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/10.3/sys/kern/kern_exit.c 293473 2016-01-09 14:08:10Z dchagin $");
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/capsicum.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 #include <sys/umtx.h>
77 #ifdef KTRACE
78 #include <sys/ktrace.h>
79 #endif
80
81 #include <security/audit/audit.h>
82 #include <security/mac/mac_framework.h>
83
84 #include <vm/vm.h>
85 #include <vm/vm_extern.h>
86 #include <vm/vm_param.h>
87 #include <vm/pmap.h>
88 #include <vm/vm_map.h>
89 #include <vm/vm_page.h>
90 #include <vm/uma.h>
91
92 #ifdef KDTRACE_HOOKS
93 #include <sys/dtrace_bsd.h>
94 dtrace_execexit_func_t dtrace_fasttrap_exit;
95 #endif
96
97 SDT_PROVIDER_DECLARE(proc);
98 SDT_PROBE_DEFINE1(proc, kernel, , exit, "int");
99
100 /* Hook for NFS teardown procedure. */
101 void (*nlminfo_release_p)(struct proc *p);
102
103 struct proc *
104 proc_realparent(struct proc *child)
105 {
106 struct proc *p, *parent;
107
108 sx_assert(&proctree_lock, SX_LOCKED);
109 if ((child->p_treeflag & P_TREE_ORPHANED) == 0) {
110 if (child->p_oppid == 0 ||
111 child->p_pptr->p_pid == child->p_oppid)
112 parent = child->p_pptr;
113 else
114 parent = initproc;
115 return (parent);
116 }
117 for (p = child; (p->p_treeflag & P_TREE_FIRST_ORPHAN) == 0;) {
118 /* Cannot use LIST_PREV(), since the list head is not known. */
119 p = __containerof(p->p_orphan.le_prev, struct proc,
120 p_orphan.le_next);
121 KASSERT((p->p_treeflag & P_TREE_ORPHANED) != 0,
122 ("missing P_ORPHAN %p", p));
123 }
124 parent = __containerof(p->p_orphan.le_prev, struct proc,
125 p_orphans.lh_first);
126 return (parent);
127 }
128
129 void
130 reaper_abandon_children(struct proc *p, bool exiting)
131 {
132 struct proc *p1, *p2, *ptmp;
133
134 sx_assert(&proctree_lock, SX_LOCKED);
135 KASSERT(p != initproc, ("reaper_abandon_children for initproc"));
136 if ((p->p_treeflag & P_TREE_REAPER) == 0)
137 return;
138 p1 = p->p_reaper;
139 LIST_FOREACH_SAFE(p2, &p->p_reaplist, p_reapsibling, ptmp) {
140 LIST_REMOVE(p2, p_reapsibling);
141 p2->p_reaper = p1;
142 p2->p_reapsubtree = p->p_reapsubtree;
143 LIST_INSERT_HEAD(&p1->p_reaplist, p2, p_reapsibling);
144 if (exiting && p2->p_pptr == p) {
145 PROC_LOCK(p2);
146 proc_reparent(p2, p1);
147 PROC_UNLOCK(p2);
148 }
149 }
150 KASSERT(LIST_EMPTY(&p->p_reaplist), ("p_reaplist not empty"));
151 p->p_treeflag &= ~P_TREE_REAPER;
152 }
153
154 static void
155 clear_orphan(struct proc *p)
156 {
157 struct proc *p1;
158
159 sx_assert(&proctree_lock, SA_XLOCKED);
160 if ((p->p_treeflag & P_TREE_ORPHANED) == 0)
161 return;
162 if ((p->p_treeflag & P_TREE_FIRST_ORPHAN) != 0) {
163 p1 = LIST_NEXT(p, p_orphan);
164 if (p1 != NULL)
165 p1->p_treeflag |= P_TREE_FIRST_ORPHAN;
166 p->p_treeflag &= ~P_TREE_FIRST_ORPHAN;
167 }
168 LIST_REMOVE(p, p_orphan);
169 p->p_treeflag &= ~P_TREE_ORPHANED;
170 }
171
172 /*
173 * exit -- death of process.
174 */
175 void
176 sys_sys_exit(struct thread *td, struct sys_exit_args *uap)
177 {
178
179 exit1(td, W_EXITCODE(uap->rval, 0));
180 /* NOTREACHED */
181 }
182
183 /*
184 * Exit: deallocate address space and other resources, change proc state to
185 * zombie, and unlink proc from allproc and parent's lists. Save exit status
186 * and rusage for wait(). Check for child processes and orphan them.
187 */
188 void
189 exit1(struct thread *td, int rv)
190 {
191 struct proc *p, *nq, *q, *t;
192 struct thread *tdt;
193 struct vnode *ttyvp = NULL;
194
195 mtx_assert(&Giant, MA_NOTOWNED);
196
197 p = td->td_proc;
198 /*
199 * XXX in case we're rebooting we just let init die in order to
200 * work around an unsolved stack overflow seen very late during
201 * shutdown on sparc64 when the gmirror worker process exists.
202 */
203 if (p == initproc && rebooting == 0) {
204 printf("init died (signal %d, exit %d)\n",
205 WTERMSIG(rv), WEXITSTATUS(rv));
206 panic("Going nowhere without my init!");
207 }
208
209 /*
210 * Deref SU mp, since the thread does not return to userspace.
211 */
212 if (softdep_ast_cleanup != NULL)
213 softdep_ast_cleanup();
214
215 /*
216 * MUST abort all other threads before proceeding past here.
217 */
218 PROC_LOCK(p);
219 /*
220 * First check if some other thread or external request got
221 * here before us. If so, act appropriately: exit or suspend.
222 * We must ensure that stop requests are handled before we set
223 * P_WEXIT.
224 */
225 thread_suspend_check(0);
226 while (p->p_flag & P_HADTHREADS) {
227 /*
228 * Kill off the other threads. This requires
229 * some co-operation from other parts of the kernel
230 * so it may not be instantaneous. With this state set
231 * any thread entering the kernel from userspace will
232 * thread_exit() in trap(). Any thread attempting to
233 * sleep will return immediately with EINTR or EWOULDBLOCK
234 * which will hopefully force them to back out to userland
235 * freeing resources as they go. Any thread attempting
236 * to return to userland will thread_exit() from userret().
237 * thread_exit() will unsuspend us when the last of the
238 * other threads exits.
239 * If there is already a thread singler after resumption,
240 * calling thread_single will fail; in that case, we just
241 * re-check all suspension request, the thread should
242 * either be suspended there or exit.
243 */
244 if (!thread_single(p, SINGLE_EXIT))
245 /*
246 * All other activity in this process is now
247 * stopped. Threading support has been turned
248 * off.
249 */
250 break;
251 /*
252 * Recheck for new stop or suspend requests which
253 * might appear while process lock was dropped in
254 * thread_single().
255 */
256 thread_suspend_check(0);
257 }
258 KASSERT(p->p_numthreads == 1,
259 ("exit1: proc %p exiting with %d threads", p, p->p_numthreads));
260 racct_sub(p, RACCT_NTHR, 1);
261 /*
262 * Wakeup anyone in procfs' PIOCWAIT. They should have a hold
263 * on our vmspace, so we should block below until they have
264 * released their reference to us. Note that if they have
265 * requested S_EXIT stops we will block here until they ack
266 * via PIOCCONT.
267 */
268 _STOPEVENT(p, S_EXIT, rv);
269
270 /*
271 * Ignore any pending request to stop due to a stop signal.
272 * Once P_WEXIT is set, future requests will be ignored as
273 * well.
274 */
275 p->p_flag &= ~P_STOPPED_SIG;
276 KASSERT(!P_SHOULDSTOP(p), ("exiting process is stopped"));
277
278 /*
279 * Note that we are exiting and do another wakeup of anyone in
280 * PIOCWAIT in case they aren't listening for S_EXIT stops or
281 * decided to wait again after we told them we are exiting.
282 */
283 p->p_flag |= P_WEXIT;
284 wakeup(&p->p_stype);
285
286 /*
287 * Wait for any processes that have a hold on our vmspace to
288 * release their reference.
289 */
290 while (p->p_lock > 0)
291 msleep(&p->p_lock, &p->p_mtx, PWAIT, "exithold", 0);
292
293 p->p_xstat = rv; /* Let event handler change exit status */
294 PROC_UNLOCK(p);
295 /* Drain the limit callout while we don't have the proc locked */
296 callout_drain(&p->p_limco);
297
298 #ifdef AUDIT
299 /*
300 * The Sun BSM exit token contains two components: an exit status as
301 * passed to exit(), and a return value to indicate what sort of exit
302 * it was. The exit status is WEXITSTATUS(rv), but it's not clear
303 * what the return value is.
304 */
305 AUDIT_ARG_EXIT(WEXITSTATUS(rv), 0);
306 AUDIT_SYSCALL_EXIT(0, td);
307 #endif
308
309 /* Are we a task leader? */
310 if (p == p->p_leader) {
311 mtx_lock(&ppeers_lock);
312 q = p->p_peers;
313 while (q != NULL) {
314 PROC_LOCK(q);
315 kern_psignal(q, SIGKILL);
316 PROC_UNLOCK(q);
317 q = q->p_peers;
318 }
319 while (p->p_peers != NULL)
320 msleep(p, &ppeers_lock, PWAIT, "exit1", 0);
321 mtx_unlock(&ppeers_lock);
322 }
323
324 /*
325 * Check if any loadable modules need anything done at process exit.
326 * E.g. SYSV IPC stuff
327 * XXX what if one of these generates an error?
328 */
329 EVENTHANDLER_INVOKE(process_exit, p);
330
331 /*
332 * If parent is waiting for us to exit or exec,
333 * P_PPWAIT is set; we will wakeup the parent below.
334 */
335 PROC_LOCK(p);
336 rv = p->p_xstat; /* Event handler could change exit status */
337 stopprofclock(p);
338 p->p_flag &= ~(P_TRACED | P_PPWAIT | P_PPTRACE);
339
340 /*
341 * Stop the real interval timer. If the handler is currently
342 * executing, prevent it from rearming itself and let it finish.
343 */
344 if (timevalisset(&p->p_realtimer.it_value) &&
345 callout_stop(&p->p_itcallout) == 0) {
346 timevalclear(&p->p_realtimer.it_interval);
347 msleep(&p->p_itcallout, &p->p_mtx, PWAIT, "ritwait", 0);
348 KASSERT(!timevalisset(&p->p_realtimer.it_value),
349 ("realtime timer is still armed"));
350 }
351 PROC_UNLOCK(p);
352
353 /*
354 * Reset any sigio structures pointing to us as a result of
355 * F_SETOWN with our pid.
356 */
357 funsetownlst(&p->p_sigiolst);
358
359 /*
360 * If this process has an nlminfo data area (for lockd), release it
361 */
362 if (nlminfo_release_p != NULL && p->p_nlminfo != NULL)
363 (*nlminfo_release_p)(p);
364
365 /*
366 * Close open files and release open-file table.
367 * This may block!
368 */
369 fdescfree(td);
370
371 /*
372 * If this thread tickled GEOM, we need to wait for the giggling to
373 * stop before we return to userland
374 */
375 if (td->td_pflags & TDP_GEOM)
376 g_waitidle();
377
378 /*
379 * Remove ourself from our leader's peer list and wake our leader.
380 */
381 mtx_lock(&ppeers_lock);
382 if (p->p_leader->p_peers) {
383 q = p->p_leader;
384 while (q->p_peers != p)
385 q = q->p_peers;
386 q->p_peers = p->p_peers;
387 wakeup(p->p_leader);
388 }
389 mtx_unlock(&ppeers_lock);
390
391 vmspace_exit(td);
392
393 sx_xlock(&proctree_lock);
394 if (SESS_LEADER(p)) {
395 struct session *sp = p->p_session;
396 struct tty *tp;
397
398 /*
399 * s_ttyp is not zero'd; we use this to indicate that
400 * the session once had a controlling terminal. (for
401 * logging and informational purposes)
402 */
403 SESS_LOCK(sp);
404 ttyvp = sp->s_ttyvp;
405 tp = sp->s_ttyp;
406 sp->s_ttyvp = NULL;
407 sp->s_ttydp = NULL;
408 sp->s_leader = NULL;
409 SESS_UNLOCK(sp);
410
411 /*
412 * Signal foreground pgrp and revoke access to
413 * controlling terminal if it has not been revoked
414 * already.
415 *
416 * Because the TTY may have been revoked in the mean
417 * time and could already have a new session associated
418 * with it, make sure we don't send a SIGHUP to a
419 * foreground process group that does not belong to this
420 * session.
421 */
422
423 if (tp != NULL) {
424 tty_lock(tp);
425 if (tp->t_session == sp)
426 tty_signal_pgrp(tp, SIGHUP);
427 tty_unlock(tp);
428 }
429
430 if (ttyvp != NULL) {
431 sx_xunlock(&proctree_lock);
432 if (vn_lock(ttyvp, LK_EXCLUSIVE) == 0) {
433 VOP_REVOKE(ttyvp, REVOKEALL);
434 VOP_UNLOCK(ttyvp, 0);
435 }
436 sx_xlock(&proctree_lock);
437 }
438 }
439 fixjobc(p, p->p_pgrp, 0);
440 sx_xunlock(&proctree_lock);
441 (void)acct_process(td);
442
443 /* Release the TTY now we've unlocked everything. */
444 if (ttyvp != NULL)
445 vrele(ttyvp);
446 #ifdef KTRACE
447 ktrprocexit(td);
448 #endif
449 /*
450 * Release reference to text vnode
451 */
452 if (p->p_textvp != NULL) {
453 vrele(p->p_textvp);
454 p->p_textvp = NULL;
455 }
456
457 /*
458 * Release our limits structure.
459 */
460 lim_free(p->p_limit);
461 p->p_limit = NULL;
462
463 tidhash_remove(td);
464
465 /*
466 * Remove proc from allproc queue and pidhash chain.
467 * Place onto zombproc. Unlink from parent's child list.
468 */
469 sx_xlock(&allproc_lock);
470 LIST_REMOVE(p, p_list);
471 LIST_INSERT_HEAD(&zombproc, p, p_list);
472 LIST_REMOVE(p, p_hash);
473 sx_xunlock(&allproc_lock);
474
475 /*
476 * Call machine-dependent code to release any
477 * machine-dependent resources other than the address space.
478 * The address space is released by "vmspace_exitfree(p)" in
479 * vm_waitproc().
480 */
481 cpu_exit(td);
482
483 WITNESS_WARN(WARN_PANIC, NULL, "process (pid %d) exiting", p->p_pid);
484
485 /*
486 * Reparent all children processes:
487 * - traced ones to the original parent (or init if we are that parent)
488 * - the rest to init
489 */
490 sx_xlock(&proctree_lock);
491 q = LIST_FIRST(&p->p_children);
492 if (q != NULL) /* only need this if any child is S_ZOMB */
493 wakeup(q->p_reaper);
494 for (; q != NULL; q = nq) {
495 nq = LIST_NEXT(q, p_sibling);
496 PROC_LOCK(q);
497 q->p_sigparent = SIGCHLD;
498
499 if (!(q->p_flag & P_TRACED)) {
500 proc_reparent(q, q->p_reaper);
501 } else {
502 /*
503 * Traced processes are killed since their existence
504 * means someone is screwing up.
505 */
506 t = proc_realparent(q);
507 if (t == p) {
508 proc_reparent(q, q->p_reaper);
509 } else {
510 PROC_LOCK(t);
511 proc_reparent(q, t);
512 PROC_UNLOCK(t);
513 }
514 /*
515 * Since q was found on our children list, the
516 * proc_reparent() call moved q to the orphan
517 * list due to present P_TRACED flag. Clear
518 * orphan link for q now while q is locked.
519 */
520 clear_orphan(q);
521 q->p_flag &= ~(P_TRACED | P_STOPPED_TRACE);
522 FOREACH_THREAD_IN_PROC(q, tdt)
523 tdt->td_dbgflags &= ~TDB_SUSPEND;
524 kern_psignal(q, SIGKILL);
525 }
526 PROC_UNLOCK(q);
527 }
528
529 /*
530 * Also get rid of our orphans.
531 */
532 while ((q = LIST_FIRST(&p->p_orphans)) != NULL) {
533 PROC_LOCK(q);
534 CTR2(KTR_PTRACE, "exit: pid %d, clearing orphan %d", p->p_pid,
535 q->p_pid);
536 clear_orphan(q);
537 PROC_UNLOCK(q);
538 }
539
540 /* Save exit status. */
541 PROC_LOCK(p);
542 p->p_xthread = td;
543
544 /* Tell the prison that we are gone. */
545 prison_proc_free(p->p_ucred->cr_prison);
546
547 #ifdef KDTRACE_HOOKS
548 /*
549 * Tell the DTrace fasttrap provider about the exit if it
550 * has declared an interest.
551 */
552 if (dtrace_fasttrap_exit)
553 dtrace_fasttrap_exit(p);
554 #endif
555
556 /*
557 * Notify interested parties of our demise.
558 */
559 KNOTE_LOCKED(&p->p_klist, NOTE_EXIT);
560
561 #ifdef KDTRACE_HOOKS
562 int reason = CLD_EXITED;
563 if (WCOREDUMP(rv))
564 reason = CLD_DUMPED;
565 else if (WIFSIGNALED(rv))
566 reason = CLD_KILLED;
567 SDT_PROBE1(proc, kernel, , exit, reason);
568 #endif
569
570 /*
571 * Just delete all entries in the p_klist. At this point we won't
572 * report any more events, and there are nasty race conditions that
573 * can beat us if we don't.
574 */
575 knlist_clear(&p->p_klist, 1);
576
577 /*
578 * If this is a process with a descriptor, we may not need to deliver
579 * a signal to the parent. proctree_lock is held over
580 * procdesc_exit() to serialize concurrent calls to close() and
581 * exit().
582 */
583 #ifdef PROCDESC
584 if (p->p_procdesc == NULL || procdesc_exit(p)) {
585 #endif
586 /*
587 * Notify parent that we're gone. If parent has the
588 * PS_NOCLDWAIT flag set, or if the handler is set to SIG_IGN,
589 * notify process 1 instead (and hope it will handle this
590 * situation).
591 */
592 PROC_LOCK(p->p_pptr);
593 mtx_lock(&p->p_pptr->p_sigacts->ps_mtx);
594 if (p->p_pptr->p_sigacts->ps_flag &
595 (PS_NOCLDWAIT | PS_CLDSIGIGN)) {
596 struct proc *pp;
597
598 mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx);
599 pp = p->p_pptr;
600 PROC_UNLOCK(pp);
601 proc_reparent(p, p->p_reaper);
602 p->p_sigparent = SIGCHLD;
603 PROC_LOCK(p->p_pptr);
604
605 /*
606 * Notify parent, so in case he was wait(2)ing or
607 * executing waitpid(2) with our pid, he will
608 * continue.
609 */
610 wakeup(pp);
611 } else
612 mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx);
613
614 if (p->p_pptr == p->p_reaper || p->p_pptr == initproc)
615 childproc_exited(p);
616 else if (p->p_sigparent != 0) {
617 if (p->p_sigparent == SIGCHLD)
618 childproc_exited(p);
619 else /* LINUX thread */
620 kern_psignal(p->p_pptr, p->p_sigparent);
621 }
622 #ifdef PROCDESC
623 } else
624 PROC_LOCK(p->p_pptr);
625 #endif
626 sx_xunlock(&proctree_lock);
627
628 /*
629 * The state PRS_ZOMBIE prevents other proesses from sending
630 * signal to the process, to avoid memory leak, we free memory
631 * for signal queue at the time when the state is set.
632 */
633 sigqueue_flush(&p->p_sigqueue);
634 sigqueue_flush(&td->td_sigqueue);
635
636 /*
637 * We have to wait until after acquiring all locks before
638 * changing p_state. We need to avoid all possible context
639 * switches (including ones from blocking on a mutex) while
640 * marked as a zombie. We also have to set the zombie state
641 * before we release the parent process' proc lock to avoid
642 * a lost wakeup. So, we first call wakeup, then we grab the
643 * sched lock, update the state, and release the parent process'
644 * proc lock.
645 */
646 wakeup(p->p_pptr);
647 cv_broadcast(&p->p_pwait);
648 sched_exit(p->p_pptr, td);
649 umtx_thread_exit(td);
650 PROC_SLOCK(p);
651 p->p_state = PRS_ZOMBIE;
652 PROC_UNLOCK(p->p_pptr);
653
654 /*
655 * Hopefully no one will try to deliver a signal to the process this
656 * late in the game.
657 */
658 knlist_destroy(&p->p_klist);
659
660 /*
661 * Save our children's rusage information in our exit rusage.
662 */
663 PROC_STATLOCK(p);
664 ruadd(&p->p_ru, &p->p_rux, &p->p_stats->p_cru, &p->p_crux);
665 PROC_STATUNLOCK(p);
666
667 /*
668 * Make sure the scheduler takes this thread out of its tables etc.
669 * This will also release this thread's reference to the ucred.
670 * Other thread parts to release include pcb bits and such.
671 */
672 thread_exit();
673 }
674
675
676 #ifndef _SYS_SYSPROTO_H_
677 struct abort2_args {
678 char *why;
679 int nargs;
680 void **args;
681 };
682 #endif
683
684 int
685 sys_abort2(struct thread *td, struct abort2_args *uap)
686 {
687 struct proc *p = td->td_proc;
688 struct sbuf *sb;
689 void *uargs[16];
690 int error, i, sig;
691
692 /*
693 * Do it right now so we can log either proper call of abort2(), or
694 * note, that invalid argument was passed. 512 is big enough to
695 * handle 16 arguments' descriptions with additional comments.
696 */
697 sb = sbuf_new(NULL, NULL, 512, SBUF_FIXEDLEN);
698 sbuf_clear(sb);
699 sbuf_printf(sb, "%s(pid %d uid %d) aborted: ",
700 p->p_comm, p->p_pid, td->td_ucred->cr_uid);
701 /*
702 * Since we can't return from abort2(), send SIGKILL in cases, where
703 * abort2() was called improperly
704 */
705 sig = SIGKILL;
706 /* Prevent from DoSes from user-space. */
707 if (uap->nargs < 0 || uap->nargs > 16)
708 goto out;
709 if (uap->nargs > 0) {
710 if (uap->args == NULL)
711 goto out;
712 error = copyin(uap->args, uargs, uap->nargs * sizeof(void *));
713 if (error != 0)
714 goto out;
715 }
716 /*
717 * Limit size of 'reason' string to 128. Will fit even when
718 * maximal number of arguments was chosen to be logged.
719 */
720 if (uap->why != NULL) {
721 error = sbuf_copyin(sb, uap->why, 128);
722 if (error < 0)
723 goto out;
724 } else {
725 sbuf_printf(sb, "(null)");
726 }
727 if (uap->nargs > 0) {
728 sbuf_printf(sb, "(");
729 for (i = 0;i < uap->nargs; i++)
730 sbuf_printf(sb, "%s%p", i == 0 ? "" : ", ", uargs[i]);
731 sbuf_printf(sb, ")");
732 }
733 /*
734 * Final stage: arguments were proper, string has been
735 * successfully copied from userspace, and copying pointers
736 * from user-space succeed.
737 */
738 sig = SIGABRT;
739 out:
740 if (sig == SIGKILL) {
741 sbuf_trim(sb);
742 sbuf_printf(sb, " (Reason text inaccessible)");
743 }
744 sbuf_cat(sb, "\n");
745 sbuf_finish(sb);
746 log(LOG_INFO, "%s", sbuf_data(sb));
747 sbuf_delete(sb);
748 exit1(td, W_EXITCODE(0, sig));
749 return (0);
750 }
751
752
753 #ifdef COMPAT_43
754 /*
755 * The dirty work is handled by kern_wait().
756 */
757 int
758 owait(struct thread *td, struct owait_args *uap __unused)
759 {
760 int error, status;
761
762 error = kern_wait(td, WAIT_ANY, &status, 0, NULL);
763 if (error == 0)
764 td->td_retval[1] = status;
765 return (error);
766 }
767 #endif /* COMPAT_43 */
768
769 /*
770 * The dirty work is handled by kern_wait().
771 */
772 int
773 sys_wait4(struct thread *td, struct wait4_args *uap)
774 {
775 struct rusage ru, *rup;
776 int error, status;
777
778 if (uap->rusage != NULL)
779 rup = &ru;
780 else
781 rup = NULL;
782 error = kern_wait(td, uap->pid, &status, uap->options, rup);
783 if (uap->status != NULL && error == 0)
784 error = copyout(&status, uap->status, sizeof(status));
785 if (uap->rusage != NULL && error == 0)
786 error = copyout(&ru, uap->rusage, sizeof(struct rusage));
787 return (error);
788 }
789
790 int
791 sys_wait6(struct thread *td, struct wait6_args *uap)
792 {
793 struct __wrusage wru, *wrup;
794 siginfo_t si, *sip;
795 idtype_t idtype;
796 id_t id;
797 int error, status;
798
799 idtype = uap->idtype;
800 id = uap->id;
801
802 if (uap->wrusage != NULL)
803 wrup = &wru;
804 else
805 wrup = NULL;
806
807 if (uap->info != NULL) {
808 sip = &si;
809 bzero(sip, sizeof(*sip));
810 } else
811 sip = NULL;
812
813 /*
814 * We expect all callers of wait6() to know about WEXITED and
815 * WTRAPPED.
816 */
817 error = kern_wait6(td, idtype, id, &status, uap->options, wrup, sip);
818
819 if (uap->status != NULL && error == 0)
820 error = copyout(&status, uap->status, sizeof(status));
821 if (uap->wrusage != NULL && error == 0)
822 error = copyout(&wru, uap->wrusage, sizeof(wru));
823 if (uap->info != NULL && error == 0)
824 error = copyout(&si, uap->info, sizeof(si));
825 return (error);
826 }
827
828 /*
829 * Reap the remains of a zombie process and optionally return status and
830 * rusage. Asserts and will release both the proctree_lock and the process
831 * lock as part of its work.
832 */
833 void
834 proc_reap(struct thread *td, struct proc *p, int *status, int options)
835 {
836 struct proc *q, *t;
837
838 sx_assert(&proctree_lock, SA_XLOCKED);
839 PROC_LOCK_ASSERT(p, MA_OWNED);
840 PROC_SLOCK_ASSERT(p, MA_OWNED);
841 KASSERT(p->p_state == PRS_ZOMBIE, ("proc_reap: !PRS_ZOMBIE"));
842
843 q = td->td_proc;
844
845 PROC_SUNLOCK(p);
846 td->td_retval[0] = p->p_pid;
847 if (status)
848 *status = p->p_xstat; /* convert to int */
849 if (options & WNOWAIT) {
850 /*
851 * Only poll, returning the status. Caller does not wish to
852 * release the proc struct just yet.
853 */
854 PROC_UNLOCK(p);
855 sx_xunlock(&proctree_lock);
856 return;
857 }
858
859 PROC_LOCK(q);
860 sigqueue_take(p->p_ksi);
861 PROC_UNLOCK(q);
862
863 /*
864 * If we got the child via a ptrace 'attach', we need to give it back
865 * to the old parent.
866 */
867 if (p->p_oppid != 0 && p->p_oppid != p->p_pptr->p_pid) {
868 PROC_UNLOCK(p);
869 t = proc_realparent(p);
870 PROC_LOCK(t);
871 PROC_LOCK(p);
872 CTR2(KTR_PTRACE,
873 "wait: traced child %d moved back to parent %d", p->p_pid,
874 t->p_pid);
875 proc_reparent(p, t);
876 p->p_oppid = 0;
877 PROC_UNLOCK(p);
878 pksignal(t, SIGCHLD, p->p_ksi);
879 wakeup(t);
880 cv_broadcast(&p->p_pwait);
881 PROC_UNLOCK(t);
882 sx_xunlock(&proctree_lock);
883 return;
884 }
885 p->p_oppid = 0;
886 PROC_UNLOCK(p);
887
888 /*
889 * Remove other references to this process to ensure we have an
890 * exclusive reference.
891 */
892 sx_xlock(&allproc_lock);
893 LIST_REMOVE(p, p_list); /* off zombproc */
894 sx_xunlock(&allproc_lock);
895 LIST_REMOVE(p, p_sibling);
896 reaper_abandon_children(p, true);
897 LIST_REMOVE(p, p_reapsibling);
898 PROC_LOCK(p);
899 clear_orphan(p);
900 PROC_UNLOCK(p);
901 leavepgrp(p);
902 #ifdef PROCDESC
903 if (p->p_procdesc != NULL)
904 procdesc_reap(p);
905 #endif
906 sx_xunlock(&proctree_lock);
907
908 /*
909 * As a side effect of this lock, we know that all other writes to
910 * this proc are visible now, so no more locking is needed for p.
911 */
912 PROC_LOCK(p);
913 p->p_xstat = 0; /* XXX: why? */
914 PROC_UNLOCK(p);
915 PROC_LOCK(q);
916 ruadd(&q->p_stats->p_cru, &q->p_crux, &p->p_ru, &p->p_rux);
917 PROC_UNLOCK(q);
918
919 /*
920 * Decrement the count of procs running with this uid.
921 */
922 (void)chgproccnt(p->p_ucred->cr_ruidinfo, -1, 0);
923
924 /*
925 * Destroy resource accounting information associated with the process.
926 */
927 #ifdef RACCT
928 if (racct_enable) {
929 PROC_LOCK(p);
930 racct_sub(p, RACCT_NPROC, 1);
931 PROC_UNLOCK(p);
932 }
933 #endif
934 racct_proc_exit(p);
935
936 /*
937 * Free credentials, arguments, and sigacts.
938 */
939 crfree(p->p_ucred);
940 p->p_ucred = NULL;
941 pargs_drop(p->p_args);
942 p->p_args = NULL;
943 sigacts_free(p->p_sigacts);
944 p->p_sigacts = NULL;
945
946 /*
947 * Do any thread-system specific cleanups.
948 */
949 thread_wait(p);
950
951 /*
952 * Give vm and machine-dependent layer a chance to free anything that
953 * cpu_exit couldn't release while still running in process context.
954 */
955 vm_waitproc(p);
956 #ifdef MAC
957 mac_proc_destroy(p);
958 #endif
959 KASSERT(FIRST_THREAD_IN_PROC(p),
960 ("proc_reap: no residual thread!"));
961 uma_zfree(proc_zone, p);
962 atomic_add_int(&nprocs, -1);
963 }
964
965 static int
966 proc_to_reap(struct thread *td, struct proc *p, idtype_t idtype, id_t id,
967 int *status, int options, struct __wrusage *wrusage, siginfo_t *siginfo,
968 int check_only)
969 {
970 struct rusage *rup;
971
972 sx_assert(&proctree_lock, SA_XLOCKED);
973
974 PROC_LOCK(p);
975
976 switch (idtype) {
977 case P_ALL:
978 break;
979 case P_PID:
980 if (p->p_pid != (pid_t)id) {
981 PROC_UNLOCK(p);
982 return (0);
983 }
984 break;
985 case P_PGID:
986 if (p->p_pgid != (pid_t)id) {
987 PROC_UNLOCK(p);
988 return (0);
989 }
990 break;
991 case P_SID:
992 if (p->p_session->s_sid != (pid_t)id) {
993 PROC_UNLOCK(p);
994 return (0);
995 }
996 break;
997 case P_UID:
998 if (p->p_ucred->cr_uid != (uid_t)id) {
999 PROC_UNLOCK(p);
1000 return (0);
1001 }
1002 break;
1003 case P_GID:
1004 if (p->p_ucred->cr_gid != (gid_t)id) {
1005 PROC_UNLOCK(p);
1006 return (0);
1007 }
1008 break;
1009 case P_JAILID:
1010 if (p->p_ucred->cr_prison->pr_id != (int)id) {
1011 PROC_UNLOCK(p);
1012 return (0);
1013 }
1014 break;
1015 /*
1016 * It seems that the thread structures get zeroed out
1017 * at process exit. This makes it impossible to
1018 * support P_SETID, P_CID or P_CPUID.
1019 */
1020 default:
1021 PROC_UNLOCK(p);
1022 return (0);
1023 }
1024
1025 if (p_canwait(td, p)) {
1026 PROC_UNLOCK(p);
1027 return (0);
1028 }
1029
1030 if (((options & WEXITED) == 0) && (p->p_state == PRS_ZOMBIE)) {
1031 PROC_UNLOCK(p);
1032 return (0);
1033 }
1034
1035 /*
1036 * This special case handles a kthread spawned by linux_clone
1037 * (see linux_misc.c). The linux_wait4 and linux_waitpid
1038 * functions need to be able to distinguish between waiting
1039 * on a process and waiting on a thread. It is a thread if
1040 * p_sigparent is not SIGCHLD, and the WLINUXCLONE option
1041 * signifies we want to wait for threads and not processes.
1042 */
1043 if ((p->p_sigparent != SIGCHLD) ^
1044 ((options & WLINUXCLONE) != 0)) {
1045 PROC_UNLOCK(p);
1046 return (0);
1047 }
1048
1049 if (siginfo != NULL) {
1050 bzero(siginfo, sizeof(*siginfo));
1051 siginfo->si_errno = 0;
1052
1053 /*
1054 * SUSv4 requires that the si_signo value is always
1055 * SIGCHLD. Obey it despite the rfork(2) interface
1056 * allows to request other signal for child exit
1057 * notification.
1058 */
1059 siginfo->si_signo = SIGCHLD;
1060
1061 /*
1062 * This is still a rough estimate. We will fix the
1063 * cases TRAPPED, STOPPED, and CONTINUED later.
1064 */
1065 if (WCOREDUMP(p->p_xstat)) {
1066 siginfo->si_code = CLD_DUMPED;
1067 siginfo->si_status = WTERMSIG(p->p_xstat);
1068 } else if (WIFSIGNALED(p->p_xstat)) {
1069 siginfo->si_code = CLD_KILLED;
1070 siginfo->si_status = WTERMSIG(p->p_xstat);
1071 } else {
1072 siginfo->si_code = CLD_EXITED;
1073 siginfo->si_status = WEXITSTATUS(p->p_xstat);
1074 }
1075
1076 siginfo->si_pid = p->p_pid;
1077 siginfo->si_uid = p->p_ucred->cr_uid;
1078
1079 /*
1080 * The si_addr field would be useful additional
1081 * detail, but apparently the PC value may be lost
1082 * when we reach this point. bzero() above sets
1083 * siginfo->si_addr to NULL.
1084 */
1085 }
1086
1087 /*
1088 * There should be no reason to limit resources usage info to
1089 * exited processes only. A snapshot about any resources used
1090 * by a stopped process may be exactly what is needed.
1091 */
1092 if (wrusage != NULL) {
1093 rup = &wrusage->wru_self;
1094 *rup = p->p_ru;
1095 PROC_STATLOCK(p);
1096 calcru(p, &rup->ru_utime, &rup->ru_stime);
1097 PROC_STATUNLOCK(p);
1098
1099 rup = &wrusage->wru_children;
1100 *rup = p->p_stats->p_cru;
1101 calccru(p, &rup->ru_utime, &rup->ru_stime);
1102 }
1103
1104 if (p->p_state == PRS_ZOMBIE && !check_only) {
1105 PROC_SLOCK(p);
1106 proc_reap(td, p, status, options);
1107 return (-1);
1108 }
1109 PROC_UNLOCK(p);
1110 return (1);
1111 }
1112
1113 int
1114 kern_wait(struct thread *td, pid_t pid, int *status, int options,
1115 struct rusage *rusage)
1116 {
1117 struct __wrusage wru, *wrup;
1118 idtype_t idtype;
1119 id_t id;
1120 int ret;
1121
1122 /*
1123 * Translate the special pid values into the (idtype, pid)
1124 * pair for kern_wait6. The WAIT_MYPGRP case is handled by
1125 * kern_wait6() on its own.
1126 */
1127 if (pid == WAIT_ANY) {
1128 idtype = P_ALL;
1129 id = 0;
1130 } else if (pid < 0) {
1131 idtype = P_PGID;
1132 id = (id_t)-pid;
1133 } else {
1134 idtype = P_PID;
1135 id = (id_t)pid;
1136 }
1137
1138 if (rusage != NULL)
1139 wrup = &wru;
1140 else
1141 wrup = NULL;
1142
1143 /*
1144 * For backward compatibility we implicitly add flags WEXITED
1145 * and WTRAPPED here.
1146 */
1147 options |= WEXITED | WTRAPPED;
1148 ret = kern_wait6(td, idtype, id, status, options, wrup, NULL);
1149 if (rusage != NULL)
1150 *rusage = wru.wru_self;
1151 return (ret);
1152 }
1153
1154 int
1155 kern_wait6(struct thread *td, idtype_t idtype, id_t id, int *status,
1156 int options, struct __wrusage *wrusage, siginfo_t *siginfo)
1157 {
1158 struct proc *p, *q;
1159 int error, nfound, ret;
1160
1161 AUDIT_ARG_VALUE((int)idtype); /* XXX - This is likely wrong! */
1162 AUDIT_ARG_PID((pid_t)id); /* XXX - This may be wrong! */
1163 AUDIT_ARG_VALUE(options);
1164
1165 q = td->td_proc;
1166
1167 if ((pid_t)id == WAIT_MYPGRP && (idtype == P_PID || idtype == P_PGID)) {
1168 PROC_LOCK(q);
1169 id = (id_t)q->p_pgid;
1170 PROC_UNLOCK(q);
1171 idtype = P_PGID;
1172 }
1173
1174 /* If we don't know the option, just return. */
1175 if ((options & ~(WUNTRACED | WNOHANG | WCONTINUED | WNOWAIT |
1176 WEXITED | WTRAPPED | WLINUXCLONE)) != 0)
1177 return (EINVAL);
1178 if ((options & (WEXITED | WUNTRACED | WCONTINUED | WTRAPPED)) == 0) {
1179 /*
1180 * We will be unable to find any matching processes,
1181 * because there are no known events to look for.
1182 * Prefer to return error instead of blocking
1183 * indefinitely.
1184 */
1185 return (EINVAL);
1186 }
1187
1188 loop:
1189 if (q->p_flag & P_STATCHILD) {
1190 PROC_LOCK(q);
1191 q->p_flag &= ~P_STATCHILD;
1192 PROC_UNLOCK(q);
1193 }
1194 nfound = 0;
1195 sx_xlock(&proctree_lock);
1196 LIST_FOREACH(p, &q->p_children, p_sibling) {
1197 ret = proc_to_reap(td, p, idtype, id, status, options,
1198 wrusage, siginfo, 0);
1199 if (ret == 0)
1200 continue;
1201 else if (ret == 1)
1202 nfound++;
1203 else
1204 return (0);
1205
1206 PROC_LOCK(p);
1207 PROC_SLOCK(p);
1208
1209 if ((options & WTRAPPED) != 0 &&
1210 (p->p_flag & P_TRACED) != 0 &&
1211 (p->p_flag & (P_STOPPED_TRACE | P_STOPPED_SIG)) != 0 &&
1212 (p->p_suspcount == p->p_numthreads) &&
1213 ((p->p_flag & P_WAITED) == 0)) {
1214 PROC_SUNLOCK(p);
1215 if ((options & WNOWAIT) == 0)
1216 p->p_flag |= P_WAITED;
1217 sx_xunlock(&proctree_lock);
1218 td->td_retval[0] = p->p_pid;
1219
1220 if (status != NULL)
1221 *status = W_STOPCODE(p->p_xstat);
1222 if (siginfo != NULL) {
1223 siginfo->si_status = p->p_xstat;
1224 siginfo->si_code = CLD_TRAPPED;
1225 }
1226 if ((options & WNOWAIT) == 0) {
1227 PROC_LOCK(q);
1228 sigqueue_take(p->p_ksi);
1229 PROC_UNLOCK(q);
1230 }
1231
1232 CTR4(KTR_PTRACE,
1233 "wait: returning trapped pid %d status %#x (xstat %d) xthread %d",
1234 p->p_pid, W_STOPCODE(p->p_xstat), p->p_xstat,
1235 p->p_xthread != NULL ? p->p_xthread->td_tid : -1);
1236 PROC_UNLOCK(p);
1237 return (0);
1238 }
1239 if ((options & WUNTRACED) != 0 &&
1240 (p->p_flag & P_STOPPED_SIG) != 0 &&
1241 (p->p_suspcount == p->p_numthreads) &&
1242 ((p->p_flag & P_WAITED) == 0)) {
1243 PROC_SUNLOCK(p);
1244 if ((options & WNOWAIT) == 0)
1245 p->p_flag |= P_WAITED;
1246 sx_xunlock(&proctree_lock);
1247 td->td_retval[0] = p->p_pid;
1248
1249 if (status != NULL)
1250 *status = W_STOPCODE(p->p_xstat);
1251 if (siginfo != NULL) {
1252 siginfo->si_status = p->p_xstat;
1253 siginfo->si_code = CLD_STOPPED;
1254 }
1255 if ((options & WNOWAIT) == 0) {
1256 PROC_LOCK(q);
1257 sigqueue_take(p->p_ksi);
1258 PROC_UNLOCK(q);
1259 }
1260
1261 PROC_UNLOCK(p);
1262 return (0);
1263 }
1264 PROC_SUNLOCK(p);
1265 if ((options & WCONTINUED) != 0 &&
1266 (p->p_flag & P_CONTINUED) != 0) {
1267 sx_xunlock(&proctree_lock);
1268 td->td_retval[0] = p->p_pid;
1269 if ((options & WNOWAIT) == 0) {
1270 p->p_flag &= ~P_CONTINUED;
1271 PROC_LOCK(q);
1272 sigqueue_take(p->p_ksi);
1273 PROC_UNLOCK(q);
1274 }
1275 PROC_UNLOCK(p);
1276
1277 if (status != NULL)
1278 *status = SIGCONT;
1279 if (siginfo != NULL) {
1280 siginfo->si_status = SIGCONT;
1281 siginfo->si_code = CLD_CONTINUED;
1282 }
1283 return (0);
1284 }
1285 PROC_UNLOCK(p);
1286 }
1287
1288 /*
1289 * Look in the orphans list too, to allow the parent to
1290 * collect it's child exit status even if child is being
1291 * debugged.
1292 *
1293 * Debugger detaches from the parent upon successful
1294 * switch-over from parent to child. At this point due to
1295 * re-parenting the parent loses the child to debugger and a
1296 * wait4(2) call would report that it has no children to wait
1297 * for. By maintaining a list of orphans we allow the parent
1298 * to successfully wait until the child becomes a zombie.
1299 */
1300 if (nfound == 0) {
1301 LIST_FOREACH(p, &q->p_orphans, p_orphan) {
1302 ret = proc_to_reap(td, p, idtype, id, NULL, options,
1303 NULL, NULL, 1);
1304 if (ret != 0) {
1305 KASSERT(ret != -1, ("reaped an orphan (pid %d)",
1306 (int)td->td_retval[0]));
1307 nfound++;
1308 break;
1309 }
1310 }
1311 }
1312 if (nfound == 0) {
1313 sx_xunlock(&proctree_lock);
1314 return (ECHILD);
1315 }
1316 if (options & WNOHANG) {
1317 sx_xunlock(&proctree_lock);
1318 td->td_retval[0] = 0;
1319 return (0);
1320 }
1321 PROC_LOCK(q);
1322 sx_xunlock(&proctree_lock);
1323 if (q->p_flag & P_STATCHILD) {
1324 q->p_flag &= ~P_STATCHILD;
1325 error = 0;
1326 } else
1327 error = msleep(q, &q->p_mtx, PWAIT | PCATCH, "wait", 0);
1328 PROC_UNLOCK(q);
1329 if (error)
1330 return (error);
1331 goto loop;
1332 }
1333
1334 /*
1335 * Make process 'parent' the new parent of process 'child'.
1336 * Must be called with an exclusive hold of proctree lock.
1337 */
1338 void
1339 proc_reparent(struct proc *child, struct proc *parent)
1340 {
1341
1342 sx_assert(&proctree_lock, SX_XLOCKED);
1343 PROC_LOCK_ASSERT(child, MA_OWNED);
1344 if (child->p_pptr == parent)
1345 return;
1346
1347 PROC_LOCK(child->p_pptr);
1348 sigqueue_take(child->p_ksi);
1349 PROC_UNLOCK(child->p_pptr);
1350 LIST_REMOVE(child, p_sibling);
1351 LIST_INSERT_HEAD(&parent->p_children, child, p_sibling);
1352
1353 clear_orphan(child);
1354 if (child->p_flag & P_TRACED) {
1355 if (LIST_EMPTY(&child->p_pptr->p_orphans)) {
1356 child->p_treeflag |= P_TREE_FIRST_ORPHAN;
1357 LIST_INSERT_HEAD(&child->p_pptr->p_orphans, child,
1358 p_orphan);
1359 } else {
1360 LIST_INSERT_AFTER(LIST_FIRST(&child->p_pptr->p_orphans),
1361 child, p_orphan);
1362 }
1363 child->p_treeflag |= P_TREE_ORPHANED;
1364 }
1365
1366 child->p_pptr = parent;
1367 }
Cache object: 6f66eb121e1079b6e1d87dc05d322a6e
|