xref: /freebsd/sys/kern/kern_exit.c (revision b28624fde638caadd4a89f50c9b7e7da0f98c4d2)
1 /*-
2  * Copyright (c) 1982, 1986, 1989, 1991, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  * (c) UNIX System Laboratories, Inc.
5  * All or some portions of this file are derived from material licensed
6  * to the University of California by American Telephone and Telegraph
7  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8  * the permission of UNIX System Laboratories, Inc.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 4. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  *	@(#)kern_exit.c	8.7 (Berkeley) 2/12/94
35  */
36 
37 #include <sys/cdefs.h>
38 __FBSDID("$FreeBSD$");
39 
40 #include "opt_compat.h"
41 #include "opt_ktrace.h"
42 #include "opt_mac.h"
43 
44 #include <sys/param.h>
45 #include <sys/systm.h>
46 #include <sys/sysproto.h>
47 #include <sys/eventhandler.h>
48 #include <sys/kernel.h>
49 #include <sys/malloc.h>
50 #include <sys/lock.h>
51 #include <sys/mutex.h>
52 #include <sys/proc.h>
53 #include <sys/pioctl.h>
54 #include <sys/tty.h>
55 #include <sys/wait.h>
56 #include <sys/vmmeter.h>
57 #include <sys/vnode.h>
58 #include <sys/resourcevar.h>
59 #include <sys/sbuf.h>
60 #include <sys/signalvar.h>
61 #include <sys/sched.h>
62 #include <sys/sx.h>
63 #include <sys/syscallsubr.h>
64 #include <sys/syslog.h>
65 #include <sys/ptrace.h>
66 #include <sys/acct.h>		/* for acct_process() function prototype */
67 #include <sys/filedesc.h>
68 #include <sys/shm.h>
69 #include <sys/sem.h>
70 #ifdef KTRACE
71 #include <sys/ktrace.h>
72 #endif
73 
74 #include <security/audit/audit.h>
75 #include <security/mac/mac_framework.h>
76 
77 #include <vm/vm.h>
78 #include <vm/vm_extern.h>
79 #include <vm/vm_param.h>
80 #include <vm/pmap.h>
81 #include <vm/vm_map.h>
82 #include <vm/vm_page.h>
83 #include <vm/uma.h>
84 
85 /* Required to be non-static for SysVR4 emulator */
86 MALLOC_DEFINE(M_ZOMBIE, "zombie", "zombie proc status");
87 
88 /* Hook for NFS teardown procedure. */
89 void (*nlminfo_release_p)(struct proc *p);
90 
91 /*
92  * exit -- death of process.
93  */
94 void
95 sys_exit(struct thread *td, struct sys_exit_args *uap)
96 {
97 
98 	exit1(td, W_EXITCODE(uap->rval, 0));
99 	/* NOTREACHED */
100 }
101 
102 /*
103  * Exit: deallocate address space and other resources, change proc state to
104  * zombie, and unlink proc from allproc and parent's lists.  Save exit status
105  * and rusage for wait().  Check for child processes and orphan them.
106  */
107 void
108 exit1(struct thread *td, int rv)
109 {
110 	struct proc *p, *nq, *q;
111 	struct tty *tp;
112 	struct vnode *ttyvp;
113 	struct vnode *vtmp;
114 #ifdef KTRACE
115 	struct vnode *tracevp;
116 	struct ucred *tracecred;
117 #endif
118 	struct plimit *plim;
119 	int locked;
120 
121 	/*
122 	 * Drop Giant if caller has it.  Eventually we should warn about
123 	 * being called with Giant held.
124 	 */
125 	while (mtx_owned(&Giant))
126 		mtx_unlock(&Giant);
127 
128 	p = td->td_proc;
129 	if (p == initproc) {
130 		printf("init died (signal %d, exit %d)\n",
131 		    WTERMSIG(rv), WEXITSTATUS(rv));
132 		panic("Going nowhere without my init!");
133 	}
134 
135 	/*
136 	 * MUST abort all other threads before proceeding past here.
137 	 */
138 	PROC_LOCK(p);
139 	if (p->p_flag & P_HADTHREADS) {
140 retry:
141 		/*
142 		 * First check if some other thread got here before us..
143 		 * if so, act apropriatly, (exit or suspend);
144 		 */
145 		thread_suspend_check(0);
146 
147 		/*
148 		 * Kill off the other threads. This requires
149 		 * some co-operation from other parts of the kernel
150 		 * so it may not be instantaneous.  With this state set
151 		 * any thread entering the kernel from userspace will
152 		 * thread_exit() in trap().  Any thread attempting to
153 		 * sleep will return immediately with EINTR or EWOULDBLOCK
154 		 * which will hopefully force them to back out to userland
155 		 * freeing resources as they go.  Any thread attempting
156 		 * to return to userland will thread_exit() from userret().
157 		 * thread_exit() will unsuspend us when the last of the
158 		 * other threads exits.
159 		 * If there is already a thread singler after resumption,
160 		 * calling thread_single will fail; in that case, we just
161 		 * re-check all suspension request, the thread should
162 		 * either be suspended there or exit.
163 		 */
164 		if (thread_single(SINGLE_EXIT))
165 			goto retry;
166 
167 		/*
168 		 * All other activity in this process is now stopped.
169 		 * Threading support has been turned off.
170 		 */
171 	}
172 	KASSERT(p->p_numthreads == 1,
173 	    ("exit1: proc %p exiting with %d threads", p, p->p_numthreads));
174 	/*
175 	 * Wakeup anyone in procfs' PIOCWAIT.  They should have a hold
176 	 * on our vmspace, so we should block below until they have
177 	 * released their reference to us.  Note that if they have
178 	 * requested S_EXIT stops we will block here until they ack
179 	 * via PIOCCONT.
180 	 */
181 	_STOPEVENT(p, S_EXIT, rv);
182 
183 	/*
184 	 * Note that we are exiting and do another wakeup of anyone in
185 	 * PIOCWAIT in case they aren't listening for S_EXIT stops or
186 	 * decided to wait again after we told them we are exiting.
187 	 */
188 	p->p_flag |= P_WEXIT;
189 	wakeup(&p->p_stype);
190 
191 	/*
192 	 * Wait for any processes that have a hold on our vmspace to
193 	 * release their reference.
194 	 */
195 	while (p->p_lock > 0)
196 		msleep(&p->p_lock, &p->p_mtx, PWAIT, "exithold", 0);
197 
198 	PROC_UNLOCK(p);
199 	/* Drain the limit callout while we don't have the proc locked */
200 	callout_drain(&p->p_limco);
201 
202 #ifdef AUDIT
203 	/*
204 	 * The Sun BSM exit token contains two components: an exit status as
205 	 * passed to exit(), and a return value to indicate what sort of exit
206 	 * it was.  The exit status is WEXITSTATUS(rv), but it's not clear
207 	 * what the return value is.
208 	 */
209 	AUDIT_ARG(exit, WEXITSTATUS(rv), 0);
210 	AUDIT_SYSCALL_EXIT(0, td);
211 #endif
212 
213 	/* Are we a task leader? */
214 	if (p == p->p_leader) {
215 		mtx_lock(&ppeers_lock);
216 		q = p->p_peers;
217 		while (q != NULL) {
218 			PROC_LOCK(q);
219 			psignal(q, SIGKILL);
220 			PROC_UNLOCK(q);
221 			q = q->p_peers;
222 		}
223 		while (p->p_peers != NULL)
224 			msleep(p, &ppeers_lock, PWAIT, "exit1", 0);
225 		mtx_unlock(&ppeers_lock);
226 	}
227 
228 	/*
229 	 * Check if any loadable modules need anything done at process exit.
230 	 * E.g. SYSV IPC stuff
231 	 * XXX what if one of these generates an error?
232 	 */
233 	EVENTHANDLER_INVOKE(process_exit, p);
234 
235 	/*
236 	 * If parent is waiting for us to exit or exec,
237 	 * P_PPWAIT is set; we will wakeup the parent below.
238 	 */
239 	PROC_LOCK(p);
240 	stopprofclock(p);
241 	p->p_flag &= ~(P_TRACED | P_PPWAIT);
242 
243 	/*
244 	 * Stop the real interval timer.  If the handler is currently
245 	 * executing, prevent it from rearming itself and let it finish.
246 	 */
247 	if (timevalisset(&p->p_realtimer.it_value) &&
248 	    callout_stop(&p->p_itcallout) == 0) {
249 		timevalclear(&p->p_realtimer.it_interval);
250 		msleep(&p->p_itcallout, &p->p_mtx, PWAIT, "ritwait", 0);
251 		KASSERT(!timevalisset(&p->p_realtimer.it_value),
252 		    ("realtime timer is still armed"));
253 	}
254 	PROC_UNLOCK(p);
255 
256 	/*
257 	 * Reset any sigio structures pointing to us as a result of
258 	 * F_SETOWN with our pid.
259 	 */
260 	funsetownlst(&p->p_sigiolst);
261 
262 	/*
263 	 * If this process has an nlminfo data area (for lockd), release it
264 	 */
265 	if (nlminfo_release_p != NULL && p->p_nlminfo != NULL)
266 		(*nlminfo_release_p)(p);
267 
268 	/*
269 	 * Close open files and release open-file table.
270 	 * This may block!
271 	 */
272 	fdfree(td);
273 
274 	/*
275 	 * If this thread tickled GEOM, we need to wait for the giggling to
276 	 * stop before we return to userland
277 	 */
278 	if (td->td_pflags & TDP_GEOM)
279 		g_waitidle();
280 
281 	/*
282 	 * Remove ourself from our leader's peer list and wake our leader.
283 	 */
284 	mtx_lock(&ppeers_lock);
285 	if (p->p_leader->p_peers) {
286 		q = p->p_leader;
287 		while (q->p_peers != p)
288 			q = q->p_peers;
289 		q->p_peers = p->p_peers;
290 		wakeup(p->p_leader);
291 	}
292 	mtx_unlock(&ppeers_lock);
293 
294 	vmspace_exit(td);
295 
296 	mtx_lock(&Giant);	/* XXX TTY */
297 	sx_xlock(&proctree_lock);
298 	if (SESS_LEADER(p)) {
299 		struct session *sp;
300 
301 		sp = p->p_session;
302 		if (sp->s_ttyvp) {
303 			/*
304 			 * Controlling process.
305 			 * Signal foreground pgrp,
306 			 * drain controlling terminal
307 			 * and revoke access to controlling terminal.
308 			 */
309 			if (sp->s_ttyp && (sp->s_ttyp->t_session == sp)) {
310 				tp = sp->s_ttyp;
311 				if (sp->s_ttyp->t_pgrp) {
312 					PGRP_LOCK(sp->s_ttyp->t_pgrp);
313 					pgsignal(sp->s_ttyp->t_pgrp, SIGHUP, 1);
314 					PGRP_UNLOCK(sp->s_ttyp->t_pgrp);
315 				}
316 				/* XXX tp should be locked. */
317 				sx_xunlock(&proctree_lock);
318 				(void) ttywait(tp);
319 				sx_xlock(&proctree_lock);
320 				/*
321 				 * The tty could have been revoked
322 				 * if we blocked.
323 				 */
324 				if (sp->s_ttyvp) {
325 					ttyvp = sp->s_ttyvp;
326 					SESS_LOCK(p->p_session);
327 					sp->s_ttyvp = NULL;
328 					SESS_UNLOCK(p->p_session);
329 					sx_xunlock(&proctree_lock);
330 					VOP_LOCK(ttyvp, LK_EXCLUSIVE, td);
331 					VOP_REVOKE(ttyvp, REVOKEALL);
332 					vput(ttyvp);
333 					sx_xlock(&proctree_lock);
334 				}
335 			}
336 			if (sp->s_ttyvp) {
337 				ttyvp = sp->s_ttyvp;
338 				SESS_LOCK(p->p_session);
339 				sp->s_ttyvp = NULL;
340 				SESS_UNLOCK(p->p_session);
341 				vrele(ttyvp);
342 			}
343 			/*
344 			 * s_ttyp is not zero'd; we use this to indicate
345 			 * that the session once had a controlling terminal.
346 			 * (for logging and informational purposes)
347 			 */
348 		}
349 		SESS_LOCK(p->p_session);
350 		sp->s_leader = NULL;
351 		SESS_UNLOCK(p->p_session);
352 	}
353 	fixjobc(p, p->p_pgrp, 0);
354 	sx_xunlock(&proctree_lock);
355 	(void)acct_process(td);
356 	mtx_unlock(&Giant);
357 #ifdef KTRACE
358 	/*
359 	 * Disable tracing, then drain any pending records and release
360 	 * the trace file.
361 	 */
362 	if (p->p_traceflag != 0) {
363 		PROC_LOCK(p);
364 		mtx_lock(&ktrace_mtx);
365 		p->p_traceflag = 0;
366 		mtx_unlock(&ktrace_mtx);
367 		PROC_UNLOCK(p);
368 		ktrprocexit(td);
369 		PROC_LOCK(p);
370 		mtx_lock(&ktrace_mtx);
371 		tracevp = p->p_tracevp;
372 		p->p_tracevp = NULL;
373 		tracecred = p->p_tracecred;
374 		p->p_tracecred = NULL;
375 		mtx_unlock(&ktrace_mtx);
376 		PROC_UNLOCK(p);
377 		if (tracevp != NULL) {
378 			locked = VFS_LOCK_GIANT(tracevp->v_mount);
379 			vrele(tracevp);
380 			VFS_UNLOCK_GIANT(locked);
381 		}
382 		if (tracecred != NULL)
383 			crfree(tracecred);
384 	}
385 #endif
386 	/*
387 	 * Release reference to text vnode
388 	 */
389 	if ((vtmp = p->p_textvp) != NULL) {
390 		p->p_textvp = NULL;
391 		locked = VFS_LOCK_GIANT(vtmp->v_mount);
392 		vrele(vtmp);
393 		VFS_UNLOCK_GIANT(locked);
394 	}
395 
396 	/*
397 	 * Release our limits structure.
398 	 */
399 	PROC_LOCK(p);
400 	plim = p->p_limit;
401 	p->p_limit = NULL;
402 	PROC_UNLOCK(p);
403 	lim_free(plim);
404 
405 	/*
406 	 * Remove proc from allproc queue and pidhash chain.
407 	 * Place onto zombproc.  Unlink from parent's child list.
408 	 */
409 	sx_xlock(&allproc_lock);
410 	LIST_REMOVE(p, p_list);
411 	LIST_INSERT_HEAD(&zombproc, p, p_list);
412 	LIST_REMOVE(p, p_hash);
413 	sx_xunlock(&allproc_lock);
414 
415 	/*
416 	 * Call machine-dependent code to release any
417 	 * machine-dependent resources other than the address space.
418 	 * The address space is released by "vmspace_exitfree(p)" in
419 	 * vm_waitproc().
420 	 */
421 	cpu_exit(td);
422 
423 	WITNESS_WARN(WARN_PANIC, NULL, "process (pid %d) exiting", p->p_pid);
424 
425 	/*
426 	 * Reparent all of our children to init.
427 	 */
428 	sx_xlock(&proctree_lock);
429 	q = LIST_FIRST(&p->p_children);
430 	if (q != NULL)		/* only need this if any child is S_ZOMB */
431 		wakeup(initproc);
432 	for (; q != NULL; q = nq) {
433 		nq = LIST_NEXT(q, p_sibling);
434 		PROC_LOCK(q);
435 		proc_reparent(q, initproc);
436 		q->p_sigparent = SIGCHLD;
437 		/*
438 		 * Traced processes are killed
439 		 * since their existence means someone is screwing up.
440 		 */
441 		if (q->p_flag & P_TRACED) {
442 			q->p_flag &= ~(P_TRACED | P_STOPPED_TRACE);
443 			psignal(q, SIGKILL);
444 		}
445 		PROC_UNLOCK(q);
446 	}
447 
448 	/* Save exit status. */
449 	PROC_LOCK(p);
450 	p->p_xstat = rv;
451 	p->p_xthread = td;
452 	/*
453 	 * Notify interested parties of our demise.
454 	 */
455 	KNOTE_LOCKED(&p->p_klist, NOTE_EXIT);
456 
457 	/*
458 	 * Just delete all entries in the p_klist. At this point we won't
459 	 * report any more events, and there are nasty race conditions that
460 	 * can beat us if we don't.
461 	 */
462 	knlist_clear(&p->p_klist, 1);
463 
464 	/*
465 	 * Notify parent that we're gone.  If parent has the PS_NOCLDWAIT
466 	 * flag set, or if the handler is set to SIG_IGN, notify process
467 	 * 1 instead (and hope it will handle this situation).
468 	 */
469 	PROC_LOCK(p->p_pptr);
470 	mtx_lock(&p->p_pptr->p_sigacts->ps_mtx);
471 	if (p->p_pptr->p_sigacts->ps_flag & (PS_NOCLDWAIT | PS_CLDSIGIGN)) {
472 		struct proc *pp;
473 
474 		mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx);
475 		pp = p->p_pptr;
476 		PROC_UNLOCK(pp);
477 		proc_reparent(p, initproc);
478 		p->p_sigparent = SIGCHLD;
479 		PROC_LOCK(p->p_pptr);
480 		/*
481 		 * If this was the last child of our parent, notify
482 		 * parent, so in case he was wait(2)ing, he will
483 		 * continue.
484 		 */
485 		if (LIST_EMPTY(&pp->p_children))
486 			wakeup(pp);
487 	} else
488 		mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx);
489 
490 	if (p->p_pptr == initproc)
491 		psignal(p->p_pptr, SIGCHLD);
492 	else if (p->p_sigparent != 0) {
493 		if (p->p_sigparent == SIGCHLD)
494 			childproc_exited(p);
495 		else	/* LINUX thread */
496 			psignal(p->p_pptr, p->p_sigparent);
497 	}
498 	sx_xunlock(&proctree_lock);
499 
500 	/*
501 	 * The state PRS_ZOMBIE prevents other proesses from sending
502 	 * signal to the process, to avoid memory leak, we free memory
503 	 * for signal queue at the time when the state is set.
504 	 */
505 	sigqueue_flush(&p->p_sigqueue);
506 	sigqueue_flush(&td->td_sigqueue);
507 
508 	/*
509 	 * We have to wait until after acquiring all locks before
510 	 * changing p_state.  We need to avoid all possible context
511 	 * switches (including ones from blocking on a mutex) while
512 	 * marked as a zombie.  We also have to set the zombie state
513 	 * before we release the parent process' proc lock to avoid
514 	 * a lost wakeup.  So, we first call wakeup, then we grab the
515 	 * sched lock, update the state, and release the parent process'
516 	 * proc lock.
517 	 */
518 	wakeup(p->p_pptr);
519 	PROC_SLOCK(p->p_pptr);
520 	sched_exit(p->p_pptr, td);
521 	PROC_SUNLOCK(p->p_pptr);
522 	PROC_SLOCK(p);
523 	p->p_state = PRS_ZOMBIE;
524 	PROC_UNLOCK(p->p_pptr);
525 
526 	/*
527 	 * Hopefully no one will try to deliver a signal to the process this
528 	 * late in the game.
529 	 */
530 	knlist_destroy(&p->p_klist);
531 
532 	/*
533 	 * Save our children's rusage information in our exit rusage.
534 	 */
535 	ruadd(&p->p_ru, &p->p_rux, &p->p_stats->p_cru, &p->p_crux);
536 
537 	/*
538 	 * Make sure the scheduler takes this thread out of its tables etc.
539 	 * This will also release this thread's reference to the ucred.
540 	 * Other thread parts to release include pcb bits and such.
541 	 */
542 	thread_exit();
543 }
544 
545 
546 #ifndef _SYS_SYSPROTO_H_
547 struct abort2_args {
548 	char *why;
549 	int nargs;
550 	void **args;
551 };
552 #endif
553 
554 int
555 abort2(struct thread *td, struct abort2_args *uap)
556 {
557 	struct proc *p = td->td_proc;
558 	struct sbuf *sb;
559 	void *uargs[16];
560 	int error, i, sig;
561 
562 	error = 0;	/* satisfy compiler */
563 
564 	/*
565 	 * Do it right now so we can log either proper call of abort2(), or
566 	 * note, that invalid argument was passed. 512 is big enough to
567 	 * handle 16 arguments' descriptions with additional comments.
568 	 */
569 	sb = sbuf_new(NULL, NULL, 512, SBUF_FIXEDLEN);
570 	sbuf_clear(sb);
571 	sbuf_printf(sb, "%s(pid %d uid %d) aborted: ",
572 	    p->p_comm, p->p_pid, td->td_ucred->cr_uid);
573 	/*
574 	 * Since we can't return from abort2(), send SIGKILL in cases, where
575 	 * abort2() was called improperly
576 	 */
577 	sig = SIGKILL;
578 	/* Prevent from DoSes from user-space. */
579 	if (uap->nargs < 0 || uap->nargs > 16)
580 		goto out;
581 	if (uap->args == NULL)
582 		goto out;
583 	error = copyin(uap->args, uargs, uap->nargs * sizeof(void *));
584 	if (error != 0)
585 		goto out;
586 	/*
587 	 * Limit size of 'reason' string to 128. Will fit even when
588 	 * maximal number of arguments was chosen to be logged.
589 	 */
590 	if (uap->why != NULL) {
591 		error = sbuf_copyin(sb, uap->why, 128);
592 		if (error < 0)
593 			goto out;
594 	} else {
595 		sbuf_printf(sb, "(null)");
596 	}
597 	if (uap->nargs) {
598 		sbuf_printf(sb, "(");
599 		for (i = 0;i < uap->nargs; i++)
600 			sbuf_printf(sb, "%s%p", i == 0 ? "" : ", ", uargs[i]);
601 		sbuf_printf(sb, ")");
602 	}
603 	/*
604 	 * Final stage: arguments were proper, string has been
605 	 * successfully copied from userspace, and copying pointers
606 	 * from user-space succeed.
607 	 */
608 	sig = SIGABRT;
609 out:
610 	if (sig == SIGKILL) {
611 		sbuf_trim(sb);
612 		sbuf_printf(sb, " (Reason text inaccessible)");
613 	}
614 	sbuf_cat(sb, "\n");
615 	sbuf_finish(sb);
616 	log(LOG_INFO, "%s", sbuf_data(sb));
617 	sbuf_delete(sb);
618 	exit1(td, W_EXITCODE(0, sig));
619 	return (0);
620 }
621 
622 
623 #ifdef COMPAT_43
624 /*
625  * The dirty work is handled by kern_wait().
626  */
627 int
628 owait(struct thread *td, struct owait_args *uap __unused)
629 {
630 	int error, status;
631 
632 	error = kern_wait(td, WAIT_ANY, &status, 0, NULL);
633 	if (error == 0)
634 		td->td_retval[1] = status;
635 	return (error);
636 }
637 #endif /* COMPAT_43 */
638 
639 /*
640  * The dirty work is handled by kern_wait().
641  */
642 int
643 wait4(struct thread *td, struct wait_args *uap)
644 {
645 	struct rusage ru, *rup;
646 	int error, status;
647 
648 	if (uap->rusage != NULL)
649 		rup = &ru;
650 	else
651 		rup = NULL;
652 	error = kern_wait(td, uap->pid, &status, uap->options, rup);
653 	if (uap->status != NULL && error == 0)
654 		error = copyout(&status, uap->status, sizeof(status));
655 	if (uap->rusage != NULL && error == 0)
656 		error = copyout(&ru, uap->rusage, sizeof(struct rusage));
657 	return (error);
658 }
659 
660 int
661 kern_wait(struct thread *td, pid_t pid, int *status, int options,
662     struct rusage *rusage)
663 {
664 	struct proc *p, *q, *t;
665 	int error, nfound;
666 
667 	AUDIT_ARG(pid, pid);
668 
669 	q = td->td_proc;
670 	if (pid == 0) {
671 		PROC_LOCK(q);
672 		pid = -q->p_pgid;
673 		PROC_UNLOCK(q);
674 	}
675 	if (options &~ (WUNTRACED|WNOHANG|WCONTINUED|WLINUXCLONE))
676 		return (EINVAL);
677 loop:
678 	if (q->p_flag & P_STATCHILD) {
679 		PROC_LOCK(q);
680 		q->p_flag &= ~P_STATCHILD;
681 		PROC_UNLOCK(q);
682 	}
683 	nfound = 0;
684 	sx_xlock(&proctree_lock);
685 	LIST_FOREACH(p, &q->p_children, p_sibling) {
686 		PROC_LOCK(p);
687 		if (pid != WAIT_ANY &&
688 		    p->p_pid != pid && p->p_pgid != -pid) {
689 			PROC_UNLOCK(p);
690 			continue;
691 		}
692 		if (p_canwait(td, p)) {
693 			PROC_UNLOCK(p);
694 			continue;
695 		}
696 
697 		/*
698 		 * This special case handles a kthread spawned by linux_clone
699 		 * (see linux_misc.c).  The linux_wait4 and linux_waitpid
700 		 * functions need to be able to distinguish between waiting
701 		 * on a process and waiting on a thread.  It is a thread if
702 		 * p_sigparent is not SIGCHLD, and the WLINUXCLONE option
703 		 * signifies we want to wait for threads and not processes.
704 		 */
705 		if ((p->p_sigparent != SIGCHLD) ^
706 		    ((options & WLINUXCLONE) != 0)) {
707 			PROC_UNLOCK(p);
708 			continue;
709 		}
710 
711 		nfound++;
712 		PROC_SLOCK(p);
713 		if (p->p_state == PRS_ZOMBIE) {
714 			if (rusage) {
715 				*rusage = p->p_ru;
716 				calcru(p, &rusage->ru_utime, &rusage->ru_stime);
717 			}
718 			PROC_SUNLOCK(p);
719 			td->td_retval[0] = p->p_pid;
720 			if (status)
721 				*status = p->p_xstat;	/* convert to int */
722 			PROC_LOCK(q);
723 			sigqueue_take(p->p_ksi);
724 			PROC_UNLOCK(q);
725 
726 			/*
727 			 * If we got the child via a ptrace 'attach',
728 			 * we need to give it back to the old parent.
729 			 */
730 			PROC_UNLOCK(p);
731 			if (p->p_oppid && (t = pfind(p->p_oppid)) != NULL) {
732 				PROC_LOCK(p);
733 				p->p_oppid = 0;
734 				proc_reparent(p, t);
735 				PROC_UNLOCK(p);
736 				tdsignal(t, NULL, SIGCHLD, p->p_ksi);
737 				wakeup(t);
738 				PROC_UNLOCK(t);
739 				sx_xunlock(&proctree_lock);
740 				return (0);
741 			}
742 
743 			/*
744 			 * Remove other references to this process to ensure
745 			 * we have an exclusive reference.
746 			 */
747 			sx_xlock(&allproc_lock);
748 			LIST_REMOVE(p, p_list);	/* off zombproc */
749 			sx_xunlock(&allproc_lock);
750 			LIST_REMOVE(p, p_sibling);
751 			leavepgrp(p);
752 			sx_xunlock(&proctree_lock);
753 
754 			/*
755 			 * As a side effect of this lock, we know that
756 			 * all other writes to this proc are visible now, so
757 			 * no more locking is needed for p.
758 			 */
759 			PROC_LOCK(p);
760 			p->p_xstat = 0;		/* XXX: why? */
761 			PROC_UNLOCK(p);
762 			PROC_LOCK(q);
763 			ruadd(&q->p_stats->p_cru, &q->p_crux, &p->p_ru,
764 			    &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
789 			 * to free anything that cpu_exit couldn't
790 			 * release while still running in process context.
791 			 */
792 			vm_waitproc(p);
793 #ifdef MAC
794 			mac_destroy_proc(p);
795 #endif
796 			KASSERT(FIRST_THREAD_IN_PROC(p),
797 			    ("kern_wait: no residual thread!"));
798 			uma_zfree(proc_zone, p);
799 			sx_xlock(&allproc_lock);
800 			nprocs--;
801 			sx_xunlock(&allproc_lock);
802 			return (0);
803 		}
804 		if ((p->p_flag & P_STOPPED_SIG) &&
805 		    (p->p_suspcount == p->p_numthreads) &&
806 		    (p->p_flag & P_WAITED) == 0 &&
807 		    (p->p_flag & P_TRACED || options & WUNTRACED)) {
808 			PROC_SUNLOCK(p);
809 			p->p_flag |= P_WAITED;
810 			sx_xunlock(&proctree_lock);
811 			td->td_retval[0] = p->p_pid;
812 			if (status)
813 				*status = W_STOPCODE(p->p_xstat);
814 
815 			PROC_LOCK(q);
816 			sigqueue_take(p->p_ksi);
817 			PROC_UNLOCK(q);
818 			PROC_UNLOCK(p);
819 
820 			return (0);
821 		}
822 		PROC_SUNLOCK(p);
823 		if (options & WCONTINUED && (p->p_flag & P_CONTINUED)) {
824 			sx_xunlock(&proctree_lock);
825 			td->td_retval[0] = p->p_pid;
826 			p->p_flag &= ~P_CONTINUED;
827 
828 			PROC_LOCK(q);
829 			sigqueue_take(p->p_ksi);
830 			PROC_UNLOCK(q);
831 			PROC_UNLOCK(p);
832 
833 			if (status)
834 				*status = SIGCONT;
835 			return (0);
836 		}
837 		PROC_UNLOCK(p);
838 	}
839 	if (nfound == 0) {
840 		sx_xunlock(&proctree_lock);
841 		return (ECHILD);
842 	}
843 	if (options & WNOHANG) {
844 		sx_xunlock(&proctree_lock);
845 		td->td_retval[0] = 0;
846 		return (0);
847 	}
848 	PROC_LOCK(q);
849 	sx_xunlock(&proctree_lock);
850 	if (q->p_flag & P_STATCHILD) {
851 		q->p_flag &= ~P_STATCHILD;
852 		error = 0;
853 	} else
854 		error = msleep(q, &q->p_mtx, PWAIT | PCATCH, "wait", 0);
855 	PROC_UNLOCK(q);
856 	if (error)
857 		return (error);
858 	goto loop;
859 }
860 
861 /*
862  * Make process 'parent' the new parent of process 'child'.
863  * Must be called with an exclusive hold of proctree lock.
864  */
865 void
866 proc_reparent(struct proc *child, struct proc *parent)
867 {
868 
869 	sx_assert(&proctree_lock, SX_XLOCKED);
870 	PROC_LOCK_ASSERT(child, MA_OWNED);
871 	if (child->p_pptr == parent)
872 		return;
873 
874 	PROC_LOCK(child->p_pptr);
875 	sigqueue_take(child->p_ksi);
876 	PROC_UNLOCK(child->p_pptr);
877 	LIST_REMOVE(child, p_sibling);
878 	LIST_INSERT_HEAD(&parent->p_children, child, p_sibling);
879 	child->p_pptr = parent;
880 }
881