xref: /freebsd/sys/kern/kern_exit.c (revision 2b743a9e9ddc6736208dc8ca1ce06ce64ad20a19)
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_ktrace.h"
42 #include "opt_mac.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/tty.h>
55 #include <sys/wait.h>
56 #include <sys/vmmeter.h>
57 #include <sys/vnode.h>
58 #include <sys/resourcevar.h>
59 #include <sys/sbuf.h>
60 #include <sys/signalvar.h>
61 #include <sys/sched.h>
62 #include <sys/sx.h>
63 #include <sys/syscallsubr.h>
64 #include <sys/syslog.h>
65 #include <sys/ptrace.h>
66 #include <sys/acct.h>		/* for acct_process() function prototype */
67 #include <sys/filedesc.h>
68 #include <sys/shm.h>
69 #include <sys/sem.h>
70 #ifdef KTRACE
71 #include <sys/ktrace.h>
72 #endif
73 
74 #include <security/audit/audit.h>
75 #include <security/mac/mac_framework.h>
76 
77 #include <vm/vm.h>
78 #include <vm/vm_extern.h>
79 #include <vm/vm_param.h>
80 #include <vm/pmap.h>
81 #include <vm/vm_map.h>
82 #include <vm/vm_page.h>
83 #include <vm/uma.h>
84 
85 /* Required to be non-static for SysVR4 emulator */
86 MALLOC_DEFINE(M_ZOMBIE, "zombie", "zombie proc status");
87 
88 /* Hook for NFS teardown procedure. */
89 void (*nlminfo_release_p)(struct proc *p);
90 
91 /*
92  * exit --
93  *	Death of process.
94  *
95  * MPSAFE
96  */
97 void
98 sys_exit(struct thread *td, struct sys_exit_args *uap)
99 {
100 
101 	exit1(td, W_EXITCODE(uap->rval, 0));
102 	/* NOTREACHED */
103 }
104 
105 /*
106  * Exit: deallocate address space and other resources, change proc state
107  * to zombie, and unlink proc from allproc and parent's lists.  Save exit
108  * status and rusage for wait().  Check for child processes and orphan them.
109  */
110 void
111 exit1(struct thread *td, int rv)
112 {
113 	struct proc *p, *nq, *q;
114 	struct tty *tp;
115 	struct vnode *ttyvp;
116 	struct vnode *vtmp;
117 #ifdef KTRACE
118 	struct vnode *tracevp;
119 	struct ucred *tracecred;
120 #endif
121 	struct plimit *plim;
122 	int locked;
123 
124 	/*
125 	 * Drop Giant if caller has it.  Eventually we should warn about
126 	 * being called with Giant held.
127 	 */
128 	while (mtx_owned(&Giant))
129 		mtx_unlock(&Giant);
130 
131 	p = td->td_proc;
132 	if (p == initproc) {
133 		printf("init died (signal %d, exit %d)\n",
134 		    WTERMSIG(rv), WEXITSTATUS(rv));
135 		panic("Going nowhere without my init!");
136 	}
137 
138 	/*
139 	 * MUST abort all other threads before proceeding past here.
140 	 */
141 	PROC_LOCK(p);
142 	if (p->p_flag & P_HADTHREADS) {
143 retry:
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 			goto retry;
169 
170 		/*
171 		 * All other activity in this process is now stopped.
172 		 * Threading support has been turned off.
173 		 */
174 	}
175 
176 	/*
177 	 * Wakeup anyone in procfs' PIOCWAIT.  They should have a hold
178 	 * on our vmspace, so we should block below until they have
179 	 * released their reference to us.  Note that if they have
180 	 * requested S_EXIT stops we will block here until they ack
181 	 * via PIOCCONT.
182 	 */
183 	_STOPEVENT(p, S_EXIT, rv);
184 
185 	/*
186 	 * Note that we are exiting and do another wakeup of anyone in
187 	 * PIOCWAIT in case they aren't listening for S_EXIT stops or
188 	 * decided to wait again after we told them we are exiting.
189 	 */
190 	p->p_flag |= P_WEXIT;
191 	wakeup(&p->p_stype);
192 
193 	/*
194 	 * Wait for any processes that have a hold on our vmspace to
195 	 * release their reference.
196 	 */
197 	while (p->p_lock > 0)
198 		msleep(&p->p_lock, &p->p_mtx, PWAIT, "exithold", 0);
199 
200 	PROC_UNLOCK(p);
201 
202 #ifdef AUDIT
203 	/*
204 	 * The Sun BSM exit token contains two components: an exit status as
205 	 * passed to exit(), and a return value to indicate what sort of exit
206 	 * it was.  The exit status is WEXITSTATUS(rv), but it's not clear
207 	 * what the return value is.
208 	 */
209 	AUDIT_ARG(exit, WEXITSTATUS(rv), 0);
210 	AUDIT_SYSCALL_EXIT(0, td);
211 #endif
212 
213 	/* Are we a task leader? */
214 	if (p == p->p_leader) {
215 		mtx_lock(&ppeers_lock);
216 		q = p->p_peers;
217 		while (q != NULL) {
218 			PROC_LOCK(q);
219 			psignal(q, SIGKILL);
220 			PROC_UNLOCK(q);
221 			q = q->p_peers;
222 		}
223 		while (p->p_peers != NULL)
224 			msleep(p, &ppeers_lock, PWAIT, "exit1", 0);
225 		mtx_unlock(&ppeers_lock);
226 	}
227 
228 	/*
229 	 * Check if any loadable modules need anything done at process exit.
230 	 * E.g. SYSV IPC stuff
231 	 * XXX what if one of these generates an error?
232 	 */
233 	EVENTHANDLER_INVOKE(process_exit, p);
234 
235 	MALLOC(p->p_ru, struct rusage *, sizeof(struct rusage),
236 		M_ZOMBIE, M_WAITOK);
237 	/*
238 	 * If parent is waiting for us to exit or exec,
239 	 * P_PPWAIT is set; we will wakeup the parent below.
240 	 */
241 	PROC_LOCK(p);
242 	stopprofclock(p);
243 	p->p_flag &= ~(P_TRACED | P_PPWAIT);
244 
245 	/*
246 	 * Stop the real interval timer.  If the handler is currently
247 	 * executing, prevent it from rearming itself and let it finish.
248 	 */
249 	if (timevalisset(&p->p_realtimer.it_value) &&
250 	    callout_stop(&p->p_itcallout) == 0) {
251 		timevalclear(&p->p_realtimer.it_interval);
252 		msleep(&p->p_itcallout, &p->p_mtx, PWAIT, "ritwait", 0);
253 		KASSERT(!timevalisset(&p->p_realtimer.it_value),
254 		    ("realtime timer is still armed"));
255 	}
256 	PROC_UNLOCK(p);
257 
258 	/*
259 	 * Reset any sigio structures pointing to us as a result of
260 	 * F_SETOWN with our pid.
261 	 */
262 	funsetownlst(&p->p_sigiolst);
263 
264 	/*
265 	 * If this process has an nlminfo data area (for lockd), release it
266 	 */
267 	if (nlminfo_release_p != NULL && p->p_nlminfo != NULL)
268 		(*nlminfo_release_p)(p);
269 
270 	/*
271 	 * Close open files and release open-file table.
272 	 * This may block!
273 	 */
274 	fdfree(td);
275 
276 	/*
277 	 * If this thread tickled GEOM, we need to wait for the giggling to
278 	 * stop before we return to userland
279 	 */
280 	if (td->td_pflags & TDP_GEOM)
281 		g_waitidle();
282 
283 	/*
284 	 * Remove ourself from our leader's peer list and wake our leader.
285 	 */
286 	mtx_lock(&ppeers_lock);
287 	if (p->p_leader->p_peers) {
288 		q = p->p_leader;
289 		while (q->p_peers != p)
290 			q = q->p_peers;
291 		q->p_peers = p->p_peers;
292 		wakeup(p->p_leader);
293 	}
294 	mtx_unlock(&ppeers_lock);
295 
296 	vmspace_exit(td);
297 
298 	mtx_lock(&Giant);	/* XXX TTY */
299 	sx_xlock(&proctree_lock);
300 	if (SESS_LEADER(p)) {
301 		struct session *sp;
302 
303 		sp = p->p_session;
304 		if (sp->s_ttyvp) {
305 			/*
306 			 * Controlling process.
307 			 * Signal foreground pgrp,
308 			 * drain controlling terminal
309 			 * and revoke access to controlling terminal.
310 			 */
311 			if (sp->s_ttyp && (sp->s_ttyp->t_session == sp)) {
312 				tp = sp->s_ttyp;
313 				if (sp->s_ttyp->t_pgrp) {
314 					PGRP_LOCK(sp->s_ttyp->t_pgrp);
315 					pgsignal(sp->s_ttyp->t_pgrp, SIGHUP, 1);
316 					PGRP_UNLOCK(sp->s_ttyp->t_pgrp);
317 				}
318 				/* XXX tp should be locked. */
319 				sx_xunlock(&proctree_lock);
320 				(void) ttywait(tp);
321 				sx_xlock(&proctree_lock);
322 				/*
323 				 * The tty could have been revoked
324 				 * if we blocked.
325 				 */
326 				if (sp->s_ttyvp) {
327 					ttyvp = sp->s_ttyvp;
328 					SESS_LOCK(p->p_session);
329 					sp->s_ttyvp = NULL;
330 					SESS_UNLOCK(p->p_session);
331 					sx_xunlock(&proctree_lock);
332 					VOP_LOCK(ttyvp, LK_EXCLUSIVE, td);
333 					VOP_REVOKE(ttyvp, REVOKEALL);
334 					vput(ttyvp);
335 					sx_xlock(&proctree_lock);
336 				}
337 			}
338 			if (sp->s_ttyvp) {
339 				ttyvp = sp->s_ttyvp;
340 				SESS_LOCK(p->p_session);
341 				sp->s_ttyvp = NULL;
342 				SESS_UNLOCK(p->p_session);
343 				vrele(ttyvp);
344 			}
345 			/*
346 			 * s_ttyp is not zero'd; we use this to indicate
347 			 * that the session once had a controlling terminal.
348 			 * (for logging and informational purposes)
349 			 */
350 		}
351 		SESS_LOCK(p->p_session);
352 		sp->s_leader = NULL;
353 		SESS_UNLOCK(p->p_session);
354 	}
355 	fixjobc(p, p->p_pgrp, 0);
356 	sx_xunlock(&proctree_lock);
357 	(void)acct_process(td);
358 	mtx_unlock(&Giant);
359 #ifdef KTRACE
360 	/*
361 	 * Drain any pending records on the thread and release the trace
362 	 * file.  It might be better if drain-and-clear were atomic.
363 	 */
364 	ktrprocexit(td);
365 	PROC_LOCK(p);
366 	mtx_lock(&ktrace_mtx);
367 	p->p_traceflag = 0;	/* don't trace the vrele() */
368 	tracevp = p->p_tracevp;
369 	p->p_tracevp = NULL;
370 	tracecred = p->p_tracecred;
371 	p->p_tracecred = NULL;
372 	mtx_unlock(&ktrace_mtx);
373 	PROC_UNLOCK(p);
374 	if (tracevp != NULL) {
375 		locked = VFS_LOCK_GIANT(tracevp->v_mount);
376 		vrele(tracevp);
377 		VFS_UNLOCK_GIANT(locked);
378 	}
379 	if (tracecred != NULL)
380 		crfree(tracecred);
381 #endif
382 	/*
383 	 * Release reference to text vnode
384 	 */
385 	if ((vtmp = p->p_textvp) != NULL) {
386 		p->p_textvp = NULL;
387 		locked = VFS_LOCK_GIANT(vtmp->v_mount);
388 		vrele(vtmp);
389 		VFS_UNLOCK_GIANT(locked);
390 	}
391 
392 	/*
393 	 * Release our limits structure.
394 	 */
395 	PROC_LOCK(p);
396 	plim = p->p_limit;
397 	p->p_limit = NULL;
398 	PROC_UNLOCK(p);
399 	lim_free(plim);
400 
401 	/*
402 	 * Remove proc from allproc queue and pidhash chain.
403 	 * Place onto zombproc.  Unlink from parent's child list.
404 	 */
405 	sx_xlock(&allproc_lock);
406 	LIST_REMOVE(p, p_list);
407 	LIST_INSERT_HEAD(&zombproc, p, p_list);
408 	LIST_REMOVE(p, p_hash);
409 	sx_xunlock(&allproc_lock);
410 
411 	/*
412 	 * Reparent all of our children to init.
413 	 */
414 	sx_xlock(&proctree_lock);
415 	q = LIST_FIRST(&p->p_children);
416 	if (q != NULL)		/* only need this if any child is S_ZOMB */
417 		wakeup(initproc);
418 	for (; q != NULL; q = nq) {
419 		nq = LIST_NEXT(q, p_sibling);
420 		PROC_LOCK(q);
421 		proc_reparent(q, initproc);
422 		q->p_sigparent = SIGCHLD;
423 		/*
424 		 * Traced processes are killed
425 		 * since their existence means someone is screwing up.
426 		 */
427 		if (q->p_flag & P_TRACED) {
428 			q->p_flag &= ~(P_TRACED | P_STOPPED_TRACE);
429 			psignal(q, SIGKILL);
430 		}
431 		PROC_UNLOCK(q);
432 	}
433 
434 	/*
435 	 * Save exit status and finalize rusage info except for times,
436 	 * adding in child rusage info later when our time is locked.
437 	 */
438 	PROC_LOCK(p);
439 	p->p_xstat = rv;
440 	p->p_xthread = td;
441 	p->p_stats->p_ru.ru_nvcsw++;
442 	*p->p_ru = p->p_stats->p_ru;
443 
444 	/*
445 	 * Notify interested parties of our demise.
446 	 */
447 	KNOTE_LOCKED(&p->p_klist, NOTE_EXIT);
448 
449 	/*
450 	 * Just delete all entries in the p_klist. At this point we won't
451 	 * report any more events, and there are nasty race conditions that
452 	 * can beat us if we don't.
453 	 */
454 	knlist_clear(&p->p_klist, 1);
455 
456 	/*
457 	 * Notify parent that we're gone.  If parent has the PS_NOCLDWAIT
458 	 * flag set, or if the handler is set to SIG_IGN, notify process
459 	 * 1 instead (and hope it will handle this situation).
460 	 */
461 	PROC_LOCK(p->p_pptr);
462 	mtx_lock(&p->p_pptr->p_sigacts->ps_mtx);
463 	if (p->p_pptr->p_sigacts->ps_flag & (PS_NOCLDWAIT | PS_CLDSIGIGN)) {
464 		struct proc *pp;
465 
466 		mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx);
467 		pp = p->p_pptr;
468 		PROC_UNLOCK(pp);
469 		proc_reparent(p, initproc);
470 		p->p_sigparent = SIGCHLD;
471 		PROC_LOCK(p->p_pptr);
472 		/*
473 		 * If this was the last child of our parent, notify
474 		 * parent, so in case he was wait(2)ing, he will
475 		 * continue.
476 		 */
477 		if (LIST_EMPTY(&pp->p_children))
478 			wakeup(pp);
479 	} else
480 		mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx);
481 
482 	if (p->p_pptr == initproc)
483 		psignal(p->p_pptr, SIGCHLD);
484 	else if (p->p_sigparent != 0) {
485 		if (p->p_sigparent == SIGCHLD)
486 			childproc_exited(p);
487 		else	/* LINUX thread */
488 			psignal(p->p_pptr, p->p_sigparent);
489 	}
490 	PROC_UNLOCK(p->p_pptr);
491 	PROC_UNLOCK(p);
492 
493 	/*
494 	 * Finally, call machine-dependent code to release the remaining
495 	 * resources including address space.
496 	 * The address space is released by "vmspace_exitfree(p)" in
497 	 * vm_waitproc().
498 	 */
499 	cpu_exit(td);
500 
501 	WITNESS_WARN(WARN_PANIC, &proctree_lock.sx_object,
502 	    "process (pid %d) exiting", p->p_pid);
503 
504 	PROC_LOCK(p);
505 	PROC_LOCK(p->p_pptr);
506 	sx_xunlock(&proctree_lock);
507 
508 	/*
509 	 * The state PRS_ZOMBIE prevents other proesses from sending
510 	 * signal to the process, to avoid memory leak, we free memory
511 	 * for signal queue at the time when the state is set.
512 	 */
513 	sigqueue_flush(&p->p_sigqueue);
514 	sigqueue_flush(&td->td_sigqueue);
515 
516 	/*
517 	 * We have to wait until after acquiring all locks before
518 	 * changing p_state.  We need to avoid all possible context
519 	 * switches (including ones from blocking on a mutex) while
520 	 * marked as a zombie.  We also have to set the zombie state
521 	 * before we release the parent process' proc lock to avoid
522 	 * a lost wakeup.  So, we first call wakeup, then we grab the
523 	 * sched lock, update the state, and release the parent process'
524 	 * proc lock.
525 	 */
526 	wakeup(p->p_pptr);
527 	mtx_lock_spin(&sched_lock);
528 	p->p_state = PRS_ZOMBIE;
529 	PROC_UNLOCK(p->p_pptr);
530 
531 	sched_exit(p->p_pptr, td);
532 
533 	/*
534 	 * Hopefully no one will try to deliver a signal to the process this
535 	 * late in the game.
536 	 */
537 	knlist_destroy(&p->p_klist);
538 
539 	/*
540 	 * Make sure the scheduler takes this thread out of its tables etc.
541 	 * This will also release this thread's reference to the ucred.
542 	 * Other thread parts to release include pcb bits and such.
543 	 */
544 	thread_exit();
545 }
546 
547 
548 #ifndef _SYS_SYSPROTO_H_
549 struct abort2_args {
550 	char *why;
551 	int nargs;
552 	void **args;
553 };
554 #endif
555 
556 /*
557  * MPSAFE.
558  */
559 int
560 abort2(struct thread *td, struct abort2_args *uap)
561 {
562 	struct proc *p = td->td_proc;
563 	struct sbuf *sb;
564 	void *uargs[16];
565 	int error, i, sig;
566 
567 	error = 0;	/* satisfy compiler */
568 
569 	/*
570 	 * Do it right now so we can log either proper call of abort2(), or
571 	 * note, that invalid argument was passed. 512 is big enough to
572 	 * handle 16 arguments' descriptions with additional comments.
573 	 */
574 	sb = sbuf_new(NULL, NULL, 512, SBUF_FIXEDLEN);
575 	sbuf_clear(sb);
576 	sbuf_printf(sb, "%s(pid %d uid %d) aborted: ",
577 	    p->p_comm, p->p_pid, td->td_ucred->cr_uid);
578 	/*
579 	 * Since we can't return from abort2(), send SIGKILL in cases, where
580 	 * abort2() was called improperly
581 	 */
582 	sig = SIGKILL;
583 	/* Prevent from DoSes from user-space. */
584 	if (uap->nargs < 0 || uap->nargs > 16)
585 		goto out;
586 	if (uap->args == NULL)
587 		goto out;
588 	error = copyin(uap->args, uargs, uap->nargs * sizeof(void *));
589 	if (error != 0)
590 		goto out;
591 	/*
592 	 * Limit size of 'reason' string to 128. Will fit even when
593 	 * maximal number of arguments was chosen to be logged.
594 	 */
595 	if (uap->why != NULL) {
596 		error = sbuf_copyin(sb, uap->why, 128);
597 		if (error < 0)
598 			goto out;
599 	} else {
600 		sbuf_printf(sb, "(null)");
601 	}
602 	if (uap->nargs) {
603 		sbuf_printf(sb, "(");
604 		for (i = 0;i < uap->nargs; i++)
605 			sbuf_printf(sb, "%s%p", i == 0 ? "" : ", ", uargs[i]);
606 		sbuf_printf(sb, ")");
607 	}
608 	/*
609 	 * Final stage: arguments were proper, string has been
610 	 * successfully copied from userspace, and copying pointers
611 	 * from user-space succeed.
612 	 */
613 	sig = SIGABRT;
614 out:
615 	if (sig == SIGKILL) {
616 		sbuf_trim(sb);
617 		sbuf_printf(sb, " (Reason text inaccessible)");
618 	}
619 	sbuf_cat(sb, "\n");
620 	sbuf_finish(sb);
621 	log(LOG_INFO, "%s", sbuf_data(sb));
622 	sbuf_delete(sb);
623 	exit1(td, W_EXITCODE(0, sig));
624 	return (0);
625 }
626 
627 
628 #ifdef COMPAT_43
629 /*
630  * The dirty work is handled by kern_wait().
631  *
632  * MPSAFE.
633  */
634 int
635 owait(struct thread *td, struct owait_args *uap __unused)
636 {
637 	int error, status;
638 
639 	error = kern_wait(td, WAIT_ANY, &status, 0, NULL);
640 	if (error == 0)
641 		td->td_retval[1] = status;
642 	return (error);
643 }
644 #endif /* COMPAT_43 */
645 
646 /*
647  * The dirty work is handled by kern_wait().
648  *
649  * MPSAFE.
650  */
651 int
652 wait4(struct thread *td, struct wait_args *uap)
653 {
654 	struct rusage ru, *rup;
655 	int error, status;
656 
657 	if (uap->rusage != NULL)
658 		rup = &ru;
659 	else
660 		rup = NULL;
661 	error = kern_wait(td, uap->pid, &status, uap->options, rup);
662 	if (uap->status != NULL && error == 0)
663 		error = copyout(&status, uap->status, sizeof(status));
664 	if (uap->rusage != NULL && error == 0)
665 		error = copyout(&ru, uap->rusage, sizeof(struct rusage));
666 	return (error);
667 }
668 
669 int
670 kern_wait(struct thread *td, pid_t pid, int *status, int options,
671     struct rusage *rusage)
672 {
673 	struct proc *p, *q, *t;
674 	int error, nfound;
675 
676 	AUDIT_ARG(pid, pid);
677 
678 	q = td->td_proc;
679 	if (pid == 0) {
680 		PROC_LOCK(q);
681 		pid = -q->p_pgid;
682 		PROC_UNLOCK(q);
683 	}
684 	if (options &~ (WUNTRACED|WNOHANG|WCONTINUED|WLINUXCLONE))
685 		return (EINVAL);
686 loop:
687 	if (q->p_flag & P_STATCHILD) {
688 		PROC_LOCK(q);
689 		q->p_flag &= ~P_STATCHILD;
690 		PROC_UNLOCK(q);
691 	}
692 	nfound = 0;
693 	sx_xlock(&proctree_lock);
694 	LIST_FOREACH(p, &q->p_children, p_sibling) {
695 		PROC_LOCK(p);
696 		if (pid != WAIT_ANY &&
697 		    p->p_pid != pid && p->p_pgid != -pid) {
698 			PROC_UNLOCK(p);
699 			continue;
700 		}
701 		if (p_canwait(td, p)) {
702 			PROC_UNLOCK(p);
703 			continue;
704 		}
705 
706 		/*
707 		 * This special case handles a kthread spawned by linux_clone
708 		 * (see linux_misc.c).  The linux_wait4 and linux_waitpid
709 		 * functions need to be able to distinguish between waiting
710 		 * on a process and waiting on a thread.  It is a thread if
711 		 * p_sigparent is not SIGCHLD, and the WLINUXCLONE option
712 		 * signifies we want to wait for threads and not processes.
713 		 */
714 		if ((p->p_sigparent != SIGCHLD) ^
715 		    ((options & WLINUXCLONE) != 0)) {
716 			PROC_UNLOCK(p);
717 			continue;
718 		}
719 
720 		nfound++;
721 		if (p->p_state == PRS_ZOMBIE) {
722 
723 			/*
724 			 * It is possible that the last thread of this
725 			 * process is still running on another CPU
726 			 * in thread_exit() after having dropped the process
727 			 * lock via PROC_UNLOCK() but before it has completed
728 			 * cpu_throw().  In that case, the other thread must
729 			 * still hold sched_lock, so simply by acquiring
730 			 * sched_lock once we will wait long enough for the
731 			 * thread to exit in that case.
732 			 */
733 			mtx_lock_spin(&sched_lock);
734 			mtx_unlock_spin(&sched_lock);
735 
736 			td->td_retval[0] = p->p_pid;
737 			if (status)
738 				*status = p->p_xstat;	/* convert to int */
739 			if (rusage) {
740 				*rusage = *p->p_ru;
741 				calcru(p, &rusage->ru_utime, &rusage->ru_stime);
742 			}
743 
744 			PROC_LOCK(q);
745 			sigqueue_take(p->p_ksi);
746 			PROC_UNLOCK(q);
747 
748 			/*
749 			 * If we got the child via a ptrace 'attach',
750 			 * we need to give it back to the old parent.
751 			 */
752 			PROC_UNLOCK(p);
753 			if (p->p_oppid && (t = pfind(p->p_oppid)) != NULL) {
754 				PROC_LOCK(p);
755 				p->p_oppid = 0;
756 				proc_reparent(p, t);
757 				PROC_UNLOCK(p);
758 				tdsignal(t, NULL, SIGCHLD, p->p_ksi);
759 				wakeup(t);
760 				PROC_UNLOCK(t);
761 				sx_xunlock(&proctree_lock);
762 				return (0);
763 			}
764 
765 			/*
766 			 * Remove other references to this process to ensure
767 			 * we have an exclusive reference.
768 			 */
769 			sx_xlock(&allproc_lock);
770 			LIST_REMOVE(p, p_list);	/* off zombproc */
771 			sx_xunlock(&allproc_lock);
772 			LIST_REMOVE(p, p_sibling);
773 			leavepgrp(p);
774 			sx_xunlock(&proctree_lock);
775 
776 			/*
777 			 * As a side effect of this lock, we know that
778 			 * all other writes to this proc are visible now, so
779 			 * no more locking is needed for p.
780 			 */
781 			PROC_LOCK(p);
782 			p->p_xstat = 0;		/* XXX: why? */
783 			PROC_UNLOCK(p);
784 			PROC_LOCK(q);
785 			ruadd(&q->p_stats->p_cru, &q->p_crux, p->p_ru,
786 			    &p->p_rux);
787 			PROC_UNLOCK(q);
788 			FREE(p->p_ru, M_ZOMBIE);
789 			p->p_ru = NULL;
790 
791 			/*
792 			 * Decrement the count of procs running with this uid.
793 			 */
794 			(void)chgproccnt(p->p_ucred->cr_ruidinfo, -1, 0);
795 
796 			/*
797 			 * Free credentials, arguments, and sigacts.
798 			 */
799 			crfree(p->p_ucred);
800 			p->p_ucred = NULL;
801 			pargs_drop(p->p_args);
802 			p->p_args = NULL;
803 			sigacts_free(p->p_sigacts);
804 			p->p_sigacts = NULL;
805 
806 			/*
807 			 * Do any thread-system specific cleanups.
808 			 */
809 			thread_wait(p);
810 
811 			/*
812 			 * Give vm and machine-dependent layer a chance
813 			 * to free anything that cpu_exit couldn't
814 			 * release while still running in process context.
815 			 */
816 			vm_waitproc(p);
817 #ifdef MAC
818 			mac_destroy_proc(p);
819 #endif
820 #ifdef AUDIT
821 			audit_proc_free(p);
822 #endif
823 			KASSERT(FIRST_THREAD_IN_PROC(p),
824 			    ("kern_wait: no residual thread!"));
825 			uma_zfree(proc_zone, p);
826 			sx_xlock(&allproc_lock);
827 			nprocs--;
828 			sx_xunlock(&allproc_lock);
829 			return (0);
830 		}
831 		mtx_lock_spin(&sched_lock);
832 		if ((p->p_flag & P_STOPPED_SIG) &&
833 		    (p->p_suspcount == p->p_numthreads) &&
834 		    (p->p_flag & P_WAITED) == 0 &&
835 		    (p->p_flag & P_TRACED || options & WUNTRACED)) {
836 			mtx_unlock_spin(&sched_lock);
837 			p->p_flag |= P_WAITED;
838 			sx_xunlock(&proctree_lock);
839 			td->td_retval[0] = p->p_pid;
840 			if (status)
841 				*status = W_STOPCODE(p->p_xstat);
842 
843 			PROC_LOCK(q);
844 			sigqueue_take(p->p_ksi);
845 			PROC_UNLOCK(q);
846 			PROC_UNLOCK(p);
847 
848 			return (0);
849 		}
850 		mtx_unlock_spin(&sched_lock);
851 		if (options & WCONTINUED && (p->p_flag & P_CONTINUED)) {
852 			sx_xunlock(&proctree_lock);
853 			td->td_retval[0] = p->p_pid;
854 			p->p_flag &= ~P_CONTINUED;
855 
856 			PROC_LOCK(q);
857 			sigqueue_take(p->p_ksi);
858 			PROC_UNLOCK(q);
859 			PROC_UNLOCK(p);
860 
861 			if (status)
862 				*status = SIGCONT;
863 			return (0);
864 		}
865 		PROC_UNLOCK(p);
866 	}
867 	if (nfound == 0) {
868 		sx_xunlock(&proctree_lock);
869 		return (ECHILD);
870 	}
871 	if (options & WNOHANG) {
872 		sx_xunlock(&proctree_lock);
873 		td->td_retval[0] = 0;
874 		return (0);
875 	}
876 	PROC_LOCK(q);
877 	sx_xunlock(&proctree_lock);
878 	if (q->p_flag & P_STATCHILD) {
879 		q->p_flag &= ~P_STATCHILD;
880 		error = 0;
881 	} else
882 		error = msleep(q, &q->p_mtx, PWAIT | PCATCH, "wait", 0);
883 	PROC_UNLOCK(q);
884 	if (error)
885 		return (error);
886 	goto loop;
887 }
888 
889 /*
890  * Make process 'parent' the new parent of process 'child'.
891  * Must be called with an exclusive hold of proctree lock.
892  */
893 void
894 proc_reparent(struct proc *child, struct proc *parent)
895 {
896 
897 	sx_assert(&proctree_lock, SX_XLOCKED);
898 	PROC_LOCK_ASSERT(child, MA_OWNED);
899 	if (child->p_pptr == parent)
900 		return;
901 
902 	PROC_LOCK(child->p_pptr);
903 	sigqueue_take(child->p_ksi);
904 	PROC_UNLOCK(child->p_pptr);
905 	LIST_REMOVE(child, p_sibling);
906 	LIST_INSERT_HEAD(&parent->p_children, child, p_sibling);
907 	child->p_pptr = parent;
908 }
909