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 1.80 2007/04/29 18:25:34 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 * Signal remaining LWPs, interlock with LWP_WEXIT.
194 FOREACH_LWP_IN_PROC(tlp
, p
) {
195 if ((tlp
->lwp_flag
& LWP_WEXIT
) == 0) {
196 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
337 /* clean up data related to virtual kernel operation */
342 * Release user portion of address space.
343 * This releases references to vnodes,
344 * which could cause I/O if the file has been unlinked.
345 * Need to do this early enough that we can still sleep.
346 * Can't free the entire vmspace as the kernel stack
347 * may be mapped within that space also.
349 * Processes sharing the same vmspace may exit in one order, and
350 * get cleaned up by vmspace_exit() in a different order. The
351 * last exiting process to reach this point releases as much of
352 * the environment as it can, and the last process cleaned up
353 * by vmspace_exit() (which decrements exitingcnt) cleans up the
357 sysref_put(&vm
->vm_sysref
);
359 if (SESS_LEADER(p
)) {
360 struct session
*sp
= p
->p_session
;
365 * We are the controlling process. Signal the
366 * foreground process group, drain the controlling
367 * terminal, and revoke access to the controlling
370 * NOTE: while waiting for the process group to exit
371 * it is possible that one of the processes in the
372 * group will revoke the tty, so we have to recheck.
374 if (sp
->s_ttyp
&& (sp
->s_ttyp
->t_session
== sp
)) {
375 if (sp
->s_ttyp
->t_pgrp
)
376 pgsignal(sp
->s_ttyp
->t_pgrp
, SIGHUP
, 1);
377 (void) ttywait(sp
->s_ttyp
);
379 * The tty could have been revoked
382 if ((vp
= sp
->s_ttyvp
) != NULL
) {
383 ttyclosesession(sp
, 0);
385 VOP_REVOKE(vp
, REVOKEALL
);
387 vrele(vp
); /* s_ttyvp ref */
391 * Release the tty. If someone has it open via
392 * /dev/tty then close it (since they no longer can
393 * once we've NULL'd it out).
396 ttyclosesession(sp
, 1);
398 * s_ttyp is not zero'd; we use this to indicate
399 * that the session once had a controlling terminal.
400 * (for logging and informational purposes)
405 fixjobc(p
, p
->p_pgrp
, 0);
406 (void)acct_process(p
);
412 ktrdestroy(&p
->p_tracenode
);
416 * Release reference to text vnode
418 if ((vtmp
= p
->p_textvp
) != NULL
) {
424 * Move the process to the zombie list. This will block
425 * until the process p_lock count reaches 0. The process will
426 * not be reaped until TDF_EXITING is set by cpu_thread_exit(),
427 * which is called from cpu_proc_exit().
429 proc_move_allproc_zombie(p
);
431 q
= LIST_FIRST(&p
->p_children
);
432 if (q
) /* only need this if any child is S_ZOMB */
433 wakeup((caddr_t
) initproc
);
434 for (; q
!= 0; q
= nq
) {
435 nq
= LIST_NEXT(q
, p_sibling
);
436 LIST_REMOVE(q
, p_sibling
);
437 LIST_INSERT_HEAD(&initproc
->p_children
, q
, p_sibling
);
438 q
->p_pptr
= initproc
;
439 q
->p_sigparent
= SIGCHLD
;
441 * Traced processes are killed
442 * since their existence means someone is screwing up.
444 if (q
->p_flag
& P_TRACED
) {
445 q
->p_flag
&= ~P_TRACED
;
451 * Save exit status and final rusage info, adding in child rusage
452 * info and self times.
455 calcru_proc(p
, &p
->p_ru
);
456 ruadd(&p
->p_ru
, &p
->p_cru
);
459 * notify interested parties of our demise.
461 KNOTE(&p
->p_klist
, NOTE_EXIT
);
464 * Notify parent that we're gone. If parent has the PS_NOCLDWAIT
465 * flag set, notify process 1 instead (and hope it will handle
468 if (p
->p_pptr
->p_sigacts
->ps_flag
& PS_NOCLDWAIT
) {
469 struct proc
*pp
= p
->p_pptr
;
470 proc_reparent(p
, initproc
);
472 * If this was the last child of our parent, notify
473 * parent, so in case he was wait(2)ing, he will
476 if (LIST_EMPTY(&pp
->p_children
))
480 if (p
->p_sigparent
&& p
->p_pptr
!= initproc
) {
481 ksignal(p
->p_pptr
, p
->p_sigparent
);
483 ksignal(p
->p_pptr
, SIGCHLD
);
486 wakeup((caddr_t
)p
->p_pptr
);
488 * cpu_exit is responsible for clearing curproc, since
489 * it is heavily integrated with the thread/switching sequence.
491 * Other substructures are freed from wait().
493 plimit_free(&p
->p_limit
);
496 * Release the current user process designation on the process so
497 * the userland scheduler can work in someone else.
499 p
->p_usched
->release_curproc(lp
);
502 * Finally, call machine-dependent code to release as many of the
503 * lwp's resources as we can and halt execution of this thread.
509 lwp_exit(int masterexit
)
511 struct lwp
*lp
= curthread
->td_lwp
;
512 struct proc
*p
= lp
->lwp_proc
;
515 * lwp_exit() may be called without setting LWP_WEXIT, so
516 * make sure it is set here.
518 lp
->lwp_flag
|= LWP_WEXIT
;
521 * Nobody actually wakes us when the lock
522 * count reaches zero, so just wait one tick.
524 while (lp
->lwp_lock
> 0)
525 tsleep(lp
, 0, "lwpexit", 1);
527 /* Hand down resource usage to our proc */
528 ruadd(&p
->p_ru
, &lp
->lwp_ru
);
531 * If we don't hold the process until the LWP is reaped wait*()
532 * may try to dispose of its vmspace before all the LWPs have
533 * actually terminated.
538 * We have to use the reaper for all the LWPs except the one doing
539 * the master exit. The LWP doing the master exit can just be
540 * left on p_lwps and the process reaper will deal with it
541 * synchronously, which is much faster.
543 if (masterexit
== 0) {
544 LIST_REMOVE(lp
, lwp_list
);
546 wakeup(&p
->p_nthreads
);
547 LIST_INSERT_HEAD(&deadlwp_list
[mycpuid
], lp
, lwp_list
);
548 taskqueue_enqueue(taskqueue_thread
[mycpuid
], deadlwp_task
[mycpuid
]);
556 * Wait until a lwp is completely dead.
558 * If the thread is still executing, which can't be waited upon,
559 * return failure. The caller is responsible of waiting a little
560 * bit and checking again.
563 * while (!lwp_wait(lp))
564 * tsleep(lp, 0, "lwpwait", 1);
567 lwp_wait(struct lwp
*lp
)
569 struct thread
*td
= lp
->lwp_thread
;;
571 KKASSERT(lwkt_preempted_proc() != lp
);
573 while (lp
->lwp_lock
> 0)
574 tsleep(lp
, 0, "lwpwait1", 1);
579 * The lwp's thread may still be in the middle
580 * of switching away, we can't rip its stack out from
581 * under it until TDF_EXITING is set and both
582 * TDF_RUNNING and TDF_PREEMPT_LOCK are clear.
583 * TDF_PREEMPT_LOCK must be checked because TDF_RUNNING
584 * will be cleared temporarily if a thread gets
587 * YYY no wakeup occurs, so we simply return failure
588 * and let the caller deal with sleeping and calling
591 if ((td
->td_flags
& (TDF_RUNNING
|TDF_PREEMPT_LOCK
|TDF_EXITING
)) !=
599 * Release the resources associated with a lwp.
600 * The lwp must be completely dead.
603 lwp_dispose(struct lwp
*lp
)
605 struct thread
*td
= lp
->lwp_thread
;;
607 KKASSERT(lwkt_preempted_proc() != lp
);
608 KKASSERT(td
->td_refs
== 0);
609 KKASSERT((td
->td_flags
& (TDF_RUNNING
|TDF_PREEMPT_LOCK
|TDF_EXITING
)) ==
617 lp
->lwp_thread
= NULL
;
618 lwkt_free_thread(td
);
624 sys_wait4(struct wait_args
*uap
)
626 struct rusage rusage
;
629 error
= kern_wait(uap
->pid
, uap
->status
? &status
: NULL
,
630 uap
->options
, uap
->rusage
? &rusage
: NULL
, &uap
->sysmsg_fds
[0]);
632 if (error
== 0 && uap
->status
)
633 error
= copyout(&status
, uap
->status
, sizeof(*uap
->status
));
634 if (error
== 0 && uap
->rusage
)
635 error
= copyout(&rusage
, uap
->rusage
, sizeof(*uap
->rusage
));
642 * wait_args(int pid, int *status, int options, struct rusage *rusage)
645 kern_wait(pid_t pid
, int *status
, int options
, struct rusage
*rusage
, int *res
)
647 struct thread
*td
= curthread
;
648 struct proc
*q
= td
->td_proc
;
654 if (options
&~ (WUNTRACED
|WNOHANG
|WLINUXCLONE
))
658 * Hack for backwards compatibility with badly written user code.
659 * Or perhaps we have to do this anyway, it is unclear. XXX
661 * The problem is that if a process group is stopped and the parent
662 * is doing a wait*(..., WUNTRACED, ...), it will see the STOP
663 * of the child and then stop itself when it tries to return from the
664 * system call. When the process group is resumed the parent will
665 * then get the STOP status even though the child has now resumed
666 * (a followup wait*() will get the CONT status).
668 * Previously the CONT would overwrite the STOP because the tstop
669 * was handled within tsleep(), and the parent would only see
670 * the CONT when both are stopped and continued together. This litte
671 * two-line hack restores this effect.
673 while (q
->p_stat
== SSTOP
)
677 LIST_FOREACH(p
, &q
->p_children
, p_sibling
) {
678 if (pid
!= WAIT_ANY
&&
679 p
->p_pid
!= pid
&& p
->p_pgid
!= -pid
)
682 /* This special case handles a kthread spawned by linux_clone
683 * (see linux_misc.c). The linux_wait4 and linux_waitpid
684 * functions need to be able to distinguish between waiting
685 * on a process and waiting on a thread. It is a thread if
686 * p_sigparent is not SIGCHLD, and the WLINUXCLONE option
687 * signifies we want to wait for threads and not processes.
689 if ((p
->p_sigparent
!= SIGCHLD
) ^
690 ((options
& WLINUXCLONE
) != 0)) {
695 if (p
->p_stat
== SZOMB
) {
697 * Reap any LWPs left in p->p_lwps. This is usually
698 * just the last LWP. This must be done before
699 * we loop on p_lock since the lwps hold a ref on
700 * it as a vmspace interlock.
702 * Once that is accomplished p_nthreads had better
705 reaplwps(&p
->p_lwps
, 0);
706 KKASSERT(p
->p_nthreads
== 0);
709 * Don't do anything really bad until all references
710 * to the process go away. This may include other
711 * LWPs which are still in the process of being
712 * reaped. We can't just pull the rug out from under
713 * them because they may still be using the VM space.
715 * Certain kernel facilities such as /proc will also
716 * put a hold on the process for short periods of
720 tsleep(p
, 0, "reap3", hz
);
722 /* scheduling hook for heuristic */
723 /* XXX no lwp available, we need a different heuristic */
725 p->p_usched->heuristic_exiting(td->td_lwp, deadlp);
728 /* Take care of our return values. */
731 *status
= p
->p_xstat
;
735 * If we got the child via a ptrace 'attach',
736 * we need to give it back to the old parent.
738 if (p
->p_oppid
&& (t
= pfind(p
->p_oppid
))) {
746 ruadd(&q
->p_cru
, &p
->p_ru
);
749 * Decrement the count of procs running with this uid.
751 chgproccnt(p
->p_ucred
->cr_ruidinfo
, -1, 0);
754 * Free up credentials.
760 * Remove unused arguments
762 if (p
->p_args
&& --p
->p_args
->ar_ref
== 0)
763 FREE(p
->p_args
, M_PARGS
);
766 * Finally finished with old proc entry.
767 * Unlink it from its process group and free it.
769 proc_remove_zombie(p
);
772 if (--p
->p_sigacts
->ps_refcnt
== 0) {
773 kfree(p
->p_sigacts
, M_SUBPROC
);
782 if (p
->p_stat
== SSTOP
&& (p
->p_flag
& P_WAITED
) == 0 &&
783 (p
->p_flag
& P_TRACED
|| options
& WUNTRACED
)) {
784 p
->p_flag
|= P_WAITED
;
788 *status
= W_STOPCODE(p
->p_xstat
);
789 /* Zero rusage so we get something consistent. */
791 bzero(rusage
, sizeof(rusage
));
797 if (options
& WNOHANG
) {
801 error
= tsleep((caddr_t
)q
, PCATCH
, "wait", 0);
808 * make process 'parent' the new parent of process 'child'.
811 proc_reparent(struct proc
*child
, struct proc
*parent
)
814 if (child
->p_pptr
== parent
)
817 LIST_REMOVE(child
, p_sibling
);
818 LIST_INSERT_HEAD(&parent
->p_children
, child
, p_sibling
);
819 child
->p_pptr
= parent
;
823 * The next two functions are to handle adding/deleting items on the
827 * Take the arguments given and put them onto the exit callout list,
828 * However first make sure that it's not already there.
829 * returns 0 on success.
833 at_exit(exitlist_fn function
)
838 /* Be noisy if the programmer has lost track of things */
839 if (rm_at_exit(function
))
840 kprintf("WARNING: exit callout entry (%p) already present\n",
843 ep
= kmalloc(sizeof(*ep
), M_ATEXIT
, M_NOWAIT
);
846 ep
->function
= function
;
847 TAILQ_INSERT_TAIL(&exit_list
, ep
, next
);
852 * Scan the exit callout list for the given item and remove it.
853 * Returns the number of items removed (0 or 1)
856 rm_at_exit(exitlist_fn function
)
860 TAILQ_FOREACH(ep
, &exit_list
, next
) {
861 if (ep
->function
== function
) {
862 TAILQ_REMOVE(&exit_list
, ep
, next
);
871 * LWP reaper related code.
874 reaplwps(void *context
, int dummy
)
876 struct lwplist
*lwplist
= context
;
879 while ((lp
= LIST_FIRST(lwplist
))) {
880 LIST_REMOVE(lp
, lwp_list
);
881 while (lwp_wait(lp
) == 0)
882 tsleep(lp
, 0, "lwpreap", 1);
892 for (cpu
= 0; cpu
< ncpus
; cpu
++) {
893 LIST_INIT(&deadlwp_list
[cpu
]);
894 deadlwp_task
[cpu
] = kmalloc(sizeof(*deadlwp_task
[cpu
]), M_DEVBUF
, M_WAITOK
);
895 TASK_INIT(deadlwp_task
[cpu
], 0, reaplwps
, &deadlwp_list
[cpu
]);
899 SYSINIT(deadlwpinit
, SI_SUB_CONFIGURE
, SI_ORDER_ANY
, deadlwp_init
, NULL
);