xref: /freebsd/sys/kern/kern_exit.c (revision 884a2a699669ec61e2366e3e358342dbc94be24a)
1 /*-
2  * Copyright (c) 1982, 1986, 1989, 1991, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  * (c) UNIX System Laboratories, Inc.
5  * All or some portions of this file are derived from material licensed
6  * to the University of California by American Telephone and Telegraph
7  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8  * the permission of UNIX System Laboratories, Inc.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 4. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  *	@(#)kern_exit.c	8.7 (Berkeley) 2/12/94
35  */
36 
37 #include <sys/cdefs.h>
38 __FBSDID("$FreeBSD$");
39 
40 #include "opt_compat.h"
41 #include "opt_kdtrace.h"
42 #include "opt_ktrace.h"
43 
44 #include <sys/param.h>
45 #include <sys/systm.h>
46 #include <sys/sysproto.h>
47 #include <sys/eventhandler.h>
48 #include <sys/kernel.h>
49 #include <sys/malloc.h>
50 #include <sys/lock.h>
51 #include <sys/mutex.h>
52 #include <sys/proc.h>
53 #include <sys/pioctl.h>
54 #include <sys/jail.h>
55 #include <sys/tty.h>
56 #include <sys/wait.h>
57 #include <sys/vmmeter.h>
58 #include <sys/vnode.h>
59 #include <sys/racct.h>
60 #include <sys/resourcevar.h>
61 #include <sys/sbuf.h>
62 #include <sys/signalvar.h>
63 #include <sys/sched.h>
64 #include <sys/sx.h>
65 #include <sys/syscallsubr.h>
66 #include <sys/syslog.h>
67 #include <sys/ptrace.h>
68 #include <sys/acct.h>		/* for acct_process() function prototype */
69 #include <sys/filedesc.h>
70 #include <sys/sdt.h>
71 #include <sys/shm.h>
72 #include <sys/sem.h>
73 #ifdef KTRACE
74 #include <sys/ktrace.h>
75 #endif
76 
77 #include <security/audit/audit.h>
78 #include <security/mac/mac_framework.h>
79 
80 #include <vm/vm.h>
81 #include <vm/vm_extern.h>
82 #include <vm/vm_param.h>
83 #include <vm/pmap.h>
84 #include <vm/vm_map.h>
85 #include <vm/vm_page.h>
86 #include <vm/uma.h>
87 
88 #ifdef KDTRACE_HOOKS
89 #include <sys/dtrace_bsd.h>
90 dtrace_execexit_func_t	dtrace_fasttrap_exit;
91 #endif
92 
93 SDT_PROVIDER_DECLARE(proc);
94 SDT_PROBE_DEFINE(proc, kernel, , exit, exit);
95 SDT_PROBE_ARGTYPE(proc, kernel, , exit, 0, "int");
96 
97 /* Hook for NFS teardown procedure. */
98 void (*nlminfo_release_p)(struct proc *p);
99 
100 /*
101  * exit -- death of process.
102  */
103 void
104 sys_exit(struct thread *td, struct sys_exit_args *uap)
105 {
106 
107 	exit1(td, W_EXITCODE(uap->rval, 0));
108 	/* NOTREACHED */
109 }
110 
111 /*
112  * Exit: deallocate address space and other resources, change proc state to
113  * zombie, and unlink proc from allproc and parent's lists.  Save exit status
114  * and rusage for wait().  Check for child processes and orphan them.
115  */
116 void
117 exit1(struct thread *td, int rv)
118 {
119 	struct proc *p, *nq, *q;
120 	struct vnode *vtmp;
121 	struct vnode *ttyvp = NULL;
122 	struct plimit *plim;
123 	int locked;
124 
125 	mtx_assert(&Giant, MA_NOTOWNED);
126 
127 	p = td->td_proc;
128 	/*
129 	 * XXX in case we're rebooting we just let init die in order to
130 	 * work around an unsolved stack overflow seen very late during
131 	 * shutdown on sparc64 when the gmirror worker process exists.
132 	 */
133 	if (p == initproc && rebooting == 0) {
134 		printf("init died (signal %d, exit %d)\n",
135 		    WTERMSIG(rv), WEXITSTATUS(rv));
136 		panic("Going nowhere without my init!");
137 	}
138 
139 	/*
140 	 * MUST abort all other threads before proceeding past here.
141 	 */
142 	PROC_LOCK(p);
143 	while (p->p_flag & P_HADTHREADS) {
144 		/*
145 		 * First check if some other thread got here before us..
146 		 * if so, act apropriatly, (exit or suspend);
147 		 */
148 		thread_suspend_check(0);
149 
150 		/*
151 		 * Kill off the other threads. This requires
152 		 * some co-operation from other parts of the kernel
153 		 * so it may not be instantaneous.  With this state set
154 		 * any thread entering the kernel from userspace will
155 		 * thread_exit() in trap().  Any thread attempting to
156 		 * sleep will return immediately with EINTR or EWOULDBLOCK
157 		 * which will hopefully force them to back out to userland
158 		 * freeing resources as they go.  Any thread attempting
159 		 * to return to userland will thread_exit() from userret().
160 		 * thread_exit() will unsuspend us when the last of the
161 		 * other threads exits.
162 		 * If there is already a thread singler after resumption,
163 		 * calling thread_single will fail; in that case, we just
164 		 * re-check all suspension request, the thread should
165 		 * either be suspended there or exit.
166 		 */
167 		if (! thread_single(SINGLE_EXIT))
168 			break;
169 
170 		/*
171 		 * All other activity in this process is now stopped.
172 		 * Threading support has been turned off.
173 		 */
174 	}
175 	KASSERT(p->p_numthreads == 1,
176 	    ("exit1: proc %p exiting with %d threads", p, p->p_numthreads));
177 	racct_sub(p, RACCT_NTHR, 1);
178 	/*
179 	 * Wakeup anyone in procfs' PIOCWAIT.  They should have a hold
180 	 * on our vmspace, so we should block below until they have
181 	 * released their reference to us.  Note that if they have
182 	 * requested S_EXIT stops we will block here until they ack
183 	 * via PIOCCONT.
184 	 */
185 	_STOPEVENT(p, S_EXIT, rv);
186 
187 	/*
188 	 * Note that we are exiting and do another wakeup of anyone in
189 	 * PIOCWAIT in case they aren't listening for S_EXIT stops or
190 	 * decided to wait again after we told them we are exiting.
191 	 */
192 	p->p_flag |= P_WEXIT;
193 	wakeup(&p->p_stype);
194 
195 	/*
196 	 * Wait for any processes that have a hold on our vmspace to
197 	 * release their reference.
198 	 */
199 	while (p->p_lock > 0)
200 		msleep(&p->p_lock, &p->p_mtx, PWAIT, "exithold", 0);
201 
202 	p->p_xstat = rv;	/* Let event handler change exit status */
203 	PROC_UNLOCK(p);
204 	/* Drain the limit callout while we don't have the proc locked */
205 	callout_drain(&p->p_limco);
206 
207 #ifdef AUDIT
208 	/*
209 	 * The Sun BSM exit token contains two components: an exit status as
210 	 * passed to exit(), and a return value to indicate what sort of exit
211 	 * it was.  The exit status is WEXITSTATUS(rv), but it's not clear
212 	 * what the return value is.
213 	 */
214 	AUDIT_ARG_EXIT(WEXITSTATUS(rv), 0);
215 	AUDIT_SYSCALL_EXIT(0, td);
216 #endif
217 
218 	/* Are we a task leader? */
219 	if (p == p->p_leader) {
220 		mtx_lock(&ppeers_lock);
221 		q = p->p_peers;
222 		while (q != NULL) {
223 			PROC_LOCK(q);
224 			psignal(q, SIGKILL);
225 			PROC_UNLOCK(q);
226 			q = q->p_peers;
227 		}
228 		while (p->p_peers != NULL)
229 			msleep(p, &ppeers_lock, PWAIT, "exit1", 0);
230 		mtx_unlock(&ppeers_lock);
231 	}
232 
233 	/*
234 	 * Check if any loadable modules need anything done at process exit.
235 	 * E.g. SYSV IPC stuff
236 	 * XXX what if one of these generates an error?
237 	 */
238 	EVENTHANDLER_INVOKE(process_exit, p);
239 
240 	/*
241 	 * If parent is waiting for us to exit or exec,
242 	 * P_PPWAIT is set; we will wakeup the parent below.
243 	 */
244 	PROC_LOCK(p);
245 	rv = p->p_xstat;	/* Event handler could change exit status */
246 	stopprofclock(p);
247 	p->p_flag &= ~(P_TRACED | P_PPWAIT);
248 
249 	/*
250 	 * Stop the real interval timer.  If the handler is currently
251 	 * executing, prevent it from rearming itself and let it finish.
252 	 */
253 	if (timevalisset(&p->p_realtimer.it_value) &&
254 	    callout_stop(&p->p_itcallout) == 0) {
255 		timevalclear(&p->p_realtimer.it_interval);
256 		msleep(&p->p_itcallout, &p->p_mtx, PWAIT, "ritwait", 0);
257 		KASSERT(!timevalisset(&p->p_realtimer.it_value),
258 		    ("realtime timer is still armed"));
259 	}
260 	PROC_UNLOCK(p);
261 
262 	/*
263 	 * Reset any sigio structures pointing to us as a result of
264 	 * F_SETOWN with our pid.
265 	 */
266 	funsetownlst(&p->p_sigiolst);
267 
268 	/*
269 	 * If this process has an nlminfo data area (for lockd), release it
270 	 */
271 	if (nlminfo_release_p != NULL && p->p_nlminfo != NULL)
272 		(*nlminfo_release_p)(p);
273 
274 	/*
275 	 * Close open files and release open-file table.
276 	 * This may block!
277 	 */
278 	fdfree(td);
279 
280 	/*
281 	 * If this thread tickled GEOM, we need to wait for the giggling to
282 	 * stop before we return to userland
283 	 */
284 	if (td->td_pflags & TDP_GEOM)
285 		g_waitidle();
286 
287 	/*
288 	 * Remove ourself from our leader's peer list and wake our leader.
289 	 */
290 	mtx_lock(&ppeers_lock);
291 	if (p->p_leader->p_peers) {
292 		q = p->p_leader;
293 		while (q->p_peers != p)
294 			q = q->p_peers;
295 		q->p_peers = p->p_peers;
296 		wakeup(p->p_leader);
297 	}
298 	mtx_unlock(&ppeers_lock);
299 
300 	vmspace_exit(td);
301 
302 	sx_xlock(&proctree_lock);
303 	if (SESS_LEADER(p)) {
304 		struct session *sp = p->p_session;
305 		struct tty *tp;
306 
307 		/*
308 		 * s_ttyp is not zero'd; we use this to indicate that
309 		 * the session once had a controlling terminal. (for
310 		 * logging and informational purposes)
311 		 */
312 		SESS_LOCK(sp);
313 		ttyvp = sp->s_ttyvp;
314 		tp = sp->s_ttyp;
315 		sp->s_ttyvp = NULL;
316 		sp->s_ttydp = NULL;
317 		sp->s_leader = NULL;
318 		SESS_UNLOCK(sp);
319 
320 		/*
321 		 * Signal foreground pgrp and revoke access to
322 		 * controlling terminal if it has not been revoked
323 		 * already.
324 		 *
325 		 * Because the TTY may have been revoked in the mean
326 		 * time and could already have a new session associated
327 		 * with it, make sure we don't send a SIGHUP to a
328 		 * foreground process group that does not belong to this
329 		 * session.
330 		 */
331 
332 		if (tp != NULL) {
333 			tty_lock(tp);
334 			if (tp->t_session == sp)
335 				tty_signal_pgrp(tp, SIGHUP);
336 			tty_unlock(tp);
337 		}
338 
339 		if (ttyvp != NULL) {
340 			sx_xunlock(&proctree_lock);
341 			if (vn_lock(ttyvp, LK_EXCLUSIVE) == 0) {
342 				VOP_REVOKE(ttyvp, REVOKEALL);
343 				VOP_UNLOCK(ttyvp, 0);
344 			}
345 			sx_xlock(&proctree_lock);
346 		}
347 	}
348 	fixjobc(p, p->p_pgrp, 0);
349 	sx_xunlock(&proctree_lock);
350 	(void)acct_process(td);
351 
352 	/* Release the TTY now we've unlocked everything. */
353 	if (ttyvp != NULL)
354 		vrele(ttyvp);
355 #ifdef KTRACE
356 	ktrprocexit(td);
357 #endif
358 	/*
359 	 * Release reference to text vnode
360 	 */
361 	if ((vtmp = p->p_textvp) != NULL) {
362 		p->p_textvp = NULL;
363 		locked = VFS_LOCK_GIANT(vtmp->v_mount);
364 		vrele(vtmp);
365 		VFS_UNLOCK_GIANT(locked);
366 	}
367 
368 	/*
369 	 * Release our limits structure.
370 	 */
371 	PROC_LOCK(p);
372 	plim = p->p_limit;
373 	p->p_limit = NULL;
374 	PROC_UNLOCK(p);
375 	lim_free(plim);
376 
377 	tidhash_remove(td);
378 
379 	/*
380 	 * Remove proc from allproc queue and pidhash chain.
381 	 * Place onto zombproc.  Unlink from parent's child list.
382 	 */
383 	sx_xlock(&allproc_lock);
384 	LIST_REMOVE(p, p_list);
385 	LIST_INSERT_HEAD(&zombproc, p, p_list);
386 	LIST_REMOVE(p, p_hash);
387 	sx_xunlock(&allproc_lock);
388 
389 	/*
390 	 * Call machine-dependent code to release any
391 	 * machine-dependent resources other than the address space.
392 	 * The address space is released by "vmspace_exitfree(p)" in
393 	 * vm_waitproc().
394 	 */
395 	cpu_exit(td);
396 
397 	WITNESS_WARN(WARN_PANIC, NULL, "process (pid %d) exiting", p->p_pid);
398 
399 	/*
400 	 * Reparent all of our children to init.
401 	 */
402 	sx_xlock(&proctree_lock);
403 	q = LIST_FIRST(&p->p_children);
404 	if (q != NULL)		/* only need this if any child is S_ZOMB */
405 		wakeup(initproc);
406 	for (; q != NULL; q = nq) {
407 		nq = LIST_NEXT(q, p_sibling);
408 		PROC_LOCK(q);
409 		proc_reparent(q, initproc);
410 		q->p_sigparent = SIGCHLD;
411 		/*
412 		 * Traced processes are killed
413 		 * since their existence means someone is screwing up.
414 		 */
415 		if (q->p_flag & P_TRACED) {
416 			struct thread *temp;
417 
418 			q->p_flag &= ~(P_TRACED | P_STOPPED_TRACE);
419 			FOREACH_THREAD_IN_PROC(q, temp)
420 				temp->td_dbgflags &= ~TDB_SUSPEND;
421 			psignal(q, SIGKILL);
422 		}
423 		PROC_UNLOCK(q);
424 	}
425 
426 	/* Save exit status. */
427 	PROC_LOCK(p);
428 	p->p_xthread = td;
429 
430 	/* Tell the prison that we are gone. */
431 	prison_proc_free(p->p_ucred->cr_prison);
432 
433 #ifdef KDTRACE_HOOKS
434 	/*
435 	 * Tell the DTrace fasttrap provider about the exit if it
436 	 * has declared an interest.
437 	 */
438 	if (dtrace_fasttrap_exit)
439 		dtrace_fasttrap_exit(p);
440 #endif
441 
442 	/*
443 	 * Notify interested parties of our demise.
444 	 */
445 	KNOTE_LOCKED(&p->p_klist, NOTE_EXIT);
446 
447 #ifdef KDTRACE_HOOKS
448 	int reason = CLD_EXITED;
449 	if (WCOREDUMP(rv))
450 		reason = CLD_DUMPED;
451 	else if (WIFSIGNALED(rv))
452 		reason = CLD_KILLED;
453 	SDT_PROBE(proc, kernel, , exit, reason, 0, 0, 0, 0);
454 #endif
455 
456 	/*
457 	 * Just delete all entries in the p_klist. At this point we won't
458 	 * report any more events, and there are nasty race conditions that
459 	 * can beat us if we don't.
460 	 */
461 	knlist_clear(&p->p_klist, 1);
462 
463 	/*
464 	 * Notify parent that we're gone.  If parent has the PS_NOCLDWAIT
465 	 * flag set, or if the handler is set to SIG_IGN, notify process
466 	 * 1 instead (and hope it will handle this situation).
467 	 */
468 	PROC_LOCK(p->p_pptr);
469 	mtx_lock(&p->p_pptr->p_sigacts->ps_mtx);
470 	if (p->p_pptr->p_sigacts->ps_flag & (PS_NOCLDWAIT | PS_CLDSIGIGN)) {
471 		struct proc *pp;
472 
473 		mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx);
474 		pp = p->p_pptr;
475 		PROC_UNLOCK(pp);
476 		proc_reparent(p, initproc);
477 		p->p_sigparent = SIGCHLD;
478 		PROC_LOCK(p->p_pptr);
479 
480 		/*
481 		 * Notify parent, so in case he was wait(2)ing or
482 		 * executing waitpid(2) with our pid, he will
483 		 * continue.
484 		 */
485 		wakeup(pp);
486 	} else
487 		mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx);
488 
489 	if (p->p_pptr == initproc)
490 		psignal(p->p_pptr, SIGCHLD);
491 	else if (p->p_sigparent != 0) {
492 		if (p->p_sigparent == SIGCHLD)
493 			childproc_exited(p);
494 		else	/* LINUX thread */
495 			psignal(p->p_pptr, p->p_sigparent);
496 	}
497 	sx_xunlock(&proctree_lock);
498 
499 	/*
500 	 * The state PRS_ZOMBIE prevents other proesses from sending
501 	 * signal to the process, to avoid memory leak, we free memory
502 	 * for signal queue at the time when the state is set.
503 	 */
504 	sigqueue_flush(&p->p_sigqueue);
505 	sigqueue_flush(&td->td_sigqueue);
506 
507 	/*
508 	 * We have to wait until after acquiring all locks before
509 	 * changing p_state.  We need to avoid all possible context
510 	 * switches (including ones from blocking on a mutex) while
511 	 * marked as a zombie.  We also have to set the zombie state
512 	 * before we release the parent process' proc lock to avoid
513 	 * a lost wakeup.  So, we first call wakeup, then we grab the
514 	 * sched lock, update the state, and release the parent process'
515 	 * proc lock.
516 	 */
517 	wakeup(p->p_pptr);
518 	cv_broadcast(&p->p_pwait);
519 	sched_exit(p->p_pptr, td);
520 	PROC_SLOCK(p);
521 	p->p_state = PRS_ZOMBIE;
522 	PROC_UNLOCK(p->p_pptr);
523 
524 	/*
525 	 * Hopefully no one will try to deliver a signal to the process this
526 	 * late in the game.
527 	 */
528 	knlist_destroy(&p->p_klist);
529 
530 	/*
531 	 * Save our children's rusage information in our exit rusage.
532 	 */
533 	ruadd(&p->p_ru, &p->p_rux, &p->p_stats->p_cru, &p->p_crux);
534 
535 	/*
536 	 * Make sure the scheduler takes this thread out of its tables etc.
537 	 * This will also release this thread's reference to the ucred.
538 	 * Other thread parts to release include pcb bits and such.
539 	 */
540 	thread_exit();
541 }
542 
543 
544 #ifndef _SYS_SYSPROTO_H_
545 struct abort2_args {
546 	char *why;
547 	int nargs;
548 	void **args;
549 };
550 #endif
551 
552 int
553 abort2(struct thread *td, struct abort2_args *uap)
554 {
555 	struct proc *p = td->td_proc;
556 	struct sbuf *sb;
557 	void *uargs[16];
558 	int error, i, sig;
559 
560 	/*
561 	 * Do it right now so we can log either proper call of abort2(), or
562 	 * note, that invalid argument was passed. 512 is big enough to
563 	 * handle 16 arguments' descriptions with additional comments.
564 	 */
565 	sb = sbuf_new(NULL, NULL, 512, SBUF_FIXEDLEN);
566 	sbuf_clear(sb);
567 	sbuf_printf(sb, "%s(pid %d uid %d) aborted: ",
568 	    p->p_comm, p->p_pid, td->td_ucred->cr_uid);
569 	/*
570 	 * Since we can't return from abort2(), send SIGKILL in cases, where
571 	 * abort2() was called improperly
572 	 */
573 	sig = SIGKILL;
574 	/* Prevent from DoSes from user-space. */
575 	if (uap->nargs < 0 || uap->nargs > 16)
576 		goto out;
577 	if (uap->nargs > 0) {
578 		if (uap->args == NULL)
579 			goto out;
580 		error = copyin(uap->args, uargs, uap->nargs * sizeof(void *));
581 		if (error != 0)
582 			goto out;
583 	}
584 	/*
585 	 * Limit size of 'reason' string to 128. Will fit even when
586 	 * maximal number of arguments was chosen to be logged.
587 	 */
588 	if (uap->why != NULL) {
589 		error = sbuf_copyin(sb, uap->why, 128);
590 		if (error < 0)
591 			goto out;
592 	} else {
593 		sbuf_printf(sb, "(null)");
594 	}
595 	if (uap->nargs > 0) {
596 		sbuf_printf(sb, "(");
597 		for (i = 0;i < uap->nargs; i++)
598 			sbuf_printf(sb, "%s%p", i == 0 ? "" : ", ", uargs[i]);
599 		sbuf_printf(sb, ")");
600 	}
601 	/*
602 	 * Final stage: arguments were proper, string has been
603 	 * successfully copied from userspace, and copying pointers
604 	 * from user-space succeed.
605 	 */
606 	sig = SIGABRT;
607 out:
608 	if (sig == SIGKILL) {
609 		sbuf_trim(sb);
610 		sbuf_printf(sb, " (Reason text inaccessible)");
611 	}
612 	sbuf_cat(sb, "\n");
613 	sbuf_finish(sb);
614 	log(LOG_INFO, "%s", sbuf_data(sb));
615 	sbuf_delete(sb);
616 	exit1(td, W_EXITCODE(0, sig));
617 	return (0);
618 }
619 
620 
621 #ifdef COMPAT_43
622 /*
623  * The dirty work is handled by kern_wait().
624  */
625 int
626 owait(struct thread *td, struct owait_args *uap __unused)
627 {
628 	int error, status;
629 
630 	error = kern_wait(td, WAIT_ANY, &status, 0, NULL);
631 	if (error == 0)
632 		td->td_retval[1] = status;
633 	return (error);
634 }
635 #endif /* COMPAT_43 */
636 
637 /*
638  * The dirty work is handled by kern_wait().
639  */
640 int
641 wait4(struct thread *td, struct wait_args *uap)
642 {
643 	struct rusage ru, *rup;
644 	int error, status;
645 
646 	if (uap->rusage != NULL)
647 		rup = &ru;
648 	else
649 		rup = NULL;
650 	error = kern_wait(td, uap->pid, &status, uap->options, rup);
651 	if (uap->status != NULL && error == 0)
652 		error = copyout(&status, uap->status, sizeof(status));
653 	if (uap->rusage != NULL && error == 0)
654 		error = copyout(&ru, uap->rusage, sizeof(struct rusage));
655 	return (error);
656 }
657 
658 /*
659  * Reap the remains of a zombie process and optionally return status and
660  * rusage.  Asserts and will release both the proctree_lock and the process
661  * lock as part of its work.
662  */
663 static void
664 proc_reap(struct thread *td, struct proc *p, int *status, int options,
665     struct rusage *rusage)
666 {
667 	struct proc *q, *t;
668 
669 	sx_assert(&proctree_lock, SA_XLOCKED);
670 	PROC_LOCK_ASSERT(p, MA_OWNED);
671 	PROC_SLOCK_ASSERT(p, MA_OWNED);
672 	KASSERT(p->p_state == PRS_ZOMBIE, ("proc_reap: !PRS_ZOMBIE"));
673 
674 	q = td->td_proc;
675 	if (rusage) {
676 		*rusage = p->p_ru;
677 		calcru(p, &rusage->ru_utime, &rusage->ru_stime);
678 	}
679 	PROC_SUNLOCK(p);
680 	td->td_retval[0] = p->p_pid;
681 	if (status)
682 		*status = p->p_xstat;	/* convert to int */
683 	if (options & WNOWAIT) {
684 		/*
685 		 *  Only poll, returning the status.  Caller does not wish to
686 		 * release the proc struct just yet.
687 		 */
688 		PROC_UNLOCK(p);
689 		sx_xunlock(&proctree_lock);
690 		return;
691 	}
692 
693 	PROC_LOCK(q);
694 	sigqueue_take(p->p_ksi);
695 	PROC_UNLOCK(q);
696 	PROC_UNLOCK(p);
697 
698 	/*
699 	 * If we got the child via a ptrace 'attach', we need to give it back
700 	 * to the old parent.
701 	 */
702 	if (p->p_oppid && (t = pfind(p->p_oppid)) != NULL) {
703 		PROC_LOCK(p);
704 		p->p_oppid = 0;
705 		proc_reparent(p, t);
706 		PROC_UNLOCK(p);
707 		pksignal(t, SIGCHLD, p->p_ksi);
708 		wakeup(t);
709 		cv_broadcast(&p->p_pwait);
710 		PROC_UNLOCK(t);
711 		sx_xunlock(&proctree_lock);
712 		return;
713 	}
714 
715 	/*
716 	 * Remove other references to this process to ensure we have an
717 	 * exclusive reference.
718 	 */
719 	sx_xlock(&allproc_lock);
720 	LIST_REMOVE(p, p_list);	/* off zombproc */
721 	sx_xunlock(&allproc_lock);
722 	LIST_REMOVE(p, p_sibling);
723 	leavepgrp(p);
724 	sx_xunlock(&proctree_lock);
725 
726 	/*
727 	 * As a side effect of this lock, we know that all other writes to
728 	 * this proc are visible now, so no more locking is needed for p.
729 	 */
730 	PROC_LOCK(p);
731 	p->p_xstat = 0;		/* XXX: why? */
732 	PROC_UNLOCK(p);
733 	PROC_LOCK(q);
734 	ruadd(&q->p_stats->p_cru, &q->p_crux, &p->p_ru, &p->p_rux);
735 	PROC_UNLOCK(q);
736 
737 	/*
738 	 * Decrement the count of procs running with this uid.
739 	 */
740 	(void)chgproccnt(p->p_ucred->cr_ruidinfo, -1, 0);
741 
742 	/*
743 	 * Destroy resource accounting information associated with the process.
744 	 */
745 	racct_proc_exit(p);
746 	PROC_LOCK(p->p_pptr);
747 	racct_sub(p->p_pptr, RACCT_NPROC, 1);
748 	PROC_UNLOCK(p->p_pptr);
749 
750 	/*
751 	 * Free credentials, arguments, and sigacts.
752 	 */
753 	crfree(p->p_ucred);
754 	p->p_ucred = NULL;
755 	pargs_drop(p->p_args);
756 	p->p_args = NULL;
757 	sigacts_free(p->p_sigacts);
758 	p->p_sigacts = NULL;
759 
760 	/*
761 	 * Do any thread-system specific cleanups.
762 	 */
763 	thread_wait(p);
764 
765 	/*
766 	 * Give vm and machine-dependent layer a chance to free anything that
767 	 * cpu_exit couldn't release while still running in process context.
768 	 */
769 	vm_waitproc(p);
770 #ifdef MAC
771 	mac_proc_destroy(p);
772 #endif
773 	KASSERT(FIRST_THREAD_IN_PROC(p),
774 	    ("proc_reap: no residual thread!"));
775 	uma_zfree(proc_zone, p);
776 	sx_xlock(&allproc_lock);
777 	nprocs--;
778 	sx_xunlock(&allproc_lock);
779 }
780 
781 int
782 kern_wait(struct thread *td, pid_t pid, int *status, int options,
783     struct rusage *rusage)
784 {
785 	struct proc *p, *q;
786 	int error, nfound;
787 
788 	AUDIT_ARG_PID(pid);
789 	AUDIT_ARG_VALUE(options);
790 
791 	q = td->td_proc;
792 	if (pid == 0) {
793 		PROC_LOCK(q);
794 		pid = -q->p_pgid;
795 		PROC_UNLOCK(q);
796 	}
797 	if (options &~ (WUNTRACED|WNOHANG|WCONTINUED|WNOWAIT|WLINUXCLONE))
798 		return (EINVAL);
799 loop:
800 	if (q->p_flag & P_STATCHILD) {
801 		PROC_LOCK(q);
802 		q->p_flag &= ~P_STATCHILD;
803 		PROC_UNLOCK(q);
804 	}
805 	nfound = 0;
806 	sx_xlock(&proctree_lock);
807 	LIST_FOREACH(p, &q->p_children, p_sibling) {
808 		PROC_LOCK(p);
809 		if (pid != WAIT_ANY &&
810 		    p->p_pid != pid && p->p_pgid != -pid) {
811 			PROC_UNLOCK(p);
812 			continue;
813 		}
814 		if (p_canwait(td, p)) {
815 			PROC_UNLOCK(p);
816 			continue;
817 		}
818 
819 		/*
820 		 * This special case handles a kthread spawned by linux_clone
821 		 * (see linux_misc.c).  The linux_wait4 and linux_waitpid
822 		 * functions need to be able to distinguish between waiting
823 		 * on a process and waiting on a thread.  It is a thread if
824 		 * p_sigparent is not SIGCHLD, and the WLINUXCLONE option
825 		 * signifies we want to wait for threads and not processes.
826 		 */
827 		if ((p->p_sigparent != SIGCHLD) ^
828 		    ((options & WLINUXCLONE) != 0)) {
829 			PROC_UNLOCK(p);
830 			continue;
831 		}
832 
833 		nfound++;
834 		PROC_SLOCK(p);
835 		if (p->p_state == PRS_ZOMBIE) {
836 			proc_reap(td, p, status, options, rusage);
837 			return (0);
838 		}
839 		if ((p->p_flag & P_STOPPED_SIG) &&
840 		    (p->p_suspcount == p->p_numthreads) &&
841 		    (p->p_flag & P_WAITED) == 0 &&
842 		    (p->p_flag & P_TRACED || options & WUNTRACED)) {
843 			PROC_SUNLOCK(p);
844 			p->p_flag |= P_WAITED;
845 			sx_xunlock(&proctree_lock);
846 			td->td_retval[0] = p->p_pid;
847 			if (status)
848 				*status = W_STOPCODE(p->p_xstat);
849 
850 			PROC_LOCK(q);
851 			sigqueue_take(p->p_ksi);
852 			PROC_UNLOCK(q);
853 			PROC_UNLOCK(p);
854 
855 			return (0);
856 		}
857 		PROC_SUNLOCK(p);
858 		if (options & WCONTINUED && (p->p_flag & P_CONTINUED)) {
859 			sx_xunlock(&proctree_lock);
860 			td->td_retval[0] = p->p_pid;
861 			p->p_flag &= ~P_CONTINUED;
862 
863 			PROC_LOCK(q);
864 			sigqueue_take(p->p_ksi);
865 			PROC_UNLOCK(q);
866 			PROC_UNLOCK(p);
867 
868 			if (status)
869 				*status = SIGCONT;
870 			return (0);
871 		}
872 		PROC_UNLOCK(p);
873 	}
874 	if (nfound == 0) {
875 		sx_xunlock(&proctree_lock);
876 		return (ECHILD);
877 	}
878 	if (options & WNOHANG) {
879 		sx_xunlock(&proctree_lock);
880 		td->td_retval[0] = 0;
881 		return (0);
882 	}
883 	PROC_LOCK(q);
884 	sx_xunlock(&proctree_lock);
885 	if (q->p_flag & P_STATCHILD) {
886 		q->p_flag &= ~P_STATCHILD;
887 		error = 0;
888 	} else
889 		error = msleep(q, &q->p_mtx, PWAIT | PCATCH, "wait", 0);
890 	PROC_UNLOCK(q);
891 	if (error)
892 		return (error);
893 	goto loop;
894 }
895 
896 /*
897  * Make process 'parent' the new parent of process 'child'.
898  * Must be called with an exclusive hold of proctree lock.
899  */
900 void
901 proc_reparent(struct proc *child, struct proc *parent)
902 {
903 	int locked;
904 
905 	sx_assert(&proctree_lock, SX_XLOCKED);
906 	PROC_LOCK_ASSERT(child, MA_OWNED);
907 	if (child->p_pptr == parent)
908 		return;
909 
910 	locked = PROC_LOCKED(parent);
911 	if (!locked)
912 		PROC_LOCK(parent);
913 	racct_add_force(parent, RACCT_NPROC, 1);
914 	if (!locked)
915 		PROC_UNLOCK(parent);
916 	PROC_LOCK(child->p_pptr);
917 	racct_sub(child->p_pptr, RACCT_NPROC, 1);
918 	sigqueue_take(child->p_ksi);
919 	PROC_UNLOCK(child->p_pptr);
920 	LIST_REMOVE(child, p_sibling);
921 	LIST_INSERT_HEAD(&parent->p_children, child, p_sibling);
922 	child->p_pptr = parent;
923 }
924