xref: /freebsd/sys/kern/kern_exit.c (revision 3b3a8eb937bf8045231e8364bfd1b94cd4a95979)
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_kdtrace.h"
42 #include "opt_ktrace.h"
43 #include "opt_procdesc.h"
44 
45 #include <sys/param.h>
46 #include <sys/systm.h>
47 #include <sys/sysproto.h>
48 #include <sys/capability.h>
49 #include <sys/eventhandler.h>
50 #include <sys/kernel.h>
51 #include <sys/malloc.h>
52 #include <sys/lock.h>
53 #include <sys/mutex.h>
54 #include <sys/proc.h>
55 #include <sys/procdesc.h>
56 #include <sys/pioctl.h>
57 #include <sys/jail.h>
58 #include <sys/tty.h>
59 #include <sys/wait.h>
60 #include <sys/vmmeter.h>
61 #include <sys/vnode.h>
62 #include <sys/racct.h>
63 #include <sys/resourcevar.h>
64 #include <sys/sbuf.h>
65 #include <sys/signalvar.h>
66 #include <sys/sched.h>
67 #include <sys/sx.h>
68 #include <sys/syscallsubr.h>
69 #include <sys/syslog.h>
70 #include <sys/ptrace.h>
71 #include <sys/acct.h>		/* for acct_process() function prototype */
72 #include <sys/filedesc.h>
73 #include <sys/sdt.h>
74 #include <sys/shm.h>
75 #include <sys/sem.h>
76 #ifdef KTRACE
77 #include <sys/ktrace.h>
78 #endif
79 
80 #include <security/audit/audit.h>
81 #include <security/mac/mac_framework.h>
82 
83 #include <vm/vm.h>
84 #include <vm/vm_extern.h>
85 #include <vm/vm_param.h>
86 #include <vm/pmap.h>
87 #include <vm/vm_map.h>
88 #include <vm/vm_page.h>
89 #include <vm/uma.h>
90 
91 #ifdef KDTRACE_HOOKS
92 #include <sys/dtrace_bsd.h>
93 dtrace_execexit_func_t	dtrace_fasttrap_exit;
94 #endif
95 
96 SDT_PROVIDER_DECLARE(proc);
97 SDT_PROBE_DEFINE(proc, kernel, , exit, exit);
98 SDT_PROBE_ARGTYPE(proc, kernel, , exit, 0, "int");
99 
100 /* Hook for NFS teardown procedure. */
101 void (*nlminfo_release_p)(struct proc *p);
102 
103 static void
104 clear_orphan(struct proc *p)
105 {
106 
107 	PROC_LOCK_ASSERT(p, MA_OWNED);
108 
109 	if (p->p_flag & P_ORPHAN) {
110 		LIST_REMOVE(p, p_orphan);
111 		p->p_flag &= ~P_ORPHAN;
112 	}
113 }
114 
115 /*
116  * exit -- death of process.
117  */
118 void
119 sys_sys_exit(struct thread *td, struct sys_exit_args *uap)
120 {
121 
122 	exit1(td, W_EXITCODE(uap->rval, 0));
123 	/* NOTREACHED */
124 }
125 
126 /*
127  * Exit: deallocate address space and other resources, change proc state to
128  * zombie, and unlink proc from allproc and parent's lists.  Save exit status
129  * and rusage for wait().  Check for child processes and orphan them.
130  */
131 void
132 exit1(struct thread *td, int rv)
133 {
134 	struct proc *p, *nq, *q;
135 	struct vnode *vtmp;
136 	struct vnode *ttyvp = NULL;
137 	struct plimit *plim;
138 	int locked;
139 
140 	mtx_assert(&Giant, MA_NOTOWNED);
141 
142 	p = td->td_proc;
143 	/*
144 	 * XXX in case we're rebooting we just let init die in order to
145 	 * work around an unsolved stack overflow seen very late during
146 	 * shutdown on sparc64 when the gmirror worker process exists.
147 	 */
148 	if (p == initproc && rebooting == 0) {
149 		printf("init died (signal %d, exit %d)\n",
150 		    WTERMSIG(rv), WEXITSTATUS(rv));
151 		panic("Going nowhere without my init!");
152 	}
153 
154 	/*
155 	 * MUST abort all other threads before proceeding past here.
156 	 */
157 	PROC_LOCK(p);
158 	while (p->p_flag & P_HADTHREADS) {
159 		/*
160 		 * First check if some other thread got here before us.
161 		 * If so, act appropriately: exit or suspend.
162 		 */
163 		thread_suspend_check(0);
164 
165 		/*
166 		 * Kill off the other threads. This requires
167 		 * some co-operation from other parts of the kernel
168 		 * so it may not be instantaneous.  With this state set
169 		 * any thread entering the kernel from userspace will
170 		 * thread_exit() in trap().  Any thread attempting to
171 		 * sleep will return immediately with EINTR or EWOULDBLOCK
172 		 * which will hopefully force them to back out to userland
173 		 * freeing resources as they go.  Any thread attempting
174 		 * to return to userland will thread_exit() from userret().
175 		 * thread_exit() will unsuspend us when the last of the
176 		 * other threads exits.
177 		 * If there is already a thread singler after resumption,
178 		 * calling thread_single will fail; in that case, we just
179 		 * re-check all suspension request, the thread should
180 		 * either be suspended there or exit.
181 		 */
182 		if (!thread_single(SINGLE_EXIT))
183 			break;
184 
185 		/*
186 		 * All other activity in this process is now stopped.
187 		 * Threading support has been turned off.
188 		 */
189 	}
190 	KASSERT(p->p_numthreads == 1,
191 	    ("exit1: proc %p exiting with %d threads", p, p->p_numthreads));
192 	racct_sub(p, RACCT_NTHR, 1);
193 	/*
194 	 * Wakeup anyone in procfs' PIOCWAIT.  They should have a hold
195 	 * on our vmspace, so we should block below until they have
196 	 * released their reference to us.  Note that if they have
197 	 * requested S_EXIT stops we will block here until they ack
198 	 * via PIOCCONT.
199 	 */
200 	_STOPEVENT(p, S_EXIT, rv);
201 
202 	/*
203 	 * Ignore any pending request to stop due to a stop signal.
204 	 * Once P_WEXIT is set, future requests will be ignored as
205 	 * well.
206 	 */
207 	p->p_flag &= ~P_STOPPED_SIG;
208 	KASSERT(!P_SHOULDSTOP(p), ("exiting process is stopped"));
209 
210 	/*
211 	 * Note that we are exiting and do another wakeup of anyone in
212 	 * PIOCWAIT in case they aren't listening for S_EXIT stops or
213 	 * decided to wait again after we told them we are exiting.
214 	 */
215 	p->p_flag |= P_WEXIT;
216 	wakeup(&p->p_stype);
217 
218 	/*
219 	 * Wait for any processes that have a hold on our vmspace to
220 	 * release their reference.
221 	 */
222 	while (p->p_lock > 0)
223 		msleep(&p->p_lock, &p->p_mtx, PWAIT, "exithold", 0);
224 
225 	p->p_xstat = rv;	/* Let event handler change exit status */
226 	PROC_UNLOCK(p);
227 	/* Drain the limit callout while we don't have the proc locked */
228 	callout_drain(&p->p_limco);
229 
230 #ifdef AUDIT
231 	/*
232 	 * The Sun BSM exit token contains two components: an exit status as
233 	 * passed to exit(), and a return value to indicate what sort of exit
234 	 * it was.  The exit status is WEXITSTATUS(rv), but it's not clear
235 	 * what the return value is.
236 	 */
237 	AUDIT_ARG_EXIT(WEXITSTATUS(rv), 0);
238 	AUDIT_SYSCALL_EXIT(0, td);
239 #endif
240 
241 	/* Are we a task leader? */
242 	if (p == p->p_leader) {
243 		mtx_lock(&ppeers_lock);
244 		q = p->p_peers;
245 		while (q != NULL) {
246 			PROC_LOCK(q);
247 			kern_psignal(q, SIGKILL);
248 			PROC_UNLOCK(q);
249 			q = q->p_peers;
250 		}
251 		while (p->p_peers != NULL)
252 			msleep(p, &ppeers_lock, PWAIT, "exit1", 0);
253 		mtx_unlock(&ppeers_lock);
254 	}
255 
256 	/*
257 	 * Check if any loadable modules need anything done at process exit.
258 	 * E.g. SYSV IPC stuff
259 	 * XXX what if one of these generates an error?
260 	 */
261 	EVENTHANDLER_INVOKE(process_exit, p);
262 
263 	/*
264 	 * If parent is waiting for us to exit or exec,
265 	 * P_PPWAIT is set; we will wakeup the parent below.
266 	 */
267 	PROC_LOCK(p);
268 	rv = p->p_xstat;	/* Event handler could change exit status */
269 	stopprofclock(p);
270 	p->p_flag &= ~(P_TRACED | P_PPWAIT);
271 
272 	/*
273 	 * Stop the real interval timer.  If the handler is currently
274 	 * executing, prevent it from rearming itself and let it finish.
275 	 */
276 	if (timevalisset(&p->p_realtimer.it_value) &&
277 	    callout_stop(&p->p_itcallout) == 0) {
278 		timevalclear(&p->p_realtimer.it_interval);
279 		msleep(&p->p_itcallout, &p->p_mtx, PWAIT, "ritwait", 0);
280 		KASSERT(!timevalisset(&p->p_realtimer.it_value),
281 		    ("realtime timer is still armed"));
282 	}
283 	PROC_UNLOCK(p);
284 
285 	/*
286 	 * Reset any sigio structures pointing to us as a result of
287 	 * F_SETOWN with our pid.
288 	 */
289 	funsetownlst(&p->p_sigiolst);
290 
291 	/*
292 	 * If this process has an nlminfo data area (for lockd), release it
293 	 */
294 	if (nlminfo_release_p != NULL && p->p_nlminfo != NULL)
295 		(*nlminfo_release_p)(p);
296 
297 	/*
298 	 * Close open files and release open-file table.
299 	 * This may block!
300 	 */
301 	fdfree(td);
302 
303 	/*
304 	 * If this thread tickled GEOM, we need to wait for the giggling to
305 	 * stop before we return to userland
306 	 */
307 	if (td->td_pflags & TDP_GEOM)
308 		g_waitidle();
309 
310 	/*
311 	 * Remove ourself from our leader's peer list and wake our leader.
312 	 */
313 	mtx_lock(&ppeers_lock);
314 	if (p->p_leader->p_peers) {
315 		q = p->p_leader;
316 		while (q->p_peers != p)
317 			q = q->p_peers;
318 		q->p_peers = p->p_peers;
319 		wakeup(p->p_leader);
320 	}
321 	mtx_unlock(&ppeers_lock);
322 
323 	vmspace_exit(td);
324 
325 	sx_xlock(&proctree_lock);
326 	if (SESS_LEADER(p)) {
327 		struct session *sp = p->p_session;
328 		struct tty *tp;
329 
330 		/*
331 		 * s_ttyp is not zero'd; we use this to indicate that
332 		 * the session once had a controlling terminal. (for
333 		 * logging and informational purposes)
334 		 */
335 		SESS_LOCK(sp);
336 		ttyvp = sp->s_ttyvp;
337 		tp = sp->s_ttyp;
338 		sp->s_ttyvp = NULL;
339 		sp->s_ttydp = NULL;
340 		sp->s_leader = NULL;
341 		SESS_UNLOCK(sp);
342 
343 		/*
344 		 * Signal foreground pgrp and revoke access to
345 		 * controlling terminal if it has not been revoked
346 		 * already.
347 		 *
348 		 * Because the TTY may have been revoked in the mean
349 		 * time and could already have a new session associated
350 		 * with it, make sure we don't send a SIGHUP to a
351 		 * foreground process group that does not belong to this
352 		 * session.
353 		 */
354 
355 		if (tp != NULL) {
356 			tty_lock(tp);
357 			if (tp->t_session == sp)
358 				tty_signal_pgrp(tp, SIGHUP);
359 			tty_unlock(tp);
360 		}
361 
362 		if (ttyvp != NULL) {
363 			sx_xunlock(&proctree_lock);
364 			if (vn_lock(ttyvp, LK_EXCLUSIVE) == 0) {
365 				VOP_REVOKE(ttyvp, REVOKEALL);
366 				VOP_UNLOCK(ttyvp, 0);
367 			}
368 			sx_xlock(&proctree_lock);
369 		}
370 	}
371 	fixjobc(p, p->p_pgrp, 0);
372 	sx_xunlock(&proctree_lock);
373 	(void)acct_process(td);
374 
375 	/* Release the TTY now we've unlocked everything. */
376 	if (ttyvp != NULL)
377 		vrele(ttyvp);
378 #ifdef KTRACE
379 	ktrprocexit(td);
380 #endif
381 	/*
382 	 * Release reference to text vnode
383 	 */
384 	if ((vtmp = p->p_textvp) != NULL) {
385 		p->p_textvp = NULL;
386 		locked = VFS_LOCK_GIANT(vtmp->v_mount);
387 		vrele(vtmp);
388 		VFS_UNLOCK_GIANT(locked);
389 	}
390 
391 	/*
392 	 * Release our limits structure.
393 	 */
394 	PROC_LOCK(p);
395 	plim = p->p_limit;
396 	p->p_limit = NULL;
397 	PROC_UNLOCK(p);
398 	lim_free(plim);
399 
400 	tidhash_remove(td);
401 
402 	/*
403 	 * Remove proc from allproc queue and pidhash chain.
404 	 * Place onto zombproc.  Unlink from parent's child list.
405 	 */
406 	sx_xlock(&allproc_lock);
407 	LIST_REMOVE(p, p_list);
408 	LIST_INSERT_HEAD(&zombproc, p, p_list);
409 	LIST_REMOVE(p, p_hash);
410 	sx_xunlock(&allproc_lock);
411 
412 	/*
413 	 * Call machine-dependent code to release any
414 	 * machine-dependent resources other than the address space.
415 	 * The address space is released by "vmspace_exitfree(p)" in
416 	 * vm_waitproc().
417 	 */
418 	cpu_exit(td);
419 
420 	WITNESS_WARN(WARN_PANIC, NULL, "process (pid %d) exiting", p->p_pid);
421 
422 	/*
423 	 * Reparent all of our children to init.
424 	 */
425 	sx_xlock(&proctree_lock);
426 	q = LIST_FIRST(&p->p_children);
427 	if (q != NULL)		/* only need this if any child is S_ZOMB */
428 		wakeup(initproc);
429 	for (; q != NULL; q = nq) {
430 		nq = LIST_NEXT(q, p_sibling);
431 		PROC_LOCK(q);
432 		proc_reparent(q, initproc);
433 		q->p_sigparent = SIGCHLD;
434 		/*
435 		 * Traced processes are killed
436 		 * since their existence means someone is screwing up.
437 		 */
438 		if (q->p_flag & P_TRACED) {
439 			struct thread *temp;
440 
441 			/*
442 			 * Since q was found on our children list, the
443 			 * proc_reparent() call moved q to the orphan
444 			 * list due to present P_TRACED flag. Clear
445 			 * orphan link for q now while q is locked.
446 			 */
447 			clear_orphan(q);
448 			q->p_flag &= ~(P_TRACED | P_STOPPED_TRACE);
449 			FOREACH_THREAD_IN_PROC(q, temp)
450 				temp->td_dbgflags &= ~TDB_SUSPEND;
451 			kern_psignal(q, SIGKILL);
452 		}
453 		PROC_UNLOCK(q);
454 	}
455 
456 	/*
457 	 * Also get rid of our orphans.
458 	 */
459 	while ((q = LIST_FIRST(&p->p_orphans)) != NULL) {
460 		PROC_LOCK(q);
461 		clear_orphan(q);
462 		PROC_UNLOCK(q);
463 	}
464 
465 	/* Save exit status. */
466 	PROC_LOCK(p);
467 	p->p_xthread = td;
468 
469 	/* Tell the prison that we are gone. */
470 	prison_proc_free(p->p_ucred->cr_prison);
471 
472 #ifdef KDTRACE_HOOKS
473 	/*
474 	 * Tell the DTrace fasttrap provider about the exit if it
475 	 * has declared an interest.
476 	 */
477 	if (dtrace_fasttrap_exit)
478 		dtrace_fasttrap_exit(p);
479 #endif
480 
481 	/*
482 	 * Notify interested parties of our demise.
483 	 */
484 	KNOTE_LOCKED(&p->p_klist, NOTE_EXIT);
485 
486 #ifdef KDTRACE_HOOKS
487 	int reason = CLD_EXITED;
488 	if (WCOREDUMP(rv))
489 		reason = CLD_DUMPED;
490 	else if (WIFSIGNALED(rv))
491 		reason = CLD_KILLED;
492 	SDT_PROBE(proc, kernel, , exit, reason, 0, 0, 0, 0);
493 #endif
494 
495 	/*
496 	 * Just delete all entries in the p_klist. At this point we won't
497 	 * report any more events, and there are nasty race conditions that
498 	 * can beat us if we don't.
499 	 */
500 	knlist_clear(&p->p_klist, 1);
501 
502 	/*
503 	 * If this is a process with a descriptor, we may not need to deliver
504 	 * a signal to the parent.  proctree_lock is held over
505 	 * procdesc_exit() to serialize concurrent calls to close() and
506 	 * exit().
507 	 */
508 #ifdef PROCDESC
509 	if (p->p_procdesc == NULL || procdesc_exit(p)) {
510 #endif
511 		/*
512 		 * Notify parent that we're gone.  If parent has the
513 		 * PS_NOCLDWAIT flag set, or if the handler is set to SIG_IGN,
514 		 * notify process 1 instead (and hope it will handle this
515 		 * situation).
516 		 */
517 		PROC_LOCK(p->p_pptr);
518 		mtx_lock(&p->p_pptr->p_sigacts->ps_mtx);
519 		if (p->p_pptr->p_sigacts->ps_flag &
520 		    (PS_NOCLDWAIT | PS_CLDSIGIGN)) {
521 			struct proc *pp;
522 
523 			mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx);
524 			pp = p->p_pptr;
525 			PROC_UNLOCK(pp);
526 			proc_reparent(p, initproc);
527 			p->p_sigparent = SIGCHLD;
528 			PROC_LOCK(p->p_pptr);
529 
530 			/*
531 			 * Notify parent, so in case he was wait(2)ing or
532 			 * executing waitpid(2) with our pid, he will
533 			 * continue.
534 			 */
535 			wakeup(pp);
536 		} else
537 			mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx);
538 
539 		if (p->p_pptr == initproc)
540 			kern_psignal(p->p_pptr, SIGCHLD);
541 		else if (p->p_sigparent != 0) {
542 			if (p->p_sigparent == SIGCHLD)
543 				childproc_exited(p);
544 			else	/* LINUX thread */
545 				kern_psignal(p->p_pptr, p->p_sigparent);
546 		}
547 #ifdef PROCDESC
548 	} else
549 		PROC_LOCK(p->p_pptr);
550 #endif
551 	sx_xunlock(&proctree_lock);
552 
553 	/*
554 	 * The state PRS_ZOMBIE prevents other proesses from sending
555 	 * signal to the process, to avoid memory leak, we free memory
556 	 * for signal queue at the time when the state is set.
557 	 */
558 	sigqueue_flush(&p->p_sigqueue);
559 	sigqueue_flush(&td->td_sigqueue);
560 
561 	/*
562 	 * We have to wait until after acquiring all locks before
563 	 * changing p_state.  We need to avoid all possible context
564 	 * switches (including ones from blocking on a mutex) while
565 	 * marked as a zombie.  We also have to set the zombie state
566 	 * before we release the parent process' proc lock to avoid
567 	 * a lost wakeup.  So, we first call wakeup, then we grab the
568 	 * sched lock, update the state, and release the parent process'
569 	 * proc lock.
570 	 */
571 	wakeup(p->p_pptr);
572 	cv_broadcast(&p->p_pwait);
573 	sched_exit(p->p_pptr, td);
574 	PROC_SLOCK(p);
575 	p->p_state = PRS_ZOMBIE;
576 	PROC_UNLOCK(p->p_pptr);
577 
578 	/*
579 	 * Hopefully no one will try to deliver a signal to the process this
580 	 * late in the game.
581 	 */
582 	knlist_destroy(&p->p_klist);
583 
584 	/*
585 	 * Save our children's rusage information in our exit rusage.
586 	 */
587 	ruadd(&p->p_ru, &p->p_rux, &p->p_stats->p_cru, &p->p_crux);
588 
589 	/*
590 	 * Make sure the scheduler takes this thread out of its tables etc.
591 	 * This will also release this thread's reference to the ucred.
592 	 * Other thread parts to release include pcb bits and such.
593 	 */
594 	thread_exit();
595 }
596 
597 
598 #ifndef _SYS_SYSPROTO_H_
599 struct abort2_args {
600 	char *why;
601 	int nargs;
602 	void **args;
603 };
604 #endif
605 
606 int
607 sys_abort2(struct thread *td, struct abort2_args *uap)
608 {
609 	struct proc *p = td->td_proc;
610 	struct sbuf *sb;
611 	void *uargs[16];
612 	int error, i, sig;
613 
614 	/*
615 	 * Do it right now so we can log either proper call of abort2(), or
616 	 * note, that invalid argument was passed. 512 is big enough to
617 	 * handle 16 arguments' descriptions with additional comments.
618 	 */
619 	sb = sbuf_new(NULL, NULL, 512, SBUF_FIXEDLEN);
620 	sbuf_clear(sb);
621 	sbuf_printf(sb, "%s(pid %d uid %d) aborted: ",
622 	    p->p_comm, p->p_pid, td->td_ucred->cr_uid);
623 	/*
624 	 * Since we can't return from abort2(), send SIGKILL in cases, where
625 	 * abort2() was called improperly
626 	 */
627 	sig = SIGKILL;
628 	/* Prevent from DoSes from user-space. */
629 	if (uap->nargs < 0 || uap->nargs > 16)
630 		goto out;
631 	if (uap->nargs > 0) {
632 		if (uap->args == NULL)
633 			goto out;
634 		error = copyin(uap->args, uargs, uap->nargs * sizeof(void *));
635 		if (error != 0)
636 			goto out;
637 	}
638 	/*
639 	 * Limit size of 'reason' string to 128. Will fit even when
640 	 * maximal number of arguments was chosen to be logged.
641 	 */
642 	if (uap->why != NULL) {
643 		error = sbuf_copyin(sb, uap->why, 128);
644 		if (error < 0)
645 			goto out;
646 	} else {
647 		sbuf_printf(sb, "(null)");
648 	}
649 	if (uap->nargs > 0) {
650 		sbuf_printf(sb, "(");
651 		for (i = 0;i < uap->nargs; i++)
652 			sbuf_printf(sb, "%s%p", i == 0 ? "" : ", ", uargs[i]);
653 		sbuf_printf(sb, ")");
654 	}
655 	/*
656 	 * Final stage: arguments were proper, string has been
657 	 * successfully copied from userspace, and copying pointers
658 	 * from user-space succeed.
659 	 */
660 	sig = SIGABRT;
661 out:
662 	if (sig == SIGKILL) {
663 		sbuf_trim(sb);
664 		sbuf_printf(sb, " (Reason text inaccessible)");
665 	}
666 	sbuf_cat(sb, "\n");
667 	sbuf_finish(sb);
668 	log(LOG_INFO, "%s", sbuf_data(sb));
669 	sbuf_delete(sb);
670 	exit1(td, W_EXITCODE(0, sig));
671 	return (0);
672 }
673 
674 
675 #ifdef COMPAT_43
676 /*
677  * The dirty work is handled by kern_wait().
678  */
679 int
680 owait(struct thread *td, struct owait_args *uap __unused)
681 {
682 	int error, status;
683 
684 	error = kern_wait(td, WAIT_ANY, &status, 0, NULL);
685 	if (error == 0)
686 		td->td_retval[1] = status;
687 	return (error);
688 }
689 #endif /* COMPAT_43 */
690 
691 /*
692  * The dirty work is handled by kern_wait().
693  */
694 int
695 sys_wait4(struct thread *td, struct wait_args *uap)
696 {
697 	struct rusage ru, *rup;
698 	int error, status;
699 
700 	if (uap->rusage != NULL)
701 		rup = &ru;
702 	else
703 		rup = NULL;
704 	error = kern_wait(td, uap->pid, &status, uap->options, rup);
705 	if (uap->status != NULL && error == 0)
706 		error = copyout(&status, uap->status, sizeof(status));
707 	if (uap->rusage != NULL && error == 0)
708 		error = copyout(&ru, uap->rusage, sizeof(struct rusage));
709 	return (error);
710 }
711 
712 /*
713  * Reap the remains of a zombie process and optionally return status and
714  * rusage.  Asserts and will release both the proctree_lock and the process
715  * lock as part of its work.
716  */
717 void
718 proc_reap(struct thread *td, struct proc *p, int *status, int options,
719     struct rusage *rusage)
720 {
721 	struct proc *q, *t;
722 
723 	sx_assert(&proctree_lock, SA_XLOCKED);
724 	PROC_LOCK_ASSERT(p, MA_OWNED);
725 	PROC_SLOCK_ASSERT(p, MA_OWNED);
726 	KASSERT(p->p_state == PRS_ZOMBIE, ("proc_reap: !PRS_ZOMBIE"));
727 
728 	q = td->td_proc;
729 	if (rusage) {
730 		*rusage = p->p_ru;
731 		calcru(p, &rusage->ru_utime, &rusage->ru_stime);
732 	}
733 	PROC_SUNLOCK(p);
734 	td->td_retval[0] = p->p_pid;
735 	if (status)
736 		*status = p->p_xstat;	/* convert to int */
737 	if (options & WNOWAIT) {
738 		/*
739 		 *  Only poll, returning the status.  Caller does not wish to
740 		 * release the proc struct just yet.
741 		 */
742 		PROC_UNLOCK(p);
743 		sx_xunlock(&proctree_lock);
744 		return;
745 	}
746 
747 	PROC_LOCK(q);
748 	sigqueue_take(p->p_ksi);
749 	PROC_UNLOCK(q);
750 	PROC_UNLOCK(p);
751 
752 	/*
753 	 * If we got the child via a ptrace 'attach', we need to give it back
754 	 * to the old parent.
755 	 */
756 	if (p->p_oppid && (t = pfind(p->p_oppid)) != NULL) {
757 		PROC_LOCK(p);
758 		proc_reparent(p, t);
759 		p->p_oppid = 0;
760 		PROC_UNLOCK(p);
761 		pksignal(t, SIGCHLD, p->p_ksi);
762 		wakeup(t);
763 		cv_broadcast(&p->p_pwait);
764 		PROC_UNLOCK(t);
765 		sx_xunlock(&proctree_lock);
766 		return;
767 	}
768 
769 	/*
770 	 * Remove other references to this process to ensure we have an
771 	 * exclusive reference.
772 	 */
773 	sx_xlock(&allproc_lock);
774 	LIST_REMOVE(p, p_list);	/* off zombproc */
775 	sx_xunlock(&allproc_lock);
776 	LIST_REMOVE(p, p_sibling);
777 	PROC_LOCK(p);
778 	clear_orphan(p);
779 	PROC_UNLOCK(p);
780 	leavepgrp(p);
781 #ifdef PROCDESC
782 	if (p->p_procdesc != NULL)
783 		procdesc_reap(p);
784 #endif
785 	sx_xunlock(&proctree_lock);
786 
787 	/*
788 	 * As a side effect of this lock, we know that all other writes to
789 	 * this proc are visible now, so 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, &p->p_rux);
796 	PROC_UNLOCK(q);
797 
798 	/*
799 	 * Decrement the count of procs running with this uid.
800 	 */
801 	(void)chgproccnt(p->p_ucred->cr_ruidinfo, -1, 0);
802 
803 	/*
804 	 * Destroy resource accounting information associated with the process.
805 	 */
806 #ifdef RACCT
807 	PROC_LOCK(p);
808 	racct_sub(p, RACCT_NPROC, 1);
809 	PROC_UNLOCK(p);
810 #endif
811 	racct_proc_exit(p);
812 
813 	/*
814 	 * Free credentials, arguments, and sigacts.
815 	 */
816 	crfree(p->p_ucred);
817 	p->p_ucred = NULL;
818 	pargs_drop(p->p_args);
819 	p->p_args = NULL;
820 	sigacts_free(p->p_sigacts);
821 	p->p_sigacts = NULL;
822 
823 	/*
824 	 * Do any thread-system specific cleanups.
825 	 */
826 	thread_wait(p);
827 
828 	/*
829 	 * Give vm and machine-dependent layer a chance to free anything that
830 	 * cpu_exit couldn't release while still running in process context.
831 	 */
832 	vm_waitproc(p);
833 #ifdef MAC
834 	mac_proc_destroy(p);
835 #endif
836 	KASSERT(FIRST_THREAD_IN_PROC(p),
837 	    ("proc_reap: no residual thread!"));
838 	uma_zfree(proc_zone, p);
839 	sx_xlock(&allproc_lock);
840 	nprocs--;
841 	sx_xunlock(&allproc_lock);
842 }
843 
844 static int
845 proc_to_reap(struct thread *td, struct proc *p, pid_t pid, int *status,
846     int options, struct rusage *rusage)
847 {
848 	struct proc *q;
849 
850 	sx_assert(&proctree_lock, SA_XLOCKED);
851 
852 	q = td->td_proc;
853 	PROC_LOCK(p);
854 	if (pid != WAIT_ANY && p->p_pid != pid && p->p_pgid != -pid) {
855 		PROC_UNLOCK(p);
856 		return (0);
857 	}
858 	if (p_canwait(td, p)) {
859 		PROC_UNLOCK(p);
860 		return (0);
861 	}
862 
863 	/*
864 	 * This special case handles a kthread spawned by linux_clone
865 	 * (see linux_misc.c).  The linux_wait4 and linux_waitpid
866 	 * functions need to be able to distinguish between waiting
867 	 * on a process and waiting on a thread.  It is a thread if
868 	 * p_sigparent is not SIGCHLD, and the WLINUXCLONE option
869 	 * signifies we want to wait for threads and not processes.
870 	 */
871 	if ((p->p_sigparent != SIGCHLD) ^
872 	    ((options & WLINUXCLONE) != 0)) {
873 		PROC_UNLOCK(p);
874 		return (0);
875 	}
876 
877 	PROC_SLOCK(p);
878 	if (p->p_state == PRS_ZOMBIE) {
879 		proc_reap(td, p, status, options, rusage);
880 		return (-1);
881 	}
882 	PROC_SUNLOCK(p);
883 	PROC_UNLOCK(p);
884 	return (1);
885 }
886 
887 int
888 kern_wait(struct thread *td, pid_t pid, int *status, int options,
889     struct rusage *rusage)
890 {
891 	struct proc *p, *q;
892 	int error, nfound, ret;
893 
894 	AUDIT_ARG_PID(pid);
895 	AUDIT_ARG_VALUE(options);
896 
897 	q = td->td_proc;
898 	if (pid == 0) {
899 		PROC_LOCK(q);
900 		pid = -q->p_pgid;
901 		PROC_UNLOCK(q);
902 	}
903 	/* If we don't know the option, just return. */
904 	if (options & ~(WUNTRACED|WNOHANG|WCONTINUED|WNOWAIT|WLINUXCLONE))
905 		return (EINVAL);
906 loop:
907 	if (q->p_flag & P_STATCHILD) {
908 		PROC_LOCK(q);
909 		q->p_flag &= ~P_STATCHILD;
910 		PROC_UNLOCK(q);
911 	}
912 	nfound = 0;
913 	sx_xlock(&proctree_lock);
914 	LIST_FOREACH(p, &q->p_children, p_sibling) {
915 		ret = proc_to_reap(td, p, pid, status, options, rusage);
916 		if (ret == 0)
917 			continue;
918 		else if (ret == 1)
919 			nfound++;
920 		else
921 			return (0);
922 
923 		PROC_LOCK(p);
924 		PROC_SLOCK(p);
925 		if ((p->p_flag & P_STOPPED_SIG) &&
926 		    (p->p_suspcount == p->p_numthreads) &&
927 		    (p->p_flag & P_WAITED) == 0 &&
928 		    (p->p_flag & P_TRACED || options & WUNTRACED)) {
929 			PROC_SUNLOCK(p);
930 			p->p_flag |= P_WAITED;
931 			sx_xunlock(&proctree_lock);
932 			td->td_retval[0] = p->p_pid;
933 			if (status)
934 				*status = W_STOPCODE(p->p_xstat);
935 
936 			PROC_LOCK(q);
937 			sigqueue_take(p->p_ksi);
938 			PROC_UNLOCK(q);
939 			PROC_UNLOCK(p);
940 
941 			return (0);
942 		}
943 		PROC_SUNLOCK(p);
944 		if (options & WCONTINUED && (p->p_flag & P_CONTINUED)) {
945 			sx_xunlock(&proctree_lock);
946 			td->td_retval[0] = p->p_pid;
947 			p->p_flag &= ~P_CONTINUED;
948 
949 			PROC_LOCK(q);
950 			sigqueue_take(p->p_ksi);
951 			PROC_UNLOCK(q);
952 			PROC_UNLOCK(p);
953 
954 			if (status)
955 				*status = SIGCONT;
956 			return (0);
957 		}
958 		PROC_UNLOCK(p);
959 	}
960 
961 	/*
962 	 * Look in the orphans list too, to allow the parent to
963 	 * collect it's child exit status even if child is being
964 	 * debugged.
965 	 *
966 	 * Debugger detaches from the parent upon successful
967 	 * switch-over from parent to child.  At this point due to
968 	 * re-parenting the parent loses the child to debugger and a
969 	 * wait4(2) call would report that it has no children to wait
970 	 * for.  By maintaining a list of orphans we allow the parent
971 	 * to successfully wait until the child becomes a zombie.
972 	 */
973 	LIST_FOREACH(p, &q->p_orphans, p_orphan) {
974 		ret = proc_to_reap(td, p, pid, status, options, rusage);
975 		if (ret == 0)
976 			continue;
977 		else if (ret == 1)
978 			nfound++;
979 		else
980 			return (0);
981 	}
982 	if (nfound == 0) {
983 		sx_xunlock(&proctree_lock);
984 		return (ECHILD);
985 	}
986 	if (options & WNOHANG) {
987 		sx_xunlock(&proctree_lock);
988 		td->td_retval[0] = 0;
989 		return (0);
990 	}
991 	PROC_LOCK(q);
992 	sx_xunlock(&proctree_lock);
993 	if (q->p_flag & P_STATCHILD) {
994 		q->p_flag &= ~P_STATCHILD;
995 		error = 0;
996 	} else
997 		error = msleep(q, &q->p_mtx, PWAIT | PCATCH, "wait", 0);
998 	PROC_UNLOCK(q);
999 	if (error)
1000 		return (error);
1001 	goto loop;
1002 }
1003 
1004 /*
1005  * Make process 'parent' the new parent of process 'child'.
1006  * Must be called with an exclusive hold of proctree lock.
1007  */
1008 void
1009 proc_reparent(struct proc *child, struct proc *parent)
1010 {
1011 
1012 	sx_assert(&proctree_lock, SX_XLOCKED);
1013 	PROC_LOCK_ASSERT(child, MA_OWNED);
1014 	if (child->p_pptr == parent)
1015 		return;
1016 
1017 	PROC_LOCK(child->p_pptr);
1018 	sigqueue_take(child->p_ksi);
1019 	PROC_UNLOCK(child->p_pptr);
1020 	LIST_REMOVE(child, p_sibling);
1021 	LIST_INSERT_HEAD(&parent->p_children, child, p_sibling);
1022 
1023 	clear_orphan(child);
1024 	if (child->p_flag & P_TRACED) {
1025 		LIST_INSERT_HEAD(&child->p_pptr->p_orphans, child, p_orphan);
1026 		child->p_flag |= P_ORPHAN;
1027 	}
1028 
1029 	child->p_pptr = parent;
1030 }
1031