xref: /netbsd-src/sys/kern/kern_exit.c (revision 0df165c04d0a9ca1adde9ed2b890344c937954a6)
1 /*	$NetBSD: kern_exit.c,v 1.193 2007/11/12 23:11:59 ad Exp $	*/
2 
3 /*-
4  * Copyright (c) 1998, 1999, 2006, 2007 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  * 3. All advertising materials mentioning features or use of this software
20  *    must display the following acknowledgement:
21  *	This product includes software developed by the NetBSD
22  *	Foundation, Inc. and its contributors.
23  * 4. Neither the name of The NetBSD Foundation nor the names of its
24  *    contributors may be used to endorse or promote products derived
25  *    from this software without specific prior written permission.
26  *
27  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
28  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
29  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
30  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
31  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
32  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
33  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
34  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
35  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
36  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
37  * POSSIBILITY OF SUCH DAMAGE.
38  */
39 
40 /*
41  * Copyright (c) 1982, 1986, 1989, 1991, 1993
42  *	The Regents of the University of California.  All rights reserved.
43  * (c) UNIX System Laboratories, Inc.
44  * All or some portions of this file are derived from material licensed
45  * to the University of California by American Telephone and Telegraph
46  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
47  * the permission of UNIX System Laboratories, Inc.
48  *
49  * Redistribution and use in source and binary forms, with or without
50  * modification, are permitted provided that the following conditions
51  * are met:
52  * 1. Redistributions of source code must retain the above copyright
53  *    notice, this list of conditions and the following disclaimer.
54  * 2. Redistributions in binary form must reproduce the above copyright
55  *    notice, this list of conditions and the following disclaimer in the
56  *    documentation and/or other materials provided with the distribution.
57  * 3. Neither the name of the University nor the names of its contributors
58  *    may be used to endorse or promote products derived from this software
59  *    without specific prior written permission.
60  *
61  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
62  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
63  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
64  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
65  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
66  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
67  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
68  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
69  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
70  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
71  * SUCH DAMAGE.
72  *
73  *	@(#)kern_exit.c	8.10 (Berkeley) 2/23/95
74  */
75 
76 #include <sys/cdefs.h>
77 __KERNEL_RCSID(0, "$NetBSD: kern_exit.c,v 1.193 2007/11/12 23:11:59 ad Exp $");
78 
79 #include "opt_ktrace.h"
80 #include "opt_perfctrs.h"
81 #include "opt_systrace.h"
82 #include "opt_sysv.h"
83 
84 #include <sys/param.h>
85 #include <sys/aio.h>
86 #include <sys/systm.h>
87 #include <sys/ioctl.h>
88 #include <sys/tty.h>
89 #include <sys/time.h>
90 #include <sys/resource.h>
91 #include <sys/kernel.h>
92 #include <sys/proc.h>
93 #include <sys/buf.h>
94 #include <sys/wait.h>
95 #include <sys/file.h>
96 #include <sys/vnode.h>
97 #include <sys/syslog.h>
98 #include <sys/malloc.h>
99 #include <sys/pool.h>
100 #include <sys/resourcevar.h>
101 #if defined(PERFCTRS)
102 #include <sys/pmc.h>
103 #endif
104 #include <sys/ptrace.h>
105 #include <sys/acct.h>
106 #include <sys/filedesc.h>
107 #include <sys/ras.h>
108 #include <sys/signalvar.h>
109 #include <sys/sched.h>
110 #include <sys/mount.h>
111 #include <sys/syscallargs.h>
112 #include <sys/systrace.h>
113 #include <sys/kauth.h>
114 #include <sys/sleepq.h>
115 #include <sys/lockdebug.h>
116 #include <sys/ktrace.h>
117 #include <sys/cpu.h>
118 #include <sys/lwpctl.h>
119 
120 #include <uvm/uvm_extern.h>
121 
122 #define DEBUG_EXIT
123 
124 #ifdef DEBUG_EXIT
125 int debug_exit = 0;
126 #define DPRINTF(x) if (debug_exit) printf x
127 #else
128 #define DPRINTF(x)
129 #endif
130 
131 static int find_stopped_child(struct proc *, pid_t, int, struct proc **, int *);
132 static void proc_free(struct proc *, struct rusage *);
133 
134 /*
135  * Fill in the appropriate signal information, and signal the parent.
136  */
137 static void
138 exit_psignal(struct proc *p, struct proc *pp, ksiginfo_t *ksi)
139 {
140 
141 	KSI_INIT(ksi);
142 	if ((ksi->ksi_signo = P_EXITSIG(p)) == SIGCHLD) {
143 		if (WIFSIGNALED(p->p_xstat)) {
144 			if (WCOREDUMP(p->p_xstat))
145 				ksi->ksi_code = CLD_DUMPED;
146 			else
147 				ksi->ksi_code = CLD_KILLED;
148 		} else {
149 			ksi->ksi_code = CLD_EXITED;
150 		}
151 	}
152 	/*
153 	 * We fill those in, even for non-SIGCHLD.
154 	 * It's safe to access p->p_cred unlocked here.
155 	 */
156 	ksi->ksi_pid = p->p_pid;
157 	ksi->ksi_uid = kauth_cred_geteuid(p->p_cred);
158 	ksi->ksi_status = p->p_xstat;
159 	/* XXX: is this still valid? */
160 	ksi->ksi_utime = p->p_stats->p_ru.ru_utime.tv_sec;
161 	ksi->ksi_stime = p->p_stats->p_ru.ru_stime.tv_sec;
162 }
163 
164 /*
165  * exit --
166  *	Death of process.
167  */
168 int
169 sys_exit(struct lwp *l, void *v, register_t *retval)
170 {
171 	struct sys_exit_args /* {
172 		syscallarg(int)	rval;
173 	} */ *uap = v;
174 	struct proc *p = l->l_proc;
175 
176 	/* Don't call exit1() multiple times in the same process. */
177 	mutex_enter(&p->p_smutex);
178 	if (p->p_sflag & PS_WEXIT) {
179 		mutex_exit(&p->p_smutex);
180 		lwp_exit(l);
181 	}
182 
183 	/* exit1() will release the mutex. */
184 	exit1(l, W_EXITCODE(SCARG(uap, rval), 0));
185 	/* NOTREACHED */
186 	return (0);
187 }
188 
189 /*
190  * Exit: deallocate address space and other resources, change proc state
191  * to zombie, and unlink proc from allproc and parent's lists.  Save exit
192  * status and rusage for wait().  Check for child processes and orphan them.
193  *
194  * Must be called with p->p_smutex held.  Does not return.
195  */
196 void
197 exit1(struct lwp *l, int rv)
198 {
199 	struct proc	*p, *q, *nq;
200 	ksiginfo_t	ksi;
201 	ksiginfoq_t	kq;
202 	int		wakeinit;
203 
204 	p = l->l_proc;
205 
206 	KASSERT(mutex_owned(&p->p_smutex));
207 
208 	if (__predict_false(p == initproc))
209 		panic("init died (signal %d, exit %d)",
210 		    WTERMSIG(rv), WEXITSTATUS(rv));
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 	ksiginfo_queue_init(&kq);
222 
223 	/*
224 	 * If we have been asked to stop on exit, do so now.
225 	 */
226 	if (p->p_sflag & PS_STOPEXIT) {
227 		KERNEL_UNLOCK_ALL(l, &l->l_biglocks);
228 		sigclearall(p, &contsigmask, &kq);
229 		p->p_waited = 0;
230 		mb_write();
231 		p->p_stat = SSTOP;
232 		lwp_lock(l);
233 		p->p_nrlwps--;
234 		l->l_stat = LSSTOP;
235 		mutex_exit(&p->p_smutex);
236 		mi_switch(l);
237 		KERNEL_LOCK(l->l_biglocks, l);
238 	} else
239 		mutex_exit(&p->p_smutex);
240 
241 	/* Destroy any lwpctl info. */
242 	if (p->p_lwpctl != NULL)
243 		lwp_ctl_exit();
244 
245 	/* Destroy all AIO works */
246 	aio_exit(p, p->p_aio);
247 
248 	/*
249 	 * Drain all remaining references that procfs, ptrace and others may
250 	 * have on the process.
251 	 */
252 	rw_enter(&p->p_reflock, RW_WRITER);
253 
254 	/*
255 	 * Bin any remaining signals and mark the process as dying so it will
256 	 * not be found for, e.g. signals.
257 	 */
258 	mutex_enter(&p->p_smutex);
259 	sigfillset(&p->p_sigctx.ps_sigignore);
260 	sigclearall(p, NULL, &kq);
261 	p->p_stat = SDYING;
262 	mutex_exit(&p->p_smutex);
263 	ksiginfo_queue_drain(&kq);
264 
265 	DPRINTF(("exit1: %d.%d exiting.\n", p->p_pid, l->l_lid));
266 
267 #ifdef PGINPROF
268 	vmsizmon();
269 #endif
270 	timers_free(p, TIMERS_ALL);
271 #if defined(__HAVE_RAS)
272 	ras_purgeall();
273 #endif
274 
275 	/*
276 	 * Close open files, release open-file table and free signal
277 	 * actions.  This may block!
278 	 */
279 	fdfree(l);
280 	cwdfree(p->p_cwdi);
281 	p->p_cwdi = NULL;
282 	doexithooks(p);
283 	sigactsfree(p->p_sigacts);
284 
285 	/*
286 	 * Write out accounting data.
287 	 */
288 	(void)acct_process(l);
289 
290 #ifdef KTRACE
291 	/*
292 	 * Release trace file.
293 	 */
294 	if (p->p_tracep != NULL) {
295 		mutex_enter(&ktrace_lock);
296 		ktrderef(p);
297 		mutex_exit(&ktrace_lock);
298 	}
299 #endif
300 #ifdef SYSTRACE
301 	systrace_sys_exit(p);
302 #endif
303 
304 	/*
305 	 * If emulation has process exit hook, call it now.
306 	 * Set the exit status now so that the exit hook has
307 	 * an opportunity to tweak it (COMPAT_LINUX requires
308 	 * this for thread group emulation)
309 	 */
310 	p->p_xstat = rv;
311 	if (p->p_emul->e_proc_exit)
312 		(*p->p_emul->e_proc_exit)(p);
313 
314 	/* Collect child u-areas. */
315 	uvm_uarea_drain(false);
316 
317 	/*
318 	 * Free the VM resources we're still holding on to.
319 	 * We must do this from a valid thread because doing
320 	 * so may block. This frees vmspace, which we don't
321 	 * need anymore. The only remaining lwp is the one
322 	 * we run at this moment, nothing runs in userland
323 	 * anymore.
324 	 */
325 	uvm_proc_exit(p);
326 
327 	/*
328 	 * While we can still block, and mark the LWP as unswappable to
329 	 * prevent conflicts with the with the swapper.  We also shouldn't
330 	 * be swapped out, because we are about to exit and will release
331 	 * memory.
332 	 */
333 	uvm_lwp_hold(l);
334 
335 	/*
336 	 * Stop profiling.
337 	 */
338 	if ((p->p_stflag & PST_PROFIL) != 0) {
339 		mutex_spin_enter(&p->p_stmutex);
340 		stopprofclock(p);
341 		mutex_spin_exit(&p->p_stmutex);
342 	}
343 
344 	/*
345 	 * If parent is waiting for us to exit or exec, P_PPWAIT is set; we
346 	 * wake up the parent early to avoid deadlock.  We can do this once
347 	 * the VM resources are released.
348 	 */
349 	mutex_enter(&proclist_lock);
350 
351 	mutex_enter(&p->p_smutex);
352 	if (p->p_sflag & PS_PPWAIT) {
353 		p->p_sflag &= ~PS_PPWAIT;
354 		cv_broadcast(&p->p_pptr->p_waitcv);
355 	}
356 	mutex_exit(&p->p_smutex);
357 
358 	if (SESS_LEADER(p)) {
359 		struct vnode *vprele = NULL, *vprevoke = NULL;
360 		struct session *sp = p->p_session;
361 		struct tty *tp;
362 
363 		if (sp->s_ttyvp) {
364 			/*
365 			 * Controlling process.
366 			 * Signal foreground pgrp,
367 			 * drain controlling terminal
368 			 * and revoke access to controlling terminal.
369 			 */
370 			tp = sp->s_ttyp;
371 			mutex_spin_enter(&tty_lock);
372 			if (tp->t_session == sp) {
373 				if (tp->t_pgrp) {
374 					mutex_enter(&proclist_mutex);
375 					pgsignal(tp->t_pgrp, SIGHUP, 1);
376 					mutex_exit(&proclist_mutex);
377 				}
378 				/* we can't guarantee the revoke will do this */
379 				tp->t_pgrp = NULL;
380 				tp->t_session = NULL;
381 				mutex_spin_exit(&tty_lock);
382 				SESSRELE(sp);
383 				mutex_exit(&proclist_lock);
384 				(void) ttywait(tp);
385 				mutex_enter(&proclist_lock);
386 
387 				/*
388 				 * The tty could have been revoked
389 				 * if we blocked.
390 				 */
391 				vprevoke = sp->s_ttyvp;
392 			} else
393 				mutex_spin_exit(&tty_lock);
394 			vprele = sp->s_ttyvp;
395 			sp->s_ttyvp = NULL;
396 			/*
397 			 * s_ttyp is not zero'd; we use this to indicate
398 			 * that the session once had a controlling terminal.
399 			 * (for logging and informational purposes)
400 			 */
401 		}
402 		sp->s_leader = NULL;
403 
404 		if (vprevoke != NULL || vprele != NULL) {
405 			mutex_exit(&proclist_lock);
406 			if (vprevoke != NULL)
407 				VOP_REVOKE(vprevoke, REVOKEALL);
408 			if (vprele != NULL)
409 				vrele(vprele);
410 			mutex_enter(&proclist_lock);
411 		}
412 	}
413 	mutex_enter(&proclist_mutex);
414 	fixjobc(p, p->p_pgrp, 0);
415 	mutex_exit(&proclist_mutex);
416 
417 	/*
418 	 * Finalize the last LWP's specificdata, as well as the
419 	 * specificdata for the proc itself.
420 	 */
421 	lwp_finispecific(l);
422 	proc_finispecific(p);
423 
424 	/*
425 	 * Notify interested parties of our demise.
426 	 */
427 	KNOTE(&p->p_klist, NOTE_EXIT);
428 
429 #if PERFCTRS
430 	/*
431 	 * Save final PMC information in parent process & clean up.
432 	 */
433 	if (PMC_ENABLED(p)) {
434 		pmc_save_context(p);
435 		pmc_accumulate(p->p_pptr, p);
436 		pmc_process_exit(p);
437 	}
438 #endif
439 
440 	/*
441 	 * Reset p_opptr pointer of all former children which got
442 	 * traced by another process and were reparented. We reset
443 	 * it to NULL here; the trace detach code then reparents
444 	 * the child to initproc. We only check allproc list, since
445 	 * eventual former children on zombproc list won't reference
446 	 * p_opptr anymore.
447 	 */
448 	if (p->p_slflag & PSL_CHTRACED) {
449 		PROCLIST_FOREACH(q, &allproc) {
450 			if (q->p_opptr == p)
451 				q->p_opptr = NULL;
452 		}
453 	}
454 
455 	/*
456 	 * Give orphaned children to init(8).
457 	 */
458 	q = LIST_FIRST(&p->p_children);
459 	wakeinit = (q != NULL);
460 	for (; q != NULL; q = nq) {
461 		nq = LIST_NEXT(q, p_sibling);
462 
463 		/*
464 		 * Traced processes are killed since their existence
465 		 * means someone is screwing up. Since we reset the
466 		 * trace flags, the logic in sys_wait4() would not be
467 		 * triggered to reparent the process to its
468 		 * original parent, so we must do this here.
469 		 */
470 		if (q->p_slflag & PSL_TRACED) {
471 			mutex_enter(&p->p_smutex);
472 			q->p_slflag &= ~(PSL_TRACED|PSL_FSTRACE|PSL_SYSCALL);
473 			mutex_exit(&p->p_smutex);
474 			if (q->p_opptr != q->p_pptr) {
475 				struct proc *t = q->p_opptr;
476 				proc_reparent(q, t ? t : initproc);
477 				q->p_opptr = NULL;
478 			} else
479 				proc_reparent(q, initproc);
480 			killproc(q, "orphaned traced process");
481 		} else
482 			proc_reparent(q, initproc);
483 	}
484 
485 	/*
486 	 * Move proc from allproc to zombproc, it's now nearly ready to be
487 	 * collected by parent.
488 	 */
489 	mutex_enter(&proclist_mutex);
490 	LIST_REMOVE(l, l_list);
491 	LIST_REMOVE(p, p_list);
492 	LIST_INSERT_HEAD(&zombproc, p, p_list);
493 
494 	/*
495 	 * Mark the process as dead.  We must do this before we signal
496 	 * the parent.
497 	 */
498 	p->p_stat = SDEAD;
499 
500 	/* Put in front of parent's sibling list for parent to collect it */
501 	q = p->p_pptr;
502 	q->p_nstopchild++;
503 	if (LIST_FIRST(&q->p_children) != p) {
504 		/* Put child where it can be found quickly */
505 		LIST_REMOVE(p, p_sibling);
506 		LIST_INSERT_HEAD(&q->p_children, p, p_sibling);
507 	}
508 	mutex_exit(&proclist_mutex);
509 
510 	/*
511 	 * Notify parent that we're gone.  If parent has the P_NOCLDWAIT
512 	 * flag set, notify init instead (and hope it will handle
513 	 * this situation).
514 	 */
515 	mutex_enter(&q->p_mutex);
516 	if (q->p_flag & (PK_NOCLDWAIT|PK_CLDSIGIGN)) {
517 		proc_reparent(p, initproc);
518 		wakeinit = 1;
519 
520 		/*
521 		 * If this was the last child of our parent, notify
522 		 * parent, so in case he was wait(2)ing, he will
523 		 * continue.
524 		 */
525 		if (LIST_FIRST(&q->p_children) == NULL)
526 			cv_broadcast(&q->p_waitcv);
527 	}
528 	mutex_exit(&q->p_mutex);
529 
530 	/* Reload parent pointer, since p may have been reparented above */
531 	q = p->p_pptr;
532 
533 	if ((p->p_slflag & PSL_FSTRACE) == 0 && p->p_exitsig != 0) {
534 		exit_psignal(p, q, &ksi);
535 		mutex_enter(&proclist_mutex);
536 		kpsignal(q, &ksi, NULL);
537 		mutex_exit(&proclist_mutex);
538 	}
539 
540 	/* Calculate the final rusage info.  */
541 	calcru(p, &p->p_stats->p_ru.ru_utime, &p->p_stats->p_ru.ru_stime,
542 	    NULL, NULL);
543 
544 	if (wakeinit)
545 		cv_broadcast(&initproc->p_waitcv);
546 
547 	callout_destroy(&l->l_timeout_ch);
548 
549 	/*
550 	 * Remaining lwp resources will be freed in lwp_exit2() once we've
551 	 * switch to idle context; at that point, we will be marked as a
552 	 * full blown zombie.
553 	 *
554 	 * XXXSMP disable preemption.
555 	 */
556 	mutex_enter(&p->p_smutex);
557 	lwp_drainrefs(l);
558 	lwp_lock(l);
559 	l->l_prflag &= ~LPR_DETACHED;
560 	l->l_stat = LSZOMB;
561 	lwp_unlock(l);
562 	KASSERT(curlwp == l);
563 	KASSERT(p->p_nrlwps == 1);
564 	KASSERT(p->p_nlwps == 1);
565 	p->p_stat = SZOMB;
566 	p->p_nrlwps--;
567 	p->p_nzlwps++;
568 	p->p_ndlwps = 0;
569 	mutex_exit(&p->p_smutex);
570 
571 	/*
572 	 * Signal the parent to collect us, and drop the proclist lock.
573 	 * Drop debugger/procfs lock; no new references can be gained.
574 	 */
575 	cv_broadcast(&p->p_pptr->p_waitcv);
576 	mutex_exit(&proclist_lock);
577 	rw_exit(&p->p_reflock);
578 
579 	/* Verify that we hold no locks other than the kernel lock. */
580 #ifdef MULTIPROCESSOR
581 	LOCKDEBUG_BARRIER(&kernel_lock, 0);
582 #else
583 	LOCKDEBUG_BARRIER(NULL, 0);
584 #endif
585 
586 	/*
587 	 * NOTE: WE ARE NO LONGER ALLOWED TO SLEEP!
588 	 */
589 
590 	/*
591 	 * Give machine-dependent code a chance to free any MD LWP
592 	 * resources.  This must be done before uvm_lwp_exit(), in
593 	 * case these resources are in the PCB.
594 	 */
595 #ifndef __NO_CPU_LWP_FREE
596 	cpu_lwp_free(l, 1);
597 #endif
598 	pmap_deactivate(l);
599 
600 	/* This process no longer needs to hold the kernel lock. */
601 #ifdef notyet
602 	/* XXXSMP hold in lwp_userret() */
603 	KERNEL_UNLOCK_LAST(l);
604 #else
605 	KERNEL_UNLOCK_ALL(l, NULL);
606 #endif
607 
608 	lwp_exit_switchaway(l);
609 }
610 
611 void
612 exit_lwps(struct lwp *l)
613 {
614 	struct proc *p;
615 	struct lwp *l2;
616 	int error;
617 	lwpid_t waited;
618 #if defined(MULTIPROCESSOR)
619 	int nlocks;
620 #endif
621 
622 	KERNEL_UNLOCK_ALL(l, &nlocks);
623 
624 	p = l->l_proc;
625 	KASSERT(mutex_owned(&p->p_smutex));
626 
627  retry:
628 	/*
629 	 * Interrupt LWPs in interruptable sleep, unsuspend suspended
630 	 * LWPs and then wait for everyone else to finish.
631 	 */
632 	LIST_FOREACH(l2, &p->p_lwps, l_sibling) {
633 		if (l2 == l)
634 			continue;
635 		lwp_lock(l2);
636 		l2->l_flag |= LW_WEXIT;
637 		if ((l2->l_stat == LSSLEEP && (l2->l_flag & LW_SINTR)) ||
638 		    l2->l_stat == LSSUSPENDED || l2->l_stat == LSSTOP) {
639 		    	/* setrunnable() will release the lock. */
640 			setrunnable(l2);
641 			DPRINTF(("exit_lwps: Made %d.%d runnable\n",
642 			    p->p_pid, l2->l_lid));
643 			continue;
644 		}
645 		lwp_unlock(l2);
646 	}
647 	while (p->p_nlwps > 1) {
648 		DPRINTF(("exit_lwps: waiting for %d LWPs (%d zombies)\n",
649 		    p->p_nlwps, p->p_nzlwps));
650 		error = lwp_wait1(l, 0, &waited, LWPWAIT_EXITCONTROL);
651 		if (p->p_nlwps == 1)
652 			break;
653 		if (error == EDEADLK) {
654 			/*
655 			 * LWPs can get suspended/slept behind us.
656 			 * (eg. sa_setwoken)
657 			 * kick them again and retry.
658 			 */
659 			goto retry;
660 		}
661 		if (error)
662 			panic("exit_lwps: lwp_wait1 failed with error %d",
663 			    error);
664 		DPRINTF(("exit_lwps: Got LWP %d from lwp_wait1()\n", waited));
665 	}
666 
667 #if defined(MULTIPROCESSOR)
668 	if (nlocks > 0) {
669 		mutex_exit(&p->p_smutex);
670 		KERNEL_LOCK(nlocks, l);
671 		mutex_enter(&p->p_smutex);
672 	}
673 #endif /* defined(MULTIPROCESSOR) */
674 	KASSERT(p->p_nlwps == 1);
675 }
676 
677 int
678 do_sys_wait(struct lwp *l, int *pid, int *status, int options,
679     struct rusage *ru, int *was_zombie)
680 {
681 	struct proc	*child;
682 	int		error;
683 
684 	mutex_enter(&proclist_lock);
685 
686 	error = find_stopped_child(l->l_proc, *pid, options, &child, status);
687 
688 	if (child == NULL) {
689 		mutex_exit(&proclist_lock);
690 		*pid = 0;
691 		return error;
692 	}
693 
694 	*pid = child->p_pid;
695 
696 	if (child->p_stat == SZOMB) {
697 		/* proc_free() will release the proclist_lock. */
698 		*was_zombie = 1;
699 		if (options & WNOWAIT)
700 			mutex_exit(&proclist_lock);
701 		else {
702 			KERNEL_LOCK(1, l);		/* XXXSMP */
703 			proc_free(child, ru);
704 			KERNEL_UNLOCK_ONE(l);		/* XXXSMP */
705 		}
706 	} else {
707 		/* Child state must have been SSTOP. */
708 		*was_zombie = 0;
709 		mutex_exit(&proclist_lock);
710 		*status = W_STOPCODE(*status);
711 	}
712 
713 	return 0;
714 }
715 
716 int
717 sys_wait4(struct lwp *l, void *v, register_t *retval)
718 {
719 	struct sys_wait4_args /* {
720 		syscallarg(int)			pid;
721 		syscallarg(int *)		status;
722 		syscallarg(int)			options;
723 		syscallarg(struct rusage *)	rusage;
724 	} */ *uap = v;
725 	int		status, error;
726 	int		was_zombie;
727 	struct rusage	ru;
728 
729 	error = do_sys_wait(l, &SCARG(uap, pid), &status, SCARG(uap, options),
730 	    SCARG(uap, rusage) != NULL ? &ru : NULL, &was_zombie);
731 
732 	retval[0] = SCARG(uap, pid);
733 	if (SCARG(uap, pid) == 0)
734 		return error;
735 
736 	if (SCARG(uap, rusage))
737 		error = copyout(&ru, SCARG(uap, rusage), sizeof(ru));
738 
739 	if (error == 0 && SCARG(uap, status))
740 		error = copyout(&status, SCARG(uap, status), sizeof(status));
741 
742 	return error;
743 }
744 
745 /*
746  * Scan list of child processes for a child process that has stopped or
747  * exited.  Used by sys_wait4 and 'compat' equivalents.
748  *
749  * Must be called with the proclist_lock held, and may release
750  * while waiting.
751  */
752 static int
753 find_stopped_child(struct proc *parent, pid_t pid, int options,
754 		   struct proc **child_p, int *status_p)
755 {
756 	struct proc *child, *dead;
757 	int error;
758 
759 	KASSERT(mutex_owned(&proclist_lock));
760 
761 	if (options & ~(WUNTRACED|WNOHANG|WALTSIG|WALLSIG)
762 	    && !(options & WOPTSCHECKED)) {
763 		*child_p = NULL;
764 		return EINVAL;
765 	}
766 
767 	if (pid == 0 && !(options & WOPTSCHECKED))
768 		pid = -parent->p_pgid;
769 
770 	for (;;) {
771 		error = ECHILD;
772 		dead = NULL;
773 
774 		mutex_enter(&proclist_mutex);
775 		LIST_FOREACH(child, &parent->p_children, p_sibling) {
776 			if (pid >= 0) {
777 				if (child->p_pid != pid) {
778 					child = p_find(pid, PFIND_ZOMBIE |
779 					    PFIND_LOCKED);
780 					if (child == NULL ||
781 					    child->p_pptr != parent) {
782 						child = NULL;
783 						break;
784 					}
785 				}
786 			} else if (pid != WAIT_ANY && child->p_pgid != -pid) {
787 				/* Child not in correct pgrp */
788 				continue;
789 			}
790 
791 			/*
792 			 * Wait for processes with p_exitsig != SIGCHLD
793 			 * processes only if WALTSIG is set; wait for
794 			 * processes with p_exitsig == SIGCHLD only
795 			 * if WALTSIG is clear.
796 			 */
797 			if (((options & WALLSIG) == 0) &&
798 			    (options & WALTSIG ? child->p_exitsig == SIGCHLD
799 						: P_EXITSIG(child) != SIGCHLD)){
800 				if (child->p_pid == pid) {
801 					child = NULL;
802 					break;
803 				}
804 				continue;
805 			}
806 
807 			error = 0;
808 			if ((options & WNOZOMBIE) == 0) {
809 				if (child->p_stat == SZOMB)
810 					break;
811 				if (child->p_stat == SDEAD) {
812 					/*
813 					 * We may occasionally arrive here
814 					 * after receiving a signal, but
815 					 * immediatley before the child
816 					 * process is zombified.  The wait
817 					 * will be short, so avoid returning
818 					 * to userspace.
819 					 */
820 					dead = child;
821 				}
822 			}
823 
824 			if (child->p_stat == SSTOP &&
825 			    child->p_waited == 0 &&
826 			    (child->p_slflag & PSL_TRACED ||
827 			    options & WUNTRACED)) {
828 				if ((options & WNOWAIT) == 0) {
829 					child->p_waited = 1;
830 					parent->p_nstopchild--;
831 				}
832 				break;
833 			}
834 			if (parent->p_nstopchild == 0 || child->p_pid == pid) {
835 				child = NULL;
836 				break;
837 			}
838 		}
839 
840 		if (child != NULL || error != 0 ||
841 		    ((options & WNOHANG) != 0 && dead == NULL)) {
842 		    	if (child != NULL) {
843 			    	*status_p = child->p_xstat;
844 			}
845 			mutex_exit(&proclist_mutex);
846 			*child_p = child;
847 			return error;
848 		}
849 
850 		/*
851 		 * Wait for another child process to stop.
852 		 */
853 		mutex_exit(&proclist_lock);
854 		error = cv_wait_sig(&parent->p_waitcv, &proclist_mutex);
855 		mutex_exit(&proclist_mutex);
856 		mutex_enter(&proclist_lock);
857 
858 		if (error != 0) {
859 			*child_p = NULL;
860 			return error;
861 		}
862 	}
863 }
864 
865 /*
866  * Free a process after parent has taken all the state info.  Must be called
867  * with the proclist lock held, and will release before returning.
868  *
869  * *ru is returned to the caller, and must be freed by the caller.
870  */
871 static void
872 proc_free(struct proc *p, struct rusage *ru)
873 {
874 	struct proc *parent;
875 	struct lwp *l;
876 	ksiginfo_t ksi;
877 	kauth_cred_t cred1, cred2;
878 	uid_t uid;
879 
880 	KASSERT(mutex_owned(&proclist_lock));
881 	KASSERT(p->p_nlwps == 1);
882 	KASSERT(p->p_nzlwps == 1);
883 	KASSERT(p->p_nrlwps == 0);
884 	KASSERT(p->p_stat == SZOMB);
885 
886 	/*
887 	 * If we got the child via ptrace(2) or procfs, and
888 	 * the parent is different (meaning the process was
889 	 * attached, rather than run as a child), then we need
890 	 * to give it back to the old parent, and send the
891 	 * parent the exit signal.  The rest of the cleanup
892 	 * will be done when the old parent waits on the child.
893 	 */
894 	if ((p->p_slflag & PSL_TRACED) != 0) {
895 		parent = p->p_pptr;
896 		if (p->p_opptr != parent){
897 			mutex_enter(&p->p_smutex);
898 			p->p_slflag &= ~(PSL_TRACED|PSL_FSTRACE|PSL_SYSCALL);
899 			mutex_exit(&p->p_smutex);
900 			parent = p->p_opptr;
901 			if (parent == NULL)
902 				parent = initproc;
903 			proc_reparent(p, parent);
904 			p->p_opptr = NULL;
905 			if (p->p_exitsig != 0) {
906 				exit_psignal(p, parent, &ksi);
907 				mutex_enter(&proclist_mutex);
908 				kpsignal(parent, &ksi, NULL);
909 				mutex_exit(&proclist_mutex);
910 			}
911 			cv_broadcast(&parent->p_waitcv);
912 			mutex_exit(&proclist_lock);
913 			return;
914 		}
915 	}
916 
917 	/*
918 	 * Finally finished with old proc entry.  Unlink it from its process
919 	 * group.
920 	 */
921 	leavepgrp(p);
922 
923 	parent = p->p_pptr;
924 	sched_proc_exit(parent, p);
925 	/*
926 	 * Add child times of exiting process onto its own times.
927 	 * This cannot be done any earlier else it might get done twice.
928 	 */
929 	ruadd(&p->p_stats->p_ru, &p->p_stats->p_cru);
930 	ruadd(&parent->p_stats->p_cru, &p->p_stats->p_ru);
931 	if (ru != NULL)
932 		*ru = p->p_stats->p_ru;
933 	p->p_xstat = 0;
934 
935 	/*
936 	 * At this point we are going to start freeing the final resources.
937 	 * If anyone tries to access the proc structure after here they will
938 	 * get a shock - bits are missing.  Attempt to make it hard!  We
939 	 * don't bother with any further locking past this point.
940 	 */
941 	mutex_enter(&proclist_mutex);
942 	p->p_stat = SIDL;		/* not even a zombie any more */
943 	LIST_REMOVE(p, p_list);	/* off zombproc */
944 	parent = p->p_pptr;
945 	p->p_pptr->p_nstopchild--;
946 	mutex_exit(&proclist_mutex);
947 	LIST_REMOVE(p, p_sibling);
948 
949 	/*
950 	 * Let pid be reallocated.
951 	 */
952 	proc_free_pid(p);
953 	mutex_exit(&proclist_lock);
954 
955 	l = LIST_FIRST(&p->p_lwps);
956 
957 	/*
958 	 * Delay release until after lwp_free.
959 	 */
960 	cred2 = l->l_cred;
961 
962 	/*
963 	 * Free the last LWP's resources.
964 	 *
965 	 * lwp_free ensures the LWP is no longer running on another CPU.
966 	 */
967 	lwp_free(l, false, true);
968 
969 	/*
970 	 * Now no one except us can reach the process p.
971 	 */
972 
973 	/*
974 	 * Decrement the count of procs running with this uid.
975 	 */
976 	cred1 = p->p_cred;
977 	uid = kauth_cred_getuid(cred1);
978 	(void)chgproccnt(uid, -1);
979 
980 	/*
981 	 * Release substructures.
982 	 */
983 
984 	limfree(p->p_limit);
985 	pstatsfree(p->p_stats);
986 	kauth_cred_free(cred1);
987 	kauth_cred_free(cred2);
988 
989 	/*
990 	 * Release reference to text vnode
991 	 */
992 	if (p->p_textvp)
993 		vrele(p->p_textvp);
994 
995 	mutex_destroy(&p->p_raslock);
996 	mutex_destroy(&p->p_mutex);
997 	mutex_destroy(&p->p_stmutex);
998 	mutex_destroy(&p->p_smutex);
999 	cv_destroy(&p->p_waitcv);
1000 	cv_destroy(&p->p_lwpcv);
1001 	rw_destroy(&p->p_reflock);
1002 
1003 	pool_put(&proc_pool, p);
1004 
1005 	/*
1006 	 * Collect child u-areas.
1007 	 */
1008 	uvm_uarea_drain(false);
1009 }
1010 
1011 /*
1012  * make process 'parent' the new parent of process 'child'.
1013  *
1014  * Must be called with proclist_lock lock held.
1015  */
1016 void
1017 proc_reparent(struct proc *child, struct proc *parent)
1018 {
1019 
1020 	KASSERT(mutex_owned(&proclist_lock));
1021 
1022 	if (child->p_pptr == parent)
1023 		return;
1024 
1025 	mutex_enter(&proclist_mutex);
1026 	if (child->p_stat == SZOMB ||
1027 	    (child->p_stat == SSTOP && !child->p_waited)) {
1028 		child->p_pptr->p_nstopchild--;
1029 		parent->p_nstopchild++;
1030 	}
1031 	mutex_exit(&proclist_mutex);
1032 	if (parent == initproc)
1033 		child->p_exitsig = SIGCHLD;
1034 
1035 	LIST_REMOVE(child, p_sibling);
1036 	LIST_INSERT_HEAD(&parent->p_children, child, p_sibling);
1037 	child->p_pptr = parent;
1038 }
1039