xref: /netbsd-src/sys/kern/kern_exit.c (revision 313c6c94c424eed90c7b7e494aa83308a0a5d0ce)
1 /*	$NetBSD: kern_exit.c,v 1.221 2009/05/08 13:32:59 yamt 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 1.221 2009/05/08 13:32:59 yamt Exp $");
71 
72 #include "opt_ktrace.h"
73 #include "opt_perfctrs.h"
74 #include "opt_sa.h"
75 #include "opt_sysv.h"
76 
77 #include <sys/param.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/pool.h>
91 #include <sys/uidinfo.h>
92 #if defined(PERFCTRS)
93 #include <sys/pmc.h>
94 #endif
95 #include <sys/ptrace.h>
96 #include <sys/acct.h>
97 #include <sys/filedesc.h>
98 #include <sys/ras.h>
99 #include <sys/signalvar.h>
100 #include <sys/sched.h>
101 #include <sys/sa.h>
102 #include <sys/savar.h>
103 #include <sys/mount.h>
104 #include <sys/syscallargs.h>
105 #include <sys/kauth.h>
106 #include <sys/sleepq.h>
107 #include <sys/lockdebug.h>
108 #include <sys/ktrace.h>
109 #include <sys/cpu.h>
110 #include <sys/lwpctl.h>
111 #include <sys/atomic.h>
112 
113 #include <uvm/uvm_extern.h>
114 
115 #ifdef DEBUG_EXIT
116 int debug_exit = 0;
117 #define DPRINTF(x) if (debug_exit) printf x
118 #else
119 #define DPRINTF(x)
120 #endif
121 
122 static int find_stopped_child(struct proc *, pid_t, int, struct proc **, int *);
123 static void proc_free(struct proc *, struct rusage *);
124 
125 /*
126  * Fill in the appropriate signal information, and signal the parent.
127  */
128 static void
129 exit_psignal(struct proc *p, struct proc *pp, ksiginfo_t *ksi)
130 {
131 
132 	KSI_INIT(ksi);
133 	if ((ksi->ksi_signo = P_EXITSIG(p)) == SIGCHLD) {
134 		if (WIFSIGNALED(p->p_xstat)) {
135 			if (WCOREDUMP(p->p_xstat))
136 				ksi->ksi_code = CLD_DUMPED;
137 			else
138 				ksi->ksi_code = CLD_KILLED;
139 		} else {
140 			ksi->ksi_code = CLD_EXITED;
141 		}
142 	}
143 	/*
144 	 * We fill those in, even for non-SIGCHLD.
145 	 * It's safe to access p->p_cred unlocked here.
146 	 */
147 	ksi->ksi_pid = p->p_pid;
148 	ksi->ksi_uid = kauth_cred_geteuid(p->p_cred);
149 	ksi->ksi_status = p->p_xstat;
150 	/* XXX: is this still valid? */
151 	ksi->ksi_utime = p->p_stats->p_ru.ru_utime.tv_sec;
152 	ksi->ksi_stime = p->p_stats->p_ru.ru_stime.tv_sec;
153 }
154 
155 /*
156  * exit --
157  *	Death of process.
158  */
159 int
160 sys_exit(struct lwp *l, const struct sys_exit_args *uap, register_t *retval)
161 {
162 	/* {
163 		syscallarg(int)	rval;
164 	} */
165 	struct proc *p = l->l_proc;
166 
167 	/* Don't call exit1() multiple times in the same process. */
168 	mutex_enter(p->p_lock);
169 	if (p->p_sflag & PS_WEXIT) {
170 		mutex_exit(p->p_lock);
171 		lwp_exit(l);
172 	}
173 
174 	/* exit1() will release the mutex. */
175 	exit1(l, W_EXITCODE(SCARG(uap, rval), 0));
176 	/* NOTREACHED */
177 	return (0);
178 }
179 
180 /*
181  * Exit: deallocate address space and other resources, change proc state
182  * to zombie, and unlink proc from allproc and parent's lists.  Save exit
183  * status and rusage for wait().  Check for child processes and orphan them.
184  *
185  * Must be called with p->p_lock held.  Does not return.
186  */
187 void
188 exit1(struct lwp *l, int rv)
189 {
190 	struct proc	*p, *q, *nq;
191 	struct pgrp	*pgrp;
192 	ksiginfo_t	ksi;
193 	ksiginfoq_t	kq;
194 	int		wakeinit, sa;
195 
196 	p = l->l_proc;
197 
198 	KASSERT(mutex_owned(p->p_lock));
199 
200 	if (__predict_false(p == initproc))
201 		panic("init died (signal %d, exit %d)",
202 		    WTERMSIG(rv), WEXITSTATUS(rv));
203 
204 	/*
205 	 * Disable scheduler activation upcalls.  We're trying to get out of
206 	 * here.
207 	 */
208 	sa = 0;
209 #ifdef KERN_SA
210 	if ((p->p_sa != NULL)) {
211 		l->l_pflag |= LP_SA_NOBLOCK;
212 		sa = 1;
213 	}
214 #endif
215 
216 	p->p_sflag |= PS_WEXIT;
217 
218 	/*
219 	 * Force all other LWPs to exit before we do.  Only then can we
220 	 * begin to tear down the rest of the process state.
221 	 */
222 	if (sa || p->p_nlwps > 1)
223 		exit_lwps(l);
224 
225 	ksiginfo_queue_init(&kq);
226 
227 	/*
228 	 * If we have been asked to stop on exit, do so now.
229 	 */
230 	if (__predict_false(p->p_sflag & PS_STOPEXIT)) {
231 		KERNEL_UNLOCK_ALL(l, &l->l_biglocks);
232 		sigclearall(p, &contsigmask, &kq);
233 		p->p_waited = 0;
234 		membar_producer();
235 		p->p_stat = SSTOP;
236 		lwp_lock(l);
237 		p->p_nrlwps--;
238 		l->l_stat = LSSTOP;
239 		mutex_exit(p->p_lock);
240 		mi_switch(l);
241 		KERNEL_LOCK(l->l_biglocks, l);
242 		mutex_enter(p->p_lock);
243 	}
244 
245 	/*
246 	 * Bin any remaining signals and mark the process as dying so it will
247 	 * not be found for, e.g. signals.
248 	 */
249 	sigfillset(&p->p_sigctx.ps_sigignore);
250 	sigclearall(p, NULL, &kq);
251 	p->p_stat = SDYING;
252 	mutex_exit(p->p_lock);
253 	ksiginfo_queue_drain(&kq);
254 
255 	/* Destroy any lwpctl info. */
256 	if (p->p_lwpctl != NULL)
257 		lwp_ctl_exit();
258 
259 	/*
260 	 * Drain all remaining references that procfs, ptrace and others may
261 	 * have on the process.
262 	 */
263 	rw_enter(&p->p_reflock, RW_WRITER);
264 
265 	DPRINTF(("exit1: %d.%d exiting.\n", p->p_pid, l->l_lid));
266 
267 	timers_free(p, TIMERS_ALL);
268 #if defined(__HAVE_RAS)
269 	ras_purgeall();
270 #endif
271 
272 	/*
273 	 * Close open files, release open-file table and free signal
274 	 * actions.  This may block!
275 	 */
276 	fd_free();
277 	cwdfree(p->p_cwdi);
278 	p->p_cwdi = NULL;
279 	doexithooks(p);
280 	sigactsfree(p->p_sigacts);
281 
282 	/*
283 	 * Write out accounting data.
284 	 */
285 	(void)acct_process(l);
286 
287 #ifdef KTRACE
288 	/*
289 	 * Release trace file.
290 	 */
291 	if (p->p_tracep != NULL) {
292 		mutex_enter(&ktrace_lock);
293 		ktrderef(p);
294 		mutex_exit(&ktrace_lock);
295 	}
296 #endif
297 
298 	/*
299 	 * If emulation has process exit hook, call it now.
300 	 * Set the exit status now so that the exit hook has
301 	 * an opportunity to tweak it (COMPAT_LINUX requires
302 	 * this for thread group emulation)
303 	 */
304 	p->p_xstat = rv;
305 	if (p->p_emul->e_proc_exit)
306 		(*p->p_emul->e_proc_exit)(p);
307 
308 	/*
309 	 * Free the VM resources we're still holding on to.
310 	 * We must do this from a valid thread because doing
311 	 * so may block. This frees vmspace, which we don't
312 	 * need anymore. The only remaining lwp is the one
313 	 * we run at this moment, nothing runs in userland
314 	 * anymore.
315 	 */
316 	uvm_proc_exit(p);
317 
318 	/*
319 	 * Stop profiling.
320 	 */
321 	if (__predict_false((p->p_stflag & PST_PROFIL) != 0)) {
322 		mutex_spin_enter(&p->p_stmutex);
323 		stopprofclock(p);
324 		mutex_spin_exit(&p->p_stmutex);
325 	}
326 
327 	/*
328 	 * If parent is waiting for us to exit or exec, PL_PPWAIT is set; we
329 	 * wake up the parent early to avoid deadlock.  We can do this once
330 	 * the VM resources are released.
331 	 */
332 	mutex_enter(proc_lock);
333 	if (p->p_lflag & PL_PPWAIT) {
334 		p->p_lflag &= ~PL_PPWAIT;
335 		cv_broadcast(&p->p_pptr->p_waitcv);
336 	}
337 
338 	if (SESS_LEADER(p)) {
339 		struct vnode *vprele = NULL, *vprevoke = NULL;
340 		struct session *sp = p->p_session;
341 		struct tty *tp;
342 
343 		if (sp->s_ttyvp) {
344 			/*
345 			 * Controlling process.
346 			 * Signal foreground pgrp,
347 			 * drain controlling terminal
348 			 * and revoke access to controlling terminal.
349 			 */
350 			tp = sp->s_ttyp;
351 			mutex_spin_enter(&tty_lock);
352 			if (tp->t_session == sp) {
353 				/* we can't guarantee the revoke will do this */
354 				pgrp = tp->t_pgrp;
355 				tp->t_pgrp = NULL;
356 				tp->t_session = NULL;
357 				mutex_spin_exit(&tty_lock);
358 				if (pgrp != NULL) {
359 					pgsignal(pgrp, SIGHUP, 1);
360 				}
361 				mutex_exit(proc_lock);
362 				(void) ttywait(tp);
363 				mutex_enter(proc_lock);
364 
365 				/* The tty could have been revoked. */
366 				vprevoke = sp->s_ttyvp;
367 			} else
368 				mutex_spin_exit(&tty_lock);
369 			vprele = sp->s_ttyvp;
370 			sp->s_ttyvp = NULL;
371 			/*
372 			 * s_ttyp is not zero'd; we use this to indicate
373 			 * that the session once had a controlling terminal.
374 			 * (for logging and informational purposes)
375 			 */
376 		}
377 		sp->s_leader = NULL;
378 
379 		if (vprevoke != NULL || vprele != NULL) {
380 			if (vprevoke != NULL) {
381 				/* Releases proc_lock. */
382 				proc_sessrele(sp);
383 				VOP_REVOKE(vprevoke, REVOKEALL);
384 			} else
385 				mutex_exit(proc_lock);
386 			if (vprele != NULL)
387 				vrele(vprele);
388 			mutex_enter(proc_lock);
389 		}
390 	}
391 	fixjobc(p, p->p_pgrp, 0);
392 
393 	/*
394 	 * Finalize the last LWP's specificdata, as well as the
395 	 * specificdata for the proc itself.
396 	 */
397 	lwp_finispecific(l);
398 	proc_finispecific(p);
399 
400 	/*
401 	 * Notify interested parties of our demise.
402 	 */
403 	KNOTE(&p->p_klist, NOTE_EXIT);
404 
405 #if PERFCTRS
406 	/*
407 	 * Save final PMC information in parent process & clean up.
408 	 */
409 	if (PMC_ENABLED(p)) {
410 		pmc_save_context(p);
411 		pmc_accumulate(p->p_pptr, p);
412 		pmc_process_exit(p);
413 	}
414 #endif
415 
416 	/*
417 	 * Reset p_opptr pointer of all former children which got
418 	 * traced by another process and were reparented. We reset
419 	 * it to NULL here; the trace detach code then reparents
420 	 * the child to initproc. We only check allproc list, since
421 	 * eventual former children on zombproc list won't reference
422 	 * p_opptr anymore.
423 	 */
424 	if (__predict_false(p->p_slflag & PSL_CHTRACED)) {
425 		PROCLIST_FOREACH(q, &allproc) {
426 			if ((q->p_flag & PK_MARKER) != 0)
427 				continue;
428 			if (q->p_opptr == p)
429 				q->p_opptr = NULL;
430 		}
431 	}
432 
433 	/*
434 	 * Give orphaned children to init(8).
435 	 */
436 	q = LIST_FIRST(&p->p_children);
437 	wakeinit = (q != NULL);
438 	for (; q != NULL; q = nq) {
439 		nq = LIST_NEXT(q, p_sibling);
440 
441 		/*
442 		 * Traced processes are killed since their existence
443 		 * means someone is screwing up. Since we reset the
444 		 * trace flags, the logic in sys_wait4() would not be
445 		 * triggered to reparent the process to its
446 		 * original parent, so we must do this here.
447 		 */
448 		if (__predict_false(q->p_slflag & PSL_TRACED)) {
449 			mutex_enter(p->p_lock);
450 			q->p_slflag &= ~(PSL_TRACED|PSL_FSTRACE|PSL_SYSCALL);
451 			mutex_exit(p->p_lock);
452 			if (q->p_opptr != q->p_pptr) {
453 				struct proc *t = q->p_opptr;
454 				proc_reparent(q, t ? t : initproc);
455 				q->p_opptr = NULL;
456 			} else
457 				proc_reparent(q, initproc);
458 			killproc(q, "orphaned traced process");
459 		} else
460 			proc_reparent(q, initproc);
461 	}
462 
463 	/*
464 	 * Move proc from allproc to zombproc, it's now nearly ready to be
465 	 * collected by parent.
466 	 */
467 	LIST_REMOVE(l, l_list);
468 	LIST_REMOVE(p, p_list);
469 	LIST_INSERT_HEAD(&zombproc, p, p_list);
470 
471 	/*
472 	 * Mark the process as dead.  We must do this before we signal
473 	 * the parent.
474 	 */
475 	p->p_stat = SDEAD;
476 
477 	/* Put in front of parent's sibling list for parent to collect it */
478 	q = p->p_pptr;
479 	q->p_nstopchild++;
480 	if (LIST_FIRST(&q->p_children) != p) {
481 		/* Put child where it can be found quickly */
482 		LIST_REMOVE(p, p_sibling);
483 		LIST_INSERT_HEAD(&q->p_children, p, p_sibling);
484 	}
485 
486 	/*
487 	 * Notify parent that we're gone.  If parent has the P_NOCLDWAIT
488 	 * flag set, notify init instead (and hope it will handle
489 	 * this situation).
490 	 */
491 	if (q->p_flag & (PK_NOCLDWAIT|PK_CLDSIGIGN)) {
492 		proc_reparent(p, initproc);
493 		wakeinit = 1;
494 
495 		/*
496 		 * If this was the last child of our parent, notify
497 		 * parent, so in case he was wait(2)ing, he will
498 		 * continue.
499 		 */
500 		if (LIST_FIRST(&q->p_children) == NULL)
501 			cv_broadcast(&q->p_waitcv);
502 	}
503 
504 	/* Reload parent pointer, since p may have been reparented above */
505 	q = p->p_pptr;
506 
507 	if (__predict_false((p->p_slflag & PSL_FSTRACE) == 0 && p->p_exitsig != 0)) {
508 		exit_psignal(p, q, &ksi);
509 		kpsignal(q, &ksi, NULL);
510 	}
511 
512 	/* Calculate the final rusage info.  */
513 	calcru(p, &p->p_stats->p_ru.ru_utime, &p->p_stats->p_ru.ru_stime,
514 	    NULL, NULL);
515 
516 	if (wakeinit)
517 		cv_broadcast(&initproc->p_waitcv);
518 
519 	callout_destroy(&l->l_timeout_ch);
520 
521 	/*
522 	 * Remaining lwp resources will be freed in lwp_exit2() once we've
523 	 * switch to idle context; at that point, we will be marked as a
524 	 * full blown zombie.
525 	 */
526 	mutex_enter(p->p_lock);
527 	lwp_drainrefs(l);
528 	lwp_lock(l);
529 	l->l_prflag &= ~LPR_DETACHED;
530 	l->l_stat = LSZOMB;
531 	lwp_unlock(l);
532 	KASSERT(curlwp == l);
533 	KASSERT(p->p_nrlwps == 1);
534 	KASSERT(p->p_nlwps == 1);
535 	p->p_stat = SZOMB;
536 	p->p_nrlwps--;
537 	p->p_nzlwps++;
538 	p->p_ndlwps = 0;
539 	mutex_exit(p->p_lock);
540 
541 	/*
542 	 * Signal the parent to collect us, and drop the proclist lock.
543 	 * Drop debugger/procfs lock; no new references can be gained.
544 	 */
545 	cv_broadcast(&p->p_pptr->p_waitcv);
546 	rw_exit(&p->p_reflock);
547 	mutex_exit(proc_lock);
548 
549 	/* Verify that we hold no locks other than the kernel lock. */
550 	LOCKDEBUG_BARRIER(&kernel_lock, 0);
551 
552 	/*
553 	 * NOTE: WE ARE NO LONGER ALLOWED TO SLEEP!
554 	 */
555 
556 	/*
557 	 * Give machine-dependent code a chance to free any MD LWP
558 	 * resources.  This must be done before uvm_lwp_exit(), in
559 	 * case these resources are in the PCB.
560 	 */
561 #ifndef __NO_CPU_LWP_FREE
562 	cpu_lwp_free(l, 1);
563 #endif
564 	pmap_deactivate(l);
565 
566 	/* This process no longer needs to hold the kernel lock. */
567 #ifdef notyet
568 	/* XXXSMP hold in lwp_userret() */
569 	KERNEL_UNLOCK_LAST(l);
570 #else
571 	KERNEL_UNLOCK_ALL(l, NULL);
572 #endif
573 
574 	lwp_exit_switchaway(l);
575 }
576 
577 void
578 exit_lwps(struct lwp *l)
579 {
580 	struct proc *p;
581 	struct lwp *l2;
582 	int error;
583 	lwpid_t waited;
584 	int nlocks;
585 
586 	KERNEL_UNLOCK_ALL(l, &nlocks);
587 
588 	p = l->l_proc;
589 	KASSERT(mutex_owned(p->p_lock));
590 
591 #ifdef KERN_SA
592 	if (p->p_sa != NULL) {
593 		struct sadata_vp *vp;
594 		SLIST_FOREACH(vp, &p->p_sa->sa_vps, savp_next) {
595 			/*
596 			 * Make SA-cached LWPs normal process interruptable
597 			 * so that the exit code can wake them. Locking
598 			 * savp_mutex locks all the lwps on this vp that
599 			 * we need to adjust.
600 			 */
601 			mutex_enter(&vp->savp_mutex);
602 			DPRINTF(("exit_lwps: Making cached LWPs of %d on "
603 			    "VP %d interruptable: ", p->p_pid, vp->savp_id));
604 			TAILQ_FOREACH(l2, &vp->savp_lwpcache, l_sleepchain) {
605 				l2->l_flag |= LW_SINTR;
606 				DPRINTF(("%d ", l2->l_lid));
607 			}
608 			DPRINTF(("\n"));
609 
610 			DPRINTF(("exit_lwps: Making unblocking LWPs of %d on "
611 			    "VP %d interruptable: ", p->p_pid, vp->savp_id));
612 			TAILQ_FOREACH(l2, &vp->savp_woken, l_sleepchain) {
613 				vp->savp_woken_count--;
614 				l2->l_flag |= LW_SINTR;
615 				DPRINTF(("%d ", l2->l_lid));
616 			}
617 			DPRINTF(("\n"));
618 			mutex_exit(&vp->savp_mutex);
619 		}
620 	}
621 #endif
622 
623  retry:
624 	/*
625 	 * Interrupt LWPs in interruptable sleep, unsuspend suspended
626 	 * LWPs and then wait for everyone else to finish.
627 	 */
628 	LIST_FOREACH(l2, &p->p_lwps, l_sibling) {
629 		if (l2 == l)
630 			continue;
631 		lwp_lock(l2);
632 		l2->l_flag &= ~LW_SA;
633 		l2->l_flag |= LW_WEXIT;
634 		if ((l2->l_stat == LSSLEEP && (l2->l_flag & LW_SINTR)) ||
635 		    l2->l_stat == LSSUSPENDED || l2->l_stat == LSSTOP) {
636 		    	/* setrunnable() will release the lock. */
637 			setrunnable(l2);
638 			DPRINTF(("exit_lwps: Made %d.%d runnable\n",
639 			    p->p_pid, l2->l_lid));
640 			continue;
641 		}
642 		lwp_unlock(l2);
643 	}
644 	while (p->p_nlwps > 1) {
645 		DPRINTF(("exit_lwps: waiting for %d LWPs (%d zombies)\n",
646 		    p->p_nlwps, p->p_nzlwps));
647 		error = lwp_wait1(l, 0, &waited, LWPWAIT_EXITCONTROL);
648 		if (p->p_nlwps == 1)
649 			break;
650 		if (error == EDEADLK) {
651 			/*
652 			 * LWPs can get suspended/slept behind us.
653 			 * (eg. sa_setwoken)
654 			 * kick them again and retry.
655 			 */
656 			goto retry;
657 		}
658 		if (error)
659 			panic("exit_lwps: lwp_wait1 failed with error %d",
660 			    error);
661 		DPRINTF(("exit_lwps: Got LWP %d from lwp_wait1()\n", waited));
662 	}
663 
664 	KERNEL_LOCK(nlocks, l);
665 	KASSERT(p->p_nlwps == 1);
666 }
667 
668 int
669 do_sys_wait(struct lwp *l, int *pid, int *status, int options,
670     struct rusage *ru, int *was_zombie)
671 {
672 	struct proc	*child;
673 	int		error;
674 
675 	mutex_enter(proc_lock);
676 	error = find_stopped_child(l->l_proc, *pid, options, &child, status);
677 
678 	if (child == NULL) {
679 		mutex_exit(proc_lock);
680 		*pid = 0;
681 		return error;
682 	}
683 
684 	*pid = child->p_pid;
685 
686 	if (child->p_stat == SZOMB) {
687 		/* proc_free() will release the proc_lock. */
688 		*was_zombie = 1;
689 		if (options & WNOWAIT)
690 			mutex_exit(proc_lock);
691 		else {
692 			proc_free(child, ru);
693 		}
694 	} else {
695 		/* Child state must have been SSTOP. */
696 		*was_zombie = 0;
697 		mutex_exit(proc_lock);
698 		*status = W_STOPCODE(*status);
699 	}
700 
701 	return 0;
702 }
703 
704 int
705 sys___wait450(struct lwp *l, const struct sys___wait450_args *uap, register_t *retval)
706 {
707 	/* {
708 		syscallarg(int)			pid;
709 		syscallarg(int *)		status;
710 		syscallarg(int)			options;
711 		syscallarg(struct rusage *)	rusage;
712 	} */
713 	int		status, error;
714 	int		was_zombie;
715 	struct rusage	ru;
716 	int pid = SCARG(uap, pid);
717 
718 	error = do_sys_wait(l, &pid, &status, SCARG(uap, options),
719 	    SCARG(uap, rusage) != NULL ? &ru : NULL, &was_zombie);
720 
721 	retval[0] = pid;
722 	if (pid == 0)
723 		return error;
724 
725 	if (SCARG(uap, rusage))
726 		error = copyout(&ru, SCARG(uap, rusage), sizeof(ru));
727 
728 	if (error == 0 && SCARG(uap, status))
729 		error = copyout(&status, SCARG(uap, status), sizeof(status));
730 
731 	return error;
732 }
733 
734 /*
735  * Scan list of child processes for a child process that has stopped or
736  * exited.  Used by sys_wait4 and 'compat' equivalents.
737  *
738  * Must be called with the proc_lock held, and may release while waiting.
739  */
740 static int
741 find_stopped_child(struct proc *parent, pid_t pid, int options,
742 		   struct proc **child_p, int *status_p)
743 {
744 	struct proc *child, *dead;
745 	int error;
746 
747 	KASSERT(mutex_owned(proc_lock));
748 
749 	if (options & ~(WUNTRACED|WNOHANG|WALTSIG|WALLSIG)
750 	    && !(options & WOPTSCHECKED)) {
751 		*child_p = NULL;
752 		return EINVAL;
753 	}
754 
755 	if (pid == 0 && !(options & WOPTSCHECKED))
756 		pid = -parent->p_pgid;
757 
758 	for (;;) {
759 		error = ECHILD;
760 		dead = NULL;
761 
762 		LIST_FOREACH(child, &parent->p_children, p_sibling) {
763 			if (pid >= 0) {
764 				if (child->p_pid != pid) {
765 					child = p_find(pid, PFIND_ZOMBIE |
766 					    PFIND_LOCKED);
767 					if (child == NULL ||
768 					    child->p_pptr != parent) {
769 						child = NULL;
770 						break;
771 					}
772 				}
773 			} else if (pid != WAIT_ANY && child->p_pgid != -pid) {
774 				/* Child not in correct pgrp */
775 				continue;
776 			}
777 
778 			/*
779 			 * Wait for processes with p_exitsig != SIGCHLD
780 			 * processes only if WALTSIG is set; wait for
781 			 * processes with p_exitsig == SIGCHLD only
782 			 * if WALTSIG is clear.
783 			 */
784 			if (((options & WALLSIG) == 0) &&
785 			    (options & WALTSIG ? child->p_exitsig == SIGCHLD
786 						: P_EXITSIG(child) != SIGCHLD)){
787 				if (child->p_pid == pid) {
788 					child = NULL;
789 					break;
790 				}
791 				continue;
792 			}
793 
794 			error = 0;
795 			if ((options & WNOZOMBIE) == 0) {
796 				if (child->p_stat == SZOMB)
797 					break;
798 				if (child->p_stat == SDEAD) {
799 					/*
800 					 * We may occasionally arrive here
801 					 * after receiving a signal, but
802 					 * immediatley before the child
803 					 * process is zombified.  The wait
804 					 * will be short, so avoid returning
805 					 * to userspace.
806 					 */
807 					dead = child;
808 				}
809 			}
810 
811 			if (child->p_stat == SSTOP &&
812 			    child->p_waited == 0 &&
813 			    (child->p_slflag & PSL_TRACED ||
814 			    options & WUNTRACED)) {
815 				if ((options & WNOWAIT) == 0) {
816 					child->p_waited = 1;
817 					parent->p_nstopchild--;
818 				}
819 				break;
820 			}
821 			if (parent->p_nstopchild == 0 || child->p_pid == pid) {
822 				child = NULL;
823 				break;
824 			}
825 		}
826 
827 		if (child != NULL || error != 0 ||
828 		    ((options & WNOHANG) != 0 && dead == NULL)) {
829 		    	if (child != NULL) {
830 			    	*status_p = child->p_xstat;
831 			}
832 			*child_p = child;
833 			return error;
834 		}
835 
836 		/*
837 		 * Wait for another child process to stop.
838 		 */
839 		error = cv_wait_sig(&parent->p_waitcv, proc_lock);
840 
841 		if (error != 0) {
842 			*child_p = NULL;
843 			return error;
844 		}
845 	}
846 }
847 
848 /*
849  * Free a process after parent has taken all the state info.  Must be called
850  * with the proclist lock held, and will release before returning.
851  *
852  * *ru is returned to the caller, and must be freed by the caller.
853  */
854 static void
855 proc_free(struct proc *p, struct rusage *ru)
856 {
857 	struct proc *parent = p->p_pptr;
858 	struct lwp *l;
859 	ksiginfo_t ksi;
860 	kauth_cred_t cred1, cred2;
861 	uid_t uid;
862 
863 	KASSERT(mutex_owned(proc_lock));
864 	KASSERT(p->p_nlwps == 1);
865 	KASSERT(p->p_nzlwps == 1);
866 	KASSERT(p->p_nrlwps == 0);
867 	KASSERT(p->p_stat == SZOMB);
868 
869 	/*
870 	 * If we got the child via ptrace(2) or procfs, and
871 	 * the parent is different (meaning the process was
872 	 * attached, rather than run as a child), then we need
873 	 * to give it back to the old parent, and send the
874 	 * parent the exit signal.  The rest of the cleanup
875 	 * will be done when the old parent waits on the child.
876 	 */
877 	if ((p->p_slflag & PSL_TRACED) != 0 && p->p_opptr != parent) {
878 		mutex_enter(p->p_lock);
879 		p->p_slflag &= ~(PSL_TRACED|PSL_FSTRACE|PSL_SYSCALL);
880 		mutex_exit(p->p_lock);
881 		parent = (p->p_opptr == NULL) ? initproc : p->p_opptr;
882 		proc_reparent(p, parent);
883 		p->p_opptr = NULL;
884 		if (p->p_exitsig != 0) {
885 			exit_psignal(p, parent, &ksi);
886 			kpsignal(parent, &ksi, NULL);
887 		}
888 		cv_broadcast(&parent->p_waitcv);
889 		mutex_exit(proc_lock);
890 		return;
891 	}
892 
893 	sched_proc_exit(parent, p);
894 
895 	/*
896 	 * Add child times of exiting process onto its own times.
897 	 * This cannot be done any earlier else it might get done twice.
898 	 */
899 	l = LIST_FIRST(&p->p_lwps);
900 	p->p_stats->p_ru.ru_nvcsw += (l->l_ncsw - l->l_nivcsw);
901 	p->p_stats->p_ru.ru_nivcsw += l->l_nivcsw;
902 	ruadd(&p->p_stats->p_ru, &l->l_ru);
903 	ruadd(&p->p_stats->p_ru, &p->p_stats->p_cru);
904 	ruadd(&parent->p_stats->p_cru, &p->p_stats->p_ru);
905 	if (ru != NULL)
906 		*ru = p->p_stats->p_ru;
907 	p->p_xstat = 0;
908 
909 	/* Release any SA state. */
910 #ifdef KERN_SA
911 	if (p->p_sa)
912 		sa_release(p);
913 #endif
914 
915 	/*
916 	 * At this point we are going to start freeing the final resources.
917 	 * If anyone tries to access the proc structure after here they will
918 	 * get a shock - bits are missing.  Attempt to make it hard!  We
919 	 * don't bother with any further locking past this point.
920 	 */
921 	p->p_stat = SIDL;		/* not even a zombie any more */
922 	LIST_REMOVE(p, p_list);	/* off zombproc */
923 	parent->p_nstopchild--;
924 	LIST_REMOVE(p, p_sibling);
925 
926 	/*
927 	 * Let pid be reallocated.
928 	 */
929 	proc_free_pid(p);
930 
931 	/*
932 	 * Unlink process from its process group.
933 	 * Releases the proc_lock.
934 	 */
935 	proc_leavepgrp(p);
936 
937 	/*
938 	 * Delay release until after lwp_free.
939 	 */
940 	cred2 = l->l_cred;
941 
942 	/*
943 	 * Free the last LWP's resources.
944 	 *
945 	 * lwp_free ensures the LWP is no longer running on another CPU.
946 	 */
947 	lwp_free(l, false, true);
948 
949 	/*
950 	 * Now no one except us can reach the process p.
951 	 */
952 
953 	/*
954 	 * Decrement the count of procs running with this uid.
955 	 */
956 	cred1 = p->p_cred;
957 	uid = kauth_cred_getuid(cred1);
958 	(void)chgproccnt(uid, -1);
959 
960 	/*
961 	 * Release substructures.
962 	 */
963 
964 	limfree(p->p_limit);
965 	pstatsfree(p->p_stats);
966 	kauth_cred_free(cred1);
967 	kauth_cred_free(cred2);
968 
969 	/*
970 	 * Release reference to text vnode
971 	 */
972 	if (p->p_textvp)
973 		vrele(p->p_textvp);
974 
975 	mutex_destroy(&p->p_auxlock);
976 	mutex_obj_free(p->p_lock);
977 	mutex_destroy(&p->p_stmutex);
978 	cv_destroy(&p->p_waitcv);
979 	cv_destroy(&p->p_lwpcv);
980 	rw_destroy(&p->p_reflock);
981 
982 	proc_free_mem(p);
983 }
984 
985 /*
986  * make process 'parent' the new parent of process 'child'.
987  *
988  * Must be called with proc_lock held.
989  */
990 void
991 proc_reparent(struct proc *child, struct proc *parent)
992 {
993 
994 	KASSERT(mutex_owned(proc_lock));
995 
996 	if (child->p_pptr == parent)
997 		return;
998 
999 	if (child->p_stat == SZOMB ||
1000 	    (child->p_stat == SSTOP && !child->p_waited)) {
1001 		child->p_pptr->p_nstopchild--;
1002 		parent->p_nstopchild++;
1003 	}
1004 	if (parent == initproc)
1005 		child->p_exitsig = SIGCHLD;
1006 
1007 	LIST_REMOVE(child, p_sibling);
1008 	LIST_INSERT_HEAD(&parent->p_children, child, p_sibling);
1009 	child->p_pptr = parent;
1010 	child->p_ppid = parent->p_pid;
1011 }
1012