xref: /freebsd/sys/kern/kern_exit.c (revision 1f88aa09417f1cfb3929fd37531b1ab51213c2d6)
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 
232 	p = td->td_proc;
233 	/*
234 	 * XXX in case we're rebooting we just let init die in order to
235 	 * work around an unsolved stack overflow seen very late during
236 	 * shutdown on sparc64 when the gmirror worker process exists.
237 	 * XXX what to do now that sparc64 is gone... remove if?
238 	 */
239 	if (p == initproc && rebooting == 0) {
240 		printf("init died (signal %d, exit %d)\n", signo, rval);
241 		panic("Going nowhere without my init!");
242 	}
243 
244 	/*
245 	 * Deref SU mp, since the thread does not return to userspace.
246 	 */
247 	td_softdep_cleanup(td);
248 
249 	/*
250 	 * MUST abort all other threads before proceeding past here.
251 	 */
252 	PROC_LOCK(p);
253 	/*
254 	 * First check if some other thread or external request got
255 	 * here before us.  If so, act appropriately: exit or suspend.
256 	 * We must ensure that stop requests are handled before we set
257 	 * P_WEXIT.
258 	 */
259 	thread_suspend_check(0);
260 	while (p->p_flag & P_HADTHREADS) {
261 		/*
262 		 * Kill off the other threads. This requires
263 		 * some co-operation from other parts of the kernel
264 		 * so it may not be instantaneous.  With this state set
265 		 * any thread entering the kernel from userspace will
266 		 * thread_exit() in trap().  Any thread attempting to
267 		 * sleep will return immediately with EINTR or EWOULDBLOCK
268 		 * which will hopefully force them to back out to userland
269 		 * freeing resources as they go.  Any thread attempting
270 		 * to return to userland will thread_exit() from userret().
271 		 * thread_exit() will unsuspend us when the last of the
272 		 * other threads exits.
273 		 * If there is already a thread singler after resumption,
274 		 * calling thread_single will fail; in that case, we just
275 		 * re-check all suspension request, the thread should
276 		 * either be suspended there or exit.
277 		 */
278 		if (!thread_single(p, SINGLE_EXIT))
279 			/*
280 			 * All other activity in this process is now
281 			 * stopped.  Threading support has been turned
282 			 * off.
283 			 */
284 			break;
285 		/*
286 		 * Recheck for new stop or suspend requests which
287 		 * might appear while process lock was dropped in
288 		 * thread_single().
289 		 */
290 		thread_suspend_check(0);
291 	}
292 	KASSERT(p->p_numthreads == 1,
293 	    ("exit1: proc %p exiting with %d threads", p, p->p_numthreads));
294 	racct_sub(p, RACCT_NTHR, 1);
295 
296 	/* Let event handler change exit status */
297 	p->p_xexit = rval;
298 	p->p_xsig = signo;
299 
300 	/*
301 	 * Ignore any pending request to stop due to a stop signal.
302 	 * Once P_WEXIT is set, future requests will be ignored as
303 	 * well.
304 	 */
305 	p->p_flag &= ~P_STOPPED_SIG;
306 	KASSERT(!P_SHOULDSTOP(p), ("exiting process is stopped"));
307 
308 	/* Note that we are exiting. */
309 	p->p_flag |= P_WEXIT;
310 
311 	/*
312 	 * Wait for any processes that have a hold on our vmspace to
313 	 * release their reference.
314 	 */
315 	while (p->p_lock > 0)
316 		msleep(&p->p_lock, &p->p_mtx, PWAIT, "exithold", 0);
317 
318 	PROC_UNLOCK(p);
319 	/* Drain the limit callout while we don't have the proc locked */
320 	callout_drain(&p->p_limco);
321 
322 #ifdef AUDIT
323 	/*
324 	 * The Sun BSM exit token contains two components: an exit status as
325 	 * passed to exit(), and a return value to indicate what sort of exit
326 	 * it was.  The exit status is WEXITSTATUS(rv), but it's not clear
327 	 * what the return value is.
328 	 */
329 	AUDIT_ARG_EXIT(rval, 0);
330 	AUDIT_SYSCALL_EXIT(0, td);
331 #endif
332 
333 	/* Are we a task leader with peers? */
334 	if (p->p_peers != NULL && p == p->p_leader) {
335 		mtx_lock(&ppeers_lock);
336 		q = p->p_peers;
337 		while (q != NULL) {
338 			PROC_LOCK(q);
339 			kern_psignal(q, SIGKILL);
340 			PROC_UNLOCK(q);
341 			q = q->p_peers;
342 		}
343 		while (p->p_peers != NULL)
344 			msleep(p, &ppeers_lock, PWAIT, "exit1", 0);
345 		mtx_unlock(&ppeers_lock);
346 	}
347 
348 	itimers_exit(p);
349 
350 	/*
351 	 * Check if any loadable modules need anything done at process exit.
352 	 * E.g. SYSV IPC stuff.
353 	 * Event handler could change exit status.
354 	 * XXX what if one of these generates an error?
355 	 */
356 	EVENTHANDLER_DIRECT_INVOKE(process_exit, p);
357 
358 	/*
359 	 * If parent is waiting for us to exit or exec,
360 	 * P_PPWAIT is set; we will wakeup the parent below.
361 	 */
362 	PROC_LOCK(p);
363 	stopprofclock(p);
364 	p->p_ptevents = 0;
365 
366 	/*
367 	 * Stop the real interval timer.  If the handler is currently
368 	 * executing, prevent it from rearming itself and let it finish.
369 	 */
370 	if (timevalisset(&p->p_realtimer.it_value) &&
371 	    _callout_stop_safe(&p->p_itcallout, CS_EXECUTING, NULL) == 0) {
372 		timevalclear(&p->p_realtimer.it_interval);
373 		msleep(&p->p_itcallout, &p->p_mtx, PWAIT, "ritwait", 0);
374 		KASSERT(!timevalisset(&p->p_realtimer.it_value),
375 		    ("realtime timer is still armed"));
376 	}
377 
378 	PROC_UNLOCK(p);
379 
380 	if (p->p_sysent->sv_onexit != NULL)
381 		p->p_sysent->sv_onexit(p);
382 	seltdfini(td);
383 
384 	/*
385 	 * Reset any sigio structures pointing to us as a result of
386 	 * F_SETOWN with our pid.  The P_WEXIT flag interlocks with fsetown().
387 	 */
388 	funsetownlst(&p->p_sigiolst);
389 
390 	/*
391 	 * Close open files and release open-file table.
392 	 * This may block!
393 	 */
394 	pdescfree(td);
395 	fdescfree(td);
396 
397 	/*
398 	 * If this thread tickled GEOM, we need to wait for the giggling to
399 	 * stop before we return to userland
400 	 */
401 	if (td->td_pflags & TDP_GEOM)
402 		g_waitidle();
403 
404 	/*
405 	 * Remove ourself from our leader's peer list and wake our leader.
406 	 */
407 	if (p->p_leader->p_peers != NULL) {
408 		mtx_lock(&ppeers_lock);
409 		if (p->p_leader->p_peers != NULL) {
410 			q = p->p_leader;
411 			while (q->p_peers != p)
412 				q = q->p_peers;
413 			q->p_peers = p->p_peers;
414 			wakeup(p->p_leader);
415 		}
416 		mtx_unlock(&ppeers_lock);
417 	}
418 
419 	vmspace_exit(td);
420 	(void)acct_process(td);
421 
422 #ifdef KTRACE
423 	ktrprocexit(td);
424 #endif
425 	/*
426 	 * Release reference to text vnode
427 	 */
428 	if (p->p_textvp != NULL) {
429 		vrele(p->p_textvp);
430 		p->p_textvp = NULL;
431 	}
432 
433 	/*
434 	 * Release our limits structure.
435 	 */
436 	lim_free(p->p_limit);
437 	p->p_limit = NULL;
438 
439 	tidhash_remove(td);
440 
441 	/*
442 	 * Call machine-dependent code to release any
443 	 * machine-dependent resources other than the address space.
444 	 * The address space is released by "vmspace_exitfree(p)" in
445 	 * vm_waitproc().
446 	 */
447 	cpu_exit(td);
448 
449 	WITNESS_WARN(WARN_PANIC, NULL, "process (pid %d) exiting", p->p_pid);
450 
451 	/*
452 	 * Remove from allproc. It still sits in the hash.
453 	 */
454 	sx_xlock(&allproc_lock);
455 	LIST_REMOVE(p, p_list);
456 
457 #ifdef DDB
458 	/*
459 	 * Used by ddb's 'ps' command to find this process via the
460 	 * pidhash.
461 	 */
462 	p->p_list.le_prev = NULL;
463 #endif
464 	sx_xunlock(&allproc_lock);
465 
466 	sx_xlock(&proctree_lock);
467 	PROC_LOCK(p);
468 	p->p_flag &= ~(P_TRACED | P_PPWAIT | P_PPTRACE);
469 	PROC_UNLOCK(p);
470 
471 	/*
472 	 * killjobc() might drop and re-acquire proctree_lock to
473 	 * revoke control tty if exiting process was a session leader.
474 	 */
475 	killjobc();
476 
477 	/*
478 	 * Reparent all children processes:
479 	 * - traced ones to the original parent (or init if we are that parent)
480 	 * - the rest to init
481 	 */
482 	q = LIST_FIRST(&p->p_children);
483 	if (q != NULL)		/* only need this if any child is S_ZOMB */
484 		wakeup(q->p_reaper);
485 	for (; q != NULL; q = nq) {
486 		nq = LIST_NEXT(q, p_sibling);
487 		ksi = ksiginfo_alloc(TRUE);
488 		PROC_LOCK(q);
489 		q->p_sigparent = SIGCHLD;
490 
491 		if ((q->p_flag & P_TRACED) == 0) {
492 			proc_reparent(q, q->p_reaper, true);
493 			if (q->p_state == PRS_ZOMBIE) {
494 				/*
495 				 * Inform reaper about the reparented
496 				 * zombie, since wait(2) has something
497 				 * new to report.  Guarantee queueing
498 				 * of the SIGCHLD signal, similar to
499 				 * the _exit() behaviour, by providing
500 				 * our ksiginfo.  Ksi is freed by the
501 				 * signal delivery.
502 				 */
503 				if (q->p_ksi == NULL) {
504 					ksi1 = NULL;
505 				} else {
506 					ksiginfo_copy(q->p_ksi, ksi);
507 					ksi->ksi_flags |= KSI_INS;
508 					ksi1 = ksi;
509 					ksi = NULL;
510 				}
511 				PROC_LOCK(q->p_reaper);
512 				pksignal(q->p_reaper, SIGCHLD, ksi1);
513 				PROC_UNLOCK(q->p_reaper);
514 			} else if (q->p_pdeathsig > 0) {
515 				/*
516 				 * The child asked to received a signal
517 				 * when we exit.
518 				 */
519 				kern_psignal(q, q->p_pdeathsig);
520 			}
521 		} else {
522 			/*
523 			 * Traced processes are killed by default
524 			 * since their existence means someone is
525 			 * screwing up.
526 			 */
527 			t = proc_realparent(q);
528 			if (t == p) {
529 				proc_reparent(q, q->p_reaper, true);
530 			} else {
531 				PROC_LOCK(t);
532 				proc_reparent(q, t, true);
533 				PROC_UNLOCK(t);
534 			}
535 			/*
536 			 * Since q was found on our children list, the
537 			 * proc_reparent() call moved q to the orphan
538 			 * list due to present P_TRACED flag. Clear
539 			 * orphan link for q now while q is locked.
540 			 */
541 			proc_clear_orphan(q);
542 			q->p_flag &= ~P_TRACED;
543 			q->p_flag2 &= ~P2_PTRACE_FSTP;
544 			q->p_ptevents = 0;
545 			p->p_xthread = NULL;
546 			FOREACH_THREAD_IN_PROC(q, tdt) {
547 				tdt->td_dbgflags &= ~(TDB_SUSPEND | TDB_XSIG |
548 				    TDB_FSTP);
549 				tdt->td_xsig = 0;
550 			}
551 			if (kern_kill_on_dbg_exit) {
552 				q->p_flag &= ~P_STOPPED_TRACE;
553 				kern_psignal(q, SIGKILL);
554 			} else if ((q->p_flag & (P_STOPPED_TRACE |
555 			    P_STOPPED_SIG)) != 0) {
556 				sigqueue_delete_proc(q, SIGTRAP);
557 				ptrace_unsuspend(q);
558 			}
559 		}
560 		PROC_UNLOCK(q);
561 		if (ksi != NULL)
562 			ksiginfo_free(ksi);
563 	}
564 
565 	/*
566 	 * Also get rid of our orphans.
567 	 */
568 	while ((q = LIST_FIRST(&p->p_orphans)) != NULL) {
569 		PROC_LOCK(q);
570 		KASSERT(q->p_oppid == p->p_pid,
571 		    ("orphan %p of %p has unexpected oppid %d", q, p,
572 		    q->p_oppid));
573 		q->p_oppid = q->p_reaper->p_pid;
574 
575 		/*
576 		 * If we are the real parent of this process
577 		 * but it has been reparented to a debugger, then
578 		 * check if it asked for a signal when we exit.
579 		 */
580 		if (q->p_pdeathsig > 0)
581 			kern_psignal(q, q->p_pdeathsig);
582 		CTR2(KTR_PTRACE, "exit: pid %d, clearing orphan %d", p->p_pid,
583 		    q->p_pid);
584 		proc_clear_orphan(q);
585 		PROC_UNLOCK(q);
586 	}
587 
588 #ifdef KDTRACE_HOOKS
589 	if (SDT_PROBES_ENABLED()) {
590 		int reason = CLD_EXITED;
591 		if (WCOREDUMP(signo))
592 			reason = CLD_DUMPED;
593 		else if (WIFSIGNALED(signo))
594 			reason = CLD_KILLED;
595 		SDT_PROBE1(proc, , , exit, reason);
596 	}
597 #endif
598 
599 	/* Save exit status. */
600 	PROC_LOCK(p);
601 	p->p_xthread = td;
602 
603 	if (p->p_sysent->sv_ontdexit != NULL)
604 		p->p_sysent->sv_ontdexit(td);
605 
606 #ifdef KDTRACE_HOOKS
607 	/*
608 	 * Tell the DTrace fasttrap provider about the exit if it
609 	 * has declared an interest.
610 	 */
611 	if (dtrace_fasttrap_exit)
612 		dtrace_fasttrap_exit(p);
613 #endif
614 
615 	/*
616 	 * Notify interested parties of our demise.
617 	 */
618 	KNOTE_LOCKED(p->p_klist, NOTE_EXIT);
619 
620 	/*
621 	 * If this is a process with a descriptor, we may not need to deliver
622 	 * a signal to the parent.  proctree_lock is held over
623 	 * procdesc_exit() to serialize concurrent calls to close() and
624 	 * exit().
625 	 */
626 	signal_parent = 0;
627 	if (p->p_procdesc == NULL || procdesc_exit(p)) {
628 		/*
629 		 * Notify parent that we're gone.  If parent has the
630 		 * PS_NOCLDWAIT flag set, or if the handler is set to SIG_IGN,
631 		 * notify process 1 instead (and hope it will handle this
632 		 * situation).
633 		 */
634 		PROC_LOCK(p->p_pptr);
635 		mtx_lock(&p->p_pptr->p_sigacts->ps_mtx);
636 		if (p->p_pptr->p_sigacts->ps_flag &
637 		    (PS_NOCLDWAIT | PS_CLDSIGIGN)) {
638 			struct proc *pp;
639 
640 			mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx);
641 			pp = p->p_pptr;
642 			PROC_UNLOCK(pp);
643 			proc_reparent(p, p->p_reaper, true);
644 			p->p_sigparent = SIGCHLD;
645 			PROC_LOCK(p->p_pptr);
646 
647 			/*
648 			 * Notify parent, so in case he was wait(2)ing or
649 			 * executing waitpid(2) with our pid, he will
650 			 * continue.
651 			 */
652 			wakeup(pp);
653 		} else
654 			mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx);
655 
656 		if (p->p_pptr == p->p_reaper || p->p_pptr == initproc) {
657 			signal_parent = 1;
658 		} else if (p->p_sigparent != 0) {
659 			if (p->p_sigparent == SIGCHLD) {
660 				signal_parent = 1;
661 			} else { /* LINUX thread */
662 				signal_parent = 2;
663 			}
664 		}
665 	} else
666 		PROC_LOCK(p->p_pptr);
667 	sx_xunlock(&proctree_lock);
668 
669 	if (signal_parent == 1) {
670 		childproc_exited(p);
671 	} else if (signal_parent == 2) {
672 		kern_psignal(p->p_pptr, p->p_sigparent);
673 	}
674 
675 	/* Tell the prison that we are gone. */
676 	prison_proc_free(p->p_ucred->cr_prison);
677 
678 	/*
679 	 * The state PRS_ZOMBIE prevents other proesses from sending
680 	 * signal to the process, to avoid memory leak, we free memory
681 	 * for signal queue at the time when the state is set.
682 	 */
683 	sigqueue_flush(&p->p_sigqueue);
684 	sigqueue_flush(&td->td_sigqueue);
685 
686 	/*
687 	 * We have to wait until after acquiring all locks before
688 	 * changing p_state.  We need to avoid all possible context
689 	 * switches (including ones from blocking on a mutex) while
690 	 * marked as a zombie.  We also have to set the zombie state
691 	 * before we release the parent process' proc lock to avoid
692 	 * a lost wakeup.  So, we first call wakeup, then we grab the
693 	 * sched lock, update the state, and release the parent process'
694 	 * proc lock.
695 	 */
696 	wakeup(p->p_pptr);
697 	cv_broadcast(&p->p_pwait);
698 	sched_exit(p->p_pptr, td);
699 	PROC_SLOCK(p);
700 	p->p_state = PRS_ZOMBIE;
701 	PROC_UNLOCK(p->p_pptr);
702 
703 	/*
704 	 * Save our children's rusage information in our exit rusage.
705 	 */
706 	PROC_STATLOCK(p);
707 	ruadd(&p->p_ru, &p->p_rux, &p->p_stats->p_cru, &p->p_crux);
708 	PROC_STATUNLOCK(p);
709 
710 	/*
711 	 * Make sure the scheduler takes this thread out of its tables etc.
712 	 * This will also release this thread's reference to the ucred.
713 	 * Other thread parts to release include pcb bits and such.
714 	 */
715 	thread_exit();
716 }
717 
718 #ifndef _SYS_SYSPROTO_H_
719 struct abort2_args {
720 	char *why;
721 	int nargs;
722 	void **args;
723 };
724 #endif
725 
726 int
727 sys_abort2(struct thread *td, struct abort2_args *uap)
728 {
729 	struct proc *p = td->td_proc;
730 	struct sbuf *sb;
731 	void *uargs[16];
732 	int error, i, sig;
733 
734 	/*
735 	 * Do it right now so we can log either proper call of abort2(), or
736 	 * note, that invalid argument was passed. 512 is big enough to
737 	 * handle 16 arguments' descriptions with additional comments.
738 	 */
739 	sb = sbuf_new(NULL, NULL, 512, SBUF_FIXEDLEN);
740 	sbuf_clear(sb);
741 	sbuf_printf(sb, "%s(pid %d uid %d) aborted: ",
742 	    p->p_comm, p->p_pid, td->td_ucred->cr_uid);
743 	/*
744 	 * Since we can't return from abort2(), send SIGKILL in cases, where
745 	 * abort2() was called improperly
746 	 */
747 	sig = SIGKILL;
748 	/* Prevent from DoSes from user-space. */
749 	if (uap->nargs < 0 || uap->nargs > 16)
750 		goto out;
751 	if (uap->nargs > 0) {
752 		if (uap->args == NULL)
753 			goto out;
754 		error = copyin(uap->args, uargs, uap->nargs * sizeof(void *));
755 		if (error != 0)
756 			goto out;
757 	}
758 	/*
759 	 * Limit size of 'reason' string to 128. Will fit even when
760 	 * maximal number of arguments was chosen to be logged.
761 	 */
762 	if (uap->why != NULL) {
763 		error = sbuf_copyin(sb, uap->why, 128);
764 		if (error < 0)
765 			goto out;
766 	} else {
767 		sbuf_printf(sb, "(null)");
768 	}
769 	if (uap->nargs > 0) {
770 		sbuf_printf(sb, "(");
771 		for (i = 0;i < uap->nargs; i++)
772 			sbuf_printf(sb, "%s%p", i == 0 ? "" : ", ", uargs[i]);
773 		sbuf_printf(sb, ")");
774 	}
775 	/*
776 	 * Final stage: arguments were proper, string has been
777 	 * successfully copied from userspace, and copying pointers
778 	 * from user-space succeed.
779 	 */
780 	sig = SIGABRT;
781 out:
782 	if (sig == SIGKILL) {
783 		sbuf_trim(sb);
784 		sbuf_printf(sb, " (Reason text inaccessible)");
785 	}
786 	sbuf_cat(sb, "\n");
787 	sbuf_finish(sb);
788 	log(LOG_INFO, "%s", sbuf_data(sb));
789 	sbuf_delete(sb);
790 	exit1(td, 0, sig);
791 	return (0);
792 }
793 
794 #ifdef COMPAT_43
795 /*
796  * The dirty work is handled by kern_wait().
797  */
798 int
799 owait(struct thread *td, struct owait_args *uap __unused)
800 {
801 	int error, status;
802 
803 	error = kern_wait(td, WAIT_ANY, &status, 0, NULL);
804 	if (error == 0)
805 		td->td_retval[1] = status;
806 	return (error);
807 }
808 #endif /* COMPAT_43 */
809 
810 /*
811  * The dirty work is handled by kern_wait().
812  */
813 int
814 sys_wait4(struct thread *td, struct wait4_args *uap)
815 {
816 	struct rusage ru, *rup;
817 	int error, status;
818 
819 	if (uap->rusage != NULL)
820 		rup = &ru;
821 	else
822 		rup = NULL;
823 	error = kern_wait(td, uap->pid, &status, uap->options, rup);
824 	if (uap->status != NULL && error == 0 && td->td_retval[0] != 0)
825 		error = copyout(&status, uap->status, sizeof(status));
826 	if (uap->rusage != NULL && error == 0 && td->td_retval[0] != 0)
827 		error = copyout(&ru, uap->rusage, sizeof(struct rusage));
828 	return (error);
829 }
830 
831 int
832 sys_wait6(struct thread *td, struct wait6_args *uap)
833 {
834 	struct __wrusage wru, *wrup;
835 	siginfo_t si, *sip;
836 	idtype_t idtype;
837 	id_t id;
838 	int error, status;
839 
840 	idtype = uap->idtype;
841 	id = uap->id;
842 
843 	if (uap->wrusage != NULL)
844 		wrup = &wru;
845 	else
846 		wrup = NULL;
847 
848 	if (uap->info != NULL) {
849 		sip = &si;
850 		bzero(sip, sizeof(*sip));
851 	} else
852 		sip = NULL;
853 
854 	/*
855 	 *  We expect all callers of wait6() to know about WEXITED and
856 	 *  WTRAPPED.
857 	 */
858 	error = kern_wait6(td, idtype, id, &status, uap->options, wrup, sip);
859 
860 	if (uap->status != NULL && error == 0 && td->td_retval[0] != 0)
861 		error = copyout(&status, uap->status, sizeof(status));
862 	if (uap->wrusage != NULL && error == 0 && td->td_retval[0] != 0)
863 		error = copyout(&wru, uap->wrusage, sizeof(wru));
864 	if (uap->info != NULL && error == 0)
865 		error = copyout(&si, uap->info, sizeof(si));
866 	return (error);
867 }
868 
869 /*
870  * Reap the remains of a zombie process and optionally return status and
871  * rusage.  Asserts and will release both the proctree_lock and the process
872  * lock as part of its work.
873  */
874 void
875 proc_reap(struct thread *td, struct proc *p, int *status, int options)
876 {
877 	struct proc *q, *t;
878 
879 	sx_assert(&proctree_lock, SA_XLOCKED);
880 	PROC_LOCK_ASSERT(p, MA_OWNED);
881 	KASSERT(p->p_state == PRS_ZOMBIE, ("proc_reap: !PRS_ZOMBIE"));
882 
883 	mtx_spin_wait_unlocked(&p->p_slock);
884 
885 	q = td->td_proc;
886 
887 	if (status)
888 		*status = KW_EXITCODE(p->p_xexit, p->p_xsig);
889 	if (options & WNOWAIT) {
890 		/*
891 		 *  Only poll, returning the status.  Caller does not wish to
892 		 * release the proc struct just yet.
893 		 */
894 		PROC_UNLOCK(p);
895 		sx_xunlock(&proctree_lock);
896 		return;
897 	}
898 
899 	PROC_LOCK(q);
900 	sigqueue_take(p->p_ksi);
901 	PROC_UNLOCK(q);
902 
903 	/*
904 	 * If we got the child via a ptrace 'attach', we need to give it back
905 	 * to the old parent.
906 	 */
907 	if (p->p_oppid != p->p_pptr->p_pid) {
908 		PROC_UNLOCK(p);
909 		t = proc_realparent(p);
910 		PROC_LOCK(t);
911 		PROC_LOCK(p);
912 		CTR2(KTR_PTRACE,
913 		    "wait: traced child %d moved back to parent %d", p->p_pid,
914 		    t->p_pid);
915 		proc_reparent(p, t, false);
916 		PROC_UNLOCK(p);
917 		pksignal(t, SIGCHLD, p->p_ksi);
918 		wakeup(t);
919 		cv_broadcast(&p->p_pwait);
920 		PROC_UNLOCK(t);
921 		sx_xunlock(&proctree_lock);
922 		return;
923 	}
924 	PROC_UNLOCK(p);
925 
926 	/*
927 	 * Remove other references to this process to ensure we have an
928 	 * exclusive reference.
929 	 */
930 	sx_xlock(PIDHASHLOCK(p->p_pid));
931 	LIST_REMOVE(p, p_hash);
932 	sx_xunlock(PIDHASHLOCK(p->p_pid));
933 	LIST_REMOVE(p, p_sibling);
934 	reaper_abandon_children(p, true);
935 	reaper_clear(p);
936 	PROC_LOCK(p);
937 	proc_clear_orphan(p);
938 	PROC_UNLOCK(p);
939 	leavepgrp(p);
940 	if (p->p_procdesc != NULL)
941 		procdesc_reap(p);
942 	sx_xunlock(&proctree_lock);
943 
944 	proc_id_clear(PROC_ID_PID, p->p_pid);
945 
946 	PROC_LOCK(p);
947 	knlist_detach(p->p_klist);
948 	p->p_klist = NULL;
949 	PROC_UNLOCK(p);
950 
951 	/*
952 	 * Removal from allproc list and process group list paired with
953 	 * PROC_LOCK which was executed during that time should guarantee
954 	 * nothing can reach this process anymore. As such further locking
955 	 * is unnecessary.
956 	 */
957 	p->p_xexit = p->p_xsig = 0;		/* XXX: why? */
958 
959 	PROC_LOCK(q);
960 	ruadd(&q->p_stats->p_cru, &q->p_crux, &p->p_ru, &p->p_rux);
961 	PROC_UNLOCK(q);
962 
963 	/*
964 	 * Decrement the count of procs running with this uid.
965 	 */
966 	(void)chgproccnt(p->p_ucred->cr_ruidinfo, -1, 0);
967 
968 	/*
969 	 * Destroy resource accounting information associated with the process.
970 	 */
971 #ifdef RACCT
972 	if (racct_enable) {
973 		PROC_LOCK(p);
974 		racct_sub(p, RACCT_NPROC, 1);
975 		PROC_UNLOCK(p);
976 	}
977 #endif
978 	racct_proc_exit(p);
979 
980 	/*
981 	 * Free credentials, arguments, and sigacts.
982 	 */
983 	proc_unset_cred(p);
984 	pargs_drop(p->p_args);
985 	p->p_args = NULL;
986 	sigacts_free(p->p_sigacts);
987 	p->p_sigacts = NULL;
988 
989 	/*
990 	 * Do any thread-system specific cleanups.
991 	 */
992 	thread_wait(p);
993 
994 	/*
995 	 * Give vm and machine-dependent layer a chance to free anything that
996 	 * cpu_exit couldn't release while still running in process context.
997 	 */
998 	vm_waitproc(p);
999 #ifdef MAC
1000 	mac_proc_destroy(p);
1001 #endif
1002 
1003 	KASSERT(FIRST_THREAD_IN_PROC(p),
1004 	    ("proc_reap: no residual thread!"));
1005 	uma_zfree(proc_zone, p);
1006 	atomic_add_int(&nprocs, -1);
1007 }
1008 
1009 static int
1010 proc_to_reap(struct thread *td, struct proc *p, idtype_t idtype, id_t id,
1011     int *status, int options, struct __wrusage *wrusage, siginfo_t *siginfo,
1012     int check_only)
1013 {
1014 	struct rusage *rup;
1015 
1016 	sx_assert(&proctree_lock, SA_XLOCKED);
1017 
1018 	PROC_LOCK(p);
1019 
1020 	switch (idtype) {
1021 	case P_ALL:
1022 		if (p->p_procdesc == NULL ||
1023 		   (p->p_pptr == td->td_proc &&
1024 		   (p->p_flag & P_TRACED) != 0)) {
1025 			break;
1026 		}
1027 
1028 		PROC_UNLOCK(p);
1029 		return (0);
1030 	case P_PID:
1031 		if (p->p_pid != (pid_t)id) {
1032 			PROC_UNLOCK(p);
1033 			return (0);
1034 		}
1035 		break;
1036 	case P_PGID:
1037 		if (p->p_pgid != (pid_t)id) {
1038 			PROC_UNLOCK(p);
1039 			return (0);
1040 		}
1041 		break;
1042 	case P_SID:
1043 		if (p->p_session->s_sid != (pid_t)id) {
1044 			PROC_UNLOCK(p);
1045 			return (0);
1046 		}
1047 		break;
1048 	case P_UID:
1049 		if (p->p_ucred->cr_uid != (uid_t)id) {
1050 			PROC_UNLOCK(p);
1051 			return (0);
1052 		}
1053 		break;
1054 	case P_GID:
1055 		if (p->p_ucred->cr_gid != (gid_t)id) {
1056 			PROC_UNLOCK(p);
1057 			return (0);
1058 		}
1059 		break;
1060 	case P_JAILID:
1061 		if (p->p_ucred->cr_prison->pr_id != (int)id) {
1062 			PROC_UNLOCK(p);
1063 			return (0);
1064 		}
1065 		break;
1066 	/*
1067 	 * It seems that the thread structures get zeroed out
1068 	 * at process exit.  This makes it impossible to
1069 	 * support P_SETID, P_CID or P_CPUID.
1070 	 */
1071 	default:
1072 		PROC_UNLOCK(p);
1073 		return (0);
1074 	}
1075 
1076 	if (p_canwait(td, p)) {
1077 		PROC_UNLOCK(p);
1078 		return (0);
1079 	}
1080 
1081 	if (((options & WEXITED) == 0) && (p->p_state == PRS_ZOMBIE)) {
1082 		PROC_UNLOCK(p);
1083 		return (0);
1084 	}
1085 
1086 	/*
1087 	 * This special case handles a kthread spawned by linux_clone
1088 	 * (see linux_misc.c).  The linux_wait4 and linux_waitpid
1089 	 * functions need to be able to distinguish between waiting
1090 	 * on a process and waiting on a thread.  It is a thread if
1091 	 * p_sigparent is not SIGCHLD, and the WLINUXCLONE option
1092 	 * signifies we want to wait for threads and not processes.
1093 	 */
1094 	if ((p->p_sigparent != SIGCHLD) ^
1095 	    ((options & WLINUXCLONE) != 0)) {
1096 		PROC_UNLOCK(p);
1097 		return (0);
1098 	}
1099 
1100 	if (siginfo != NULL) {
1101 		bzero(siginfo, sizeof(*siginfo));
1102 		siginfo->si_errno = 0;
1103 
1104 		/*
1105 		 * SUSv4 requires that the si_signo value is always
1106 		 * SIGCHLD. Obey it despite the rfork(2) interface
1107 		 * allows to request other signal for child exit
1108 		 * notification.
1109 		 */
1110 		siginfo->si_signo = SIGCHLD;
1111 
1112 		/*
1113 		 *  This is still a rough estimate.  We will fix the
1114 		 *  cases TRAPPED, STOPPED, and CONTINUED later.
1115 		 */
1116 		if (WCOREDUMP(p->p_xsig)) {
1117 			siginfo->si_code = CLD_DUMPED;
1118 			siginfo->si_status = WTERMSIG(p->p_xsig);
1119 		} else if (WIFSIGNALED(p->p_xsig)) {
1120 			siginfo->si_code = CLD_KILLED;
1121 			siginfo->si_status = WTERMSIG(p->p_xsig);
1122 		} else {
1123 			siginfo->si_code = CLD_EXITED;
1124 			siginfo->si_status = p->p_xexit;
1125 		}
1126 
1127 		siginfo->si_pid = p->p_pid;
1128 		siginfo->si_uid = p->p_ucred->cr_uid;
1129 
1130 		/*
1131 		 * The si_addr field would be useful additional
1132 		 * detail, but apparently the PC value may be lost
1133 		 * when we reach this point.  bzero() above sets
1134 		 * siginfo->si_addr to NULL.
1135 		 */
1136 	}
1137 
1138 	/*
1139 	 * There should be no reason to limit resources usage info to
1140 	 * exited processes only.  A snapshot about any resources used
1141 	 * by a stopped process may be exactly what is needed.
1142 	 */
1143 	if (wrusage != NULL) {
1144 		rup = &wrusage->wru_self;
1145 		*rup = p->p_ru;
1146 		PROC_STATLOCK(p);
1147 		calcru(p, &rup->ru_utime, &rup->ru_stime);
1148 		PROC_STATUNLOCK(p);
1149 
1150 		rup = &wrusage->wru_children;
1151 		*rup = p->p_stats->p_cru;
1152 		calccru(p, &rup->ru_utime, &rup->ru_stime);
1153 	}
1154 
1155 	if (p->p_state == PRS_ZOMBIE && !check_only) {
1156 		proc_reap(td, p, status, options);
1157 		return (-1);
1158 	}
1159 	return (1);
1160 }
1161 
1162 int
1163 kern_wait(struct thread *td, pid_t pid, int *status, int options,
1164     struct rusage *rusage)
1165 {
1166 	struct __wrusage wru, *wrup;
1167 	idtype_t idtype;
1168 	id_t id;
1169 	int ret;
1170 
1171 	/*
1172 	 * Translate the special pid values into the (idtype, pid)
1173 	 * pair for kern_wait6.  The WAIT_MYPGRP case is handled by
1174 	 * kern_wait6() on its own.
1175 	 */
1176 	if (pid == WAIT_ANY) {
1177 		idtype = P_ALL;
1178 		id = 0;
1179 	} else if (pid < 0) {
1180 		idtype = P_PGID;
1181 		id = (id_t)-pid;
1182 	} else {
1183 		idtype = P_PID;
1184 		id = (id_t)pid;
1185 	}
1186 
1187 	if (rusage != NULL)
1188 		wrup = &wru;
1189 	else
1190 		wrup = NULL;
1191 
1192 	/*
1193 	 * For backward compatibility we implicitly add flags WEXITED
1194 	 * and WTRAPPED here.
1195 	 */
1196 	options |= WEXITED | WTRAPPED;
1197 	ret = kern_wait6(td, idtype, id, status, options, wrup, NULL);
1198 	if (rusage != NULL)
1199 		*rusage = wru.wru_self;
1200 	return (ret);
1201 }
1202 
1203 static void
1204 report_alive_proc(struct thread *td, struct proc *p, siginfo_t *siginfo,
1205     int *status, int options, int si_code)
1206 {
1207 	bool cont;
1208 
1209 	PROC_LOCK_ASSERT(p, MA_OWNED);
1210 	sx_assert(&proctree_lock, SA_XLOCKED);
1211 	MPASS(si_code == CLD_TRAPPED || si_code == CLD_STOPPED ||
1212 	    si_code == CLD_CONTINUED);
1213 
1214 	cont = si_code == CLD_CONTINUED;
1215 	if ((options & WNOWAIT) == 0) {
1216 		if (cont)
1217 			p->p_flag &= ~P_CONTINUED;
1218 		else
1219 			p->p_flag |= P_WAITED;
1220 		if (kern_wait_dequeue_sigchld &&
1221 		    (td->td_proc->p_sysent->sv_flags & SV_SIG_WAITNDQ) == 0) {
1222 			PROC_LOCK(td->td_proc);
1223 			sigqueue_take(p->p_ksi);
1224 			PROC_UNLOCK(td->td_proc);
1225 		}
1226 	}
1227 	sx_xunlock(&proctree_lock);
1228 	if (siginfo != NULL) {
1229 		siginfo->si_code = si_code;
1230 		siginfo->si_status = cont ? SIGCONT : p->p_xsig;
1231 	}
1232 	if (status != NULL)
1233 		*status = cont ? SIGCONT : W_STOPCODE(p->p_xsig);
1234 	PROC_UNLOCK(p);
1235 	td->td_retval[0] = p->p_pid;
1236 }
1237 
1238 int
1239 kern_wait6(struct thread *td, idtype_t idtype, id_t id, int *status,
1240     int options, struct __wrusage *wrusage, siginfo_t *siginfo)
1241 {
1242 	struct proc *p, *q;
1243 	pid_t pid;
1244 	int error, nfound, ret;
1245 	bool report;
1246 
1247 	AUDIT_ARG_VALUE((int)idtype);	/* XXX - This is likely wrong! */
1248 	AUDIT_ARG_PID((pid_t)id);	/* XXX - This may be wrong! */
1249 	AUDIT_ARG_VALUE(options);
1250 
1251 	q = td->td_proc;
1252 
1253 	if ((pid_t)id == WAIT_MYPGRP && (idtype == P_PID || idtype == P_PGID)) {
1254 		PROC_LOCK(q);
1255 		id = (id_t)q->p_pgid;
1256 		PROC_UNLOCK(q);
1257 		idtype = P_PGID;
1258 	}
1259 
1260 	/* If we don't know the option, just return. */
1261 	if ((options & ~(WUNTRACED | WNOHANG | WCONTINUED | WNOWAIT |
1262 	    WEXITED | WTRAPPED | WLINUXCLONE)) != 0)
1263 		return (EINVAL);
1264 	if ((options & (WEXITED | WUNTRACED | WCONTINUED | WTRAPPED)) == 0) {
1265 		/*
1266 		 * We will be unable to find any matching processes,
1267 		 * because there are no known events to look for.
1268 		 * Prefer to return error instead of blocking
1269 		 * indefinitely.
1270 		 */
1271 		return (EINVAL);
1272 	}
1273 
1274 loop:
1275 	if (q->p_flag & P_STATCHILD) {
1276 		PROC_LOCK(q);
1277 		q->p_flag &= ~P_STATCHILD;
1278 		PROC_UNLOCK(q);
1279 	}
1280 	sx_xlock(&proctree_lock);
1281 loop_locked:
1282 	nfound = 0;
1283 	LIST_FOREACH(p, &q->p_children, p_sibling) {
1284 		pid = p->p_pid;
1285 		ret = proc_to_reap(td, p, idtype, id, status, options,
1286 		    wrusage, siginfo, 0);
1287 		if (ret == 0)
1288 			continue;
1289 		else if (ret != 1) {
1290 			td->td_retval[0] = pid;
1291 			return (0);
1292 		}
1293 
1294 		nfound++;
1295 		PROC_LOCK_ASSERT(p, MA_OWNED);
1296 
1297 		if ((options & WTRAPPED) != 0 &&
1298 		    (p->p_flag & P_TRACED) != 0) {
1299 			PROC_SLOCK(p);
1300 			report =
1301 			    ((p->p_flag & (P_STOPPED_TRACE | P_STOPPED_SIG)) &&
1302 			    p->p_suspcount == p->p_numthreads &&
1303 			    (p->p_flag & P_WAITED) == 0);
1304 			PROC_SUNLOCK(p);
1305 			if (report) {
1306 			CTR4(KTR_PTRACE,
1307 			    "wait: returning trapped pid %d status %#x "
1308 			    "(xstat %d) xthread %d",
1309 			    p->p_pid, W_STOPCODE(p->p_xsig), p->p_xsig,
1310 			    p->p_xthread != NULL ?
1311 			    p->p_xthread->td_tid : -1);
1312 				report_alive_proc(td, p, siginfo, status,
1313 				    options, CLD_TRAPPED);
1314 				return (0);
1315 			}
1316 		}
1317 		if ((options & WUNTRACED) != 0 &&
1318 		    (p->p_flag & P_STOPPED_SIG) != 0) {
1319 			PROC_SLOCK(p);
1320 			report = (p->p_suspcount == p->p_numthreads &&
1321 			    ((p->p_flag & P_WAITED) == 0));
1322 			PROC_SUNLOCK(p);
1323 			if (report) {
1324 				report_alive_proc(td, p, siginfo, status,
1325 				    options, CLD_STOPPED);
1326 				return (0);
1327 			}
1328 		}
1329 		if ((options & WCONTINUED) != 0 &&
1330 		    (p->p_flag & P_CONTINUED) != 0) {
1331 			report_alive_proc(td, p, siginfo, status, options,
1332 			    CLD_CONTINUED);
1333 			return (0);
1334 		}
1335 		PROC_UNLOCK(p);
1336 	}
1337 
1338 	/*
1339 	 * Look in the orphans list too, to allow the parent to
1340 	 * collect it's child exit status even if child is being
1341 	 * debugged.
1342 	 *
1343 	 * Debugger detaches from the parent upon successful
1344 	 * switch-over from parent to child.  At this point due to
1345 	 * re-parenting the parent loses the child to debugger and a
1346 	 * wait4(2) call would report that it has no children to wait
1347 	 * for.  By maintaining a list of orphans we allow the parent
1348 	 * to successfully wait until the child becomes a zombie.
1349 	 */
1350 	if (nfound == 0) {
1351 		LIST_FOREACH(p, &q->p_orphans, p_orphan) {
1352 			ret = proc_to_reap(td, p, idtype, id, NULL, options,
1353 			    NULL, NULL, 1);
1354 			if (ret != 0) {
1355 				KASSERT(ret != -1, ("reaped an orphan (pid %d)",
1356 				    (int)td->td_retval[0]));
1357 				PROC_UNLOCK(p);
1358 				nfound++;
1359 				break;
1360 			}
1361 		}
1362 	}
1363 	if (nfound == 0) {
1364 		sx_xunlock(&proctree_lock);
1365 		return (ECHILD);
1366 	}
1367 	if (options & WNOHANG) {
1368 		sx_xunlock(&proctree_lock);
1369 		td->td_retval[0] = 0;
1370 		return (0);
1371 	}
1372 	PROC_LOCK(q);
1373 	if (q->p_flag & P_STATCHILD) {
1374 		q->p_flag &= ~P_STATCHILD;
1375 		PROC_UNLOCK(q);
1376 		goto loop_locked;
1377 	}
1378 	sx_xunlock(&proctree_lock);
1379 	error = msleep(q, &q->p_mtx, PWAIT | PCATCH | PDROP, "wait", 0);
1380 	if (error)
1381 		return (error);
1382 	goto loop;
1383 }
1384 
1385 void
1386 proc_add_orphan(struct proc *child, struct proc *parent)
1387 {
1388 
1389 	sx_assert(&proctree_lock, SX_XLOCKED);
1390 	KASSERT((child->p_flag & P_TRACED) != 0,
1391 	    ("proc_add_orphan: not traced"));
1392 
1393 	if (LIST_EMPTY(&parent->p_orphans)) {
1394 		child->p_treeflag |= P_TREE_FIRST_ORPHAN;
1395 		LIST_INSERT_HEAD(&parent->p_orphans, child, p_orphan);
1396 	} else {
1397 		LIST_INSERT_AFTER(LIST_FIRST(&parent->p_orphans),
1398 		    child, p_orphan);
1399 	}
1400 	child->p_treeflag |= P_TREE_ORPHANED;
1401 }
1402 
1403 /*
1404  * Make process 'parent' the new parent of process 'child'.
1405  * Must be called with an exclusive hold of proctree lock.
1406  */
1407 void
1408 proc_reparent(struct proc *child, struct proc *parent, bool set_oppid)
1409 {
1410 
1411 	sx_assert(&proctree_lock, SX_XLOCKED);
1412 	PROC_LOCK_ASSERT(child, MA_OWNED);
1413 	if (child->p_pptr == parent)
1414 		return;
1415 
1416 	PROC_LOCK(child->p_pptr);
1417 	sigqueue_take(child->p_ksi);
1418 	PROC_UNLOCK(child->p_pptr);
1419 	LIST_REMOVE(child, p_sibling);
1420 	LIST_INSERT_HEAD(&parent->p_children, child, p_sibling);
1421 
1422 	proc_clear_orphan(child);
1423 	if ((child->p_flag & P_TRACED) != 0) {
1424 		proc_add_orphan(child, child->p_pptr);
1425 	}
1426 
1427 	child->p_pptr = parent;
1428 	if (set_oppid)
1429 		child->p_oppid = parent->p_pid;
1430 }
1431