xref: /netbsd-src/sys/kern/kern_exit.c (revision 2e31951ce35b19b425cc71066afbe82e73ab68f4)
1 /*	$NetBSD: kern_exit.c,v 1.216 2008/11/14 23:33:45 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 1.216 2008/11/14 23:33:45 ad 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/malloc.h>
91 #include <sys/pool.h>
92 #include <sys/uidinfo.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/sa.h>
103 #include <sys/savar.h>
104 #include <sys/mount.h>
105 #include <sys/syscallargs.h>
106 #include <sys/kauth.h>
107 #include <sys/sleepq.h>
108 #include <sys/lockdebug.h>
109 #include <sys/ktrace.h>
110 #include <sys/cpu.h>
111 #include <sys/lwpctl.h>
112 #include <sys/atomic.h>
113 
114 #include <uvm/uvm_extern.h>
115 
116 #define DEBUG_EXIT
117 
118 #ifdef DEBUG_EXIT
119 int debug_exit = 0;
120 #define DPRINTF(x) if (debug_exit) printf x
121 #else
122 #define DPRINTF(x)
123 #endif
124 
125 static int find_stopped_child(struct proc *, pid_t, int, struct proc **, int *);
126 static void proc_free(struct proc *, struct rusage *);
127 
128 /*
129  * Fill in the appropriate signal information, and signal the parent.
130  */
131 static void
132 exit_psignal(struct proc *p, struct proc *pp, ksiginfo_t *ksi)
133 {
134 
135 	KSI_INIT(ksi);
136 	if ((ksi->ksi_signo = P_EXITSIG(p)) == SIGCHLD) {
137 		if (WIFSIGNALED(p->p_xstat)) {
138 			if (WCOREDUMP(p->p_xstat))
139 				ksi->ksi_code = CLD_DUMPED;
140 			else
141 				ksi->ksi_code = CLD_KILLED;
142 		} else {
143 			ksi->ksi_code = CLD_EXITED;
144 		}
145 	}
146 	/*
147 	 * We fill those in, even for non-SIGCHLD.
148 	 * It's safe to access p->p_cred unlocked here.
149 	 */
150 	ksi->ksi_pid = p->p_pid;
151 	ksi->ksi_uid = kauth_cred_geteuid(p->p_cred);
152 	ksi->ksi_status = p->p_xstat;
153 	/* XXX: is this still valid? */
154 	ksi->ksi_utime = p->p_stats->p_ru.ru_utime.tv_sec;
155 	ksi->ksi_stime = p->p_stats->p_ru.ru_stime.tv_sec;
156 }
157 
158 /*
159  * exit --
160  *	Death of process.
161  */
162 int
163 sys_exit(struct lwp *l, const struct sys_exit_args *uap, register_t *retval)
164 {
165 	/* {
166 		syscallarg(int)	rval;
167 	} */
168 	struct proc *p = l->l_proc;
169 
170 	/* Don't call exit1() multiple times in the same process. */
171 	mutex_enter(p->p_lock);
172 	if (p->p_sflag & PS_WEXIT) {
173 		mutex_exit(p->p_lock);
174 		lwp_exit(l);
175 	}
176 
177 	/* exit1() will release the mutex. */
178 	exit1(l, W_EXITCODE(SCARG(uap, rval), 0));
179 	/* NOTREACHED */
180 	return (0);
181 }
182 
183 /*
184  * Exit: deallocate address space and other resources, change proc state
185  * to zombie, and unlink proc from allproc and parent's lists.  Save exit
186  * status and rusage for wait().  Check for child processes and orphan them.
187  *
188  * Must be called with p->p_lock held.  Does not return.
189  */
190 void
191 exit1(struct lwp *l, int rv)
192 {
193 	struct proc	*p, *q, *nq;
194 	struct pgrp	*pgrp;
195 	ksiginfo_t	ksi;
196 	ksiginfoq_t	kq;
197 	int		wakeinit, sa;
198 
199 	p = l->l_proc;
200 
201 	KASSERT(mutex_owned(p->p_lock));
202 
203 	if (__predict_false(p == initproc))
204 		panic("init died (signal %d, exit %d)",
205 		    WTERMSIG(rv), WEXITSTATUS(rv));
206 
207 	/*
208 	 * Disable scheduler activation upcalls.  We're trying to get out of
209 	 * here.
210 	 */
211 	sa = 0;
212 #ifdef KERN_SA
213 	if ((p->p_sa != NULL)) {
214 		l->l_pflag |= LP_SA_NOBLOCK;
215 		sa = 1;
216 	}
217 #endif
218 
219 	p->p_sflag |= PS_WEXIT;
220 
221 	/*
222 	 * Force all other LWPs to exit before we do.  Only then can we
223 	 * begin to tear down the rest of the process state.
224 	 */
225 	if (sa || p->p_nlwps > 1)
226 		exit_lwps(l);
227 
228 	ksiginfo_queue_init(&kq);
229 
230 	/*
231 	 * If we have been asked to stop on exit, do so now.
232 	 */
233 	if (__predict_false(p->p_sflag & PS_STOPEXIT)) {
234 		KERNEL_UNLOCK_ALL(l, &l->l_biglocks);
235 		sigclearall(p, &contsigmask, &kq);
236 		p->p_waited = 0;
237 		membar_producer();
238 		p->p_stat = SSTOP;
239 		lwp_lock(l);
240 		p->p_nrlwps--;
241 		l->l_stat = LSSTOP;
242 		mutex_exit(p->p_lock);
243 		mi_switch(l);
244 		KERNEL_LOCK(l->l_biglocks, l);
245 		mutex_enter(p->p_lock);
246 	}
247 
248 	/*
249 	 * Bin any remaining signals and mark the process as dying so it will
250 	 * not be found for, e.g. signals.
251 	 */
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 	/* Destroy any lwpctl info. */
259 	if (p->p_lwpctl != NULL)
260 		lwp_ctl_exit();
261 
262 	/*
263 	 * Drain all remaining references that procfs, ptrace and others may
264 	 * have on the process.
265 	 */
266 	rw_enter(&p->p_reflock, RW_WRITER);
267 
268 	DPRINTF(("exit1: %d.%d exiting.\n", p->p_pid, l->l_lid));
269 
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 	fd_free();
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 	 * Stop profiling.
323 	 */
324 	if (__predict_false((p->p_stflag & PST_PROFIL) != 0)) {
325 		mutex_spin_enter(&p->p_stmutex);
326 		stopprofclock(p);
327 		mutex_spin_exit(&p->p_stmutex);
328 	}
329 
330 	/*
331 	 * If parent is waiting for us to exit or exec, PL_PPWAIT is set; we
332 	 * wake up the parent early to avoid deadlock.  We can do this once
333 	 * the VM resources are released.
334 	 */
335 	mutex_enter(proc_lock);
336 	if (p->p_lflag & PL_PPWAIT) {
337 		p->p_lflag &= ~PL_PPWAIT;
338 		cv_broadcast(&p->p_pptr->p_waitcv);
339 	}
340 
341 	if (SESS_LEADER(p)) {
342 		struct vnode *vprele = NULL, *vprevoke = NULL;
343 		struct session *sp = p->p_session;
344 		struct tty *tp;
345 
346 		if (sp->s_ttyvp) {
347 			/*
348 			 * Controlling process.
349 			 * Signal foreground pgrp,
350 			 * drain controlling terminal
351 			 * and revoke access to controlling terminal.
352 			 */
353 			tp = sp->s_ttyp;
354 			mutex_spin_enter(&tty_lock);
355 			if (tp->t_session == sp) {
356 				/* we can't guarantee the revoke will do this */
357 				pgrp = tp->t_pgrp;
358 				tp->t_pgrp = NULL;
359 				tp->t_session = NULL;
360 				mutex_spin_exit(&tty_lock);
361 				if (pgrp != NULL) {
362 					pgsignal(pgrp, SIGHUP, 1);
363 				}
364 				mutex_exit(proc_lock);
365 				(void) ttywait(tp);
366 				mutex_enter(proc_lock);
367 
368 				/* The tty could have been revoked. */
369 				vprevoke = sp->s_ttyvp;
370 			} else
371 				mutex_spin_exit(&tty_lock);
372 			vprele = sp->s_ttyvp;
373 			sp->s_ttyvp = NULL;
374 			/*
375 			 * s_ttyp is not zero'd; we use this to indicate
376 			 * that the session once had a controlling terminal.
377 			 * (for logging and informational purposes)
378 			 */
379 		}
380 		sp->s_leader = NULL;
381 
382 		if (vprevoke != NULL || vprele != NULL) {
383 			if (vprevoke != NULL) {
384 				SESSRELE(sp);
385 				mutex_exit(proc_lock);
386 				VOP_REVOKE(vprevoke, REVOKEALL);
387 			} else
388 				mutex_exit(proc_lock);
389 			if (vprele != NULL)
390 				vrele(vprele);
391 			mutex_enter(proc_lock);
392 		}
393 	}
394 	fixjobc(p, p->p_pgrp, 0);
395 
396 	/*
397 	 * Finalize the last LWP's specificdata, as well as the
398 	 * specificdata for the proc itself.
399 	 */
400 	lwp_finispecific(l);
401 	proc_finispecific(p);
402 
403 	/*
404 	 * Notify interested parties of our demise.
405 	 */
406 	KNOTE(&p->p_klist, NOTE_EXIT);
407 
408 #if PERFCTRS
409 	/*
410 	 * Save final PMC information in parent process & clean up.
411 	 */
412 	if (PMC_ENABLED(p)) {
413 		pmc_save_context(p);
414 		pmc_accumulate(p->p_pptr, p);
415 		pmc_process_exit(p);
416 	}
417 #endif
418 
419 	/*
420 	 * Reset p_opptr pointer of all former children which got
421 	 * traced by another process and were reparented. We reset
422 	 * it to NULL here; the trace detach code then reparents
423 	 * the child to initproc. We only check allproc list, since
424 	 * eventual former children on zombproc list won't reference
425 	 * p_opptr anymore.
426 	 */
427 	if (__predict_false(p->p_slflag & PSL_CHTRACED)) {
428 		PROCLIST_FOREACH(q, &allproc) {
429 			if ((q->p_flag & PK_MARKER) != 0)
430 				continue;
431 			if (q->p_opptr == p)
432 				q->p_opptr = NULL;
433 		}
434 	}
435 
436 	/*
437 	 * Give orphaned children to init(8).
438 	 */
439 	q = LIST_FIRST(&p->p_children);
440 	wakeinit = (q != NULL);
441 	for (; q != NULL; q = nq) {
442 		nq = LIST_NEXT(q, p_sibling);
443 
444 		/*
445 		 * Traced processes are killed since their existence
446 		 * means someone is screwing up. Since we reset the
447 		 * trace flags, the logic in sys_wait4() would not be
448 		 * triggered to reparent the process to its
449 		 * original parent, so we must do this here.
450 		 */
451 		if (__predict_false(q->p_slflag & PSL_TRACED)) {
452 			mutex_enter(p->p_lock);
453 			q->p_slflag &= ~(PSL_TRACED|PSL_FSTRACE|PSL_SYSCALL);
454 			mutex_exit(p->p_lock);
455 			if (q->p_opptr != q->p_pptr) {
456 				struct proc *t = q->p_opptr;
457 				proc_reparent(q, t ? t : initproc);
458 				q->p_opptr = NULL;
459 			} else
460 				proc_reparent(q, initproc);
461 			killproc(q, "orphaned traced process");
462 		} else
463 			proc_reparent(q, initproc);
464 	}
465 
466 	/*
467 	 * Move proc from allproc to zombproc, it's now nearly ready to be
468 	 * collected by parent.
469 	 */
470 	LIST_REMOVE(l, l_list);
471 	LIST_REMOVE(p, p_list);
472 	LIST_INSERT_HEAD(&zombproc, p, p_list);
473 
474 	/*
475 	 * Mark the process as dead.  We must do this before we signal
476 	 * the parent.
477 	 */
478 	p->p_stat = SDEAD;
479 
480 	/* Put in front of parent's sibling list for parent to collect it */
481 	q = p->p_pptr;
482 	q->p_nstopchild++;
483 	if (LIST_FIRST(&q->p_children) != p) {
484 		/* Put child where it can be found quickly */
485 		LIST_REMOVE(p, p_sibling);
486 		LIST_INSERT_HEAD(&q->p_children, p, p_sibling);
487 	}
488 
489 	/*
490 	 * Notify parent that we're gone.  If parent has the P_NOCLDWAIT
491 	 * flag set, notify init instead (and hope it will handle
492 	 * this situation).
493 	 */
494 	if (q->p_flag & (PK_NOCLDWAIT|PK_CLDSIGIGN)) {
495 		proc_reparent(p, initproc);
496 		wakeinit = 1;
497 
498 		/*
499 		 * If this was the last child of our parent, notify
500 		 * parent, so in case he was wait(2)ing, he will
501 		 * continue.
502 		 */
503 		if (LIST_FIRST(&q->p_children) == NULL)
504 			cv_broadcast(&q->p_waitcv);
505 	}
506 
507 	/* Reload parent pointer, since p may have been reparented above */
508 	q = p->p_pptr;
509 
510 	if (__predict_false((p->p_slflag & PSL_FSTRACE) == 0 && p->p_exitsig != 0)) {
511 		exit_psignal(p, q, &ksi);
512 		kpsignal(q, &ksi, NULL);
513 	}
514 
515 	/* Calculate the final rusage info.  */
516 	calcru(p, &p->p_stats->p_ru.ru_utime, &p->p_stats->p_ru.ru_stime,
517 	    NULL, NULL);
518 
519 	if (wakeinit)
520 		cv_broadcast(&initproc->p_waitcv);
521 
522 	callout_destroy(&l->l_timeout_ch);
523 
524 	/*
525 	 * Remaining lwp resources will be freed in lwp_exit2() once we've
526 	 * switch to idle context; at that point, we will be marked as a
527 	 * full blown zombie.
528 	 */
529 	mutex_enter(p->p_lock);
530 	lwp_drainrefs(l);
531 	lwp_lock(l);
532 	l->l_prflag &= ~LPR_DETACHED;
533 	l->l_stat = LSZOMB;
534 	lwp_unlock(l);
535 	KASSERT(curlwp == l);
536 	KASSERT(p->p_nrlwps == 1);
537 	KASSERT(p->p_nlwps == 1);
538 	p->p_stat = SZOMB;
539 	p->p_nrlwps--;
540 	p->p_nzlwps++;
541 	p->p_ndlwps = 0;
542 	mutex_exit(p->p_lock);
543 
544 	/*
545 	 * Signal the parent to collect us, and drop the proclist lock.
546 	 * Drop debugger/procfs lock; no new references can be gained.
547 	 */
548 	cv_broadcast(&p->p_pptr->p_waitcv);
549 	mutex_exit(proc_lock);
550 	rw_exit(&p->p_reflock);
551 
552 	/* Verify that we hold no locks other than the kernel lock. */
553 	LOCKDEBUG_BARRIER(&kernel_lock, 0);
554 
555 	/*
556 	 * NOTE: WE ARE NO LONGER ALLOWED TO SLEEP!
557 	 */
558 
559 	/*
560 	 * Give machine-dependent code a chance to free any MD LWP
561 	 * resources.  This must be done before uvm_lwp_exit(), in
562 	 * case these resources are in the PCB.
563 	 */
564 #ifndef __NO_CPU_LWP_FREE
565 	cpu_lwp_free(l, 1);
566 #endif
567 	pmap_deactivate(l);
568 
569 	/* This process no longer needs to hold the kernel lock. */
570 #ifdef notyet
571 	/* XXXSMP hold in lwp_userret() */
572 	KERNEL_UNLOCK_LAST(l);
573 #else
574 	KERNEL_UNLOCK_ALL(l, NULL);
575 #endif
576 
577 	lwp_exit_switchaway(l);
578 }
579 
580 void
581 exit_lwps(struct lwp *l)
582 {
583 	struct proc *p;
584 	struct lwp *l2;
585 	int error;
586 	lwpid_t waited;
587 	int nlocks;
588 
589 	KERNEL_UNLOCK_ALL(l, &nlocks);
590 
591 	p = l->l_proc;
592 	KASSERT(mutex_owned(p->p_lock));
593 
594 #ifdef KERN_SA
595 	if (p->p_sa != NULL) {
596 		struct sadata_vp *vp;
597 		SLIST_FOREACH(vp, &p->p_sa->sa_vps, savp_next) {
598 			/*
599 			 * Make SA-cached LWPs normal process interruptable
600 			 * so that the exit code can wake them. Locking
601 			 * savp_mutex locks all the lwps on this vp that
602 			 * we need to adjust.
603 			 */
604 			mutex_enter(&vp->savp_mutex);
605 			DPRINTF(("exit_lwps: Making cached LWPs of %d on "
606 			    "VP %d interruptable: ", p->p_pid, vp->savp_id));
607 			TAILQ_FOREACH(l2, &vp->savp_lwpcache, l_sleepchain) {
608 				l2->l_flag |= LW_SINTR;
609 				DPRINTF(("%d ", l2->l_lid));
610 			}
611 			DPRINTF(("\n"));
612 
613 			DPRINTF(("exit_lwps: Making unblocking LWPs of %d on "
614 			    "VP %d interruptable: ", p->p_pid, vp->savp_id));
615 			TAILQ_FOREACH(l2, &vp->savp_woken, l_sleepchain) {
616 				vp->savp_woken_count--;
617 				l2->l_flag |= LW_SINTR;
618 				DPRINTF(("%d ", l2->l_lid));
619 			}
620 			DPRINTF(("\n"));
621 			mutex_exit(&vp->savp_mutex);
622 		}
623 	}
624 #endif
625 
626  retry:
627 	/*
628 	 * Interrupt LWPs in interruptable sleep, unsuspend suspended
629 	 * LWPs and then wait for everyone else to finish.
630 	 */
631 	LIST_FOREACH(l2, &p->p_lwps, l_sibling) {
632 		if (l2 == l)
633 			continue;
634 		lwp_lock(l2);
635 		l2->l_flag &= ~LW_SA;
636 		l2->l_flag |= LW_WEXIT;
637 		if ((l2->l_stat == LSSLEEP && (l2->l_flag & LW_SINTR)) ||
638 		    l2->l_stat == LSSUSPENDED || l2->l_stat == LSSTOP) {
639 		    	/* setrunnable() will release the lock. */
640 			setrunnable(l2);
641 			DPRINTF(("exit_lwps: Made %d.%d runnable\n",
642 			    p->p_pid, l2->l_lid));
643 			continue;
644 		}
645 		lwp_unlock(l2);
646 	}
647 	while (p->p_nlwps > 1) {
648 		DPRINTF(("exit_lwps: waiting for %d LWPs (%d zombies)\n",
649 		    p->p_nlwps, p->p_nzlwps));
650 		error = lwp_wait1(l, 0, &waited, LWPWAIT_EXITCONTROL);
651 		if (p->p_nlwps == 1)
652 			break;
653 		if (error == EDEADLK) {
654 			/*
655 			 * LWPs can get suspended/slept behind us.
656 			 * (eg. sa_setwoken)
657 			 * kick them again and retry.
658 			 */
659 			goto retry;
660 		}
661 		if (error)
662 			panic("exit_lwps: lwp_wait1 failed with error %d",
663 			    error);
664 		DPRINTF(("exit_lwps: Got LWP %d from lwp_wait1()\n", waited));
665 	}
666 
667 	KERNEL_LOCK(nlocks, l);
668 	KASSERT(p->p_nlwps == 1);
669 }
670 
671 int
672 do_sys_wait(struct lwp *l, int *pid, int *status, int options,
673     struct rusage *ru, int *was_zombie)
674 {
675 	struct proc	*child;
676 	int		error;
677 
678 	mutex_enter(proc_lock);
679 	error = find_stopped_child(l->l_proc, *pid, options, &child, status);
680 
681 	if (child == NULL) {
682 		mutex_exit(proc_lock);
683 		*pid = 0;
684 		return error;
685 	}
686 
687 	*pid = child->p_pid;
688 
689 	if (child->p_stat == SZOMB) {
690 		/* proc_free() will release the proc_lock. */
691 		*was_zombie = 1;
692 		if (options & WNOWAIT)
693 			mutex_exit(proc_lock);
694 		else {
695 			proc_free(child, ru);
696 		}
697 	} else {
698 		/* Child state must have been SSTOP. */
699 		*was_zombie = 0;
700 		mutex_exit(proc_lock);
701 		*status = W_STOPCODE(*status);
702 	}
703 
704 	return 0;
705 }
706 
707 int
708 sys_wait4(struct lwp *l, const struct sys_wait4_args *uap, register_t *retval)
709 {
710 	/* {
711 		syscallarg(int)			pid;
712 		syscallarg(int *)		status;
713 		syscallarg(int)			options;
714 		syscallarg(struct rusage *)	rusage;
715 	} */
716 	int		status, error;
717 	int		was_zombie;
718 	struct rusage	ru;
719 	int pid = SCARG(uap, pid);
720 
721 	error = do_sys_wait(l, &pid, &status, SCARG(uap, options),
722 	    SCARG(uap, rusage) != NULL ? &ru : NULL, &was_zombie);
723 
724 	retval[0] = pid;
725 	if (pid == 0)
726 		return error;
727 
728 	if (SCARG(uap, rusage))
729 		error = copyout(&ru, SCARG(uap, rusage), sizeof(ru));
730 
731 	if (error == 0 && SCARG(uap, status))
732 		error = copyout(&status, SCARG(uap, status), sizeof(status));
733 
734 	return error;
735 }
736 
737 /*
738  * Scan list of child processes for a child process that has stopped or
739  * exited.  Used by sys_wait4 and 'compat' equivalents.
740  *
741  * Must be called with the proc_lock held, and may release while waiting.
742  */
743 static int
744 find_stopped_child(struct proc *parent, pid_t pid, int options,
745 		   struct proc **child_p, int *status_p)
746 {
747 	struct proc *child, *dead;
748 	int error;
749 
750 	KASSERT(mutex_owned(proc_lock));
751 
752 	if (options & ~(WUNTRACED|WNOHANG|WALTSIG|WALLSIG)
753 	    && !(options & WOPTSCHECKED)) {
754 		*child_p = NULL;
755 		return EINVAL;
756 	}
757 
758 	if (pid == 0 && !(options & WOPTSCHECKED))
759 		pid = -parent->p_pgid;
760 
761 	for (;;) {
762 		error = ECHILD;
763 		dead = NULL;
764 
765 		LIST_FOREACH(child, &parent->p_children, p_sibling) {
766 			if (pid >= 0) {
767 				if (child->p_pid != pid) {
768 					child = p_find(pid, PFIND_ZOMBIE |
769 					    PFIND_LOCKED);
770 					if (child == NULL ||
771 					    child->p_pptr != parent) {
772 						child = NULL;
773 						break;
774 					}
775 				}
776 			} else if (pid != WAIT_ANY && child->p_pgid != -pid) {
777 				/* Child not in correct pgrp */
778 				continue;
779 			}
780 
781 			/*
782 			 * Wait for processes with p_exitsig != SIGCHLD
783 			 * processes only if WALTSIG is set; wait for
784 			 * processes with p_exitsig == SIGCHLD only
785 			 * if WALTSIG is clear.
786 			 */
787 			if (((options & WALLSIG) == 0) &&
788 			    (options & WALTSIG ? child->p_exitsig == SIGCHLD
789 						: P_EXITSIG(child) != SIGCHLD)){
790 				if (child->p_pid == pid) {
791 					child = NULL;
792 					break;
793 				}
794 				continue;
795 			}
796 
797 			error = 0;
798 			if ((options & WNOZOMBIE) == 0) {
799 				if (child->p_stat == SZOMB)
800 					break;
801 				if (child->p_stat == SDEAD) {
802 					/*
803 					 * We may occasionally arrive here
804 					 * after receiving a signal, but
805 					 * immediatley before the child
806 					 * process is zombified.  The wait
807 					 * will be short, so avoid returning
808 					 * to userspace.
809 					 */
810 					dead = child;
811 				}
812 			}
813 
814 			if (child->p_stat == SSTOP &&
815 			    child->p_waited == 0 &&
816 			    (child->p_slflag & PSL_TRACED ||
817 			    options & WUNTRACED)) {
818 				if ((options & WNOWAIT) == 0) {
819 					child->p_waited = 1;
820 					parent->p_nstopchild--;
821 				}
822 				break;
823 			}
824 			if (parent->p_nstopchild == 0 || child->p_pid == pid) {
825 				child = NULL;
826 				break;
827 			}
828 		}
829 
830 		if (child != NULL || error != 0 ||
831 		    ((options & WNOHANG) != 0 && dead == NULL)) {
832 		    	if (child != NULL) {
833 			    	*status_p = child->p_xstat;
834 			}
835 			*child_p = child;
836 			return error;
837 		}
838 
839 		/*
840 		 * Wait for another child process to stop.
841 		 */
842 		error = cv_wait_sig(&parent->p_waitcv, proc_lock);
843 
844 		if (error != 0) {
845 			*child_p = NULL;
846 			return error;
847 		}
848 	}
849 }
850 
851 /*
852  * Free a process after parent has taken all the state info.  Must be called
853  * with the proclist lock held, and will release before returning.
854  *
855  * *ru is returned to the caller, and must be freed by the caller.
856  */
857 static void
858 proc_free(struct proc *p, struct rusage *ru)
859 {
860 	struct proc *parent;
861 	struct lwp *l;
862 	ksiginfo_t ksi;
863 	kauth_cred_t cred1, cred2;
864 	uid_t uid;
865 
866 	KASSERT(mutex_owned(proc_lock));
867 	KASSERT(p->p_nlwps == 1);
868 	KASSERT(p->p_nzlwps == 1);
869 	KASSERT(p->p_nrlwps == 0);
870 	KASSERT(p->p_stat == SZOMB);
871 
872 	/*
873 	 * If we got the child via ptrace(2) or procfs, and
874 	 * the parent is different (meaning the process was
875 	 * attached, rather than run as a child), then we need
876 	 * to give it back to the old parent, and send the
877 	 * parent the exit signal.  The rest of the cleanup
878 	 * will be done when the old parent waits on the child.
879 	 */
880 	if ((p->p_slflag & PSL_TRACED) != 0) {
881 		parent = p->p_pptr;
882 		if (p->p_opptr != parent){
883 			mutex_enter(p->p_lock);
884 			p->p_slflag &= ~(PSL_TRACED|PSL_FSTRACE|PSL_SYSCALL);
885 			mutex_exit(p->p_lock);
886 			parent = p->p_opptr;
887 			if (parent == NULL)
888 				parent = initproc;
889 			proc_reparent(p, parent);
890 			p->p_opptr = NULL;
891 			if (p->p_exitsig != 0) {
892 				exit_psignal(p, parent, &ksi);
893 				kpsignal(parent, &ksi, NULL);
894 			}
895 			cv_broadcast(&parent->p_waitcv);
896 			mutex_exit(proc_lock);
897 			return;
898 		}
899 	}
900 
901 	/*
902 	 * Finally finished with old proc entry.  Unlink it from its process
903 	 * group.
904 	 */
905 	leavepgrp(p);
906 
907 	parent = p->p_pptr;
908 	sched_proc_exit(parent, p);
909 
910 	/*
911 	 * Add child times of exiting process onto its own times.
912 	 * This cannot be done any earlier else it might get done twice.
913 	 */
914 	l = LIST_FIRST(&p->p_lwps);
915 	p->p_stats->p_ru.ru_nvcsw += (l->l_ncsw - l->l_nivcsw);
916 	p->p_stats->p_ru.ru_nivcsw += l->l_nivcsw;
917 	ruadd(&p->p_stats->p_ru, &l->l_ru);
918 	ruadd(&p->p_stats->p_ru, &p->p_stats->p_cru);
919 	ruadd(&parent->p_stats->p_cru, &p->p_stats->p_ru);
920 	if (ru != NULL)
921 		*ru = p->p_stats->p_ru;
922 	p->p_xstat = 0;
923 
924 	/* Release any SA state. */
925 #ifdef KERN_SA
926 	if (p->p_sa)
927 		sa_release(p);
928 #endif
929 
930 	/*
931 	 * At this point we are going to start freeing the final resources.
932 	 * If anyone tries to access the proc structure after here they will
933 	 * get a shock - bits are missing.  Attempt to make it hard!  We
934 	 * don't bother with any further locking past this point.
935 	 */
936 	p->p_stat = SIDL;		/* not even a zombie any more */
937 	LIST_REMOVE(p, p_list);	/* off zombproc */
938 	parent = p->p_pptr;
939 	p->p_pptr->p_nstopchild--;
940 	LIST_REMOVE(p, p_sibling);
941 
942 	/*
943 	 * Let pid be reallocated.
944 	 */
945 	proc_free_pid(p);
946 	mutex_exit(proc_lock);
947 
948 	/*
949 	 * Delay release until after lwp_free.
950 	 */
951 	cred2 = l->l_cred;
952 
953 	/*
954 	 * Free the last LWP's resources.
955 	 *
956 	 * lwp_free ensures the LWP is no longer running on another CPU.
957 	 */
958 	lwp_free(l, false, true);
959 
960 	/*
961 	 * Now no one except us can reach the process p.
962 	 */
963 
964 	/*
965 	 * Decrement the count of procs running with this uid.
966 	 */
967 	cred1 = p->p_cred;
968 	uid = kauth_cred_getuid(cred1);
969 	(void)chgproccnt(uid, -1);
970 
971 	/*
972 	 * Release substructures.
973 	 */
974 
975 	limfree(p->p_limit);
976 	pstatsfree(p->p_stats);
977 	kauth_cred_free(cred1);
978 	kauth_cred_free(cred2);
979 
980 	/*
981 	 * Release reference to text vnode
982 	 */
983 	if (p->p_textvp)
984 		vrele(p->p_textvp);
985 
986 	mutex_destroy(&p->p_auxlock);
987 	mutex_obj_free(p->p_lock);
988 	mutex_destroy(&p->p_stmutex);
989 	cv_destroy(&p->p_waitcv);
990 	cv_destroy(&p->p_lwpcv);
991 	rw_destroy(&p->p_reflock);
992 
993 	proc_free_mem(p);
994 }
995 
996 /*
997  * make process 'parent' the new parent of process 'child'.
998  *
999  * Must be called with proc_lock held.
1000  */
1001 void
1002 proc_reparent(struct proc *child, struct proc *parent)
1003 {
1004 
1005 	KASSERT(mutex_owned(proc_lock));
1006 
1007 	if (child->p_pptr == parent)
1008 		return;
1009 
1010 	if (child->p_stat == SZOMB ||
1011 	    (child->p_stat == SSTOP && !child->p_waited)) {
1012 		child->p_pptr->p_nstopchild--;
1013 		parent->p_nstopchild++;
1014 	}
1015 	if (parent == initproc)
1016 		child->p_exitsig = SIGCHLD;
1017 
1018 	LIST_REMOVE(child, p_sibling);
1019 	LIST_INSERT_HEAD(&parent->p_children, child, p_sibling);
1020 	child->p_pptr = parent;
1021 	child->p_ppid = parent->p_pid;
1022 }
1023