xref: /freebsd/sys/kern/kern_exit.c (revision 52f72944b8f5abb2386eae924357dee8aea17d5b)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1982, 1986, 1989, 1991, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  * (c) UNIX System Laboratories, Inc.
7  * All or some portions of this file are derived from material licensed
8  * to the University of California by American Telephone and Telegraph
9  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
10  * the permission of UNIX System Laboratories, Inc.
11  *
12  * Redistribution and use in source and binary forms, with or without
13  * modification, are permitted provided that the following conditions
14  * are met:
15  * 1. Redistributions of source code must retain the above copyright
16  *    notice, this list of conditions and the following disclaimer.
17  * 2. Redistributions in binary form must reproduce the above copyright
18  *    notice, this list of conditions and the following disclaimer in the
19  *    documentation and/or other materials provided with the distribution.
20  * 3. Neither the name of the University nor the names of its contributors
21  *    may be used to endorse or promote products derived from this software
22  *    without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  *
36  *	@(#)kern_exit.c	8.7 (Berkeley) 2/12/94
37  */
38 
39 #include <sys/cdefs.h>
40 __FBSDID("$FreeBSD$");
41 
42 #include "opt_compat.h"
43 #include "opt_ktrace.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, , , exit, "int");
99 
100 /* Hook for NFS teardown procedure. */
101 void (*nlminfo_release_p)(struct proc *p);
102 
103 EVENTHANDLER_LIST_DECLARE(process_exit);
104 
105 struct proc *
106 proc_realparent(struct proc *child)
107 {
108 	struct proc *p, *parent;
109 
110 	sx_assert(&proctree_lock, SX_LOCKED);
111 	if ((child->p_treeflag & P_TREE_ORPHANED) == 0) {
112 		if (child->p_oppid == 0 ||
113 		    child->p_pptr->p_pid == child->p_oppid)
114 			parent = child->p_pptr;
115 		else
116 			parent = initproc;
117 		return (parent);
118 	}
119 	for (p = child; (p->p_treeflag & P_TREE_FIRST_ORPHAN) == 0;) {
120 		/* Cannot use LIST_PREV(), since the list head is not known. */
121 		p = __containerof(p->p_orphan.le_prev, struct proc,
122 		    p_orphan.le_next);
123 		KASSERT((p->p_treeflag & P_TREE_ORPHANED) != 0,
124 		    ("missing P_ORPHAN %p", p));
125 	}
126 	parent = __containerof(p->p_orphan.le_prev, struct proc,
127 	    p_orphans.lh_first);
128 	return (parent);
129 }
130 
131 void
132 reaper_abandon_children(struct proc *p, bool exiting)
133 {
134 	struct proc *p1, *p2, *ptmp;
135 
136 	sx_assert(&proctree_lock, SX_LOCKED);
137 	KASSERT(p != initproc, ("reaper_abandon_children for initproc"));
138 	if ((p->p_treeflag & P_TREE_REAPER) == 0)
139 		return;
140 	p1 = p->p_reaper;
141 	LIST_FOREACH_SAFE(p2, &p->p_reaplist, p_reapsibling, ptmp) {
142 		LIST_REMOVE(p2, p_reapsibling);
143 		p2->p_reaper = p1;
144 		p2->p_reapsubtree = p->p_reapsubtree;
145 		LIST_INSERT_HEAD(&p1->p_reaplist, p2, p_reapsibling);
146 		if (exiting && p2->p_pptr == p) {
147 			PROC_LOCK(p2);
148 			proc_reparent(p2, p1);
149 			PROC_UNLOCK(p2);
150 		}
151 	}
152 	KASSERT(LIST_EMPTY(&p->p_reaplist), ("p_reaplist not empty"));
153 	p->p_treeflag &= ~P_TREE_REAPER;
154 }
155 
156 static void
157 clear_orphan(struct proc *p)
158 {
159 	struct proc *p1;
160 
161 	sx_assert(&proctree_lock, SA_XLOCKED);
162 	if ((p->p_treeflag & P_TREE_ORPHANED) == 0)
163 		return;
164 	if ((p->p_treeflag & P_TREE_FIRST_ORPHAN) != 0) {
165 		p1 = LIST_NEXT(p, p_orphan);
166 		if (p1 != NULL)
167 			p1->p_treeflag |= P_TREE_FIRST_ORPHAN;
168 		p->p_treeflag &= ~P_TREE_FIRST_ORPHAN;
169 	}
170 	LIST_REMOVE(p, p_orphan);
171 	p->p_treeflag &= ~P_TREE_ORPHANED;
172 }
173 
174 /*
175  * exit -- death of process.
176  */
177 void
178 sys_sys_exit(struct thread *td, struct sys_exit_args *uap)
179 {
180 
181 	exit1(td, uap->rval, 0);
182 	/* NOTREACHED */
183 }
184 
185 /*
186  * Exit: deallocate address space and other resources, change proc state to
187  * zombie, and unlink proc from allproc and parent's lists.  Save exit status
188  * and rusage for wait().  Check for child processes and orphan them.
189  */
190 void
191 exit1(struct thread *td, int rval, int signo)
192 {
193 	struct proc *p, *nq, *q, *t;
194 	struct thread *tdt;
195 	ksiginfo_t *ksi, *ksi1;
196 	int signal_parent;
197 
198 	mtx_assert(&Giant, MA_NOTOWNED);
199 	KASSERT(rval == 0 || signo == 0, ("exit1 rv %d sig %d", rval, signo));
200 
201 	p = td->td_proc;
202 	/*
203 	 * XXX in case we're rebooting we just let init die in order to
204 	 * work around an unsolved stack overflow seen very late during
205 	 * shutdown on sparc64 when the gmirror worker process exists.
206 	 */
207 	if (p == initproc && rebooting == 0) {
208 		printf("init died (signal %d, exit %d)\n", signo, rval);
209 		panic("Going nowhere without my init!");
210 	}
211 
212 	/*
213 	 * Deref SU mp, since the thread does not return to userspace.
214 	 */
215 	td_softdep_cleanup(td);
216 
217 	/*
218 	 * MUST abort all other threads before proceeding past here.
219 	 */
220 	PROC_LOCK(p);
221 	/*
222 	 * First check if some other thread or external request got
223 	 * here before us.  If so, act appropriately: exit or suspend.
224 	 * We must ensure that stop requests are handled before we set
225 	 * P_WEXIT.
226 	 */
227 	thread_suspend_check(0);
228 	while (p->p_flag & P_HADTHREADS) {
229 		/*
230 		 * Kill off the other threads. This requires
231 		 * some co-operation from other parts of the kernel
232 		 * so it may not be instantaneous.  With this state set
233 		 * any thread entering the kernel from userspace will
234 		 * thread_exit() in trap().  Any thread attempting to
235 		 * sleep will return immediately with EINTR or EWOULDBLOCK
236 		 * which will hopefully force them to back out to userland
237 		 * freeing resources as they go.  Any thread attempting
238 		 * to return to userland will thread_exit() from userret().
239 		 * thread_exit() will unsuspend us when the last of the
240 		 * other threads exits.
241 		 * If there is already a thread singler after resumption,
242 		 * calling thread_single will fail; in that case, we just
243 		 * re-check all suspension request, the thread should
244 		 * either be suspended there or exit.
245 		 */
246 		if (!thread_single(p, SINGLE_EXIT))
247 			/*
248 			 * All other activity in this process is now
249 			 * stopped.  Threading support has been turned
250 			 * off.
251 			 */
252 			break;
253 		/*
254 		 * Recheck for new stop or suspend requests which
255 		 * might appear while process lock was dropped in
256 		 * thread_single().
257 		 */
258 		thread_suspend_check(0);
259 	}
260 	KASSERT(p->p_numthreads == 1,
261 	    ("exit1: proc %p exiting with %d threads", p, p->p_numthreads));
262 	racct_sub(p, RACCT_NTHR, 1);
263 
264 	/* Let event handler change exit status */
265 	p->p_xexit = rval;
266 	p->p_xsig = signo;
267 
268 	/*
269 	 * Wakeup anyone in procfs' PIOCWAIT.  They should have a hold
270 	 * on our vmspace, so we should block below until they have
271 	 * released their reference to us.  Note that if they have
272 	 * requested S_EXIT stops we will block here until they ack
273 	 * via PIOCCONT.
274 	 */
275 	_STOPEVENT(p, S_EXIT, 0);
276 
277 	/*
278 	 * Ignore any pending request to stop due to a stop signal.
279 	 * Once P_WEXIT is set, future requests will be ignored as
280 	 * well.
281 	 */
282 	p->p_flag &= ~P_STOPPED_SIG;
283 	KASSERT(!P_SHOULDSTOP(p), ("exiting process is stopped"));
284 
285 	/*
286 	 * Note that we are exiting and do another wakeup of anyone in
287 	 * PIOCWAIT in case they aren't listening for S_EXIT stops or
288 	 * decided to wait again after we told them we are exiting.
289 	 */
290 	p->p_flag |= P_WEXIT;
291 	wakeup(&p->p_stype);
292 
293 	/*
294 	 * Wait for any processes that have a hold on our vmspace to
295 	 * release their reference.
296 	 */
297 	while (p->p_lock > 0)
298 		msleep(&p->p_lock, &p->p_mtx, PWAIT, "exithold", 0);
299 
300 	PROC_UNLOCK(p);
301 	/* Drain the limit callout while we don't have the proc locked */
302 	callout_drain(&p->p_limco);
303 
304 #ifdef AUDIT
305 	/*
306 	 * The Sun BSM exit token contains two components: an exit status as
307 	 * passed to exit(), and a return value to indicate what sort of exit
308 	 * it was.  The exit status is WEXITSTATUS(rv), but it's not clear
309 	 * what the return value is.
310 	 */
311 	AUDIT_ARG_EXIT(rval, 0);
312 	AUDIT_SYSCALL_EXIT(0, td);
313 #endif
314 
315 	/* Are we a task leader with peers? */
316 	if (p->p_peers != NULL && p == p->p_leader) {
317 		mtx_lock(&ppeers_lock);
318 		q = p->p_peers;
319 		while (q != NULL) {
320 			PROC_LOCK(q);
321 			kern_psignal(q, SIGKILL);
322 			PROC_UNLOCK(q);
323 			q = q->p_peers;
324 		}
325 		while (p->p_peers != NULL)
326 			msleep(p, &ppeers_lock, PWAIT, "exit1", 0);
327 		mtx_unlock(&ppeers_lock);
328 	}
329 
330 	/*
331 	 * Check if any loadable modules need anything done at process exit.
332 	 * E.g. SYSV IPC stuff.
333 	 * Event handler could change exit status.
334 	 * XXX what if one of these generates an error?
335 	 */
336 	EVENTHANDLER_DIRECT_INVOKE(process_exit, p);
337 
338 	/*
339 	 * If parent is waiting for us to exit or exec,
340 	 * P_PPWAIT is set; we will wakeup the parent below.
341 	 */
342 	PROC_LOCK(p);
343 	stopprofclock(p);
344 	p->p_flag &= ~(P_TRACED | P_PPWAIT | P_PPTRACE);
345 	p->p_ptevents = 0;
346 
347 	/*
348 	 * Stop the real interval timer.  If the handler is currently
349 	 * executing, prevent it from rearming itself and let it finish.
350 	 */
351 	if (timevalisset(&p->p_realtimer.it_value) &&
352 	    _callout_stop_safe(&p->p_itcallout, CS_EXECUTING, NULL) == 0) {
353 		timevalclear(&p->p_realtimer.it_interval);
354 		msleep(&p->p_itcallout, &p->p_mtx, PWAIT, "ritwait", 0);
355 		KASSERT(!timevalisset(&p->p_realtimer.it_value),
356 		    ("realtime timer is still armed"));
357 	}
358 
359 	PROC_UNLOCK(p);
360 
361 	umtx_thread_exit(td);
362 
363 	/*
364 	 * Reset any sigio structures pointing to us as a result of
365 	 * F_SETOWN with our pid.
366 	 */
367 	funsetownlst(&p->p_sigiolst);
368 
369 	/*
370 	 * If this process has an nlminfo data area (for lockd), release it
371 	 */
372 	if (nlminfo_release_p != NULL && p->p_nlminfo != NULL)
373 		(*nlminfo_release_p)(p);
374 
375 	/*
376 	 * Close open files and release open-file table.
377 	 * This may block!
378 	 */
379 	fdescfree(td);
380 
381 	/*
382 	 * If this thread tickled GEOM, we need to wait for the giggling to
383 	 * stop before we return to userland
384 	 */
385 	if (td->td_pflags & TDP_GEOM)
386 		g_waitidle();
387 
388 	/*
389 	 * Remove ourself from our leader's peer list and wake our leader.
390 	 */
391 	if (p->p_leader->p_peers != NULL) {
392 		mtx_lock(&ppeers_lock);
393 		if (p->p_leader->p_peers != NULL) {
394 			q = p->p_leader;
395 			while (q->p_peers != p)
396 				q = q->p_peers;
397 			q->p_peers = p->p_peers;
398 			wakeup(p->p_leader);
399 		}
400 		mtx_unlock(&ppeers_lock);
401 	}
402 
403 	vmspace_exit(td);
404 	killjobc();
405 	(void)acct_process(td);
406 
407 #ifdef KTRACE
408 	ktrprocexit(td);
409 #endif
410 	/*
411 	 * Release reference to text vnode
412 	 */
413 	if (p->p_textvp != NULL) {
414 		vrele(p->p_textvp);
415 		p->p_textvp = NULL;
416 	}
417 
418 	/*
419 	 * Release our limits structure.
420 	 */
421 	lim_free(p->p_limit);
422 	p->p_limit = NULL;
423 
424 	tidhash_remove(td);
425 
426 	/*
427 	 * Call machine-dependent code to release any
428 	 * machine-dependent resources other than the address space.
429 	 * The address space is released by "vmspace_exitfree(p)" in
430 	 * vm_waitproc().
431 	 */
432 	cpu_exit(td);
433 
434 	WITNESS_WARN(WARN_PANIC, NULL, "process (pid %d) exiting", p->p_pid);
435 
436 	sx_xlock(&proctree_lock);
437 	/*
438 	 * Remove proc from allproc queue and pidhash chain.
439 	 * Place onto zombproc.  Unlink from parent's child list.
440 	 */
441 	sx_xlock(&allproc_lock);
442 	LIST_REMOVE(p, p_list);
443 	LIST_INSERT_HEAD(&zombproc, p, p_list);
444 	LIST_REMOVE(p, p_hash);
445 	sx_xunlock(&allproc_lock);
446 
447 	/*
448 	 * Reparent all children processes:
449 	 * - traced ones to the original parent (or init if we are that parent)
450 	 * - the rest to init
451 	 */
452 	q = LIST_FIRST(&p->p_children);
453 	if (q != NULL)		/* only need this if any child is S_ZOMB */
454 		wakeup(q->p_reaper);
455 	for (; q != NULL; q = nq) {
456 		nq = LIST_NEXT(q, p_sibling);
457 		ksi = ksiginfo_alloc(TRUE);
458 		PROC_LOCK(q);
459 		q->p_sigparent = SIGCHLD;
460 
461 		if (!(q->p_flag & P_TRACED)) {
462 			proc_reparent(q, q->p_reaper);
463 			if (q->p_state == PRS_ZOMBIE) {
464 				/*
465 				 * Inform reaper about the reparented
466 				 * zombie, since wait(2) has something
467 				 * new to report.  Guarantee queueing
468 				 * of the SIGCHLD signal, similar to
469 				 * the _exit() behaviour, by providing
470 				 * our ksiginfo.  Ksi is freed by the
471 				 * signal delivery.
472 				 */
473 				if (q->p_ksi == NULL) {
474 					ksi1 = NULL;
475 				} else {
476 					ksiginfo_copy(q->p_ksi, ksi);
477 					ksi->ksi_flags |= KSI_INS;
478 					ksi1 = ksi;
479 					ksi = NULL;
480 				}
481 				PROC_LOCK(q->p_reaper);
482 				pksignal(q->p_reaper, SIGCHLD, ksi1);
483 				PROC_UNLOCK(q->p_reaper);
484 			}
485 		} else {
486 			/*
487 			 * Traced processes are killed since their existence
488 			 * means someone is screwing up.
489 			 */
490 			t = proc_realparent(q);
491 			if (t == p) {
492 				proc_reparent(q, q->p_reaper);
493 			} else {
494 				PROC_LOCK(t);
495 				proc_reparent(q, t);
496 				PROC_UNLOCK(t);
497 			}
498 			/*
499 			 * Since q was found on our children list, the
500 			 * proc_reparent() call moved q to the orphan
501 			 * list due to present P_TRACED flag. Clear
502 			 * orphan link for q now while q is locked.
503 			 */
504 			clear_orphan(q);
505 			q->p_flag &= ~(P_TRACED | P_STOPPED_TRACE);
506 			q->p_flag2 &= ~P2_PTRACE_FSTP;
507 			q->p_ptevents = 0;
508 			FOREACH_THREAD_IN_PROC(q, tdt) {
509 				tdt->td_dbgflags &= ~(TDB_SUSPEND | TDB_XSIG |
510 				    TDB_FSTP);
511 			}
512 			kern_psignal(q, SIGKILL);
513 		}
514 		PROC_UNLOCK(q);
515 		if (ksi != NULL)
516 			ksiginfo_free(ksi);
517 	}
518 
519 	/*
520 	 * Also get rid of our orphans.
521 	 */
522 	while ((q = LIST_FIRST(&p->p_orphans)) != NULL) {
523 		PROC_LOCK(q);
524 		CTR2(KTR_PTRACE, "exit: pid %d, clearing orphan %d", p->p_pid,
525 		    q->p_pid);
526 		clear_orphan(q);
527 		PROC_UNLOCK(q);
528 	}
529 
530 	/* Save exit status. */
531 	PROC_LOCK(p);
532 	p->p_xthread = td;
533 
534 #ifdef KDTRACE_HOOKS
535 	/*
536 	 * Tell the DTrace fasttrap provider about the exit if it
537 	 * has declared an interest.
538 	 */
539 	if (dtrace_fasttrap_exit)
540 		dtrace_fasttrap_exit(p);
541 #endif
542 
543 	/*
544 	 * Notify interested parties of our demise.
545 	 */
546 	KNOTE_LOCKED(p->p_klist, NOTE_EXIT);
547 
548 #ifdef KDTRACE_HOOKS
549 	int reason = CLD_EXITED;
550 	if (WCOREDUMP(signo))
551 		reason = CLD_DUMPED;
552 	else if (WIFSIGNALED(signo))
553 		reason = CLD_KILLED;
554 	SDT_PROBE1(proc, , , exit, reason);
555 #endif
556 
557 	/*
558 	 * If this is a process with a descriptor, we may not need to deliver
559 	 * a signal to the parent.  proctree_lock is held over
560 	 * procdesc_exit() to serialize concurrent calls to close() and
561 	 * exit().
562 	 */
563 	signal_parent = 0;
564 	if (p->p_procdesc == NULL || procdesc_exit(p)) {
565 		/*
566 		 * Notify parent that we're gone.  If parent has the
567 		 * PS_NOCLDWAIT flag set, or if the handler is set to SIG_IGN,
568 		 * notify process 1 instead (and hope it will handle this
569 		 * situation).
570 		 */
571 		PROC_LOCK(p->p_pptr);
572 		mtx_lock(&p->p_pptr->p_sigacts->ps_mtx);
573 		if (p->p_pptr->p_sigacts->ps_flag &
574 		    (PS_NOCLDWAIT | PS_CLDSIGIGN)) {
575 			struct proc *pp;
576 
577 			mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx);
578 			pp = p->p_pptr;
579 			PROC_UNLOCK(pp);
580 			proc_reparent(p, p->p_reaper);
581 			p->p_sigparent = SIGCHLD;
582 			PROC_LOCK(p->p_pptr);
583 
584 			/*
585 			 * Notify parent, so in case he was wait(2)ing or
586 			 * executing waitpid(2) with our pid, he will
587 			 * continue.
588 			 */
589 			wakeup(pp);
590 		} else
591 			mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx);
592 
593 		if (p->p_pptr == p->p_reaper || p->p_pptr == initproc) {
594 			signal_parent = 1;
595 		} else if (p->p_sigparent != 0) {
596 			if (p->p_sigparent == SIGCHLD) {
597 				signal_parent = 1;
598 			} else { /* LINUX thread */
599 				signal_parent = 2;
600 			}
601 		}
602 	} else
603 		PROC_LOCK(p->p_pptr);
604 	sx_xunlock(&proctree_lock);
605 
606 	if (signal_parent == 1) {
607 		childproc_exited(p);
608 	} else if (signal_parent == 2) {
609 		kern_psignal(p->p_pptr, p->p_sigparent);
610 	}
611 
612 	/* Tell the prison that we are gone. */
613 	prison_proc_free(p->p_ucred->cr_prison);
614 
615 	/*
616 	 * The state PRS_ZOMBIE prevents other proesses from sending
617 	 * signal to the process, to avoid memory leak, we free memory
618 	 * for signal queue at the time when the state is set.
619 	 */
620 	sigqueue_flush(&p->p_sigqueue);
621 	sigqueue_flush(&td->td_sigqueue);
622 
623 	/*
624 	 * We have to wait until after acquiring all locks before
625 	 * changing p_state.  We need to avoid all possible context
626 	 * switches (including ones from blocking on a mutex) while
627 	 * marked as a zombie.  We also have to set the zombie state
628 	 * before we release the parent process' proc lock to avoid
629 	 * a lost wakeup.  So, we first call wakeup, then we grab the
630 	 * sched lock, update the state, and release the parent process'
631 	 * proc lock.
632 	 */
633 	wakeup(p->p_pptr);
634 	cv_broadcast(&p->p_pwait);
635 	sched_exit(p->p_pptr, td);
636 	PROC_SLOCK(p);
637 	p->p_state = PRS_ZOMBIE;
638 	PROC_UNLOCK(p->p_pptr);
639 
640 	/*
641 	 * Save our children's rusage information in our exit rusage.
642 	 */
643 	PROC_STATLOCK(p);
644 	ruadd(&p->p_ru, &p->p_rux, &p->p_stats->p_cru, &p->p_crux);
645 	PROC_STATUNLOCK(p);
646 
647 	/*
648 	 * Make sure the scheduler takes this thread out of its tables etc.
649 	 * This will also release this thread's reference to the ucred.
650 	 * Other thread parts to release include pcb bits and such.
651 	 */
652 	thread_exit();
653 }
654 
655 
656 #ifndef _SYS_SYSPROTO_H_
657 struct abort2_args {
658 	char *why;
659 	int nargs;
660 	void **args;
661 };
662 #endif
663 
664 int
665 sys_abort2(struct thread *td, struct abort2_args *uap)
666 {
667 	struct proc *p = td->td_proc;
668 	struct sbuf *sb;
669 	void *uargs[16];
670 	int error, i, sig;
671 
672 	/*
673 	 * Do it right now so we can log either proper call of abort2(), or
674 	 * note, that invalid argument was passed. 512 is big enough to
675 	 * handle 16 arguments' descriptions with additional comments.
676 	 */
677 	sb = sbuf_new(NULL, NULL, 512, SBUF_FIXEDLEN);
678 	sbuf_clear(sb);
679 	sbuf_printf(sb, "%s(pid %d uid %d) aborted: ",
680 	    p->p_comm, p->p_pid, td->td_ucred->cr_uid);
681 	/*
682 	 * Since we can't return from abort2(), send SIGKILL in cases, where
683 	 * abort2() was called improperly
684 	 */
685 	sig = SIGKILL;
686 	/* Prevent from DoSes from user-space. */
687 	if (uap->nargs < 0 || uap->nargs > 16)
688 		goto out;
689 	if (uap->nargs > 0) {
690 		if (uap->args == NULL)
691 			goto out;
692 		error = copyin(uap->args, uargs, uap->nargs * sizeof(void *));
693 		if (error != 0)
694 			goto out;
695 	}
696 	/*
697 	 * Limit size of 'reason' string to 128. Will fit even when
698 	 * maximal number of arguments was chosen to be logged.
699 	 */
700 	if (uap->why != NULL) {
701 		error = sbuf_copyin(sb, uap->why, 128);
702 		if (error < 0)
703 			goto out;
704 	} else {
705 		sbuf_printf(sb, "(null)");
706 	}
707 	if (uap->nargs > 0) {
708 		sbuf_printf(sb, "(");
709 		for (i = 0;i < uap->nargs; i++)
710 			sbuf_printf(sb, "%s%p", i == 0 ? "" : ", ", uargs[i]);
711 		sbuf_printf(sb, ")");
712 	}
713 	/*
714 	 * Final stage: arguments were proper, string has been
715 	 * successfully copied from userspace, and copying pointers
716 	 * from user-space succeed.
717 	 */
718 	sig = SIGABRT;
719 out:
720 	if (sig == SIGKILL) {
721 		sbuf_trim(sb);
722 		sbuf_printf(sb, " (Reason text inaccessible)");
723 	}
724 	sbuf_cat(sb, "\n");
725 	sbuf_finish(sb);
726 	log(LOG_INFO, "%s", sbuf_data(sb));
727 	sbuf_delete(sb);
728 	exit1(td, 0, sig);
729 	return (0);
730 }
731 
732 
733 #ifdef COMPAT_43
734 /*
735  * The dirty work is handled by kern_wait().
736  */
737 int
738 owait(struct thread *td, struct owait_args *uap __unused)
739 {
740 	int error, status;
741 
742 	error = kern_wait(td, WAIT_ANY, &status, 0, NULL);
743 	if (error == 0)
744 		td->td_retval[1] = status;
745 	return (error);
746 }
747 #endif /* COMPAT_43 */
748 
749 /*
750  * The dirty work is handled by kern_wait().
751  */
752 int
753 sys_wait4(struct thread *td, struct wait4_args *uap)
754 {
755 	struct rusage ru, *rup;
756 	int error, status;
757 
758 	if (uap->rusage != NULL)
759 		rup = &ru;
760 	else
761 		rup = NULL;
762 	error = kern_wait(td, uap->pid, &status, uap->options, rup);
763 	if (uap->status != NULL && error == 0 && td->td_retval[0] != 0)
764 		error = copyout(&status, uap->status, sizeof(status));
765 	if (uap->rusage != NULL && error == 0 && td->td_retval[0] != 0)
766 		error = copyout(&ru, uap->rusage, sizeof(struct rusage));
767 	return (error);
768 }
769 
770 int
771 sys_wait6(struct thread *td, struct wait6_args *uap)
772 {
773 	struct __wrusage wru, *wrup;
774 	siginfo_t si, *sip;
775 	idtype_t idtype;
776 	id_t id;
777 	int error, status;
778 
779 	idtype = uap->idtype;
780 	id = uap->id;
781 
782 	if (uap->wrusage != NULL)
783 		wrup = &wru;
784 	else
785 		wrup = NULL;
786 
787 	if (uap->info != NULL) {
788 		sip = &si;
789 		bzero(sip, sizeof(*sip));
790 	} else
791 		sip = NULL;
792 
793 	/*
794 	 *  We expect all callers of wait6() to know about WEXITED and
795 	 *  WTRAPPED.
796 	 */
797 	error = kern_wait6(td, idtype, id, &status, uap->options, wrup, sip);
798 
799 	if (uap->status != NULL && error == 0 && td->td_retval[0] != 0)
800 		error = copyout(&status, uap->status, sizeof(status));
801 	if (uap->wrusage != NULL && error == 0 && td->td_retval[0] != 0)
802 		error = copyout(&wru, uap->wrusage, sizeof(wru));
803 	if (uap->info != NULL && error == 0)
804 		error = copyout(&si, uap->info, sizeof(si));
805 	return (error);
806 }
807 
808 /*
809  * Reap the remains of a zombie process and optionally return status and
810  * rusage.  Asserts and will release both the proctree_lock and the process
811  * lock as part of its work.
812  */
813 void
814 proc_reap(struct thread *td, struct proc *p, int *status, int options)
815 {
816 	struct proc *q, *t;
817 
818 	sx_assert(&proctree_lock, SA_XLOCKED);
819 	PROC_LOCK_ASSERT(p, MA_OWNED);
820 	KASSERT(p->p_state == PRS_ZOMBIE, ("proc_reap: !PRS_ZOMBIE"));
821 
822 	mtx_spin_wait_unlocked(&p->p_slock);
823 
824 	q = td->td_proc;
825 
826 	if (status)
827 		*status = KW_EXITCODE(p->p_xexit, p->p_xsig);
828 	if (options & WNOWAIT) {
829 		/*
830 		 *  Only poll, returning the status.  Caller does not wish to
831 		 * release the proc struct just yet.
832 		 */
833 		PROC_UNLOCK(p);
834 		sx_xunlock(&proctree_lock);
835 		return;
836 	}
837 
838 	PROC_LOCK(q);
839 	sigqueue_take(p->p_ksi);
840 	PROC_UNLOCK(q);
841 
842 	/*
843 	 * If we got the child via a ptrace 'attach', we need to give it back
844 	 * to the old parent.
845 	 */
846 	if (p->p_oppid != 0 && p->p_oppid != p->p_pptr->p_pid) {
847 		PROC_UNLOCK(p);
848 		t = proc_realparent(p);
849 		PROC_LOCK(t);
850 		PROC_LOCK(p);
851 		CTR2(KTR_PTRACE,
852 		    "wait: traced child %d moved back to parent %d", p->p_pid,
853 		    t->p_pid);
854 		proc_reparent(p, t);
855 		p->p_oppid = 0;
856 		PROC_UNLOCK(p);
857 		pksignal(t, SIGCHLD, p->p_ksi);
858 		wakeup(t);
859 		cv_broadcast(&p->p_pwait);
860 		PROC_UNLOCK(t);
861 		sx_xunlock(&proctree_lock);
862 		return;
863 	}
864 	p->p_oppid = 0;
865 	PROC_UNLOCK(p);
866 
867 	/*
868 	 * Remove other references to this process to ensure we have an
869 	 * exclusive reference.
870 	 */
871 	sx_xlock(&allproc_lock);
872 	LIST_REMOVE(p, p_list);	/* off zombproc */
873 	sx_xunlock(&allproc_lock);
874 	LIST_REMOVE(p, p_sibling);
875 	reaper_abandon_children(p, true);
876 	LIST_REMOVE(p, p_reapsibling);
877 	PROC_LOCK(p);
878 	clear_orphan(p);
879 	PROC_UNLOCK(p);
880 	leavepgrp(p);
881 	if (p->p_procdesc != NULL)
882 		procdesc_reap(p);
883 	sx_xunlock(&proctree_lock);
884 
885 	PROC_LOCK(p);
886 	knlist_detach(p->p_klist);
887 	p->p_klist = NULL;
888 	PROC_UNLOCK(p);
889 
890 	/*
891 	 * Removal from allproc list and process group list paired with
892 	 * PROC_LOCK which was executed during that time should guarantee
893 	 * nothing can reach this process anymore. As such further locking
894 	 * is unnecessary.
895 	 */
896 	p->p_xexit = p->p_xsig = 0;		/* XXX: why? */
897 
898 	PROC_LOCK(q);
899 	ruadd(&q->p_stats->p_cru, &q->p_crux, &p->p_ru, &p->p_rux);
900 	PROC_UNLOCK(q);
901 
902 	/*
903 	 * Decrement the count of procs running with this uid.
904 	 */
905 	(void)chgproccnt(p->p_ucred->cr_ruidinfo, -1, 0);
906 
907 	/*
908 	 * Destroy resource accounting information associated with the process.
909 	 */
910 #ifdef RACCT
911 	if (racct_enable) {
912 		PROC_LOCK(p);
913 		racct_sub(p, RACCT_NPROC, 1);
914 		PROC_UNLOCK(p);
915 	}
916 #endif
917 	racct_proc_exit(p);
918 
919 	/*
920 	 * Free credentials, arguments, and sigacts.
921 	 */
922 	crfree(p->p_ucred);
923 	proc_set_cred(p, NULL);
924 	pargs_drop(p->p_args);
925 	p->p_args = NULL;
926 	sigacts_free(p->p_sigacts);
927 	p->p_sigacts = NULL;
928 
929 	/*
930 	 * Do any thread-system specific cleanups.
931 	 */
932 	thread_wait(p);
933 
934 	/*
935 	 * Give vm and machine-dependent layer a chance to free anything that
936 	 * cpu_exit couldn't release while still running in process context.
937 	 */
938 	vm_waitproc(p);
939 #ifdef MAC
940 	mac_proc_destroy(p);
941 #endif
942 
943 	KASSERT(FIRST_THREAD_IN_PROC(p),
944 	    ("proc_reap: no residual thread!"));
945 	uma_zfree(proc_zone, p);
946 	atomic_add_int(&nprocs, -1);
947 }
948 
949 static int
950 proc_to_reap(struct thread *td, struct proc *p, idtype_t idtype, id_t id,
951     int *status, int options, struct __wrusage *wrusage, siginfo_t *siginfo,
952     int check_only)
953 {
954 	struct rusage *rup;
955 
956 	sx_assert(&proctree_lock, SA_XLOCKED);
957 
958 	PROC_LOCK(p);
959 
960 	switch (idtype) {
961 	case P_ALL:
962 		if (p->p_procdesc != NULL) {
963 			PROC_UNLOCK(p);
964 			return (0);
965 		}
966 		break;
967 	case P_PID:
968 		if (p->p_pid != (pid_t)id) {
969 			PROC_UNLOCK(p);
970 			return (0);
971 		}
972 		break;
973 	case P_PGID:
974 		if (p->p_pgid != (pid_t)id) {
975 			PROC_UNLOCK(p);
976 			return (0);
977 		}
978 		break;
979 	case P_SID:
980 		if (p->p_session->s_sid != (pid_t)id) {
981 			PROC_UNLOCK(p);
982 			return (0);
983 		}
984 		break;
985 	case P_UID:
986 		if (p->p_ucred->cr_uid != (uid_t)id) {
987 			PROC_UNLOCK(p);
988 			return (0);
989 		}
990 		break;
991 	case P_GID:
992 		if (p->p_ucred->cr_gid != (gid_t)id) {
993 			PROC_UNLOCK(p);
994 			return (0);
995 		}
996 		break;
997 	case P_JAILID:
998 		if (p->p_ucred->cr_prison->pr_id != (int)id) {
999 			PROC_UNLOCK(p);
1000 			return (0);
1001 		}
1002 		break;
1003 	/*
1004 	 * It seems that the thread structures get zeroed out
1005 	 * at process exit.  This makes it impossible to
1006 	 * support P_SETID, P_CID or P_CPUID.
1007 	 */
1008 	default:
1009 		PROC_UNLOCK(p);
1010 		return (0);
1011 	}
1012 
1013 	if (p_canwait(td, p)) {
1014 		PROC_UNLOCK(p);
1015 		return (0);
1016 	}
1017 
1018 	if (((options & WEXITED) == 0) && (p->p_state == PRS_ZOMBIE)) {
1019 		PROC_UNLOCK(p);
1020 		return (0);
1021 	}
1022 
1023 	/*
1024 	 * This special case handles a kthread spawned by linux_clone
1025 	 * (see linux_misc.c).  The linux_wait4 and linux_waitpid
1026 	 * functions need to be able to distinguish between waiting
1027 	 * on a process and waiting on a thread.  It is a thread if
1028 	 * p_sigparent is not SIGCHLD, and the WLINUXCLONE option
1029 	 * signifies we want to wait for threads and not processes.
1030 	 */
1031 	if ((p->p_sigparent != SIGCHLD) ^
1032 	    ((options & WLINUXCLONE) != 0)) {
1033 		PROC_UNLOCK(p);
1034 		return (0);
1035 	}
1036 
1037 	if (siginfo != NULL) {
1038 		bzero(siginfo, sizeof(*siginfo));
1039 		siginfo->si_errno = 0;
1040 
1041 		/*
1042 		 * SUSv4 requires that the si_signo value is always
1043 		 * SIGCHLD. Obey it despite the rfork(2) interface
1044 		 * allows to request other signal for child exit
1045 		 * notification.
1046 		 */
1047 		siginfo->si_signo = SIGCHLD;
1048 
1049 		/*
1050 		 *  This is still a rough estimate.  We will fix the
1051 		 *  cases TRAPPED, STOPPED, and CONTINUED later.
1052 		 */
1053 		if (WCOREDUMP(p->p_xsig)) {
1054 			siginfo->si_code = CLD_DUMPED;
1055 			siginfo->si_status = WTERMSIG(p->p_xsig);
1056 		} else if (WIFSIGNALED(p->p_xsig)) {
1057 			siginfo->si_code = CLD_KILLED;
1058 			siginfo->si_status = WTERMSIG(p->p_xsig);
1059 		} else {
1060 			siginfo->si_code = CLD_EXITED;
1061 			siginfo->si_status = p->p_xexit;
1062 		}
1063 
1064 		siginfo->si_pid = p->p_pid;
1065 		siginfo->si_uid = p->p_ucred->cr_uid;
1066 
1067 		/*
1068 		 * The si_addr field would be useful additional
1069 		 * detail, but apparently the PC value may be lost
1070 		 * when we reach this point.  bzero() above sets
1071 		 * siginfo->si_addr to NULL.
1072 		 */
1073 	}
1074 
1075 	/*
1076 	 * There should be no reason to limit resources usage info to
1077 	 * exited processes only.  A snapshot about any resources used
1078 	 * by a stopped process may be exactly what is needed.
1079 	 */
1080 	if (wrusage != NULL) {
1081 		rup = &wrusage->wru_self;
1082 		*rup = p->p_ru;
1083 		PROC_STATLOCK(p);
1084 		calcru(p, &rup->ru_utime, &rup->ru_stime);
1085 		PROC_STATUNLOCK(p);
1086 
1087 		rup = &wrusage->wru_children;
1088 		*rup = p->p_stats->p_cru;
1089 		calccru(p, &rup->ru_utime, &rup->ru_stime);
1090 	}
1091 
1092 	if (p->p_state == PRS_ZOMBIE && !check_only) {
1093 		proc_reap(td, p, status, options);
1094 		return (-1);
1095 	}
1096 	return (1);
1097 }
1098 
1099 int
1100 kern_wait(struct thread *td, pid_t pid, int *status, int options,
1101     struct rusage *rusage)
1102 {
1103 	struct __wrusage wru, *wrup;
1104 	idtype_t idtype;
1105 	id_t id;
1106 	int ret;
1107 
1108 	/*
1109 	 * Translate the special pid values into the (idtype, pid)
1110 	 * pair for kern_wait6.  The WAIT_MYPGRP case is handled by
1111 	 * kern_wait6() on its own.
1112 	 */
1113 	if (pid == WAIT_ANY) {
1114 		idtype = P_ALL;
1115 		id = 0;
1116 	} else if (pid < 0) {
1117 		idtype = P_PGID;
1118 		id = (id_t)-pid;
1119 	} else {
1120 		idtype = P_PID;
1121 		id = (id_t)pid;
1122 	}
1123 
1124 	if (rusage != NULL)
1125 		wrup = &wru;
1126 	else
1127 		wrup = NULL;
1128 
1129 	/*
1130 	 * For backward compatibility we implicitly add flags WEXITED
1131 	 * and WTRAPPED here.
1132 	 */
1133 	options |= WEXITED | WTRAPPED;
1134 	ret = kern_wait6(td, idtype, id, status, options, wrup, NULL);
1135 	if (rusage != NULL)
1136 		*rusage = wru.wru_self;
1137 	return (ret);
1138 }
1139 
1140 static void
1141 report_alive_proc(struct thread *td, struct proc *p, siginfo_t *siginfo,
1142     int *status, int options, int si_code)
1143 {
1144 	bool cont;
1145 
1146 	PROC_LOCK_ASSERT(p, MA_OWNED);
1147 	sx_assert(&proctree_lock, SA_XLOCKED);
1148 	MPASS(si_code == CLD_TRAPPED || si_code == CLD_STOPPED ||
1149 	    si_code == CLD_CONTINUED);
1150 
1151 	cont = si_code == CLD_CONTINUED;
1152 	if ((options & WNOWAIT) == 0) {
1153 		if (cont)
1154 			p->p_flag &= ~P_CONTINUED;
1155 		else
1156 			p->p_flag |= P_WAITED;
1157 		PROC_LOCK(td->td_proc);
1158 		sigqueue_take(p->p_ksi);
1159 		PROC_UNLOCK(td->td_proc);
1160 	}
1161 	sx_xunlock(&proctree_lock);
1162 	if (siginfo != NULL) {
1163 		siginfo->si_code = si_code;
1164 		siginfo->si_status = cont ? SIGCONT : p->p_xsig;
1165 	}
1166 	if (status != NULL)
1167 		*status = cont ? SIGCONT : W_STOPCODE(p->p_xsig);
1168 	PROC_UNLOCK(p);
1169 	td->td_retval[0] = p->p_pid;
1170 }
1171 
1172 int
1173 kern_wait6(struct thread *td, idtype_t idtype, id_t id, int *status,
1174     int options, struct __wrusage *wrusage, siginfo_t *siginfo)
1175 {
1176 	struct proc *p, *q;
1177 	pid_t pid;
1178 	int error, nfound, ret;
1179 	bool report;
1180 
1181 	AUDIT_ARG_VALUE((int)idtype);	/* XXX - This is likely wrong! */
1182 	AUDIT_ARG_PID((pid_t)id);	/* XXX - This may be wrong! */
1183 	AUDIT_ARG_VALUE(options);
1184 
1185 	q = td->td_proc;
1186 
1187 	if ((pid_t)id == WAIT_MYPGRP && (idtype == P_PID || idtype == P_PGID)) {
1188 		PROC_LOCK(q);
1189 		id = (id_t)q->p_pgid;
1190 		PROC_UNLOCK(q);
1191 		idtype = P_PGID;
1192 	}
1193 
1194 	/* If we don't know the option, just return. */
1195 	if ((options & ~(WUNTRACED | WNOHANG | WCONTINUED | WNOWAIT |
1196 	    WEXITED | WTRAPPED | WLINUXCLONE)) != 0)
1197 		return (EINVAL);
1198 	if ((options & (WEXITED | WUNTRACED | WCONTINUED | WTRAPPED)) == 0) {
1199 		/*
1200 		 * We will be unable to find any matching processes,
1201 		 * because there are no known events to look for.
1202 		 * Prefer to return error instead of blocking
1203 		 * indefinitely.
1204 		 */
1205 		return (EINVAL);
1206 	}
1207 
1208 loop:
1209 	if (q->p_flag & P_STATCHILD) {
1210 		PROC_LOCK(q);
1211 		q->p_flag &= ~P_STATCHILD;
1212 		PROC_UNLOCK(q);
1213 	}
1214 	sx_xlock(&proctree_lock);
1215 loop_locked:
1216 	nfound = 0;
1217 	LIST_FOREACH(p, &q->p_children, p_sibling) {
1218 		pid = p->p_pid;
1219 		ret = proc_to_reap(td, p, idtype, id, status, options,
1220 		    wrusage, siginfo, 0);
1221 		if (ret == 0)
1222 			continue;
1223 		else if (ret != 1) {
1224 			td->td_retval[0] = pid;
1225 			return (0);
1226 		}
1227 
1228 		nfound++;
1229 		PROC_LOCK_ASSERT(p, MA_OWNED);
1230 
1231 		if ((options & WTRAPPED) != 0 &&
1232 		    (p->p_flag & P_TRACED) != 0) {
1233 			PROC_SLOCK(p);
1234 			report =
1235 			    ((p->p_flag & (P_STOPPED_TRACE | P_STOPPED_SIG)) &&
1236 			    p->p_suspcount == p->p_numthreads &&
1237 			    (p->p_flag & P_WAITED) == 0);
1238 			PROC_SUNLOCK(p);
1239 			if (report) {
1240 			CTR4(KTR_PTRACE,
1241 			    "wait: returning trapped pid %d status %#x "
1242 			    "(xstat %d) xthread %d",
1243 			    p->p_pid, W_STOPCODE(p->p_xsig), p->p_xsig,
1244 			    p->p_xthread != NULL ?
1245 			    p->p_xthread->td_tid : -1);
1246 				report_alive_proc(td, p, siginfo, status,
1247 				    options, CLD_TRAPPED);
1248 				return (0);
1249 			}
1250 		}
1251 		if ((options & WUNTRACED) != 0 &&
1252 		    (p->p_flag & P_STOPPED_SIG) != 0) {
1253 			PROC_SLOCK(p);
1254 			report = (p->p_suspcount == p->p_numthreads &&
1255 			    ((p->p_flag & P_WAITED) == 0));
1256 			PROC_SUNLOCK(p);
1257 			if (report) {
1258 				report_alive_proc(td, p, siginfo, status,
1259 				    options, CLD_STOPPED);
1260 				return (0);
1261 			}
1262 		}
1263 		if ((options & WCONTINUED) != 0 &&
1264 		    (p->p_flag & P_CONTINUED) != 0) {
1265 			report_alive_proc(td, p, siginfo, status, options,
1266 			    CLD_CONTINUED);
1267 			return (0);
1268 		}
1269 		PROC_UNLOCK(p);
1270 	}
1271 
1272 	/*
1273 	 * Look in the orphans list too, to allow the parent to
1274 	 * collect it's child exit status even if child is being
1275 	 * debugged.
1276 	 *
1277 	 * Debugger detaches from the parent upon successful
1278 	 * switch-over from parent to child.  At this point due to
1279 	 * re-parenting the parent loses the child to debugger and a
1280 	 * wait4(2) call would report that it has no children to wait
1281 	 * for.  By maintaining a list of orphans we allow the parent
1282 	 * to successfully wait until the child becomes a zombie.
1283 	 */
1284 	if (nfound == 0) {
1285 		LIST_FOREACH(p, &q->p_orphans, p_orphan) {
1286 			ret = proc_to_reap(td, p, idtype, id, NULL, options,
1287 			    NULL, NULL, 1);
1288 			if (ret != 0) {
1289 				KASSERT(ret != -1, ("reaped an orphan (pid %d)",
1290 				    (int)td->td_retval[0]));
1291 				PROC_UNLOCK(p);
1292 				nfound++;
1293 				break;
1294 			}
1295 		}
1296 	}
1297 	if (nfound == 0) {
1298 		sx_xunlock(&proctree_lock);
1299 		return (ECHILD);
1300 	}
1301 	if (options & WNOHANG) {
1302 		sx_xunlock(&proctree_lock);
1303 		td->td_retval[0] = 0;
1304 		return (0);
1305 	}
1306 	PROC_LOCK(q);
1307 	if (q->p_flag & P_STATCHILD) {
1308 		q->p_flag &= ~P_STATCHILD;
1309 		PROC_UNLOCK(q);
1310 		goto loop_locked;
1311 	}
1312 	sx_xunlock(&proctree_lock);
1313 	error = msleep(q, &q->p_mtx, PWAIT | PCATCH | PDROP, "wait", 0);
1314 	if (error)
1315 		return (error);
1316 	goto loop;
1317 }
1318 
1319 /*
1320  * Make process 'parent' the new parent of process 'child'.
1321  * Must be called with an exclusive hold of proctree lock.
1322  */
1323 void
1324 proc_reparent(struct proc *child, struct proc *parent)
1325 {
1326 
1327 	sx_assert(&proctree_lock, SX_XLOCKED);
1328 	PROC_LOCK_ASSERT(child, MA_OWNED);
1329 	if (child->p_pptr == parent)
1330 		return;
1331 
1332 	PROC_LOCK(child->p_pptr);
1333 	sigqueue_take(child->p_ksi);
1334 	PROC_UNLOCK(child->p_pptr);
1335 	LIST_REMOVE(child, p_sibling);
1336 	LIST_INSERT_HEAD(&parent->p_children, child, p_sibling);
1337 
1338 	clear_orphan(child);
1339 	if (child->p_flag & P_TRACED) {
1340 		if (LIST_EMPTY(&child->p_pptr->p_orphans)) {
1341 			child->p_treeflag |= P_TREE_FIRST_ORPHAN;
1342 			LIST_INSERT_HEAD(&child->p_pptr->p_orphans, child,
1343 			    p_orphan);
1344 		} else {
1345 			LIST_INSERT_AFTER(LIST_FIRST(&child->p_pptr->p_orphans),
1346 			    child, p_orphan);
1347 		}
1348 		child->p_treeflag |= P_TREE_ORPHANED;
1349 	}
1350 
1351 	child->p_pptr = parent;
1352 }
1353