xref: /netbsd-src/sys/kern/kern_exit.c (revision e39ef1d61eee3ccba837ee281f1e098c864487aa)
1 /*	$NetBSD: kern_exit.c,v 1.235 2011/10/09 20:30:37 chs 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 1.235 2011/10/09 20:30:37 chs 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 #include <sys/sdt.h>
113 
114 #include <uvm/uvm_extern.h>
115 
116 #ifdef DEBUG_EXIT
117 int debug_exit = 0;
118 #define DPRINTF(x) if (debug_exit) printf x
119 #else
120 #define DPRINTF(x)
121 #endif
122 
123 static int find_stopped_child(struct proc *, pid_t, int, struct proc **, int *);
124 static void proc_free(struct proc *, struct rusage *);
125 
126 /*
127  * DTrace SDT provider definitions
128  */
129 SDT_PROBE_DEFINE(proc,,,exit,
130 	    "int", NULL, 		/* reason */
131 	    NULL, NULL, NULL, NULL,
132 	    NULL, NULL, NULL, NULL);
133 /*
134  * Fill in the appropriate signal information, and signal the parent.
135  */
136 /* XXX noclone works around a gcc 4.5 bug on arm */
137 static void __noclone
138 exit_psignal(struct proc *p, struct proc *pp, ksiginfo_t *ksi)
139 {
140 
141 	KSI_INIT(ksi);
142 	if ((ksi->ksi_signo = P_EXITSIG(p)) == SIGCHLD) {
143 		if (WIFSIGNALED(p->p_xstat)) {
144 			if (WCOREDUMP(p->p_xstat))
145 				ksi->ksi_code = CLD_DUMPED;
146 			else
147 				ksi->ksi_code = CLD_KILLED;
148 		} else {
149 			ksi->ksi_code = CLD_EXITED;
150 		}
151 	}
152 	/*
153 	 * We fill those in, even for non-SIGCHLD.
154 	 * It's safe to access p->p_cred unlocked here.
155 	 */
156 	ksi->ksi_pid = p->p_pid;
157 	ksi->ksi_uid = kauth_cred_geteuid(p->p_cred);
158 	ksi->ksi_status = p->p_xstat;
159 	/* XXX: is this still valid? */
160 	ksi->ksi_utime = p->p_stats->p_ru.ru_utime.tv_sec;
161 	ksi->ksi_stime = p->p_stats->p_ru.ru_stime.tv_sec;
162 }
163 
164 /*
165  * exit --
166  *	Death of process.
167  */
168 int
169 sys_exit(struct lwp *l, const struct sys_exit_args *uap, register_t *retval)
170 {
171 	/* {
172 		syscallarg(int)	rval;
173 	} */
174 	struct proc *p = l->l_proc;
175 
176 	/* Don't call exit1() multiple times in the same process. */
177 	mutex_enter(p->p_lock);
178 	if (p->p_sflag & PS_WEXIT) {
179 		mutex_exit(p->p_lock);
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_lock held.  Does not return.
195  */
196 void
197 exit1(struct lwp *l, int rv)
198 {
199 	struct proc	*p, *q, *nq;
200 	struct pgrp	*pgrp;
201 	ksiginfo_t	ksi;
202 	ksiginfoq_t	kq;
203 	int		wakeinit, sa;
204 
205 	p = l->l_proc;
206 
207 	KASSERT(mutex_owned(p->p_lock));
208 
209 	if (__predict_false(p == initproc))
210 		panic("init died (signal %d, exit %d)",
211 		    WTERMSIG(rv), WEXITSTATUS(rv));
212 
213 	/*
214 	 * Disable scheduler activation upcalls.  We're trying to get out of
215 	 * here.
216 	 */
217 	sa = 0;
218 #ifdef KERN_SA
219 	if ((p->p_sa != NULL)) {
220 		l->l_pflag |= LP_SA_NOBLOCK;
221 		sa = 1;
222 	}
223 #endif
224 
225 	p->p_sflag |= PS_WEXIT;
226 
227 	/*
228 	 * Force all other LWPs to exit before we do.  Only then can we
229 	 * begin to tear down the rest of the process state.
230 	 */
231 	if (sa || p->p_nlwps > 1)
232 		exit_lwps(l);
233 
234 	ksiginfo_queue_init(&kq);
235 
236 	/*
237 	 * If we have been asked to stop on exit, do so now.
238 	 */
239 	if (__predict_false(p->p_sflag & PS_STOPEXIT)) {
240 		KERNEL_UNLOCK_ALL(l, &l->l_biglocks);
241 		sigclearall(p, &contsigmask, &kq);
242 		p->p_waited = 0;
243 		membar_producer();
244 		p->p_stat = SSTOP;
245 		lwp_lock(l);
246 		p->p_nrlwps--;
247 		l->l_stat = LSSTOP;
248 		lwp_unlock(l);
249 		mutex_exit(p->p_lock);
250 		lwp_lock(l);
251 		mi_switch(l);
252 		KERNEL_LOCK(l->l_biglocks, l);
253 		mutex_enter(p->p_lock);
254 	}
255 
256 	/*
257 	 * Bin any remaining signals and mark the process as dying so it will
258 	 * not be found for, e.g. signals.
259 	 */
260 	sigfillset(&p->p_sigctx.ps_sigignore);
261 	sigclearall(p, NULL, &kq);
262 	p->p_stat = SDYING;
263 	mutex_exit(p->p_lock);
264 	ksiginfo_queue_drain(&kq);
265 
266 	/* Destroy any lwpctl info. */
267 	if (p->p_lwpctl != NULL)
268 		lwp_ctl_exit();
269 
270 	/*
271 	 * Drain all remaining references that procfs, ptrace and others may
272 	 * have on the process.
273 	 */
274 	rw_enter(&p->p_reflock, RW_WRITER);
275 
276 	DPRINTF(("exit1: %d.%d exiting.\n", p->p_pid, l->l_lid));
277 
278 	timers_free(p, TIMERS_ALL);
279 #if defined(__HAVE_RAS)
280 	ras_purgeall();
281 #endif
282 
283 	/*
284 	 * Close open files, release open-file table and free signal
285 	 * actions.  This may block!
286 	 */
287 	fd_free();
288 	cwdfree(p->p_cwdi);
289 	p->p_cwdi = NULL;
290 	doexithooks(p);
291 	sigactsfree(p->p_sigacts);
292 
293 	/*
294 	 * Write out accounting data.
295 	 */
296 	(void)acct_process(l);
297 
298 #ifdef KTRACE
299 	/*
300 	 * Release trace file.
301 	 */
302 	if (p->p_tracep != NULL) {
303 		mutex_enter(&ktrace_lock);
304 		ktrderef(p);
305 		mutex_exit(&ktrace_lock);
306 	}
307 #endif
308 
309 	/*
310 	 * If emulation has process exit hook, call it now.
311 	 * Set the exit status now so that the exit hook has
312 	 * an opportunity to tweak it (COMPAT_LINUX requires
313 	 * this for thread group emulation)
314 	 */
315 	p->p_xstat = rv;
316 	if (p->p_emul->e_proc_exit)
317 		(*p->p_emul->e_proc_exit)(p);
318 
319 	/*
320 	 * Free the VM resources we're still holding on to.
321 	 * We must do this from a valid thread because doing
322 	 * so may block. This frees vmspace, which we don't
323 	 * need anymore. The only remaining lwp is the one
324 	 * we run at this moment, nothing runs in userland
325 	 * anymore.
326 	 */
327 	uvm_proc_exit(p);
328 
329 	/*
330 	 * Stop profiling.
331 	 */
332 	if (__predict_false((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, PL_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(proc_lock);
344 	if (p->p_lflag & PL_PPWAIT) {
345 		l->l_lwpctl = NULL; /* was on loan from blocked parent */
346 		p->p_lflag &= ~PL_PPWAIT;
347 		cv_broadcast(&p->p_pptr->p_waitcv);
348 	}
349 
350 	if (SESS_LEADER(p)) {
351 		struct vnode *vprele = NULL, *vprevoke = NULL;
352 		struct session *sp = p->p_session;
353 		struct tty *tp;
354 
355 		if (sp->s_ttyvp) {
356 			/*
357 			 * Controlling process.
358 			 * Signal foreground pgrp,
359 			 * drain controlling terminal
360 			 * and revoke access to controlling terminal.
361 			 */
362 			tp = sp->s_ttyp;
363 			mutex_spin_enter(&tty_lock);
364 			if (tp->t_session == sp) {
365 				/* we can't guarantee the revoke will do this */
366 				pgrp = tp->t_pgrp;
367 				tp->t_pgrp = NULL;
368 				tp->t_session = NULL;
369 				mutex_spin_exit(&tty_lock);
370 				if (pgrp != NULL) {
371 					pgsignal(pgrp, SIGHUP, 1);
372 				}
373 				mutex_exit(proc_lock);
374 				(void) ttywait(tp);
375 				mutex_enter(proc_lock);
376 
377 				/* The tty could have been revoked. */
378 				vprevoke = sp->s_ttyvp;
379 			} else
380 				mutex_spin_exit(&tty_lock);
381 			vprele = sp->s_ttyvp;
382 			sp->s_ttyvp = NULL;
383 			/*
384 			 * s_ttyp is not zero'd; we use this to indicate
385 			 * that the session once had a controlling terminal.
386 			 * (for logging and informational purposes)
387 			 */
388 		}
389 		sp->s_leader = NULL;
390 
391 		if (vprevoke != NULL || vprele != NULL) {
392 			if (vprevoke != NULL) {
393 				/* Releases proc_lock. */
394 				proc_sessrele(sp);
395 				VOP_REVOKE(vprevoke, REVOKEALL);
396 			} else
397 				mutex_exit(proc_lock);
398 			if (vprele != NULL)
399 				vrele(vprele);
400 			mutex_enter(proc_lock);
401 		}
402 	}
403 	fixjobc(p, p->p_pgrp, 0);
404 
405 	/*
406 	 * Finalize the last LWP's specificdata, as well as the
407 	 * specificdata for the proc itself.
408 	 */
409 	lwp_finispecific(l);
410 	proc_finispecific(p);
411 
412 	/*
413 	 * Notify interested parties of our demise.
414 	 */
415 	KNOTE(&p->p_klist, NOTE_EXIT);
416 
417 	SDT_PROBE(proc,,,exit,
418 		(WCOREDUMP(rv) ? CLD_DUMPED :
419 		 (WIFSIGNALED(rv) ? CLD_KILLED : CLD_EXITED)),
420 		0,0,0,0);
421 
422 #if PERFCTRS
423 	/*
424 	 * Save final PMC information in parent process & clean up.
425 	 */
426 	if (PMC_ENABLED(p)) {
427 		pmc_save_context(p);
428 		pmc_accumulate(p->p_pptr, p);
429 		pmc_process_exit(p);
430 	}
431 #endif
432 
433 	/*
434 	 * Reset p_opptr pointer of all former children which got
435 	 * traced by another process and were reparented. We reset
436 	 * it to NULL here; the trace detach code then reparents
437 	 * the child to initproc. We only check allproc list, since
438 	 * eventual former children on zombproc list won't reference
439 	 * p_opptr anymore.
440 	 */
441 	if (__predict_false(p->p_slflag & PSL_CHTRACED)) {
442 		PROCLIST_FOREACH(q, &allproc) {
443 			if (q->p_opptr == p)
444 				q->p_opptr = NULL;
445 		}
446 	}
447 
448 	/*
449 	 * Give orphaned children to init(8).
450 	 */
451 	q = LIST_FIRST(&p->p_children);
452 	wakeinit = (q != NULL);
453 	for (; q != NULL; q = nq) {
454 		nq = LIST_NEXT(q, p_sibling);
455 
456 		/*
457 		 * Traced processes are killed since their existence
458 		 * means someone is screwing up. Since we reset the
459 		 * trace flags, the logic in sys_wait4() would not be
460 		 * triggered to reparent the process to its
461 		 * original parent, so we must do this here.
462 		 */
463 		if (__predict_false(q->p_slflag & PSL_TRACED)) {
464 			mutex_enter(p->p_lock);
465 			q->p_slflag &= ~(PSL_TRACED|PSL_FSTRACE|PSL_SYSCALL);
466 			mutex_exit(p->p_lock);
467 			if (q->p_opptr != q->p_pptr) {
468 				struct proc *t = q->p_opptr;
469 				proc_reparent(q, t ? t : initproc);
470 				q->p_opptr = NULL;
471 			} else
472 				proc_reparent(q, initproc);
473 			killproc(q, "orphaned traced process");
474 		} else
475 			proc_reparent(q, initproc);
476 	}
477 
478 	/*
479 	 * Move proc from allproc to zombproc, it's now nearly ready to be
480 	 * collected by parent.
481 	 */
482 	LIST_REMOVE(l, l_list);
483 	LIST_REMOVE(p, p_list);
484 	LIST_INSERT_HEAD(&zombproc, p, p_list);
485 
486 	/*
487 	 * Mark the process as dead.  We must do this before we signal
488 	 * the parent.
489 	 */
490 	p->p_stat = SDEAD;
491 
492 	/* Put in front of parent's sibling list for parent to collect it */
493 	q = p->p_pptr;
494 	q->p_nstopchild++;
495 	if (LIST_FIRST(&q->p_children) != p) {
496 		/* Put child where it can be found quickly */
497 		LIST_REMOVE(p, p_sibling);
498 		LIST_INSERT_HEAD(&q->p_children, p, p_sibling);
499 	}
500 
501 	/*
502 	 * Notify parent that we're gone.  If parent has the P_NOCLDWAIT
503 	 * flag set, notify init instead (and hope it will handle
504 	 * this situation).
505 	 */
506 	if (q->p_flag & (PK_NOCLDWAIT|PK_CLDSIGIGN)) {
507 		proc_reparent(p, initproc);
508 		wakeinit = 1;
509 
510 		/*
511 		 * If this was the last child of our parent, notify
512 		 * parent, so in case he was wait(2)ing, he will
513 		 * continue.
514 		 */
515 		if (LIST_FIRST(&q->p_children) == NULL)
516 			cv_broadcast(&q->p_waitcv);
517 	}
518 
519 	/* Reload parent pointer, since p may have been reparented above */
520 	q = p->p_pptr;
521 
522 	if (__predict_false((p->p_slflag & PSL_FSTRACE) == 0 &&
523 	    p->p_exitsig != 0)) {
524 		exit_psignal(p, q, &ksi);
525 		kpsignal(q, &ksi, NULL);
526 	}
527 
528 	/* Calculate the final rusage info.  */
529 	calcru(p, &p->p_stats->p_ru.ru_utime, &p->p_stats->p_ru.ru_stime,
530 	    NULL, NULL);
531 
532 	if (wakeinit)
533 		cv_broadcast(&initproc->p_waitcv);
534 
535 	callout_destroy(&l->l_timeout_ch);
536 
537 	/*
538 	 * Release any PCU resources before becoming a zombie.
539 	 */
540 	pcu_discard_all(l);
541 
542 	/*
543 	 * Remaining lwp resources will be freed in lwp_exit2() once we've
544 	 * switch to idle context; at that point, we will be marked as a
545 	 * full blown zombie.
546 	 */
547 	mutex_enter(p->p_lock);
548 	lwp_drainrefs(l);
549 	lwp_lock(l);
550 	l->l_prflag &= ~LPR_DETACHED;
551 	l->l_stat = LSZOMB;
552 	lwp_unlock(l);
553 	KASSERT(curlwp == l);
554 	KASSERT(p->p_nrlwps == 1);
555 	KASSERT(p->p_nlwps == 1);
556 	p->p_stat = SZOMB;
557 	p->p_nrlwps--;
558 	p->p_nzlwps++;
559 	p->p_ndlwps = 0;
560 	mutex_exit(p->p_lock);
561 
562 	/*
563 	 * Signal the parent to collect us, and drop the proclist lock.
564 	 * Drop debugger/procfs lock; no new references can be gained.
565 	 */
566 	cv_broadcast(&p->p_pptr->p_waitcv);
567 	rw_exit(&p->p_reflock);
568 	mutex_exit(proc_lock);
569 
570 	/* Verify that we hold no locks other than the kernel lock. */
571 	LOCKDEBUG_BARRIER(&kernel_lock, 0);
572 
573 	/*
574 	 * NOTE: WE ARE NO LONGER ALLOWED TO SLEEP!
575 	 */
576 
577 	/*
578 	 * Give machine-dependent code a chance to free any MD LWP
579 	 * resources.  This must be done before uvm_lwp_exit(), in
580 	 * case these resources are in the PCB.
581 	 */
582 	cpu_lwp_free(l, 1);
583 	pmap_deactivate(l);
584 
585 	/* This process no longer needs to hold the kernel lock. */
586 #ifdef notyet
587 	/* XXXSMP hold in lwp_userret() */
588 	KERNEL_UNLOCK_LAST(l);
589 #else
590 	KERNEL_UNLOCK_ALL(l, NULL);
591 #endif
592 
593 	lwp_exit_switchaway(l);
594 }
595 
596 void
597 exit_lwps(struct lwp *l)
598 {
599 	struct proc *p;
600 	struct lwp *l2;
601 	int error;
602 	lwpid_t waited;
603 	int nlocks;
604 
605 	KERNEL_UNLOCK_ALL(l, &nlocks);
606 
607 	p = l->l_proc;
608 	KASSERT(mutex_owned(p->p_lock));
609 
610 #ifdef KERN_SA
611 	if (p->p_sa != NULL) {
612 		struct sadata_vp *vp;
613 		SLIST_FOREACH(vp, &p->p_sa->sa_vps, savp_next) {
614 			/*
615 			 * Make SA-cached LWPs normal process interruptable
616 			 * so that the exit code can wake them. Locking
617 			 * savp_mutex locks all the lwps on this vp that
618 			 * we need to adjust.
619 			 */
620 			mutex_enter(&vp->savp_mutex);
621 			DPRINTF(("exit_lwps: Making cached LWPs of %d on "
622 			    "VP %d interruptable: ", p->p_pid, vp->savp_id));
623 			TAILQ_FOREACH(l2, &vp->savp_lwpcache, l_sleepchain) {
624 				l2->l_flag |= LW_SINTR;
625 				DPRINTF(("%d ", l2->l_lid));
626 			}
627 			DPRINTF(("\n"));
628 
629 			DPRINTF(("exit_lwps: Making unblocking LWPs of %d on "
630 			    "VP %d interruptable: ", p->p_pid, vp->savp_id));
631 			TAILQ_FOREACH(l2, &vp->savp_woken, l_sleepchain) {
632 				vp->savp_woken_count--;
633 				l2->l_flag |= LW_SINTR;
634 				DPRINTF(("%d ", l2->l_lid));
635 			}
636 			DPRINTF(("\n"));
637 			mutex_exit(&vp->savp_mutex);
638 		}
639 	}
640 #endif
641 
642  retry:
643 	/*
644 	 * Interrupt LWPs in interruptable sleep, unsuspend suspended
645 	 * LWPs and then wait for everyone else to finish.
646 	 */
647 	LIST_FOREACH(l2, &p->p_lwps, l_sibling) {
648 		if (l2 == l)
649 			continue;
650 		lwp_lock(l2);
651 		l2->l_flag &= ~LW_SA;
652 		l2->l_flag |= LW_WEXIT;
653 		if ((l2->l_stat == LSSLEEP && (l2->l_flag & LW_SINTR)) ||
654 		    l2->l_stat == LSSUSPENDED || l2->l_stat == LSSTOP) {
655 		    	/* setrunnable() will release the lock. */
656 			setrunnable(l2);
657 			DPRINTF(("exit_lwps: Made %d.%d runnable\n",
658 			    p->p_pid, l2->l_lid));
659 			continue;
660 		}
661 		lwp_unlock(l2);
662 	}
663 	while (p->p_nlwps > 1) {
664 		DPRINTF(("exit_lwps: waiting for %d LWPs (%d zombies)\n",
665 		    p->p_nlwps, p->p_nzlwps));
666 		error = lwp_wait1(l, 0, &waited, LWPWAIT_EXITCONTROL);
667 		if (p->p_nlwps == 1)
668 			break;
669 		if (error == EDEADLK) {
670 			/*
671 			 * LWPs can get suspended/slept behind us.
672 			 * (eg. sa_setwoken)
673 			 * kick them again and retry.
674 			 */
675 			goto retry;
676 		}
677 		if (error)
678 			panic("exit_lwps: lwp_wait1 failed with error %d",
679 			    error);
680 		DPRINTF(("exit_lwps: Got LWP %d from lwp_wait1()\n", waited));
681 	}
682 
683 	KERNEL_LOCK(nlocks, l);
684 	KASSERT(p->p_nlwps == 1);
685 }
686 
687 int
688 do_sys_wait(int *pid, int *status, int options, struct rusage *ru)
689 {
690 	proc_t *child;
691 	int error;
692 
693 	if (ru != NULL) {
694 		memset(ru, 0, sizeof(*ru));
695 	}
696 	mutex_enter(proc_lock);
697 	error = find_stopped_child(curproc, *pid, options, &child, status);
698 	if (child == NULL) {
699 		mutex_exit(proc_lock);
700 		*pid = 0;
701 		return error;
702 	}
703 	*pid = child->p_pid;
704 
705 	if (child->p_stat == SZOMB) {
706 		/* proc_free() will release the proc_lock. */
707 		if (options & WNOWAIT) {
708 			mutex_exit(proc_lock);
709 		} else {
710 			proc_free(child, ru);
711 		}
712 	} else {
713 		/* Child state must have been SSTOP. */
714 		mutex_exit(proc_lock);
715 		*status = W_STOPCODE(*status);
716 	}
717 	return 0;
718 }
719 
720 int
721 sys___wait450(struct lwp *l, const struct sys___wait450_args *uap,
722     register_t *retval)
723 {
724 	/* {
725 		syscallarg(int)			pid;
726 		syscallarg(int *)		status;
727 		syscallarg(int)			options;
728 		syscallarg(struct rusage *)	rusage;
729 	} */
730 	int error, status, pid = SCARG(uap, pid);
731 	struct rusage ru;
732 
733 	error = do_sys_wait(&pid, &status, SCARG(uap, options),
734 	    SCARG(uap, rusage) != NULL ? &ru : NULL);
735 
736 	retval[0] = pid;
737 	if (pid == 0) {
738 		return error;
739 	}
740 	if (SCARG(uap, status)) {
741 		error = copyout(&status, SCARG(uap, status), sizeof(status));
742 	}
743 	if (SCARG(uap, rusage) && error == 0) {
744 		error = copyout(&ru, SCARG(uap, rusage), sizeof(ru));
745 	}
746 	return error;
747 }
748 
749 /*
750  * Scan list of child processes for a child process that has stopped or
751  * exited.  Used by sys_wait4 and 'compat' equivalents.
752  *
753  * Must be called with the proc_lock held, and may release while waiting.
754  */
755 static int
756 find_stopped_child(struct proc *parent, pid_t pid, int options,
757 		   struct proc **child_p, int *status_p)
758 {
759 	struct proc *child, *dead;
760 	int error;
761 
762 	KASSERT(mutex_owned(proc_lock));
763 
764 	if (options & ~(WUNTRACED|WNOHANG|WALTSIG|WALLSIG)
765 	    && !(options & WOPTSCHECKED)) {
766 		*child_p = NULL;
767 		return EINVAL;
768 	}
769 
770 	if (pid == 0 && !(options & WOPTSCHECKED))
771 		pid = -parent->p_pgid;
772 
773 	for (;;) {
774 		error = ECHILD;
775 		dead = NULL;
776 
777 		LIST_FOREACH(child, &parent->p_children, p_sibling) {
778 			if (pid >= 0) {
779 				if (child->p_pid != pid) {
780 					child = proc_find_raw(pid);
781 					if (child == NULL ||
782 					    child->p_stat == SIDL ||
783 					    child->p_pptr != parent) {
784 						child = NULL;
785 						break;
786 					}
787 				}
788 			} else if (pid != WAIT_ANY && child->p_pgid != -pid) {
789 				/* Child not in correct pgrp */
790 				continue;
791 			}
792 
793 			/*
794 			 * Wait for processes with p_exitsig != SIGCHLD
795 			 * processes only if WALTSIG is set; wait for
796 			 * processes with p_exitsig == SIGCHLD only
797 			 * if WALTSIG is clear.
798 			 */
799 			if (((options & WALLSIG) == 0) &&
800 			    (options & WALTSIG ? child->p_exitsig == SIGCHLD
801 						: P_EXITSIG(child) != SIGCHLD)){
802 				if (child->p_pid == pid) {
803 					child = NULL;
804 					break;
805 				}
806 				continue;
807 			}
808 
809 			error = 0;
810 			if ((options & WNOZOMBIE) == 0) {
811 				if (child->p_stat == SZOMB)
812 					break;
813 				if (child->p_stat == SDEAD) {
814 					/*
815 					 * We may occasionally arrive here
816 					 * after receiving a signal, but
817 					 * immediately before the child
818 					 * process is zombified.  The wait
819 					 * will be short, so avoid returning
820 					 * to userspace.
821 					 */
822 					dead = child;
823 				}
824 			}
825 
826 			if (child->p_stat == SSTOP &&
827 			    child->p_waited == 0 &&
828 			    (child->p_slflag & PSL_TRACED ||
829 			    options & WUNTRACED)) {
830 				if ((options & WNOWAIT) == 0) {
831 					child->p_waited = 1;
832 					parent->p_nstopchild--;
833 				}
834 				break;
835 			}
836 			if (parent->p_nstopchild == 0 || child->p_pid == pid) {
837 				child = NULL;
838 				break;
839 			}
840 		}
841 
842 		if (child != NULL || error != 0 ||
843 		    ((options & WNOHANG) != 0 && dead == NULL)) {
844 		    	if (child != NULL) {
845 			    	*status_p = child->p_xstat;
846 			}
847 			*child_p = child;
848 			return error;
849 		}
850 
851 		/*
852 		 * Wait for another child process to stop.
853 		 */
854 		error = cv_wait_sig(&parent->p_waitcv, proc_lock);
855 
856 		if (error != 0) {
857 			*child_p = NULL;
858 			return error;
859 		}
860 	}
861 }
862 
863 /*
864  * Free a process after parent has taken all the state info.  Must be called
865  * with the proclist lock held, and will release before returning.
866  *
867  * *ru is returned to the caller, and must be freed by the caller.
868  */
869 static void
870 proc_free(struct proc *p, struct rusage *ru)
871 {
872 	struct proc *parent = p->p_pptr;
873 	struct lwp *l;
874 	ksiginfo_t ksi;
875 	kauth_cred_t cred1, cred2;
876 	uid_t uid;
877 
878 	KASSERT(mutex_owned(proc_lock));
879 	KASSERT(p->p_nlwps == 1);
880 	KASSERT(p->p_nzlwps == 1);
881 	KASSERT(p->p_nrlwps == 0);
882 	KASSERT(p->p_stat == SZOMB);
883 
884 	/*
885 	 * If we got the child via ptrace(2) or procfs, and
886 	 * the parent is different (meaning the process was
887 	 * attached, rather than run as a child), then we need
888 	 * to give it back to the old parent, and send the
889 	 * parent the exit signal.  The rest of the cleanup
890 	 * will be done when the old parent waits on the child.
891 	 */
892 	if ((p->p_slflag & PSL_TRACED) != 0 && p->p_opptr != parent) {
893 		mutex_enter(p->p_lock);
894 		p->p_slflag &= ~(PSL_TRACED|PSL_FSTRACE|PSL_SYSCALL);
895 		mutex_exit(p->p_lock);
896 		parent = (p->p_opptr == NULL) ? initproc : p->p_opptr;
897 		proc_reparent(p, parent);
898 		p->p_opptr = NULL;
899 		if (p->p_exitsig != 0) {
900 			exit_psignal(p, parent, &ksi);
901 			kpsignal(parent, &ksi, NULL);
902 		}
903 		cv_broadcast(&parent->p_waitcv);
904 		mutex_exit(proc_lock);
905 		return;
906 	}
907 
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_nstopchild--;
939 	LIST_REMOVE(p, p_sibling);
940 
941 	/*
942 	 * Let pid be reallocated.
943 	 */
944 	proc_free_pid(p->p_pid);
945 
946 	/*
947 	 * Unlink process from its process group.
948 	 * Releases the proc_lock.
949 	 */
950 	proc_leavepgrp(p);
951 
952 	/*
953 	 * Delay release until after lwp_free.
954 	 */
955 	cred2 = l->l_cred;
956 
957 	/*
958 	 * Free the last LWP's resources.
959 	 *
960 	 * lwp_free ensures the LWP is no longer running on another CPU.
961 	 */
962 	lwp_free(l, false, true);
963 
964 	/*
965 	 * Now no one except us can reach the process p.
966 	 */
967 
968 	/*
969 	 * Decrement the count of procs running with this uid.
970 	 */
971 	cred1 = p->p_cred;
972 	uid = kauth_cred_getuid(cred1);
973 	(void)chgproccnt(uid, -1);
974 
975 	/*
976 	 * Release substructures.
977 	 */
978 
979 	lim_free(p->p_limit);
980 	pstatsfree(p->p_stats);
981 	kauth_cred_free(cred1);
982 	kauth_cred_free(cred2);
983 
984 	/*
985 	 * Release reference to text vnode
986 	 */
987 	if (p->p_textvp)
988 		vrele(p->p_textvp);
989 
990 	mutex_destroy(&p->p_auxlock);
991 	mutex_obj_free(p->p_lock);
992 	mutex_destroy(&p->p_stmutex);
993 	cv_destroy(&p->p_waitcv);
994 	cv_destroy(&p->p_lwpcv);
995 	rw_destroy(&p->p_reflock);
996 
997 	proc_free_mem(p);
998 }
999 
1000 /*
1001  * make process 'parent' the new parent of process 'child'.
1002  *
1003  * Must be called with proc_lock held.
1004  */
1005 void
1006 proc_reparent(struct proc *child, struct proc *parent)
1007 {
1008 
1009 	KASSERT(mutex_owned(proc_lock));
1010 
1011 	if (child->p_pptr == parent)
1012 		return;
1013 
1014 	if (child->p_stat == SZOMB ||
1015 	    (child->p_stat == SSTOP && !child->p_waited)) {
1016 		child->p_pptr->p_nstopchild--;
1017 		parent->p_nstopchild++;
1018 	}
1019 	if (parent == initproc)
1020 		child->p_exitsig = SIGCHLD;
1021 
1022 	LIST_REMOVE(child, p_sibling);
1023 	LIST_INSERT_HEAD(&parent->p_children, child, p_sibling);
1024 	child->p_pptr = parent;
1025 	child->p_ppid = parent->p_pid;
1026 }
1027