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/10.0/sys/kern/kern_exit.c 254350 2013-08-15 04:08:55Z markj $");
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_DEFINE1(proc, kernel, , exit, exit, "int");
98
99 /* Hook for NFS teardown procedure. */
100 void (*nlminfo_release_p)(struct proc *p);
101
102 static void
103 clear_orphan(struct proc *p)
104 {
105
106 PROC_LOCK_ASSERT(p, MA_OWNED);
107
108 if (p->p_flag & P_ORPHAN) {
109 LIST_REMOVE(p, p_orphan);
110 p->p_flag &= ~P_ORPHAN;
111 }
112 }
113
114 /*
115 * exit -- death of process.
116 */
117 void
118 sys_sys_exit(struct thread *td, struct sys_exit_args *uap)
119 {
120
121 exit1(td, W_EXITCODE(uap->rval, 0));
122 /* NOTREACHED */
123 }
124
125 /*
126 * Exit: deallocate address space and other resources, change proc state to
127 * zombie, and unlink proc from allproc and parent's lists. Save exit status
128 * and rusage for wait(). Check for child processes and orphan them.
129 */
130 void
131 exit1(struct thread *td, int rv)
132 {
133 struct proc *p, *nq, *q;
134 struct vnode *vtmp;
135 struct vnode *ttyvp = NULL;
136 struct plimit *plim;
137
138 mtx_assert(&Giant, MA_NOTOWNED);
139
140 p = td->td_proc;
141 /*
142 * XXX in case we're rebooting we just let init die in order to
143 * work around an unsolved stack overflow seen very late during
144 * shutdown on sparc64 when the gmirror worker process exists.
145 */
146 if (p == initproc && rebooting == 0) {
147 printf("init died (signal %d, exit %d)\n",
148 WTERMSIG(rv), WEXITSTATUS(rv));
149 panic("Going nowhere without my init!");
150 }
151
152 /*
153 * MUST abort all other threads before proceeding past here.
154 */
155 PROC_LOCK(p);
156 while (p->p_flag & P_HADTHREADS) {
157 /*
158 * First check if some other thread got here before us.
159 * If so, act appropriately: exit or suspend.
160 */
161 thread_suspend_check(0);
162
163 /*
164 * Kill off the other threads. This requires
165 * some co-operation from other parts of the kernel
166 * so it may not be instantaneous. With this state set
167 * any thread entering the kernel from userspace will
168 * thread_exit() in trap(). Any thread attempting to
169 * sleep will return immediately with EINTR or EWOULDBLOCK
170 * which will hopefully force them to back out to userland
171 * freeing resources as they go. Any thread attempting
172 * to return to userland will thread_exit() from userret().
173 * thread_exit() will unsuspend us when the last of the
174 * other threads exits.
175 * If there is already a thread singler after resumption,
176 * calling thread_single will fail; in that case, we just
177 * re-check all suspension request, the thread should
178 * either be suspended there or exit.
179 */
180 if (!thread_single(SINGLE_EXIT))
181 break;
182
183 /*
184 * All other activity in this process is now stopped.
185 * Threading support has been turned off.
186 */
187 }
188 KASSERT(p->p_numthreads == 1,
189 ("exit1: proc %p exiting with %d threads", p, p->p_numthreads));
190 racct_sub(p, RACCT_NTHR, 1);
191 /*
192 * Wakeup anyone in procfs' PIOCWAIT. They should have a hold
193 * on our vmspace, so we should block below until they have
194 * released their reference to us. Note that if they have
195 * requested S_EXIT stops we will block here until they ack
196 * via PIOCCONT.
197 */
198 _STOPEVENT(p, S_EXIT, rv);
199
200 /*
201 * Ignore any pending request to stop due to a stop signal.
202 * Once P_WEXIT is set, future requests will be ignored as
203 * well.
204 */
205 p->p_flag &= ~P_STOPPED_SIG;
206 KASSERT(!P_SHOULDSTOP(p), ("exiting process is stopped"));
207
208 /*
209 * Note that we are exiting and do another wakeup of anyone in
210 * PIOCWAIT in case they aren't listening for S_EXIT stops or
211 * decided to wait again after we told them we are exiting.
212 */
213 p->p_flag |= P_WEXIT;
214 wakeup(&p->p_stype);
215
216 /*
217 * Wait for any processes that have a hold on our vmspace to
218 * release their reference.
219 */
220 while (p->p_lock > 0)
221 msleep(&p->p_lock, &p->p_mtx, PWAIT, "exithold", 0);
222
223 p->p_xstat = rv; /* Let event handler change exit status */
224 PROC_UNLOCK(p);
225 /* Drain the limit callout while we don't have the proc locked */
226 callout_drain(&p->p_limco);
227
228 #ifdef AUDIT
229 /*
230 * The Sun BSM exit token contains two components: an exit status as
231 * passed to exit(), and a return value to indicate what sort of exit
232 * it was. The exit status is WEXITSTATUS(rv), but it's not clear
233 * what the return value is.
234 */
235 AUDIT_ARG_EXIT(WEXITSTATUS(rv), 0);
236 AUDIT_SYSCALL_EXIT(0, td);
237 #endif
238
239 /* Are we a task leader? */
240 if (p == p->p_leader) {
241 mtx_lock(&ppeers_lock);
242 q = p->p_peers;
243 while (q != NULL) {
244 PROC_LOCK(q);
245 kern_psignal(q, SIGKILL);
246 PROC_UNLOCK(q);
247 q = q->p_peers;
248 }
249 while (p->p_peers != NULL)
250 msleep(p, &ppeers_lock, PWAIT, "exit1", 0);
251 mtx_unlock(&ppeers_lock);
252 }
253
254 /*
255 * Check if any loadable modules need anything done at process exit.
256 * E.g. SYSV IPC stuff
257 * XXX what if one of these generates an error?
258 */
259 EVENTHANDLER_INVOKE(process_exit, p);
260
261 /*
262 * If parent is waiting for us to exit or exec,
263 * P_PPWAIT is set; we will wakeup the parent below.
264 */
265 PROC_LOCK(p);
266 rv = p->p_xstat; /* Event handler could change exit status */
267 stopprofclock(p);
268 p->p_flag &= ~(P_TRACED | P_PPWAIT | P_PPTRACE);
269
270 /*
271 * Stop the real interval timer. If the handler is currently
272 * executing, prevent it from rearming itself and let it finish.
273 */
274 if (timevalisset(&p->p_realtimer.it_value) &&
275 callout_stop(&p->p_itcallout) == 0) {
276 timevalclear(&p->p_realtimer.it_interval);
277 msleep(&p->p_itcallout, &p->p_mtx, PWAIT, "ritwait", 0);
278 KASSERT(!timevalisset(&p->p_realtimer.it_value),
279 ("realtime timer is still armed"));
280 }
281 PROC_UNLOCK(p);
282
283 /*
284 * Reset any sigio structures pointing to us as a result of
285 * F_SETOWN with our pid.
286 */
287 funsetownlst(&p->p_sigiolst);
288
289 /*
290 * If this process has an nlminfo data area (for lockd), release it
291 */
292 if (nlminfo_release_p != NULL && p->p_nlminfo != NULL)
293 (*nlminfo_release_p)(p);
294
295 /*
296 * Close open files and release open-file table.
297 * This may block!
298 */
299 fdescfree(td);
300
301 /*
302 * If this thread tickled GEOM, we need to wait for the giggling to
303 * stop before we return to userland
304 */
305 if (td->td_pflags & TDP_GEOM)
306 g_waitidle();
307
308 /*
309 * Remove ourself from our leader's peer list and wake our leader.
310 */
311 mtx_lock(&ppeers_lock);
312 if (p->p_leader->p_peers) {
313 q = p->p_leader;
314 while (q->p_peers != p)
315 q = q->p_peers;
316 q->p_peers = p->p_peers;
317 wakeup(p->p_leader);
318 }
319 mtx_unlock(&ppeers_lock);
320
321 vmspace_exit(td);
322
323 sx_xlock(&proctree_lock);
324 if (SESS_LEADER(p)) {
325 struct session *sp = p->p_session;
326 struct tty *tp;
327
328 /*
329 * s_ttyp is not zero'd; we use this to indicate that
330 * the session once had a controlling terminal. (for
331 * logging and informational purposes)
332 */
333 SESS_LOCK(sp);
334 ttyvp = sp->s_ttyvp;
335 tp = sp->s_ttyp;
336 sp->s_ttyvp = NULL;
337 sp->s_ttydp = NULL;
338 sp->s_leader = NULL;
339 SESS_UNLOCK(sp);
340
341 /*
342 * Signal foreground pgrp and revoke access to
343 * controlling terminal if it has not been revoked
344 * already.
345 *
346 * Because the TTY may have been revoked in the mean
347 * time and could already have a new session associated
348 * with it, make sure we don't send a SIGHUP to a
349 * foreground process group that does not belong to this
350 * session.
351 */
352
353 if (tp != NULL) {
354 tty_lock(tp);
355 if (tp->t_session == sp)
356 tty_signal_pgrp(tp, SIGHUP);
357 tty_unlock(tp);
358 }
359
360 if (ttyvp != NULL) {
361 sx_xunlock(&proctree_lock);
362 if (vn_lock(ttyvp, LK_EXCLUSIVE) == 0) {
363 VOP_REVOKE(ttyvp, REVOKEALL);
364 VOP_UNLOCK(ttyvp, 0);
365 }
366 sx_xlock(&proctree_lock);
367 }
368 }
369 fixjobc(p, p->p_pgrp, 0);
370 sx_xunlock(&proctree_lock);
371 (void)acct_process(td);
372
373 /* Release the TTY now we've unlocked everything. */
374 if (ttyvp != NULL)
375 vrele(ttyvp);
376 #ifdef KTRACE
377 ktrprocexit(td);
378 #endif
379 /*
380 * Release reference to text vnode
381 */
382 if ((vtmp = p->p_textvp) != NULL) {
383 p->p_textvp = NULL;
384 vrele(vtmp);
385 }
386
387 /*
388 * Release our limits structure.
389 */
390 PROC_LOCK(p);
391 plim = p->p_limit;
392 p->p_limit = NULL;
393 PROC_UNLOCK(p);
394 lim_free(plim);
395
396 tidhash_remove(td);
397
398 /*
399 * Remove proc from allproc queue and pidhash chain.
400 * Place onto zombproc. Unlink from parent's child list.
401 */
402 sx_xlock(&allproc_lock);
403 LIST_REMOVE(p, p_list);
404 LIST_INSERT_HEAD(&zombproc, p, p_list);
405 LIST_REMOVE(p, p_hash);
406 sx_xunlock(&allproc_lock);
407
408 /*
409 * Call machine-dependent code to release any
410 * machine-dependent resources other than the address space.
411 * The address space is released by "vmspace_exitfree(p)" in
412 * vm_waitproc().
413 */
414 cpu_exit(td);
415
416 WITNESS_WARN(WARN_PANIC, NULL, "process (pid %d) exiting", p->p_pid);
417
418 /*
419 * Reparent all of our children to init.
420 */
421 sx_xlock(&proctree_lock);
422 q = LIST_FIRST(&p->p_children);
423 if (q != NULL) /* only need this if any child is S_ZOMB */
424 wakeup(initproc);
425 for (; q != NULL; q = nq) {
426 nq = LIST_NEXT(q, p_sibling);
427 PROC_LOCK(q);
428 proc_reparent(q, initproc);
429 q->p_sigparent = SIGCHLD;
430 /*
431 * Traced processes are killed
432 * since their existence means someone is screwing up.
433 */
434 if (q->p_flag & P_TRACED) {
435 struct thread *temp;
436
437 /*
438 * Since q was found on our children list, the
439 * proc_reparent() call moved q to the orphan
440 * list due to present P_TRACED flag. Clear
441 * orphan link for q now while q is locked.
442 */
443 clear_orphan(q);
444 q->p_flag &= ~(P_TRACED | P_STOPPED_TRACE);
445 FOREACH_THREAD_IN_PROC(q, temp)
446 temp->td_dbgflags &= ~TDB_SUSPEND;
447 kern_psignal(q, SIGKILL);
448 }
449 PROC_UNLOCK(q);
450 }
451
452 /*
453 * Also get rid of our orphans.
454 */
455 while ((q = LIST_FIRST(&p->p_orphans)) != NULL) {
456 PROC_LOCK(q);
457 clear_orphan(q);
458 PROC_UNLOCK(q);
459 }
460
461 /* Save exit status. */
462 PROC_LOCK(p);
463 p->p_xthread = td;
464
465 /* Tell the prison that we are gone. */
466 prison_proc_free(p->p_ucred->cr_prison);
467
468 #ifdef KDTRACE_HOOKS
469 /*
470 * Tell the DTrace fasttrap provider about the exit if it
471 * has declared an interest.
472 */
473 if (dtrace_fasttrap_exit)
474 dtrace_fasttrap_exit(p);
475 #endif
476
477 /*
478 * Notify interested parties of our demise.
479 */
480 KNOTE_LOCKED(&p->p_klist, NOTE_EXIT);
481
482 #ifdef KDTRACE_HOOKS
483 int reason = CLD_EXITED;
484 if (WCOREDUMP(rv))
485 reason = CLD_DUMPED;
486 else if (WIFSIGNALED(rv))
487 reason = CLD_KILLED;
488 SDT_PROBE(proc, kernel, , exit, reason, 0, 0, 0, 0);
489 #endif
490
491 /*
492 * Just delete all entries in the p_klist. At this point we won't
493 * report any more events, and there are nasty race conditions that
494 * can beat us if we don't.
495 */
496 knlist_clear(&p->p_klist, 1);
497
498 /*
499 * If this is a process with a descriptor, we may not need to deliver
500 * a signal to the parent. proctree_lock is held over
501 * procdesc_exit() to serialize concurrent calls to close() and
502 * exit().
503 */
504 #ifdef PROCDESC
505 if (p->p_procdesc == NULL || procdesc_exit(p)) {
506 #endif
507 /*
508 * Notify parent that we're gone. If parent has the
509 * PS_NOCLDWAIT flag set, or if the handler is set to SIG_IGN,
510 * notify process 1 instead (and hope it will handle this
511 * situation).
512 */
513 PROC_LOCK(p->p_pptr);
514 mtx_lock(&p->p_pptr->p_sigacts->ps_mtx);
515 if (p->p_pptr->p_sigacts->ps_flag &
516 (PS_NOCLDWAIT | PS_CLDSIGIGN)) {
517 struct proc *pp;
518
519 mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx);
520 pp = p->p_pptr;
521 PROC_UNLOCK(pp);
522 proc_reparent(p, initproc);
523 p->p_sigparent = SIGCHLD;
524 PROC_LOCK(p->p_pptr);
525
526 /*
527 * Notify parent, so in case he was wait(2)ing or
528 * executing waitpid(2) with our pid, he will
529 * continue.
530 */
531 wakeup(pp);
532 } else
533 mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx);
534
535 if (p->p_pptr == initproc)
536 kern_psignal(p->p_pptr, SIGCHLD);
537 else if (p->p_sigparent != 0) {
538 if (p->p_sigparent == SIGCHLD)
539 childproc_exited(p);
540 else /* LINUX thread */
541 kern_psignal(p->p_pptr, p->p_sigparent);
542 }
543 #ifdef PROCDESC
544 } else
545 PROC_LOCK(p->p_pptr);
546 #endif
547 sx_xunlock(&proctree_lock);
548
549 /*
550 * The state PRS_ZOMBIE prevents other proesses from sending
551 * signal to the process, to avoid memory leak, we free memory
552 * for signal queue at the time when the state is set.
553 */
554 sigqueue_flush(&p->p_sigqueue);
555 sigqueue_flush(&td->td_sigqueue);
556
557 /*
558 * We have to wait until after acquiring all locks before
559 * changing p_state. We need to avoid all possible context
560 * switches (including ones from blocking on a mutex) while
561 * marked as a zombie. We also have to set the zombie state
562 * before we release the parent process' proc lock to avoid
563 * a lost wakeup. So, we first call wakeup, then we grab the
564 * sched lock, update the state, and release the parent process'
565 * proc lock.
566 */
567 wakeup(p->p_pptr);
568 cv_broadcast(&p->p_pwait);
569 sched_exit(p->p_pptr, td);
570 PROC_SLOCK(p);
571 p->p_state = PRS_ZOMBIE;
572 PROC_UNLOCK(p->p_pptr);
573
574 /*
575 * Hopefully no one will try to deliver a signal to the process this
576 * late in the game.
577 */
578 knlist_destroy(&p->p_klist);
579
580 /*
581 * Save our children's rusage information in our exit rusage.
582 */
583 ruadd(&p->p_ru, &p->p_rux, &p->p_stats->p_cru, &p->p_crux);
584
585 /*
586 * Make sure the scheduler takes this thread out of its tables etc.
587 * This will also release this thread's reference to the ucred.
588 * Other thread parts to release include pcb bits and such.
589 */
590 thread_exit();
591 }
592
593
594 #ifndef _SYS_SYSPROTO_H_
595 struct abort2_args {
596 char *why;
597 int nargs;
598 void **args;
599 };
600 #endif
601
602 int
603 sys_abort2(struct thread *td, struct abort2_args *uap)
604 {
605 struct proc *p = td->td_proc;
606 struct sbuf *sb;
607 void *uargs[16];
608 int error, i, sig;
609
610 /*
611 * Do it right now so we can log either proper call of abort2(), or
612 * note, that invalid argument was passed. 512 is big enough to
613 * handle 16 arguments' descriptions with additional comments.
614 */
615 sb = sbuf_new(NULL, NULL, 512, SBUF_FIXEDLEN);
616 sbuf_clear(sb);
617 sbuf_printf(sb, "%s(pid %d uid %d) aborted: ",
618 p->p_comm, p->p_pid, td->td_ucred->cr_uid);
619 /*
620 * Since we can't return from abort2(), send SIGKILL in cases, where
621 * abort2() was called improperly
622 */
623 sig = SIGKILL;
624 /* Prevent from DoSes from user-space. */
625 if (uap->nargs < 0 || uap->nargs > 16)
626 goto out;
627 if (uap->nargs > 0) {
628 if (uap->args == NULL)
629 goto out;
630 error = copyin(uap->args, uargs, uap->nargs * sizeof(void *));
631 if (error != 0)
632 goto out;
633 }
634 /*
635 * Limit size of 'reason' string to 128. Will fit even when
636 * maximal number of arguments was chosen to be logged.
637 */
638 if (uap->why != NULL) {
639 error = sbuf_copyin(sb, uap->why, 128);
640 if (error < 0)
641 goto out;
642 } else {
643 sbuf_printf(sb, "(null)");
644 }
645 if (uap->nargs > 0) {
646 sbuf_printf(sb, "(");
647 for (i = 0;i < uap->nargs; i++)
648 sbuf_printf(sb, "%s%p", i == 0 ? "" : ", ", uargs[i]);
649 sbuf_printf(sb, ")");
650 }
651 /*
652 * Final stage: arguments were proper, string has been
653 * successfully copied from userspace, and copying pointers
654 * from user-space succeed.
655 */
656 sig = SIGABRT;
657 out:
658 if (sig == SIGKILL) {
659 sbuf_trim(sb);
660 sbuf_printf(sb, " (Reason text inaccessible)");
661 }
662 sbuf_cat(sb, "\n");
663 sbuf_finish(sb);
664 log(LOG_INFO, "%s", sbuf_data(sb));
665 sbuf_delete(sb);
666 exit1(td, W_EXITCODE(0, sig));
667 return (0);
668 }
669
670
671 #ifdef COMPAT_43
672 /*
673 * The dirty work is handled by kern_wait().
674 */
675 int
676 owait(struct thread *td, struct owait_args *uap __unused)
677 {
678 int error, status;
679
680 error = kern_wait(td, WAIT_ANY, &status, 0, NULL);
681 if (error == 0)
682 td->td_retval[1] = status;
683 return (error);
684 }
685 #endif /* COMPAT_43 */
686
687 /*
688 * The dirty work is handled by kern_wait().
689 */
690 int
691 sys_wait4(struct thread *td, struct wait4_args *uap)
692 {
693 struct rusage ru, *rup;
694 int error, status;
695
696 if (uap->rusage != NULL)
697 rup = &ru;
698 else
699 rup = NULL;
700 error = kern_wait(td, uap->pid, &status, uap->options, rup);
701 if (uap->status != NULL && error == 0)
702 error = copyout(&status, uap->status, sizeof(status));
703 if (uap->rusage != NULL && error == 0)
704 error = copyout(&ru, uap->rusage, sizeof(struct rusage));
705 return (error);
706 }
707
708 int
709 sys_wait6(struct thread *td, struct wait6_args *uap)
710 {
711 struct __wrusage wru, *wrup;
712 siginfo_t si, *sip;
713 idtype_t idtype;
714 id_t id;
715 int error, status;
716
717 idtype = uap->idtype;
718 id = uap->id;
719
720 if (uap->wrusage != NULL)
721 wrup = &wru;
722 else
723 wrup = NULL;
724
725 if (uap->info != NULL) {
726 sip = &si;
727 bzero(sip, sizeof(*sip));
728 } else
729 sip = NULL;
730
731 /*
732 * We expect all callers of wait6() to know about WEXITED and
733 * WTRAPPED.
734 */
735 error = kern_wait6(td, idtype, id, &status, uap->options, wrup, sip);
736
737 if (uap->status != NULL && error == 0)
738 error = copyout(&status, uap->status, sizeof(status));
739 if (uap->wrusage != NULL && error == 0)
740 error = copyout(&wru, uap->wrusage, sizeof(wru));
741 if (uap->info != NULL && error == 0)
742 error = copyout(&si, uap->info, sizeof(si));
743 return (error);
744 }
745
746 /*
747 * Reap the remains of a zombie process and optionally return status and
748 * rusage. Asserts and will release both the proctree_lock and the process
749 * lock as part of its work.
750 */
751 void
752 proc_reap(struct thread *td, struct proc *p, int *status, int options)
753 {
754 struct proc *q, *t;
755
756 sx_assert(&proctree_lock, SA_XLOCKED);
757 PROC_LOCK_ASSERT(p, MA_OWNED);
758 PROC_SLOCK_ASSERT(p, MA_OWNED);
759 KASSERT(p->p_state == PRS_ZOMBIE, ("proc_reap: !PRS_ZOMBIE"));
760
761 q = td->td_proc;
762
763 PROC_SUNLOCK(p);
764 td->td_retval[0] = p->p_pid;
765 if (status)
766 *status = p->p_xstat; /* convert to int */
767 if (options & WNOWAIT) {
768 /*
769 * Only poll, returning the status. Caller does not wish to
770 * release the proc struct just yet.
771 */
772 PROC_UNLOCK(p);
773 sx_xunlock(&proctree_lock);
774 return;
775 }
776
777 PROC_LOCK(q);
778 sigqueue_take(p->p_ksi);
779 PROC_UNLOCK(q);
780 PROC_UNLOCK(p);
781
782 /*
783 * If we got the child via a ptrace 'attach', we need to give it back
784 * to the old parent.
785 */
786 if (p->p_oppid && (t = pfind(p->p_oppid)) != NULL) {
787 PROC_LOCK(p);
788 proc_reparent(p, t);
789 p->p_oppid = 0;
790 PROC_UNLOCK(p);
791 pksignal(t, SIGCHLD, p->p_ksi);
792 wakeup(t);
793 cv_broadcast(&p->p_pwait);
794 PROC_UNLOCK(t);
795 sx_xunlock(&proctree_lock);
796 return;
797 }
798
799 /*
800 * Remove other references to this process to ensure we have an
801 * exclusive reference.
802 */
803 sx_xlock(&allproc_lock);
804 LIST_REMOVE(p, p_list); /* off zombproc */
805 sx_xunlock(&allproc_lock);
806 LIST_REMOVE(p, p_sibling);
807 PROC_LOCK(p);
808 clear_orphan(p);
809 PROC_UNLOCK(p);
810 leavepgrp(p);
811 #ifdef PROCDESC
812 if (p->p_procdesc != NULL)
813 procdesc_reap(p);
814 #endif
815 sx_xunlock(&proctree_lock);
816
817 /*
818 * As a side effect of this lock, we know that all other writes to
819 * this proc are visible now, so no more locking is needed for p.
820 */
821 PROC_LOCK(p);
822 p->p_xstat = 0; /* XXX: why? */
823 PROC_UNLOCK(p);
824 PROC_LOCK(q);
825 ruadd(&q->p_stats->p_cru, &q->p_crux, &p->p_ru, &p->p_rux);
826 PROC_UNLOCK(q);
827
828 /*
829 * Decrement the count of procs running with this uid.
830 */
831 (void)chgproccnt(p->p_ucred->cr_ruidinfo, -1, 0);
832
833 /*
834 * Destroy resource accounting information associated with the process.
835 */
836 #ifdef RACCT
837 PROC_LOCK(p);
838 racct_sub(p, RACCT_NPROC, 1);
839 PROC_UNLOCK(p);
840 #endif
841 racct_proc_exit(p);
842
843 /*
844 * Free credentials, arguments, and sigacts.
845 */
846 crfree(p->p_ucred);
847 p->p_ucred = NULL;
848 pargs_drop(p->p_args);
849 p->p_args = NULL;
850 sigacts_free(p->p_sigacts);
851 p->p_sigacts = NULL;
852
853 /*
854 * Do any thread-system specific cleanups.
855 */
856 thread_wait(p);
857
858 /*
859 * Give vm and machine-dependent layer a chance to free anything that
860 * cpu_exit couldn't release while still running in process context.
861 */
862 vm_waitproc(p);
863 #ifdef MAC
864 mac_proc_destroy(p);
865 #endif
866 KASSERT(FIRST_THREAD_IN_PROC(p),
867 ("proc_reap: no residual thread!"));
868 uma_zfree(proc_zone, p);
869 sx_xlock(&allproc_lock);
870 nprocs--;
871 sx_xunlock(&allproc_lock);
872 }
873
874 static int
875 proc_to_reap(struct thread *td, struct proc *p, idtype_t idtype, id_t id,
876 int *status, int options, struct __wrusage *wrusage, siginfo_t *siginfo)
877 {
878 struct proc *q;
879 struct rusage *rup;
880
881 sx_assert(&proctree_lock, SA_XLOCKED);
882
883 q = td->td_proc;
884 PROC_LOCK(p);
885
886 switch (idtype) {
887 case P_ALL:
888 break;
889 case P_PID:
890 if (p->p_pid != (pid_t)id) {
891 PROC_UNLOCK(p);
892 return (0);
893 }
894 break;
895 case P_PGID:
896 if (p->p_pgid != (pid_t)id) {
897 PROC_UNLOCK(p);
898 return (0);
899 }
900 break;
901 case P_SID:
902 if (p->p_session->s_sid != (pid_t)id) {
903 PROC_UNLOCK(p);
904 return (0);
905 }
906 break;
907 case P_UID:
908 if (p->p_ucred->cr_uid != (uid_t)id) {
909 PROC_UNLOCK(p);
910 return (0);
911 }
912 break;
913 case P_GID:
914 if (p->p_ucred->cr_gid != (gid_t)id) {
915 PROC_UNLOCK(p);
916 return (0);
917 }
918 break;
919 case P_JAILID:
920 if (p->p_ucred->cr_prison->pr_id != (int)id) {
921 PROC_UNLOCK(p);
922 return (0);
923 }
924 break;
925 /*
926 * It seems that the thread structures get zeroed out
927 * at process exit. This makes it impossible to
928 * support P_SETID, P_CID or P_CPUID.
929 */
930 default:
931 PROC_UNLOCK(p);
932 return (0);
933 }
934
935 if (p_canwait(td, p)) {
936 PROC_UNLOCK(p);
937 return (0);
938 }
939
940 if (((options & WEXITED) == 0) && (p->p_state == PRS_ZOMBIE)) {
941 PROC_UNLOCK(p);
942 return (0);
943 }
944
945 /*
946 * This special case handles a kthread spawned by linux_clone
947 * (see linux_misc.c). The linux_wait4 and linux_waitpid
948 * functions need to be able to distinguish between waiting
949 * on a process and waiting on a thread. It is a thread if
950 * p_sigparent is not SIGCHLD, and the WLINUXCLONE option
951 * signifies we want to wait for threads and not processes.
952 */
953 if ((p->p_sigparent != SIGCHLD) ^
954 ((options & WLINUXCLONE) != 0)) {
955 PROC_UNLOCK(p);
956 return (0);
957 }
958
959 PROC_SLOCK(p);
960
961 if (siginfo != NULL) {
962 bzero(siginfo, sizeof(*siginfo));
963 siginfo->si_errno = 0;
964
965 /*
966 * SUSv4 requires that the si_signo value is always
967 * SIGCHLD. Obey it despite the rfork(2) interface
968 * allows to request other signal for child exit
969 * notification.
970 */
971 siginfo->si_signo = SIGCHLD;
972
973 /*
974 * This is still a rough estimate. We will fix the
975 * cases TRAPPED, STOPPED, and CONTINUED later.
976 */
977 if (WCOREDUMP(p->p_xstat))
978 siginfo->si_code = CLD_DUMPED;
979 else if (WIFSIGNALED(p->p_xstat))
980 siginfo->si_code = CLD_KILLED;
981 else
982 siginfo->si_code = CLD_EXITED;
983
984 siginfo->si_pid = p->p_pid;
985 siginfo->si_uid = p->p_ucred->cr_uid;
986 siginfo->si_status = p->p_xstat;
987
988 /*
989 * The si_addr field would be useful additional
990 * detail, but apparently the PC value may be lost
991 * when we reach this point. bzero() above sets
992 * siginfo->si_addr to NULL.
993 */
994 }
995
996 /*
997 * There should be no reason to limit resources usage info to
998 * exited processes only. A snapshot about any resources used
999 * by a stopped process may be exactly what is needed.
1000 */
1001 if (wrusage != NULL) {
1002 rup = &wrusage->wru_self;
1003 *rup = p->p_ru;
1004 calcru(p, &rup->ru_utime, &rup->ru_stime);
1005
1006 rup = &wrusage->wru_children;
1007 *rup = p->p_stats->p_cru;
1008 calccru(p, &rup->ru_utime, &rup->ru_stime);
1009 }
1010
1011 if (p->p_state == PRS_ZOMBIE) {
1012 proc_reap(td, p, status, options);
1013 return (-1);
1014 }
1015 PROC_SUNLOCK(p);
1016 PROC_UNLOCK(p);
1017 return (1);
1018 }
1019
1020 int
1021 kern_wait(struct thread *td, pid_t pid, int *status, int options,
1022 struct rusage *rusage)
1023 {
1024 struct __wrusage wru, *wrup;
1025 idtype_t idtype;
1026 id_t id;
1027 int ret;
1028
1029 /*
1030 * Translate the special pid values into the (idtype, pid)
1031 * pair for kern_wait6. The WAIT_MYPGRP case is handled by
1032 * kern_wait6() on its own.
1033 */
1034 if (pid == WAIT_ANY) {
1035 idtype = P_ALL;
1036 id = 0;
1037 } else if (pid < 0) {
1038 idtype = P_PGID;
1039 id = (id_t)-pid;
1040 } else {
1041 idtype = P_PID;
1042 id = (id_t)pid;
1043 }
1044
1045 if (rusage != NULL)
1046 wrup = &wru;
1047 else
1048 wrup = NULL;
1049
1050 /*
1051 * For backward compatibility we implicitly add flags WEXITED
1052 * and WTRAPPED here.
1053 */
1054 options |= WEXITED | WTRAPPED;
1055 ret = kern_wait6(td, idtype, id, status, options, wrup, NULL);
1056 if (rusage != NULL)
1057 *rusage = wru.wru_self;
1058 return (ret);
1059 }
1060
1061 int
1062 kern_wait6(struct thread *td, idtype_t idtype, id_t id, int *status,
1063 int options, struct __wrusage *wrusage, siginfo_t *siginfo)
1064 {
1065 struct proc *p, *q;
1066 int error, nfound, ret;
1067
1068 AUDIT_ARG_VALUE((int)idtype); /* XXX - This is likely wrong! */
1069 AUDIT_ARG_PID((pid_t)id); /* XXX - This may be wrong! */
1070 AUDIT_ARG_VALUE(options);
1071
1072 q = td->td_proc;
1073
1074 if ((pid_t)id == WAIT_MYPGRP && (idtype == P_PID || idtype == P_PGID)) {
1075 PROC_LOCK(q);
1076 id = (id_t)q->p_pgid;
1077 PROC_UNLOCK(q);
1078 idtype = P_PGID;
1079 }
1080
1081 /* If we don't know the option, just return. */
1082 if ((options & ~(WUNTRACED | WNOHANG | WCONTINUED | WNOWAIT |
1083 WEXITED | WTRAPPED | WLINUXCLONE)) != 0)
1084 return (EINVAL);
1085 if ((options & (WEXITED | WUNTRACED | WCONTINUED | WTRAPPED)) == 0) {
1086 /*
1087 * We will be unable to find any matching processes,
1088 * because there are no known events to look for.
1089 * Prefer to return error instead of blocking
1090 * indefinitely.
1091 */
1092 return (EINVAL);
1093 }
1094
1095 loop:
1096 if (q->p_flag & P_STATCHILD) {
1097 PROC_LOCK(q);
1098 q->p_flag &= ~P_STATCHILD;
1099 PROC_UNLOCK(q);
1100 }
1101 nfound = 0;
1102 sx_xlock(&proctree_lock);
1103 LIST_FOREACH(p, &q->p_children, p_sibling) {
1104 ret = proc_to_reap(td, p, idtype, id, status, options,
1105 wrusage, siginfo);
1106 if (ret == 0)
1107 continue;
1108 else if (ret == 1)
1109 nfound++;
1110 else
1111 return (0);
1112
1113 PROC_LOCK(p);
1114 PROC_SLOCK(p);
1115
1116 if ((options & WTRAPPED) != 0 &&
1117 (p->p_flag & P_TRACED) != 0 &&
1118 (p->p_flag & (P_STOPPED_TRACE | P_STOPPED_SIG)) != 0 &&
1119 (p->p_suspcount == p->p_numthreads) &&
1120 ((p->p_flag & P_WAITED) == 0)) {
1121 PROC_SUNLOCK(p);
1122 if ((options & WNOWAIT) == 0)
1123 p->p_flag |= P_WAITED;
1124 sx_xunlock(&proctree_lock);
1125 td->td_retval[0] = p->p_pid;
1126
1127 if (status != NULL)
1128 *status = W_STOPCODE(p->p_xstat);
1129 if (siginfo != NULL) {
1130 siginfo->si_status = p->p_xstat;
1131 siginfo->si_code = CLD_TRAPPED;
1132 }
1133 if ((options & WNOWAIT) == 0) {
1134 PROC_LOCK(q);
1135 sigqueue_take(p->p_ksi);
1136 PROC_UNLOCK(q);
1137 }
1138
1139 PROC_UNLOCK(p);
1140 return (0);
1141 }
1142 if ((options & WUNTRACED) != 0 &&
1143 (p->p_flag & P_STOPPED_SIG) != 0 &&
1144 (p->p_suspcount == p->p_numthreads) &&
1145 ((p->p_flag & P_WAITED) == 0)) {
1146 PROC_SUNLOCK(p);
1147 if ((options & WNOWAIT) == 0)
1148 p->p_flag |= P_WAITED;
1149 sx_xunlock(&proctree_lock);
1150 td->td_retval[0] = p->p_pid;
1151
1152 if (status != NULL)
1153 *status = W_STOPCODE(p->p_xstat);
1154 if (siginfo != NULL) {
1155 siginfo->si_status = p->p_xstat;
1156 siginfo->si_code = CLD_STOPPED;
1157 }
1158 if ((options & WNOWAIT) == 0) {
1159 PROC_LOCK(q);
1160 sigqueue_take(p->p_ksi);
1161 PROC_UNLOCK(q);
1162 }
1163
1164 PROC_UNLOCK(p);
1165 return (0);
1166 }
1167 PROC_SUNLOCK(p);
1168 if ((options & WCONTINUED) != 0 &&
1169 (p->p_flag & P_CONTINUED) != 0) {
1170 sx_xunlock(&proctree_lock);
1171 td->td_retval[0] = p->p_pid;
1172 if ((options & WNOWAIT) == 0) {
1173 p->p_flag &= ~P_CONTINUED;
1174 PROC_LOCK(q);
1175 sigqueue_take(p->p_ksi);
1176 PROC_UNLOCK(q);
1177 }
1178 PROC_UNLOCK(p);
1179
1180 if (status != NULL)
1181 *status = SIGCONT;
1182 if (siginfo != NULL) {
1183 siginfo->si_status = SIGCONT;
1184 siginfo->si_code = CLD_CONTINUED;
1185 }
1186 return (0);
1187 }
1188 PROC_UNLOCK(p);
1189 }
1190
1191 /*
1192 * Look in the orphans list too, to allow the parent to
1193 * collect it's child exit status even if child is being
1194 * debugged.
1195 *
1196 * Debugger detaches from the parent upon successful
1197 * switch-over from parent to child. At this point due to
1198 * re-parenting the parent loses the child to debugger and a
1199 * wait4(2) call would report that it has no children to wait
1200 * for. By maintaining a list of orphans we allow the parent
1201 * to successfully wait until the child becomes a zombie.
1202 */
1203 LIST_FOREACH(p, &q->p_orphans, p_orphan) {
1204 ret = proc_to_reap(td, p, idtype, id, status, options,
1205 wrusage, siginfo);
1206 if (ret == 0)
1207 continue;
1208 else if (ret == 1)
1209 nfound++;
1210 else
1211 return (0);
1212 }
1213 if (nfound == 0) {
1214 sx_xunlock(&proctree_lock);
1215 return (ECHILD);
1216 }
1217 if (options & WNOHANG) {
1218 sx_xunlock(&proctree_lock);
1219 td->td_retval[0] = 0;
1220 return (0);
1221 }
1222 PROC_LOCK(q);
1223 sx_xunlock(&proctree_lock);
1224 if (q->p_flag & P_STATCHILD) {
1225 q->p_flag &= ~P_STATCHILD;
1226 error = 0;
1227 } else
1228 error = msleep(q, &q->p_mtx, PWAIT | PCATCH, "wait", 0);
1229 PROC_UNLOCK(q);
1230 if (error)
1231 return (error);
1232 goto loop;
1233 }
1234
1235 /*
1236 * Make process 'parent' the new parent of process 'child'.
1237 * Must be called with an exclusive hold of proctree lock.
1238 */
1239 void
1240 proc_reparent(struct proc *child, struct proc *parent)
1241 {
1242
1243 sx_assert(&proctree_lock, SX_XLOCKED);
1244 PROC_LOCK_ASSERT(child, MA_OWNED);
1245 if (child->p_pptr == parent)
1246 return;
1247
1248 PROC_LOCK(child->p_pptr);
1249 sigqueue_take(child->p_ksi);
1250 PROC_UNLOCK(child->p_pptr);
1251 LIST_REMOVE(child, p_sibling);
1252 LIST_INSERT_HEAD(&parent->p_children, child, p_sibling);
1253
1254 clear_orphan(child);
1255 if (child->p_flag & P_TRACED) {
1256 LIST_INSERT_HEAD(&child->p_pptr->p_orphans, child, p_orphan);
1257 child->p_flag |= P_ORPHAN;
1258 }
1259
1260 child->p_pptr = parent;
1261 }
Cache object: 6b24bc15f720471890e675195219ba94
|