xref: /freebsd/sys/kern/kern_exit.c (revision f5b7695d2d5abd735064870ad43f4b9c723940c1)
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_ktrace.h"
43 
44 #include <sys/param.h>
45 #include <sys/systm.h>
46 #include <sys/sysproto.h>
47 #include <sys/capsicum.h>
48 #include <sys/eventhandler.h>
49 #include <sys/kernel.h>
50 #include <sys/ktr.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/jail.h>
57 #include <sys/tty.h>
58 #include <sys/wait.h>
59 #include <sys/vmmeter.h>
60 #include <sys/vnode.h>
61 #include <sys/racct.h>
62 #include <sys/resourcevar.h>
63 #include <sys/sbuf.h>
64 #include <sys/signalvar.h>
65 #include <sys/sched.h>
66 #include <sys/sx.h>
67 #include <sys/syscallsubr.h>
68 #include <sys/syslog.h>
69 #include <sys/ptrace.h>
70 #include <sys/acct.h>		/* for acct_process() function prototype */
71 #include <sys/filedesc.h>
72 #include <sys/sdt.h>
73 #include <sys/shm.h>
74 #include <sys/sem.h>
75 #include <sys/umtx.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, , , 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 		return (child->p_pptr->p_pid == child->p_oppid ?
110 			    child->p_pptr : initproc);
111 	for (p = child; (p->p_treeflag & P_TREE_FIRST_ORPHAN) == 0;) {
112 		/* Cannot use LIST_PREV(), since the list head is not known. */
113 		p = __containerof(p->p_orphan.le_prev, struct proc,
114 		    p_orphan.le_next);
115 		KASSERT((p->p_treeflag & P_TREE_ORPHANED) != 0,
116 		    ("missing P_ORPHAN %p", p));
117 	}
118 	parent = __containerof(p->p_orphan.le_prev, struct proc,
119 	    p_orphans.lh_first);
120 	return (parent);
121 }
122 
123 void
124 reaper_abandon_children(struct proc *p, bool exiting)
125 {
126 	struct proc *p1, *p2, *ptmp;
127 
128 	sx_assert(&proctree_lock, SX_LOCKED);
129 	KASSERT(p != initproc, ("reaper_abandon_children for initproc"));
130 	if ((p->p_treeflag & P_TREE_REAPER) == 0)
131 		return;
132 	p1 = p->p_reaper;
133 	LIST_FOREACH_SAFE(p2, &p->p_reaplist, p_reapsibling, ptmp) {
134 		LIST_REMOVE(p2, p_reapsibling);
135 		p2->p_reaper = p1;
136 		p2->p_reapsubtree = p->p_reapsubtree;
137 		LIST_INSERT_HEAD(&p1->p_reaplist, p2, p_reapsibling);
138 		if (exiting && p2->p_pptr == p) {
139 			PROC_LOCK(p2);
140 			proc_reparent(p2, p1, true);
141 			PROC_UNLOCK(p2);
142 		}
143 	}
144 	KASSERT(LIST_EMPTY(&p->p_reaplist), ("p_reaplist not empty"));
145 	p->p_treeflag &= ~P_TREE_REAPER;
146 }
147 
148 static void
149 reaper_clear(struct proc *p)
150 {
151 	struct proc *p1;
152 	bool clear;
153 
154 	sx_assert(&proctree_lock, SX_LOCKED);
155 	LIST_REMOVE(p, p_reapsibling);
156 	if (p->p_reapsubtree == 1)
157 		return;
158 	clear = true;
159 	LIST_FOREACH(p1, &p->p_reaper->p_reaplist, p_reapsibling) {
160 		if (p1->p_reapsubtree == p->p_reapsubtree) {
161 			clear = false;
162 			break;
163 		}
164 	}
165 	if (clear)
166 		proc_id_clear(PROC_ID_REAP, p->p_reapsubtree);
167 }
168 
169 void
170 proc_clear_orphan(struct proc *p)
171 {
172 	struct proc *p1;
173 
174 	sx_assert(&proctree_lock, SA_XLOCKED);
175 	if ((p->p_treeflag & P_TREE_ORPHANED) == 0)
176 		return;
177 	if ((p->p_treeflag & P_TREE_FIRST_ORPHAN) != 0) {
178 		p1 = LIST_NEXT(p, p_orphan);
179 		if (p1 != NULL)
180 			p1->p_treeflag |= P_TREE_FIRST_ORPHAN;
181 		p->p_treeflag &= ~P_TREE_FIRST_ORPHAN;
182 	}
183 	LIST_REMOVE(p, p_orphan);
184 	p->p_treeflag &= ~P_TREE_ORPHANED;
185 }
186 
187 /*
188  * exit -- death of process.
189  */
190 void
191 sys_sys_exit(struct thread *td, struct sys_exit_args *uap)
192 {
193 
194 	exit1(td, uap->rval, 0);
195 	/* NOTREACHED */
196 }
197 
198 /*
199  * Exit: deallocate address space and other resources, change proc state to
200  * zombie, and unlink proc from allproc and parent's lists.  Save exit status
201  * and rusage for wait().  Check for child processes and orphan them.
202  */
203 void
204 exit1(struct thread *td, int rval, int signo)
205 {
206 	struct proc *p, *nq, *q, *t;
207 	struct thread *tdt;
208 	ksiginfo_t *ksi, *ksi1;
209 	int signal_parent;
210 
211 	mtx_assert(&Giant, MA_NOTOWNED);
212 	KASSERT(rval == 0 || signo == 0, ("exit1 rv %d sig %d", rval, signo));
213 
214 	p = td->td_proc;
215 	/*
216 	 * XXX in case we're rebooting we just let init die in order to
217 	 * work around an unsolved stack overflow seen very late during
218 	 * shutdown on sparc64 when the gmirror worker process exists.
219 	 * XXX what to do now that sparc64 is gone... remove if?
220 	 */
221 	if (p == initproc && rebooting == 0) {
222 		printf("init died (signal %d, exit %d)\n", signo, rval);
223 		panic("Going nowhere without my init!");
224 	}
225 
226 	/*
227 	 * Deref SU mp, since the thread does not return to userspace.
228 	 */
229 	td_softdep_cleanup(td);
230 
231 	/*
232 	 * MUST abort all other threads before proceeding past here.
233 	 */
234 	PROC_LOCK(p);
235 	/*
236 	 * First check if some other thread or external request got
237 	 * here before us.  If so, act appropriately: exit or suspend.
238 	 * We must ensure that stop requests are handled before we set
239 	 * P_WEXIT.
240 	 */
241 	thread_suspend_check(0);
242 	while (p->p_flag & P_HADTHREADS) {
243 		/*
244 		 * Kill off the other threads. This requires
245 		 * some co-operation from other parts of the kernel
246 		 * so it may not be instantaneous.  With this state set
247 		 * any thread entering the kernel from userspace will
248 		 * thread_exit() in trap().  Any thread attempting to
249 		 * sleep will return immediately with EINTR or EWOULDBLOCK
250 		 * which will hopefully force them to back out to userland
251 		 * freeing resources as they go.  Any thread attempting
252 		 * to return to userland will thread_exit() from userret().
253 		 * thread_exit() will unsuspend us when the last of the
254 		 * other threads exits.
255 		 * If there is already a thread singler after resumption,
256 		 * calling thread_single will fail; in that case, we just
257 		 * re-check all suspension request, the thread should
258 		 * either be suspended there or exit.
259 		 */
260 		if (!thread_single(p, SINGLE_EXIT))
261 			/*
262 			 * All other activity in this process is now
263 			 * stopped.  Threading support has been turned
264 			 * off.
265 			 */
266 			break;
267 		/*
268 		 * Recheck for new stop or suspend requests which
269 		 * might appear while process lock was dropped in
270 		 * thread_single().
271 		 */
272 		thread_suspend_check(0);
273 	}
274 	KASSERT(p->p_numthreads == 1,
275 	    ("exit1: proc %p exiting with %d threads", p, p->p_numthreads));
276 	racct_sub(p, RACCT_NTHR, 1);
277 
278 	/* Let event handler change exit status */
279 	p->p_xexit = rval;
280 	p->p_xsig = signo;
281 
282 	/*
283 	 * Ignore any pending request to stop due to a stop signal.
284 	 * Once P_WEXIT is set, future requests will be ignored as
285 	 * well.
286 	 */
287 	p->p_flag &= ~P_STOPPED_SIG;
288 	KASSERT(!P_SHOULDSTOP(p), ("exiting process is stopped"));
289 
290 	/* Note that we are exiting. */
291 	p->p_flag |= P_WEXIT;
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_ptevents = 0;
345 
346 	/*
347 	 * Stop the real interval timer.  If the handler is currently
348 	 * executing, prevent it from rearming itself and let it finish.
349 	 */
350 	if (timevalisset(&p->p_realtimer.it_value) &&
351 	    _callout_stop_safe(&p->p_itcallout, CS_EXECUTING, NULL) == 0) {
352 		timevalclear(&p->p_realtimer.it_interval);
353 		msleep(&p->p_itcallout, &p->p_mtx, PWAIT, "ritwait", 0);
354 		KASSERT(!timevalisset(&p->p_realtimer.it_value),
355 		    ("realtime timer is still armed"));
356 	}
357 
358 	PROC_UNLOCK(p);
359 
360 	umtx_thread_exit(td);
361 
362 	/*
363 	 * Reset any sigio structures pointing to us as a result of
364 	 * F_SETOWN with our pid.
365 	 */
366 	funsetownlst(&p->p_sigiolst);
367 
368 	/*
369 	 * If this process has an nlminfo data area (for lockd), release it
370 	 */
371 	if (nlminfo_release_p != NULL && p->p_nlminfo != NULL)
372 		(*nlminfo_release_p)(p);
373 
374 	/*
375 	 * Close open files and release open-file table.
376 	 * This may block!
377 	 */
378 	fdescfree(td);
379 
380 	/*
381 	 * If this thread tickled GEOM, we need to wait for the giggling to
382 	 * stop before we return to userland
383 	 */
384 	if (td->td_pflags & TDP_GEOM)
385 		g_waitidle();
386 
387 	/*
388 	 * Remove ourself from our leader's peer list and wake our leader.
389 	 */
390 	if (p->p_leader->p_peers != NULL) {
391 		mtx_lock(&ppeers_lock);
392 		if (p->p_leader->p_peers != NULL) {
393 			q = p->p_leader;
394 			while (q->p_peers != p)
395 				q = q->p_peers;
396 			q->p_peers = p->p_peers;
397 			wakeup(p->p_leader);
398 		}
399 		mtx_unlock(&ppeers_lock);
400 	}
401 
402 	vmspace_exit(td);
403 	killjobc();
404 	(void)acct_process(td);
405 
406 #ifdef KTRACE
407 	ktrprocexit(td);
408 #endif
409 	/*
410 	 * Release reference to text vnode
411 	 */
412 	if (p->p_textvp != NULL) {
413 		vrele(p->p_textvp);
414 		p->p_textvp = NULL;
415 	}
416 
417 	/*
418 	 * Release our limits structure.
419 	 */
420 	lim_free(p->p_limit);
421 	p->p_limit = NULL;
422 
423 	tidhash_remove(td);
424 
425 	/*
426 	 * Call machine-dependent code to release any
427 	 * machine-dependent resources other than the address space.
428 	 * The address space is released by "vmspace_exitfree(p)" in
429 	 * vm_waitproc().
430 	 */
431 	cpu_exit(td);
432 
433 	WITNESS_WARN(WARN_PANIC, NULL, "process (pid %d) exiting", p->p_pid);
434 
435 	/*
436 	 * Remove from allproc. It still sits in the hash.
437 	 */
438 	sx_xlock(&allproc_lock);
439 	LIST_REMOVE(p, p_list);
440 	sx_xunlock(&allproc_lock);
441 
442 	sx_xlock(&proctree_lock);
443 	PROC_LOCK(p);
444 	p->p_flag &= ~(P_TRACED | P_PPWAIT | P_PPTRACE);
445 	PROC_UNLOCK(p);
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) == 0) {
462 			proc_reparent(q, q->p_reaper, true);
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 			} else if (q->p_pdeathsig > 0) {
485 				/*
486 				 * The child asked to received a signal
487 				 * when we exit.
488 				 */
489 				kern_psignal(q, q->p_pdeathsig);
490 			}
491 		} else {
492 			/*
493 			 * Traced processes are killed since their existence
494 			 * means someone is screwing up.
495 			 */
496 			t = proc_realparent(q);
497 			if (t == p) {
498 				proc_reparent(q, q->p_reaper, true);
499 			} else {
500 				PROC_LOCK(t);
501 				proc_reparent(q, t, true);
502 				PROC_UNLOCK(t);
503 			}
504 			/*
505 			 * Since q was found on our children list, the
506 			 * proc_reparent() call moved q to the orphan
507 			 * list due to present P_TRACED flag. Clear
508 			 * orphan link for q now while q is locked.
509 			 */
510 			proc_clear_orphan(q);
511 			q->p_flag &= ~(P_TRACED | P_STOPPED_TRACE);
512 			q->p_flag2 &= ~P2_PTRACE_FSTP;
513 			q->p_ptevents = 0;
514 			FOREACH_THREAD_IN_PROC(q, tdt) {
515 				tdt->td_dbgflags &= ~(TDB_SUSPEND | TDB_XSIG |
516 				    TDB_FSTP);
517 			}
518 			kern_psignal(q, SIGKILL);
519 		}
520 		PROC_UNLOCK(q);
521 		if (ksi != NULL)
522 			ksiginfo_free(ksi);
523 	}
524 
525 	/*
526 	 * Also get rid of our orphans.
527 	 */
528 	while ((q = LIST_FIRST(&p->p_orphans)) != NULL) {
529 		PROC_LOCK(q);
530 		KASSERT(q->p_oppid == p->p_pid,
531 		    ("orphan %p of %p has unexpected oppid %d", q, p,
532 		    q->p_oppid));
533 		q->p_oppid = q->p_reaper->p_pid;
534 
535 		/*
536 		 * If we are the real parent of this process
537 		 * but it has been reparented to a debugger, then
538 		 * check if it asked for a signal when we exit.
539 		 */
540 		if (q->p_pdeathsig > 0)
541 			kern_psignal(q, q->p_pdeathsig);
542 		CTR2(KTR_PTRACE, "exit: pid %d, clearing orphan %d", p->p_pid,
543 		    q->p_pid);
544 		proc_clear_orphan(q);
545 		PROC_UNLOCK(q);
546 	}
547 
548 #ifdef KDTRACE_HOOKS
549 	if (SDT_PROBES_ENABLED()) {
550 		int reason = CLD_EXITED;
551 		if (WCOREDUMP(signo))
552 			reason = CLD_DUMPED;
553 		else if (WIFSIGNALED(signo))
554 			reason = CLD_KILLED;
555 		SDT_PROBE1(proc, , , exit, reason);
556 	}
557 #endif
558 
559 	/* Save exit status. */
560 	PROC_LOCK(p);
561 	p->p_xthread = td;
562 
563 #ifdef KDTRACE_HOOKS
564 	/*
565 	 * Tell the DTrace fasttrap provider about the exit if it
566 	 * has declared an interest.
567 	 */
568 	if (dtrace_fasttrap_exit)
569 		dtrace_fasttrap_exit(p);
570 #endif
571 
572 	/*
573 	 * Notify interested parties of our demise.
574 	 */
575 	KNOTE_LOCKED(p->p_klist, NOTE_EXIT);
576 
577 	/*
578 	 * If this is a process with a descriptor, we may not need to deliver
579 	 * a signal to the parent.  proctree_lock is held over
580 	 * procdesc_exit() to serialize concurrent calls to close() and
581 	 * exit().
582 	 */
583 	signal_parent = 0;
584 	if (p->p_procdesc == NULL || procdesc_exit(p)) {
585 		/*
586 		 * Notify parent that we're gone.  If parent has the
587 		 * PS_NOCLDWAIT flag set, or if the handler is set to SIG_IGN,
588 		 * notify process 1 instead (and hope it will handle this
589 		 * situation).
590 		 */
591 		PROC_LOCK(p->p_pptr);
592 		mtx_lock(&p->p_pptr->p_sigacts->ps_mtx);
593 		if (p->p_pptr->p_sigacts->ps_flag &
594 		    (PS_NOCLDWAIT | PS_CLDSIGIGN)) {
595 			struct proc *pp;
596 
597 			mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx);
598 			pp = p->p_pptr;
599 			PROC_UNLOCK(pp);
600 			proc_reparent(p, p->p_reaper, true);
601 			p->p_sigparent = SIGCHLD;
602 			PROC_LOCK(p->p_pptr);
603 
604 			/*
605 			 * Notify parent, so in case he was wait(2)ing or
606 			 * executing waitpid(2) with our pid, he will
607 			 * continue.
608 			 */
609 			wakeup(pp);
610 		} else
611 			mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx);
612 
613 		if (p->p_pptr == p->p_reaper || p->p_pptr == initproc) {
614 			signal_parent = 1;
615 		} else if (p->p_sigparent != 0) {
616 			if (p->p_sigparent == SIGCHLD) {
617 				signal_parent = 1;
618 			} else { /* LINUX thread */
619 				signal_parent = 2;
620 			}
621 		}
622 	} else
623 		PROC_LOCK(p->p_pptr);
624 	sx_xunlock(&proctree_lock);
625 
626 	if (signal_parent == 1) {
627 		childproc_exited(p);
628 	} else if (signal_parent == 2) {
629 		kern_psignal(p->p_pptr, p->p_sigparent);
630 	}
631 
632 	/* Tell the prison that we are gone. */
633 	prison_proc_free(p->p_ucred->cr_prison);
634 
635 	/*
636 	 * The state PRS_ZOMBIE prevents other proesses from sending
637 	 * signal to the process, to avoid memory leak, we free memory
638 	 * for signal queue at the time when the state is set.
639 	 */
640 	sigqueue_flush(&p->p_sigqueue);
641 	sigqueue_flush(&td->td_sigqueue);
642 
643 	/*
644 	 * We have to wait until after acquiring all locks before
645 	 * changing p_state.  We need to avoid all possible context
646 	 * switches (including ones from blocking on a mutex) while
647 	 * marked as a zombie.  We also have to set the zombie state
648 	 * before we release the parent process' proc lock to avoid
649 	 * a lost wakeup.  So, we first call wakeup, then we grab the
650 	 * sched lock, update the state, and release the parent process'
651 	 * proc lock.
652 	 */
653 	wakeup(p->p_pptr);
654 	cv_broadcast(&p->p_pwait);
655 	sched_exit(p->p_pptr, td);
656 	PROC_SLOCK(p);
657 	p->p_state = PRS_ZOMBIE;
658 	PROC_UNLOCK(p->p_pptr);
659 
660 	/*
661 	 * Save our children's rusage information in our exit rusage.
662 	 */
663 	PROC_STATLOCK(p);
664 	ruadd(&p->p_ru, &p->p_rux, &p->p_stats->p_cru, &p->p_crux);
665 	PROC_STATUNLOCK(p);
666 
667 	/*
668 	 * Make sure the scheduler takes this thread out of its tables etc.
669 	 * This will also release this thread's reference to the ucred.
670 	 * Other thread parts to release include pcb bits and such.
671 	 */
672 	thread_exit();
673 }
674 
675 #ifndef _SYS_SYSPROTO_H_
676 struct abort2_args {
677 	char *why;
678 	int nargs;
679 	void **args;
680 };
681 #endif
682 
683 int
684 sys_abort2(struct thread *td, struct abort2_args *uap)
685 {
686 	struct proc *p = td->td_proc;
687 	struct sbuf *sb;
688 	void *uargs[16];
689 	int error, i, sig;
690 
691 	/*
692 	 * Do it right now so we can log either proper call of abort2(), or
693 	 * note, that invalid argument was passed. 512 is big enough to
694 	 * handle 16 arguments' descriptions with additional comments.
695 	 */
696 	sb = sbuf_new(NULL, NULL, 512, SBUF_FIXEDLEN);
697 	sbuf_clear(sb);
698 	sbuf_printf(sb, "%s(pid %d uid %d) aborted: ",
699 	    p->p_comm, p->p_pid, td->td_ucred->cr_uid);
700 	/*
701 	 * Since we can't return from abort2(), send SIGKILL in cases, where
702 	 * abort2() was called improperly
703 	 */
704 	sig = SIGKILL;
705 	/* Prevent from DoSes from user-space. */
706 	if (uap->nargs < 0 || uap->nargs > 16)
707 		goto out;
708 	if (uap->nargs > 0) {
709 		if (uap->args == NULL)
710 			goto out;
711 		error = copyin(uap->args, uargs, uap->nargs * sizeof(void *));
712 		if (error != 0)
713 			goto out;
714 	}
715 	/*
716 	 * Limit size of 'reason' string to 128. Will fit even when
717 	 * maximal number of arguments was chosen to be logged.
718 	 */
719 	if (uap->why != NULL) {
720 		error = sbuf_copyin(sb, uap->why, 128);
721 		if (error < 0)
722 			goto out;
723 	} else {
724 		sbuf_printf(sb, "(null)");
725 	}
726 	if (uap->nargs > 0) {
727 		sbuf_printf(sb, "(");
728 		for (i = 0;i < uap->nargs; i++)
729 			sbuf_printf(sb, "%s%p", i == 0 ? "" : ", ", uargs[i]);
730 		sbuf_printf(sb, ")");
731 	}
732 	/*
733 	 * Final stage: arguments were proper, string has been
734 	 * successfully copied from userspace, and copying pointers
735 	 * from user-space succeed.
736 	 */
737 	sig = SIGABRT;
738 out:
739 	if (sig == SIGKILL) {
740 		sbuf_trim(sb);
741 		sbuf_printf(sb, " (Reason text inaccessible)");
742 	}
743 	sbuf_cat(sb, "\n");
744 	sbuf_finish(sb);
745 	log(LOG_INFO, "%s", sbuf_data(sb));
746 	sbuf_delete(sb);
747 	exit1(td, 0, sig);
748 	return (0);
749 }
750 
751 #ifdef COMPAT_43
752 /*
753  * The dirty work is handled by kern_wait().
754  */
755 int
756 owait(struct thread *td, struct owait_args *uap __unused)
757 {
758 	int error, status;
759 
760 	error = kern_wait(td, WAIT_ANY, &status, 0, NULL);
761 	if (error == 0)
762 		td->td_retval[1] = status;
763 	return (error);
764 }
765 #endif /* COMPAT_43 */
766 
767 /*
768  * The dirty work is handled by kern_wait().
769  */
770 int
771 sys_wait4(struct thread *td, struct wait4_args *uap)
772 {
773 	struct rusage ru, *rup;
774 	int error, status;
775 
776 	if (uap->rusage != NULL)
777 		rup = &ru;
778 	else
779 		rup = NULL;
780 	error = kern_wait(td, uap->pid, &status, uap->options, rup);
781 	if (uap->status != NULL && error == 0 && td->td_retval[0] != 0)
782 		error = copyout(&status, uap->status, sizeof(status));
783 	if (uap->rusage != NULL && error == 0 && td->td_retval[0] != 0)
784 		error = copyout(&ru, uap->rusage, sizeof(struct rusage));
785 	return (error);
786 }
787 
788 int
789 sys_wait6(struct thread *td, struct wait6_args *uap)
790 {
791 	struct __wrusage wru, *wrup;
792 	siginfo_t si, *sip;
793 	idtype_t idtype;
794 	id_t id;
795 	int error, status;
796 
797 	idtype = uap->idtype;
798 	id = uap->id;
799 
800 	if (uap->wrusage != NULL)
801 		wrup = &wru;
802 	else
803 		wrup = NULL;
804 
805 	if (uap->info != NULL) {
806 		sip = &si;
807 		bzero(sip, sizeof(*sip));
808 	} else
809 		sip = NULL;
810 
811 	/*
812 	 *  We expect all callers of wait6() to know about WEXITED and
813 	 *  WTRAPPED.
814 	 */
815 	error = kern_wait6(td, idtype, id, &status, uap->options, wrup, sip);
816 
817 	if (uap->status != NULL && error == 0 && td->td_retval[0] != 0)
818 		error = copyout(&status, uap->status, sizeof(status));
819 	if (uap->wrusage != NULL && error == 0 && td->td_retval[0] != 0)
820 		error = copyout(&wru, uap->wrusage, sizeof(wru));
821 	if (uap->info != NULL && error == 0)
822 		error = copyout(&si, uap->info, sizeof(si));
823 	return (error);
824 }
825 
826 /*
827  * Reap the remains of a zombie process and optionally return status and
828  * rusage.  Asserts and will release both the proctree_lock and the process
829  * lock as part of its work.
830  */
831 void
832 proc_reap(struct thread *td, struct proc *p, int *status, int options)
833 {
834 	struct proc *q, *t;
835 
836 	sx_assert(&proctree_lock, SA_XLOCKED);
837 	PROC_LOCK_ASSERT(p, MA_OWNED);
838 	KASSERT(p->p_state == PRS_ZOMBIE, ("proc_reap: !PRS_ZOMBIE"));
839 
840 	mtx_spin_wait_unlocked(&p->p_slock);
841 
842 	q = td->td_proc;
843 
844 	if (status)
845 		*status = KW_EXITCODE(p->p_xexit, p->p_xsig);
846 	if (options & WNOWAIT) {
847 		/*
848 		 *  Only poll, returning the status.  Caller does not wish to
849 		 * release the proc struct just yet.
850 		 */
851 		PROC_UNLOCK(p);
852 		sx_xunlock(&proctree_lock);
853 		return;
854 	}
855 
856 	PROC_LOCK(q);
857 	sigqueue_take(p->p_ksi);
858 	PROC_UNLOCK(q);
859 
860 	/*
861 	 * If we got the child via a ptrace 'attach', we need to give it back
862 	 * to the old parent.
863 	 */
864 	if (p->p_oppid != p->p_pptr->p_pid) {
865 		PROC_UNLOCK(p);
866 		t = proc_realparent(p);
867 		PROC_LOCK(t);
868 		PROC_LOCK(p);
869 		CTR2(KTR_PTRACE,
870 		    "wait: traced child %d moved back to parent %d", p->p_pid,
871 		    t->p_pid);
872 		proc_reparent(p, t, false);
873 		PROC_UNLOCK(p);
874 		pksignal(t, SIGCHLD, p->p_ksi);
875 		wakeup(t);
876 		cv_broadcast(&p->p_pwait);
877 		PROC_UNLOCK(t);
878 		sx_xunlock(&proctree_lock);
879 		return;
880 	}
881 	PROC_UNLOCK(p);
882 
883 	/*
884 	 * Remove other references to this process to ensure we have an
885 	 * exclusive reference.
886 	 */
887 	sx_xlock(PIDHASHLOCK(p->p_pid));
888 	LIST_REMOVE(p, p_hash);
889 	sx_xunlock(PIDHASHLOCK(p->p_pid));
890 	LIST_REMOVE(p, p_sibling);
891 	reaper_abandon_children(p, true);
892 	reaper_clear(p);
893 	PROC_LOCK(p);
894 	proc_clear_orphan(p);
895 	PROC_UNLOCK(p);
896 	leavepgrp(p);
897 	if (p->p_procdesc != NULL)
898 		procdesc_reap(p);
899 	sx_xunlock(&proctree_lock);
900 
901 	proc_id_clear(PROC_ID_PID, p->p_pid);
902 
903 	PROC_LOCK(p);
904 	knlist_detach(p->p_klist);
905 	p->p_klist = NULL;
906 	PROC_UNLOCK(p);
907 
908 	/*
909 	 * Removal from allproc list and process group list paired with
910 	 * PROC_LOCK which was executed during that time should guarantee
911 	 * nothing can reach this process anymore. As such further locking
912 	 * is unnecessary.
913 	 */
914 	p->p_xexit = p->p_xsig = 0;		/* XXX: why? */
915 
916 	PROC_LOCK(q);
917 	ruadd(&q->p_stats->p_cru, &q->p_crux, &p->p_ru, &p->p_rux);
918 	PROC_UNLOCK(q);
919 
920 	/*
921 	 * Decrement the count of procs running with this uid.
922 	 */
923 	(void)chgproccnt(p->p_ucred->cr_ruidinfo, -1, 0);
924 
925 	/*
926 	 * Destroy resource accounting information associated with the process.
927 	 */
928 #ifdef RACCT
929 	if (racct_enable) {
930 		PROC_LOCK(p);
931 		racct_sub(p, RACCT_NPROC, 1);
932 		PROC_UNLOCK(p);
933 	}
934 #endif
935 	racct_proc_exit(p);
936 
937 	/*
938 	 * Free credentials, arguments, and sigacts.
939 	 */
940 	crfree(p->p_ucred);
941 	proc_set_cred(p, NULL);
942 	pargs_drop(p->p_args);
943 	p->p_args = NULL;
944 	sigacts_free(p->p_sigacts);
945 	p->p_sigacts = NULL;
946 
947 	/*
948 	 * Do any thread-system specific cleanups.
949 	 */
950 	thread_wait(p);
951 
952 	/*
953 	 * Give vm and machine-dependent layer a chance to free anything that
954 	 * cpu_exit couldn't release while still running in process context.
955 	 */
956 	vm_waitproc(p);
957 #ifdef MAC
958 	mac_proc_destroy(p);
959 #endif
960 
961 	KASSERT(FIRST_THREAD_IN_PROC(p),
962 	    ("proc_reap: no residual thread!"));
963 	uma_zfree(proc_zone, p);
964 	atomic_add_int(&nprocs, -1);
965 }
966 
967 static int
968 proc_to_reap(struct thread *td, struct proc *p, idtype_t idtype, id_t id,
969     int *status, int options, struct __wrusage *wrusage, siginfo_t *siginfo,
970     int check_only)
971 {
972 	struct rusage *rup;
973 
974 	sx_assert(&proctree_lock, SA_XLOCKED);
975 
976 	PROC_LOCK(p);
977 
978 	switch (idtype) {
979 	case P_ALL:
980 		if (p->p_procdesc == NULL ||
981 		   (p->p_pptr == td->td_proc &&
982 		   (p->p_flag & P_TRACED) != 0)) {
983 			break;
984 		}
985 
986 		PROC_UNLOCK(p);
987 		return (0);
988 	case P_PID:
989 		if (p->p_pid != (pid_t)id) {
990 			PROC_UNLOCK(p);
991 			return (0);
992 		}
993 		break;
994 	case P_PGID:
995 		if (p->p_pgid != (pid_t)id) {
996 			PROC_UNLOCK(p);
997 			return (0);
998 		}
999 		break;
1000 	case P_SID:
1001 		if (p->p_session->s_sid != (pid_t)id) {
1002 			PROC_UNLOCK(p);
1003 			return (0);
1004 		}
1005 		break;
1006 	case P_UID:
1007 		if (p->p_ucred->cr_uid != (uid_t)id) {
1008 			PROC_UNLOCK(p);
1009 			return (0);
1010 		}
1011 		break;
1012 	case P_GID:
1013 		if (p->p_ucred->cr_gid != (gid_t)id) {
1014 			PROC_UNLOCK(p);
1015 			return (0);
1016 		}
1017 		break;
1018 	case P_JAILID:
1019 		if (p->p_ucred->cr_prison->pr_id != (int)id) {
1020 			PROC_UNLOCK(p);
1021 			return (0);
1022 		}
1023 		break;
1024 	/*
1025 	 * It seems that the thread structures get zeroed out
1026 	 * at process exit.  This makes it impossible to
1027 	 * support P_SETID, P_CID or P_CPUID.
1028 	 */
1029 	default:
1030 		PROC_UNLOCK(p);
1031 		return (0);
1032 	}
1033 
1034 	if (p_canwait(td, p)) {
1035 		PROC_UNLOCK(p);
1036 		return (0);
1037 	}
1038 
1039 	if (((options & WEXITED) == 0) && (p->p_state == PRS_ZOMBIE)) {
1040 		PROC_UNLOCK(p);
1041 		return (0);
1042 	}
1043 
1044 	/*
1045 	 * This special case handles a kthread spawned by linux_clone
1046 	 * (see linux_misc.c).  The linux_wait4 and linux_waitpid
1047 	 * functions need to be able to distinguish between waiting
1048 	 * on a process and waiting on a thread.  It is a thread if
1049 	 * p_sigparent is not SIGCHLD, and the WLINUXCLONE option
1050 	 * signifies we want to wait for threads and not processes.
1051 	 */
1052 	if ((p->p_sigparent != SIGCHLD) ^
1053 	    ((options & WLINUXCLONE) != 0)) {
1054 		PROC_UNLOCK(p);
1055 		return (0);
1056 	}
1057 
1058 	if (siginfo != NULL) {
1059 		bzero(siginfo, sizeof(*siginfo));
1060 		siginfo->si_errno = 0;
1061 
1062 		/*
1063 		 * SUSv4 requires that the si_signo value is always
1064 		 * SIGCHLD. Obey it despite the rfork(2) interface
1065 		 * allows to request other signal for child exit
1066 		 * notification.
1067 		 */
1068 		siginfo->si_signo = SIGCHLD;
1069 
1070 		/*
1071 		 *  This is still a rough estimate.  We will fix the
1072 		 *  cases TRAPPED, STOPPED, and CONTINUED later.
1073 		 */
1074 		if (WCOREDUMP(p->p_xsig)) {
1075 			siginfo->si_code = CLD_DUMPED;
1076 			siginfo->si_status = WTERMSIG(p->p_xsig);
1077 		} else if (WIFSIGNALED(p->p_xsig)) {
1078 			siginfo->si_code = CLD_KILLED;
1079 			siginfo->si_status = WTERMSIG(p->p_xsig);
1080 		} else {
1081 			siginfo->si_code = CLD_EXITED;
1082 			siginfo->si_status = p->p_xexit;
1083 		}
1084 
1085 		siginfo->si_pid = p->p_pid;
1086 		siginfo->si_uid = p->p_ucred->cr_uid;
1087 
1088 		/*
1089 		 * The si_addr field would be useful additional
1090 		 * detail, but apparently the PC value may be lost
1091 		 * when we reach this point.  bzero() above sets
1092 		 * siginfo->si_addr to NULL.
1093 		 */
1094 	}
1095 
1096 	/*
1097 	 * There should be no reason to limit resources usage info to
1098 	 * exited processes only.  A snapshot about any resources used
1099 	 * by a stopped process may be exactly what is needed.
1100 	 */
1101 	if (wrusage != NULL) {
1102 		rup = &wrusage->wru_self;
1103 		*rup = p->p_ru;
1104 		PROC_STATLOCK(p);
1105 		calcru(p, &rup->ru_utime, &rup->ru_stime);
1106 		PROC_STATUNLOCK(p);
1107 
1108 		rup = &wrusage->wru_children;
1109 		*rup = p->p_stats->p_cru;
1110 		calccru(p, &rup->ru_utime, &rup->ru_stime);
1111 	}
1112 
1113 	if (p->p_state == PRS_ZOMBIE && !check_only) {
1114 		proc_reap(td, p, status, options);
1115 		return (-1);
1116 	}
1117 	return (1);
1118 }
1119 
1120 int
1121 kern_wait(struct thread *td, pid_t pid, int *status, int options,
1122     struct rusage *rusage)
1123 {
1124 	struct __wrusage wru, *wrup;
1125 	idtype_t idtype;
1126 	id_t id;
1127 	int ret;
1128 
1129 	/*
1130 	 * Translate the special pid values into the (idtype, pid)
1131 	 * pair for kern_wait6.  The WAIT_MYPGRP case is handled by
1132 	 * kern_wait6() on its own.
1133 	 */
1134 	if (pid == WAIT_ANY) {
1135 		idtype = P_ALL;
1136 		id = 0;
1137 	} else if (pid < 0) {
1138 		idtype = P_PGID;
1139 		id = (id_t)-pid;
1140 	} else {
1141 		idtype = P_PID;
1142 		id = (id_t)pid;
1143 	}
1144 
1145 	if (rusage != NULL)
1146 		wrup = &wru;
1147 	else
1148 		wrup = NULL;
1149 
1150 	/*
1151 	 * For backward compatibility we implicitly add flags WEXITED
1152 	 * and WTRAPPED here.
1153 	 */
1154 	options |= WEXITED | WTRAPPED;
1155 	ret = kern_wait6(td, idtype, id, status, options, wrup, NULL);
1156 	if (rusage != NULL)
1157 		*rusage = wru.wru_self;
1158 	return (ret);
1159 }
1160 
1161 static void
1162 report_alive_proc(struct thread *td, struct proc *p, siginfo_t *siginfo,
1163     int *status, int options, int si_code)
1164 {
1165 	bool cont;
1166 
1167 	PROC_LOCK_ASSERT(p, MA_OWNED);
1168 	sx_assert(&proctree_lock, SA_XLOCKED);
1169 	MPASS(si_code == CLD_TRAPPED || si_code == CLD_STOPPED ||
1170 	    si_code == CLD_CONTINUED);
1171 
1172 	cont = si_code == CLD_CONTINUED;
1173 	if ((options & WNOWAIT) == 0) {
1174 		if (cont)
1175 			p->p_flag &= ~P_CONTINUED;
1176 		else
1177 			p->p_flag |= P_WAITED;
1178 		PROC_LOCK(td->td_proc);
1179 		sigqueue_take(p->p_ksi);
1180 		PROC_UNLOCK(td->td_proc);
1181 	}
1182 	sx_xunlock(&proctree_lock);
1183 	if (siginfo != NULL) {
1184 		siginfo->si_code = si_code;
1185 		siginfo->si_status = cont ? SIGCONT : p->p_xsig;
1186 	}
1187 	if (status != NULL)
1188 		*status = cont ? SIGCONT : W_STOPCODE(p->p_xsig);
1189 	PROC_UNLOCK(p);
1190 	td->td_retval[0] = p->p_pid;
1191 }
1192 
1193 int
1194 kern_wait6(struct thread *td, idtype_t idtype, id_t id, int *status,
1195     int options, struct __wrusage *wrusage, siginfo_t *siginfo)
1196 {
1197 	struct proc *p, *q;
1198 	pid_t pid;
1199 	int error, nfound, ret;
1200 	bool report;
1201 
1202 	AUDIT_ARG_VALUE((int)idtype);	/* XXX - This is likely wrong! */
1203 	AUDIT_ARG_PID((pid_t)id);	/* XXX - This may be wrong! */
1204 	AUDIT_ARG_VALUE(options);
1205 
1206 	q = td->td_proc;
1207 
1208 	if ((pid_t)id == WAIT_MYPGRP && (idtype == P_PID || idtype == P_PGID)) {
1209 		PROC_LOCK(q);
1210 		id = (id_t)q->p_pgid;
1211 		PROC_UNLOCK(q);
1212 		idtype = P_PGID;
1213 	}
1214 
1215 	/* If we don't know the option, just return. */
1216 	if ((options & ~(WUNTRACED | WNOHANG | WCONTINUED | WNOWAIT |
1217 	    WEXITED | WTRAPPED | WLINUXCLONE)) != 0)
1218 		return (EINVAL);
1219 	if ((options & (WEXITED | WUNTRACED | WCONTINUED | WTRAPPED)) == 0) {
1220 		/*
1221 		 * We will be unable to find any matching processes,
1222 		 * because there are no known events to look for.
1223 		 * Prefer to return error instead of blocking
1224 		 * indefinitely.
1225 		 */
1226 		return (EINVAL);
1227 	}
1228 
1229 loop:
1230 	if (q->p_flag & P_STATCHILD) {
1231 		PROC_LOCK(q);
1232 		q->p_flag &= ~P_STATCHILD;
1233 		PROC_UNLOCK(q);
1234 	}
1235 	sx_xlock(&proctree_lock);
1236 loop_locked:
1237 	nfound = 0;
1238 	LIST_FOREACH(p, &q->p_children, p_sibling) {
1239 		pid = p->p_pid;
1240 		ret = proc_to_reap(td, p, idtype, id, status, options,
1241 		    wrusage, siginfo, 0);
1242 		if (ret == 0)
1243 			continue;
1244 		else if (ret != 1) {
1245 			td->td_retval[0] = pid;
1246 			return (0);
1247 		}
1248 
1249 		nfound++;
1250 		PROC_LOCK_ASSERT(p, MA_OWNED);
1251 
1252 		if ((options & WTRAPPED) != 0 &&
1253 		    (p->p_flag & P_TRACED) != 0) {
1254 			PROC_SLOCK(p);
1255 			report =
1256 			    ((p->p_flag & (P_STOPPED_TRACE | P_STOPPED_SIG)) &&
1257 			    p->p_suspcount == p->p_numthreads &&
1258 			    (p->p_flag & P_WAITED) == 0);
1259 			PROC_SUNLOCK(p);
1260 			if (report) {
1261 			CTR4(KTR_PTRACE,
1262 			    "wait: returning trapped pid %d status %#x "
1263 			    "(xstat %d) xthread %d",
1264 			    p->p_pid, W_STOPCODE(p->p_xsig), p->p_xsig,
1265 			    p->p_xthread != NULL ?
1266 			    p->p_xthread->td_tid : -1);
1267 				report_alive_proc(td, p, siginfo, status,
1268 				    options, CLD_TRAPPED);
1269 				return (0);
1270 			}
1271 		}
1272 		if ((options & WUNTRACED) != 0 &&
1273 		    (p->p_flag & P_STOPPED_SIG) != 0) {
1274 			PROC_SLOCK(p);
1275 			report = (p->p_suspcount == p->p_numthreads &&
1276 			    ((p->p_flag & P_WAITED) == 0));
1277 			PROC_SUNLOCK(p);
1278 			if (report) {
1279 				report_alive_proc(td, p, siginfo, status,
1280 				    options, CLD_STOPPED);
1281 				return (0);
1282 			}
1283 		}
1284 		if ((options & WCONTINUED) != 0 &&
1285 		    (p->p_flag & P_CONTINUED) != 0) {
1286 			report_alive_proc(td, p, siginfo, status, options,
1287 			    CLD_CONTINUED);
1288 			return (0);
1289 		}
1290 		PROC_UNLOCK(p);
1291 	}
1292 
1293 	/*
1294 	 * Look in the orphans list too, to allow the parent to
1295 	 * collect it's child exit status even if child is being
1296 	 * debugged.
1297 	 *
1298 	 * Debugger detaches from the parent upon successful
1299 	 * switch-over from parent to child.  At this point due to
1300 	 * re-parenting the parent loses the child to debugger and a
1301 	 * wait4(2) call would report that it has no children to wait
1302 	 * for.  By maintaining a list of orphans we allow the parent
1303 	 * to successfully wait until the child becomes a zombie.
1304 	 */
1305 	if (nfound == 0) {
1306 		LIST_FOREACH(p, &q->p_orphans, p_orphan) {
1307 			ret = proc_to_reap(td, p, idtype, id, NULL, options,
1308 			    NULL, NULL, 1);
1309 			if (ret != 0) {
1310 				KASSERT(ret != -1, ("reaped an orphan (pid %d)",
1311 				    (int)td->td_retval[0]));
1312 				PROC_UNLOCK(p);
1313 				nfound++;
1314 				break;
1315 			}
1316 		}
1317 	}
1318 	if (nfound == 0) {
1319 		sx_xunlock(&proctree_lock);
1320 		return (ECHILD);
1321 	}
1322 	if (options & WNOHANG) {
1323 		sx_xunlock(&proctree_lock);
1324 		td->td_retval[0] = 0;
1325 		return (0);
1326 	}
1327 	PROC_LOCK(q);
1328 	if (q->p_flag & P_STATCHILD) {
1329 		q->p_flag &= ~P_STATCHILD;
1330 		PROC_UNLOCK(q);
1331 		goto loop_locked;
1332 	}
1333 	sx_xunlock(&proctree_lock);
1334 	error = msleep(q, &q->p_mtx, PWAIT | PCATCH | PDROP, "wait", 0);
1335 	if (error)
1336 		return (error);
1337 	goto loop;
1338 }
1339 
1340 void
1341 proc_add_orphan(struct proc *child, struct proc *parent)
1342 {
1343 
1344 	sx_assert(&proctree_lock, SX_XLOCKED);
1345 	KASSERT((child->p_flag & P_TRACED) != 0,
1346 	    ("proc_add_orphan: not traced"));
1347 
1348 	if (LIST_EMPTY(&parent->p_orphans)) {
1349 		child->p_treeflag |= P_TREE_FIRST_ORPHAN;
1350 		LIST_INSERT_HEAD(&parent->p_orphans, child, p_orphan);
1351 	} else {
1352 		LIST_INSERT_AFTER(LIST_FIRST(&parent->p_orphans),
1353 		    child, p_orphan);
1354 	}
1355 	child->p_treeflag |= P_TREE_ORPHANED;
1356 }
1357 
1358 /*
1359  * Make process 'parent' the new parent of process 'child'.
1360  * Must be called with an exclusive hold of proctree lock.
1361  */
1362 void
1363 proc_reparent(struct proc *child, struct proc *parent, bool set_oppid)
1364 {
1365 
1366 	sx_assert(&proctree_lock, SX_XLOCKED);
1367 	PROC_LOCK_ASSERT(child, MA_OWNED);
1368 	if (child->p_pptr == parent)
1369 		return;
1370 
1371 	PROC_LOCK(child->p_pptr);
1372 	sigqueue_take(child->p_ksi);
1373 	PROC_UNLOCK(child->p_pptr);
1374 	LIST_REMOVE(child, p_sibling);
1375 	LIST_INSERT_HEAD(&parent->p_children, child, p_sibling);
1376 
1377 	proc_clear_orphan(child);
1378 	if ((child->p_flag & P_TRACED) != 0) {
1379 		proc_add_orphan(child, child->p_pptr);
1380 	}
1381 
1382 	child->p_pptr = parent;
1383 	if (set_oppid)
1384 		child->p_oppid = parent->p_pid;
1385 }
1386