FreeBSD/Linux Kernel Cross Reference
sys/kern/kern_exit.c
1 /*-
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.
9 *
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
12 * are met:
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 * 4. Neither the name of the University nor the names of its contributors
19 * may be used to endorse or promote products derived from this software
20 * without specific prior written permission.
21 *
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 * SUCH DAMAGE.
33 *
34 * @(#)kern_exit.c 8.7 (Berkeley) 2/12/94
35 */
36
37 #include <sys/cdefs.h>
38 __FBSDID("$FreeBSD: releng/9.2/sys/kern/kern_exit.c 247349 2013-02-26 21:04:58Z jhb $");
39
40 #include "opt_compat.h"
41 #include "opt_kdtrace.h"
42 #include "opt_ktrace.h"
43 #include "opt_procdesc.h"
44
45 #include <sys/param.h>
46 #include <sys/systm.h>
47 #include <sys/sysproto.h>
48 #include <sys/capability.h>
49 #include <sys/eventhandler.h>
50 #include <sys/kernel.h>
51 #include <sys/malloc.h>
52 #include <sys/lock.h>
53 #include <sys/mutex.h>
54 #include <sys/proc.h>
55 #include <sys/procdesc.h>
56 #include <sys/pioctl.h>
57 #include <sys/jail.h>
58 #include <sys/tty.h>
59 #include <sys/wait.h>
60 #include <sys/vmmeter.h>
61 #include <sys/vnode.h>
62 #include <sys/racct.h>
63 #include <sys/resourcevar.h>
64 #include <sys/sbuf.h>
65 #include <sys/signalvar.h>
66 #include <sys/sched.h>
67 #include <sys/sx.h>
68 #include <sys/syscallsubr.h>
69 #include <sys/syslog.h>
70 #include <sys/ptrace.h>
71 #include <sys/acct.h> /* for acct_process() function prototype */
72 #include <sys/filedesc.h>
73 #include <sys/sdt.h>
74 #include <sys/shm.h>
75 #include <sys/sem.h>
76 #ifdef KTRACE
77 #include <sys/ktrace.h>
78 #endif
79
80 #include <security/audit/audit.h>
81 #include <security/mac/mac_framework.h>
82
83 #include <vm/vm.h>
84 #include <vm/vm_extern.h>
85 #include <vm/vm_param.h>
86 #include <vm/pmap.h>
87 #include <vm/vm_map.h>
88 #include <vm/vm_page.h>
89 #include <vm/uma.h>
90
91 #ifdef KDTRACE_HOOKS
92 #include <sys/dtrace_bsd.h>
93 dtrace_execexit_func_t dtrace_fasttrap_exit;
94 #endif
95
96 SDT_PROVIDER_DECLARE(proc);
97 SDT_PROBE_DEFINE(proc, kernel, , exit, exit);
98 SDT_PROBE_ARGTYPE(proc, kernel, , exit, 0, "int");
99
100 /* Hook for NFS teardown procedure. */
101 void (*nlminfo_release_p)(struct proc *p);
102
103 static void
104 clear_orphan(struct proc *p)
105 {
106
107 PROC_LOCK_ASSERT(p, MA_OWNED);
108
109 if (p->p_flag & P_ORPHAN) {
110 LIST_REMOVE(p, p_orphan);
111 p->p_flag &= ~P_ORPHAN;
112 }
113 }
114
115 /*
116 * exit -- death of process.
117 */
118 void
119 sys_sys_exit(struct thread *td, struct sys_exit_args *uap)
120 {
121
122 exit1(td, W_EXITCODE(uap->rval, 0));
123 /* NOTREACHED */
124 }
125
126 /*
127 * Exit: deallocate address space and other resources, change proc state to
128 * zombie, and unlink proc from allproc and parent's lists. Save exit status
129 * and rusage for wait(). Check for child processes and orphan them.
130 */
131 void
132 exit1(struct thread *td, int rv)
133 {
134 struct proc *p, *nq, *q;
135 struct vnode *vtmp;
136 struct vnode *ttyvp = NULL;
137 struct plimit *plim;
138 int locked;
139
140 mtx_assert(&Giant, MA_NOTOWNED);
141
142 p = td->td_proc;
143 /*
144 * XXX in case we're rebooting we just let init die in order to
145 * work around an unsolved stack overflow seen very late during
146 * shutdown on sparc64 when the gmirror worker process exists.
147 */
148 if (p == initproc && rebooting == 0) {
149 printf("init died (signal %d, exit %d)\n",
150 WTERMSIG(rv), WEXITSTATUS(rv));
151 panic("Going nowhere without my init!");
152 }
153
154 /*
155 * MUST abort all other threads before proceeding past here.
156 */
157 PROC_LOCK(p);
158 while (p->p_flag & P_HADTHREADS) {
159 /*
160 * First check if some other thread got here before us..
161 * if so, act apropriatly, (exit or suspend);
162 */
163 thread_suspend_check(0);
164
165 /*
166 * Kill off the other threads. This requires
167 * some co-operation from other parts of the kernel
168 * so it may not be instantaneous. With this state set
169 * any thread entering the kernel from userspace will
170 * thread_exit() in trap(). Any thread attempting to
171 * sleep will return immediately with EINTR or EWOULDBLOCK
172 * which will hopefully force them to back out to userland
173 * freeing resources as they go. Any thread attempting
174 * to return to userland will thread_exit() from userret().
175 * thread_exit() will unsuspend us when the last of the
176 * other threads exits.
177 * If there is already a thread singler after resumption,
178 * calling thread_single will fail; in that case, we just
179 * re-check all suspension request, the thread should
180 * either be suspended there or exit.
181 */
182 if (! thread_single(SINGLE_EXIT))
183 break;
184
185 /*
186 * All other activity in this process is now stopped.
187 * Threading support has been turned off.
188 */
189 }
190 KASSERT(p->p_numthreads == 1,
191 ("exit1: proc %p exiting with %d threads", p, p->p_numthreads));
192 racct_sub(p, RACCT_NTHR, 1);
193 /*
194 * Wakeup anyone in procfs' PIOCWAIT. They should have a hold
195 * on our vmspace, so we should block below until they have
196 * released their reference to us. Note that if they have
197 * requested S_EXIT stops we will block here until they ack
198 * via PIOCCONT.
199 */
200 _STOPEVENT(p, S_EXIT, rv);
201
202 /*
203 * Ignore any pending request to stop due to a stop signal.
204 * Once P_WEXIT is set, future requests will be ignored as
205 * well.
206 */
207 p->p_flag &= ~P_STOPPED_SIG;
208 KASSERT(!P_SHOULDSTOP(p), ("exiting process is stopped"));
209
210 /*
211 * Note that we are exiting and do another wakeup of anyone in
212 * PIOCWAIT in case they aren't listening for S_EXIT stops or
213 * decided to wait again after we told them we are exiting.
214 */
215 p->p_flag |= P_WEXIT;
216 wakeup(&p->p_stype);
217
218 /*
219 * Wait for any processes that have a hold on our vmspace to
220 * release their reference.
221 */
222 while (p->p_lock > 0)
223 msleep(&p->p_lock, &p->p_mtx, PWAIT, "exithold", 0);
224
225 p->p_xstat = rv; /* Let event handler change exit status */
226 PROC_UNLOCK(p);
227 /* Drain the limit callout while we don't have the proc locked */
228 callout_drain(&p->p_limco);
229
230 #ifdef AUDIT
231 /*
232 * The Sun BSM exit token contains two components: an exit status as
233 * passed to exit(), and a return value to indicate what sort of exit
234 * it was. The exit status is WEXITSTATUS(rv), but it's not clear
235 * what the return value is.
236 */
237 AUDIT_ARG_EXIT(WEXITSTATUS(rv), 0);
238 AUDIT_SYSCALL_EXIT(0, td);
239 #endif
240
241 /* Are we a task leader? */
242 if (p == p->p_leader) {
243 mtx_lock(&ppeers_lock);
244 q = p->p_peers;
245 while (q != NULL) {
246 PROC_LOCK(q);
247 kern_psignal(q, SIGKILL);
248 PROC_UNLOCK(q);
249 q = q->p_peers;
250 }
251 while (p->p_peers != NULL)
252 msleep(p, &ppeers_lock, PWAIT, "exit1", 0);
253 mtx_unlock(&ppeers_lock);
254 }
255
256 /*
257 * Check if any loadable modules need anything done at process exit.
258 * E.g. SYSV IPC stuff
259 * XXX what if one of these generates an error?
260 */
261 EVENTHANDLER_INVOKE(process_exit, p);
262
263 /*
264 * If parent is waiting for us to exit or exec,
265 * P_PPWAIT is set; we will wakeup the parent below.
266 */
267 PROC_LOCK(p);
268 rv = p->p_xstat; /* Event handler could change exit status */
269 stopprofclock(p);
270 p->p_flag &= ~(P_TRACED | P_PPWAIT | P_PPTRACE);
271
272 /*
273 * Stop the real interval timer. If the handler is currently
274 * executing, prevent it from rearming itself and let it finish.
275 */
276 if (timevalisset(&p->p_realtimer.it_value) &&
277 callout_stop(&p->p_itcallout) == 0) {
278 timevalclear(&p->p_realtimer.it_interval);
279 msleep(&p->p_itcallout, &p->p_mtx, PWAIT, "ritwait", 0);
280 KASSERT(!timevalisset(&p->p_realtimer.it_value),
281 ("realtime timer is still armed"));
282 }
283 PROC_UNLOCK(p);
284
285 /*
286 * Reset any sigio structures pointing to us as a result of
287 * F_SETOWN with our pid.
288 */
289 funsetownlst(&p->p_sigiolst);
290
291 /*
292 * If this process has an nlminfo data area (for lockd), release it
293 */
294 if (nlminfo_release_p != NULL && p->p_nlminfo != NULL)
295 (*nlminfo_release_p)(p);
296
297 /*
298 * Close open files and release open-file table.
299 * This may block!
300 */
301 fdfree(td);
302
303 /*
304 * If this thread tickled GEOM, we need to wait for the giggling to
305 * stop before we return to userland
306 */
307 if (td->td_pflags & TDP_GEOM)
308 g_waitidle();
309
310 /*
311 * Remove ourself from our leader's peer list and wake our leader.
312 */
313 mtx_lock(&ppeers_lock);
314 if (p->p_leader->p_peers) {
315 q = p->p_leader;
316 while (q->p_peers != p)
317 q = q->p_peers;
318 q->p_peers = p->p_peers;
319 wakeup(p->p_leader);
320 }
321 mtx_unlock(&ppeers_lock);
322
323 vmspace_exit(td);
324
325 sx_xlock(&proctree_lock);
326 if (SESS_LEADER(p)) {
327 struct session *sp = p->p_session;
328 struct tty *tp;
329
330 /*
331 * s_ttyp is not zero'd; we use this to indicate that
332 * the session once had a controlling terminal. (for
333 * logging and informational purposes)
334 */
335 SESS_LOCK(sp);
336 ttyvp = sp->s_ttyvp;
337 tp = sp->s_ttyp;
338 sp->s_ttyvp = NULL;
339 sp->s_ttydp = NULL;
340 sp->s_leader = NULL;
341 SESS_UNLOCK(sp);
342
343 /*
344 * Signal foreground pgrp and revoke access to
345 * controlling terminal if it has not been revoked
346 * already.
347 *
348 * Because the TTY may have been revoked in the mean
349 * time and could already have a new session associated
350 * with it, make sure we don't send a SIGHUP to a
351 * foreground process group that does not belong to this
352 * session.
353 */
354
355 if (tp != NULL) {
356 tty_lock(tp);
357 if (tp->t_session == sp)
358 tty_signal_pgrp(tp, SIGHUP);
359 tty_unlock(tp);
360 }
361
362 if (ttyvp != NULL) {
363 sx_xunlock(&proctree_lock);
364 if (vn_lock(ttyvp, LK_EXCLUSIVE) == 0) {
365 VOP_REVOKE(ttyvp, REVOKEALL);
366 VOP_UNLOCK(ttyvp, 0);
367 }
368 sx_xlock(&proctree_lock);
369 }
370 }
371 fixjobc(p, p->p_pgrp, 0);
372 sx_xunlock(&proctree_lock);
373 (void)acct_process(td);
374
375 /* Release the TTY now we've unlocked everything. */
376 if (ttyvp != NULL)
377 vrele(ttyvp);
378 #ifdef KTRACE
379 ktrprocexit(td);
380 #endif
381 /*
382 * Release reference to text vnode
383 */
384 if ((vtmp = p->p_textvp) != NULL) {
385 p->p_textvp = NULL;
386 locked = VFS_LOCK_GIANT(vtmp->v_mount);
387 vrele(vtmp);
388 VFS_UNLOCK_GIANT(locked);
389 }
390
391 /*
392 * Release our limits structure.
393 */
394 PROC_LOCK(p);
395 plim = p->p_limit;
396 p->p_limit = NULL;
397 PROC_UNLOCK(p);
398 lim_free(plim);
399
400 tidhash_remove(td);
401
402 /*
403 * Remove proc from allproc queue and pidhash chain.
404 * Place onto zombproc. Unlink from parent's child list.
405 */
406 sx_xlock(&allproc_lock);
407 LIST_REMOVE(p, p_list);
408 LIST_INSERT_HEAD(&zombproc, p, p_list);
409 LIST_REMOVE(p, p_hash);
410 sx_xunlock(&allproc_lock);
411
412 /*
413 * Call machine-dependent code to release any
414 * machine-dependent resources other than the address space.
415 * The address space is released by "vmspace_exitfree(p)" in
416 * vm_waitproc().
417 */
418 cpu_exit(td);
419
420 WITNESS_WARN(WARN_PANIC, NULL, "process (pid %d) exiting", p->p_pid);
421
422 /*
423 * Reparent all of our children to init.
424 */
425 sx_xlock(&proctree_lock);
426 q = LIST_FIRST(&p->p_children);
427 if (q != NULL) /* only need this if any child is S_ZOMB */
428 wakeup(initproc);
429 for (; q != NULL; q = nq) {
430 nq = LIST_NEXT(q, p_sibling);
431 PROC_LOCK(q);
432 proc_reparent(q, initproc);
433 q->p_sigparent = SIGCHLD;
434 /*
435 * Traced processes are killed
436 * since their existence means someone is screwing up.
437 */
438 if (q->p_flag & P_TRACED) {
439 struct thread *temp;
440
441 /*
442 * Since q was found on our children list, the
443 * proc_reparent() call moved q to the orphan
444 * list due to present P_TRACED flag. Clear
445 * orphan link for q now while q is locked.
446 */
447 clear_orphan(q);
448 q->p_flag &= ~(P_TRACED | P_STOPPED_TRACE);
449 FOREACH_THREAD_IN_PROC(q, temp)
450 temp->td_dbgflags &= ~TDB_SUSPEND;
451 kern_psignal(q, SIGKILL);
452 }
453 PROC_UNLOCK(q);
454 }
455
456 /*
457 * Also get rid of our orphans.
458 */
459 while ((q = LIST_FIRST(&p->p_orphans)) != NULL) {
460 PROC_LOCK(q);
461 clear_orphan(q);
462 PROC_UNLOCK(q);
463 }
464
465 /* Save exit status. */
466 PROC_LOCK(p);
467 p->p_xthread = td;
468
469 /* Tell the prison that we are gone. */
470 prison_proc_free(p->p_ucred->cr_prison);
471
472 #ifdef KDTRACE_HOOKS
473 /*
474 * Tell the DTrace fasttrap provider about the exit if it
475 * has declared an interest.
476 */
477 if (dtrace_fasttrap_exit)
478 dtrace_fasttrap_exit(p);
479 #endif
480
481 /*
482 * Notify interested parties of our demise.
483 */
484 KNOTE_LOCKED(&p->p_klist, NOTE_EXIT);
485
486 #ifdef KDTRACE_HOOKS
487 int reason = CLD_EXITED;
488 if (WCOREDUMP(rv))
489 reason = CLD_DUMPED;
490 else if (WIFSIGNALED(rv))
491 reason = CLD_KILLED;
492 SDT_PROBE(proc, kernel, , exit, reason, 0, 0, 0, 0);
493 #endif
494
495 /*
496 * Just delete all entries in the p_klist. At this point we won't
497 * report any more events, and there are nasty race conditions that
498 * can beat us if we don't.
499 */
500 knlist_clear(&p->p_klist, 1);
501
502 /*
503 * If this is a process with a descriptor, we may not need to deliver
504 * a signal to the parent. proctree_lock is held over
505 * procdesc_exit() to serialize concurrent calls to close() and
506 * exit().
507 */
508 #ifdef PROCDESC
509 if (p->p_procdesc == NULL || procdesc_exit(p)) {
510 #endif
511 /*
512 * Notify parent that we're gone. If parent has the
513 * PS_NOCLDWAIT flag set, or if the handler is set to SIG_IGN,
514 * notify process 1 instead (and hope it will handle this
515 * situation).
516 */
517 PROC_LOCK(p->p_pptr);
518 mtx_lock(&p->p_pptr->p_sigacts->ps_mtx);
519 if (p->p_pptr->p_sigacts->ps_flag &
520 (PS_NOCLDWAIT | PS_CLDSIGIGN)) {
521 struct proc *pp;
522
523 mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx);
524 pp = p->p_pptr;
525 PROC_UNLOCK(pp);
526 proc_reparent(p, initproc);
527 p->p_sigparent = SIGCHLD;
528 PROC_LOCK(p->p_pptr);
529
530 /*
531 * Notify parent, so in case he was wait(2)ing or
532 * executing waitpid(2) with our pid, he will
533 * continue.
534 */
535 wakeup(pp);
536 } else
537 mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx);
538
539 if (p->p_pptr == initproc)
540 kern_psignal(p->p_pptr, SIGCHLD);
541 else if (p->p_sigparent != 0) {
542 if (p->p_sigparent == SIGCHLD)
543 childproc_exited(p);
544 else /* LINUX thread */
545 kern_psignal(p->p_pptr, p->p_sigparent);
546 }
547 #ifdef PROCDESC
548 } else
549 PROC_LOCK(p->p_pptr);
550 #endif
551 sx_xunlock(&proctree_lock);
552
553 /*
554 * The state PRS_ZOMBIE prevents other proesses from sending
555 * signal to the process, to avoid memory leak, we free memory
556 * for signal queue at the time when the state is set.
557 */
558 sigqueue_flush(&p->p_sigqueue);
559 sigqueue_flush(&td->td_sigqueue);
560
561 /*
562 * We have to wait until after acquiring all locks before
563 * changing p_state. We need to avoid all possible context
564 * switches (including ones from blocking on a mutex) while
565 * marked as a zombie. We also have to set the zombie state
566 * before we release the parent process' proc lock to avoid
567 * a lost wakeup. So, we first call wakeup, then we grab the
568 * sched lock, update the state, and release the parent process'
569 * proc lock.
570 */
571 wakeup(p->p_pptr);
572 cv_broadcast(&p->p_pwait);
573 sched_exit(p->p_pptr, td);
574 PROC_SLOCK(p);
575 p->p_state = PRS_ZOMBIE;
576 PROC_UNLOCK(p->p_pptr);
577
578 /*
579 * Hopefully no one will try to deliver a signal to the process this
580 * late in the game.
581 */
582 knlist_destroy(&p->p_klist);
583
584 /*
585 * Save our children's rusage information in our exit rusage.
586 */
587 ruadd(&p->p_ru, &p->p_rux, &p->p_stats->p_cru, &p->p_crux);
588
589 /*
590 * Make sure the scheduler takes this thread out of its tables etc.
591 * This will also release this thread's reference to the ucred.
592 * Other thread parts to release include pcb bits and such.
593 */
594 thread_exit();
595 }
596
597
598 #ifndef _SYS_SYSPROTO_H_
599 struct abort2_args {
600 char *why;
601 int nargs;
602 void **args;
603 };
604 #endif
605
606 int
607 sys_abort2(struct thread *td, struct abort2_args *uap)
608 {
609 struct proc *p = td->td_proc;
610 struct sbuf *sb;
611 void *uargs[16];
612 int error, i, sig;
613
614 /*
615 * Do it right now so we can log either proper call of abort2(), or
616 * note, that invalid argument was passed. 512 is big enough to
617 * handle 16 arguments' descriptions with additional comments.
618 */
619 sb = sbuf_new(NULL, NULL, 512, SBUF_FIXEDLEN);
620 sbuf_clear(sb);
621 sbuf_printf(sb, "%s(pid %d uid %d) aborted: ",
622 p->p_comm, p->p_pid, td->td_ucred->cr_uid);
623 /*
624 * Since we can't return from abort2(), send SIGKILL in cases, where
625 * abort2() was called improperly
626 */
627 sig = SIGKILL;
628 /* Prevent from DoSes from user-space. */
629 if (uap->nargs < 0 || uap->nargs > 16)
630 goto out;
631 if (uap->nargs > 0) {
632 if (uap->args == NULL)
633 goto out;
634 error = copyin(uap->args, uargs, uap->nargs * sizeof(void *));
635 if (error != 0)
636 goto out;
637 }
638 /*
639 * Limit size of 'reason' string to 128. Will fit even when
640 * maximal number of arguments was chosen to be logged.
641 */
642 if (uap->why != NULL) {
643 error = sbuf_copyin(sb, uap->why, 128);
644 if (error < 0)
645 goto out;
646 } else {
647 sbuf_printf(sb, "(null)");
648 }
649 if (uap->nargs > 0) {
650 sbuf_printf(sb, "(");
651 for (i = 0;i < uap->nargs; i++)
652 sbuf_printf(sb, "%s%p", i == 0 ? "" : ", ", uargs[i]);
653 sbuf_printf(sb, ")");
654 }
655 /*
656 * Final stage: arguments were proper, string has been
657 * successfully copied from userspace, and copying pointers
658 * from user-space succeed.
659 */
660 sig = SIGABRT;
661 out:
662 if (sig == SIGKILL) {
663 sbuf_trim(sb);
664 sbuf_printf(sb, " (Reason text inaccessible)");
665 }
666 sbuf_cat(sb, "\n");
667 sbuf_finish(sb);
668 log(LOG_INFO, "%s", sbuf_data(sb));
669 sbuf_delete(sb);
670 exit1(td, W_EXITCODE(0, sig));
671 return (0);
672 }
673
674
675 #ifdef COMPAT_43
676 /*
677 * The dirty work is handled by kern_wait().
678 */
679 int
680 owait(struct thread *td, struct owait_args *uap __unused)
681 {
682 int error, status;
683
684 error = kern_wait(td, WAIT_ANY, &status, 0, NULL);
685 if (error == 0)
686 td->td_retval[1] = status;
687 return (error);
688 }
689 #endif /* COMPAT_43 */
690
691 /*
692 * The dirty work is handled by kern_wait().
693 */
694 int
695 sys_wait4(struct thread *td, struct wait4_args *uap)
696 {
697 struct rusage ru, *rup;
698 int error, status;
699
700 if (uap->rusage != NULL)
701 rup = &ru;
702 else
703 rup = NULL;
704 error = kern_wait(td, uap->pid, &status, uap->options, rup);
705 if (uap->status != NULL && error == 0)
706 error = copyout(&status, uap->status, sizeof(status));
707 if (uap->rusage != NULL && error == 0)
708 error = copyout(&ru, uap->rusage, sizeof(struct rusage));
709 return (error);
710 }
711
712 int
713 sys_wait6(struct thread *td, struct wait6_args *uap)
714 {
715 struct __wrusage wru, *wrup;
716 siginfo_t si, *sip;
717 idtype_t idtype;
718 id_t id;
719 int error, status;
720
721 idtype = uap->idtype;
722 id = uap->id;
723
724 if (uap->wrusage != NULL)
725 wrup = &wru;
726 else
727 wrup = NULL;
728
729 if (uap->info != NULL) {
730 sip = &si;
731 bzero(sip, sizeof(*sip));
732 } else
733 sip = NULL;
734
735 /*
736 * We expect all callers of wait6() to know about WEXITED and
737 * WTRAPPED.
738 */
739 error = kern_wait6(td, idtype, id, &status, uap->options, wrup, sip);
740
741 if (uap->status != NULL && error == 0)
742 error = copyout(&status, uap->status, sizeof(status));
743 if (uap->wrusage != NULL && error == 0)
744 error = copyout(&wru, uap->wrusage, sizeof(wru));
745 if (uap->info != NULL && error == 0)
746 error = copyout(&si, uap->info, sizeof(si));
747 return (error);
748 }
749
750 /*
751 * Reap the remains of a zombie process and optionally return status and
752 * rusage. Asserts and will release both the proctree_lock and the process
753 * lock as part of its work.
754 */
755 void
756 proc_reap(struct thread *td, struct proc *p, int *status, int options)
757 {
758 struct proc *q, *t;
759
760 sx_assert(&proctree_lock, SA_XLOCKED);
761 PROC_LOCK_ASSERT(p, MA_OWNED);
762 PROC_SLOCK_ASSERT(p, MA_OWNED);
763 KASSERT(p->p_state == PRS_ZOMBIE, ("proc_reap: !PRS_ZOMBIE"));
764
765 q = td->td_proc;
766
767 PROC_SUNLOCK(p);
768 td->td_retval[0] = p->p_pid;
769 if (status)
770 *status = p->p_xstat; /* convert to int */
771 if (options & WNOWAIT) {
772 /*
773 * Only poll, returning the status. Caller does not wish to
774 * release the proc struct just yet.
775 */
776 PROC_UNLOCK(p);
777 sx_xunlock(&proctree_lock);
778 return;
779 }
780
781 PROC_LOCK(q);
782 sigqueue_take(p->p_ksi);
783 PROC_UNLOCK(q);
784 PROC_UNLOCK(p);
785
786 /*
787 * If we got the child via a ptrace 'attach', we need to give it back
788 * to the old parent.
789 */
790 if (p->p_oppid && (t = pfind(p->p_oppid)) != NULL) {
791 PROC_LOCK(p);
792 proc_reparent(p, t);
793 p->p_oppid = 0;
794 PROC_UNLOCK(p);
795 pksignal(t, SIGCHLD, p->p_ksi);
796 wakeup(t);
797 cv_broadcast(&p->p_pwait);
798 PROC_UNLOCK(t);
799 sx_xunlock(&proctree_lock);
800 return;
801 }
802
803 /*
804 * Remove other references to this process to ensure we have an
805 * exclusive reference.
806 */
807 sx_xlock(&allproc_lock);
808 LIST_REMOVE(p, p_list); /* off zombproc */
809 sx_xunlock(&allproc_lock);
810 LIST_REMOVE(p, p_sibling);
811 PROC_LOCK(p);
812 clear_orphan(p);
813 PROC_UNLOCK(p);
814 leavepgrp(p);
815 #ifdef PROCDESC
816 if (p->p_procdesc != NULL)
817 procdesc_reap(p);
818 #endif
819 sx_xunlock(&proctree_lock);
820
821 /*
822 * As a side effect of this lock, we know that all other writes to
823 * this proc are visible now, so no more locking is needed for p.
824 */
825 PROC_LOCK(p);
826 p->p_xstat = 0; /* XXX: why? */
827 PROC_UNLOCK(p);
828 PROC_LOCK(q);
829 ruadd(&q->p_stats->p_cru, &q->p_crux, &p->p_ru, &p->p_rux);
830 PROC_UNLOCK(q);
831
832 /*
833 * Decrement the count of procs running with this uid.
834 */
835 (void)chgproccnt(p->p_ucred->cr_ruidinfo, -1, 0);
836
837 /*
838 * Destroy resource accounting information associated with the process.
839 */
840 #ifdef RACCT
841 PROC_LOCK(p);
842 racct_sub(p, RACCT_NPROC, 1);
843 PROC_UNLOCK(p);
844 #endif
845 racct_proc_exit(p);
846
847 /*
848 * Free credentials, arguments, and sigacts.
849 */
850 crfree(p->p_ucred);
851 p->p_ucred = NULL;
852 pargs_drop(p->p_args);
853 p->p_args = NULL;
854 sigacts_free(p->p_sigacts);
855 p->p_sigacts = NULL;
856
857 /*
858 * Do any thread-system specific cleanups.
859 */
860 thread_wait(p);
861
862 /*
863 * Give vm and machine-dependent layer a chance to free anything that
864 * cpu_exit couldn't release while still running in process context.
865 */
866 vm_waitproc(p);
867 #ifdef MAC
868 mac_proc_destroy(p);
869 #endif
870 KASSERT(FIRST_THREAD_IN_PROC(p),
871 ("proc_reap: no residual thread!"));
872 uma_zfree(proc_zone, p);
873 sx_xlock(&allproc_lock);
874 nprocs--;
875 sx_xunlock(&allproc_lock);
876 }
877
878 static int
879 proc_to_reap(struct thread *td, struct proc *p, idtype_t idtype, id_t id,
880 int *status, int options, struct __wrusage *wrusage, siginfo_t *siginfo)
881 {
882 struct proc *q;
883 struct rusage *rup;
884
885 sx_assert(&proctree_lock, SA_XLOCKED);
886
887 q = td->td_proc;
888 PROC_LOCK(p);
889
890 switch (idtype) {
891 case P_ALL:
892 break;
893 case P_PID:
894 if (p->p_pid != (pid_t)id) {
895 PROC_UNLOCK(p);
896 return (0);
897 }
898 break;
899 case P_PGID:
900 if (p->p_pgid != (pid_t)id) {
901 PROC_UNLOCK(p);
902 return (0);
903 }
904 break;
905 case P_SID:
906 if (p->p_session->s_sid != (pid_t)id) {
907 PROC_UNLOCK(p);
908 return (0);
909 }
910 break;
911 case P_UID:
912 if (p->p_ucred->cr_uid != (uid_t)id) {
913 PROC_UNLOCK(p);
914 return (0);
915 }
916 break;
917 case P_GID:
918 if (p->p_ucred->cr_gid != (gid_t)id) {
919 PROC_UNLOCK(p);
920 return (0);
921 }
922 break;
923 case P_JAILID:
924 if (p->p_ucred->cr_prison == NULL ||
925 (p->p_ucred->cr_prison->pr_id != (int)id)) {
926 PROC_UNLOCK(p);
927 return (0);
928 }
929 break;
930 /*
931 * It seems that the thread structures get zeroed out
932 * at process exit. This makes it impossible to
933 * support P_SETID, P_CID or P_CPUID.
934 */
935 default:
936 PROC_UNLOCK(p);
937 return (0);
938 }
939
940 if (p_canwait(td, p)) {
941 PROC_UNLOCK(p);
942 return (0);
943 }
944
945 if (((options & WEXITED) == 0) && (p->p_state == PRS_ZOMBIE)) {
946 PROC_UNLOCK(p);
947 return (0);
948 }
949
950 /*
951 * This special case handles a kthread spawned by linux_clone
952 * (see linux_misc.c). The linux_wait4 and linux_waitpid
953 * functions need to be able to distinguish between waiting
954 * on a process and waiting on a thread. It is a thread if
955 * p_sigparent is not SIGCHLD, and the WLINUXCLONE option
956 * signifies we want to wait for threads and not processes.
957 */
958 if ((p->p_sigparent != SIGCHLD) ^
959 ((options & WLINUXCLONE) != 0)) {
960 PROC_UNLOCK(p);
961 return (0);
962 }
963
964 PROC_SLOCK(p);
965
966 if (siginfo != NULL) {
967 bzero(siginfo, sizeof(*siginfo));
968 siginfo->si_errno = 0;
969
970 /*
971 * SUSv4 requires that the si_signo value is always
972 * SIGCHLD. Obey it despite the rfork(2) interface
973 * allows to request other signal for child exit
974 * notification.
975 */
976 siginfo->si_signo = SIGCHLD;
977
978 /*
979 * This is still a rough estimate. We will fix the
980 * cases TRAPPED, STOPPED, and CONTINUED later.
981 */
982 if (WCOREDUMP(p->p_xstat))
983 siginfo->si_code = CLD_DUMPED;
984 else if (WIFSIGNALED(p->p_xstat))
985 siginfo->si_code = CLD_KILLED;
986 else
987 siginfo->si_code = CLD_EXITED;
988
989 siginfo->si_pid = p->p_pid;
990 siginfo->si_uid = p->p_ucred->cr_uid;
991 siginfo->si_status = p->p_xstat;
992
993 /*
994 * The si_addr field would be useful additional
995 * detail, but apparently the PC value may be lost
996 * when we reach this point. bzero() above sets
997 * siginfo->si_addr to NULL.
998 */
999 }
1000
1001 /*
1002 * There should be no reason to limit resources usage info to
1003 * exited processes only. A snapshot about any resources used
1004 * by a stopped process may be exactly what is needed.
1005 */
1006 if (wrusage != NULL) {
1007 rup = &wrusage->wru_self;
1008 *rup = p->p_ru;
1009 calcru(p, &rup->ru_utime, &rup->ru_stime);
1010
1011 rup = &wrusage->wru_children;
1012 *rup = p->p_stats->p_cru;
1013 calccru(p, &rup->ru_utime, &rup->ru_stime);
1014 }
1015
1016 if (p->p_state == PRS_ZOMBIE) {
1017 proc_reap(td, p, status, options);
1018 return (-1);
1019 }
1020 PROC_SUNLOCK(p);
1021 PROC_UNLOCK(p);
1022 return (1);
1023 }
1024
1025 int
1026 kern_wait(struct thread *td, pid_t pid, int *status, int options,
1027 struct rusage *rusage)
1028 {
1029 struct __wrusage wru, *wrup;
1030 idtype_t idtype;
1031 id_t id;
1032 int ret;
1033
1034 /*
1035 * Translate the special pid values into the (idtype, pid)
1036 * pair for kern_wait6. The WAIT_MYPGRP case is handled by
1037 * kern_wait6() on its own.
1038 */
1039 if (pid == WAIT_ANY) {
1040 idtype = P_ALL;
1041 id = 0;
1042 } else if (pid < 0) {
1043 idtype = P_PGID;
1044 id = (id_t)-pid;
1045 } else {
1046 idtype = P_PID;
1047 id = (id_t)pid;
1048 }
1049
1050 if (rusage != NULL)
1051 wrup = &wru;
1052 else
1053 wrup = NULL;
1054
1055 /*
1056 * For backward compatibility we implicitly add flags WEXITED
1057 * and WTRAPPED here.
1058 */
1059 options |= WEXITED | WTRAPPED;
1060 ret = kern_wait6(td, idtype, id, status, options, wrup, NULL);
1061 if (rusage != NULL)
1062 *rusage = wru.wru_self;
1063 return (ret);
1064 }
1065
1066 int
1067 kern_wait6(struct thread *td, idtype_t idtype, id_t id, int *status,
1068 int options, struct __wrusage *wrusage, siginfo_t *siginfo)
1069 {
1070 struct proc *p, *q;
1071 int error, nfound, ret;
1072
1073 AUDIT_ARG_VALUE((int)idtype); /* XXX - This is likely wrong! */
1074 AUDIT_ARG_PID((pid_t)id); /* XXX - This may be wrong! */
1075 AUDIT_ARG_VALUE(options);
1076
1077 q = td->td_proc;
1078
1079 if ((pid_t)id == WAIT_MYPGRP && (idtype == P_PID || idtype == P_PGID)) {
1080 PROC_LOCK(q);
1081 id = (id_t)q->p_pgid;
1082 PROC_UNLOCK(q);
1083 idtype = P_PGID;
1084 }
1085
1086 /* If we don't know the option, just return. */
1087 if ((options & ~(WUNTRACED | WNOHANG | WCONTINUED | WNOWAIT |
1088 WEXITED | WTRAPPED | WLINUXCLONE)) != 0)
1089 return (EINVAL);
1090 if ((options & (WEXITED | WUNTRACED | WCONTINUED | WTRAPPED)) == 0) {
1091 /*
1092 * We will be unable to find any matching processes,
1093 * because there are no known events to look for.
1094 * Prefer to return error instead of blocking
1095 * indefinitely.
1096 */
1097 return (EINVAL);
1098 }
1099
1100 loop:
1101 if (q->p_flag & P_STATCHILD) {
1102 PROC_LOCK(q);
1103 q->p_flag &= ~P_STATCHILD;
1104 PROC_UNLOCK(q);
1105 }
1106 nfound = 0;
1107 sx_xlock(&proctree_lock);
1108 LIST_FOREACH(p, &q->p_children, p_sibling) {
1109 ret = proc_to_reap(td, p, idtype, id, status, options,
1110 wrusage, siginfo);
1111 if (ret == 0)
1112 continue;
1113 else if (ret == 1)
1114 nfound++;
1115 else
1116 return (0);
1117
1118 PROC_LOCK(p);
1119 PROC_SLOCK(p);
1120
1121 if ((options & WTRAPPED) != 0 &&
1122 (p->p_flag & P_TRACED) != 0 &&
1123 (p->p_flag & (P_STOPPED_TRACE | P_STOPPED_SIG)) != 0 &&
1124 (p->p_suspcount == p->p_numthreads) &&
1125 ((p->p_flag & P_WAITED) == 0)) {
1126 PROC_SUNLOCK(p);
1127 if ((options & WNOWAIT) == 0)
1128 p->p_flag |= P_WAITED;
1129 sx_xunlock(&proctree_lock);
1130 td->td_retval[0] = p->p_pid;
1131
1132 if (status != NULL)
1133 *status = W_STOPCODE(p->p_xstat);
1134 if (siginfo != NULL) {
1135 siginfo->si_status = p->p_xstat;
1136 siginfo->si_code = CLD_TRAPPED;
1137 }
1138 if ((options & WNOWAIT) == 0) {
1139 PROC_LOCK(q);
1140 sigqueue_take(p->p_ksi);
1141 PROC_UNLOCK(q);
1142 }
1143
1144 PROC_UNLOCK(p);
1145 return (0);
1146 }
1147 if ((options & WUNTRACED) != 0 &&
1148 (p->p_flag & P_STOPPED_SIG) != 0 &&
1149 (p->p_suspcount == p->p_numthreads) &&
1150 ((p->p_flag & P_WAITED) == 0)) {
1151 PROC_SUNLOCK(p);
1152 if ((options & WNOWAIT) == 0)
1153 p->p_flag |= P_WAITED;
1154 sx_xunlock(&proctree_lock);
1155 td->td_retval[0] = p->p_pid;
1156
1157 if (status != NULL)
1158 *status = W_STOPCODE(p->p_xstat);
1159 if (siginfo != NULL) {
1160 siginfo->si_status = p->p_xstat;
1161 siginfo->si_code = CLD_STOPPED;
1162 }
1163 if ((options & WNOWAIT) == 0) {
1164 PROC_LOCK(q);
1165 sigqueue_take(p->p_ksi);
1166 PROC_UNLOCK(q);
1167 }
1168
1169 PROC_UNLOCK(p);
1170 return (0);
1171 }
1172 PROC_SUNLOCK(p);
1173 if ((options & WCONTINUED) != 0 &&
1174 (p->p_flag & P_CONTINUED) != 0) {
1175 sx_xunlock(&proctree_lock);
1176 td->td_retval[0] = p->p_pid;
1177 if ((options & WNOWAIT) == 0) {
1178 p->p_flag &= ~P_CONTINUED;
1179 PROC_LOCK(q);
1180 sigqueue_take(p->p_ksi);
1181 PROC_UNLOCK(q);
1182 }
1183 PROC_UNLOCK(p);
1184
1185 if (status != NULL)
1186 *status = SIGCONT;
1187 if (siginfo != NULL) {
1188 siginfo->si_status = SIGCONT;
1189 siginfo->si_code = CLD_CONTINUED;
1190 }
1191 return (0);
1192 }
1193 PROC_UNLOCK(p);
1194 }
1195
1196 /*
1197 * Look in the orphans list too, to allow the parent to
1198 * collect it's child exit status even if child is being
1199 * debugged.
1200 *
1201 * Debugger detaches from the parent upon successful
1202 * switch-over from parent to child. At this point due to
1203 * re-parenting the parent loses the child to debugger and a
1204 * wait4(2) call would report that it has no children to wait
1205 * for. By maintaining a list of orphans we allow the parent
1206 * to successfully wait until the child becomes a zombie.
1207 */
1208 LIST_FOREACH(p, &q->p_orphans, p_orphan) {
1209 ret = proc_to_reap(td, p, idtype, id, status, options,
1210 wrusage, siginfo);
1211 if (ret == 0)
1212 continue;
1213 else if (ret == 1)
1214 nfound++;
1215 else
1216 return (0);
1217 }
1218 if (nfound == 0) {
1219 sx_xunlock(&proctree_lock);
1220 return (ECHILD);
1221 }
1222 if (options & WNOHANG) {
1223 sx_xunlock(&proctree_lock);
1224 td->td_retval[0] = 0;
1225 return (0);
1226 }
1227 PROC_LOCK(q);
1228 sx_xunlock(&proctree_lock);
1229 if (q->p_flag & P_STATCHILD) {
1230 q->p_flag &= ~P_STATCHILD;
1231 error = 0;
1232 } else
1233 error = msleep(q, &q->p_mtx, PWAIT | PCATCH, "wait", 0);
1234 PROC_UNLOCK(q);
1235 if (error)
1236 return (error);
1237 goto loop;
1238 }
1239
1240 /*
1241 * Make process 'parent' the new parent of process 'child'.
1242 * Must be called with an exclusive hold of proctree lock.
1243 */
1244 void
1245 proc_reparent(struct proc *child, struct proc *parent)
1246 {
1247
1248 sx_assert(&proctree_lock, SX_XLOCKED);
1249 PROC_LOCK_ASSERT(child, MA_OWNED);
1250 if (child->p_pptr == parent)
1251 return;
1252
1253 PROC_LOCK(child->p_pptr);
1254 sigqueue_take(child->p_ksi);
1255 PROC_UNLOCK(child->p_pptr);
1256 LIST_REMOVE(child, p_sibling);
1257 LIST_INSERT_HEAD(&parent->p_children, child, p_sibling);
1258
1259 clear_orphan(child);
1260 if (child->p_flag & P_TRACED) {
1261 LIST_INSERT_HEAD(&child->p_pptr->p_orphans, child, p_orphan);
1262 child->p_flag |= P_ORPHAN;
1263 }
1264
1265 child->p_pptr = parent;
1266 }
Cache object: ee5e37e1efb7af05779ceec848a6a32f
|