xref: /freebsd/sys/kern/kern_exit.c (revision 6aae3517ed2500fb963ba0a4264b4756088dd0f4)
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_ddb.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/ktr.h>
52 #include <sys/malloc.h>
53 #include <sys/lock.h>
54 #include <sys/mutex.h>
55 #include <sys/proc.h>
56 #include <sys/procdesc.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/sysctl.h>
70 #include <sys/syslog.h>
71 #include <sys/ptrace.h>
72 #include <sys/acct.h>		/* for acct_process() function prototype */
73 #include <sys/filedesc.h>
74 #include <sys/sdt.h>
75 #include <sys/shm.h>
76 #include <sys/sem.h>
77 #include <sys/sysent.h>
78 #include <sys/timers.h>
79 #include <sys/umtxvar.h>
80 #ifdef KTRACE
81 #include <sys/ktrace.h>
82 #endif
83 
84 #include <security/audit/audit.h>
85 #include <security/mac/mac_framework.h>
86 
87 #include <vm/vm.h>
88 #include <vm/vm_extern.h>
89 #include <vm/vm_param.h>
90 #include <vm/pmap.h>
91 #include <vm/vm_map.h>
92 #include <vm/vm_page.h>
93 #include <vm/uma.h>
94 
95 #ifdef KDTRACE_HOOKS
96 #include <sys/dtrace_bsd.h>
97 dtrace_execexit_func_t	dtrace_fasttrap_exit;
98 #endif
99 
100 SDT_PROVIDER_DECLARE(proc);
101 SDT_PROBE_DEFINE1(proc, , , exit, "int");
102 
103 static int kern_kill_on_dbg_exit = 1;
104 SYSCTL_INT(_kern, OID_AUTO, kill_on_debugger_exit, CTLFLAG_RWTUN,
105     &kern_kill_on_dbg_exit, 0,
106     "Kill ptraced processes when debugger exits");
107 
108 static bool kern_wait_dequeue_sigchld = 1;
109 SYSCTL_BOOL(_kern, OID_AUTO, wait_dequeue_sigchld, CTLFLAG_RWTUN,
110     &kern_wait_dequeue_sigchld, 0,
111     "Dequeue SIGCHLD on wait(2) for live process");
112 
113 struct proc *
114 proc_realparent(struct proc *child)
115 {
116 	struct proc *p, *parent;
117 
118 	sx_assert(&proctree_lock, SX_LOCKED);
119 	if ((child->p_treeflag & P_TREE_ORPHANED) == 0)
120 		return (child->p_pptr->p_pid == child->p_oppid ?
121 		    child->p_pptr : child->p_reaper);
122 	for (p = child; (p->p_treeflag & P_TREE_FIRST_ORPHAN) == 0;) {
123 		/* Cannot use LIST_PREV(), since the list head is not known. */
124 		p = __containerof(p->p_orphan.le_prev, struct proc,
125 		    p_orphan.le_next);
126 		KASSERT((p->p_treeflag & P_TREE_ORPHANED) != 0,
127 		    ("missing P_ORPHAN %p", p));
128 	}
129 	parent = __containerof(p->p_orphan.le_prev, struct proc,
130 	    p_orphans.lh_first);
131 	return (parent);
132 }
133 
134 void
135 reaper_abandon_children(struct proc *p, bool exiting)
136 {
137 	struct proc *p1, *p2, *ptmp;
138 
139 	sx_assert(&proctree_lock, SX_LOCKED);
140 	KASSERT(p != initproc, ("reaper_abandon_children for initproc"));
141 	if ((p->p_treeflag & P_TREE_REAPER) == 0)
142 		return;
143 	p1 = p->p_reaper;
144 	LIST_FOREACH_SAFE(p2, &p->p_reaplist, p_reapsibling, ptmp) {
145 		LIST_REMOVE(p2, p_reapsibling);
146 		p2->p_reaper = p1;
147 		p2->p_reapsubtree = p->p_reapsubtree;
148 		LIST_INSERT_HEAD(&p1->p_reaplist, p2, p_reapsibling);
149 		if (exiting && p2->p_pptr == p) {
150 			PROC_LOCK(p2);
151 			proc_reparent(p2, p1, true);
152 			PROC_UNLOCK(p2);
153 		}
154 	}
155 	KASSERT(LIST_EMPTY(&p->p_reaplist), ("p_reaplist not empty"));
156 	p->p_treeflag &= ~P_TREE_REAPER;
157 }
158 
159 static void
160 reaper_clear(struct proc *p)
161 {
162 	struct proc *p1;
163 	bool clear;
164 
165 	sx_assert(&proctree_lock, SX_LOCKED);
166 	LIST_REMOVE(p, p_reapsibling);
167 	if (p->p_reapsubtree == 1)
168 		return;
169 	clear = true;
170 	LIST_FOREACH(p1, &p->p_reaper->p_reaplist, p_reapsibling) {
171 		if (p1->p_reapsubtree == p->p_reapsubtree) {
172 			clear = false;
173 			break;
174 		}
175 	}
176 	if (clear)
177 		proc_id_clear(PROC_ID_REAP, p->p_reapsubtree);
178 }
179 
180 void
181 proc_clear_orphan(struct proc *p)
182 {
183 	struct proc *p1;
184 
185 	sx_assert(&proctree_lock, SA_XLOCKED);
186 	if ((p->p_treeflag & P_TREE_ORPHANED) == 0)
187 		return;
188 	if ((p->p_treeflag & P_TREE_FIRST_ORPHAN) != 0) {
189 		p1 = LIST_NEXT(p, p_orphan);
190 		if (p1 != NULL)
191 			p1->p_treeflag |= P_TREE_FIRST_ORPHAN;
192 		p->p_treeflag &= ~P_TREE_FIRST_ORPHAN;
193 	}
194 	LIST_REMOVE(p, p_orphan);
195 	p->p_treeflag &= ~P_TREE_ORPHANED;
196 }
197 
198 void
199 exit_onexit(struct proc *p)
200 {
201 	MPASS(p->p_numthreads == 1);
202 	umtx_thread_exit(FIRST_THREAD_IN_PROC(p));
203 }
204 
205 /*
206  * exit -- death of process.
207  */
208 void
209 sys_sys_exit(struct thread *td, struct sys_exit_args *uap)
210 {
211 
212 	exit1(td, uap->rval, 0);
213 	/* NOTREACHED */
214 }
215 
216 /*
217  * Exit: deallocate address space and other resources, change proc state to
218  * zombie, and unlink proc from allproc and parent's lists.  Save exit status
219  * and rusage for wait().  Check for child processes and orphan them.
220  */
221 void
222 exit1(struct thread *td, int rval, int signo)
223 {
224 	struct proc *p, *nq, *q, *t;
225 	struct thread *tdt;
226 	ksiginfo_t *ksi, *ksi1;
227 	int signal_parent;
228 
229 	mtx_assert(&Giant, MA_NOTOWNED);
230 	KASSERT(rval == 0 || signo == 0, ("exit1 rv %d sig %d", rval, signo));
231 	TSPROCEXIT(td->td_proc->p_pid);
232 
233 	p = td->td_proc;
234 	/*
235 	 * XXX in case we're rebooting we just let init die in order to
236 	 * work around an unsolved stack overflow seen very late during
237 	 * shutdown on sparc64 when the gmirror worker process exists.
238 	 * XXX what to do now that sparc64 is gone... remove if?
239 	 */
240 	if (p == initproc && rebooting == 0) {
241 		printf("init died (signal %d, exit %d)\n", signo, rval);
242 		panic("Going nowhere without my init!");
243 	}
244 
245 	/*
246 	 * Deref SU mp, since the thread does not return to userspace.
247 	 */
248 	td_softdep_cleanup(td);
249 
250 	/*
251 	 * MUST abort all other threads before proceeding past here.
252 	 */
253 	PROC_LOCK(p);
254 	/*
255 	 * First check if some other thread or external request got
256 	 * here before us.  If so, act appropriately: exit or suspend.
257 	 * We must ensure that stop requests are handled before we set
258 	 * P_WEXIT.
259 	 */
260 	thread_suspend_check(0);
261 	while (p->p_flag & P_HADTHREADS) {
262 		/*
263 		 * Kill off the other threads. This requires
264 		 * some co-operation from other parts of the kernel
265 		 * so it may not be instantaneous.  With this state set
266 		 * any thread entering the kernel from userspace will
267 		 * thread_exit() in trap().  Any thread attempting to
268 		 * sleep will return immediately with EINTR or EWOULDBLOCK
269 		 * which will hopefully force them to back out to userland
270 		 * freeing resources as they go.  Any thread attempting
271 		 * to return to userland will thread_exit() from userret().
272 		 * thread_exit() will unsuspend us when the last of the
273 		 * other threads exits.
274 		 * If there is already a thread singler after resumption,
275 		 * calling thread_single will fail; in that case, we just
276 		 * re-check all suspension request, the thread should
277 		 * either be suspended there or exit.
278 		 */
279 		if (!thread_single(p, SINGLE_EXIT))
280 			/*
281 			 * All other activity in this process is now
282 			 * stopped.  Threading support has been turned
283 			 * off.
284 			 */
285 			break;
286 		/*
287 		 * Recheck for new stop or suspend requests which
288 		 * might appear while process lock was dropped in
289 		 * thread_single().
290 		 */
291 		thread_suspend_check(0);
292 	}
293 	KASSERT(p->p_numthreads == 1,
294 	    ("exit1: proc %p exiting with %d threads", p, p->p_numthreads));
295 	racct_sub(p, RACCT_NTHR, 1);
296 
297 	/* Let event handler change exit status */
298 	p->p_xexit = rval;
299 	p->p_xsig = signo;
300 
301 	/*
302 	 * Ignore any pending request to stop due to a stop signal.
303 	 * Once P_WEXIT is set, future requests will be ignored as
304 	 * well.
305 	 */
306 	p->p_flag &= ~P_STOPPED_SIG;
307 	KASSERT(!P_SHOULDSTOP(p), ("exiting process is stopped"));
308 
309 	/* Note that we are exiting. */
310 	p->p_flag |= P_WEXIT;
311 
312 	/*
313 	 * Wait for any processes that have a hold on our vmspace to
314 	 * release their reference.
315 	 */
316 	while (p->p_lock > 0)
317 		msleep(&p->p_lock, &p->p_mtx, PWAIT, "exithold", 0);
318 
319 	PROC_UNLOCK(p);
320 	/* Drain the limit callout while we don't have the proc locked */
321 	callout_drain(&p->p_limco);
322 
323 #ifdef AUDIT
324 	/*
325 	 * The Sun BSM exit token contains two components: an exit status as
326 	 * passed to exit(), and a return value to indicate what sort of exit
327 	 * it was.  The exit status is WEXITSTATUS(rv), but it's not clear
328 	 * what the return value is.
329 	 */
330 	AUDIT_ARG_EXIT(rval, 0);
331 	AUDIT_SYSCALL_EXIT(0, td);
332 #endif
333 
334 	/* Are we a task leader with peers? */
335 	if (p->p_peers != NULL && p == p->p_leader) {
336 		mtx_lock(&ppeers_lock);
337 		q = p->p_peers;
338 		while (q != NULL) {
339 			PROC_LOCK(q);
340 			kern_psignal(q, SIGKILL);
341 			PROC_UNLOCK(q);
342 			q = q->p_peers;
343 		}
344 		while (p->p_peers != NULL)
345 			msleep(p, &ppeers_lock, PWAIT, "exit1", 0);
346 		mtx_unlock(&ppeers_lock);
347 	}
348 
349 	itimers_exit(p);
350 
351 	/*
352 	 * Check if any loadable modules need anything done at process exit.
353 	 * E.g. SYSV IPC stuff.
354 	 * Event handler could change exit status.
355 	 * XXX what if one of these generates an error?
356 	 */
357 	EVENTHANDLER_DIRECT_INVOKE(process_exit, p);
358 
359 	/*
360 	 * If parent is waiting for us to exit or exec,
361 	 * P_PPWAIT is set; we will wakeup the parent below.
362 	 */
363 	PROC_LOCK(p);
364 	stopprofclock(p);
365 	p->p_ptevents = 0;
366 
367 	/*
368 	 * Stop the real interval timer.  If the handler is currently
369 	 * executing, prevent it from rearming itself and let it finish.
370 	 */
371 	if (timevalisset(&p->p_realtimer.it_value) &&
372 	    _callout_stop_safe(&p->p_itcallout, CS_EXECUTING, NULL) == 0) {
373 		timevalclear(&p->p_realtimer.it_interval);
374 		msleep(&p->p_itcallout, &p->p_mtx, PWAIT, "ritwait", 0);
375 		KASSERT(!timevalisset(&p->p_realtimer.it_value),
376 		    ("realtime timer is still armed"));
377 	}
378 
379 	PROC_UNLOCK(p);
380 
381 	if (p->p_sysent->sv_onexit != NULL)
382 		p->p_sysent->sv_onexit(p);
383 	seltdfini(td);
384 
385 	/*
386 	 * Reset any sigio structures pointing to us as a result of
387 	 * F_SETOWN with our pid.  The P_WEXIT flag interlocks with fsetown().
388 	 */
389 	funsetownlst(&p->p_sigiolst);
390 
391 	/*
392 	 * Close open files and release open-file table.
393 	 * This may block!
394 	 */
395 	pdescfree(td);
396 	fdescfree(td);
397 
398 	/*
399 	 * If this thread tickled GEOM, we need to wait for the giggling to
400 	 * stop before we return to userland
401 	 */
402 	if (td->td_pflags & TDP_GEOM)
403 		g_waitidle();
404 
405 	/*
406 	 * Remove ourself from our leader's peer list and wake our leader.
407 	 */
408 	if (p->p_leader->p_peers != NULL) {
409 		mtx_lock(&ppeers_lock);
410 		if (p->p_leader->p_peers != NULL) {
411 			q = p->p_leader;
412 			while (q->p_peers != p)
413 				q = q->p_peers;
414 			q->p_peers = p->p_peers;
415 			wakeup(p->p_leader);
416 		}
417 		mtx_unlock(&ppeers_lock);
418 	}
419 
420 	vmspace_exit(td);
421 	(void)acct_process(td);
422 
423 #ifdef KTRACE
424 	ktrprocexit(td);
425 #endif
426 	/*
427 	 * Release reference to text vnode
428 	 */
429 	if (p->p_textvp != NULL) {
430 		vrele(p->p_textvp);
431 		p->p_textvp = NULL;
432 	}
433 
434 	/*
435 	 * Release our limits structure.
436 	 */
437 	lim_free(p->p_limit);
438 	p->p_limit = NULL;
439 
440 	tidhash_remove(td);
441 
442 	/*
443 	 * Call machine-dependent code to release any
444 	 * machine-dependent resources other than the address space.
445 	 * The address space is released by "vmspace_exitfree(p)" in
446 	 * vm_waitproc().
447 	 */
448 	cpu_exit(td);
449 
450 	WITNESS_WARN(WARN_PANIC, NULL, "process (pid %d) exiting", p->p_pid);
451 
452 	/*
453 	 * Remove from allproc. It still sits in the hash.
454 	 */
455 	sx_xlock(&allproc_lock);
456 	LIST_REMOVE(p, p_list);
457 
458 #ifdef DDB
459 	/*
460 	 * Used by ddb's 'ps' command to find this process via the
461 	 * pidhash.
462 	 */
463 	p->p_list.le_prev = NULL;
464 #endif
465 	sx_xunlock(&allproc_lock);
466 
467 	sx_xlock(&proctree_lock);
468 	PROC_LOCK(p);
469 	p->p_flag &= ~(P_TRACED | P_PPWAIT | P_PPTRACE);
470 	PROC_UNLOCK(p);
471 
472 	/*
473 	 * killjobc() might drop and re-acquire proctree_lock to
474 	 * revoke control tty if exiting process was a session leader.
475 	 */
476 	killjobc();
477 
478 	/*
479 	 * Reparent all children processes:
480 	 * - traced ones to the original parent (or init if we are that parent)
481 	 * - the rest to init
482 	 */
483 	q = LIST_FIRST(&p->p_children);
484 	if (q != NULL)		/* only need this if any child is S_ZOMB */
485 		wakeup(q->p_reaper);
486 	for (; q != NULL; q = nq) {
487 		nq = LIST_NEXT(q, p_sibling);
488 		ksi = ksiginfo_alloc(TRUE);
489 		PROC_LOCK(q);
490 		q->p_sigparent = SIGCHLD;
491 
492 		if ((q->p_flag & P_TRACED) == 0) {
493 			proc_reparent(q, q->p_reaper, true);
494 			if (q->p_state == PRS_ZOMBIE) {
495 				/*
496 				 * Inform reaper about the reparented
497 				 * zombie, since wait(2) has something
498 				 * new to report.  Guarantee queueing
499 				 * of the SIGCHLD signal, similar to
500 				 * the _exit() behaviour, by providing
501 				 * our ksiginfo.  Ksi is freed by the
502 				 * signal delivery.
503 				 */
504 				if (q->p_ksi == NULL) {
505 					ksi1 = NULL;
506 				} else {
507 					ksiginfo_copy(q->p_ksi, ksi);
508 					ksi->ksi_flags |= KSI_INS;
509 					ksi1 = ksi;
510 					ksi = NULL;
511 				}
512 				PROC_LOCK(q->p_reaper);
513 				pksignal(q->p_reaper, SIGCHLD, ksi1);
514 				PROC_UNLOCK(q->p_reaper);
515 			} else if (q->p_pdeathsig > 0) {
516 				/*
517 				 * The child asked to received a signal
518 				 * when we exit.
519 				 */
520 				kern_psignal(q, q->p_pdeathsig);
521 			}
522 		} else {
523 			/*
524 			 * Traced processes are killed by default
525 			 * since their existence means someone is
526 			 * screwing up.
527 			 */
528 			t = proc_realparent(q);
529 			if (t == p) {
530 				proc_reparent(q, q->p_reaper, true);
531 			} else {
532 				PROC_LOCK(t);
533 				proc_reparent(q, t, true);
534 				PROC_UNLOCK(t);
535 			}
536 			/*
537 			 * Since q was found on our children list, the
538 			 * proc_reparent() call moved q to the orphan
539 			 * list due to present P_TRACED flag. Clear
540 			 * orphan link for q now while q is locked.
541 			 */
542 			proc_clear_orphan(q);
543 			q->p_flag &= ~P_TRACED;
544 			q->p_flag2 &= ~P2_PTRACE_FSTP;
545 			q->p_ptevents = 0;
546 			p->p_xthread = NULL;
547 			FOREACH_THREAD_IN_PROC(q, tdt) {
548 				tdt->td_dbgflags &= ~(TDB_SUSPEND | TDB_XSIG |
549 				    TDB_FSTP);
550 				tdt->td_xsig = 0;
551 			}
552 			if (kern_kill_on_dbg_exit) {
553 				q->p_flag &= ~P_STOPPED_TRACE;
554 				kern_psignal(q, SIGKILL);
555 			} else if ((q->p_flag & (P_STOPPED_TRACE |
556 			    P_STOPPED_SIG)) != 0) {
557 				sigqueue_delete_proc(q, SIGTRAP);
558 				ptrace_unsuspend(q);
559 			}
560 		}
561 		PROC_UNLOCK(q);
562 		if (ksi != NULL)
563 			ksiginfo_free(ksi);
564 	}
565 
566 	/*
567 	 * Also get rid of our orphans.
568 	 */
569 	while ((q = LIST_FIRST(&p->p_orphans)) != NULL) {
570 		PROC_LOCK(q);
571 		KASSERT(q->p_oppid == p->p_pid,
572 		    ("orphan %p of %p has unexpected oppid %d", q, p,
573 		    q->p_oppid));
574 		q->p_oppid = q->p_reaper->p_pid;
575 
576 		/*
577 		 * If we are the real parent of this process
578 		 * but it has been reparented to a debugger, then
579 		 * check if it asked for a signal when we exit.
580 		 */
581 		if (q->p_pdeathsig > 0)
582 			kern_psignal(q, q->p_pdeathsig);
583 		CTR2(KTR_PTRACE, "exit: pid %d, clearing orphan %d", p->p_pid,
584 		    q->p_pid);
585 		proc_clear_orphan(q);
586 		PROC_UNLOCK(q);
587 	}
588 
589 #ifdef KDTRACE_HOOKS
590 	if (SDT_PROBES_ENABLED()) {
591 		int reason = CLD_EXITED;
592 		if (WCOREDUMP(signo))
593 			reason = CLD_DUMPED;
594 		else if (WIFSIGNALED(signo))
595 			reason = CLD_KILLED;
596 		SDT_PROBE1(proc, , , exit, reason);
597 	}
598 #endif
599 
600 	/* Save exit status. */
601 	PROC_LOCK(p);
602 	p->p_xthread = td;
603 
604 	if (p->p_sysent->sv_ontdexit != NULL)
605 		p->p_sysent->sv_ontdexit(td);
606 
607 #ifdef KDTRACE_HOOKS
608 	/*
609 	 * Tell the DTrace fasttrap provider about the exit if it
610 	 * has declared an interest.
611 	 */
612 	if (dtrace_fasttrap_exit)
613 		dtrace_fasttrap_exit(p);
614 #endif
615 
616 	/*
617 	 * Notify interested parties of our demise.
618 	 */
619 	KNOTE_LOCKED(p->p_klist, NOTE_EXIT);
620 
621 	/*
622 	 * If this is a process with a descriptor, we may not need to deliver
623 	 * a signal to the parent.  proctree_lock is held over
624 	 * procdesc_exit() to serialize concurrent calls to close() and
625 	 * exit().
626 	 */
627 	signal_parent = 0;
628 	if (p->p_procdesc == NULL || procdesc_exit(p)) {
629 		/*
630 		 * Notify parent that we're gone.  If parent has the
631 		 * PS_NOCLDWAIT flag set, or if the handler is set to SIG_IGN,
632 		 * notify process 1 instead (and hope it will handle this
633 		 * situation).
634 		 */
635 		PROC_LOCK(p->p_pptr);
636 		mtx_lock(&p->p_pptr->p_sigacts->ps_mtx);
637 		if (p->p_pptr->p_sigacts->ps_flag &
638 		    (PS_NOCLDWAIT | PS_CLDSIGIGN)) {
639 			struct proc *pp;
640 
641 			mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx);
642 			pp = p->p_pptr;
643 			PROC_UNLOCK(pp);
644 			proc_reparent(p, p->p_reaper, true);
645 			p->p_sigparent = SIGCHLD;
646 			PROC_LOCK(p->p_pptr);
647 
648 			/*
649 			 * Notify parent, so in case he was wait(2)ing or
650 			 * executing waitpid(2) with our pid, he will
651 			 * continue.
652 			 */
653 			wakeup(pp);
654 		} else
655 			mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx);
656 
657 		if (p->p_pptr == p->p_reaper || p->p_pptr == initproc) {
658 			signal_parent = 1;
659 		} else if (p->p_sigparent != 0) {
660 			if (p->p_sigparent == SIGCHLD) {
661 				signal_parent = 1;
662 			} else { /* LINUX thread */
663 				signal_parent = 2;
664 			}
665 		}
666 	} else
667 		PROC_LOCK(p->p_pptr);
668 	sx_xunlock(&proctree_lock);
669 
670 	if (signal_parent == 1) {
671 		childproc_exited(p);
672 	} else if (signal_parent == 2) {
673 		kern_psignal(p->p_pptr, p->p_sigparent);
674 	}
675 
676 	/* Tell the prison that we are gone. */
677 	prison_proc_free(p->p_ucred->cr_prison);
678 
679 	/*
680 	 * The state PRS_ZOMBIE prevents other proesses from sending
681 	 * signal to the process, to avoid memory leak, we free memory
682 	 * for signal queue at the time when the state is set.
683 	 */
684 	sigqueue_flush(&p->p_sigqueue);
685 	sigqueue_flush(&td->td_sigqueue);
686 
687 	/*
688 	 * We have to wait until after acquiring all locks before
689 	 * changing p_state.  We need to avoid all possible context
690 	 * switches (including ones from blocking on a mutex) while
691 	 * marked as a zombie.  We also have to set the zombie state
692 	 * before we release the parent process' proc lock to avoid
693 	 * a lost wakeup.  So, we first call wakeup, then we grab the
694 	 * sched lock, update the state, and release the parent process'
695 	 * proc lock.
696 	 */
697 	wakeup(p->p_pptr);
698 	cv_broadcast(&p->p_pwait);
699 	sched_exit(p->p_pptr, td);
700 	PROC_SLOCK(p);
701 	p->p_state = PRS_ZOMBIE;
702 	PROC_UNLOCK(p->p_pptr);
703 
704 	/*
705 	 * Save our children's rusage information in our exit rusage.
706 	 */
707 	PROC_STATLOCK(p);
708 	ruadd(&p->p_ru, &p->p_rux, &p->p_stats->p_cru, &p->p_crux);
709 	PROC_STATUNLOCK(p);
710 
711 	/*
712 	 * Make sure the scheduler takes this thread out of its tables etc.
713 	 * This will also release this thread's reference to the ucred.
714 	 * Other thread parts to release include pcb bits and such.
715 	 */
716 	thread_exit();
717 }
718 
719 #ifndef _SYS_SYSPROTO_H_
720 struct abort2_args {
721 	char *why;
722 	int nargs;
723 	void **args;
724 };
725 #endif
726 
727 int
728 sys_abort2(struct thread *td, struct abort2_args *uap)
729 {
730 	struct proc *p = td->td_proc;
731 	struct sbuf *sb;
732 	void *uargs[16];
733 	int error, i, sig;
734 
735 	/*
736 	 * Do it right now so we can log either proper call of abort2(), or
737 	 * note, that invalid argument was passed. 512 is big enough to
738 	 * handle 16 arguments' descriptions with additional comments.
739 	 */
740 	sb = sbuf_new(NULL, NULL, 512, SBUF_FIXEDLEN);
741 	sbuf_clear(sb);
742 	sbuf_printf(sb, "%s(pid %d uid %d) aborted: ",
743 	    p->p_comm, p->p_pid, td->td_ucred->cr_uid);
744 	/*
745 	 * Since we can't return from abort2(), send SIGKILL in cases, where
746 	 * abort2() was called improperly
747 	 */
748 	sig = SIGKILL;
749 	/* Prevent from DoSes from user-space. */
750 	if (uap->nargs < 0 || uap->nargs > 16)
751 		goto out;
752 	if (uap->nargs > 0) {
753 		if (uap->args == NULL)
754 			goto out;
755 		error = copyin(uap->args, uargs, uap->nargs * sizeof(void *));
756 		if (error != 0)
757 			goto out;
758 	}
759 	/*
760 	 * Limit size of 'reason' string to 128. Will fit even when
761 	 * maximal number of arguments was chosen to be logged.
762 	 */
763 	if (uap->why != NULL) {
764 		error = sbuf_copyin(sb, uap->why, 128);
765 		if (error < 0)
766 			goto out;
767 	} else {
768 		sbuf_printf(sb, "(null)");
769 	}
770 	if (uap->nargs > 0) {
771 		sbuf_printf(sb, "(");
772 		for (i = 0;i < uap->nargs; i++)
773 			sbuf_printf(sb, "%s%p", i == 0 ? "" : ", ", uargs[i]);
774 		sbuf_printf(sb, ")");
775 	}
776 	/*
777 	 * Final stage: arguments were proper, string has been
778 	 * successfully copied from userspace, and copying pointers
779 	 * from user-space succeed.
780 	 */
781 	sig = SIGABRT;
782 out:
783 	if (sig == SIGKILL) {
784 		sbuf_trim(sb);
785 		sbuf_printf(sb, " (Reason text inaccessible)");
786 	}
787 	sbuf_cat(sb, "\n");
788 	sbuf_finish(sb);
789 	log(LOG_INFO, "%s", sbuf_data(sb));
790 	sbuf_delete(sb);
791 	exit1(td, 0, sig);
792 	return (0);
793 }
794 
795 #ifdef COMPAT_43
796 /*
797  * The dirty work is handled by kern_wait().
798  */
799 int
800 owait(struct thread *td, struct owait_args *uap __unused)
801 {
802 	int error, status;
803 
804 	error = kern_wait(td, WAIT_ANY, &status, 0, NULL);
805 	if (error == 0)
806 		td->td_retval[1] = status;
807 	return (error);
808 }
809 #endif /* COMPAT_43 */
810 
811 /*
812  * The dirty work is handled by kern_wait().
813  */
814 int
815 sys_wait4(struct thread *td, struct wait4_args *uap)
816 {
817 	struct rusage ru, *rup;
818 	int error, status;
819 
820 	if (uap->rusage != NULL)
821 		rup = &ru;
822 	else
823 		rup = NULL;
824 	error = kern_wait(td, uap->pid, &status, uap->options, rup);
825 	if (uap->status != NULL && error == 0 && td->td_retval[0] != 0)
826 		error = copyout(&status, uap->status, sizeof(status));
827 	if (uap->rusage != NULL && error == 0 && td->td_retval[0] != 0)
828 		error = copyout(&ru, uap->rusage, sizeof(struct rusage));
829 	return (error);
830 }
831 
832 int
833 sys_wait6(struct thread *td, struct wait6_args *uap)
834 {
835 	struct __wrusage wru, *wrup;
836 	siginfo_t si, *sip;
837 	idtype_t idtype;
838 	id_t id;
839 	int error, status;
840 
841 	idtype = uap->idtype;
842 	id = uap->id;
843 
844 	if (uap->wrusage != NULL)
845 		wrup = &wru;
846 	else
847 		wrup = NULL;
848 
849 	if (uap->info != NULL) {
850 		sip = &si;
851 		bzero(sip, sizeof(*sip));
852 	} else
853 		sip = NULL;
854 
855 	/*
856 	 *  We expect all callers of wait6() to know about WEXITED and
857 	 *  WTRAPPED.
858 	 */
859 	error = kern_wait6(td, idtype, id, &status, uap->options, wrup, sip);
860 
861 	if (uap->status != NULL && error == 0 && td->td_retval[0] != 0)
862 		error = copyout(&status, uap->status, sizeof(status));
863 	if (uap->wrusage != NULL && error == 0 && td->td_retval[0] != 0)
864 		error = copyout(&wru, uap->wrusage, sizeof(wru));
865 	if (uap->info != NULL && error == 0)
866 		error = copyout(&si, uap->info, sizeof(si));
867 	return (error);
868 }
869 
870 /*
871  * Reap the remains of a zombie process and optionally return status and
872  * rusage.  Asserts and will release both the proctree_lock and the process
873  * lock as part of its work.
874  */
875 void
876 proc_reap(struct thread *td, struct proc *p, int *status, int options)
877 {
878 	struct proc *q, *t;
879 
880 	sx_assert(&proctree_lock, SA_XLOCKED);
881 	PROC_LOCK_ASSERT(p, MA_OWNED);
882 	KASSERT(p->p_state == PRS_ZOMBIE, ("proc_reap: !PRS_ZOMBIE"));
883 
884 	mtx_spin_wait_unlocked(&p->p_slock);
885 
886 	q = td->td_proc;
887 
888 	if (status)
889 		*status = KW_EXITCODE(p->p_xexit, p->p_xsig);
890 	if (options & WNOWAIT) {
891 		/*
892 		 *  Only poll, returning the status.  Caller does not wish to
893 		 * release the proc struct just yet.
894 		 */
895 		PROC_UNLOCK(p);
896 		sx_xunlock(&proctree_lock);
897 		return;
898 	}
899 
900 	PROC_LOCK(q);
901 	sigqueue_take(p->p_ksi);
902 	PROC_UNLOCK(q);
903 
904 	/*
905 	 * If we got the child via a ptrace 'attach', we need to give it back
906 	 * to the old parent.
907 	 */
908 	if (p->p_oppid != p->p_pptr->p_pid) {
909 		PROC_UNLOCK(p);
910 		t = proc_realparent(p);
911 		PROC_LOCK(t);
912 		PROC_LOCK(p);
913 		CTR2(KTR_PTRACE,
914 		    "wait: traced child %d moved back to parent %d", p->p_pid,
915 		    t->p_pid);
916 		proc_reparent(p, t, false);
917 		PROC_UNLOCK(p);
918 		pksignal(t, SIGCHLD, p->p_ksi);
919 		wakeup(t);
920 		cv_broadcast(&p->p_pwait);
921 		PROC_UNLOCK(t);
922 		sx_xunlock(&proctree_lock);
923 		return;
924 	}
925 	PROC_UNLOCK(p);
926 
927 	/*
928 	 * Remove other references to this process to ensure we have an
929 	 * exclusive reference.
930 	 */
931 	sx_xlock(PIDHASHLOCK(p->p_pid));
932 	LIST_REMOVE(p, p_hash);
933 	sx_xunlock(PIDHASHLOCK(p->p_pid));
934 	LIST_REMOVE(p, p_sibling);
935 	reaper_abandon_children(p, true);
936 	reaper_clear(p);
937 	PROC_LOCK(p);
938 	proc_clear_orphan(p);
939 	PROC_UNLOCK(p);
940 	leavepgrp(p);
941 	if (p->p_procdesc != NULL)
942 		procdesc_reap(p);
943 	sx_xunlock(&proctree_lock);
944 
945 	proc_id_clear(PROC_ID_PID, p->p_pid);
946 
947 	PROC_LOCK(p);
948 	knlist_detach(p->p_klist);
949 	p->p_klist = NULL;
950 	PROC_UNLOCK(p);
951 
952 	/*
953 	 * Removal from allproc list and process group list paired with
954 	 * PROC_LOCK which was executed during that time should guarantee
955 	 * nothing can reach this process anymore. As such further locking
956 	 * is unnecessary.
957 	 */
958 	p->p_xexit = p->p_xsig = 0;		/* XXX: why? */
959 
960 	PROC_LOCK(q);
961 	ruadd(&q->p_stats->p_cru, &q->p_crux, &p->p_ru, &p->p_rux);
962 	PROC_UNLOCK(q);
963 
964 	/*
965 	 * Decrement the count of procs running with this uid.
966 	 */
967 	(void)chgproccnt(p->p_ucred->cr_ruidinfo, -1, 0);
968 
969 	/*
970 	 * Destroy resource accounting information associated with the process.
971 	 */
972 #ifdef RACCT
973 	if (racct_enable) {
974 		PROC_LOCK(p);
975 		racct_sub(p, RACCT_NPROC, 1);
976 		PROC_UNLOCK(p);
977 	}
978 #endif
979 	racct_proc_exit(p);
980 
981 	/*
982 	 * Free credentials, arguments, and sigacts.
983 	 */
984 	proc_unset_cred(p);
985 	pargs_drop(p->p_args);
986 	p->p_args = NULL;
987 	sigacts_free(p->p_sigacts);
988 	p->p_sigacts = NULL;
989 
990 	/*
991 	 * Do any thread-system specific cleanups.
992 	 */
993 	thread_wait(p);
994 
995 	/*
996 	 * Give vm and machine-dependent layer a chance to free anything that
997 	 * cpu_exit couldn't release while still running in process context.
998 	 */
999 	vm_waitproc(p);
1000 #ifdef MAC
1001 	mac_proc_destroy(p);
1002 #endif
1003 
1004 	KASSERT(FIRST_THREAD_IN_PROC(p),
1005 	    ("proc_reap: no residual thread!"));
1006 	uma_zfree(proc_zone, p);
1007 	atomic_add_int(&nprocs, -1);
1008 }
1009 
1010 static int
1011 proc_to_reap(struct thread *td, struct proc *p, idtype_t idtype, id_t id,
1012     int *status, int options, struct __wrusage *wrusage, siginfo_t *siginfo,
1013     int check_only)
1014 {
1015 	struct rusage *rup;
1016 
1017 	sx_assert(&proctree_lock, SA_XLOCKED);
1018 
1019 	PROC_LOCK(p);
1020 
1021 	switch (idtype) {
1022 	case P_ALL:
1023 		if (p->p_procdesc == NULL ||
1024 		   (p->p_pptr == td->td_proc &&
1025 		   (p->p_flag & P_TRACED) != 0)) {
1026 			break;
1027 		}
1028 
1029 		PROC_UNLOCK(p);
1030 		return (0);
1031 	case P_PID:
1032 		if (p->p_pid != (pid_t)id) {
1033 			PROC_UNLOCK(p);
1034 			return (0);
1035 		}
1036 		break;
1037 	case P_PGID:
1038 		if (p->p_pgid != (pid_t)id) {
1039 			PROC_UNLOCK(p);
1040 			return (0);
1041 		}
1042 		break;
1043 	case P_SID:
1044 		if (p->p_session->s_sid != (pid_t)id) {
1045 			PROC_UNLOCK(p);
1046 			return (0);
1047 		}
1048 		break;
1049 	case P_UID:
1050 		if (p->p_ucred->cr_uid != (uid_t)id) {
1051 			PROC_UNLOCK(p);
1052 			return (0);
1053 		}
1054 		break;
1055 	case P_GID:
1056 		if (p->p_ucred->cr_gid != (gid_t)id) {
1057 			PROC_UNLOCK(p);
1058 			return (0);
1059 		}
1060 		break;
1061 	case P_JAILID:
1062 		if (p->p_ucred->cr_prison->pr_id != (int)id) {
1063 			PROC_UNLOCK(p);
1064 			return (0);
1065 		}
1066 		break;
1067 	/*
1068 	 * It seems that the thread structures get zeroed out
1069 	 * at process exit.  This makes it impossible to
1070 	 * support P_SETID, P_CID or P_CPUID.
1071 	 */
1072 	default:
1073 		PROC_UNLOCK(p);
1074 		return (0);
1075 	}
1076 
1077 	if (p_canwait(td, p)) {
1078 		PROC_UNLOCK(p);
1079 		return (0);
1080 	}
1081 
1082 	if (((options & WEXITED) == 0) && (p->p_state == PRS_ZOMBIE)) {
1083 		PROC_UNLOCK(p);
1084 		return (0);
1085 	}
1086 
1087 	/*
1088 	 * This special case handles a kthread spawned by linux_clone
1089 	 * (see linux_misc.c).  The linux_wait4 and linux_waitpid
1090 	 * functions need to be able to distinguish between waiting
1091 	 * on a process and waiting on a thread.  It is a thread if
1092 	 * p_sigparent is not SIGCHLD, and the WLINUXCLONE option
1093 	 * signifies we want to wait for threads and not processes.
1094 	 */
1095 	if ((p->p_sigparent != SIGCHLD) ^
1096 	    ((options & WLINUXCLONE) != 0)) {
1097 		PROC_UNLOCK(p);
1098 		return (0);
1099 	}
1100 
1101 	if (siginfo != NULL) {
1102 		bzero(siginfo, sizeof(*siginfo));
1103 		siginfo->si_errno = 0;
1104 
1105 		/*
1106 		 * SUSv4 requires that the si_signo value is always
1107 		 * SIGCHLD. Obey it despite the rfork(2) interface
1108 		 * allows to request other signal for child exit
1109 		 * notification.
1110 		 */
1111 		siginfo->si_signo = SIGCHLD;
1112 
1113 		/*
1114 		 *  This is still a rough estimate.  We will fix the
1115 		 *  cases TRAPPED, STOPPED, and CONTINUED later.
1116 		 */
1117 		if (WCOREDUMP(p->p_xsig)) {
1118 			siginfo->si_code = CLD_DUMPED;
1119 			siginfo->si_status = WTERMSIG(p->p_xsig);
1120 		} else if (WIFSIGNALED(p->p_xsig)) {
1121 			siginfo->si_code = CLD_KILLED;
1122 			siginfo->si_status = WTERMSIG(p->p_xsig);
1123 		} else {
1124 			siginfo->si_code = CLD_EXITED;
1125 			siginfo->si_status = p->p_xexit;
1126 		}
1127 
1128 		siginfo->si_pid = p->p_pid;
1129 		siginfo->si_uid = p->p_ucred->cr_uid;
1130 
1131 		/*
1132 		 * The si_addr field would be useful additional
1133 		 * detail, but apparently the PC value may be lost
1134 		 * when we reach this point.  bzero() above sets
1135 		 * siginfo->si_addr to NULL.
1136 		 */
1137 	}
1138 
1139 	/*
1140 	 * There should be no reason to limit resources usage info to
1141 	 * exited processes only.  A snapshot about any resources used
1142 	 * by a stopped process may be exactly what is needed.
1143 	 */
1144 	if (wrusage != NULL) {
1145 		rup = &wrusage->wru_self;
1146 		*rup = p->p_ru;
1147 		PROC_STATLOCK(p);
1148 		calcru(p, &rup->ru_utime, &rup->ru_stime);
1149 		PROC_STATUNLOCK(p);
1150 
1151 		rup = &wrusage->wru_children;
1152 		*rup = p->p_stats->p_cru;
1153 		calccru(p, &rup->ru_utime, &rup->ru_stime);
1154 	}
1155 
1156 	if (p->p_state == PRS_ZOMBIE && !check_only) {
1157 		proc_reap(td, p, status, options);
1158 		return (-1);
1159 	}
1160 	return (1);
1161 }
1162 
1163 int
1164 kern_wait(struct thread *td, pid_t pid, int *status, int options,
1165     struct rusage *rusage)
1166 {
1167 	struct __wrusage wru, *wrup;
1168 	idtype_t idtype;
1169 	id_t id;
1170 	int ret;
1171 
1172 	/*
1173 	 * Translate the special pid values into the (idtype, pid)
1174 	 * pair for kern_wait6.  The WAIT_MYPGRP case is handled by
1175 	 * kern_wait6() on its own.
1176 	 */
1177 	if (pid == WAIT_ANY) {
1178 		idtype = P_ALL;
1179 		id = 0;
1180 	} else if (pid < 0) {
1181 		idtype = P_PGID;
1182 		id = (id_t)-pid;
1183 	} else {
1184 		idtype = P_PID;
1185 		id = (id_t)pid;
1186 	}
1187 
1188 	if (rusage != NULL)
1189 		wrup = &wru;
1190 	else
1191 		wrup = NULL;
1192 
1193 	/*
1194 	 * For backward compatibility we implicitly add flags WEXITED
1195 	 * and WTRAPPED here.
1196 	 */
1197 	options |= WEXITED | WTRAPPED;
1198 	ret = kern_wait6(td, idtype, id, status, options, wrup, NULL);
1199 	if (rusage != NULL)
1200 		*rusage = wru.wru_self;
1201 	return (ret);
1202 }
1203 
1204 static void
1205 report_alive_proc(struct thread *td, struct proc *p, siginfo_t *siginfo,
1206     int *status, int options, int si_code)
1207 {
1208 	bool cont;
1209 
1210 	PROC_LOCK_ASSERT(p, MA_OWNED);
1211 	sx_assert(&proctree_lock, SA_XLOCKED);
1212 	MPASS(si_code == CLD_TRAPPED || si_code == CLD_STOPPED ||
1213 	    si_code == CLD_CONTINUED);
1214 
1215 	cont = si_code == CLD_CONTINUED;
1216 	if ((options & WNOWAIT) == 0) {
1217 		if (cont)
1218 			p->p_flag &= ~P_CONTINUED;
1219 		else
1220 			p->p_flag |= P_WAITED;
1221 		if (kern_wait_dequeue_sigchld &&
1222 		    (td->td_proc->p_sysent->sv_flags & SV_SIG_WAITNDQ) == 0) {
1223 			PROC_LOCK(td->td_proc);
1224 			sigqueue_take(p->p_ksi);
1225 			PROC_UNLOCK(td->td_proc);
1226 		}
1227 	}
1228 	sx_xunlock(&proctree_lock);
1229 	if (siginfo != NULL) {
1230 		siginfo->si_code = si_code;
1231 		siginfo->si_status = cont ? SIGCONT : p->p_xsig;
1232 	}
1233 	if (status != NULL)
1234 		*status = cont ? SIGCONT : W_STOPCODE(p->p_xsig);
1235 	PROC_UNLOCK(p);
1236 	td->td_retval[0] = p->p_pid;
1237 }
1238 
1239 int
1240 kern_wait6(struct thread *td, idtype_t idtype, id_t id, int *status,
1241     int options, struct __wrusage *wrusage, siginfo_t *siginfo)
1242 {
1243 	struct proc *p, *q;
1244 	pid_t pid;
1245 	int error, nfound, ret;
1246 	bool report;
1247 
1248 	AUDIT_ARG_VALUE((int)idtype);	/* XXX - This is likely wrong! */
1249 	AUDIT_ARG_PID((pid_t)id);	/* XXX - This may be wrong! */
1250 	AUDIT_ARG_VALUE(options);
1251 
1252 	q = td->td_proc;
1253 
1254 	if ((pid_t)id == WAIT_MYPGRP && (idtype == P_PID || idtype == P_PGID)) {
1255 		PROC_LOCK(q);
1256 		id = (id_t)q->p_pgid;
1257 		PROC_UNLOCK(q);
1258 		idtype = P_PGID;
1259 	}
1260 
1261 	/* If we don't know the option, just return. */
1262 	if ((options & ~(WUNTRACED | WNOHANG | WCONTINUED | WNOWAIT |
1263 	    WEXITED | WTRAPPED | WLINUXCLONE)) != 0)
1264 		return (EINVAL);
1265 	if ((options & (WEXITED | WUNTRACED | WCONTINUED | WTRAPPED)) == 0) {
1266 		/*
1267 		 * We will be unable to find any matching processes,
1268 		 * because there are no known events to look for.
1269 		 * Prefer to return error instead of blocking
1270 		 * indefinitely.
1271 		 */
1272 		return (EINVAL);
1273 	}
1274 
1275 loop:
1276 	if (q->p_flag & P_STATCHILD) {
1277 		PROC_LOCK(q);
1278 		q->p_flag &= ~P_STATCHILD;
1279 		PROC_UNLOCK(q);
1280 	}
1281 	sx_xlock(&proctree_lock);
1282 loop_locked:
1283 	nfound = 0;
1284 	LIST_FOREACH(p, &q->p_children, p_sibling) {
1285 		pid = p->p_pid;
1286 		ret = proc_to_reap(td, p, idtype, id, status, options,
1287 		    wrusage, siginfo, 0);
1288 		if (ret == 0)
1289 			continue;
1290 		else if (ret != 1) {
1291 			td->td_retval[0] = pid;
1292 			return (0);
1293 		}
1294 
1295 		nfound++;
1296 		PROC_LOCK_ASSERT(p, MA_OWNED);
1297 
1298 		if ((options & WTRAPPED) != 0 &&
1299 		    (p->p_flag & P_TRACED) != 0) {
1300 			PROC_SLOCK(p);
1301 			report =
1302 			    ((p->p_flag & (P_STOPPED_TRACE | P_STOPPED_SIG)) &&
1303 			    p->p_suspcount == p->p_numthreads &&
1304 			    (p->p_flag & P_WAITED) == 0);
1305 			PROC_SUNLOCK(p);
1306 			if (report) {
1307 			CTR4(KTR_PTRACE,
1308 			    "wait: returning trapped pid %d status %#x "
1309 			    "(xstat %d) xthread %d",
1310 			    p->p_pid, W_STOPCODE(p->p_xsig), p->p_xsig,
1311 			    p->p_xthread != NULL ?
1312 			    p->p_xthread->td_tid : -1);
1313 				report_alive_proc(td, p, siginfo, status,
1314 				    options, CLD_TRAPPED);
1315 				return (0);
1316 			}
1317 		}
1318 		if ((options & WUNTRACED) != 0 &&
1319 		    (p->p_flag & P_STOPPED_SIG) != 0) {
1320 			PROC_SLOCK(p);
1321 			report = (p->p_suspcount == p->p_numthreads &&
1322 			    ((p->p_flag & P_WAITED) == 0));
1323 			PROC_SUNLOCK(p);
1324 			if (report) {
1325 				report_alive_proc(td, p, siginfo, status,
1326 				    options, CLD_STOPPED);
1327 				return (0);
1328 			}
1329 		}
1330 		if ((options & WCONTINUED) != 0 &&
1331 		    (p->p_flag & P_CONTINUED) != 0) {
1332 			report_alive_proc(td, p, siginfo, status, options,
1333 			    CLD_CONTINUED);
1334 			return (0);
1335 		}
1336 		PROC_UNLOCK(p);
1337 	}
1338 
1339 	/*
1340 	 * Look in the orphans list too, to allow the parent to
1341 	 * collect it's child exit status even if child is being
1342 	 * debugged.
1343 	 *
1344 	 * Debugger detaches from the parent upon successful
1345 	 * switch-over from parent to child.  At this point due to
1346 	 * re-parenting the parent loses the child to debugger and a
1347 	 * wait4(2) call would report that it has no children to wait
1348 	 * for.  By maintaining a list of orphans we allow the parent
1349 	 * to successfully wait until the child becomes a zombie.
1350 	 */
1351 	if (nfound == 0) {
1352 		LIST_FOREACH(p, &q->p_orphans, p_orphan) {
1353 			ret = proc_to_reap(td, p, idtype, id, NULL, options,
1354 			    NULL, NULL, 1);
1355 			if (ret != 0) {
1356 				KASSERT(ret != -1, ("reaped an orphan (pid %d)",
1357 				    (int)td->td_retval[0]));
1358 				PROC_UNLOCK(p);
1359 				nfound++;
1360 				break;
1361 			}
1362 		}
1363 	}
1364 	if (nfound == 0) {
1365 		sx_xunlock(&proctree_lock);
1366 		return (ECHILD);
1367 	}
1368 	if (options & WNOHANG) {
1369 		sx_xunlock(&proctree_lock);
1370 		td->td_retval[0] = 0;
1371 		return (0);
1372 	}
1373 	PROC_LOCK(q);
1374 	if (q->p_flag & P_STATCHILD) {
1375 		q->p_flag &= ~P_STATCHILD;
1376 		PROC_UNLOCK(q);
1377 		goto loop_locked;
1378 	}
1379 	sx_xunlock(&proctree_lock);
1380 	error = msleep(q, &q->p_mtx, PWAIT | PCATCH | PDROP, "wait", 0);
1381 	if (error)
1382 		return (error);
1383 	goto loop;
1384 }
1385 
1386 void
1387 proc_add_orphan(struct proc *child, struct proc *parent)
1388 {
1389 
1390 	sx_assert(&proctree_lock, SX_XLOCKED);
1391 	KASSERT((child->p_flag & P_TRACED) != 0,
1392 	    ("proc_add_orphan: not traced"));
1393 
1394 	if (LIST_EMPTY(&parent->p_orphans)) {
1395 		child->p_treeflag |= P_TREE_FIRST_ORPHAN;
1396 		LIST_INSERT_HEAD(&parent->p_orphans, child, p_orphan);
1397 	} else {
1398 		LIST_INSERT_AFTER(LIST_FIRST(&parent->p_orphans),
1399 		    child, p_orphan);
1400 	}
1401 	child->p_treeflag |= P_TREE_ORPHANED;
1402 }
1403 
1404 /*
1405  * Make process 'parent' the new parent of process 'child'.
1406  * Must be called with an exclusive hold of proctree lock.
1407  */
1408 void
1409 proc_reparent(struct proc *child, struct proc *parent, bool set_oppid)
1410 {
1411 
1412 	sx_assert(&proctree_lock, SX_XLOCKED);
1413 	PROC_LOCK_ASSERT(child, MA_OWNED);
1414 	if (child->p_pptr == parent)
1415 		return;
1416 
1417 	PROC_LOCK(child->p_pptr);
1418 	sigqueue_take(child->p_ksi);
1419 	PROC_UNLOCK(child->p_pptr);
1420 	LIST_REMOVE(child, p_sibling);
1421 	LIST_INSERT_HEAD(&parent->p_children, child, p_sibling);
1422 
1423 	proc_clear_orphan(child);
1424 	if ((child->p_flag & P_TRACED) != 0) {
1425 		proc_add_orphan(child, child->p_pptr);
1426 	}
1427 
1428 	child->p_pptr = parent;
1429 	if (set_oppid)
1430 		child->p_oppid = parent->p_pid;
1431 }
1432