xref: /freebsd/sys/kern/kern_exit.c (revision dadef94c7a762d05890e2891bc4a7d1dfe0cf758)
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 
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/jail.h>
55 #include <sys/tty.h>
56 #include <sys/wait.h>
57 #include <sys/vmmeter.h>
58 #include <sys/vnode.h>
59 #include <sys/resourcevar.h>
60 #include <sys/sbuf.h>
61 #include <sys/signalvar.h>
62 #include <sys/sched.h>
63 #include <sys/sx.h>
64 #include <sys/syscallsubr.h>
65 #include <sys/syslog.h>
66 #include <sys/ptrace.h>
67 #include <sys/acct.h>		/* for acct_process() function prototype */
68 #include <sys/filedesc.h>
69 #include <sys/sdt.h>
70 #include <sys/shm.h>
71 #include <sys/sem.h>
72 #ifdef KTRACE
73 #include <sys/ktrace.h>
74 #endif
75 
76 #include <security/audit/audit.h>
77 #include <security/mac/mac_framework.h>
78 
79 #include <vm/vm.h>
80 #include <vm/vm_extern.h>
81 #include <vm/vm_param.h>
82 #include <vm/pmap.h>
83 #include <vm/vm_map.h>
84 #include <vm/vm_page.h>
85 #include <vm/uma.h>
86 
87 #ifdef KDTRACE_HOOKS
88 #include <sys/dtrace_bsd.h>
89 dtrace_execexit_func_t	dtrace_fasttrap_exit;
90 #endif
91 
92 SDT_PROVIDER_DECLARE(proc);
93 SDT_PROBE_DEFINE(proc, kernel, , exit, exit);
94 SDT_PROBE_ARGTYPE(proc, kernel, , exit, 0, "int");
95 
96 /* Required to be non-static for SysVR4 emulator */
97 MALLOC_DEFINE(M_ZOMBIE, "zombie", "zombie proc status");
98 
99 /* Hook for NFS teardown procedure. */
100 void (*nlminfo_release_p)(struct proc *p);
101 
102 /*
103  * exit -- death of process.
104  */
105 void
106 sys_exit(struct thread *td, struct sys_exit_args *uap)
107 {
108 
109 	exit1(td, W_EXITCODE(uap->rval, 0));
110 	/* NOTREACHED */
111 }
112 
113 /*
114  * Exit: deallocate address space and other resources, change proc state to
115  * zombie, and unlink proc from allproc and parent's lists.  Save exit status
116  * and rusage for wait().  Check for child processes and orphan them.
117  */
118 void
119 exit1(struct thread *td, int rv)
120 {
121 	struct proc *p, *nq, *q;
122 	struct vnode *vtmp;
123 	struct vnode *ttyvp = NULL;
124 #ifdef KTRACE
125 	struct vnode *tracevp;
126 	struct ucred *tracecred;
127 #endif
128 	struct plimit *plim;
129 	int locked;
130 
131 	mtx_assert(&Giant, MA_NOTOWNED);
132 
133 	p = td->td_proc;
134 	/*
135 	 * XXX in case we're rebooting we just let init die in order to
136 	 * work around an unsolved stack overflow seen very late during
137 	 * shutdown on sparc64 when the gmirror worker process exists.
138 	 */
139 	if (p == initproc && rebooting == 0) {
140 		printf("init died (signal %d, exit %d)\n",
141 		    WTERMSIG(rv), WEXITSTATUS(rv));
142 		panic("Going nowhere without my init!");
143 	}
144 
145 	/*
146 	 * MUST abort all other threads before proceeding past here.
147 	 */
148 	PROC_LOCK(p);
149 	while (p->p_flag & P_HADTHREADS) {
150 		/*
151 		 * First check if some other thread got here before us..
152 		 * if so, act apropriatly, (exit or suspend);
153 		 */
154 		thread_suspend_check(0);
155 
156 		/*
157 		 * Kill off the other threads. This requires
158 		 * some co-operation from other parts of the kernel
159 		 * so it may not be instantaneous.  With this state set
160 		 * any thread entering the kernel from userspace will
161 		 * thread_exit() in trap().  Any thread attempting to
162 		 * sleep will return immediately with EINTR or EWOULDBLOCK
163 		 * which will hopefully force them to back out to userland
164 		 * freeing resources as they go.  Any thread attempting
165 		 * to return to userland will thread_exit() from userret().
166 		 * thread_exit() will unsuspend us when the last of the
167 		 * other threads exits.
168 		 * If there is already a thread singler after resumption,
169 		 * calling thread_single will fail; in that case, we just
170 		 * re-check all suspension request, the thread should
171 		 * either be suspended there or exit.
172 		 */
173 		if (! thread_single(SINGLE_EXIT))
174 			break;
175 
176 		/*
177 		 * All other activity in this process is now stopped.
178 		 * Threading support has been turned off.
179 		 */
180 	}
181 	KASSERT(p->p_numthreads == 1,
182 	    ("exit1: proc %p exiting with %d threads", p, p->p_numthreads));
183 	/*
184 	 * Wakeup anyone in procfs' PIOCWAIT.  They should have a hold
185 	 * on our vmspace, so we should block below until they have
186 	 * released their reference to us.  Note that if they have
187 	 * requested S_EXIT stops we will block here until they ack
188 	 * via PIOCCONT.
189 	 */
190 	_STOPEVENT(p, S_EXIT, rv);
191 
192 	/*
193 	 * Note that we are exiting and do another wakeup of anyone in
194 	 * PIOCWAIT in case they aren't listening for S_EXIT stops or
195 	 * decided to wait again after we told them we are exiting.
196 	 */
197 	p->p_flag |= P_WEXIT;
198 	wakeup(&p->p_stype);
199 
200 	/*
201 	 * Wait for any processes that have a hold on our vmspace to
202 	 * release their reference.
203 	 */
204 	while (p->p_lock > 0)
205 		msleep(&p->p_lock, &p->p_mtx, PWAIT, "exithold", 0);
206 
207 	PROC_UNLOCK(p);
208 	/* Drain the limit callout while we don't have the proc locked */
209 	callout_drain(&p->p_limco);
210 
211 #ifdef AUDIT
212 	/*
213 	 * The Sun BSM exit token contains two components: an exit status as
214 	 * passed to exit(), and a return value to indicate what sort of exit
215 	 * it was.  The exit status is WEXITSTATUS(rv), but it's not clear
216 	 * what the return value is.
217 	 */
218 	AUDIT_ARG_EXIT(WEXITSTATUS(rv), 0);
219 	AUDIT_SYSCALL_EXIT(0, td);
220 #endif
221 
222 	/* Are we a task leader? */
223 	if (p == p->p_leader) {
224 		mtx_lock(&ppeers_lock);
225 		q = p->p_peers;
226 		while (q != NULL) {
227 			PROC_LOCK(q);
228 			psignal(q, SIGKILL);
229 			PROC_UNLOCK(q);
230 			q = q->p_peers;
231 		}
232 		while (p->p_peers != NULL)
233 			msleep(p, &ppeers_lock, PWAIT, "exit1", 0);
234 		mtx_unlock(&ppeers_lock);
235 	}
236 
237 	/*
238 	 * Check if any loadable modules need anything done at process exit.
239 	 * E.g. SYSV IPC stuff
240 	 * XXX what if one of these generates an error?
241 	 */
242 	EVENTHANDLER_INVOKE(process_exit, p);
243 
244 	/*
245 	 * If parent is waiting for us to exit or exec,
246 	 * P_PPWAIT is set; we will wakeup the parent below.
247 	 */
248 	PROC_LOCK(p);
249 	stopprofclock(p);
250 	p->p_flag &= ~(P_TRACED | P_PPWAIT);
251 
252 	/*
253 	 * Stop the real interval timer.  If the handler is currently
254 	 * executing, prevent it from rearming itself and let it finish.
255 	 */
256 	if (timevalisset(&p->p_realtimer.it_value) &&
257 	    callout_stop(&p->p_itcallout) == 0) {
258 		timevalclear(&p->p_realtimer.it_interval);
259 		msleep(&p->p_itcallout, &p->p_mtx, PWAIT, "ritwait", 0);
260 		KASSERT(!timevalisset(&p->p_realtimer.it_value),
261 		    ("realtime timer is still armed"));
262 	}
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 	funsetownlst(&p->p_sigiolst);
270 
271 	/*
272 	 * If this process has an nlminfo data area (for lockd), release it
273 	 */
274 	if (nlminfo_release_p != NULL && p->p_nlminfo != NULL)
275 		(*nlminfo_release_p)(p);
276 
277 	/*
278 	 * Close open files and release open-file table.
279 	 * This may block!
280 	 */
281 	fdfree(td);
282 
283 	/*
284 	 * If this thread tickled GEOM, we need to wait for the giggling to
285 	 * stop before we return to userland
286 	 */
287 	if (td->td_pflags & TDP_GEOM)
288 		g_waitidle();
289 
290 	/*
291 	 * Remove ourself from our leader's peer list and wake our leader.
292 	 */
293 	mtx_lock(&ppeers_lock);
294 	if (p->p_leader->p_peers) {
295 		q = p->p_leader;
296 		while (q->p_peers != p)
297 			q = q->p_peers;
298 		q->p_peers = p->p_peers;
299 		wakeup(p->p_leader);
300 	}
301 	mtx_unlock(&ppeers_lock);
302 
303 	vmspace_exit(td);
304 
305 	sx_xlock(&proctree_lock);
306 	if (SESS_LEADER(p)) {
307 		struct session *sp = p->p_session;
308 		struct tty *tp;
309 
310 		/*
311 		 * s_ttyp is not zero'd; we use this to indicate that
312 		 * the session once had a controlling terminal. (for
313 		 * logging and informational purposes)
314 		 */
315 		SESS_LOCK(sp);
316 		ttyvp = sp->s_ttyvp;
317 		tp = sp->s_ttyp;
318 		sp->s_ttyvp = NULL;
319 		sp->s_ttydp = NULL;
320 		sp->s_leader = NULL;
321 		SESS_UNLOCK(sp);
322 
323 		/*
324 		 * Signal foreground pgrp and revoke access to
325 		 * controlling terminal if it has not been revoked
326 		 * already.
327 		 *
328 		 * Because the TTY may have been revoked in the mean
329 		 * time and could already have a new session associated
330 		 * with it, make sure we don't send a SIGHUP to a
331 		 * foreground process group that does not belong to this
332 		 * session.
333 		 */
334 
335 		if (tp != NULL) {
336 			tty_lock(tp);
337 			if (tp->t_session == sp)
338 				tty_signal_pgrp(tp, SIGHUP);
339 			tty_unlock(tp);
340 		}
341 
342 		if (ttyvp != NULL) {
343 			sx_xunlock(&proctree_lock);
344 			if (vn_lock(ttyvp, LK_EXCLUSIVE) == 0) {
345 				VOP_REVOKE(ttyvp, REVOKEALL);
346 				VOP_UNLOCK(ttyvp, 0);
347 			}
348 			sx_xlock(&proctree_lock);
349 		}
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 	tidhash_remove(td);
407 
408 	/*
409 	 * Remove proc from allproc queue and pidhash chain.
410 	 * Place onto zombproc.  Unlink from parent's child list.
411 	 */
412 	sx_xlock(&allproc_lock);
413 	LIST_REMOVE(p, p_list);
414 	LIST_INSERT_HEAD(&zombproc, p, p_list);
415 	LIST_REMOVE(p, p_hash);
416 	sx_xunlock(&allproc_lock);
417 
418 	/*
419 	 * Call machine-dependent code to release any
420 	 * machine-dependent resources other than the address space.
421 	 * The address space is released by "vmspace_exitfree(p)" in
422 	 * vm_waitproc().
423 	 */
424 	cpu_exit(td);
425 
426 	WITNESS_WARN(WARN_PANIC, NULL, "process (pid %d) exiting", p->p_pid);
427 
428 	/*
429 	 * Reparent all of our children to init.
430 	 */
431 	sx_xlock(&proctree_lock);
432 	q = LIST_FIRST(&p->p_children);
433 	if (q != NULL)		/* only need this if any child is S_ZOMB */
434 		wakeup(initproc);
435 	for (; q != NULL; q = nq) {
436 		nq = LIST_NEXT(q, p_sibling);
437 		PROC_LOCK(q);
438 		proc_reparent(q, initproc);
439 		q->p_sigparent = SIGCHLD;
440 		/*
441 		 * Traced processes are killed
442 		 * since their existence means someone is screwing up.
443 		 */
444 		if (q->p_flag & P_TRACED) {
445 			struct thread *temp;
446 
447 			q->p_flag &= ~(P_TRACED | P_STOPPED_TRACE);
448 			FOREACH_THREAD_IN_PROC(q, temp)
449 				temp->td_dbgflags &= ~TDB_SUSPEND;
450 			psignal(q, SIGKILL);
451 		}
452 		PROC_UNLOCK(q);
453 	}
454 
455 	/* Save exit status. */
456 	PROC_LOCK(p);
457 	p->p_xstat = rv;
458 	p->p_xthread = td;
459 
460 	/* Tell the prison that we are gone. */
461 	prison_proc_free(p->p_ucred->cr_prison);
462 
463 #ifdef KDTRACE_HOOKS
464 	/*
465 	 * Tell the DTrace fasttrap provider about the exit if it
466 	 * has declared an interest.
467 	 */
468 	if (dtrace_fasttrap_exit)
469 		dtrace_fasttrap_exit(p);
470 #endif
471 
472 	/*
473 	 * Notify interested parties of our demise.
474 	 */
475 	KNOTE_LOCKED(&p->p_klist, NOTE_EXIT);
476 
477 #ifdef KDTRACE_HOOKS
478 	int reason = CLD_EXITED;
479 	if (WCOREDUMP(rv))
480 		reason = CLD_DUMPED;
481 	else if (WIFSIGNALED(rv))
482 		reason = CLD_KILLED;
483 	SDT_PROBE(proc, kernel, , exit, reason, 0, 0, 0, 0);
484 #endif
485 
486 	/*
487 	 * Just delete all entries in the p_klist. At this point we won't
488 	 * report any more events, and there are nasty race conditions that
489 	 * can beat us if we don't.
490 	 */
491 	knlist_clear(&p->p_klist, 1);
492 
493 	/*
494 	 * Notify parent that we're gone.  If parent has the PS_NOCLDWAIT
495 	 * flag set, or if the handler is set to SIG_IGN, notify process
496 	 * 1 instead (and hope it will handle this situation).
497 	 */
498 	PROC_LOCK(p->p_pptr);
499 	mtx_lock(&p->p_pptr->p_sigacts->ps_mtx);
500 	if (p->p_pptr->p_sigacts->ps_flag & (PS_NOCLDWAIT | PS_CLDSIGIGN)) {
501 		struct proc *pp;
502 
503 		mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx);
504 		pp = p->p_pptr;
505 		PROC_UNLOCK(pp);
506 		proc_reparent(p, initproc);
507 		p->p_sigparent = SIGCHLD;
508 		PROC_LOCK(p->p_pptr);
509 
510 		/*
511 		 * Notify parent, so in case he was wait(2)ing or
512 		 * executing waitpid(2) with our pid, he will
513 		 * continue.
514 		 */
515 		wakeup(pp);
516 	} else
517 		mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx);
518 
519 	if (p->p_pptr == initproc)
520 		psignal(p->p_pptr, SIGCHLD);
521 	else if (p->p_sigparent != 0) {
522 		if (p->p_sigparent == SIGCHLD)
523 			childproc_exited(p);
524 		else	/* LINUX thread */
525 			psignal(p->p_pptr, p->p_sigparent);
526 	}
527 	sx_xunlock(&proctree_lock);
528 
529 	/*
530 	 * The state PRS_ZOMBIE prevents other proesses from sending
531 	 * signal to the process, to avoid memory leak, we free memory
532 	 * for signal queue at the time when the state is set.
533 	 */
534 	sigqueue_flush(&p->p_sigqueue);
535 	sigqueue_flush(&td->td_sigqueue);
536 
537 	/*
538 	 * We have to wait until after acquiring all locks before
539 	 * changing p_state.  We need to avoid all possible context
540 	 * switches (including ones from blocking on a mutex) while
541 	 * marked as a zombie.  We also have to set the zombie state
542 	 * before we release the parent process' proc lock to avoid
543 	 * a lost wakeup.  So, we first call wakeup, then we grab the
544 	 * sched lock, update the state, and release the parent process'
545 	 * proc lock.
546 	 */
547 	wakeup(p->p_pptr);
548 	cv_broadcast(&p->p_pwait);
549 	sched_exit(p->p_pptr, td);
550 	PROC_SLOCK(p);
551 	p->p_state = PRS_ZOMBIE;
552 	PROC_UNLOCK(p->p_pptr);
553 
554 	/*
555 	 * Hopefully no one will try to deliver a signal to the process this
556 	 * late in the game.
557 	 */
558 	knlist_destroy(&p->p_klist);
559 
560 	/*
561 	 * Save our children's rusage information in our exit rusage.
562 	 */
563 	ruadd(&p->p_ru, &p->p_rux, &p->p_stats->p_cru, &p->p_crux);
564 
565 	/*
566 	 * Make sure the scheduler takes this thread out of its tables etc.
567 	 * This will also release this thread's reference to the ucred.
568 	 * Other thread parts to release include pcb bits and such.
569 	 */
570 	thread_exit();
571 }
572 
573 
574 #ifndef _SYS_SYSPROTO_H_
575 struct abort2_args {
576 	char *why;
577 	int nargs;
578 	void **args;
579 };
580 #endif
581 
582 int
583 abort2(struct thread *td, struct abort2_args *uap)
584 {
585 	struct proc *p = td->td_proc;
586 	struct sbuf *sb;
587 	void *uargs[16];
588 	int error, i, sig;
589 
590 	/*
591 	 * Do it right now so we can log either proper call of abort2(), or
592 	 * note, that invalid argument was passed. 512 is big enough to
593 	 * handle 16 arguments' descriptions with additional comments.
594 	 */
595 	sb = sbuf_new(NULL, NULL, 512, SBUF_FIXEDLEN);
596 	sbuf_clear(sb);
597 	sbuf_printf(sb, "%s(pid %d uid %d) aborted: ",
598 	    p->p_comm, p->p_pid, td->td_ucred->cr_uid);
599 	/*
600 	 * Since we can't return from abort2(), send SIGKILL in cases, where
601 	 * abort2() was called improperly
602 	 */
603 	sig = SIGKILL;
604 	/* Prevent from DoSes from user-space. */
605 	if (uap->nargs < 0 || uap->nargs > 16)
606 		goto out;
607 	if (uap->nargs > 0) {
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 	/*
615 	 * Limit size of 'reason' string to 128. Will fit even when
616 	 * maximal number of arguments was chosen to be logged.
617 	 */
618 	if (uap->why != NULL) {
619 		error = sbuf_copyin(sb, uap->why, 128);
620 		if (error < 0)
621 			goto out;
622 	} else {
623 		sbuf_printf(sb, "(null)");
624 	}
625 	if (uap->nargs > 0) {
626 		sbuf_printf(sb, "(");
627 		for (i = 0;i < uap->nargs; i++)
628 			sbuf_printf(sb, "%s%p", i == 0 ? "" : ", ", uargs[i]);
629 		sbuf_printf(sb, ")");
630 	}
631 	/*
632 	 * Final stage: arguments were proper, string has been
633 	 * successfully copied from userspace, and copying pointers
634 	 * from user-space succeed.
635 	 */
636 	sig = SIGABRT;
637 out:
638 	if (sig == SIGKILL) {
639 		sbuf_trim(sb);
640 		sbuf_printf(sb, " (Reason text inaccessible)");
641 	}
642 	sbuf_cat(sb, "\n");
643 	sbuf_finish(sb);
644 	log(LOG_INFO, "%s", sbuf_data(sb));
645 	sbuf_delete(sb);
646 	exit1(td, W_EXITCODE(0, sig));
647 	return (0);
648 }
649 
650 
651 #ifdef COMPAT_43
652 /*
653  * The dirty work is handled by kern_wait().
654  */
655 int
656 owait(struct thread *td, struct owait_args *uap __unused)
657 {
658 	int error, status;
659 
660 	error = kern_wait(td, WAIT_ANY, &status, 0, NULL);
661 	if (error == 0)
662 		td->td_retval[1] = status;
663 	return (error);
664 }
665 #endif /* COMPAT_43 */
666 
667 /*
668  * The dirty work is handled by kern_wait().
669  */
670 int
671 wait4(struct thread *td, struct wait_args *uap)
672 {
673 	struct rusage ru, *rup;
674 	int error, status;
675 
676 	if (uap->rusage != NULL)
677 		rup = &ru;
678 	else
679 		rup = NULL;
680 	error = kern_wait(td, uap->pid, &status, uap->options, rup);
681 	if (uap->status != NULL && error == 0)
682 		error = copyout(&status, uap->status, sizeof(status));
683 	if (uap->rusage != NULL && error == 0)
684 		error = copyout(&ru, uap->rusage, sizeof(struct rusage));
685 	return (error);
686 }
687 
688 /*
689  * Reap the remains of a zombie process and optionally return status and
690  * rusage.  Asserts and will release both the proctree_lock and the process
691  * lock as part of its work.
692  */
693 static void
694 proc_reap(struct thread *td, struct proc *p, int *status, int options,
695     struct rusage *rusage)
696 {
697 	struct proc *q, *t;
698 
699 	sx_assert(&proctree_lock, SA_XLOCKED);
700 	PROC_LOCK_ASSERT(p, MA_OWNED);
701 	PROC_SLOCK_ASSERT(p, MA_OWNED);
702 	KASSERT(p->p_state == PRS_ZOMBIE, ("proc_reap: !PRS_ZOMBIE"));
703 
704 	q = td->td_proc;
705 	if (rusage) {
706 		*rusage = p->p_ru;
707 		calcru(p, &rusage->ru_utime, &rusage->ru_stime);
708 	}
709 	PROC_SUNLOCK(p);
710 	td->td_retval[0] = p->p_pid;
711 	if (status)
712 		*status = p->p_xstat;	/* convert to int */
713 	if (options & WNOWAIT) {
714 		/*
715 		 *  Only poll, returning the status.  Caller does not wish to
716 		 * release the proc struct just yet.
717 		 */
718 		PROC_UNLOCK(p);
719 		sx_xunlock(&proctree_lock);
720 		return;
721 	}
722 
723 	PROC_LOCK(q);
724 	sigqueue_take(p->p_ksi);
725 	PROC_UNLOCK(q);
726 	PROC_UNLOCK(p);
727 
728 	/*
729 	 * If we got the child via a ptrace 'attach', we need to give it back
730 	 * to the old parent.
731 	 */
732 	if (p->p_oppid && (t = pfind(p->p_oppid)) != NULL) {
733 		PROC_LOCK(p);
734 		p->p_oppid = 0;
735 		proc_reparent(p, t);
736 		PROC_UNLOCK(p);
737 		pksignal(t, SIGCHLD, p->p_ksi);
738 		wakeup(t);
739 		cv_broadcast(&p->p_pwait);
740 		PROC_UNLOCK(t);
741 		sx_xunlock(&proctree_lock);
742 		return;
743 	}
744 
745 	/*
746 	 * Remove other references to this process to ensure we have an
747 	 * exclusive reference.
748 	 */
749 	sx_xlock(&allproc_lock);
750 	LIST_REMOVE(p, p_list);	/* off zombproc */
751 	sx_xunlock(&allproc_lock);
752 	LIST_REMOVE(p, p_sibling);
753 	leavepgrp(p);
754 	sx_xunlock(&proctree_lock);
755 
756 	/*
757 	 * As a side effect of this lock, we know that all other writes to
758 	 * this proc are visible now, so no more locking is needed for p.
759 	 */
760 	PROC_LOCK(p);
761 	p->p_xstat = 0;		/* XXX: why? */
762 	PROC_UNLOCK(p);
763 	PROC_LOCK(q);
764 	ruadd(&q->p_stats->p_cru, &q->p_crux, &p->p_ru, &p->p_rux);
765 	PROC_UNLOCK(q);
766 
767 	/*
768 	 * Decrement the count of procs running with this uid.
769 	 */
770 	(void)chgproccnt(p->p_ucred->cr_ruidinfo, -1, 0);
771 
772 	/*
773 	 * Free credentials, arguments, and sigacts.
774 	 */
775 	crfree(p->p_ucred);
776 	p->p_ucred = NULL;
777 	pargs_drop(p->p_args);
778 	p->p_args = NULL;
779 	sigacts_free(p->p_sigacts);
780 	p->p_sigacts = NULL;
781 
782 	/*
783 	 * Do any thread-system specific cleanups.
784 	 */
785 	thread_wait(p);
786 
787 	/*
788 	 * Give vm and machine-dependent layer a chance to free anything that
789 	 * cpu_exit couldn't release while still running in process context.
790 	 */
791 	vm_waitproc(p);
792 #ifdef MAC
793 	mac_proc_destroy(p);
794 #endif
795 	KASSERT(FIRST_THREAD_IN_PROC(p),
796 	    ("proc_reap: no residual thread!"));
797 	uma_zfree(proc_zone, p);
798 	sx_xlock(&allproc_lock);
799 	nprocs--;
800 	sx_xunlock(&allproc_lock);
801 }
802 
803 int
804 kern_wait(struct thread *td, pid_t pid, int *status, int options,
805     struct rusage *rusage)
806 {
807 	struct proc *p, *q;
808 	int error, nfound;
809 
810 	AUDIT_ARG_PID(pid);
811 	AUDIT_ARG_VALUE(options);
812 
813 	q = td->td_proc;
814 	if (pid == 0) {
815 		PROC_LOCK(q);
816 		pid = -q->p_pgid;
817 		PROC_UNLOCK(q);
818 	}
819 	if (options &~ (WUNTRACED|WNOHANG|WCONTINUED|WNOWAIT|WLINUXCLONE))
820 		return (EINVAL);
821 loop:
822 	if (q->p_flag & P_STATCHILD) {
823 		PROC_LOCK(q);
824 		q->p_flag &= ~P_STATCHILD;
825 		PROC_UNLOCK(q);
826 	}
827 	nfound = 0;
828 	sx_xlock(&proctree_lock);
829 	LIST_FOREACH(p, &q->p_children, p_sibling) {
830 		PROC_LOCK(p);
831 		if (pid != WAIT_ANY &&
832 		    p->p_pid != pid && p->p_pgid != -pid) {
833 			PROC_UNLOCK(p);
834 			continue;
835 		}
836 		if (p_canwait(td, p)) {
837 			PROC_UNLOCK(p);
838 			continue;
839 		}
840 
841 		/*
842 		 * This special case handles a kthread spawned by linux_clone
843 		 * (see linux_misc.c).  The linux_wait4 and linux_waitpid
844 		 * functions need to be able to distinguish between waiting
845 		 * on a process and waiting on a thread.  It is a thread if
846 		 * p_sigparent is not SIGCHLD, and the WLINUXCLONE option
847 		 * signifies we want to wait for threads and not processes.
848 		 */
849 		if ((p->p_sigparent != SIGCHLD) ^
850 		    ((options & WLINUXCLONE) != 0)) {
851 			PROC_UNLOCK(p);
852 			continue;
853 		}
854 
855 		nfound++;
856 		PROC_SLOCK(p);
857 		if (p->p_state == PRS_ZOMBIE) {
858 			proc_reap(td, p, status, options, rusage);
859 			return (0);
860 		}
861 		if ((p->p_flag & P_STOPPED_SIG) &&
862 		    (p->p_suspcount == p->p_numthreads) &&
863 		    (p->p_flag & P_WAITED) == 0 &&
864 		    (p->p_flag & P_TRACED || options & WUNTRACED)) {
865 			PROC_SUNLOCK(p);
866 			p->p_flag |= P_WAITED;
867 			sx_xunlock(&proctree_lock);
868 			td->td_retval[0] = p->p_pid;
869 			if (status)
870 				*status = W_STOPCODE(p->p_xstat);
871 
872 			PROC_LOCK(q);
873 			sigqueue_take(p->p_ksi);
874 			PROC_UNLOCK(q);
875 			PROC_UNLOCK(p);
876 
877 			return (0);
878 		}
879 		PROC_SUNLOCK(p);
880 		if (options & WCONTINUED && (p->p_flag & P_CONTINUED)) {
881 			sx_xunlock(&proctree_lock);
882 			td->td_retval[0] = p->p_pid;
883 			p->p_flag &= ~P_CONTINUED;
884 
885 			PROC_LOCK(q);
886 			sigqueue_take(p->p_ksi);
887 			PROC_UNLOCK(q);
888 			PROC_UNLOCK(p);
889 
890 			if (status)
891 				*status = SIGCONT;
892 			return (0);
893 		}
894 		PROC_UNLOCK(p);
895 	}
896 	if (nfound == 0) {
897 		sx_xunlock(&proctree_lock);
898 		return (ECHILD);
899 	}
900 	if (options & WNOHANG) {
901 		sx_xunlock(&proctree_lock);
902 		td->td_retval[0] = 0;
903 		return (0);
904 	}
905 	PROC_LOCK(q);
906 	sx_xunlock(&proctree_lock);
907 	if (q->p_flag & P_STATCHILD) {
908 		q->p_flag &= ~P_STATCHILD;
909 		error = 0;
910 	} else
911 		error = msleep(q, &q->p_mtx, PWAIT | PCATCH, "wait", 0);
912 	PROC_UNLOCK(q);
913 	if (error)
914 		return (error);
915 	goto loop;
916 }
917 
918 /*
919  * Make process 'parent' the new parent of process 'child'.
920  * Must be called with an exclusive hold of proctree lock.
921  */
922 void
923 proc_reparent(struct proc *child, struct proc *parent)
924 {
925 
926 	sx_assert(&proctree_lock, SX_XLOCKED);
927 	PROC_LOCK_ASSERT(child, MA_OWNED);
928 	if (child->p_pptr == parent)
929 		return;
930 
931 	PROC_LOCK(child->p_pptr);
932 	sigqueue_take(child->p_ksi);
933 	PROC_UNLOCK(child->p_pptr);
934 	LIST_REMOVE(child, p_sibling);
935 	LIST_INSERT_HEAD(&parent->p_children, child, p_sibling);
936 	child->p_pptr = parent;
937 }
938