xref: /netbsd-src/sys/kern/kern_exit.c (revision 267197ec1eebfcb9810ea27a89625b6ddf68e3e7)
1 /*	$NetBSD: kern_exit.c,v 1.199 2008/01/28 12:22:46 yamt 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 1.199 2008/01/28 12:22:46 yamt Exp $");
78 
79 #include "opt_ktrace.h"
80 #include "opt_perfctrs.h"
81 #include "opt_sysv.h"
82 
83 #include <sys/param.h>
84 #include <sys/aio.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/proc.h>
92 #include <sys/buf.h>
93 #include <sys/wait.h>
94 #include <sys/file.h>
95 #include <sys/vnode.h>
96 #include <sys/syslog.h>
97 #include <sys/malloc.h>
98 #include <sys/pool.h>
99 #include <sys/resourcevar.h>
100 #if defined(PERFCTRS)
101 #include <sys/pmc.h>
102 #endif
103 #include <sys/ptrace.h>
104 #include <sys/acct.h>
105 #include <sys/filedesc.h>
106 #include <sys/ras.h>
107 #include <sys/signalvar.h>
108 #include <sys/sched.h>
109 #include <sys/mount.h>
110 #include <sys/syscallargs.h>
111 #include <sys/kauth.h>
112 #include <sys/sleepq.h>
113 #include <sys/lockdebug.h>
114 #include <sys/ktrace.h>
115 #include <sys/cpu.h>
116 #include <sys/lwpctl.h>
117 #include <sys/atomic.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 int find_stopped_child(struct proc *, pid_t, int, struct proc **, int *);
131 static void proc_free(struct proc *, struct rusage *);
132 
133 /*
134  * Fill in the appropriate signal information, and signal the parent.
135  */
136 static void
137 exit_psignal(struct proc *p, struct proc *pp, ksiginfo_t *ksi)
138 {
139 
140 	KSI_INIT(ksi);
141 	if ((ksi->ksi_signo = P_EXITSIG(p)) == SIGCHLD) {
142 		if (WIFSIGNALED(p->p_xstat)) {
143 			if (WCOREDUMP(p->p_xstat))
144 				ksi->ksi_code = CLD_DUMPED;
145 			else
146 				ksi->ksi_code = CLD_KILLED;
147 		} else {
148 			ksi->ksi_code = CLD_EXITED;
149 		}
150 	}
151 	/*
152 	 * We fill those in, even for non-SIGCHLD.
153 	 * It's safe to access p->p_cred unlocked here.
154 	 */
155 	ksi->ksi_pid = p->p_pid;
156 	ksi->ksi_uid = kauth_cred_geteuid(p->p_cred);
157 	ksi->ksi_status = p->p_xstat;
158 	/* XXX: is this still valid? */
159 	ksi->ksi_utime = p->p_stats->p_ru.ru_utime.tv_sec;
160 	ksi->ksi_stime = p->p_stats->p_ru.ru_stime.tv_sec;
161 }
162 
163 /*
164  * exit --
165  *	Death of process.
166  */
167 int
168 sys_exit(struct lwp *l, const struct sys_exit_args *uap, register_t *retval)
169 {
170 	/* {
171 		syscallarg(int)	rval;
172 	} */
173 	struct proc *p = l->l_proc;
174 
175 	/* Don't call exit1() multiple times in the same process. */
176 	KERNEL_LOCK(1, NULL);
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 		membar_producer();
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 
301 	/*
302 	 * If emulation has process exit hook, call it now.
303 	 * Set the exit status now so that the exit hook has
304 	 * an opportunity to tweak it (COMPAT_LINUX requires
305 	 * this for thread group emulation)
306 	 */
307 	p->p_xstat = rv;
308 	if (p->p_emul->e_proc_exit)
309 		(*p->p_emul->e_proc_exit)(p);
310 
311 	/*
312 	 * Free the VM resources we're still holding on to.
313 	 * We must do this from a valid thread because doing
314 	 * so may block. This frees vmspace, which we don't
315 	 * need anymore. The only remaining lwp is the one
316 	 * we run at this moment, nothing runs in userland
317 	 * anymore.
318 	 */
319 	uvm_proc_exit(p);
320 
321 	/*
322 	 * While we can still block, and mark the LWP as unswappable to
323 	 * prevent conflicts with the with the swapper.  We also shouldn't
324 	 * be swapped out, because we are about to exit and will release
325 	 * memory.
326 	 */
327 	uvm_lwp_hold(l);
328 
329 	/*
330 	 * Stop profiling.
331 	 */
332 	if ((p->p_stflag & PST_PROFIL) != 0) {
333 		mutex_spin_enter(&p->p_stmutex);
334 		stopprofclock(p);
335 		mutex_spin_exit(&p->p_stmutex);
336 	}
337 
338 	/*
339 	 * If parent is waiting for us to exit or exec, P_PPWAIT is set; we
340 	 * wake up the parent early to avoid deadlock.  We can do this once
341 	 * the VM resources are released.
342 	 */
343 	mutex_enter(&proclist_lock);
344 
345 	mutex_enter(&p->p_smutex);
346 	if (p->p_sflag & PS_PPWAIT) {
347 		p->p_sflag &= ~PS_PPWAIT;
348 		cv_broadcast(&p->p_pptr->p_waitcv);
349 	}
350 	mutex_exit(&p->p_smutex);
351 
352 	if (SESS_LEADER(p)) {
353 		struct vnode *vprele = NULL, *vprevoke = NULL;
354 		struct session *sp = p->p_session;
355 		struct tty *tp;
356 
357 		if (sp->s_ttyvp) {
358 			/*
359 			 * Controlling process.
360 			 * Signal foreground pgrp,
361 			 * drain controlling terminal
362 			 * and revoke access to controlling terminal.
363 			 */
364 			tp = sp->s_ttyp;
365 			mutex_spin_enter(&tty_lock);
366 			if (tp->t_session == sp) {
367 				if (tp->t_pgrp) {
368 					mutex_enter(&proclist_mutex);
369 					pgsignal(tp->t_pgrp, SIGHUP, 1);
370 					mutex_exit(&proclist_mutex);
371 				}
372 				/* we can't guarantee the revoke will do this */
373 				tp->t_pgrp = NULL;
374 				tp->t_session = NULL;
375 				mutex_spin_exit(&tty_lock);
376 				mutex_exit(&proclist_lock);
377 				(void) ttywait(tp);
378 				mutex_enter(&proclist_lock);
379 
380 				/* The tty could have been revoked. */
381 				vprevoke = sp->s_ttyvp;
382 			} else
383 				mutex_spin_exit(&tty_lock);
384 			vprele = sp->s_ttyvp;
385 			sp->s_ttyvp = NULL;
386 			/*
387 			 * s_ttyp is not zero'd; we use this to indicate
388 			 * that the session once had a controlling terminal.
389 			 * (for logging and informational purposes)
390 			 */
391 		}
392 		sp->s_leader = NULL;
393 
394 		if (vprevoke != NULL || vprele != NULL) {
395 			if (vprevoke != NULL) {
396 				SESSRELE(sp);
397 				mutex_exit(&proclist_lock);
398 				VOP_REVOKE(vprevoke, REVOKEALL);
399 			} else
400 				mutex_exit(&proclist_lock);
401 			if (vprele != NULL)
402 				vrele(vprele);
403 			mutex_enter(&proclist_lock);
404 		}
405 	}
406 	mutex_enter(&proclist_mutex);
407 	fixjobc(p, p->p_pgrp, 0);
408 	mutex_exit(&proclist_mutex);
409 
410 	/*
411 	 * Finalize the last LWP's specificdata, as well as the
412 	 * specificdata for the proc itself.
413 	 */
414 	lwp_finispecific(l);
415 	proc_finispecific(p);
416 
417 	/*
418 	 * Notify interested parties of our demise.
419 	 */
420 	KNOTE(&p->p_klist, NOTE_EXIT);
421 
422 
423 
424 #if PERFCTRS
425 	/*
426 	 * Save final PMC information in parent process & clean up.
427 	 */
428 	if (PMC_ENABLED(p)) {
429 		pmc_save_context(p);
430 		pmc_accumulate(p->p_pptr, p);
431 		pmc_process_exit(p);
432 	}
433 #endif
434 
435 	/*
436 	 * Reset p_opptr pointer of all former children which got
437 	 * traced by another process and were reparented. We reset
438 	 * it to NULL here; the trace detach code then reparents
439 	 * the child to initproc. We only check allproc list, since
440 	 * eventual former children on zombproc list won't reference
441 	 * p_opptr anymore.
442 	 */
443 	if (p->p_slflag & PSL_CHTRACED) {
444 		PROCLIST_FOREACH(q, &allproc) {
445 			if (q->p_opptr == p)
446 				q->p_opptr = NULL;
447 		}
448 	}
449 
450 	/*
451 	 * Give orphaned children to init(8).
452 	 */
453 	q = LIST_FIRST(&p->p_children);
454 	wakeinit = (q != NULL);
455 	for (; q != NULL; q = nq) {
456 		nq = LIST_NEXT(q, p_sibling);
457 
458 		/*
459 		 * Traced processes are killed since their existence
460 		 * means someone is screwing up. Since we reset the
461 		 * trace flags, the logic in sys_wait4() would not be
462 		 * triggered to reparent the process to its
463 		 * original parent, so we must do this here.
464 		 */
465 		if (q->p_slflag & PSL_TRACED) {
466 			mutex_enter(&p->p_smutex);
467 			q->p_slflag &= ~(PSL_TRACED|PSL_FSTRACE|PSL_SYSCALL);
468 			mutex_exit(&p->p_smutex);
469 			if (q->p_opptr != q->p_pptr) {
470 				struct proc *t = q->p_opptr;
471 				proc_reparent(q, t ? t : initproc);
472 				q->p_opptr = NULL;
473 			} else
474 				proc_reparent(q, initproc);
475 			killproc(q, "orphaned traced process");
476 		} else
477 			proc_reparent(q, initproc);
478 	}
479 
480 	/*
481 	 * Move proc from allproc to zombproc, it's now nearly ready to be
482 	 * collected by parent.
483 	 */
484 	mutex_enter(&proclist_mutex);
485 	LIST_REMOVE(l, l_list);
486 	LIST_REMOVE(p, p_list);
487 	LIST_INSERT_HEAD(&zombproc, p, p_list);
488 
489 	/*
490 	 * Mark the process as dead.  We must do this before we signal
491 	 * the parent.
492 	 */
493 	p->p_stat = SDEAD;
494 
495 	/* Put in front of parent's sibling list for parent to collect it */
496 	q = p->p_pptr;
497 	q->p_nstopchild++;
498 	if (LIST_FIRST(&q->p_children) != p) {
499 		/* Put child where it can be found quickly */
500 		LIST_REMOVE(p, p_sibling);
501 		LIST_INSERT_HEAD(&q->p_children, p, p_sibling);
502 	}
503 	mutex_exit(&proclist_mutex);
504 
505 	/*
506 	 * Notify parent that we're gone.  If parent has the P_NOCLDWAIT
507 	 * flag set, notify init instead (and hope it will handle
508 	 * this situation).
509 	 */
510 	mutex_enter(&q->p_mutex);
511 	if (q->p_flag & (PK_NOCLDWAIT|PK_CLDSIGIGN)) {
512 		proc_reparent(p, initproc);
513 		wakeinit = 1;
514 
515 		/*
516 		 * If this was the last child of our parent, notify
517 		 * parent, so in case he was wait(2)ing, he will
518 		 * continue.
519 		 */
520 		if (LIST_FIRST(&q->p_children) == NULL)
521 			cv_broadcast(&q->p_waitcv);
522 	}
523 	mutex_exit(&q->p_mutex);
524 
525 	/* Reload parent pointer, since p may have been reparented above */
526 	q = p->p_pptr;
527 
528 	if ((p->p_slflag & PSL_FSTRACE) == 0 && p->p_exitsig != 0) {
529 		exit_psignal(p, q, &ksi);
530 		mutex_enter(&proclist_mutex);
531 		kpsignal(q, &ksi, NULL);
532 		mutex_exit(&proclist_mutex);
533 	}
534 
535 	/* Calculate the final rusage info.  */
536 	calcru(p, &p->p_stats->p_ru.ru_utime, &p->p_stats->p_ru.ru_stime,
537 	    NULL, NULL);
538 
539 	if (wakeinit)
540 		cv_broadcast(&initproc->p_waitcv);
541 
542 	callout_destroy(&l->l_timeout_ch);
543 
544 	/*
545 	 * Remaining lwp resources will be freed in lwp_exit2() once we've
546 	 * switch to idle context; at that point, we will be marked as a
547 	 * full blown zombie.
548 	 *
549 	 * XXXSMP disable preemption.
550 	 */
551 	mutex_enter(&p->p_smutex);
552 	lwp_drainrefs(l);
553 	lwp_lock(l);
554 	l->l_prflag &= ~LPR_DETACHED;
555 	l->l_stat = LSZOMB;
556 	lwp_unlock(l);
557 	KASSERT(curlwp == l);
558 	KASSERT(p->p_nrlwps == 1);
559 	KASSERT(p->p_nlwps == 1);
560 	p->p_stat = SZOMB;
561 	p->p_nrlwps--;
562 	p->p_nzlwps++;
563 	p->p_ndlwps = 0;
564 	mutex_exit(&p->p_smutex);
565 
566 	/*
567 	 * Signal the parent to collect us, and drop the proclist lock.
568 	 * Drop debugger/procfs lock; no new references can be gained.
569 	 */
570 	cv_broadcast(&p->p_pptr->p_waitcv);
571 	mutex_exit(&proclist_lock);
572 	rw_exit(&p->p_reflock);
573 
574 	/* Verify that we hold no locks other than the kernel lock. */
575 #ifdef MULTIPROCESSOR
576 	LOCKDEBUG_BARRIER(&kernel_lock, 0);
577 #else
578 	LOCKDEBUG_BARRIER(NULL, 0);
579 #endif
580 
581 	/*
582 	 * NOTE: WE ARE NO LONGER ALLOWED TO SLEEP!
583 	 */
584 
585 	/*
586 	 * Give machine-dependent code a chance to free any MD LWP
587 	 * resources.  This must be done before uvm_lwp_exit(), in
588 	 * case these resources are in the PCB.
589 	 */
590 #ifndef __NO_CPU_LWP_FREE
591 	cpu_lwp_free(l, 1);
592 #endif
593 	pmap_deactivate(l);
594 
595 	/* This process no longer needs to hold the kernel lock. */
596 #ifdef notyet
597 	/* XXXSMP hold in lwp_userret() */
598 	KERNEL_UNLOCK_LAST(l);
599 #else
600 	KERNEL_UNLOCK_ALL(l, NULL);
601 #endif
602 
603 	lwp_exit_switchaway(l);
604 }
605 
606 void
607 exit_lwps(struct lwp *l)
608 {
609 	struct proc *p;
610 	struct lwp *l2;
611 	int error;
612 	lwpid_t waited;
613 #if defined(MULTIPROCESSOR)
614 	int nlocks;
615 #endif
616 
617 	KERNEL_UNLOCK_ALL(l, &nlocks);
618 
619 	p = l->l_proc;
620 	KASSERT(mutex_owned(&p->p_smutex));
621 
622  retry:
623 	/*
624 	 * Interrupt LWPs in interruptable sleep, unsuspend suspended
625 	 * LWPs and then wait for everyone else to finish.
626 	 */
627 	LIST_FOREACH(l2, &p->p_lwps, l_sibling) {
628 		if (l2 == l)
629 			continue;
630 		lwp_lock(l2);
631 		l2->l_flag |= LW_WEXIT;
632 		if ((l2->l_stat == LSSLEEP && (l2->l_flag & LW_SINTR)) ||
633 		    l2->l_stat == LSSUSPENDED || l2->l_stat == LSSTOP) {
634 		    	/* setrunnable() will release the lock. */
635 			setrunnable(l2);
636 			DPRINTF(("exit_lwps: Made %d.%d runnable\n",
637 			    p->p_pid, l2->l_lid));
638 			continue;
639 		}
640 		lwp_unlock(l2);
641 	}
642 	while (p->p_nlwps > 1) {
643 		DPRINTF(("exit_lwps: waiting for %d LWPs (%d zombies)\n",
644 		    p->p_nlwps, p->p_nzlwps));
645 		error = lwp_wait1(l, 0, &waited, LWPWAIT_EXITCONTROL);
646 		if (p->p_nlwps == 1)
647 			break;
648 		if (error == EDEADLK) {
649 			/*
650 			 * LWPs can get suspended/slept behind us.
651 			 * (eg. sa_setwoken)
652 			 * kick them again and retry.
653 			 */
654 			goto retry;
655 		}
656 		if (error)
657 			panic("exit_lwps: lwp_wait1 failed with error %d",
658 			    error);
659 		DPRINTF(("exit_lwps: Got LWP %d from lwp_wait1()\n", waited));
660 	}
661 
662 #if defined(MULTIPROCESSOR)
663 	if (nlocks > 0) {
664 		mutex_exit(&p->p_smutex);
665 		KERNEL_LOCK(nlocks, l);
666 		mutex_enter(&p->p_smutex);
667 	}
668 #endif /* defined(MULTIPROCESSOR) */
669 	KASSERT(p->p_nlwps == 1);
670 }
671 
672 int
673 do_sys_wait(struct lwp *l, int *pid, int *status, int options,
674     struct rusage *ru, int *was_zombie)
675 {
676 	struct proc	*child;
677 	int		error;
678 
679 	KERNEL_LOCK(1, NULL);		/* XXXSMP */
680 	mutex_enter(&proclist_lock);
681 	error = find_stopped_child(l->l_proc, *pid, options, &child, status);
682 	KERNEL_UNLOCK_ONE(NULL);	/* XXXSMP */
683 
684 	if (child == NULL) {
685 		mutex_exit(&proclist_lock);
686 		*pid = 0;
687 		return error;
688 	}
689 
690 	*pid = child->p_pid;
691 
692 	if (child->p_stat == SZOMB) {
693 		/* proc_free() will release the proclist_lock. */
694 		*was_zombie = 1;
695 		if (options & WNOWAIT)
696 			mutex_exit(&proclist_lock);
697 		else {
698 			proc_free(child, ru);
699 		}
700 	} else {
701 		/* Child state must have been SSTOP. */
702 		*was_zombie = 0;
703 		mutex_exit(&proclist_lock);
704 		*status = W_STOPCODE(*status);
705 	}
706 
707 	return 0;
708 }
709 
710 int
711 sys_wait4(struct lwp *l, const struct sys_wait4_args *uap, register_t *retval)
712 {
713 	/* {
714 		syscallarg(int)			pid;
715 		syscallarg(int *)		status;
716 		syscallarg(int)			options;
717 		syscallarg(struct rusage *)	rusage;
718 	} */
719 	int		status, error;
720 	int		was_zombie;
721 	struct rusage	ru;
722 	int pid = SCARG(uap, pid);
723 
724 	error = do_sys_wait(l, &pid, &status, SCARG(uap, options),
725 	    SCARG(uap, rusage) != NULL ? &ru : NULL, &was_zombie);
726 
727 	retval[0] = pid;
728 	if (pid == 0)
729 		return error;
730 
731 	if (SCARG(uap, rusage))
732 		error = copyout(&ru, SCARG(uap, rusage), sizeof(ru));
733 
734 	if (error == 0 && SCARG(uap, status))
735 		error = copyout(&status, SCARG(uap, status), sizeof(status));
736 
737 	return error;
738 }
739 
740 /*
741  * Scan list of child processes for a child process that has stopped or
742  * exited.  Used by sys_wait4 and 'compat' equivalents.
743  *
744  * Must be called with the proclist_lock held, and may release
745  * while waiting.
746  */
747 static int
748 find_stopped_child(struct proc *parent, pid_t pid, int options,
749 		   struct proc **child_p, int *status_p)
750 {
751 	struct proc *child, *dead;
752 	int error;
753 
754 	KASSERT(mutex_owned(&proclist_lock));
755 
756 	if (options & ~(WUNTRACED|WNOHANG|WALTSIG|WALLSIG)
757 	    && !(options & WOPTSCHECKED)) {
758 		*child_p = NULL;
759 		return EINVAL;
760 	}
761 
762 	if (pid == 0 && !(options & WOPTSCHECKED))
763 		pid = -parent->p_pgid;
764 
765 	for (;;) {
766 		error = ECHILD;
767 		dead = NULL;
768 
769 		mutex_enter(&proclist_mutex);
770 		LIST_FOREACH(child, &parent->p_children, p_sibling) {
771 			if (pid >= 0) {
772 				if (child->p_pid != pid) {
773 					child = p_find(pid, PFIND_ZOMBIE |
774 					    PFIND_LOCKED);
775 					if (child == NULL ||
776 					    child->p_pptr != parent) {
777 						child = NULL;
778 						break;
779 					}
780 				}
781 			} else if (pid != WAIT_ANY && child->p_pgid != -pid) {
782 				/* Child not in correct pgrp */
783 				continue;
784 			}
785 
786 			/*
787 			 * Wait for processes with p_exitsig != SIGCHLD
788 			 * processes only if WALTSIG is set; wait for
789 			 * processes with p_exitsig == SIGCHLD only
790 			 * if WALTSIG is clear.
791 			 */
792 			if (((options & WALLSIG) == 0) &&
793 			    (options & WALTSIG ? child->p_exitsig == SIGCHLD
794 						: P_EXITSIG(child) != SIGCHLD)){
795 				if (child->p_pid == pid) {
796 					child = NULL;
797 					break;
798 				}
799 				continue;
800 			}
801 
802 			error = 0;
803 			if ((options & WNOZOMBIE) == 0) {
804 				if (child->p_stat == SZOMB)
805 					break;
806 				if (child->p_stat == SDEAD) {
807 					/*
808 					 * We may occasionally arrive here
809 					 * after receiving a signal, but
810 					 * immediatley before the child
811 					 * process is zombified.  The wait
812 					 * will be short, so avoid returning
813 					 * to userspace.
814 					 */
815 					dead = child;
816 				}
817 			}
818 
819 			if (child->p_stat == SSTOP &&
820 			    child->p_waited == 0 &&
821 			    (child->p_slflag & PSL_TRACED ||
822 			    options & WUNTRACED)) {
823 				if ((options & WNOWAIT) == 0) {
824 					child->p_waited = 1;
825 					parent->p_nstopchild--;
826 				}
827 				break;
828 			}
829 			if (parent->p_nstopchild == 0 || child->p_pid == pid) {
830 				child = NULL;
831 				break;
832 			}
833 		}
834 
835 		if (child != NULL || error != 0 ||
836 		    ((options & WNOHANG) != 0 && dead == NULL)) {
837 		    	if (child != NULL) {
838 			    	*status_p = child->p_xstat;
839 			}
840 			mutex_exit(&proclist_mutex);
841 			*child_p = child;
842 			return error;
843 		}
844 
845 		/*
846 		 * Wait for another child process to stop.
847 		 */
848 		mutex_exit(&proclist_lock);
849 		error = cv_wait_sig(&parent->p_waitcv, &proclist_mutex);
850 		mutex_exit(&proclist_mutex);
851 		mutex_enter(&proclist_lock);
852 
853 		if (error != 0) {
854 			*child_p = NULL;
855 			return error;
856 		}
857 	}
858 }
859 
860 /*
861  * Free a process after parent has taken all the state info.  Must be called
862  * with the proclist lock held, and will release before returning.
863  *
864  * *ru is returned to the caller, and must be freed by the caller.
865  */
866 static void
867 proc_free(struct proc *p, struct rusage *ru)
868 {
869 	struct proc *parent;
870 	struct lwp *l;
871 	ksiginfo_t ksi;
872 	kauth_cred_t cred1, cred2;
873 	uid_t uid;
874 
875 	KASSERT(mutex_owned(&proclist_lock));
876 	KASSERT(p->p_nlwps == 1);
877 	KASSERT(p->p_nzlwps == 1);
878 	KASSERT(p->p_nrlwps == 0);
879 	KASSERT(p->p_stat == SZOMB);
880 
881 	/*
882 	 * If we got the child via ptrace(2) or procfs, and
883 	 * the parent is different (meaning the process was
884 	 * attached, rather than run as a child), then we need
885 	 * to give it back to the old parent, and send the
886 	 * parent the exit signal.  The rest of the cleanup
887 	 * will be done when the old parent waits on the child.
888 	 */
889 	if ((p->p_slflag & PSL_TRACED) != 0) {
890 		parent = p->p_pptr;
891 		if (p->p_opptr != parent){
892 			mutex_enter(&p->p_smutex);
893 			p->p_slflag &= ~(PSL_TRACED|PSL_FSTRACE|PSL_SYSCALL);
894 			mutex_exit(&p->p_smutex);
895 			parent = p->p_opptr;
896 			if (parent == NULL)
897 				parent = initproc;
898 			proc_reparent(p, parent);
899 			p->p_opptr = NULL;
900 			if (p->p_exitsig != 0) {
901 				exit_psignal(p, parent, &ksi);
902 				mutex_enter(&proclist_mutex);
903 				kpsignal(parent, &ksi, NULL);
904 				mutex_exit(&proclist_mutex);
905 			}
906 			KERNEL_LOCK(1, NULL);		/* XXXSMP */
907 			cv_broadcast(&parent->p_waitcv);
908 			KERNEL_UNLOCK_ONE(NULL);	/* XXXSMP */
909 			mutex_exit(&proclist_lock);
910 			return;
911 		}
912 	}
913 
914 	/*
915 	 * Finally finished with old proc entry.  Unlink it from its process
916 	 * group.
917 	 */
918 	leavepgrp(p);
919 
920 	parent = p->p_pptr;
921 	sched_proc_exit(parent, p);
922 	/*
923 	 * Add child times of exiting process onto its own times.
924 	 * This cannot be done any earlier else it might get done twice.
925 	 */
926 	ruadd(&p->p_stats->p_ru, &p->p_stats->p_cru);
927 	ruadd(&parent->p_stats->p_cru, &p->p_stats->p_ru);
928 	if (ru != NULL)
929 		*ru = p->p_stats->p_ru;
930 	p->p_xstat = 0;
931 
932 	/*
933 	 * At this point we are going to start freeing the final resources.
934 	 * If anyone tries to access the proc structure after here they will
935 	 * get a shock - bits are missing.  Attempt to make it hard!  We
936 	 * don't bother with any further locking past this point.
937 	 */
938 	mutex_enter(&proclist_mutex);
939 	p->p_stat = SIDL;		/* not even a zombie any more */
940 	LIST_REMOVE(p, p_list);	/* off zombproc */
941 	parent = p->p_pptr;
942 	p->p_pptr->p_nstopchild--;
943 	mutex_exit(&proclist_mutex);
944 	LIST_REMOVE(p, p_sibling);
945 
946 	/*
947 	 * Let pid be reallocated.
948 	 */
949 	proc_free_pid(p);
950 	mutex_exit(&proclist_lock);
951 
952 	l = LIST_FIRST(&p->p_lwps);
953 
954 	/*
955 	 * Delay release until after lwp_free.
956 	 */
957 	cred2 = l->l_cred;
958 
959 	/*
960 	 * Free the last LWP's resources.
961 	 *
962 	 * lwp_free ensures the LWP is no longer running on another CPU.
963 	 */
964 	lwp_free(l, false, true);
965 
966 	/*
967 	 * Now no one except us can reach the process p.
968 	 */
969 
970 	/*
971 	 * Decrement the count of procs running with this uid.
972 	 */
973 	cred1 = p->p_cred;
974 	uid = kauth_cred_getuid(cred1);
975 	(void)chgproccnt(uid, -1);
976 
977 	/*
978 	 * Release substructures.
979 	 */
980 
981 	limfree(p->p_limit);
982 	pstatsfree(p->p_stats);
983 	kauth_cred_free(cred1);
984 	kauth_cred_free(cred2);
985 
986 	/*
987 	 * Release reference to text vnode
988 	 */
989 	if (p->p_textvp)
990 		vrele(p->p_textvp);
991 
992 	mutex_destroy(&p->p_auxlock);
993 	mutex_destroy(&p->p_mutex);
994 	mutex_destroy(&p->p_stmutex);
995 	mutex_destroy(&p->p_smutex);
996 	cv_destroy(&p->p_waitcv);
997 	cv_destroy(&p->p_lwpcv);
998 	rw_destroy(&p->p_reflock);
999 
1000 	proc_free_mem(p);
1001 }
1002 
1003 /*
1004  * make process 'parent' the new parent of process 'child'.
1005  *
1006  * Must be called with proclist_lock lock held.
1007  */
1008 void
1009 proc_reparent(struct proc *child, struct proc *parent)
1010 {
1011 
1012 	KASSERT(mutex_owned(&proclist_lock));
1013 
1014 	if (child->p_pptr == parent)
1015 		return;
1016 
1017 	mutex_enter(&proclist_mutex);
1018 	if (child->p_stat == SZOMB ||
1019 	    (child->p_stat == SSTOP && !child->p_waited)) {
1020 		child->p_pptr->p_nstopchild--;
1021 		parent->p_nstopchild++;
1022 	}
1023 	mutex_exit(&proclist_mutex);
1024 	if (parent == initproc)
1025 		child->p_exitsig = SIGCHLD;
1026 
1027 	LIST_REMOVE(child, p_sibling);
1028 	LIST_INSERT_HEAD(&parent->p_children, child, p_sibling);
1029 	child->p_pptr = parent;
1030 }
1031