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 1.79 2007/03/12 21:05:48 corecode 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>
83 static MALLOC_DEFINE(M_ATEXIT
, "atexit", "atexit callback");
84 static MALLOC_DEFINE(M_ZOMBIE
, "zombie", "zombie proc status");
87 * callout list for things to do at exit time
91 TAILQ_ENTRY(exitlist
) next
;
94 TAILQ_HEAD(exit_list_head
, exitlist
);
95 static struct exit_list_head exit_list
= TAILQ_HEAD_INITIALIZER(exit_list
);
100 struct task
*deadlwp_task
[MAXCPU
];
101 struct lwplist deadlwp_list
[MAXCPU
];
107 * SYS_EXIT_ARGS(int rval)
110 sys_exit(struct exit_args
*uap
)
112 exit1(W_EXITCODE(uap
->rval
, 0));
118 * Death of a lwp or process with optional bells and whistles.
121 sys_extexit(struct extexit_args
*uap
)
126 action
= EXTEXIT_ACTION(uap
->how
);
127 who
= EXTEXIT_WHO(uap
->how
);
129 /* Check parameters before we might perform some action */
144 error
= copyout(&uap
->status
, uap
->addr
, sizeof(uap
->status
));
156 * Be sure only to perform a simple lwp exit if there is at
157 * least one more lwp in the proc, which will call exit1()
158 * later, otherwise the proc will be an UNDEAD and not even a
161 if (curproc
->p_nthreads
> 1)
163 /* else last lwp in proc: do the real thing */
166 default: /* to help gcc */
168 exit1(W_EXITCODE(uap
->status
, 0));
176 killlwps(struct lwp
*lp
)
178 struct proc
*p
= lp
->lwp_proc
;
181 KKASSERT((lp
->lwp_flag
& LWP_WEXIT
) == 0);
183 FOREACH_LWP_IN_PROC(tlp
, p
) {
185 continue; /* don't kill the current lwp */
186 tlp
->lwp_flag
|= LWP_WEXIT
;
190 while (p
->p_nthreads
> 1) {
192 kprintf("killlwps: waiting for %d lwps of pid %d to die\n",
193 p
->p_nthreads
- 1, p
->p_pid
);
194 tsleep(&p
->p_nthreads
, 0, "killlwps", hz
);
199 * Exit: deallocate address space and other resources, change proc state
200 * to zombie, and unlink proc from allproc and parent's lists. Save exit
201 * status and rusage for wait(). Check for child processes and orphan them.
206 struct thread
*td
= curthread
;
207 struct proc
*p
= td
->td_proc
;
208 struct lwp
*lp
= td
->td_lwp
;
215 kprintf("init died (signal %d, exit %d)\n",
216 WTERMSIG(rv
), WEXITSTATUS(rv
));
217 panic("Going nowhere without my init!");
221 * Kill all other threads if there are any.
223 * If some other thread initiated our exit, do so.
225 if (lp
->lwp_flag
& LWP_WEXIT
) {
226 KKASSERT(p
->p_nthreads
> 1);
229 if (p
->p_nthreads
> 1)
232 caps_exit(lp
->lwp_thread
);
235 /* are we a task leader? */
236 if(p
== p
->p_leader
) {
237 struct kill_args killArgs
;
238 killArgs
.signum
= SIGKILL
;
241 killArgs
.pid
= q
->p_pid
;
243 * The interface for kill is better
244 * than the internal signal
251 tsleep((caddr_t
)p
, 0, "exit1", 0);
257 STOPEVENT(p
, S_EXIT
, rv
);
258 wakeup(&p
->p_stype
); /* Wakeup anyone in procfs' PIOCWAIT */
261 * Check if any loadable modules need anything done at process exit.
262 * e.g. SYSV IPC stuff
263 * XXX what if one of these generates an error?
265 TAILQ_FOREACH(ep
, &exit_list
, next
)
268 if (p
->p_flag
& P_PROFIL
)
271 * If parent is waiting for us to exit or exec,
272 * P_PPWAIT is set; we will wakeup the parent below.
274 p
->p_flag
&= ~(P_TRACED
| P_PPWAIT
);
275 p
->p_flag
|= P_WEXIT
;
276 SIGEMPTYSET(p
->p_siglist
);
277 SIGEMPTYSET(lp
->lwp_siglist
);
278 if (timevalisset(&p
->p_realtimer
.it_value
))
279 callout_stop(&p
->p_ithandle
);
282 * Reset any sigio structures pointing to us as a result of
283 * F_SETOWN with our pid.
285 funsetownlst(&p
->p_sigiolst
);
288 * Close open files and release open-file table.
294 if(p
->p_leader
->p_peers
) {
296 while(q
->p_peers
!= p
)
298 q
->p_peers
= p
->p_peers
;
299 wakeup((caddr_t
)p
->p_leader
);
303 * XXX Shutdown SYSV semaphores
307 KKASSERT(p
->p_numposixlocks
== 0);
309 /* The next two chunks should probably be moved to vmspace_exit. */
313 * Release upcalls associated with this process
318 /* clean up data related to virtual kernel operation */
323 * Release user portion of address space.
324 * This releases references to vnodes,
325 * which could cause I/O if the file has been unlinked.
326 * Need to do this early enough that we can still sleep.
327 * Can't free the entire vmspace as the kernel stack
328 * may be mapped within that space also.
330 * Processes sharing the same vmspace may exit in one order, and
331 * get cleaned up by vmspace_exit() in a different order. The
332 * last exiting process to reach this point releases as much of
333 * the environment as it can, and the last process cleaned up
334 * by vmspace_exit() (which decrements exitingcnt) cleans up the
338 if (--vm
->vm_refcnt
== 0) {
340 pmap_remove_pages(vmspace_pmap(vm
), VM_MIN_USER_ADDRESS
,
341 VM_MAX_USER_ADDRESS
);
342 vm_map_remove(&vm
->vm_map
, VM_MIN_USER_ADDRESS
,
343 VM_MAX_USER_ADDRESS
);
346 if (SESS_LEADER(p
)) {
347 struct session
*sp
= p
->p_session
;
352 * We are the controlling process. Signal the
353 * foreground process group, drain the controlling
354 * terminal, and revoke access to the controlling
357 * NOTE: while waiting for the process group to exit
358 * it is possible that one of the processes in the
359 * group will revoke the tty, so we have to recheck.
361 if (sp
->s_ttyp
&& (sp
->s_ttyp
->t_session
== sp
)) {
362 if (sp
->s_ttyp
->t_pgrp
)
363 pgsignal(sp
->s_ttyp
->t_pgrp
, SIGHUP
, 1);
364 (void) ttywait(sp
->s_ttyp
);
366 * The tty could have been revoked
369 if ((vp
= sp
->s_ttyvp
) != NULL
) {
370 ttyclosesession(sp
, 0);
372 VOP_REVOKE(vp
, REVOKEALL
);
374 vrele(vp
); /* s_ttyvp ref */
378 * Release the tty. If someone has it open via
379 * /dev/tty then close it (since they no longer can
380 * once we've NULL'd it out).
383 ttyclosesession(sp
, 1);
385 * s_ttyp is not zero'd; we use this to indicate
386 * that the session once had a controlling terminal.
387 * (for logging and informational purposes)
392 fixjobc(p
, p
->p_pgrp
, 0);
393 (void)acct_process(p
);
399 ktrdestroy(&p
->p_tracenode
);
403 * Release reference to text vnode
405 if ((vtmp
= p
->p_textvp
) != NULL
) {
411 * Move the process to the zombie list. This will block
412 * until the process p_lock count reaches 0. The process will
413 * not be reaped until TDF_EXITING is set by cpu_thread_exit(),
414 * which is called from cpu_proc_exit().
416 proc_move_allproc_zombie(p
);
418 q
= LIST_FIRST(&p
->p_children
);
419 if (q
) /* only need this if any child is S_ZOMB */
420 wakeup((caddr_t
) initproc
);
421 for (; q
!= 0; q
= nq
) {
422 nq
= LIST_NEXT(q
, p_sibling
);
423 LIST_REMOVE(q
, p_sibling
);
424 LIST_INSERT_HEAD(&initproc
->p_children
, q
, p_sibling
);
425 q
->p_pptr
= initproc
;
426 q
->p_sigparent
= SIGCHLD
;
428 * Traced processes are killed
429 * since their existence means someone is screwing up.
431 if (q
->p_flag
& P_TRACED
) {
432 q
->p_flag
&= ~P_TRACED
;
438 * Save exit status and final rusage info, adding in child rusage
439 * info and self times.
442 calcru_proc(p
, &p
->p_ru
);
443 ruadd(&p
->p_ru
, &p
->p_cru
);
446 * notify interested parties of our demise.
448 KNOTE(&p
->p_klist
, NOTE_EXIT
);
451 * Notify parent that we're gone. If parent has the PS_NOCLDWAIT
452 * flag set, notify process 1 instead (and hope it will handle
455 if (p
->p_pptr
->p_sigacts
->ps_flag
& PS_NOCLDWAIT
) {
456 struct proc
*pp
= p
->p_pptr
;
457 proc_reparent(p
, initproc
);
459 * If this was the last child of our parent, notify
460 * parent, so in case he was wait(2)ing, he will
463 if (LIST_EMPTY(&pp
->p_children
))
467 if (p
->p_sigparent
&& p
->p_pptr
!= initproc
) {
468 ksignal(p
->p_pptr
, p
->p_sigparent
);
470 ksignal(p
->p_pptr
, SIGCHLD
);
473 wakeup((caddr_t
)p
->p_pptr
);
475 * cpu_exit is responsible for clearing curproc, since
476 * it is heavily integrated with the thread/switching sequence.
478 * Other substructures are freed from wait().
480 plimit_free(&p
->p_limit
);
483 * Release the current user process designation on the process so
484 * the userland scheduler can work in someone else.
486 p
->p_usched
->release_curproc(lp
);
489 * Finally, call machine-dependent code to release the remaining
490 * resources including address space, the kernel stack and pcb.
491 * The address space is released by "vmspace_free(p->p_vmspace)";
492 * This is machine-dependent, as we may have to change stacks
493 * or ensure that the current one isn't reallocated before we
494 * finish. cpu_exit will end with a call to cpu_switch(), finishing
495 * our execution (pun intended).
503 struct lwp
*lp
= curthread
->td_lwp
;
504 struct proc
*p
= lp
->lwp_proc
;
507 * Nobody actually wakes us when the lock
508 * count reaches zero, so just wait one tick.
510 while (lp
->lwp_lock
> 0)
511 tsleep(lp
, 0, "lwpexit", 1);
513 /* Hand down resource usage to our proc */
514 ruadd(&p
->p_ru
, &lp
->lwp_ru
);
517 LIST_REMOVE(lp
, lwp_list
);
518 wakeup(&p
->p_nthreads
);
519 LIST_INSERT_HEAD(&deadlwp_list
[mycpuid
], lp
, lwp_list
);
520 taskqueue_enqueue(taskqueue_thread
[mycpuid
], deadlwp_task
[mycpuid
]);
525 * Wait until a lwp is completely dead.
527 * If the thread is still executing, which can't be waited upon,
528 * return failure. The caller is responsible of waiting a little
529 * bit and checking again.
532 * while (!lwp_wait(lp))
533 * tsleep(lp, 0, "lwpwait", 1);
536 lwp_wait(struct lwp
*lp
)
538 struct thread
*td
= lp
->lwp_thread
;;
540 KKASSERT(lwkt_preempted_proc() != lp
);
542 while (lp
->lwp_lock
> 0)
543 tsleep(lp
, 0, "lwpwait1", 1);
548 * The lwp's thread may still be in the middle
549 * of switching away, we can't rip its stack out from
550 * under it until TDF_EXITING is set and both
551 * TDF_RUNNING and TDF_PREEMPT_LOCK are clear.
552 * TDF_PREEMPT_LOCK must be checked because TDF_RUNNING
553 * will be cleared temporarily if a thread gets
556 * YYY no wakeup occurs, so we simply return failure
557 * and let the caller deal with sleeping and calling
560 if ((td
->td_flags
& (TDF_RUNNING
|TDF_PREEMPT_LOCK
|TDF_EXITING
)) !=
568 * Release the resources associated with a lwp.
569 * The lwp must be completely dead.
572 lwp_dispose(struct lwp
*lp
)
574 struct thread
*td
= lp
->lwp_thread
;;
576 KKASSERT(lwkt_preempted_proc() != lp
);
577 KKASSERT(td
->td_refs
== 0);
578 KKASSERT((td
->td_flags
& (TDF_RUNNING
|TDF_PREEMPT_LOCK
|TDF_EXITING
)) ==
584 lp
->lwp_thread
= NULL
;
585 lwkt_free_thread(td
);
591 sys_wait4(struct wait_args
*uap
)
593 struct rusage rusage
;
596 error
= kern_wait(uap
->pid
, uap
->status
? &status
: NULL
,
597 uap
->options
, uap
->rusage
? &rusage
: NULL
, &uap
->sysmsg_fds
[0]);
599 if (error
== 0 && uap
->status
)
600 error
= copyout(&status
, uap
->status
, sizeof(*uap
->status
));
601 if (error
== 0 && uap
->rusage
)
602 error
= copyout(&rusage
, uap
->rusage
, sizeof(*uap
->rusage
));
609 * wait_args(int pid, int *status, int options, struct rusage *rusage)
612 kern_wait(pid_t pid
, int *status
, int options
, struct rusage
*rusage
, int *res
)
614 struct thread
*td
= curthread
;
615 struct proc
*q
= td
->td_proc
;
621 if (options
&~ (WUNTRACED
|WNOHANG
|WLINUXCLONE
))
625 * Hack for backwards compatibility with badly written user code.
626 * Or perhaps we have to do this anyway, it is unclear. XXX
628 * The problem is that if a process group is stopped and the parent
629 * is doing a wait*(..., WUNTRACED, ...), it will see the STOP
630 * of the child and then stop itself when it tries to return from the
631 * system call. When the process group is resumed the parent will
632 * then get the STOP status even though the child has now resumed
633 * (a followup wait*() will get the CONT status).
635 * Previously the CONT would overwrite the STOP because the tstop
636 * was handled within tsleep(), and the parent would only see
637 * the CONT when both are stopped and continued together. This litte
638 * two-line hack restores this effect.
640 while (q
->p_stat
== SSTOP
)
644 LIST_FOREACH(p
, &q
->p_children
, p_sibling
) {
645 if (pid
!= WAIT_ANY
&&
646 p
->p_pid
!= pid
&& p
->p_pgid
!= -pid
)
649 /* This special case handles a kthread spawned by linux_clone
650 * (see linux_misc.c). The linux_wait4 and linux_waitpid
651 * functions need to be able to distinguish between waiting
652 * on a process and waiting on a thread. It is a thread if
653 * p_sigparent is not SIGCHLD, and the WLINUXCLONE option
654 * signifies we want to wait for threads and not processes.
656 if ((p
->p_sigparent
!= SIGCHLD
) ^
657 ((options
& WLINUXCLONE
) != 0)) {
662 if (p
->p_stat
== SZOMB
) {
664 * Other kernel threads may be in the middle of
665 * accessing the proc. For example, kern/kern_proc.c
666 * could be blocked writing proc data to a sysctl.
667 * At the moment, if this occurs, we are not woken
668 * up and rely on a one-second retry.
671 tsleep(p
, 0, "reap3", hz
);
673 /* scheduling hook for heuristic */
674 /* XXX no lwp available, we need a different heuristic */
676 p->p_usched->heuristic_exiting(td->td_lwp, deadlp);
679 /* Take care of our return values. */
682 *status
= p
->p_xstat
;
686 * If we got the child via a ptrace 'attach',
687 * we need to give it back to the old parent.
689 if (p
->p_oppid
&& (t
= pfind(p
->p_oppid
))) {
697 ruadd(&q
->p_cru
, &p
->p_ru
);
700 * Decrement the count of procs running with this uid.
702 chgproccnt(p
->p_ucred
->cr_ruidinfo
, -1, 0);
705 * Free up credentials.
711 * Remove unused arguments
713 if (p
->p_args
&& --p
->p_args
->ar_ref
== 0)
714 FREE(p
->p_args
, M_PARGS
);
717 * Finally finished with old proc entry.
718 * Unlink it from its process group and free it.
720 proc_remove_zombie(p
);
723 if (--p
->p_sigacts
->ps_refcnt
== 0) {
724 kfree(p
->p_sigacts
, M_SUBPROC
);
733 if (p
->p_stat
== SSTOP
&& (p
->p_flag
& P_WAITED
) == 0 &&
734 (p
->p_flag
& P_TRACED
|| options
& WUNTRACED
)) {
735 p
->p_flag
|= P_WAITED
;
739 *status
= W_STOPCODE(p
->p_xstat
);
740 /* Zero rusage so we get something consistent. */
742 bzero(rusage
, sizeof(rusage
));
748 if (options
& WNOHANG
) {
752 error
= tsleep((caddr_t
)q
, PCATCH
, "wait", 0);
759 * make process 'parent' the new parent of process 'child'.
762 proc_reparent(struct proc
*child
, struct proc
*parent
)
765 if (child
->p_pptr
== parent
)
768 LIST_REMOVE(child
, p_sibling
);
769 LIST_INSERT_HEAD(&parent
->p_children
, child
, p_sibling
);
770 child
->p_pptr
= parent
;
774 * The next two functions are to handle adding/deleting items on the
778 * Take the arguments given and put them onto the exit callout list,
779 * However first make sure that it's not already there.
780 * returns 0 on success.
784 at_exit(exitlist_fn function
)
789 /* Be noisy if the programmer has lost track of things */
790 if (rm_at_exit(function
))
791 kprintf("WARNING: exit callout entry (%p) already present\n",
794 ep
= kmalloc(sizeof(*ep
), M_ATEXIT
, M_NOWAIT
);
797 ep
->function
= function
;
798 TAILQ_INSERT_TAIL(&exit_list
, ep
, next
);
803 * Scan the exit callout list for the given item and remove it.
804 * Returns the number of items removed (0 or 1)
807 rm_at_exit(exitlist_fn function
)
811 TAILQ_FOREACH(ep
, &exit_list
, next
) {
812 if (ep
->function
== function
) {
813 TAILQ_REMOVE(&exit_list
, ep
, next
);
823 * LWP reaper related code.
827 reaplwps(void *context
, int dummy
)
829 struct lwplist
*lwplist
= context
;
832 while ((lp
= LIST_FIRST(lwplist
))) {
834 tsleep(lp
, 0, "lwpreap", 1);
835 LIST_REMOVE(lp
, lwp_list
);
845 for (cpu
= 0; cpu
< ncpus
; cpu
++) {
846 LIST_INIT(&deadlwp_list
[cpu
]);
847 deadlwp_task
[cpu
] = kmalloc(sizeof(*deadlwp_task
[cpu
]), M_DEVBUF
, M_WAITOK
);
848 TASK_INIT(deadlwp_task
[cpu
], 0, reaplwps
, &deadlwp_list
[cpu
]);
852 SYSINIT(deadlwpinit
, SI_SUB_CONFIGURE
, SI_ORDER_ANY
, deadlwp_init
, NULL
);