xref: /netbsd-src/sys/kern/kern_exit.c (revision 96230fab84e26a6435963032070e916a951a8b2e)
1 /*	$NetBSD: kern_exit.c,v 1.213 2008/10/15 06:51:20 wrstuden 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 1.213 2008/10/15 06:51:20 wrstuden 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/aio.h>
79 #include <sys/systm.h>
80 #include <sys/ioctl.h>
81 #include <sys/tty.h>
82 #include <sys/time.h>
83 #include <sys/resource.h>
84 #include <sys/kernel.h>
85 #include <sys/proc.h>
86 #include <sys/buf.h>
87 #include <sys/wait.h>
88 #include <sys/file.h>
89 #include <sys/vnode.h>
90 #include <sys/syslog.h>
91 #include <sys/malloc.h>
92 #include <sys/pool.h>
93 #include <sys/uidinfo.h>
94 #if defined(PERFCTRS)
95 #include <sys/pmc.h>
96 #endif
97 #include <sys/ptrace.h>
98 #include <sys/acct.h>
99 #include <sys/filedesc.h>
100 #include <sys/ras.h>
101 #include <sys/signalvar.h>
102 #include <sys/sched.h>
103 #include <sys/sa.h>
104 #include <sys/savar.h>
105 #include <sys/mount.h>
106 #include <sys/syscallargs.h>
107 #include <sys/kauth.h>
108 #include <sys/sleepq.h>
109 #include <sys/lockdebug.h>
110 #include <sys/ktrace.h>
111 #include <sys/cpu.h>
112 #include <sys/lwpctl.h>
113 #include <sys/atomic.h>
114 
115 #include <uvm/uvm_extern.h>
116 
117 #define DEBUG_EXIT
118 
119 #ifdef DEBUG_EXIT
120 int debug_exit = 0;
121 #define DPRINTF(x) if (debug_exit) printf x
122 #else
123 #define DPRINTF(x)
124 #endif
125 
126 static int find_stopped_child(struct proc *, pid_t, int, struct proc **, int *);
127 static void proc_free(struct proc *, struct rusage *);
128 
129 /*
130  * Fill in the appropriate signal information, and signal the parent.
131  */
132 static void
133 exit_psignal(struct proc *p, struct proc *pp, ksiginfo_t *ksi)
134 {
135 
136 	KSI_INIT(ksi);
137 	if ((ksi->ksi_signo = P_EXITSIG(p)) == SIGCHLD) {
138 		if (WIFSIGNALED(p->p_xstat)) {
139 			if (WCOREDUMP(p->p_xstat))
140 				ksi->ksi_code = CLD_DUMPED;
141 			else
142 				ksi->ksi_code = CLD_KILLED;
143 		} else {
144 			ksi->ksi_code = CLD_EXITED;
145 		}
146 	}
147 	/*
148 	 * We fill those in, even for non-SIGCHLD.
149 	 * It's safe to access p->p_cred unlocked here.
150 	 */
151 	ksi->ksi_pid = p->p_pid;
152 	ksi->ksi_uid = kauth_cred_geteuid(p->p_cred);
153 	ksi->ksi_status = p->p_xstat;
154 	/* XXX: is this still valid? */
155 	ksi->ksi_utime = p->p_stats->p_ru.ru_utime.tv_sec;
156 	ksi->ksi_stime = p->p_stats->p_ru.ru_stime.tv_sec;
157 }
158 
159 /*
160  * exit --
161  *	Death of process.
162  */
163 int
164 sys_exit(struct lwp *l, const struct sys_exit_args *uap, register_t *retval)
165 {
166 	/* {
167 		syscallarg(int)	rval;
168 	} */
169 	struct proc *p = l->l_proc;
170 
171 	/* Don't call exit1() multiple times in the same process. */
172 	mutex_enter(p->p_lock);
173 	if (p->p_sflag & PS_WEXIT) {
174 		mutex_exit(p->p_lock);
175 		lwp_exit(l);
176 	}
177 
178 	/* exit1() will release the mutex. */
179 	exit1(l, W_EXITCODE(SCARG(uap, rval), 0));
180 	/* NOTREACHED */
181 	return (0);
182 }
183 
184 /*
185  * Exit: deallocate address space and other resources, change proc state
186  * to zombie, and unlink proc from allproc and parent's lists.  Save exit
187  * status and rusage for wait().  Check for child processes and orphan them.
188  *
189  * Must be called with p->p_lock held.  Does not return.
190  */
191 void
192 exit1(struct lwp *l, int rv)
193 {
194 	struct proc	*p, *q, *nq;
195 	struct pgrp	*pgrp;
196 	ksiginfo_t	ksi;
197 	ksiginfoq_t	kq;
198 	int		wakeinit, sa;
199 
200 	p = l->l_proc;
201 
202 	KASSERT(mutex_owned(p->p_lock));
203 
204 	if (__predict_false(p == initproc))
205 		panic("init died (signal %d, exit %d)",
206 		    WTERMSIG(rv), WEXITSTATUS(rv));
207 
208 	/*
209 	 * Disable scheduler activation upcalls.  We're trying to get out of
210 	 * here.
211 	 */
212 	sa = 0;
213 #ifdef KERN_SA
214 	if ((p->p_sa != NULL)) {
215 		l->l_pflag |= LP_SA_NOBLOCK;
216 		sa = 1;
217 	}
218 #endif
219 
220 	p->p_sflag |= PS_WEXIT;
221 
222 	/*
223 	 * Force all other LWPs to exit before we do.  Only then can we
224 	 * begin to tear down the rest of the process state.
225 	 */
226 	if (sa || p->p_nlwps > 1)
227 		exit_lwps(l);
228 
229 	ksiginfo_queue_init(&kq);
230 
231 	/*
232 	 * If we have been asked to stop on exit, do so now.
233 	 */
234 	if (__predict_false(p->p_sflag & PS_STOPEXIT)) {
235 		KERNEL_UNLOCK_ALL(l, &l->l_biglocks);
236 		sigclearall(p, &contsigmask, &kq);
237 		p->p_waited = 0;
238 		membar_producer();
239 		p->p_stat = SSTOP;
240 		lwp_lock(l);
241 		p->p_nrlwps--;
242 		l->l_stat = LSSTOP;
243 		mutex_exit(p->p_lock);
244 		mi_switch(l);
245 		KERNEL_LOCK(l->l_biglocks, l);
246 		mutex_enter(p->p_lock);
247 	}
248 
249 	/*
250 	 * Bin any remaining signals and mark the process as dying so it will
251 	 * not be found for, e.g. signals.
252 	 */
253 	sigfillset(&p->p_sigctx.ps_sigignore);
254 	sigclearall(p, NULL, &kq);
255 	p->p_stat = SDYING;
256 	mutex_exit(p->p_lock);
257 	ksiginfo_queue_drain(&kq);
258 
259 	/* Destroy any lwpctl info. */
260 	if (p->p_lwpctl != NULL)
261 		lwp_ctl_exit();
262 
263 	/* Destroy all AIO works */
264 	aio_exit(p, p->p_aio);
265 
266 	/*
267 	 * Drain all remaining references that procfs, ptrace and others may
268 	 * have on the process.
269 	 */
270 	rw_enter(&p->p_reflock, RW_WRITER);
271 
272 	DPRINTF(("exit1: %d.%d exiting.\n", p->p_pid, l->l_lid));
273 
274 	timers_free(p, TIMERS_ALL);
275 #if defined(__HAVE_RAS)
276 	ras_purgeall();
277 #endif
278 
279 	/*
280 	 * Close open files, release open-file table and free signal
281 	 * actions.  This may block!
282 	 */
283 	fd_free();
284 	cwdfree(p->p_cwdi);
285 	p->p_cwdi = NULL;
286 	doexithooks(p);
287 	sigactsfree(p->p_sigacts);
288 
289 	/*
290 	 * Write out accounting data.
291 	 */
292 	(void)acct_process(l);
293 
294 #ifdef KTRACE
295 	/*
296 	 * Release trace file.
297 	 */
298 	if (p->p_tracep != NULL) {
299 		mutex_enter(&ktrace_lock);
300 		ktrderef(p);
301 		mutex_exit(&ktrace_lock);
302 	}
303 #endif
304 
305 	/*
306 	 * If emulation has process exit hook, call it now.
307 	 * Set the exit status now so that the exit hook has
308 	 * an opportunity to tweak it (COMPAT_LINUX requires
309 	 * this for thread group emulation)
310 	 */
311 	p->p_xstat = rv;
312 	if (p->p_emul->e_proc_exit)
313 		(*p->p_emul->e_proc_exit)(p);
314 
315 	/*
316 	 * Free the VM resources we're still holding on to.
317 	 * We must do this from a valid thread because doing
318 	 * so may block. This frees vmspace, which we don't
319 	 * need anymore. The only remaining lwp is the one
320 	 * we run at this moment, nothing runs in userland
321 	 * anymore.
322 	 */
323 	uvm_proc_exit(p);
324 
325 	/*
326 	 * Stop profiling.
327 	 */
328 	if (__predict_false((p->p_stflag & PST_PROFIL) != 0)) {
329 		mutex_spin_enter(&p->p_stmutex);
330 		stopprofclock(p);
331 		mutex_spin_exit(&p->p_stmutex);
332 	}
333 
334 	/*
335 	 * If parent is waiting for us to exit or exec, P_PPWAIT is set; we
336 	 * wake up the parent early to avoid deadlock.  We can do this once
337 	 * the VM resources are released.
338 	 */
339 	mutex_enter(proc_lock);
340 	if (p->p_lflag & PL_PPWAIT) {
341 		p->p_lflag &= ~PL_PPWAIT;
342 		cv_broadcast(&p->p_pptr->p_waitcv);
343 	}
344 
345 	if (SESS_LEADER(p)) {
346 		struct vnode *vprele = NULL, *vprevoke = NULL;
347 		struct session *sp = p->p_session;
348 		struct tty *tp;
349 
350 		if (sp->s_ttyvp) {
351 			/*
352 			 * Controlling process.
353 			 * Signal foreground pgrp,
354 			 * drain controlling terminal
355 			 * and revoke access to controlling terminal.
356 			 */
357 			tp = sp->s_ttyp;
358 			mutex_spin_enter(&tty_lock);
359 			if (tp->t_session == sp) {
360 				/* we can't guarantee the revoke will do this */
361 				pgrp = tp->t_pgrp;
362 				tp->t_pgrp = NULL;
363 				tp->t_session = NULL;
364 				mutex_spin_exit(&tty_lock);
365 				if (pgrp != NULL) {
366 					pgsignal(pgrp, SIGHUP, 1);
367 				}
368 				mutex_exit(proc_lock);
369 				(void) ttywait(tp);
370 				mutex_enter(proc_lock);
371 
372 				/* The tty could have been revoked. */
373 				vprevoke = sp->s_ttyvp;
374 			} else
375 				mutex_spin_exit(&tty_lock);
376 			vprele = sp->s_ttyvp;
377 			sp->s_ttyvp = NULL;
378 			/*
379 			 * s_ttyp is not zero'd; we use this to indicate
380 			 * that the session once had a controlling terminal.
381 			 * (for logging and informational purposes)
382 			 */
383 		}
384 		sp->s_leader = NULL;
385 
386 		if (vprevoke != NULL || vprele != NULL) {
387 			if (vprevoke != NULL) {
388 				SESSRELE(sp);
389 				mutex_exit(proc_lock);
390 				VOP_REVOKE(vprevoke, REVOKEALL);
391 			} else
392 				mutex_exit(proc_lock);
393 			if (vprele != NULL)
394 				vrele(vprele);
395 			mutex_enter(proc_lock);
396 		}
397 	}
398 	fixjobc(p, p->p_pgrp, 0);
399 
400 	/*
401 	 * Finalize the last LWP's specificdata, as well as the
402 	 * specificdata for the proc itself.
403 	 */
404 	lwp_finispecific(l);
405 	proc_finispecific(p);
406 
407 	/*
408 	 * Notify interested parties of our demise.
409 	 */
410 	KNOTE(&p->p_klist, NOTE_EXIT);
411 
412 #if PERFCTRS
413 	/*
414 	 * Save final PMC information in parent process & clean up.
415 	 */
416 	if (PMC_ENABLED(p)) {
417 		pmc_save_context(p);
418 		pmc_accumulate(p->p_pptr, p);
419 		pmc_process_exit(p);
420 	}
421 #endif
422 
423 	/*
424 	 * Reset p_opptr pointer of all former children which got
425 	 * traced by another process and were reparented. We reset
426 	 * it to NULL here; the trace detach code then reparents
427 	 * the child to initproc. We only check allproc list, since
428 	 * eventual former children on zombproc list won't reference
429 	 * p_opptr anymore.
430 	 */
431 	if (__predict_false(p->p_slflag & PSL_CHTRACED)) {
432 		PROCLIST_FOREACH(q, &allproc) {
433 			if ((q->p_flag & PK_MARKER) != 0)
434 				continue;
435 			if (q->p_opptr == p)
436 				q->p_opptr = NULL;
437 		}
438 	}
439 
440 	/*
441 	 * Give orphaned children to init(8).
442 	 */
443 	q = LIST_FIRST(&p->p_children);
444 	wakeinit = (q != NULL);
445 	for (; q != NULL; q = nq) {
446 		nq = LIST_NEXT(q, p_sibling);
447 
448 		/*
449 		 * Traced processes are killed since their existence
450 		 * means someone is screwing up. Since we reset the
451 		 * trace flags, the logic in sys_wait4() would not be
452 		 * triggered to reparent the process to its
453 		 * original parent, so we must do this here.
454 		 */
455 		if (__predict_false(q->p_slflag & PSL_TRACED)) {
456 			mutex_enter(p->p_lock);
457 			q->p_slflag &= ~(PSL_TRACED|PSL_FSTRACE|PSL_SYSCALL);
458 			mutex_exit(p->p_lock);
459 			if (q->p_opptr != q->p_pptr) {
460 				struct proc *t = q->p_opptr;
461 				proc_reparent(q, t ? t : initproc);
462 				q->p_opptr = NULL;
463 			} else
464 				proc_reparent(q, initproc);
465 			killproc(q, "orphaned traced process");
466 		} else
467 			proc_reparent(q, initproc);
468 	}
469 
470 	/*
471 	 * Move proc from allproc to zombproc, it's now nearly ready to be
472 	 * collected by parent.
473 	 */
474 	LIST_REMOVE(l, l_list);
475 	LIST_REMOVE(p, p_list);
476 	LIST_INSERT_HEAD(&zombproc, p, p_list);
477 
478 	/*
479 	 * Mark the process as dead.  We must do this before we signal
480 	 * the parent.
481 	 */
482 	p->p_stat = SDEAD;
483 
484 	/* Put in front of parent's sibling list for parent to collect it */
485 	q = p->p_pptr;
486 	q->p_nstopchild++;
487 	if (LIST_FIRST(&q->p_children) != p) {
488 		/* Put child where it can be found quickly */
489 		LIST_REMOVE(p, p_sibling);
490 		LIST_INSERT_HEAD(&q->p_children, p, p_sibling);
491 	}
492 
493 	/*
494 	 * Notify parent that we're gone.  If parent has the P_NOCLDWAIT
495 	 * flag set, notify init instead (and hope it will handle
496 	 * this situation).
497 	 */
498 	if (q->p_flag & (PK_NOCLDWAIT|PK_CLDSIGIGN)) {
499 		proc_reparent(p, initproc);
500 		wakeinit = 1;
501 
502 		/*
503 		 * If this was the last child of our parent, notify
504 		 * parent, so in case he was wait(2)ing, he will
505 		 * continue.
506 		 */
507 		if (LIST_FIRST(&q->p_children) == NULL)
508 			cv_broadcast(&q->p_waitcv);
509 	}
510 
511 	/* Reload parent pointer, since p may have been reparented above */
512 	q = p->p_pptr;
513 
514 	if (__predict_false((p->p_slflag & PSL_FSTRACE) == 0 && p->p_exitsig != 0)) {
515 		exit_psignal(p, q, &ksi);
516 		kpsignal(q, &ksi, NULL);
517 	}
518 
519 	/* Calculate the final rusage info.  */
520 	calcru(p, &p->p_stats->p_ru.ru_utime, &p->p_stats->p_ru.ru_stime,
521 	    NULL, NULL);
522 
523 	if (wakeinit)
524 		cv_broadcast(&initproc->p_waitcv);
525 
526 	callout_destroy(&l->l_timeout_ch);
527 
528 	/*
529 	 * Remaining lwp resources will be freed in lwp_exit2() once we've
530 	 * switch to idle context; at that point, we will be marked as a
531 	 * full blown zombie.
532 	 */
533 	mutex_enter(p->p_lock);
534 	lwp_drainrefs(l);
535 	lwp_lock(l);
536 	l->l_prflag &= ~LPR_DETACHED;
537 	l->l_stat = LSZOMB;
538 	lwp_unlock(l);
539 	KASSERT(curlwp == l);
540 	KASSERT(p->p_nrlwps == 1);
541 	KASSERT(p->p_nlwps == 1);
542 	p->p_stat = SZOMB;
543 	p->p_nrlwps--;
544 	p->p_nzlwps++;
545 	p->p_ndlwps = 0;
546 	mutex_exit(p->p_lock);
547 
548 	/*
549 	 * Signal the parent to collect us, and drop the proclist lock.
550 	 * Drop debugger/procfs lock; no new references can be gained.
551 	 */
552 	cv_broadcast(&p->p_pptr->p_waitcv);
553 	mutex_exit(proc_lock);
554 	rw_exit(&p->p_reflock);
555 
556 	/* Verify that we hold no locks other than the kernel lock. */
557 	LOCKDEBUG_BARRIER(&kernel_lock, 0);
558 
559 	/*
560 	 * NOTE: WE ARE NO LONGER ALLOWED TO SLEEP!
561 	 */
562 
563 	/*
564 	 * Give machine-dependent code a chance to free any MD LWP
565 	 * resources.  This must be done before uvm_lwp_exit(), in
566 	 * case these resources are in the PCB.
567 	 */
568 #ifndef __NO_CPU_LWP_FREE
569 	cpu_lwp_free(l, 1);
570 #endif
571 	pmap_deactivate(l);
572 
573 	/* This process no longer needs to hold the kernel lock. */
574 #ifdef notyet
575 	/* XXXSMP hold in lwp_userret() */
576 	KERNEL_UNLOCK_LAST(l);
577 #else
578 	KERNEL_UNLOCK_ALL(l, NULL);
579 #endif
580 
581 	lwp_exit_switchaway(l);
582 }
583 
584 void
585 exit_lwps(struct lwp *l)
586 {
587 	struct proc *p;
588 	struct lwp *l2;
589 	int error;
590 	lwpid_t waited;
591 	int nlocks;
592 
593 	KERNEL_UNLOCK_ALL(l, &nlocks);
594 
595 	p = l->l_proc;
596 	KASSERT(mutex_owned(p->p_lock));
597 
598 #if 0
599 	if (p->p_sa != NULL) {
600 		SLIST_FOREACH(vp, &p->p_sa->sa_vps, savp_next) {
601 			/*
602 			 * Make SA-cached LWPs normal process runnable
603 			 * LWPs so that they'll also self-destruct.
604 			 */
605 			DPRINTF(("exit_lwps: Making cached LWPs of %d on "
606 			    "VP %d runnable: ", p->p_pid, vp->savp_id));
607 			while ((l2 = sa_getcachelwp(p, vp)) != 0) {
608 				lwp_lock(l2);
609 				l2->l_flag = (l2->l_flag & ~LW_SA) | LW_WEXIT;
610 				l2->l_priority = MAXPRI_USER; /* XXX WRS needs thought */
611 
612 				/* setrunnable() will release the mutex. */
613 				setrunnable(l2);
614 				DPRINTF(("%d ", l2->l_lid));
615 			}
616 			DPRINTF(("\n"));
617 
618 			/*
619 			 * Clear wokenq, the LWPs on the queue will
620 			 * run below. Workes as these threads are still
621 			 * on the p_lwps list (even though they are no longer
622 			 * counted).
623 			 */
624 			TAILQ_INIT(&vp->savp_woken);
625 		}
626 	}
627 #endif
628 
629  retry:
630 	/*
631 	 * Interrupt LWPs in interruptable sleep, unsuspend suspended
632 	 * LWPs and then wait for everyone else to finish.
633 	 */
634 	LIST_FOREACH(l2, &p->p_lwps, l_sibling) {
635 		if (l2 == l)
636 			continue;
637 		lwp_lock(l2);
638 		l2->l_flag &= ~LW_SA;
639 		l2->l_flag |= LW_WEXIT;
640 		if ((l2->l_stat == LSSLEEP && (l2->l_flag & LW_SINTR)) ||
641 		    l2->l_stat == LSSUSPENDED || l2->l_stat == LSSTOP) {
642 		    	/* setrunnable() will release the lock. */
643 			setrunnable(l2);
644 			DPRINTF(("exit_lwps: Made %d.%d runnable\n",
645 			    p->p_pid, l2->l_lid));
646 			continue;
647 		}
648 		lwp_unlock(l2);
649 	}
650 	while (p->p_nlwps > 1) {
651 		DPRINTF(("exit_lwps: waiting for %d LWPs (%d zombies)\n",
652 		    p->p_nlwps, p->p_nzlwps));
653 		error = lwp_wait1(l, 0, &waited, LWPWAIT_EXITCONTROL);
654 		if (p->p_nlwps == 1)
655 			break;
656 		if (error == EDEADLK) {
657 			/*
658 			 * LWPs can get suspended/slept behind us.
659 			 * (eg. sa_setwoken)
660 			 * kick them again and retry.
661 			 */
662 			goto retry;
663 		}
664 		if (error)
665 			panic("exit_lwps: lwp_wait1 failed with error %d",
666 			    error);
667 		DPRINTF(("exit_lwps: Got LWP %d from lwp_wait1()\n", waited));
668 	}
669 
670 	KERNEL_LOCK(nlocks, l);
671 	KASSERT(p->p_nlwps == 1);
672 }
673 
674 int
675 do_sys_wait(struct lwp *l, int *pid, int *status, int options,
676     struct rusage *ru, int *was_zombie)
677 {
678 	struct proc	*child;
679 	int		error;
680 
681 	mutex_enter(proc_lock);
682 	error = find_stopped_child(l->l_proc, *pid, options, &child, status);
683 
684 	if (child == NULL) {
685 		mutex_exit(proc_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 proc_lock. */
694 		*was_zombie = 1;
695 		if (options & WNOWAIT)
696 			mutex_exit(proc_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(proc_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 proc_lock held, and may release while waiting.
745  */
746 static int
747 find_stopped_child(struct proc *parent, pid_t pid, int options,
748 		   struct proc **child_p, int *status_p)
749 {
750 	struct proc *child, *dead;
751 	int error;
752 
753 	KASSERT(mutex_owned(proc_lock));
754 
755 	if (options & ~(WUNTRACED|WNOHANG|WALTSIG|WALLSIG)
756 	    && !(options & WOPTSCHECKED)) {
757 		*child_p = NULL;
758 		return EINVAL;
759 	}
760 
761 	if (pid == 0 && !(options & WOPTSCHECKED))
762 		pid = -parent->p_pgid;
763 
764 	for (;;) {
765 		error = ECHILD;
766 		dead = NULL;
767 
768 		LIST_FOREACH(child, &parent->p_children, p_sibling) {
769 			if (pid >= 0) {
770 				if (child->p_pid != pid) {
771 					child = p_find(pid, PFIND_ZOMBIE |
772 					    PFIND_LOCKED);
773 					if (child == NULL ||
774 					    child->p_pptr != parent) {
775 						child = NULL;
776 						break;
777 					}
778 				}
779 			} else if (pid != WAIT_ANY && child->p_pgid != -pid) {
780 				/* Child not in correct pgrp */
781 				continue;
782 			}
783 
784 			/*
785 			 * Wait for processes with p_exitsig != SIGCHLD
786 			 * processes only if WALTSIG is set; wait for
787 			 * processes with p_exitsig == SIGCHLD only
788 			 * if WALTSIG is clear.
789 			 */
790 			if (((options & WALLSIG) == 0) &&
791 			    (options & WALTSIG ? child->p_exitsig == SIGCHLD
792 						: P_EXITSIG(child) != SIGCHLD)){
793 				if (child->p_pid == pid) {
794 					child = NULL;
795 					break;
796 				}
797 				continue;
798 			}
799 
800 			error = 0;
801 			if ((options & WNOZOMBIE) == 0) {
802 				if (child->p_stat == SZOMB)
803 					break;
804 				if (child->p_stat == SDEAD) {
805 					/*
806 					 * We may occasionally arrive here
807 					 * after receiving a signal, but
808 					 * immediatley before the child
809 					 * process is zombified.  The wait
810 					 * will be short, so avoid returning
811 					 * to userspace.
812 					 */
813 					dead = child;
814 				}
815 			}
816 
817 			if (child->p_stat == SSTOP &&
818 			    child->p_waited == 0 &&
819 			    (child->p_slflag & PSL_TRACED ||
820 			    options & WUNTRACED)) {
821 				if ((options & WNOWAIT) == 0) {
822 					child->p_waited = 1;
823 					parent->p_nstopchild--;
824 				}
825 				break;
826 			}
827 			if (parent->p_nstopchild == 0 || child->p_pid == pid) {
828 				child = NULL;
829 				break;
830 			}
831 		}
832 
833 		if (child != NULL || error != 0 ||
834 		    ((options & WNOHANG) != 0 && dead == NULL)) {
835 		    	if (child != NULL) {
836 			    	*status_p = child->p_xstat;
837 			}
838 			*child_p = child;
839 			return error;
840 		}
841 
842 		/*
843 		 * Wait for another child process to stop.
844 		 */
845 		error = cv_wait_sig(&parent->p_waitcv, proc_lock);
846 
847 		if (error != 0) {
848 			*child_p = NULL;
849 			return error;
850 		}
851 	}
852 }
853 
854 /*
855  * Free a process after parent has taken all the state info.  Must be called
856  * with the proclist lock held, and will release before returning.
857  *
858  * *ru is returned to the caller, and must be freed by the caller.
859  */
860 static void
861 proc_free(struct proc *p, struct rusage *ru)
862 {
863 	struct proc *parent;
864 	struct lwp *l;
865 	ksiginfo_t ksi;
866 	kauth_cred_t cred1, cred2;
867 	uid_t uid;
868 
869 	KASSERT(mutex_owned(proc_lock));
870 	KASSERT(p->p_nlwps == 1);
871 	KASSERT(p->p_nzlwps == 1);
872 	KASSERT(p->p_nrlwps == 0);
873 	KASSERT(p->p_stat == SZOMB);
874 
875 	/*
876 	 * If we got the child via ptrace(2) or procfs, and
877 	 * the parent is different (meaning the process was
878 	 * attached, rather than run as a child), then we need
879 	 * to give it back to the old parent, and send the
880 	 * parent the exit signal.  The rest of the cleanup
881 	 * will be done when the old parent waits on the child.
882 	 */
883 	if ((p->p_slflag & PSL_TRACED) != 0) {
884 		parent = p->p_pptr;
885 		if (p->p_opptr != parent){
886 			mutex_enter(p->p_lock);
887 			p->p_slflag &= ~(PSL_TRACED|PSL_FSTRACE|PSL_SYSCALL);
888 			mutex_exit(p->p_lock);
889 			parent = p->p_opptr;
890 			if (parent == NULL)
891 				parent = initproc;
892 			proc_reparent(p, parent);
893 			p->p_opptr = NULL;
894 			if (p->p_exitsig != 0) {
895 				exit_psignal(p, parent, &ksi);
896 				kpsignal(parent, &ksi, NULL);
897 			}
898 			cv_broadcast(&parent->p_waitcv);
899 			mutex_exit(proc_lock);
900 			return;
901 		}
902 	}
903 
904 	/*
905 	 * Finally finished with old proc entry.  Unlink it from its process
906 	 * group.
907 	 */
908 	leavepgrp(p);
909 
910 	parent = p->p_pptr;
911 	sched_proc_exit(parent, p);
912 
913 	/*
914 	 * Add child times of exiting process onto its own times.
915 	 * This cannot be done any earlier else it might get done twice.
916 	 */
917 	l = LIST_FIRST(&p->p_lwps);
918 	p->p_stats->p_ru.ru_nvcsw += (l->l_ncsw - l->l_nivcsw);
919 	p->p_stats->p_ru.ru_nivcsw += l->l_nivcsw;
920 	ruadd(&p->p_stats->p_ru, &l->l_ru);
921 	ruadd(&p->p_stats->p_ru, &p->p_stats->p_cru);
922 	ruadd(&parent->p_stats->p_cru, &p->p_stats->p_ru);
923 	if (ru != NULL)
924 		*ru = p->p_stats->p_ru;
925 	p->p_xstat = 0;
926 
927 	/* Release any SA state. */
928 #ifdef KERN_SA
929 	if (p->p_sa)
930 		sa_release(p);
931 #endif
932 
933 	/*
934 	 * At this point we are going to start freeing the final resources.
935 	 * If anyone tries to access the proc structure after here they will
936 	 * get a shock - bits are missing.  Attempt to make it hard!  We
937 	 * don't bother with any further locking past this point.
938 	 */
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 	LIST_REMOVE(p, p_sibling);
944 
945 	/*
946 	 * Let pid be reallocated.
947 	 */
948 	proc_free_pid(p);
949 	mutex_exit(proc_lock);
950 
951 	/*
952 	 * Delay release until after lwp_free.
953 	 */
954 	cred2 = l->l_cred;
955 
956 	/*
957 	 * Free the last LWP's resources.
958 	 *
959 	 * lwp_free ensures the LWP is no longer running on another CPU.
960 	 */
961 	lwp_free(l, false, true);
962 
963 	/*
964 	 * Now no one except us can reach the process p.
965 	 */
966 
967 	/*
968 	 * Decrement the count of procs running with this uid.
969 	 */
970 	cred1 = p->p_cred;
971 	uid = kauth_cred_getuid(cred1);
972 	(void)chgproccnt(uid, -1);
973 
974 	/*
975 	 * Release substructures.
976 	 */
977 
978 	limfree(p->p_limit);
979 	pstatsfree(p->p_stats);
980 	kauth_cred_free(cred1);
981 	kauth_cred_free(cred2);
982 
983 	/*
984 	 * Release reference to text vnode
985 	 */
986 	if (p->p_textvp)
987 		vrele(p->p_textvp);
988 
989 	mutex_destroy(&p->p_auxlock);
990 	mutex_obj_free(p->p_lock);
991 	mutex_destroy(&p->p_stmutex);
992 	cv_destroy(&p->p_waitcv);
993 	cv_destroy(&p->p_lwpcv);
994 	rw_destroy(&p->p_reflock);
995 
996 	proc_free_mem(p);
997 }
998 
999 /*
1000  * make process 'parent' the new parent of process 'child'.
1001  *
1002  * Must be called with proc_lock held.
1003  */
1004 void
1005 proc_reparent(struct proc *child, struct proc *parent)
1006 {
1007 
1008 	KASSERT(mutex_owned(proc_lock));
1009 
1010 	if (child->p_pptr == parent)
1011 		return;
1012 
1013 	if (child->p_stat == SZOMB ||
1014 	    (child->p_stat == SSTOP && !child->p_waited)) {
1015 		child->p_pptr->p_nstopchild--;
1016 		parent->p_nstopchild++;
1017 	}
1018 	if (parent == initproc)
1019 		child->p_exitsig = SIGCHLD;
1020 
1021 	LIST_REMOVE(child, p_sibling);
1022 	LIST_INSERT_HEAD(&parent->p_children, child, p_sibling);
1023 	child->p_pptr = parent;
1024 	child->p_ppid = parent->p_pid;
1025 }
1026