xref: /netbsd-src/sys/kern/kern_exit.c (revision 4b71a66d0f279143147d63ebfcfd8a59499a3684)
1 /*	$NetBSD: kern_exit.c,v 1.209 2008/05/19 17:06:02 ad 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 1.209 2008/05/19 17:06:02 ad 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/aio.h>
78 #include <sys/systm.h>
79 #include <sys/ioctl.h>
80 #include <sys/tty.h>
81 #include <sys/time.h>
82 #include <sys/resource.h>
83 #include <sys/kernel.h>
84 #include <sys/proc.h>
85 #include <sys/buf.h>
86 #include <sys/wait.h>
87 #include <sys/file.h>
88 #include <sys/vnode.h>
89 #include <sys/syslog.h>
90 #include <sys/malloc.h>
91 #include <sys/pool.h>
92 #include <sys/resourcevar.h>
93 #if defined(PERFCTRS)
94 #include <sys/pmc.h>
95 #endif
96 #include <sys/ptrace.h>
97 #include <sys/acct.h>
98 #include <sys/filedesc.h>
99 #include <sys/ras.h>
100 #include <sys/signalvar.h>
101 #include <sys/sched.h>
102 #include <sys/mount.h>
103 #include <sys/syscallargs.h>
104 #include <sys/kauth.h>
105 #include <sys/sleepq.h>
106 #include <sys/lockdebug.h>
107 #include <sys/ktrace.h>
108 #include <sys/cpu.h>
109 #include <sys/lwpctl.h>
110 #include <sys/atomic.h>
111 
112 #include <uvm/uvm_extern.h>
113 
114 #define DEBUG_EXIT
115 
116 #ifdef DEBUG_EXIT
117 int debug_exit = 0;
118 #define DPRINTF(x) if (debug_exit) printf x
119 #else
120 #define DPRINTF(x)
121 #endif
122 
123 static int find_stopped_child(struct proc *, pid_t, int, struct proc **, int *);
124 static void proc_free(struct proc *, struct rusage *);
125 
126 /*
127  * Fill in the appropriate signal information, and signal the parent.
128  */
129 static void
130 exit_psignal(struct proc *p, struct proc *pp, ksiginfo_t *ksi)
131 {
132 
133 	KSI_INIT(ksi);
134 	if ((ksi->ksi_signo = P_EXITSIG(p)) == SIGCHLD) {
135 		if (WIFSIGNALED(p->p_xstat)) {
136 			if (WCOREDUMP(p->p_xstat))
137 				ksi->ksi_code = CLD_DUMPED;
138 			else
139 				ksi->ksi_code = CLD_KILLED;
140 		} else {
141 			ksi->ksi_code = CLD_EXITED;
142 		}
143 	}
144 	/*
145 	 * We fill those in, even for non-SIGCHLD.
146 	 * It's safe to access p->p_cred unlocked here.
147 	 */
148 	ksi->ksi_pid = p->p_pid;
149 	ksi->ksi_uid = kauth_cred_geteuid(p->p_cred);
150 	ksi->ksi_status = p->p_xstat;
151 	/* XXX: is this still valid? */
152 	ksi->ksi_utime = p->p_stats->p_ru.ru_utime.tv_sec;
153 	ksi->ksi_stime = p->p_stats->p_ru.ru_stime.tv_sec;
154 }
155 
156 /*
157  * exit --
158  *	Death of process.
159  */
160 int
161 sys_exit(struct lwp *l, const struct sys_exit_args *uap, register_t *retval)
162 {
163 	/* {
164 		syscallarg(int)	rval;
165 	} */
166 	struct proc *p = l->l_proc;
167 
168 	/* Don't call exit1() multiple times in the same process. */
169 	mutex_enter(p->p_lock);
170 	if (p->p_sflag & PS_WEXIT) {
171 		mutex_exit(p->p_lock);
172 		lwp_exit(l);
173 	}
174 
175 	/* exit1() will release the mutex. */
176 	exit1(l, W_EXITCODE(SCARG(uap, rval), 0));
177 	/* NOTREACHED */
178 	return (0);
179 }
180 
181 /*
182  * Exit: deallocate address space and other resources, change proc state
183  * to zombie, and unlink proc from allproc and parent's lists.  Save exit
184  * status and rusage for wait().  Check for child processes and orphan them.
185  *
186  * Must be called with p->p_lock held.  Does not return.
187  */
188 void
189 exit1(struct lwp *l, int rv)
190 {
191 	struct proc	*p, *q, *nq;
192 	struct pgrp	*pgrp;
193 	ksiginfo_t	ksi;
194 	ksiginfoq_t	kq;
195 	int		wakeinit;
196 
197 	p = l->l_proc;
198 
199 	KASSERT(mutex_owned(p->p_lock));
200 
201 	if (__predict_false(p == initproc))
202 		panic("init died (signal %d, exit %d)",
203 		    WTERMSIG(rv), WEXITSTATUS(rv));
204 
205 	p->p_sflag |= PS_WEXIT;
206 
207 	/*
208 	 * Force all other LWPs to exit before we do.  Only then can we
209 	 * begin to tear down the rest of the process state.
210 	 */
211 	if (p->p_nlwps > 1)
212 		exit_lwps(l);
213 
214 	ksiginfo_queue_init(&kq);
215 
216 	/*
217 	 * If we have been asked to stop on exit, do so now.
218 	 */
219 	if (p->p_sflag & PS_STOPEXIT) {
220 		KERNEL_UNLOCK_ALL(l, &l->l_biglocks);
221 		sigclearall(p, &contsigmask, &kq);
222 		p->p_waited = 0;
223 		membar_producer();
224 		p->p_stat = SSTOP;
225 		lwp_lock(l);
226 		p->p_nrlwps--;
227 		l->l_stat = LSSTOP;
228 		mutex_exit(p->p_lock);
229 		mi_switch(l);
230 		KERNEL_LOCK(l->l_biglocks, l);
231 	} else
232 		mutex_exit(p->p_lock);
233 
234 	/* Destroy any lwpctl info. */
235 	if (p->p_lwpctl != NULL)
236 		lwp_ctl_exit();
237 
238 	/* Destroy all AIO works */
239 	aio_exit(p, p->p_aio);
240 
241 	/*
242 	 * Drain all remaining references that procfs, ptrace and others may
243 	 * have on the process.
244 	 */
245 	rw_enter(&p->p_reflock, RW_WRITER);
246 
247 	/*
248 	 * Bin any remaining signals and mark the process as dying so it will
249 	 * not be found for, e.g. signals.
250 	 */
251 	mutex_enter(p->p_lock);
252 	sigfillset(&p->p_sigctx.ps_sigignore);
253 	sigclearall(p, NULL, &kq);
254 	p->p_stat = SDYING;
255 	mutex_exit(p->p_lock);
256 	ksiginfo_queue_drain(&kq);
257 
258 	DPRINTF(("exit1: %d.%d exiting.\n", p->p_pid, l->l_lid));
259 
260 #ifdef PGINPROF
261 	vmsizmon();
262 #endif
263 	timers_free(p, TIMERS_ALL);
264 #if defined(__HAVE_RAS)
265 	ras_purgeall();
266 #endif
267 
268 	/*
269 	 * Close open files, release open-file table and free signal
270 	 * actions.  This may block!
271 	 */
272 	fd_free();
273 	cwdfree(p->p_cwdi);
274 	p->p_cwdi = NULL;
275 	doexithooks(p);
276 	sigactsfree(p->p_sigacts);
277 
278 	/*
279 	 * Write out accounting data.
280 	 */
281 	(void)acct_process(l);
282 
283 #ifdef KTRACE
284 	/*
285 	 * Release trace file.
286 	 */
287 	if (p->p_tracep != NULL) {
288 		mutex_enter(&ktrace_lock);
289 		ktrderef(p);
290 		mutex_exit(&ktrace_lock);
291 	}
292 #endif
293 
294 	/*
295 	 * If emulation has process exit hook, call it now.
296 	 * Set the exit status now so that the exit hook has
297 	 * an opportunity to tweak it (COMPAT_LINUX requires
298 	 * this for thread group emulation)
299 	 */
300 	p->p_xstat = rv;
301 	if (p->p_emul->e_proc_exit)
302 		(*p->p_emul->e_proc_exit)(p);
303 
304 	/*
305 	 * Free the VM resources we're still holding on to.
306 	 * We must do this from a valid thread because doing
307 	 * so may block. This frees vmspace, which we don't
308 	 * need anymore. The only remaining lwp is the one
309 	 * we run at this moment, nothing runs in userland
310 	 * anymore.
311 	 */
312 	uvm_proc_exit(p);
313 
314 	/*
315 	 * While we can still block, and mark the LWP as unswappable to
316 	 * prevent conflicts with the with the swapper.  We also shouldn't
317 	 * be swapped out, because we are about to exit and will release
318 	 * memory.
319 	 */
320 	uvm_lwp_hold(l);
321 
322 	/*
323 	 * Stop profiling.
324 	 */
325 	if ((p->p_stflag & PST_PROFIL) != 0) {
326 		mutex_spin_enter(&p->p_stmutex);
327 		stopprofclock(p);
328 		mutex_spin_exit(&p->p_stmutex);
329 	}
330 
331 	/*
332 	 * If parent is waiting for us to exit or exec, P_PPWAIT is set; we
333 	 * wake up the parent early to avoid deadlock.  We can do this once
334 	 * the VM resources are released.
335 	 */
336 	mutex_enter(proc_lock);
337 	mutex_enter(p->p_lock);
338 	if (p->p_sflag & PS_PPWAIT) {
339 		p->p_sflag &= ~PS_PPWAIT;
340 		cv_broadcast(&p->p_pptr->p_waitcv);
341 	}
342 	mutex_exit(p->p_lock);
343 
344 	if (SESS_LEADER(p)) {
345 		struct vnode *vprele = NULL, *vprevoke = NULL;
346 		struct session *sp = p->p_session;
347 		struct tty *tp;
348 
349 		if (sp->s_ttyvp) {
350 			/*
351 			 * Controlling process.
352 			 * Signal foreground pgrp,
353 			 * drain controlling terminal
354 			 * and revoke access to controlling terminal.
355 			 */
356 			tp = sp->s_ttyp;
357 			mutex_spin_enter(&tty_lock);
358 			if (tp->t_session == sp) {
359 				/* we can't guarantee the revoke will do this */
360 				pgrp = tp->t_pgrp;
361 				tp->t_pgrp = NULL;
362 				tp->t_session = NULL;
363 				mutex_spin_exit(&tty_lock);
364 				if (pgrp != NULL) {
365 					pgsignal(pgrp, SIGHUP, 1);
366 				}
367 				mutex_exit(proc_lock);
368 				(void) ttywait(tp);
369 				mutex_enter(proc_lock);
370 
371 				/* The tty could have been revoked. */
372 				vprevoke = sp->s_ttyvp;
373 			} else
374 				mutex_spin_exit(&tty_lock);
375 			vprele = sp->s_ttyvp;
376 			sp->s_ttyvp = NULL;
377 			/*
378 			 * s_ttyp is not zero'd; we use this to indicate
379 			 * that the session once had a controlling terminal.
380 			 * (for logging and informational purposes)
381 			 */
382 		}
383 		sp->s_leader = NULL;
384 
385 		if (vprevoke != NULL || vprele != NULL) {
386 			if (vprevoke != NULL) {
387 				SESSRELE(sp);
388 				mutex_exit(proc_lock);
389 				VOP_REVOKE(vprevoke, REVOKEALL);
390 			} else
391 				mutex_exit(proc_lock);
392 			if (vprele != NULL)
393 				vrele(vprele);
394 			mutex_enter(proc_lock);
395 		}
396 	}
397 	fixjobc(p, p->p_pgrp, 0);
398 
399 	/*
400 	 * Finalize the last LWP's specificdata, as well as the
401 	 * specificdata for the proc itself.
402 	 */
403 	lwp_finispecific(l);
404 	proc_finispecific(p);
405 
406 	/*
407 	 * Notify interested parties of our demise.
408 	 */
409 	KNOTE(&p->p_klist, NOTE_EXIT);
410 
411 
412 
413 #if PERFCTRS
414 	/*
415 	 * Save final PMC information in parent process & clean up.
416 	 */
417 	if (PMC_ENABLED(p)) {
418 		pmc_save_context(p);
419 		pmc_accumulate(p->p_pptr, p);
420 		pmc_process_exit(p);
421 	}
422 #endif
423 
424 	/*
425 	 * Reset p_opptr pointer of all former children which got
426 	 * traced by another process and were reparented. We reset
427 	 * it to NULL here; the trace detach code then reparents
428 	 * the child to initproc. We only check allproc list, since
429 	 * eventual former children on zombproc list won't reference
430 	 * p_opptr anymore.
431 	 */
432 	if (__predict_false(p->p_slflag & PSL_CHTRACED)) {
433 		PROCLIST_FOREACH(q, &allproc) {
434 			if ((q->p_flag & PK_MARKER) != 0)
435 				continue;
436 			if (q->p_opptr == p)
437 				q->p_opptr = NULL;
438 		}
439 	}
440 
441 	/*
442 	 * Give orphaned children to init(8).
443 	 */
444 	q = LIST_FIRST(&p->p_children);
445 	wakeinit = (q != NULL);
446 	for (; q != NULL; q = nq) {
447 		nq = LIST_NEXT(q, p_sibling);
448 
449 		/*
450 		 * Traced processes are killed since their existence
451 		 * means someone is screwing up. Since we reset the
452 		 * trace flags, the logic in sys_wait4() would not be
453 		 * triggered to reparent the process to its
454 		 * original parent, so we must do this here.
455 		 */
456 		if (q->p_slflag & PSL_TRACED) {
457 			mutex_enter(p->p_lock);
458 			q->p_slflag &= ~(PSL_TRACED|PSL_FSTRACE|PSL_SYSCALL);
459 			mutex_exit(p->p_lock);
460 			if (q->p_opptr != q->p_pptr) {
461 				struct proc *t = q->p_opptr;
462 				proc_reparent(q, t ? t : initproc);
463 				q->p_opptr = NULL;
464 			} else
465 				proc_reparent(q, initproc);
466 			killproc(q, "orphaned traced process");
467 		} else
468 			proc_reparent(q, initproc);
469 	}
470 
471 	/*
472 	 * Move proc from allproc to zombproc, it's now nearly ready to be
473 	 * collected by parent.
474 	 */
475 	LIST_REMOVE(l, l_list);
476 	LIST_REMOVE(p, p_list);
477 	LIST_INSERT_HEAD(&zombproc, p, p_list);
478 
479 	/*
480 	 * Mark the process as dead.  We must do this before we signal
481 	 * the parent.
482 	 */
483 	p->p_stat = SDEAD;
484 
485 	/* Put in front of parent's sibling list for parent to collect it */
486 	q = p->p_pptr;
487 	q->p_nstopchild++;
488 	if (LIST_FIRST(&q->p_children) != p) {
489 		/* Put child where it can be found quickly */
490 		LIST_REMOVE(p, p_sibling);
491 		LIST_INSERT_HEAD(&q->p_children, p, p_sibling);
492 	}
493 
494 	/*
495 	 * Notify parent that we're gone.  If parent has the P_NOCLDWAIT
496 	 * flag set, notify init instead (and hope it will handle
497 	 * this situation).
498 	 */
499 	mutex_enter(q->p_lock);
500 	if (q->p_flag & (PK_NOCLDWAIT|PK_CLDSIGIGN)) {
501 		proc_reparent(p, initproc);
502 		wakeinit = 1;
503 
504 		/*
505 		 * If this was the last child of our parent, notify
506 		 * parent, so in case he was wait(2)ing, he will
507 		 * continue.
508 		 */
509 		if (LIST_FIRST(&q->p_children) == NULL)
510 			cv_broadcast(&q->p_waitcv);
511 	}
512 	mutex_exit(q->p_lock);
513 
514 	/* Reload parent pointer, since p may have been reparented above */
515 	q = p->p_pptr;
516 
517 	if ((p->p_slflag & PSL_FSTRACE) == 0 && p->p_exitsig != 0) {
518 		exit_psignal(p, q, &ksi);
519 		kpsignal(q, &ksi, NULL);
520 	}
521 
522 	/* Calculate the final rusage info.  */
523 	calcru(p, &p->p_stats->p_ru.ru_utime, &p->p_stats->p_ru.ru_stime,
524 	    NULL, NULL);
525 
526 	if (wakeinit)
527 		cv_broadcast(&initproc->p_waitcv);
528 
529 	callout_destroy(&l->l_timeout_ch);
530 
531 	/*
532 	 * Remaining lwp resources will be freed in lwp_exit2() once we've
533 	 * switch to idle context; at that point, we will be marked as a
534 	 * full blown zombie.
535 	 */
536 	mutex_enter(p->p_lock);
537 	lwp_drainrefs(l);
538 	lwp_lock(l);
539 	l->l_prflag &= ~LPR_DETACHED;
540 	l->l_stat = LSZOMB;
541 	lwp_unlock(l);
542 	KASSERT(curlwp == l);
543 	KASSERT(p->p_nrlwps == 1);
544 	KASSERT(p->p_nlwps == 1);
545 	p->p_stat = SZOMB;
546 	p->p_nrlwps--;
547 	p->p_nzlwps++;
548 	p->p_ndlwps = 0;
549 	mutex_exit(p->p_lock);
550 
551 	/*
552 	 * Signal the parent to collect us, and drop the proclist lock.
553 	 * Drop debugger/procfs lock; no new references can be gained.
554 	 */
555 	cv_broadcast(&p->p_pptr->p_waitcv);
556 	mutex_exit(proc_lock);
557 	rw_exit(&p->p_reflock);
558 
559 	/* Verify that we hold no locks other than the kernel lock. */
560 	LOCKDEBUG_BARRIER(&kernel_lock, 0);
561 
562 	/*
563 	 * NOTE: WE ARE NO LONGER ALLOWED TO SLEEP!
564 	 */
565 
566 	/*
567 	 * Give machine-dependent code a chance to free any MD LWP
568 	 * resources.  This must be done before uvm_lwp_exit(), in
569 	 * case these resources are in the PCB.
570 	 */
571 #ifndef __NO_CPU_LWP_FREE
572 	cpu_lwp_free(l, 1);
573 #endif
574 	pmap_deactivate(l);
575 
576 	/* This process no longer needs to hold the kernel lock. */
577 #ifdef notyet
578 	/* XXXSMP hold in lwp_userret() */
579 	KERNEL_UNLOCK_LAST(l);
580 #else
581 	KERNEL_UNLOCK_ALL(l, NULL);
582 #endif
583 
584 	lwp_exit_switchaway(l);
585 }
586 
587 void
588 exit_lwps(struct lwp *l)
589 {
590 	struct proc *p;
591 	struct lwp *l2;
592 	int error;
593 	lwpid_t waited;
594 	int nlocks;
595 
596 	KERNEL_UNLOCK_ALL(l, &nlocks);
597 
598 	p = l->l_proc;
599 	KASSERT(mutex_owned(p->p_lock));
600 
601  retry:
602 	/*
603 	 * Interrupt LWPs in interruptable sleep, unsuspend suspended
604 	 * LWPs and then wait for everyone else to finish.
605 	 */
606 	LIST_FOREACH(l2, &p->p_lwps, l_sibling) {
607 		if (l2 == l)
608 			continue;
609 		lwp_lock(l2);
610 		l2->l_flag |= LW_WEXIT;
611 		if ((l2->l_stat == LSSLEEP && (l2->l_flag & LW_SINTR)) ||
612 		    l2->l_stat == LSSUSPENDED || l2->l_stat == LSSTOP) {
613 		    	/* setrunnable() will release the lock. */
614 			setrunnable(l2);
615 			DPRINTF(("exit_lwps: Made %d.%d runnable\n",
616 			    p->p_pid, l2->l_lid));
617 			continue;
618 		}
619 		lwp_unlock(l2);
620 	}
621 	while (p->p_nlwps > 1) {
622 		DPRINTF(("exit_lwps: waiting for %d LWPs (%d zombies)\n",
623 		    p->p_nlwps, p->p_nzlwps));
624 		error = lwp_wait1(l, 0, &waited, LWPWAIT_EXITCONTROL);
625 		if (p->p_nlwps == 1)
626 			break;
627 		if (error == EDEADLK) {
628 			/*
629 			 * LWPs can get suspended/slept behind us.
630 			 * (eg. sa_setwoken)
631 			 * kick them again and retry.
632 			 */
633 			goto retry;
634 		}
635 		if (error)
636 			panic("exit_lwps: lwp_wait1 failed with error %d",
637 			    error);
638 		DPRINTF(("exit_lwps: Got LWP %d from lwp_wait1()\n", waited));
639 	}
640 
641 	KERNEL_LOCK(nlocks, l);
642 	KASSERT(p->p_nlwps == 1);
643 }
644 
645 int
646 do_sys_wait(struct lwp *l, int *pid, int *status, int options,
647     struct rusage *ru, int *was_zombie)
648 {
649 	struct proc	*child;
650 	int		error;
651 
652 	mutex_enter(proc_lock);
653 	error = find_stopped_child(l->l_proc, *pid, options, &child, status);
654 
655 	if (child == NULL) {
656 		mutex_exit(proc_lock);
657 		*pid = 0;
658 		return error;
659 	}
660 
661 	*pid = child->p_pid;
662 
663 	if (child->p_stat == SZOMB) {
664 		/* proc_free() will release the proc_lock. */
665 		*was_zombie = 1;
666 		if (options & WNOWAIT)
667 			mutex_exit(proc_lock);
668 		else {
669 			proc_free(child, ru);
670 		}
671 	} else {
672 		/* Child state must have been SSTOP. */
673 		*was_zombie = 0;
674 		mutex_exit(proc_lock);
675 		*status = W_STOPCODE(*status);
676 	}
677 
678 	return 0;
679 }
680 
681 int
682 sys_wait4(struct lwp *l, const struct sys_wait4_args *uap, register_t *retval)
683 {
684 	/* {
685 		syscallarg(int)			pid;
686 		syscallarg(int *)		status;
687 		syscallarg(int)			options;
688 		syscallarg(struct rusage *)	rusage;
689 	} */
690 	int		status, error;
691 	int		was_zombie;
692 	struct rusage	ru;
693 	int pid = SCARG(uap, pid);
694 
695 	error = do_sys_wait(l, &pid, &status, SCARG(uap, options),
696 	    SCARG(uap, rusage) != NULL ? &ru : NULL, &was_zombie);
697 
698 	retval[0] = pid;
699 	if (pid == 0)
700 		return error;
701 
702 	if (SCARG(uap, rusage))
703 		error = copyout(&ru, SCARG(uap, rusage), sizeof(ru));
704 
705 	if (error == 0 && SCARG(uap, status))
706 		error = copyout(&status, SCARG(uap, status), sizeof(status));
707 
708 	return error;
709 }
710 
711 /*
712  * Scan list of child processes for a child process that has stopped or
713  * exited.  Used by sys_wait4 and 'compat' equivalents.
714  *
715  * Must be called with the proc_lock held, and may release while waiting.
716  */
717 static int
718 find_stopped_child(struct proc *parent, pid_t pid, int options,
719 		   struct proc **child_p, int *status_p)
720 {
721 	struct proc *child, *dead;
722 	int error;
723 
724 	KASSERT(mutex_owned(proc_lock));
725 
726 	if (options & ~(WUNTRACED|WNOHANG|WALTSIG|WALLSIG)
727 	    && !(options & WOPTSCHECKED)) {
728 		*child_p = NULL;
729 		return EINVAL;
730 	}
731 
732 	if (pid == 0 && !(options & WOPTSCHECKED))
733 		pid = -parent->p_pgid;
734 
735 	for (;;) {
736 		error = ECHILD;
737 		dead = NULL;
738 
739 		LIST_FOREACH(child, &parent->p_children, p_sibling) {
740 			if (pid >= 0) {
741 				if (child->p_pid != pid) {
742 					child = p_find(pid, PFIND_ZOMBIE |
743 					    PFIND_LOCKED);
744 					if (child == NULL ||
745 					    child->p_pptr != parent) {
746 						child = NULL;
747 						break;
748 					}
749 				}
750 			} else if (pid != WAIT_ANY && child->p_pgid != -pid) {
751 				/* Child not in correct pgrp */
752 				continue;
753 			}
754 
755 			/*
756 			 * Wait for processes with p_exitsig != SIGCHLD
757 			 * processes only if WALTSIG is set; wait for
758 			 * processes with p_exitsig == SIGCHLD only
759 			 * if WALTSIG is clear.
760 			 */
761 			if (((options & WALLSIG) == 0) &&
762 			    (options & WALTSIG ? child->p_exitsig == SIGCHLD
763 						: P_EXITSIG(child) != SIGCHLD)){
764 				if (child->p_pid == pid) {
765 					child = NULL;
766 					break;
767 				}
768 				continue;
769 			}
770 
771 			error = 0;
772 			if ((options & WNOZOMBIE) == 0) {
773 				if (child->p_stat == SZOMB)
774 					break;
775 				if (child->p_stat == SDEAD) {
776 					/*
777 					 * We may occasionally arrive here
778 					 * after receiving a signal, but
779 					 * immediatley before the child
780 					 * process is zombified.  The wait
781 					 * will be short, so avoid returning
782 					 * to userspace.
783 					 */
784 					dead = child;
785 				}
786 			}
787 
788 			if (child->p_stat == SSTOP &&
789 			    child->p_waited == 0 &&
790 			    (child->p_slflag & PSL_TRACED ||
791 			    options & WUNTRACED)) {
792 				if ((options & WNOWAIT) == 0) {
793 					child->p_waited = 1;
794 					parent->p_nstopchild--;
795 				}
796 				break;
797 			}
798 			if (parent->p_nstopchild == 0 || child->p_pid == pid) {
799 				child = NULL;
800 				break;
801 			}
802 		}
803 
804 		if (child != NULL || error != 0 ||
805 		    ((options & WNOHANG) != 0 && dead == NULL)) {
806 		    	if (child != NULL) {
807 			    	*status_p = child->p_xstat;
808 			}
809 			*child_p = child;
810 			return error;
811 		}
812 
813 		/*
814 		 * Wait for another child process to stop.
815 		 */
816 		error = cv_wait_sig(&parent->p_waitcv, proc_lock);
817 
818 		if (error != 0) {
819 			*child_p = NULL;
820 			return error;
821 		}
822 	}
823 }
824 
825 /*
826  * Free a process after parent has taken all the state info.  Must be called
827  * with the proclist lock held, and will release before returning.
828  *
829  * *ru is returned to the caller, and must be freed by the caller.
830  */
831 static void
832 proc_free(struct proc *p, struct rusage *ru)
833 {
834 	struct proc *parent;
835 	struct lwp *l;
836 	ksiginfo_t ksi;
837 	kauth_cred_t cred1, cred2;
838 	uid_t uid;
839 
840 	KASSERT(mutex_owned(proc_lock));
841 	KASSERT(p->p_nlwps == 1);
842 	KASSERT(p->p_nzlwps == 1);
843 	KASSERT(p->p_nrlwps == 0);
844 	KASSERT(p->p_stat == SZOMB);
845 
846 	/*
847 	 * If we got the child via ptrace(2) or procfs, and
848 	 * the parent is different (meaning the process was
849 	 * attached, rather than run as a child), then we need
850 	 * to give it back to the old parent, and send the
851 	 * parent the exit signal.  The rest of the cleanup
852 	 * will be done when the old parent waits on the child.
853 	 */
854 	if ((p->p_slflag & PSL_TRACED) != 0) {
855 		parent = p->p_pptr;
856 		if (p->p_opptr != parent){
857 			mutex_enter(p->p_lock);
858 			p->p_slflag &= ~(PSL_TRACED|PSL_FSTRACE|PSL_SYSCALL);
859 			mutex_exit(p->p_lock);
860 			parent = p->p_opptr;
861 			if (parent == NULL)
862 				parent = initproc;
863 			proc_reparent(p, parent);
864 			p->p_opptr = NULL;
865 			if (p->p_exitsig != 0) {
866 				exit_psignal(p, parent, &ksi);
867 				kpsignal(parent, &ksi, NULL);
868 			}
869 			cv_broadcast(&parent->p_waitcv);
870 			mutex_exit(proc_lock);
871 			return;
872 		}
873 	}
874 
875 	/*
876 	 * Finally finished with old proc entry.  Unlink it from its process
877 	 * group.
878 	 */
879 	leavepgrp(p);
880 
881 	parent = p->p_pptr;
882 	sched_proc_exit(parent, p);
883 
884 	/*
885 	 * Add child times of exiting process onto its own times.
886 	 * This cannot be done any earlier else it might get done twice.
887 	 */
888 	l = LIST_FIRST(&p->p_lwps);
889 	p->p_stats->p_ru.ru_nvcsw += (l->l_ncsw - l->l_nivcsw);
890 	p->p_stats->p_ru.ru_nivcsw += l->l_nivcsw;
891 	ruadd(&p->p_stats->p_ru, &l->l_ru);
892 	ruadd(&p->p_stats->p_ru, &p->p_stats->p_cru);
893 	ruadd(&parent->p_stats->p_cru, &p->p_stats->p_ru);
894 	if (ru != NULL)
895 		*ru = p->p_stats->p_ru;
896 	p->p_xstat = 0;
897 
898 	/*
899 	 * At this point we are going to start freeing the final resources.
900 	 * If anyone tries to access the proc structure after here they will
901 	 * get a shock - bits are missing.  Attempt to make it hard!  We
902 	 * don't bother with any further locking past this point.
903 	 */
904 	p->p_stat = SIDL;		/* not even a zombie any more */
905 	LIST_REMOVE(p, p_list);	/* off zombproc */
906 	parent = p->p_pptr;
907 	p->p_pptr->p_nstopchild--;
908 	LIST_REMOVE(p, p_sibling);
909 
910 	/*
911 	 * Let pid be reallocated.
912 	 */
913 	proc_free_pid(p);
914 	mutex_exit(proc_lock);
915 
916 	/*
917 	 * Delay release until after lwp_free.
918 	 */
919 	cred2 = l->l_cred;
920 
921 	/*
922 	 * Free the last LWP's resources.
923 	 *
924 	 * lwp_free ensures the LWP is no longer running on another CPU.
925 	 */
926 	lwp_free(l, false, true);
927 
928 	/*
929 	 * Now no one except us can reach the process p.
930 	 */
931 
932 	/*
933 	 * Decrement the count of procs running with this uid.
934 	 */
935 	cred1 = p->p_cred;
936 	uid = kauth_cred_getuid(cred1);
937 	(void)chgproccnt(uid, -1);
938 
939 	/*
940 	 * Release substructures.
941 	 */
942 
943 	limfree(p->p_limit);
944 	pstatsfree(p->p_stats);
945 	kauth_cred_free(cred1);
946 	kauth_cred_free(cred2);
947 
948 	/*
949 	 * Release reference to text vnode
950 	 */
951 	if (p->p_textvp)
952 		vrele(p->p_textvp);
953 
954 	mutex_destroy(&p->p_auxlock);
955 	mutex_obj_free(p->p_lock);
956 	mutex_destroy(&p->p_stmutex);
957 	cv_destroy(&p->p_waitcv);
958 	cv_destroy(&p->p_lwpcv);
959 	rw_destroy(&p->p_reflock);
960 
961 	proc_free_mem(p);
962 }
963 
964 /*
965  * make process 'parent' the new parent of process 'child'.
966  *
967  * Must be called with proc_lock held.
968  */
969 void
970 proc_reparent(struct proc *child, struct proc *parent)
971 {
972 
973 	KASSERT(mutex_owned(proc_lock));
974 
975 	if (child->p_pptr == parent)
976 		return;
977 
978 	if (child->p_stat == SZOMB ||
979 	    (child->p_stat == SSTOP && !child->p_waited)) {
980 		child->p_pptr->p_nstopchild--;
981 		parent->p_nstopchild++;
982 	}
983 	if (parent == initproc)
984 		child->p_exitsig = SIGCHLD;
985 
986 	LIST_REMOVE(child, p_sibling);
987 	LIST_INSERT_HEAD(&parent->p_children, child, p_sibling);
988 	child->p_pptr = parent;
989 }
990