xref: /freebsd/sys/kern/kern_exit.c (revision 3dc7424242553c18368d96ed8cccc0407134cd04)
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 <vm/vm.h>
76 #include <vm/vm_extern.h>
77 #include <vm/vm_param.h>
78 #include <vm/pmap.h>
79 #include <vm/vm_map.h>
80 #include <vm/vm_page.h>
81 #include <vm/uma.h>
82 
83 /* Required to be non-static for SysVR4 emulator */
84 MALLOC_DEFINE(M_ZOMBIE, "zombie", "zombie proc status");
85 
86 /* Hook for NFS teardown procedure. */
87 void (*nlminfo_release_p)(struct proc *p);
88 
89 /*
90  * exit --
91  *	Death of process.
92  *
93  * MPSAFE
94  */
95 void
96 sys_exit(struct thread *td, struct sys_exit_args *uap)
97 {
98 
99 	exit1(td, W_EXITCODE(uap->rval, 0));
100 	/* NOTREACHED */
101 }
102 
103 /*
104  * Exit: deallocate address space and other resources, change proc state
105  * to zombie, and unlink proc from allproc and parent's lists.  Save exit
106  * status and rusage for wait().  Check for child processes and orphan them.
107  */
108 void
109 exit1(struct thread *td, int rv)
110 {
111 	struct bintime new_switchtime;
112 	struct proc *p, *nq, *q;
113 	struct tty *tp;
114 	struct vnode *ttyvp;
115 	struct vmspace *vm;
116 	struct vnode *vtmp;
117 #ifdef KTRACE
118 	struct vnode *tracevp;
119 	struct ucred *tracecred;
120 #endif
121 	struct plimit *plim;
122 	int locked, refcnt;
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 	p->p_flag |= P_WEXIT;
177 
178 	PROC_LOCK(p->p_pptr);
179 	sigqueue_take(p->p_ksi);
180 	PROC_UNLOCK(p->p_pptr);
181 
182 	PROC_UNLOCK(p);
183 
184 	/* Are we a task leader? */
185 	if (p == p->p_leader) {
186 		mtx_lock(&ppeers_lock);
187 		q = p->p_peers;
188 		while (q != NULL) {
189 			PROC_LOCK(q);
190 			psignal(q, SIGKILL);
191 			PROC_UNLOCK(q);
192 			q = q->p_peers;
193 		}
194 		while (p->p_peers != NULL)
195 			msleep(p, &ppeers_lock, PWAIT, "exit1", 0);
196 		mtx_unlock(&ppeers_lock);
197 	}
198 
199 	PROC_LOCK(p);
200 	_STOPEVENT(p, S_EXIT, rv);
201 	wakeup(&p->p_stype);	/* Wakeup anyone in procfs' PIOCWAIT */
202 	PROC_UNLOCK(p);
203 
204 	/*
205 	 * Check if any loadable modules need anything done at process exit.
206 	 * E.g. SYSV IPC stuff
207 	 * XXX what if one of these generates an error?
208 	 */
209 	EVENTHANDLER_INVOKE(process_exit, p);
210 
211 	MALLOC(p->p_ru, struct rusage *, sizeof(struct rusage),
212 		M_ZOMBIE, M_WAITOK);
213 	/*
214 	 * If parent is waiting for us to exit or exec,
215 	 * P_PPWAIT is set; we will wakeup the parent below.
216 	 */
217 	PROC_LOCK(p);
218 	stopprofclock(p);
219 	p->p_flag &= ~(P_TRACED | P_PPWAIT);
220 
221 	/*
222 	 * Stop the real interval timer.  If the handler is currently
223 	 * executing, prevent it from rearming itself and let it finish.
224 	 */
225 	if (timevalisset(&p->p_realtimer.it_value) &&
226 	    callout_stop(&p->p_itcallout) == 0) {
227 		timevalclear(&p->p_realtimer.it_interval);
228 		msleep(&p->p_itcallout, &p->p_mtx, PWAIT, "ritwait", 0);
229 		KASSERT(!timevalisset(&p->p_realtimer.it_value),
230 		    ("realtime timer is still armed"));
231 	}
232 	sigqueue_flush(&p->p_sigqueue);
233 	sigqueue_flush(&td->td_sigqueue);
234 	PROC_UNLOCK(p);
235 
236 	/*
237 	 * Reset any sigio structures pointing to us as a result of
238 	 * F_SETOWN with our pid.
239 	 */
240 	mtx_lock(&Giant);	/* XXX: not sure if needed */
241 	funsetownlst(&p->p_sigiolst);
242 	mtx_unlock(&Giant);
243 
244 	/*
245 	 * If this process has an nlminfo data area (for lockd), release it
246 	 */
247 	if (nlminfo_release_p != NULL && p->p_nlminfo != NULL)
248 		(*nlminfo_release_p)(p);
249 
250 	/*
251 	 * Close open files and release open-file table.
252 	 * This may block!
253 	 */
254 	fdfree(td);
255 
256 	/*
257 	 * If this thread tickled GEOM, we need to wait for the giggling to
258 	 * stop before we return to userland
259 	 */
260 	if (td->td_pflags & TDP_GEOM)
261 		g_waitidle();
262 
263 	/*
264 	 * Remove ourself from our leader's peer list and wake our leader.
265 	 */
266 	mtx_lock(&ppeers_lock);
267 	if (p->p_leader->p_peers) {
268 		q = p->p_leader;
269 		while (q->p_peers != p)
270 			q = q->p_peers;
271 		q->p_peers = p->p_peers;
272 		wakeup(p->p_leader);
273 	}
274 	mtx_unlock(&ppeers_lock);
275 
276 	/* The next two chunks should probably be moved to vmspace_exit. */
277 	vm = p->p_vmspace;
278 	/*
279 	 * Release user portion of address space.
280 	 * This releases references to vnodes,
281 	 * which could cause I/O if the file has been unlinked.
282 	 * Need to do this early enough that we can still sleep.
283 	 * Can't free the entire vmspace as the kernel stack
284 	 * may be mapped within that space also.
285 	 *
286 	 * Processes sharing the same vmspace may exit in one order, and
287 	 * get cleaned up by vmspace_exit() in a different order.  The
288 	 * last exiting process to reach this point releases as much of
289 	 * the environment as it can, and the last process cleaned up
290 	 * by vmspace_exit() (which decrements exitingcnt) cleans up the
291 	 * remainder.
292 	 */
293 	atomic_add_int(&vm->vm_exitingcnt, 1);
294 	do
295 		refcnt = vm->vm_refcnt;
296 	while (!atomic_cmpset_int(&vm->vm_refcnt, refcnt, refcnt - 1));
297 	if (refcnt == 1) {
298 		shmexit(vm);
299 		pmap_remove_pages(vmspace_pmap(vm), vm_map_min(&vm->vm_map),
300 		    vm_map_max(&vm->vm_map));
301 		(void) vm_map_remove(&vm->vm_map, vm_map_min(&vm->vm_map),
302 		    vm_map_max(&vm->vm_map));
303 	}
304 
305 	sx_xlock(&proctree_lock);
306 	if (SESS_LEADER(p)) {
307 		struct session *sp;
308 
309 		sp = p->p_session;
310 		if (sp->s_ttyvp) {
311 			locked = VFS_LOCK_GIANT(sp->s_ttyvp->v_mount);
312 			/*
313 			 * Controlling process.
314 			 * Signal foreground pgrp,
315 			 * drain controlling terminal
316 			 * and revoke access to controlling terminal.
317 			 */
318 			if (sp->s_ttyp && (sp->s_ttyp->t_session == sp)) {
319 				tp = sp->s_ttyp;
320 				if (sp->s_ttyp->t_pgrp) {
321 					PGRP_LOCK(sp->s_ttyp->t_pgrp);
322 					pgsignal(sp->s_ttyp->t_pgrp, SIGHUP, 1);
323 					PGRP_UNLOCK(sp->s_ttyp->t_pgrp);
324 				}
325 				/* XXX tp should be locked. */
326 				sx_xunlock(&proctree_lock);
327 				(void) ttywait(tp);
328 				sx_xlock(&proctree_lock);
329 				/*
330 				 * The tty could have been revoked
331 				 * if we blocked.
332 				 */
333 				if (sp->s_ttyvp) {
334 					ttyvp = sp->s_ttyvp;
335 					SESS_LOCK(p->p_session);
336 					sp->s_ttyvp = NULL;
337 					SESS_UNLOCK(p->p_session);
338 					sx_xunlock(&proctree_lock);
339 					VOP_LOCK(ttyvp, LK_EXCLUSIVE, td);
340 					VOP_REVOKE(ttyvp, REVOKEALL);
341 					vput(ttyvp);
342 					sx_xlock(&proctree_lock);
343 				}
344 			}
345 			if (sp->s_ttyvp) {
346 				ttyvp = sp->s_ttyvp;
347 				SESS_LOCK(p->p_session);
348 				sp->s_ttyvp = NULL;
349 				SESS_UNLOCK(p->p_session);
350 				vrele(ttyvp);
351 			}
352 			/*
353 			 * s_ttyp is not zero'd; we use this to indicate
354 			 * that the session once had a controlling terminal.
355 			 * (for logging and informational purposes)
356 			 */
357 			VFS_UNLOCK_GIANT(locked);
358 		}
359 		SESS_LOCK(p->p_session);
360 		sp->s_leader = NULL;
361 		SESS_UNLOCK(p->p_session);
362 	}
363 	fixjobc(p, p->p_pgrp, 0);
364 	sx_xunlock(&proctree_lock);
365 	(void)acct_process(td);
366 #ifdef KTRACE
367 	/*
368 	 * Drain any pending records on the thread and release the trace
369 	 * file.  It might be better if drain-and-clear were atomic.
370 	 */
371 	ktrprocexit(td);
372 	PROC_LOCK(p);
373 	mtx_lock(&ktrace_mtx);
374 	p->p_traceflag = 0;	/* don't trace the vrele() */
375 	tracevp = p->p_tracevp;
376 	p->p_tracevp = NULL;
377 	tracecred = p->p_tracecred;
378 	p->p_tracecred = NULL;
379 	mtx_unlock(&ktrace_mtx);
380 	PROC_UNLOCK(p);
381 	if (tracevp != NULL) {
382 		locked = VFS_LOCK_GIANT(tracevp->v_mount);
383 		vrele(tracevp);
384 		VFS_UNLOCK_GIANT(locked);
385 	}
386 	if (tracecred != NULL)
387 		crfree(tracecred);
388 #endif
389 	/*
390 	 * Release reference to text vnode
391 	 */
392 	if ((vtmp = p->p_textvp) != NULL) {
393 		p->p_textvp = NULL;
394 		locked = VFS_LOCK_GIANT(vtmp->v_mount);
395 		vrele(vtmp);
396 		VFS_UNLOCK_GIANT(locked);
397 	}
398 
399 	/*
400 	 * Release our limits structure.
401 	 */
402 	PROC_LOCK(p);
403 	plim = p->p_limit;
404 	p->p_limit = NULL;
405 	PROC_UNLOCK(p);
406 	lim_free(plim);
407 
408 	/*
409 	 * Remove proc from allproc queue and pidhash chain.
410 	 * Place onto zombproc.  Unlink from parent's child list.
411 	 */
412 	sx_xlock(&allproc_lock);
413 	LIST_REMOVE(p, p_list);
414 	LIST_INSERT_HEAD(&zombproc, p, p_list);
415 	LIST_REMOVE(p, p_hash);
416 	sx_xunlock(&allproc_lock);
417 
418 	sx_xlock(&proctree_lock);
419 	q = LIST_FIRST(&p->p_children);
420 	if (q != NULL)		/* only need this if any child is S_ZOMB */
421 		wakeup(initproc);
422 	for (; q != NULL; q = nq) {
423 		nq = LIST_NEXT(q, p_sibling);
424 		PROC_LOCK(q);
425 		proc_reparent(q, initproc);
426 		q->p_sigparent = SIGCHLD;
427 		/*
428 		 * Traced processes are killed
429 		 * since their existence means someone is screwing up.
430 		 */
431 		if (q->p_flag & P_TRACED) {
432 			q->p_flag &= ~(P_TRACED | P_STOPPED_TRACE);
433 			psignal(q, SIGKILL);
434 		}
435 		PROC_UNLOCK(q);
436 	}
437 
438 	/*
439 	 * Save exit status and finalize rusage info except for times,
440 	 * adding in child rusage info later when our time is locked.
441 	 */
442 	PROC_LOCK(p);
443 	p->p_xstat = rv;
444 	p->p_xthread = td;
445 	p->p_stats->p_ru.ru_nvcsw++;
446 	*p->p_ru = p->p_stats->p_ru;
447 
448 	/*
449 	 * Notify interested parties of our demise.
450 	 */
451 	KNOTE_LOCKED(&p->p_klist, NOTE_EXIT);
452 
453 	/*
454 	 * Just delete all entries in the p_klist. At this point we won't
455 	 * report any more events, and there are nasty race conditions that
456 	 * can beat us if we don't.
457 	 */
458 	knlist_clear(&p->p_klist, 1);
459 
460 	/*
461 	 * Notify parent that we're gone.  If parent has the PS_NOCLDWAIT
462 	 * flag set, or if the handler is set to SIG_IGN, notify process
463 	 * 1 instead (and hope it will handle this situation).
464 	 */
465 	PROC_LOCK(p->p_pptr);
466 	mtx_lock(&p->p_pptr->p_sigacts->ps_mtx);
467 	if (p->p_pptr->p_sigacts->ps_flag & (PS_NOCLDWAIT | PS_CLDSIGIGN)) {
468 		struct proc *pp;
469 
470 		mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx);
471 		pp = p->p_pptr;
472 		PROC_UNLOCK(pp);
473 		proc_reparent(p, initproc);
474 		p->p_sigparent = SIGCHLD;
475 		PROC_LOCK(p->p_pptr);
476 		/*
477 		 * If this was the last child of our parent, notify
478 		 * parent, so in case he was wait(2)ing, he will
479 		 * continue.
480 		 */
481 		if (LIST_EMPTY(&pp->p_children))
482 			wakeup(pp);
483 	} else
484 		mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx);
485 
486 	if (p->p_pptr == initproc)
487 		psignal(p->p_pptr, SIGCHLD);
488 	else if (p->p_sigparent != 0) {
489 		if (p->p_sigparent == SIGCHLD)
490 			childproc_exited(p);
491 		else	/* LINUX thread */
492 			psignal(p->p_pptr, p->p_sigparent);
493 	}
494 	PROC_UNLOCK(p->p_pptr);
495 
496 	/*
497 	 * If this is a kthread, then wakeup anyone waiting for it to exit.
498 	 */
499 	if (p->p_flag & P_KTHREAD)
500 		wakeup(p);
501 	PROC_UNLOCK(p);
502 
503 	/*
504 	 * Finally, call machine-dependent code to release the remaining
505 	 * resources including address space.
506 	 * The address space is released by "vmspace_exitfree(p)" in
507 	 * vm_waitproc().
508 	 */
509 	cpu_exit(td);
510 
511 	WITNESS_WARN(WARN_PANIC, &proctree_lock.sx_object,
512 	    "process (pid %d) exiting", p->p_pid);
513 
514 	PROC_LOCK(p);
515 	PROC_LOCK(p->p_pptr);
516 	sx_xunlock(&proctree_lock);
517 
518 	/*
519 	 * We have to wait until after acquiring all locks before
520 	 * changing p_state.  We need to avoid all possible context
521 	 * switches (including ones from blocking on a mutex) while
522 	 * marked as a zombie.  We also have to set the zombie state
523 	 * before we release the parent process' proc lock to avoid
524 	 * a lost wakeup.  So, we first call wakeup, then we grab the
525 	 * sched lock, update the state, and release the parent process'
526 	 * proc lock.
527 	 */
528 	wakeup(p->p_pptr);
529 	mtx_lock_spin(&sched_lock);
530 	p->p_state = PRS_ZOMBIE;
531 	PROC_UNLOCK(p->p_pptr);
532 
533 	ruadd(p->p_ru, &p->p_rux, &p->p_stats->p_cru, &p->p_crux);
534 
535 	/* Do the same timestamp bookkeeping that mi_switch() would do. */
536 	binuptime(&new_switchtime);
537 	bintime_add(&p->p_rux.rux_runtime, &new_switchtime);
538 	bintime_sub(&p->p_rux.rux_runtime, PCPU_PTR(switchtime));
539 	PCPU_SET(switchtime, new_switchtime);
540 	PCPU_SET(switchticks, ticks);
541 	cnt.v_swtch++;
542 
543 	sched_exit(p->p_pptr, td);
544 
545 	/*
546 	 * Hopefully no one will try to deliver a signal to the process this
547 	 * late in the game.
548 	 */
549 	knlist_destroy(&p->p_klist);
550 
551 	/*
552 	 * Make sure the scheduler takes this thread out of its tables etc.
553 	 * This will also release this thread's reference to the ucred.
554 	 * Other thread parts to release include pcb bits and such.
555 	 */
556 	thread_exit();
557 }
558 
559 
560 #ifndef _SYS_SYSPROTO_H_
561 struct abort2_args {
562 	char *why;
563 	int nargs;
564 	void **args;
565 };
566 #endif
567 
568 /*
569  * MPSAFE.
570  */
571 int
572 abort2(struct thread *td, struct abort2_args *uap)
573 {
574 	struct proc *p = td->td_proc;
575 	struct sbuf *sb;
576 	void *uargs[16];
577 	int error, i, sig;
578 
579 	error = 0;	/* satisfy compiler */
580 
581 	/*
582 	 * Do it right now so we can log either proper call of abort2(), or
583 	 * note, that invalid argument was passed. 512 is big enough to
584 	 * handle 16 arguments' descriptions with additional comments.
585 	 */
586 	sb = sbuf_new(NULL, NULL, 512, SBUF_FIXEDLEN);
587 	sbuf_clear(sb);
588 	sbuf_printf(sb, "%s(pid %d uid %d) aborted: ",
589 	    p->p_comm, p->p_pid, td->td_ucred->cr_uid);
590 	/*
591 	 * Since we can't return from abort2(), send SIGKILL in cases, where
592 	 * abort2() was called improperly
593 	 */
594 	sig = SIGKILL;
595 	/* Prevent from DoSes from user-space. */
596 	if (uap->nargs < 0 || uap->nargs > 16)
597 		goto out;
598 	if (uap->args == NULL)
599 		goto out;
600 	error = copyin(uap->args, uargs, uap->nargs * sizeof(void *));
601 	if (error != 0)
602 		goto out;
603 	/*
604 	 * Limit size of 'reason' string to 128. Will fit even when
605 	 * maximal number of arguments was chosen to be logged.
606 	 */
607 	if (uap->why != NULL) {
608 		error = sbuf_copyin(sb, uap->why, 128);
609 		if (error < 0)
610 			goto out;
611 	} else {
612 		sbuf_printf(sb, "(null)");
613 	}
614 	if (uap->nargs) {
615 		sbuf_printf(sb, "(");
616 		for (i = 0;i < uap->nargs; i++)
617 			sbuf_printf(sb, "%s%p", i == 0 ? "" : ", ", uargs[i]);
618 		sbuf_printf(sb, ")");
619 	}
620 	/*
621 	 * Final stage: arguments were proper, string has been
622 	 * successfully copied from userspace, and copying pointers
623 	 * from user-space succeed.
624 	 */
625 	sig = SIGABRT;
626 out:
627 	if (sig == SIGKILL) {
628 		sbuf_trim(sb);
629 		sbuf_printf(sb, " (Reason text inaccessible)");
630 	}
631 	sbuf_cat(sb, "\n");
632 	sbuf_finish(sb);
633 	log(LOG_INFO, "%s", sbuf_data(sb));
634 	sbuf_delete(sb);
635 	exit1(td, W_EXITCODE(0, sig));
636 	return (0);
637 }
638 
639 
640 #ifdef COMPAT_43
641 /*
642  * The dirty work is handled by kern_wait().
643  *
644  * MPSAFE.
645  */
646 int
647 owait(struct thread *td, struct owait_args *uap __unused)
648 {
649 	int error, status;
650 
651 	error = kern_wait(td, WAIT_ANY, &status, 0, NULL);
652 	if (error == 0)
653 		td->td_retval[1] = status;
654 	return (error);
655 }
656 #endif /* COMPAT_43 */
657 
658 /*
659  * The dirty work is handled by kern_wait().
660  *
661  * MPSAFE.
662  */
663 int
664 wait4(struct thread *td, struct wait_args *uap)
665 {
666 	struct rusage ru, *rup;
667 	int error, status;
668 
669 	if (uap->rusage != NULL)
670 		rup = &ru;
671 	else
672 		rup = NULL;
673 	error = kern_wait(td, uap->pid, &status, uap->options, rup);
674 	if (uap->status != NULL && error == 0)
675 		error = copyout(&status, uap->status, sizeof(status));
676 	if (uap->rusage != NULL && error == 0)
677 		error = copyout(&ru, uap->rusage, sizeof(struct rusage));
678 	return (error);
679 }
680 
681 int
682 kern_wait(struct thread *td, pid_t pid, int *status, int options,
683     struct rusage *rusage)
684 {
685 	struct proc *p, *q, *t;
686 	int error, nfound;
687 
688 	q = td->td_proc;
689 	if (pid == 0) {
690 		PROC_LOCK(q);
691 		pid = -q->p_pgid;
692 		PROC_UNLOCK(q);
693 	}
694 	if (options &~ (WUNTRACED|WNOHANG|WCONTINUED|WLINUXCLONE))
695 		return (EINVAL);
696 loop:
697 	if (q->p_flag & P_STATCHILD) {
698 		PROC_LOCK(q);
699 		q->p_flag &= ~P_STATCHILD;
700 		PROC_UNLOCK(q);
701 	}
702 	nfound = 0;
703 	sx_xlock(&proctree_lock);
704 	LIST_FOREACH(p, &q->p_children, p_sibling) {
705 		PROC_LOCK(p);
706 		if (pid != WAIT_ANY &&
707 		    p->p_pid != pid && p->p_pgid != -pid) {
708 			PROC_UNLOCK(p);
709 			continue;
710 		}
711 		if (p_canwait(td, p)) {
712 			PROC_UNLOCK(p);
713 			continue;
714 		}
715 
716 		/*
717 		 * This special case handles a kthread spawned by linux_clone
718 		 * (see linux_misc.c).  The linux_wait4 and linux_waitpid
719 		 * functions need to be able to distinguish between waiting
720 		 * on a process and waiting on a thread.  It is a thread if
721 		 * p_sigparent is not SIGCHLD, and the WLINUXCLONE option
722 		 * signifies we want to wait for threads and not processes.
723 		 */
724 		if ((p->p_sigparent != SIGCHLD) ^
725 		    ((options & WLINUXCLONE) != 0)) {
726 			PROC_UNLOCK(p);
727 			continue;
728 		}
729 
730 		nfound++;
731 		if (p->p_state == PRS_ZOMBIE) {
732 
733 			/*
734 			 * It is possible that the last thread of this
735 			 * process is still running on another CPU
736 			 * in thread_exit() after having dropped the process
737 			 * lock via PROC_UNLOCK() but before it has completed
738 			 * cpu_throw().  In that case, the other thread must
739 			 * still hold sched_lock, so simply by acquiring
740 			 * sched_lock once we will wait long enough for the
741 			 * thread to exit in that case.
742 			 */
743 			mtx_lock_spin(&sched_lock);
744 			mtx_unlock_spin(&sched_lock);
745 
746 			td->td_retval[0] = p->p_pid;
747 			if (status)
748 				*status = p->p_xstat;	/* convert to int */
749 			if (rusage) {
750 				*rusage = *p->p_ru;
751 				calcru(p, &rusage->ru_utime, &rusage->ru_stime);
752 			}
753 
754 			PROC_LOCK(q);
755 			sigqueue_take(p->p_ksi);
756 			PROC_UNLOCK(q);
757 
758 			/*
759 			 * If we got the child via a ptrace 'attach',
760 			 * we need to give it back to the old parent.
761 			 */
762 			PROC_UNLOCK(p);
763 			if (p->p_oppid && (t = pfind(p->p_oppid)) != NULL) {
764 				PROC_LOCK(p);
765 				p->p_oppid = 0;
766 				proc_reparent(p, t);
767 				PROC_UNLOCK(p);
768 				tdsignal(t, NULL, SIGCHLD, p->p_ksi);
769 				wakeup(t);
770 				PROC_UNLOCK(t);
771 				sx_xunlock(&proctree_lock);
772 				return (0);
773 			}
774 
775 			/*
776 			 * Remove other references to this process to ensure
777 			 * we have an exclusive reference.
778 			 */
779 			sx_xlock(&allproc_lock);
780 			LIST_REMOVE(p, p_list);	/* off zombproc */
781 			sx_xunlock(&allproc_lock);
782 			LIST_REMOVE(p, p_sibling);
783 			leavepgrp(p);
784 			sx_xunlock(&proctree_lock);
785 
786 			/*
787 			 * As a side effect of this lock, we know that
788 			 * all other writes to this proc are visible now, so
789 			 * no more locking is needed for p.
790 			 */
791 			PROC_LOCK(p);
792 			p->p_xstat = 0;		/* XXX: why? */
793 			PROC_UNLOCK(p);
794 			PROC_LOCK(q);
795 			ruadd(&q->p_stats->p_cru, &q->p_crux, p->p_ru,
796 			    &p->p_rux);
797 			PROC_UNLOCK(q);
798 			FREE(p->p_ru, M_ZOMBIE);
799 			p->p_ru = NULL;
800 
801 			/*
802 			 * Decrement the count of procs running with this uid.
803 			 */
804 			(void)chgproccnt(p->p_ucred->cr_ruidinfo, -1, 0);
805 
806 			/*
807 			 * Free credentials, arguments, and sigacts.
808 			 */
809 			crfree(p->p_ucred);
810 			p->p_ucred = NULL;
811 			pargs_drop(p->p_args);
812 			p->p_args = NULL;
813 			sigacts_free(p->p_sigacts);
814 			p->p_sigacts = NULL;
815 
816 			/*
817 			 * Do any thread-system specific cleanups.
818 			 */
819 			thread_wait(p);
820 
821 			/*
822 			 * Give vm and machine-dependent layer a chance
823 			 * to free anything that cpu_exit couldn't
824 			 * release while still running in process context.
825 			 */
826 			vm_waitproc(p);
827 #ifdef MAC
828 			mac_destroy_proc(p);
829 #endif
830 			KASSERT(FIRST_THREAD_IN_PROC(p),
831 			    ("kern_wait: no residual thread!"));
832 			uma_zfree(proc_zone, p);
833 			sx_xlock(&allproc_lock);
834 			nprocs--;
835 			sx_xunlock(&allproc_lock);
836 			return (0);
837 		}
838 		mtx_lock_spin(&sched_lock);
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 			mtx_unlock_spin(&sched_lock);
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 			PROC_UNLOCK(p);
850 
851 			PROC_LOCK(q);
852 			sigqueue_take(p->p_ksi);
853 			PROC_UNLOCK(q);
854 
855 			return (0);
856 		}
857 		mtx_unlock_spin(&sched_lock);
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 			PROC_UNLOCK(p);
863 
864 			PROC_LOCK(q);
865 			sigqueue_take(p->p_ksi);
866 			PROC_UNLOCK(q);
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 
904 	sx_assert(&proctree_lock, SX_XLOCKED);
905 	PROC_LOCK_ASSERT(child, MA_OWNED);
906 	if (child->p_pptr == parent)
907 		return;
908 
909 	LIST_REMOVE(child, p_sibling);
910 	LIST_INSERT_HEAD(&parent->p_children, child, p_sibling);
911 	child->p_pptr = parent;
912 }
913