xref: /freebsd/sys/kern/kern_exit.c (revision acb1f1269c6f4ff89a0d28ba742f6687e9ef779d)
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/umtx.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 /*
199  * exit -- death of process.
200  */
201 void
202 sys_sys_exit(struct thread *td, struct sys_exit_args *uap)
203 {
204 
205 	exit1(td, uap->rval, 0);
206 	/* NOTREACHED */
207 }
208 
209 /*
210  * Exit: deallocate address space and other resources, change proc state to
211  * zombie, and unlink proc from allproc and parent's lists.  Save exit status
212  * and rusage for wait().  Check for child processes and orphan them.
213  */
214 void
215 exit1(struct thread *td, int rval, int signo)
216 {
217 	struct proc *p, *nq, *q, *t;
218 	struct thread *tdt;
219 	ksiginfo_t *ksi, *ksi1;
220 	int signal_parent;
221 
222 	mtx_assert(&Giant, MA_NOTOWNED);
223 	KASSERT(rval == 0 || signo == 0, ("exit1 rv %d sig %d", rval, signo));
224 
225 	p = td->td_proc;
226 	/*
227 	 * XXX in case we're rebooting we just let init die in order to
228 	 * work around an unsolved stack overflow seen very late during
229 	 * shutdown on sparc64 when the gmirror worker process exists.
230 	 * XXX what to do now that sparc64 is gone... remove if?
231 	 */
232 	if (p == initproc && rebooting == 0) {
233 		printf("init died (signal %d, exit %d)\n", signo, rval);
234 		panic("Going nowhere without my init!");
235 	}
236 
237 	/*
238 	 * Deref SU mp, since the thread does not return to userspace.
239 	 */
240 	td_softdep_cleanup(td);
241 
242 	/*
243 	 * MUST abort all other threads before proceeding past here.
244 	 */
245 	PROC_LOCK(p);
246 	/*
247 	 * First check if some other thread or external request got
248 	 * here before us.  If so, act appropriately: exit or suspend.
249 	 * We must ensure that stop requests are handled before we set
250 	 * P_WEXIT.
251 	 */
252 	thread_suspend_check(0);
253 	while (p->p_flag & P_HADTHREADS) {
254 		/*
255 		 * Kill off the other threads. This requires
256 		 * some co-operation from other parts of the kernel
257 		 * so it may not be instantaneous.  With this state set
258 		 * any thread entering the kernel from userspace will
259 		 * thread_exit() in trap().  Any thread attempting to
260 		 * sleep will return immediately with EINTR or EWOULDBLOCK
261 		 * which will hopefully force them to back out to userland
262 		 * freeing resources as they go.  Any thread attempting
263 		 * to return to userland will thread_exit() from userret().
264 		 * thread_exit() will unsuspend us when the last of the
265 		 * other threads exits.
266 		 * If there is already a thread singler after resumption,
267 		 * calling thread_single will fail; in that case, we just
268 		 * re-check all suspension request, the thread should
269 		 * either be suspended there or exit.
270 		 */
271 		if (!thread_single(p, SINGLE_EXIT))
272 			/*
273 			 * All other activity in this process is now
274 			 * stopped.  Threading support has been turned
275 			 * off.
276 			 */
277 			break;
278 		/*
279 		 * Recheck for new stop or suspend requests which
280 		 * might appear while process lock was dropped in
281 		 * thread_single().
282 		 */
283 		thread_suspend_check(0);
284 	}
285 	KASSERT(p->p_numthreads == 1,
286 	    ("exit1: proc %p exiting with %d threads", p, p->p_numthreads));
287 	racct_sub(p, RACCT_NTHR, 1);
288 
289 	/* Let event handler change exit status */
290 	p->p_xexit = rval;
291 	p->p_xsig = signo;
292 
293 	/*
294 	 * Ignore any pending request to stop due to a stop signal.
295 	 * Once P_WEXIT is set, future requests will be ignored as
296 	 * well.
297 	 */
298 	p->p_flag &= ~P_STOPPED_SIG;
299 	KASSERT(!P_SHOULDSTOP(p), ("exiting process is stopped"));
300 
301 	/* Note that we are exiting. */
302 	p->p_flag |= P_WEXIT;
303 
304 	/*
305 	 * Wait for any processes that have a hold on our vmspace to
306 	 * release their reference.
307 	 */
308 	while (p->p_lock > 0)
309 		msleep(&p->p_lock, &p->p_mtx, PWAIT, "exithold", 0);
310 
311 	PROC_UNLOCK(p);
312 	/* Drain the limit callout while we don't have the proc locked */
313 	callout_drain(&p->p_limco);
314 
315 #ifdef AUDIT
316 	/*
317 	 * The Sun BSM exit token contains two components: an exit status as
318 	 * passed to exit(), and a return value to indicate what sort of exit
319 	 * it was.  The exit status is WEXITSTATUS(rv), but it's not clear
320 	 * what the return value is.
321 	 */
322 	AUDIT_ARG_EXIT(rval, 0);
323 	AUDIT_SYSCALL_EXIT(0, td);
324 #endif
325 
326 	/* Are we a task leader with peers? */
327 	if (p->p_peers != NULL && p == p->p_leader) {
328 		mtx_lock(&ppeers_lock);
329 		q = p->p_peers;
330 		while (q != NULL) {
331 			PROC_LOCK(q);
332 			kern_psignal(q, SIGKILL);
333 			PROC_UNLOCK(q);
334 			q = q->p_peers;
335 		}
336 		while (p->p_peers != NULL)
337 			msleep(p, &ppeers_lock, PWAIT, "exit1", 0);
338 		mtx_unlock(&ppeers_lock);
339 	}
340 
341 	itimers_exit(p);
342 
343 	if (p->p_sysent->sv_onexit != NULL)
344 		p->p_sysent->sv_onexit(p);
345 
346 	/*
347 	 * Check if any loadable modules need anything done at process exit.
348 	 * E.g. SYSV IPC stuff.
349 	 * Event handler could change exit status.
350 	 * XXX what if one of these generates an error?
351 	 */
352 	EVENTHANDLER_DIRECT_INVOKE(process_exit, p);
353 
354 	/*
355 	 * If parent is waiting for us to exit or exec,
356 	 * P_PPWAIT is set; we will wakeup the parent below.
357 	 */
358 	PROC_LOCK(p);
359 	stopprofclock(p);
360 	p->p_ptevents = 0;
361 
362 	/*
363 	 * Stop the real interval timer.  If the handler is currently
364 	 * executing, prevent it from rearming itself and let it finish.
365 	 */
366 	if (timevalisset(&p->p_realtimer.it_value) &&
367 	    _callout_stop_safe(&p->p_itcallout, CS_EXECUTING, NULL) == 0) {
368 		timevalclear(&p->p_realtimer.it_interval);
369 		msleep(&p->p_itcallout, &p->p_mtx, PWAIT, "ritwait", 0);
370 		KASSERT(!timevalisset(&p->p_realtimer.it_value),
371 		    ("realtime timer is still armed"));
372 	}
373 
374 	PROC_UNLOCK(p);
375 
376 	umtx_thread_exit(td);
377 	seltdfini(td);
378 
379 	/*
380 	 * Reset any sigio structures pointing to us as a result of
381 	 * F_SETOWN with our pid.  The P_WEXIT flag interlocks with fsetown().
382 	 */
383 	funsetownlst(&p->p_sigiolst);
384 
385 	/*
386 	 * Close open files and release open-file table.
387 	 * This may block!
388 	 */
389 	pdescfree(td);
390 	fdescfree(td);
391 
392 	/*
393 	 * If this thread tickled GEOM, we need to wait for the giggling to
394 	 * stop before we return to userland
395 	 */
396 	if (td->td_pflags & TDP_GEOM)
397 		g_waitidle();
398 
399 	/*
400 	 * Remove ourself from our leader's peer list and wake our leader.
401 	 */
402 	if (p->p_leader->p_peers != NULL) {
403 		mtx_lock(&ppeers_lock);
404 		if (p->p_leader->p_peers != NULL) {
405 			q = p->p_leader;
406 			while (q->p_peers != p)
407 				q = q->p_peers;
408 			q->p_peers = p->p_peers;
409 			wakeup(p->p_leader);
410 		}
411 		mtx_unlock(&ppeers_lock);
412 	}
413 
414 	vmspace_exit(td);
415 	(void)acct_process(td);
416 
417 #ifdef KTRACE
418 	ktrprocexit(td);
419 #endif
420 	/*
421 	 * Release reference to text vnode
422 	 */
423 	if (p->p_textvp != NULL) {
424 		vrele(p->p_textvp);
425 		p->p_textvp = NULL;
426 	}
427 
428 	/*
429 	 * Release our limits structure.
430 	 */
431 	lim_free(p->p_limit);
432 	p->p_limit = NULL;
433 
434 	tidhash_remove(td);
435 
436 	/*
437 	 * Call machine-dependent code to release any
438 	 * machine-dependent resources other than the address space.
439 	 * The address space is released by "vmspace_exitfree(p)" in
440 	 * vm_waitproc().
441 	 */
442 	cpu_exit(td);
443 
444 	WITNESS_WARN(WARN_PANIC, NULL, "process (pid %d) exiting", p->p_pid);
445 
446 	/*
447 	 * Remove from allproc. It still sits in the hash.
448 	 */
449 	sx_xlock(&allproc_lock);
450 	LIST_REMOVE(p, p_list);
451 
452 #ifdef DDB
453 	/*
454 	 * Used by ddb's 'ps' command to find this process via the
455 	 * pidhash.
456 	 */
457 	p->p_list.le_prev = NULL;
458 #endif
459 	sx_xunlock(&allproc_lock);
460 
461 	sx_xlock(&proctree_lock);
462 	PROC_LOCK(p);
463 	p->p_flag &= ~(P_TRACED | P_PPWAIT | P_PPTRACE);
464 	PROC_UNLOCK(p);
465 
466 	/*
467 	 * killjobc() might drop and re-acquire proctree_lock to
468 	 * revoke control tty if exiting process was a session leader.
469 	 */
470 	killjobc();
471 
472 	/*
473 	 * Reparent all children processes:
474 	 * - traced ones to the original parent (or init if we are that parent)
475 	 * - the rest to init
476 	 */
477 	q = LIST_FIRST(&p->p_children);
478 	if (q != NULL)		/* only need this if any child is S_ZOMB */
479 		wakeup(q->p_reaper);
480 	for (; q != NULL; q = nq) {
481 		nq = LIST_NEXT(q, p_sibling);
482 		ksi = ksiginfo_alloc(TRUE);
483 		PROC_LOCK(q);
484 		q->p_sigparent = SIGCHLD;
485 
486 		if ((q->p_flag & P_TRACED) == 0) {
487 			proc_reparent(q, q->p_reaper, true);
488 			if (q->p_state == PRS_ZOMBIE) {
489 				/*
490 				 * Inform reaper about the reparented
491 				 * zombie, since wait(2) has something
492 				 * new to report.  Guarantee queueing
493 				 * of the SIGCHLD signal, similar to
494 				 * the _exit() behaviour, by providing
495 				 * our ksiginfo.  Ksi is freed by the
496 				 * signal delivery.
497 				 */
498 				if (q->p_ksi == NULL) {
499 					ksi1 = NULL;
500 				} else {
501 					ksiginfo_copy(q->p_ksi, ksi);
502 					ksi->ksi_flags |= KSI_INS;
503 					ksi1 = ksi;
504 					ksi = NULL;
505 				}
506 				PROC_LOCK(q->p_reaper);
507 				pksignal(q->p_reaper, SIGCHLD, ksi1);
508 				PROC_UNLOCK(q->p_reaper);
509 			} else if (q->p_pdeathsig > 0) {
510 				/*
511 				 * The child asked to received a signal
512 				 * when we exit.
513 				 */
514 				kern_psignal(q, q->p_pdeathsig);
515 			}
516 		} else {
517 			/*
518 			 * Traced processes are killed by default
519 			 * since their existence means someone is
520 			 * screwing up.
521 			 */
522 			t = proc_realparent(q);
523 			if (t == p) {
524 				proc_reparent(q, q->p_reaper, true);
525 			} else {
526 				PROC_LOCK(t);
527 				proc_reparent(q, t, true);
528 				PROC_UNLOCK(t);
529 			}
530 			/*
531 			 * Since q was found on our children list, the
532 			 * proc_reparent() call moved q to the orphan
533 			 * list due to present P_TRACED flag. Clear
534 			 * orphan link for q now while q is locked.
535 			 */
536 			proc_clear_orphan(q);
537 			q->p_flag &= ~P_TRACED;
538 			q->p_flag2 &= ~P2_PTRACE_FSTP;
539 			q->p_ptevents = 0;
540 			p->p_xthread = NULL;
541 			FOREACH_THREAD_IN_PROC(q, tdt) {
542 				tdt->td_dbgflags &= ~(TDB_SUSPEND | TDB_XSIG |
543 				    TDB_FSTP);
544 				tdt->td_xsig = 0;
545 			}
546 			if (kern_kill_on_dbg_exit) {
547 				q->p_flag &= ~P_STOPPED_TRACE;
548 				kern_psignal(q, SIGKILL);
549 			} else if ((q->p_flag & (P_STOPPED_TRACE |
550 			    P_STOPPED_SIG)) != 0) {
551 				sigqueue_delete_proc(q, SIGTRAP);
552 				ptrace_unsuspend(q);
553 			}
554 		}
555 		PROC_UNLOCK(q);
556 		if (ksi != NULL)
557 			ksiginfo_free(ksi);
558 	}
559 
560 	/*
561 	 * Also get rid of our orphans.
562 	 */
563 	while ((q = LIST_FIRST(&p->p_orphans)) != NULL) {
564 		PROC_LOCK(q);
565 		KASSERT(q->p_oppid == p->p_pid,
566 		    ("orphan %p of %p has unexpected oppid %d", q, p,
567 		    q->p_oppid));
568 		q->p_oppid = q->p_reaper->p_pid;
569 
570 		/*
571 		 * If we are the real parent of this process
572 		 * but it has been reparented to a debugger, then
573 		 * check if it asked for a signal when we exit.
574 		 */
575 		if (q->p_pdeathsig > 0)
576 			kern_psignal(q, q->p_pdeathsig);
577 		CTR2(KTR_PTRACE, "exit: pid %d, clearing orphan %d", p->p_pid,
578 		    q->p_pid);
579 		proc_clear_orphan(q);
580 		PROC_UNLOCK(q);
581 	}
582 
583 #ifdef KDTRACE_HOOKS
584 	if (SDT_PROBES_ENABLED()) {
585 		int reason = CLD_EXITED;
586 		if (WCOREDUMP(signo))
587 			reason = CLD_DUMPED;
588 		else if (WIFSIGNALED(signo))
589 			reason = CLD_KILLED;
590 		SDT_PROBE1(proc, , , exit, reason);
591 	}
592 #endif
593 
594 	/* Save exit status. */
595 	PROC_LOCK(p);
596 	p->p_xthread = td;
597 
598 	if (p->p_sysent->sv_ontdexit != NULL)
599 		p->p_sysent->sv_ontdexit(td);
600 
601 #ifdef KDTRACE_HOOKS
602 	/*
603 	 * Tell the DTrace fasttrap provider about the exit if it
604 	 * has declared an interest.
605 	 */
606 	if (dtrace_fasttrap_exit)
607 		dtrace_fasttrap_exit(p);
608 #endif
609 
610 	/*
611 	 * Notify interested parties of our demise.
612 	 */
613 	KNOTE_LOCKED(p->p_klist, NOTE_EXIT);
614 
615 	/*
616 	 * If this is a process with a descriptor, we may not need to deliver
617 	 * a signal to the parent.  proctree_lock is held over
618 	 * procdesc_exit() to serialize concurrent calls to close() and
619 	 * exit().
620 	 */
621 	signal_parent = 0;
622 	if (p->p_procdesc == NULL || procdesc_exit(p)) {
623 		/*
624 		 * Notify parent that we're gone.  If parent has the
625 		 * PS_NOCLDWAIT flag set, or if the handler is set to SIG_IGN,
626 		 * notify process 1 instead (and hope it will handle this
627 		 * situation).
628 		 */
629 		PROC_LOCK(p->p_pptr);
630 		mtx_lock(&p->p_pptr->p_sigacts->ps_mtx);
631 		if (p->p_pptr->p_sigacts->ps_flag &
632 		    (PS_NOCLDWAIT | PS_CLDSIGIGN)) {
633 			struct proc *pp;
634 
635 			mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx);
636 			pp = p->p_pptr;
637 			PROC_UNLOCK(pp);
638 			proc_reparent(p, p->p_reaper, true);
639 			p->p_sigparent = SIGCHLD;
640 			PROC_LOCK(p->p_pptr);
641 
642 			/*
643 			 * Notify parent, so in case he was wait(2)ing or
644 			 * executing waitpid(2) with our pid, he will
645 			 * continue.
646 			 */
647 			wakeup(pp);
648 		} else
649 			mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx);
650 
651 		if (p->p_pptr == p->p_reaper || p->p_pptr == initproc) {
652 			signal_parent = 1;
653 		} else if (p->p_sigparent != 0) {
654 			if (p->p_sigparent == SIGCHLD) {
655 				signal_parent = 1;
656 			} else { /* LINUX thread */
657 				signal_parent = 2;
658 			}
659 		}
660 	} else
661 		PROC_LOCK(p->p_pptr);
662 	sx_xunlock(&proctree_lock);
663 
664 	if (signal_parent == 1) {
665 		childproc_exited(p);
666 	} else if (signal_parent == 2) {
667 		kern_psignal(p->p_pptr, p->p_sigparent);
668 	}
669 
670 	/* Tell the prison that we are gone. */
671 	prison_proc_free(p->p_ucred->cr_prison);
672 
673 	/*
674 	 * The state PRS_ZOMBIE prevents other proesses from sending
675 	 * signal to the process, to avoid memory leak, we free memory
676 	 * for signal queue at the time when the state is set.
677 	 */
678 	sigqueue_flush(&p->p_sigqueue);
679 	sigqueue_flush(&td->td_sigqueue);
680 
681 	/*
682 	 * We have to wait until after acquiring all locks before
683 	 * changing p_state.  We need to avoid all possible context
684 	 * switches (including ones from blocking on a mutex) while
685 	 * marked as a zombie.  We also have to set the zombie state
686 	 * before we release the parent process' proc lock to avoid
687 	 * a lost wakeup.  So, we first call wakeup, then we grab the
688 	 * sched lock, update the state, and release the parent process'
689 	 * proc lock.
690 	 */
691 	wakeup(p->p_pptr);
692 	cv_broadcast(&p->p_pwait);
693 	sched_exit(p->p_pptr, td);
694 	PROC_SLOCK(p);
695 	p->p_state = PRS_ZOMBIE;
696 	PROC_UNLOCK(p->p_pptr);
697 
698 	/*
699 	 * Save our children's rusage information in our exit rusage.
700 	 */
701 	PROC_STATLOCK(p);
702 	ruadd(&p->p_ru, &p->p_rux, &p->p_stats->p_cru, &p->p_crux);
703 	PROC_STATUNLOCK(p);
704 
705 	/*
706 	 * Make sure the scheduler takes this thread out of its tables etc.
707 	 * This will also release this thread's reference to the ucred.
708 	 * Other thread parts to release include pcb bits and such.
709 	 */
710 	thread_exit();
711 }
712 
713 #ifndef _SYS_SYSPROTO_H_
714 struct abort2_args {
715 	char *why;
716 	int nargs;
717 	void **args;
718 };
719 #endif
720 
721 int
722 sys_abort2(struct thread *td, struct abort2_args *uap)
723 {
724 	struct proc *p = td->td_proc;
725 	struct sbuf *sb;
726 	void *uargs[16];
727 	int error, i, sig;
728 
729 	/*
730 	 * Do it right now so we can log either proper call of abort2(), or
731 	 * note, that invalid argument was passed. 512 is big enough to
732 	 * handle 16 arguments' descriptions with additional comments.
733 	 */
734 	sb = sbuf_new(NULL, NULL, 512, SBUF_FIXEDLEN);
735 	sbuf_clear(sb);
736 	sbuf_printf(sb, "%s(pid %d uid %d) aborted: ",
737 	    p->p_comm, p->p_pid, td->td_ucred->cr_uid);
738 	/*
739 	 * Since we can't return from abort2(), send SIGKILL in cases, where
740 	 * abort2() was called improperly
741 	 */
742 	sig = SIGKILL;
743 	/* Prevent from DoSes from user-space. */
744 	if (uap->nargs < 0 || uap->nargs > 16)
745 		goto out;
746 	if (uap->nargs > 0) {
747 		if (uap->args == NULL)
748 			goto out;
749 		error = copyin(uap->args, uargs, uap->nargs * sizeof(void *));
750 		if (error != 0)
751 			goto out;
752 	}
753 	/*
754 	 * Limit size of 'reason' string to 128. Will fit even when
755 	 * maximal number of arguments was chosen to be logged.
756 	 */
757 	if (uap->why != NULL) {
758 		error = sbuf_copyin(sb, uap->why, 128);
759 		if (error < 0)
760 			goto out;
761 	} else {
762 		sbuf_printf(sb, "(null)");
763 	}
764 	if (uap->nargs > 0) {
765 		sbuf_printf(sb, "(");
766 		for (i = 0;i < uap->nargs; i++)
767 			sbuf_printf(sb, "%s%p", i == 0 ? "" : ", ", uargs[i]);
768 		sbuf_printf(sb, ")");
769 	}
770 	/*
771 	 * Final stage: arguments were proper, string has been
772 	 * successfully copied from userspace, and copying pointers
773 	 * from user-space succeed.
774 	 */
775 	sig = SIGABRT;
776 out:
777 	if (sig == SIGKILL) {
778 		sbuf_trim(sb);
779 		sbuf_printf(sb, " (Reason text inaccessible)");
780 	}
781 	sbuf_cat(sb, "\n");
782 	sbuf_finish(sb);
783 	log(LOG_INFO, "%s", sbuf_data(sb));
784 	sbuf_delete(sb);
785 	exit1(td, 0, sig);
786 	return (0);
787 }
788 
789 #ifdef COMPAT_43
790 /*
791  * The dirty work is handled by kern_wait().
792  */
793 int
794 owait(struct thread *td, struct owait_args *uap __unused)
795 {
796 	int error, status;
797 
798 	error = kern_wait(td, WAIT_ANY, &status, 0, NULL);
799 	if (error == 0)
800 		td->td_retval[1] = status;
801 	return (error);
802 }
803 #endif /* COMPAT_43 */
804 
805 /*
806  * The dirty work is handled by kern_wait().
807  */
808 int
809 sys_wait4(struct thread *td, struct wait4_args *uap)
810 {
811 	struct rusage ru, *rup;
812 	int error, status;
813 
814 	if (uap->rusage != NULL)
815 		rup = &ru;
816 	else
817 		rup = NULL;
818 	error = kern_wait(td, uap->pid, &status, uap->options, rup);
819 	if (uap->status != NULL && error == 0 && td->td_retval[0] != 0)
820 		error = copyout(&status, uap->status, sizeof(status));
821 	if (uap->rusage != NULL && error == 0 && td->td_retval[0] != 0)
822 		error = copyout(&ru, uap->rusage, sizeof(struct rusage));
823 	return (error);
824 }
825 
826 int
827 sys_wait6(struct thread *td, struct wait6_args *uap)
828 {
829 	struct __wrusage wru, *wrup;
830 	siginfo_t si, *sip;
831 	idtype_t idtype;
832 	id_t id;
833 	int error, status;
834 
835 	idtype = uap->idtype;
836 	id = uap->id;
837 
838 	if (uap->wrusage != NULL)
839 		wrup = &wru;
840 	else
841 		wrup = NULL;
842 
843 	if (uap->info != NULL) {
844 		sip = &si;
845 		bzero(sip, sizeof(*sip));
846 	} else
847 		sip = NULL;
848 
849 	/*
850 	 *  We expect all callers of wait6() to know about WEXITED and
851 	 *  WTRAPPED.
852 	 */
853 	error = kern_wait6(td, idtype, id, &status, uap->options, wrup, sip);
854 
855 	if (uap->status != NULL && error == 0 && td->td_retval[0] != 0)
856 		error = copyout(&status, uap->status, sizeof(status));
857 	if (uap->wrusage != NULL && error == 0 && td->td_retval[0] != 0)
858 		error = copyout(&wru, uap->wrusage, sizeof(wru));
859 	if (uap->info != NULL && error == 0)
860 		error = copyout(&si, uap->info, sizeof(si));
861 	return (error);
862 }
863 
864 /*
865  * Reap the remains of a zombie process and optionally return status and
866  * rusage.  Asserts and will release both the proctree_lock and the process
867  * lock as part of its work.
868  */
869 void
870 proc_reap(struct thread *td, struct proc *p, int *status, int options)
871 {
872 	struct proc *q, *t;
873 
874 	sx_assert(&proctree_lock, SA_XLOCKED);
875 	PROC_LOCK_ASSERT(p, MA_OWNED);
876 	KASSERT(p->p_state == PRS_ZOMBIE, ("proc_reap: !PRS_ZOMBIE"));
877 
878 	mtx_spin_wait_unlocked(&p->p_slock);
879 
880 	q = td->td_proc;
881 
882 	if (status)
883 		*status = KW_EXITCODE(p->p_xexit, p->p_xsig);
884 	if (options & WNOWAIT) {
885 		/*
886 		 *  Only poll, returning the status.  Caller does not wish to
887 		 * release the proc struct just yet.
888 		 */
889 		PROC_UNLOCK(p);
890 		sx_xunlock(&proctree_lock);
891 		return;
892 	}
893 
894 	PROC_LOCK(q);
895 	sigqueue_take(p->p_ksi);
896 	PROC_UNLOCK(q);
897 
898 	/*
899 	 * If we got the child via a ptrace 'attach', we need to give it back
900 	 * to the old parent.
901 	 */
902 	if (p->p_oppid != p->p_pptr->p_pid) {
903 		PROC_UNLOCK(p);
904 		t = proc_realparent(p);
905 		PROC_LOCK(t);
906 		PROC_LOCK(p);
907 		CTR2(KTR_PTRACE,
908 		    "wait: traced child %d moved back to parent %d", p->p_pid,
909 		    t->p_pid);
910 		proc_reparent(p, t, false);
911 		PROC_UNLOCK(p);
912 		pksignal(t, SIGCHLD, p->p_ksi);
913 		wakeup(t);
914 		cv_broadcast(&p->p_pwait);
915 		PROC_UNLOCK(t);
916 		sx_xunlock(&proctree_lock);
917 		return;
918 	}
919 	PROC_UNLOCK(p);
920 
921 	/*
922 	 * Remove other references to this process to ensure we have an
923 	 * exclusive reference.
924 	 */
925 	sx_xlock(PIDHASHLOCK(p->p_pid));
926 	LIST_REMOVE(p, p_hash);
927 	sx_xunlock(PIDHASHLOCK(p->p_pid));
928 	LIST_REMOVE(p, p_sibling);
929 	reaper_abandon_children(p, true);
930 	reaper_clear(p);
931 	PROC_LOCK(p);
932 	proc_clear_orphan(p);
933 	PROC_UNLOCK(p);
934 	leavepgrp(p);
935 	if (p->p_procdesc != NULL)
936 		procdesc_reap(p);
937 	sx_xunlock(&proctree_lock);
938 
939 	proc_id_clear(PROC_ID_PID, p->p_pid);
940 
941 	PROC_LOCK(p);
942 	knlist_detach(p->p_klist);
943 	p->p_klist = NULL;
944 	PROC_UNLOCK(p);
945 
946 	/*
947 	 * Removal from allproc list and process group list paired with
948 	 * PROC_LOCK which was executed during that time should guarantee
949 	 * nothing can reach this process anymore. As such further locking
950 	 * is unnecessary.
951 	 */
952 	p->p_xexit = p->p_xsig = 0;		/* XXX: why? */
953 
954 	PROC_LOCK(q);
955 	ruadd(&q->p_stats->p_cru, &q->p_crux, &p->p_ru, &p->p_rux);
956 	PROC_UNLOCK(q);
957 
958 	/*
959 	 * Decrement the count of procs running with this uid.
960 	 */
961 	(void)chgproccnt(p->p_ucred->cr_ruidinfo, -1, 0);
962 
963 	/*
964 	 * Destroy resource accounting information associated with the process.
965 	 */
966 #ifdef RACCT
967 	if (racct_enable) {
968 		PROC_LOCK(p);
969 		racct_sub(p, RACCT_NPROC, 1);
970 		PROC_UNLOCK(p);
971 	}
972 #endif
973 	racct_proc_exit(p);
974 
975 	/*
976 	 * Free credentials, arguments, and sigacts.
977 	 */
978 	proc_unset_cred(p);
979 	pargs_drop(p->p_args);
980 	p->p_args = NULL;
981 	sigacts_free(p->p_sigacts);
982 	p->p_sigacts = NULL;
983 
984 	/*
985 	 * Do any thread-system specific cleanups.
986 	 */
987 	thread_wait(p);
988 
989 	/*
990 	 * Give vm and machine-dependent layer a chance to free anything that
991 	 * cpu_exit couldn't release while still running in process context.
992 	 */
993 	vm_waitproc(p);
994 #ifdef MAC
995 	mac_proc_destroy(p);
996 #endif
997 
998 	KASSERT(FIRST_THREAD_IN_PROC(p),
999 	    ("proc_reap: no residual thread!"));
1000 	uma_zfree(proc_zone, p);
1001 	atomic_add_int(&nprocs, -1);
1002 }
1003 
1004 static int
1005 proc_to_reap(struct thread *td, struct proc *p, idtype_t idtype, id_t id,
1006     int *status, int options, struct __wrusage *wrusage, siginfo_t *siginfo,
1007     int check_only)
1008 {
1009 	struct rusage *rup;
1010 
1011 	sx_assert(&proctree_lock, SA_XLOCKED);
1012 
1013 	PROC_LOCK(p);
1014 
1015 	switch (idtype) {
1016 	case P_ALL:
1017 		if (p->p_procdesc == NULL ||
1018 		   (p->p_pptr == td->td_proc &&
1019 		   (p->p_flag & P_TRACED) != 0)) {
1020 			break;
1021 		}
1022 
1023 		PROC_UNLOCK(p);
1024 		return (0);
1025 	case P_PID:
1026 		if (p->p_pid != (pid_t)id) {
1027 			PROC_UNLOCK(p);
1028 			return (0);
1029 		}
1030 		break;
1031 	case P_PGID:
1032 		if (p->p_pgid != (pid_t)id) {
1033 			PROC_UNLOCK(p);
1034 			return (0);
1035 		}
1036 		break;
1037 	case P_SID:
1038 		if (p->p_session->s_sid != (pid_t)id) {
1039 			PROC_UNLOCK(p);
1040 			return (0);
1041 		}
1042 		break;
1043 	case P_UID:
1044 		if (p->p_ucred->cr_uid != (uid_t)id) {
1045 			PROC_UNLOCK(p);
1046 			return (0);
1047 		}
1048 		break;
1049 	case P_GID:
1050 		if (p->p_ucred->cr_gid != (gid_t)id) {
1051 			PROC_UNLOCK(p);
1052 			return (0);
1053 		}
1054 		break;
1055 	case P_JAILID:
1056 		if (p->p_ucred->cr_prison->pr_id != (int)id) {
1057 			PROC_UNLOCK(p);
1058 			return (0);
1059 		}
1060 		break;
1061 	/*
1062 	 * It seems that the thread structures get zeroed out
1063 	 * at process exit.  This makes it impossible to
1064 	 * support P_SETID, P_CID or P_CPUID.
1065 	 */
1066 	default:
1067 		PROC_UNLOCK(p);
1068 		return (0);
1069 	}
1070 
1071 	if (p_canwait(td, p)) {
1072 		PROC_UNLOCK(p);
1073 		return (0);
1074 	}
1075 
1076 	if (((options & WEXITED) == 0) && (p->p_state == PRS_ZOMBIE)) {
1077 		PROC_UNLOCK(p);
1078 		return (0);
1079 	}
1080 
1081 	/*
1082 	 * This special case handles a kthread spawned by linux_clone
1083 	 * (see linux_misc.c).  The linux_wait4 and linux_waitpid
1084 	 * functions need to be able to distinguish between waiting
1085 	 * on a process and waiting on a thread.  It is a thread if
1086 	 * p_sigparent is not SIGCHLD, and the WLINUXCLONE option
1087 	 * signifies we want to wait for threads and not processes.
1088 	 */
1089 	if ((p->p_sigparent != SIGCHLD) ^
1090 	    ((options & WLINUXCLONE) != 0)) {
1091 		PROC_UNLOCK(p);
1092 		return (0);
1093 	}
1094 
1095 	if (siginfo != NULL) {
1096 		bzero(siginfo, sizeof(*siginfo));
1097 		siginfo->si_errno = 0;
1098 
1099 		/*
1100 		 * SUSv4 requires that the si_signo value is always
1101 		 * SIGCHLD. Obey it despite the rfork(2) interface
1102 		 * allows to request other signal for child exit
1103 		 * notification.
1104 		 */
1105 		siginfo->si_signo = SIGCHLD;
1106 
1107 		/*
1108 		 *  This is still a rough estimate.  We will fix the
1109 		 *  cases TRAPPED, STOPPED, and CONTINUED later.
1110 		 */
1111 		if (WCOREDUMP(p->p_xsig)) {
1112 			siginfo->si_code = CLD_DUMPED;
1113 			siginfo->si_status = WTERMSIG(p->p_xsig);
1114 		} else if (WIFSIGNALED(p->p_xsig)) {
1115 			siginfo->si_code = CLD_KILLED;
1116 			siginfo->si_status = WTERMSIG(p->p_xsig);
1117 		} else {
1118 			siginfo->si_code = CLD_EXITED;
1119 			siginfo->si_status = p->p_xexit;
1120 		}
1121 
1122 		siginfo->si_pid = p->p_pid;
1123 		siginfo->si_uid = p->p_ucred->cr_uid;
1124 
1125 		/*
1126 		 * The si_addr field would be useful additional
1127 		 * detail, but apparently the PC value may be lost
1128 		 * when we reach this point.  bzero() above sets
1129 		 * siginfo->si_addr to NULL.
1130 		 */
1131 	}
1132 
1133 	/*
1134 	 * There should be no reason to limit resources usage info to
1135 	 * exited processes only.  A snapshot about any resources used
1136 	 * by a stopped process may be exactly what is needed.
1137 	 */
1138 	if (wrusage != NULL) {
1139 		rup = &wrusage->wru_self;
1140 		*rup = p->p_ru;
1141 		PROC_STATLOCK(p);
1142 		calcru(p, &rup->ru_utime, &rup->ru_stime);
1143 		PROC_STATUNLOCK(p);
1144 
1145 		rup = &wrusage->wru_children;
1146 		*rup = p->p_stats->p_cru;
1147 		calccru(p, &rup->ru_utime, &rup->ru_stime);
1148 	}
1149 
1150 	if (p->p_state == PRS_ZOMBIE && !check_only) {
1151 		proc_reap(td, p, status, options);
1152 		return (-1);
1153 	}
1154 	return (1);
1155 }
1156 
1157 int
1158 kern_wait(struct thread *td, pid_t pid, int *status, int options,
1159     struct rusage *rusage)
1160 {
1161 	struct __wrusage wru, *wrup;
1162 	idtype_t idtype;
1163 	id_t id;
1164 	int ret;
1165 
1166 	/*
1167 	 * Translate the special pid values into the (idtype, pid)
1168 	 * pair for kern_wait6.  The WAIT_MYPGRP case is handled by
1169 	 * kern_wait6() on its own.
1170 	 */
1171 	if (pid == WAIT_ANY) {
1172 		idtype = P_ALL;
1173 		id = 0;
1174 	} else if (pid < 0) {
1175 		idtype = P_PGID;
1176 		id = (id_t)-pid;
1177 	} else {
1178 		idtype = P_PID;
1179 		id = (id_t)pid;
1180 	}
1181 
1182 	if (rusage != NULL)
1183 		wrup = &wru;
1184 	else
1185 		wrup = NULL;
1186 
1187 	/*
1188 	 * For backward compatibility we implicitly add flags WEXITED
1189 	 * and WTRAPPED here.
1190 	 */
1191 	options |= WEXITED | WTRAPPED;
1192 	ret = kern_wait6(td, idtype, id, status, options, wrup, NULL);
1193 	if (rusage != NULL)
1194 		*rusage = wru.wru_self;
1195 	return (ret);
1196 }
1197 
1198 static void
1199 report_alive_proc(struct thread *td, struct proc *p, siginfo_t *siginfo,
1200     int *status, int options, int si_code)
1201 {
1202 	bool cont;
1203 
1204 	PROC_LOCK_ASSERT(p, MA_OWNED);
1205 	sx_assert(&proctree_lock, SA_XLOCKED);
1206 	MPASS(si_code == CLD_TRAPPED || si_code == CLD_STOPPED ||
1207 	    si_code == CLD_CONTINUED);
1208 
1209 	cont = si_code == CLD_CONTINUED;
1210 	if ((options & WNOWAIT) == 0) {
1211 		if (cont)
1212 			p->p_flag &= ~P_CONTINUED;
1213 		else
1214 			p->p_flag |= P_WAITED;
1215 		if (kern_wait_dequeue_sigchld &&
1216 		    (td->td_proc->p_sysent->sv_flags & SV_SIG_WAITNDQ) == 0) {
1217 			PROC_LOCK(td->td_proc);
1218 			sigqueue_take(p->p_ksi);
1219 			PROC_UNLOCK(td->td_proc);
1220 		}
1221 	}
1222 	sx_xunlock(&proctree_lock);
1223 	if (siginfo != NULL) {
1224 		siginfo->si_code = si_code;
1225 		siginfo->si_status = cont ? SIGCONT : p->p_xsig;
1226 	}
1227 	if (status != NULL)
1228 		*status = cont ? SIGCONT : W_STOPCODE(p->p_xsig);
1229 	PROC_UNLOCK(p);
1230 	td->td_retval[0] = p->p_pid;
1231 }
1232 
1233 int
1234 kern_wait6(struct thread *td, idtype_t idtype, id_t id, int *status,
1235     int options, struct __wrusage *wrusage, siginfo_t *siginfo)
1236 {
1237 	struct proc *p, *q;
1238 	pid_t pid;
1239 	int error, nfound, ret;
1240 	bool report;
1241 
1242 	AUDIT_ARG_VALUE((int)idtype);	/* XXX - This is likely wrong! */
1243 	AUDIT_ARG_PID((pid_t)id);	/* XXX - This may be wrong! */
1244 	AUDIT_ARG_VALUE(options);
1245 
1246 	q = td->td_proc;
1247 
1248 	if ((pid_t)id == WAIT_MYPGRP && (idtype == P_PID || idtype == P_PGID)) {
1249 		PROC_LOCK(q);
1250 		id = (id_t)q->p_pgid;
1251 		PROC_UNLOCK(q);
1252 		idtype = P_PGID;
1253 	}
1254 
1255 	/* If we don't know the option, just return. */
1256 	if ((options & ~(WUNTRACED | WNOHANG | WCONTINUED | WNOWAIT |
1257 	    WEXITED | WTRAPPED | WLINUXCLONE)) != 0)
1258 		return (EINVAL);
1259 	if ((options & (WEXITED | WUNTRACED | WCONTINUED | WTRAPPED)) == 0) {
1260 		/*
1261 		 * We will be unable to find any matching processes,
1262 		 * because there are no known events to look for.
1263 		 * Prefer to return error instead of blocking
1264 		 * indefinitely.
1265 		 */
1266 		return (EINVAL);
1267 	}
1268 
1269 loop:
1270 	if (q->p_flag & P_STATCHILD) {
1271 		PROC_LOCK(q);
1272 		q->p_flag &= ~P_STATCHILD;
1273 		PROC_UNLOCK(q);
1274 	}
1275 	sx_xlock(&proctree_lock);
1276 loop_locked:
1277 	nfound = 0;
1278 	LIST_FOREACH(p, &q->p_children, p_sibling) {
1279 		pid = p->p_pid;
1280 		ret = proc_to_reap(td, p, idtype, id, status, options,
1281 		    wrusage, siginfo, 0);
1282 		if (ret == 0)
1283 			continue;
1284 		else if (ret != 1) {
1285 			td->td_retval[0] = pid;
1286 			return (0);
1287 		}
1288 
1289 		nfound++;
1290 		PROC_LOCK_ASSERT(p, MA_OWNED);
1291 
1292 		if ((options & WTRAPPED) != 0 &&
1293 		    (p->p_flag & P_TRACED) != 0) {
1294 			PROC_SLOCK(p);
1295 			report =
1296 			    ((p->p_flag & (P_STOPPED_TRACE | P_STOPPED_SIG)) &&
1297 			    p->p_suspcount == p->p_numthreads &&
1298 			    (p->p_flag & P_WAITED) == 0);
1299 			PROC_SUNLOCK(p);
1300 			if (report) {
1301 			CTR4(KTR_PTRACE,
1302 			    "wait: returning trapped pid %d status %#x "
1303 			    "(xstat %d) xthread %d",
1304 			    p->p_pid, W_STOPCODE(p->p_xsig), p->p_xsig,
1305 			    p->p_xthread != NULL ?
1306 			    p->p_xthread->td_tid : -1);
1307 				report_alive_proc(td, p, siginfo, status,
1308 				    options, CLD_TRAPPED);
1309 				return (0);
1310 			}
1311 		}
1312 		if ((options & WUNTRACED) != 0 &&
1313 		    (p->p_flag & P_STOPPED_SIG) != 0) {
1314 			PROC_SLOCK(p);
1315 			report = (p->p_suspcount == p->p_numthreads &&
1316 			    ((p->p_flag & P_WAITED) == 0));
1317 			PROC_SUNLOCK(p);
1318 			if (report) {
1319 				report_alive_proc(td, p, siginfo, status,
1320 				    options, CLD_STOPPED);
1321 				return (0);
1322 			}
1323 		}
1324 		if ((options & WCONTINUED) != 0 &&
1325 		    (p->p_flag & P_CONTINUED) != 0) {
1326 			report_alive_proc(td, p, siginfo, status, options,
1327 			    CLD_CONTINUED);
1328 			return (0);
1329 		}
1330 		PROC_UNLOCK(p);
1331 	}
1332 
1333 	/*
1334 	 * Look in the orphans list too, to allow the parent to
1335 	 * collect it's child exit status even if child is being
1336 	 * debugged.
1337 	 *
1338 	 * Debugger detaches from the parent upon successful
1339 	 * switch-over from parent to child.  At this point due to
1340 	 * re-parenting the parent loses the child to debugger and a
1341 	 * wait4(2) call would report that it has no children to wait
1342 	 * for.  By maintaining a list of orphans we allow the parent
1343 	 * to successfully wait until the child becomes a zombie.
1344 	 */
1345 	if (nfound == 0) {
1346 		LIST_FOREACH(p, &q->p_orphans, p_orphan) {
1347 			ret = proc_to_reap(td, p, idtype, id, NULL, options,
1348 			    NULL, NULL, 1);
1349 			if (ret != 0) {
1350 				KASSERT(ret != -1, ("reaped an orphan (pid %d)",
1351 				    (int)td->td_retval[0]));
1352 				PROC_UNLOCK(p);
1353 				nfound++;
1354 				break;
1355 			}
1356 		}
1357 	}
1358 	if (nfound == 0) {
1359 		sx_xunlock(&proctree_lock);
1360 		return (ECHILD);
1361 	}
1362 	if (options & WNOHANG) {
1363 		sx_xunlock(&proctree_lock);
1364 		td->td_retval[0] = 0;
1365 		return (0);
1366 	}
1367 	PROC_LOCK(q);
1368 	if (q->p_flag & P_STATCHILD) {
1369 		q->p_flag &= ~P_STATCHILD;
1370 		PROC_UNLOCK(q);
1371 		goto loop_locked;
1372 	}
1373 	sx_xunlock(&proctree_lock);
1374 	error = msleep(q, &q->p_mtx, PWAIT | PCATCH | PDROP, "wait", 0);
1375 	if (error)
1376 		return (error);
1377 	goto loop;
1378 }
1379 
1380 void
1381 proc_add_orphan(struct proc *child, struct proc *parent)
1382 {
1383 
1384 	sx_assert(&proctree_lock, SX_XLOCKED);
1385 	KASSERT((child->p_flag & P_TRACED) != 0,
1386 	    ("proc_add_orphan: not traced"));
1387 
1388 	if (LIST_EMPTY(&parent->p_orphans)) {
1389 		child->p_treeflag |= P_TREE_FIRST_ORPHAN;
1390 		LIST_INSERT_HEAD(&parent->p_orphans, child, p_orphan);
1391 	} else {
1392 		LIST_INSERT_AFTER(LIST_FIRST(&parent->p_orphans),
1393 		    child, p_orphan);
1394 	}
1395 	child->p_treeflag |= P_TREE_ORPHANED;
1396 }
1397 
1398 /*
1399  * Make process 'parent' the new parent of process 'child'.
1400  * Must be called with an exclusive hold of proctree lock.
1401  */
1402 void
1403 proc_reparent(struct proc *child, struct proc *parent, bool set_oppid)
1404 {
1405 
1406 	sx_assert(&proctree_lock, SX_XLOCKED);
1407 	PROC_LOCK_ASSERT(child, MA_OWNED);
1408 	if (child->p_pptr == parent)
1409 		return;
1410 
1411 	PROC_LOCK(child->p_pptr);
1412 	sigqueue_take(child->p_ksi);
1413 	PROC_UNLOCK(child->p_pptr);
1414 	LIST_REMOVE(child, p_sibling);
1415 	LIST_INSERT_HEAD(&parent->p_children, child, p_sibling);
1416 
1417 	proc_clear_orphan(child);
1418 	if ((child->p_flag & P_TRACED) != 0) {
1419 		proc_add_orphan(child, child->p_pptr);
1420 	}
1421 
1422 	child->p_pptr = parent;
1423 	if (set_oppid)
1424 		child->p_oppid = parent->p_pid;
1425 }
1426