xref: /freebsd/sys/kern/kern_exit.c (revision 22cf89c938886d14f5796fc49f9f020c23ea8eaf)
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 #include "opt_ddb.h"
41 #include "opt_ktrace.h"
42 
43 #include <sys/param.h>
44 #include <sys/systm.h>
45 #include <sys/sysproto.h>
46 #include <sys/capsicum.h>
47 #include <sys/eventhandler.h>
48 #include <sys/kernel.h>
49 #include <sys/ktr.h>
50 #include <sys/malloc.h>
51 #include <sys/lock.h>
52 #include <sys/mutex.h>
53 #include <sys/proc.h>
54 #include <sys/procdesc.h>
55 #include <sys/jail.h>
56 #include <sys/tty.h>
57 #include <sys/wait.h>
58 #include <sys/vmmeter.h>
59 #include <sys/vnode.h>
60 #include <sys/racct.h>
61 #include <sys/resourcevar.h>
62 #include <sys/sbuf.h>
63 #include <sys/signalvar.h>
64 #include <sys/sched.h>
65 #include <sys/sx.h>
66 #include <sys/syscallsubr.h>
67 #include <sys/sysctl.h>
68 #include <sys/syslog.h>
69 #include <sys/ptrace.h>
70 #include <sys/acct.h>		/* for acct_process() function prototype */
71 #include <sys/filedesc.h>
72 #include <sys/sdt.h>
73 #include <sys/shm.h>
74 #include <sys/sem.h>
75 #include <sys/sysent.h>
76 #include <sys/timers.h>
77 #include <sys/umtxvar.h>
78 #ifdef KTRACE
79 #include <sys/ktrace.h>
80 #endif
81 
82 #include <security/audit/audit.h>
83 #include <security/mac/mac_framework.h>
84 
85 #include <vm/vm.h>
86 #include <vm/vm_extern.h>
87 #include <vm/vm_param.h>
88 #include <vm/pmap.h>
89 #include <vm/vm_map.h>
90 #include <vm/vm_page.h>
91 #include <vm/uma.h>
92 
93 #ifdef KDTRACE_HOOKS
94 #include <sys/dtrace_bsd.h>
95 dtrace_execexit_func_t	dtrace_fasttrap_exit;
96 #endif
97 
98 SDT_PROVIDER_DECLARE(proc);
99 SDT_PROBE_DEFINE1(proc, , , exit, "int");
100 
101 static int kern_kill_on_dbg_exit = 1;
102 SYSCTL_INT(_kern, OID_AUTO, kill_on_debugger_exit, CTLFLAG_RWTUN,
103     &kern_kill_on_dbg_exit, 0,
104     "Kill ptraced processes when debugger exits");
105 
106 static bool kern_wait_dequeue_sigchld = 1;
107 SYSCTL_BOOL(_kern, OID_AUTO, wait_dequeue_sigchld, CTLFLAG_RWTUN,
108     &kern_wait_dequeue_sigchld, 0,
109     "Dequeue SIGCHLD on wait(2) for live process");
110 
111 struct proc *
112 proc_realparent(struct proc *child)
113 {
114 	struct proc *p, *parent;
115 
116 	sx_assert(&proctree_lock, SX_LOCKED);
117 	if ((child->p_treeflag & P_TREE_ORPHANED) == 0)
118 		return (child->p_pptr->p_pid == child->p_oppid ?
119 		    child->p_pptr : child->p_reaper);
120 	for (p = child; (p->p_treeflag & P_TREE_FIRST_ORPHAN) == 0;) {
121 		/* Cannot use LIST_PREV(), since the list head is not known. */
122 		p = __containerof(p->p_orphan.le_prev, struct proc,
123 		    p_orphan.le_next);
124 		KASSERT((p->p_treeflag & P_TREE_ORPHANED) != 0,
125 		    ("missing P_ORPHAN %p", p));
126 	}
127 	parent = __containerof(p->p_orphan.le_prev, struct proc,
128 	    p_orphans.lh_first);
129 	return (parent);
130 }
131 
132 void
133 reaper_abandon_children(struct proc *p, bool exiting)
134 {
135 	struct proc *p1, *p2, *ptmp;
136 
137 	sx_assert(&proctree_lock, SX_XLOCKED);
138 	KASSERT(p != initproc, ("reaper_abandon_children for initproc"));
139 	if ((p->p_treeflag & P_TREE_REAPER) == 0)
140 		return;
141 	p1 = p->p_reaper;
142 	LIST_FOREACH_SAFE(p2, &p->p_reaplist, p_reapsibling, ptmp) {
143 		LIST_REMOVE(p2, p_reapsibling);
144 		p2->p_reaper = p1;
145 		p2->p_reapsubtree = p->p_reapsubtree;
146 		LIST_INSERT_HEAD(&p1->p_reaplist, p2, p_reapsibling);
147 		if (exiting && p2->p_pptr == p) {
148 			PROC_LOCK(p2);
149 			proc_reparent(p2, p1, true);
150 			PROC_UNLOCK(p2);
151 		}
152 	}
153 	KASSERT(LIST_EMPTY(&p->p_reaplist), ("p_reaplist not empty"));
154 	p->p_treeflag &= ~P_TREE_REAPER;
155 }
156 
157 static void
158 reaper_clear(struct proc *p)
159 {
160 	struct proc *p1;
161 	bool clear;
162 
163 	sx_assert(&proctree_lock, SX_LOCKED);
164 	LIST_REMOVE(p, p_reapsibling);
165 	if (p->p_reapsubtree == 1)
166 		return;
167 	clear = true;
168 	LIST_FOREACH(p1, &p->p_reaper->p_reaplist, p_reapsibling) {
169 		if (p1->p_reapsubtree == p->p_reapsubtree) {
170 			clear = false;
171 			break;
172 		}
173 	}
174 	if (clear)
175 		proc_id_clear(PROC_ID_REAP, p->p_reapsubtree);
176 }
177 
178 void
179 proc_clear_orphan(struct proc *p)
180 {
181 	struct proc *p1;
182 
183 	sx_assert(&proctree_lock, SA_XLOCKED);
184 	if ((p->p_treeflag & P_TREE_ORPHANED) == 0)
185 		return;
186 	if ((p->p_treeflag & P_TREE_FIRST_ORPHAN) != 0) {
187 		p1 = LIST_NEXT(p, p_orphan);
188 		if (p1 != NULL)
189 			p1->p_treeflag |= P_TREE_FIRST_ORPHAN;
190 		p->p_treeflag &= ~P_TREE_FIRST_ORPHAN;
191 	}
192 	LIST_REMOVE(p, p_orphan);
193 	p->p_treeflag &= ~P_TREE_ORPHANED;
194 }
195 
196 void
197 exit_onexit(struct proc *p)
198 {
199 	MPASS(p->p_numthreads == 1);
200 	umtx_thread_exit(FIRST_THREAD_IN_PROC(p));
201 }
202 
203 /*
204  * exit -- death of process.
205  */
206 int
207 sys_exit(struct thread *td, struct exit_args *uap)
208 {
209 
210 	exit1(td, uap->rval, 0);
211 	__unreachable();
212 }
213 
214 void
215 proc_set_p2_wexit(struct proc *p)
216 {
217 	PROC_LOCK_ASSERT(p, MA_OWNED);
218 	p->p_flag2 |= P2_WEXIT;
219 }
220 
221 /*
222  * Exit: deallocate address space and other resources, change proc state to
223  * zombie, and unlink proc from allproc and parent's lists.  Save exit status
224  * and rusage for wait().  Check for child processes and orphan them.
225  */
226 void
227 exit1(struct thread *td, int rval, int signo)
228 {
229 	struct proc *p, *nq, *q, *t;
230 	struct thread *tdt;
231 	ksiginfo_t *ksi, *ksi1;
232 	int signal_parent;
233 
234 	mtx_assert(&Giant, MA_NOTOWNED);
235 	KASSERT(rval == 0 || signo == 0, ("exit1 rv %d sig %d", rval, signo));
236 	TSPROCEXIT(td->td_proc->p_pid);
237 
238 	p = td->td_proc;
239 	/*
240 	 * In case we're rebooting we just let init die in order to
241 	 * work around an issues where pid 1 might get a fatal signal.
242 	 * For instance, if network interface serving NFS root is
243 	 * going down due to reboot, page-in requests for text are
244 	 * failing.
245 	 */
246 	if (p == initproc && rebooting == 0) {
247 		printf("init died (signal %d, exit %d)\n", signo, rval);
248 		panic("Going nowhere without my init!");
249 	}
250 
251 	/*
252 	 * Process deferred operations, designated with ASTF_KCLEAR.
253 	 * For instance, we need to deref SU mp, since the thread does
254 	 * not return to userspace, and wait for geom to stabilize.
255 	 */
256 	ast_kclear(td);
257 
258 	/*
259 	 * MUST abort all other threads before proceeding past here.
260 	 */
261 	PROC_LOCK(p);
262 	proc_set_p2_wexit(p);
263 
264 	/*
265 	 * First check if some other thread or external request got
266 	 * here before us.  If so, act appropriately: exit or suspend.
267 	 * We must ensure that stop requests are handled before we set
268 	 * P_WEXIT.
269 	 */
270 	thread_suspend_check(0);
271 	while (p->p_flag & P_HADTHREADS) {
272 		/*
273 		 * Kill off the other threads. This requires
274 		 * some co-operation from other parts of the kernel
275 		 * so it may not be instantaneous.  With this state set
276 		 * any thread attempting to interruptibly
277 		 * sleep will return immediately with EINTR or EWOULDBLOCK
278 		 * which will hopefully force them to back out to userland
279 		 * freeing resources as they go.  Any thread attempting
280 		 * to return to userland will thread_exit() from ast().
281 		 * thread_exit() will unsuspend us when the last of the
282 		 * other threads exits.
283 		 * If there is already a thread singler after resumption,
284 		 * calling thread_single() will fail; in that case, we just
285 		 * re-check all suspension request, the thread should
286 		 * either be suspended there or exit.
287 		 */
288 		if (!thread_single(p, SINGLE_EXIT))
289 			/*
290 			 * All other activity in this process is now
291 			 * stopped.  Threading support has been turned
292 			 * off.
293 			 */
294 			break;
295 		/*
296 		 * Recheck for new stop or suspend requests which
297 		 * might appear while process lock was dropped in
298 		 * thread_single().
299 		 */
300 		thread_suspend_check(0);
301 	}
302 	KASSERT(p->p_numthreads == 1,
303 	    ("exit1: proc %p exiting with %d threads", p, p->p_numthreads));
304 	racct_sub(p, RACCT_NTHR, 1);
305 
306 	/* Let event handler change exit status */
307 	p->p_xexit = rval;
308 	p->p_xsig = signo;
309 
310 	/*
311 	 * Ignore any pending request to stop due to a stop signal.
312 	 * Once P_WEXIT is set, future requests will be ignored as
313 	 * well.
314 	 */
315 	p->p_flag &= ~P_STOPPED_SIG;
316 	KASSERT(!P_SHOULDSTOP(p), ("exiting process is stopped"));
317 
318 	/* Note that we are exiting. */
319 	p->p_flag |= P_WEXIT;
320 
321 	/*
322 	 * Wait for any processes that have a hold on our vmspace to
323 	 * release their reference.
324 	 */
325 	while (p->p_lock > 0)
326 		msleep(&p->p_lock, &p->p_mtx, PWAIT, "exithold", 0);
327 
328 	PROC_UNLOCK(p);
329 	/* Drain the limit callout while we don't have the proc locked */
330 	callout_drain(&p->p_limco);
331 
332 #ifdef AUDIT
333 	/*
334 	 * The Sun BSM exit token contains two components: an exit status as
335 	 * passed to exit(), and a return value to indicate what sort of exit
336 	 * it was.  The exit status is WEXITSTATUS(rv), but it's not clear
337 	 * what the return value is.
338 	 */
339 	AUDIT_ARG_EXIT(rval, 0);
340 	AUDIT_SYSCALL_EXIT(0, td);
341 #endif
342 
343 	/* Are we a task leader with peers? */
344 	if (p->p_peers != NULL && p == p->p_leader) {
345 		mtx_lock(&ppeers_lock);
346 		q = p->p_peers;
347 		while (q != NULL) {
348 			PROC_LOCK(q);
349 			kern_psignal(q, SIGKILL);
350 			PROC_UNLOCK(q);
351 			q = q->p_peers;
352 		}
353 		while (p->p_peers != NULL)
354 			msleep(p, &ppeers_lock, PWAIT, "exit1", 0);
355 		mtx_unlock(&ppeers_lock);
356 	}
357 
358 	itimers_exit(p);
359 
360 	/*
361 	 * Check if any loadable modules need anything done at process exit.
362 	 * E.g. SYSV IPC stuff.
363 	 * Event handler could change exit status.
364 	 * XXX what if one of these generates an error?
365 	 */
366 	EVENTHANDLER_DIRECT_INVOKE(process_exit, p);
367 
368 	/*
369 	 * If parent is waiting for us to exit or exec,
370 	 * P_PPWAIT is set; we will wakeup the parent below.
371 	 */
372 	PROC_LOCK(p);
373 	stopprofclock(p);
374 	p->p_ptevents = 0;
375 
376 	/*
377 	 * Stop the real interval timer.  If the handler is currently
378 	 * executing, prevent it from rearming itself and let it finish.
379 	 */
380 	if (timevalisset(&p->p_realtimer.it_value) &&
381 	    callout_stop(&p->p_itcallout) == 0) {
382 		timevalclear(&p->p_realtimer.it_interval);
383 		PROC_UNLOCK(p);
384 		callout_drain(&p->p_itcallout);
385 	} else {
386 		PROC_UNLOCK(p);
387 	}
388 
389 	if (p->p_sysent->sv_onexit != NULL)
390 		p->p_sysent->sv_onexit(p);
391 	seltdfini(td);
392 
393 	/*
394 	 * Reset any sigio structures pointing to us as a result of
395 	 * F_SETOWN with our pid.  The P_WEXIT flag interlocks with fsetown().
396 	 */
397 	funsetownlst(&p->p_sigiolst);
398 
399 	/*
400 	 * Close open files and release open-file table.
401 	 * This may block!
402 	 */
403 	pdescfree(td);
404 	fdescfree(td);
405 
406 	/*
407 	 * Remove ourself from our leader's peer list and wake our leader.
408 	 */
409 	if (p->p_leader->p_peers != NULL) {
410 		mtx_lock(&ppeers_lock);
411 		if (p->p_leader->p_peers != NULL) {
412 			q = p->p_leader;
413 			while (q->p_peers != p)
414 				q = q->p_peers;
415 			q->p_peers = p->p_peers;
416 			wakeup(p->p_leader);
417 		}
418 		mtx_unlock(&ppeers_lock);
419 	}
420 
421 	exec_free_abi_mappings(p);
422 	vmspace_exit(td);
423 	(void)acct_process(td);
424 
425 #ifdef KTRACE
426 	ktrprocexit(td);
427 #endif
428 	/*
429 	 * Release reference to text vnode etc
430 	 */
431 	if (p->p_textvp != NULL) {
432 		vrele(p->p_textvp);
433 		p->p_textvp = NULL;
434 	}
435 	if (p->p_textdvp != NULL) {
436 		vrele(p->p_textdvp);
437 		p->p_textdvp = NULL;
438 	}
439 	if (p->p_binname != NULL) {
440 		free(p->p_binname, M_PARGS);
441 		p->p_binname = NULL;
442 	}
443 
444 	/*
445 	 * Release our limits structure.
446 	 */
447 	lim_free(p->p_limit);
448 	p->p_limit = NULL;
449 
450 	tidhash_remove(td);
451 
452 	/*
453 	 * Call machine-dependent code to release any
454 	 * machine-dependent resources other than the address space.
455 	 * The address space is released by "vmspace_exitfree(p)" in
456 	 * vm_waitproc().
457 	 */
458 	cpu_exit(td);
459 
460 	WITNESS_WARN(WARN_PANIC, NULL, "process (pid %d) exiting", p->p_pid);
461 
462 	/*
463 	 * Remove from allproc. It still sits in the hash.
464 	 */
465 	sx_xlock(&allproc_lock);
466 	LIST_REMOVE(p, p_list);
467 
468 #ifdef DDB
469 	/*
470 	 * Used by ddb's 'ps' command to find this process via the
471 	 * pidhash.
472 	 */
473 	p->p_list.le_prev = NULL;
474 #endif
475 	prison_proc_unlink(p->p_ucred->cr_prison, p);
476 	sx_xunlock(&allproc_lock);
477 
478 	sx_xlock(&proctree_lock);
479 	if ((p->p_flag & (P_TRACED | P_PPWAIT | P_PPTRACE)) != 0) {
480 		PROC_LOCK(p);
481 		p->p_flag &= ~(P_TRACED | P_PPWAIT | P_PPTRACE);
482 		PROC_UNLOCK(p);
483 	}
484 
485 	/*
486 	 * killjobc() might drop and re-acquire proctree_lock to
487 	 * revoke control tty if exiting process was a session leader.
488 	 */
489 	killjobc();
490 
491 	/*
492 	 * Reparent all children processes:
493 	 * - traced ones to the original parent (or init if we are that parent)
494 	 * - the rest to init
495 	 */
496 	q = LIST_FIRST(&p->p_children);
497 	if (q != NULL)		/* only need this if any child is S_ZOMB */
498 		wakeup(q->p_reaper);
499 	for (; q != NULL; q = nq) {
500 		nq = LIST_NEXT(q, p_sibling);
501 		ksi = ksiginfo_alloc(M_WAITOK);
502 		PROC_LOCK(q);
503 		q->p_sigparent = SIGCHLD;
504 
505 		if ((q->p_flag & P_TRACED) == 0) {
506 			proc_reparent(q, q->p_reaper, true);
507 			if (q->p_state == PRS_ZOMBIE) {
508 				/*
509 				 * Inform reaper about the reparented
510 				 * zombie, since wait(2) has something
511 				 * new to report.  Guarantee queueing
512 				 * of the SIGCHLD signal, similar to
513 				 * the _exit() behaviour, by providing
514 				 * our ksiginfo.  Ksi is freed by the
515 				 * signal delivery.
516 				 */
517 				if (q->p_ksi == NULL) {
518 					ksi1 = NULL;
519 				} else {
520 					ksiginfo_copy(q->p_ksi, ksi);
521 					ksi->ksi_flags |= KSI_INS;
522 					ksi1 = ksi;
523 					ksi = NULL;
524 				}
525 				PROC_LOCK(q->p_reaper);
526 				pksignal(q->p_reaper, SIGCHLD, ksi1);
527 				PROC_UNLOCK(q->p_reaper);
528 			} else if (q->p_pdeathsig > 0) {
529 				/*
530 				 * The child asked to received a signal
531 				 * when we exit.
532 				 */
533 				kern_psignal(q, q->p_pdeathsig);
534 			}
535 		} else {
536 			/*
537 			 * Traced processes are killed by default
538 			 * since their existence means someone is
539 			 * screwing up.
540 			 */
541 			t = proc_realparent(q);
542 			if (t == p) {
543 				proc_reparent(q, q->p_reaper, true);
544 			} else {
545 				PROC_LOCK(t);
546 				proc_reparent(q, t, true);
547 				PROC_UNLOCK(t);
548 			}
549 			/*
550 			 * Since q was found on our children list, the
551 			 * proc_reparent() call moved q to the orphan
552 			 * list due to present P_TRACED flag. Clear
553 			 * orphan link for q now while q is locked.
554 			 */
555 			proc_clear_orphan(q);
556 			q->p_flag &= ~P_TRACED;
557 			q->p_flag2 &= ~P2_PTRACE_FSTP;
558 			q->p_ptevents = 0;
559 			p->p_xthread = NULL;
560 			FOREACH_THREAD_IN_PROC(q, tdt) {
561 				tdt->td_dbgflags &= ~(TDB_SUSPEND | TDB_XSIG |
562 				    TDB_FSTP);
563 				tdt->td_xsig = 0;
564 			}
565 			if (kern_kill_on_dbg_exit) {
566 				q->p_flag &= ~P_STOPPED_TRACE;
567 				kern_psignal(q, SIGKILL);
568 			} else if ((q->p_flag & (P_STOPPED_TRACE |
569 			    P_STOPPED_SIG)) != 0) {
570 				sigqueue_delete_proc(q, SIGTRAP);
571 				ptrace_unsuspend(q);
572 			}
573 		}
574 		PROC_UNLOCK(q);
575 		if (ksi != NULL)
576 			ksiginfo_free(ksi);
577 	}
578 
579 	/*
580 	 * Also get rid of our orphans.
581 	 */
582 	while ((q = LIST_FIRST(&p->p_orphans)) != NULL) {
583 		PROC_LOCK(q);
584 		KASSERT(q->p_oppid == p->p_pid,
585 		    ("orphan %p of %p has unexpected oppid %d", q, p,
586 		    q->p_oppid));
587 		q->p_oppid = q->p_reaper->p_pid;
588 
589 		/*
590 		 * If we are the real parent of this process
591 		 * but it has been reparented to a debugger, then
592 		 * check if it asked for a signal when we exit.
593 		 */
594 		if (q->p_pdeathsig > 0)
595 			kern_psignal(q, q->p_pdeathsig);
596 		CTR2(KTR_PTRACE, "exit: pid %d, clearing orphan %d", p->p_pid,
597 		    q->p_pid);
598 		proc_clear_orphan(q);
599 		PROC_UNLOCK(q);
600 	}
601 
602 #ifdef KDTRACE_HOOKS
603 	if (SDT_PROBES_ENABLED()) {
604 		int reason = CLD_EXITED;
605 		if (WCOREDUMP(signo))
606 			reason = CLD_DUMPED;
607 		else if (WIFSIGNALED(signo))
608 			reason = CLD_KILLED;
609 		SDT_PROBE1(proc, , , exit, reason);
610 	}
611 #endif
612 
613 	/* Save exit status. */
614 	PROC_LOCK(p);
615 	p->p_xthread = td;
616 
617 	if (p->p_sysent->sv_ontdexit != NULL)
618 		p->p_sysent->sv_ontdexit(td);
619 
620 #ifdef KDTRACE_HOOKS
621 	/*
622 	 * Tell the DTrace fasttrap provider about the exit if it
623 	 * has declared an interest.
624 	 */
625 	if (dtrace_fasttrap_exit)
626 		dtrace_fasttrap_exit(p);
627 #endif
628 
629 	/*
630 	 * Notify interested parties of our demise.
631 	 */
632 	KNOTE_LOCKED(p->p_klist, NOTE_EXIT);
633 
634 	/*
635 	 * If this is a process with a descriptor, we may not need to deliver
636 	 * a signal to the parent.  proctree_lock is held over
637 	 * procdesc_exit() to serialize concurrent calls to close() and
638 	 * exit().
639 	 */
640 	signal_parent = 0;
641 	if (p->p_procdesc == NULL || procdesc_exit(p)) {
642 		/*
643 		 * Notify parent that we're gone.  If parent has the
644 		 * PS_NOCLDWAIT flag set, or if the handler is set to SIG_IGN,
645 		 * notify process 1 instead (and hope it will handle this
646 		 * situation).
647 		 */
648 		PROC_LOCK(p->p_pptr);
649 		mtx_lock(&p->p_pptr->p_sigacts->ps_mtx);
650 		if (p->p_pptr->p_sigacts->ps_flag &
651 		    (PS_NOCLDWAIT | PS_CLDSIGIGN)) {
652 			struct proc *pp;
653 
654 			mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx);
655 			pp = p->p_pptr;
656 			PROC_UNLOCK(pp);
657 			proc_reparent(p, p->p_reaper, true);
658 			p->p_sigparent = SIGCHLD;
659 			PROC_LOCK(p->p_pptr);
660 
661 			/*
662 			 * Notify parent, so in case he was wait(2)ing or
663 			 * executing waitpid(2) with our pid, he will
664 			 * continue.
665 			 */
666 			wakeup(pp);
667 		} else
668 			mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx);
669 
670 		if (p->p_pptr == p->p_reaper || p->p_pptr == initproc) {
671 			signal_parent = 1;
672 		} else if (p->p_sigparent != 0) {
673 			if (p->p_sigparent == SIGCHLD) {
674 				signal_parent = 1;
675 			} else { /* LINUX thread */
676 				signal_parent = 2;
677 			}
678 		}
679 	} else
680 		PROC_LOCK(p->p_pptr);
681 	sx_xunlock(&proctree_lock);
682 
683 	if (signal_parent == 1) {
684 		childproc_exited(p);
685 	} else if (signal_parent == 2) {
686 		kern_psignal(p->p_pptr, p->p_sigparent);
687 	}
688 
689 	/* Tell the prison that we are gone. */
690 	prison_proc_free(p->p_ucred->cr_prison);
691 
692 	/*
693 	 * The state PRS_ZOMBIE prevents other processes from sending
694 	 * signal to the process, to avoid memory leak, we free memory
695 	 * for signal queue at the time when the state is set.
696 	 */
697 	sigqueue_flush(&p->p_sigqueue);
698 	sigqueue_flush(&td->td_sigqueue);
699 
700 	/*
701 	 * We have to wait until after acquiring all locks before
702 	 * changing p_state.  We need to avoid all possible context
703 	 * switches (including ones from blocking on a mutex) while
704 	 * marked as a zombie.  We also have to set the zombie state
705 	 * before we release the parent process' proc lock to avoid
706 	 * a lost wakeup.  So, we first call wakeup, then we grab the
707 	 * sched lock, update the state, and release the parent process'
708 	 * proc lock.
709 	 */
710 	wakeup(p->p_pptr);
711 	cv_broadcast(&p->p_pwait);
712 	sched_exit(p->p_pptr, td);
713 	PROC_SLOCK(p);
714 	p->p_state = PRS_ZOMBIE;
715 	PROC_UNLOCK(p->p_pptr);
716 
717 	/*
718 	 * Save our children's rusage information in our exit rusage.
719 	 */
720 	PROC_STATLOCK(p);
721 	ruadd(&p->p_ru, &p->p_rux, &p->p_stats->p_cru, &p->p_crux);
722 	PROC_STATUNLOCK(p);
723 
724 	/*
725 	 * Make sure the scheduler takes this thread out of its tables etc.
726 	 * This will also release this thread's reference to the ucred.
727 	 * Other thread parts to release include pcb bits and such.
728 	 */
729 	thread_exit();
730 }
731 
732 #ifndef _SYS_SYSPROTO_H_
733 struct abort2_args {
734 	char *why;
735 	int nargs;
736 	void **args;
737 };
738 #endif
739 
740 int
741 sys_abort2(struct thread *td, struct abort2_args *uap)
742 {
743 	void *uargs[16];
744 	void **uargsp;
745 	int error, nargs;
746 
747 	nargs = uap->nargs;
748 	if (nargs < 0 || nargs > nitems(uargs))
749 		nargs = -1;
750 	uargsp = NULL;
751 	if (nargs > 0) {
752 		if (uap->args != NULL) {
753 			error = copyin(uap->args, uargs,
754 			    nargs * sizeof(void *));
755 			if (error != 0)
756 				nargs = -1;
757 			else
758 				uargsp = uargs;
759 		} else
760 			nargs = -1;
761 	}
762 	return (kern_abort2(td, uap->why, nargs, uargsp));
763 }
764 
765 /*
766  * kern_abort2()
767  * Arguments:
768  *  why - user pointer to why
769  *  nargs - number of arguments copied or -1 if an error occurred in copying
770  *  args - pointer to an array of pointers in kernel format
771  */
772 int
773 kern_abort2(struct thread *td, const char *why, int nargs, void **uargs)
774 {
775 	struct proc *p = td->td_proc;
776 	struct sbuf *sb;
777 	int error, i, sig;
778 
779 	/*
780 	 * Do it right now so we can log either proper call of abort2(), or
781 	 * note, that invalid argument was passed. 512 is big enough to
782 	 * handle 16 arguments' descriptions with additional comments.
783 	 */
784 	sb = sbuf_new(NULL, NULL, 512, SBUF_FIXEDLEN);
785 	sbuf_clear(sb);
786 	sbuf_printf(sb, "%s(pid %d uid %d) aborted: ",
787 	    p->p_comm, p->p_pid, td->td_ucred->cr_uid);
788 	/*
789 	 * Since we can't return from abort2(), send SIGKILL in cases, where
790 	 * abort2() was called improperly
791 	 */
792 	sig = SIGKILL;
793 	/* Prevent from DoSes from user-space. */
794 	if (nargs == -1)
795 		goto out;
796 	KASSERT(nargs >= 0 && nargs <= 16, ("called with too many args (%d)",
797 	    nargs));
798 	/*
799 	 * Limit size of 'reason' string to 128. Will fit even when
800 	 * maximal number of arguments was chosen to be logged.
801 	 */
802 	if (why != NULL) {
803 		error = sbuf_copyin(sb, why, 128);
804 		if (error < 0)
805 			goto out;
806 	} else {
807 		sbuf_printf(sb, "(null)");
808 	}
809 	if (nargs > 0) {
810 		sbuf_printf(sb, "(");
811 		for (i = 0;i < nargs; i++)
812 			sbuf_printf(sb, "%s%p", i == 0 ? "" : ", ", uargs[i]);
813 		sbuf_printf(sb, ")");
814 	}
815 	/*
816 	 * Final stage: arguments were proper, string has been
817 	 * successfully copied from userspace, and copying pointers
818 	 * from user-space succeed.
819 	 */
820 	sig = SIGABRT;
821 out:
822 	if (sig == SIGKILL) {
823 		sbuf_trim(sb);
824 		sbuf_printf(sb, " (Reason text inaccessible)");
825 	}
826 	sbuf_cat(sb, "\n");
827 	sbuf_finish(sb);
828 	log(LOG_INFO, "%s", sbuf_data(sb));
829 	sbuf_delete(sb);
830 	exit1(td, 0, sig);
831 	return (0);
832 }
833 
834 #ifdef COMPAT_43
835 /*
836  * The dirty work is handled by kern_wait().
837  */
838 int
839 owait(struct thread *td, struct owait_args *uap __unused)
840 {
841 	int error, status;
842 
843 	error = kern_wait(td, WAIT_ANY, &status, 0, NULL);
844 	if (error == 0)
845 		td->td_retval[1] = status;
846 	return (error);
847 }
848 #endif /* COMPAT_43 */
849 
850 /*
851  * The dirty work is handled by kern_wait().
852  */
853 int
854 sys_wait4(struct thread *td, struct wait4_args *uap)
855 {
856 	struct rusage ru, *rup;
857 	int error, status;
858 
859 	if (uap->rusage != NULL)
860 		rup = &ru;
861 	else
862 		rup = NULL;
863 	error = kern_wait(td, uap->pid, &status, uap->options, rup);
864 	if (uap->status != NULL && error == 0 && td->td_retval[0] != 0)
865 		error = copyout(&status, uap->status, sizeof(status));
866 	if (uap->rusage != NULL && error == 0 && td->td_retval[0] != 0)
867 		error = copyout(&ru, uap->rusage, sizeof(struct rusage));
868 	return (error);
869 }
870 
871 int
872 sys_wait6(struct thread *td, struct wait6_args *uap)
873 {
874 	struct __wrusage wru, *wrup;
875 	siginfo_t si, *sip;
876 	idtype_t idtype;
877 	id_t id;
878 	int error, status;
879 
880 	idtype = uap->idtype;
881 	id = uap->id;
882 
883 	if (uap->wrusage != NULL)
884 		wrup = &wru;
885 	else
886 		wrup = NULL;
887 
888 	if (uap->info != NULL) {
889 		sip = &si;
890 		bzero(sip, sizeof(*sip));
891 	} else
892 		sip = NULL;
893 
894 	/*
895 	 *  We expect all callers of wait6() to know about WEXITED and
896 	 *  WTRAPPED.
897 	 */
898 	error = kern_wait6(td, idtype, id, &status, uap->options, wrup, sip);
899 
900 	if (uap->status != NULL && error == 0 && td->td_retval[0] != 0)
901 		error = copyout(&status, uap->status, sizeof(status));
902 	if (uap->wrusage != NULL && error == 0 && td->td_retval[0] != 0)
903 		error = copyout(&wru, uap->wrusage, sizeof(wru));
904 	if (uap->info != NULL && error == 0)
905 		error = copyout(&si, uap->info, sizeof(si));
906 	return (error);
907 }
908 
909 /*
910  * Reap the remains of a zombie process and optionally return status and
911  * rusage.  Asserts and will release both the proctree_lock and the process
912  * lock as part of its work.
913  */
914 void
915 proc_reap(struct thread *td, struct proc *p, int *status, int options)
916 {
917 	struct proc *q, *t;
918 
919 	sx_assert(&proctree_lock, SA_XLOCKED);
920 	PROC_LOCK_ASSERT(p, MA_OWNED);
921 	KASSERT(p->p_state == PRS_ZOMBIE, ("proc_reap: !PRS_ZOMBIE"));
922 
923 	mtx_spin_wait_unlocked(&p->p_slock);
924 
925 	q = td->td_proc;
926 
927 	if (status)
928 		*status = KW_EXITCODE(p->p_xexit, p->p_xsig);
929 	if (options & WNOWAIT) {
930 		/*
931 		 *  Only poll, returning the status.  Caller does not wish to
932 		 * release the proc struct just yet.
933 		 */
934 		PROC_UNLOCK(p);
935 		sx_xunlock(&proctree_lock);
936 		return;
937 	}
938 
939 	PROC_LOCK(q);
940 	sigqueue_take(p->p_ksi);
941 	PROC_UNLOCK(q);
942 
943 	/*
944 	 * If we got the child via a ptrace 'attach', we need to give it back
945 	 * to the old parent.
946 	 */
947 	if (p->p_oppid != p->p_pptr->p_pid) {
948 		PROC_UNLOCK(p);
949 		t = proc_realparent(p);
950 		PROC_LOCK(t);
951 		PROC_LOCK(p);
952 		CTR2(KTR_PTRACE,
953 		    "wait: traced child %d moved back to parent %d", p->p_pid,
954 		    t->p_pid);
955 		proc_reparent(p, t, false);
956 		PROC_UNLOCK(p);
957 		pksignal(t, SIGCHLD, p->p_ksi);
958 		wakeup(t);
959 		cv_broadcast(&p->p_pwait);
960 		PROC_UNLOCK(t);
961 		sx_xunlock(&proctree_lock);
962 		return;
963 	}
964 	PROC_UNLOCK(p);
965 
966 	/*
967 	 * Remove other references to this process to ensure we have an
968 	 * exclusive reference.
969 	 */
970 	sx_xlock(PIDHASHLOCK(p->p_pid));
971 	LIST_REMOVE(p, p_hash);
972 	sx_xunlock(PIDHASHLOCK(p->p_pid));
973 	LIST_REMOVE(p, p_sibling);
974 	reaper_abandon_children(p, true);
975 	reaper_clear(p);
976 	PROC_LOCK(p);
977 	proc_clear_orphan(p);
978 	PROC_UNLOCK(p);
979 	leavepgrp(p);
980 	if (p->p_procdesc != NULL)
981 		procdesc_reap(p);
982 	sx_xunlock(&proctree_lock);
983 
984 	proc_id_clear(PROC_ID_PID, p->p_pid);
985 
986 	PROC_LOCK(p);
987 	knlist_detach(p->p_klist);
988 	p->p_klist = NULL;
989 	PROC_UNLOCK(p);
990 
991 	/*
992 	 * Removal from allproc list and process group list paired with
993 	 * PROC_LOCK which was executed during that time should guarantee
994 	 * nothing can reach this process anymore. As such further locking
995 	 * is unnecessary.
996 	 */
997 	p->p_xexit = p->p_xsig = 0;		/* XXX: why? */
998 
999 	PROC_LOCK(q);
1000 	ruadd(&q->p_stats->p_cru, &q->p_crux, &p->p_ru, &p->p_rux);
1001 	PROC_UNLOCK(q);
1002 
1003 	/*
1004 	 * Decrement the count of procs running with this uid.
1005 	 */
1006 	(void)chgproccnt(p->p_ucred->cr_ruidinfo, -1, 0);
1007 
1008 	/*
1009 	 * Destroy resource accounting information associated with the process.
1010 	 */
1011 #ifdef RACCT
1012 	if (racct_enable) {
1013 		PROC_LOCK(p);
1014 		racct_sub(p, RACCT_NPROC, 1);
1015 		PROC_UNLOCK(p);
1016 	}
1017 #endif
1018 	racct_proc_exit(p);
1019 
1020 	/*
1021 	 * Free credentials, arguments, and sigacts.
1022 	 */
1023 	proc_unset_cred(p);
1024 	pargs_drop(p->p_args);
1025 	p->p_args = NULL;
1026 	sigacts_free(p->p_sigacts);
1027 	p->p_sigacts = NULL;
1028 
1029 	/*
1030 	 * Do any thread-system specific cleanups.
1031 	 */
1032 	thread_wait(p);
1033 
1034 	/*
1035 	 * Give vm and machine-dependent layer a chance to free anything that
1036 	 * cpu_exit couldn't release while still running in process context.
1037 	 */
1038 	vm_waitproc(p);
1039 #ifdef MAC
1040 	mac_proc_destroy(p);
1041 #endif
1042 
1043 	KASSERT(FIRST_THREAD_IN_PROC(p),
1044 	    ("proc_reap: no residual thread!"));
1045 	uma_zfree(proc_zone, p);
1046 	atomic_add_int(&nprocs, -1);
1047 }
1048 
1049 static int
1050 proc_to_reap(struct thread *td, struct proc *p, idtype_t idtype, id_t id,
1051     int *status, int options, struct __wrusage *wrusage, siginfo_t *siginfo,
1052     int check_only)
1053 {
1054 	struct rusage *rup;
1055 
1056 	sx_assert(&proctree_lock, SA_XLOCKED);
1057 
1058 	PROC_LOCK(p);
1059 
1060 	switch (idtype) {
1061 	case P_ALL:
1062 		if (p->p_procdesc == NULL ||
1063 		   (p->p_pptr == td->td_proc &&
1064 		   (p->p_flag & P_TRACED) != 0)) {
1065 			break;
1066 		}
1067 
1068 		PROC_UNLOCK(p);
1069 		return (0);
1070 	case P_PID:
1071 		if (p->p_pid != (pid_t)id) {
1072 			PROC_UNLOCK(p);
1073 			return (0);
1074 		}
1075 		break;
1076 	case P_PGID:
1077 		if (p->p_pgid != (pid_t)id) {
1078 			PROC_UNLOCK(p);
1079 			return (0);
1080 		}
1081 		break;
1082 	case P_SID:
1083 		if (p->p_session->s_sid != (pid_t)id) {
1084 			PROC_UNLOCK(p);
1085 			return (0);
1086 		}
1087 		break;
1088 	case P_UID:
1089 		if (p->p_ucred->cr_uid != (uid_t)id) {
1090 			PROC_UNLOCK(p);
1091 			return (0);
1092 		}
1093 		break;
1094 	case P_GID:
1095 		if (p->p_ucred->cr_gid != (gid_t)id) {
1096 			PROC_UNLOCK(p);
1097 			return (0);
1098 		}
1099 		break;
1100 	case P_JAILID:
1101 		if (p->p_ucred->cr_prison->pr_id != (int)id) {
1102 			PROC_UNLOCK(p);
1103 			return (0);
1104 		}
1105 		break;
1106 	/*
1107 	 * It seems that the thread structures get zeroed out
1108 	 * at process exit.  This makes it impossible to
1109 	 * support P_SETID, P_CID or P_CPUID.
1110 	 */
1111 	default:
1112 		PROC_UNLOCK(p);
1113 		return (0);
1114 	}
1115 
1116 	if (p_canwait(td, p)) {
1117 		PROC_UNLOCK(p);
1118 		return (0);
1119 	}
1120 
1121 	if (((options & WEXITED) == 0) && (p->p_state == PRS_ZOMBIE)) {
1122 		PROC_UNLOCK(p);
1123 		return (0);
1124 	}
1125 
1126 	/*
1127 	 * This special case handles a kthread spawned by linux_clone
1128 	 * (see linux_misc.c).  The linux_wait4 and linux_waitpid
1129 	 * functions need to be able to distinguish between waiting
1130 	 * on a process and waiting on a thread.  It is a thread if
1131 	 * p_sigparent is not SIGCHLD, and the WLINUXCLONE option
1132 	 * signifies we want to wait for threads and not processes.
1133 	 */
1134 	if ((p->p_sigparent != SIGCHLD) ^
1135 	    ((options & WLINUXCLONE) != 0)) {
1136 		PROC_UNLOCK(p);
1137 		return (0);
1138 	}
1139 
1140 	if (siginfo != NULL) {
1141 		bzero(siginfo, sizeof(*siginfo));
1142 		siginfo->si_errno = 0;
1143 
1144 		/*
1145 		 * SUSv4 requires that the si_signo value is always
1146 		 * SIGCHLD. Obey it despite the rfork(2) interface
1147 		 * allows to request other signal for child exit
1148 		 * notification.
1149 		 */
1150 		siginfo->si_signo = SIGCHLD;
1151 
1152 		/*
1153 		 *  This is still a rough estimate.  We will fix the
1154 		 *  cases TRAPPED, STOPPED, and CONTINUED later.
1155 		 */
1156 		if (WCOREDUMP(p->p_xsig)) {
1157 			siginfo->si_code = CLD_DUMPED;
1158 			siginfo->si_status = WTERMSIG(p->p_xsig);
1159 		} else if (WIFSIGNALED(p->p_xsig)) {
1160 			siginfo->si_code = CLD_KILLED;
1161 			siginfo->si_status = WTERMSIG(p->p_xsig);
1162 		} else {
1163 			siginfo->si_code = CLD_EXITED;
1164 			siginfo->si_status = p->p_xexit;
1165 		}
1166 
1167 		siginfo->si_pid = p->p_pid;
1168 		siginfo->si_uid = p->p_ucred->cr_uid;
1169 
1170 		/*
1171 		 * The si_addr field would be useful additional
1172 		 * detail, but apparently the PC value may be lost
1173 		 * when we reach this point.  bzero() above sets
1174 		 * siginfo->si_addr to NULL.
1175 		 */
1176 	}
1177 
1178 	/*
1179 	 * There should be no reason to limit resources usage info to
1180 	 * exited processes only.  A snapshot about any resources used
1181 	 * by a stopped process may be exactly what is needed.
1182 	 */
1183 	if (wrusage != NULL) {
1184 		rup = &wrusage->wru_self;
1185 		*rup = p->p_ru;
1186 		PROC_STATLOCK(p);
1187 		calcru(p, &rup->ru_utime, &rup->ru_stime);
1188 		PROC_STATUNLOCK(p);
1189 
1190 		rup = &wrusage->wru_children;
1191 		*rup = p->p_stats->p_cru;
1192 		calccru(p, &rup->ru_utime, &rup->ru_stime);
1193 	}
1194 
1195 	if (p->p_state == PRS_ZOMBIE && !check_only) {
1196 		proc_reap(td, p, status, options);
1197 		return (-1);
1198 	}
1199 	return (1);
1200 }
1201 
1202 int
1203 kern_wait(struct thread *td, pid_t pid, int *status, int options,
1204     struct rusage *rusage)
1205 {
1206 	struct __wrusage wru, *wrup;
1207 	idtype_t idtype;
1208 	id_t id;
1209 	int ret;
1210 
1211 	/*
1212 	 * Translate the special pid values into the (idtype, pid)
1213 	 * pair for kern_wait6.  The WAIT_MYPGRP case is handled by
1214 	 * kern_wait6() on its own.
1215 	 */
1216 	if (pid == WAIT_ANY) {
1217 		idtype = P_ALL;
1218 		id = 0;
1219 	} else if (pid < 0) {
1220 		idtype = P_PGID;
1221 		id = (id_t)-pid;
1222 	} else {
1223 		idtype = P_PID;
1224 		id = (id_t)pid;
1225 	}
1226 
1227 	if (rusage != NULL)
1228 		wrup = &wru;
1229 	else
1230 		wrup = NULL;
1231 
1232 	/*
1233 	 * For backward compatibility we implicitly add flags WEXITED
1234 	 * and WTRAPPED here.
1235 	 */
1236 	options |= WEXITED | WTRAPPED;
1237 	ret = kern_wait6(td, idtype, id, status, options, wrup, NULL);
1238 	if (rusage != NULL)
1239 		*rusage = wru.wru_self;
1240 	return (ret);
1241 }
1242 
1243 static void
1244 report_alive_proc(struct thread *td, struct proc *p, siginfo_t *siginfo,
1245     int *status, int options, int si_code)
1246 {
1247 	bool cont;
1248 
1249 	PROC_LOCK_ASSERT(p, MA_OWNED);
1250 	sx_assert(&proctree_lock, SA_XLOCKED);
1251 	MPASS(si_code == CLD_TRAPPED || si_code == CLD_STOPPED ||
1252 	    si_code == CLD_CONTINUED);
1253 
1254 	cont = si_code == CLD_CONTINUED;
1255 	if ((options & WNOWAIT) == 0) {
1256 		if (cont)
1257 			p->p_flag &= ~P_CONTINUED;
1258 		else
1259 			p->p_flag |= P_WAITED;
1260 		if (kern_wait_dequeue_sigchld &&
1261 		    (td->td_proc->p_sysent->sv_flags & SV_SIG_WAITNDQ) == 0) {
1262 			PROC_LOCK(td->td_proc);
1263 			sigqueue_take(p->p_ksi);
1264 			PROC_UNLOCK(td->td_proc);
1265 		}
1266 	}
1267 	sx_xunlock(&proctree_lock);
1268 	if (siginfo != NULL) {
1269 		siginfo->si_code = si_code;
1270 		siginfo->si_status = cont ? SIGCONT : p->p_xsig;
1271 	}
1272 	if (status != NULL)
1273 		*status = cont ? SIGCONT : W_STOPCODE(p->p_xsig);
1274 	td->td_retval[0] = p->p_pid;
1275 	PROC_UNLOCK(p);
1276 }
1277 
1278 int
1279 kern_wait6(struct thread *td, idtype_t idtype, id_t id, int *status,
1280     int options, struct __wrusage *wrusage, siginfo_t *siginfo)
1281 {
1282 	struct proc *p, *q;
1283 	pid_t pid;
1284 	int error, nfound, ret;
1285 	bool report;
1286 
1287 	AUDIT_ARG_VALUE((int)idtype);	/* XXX - This is likely wrong! */
1288 	AUDIT_ARG_PID((pid_t)id);	/* XXX - This may be wrong! */
1289 	AUDIT_ARG_VALUE(options);
1290 
1291 	q = td->td_proc;
1292 
1293 	if ((pid_t)id == WAIT_MYPGRP && (idtype == P_PID || idtype == P_PGID)) {
1294 		PROC_LOCK(q);
1295 		id = (id_t)q->p_pgid;
1296 		PROC_UNLOCK(q);
1297 		idtype = P_PGID;
1298 	}
1299 
1300 	/* If we don't know the option, just return. */
1301 	if ((options & ~(WUNTRACED | WNOHANG | WCONTINUED | WNOWAIT |
1302 	    WEXITED | WTRAPPED | WLINUXCLONE)) != 0)
1303 		return (EINVAL);
1304 	if ((options & (WEXITED | WUNTRACED | WCONTINUED | WTRAPPED)) == 0) {
1305 		/*
1306 		 * We will be unable to find any matching processes,
1307 		 * because there are no known events to look for.
1308 		 * Prefer to return error instead of blocking
1309 		 * indefinitely.
1310 		 */
1311 		return (EINVAL);
1312 	}
1313 
1314 loop:
1315 	if (q->p_flag & P_STATCHILD) {
1316 		PROC_LOCK(q);
1317 		q->p_flag &= ~P_STATCHILD;
1318 		PROC_UNLOCK(q);
1319 	}
1320 	sx_xlock(&proctree_lock);
1321 loop_locked:
1322 	nfound = 0;
1323 	LIST_FOREACH(p, &q->p_children, p_sibling) {
1324 		pid = p->p_pid;
1325 		ret = proc_to_reap(td, p, idtype, id, status, options,
1326 		    wrusage, siginfo, 0);
1327 		if (ret == 0)
1328 			continue;
1329 		else if (ret != 1) {
1330 			td->td_retval[0] = pid;
1331 			return (0);
1332 		}
1333 
1334 		nfound++;
1335 		PROC_LOCK_ASSERT(p, MA_OWNED);
1336 
1337 		if ((options & WTRAPPED) != 0 &&
1338 		    (p->p_flag & P_TRACED) != 0) {
1339 			PROC_SLOCK(p);
1340 			report =
1341 			    ((p->p_flag & (P_STOPPED_TRACE | P_STOPPED_SIG)) &&
1342 			    p->p_suspcount == p->p_numthreads &&
1343 			    (p->p_flag & P_WAITED) == 0);
1344 			PROC_SUNLOCK(p);
1345 			if (report) {
1346 			CTR4(KTR_PTRACE,
1347 			    "wait: returning trapped pid %d status %#x "
1348 			    "(xstat %d) xthread %d",
1349 			    p->p_pid, W_STOPCODE(p->p_xsig), p->p_xsig,
1350 			    p->p_xthread != NULL ?
1351 			    p->p_xthread->td_tid : -1);
1352 				report_alive_proc(td, p, siginfo, status,
1353 				    options, CLD_TRAPPED);
1354 				return (0);
1355 			}
1356 		}
1357 		if ((options & WUNTRACED) != 0 &&
1358 		    (p->p_flag & P_STOPPED_SIG) != 0) {
1359 			PROC_SLOCK(p);
1360 			report = (p->p_suspcount == p->p_numthreads &&
1361 			    ((p->p_flag & P_WAITED) == 0));
1362 			PROC_SUNLOCK(p);
1363 			if (report) {
1364 				report_alive_proc(td, p, siginfo, status,
1365 				    options, CLD_STOPPED);
1366 				return (0);
1367 			}
1368 		}
1369 		if ((options & WCONTINUED) != 0 &&
1370 		    (p->p_flag & P_CONTINUED) != 0) {
1371 			report_alive_proc(td, p, siginfo, status, options,
1372 			    CLD_CONTINUED);
1373 			return (0);
1374 		}
1375 		PROC_UNLOCK(p);
1376 	}
1377 
1378 	/*
1379 	 * Look in the orphans list too, to allow the parent to
1380 	 * collect it's child exit status even if child is being
1381 	 * debugged.
1382 	 *
1383 	 * Debugger detaches from the parent upon successful
1384 	 * switch-over from parent to child.  At this point due to
1385 	 * re-parenting the parent loses the child to debugger and a
1386 	 * wait4(2) call would report that it has no children to wait
1387 	 * for.  By maintaining a list of orphans we allow the parent
1388 	 * to successfully wait until the child becomes a zombie.
1389 	 */
1390 	if (nfound == 0) {
1391 		LIST_FOREACH(p, &q->p_orphans, p_orphan) {
1392 			ret = proc_to_reap(td, p, idtype, id, NULL, options,
1393 			    NULL, NULL, 1);
1394 			if (ret != 0) {
1395 				KASSERT(ret != -1, ("reaped an orphan (pid %d)",
1396 				    (int)td->td_retval[0]));
1397 				PROC_UNLOCK(p);
1398 				nfound++;
1399 				break;
1400 			}
1401 		}
1402 	}
1403 	if (nfound == 0) {
1404 		sx_xunlock(&proctree_lock);
1405 		return (ECHILD);
1406 	}
1407 	if (options & WNOHANG) {
1408 		sx_xunlock(&proctree_lock);
1409 		td->td_retval[0] = 0;
1410 		return (0);
1411 	}
1412 	PROC_LOCK(q);
1413 	if (q->p_flag & P_STATCHILD) {
1414 		q->p_flag &= ~P_STATCHILD;
1415 		PROC_UNLOCK(q);
1416 		goto loop_locked;
1417 	}
1418 	sx_xunlock(&proctree_lock);
1419 	error = msleep(q, &q->p_mtx, PWAIT | PCATCH | PDROP, "wait", 0);
1420 	if (error)
1421 		return (error);
1422 	goto loop;
1423 }
1424 
1425 void
1426 proc_add_orphan(struct proc *child, struct proc *parent)
1427 {
1428 
1429 	sx_assert(&proctree_lock, SX_XLOCKED);
1430 	KASSERT((child->p_flag & P_TRACED) != 0,
1431 	    ("proc_add_orphan: not traced"));
1432 
1433 	if (LIST_EMPTY(&parent->p_orphans)) {
1434 		child->p_treeflag |= P_TREE_FIRST_ORPHAN;
1435 		LIST_INSERT_HEAD(&parent->p_orphans, child, p_orphan);
1436 	} else {
1437 		LIST_INSERT_AFTER(LIST_FIRST(&parent->p_orphans),
1438 		    child, p_orphan);
1439 	}
1440 	child->p_treeflag |= P_TREE_ORPHANED;
1441 }
1442 
1443 /*
1444  * Make process 'parent' the new parent of process 'child'.
1445  * Must be called with an exclusive hold of proctree lock.
1446  */
1447 void
1448 proc_reparent(struct proc *child, struct proc *parent, bool set_oppid)
1449 {
1450 
1451 	sx_assert(&proctree_lock, SX_XLOCKED);
1452 	PROC_LOCK_ASSERT(child, MA_OWNED);
1453 	if (child->p_pptr == parent)
1454 		return;
1455 
1456 	PROC_LOCK(child->p_pptr);
1457 	sigqueue_take(child->p_ksi);
1458 	PROC_UNLOCK(child->p_pptr);
1459 	LIST_REMOVE(child, p_sibling);
1460 	LIST_INSERT_HEAD(&parent->p_children, child, p_sibling);
1461 
1462 	proc_clear_orphan(child);
1463 	if ((child->p_flag & P_TRACED) != 0) {
1464 		proc_add_orphan(child, child->p_pptr);
1465 	}
1466 
1467 	child->p_pptr = parent;
1468 	if (set_oppid)
1469 		child->p_oppid = parent->p_pid;
1470 }
1471