xref: /freebsd/sys/kern/kern_exit.c (revision 10b59a9b4add0320d52c15ce057dd697261e7dfc)
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_DEFINE(proc, kernel, , exit, exit);
98 SDT_PROBE_ARGTYPE(proc, kernel, , exit, 0, "int");
99 
100 /* Hook for NFS teardown procedure. */
101 void (*nlminfo_release_p)(struct proc *p);
102 
103 /*
104  * exit -- death of process.
105  */
106 void
107 sys_sys_exit(struct thread *td, struct sys_exit_args *uap)
108 {
109 
110 	exit1(td, W_EXITCODE(uap->rval, 0));
111 	/* NOTREACHED */
112 }
113 
114 /*
115  * Exit: deallocate address space and other resources, change proc state to
116  * zombie, and unlink proc from allproc and parent's lists.  Save exit status
117  * and rusage for wait().  Check for child processes and orphan them.
118  */
119 void
120 exit1(struct thread *td, int rv)
121 {
122 	struct proc *p, *nq, *q;
123 	struct vnode *vtmp;
124 	struct vnode *ttyvp = NULL;
125 	struct plimit *plim;
126 	int locked;
127 
128 	mtx_assert(&Giant, MA_NOTOWNED);
129 
130 	p = td->td_proc;
131 	/*
132 	 * XXX in case we're rebooting we just let init die in order to
133 	 * work around an unsolved stack overflow seen very late during
134 	 * shutdown on sparc64 when the gmirror worker process exists.
135 	 */
136 	if (p == initproc && rebooting == 0) {
137 		printf("init died (signal %d, exit %d)\n",
138 		    WTERMSIG(rv), WEXITSTATUS(rv));
139 		panic("Going nowhere without my init!");
140 	}
141 
142 	/*
143 	 * MUST abort all other threads before proceeding past here.
144 	 */
145 	PROC_LOCK(p);
146 	while (p->p_flag & P_HADTHREADS) {
147 		/*
148 		 * First check if some other thread got here before us..
149 		 * if so, act apropriatly, (exit or suspend);
150 		 */
151 		thread_suspend_check(0);
152 
153 		/*
154 		 * Kill off the other threads. This requires
155 		 * some co-operation from other parts of the kernel
156 		 * so it may not be instantaneous.  With this state set
157 		 * any thread entering the kernel from userspace will
158 		 * thread_exit() in trap().  Any thread attempting to
159 		 * sleep will return immediately with EINTR or EWOULDBLOCK
160 		 * which will hopefully force them to back out to userland
161 		 * freeing resources as they go.  Any thread attempting
162 		 * to return to userland will thread_exit() from userret().
163 		 * thread_exit() will unsuspend us when the last of the
164 		 * other threads exits.
165 		 * If there is already a thread singler after resumption,
166 		 * calling thread_single will fail; in that case, we just
167 		 * re-check all suspension request, the thread should
168 		 * either be suspended there or exit.
169 		 */
170 		if (! thread_single(SINGLE_EXIT))
171 			break;
172 
173 		/*
174 		 * All other activity in this process is now stopped.
175 		 * Threading support has been turned off.
176 		 */
177 	}
178 	KASSERT(p->p_numthreads == 1,
179 	    ("exit1: proc %p exiting with %d threads", p, p->p_numthreads));
180 	racct_sub(p, RACCT_NTHR, 1);
181 	/*
182 	 * Wakeup anyone in procfs' PIOCWAIT.  They should have a hold
183 	 * on our vmspace, so we should block below until they have
184 	 * released their reference to us.  Note that if they have
185 	 * requested S_EXIT stops we will block here until they ack
186 	 * via PIOCCONT.
187 	 */
188 	_STOPEVENT(p, S_EXIT, rv);
189 
190 	/*
191 	 * Note that we are exiting and do another wakeup of anyone in
192 	 * PIOCWAIT in case they aren't listening for S_EXIT stops or
193 	 * decided to wait again after we told them we are exiting.
194 	 */
195 	p->p_flag |= P_WEXIT;
196 	wakeup(&p->p_stype);
197 
198 	/*
199 	 * Wait for any processes that have a hold on our vmspace to
200 	 * release their reference.
201 	 */
202 	while (p->p_lock > 0)
203 		msleep(&p->p_lock, &p->p_mtx, PWAIT, "exithold", 0);
204 
205 	p->p_xstat = rv;	/* Let event handler change exit status */
206 	PROC_UNLOCK(p);
207 	/* Drain the limit callout while we don't have the proc locked */
208 	callout_drain(&p->p_limco);
209 
210 #ifdef AUDIT
211 	/*
212 	 * The Sun BSM exit token contains two components: an exit status as
213 	 * passed to exit(), and a return value to indicate what sort of exit
214 	 * it was.  The exit status is WEXITSTATUS(rv), but it's not clear
215 	 * what the return value is.
216 	 */
217 	AUDIT_ARG_EXIT(WEXITSTATUS(rv), 0);
218 	AUDIT_SYSCALL_EXIT(0, td);
219 #endif
220 
221 	/* Are we a task leader? */
222 	if (p == p->p_leader) {
223 		mtx_lock(&ppeers_lock);
224 		q = p->p_peers;
225 		while (q != NULL) {
226 			PROC_LOCK(q);
227 			kern_psignal(q, SIGKILL);
228 			PROC_UNLOCK(q);
229 			q = q->p_peers;
230 		}
231 		while (p->p_peers != NULL)
232 			msleep(p, &ppeers_lock, PWAIT, "exit1", 0);
233 		mtx_unlock(&ppeers_lock);
234 	}
235 
236 	/*
237 	 * Check if any loadable modules need anything done at process exit.
238 	 * E.g. SYSV IPC stuff
239 	 * XXX what if one of these generates an error?
240 	 */
241 	EVENTHANDLER_INVOKE(process_exit, p);
242 
243 	/*
244 	 * If parent is waiting for us to exit or exec,
245 	 * P_PPWAIT is set; we will wakeup the parent below.
246 	 */
247 	PROC_LOCK(p);
248 	rv = p->p_xstat;	/* Event handler could change exit status */
249 	stopprofclock(p);
250 	p->p_flag &= ~(P_TRACED | P_PPWAIT);
251 
252 	/*
253 	 * Stop the real interval timer.  If the handler is currently
254 	 * executing, prevent it from rearming itself and let it finish.
255 	 */
256 	if (timevalisset(&p->p_realtimer.it_value) &&
257 	    callout_stop(&p->p_itcallout) == 0) {
258 		timevalclear(&p->p_realtimer.it_interval);
259 		msleep(&p->p_itcallout, &p->p_mtx, PWAIT, "ritwait", 0);
260 		KASSERT(!timevalisset(&p->p_realtimer.it_value),
261 		    ("realtime timer is still armed"));
262 	}
263 	PROC_UNLOCK(p);
264 
265 	/*
266 	 * Reset any sigio structures pointing to us as a result of
267 	 * F_SETOWN with our pid.
268 	 */
269 	funsetownlst(&p->p_sigiolst);
270 
271 	/*
272 	 * If this process has an nlminfo data area (for lockd), release it
273 	 */
274 	if (nlminfo_release_p != NULL && p->p_nlminfo != NULL)
275 		(*nlminfo_release_p)(p);
276 
277 	/*
278 	 * Close open files and release open-file table.
279 	 * This may block!
280 	 */
281 	fdfree(td);
282 
283 	/*
284 	 * If this thread tickled GEOM, we need to wait for the giggling to
285 	 * stop before we return to userland
286 	 */
287 	if (td->td_pflags & TDP_GEOM)
288 		g_waitidle();
289 
290 	/*
291 	 * Remove ourself from our leader's peer list and wake our leader.
292 	 */
293 	mtx_lock(&ppeers_lock);
294 	if (p->p_leader->p_peers) {
295 		q = p->p_leader;
296 		while (q->p_peers != p)
297 			q = q->p_peers;
298 		q->p_peers = p->p_peers;
299 		wakeup(p->p_leader);
300 	}
301 	mtx_unlock(&ppeers_lock);
302 
303 	vmspace_exit(td);
304 
305 	sx_xlock(&proctree_lock);
306 	if (SESS_LEADER(p)) {
307 		struct session *sp = p->p_session;
308 		struct tty *tp;
309 
310 		/*
311 		 * s_ttyp is not zero'd; we use this to indicate that
312 		 * the session once had a controlling terminal. (for
313 		 * logging and informational purposes)
314 		 */
315 		SESS_LOCK(sp);
316 		ttyvp = sp->s_ttyvp;
317 		tp = sp->s_ttyp;
318 		sp->s_ttyvp = NULL;
319 		sp->s_ttydp = NULL;
320 		sp->s_leader = NULL;
321 		SESS_UNLOCK(sp);
322 
323 		/*
324 		 * Signal foreground pgrp and revoke access to
325 		 * controlling terminal if it has not been revoked
326 		 * already.
327 		 *
328 		 * Because the TTY may have been revoked in the mean
329 		 * time and could already have a new session associated
330 		 * with it, make sure we don't send a SIGHUP to a
331 		 * foreground process group that does not belong to this
332 		 * session.
333 		 */
334 
335 		if (tp != NULL) {
336 			tty_lock(tp);
337 			if (tp->t_session == sp)
338 				tty_signal_pgrp(tp, SIGHUP);
339 			tty_unlock(tp);
340 		}
341 
342 		if (ttyvp != NULL) {
343 			sx_xunlock(&proctree_lock);
344 			if (vn_lock(ttyvp, LK_EXCLUSIVE) == 0) {
345 				VOP_REVOKE(ttyvp, REVOKEALL);
346 				VOP_UNLOCK(ttyvp, 0);
347 			}
348 			sx_xlock(&proctree_lock);
349 		}
350 	}
351 	fixjobc(p, p->p_pgrp, 0);
352 	sx_xunlock(&proctree_lock);
353 	(void)acct_process(td);
354 
355 	/* Release the TTY now we've unlocked everything. */
356 	if (ttyvp != NULL)
357 		vrele(ttyvp);
358 #ifdef KTRACE
359 	ktrprocexit(td);
360 #endif
361 	/*
362 	 * Release reference to text vnode
363 	 */
364 	if ((vtmp = p->p_textvp) != NULL) {
365 		p->p_textvp = NULL;
366 		locked = VFS_LOCK_GIANT(vtmp->v_mount);
367 		vrele(vtmp);
368 		VFS_UNLOCK_GIANT(locked);
369 	}
370 
371 	/*
372 	 * Release our limits structure.
373 	 */
374 	PROC_LOCK(p);
375 	plim = p->p_limit;
376 	p->p_limit = NULL;
377 	PROC_UNLOCK(p);
378 	lim_free(plim);
379 
380 	tidhash_remove(td);
381 
382 	/*
383 	 * Remove proc from allproc queue and pidhash chain.
384 	 * Place onto zombproc.  Unlink from parent's child list.
385 	 */
386 	sx_xlock(&allproc_lock);
387 	LIST_REMOVE(p, p_list);
388 	LIST_INSERT_HEAD(&zombproc, p, p_list);
389 	LIST_REMOVE(p, p_hash);
390 	sx_xunlock(&allproc_lock);
391 
392 	/*
393 	 * Call machine-dependent code to release any
394 	 * machine-dependent resources other than the address space.
395 	 * The address space is released by "vmspace_exitfree(p)" in
396 	 * vm_waitproc().
397 	 */
398 	cpu_exit(td);
399 
400 	WITNESS_WARN(WARN_PANIC, NULL, "process (pid %d) exiting", p->p_pid);
401 
402 	/*
403 	 * Reparent all of our children to init.
404 	 */
405 	sx_xlock(&proctree_lock);
406 	q = LIST_FIRST(&p->p_children);
407 	if (q != NULL)		/* only need this if any child is S_ZOMB */
408 		wakeup(initproc);
409 	for (; q != NULL; q = nq) {
410 		nq = LIST_NEXT(q, p_sibling);
411 		PROC_LOCK(q);
412 		proc_reparent(q, initproc);
413 		q->p_sigparent = SIGCHLD;
414 		/*
415 		 * Traced processes are killed
416 		 * since their existence means someone is screwing up.
417 		 */
418 		if (q->p_flag & P_TRACED) {
419 			struct thread *temp;
420 
421 			q->p_flag &= ~(P_TRACED | P_STOPPED_TRACE);
422 			FOREACH_THREAD_IN_PROC(q, temp)
423 				temp->td_dbgflags &= ~TDB_SUSPEND;
424 			kern_psignal(q, SIGKILL);
425 		}
426 		PROC_UNLOCK(q);
427 	}
428 
429 	/* Save exit status. */
430 	PROC_LOCK(p);
431 	p->p_xthread = td;
432 
433 	/* Tell the prison that we are gone. */
434 	prison_proc_free(p->p_ucred->cr_prison);
435 
436 #ifdef KDTRACE_HOOKS
437 	/*
438 	 * Tell the DTrace fasttrap provider about the exit if it
439 	 * has declared an interest.
440 	 */
441 	if (dtrace_fasttrap_exit)
442 		dtrace_fasttrap_exit(p);
443 #endif
444 
445 	/*
446 	 * Notify interested parties of our demise.
447 	 */
448 	KNOTE_LOCKED(&p->p_klist, NOTE_EXIT);
449 
450 #ifdef KDTRACE_HOOKS
451 	int reason = CLD_EXITED;
452 	if (WCOREDUMP(rv))
453 		reason = CLD_DUMPED;
454 	else if (WIFSIGNALED(rv))
455 		reason = CLD_KILLED;
456 	SDT_PROBE(proc, kernel, , exit, reason, 0, 0, 0, 0);
457 #endif
458 
459 	/*
460 	 * Just delete all entries in the p_klist. At this point we won't
461 	 * report any more events, and there are nasty race conditions that
462 	 * can beat us if we don't.
463 	 */
464 	knlist_clear(&p->p_klist, 1);
465 
466 	/*
467 	 * If this is a process with a descriptor, we may not need to deliver
468 	 * a signal to the parent.  proctree_lock is held over
469 	 * procdesc_exit() to serialize concurrent calls to close() and
470 	 * exit().
471 	 */
472 #ifdef PROCDESC
473 	if (p->p_procdesc == NULL || procdesc_exit(p)) {
474 #endif
475 		/*
476 		 * Notify parent that we're gone.  If parent has the
477 		 * PS_NOCLDWAIT flag set, or if the handler is set to SIG_IGN,
478 		 * notify process 1 instead (and hope it will handle this
479 		 * situation).
480 		 */
481 		PROC_LOCK(p->p_pptr);
482 		mtx_lock(&p->p_pptr->p_sigacts->ps_mtx);
483 		if (p->p_pptr->p_sigacts->ps_flag &
484 		    (PS_NOCLDWAIT | PS_CLDSIGIGN)) {
485 			struct proc *pp;
486 
487 			mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx);
488 			pp = p->p_pptr;
489 			PROC_UNLOCK(pp);
490 			proc_reparent(p, initproc);
491 			p->p_sigparent = SIGCHLD;
492 			PROC_LOCK(p->p_pptr);
493 
494 			/*
495 			 * Notify parent, so in case he was wait(2)ing or
496 			 * executing waitpid(2) with our pid, he will
497 			 * continue.
498 			 */
499 			wakeup(pp);
500 		} else
501 			mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx);
502 
503 		if (p->p_pptr == initproc)
504 			kern_psignal(p->p_pptr, SIGCHLD);
505 		else if (p->p_sigparent != 0) {
506 			if (p->p_sigparent == SIGCHLD)
507 				childproc_exited(p);
508 			else	/* LINUX thread */
509 				kern_psignal(p->p_pptr, p->p_sigparent);
510 		}
511 #ifdef PROCDESC
512 	} else
513 		PROC_LOCK(p->p_pptr);
514 #endif
515 	sx_xunlock(&proctree_lock);
516 
517 	/*
518 	 * The state PRS_ZOMBIE prevents other proesses from sending
519 	 * signal to the process, to avoid memory leak, we free memory
520 	 * for signal queue at the time when the state is set.
521 	 */
522 	sigqueue_flush(&p->p_sigqueue);
523 	sigqueue_flush(&td->td_sigqueue);
524 
525 	/*
526 	 * We have to wait until after acquiring all locks before
527 	 * changing p_state.  We need to avoid all possible context
528 	 * switches (including ones from blocking on a mutex) while
529 	 * marked as a zombie.  We also have to set the zombie state
530 	 * before we release the parent process' proc lock to avoid
531 	 * a lost wakeup.  So, we first call wakeup, then we grab the
532 	 * sched lock, update the state, and release the parent process'
533 	 * proc lock.
534 	 */
535 	wakeup(p->p_pptr);
536 	cv_broadcast(&p->p_pwait);
537 	sched_exit(p->p_pptr, td);
538 	PROC_SLOCK(p);
539 	p->p_state = PRS_ZOMBIE;
540 	PROC_UNLOCK(p->p_pptr);
541 
542 	/*
543 	 * Hopefully no one will try to deliver a signal to the process this
544 	 * late in the game.
545 	 */
546 	knlist_destroy(&p->p_klist);
547 
548 	/*
549 	 * Save our children's rusage information in our exit rusage.
550 	 */
551 	ruadd(&p->p_ru, &p->p_rux, &p->p_stats->p_cru, &p->p_crux);
552 
553 	/*
554 	 * Make sure the scheduler takes this thread out of its tables etc.
555 	 * This will also release this thread's reference to the ucred.
556 	 * Other thread parts to release include pcb bits and such.
557 	 */
558 	thread_exit();
559 }
560 
561 
562 #ifndef _SYS_SYSPROTO_H_
563 struct abort2_args {
564 	char *why;
565 	int nargs;
566 	void **args;
567 };
568 #endif
569 
570 int
571 sys_abort2(struct thread *td, struct abort2_args *uap)
572 {
573 	struct proc *p = td->td_proc;
574 	struct sbuf *sb;
575 	void *uargs[16];
576 	int error, i, sig;
577 
578 	/*
579 	 * Do it right now so we can log either proper call of abort2(), or
580 	 * note, that invalid argument was passed. 512 is big enough to
581 	 * handle 16 arguments' descriptions with additional comments.
582 	 */
583 	sb = sbuf_new(NULL, NULL, 512, SBUF_FIXEDLEN);
584 	sbuf_clear(sb);
585 	sbuf_printf(sb, "%s(pid %d uid %d) aborted: ",
586 	    p->p_comm, p->p_pid, td->td_ucred->cr_uid);
587 	/*
588 	 * Since we can't return from abort2(), send SIGKILL in cases, where
589 	 * abort2() was called improperly
590 	 */
591 	sig = SIGKILL;
592 	/* Prevent from DoSes from user-space. */
593 	if (uap->nargs < 0 || uap->nargs > 16)
594 		goto out;
595 	if (uap->nargs > 0) {
596 		if (uap->args == NULL)
597 			goto out;
598 		error = copyin(uap->args, uargs, uap->nargs * sizeof(void *));
599 		if (error != 0)
600 			goto out;
601 	}
602 	/*
603 	 * Limit size of 'reason' string to 128. Will fit even when
604 	 * maximal number of arguments was chosen to be logged.
605 	 */
606 	if (uap->why != NULL) {
607 		error = sbuf_copyin(sb, uap->why, 128);
608 		if (error < 0)
609 			goto out;
610 	} else {
611 		sbuf_printf(sb, "(null)");
612 	}
613 	if (uap->nargs > 0) {
614 		sbuf_printf(sb, "(");
615 		for (i = 0;i < uap->nargs; i++)
616 			sbuf_printf(sb, "%s%p", i == 0 ? "" : ", ", uargs[i]);
617 		sbuf_printf(sb, ")");
618 	}
619 	/*
620 	 * Final stage: arguments were proper, string has been
621 	 * successfully copied from userspace, and copying pointers
622 	 * from user-space succeed.
623 	 */
624 	sig = SIGABRT;
625 out:
626 	if (sig == SIGKILL) {
627 		sbuf_trim(sb);
628 		sbuf_printf(sb, " (Reason text inaccessible)");
629 	}
630 	sbuf_cat(sb, "\n");
631 	sbuf_finish(sb);
632 	log(LOG_INFO, "%s", sbuf_data(sb));
633 	sbuf_delete(sb);
634 	exit1(td, W_EXITCODE(0, sig));
635 	return (0);
636 }
637 
638 
639 #ifdef COMPAT_43
640 /*
641  * The dirty work is handled by kern_wait().
642  */
643 int
644 owait(struct thread *td, struct owait_args *uap __unused)
645 {
646 	int error, status;
647 
648 	error = kern_wait(td, WAIT_ANY, &status, 0, NULL);
649 	if (error == 0)
650 		td->td_retval[1] = status;
651 	return (error);
652 }
653 #endif /* COMPAT_43 */
654 
655 /*
656  * The dirty work is handled by kern_wait().
657  */
658 int
659 sys_wait4(struct thread *td, struct wait_args *uap)
660 {
661 	struct rusage ru, *rup;
662 	int error, status;
663 
664 	if (uap->rusage != NULL)
665 		rup = &ru;
666 	else
667 		rup = NULL;
668 	error = kern_wait(td, uap->pid, &status, uap->options, rup);
669 	if (uap->status != NULL && error == 0)
670 		error = copyout(&status, uap->status, sizeof(status));
671 	if (uap->rusage != NULL && error == 0)
672 		error = copyout(&ru, uap->rusage, sizeof(struct rusage));
673 	return (error);
674 }
675 
676 /*
677  * Reap the remains of a zombie process and optionally return status and
678  * rusage.  Asserts and will release both the proctree_lock and the process
679  * lock as part of its work.
680  */
681 void
682 proc_reap(struct thread *td, struct proc *p, int *status, int options,
683     struct rusage *rusage)
684 {
685 	struct proc *q, *t;
686 
687 	sx_assert(&proctree_lock, SA_XLOCKED);
688 	PROC_LOCK_ASSERT(p, MA_OWNED);
689 	PROC_SLOCK_ASSERT(p, MA_OWNED);
690 	KASSERT(p->p_state == PRS_ZOMBIE, ("proc_reap: !PRS_ZOMBIE"));
691 
692 	q = td->td_proc;
693 	if (rusage) {
694 		*rusage = p->p_ru;
695 		calcru(p, &rusage->ru_utime, &rusage->ru_stime);
696 	}
697 	PROC_SUNLOCK(p);
698 	td->td_retval[0] = p->p_pid;
699 	if (status)
700 		*status = p->p_xstat;	/* convert to int */
701 	if (options & WNOWAIT) {
702 		/*
703 		 *  Only poll, returning the status.  Caller does not wish to
704 		 * release the proc struct just yet.
705 		 */
706 		PROC_UNLOCK(p);
707 		sx_xunlock(&proctree_lock);
708 		return;
709 	}
710 
711 	PROC_LOCK(q);
712 	sigqueue_take(p->p_ksi);
713 	PROC_UNLOCK(q);
714 	PROC_UNLOCK(p);
715 
716 	/*
717 	 * If we got the child via a ptrace 'attach', we need to give it back
718 	 * to the old parent.
719 	 */
720 	if (p->p_oppid && (t = pfind(p->p_oppid)) != NULL) {
721 		PROC_LOCK(p);
722 		proc_reparent(p, t);
723 		p->p_pptr->p_dbg_child--;
724 		p->p_oppid = 0;
725 		PROC_UNLOCK(p);
726 		pksignal(t, SIGCHLD, p->p_ksi);
727 		wakeup(t);
728 		cv_broadcast(&p->p_pwait);
729 		PROC_UNLOCK(t);
730 		sx_xunlock(&proctree_lock);
731 		return;
732 	}
733 
734 	/*
735 	 * Remove other references to this process to ensure we have an
736 	 * exclusive reference.
737 	 */
738 	sx_xlock(&allproc_lock);
739 	LIST_REMOVE(p, p_list);	/* off zombproc */
740 	sx_xunlock(&allproc_lock);
741 	LIST_REMOVE(p, p_sibling);
742 	leavepgrp(p);
743 #ifdef PROCDESC
744 	if (p->p_procdesc != NULL)
745 		procdesc_reap(p);
746 #endif
747 	sx_xunlock(&proctree_lock);
748 
749 	/*
750 	 * As a side effect of this lock, we know that all other writes to
751 	 * this proc are visible now, so no more locking is needed for p.
752 	 */
753 	PROC_LOCK(p);
754 	p->p_xstat = 0;		/* XXX: why? */
755 	PROC_UNLOCK(p);
756 	PROC_LOCK(q);
757 	ruadd(&q->p_stats->p_cru, &q->p_crux, &p->p_ru, &p->p_rux);
758 	PROC_UNLOCK(q);
759 
760 	/*
761 	 * Decrement the count of procs running with this uid.
762 	 */
763 	(void)chgproccnt(p->p_ucred->cr_ruidinfo, -1, 0);
764 
765 	/*
766 	 * Destroy resource accounting information associated with the process.
767 	 */
768 #ifdef RACCT
769 	PROC_LOCK(p);
770 	racct_sub(p, RACCT_NPROC, 1);
771 	PROC_UNLOCK(p);
772 #endif
773 	racct_proc_exit(p);
774 
775 	/*
776 	 * Free credentials, arguments, and sigacts.
777 	 */
778 	crfree(p->p_ucred);
779 	p->p_ucred = NULL;
780 	pargs_drop(p->p_args);
781 	p->p_args = NULL;
782 	sigacts_free(p->p_sigacts);
783 	p->p_sigacts = NULL;
784 
785 	/*
786 	 * Do any thread-system specific cleanups.
787 	 */
788 	thread_wait(p);
789 
790 	/*
791 	 * Give vm and machine-dependent layer a chance to free anything that
792 	 * cpu_exit couldn't release while still running in process context.
793 	 */
794 	vm_waitproc(p);
795 #ifdef MAC
796 	mac_proc_destroy(p);
797 #endif
798 	KASSERT(FIRST_THREAD_IN_PROC(p),
799 	    ("proc_reap: no residual thread!"));
800 	uma_zfree(proc_zone, p);
801 	sx_xlock(&allproc_lock);
802 	nprocs--;
803 	sx_xunlock(&allproc_lock);
804 }
805 
806 int
807 kern_wait(struct thread *td, pid_t pid, int *status, int options,
808     struct rusage *rusage)
809 {
810 	struct proc *p, *q;
811 	int error, nfound;
812 
813 	AUDIT_ARG_PID(pid);
814 	AUDIT_ARG_VALUE(options);
815 
816 	q = td->td_proc;
817 	if (pid == 0) {
818 		PROC_LOCK(q);
819 		pid = -q->p_pgid;
820 		PROC_UNLOCK(q);
821 	}
822 	/* If we don't know the option, just return. */
823 	if (options & ~(WUNTRACED|WNOHANG|WCONTINUED|WNOWAIT|WLINUXCLONE))
824 		return (EINVAL);
825 loop:
826 	if (q->p_flag & P_STATCHILD) {
827 		PROC_LOCK(q);
828 		q->p_flag &= ~P_STATCHILD;
829 		PROC_UNLOCK(q);
830 	}
831 	nfound = 0;
832 	sx_xlock(&proctree_lock);
833 	LIST_FOREACH(p, &q->p_children, p_sibling) {
834 		PROC_LOCK(p);
835 		if (pid != WAIT_ANY &&
836 		    p->p_pid != pid && p->p_pgid != -pid) {
837 			PROC_UNLOCK(p);
838 			continue;
839 		}
840 		if (p_canwait(td, p)) {
841 			PROC_UNLOCK(p);
842 			continue;
843 		}
844 
845 		/*
846 		 * This special case handles a kthread spawned by linux_clone
847 		 * (see linux_misc.c).  The linux_wait4 and linux_waitpid
848 		 * functions need to be able to distinguish between waiting
849 		 * on a process and waiting on a thread.  It is a thread if
850 		 * p_sigparent is not SIGCHLD, and the WLINUXCLONE option
851 		 * signifies we want to wait for threads and not processes.
852 		 */
853 		if ((p->p_sigparent != SIGCHLD) ^
854 		    ((options & WLINUXCLONE) != 0)) {
855 			PROC_UNLOCK(p);
856 			continue;
857 		}
858 
859 		nfound++;
860 		PROC_SLOCK(p);
861 		if (p->p_state == PRS_ZOMBIE) {
862 			proc_reap(td, p, status, options, rusage);
863 			return (0);
864 		}
865 		if ((p->p_flag & P_STOPPED_SIG) &&
866 		    (p->p_suspcount == p->p_numthreads) &&
867 		    (p->p_flag & P_WAITED) == 0 &&
868 		    (p->p_flag & P_TRACED || options & WUNTRACED)) {
869 			PROC_SUNLOCK(p);
870 			p->p_flag |= P_WAITED;
871 			sx_xunlock(&proctree_lock);
872 			td->td_retval[0] = p->p_pid;
873 			if (status)
874 				*status = W_STOPCODE(p->p_xstat);
875 
876 			PROC_LOCK(q);
877 			sigqueue_take(p->p_ksi);
878 			PROC_UNLOCK(q);
879 			PROC_UNLOCK(p);
880 
881 			return (0);
882 		}
883 		PROC_SUNLOCK(p);
884 		if (options & WCONTINUED && (p->p_flag & P_CONTINUED)) {
885 			sx_xunlock(&proctree_lock);
886 			td->td_retval[0] = p->p_pid;
887 			p->p_flag &= ~P_CONTINUED;
888 
889 			PROC_LOCK(q);
890 			sigqueue_take(p->p_ksi);
891 			PROC_UNLOCK(q);
892 			PROC_UNLOCK(p);
893 
894 			if (status)
895 				*status = SIGCONT;
896 			return (0);
897 		}
898 		PROC_UNLOCK(p);
899 	}
900 	if (nfound == 0) {
901 		sx_xunlock(&proctree_lock);
902 		if (td->td_proc->p_dbg_child)
903 			return (0);
904 		else
905 			return (ECHILD);
906 	}
907 	if (options & WNOHANG) {
908 		sx_xunlock(&proctree_lock);
909 		td->td_retval[0] = 0;
910 		return (0);
911 	}
912 	PROC_LOCK(q);
913 	sx_xunlock(&proctree_lock);
914 	if (q->p_flag & P_STATCHILD) {
915 		q->p_flag &= ~P_STATCHILD;
916 		error = 0;
917 	} else
918 		error = msleep(q, &q->p_mtx, PWAIT | PCATCH, "wait", 0);
919 	PROC_UNLOCK(q);
920 	if (error)
921 		return (error);
922 	goto loop;
923 }
924 
925 /*
926  * Make process 'parent' the new parent of process 'child'.
927  * Must be called with an exclusive hold of proctree lock.
928  */
929 void
930 proc_reparent(struct proc *child, struct proc *parent)
931 {
932 
933 	sx_assert(&proctree_lock, SX_XLOCKED);
934 	PROC_LOCK_ASSERT(child, MA_OWNED);
935 	if (child->p_pptr == parent)
936 		return;
937 
938 	PROC_LOCK(child->p_pptr);
939 	sigqueue_take(child->p_ksi);
940 	PROC_UNLOCK(child->p_pptr);
941 	LIST_REMOVE(child, p_sibling);
942 	LIST_INSERT_HEAD(&parent->p_children, child, p_sibling);
943 	child->p_pptr = parent;
944 }
945