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 1.91 2008/05/18 20:02:02 nth 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_extern.h>
80 #include <sys/thread2.h>
81 #include <sys/sysref2.h>
83 static void reaplwps(void *context, int dummy);
84 static void reaplwp(struct lwp *lp);
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)
116 sys_exit(struct exit_args *uap)
119 exit1(W_EXITCODE(uap->rval, 0));
126 * Death of a lwp or process with optional bells and whistles.
131 sys_extexit(struct extexit_args *uap)
136 action = EXTEXIT_ACTION(uap->how);
137 who = EXTEXIT_WHO(uap->how);
139 /* Check parameters before we might perform some action */
152 error = copyout(&uap->status, uap->addr, sizeof(uap->status));
165 * Be sure only to perform a simple lwp exit if there is at
166 * least one more lwp in the proc, which will call exit1()
167 * later, otherwise the proc will be an UNDEAD and not even a
170 if (curproc->p_nthreads > 1) {
174 /* else last lwp in proc: do the real thing */
176 default: /* to help gcc */
178 exit1(W_EXITCODE(uap->status, 0));
183 rel_mplock(); /* safety */
187 * Kill all lwps associated with the current process except the
188 * current lwp. Return an error if we race another thread trying to
189 * do the same thing and lose the race.
191 * If forexec is non-zero the current thread and process flags are
192 * cleaned up so they can be reused.
195 killalllwps(int forexec)
197 struct lwp *lp = curthread->td_lwp;
198 struct proc *p = lp->lwp_proc;
201 * Interlock against P_WEXIT. Only one of the process's thread
202 * is allowed to do the master exit.
204 if (p->p_flag & P_WEXIT)
206 p->p_flag |= P_WEXIT;
209 * Interlock with LWP_WEXIT and kill any remaining LWPs
211 lp->lwp_flag |= LWP_WEXIT;
212 if (p->p_nthreads > 1)
216 * If doing this for an exec, clean up the remaining thread
217 * (us) for continuing operation after all the other threads
221 lp->lwp_flag &= ~LWP_WEXIT;
222 p->p_flag &= ~P_WEXIT;
228 * Kill all LWPs except the current one. Do not try to signal
229 * LWPs which have exited on their own or have already been
233 killlwps(struct lwp *lp)
235 struct proc *p = lp->lwp_proc;
239 * Kill the remaining LWPs. We must send the signal before setting
240 * LWP_WEXIT. The setting of WEXIT is optional but helps reduce
241 * races. tlp must be held across the call as it might block and
242 * allow the target lwp to rip itself out from under our loop.
244 FOREACH_LWP_IN_PROC(tlp, p) {
246 if ((tlp->lwp_flag & LWP_WEXIT) == 0) {
247 lwpsignal(p, tlp, SIGKILL);
248 tlp->lwp_flag |= LWP_WEXIT;
254 * Wait for everything to clear out.
256 while (p->p_nthreads > 1) {
257 tsleep(&p->p_nthreads, 0, "killlwps", 0);
262 * Exit: deallocate address space and other resources, change proc state
263 * to zombie, and unlink proc from allproc and parent's lists. Save exit
264 * status and rusage for wait(). Check for child processes and orphan them.
269 struct thread *td = curthread;
270 struct proc *p = td->td_proc;
271 struct lwp *lp = td->td_lwp;
279 kprintf("init died (signal %d, exit %d)\n",
280 WTERMSIG(rv), WEXITSTATUS(rv));
281 panic("Going nowhere without my init!");
284 varsymset_clean(&p->p_varsymset);
285 lockuninit(&p->p_varsymset.vx_lock);
287 * Kill all lwps associated with the current process, return an
288 * error if we race another thread trying to do the same thing
291 error = killalllwps(0);
297 caps_exit(lp->lwp_thread);
300 /* are we a task leader? */
301 if (p == p->p_leader) {
302 struct kill_args killArgs;
303 killArgs.signum = SIGKILL;
306 killArgs.pid = q->p_pid;
308 * The interface for kill is better
309 * than the internal signal
316 tsleep((caddr_t)p, 0, "exit1", 0);
322 STOPEVENT(p, S_EXIT, rv);
323 wakeup(&p->p_stype); /* Wakeup anyone in procfs' PIOCWAIT */
326 * Check if any loadable modules need anything done at process exit.
327 * e.g. SYSV IPC stuff
328 * XXX what if one of these generates an error?
330 TAILQ_FOREACH(ep, &exit_list, next)
333 if (p->p_flag & P_PROFIL)
336 * If parent is waiting for us to exit or exec,
337 * P_PPWAIT is set; we will wakeup the parent below.
339 p->p_flag &= ~(P_TRACED | P_PPWAIT);
340 SIGEMPTYSET(p->p_siglist);
341 SIGEMPTYSET(lp->lwp_siglist);
342 if (timevalisset(&p->p_realtimer.it_value))
343 callout_stop(&p->p_ithandle);
346 * Reset any sigio structures pointing to us as a result of
347 * F_SETOWN with our pid.
349 funsetownlst(&p->p_sigiolst);
352 * Close open files and release open-file table.
357 if(p->p_leader->p_peers) {
359 while(q->p_peers != p)
361 q->p_peers = p->p_peers;
362 wakeup((caddr_t)p->p_leader);
366 * XXX Shutdown SYSV semaphores
370 KKASSERT(p->p_numposixlocks == 0);
372 /* The next two chunks should probably be moved to vmspace_exit. */
376 * Release upcalls associated with this process
382 * Clean up data related to virtual kernel operation. Clean up
383 * any vkernel context related to the current lwp now so we can
387 vkernel_lwp_exit(lp);
392 * Release user portion of address space.
393 * This releases references to vnodes,
394 * which could cause I/O if the file has been unlinked.
395 * Need to do this early enough that we can still sleep.
396 * Can't free the entire vmspace as the kernel stack
397 * may be mapped within that space also.
399 * Processes sharing the same vmspace may exit in one order, and
400 * get cleaned up by vmspace_exit() in a different order. The
401 * last exiting process to reach this point releases as much of
402 * the environment as it can, and the last process cleaned up
403 * by vmspace_exit() (which decrements exitingcnt) cleans up the
407 sysref_put(&vm->vm_sysref);
409 if (SESS_LEADER(p)) {
410 struct session *sp = p->p_session;
414 * We are the controlling process. Signal the
415 * foreground process group, drain the controlling
416 * terminal, and revoke access to the controlling
419 * NOTE: while waiting for the process group to exit
420 * it is possible that one of the processes in the
421 * group will revoke the tty, so the ttyclosesession()
422 * function will re-check sp->s_ttyvp.
424 if (sp->s_ttyp && (sp->s_ttyp->t_session == sp)) {
425 if (sp->s_ttyp->t_pgrp)
426 pgsignal(sp->s_ttyp->t_pgrp, SIGHUP, 1);
428 ttyclosesession(sp, 1); /* also revoke */
431 * Release the tty. If someone has it open via
432 * /dev/tty then close it (since they no longer can
433 * once we've NULL'd it out).
435 ttyclosesession(sp, 0);
438 * s_ttyp is not zero'd; we use this to indicate
439 * that the session once had a controlling terminal.
440 * (for logging and informational purposes)
445 fixjobc(p, p->p_pgrp, 0);
446 (void)acct_process(p);
452 ktrdestroy(&p->p_tracenode);
456 * Release reference to text vnode
458 if ((vtmp = p->p_textvp) != NULL) {
464 * Move the process to the zombie list. This will block
465 * until the process p_lock count reaches 0. The process will
466 * not be reaped until TDF_EXITING is set by cpu_thread_exit(),
467 * which is called from cpu_proc_exit().
469 proc_move_allproc_zombie(p);
471 q = LIST_FIRST(&p->p_children);
472 if (q) /* only need this if any child is S_ZOMB */
473 wakeup((caddr_t) initproc);
474 for (; q != 0; q = nq) {
475 nq = LIST_NEXT(q, p_sibling);
476 LIST_REMOVE(q, p_sibling);
477 LIST_INSERT_HEAD(&initproc->p_children, q, p_sibling);
478 q->p_pptr = initproc;
479 q->p_sigparent = SIGCHLD;
481 * Traced processes are killed
482 * since their existence means someone is screwing up.
484 if (q->p_flag & P_TRACED) {
485 q->p_flag &= ~P_TRACED;
491 * Save exit status and final rusage info, adding in child rusage
492 * info and self times.
495 calcru_proc(p, &p->p_ru);
496 ruadd(&p->p_ru, &p->p_cru);
499 * notify interested parties of our demise.
501 KNOTE(&p->p_klist, NOTE_EXIT);
504 * Notify parent that we're gone. If parent has the PS_NOCLDWAIT
505 * flag set, notify process 1 instead (and hope it will handle
508 if (p->p_pptr->p_sigacts->ps_flag & PS_NOCLDWAIT) {
509 struct proc *pp = p->p_pptr;
510 proc_reparent(p, initproc);
512 * If this was the last child of our parent, notify
513 * parent, so in case he was wait(2)ing, he will
516 if (LIST_EMPTY(&pp->p_children))
520 if (p->p_sigparent && p->p_pptr != initproc) {
521 ksignal(p->p_pptr, p->p_sigparent);
523 ksignal(p->p_pptr, SIGCHLD);
526 wakeup((caddr_t)p->p_pptr);
528 * cpu_exit is responsible for clearing curproc, since
529 * it is heavily integrated with the thread/switching sequence.
531 * Other substructures are freed from wait().
536 * Release the current user process designation on the process so
537 * the userland scheduler can work in someone else.
539 p->p_usched->release_curproc(lp);
542 * Finally, call machine-dependent code to release as many of the
543 * lwp's resources as we can and halt execution of this thread.
549 * Eventually called by every exiting LWP
552 lwp_exit(int masterexit)
554 struct lwp *lp = curthread->td_lwp;
555 struct proc *p = lp->lwp_proc;
558 * lwp_exit() may be called without setting LWP_WEXIT, so
559 * make sure it is set here.
561 lp->lwp_flag |= LWP_WEXIT;
564 * Clean up any virtualization
567 vkernel_lwp_exit(lp);
570 * Clean up any syscall-cached ucred
572 if (lp->lwp_syscall_ucred) {
573 crfree(lp->lwp_syscall_ucred);
574 lp->lwp_syscall_ucred = NULL;
578 * Nobody actually wakes us when the lock
579 * count reaches zero, so just wait one tick.
581 while (lp->lwp_lock > 0)
582 tsleep(lp, 0, "lwpexit", 1);
584 /* Hand down resource usage to our proc */
585 ruadd(&p->p_ru, &lp->lwp_ru);
588 * If we don't hold the process until the LWP is reaped wait*()
589 * may try to dispose of its vmspace before all the LWPs have
590 * actually terminated.
595 * We have to use the reaper for all the LWPs except the one doing
596 * the master exit. The LWP doing the master exit can just be
597 * left on p_lwps and the process reaper will deal with it
598 * synchronously, which is much faster.
600 if (masterexit == 0) {
601 lwp_rb_tree_RB_REMOVE(&p->p_lwp_tree, lp);
603 wakeup(&p->p_nthreads);
604 LIST_INSERT_HEAD(&deadlwp_list[mycpuid], lp, u.lwp_reap_entry);
605 taskqueue_enqueue(taskqueue_thread[mycpuid], deadlwp_task[mycpuid]);
609 biosched_done(curthread);
614 * Wait until a lwp is completely dead.
616 * If the thread is still executing, which can't be waited upon,
617 * return failure. The caller is responsible of waiting a little
618 * bit and checking again.
621 * while (!lwp_wait(lp))
622 * tsleep(lp, 0, "lwpwait", 1);
625 lwp_wait(struct lwp *lp)
627 struct thread *td = lp->lwp_thread;;
629 KKASSERT(lwkt_preempted_proc() != lp);
631 while (lp->lwp_lock > 0)
632 tsleep(lp, 0, "lwpwait1", 1);
637 * The lwp's thread may still be in the middle
638 * of switching away, we can't rip its stack out from
639 * under it until TDF_EXITING is set and both
640 * TDF_RUNNING and TDF_PREEMPT_LOCK are clear.
641 * TDF_PREEMPT_LOCK must be checked because TDF_RUNNING
642 * will be cleared temporarily if a thread gets
645 * YYY no wakeup occurs, so we simply return failure
646 * and let the caller deal with sleeping and calling
649 if ((td->td_flags & (TDF_RUNNING|TDF_PREEMPT_LOCK|TDF_EXITING)) !=
657 * Release the resources associated with a lwp.
658 * The lwp must be completely dead.
661 lwp_dispose(struct lwp *lp)
663 struct thread *td = lp->lwp_thread;;
665 KKASSERT(lwkt_preempted_proc() != lp);
666 KKASSERT(td->td_refs == 0);
667 KKASSERT((td->td_flags & (TDF_RUNNING|TDF_PREEMPT_LOCK|TDF_EXITING)) ==
675 lp->lwp_thread = NULL;
676 lwkt_free_thread(td);
685 sys_wait4(struct wait_args *uap)
687 struct rusage rusage;
690 error = kern_wait(uap->pid, (uap->status ? &status : NULL),
691 uap->options, (uap->rusage ? &rusage : NULL),
692 &uap->sysmsg_result);
694 if (error == 0 && uap->status)
695 error = copyout(&status, uap->status, sizeof(*uap->status));
696 if (error == 0 && uap->rusage)
697 error = copyout(&rusage, uap->rusage, sizeof(*uap->rusage));
704 * wait_args(int pid, int *status, int options, struct rusage *rusage)
709 kern_wait(pid_t pid, int *status, int options, struct rusage *rusage, int *res)
711 struct thread *td = curthread;
713 struct proc *q = td->td_proc;
719 if (options &~ (WUNTRACED|WNOHANG|WCONTINUED|WLINUXCLONE))
724 * Hack for backwards compatibility with badly written user code.
725 * Or perhaps we have to do this anyway, it is unclear. XXX
727 * The problem is that if a process group is stopped and the parent
728 * is doing a wait*(..., WUNTRACED, ...), it will see the STOP
729 * of the child and then stop itself when it tries to return from the
730 * system call. When the process group is resumed the parent will
731 * then get the STOP status even though the child has now resumed
732 * (a followup wait*() will get the CONT status).
734 * Previously the CONT would overwrite the STOP because the tstop
735 * was handled within tsleep(), and the parent would only see
736 * the CONT when both are stopped and continued together. This litte
737 * two-line hack restores this effect.
739 while (q->p_stat == SSTOP)
743 LIST_FOREACH(p, &q->p_children, p_sibling) {
744 if (pid != WAIT_ANY &&
745 p->p_pid != pid && p->p_pgid != -pid)
748 /* This special case handles a kthread spawned by linux_clone
749 * (see linux_misc.c). The linux_wait4 and linux_waitpid
750 * functions need to be able to distinguish between waiting
751 * on a process and waiting on a thread. It is a thread if
752 * p_sigparent is not SIGCHLD, and the WLINUXCLONE option
753 * signifies we want to wait for threads and not processes.
755 if ((p->p_sigparent != SIGCHLD) ^
756 ((options & WLINUXCLONE) != 0)) {
761 if (p->p_stat == SZOMB) {
763 * We may go into SZOMB with threads still present.
764 * We must wait for them to exit before we can reap
765 * the master thread, otherwise we may race reaping
766 * non-master threads.
768 while (p->p_nthreads > 0) {
769 tsleep(&p->p_nthreads, 0, "lwpzomb", hz);
773 * Reap any LWPs left in p->p_lwps. This is usually
774 * just the last LWP. This must be done before
775 * we loop on p_lock since the lwps hold a ref on
776 * it as a vmspace interlock.
778 * Once that is accomplished p_nthreads had better
781 while ((lp = RB_ROOT(&p->p_lwp_tree)) != NULL) {
782 lwp_rb_tree_RB_REMOVE(&p->p_lwp_tree, lp);
785 KKASSERT(p->p_nthreads == 0);
788 * Don't do anything really bad until all references
789 * to the process go away. This may include other
790 * LWPs which are still in the process of being
791 * reaped. We can't just pull the rug out from under
792 * them because they may still be using the VM space.
794 * Certain kernel facilities such as /proc will also
795 * put a hold on the process for short periods of
799 tsleep(p, 0, "reap3", hz);
801 /* scheduling hook for heuristic */
802 /* XXX no lwp available, we need a different heuristic */
804 p->p_usched->heuristic_exiting(td->td_lwp, deadlp);
807 /* Take care of our return values. */
810 *status = p->p_xstat;
814 * If we got the child via a ptrace 'attach',
815 * we need to give it back to the old parent.
817 if (p->p_oppid && (t = pfind(p->p_oppid))) {
827 * Unlink the proc from its process group so that
828 * the following operations won't lead to an
829 * inconsistent state for processes running down
832 KKASSERT(p->p_lock == 0);
833 proc_remove_zombie(p);
837 ruadd(&q->p_cru, &p->p_ru);
840 * Decrement the count of procs running with this uid.
842 chgproccnt(p->p_ucred->cr_ruidinfo, -1, 0);
845 * Free up credentials.
851 * Remove unused arguments
853 if (p->p_args && --p->p_args->ar_ref == 0)
854 FREE(p->p_args, M_PARGS);
856 if (--p->p_sigacts->ps_refcnt == 0) {
857 kfree(p->p_sigacts, M_SUBPROC);
867 if (p->p_stat == SSTOP && (p->p_flag & P_WAITED) == 0 &&
868 (p->p_flag & P_TRACED || options & WUNTRACED)) {
869 p->p_flag |= P_WAITED;
873 *status = W_STOPCODE(p->p_xstat);
874 /* Zero rusage so we get something consistent. */
876 bzero(rusage, sizeof(rusage));
880 if (options & WCONTINUED && (p->p_flag & P_CONTINUED)) {
882 p->p_flag &= ~P_CONTINUED;
894 if (options & WNOHANG) {
899 error = tsleep((caddr_t)q, PCATCH, "wait", 0);
909 * make process 'parent' the new parent of process 'child'.
912 proc_reparent(struct proc *child, struct proc *parent)
915 if (child->p_pptr == parent)
918 LIST_REMOVE(child, p_sibling);
919 LIST_INSERT_HEAD(&parent->p_children, child, p_sibling);
920 child->p_pptr = parent;
924 * The next two functions are to handle adding/deleting items on the
928 * Take the arguments given and put them onto the exit callout list,
929 * However first make sure that it's not already there.
930 * returns 0 on success.
934 at_exit(exitlist_fn function)
939 /* Be noisy if the programmer has lost track of things */
940 if (rm_at_exit(function))
941 kprintf("WARNING: exit callout entry (%p) already present\n",
944 ep = kmalloc(sizeof(*ep), M_ATEXIT, M_NOWAIT);
947 ep->function = function;
948 TAILQ_INSERT_TAIL(&exit_list, ep, next);
953 * Scan the exit callout list for the given item and remove it.
954 * Returns the number of items removed (0 or 1)
957 rm_at_exit(exitlist_fn function)
961 TAILQ_FOREACH(ep, &exit_list, next) {
962 if (ep->function == function) {
963 TAILQ_REMOVE(&exit_list, ep, next);
972 * LWP reaper related code.
975 reaplwps(void *context, int dummy)
977 struct lwplist *lwplist = context;
981 while ((lp = LIST_FIRST(lwplist))) {
982 LIST_REMOVE(lp, u.lwp_reap_entry);
989 reaplwp(struct lwp *lp)
991 while (lwp_wait(lp) == 0)
992 tsleep(lp, 0, "lwpreap", 1);
1001 for (cpu = 0; cpu < ncpus; cpu++) {
1002 LIST_INIT(&deadlwp_list[cpu]);
1003 deadlwp_task[cpu] = kmalloc(sizeof(*deadlwp_task[cpu]), M_DEVBUF, M_WAITOK);
1004 TASK_INIT(deadlwp_task[cpu], 0, reaplwps, &deadlwp_list[cpu]);
1008 SYSINIT(deadlwpinit, SI_SUB_CONFIGURE, SI_ORDER_ANY, deadlwp_init, NULL);