xref: /freebsd/sys/kern/kern_exit.c (revision d2387d42b8da231a5b95cbc313825fb2aadf26f6)
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  * 3. All advertising materials mentioning features or use of this software
19  *    must display the following acknowledgement:
20  *	This product includes software developed by the University of
21  *	California, Berkeley and its contributors.
22  * 4. Neither the name of the University nor the names of its contributors
23  *    may be used to endorse or promote products derived from this software
24  *    without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36  * SUCH DAMAGE.
37  *
38  *	@(#)kern_exit.c	8.7 (Berkeley) 2/12/94
39  */
40 
41 #include <sys/cdefs.h>
42 __FBSDID("$FreeBSD$");
43 
44 #include "opt_compat.h"
45 #include "opt_ktrace.h"
46 #include "opt_mac.h"
47 
48 #include <sys/param.h>
49 #include <sys/systm.h>
50 #include <sys/sysproto.h>
51 #include <sys/eventhandler.h>
52 #include <sys/kernel.h>
53 #include <sys/malloc.h>
54 #include <sys/lock.h>
55 #include <sys/mutex.h>
56 #include <sys/proc.h>
57 #include <sys/pioctl.h>
58 #include <sys/tty.h>
59 #include <sys/wait.h>
60 #include <sys/vmmeter.h>
61 #include <sys/vnode.h>
62 #include <sys/resourcevar.h>
63 #include <sys/signalvar.h>
64 #include <sys/sched.h>
65 #include <sys/sx.h>
66 #include <sys/ptrace.h>
67 #include <sys/acct.h>		/* for acct_process() function prototype */
68 #include <sys/filedesc.h>
69 #include <sys/mac.h>
70 #include <sys/shm.h>
71 #include <sys/sem.h>
72 #ifdef KTRACE
73 #include <sys/ktrace.h>
74 #endif
75 
76 #include <vm/vm.h>
77 #include <vm/vm_extern.h>
78 #include <vm/vm_param.h>
79 #include <vm/pmap.h>
80 #include <vm/vm_map.h>
81 #include <vm/vm_page.h>
82 #include <vm/uma.h>
83 #include <sys/user.h>
84 
85 /* Required to be non-static for SysVR4 emulator */
86 MALLOC_DEFINE(M_ZOMBIE, "zombie", "zombie proc status");
87 
88 /*
89  * exit --
90  *	Death of process.
91  *
92  * MPSAFE
93  */
94 void
95 sys_exit(struct thread *td, struct sys_exit_args *uap)
96 {
97 
98 	exit1(td, W_EXITCODE(uap->rval, 0));
99 	/* NOTREACHED */
100 }
101 
102 /*
103  * Exit: deallocate address space and other resources, change proc state
104  * to zombie, and unlink proc from allproc and parent's lists.  Save exit
105  * status and rusage for wait().  Check for child processes and orphan them.
106  */
107 void
108 exit1(struct thread *td, int rv)
109 {
110 	struct proc *p, *nq, *q;
111 	struct tty *tp;
112 	struct vnode *ttyvp;
113 	struct vmspace *vm;
114 	struct vnode *vtmp;
115 #ifdef KTRACE
116 	struct vnode *tracevp;
117 	struct ucred *tracecred;
118 #endif
119 	struct plimit *plim;
120 
121 	/*
122 	 * Drop Giant if caller has it.  Eventually we should warn about
123 	 * being called with Giant held.
124 	 */
125 	while (mtx_owned(&Giant))
126 		mtx_unlock(&Giant);
127 
128 	p = td->td_proc;
129 	if (p == initproc) {
130 		printf("init died (signal %d, exit %d)\n",
131 		    WTERMSIG(rv), WEXITSTATUS(rv));
132 		panic("Going nowhere without my init!");
133 	}
134 
135 	/*
136 	 * MUST abort all other threads before proceeding past here.
137 	 */
138 	PROC_LOCK(p);
139 	if (p->p_flag & P_SA || p->p_numthreads > 1) {
140 		/*
141 		 * First check if some other thread got here before us..
142 		 * if so, act apropriatly, (exit or suspend);
143 		 */
144 		thread_suspend_check(0);
145 
146 		/*
147 		 * Kill off the other threads. This requires
148 		 * Some co-operation from other parts of the kernel
149 		 * so it may not be instant.
150 		 * With this state set:
151 		 * Any thread entering the kernel from userspace will
152 		 * thread_exit() in trap().  Any thread attempting to
153 		 * sleep will return immediatly
154 		 * with EINTR or EWOULDBLOCK, which will hopefully force them
155 		 * to back out to userland, freeing resources as they go, and
156 		 * anything attempting to return to userland will thread_exit()
157 		 * from userret().  thread_exit() will unsuspend us
158 		 * when the last other thread exits.
159 		 */
160 		if (thread_single(SINGLE_EXIT))
161 			panic ("Exit: Single threading fouled up");
162 		/*
163 		 * All other activity in this process is now stopped.
164 		 * Remove excess KSEs and KSEGRPS. XXXKSE (when we have them)
165 		 * ...
166 		 * Turn off threading support.
167 		 */
168 		p->p_flag &= ~P_SA;
169 		thread_single_end();	/* Don't need this any more. */
170 	}
171 	/*
172 	 * With this state set:
173 	 * Any thread entering the kernel from userspace will thread_exit()
174 	 * in trap().  Any thread attempting to sleep will return immediatly
175 	 * with EINTR or EWOULDBLOCK, which will hopefully force them
176 	 * to back out to userland, freeing resources as they go, and
177 	 * anything attempting to return to userland will thread_exit()
178 	 * from userret().  thread_exit() will do a wakeup on p->p_numthreads
179 	 * if it transitions to 1.
180 	 */
181 
182 	p->p_flag |= P_WEXIT;
183 	PROC_UNLOCK(p);
184 
185 	/* Are we a task leader? */
186 	if (p == p->p_leader) {
187 		mtx_lock(&ppeers_lock);
188 		q = p->p_peers;
189 		while (q != NULL) {
190 			PROC_LOCK(q);
191 			psignal(q, SIGKILL);
192 			PROC_UNLOCK(q);
193 			q = q->p_peers;
194 		}
195 		while (p->p_peers != NULL)
196 			msleep(p, &ppeers_lock, PWAIT, "exit1", 0);
197 		mtx_unlock(&ppeers_lock);
198 	}
199 
200 #ifdef PGINPROF
201 	mtx_lock(&Giant);
202 	vmsizmon();
203 	mtx_unlock(&Giant);
204 #endif
205 	PROC_LOCK(p);
206 	_STOPEVENT(p, S_EXIT, rv);
207 	wakeup(&p->p_stype);	/* Wakeup anyone in procfs' PIOCWAIT */
208 	PROC_UNLOCK(p);
209 
210 	/*
211 	 * Check if any loadable modules need anything done at process exit.
212 	 * e.g. SYSV IPC stuff
213 	 * XXX what if one of these generates an error?
214 	 */
215 	EVENTHANDLER_INVOKE(process_exit, p);
216 
217 	MALLOC(p->p_ru, struct rusage *, sizeof(struct rusage),
218 		M_ZOMBIE, M_WAITOK);
219 	/*
220 	 * If parent is waiting for us to exit or exec,
221 	 * P_PPWAIT is set; we will wakeup the parent below.
222 	 */
223 	PROC_LOCK(p);
224 	stopprofclock(p);
225 	p->p_flag &= ~(P_TRACED | P_PPWAIT);
226 	SIGEMPTYSET(p->p_siglist);
227 	SIGEMPTYSET(td->td_siglist);
228 
229 	/*
230 	 * Stop the real interval timer.  If the handler is currently
231 	 * executing, prevent it from rearming itself and let it finish.
232 	 */
233 	if (timevalisset(&p->p_realtimer.it_value) &&
234 	    callout_stop(&p->p_itcallout) == 0) {
235 		timevalclear(&p->p_realtimer.it_interval);
236 		msleep(&p->p_itcallout, &p->p_mtx, PWAIT, "ritwait", 0);
237 		KASSERT(!timevalisset(&p->p_realtimer.it_value),
238 		    ("realtime timer is still armed"));
239 	}
240 	PROC_UNLOCK(p);
241 
242 	/*
243 	 * Reset any sigio structures pointing to us as a result of
244 	 * F_SETOWN with our pid.
245 	 */
246 	mtx_lock(&Giant);	/* XXX: not sure if needed */
247 	funsetownlst(&p->p_sigiolst);
248 
249 	/*
250 	 * Close open files and release open-file table.
251 	 * This may block!
252 	 */
253 	fdfree(td);
254 	mtx_unlock(&Giant);
255 
256 	/*
257 	 * Remove ourself from our leader's peer list and wake our leader.
258 	 */
259 	mtx_lock(&ppeers_lock);
260 	if (p->p_leader->p_peers) {
261 		q = p->p_leader;
262 		while (q->p_peers != p)
263 			q = q->p_peers;
264 		q->p_peers = p->p_peers;
265 		wakeup(p->p_leader);
266 	}
267 	mtx_unlock(&ppeers_lock);
268 
269 	mtx_lock(&Giant);
270 	/* The next two chunks should probably be moved to vmspace_exit. */
271 	vm = p->p_vmspace;
272 	/*
273 	 * Release user portion of address space.
274 	 * This releases references to vnodes,
275 	 * which could cause I/O if the file has been unlinked.
276 	 * Need to do this early enough that we can still sleep.
277 	 * Can't free the entire vmspace as the kernel stack
278 	 * may be mapped within that space also.
279 	 *
280 	 * Processes sharing the same vmspace may exit in one order, and
281 	 * get cleaned up by vmspace_exit() in a different order.  The
282 	 * last exiting process to reach this point releases as much of
283 	 * the environment as it can, and the last process cleaned up
284 	 * by vmspace_exit() (which decrements exitingcnt) cleans up the
285 	 * remainder.
286 	 */
287 	++vm->vm_exitingcnt;
288 	if (--vm->vm_refcnt == 0) {
289 		shmexit(vm);
290 		vm_page_lock_queues();
291 		pmap_remove_pages(vmspace_pmap(vm), vm_map_min(&vm->vm_map),
292 		    vm_map_max(&vm->vm_map));
293 		vm_page_unlock_queues();
294 		(void) vm_map_remove(&vm->vm_map, vm_map_min(&vm->vm_map),
295 		    vm_map_max(&vm->vm_map));
296 	}
297 
298 	sx_xlock(&proctree_lock);
299 	if (SESS_LEADER(p)) {
300 		struct session *sp;
301 
302 		sp = p->p_session;
303 		if (sp->s_ttyvp) {
304 			/*
305 			 * Controlling process.
306 			 * Signal foreground pgrp,
307 			 * drain controlling terminal
308 			 * and revoke access to controlling terminal.
309 			 */
310 			if (sp->s_ttyp && (sp->s_ttyp->t_session == sp)) {
311 				tp = sp->s_ttyp;
312 				if (sp->s_ttyp->t_pgrp) {
313 					PGRP_LOCK(sp->s_ttyp->t_pgrp);
314 					pgsignal(sp->s_ttyp->t_pgrp, SIGHUP, 1);
315 					PGRP_UNLOCK(sp->s_ttyp->t_pgrp);
316 				}
317 				/* XXX tp should be locked. */
318 				sx_xunlock(&proctree_lock);
319 				(void) ttywait(tp);
320 				sx_xlock(&proctree_lock);
321 				/*
322 				 * The tty could have been revoked
323 				 * if we blocked.
324 				 */
325 				if (sp->s_ttyvp) {
326 					ttyvp = sp->s_ttyvp;
327 					SESS_LOCK(p->p_session);
328 					sp->s_ttyvp = NULL;
329 					SESS_UNLOCK(p->p_session);
330 					sx_xunlock(&proctree_lock);
331 					VOP_REVOKE(ttyvp, REVOKEALL);
332 					vrele(ttyvp);
333 					sx_xlock(&proctree_lock);
334 				}
335 			}
336 			if (sp->s_ttyvp) {
337 				ttyvp = sp->s_ttyvp;
338 				SESS_LOCK(p->p_session);
339 				sp->s_ttyvp = NULL;
340 				SESS_UNLOCK(p->p_session);
341 				vrele(ttyvp);
342 			}
343 			/*
344 			 * s_ttyp is not zero'd; we use this to indicate
345 			 * that the session once had a controlling terminal.
346 			 * (for logging and informational purposes)
347 			 */
348 		}
349 		SESS_LOCK(p->p_session);
350 		sp->s_leader = NULL;
351 		SESS_UNLOCK(p->p_session);
352 	}
353 	fixjobc(p, p->p_pgrp, 0);
354 	sx_xunlock(&proctree_lock);
355 	(void)acct_process(td);
356 	mtx_unlock(&Giant);
357 #ifdef KTRACE
358 	/*
359 	 * release trace file
360 	 */
361 	PROC_LOCK(p);
362 	mtx_lock(&ktrace_mtx);
363 	p->p_traceflag = 0;	/* don't trace the vrele() */
364 	tracevp = p->p_tracevp;
365 	p->p_tracevp = NULL;
366 	tracecred = p->p_tracecred;
367 	p->p_tracecred = NULL;
368 	mtx_unlock(&ktrace_mtx);
369 	PROC_UNLOCK(p);
370 	if (tracevp != NULL) {
371 		mtx_lock(&Giant);
372 		vrele(tracevp);
373 		mtx_unlock(&Giant);
374 	}
375 	if (tracecred != NULL)
376 		crfree(tracecred);
377 #endif
378 	/*
379 	 * Release reference to text vnode
380 	 */
381 	if ((vtmp = p->p_textvp) != NULL) {
382 		p->p_textvp = NULL;
383 		mtx_lock(&Giant);
384 		vrele(vtmp);
385 		mtx_unlock(&Giant);
386 	}
387 
388 	/*
389 	 * Release our limits structure.
390 	 */
391 	PROC_LOCK(p);
392 	plim = p->p_limit;
393 	p->p_limit = NULL;
394 	PROC_UNLOCK(p);
395 	lim_free(plim);
396 
397 	/*
398 	 * Release this thread's reference to the ucred.  The actual proc
399 	 * reference will stay around until the proc is harvested by
400 	 * wait().  At this point the ucred is immutable (no other threads
401 	 * from this proc are around that can change it) so we leave the
402 	 * per-thread ucred pointer intact in case it is needed although
403 	 * in theory nothing should be using it at this point.
404 	 */
405 	crfree(td->td_ucred);
406 
407 	/*
408 	 * Remove proc from allproc queue and pidhash chain.
409 	 * Place onto zombproc.  Unlink from parent's child list.
410 	 */
411 	sx_xlock(&allproc_lock);
412 	LIST_REMOVE(p, p_list);
413 	LIST_INSERT_HEAD(&zombproc, p, p_list);
414 	LIST_REMOVE(p, p_hash);
415 	sx_xunlock(&allproc_lock);
416 
417 	sx_xlock(&proctree_lock);
418 	q = LIST_FIRST(&p->p_children);
419 	if (q != NULL)		/* only need this if any child is S_ZOMB */
420 		wakeup(initproc);
421 	for (; q != NULL; q = nq) {
422 		nq = LIST_NEXT(q, p_sibling);
423 		PROC_LOCK(q);
424 		proc_reparent(q, initproc);
425 		q->p_sigparent = SIGCHLD;
426 		/*
427 		 * Traced processes are killed
428 		 * since their existence means someone is screwing up.
429 		 */
430 		if (q->p_flag & P_TRACED) {
431 			q->p_flag &= ~P_TRACED;
432 			psignal(q, SIGKILL);
433 		}
434 		PROC_UNLOCK(q);
435 	}
436 
437 	/*
438 	 * Save exit status and final rusage info, adding in child rusage
439 	 * info and self times.
440 	 */
441 	mtx_lock(&Giant);
442 	PROC_LOCK(p);
443 	p->p_xstat = rv;
444 	*p->p_ru = p->p_stats->p_ru;
445 	mtx_lock_spin(&sched_lock);
446 	calcru(p, &p->p_ru->ru_utime, &p->p_ru->ru_stime, NULL);
447 	mtx_unlock_spin(&sched_lock);
448 	ruadd(p->p_ru, &p->p_stats->p_cru);
449 
450 	/*
451 	 * Notify interested parties of our demise.
452 	 */
453 	KNOTE(&p->p_klist, NOTE_EXIT);
454 	mtx_unlock(&Giant);
455 	/*
456 	 * Just delete all entries in the p_klist. At this point we won't
457 	 * report any more events, and there are nasty race conditions that
458 	 * can beat us if we don't.
459 	 */
460 	while (SLIST_FIRST(&p->p_klist))
461 		SLIST_REMOVE_HEAD(&p->p_klist, kn_selnext);
462 
463 	/*
464 	 * Notify parent that we're gone.  If parent has the PS_NOCLDWAIT
465 	 * flag set, or if the handler is set to SIG_IGN, notify process
466 	 * 1 instead (and hope it will handle this situation).
467 	 */
468 	PROC_LOCK(p->p_pptr);
469 	mtx_lock(&p->p_pptr->p_sigacts->ps_mtx);
470 	if (p->p_pptr->p_sigacts->ps_flag & (PS_NOCLDWAIT | PS_CLDSIGIGN)) {
471 		struct proc *pp;
472 
473 		mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx);
474 		pp = p->p_pptr;
475 		PROC_UNLOCK(pp);
476 		proc_reparent(p, initproc);
477 		p->p_sigparent = SIGCHLD;
478 		PROC_LOCK(p->p_pptr);
479 		/*
480 		 * If this was the last child of our parent, notify
481 		 * parent, so in case he was wait(2)ing, he will
482 		 * continue.
483 		 */
484 		if (LIST_EMPTY(&pp->p_children))
485 			wakeup(pp);
486 	} else
487 		mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx);
488 
489 	if (p->p_pptr == initproc)
490 		psignal(p->p_pptr, SIGCHLD);
491 	else if (p->p_sigparent != 0)
492 		psignal(p->p_pptr, p->p_sigparent);
493 	PROC_UNLOCK(p->p_pptr);
494 
495 	/*
496 	 * If this is a kthread, then wakeup anyone waiting for it to exit.
497 	 */
498 	if (p->p_flag & P_KTHREAD)
499 		wakeup(p);
500 	PROC_UNLOCK(p);
501 
502 	/*
503 	 * Finally, call machine-dependent code to release the remaining
504 	 * resources including address space.
505 	 * The address space is released by "vmspace_exitfree(p)" in
506 	 * vm_waitproc().
507 	 */
508 	cpu_exit(td);
509 
510 	PROC_LOCK(p);
511 	PROC_LOCK(p->p_pptr);
512 	sx_xunlock(&proctree_lock);
513 
514 	while (mtx_owned(&Giant))
515 		mtx_unlock(&Giant);
516 
517 	/*
518 	 * We have to wait until after acquiring all locks before
519 	 * changing p_state.  We need to avoid any possibly context
520 	 * switches while marked as a zombie including blocking on
521 	 * a mutex.
522 	 */
523 	mtx_lock_spin(&sched_lock);
524 	p->p_state = PRS_ZOMBIE;
525 	critical_enter();
526 	mtx_unlock_spin(&sched_lock);
527 
528 	wakeup(p->p_pptr);
529 	PROC_UNLOCK(p->p_pptr);
530 
531 	mtx_lock_spin(&sched_lock);
532 	critical_exit();
533 	cnt.v_swtch++;
534 	binuptime(PCPU_PTR(switchtime));
535 	PCPU_SET(switchticks, ticks);
536 
537 	cpu_sched_exit(td); /* XXXKSE check if this should be in thread_exit */
538 	/*
539 	 * Allow the scheduler to adjust the priority of the
540 	 * parent when a kseg is exiting.
541 	 */
542 	if (p->p_pid != 1)
543 		sched_exit(p->p_pptr, p);
544 
545 	/*
546 	 * Make sure the scheduler takes this thread out of its tables etc.
547 	 * This will also release this thread's reference to the ucred.
548 	 * Other thread parts to release include pcb bits and such.
549 	 */
550 	thread_exit();
551 }
552 
553 #ifdef COMPAT_43
554 /*
555  * MPSAFE.  The dirty work is handled by kern_wait().
556  */
557 int
558 owait(struct thread *td, struct owait_args *uap __unused)
559 {
560 	int error, status;
561 
562 	error = kern_wait(td, WAIT_ANY, &status, 0, NULL);
563 	if (error == 0)
564 		td->td_retval[1] = status;
565 	return (error);
566 }
567 #endif /* COMPAT_43 */
568 
569 /*
570  * MPSAFE.  The dirty work is handled by kern_wait().
571  */
572 int
573 wait4(struct thread *td, struct wait_args *uap)
574 {
575 	struct rusage ru;
576 	int error, status;
577 
578 	error = kern_wait(td, uap->pid, &status, uap->options, &ru);
579 	if (uap->status != NULL && error == 0)
580 		error = copyout(&status, uap->status, sizeof(status));
581 	if (uap->rusage != NULL && error == 0)
582 		error = copyout(&ru, uap->rusage, sizeof(struct rusage));
583 	return (error);
584 }
585 
586 int
587 kern_wait(struct thread *td, pid_t pid, int *status, int options, struct rusage *rusage)
588 {
589 	int nfound;
590 	struct proc *p, *q, *t;
591 	int error;
592 
593 	q = td->td_proc;
594 	if (pid == 0) {
595 		PROC_LOCK(q);
596 		pid = -q->p_pgid;
597 		PROC_UNLOCK(q);
598 	}
599 	if (options &~ (WUNTRACED|WNOHANG|WCONTINUED|WLINUXCLONE))
600 		return (EINVAL);
601 loop:
602 	nfound = 0;
603 	sx_xlock(&proctree_lock);
604 	LIST_FOREACH(p, &q->p_children, p_sibling) {
605 		PROC_LOCK(p);
606 		if (pid != WAIT_ANY &&
607 		    p->p_pid != pid && p->p_pgid != -pid) {
608 			PROC_UNLOCK(p);
609 			continue;
610 		}
611 
612 		/*
613 		 * This special case handles a kthread spawned by linux_clone
614 		 * (see linux_misc.c).  The linux_wait4 and linux_waitpid
615 		 * functions need to be able to distinguish between waiting
616 		 * on a process and waiting on a thread.  It is a thread if
617 		 * p_sigparent is not SIGCHLD, and the WLINUXCLONE option
618 		 * signifies we want to wait for threads and not processes.
619 		 */
620 		if ((p->p_sigparent != SIGCHLD) ^
621 		    ((options & WLINUXCLONE) != 0)) {
622 			PROC_UNLOCK(p);
623 			continue;
624 		}
625 
626 		nfound++;
627 		if (p->p_state == PRS_ZOMBIE) {
628 			td->td_retval[0] = p->p_pid;
629 			if (status)
630 				*status = p->p_xstat;	/* convert to int */
631 			if (rusage)
632 				bcopy(p->p_ru, rusage, sizeof(struct rusage));
633 
634 			/*
635 			 * If we got the child via a ptrace 'attach',
636 			 * we need to give it back to the old parent.
637 			 */
638 			PROC_UNLOCK(p);
639 			if (p->p_oppid && (t = pfind(p->p_oppid)) != NULL) {
640 				PROC_LOCK(p);
641 				p->p_oppid = 0;
642 				proc_reparent(p, t);
643 				PROC_UNLOCK(p);
644 				psignal(t, SIGCHLD);
645 				wakeup(t);
646 				PROC_UNLOCK(t);
647 				sx_xunlock(&proctree_lock);
648 				return (0);
649 			}
650 
651 			/*
652 			 * Remove other references to this process to ensure
653 			 * we have an exclusive reference.
654 			 */
655 			sx_xlock(&allproc_lock);
656 			LIST_REMOVE(p, p_list);	/* off zombproc */
657 			sx_xunlock(&allproc_lock);
658 			LIST_REMOVE(p, p_sibling);
659 			leavepgrp(p);
660 			sx_xunlock(&proctree_lock);
661 
662 			/*
663 			 * As a side effect of this lock, we know that
664 			 * all other writes to this proc are visible now, so
665 			 * no more locking is needed for p.
666 			 */
667 			mtx_lock(&Giant);
668 			PROC_LOCK(p);
669 			p->p_xstat = 0;		/* XXX: why? */
670 			PROC_UNLOCK(p);
671 			PROC_LOCK(q);
672 			ruadd(&q->p_stats->p_cru, p->p_ru);
673 			PROC_UNLOCK(q);
674 			FREE(p->p_ru, M_ZOMBIE);
675 			p->p_ru = NULL;
676 			mtx_unlock(&Giant);
677 
678 			/*
679 			 * Decrement the count of procs running with this uid.
680 			 */
681 			(void)chgproccnt(p->p_ucred->cr_ruidinfo, -1, 0);
682 
683 			/*
684 			 * Free credentials, arguments, and sigacts
685 			 */
686 			crfree(p->p_ucred);
687 			p->p_ucred = NULL;
688 			pargs_drop(p->p_args);
689 			p->p_args = NULL;
690 			sigacts_free(p->p_sigacts);
691 			p->p_sigacts = NULL;
692 
693 			/*
694 			 * do any thread-system specific cleanups
695 			 */
696 			thread_wait(p);
697 
698 			/*
699 			 * Give vm and machine-dependent layer a chance
700 			 * to free anything that cpu_exit couldn't
701 			 * release while still running in process context.
702 			 */
703 			mtx_lock(&Giant);
704 			vm_waitproc(p);
705 			mtx_unlock(&Giant);
706 #ifdef MAC
707 			mac_destroy_proc(p);
708 #endif
709 			KASSERT(FIRST_THREAD_IN_PROC(p),
710 			    ("kern_wait: no residual thread!"));
711 			uma_zfree(proc_zone, p);
712 			sx_xlock(&allproc_lock);
713 			nprocs--;
714 			sx_xunlock(&allproc_lock);
715 			return (0);
716 		}
717 		mtx_lock_spin(&sched_lock);
718 		if (P_SHOULDSTOP(p) && (p->p_suspcount == p->p_numthreads) &&
719 		    ((p->p_flag & P_WAITED) == 0) &&
720 		    (p->p_flag & P_TRACED || options & WUNTRACED)) {
721 			mtx_unlock_spin(&sched_lock);
722 			p->p_flag |= P_WAITED;
723 			sx_xunlock(&proctree_lock);
724 			td->td_retval[0] = p->p_pid;
725 			if (status)
726 				*status = W_STOPCODE(p->p_xstat);
727 			PROC_UNLOCK(p);
728 			return (0);
729 		}
730 		mtx_unlock_spin(&sched_lock);
731 		if (options & WCONTINUED && (p->p_flag & P_CONTINUED)) {
732 			sx_xunlock(&proctree_lock);
733 			td->td_retval[0] = p->p_pid;
734 			p->p_flag &= ~P_CONTINUED;
735 			PROC_UNLOCK(p);
736 
737 			if (status)
738 				*status = SIGCONT;
739 			return (0);
740 		}
741 		PROC_UNLOCK(p);
742 	}
743 	if (nfound == 0) {
744 		sx_xunlock(&proctree_lock);
745 		return (ECHILD);
746 	}
747 	if (options & WNOHANG) {
748 		sx_xunlock(&proctree_lock);
749 		td->td_retval[0] = 0;
750 		return (0);
751 	}
752 	PROC_LOCK(q);
753 	sx_xunlock(&proctree_lock);
754 	error = msleep(q, &q->p_mtx, PWAIT | PCATCH, "wait", 0);
755 	PROC_UNLOCK(q);
756 	if (error)
757 		return (error);
758 	goto loop;
759 }
760 
761 /*
762  * Make process 'parent' the new parent of process 'child'.
763  * Must be called with an exclusive hold of proctree lock.
764  */
765 void
766 proc_reparent(struct proc *child, struct proc *parent)
767 {
768 
769 	sx_assert(&proctree_lock, SX_XLOCKED);
770 	PROC_LOCK_ASSERT(child, MA_OWNED);
771 	if (child->p_pptr == parent)
772 		return;
773 
774 	LIST_REMOVE(child, p_sibling);
775 	LIST_INSERT_HEAD(&parent->p_children, child, p_sibling);
776 	child->p_pptr = parent;
777 }
778