xref: /freebsd/sys/kern/kern_exit.c (revision 9f23cbd6cae82fd77edfad7173432fa8dccd0a95)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1982, 1986, 1989, 1991, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  * (c) UNIX System Laboratories, Inc.
7  * All or some portions of this file are derived from material licensed
8  * to the University of California by American Telephone and Telegraph
9  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
10  * the permission of UNIX System Laboratories, Inc.
11  *
12  * Redistribution and use in source and binary forms, with or without
13  * modification, are permitted provided that the following conditions
14  * are met:
15  * 1. Redistributions of source code must retain the above copyright
16  *    notice, this list of conditions and the following disclaimer.
17  * 2. Redistributions in binary form must reproduce the above copyright
18  *    notice, this list of conditions and the following disclaimer in the
19  *    documentation and/or other materials provided with the distribution.
20  * 3. Neither the name of the University nor the names of its contributors
21  *    may be used to endorse or promote products derived from this software
22  *    without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  *
36  *	@(#)kern_exit.c	8.7 (Berkeley) 2/12/94
37  */
38 
39 #include <sys/cdefs.h>
40 __FBSDID("$FreeBSD$");
41 
42 #include "opt_ddb.h"
43 #include "opt_ktrace.h"
44 
45 #include <sys/param.h>
46 #include <sys/systm.h>
47 #include <sys/sysproto.h>
48 #include <sys/capsicum.h>
49 #include <sys/eventhandler.h>
50 #include <sys/kernel.h>
51 #include <sys/ktr.h>
52 #include <sys/malloc.h>
53 #include <sys/lock.h>
54 #include <sys/mutex.h>
55 #include <sys/proc.h>
56 #include <sys/procdesc.h>
57 #include <sys/jail.h>
58 #include <sys/tty.h>
59 #include <sys/wait.h>
60 #include <sys/vmmeter.h>
61 #include <sys/vnode.h>
62 #include <sys/racct.h>
63 #include <sys/resourcevar.h>
64 #include <sys/sbuf.h>
65 #include <sys/signalvar.h>
66 #include <sys/sched.h>
67 #include <sys/sx.h>
68 #include <sys/syscallsubr.h>
69 #include <sys/sysctl.h>
70 #include <sys/syslog.h>
71 #include <sys/ptrace.h>
72 #include <sys/acct.h>		/* for acct_process() function prototype */
73 #include <sys/filedesc.h>
74 #include <sys/sdt.h>
75 #include <sys/shm.h>
76 #include <sys/sem.h>
77 #include <sys/sysent.h>
78 #include <sys/timers.h>
79 #include <sys/umtxvar.h>
80 #ifdef KTRACE
81 #include <sys/ktrace.h>
82 #endif
83 
84 #include <security/audit/audit.h>
85 #include <security/mac/mac_framework.h>
86 
87 #include <vm/vm.h>
88 #include <vm/vm_extern.h>
89 #include <vm/vm_param.h>
90 #include <vm/pmap.h>
91 #include <vm/vm_map.h>
92 #include <vm/vm_page.h>
93 #include <vm/uma.h>
94 
95 #ifdef KDTRACE_HOOKS
96 #include <sys/dtrace_bsd.h>
97 dtrace_execexit_func_t	dtrace_fasttrap_exit;
98 #endif
99 
100 SDT_PROVIDER_DECLARE(proc);
101 SDT_PROBE_DEFINE1(proc, , , exit, "int");
102 
103 static int kern_kill_on_dbg_exit = 1;
104 SYSCTL_INT(_kern, OID_AUTO, kill_on_debugger_exit, CTLFLAG_RWTUN,
105     &kern_kill_on_dbg_exit, 0,
106     "Kill ptraced processes when debugger exits");
107 
108 static bool kern_wait_dequeue_sigchld = 1;
109 SYSCTL_BOOL(_kern, OID_AUTO, wait_dequeue_sigchld, CTLFLAG_RWTUN,
110     &kern_wait_dequeue_sigchld, 0,
111     "Dequeue SIGCHLD on wait(2) for live process");
112 
113 struct proc *
114 proc_realparent(struct proc *child)
115 {
116 	struct proc *p, *parent;
117 
118 	sx_assert(&proctree_lock, SX_LOCKED);
119 	if ((child->p_treeflag & P_TREE_ORPHANED) == 0)
120 		return (child->p_pptr->p_pid == child->p_oppid ?
121 		    child->p_pptr : child->p_reaper);
122 	for (p = child; (p->p_treeflag & P_TREE_FIRST_ORPHAN) == 0;) {
123 		/* Cannot use LIST_PREV(), since the list head is not known. */
124 		p = __containerof(p->p_orphan.le_prev, struct proc,
125 		    p_orphan.le_next);
126 		KASSERT((p->p_treeflag & P_TREE_ORPHANED) != 0,
127 		    ("missing P_ORPHAN %p", p));
128 	}
129 	parent = __containerof(p->p_orphan.le_prev, struct proc,
130 	    p_orphans.lh_first);
131 	return (parent);
132 }
133 
134 void
135 reaper_abandon_children(struct proc *p, bool exiting)
136 {
137 	struct proc *p1, *p2, *ptmp;
138 
139 	sx_assert(&proctree_lock, SX_XLOCKED);
140 	KASSERT(p != initproc, ("reaper_abandon_children for initproc"));
141 	if ((p->p_treeflag & P_TREE_REAPER) == 0)
142 		return;
143 	p1 = p->p_reaper;
144 	LIST_FOREACH_SAFE(p2, &p->p_reaplist, p_reapsibling, ptmp) {
145 		LIST_REMOVE(p2, p_reapsibling);
146 		p2->p_reaper = p1;
147 		p2->p_reapsubtree = p->p_reapsubtree;
148 		LIST_INSERT_HEAD(&p1->p_reaplist, p2, p_reapsibling);
149 		if (exiting && p2->p_pptr == p) {
150 			PROC_LOCK(p2);
151 			proc_reparent(p2, p1, true);
152 			PROC_UNLOCK(p2);
153 		}
154 	}
155 	KASSERT(LIST_EMPTY(&p->p_reaplist), ("p_reaplist not empty"));
156 	p->p_treeflag &= ~P_TREE_REAPER;
157 }
158 
159 static void
160 reaper_clear(struct proc *p)
161 {
162 	struct proc *p1;
163 	bool clear;
164 
165 	sx_assert(&proctree_lock, SX_LOCKED);
166 	LIST_REMOVE(p, p_reapsibling);
167 	if (p->p_reapsubtree == 1)
168 		return;
169 	clear = true;
170 	LIST_FOREACH(p1, &p->p_reaper->p_reaplist, p_reapsibling) {
171 		if (p1->p_reapsubtree == p->p_reapsubtree) {
172 			clear = false;
173 			break;
174 		}
175 	}
176 	if (clear)
177 		proc_id_clear(PROC_ID_REAP, p->p_reapsubtree);
178 }
179 
180 void
181 proc_clear_orphan(struct proc *p)
182 {
183 	struct proc *p1;
184 
185 	sx_assert(&proctree_lock, SA_XLOCKED);
186 	if ((p->p_treeflag & P_TREE_ORPHANED) == 0)
187 		return;
188 	if ((p->p_treeflag & P_TREE_FIRST_ORPHAN) != 0) {
189 		p1 = LIST_NEXT(p, p_orphan);
190 		if (p1 != NULL)
191 			p1->p_treeflag |= P_TREE_FIRST_ORPHAN;
192 		p->p_treeflag &= ~P_TREE_FIRST_ORPHAN;
193 	}
194 	LIST_REMOVE(p, p_orphan);
195 	p->p_treeflag &= ~P_TREE_ORPHANED;
196 }
197 
198 void
199 exit_onexit(struct proc *p)
200 {
201 	MPASS(p->p_numthreads == 1);
202 	umtx_thread_exit(FIRST_THREAD_IN_PROC(p));
203 }
204 
205 /*
206  * exit -- death of process.
207  */
208 int
209 sys_exit(struct thread *td, struct exit_args *uap)
210 {
211 
212 	exit1(td, uap->rval, 0);
213 	__unreachable();
214 }
215 
216 void
217 proc_set_p2_wexit(struct proc *p)
218 {
219 	PROC_LOCK_ASSERT(p, MA_OWNED);
220 	p->p_flag2 |= P2_WEXIT;
221 }
222 
223 void
224 exit1(struct thread *td, int rval, int signo)
225 {
226 	exit2(td, rval, signo, false);
227 }
228 
229 /*
230  * Exit: deallocate address space and other resources, change proc state to
231  * zombie, and unlink proc from allproc and parent's lists.  Save exit status
232  * and rusage for wait().  Check for child processes and orphan them.
233  */
234 void
235 exit2(struct thread *td, int rval, int signo, bool dec_killpg_cnt)
236 {
237 	struct proc *p, *nq, *q, *t;
238 	struct thread *tdt;
239 	ksiginfo_t *ksi, *ksi1;
240 	int signal_parent;
241 
242 	mtx_assert(&Giant, MA_NOTOWNED);
243 	KASSERT(rval == 0 || signo == 0, ("exit1 rv %d sig %d", rval, signo));
244 	TSPROCEXIT(td->td_proc->p_pid);
245 
246 	p = td->td_proc;
247 	/*
248 	 * XXX in case we're rebooting we just let init die in order to
249 	 * work around an unsolved stack overflow seen very late during
250 	 * shutdown on sparc64 when the gmirror worker process exists.
251 	 * XXX what to do now that sparc64 is gone... remove if?
252 	 */
253 	if (p == initproc && rebooting == 0) {
254 		printf("init died (signal %d, exit %d)\n", signo, rval);
255 		panic("Going nowhere without my init!");
256 	}
257 
258 	/*
259 	 * Process deferred operations, designated with ASTF_KCLEAR.
260 	 * For instance, we need to deref SU mp, since the thread does
261 	 * not return to userspace, and wait for geom to stabilize.
262 	 */
263 	ast_kclear(td);
264 
265 	/*
266 	 * MUST abort all other threads before proceeding past here.
267 	 */
268 	PROC_LOCK(p);
269 	proc_set_p2_wexit(p);
270 
271 	/*
272 	 * First check if some other thread or external request got
273 	 * here before us.  If so, act appropriately: exit or suspend.
274 	 * We must ensure that stop requests are handled before we set
275 	 * P_WEXIT.
276 	 */
277 	thread_suspend_check(0);
278 	while (p->p_flag & P_HADTHREADS) {
279 		/*
280 		 * Kill off the other threads. This requires
281 		 * some co-operation from other parts of the kernel
282 		 * so it may not be instantaneous.  With this state set
283 		 * any thread attempting to interruptibly
284 		 * sleep will return immediately with EINTR or EWOULDBLOCK
285 		 * which will hopefully force them to back out to userland
286 		 * freeing resources as they go.  Any thread attempting
287 		 * to return to userland will thread_exit() from ast().
288 		 * thread_exit() will unsuspend us when the last of the
289 		 * other threads exits.
290 		 * If there is already a thread singler after resumption,
291 		 * calling thread_single() will fail; in that case, we just
292 		 * re-check all suspension request, the thread should
293 		 * either be suspended there or exit.
294 		 */
295 		if (!thread_single(p, SINGLE_EXIT))
296 			/*
297 			 * All other activity in this process is now
298 			 * stopped.  Threading support has been turned
299 			 * off.
300 			 */
301 			break;
302 		/*
303 		 * Recheck for new stop or suspend requests which
304 		 * might appear while process lock was dropped in
305 		 * thread_single().
306 		 */
307 		thread_suspend_check(0);
308 	}
309 	KASSERT(p->p_numthreads == 1,
310 	    ("exit1: proc %p exiting with %d threads", p, p->p_numthreads));
311 	racct_sub(p, RACCT_NTHR, 1);
312 
313 	if (dec_killpg_cnt) {
314 		MPASS(atomic_load_int(&p->p_killpg_cnt) > 0);
315 		atomic_add_int(&p->p_killpg_cnt, -1);
316 	}
317 
318 	/* Let event handler change exit status */
319 	p->p_xexit = rval;
320 	p->p_xsig = signo;
321 
322 	/*
323 	 * Ignore any pending request to stop due to a stop signal.
324 	 * Once P_WEXIT is set, future requests will be ignored as
325 	 * well.
326 	 */
327 	p->p_flag &= ~P_STOPPED_SIG;
328 	KASSERT(!P_SHOULDSTOP(p), ("exiting process is stopped"));
329 
330 	/* Note that we are exiting. */
331 	p->p_flag |= P_WEXIT;
332 
333 	/*
334 	 * Wait for any processes that have a hold on our vmspace to
335 	 * release their reference.
336 	 */
337 	while (p->p_lock > 0)
338 		msleep(&p->p_lock, &p->p_mtx, PWAIT, "exithold", 0);
339 
340 	PROC_UNLOCK(p);
341 	/* Drain the limit callout while we don't have the proc locked */
342 	callout_drain(&p->p_limco);
343 
344 #ifdef AUDIT
345 	/*
346 	 * The Sun BSM exit token contains two components: an exit status as
347 	 * passed to exit(), and a return value to indicate what sort of exit
348 	 * it was.  The exit status is WEXITSTATUS(rv), but it's not clear
349 	 * what the return value is.
350 	 */
351 	AUDIT_ARG_EXIT(rval, 0);
352 	AUDIT_SYSCALL_EXIT(0, td);
353 #endif
354 
355 	/* Are we a task leader with peers? */
356 	if (p->p_peers != NULL && p == p->p_leader) {
357 		mtx_lock(&ppeers_lock);
358 		q = p->p_peers;
359 		while (q != NULL) {
360 			PROC_LOCK(q);
361 			kern_psignal(q, SIGKILL);
362 			PROC_UNLOCK(q);
363 			q = q->p_peers;
364 		}
365 		while (p->p_peers != NULL)
366 			msleep(p, &ppeers_lock, PWAIT, "exit1", 0);
367 		mtx_unlock(&ppeers_lock);
368 	}
369 
370 	itimers_exit(p);
371 
372 	/*
373 	 * Check if any loadable modules need anything done at process exit.
374 	 * E.g. SYSV IPC stuff.
375 	 * Event handler could change exit status.
376 	 * XXX what if one of these generates an error?
377 	 */
378 	EVENTHANDLER_DIRECT_INVOKE(process_exit, p);
379 
380 	/*
381 	 * If parent is waiting for us to exit or exec,
382 	 * P_PPWAIT is set; we will wakeup the parent below.
383 	 */
384 	PROC_LOCK(p);
385 	stopprofclock(p);
386 	p->p_ptevents = 0;
387 
388 	/*
389 	 * Stop the real interval timer.  If the handler is currently
390 	 * executing, prevent it from rearming itself and let it finish.
391 	 */
392 	if (timevalisset(&p->p_realtimer.it_value) &&
393 	    callout_stop(&p->p_itcallout) == 0) {
394 		timevalclear(&p->p_realtimer.it_interval);
395 		PROC_UNLOCK(p);
396 		callout_drain(&p->p_itcallout);
397 	} else {
398 		PROC_UNLOCK(p);
399 	}
400 
401 	if (p->p_sysent->sv_onexit != NULL)
402 		p->p_sysent->sv_onexit(p);
403 	seltdfini(td);
404 
405 	/*
406 	 * Reset any sigio structures pointing to us as a result of
407 	 * F_SETOWN with our pid.  The P_WEXIT flag interlocks with fsetown().
408 	 */
409 	funsetownlst(&p->p_sigiolst);
410 
411 	/*
412 	 * Close open files and release open-file table.
413 	 * This may block!
414 	 */
415 	pdescfree(td);
416 	fdescfree(td);
417 
418 	/*
419 	 * Remove ourself from our leader's peer list and wake our leader.
420 	 */
421 	if (p->p_leader->p_peers != NULL) {
422 		mtx_lock(&ppeers_lock);
423 		if (p->p_leader->p_peers != NULL) {
424 			q = p->p_leader;
425 			while (q->p_peers != p)
426 				q = q->p_peers;
427 			q->p_peers = p->p_peers;
428 			wakeup(p->p_leader);
429 		}
430 		mtx_unlock(&ppeers_lock);
431 	}
432 
433 	exec_free_abi_mappings(p);
434 	vmspace_exit(td);
435 	(void)acct_process(td);
436 
437 #ifdef KTRACE
438 	ktrprocexit(td);
439 #endif
440 	/*
441 	 * Release reference to text vnode etc
442 	 */
443 	if (p->p_textvp != NULL) {
444 		vrele(p->p_textvp);
445 		p->p_textvp = NULL;
446 	}
447 	if (p->p_textdvp != NULL) {
448 		vrele(p->p_textdvp);
449 		p->p_textdvp = NULL;
450 	}
451 	if (p->p_binname != NULL) {
452 		free(p->p_binname, M_PARGS);
453 		p->p_binname = NULL;
454 	}
455 
456 	/*
457 	 * Release our limits structure.
458 	 */
459 	lim_free(p->p_limit);
460 	p->p_limit = NULL;
461 
462 	tidhash_remove(td);
463 
464 	/*
465 	 * Call machine-dependent code to release any
466 	 * machine-dependent resources other than the address space.
467 	 * The address space is released by "vmspace_exitfree(p)" in
468 	 * vm_waitproc().
469 	 */
470 	cpu_exit(td);
471 
472 	WITNESS_WARN(WARN_PANIC, NULL, "process (pid %d) exiting", p->p_pid);
473 
474 	/*
475 	 * Remove from allproc. It still sits in the hash.
476 	 */
477 	sx_xlock(&allproc_lock);
478 	LIST_REMOVE(p, p_list);
479 
480 #ifdef DDB
481 	/*
482 	 * Used by ddb's 'ps' command to find this process via the
483 	 * pidhash.
484 	 */
485 	p->p_list.le_prev = NULL;
486 #endif
487 	prison_proc_unlink(p->p_ucred->cr_prison, p);
488 	sx_xunlock(&allproc_lock);
489 
490 	sx_xlock(&proctree_lock);
491 	if ((p->p_flag & (P_TRACED | P_PPWAIT | P_PPTRACE)) != 0) {
492 		PROC_LOCK(p);
493 		p->p_flag &= ~(P_TRACED | P_PPWAIT | P_PPTRACE);
494 		PROC_UNLOCK(p);
495 	}
496 
497 	/*
498 	 * killjobc() might drop and re-acquire proctree_lock to
499 	 * revoke control tty if exiting process was a session leader.
500 	 */
501 	killjobc();
502 
503 	/*
504 	 * Reparent all children processes:
505 	 * - traced ones to the original parent (or init if we are that parent)
506 	 * - the rest to init
507 	 */
508 	q = LIST_FIRST(&p->p_children);
509 	if (q != NULL)		/* only need this if any child is S_ZOMB */
510 		wakeup(q->p_reaper);
511 	for (; q != NULL; q = nq) {
512 		nq = LIST_NEXT(q, p_sibling);
513 		ksi = ksiginfo_alloc(M_WAITOK);
514 		PROC_LOCK(q);
515 		q->p_sigparent = SIGCHLD;
516 
517 		if ((q->p_flag & P_TRACED) == 0) {
518 			proc_reparent(q, q->p_reaper, true);
519 			if (q->p_state == PRS_ZOMBIE) {
520 				/*
521 				 * Inform reaper about the reparented
522 				 * zombie, since wait(2) has something
523 				 * new to report.  Guarantee queueing
524 				 * of the SIGCHLD signal, similar to
525 				 * the _exit() behaviour, by providing
526 				 * our ksiginfo.  Ksi is freed by the
527 				 * signal delivery.
528 				 */
529 				if (q->p_ksi == NULL) {
530 					ksi1 = NULL;
531 				} else {
532 					ksiginfo_copy(q->p_ksi, ksi);
533 					ksi->ksi_flags |= KSI_INS;
534 					ksi1 = ksi;
535 					ksi = NULL;
536 				}
537 				PROC_LOCK(q->p_reaper);
538 				pksignal(q->p_reaper, SIGCHLD, ksi1);
539 				PROC_UNLOCK(q->p_reaper);
540 			} else if (q->p_pdeathsig > 0) {
541 				/*
542 				 * The child asked to received a signal
543 				 * when we exit.
544 				 */
545 				kern_psignal(q, q->p_pdeathsig);
546 			}
547 		} else {
548 			/*
549 			 * Traced processes are killed by default
550 			 * since their existence means someone is
551 			 * screwing up.
552 			 */
553 			t = proc_realparent(q);
554 			if (t == p) {
555 				proc_reparent(q, q->p_reaper, true);
556 			} else {
557 				PROC_LOCK(t);
558 				proc_reparent(q, t, true);
559 				PROC_UNLOCK(t);
560 			}
561 			/*
562 			 * Since q was found on our children list, the
563 			 * proc_reparent() call moved q to the orphan
564 			 * list due to present P_TRACED flag. Clear
565 			 * orphan link for q now while q is locked.
566 			 */
567 			proc_clear_orphan(q);
568 			q->p_flag &= ~P_TRACED;
569 			q->p_flag2 &= ~P2_PTRACE_FSTP;
570 			q->p_ptevents = 0;
571 			p->p_xthread = NULL;
572 			FOREACH_THREAD_IN_PROC(q, tdt) {
573 				tdt->td_dbgflags &= ~(TDB_SUSPEND | TDB_XSIG |
574 				    TDB_FSTP);
575 				tdt->td_xsig = 0;
576 			}
577 			if (kern_kill_on_dbg_exit) {
578 				q->p_flag &= ~P_STOPPED_TRACE;
579 				kern_psignal(q, SIGKILL);
580 			} else if ((q->p_flag & (P_STOPPED_TRACE |
581 			    P_STOPPED_SIG)) != 0) {
582 				sigqueue_delete_proc(q, SIGTRAP);
583 				ptrace_unsuspend(q);
584 			}
585 		}
586 		PROC_UNLOCK(q);
587 		if (ksi != NULL)
588 			ksiginfo_free(ksi);
589 	}
590 
591 	/*
592 	 * Also get rid of our orphans.
593 	 */
594 	while ((q = LIST_FIRST(&p->p_orphans)) != NULL) {
595 		PROC_LOCK(q);
596 		KASSERT(q->p_oppid == p->p_pid,
597 		    ("orphan %p of %p has unexpected oppid %d", q, p,
598 		    q->p_oppid));
599 		q->p_oppid = q->p_reaper->p_pid;
600 
601 		/*
602 		 * If we are the real parent of this process
603 		 * but it has been reparented to a debugger, then
604 		 * check if it asked for a signal when we exit.
605 		 */
606 		if (q->p_pdeathsig > 0)
607 			kern_psignal(q, q->p_pdeathsig);
608 		CTR2(KTR_PTRACE, "exit: pid %d, clearing orphan %d", p->p_pid,
609 		    q->p_pid);
610 		proc_clear_orphan(q);
611 		PROC_UNLOCK(q);
612 	}
613 
614 #ifdef KDTRACE_HOOKS
615 	if (SDT_PROBES_ENABLED()) {
616 		int reason = CLD_EXITED;
617 		if (WCOREDUMP(signo))
618 			reason = CLD_DUMPED;
619 		else if (WIFSIGNALED(signo))
620 			reason = CLD_KILLED;
621 		SDT_PROBE1(proc, , , exit, reason);
622 	}
623 #endif
624 
625 	/* Save exit status. */
626 	PROC_LOCK(p);
627 	p->p_xthread = td;
628 
629 	if (p->p_sysent->sv_ontdexit != NULL)
630 		p->p_sysent->sv_ontdexit(td);
631 
632 #ifdef KDTRACE_HOOKS
633 	/*
634 	 * Tell the DTrace fasttrap provider about the exit if it
635 	 * has declared an interest.
636 	 */
637 	if (dtrace_fasttrap_exit)
638 		dtrace_fasttrap_exit(p);
639 #endif
640 
641 	/*
642 	 * Notify interested parties of our demise.
643 	 */
644 	KNOTE_LOCKED(p->p_klist, NOTE_EXIT);
645 
646 	/*
647 	 * If this is a process with a descriptor, we may not need to deliver
648 	 * a signal to the parent.  proctree_lock is held over
649 	 * procdesc_exit() to serialize concurrent calls to close() and
650 	 * exit().
651 	 */
652 	signal_parent = 0;
653 	if (p->p_procdesc == NULL || procdesc_exit(p)) {
654 		/*
655 		 * Notify parent that we're gone.  If parent has the
656 		 * PS_NOCLDWAIT flag set, or if the handler is set to SIG_IGN,
657 		 * notify process 1 instead (and hope it will handle this
658 		 * situation).
659 		 */
660 		PROC_LOCK(p->p_pptr);
661 		mtx_lock(&p->p_pptr->p_sigacts->ps_mtx);
662 		if (p->p_pptr->p_sigacts->ps_flag &
663 		    (PS_NOCLDWAIT | PS_CLDSIGIGN)) {
664 			struct proc *pp;
665 
666 			mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx);
667 			pp = p->p_pptr;
668 			PROC_UNLOCK(pp);
669 			proc_reparent(p, p->p_reaper, true);
670 			p->p_sigparent = SIGCHLD;
671 			PROC_LOCK(p->p_pptr);
672 
673 			/*
674 			 * Notify parent, so in case he was wait(2)ing or
675 			 * executing waitpid(2) with our pid, he will
676 			 * continue.
677 			 */
678 			wakeup(pp);
679 		} else
680 			mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx);
681 
682 		if (p->p_pptr == p->p_reaper || p->p_pptr == initproc) {
683 			signal_parent = 1;
684 		} else if (p->p_sigparent != 0) {
685 			if (p->p_sigparent == SIGCHLD) {
686 				signal_parent = 1;
687 			} else { /* LINUX thread */
688 				signal_parent = 2;
689 			}
690 		}
691 	} else
692 		PROC_LOCK(p->p_pptr);
693 	sx_xunlock(&proctree_lock);
694 
695 	if (signal_parent == 1) {
696 		childproc_exited(p);
697 	} else if (signal_parent == 2) {
698 		kern_psignal(p->p_pptr, p->p_sigparent);
699 	}
700 
701 	/* Tell the prison that we are gone. */
702 	prison_proc_free(p->p_ucred->cr_prison);
703 
704 	/*
705 	 * The state PRS_ZOMBIE prevents other proesses from sending
706 	 * signal to the process, to avoid memory leak, we free memory
707 	 * for signal queue at the time when the state is set.
708 	 */
709 	sigqueue_flush(&p->p_sigqueue);
710 	sigqueue_flush(&td->td_sigqueue);
711 
712 	/*
713 	 * We have to wait until after acquiring all locks before
714 	 * changing p_state.  We need to avoid all possible context
715 	 * switches (including ones from blocking on a mutex) while
716 	 * marked as a zombie.  We also have to set the zombie state
717 	 * before we release the parent process' proc lock to avoid
718 	 * a lost wakeup.  So, we first call wakeup, then we grab the
719 	 * sched lock, update the state, and release the parent process'
720 	 * proc lock.
721 	 */
722 	wakeup(p->p_pptr);
723 	cv_broadcast(&p->p_pwait);
724 	sched_exit(p->p_pptr, td);
725 	PROC_SLOCK(p);
726 	p->p_state = PRS_ZOMBIE;
727 	PROC_UNLOCK(p->p_pptr);
728 
729 	/*
730 	 * Save our children's rusage information in our exit rusage.
731 	 */
732 	PROC_STATLOCK(p);
733 	ruadd(&p->p_ru, &p->p_rux, &p->p_stats->p_cru, &p->p_crux);
734 	PROC_STATUNLOCK(p);
735 
736 	/*
737 	 * Make sure the scheduler takes this thread out of its tables etc.
738 	 * This will also release this thread's reference to the ucred.
739 	 * Other thread parts to release include pcb bits and such.
740 	 */
741 	thread_exit();
742 }
743 
744 #ifndef _SYS_SYSPROTO_H_
745 struct abort2_args {
746 	char *why;
747 	int nargs;
748 	void **args;
749 };
750 #endif
751 
752 int
753 sys_abort2(struct thread *td, struct abort2_args *uap)
754 {
755 	void *uargs[16];
756 	void **uargsp;
757 	int error, nargs;
758 
759 	nargs = uap->nargs;
760 	if (nargs < 0 || nargs > nitems(uargs))
761 		nargs = -1;
762 	uargsp = NULL;
763 	if (nargs > 0) {
764 		if (uap->args != NULL) {
765 			error = copyin(uap->args, uargs,
766 			    nargs * sizeof(void *));
767 			if (error != 0)
768 				nargs = -1;
769 			else
770 				uargsp = uargs;
771 		} else
772 			nargs = -1;
773 	}
774 	return (kern_abort2(td, uap->why, nargs, uargsp));
775 }
776 
777 /*
778  * kern_abort2()
779  * Arguments:
780  *  why - user pointer to why
781  *  nargs - number of arguments copied or -1 if an error occurred in copying
782  *  args - pointer to an array of pointers in kernel format
783  */
784 int
785 kern_abort2(struct thread *td, const char *why, int nargs, void **uargs)
786 {
787 	struct proc *p = td->td_proc;
788 	struct sbuf *sb;
789 	int error, i, sig;
790 
791 	/*
792 	 * Do it right now so we can log either proper call of abort2(), or
793 	 * note, that invalid argument was passed. 512 is big enough to
794 	 * handle 16 arguments' descriptions with additional comments.
795 	 */
796 	sb = sbuf_new(NULL, NULL, 512, SBUF_FIXEDLEN);
797 	sbuf_clear(sb);
798 	sbuf_printf(sb, "%s(pid %d uid %d) aborted: ",
799 	    p->p_comm, p->p_pid, td->td_ucred->cr_uid);
800 	/*
801 	 * Since we can't return from abort2(), send SIGKILL in cases, where
802 	 * abort2() was called improperly
803 	 */
804 	sig = SIGKILL;
805 	/* Prevent from DoSes from user-space. */
806 	if (nargs == -1)
807 		goto out;
808 	KASSERT(nargs >= 0 && nargs <= 16, ("called with too many args (%d)",
809 	    nargs));
810 	/*
811 	 * Limit size of 'reason' string to 128. Will fit even when
812 	 * maximal number of arguments was chosen to be logged.
813 	 */
814 	if (why != NULL) {
815 		error = sbuf_copyin(sb, why, 128);
816 		if (error < 0)
817 			goto out;
818 	} else {
819 		sbuf_printf(sb, "(null)");
820 	}
821 	if (nargs > 0) {
822 		sbuf_printf(sb, "(");
823 		for (i = 0;i < nargs; i++)
824 			sbuf_printf(sb, "%s%p", i == 0 ? "" : ", ", uargs[i]);
825 		sbuf_printf(sb, ")");
826 	}
827 	/*
828 	 * Final stage: arguments were proper, string has been
829 	 * successfully copied from userspace, and copying pointers
830 	 * from user-space succeed.
831 	 */
832 	sig = SIGABRT;
833 out:
834 	if (sig == SIGKILL) {
835 		sbuf_trim(sb);
836 		sbuf_printf(sb, " (Reason text inaccessible)");
837 	}
838 	sbuf_cat(sb, "\n");
839 	sbuf_finish(sb);
840 	log(LOG_INFO, "%s", sbuf_data(sb));
841 	sbuf_delete(sb);
842 	exit1(td, 0, sig);
843 	return (0);
844 }
845 
846 #ifdef COMPAT_43
847 /*
848  * The dirty work is handled by kern_wait().
849  */
850 int
851 owait(struct thread *td, struct owait_args *uap __unused)
852 {
853 	int error, status;
854 
855 	error = kern_wait(td, WAIT_ANY, &status, 0, NULL);
856 	if (error == 0)
857 		td->td_retval[1] = status;
858 	return (error);
859 }
860 #endif /* COMPAT_43 */
861 
862 /*
863  * The dirty work is handled by kern_wait().
864  */
865 int
866 sys_wait4(struct thread *td, struct wait4_args *uap)
867 {
868 	struct rusage ru, *rup;
869 	int error, status;
870 
871 	if (uap->rusage != NULL)
872 		rup = &ru;
873 	else
874 		rup = NULL;
875 	error = kern_wait(td, uap->pid, &status, uap->options, rup);
876 	if (uap->status != NULL && error == 0 && td->td_retval[0] != 0)
877 		error = copyout(&status, uap->status, sizeof(status));
878 	if (uap->rusage != NULL && error == 0 && td->td_retval[0] != 0)
879 		error = copyout(&ru, uap->rusage, sizeof(struct rusage));
880 	return (error);
881 }
882 
883 int
884 sys_wait6(struct thread *td, struct wait6_args *uap)
885 {
886 	struct __wrusage wru, *wrup;
887 	siginfo_t si, *sip;
888 	idtype_t idtype;
889 	id_t id;
890 	int error, status;
891 
892 	idtype = uap->idtype;
893 	id = uap->id;
894 
895 	if (uap->wrusage != NULL)
896 		wrup = &wru;
897 	else
898 		wrup = NULL;
899 
900 	if (uap->info != NULL) {
901 		sip = &si;
902 		bzero(sip, sizeof(*sip));
903 	} else
904 		sip = NULL;
905 
906 	/*
907 	 *  We expect all callers of wait6() to know about WEXITED and
908 	 *  WTRAPPED.
909 	 */
910 	error = kern_wait6(td, idtype, id, &status, uap->options, wrup, sip);
911 
912 	if (uap->status != NULL && error == 0 && td->td_retval[0] != 0)
913 		error = copyout(&status, uap->status, sizeof(status));
914 	if (uap->wrusage != NULL && error == 0 && td->td_retval[0] != 0)
915 		error = copyout(&wru, uap->wrusage, sizeof(wru));
916 	if (uap->info != NULL && error == 0)
917 		error = copyout(&si, uap->info, sizeof(si));
918 	return (error);
919 }
920 
921 /*
922  * Reap the remains of a zombie process and optionally return status and
923  * rusage.  Asserts and will release both the proctree_lock and the process
924  * lock as part of its work.
925  */
926 void
927 proc_reap(struct thread *td, struct proc *p, int *status, int options)
928 {
929 	struct proc *q, *t;
930 
931 	sx_assert(&proctree_lock, SA_XLOCKED);
932 	PROC_LOCK_ASSERT(p, MA_OWNED);
933 	KASSERT(p->p_state == PRS_ZOMBIE, ("proc_reap: !PRS_ZOMBIE"));
934 
935 	mtx_spin_wait_unlocked(&p->p_slock);
936 
937 	q = td->td_proc;
938 
939 	if (status)
940 		*status = KW_EXITCODE(p->p_xexit, p->p_xsig);
941 	if (options & WNOWAIT) {
942 		/*
943 		 *  Only poll, returning the status.  Caller does not wish to
944 		 * release the proc struct just yet.
945 		 */
946 		PROC_UNLOCK(p);
947 		sx_xunlock(&proctree_lock);
948 		return;
949 	}
950 
951 	PROC_LOCK(q);
952 	sigqueue_take(p->p_ksi);
953 	PROC_UNLOCK(q);
954 
955 	/*
956 	 * If we got the child via a ptrace 'attach', we need to give it back
957 	 * to the old parent.
958 	 */
959 	if (p->p_oppid != p->p_pptr->p_pid) {
960 		PROC_UNLOCK(p);
961 		t = proc_realparent(p);
962 		PROC_LOCK(t);
963 		PROC_LOCK(p);
964 		CTR2(KTR_PTRACE,
965 		    "wait: traced child %d moved back to parent %d", p->p_pid,
966 		    t->p_pid);
967 		proc_reparent(p, t, false);
968 		PROC_UNLOCK(p);
969 		pksignal(t, SIGCHLD, p->p_ksi);
970 		wakeup(t);
971 		cv_broadcast(&p->p_pwait);
972 		PROC_UNLOCK(t);
973 		sx_xunlock(&proctree_lock);
974 		return;
975 	}
976 	PROC_UNLOCK(p);
977 
978 	/*
979 	 * Remove other references to this process to ensure we have an
980 	 * exclusive reference.
981 	 */
982 	sx_xlock(PIDHASHLOCK(p->p_pid));
983 	LIST_REMOVE(p, p_hash);
984 	sx_xunlock(PIDHASHLOCK(p->p_pid));
985 	LIST_REMOVE(p, p_sibling);
986 	reaper_abandon_children(p, true);
987 	reaper_clear(p);
988 	PROC_LOCK(p);
989 	proc_clear_orphan(p);
990 	PROC_UNLOCK(p);
991 	leavepgrp(p);
992 	if (p->p_procdesc != NULL)
993 		procdesc_reap(p);
994 	sx_xunlock(&proctree_lock);
995 
996 	proc_id_clear(PROC_ID_PID, p->p_pid);
997 
998 	PROC_LOCK(p);
999 	knlist_detach(p->p_klist);
1000 	p->p_klist = NULL;
1001 	PROC_UNLOCK(p);
1002 
1003 	/*
1004 	 * Removal from allproc list and process group list paired with
1005 	 * PROC_LOCK which was executed during that time should guarantee
1006 	 * nothing can reach this process anymore. As such further locking
1007 	 * is unnecessary.
1008 	 */
1009 	p->p_xexit = p->p_xsig = 0;		/* XXX: why? */
1010 
1011 	PROC_LOCK(q);
1012 	ruadd(&q->p_stats->p_cru, &q->p_crux, &p->p_ru, &p->p_rux);
1013 	PROC_UNLOCK(q);
1014 
1015 	/*
1016 	 * Decrement the count of procs running with this uid.
1017 	 */
1018 	(void)chgproccnt(p->p_ucred->cr_ruidinfo, -1, 0);
1019 
1020 	/*
1021 	 * Destroy resource accounting information associated with the process.
1022 	 */
1023 #ifdef RACCT
1024 	if (racct_enable) {
1025 		PROC_LOCK(p);
1026 		racct_sub(p, RACCT_NPROC, 1);
1027 		PROC_UNLOCK(p);
1028 	}
1029 #endif
1030 	racct_proc_exit(p);
1031 
1032 	/*
1033 	 * Free credentials, arguments, and sigacts.
1034 	 */
1035 	proc_unset_cred(p);
1036 	pargs_drop(p->p_args);
1037 	p->p_args = NULL;
1038 	sigacts_free(p->p_sigacts);
1039 	p->p_sigacts = NULL;
1040 
1041 	/*
1042 	 * Do any thread-system specific cleanups.
1043 	 */
1044 	thread_wait(p);
1045 
1046 	/*
1047 	 * Give vm and machine-dependent layer a chance to free anything that
1048 	 * cpu_exit couldn't release while still running in process context.
1049 	 */
1050 	vm_waitproc(p);
1051 #ifdef MAC
1052 	mac_proc_destroy(p);
1053 #endif
1054 
1055 	KASSERT(FIRST_THREAD_IN_PROC(p),
1056 	    ("proc_reap: no residual thread!"));
1057 	uma_zfree(proc_zone, p);
1058 	atomic_add_int(&nprocs, -1);
1059 }
1060 
1061 static int
1062 proc_to_reap(struct thread *td, struct proc *p, idtype_t idtype, id_t id,
1063     int *status, int options, struct __wrusage *wrusage, siginfo_t *siginfo,
1064     int check_only)
1065 {
1066 	struct rusage *rup;
1067 
1068 	sx_assert(&proctree_lock, SA_XLOCKED);
1069 
1070 	PROC_LOCK(p);
1071 
1072 	switch (idtype) {
1073 	case P_ALL:
1074 		if (p->p_procdesc == NULL ||
1075 		   (p->p_pptr == td->td_proc &&
1076 		   (p->p_flag & P_TRACED) != 0)) {
1077 			break;
1078 		}
1079 
1080 		PROC_UNLOCK(p);
1081 		return (0);
1082 	case P_PID:
1083 		if (p->p_pid != (pid_t)id) {
1084 			PROC_UNLOCK(p);
1085 			return (0);
1086 		}
1087 		break;
1088 	case P_PGID:
1089 		if (p->p_pgid != (pid_t)id) {
1090 			PROC_UNLOCK(p);
1091 			return (0);
1092 		}
1093 		break;
1094 	case P_SID:
1095 		if (p->p_session->s_sid != (pid_t)id) {
1096 			PROC_UNLOCK(p);
1097 			return (0);
1098 		}
1099 		break;
1100 	case P_UID:
1101 		if (p->p_ucred->cr_uid != (uid_t)id) {
1102 			PROC_UNLOCK(p);
1103 			return (0);
1104 		}
1105 		break;
1106 	case P_GID:
1107 		if (p->p_ucred->cr_gid != (gid_t)id) {
1108 			PROC_UNLOCK(p);
1109 			return (0);
1110 		}
1111 		break;
1112 	case P_JAILID:
1113 		if (p->p_ucred->cr_prison->pr_id != (int)id) {
1114 			PROC_UNLOCK(p);
1115 			return (0);
1116 		}
1117 		break;
1118 	/*
1119 	 * It seems that the thread structures get zeroed out
1120 	 * at process exit.  This makes it impossible to
1121 	 * support P_SETID, P_CID or P_CPUID.
1122 	 */
1123 	default:
1124 		PROC_UNLOCK(p);
1125 		return (0);
1126 	}
1127 
1128 	if (p_canwait(td, p)) {
1129 		PROC_UNLOCK(p);
1130 		return (0);
1131 	}
1132 
1133 	if (((options & WEXITED) == 0) && (p->p_state == PRS_ZOMBIE)) {
1134 		PROC_UNLOCK(p);
1135 		return (0);
1136 	}
1137 
1138 	/*
1139 	 * This special case handles a kthread spawned by linux_clone
1140 	 * (see linux_misc.c).  The linux_wait4 and linux_waitpid
1141 	 * functions need to be able to distinguish between waiting
1142 	 * on a process and waiting on a thread.  It is a thread if
1143 	 * p_sigparent is not SIGCHLD, and the WLINUXCLONE option
1144 	 * signifies we want to wait for threads and not processes.
1145 	 */
1146 	if ((p->p_sigparent != SIGCHLD) ^
1147 	    ((options & WLINUXCLONE) != 0)) {
1148 		PROC_UNLOCK(p);
1149 		return (0);
1150 	}
1151 
1152 	if (siginfo != NULL) {
1153 		bzero(siginfo, sizeof(*siginfo));
1154 		siginfo->si_errno = 0;
1155 
1156 		/*
1157 		 * SUSv4 requires that the si_signo value is always
1158 		 * SIGCHLD. Obey it despite the rfork(2) interface
1159 		 * allows to request other signal for child exit
1160 		 * notification.
1161 		 */
1162 		siginfo->si_signo = SIGCHLD;
1163 
1164 		/*
1165 		 *  This is still a rough estimate.  We will fix the
1166 		 *  cases TRAPPED, STOPPED, and CONTINUED later.
1167 		 */
1168 		if (WCOREDUMP(p->p_xsig)) {
1169 			siginfo->si_code = CLD_DUMPED;
1170 			siginfo->si_status = WTERMSIG(p->p_xsig);
1171 		} else if (WIFSIGNALED(p->p_xsig)) {
1172 			siginfo->si_code = CLD_KILLED;
1173 			siginfo->si_status = WTERMSIG(p->p_xsig);
1174 		} else {
1175 			siginfo->si_code = CLD_EXITED;
1176 			siginfo->si_status = p->p_xexit;
1177 		}
1178 
1179 		siginfo->si_pid = p->p_pid;
1180 		siginfo->si_uid = p->p_ucred->cr_uid;
1181 
1182 		/*
1183 		 * The si_addr field would be useful additional
1184 		 * detail, but apparently the PC value may be lost
1185 		 * when we reach this point.  bzero() above sets
1186 		 * siginfo->si_addr to NULL.
1187 		 */
1188 	}
1189 
1190 	/*
1191 	 * There should be no reason to limit resources usage info to
1192 	 * exited processes only.  A snapshot about any resources used
1193 	 * by a stopped process may be exactly what is needed.
1194 	 */
1195 	if (wrusage != NULL) {
1196 		rup = &wrusage->wru_self;
1197 		*rup = p->p_ru;
1198 		PROC_STATLOCK(p);
1199 		calcru(p, &rup->ru_utime, &rup->ru_stime);
1200 		PROC_STATUNLOCK(p);
1201 
1202 		rup = &wrusage->wru_children;
1203 		*rup = p->p_stats->p_cru;
1204 		calccru(p, &rup->ru_utime, &rup->ru_stime);
1205 	}
1206 
1207 	if (p->p_state == PRS_ZOMBIE && !check_only) {
1208 		proc_reap(td, p, status, options);
1209 		return (-1);
1210 	}
1211 	return (1);
1212 }
1213 
1214 int
1215 kern_wait(struct thread *td, pid_t pid, int *status, int options,
1216     struct rusage *rusage)
1217 {
1218 	struct __wrusage wru, *wrup;
1219 	idtype_t idtype;
1220 	id_t id;
1221 	int ret;
1222 
1223 	/*
1224 	 * Translate the special pid values into the (idtype, pid)
1225 	 * pair for kern_wait6.  The WAIT_MYPGRP case is handled by
1226 	 * kern_wait6() on its own.
1227 	 */
1228 	if (pid == WAIT_ANY) {
1229 		idtype = P_ALL;
1230 		id = 0;
1231 	} else if (pid < 0) {
1232 		idtype = P_PGID;
1233 		id = (id_t)-pid;
1234 	} else {
1235 		idtype = P_PID;
1236 		id = (id_t)pid;
1237 	}
1238 
1239 	if (rusage != NULL)
1240 		wrup = &wru;
1241 	else
1242 		wrup = NULL;
1243 
1244 	/*
1245 	 * For backward compatibility we implicitly add flags WEXITED
1246 	 * and WTRAPPED here.
1247 	 */
1248 	options |= WEXITED | WTRAPPED;
1249 	ret = kern_wait6(td, idtype, id, status, options, wrup, NULL);
1250 	if (rusage != NULL)
1251 		*rusage = wru.wru_self;
1252 	return (ret);
1253 }
1254 
1255 static void
1256 report_alive_proc(struct thread *td, struct proc *p, siginfo_t *siginfo,
1257     int *status, int options, int si_code)
1258 {
1259 	bool cont;
1260 
1261 	PROC_LOCK_ASSERT(p, MA_OWNED);
1262 	sx_assert(&proctree_lock, SA_XLOCKED);
1263 	MPASS(si_code == CLD_TRAPPED || si_code == CLD_STOPPED ||
1264 	    si_code == CLD_CONTINUED);
1265 
1266 	cont = si_code == CLD_CONTINUED;
1267 	if ((options & WNOWAIT) == 0) {
1268 		if (cont)
1269 			p->p_flag &= ~P_CONTINUED;
1270 		else
1271 			p->p_flag |= P_WAITED;
1272 		if (kern_wait_dequeue_sigchld &&
1273 		    (td->td_proc->p_sysent->sv_flags & SV_SIG_WAITNDQ) == 0) {
1274 			PROC_LOCK(td->td_proc);
1275 			sigqueue_take(p->p_ksi);
1276 			PROC_UNLOCK(td->td_proc);
1277 		}
1278 	}
1279 	sx_xunlock(&proctree_lock);
1280 	if (siginfo != NULL) {
1281 		siginfo->si_code = si_code;
1282 		siginfo->si_status = cont ? SIGCONT : p->p_xsig;
1283 	}
1284 	if (status != NULL)
1285 		*status = cont ? SIGCONT : W_STOPCODE(p->p_xsig);
1286 	td->td_retval[0] = p->p_pid;
1287 	PROC_UNLOCK(p);
1288 }
1289 
1290 int
1291 kern_wait6(struct thread *td, idtype_t idtype, id_t id, int *status,
1292     int options, struct __wrusage *wrusage, siginfo_t *siginfo)
1293 {
1294 	struct proc *p, *q;
1295 	pid_t pid;
1296 	int error, nfound, ret;
1297 	bool report;
1298 
1299 	AUDIT_ARG_VALUE((int)idtype);	/* XXX - This is likely wrong! */
1300 	AUDIT_ARG_PID((pid_t)id);	/* XXX - This may be wrong! */
1301 	AUDIT_ARG_VALUE(options);
1302 
1303 	q = td->td_proc;
1304 
1305 	if ((pid_t)id == WAIT_MYPGRP && (idtype == P_PID || idtype == P_PGID)) {
1306 		PROC_LOCK(q);
1307 		id = (id_t)q->p_pgid;
1308 		PROC_UNLOCK(q);
1309 		idtype = P_PGID;
1310 	}
1311 
1312 	/* If we don't know the option, just return. */
1313 	if ((options & ~(WUNTRACED | WNOHANG | WCONTINUED | WNOWAIT |
1314 	    WEXITED | WTRAPPED | WLINUXCLONE)) != 0)
1315 		return (EINVAL);
1316 	if ((options & (WEXITED | WUNTRACED | WCONTINUED | WTRAPPED)) == 0) {
1317 		/*
1318 		 * We will be unable to find any matching processes,
1319 		 * because there are no known events to look for.
1320 		 * Prefer to return error instead of blocking
1321 		 * indefinitely.
1322 		 */
1323 		return (EINVAL);
1324 	}
1325 
1326 loop:
1327 	if (q->p_flag & P_STATCHILD) {
1328 		PROC_LOCK(q);
1329 		q->p_flag &= ~P_STATCHILD;
1330 		PROC_UNLOCK(q);
1331 	}
1332 	sx_xlock(&proctree_lock);
1333 loop_locked:
1334 	nfound = 0;
1335 	LIST_FOREACH(p, &q->p_children, p_sibling) {
1336 		pid = p->p_pid;
1337 		ret = proc_to_reap(td, p, idtype, id, status, options,
1338 		    wrusage, siginfo, 0);
1339 		if (ret == 0)
1340 			continue;
1341 		else if (ret != 1) {
1342 			td->td_retval[0] = pid;
1343 			return (0);
1344 		}
1345 
1346 		nfound++;
1347 		PROC_LOCK_ASSERT(p, MA_OWNED);
1348 
1349 		if ((options & WTRAPPED) != 0 &&
1350 		    (p->p_flag & P_TRACED) != 0) {
1351 			PROC_SLOCK(p);
1352 			report =
1353 			    ((p->p_flag & (P_STOPPED_TRACE | P_STOPPED_SIG)) &&
1354 			    p->p_suspcount == p->p_numthreads &&
1355 			    (p->p_flag & P_WAITED) == 0);
1356 			PROC_SUNLOCK(p);
1357 			if (report) {
1358 			CTR4(KTR_PTRACE,
1359 			    "wait: returning trapped pid %d status %#x "
1360 			    "(xstat %d) xthread %d",
1361 			    p->p_pid, W_STOPCODE(p->p_xsig), p->p_xsig,
1362 			    p->p_xthread != NULL ?
1363 			    p->p_xthread->td_tid : -1);
1364 				report_alive_proc(td, p, siginfo, status,
1365 				    options, CLD_TRAPPED);
1366 				return (0);
1367 			}
1368 		}
1369 		if ((options & WUNTRACED) != 0 &&
1370 		    (p->p_flag & P_STOPPED_SIG) != 0) {
1371 			PROC_SLOCK(p);
1372 			report = (p->p_suspcount == p->p_numthreads &&
1373 			    ((p->p_flag & P_WAITED) == 0));
1374 			PROC_SUNLOCK(p);
1375 			if (report) {
1376 				report_alive_proc(td, p, siginfo, status,
1377 				    options, CLD_STOPPED);
1378 				return (0);
1379 			}
1380 		}
1381 		if ((options & WCONTINUED) != 0 &&
1382 		    (p->p_flag & P_CONTINUED) != 0) {
1383 			report_alive_proc(td, p, siginfo, status, options,
1384 			    CLD_CONTINUED);
1385 			return (0);
1386 		}
1387 		PROC_UNLOCK(p);
1388 	}
1389 
1390 	/*
1391 	 * Look in the orphans list too, to allow the parent to
1392 	 * collect it's child exit status even if child is being
1393 	 * debugged.
1394 	 *
1395 	 * Debugger detaches from the parent upon successful
1396 	 * switch-over from parent to child.  At this point due to
1397 	 * re-parenting the parent loses the child to debugger and a
1398 	 * wait4(2) call would report that it has no children to wait
1399 	 * for.  By maintaining a list of orphans we allow the parent
1400 	 * to successfully wait until the child becomes a zombie.
1401 	 */
1402 	if (nfound == 0) {
1403 		LIST_FOREACH(p, &q->p_orphans, p_orphan) {
1404 			ret = proc_to_reap(td, p, idtype, id, NULL, options,
1405 			    NULL, NULL, 1);
1406 			if (ret != 0) {
1407 				KASSERT(ret != -1, ("reaped an orphan (pid %d)",
1408 				    (int)td->td_retval[0]));
1409 				PROC_UNLOCK(p);
1410 				nfound++;
1411 				break;
1412 			}
1413 		}
1414 	}
1415 	if (nfound == 0) {
1416 		sx_xunlock(&proctree_lock);
1417 		return (ECHILD);
1418 	}
1419 	if (options & WNOHANG) {
1420 		sx_xunlock(&proctree_lock);
1421 		td->td_retval[0] = 0;
1422 		return (0);
1423 	}
1424 	PROC_LOCK(q);
1425 	if (q->p_flag & P_STATCHILD) {
1426 		q->p_flag &= ~P_STATCHILD;
1427 		PROC_UNLOCK(q);
1428 		goto loop_locked;
1429 	}
1430 	sx_xunlock(&proctree_lock);
1431 	error = msleep(q, &q->p_mtx, PWAIT | PCATCH | PDROP, "wait", 0);
1432 	if (error)
1433 		return (error);
1434 	goto loop;
1435 }
1436 
1437 void
1438 proc_add_orphan(struct proc *child, struct proc *parent)
1439 {
1440 
1441 	sx_assert(&proctree_lock, SX_XLOCKED);
1442 	KASSERT((child->p_flag & P_TRACED) != 0,
1443 	    ("proc_add_orphan: not traced"));
1444 
1445 	if (LIST_EMPTY(&parent->p_orphans)) {
1446 		child->p_treeflag |= P_TREE_FIRST_ORPHAN;
1447 		LIST_INSERT_HEAD(&parent->p_orphans, child, p_orphan);
1448 	} else {
1449 		LIST_INSERT_AFTER(LIST_FIRST(&parent->p_orphans),
1450 		    child, p_orphan);
1451 	}
1452 	child->p_treeflag |= P_TREE_ORPHANED;
1453 }
1454 
1455 /*
1456  * Make process 'parent' the new parent of process 'child'.
1457  * Must be called with an exclusive hold of proctree lock.
1458  */
1459 void
1460 proc_reparent(struct proc *child, struct proc *parent, bool set_oppid)
1461 {
1462 
1463 	sx_assert(&proctree_lock, SX_XLOCKED);
1464 	PROC_LOCK_ASSERT(child, MA_OWNED);
1465 	if (child->p_pptr == parent)
1466 		return;
1467 
1468 	PROC_LOCK(child->p_pptr);
1469 	sigqueue_take(child->p_ksi);
1470 	PROC_UNLOCK(child->p_pptr);
1471 	LIST_REMOVE(child, p_sibling);
1472 	LIST_INSERT_HEAD(&parent->p_children, child, p_sibling);
1473 
1474 	proc_clear_orphan(child);
1475 	if ((child->p_flag & P_TRACED) != 0) {
1476 		proc_add_orphan(child, child->p_pptr);
1477 	}
1478 
1479 	child->p_pptr = parent;
1480 	if (set_oppid)
1481 		child->p_oppid = parent->p_pid;
1482 }
1483