xref: /freebsd/sys/kern/kern_exit.c (revision 74b07c5412065adfab90021f6ded2bdf7304744d)
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 bintime new_switchtime;
114 	struct proc *p, *nq, *q;
115 	struct tty *tp;
116 	struct vnode *ttyvp;
117 	struct vmspace *vm;
118 	struct vnode *vtmp;
119 #ifdef KTRACE
120 	struct vnode *tracevp;
121 	struct ucred *tracecred;
122 #endif
123 	struct plimit *plim;
124 	int locked, refcnt;
125 
126 	/*
127 	 * Drop Giant if caller has it.  Eventually we should warn about
128 	 * being called with Giant held.
129 	 */
130 	while (mtx_owned(&Giant))
131 		mtx_unlock(&Giant);
132 
133 	p = td->td_proc;
134 	if (p == initproc) {
135 		printf("init died (signal %d, exit %d)\n",
136 		    WTERMSIG(rv), WEXITSTATUS(rv));
137 		panic("Going nowhere without my init!");
138 	}
139 
140 	/*
141 	 * MUST abort all other threads before proceeding past here.
142 	 */
143 	PROC_LOCK(p);
144 	if (p->p_flag & P_HADTHREADS) {
145 retry:
146 		/*
147 		 * First check if some other thread got here before us..
148 		 * if so, act apropriatly, (exit or suspend);
149 		 */
150 		thread_suspend_check(0);
151 
152 		/*
153 		 * Kill off the other threads. This requires
154 		 * some co-operation from other parts of the kernel
155 		 * so it may not be instantaneous.  With this state set
156 		 * any thread entering the kernel from userspace will
157 		 * thread_exit() in trap().  Any thread attempting to
158 		 * sleep will return immediately with EINTR or EWOULDBLOCK
159 		 * which will hopefully force them to back out to userland
160 		 * freeing resources as they go.  Any thread attempting
161 		 * to return to userland will thread_exit() from userret().
162 		 * thread_exit() will unsuspend us when the last of the
163 		 * other threads exits.
164 		 * If there is already a thread singler after resumption,
165 		 * calling thread_single will fail; in that case, we just
166 		 * re-check all suspension request, the thread should
167 		 * either be suspended there or exit.
168 		 */
169 		if (thread_single(SINGLE_EXIT))
170 			goto retry;
171 
172 		/*
173 		 * All other activity in this process is now stopped.
174 		 * Threading support has been turned off.
175 		 */
176 	}
177 
178 	p->p_flag |= P_WEXIT;
179 
180 	PROC_LOCK(p->p_pptr);
181 	sigqueue_take(p->p_ksi);
182 	PROC_UNLOCK(p->p_pptr);
183 
184 	PROC_UNLOCK(p);
185 
186 #ifdef AUDIT
187 	/*
188 	 * The Sun BSM exit token contains two components: an exit status as
189 	 * passed to exit(), and a return value to indicate what sort of exit
190 	 * it was.  The exit status is WEXITSTATUS(rv), but it's not clear
191 	 * what the return value is.
192 	 */
193 	AUDIT_ARG(exit, WEXITSTATUS(rv), 0);
194 	AUDIT_SYSCALL_EXIT(0, td);
195 #endif
196 
197 	/* Are we a task leader? */
198 	if (p == p->p_leader) {
199 		mtx_lock(&ppeers_lock);
200 		q = p->p_peers;
201 		while (q != NULL) {
202 			PROC_LOCK(q);
203 			psignal(q, SIGKILL);
204 			PROC_UNLOCK(q);
205 			q = q->p_peers;
206 		}
207 		while (p->p_peers != NULL)
208 			msleep(p, &ppeers_lock, PWAIT, "exit1", 0);
209 		mtx_unlock(&ppeers_lock);
210 	}
211 
212 	PROC_LOCK(p);
213 	_STOPEVENT(p, S_EXIT, rv);
214 	wakeup(&p->p_stype);	/* Wakeup anyone in procfs' PIOCWAIT */
215 	PROC_UNLOCK(p);
216 
217 	/*
218 	 * Check if any loadable modules need anything done at process exit.
219 	 * E.g. SYSV IPC stuff
220 	 * XXX what if one of these generates an error?
221 	 */
222 	EVENTHANDLER_INVOKE(process_exit, p);
223 
224 	MALLOC(p->p_ru, struct rusage *, sizeof(struct rusage),
225 		M_ZOMBIE, M_WAITOK);
226 	/*
227 	 * If parent is waiting for us to exit or exec,
228 	 * P_PPWAIT is set; we will wakeup the parent below.
229 	 */
230 	PROC_LOCK(p);
231 	stopprofclock(p);
232 	p->p_flag &= ~(P_TRACED | P_PPWAIT);
233 
234 	/*
235 	 * Stop the real interval timer.  If the handler is currently
236 	 * executing, prevent it from rearming itself and let it finish.
237 	 */
238 	if (timevalisset(&p->p_realtimer.it_value) &&
239 	    callout_stop(&p->p_itcallout) == 0) {
240 		timevalclear(&p->p_realtimer.it_interval);
241 		msleep(&p->p_itcallout, &p->p_mtx, PWAIT, "ritwait", 0);
242 		KASSERT(!timevalisset(&p->p_realtimer.it_value),
243 		    ("realtime timer is still armed"));
244 	}
245 	sigqueue_flush(&p->p_sigqueue);
246 	sigqueue_flush(&td->td_sigqueue);
247 	PROC_UNLOCK(p);
248 
249 	/*
250 	 * Reset any sigio structures pointing to us as a result of
251 	 * F_SETOWN with our pid.
252 	 */
253 	mtx_lock(&Giant);	/* XXX: not sure if needed */
254 	funsetownlst(&p->p_sigiolst);
255 	mtx_unlock(&Giant);
256 
257 	/*
258 	 * If this process has an nlminfo data area (for lockd), release it
259 	 */
260 	if (nlminfo_release_p != NULL && p->p_nlminfo != NULL)
261 		(*nlminfo_release_p)(p);
262 
263 	/*
264 	 * Close open files and release open-file table.
265 	 * This may block!
266 	 */
267 	fdfree(td);
268 
269 	/*
270 	 * If this thread tickled GEOM, we need to wait for the giggling to
271 	 * stop before we return to userland
272 	 */
273 	if (td->td_pflags & TDP_GEOM)
274 		g_waitidle();
275 
276 	/*
277 	 * Remove ourself from our leader's peer list and wake our leader.
278 	 */
279 	mtx_lock(&ppeers_lock);
280 	if (p->p_leader->p_peers) {
281 		q = p->p_leader;
282 		while (q->p_peers != p)
283 			q = q->p_peers;
284 		q->p_peers = p->p_peers;
285 		wakeup(p->p_leader);
286 	}
287 	mtx_unlock(&ppeers_lock);
288 
289 	/* The next two chunks should probably be moved to vmspace_exit. */
290 	vm = p->p_vmspace;
291 	/*
292 	 * Release user portion of address space.
293 	 * This releases references to vnodes,
294 	 * which could cause I/O if the file has been unlinked.
295 	 * Need to do this early enough that we can still sleep.
296 	 * Can't free the entire vmspace as the kernel stack
297 	 * may be mapped within that space also.
298 	 *
299 	 * Processes sharing the same vmspace may exit in one order, and
300 	 * get cleaned up by vmspace_exit() in a different order.  The
301 	 * last exiting process to reach this point releases as much of
302 	 * the environment as it can, and the last process cleaned up
303 	 * by vmspace_exit() (which decrements exitingcnt) cleans up the
304 	 * remainder.
305 	 */
306 	atomic_add_int(&vm->vm_exitingcnt, 1);
307 	do
308 		refcnt = vm->vm_refcnt;
309 	while (!atomic_cmpset_int(&vm->vm_refcnt, refcnt, refcnt - 1));
310 	if (refcnt == 1) {
311 		shmexit(vm);
312 		pmap_remove_pages(vmspace_pmap(vm), vm_map_min(&vm->vm_map),
313 		    vm_map_max(&vm->vm_map));
314 		(void) vm_map_remove(&vm->vm_map, vm_map_min(&vm->vm_map),
315 		    vm_map_max(&vm->vm_map));
316 	}
317 
318 	sx_xlock(&proctree_lock);
319 	if (SESS_LEADER(p)) {
320 		struct session *sp;
321 
322 		sp = p->p_session;
323 		if (sp->s_ttyvp) {
324 			locked = VFS_LOCK_GIANT(sp->s_ttyvp->v_mount);
325 			/*
326 			 * Controlling process.
327 			 * Signal foreground pgrp,
328 			 * drain controlling terminal
329 			 * and revoke access to controlling terminal.
330 			 */
331 			if (sp->s_ttyp && (sp->s_ttyp->t_session == sp)) {
332 				tp = sp->s_ttyp;
333 				if (sp->s_ttyp->t_pgrp) {
334 					PGRP_LOCK(sp->s_ttyp->t_pgrp);
335 					pgsignal(sp->s_ttyp->t_pgrp, SIGHUP, 1);
336 					PGRP_UNLOCK(sp->s_ttyp->t_pgrp);
337 				}
338 				/* XXX tp should be locked. */
339 				sx_xunlock(&proctree_lock);
340 				(void) ttywait(tp);
341 				sx_xlock(&proctree_lock);
342 				/*
343 				 * The tty could have been revoked
344 				 * if we blocked.
345 				 */
346 				if (sp->s_ttyvp) {
347 					ttyvp = sp->s_ttyvp;
348 					SESS_LOCK(p->p_session);
349 					sp->s_ttyvp = NULL;
350 					SESS_UNLOCK(p->p_session);
351 					sx_xunlock(&proctree_lock);
352 					VOP_LOCK(ttyvp, LK_EXCLUSIVE, td);
353 					VOP_REVOKE(ttyvp, REVOKEALL);
354 					vput(ttyvp);
355 					sx_xlock(&proctree_lock);
356 				}
357 			}
358 			if (sp->s_ttyvp) {
359 				ttyvp = sp->s_ttyvp;
360 				SESS_LOCK(p->p_session);
361 				sp->s_ttyvp = NULL;
362 				SESS_UNLOCK(p->p_session);
363 				vrele(ttyvp);
364 			}
365 			/*
366 			 * s_ttyp is not zero'd; we use this to indicate
367 			 * that the session once had a controlling terminal.
368 			 * (for logging and informational purposes)
369 			 */
370 			VFS_UNLOCK_GIANT(locked);
371 		}
372 		SESS_LOCK(p->p_session);
373 		sp->s_leader = NULL;
374 		SESS_UNLOCK(p->p_session);
375 	}
376 	fixjobc(p, p->p_pgrp, 0);
377 	sx_xunlock(&proctree_lock);
378 	(void)acct_process(td);
379 #ifdef KTRACE
380 	/*
381 	 * Drain any pending records on the thread and release the trace
382 	 * file.  It might be better if drain-and-clear were atomic.
383 	 */
384 	ktrprocexit(td);
385 	PROC_LOCK(p);
386 	mtx_lock(&ktrace_mtx);
387 	p->p_traceflag = 0;	/* don't trace the vrele() */
388 	tracevp = p->p_tracevp;
389 	p->p_tracevp = NULL;
390 	tracecred = p->p_tracecred;
391 	p->p_tracecred = NULL;
392 	mtx_unlock(&ktrace_mtx);
393 	PROC_UNLOCK(p);
394 	if (tracevp != NULL) {
395 		locked = VFS_LOCK_GIANT(tracevp->v_mount);
396 		vrele(tracevp);
397 		VFS_UNLOCK_GIANT(locked);
398 	}
399 	if (tracecred != NULL)
400 		crfree(tracecred);
401 #endif
402 	/*
403 	 * Release reference to text vnode
404 	 */
405 	if ((vtmp = p->p_textvp) != NULL) {
406 		p->p_textvp = NULL;
407 		locked = VFS_LOCK_GIANT(vtmp->v_mount);
408 		vrele(vtmp);
409 		VFS_UNLOCK_GIANT(locked);
410 	}
411 
412 	/*
413 	 * Release our limits structure.
414 	 */
415 	PROC_LOCK(p);
416 	plim = p->p_limit;
417 	p->p_limit = NULL;
418 	PROC_UNLOCK(p);
419 	lim_free(plim);
420 
421 	/*
422 	 * Remove proc from allproc queue and pidhash chain.
423 	 * Place onto zombproc.  Unlink from parent's child list.
424 	 */
425 	sx_xlock(&allproc_lock);
426 	LIST_REMOVE(p, p_list);
427 	LIST_INSERT_HEAD(&zombproc, p, p_list);
428 	LIST_REMOVE(p, p_hash);
429 	sx_xunlock(&allproc_lock);
430 
431 	/*
432 	 * Reparent all of our children to init.
433 	 */
434 	sx_xlock(&proctree_lock);
435 	q = LIST_FIRST(&p->p_children);
436 	if (q != NULL)		/* only need this if any child is S_ZOMB */
437 		wakeup(initproc);
438 	for (; q != NULL; q = nq) {
439 		nq = LIST_NEXT(q, p_sibling);
440 		PROC_LOCK(q);
441 		proc_reparent(q, initproc);
442 		q->p_sigparent = SIGCHLD;
443 		/*
444 		 * Traced processes are killed
445 		 * since their existence means someone is screwing up.
446 		 */
447 		if (q->p_flag & P_TRACED) {
448 			q->p_flag &= ~(P_TRACED | P_STOPPED_TRACE);
449 			psignal(q, SIGKILL);
450 		}
451 		PROC_UNLOCK(q);
452 	}
453 
454 	/*
455 	 * Save exit status and finalize rusage info except for times,
456 	 * adding in child rusage info later when our time is locked.
457 	 */
458 	PROC_LOCK(p);
459 	p->p_xstat = rv;
460 	p->p_xthread = td;
461 	p->p_stats->p_ru.ru_nvcsw++;
462 	*p->p_ru = p->p_stats->p_ru;
463 
464 	/*
465 	 * Notify interested parties of our demise.
466 	 */
467 	KNOTE_LOCKED(&p->p_klist, NOTE_EXIT);
468 
469 	/*
470 	 * Just delete all entries in the p_klist. At this point we won't
471 	 * report any more events, and there are nasty race conditions that
472 	 * can beat us if we don't.
473 	 */
474 	knlist_clear(&p->p_klist, 1);
475 
476 	/*
477 	 * Notify parent that we're gone.  If parent has the PS_NOCLDWAIT
478 	 * flag set, or if the handler is set to SIG_IGN, notify process
479 	 * 1 instead (and hope it will handle this situation).
480 	 */
481 	PROC_LOCK(p->p_pptr);
482 	mtx_lock(&p->p_pptr->p_sigacts->ps_mtx);
483 	if (p->p_pptr->p_sigacts->ps_flag & (PS_NOCLDWAIT | PS_CLDSIGIGN)) {
484 		struct proc *pp;
485 
486 		mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx);
487 		pp = p->p_pptr;
488 		PROC_UNLOCK(pp);
489 		proc_reparent(p, initproc);
490 		p->p_sigparent = SIGCHLD;
491 		PROC_LOCK(p->p_pptr);
492 		/*
493 		 * If this was the last child of our parent, notify
494 		 * parent, so in case he was wait(2)ing, he will
495 		 * continue.
496 		 */
497 		if (LIST_EMPTY(&pp->p_children))
498 			wakeup(pp);
499 	} else
500 		mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx);
501 
502 	if (p->p_pptr == initproc)
503 		psignal(p->p_pptr, SIGCHLD);
504 	else if (p->p_sigparent != 0) {
505 		if (p->p_sigparent == SIGCHLD)
506 			childproc_exited(p);
507 		else	/* LINUX thread */
508 			psignal(p->p_pptr, p->p_sigparent);
509 	}
510 	PROC_UNLOCK(p->p_pptr);
511 	PROC_UNLOCK(p);
512 
513 	/*
514 	 * Finally, call machine-dependent code to release the remaining
515 	 * resources including address space.
516 	 * The address space is released by "vmspace_exitfree(p)" in
517 	 * vm_waitproc().
518 	 */
519 	cpu_exit(td);
520 
521 	WITNESS_WARN(WARN_PANIC, &proctree_lock.sx_object,
522 	    "process (pid %d) exiting", p->p_pid);
523 
524 	PROC_LOCK(p);
525 	PROC_LOCK(p->p_pptr);
526 	sx_xunlock(&proctree_lock);
527 
528 	/*
529 	 * We have to wait until after acquiring all locks before
530 	 * changing p_state.  We need to avoid all possible context
531 	 * switches (including ones from blocking on a mutex) while
532 	 * marked as a zombie.  We also have to set the zombie state
533 	 * before we release the parent process' proc lock to avoid
534 	 * a lost wakeup.  So, we first call wakeup, then we grab the
535 	 * sched lock, update the state, and release the parent process'
536 	 * proc lock.
537 	 */
538 	wakeup(p->p_pptr);
539 	mtx_lock_spin(&sched_lock);
540 	p->p_state = PRS_ZOMBIE;
541 	PROC_UNLOCK(p->p_pptr);
542 
543 	ruadd(p->p_ru, &p->p_rux, &p->p_stats->p_cru, &p->p_crux);
544 
545 	/* Do the same timestamp bookkeeping that mi_switch() would do. */
546 	binuptime(&new_switchtime);
547 	bintime_add(&p->p_rux.rux_runtime, &new_switchtime);
548 	bintime_sub(&p->p_rux.rux_runtime, PCPU_PTR(switchtime));
549 	PCPU_SET(switchtime, new_switchtime);
550 	PCPU_SET(switchticks, ticks);
551 	cnt.v_swtch++;
552 
553 	sched_exit(p->p_pptr, td);
554 
555 	/*
556 	 * Hopefully no one will try to deliver a signal to the process this
557 	 * late in the game.
558 	 */
559 	knlist_destroy(&p->p_klist);
560 
561 	/*
562 	 * Make sure the scheduler takes this thread out of its tables etc.
563 	 * This will also release this thread's reference to the ucred.
564 	 * Other thread parts to release include pcb bits and such.
565 	 */
566 	thread_exit();
567 }
568 
569 
570 #ifndef _SYS_SYSPROTO_H_
571 struct abort2_args {
572 	char *why;
573 	int nargs;
574 	void **args;
575 };
576 #endif
577 
578 /*
579  * MPSAFE.
580  */
581 int
582 abort2(struct thread *td, struct abort2_args *uap)
583 {
584 	struct proc *p = td->td_proc;
585 	struct sbuf *sb;
586 	void *uargs[16];
587 	int error, i, sig;
588 
589 	error = 0;	/* satisfy compiler */
590 
591 	/*
592 	 * Do it right now so we can log either proper call of abort2(), or
593 	 * note, that invalid argument was passed. 512 is big enough to
594 	 * handle 16 arguments' descriptions with additional comments.
595 	 */
596 	sb = sbuf_new(NULL, NULL, 512, SBUF_FIXEDLEN);
597 	sbuf_clear(sb);
598 	sbuf_printf(sb, "%s(pid %d uid %d) aborted: ",
599 	    p->p_comm, p->p_pid, td->td_ucred->cr_uid);
600 	/*
601 	 * Since we can't return from abort2(), send SIGKILL in cases, where
602 	 * abort2() was called improperly
603 	 */
604 	sig = SIGKILL;
605 	/* Prevent from DoSes from user-space. */
606 	if (uap->nargs < 0 || uap->nargs > 16)
607 		goto out;
608 	if (uap->args == NULL)
609 		goto out;
610 	error = copyin(uap->args, uargs, uap->nargs * sizeof(void *));
611 	if (error != 0)
612 		goto out;
613 	/*
614 	 * Limit size of 'reason' string to 128. Will fit even when
615 	 * maximal number of arguments was chosen to be logged.
616 	 */
617 	if (uap->why != NULL) {
618 		error = sbuf_copyin(sb, uap->why, 128);
619 		if (error < 0)
620 			goto out;
621 	} else {
622 		sbuf_printf(sb, "(null)");
623 	}
624 	if (uap->nargs) {
625 		sbuf_printf(sb, "(");
626 		for (i = 0;i < uap->nargs; i++)
627 			sbuf_printf(sb, "%s%p", i == 0 ? "" : ", ", uargs[i]);
628 		sbuf_printf(sb, ")");
629 	}
630 	/*
631 	 * Final stage: arguments were proper, string has been
632 	 * successfully copied from userspace, and copying pointers
633 	 * from user-space succeed.
634 	 */
635 	sig = SIGABRT;
636 out:
637 	if (sig == SIGKILL) {
638 		sbuf_trim(sb);
639 		sbuf_printf(sb, " (Reason text inaccessible)");
640 	}
641 	sbuf_cat(sb, "\n");
642 	sbuf_finish(sb);
643 	log(LOG_INFO, "%s", sbuf_data(sb));
644 	sbuf_delete(sb);
645 	exit1(td, W_EXITCODE(0, sig));
646 	return (0);
647 }
648 
649 
650 #ifdef COMPAT_43
651 /*
652  * The dirty work is handled by kern_wait().
653  *
654  * MPSAFE.
655  */
656 int
657 owait(struct thread *td, struct owait_args *uap __unused)
658 {
659 	int error, status;
660 
661 	error = kern_wait(td, WAIT_ANY, &status, 0, NULL);
662 	if (error == 0)
663 		td->td_retval[1] = status;
664 	return (error);
665 }
666 #endif /* COMPAT_43 */
667 
668 /*
669  * The dirty work is handled by kern_wait().
670  *
671  * MPSAFE.
672  */
673 int
674 wait4(struct thread *td, struct wait_args *uap)
675 {
676 	struct rusage ru, *rup;
677 	int error, status;
678 
679 	if (uap->rusage != NULL)
680 		rup = &ru;
681 	else
682 		rup = NULL;
683 	error = kern_wait(td, uap->pid, &status, uap->options, rup);
684 	if (uap->status != NULL && error == 0)
685 		error = copyout(&status, uap->status, sizeof(status));
686 	if (uap->rusage != NULL && error == 0)
687 		error = copyout(&ru, uap->rusage, sizeof(struct rusage));
688 	return (error);
689 }
690 
691 int
692 kern_wait(struct thread *td, pid_t pid, int *status, int options,
693     struct rusage *rusage)
694 {
695 	struct proc *p, *q, *t;
696 	int error, nfound;
697 
698 	AUDIT_ARG(pid, pid);
699 
700 	q = td->td_proc;
701 	if (pid == 0) {
702 		PROC_LOCK(q);
703 		pid = -q->p_pgid;
704 		PROC_UNLOCK(q);
705 	}
706 	if (options &~ (WUNTRACED|WNOHANG|WCONTINUED|WLINUXCLONE))
707 		return (EINVAL);
708 loop:
709 	if (q->p_flag & P_STATCHILD) {
710 		PROC_LOCK(q);
711 		q->p_flag &= ~P_STATCHILD;
712 		PROC_UNLOCK(q);
713 	}
714 	nfound = 0;
715 	sx_xlock(&proctree_lock);
716 	LIST_FOREACH(p, &q->p_children, p_sibling) {
717 		PROC_LOCK(p);
718 		if (pid != WAIT_ANY &&
719 		    p->p_pid != pid && p->p_pgid != -pid) {
720 			PROC_UNLOCK(p);
721 			continue;
722 		}
723 		if (p_canwait(td, p)) {
724 			PROC_UNLOCK(p);
725 			continue;
726 		}
727 
728 		/*
729 		 * This special case handles a kthread spawned by linux_clone
730 		 * (see linux_misc.c).  The linux_wait4 and linux_waitpid
731 		 * functions need to be able to distinguish between waiting
732 		 * on a process and waiting on a thread.  It is a thread if
733 		 * p_sigparent is not SIGCHLD, and the WLINUXCLONE option
734 		 * signifies we want to wait for threads and not processes.
735 		 */
736 		if ((p->p_sigparent != SIGCHLD) ^
737 		    ((options & WLINUXCLONE) != 0)) {
738 			PROC_UNLOCK(p);
739 			continue;
740 		}
741 
742 		nfound++;
743 		if (p->p_state == PRS_ZOMBIE) {
744 
745 			/*
746 			 * It is possible that the last thread of this
747 			 * process is still running on another CPU
748 			 * in thread_exit() after having dropped the process
749 			 * lock via PROC_UNLOCK() but before it has completed
750 			 * cpu_throw().  In that case, the other thread must
751 			 * still hold sched_lock, so simply by acquiring
752 			 * sched_lock once we will wait long enough for the
753 			 * thread to exit in that case.
754 			 */
755 			mtx_lock_spin(&sched_lock);
756 			mtx_unlock_spin(&sched_lock);
757 
758 			td->td_retval[0] = p->p_pid;
759 			if (status)
760 				*status = p->p_xstat;	/* convert to int */
761 			if (rusage) {
762 				*rusage = *p->p_ru;
763 				calcru(p, &rusage->ru_utime, &rusage->ru_stime);
764 			}
765 
766 			PROC_LOCK(q);
767 			sigqueue_take(p->p_ksi);
768 			PROC_UNLOCK(q);
769 
770 			/*
771 			 * If we got the child via a ptrace 'attach',
772 			 * we need to give it back to the old parent.
773 			 */
774 			PROC_UNLOCK(p);
775 			if (p->p_oppid && (t = pfind(p->p_oppid)) != NULL) {
776 				PROC_LOCK(p);
777 				p->p_oppid = 0;
778 				proc_reparent(p, t);
779 				PROC_UNLOCK(p);
780 				tdsignal(t, NULL, SIGCHLD, p->p_ksi);
781 				wakeup(t);
782 				PROC_UNLOCK(t);
783 				sx_xunlock(&proctree_lock);
784 				return (0);
785 			}
786 
787 			/*
788 			 * Remove other references to this process to ensure
789 			 * we have an exclusive reference.
790 			 */
791 			sx_xlock(&allproc_lock);
792 			LIST_REMOVE(p, p_list);	/* off zombproc */
793 			sx_xunlock(&allproc_lock);
794 			LIST_REMOVE(p, p_sibling);
795 			leavepgrp(p);
796 			sx_xunlock(&proctree_lock);
797 
798 			/*
799 			 * As a side effect of this lock, we know that
800 			 * all other writes to this proc are visible now, so
801 			 * no more locking is needed for p.
802 			 */
803 			PROC_LOCK(p);
804 			p->p_xstat = 0;		/* XXX: why? */
805 			PROC_UNLOCK(p);
806 			PROC_LOCK(q);
807 			ruadd(&q->p_stats->p_cru, &q->p_crux, p->p_ru,
808 			    &p->p_rux);
809 			PROC_UNLOCK(q);
810 			FREE(p->p_ru, M_ZOMBIE);
811 			p->p_ru = NULL;
812 
813 			/*
814 			 * Decrement the count of procs running with this uid.
815 			 */
816 			(void)chgproccnt(p->p_ucred->cr_ruidinfo, -1, 0);
817 
818 			/*
819 			 * Free credentials, arguments, and sigacts.
820 			 */
821 			crfree(p->p_ucred);
822 			p->p_ucred = NULL;
823 			pargs_drop(p->p_args);
824 			p->p_args = NULL;
825 			sigacts_free(p->p_sigacts);
826 			p->p_sigacts = NULL;
827 
828 			/*
829 			 * Do any thread-system specific cleanups.
830 			 */
831 			thread_wait(p);
832 
833 			/*
834 			 * Give vm and machine-dependent layer a chance
835 			 * to free anything that cpu_exit couldn't
836 			 * release while still running in process context.
837 			 */
838 			vm_waitproc(p);
839 #ifdef MAC
840 			mac_destroy_proc(p);
841 #endif
842 #ifdef AUDIT
843 			audit_proc_free(p);
844 #endif
845 			KASSERT(FIRST_THREAD_IN_PROC(p),
846 			    ("kern_wait: no residual thread!"));
847 			uma_zfree(proc_zone, p);
848 			sx_xlock(&allproc_lock);
849 			nprocs--;
850 			sx_xunlock(&allproc_lock);
851 			return (0);
852 		}
853 		mtx_lock_spin(&sched_lock);
854 		if ((p->p_flag & P_STOPPED_SIG) &&
855 		    (p->p_suspcount == p->p_numthreads) &&
856 		    (p->p_flag & P_WAITED) == 0 &&
857 		    (p->p_flag & P_TRACED || options & WUNTRACED)) {
858 			mtx_unlock_spin(&sched_lock);
859 			p->p_flag |= P_WAITED;
860 			sx_xunlock(&proctree_lock);
861 			td->td_retval[0] = p->p_pid;
862 			if (status)
863 				*status = W_STOPCODE(p->p_xstat);
864 			PROC_UNLOCK(p);
865 
866 			PROC_LOCK(q);
867 			sigqueue_take(p->p_ksi);
868 			PROC_UNLOCK(q);
869 
870 			return (0);
871 		}
872 		mtx_unlock_spin(&sched_lock);
873 		if (options & WCONTINUED && (p->p_flag & P_CONTINUED)) {
874 			sx_xunlock(&proctree_lock);
875 			td->td_retval[0] = p->p_pid;
876 			p->p_flag &= ~P_CONTINUED;
877 			PROC_UNLOCK(p);
878 
879 			PROC_LOCK(q);
880 			sigqueue_take(p->p_ksi);
881 			PROC_UNLOCK(q);
882 
883 			if (status)
884 				*status = SIGCONT;
885 			return (0);
886 		}
887 		PROC_UNLOCK(p);
888 	}
889 	if (nfound == 0) {
890 		sx_xunlock(&proctree_lock);
891 		return (ECHILD);
892 	}
893 	if (options & WNOHANG) {
894 		sx_xunlock(&proctree_lock);
895 		td->td_retval[0] = 0;
896 		return (0);
897 	}
898 	PROC_LOCK(q);
899 	sx_xunlock(&proctree_lock);
900 	if (q->p_flag & P_STATCHILD) {
901 		q->p_flag &= ~P_STATCHILD;
902 		error = 0;
903 	} else
904 		error = msleep(q, &q->p_mtx, PWAIT | PCATCH, "wait", 0);
905 	PROC_UNLOCK(q);
906 	if (error)
907 		return (error);
908 	goto loop;
909 }
910 
911 /*
912  * Make process 'parent' the new parent of process 'child'.
913  * Must be called with an exclusive hold of proctree lock.
914  */
915 void
916 proc_reparent(struct proc *child, struct proc *parent)
917 {
918 
919 	sx_assert(&proctree_lock, SX_XLOCKED);
920 	PROC_LOCK_ASSERT(child, MA_OWNED);
921 	if (child->p_pptr == parent)
922 		return;
923 
924 	LIST_REMOVE(child, p_sibling);
925 	LIST_INSERT_HEAD(&parent->p_children, child, p_sibling);
926 	child->p_pptr = parent;
927 }
928