xref: /netbsd-src/sys/kern/kern_exit.c (revision 6cf6fe02a981b55727c49c3d37b0d8191a98c0ee)
1 /*	$NetBSD: kern_exit.c,v 1.244 2014/05/05 15:45:32 christos Exp $	*/
2 
3 /*-
4  * Copyright (c) 1998, 1999, 2006, 2007, 2008 The NetBSD Foundation, Inc.
5  * All rights reserved.
6  *
7  * This code is derived from software contributed to The NetBSD Foundation
8  * by Jason R. Thorpe of the Numerical Aerospace Simulation Facility,
9  * NASA Ames Research Center, and by Andrew Doran.
10  *
11  * Redistribution and use in source and binary forms, with or without
12  * modification, are permitted provided that the following conditions
13  * are met:
14  * 1. Redistributions of source code must retain the above copyright
15  *    notice, this list of conditions and the following disclaimer.
16  * 2. Redistributions in binary form must reproduce the above copyright
17  *    notice, this list of conditions and the following disclaimer in the
18  *    documentation and/or other materials provided with the distribution.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
21  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
22  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
23  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
24  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
25  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
26  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
27  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
28  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
29  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
30  * POSSIBILITY OF SUCH DAMAGE.
31  */
32 
33 /*
34  * Copyright (c) 1982, 1986, 1989, 1991, 1993
35  *	The Regents of the University of California.  All rights reserved.
36  * (c) UNIX System Laboratories, Inc.
37  * All or some portions of this file are derived from material licensed
38  * to the University of California by American Telephone and Telegraph
39  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
40  * the permission of UNIX System Laboratories, Inc.
41  *
42  * Redistribution and use in source and binary forms, with or without
43  * modification, are permitted provided that the following conditions
44  * are met:
45  * 1. Redistributions of source code must retain the above copyright
46  *    notice, this list of conditions and the following disclaimer.
47  * 2. Redistributions in binary form must reproduce the above copyright
48  *    notice, this list of conditions and the following disclaimer in the
49  *    documentation and/or other materials provided with the distribution.
50  * 3. Neither the name of the University nor the names of its contributors
51  *    may be used to endorse or promote products derived from this software
52  *    without specific prior written permission.
53  *
54  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
55  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
56  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
57  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
58  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
59  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
60  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
61  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
62  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
63  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
64  * SUCH DAMAGE.
65  *
66  *	@(#)kern_exit.c	8.10 (Berkeley) 2/23/95
67  */
68 
69 #include <sys/cdefs.h>
70 __KERNEL_RCSID(0, "$NetBSD: kern_exit.c,v 1.244 2014/05/05 15:45:32 christos Exp $");
71 
72 #include "opt_ktrace.h"
73 #include "opt_perfctrs.h"
74 #include "opt_sysv.h"
75 
76 #include <sys/param.h>
77 #include <sys/systm.h>
78 #include <sys/ioctl.h>
79 #include <sys/tty.h>
80 #include <sys/time.h>
81 #include <sys/resource.h>
82 #include <sys/kernel.h>
83 #include <sys/proc.h>
84 #include <sys/buf.h>
85 #include <sys/wait.h>
86 #include <sys/file.h>
87 #include <sys/vnode.h>
88 #include <sys/syslog.h>
89 #include <sys/pool.h>
90 #include <sys/uidinfo.h>
91 #if defined(PERFCTRS)
92 #include <sys/pmc.h>
93 #endif
94 #include <sys/ptrace.h>
95 #include <sys/acct.h>
96 #include <sys/filedesc.h>
97 #include <sys/ras.h>
98 #include <sys/signalvar.h>
99 #include <sys/sched.h>
100 #include <sys/mount.h>
101 #include <sys/syscallargs.h>
102 #include <sys/kauth.h>
103 #include <sys/sleepq.h>
104 #include <sys/lockdebug.h>
105 #include <sys/ktrace.h>
106 #include <sys/cpu.h>
107 #include <sys/lwpctl.h>
108 #include <sys/atomic.h>
109 #include <sys/sdt.h>
110 
111 #include <uvm/uvm_extern.h>
112 
113 #ifdef DEBUG_EXIT
114 int debug_exit = 0;
115 #define DPRINTF(x) if (debug_exit) printf x
116 #else
117 #define DPRINTF(x)
118 #endif
119 
120 static int find_stopped_child(struct proc *, pid_t, int, struct proc **, int *);
121 static void proc_free(struct proc *, struct rusage *);
122 
123 /*
124  * DTrace SDT provider definitions
125  */
126 SDT_PROBE_DEFINE(proc,,,exit,exit,
127 	    "int", NULL, 		/* reason */
128 	    NULL, NULL, NULL, NULL,
129 	    NULL, NULL, NULL, NULL);
130 /*
131  * Fill in the appropriate signal information, and signal the parent.
132  */
133 /* XXX noclone works around a gcc 4.5 bug on arm */
134 static void __noclone
135 exit_psignal(struct proc *p, struct proc *pp, ksiginfo_t *ksi)
136 {
137 
138 	KSI_INIT(ksi);
139 	if ((ksi->ksi_signo = P_EXITSIG(p)) == SIGCHLD) {
140 		if (WIFSIGNALED(p->p_xstat)) {
141 			if (WCOREDUMP(p->p_xstat))
142 				ksi->ksi_code = CLD_DUMPED;
143 			else
144 				ksi->ksi_code = CLD_KILLED;
145 		} else {
146 			ksi->ksi_code = CLD_EXITED;
147 		}
148 	}
149 	/*
150 	 * We fill those in, even for non-SIGCHLD.
151 	 * It's safe to access p->p_cred unlocked here.
152 	 */
153 	ksi->ksi_pid = p->p_pid;
154 	ksi->ksi_uid = kauth_cred_geteuid(p->p_cred);
155 	ksi->ksi_status = p->p_xstat;
156 	/* XXX: is this still valid? */
157 	ksi->ksi_utime = p->p_stats->p_ru.ru_utime.tv_sec;
158 	ksi->ksi_stime = p->p_stats->p_ru.ru_stime.tv_sec;
159 }
160 
161 /*
162  * exit --
163  *	Death of process.
164  */
165 int
166 sys_exit(struct lwp *l, const struct sys_exit_args *uap, register_t *retval)
167 {
168 	/* {
169 		syscallarg(int)	rval;
170 	} */
171 	struct proc *p = l->l_proc;
172 
173 	/* Don't call exit1() multiple times in the same process. */
174 	mutex_enter(p->p_lock);
175 	if (p->p_sflag & PS_WEXIT) {
176 		mutex_exit(p->p_lock);
177 		lwp_exit(l);
178 	}
179 
180 	/* exit1() will release the mutex. */
181 	exit1(l, W_EXITCODE(SCARG(uap, rval), 0));
182 	/* NOTREACHED */
183 	return (0);
184 }
185 
186 /*
187  * Exit: deallocate address space and other resources, change proc state
188  * to zombie, and unlink proc from allproc and parent's lists.  Save exit
189  * status and rusage for wait().  Check for child processes and orphan them.
190  *
191  * Must be called with p->p_lock held.  Does not return.
192  */
193 void
194 exit1(struct lwp *l, int rv)
195 {
196 	struct proc	*p, *child, *next_child, *old_parent, *new_parent;
197 	struct pgrp	*pgrp;
198 	ksiginfo_t	ksi;
199 	ksiginfoq_t	kq;
200 	int		wakeinit;
201 
202 	p = l->l_proc;
203 
204 	KASSERT(mutex_owned(p->p_lock));
205 	KASSERT(p->p_vmspace != NULL);
206 
207 	if (__predict_false(p == initproc)) {
208 		panic("init died (signal %d, exit %d)",
209 		    WTERMSIG(rv), WEXITSTATUS(rv));
210 	}
211 
212 	p->p_sflag |= PS_WEXIT;
213 
214 	/*
215 	 * Force all other LWPs to exit before we do.  Only then can we
216 	 * begin to tear down the rest of the process state.
217 	 */
218 	if (p->p_nlwps > 1) {
219 		exit_lwps(l);
220 	}
221 
222 	ksiginfo_queue_init(&kq);
223 
224 	/*
225 	 * If we have been asked to stop on exit, do so now.
226 	 */
227 	if (__predict_false(p->p_sflag & PS_STOPEXIT)) {
228 		KERNEL_UNLOCK_ALL(l, &l->l_biglocks);
229 		sigclearall(p, &contsigmask, &kq);
230 		p->p_waited = 0;
231 		membar_producer();
232 		p->p_stat = SSTOP;
233 		lwp_lock(l);
234 		p->p_nrlwps--;
235 		l->l_stat = LSSTOP;
236 		lwp_unlock(l);
237 		mutex_exit(p->p_lock);
238 		lwp_lock(l);
239 		mi_switch(l);
240 		KERNEL_LOCK(l->l_biglocks, l);
241 		mutex_enter(p->p_lock);
242 	}
243 
244 	/*
245 	 * Bin any remaining signals and mark the process as dying so it will
246 	 * not be found for, e.g. signals.
247 	 */
248 	sigfillset(&p->p_sigctx.ps_sigignore);
249 	sigclearall(p, NULL, &kq);
250 	p->p_stat = SDYING;
251 	mutex_exit(p->p_lock);
252 	ksiginfo_queue_drain(&kq);
253 
254 	/* Destroy any lwpctl info. */
255 	if (p->p_lwpctl != NULL)
256 		lwp_ctl_exit();
257 
258 	/*
259 	 * Drain all remaining references that procfs, ptrace and others may
260 	 * have on the process.
261 	 */
262 	rw_enter(&p->p_reflock, RW_WRITER);
263 
264 	DPRINTF(("exit1: %d.%d exiting.\n", p->p_pid, l->l_lid));
265 
266 	timers_free(p, TIMERS_ALL);
267 #if defined(__HAVE_RAS)
268 	ras_purgeall();
269 #endif
270 
271 	/*
272 	 * Close open files, release open-file table and free signal
273 	 * actions.  This may block!
274 	 */
275 	fd_free();
276 	cwdfree(p->p_cwdi);
277 	p->p_cwdi = NULL;
278 	doexithooks(p);
279 	sigactsfree(p->p_sigacts);
280 
281 	/*
282 	 * Write out accounting data.
283 	 */
284 	(void)acct_process(l);
285 
286 #ifdef KTRACE
287 	/*
288 	 * Release trace file.
289 	 */
290 	if (p->p_tracep != NULL) {
291 		mutex_enter(&ktrace_lock);
292 		ktrderef(p);
293 		mutex_exit(&ktrace_lock);
294 	}
295 #endif
296 
297 	/*
298 	 * If emulation has process exit hook, call it now.
299 	 * Set the exit status now so that the exit hook has
300 	 * an opportunity to tweak it (COMPAT_LINUX requires
301 	 * this for thread group emulation)
302 	 */
303 	p->p_xstat = rv;
304 	if (p->p_emul->e_proc_exit)
305 		(*p->p_emul->e_proc_exit)(p);
306 
307 	/*
308 	 * Free the VM resources we're still holding on to.
309 	 * We must do this from a valid thread because doing
310 	 * so may block. This frees vmspace, which we don't
311 	 * need anymore. The only remaining lwp is the one
312 	 * we run at this moment, nothing runs in userland
313 	 * anymore.
314 	 */
315 	uvm_proc_exit(p);
316 
317 	/*
318 	 * Stop profiling.
319 	 */
320 	if (__predict_false((p->p_stflag & PST_PROFIL) != 0)) {
321 		mutex_spin_enter(&p->p_stmutex);
322 		stopprofclock(p);
323 		mutex_spin_exit(&p->p_stmutex);
324 	}
325 
326 	/*
327 	 * If parent is waiting for us to exit or exec, PL_PPWAIT is set; we
328 	 * wake up the parent early to avoid deadlock.  We can do this once
329 	 * the VM resources are released.
330 	 */
331 	mutex_enter(proc_lock);
332 	if (p->p_lflag & PL_PPWAIT) {
333 #if 0
334 		lwp_t *lp;
335 
336 		l->l_lwpctl = NULL; /* was on loan from blocked parent */
337 		p->p_lflag &= ~PL_PPWAIT;
338 
339 		lp = p->p_vforklwp;
340 		p->p_vforklwp = NULL;
341 		lp->l_pflag &= ~LP_VFORKWAIT; /* XXX */
342 		cv_broadcast(&lp->l_waitcv);
343 #else
344 		l->l_lwpctl = NULL; /* was on loan from blocked parent */
345 		p->p_lflag &= ~PL_PPWAIT;
346 		cv_broadcast(&p->p_pptr->p_waitcv);
347 #endif
348 	}
349 
350 	if (SESS_LEADER(p)) {
351 		struct vnode *vprele = NULL, *vprevoke = NULL;
352 		struct session *sp = p->p_session;
353 		struct tty *tp;
354 
355 		if (sp->s_ttyvp) {
356 			/*
357 			 * Controlling process.
358 			 * Signal foreground pgrp,
359 			 * drain controlling terminal
360 			 * and revoke access to controlling terminal.
361 			 */
362 			tp = sp->s_ttyp;
363 			mutex_spin_enter(&tty_lock);
364 			if (tp->t_session == sp) {
365 				/* we can't guarantee the revoke will do this */
366 				pgrp = tp->t_pgrp;
367 				tp->t_pgrp = NULL;
368 				tp->t_session = NULL;
369 				mutex_spin_exit(&tty_lock);
370 				if (pgrp != NULL) {
371 					pgsignal(pgrp, SIGHUP, 1);
372 				}
373 				mutex_exit(proc_lock);
374 				(void) ttywait(tp);
375 				mutex_enter(proc_lock);
376 
377 				/* The tty could have been revoked. */
378 				vprevoke = sp->s_ttyvp;
379 			} else
380 				mutex_spin_exit(&tty_lock);
381 			vprele = sp->s_ttyvp;
382 			sp->s_ttyvp = NULL;
383 			/*
384 			 * s_ttyp is not zero'd; we use this to indicate
385 			 * that the session once had a controlling terminal.
386 			 * (for logging and informational purposes)
387 			 */
388 		}
389 		sp->s_leader = NULL;
390 
391 		if (vprevoke != NULL || vprele != NULL) {
392 			if (vprevoke != NULL) {
393 				/* Releases proc_lock. */
394 				proc_sessrele(sp);
395 				VOP_REVOKE(vprevoke, REVOKEALL);
396 			} else
397 				mutex_exit(proc_lock);
398 			if (vprele != NULL)
399 				vrele(vprele);
400 			mutex_enter(proc_lock);
401 		}
402 	}
403 	fixjobc(p, p->p_pgrp, 0);
404 
405 	/*
406 	 * Finalize the last LWP's specificdata, as well as the
407 	 * specificdata for the proc itself.
408 	 */
409 	lwp_finispecific(l);
410 	proc_finispecific(p);
411 
412 	/*
413 	 * Notify interested parties of our demise.
414 	 */
415 	KNOTE(&p->p_klist, NOTE_EXIT);
416 
417 	SDT_PROBE(proc,,,exit,
418 		(WCOREDUMP(rv) ? CLD_DUMPED :
419 		 (WIFSIGNALED(rv) ? CLD_KILLED : CLD_EXITED)),
420 		0,0,0,0);
421 
422 #if PERFCTRS
423 	/*
424 	 * Save final PMC information in parent process & clean up.
425 	 */
426 	if (PMC_ENABLED(p)) {
427 		pmc_save_context(p);
428 		pmc_accumulate(p->p_pptr, p);
429 		pmc_process_exit(p);
430 	}
431 #endif
432 
433 	/*
434 	 * Reset p_opptr pointer of all former children which got
435 	 * traced by another process and were reparented. We reset
436 	 * it to NULL here; the trace detach code then reparents
437 	 * the child to initproc. We only check allproc list, since
438 	 * eventual former children on zombproc list won't reference
439 	 * p_opptr anymore.
440 	 */
441 	if (__predict_false(p->p_slflag & PSL_CHTRACED)) {
442 		struct proc *q;
443 		PROCLIST_FOREACH(q, &allproc) {
444 			if (q->p_opptr == p)
445 				q->p_opptr = NULL;
446 		}
447 	}
448 
449 	/*
450 	 * Give orphaned children to init(8).
451 	 */
452 	child = LIST_FIRST(&p->p_children);
453 	wakeinit = (child != NULL);
454 	for (; child != NULL; child = next_child) {
455 		next_child = LIST_NEXT(child, p_sibling);
456 
457 		/*
458 		 * Traced processes are killed since their existence
459 		 * means someone is screwing up. Since we reset the
460 		 * trace flags, the logic in sys_wait4() would not be
461 		 * triggered to reparent the process to its
462 		 * original parent, so we must do this here.
463 		 */
464 		if (__predict_false(child->p_slflag & PSL_TRACED)) {
465 			mutex_enter(p->p_lock);
466 			child->p_slflag &=
467 			    ~(PSL_TRACED|PSL_FSTRACE|PSL_SYSCALL);
468 			mutex_exit(p->p_lock);
469 			if (child->p_opptr != child->p_pptr) {
470 				struct proc *t = child->p_opptr;
471 				proc_reparent(child, t ? t : initproc);
472 				child->p_opptr = NULL;
473 			} else
474 				proc_reparent(child, initproc);
475 			killproc(child, "orphaned traced process");
476 		} else
477 			proc_reparent(child, initproc);
478 	}
479 
480 	/*
481 	 * Move proc from allproc to zombproc, it's now nearly ready to be
482 	 * collected by parent.
483 	 */
484 	LIST_REMOVE(l, l_list);
485 	LIST_REMOVE(p, p_list);
486 	LIST_INSERT_HEAD(&zombproc, p, p_list);
487 
488 	/*
489 	 * Mark the process as dead.  We must do this before we signal
490 	 * the parent.
491 	 */
492 	p->p_stat = SDEAD;
493 
494 	/* Put in front of parent's sibling list for parent to collect it */
495 	old_parent = p->p_pptr;
496 	old_parent->p_nstopchild++;
497 	if (LIST_FIRST(&old_parent->p_children) != p) {
498 		/* Put child where it can be found quickly */
499 		LIST_REMOVE(p, p_sibling);
500 		LIST_INSERT_HEAD(&old_parent->p_children, p, p_sibling);
501 	}
502 
503 	/*
504 	 * Notify parent that we're gone.  If parent has the P_NOCLDWAIT
505 	 * flag set, notify init instead (and hope it will handle
506 	 * this situation).
507 	 */
508 	if (old_parent->p_flag & (PK_NOCLDWAIT|PK_CLDSIGIGN)) {
509 		proc_reparent(p, initproc);
510 		wakeinit = 1;
511 
512 		/*
513 		 * If this was the last child of our parent, notify
514 		 * parent, so in case he was wait(2)ing, he will
515 		 * continue.
516 		 */
517 		if (LIST_FIRST(&old_parent->p_children) == NULL)
518 			cv_broadcast(&old_parent->p_waitcv);
519 	}
520 
521 	/* Reload parent pointer, since p may have been reparented above */
522 	new_parent = p->p_pptr;
523 
524 	if (__predict_false((p->p_slflag & PSL_FSTRACE) == 0 &&
525 	    p->p_exitsig != 0)) {
526 		exit_psignal(p, new_parent, &ksi);
527 		kpsignal(new_parent, &ksi, NULL);
528 	}
529 
530 	/* Calculate the final rusage info.  */
531 	calcru(p, &p->p_stats->p_ru.ru_utime, &p->p_stats->p_ru.ru_stime,
532 	    NULL, NULL);
533 
534 	if (wakeinit)
535 		cv_broadcast(&initproc->p_waitcv);
536 
537 	callout_destroy(&l->l_timeout_ch);
538 
539 	/*
540 	 * Release any PCU resources before becoming a zombie.
541 	 */
542 	pcu_discard_all(l);
543 
544 	mutex_enter(p->p_lock);
545 	/* Free the linux lwp id */
546 	if ((l->l_pflag & LP_PIDLID) != 0 && l->l_lid != p->p_pid)
547 		proc_free_pid(l->l_lid);
548 	lwp_drainrefs(l);
549 	lwp_lock(l);
550 	l->l_prflag &= ~LPR_DETACHED;
551 	l->l_stat = LSZOMB;
552 	lwp_unlock(l);
553 	KASSERT(curlwp == l);
554 	KASSERT(p->p_nrlwps == 1);
555 	KASSERT(p->p_nlwps == 1);
556 	p->p_stat = SZOMB;
557 	p->p_nrlwps--;
558 	p->p_nzlwps++;
559 	p->p_ndlwps = 0;
560 	mutex_exit(p->p_lock);
561 
562 	/*
563 	 * Signal the parent to collect us, and drop the proclist lock.
564 	 * Drop debugger/procfs lock; no new references can be gained.
565 	 */
566 	cv_broadcast(&p->p_pptr->p_waitcv);
567 	rw_exit(&p->p_reflock);
568 	mutex_exit(proc_lock);
569 
570 	/* Verify that we hold no locks other than the kernel lock. */
571 	LOCKDEBUG_BARRIER(&kernel_lock, 0);
572 
573 	/*
574 	 * NOTE: WE ARE NO LONGER ALLOWED TO SLEEP!
575 	 */
576 
577 	/*
578 	 * Give machine-dependent code a chance to free any MD LWP
579 	 * resources.  This must be done before uvm_lwp_exit(), in
580 	 * case these resources are in the PCB.
581 	 */
582 	cpu_lwp_free(l, 1);
583 
584 	pmap_deactivate(l);
585 
586 	/* This process no longer needs to hold the kernel lock. */
587 #ifdef notyet
588 	/* XXXSMP hold in lwp_userret() */
589 	KERNEL_UNLOCK_LAST(l);
590 #else
591 	KERNEL_UNLOCK_ALL(l, NULL);
592 #endif
593 
594 	lwp_exit_switchaway(l);
595 }
596 
597 void
598 exit_lwps(struct lwp *l)
599 {
600 	proc_t *p = l->l_proc;
601 	lwp_t *l2;
602 	int nlocks;
603 
604 	KERNEL_UNLOCK_ALL(l, &nlocks);
605 retry:
606 	KASSERT(mutex_owned(p->p_lock));
607 
608 	/*
609 	 * Interrupt LWPs in interruptable sleep, unsuspend suspended
610 	 * LWPs and then wait for everyone else to finish.
611 	 */
612 	LIST_FOREACH(l2, &p->p_lwps, l_sibling) {
613 		if (l2 == l)
614 			continue;
615 		lwp_lock(l2);
616 		l2->l_flag |= LW_WEXIT;
617 		if ((l2->l_stat == LSSLEEP && (l2->l_flag & LW_SINTR)) ||
618 		    l2->l_stat == LSSUSPENDED || l2->l_stat == LSSTOP) {
619 		    	/* setrunnable() will release the lock. */
620 			setrunnable(l2);
621 			continue;
622 		}
623 		lwp_unlock(l2);
624 	}
625 
626 	/*
627 	 * Wait for every LWP to exit.  Note: LWPs can get suspended/slept
628 	 * behind us or there may even be new LWPs created.  Therefore, a
629 	 * full retry is required on error.
630 	 */
631 	while (p->p_nlwps > 1) {
632 		if (lwp_wait(l, 0, NULL, true)) {
633 			goto retry;
634 		}
635 	}
636 
637 	KERNEL_LOCK(nlocks, l);
638 	KASSERT(p->p_nlwps == 1);
639 }
640 
641 int
642 do_sys_wait(int *pid, int *status, int options, struct rusage *ru)
643 {
644 	proc_t *child;
645 	int error;
646 
647 	if (ru != NULL) {
648 		memset(ru, 0, sizeof(*ru));
649 	}
650 	mutex_enter(proc_lock);
651 	error = find_stopped_child(curproc, *pid, options, &child, status);
652 	if (child == NULL) {
653 		mutex_exit(proc_lock);
654 		*pid = 0;
655 		return error;
656 	}
657 	*pid = child->p_pid;
658 
659 	if (child->p_stat == SZOMB) {
660 		/* proc_free() will release the proc_lock. */
661 		if (options & WNOWAIT) {
662 			mutex_exit(proc_lock);
663 		} else {
664 			proc_free(child, ru);
665 		}
666 	} else {
667 		/* Child state must have been SSTOP. */
668 		mutex_exit(proc_lock);
669 		*status = W_STOPCODE(*status);
670 	}
671 	return 0;
672 }
673 
674 int
675 sys___wait450(struct lwp *l, const struct sys___wait450_args *uap,
676     register_t *retval)
677 {
678 	/* {
679 		syscallarg(int)			pid;
680 		syscallarg(int *)		status;
681 		syscallarg(int)			options;
682 		syscallarg(struct rusage *)	rusage;
683 	} */
684 	int error, status, pid = SCARG(uap, pid);
685 	struct rusage ru;
686 
687 	error = do_sys_wait(&pid, &status, SCARG(uap, options),
688 	    SCARG(uap, rusage) != NULL ? &ru : NULL);
689 
690 	retval[0] = pid;
691 	if (pid == 0) {
692 		return error;
693 	}
694 	if (SCARG(uap, status)) {
695 		error = copyout(&status, SCARG(uap, status), sizeof(status));
696 	}
697 	if (SCARG(uap, rusage) && error == 0) {
698 		error = copyout(&ru, SCARG(uap, rusage), sizeof(ru));
699 	}
700 	return error;
701 }
702 
703 /*
704  * Scan list of child processes for a child process that has stopped or
705  * exited.  Used by sys_wait4 and 'compat' equivalents.
706  *
707  * Must be called with the proc_lock held, and may release while waiting.
708  */
709 static int
710 find_stopped_child(struct proc *parent, pid_t pid, int options,
711 		   struct proc **child_p, int *status_p)
712 {
713 	struct proc *child, *dead;
714 	int error;
715 
716 	KASSERT(mutex_owned(proc_lock));
717 
718 	if (options & ~(WUNTRACED|WNOHANG|WALTSIG|WALLSIG)
719 	    && !(options & WOPTSCHECKED)) {
720 		*child_p = NULL;
721 		return EINVAL;
722 	}
723 
724 	if (pid == 0 && !(options & WOPTSCHECKED))
725 		pid = -parent->p_pgid;
726 
727 	for (;;) {
728 		error = ECHILD;
729 		dead = NULL;
730 
731 		LIST_FOREACH(child, &parent->p_children, p_sibling) {
732 			if (pid >= 0) {
733 				if (child->p_pid != pid) {
734 					child = proc_find_raw(pid);
735 					if (child == NULL ||
736 					    child->p_stat == SIDL ||
737 					    child->p_pptr != parent) {
738 						child = NULL;
739 						break;
740 					}
741 				}
742 			} else if (pid != WAIT_ANY && child->p_pgid != -pid) {
743 				/* Child not in correct pgrp */
744 				continue;
745 			}
746 
747 			/*
748 			 * Wait for processes with p_exitsig != SIGCHLD
749 			 * processes only if WALTSIG is set; wait for
750 			 * processes with p_exitsig == SIGCHLD only
751 			 * if WALTSIG is clear.
752 			 */
753 			if (((options & WALLSIG) == 0) &&
754 			    (options & WALTSIG ? child->p_exitsig == SIGCHLD
755 						: P_EXITSIG(child) != SIGCHLD)){
756 				if (child->p_pid == pid) {
757 					child = NULL;
758 					break;
759 				}
760 				continue;
761 			}
762 
763 			error = 0;
764 			if ((options & WNOZOMBIE) == 0) {
765 				if (child->p_stat == SZOMB)
766 					break;
767 				if (child->p_stat == SDEAD) {
768 					/*
769 					 * We may occasionally arrive here
770 					 * after receiving a signal, but
771 					 * immediately before the child
772 					 * process is zombified.  The wait
773 					 * will be short, so avoid returning
774 					 * to userspace.
775 					 */
776 					dead = child;
777 				}
778 			}
779 
780 			if (child->p_stat == SSTOP &&
781 			    child->p_waited == 0 &&
782 			    (child->p_slflag & PSL_TRACED ||
783 			    options & WUNTRACED)) {
784 				if ((options & WNOWAIT) == 0) {
785 					child->p_waited = 1;
786 					parent->p_nstopchild--;
787 				}
788 				break;
789 			}
790 			if (parent->p_nstopchild == 0 || child->p_pid == pid) {
791 				child = NULL;
792 				break;
793 			}
794 		}
795 
796 		if (child != NULL || error != 0 ||
797 		    ((options & WNOHANG) != 0 && dead == NULL)) {
798 		    	if (child != NULL) {
799 			    	*status_p = child->p_xstat;
800 			}
801 			*child_p = child;
802 			return error;
803 		}
804 
805 		/*
806 		 * Wait for another child process to stop.
807 		 */
808 		error = cv_wait_sig(&parent->p_waitcv, proc_lock);
809 
810 		if (error != 0) {
811 			*child_p = NULL;
812 			return error;
813 		}
814 	}
815 }
816 
817 /*
818  * Free a process after parent has taken all the state info.  Must be called
819  * with the proclist lock held, and will release before returning.
820  *
821  * *ru is returned to the caller, and must be freed by the caller.
822  */
823 static void
824 proc_free(struct proc *p, struct rusage *ru)
825 {
826 	struct proc *parent = p->p_pptr;
827 	struct lwp *l;
828 	ksiginfo_t ksi;
829 	kauth_cred_t cred1, cred2;
830 	uid_t uid;
831 
832 	KASSERT(mutex_owned(proc_lock));
833 	KASSERT(p->p_nlwps == 1);
834 	KASSERT(p->p_nzlwps == 1);
835 	KASSERT(p->p_nrlwps == 0);
836 	KASSERT(p->p_stat == SZOMB);
837 
838 	/*
839 	 * If we got the child via ptrace(2) or procfs, and
840 	 * the parent is different (meaning the process was
841 	 * attached, rather than run as a child), then we need
842 	 * to give it back to the old parent, and send the
843 	 * parent the exit signal.  The rest of the cleanup
844 	 * will be done when the old parent waits on the child.
845 	 */
846 	if ((p->p_slflag & PSL_TRACED) != 0 && p->p_opptr != parent) {
847 		mutex_enter(p->p_lock);
848 		p->p_slflag &= ~(PSL_TRACED|PSL_FSTRACE|PSL_SYSCALL);
849 		mutex_exit(p->p_lock);
850 		parent = (p->p_opptr == NULL) ? initproc : p->p_opptr;
851 		proc_reparent(p, parent);
852 		p->p_opptr = NULL;
853 		if (p->p_exitsig != 0) {
854 			exit_psignal(p, parent, &ksi);
855 			kpsignal(parent, &ksi, NULL);
856 		}
857 		cv_broadcast(&parent->p_waitcv);
858 		mutex_exit(proc_lock);
859 		return;
860 	}
861 
862 	sched_proc_exit(parent, p);
863 
864 	/*
865 	 * Add child times of exiting process onto its own times.
866 	 * This cannot be done any earlier else it might get done twice.
867 	 */
868 	l = LIST_FIRST(&p->p_lwps);
869 	p->p_stats->p_ru.ru_nvcsw += (l->l_ncsw - l->l_nivcsw);
870 	p->p_stats->p_ru.ru_nivcsw += l->l_nivcsw;
871 	ruadd(&p->p_stats->p_ru, &l->l_ru);
872 	ruadd(&p->p_stats->p_ru, &p->p_stats->p_cru);
873 	ruadd(&parent->p_stats->p_cru, &p->p_stats->p_ru);
874 	if (ru != NULL)
875 		*ru = p->p_stats->p_ru;
876 	p->p_xstat = 0;
877 
878 	/*
879 	 * At this point we are going to start freeing the final resources.
880 	 * If anyone tries to access the proc structure after here they will
881 	 * get a shock - bits are missing.  Attempt to make it hard!  We
882 	 * don't bother with any further locking past this point.
883 	 */
884 	p->p_stat = SIDL;		/* not even a zombie any more */
885 	LIST_REMOVE(p, p_list);	/* off zombproc */
886 	parent->p_nstopchild--;
887 	LIST_REMOVE(p, p_sibling);
888 
889 	/*
890 	 * Let pid be reallocated.
891 	 */
892 	proc_free_pid(p->p_pid);
893 
894 	/*
895 	 * Unlink process from its process group.
896 	 * Releases the proc_lock.
897 	 */
898 	proc_leavepgrp(p);
899 
900 	/*
901 	 * Delay release until after lwp_free.
902 	 */
903 	cred2 = l->l_cred;
904 
905 	/*
906 	 * Free the last LWP's resources.
907 	 *
908 	 * lwp_free ensures the LWP is no longer running on another CPU.
909 	 */
910 	lwp_free(l, false, true);
911 
912 	/*
913 	 * Now no one except us can reach the process p.
914 	 */
915 
916 	/*
917 	 * Decrement the count of procs running with this uid.
918 	 */
919 	cred1 = p->p_cred;
920 	uid = kauth_cred_getuid(cred1);
921 	(void)chgproccnt(uid, -1);
922 
923 	/*
924 	 * Release substructures.
925 	 */
926 
927 	lim_free(p->p_limit);
928 	pstatsfree(p->p_stats);
929 	kauth_cred_free(cred1);
930 	kauth_cred_free(cred2);
931 
932 	/*
933 	 * Release reference to text vnode
934 	 */
935 	if (p->p_textvp)
936 		vrele(p->p_textvp);
937 
938 	mutex_destroy(&p->p_auxlock);
939 	mutex_obj_free(p->p_lock);
940 	mutex_destroy(&p->p_stmutex);
941 	cv_destroy(&p->p_waitcv);
942 	cv_destroy(&p->p_lwpcv);
943 	rw_destroy(&p->p_reflock);
944 
945 	proc_free_mem(p);
946 }
947 
948 /*
949  * make process 'parent' the new parent of process 'child'.
950  *
951  * Must be called with proc_lock held.
952  */
953 void
954 proc_reparent(struct proc *child, struct proc *parent)
955 {
956 
957 	KASSERT(mutex_owned(proc_lock));
958 
959 	if (child->p_pptr == parent)
960 		return;
961 
962 	if (child->p_stat == SZOMB ||
963 	    (child->p_stat == SSTOP && !child->p_waited)) {
964 		child->p_pptr->p_nstopchild--;
965 		parent->p_nstopchild++;
966 	}
967 	if (parent == initproc)
968 		child->p_exitsig = SIGCHLD;
969 
970 	LIST_REMOVE(child, p_sibling);
971 	LIST_INSERT_HEAD(&parent->p_children, child, p_sibling);
972 	child->p_pptr = parent;
973 	child->p_ppid = parent->p_pid;
974 }
975