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/10.1/sys/kern/kern_exit.c 271423 2014-09-11 11:25:10Z kib $");
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, "int");
98
99 /* Hook for NFS teardown procedure. */
100 void (*nlminfo_release_p)(struct proc *p);
101
102 struct proc *
103 proc_realparent(struct proc *child)
104 {
105 struct proc *p, *parent;
106
107 sx_assert(&proctree_lock, SX_LOCKED);
108 if ((child->p_treeflag & P_TREE_ORPHANED) == 0) {
109 if (child->p_oppid == 0 ||
110 child->p_pptr->p_pid == child->p_oppid)
111 parent = child->p_pptr;
112 else
113 parent = initproc;
114 return (parent);
115 }
116 for (p = child; (p->p_treeflag & P_TREE_FIRST_ORPHAN) == 0;) {
117 /* Cannot use LIST_PREV(), since the list head is not known. */
118 p = __containerof(p->p_orphan.le_prev, struct proc,
119 p_orphan.le_next);
120 KASSERT((p->p_treeflag & P_TREE_ORPHANED) != 0,
121 ("missing P_ORPHAN %p", p));
122 }
123 parent = __containerof(p->p_orphan.le_prev, struct proc,
124 p_orphans.lh_first);
125 return (parent);
126 }
127
128 static void
129 clear_orphan(struct proc *p)
130 {
131 struct proc *p1;
132
133 sx_assert(&proctree_lock, SA_XLOCKED);
134 if ((p->p_treeflag & P_TREE_ORPHANED) == 0)
135 return;
136 if ((p->p_treeflag & P_TREE_FIRST_ORPHAN) != 0) {
137 p1 = LIST_NEXT(p, p_orphan);
138 if (p1 != NULL)
139 p1->p_treeflag |= P_TREE_FIRST_ORPHAN;
140 p->p_treeflag &= ~P_TREE_FIRST_ORPHAN;
141 }
142 LIST_REMOVE(p, p_orphan);
143 p->p_treeflag &= ~P_TREE_ORPHANED;
144 }
145
146 /*
147 * exit -- death of process.
148 */
149 void
150 sys_sys_exit(struct thread *td, struct sys_exit_args *uap)
151 {
152
153 exit1(td, W_EXITCODE(uap->rval, 0));
154 /* NOTREACHED */
155 }
156
157 /*
158 * Exit: deallocate address space and other resources, change proc state to
159 * zombie, and unlink proc from allproc and parent's lists. Save exit status
160 * and rusage for wait(). Check for child processes and orphan them.
161 */
162 void
163 exit1(struct thread *td, int rv)
164 {
165 struct proc *p, *nq, *q;
166 struct vnode *ttyvp = NULL;
167
168 mtx_assert(&Giant, MA_NOTOWNED);
169
170 p = td->td_proc;
171 /*
172 * XXX in case we're rebooting we just let init die in order to
173 * work around an unsolved stack overflow seen very late during
174 * shutdown on sparc64 when the gmirror worker process exists.
175 */
176 if (p == initproc && rebooting == 0) {
177 printf("init died (signal %d, exit %d)\n",
178 WTERMSIG(rv), WEXITSTATUS(rv));
179 panic("Going nowhere without my init!");
180 }
181
182 /*
183 * MUST abort all other threads before proceeding past here.
184 */
185 PROC_LOCK(p);
186 while (p->p_flag & P_HADTHREADS) {
187 /*
188 * First check if some other thread got here before us.
189 * If so, act appropriately: exit or suspend.
190 */
191 thread_suspend_check(0);
192
193 /*
194 * Kill off the other threads. This requires
195 * some co-operation from other parts of the kernel
196 * so it may not be instantaneous. With this state set
197 * any thread entering the kernel from userspace will
198 * thread_exit() in trap(). Any thread attempting to
199 * sleep will return immediately with EINTR or EWOULDBLOCK
200 * which will hopefully force them to back out to userland
201 * freeing resources as they go. Any thread attempting
202 * to return to userland will thread_exit() from userret().
203 * thread_exit() will unsuspend us when the last of the
204 * other threads exits.
205 * If there is already a thread singler after resumption,
206 * calling thread_single will fail; in that case, we just
207 * re-check all suspension request, the thread should
208 * either be suspended there or exit.
209 */
210 if (!thread_single(SINGLE_EXIT))
211 break;
212
213 /*
214 * All other activity in this process is now stopped.
215 * Threading support has been turned off.
216 */
217 }
218 KASSERT(p->p_numthreads == 1,
219 ("exit1: proc %p exiting with %d threads", p, p->p_numthreads));
220 racct_sub(p, RACCT_NTHR, 1);
221 /*
222 * Wakeup anyone in procfs' PIOCWAIT. They should have a hold
223 * on our vmspace, so we should block below until they have
224 * released their reference to us. Note that if they have
225 * requested S_EXIT stops we will block here until they ack
226 * via PIOCCONT.
227 */
228 _STOPEVENT(p, S_EXIT, rv);
229
230 /*
231 * Ignore any pending request to stop due to a stop signal.
232 * Once P_WEXIT is set, future requests will be ignored as
233 * well.
234 */
235 p->p_flag &= ~P_STOPPED_SIG;
236 KASSERT(!P_SHOULDSTOP(p), ("exiting process is stopped"));
237
238 /*
239 * Note that we are exiting and do another wakeup of anyone in
240 * PIOCWAIT in case they aren't listening for S_EXIT stops or
241 * decided to wait again after we told them we are exiting.
242 */
243 p->p_flag |= P_WEXIT;
244 wakeup(&p->p_stype);
245
246 /*
247 * Wait for any processes that have a hold on our vmspace to
248 * release their reference.
249 */
250 while (p->p_lock > 0)
251 msleep(&p->p_lock, &p->p_mtx, PWAIT, "exithold", 0);
252
253 p->p_xstat = rv; /* Let event handler change exit status */
254 PROC_UNLOCK(p);
255 /* Drain the limit callout while we don't have the proc locked */
256 callout_drain(&p->p_limco);
257
258 #ifdef AUDIT
259 /*
260 * The Sun BSM exit token contains two components: an exit status as
261 * passed to exit(), and a return value to indicate what sort of exit
262 * it was. The exit status is WEXITSTATUS(rv), but it's not clear
263 * what the return value is.
264 */
265 AUDIT_ARG_EXIT(WEXITSTATUS(rv), 0);
266 AUDIT_SYSCALL_EXIT(0, td);
267 #endif
268
269 /* Are we a task leader? */
270 if (p == p->p_leader) {
271 mtx_lock(&ppeers_lock);
272 q = p->p_peers;
273 while (q != NULL) {
274 PROC_LOCK(q);
275 kern_psignal(q, SIGKILL);
276 PROC_UNLOCK(q);
277 q = q->p_peers;
278 }
279 while (p->p_peers != NULL)
280 msleep(p, &ppeers_lock, PWAIT, "exit1", 0);
281 mtx_unlock(&ppeers_lock);
282 }
283
284 /*
285 * Check if any loadable modules need anything done at process exit.
286 * E.g. SYSV IPC stuff
287 * XXX what if one of these generates an error?
288 */
289 EVENTHANDLER_INVOKE(process_exit, p);
290
291 /*
292 * If parent is waiting for us to exit or exec,
293 * P_PPWAIT is set; we will wakeup the parent below.
294 */
295 PROC_LOCK(p);
296 rv = p->p_xstat; /* Event handler could change exit status */
297 stopprofclock(p);
298 p->p_flag &= ~(P_TRACED | P_PPWAIT | P_PPTRACE);
299
300 /*
301 * Stop the real interval timer. If the handler is currently
302 * executing, prevent it from rearming itself and let it finish.
303 */
304 if (timevalisset(&p->p_realtimer.it_value) &&
305 callout_stop(&p->p_itcallout) == 0) {
306 timevalclear(&p->p_realtimer.it_interval);
307 msleep(&p->p_itcallout, &p->p_mtx, PWAIT, "ritwait", 0);
308 KASSERT(!timevalisset(&p->p_realtimer.it_value),
309 ("realtime timer is still armed"));
310 }
311 PROC_UNLOCK(p);
312
313 /*
314 * Reset any sigio structures pointing to us as a result of
315 * F_SETOWN with our pid.
316 */
317 funsetownlst(&p->p_sigiolst);
318
319 /*
320 * If this process has an nlminfo data area (for lockd), release it
321 */
322 if (nlminfo_release_p != NULL && p->p_nlminfo != NULL)
323 (*nlminfo_release_p)(p);
324
325 /*
326 * Close open files and release open-file table.
327 * This may block!
328 */
329 fdescfree(td);
330
331 /*
332 * If this thread tickled GEOM, we need to wait for the giggling to
333 * stop before we return to userland
334 */
335 if (td->td_pflags & TDP_GEOM)
336 g_waitidle();
337
338 /*
339 * Remove ourself from our leader's peer list and wake our leader.
340 */
341 mtx_lock(&ppeers_lock);
342 if (p->p_leader->p_peers) {
343 q = p->p_leader;
344 while (q->p_peers != p)
345 q = q->p_peers;
346 q->p_peers = p->p_peers;
347 wakeup(p->p_leader);
348 }
349 mtx_unlock(&ppeers_lock);
350
351 vmspace_exit(td);
352
353 sx_xlock(&proctree_lock);
354 if (SESS_LEADER(p)) {
355 struct session *sp = p->p_session;
356 struct tty *tp;
357
358 /*
359 * s_ttyp is not zero'd; we use this to indicate that
360 * the session once had a controlling terminal. (for
361 * logging and informational purposes)
362 */
363 SESS_LOCK(sp);
364 ttyvp = sp->s_ttyvp;
365 tp = sp->s_ttyp;
366 sp->s_ttyvp = NULL;
367 sp->s_ttydp = NULL;
368 sp->s_leader = NULL;
369 SESS_UNLOCK(sp);
370
371 /*
372 * Signal foreground pgrp and revoke access to
373 * controlling terminal if it has not been revoked
374 * already.
375 *
376 * Because the TTY may have been revoked in the mean
377 * time and could already have a new session associated
378 * with it, make sure we don't send a SIGHUP to a
379 * foreground process group that does not belong to this
380 * session.
381 */
382
383 if (tp != NULL) {
384 tty_lock(tp);
385 if (tp->t_session == sp)
386 tty_signal_pgrp(tp, SIGHUP);
387 tty_unlock(tp);
388 }
389
390 if (ttyvp != NULL) {
391 sx_xunlock(&proctree_lock);
392 if (vn_lock(ttyvp, LK_EXCLUSIVE) == 0) {
393 VOP_REVOKE(ttyvp, REVOKEALL);
394 VOP_UNLOCK(ttyvp, 0);
395 }
396 sx_xlock(&proctree_lock);
397 }
398 }
399 fixjobc(p, p->p_pgrp, 0);
400 sx_xunlock(&proctree_lock);
401 (void)acct_process(td);
402
403 /* Release the TTY now we've unlocked everything. */
404 if (ttyvp != NULL)
405 vrele(ttyvp);
406 #ifdef KTRACE
407 ktrprocexit(td);
408 #endif
409 /*
410 * Release reference to text vnode
411 */
412 if (p->p_textvp != NULL) {
413 vrele(p->p_textvp);
414 p->p_textvp = NULL;
415 }
416
417 /*
418 * Release our limits structure.
419 */
420 lim_free(p->p_limit);
421 p->p_limit = NULL;
422
423 tidhash_remove(td);
424
425 /*
426 * Remove proc from allproc queue and pidhash chain.
427 * Place onto zombproc. Unlink from parent's child list.
428 */
429 sx_xlock(&allproc_lock);
430 LIST_REMOVE(p, p_list);
431 LIST_INSERT_HEAD(&zombproc, p, p_list);
432 LIST_REMOVE(p, p_hash);
433 sx_xunlock(&allproc_lock);
434
435 /*
436 * Call machine-dependent code to release any
437 * machine-dependent resources other than the address space.
438 * The address space is released by "vmspace_exitfree(p)" in
439 * vm_waitproc().
440 */
441 cpu_exit(td);
442
443 WITNESS_WARN(WARN_PANIC, NULL, "process (pid %d) exiting", p->p_pid);
444
445 /*
446 * Reparent all of our children to init.
447 */
448 sx_xlock(&proctree_lock);
449 q = LIST_FIRST(&p->p_children);
450 if (q != NULL) /* only need this if any child is S_ZOMB */
451 wakeup(initproc);
452 for (; q != NULL; q = nq) {
453 nq = LIST_NEXT(q, p_sibling);
454 PROC_LOCK(q);
455 proc_reparent(q, initproc);
456 q->p_sigparent = SIGCHLD;
457 /*
458 * Traced processes are killed
459 * since their existence means someone is screwing up.
460 */
461 if (q->p_flag & P_TRACED) {
462 struct thread *temp;
463
464 /*
465 * Since q was found on our children list, the
466 * proc_reparent() call moved q to the orphan
467 * list due to present P_TRACED flag. Clear
468 * orphan link for q now while q is locked.
469 */
470 clear_orphan(q);
471 q->p_flag &= ~(P_TRACED | P_STOPPED_TRACE);
472 FOREACH_THREAD_IN_PROC(q, temp)
473 temp->td_dbgflags &= ~TDB_SUSPEND;
474 kern_psignal(q, SIGKILL);
475 }
476 PROC_UNLOCK(q);
477 }
478
479 /*
480 * Also get rid of our orphans.
481 */
482 while ((q = LIST_FIRST(&p->p_orphans)) != NULL) {
483 PROC_LOCK(q);
484 clear_orphan(q);
485 PROC_UNLOCK(q);
486 }
487
488 /* Save exit status. */
489 PROC_LOCK(p);
490 p->p_xthread = td;
491
492 /* Tell the prison that we are gone. */
493 prison_proc_free(p->p_ucred->cr_prison);
494
495 #ifdef KDTRACE_HOOKS
496 /*
497 * Tell the DTrace fasttrap provider about the exit if it
498 * has declared an interest.
499 */
500 if (dtrace_fasttrap_exit)
501 dtrace_fasttrap_exit(p);
502 #endif
503
504 /*
505 * Notify interested parties of our demise.
506 */
507 KNOTE_LOCKED(&p->p_klist, NOTE_EXIT);
508
509 #ifdef KDTRACE_HOOKS
510 int reason = CLD_EXITED;
511 if (WCOREDUMP(rv))
512 reason = CLD_DUMPED;
513 else if (WIFSIGNALED(rv))
514 reason = CLD_KILLED;
515 SDT_PROBE(proc, kernel, , exit, reason, 0, 0, 0, 0);
516 #endif
517
518 /*
519 * Just delete all entries in the p_klist. At this point we won't
520 * report any more events, and there are nasty race conditions that
521 * can beat us if we don't.
522 */
523 knlist_clear(&p->p_klist, 1);
524
525 /*
526 * If this is a process with a descriptor, we may not need to deliver
527 * a signal to the parent. proctree_lock is held over
528 * procdesc_exit() to serialize concurrent calls to close() and
529 * exit().
530 */
531 #ifdef PROCDESC
532 if (p->p_procdesc == NULL || procdesc_exit(p)) {
533 #endif
534 /*
535 * Notify parent that we're gone. If parent has the
536 * PS_NOCLDWAIT flag set, or if the handler is set to SIG_IGN,
537 * notify process 1 instead (and hope it will handle this
538 * situation).
539 */
540 PROC_LOCK(p->p_pptr);
541 mtx_lock(&p->p_pptr->p_sigacts->ps_mtx);
542 if (p->p_pptr->p_sigacts->ps_flag &
543 (PS_NOCLDWAIT | PS_CLDSIGIGN)) {
544 struct proc *pp;
545
546 mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx);
547 pp = p->p_pptr;
548 PROC_UNLOCK(pp);
549 proc_reparent(p, initproc);
550 p->p_sigparent = SIGCHLD;
551 PROC_LOCK(p->p_pptr);
552
553 /*
554 * Notify parent, so in case he was wait(2)ing or
555 * executing waitpid(2) with our pid, he will
556 * continue.
557 */
558 wakeup(pp);
559 } else
560 mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx);
561
562 if (p->p_pptr == initproc)
563 kern_psignal(p->p_pptr, SIGCHLD);
564 else if (p->p_sigparent != 0) {
565 if (p->p_sigparent == SIGCHLD)
566 childproc_exited(p);
567 else /* LINUX thread */
568 kern_psignal(p->p_pptr, p->p_sigparent);
569 }
570 #ifdef PROCDESC
571 } else
572 PROC_LOCK(p->p_pptr);
573 #endif
574 sx_xunlock(&proctree_lock);
575
576 /*
577 * The state PRS_ZOMBIE prevents other proesses from sending
578 * signal to the process, to avoid memory leak, we free memory
579 * for signal queue at the time when the state is set.
580 */
581 sigqueue_flush(&p->p_sigqueue);
582 sigqueue_flush(&td->td_sigqueue);
583
584 /*
585 * We have to wait until after acquiring all locks before
586 * changing p_state. We need to avoid all possible context
587 * switches (including ones from blocking on a mutex) while
588 * marked as a zombie. We also have to set the zombie state
589 * before we release the parent process' proc lock to avoid
590 * a lost wakeup. So, we first call wakeup, then we grab the
591 * sched lock, update the state, and release the parent process'
592 * proc lock.
593 */
594 wakeup(p->p_pptr);
595 cv_broadcast(&p->p_pwait);
596 sched_exit(p->p_pptr, td);
597 PROC_SLOCK(p);
598 p->p_state = PRS_ZOMBIE;
599 PROC_UNLOCK(p->p_pptr);
600
601 /*
602 * Hopefully no one will try to deliver a signal to the process this
603 * late in the game.
604 */
605 knlist_destroy(&p->p_klist);
606
607 /*
608 * Save our children's rusage information in our exit rusage.
609 */
610 ruadd(&p->p_ru, &p->p_rux, &p->p_stats->p_cru, &p->p_crux);
611
612 /*
613 * Make sure the scheduler takes this thread out of its tables etc.
614 * This will also release this thread's reference to the ucred.
615 * Other thread parts to release include pcb bits and such.
616 */
617 thread_exit();
618 }
619
620
621 #ifndef _SYS_SYSPROTO_H_
622 struct abort2_args {
623 char *why;
624 int nargs;
625 void **args;
626 };
627 #endif
628
629 int
630 sys_abort2(struct thread *td, struct abort2_args *uap)
631 {
632 struct proc *p = td->td_proc;
633 struct sbuf *sb;
634 void *uargs[16];
635 int error, i, sig;
636
637 /*
638 * Do it right now so we can log either proper call of abort2(), or
639 * note, that invalid argument was passed. 512 is big enough to
640 * handle 16 arguments' descriptions with additional comments.
641 */
642 sb = sbuf_new(NULL, NULL, 512, SBUF_FIXEDLEN);
643 sbuf_clear(sb);
644 sbuf_printf(sb, "%s(pid %d uid %d) aborted: ",
645 p->p_comm, p->p_pid, td->td_ucred->cr_uid);
646 /*
647 * Since we can't return from abort2(), send SIGKILL in cases, where
648 * abort2() was called improperly
649 */
650 sig = SIGKILL;
651 /* Prevent from DoSes from user-space. */
652 if (uap->nargs < 0 || uap->nargs > 16)
653 goto out;
654 if (uap->nargs > 0) {
655 if (uap->args == NULL)
656 goto out;
657 error = copyin(uap->args, uargs, uap->nargs * sizeof(void *));
658 if (error != 0)
659 goto out;
660 }
661 /*
662 * Limit size of 'reason' string to 128. Will fit even when
663 * maximal number of arguments was chosen to be logged.
664 */
665 if (uap->why != NULL) {
666 error = sbuf_copyin(sb, uap->why, 128);
667 if (error < 0)
668 goto out;
669 } else {
670 sbuf_printf(sb, "(null)");
671 }
672 if (uap->nargs > 0) {
673 sbuf_printf(sb, "(");
674 for (i = 0;i < uap->nargs; i++)
675 sbuf_printf(sb, "%s%p", i == 0 ? "" : ", ", uargs[i]);
676 sbuf_printf(sb, ")");
677 }
678 /*
679 * Final stage: arguments were proper, string has been
680 * successfully copied from userspace, and copying pointers
681 * from user-space succeed.
682 */
683 sig = SIGABRT;
684 out:
685 if (sig == SIGKILL) {
686 sbuf_trim(sb);
687 sbuf_printf(sb, " (Reason text inaccessible)");
688 }
689 sbuf_cat(sb, "\n");
690 sbuf_finish(sb);
691 log(LOG_INFO, "%s", sbuf_data(sb));
692 sbuf_delete(sb);
693 exit1(td, W_EXITCODE(0, sig));
694 return (0);
695 }
696
697
698 #ifdef COMPAT_43
699 /*
700 * The dirty work is handled by kern_wait().
701 */
702 int
703 owait(struct thread *td, struct owait_args *uap __unused)
704 {
705 int error, status;
706
707 error = kern_wait(td, WAIT_ANY, &status, 0, NULL);
708 if (error == 0)
709 td->td_retval[1] = status;
710 return (error);
711 }
712 #endif /* COMPAT_43 */
713
714 /*
715 * The dirty work is handled by kern_wait().
716 */
717 int
718 sys_wait4(struct thread *td, struct wait4_args *uap)
719 {
720 struct rusage ru, *rup;
721 int error, status;
722
723 if (uap->rusage != NULL)
724 rup = &ru;
725 else
726 rup = NULL;
727 error = kern_wait(td, uap->pid, &status, uap->options, rup);
728 if (uap->status != NULL && error == 0)
729 error = copyout(&status, uap->status, sizeof(status));
730 if (uap->rusage != NULL && error == 0)
731 error = copyout(&ru, uap->rusage, sizeof(struct rusage));
732 return (error);
733 }
734
735 int
736 sys_wait6(struct thread *td, struct wait6_args *uap)
737 {
738 struct __wrusage wru, *wrup;
739 siginfo_t si, *sip;
740 idtype_t idtype;
741 id_t id;
742 int error, status;
743
744 idtype = uap->idtype;
745 id = uap->id;
746
747 if (uap->wrusage != NULL)
748 wrup = &wru;
749 else
750 wrup = NULL;
751
752 if (uap->info != NULL) {
753 sip = &si;
754 bzero(sip, sizeof(*sip));
755 } else
756 sip = NULL;
757
758 /*
759 * We expect all callers of wait6() to know about WEXITED and
760 * WTRAPPED.
761 */
762 error = kern_wait6(td, idtype, id, &status, uap->options, wrup, sip);
763
764 if (uap->status != NULL && error == 0)
765 error = copyout(&status, uap->status, sizeof(status));
766 if (uap->wrusage != NULL && error == 0)
767 error = copyout(&wru, uap->wrusage, sizeof(wru));
768 if (uap->info != NULL && error == 0)
769 error = copyout(&si, uap->info, sizeof(si));
770 return (error);
771 }
772
773 /*
774 * Reap the remains of a zombie process and optionally return status and
775 * rusage. Asserts and will release both the proctree_lock and the process
776 * lock as part of its work.
777 */
778 void
779 proc_reap(struct thread *td, struct proc *p, int *status, int options)
780 {
781 struct proc *q, *t;
782
783 sx_assert(&proctree_lock, SA_XLOCKED);
784 PROC_LOCK_ASSERT(p, MA_OWNED);
785 PROC_SLOCK_ASSERT(p, MA_OWNED);
786 KASSERT(p->p_state == PRS_ZOMBIE, ("proc_reap: !PRS_ZOMBIE"));
787
788 q = td->td_proc;
789
790 PROC_SUNLOCK(p);
791 td->td_retval[0] = p->p_pid;
792 if (status)
793 *status = p->p_xstat; /* convert to int */
794 if (options & WNOWAIT) {
795 /*
796 * Only poll, returning the status. Caller does not wish to
797 * release the proc struct just yet.
798 */
799 PROC_UNLOCK(p);
800 sx_xunlock(&proctree_lock);
801 return;
802 }
803
804 PROC_LOCK(q);
805 sigqueue_take(p->p_ksi);
806 PROC_UNLOCK(q);
807 PROC_UNLOCK(p);
808
809 /*
810 * If we got the child via a ptrace 'attach', we need to give it back
811 * to the old parent.
812 */
813 if (p->p_oppid != 0) {
814 t = proc_realparent(p);
815 PROC_LOCK(t);
816 PROC_LOCK(p);
817 proc_reparent(p, t);
818 p->p_oppid = 0;
819 PROC_UNLOCK(p);
820 pksignal(t, SIGCHLD, p->p_ksi);
821 wakeup(t);
822 cv_broadcast(&p->p_pwait);
823 PROC_UNLOCK(t);
824 sx_xunlock(&proctree_lock);
825 return;
826 }
827
828 /*
829 * Remove other references to this process to ensure we have an
830 * exclusive reference.
831 */
832 sx_xlock(&allproc_lock);
833 LIST_REMOVE(p, p_list); /* off zombproc */
834 sx_xunlock(&allproc_lock);
835 LIST_REMOVE(p, p_sibling);
836 PROC_LOCK(p);
837 clear_orphan(p);
838 PROC_UNLOCK(p);
839 leavepgrp(p);
840 #ifdef PROCDESC
841 if (p->p_procdesc != NULL)
842 procdesc_reap(p);
843 #endif
844 sx_xunlock(&proctree_lock);
845
846 /*
847 * As a side effect of this lock, we know that all other writes to
848 * this proc are visible now, so no more locking is needed for p.
849 */
850 PROC_LOCK(p);
851 p->p_xstat = 0; /* XXX: why? */
852 PROC_UNLOCK(p);
853 PROC_LOCK(q);
854 ruadd(&q->p_stats->p_cru, &q->p_crux, &p->p_ru, &p->p_rux);
855 PROC_UNLOCK(q);
856
857 /*
858 * Decrement the count of procs running with this uid.
859 */
860 (void)chgproccnt(p->p_ucred->cr_ruidinfo, -1, 0);
861
862 /*
863 * Destroy resource accounting information associated with the process.
864 */
865 #ifdef RACCT
866 PROC_LOCK(p);
867 racct_sub(p, RACCT_NPROC, 1);
868 PROC_UNLOCK(p);
869 #endif
870 racct_proc_exit(p);
871
872 /*
873 * Free credentials, arguments, and sigacts.
874 */
875 crfree(p->p_ucred);
876 p->p_ucred = NULL;
877 pargs_drop(p->p_args);
878 p->p_args = NULL;
879 sigacts_free(p->p_sigacts);
880 p->p_sigacts = NULL;
881
882 /*
883 * Do any thread-system specific cleanups.
884 */
885 thread_wait(p);
886
887 /*
888 * Give vm and machine-dependent layer a chance to free anything that
889 * cpu_exit couldn't release while still running in process context.
890 */
891 vm_waitproc(p);
892 #ifdef MAC
893 mac_proc_destroy(p);
894 #endif
895 KASSERT(FIRST_THREAD_IN_PROC(p),
896 ("proc_reap: no residual thread!"));
897 uma_zfree(proc_zone, p);
898 sx_xlock(&allproc_lock);
899 nprocs--;
900 sx_xunlock(&allproc_lock);
901 }
902
903 static int
904 proc_to_reap(struct thread *td, struct proc *p, idtype_t idtype, id_t id,
905 int *status, int options, struct __wrusage *wrusage, siginfo_t *siginfo)
906 {
907 struct proc *q;
908 struct rusage *rup;
909
910 sx_assert(&proctree_lock, SA_XLOCKED);
911
912 q = td->td_proc;
913 PROC_LOCK(p);
914
915 switch (idtype) {
916 case P_ALL:
917 break;
918 case P_PID:
919 if (p->p_pid != (pid_t)id) {
920 PROC_UNLOCK(p);
921 return (0);
922 }
923 break;
924 case P_PGID:
925 if (p->p_pgid != (pid_t)id) {
926 PROC_UNLOCK(p);
927 return (0);
928 }
929 break;
930 case P_SID:
931 if (p->p_session->s_sid != (pid_t)id) {
932 PROC_UNLOCK(p);
933 return (0);
934 }
935 break;
936 case P_UID:
937 if (p->p_ucred->cr_uid != (uid_t)id) {
938 PROC_UNLOCK(p);
939 return (0);
940 }
941 break;
942 case P_GID:
943 if (p->p_ucred->cr_gid != (gid_t)id) {
944 PROC_UNLOCK(p);
945 return (0);
946 }
947 break;
948 case P_JAILID:
949 if (p->p_ucred->cr_prison->pr_id != (int)id) {
950 PROC_UNLOCK(p);
951 return (0);
952 }
953 break;
954 /*
955 * It seems that the thread structures get zeroed out
956 * at process exit. This makes it impossible to
957 * support P_SETID, P_CID or P_CPUID.
958 */
959 default:
960 PROC_UNLOCK(p);
961 return (0);
962 }
963
964 if (p_canwait(td, p)) {
965 PROC_UNLOCK(p);
966 return (0);
967 }
968
969 if (((options & WEXITED) == 0) && (p->p_state == PRS_ZOMBIE)) {
970 PROC_UNLOCK(p);
971 return (0);
972 }
973
974 /*
975 * This special case handles a kthread spawned by linux_clone
976 * (see linux_misc.c). The linux_wait4 and linux_waitpid
977 * functions need to be able to distinguish between waiting
978 * on a process and waiting on a thread. It is a thread if
979 * p_sigparent is not SIGCHLD, and the WLINUXCLONE option
980 * signifies we want to wait for threads and not processes.
981 */
982 if ((p->p_sigparent != SIGCHLD) ^
983 ((options & WLINUXCLONE) != 0)) {
984 PROC_UNLOCK(p);
985 return (0);
986 }
987
988 PROC_SLOCK(p);
989
990 if (siginfo != NULL) {
991 bzero(siginfo, sizeof(*siginfo));
992 siginfo->si_errno = 0;
993
994 /*
995 * SUSv4 requires that the si_signo value is always
996 * SIGCHLD. Obey it despite the rfork(2) interface
997 * allows to request other signal for child exit
998 * notification.
999 */
1000 siginfo->si_signo = SIGCHLD;
1001
1002 /*
1003 * This is still a rough estimate. We will fix the
1004 * cases TRAPPED, STOPPED, and CONTINUED later.
1005 */
1006 if (WCOREDUMP(p->p_xstat)) {
1007 siginfo->si_code = CLD_DUMPED;
1008 siginfo->si_status = WTERMSIG(p->p_xstat);
1009 } else if (WIFSIGNALED(p->p_xstat)) {
1010 siginfo->si_code = CLD_KILLED;
1011 siginfo->si_status = WTERMSIG(p->p_xstat);
1012 } else {
1013 siginfo->si_code = CLD_EXITED;
1014 siginfo->si_status = WEXITSTATUS(p->p_xstat);
1015 }
1016
1017 siginfo->si_pid = p->p_pid;
1018 siginfo->si_uid = p->p_ucred->cr_uid;
1019
1020 /*
1021 * The si_addr field would be useful additional
1022 * detail, but apparently the PC value may be lost
1023 * when we reach this point. bzero() above sets
1024 * siginfo->si_addr to NULL.
1025 */
1026 }
1027
1028 /*
1029 * There should be no reason to limit resources usage info to
1030 * exited processes only. A snapshot about any resources used
1031 * by a stopped process may be exactly what is needed.
1032 */
1033 if (wrusage != NULL) {
1034 rup = &wrusage->wru_self;
1035 *rup = p->p_ru;
1036 calcru(p, &rup->ru_utime, &rup->ru_stime);
1037
1038 rup = &wrusage->wru_children;
1039 *rup = p->p_stats->p_cru;
1040 calccru(p, &rup->ru_utime, &rup->ru_stime);
1041 }
1042
1043 if (p->p_state == PRS_ZOMBIE) {
1044 proc_reap(td, p, status, options);
1045 return (-1);
1046 }
1047 PROC_SUNLOCK(p);
1048 PROC_UNLOCK(p);
1049 return (1);
1050 }
1051
1052 int
1053 kern_wait(struct thread *td, pid_t pid, int *status, int options,
1054 struct rusage *rusage)
1055 {
1056 struct __wrusage wru, *wrup;
1057 idtype_t idtype;
1058 id_t id;
1059 int ret;
1060
1061 /*
1062 * Translate the special pid values into the (idtype, pid)
1063 * pair for kern_wait6. The WAIT_MYPGRP case is handled by
1064 * kern_wait6() on its own.
1065 */
1066 if (pid == WAIT_ANY) {
1067 idtype = P_ALL;
1068 id = 0;
1069 } else if (pid < 0) {
1070 idtype = P_PGID;
1071 id = (id_t)-pid;
1072 } else {
1073 idtype = P_PID;
1074 id = (id_t)pid;
1075 }
1076
1077 if (rusage != NULL)
1078 wrup = &wru;
1079 else
1080 wrup = NULL;
1081
1082 /*
1083 * For backward compatibility we implicitly add flags WEXITED
1084 * and WTRAPPED here.
1085 */
1086 options |= WEXITED | WTRAPPED;
1087 ret = kern_wait6(td, idtype, id, status, options, wrup, NULL);
1088 if (rusage != NULL)
1089 *rusage = wru.wru_self;
1090 return (ret);
1091 }
1092
1093 int
1094 kern_wait6(struct thread *td, idtype_t idtype, id_t id, int *status,
1095 int options, struct __wrusage *wrusage, siginfo_t *siginfo)
1096 {
1097 struct proc *p, *q;
1098 int error, nfound, ret;
1099
1100 AUDIT_ARG_VALUE((int)idtype); /* XXX - This is likely wrong! */
1101 AUDIT_ARG_PID((pid_t)id); /* XXX - This may be wrong! */
1102 AUDIT_ARG_VALUE(options);
1103
1104 q = td->td_proc;
1105
1106 if ((pid_t)id == WAIT_MYPGRP && (idtype == P_PID || idtype == P_PGID)) {
1107 PROC_LOCK(q);
1108 id = (id_t)q->p_pgid;
1109 PROC_UNLOCK(q);
1110 idtype = P_PGID;
1111 }
1112
1113 /* If we don't know the option, just return. */
1114 if ((options & ~(WUNTRACED | WNOHANG | WCONTINUED | WNOWAIT |
1115 WEXITED | WTRAPPED | WLINUXCLONE)) != 0)
1116 return (EINVAL);
1117 if ((options & (WEXITED | WUNTRACED | WCONTINUED | WTRAPPED)) == 0) {
1118 /*
1119 * We will be unable to find any matching processes,
1120 * because there are no known events to look for.
1121 * Prefer to return error instead of blocking
1122 * indefinitely.
1123 */
1124 return (EINVAL);
1125 }
1126
1127 loop:
1128 if (q->p_flag & P_STATCHILD) {
1129 PROC_LOCK(q);
1130 q->p_flag &= ~P_STATCHILD;
1131 PROC_UNLOCK(q);
1132 }
1133 nfound = 0;
1134 sx_xlock(&proctree_lock);
1135 LIST_FOREACH(p, &q->p_children, p_sibling) {
1136 ret = proc_to_reap(td, p, idtype, id, status, options,
1137 wrusage, siginfo);
1138 if (ret == 0)
1139 continue;
1140 else if (ret == 1)
1141 nfound++;
1142 else
1143 return (0);
1144
1145 PROC_LOCK(p);
1146 PROC_SLOCK(p);
1147
1148 if ((options & WTRAPPED) != 0 &&
1149 (p->p_flag & P_TRACED) != 0 &&
1150 (p->p_flag & (P_STOPPED_TRACE | P_STOPPED_SIG)) != 0 &&
1151 (p->p_suspcount == p->p_numthreads) &&
1152 ((p->p_flag & P_WAITED) == 0)) {
1153 PROC_SUNLOCK(p);
1154 if ((options & WNOWAIT) == 0)
1155 p->p_flag |= P_WAITED;
1156 sx_xunlock(&proctree_lock);
1157 td->td_retval[0] = p->p_pid;
1158
1159 if (status != NULL)
1160 *status = W_STOPCODE(p->p_xstat);
1161 if (siginfo != NULL) {
1162 siginfo->si_status = p->p_xstat;
1163 siginfo->si_code = CLD_TRAPPED;
1164 }
1165 if ((options & WNOWAIT) == 0) {
1166 PROC_LOCK(q);
1167 sigqueue_take(p->p_ksi);
1168 PROC_UNLOCK(q);
1169 }
1170
1171 PROC_UNLOCK(p);
1172 return (0);
1173 }
1174 if ((options & WUNTRACED) != 0 &&
1175 (p->p_flag & P_STOPPED_SIG) != 0 &&
1176 (p->p_suspcount == p->p_numthreads) &&
1177 ((p->p_flag & P_WAITED) == 0)) {
1178 PROC_SUNLOCK(p);
1179 if ((options & WNOWAIT) == 0)
1180 p->p_flag |= P_WAITED;
1181 sx_xunlock(&proctree_lock);
1182 td->td_retval[0] = p->p_pid;
1183
1184 if (status != NULL)
1185 *status = W_STOPCODE(p->p_xstat);
1186 if (siginfo != NULL) {
1187 siginfo->si_status = p->p_xstat;
1188 siginfo->si_code = CLD_STOPPED;
1189 }
1190 if ((options & WNOWAIT) == 0) {
1191 PROC_LOCK(q);
1192 sigqueue_take(p->p_ksi);
1193 PROC_UNLOCK(q);
1194 }
1195
1196 PROC_UNLOCK(p);
1197 return (0);
1198 }
1199 PROC_SUNLOCK(p);
1200 if ((options & WCONTINUED) != 0 &&
1201 (p->p_flag & P_CONTINUED) != 0) {
1202 sx_xunlock(&proctree_lock);
1203 td->td_retval[0] = p->p_pid;
1204 if ((options & WNOWAIT) == 0) {
1205 p->p_flag &= ~P_CONTINUED;
1206 PROC_LOCK(q);
1207 sigqueue_take(p->p_ksi);
1208 PROC_UNLOCK(q);
1209 }
1210 PROC_UNLOCK(p);
1211
1212 if (status != NULL)
1213 *status = SIGCONT;
1214 if (siginfo != NULL) {
1215 siginfo->si_status = SIGCONT;
1216 siginfo->si_code = CLD_CONTINUED;
1217 }
1218 return (0);
1219 }
1220 PROC_UNLOCK(p);
1221 }
1222
1223 /*
1224 * Look in the orphans list too, to allow the parent to
1225 * collect it's child exit status even if child is being
1226 * debugged.
1227 *
1228 * Debugger detaches from the parent upon successful
1229 * switch-over from parent to child. At this point due to
1230 * re-parenting the parent loses the child to debugger and a
1231 * wait4(2) call would report that it has no children to wait
1232 * for. By maintaining a list of orphans we allow the parent
1233 * to successfully wait until the child becomes a zombie.
1234 */
1235 LIST_FOREACH(p, &q->p_orphans, p_orphan) {
1236 ret = proc_to_reap(td, p, idtype, id, status, options,
1237 wrusage, siginfo);
1238 if (ret == 0)
1239 continue;
1240 else if (ret == 1)
1241 nfound++;
1242 else
1243 return (0);
1244 }
1245 if (nfound == 0) {
1246 sx_xunlock(&proctree_lock);
1247 return (ECHILD);
1248 }
1249 if (options & WNOHANG) {
1250 sx_xunlock(&proctree_lock);
1251 td->td_retval[0] = 0;
1252 return (0);
1253 }
1254 PROC_LOCK(q);
1255 sx_xunlock(&proctree_lock);
1256 if (q->p_flag & P_STATCHILD) {
1257 q->p_flag &= ~P_STATCHILD;
1258 error = 0;
1259 } else
1260 error = msleep(q, &q->p_mtx, PWAIT | PCATCH, "wait", 0);
1261 PROC_UNLOCK(q);
1262 if (error)
1263 return (error);
1264 goto loop;
1265 }
1266
1267 /*
1268 * Make process 'parent' the new parent of process 'child'.
1269 * Must be called with an exclusive hold of proctree lock.
1270 */
1271 void
1272 proc_reparent(struct proc *child, struct proc *parent)
1273 {
1274
1275 sx_assert(&proctree_lock, SX_XLOCKED);
1276 PROC_LOCK_ASSERT(child, MA_OWNED);
1277 if (child->p_pptr == parent)
1278 return;
1279
1280 PROC_LOCK(child->p_pptr);
1281 sigqueue_take(child->p_ksi);
1282 PROC_UNLOCK(child->p_pptr);
1283 LIST_REMOVE(child, p_sibling);
1284 LIST_INSERT_HEAD(&parent->p_children, child, p_sibling);
1285
1286 clear_orphan(child);
1287 if (child->p_flag & P_TRACED) {
1288 if (LIST_EMPTY(&child->p_pptr->p_orphans)) {
1289 child->p_treeflag |= P_TREE_FIRST_ORPHAN;
1290 LIST_INSERT_HEAD(&child->p_pptr->p_orphans, child,
1291 p_orphan);
1292 } else {
1293 LIST_INSERT_AFTER(LIST_FIRST(&child->p_pptr->p_orphans),
1294 child, p_orphan);
1295 }
1296 child->p_treeflag |= P_TREE_ORPHANED;
1297 }
1298
1299 child->p_pptr = parent;
1300 }
Cache object: 0308017bd7a2aa8a2685f3ae5f8e1de9
|