xref: /freebsd/sys/kern/kern_exit.c (revision 43a5ec4eb41567cc92586503212743d89686d78f)
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 int
209 sys_exit(struct thread *td, struct exit_args *uap)
210 {
211 
212 	exit1(td, uap->rval, 0);
213 	__unreachable();
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 	exec_free_abi_mappings(p);
421 	vmspace_exit(td);
422 	(void)acct_process(td);
423 
424 #ifdef KTRACE
425 	ktrprocexit(td);
426 #endif
427 	/*
428 	 * Release reference to text vnode etc
429 	 */
430 	if (p->p_textvp != NULL) {
431 		vrele(p->p_textvp);
432 		p->p_textvp = NULL;
433 	}
434 	if (p->p_textdvp != NULL) {
435 		vrele(p->p_textdvp);
436 		p->p_textdvp = NULL;
437 	}
438 	if (p->p_binname != NULL) {
439 		free(p->p_binname, M_PARGS);
440 		p->p_binname = NULL;
441 	}
442 
443 	/*
444 	 * Release our limits structure.
445 	 */
446 	lim_free(p->p_limit);
447 	p->p_limit = NULL;
448 
449 	tidhash_remove(td);
450 
451 	/*
452 	 * Call machine-dependent code to release any
453 	 * machine-dependent resources other than the address space.
454 	 * The address space is released by "vmspace_exitfree(p)" in
455 	 * vm_waitproc().
456 	 */
457 	cpu_exit(td);
458 
459 	WITNESS_WARN(WARN_PANIC, NULL, "process (pid %d) exiting", p->p_pid);
460 
461 	/*
462 	 * Remove from allproc. It still sits in the hash.
463 	 */
464 	sx_xlock(&allproc_lock);
465 	LIST_REMOVE(p, p_list);
466 
467 #ifdef DDB
468 	/*
469 	 * Used by ddb's 'ps' command to find this process via the
470 	 * pidhash.
471 	 */
472 	p->p_list.le_prev = NULL;
473 #endif
474 	sx_xunlock(&allproc_lock);
475 
476 	sx_xlock(&proctree_lock);
477 	PROC_LOCK(p);
478 	p->p_flag &= ~(P_TRACED | P_PPWAIT | P_PPTRACE);
479 	PROC_UNLOCK(p);
480 
481 	/*
482 	 * killjobc() might drop and re-acquire proctree_lock to
483 	 * revoke control tty if exiting process was a session leader.
484 	 */
485 	killjobc();
486 
487 	/*
488 	 * Reparent all children processes:
489 	 * - traced ones to the original parent (or init if we are that parent)
490 	 * - the rest to init
491 	 */
492 	q = LIST_FIRST(&p->p_children);
493 	if (q != NULL)		/* only need this if any child is S_ZOMB */
494 		wakeup(q->p_reaper);
495 	for (; q != NULL; q = nq) {
496 		nq = LIST_NEXT(q, p_sibling);
497 		ksi = ksiginfo_alloc(TRUE);
498 		PROC_LOCK(q);
499 		q->p_sigparent = SIGCHLD;
500 
501 		if ((q->p_flag & P_TRACED) == 0) {
502 			proc_reparent(q, q->p_reaper, true);
503 			if (q->p_state == PRS_ZOMBIE) {
504 				/*
505 				 * Inform reaper about the reparented
506 				 * zombie, since wait(2) has something
507 				 * new to report.  Guarantee queueing
508 				 * of the SIGCHLD signal, similar to
509 				 * the _exit() behaviour, by providing
510 				 * our ksiginfo.  Ksi is freed by the
511 				 * signal delivery.
512 				 */
513 				if (q->p_ksi == NULL) {
514 					ksi1 = NULL;
515 				} else {
516 					ksiginfo_copy(q->p_ksi, ksi);
517 					ksi->ksi_flags |= KSI_INS;
518 					ksi1 = ksi;
519 					ksi = NULL;
520 				}
521 				PROC_LOCK(q->p_reaper);
522 				pksignal(q->p_reaper, SIGCHLD, ksi1);
523 				PROC_UNLOCK(q->p_reaper);
524 			} else if (q->p_pdeathsig > 0) {
525 				/*
526 				 * The child asked to received a signal
527 				 * when we exit.
528 				 */
529 				kern_psignal(q, q->p_pdeathsig);
530 			}
531 		} else {
532 			/*
533 			 * Traced processes are killed by default
534 			 * since their existence means someone is
535 			 * screwing up.
536 			 */
537 			t = proc_realparent(q);
538 			if (t == p) {
539 				proc_reparent(q, q->p_reaper, true);
540 			} else {
541 				PROC_LOCK(t);
542 				proc_reparent(q, t, true);
543 				PROC_UNLOCK(t);
544 			}
545 			/*
546 			 * Since q was found on our children list, the
547 			 * proc_reparent() call moved q to the orphan
548 			 * list due to present P_TRACED flag. Clear
549 			 * orphan link for q now while q is locked.
550 			 */
551 			proc_clear_orphan(q);
552 			q->p_flag &= ~P_TRACED;
553 			q->p_flag2 &= ~P2_PTRACE_FSTP;
554 			q->p_ptevents = 0;
555 			p->p_xthread = NULL;
556 			FOREACH_THREAD_IN_PROC(q, tdt) {
557 				tdt->td_dbgflags &= ~(TDB_SUSPEND | TDB_XSIG |
558 				    TDB_FSTP);
559 				tdt->td_xsig = 0;
560 			}
561 			if (kern_kill_on_dbg_exit) {
562 				q->p_flag &= ~P_STOPPED_TRACE;
563 				kern_psignal(q, SIGKILL);
564 			} else if ((q->p_flag & (P_STOPPED_TRACE |
565 			    P_STOPPED_SIG)) != 0) {
566 				sigqueue_delete_proc(q, SIGTRAP);
567 				ptrace_unsuspend(q);
568 			}
569 		}
570 		PROC_UNLOCK(q);
571 		if (ksi != NULL)
572 			ksiginfo_free(ksi);
573 	}
574 
575 	/*
576 	 * Also get rid of our orphans.
577 	 */
578 	while ((q = LIST_FIRST(&p->p_orphans)) != NULL) {
579 		PROC_LOCK(q);
580 		KASSERT(q->p_oppid == p->p_pid,
581 		    ("orphan %p of %p has unexpected oppid %d", q, p,
582 		    q->p_oppid));
583 		q->p_oppid = q->p_reaper->p_pid;
584 
585 		/*
586 		 * If we are the real parent of this process
587 		 * but it has been reparented to a debugger, then
588 		 * check if it asked for a signal when we exit.
589 		 */
590 		if (q->p_pdeathsig > 0)
591 			kern_psignal(q, q->p_pdeathsig);
592 		CTR2(KTR_PTRACE, "exit: pid %d, clearing orphan %d", p->p_pid,
593 		    q->p_pid);
594 		proc_clear_orphan(q);
595 		PROC_UNLOCK(q);
596 	}
597 
598 #ifdef KDTRACE_HOOKS
599 	if (SDT_PROBES_ENABLED()) {
600 		int reason = CLD_EXITED;
601 		if (WCOREDUMP(signo))
602 			reason = CLD_DUMPED;
603 		else if (WIFSIGNALED(signo))
604 			reason = CLD_KILLED;
605 		SDT_PROBE1(proc, , , exit, reason);
606 	}
607 #endif
608 
609 	/* Save exit status. */
610 	PROC_LOCK(p);
611 	p->p_xthread = td;
612 
613 	if (p->p_sysent->sv_ontdexit != NULL)
614 		p->p_sysent->sv_ontdexit(td);
615 
616 #ifdef KDTRACE_HOOKS
617 	/*
618 	 * Tell the DTrace fasttrap provider about the exit if it
619 	 * has declared an interest.
620 	 */
621 	if (dtrace_fasttrap_exit)
622 		dtrace_fasttrap_exit(p);
623 #endif
624 
625 	/*
626 	 * Notify interested parties of our demise.
627 	 */
628 	KNOTE_LOCKED(p->p_klist, NOTE_EXIT);
629 
630 	/*
631 	 * If this is a process with a descriptor, we may not need to deliver
632 	 * a signal to the parent.  proctree_lock is held over
633 	 * procdesc_exit() to serialize concurrent calls to close() and
634 	 * exit().
635 	 */
636 	signal_parent = 0;
637 	if (p->p_procdesc == NULL || procdesc_exit(p)) {
638 		/*
639 		 * Notify parent that we're gone.  If parent has the
640 		 * PS_NOCLDWAIT flag set, or if the handler is set to SIG_IGN,
641 		 * notify process 1 instead (and hope it will handle this
642 		 * situation).
643 		 */
644 		PROC_LOCK(p->p_pptr);
645 		mtx_lock(&p->p_pptr->p_sigacts->ps_mtx);
646 		if (p->p_pptr->p_sigacts->ps_flag &
647 		    (PS_NOCLDWAIT | PS_CLDSIGIGN)) {
648 			struct proc *pp;
649 
650 			mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx);
651 			pp = p->p_pptr;
652 			PROC_UNLOCK(pp);
653 			proc_reparent(p, p->p_reaper, true);
654 			p->p_sigparent = SIGCHLD;
655 			PROC_LOCK(p->p_pptr);
656 
657 			/*
658 			 * Notify parent, so in case he was wait(2)ing or
659 			 * executing waitpid(2) with our pid, he will
660 			 * continue.
661 			 */
662 			wakeup(pp);
663 		} else
664 			mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx);
665 
666 		if (p->p_pptr == p->p_reaper || p->p_pptr == initproc) {
667 			signal_parent = 1;
668 		} else if (p->p_sigparent != 0) {
669 			if (p->p_sigparent == SIGCHLD) {
670 				signal_parent = 1;
671 			} else { /* LINUX thread */
672 				signal_parent = 2;
673 			}
674 		}
675 	} else
676 		PROC_LOCK(p->p_pptr);
677 	sx_xunlock(&proctree_lock);
678 
679 	if (signal_parent == 1) {
680 		childproc_exited(p);
681 	} else if (signal_parent == 2) {
682 		kern_psignal(p->p_pptr, p->p_sigparent);
683 	}
684 
685 	/* Tell the prison that we are gone. */
686 	prison_proc_free(p->p_ucred->cr_prison);
687 
688 	/*
689 	 * The state PRS_ZOMBIE prevents other proesses from sending
690 	 * signal to the process, to avoid memory leak, we free memory
691 	 * for signal queue at the time when the state is set.
692 	 */
693 	sigqueue_flush(&p->p_sigqueue);
694 	sigqueue_flush(&td->td_sigqueue);
695 
696 	/*
697 	 * We have to wait until after acquiring all locks before
698 	 * changing p_state.  We need to avoid all possible context
699 	 * switches (including ones from blocking on a mutex) while
700 	 * marked as a zombie.  We also have to set the zombie state
701 	 * before we release the parent process' proc lock to avoid
702 	 * a lost wakeup.  So, we first call wakeup, then we grab the
703 	 * sched lock, update the state, and release the parent process'
704 	 * proc lock.
705 	 */
706 	wakeup(p->p_pptr);
707 	cv_broadcast(&p->p_pwait);
708 	sched_exit(p->p_pptr, td);
709 	PROC_SLOCK(p);
710 	p->p_state = PRS_ZOMBIE;
711 	PROC_UNLOCK(p->p_pptr);
712 
713 	/*
714 	 * Save our children's rusage information in our exit rusage.
715 	 */
716 	PROC_STATLOCK(p);
717 	ruadd(&p->p_ru, &p->p_rux, &p->p_stats->p_cru, &p->p_crux);
718 	PROC_STATUNLOCK(p);
719 
720 	/*
721 	 * Make sure the scheduler takes this thread out of its tables etc.
722 	 * This will also release this thread's reference to the ucred.
723 	 * Other thread parts to release include pcb bits and such.
724 	 */
725 	thread_exit();
726 }
727 
728 #ifndef _SYS_SYSPROTO_H_
729 struct abort2_args {
730 	char *why;
731 	int nargs;
732 	void **args;
733 };
734 #endif
735 
736 int
737 sys_abort2(struct thread *td, struct abort2_args *uap)
738 {
739 	void *uargs[16];
740 	void **uargsp;
741 	int error, nargs;
742 
743 	nargs = uap->nargs;
744 	if (nargs < 0 || nargs > nitems(uargs))
745 		nargs = -1;
746 	uargsp = NULL;
747 	if (nargs > 0) {
748 		if (uap->args != NULL) {
749 			error = copyin(uap->args, uargs,
750 			    nargs * sizeof(void *));
751 			if (error != 0)
752 				nargs = -1;
753 			else
754 				uargsp = uargs;
755 		} else
756 			nargs = -1;
757 	}
758 	return (kern_abort2(td, uap->why, nargs, uargsp));
759 }
760 
761 /*
762  * kern_abort2()
763  * Arguments:
764  *  why - user pointer to why
765  *  nargs - number of arguments copied or -1 if an error occured in copying
766  *  args - pointer to an array of pointers in kernel format
767  */
768 int
769 kern_abort2(struct thread *td, const char *why, int nargs, void **uargs)
770 {
771 	struct proc *p = td->td_proc;
772 	struct sbuf *sb;
773 	int error, i, sig;
774 
775 	/*
776 	 * Do it right now so we can log either proper call of abort2(), or
777 	 * note, that invalid argument was passed. 512 is big enough to
778 	 * handle 16 arguments' descriptions with additional comments.
779 	 */
780 	sb = sbuf_new(NULL, NULL, 512, SBUF_FIXEDLEN);
781 	sbuf_clear(sb);
782 	sbuf_printf(sb, "%s(pid %d uid %d) aborted: ",
783 	    p->p_comm, p->p_pid, td->td_ucred->cr_uid);
784 	/*
785 	 * Since we can't return from abort2(), send SIGKILL in cases, where
786 	 * abort2() was called improperly
787 	 */
788 	sig = SIGKILL;
789 	/* Prevent from DoSes from user-space. */
790 	if (nargs == -1)
791 		goto out;
792 	KASSERT(nargs >= 0 && nargs <= 16, ("called with too many args (%d)",
793 	    nargs));
794 	/*
795 	 * Limit size of 'reason' string to 128. Will fit even when
796 	 * maximal number of arguments was chosen to be logged.
797 	 */
798 	if (why != NULL) {
799 		error = sbuf_copyin(sb, why, 128);
800 		if (error < 0)
801 			goto out;
802 	} else {
803 		sbuf_printf(sb, "(null)");
804 	}
805 	if (nargs > 0) {
806 		sbuf_printf(sb, "(");
807 		for (i = 0;i < nargs; i++)
808 			sbuf_printf(sb, "%s%p", i == 0 ? "" : ", ", uargs[i]);
809 		sbuf_printf(sb, ")");
810 	}
811 	/*
812 	 * Final stage: arguments were proper, string has been
813 	 * successfully copied from userspace, and copying pointers
814 	 * from user-space succeed.
815 	 */
816 	sig = SIGABRT;
817 out:
818 	if (sig == SIGKILL) {
819 		sbuf_trim(sb);
820 		sbuf_printf(sb, " (Reason text inaccessible)");
821 	}
822 	sbuf_cat(sb, "\n");
823 	sbuf_finish(sb);
824 	log(LOG_INFO, "%s", sbuf_data(sb));
825 	sbuf_delete(sb);
826 	exit1(td, 0, sig);
827 	return (0);
828 }
829 
830 #ifdef COMPAT_43
831 /*
832  * The dirty work is handled by kern_wait().
833  */
834 int
835 owait(struct thread *td, struct owait_args *uap __unused)
836 {
837 	int error, status;
838 
839 	error = kern_wait(td, WAIT_ANY, &status, 0, NULL);
840 	if (error == 0)
841 		td->td_retval[1] = status;
842 	return (error);
843 }
844 #endif /* COMPAT_43 */
845 
846 /*
847  * The dirty work is handled by kern_wait().
848  */
849 int
850 sys_wait4(struct thread *td, struct wait4_args *uap)
851 {
852 	struct rusage ru, *rup;
853 	int error, status;
854 
855 	if (uap->rusage != NULL)
856 		rup = &ru;
857 	else
858 		rup = NULL;
859 	error = kern_wait(td, uap->pid, &status, uap->options, rup);
860 	if (uap->status != NULL && error == 0 && td->td_retval[0] != 0)
861 		error = copyout(&status, uap->status, sizeof(status));
862 	if (uap->rusage != NULL && error == 0 && td->td_retval[0] != 0)
863 		error = copyout(&ru, uap->rusage, sizeof(struct rusage));
864 	return (error);
865 }
866 
867 int
868 sys_wait6(struct thread *td, struct wait6_args *uap)
869 {
870 	struct __wrusage wru, *wrup;
871 	siginfo_t si, *sip;
872 	idtype_t idtype;
873 	id_t id;
874 	int error, status;
875 
876 	idtype = uap->idtype;
877 	id = uap->id;
878 
879 	if (uap->wrusage != NULL)
880 		wrup = &wru;
881 	else
882 		wrup = NULL;
883 
884 	if (uap->info != NULL) {
885 		sip = &si;
886 		bzero(sip, sizeof(*sip));
887 	} else
888 		sip = NULL;
889 
890 	/*
891 	 *  We expect all callers of wait6() to know about WEXITED and
892 	 *  WTRAPPED.
893 	 */
894 	error = kern_wait6(td, idtype, id, &status, uap->options, wrup, sip);
895 
896 	if (uap->status != NULL && error == 0 && td->td_retval[0] != 0)
897 		error = copyout(&status, uap->status, sizeof(status));
898 	if (uap->wrusage != NULL && error == 0 && td->td_retval[0] != 0)
899 		error = copyout(&wru, uap->wrusage, sizeof(wru));
900 	if (uap->info != NULL && error == 0)
901 		error = copyout(&si, uap->info, sizeof(si));
902 	return (error);
903 }
904 
905 /*
906  * Reap the remains of a zombie process and optionally return status and
907  * rusage.  Asserts and will release both the proctree_lock and the process
908  * lock as part of its work.
909  */
910 void
911 proc_reap(struct thread *td, struct proc *p, int *status, int options)
912 {
913 	struct proc *q, *t;
914 
915 	sx_assert(&proctree_lock, SA_XLOCKED);
916 	PROC_LOCK_ASSERT(p, MA_OWNED);
917 	KASSERT(p->p_state == PRS_ZOMBIE, ("proc_reap: !PRS_ZOMBIE"));
918 
919 	mtx_spin_wait_unlocked(&p->p_slock);
920 
921 	q = td->td_proc;
922 
923 	if (status)
924 		*status = KW_EXITCODE(p->p_xexit, p->p_xsig);
925 	if (options & WNOWAIT) {
926 		/*
927 		 *  Only poll, returning the status.  Caller does not wish to
928 		 * release the proc struct just yet.
929 		 */
930 		PROC_UNLOCK(p);
931 		sx_xunlock(&proctree_lock);
932 		return;
933 	}
934 
935 	PROC_LOCK(q);
936 	sigqueue_take(p->p_ksi);
937 	PROC_UNLOCK(q);
938 
939 	/*
940 	 * If we got the child via a ptrace 'attach', we need to give it back
941 	 * to the old parent.
942 	 */
943 	if (p->p_oppid != p->p_pptr->p_pid) {
944 		PROC_UNLOCK(p);
945 		t = proc_realparent(p);
946 		PROC_LOCK(t);
947 		PROC_LOCK(p);
948 		CTR2(KTR_PTRACE,
949 		    "wait: traced child %d moved back to parent %d", p->p_pid,
950 		    t->p_pid);
951 		proc_reparent(p, t, false);
952 		PROC_UNLOCK(p);
953 		pksignal(t, SIGCHLD, p->p_ksi);
954 		wakeup(t);
955 		cv_broadcast(&p->p_pwait);
956 		PROC_UNLOCK(t);
957 		sx_xunlock(&proctree_lock);
958 		return;
959 	}
960 	PROC_UNLOCK(p);
961 
962 	/*
963 	 * Remove other references to this process to ensure we have an
964 	 * exclusive reference.
965 	 */
966 	sx_xlock(PIDHASHLOCK(p->p_pid));
967 	LIST_REMOVE(p, p_hash);
968 	sx_xunlock(PIDHASHLOCK(p->p_pid));
969 	LIST_REMOVE(p, p_sibling);
970 	reaper_abandon_children(p, true);
971 	reaper_clear(p);
972 	PROC_LOCK(p);
973 	proc_clear_orphan(p);
974 	PROC_UNLOCK(p);
975 	leavepgrp(p);
976 	if (p->p_procdesc != NULL)
977 		procdesc_reap(p);
978 	sx_xunlock(&proctree_lock);
979 
980 	proc_id_clear(PROC_ID_PID, p->p_pid);
981 
982 	PROC_LOCK(p);
983 	knlist_detach(p->p_klist);
984 	p->p_klist = NULL;
985 	PROC_UNLOCK(p);
986 
987 	/*
988 	 * Removal from allproc list and process group list paired with
989 	 * PROC_LOCK which was executed during that time should guarantee
990 	 * nothing can reach this process anymore. As such further locking
991 	 * is unnecessary.
992 	 */
993 	p->p_xexit = p->p_xsig = 0;		/* XXX: why? */
994 
995 	PROC_LOCK(q);
996 	ruadd(&q->p_stats->p_cru, &q->p_crux, &p->p_ru, &p->p_rux);
997 	PROC_UNLOCK(q);
998 
999 	/*
1000 	 * Decrement the count of procs running with this uid.
1001 	 */
1002 	(void)chgproccnt(p->p_ucred->cr_ruidinfo, -1, 0);
1003 
1004 	/*
1005 	 * Destroy resource accounting information associated with the process.
1006 	 */
1007 #ifdef RACCT
1008 	if (racct_enable) {
1009 		PROC_LOCK(p);
1010 		racct_sub(p, RACCT_NPROC, 1);
1011 		PROC_UNLOCK(p);
1012 	}
1013 #endif
1014 	racct_proc_exit(p);
1015 
1016 	/*
1017 	 * Free credentials, arguments, and sigacts.
1018 	 */
1019 	proc_unset_cred(p);
1020 	pargs_drop(p->p_args);
1021 	p->p_args = NULL;
1022 	sigacts_free(p->p_sigacts);
1023 	p->p_sigacts = NULL;
1024 
1025 	/*
1026 	 * Do any thread-system specific cleanups.
1027 	 */
1028 	thread_wait(p);
1029 
1030 	/*
1031 	 * Give vm and machine-dependent layer a chance to free anything that
1032 	 * cpu_exit couldn't release while still running in process context.
1033 	 */
1034 	vm_waitproc(p);
1035 #ifdef MAC
1036 	mac_proc_destroy(p);
1037 #endif
1038 
1039 	KASSERT(FIRST_THREAD_IN_PROC(p),
1040 	    ("proc_reap: no residual thread!"));
1041 	uma_zfree(proc_zone, p);
1042 	atomic_add_int(&nprocs, -1);
1043 }
1044 
1045 static int
1046 proc_to_reap(struct thread *td, struct proc *p, idtype_t idtype, id_t id,
1047     int *status, int options, struct __wrusage *wrusage, siginfo_t *siginfo,
1048     int check_only)
1049 {
1050 	struct rusage *rup;
1051 
1052 	sx_assert(&proctree_lock, SA_XLOCKED);
1053 
1054 	PROC_LOCK(p);
1055 
1056 	switch (idtype) {
1057 	case P_ALL:
1058 		if (p->p_procdesc == NULL ||
1059 		   (p->p_pptr == td->td_proc &&
1060 		   (p->p_flag & P_TRACED) != 0)) {
1061 			break;
1062 		}
1063 
1064 		PROC_UNLOCK(p);
1065 		return (0);
1066 	case P_PID:
1067 		if (p->p_pid != (pid_t)id) {
1068 			PROC_UNLOCK(p);
1069 			return (0);
1070 		}
1071 		break;
1072 	case P_PGID:
1073 		if (p->p_pgid != (pid_t)id) {
1074 			PROC_UNLOCK(p);
1075 			return (0);
1076 		}
1077 		break;
1078 	case P_SID:
1079 		if (p->p_session->s_sid != (pid_t)id) {
1080 			PROC_UNLOCK(p);
1081 			return (0);
1082 		}
1083 		break;
1084 	case P_UID:
1085 		if (p->p_ucred->cr_uid != (uid_t)id) {
1086 			PROC_UNLOCK(p);
1087 			return (0);
1088 		}
1089 		break;
1090 	case P_GID:
1091 		if (p->p_ucred->cr_gid != (gid_t)id) {
1092 			PROC_UNLOCK(p);
1093 			return (0);
1094 		}
1095 		break;
1096 	case P_JAILID:
1097 		if (p->p_ucred->cr_prison->pr_id != (int)id) {
1098 			PROC_UNLOCK(p);
1099 			return (0);
1100 		}
1101 		break;
1102 	/*
1103 	 * It seems that the thread structures get zeroed out
1104 	 * at process exit.  This makes it impossible to
1105 	 * support P_SETID, P_CID or P_CPUID.
1106 	 */
1107 	default:
1108 		PROC_UNLOCK(p);
1109 		return (0);
1110 	}
1111 
1112 	if (p_canwait(td, p)) {
1113 		PROC_UNLOCK(p);
1114 		return (0);
1115 	}
1116 
1117 	if (((options & WEXITED) == 0) && (p->p_state == PRS_ZOMBIE)) {
1118 		PROC_UNLOCK(p);
1119 		return (0);
1120 	}
1121 
1122 	/*
1123 	 * This special case handles a kthread spawned by linux_clone
1124 	 * (see linux_misc.c).  The linux_wait4 and linux_waitpid
1125 	 * functions need to be able to distinguish between waiting
1126 	 * on a process and waiting on a thread.  It is a thread if
1127 	 * p_sigparent is not SIGCHLD, and the WLINUXCLONE option
1128 	 * signifies we want to wait for threads and not processes.
1129 	 */
1130 	if ((p->p_sigparent != SIGCHLD) ^
1131 	    ((options & WLINUXCLONE) != 0)) {
1132 		PROC_UNLOCK(p);
1133 		return (0);
1134 	}
1135 
1136 	if (siginfo != NULL) {
1137 		bzero(siginfo, sizeof(*siginfo));
1138 		siginfo->si_errno = 0;
1139 
1140 		/*
1141 		 * SUSv4 requires that the si_signo value is always
1142 		 * SIGCHLD. Obey it despite the rfork(2) interface
1143 		 * allows to request other signal for child exit
1144 		 * notification.
1145 		 */
1146 		siginfo->si_signo = SIGCHLD;
1147 
1148 		/*
1149 		 *  This is still a rough estimate.  We will fix the
1150 		 *  cases TRAPPED, STOPPED, and CONTINUED later.
1151 		 */
1152 		if (WCOREDUMP(p->p_xsig)) {
1153 			siginfo->si_code = CLD_DUMPED;
1154 			siginfo->si_status = WTERMSIG(p->p_xsig);
1155 		} else if (WIFSIGNALED(p->p_xsig)) {
1156 			siginfo->si_code = CLD_KILLED;
1157 			siginfo->si_status = WTERMSIG(p->p_xsig);
1158 		} else {
1159 			siginfo->si_code = CLD_EXITED;
1160 			siginfo->si_status = p->p_xexit;
1161 		}
1162 
1163 		siginfo->si_pid = p->p_pid;
1164 		siginfo->si_uid = p->p_ucred->cr_uid;
1165 
1166 		/*
1167 		 * The si_addr field would be useful additional
1168 		 * detail, but apparently the PC value may be lost
1169 		 * when we reach this point.  bzero() above sets
1170 		 * siginfo->si_addr to NULL.
1171 		 */
1172 	}
1173 
1174 	/*
1175 	 * There should be no reason to limit resources usage info to
1176 	 * exited processes only.  A snapshot about any resources used
1177 	 * by a stopped process may be exactly what is needed.
1178 	 */
1179 	if (wrusage != NULL) {
1180 		rup = &wrusage->wru_self;
1181 		*rup = p->p_ru;
1182 		PROC_STATLOCK(p);
1183 		calcru(p, &rup->ru_utime, &rup->ru_stime);
1184 		PROC_STATUNLOCK(p);
1185 
1186 		rup = &wrusage->wru_children;
1187 		*rup = p->p_stats->p_cru;
1188 		calccru(p, &rup->ru_utime, &rup->ru_stime);
1189 	}
1190 
1191 	if (p->p_state == PRS_ZOMBIE && !check_only) {
1192 		proc_reap(td, p, status, options);
1193 		return (-1);
1194 	}
1195 	return (1);
1196 }
1197 
1198 int
1199 kern_wait(struct thread *td, pid_t pid, int *status, int options,
1200     struct rusage *rusage)
1201 {
1202 	struct __wrusage wru, *wrup;
1203 	idtype_t idtype;
1204 	id_t id;
1205 	int ret;
1206 
1207 	/*
1208 	 * Translate the special pid values into the (idtype, pid)
1209 	 * pair for kern_wait6.  The WAIT_MYPGRP case is handled by
1210 	 * kern_wait6() on its own.
1211 	 */
1212 	if (pid == WAIT_ANY) {
1213 		idtype = P_ALL;
1214 		id = 0;
1215 	} else if (pid < 0) {
1216 		idtype = P_PGID;
1217 		id = (id_t)-pid;
1218 	} else {
1219 		idtype = P_PID;
1220 		id = (id_t)pid;
1221 	}
1222 
1223 	if (rusage != NULL)
1224 		wrup = &wru;
1225 	else
1226 		wrup = NULL;
1227 
1228 	/*
1229 	 * For backward compatibility we implicitly add flags WEXITED
1230 	 * and WTRAPPED here.
1231 	 */
1232 	options |= WEXITED | WTRAPPED;
1233 	ret = kern_wait6(td, idtype, id, status, options, wrup, NULL);
1234 	if (rusage != NULL)
1235 		*rusage = wru.wru_self;
1236 	return (ret);
1237 }
1238 
1239 static void
1240 report_alive_proc(struct thread *td, struct proc *p, siginfo_t *siginfo,
1241     int *status, int options, int si_code)
1242 {
1243 	bool cont;
1244 
1245 	PROC_LOCK_ASSERT(p, MA_OWNED);
1246 	sx_assert(&proctree_lock, SA_XLOCKED);
1247 	MPASS(si_code == CLD_TRAPPED || si_code == CLD_STOPPED ||
1248 	    si_code == CLD_CONTINUED);
1249 
1250 	cont = si_code == CLD_CONTINUED;
1251 	if ((options & WNOWAIT) == 0) {
1252 		if (cont)
1253 			p->p_flag &= ~P_CONTINUED;
1254 		else
1255 			p->p_flag |= P_WAITED;
1256 		if (kern_wait_dequeue_sigchld &&
1257 		    (td->td_proc->p_sysent->sv_flags & SV_SIG_WAITNDQ) == 0) {
1258 			PROC_LOCK(td->td_proc);
1259 			sigqueue_take(p->p_ksi);
1260 			PROC_UNLOCK(td->td_proc);
1261 		}
1262 	}
1263 	sx_xunlock(&proctree_lock);
1264 	if (siginfo != NULL) {
1265 		siginfo->si_code = si_code;
1266 		siginfo->si_status = cont ? SIGCONT : p->p_xsig;
1267 	}
1268 	if (status != NULL)
1269 		*status = cont ? SIGCONT : W_STOPCODE(p->p_xsig);
1270 	PROC_UNLOCK(p);
1271 	td->td_retval[0] = p->p_pid;
1272 }
1273 
1274 int
1275 kern_wait6(struct thread *td, idtype_t idtype, id_t id, int *status,
1276     int options, struct __wrusage *wrusage, siginfo_t *siginfo)
1277 {
1278 	struct proc *p, *q;
1279 	pid_t pid;
1280 	int error, nfound, ret;
1281 	bool report;
1282 
1283 	AUDIT_ARG_VALUE((int)idtype);	/* XXX - This is likely wrong! */
1284 	AUDIT_ARG_PID((pid_t)id);	/* XXX - This may be wrong! */
1285 	AUDIT_ARG_VALUE(options);
1286 
1287 	q = td->td_proc;
1288 
1289 	if ((pid_t)id == WAIT_MYPGRP && (idtype == P_PID || idtype == P_PGID)) {
1290 		PROC_LOCK(q);
1291 		id = (id_t)q->p_pgid;
1292 		PROC_UNLOCK(q);
1293 		idtype = P_PGID;
1294 	}
1295 
1296 	/* If we don't know the option, just return. */
1297 	if ((options & ~(WUNTRACED | WNOHANG | WCONTINUED | WNOWAIT |
1298 	    WEXITED | WTRAPPED | WLINUXCLONE)) != 0)
1299 		return (EINVAL);
1300 	if ((options & (WEXITED | WUNTRACED | WCONTINUED | WTRAPPED)) == 0) {
1301 		/*
1302 		 * We will be unable to find any matching processes,
1303 		 * because there are no known events to look for.
1304 		 * Prefer to return error instead of blocking
1305 		 * indefinitely.
1306 		 */
1307 		return (EINVAL);
1308 	}
1309 
1310 loop:
1311 	if (q->p_flag & P_STATCHILD) {
1312 		PROC_LOCK(q);
1313 		q->p_flag &= ~P_STATCHILD;
1314 		PROC_UNLOCK(q);
1315 	}
1316 	sx_xlock(&proctree_lock);
1317 loop_locked:
1318 	nfound = 0;
1319 	LIST_FOREACH(p, &q->p_children, p_sibling) {
1320 		pid = p->p_pid;
1321 		ret = proc_to_reap(td, p, idtype, id, status, options,
1322 		    wrusage, siginfo, 0);
1323 		if (ret == 0)
1324 			continue;
1325 		else if (ret != 1) {
1326 			td->td_retval[0] = pid;
1327 			return (0);
1328 		}
1329 
1330 		nfound++;
1331 		PROC_LOCK_ASSERT(p, MA_OWNED);
1332 
1333 		if ((options & WTRAPPED) != 0 &&
1334 		    (p->p_flag & P_TRACED) != 0) {
1335 			PROC_SLOCK(p);
1336 			report =
1337 			    ((p->p_flag & (P_STOPPED_TRACE | P_STOPPED_SIG)) &&
1338 			    p->p_suspcount == p->p_numthreads &&
1339 			    (p->p_flag & P_WAITED) == 0);
1340 			PROC_SUNLOCK(p);
1341 			if (report) {
1342 			CTR4(KTR_PTRACE,
1343 			    "wait: returning trapped pid %d status %#x "
1344 			    "(xstat %d) xthread %d",
1345 			    p->p_pid, W_STOPCODE(p->p_xsig), p->p_xsig,
1346 			    p->p_xthread != NULL ?
1347 			    p->p_xthread->td_tid : -1);
1348 				report_alive_proc(td, p, siginfo, status,
1349 				    options, CLD_TRAPPED);
1350 				return (0);
1351 			}
1352 		}
1353 		if ((options & WUNTRACED) != 0 &&
1354 		    (p->p_flag & P_STOPPED_SIG) != 0) {
1355 			PROC_SLOCK(p);
1356 			report = (p->p_suspcount == p->p_numthreads &&
1357 			    ((p->p_flag & P_WAITED) == 0));
1358 			PROC_SUNLOCK(p);
1359 			if (report) {
1360 				report_alive_proc(td, p, siginfo, status,
1361 				    options, CLD_STOPPED);
1362 				return (0);
1363 			}
1364 		}
1365 		if ((options & WCONTINUED) != 0 &&
1366 		    (p->p_flag & P_CONTINUED) != 0) {
1367 			report_alive_proc(td, p, siginfo, status, options,
1368 			    CLD_CONTINUED);
1369 			return (0);
1370 		}
1371 		PROC_UNLOCK(p);
1372 	}
1373 
1374 	/*
1375 	 * Look in the orphans list too, to allow the parent to
1376 	 * collect it's child exit status even if child is being
1377 	 * debugged.
1378 	 *
1379 	 * Debugger detaches from the parent upon successful
1380 	 * switch-over from parent to child.  At this point due to
1381 	 * re-parenting the parent loses the child to debugger and a
1382 	 * wait4(2) call would report that it has no children to wait
1383 	 * for.  By maintaining a list of orphans we allow the parent
1384 	 * to successfully wait until the child becomes a zombie.
1385 	 */
1386 	if (nfound == 0) {
1387 		LIST_FOREACH(p, &q->p_orphans, p_orphan) {
1388 			ret = proc_to_reap(td, p, idtype, id, NULL, options,
1389 			    NULL, NULL, 1);
1390 			if (ret != 0) {
1391 				KASSERT(ret != -1, ("reaped an orphan (pid %d)",
1392 				    (int)td->td_retval[0]));
1393 				PROC_UNLOCK(p);
1394 				nfound++;
1395 				break;
1396 			}
1397 		}
1398 	}
1399 	if (nfound == 0) {
1400 		sx_xunlock(&proctree_lock);
1401 		return (ECHILD);
1402 	}
1403 	if (options & WNOHANG) {
1404 		sx_xunlock(&proctree_lock);
1405 		td->td_retval[0] = 0;
1406 		return (0);
1407 	}
1408 	PROC_LOCK(q);
1409 	if (q->p_flag & P_STATCHILD) {
1410 		q->p_flag &= ~P_STATCHILD;
1411 		PROC_UNLOCK(q);
1412 		goto loop_locked;
1413 	}
1414 	sx_xunlock(&proctree_lock);
1415 	error = msleep(q, &q->p_mtx, PWAIT | PCATCH | PDROP, "wait", 0);
1416 	if (error)
1417 		return (error);
1418 	goto loop;
1419 }
1420 
1421 void
1422 proc_add_orphan(struct proc *child, struct proc *parent)
1423 {
1424 
1425 	sx_assert(&proctree_lock, SX_XLOCKED);
1426 	KASSERT((child->p_flag & P_TRACED) != 0,
1427 	    ("proc_add_orphan: not traced"));
1428 
1429 	if (LIST_EMPTY(&parent->p_orphans)) {
1430 		child->p_treeflag |= P_TREE_FIRST_ORPHAN;
1431 		LIST_INSERT_HEAD(&parent->p_orphans, child, p_orphan);
1432 	} else {
1433 		LIST_INSERT_AFTER(LIST_FIRST(&parent->p_orphans),
1434 		    child, p_orphan);
1435 	}
1436 	child->p_treeflag |= P_TREE_ORPHANED;
1437 }
1438 
1439 /*
1440  * Make process 'parent' the new parent of process 'child'.
1441  * Must be called with an exclusive hold of proctree lock.
1442  */
1443 void
1444 proc_reparent(struct proc *child, struct proc *parent, bool set_oppid)
1445 {
1446 
1447 	sx_assert(&proctree_lock, SX_XLOCKED);
1448 	PROC_LOCK_ASSERT(child, MA_OWNED);
1449 	if (child->p_pptr == parent)
1450 		return;
1451 
1452 	PROC_LOCK(child->p_pptr);
1453 	sigqueue_take(child->p_ksi);
1454 	PROC_UNLOCK(child->p_pptr);
1455 	LIST_REMOVE(child, p_sibling);
1456 	LIST_INSERT_HEAD(&parent->p_children, child, p_sibling);
1457 
1458 	proc_clear_orphan(child);
1459 	if ((child->p_flag & P_TRACED) != 0) {
1460 		proc_add_orphan(child, child->p_pptr);
1461 	}
1462 
1463 	child->p_pptr = parent;
1464 	if (set_oppid)
1465 		child->p_oppid = parent->p_pid;
1466 }
1467