xref: /netbsd-src/sys/kern/kern_exit.c (revision b7ae68fde0d8ef1c03714e8bbb1ee7c6118ea93b)
1 /*	$NetBSD: kern_exit.c,v 1.159 2006/09/23 15:36:12 xtraeme Exp $	*/
2 
3 /*-
4  * Copyright (c) 1998, 1999 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.
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 1.159 2006/09/23 15:36:12 xtraeme 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/systm.h>
86 #include <sys/ioctl.h>
87 #include <sys/tty.h>
88 #include <sys/time.h>
89 #include <sys/resource.h>
90 #include <sys/kernel.h>
91 #include <sys/ktrace.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/sa.h>
111 #include <sys/savar.h>
112 #include <sys/mount.h>
113 #include <sys/syscallargs.h>
114 #include <sys/systrace.h>
115 #include <sys/kauth.h>
116 
117 #include <machine/cpu.h>
118 
119 #include <uvm/uvm_extern.h>
120 
121 #define DEBUG_EXIT
122 
123 #ifdef DEBUG_EXIT
124 int debug_exit = 0;
125 #define DPRINTF(x) if (debug_exit) printf x
126 #else
127 #define DPRINTF(x)
128 #endif
129 
130 static void lwp_exit_hook(struct lwp *, void *);
131 
132 /*
133  * Fill in the appropriate signal information, and signal the parent.
134  */
135 static void
136 exit_psignal(struct proc *p, struct proc *pp, ksiginfo_t *ksi)
137 {
138 
139 	(void)memset(ksi, 0, sizeof(ksiginfo_t));
140 	if ((ksi->ksi_signo = P_EXITSIG(p)) == SIGCHLD) {
141 		if (WIFSIGNALED(p->p_xstat)) {
142 			if (WCOREDUMP(p->p_xstat))
143 				ksi->ksi_code = CLD_DUMPED;
144 			else
145 				ksi->ksi_code = CLD_KILLED;
146 		} else {
147 			ksi->ksi_code = CLD_EXITED;
148 		}
149 	}
150 	/*
151 	 * we fill those in, even for non-SIGCHLD.
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_ru->ru_utime.tv_sec;
158 	ksi->ksi_stime = p->p_ru->ru_stime.tv_sec;
159 }
160 
161 /*
162  * exit --
163  *	Death of process.
164  */
165 int
166 sys_exit(struct lwp *l, void *v, register_t *retval)
167 {
168 	struct sys_exit_args /* {
169 		syscallarg(int)	rval;
170 	} */ *uap = v;
171 
172 	/* Don't call exit1() multiple times in the same process.*/
173 	if (l->l_proc->p_flag & P_WEXIT)
174 		lwp_exit(l);
175 
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 void
187 exit1(struct lwp *l, int rv)
188 {
189 	struct proc	*p, *q, *nq;
190 	int		s, sa;
191 	struct plimit	*plim;
192 	struct pstats	*pstats;
193 	struct sigacts	*ps;
194 	ksiginfo_t	ksi;
195 	int		do_psignal = 0;
196 
197 	p = l->l_proc;
198 
199 	if (__predict_false(p == initproc))
200 		panic("init died (signal %d, exit %d)",
201 		    WTERMSIG(rv), WEXITSTATUS(rv));
202 
203 	p->p_flag |= P_WEXIT;
204 	if (p->p_flag & P_STOPEXIT) {
205 		sigminusset(&contsigmask, &p->p_sigctx.ps_siglist);
206 		SCHED_LOCK(s);
207 		p->p_stat = SSTOP;
208 		l->l_stat = LSSTOP;
209 		p->p_nrlwps--;
210 		mi_switch(l, NULL);
211 		SCHED_ASSERT_UNLOCKED();
212 		splx(s);
213 	}
214 
215 	DPRINTF(("exit1: %d.%d exiting.\n", p->p_pid, l->l_lid));
216 	/*
217 	 * Disable scheduler activation upcalls.
218 	 * We're trying to get out of here.
219 	 */
220 	sa = 0;
221 	if (p->p_sa != NULL) {
222 		l->l_flag &= ~L_SA;
223 #if 0
224 		p->p_flag &= ~P_SA;
225 #endif
226 		sa = 1;
227 	}
228 
229 #ifdef PGINPROF
230 	vmsizmon();
231 #endif
232 	if (p->p_flag & P_PROFIL)
233 		stopprofclock(p);
234 	p->p_ru = pool_get(&rusage_pool, PR_WAITOK);
235 	/*
236 	 * If parent is waiting for us to exit or exec, P_PPWAIT is set; we
237 	 * wake up the parent early to avoid deadlock.
238 	 */
239 	if (p->p_flag & P_PPWAIT) {
240 		p->p_flag &= ~P_PPWAIT;
241 		wakeup(p->p_pptr);
242 	}
243 	sigfillset(&p->p_sigctx.ps_sigignore);
244 	sigemptyset(&p->p_sigctx.ps_siglist);
245 	p->p_sigctx.ps_sigcheck = 0;
246 	timers_free(p, TIMERS_ALL);
247 
248 	if (sa || (p->p_nlwps > 1)) {
249 		exit_lwps(l);
250 
251 		/*
252 		 * Collect thread u-areas.
253 		 */
254 		uvm_uarea_drain(FALSE);
255 	}
256 
257 #if defined(__HAVE_RAS)
258 	ras_purgeall(p);
259 #endif
260 
261 	/*
262 	 * Close open files and release open-file table.
263 	 * This may block!
264 	 */
265 	fdfree(l);
266 	cwdfree(p->p_cwdi);
267 	p->p_cwdi = 0;
268 
269 	doexithooks(p);
270 
271 	if (SESS_LEADER(p)) {
272 		struct session *sp = p->p_session;
273 		struct tty *tp;
274 
275 		if (sp->s_ttyvp) {
276 			/*
277 			 * Controlling process.
278 			 * Signal foreground pgrp,
279 			 * drain controlling terminal
280 			 * and revoke access to controlling terminal.
281 			 */
282 			tp = sp->s_ttyp;
283 			s = spltty();
284 			TTY_LOCK(tp);
285 			if (tp->t_session == sp) {
286 				if (tp->t_pgrp)
287 					pgsignal(tp->t_pgrp, SIGHUP, 1);
288 				/* we can't guarantee the revoke will do this */
289 				tp->t_pgrp = NULL;
290 				tp->t_session = NULL;
291 				TTY_UNLOCK(tp);
292 				splx(s);
293 				SESSRELE(sp);
294 				(void) ttywait(tp);
295 				/*
296 				 * The tty could have been revoked
297 				 * if we blocked.
298 				 */
299 				if (sp->s_ttyvp)
300 					VOP_REVOKE(sp->s_ttyvp, REVOKEALL);
301 			} else {
302 				TTY_UNLOCK(tp);
303 				splx(s);
304 			}
305 			if (sp->s_ttyvp)
306 				vrele(sp->s_ttyvp);
307 			sp->s_ttyvp = NULL;
308 			/*
309 			 * s_ttyp is not zero'd; we use this to indicate
310 			 * that the session once had a controlling terminal.
311 			 * (for logging and informational purposes)
312 			 */
313 		}
314 		sp->s_leader = NULL;
315 	}
316 	fixjobc(p, p->p_pgrp, 0);
317 
318 	/*
319 	 * Collect accounting flags from the last remaining LWP (this one),
320 	 * and write out accounting data.
321 	 */
322 	p->p_acflag |= l->l_acflag;
323 	(void)acct_process(l);
324 
325 #ifdef KTRACE
326 	/*
327 	 * Release trace file.
328 	 */
329 	ktrderef(p);
330 #endif
331 #ifdef SYSTRACE
332 	systrace_sys_exit(p);
333 #endif
334 
335 	/*
336 	 * If emulation has process exit hook, call it now.
337 	 * Set the exit status now so that the exit hook has
338 	 * an opportunity to tweak it (COMPAT_LINUX requires
339 	 * this for thread group emulation)
340 	 */
341 	p->p_xstat = rv;
342 	if (p->p_emul->e_proc_exit)
343 		(*p->p_emul->e_proc_exit)(p);
344 
345 	/*
346 	 * Free the VM resources we're still holding on to.
347 	 * We must do this from a valid thread because doing
348 	 * so may block. This frees vmspace, which we don't
349 	 * need anymore. The only remaining lwp is the one
350 	 * we run at this moment, nothing runs in userland
351 	 * anymore.
352 	 */
353 	uvm_proc_exit(p);
354 
355 	/*
356 	 * Give machine-dependent code a chance to free any
357 	 * MD LWP resources while we can still block. This must be done
358 	 * before uvm_lwp_exit(), in case these resources are in the
359 	 * PCB.
360 	 * THIS IS LAST BLOCKING OPERATION.
361 	 */
362 #ifndef __NO_CPU_LWP_FREE
363 	cpu_lwp_free(l, 1);
364 #endif
365 
366 	pmap_deactivate(l);
367 
368 	/*
369 	 * NOTE: WE ARE NO LONGER ALLOWED TO SLEEP!
370 	 */
371 
372 	/*
373 	 * Save final rusage info, adding in child rusage info and self times.
374 	 * In order to pick up the time for the current execution, we must
375 	 * do this before unlinking the lwp from l_list.
376 	 */
377 	*p->p_ru = p->p_stats->p_ru;
378 	calcru(p, &p->p_ru->ru_utime, &p->p_ru->ru_stime, NULL);
379 	ruadd(p->p_ru, &p->p_stats->p_cru);
380 
381 	/*
382 	 * Notify interested parties of our demise.
383 	 */
384 	KNOTE(&p->p_klist, NOTE_EXIT);
385 
386 #if PERFCTRS
387 	/*
388 	 * Save final PMC information in parent process & clean up.
389 	 */
390 	if (PMC_ENABLED(p)) {
391 		pmc_save_context(p);
392 		pmc_accumulate(p->p_pptr, p);
393 		pmc_process_exit(p);
394 	}
395 #endif
396 
397 	s = proclist_lock_write();
398 	/*
399 	 * Reset p_opptr pointer of all former children which got
400 	 * traced by another process and were reparented. We reset
401 	 * it to NULL here; the trace detach code then reparents
402 	 * the child to initproc. We only check allproc list, since
403 	 * eventual former children on zombproc list won't reference
404 	 * p_opptr anymore.
405 	 */
406 	if (p->p_flag & P_CHTRACED) {
407 		PROCLIST_FOREACH(q, &allproc) {
408 			if (q->p_opptr == p)
409 				q->p_opptr = NULL;
410 		}
411 	}
412 
413 	/*
414 	 * Give orphaned children to init(8).
415 	 */
416 	q = LIST_FIRST(&p->p_children);
417 	if (q)		/* only need this if any child is SZOMB */
418 		wakeup(initproc);
419 	for (; q != NULL; q = nq) {
420 		nq = LIST_NEXT(q, p_sibling);
421 
422 		/*
423 		 * Traced processes are killed since their existence
424 		 * means someone is screwing up. Since we reset the
425 		 * trace flags, the logic in sys_wait4() would not be
426 		 * triggered to reparent the process to its
427 		 * original parent, so we must do this here.
428 		 */
429 		if (q->p_flag & P_TRACED) {
430 			if (q->p_opptr != q->p_pptr) {
431 				struct proc *t = q->p_opptr;
432 				proc_reparent(q, t ? t : initproc);
433 				q->p_opptr = NULL;
434 			} else
435 				proc_reparent(q, initproc);
436 			q->p_flag &= ~(P_TRACED|P_WAITED|P_FSTRACE|P_SYSCALL);
437 			killproc(q, "orphaned traced process");
438 		} else {
439 			proc_reparent(q, initproc);
440 		}
441 	}
442 
443 	/*
444 	 * Move proc from allproc to zombproc, it's now ready
445 	 * to be collected by parent. Remaining lwp resources
446 	 * will be freed in lwp_exit2() once we've switch to idle
447 	 * context.
448 	 * Changing the state to SZOMB stops it being found by pfind().
449 	 */
450 	LIST_REMOVE(p, p_list);
451 	LIST_INSERT_HEAD(&zombproc, p, p_list);
452 	p->p_stat = SZOMB;
453 
454 	LIST_REMOVE(l, l_list);
455 	LIST_REMOVE(l, l_sibling);
456 	l->l_flag |= L_DETACHED;	/* detached from proc too */
457 	l->l_stat = LSDEAD;
458 
459 	KASSERT(p->p_nrlwps == 1);
460 	KASSERT(p->p_nlwps == 1);
461 	p->p_nrlwps--;
462 	p->p_nlwps--;
463 
464 	/* Put in front of parent's sibling list for parent to collect it */
465 	q = p->p_pptr;
466 	q->p_nstopchild++;
467 	if (LIST_FIRST(&q->p_children) != p) {
468 		/* Put child where it can be found quickly */
469 		LIST_REMOVE(p, p_sibling);
470 		LIST_INSERT_HEAD(&q->p_children, p, p_sibling);
471 	}
472 
473 	/*
474 	 * Notify parent that we're gone.  If parent has the P_NOCLDWAIT
475 	 * flag set, notify init instead (and hope it will handle
476 	 * this situation).
477 	 */
478 	if (q->p_flag & (P_NOCLDWAIT|P_CLDSIGIGN)) {
479 		proc_reparent(p, initproc);
480 
481 		/*
482 		 * If this was the last child of our parent, notify
483 		 * parent, so in case he was wait(2)ing, he will
484 		 * continue.
485 		 */
486 		if (LIST_FIRST(&q->p_children) == NULL)
487 			wakeup(q);
488 	}
489 
490 	/*
491 	 * Clear curlwp after we've done all operations
492 	 * that could block, and before tearing down the rest
493 	 * of the process state that might be used from clock, etc.
494 	 * Also, can't clear curlwp while we're still runnable,
495 	 * as we're not on a run queue (we are current, just not
496 	 * a proper proc any longer!).
497 	 *
498 	 * Other substructures are freed from wait().
499 	 */
500 	curlwp = NULL;
501 
502 	/* Delay release until after dropping the proclist lock */
503 	plim = p->p_limit;
504 	pstats = p->p_stats;
505 	ps = p->p_sigacts;
506 
507 	p->p_limit = NULL;
508 	p->p_stats = NULL;
509 	p->p_sigacts = NULL;
510 
511 	/* Reload parent pointer, since p may have been reparented above */
512 	q = p->p_pptr;
513 
514 	if ((p->p_flag & P_FSTRACE) == 0 && p->p_exitsig != 0) {
515 		exit_psignal(p, q, &ksi);
516 		do_psignal = 1;
517 	}
518 
519 	/*
520 	 * Once we release the proclist lock, we shouldn't touch the
521 	 * process structure anymore, since it's now on the zombie
522 	 * list and available for collection by the parent.
523 	 */
524 	proclist_unlock_write(s);
525 
526 	if (do_psignal)
527 		kpsignal(q, &ksi, NULL);
528 
529 	/* Wake up the parent so it can get exit status. */
530 	wakeup(q);
531 
532 	/* Release substructures */
533 	sigactsfree(ps);
534 	limfree(plim);
535 	pstatsfree(pstats);
536 
537 	/* Release cached credentials. */
538 	kauth_cred_free(l->l_cred);
539 
540 #ifdef DEBUG
541 	/* Nothing should use the process link anymore */
542 	l->l_proc = NULL;
543 #endif
544 
545 	/* This process no longer needs to hold the kernel lock. */
546 	KERNEL_PROC_UNLOCK(l);
547 
548 	/*
549 	 * Finally, call machine-dependent code to switch to a new
550 	 * context (possibly the idle context).  Once we are no longer
551 	 * using the dead lwp's stack, lwp_exit2() will be called
552 	 * to arrange for the resources to be released.
553 	 *
554 	 * Note that cpu_exit() will end with a call equivalent to
555 	 * cpu_switch(), finishing our execution (pun intended).
556 	 */
557 
558 	uvmexp.swtch++;
559 	cpu_exit(l);
560 }
561 
562 void
563 exit_lwps(struct lwp *l)
564 {
565 	struct proc *p;
566 	struct lwp *l2;
567 	struct sadata_vp *vp;
568 	int s, error;
569 	lwpid_t		waited;
570 
571 	p = l->l_proc;
572 
573 	/* XXX SMP
574 	 * This would be the right place to IPI any LWPs running on
575 	 * other processors so that they can notice the userret exit hook.
576 	 */
577 	p->p_userret = lwp_exit_hook;
578 	p->p_userret_arg = NULL;
579 
580 	if (p->p_sa) {
581 		SLIST_FOREACH(vp, &p->p_sa->sa_vps, savp_next) {
582 			/*
583 			 * Make SA-cached LWPs normal process runnable
584 			 * LWPs so that they'll also self-destruct.
585 			 */
586 			DPRINTF(("exit_lwps: Making cached LWPs of %d on VP %d runnable: ",
587 				    p->p_pid, vp->savp_id));
588 			SCHED_LOCK(s);
589 			while ((l2 = sa_getcachelwp(vp)) != 0) {
590 				l2->l_priority = l2->l_usrpri;
591 				setrunnable(l2);
592 				DPRINTF(("%d ", l2->l_lid));
593 			}
594 			SCHED_UNLOCK(s);
595 			DPRINTF(("\n"));
596 
597 			/*
598 			 * Clear wokenq, the LWPs on the queue will
599 			 * run below.
600 			 */
601 			vp->savp_wokenq_head = NULL;
602 		}
603 	}
604 
605 retry:
606 	/*
607 	 * Interrupt LWPs in interruptable sleep, unsuspend suspended
608 	 * LWPs, make detached LWPs undetached (so we can wait for
609 	 * them) and then wait for everyone else to finish.
610 	 */
611 	LIST_FOREACH(l2, &p->p_lwps, l_sibling) {
612 		l2->l_flag &= ~(L_DETACHED|L_SA);
613 
614 		SCHED_LOCK(s);
615 		if ((l2->l_stat == LSSLEEP && (l2->l_flag & L_SINTR)) ||
616 		    l2->l_stat == LSSUSPENDED || l2->l_stat == LSSTOP) {
617 			setrunnable(l2);
618 			DPRINTF(("exit_lwps: Made %d.%d runnable\n",
619 			    p->p_pid, l2->l_lid));
620 		}
621 		SCHED_UNLOCK(s);
622 	}
623 
624 
625 	while (p->p_nlwps > 1) {
626 		DPRINTF(("exit_lwps: waiting for %d LWPs (%d runnable, %d zombies)\n",
627 		    p->p_nlwps, p->p_nrlwps, p->p_nzlwps));
628 		error = lwp_wait1(l, 0, &waited, LWPWAIT_EXITCONTROL);
629 		if (error == EDEADLK) {
630 			/*
631 			 * LWPs can get suspended/slept behind us.
632 			 * (eg. sa_setwoken)
633 			 * kick them again and retry.
634 			 */
635 			goto retry;
636 		}
637 		if (error)
638 			panic("exit_lwps: lwp_wait1 failed with error %d",
639 			    error);
640 		DPRINTF(("exit_lwps: Got LWP %d from lwp_wait1()\n", waited));
641 	}
642 
643 	p->p_userret = NULL;
644 }
645 
646 /* Wrapper function for use in p_userret */
647 static void
648 lwp_exit_hook(struct lwp *l, void *arg)
649 {
650 	KERNEL_PROC_LOCK(l);
651 	lwp_exit(l);
652 }
653 
654 int
655 sys_wait4(struct lwp *l, void *v, register_t *retval)
656 {
657 	struct sys_wait4_args /* {
658 		syscallarg(int)			pid;
659 		syscallarg(int *)		status;
660 		syscallarg(int)			options;
661 		syscallarg(struct rusage *)	rusage;
662 	} */ *uap = v;
663 	struct proc	*child, *parent;
664 	int		status, error;
665 
666 	parent = l->l_proc;
667 
668 	if (SCARG(uap, pid) == 0)
669 		SCARG(uap, pid) = -parent->p_pgid;
670 	if (SCARG(uap, options) & ~(WUNTRACED|WNOHANG|WALTSIG|WALLSIG))
671 		return (EINVAL);
672 
673 	error = find_stopped_child(parent, SCARG(uap,pid), SCARG(uap,options),
674 				&child);
675 	if (error != 0)
676 		return error;
677 	if (child == NULL) {
678 		*retval = 0;
679 		return 0;
680 	}
681 
682 	/*
683 	 * Collect child u-areas.
684 	 */
685 	uvm_uarea_drain(FALSE);
686 
687 	retval[0] = child->p_pid;
688 
689 	if (P_ZOMBIE(child)) {
690 		if (SCARG(uap, status)) {
691 			status = child->p_xstat;	/* convert to int */
692 			error = copyout(&status, SCARG(uap, status),
693 					sizeof(status));
694 			if (error)
695 				return (error);
696 		}
697 		if (SCARG(uap, rusage)) {
698 			error = copyout(child->p_ru, SCARG(uap, rusage),
699 				    sizeof(struct rusage));
700 			if (error)
701 				return (error);
702 		}
703 
704 		proc_free(child);
705 		return 0;
706 	}
707 
708 	/* child state must be SSTOP */
709 	if (SCARG(uap, status)) {
710 		status = W_STOPCODE(child->p_xstat);
711 		return copyout(&status, SCARG(uap, status), sizeof(status));
712 	}
713 	return 0;
714 }
715 
716 /*
717  * Scan list of child processes for a child process that has stopped or
718  * exited.  Used by sys_wait4 and 'compat' equivalents.
719  */
720 int
721 find_stopped_child(struct proc *parent, pid_t pid, int options,
722 	struct proc **child_p)
723 {
724 	struct proc *child;
725 	int error;
726 
727 	for (;;) {
728 		proclist_lock_read();
729 		error = ECHILD;
730 		LIST_FOREACH(child, &parent->p_children, p_sibling) {
731 			if (pid >= 0) {
732 				if (child->p_pid != pid) {
733 					child = p_find(pid, PFIND_ZOMBIE |
734 								PFIND_LOCKED);
735 					if (child == NULL
736 					    || child->p_pptr != parent) {
737 						child = NULL;
738 						break;
739 					}
740 				}
741 			} else
742 				if (pid != WAIT_ANY && child->p_pgid != -pid)
743 					/* child not in correct pgrp */
744 					continue;
745 			/*
746 			 * Wait for processes with p_exitsig != SIGCHLD
747 			 * processes only if WALTSIG is set; wait for
748 			 * processes with p_exitsig == SIGCHLD only
749 			 * if WALTSIG is clear.
750 			 */
751 			if (((options & WALLSIG) == 0) &&
752 			    (options & WALTSIG ? child->p_exitsig == SIGCHLD
753 						: P_EXITSIG(child) != SIGCHLD)){
754 				if (child->p_pid == pid) {
755 					child = NULL;
756 					break;
757 				}
758 				continue;
759 			}
760 
761 			error = 0;
762 			if (child->p_stat == SZOMB && !(options & WNOZOMBIE))
763 				break;
764 
765 			if (child->p_stat == SSTOP &&
766 			    (child->p_flag & P_WAITED) == 0 &&
767 			    (child->p_flag & P_TRACED || options & WUNTRACED)) {
768 				if ((options & WNOWAIT) == 0) {
769 					child->p_flag |= P_WAITED;
770 					parent->p_nstopchild--;
771 				}
772 				break;
773 			}
774 			if (parent->p_nstopchild == 0 || child->p_pid == pid) {
775 				child = NULL;
776 				break;
777 			}
778 		}
779 		proclist_unlock_read();
780 		if (child != NULL || error != 0 || options & WNOHANG) {
781 			*child_p = child;
782 			return error;
783 		}
784 		error = tsleep(parent, PWAIT | PCATCH, "wait", 0);
785 		if (error != 0)
786 			return error;
787 	}
788 }
789 
790 /*
791  * Free a process after parent has taken all the state info.
792  */
793 void
794 proc_free(struct proc *p)
795 {
796 	struct proc *parent = p->p_pptr;
797 	ksiginfo_t ksi;
798 	int s;
799 
800 	KASSERT(p->p_nlwps == 0);
801 	KASSERT(p->p_nzlwps == 0);
802 	KASSERT(p->p_nrlwps == 0);
803 	KASSERT(LIST_EMPTY(&p->p_lwps));
804 
805 	/*
806 	 * If we got the child via ptrace(2) or procfs, and
807 	 * the parent is different (meaning the process was
808 	 * attached, rather than run as a child), then we need
809 	 * to give it back to the old parent, and send the
810 	 * parent the exit signal.  The rest of the cleanup
811 	 * will be done when the old parent waits on the child.
812 	 */
813 	if ((p->p_flag & P_TRACED) && p->p_opptr != parent){
814 		parent = p->p_opptr;
815 		if (parent == NULL)
816 			parent = initproc;
817 		proc_reparent(p, parent);
818 		p->p_opptr = NULL;
819 		p->p_flag &= ~(P_TRACED|P_WAITED|P_FSTRACE|P_SYSCALL);
820 		if (p->p_exitsig != 0) {
821 			exit_psignal(p, parent, &ksi);
822 			kpsignal(parent, &ksi, NULL);
823 		}
824 		wakeup(parent);
825 		return;
826 	}
827 
828 	scheduler_wait_hook(parent, p);
829 	p->p_xstat = 0;
830 
831 	ruadd(&parent->p_stats->p_cru, p->p_ru);
832 
833 	/*
834 	 * At this point we are going to start freeing the final resources.
835 	 * If anyone tries to access the proc structure after here they
836 	 * will get a shock - bits are missing.
837 	 * Attempt to make it hard!
838 	 */
839 
840 	p->p_stat = SIDL;		/* not even a zombie any more */
841 
842 	pool_put(&rusage_pool, p->p_ru);
843 
844 	/*
845 	 * Finally finished with old proc entry.
846 	 * Unlink it from its process group and free it.
847 	 */
848 	leavepgrp(p);
849 
850 	s = proclist_lock_write();
851 	LIST_REMOVE(p, p_list);	/* off zombproc */
852 	p->p_pptr->p_nstopchild--;
853 	LIST_REMOVE(p, p_sibling);
854 	proclist_unlock_write(s);
855 
856 	/*
857 	 * Decrement the count of procs running with this uid.
858 	 */
859 	(void)chgproccnt(kauth_cred_getuid(p->p_cred), -1);
860 
861 	/*
862 	 * Free up credentials.
863 	 */
864 	kauth_cred_free(p->p_cred);
865 
866 	/*
867 	 * Release reference to text vnode
868 	 */
869 	if (p->p_textvp)
870 		vrele(p->p_textvp);
871 
872 	/* Release any SA state. */
873 	if (p->p_sa)
874 		sa_release(p);
875 
876 	/* Free proc structure and let pid be reallocated */
877 	proc_free_mem(p);
878 }
879 
880 /*
881  * make process 'parent' the new parent of process 'child'.
882  *
883  * Must be called with proclist_lock_write() held.
884  */
885 void
886 proc_reparent(struct proc *child, struct proc *parent)
887 {
888 
889 	if (child->p_pptr == parent)
890 		return;
891 
892 	if (child->p_stat == SZOMB
893 	    || (child->p_stat == SSTOP && !(child->p_flag & P_WAITED))) {
894 		child->p_pptr->p_nstopchild--;
895 		parent->p_nstopchild++;
896 	}
897 	if (parent == initproc)
898 		child->p_exitsig = SIGCHLD;
899 
900 	LIST_REMOVE(child, p_sibling);
901 	LIST_INSERT_HEAD(&parent->p_children, child, p_sibling);
902 	child->p_pptr = parent;
903 }
904