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("$FreeBSD$");
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, *t;
166 struct thread *tdt;
167 struct vnode *vtmp;
168 struct vnode *ttyvp = NULL;
169 struct plimit *plim;
170 int locked;
171
172 mtx_assert(&Giant, MA_NOTOWNED);
173
174 p = td->td_proc;
175 /*
176 * XXX in case we're rebooting we just let init die in order to
177 * work around an unsolved stack overflow seen very late during
178 * shutdown on sparc64 when the gmirror worker process exists.
179 */
180 if (p == initproc && rebooting == 0) {
181 printf("init died (signal %d, exit %d)\n",
182 WTERMSIG(rv), WEXITSTATUS(rv));
183 panic("Going nowhere without my init!");
184 }
185
186 /*
187 * MUST abort all other threads before proceeding past here.
188 */
189 PROC_LOCK(p);
190 while (p->p_flag & P_HADTHREADS) {
191 /*
192 * First check if some other thread got here before us..
193 * if so, act apropriatly, (exit or suspend);
194 */
195 thread_suspend_check(0);
196
197 /*
198 * Kill off the other threads. This requires
199 * some co-operation from other parts of the kernel
200 * so it may not be instantaneous. With this state set
201 * any thread entering the kernel from userspace will
202 * thread_exit() in trap(). Any thread attempting to
203 * sleep will return immediately with EINTR or EWOULDBLOCK
204 * which will hopefully force them to back out to userland
205 * freeing resources as they go. Any thread attempting
206 * to return to userland will thread_exit() from userret().
207 * thread_exit() will unsuspend us when the last of the
208 * other threads exits.
209 * If there is already a thread singler after resumption,
210 * calling thread_single will fail; in that case, we just
211 * re-check all suspension request, the thread should
212 * either be suspended there or exit.
213 */
214 if (! thread_single(SINGLE_EXIT))
215 break;
216
217 /*
218 * All other activity in this process is now stopped.
219 * Threading support has been turned off.
220 */
221 }
222 KASSERT(p->p_numthreads == 1,
223 ("exit1: proc %p exiting with %d threads", p, p->p_numthreads));
224 racct_sub(p, RACCT_NTHR, 1);
225 /*
226 * Wakeup anyone in procfs' PIOCWAIT. They should have a hold
227 * on our vmspace, so we should block below until they have
228 * released their reference to us. Note that if they have
229 * requested S_EXIT stops we will block here until they ack
230 * via PIOCCONT.
231 */
232 _STOPEVENT(p, S_EXIT, rv);
233
234 /*
235 * Ignore any pending request to stop due to a stop signal.
236 * Once P_WEXIT is set, future requests will be ignored as
237 * well.
238 */
239 p->p_flag &= ~P_STOPPED_SIG;
240 KASSERT(!P_SHOULDSTOP(p), ("exiting process is stopped"));
241
242 /*
243 * Note that we are exiting and do another wakeup of anyone in
244 * PIOCWAIT in case they aren't listening for S_EXIT stops or
245 * decided to wait again after we told them we are exiting.
246 */
247 p->p_flag |= P_WEXIT;
248 wakeup(&p->p_stype);
249
250 /*
251 * Wait for any processes that have a hold on our vmspace to
252 * release their reference.
253 */
254 while (p->p_lock > 0)
255 msleep(&p->p_lock, &p->p_mtx, PWAIT, "exithold", 0);
256
257 p->p_xstat = rv; /* Let event handler change exit status */
258 PROC_UNLOCK(p);
259 /* Drain the limit callout while we don't have the proc locked */
260 callout_drain(&p->p_limco);
261
262 #ifdef AUDIT
263 /*
264 * The Sun BSM exit token contains two components: an exit status as
265 * passed to exit(), and a return value to indicate what sort of exit
266 * it was. The exit status is WEXITSTATUS(rv), but it's not clear
267 * what the return value is.
268 */
269 AUDIT_ARG_EXIT(WEXITSTATUS(rv), 0);
270 AUDIT_SYSCALL_EXIT(0, td);
271 #endif
272
273 /* Are we a task leader? */
274 if (p == p->p_leader) {
275 mtx_lock(&ppeers_lock);
276 q = p->p_peers;
277 while (q != NULL) {
278 PROC_LOCK(q);
279 kern_psignal(q, SIGKILL);
280 PROC_UNLOCK(q);
281 q = q->p_peers;
282 }
283 while (p->p_peers != NULL)
284 msleep(p, &ppeers_lock, PWAIT, "exit1", 0);
285 mtx_unlock(&ppeers_lock);
286 }
287
288 /*
289 * Check if any loadable modules need anything done at process exit.
290 * E.g. SYSV IPC stuff
291 * XXX what if one of these generates an error?
292 */
293 EVENTHANDLER_INVOKE(process_exit, p);
294
295 /*
296 * If parent is waiting for us to exit or exec,
297 * P_PPWAIT is set; we will wakeup the parent below.
298 */
299 PROC_LOCK(p);
300 rv = p->p_xstat; /* Event handler could change exit status */
301 stopprofclock(p);
302 p->p_flag &= ~(P_TRACED | P_PPWAIT | P_PPTRACE);
303
304 /*
305 * Stop the real interval timer. If the handler is currently
306 * executing, prevent it from rearming itself and let it finish.
307 */
308 if (timevalisset(&p->p_realtimer.it_value) &&
309 callout_stop(&p->p_itcallout) == 0) {
310 timevalclear(&p->p_realtimer.it_interval);
311 msleep(&p->p_itcallout, &p->p_mtx, PWAIT, "ritwait", 0);
312 KASSERT(!timevalisset(&p->p_realtimer.it_value),
313 ("realtime timer is still armed"));
314 }
315 PROC_UNLOCK(p);
316
317 /*
318 * Reset any sigio structures pointing to us as a result of
319 * F_SETOWN with our pid.
320 */
321 funsetownlst(&p->p_sigiolst);
322
323 /*
324 * If this process has an nlminfo data area (for lockd), release it
325 */
326 if (nlminfo_release_p != NULL && p->p_nlminfo != NULL)
327 (*nlminfo_release_p)(p);
328
329 /*
330 * Close open files and release open-file table.
331 * This may block!
332 */
333 fdfree(td);
334
335 /*
336 * If this thread tickled GEOM, we need to wait for the giggling to
337 * stop before we return to userland
338 */
339 if (td->td_pflags & TDP_GEOM)
340 g_waitidle();
341
342 /*
343 * Remove ourself from our leader's peer list and wake our leader.
344 */
345 mtx_lock(&ppeers_lock);
346 if (p->p_leader->p_peers) {
347 q = p->p_leader;
348 while (q->p_peers != p)
349 q = q->p_peers;
350 q->p_peers = p->p_peers;
351 wakeup(p->p_leader);
352 }
353 mtx_unlock(&ppeers_lock);
354
355 vmspace_exit(td);
356
357 sx_xlock(&proctree_lock);
358 if (SESS_LEADER(p)) {
359 struct session *sp = p->p_session;
360 struct tty *tp;
361
362 /*
363 * s_ttyp is not zero'd; we use this to indicate that
364 * the session once had a controlling terminal. (for
365 * logging and informational purposes)
366 */
367 SESS_LOCK(sp);
368 ttyvp = sp->s_ttyvp;
369 tp = sp->s_ttyp;
370 sp->s_ttyvp = NULL;
371 sp->s_ttydp = NULL;
372 sp->s_leader = NULL;
373 SESS_UNLOCK(sp);
374
375 /*
376 * Signal foreground pgrp and revoke access to
377 * controlling terminal if it has not been revoked
378 * already.
379 *
380 * Because the TTY may have been revoked in the mean
381 * time and could already have a new session associated
382 * with it, make sure we don't send a SIGHUP to a
383 * foreground process group that does not belong to this
384 * session.
385 */
386
387 if (tp != NULL) {
388 tty_lock(tp);
389 if (tp->t_session == sp)
390 tty_signal_pgrp(tp, SIGHUP);
391 tty_unlock(tp);
392 }
393
394 if (ttyvp != NULL) {
395 sx_xunlock(&proctree_lock);
396 if (vn_lock(ttyvp, LK_EXCLUSIVE) == 0) {
397 VOP_REVOKE(ttyvp, REVOKEALL);
398 VOP_UNLOCK(ttyvp, 0);
399 }
400 sx_xlock(&proctree_lock);
401 }
402 }
403 fixjobc(p, p->p_pgrp, 0);
404 sx_xunlock(&proctree_lock);
405 (void)acct_process(td);
406
407 /* Release the TTY now we've unlocked everything. */
408 if (ttyvp != NULL)
409 vrele(ttyvp);
410 #ifdef KTRACE
411 ktrprocexit(td);
412 #endif
413 /*
414 * Release reference to text vnode
415 */
416 if ((vtmp = p->p_textvp) != NULL) {
417 p->p_textvp = NULL;
418 locked = VFS_LOCK_GIANT(vtmp->v_mount);
419 vrele(vtmp);
420 VFS_UNLOCK_GIANT(locked);
421 }
422
423 /*
424 * Release our limits structure.
425 */
426 PROC_LOCK(p);
427 plim = p->p_limit;
428 p->p_limit = NULL;
429 PROC_UNLOCK(p);
430 lim_free(plim);
431
432 tidhash_remove(td);
433
434 /*
435 * Remove proc from allproc queue and pidhash chain.
436 * Place onto zombproc. Unlink from parent's child list.
437 */
438 sx_xlock(&allproc_lock);
439 LIST_REMOVE(p, p_list);
440 LIST_INSERT_HEAD(&zombproc, p, p_list);
441 LIST_REMOVE(p, p_hash);
442 sx_xunlock(&allproc_lock);
443
444 /*
445 * Call machine-dependent code to release any
446 * machine-dependent resources other than the address space.
447 * The address space is released by "vmspace_exitfree(p)" in
448 * vm_waitproc().
449 */
450 cpu_exit(td);
451
452 WITNESS_WARN(WARN_PANIC, NULL, "process (pid %d) exiting", p->p_pid);
453
454 /*
455 * Reparent all children processes:
456 * - traced ones to the original parent (or init if we are that parent)
457 * - the rest to init
458 */
459 sx_xlock(&proctree_lock);
460 q = LIST_FIRST(&p->p_children);
461 if (q != NULL) /* only need this if any child is S_ZOMB */
462 wakeup(initproc);
463 for (; q != NULL; q = nq) {
464 nq = LIST_NEXT(q, p_sibling);
465 PROC_LOCK(q);
466 q->p_sigparent = SIGCHLD;
467
468 if (!(q->p_flag & P_TRACED)) {
469 proc_reparent(q, initproc);
470 } else {
471 /*
472 * Traced processes are killed since their existence
473 * means someone is screwing up.
474 */
475 t = proc_realparent(q);
476 if (t == p) {
477 proc_reparent(q, initproc);
478 } else {
479 PROC_LOCK(t);
480 proc_reparent(q, t);
481 PROC_UNLOCK(t);
482 }
483 /*
484 * Since q was found on our children list, the
485 * proc_reparent() call moved q to the orphan
486 * list due to present P_TRACED flag. Clear
487 * orphan link for q now while q is locked.
488 */
489 clear_orphan(q);
490 q->p_flag &= ~(P_TRACED | P_STOPPED_TRACE);
491 FOREACH_THREAD_IN_PROC(q, tdt)
492 tdt->td_dbgflags &= ~TDB_SUSPEND;
493 kern_psignal(q, SIGKILL);
494 }
495 PROC_UNLOCK(q);
496 }
497
498 /*
499 * Also get rid of our orphans.
500 */
501 while ((q = LIST_FIRST(&p->p_orphans)) != NULL) {
502 PROC_LOCK(q);
503 CTR2(KTR_PTRACE, "exit: pid %d, clearing orphan %d", p->p_pid,
504 q->p_pid);
505 clear_orphan(q);
506 PROC_UNLOCK(q);
507 }
508
509 /* Save exit status. */
510 PROC_LOCK(p);
511 p->p_xthread = td;
512
513 /* Tell the prison that we are gone. */
514 prison_proc_free(p->p_ucred->cr_prison);
515
516 #ifdef KDTRACE_HOOKS
517 /*
518 * Tell the DTrace fasttrap provider about the exit if it
519 * has declared an interest.
520 */
521 if (dtrace_fasttrap_exit)
522 dtrace_fasttrap_exit(p);
523 #endif
524
525 /*
526 * Notify interested parties of our demise.
527 */
528 KNOTE_LOCKED(&p->p_klist, NOTE_EXIT);
529
530 #ifdef KDTRACE_HOOKS
531 int reason = CLD_EXITED;
532 if (WCOREDUMP(rv))
533 reason = CLD_DUMPED;
534 else if (WIFSIGNALED(rv))
535 reason = CLD_KILLED;
536 SDT_PROBE1(proc, kernel, , exit, reason);
537 #endif
538
539 /*
540 * Just delete all entries in the p_klist. At this point we won't
541 * report any more events, and there are nasty race conditions that
542 * can beat us if we don't.
543 */
544 knlist_clear(&p->p_klist, 1);
545
546 /*
547 * If this is a process with a descriptor, we may not need to deliver
548 * a signal to the parent. proctree_lock is held over
549 * procdesc_exit() to serialize concurrent calls to close() and
550 * exit().
551 */
552 #ifdef PROCDESC
553 if (p->p_procdesc == NULL || procdesc_exit(p)) {
554 #endif
555 /*
556 * Notify parent that we're gone. If parent has the
557 * PS_NOCLDWAIT flag set, or if the handler is set to SIG_IGN,
558 * notify process 1 instead (and hope it will handle this
559 * situation).
560 */
561 PROC_LOCK(p->p_pptr);
562 mtx_lock(&p->p_pptr->p_sigacts->ps_mtx);
563 if (p->p_pptr->p_sigacts->ps_flag &
564 (PS_NOCLDWAIT | PS_CLDSIGIGN)) {
565 struct proc *pp;
566
567 mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx);
568 pp = p->p_pptr;
569 PROC_UNLOCK(pp);
570 proc_reparent(p, initproc);
571 p->p_sigparent = SIGCHLD;
572 PROC_LOCK(p->p_pptr);
573
574 /*
575 * Notify parent, so in case he was wait(2)ing or
576 * executing waitpid(2) with our pid, he will
577 * continue.
578 */
579 wakeup(pp);
580 } else
581 mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx);
582
583 if (p->p_pptr == initproc)
584 kern_psignal(p->p_pptr, SIGCHLD);
585 else if (p->p_sigparent != 0) {
586 if (p->p_sigparent == SIGCHLD)
587 childproc_exited(p);
588 else /* LINUX thread */
589 kern_psignal(p->p_pptr, p->p_sigparent);
590 }
591 #ifdef PROCDESC
592 } else
593 PROC_LOCK(p->p_pptr);
594 #endif
595 sx_xunlock(&proctree_lock);
596
597 /*
598 * The state PRS_ZOMBIE prevents other proesses from sending
599 * signal to the process, to avoid memory leak, we free memory
600 * for signal queue at the time when the state is set.
601 */
602 sigqueue_flush(&p->p_sigqueue);
603 sigqueue_flush(&td->td_sigqueue);
604
605 /*
606 * We have to wait until after acquiring all locks before
607 * changing p_state. We need to avoid all possible context
608 * switches (including ones from blocking on a mutex) while
609 * marked as a zombie. We also have to set the zombie state
610 * before we release the parent process' proc lock to avoid
611 * a lost wakeup. So, we first call wakeup, then we grab the
612 * sched lock, update the state, and release the parent process'
613 * proc lock.
614 */
615 wakeup(p->p_pptr);
616 cv_broadcast(&p->p_pwait);
617 sched_exit(p->p_pptr, td);
618 PROC_SLOCK(p);
619 p->p_state = PRS_ZOMBIE;
620 PROC_UNLOCK(p->p_pptr);
621
622 /*
623 * Hopefully no one will try to deliver a signal to the process this
624 * late in the game.
625 */
626 knlist_destroy(&p->p_klist);
627
628 /*
629 * Save our children's rusage information in our exit rusage.
630 */
631 ruadd(&p->p_ru, &p->p_rux, &p->p_stats->p_cru, &p->p_crux);
632
633 /*
634 * Make sure the scheduler takes this thread out of its tables etc.
635 * This will also release this thread's reference to the ucred.
636 * Other thread parts to release include pcb bits and such.
637 */
638 thread_exit();
639 }
640
641
642 #ifndef _SYS_SYSPROTO_H_
643 struct abort2_args {
644 char *why;
645 int nargs;
646 void **args;
647 };
648 #endif
649
650 int
651 sys_abort2(struct thread *td, struct abort2_args *uap)
652 {
653 struct proc *p = td->td_proc;
654 struct sbuf *sb;
655 void *uargs[16];
656 int error, i, sig;
657
658 /*
659 * Do it right now so we can log either proper call of abort2(), or
660 * note, that invalid argument was passed. 512 is big enough to
661 * handle 16 arguments' descriptions with additional comments.
662 */
663 sb = sbuf_new(NULL, NULL, 512, SBUF_FIXEDLEN);
664 sbuf_clear(sb);
665 sbuf_printf(sb, "%s(pid %d uid %d) aborted: ",
666 p->p_comm, p->p_pid, td->td_ucred->cr_uid);
667 /*
668 * Since we can't return from abort2(), send SIGKILL in cases, where
669 * abort2() was called improperly
670 */
671 sig = SIGKILL;
672 /* Prevent from DoSes from user-space. */
673 if (uap->nargs < 0 || uap->nargs > 16)
674 goto out;
675 if (uap->nargs > 0) {
676 if (uap->args == NULL)
677 goto out;
678 error = copyin(uap->args, uargs, uap->nargs * sizeof(void *));
679 if (error != 0)
680 goto out;
681 }
682 /*
683 * Limit size of 'reason' string to 128. Will fit even when
684 * maximal number of arguments was chosen to be logged.
685 */
686 if (uap->why != NULL) {
687 error = sbuf_copyin(sb, uap->why, 128);
688 if (error < 0)
689 goto out;
690 } else {
691 sbuf_printf(sb, "(null)");
692 }
693 if (uap->nargs > 0) {
694 sbuf_printf(sb, "(");
695 for (i = 0;i < uap->nargs; i++)
696 sbuf_printf(sb, "%s%p", i == 0 ? "" : ", ", uargs[i]);
697 sbuf_printf(sb, ")");
698 }
699 /*
700 * Final stage: arguments were proper, string has been
701 * successfully copied from userspace, and copying pointers
702 * from user-space succeed.
703 */
704 sig = SIGABRT;
705 out:
706 if (sig == SIGKILL) {
707 sbuf_trim(sb);
708 sbuf_printf(sb, " (Reason text inaccessible)");
709 }
710 sbuf_cat(sb, "\n");
711 sbuf_finish(sb);
712 log(LOG_INFO, "%s", sbuf_data(sb));
713 sbuf_delete(sb);
714 exit1(td, W_EXITCODE(0, sig));
715 return (0);
716 }
717
718
719 #ifdef COMPAT_43
720 /*
721 * The dirty work is handled by kern_wait().
722 */
723 int
724 owait(struct thread *td, struct owait_args *uap __unused)
725 {
726 int error, status;
727
728 error = kern_wait(td, WAIT_ANY, &status, 0, NULL);
729 if (error == 0)
730 td->td_retval[1] = status;
731 return (error);
732 }
733 #endif /* COMPAT_43 */
734
735 /*
736 * The dirty work is handled by kern_wait().
737 */
738 int
739 sys_wait4(struct thread *td, struct wait4_args *uap)
740 {
741 struct rusage ru, *rup;
742 int error, status;
743
744 if (uap->rusage != NULL)
745 rup = &ru;
746 else
747 rup = NULL;
748 error = kern_wait(td, uap->pid, &status, uap->options, rup);
749 if (uap->status != NULL && error == 0)
750 error = copyout(&status, uap->status, sizeof(status));
751 if (uap->rusage != NULL && error == 0)
752 error = copyout(&ru, uap->rusage, sizeof(struct rusage));
753 return (error);
754 }
755
756 int
757 sys_wait6(struct thread *td, struct wait6_args *uap)
758 {
759 struct __wrusage wru, *wrup;
760 siginfo_t si, *sip;
761 idtype_t idtype;
762 id_t id;
763 int error, status;
764
765 idtype = uap->idtype;
766 id = uap->id;
767
768 if (uap->wrusage != NULL)
769 wrup = &wru;
770 else
771 wrup = NULL;
772
773 if (uap->info != NULL) {
774 sip = &si;
775 bzero(sip, sizeof(*sip));
776 } else
777 sip = NULL;
778
779 /*
780 * We expect all callers of wait6() to know about WEXITED and
781 * WTRAPPED.
782 */
783 error = kern_wait6(td, idtype, id, &status, uap->options, wrup, sip);
784
785 if (uap->status != NULL && error == 0)
786 error = copyout(&status, uap->status, sizeof(status));
787 if (uap->wrusage != NULL && error == 0)
788 error = copyout(&wru, uap->wrusage, sizeof(wru));
789 if (uap->info != NULL && error == 0)
790 error = copyout(&si, uap->info, sizeof(si));
791 return (error);
792 }
793
794 /*
795 * Reap the remains of a zombie process and optionally return status and
796 * rusage. Asserts and will release both the proctree_lock and the process
797 * lock as part of its work.
798 */
799 void
800 proc_reap(struct thread *td, struct proc *p, int *status, int options)
801 {
802 struct proc *q, *t;
803
804 sx_assert(&proctree_lock, SA_XLOCKED);
805 PROC_LOCK_ASSERT(p, MA_OWNED);
806 PROC_SLOCK_ASSERT(p, MA_OWNED);
807 KASSERT(p->p_state == PRS_ZOMBIE, ("proc_reap: !PRS_ZOMBIE"));
808
809 q = td->td_proc;
810
811 PROC_SUNLOCK(p);
812 td->td_retval[0] = p->p_pid;
813 if (status)
814 *status = p->p_xstat; /* convert to int */
815 if (options & WNOWAIT) {
816 /*
817 * Only poll, returning the status. Caller does not wish to
818 * release the proc struct just yet.
819 */
820 PROC_UNLOCK(p);
821 sx_xunlock(&proctree_lock);
822 return;
823 }
824
825 PROC_LOCK(q);
826 sigqueue_take(p->p_ksi);
827 PROC_UNLOCK(q);
828
829 /*
830 * If we got the child via a ptrace 'attach', we need to give it back
831 * to the old parent.
832 */
833 if (p->p_oppid != 0 && p->p_oppid != p->p_pptr->p_pid) {
834 PROC_UNLOCK(p);
835 t = proc_realparent(p);
836 PROC_LOCK(t);
837 PROC_LOCK(p);
838 CTR2(KTR_PTRACE,
839 "wait: traced child %d moved back to parent %d", p->p_pid,
840 t->p_pid);
841 proc_reparent(p, t);
842 p->p_oppid = 0;
843 PROC_UNLOCK(p);
844 pksignal(t, SIGCHLD, p->p_ksi);
845 wakeup(t);
846 cv_broadcast(&p->p_pwait);
847 PROC_UNLOCK(t);
848 sx_xunlock(&proctree_lock);
849 return;
850 }
851 p->p_oppid = 0;
852 PROC_UNLOCK(p);
853
854 /*
855 * Remove other references to this process to ensure we have an
856 * exclusive reference.
857 */
858 sx_xlock(&allproc_lock);
859 LIST_REMOVE(p, p_list); /* off zombproc */
860 sx_xunlock(&allproc_lock);
861 LIST_REMOVE(p, p_sibling);
862 PROC_LOCK(p);
863 clear_orphan(p);
864 PROC_UNLOCK(p);
865 leavepgrp(p);
866 #ifdef PROCDESC
867 if (p->p_procdesc != NULL)
868 procdesc_reap(p);
869 #endif
870 sx_xunlock(&proctree_lock);
871
872 /*
873 * As a side effect of this lock, we know that all other writes to
874 * this proc are visible now, so no more locking is needed for p.
875 */
876 PROC_LOCK(p);
877 p->p_xstat = 0; /* XXX: why? */
878 PROC_UNLOCK(p);
879 PROC_LOCK(q);
880 ruadd(&q->p_stats->p_cru, &q->p_crux, &p->p_ru, &p->p_rux);
881 PROC_UNLOCK(q);
882
883 /*
884 * Decrement the count of procs running with this uid.
885 */
886 (void)chgproccnt(p->p_ucred->cr_ruidinfo, -1, 0);
887
888 /*
889 * Destroy resource accounting information associated with the process.
890 */
891 #ifdef RACCT
892 PROC_LOCK(p);
893 racct_sub(p, RACCT_NPROC, 1);
894 PROC_UNLOCK(p);
895 #endif
896 racct_proc_exit(p);
897
898 /*
899 * Free credentials, arguments, and sigacts.
900 */
901 crfree(p->p_ucred);
902 p->p_ucred = NULL;
903 pargs_drop(p->p_args);
904 p->p_args = NULL;
905 sigacts_free(p->p_sigacts);
906 p->p_sigacts = NULL;
907
908 /*
909 * Do any thread-system specific cleanups.
910 */
911 thread_wait(p);
912
913 /*
914 * Give vm and machine-dependent layer a chance to free anything that
915 * cpu_exit couldn't release while still running in process context.
916 */
917 vm_waitproc(p);
918 #ifdef MAC
919 mac_proc_destroy(p);
920 #endif
921 KASSERT(FIRST_THREAD_IN_PROC(p),
922 ("proc_reap: no residual thread!"));
923 uma_zfree(proc_zone, p);
924 sx_xlock(&allproc_lock);
925 nprocs--;
926 sx_xunlock(&allproc_lock);
927 }
928
929 static int
930 proc_to_reap(struct thread *td, struct proc *p, idtype_t idtype, id_t id,
931 int *status, int options, struct __wrusage *wrusage, siginfo_t *siginfo,
932 int check_only)
933 {
934 struct proc *q;
935 struct rusage *rup;
936
937 sx_assert(&proctree_lock, SA_XLOCKED);
938
939 q = td->td_proc;
940 PROC_LOCK(p);
941
942 switch (idtype) {
943 case P_ALL:
944 break;
945 case P_PID:
946 if (p->p_pid != (pid_t)id) {
947 PROC_UNLOCK(p);
948 return (0);
949 }
950 break;
951 case P_PGID:
952 if (p->p_pgid != (pid_t)id) {
953 PROC_UNLOCK(p);
954 return (0);
955 }
956 break;
957 case P_SID:
958 if (p->p_session->s_sid != (pid_t)id) {
959 PROC_UNLOCK(p);
960 return (0);
961 }
962 break;
963 case P_UID:
964 if (p->p_ucred->cr_uid != (uid_t)id) {
965 PROC_UNLOCK(p);
966 return (0);
967 }
968 break;
969 case P_GID:
970 if (p->p_ucred->cr_gid != (gid_t)id) {
971 PROC_UNLOCK(p);
972 return (0);
973 }
974 break;
975 case P_JAILID:
976 if (p->p_ucred->cr_prison == NULL ||
977 (p->p_ucred->cr_prison->pr_id != (int)id)) {
978 PROC_UNLOCK(p);
979 return (0);
980 }
981 break;
982 /*
983 * It seems that the thread structures get zeroed out
984 * at process exit. This makes it impossible to
985 * support P_SETID, P_CID or P_CPUID.
986 */
987 default:
988 PROC_UNLOCK(p);
989 return (0);
990 }
991
992 if (p_canwait(td, p)) {
993 PROC_UNLOCK(p);
994 return (0);
995 }
996
997 if (((options & WEXITED) == 0) && (p->p_state == PRS_ZOMBIE)) {
998 PROC_UNLOCK(p);
999 return (0);
1000 }
1001
1002 /*
1003 * This special case handles a kthread spawned by linux_clone
1004 * (see linux_misc.c). The linux_wait4 and linux_waitpid
1005 * functions need to be able to distinguish between waiting
1006 * on a process and waiting on a thread. It is a thread if
1007 * p_sigparent is not SIGCHLD, and the WLINUXCLONE option
1008 * signifies we want to wait for threads and not processes.
1009 */
1010 if ((p->p_sigparent != SIGCHLD) ^
1011 ((options & WLINUXCLONE) != 0)) {
1012 PROC_UNLOCK(p);
1013 return (0);
1014 }
1015
1016 PROC_SLOCK(p);
1017
1018 if (siginfo != NULL) {
1019 bzero(siginfo, sizeof(*siginfo));
1020 siginfo->si_errno = 0;
1021
1022 /*
1023 * SUSv4 requires that the si_signo value is always
1024 * SIGCHLD. Obey it despite the rfork(2) interface
1025 * allows to request other signal for child exit
1026 * notification.
1027 */
1028 siginfo->si_signo = SIGCHLD;
1029
1030 /*
1031 * This is still a rough estimate. We will fix the
1032 * cases TRAPPED, STOPPED, and CONTINUED later.
1033 */
1034 if (WCOREDUMP(p->p_xstat)) {
1035 siginfo->si_code = CLD_DUMPED;
1036 siginfo->si_status = WTERMSIG(p->p_xstat);
1037 } else if (WIFSIGNALED(p->p_xstat)) {
1038 siginfo->si_code = CLD_KILLED;
1039 siginfo->si_status = WTERMSIG(p->p_xstat);
1040 } else {
1041 siginfo->si_code = CLD_EXITED;
1042 siginfo->si_status = WEXITSTATUS(p->p_xstat);
1043 }
1044
1045 siginfo->si_pid = p->p_pid;
1046 siginfo->si_uid = p->p_ucred->cr_uid;
1047
1048 /*
1049 * The si_addr field would be useful additional
1050 * detail, but apparently the PC value may be lost
1051 * when we reach this point. bzero() above sets
1052 * siginfo->si_addr to NULL.
1053 */
1054 }
1055
1056 /*
1057 * There should be no reason to limit resources usage info to
1058 * exited processes only. A snapshot about any resources used
1059 * by a stopped process may be exactly what is needed.
1060 */
1061 if (wrusage != NULL) {
1062 rup = &wrusage->wru_self;
1063 *rup = p->p_ru;
1064 calcru(p, &rup->ru_utime, &rup->ru_stime);
1065
1066 rup = &wrusage->wru_children;
1067 *rup = p->p_stats->p_cru;
1068 calccru(p, &rup->ru_utime, &rup->ru_stime);
1069 }
1070
1071 if (p->p_state == PRS_ZOMBIE && !check_only) {
1072 proc_reap(td, p, status, options);
1073 return (-1);
1074 }
1075 PROC_SUNLOCK(p);
1076 PROC_UNLOCK(p);
1077 return (1);
1078 }
1079
1080 int
1081 kern_wait(struct thread *td, pid_t pid, int *status, int options,
1082 struct rusage *rusage)
1083 {
1084 struct __wrusage wru, *wrup;
1085 idtype_t idtype;
1086 id_t id;
1087 int ret;
1088
1089 /*
1090 * Translate the special pid values into the (idtype, pid)
1091 * pair for kern_wait6. The WAIT_MYPGRP case is handled by
1092 * kern_wait6() on its own.
1093 */
1094 if (pid == WAIT_ANY) {
1095 idtype = P_ALL;
1096 id = 0;
1097 } else if (pid < 0) {
1098 idtype = P_PGID;
1099 id = (id_t)-pid;
1100 } else {
1101 idtype = P_PID;
1102 id = (id_t)pid;
1103 }
1104
1105 if (rusage != NULL)
1106 wrup = &wru;
1107 else
1108 wrup = NULL;
1109
1110 /*
1111 * For backward compatibility we implicitly add flags WEXITED
1112 * and WTRAPPED here.
1113 */
1114 options |= WEXITED | WTRAPPED;
1115 ret = kern_wait6(td, idtype, id, status, options, wrup, NULL);
1116 if (rusage != NULL)
1117 *rusage = wru.wru_self;
1118 return (ret);
1119 }
1120
1121 int
1122 kern_wait6(struct thread *td, idtype_t idtype, id_t id, int *status,
1123 int options, struct __wrusage *wrusage, siginfo_t *siginfo)
1124 {
1125 struct proc *p, *q;
1126 int error, nfound, ret;
1127
1128 AUDIT_ARG_VALUE((int)idtype); /* XXX - This is likely wrong! */
1129 AUDIT_ARG_PID((pid_t)id); /* XXX - This may be wrong! */
1130 AUDIT_ARG_VALUE(options);
1131
1132 q = td->td_proc;
1133
1134 if ((pid_t)id == WAIT_MYPGRP && (idtype == P_PID || idtype == P_PGID)) {
1135 PROC_LOCK(q);
1136 id = (id_t)q->p_pgid;
1137 PROC_UNLOCK(q);
1138 idtype = P_PGID;
1139 }
1140
1141 /* If we don't know the option, just return. */
1142 if ((options & ~(WUNTRACED | WNOHANG | WCONTINUED | WNOWAIT |
1143 WEXITED | WTRAPPED | WLINUXCLONE)) != 0)
1144 return (EINVAL);
1145 if ((options & (WEXITED | WUNTRACED | WCONTINUED | WTRAPPED)) == 0) {
1146 /*
1147 * We will be unable to find any matching processes,
1148 * because there are no known events to look for.
1149 * Prefer to return error instead of blocking
1150 * indefinitely.
1151 */
1152 return (EINVAL);
1153 }
1154
1155 loop:
1156 if (q->p_flag & P_STATCHILD) {
1157 PROC_LOCK(q);
1158 q->p_flag &= ~P_STATCHILD;
1159 PROC_UNLOCK(q);
1160 }
1161 nfound = 0;
1162 sx_xlock(&proctree_lock);
1163 LIST_FOREACH(p, &q->p_children, p_sibling) {
1164 ret = proc_to_reap(td, p, idtype, id, status, options,
1165 wrusage, siginfo, 0);
1166 if (ret == 0)
1167 continue;
1168 else if (ret == 1)
1169 nfound++;
1170 else
1171 return (0);
1172
1173 PROC_LOCK(p);
1174 PROC_SLOCK(p);
1175
1176 if ((options & WTRAPPED) != 0 &&
1177 (p->p_flag & P_TRACED) != 0 &&
1178 (p->p_flag & (P_STOPPED_TRACE | P_STOPPED_SIG)) != 0 &&
1179 (p->p_suspcount == p->p_numthreads) &&
1180 ((p->p_flag & P_WAITED) == 0)) {
1181 PROC_SUNLOCK(p);
1182 if ((options & WNOWAIT) == 0)
1183 p->p_flag |= P_WAITED;
1184 sx_xunlock(&proctree_lock);
1185 td->td_retval[0] = p->p_pid;
1186
1187 if (status != NULL)
1188 *status = W_STOPCODE(p->p_xstat);
1189 if (siginfo != NULL) {
1190 siginfo->si_status = p->p_xstat;
1191 siginfo->si_code = CLD_TRAPPED;
1192 }
1193 if ((options & WNOWAIT) == 0) {
1194 PROC_LOCK(q);
1195 sigqueue_take(p->p_ksi);
1196 PROC_UNLOCK(q);
1197 }
1198
1199 CTR4(KTR_PTRACE,
1200 "wait: returning trapped pid %d status %#x (xstat %d) xthread %d",
1201 p->p_pid, W_STOPCODE(p->p_xstat), p->p_xstat,
1202 p->p_xthread != NULL ? p->p_xthread->td_tid : -1);
1203 PROC_UNLOCK(p);
1204 return (0);
1205 }
1206 if ((options & WUNTRACED) != 0 &&
1207 (p->p_flag & P_STOPPED_SIG) != 0 &&
1208 (p->p_suspcount == p->p_numthreads) &&
1209 ((p->p_flag & P_WAITED) == 0)) {
1210 PROC_SUNLOCK(p);
1211 if ((options & WNOWAIT) == 0)
1212 p->p_flag |= P_WAITED;
1213 sx_xunlock(&proctree_lock);
1214 td->td_retval[0] = p->p_pid;
1215
1216 if (status != NULL)
1217 *status = W_STOPCODE(p->p_xstat);
1218 if (siginfo != NULL) {
1219 siginfo->si_status = p->p_xstat;
1220 siginfo->si_code = CLD_STOPPED;
1221 }
1222 if ((options & WNOWAIT) == 0) {
1223 PROC_LOCK(q);
1224 sigqueue_take(p->p_ksi);
1225 PROC_UNLOCK(q);
1226 }
1227
1228 PROC_UNLOCK(p);
1229 return (0);
1230 }
1231 PROC_SUNLOCK(p);
1232 if ((options & WCONTINUED) != 0 &&
1233 (p->p_flag & P_CONTINUED) != 0) {
1234 sx_xunlock(&proctree_lock);
1235 td->td_retval[0] = p->p_pid;
1236 if ((options & WNOWAIT) == 0) {
1237 p->p_flag &= ~P_CONTINUED;
1238 PROC_LOCK(q);
1239 sigqueue_take(p->p_ksi);
1240 PROC_UNLOCK(q);
1241 }
1242 PROC_UNLOCK(p);
1243
1244 if (status != NULL)
1245 *status = SIGCONT;
1246 if (siginfo != NULL) {
1247 siginfo->si_status = SIGCONT;
1248 siginfo->si_code = CLD_CONTINUED;
1249 }
1250 return (0);
1251 }
1252 PROC_UNLOCK(p);
1253 }
1254
1255 /*
1256 * Look in the orphans list too, to allow the parent to
1257 * collect it's child exit status even if child is being
1258 * debugged.
1259 *
1260 * Debugger detaches from the parent upon successful
1261 * switch-over from parent to child. At this point due to
1262 * re-parenting the parent loses the child to debugger and a
1263 * wait4(2) call would report that it has no children to wait
1264 * for. By maintaining a list of orphans we allow the parent
1265 * to successfully wait until the child becomes a zombie.
1266 */
1267 if (nfound == 0) {
1268 LIST_FOREACH(p, &q->p_orphans, p_orphan) {
1269 ret = proc_to_reap(td, p, idtype, id, NULL, options,
1270 NULL, NULL, 1);
1271 if (ret != 0) {
1272 KASSERT(ret != -1, ("reaped an orphan (pid %d)",
1273 (int)td->td_retval[0]));
1274 nfound++;
1275 break;
1276 }
1277 }
1278 }
1279 if (nfound == 0) {
1280 sx_xunlock(&proctree_lock);
1281 return (ECHILD);
1282 }
1283 if (options & WNOHANG) {
1284 sx_xunlock(&proctree_lock);
1285 td->td_retval[0] = 0;
1286 return (0);
1287 }
1288 PROC_LOCK(q);
1289 sx_xunlock(&proctree_lock);
1290 if (q->p_flag & P_STATCHILD) {
1291 q->p_flag &= ~P_STATCHILD;
1292 error = 0;
1293 } else
1294 error = msleep(q, &q->p_mtx, PWAIT | PCATCH, "wait", 0);
1295 PROC_UNLOCK(q);
1296 if (error)
1297 return (error);
1298 goto loop;
1299 }
1300
1301 /*
1302 * Make process 'parent' the new parent of process 'child'.
1303 * Must be called with an exclusive hold of proctree lock.
1304 */
1305 void
1306 proc_reparent(struct proc *child, struct proc *parent)
1307 {
1308
1309 sx_assert(&proctree_lock, SX_XLOCKED);
1310 PROC_LOCK_ASSERT(child, MA_OWNED);
1311 if (child->p_pptr == parent)
1312 return;
1313
1314 PROC_LOCK(child->p_pptr);
1315 sigqueue_take(child->p_ksi);
1316 PROC_UNLOCK(child->p_pptr);
1317 LIST_REMOVE(child, p_sibling);
1318 LIST_INSERT_HEAD(&parent->p_children, child, p_sibling);
1319
1320 clear_orphan(child);
1321 if (child->p_flag & P_TRACED) {
1322 if (LIST_EMPTY(&child->p_pptr->p_orphans)) {
1323 child->p_treeflag |= P_TREE_FIRST_ORPHAN;
1324 LIST_INSERT_HEAD(&child->p_pptr->p_orphans, child,
1325 p_orphan);
1326 } else {
1327 LIST_INSERT_AFTER(LIST_FIRST(&child->p_pptr->p_orphans),
1328 child, p_orphan);
1329 }
1330 child->p_treeflag |= P_TREE_ORPHANED;
1331 }
1332
1333 child->p_pptr = parent;
1334 }
Cache object: f534a5c72842e6b634055051e3ca4d5d
|