xref: /freebsd/sys/kern/kern_exit.c (revision 25ecdc7d52770caf1c9b44b5ec11f468f6b636f3)
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 struct proc *
109 proc_realparent(struct proc *child)
110 {
111 	struct proc *p, *parent;
112 
113 	sx_assert(&proctree_lock, SX_LOCKED);
114 	if ((child->p_treeflag & P_TREE_ORPHANED) == 0)
115 		return (child->p_pptr->p_pid == child->p_oppid ?
116 		    child->p_pptr : child->p_reaper);
117 	for (p = child; (p->p_treeflag & P_TREE_FIRST_ORPHAN) == 0;) {
118 		/* Cannot use LIST_PREV(), since the list head is not known. */
119 		p = __containerof(p->p_orphan.le_prev, struct proc,
120 		    p_orphan.le_next);
121 		KASSERT((p->p_treeflag & P_TREE_ORPHANED) != 0,
122 		    ("missing P_ORPHAN %p", p));
123 	}
124 	parent = __containerof(p->p_orphan.le_prev, struct proc,
125 	    p_orphans.lh_first);
126 	return (parent);
127 }
128 
129 void
130 reaper_abandon_children(struct proc *p, bool exiting)
131 {
132 	struct proc *p1, *p2, *ptmp;
133 
134 	sx_assert(&proctree_lock, SX_LOCKED);
135 	KASSERT(p != initproc, ("reaper_abandon_children for initproc"));
136 	if ((p->p_treeflag & P_TREE_REAPER) == 0)
137 		return;
138 	p1 = p->p_reaper;
139 	LIST_FOREACH_SAFE(p2, &p->p_reaplist, p_reapsibling, ptmp) {
140 		LIST_REMOVE(p2, p_reapsibling);
141 		p2->p_reaper = p1;
142 		p2->p_reapsubtree = p->p_reapsubtree;
143 		LIST_INSERT_HEAD(&p1->p_reaplist, p2, p_reapsibling);
144 		if (exiting && p2->p_pptr == p) {
145 			PROC_LOCK(p2);
146 			proc_reparent(p2, p1, true);
147 			PROC_UNLOCK(p2);
148 		}
149 	}
150 	KASSERT(LIST_EMPTY(&p->p_reaplist), ("p_reaplist not empty"));
151 	p->p_treeflag &= ~P_TREE_REAPER;
152 }
153 
154 static void
155 reaper_clear(struct proc *p)
156 {
157 	struct proc *p1;
158 	bool clear;
159 
160 	sx_assert(&proctree_lock, SX_LOCKED);
161 	LIST_REMOVE(p, p_reapsibling);
162 	if (p->p_reapsubtree == 1)
163 		return;
164 	clear = true;
165 	LIST_FOREACH(p1, &p->p_reaper->p_reaplist, p_reapsibling) {
166 		if (p1->p_reapsubtree == p->p_reapsubtree) {
167 			clear = false;
168 			break;
169 		}
170 	}
171 	if (clear)
172 		proc_id_clear(PROC_ID_REAP, p->p_reapsubtree);
173 }
174 
175 void
176 proc_clear_orphan(struct proc *p)
177 {
178 	struct proc *p1;
179 
180 	sx_assert(&proctree_lock, SA_XLOCKED);
181 	if ((p->p_treeflag & P_TREE_ORPHANED) == 0)
182 		return;
183 	if ((p->p_treeflag & P_TREE_FIRST_ORPHAN) != 0) {
184 		p1 = LIST_NEXT(p, p_orphan);
185 		if (p1 != NULL)
186 			p1->p_treeflag |= P_TREE_FIRST_ORPHAN;
187 		p->p_treeflag &= ~P_TREE_FIRST_ORPHAN;
188 	}
189 	LIST_REMOVE(p, p_orphan);
190 	p->p_treeflag &= ~P_TREE_ORPHANED;
191 }
192 
193 /*
194  * exit -- death of process.
195  */
196 void
197 sys_sys_exit(struct thread *td, struct sys_exit_args *uap)
198 {
199 
200 	exit1(td, uap->rval, 0);
201 	/* NOTREACHED */
202 }
203 
204 /*
205  * Exit: deallocate address space and other resources, change proc state to
206  * zombie, and unlink proc from allproc and parent's lists.  Save exit status
207  * and rusage for wait().  Check for child processes and orphan them.
208  */
209 void
210 exit1(struct thread *td, int rval, int signo)
211 {
212 	struct proc *p, *nq, *q, *t;
213 	struct thread *tdt;
214 	ksiginfo_t *ksi, *ksi1;
215 	int signal_parent;
216 
217 	mtx_assert(&Giant, MA_NOTOWNED);
218 	KASSERT(rval == 0 || signo == 0, ("exit1 rv %d sig %d", rval, signo));
219 
220 	p = td->td_proc;
221 	/*
222 	 * XXX in case we're rebooting we just let init die in order to
223 	 * work around an unsolved stack overflow seen very late during
224 	 * shutdown on sparc64 when the gmirror worker process exists.
225 	 * XXX what to do now that sparc64 is gone... remove if?
226 	 */
227 	if (p == initproc && rebooting == 0) {
228 		printf("init died (signal %d, exit %d)\n", signo, rval);
229 		panic("Going nowhere without my init!");
230 	}
231 
232 	/*
233 	 * Deref SU mp, since the thread does not return to userspace.
234 	 */
235 	td_softdep_cleanup(td);
236 
237 	/*
238 	 * MUST abort all other threads before proceeding past here.
239 	 */
240 	PROC_LOCK(p);
241 	/*
242 	 * First check if some other thread or external request got
243 	 * here before us.  If so, act appropriately: exit or suspend.
244 	 * We must ensure that stop requests are handled before we set
245 	 * P_WEXIT.
246 	 */
247 	thread_suspend_check(0);
248 	while (p->p_flag & P_HADTHREADS) {
249 		/*
250 		 * Kill off the other threads. This requires
251 		 * some co-operation from other parts of the kernel
252 		 * so it may not be instantaneous.  With this state set
253 		 * any thread entering the kernel from userspace will
254 		 * thread_exit() in trap().  Any thread attempting to
255 		 * sleep will return immediately with EINTR or EWOULDBLOCK
256 		 * which will hopefully force them to back out to userland
257 		 * freeing resources as they go.  Any thread attempting
258 		 * to return to userland will thread_exit() from userret().
259 		 * thread_exit() will unsuspend us when the last of the
260 		 * other threads exits.
261 		 * If there is already a thread singler after resumption,
262 		 * calling thread_single will fail; in that case, we just
263 		 * re-check all suspension request, the thread should
264 		 * either be suspended there or exit.
265 		 */
266 		if (!thread_single(p, SINGLE_EXIT))
267 			/*
268 			 * All other activity in this process is now
269 			 * stopped.  Threading support has been turned
270 			 * off.
271 			 */
272 			break;
273 		/*
274 		 * Recheck for new stop or suspend requests which
275 		 * might appear while process lock was dropped in
276 		 * thread_single().
277 		 */
278 		thread_suspend_check(0);
279 	}
280 	KASSERT(p->p_numthreads == 1,
281 	    ("exit1: proc %p exiting with %d threads", p, p->p_numthreads));
282 	racct_sub(p, RACCT_NTHR, 1);
283 
284 	/* Let event handler change exit status */
285 	p->p_xexit = rval;
286 	p->p_xsig = signo;
287 
288 	/*
289 	 * Ignore any pending request to stop due to a stop signal.
290 	 * Once P_WEXIT is set, future requests will be ignored as
291 	 * well.
292 	 */
293 	p->p_flag &= ~P_STOPPED_SIG;
294 	KASSERT(!P_SHOULDSTOP(p), ("exiting process is stopped"));
295 
296 	/* Note that we are exiting. */
297 	p->p_flag |= P_WEXIT;
298 
299 	/*
300 	 * Wait for any processes that have a hold on our vmspace to
301 	 * release their reference.
302 	 */
303 	while (p->p_lock > 0)
304 		msleep(&p->p_lock, &p->p_mtx, PWAIT, "exithold", 0);
305 
306 	PROC_UNLOCK(p);
307 	/* Drain the limit callout while we don't have the proc locked */
308 	callout_drain(&p->p_limco);
309 
310 #ifdef AUDIT
311 	/*
312 	 * The Sun BSM exit token contains two components: an exit status as
313 	 * passed to exit(), and a return value to indicate what sort of exit
314 	 * it was.  The exit status is WEXITSTATUS(rv), but it's not clear
315 	 * what the return value is.
316 	 */
317 	AUDIT_ARG_EXIT(rval, 0);
318 	AUDIT_SYSCALL_EXIT(0, td);
319 #endif
320 
321 	/* Are we a task leader with peers? */
322 	if (p->p_peers != NULL && p == p->p_leader) {
323 		mtx_lock(&ppeers_lock);
324 		q = p->p_peers;
325 		while (q != NULL) {
326 			PROC_LOCK(q);
327 			kern_psignal(q, SIGKILL);
328 			PROC_UNLOCK(q);
329 			q = q->p_peers;
330 		}
331 		while (p->p_peers != NULL)
332 			msleep(p, &ppeers_lock, PWAIT, "exit1", 0);
333 		mtx_unlock(&ppeers_lock);
334 	}
335 
336 	itimers_exit(p);
337 
338 	if (p->p_sysent->sv_onexit != NULL)
339 		p->p_sysent->sv_onexit(p);
340 
341 	/*
342 	 * Check if any loadable modules need anything done at process exit.
343 	 * E.g. SYSV IPC stuff.
344 	 * Event handler could change exit status.
345 	 * XXX what if one of these generates an error?
346 	 */
347 	EVENTHANDLER_DIRECT_INVOKE(process_exit, p);
348 
349 	/*
350 	 * If parent is waiting for us to exit or exec,
351 	 * P_PPWAIT is set; we will wakeup the parent below.
352 	 */
353 	PROC_LOCK(p);
354 	stopprofclock(p);
355 	p->p_ptevents = 0;
356 
357 	/*
358 	 * Stop the real interval timer.  If the handler is currently
359 	 * executing, prevent it from rearming itself and let it finish.
360 	 */
361 	if (timevalisset(&p->p_realtimer.it_value) &&
362 	    _callout_stop_safe(&p->p_itcallout, CS_EXECUTING, NULL) == 0) {
363 		timevalclear(&p->p_realtimer.it_interval);
364 		msleep(&p->p_itcallout, &p->p_mtx, PWAIT, "ritwait", 0);
365 		KASSERT(!timevalisset(&p->p_realtimer.it_value),
366 		    ("realtime timer is still armed"));
367 	}
368 
369 	PROC_UNLOCK(p);
370 
371 	umtx_thread_exit(td);
372 	seltdfini(td);
373 
374 	/*
375 	 * Reset any sigio structures pointing to us as a result of
376 	 * F_SETOWN with our pid.  The P_WEXIT flag interlocks with fsetown().
377 	 */
378 	funsetownlst(&p->p_sigiolst);
379 
380 	/*
381 	 * Close open files and release open-file table.
382 	 * This may block!
383 	 */
384 	pdescfree(td);
385 	fdescfree(td);
386 
387 	/*
388 	 * If this thread tickled GEOM, we need to wait for the giggling to
389 	 * stop before we return to userland
390 	 */
391 	if (td->td_pflags & TDP_GEOM)
392 		g_waitidle();
393 
394 	/*
395 	 * Remove ourself from our leader's peer list and wake our leader.
396 	 */
397 	if (p->p_leader->p_peers != NULL) {
398 		mtx_lock(&ppeers_lock);
399 		if (p->p_leader->p_peers != NULL) {
400 			q = p->p_leader;
401 			while (q->p_peers != p)
402 				q = q->p_peers;
403 			q->p_peers = p->p_peers;
404 			wakeup(p->p_leader);
405 		}
406 		mtx_unlock(&ppeers_lock);
407 	}
408 
409 	vmspace_exit(td);
410 	(void)acct_process(td);
411 
412 #ifdef KTRACE
413 	ktrprocexit(td);
414 #endif
415 	/*
416 	 * Release reference to text vnode
417 	 */
418 	if (p->p_textvp != NULL) {
419 		vrele(p->p_textvp);
420 		p->p_textvp = NULL;
421 	}
422 
423 	/*
424 	 * Release our limits structure.
425 	 */
426 	lim_free(p->p_limit);
427 	p->p_limit = NULL;
428 
429 	tidhash_remove(td);
430 
431 	/*
432 	 * Call machine-dependent code to release any
433 	 * machine-dependent resources other than the address space.
434 	 * The address space is released by "vmspace_exitfree(p)" in
435 	 * vm_waitproc().
436 	 */
437 	cpu_exit(td);
438 
439 	WITNESS_WARN(WARN_PANIC, NULL, "process (pid %d) exiting", p->p_pid);
440 
441 	/*
442 	 * Remove from allproc. It still sits in the hash.
443 	 */
444 	sx_xlock(&allproc_lock);
445 	LIST_REMOVE(p, p_list);
446 
447 #ifdef DDB
448 	/*
449 	 * Used by ddb's 'ps' command to find this process via the
450 	 * pidhash.
451 	 */
452 	p->p_list.le_prev = NULL;
453 #endif
454 	sx_xunlock(&allproc_lock);
455 
456 	sx_xlock(&proctree_lock);
457 	PROC_LOCK(p);
458 	p->p_flag &= ~(P_TRACED | P_PPWAIT | P_PPTRACE);
459 	PROC_UNLOCK(p);
460 
461 	/*
462 	 * killjobc() might drop and re-acquire proctree_lock to
463 	 * revoke control tty if exiting process was a session leader.
464 	 */
465 	killjobc();
466 
467 	/*
468 	 * Reparent all children processes:
469 	 * - traced ones to the original parent (or init if we are that parent)
470 	 * - the rest to init
471 	 */
472 	q = LIST_FIRST(&p->p_children);
473 	if (q != NULL)		/* only need this if any child is S_ZOMB */
474 		wakeup(q->p_reaper);
475 	for (; q != NULL; q = nq) {
476 		nq = LIST_NEXT(q, p_sibling);
477 		ksi = ksiginfo_alloc(TRUE);
478 		PROC_LOCK(q);
479 		q->p_sigparent = SIGCHLD;
480 
481 		if ((q->p_flag & P_TRACED) == 0) {
482 			proc_reparent(q, q->p_reaper, true);
483 			if (q->p_state == PRS_ZOMBIE) {
484 				/*
485 				 * Inform reaper about the reparented
486 				 * zombie, since wait(2) has something
487 				 * new to report.  Guarantee queueing
488 				 * of the SIGCHLD signal, similar to
489 				 * the _exit() behaviour, by providing
490 				 * our ksiginfo.  Ksi is freed by the
491 				 * signal delivery.
492 				 */
493 				if (q->p_ksi == NULL) {
494 					ksi1 = NULL;
495 				} else {
496 					ksiginfo_copy(q->p_ksi, ksi);
497 					ksi->ksi_flags |= KSI_INS;
498 					ksi1 = ksi;
499 					ksi = NULL;
500 				}
501 				PROC_LOCK(q->p_reaper);
502 				pksignal(q->p_reaper, SIGCHLD, ksi1);
503 				PROC_UNLOCK(q->p_reaper);
504 			} else if (q->p_pdeathsig > 0) {
505 				/*
506 				 * The child asked to received a signal
507 				 * when we exit.
508 				 */
509 				kern_psignal(q, q->p_pdeathsig);
510 			}
511 		} else {
512 			/*
513 			 * Traced processes are killed by default
514 			 * since their existence means someone is
515 			 * screwing up.
516 			 */
517 			t = proc_realparent(q);
518 			if (t == p) {
519 				proc_reparent(q, q->p_reaper, true);
520 			} else {
521 				PROC_LOCK(t);
522 				proc_reparent(q, t, true);
523 				PROC_UNLOCK(t);
524 			}
525 			/*
526 			 * Since q was found on our children list, the
527 			 * proc_reparent() call moved q to the orphan
528 			 * list due to present P_TRACED flag. Clear
529 			 * orphan link for q now while q is locked.
530 			 */
531 			proc_clear_orphan(q);
532 			q->p_flag &= ~P_TRACED;
533 			q->p_flag2 &= ~P2_PTRACE_FSTP;
534 			q->p_ptevents = 0;
535 			p->p_xthread = NULL;
536 			FOREACH_THREAD_IN_PROC(q, tdt) {
537 				tdt->td_dbgflags &= ~(TDB_SUSPEND | TDB_XSIG |
538 				    TDB_FSTP);
539 				tdt->td_xsig = 0;
540 			}
541 			if (kern_kill_on_dbg_exit) {
542 				q->p_flag &= ~P_STOPPED_TRACE;
543 				kern_psignal(q, SIGKILL);
544 			} else if ((q->p_flag & (P_STOPPED_TRACE |
545 			    P_STOPPED_SIG)) != 0) {
546 				sigqueue_delete_proc(q, SIGTRAP);
547 				ptrace_unsuspend(q);
548 			}
549 		}
550 		PROC_UNLOCK(q);
551 		if (ksi != NULL)
552 			ksiginfo_free(ksi);
553 	}
554 
555 	/*
556 	 * Also get rid of our orphans.
557 	 */
558 	while ((q = LIST_FIRST(&p->p_orphans)) != NULL) {
559 		PROC_LOCK(q);
560 		KASSERT(q->p_oppid == p->p_pid,
561 		    ("orphan %p of %p has unexpected oppid %d", q, p,
562 		    q->p_oppid));
563 		q->p_oppid = q->p_reaper->p_pid;
564 
565 		/*
566 		 * If we are the real parent of this process
567 		 * but it has been reparented to a debugger, then
568 		 * check if it asked for a signal when we exit.
569 		 */
570 		if (q->p_pdeathsig > 0)
571 			kern_psignal(q, q->p_pdeathsig);
572 		CTR2(KTR_PTRACE, "exit: pid %d, clearing orphan %d", p->p_pid,
573 		    q->p_pid);
574 		proc_clear_orphan(q);
575 		PROC_UNLOCK(q);
576 	}
577 
578 #ifdef KDTRACE_HOOKS
579 	if (SDT_PROBES_ENABLED()) {
580 		int reason = CLD_EXITED;
581 		if (WCOREDUMP(signo))
582 			reason = CLD_DUMPED;
583 		else if (WIFSIGNALED(signo))
584 			reason = CLD_KILLED;
585 		SDT_PROBE1(proc, , , exit, reason);
586 	}
587 #endif
588 
589 	/* Save exit status. */
590 	PROC_LOCK(p);
591 	p->p_xthread = td;
592 
593 	if (p->p_sysent->sv_ontdexit != NULL)
594 		p->p_sysent->sv_ontdexit(td);
595 
596 #ifdef KDTRACE_HOOKS
597 	/*
598 	 * Tell the DTrace fasttrap provider about the exit if it
599 	 * has declared an interest.
600 	 */
601 	if (dtrace_fasttrap_exit)
602 		dtrace_fasttrap_exit(p);
603 #endif
604 
605 	/*
606 	 * Notify interested parties of our demise.
607 	 */
608 	KNOTE_LOCKED(p->p_klist, NOTE_EXIT);
609 
610 	/*
611 	 * If this is a process with a descriptor, we may not need to deliver
612 	 * a signal to the parent.  proctree_lock is held over
613 	 * procdesc_exit() to serialize concurrent calls to close() and
614 	 * exit().
615 	 */
616 	signal_parent = 0;
617 	if (p->p_procdesc == NULL || procdesc_exit(p)) {
618 		/*
619 		 * Notify parent that we're gone.  If parent has the
620 		 * PS_NOCLDWAIT flag set, or if the handler is set to SIG_IGN,
621 		 * notify process 1 instead (and hope it will handle this
622 		 * situation).
623 		 */
624 		PROC_LOCK(p->p_pptr);
625 		mtx_lock(&p->p_pptr->p_sigacts->ps_mtx);
626 		if (p->p_pptr->p_sigacts->ps_flag &
627 		    (PS_NOCLDWAIT | PS_CLDSIGIGN)) {
628 			struct proc *pp;
629 
630 			mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx);
631 			pp = p->p_pptr;
632 			PROC_UNLOCK(pp);
633 			proc_reparent(p, p->p_reaper, true);
634 			p->p_sigparent = SIGCHLD;
635 			PROC_LOCK(p->p_pptr);
636 
637 			/*
638 			 * Notify parent, so in case he was wait(2)ing or
639 			 * executing waitpid(2) with our pid, he will
640 			 * continue.
641 			 */
642 			wakeup(pp);
643 		} else
644 			mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx);
645 
646 		if (p->p_pptr == p->p_reaper || p->p_pptr == initproc) {
647 			signal_parent = 1;
648 		} else if (p->p_sigparent != 0) {
649 			if (p->p_sigparent == SIGCHLD) {
650 				signal_parent = 1;
651 			} else { /* LINUX thread */
652 				signal_parent = 2;
653 			}
654 		}
655 	} else
656 		PROC_LOCK(p->p_pptr);
657 	sx_xunlock(&proctree_lock);
658 
659 	if (signal_parent == 1) {
660 		childproc_exited(p);
661 	} else if (signal_parent == 2) {
662 		kern_psignal(p->p_pptr, p->p_sigparent);
663 	}
664 
665 	/* Tell the prison that we are gone. */
666 	prison_proc_free(p->p_ucred->cr_prison);
667 
668 	/*
669 	 * The state PRS_ZOMBIE prevents other proesses from sending
670 	 * signal to the process, to avoid memory leak, we free memory
671 	 * for signal queue at the time when the state is set.
672 	 */
673 	sigqueue_flush(&p->p_sigqueue);
674 	sigqueue_flush(&td->td_sigqueue);
675 
676 	/*
677 	 * We have to wait until after acquiring all locks before
678 	 * changing p_state.  We need to avoid all possible context
679 	 * switches (including ones from blocking on a mutex) while
680 	 * marked as a zombie.  We also have to set the zombie state
681 	 * before we release the parent process' proc lock to avoid
682 	 * a lost wakeup.  So, we first call wakeup, then we grab the
683 	 * sched lock, update the state, and release the parent process'
684 	 * proc lock.
685 	 */
686 	wakeup(p->p_pptr);
687 	cv_broadcast(&p->p_pwait);
688 	sched_exit(p->p_pptr, td);
689 	PROC_SLOCK(p);
690 	p->p_state = PRS_ZOMBIE;
691 	PROC_UNLOCK(p->p_pptr);
692 
693 	/*
694 	 * Save our children's rusage information in our exit rusage.
695 	 */
696 	PROC_STATLOCK(p);
697 	ruadd(&p->p_ru, &p->p_rux, &p->p_stats->p_cru, &p->p_crux);
698 	PROC_STATUNLOCK(p);
699 
700 	/*
701 	 * Make sure the scheduler takes this thread out of its tables etc.
702 	 * This will also release this thread's reference to the ucred.
703 	 * Other thread parts to release include pcb bits and such.
704 	 */
705 	thread_exit();
706 }
707 
708 #ifndef _SYS_SYSPROTO_H_
709 struct abort2_args {
710 	char *why;
711 	int nargs;
712 	void **args;
713 };
714 #endif
715 
716 int
717 sys_abort2(struct thread *td, struct abort2_args *uap)
718 {
719 	struct proc *p = td->td_proc;
720 	struct sbuf *sb;
721 	void *uargs[16];
722 	int error, i, sig;
723 
724 	/*
725 	 * Do it right now so we can log either proper call of abort2(), or
726 	 * note, that invalid argument was passed. 512 is big enough to
727 	 * handle 16 arguments' descriptions with additional comments.
728 	 */
729 	sb = sbuf_new(NULL, NULL, 512, SBUF_FIXEDLEN);
730 	sbuf_clear(sb);
731 	sbuf_printf(sb, "%s(pid %d uid %d) aborted: ",
732 	    p->p_comm, p->p_pid, td->td_ucred->cr_uid);
733 	/*
734 	 * Since we can't return from abort2(), send SIGKILL in cases, where
735 	 * abort2() was called improperly
736 	 */
737 	sig = SIGKILL;
738 	/* Prevent from DoSes from user-space. */
739 	if (uap->nargs < 0 || uap->nargs > 16)
740 		goto out;
741 	if (uap->nargs > 0) {
742 		if (uap->args == NULL)
743 			goto out;
744 		error = copyin(uap->args, uargs, uap->nargs * sizeof(void *));
745 		if (error != 0)
746 			goto out;
747 	}
748 	/*
749 	 * Limit size of 'reason' string to 128. Will fit even when
750 	 * maximal number of arguments was chosen to be logged.
751 	 */
752 	if (uap->why != NULL) {
753 		error = sbuf_copyin(sb, uap->why, 128);
754 		if (error < 0)
755 			goto out;
756 	} else {
757 		sbuf_printf(sb, "(null)");
758 	}
759 	if (uap->nargs > 0) {
760 		sbuf_printf(sb, "(");
761 		for (i = 0;i < uap->nargs; i++)
762 			sbuf_printf(sb, "%s%p", i == 0 ? "" : ", ", uargs[i]);
763 		sbuf_printf(sb, ")");
764 	}
765 	/*
766 	 * Final stage: arguments were proper, string has been
767 	 * successfully copied from userspace, and copying pointers
768 	 * from user-space succeed.
769 	 */
770 	sig = SIGABRT;
771 out:
772 	if (sig == SIGKILL) {
773 		sbuf_trim(sb);
774 		sbuf_printf(sb, " (Reason text inaccessible)");
775 	}
776 	sbuf_cat(sb, "\n");
777 	sbuf_finish(sb);
778 	log(LOG_INFO, "%s", sbuf_data(sb));
779 	sbuf_delete(sb);
780 	exit1(td, 0, sig);
781 	return (0);
782 }
783 
784 #ifdef COMPAT_43
785 /*
786  * The dirty work is handled by kern_wait().
787  */
788 int
789 owait(struct thread *td, struct owait_args *uap __unused)
790 {
791 	int error, status;
792 
793 	error = kern_wait(td, WAIT_ANY, &status, 0, NULL);
794 	if (error == 0)
795 		td->td_retval[1] = status;
796 	return (error);
797 }
798 #endif /* COMPAT_43 */
799 
800 /*
801  * The dirty work is handled by kern_wait().
802  */
803 int
804 sys_wait4(struct thread *td, struct wait4_args *uap)
805 {
806 	struct rusage ru, *rup;
807 	int error, status;
808 
809 	if (uap->rusage != NULL)
810 		rup = &ru;
811 	else
812 		rup = NULL;
813 	error = kern_wait(td, uap->pid, &status, uap->options, rup);
814 	if (uap->status != NULL && error == 0 && td->td_retval[0] != 0)
815 		error = copyout(&status, uap->status, sizeof(status));
816 	if (uap->rusage != NULL && error == 0 && td->td_retval[0] != 0)
817 		error = copyout(&ru, uap->rusage, sizeof(struct rusage));
818 	return (error);
819 }
820 
821 int
822 sys_wait6(struct thread *td, struct wait6_args *uap)
823 {
824 	struct __wrusage wru, *wrup;
825 	siginfo_t si, *sip;
826 	idtype_t idtype;
827 	id_t id;
828 	int error, status;
829 
830 	idtype = uap->idtype;
831 	id = uap->id;
832 
833 	if (uap->wrusage != NULL)
834 		wrup = &wru;
835 	else
836 		wrup = NULL;
837 
838 	if (uap->info != NULL) {
839 		sip = &si;
840 		bzero(sip, sizeof(*sip));
841 	} else
842 		sip = NULL;
843 
844 	/*
845 	 *  We expect all callers of wait6() to know about WEXITED and
846 	 *  WTRAPPED.
847 	 */
848 	error = kern_wait6(td, idtype, id, &status, uap->options, wrup, sip);
849 
850 	if (uap->status != NULL && error == 0 && td->td_retval[0] != 0)
851 		error = copyout(&status, uap->status, sizeof(status));
852 	if (uap->wrusage != NULL && error == 0 && td->td_retval[0] != 0)
853 		error = copyout(&wru, uap->wrusage, sizeof(wru));
854 	if (uap->info != NULL && error == 0)
855 		error = copyout(&si, uap->info, sizeof(si));
856 	return (error);
857 }
858 
859 /*
860  * Reap the remains of a zombie process and optionally return status and
861  * rusage.  Asserts and will release both the proctree_lock and the process
862  * lock as part of its work.
863  */
864 void
865 proc_reap(struct thread *td, struct proc *p, int *status, int options)
866 {
867 	struct proc *q, *t;
868 
869 	sx_assert(&proctree_lock, SA_XLOCKED);
870 	PROC_LOCK_ASSERT(p, MA_OWNED);
871 	KASSERT(p->p_state == PRS_ZOMBIE, ("proc_reap: !PRS_ZOMBIE"));
872 
873 	mtx_spin_wait_unlocked(&p->p_slock);
874 
875 	q = td->td_proc;
876 
877 	if (status)
878 		*status = KW_EXITCODE(p->p_xexit, p->p_xsig);
879 	if (options & WNOWAIT) {
880 		/*
881 		 *  Only poll, returning the status.  Caller does not wish to
882 		 * release the proc struct just yet.
883 		 */
884 		PROC_UNLOCK(p);
885 		sx_xunlock(&proctree_lock);
886 		return;
887 	}
888 
889 	PROC_LOCK(q);
890 	sigqueue_take(p->p_ksi);
891 	PROC_UNLOCK(q);
892 
893 	/*
894 	 * If we got the child via a ptrace 'attach', we need to give it back
895 	 * to the old parent.
896 	 */
897 	if (p->p_oppid != p->p_pptr->p_pid) {
898 		PROC_UNLOCK(p);
899 		t = proc_realparent(p);
900 		PROC_LOCK(t);
901 		PROC_LOCK(p);
902 		CTR2(KTR_PTRACE,
903 		    "wait: traced child %d moved back to parent %d", p->p_pid,
904 		    t->p_pid);
905 		proc_reparent(p, t, false);
906 		PROC_UNLOCK(p);
907 		pksignal(t, SIGCHLD, p->p_ksi);
908 		wakeup(t);
909 		cv_broadcast(&p->p_pwait);
910 		PROC_UNLOCK(t);
911 		sx_xunlock(&proctree_lock);
912 		return;
913 	}
914 	PROC_UNLOCK(p);
915 
916 	/*
917 	 * Remove other references to this process to ensure we have an
918 	 * exclusive reference.
919 	 */
920 	sx_xlock(PIDHASHLOCK(p->p_pid));
921 	LIST_REMOVE(p, p_hash);
922 	sx_xunlock(PIDHASHLOCK(p->p_pid));
923 	LIST_REMOVE(p, p_sibling);
924 	reaper_abandon_children(p, true);
925 	reaper_clear(p);
926 	PROC_LOCK(p);
927 	proc_clear_orphan(p);
928 	PROC_UNLOCK(p);
929 	leavepgrp(p);
930 	if (p->p_procdesc != NULL)
931 		procdesc_reap(p);
932 	sx_xunlock(&proctree_lock);
933 
934 	proc_id_clear(PROC_ID_PID, p->p_pid);
935 
936 	PROC_LOCK(p);
937 	knlist_detach(p->p_klist);
938 	p->p_klist = NULL;
939 	PROC_UNLOCK(p);
940 
941 	/*
942 	 * Removal from allproc list and process group list paired with
943 	 * PROC_LOCK which was executed during that time should guarantee
944 	 * nothing can reach this process anymore. As such further locking
945 	 * is unnecessary.
946 	 */
947 	p->p_xexit = p->p_xsig = 0;		/* XXX: why? */
948 
949 	PROC_LOCK(q);
950 	ruadd(&q->p_stats->p_cru, &q->p_crux, &p->p_ru, &p->p_rux);
951 	PROC_UNLOCK(q);
952 
953 	/*
954 	 * Decrement the count of procs running with this uid.
955 	 */
956 	(void)chgproccnt(p->p_ucred->cr_ruidinfo, -1, 0);
957 
958 	/*
959 	 * Destroy resource accounting information associated with the process.
960 	 */
961 #ifdef RACCT
962 	if (racct_enable) {
963 		PROC_LOCK(p);
964 		racct_sub(p, RACCT_NPROC, 1);
965 		PROC_UNLOCK(p);
966 	}
967 #endif
968 	racct_proc_exit(p);
969 
970 	/*
971 	 * Free credentials, arguments, and sigacts.
972 	 */
973 	proc_unset_cred(p);
974 	pargs_drop(p->p_args);
975 	p->p_args = NULL;
976 	sigacts_free(p->p_sigacts);
977 	p->p_sigacts = NULL;
978 
979 	/*
980 	 * Do any thread-system specific cleanups.
981 	 */
982 	thread_wait(p);
983 
984 	/*
985 	 * Give vm and machine-dependent layer a chance to free anything that
986 	 * cpu_exit couldn't release while still running in process context.
987 	 */
988 	vm_waitproc(p);
989 #ifdef MAC
990 	mac_proc_destroy(p);
991 #endif
992 
993 	KASSERT(FIRST_THREAD_IN_PROC(p),
994 	    ("proc_reap: no residual thread!"));
995 	uma_zfree(proc_zone, p);
996 	atomic_add_int(&nprocs, -1);
997 }
998 
999 static int
1000 proc_to_reap(struct thread *td, struct proc *p, idtype_t idtype, id_t id,
1001     int *status, int options, struct __wrusage *wrusage, siginfo_t *siginfo,
1002     int check_only)
1003 {
1004 	struct rusage *rup;
1005 
1006 	sx_assert(&proctree_lock, SA_XLOCKED);
1007 
1008 	PROC_LOCK(p);
1009 
1010 	switch (idtype) {
1011 	case P_ALL:
1012 		if (p->p_procdesc == NULL ||
1013 		   (p->p_pptr == td->td_proc &&
1014 		   (p->p_flag & P_TRACED) != 0)) {
1015 			break;
1016 		}
1017 
1018 		PROC_UNLOCK(p);
1019 		return (0);
1020 	case P_PID:
1021 		if (p->p_pid != (pid_t)id) {
1022 			PROC_UNLOCK(p);
1023 			return (0);
1024 		}
1025 		break;
1026 	case P_PGID:
1027 		if (p->p_pgid != (pid_t)id) {
1028 			PROC_UNLOCK(p);
1029 			return (0);
1030 		}
1031 		break;
1032 	case P_SID:
1033 		if (p->p_session->s_sid != (pid_t)id) {
1034 			PROC_UNLOCK(p);
1035 			return (0);
1036 		}
1037 		break;
1038 	case P_UID:
1039 		if (p->p_ucred->cr_uid != (uid_t)id) {
1040 			PROC_UNLOCK(p);
1041 			return (0);
1042 		}
1043 		break;
1044 	case P_GID:
1045 		if (p->p_ucred->cr_gid != (gid_t)id) {
1046 			PROC_UNLOCK(p);
1047 			return (0);
1048 		}
1049 		break;
1050 	case P_JAILID:
1051 		if (p->p_ucred->cr_prison->pr_id != (int)id) {
1052 			PROC_UNLOCK(p);
1053 			return (0);
1054 		}
1055 		break;
1056 	/*
1057 	 * It seems that the thread structures get zeroed out
1058 	 * at process exit.  This makes it impossible to
1059 	 * support P_SETID, P_CID or P_CPUID.
1060 	 */
1061 	default:
1062 		PROC_UNLOCK(p);
1063 		return (0);
1064 	}
1065 
1066 	if (p_canwait(td, p)) {
1067 		PROC_UNLOCK(p);
1068 		return (0);
1069 	}
1070 
1071 	if (((options & WEXITED) == 0) && (p->p_state == PRS_ZOMBIE)) {
1072 		PROC_UNLOCK(p);
1073 		return (0);
1074 	}
1075 
1076 	/*
1077 	 * This special case handles a kthread spawned by linux_clone
1078 	 * (see linux_misc.c).  The linux_wait4 and linux_waitpid
1079 	 * functions need to be able to distinguish between waiting
1080 	 * on a process and waiting on a thread.  It is a thread if
1081 	 * p_sigparent is not SIGCHLD, and the WLINUXCLONE option
1082 	 * signifies we want to wait for threads and not processes.
1083 	 */
1084 	if ((p->p_sigparent != SIGCHLD) ^
1085 	    ((options & WLINUXCLONE) != 0)) {
1086 		PROC_UNLOCK(p);
1087 		return (0);
1088 	}
1089 
1090 	if (siginfo != NULL) {
1091 		bzero(siginfo, sizeof(*siginfo));
1092 		siginfo->si_errno = 0;
1093 
1094 		/*
1095 		 * SUSv4 requires that the si_signo value is always
1096 		 * SIGCHLD. Obey it despite the rfork(2) interface
1097 		 * allows to request other signal for child exit
1098 		 * notification.
1099 		 */
1100 		siginfo->si_signo = SIGCHLD;
1101 
1102 		/*
1103 		 *  This is still a rough estimate.  We will fix the
1104 		 *  cases TRAPPED, STOPPED, and CONTINUED later.
1105 		 */
1106 		if (WCOREDUMP(p->p_xsig)) {
1107 			siginfo->si_code = CLD_DUMPED;
1108 			siginfo->si_status = WTERMSIG(p->p_xsig);
1109 		} else if (WIFSIGNALED(p->p_xsig)) {
1110 			siginfo->si_code = CLD_KILLED;
1111 			siginfo->si_status = WTERMSIG(p->p_xsig);
1112 		} else {
1113 			siginfo->si_code = CLD_EXITED;
1114 			siginfo->si_status = p->p_xexit;
1115 		}
1116 
1117 		siginfo->si_pid = p->p_pid;
1118 		siginfo->si_uid = p->p_ucred->cr_uid;
1119 
1120 		/*
1121 		 * The si_addr field would be useful additional
1122 		 * detail, but apparently the PC value may be lost
1123 		 * when we reach this point.  bzero() above sets
1124 		 * siginfo->si_addr to NULL.
1125 		 */
1126 	}
1127 
1128 	/*
1129 	 * There should be no reason to limit resources usage info to
1130 	 * exited processes only.  A snapshot about any resources used
1131 	 * by a stopped process may be exactly what is needed.
1132 	 */
1133 	if (wrusage != NULL) {
1134 		rup = &wrusage->wru_self;
1135 		*rup = p->p_ru;
1136 		PROC_STATLOCK(p);
1137 		calcru(p, &rup->ru_utime, &rup->ru_stime);
1138 		PROC_STATUNLOCK(p);
1139 
1140 		rup = &wrusage->wru_children;
1141 		*rup = p->p_stats->p_cru;
1142 		calccru(p, &rup->ru_utime, &rup->ru_stime);
1143 	}
1144 
1145 	if (p->p_state == PRS_ZOMBIE && !check_only) {
1146 		proc_reap(td, p, status, options);
1147 		return (-1);
1148 	}
1149 	return (1);
1150 }
1151 
1152 int
1153 kern_wait(struct thread *td, pid_t pid, int *status, int options,
1154     struct rusage *rusage)
1155 {
1156 	struct __wrusage wru, *wrup;
1157 	idtype_t idtype;
1158 	id_t id;
1159 	int ret;
1160 
1161 	/*
1162 	 * Translate the special pid values into the (idtype, pid)
1163 	 * pair for kern_wait6.  The WAIT_MYPGRP case is handled by
1164 	 * kern_wait6() on its own.
1165 	 */
1166 	if (pid == WAIT_ANY) {
1167 		idtype = P_ALL;
1168 		id = 0;
1169 	} else if (pid < 0) {
1170 		idtype = P_PGID;
1171 		id = (id_t)-pid;
1172 	} else {
1173 		idtype = P_PID;
1174 		id = (id_t)pid;
1175 	}
1176 
1177 	if (rusage != NULL)
1178 		wrup = &wru;
1179 	else
1180 		wrup = NULL;
1181 
1182 	/*
1183 	 * For backward compatibility we implicitly add flags WEXITED
1184 	 * and WTRAPPED here.
1185 	 */
1186 	options |= WEXITED | WTRAPPED;
1187 	ret = kern_wait6(td, idtype, id, status, options, wrup, NULL);
1188 	if (rusage != NULL)
1189 		*rusage = wru.wru_self;
1190 	return (ret);
1191 }
1192 
1193 static void
1194 report_alive_proc(struct thread *td, struct proc *p, siginfo_t *siginfo,
1195     int *status, int options, int si_code)
1196 {
1197 	bool cont;
1198 
1199 	PROC_LOCK_ASSERT(p, MA_OWNED);
1200 	sx_assert(&proctree_lock, SA_XLOCKED);
1201 	MPASS(si_code == CLD_TRAPPED || si_code == CLD_STOPPED ||
1202 	    si_code == CLD_CONTINUED);
1203 
1204 	cont = si_code == CLD_CONTINUED;
1205 	if ((options & WNOWAIT) == 0) {
1206 		if (cont)
1207 			p->p_flag &= ~P_CONTINUED;
1208 		else
1209 			p->p_flag |= P_WAITED;
1210 		PROC_LOCK(td->td_proc);
1211 		sigqueue_take(p->p_ksi);
1212 		PROC_UNLOCK(td->td_proc);
1213 	}
1214 	sx_xunlock(&proctree_lock);
1215 	if (siginfo != NULL) {
1216 		siginfo->si_code = si_code;
1217 		siginfo->si_status = cont ? SIGCONT : p->p_xsig;
1218 	}
1219 	if (status != NULL)
1220 		*status = cont ? SIGCONT : W_STOPCODE(p->p_xsig);
1221 	PROC_UNLOCK(p);
1222 	td->td_retval[0] = p->p_pid;
1223 }
1224 
1225 int
1226 kern_wait6(struct thread *td, idtype_t idtype, id_t id, int *status,
1227     int options, struct __wrusage *wrusage, siginfo_t *siginfo)
1228 {
1229 	struct proc *p, *q;
1230 	pid_t pid;
1231 	int error, nfound, ret;
1232 	bool report;
1233 
1234 	AUDIT_ARG_VALUE((int)idtype);	/* XXX - This is likely wrong! */
1235 	AUDIT_ARG_PID((pid_t)id);	/* XXX - This may be wrong! */
1236 	AUDIT_ARG_VALUE(options);
1237 
1238 	q = td->td_proc;
1239 
1240 	if ((pid_t)id == WAIT_MYPGRP && (idtype == P_PID || idtype == P_PGID)) {
1241 		PROC_LOCK(q);
1242 		id = (id_t)q->p_pgid;
1243 		PROC_UNLOCK(q);
1244 		idtype = P_PGID;
1245 	}
1246 
1247 	/* If we don't know the option, just return. */
1248 	if ((options & ~(WUNTRACED | WNOHANG | WCONTINUED | WNOWAIT |
1249 	    WEXITED | WTRAPPED | WLINUXCLONE)) != 0)
1250 		return (EINVAL);
1251 	if ((options & (WEXITED | WUNTRACED | WCONTINUED | WTRAPPED)) == 0) {
1252 		/*
1253 		 * We will be unable to find any matching processes,
1254 		 * because there are no known events to look for.
1255 		 * Prefer to return error instead of blocking
1256 		 * indefinitely.
1257 		 */
1258 		return (EINVAL);
1259 	}
1260 
1261 loop:
1262 	if (q->p_flag & P_STATCHILD) {
1263 		PROC_LOCK(q);
1264 		q->p_flag &= ~P_STATCHILD;
1265 		PROC_UNLOCK(q);
1266 	}
1267 	sx_xlock(&proctree_lock);
1268 loop_locked:
1269 	nfound = 0;
1270 	LIST_FOREACH(p, &q->p_children, p_sibling) {
1271 		pid = p->p_pid;
1272 		ret = proc_to_reap(td, p, idtype, id, status, options,
1273 		    wrusage, siginfo, 0);
1274 		if (ret == 0)
1275 			continue;
1276 		else if (ret != 1) {
1277 			td->td_retval[0] = pid;
1278 			return (0);
1279 		}
1280 
1281 		nfound++;
1282 		PROC_LOCK_ASSERT(p, MA_OWNED);
1283 
1284 		if ((options & WTRAPPED) != 0 &&
1285 		    (p->p_flag & P_TRACED) != 0) {
1286 			PROC_SLOCK(p);
1287 			report =
1288 			    ((p->p_flag & (P_STOPPED_TRACE | P_STOPPED_SIG)) &&
1289 			    p->p_suspcount == p->p_numthreads &&
1290 			    (p->p_flag & P_WAITED) == 0);
1291 			PROC_SUNLOCK(p);
1292 			if (report) {
1293 			CTR4(KTR_PTRACE,
1294 			    "wait: returning trapped pid %d status %#x "
1295 			    "(xstat %d) xthread %d",
1296 			    p->p_pid, W_STOPCODE(p->p_xsig), p->p_xsig,
1297 			    p->p_xthread != NULL ?
1298 			    p->p_xthread->td_tid : -1);
1299 				report_alive_proc(td, p, siginfo, status,
1300 				    options, CLD_TRAPPED);
1301 				return (0);
1302 			}
1303 		}
1304 		if ((options & WUNTRACED) != 0 &&
1305 		    (p->p_flag & P_STOPPED_SIG) != 0) {
1306 			PROC_SLOCK(p);
1307 			report = (p->p_suspcount == p->p_numthreads &&
1308 			    ((p->p_flag & P_WAITED) == 0));
1309 			PROC_SUNLOCK(p);
1310 			if (report) {
1311 				report_alive_proc(td, p, siginfo, status,
1312 				    options, CLD_STOPPED);
1313 				return (0);
1314 			}
1315 		}
1316 		if ((options & WCONTINUED) != 0 &&
1317 		    (p->p_flag & P_CONTINUED) != 0) {
1318 			report_alive_proc(td, p, siginfo, status, options,
1319 			    CLD_CONTINUED);
1320 			return (0);
1321 		}
1322 		PROC_UNLOCK(p);
1323 	}
1324 
1325 	/*
1326 	 * Look in the orphans list too, to allow the parent to
1327 	 * collect it's child exit status even if child is being
1328 	 * debugged.
1329 	 *
1330 	 * Debugger detaches from the parent upon successful
1331 	 * switch-over from parent to child.  At this point due to
1332 	 * re-parenting the parent loses the child to debugger and a
1333 	 * wait4(2) call would report that it has no children to wait
1334 	 * for.  By maintaining a list of orphans we allow the parent
1335 	 * to successfully wait until the child becomes a zombie.
1336 	 */
1337 	if (nfound == 0) {
1338 		LIST_FOREACH(p, &q->p_orphans, p_orphan) {
1339 			ret = proc_to_reap(td, p, idtype, id, NULL, options,
1340 			    NULL, NULL, 1);
1341 			if (ret != 0) {
1342 				KASSERT(ret != -1, ("reaped an orphan (pid %d)",
1343 				    (int)td->td_retval[0]));
1344 				PROC_UNLOCK(p);
1345 				nfound++;
1346 				break;
1347 			}
1348 		}
1349 	}
1350 	if (nfound == 0) {
1351 		sx_xunlock(&proctree_lock);
1352 		return (ECHILD);
1353 	}
1354 	if (options & WNOHANG) {
1355 		sx_xunlock(&proctree_lock);
1356 		td->td_retval[0] = 0;
1357 		return (0);
1358 	}
1359 	PROC_LOCK(q);
1360 	if (q->p_flag & P_STATCHILD) {
1361 		q->p_flag &= ~P_STATCHILD;
1362 		PROC_UNLOCK(q);
1363 		goto loop_locked;
1364 	}
1365 	sx_xunlock(&proctree_lock);
1366 	error = msleep(q, &q->p_mtx, PWAIT | PCATCH | PDROP, "wait", 0);
1367 	if (error)
1368 		return (error);
1369 	goto loop;
1370 }
1371 
1372 void
1373 proc_add_orphan(struct proc *child, struct proc *parent)
1374 {
1375 
1376 	sx_assert(&proctree_lock, SX_XLOCKED);
1377 	KASSERT((child->p_flag & P_TRACED) != 0,
1378 	    ("proc_add_orphan: not traced"));
1379 
1380 	if (LIST_EMPTY(&parent->p_orphans)) {
1381 		child->p_treeflag |= P_TREE_FIRST_ORPHAN;
1382 		LIST_INSERT_HEAD(&parent->p_orphans, child, p_orphan);
1383 	} else {
1384 		LIST_INSERT_AFTER(LIST_FIRST(&parent->p_orphans),
1385 		    child, p_orphan);
1386 	}
1387 	child->p_treeflag |= P_TREE_ORPHANED;
1388 }
1389 
1390 /*
1391  * Make process 'parent' the new parent of process 'child'.
1392  * Must be called with an exclusive hold of proctree lock.
1393  */
1394 void
1395 proc_reparent(struct proc *child, struct proc *parent, bool set_oppid)
1396 {
1397 
1398 	sx_assert(&proctree_lock, SX_XLOCKED);
1399 	PROC_LOCK_ASSERT(child, MA_OWNED);
1400 	if (child->p_pptr == parent)
1401 		return;
1402 
1403 	PROC_LOCK(child->p_pptr);
1404 	sigqueue_take(child->p_ksi);
1405 	PROC_UNLOCK(child->p_pptr);
1406 	LIST_REMOVE(child, p_sibling);
1407 	LIST_INSERT_HEAD(&parent->p_children, child, p_sibling);
1408 
1409 	proc_clear_orphan(child);
1410 	if ((child->p_flag & P_TRACED) != 0) {
1411 		proc_add_orphan(child, child->p_pptr);
1412 	}
1413 
1414 	child->p_pptr = parent;
1415 	if (set_oppid)
1416 		child->p_oppid = parent->p_pid;
1417 }
1418