xref: /freebsd/sys/kern/kern_exit.c (revision 195ebc7e9e4b129de810833791a19dfb4349d6a9)
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_mac.h"
44 
45 #include <sys/param.h>
46 #include <sys/systm.h>
47 #include <sys/sysproto.h>
48 #include <sys/eventhandler.h>
49 #include <sys/kernel.h>
50 #include <sys/malloc.h>
51 #include <sys/lock.h>
52 #include <sys/mutex.h>
53 #include <sys/proc.h>
54 #include <sys/pioctl.h>
55 #include <sys/jail.h>
56 #include <sys/tty.h>
57 #include <sys/wait.h>
58 #include <sys/vmmeter.h>
59 #include <sys/vnode.h>
60 #include <sys/resourcevar.h>
61 #include <sys/sbuf.h>
62 #include <sys/signalvar.h>
63 #include <sys/sched.h>
64 #include <sys/sx.h>
65 #include <sys/syscallsubr.h>
66 #include <sys/syslog.h>
67 #include <sys/ptrace.h>
68 #include <sys/acct.h>		/* for acct_process() function prototype */
69 #include <sys/filedesc.h>
70 #include <sys/sdt.h>
71 #include <sys/shm.h>
72 #include <sys/sem.h>
73 #include <sys/vimage.h>
74 #ifdef KTRACE
75 #include <sys/ktrace.h>
76 #endif
77 
78 #include <security/audit/audit.h>
79 #include <security/mac/mac_framework.h>
80 
81 #include <vm/vm.h>
82 #include <vm/vm_extern.h>
83 #include <vm/vm_param.h>
84 #include <vm/pmap.h>
85 #include <vm/vm_map.h>
86 #include <vm/vm_page.h>
87 #include <vm/uma.h>
88 
89 #ifdef KDTRACE_HOOKS
90 #include <sys/dtrace_bsd.h>
91 dtrace_execexit_func_t	dtrace_fasttrap_exit;
92 #endif
93 
94 SDT_PROVIDER_DECLARE(proc);
95 SDT_PROBE_DEFINE(proc, kernel, , exit);
96 SDT_PROBE_ARGTYPE(proc, kernel, , exit, 0, "int");
97 
98 /* Required to be non-static for SysVR4 emulator */
99 MALLOC_DEFINE(M_ZOMBIE, "zombie", "zombie proc status");
100 
101 /* Hook for NFS teardown procedure. */
102 void (*nlminfo_release_p)(struct proc *p);
103 
104 /*
105  * exit -- death of process.
106  */
107 void
108 sys_exit(struct thread *td, struct sys_exit_args *uap)
109 {
110 
111 	exit1(td, W_EXITCODE(uap->rval, 0));
112 	/* NOTREACHED */
113 }
114 
115 /*
116  * Exit: deallocate address space and other resources, change proc state to
117  * zombie, and unlink proc from allproc and parent's lists.  Save exit status
118  * and rusage for wait().  Check for child processes and orphan them.
119  */
120 void
121 exit1(struct thread *td, int rv)
122 {
123 	struct proc *p, *nq, *q;
124 	struct vnode *vtmp;
125 	struct vnode *ttyvp = NULL;
126 #ifdef KTRACE
127 	struct vnode *tracevp;
128 	struct ucred *tracecred;
129 #endif
130 	struct plimit *plim;
131 	int locked;
132 
133 	mtx_assert(&Giant, MA_NOTOWNED);
134 
135 	p = td->td_proc;
136 	if (p == initproc) {
137 		printf("init died (signal %d, exit %d)\n",
138 		    WTERMSIG(rv), WEXITSTATUS(rv));
139 		panic("Going nowhere without my init!");
140 	}
141 
142 	/*
143 	 * MUST abort all other threads before proceeding past here.
144 	 */
145 	PROC_LOCK(p);
146 	while (p->p_flag & P_HADTHREADS) {
147 		/*
148 		 * First check if some other thread got here before us..
149 		 * if so, act apropriatly, (exit or suspend);
150 		 */
151 		thread_suspend_check(0);
152 
153 		/*
154 		 * Kill off the other threads. This requires
155 		 * some co-operation from other parts of the kernel
156 		 * so it may not be instantaneous.  With this state set
157 		 * any thread entering the kernel from userspace will
158 		 * thread_exit() in trap().  Any thread attempting to
159 		 * sleep will return immediately with EINTR or EWOULDBLOCK
160 		 * which will hopefully force them to back out to userland
161 		 * freeing resources as they go.  Any thread attempting
162 		 * to return to userland will thread_exit() from userret().
163 		 * thread_exit() will unsuspend us when the last of the
164 		 * other threads exits.
165 		 * If there is already a thread singler after resumption,
166 		 * calling thread_single will fail; in that case, we just
167 		 * re-check all suspension request, the thread should
168 		 * either be suspended there or exit.
169 		 */
170 		if (! thread_single(SINGLE_EXIT))
171 			break;
172 
173 		/*
174 		 * All other activity in this process is now stopped.
175 		 * Threading support has been turned off.
176 		 */
177 	}
178 	KASSERT(p->p_numthreads == 1,
179 	    ("exit1: proc %p exiting with %d threads", p, p->p_numthreads));
180 	/*
181 	 * Wakeup anyone in procfs' PIOCWAIT.  They should have a hold
182 	 * on our vmspace, so we should block below until they have
183 	 * released their reference to us.  Note that if they have
184 	 * requested S_EXIT stops we will block here until they ack
185 	 * via PIOCCONT.
186 	 */
187 	_STOPEVENT(p, S_EXIT, rv);
188 
189 	/*
190 	 * Note that we are exiting and do another wakeup of anyone in
191 	 * PIOCWAIT in case they aren't listening for S_EXIT stops or
192 	 * decided to wait again after we told them we are exiting.
193 	 */
194 	p->p_flag |= P_WEXIT;
195 	wakeup(&p->p_stype);
196 
197 	/*
198 	 * Wait for any processes that have a hold on our vmspace to
199 	 * release their reference.
200 	 */
201 	while (p->p_lock > 0)
202 		msleep(&p->p_lock, &p->p_mtx, PWAIT, "exithold", 0);
203 
204 	PROC_UNLOCK(p);
205 	/* Drain the limit callout while we don't have the proc locked */
206 	callout_drain(&p->p_limco);
207 
208 #ifdef AUDIT
209 	/*
210 	 * The Sun BSM exit token contains two components: an exit status as
211 	 * passed to exit(), and a return value to indicate what sort of exit
212 	 * it was.  The exit status is WEXITSTATUS(rv), but it's not clear
213 	 * what the return value is.
214 	 */
215 	AUDIT_ARG(exit, WEXITSTATUS(rv), 0);
216 	AUDIT_SYSCALL_EXIT(0, td);
217 #endif
218 
219 	/* Are we a task leader? */
220 	if (p == p->p_leader) {
221 		mtx_lock(&ppeers_lock);
222 		q = p->p_peers;
223 		while (q != NULL) {
224 			PROC_LOCK(q);
225 			psignal(q, SIGKILL);
226 			PROC_UNLOCK(q);
227 			q = q->p_peers;
228 		}
229 		while (p->p_peers != NULL)
230 			msleep(p, &ppeers_lock, PWAIT, "exit1", 0);
231 		mtx_unlock(&ppeers_lock);
232 	}
233 
234 	/*
235 	 * Check if any loadable modules need anything done at process exit.
236 	 * E.g. SYSV IPC stuff
237 	 * XXX what if one of these generates an error?
238 	 */
239 	EVENTHANDLER_INVOKE(process_exit, p);
240 
241 	/*
242 	 * If parent is waiting for us to exit or exec,
243 	 * P_PPWAIT is set; we will wakeup the parent below.
244 	 */
245 	PROC_LOCK(p);
246 	stopprofclock(p);
247 	p->p_flag &= ~(P_TRACED | P_PPWAIT);
248 
249 	/*
250 	 * Stop the real interval timer.  If the handler is currently
251 	 * executing, prevent it from rearming itself and let it finish.
252 	 */
253 	if (timevalisset(&p->p_realtimer.it_value) &&
254 	    callout_stop(&p->p_itcallout) == 0) {
255 		timevalclear(&p->p_realtimer.it_interval);
256 		msleep(&p->p_itcallout, &p->p_mtx, PWAIT, "ritwait", 0);
257 		KASSERT(!timevalisset(&p->p_realtimer.it_value),
258 		    ("realtime timer is still armed"));
259 	}
260 	PROC_UNLOCK(p);
261 
262 	/*
263 	 * Reset any sigio structures pointing to us as a result of
264 	 * F_SETOWN with our pid.
265 	 */
266 	funsetownlst(&p->p_sigiolst);
267 
268 	/*
269 	 * If this process has an nlminfo data area (for lockd), release it
270 	 */
271 	if (nlminfo_release_p != NULL && p->p_nlminfo != NULL)
272 		(*nlminfo_release_p)(p);
273 
274 	/*
275 	 * Close open files and release open-file table.
276 	 * This may block!
277 	 */
278 	fdfree(td);
279 
280 	/*
281 	 * If this thread tickled GEOM, we need to wait for the giggling to
282 	 * stop before we return to userland
283 	 */
284 	if (td->td_pflags & TDP_GEOM)
285 		g_waitidle();
286 
287 	/*
288 	 * Remove ourself from our leader's peer list and wake our leader.
289 	 */
290 	mtx_lock(&ppeers_lock);
291 	if (p->p_leader->p_peers) {
292 		q = p->p_leader;
293 		while (q->p_peers != p)
294 			q = q->p_peers;
295 		q->p_peers = p->p_peers;
296 		wakeup(p->p_leader);
297 	}
298 	mtx_unlock(&ppeers_lock);
299 
300 	vmspace_exit(td);
301 
302 	sx_xlock(&proctree_lock);
303 	if (SESS_LEADER(p)) {
304 		struct session *sp;
305 
306 		sp = p->p_session;
307 
308 		SESS_LOCK(sp);
309 		ttyvp = sp->s_ttyvp;
310 		sp->s_ttyvp = NULL;
311 		SESS_UNLOCK(sp);
312 
313 		if (ttyvp != NULL) {
314 			/*
315 			 * Controlling process.
316 			 * Signal foreground pgrp and revoke access to
317 			 * controlling terminal.
318 			 *
319 			 * There is no need to drain the terminal here,
320 			 * because this will be done on revocation.
321 			 */
322 			if (sp->s_ttyp != NULL) {
323 				struct tty *tp = sp->s_ttyp;
324 
325 				tty_lock(tp);
326 				tty_signal_pgrp(tp, SIGHUP);
327 				tty_unlock(tp);
328 
329 				/*
330 				 * The tty could have been revoked
331 				 * if we blocked.
332 				 */
333 				if (ttyvp->v_type != VBAD) {
334 					sx_xunlock(&proctree_lock);
335 					VOP_LOCK(ttyvp, LK_EXCLUSIVE);
336 					VOP_REVOKE(ttyvp, REVOKEALL);
337 					VOP_UNLOCK(ttyvp, 0);
338 					sx_xlock(&proctree_lock);
339 				}
340 			}
341 			/*
342 			 * s_ttyp is not zero'd; we use this to indicate that
343 			 * the session once had a controlling terminal.
344 			 * (for logging and informational purposes)
345 			 */
346 		}
347 		SESS_LOCK(p->p_session);
348 		sp->s_leader = NULL;
349 		SESS_UNLOCK(p->p_session);
350 	}
351 	fixjobc(p, p->p_pgrp, 0);
352 	sx_xunlock(&proctree_lock);
353 	(void)acct_process(td);
354 
355 	/* Release the TTY now we've unlocked everything. */
356 	if (ttyvp != NULL)
357 		vrele(ttyvp);
358 #ifdef KTRACE
359 	/*
360 	 * Disable tracing, then drain any pending records and release
361 	 * the trace file.
362 	 */
363 	if (p->p_traceflag != 0) {
364 		PROC_LOCK(p);
365 		mtx_lock(&ktrace_mtx);
366 		p->p_traceflag = 0;
367 		mtx_unlock(&ktrace_mtx);
368 		PROC_UNLOCK(p);
369 		ktrprocexit(td);
370 		PROC_LOCK(p);
371 		mtx_lock(&ktrace_mtx);
372 		tracevp = p->p_tracevp;
373 		p->p_tracevp = NULL;
374 		tracecred = p->p_tracecred;
375 		p->p_tracecred = NULL;
376 		mtx_unlock(&ktrace_mtx);
377 		PROC_UNLOCK(p);
378 		if (tracevp != NULL) {
379 			locked = VFS_LOCK_GIANT(tracevp->v_mount);
380 			vrele(tracevp);
381 			VFS_UNLOCK_GIANT(locked);
382 		}
383 		if (tracecred != NULL)
384 			crfree(tracecred);
385 	}
386 #endif
387 	/*
388 	 * Release reference to text vnode
389 	 */
390 	if ((vtmp = p->p_textvp) != NULL) {
391 		p->p_textvp = NULL;
392 		locked = VFS_LOCK_GIANT(vtmp->v_mount);
393 		vrele(vtmp);
394 		VFS_UNLOCK_GIANT(locked);
395 	}
396 
397 	/*
398 	 * Release our limits structure.
399 	 */
400 	PROC_LOCK(p);
401 	plim = p->p_limit;
402 	p->p_limit = NULL;
403 	PROC_UNLOCK(p);
404 	lim_free(plim);
405 
406 	/*
407 	 * Remove proc from allproc queue and pidhash chain.
408 	 * Place onto zombproc.  Unlink from parent's child list.
409 	 */
410 	sx_xlock(&allproc_lock);
411 	LIST_REMOVE(p, p_list);
412 	LIST_INSERT_HEAD(&zombproc, p, p_list);
413 	LIST_REMOVE(p, p_hash);
414 	sx_xunlock(&allproc_lock);
415 
416 	/*
417 	 * Call machine-dependent code to release any
418 	 * machine-dependent resources other than the address space.
419 	 * The address space is released by "vmspace_exitfree(p)" in
420 	 * vm_waitproc().
421 	 */
422 	cpu_exit(td);
423 
424 	WITNESS_WARN(WARN_PANIC, NULL, "process (pid %d) exiting", p->p_pid);
425 
426 	/*
427 	 * Reparent all of our children to init.
428 	 */
429 	sx_xlock(&proctree_lock);
430 	q = LIST_FIRST(&p->p_children);
431 	if (q != NULL)		/* only need this if any child is S_ZOMB */
432 		wakeup(initproc);
433 	for (; q != NULL; q = nq) {
434 		nq = LIST_NEXT(q, p_sibling);
435 		PROC_LOCK(q);
436 		proc_reparent(q, initproc);
437 		q->p_sigparent = SIGCHLD;
438 		/*
439 		 * Traced processes are killed
440 		 * since their existence means someone is screwing up.
441 		 */
442 		if (q->p_flag & P_TRACED) {
443 			struct thread *temp;
444 
445 			q->p_flag &= ~(P_TRACED | P_STOPPED_TRACE);
446 			FOREACH_THREAD_IN_PROC(q, temp)
447 				temp->td_dbgflags &= ~TDB_SUSPEND;
448 			psignal(q, SIGKILL);
449 		}
450 		PROC_UNLOCK(q);
451 	}
452 
453 	/* Save exit status. */
454 	PROC_LOCK(p);
455 	p->p_xstat = rv;
456 	p->p_xthread = td;
457 
458 	/* Tell the prison that we are gone. */
459 	prison_proc_free(p->p_ucred->cr_prison);
460 
461 #ifdef KDTRACE_HOOKS
462 	/*
463 	 * Tell the DTrace fasttrap provider about the exit if it
464 	 * has declared an interest.
465 	 */
466 	if (dtrace_fasttrap_exit)
467 		dtrace_fasttrap_exit(p);
468 #endif
469 
470 	/*
471 	 * Notify interested parties of our demise.
472 	 */
473 	KNOTE_LOCKED(&p->p_klist, NOTE_EXIT);
474 
475 #ifdef KDTRACE_HOOKS
476 	int reason = CLD_EXITED;
477 	if (WCOREDUMP(rv))
478 		reason = CLD_DUMPED;
479 	else if (WIFSIGNALED(rv))
480 		reason = CLD_KILLED;
481 	SDT_PROBE(proc, kernel, , exit, reason, 0, 0, 0, 0);
482 #endif
483 
484 	/*
485 	 * Just delete all entries in the p_klist. At this point we won't
486 	 * report any more events, and there are nasty race conditions that
487 	 * can beat us if we don't.
488 	 */
489 	knlist_clear(&p->p_klist, 1);
490 
491 	/*
492 	 * Notify parent that we're gone.  If parent has the PS_NOCLDWAIT
493 	 * flag set, or if the handler is set to SIG_IGN, notify process
494 	 * 1 instead (and hope it will handle this situation).
495 	 */
496 	PROC_LOCK(p->p_pptr);
497 	mtx_lock(&p->p_pptr->p_sigacts->ps_mtx);
498 	if (p->p_pptr->p_sigacts->ps_flag & (PS_NOCLDWAIT | PS_CLDSIGIGN)) {
499 		struct proc *pp;
500 
501 		mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx);
502 		pp = p->p_pptr;
503 		PROC_UNLOCK(pp);
504 		proc_reparent(p, initproc);
505 		p->p_sigparent = SIGCHLD;
506 		PROC_LOCK(p->p_pptr);
507 
508 		/*
509 		 * Notify parent, so in case he was wait(2)ing or
510 		 * executing waitpid(2) with our pid, he will
511 		 * continue.
512 		 */
513 		wakeup(pp);
514 	} else
515 		mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx);
516 
517 	if (p->p_pptr == initproc)
518 		psignal(p->p_pptr, SIGCHLD);
519 	else if (p->p_sigparent != 0) {
520 		if (p->p_sigparent == SIGCHLD)
521 			childproc_exited(p);
522 		else	/* LINUX thread */
523 			psignal(p->p_pptr, p->p_sigparent);
524 	}
525 	sx_xunlock(&proctree_lock);
526 
527 	/*
528 	 * The state PRS_ZOMBIE prevents other proesses from sending
529 	 * signal to the process, to avoid memory leak, we free memory
530 	 * for signal queue at the time when the state is set.
531 	 */
532 	sigqueue_flush(&p->p_sigqueue);
533 	sigqueue_flush(&td->td_sigqueue);
534 
535 	/*
536 	 * We have to wait until after acquiring all locks before
537 	 * changing p_state.  We need to avoid all possible context
538 	 * switches (including ones from blocking on a mutex) while
539 	 * marked as a zombie.  We also have to set the zombie state
540 	 * before we release the parent process' proc lock to avoid
541 	 * a lost wakeup.  So, we first call wakeup, then we grab the
542 	 * sched lock, update the state, and release the parent process'
543 	 * proc lock.
544 	 */
545 	wakeup(p->p_pptr);
546 	cv_broadcast(&p->p_pwait);
547 	sched_exit(p->p_pptr, td);
548 	PROC_SLOCK(p);
549 	p->p_state = PRS_ZOMBIE;
550 	PROC_UNLOCK(p->p_pptr);
551 
552 	/*
553 	 * Hopefully no one will try to deliver a signal to the process this
554 	 * late in the game.
555 	 */
556 	knlist_destroy(&p->p_klist);
557 
558 	/*
559 	 * Save our children's rusage information in our exit rusage.
560 	 */
561 	ruadd(&p->p_ru, &p->p_rux, &p->p_stats->p_cru, &p->p_crux);
562 
563 	/*
564 	 * Make sure the scheduler takes this thread out of its tables etc.
565 	 * This will also release this thread's reference to the ucred.
566 	 * Other thread parts to release include pcb bits and such.
567 	 */
568 	thread_exit();
569 }
570 
571 
572 #ifndef _SYS_SYSPROTO_H_
573 struct abort2_args {
574 	char *why;
575 	int nargs;
576 	void **args;
577 };
578 #endif
579 
580 int
581 abort2(struct thread *td, struct abort2_args *uap)
582 {
583 	struct proc *p = td->td_proc;
584 	struct sbuf *sb;
585 	void *uargs[16];
586 	int error, i, sig;
587 
588 	/*
589 	 * Do it right now so we can log either proper call of abort2(), or
590 	 * note, that invalid argument was passed. 512 is big enough to
591 	 * handle 16 arguments' descriptions with additional comments.
592 	 */
593 	sb = sbuf_new(NULL, NULL, 512, SBUF_FIXEDLEN);
594 	sbuf_clear(sb);
595 	sbuf_printf(sb, "%s(pid %d uid %d) aborted: ",
596 	    p->p_comm, p->p_pid, td->td_ucred->cr_uid);
597 	/*
598 	 * Since we can't return from abort2(), send SIGKILL in cases, where
599 	 * abort2() was called improperly
600 	 */
601 	sig = SIGKILL;
602 	/* Prevent from DoSes from user-space. */
603 	if (uap->nargs < 0 || uap->nargs > 16)
604 		goto out;
605 	if (uap->nargs > 0) {
606 		if (uap->args == NULL)
607 			goto out;
608 		error = copyin(uap->args, uargs, uap->nargs * sizeof(void *));
609 		if (error != 0)
610 			goto out;
611 	}
612 	/*
613 	 * Limit size of 'reason' string to 128. Will fit even when
614 	 * maximal number of arguments was chosen to be logged.
615 	 */
616 	if (uap->why != NULL) {
617 		error = sbuf_copyin(sb, uap->why, 128);
618 		if (error < 0)
619 			goto out;
620 	} else {
621 		sbuf_printf(sb, "(null)");
622 	}
623 	if (uap->nargs > 0) {
624 		sbuf_printf(sb, "(");
625 		for (i = 0;i < uap->nargs; i++)
626 			sbuf_printf(sb, "%s%p", i == 0 ? "" : ", ", uargs[i]);
627 		sbuf_printf(sb, ")");
628 	}
629 	/*
630 	 * Final stage: arguments were proper, string has been
631 	 * successfully copied from userspace, and copying pointers
632 	 * from user-space succeed.
633 	 */
634 	sig = SIGABRT;
635 out:
636 	if (sig == SIGKILL) {
637 		sbuf_trim(sb);
638 		sbuf_printf(sb, " (Reason text inaccessible)");
639 	}
640 	sbuf_cat(sb, "\n");
641 	sbuf_finish(sb);
642 	log(LOG_INFO, "%s", sbuf_data(sb));
643 	sbuf_delete(sb);
644 	exit1(td, W_EXITCODE(0, sig));
645 	return (0);
646 }
647 
648 
649 #ifdef COMPAT_43
650 /*
651  * The dirty work is handled by kern_wait().
652  */
653 int
654 owait(struct thread *td, struct owait_args *uap __unused)
655 {
656 	int error, status;
657 
658 	error = kern_wait(td, WAIT_ANY, &status, 0, NULL);
659 	if (error == 0)
660 		td->td_retval[1] = status;
661 	return (error);
662 }
663 #endif /* COMPAT_43 */
664 
665 /*
666  * The dirty work is handled by kern_wait().
667  */
668 int
669 wait4(struct thread *td, struct wait_args *uap)
670 {
671 	struct rusage ru, *rup;
672 	int error, status;
673 
674 	if (uap->rusage != NULL)
675 		rup = &ru;
676 	else
677 		rup = NULL;
678 	error = kern_wait(td, uap->pid, &status, uap->options, rup);
679 	if (uap->status != NULL && error == 0)
680 		error = copyout(&status, uap->status, sizeof(status));
681 	if (uap->rusage != NULL && error == 0)
682 		error = copyout(&ru, uap->rusage, sizeof(struct rusage));
683 	return (error);
684 }
685 
686 int
687 kern_wait(struct thread *td, pid_t pid, int *status, int options,
688     struct rusage *rusage)
689 {
690 	struct proc *p, *q, *t;
691 	int error, nfound;
692 
693 	AUDIT_ARG(pid, pid);
694 
695 	q = td->td_proc;
696 	if (pid == 0) {
697 		PROC_LOCK(q);
698 		pid = -q->p_pgid;
699 		PROC_UNLOCK(q);
700 	}
701 	if (options &~ (WUNTRACED|WNOHANG|WCONTINUED|WNOWAIT|WLINUXCLONE))
702 		return (EINVAL);
703 loop:
704 	if (q->p_flag & P_STATCHILD) {
705 		PROC_LOCK(q);
706 		q->p_flag &= ~P_STATCHILD;
707 		PROC_UNLOCK(q);
708 	}
709 	nfound = 0;
710 	sx_xlock(&proctree_lock);
711 	LIST_FOREACH(p, &q->p_children, p_sibling) {
712 		PROC_LOCK(p);
713 		if (pid != WAIT_ANY &&
714 		    p->p_pid != pid && p->p_pgid != -pid) {
715 			PROC_UNLOCK(p);
716 			continue;
717 		}
718 		if (p_canwait(td, p)) {
719 			PROC_UNLOCK(p);
720 			continue;
721 		}
722 
723 		/*
724 		 * This special case handles a kthread spawned by linux_clone
725 		 * (see linux_misc.c).  The linux_wait4 and linux_waitpid
726 		 * functions need to be able to distinguish between waiting
727 		 * on a process and waiting on a thread.  It is a thread if
728 		 * p_sigparent is not SIGCHLD, and the WLINUXCLONE option
729 		 * signifies we want to wait for threads and not processes.
730 		 */
731 		if ((p->p_sigparent != SIGCHLD) ^
732 		    ((options & WLINUXCLONE) != 0)) {
733 			PROC_UNLOCK(p);
734 			continue;
735 		}
736 
737 		nfound++;
738 		PROC_SLOCK(p);
739 		if (p->p_state == PRS_ZOMBIE) {
740 			INIT_VPROCG(P_TO_VPROCG(p));
741 			if (rusage) {
742 				*rusage = p->p_ru;
743 				calcru(p, &rusage->ru_utime, &rusage->ru_stime);
744 			}
745 			PROC_SUNLOCK(p);
746 			td->td_retval[0] = p->p_pid;
747 			if (status)
748 				*status = p->p_xstat;	/* convert to int */
749 			if (options & WNOWAIT) {
750 
751 				/*
752 				 *  Only poll, returning the status.
753 				 *  Caller does not wish to release the proc
754 				 *  struct just yet.
755 				 */
756 				PROC_UNLOCK(p);
757 				sx_xunlock(&proctree_lock);
758 				return (0);
759 			}
760 
761 			PROC_LOCK(q);
762 			sigqueue_take(p->p_ksi);
763 			PROC_UNLOCK(q);
764 			PROC_UNLOCK(p);
765 
766 			/*
767 			 * If we got the child via a ptrace 'attach',
768 			 * we need to give it back to the old parent.
769 			 */
770 			if (p->p_oppid && (t = pfind(p->p_oppid)) != NULL) {
771 				PROC_LOCK(p);
772 				p->p_oppid = 0;
773 				proc_reparent(p, t);
774 				PROC_UNLOCK(p);
775 				tdsignal(t, NULL, SIGCHLD, p->p_ksi);
776 				wakeup(t);
777 				cv_broadcast(&p->p_pwait);
778 				PROC_UNLOCK(t);
779 				sx_xunlock(&proctree_lock);
780 				return (0);
781 			}
782 
783 			/*
784 			 * Remove other references to this process to ensure
785 			 * we have an exclusive reference.
786 			 */
787 			sx_xlock(&allproc_lock);
788 			LIST_REMOVE(p, p_list);	/* off zombproc */
789 			sx_xunlock(&allproc_lock);
790 			LIST_REMOVE(p, p_sibling);
791 			leavepgrp(p);
792 			sx_xunlock(&proctree_lock);
793 
794 			/*
795 			 * As a side effect of this lock, we know that
796 			 * all other writes to this proc are visible now, so
797 			 * no more locking is needed for p.
798 			 */
799 			PROC_LOCK(p);
800 			p->p_xstat = 0;		/* XXX: why? */
801 			PROC_UNLOCK(p);
802 			PROC_LOCK(q);
803 			ruadd(&q->p_stats->p_cru, &q->p_crux, &p->p_ru,
804 			    &p->p_rux);
805 			PROC_UNLOCK(q);
806 
807 			/*
808 			 * Decrement the count of procs running with this uid.
809 			 */
810 			(void)chgproccnt(p->p_ucred->cr_ruidinfo, -1, 0);
811 
812 			/*
813 			 * Free credentials, arguments, and sigacts.
814 			 */
815 			crfree(p->p_ucred);
816 			p->p_ucred = NULL;
817 			pargs_drop(p->p_args);
818 			p->p_args = NULL;
819 			sigacts_free(p->p_sigacts);
820 			p->p_sigacts = NULL;
821 
822 			/*
823 			 * Do any thread-system specific cleanups.
824 			 */
825 			thread_wait(p);
826 
827 			/*
828 			 * Give vm and machine-dependent layer a chance
829 			 * to free anything that cpu_exit couldn't
830 			 * 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 			    ("kern_wait: no residual thread!"));
838 			uma_zfree(proc_zone, p);
839 			sx_xlock(&allproc_lock);
840 			nprocs--;
841 #ifdef VIMAGE
842 			vprocg->nprocs--;
843 #endif
844 			sx_xunlock(&allproc_lock);
845 			return (0);
846 		}
847 		if ((p->p_flag & P_STOPPED_SIG) &&
848 		    (p->p_suspcount == p->p_numthreads) &&
849 		    (p->p_flag & P_WAITED) == 0 &&
850 		    (p->p_flag & P_TRACED || options & WUNTRACED)) {
851 			PROC_SUNLOCK(p);
852 			p->p_flag |= P_WAITED;
853 			sx_xunlock(&proctree_lock);
854 			td->td_retval[0] = p->p_pid;
855 			if (status)
856 				*status = W_STOPCODE(p->p_xstat);
857 
858 			PROC_LOCK(q);
859 			sigqueue_take(p->p_ksi);
860 			PROC_UNLOCK(q);
861 			PROC_UNLOCK(p);
862 
863 			return (0);
864 		}
865 		PROC_SUNLOCK(p);
866 		if (options & WCONTINUED && (p->p_flag & P_CONTINUED)) {
867 			sx_xunlock(&proctree_lock);
868 			td->td_retval[0] = p->p_pid;
869 			p->p_flag &= ~P_CONTINUED;
870 
871 			PROC_LOCK(q);
872 			sigqueue_take(p->p_ksi);
873 			PROC_UNLOCK(q);
874 			PROC_UNLOCK(p);
875 
876 			if (status)
877 				*status = SIGCONT;
878 			return (0);
879 		}
880 		PROC_UNLOCK(p);
881 	}
882 	if (nfound == 0) {
883 		sx_xunlock(&proctree_lock);
884 		return (ECHILD);
885 	}
886 	if (options & WNOHANG) {
887 		sx_xunlock(&proctree_lock);
888 		td->td_retval[0] = 0;
889 		return (0);
890 	}
891 	PROC_LOCK(q);
892 	sx_xunlock(&proctree_lock);
893 	if (q->p_flag & P_STATCHILD) {
894 		q->p_flag &= ~P_STATCHILD;
895 		error = 0;
896 	} else
897 		error = msleep(q, &q->p_mtx, PWAIT | PCATCH, "wait", 0);
898 	PROC_UNLOCK(q);
899 	if (error)
900 		return (error);
901 	goto loop;
902 }
903 
904 /*
905  * Make process 'parent' the new parent of process 'child'.
906  * Must be called with an exclusive hold of proctree lock.
907  */
908 void
909 proc_reparent(struct proc *child, struct proc *parent)
910 {
911 
912 	sx_assert(&proctree_lock, SX_XLOCKED);
913 	PROC_LOCK_ASSERT(child, MA_OWNED);
914 	if (child->p_pptr == parent)
915 		return;
916 
917 	PROC_LOCK(child->p_pptr);
918 	sigqueue_take(child->p_ksi);
919 	PROC_UNLOCK(child->p_pptr);
920 	LIST_REMOVE(child, p_sibling);
921 	LIST_INSERT_HEAD(&parent->p_children, child, p_sibling);
922 	child->p_pptr = parent;
923 }
924