xref: /freebsd/sys/kern/kern_exit.c (revision 7029da5c36f2d3cf6bb6c81bf551229f416399e8)
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/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 struct proc *
104 proc_realparent(struct proc *child)
105 {
106 	struct proc *p, *parent;
107 
108 	sx_assert(&proctree_lock, SX_LOCKED);
109 	if ((child->p_treeflag & P_TREE_ORPHANED) == 0)
110 		return (child->p_pptr->p_pid == child->p_oppid ?
111 			    child->p_pptr : initproc);
112 	for (p = child; (p->p_treeflag & P_TREE_FIRST_ORPHAN) == 0;) {
113 		/* Cannot use LIST_PREV(), since the list head is not known. */
114 		p = __containerof(p->p_orphan.le_prev, struct proc,
115 		    p_orphan.le_next);
116 		KASSERT((p->p_treeflag & P_TREE_ORPHANED) != 0,
117 		    ("missing P_ORPHAN %p", p));
118 	}
119 	parent = __containerof(p->p_orphan.le_prev, struct proc,
120 	    p_orphans.lh_first);
121 	return (parent);
122 }
123 
124 void
125 reaper_abandon_children(struct proc *p, bool exiting)
126 {
127 	struct proc *p1, *p2, *ptmp;
128 
129 	sx_assert(&proctree_lock, SX_LOCKED);
130 	KASSERT(p != initproc, ("reaper_abandon_children for initproc"));
131 	if ((p->p_treeflag & P_TREE_REAPER) == 0)
132 		return;
133 	p1 = p->p_reaper;
134 	LIST_FOREACH_SAFE(p2, &p->p_reaplist, p_reapsibling, ptmp) {
135 		LIST_REMOVE(p2, p_reapsibling);
136 		p2->p_reaper = p1;
137 		p2->p_reapsubtree = p->p_reapsubtree;
138 		LIST_INSERT_HEAD(&p1->p_reaplist, p2, p_reapsibling);
139 		if (exiting && p2->p_pptr == p) {
140 			PROC_LOCK(p2);
141 			proc_reparent(p2, p1, true);
142 			PROC_UNLOCK(p2);
143 		}
144 	}
145 	KASSERT(LIST_EMPTY(&p->p_reaplist), ("p_reaplist not empty"));
146 	p->p_treeflag &= ~P_TREE_REAPER;
147 }
148 
149 static void
150 reaper_clear(struct proc *p)
151 {
152 	struct proc *p1;
153 	bool clear;
154 
155 	sx_assert(&proctree_lock, SX_LOCKED);
156 	LIST_REMOVE(p, p_reapsibling);
157 	if (p->p_reapsubtree == 1)
158 		return;
159 	clear = true;
160 	LIST_FOREACH(p1, &p->p_reaper->p_reaplist, p_reapsibling) {
161 		if (p1->p_reapsubtree == p->p_reapsubtree) {
162 			clear = false;
163 			break;
164 		}
165 	}
166 	if (clear)
167 		proc_id_clear(PROC_ID_REAP, p->p_reapsubtree);
168 }
169 
170 void
171 proc_clear_orphan(struct proc *p)
172 {
173 	struct proc *p1;
174 
175 	sx_assert(&proctree_lock, SA_XLOCKED);
176 	if ((p->p_treeflag & P_TREE_ORPHANED) == 0)
177 		return;
178 	if ((p->p_treeflag & P_TREE_FIRST_ORPHAN) != 0) {
179 		p1 = LIST_NEXT(p, p_orphan);
180 		if (p1 != NULL)
181 			p1->p_treeflag |= P_TREE_FIRST_ORPHAN;
182 		p->p_treeflag &= ~P_TREE_FIRST_ORPHAN;
183 	}
184 	LIST_REMOVE(p, p_orphan);
185 	p->p_treeflag &= ~P_TREE_ORPHANED;
186 }
187 
188 /*
189  * exit -- death of process.
190  */
191 void
192 sys_sys_exit(struct thread *td, struct sys_exit_args *uap)
193 {
194 
195 	exit1(td, uap->rval, 0);
196 	/* NOTREACHED */
197 }
198 
199 /*
200  * Exit: deallocate address space and other resources, change proc state to
201  * zombie, and unlink proc from allproc and parent's lists.  Save exit status
202  * and rusage for wait().  Check for child processes and orphan them.
203  */
204 void
205 exit1(struct thread *td, int rval, int signo)
206 {
207 	struct proc *p, *nq, *q, *t;
208 	struct thread *tdt;
209 	ksiginfo_t *ksi, *ksi1;
210 	int signal_parent;
211 
212 	mtx_assert(&Giant, MA_NOTOWNED);
213 	KASSERT(rval == 0 || signo == 0, ("exit1 rv %d sig %d", rval, signo));
214 
215 	p = td->td_proc;
216 	/*
217 	 * XXX in case we're rebooting we just let init die in order to
218 	 * work around an unsolved stack overflow seen very late during
219 	 * shutdown on sparc64 when the gmirror worker process exists.
220 	 * XXX what to do now that sparc64 is gone... remove if?
221 	 */
222 	if (p == initproc && rebooting == 0) {
223 		printf("init died (signal %d, exit %d)\n", signo, rval);
224 		panic("Going nowhere without my init!");
225 	}
226 
227 	/*
228 	 * Deref SU mp, since the thread does not return to userspace.
229 	 */
230 	td_softdep_cleanup(td);
231 
232 	/*
233 	 * MUST abort all other threads before proceeding past here.
234 	 */
235 	PROC_LOCK(p);
236 	/*
237 	 * First check if some other thread or external request got
238 	 * here before us.  If so, act appropriately: exit or suspend.
239 	 * We must ensure that stop requests are handled before we set
240 	 * P_WEXIT.
241 	 */
242 	thread_suspend_check(0);
243 	while (p->p_flag & P_HADTHREADS) {
244 		/*
245 		 * Kill off the other threads. This requires
246 		 * some co-operation from other parts of the kernel
247 		 * so it may not be instantaneous.  With this state set
248 		 * any thread entering the kernel from userspace will
249 		 * thread_exit() in trap().  Any thread attempting to
250 		 * sleep will return immediately with EINTR or EWOULDBLOCK
251 		 * which will hopefully force them to back out to userland
252 		 * freeing resources as they go.  Any thread attempting
253 		 * to return to userland will thread_exit() from userret().
254 		 * thread_exit() will unsuspend us when the last of the
255 		 * other threads exits.
256 		 * If there is already a thread singler after resumption,
257 		 * calling thread_single will fail; in that case, we just
258 		 * re-check all suspension request, the thread should
259 		 * either be suspended there or exit.
260 		 */
261 		if (!thread_single(p, SINGLE_EXIT))
262 			/*
263 			 * All other activity in this process is now
264 			 * stopped.  Threading support has been turned
265 			 * off.
266 			 */
267 			break;
268 		/*
269 		 * Recheck for new stop or suspend requests which
270 		 * might appear while process lock was dropped in
271 		 * thread_single().
272 		 */
273 		thread_suspend_check(0);
274 	}
275 	KASSERT(p->p_numthreads == 1,
276 	    ("exit1: proc %p exiting with %d threads", p, p->p_numthreads));
277 	racct_sub(p, RACCT_NTHR, 1);
278 
279 	/* Let event handler change exit status */
280 	p->p_xexit = rval;
281 	p->p_xsig = signo;
282 
283 	/*
284 	 * Wakeup anyone in procfs' PIOCWAIT.  They should have a hold
285 	 * on our vmspace, so we should block below until they have
286 	 * released their reference to us.  Note that if they have
287 	 * requested S_EXIT stops we will block here until they ack
288 	 * via PIOCCONT.
289 	 */
290 	_STOPEVENT(p, S_EXIT, 0);
291 
292 	/*
293 	 * Ignore any pending request to stop due to a stop signal.
294 	 * Once P_WEXIT is set, future requests will be ignored as
295 	 * well.
296 	 */
297 	p->p_flag &= ~P_STOPPED_SIG;
298 	KASSERT(!P_SHOULDSTOP(p), ("exiting process is stopped"));
299 
300 	/*
301 	 * Note that we are exiting and do another wakeup of anyone in
302 	 * PIOCWAIT in case they aren't listening for S_EXIT stops or
303 	 * decided to wait again after we told them we are exiting.
304 	 */
305 	p->p_flag |= P_WEXIT;
306 	wakeup(&p->p_stype);
307 
308 	/*
309 	 * Wait for any processes that have a hold on our vmspace to
310 	 * release their reference.
311 	 */
312 	while (p->p_lock > 0)
313 		msleep(&p->p_lock, &p->p_mtx, PWAIT, "exithold", 0);
314 
315 	PROC_UNLOCK(p);
316 	/* Drain the limit callout while we don't have the proc locked */
317 	callout_drain(&p->p_limco);
318 
319 #ifdef AUDIT
320 	/*
321 	 * The Sun BSM exit token contains two components: an exit status as
322 	 * passed to exit(), and a return value to indicate what sort of exit
323 	 * it was.  The exit status is WEXITSTATUS(rv), but it's not clear
324 	 * what the return value is.
325 	 */
326 	AUDIT_ARG_EXIT(rval, 0);
327 	AUDIT_SYSCALL_EXIT(0, td);
328 #endif
329 
330 	/* Are we a task leader with peers? */
331 	if (p->p_peers != NULL && p == p->p_leader) {
332 		mtx_lock(&ppeers_lock);
333 		q = p->p_peers;
334 		while (q != NULL) {
335 			PROC_LOCK(q);
336 			kern_psignal(q, SIGKILL);
337 			PROC_UNLOCK(q);
338 			q = q->p_peers;
339 		}
340 		while (p->p_peers != NULL)
341 			msleep(p, &ppeers_lock, PWAIT, "exit1", 0);
342 		mtx_unlock(&ppeers_lock);
343 	}
344 
345 	/*
346 	 * Check if any loadable modules need anything done at process exit.
347 	 * E.g. SYSV IPC stuff.
348 	 * Event handler could change exit status.
349 	 * XXX what if one of these generates an error?
350 	 */
351 	EVENTHANDLER_DIRECT_INVOKE(process_exit, p);
352 
353 	/*
354 	 * If parent is waiting for us to exit or exec,
355 	 * P_PPWAIT is set; we will wakeup the parent below.
356 	 */
357 	PROC_LOCK(p);
358 	stopprofclock(p);
359 	p->p_ptevents = 0;
360 
361 	/*
362 	 * Stop the real interval timer.  If the handler is currently
363 	 * executing, prevent it from rearming itself and let it finish.
364 	 */
365 	if (timevalisset(&p->p_realtimer.it_value) &&
366 	    _callout_stop_safe(&p->p_itcallout, CS_EXECUTING, NULL) == 0) {
367 		timevalclear(&p->p_realtimer.it_interval);
368 		msleep(&p->p_itcallout, &p->p_mtx, PWAIT, "ritwait", 0);
369 		KASSERT(!timevalisset(&p->p_realtimer.it_value),
370 		    ("realtime timer is still armed"));
371 	}
372 
373 	PROC_UNLOCK(p);
374 
375 	umtx_thread_exit(td);
376 
377 	/*
378 	 * Reset any sigio structures pointing to us as a result of
379 	 * F_SETOWN with our pid.
380 	 */
381 	funsetownlst(&p->p_sigiolst);
382 
383 	/*
384 	 * If this process has an nlminfo data area (for lockd), release it
385 	 */
386 	if (nlminfo_release_p != NULL && p->p_nlminfo != NULL)
387 		(*nlminfo_release_p)(p);
388 
389 	/*
390 	 * Close open files and release open-file table.
391 	 * This may block!
392 	 */
393 	fdescfree(td);
394 
395 	/*
396 	 * If this thread tickled GEOM, we need to wait for the giggling to
397 	 * stop before we return to userland
398 	 */
399 	if (td->td_pflags & TDP_GEOM)
400 		g_waitidle();
401 
402 	/*
403 	 * Remove ourself from our leader's peer list and wake our leader.
404 	 */
405 	if (p->p_leader->p_peers != NULL) {
406 		mtx_lock(&ppeers_lock);
407 		if (p->p_leader->p_peers != NULL) {
408 			q = p->p_leader;
409 			while (q->p_peers != p)
410 				q = q->p_peers;
411 			q->p_peers = p->p_peers;
412 			wakeup(p->p_leader);
413 		}
414 		mtx_unlock(&ppeers_lock);
415 	}
416 
417 	vmspace_exit(td);
418 	killjobc();
419 	(void)acct_process(td);
420 
421 #ifdef KTRACE
422 	ktrprocexit(td);
423 #endif
424 	/*
425 	 * Release reference to text vnode
426 	 */
427 	if (p->p_textvp != NULL) {
428 		vrele(p->p_textvp);
429 		p->p_textvp = NULL;
430 	}
431 
432 	/*
433 	 * Release our limits structure.
434 	 */
435 	lim_free(p->p_limit);
436 	p->p_limit = NULL;
437 
438 	tidhash_remove(td);
439 
440 	/*
441 	 * Call machine-dependent code to release any
442 	 * machine-dependent resources other than the address space.
443 	 * The address space is released by "vmspace_exitfree(p)" in
444 	 * vm_waitproc().
445 	 */
446 	cpu_exit(td);
447 
448 	WITNESS_WARN(WARN_PANIC, NULL, "process (pid %d) exiting", p->p_pid);
449 
450 	/*
451 	 * Remove from allproc. It still sits in the hash.
452 	 */
453 	sx_xlock(&allproc_lock);
454 	LIST_REMOVE(p, p_list);
455 	sx_xunlock(&allproc_lock);
456 
457 	sx_xlock(&proctree_lock);
458 	PROC_LOCK(p);
459 	p->p_flag &= ~(P_TRACED | P_PPWAIT | P_PPTRACE);
460 	PROC_UNLOCK(p);
461 
462 	/*
463 	 * Reparent all children processes:
464 	 * - traced ones to the original parent (or init if we are that parent)
465 	 * - the rest to init
466 	 */
467 	q = LIST_FIRST(&p->p_children);
468 	if (q != NULL)		/* only need this if any child is S_ZOMB */
469 		wakeup(q->p_reaper);
470 	for (; q != NULL; q = nq) {
471 		nq = LIST_NEXT(q, p_sibling);
472 		ksi = ksiginfo_alloc(TRUE);
473 		PROC_LOCK(q);
474 		q->p_sigparent = SIGCHLD;
475 
476 		if ((q->p_flag & P_TRACED) == 0) {
477 			proc_reparent(q, q->p_reaper, true);
478 			if (q->p_state == PRS_ZOMBIE) {
479 				/*
480 				 * Inform reaper about the reparented
481 				 * zombie, since wait(2) has something
482 				 * new to report.  Guarantee queueing
483 				 * of the SIGCHLD signal, similar to
484 				 * the _exit() behaviour, by providing
485 				 * our ksiginfo.  Ksi is freed by the
486 				 * signal delivery.
487 				 */
488 				if (q->p_ksi == NULL) {
489 					ksi1 = NULL;
490 				} else {
491 					ksiginfo_copy(q->p_ksi, ksi);
492 					ksi->ksi_flags |= KSI_INS;
493 					ksi1 = ksi;
494 					ksi = NULL;
495 				}
496 				PROC_LOCK(q->p_reaper);
497 				pksignal(q->p_reaper, SIGCHLD, ksi1);
498 				PROC_UNLOCK(q->p_reaper);
499 			} else if (q->p_pdeathsig > 0) {
500 				/*
501 				 * The child asked to received a signal
502 				 * when we exit.
503 				 */
504 				kern_psignal(q, q->p_pdeathsig);
505 			}
506 		} else {
507 			/*
508 			 * Traced processes are killed since their existence
509 			 * means someone is screwing up.
510 			 */
511 			t = proc_realparent(q);
512 			if (t == p) {
513 				proc_reparent(q, q->p_reaper, true);
514 			} else {
515 				PROC_LOCK(t);
516 				proc_reparent(q, t, true);
517 				PROC_UNLOCK(t);
518 			}
519 			/*
520 			 * Since q was found on our children list, the
521 			 * proc_reparent() call moved q to the orphan
522 			 * list due to present P_TRACED flag. Clear
523 			 * orphan link for q now while q is locked.
524 			 */
525 			proc_clear_orphan(q);
526 			q->p_flag &= ~(P_TRACED | P_STOPPED_TRACE);
527 			q->p_flag2 &= ~P2_PTRACE_FSTP;
528 			q->p_ptevents = 0;
529 			FOREACH_THREAD_IN_PROC(q, tdt) {
530 				tdt->td_dbgflags &= ~(TDB_SUSPEND | TDB_XSIG |
531 				    TDB_FSTP);
532 			}
533 			kern_psignal(q, SIGKILL);
534 		}
535 		PROC_UNLOCK(q);
536 		if (ksi != NULL)
537 			ksiginfo_free(ksi);
538 	}
539 
540 	/*
541 	 * Also get rid of our orphans.
542 	 */
543 	while ((q = LIST_FIRST(&p->p_orphans)) != NULL) {
544 		PROC_LOCK(q);
545 		KASSERT(q->p_oppid == p->p_pid,
546 		    ("orphan %p of %p has unexpected oppid %d", q, p,
547 		    q->p_oppid));
548 		q->p_oppid = q->p_reaper->p_pid;
549 
550 		/*
551 		 * If we are the real parent of this process
552 		 * but it has been reparented to a debugger, then
553 		 * check if it asked for a signal when we exit.
554 		 */
555 		if (q->p_pdeathsig > 0)
556 			kern_psignal(q, q->p_pdeathsig);
557 		CTR2(KTR_PTRACE, "exit: pid %d, clearing orphan %d", p->p_pid,
558 		    q->p_pid);
559 		proc_clear_orphan(q);
560 		PROC_UNLOCK(q);
561 	}
562 
563 #ifdef KDTRACE_HOOKS
564 	if (SDT_PROBES_ENABLED()) {
565 		int reason = CLD_EXITED;
566 		if (WCOREDUMP(signo))
567 			reason = CLD_DUMPED;
568 		else if (WIFSIGNALED(signo))
569 			reason = CLD_KILLED;
570 		SDT_PROBE1(proc, , , exit, reason);
571 	}
572 #endif
573 
574 	/* Save exit status. */
575 	PROC_LOCK(p);
576 	p->p_xthread = td;
577 
578 #ifdef KDTRACE_HOOKS
579 	/*
580 	 * Tell the DTrace fasttrap provider about the exit if it
581 	 * has declared an interest.
582 	 */
583 	if (dtrace_fasttrap_exit)
584 		dtrace_fasttrap_exit(p);
585 #endif
586 
587 	/*
588 	 * Notify interested parties of our demise.
589 	 */
590 	KNOTE_LOCKED(p->p_klist, NOTE_EXIT);
591 
592 	/*
593 	 * If this is a process with a descriptor, we may not need to deliver
594 	 * a signal to the parent.  proctree_lock is held over
595 	 * procdesc_exit() to serialize concurrent calls to close() and
596 	 * exit().
597 	 */
598 	signal_parent = 0;
599 	if (p->p_procdesc == NULL || procdesc_exit(p)) {
600 		/*
601 		 * Notify parent that we're gone.  If parent has the
602 		 * PS_NOCLDWAIT flag set, or if the handler is set to SIG_IGN,
603 		 * notify process 1 instead (and hope it will handle this
604 		 * situation).
605 		 */
606 		PROC_LOCK(p->p_pptr);
607 		mtx_lock(&p->p_pptr->p_sigacts->ps_mtx);
608 		if (p->p_pptr->p_sigacts->ps_flag &
609 		    (PS_NOCLDWAIT | PS_CLDSIGIGN)) {
610 			struct proc *pp;
611 
612 			mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx);
613 			pp = p->p_pptr;
614 			PROC_UNLOCK(pp);
615 			proc_reparent(p, p->p_reaper, true);
616 			p->p_sigparent = SIGCHLD;
617 			PROC_LOCK(p->p_pptr);
618 
619 			/*
620 			 * Notify parent, so in case he was wait(2)ing or
621 			 * executing waitpid(2) with our pid, he will
622 			 * continue.
623 			 */
624 			wakeup(pp);
625 		} else
626 			mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx);
627 
628 		if (p->p_pptr == p->p_reaper || p->p_pptr == initproc) {
629 			signal_parent = 1;
630 		} else if (p->p_sigparent != 0) {
631 			if (p->p_sigparent == SIGCHLD) {
632 				signal_parent = 1;
633 			} else { /* LINUX thread */
634 				signal_parent = 2;
635 			}
636 		}
637 	} else
638 		PROC_LOCK(p->p_pptr);
639 	sx_xunlock(&proctree_lock);
640 
641 	if (signal_parent == 1) {
642 		childproc_exited(p);
643 	} else if (signal_parent == 2) {
644 		kern_psignal(p->p_pptr, p->p_sigparent);
645 	}
646 
647 	/* Tell the prison that we are gone. */
648 	prison_proc_free(p->p_ucred->cr_prison);
649 
650 	/*
651 	 * The state PRS_ZOMBIE prevents other proesses from sending
652 	 * signal to the process, to avoid memory leak, we free memory
653 	 * for signal queue at the time when the state is set.
654 	 */
655 	sigqueue_flush(&p->p_sigqueue);
656 	sigqueue_flush(&td->td_sigqueue);
657 
658 	/*
659 	 * We have to wait until after acquiring all locks before
660 	 * changing p_state.  We need to avoid all possible context
661 	 * switches (including ones from blocking on a mutex) while
662 	 * marked as a zombie.  We also have to set the zombie state
663 	 * before we release the parent process' proc lock to avoid
664 	 * a lost wakeup.  So, we first call wakeup, then we grab the
665 	 * sched lock, update the state, and release the parent process'
666 	 * proc lock.
667 	 */
668 	wakeup(p->p_pptr);
669 	cv_broadcast(&p->p_pwait);
670 	sched_exit(p->p_pptr, td);
671 	PROC_SLOCK(p);
672 	p->p_state = PRS_ZOMBIE;
673 	PROC_UNLOCK(p->p_pptr);
674 
675 	/*
676 	 * Save our children's rusage information in our exit rusage.
677 	 */
678 	PROC_STATLOCK(p);
679 	ruadd(&p->p_ru, &p->p_rux, &p->p_stats->p_cru, &p->p_crux);
680 	PROC_STATUNLOCK(p);
681 
682 	/*
683 	 * Make sure the scheduler takes this thread out of its tables etc.
684 	 * This will also release this thread's reference to the ucred.
685 	 * Other thread parts to release include pcb bits and such.
686 	 */
687 	thread_exit();
688 }
689 
690 #ifndef _SYS_SYSPROTO_H_
691 struct abort2_args {
692 	char *why;
693 	int nargs;
694 	void **args;
695 };
696 #endif
697 
698 int
699 sys_abort2(struct thread *td, struct abort2_args *uap)
700 {
701 	struct proc *p = td->td_proc;
702 	struct sbuf *sb;
703 	void *uargs[16];
704 	int error, i, sig;
705 
706 	/*
707 	 * Do it right now so we can log either proper call of abort2(), or
708 	 * note, that invalid argument was passed. 512 is big enough to
709 	 * handle 16 arguments' descriptions with additional comments.
710 	 */
711 	sb = sbuf_new(NULL, NULL, 512, SBUF_FIXEDLEN);
712 	sbuf_clear(sb);
713 	sbuf_printf(sb, "%s(pid %d uid %d) aborted: ",
714 	    p->p_comm, p->p_pid, td->td_ucred->cr_uid);
715 	/*
716 	 * Since we can't return from abort2(), send SIGKILL in cases, where
717 	 * abort2() was called improperly
718 	 */
719 	sig = SIGKILL;
720 	/* Prevent from DoSes from user-space. */
721 	if (uap->nargs < 0 || uap->nargs > 16)
722 		goto out;
723 	if (uap->nargs > 0) {
724 		if (uap->args == NULL)
725 			goto out;
726 		error = copyin(uap->args, uargs, uap->nargs * sizeof(void *));
727 		if (error != 0)
728 			goto out;
729 	}
730 	/*
731 	 * Limit size of 'reason' string to 128. Will fit even when
732 	 * maximal number of arguments was chosen to be logged.
733 	 */
734 	if (uap->why != NULL) {
735 		error = sbuf_copyin(sb, uap->why, 128);
736 		if (error < 0)
737 			goto out;
738 	} else {
739 		sbuf_printf(sb, "(null)");
740 	}
741 	if (uap->nargs > 0) {
742 		sbuf_printf(sb, "(");
743 		for (i = 0;i < uap->nargs; i++)
744 			sbuf_printf(sb, "%s%p", i == 0 ? "" : ", ", uargs[i]);
745 		sbuf_printf(sb, ")");
746 	}
747 	/*
748 	 * Final stage: arguments were proper, string has been
749 	 * successfully copied from userspace, and copying pointers
750 	 * from user-space succeed.
751 	 */
752 	sig = SIGABRT;
753 out:
754 	if (sig == SIGKILL) {
755 		sbuf_trim(sb);
756 		sbuf_printf(sb, " (Reason text inaccessible)");
757 	}
758 	sbuf_cat(sb, "\n");
759 	sbuf_finish(sb);
760 	log(LOG_INFO, "%s", sbuf_data(sb));
761 	sbuf_delete(sb);
762 	exit1(td, 0, sig);
763 	return (0);
764 }
765 
766 #ifdef COMPAT_43
767 /*
768  * The dirty work is handled by kern_wait().
769  */
770 int
771 owait(struct thread *td, struct owait_args *uap __unused)
772 {
773 	int error, status;
774 
775 	error = kern_wait(td, WAIT_ANY, &status, 0, NULL);
776 	if (error == 0)
777 		td->td_retval[1] = status;
778 	return (error);
779 }
780 #endif /* COMPAT_43 */
781 
782 /*
783  * The dirty work is handled by kern_wait().
784  */
785 int
786 sys_wait4(struct thread *td, struct wait4_args *uap)
787 {
788 	struct rusage ru, *rup;
789 	int error, status;
790 
791 	if (uap->rusage != NULL)
792 		rup = &ru;
793 	else
794 		rup = NULL;
795 	error = kern_wait(td, uap->pid, &status, uap->options, rup);
796 	if (uap->status != NULL && error == 0 && td->td_retval[0] != 0)
797 		error = copyout(&status, uap->status, sizeof(status));
798 	if (uap->rusage != NULL && error == 0 && td->td_retval[0] != 0)
799 		error = copyout(&ru, uap->rusage, sizeof(struct rusage));
800 	return (error);
801 }
802 
803 int
804 sys_wait6(struct thread *td, struct wait6_args *uap)
805 {
806 	struct __wrusage wru, *wrup;
807 	siginfo_t si, *sip;
808 	idtype_t idtype;
809 	id_t id;
810 	int error, status;
811 
812 	idtype = uap->idtype;
813 	id = uap->id;
814 
815 	if (uap->wrusage != NULL)
816 		wrup = &wru;
817 	else
818 		wrup = NULL;
819 
820 	if (uap->info != NULL) {
821 		sip = &si;
822 		bzero(sip, sizeof(*sip));
823 	} else
824 		sip = NULL;
825 
826 	/*
827 	 *  We expect all callers of wait6() to know about WEXITED and
828 	 *  WTRAPPED.
829 	 */
830 	error = kern_wait6(td, idtype, id, &status, uap->options, wrup, sip);
831 
832 	if (uap->status != NULL && error == 0 && td->td_retval[0] != 0)
833 		error = copyout(&status, uap->status, sizeof(status));
834 	if (uap->wrusage != NULL && error == 0 && td->td_retval[0] != 0)
835 		error = copyout(&wru, uap->wrusage, sizeof(wru));
836 	if (uap->info != NULL && error == 0)
837 		error = copyout(&si, uap->info, sizeof(si));
838 	return (error);
839 }
840 
841 /*
842  * Reap the remains of a zombie process and optionally return status and
843  * rusage.  Asserts and will release both the proctree_lock and the process
844  * lock as part of its work.
845  */
846 void
847 proc_reap(struct thread *td, struct proc *p, int *status, int options)
848 {
849 	struct proc *q, *t;
850 
851 	sx_assert(&proctree_lock, SA_XLOCKED);
852 	PROC_LOCK_ASSERT(p, MA_OWNED);
853 	KASSERT(p->p_state == PRS_ZOMBIE, ("proc_reap: !PRS_ZOMBIE"));
854 
855 	mtx_spin_wait_unlocked(&p->p_slock);
856 
857 	q = td->td_proc;
858 
859 	if (status)
860 		*status = KW_EXITCODE(p->p_xexit, p->p_xsig);
861 	if (options & WNOWAIT) {
862 		/*
863 		 *  Only poll, returning the status.  Caller does not wish to
864 		 * release the proc struct just yet.
865 		 */
866 		PROC_UNLOCK(p);
867 		sx_xunlock(&proctree_lock);
868 		return;
869 	}
870 
871 	PROC_LOCK(q);
872 	sigqueue_take(p->p_ksi);
873 	PROC_UNLOCK(q);
874 
875 	/*
876 	 * If we got the child via a ptrace 'attach', we need to give it back
877 	 * to the old parent.
878 	 */
879 	if (p->p_oppid != p->p_pptr->p_pid) {
880 		PROC_UNLOCK(p);
881 		t = proc_realparent(p);
882 		PROC_LOCK(t);
883 		PROC_LOCK(p);
884 		CTR2(KTR_PTRACE,
885 		    "wait: traced child %d moved back to parent %d", p->p_pid,
886 		    t->p_pid);
887 		proc_reparent(p, t, false);
888 		PROC_UNLOCK(p);
889 		pksignal(t, SIGCHLD, p->p_ksi);
890 		wakeup(t);
891 		cv_broadcast(&p->p_pwait);
892 		PROC_UNLOCK(t);
893 		sx_xunlock(&proctree_lock);
894 		return;
895 	}
896 	PROC_UNLOCK(p);
897 
898 	/*
899 	 * Remove other references to this process to ensure we have an
900 	 * exclusive reference.
901 	 */
902 	sx_xlock(PIDHASHLOCK(p->p_pid));
903 	LIST_REMOVE(p, p_hash);
904 	sx_xunlock(PIDHASHLOCK(p->p_pid));
905 	LIST_REMOVE(p, p_sibling);
906 	reaper_abandon_children(p, true);
907 	reaper_clear(p);
908 	PROC_LOCK(p);
909 	proc_clear_orphan(p);
910 	PROC_UNLOCK(p);
911 	leavepgrp(p);
912 	if (p->p_procdesc != NULL)
913 		procdesc_reap(p);
914 	sx_xunlock(&proctree_lock);
915 
916 	proc_id_clear(PROC_ID_PID, p->p_pid);
917 
918 	PROC_LOCK(p);
919 	knlist_detach(p->p_klist);
920 	p->p_klist = NULL;
921 	PROC_UNLOCK(p);
922 
923 	/*
924 	 * Removal from allproc list and process group list paired with
925 	 * PROC_LOCK which was executed during that time should guarantee
926 	 * nothing can reach this process anymore. As such further locking
927 	 * is unnecessary.
928 	 */
929 	p->p_xexit = p->p_xsig = 0;		/* XXX: why? */
930 
931 	PROC_LOCK(q);
932 	ruadd(&q->p_stats->p_cru, &q->p_crux, &p->p_ru, &p->p_rux);
933 	PROC_UNLOCK(q);
934 
935 	/*
936 	 * Decrement the count of procs running with this uid.
937 	 */
938 	(void)chgproccnt(p->p_ucred->cr_ruidinfo, -1, 0);
939 
940 	/*
941 	 * Destroy resource accounting information associated with the process.
942 	 */
943 #ifdef RACCT
944 	if (racct_enable) {
945 		PROC_LOCK(p);
946 		racct_sub(p, RACCT_NPROC, 1);
947 		PROC_UNLOCK(p);
948 	}
949 #endif
950 	racct_proc_exit(p);
951 
952 	/*
953 	 * Free credentials, arguments, and sigacts.
954 	 */
955 	crfree(p->p_ucred);
956 	proc_set_cred(p, NULL);
957 	pargs_drop(p->p_args);
958 	p->p_args = NULL;
959 	sigacts_free(p->p_sigacts);
960 	p->p_sigacts = NULL;
961 
962 	/*
963 	 * Do any thread-system specific cleanups.
964 	 */
965 	thread_wait(p);
966 
967 	/*
968 	 * Give vm and machine-dependent layer a chance to free anything that
969 	 * cpu_exit couldn't release while still running in process context.
970 	 */
971 	vm_waitproc(p);
972 #ifdef MAC
973 	mac_proc_destroy(p);
974 #endif
975 
976 	KASSERT(FIRST_THREAD_IN_PROC(p),
977 	    ("proc_reap: no residual thread!"));
978 	uma_zfree(proc_zone, p);
979 	atomic_add_int(&nprocs, -1);
980 }
981 
982 static int
983 proc_to_reap(struct thread *td, struct proc *p, idtype_t idtype, id_t id,
984     int *status, int options, struct __wrusage *wrusage, siginfo_t *siginfo,
985     int check_only)
986 {
987 	struct rusage *rup;
988 
989 	sx_assert(&proctree_lock, SA_XLOCKED);
990 
991 	PROC_LOCK(p);
992 
993 	switch (idtype) {
994 	case P_ALL:
995 		if (p->p_procdesc == NULL ||
996 		   (p->p_pptr == td->td_proc &&
997 		   (p->p_flag & P_TRACED) != 0)) {
998 			break;
999 		}
1000 
1001 		PROC_UNLOCK(p);
1002 		return (0);
1003 	case P_PID:
1004 		if (p->p_pid != (pid_t)id) {
1005 			PROC_UNLOCK(p);
1006 			return (0);
1007 		}
1008 		break;
1009 	case P_PGID:
1010 		if (p->p_pgid != (pid_t)id) {
1011 			PROC_UNLOCK(p);
1012 			return (0);
1013 		}
1014 		break;
1015 	case P_SID:
1016 		if (p->p_session->s_sid != (pid_t)id) {
1017 			PROC_UNLOCK(p);
1018 			return (0);
1019 		}
1020 		break;
1021 	case P_UID:
1022 		if (p->p_ucred->cr_uid != (uid_t)id) {
1023 			PROC_UNLOCK(p);
1024 			return (0);
1025 		}
1026 		break;
1027 	case P_GID:
1028 		if (p->p_ucred->cr_gid != (gid_t)id) {
1029 			PROC_UNLOCK(p);
1030 			return (0);
1031 		}
1032 		break;
1033 	case P_JAILID:
1034 		if (p->p_ucred->cr_prison->pr_id != (int)id) {
1035 			PROC_UNLOCK(p);
1036 			return (0);
1037 		}
1038 		break;
1039 	/*
1040 	 * It seems that the thread structures get zeroed out
1041 	 * at process exit.  This makes it impossible to
1042 	 * support P_SETID, P_CID or P_CPUID.
1043 	 */
1044 	default:
1045 		PROC_UNLOCK(p);
1046 		return (0);
1047 	}
1048 
1049 	if (p_canwait(td, p)) {
1050 		PROC_UNLOCK(p);
1051 		return (0);
1052 	}
1053 
1054 	if (((options & WEXITED) == 0) && (p->p_state == PRS_ZOMBIE)) {
1055 		PROC_UNLOCK(p);
1056 		return (0);
1057 	}
1058 
1059 	/*
1060 	 * This special case handles a kthread spawned by linux_clone
1061 	 * (see linux_misc.c).  The linux_wait4 and linux_waitpid
1062 	 * functions need to be able to distinguish between waiting
1063 	 * on a process and waiting on a thread.  It is a thread if
1064 	 * p_sigparent is not SIGCHLD, and the WLINUXCLONE option
1065 	 * signifies we want to wait for threads and not processes.
1066 	 */
1067 	if ((p->p_sigparent != SIGCHLD) ^
1068 	    ((options & WLINUXCLONE) != 0)) {
1069 		PROC_UNLOCK(p);
1070 		return (0);
1071 	}
1072 
1073 	if (siginfo != NULL) {
1074 		bzero(siginfo, sizeof(*siginfo));
1075 		siginfo->si_errno = 0;
1076 
1077 		/*
1078 		 * SUSv4 requires that the si_signo value is always
1079 		 * SIGCHLD. Obey it despite the rfork(2) interface
1080 		 * allows to request other signal for child exit
1081 		 * notification.
1082 		 */
1083 		siginfo->si_signo = SIGCHLD;
1084 
1085 		/*
1086 		 *  This is still a rough estimate.  We will fix the
1087 		 *  cases TRAPPED, STOPPED, and CONTINUED later.
1088 		 */
1089 		if (WCOREDUMP(p->p_xsig)) {
1090 			siginfo->si_code = CLD_DUMPED;
1091 			siginfo->si_status = WTERMSIG(p->p_xsig);
1092 		} else if (WIFSIGNALED(p->p_xsig)) {
1093 			siginfo->si_code = CLD_KILLED;
1094 			siginfo->si_status = WTERMSIG(p->p_xsig);
1095 		} else {
1096 			siginfo->si_code = CLD_EXITED;
1097 			siginfo->si_status = p->p_xexit;
1098 		}
1099 
1100 		siginfo->si_pid = p->p_pid;
1101 		siginfo->si_uid = p->p_ucred->cr_uid;
1102 
1103 		/*
1104 		 * The si_addr field would be useful additional
1105 		 * detail, but apparently the PC value may be lost
1106 		 * when we reach this point.  bzero() above sets
1107 		 * siginfo->si_addr to NULL.
1108 		 */
1109 	}
1110 
1111 	/*
1112 	 * There should be no reason to limit resources usage info to
1113 	 * exited processes only.  A snapshot about any resources used
1114 	 * by a stopped process may be exactly what is needed.
1115 	 */
1116 	if (wrusage != NULL) {
1117 		rup = &wrusage->wru_self;
1118 		*rup = p->p_ru;
1119 		PROC_STATLOCK(p);
1120 		calcru(p, &rup->ru_utime, &rup->ru_stime);
1121 		PROC_STATUNLOCK(p);
1122 
1123 		rup = &wrusage->wru_children;
1124 		*rup = p->p_stats->p_cru;
1125 		calccru(p, &rup->ru_utime, &rup->ru_stime);
1126 	}
1127 
1128 	if (p->p_state == PRS_ZOMBIE && !check_only) {
1129 		proc_reap(td, p, status, options);
1130 		return (-1);
1131 	}
1132 	return (1);
1133 }
1134 
1135 int
1136 kern_wait(struct thread *td, pid_t pid, int *status, int options,
1137     struct rusage *rusage)
1138 {
1139 	struct __wrusage wru, *wrup;
1140 	idtype_t idtype;
1141 	id_t id;
1142 	int ret;
1143 
1144 	/*
1145 	 * Translate the special pid values into the (idtype, pid)
1146 	 * pair for kern_wait6.  The WAIT_MYPGRP case is handled by
1147 	 * kern_wait6() on its own.
1148 	 */
1149 	if (pid == WAIT_ANY) {
1150 		idtype = P_ALL;
1151 		id = 0;
1152 	} else if (pid < 0) {
1153 		idtype = P_PGID;
1154 		id = (id_t)-pid;
1155 	} else {
1156 		idtype = P_PID;
1157 		id = (id_t)pid;
1158 	}
1159 
1160 	if (rusage != NULL)
1161 		wrup = &wru;
1162 	else
1163 		wrup = NULL;
1164 
1165 	/*
1166 	 * For backward compatibility we implicitly add flags WEXITED
1167 	 * and WTRAPPED here.
1168 	 */
1169 	options |= WEXITED | WTRAPPED;
1170 	ret = kern_wait6(td, idtype, id, status, options, wrup, NULL);
1171 	if (rusage != NULL)
1172 		*rusage = wru.wru_self;
1173 	return (ret);
1174 }
1175 
1176 static void
1177 report_alive_proc(struct thread *td, struct proc *p, siginfo_t *siginfo,
1178     int *status, int options, int si_code)
1179 {
1180 	bool cont;
1181 
1182 	PROC_LOCK_ASSERT(p, MA_OWNED);
1183 	sx_assert(&proctree_lock, SA_XLOCKED);
1184 	MPASS(si_code == CLD_TRAPPED || si_code == CLD_STOPPED ||
1185 	    si_code == CLD_CONTINUED);
1186 
1187 	cont = si_code == CLD_CONTINUED;
1188 	if ((options & WNOWAIT) == 0) {
1189 		if (cont)
1190 			p->p_flag &= ~P_CONTINUED;
1191 		else
1192 			p->p_flag |= P_WAITED;
1193 		PROC_LOCK(td->td_proc);
1194 		sigqueue_take(p->p_ksi);
1195 		PROC_UNLOCK(td->td_proc);
1196 	}
1197 	sx_xunlock(&proctree_lock);
1198 	if (siginfo != NULL) {
1199 		siginfo->si_code = si_code;
1200 		siginfo->si_status = cont ? SIGCONT : p->p_xsig;
1201 	}
1202 	if (status != NULL)
1203 		*status = cont ? SIGCONT : W_STOPCODE(p->p_xsig);
1204 	PROC_UNLOCK(p);
1205 	td->td_retval[0] = p->p_pid;
1206 }
1207 
1208 int
1209 kern_wait6(struct thread *td, idtype_t idtype, id_t id, int *status,
1210     int options, struct __wrusage *wrusage, siginfo_t *siginfo)
1211 {
1212 	struct proc *p, *q;
1213 	pid_t pid;
1214 	int error, nfound, ret;
1215 	bool report;
1216 
1217 	AUDIT_ARG_VALUE((int)idtype);	/* XXX - This is likely wrong! */
1218 	AUDIT_ARG_PID((pid_t)id);	/* XXX - This may be wrong! */
1219 	AUDIT_ARG_VALUE(options);
1220 
1221 	q = td->td_proc;
1222 
1223 	if ((pid_t)id == WAIT_MYPGRP && (idtype == P_PID || idtype == P_PGID)) {
1224 		PROC_LOCK(q);
1225 		id = (id_t)q->p_pgid;
1226 		PROC_UNLOCK(q);
1227 		idtype = P_PGID;
1228 	}
1229 
1230 	/* If we don't know the option, just return. */
1231 	if ((options & ~(WUNTRACED | WNOHANG | WCONTINUED | WNOWAIT |
1232 	    WEXITED | WTRAPPED | WLINUXCLONE)) != 0)
1233 		return (EINVAL);
1234 	if ((options & (WEXITED | WUNTRACED | WCONTINUED | WTRAPPED)) == 0) {
1235 		/*
1236 		 * We will be unable to find any matching processes,
1237 		 * because there are no known events to look for.
1238 		 * Prefer to return error instead of blocking
1239 		 * indefinitely.
1240 		 */
1241 		return (EINVAL);
1242 	}
1243 
1244 loop:
1245 	if (q->p_flag & P_STATCHILD) {
1246 		PROC_LOCK(q);
1247 		q->p_flag &= ~P_STATCHILD;
1248 		PROC_UNLOCK(q);
1249 	}
1250 	sx_xlock(&proctree_lock);
1251 loop_locked:
1252 	nfound = 0;
1253 	LIST_FOREACH(p, &q->p_children, p_sibling) {
1254 		pid = p->p_pid;
1255 		ret = proc_to_reap(td, p, idtype, id, status, options,
1256 		    wrusage, siginfo, 0);
1257 		if (ret == 0)
1258 			continue;
1259 		else if (ret != 1) {
1260 			td->td_retval[0] = pid;
1261 			return (0);
1262 		}
1263 
1264 		nfound++;
1265 		PROC_LOCK_ASSERT(p, MA_OWNED);
1266 
1267 		if ((options & WTRAPPED) != 0 &&
1268 		    (p->p_flag & P_TRACED) != 0) {
1269 			PROC_SLOCK(p);
1270 			report =
1271 			    ((p->p_flag & (P_STOPPED_TRACE | P_STOPPED_SIG)) &&
1272 			    p->p_suspcount == p->p_numthreads &&
1273 			    (p->p_flag & P_WAITED) == 0);
1274 			PROC_SUNLOCK(p);
1275 			if (report) {
1276 			CTR4(KTR_PTRACE,
1277 			    "wait: returning trapped pid %d status %#x "
1278 			    "(xstat %d) xthread %d",
1279 			    p->p_pid, W_STOPCODE(p->p_xsig), p->p_xsig,
1280 			    p->p_xthread != NULL ?
1281 			    p->p_xthread->td_tid : -1);
1282 				report_alive_proc(td, p, siginfo, status,
1283 				    options, CLD_TRAPPED);
1284 				return (0);
1285 			}
1286 		}
1287 		if ((options & WUNTRACED) != 0 &&
1288 		    (p->p_flag & P_STOPPED_SIG) != 0) {
1289 			PROC_SLOCK(p);
1290 			report = (p->p_suspcount == p->p_numthreads &&
1291 			    ((p->p_flag & P_WAITED) == 0));
1292 			PROC_SUNLOCK(p);
1293 			if (report) {
1294 				report_alive_proc(td, p, siginfo, status,
1295 				    options, CLD_STOPPED);
1296 				return (0);
1297 			}
1298 		}
1299 		if ((options & WCONTINUED) != 0 &&
1300 		    (p->p_flag & P_CONTINUED) != 0) {
1301 			report_alive_proc(td, p, siginfo, status, options,
1302 			    CLD_CONTINUED);
1303 			return (0);
1304 		}
1305 		PROC_UNLOCK(p);
1306 	}
1307 
1308 	/*
1309 	 * Look in the orphans list too, to allow the parent to
1310 	 * collect it's child exit status even if child is being
1311 	 * debugged.
1312 	 *
1313 	 * Debugger detaches from the parent upon successful
1314 	 * switch-over from parent to child.  At this point due to
1315 	 * re-parenting the parent loses the child to debugger and a
1316 	 * wait4(2) call would report that it has no children to wait
1317 	 * for.  By maintaining a list of orphans we allow the parent
1318 	 * to successfully wait until the child becomes a zombie.
1319 	 */
1320 	if (nfound == 0) {
1321 		LIST_FOREACH(p, &q->p_orphans, p_orphan) {
1322 			ret = proc_to_reap(td, p, idtype, id, NULL, options,
1323 			    NULL, NULL, 1);
1324 			if (ret != 0) {
1325 				KASSERT(ret != -1, ("reaped an orphan (pid %d)",
1326 				    (int)td->td_retval[0]));
1327 				PROC_UNLOCK(p);
1328 				nfound++;
1329 				break;
1330 			}
1331 		}
1332 	}
1333 	if (nfound == 0) {
1334 		sx_xunlock(&proctree_lock);
1335 		return (ECHILD);
1336 	}
1337 	if (options & WNOHANG) {
1338 		sx_xunlock(&proctree_lock);
1339 		td->td_retval[0] = 0;
1340 		return (0);
1341 	}
1342 	PROC_LOCK(q);
1343 	if (q->p_flag & P_STATCHILD) {
1344 		q->p_flag &= ~P_STATCHILD;
1345 		PROC_UNLOCK(q);
1346 		goto loop_locked;
1347 	}
1348 	sx_xunlock(&proctree_lock);
1349 	error = msleep(q, &q->p_mtx, PWAIT | PCATCH | PDROP, "wait", 0);
1350 	if (error)
1351 		return (error);
1352 	goto loop;
1353 }
1354 
1355 void
1356 proc_add_orphan(struct proc *child, struct proc *parent)
1357 {
1358 
1359 	sx_assert(&proctree_lock, SX_XLOCKED);
1360 	KASSERT((child->p_flag & P_TRACED) != 0,
1361 	    ("proc_add_orphan: not traced"));
1362 
1363 	if (LIST_EMPTY(&parent->p_orphans)) {
1364 		child->p_treeflag |= P_TREE_FIRST_ORPHAN;
1365 		LIST_INSERT_HEAD(&parent->p_orphans, child, p_orphan);
1366 	} else {
1367 		LIST_INSERT_AFTER(LIST_FIRST(&parent->p_orphans),
1368 		    child, p_orphan);
1369 	}
1370 	child->p_treeflag |= P_TREE_ORPHANED;
1371 }
1372 
1373 /*
1374  * Make process 'parent' the new parent of process 'child'.
1375  * Must be called with an exclusive hold of proctree lock.
1376  */
1377 void
1378 proc_reparent(struct proc *child, struct proc *parent, bool set_oppid)
1379 {
1380 
1381 	sx_assert(&proctree_lock, SX_XLOCKED);
1382 	PROC_LOCK_ASSERT(child, MA_OWNED);
1383 	if (child->p_pptr == parent)
1384 		return;
1385 
1386 	PROC_LOCK(child->p_pptr);
1387 	sigqueue_take(child->p_ksi);
1388 	PROC_UNLOCK(child->p_pptr);
1389 	LIST_REMOVE(child, p_sibling);
1390 	LIST_INSERT_HEAD(&parent->p_children, child, p_sibling);
1391 
1392 	proc_clear_orphan(child);
1393 	if ((child->p_flag & P_TRACED) != 0) {
1394 		proc_add_orphan(child, child->p_pptr);
1395 	}
1396 
1397 	child->p_pptr = parent;
1398 	if (set_oppid)
1399 		child->p_oppid = parent->p_pid;
1400 }
1401