xref: /freebsd/sys/kern/kern_exit.c (revision d056fa046c6a91b90cd98165face0e42a33a5173)
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/mac.h>
69 #include <sys/shm.h>
70 #include <sys/sem.h>
71 #ifdef KTRACE
72 #include <sys/ktrace.h>
73 #endif
74 
75 #include <security/audit/audit.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_LOCK(p->p_pptr);
201 	sigqueue_take(p->p_ksi);
202 	PROC_UNLOCK(p->p_pptr);
203 
204 	PROC_UNLOCK(p);
205 
206 #ifdef AUDIT
207 	/*
208 	 * The Sun BSM exit token contains two components: an exit status as
209 	 * passed to exit(), and a return value to indicate what sort of exit
210 	 * it was.  The exit status is WEXITSTATUS(rv), but it's not clear
211 	 * what the return value is.
212 	 */
213 	AUDIT_ARG(exit, WEXITSTATUS(rv), 0);
214 	AUDIT_SYSCALL_EXIT(0, td);
215 #endif
216 
217 	/* Are we a task leader? */
218 	if (p == p->p_leader) {
219 		mtx_lock(&ppeers_lock);
220 		q = p->p_peers;
221 		while (q != NULL) {
222 			PROC_LOCK(q);
223 			psignal(q, SIGKILL);
224 			PROC_UNLOCK(q);
225 			q = q->p_peers;
226 		}
227 		while (p->p_peers != NULL)
228 			msleep(p, &ppeers_lock, PWAIT, "exit1", 0);
229 		mtx_unlock(&ppeers_lock);
230 	}
231 
232 	/*
233 	 * Check if any loadable modules need anything done at process exit.
234 	 * E.g. SYSV IPC stuff
235 	 * XXX what if one of these generates an error?
236 	 */
237 	EVENTHANDLER_INVOKE(process_exit, p);
238 
239 	MALLOC(p->p_ru, struct rusage *, sizeof(struct rusage),
240 		M_ZOMBIE, M_WAITOK);
241 	/*
242 	 * If parent is waiting for us to exit or exec,
243 	 * P_PPWAIT is set; we will wakeup the parent below.
244 	 */
245 	PROC_LOCK(p);
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 	sigqueue_flush(&p->p_sigqueue);
261 	sigqueue_flush(&td->td_sigqueue);
262 	PROC_UNLOCK(p);
263 
264 	/*
265 	 * Reset any sigio structures pointing to us as a result of
266 	 * F_SETOWN with our pid.
267 	 */
268 	funsetownlst(&p->p_sigiolst);
269 
270 	/*
271 	 * If this process has an nlminfo data area (for lockd), release it
272 	 */
273 	if (nlminfo_release_p != NULL && p->p_nlminfo != NULL)
274 		(*nlminfo_release_p)(p);
275 
276 	/*
277 	 * Close open files and release open-file table.
278 	 * This may block!
279 	 */
280 	fdfree(td);
281 
282 	/*
283 	 * If this thread tickled GEOM, we need to wait for the giggling to
284 	 * stop before we return to userland
285 	 */
286 	if (td->td_pflags & TDP_GEOM)
287 		g_waitidle();
288 
289 	/*
290 	 * Remove ourself from our leader's peer list and wake our leader.
291 	 */
292 	mtx_lock(&ppeers_lock);
293 	if (p->p_leader->p_peers) {
294 		q = p->p_leader;
295 		while (q->p_peers != p)
296 			q = q->p_peers;
297 		q->p_peers = p->p_peers;
298 		wakeup(p->p_leader);
299 	}
300 	mtx_unlock(&ppeers_lock);
301 
302 	vmspace_exit(td);
303 
304 	sx_xlock(&proctree_lock);
305 	if (SESS_LEADER(p)) {
306 		struct session *sp;
307 
308 		sp = p->p_session;
309 		if (sp->s_ttyvp) {
310 			locked = VFS_LOCK_GIANT(sp->s_ttyvp->v_mount);
311 			/*
312 			 * Controlling process.
313 			 * Signal foreground pgrp,
314 			 * drain controlling terminal
315 			 * and revoke access to controlling terminal.
316 			 */
317 			if (sp->s_ttyp && (sp->s_ttyp->t_session == sp)) {
318 				tp = sp->s_ttyp;
319 				if (sp->s_ttyp->t_pgrp) {
320 					PGRP_LOCK(sp->s_ttyp->t_pgrp);
321 					pgsignal(sp->s_ttyp->t_pgrp, SIGHUP, 1);
322 					PGRP_UNLOCK(sp->s_ttyp->t_pgrp);
323 				}
324 				/* XXX tp should be locked. */
325 				sx_xunlock(&proctree_lock);
326 				(void) ttywait(tp);
327 				sx_xlock(&proctree_lock);
328 				/*
329 				 * The tty could have been revoked
330 				 * if we blocked.
331 				 */
332 				if (sp->s_ttyvp) {
333 					ttyvp = sp->s_ttyvp;
334 					SESS_LOCK(p->p_session);
335 					sp->s_ttyvp = NULL;
336 					SESS_UNLOCK(p->p_session);
337 					sx_xunlock(&proctree_lock);
338 					VOP_LOCK(ttyvp, LK_EXCLUSIVE, td);
339 					VOP_REVOKE(ttyvp, REVOKEALL);
340 					vput(ttyvp);
341 					sx_xlock(&proctree_lock);
342 				}
343 			}
344 			if (sp->s_ttyvp) {
345 				ttyvp = sp->s_ttyvp;
346 				SESS_LOCK(p->p_session);
347 				sp->s_ttyvp = NULL;
348 				SESS_UNLOCK(p->p_session);
349 				vrele(ttyvp);
350 			}
351 			/*
352 			 * s_ttyp is not zero'd; we use this to indicate
353 			 * that the session once had a controlling terminal.
354 			 * (for logging and informational purposes)
355 			 */
356 			VFS_UNLOCK_GIANT(locked);
357 		}
358 		SESS_LOCK(p->p_session);
359 		sp->s_leader = NULL;
360 		SESS_UNLOCK(p->p_session);
361 	}
362 	fixjobc(p, p->p_pgrp, 0);
363 	sx_xunlock(&proctree_lock);
364 	(void)acct_process(td);
365 #ifdef KTRACE
366 	/*
367 	 * Drain any pending records on the thread and release the trace
368 	 * file.  It might be better if drain-and-clear were atomic.
369 	 */
370 	ktrprocexit(td);
371 	PROC_LOCK(p);
372 	mtx_lock(&ktrace_mtx);
373 	p->p_traceflag = 0;	/* don't trace the vrele() */
374 	tracevp = p->p_tracevp;
375 	p->p_tracevp = NULL;
376 	tracecred = p->p_tracecred;
377 	p->p_tracecred = NULL;
378 	mtx_unlock(&ktrace_mtx);
379 	PROC_UNLOCK(p);
380 	if (tracevp != NULL) {
381 		locked = VFS_LOCK_GIANT(tracevp->v_mount);
382 		vrele(tracevp);
383 		VFS_UNLOCK_GIANT(locked);
384 	}
385 	if (tracecred != NULL)
386 		crfree(tracecred);
387 #endif
388 	/*
389 	 * Release reference to text vnode
390 	 */
391 	if ((vtmp = p->p_textvp) != NULL) {
392 		p->p_textvp = NULL;
393 		locked = VFS_LOCK_GIANT(vtmp->v_mount);
394 		vrele(vtmp);
395 		VFS_UNLOCK_GIANT(locked);
396 	}
397 
398 	/*
399 	 * Release our limits structure.
400 	 */
401 	PROC_LOCK(p);
402 	plim = p->p_limit;
403 	p->p_limit = NULL;
404 	PROC_UNLOCK(p);
405 	lim_free(plim);
406 
407 	/*
408 	 * Remove proc from allproc queue and pidhash chain.
409 	 * Place onto zombproc.  Unlink from parent's child list.
410 	 */
411 	sx_xlock(&allproc_lock);
412 	LIST_REMOVE(p, p_list);
413 	LIST_INSERT_HEAD(&zombproc, p, p_list);
414 	LIST_REMOVE(p, p_hash);
415 	sx_xunlock(&allproc_lock);
416 
417 	/*
418 	 * Reparent all of our children to init.
419 	 */
420 	sx_xlock(&proctree_lock);
421 	q = LIST_FIRST(&p->p_children);
422 	if (q != NULL)		/* only need this if any child is S_ZOMB */
423 		wakeup(initproc);
424 	for (; q != NULL; q = nq) {
425 		nq = LIST_NEXT(q, p_sibling);
426 		PROC_LOCK(q);
427 		proc_reparent(q, initproc);
428 		q->p_sigparent = SIGCHLD;
429 		/*
430 		 * Traced processes are killed
431 		 * since their existence means someone is screwing up.
432 		 */
433 		if (q->p_flag & P_TRACED) {
434 			q->p_flag &= ~(P_TRACED | P_STOPPED_TRACE);
435 			psignal(q, SIGKILL);
436 		}
437 		PROC_UNLOCK(q);
438 	}
439 
440 	/*
441 	 * Save exit status and finalize rusage info except for times,
442 	 * adding in child rusage info later when our time is locked.
443 	 */
444 	PROC_LOCK(p);
445 	p->p_xstat = rv;
446 	p->p_xthread = td;
447 	p->p_stats->p_ru.ru_nvcsw++;
448 	*p->p_ru = p->p_stats->p_ru;
449 
450 	/*
451 	 * Notify interested parties of our demise.
452 	 */
453 	KNOTE_LOCKED(&p->p_klist, NOTE_EXIT);
454 
455 	/*
456 	 * Just delete all entries in the p_klist. At this point we won't
457 	 * report any more events, and there are nasty race conditions that
458 	 * can beat us if we don't.
459 	 */
460 	knlist_clear(&p->p_klist, 1);
461 
462 	/*
463 	 * Notify parent that we're gone.  If parent has the PS_NOCLDWAIT
464 	 * flag set, or if the handler is set to SIG_IGN, notify process
465 	 * 1 instead (and hope it will handle this situation).
466 	 */
467 	PROC_LOCK(p->p_pptr);
468 	mtx_lock(&p->p_pptr->p_sigacts->ps_mtx);
469 	if (p->p_pptr->p_sigacts->ps_flag & (PS_NOCLDWAIT | PS_CLDSIGIGN)) {
470 		struct proc *pp;
471 
472 		mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx);
473 		pp = p->p_pptr;
474 		PROC_UNLOCK(pp);
475 		proc_reparent(p, initproc);
476 		p->p_sigparent = SIGCHLD;
477 		PROC_LOCK(p->p_pptr);
478 		/*
479 		 * If this was the last child of our parent, notify
480 		 * parent, so in case he was wait(2)ing, he will
481 		 * continue.
482 		 */
483 		if (LIST_EMPTY(&pp->p_children))
484 			wakeup(pp);
485 	} else
486 		mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx);
487 
488 	if (p->p_pptr == initproc)
489 		psignal(p->p_pptr, SIGCHLD);
490 	else if (p->p_sigparent != 0) {
491 		if (p->p_sigparent == SIGCHLD)
492 			childproc_exited(p);
493 		else	/* LINUX thread */
494 			psignal(p->p_pptr, p->p_sigparent);
495 	}
496 	PROC_UNLOCK(p->p_pptr);
497 	PROC_UNLOCK(p);
498 
499 	/*
500 	 * Finally, call machine-dependent code to release the remaining
501 	 * resources including address space.
502 	 * The address space is released by "vmspace_exitfree(p)" in
503 	 * vm_waitproc().
504 	 */
505 	cpu_exit(td);
506 
507 	WITNESS_WARN(WARN_PANIC, &proctree_lock.sx_object,
508 	    "process (pid %d) exiting", p->p_pid);
509 
510 	PROC_LOCK(p);
511 	PROC_LOCK(p->p_pptr);
512 	sx_xunlock(&proctree_lock);
513 
514 	/*
515 	 * We have to wait until after acquiring all locks before
516 	 * changing p_state.  We need to avoid all possible context
517 	 * switches (including ones from blocking on a mutex) while
518 	 * marked as a zombie.  We also have to set the zombie state
519 	 * before we release the parent process' proc lock to avoid
520 	 * a lost wakeup.  So, we first call wakeup, then we grab the
521 	 * sched lock, update the state, and release the parent process'
522 	 * proc lock.
523 	 */
524 	wakeup(p->p_pptr);
525 	mtx_lock_spin(&sched_lock);
526 	p->p_state = PRS_ZOMBIE;
527 	PROC_UNLOCK(p->p_pptr);
528 
529 	sched_exit(p->p_pptr, td);
530 
531 	/*
532 	 * Hopefully no one will try to deliver a signal to the process this
533 	 * late in the game.
534 	 */
535 	knlist_destroy(&p->p_klist);
536 
537 	/*
538 	 * Make sure the scheduler takes this thread out of its tables etc.
539 	 * This will also release this thread's reference to the ucred.
540 	 * Other thread parts to release include pcb bits and such.
541 	 */
542 	thread_exit();
543 }
544 
545 
546 #ifndef _SYS_SYSPROTO_H_
547 struct abort2_args {
548 	char *why;
549 	int nargs;
550 	void **args;
551 };
552 #endif
553 
554 /*
555  * MPSAFE.
556  */
557 int
558 abort2(struct thread *td, struct abort2_args *uap)
559 {
560 	struct proc *p = td->td_proc;
561 	struct sbuf *sb;
562 	void *uargs[16];
563 	int error, i, sig;
564 
565 	error = 0;	/* satisfy compiler */
566 
567 	/*
568 	 * Do it right now so we can log either proper call of abort2(), or
569 	 * note, that invalid argument was passed. 512 is big enough to
570 	 * handle 16 arguments' descriptions with additional comments.
571 	 */
572 	sb = sbuf_new(NULL, NULL, 512, SBUF_FIXEDLEN);
573 	sbuf_clear(sb);
574 	sbuf_printf(sb, "%s(pid %d uid %d) aborted: ",
575 	    p->p_comm, p->p_pid, td->td_ucred->cr_uid);
576 	/*
577 	 * Since we can't return from abort2(), send SIGKILL in cases, where
578 	 * abort2() was called improperly
579 	 */
580 	sig = SIGKILL;
581 	/* Prevent from DoSes from user-space. */
582 	if (uap->nargs < 0 || uap->nargs > 16)
583 		goto out;
584 	if (uap->args == NULL)
585 		goto out;
586 	error = copyin(uap->args, uargs, uap->nargs * sizeof(void *));
587 	if (error != 0)
588 		goto out;
589 	/*
590 	 * Limit size of 'reason' string to 128. Will fit even when
591 	 * maximal number of arguments was chosen to be logged.
592 	 */
593 	if (uap->why != NULL) {
594 		error = sbuf_copyin(sb, uap->why, 128);
595 		if (error < 0)
596 			goto out;
597 	} else {
598 		sbuf_printf(sb, "(null)");
599 	}
600 	if (uap->nargs) {
601 		sbuf_printf(sb, "(");
602 		for (i = 0;i < uap->nargs; i++)
603 			sbuf_printf(sb, "%s%p", i == 0 ? "" : ", ", uargs[i]);
604 		sbuf_printf(sb, ")");
605 	}
606 	/*
607 	 * Final stage: arguments were proper, string has been
608 	 * successfully copied from userspace, and copying pointers
609 	 * from user-space succeed.
610 	 */
611 	sig = SIGABRT;
612 out:
613 	if (sig == SIGKILL) {
614 		sbuf_trim(sb);
615 		sbuf_printf(sb, " (Reason text inaccessible)");
616 	}
617 	sbuf_cat(sb, "\n");
618 	sbuf_finish(sb);
619 	log(LOG_INFO, "%s", sbuf_data(sb));
620 	sbuf_delete(sb);
621 	exit1(td, W_EXITCODE(0, sig));
622 	return (0);
623 }
624 
625 
626 #ifdef COMPAT_43
627 /*
628  * The dirty work is handled by kern_wait().
629  *
630  * MPSAFE.
631  */
632 int
633 owait(struct thread *td, struct owait_args *uap __unused)
634 {
635 	int error, status;
636 
637 	error = kern_wait(td, WAIT_ANY, &status, 0, NULL);
638 	if (error == 0)
639 		td->td_retval[1] = status;
640 	return (error);
641 }
642 #endif /* COMPAT_43 */
643 
644 /*
645  * The dirty work is handled by kern_wait().
646  *
647  * MPSAFE.
648  */
649 int
650 wait4(struct thread *td, struct wait_args *uap)
651 {
652 	struct rusage ru, *rup;
653 	int error, status;
654 
655 	if (uap->rusage != NULL)
656 		rup = &ru;
657 	else
658 		rup = NULL;
659 	error = kern_wait(td, uap->pid, &status, uap->options, rup);
660 	if (uap->status != NULL && error == 0)
661 		error = copyout(&status, uap->status, sizeof(status));
662 	if (uap->rusage != NULL && error == 0)
663 		error = copyout(&ru, uap->rusage, sizeof(struct rusage));
664 	return (error);
665 }
666 
667 int
668 kern_wait(struct thread *td, pid_t pid, int *status, int options,
669     struct rusage *rusage)
670 {
671 	struct proc *p, *q, *t;
672 	int error, nfound;
673 
674 	AUDIT_ARG(pid, pid);
675 
676 	q = td->td_proc;
677 	if (pid == 0) {
678 		PROC_LOCK(q);
679 		pid = -q->p_pgid;
680 		PROC_UNLOCK(q);
681 	}
682 	if (options &~ (WUNTRACED|WNOHANG|WCONTINUED|WLINUXCLONE))
683 		return (EINVAL);
684 loop:
685 	if (q->p_flag & P_STATCHILD) {
686 		PROC_LOCK(q);
687 		q->p_flag &= ~P_STATCHILD;
688 		PROC_UNLOCK(q);
689 	}
690 	nfound = 0;
691 	sx_xlock(&proctree_lock);
692 	LIST_FOREACH(p, &q->p_children, p_sibling) {
693 		PROC_LOCK(p);
694 		if (pid != WAIT_ANY &&
695 		    p->p_pid != pid && p->p_pgid != -pid) {
696 			PROC_UNLOCK(p);
697 			continue;
698 		}
699 		if (p_canwait(td, p)) {
700 			PROC_UNLOCK(p);
701 			continue;
702 		}
703 
704 		/*
705 		 * This special case handles a kthread spawned by linux_clone
706 		 * (see linux_misc.c).  The linux_wait4 and linux_waitpid
707 		 * functions need to be able to distinguish between waiting
708 		 * on a process and waiting on a thread.  It is a thread if
709 		 * p_sigparent is not SIGCHLD, and the WLINUXCLONE option
710 		 * signifies we want to wait for threads and not processes.
711 		 */
712 		if ((p->p_sigparent != SIGCHLD) ^
713 		    ((options & WLINUXCLONE) != 0)) {
714 			PROC_UNLOCK(p);
715 			continue;
716 		}
717 
718 		nfound++;
719 		if (p->p_state == PRS_ZOMBIE) {
720 
721 			/*
722 			 * It is possible that the last thread of this
723 			 * process is still running on another CPU
724 			 * in thread_exit() after having dropped the process
725 			 * lock via PROC_UNLOCK() but before it has completed
726 			 * cpu_throw().  In that case, the other thread must
727 			 * still hold sched_lock, so simply by acquiring
728 			 * sched_lock once we will wait long enough for the
729 			 * thread to exit in that case.
730 			 */
731 			mtx_lock_spin(&sched_lock);
732 			mtx_unlock_spin(&sched_lock);
733 
734 			td->td_retval[0] = p->p_pid;
735 			if (status)
736 				*status = p->p_xstat;	/* convert to int */
737 			if (rusage) {
738 				*rusage = *p->p_ru;
739 				calcru(p, &rusage->ru_utime, &rusage->ru_stime);
740 			}
741 
742 			PROC_LOCK(q);
743 			sigqueue_take(p->p_ksi);
744 			PROC_UNLOCK(q);
745 
746 			/*
747 			 * If we got the child via a ptrace 'attach',
748 			 * we need to give it back to the old parent.
749 			 */
750 			PROC_UNLOCK(p);
751 			if (p->p_oppid && (t = pfind(p->p_oppid)) != NULL) {
752 				PROC_LOCK(p);
753 				p->p_oppid = 0;
754 				proc_reparent(p, t);
755 				PROC_UNLOCK(p);
756 				tdsignal(t, NULL, SIGCHLD, p->p_ksi);
757 				wakeup(t);
758 				PROC_UNLOCK(t);
759 				sx_xunlock(&proctree_lock);
760 				return (0);
761 			}
762 
763 			/*
764 			 * Remove other references to this process to ensure
765 			 * we have an exclusive reference.
766 			 */
767 			sx_xlock(&allproc_lock);
768 			LIST_REMOVE(p, p_list);	/* off zombproc */
769 			sx_xunlock(&allproc_lock);
770 			LIST_REMOVE(p, p_sibling);
771 			leavepgrp(p);
772 			sx_xunlock(&proctree_lock);
773 
774 			/*
775 			 * As a side effect of this lock, we know that
776 			 * all other writes to this proc are visible now, so
777 			 * no more locking is needed for p.
778 			 */
779 			PROC_LOCK(p);
780 			p->p_xstat = 0;		/* XXX: why? */
781 			PROC_UNLOCK(p);
782 			PROC_LOCK(q);
783 			ruadd(&q->p_stats->p_cru, &q->p_crux, p->p_ru,
784 			    &p->p_rux);
785 			PROC_UNLOCK(q);
786 			FREE(p->p_ru, M_ZOMBIE);
787 			p->p_ru = NULL;
788 
789 			/*
790 			 * Decrement the count of procs running with this uid.
791 			 */
792 			(void)chgproccnt(p->p_ucred->cr_ruidinfo, -1, 0);
793 
794 			/*
795 			 * Free credentials, arguments, and sigacts.
796 			 */
797 			crfree(p->p_ucred);
798 			p->p_ucred = NULL;
799 			pargs_drop(p->p_args);
800 			p->p_args = NULL;
801 			sigacts_free(p->p_sigacts);
802 			p->p_sigacts = NULL;
803 
804 			/*
805 			 * Do any thread-system specific cleanups.
806 			 */
807 			thread_wait(p);
808 
809 			/*
810 			 * Give vm and machine-dependent layer a chance
811 			 * to free anything that cpu_exit couldn't
812 			 * release while still running in process context.
813 			 */
814 			vm_waitproc(p);
815 #ifdef MAC
816 			mac_destroy_proc(p);
817 #endif
818 #ifdef AUDIT
819 			audit_proc_free(p);
820 #endif
821 			KASSERT(FIRST_THREAD_IN_PROC(p),
822 			    ("kern_wait: no residual thread!"));
823 			uma_zfree(proc_zone, p);
824 			sx_xlock(&allproc_lock);
825 			nprocs--;
826 			sx_xunlock(&allproc_lock);
827 			return (0);
828 		}
829 		mtx_lock_spin(&sched_lock);
830 		if ((p->p_flag & P_STOPPED_SIG) &&
831 		    (p->p_suspcount == p->p_numthreads) &&
832 		    (p->p_flag & P_WAITED) == 0 &&
833 		    (p->p_flag & P_TRACED || options & WUNTRACED)) {
834 			mtx_unlock_spin(&sched_lock);
835 			p->p_flag |= P_WAITED;
836 			sx_xunlock(&proctree_lock);
837 			td->td_retval[0] = p->p_pid;
838 			if (status)
839 				*status = W_STOPCODE(p->p_xstat);
840 			PROC_UNLOCK(p);
841 
842 			PROC_LOCK(q);
843 			sigqueue_take(p->p_ksi);
844 			PROC_UNLOCK(q);
845 
846 			return (0);
847 		}
848 		mtx_unlock_spin(&sched_lock);
849 		if (options & WCONTINUED && (p->p_flag & P_CONTINUED)) {
850 			sx_xunlock(&proctree_lock);
851 			td->td_retval[0] = p->p_pid;
852 			p->p_flag &= ~P_CONTINUED;
853 			PROC_UNLOCK(p);
854 
855 			PROC_LOCK(q);
856 			sigqueue_take(p->p_ksi);
857 			PROC_UNLOCK(q);
858 
859 			if (status)
860 				*status = SIGCONT;
861 			return (0);
862 		}
863 		PROC_UNLOCK(p);
864 	}
865 	if (nfound == 0) {
866 		sx_xunlock(&proctree_lock);
867 		return (ECHILD);
868 	}
869 	if (options & WNOHANG) {
870 		sx_xunlock(&proctree_lock);
871 		td->td_retval[0] = 0;
872 		return (0);
873 	}
874 	PROC_LOCK(q);
875 	sx_xunlock(&proctree_lock);
876 	if (q->p_flag & P_STATCHILD) {
877 		q->p_flag &= ~P_STATCHILD;
878 		error = 0;
879 	} else
880 		error = msleep(q, &q->p_mtx, PWAIT | PCATCH, "wait", 0);
881 	PROC_UNLOCK(q);
882 	if (error)
883 		return (error);
884 	goto loop;
885 }
886 
887 /*
888  * Make process 'parent' the new parent of process 'child'.
889  * Must be called with an exclusive hold of proctree lock.
890  */
891 void
892 proc_reparent(struct proc *child, struct proc *parent)
893 {
894 
895 	sx_assert(&proctree_lock, SX_XLOCKED);
896 	PROC_LOCK_ASSERT(child, MA_OWNED);
897 	if (child->p_pptr == parent)
898 		return;
899 
900 	LIST_REMOVE(child, p_sibling);
901 	LIST_INSERT_HEAD(&parent->p_children, child, p_sibling);
902 	child->p_pptr = parent;
903 }
904