xref: /freebsd/sys/kern/kern_exit.c (revision 7ef62cebc2f965b0f640263e179276928885e33d)
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_XLOCKED);
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 int
209 sys_exit(struct thread *td, struct exit_args *uap)
210 {
211 
212 	exit1(td, uap->rval, 0);
213 	__unreachable();
214 }
215 
216 void
217 proc_set_p2_wexit(struct proc *p)
218 {
219 	PROC_LOCK_ASSERT(p, MA_OWNED);
220 	p->p_flag2 |= P2_WEXIT;
221 }
222 
223 /*
224  * Exit: deallocate address space and other resources, change proc state to
225  * zombie, and unlink proc from allproc and parent's lists.  Save exit status
226  * and rusage for wait().  Check for child processes and orphan them.
227  */
228 void
229 exit1(struct thread *td, int rval, int signo)
230 {
231 	struct proc *p, *nq, *q, *t;
232 	struct thread *tdt;
233 	ksiginfo_t *ksi, *ksi1;
234 	int signal_parent;
235 
236 	mtx_assert(&Giant, MA_NOTOWNED);
237 	KASSERT(rval == 0 || signo == 0, ("exit1 rv %d sig %d", rval, signo));
238 	TSPROCEXIT(td->td_proc->p_pid);
239 
240 	p = td->td_proc;
241 	/*
242 	 * In case we're rebooting we just let init die in order to
243 	 * work around an issues where pid 1 might get a fatal signal.
244 	 * For instance, if network interface serving NFS root is
245 	 * going down due to reboot, page-in requests for text are
246 	 * failing.
247 	 */
248 	if (p == initproc && rebooting == 0) {
249 		printf("init died (signal %d, exit %d)\n", signo, rval);
250 		panic("Going nowhere without my init!");
251 	}
252 
253 	/*
254 	 * Process deferred operations, designated with ASTF_KCLEAR.
255 	 * For instance, we need to deref SU mp, since the thread does
256 	 * not return to userspace, and wait for geom to stabilize.
257 	 */
258 	ast_kclear(td);
259 
260 	/*
261 	 * MUST abort all other threads before proceeding past here.
262 	 */
263 	PROC_LOCK(p);
264 	proc_set_p2_wexit(p);
265 
266 	/*
267 	 * First check if some other thread or external request got
268 	 * here before us.  If so, act appropriately: exit or suspend.
269 	 * We must ensure that stop requests are handled before we set
270 	 * P_WEXIT.
271 	 */
272 	thread_suspend_check(0);
273 	while (p->p_flag & P_HADTHREADS) {
274 		/*
275 		 * Kill off the other threads. This requires
276 		 * some co-operation from other parts of the kernel
277 		 * so it may not be instantaneous.  With this state set
278 		 * any thread attempting to interruptibly
279 		 * sleep will return immediately with EINTR or EWOULDBLOCK
280 		 * which will hopefully force them to back out to userland
281 		 * freeing resources as they go.  Any thread attempting
282 		 * to return to userland will thread_exit() from ast().
283 		 * thread_exit() will unsuspend us when the last of the
284 		 * other threads exits.
285 		 * If there is already a thread singler after resumption,
286 		 * calling thread_single() will fail; in that case, we just
287 		 * re-check all suspension request, the thread should
288 		 * either be suspended there or exit.
289 		 */
290 		if (!thread_single(p, SINGLE_EXIT))
291 			/*
292 			 * All other activity in this process is now
293 			 * stopped.  Threading support has been turned
294 			 * off.
295 			 */
296 			break;
297 		/*
298 		 * Recheck for new stop or suspend requests which
299 		 * might appear while process lock was dropped in
300 		 * thread_single().
301 		 */
302 		thread_suspend_check(0);
303 	}
304 	KASSERT(p->p_numthreads == 1,
305 	    ("exit1: proc %p exiting with %d threads", p, p->p_numthreads));
306 	racct_sub(p, RACCT_NTHR, 1);
307 
308 	/* Let event handler change exit status */
309 	p->p_xexit = rval;
310 	p->p_xsig = signo;
311 
312 	/*
313 	 * Ignore any pending request to stop due to a stop signal.
314 	 * Once P_WEXIT is set, future requests will be ignored as
315 	 * well.
316 	 */
317 	p->p_flag &= ~P_STOPPED_SIG;
318 	KASSERT(!P_SHOULDSTOP(p), ("exiting process is stopped"));
319 
320 	/* Note that we are exiting. */
321 	p->p_flag |= P_WEXIT;
322 
323 	/*
324 	 * Wait for any processes that have a hold on our vmspace to
325 	 * release their reference.
326 	 */
327 	while (p->p_lock > 0)
328 		msleep(&p->p_lock, &p->p_mtx, PWAIT, "exithold", 0);
329 
330 	PROC_UNLOCK(p);
331 	/* Drain the limit callout while we don't have the proc locked */
332 	callout_drain(&p->p_limco);
333 
334 #ifdef AUDIT
335 	/*
336 	 * The Sun BSM exit token contains two components: an exit status as
337 	 * passed to exit(), and a return value to indicate what sort of exit
338 	 * it was.  The exit status is WEXITSTATUS(rv), but it's not clear
339 	 * what the return value is.
340 	 */
341 	AUDIT_ARG_EXIT(rval, 0);
342 	AUDIT_SYSCALL_EXIT(0, td);
343 #endif
344 
345 	/* Are we a task leader with peers? */
346 	if (p->p_peers != NULL && p == p->p_leader) {
347 		mtx_lock(&ppeers_lock);
348 		q = p->p_peers;
349 		while (q != NULL) {
350 			PROC_LOCK(q);
351 			kern_psignal(q, SIGKILL);
352 			PROC_UNLOCK(q);
353 			q = q->p_peers;
354 		}
355 		while (p->p_peers != NULL)
356 			msleep(p, &ppeers_lock, PWAIT, "exit1", 0);
357 		mtx_unlock(&ppeers_lock);
358 	}
359 
360 	itimers_exit(p);
361 
362 	/*
363 	 * Check if any loadable modules need anything done at process exit.
364 	 * E.g. SYSV IPC stuff.
365 	 * Event handler could change exit status.
366 	 * XXX what if one of these generates an error?
367 	 */
368 	EVENTHANDLER_DIRECT_INVOKE(process_exit, p);
369 
370 	/*
371 	 * If parent is waiting for us to exit or exec,
372 	 * P_PPWAIT is set; we will wakeup the parent below.
373 	 */
374 	PROC_LOCK(p);
375 	stopprofclock(p);
376 	p->p_ptevents = 0;
377 
378 	/*
379 	 * Stop the real interval timer.  If the handler is currently
380 	 * executing, prevent it from rearming itself and let it finish.
381 	 */
382 	if (timevalisset(&p->p_realtimer.it_value) &&
383 	    callout_stop(&p->p_itcallout) == 0) {
384 		timevalclear(&p->p_realtimer.it_interval);
385 		PROC_UNLOCK(p);
386 		callout_drain(&p->p_itcallout);
387 	} else {
388 		PROC_UNLOCK(p);
389 	}
390 
391 	if (p->p_sysent->sv_onexit != NULL)
392 		p->p_sysent->sv_onexit(p);
393 	seltdfini(td);
394 
395 	/*
396 	 * Reset any sigio structures pointing to us as a result of
397 	 * F_SETOWN with our pid.  The P_WEXIT flag interlocks with fsetown().
398 	 */
399 	funsetownlst(&p->p_sigiolst);
400 
401 	/*
402 	 * Close open files and release open-file table.
403 	 * This may block!
404 	 */
405 	pdescfree(td);
406 	fdescfree(td);
407 
408 	/*
409 	 * Remove ourself from our leader's peer list and wake our leader.
410 	 */
411 	if (p->p_leader->p_peers != NULL) {
412 		mtx_lock(&ppeers_lock);
413 		if (p->p_leader->p_peers != NULL) {
414 			q = p->p_leader;
415 			while (q->p_peers != p)
416 				q = q->p_peers;
417 			q->p_peers = p->p_peers;
418 			wakeup(p->p_leader);
419 		}
420 		mtx_unlock(&ppeers_lock);
421 	}
422 
423 	exec_free_abi_mappings(p);
424 	vmspace_exit(td);
425 	(void)acct_process(td);
426 
427 #ifdef KTRACE
428 	ktrprocexit(td);
429 #endif
430 	/*
431 	 * Release reference to text vnode etc
432 	 */
433 	if (p->p_textvp != NULL) {
434 		vrele(p->p_textvp);
435 		p->p_textvp = NULL;
436 	}
437 	if (p->p_textdvp != NULL) {
438 		vrele(p->p_textdvp);
439 		p->p_textdvp = NULL;
440 	}
441 	if (p->p_binname != NULL) {
442 		free(p->p_binname, M_PARGS);
443 		p->p_binname = NULL;
444 	}
445 
446 	/*
447 	 * Release our limits structure.
448 	 */
449 	lim_free(p->p_limit);
450 	p->p_limit = NULL;
451 
452 	tidhash_remove(td);
453 
454 	/*
455 	 * Call machine-dependent code to release any
456 	 * machine-dependent resources other than the address space.
457 	 * The address space is released by "vmspace_exitfree(p)" in
458 	 * vm_waitproc().
459 	 */
460 	cpu_exit(td);
461 
462 	WITNESS_WARN(WARN_PANIC, NULL, "process (pid %d) exiting", p->p_pid);
463 
464 	/*
465 	 * Remove from allproc. It still sits in the hash.
466 	 */
467 	sx_xlock(&allproc_lock);
468 	LIST_REMOVE(p, p_list);
469 
470 #ifdef DDB
471 	/*
472 	 * Used by ddb's 'ps' command to find this process via the
473 	 * pidhash.
474 	 */
475 	p->p_list.le_prev = NULL;
476 #endif
477 	prison_proc_unlink(p->p_ucred->cr_prison, p);
478 	sx_xunlock(&allproc_lock);
479 
480 	sx_xlock(&proctree_lock);
481 	if ((p->p_flag & (P_TRACED | P_PPWAIT | P_PPTRACE)) != 0) {
482 		PROC_LOCK(p);
483 		p->p_flag &= ~(P_TRACED | P_PPWAIT | P_PPTRACE);
484 		PROC_UNLOCK(p);
485 	}
486 
487 	/*
488 	 * killjobc() might drop and re-acquire proctree_lock to
489 	 * revoke control tty if exiting process was a session leader.
490 	 */
491 	killjobc();
492 
493 	/*
494 	 * Reparent all children processes:
495 	 * - traced ones to the original parent (or init if we are that parent)
496 	 * - the rest to init
497 	 */
498 	q = LIST_FIRST(&p->p_children);
499 	if (q != NULL)		/* only need this if any child is S_ZOMB */
500 		wakeup(q->p_reaper);
501 	for (; q != NULL; q = nq) {
502 		nq = LIST_NEXT(q, p_sibling);
503 		ksi = ksiginfo_alloc(M_WAITOK);
504 		PROC_LOCK(q);
505 		q->p_sigparent = SIGCHLD;
506 
507 		if ((q->p_flag & P_TRACED) == 0) {
508 			proc_reparent(q, q->p_reaper, true);
509 			if (q->p_state == PRS_ZOMBIE) {
510 				/*
511 				 * Inform reaper about the reparented
512 				 * zombie, since wait(2) has something
513 				 * new to report.  Guarantee queueing
514 				 * of the SIGCHLD signal, similar to
515 				 * the _exit() behaviour, by providing
516 				 * our ksiginfo.  Ksi is freed by the
517 				 * signal delivery.
518 				 */
519 				if (q->p_ksi == NULL) {
520 					ksi1 = NULL;
521 				} else {
522 					ksiginfo_copy(q->p_ksi, ksi);
523 					ksi->ksi_flags |= KSI_INS;
524 					ksi1 = ksi;
525 					ksi = NULL;
526 				}
527 				PROC_LOCK(q->p_reaper);
528 				pksignal(q->p_reaper, SIGCHLD, ksi1);
529 				PROC_UNLOCK(q->p_reaper);
530 			} else if (q->p_pdeathsig > 0) {
531 				/*
532 				 * The child asked to received a signal
533 				 * when we exit.
534 				 */
535 				kern_psignal(q, q->p_pdeathsig);
536 			}
537 		} else {
538 			/*
539 			 * Traced processes are killed by default
540 			 * since their existence means someone is
541 			 * screwing up.
542 			 */
543 			t = proc_realparent(q);
544 			if (t == p) {
545 				proc_reparent(q, q->p_reaper, true);
546 			} else {
547 				PROC_LOCK(t);
548 				proc_reparent(q, t, true);
549 				PROC_UNLOCK(t);
550 			}
551 			/*
552 			 * Since q was found on our children list, the
553 			 * proc_reparent() call moved q to the orphan
554 			 * list due to present P_TRACED flag. Clear
555 			 * orphan link for q now while q is locked.
556 			 */
557 			proc_clear_orphan(q);
558 			q->p_flag &= ~P_TRACED;
559 			q->p_flag2 &= ~P2_PTRACE_FSTP;
560 			q->p_ptevents = 0;
561 			p->p_xthread = NULL;
562 			FOREACH_THREAD_IN_PROC(q, tdt) {
563 				tdt->td_dbgflags &= ~(TDB_SUSPEND | TDB_XSIG |
564 				    TDB_FSTP);
565 				tdt->td_xsig = 0;
566 			}
567 			if (kern_kill_on_dbg_exit) {
568 				q->p_flag &= ~P_STOPPED_TRACE;
569 				kern_psignal(q, SIGKILL);
570 			} else if ((q->p_flag & (P_STOPPED_TRACE |
571 			    P_STOPPED_SIG)) != 0) {
572 				sigqueue_delete_proc(q, SIGTRAP);
573 				ptrace_unsuspend(q);
574 			}
575 		}
576 		PROC_UNLOCK(q);
577 		if (ksi != NULL)
578 			ksiginfo_free(ksi);
579 	}
580 
581 	/*
582 	 * Also get rid of our orphans.
583 	 */
584 	while ((q = LIST_FIRST(&p->p_orphans)) != NULL) {
585 		PROC_LOCK(q);
586 		KASSERT(q->p_oppid == p->p_pid,
587 		    ("orphan %p of %p has unexpected oppid %d", q, p,
588 		    q->p_oppid));
589 		q->p_oppid = q->p_reaper->p_pid;
590 
591 		/*
592 		 * If we are the real parent of this process
593 		 * but it has been reparented to a debugger, then
594 		 * check if it asked for a signal when we exit.
595 		 */
596 		if (q->p_pdeathsig > 0)
597 			kern_psignal(q, q->p_pdeathsig);
598 		CTR2(KTR_PTRACE, "exit: pid %d, clearing orphan %d", p->p_pid,
599 		    q->p_pid);
600 		proc_clear_orphan(q);
601 		PROC_UNLOCK(q);
602 	}
603 
604 #ifdef KDTRACE_HOOKS
605 	if (SDT_PROBES_ENABLED()) {
606 		int reason = CLD_EXITED;
607 		if (WCOREDUMP(signo))
608 			reason = CLD_DUMPED;
609 		else if (WIFSIGNALED(signo))
610 			reason = CLD_KILLED;
611 		SDT_PROBE1(proc, , , exit, reason);
612 	}
613 #endif
614 
615 	/* Save exit status. */
616 	PROC_LOCK(p);
617 	p->p_xthread = td;
618 
619 	if (p->p_sysent->sv_ontdexit != NULL)
620 		p->p_sysent->sv_ontdexit(td);
621 
622 #ifdef KDTRACE_HOOKS
623 	/*
624 	 * Tell the DTrace fasttrap provider about the exit if it
625 	 * has declared an interest.
626 	 */
627 	if (dtrace_fasttrap_exit)
628 		dtrace_fasttrap_exit(p);
629 #endif
630 
631 	/*
632 	 * Notify interested parties of our demise.
633 	 */
634 	KNOTE_LOCKED(p->p_klist, NOTE_EXIT);
635 
636 	/*
637 	 * If this is a process with a descriptor, we may not need to deliver
638 	 * a signal to the parent.  proctree_lock is held over
639 	 * procdesc_exit() to serialize concurrent calls to close() and
640 	 * exit().
641 	 */
642 	signal_parent = 0;
643 	if (p->p_procdesc == NULL || procdesc_exit(p)) {
644 		/*
645 		 * Notify parent that we're gone.  If parent has the
646 		 * PS_NOCLDWAIT flag set, or if the handler is set to SIG_IGN,
647 		 * notify process 1 instead (and hope it will handle this
648 		 * situation).
649 		 */
650 		PROC_LOCK(p->p_pptr);
651 		mtx_lock(&p->p_pptr->p_sigacts->ps_mtx);
652 		if (p->p_pptr->p_sigacts->ps_flag &
653 		    (PS_NOCLDWAIT | PS_CLDSIGIGN)) {
654 			struct proc *pp;
655 
656 			mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx);
657 			pp = p->p_pptr;
658 			PROC_UNLOCK(pp);
659 			proc_reparent(p, p->p_reaper, true);
660 			p->p_sigparent = SIGCHLD;
661 			PROC_LOCK(p->p_pptr);
662 
663 			/*
664 			 * Notify parent, so in case he was wait(2)ing or
665 			 * executing waitpid(2) with our pid, he will
666 			 * continue.
667 			 */
668 			wakeup(pp);
669 		} else
670 			mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx);
671 
672 		if (p->p_pptr == p->p_reaper || p->p_pptr == initproc) {
673 			signal_parent = 1;
674 		} else if (p->p_sigparent != 0) {
675 			if (p->p_sigparent == SIGCHLD) {
676 				signal_parent = 1;
677 			} else { /* LINUX thread */
678 				signal_parent = 2;
679 			}
680 		}
681 	} else
682 		PROC_LOCK(p->p_pptr);
683 	sx_xunlock(&proctree_lock);
684 
685 	if (signal_parent == 1) {
686 		childproc_exited(p);
687 	} else if (signal_parent == 2) {
688 		kern_psignal(p->p_pptr, p->p_sigparent);
689 	}
690 
691 	/* Tell the prison that we are gone. */
692 	prison_proc_free(p->p_ucred->cr_prison);
693 
694 	/*
695 	 * The state PRS_ZOMBIE prevents other processes from sending
696 	 * signal to the process, to avoid memory leak, we free memory
697 	 * for signal queue at the time when the state is set.
698 	 */
699 	sigqueue_flush(&p->p_sigqueue);
700 	sigqueue_flush(&td->td_sigqueue);
701 
702 	/*
703 	 * We have to wait until after acquiring all locks before
704 	 * changing p_state.  We need to avoid all possible context
705 	 * switches (including ones from blocking on a mutex) while
706 	 * marked as a zombie.  We also have to set the zombie state
707 	 * before we release the parent process' proc lock to avoid
708 	 * a lost wakeup.  So, we first call wakeup, then we grab the
709 	 * sched lock, update the state, and release the parent process'
710 	 * proc lock.
711 	 */
712 	wakeup(p->p_pptr);
713 	cv_broadcast(&p->p_pwait);
714 	sched_exit(p->p_pptr, td);
715 	PROC_SLOCK(p);
716 	p->p_state = PRS_ZOMBIE;
717 	PROC_UNLOCK(p->p_pptr);
718 
719 	/*
720 	 * Save our children's rusage information in our exit rusage.
721 	 */
722 	PROC_STATLOCK(p);
723 	ruadd(&p->p_ru, &p->p_rux, &p->p_stats->p_cru, &p->p_crux);
724 	PROC_STATUNLOCK(p);
725 
726 	/*
727 	 * Make sure the scheduler takes this thread out of its tables etc.
728 	 * This will also release this thread's reference to the ucred.
729 	 * Other thread parts to release include pcb bits and such.
730 	 */
731 	thread_exit();
732 }
733 
734 #ifndef _SYS_SYSPROTO_H_
735 struct abort2_args {
736 	char *why;
737 	int nargs;
738 	void **args;
739 };
740 #endif
741 
742 int
743 sys_abort2(struct thread *td, struct abort2_args *uap)
744 {
745 	void *uargs[16];
746 	void **uargsp;
747 	int error, nargs;
748 
749 	nargs = uap->nargs;
750 	if (nargs < 0 || nargs > nitems(uargs))
751 		nargs = -1;
752 	uargsp = NULL;
753 	if (nargs > 0) {
754 		if (uap->args != NULL) {
755 			error = copyin(uap->args, uargs,
756 			    nargs * sizeof(void *));
757 			if (error != 0)
758 				nargs = -1;
759 			else
760 				uargsp = uargs;
761 		} else
762 			nargs = -1;
763 	}
764 	return (kern_abort2(td, uap->why, nargs, uargsp));
765 }
766 
767 /*
768  * kern_abort2()
769  * Arguments:
770  *  why - user pointer to why
771  *  nargs - number of arguments copied or -1 if an error occurred in copying
772  *  args - pointer to an array of pointers in kernel format
773  */
774 int
775 kern_abort2(struct thread *td, const char *why, int nargs, void **uargs)
776 {
777 	struct proc *p = td->td_proc;
778 	struct sbuf *sb;
779 	int error, i, sig;
780 
781 	/*
782 	 * Do it right now so we can log either proper call of abort2(), or
783 	 * note, that invalid argument was passed. 512 is big enough to
784 	 * handle 16 arguments' descriptions with additional comments.
785 	 */
786 	sb = sbuf_new(NULL, NULL, 512, SBUF_FIXEDLEN);
787 	sbuf_clear(sb);
788 	sbuf_printf(sb, "%s(pid %d uid %d) aborted: ",
789 	    p->p_comm, p->p_pid, td->td_ucred->cr_uid);
790 	/*
791 	 * Since we can't return from abort2(), send SIGKILL in cases, where
792 	 * abort2() was called improperly
793 	 */
794 	sig = SIGKILL;
795 	/* Prevent from DoSes from user-space. */
796 	if (nargs == -1)
797 		goto out;
798 	KASSERT(nargs >= 0 && nargs <= 16, ("called with too many args (%d)",
799 	    nargs));
800 	/*
801 	 * Limit size of 'reason' string to 128. Will fit even when
802 	 * maximal number of arguments was chosen to be logged.
803 	 */
804 	if (why != NULL) {
805 		error = sbuf_copyin(sb, why, 128);
806 		if (error < 0)
807 			goto out;
808 	} else {
809 		sbuf_printf(sb, "(null)");
810 	}
811 	if (nargs > 0) {
812 		sbuf_printf(sb, "(");
813 		for (i = 0;i < nargs; i++)
814 			sbuf_printf(sb, "%s%p", i == 0 ? "" : ", ", uargs[i]);
815 		sbuf_printf(sb, ")");
816 	}
817 	/*
818 	 * Final stage: arguments were proper, string has been
819 	 * successfully copied from userspace, and copying pointers
820 	 * from user-space succeed.
821 	 */
822 	sig = SIGABRT;
823 out:
824 	if (sig == SIGKILL) {
825 		sbuf_trim(sb);
826 		sbuf_printf(sb, " (Reason text inaccessible)");
827 	}
828 	sbuf_cat(sb, "\n");
829 	sbuf_finish(sb);
830 	log(LOG_INFO, "%s", sbuf_data(sb));
831 	sbuf_delete(sb);
832 	exit1(td, 0, sig);
833 	return (0);
834 }
835 
836 #ifdef COMPAT_43
837 /*
838  * The dirty work is handled by kern_wait().
839  */
840 int
841 owait(struct thread *td, struct owait_args *uap __unused)
842 {
843 	int error, status;
844 
845 	error = kern_wait(td, WAIT_ANY, &status, 0, NULL);
846 	if (error == 0)
847 		td->td_retval[1] = status;
848 	return (error);
849 }
850 #endif /* COMPAT_43 */
851 
852 /*
853  * The dirty work is handled by kern_wait().
854  */
855 int
856 sys_wait4(struct thread *td, struct wait4_args *uap)
857 {
858 	struct rusage ru, *rup;
859 	int error, status;
860 
861 	if (uap->rusage != NULL)
862 		rup = &ru;
863 	else
864 		rup = NULL;
865 	error = kern_wait(td, uap->pid, &status, uap->options, rup);
866 	if (uap->status != NULL && error == 0 && td->td_retval[0] != 0)
867 		error = copyout(&status, uap->status, sizeof(status));
868 	if (uap->rusage != NULL && error == 0 && td->td_retval[0] != 0)
869 		error = copyout(&ru, uap->rusage, sizeof(struct rusage));
870 	return (error);
871 }
872 
873 int
874 sys_wait6(struct thread *td, struct wait6_args *uap)
875 {
876 	struct __wrusage wru, *wrup;
877 	siginfo_t si, *sip;
878 	idtype_t idtype;
879 	id_t id;
880 	int error, status;
881 
882 	idtype = uap->idtype;
883 	id = uap->id;
884 
885 	if (uap->wrusage != NULL)
886 		wrup = &wru;
887 	else
888 		wrup = NULL;
889 
890 	if (uap->info != NULL) {
891 		sip = &si;
892 		bzero(sip, sizeof(*sip));
893 	} else
894 		sip = NULL;
895 
896 	/*
897 	 *  We expect all callers of wait6() to know about WEXITED and
898 	 *  WTRAPPED.
899 	 */
900 	error = kern_wait6(td, idtype, id, &status, uap->options, wrup, sip);
901 
902 	if (uap->status != NULL && error == 0 && td->td_retval[0] != 0)
903 		error = copyout(&status, uap->status, sizeof(status));
904 	if (uap->wrusage != NULL && error == 0 && td->td_retval[0] != 0)
905 		error = copyout(&wru, uap->wrusage, sizeof(wru));
906 	if (uap->info != NULL && error == 0)
907 		error = copyout(&si, uap->info, sizeof(si));
908 	return (error);
909 }
910 
911 /*
912  * Reap the remains of a zombie process and optionally return status and
913  * rusage.  Asserts and will release both the proctree_lock and the process
914  * lock as part of its work.
915  */
916 void
917 proc_reap(struct thread *td, struct proc *p, int *status, int options)
918 {
919 	struct proc *q, *t;
920 
921 	sx_assert(&proctree_lock, SA_XLOCKED);
922 	PROC_LOCK_ASSERT(p, MA_OWNED);
923 	KASSERT(p->p_state == PRS_ZOMBIE, ("proc_reap: !PRS_ZOMBIE"));
924 
925 	mtx_spin_wait_unlocked(&p->p_slock);
926 
927 	q = td->td_proc;
928 
929 	if (status)
930 		*status = KW_EXITCODE(p->p_xexit, p->p_xsig);
931 	if (options & WNOWAIT) {
932 		/*
933 		 *  Only poll, returning the status.  Caller does not wish to
934 		 * release the proc struct just yet.
935 		 */
936 		PROC_UNLOCK(p);
937 		sx_xunlock(&proctree_lock);
938 		return;
939 	}
940 
941 	PROC_LOCK(q);
942 	sigqueue_take(p->p_ksi);
943 	PROC_UNLOCK(q);
944 
945 	/*
946 	 * If we got the child via a ptrace 'attach', we need to give it back
947 	 * to the old parent.
948 	 */
949 	if (p->p_oppid != p->p_pptr->p_pid) {
950 		PROC_UNLOCK(p);
951 		t = proc_realparent(p);
952 		PROC_LOCK(t);
953 		PROC_LOCK(p);
954 		CTR2(KTR_PTRACE,
955 		    "wait: traced child %d moved back to parent %d", p->p_pid,
956 		    t->p_pid);
957 		proc_reparent(p, t, false);
958 		PROC_UNLOCK(p);
959 		pksignal(t, SIGCHLD, p->p_ksi);
960 		wakeup(t);
961 		cv_broadcast(&p->p_pwait);
962 		PROC_UNLOCK(t);
963 		sx_xunlock(&proctree_lock);
964 		return;
965 	}
966 	PROC_UNLOCK(p);
967 
968 	/*
969 	 * Remove other references to this process to ensure we have an
970 	 * exclusive reference.
971 	 */
972 	sx_xlock(PIDHASHLOCK(p->p_pid));
973 	LIST_REMOVE(p, p_hash);
974 	sx_xunlock(PIDHASHLOCK(p->p_pid));
975 	LIST_REMOVE(p, p_sibling);
976 	reaper_abandon_children(p, true);
977 	reaper_clear(p);
978 	PROC_LOCK(p);
979 	proc_clear_orphan(p);
980 	PROC_UNLOCK(p);
981 	leavepgrp(p);
982 	if (p->p_procdesc != NULL)
983 		procdesc_reap(p);
984 	sx_xunlock(&proctree_lock);
985 
986 	proc_id_clear(PROC_ID_PID, p->p_pid);
987 
988 	PROC_LOCK(p);
989 	knlist_detach(p->p_klist);
990 	p->p_klist = NULL;
991 	PROC_UNLOCK(p);
992 
993 	/*
994 	 * Removal from allproc list and process group list paired with
995 	 * PROC_LOCK which was executed during that time should guarantee
996 	 * nothing can reach this process anymore. As such further locking
997 	 * is unnecessary.
998 	 */
999 	p->p_xexit = p->p_xsig = 0;		/* XXX: why? */
1000 
1001 	PROC_LOCK(q);
1002 	ruadd(&q->p_stats->p_cru, &q->p_crux, &p->p_ru, &p->p_rux);
1003 	PROC_UNLOCK(q);
1004 
1005 	/*
1006 	 * Decrement the count of procs running with this uid.
1007 	 */
1008 	(void)chgproccnt(p->p_ucred->cr_ruidinfo, -1, 0);
1009 
1010 	/*
1011 	 * Destroy resource accounting information associated with the process.
1012 	 */
1013 #ifdef RACCT
1014 	if (racct_enable) {
1015 		PROC_LOCK(p);
1016 		racct_sub(p, RACCT_NPROC, 1);
1017 		PROC_UNLOCK(p);
1018 	}
1019 #endif
1020 	racct_proc_exit(p);
1021 
1022 	/*
1023 	 * Free credentials, arguments, and sigacts.
1024 	 */
1025 	proc_unset_cred(p);
1026 	pargs_drop(p->p_args);
1027 	p->p_args = NULL;
1028 	sigacts_free(p->p_sigacts);
1029 	p->p_sigacts = NULL;
1030 
1031 	/*
1032 	 * Do any thread-system specific cleanups.
1033 	 */
1034 	thread_wait(p);
1035 
1036 	/*
1037 	 * Give vm and machine-dependent layer a chance to free anything that
1038 	 * cpu_exit couldn't release while still running in process context.
1039 	 */
1040 	vm_waitproc(p);
1041 #ifdef MAC
1042 	mac_proc_destroy(p);
1043 #endif
1044 
1045 	KASSERT(FIRST_THREAD_IN_PROC(p),
1046 	    ("proc_reap: no residual thread!"));
1047 	uma_zfree(proc_zone, p);
1048 	atomic_add_int(&nprocs, -1);
1049 }
1050 
1051 static int
1052 proc_to_reap(struct thread *td, struct proc *p, idtype_t idtype, id_t id,
1053     int *status, int options, struct __wrusage *wrusage, siginfo_t *siginfo,
1054     int check_only)
1055 {
1056 	struct rusage *rup;
1057 
1058 	sx_assert(&proctree_lock, SA_XLOCKED);
1059 
1060 	PROC_LOCK(p);
1061 
1062 	switch (idtype) {
1063 	case P_ALL:
1064 		if (p->p_procdesc == NULL ||
1065 		   (p->p_pptr == td->td_proc &&
1066 		   (p->p_flag & P_TRACED) != 0)) {
1067 			break;
1068 		}
1069 
1070 		PROC_UNLOCK(p);
1071 		return (0);
1072 	case P_PID:
1073 		if (p->p_pid != (pid_t)id) {
1074 			PROC_UNLOCK(p);
1075 			return (0);
1076 		}
1077 		break;
1078 	case P_PGID:
1079 		if (p->p_pgid != (pid_t)id) {
1080 			PROC_UNLOCK(p);
1081 			return (0);
1082 		}
1083 		break;
1084 	case P_SID:
1085 		if (p->p_session->s_sid != (pid_t)id) {
1086 			PROC_UNLOCK(p);
1087 			return (0);
1088 		}
1089 		break;
1090 	case P_UID:
1091 		if (p->p_ucred->cr_uid != (uid_t)id) {
1092 			PROC_UNLOCK(p);
1093 			return (0);
1094 		}
1095 		break;
1096 	case P_GID:
1097 		if (p->p_ucred->cr_gid != (gid_t)id) {
1098 			PROC_UNLOCK(p);
1099 			return (0);
1100 		}
1101 		break;
1102 	case P_JAILID:
1103 		if (p->p_ucred->cr_prison->pr_id != (int)id) {
1104 			PROC_UNLOCK(p);
1105 			return (0);
1106 		}
1107 		break;
1108 	/*
1109 	 * It seems that the thread structures get zeroed out
1110 	 * at process exit.  This makes it impossible to
1111 	 * support P_SETID, P_CID or P_CPUID.
1112 	 */
1113 	default:
1114 		PROC_UNLOCK(p);
1115 		return (0);
1116 	}
1117 
1118 	if (p_canwait(td, p)) {
1119 		PROC_UNLOCK(p);
1120 		return (0);
1121 	}
1122 
1123 	if (((options & WEXITED) == 0) && (p->p_state == PRS_ZOMBIE)) {
1124 		PROC_UNLOCK(p);
1125 		return (0);
1126 	}
1127 
1128 	/*
1129 	 * This special case handles a kthread spawned by linux_clone
1130 	 * (see linux_misc.c).  The linux_wait4 and linux_waitpid
1131 	 * functions need to be able to distinguish between waiting
1132 	 * on a process and waiting on a thread.  It is a thread if
1133 	 * p_sigparent is not SIGCHLD, and the WLINUXCLONE option
1134 	 * signifies we want to wait for threads and not processes.
1135 	 */
1136 	if ((p->p_sigparent != SIGCHLD) ^
1137 	    ((options & WLINUXCLONE) != 0)) {
1138 		PROC_UNLOCK(p);
1139 		return (0);
1140 	}
1141 
1142 	if (siginfo != NULL) {
1143 		bzero(siginfo, sizeof(*siginfo));
1144 		siginfo->si_errno = 0;
1145 
1146 		/*
1147 		 * SUSv4 requires that the si_signo value is always
1148 		 * SIGCHLD. Obey it despite the rfork(2) interface
1149 		 * allows to request other signal for child exit
1150 		 * notification.
1151 		 */
1152 		siginfo->si_signo = SIGCHLD;
1153 
1154 		/*
1155 		 *  This is still a rough estimate.  We will fix the
1156 		 *  cases TRAPPED, STOPPED, and CONTINUED later.
1157 		 */
1158 		if (WCOREDUMP(p->p_xsig)) {
1159 			siginfo->si_code = CLD_DUMPED;
1160 			siginfo->si_status = WTERMSIG(p->p_xsig);
1161 		} else if (WIFSIGNALED(p->p_xsig)) {
1162 			siginfo->si_code = CLD_KILLED;
1163 			siginfo->si_status = WTERMSIG(p->p_xsig);
1164 		} else {
1165 			siginfo->si_code = CLD_EXITED;
1166 			siginfo->si_status = p->p_xexit;
1167 		}
1168 
1169 		siginfo->si_pid = p->p_pid;
1170 		siginfo->si_uid = p->p_ucred->cr_uid;
1171 
1172 		/*
1173 		 * The si_addr field would be useful additional
1174 		 * detail, but apparently the PC value may be lost
1175 		 * when we reach this point.  bzero() above sets
1176 		 * siginfo->si_addr to NULL.
1177 		 */
1178 	}
1179 
1180 	/*
1181 	 * There should be no reason to limit resources usage info to
1182 	 * exited processes only.  A snapshot about any resources used
1183 	 * by a stopped process may be exactly what is needed.
1184 	 */
1185 	if (wrusage != NULL) {
1186 		rup = &wrusage->wru_self;
1187 		*rup = p->p_ru;
1188 		PROC_STATLOCK(p);
1189 		calcru(p, &rup->ru_utime, &rup->ru_stime);
1190 		PROC_STATUNLOCK(p);
1191 
1192 		rup = &wrusage->wru_children;
1193 		*rup = p->p_stats->p_cru;
1194 		calccru(p, &rup->ru_utime, &rup->ru_stime);
1195 	}
1196 
1197 	if (p->p_state == PRS_ZOMBIE && !check_only) {
1198 		proc_reap(td, p, status, options);
1199 		return (-1);
1200 	}
1201 	return (1);
1202 }
1203 
1204 int
1205 kern_wait(struct thread *td, pid_t pid, int *status, int options,
1206     struct rusage *rusage)
1207 {
1208 	struct __wrusage wru, *wrup;
1209 	idtype_t idtype;
1210 	id_t id;
1211 	int ret;
1212 
1213 	/*
1214 	 * Translate the special pid values into the (idtype, pid)
1215 	 * pair for kern_wait6.  The WAIT_MYPGRP case is handled by
1216 	 * kern_wait6() on its own.
1217 	 */
1218 	if (pid == WAIT_ANY) {
1219 		idtype = P_ALL;
1220 		id = 0;
1221 	} else if (pid < 0) {
1222 		idtype = P_PGID;
1223 		id = (id_t)-pid;
1224 	} else {
1225 		idtype = P_PID;
1226 		id = (id_t)pid;
1227 	}
1228 
1229 	if (rusage != NULL)
1230 		wrup = &wru;
1231 	else
1232 		wrup = NULL;
1233 
1234 	/*
1235 	 * For backward compatibility we implicitly add flags WEXITED
1236 	 * and WTRAPPED here.
1237 	 */
1238 	options |= WEXITED | WTRAPPED;
1239 	ret = kern_wait6(td, idtype, id, status, options, wrup, NULL);
1240 	if (rusage != NULL)
1241 		*rusage = wru.wru_self;
1242 	return (ret);
1243 }
1244 
1245 static void
1246 report_alive_proc(struct thread *td, struct proc *p, siginfo_t *siginfo,
1247     int *status, int options, int si_code)
1248 {
1249 	bool cont;
1250 
1251 	PROC_LOCK_ASSERT(p, MA_OWNED);
1252 	sx_assert(&proctree_lock, SA_XLOCKED);
1253 	MPASS(si_code == CLD_TRAPPED || si_code == CLD_STOPPED ||
1254 	    si_code == CLD_CONTINUED);
1255 
1256 	cont = si_code == CLD_CONTINUED;
1257 	if ((options & WNOWAIT) == 0) {
1258 		if (cont)
1259 			p->p_flag &= ~P_CONTINUED;
1260 		else
1261 			p->p_flag |= P_WAITED;
1262 		if (kern_wait_dequeue_sigchld &&
1263 		    (td->td_proc->p_sysent->sv_flags & SV_SIG_WAITNDQ) == 0) {
1264 			PROC_LOCK(td->td_proc);
1265 			sigqueue_take(p->p_ksi);
1266 			PROC_UNLOCK(td->td_proc);
1267 		}
1268 	}
1269 	sx_xunlock(&proctree_lock);
1270 	if (siginfo != NULL) {
1271 		siginfo->si_code = si_code;
1272 		siginfo->si_status = cont ? SIGCONT : p->p_xsig;
1273 	}
1274 	if (status != NULL)
1275 		*status = cont ? SIGCONT : W_STOPCODE(p->p_xsig);
1276 	td->td_retval[0] = p->p_pid;
1277 	PROC_UNLOCK(p);
1278 }
1279 
1280 int
1281 kern_wait6(struct thread *td, idtype_t idtype, id_t id, int *status,
1282     int options, struct __wrusage *wrusage, siginfo_t *siginfo)
1283 {
1284 	struct proc *p, *q;
1285 	pid_t pid;
1286 	int error, nfound, ret;
1287 	bool report;
1288 
1289 	AUDIT_ARG_VALUE((int)idtype);	/* XXX - This is likely wrong! */
1290 	AUDIT_ARG_PID((pid_t)id);	/* XXX - This may be wrong! */
1291 	AUDIT_ARG_VALUE(options);
1292 
1293 	q = td->td_proc;
1294 
1295 	if ((pid_t)id == WAIT_MYPGRP && (idtype == P_PID || idtype == P_PGID)) {
1296 		PROC_LOCK(q);
1297 		id = (id_t)q->p_pgid;
1298 		PROC_UNLOCK(q);
1299 		idtype = P_PGID;
1300 	}
1301 
1302 	/* If we don't know the option, just return. */
1303 	if ((options & ~(WUNTRACED | WNOHANG | WCONTINUED | WNOWAIT |
1304 	    WEXITED | WTRAPPED | WLINUXCLONE)) != 0)
1305 		return (EINVAL);
1306 	if ((options & (WEXITED | WUNTRACED | WCONTINUED | WTRAPPED)) == 0) {
1307 		/*
1308 		 * We will be unable to find any matching processes,
1309 		 * because there are no known events to look for.
1310 		 * Prefer to return error instead of blocking
1311 		 * indefinitely.
1312 		 */
1313 		return (EINVAL);
1314 	}
1315 
1316 loop:
1317 	if (q->p_flag & P_STATCHILD) {
1318 		PROC_LOCK(q);
1319 		q->p_flag &= ~P_STATCHILD;
1320 		PROC_UNLOCK(q);
1321 	}
1322 	sx_xlock(&proctree_lock);
1323 loop_locked:
1324 	nfound = 0;
1325 	LIST_FOREACH(p, &q->p_children, p_sibling) {
1326 		pid = p->p_pid;
1327 		ret = proc_to_reap(td, p, idtype, id, status, options,
1328 		    wrusage, siginfo, 0);
1329 		if (ret == 0)
1330 			continue;
1331 		else if (ret != 1) {
1332 			td->td_retval[0] = pid;
1333 			return (0);
1334 		}
1335 
1336 		nfound++;
1337 		PROC_LOCK_ASSERT(p, MA_OWNED);
1338 
1339 		if ((options & WTRAPPED) != 0 &&
1340 		    (p->p_flag & P_TRACED) != 0) {
1341 			PROC_SLOCK(p);
1342 			report =
1343 			    ((p->p_flag & (P_STOPPED_TRACE | P_STOPPED_SIG)) &&
1344 			    p->p_suspcount == p->p_numthreads &&
1345 			    (p->p_flag & P_WAITED) == 0);
1346 			PROC_SUNLOCK(p);
1347 			if (report) {
1348 			CTR4(KTR_PTRACE,
1349 			    "wait: returning trapped pid %d status %#x "
1350 			    "(xstat %d) xthread %d",
1351 			    p->p_pid, W_STOPCODE(p->p_xsig), p->p_xsig,
1352 			    p->p_xthread != NULL ?
1353 			    p->p_xthread->td_tid : -1);
1354 				report_alive_proc(td, p, siginfo, status,
1355 				    options, CLD_TRAPPED);
1356 				return (0);
1357 			}
1358 		}
1359 		if ((options & WUNTRACED) != 0 &&
1360 		    (p->p_flag & P_STOPPED_SIG) != 0) {
1361 			PROC_SLOCK(p);
1362 			report = (p->p_suspcount == p->p_numthreads &&
1363 			    ((p->p_flag & P_WAITED) == 0));
1364 			PROC_SUNLOCK(p);
1365 			if (report) {
1366 				report_alive_proc(td, p, siginfo, status,
1367 				    options, CLD_STOPPED);
1368 				return (0);
1369 			}
1370 		}
1371 		if ((options & WCONTINUED) != 0 &&
1372 		    (p->p_flag & P_CONTINUED) != 0) {
1373 			report_alive_proc(td, p, siginfo, status, options,
1374 			    CLD_CONTINUED);
1375 			return (0);
1376 		}
1377 		PROC_UNLOCK(p);
1378 	}
1379 
1380 	/*
1381 	 * Look in the orphans list too, to allow the parent to
1382 	 * collect it's child exit status even if child is being
1383 	 * debugged.
1384 	 *
1385 	 * Debugger detaches from the parent upon successful
1386 	 * switch-over from parent to child.  At this point due to
1387 	 * re-parenting the parent loses the child to debugger and a
1388 	 * wait4(2) call would report that it has no children to wait
1389 	 * for.  By maintaining a list of orphans we allow the parent
1390 	 * to successfully wait until the child becomes a zombie.
1391 	 */
1392 	if (nfound == 0) {
1393 		LIST_FOREACH(p, &q->p_orphans, p_orphan) {
1394 			ret = proc_to_reap(td, p, idtype, id, NULL, options,
1395 			    NULL, NULL, 1);
1396 			if (ret != 0) {
1397 				KASSERT(ret != -1, ("reaped an orphan (pid %d)",
1398 				    (int)td->td_retval[0]));
1399 				PROC_UNLOCK(p);
1400 				nfound++;
1401 				break;
1402 			}
1403 		}
1404 	}
1405 	if (nfound == 0) {
1406 		sx_xunlock(&proctree_lock);
1407 		return (ECHILD);
1408 	}
1409 	if (options & WNOHANG) {
1410 		sx_xunlock(&proctree_lock);
1411 		td->td_retval[0] = 0;
1412 		return (0);
1413 	}
1414 	PROC_LOCK(q);
1415 	if (q->p_flag & P_STATCHILD) {
1416 		q->p_flag &= ~P_STATCHILD;
1417 		PROC_UNLOCK(q);
1418 		goto loop_locked;
1419 	}
1420 	sx_xunlock(&proctree_lock);
1421 	error = msleep(q, &q->p_mtx, PWAIT | PCATCH | PDROP, "wait", 0);
1422 	if (error)
1423 		return (error);
1424 	goto loop;
1425 }
1426 
1427 void
1428 proc_add_orphan(struct proc *child, struct proc *parent)
1429 {
1430 
1431 	sx_assert(&proctree_lock, SX_XLOCKED);
1432 	KASSERT((child->p_flag & P_TRACED) != 0,
1433 	    ("proc_add_orphan: not traced"));
1434 
1435 	if (LIST_EMPTY(&parent->p_orphans)) {
1436 		child->p_treeflag |= P_TREE_FIRST_ORPHAN;
1437 		LIST_INSERT_HEAD(&parent->p_orphans, child, p_orphan);
1438 	} else {
1439 		LIST_INSERT_AFTER(LIST_FIRST(&parent->p_orphans),
1440 		    child, p_orphan);
1441 	}
1442 	child->p_treeflag |= P_TREE_ORPHANED;
1443 }
1444 
1445 /*
1446  * Make process 'parent' the new parent of process 'child'.
1447  * Must be called with an exclusive hold of proctree lock.
1448  */
1449 void
1450 proc_reparent(struct proc *child, struct proc *parent, bool set_oppid)
1451 {
1452 
1453 	sx_assert(&proctree_lock, SX_XLOCKED);
1454 	PROC_LOCK_ASSERT(child, MA_OWNED);
1455 	if (child->p_pptr == parent)
1456 		return;
1457 
1458 	PROC_LOCK(child->p_pptr);
1459 	sigqueue_take(child->p_ksi);
1460 	PROC_UNLOCK(child->p_pptr);
1461 	LIST_REMOVE(child, p_sibling);
1462 	LIST_INSERT_HEAD(&parent->p_children, child, p_sibling);
1463 
1464 	proc_clear_orphan(child);
1465 	if ((child->p_flag & P_TRACED) != 0) {
1466 		proc_add_orphan(child, child->p_pptr);
1467 	}
1468 
1469 	child->p_pptr = parent;
1470 	if (set_oppid)
1471 		child->p_oppid = parent->p_pid;
1472 }
1473