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>
71 #include <sys/eventhandler.h>
74 #include <vm/vm_param.h>
77 #include <vm/vm_map.h>
78 #include <vm/vm_extern.h>
81 #include <sys/thread2.h>
82 #include <sys/sysref2.h>
83 #include <sys/mplock2.h>
85 static void reaplwps(void *context, int dummy);
86 static void reaplwp(struct lwp *lp);
87 static void killlwps(struct lwp *lp);
89 static MALLOC_DEFINE(M_ATEXIT, "atexit", "atexit callback");
90 static MALLOC_DEFINE(M_ZOMBIE, "zombie", "zombie proc status");
93 * callout list for things to do at exit time
97 TAILQ_ENTRY(exitlist) next;
100 TAILQ_HEAD(exit_list_head, exitlist);
101 static struct exit_list_head exit_list = TAILQ_HEAD_INITIALIZER(exit_list);
106 struct task *deadlwp_task[MAXCPU];
107 struct lwplist deadlwp_list[MAXCPU];
113 * SYS_EXIT_ARGS(int rval)
118 sys_exit(struct exit_args *uap)
121 exit1(W_EXITCODE(uap->rval, 0));
128 * Death of a lwp or process with optional bells and whistles.
133 sys_extexit(struct extexit_args *uap)
138 action = EXTEXIT_ACTION(uap->how);
139 who = EXTEXIT_WHO(uap->how);
141 /* Check parameters before we might perform some action */
154 error = copyout(&uap->status, uap->addr, sizeof(uap->status));
167 * Be sure only to perform a simple lwp exit if there is at
168 * least one more lwp in the proc, which will call exit1()
169 * later, otherwise the proc will be an UNDEAD and not even a
172 if (curproc->p_nthreads > 1) {
176 /* else last lwp in proc: do the real thing */
178 default: /* to help gcc */
180 exit1(W_EXITCODE(uap->status, 0));
185 rel_mplock(); /* safety */
189 * Kill all lwps associated with the current process except the
190 * current lwp. Return an error if we race another thread trying to
191 * do the same thing and lose the race.
193 * If forexec is non-zero the current thread and process flags are
194 * cleaned up so they can be reused.
197 killalllwps(int forexec)
199 struct lwp *lp = curthread->td_lwp;
200 struct proc *p = lp->lwp_proc;
203 * Interlock against P_WEXIT. Only one of the process's thread
204 * is allowed to do the master exit.
206 if (p->p_flag & P_WEXIT)
208 p->p_flag |= P_WEXIT;
211 * Interlock with LWP_WEXIT and kill any remaining LWPs
213 lp->lwp_flag |= LWP_WEXIT;
214 if (p->p_nthreads > 1)
218 * If doing this for an exec, clean up the remaining thread
219 * (us) for continuing operation after all the other threads
223 lp->lwp_flag &= ~LWP_WEXIT;
224 p->p_flag &= ~P_WEXIT;
230 * Kill all LWPs except the current one. Do not try to signal
231 * LWPs which have exited on their own or have already been
235 killlwps(struct lwp *lp)
237 struct proc *p = lp->lwp_proc;
241 * Kill the remaining LWPs. We must send the signal before setting
242 * LWP_WEXIT. The setting of WEXIT is optional but helps reduce
243 * races. tlp must be held across the call as it might block and
244 * allow the target lwp to rip itself out from under our loop.
246 FOREACH_LWP_IN_PROC(tlp, p) {
248 if ((tlp->lwp_flag & LWP_WEXIT) == 0) {
249 lwpsignal(p, tlp, SIGKILL);
250 tlp->lwp_flag |= LWP_WEXIT;
256 * Wait for everything to clear out.
258 while (p->p_nthreads > 1) {
259 tsleep(&p->p_nthreads, 0, "killlwps", 0);
264 * Exit: deallocate address space and other resources, change proc state
265 * to zombie, and unlink proc from allproc and parent's lists. Save exit
266 * status and rusage for wait(). Check for child processes and orphan them.
271 struct thread *td = curthread;
272 struct proc *p = td->td_proc;
273 struct lwp *lp = td->td_lwp;
281 kprintf("init died (signal %d, exit %d)\n",
282 WTERMSIG(rv), WEXITSTATUS(rv));
283 panic("Going nowhere without my init!");
286 varsymset_clean(&p->p_varsymset);
287 lockuninit(&p->p_varsymset.vx_lock);
289 * Kill all lwps associated with the current process, return an
290 * error if we race another thread trying to do the same thing
293 error = killalllwps(0);
299 caps_exit(lp->lwp_thread);
302 /* are we a task leader? */
303 if (p == p->p_leader) {
304 struct kill_args killArgs;
305 killArgs.signum = SIGKILL;
308 killArgs.pid = q->p_pid;
310 * The interface for kill is better
311 * than the internal signal
318 tsleep((caddr_t)p, 0, "exit1", 0);
324 STOPEVENT(p, S_EXIT, rv);
325 wakeup(&p->p_stype); /* Wakeup anyone in procfs' PIOCWAIT */
328 * Check if any loadable modules need anything done at process exit.
329 * e.g. SYSV IPC stuff
330 * XXX what if one of these generates an error?
333 EVENTHANDLER_INVOKE(process_exit, p);
336 * XXX: imho, the eventhandler stuff is much cleaner than this.
337 * Maybe we should move everything to use eventhandler.
339 TAILQ_FOREACH(ep, &exit_list, next)
342 if (p->p_flag & P_PROFIL)
345 * If parent is waiting for us to exit or exec,
346 * P_PPWAIT is set; we will wakeup the parent below.
348 p->p_flag &= ~(P_TRACED | P_PPWAIT);
349 SIGEMPTYSET(p->p_siglist);
350 SIGEMPTYSET(lp->lwp_siglist);
351 if (timevalisset(&p->p_realtimer.it_value))
352 callout_stop(&p->p_ithandle);
355 * Reset any sigio structures pointing to us as a result of
356 * F_SETOWN with our pid.
358 funsetownlst(&p->p_sigiolst);
361 * Close open files and release open-file table.
366 if(p->p_leader->p_peers) {
368 while(q->p_peers != p)
370 q->p_peers = p->p_peers;
371 wakeup((caddr_t)p->p_leader);
375 * XXX Shutdown SYSV semaphores
379 KKASSERT(p->p_numposixlocks == 0);
381 /* The next two chunks should probably be moved to vmspace_exit. */
385 * Release upcalls associated with this process
391 * Clean up data related to virtual kernel operation. Clean up
392 * any vkernel context related to the current lwp now so we can
396 vkernel_lwp_exit(lp);
401 * Release user portion of address space.
402 * This releases references to vnodes,
403 * which could cause I/O if the file has been unlinked.
404 * Need to do this early enough that we can still sleep.
405 * Can't free the entire vmspace as the kernel stack
406 * may be mapped within that space also.
408 * Processes sharing the same vmspace may exit in one order, and
409 * get cleaned up by vmspace_exit() in a different order. The
410 * last exiting process to reach this point releases as much of
411 * the environment as it can, and the last process cleaned up
412 * by vmspace_exit() (which decrements exitingcnt) cleans up the
415 vmspace_exitbump(vm);
416 sysref_put(&vm->vm_sysref);
418 if (SESS_LEADER(p)) {
419 struct session *sp = p->p_session;
423 * We are the controlling process. Signal the
424 * foreground process group, drain the controlling
425 * terminal, and revoke access to the controlling
428 * NOTE: while waiting for the process group to exit
429 * it is possible that one of the processes in the
430 * group will revoke the tty, so the ttyclosesession()
431 * function will re-check sp->s_ttyvp.
433 if (sp->s_ttyp && (sp->s_ttyp->t_session == sp)) {
434 if (sp->s_ttyp->t_pgrp)
435 pgsignal(sp->s_ttyp->t_pgrp, SIGHUP, 1);
437 ttyclosesession(sp, 1); /* also revoke */
440 * Release the tty. If someone has it open via
441 * /dev/tty then close it (since they no longer can
442 * once we've NULL'd it out).
444 ttyclosesession(sp, 0);
447 * s_ttyp is not zero'd; we use this to indicate
448 * that the session once had a controlling terminal.
449 * (for logging and informational purposes)
454 fixjobc(p, p->p_pgrp, 0);
455 (void)acct_process(p);
461 ktrdestroy(&p->p_tracenode);
465 * Release reference to text vnode
467 if ((vtmp = p->p_textvp) != NULL) {
472 /* Release namecache handle to text file */
473 if (p->p_textnch.ncp)
474 cache_drop(&p->p_textnch);
477 * Move the process to the zombie list. This will block
478 * until the process p_lock count reaches 0. The process will
479 * not be reaped until TDF_EXITING is set by cpu_thread_exit(),
480 * which is called from cpu_proc_exit().
482 proc_move_allproc_zombie(p);
484 q = LIST_FIRST(&p->p_children);
485 if (q) /* only need this if any child is S_ZOMB */
486 wakeup((caddr_t) initproc);
487 for (; q != 0; q = nq) {
488 nq = LIST_NEXT(q, p_sibling);
489 LIST_REMOVE(q, p_sibling);
490 LIST_INSERT_HEAD(&initproc->p_children, q, p_sibling);
491 q->p_pptr = initproc;
492 q->p_sigparent = SIGCHLD;
494 * Traced processes are killed
495 * since their existence means someone is screwing up.
497 if (q->p_flag & P_TRACED) {
498 q->p_flag &= ~P_TRACED;
504 * Save exit status and final rusage info, adding in child rusage
505 * info and self times.
507 calcru_proc(p, &p->p_ru);
508 ruadd(&p->p_ru, &p->p_cru);
511 * notify interested parties of our demise.
513 KNOTE(&p->p_klist, NOTE_EXIT);
516 * Notify parent that we're gone. If parent has the PS_NOCLDWAIT
517 * flag set, notify process 1 instead (and hope it will handle
520 if (p->p_pptr->p_sigacts->ps_flag & PS_NOCLDWAIT) {
521 struct proc *pp = p->p_pptr;
522 proc_reparent(p, initproc);
524 * If this was the last child of our parent, notify
525 * parent, so in case he was wait(2)ing, he will
528 if (LIST_EMPTY(&pp->p_children))
532 if (p->p_sigparent && p->p_pptr != initproc) {
533 ksignal(p->p_pptr, p->p_sigparent);
535 ksignal(p->p_pptr, SIGCHLD);
538 wakeup((caddr_t)p->p_pptr);
540 * cpu_exit is responsible for clearing curproc, since
541 * it is heavily integrated with the thread/switching sequence.
543 * Other substructures are freed from wait().
548 * Release the current user process designation on the process so
549 * the userland scheduler can work in someone else.
551 p->p_usched->release_curproc(lp);
554 * Finally, call machine-dependent code to release as many of the
555 * lwp's resources as we can and halt execution of this thread.
561 * Eventually called by every exiting LWP
564 lwp_exit(int masterexit)
566 struct thread *td = curthread;
567 struct lwp *lp = td->td_lwp;
568 struct proc *p = lp->lwp_proc;
571 * lwp_exit() may be called without setting LWP_WEXIT, so
572 * make sure it is set here.
574 lp->lwp_flag |= LWP_WEXIT;
577 * Clean up any virtualization
580 vkernel_lwp_exit(lp);
583 * Clean up select/poll support
585 kqueue_terminate(&lp->lwp_kqueue);
588 * Clean up any syscall-cached ucred
591 crfree(td->td_ucred);
596 * Nobody actually wakes us when the lock
597 * count reaches zero, so just wait one tick.
599 while (lp->lwp_lock > 0)
600 tsleep(lp, 0, "lwpexit", 1);
602 /* Hand down resource usage to our proc */
603 ruadd(&p->p_ru, &lp->lwp_ru);
606 * If we don't hold the process until the LWP is reaped wait*()
607 * may try to dispose of its vmspace before all the LWPs have
608 * actually terminated.
613 * We have to use the reaper for all the LWPs except the one doing
614 * the master exit. The LWP doing the master exit can just be
615 * left on p_lwps and the process reaper will deal with it
616 * synchronously, which is much faster.
618 if (masterexit == 0) {
619 lwp_rb_tree_RB_REMOVE(&p->p_lwp_tree, lp);
621 wakeup(&p->p_nthreads);
622 LIST_INSERT_HEAD(&deadlwp_list[mycpuid], lp, u.lwp_reap_entry);
623 taskqueue_enqueue(taskqueue_thread[mycpuid], deadlwp_task[mycpuid]);
627 biosched_done(curthread);
632 * Wait until a lwp is completely dead.
634 * If the thread is still executing, which can't be waited upon,
635 * return failure. The caller is responsible of waiting a little
636 * bit and checking again.
639 * while (!lwp_wait(lp))
640 * tsleep(lp, 0, "lwpwait", 1);
643 lwp_wait(struct lwp *lp)
645 struct thread *td = lp->lwp_thread;;
647 KKASSERT(lwkt_preempted_proc() != lp);
649 while (lp->lwp_lock > 0)
650 tsleep(lp, 0, "lwpwait1", 1);
655 * The lwp's thread may still be in the middle
656 * of switching away, we can't rip its stack out from
657 * under it until TDF_EXITING is set and both
658 * TDF_RUNNING and TDF_PREEMPT_LOCK are clear.
659 * TDF_PREEMPT_LOCK must be checked because TDF_RUNNING
660 * will be cleared temporarily if a thread gets
663 * YYY no wakeup occurs, so we simply return failure
664 * and let the caller deal with sleeping and calling
667 if ((td->td_flags & (TDF_RUNNING|TDF_PREEMPT_LOCK|TDF_EXITING)) !=
675 * Release the resources associated with a lwp.
676 * The lwp must be completely dead.
679 lwp_dispose(struct lwp *lp)
681 struct thread *td = lp->lwp_thread;;
683 KKASSERT(lwkt_preempted_proc() != lp);
684 KKASSERT(td->td_refs == 0);
685 KKASSERT((td->td_flags & (TDF_RUNNING|TDF_PREEMPT_LOCK|TDF_EXITING)) ==
693 lp->lwp_thread = NULL;
694 lwkt_free_thread(td);
703 sys_wait4(struct wait_args *uap)
705 struct rusage rusage;
708 error = kern_wait(uap->pid, (uap->status ? &status : NULL),
709 uap->options, (uap->rusage ? &rusage : NULL),
710 &uap->sysmsg_result);
712 if (error == 0 && uap->status)
713 error = copyout(&status, uap->status, sizeof(*uap->status));
714 if (error == 0 && uap->rusage)
715 error = copyout(&rusage, uap->rusage, sizeof(*uap->rusage));
722 * wait_args(int pid, int *status, int options, struct rusage *rusage)
727 kern_wait(pid_t pid, int *status, int options, struct rusage *rusage, int *res)
729 struct thread *td = curthread;
731 struct proc *q = td->td_proc;
737 if (options &~ (WUNTRACED|WNOHANG|WCONTINUED|WLINUXCLONE))
742 * Hack for backwards compatibility with badly written user code.
743 * Or perhaps we have to do this anyway, it is unclear. XXX
745 * The problem is that if a process group is stopped and the parent
746 * is doing a wait*(..., WUNTRACED, ...), it will see the STOP
747 * of the child and then stop itself when it tries to return from the
748 * system call. When the process group is resumed the parent will
749 * then get the STOP status even though the child has now resumed
750 * (a followup wait*() will get the CONT status).
752 * Previously the CONT would overwrite the STOP because the tstop
753 * was handled within tsleep(), and the parent would only see
754 * the CONT when both are stopped and continued together. This litte
755 * two-line hack restores this effect.
757 while (q->p_stat == SSTOP)
761 LIST_FOREACH(p, &q->p_children, p_sibling) {
762 if (pid != WAIT_ANY &&
763 p->p_pid != pid && p->p_pgid != -pid)
766 /* This special case handles a kthread spawned by linux_clone
767 * (see linux_misc.c). The linux_wait4 and linux_waitpid
768 * functions need to be able to distinguish between waiting
769 * on a process and waiting on a thread. It is a thread if
770 * p_sigparent is not SIGCHLD, and the WLINUXCLONE option
771 * signifies we want to wait for threads and not processes.
773 if ((p->p_sigparent != SIGCHLD) ^
774 ((options & WLINUXCLONE) != 0)) {
779 if (p->p_stat == SZOMB) {
781 * We may go into SZOMB with threads still present.
782 * We must wait for them to exit before we can reap
783 * the master thread, otherwise we may race reaping
784 * non-master threads.
786 while (p->p_nthreads > 0) {
787 tsleep(&p->p_nthreads, 0, "lwpzomb", hz);
791 * Reap any LWPs left in p->p_lwps. This is usually
792 * just the last LWP. This must be done before
793 * we loop on p_lock since the lwps hold a ref on
794 * it as a vmspace interlock.
796 * Once that is accomplished p_nthreads had better
799 while ((lp = RB_ROOT(&p->p_lwp_tree)) != NULL) {
800 lwp_rb_tree_RB_REMOVE(&p->p_lwp_tree, lp);
803 KKASSERT(p->p_nthreads == 0);
806 * Don't do anything really bad until all references
807 * to the process go away. This may include other
808 * LWPs which are still in the process of being
809 * reaped. We can't just pull the rug out from under
810 * them because they may still be using the VM space.
812 * Certain kernel facilities such as /proc will also
813 * put a hold on the process for short periods of
817 tsleep(p, 0, "reap3", hz);
819 /* scheduling hook for heuristic */
820 /* XXX no lwp available, we need a different heuristic */
822 p->p_usched->heuristic_exiting(td->td_lwp, deadlp);
825 /* Take care of our return values. */
828 *status = p->p_xstat;
832 * If we got the child via a ptrace 'attach',
833 * we need to give it back to the old parent.
835 if (p->p_oppid && (t = pfind(p->p_oppid))) {
845 * Unlink the proc from its process group so that
846 * the following operations won't lead to an
847 * inconsistent state for processes running down
850 KKASSERT(p->p_lock == 0);
851 proc_remove_zombie(p);
855 ruadd(&q->p_cru, &p->p_ru);
858 * Decrement the count of procs running with this uid.
860 chgproccnt(p->p_ucred->cr_ruidinfo, -1, 0);
863 * Free up credentials.
869 * Remove unused arguments
871 if (p->p_args && --p->p_args->ar_ref == 0)
872 FREE(p->p_args, M_PARGS);
874 if (--p->p_sigacts->ps_refcnt == 0) {
875 kfree(p->p_sigacts, M_SUBPROC);
885 if (p->p_stat == SSTOP && (p->p_flag & P_WAITED) == 0 &&
886 (p->p_flag & P_TRACED || options & WUNTRACED)) {
887 p->p_flag |= P_WAITED;
891 *status = W_STOPCODE(p->p_xstat);
892 /* Zero rusage so we get something consistent. */
894 bzero(rusage, sizeof(rusage));
898 if (options & WCONTINUED && (p->p_flag & P_CONTINUED)) {
900 p->p_flag &= ~P_CONTINUED;
912 if (options & WNOHANG) {
917 error = tsleep((caddr_t)q, PCATCH, "wait", 0);
927 * make process 'parent' the new parent of process 'child'.
930 proc_reparent(struct proc *child, struct proc *parent)
933 if (child->p_pptr == parent)
936 LIST_REMOVE(child, p_sibling);
937 LIST_INSERT_HEAD(&parent->p_children, child, p_sibling);
938 child->p_pptr = parent;
942 * The next two functions are to handle adding/deleting items on the
946 * Take the arguments given and put them onto the exit callout list,
947 * However first make sure that it's not already there.
948 * returns 0 on success.
952 at_exit(exitlist_fn function)
957 /* Be noisy if the programmer has lost track of things */
958 if (rm_at_exit(function))
959 kprintf("WARNING: exit callout entry (%p) already present\n",
962 ep = kmalloc(sizeof(*ep), M_ATEXIT, M_NOWAIT);
965 ep->function = function;
966 TAILQ_INSERT_TAIL(&exit_list, ep, next);
971 * Scan the exit callout list for the given item and remove it.
972 * Returns the number of items removed (0 or 1)
975 rm_at_exit(exitlist_fn function)
979 TAILQ_FOREACH(ep, &exit_list, next) {
980 if (ep->function == function) {
981 TAILQ_REMOVE(&exit_list, ep, next);
990 * LWP reaper related code.
993 reaplwps(void *context, int dummy)
995 struct lwplist *lwplist = context;
999 while ((lp = LIST_FIRST(lwplist))) {
1000 LIST_REMOVE(lp, u.lwp_reap_entry);
1007 reaplwp(struct lwp *lp)
1009 while (lwp_wait(lp) == 0)
1010 tsleep(lp, 0, "lwpreap", 1);
1019 for (cpu = 0; cpu < ncpus; cpu++) {
1020 LIST_INIT(&deadlwp_list[cpu]);
1021 deadlwp_task[cpu] = kmalloc(sizeof(*deadlwp_task[cpu]), M_DEVBUF, M_WAITOK);
1022 TASK_INIT(deadlwp_task[cpu], 0, reaplwps, &deadlwp_list[cpu]);
1026 SYSINIT(deadlwpinit, SI_SUB_CONFIGURE, SI_ORDER_ANY, deadlwp_init, NULL);