * Processes sharing the same vmspace may exit in one order, and
356 * get cleaned up by vmspace_exit() in a different order. The
357 * last exiting process to reach this point releases as much of
358 * the environment as it can, and the last process cleaned up
359 * by vmspace_exit() (which decrements exitingcnt) cleans up the
363 sysref_put(&vm->vm_sysref);
365 if (SESS_LEADER(p)) {
366 struct session *sp = p->p_session;
370 * We are the controlling process. Signal the
371 * foreground process group, drain the controlling
372 * terminal, and revoke access to the controlling
375 * NOTE: while waiting for the process group to exit
376 * it is possible that one of the processes in the
377 * group will revoke the tty, so the ttyclosesession()
378 * function will re-check sp->s_ttyvp.
380 if (sp->s_ttyp && (sp->s_ttyp->t_session == sp)) {
381 if (sp->s_ttyp->t_pgrp)
382 pgsignal(sp->s_ttyp->t_pgrp, SIGHUP, 1);
384 ttyclosesession(sp, 1); /* also revoke */
387 * Release the tty. If someone has it open via
388 * /dev/tty then close it (since they no longer can
389 * once we've NULL'd it out).
391 ttyclosesession(sp, 0);
394 * s_ttyp is not zero'd; we use this to indicate
395 * that the session once had a controlling terminal.
396 * (for logging and informational purposes)
401 fixjobc(p, p->p_pgrp, 0);
402 (void)acct_process(p);
408 ktrdestroy(&p->p_tracenode);
412 * Release reference to text vnode
414 if ((vtmp = p->p_textvp) != NULL) {
420 * Move the process to the zombie list. This will block
421 * until the process p_lock count reaches 0. The process will
422 * not be reaped until TDF_EXITING is set by cpu_thread_exit(),
423 * which is called from cpu_proc_exit().
425 proc_move_allproc_zombie(p);
427 q = LIST_FIRST(&p->p_children);
428 if (q) /* only need this if any child is S_ZOMB */
429 wakeup((caddr_t) initproc);
430 for (; q != 0; q = nq) {
431 nq = LIST_NEXT(q, p_sibling);
432 LIST_REMOVE(q, p_sibling);
433 LIST_INSERT_HEAD(&initproc->p_children, q, p_sibling);
434 q->p_pptr = initproc;
435 q->p_sigparent = SIGCHLD;
437 * Traced processes are killed
438 * since their existence means someone is screwing up.
440 if (q->p_flag & P_TRACED) {
441 q->p_flag &= ~P_TRACED;
447 * Save exit status and final rusage info, adding in child rusage
448 * info and self times.
451 calcru_proc(p, &p->p_ru);
452 ruadd(&p->p_ru, &p->p_cru);
455 * notify interested parties of our demise.
457 KNOTE(&p->p_klist, NOTE_EXIT);
460 * Notify parent that we're gone. If parent has the PS_NOCLDWAIT
461 * flag set, notify process 1 instead (and hope it will handle
464 if (p->p_pptr->p_sigacts->ps_flag & PS_NOCLDWAIT) {
465 struct proc *pp = p->p_pptr;
466 proc_reparent(p, initproc);
468 * If this was the last child of our parent, notify
469 * parent, so in case he was wait(2)ing, he will
472 if (LIST_EMPTY(&pp->p_children))
476 if (p->p_sigparent && p->p_pptr != initproc) {
477 ksignal(p->p_pptr, p->p_sigparent);
479 ksignal(p->p_pptr, SIGCHLD);
482 wakeup((caddr_t)p->p_pptr);
484 * cpu_exit is responsible for clearing curproc, since
485 * it is heavily integrated with the thread/switching sequence.
487 * Other substructures are freed from wait().
489 plimit_free(&p->p_limit);
492 * Release the current user process designation on the process so
493 * the userland scheduler can work in someone else.
495 p->p_usched->release_curproc(lp);
498 * Finally, call machine-dependent code to release as many of the
499 * lwp's resources as we can and halt execution of this thread.
505 lwp_exit(int masterexit)
507 struct lwp *lp = curthread->td_lwp;
508 struct proc *p = lp->lwp_proc;
511 * lwp_exit() may be called without setting LWP_WEXIT, so
512 * make sure it is set here.
514 lp->lwp_flag |= LWP_WEXIT;
517 * Clean up any virtualization
520 vkernel_lwp_exit(lp);
523 * Nobody actually wakes us when the lock
524 * count reaches zero, so just wait one tick.
526 while (lp->lwp_lock > 0)
527 tsleep(lp, 0, "lwpexit", 1);
529 /* Hand down resource usage to our proc */
530 ruadd(&p->p_ru, &lp->lwp_ru);
533 * If we don't hold the process until the LWP is reaped wait*()
534 * may try to dispose of its vmspace before all the LWPs have
535 * actually terminated.
540 * We have to use the reaper for all the LWPs except the one doing
541 * the master exit. The LWP doing the master exit can just be
542 * left on p_lwps and the process reaper will deal with it
543 * synchronously, which is much faster.
545 if (masterexit == 0) {
546 LIST_REMOVE(lp, lwp_list);
548 wakeup(&p->p_nthreads);
549 LIST_INSERT_HEAD(&deadlwp_list[mycpuid], lp, lwp_list);
550 taskqueue_enqueue(taskqueue_thread[mycpuid], deadlwp_task[mycpuid]);
558 * Wait until a lwp is completely dead.
560 * If the thread is still executing, which can't be waited upon,
561 * return failure. The caller is responsible of waiting a little
562 * bit and checking again.
565 * while (!lwp_wait(lp))
566 * tsleep(lp, 0, "lwpwait", 1);
569 lwp_wait(struct lwp *lp)
571 struct thread *td = lp->lwp_thread;;
573 KKASSERT(lwkt_preempted_proc() != lp);
575 while (lp->lwp_lock > 0)
576 tsleep(lp, 0, "lwpwait1", 1);
581 * The lwp's thread may still be in the middle
582 * of switching away, we can't rip its stack out from
583 * under it until TDF_EXITING is set and both
584 * TDF_RUNNING and TDF_PREEMPT_LOCK are clear.
585 * TDF_PREEMPT_LOCK must be checked because TDF_RUNNING
586 * will be cleared temporarily if a thread gets
589 * YYY no wakeup occurs, so we simply return failure
590 * and let the caller deal with sleeping and calling
593 if ((td->td_flags & (TDF_RUNNING|TDF_PREEMPT_LOCK|TDF_EXITING)) !=
601 * Release the resources associated with a lwp.
602 * The lwp must be completely dead.
605 lwp_dispose(struct lwp *lp)
607 struct thread *td = lp->lwp_thread;;
609 KKASSERT(lwkt_preempted_proc() != lp);
610 KKASSERT(td->td_refs == 0);
611 KKASSERT((td->td_flags & (TDF_RUNNING|TDF_PREEMPT_LOCK|TDF_EXITING)) ==
619 lp->lwp_thread = NULL;
620 lwkt_free_thread(td);
626 sys_wait4(struct wait_args *uap)
628 struct rusage rusage;
631 error = kern_wait(uap->pid, uap->status ? &status : NULL,
632 uap->options, uap->rusage ? &rusage : NULL, &uap->sysmsg_fds[0]);
634 if (error == 0 && uap->status)
635 error = copyout(&status, uap->status, sizeof(*uap->status));
636 if (error == 0 && uap->rusage)
637 error = copyout(&rusage, uap->rusage, sizeof(*uap->rusage));
644 * wait_args(int pid, int *status, int options, struct rusage *rusage)
647 kern_wait(pid_t pid, int *status, int options, struct rusage *rusage, int *res)
649 struct thread *td = curthread;
650 struct proc *q = td->td_proc;
656 if (options &~ (WUNTRACED|WNOHANG|WLINUXCLONE))
660 * Hack for backwards compatibility with badly written user code.
661 * Or perhaps we have to do this anyway, it is unclear. XXX
663 * The problem is that if a process group is stopped and the parent
664 * is doing a wait*(..., WUNTRACED, ...), it will see the STOP
665 * of the child and then stop itself when it tries to return from the
666 * system call. When the process group is resumed the parent will
667 * then get the STOP status even though the child has now resumed
668 * (a followup wait*() will get the CONT status).
670 * Previously the CONT would overwrite the STOP because the tstop
671 * was handled within tsleep(), and the parent would only see
672 * the CONT when both are stopped and continued together. This litte
673 * two-line hack restores this effect.
675 while (q->p_stat == SSTOP)
679 LIST_FOREACH(p, &q->p_children, p_sibling) {
680 if (pid != WAIT_ANY &&
681 p->p_pid != pid && p->p_pgid != -pid)
684 /* This special case handles a kthread spawned by linux_clone
685 * (see linux_misc.c). The linux_wait4 and linux_waitpid
686 * functions need to be able to distinguish between waiting
687 * on a process and waiting on a thread. It is a thread if
688 * p_sigparent is not SIGCHLD, and the WLINUXCLONE option
689 * signifies we want to wait for threads and not processes.
691 if ((p->p_sigparent != SIGCHLD) ^
692 ((options & WLINUXCLONE) != 0)) {
697 if (p->p_stat == SZOMB) {
699 * Reap any LWPs left in p->p_lwps. This is usually
700 * just the last LWP. This must be done before
701 * we loop on p_lock since the lwps hold a ref on
702 * it as a vmspace interlock.
704 * Once that is accomplished p_nthreads had better
707 reaplwps(&p->p_lwps, 0);
708 KKASSERT(p->p_nthreads == 0);
711 * Don't do anything really bad until all references
712 * to the process go away. This may include other
713 * LWPs which are still in the process of being
714 * reaped. We can't just pull the rug out from under
715 * them because they may still be using the VM space.
717 * Certain kernel facilities such as /proc will also
718 * put a hold on the process for short periods of
722 tsleep(p, 0, "reap3", hz);
724 /* scheduling hook for heuristic */
725 /* XXX no lwp available, we need a different heuristic */
727 p->p_usched->heuristic_exiting(td->td_lwp, deadlp);
730 /* Take care of our return values. */
733 *status = p->p_xstat;
737 * If we got the child via a ptrace 'attach',
738 * we need to give it back to the old parent.
740 if (p->p_oppid && (t = pfind(p->p_oppid))) {
748 ruadd(&q->p_cru, &p->p_ru);
751 * Decrement the count of procs running with this uid.
753 chgproccnt(p->p_ucred->cr_ruidinfo, -1, 0);
756 * Free up credentials.
762 * Remove unused arguments
764 if (p->p_args && --p->p_args->ar_ref == 0)
765 FREE(p->p_args, M_PARGS);
768 * Finally finished with old proc entry.
769 * Unlink it from its process group and free it.
771 proc_remove_zombie(p);
774 if (--p->p_sigacts->ps_refcnt == 0) {
775 kfree(p->p_sigacts, M_SUBPROC);
784 if (p->p_stat == SSTOP && (p->p_flag & P_WAITED) == 0 &&
785 (p->p_flag & P_TRACED || options & WUNTRACED)) {
786 p->p_flag |= P_WAITED;
790 *status = W_STOPCODE(p->p_xstat);
791 /* Zero rusage so we get something consistent. */
793 bzero(rusage, sizeof(rusage));
799 if (options & WNOHANG) {
803 error = tsleep((caddr_t)q, PCATCH, "wait", 0);
810 * make process 'parent' the new parent of process 'child'.
813 proc_reparent(struct proc *child, struct proc *parent)
816 if (child->p_pptr == parent)
819 LIST_REMOVE(child, p_sibling);
820 LIST_INSERT_HEAD(&parent->p_children, child, p_sibling);
821 child->p_pptr = parent;
825 * The next two functions are to handle adding/deleting items on the
829 * Take the arguments given and put them onto the exit callout list,
830 * However first make sure that it's not already there.
831 * returns 0 on success.
835 at_exit(exitlist_fn function)
840 /* Be noisy if the programmer has lost track of things */
841 if (rm_at_exit(function))
842 kprintf("WARNING: exit callout entry (%p) already present\n",
845 ep = kmalloc(sizeof(*ep), M_ATEXIT, M_NOWAIT);
848 ep->function = function;
849 TAILQ_INSERT_TAIL(&exit_list, ep, next);
854 * Scan the exit callout list for the given item and remove it.
855 * Returns the number of items removed (0 or 1)
858 rm_at_exit(exitlist_fn function)
862 TAILQ_FOREACH(ep, &exit_list, next) {
863 if (ep->function == function) {
864 TAILQ_REMOVE(&exit_list, ep, next);
873 * LWP reaper related code.
876 reaplwps(void *context, int dummy)
878 struct lwplist *lwplist = context;
881 while ((lp = LIST_FIRST(lwplist))) {
882 LIST_REMOVE(lp, lwp_list);
883 while (lwp_wait(lp) == 0)
884 tsleep(lp, 0, "lwpreap", 1);
894 for (cpu = 0; cpu < ncpus; cpu++) {
895 LIST_INIT(&deadlwp_list[cpu]);
896 deadlwp_task[cpu] = kmalloc(sizeof(*deadlwp_task[cpu]), M_DEVBUF, M_WAITOK);
897 TASK_INIT(deadlwp_task[cpu], 0, reaplwps, &deadlwp_list[cpu]);
901 SYSINIT(deadlwpinit, SI_SUB_CONFIGURE, SI_ORDER_ANY, deadlwp_init, NULL);