xref: /freebsd/sys/kern/kern_exit.c (revision 87569f75a91f298c52a71823c04d41cf53c88889)
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 vmspace *vm;
117 	struct vnode *vtmp;
118 #ifdef KTRACE
119 	struct vnode *tracevp;
120 	struct ucred *tracecred;
121 #endif
122 	struct plimit *plim;
123 	int locked, refcnt;
124 
125 	/*
126 	 * Drop Giant if caller has it.  Eventually we should warn about
127 	 * being called with Giant held.
128 	 */
129 	while (mtx_owned(&Giant))
130 		mtx_unlock(&Giant);
131 
132 	p = td->td_proc;
133 	if (p == initproc) {
134 		printf("init died (signal %d, exit %d)\n",
135 		    WTERMSIG(rv), WEXITSTATUS(rv));
136 		panic("Going nowhere without my init!");
137 	}
138 
139 	/*
140 	 * MUST abort all other threads before proceeding past here.
141 	 */
142 	PROC_LOCK(p);
143 	if (p->p_flag & P_HADTHREADS) {
144 retry:
145 		/*
146 		 * First check if some other thread got here before us..
147 		 * if so, act apropriatly, (exit or suspend);
148 		 */
149 		thread_suspend_check(0);
150 
151 		/*
152 		 * Kill off the other threads. This requires
153 		 * some co-operation from other parts of the kernel
154 		 * so it may not be instantaneous.  With this state set
155 		 * any thread entering the kernel from userspace will
156 		 * thread_exit() in trap().  Any thread attempting to
157 		 * sleep will return immediately with EINTR or EWOULDBLOCK
158 		 * which will hopefully force them to back out to userland
159 		 * freeing resources as they go.  Any thread attempting
160 		 * to return to userland will thread_exit() from userret().
161 		 * thread_exit() will unsuspend us when the last of the
162 		 * other threads exits.
163 		 * If there is already a thread singler after resumption,
164 		 * calling thread_single will fail; in that case, we just
165 		 * re-check all suspension request, the thread should
166 		 * either be suspended there or exit.
167 		 */
168 		if (thread_single(SINGLE_EXIT))
169 			goto retry;
170 
171 		/*
172 		 * All other activity in this process is now stopped.
173 		 * Threading support has been turned off.
174 		 */
175 	}
176 
177 	/*
178 	 * Wakeup anyone in procfs' PIOCWAIT.  They should have a hold
179 	 * on our vmspace, so we should block below until they have
180 	 * released their reference to us.  Note that if they have
181 	 * requested S_EXIT stops we will block here until they ack
182 	 * via PIOCCONT.
183 	 */
184 	_STOPEVENT(p, S_EXIT, rv);
185 
186 	/*
187 	 * Note that we are exiting and do another wakeup of anyone in
188 	 * PIOCWAIT in case they aren't listening for S_EXIT stops or
189 	 * decided to wait again after we told them we are exiting.
190 	 */
191 	p->p_flag |= P_WEXIT;
192 	wakeup(&p->p_stype);
193 
194 	/*
195 	 * Wait for any processes that have a hold on our vmspace to
196 	 * release their reference.
197 	 */
198 	while (p->p_lock > 0)
199 		msleep(&p->p_lock, &p->p_mtx, PWAIT, "exithold", 0);
200 
201 	PROC_LOCK(p->p_pptr);
202 	sigqueue_take(p->p_ksi);
203 	PROC_UNLOCK(p->p_pptr);
204 
205 	PROC_UNLOCK(p);
206 
207 #ifdef AUDIT
208 	/*
209 	 * The Sun BSM exit token contains two components: an exit status as
210 	 * passed to exit(), and a return value to indicate what sort of exit
211 	 * it was.  The exit status is WEXITSTATUS(rv), but it's not clear
212 	 * what the return value is.
213 	 */
214 	AUDIT_ARG(exit, WEXITSTATUS(rv), 0);
215 	AUDIT_SYSCALL_EXIT(0, td);
216 #endif
217 
218 	/* Are we a task leader? */
219 	if (p == p->p_leader) {
220 		mtx_lock(&ppeers_lock);
221 		q = p->p_peers;
222 		while (q != NULL) {
223 			PROC_LOCK(q);
224 			psignal(q, SIGKILL);
225 			PROC_UNLOCK(q);
226 			q = q->p_peers;
227 		}
228 		while (p->p_peers != NULL)
229 			msleep(p, &ppeers_lock, PWAIT, "exit1", 0);
230 		mtx_unlock(&ppeers_lock);
231 	}
232 
233 	/*
234 	 * Check if any loadable modules need anything done at process exit.
235 	 * E.g. SYSV IPC stuff
236 	 * XXX what if one of these generates an error?
237 	 */
238 	EVENTHANDLER_INVOKE(process_exit, p);
239 
240 	MALLOC(p->p_ru, struct rusage *, sizeof(struct rusage),
241 		M_ZOMBIE, M_WAITOK);
242 	/*
243 	 * If parent is waiting for us to exit or exec,
244 	 * P_PPWAIT is set; we will wakeup the parent below.
245 	 */
246 	PROC_LOCK(p);
247 	stopprofclock(p);
248 	p->p_flag &= ~(P_TRACED | P_PPWAIT);
249 
250 	/*
251 	 * Stop the real interval timer.  If the handler is currently
252 	 * executing, prevent it from rearming itself and let it finish.
253 	 */
254 	if (timevalisset(&p->p_realtimer.it_value) &&
255 	    callout_stop(&p->p_itcallout) == 0) {
256 		timevalclear(&p->p_realtimer.it_interval);
257 		msleep(&p->p_itcallout, &p->p_mtx, PWAIT, "ritwait", 0);
258 		KASSERT(!timevalisset(&p->p_realtimer.it_value),
259 		    ("realtime timer is still armed"));
260 	}
261 	sigqueue_flush(&p->p_sigqueue);
262 	sigqueue_flush(&td->td_sigqueue);
263 	PROC_UNLOCK(p);
264 
265 	/*
266 	 * Reset any sigio structures pointing to us as a result of
267 	 * F_SETOWN with our pid.
268 	 */
269 	mtx_lock(&Giant);	/* XXX: not sure if needed */
270 	funsetownlst(&p->p_sigiolst);
271 	mtx_unlock(&Giant);
272 
273 	/*
274 	 * If this process has an nlminfo data area (for lockd), release it
275 	 */
276 	if (nlminfo_release_p != NULL && p->p_nlminfo != NULL)
277 		(*nlminfo_release_p)(p);
278 
279 	/*
280 	 * Close open files and release open-file table.
281 	 * This may block!
282 	 */
283 	fdfree(td);
284 
285 	/*
286 	 * If this thread tickled GEOM, we need to wait for the giggling to
287 	 * stop before we return to userland
288 	 */
289 	if (td->td_pflags & TDP_GEOM)
290 		g_waitidle();
291 
292 	/*
293 	 * Remove ourself from our leader's peer list and wake our leader.
294 	 */
295 	mtx_lock(&ppeers_lock);
296 	if (p->p_leader->p_peers) {
297 		q = p->p_leader;
298 		while (q->p_peers != p)
299 			q = q->p_peers;
300 		q->p_peers = p->p_peers;
301 		wakeup(p->p_leader);
302 	}
303 	mtx_unlock(&ppeers_lock);
304 
305 	/* The next two chunks should probably be moved to vmspace_exit. */
306 	vm = p->p_vmspace;
307 	/*
308 	 * Release user portion of address space.
309 	 * This releases references to vnodes,
310 	 * which could cause I/O if the file has been unlinked.
311 	 * Need to do this early enough that we can still sleep.
312 	 * Can't free the entire vmspace as the kernel stack
313 	 * may be mapped within that space also.
314 	 *
315 	 * Processes sharing the same vmspace may exit in one order, and
316 	 * get cleaned up by vmspace_exit() in a different order.  The
317 	 * last exiting process to reach this point releases as much of
318 	 * the environment as it can, and the last process cleaned up
319 	 * by vmspace_exit() (which decrements exitingcnt) cleans up the
320 	 * remainder.
321 	 */
322 	atomic_add_int(&vm->vm_exitingcnt, 1);
323 	do
324 		refcnt = vm->vm_refcnt;
325 	while (!atomic_cmpset_int(&vm->vm_refcnt, refcnt, refcnt - 1));
326 	if (refcnt == 1) {
327 		shmexit(vm);
328 		pmap_remove_pages(vmspace_pmap(vm), vm_map_min(&vm->vm_map),
329 		    vm_map_max(&vm->vm_map));
330 		(void) vm_map_remove(&vm->vm_map, vm_map_min(&vm->vm_map),
331 		    vm_map_max(&vm->vm_map));
332 	}
333 
334 	sx_xlock(&proctree_lock);
335 	if (SESS_LEADER(p)) {
336 		struct session *sp;
337 
338 		sp = p->p_session;
339 		if (sp->s_ttyvp) {
340 			locked = VFS_LOCK_GIANT(sp->s_ttyvp->v_mount);
341 			/*
342 			 * Controlling process.
343 			 * Signal foreground pgrp,
344 			 * drain controlling terminal
345 			 * and revoke access to controlling terminal.
346 			 */
347 			if (sp->s_ttyp && (sp->s_ttyp->t_session == sp)) {
348 				tp = sp->s_ttyp;
349 				if (sp->s_ttyp->t_pgrp) {
350 					PGRP_LOCK(sp->s_ttyp->t_pgrp);
351 					pgsignal(sp->s_ttyp->t_pgrp, SIGHUP, 1);
352 					PGRP_UNLOCK(sp->s_ttyp->t_pgrp);
353 				}
354 				/* XXX tp should be locked. */
355 				sx_xunlock(&proctree_lock);
356 				(void) ttywait(tp);
357 				sx_xlock(&proctree_lock);
358 				/*
359 				 * The tty could have been revoked
360 				 * if we blocked.
361 				 */
362 				if (sp->s_ttyvp) {
363 					ttyvp = sp->s_ttyvp;
364 					SESS_LOCK(p->p_session);
365 					sp->s_ttyvp = NULL;
366 					SESS_UNLOCK(p->p_session);
367 					sx_xunlock(&proctree_lock);
368 					VOP_LOCK(ttyvp, LK_EXCLUSIVE, td);
369 					VOP_REVOKE(ttyvp, REVOKEALL);
370 					vput(ttyvp);
371 					sx_xlock(&proctree_lock);
372 				}
373 			}
374 			if (sp->s_ttyvp) {
375 				ttyvp = sp->s_ttyvp;
376 				SESS_LOCK(p->p_session);
377 				sp->s_ttyvp = NULL;
378 				SESS_UNLOCK(p->p_session);
379 				vrele(ttyvp);
380 			}
381 			/*
382 			 * s_ttyp is not zero'd; we use this to indicate
383 			 * that the session once had a controlling terminal.
384 			 * (for logging and informational purposes)
385 			 */
386 			VFS_UNLOCK_GIANT(locked);
387 		}
388 		SESS_LOCK(p->p_session);
389 		sp->s_leader = NULL;
390 		SESS_UNLOCK(p->p_session);
391 	}
392 	fixjobc(p, p->p_pgrp, 0);
393 	sx_xunlock(&proctree_lock);
394 	(void)acct_process(td);
395 #ifdef KTRACE
396 	/*
397 	 * Drain any pending records on the thread and release the trace
398 	 * file.  It might be better if drain-and-clear were atomic.
399 	 */
400 	ktrprocexit(td);
401 	PROC_LOCK(p);
402 	mtx_lock(&ktrace_mtx);
403 	p->p_traceflag = 0;	/* don't trace the vrele() */
404 	tracevp = p->p_tracevp;
405 	p->p_tracevp = NULL;
406 	tracecred = p->p_tracecred;
407 	p->p_tracecred = NULL;
408 	mtx_unlock(&ktrace_mtx);
409 	PROC_UNLOCK(p);
410 	if (tracevp != NULL) {
411 		locked = VFS_LOCK_GIANT(tracevp->v_mount);
412 		vrele(tracevp);
413 		VFS_UNLOCK_GIANT(locked);
414 	}
415 	if (tracecred != NULL)
416 		crfree(tracecred);
417 #endif
418 	/*
419 	 * Release reference to text vnode
420 	 */
421 	if ((vtmp = p->p_textvp) != NULL) {
422 		p->p_textvp = NULL;
423 		locked = VFS_LOCK_GIANT(vtmp->v_mount);
424 		vrele(vtmp);
425 		VFS_UNLOCK_GIANT(locked);
426 	}
427 
428 	/*
429 	 * Release our limits structure.
430 	 */
431 	PROC_LOCK(p);
432 	plim = p->p_limit;
433 	p->p_limit = NULL;
434 	PROC_UNLOCK(p);
435 	lim_free(plim);
436 
437 	/*
438 	 * Remove proc from allproc queue and pidhash chain.
439 	 * Place onto zombproc.  Unlink from parent's child list.
440 	 */
441 	sx_xlock(&allproc_lock);
442 	LIST_REMOVE(p, p_list);
443 	LIST_INSERT_HEAD(&zombproc, p, p_list);
444 	LIST_REMOVE(p, p_hash);
445 	sx_xunlock(&allproc_lock);
446 
447 	/*
448 	 * Reparent all of our children to init.
449 	 */
450 	sx_xlock(&proctree_lock);
451 	q = LIST_FIRST(&p->p_children);
452 	if (q != NULL)		/* only need this if any child is S_ZOMB */
453 		wakeup(initproc);
454 	for (; q != NULL; q = nq) {
455 		nq = LIST_NEXT(q, p_sibling);
456 		PROC_LOCK(q);
457 		proc_reparent(q, initproc);
458 		q->p_sigparent = SIGCHLD;
459 		/*
460 		 * Traced processes are killed
461 		 * since their existence means someone is screwing up.
462 		 */
463 		if (q->p_flag & P_TRACED) {
464 			q->p_flag &= ~(P_TRACED | P_STOPPED_TRACE);
465 			psignal(q, SIGKILL);
466 		}
467 		PROC_UNLOCK(q);
468 	}
469 
470 	/*
471 	 * Save exit status and finalize rusage info except for times,
472 	 * adding in child rusage info later when our time is locked.
473 	 */
474 	PROC_LOCK(p);
475 	p->p_xstat = rv;
476 	p->p_xthread = td;
477 	p->p_stats->p_ru.ru_nvcsw++;
478 	*p->p_ru = p->p_stats->p_ru;
479 
480 	/*
481 	 * Notify interested parties of our demise.
482 	 */
483 	KNOTE_LOCKED(&p->p_klist, NOTE_EXIT);
484 
485 	/*
486 	 * Just delete all entries in the p_klist. At this point we won't
487 	 * report any more events, and there are nasty race conditions that
488 	 * can beat us if we don't.
489 	 */
490 	knlist_clear(&p->p_klist, 1);
491 
492 	/*
493 	 * Notify parent that we're gone.  If parent has the PS_NOCLDWAIT
494 	 * flag set, or if the handler is set to SIG_IGN, notify process
495 	 * 1 instead (and hope it will handle this situation).
496 	 */
497 	PROC_LOCK(p->p_pptr);
498 	mtx_lock(&p->p_pptr->p_sigacts->ps_mtx);
499 	if (p->p_pptr->p_sigacts->ps_flag & (PS_NOCLDWAIT | PS_CLDSIGIGN)) {
500 		struct proc *pp;
501 
502 		mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx);
503 		pp = p->p_pptr;
504 		PROC_UNLOCK(pp);
505 		proc_reparent(p, initproc);
506 		p->p_sigparent = SIGCHLD;
507 		PROC_LOCK(p->p_pptr);
508 		/*
509 		 * If this was the last child of our parent, notify
510 		 * parent, so in case he was wait(2)ing, he will
511 		 * continue.
512 		 */
513 		if (LIST_EMPTY(&pp->p_children))
514 			wakeup(pp);
515 	} else
516 		mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx);
517 
518 	if (p->p_pptr == initproc)
519 		psignal(p->p_pptr, SIGCHLD);
520 	else if (p->p_sigparent != 0) {
521 		if (p->p_sigparent == SIGCHLD)
522 			childproc_exited(p);
523 		else	/* LINUX thread */
524 			psignal(p->p_pptr, p->p_sigparent);
525 	}
526 	PROC_UNLOCK(p->p_pptr);
527 	PROC_UNLOCK(p);
528 
529 	/*
530 	 * Finally, call machine-dependent code to release the remaining
531 	 * resources including address space.
532 	 * The address space is released by "vmspace_exitfree(p)" in
533 	 * vm_waitproc().
534 	 */
535 	cpu_exit(td);
536 
537 	WITNESS_WARN(WARN_PANIC, &proctree_lock.sx_object,
538 	    "process (pid %d) exiting", p->p_pid);
539 
540 	PROC_LOCK(p);
541 	PROC_LOCK(p->p_pptr);
542 	sx_xunlock(&proctree_lock);
543 
544 	/*
545 	 * We have to wait until after acquiring all locks before
546 	 * changing p_state.  We need to avoid all possible context
547 	 * switches (including ones from blocking on a mutex) while
548 	 * marked as a zombie.  We also have to set the zombie state
549 	 * before we release the parent process' proc lock to avoid
550 	 * a lost wakeup.  So, we first call wakeup, then we grab the
551 	 * sched lock, update the state, and release the parent process'
552 	 * proc lock.
553 	 */
554 	wakeup(p->p_pptr);
555 	mtx_lock_spin(&sched_lock);
556 	p->p_state = PRS_ZOMBIE;
557 	PROC_UNLOCK(p->p_pptr);
558 
559 	sched_exit(p->p_pptr, td);
560 
561 	/*
562 	 * Hopefully no one will try to deliver a signal to the process this
563 	 * late in the game.
564 	 */
565 	knlist_destroy(&p->p_klist);
566 
567 	/*
568 	 * Make sure the scheduler takes this thread out of its tables etc.
569 	 * This will also release this thread's reference to the ucred.
570 	 * Other thread parts to release include pcb bits and such.
571 	 */
572 	thread_exit();
573 }
574 
575 
576 #ifndef _SYS_SYSPROTO_H_
577 struct abort2_args {
578 	char *why;
579 	int nargs;
580 	void **args;
581 };
582 #endif
583 
584 /*
585  * MPSAFE.
586  */
587 int
588 abort2(struct thread *td, struct abort2_args *uap)
589 {
590 	struct proc *p = td->td_proc;
591 	struct sbuf *sb;
592 	void *uargs[16];
593 	int error, i, sig;
594 
595 	error = 0;	/* satisfy compiler */
596 
597 	/*
598 	 * Do it right now so we can log either proper call of abort2(), or
599 	 * note, that invalid argument was passed. 512 is big enough to
600 	 * handle 16 arguments' descriptions with additional comments.
601 	 */
602 	sb = sbuf_new(NULL, NULL, 512, SBUF_FIXEDLEN);
603 	sbuf_clear(sb);
604 	sbuf_printf(sb, "%s(pid %d uid %d) aborted: ",
605 	    p->p_comm, p->p_pid, td->td_ucred->cr_uid);
606 	/*
607 	 * Since we can't return from abort2(), send SIGKILL in cases, where
608 	 * abort2() was called improperly
609 	 */
610 	sig = SIGKILL;
611 	/* Prevent from DoSes from user-space. */
612 	if (uap->nargs < 0 || uap->nargs > 16)
613 		goto out;
614 	if (uap->args == NULL)
615 		goto out;
616 	error = copyin(uap->args, uargs, uap->nargs * sizeof(void *));
617 	if (error != 0)
618 		goto out;
619 	/*
620 	 * Limit size of 'reason' string to 128. Will fit even when
621 	 * maximal number of arguments was chosen to be logged.
622 	 */
623 	if (uap->why != NULL) {
624 		error = sbuf_copyin(sb, uap->why, 128);
625 		if (error < 0)
626 			goto out;
627 	} else {
628 		sbuf_printf(sb, "(null)");
629 	}
630 	if (uap->nargs) {
631 		sbuf_printf(sb, "(");
632 		for (i = 0;i < uap->nargs; i++)
633 			sbuf_printf(sb, "%s%p", i == 0 ? "" : ", ", uargs[i]);
634 		sbuf_printf(sb, ")");
635 	}
636 	/*
637 	 * Final stage: arguments were proper, string has been
638 	 * successfully copied from userspace, and copying pointers
639 	 * from user-space succeed.
640 	 */
641 	sig = SIGABRT;
642 out:
643 	if (sig == SIGKILL) {
644 		sbuf_trim(sb);
645 		sbuf_printf(sb, " (Reason text inaccessible)");
646 	}
647 	sbuf_cat(sb, "\n");
648 	sbuf_finish(sb);
649 	log(LOG_INFO, "%s", sbuf_data(sb));
650 	sbuf_delete(sb);
651 	exit1(td, W_EXITCODE(0, sig));
652 	return (0);
653 }
654 
655 
656 #ifdef COMPAT_43
657 /*
658  * The dirty work is handled by kern_wait().
659  *
660  * MPSAFE.
661  */
662 int
663 owait(struct thread *td, struct owait_args *uap __unused)
664 {
665 	int error, status;
666 
667 	error = kern_wait(td, WAIT_ANY, &status, 0, NULL);
668 	if (error == 0)
669 		td->td_retval[1] = status;
670 	return (error);
671 }
672 #endif /* COMPAT_43 */
673 
674 /*
675  * The dirty work is handled by kern_wait().
676  *
677  * MPSAFE.
678  */
679 int
680 wait4(struct thread *td, struct wait_args *uap)
681 {
682 	struct rusage ru, *rup;
683 	int error, status;
684 
685 	if (uap->rusage != NULL)
686 		rup = &ru;
687 	else
688 		rup = NULL;
689 	error = kern_wait(td, uap->pid, &status, uap->options, rup);
690 	if (uap->status != NULL && error == 0)
691 		error = copyout(&status, uap->status, sizeof(status));
692 	if (uap->rusage != NULL && error == 0)
693 		error = copyout(&ru, uap->rusage, sizeof(struct rusage));
694 	return (error);
695 }
696 
697 int
698 kern_wait(struct thread *td, pid_t pid, int *status, int options,
699     struct rusage *rusage)
700 {
701 	struct proc *p, *q, *t;
702 	int error, nfound;
703 
704 	AUDIT_ARG(pid, pid);
705 
706 	q = td->td_proc;
707 	if (pid == 0) {
708 		PROC_LOCK(q);
709 		pid = -q->p_pgid;
710 		PROC_UNLOCK(q);
711 	}
712 	if (options &~ (WUNTRACED|WNOHANG|WCONTINUED|WLINUXCLONE))
713 		return (EINVAL);
714 loop:
715 	if (q->p_flag & P_STATCHILD) {
716 		PROC_LOCK(q);
717 		q->p_flag &= ~P_STATCHILD;
718 		PROC_UNLOCK(q);
719 	}
720 	nfound = 0;
721 	sx_xlock(&proctree_lock);
722 	LIST_FOREACH(p, &q->p_children, p_sibling) {
723 		PROC_LOCK(p);
724 		if (pid != WAIT_ANY &&
725 		    p->p_pid != pid && p->p_pgid != -pid) {
726 			PROC_UNLOCK(p);
727 			continue;
728 		}
729 		if (p_canwait(td, p)) {
730 			PROC_UNLOCK(p);
731 			continue;
732 		}
733 
734 		/*
735 		 * This special case handles a kthread spawned by linux_clone
736 		 * (see linux_misc.c).  The linux_wait4 and linux_waitpid
737 		 * functions need to be able to distinguish between waiting
738 		 * on a process and waiting on a thread.  It is a thread if
739 		 * p_sigparent is not SIGCHLD, and the WLINUXCLONE option
740 		 * signifies we want to wait for threads and not processes.
741 		 */
742 		if ((p->p_sigparent != SIGCHLD) ^
743 		    ((options & WLINUXCLONE) != 0)) {
744 			PROC_UNLOCK(p);
745 			continue;
746 		}
747 
748 		nfound++;
749 		if (p->p_state == PRS_ZOMBIE) {
750 
751 			/*
752 			 * It is possible that the last thread of this
753 			 * process is still running on another CPU
754 			 * in thread_exit() after having dropped the process
755 			 * lock via PROC_UNLOCK() but before it has completed
756 			 * cpu_throw().  In that case, the other thread must
757 			 * still hold sched_lock, so simply by acquiring
758 			 * sched_lock once we will wait long enough for the
759 			 * thread to exit in that case.
760 			 */
761 			mtx_lock_spin(&sched_lock);
762 			mtx_unlock_spin(&sched_lock);
763 
764 			td->td_retval[0] = p->p_pid;
765 			if (status)
766 				*status = p->p_xstat;	/* convert to int */
767 			if (rusage) {
768 				*rusage = *p->p_ru;
769 				calcru(p, &rusage->ru_utime, &rusage->ru_stime);
770 			}
771 
772 			PROC_LOCK(q);
773 			sigqueue_take(p->p_ksi);
774 			PROC_UNLOCK(q);
775 
776 			/*
777 			 * If we got the child via a ptrace 'attach',
778 			 * we need to give it back to the old parent.
779 			 */
780 			PROC_UNLOCK(p);
781 			if (p->p_oppid && (t = pfind(p->p_oppid)) != NULL) {
782 				PROC_LOCK(p);
783 				p->p_oppid = 0;
784 				proc_reparent(p, t);
785 				PROC_UNLOCK(p);
786 				tdsignal(t, NULL, SIGCHLD, p->p_ksi);
787 				wakeup(t);
788 				PROC_UNLOCK(t);
789 				sx_xunlock(&proctree_lock);
790 				return (0);
791 			}
792 
793 			/*
794 			 * Remove other references to this process to ensure
795 			 * we have an exclusive reference.
796 			 */
797 			sx_xlock(&allproc_lock);
798 			LIST_REMOVE(p, p_list);	/* off zombproc */
799 			sx_xunlock(&allproc_lock);
800 			LIST_REMOVE(p, p_sibling);
801 			leavepgrp(p);
802 			sx_xunlock(&proctree_lock);
803 
804 			/*
805 			 * As a side effect of this lock, we know that
806 			 * all other writes to this proc are visible now, so
807 			 * no more locking is needed for p.
808 			 */
809 			PROC_LOCK(p);
810 			p->p_xstat = 0;		/* XXX: why? */
811 			PROC_UNLOCK(p);
812 			PROC_LOCK(q);
813 			ruadd(&q->p_stats->p_cru, &q->p_crux, p->p_ru,
814 			    &p->p_rux);
815 			PROC_UNLOCK(q);
816 			FREE(p->p_ru, M_ZOMBIE);
817 			p->p_ru = NULL;
818 
819 			/*
820 			 * Decrement the count of procs running with this uid.
821 			 */
822 			(void)chgproccnt(p->p_ucred->cr_ruidinfo, -1, 0);
823 
824 			/*
825 			 * Free credentials, arguments, and sigacts.
826 			 */
827 			crfree(p->p_ucred);
828 			p->p_ucred = NULL;
829 			pargs_drop(p->p_args);
830 			p->p_args = NULL;
831 			sigacts_free(p->p_sigacts);
832 			p->p_sigacts = NULL;
833 
834 			/*
835 			 * Do any thread-system specific cleanups.
836 			 */
837 			thread_wait(p);
838 
839 			/*
840 			 * Give vm and machine-dependent layer a chance
841 			 * to free anything that cpu_exit couldn't
842 			 * release while still running in process context.
843 			 */
844 			vm_waitproc(p);
845 #ifdef MAC
846 			mac_destroy_proc(p);
847 #endif
848 #ifdef AUDIT
849 			audit_proc_free(p);
850 #endif
851 			KASSERT(FIRST_THREAD_IN_PROC(p),
852 			    ("kern_wait: no residual thread!"));
853 			uma_zfree(proc_zone, p);
854 			sx_xlock(&allproc_lock);
855 			nprocs--;
856 			sx_xunlock(&allproc_lock);
857 			return (0);
858 		}
859 		mtx_lock_spin(&sched_lock);
860 		if ((p->p_flag & P_STOPPED_SIG) &&
861 		    (p->p_suspcount == p->p_numthreads) &&
862 		    (p->p_flag & P_WAITED) == 0 &&
863 		    (p->p_flag & P_TRACED || options & WUNTRACED)) {
864 			mtx_unlock_spin(&sched_lock);
865 			p->p_flag |= P_WAITED;
866 			sx_xunlock(&proctree_lock);
867 			td->td_retval[0] = p->p_pid;
868 			if (status)
869 				*status = W_STOPCODE(p->p_xstat);
870 			PROC_UNLOCK(p);
871 
872 			PROC_LOCK(q);
873 			sigqueue_take(p->p_ksi);
874 			PROC_UNLOCK(q);
875 
876 			return (0);
877 		}
878 		mtx_unlock_spin(&sched_lock);
879 		if (options & WCONTINUED && (p->p_flag & P_CONTINUED)) {
880 			sx_xunlock(&proctree_lock);
881 			td->td_retval[0] = p->p_pid;
882 			p->p_flag &= ~P_CONTINUED;
883 			PROC_UNLOCK(p);
884 
885 			PROC_LOCK(q);
886 			sigqueue_take(p->p_ksi);
887 			PROC_UNLOCK(q);
888 
889 			if (status)
890 				*status = SIGCONT;
891 			return (0);
892 		}
893 		PROC_UNLOCK(p);
894 	}
895 	if (nfound == 0) {
896 		sx_xunlock(&proctree_lock);
897 		return (ECHILD);
898 	}
899 	if (options & WNOHANG) {
900 		sx_xunlock(&proctree_lock);
901 		td->td_retval[0] = 0;
902 		return (0);
903 	}
904 	PROC_LOCK(q);
905 	sx_xunlock(&proctree_lock);
906 	if (q->p_flag & P_STATCHILD) {
907 		q->p_flag &= ~P_STATCHILD;
908 		error = 0;
909 	} else
910 		error = msleep(q, &q->p_mtx, PWAIT | PCATCH, "wait", 0);
911 	PROC_UNLOCK(q);
912 	if (error)
913 		return (error);
914 	goto loop;
915 }
916 
917 /*
918  * Make process 'parent' the new parent of process 'child'.
919  * Must be called with an exclusive hold of proctree lock.
920  */
921 void
922 proc_reparent(struct proc *child, struct proc *parent)
923 {
924 
925 	sx_assert(&proctree_lock, SX_XLOCKED);
926 	PROC_LOCK_ASSERT(child, MA_OWNED);
927 	if (child->p_pptr == parent)
928 		return;
929 
930 	LIST_REMOVE(child, p_sibling);
931 	LIST_INSERT_HEAD(&parent->p_children, child, p_sibling);
932 	child->p_pptr = parent;
933 }
934