xref: /freebsd/sys/kern/kern_exit.c (revision 6356dba0b403daa023dec24559ab1f8e602e4f14)
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/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);
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 	if (p == initproc) {
135 		printf("init died (signal %d, exit %d)\n",
136 		    WTERMSIG(rv), WEXITSTATUS(rv));
137 		panic("Going nowhere without my init!");
138 	}
139 
140 	/*
141 	 * MUST abort all other threads before proceeding past here.
142 	 */
143 	PROC_LOCK(p);
144 	while (p->p_flag & P_HADTHREADS) {
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 			break;
170 
171 		/*
172 		 * All other activity in this process is now stopped.
173 		 * Threading support has been turned off.
174 		 */
175 	}
176 	KASSERT(p->p_numthreads == 1,
177 	    ("exit1: proc %p exiting with %d threads", p, p->p_numthreads));
178 	/*
179 	 * Wakeup anyone in procfs' PIOCWAIT.  They should have a hold
180 	 * on our vmspace, so we should block below until they have
181 	 * released their reference to us.  Note that if they have
182 	 * requested S_EXIT stops we will block here until they ack
183 	 * via PIOCCONT.
184 	 */
185 	_STOPEVENT(p, S_EXIT, rv);
186 
187 	/*
188 	 * Note that we are exiting and do another wakeup of anyone in
189 	 * PIOCWAIT in case they aren't listening for S_EXIT stops or
190 	 * decided to wait again after we told them we are exiting.
191 	 */
192 	p->p_flag |= P_WEXIT;
193 	wakeup(&p->p_stype);
194 
195 	/*
196 	 * Wait for any processes that have a hold on our vmspace to
197 	 * release their reference.
198 	 */
199 	while (p->p_lock > 0)
200 		msleep(&p->p_lock, &p->p_mtx, PWAIT, "exithold", 0);
201 
202 	PROC_UNLOCK(p);
203 	/* Drain the limit callout while we don't have the proc locked */
204 	callout_drain(&p->p_limco);
205 
206 #ifdef AUDIT
207 	/*
208 	 * The Sun BSM exit token contains two components: an exit status as
209 	 * passed to exit(), and a return value to indicate what sort of exit
210 	 * it was.  The exit status is WEXITSTATUS(rv), but it's not clear
211 	 * what the return value is.
212 	 */
213 	AUDIT_ARG(exit, WEXITSTATUS(rv), 0);
214 	AUDIT_SYSCALL_EXIT(0, td);
215 #endif
216 
217 	/* Are we a task leader? */
218 	if (p == p->p_leader) {
219 		mtx_lock(&ppeers_lock);
220 		q = p->p_peers;
221 		while (q != NULL) {
222 			PROC_LOCK(q);
223 			psignal(q, SIGKILL);
224 			PROC_UNLOCK(q);
225 			q = q->p_peers;
226 		}
227 		while (p->p_peers != NULL)
228 			msleep(p, &ppeers_lock, PWAIT, "exit1", 0);
229 		mtx_unlock(&ppeers_lock);
230 	}
231 
232 	/*
233 	 * Check if any loadable modules need anything done at process exit.
234 	 * E.g. SYSV IPC stuff
235 	 * XXX what if one of these generates an error?
236 	 */
237 	EVENTHANDLER_INVOKE(process_exit, p);
238 
239 	/*
240 	 * If parent is waiting for us to exit or exec,
241 	 * P_PPWAIT is set; we will wakeup the parent below.
242 	 */
243 	PROC_LOCK(p);
244 	stopprofclock(p);
245 	p->p_flag &= ~(P_TRACED | P_PPWAIT);
246 
247 	/*
248 	 * Stop the real interval timer.  If the handler is currently
249 	 * executing, prevent it from rearming itself and let it finish.
250 	 */
251 	if (timevalisset(&p->p_realtimer.it_value) &&
252 	    callout_stop(&p->p_itcallout) == 0) {
253 		timevalclear(&p->p_realtimer.it_interval);
254 		msleep(&p->p_itcallout, &p->p_mtx, PWAIT, "ritwait", 0);
255 		KASSERT(!timevalisset(&p->p_realtimer.it_value),
256 		    ("realtime timer is still armed"));
257 	}
258 	PROC_UNLOCK(p);
259 
260 	/*
261 	 * Reset any sigio structures pointing to us as a result of
262 	 * F_SETOWN with our pid.
263 	 */
264 	funsetownlst(&p->p_sigiolst);
265 
266 	/*
267 	 * If this process has an nlminfo data area (for lockd), release it
268 	 */
269 	if (nlminfo_release_p != NULL && p->p_nlminfo != NULL)
270 		(*nlminfo_release_p)(p);
271 
272 	/*
273 	 * Close open files and release open-file table.
274 	 * This may block!
275 	 */
276 	fdfree(td);
277 
278 	/*
279 	 * If this thread tickled GEOM, we need to wait for the giggling to
280 	 * stop before we return to userland
281 	 */
282 	if (td->td_pflags & TDP_GEOM)
283 		g_waitidle();
284 
285 	/*
286 	 * Remove ourself from our leader's peer list and wake our leader.
287 	 */
288 	mtx_lock(&ppeers_lock);
289 	if (p->p_leader->p_peers) {
290 		q = p->p_leader;
291 		while (q->p_peers != p)
292 			q = q->p_peers;
293 		q->p_peers = p->p_peers;
294 		wakeup(p->p_leader);
295 	}
296 	mtx_unlock(&ppeers_lock);
297 
298 	vmspace_exit(td);
299 
300 	sx_xlock(&proctree_lock);
301 	if (SESS_LEADER(p)) {
302 		struct session *sp;
303 
304 		sp = p->p_session;
305 
306 		SESS_LOCK(sp);
307 		ttyvp = sp->s_ttyvp;
308 		sp->s_ttyvp = NULL;
309 		SESS_UNLOCK(sp);
310 
311 		if (ttyvp != NULL) {
312 			/*
313 			 * Controlling process.
314 			 * Signal foreground pgrp and revoke access to
315 			 * controlling terminal.
316 			 *
317 			 * There is no need to drain the terminal here,
318 			 * because this will be done on revocation.
319 			 */
320 			if (sp->s_ttyp != NULL) {
321 				struct tty *tp = sp->s_ttyp;
322 
323 				tty_lock(tp);
324 				tty_signal_pgrp(tp, SIGHUP);
325 				tty_unlock(tp);
326 
327 				/*
328 				 * The tty could have been revoked
329 				 * if we blocked.
330 				 */
331 				if (ttyvp->v_type != VBAD) {
332 					sx_xunlock(&proctree_lock);
333 					VOP_LOCK(ttyvp, LK_EXCLUSIVE);
334 					VOP_REVOKE(ttyvp, REVOKEALL);
335 					VOP_UNLOCK(ttyvp, 0);
336 					sx_xlock(&proctree_lock);
337 				}
338 			}
339 			/*
340 			 * s_ttyp is not zero'd; we use this to indicate that
341 			 * the session once had a controlling terminal.
342 			 * (for logging and informational purposes)
343 			 */
344 		}
345 		SESS_LOCK(p->p_session);
346 		sp->s_leader = NULL;
347 		SESS_UNLOCK(p->p_session);
348 	}
349 	fixjobc(p, p->p_pgrp, 0);
350 	sx_xunlock(&proctree_lock);
351 	(void)acct_process(td);
352 
353 	/* Release the TTY now we've unlocked everything. */
354 	if (ttyvp != NULL)
355 		vrele(ttyvp);
356 #ifdef KTRACE
357 	/*
358 	 * Disable tracing, then drain any pending records and release
359 	 * the trace file.
360 	 */
361 	if (p->p_traceflag != 0) {
362 		PROC_LOCK(p);
363 		mtx_lock(&ktrace_mtx);
364 		p->p_traceflag = 0;
365 		mtx_unlock(&ktrace_mtx);
366 		PROC_UNLOCK(p);
367 		ktrprocexit(td);
368 		PROC_LOCK(p);
369 		mtx_lock(&ktrace_mtx);
370 		tracevp = p->p_tracevp;
371 		p->p_tracevp = NULL;
372 		tracecred = p->p_tracecred;
373 		p->p_tracecred = NULL;
374 		mtx_unlock(&ktrace_mtx);
375 		PROC_UNLOCK(p);
376 		if (tracevp != NULL) {
377 			locked = VFS_LOCK_GIANT(tracevp->v_mount);
378 			vrele(tracevp);
379 			VFS_UNLOCK_GIANT(locked);
380 		}
381 		if (tracecred != NULL)
382 			crfree(tracecred);
383 	}
384 #endif
385 	/*
386 	 * Release reference to text vnode
387 	 */
388 	if ((vtmp = p->p_textvp) != NULL) {
389 		p->p_textvp = NULL;
390 		locked = VFS_LOCK_GIANT(vtmp->v_mount);
391 		vrele(vtmp);
392 		VFS_UNLOCK_GIANT(locked);
393 	}
394 
395 	/*
396 	 * Release our limits structure.
397 	 */
398 	PROC_LOCK(p);
399 	plim = p->p_limit;
400 	p->p_limit = NULL;
401 	PROC_UNLOCK(p);
402 	lim_free(plim);
403 
404 	/*
405 	 * Remove proc from allproc queue and pidhash chain.
406 	 * Place onto zombproc.  Unlink from parent's child list.
407 	 */
408 	sx_xlock(&allproc_lock);
409 	LIST_REMOVE(p, p_list);
410 	LIST_INSERT_HEAD(&zombproc, p, p_list);
411 	LIST_REMOVE(p, p_hash);
412 	sx_xunlock(&allproc_lock);
413 
414 	/*
415 	 * Call machine-dependent code to release any
416 	 * machine-dependent resources other than the address space.
417 	 * The address space is released by "vmspace_exitfree(p)" in
418 	 * vm_waitproc().
419 	 */
420 	cpu_exit(td);
421 
422 	WITNESS_WARN(WARN_PANIC, NULL, "process (pid %d) exiting", p->p_pid);
423 
424 	/*
425 	 * Reparent all of our children to init.
426 	 */
427 	sx_xlock(&proctree_lock);
428 	q = LIST_FIRST(&p->p_children);
429 	if (q != NULL)		/* only need this if any child is S_ZOMB */
430 		wakeup(initproc);
431 	for (; q != NULL; q = nq) {
432 		nq = LIST_NEXT(q, p_sibling);
433 		PROC_LOCK(q);
434 		proc_reparent(q, initproc);
435 		q->p_sigparent = SIGCHLD;
436 		/*
437 		 * Traced processes are killed
438 		 * since their existence means someone is screwing up.
439 		 */
440 		if (q->p_flag & P_TRACED) {
441 			q->p_flag &= ~(P_TRACED | P_STOPPED_TRACE);
442 			psignal(q, SIGKILL);
443 		}
444 		PROC_UNLOCK(q);
445 	}
446 
447 	/* Save exit status. */
448 	PROC_LOCK(p);
449 	p->p_xstat = rv;
450 	p->p_xthread = td;
451 
452 #ifdef KDTRACE_HOOKS
453 	/*
454 	 * Tell the DTrace fasttrap provider about the exit if it
455 	 * has declared an interest.
456 	 */
457 	if (dtrace_fasttrap_exit)
458 		dtrace_fasttrap_exit(p);
459 #endif
460 
461 	/*
462 	 * Notify interested parties of our demise.
463 	 */
464 	KNOTE_LOCKED(&p->p_klist, NOTE_EXIT);
465 
466 #ifdef KDTRACE_HOOKS
467 	int reason = CLD_EXITED;
468 	if (WCOREDUMP(rv))
469 		reason = CLD_DUMPED;
470 	else if (WIFSIGNALED(rv))
471 		reason = CLD_KILLED;
472 	SDT_PROBE(proc, kernel, , exit, reason, 0, 0, 0, 0);
473 #endif
474 
475 	/*
476 	 * Just delete all entries in the p_klist. At this point we won't
477 	 * report any more events, and there are nasty race conditions that
478 	 * can beat us if we don't.
479 	 */
480 	knlist_clear(&p->p_klist, 1);
481 
482 	/*
483 	 * Notify parent that we're gone.  If parent has the PS_NOCLDWAIT
484 	 * flag set, or if the handler is set to SIG_IGN, notify process
485 	 * 1 instead (and hope it will handle this situation).
486 	 */
487 	PROC_LOCK(p->p_pptr);
488 	mtx_lock(&p->p_pptr->p_sigacts->ps_mtx);
489 	if (p->p_pptr->p_sigacts->ps_flag & (PS_NOCLDWAIT | PS_CLDSIGIGN)) {
490 		struct proc *pp;
491 
492 		mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx);
493 		pp = p->p_pptr;
494 		PROC_UNLOCK(pp);
495 		proc_reparent(p, initproc);
496 		p->p_sigparent = SIGCHLD;
497 		PROC_LOCK(p->p_pptr);
498 		/*
499 		 * If this was the last child of our parent, notify
500 		 * parent, so in case he was wait(2)ing, he will
501 		 * continue.
502 		 */
503 		if (LIST_EMPTY(&pp->p_children))
504 			wakeup(pp);
505 	} else
506 		mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx);
507 
508 	if (p->p_pptr == initproc)
509 		psignal(p->p_pptr, SIGCHLD);
510 	else if (p->p_sigparent != 0) {
511 		if (p->p_sigparent == SIGCHLD)
512 			childproc_exited(p);
513 		else	/* LINUX thread */
514 			psignal(p->p_pptr, p->p_sigparent);
515 	}
516 	sx_xunlock(&proctree_lock);
517 
518 	/*
519 	 * The state PRS_ZOMBIE prevents other proesses from sending
520 	 * signal to the process, to avoid memory leak, we free memory
521 	 * for signal queue at the time when the state is set.
522 	 */
523 	sigqueue_flush(&p->p_sigqueue);
524 	sigqueue_flush(&td->td_sigqueue);
525 
526 	/*
527 	 * We have to wait until after acquiring all locks before
528 	 * changing p_state.  We need to avoid all possible context
529 	 * switches (including ones from blocking on a mutex) while
530 	 * marked as a zombie.  We also have to set the zombie state
531 	 * before we release the parent process' proc lock to avoid
532 	 * a lost wakeup.  So, we first call wakeup, then we grab the
533 	 * sched lock, update the state, and release the parent process'
534 	 * proc lock.
535 	 */
536 	wakeup(p->p_pptr);
537 	sched_exit(p->p_pptr, td);
538 	PROC_SLOCK(p);
539 	p->p_state = PRS_ZOMBIE;
540 	PROC_UNLOCK(p->p_pptr);
541 
542 	/*
543 	 * Hopefully no one will try to deliver a signal to the process this
544 	 * late in the game.
545 	 */
546 	knlist_destroy(&p->p_klist);
547 
548 	/*
549 	 * Save our children's rusage information in our exit rusage.
550 	 */
551 	ruadd(&p->p_ru, &p->p_rux, &p->p_stats->p_cru, &p->p_crux);
552 
553 	/*
554 	 * Make sure the scheduler takes this thread out of its tables etc.
555 	 * This will also release this thread's reference to the ucred.
556 	 * Other thread parts to release include pcb bits and such.
557 	 */
558 	thread_exit();
559 }
560 
561 
562 #ifndef _SYS_SYSPROTO_H_
563 struct abort2_args {
564 	char *why;
565 	int nargs;
566 	void **args;
567 };
568 #endif
569 
570 int
571 abort2(struct thread *td, struct abort2_args *uap)
572 {
573 	struct proc *p = td->td_proc;
574 	struct sbuf *sb;
575 	void *uargs[16];
576 	int error, i, sig;
577 
578 	error = 0;	/* satisfy compiler */
579 
580 	/*
581 	 * Do it right now so we can log either proper call of abort2(), or
582 	 * note, that invalid argument was passed. 512 is big enough to
583 	 * handle 16 arguments' descriptions with additional comments.
584 	 */
585 	sb = sbuf_new(NULL, NULL, 512, SBUF_FIXEDLEN);
586 	sbuf_clear(sb);
587 	sbuf_printf(sb, "%s(pid %d uid %d) aborted: ",
588 	    p->p_comm, p->p_pid, td->td_ucred->cr_uid);
589 	/*
590 	 * Since we can't return from abort2(), send SIGKILL in cases, where
591 	 * abort2() was called improperly
592 	 */
593 	sig = SIGKILL;
594 	/* Prevent from DoSes from user-space. */
595 	if (uap->nargs < 0 || uap->nargs > 16)
596 		goto out;
597 	if (uap->nargs > 0) {
598 		if (uap->args == NULL)
599 			goto out;
600 		error = copyin(uap->args, uargs, uap->nargs * sizeof(void *));
601 		if (error != 0)
602 			goto out;
603 	}
604 	/*
605 	 * Limit size of 'reason' string to 128. Will fit even when
606 	 * maximal number of arguments was chosen to be logged.
607 	 */
608 	if (uap->why != NULL) {
609 		error = sbuf_copyin(sb, uap->why, 128);
610 		if (error < 0)
611 			goto out;
612 	} else {
613 		sbuf_printf(sb, "(null)");
614 	}
615 	if (uap->nargs > 0) {
616 		sbuf_printf(sb, "(");
617 		for (i = 0;i < uap->nargs; i++)
618 			sbuf_printf(sb, "%s%p", i == 0 ? "" : ", ", uargs[i]);
619 		sbuf_printf(sb, ")");
620 	}
621 	/*
622 	 * Final stage: arguments were proper, string has been
623 	 * successfully copied from userspace, and copying pointers
624 	 * from user-space succeed.
625 	 */
626 	sig = SIGABRT;
627 out:
628 	if (sig == SIGKILL) {
629 		sbuf_trim(sb);
630 		sbuf_printf(sb, " (Reason text inaccessible)");
631 	}
632 	sbuf_cat(sb, "\n");
633 	sbuf_finish(sb);
634 	log(LOG_INFO, "%s", sbuf_data(sb));
635 	sbuf_delete(sb);
636 	exit1(td, W_EXITCODE(0, sig));
637 	return (0);
638 }
639 
640 
641 #ifdef COMPAT_43
642 /*
643  * The dirty work is handled by kern_wait().
644  */
645 int
646 owait(struct thread *td, struct owait_args *uap __unused)
647 {
648 	int error, status;
649 
650 	error = kern_wait(td, WAIT_ANY, &status, 0, NULL);
651 	if (error == 0)
652 		td->td_retval[1] = status;
653 	return (error);
654 }
655 #endif /* COMPAT_43 */
656 
657 /*
658  * The dirty work is handled by kern_wait().
659  */
660 int
661 wait4(struct thread *td, struct wait_args *uap)
662 {
663 	struct rusage ru, *rup;
664 	int error, status;
665 
666 	if (uap->rusage != NULL)
667 		rup = &ru;
668 	else
669 		rup = NULL;
670 	error = kern_wait(td, uap->pid, &status, uap->options, rup);
671 	if (uap->status != NULL && error == 0)
672 		error = copyout(&status, uap->status, sizeof(status));
673 	if (uap->rusage != NULL && error == 0)
674 		error = copyout(&ru, uap->rusage, sizeof(struct rusage));
675 	return (error);
676 }
677 
678 int
679 kern_wait(struct thread *td, pid_t pid, int *status, int options,
680     struct rusage *rusage)
681 {
682 	struct proc *p, *q, *t;
683 	int error, nfound;
684 
685 	AUDIT_ARG(pid, pid);
686 
687 	q = td->td_proc;
688 	if (pid == 0) {
689 		PROC_LOCK(q);
690 		pid = -q->p_pgid;
691 		PROC_UNLOCK(q);
692 	}
693 	if (options &~ (WUNTRACED|WNOHANG|WCONTINUED|WLINUXCLONE))
694 		return (EINVAL);
695 loop:
696 	if (q->p_flag & P_STATCHILD) {
697 		PROC_LOCK(q);
698 		q->p_flag &= ~P_STATCHILD;
699 		PROC_UNLOCK(q);
700 	}
701 	nfound = 0;
702 	sx_xlock(&proctree_lock);
703 	LIST_FOREACH(p, &q->p_children, p_sibling) {
704 		PROC_LOCK(p);
705 		if (pid != WAIT_ANY &&
706 		    p->p_pid != pid && p->p_pgid != -pid) {
707 			PROC_UNLOCK(p);
708 			continue;
709 		}
710 		if (p_canwait(td, p)) {
711 			PROC_UNLOCK(p);
712 			continue;
713 		}
714 
715 		/*
716 		 * This special case handles a kthread spawned by linux_clone
717 		 * (see linux_misc.c).  The linux_wait4 and linux_waitpid
718 		 * functions need to be able to distinguish between waiting
719 		 * on a process and waiting on a thread.  It is a thread if
720 		 * p_sigparent is not SIGCHLD, and the WLINUXCLONE option
721 		 * signifies we want to wait for threads and not processes.
722 		 */
723 		if ((p->p_sigparent != SIGCHLD) ^
724 		    ((options & WLINUXCLONE) != 0)) {
725 			PROC_UNLOCK(p);
726 			continue;
727 		}
728 
729 		nfound++;
730 		PROC_SLOCK(p);
731 		if (p->p_state == PRS_ZOMBIE) {
732 			if (rusage) {
733 				*rusage = p->p_ru;
734 				calcru(p, &rusage->ru_utime, &rusage->ru_stime);
735 			}
736 			PROC_SUNLOCK(p);
737 			td->td_retval[0] = p->p_pid;
738 			if (status)
739 				*status = p->p_xstat;	/* convert to int */
740 			PROC_LOCK(q);
741 			sigqueue_take(p->p_ksi);
742 			PROC_UNLOCK(q);
743 
744 			/*
745 			 * If we got the child via a ptrace 'attach',
746 			 * we need to give it back to the old parent.
747 			 */
748 			PROC_UNLOCK(p);
749 			if (p->p_oppid && (t = pfind(p->p_oppid)) != NULL) {
750 				PROC_LOCK(p);
751 				p->p_oppid = 0;
752 				proc_reparent(p, t);
753 				PROC_UNLOCK(p);
754 				tdsignal(t, NULL, SIGCHLD, p->p_ksi);
755 				wakeup(t);
756 				PROC_UNLOCK(t);
757 				sx_xunlock(&proctree_lock);
758 				return (0);
759 			}
760 
761 			/*
762 			 * Remove other references to this process to ensure
763 			 * we have an exclusive reference.
764 			 */
765 			sx_xlock(&allproc_lock);
766 			LIST_REMOVE(p, p_list);	/* off zombproc */
767 			sx_xunlock(&allproc_lock);
768 			LIST_REMOVE(p, p_sibling);
769 			leavepgrp(p);
770 			sx_xunlock(&proctree_lock);
771 
772 			/*
773 			 * As a side effect of this lock, we know that
774 			 * all other writes to this proc are visible now, so
775 			 * no more locking is needed for p.
776 			 */
777 			PROC_LOCK(p);
778 			p->p_xstat = 0;		/* XXX: why? */
779 			PROC_UNLOCK(p);
780 			PROC_LOCK(q);
781 			ruadd(&q->p_stats->p_cru, &q->p_crux, &p->p_ru,
782 			    &p->p_rux);
783 			PROC_UNLOCK(q);
784 
785 			/*
786 			 * Decrement the count of procs running with this uid.
787 			 */
788 			(void)chgproccnt(p->p_ucred->cr_ruidinfo, -1, 0);
789 
790 			/*
791 			 * Free credentials, arguments, and sigacts.
792 			 */
793 			crfree(p->p_ucred);
794 			p->p_ucred = NULL;
795 			pargs_drop(p->p_args);
796 			p->p_args = NULL;
797 			sigacts_free(p->p_sigacts);
798 			p->p_sigacts = NULL;
799 
800 			/*
801 			 * Do any thread-system specific cleanups.
802 			 */
803 			thread_wait(p);
804 
805 			/*
806 			 * Give vm and machine-dependent layer a chance
807 			 * to free anything that cpu_exit couldn't
808 			 * release while still running in process context.
809 			 */
810 			vm_waitproc(p);
811 #ifdef MAC
812 			mac_proc_destroy(p);
813 #endif
814 			KASSERT(FIRST_THREAD_IN_PROC(p),
815 			    ("kern_wait: no residual thread!"));
816 			uma_zfree(proc_zone, p);
817 			sx_xlock(&allproc_lock);
818 			nprocs--;
819 			sx_xunlock(&allproc_lock);
820 			return (0);
821 		}
822 		if ((p->p_flag & P_STOPPED_SIG) &&
823 		    (p->p_suspcount == p->p_numthreads) &&
824 		    (p->p_flag & P_WAITED) == 0 &&
825 		    (p->p_flag & P_TRACED || options & WUNTRACED)) {
826 			PROC_SUNLOCK(p);
827 			p->p_flag |= P_WAITED;
828 			sx_xunlock(&proctree_lock);
829 			td->td_retval[0] = p->p_pid;
830 			if (status)
831 				*status = W_STOPCODE(p->p_xstat);
832 
833 			PROC_LOCK(q);
834 			sigqueue_take(p->p_ksi);
835 			PROC_UNLOCK(q);
836 			PROC_UNLOCK(p);
837 
838 			return (0);
839 		}
840 		PROC_SUNLOCK(p);
841 		if (options & WCONTINUED && (p->p_flag & P_CONTINUED)) {
842 			sx_xunlock(&proctree_lock);
843 			td->td_retval[0] = p->p_pid;
844 			p->p_flag &= ~P_CONTINUED;
845 
846 			PROC_LOCK(q);
847 			sigqueue_take(p->p_ksi);
848 			PROC_UNLOCK(q);
849 			PROC_UNLOCK(p);
850 
851 			if (status)
852 				*status = SIGCONT;
853 			return (0);
854 		}
855 		PROC_UNLOCK(p);
856 	}
857 	if (nfound == 0) {
858 		sx_xunlock(&proctree_lock);
859 		return (ECHILD);
860 	}
861 	if (options & WNOHANG) {
862 		sx_xunlock(&proctree_lock);
863 		td->td_retval[0] = 0;
864 		return (0);
865 	}
866 	PROC_LOCK(q);
867 	sx_xunlock(&proctree_lock);
868 	if (q->p_flag & P_STATCHILD) {
869 		q->p_flag &= ~P_STATCHILD;
870 		error = 0;
871 	} else
872 		error = msleep(q, &q->p_mtx, PWAIT | PCATCH, "wait", 0);
873 	PROC_UNLOCK(q);
874 	if (error)
875 		return (error);
876 	goto loop;
877 }
878 
879 /*
880  * Make process 'parent' the new parent of process 'child'.
881  * Must be called with an exclusive hold of proctree lock.
882  */
883 void
884 proc_reparent(struct proc *child, struct proc *parent)
885 {
886 
887 	sx_assert(&proctree_lock, SX_XLOCKED);
888 	PROC_LOCK_ASSERT(child, MA_OWNED);
889 	if (child->p_pptr == parent)
890 		return;
891 
892 	PROC_LOCK(child->p_pptr);
893 	sigqueue_take(child->p_ksi);
894 	PROC_UNLOCK(child->p_pptr);
895 	LIST_REMOVE(child, p_sibling);
896 	LIST_INSERT_HEAD(&parent->p_children, child, p_sibling);
897 	child->p_pptr = parent;
898 }
899