xref: /netbsd-src/sys/kern/kern_exit.c (revision 8b0f9554ff8762542c4defc4f70e1eb76fb508fa)
1 /*	$NetBSD: kern_exit.c,v 1.194 2007/11/30 23:05:43 ad Exp $	*/
2 
3 /*-
4  * Copyright (c) 1998, 1999, 2006, 2007 The NetBSD Foundation, Inc.
5  * All rights reserved.
6  *
7  * This code is derived from software contributed to The NetBSD Foundation
8  * by Jason R. Thorpe of the Numerical Aerospace Simulation Facility,
9  * NASA Ames Research Center, and by Andrew Doran.
10  *
11  * Redistribution and use in source and binary forms, with or without
12  * modification, are permitted provided that the following conditions
13  * are met:
14  * 1. Redistributions of source code must retain the above copyright
15  *    notice, this list of conditions and the following disclaimer.
16  * 2. Redistributions in binary form must reproduce the above copyright
17  *    notice, this list of conditions and the following disclaimer in the
18  *    documentation and/or other materials provided with the distribution.
19  * 3. All advertising materials mentioning features or use of this software
20  *    must display the following acknowledgement:
21  *	This product includes software developed by the NetBSD
22  *	Foundation, Inc. and its contributors.
23  * 4. Neither the name of The NetBSD Foundation nor the names of its
24  *    contributors may be used to endorse or promote products derived
25  *    from this software without specific prior written permission.
26  *
27  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
28  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
29  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
30  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
31  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
32  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
33  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
34  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
35  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
36  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
37  * POSSIBILITY OF SUCH DAMAGE.
38  */
39 
40 /*
41  * Copyright (c) 1982, 1986, 1989, 1991, 1993
42  *	The Regents of the University of California.  All rights reserved.
43  * (c) UNIX System Laboratories, Inc.
44  * All or some portions of this file are derived from material licensed
45  * to the University of California by American Telephone and Telegraph
46  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
47  * the permission of UNIX System Laboratories, Inc.
48  *
49  * Redistribution and use in source and binary forms, with or without
50  * modification, are permitted provided that the following conditions
51  * are met:
52  * 1. Redistributions of source code must retain the above copyright
53  *    notice, this list of conditions and the following disclaimer.
54  * 2. Redistributions in binary form must reproduce the above copyright
55  *    notice, this list of conditions and the following disclaimer in the
56  *    documentation and/or other materials provided with the distribution.
57  * 3. Neither the name of the University nor the names of its contributors
58  *    may be used to endorse or promote products derived from this software
59  *    without specific prior written permission.
60  *
61  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
62  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
63  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
64  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
65  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
66  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
67  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
68  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
69  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
70  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
71  * SUCH DAMAGE.
72  *
73  *	@(#)kern_exit.c	8.10 (Berkeley) 2/23/95
74  */
75 
76 #include <sys/cdefs.h>
77 __KERNEL_RCSID(0, "$NetBSD: kern_exit.c,v 1.194 2007/11/30 23:05:43 ad Exp $");
78 
79 #include "opt_ktrace.h"
80 #include "opt_perfctrs.h"
81 #include "opt_systrace.h"
82 #include "opt_sysv.h"
83 
84 #include <sys/param.h>
85 #include <sys/aio.h>
86 #include <sys/systm.h>
87 #include <sys/ioctl.h>
88 #include <sys/tty.h>
89 #include <sys/time.h>
90 #include <sys/resource.h>
91 #include <sys/kernel.h>
92 #include <sys/proc.h>
93 #include <sys/buf.h>
94 #include <sys/wait.h>
95 #include <sys/file.h>
96 #include <sys/vnode.h>
97 #include <sys/syslog.h>
98 #include <sys/malloc.h>
99 #include <sys/pool.h>
100 #include <sys/resourcevar.h>
101 #if defined(PERFCTRS)
102 #include <sys/pmc.h>
103 #endif
104 #include <sys/ptrace.h>
105 #include <sys/acct.h>
106 #include <sys/filedesc.h>
107 #include <sys/ras.h>
108 #include <sys/signalvar.h>
109 #include <sys/sched.h>
110 #include <sys/mount.h>
111 #include <sys/syscallargs.h>
112 #include <sys/systrace.h>
113 #include <sys/kauth.h>
114 #include <sys/sleepq.h>
115 #include <sys/lockdebug.h>
116 #include <sys/ktrace.h>
117 #include <sys/cpu.h>
118 #include <sys/lwpctl.h>
119 #include <sys/atomic.h>
120 
121 #include <uvm/uvm_extern.h>
122 
123 #define DEBUG_EXIT
124 
125 #ifdef DEBUG_EXIT
126 int debug_exit = 0;
127 #define DPRINTF(x) if (debug_exit) printf x
128 #else
129 #define DPRINTF(x)
130 #endif
131 
132 static int find_stopped_child(struct proc *, pid_t, int, struct proc **, int *);
133 static void proc_free(struct proc *, struct rusage *);
134 
135 /*
136  * Fill in the appropriate signal information, and signal the parent.
137  */
138 static void
139 exit_psignal(struct proc *p, struct proc *pp, ksiginfo_t *ksi)
140 {
141 
142 	KSI_INIT(ksi);
143 	if ((ksi->ksi_signo = P_EXITSIG(p)) == SIGCHLD) {
144 		if (WIFSIGNALED(p->p_xstat)) {
145 			if (WCOREDUMP(p->p_xstat))
146 				ksi->ksi_code = CLD_DUMPED;
147 			else
148 				ksi->ksi_code = CLD_KILLED;
149 		} else {
150 			ksi->ksi_code = CLD_EXITED;
151 		}
152 	}
153 	/*
154 	 * We fill those in, even for non-SIGCHLD.
155 	 * It's safe to access p->p_cred unlocked here.
156 	 */
157 	ksi->ksi_pid = p->p_pid;
158 	ksi->ksi_uid = kauth_cred_geteuid(p->p_cred);
159 	ksi->ksi_status = p->p_xstat;
160 	/* XXX: is this still valid? */
161 	ksi->ksi_utime = p->p_stats->p_ru.ru_utime.tv_sec;
162 	ksi->ksi_stime = p->p_stats->p_ru.ru_stime.tv_sec;
163 }
164 
165 /*
166  * exit --
167  *	Death of process.
168  */
169 int
170 sys_exit(struct lwp *l, void *v, register_t *retval)
171 {
172 	struct sys_exit_args /* {
173 		syscallarg(int)	rval;
174 	} */ *uap = v;
175 	struct proc *p = l->l_proc;
176 
177 	/* Don't call exit1() multiple times in the same process. */
178 	mutex_enter(&p->p_smutex);
179 	if (p->p_sflag & PS_WEXIT) {
180 		mutex_exit(&p->p_smutex);
181 		lwp_exit(l);
182 	}
183 
184 	/* exit1() will release the mutex. */
185 	exit1(l, W_EXITCODE(SCARG(uap, rval), 0));
186 	/* NOTREACHED */
187 	return (0);
188 }
189 
190 /*
191  * Exit: deallocate address space and other resources, change proc state
192  * to zombie, and unlink proc from allproc and parent's lists.  Save exit
193  * status and rusage for wait().  Check for child processes and orphan them.
194  *
195  * Must be called with p->p_smutex held.  Does not return.
196  */
197 void
198 exit1(struct lwp *l, int rv)
199 {
200 	struct proc	*p, *q, *nq;
201 	ksiginfo_t	ksi;
202 	ksiginfoq_t	kq;
203 	int		wakeinit;
204 
205 	p = l->l_proc;
206 
207 	KASSERT(mutex_owned(&p->p_smutex));
208 
209 	if (__predict_false(p == initproc))
210 		panic("init died (signal %d, exit %d)",
211 		    WTERMSIG(rv), WEXITSTATUS(rv));
212 
213 	p->p_sflag |= PS_WEXIT;
214 
215 	/*
216 	 * Force all other LWPs to exit before we do.  Only then can we
217 	 * begin to tear down the rest of the process state.
218 	 */
219 	if (p->p_nlwps > 1)
220 		exit_lwps(l);
221 
222 	ksiginfo_queue_init(&kq);
223 
224 	/*
225 	 * If we have been asked to stop on exit, do so now.
226 	 */
227 	if (p->p_sflag & PS_STOPEXIT) {
228 		KERNEL_UNLOCK_ALL(l, &l->l_biglocks);
229 		sigclearall(p, &contsigmask, &kq);
230 		p->p_waited = 0;
231 		membar_producer();
232 		p->p_stat = SSTOP;
233 		lwp_lock(l);
234 		p->p_nrlwps--;
235 		l->l_stat = LSSTOP;
236 		mutex_exit(&p->p_smutex);
237 		mi_switch(l);
238 		KERNEL_LOCK(l->l_biglocks, l);
239 	} else
240 		mutex_exit(&p->p_smutex);
241 
242 	/* Destroy any lwpctl info. */
243 	if (p->p_lwpctl != NULL)
244 		lwp_ctl_exit();
245 
246 	/* Destroy all AIO works */
247 	aio_exit(p, p->p_aio);
248 
249 	/*
250 	 * Drain all remaining references that procfs, ptrace and others may
251 	 * have on the process.
252 	 */
253 	rw_enter(&p->p_reflock, RW_WRITER);
254 
255 	/*
256 	 * Bin any remaining signals and mark the process as dying so it will
257 	 * not be found for, e.g. signals.
258 	 */
259 	mutex_enter(&p->p_smutex);
260 	sigfillset(&p->p_sigctx.ps_sigignore);
261 	sigclearall(p, NULL, &kq);
262 	p->p_stat = SDYING;
263 	mutex_exit(&p->p_smutex);
264 	ksiginfo_queue_drain(&kq);
265 
266 	DPRINTF(("exit1: %d.%d exiting.\n", p->p_pid, l->l_lid));
267 
268 #ifdef PGINPROF
269 	vmsizmon();
270 #endif
271 	timers_free(p, TIMERS_ALL);
272 #if defined(__HAVE_RAS)
273 	ras_purgeall();
274 #endif
275 
276 	/*
277 	 * Close open files, release open-file table and free signal
278 	 * actions.  This may block!
279 	 */
280 	fdfree(l);
281 	cwdfree(p->p_cwdi);
282 	p->p_cwdi = NULL;
283 	doexithooks(p);
284 	sigactsfree(p->p_sigacts);
285 
286 	/*
287 	 * Write out accounting data.
288 	 */
289 	(void)acct_process(l);
290 
291 #ifdef KTRACE
292 	/*
293 	 * Release trace file.
294 	 */
295 	if (p->p_tracep != NULL) {
296 		mutex_enter(&ktrace_lock);
297 		ktrderef(p);
298 		mutex_exit(&ktrace_lock);
299 	}
300 #endif
301 #ifdef SYSTRACE
302 	systrace_sys_exit(p);
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 	/* Collect child u-areas. */
316 	uvm_uarea_drain(false);
317 
318 	/*
319 	 * Free the VM resources we're still holding on to.
320 	 * We must do this from a valid thread because doing
321 	 * so may block. This frees vmspace, which we don't
322 	 * need anymore. The only remaining lwp is the one
323 	 * we run at this moment, nothing runs in userland
324 	 * anymore.
325 	 */
326 	uvm_proc_exit(p);
327 
328 	/*
329 	 * While we can still block, and mark the LWP as unswappable to
330 	 * prevent conflicts with the with the swapper.  We also shouldn't
331 	 * be swapped out, because we are about to exit and will release
332 	 * memory.
333 	 */
334 	uvm_lwp_hold(l);
335 
336 	/*
337 	 * Stop profiling.
338 	 */
339 	if ((p->p_stflag & PST_PROFIL) != 0) {
340 		mutex_spin_enter(&p->p_stmutex);
341 		stopprofclock(p);
342 		mutex_spin_exit(&p->p_stmutex);
343 	}
344 
345 	/*
346 	 * If parent is waiting for us to exit or exec, P_PPWAIT is set; we
347 	 * wake up the parent early to avoid deadlock.  We can do this once
348 	 * the VM resources are released.
349 	 */
350 	mutex_enter(&proclist_lock);
351 
352 	mutex_enter(&p->p_smutex);
353 	if (p->p_sflag & PS_PPWAIT) {
354 		p->p_sflag &= ~PS_PPWAIT;
355 		cv_broadcast(&p->p_pptr->p_waitcv);
356 	}
357 	mutex_exit(&p->p_smutex);
358 
359 	if (SESS_LEADER(p)) {
360 		struct vnode *vprele = NULL, *vprevoke = NULL;
361 		struct session *sp = p->p_session;
362 		struct tty *tp;
363 
364 		if (sp->s_ttyvp) {
365 			/*
366 			 * Controlling process.
367 			 * Signal foreground pgrp,
368 			 * drain controlling terminal
369 			 * and revoke access to controlling terminal.
370 			 */
371 			tp = sp->s_ttyp;
372 			mutex_spin_enter(&tty_lock);
373 			if (tp->t_session == sp) {
374 				if (tp->t_pgrp) {
375 					mutex_enter(&proclist_mutex);
376 					pgsignal(tp->t_pgrp, SIGHUP, 1);
377 					mutex_exit(&proclist_mutex);
378 				}
379 				/* we can't guarantee the revoke will do this */
380 				tp->t_pgrp = NULL;
381 				tp->t_session = NULL;
382 				mutex_spin_exit(&tty_lock);
383 				SESSRELE(sp);
384 				mutex_exit(&proclist_lock);
385 				(void) ttywait(tp);
386 				mutex_enter(&proclist_lock);
387 
388 				/*
389 				 * The tty could have been revoked
390 				 * if we blocked.
391 				 */
392 				vprevoke = sp->s_ttyvp;
393 			} else
394 				mutex_spin_exit(&tty_lock);
395 			vprele = sp->s_ttyvp;
396 			sp->s_ttyvp = NULL;
397 			/*
398 			 * s_ttyp is not zero'd; we use this to indicate
399 			 * that the session once had a controlling terminal.
400 			 * (for logging and informational purposes)
401 			 */
402 		}
403 		sp->s_leader = NULL;
404 
405 		if (vprevoke != NULL || vprele != NULL) {
406 			mutex_exit(&proclist_lock);
407 			if (vprevoke != NULL)
408 				VOP_REVOKE(vprevoke, REVOKEALL);
409 			if (vprele != NULL)
410 				vrele(vprele);
411 			mutex_enter(&proclist_lock);
412 		}
413 	}
414 	mutex_enter(&proclist_mutex);
415 	fixjobc(p, p->p_pgrp, 0);
416 	mutex_exit(&proclist_mutex);
417 
418 	/*
419 	 * Finalize the last LWP's specificdata, as well as the
420 	 * specificdata for the proc itself.
421 	 */
422 	lwp_finispecific(l);
423 	proc_finispecific(p);
424 
425 	/*
426 	 * Notify interested parties of our demise.
427 	 */
428 	KNOTE(&p->p_klist, NOTE_EXIT);
429 
430 #if PERFCTRS
431 	/*
432 	 * Save final PMC information in parent process & clean up.
433 	 */
434 	if (PMC_ENABLED(p)) {
435 		pmc_save_context(p);
436 		pmc_accumulate(p->p_pptr, p);
437 		pmc_process_exit(p);
438 	}
439 #endif
440 
441 	/*
442 	 * Reset p_opptr pointer of all former children which got
443 	 * traced by another process and were reparented. We reset
444 	 * it to NULL here; the trace detach code then reparents
445 	 * the child to initproc. We only check allproc list, since
446 	 * eventual former children on zombproc list won't reference
447 	 * p_opptr anymore.
448 	 */
449 	if (p->p_slflag & PSL_CHTRACED) {
450 		PROCLIST_FOREACH(q, &allproc) {
451 			if (q->p_opptr == p)
452 				q->p_opptr = NULL;
453 		}
454 	}
455 
456 	/*
457 	 * Give orphaned children to init(8).
458 	 */
459 	q = LIST_FIRST(&p->p_children);
460 	wakeinit = (q != NULL);
461 	for (; q != NULL; q = nq) {
462 		nq = LIST_NEXT(q, p_sibling);
463 
464 		/*
465 		 * Traced processes are killed since their existence
466 		 * means someone is screwing up. Since we reset the
467 		 * trace flags, the logic in sys_wait4() would not be
468 		 * triggered to reparent the process to its
469 		 * original parent, so we must do this here.
470 		 */
471 		if (q->p_slflag & PSL_TRACED) {
472 			mutex_enter(&p->p_smutex);
473 			q->p_slflag &= ~(PSL_TRACED|PSL_FSTRACE|PSL_SYSCALL);
474 			mutex_exit(&p->p_smutex);
475 			if (q->p_opptr != q->p_pptr) {
476 				struct proc *t = q->p_opptr;
477 				proc_reparent(q, t ? t : initproc);
478 				q->p_opptr = NULL;
479 			} else
480 				proc_reparent(q, initproc);
481 			killproc(q, "orphaned traced process");
482 		} else
483 			proc_reparent(q, initproc);
484 	}
485 
486 	/*
487 	 * Move proc from allproc to zombproc, it's now nearly ready to be
488 	 * collected by parent.
489 	 */
490 	mutex_enter(&proclist_mutex);
491 	LIST_REMOVE(l, l_list);
492 	LIST_REMOVE(p, p_list);
493 	LIST_INSERT_HEAD(&zombproc, p, p_list);
494 
495 	/*
496 	 * Mark the process as dead.  We must do this before we signal
497 	 * the parent.
498 	 */
499 	p->p_stat = SDEAD;
500 
501 	/* Put in front of parent's sibling list for parent to collect it */
502 	q = p->p_pptr;
503 	q->p_nstopchild++;
504 	if (LIST_FIRST(&q->p_children) != p) {
505 		/* Put child where it can be found quickly */
506 		LIST_REMOVE(p, p_sibling);
507 		LIST_INSERT_HEAD(&q->p_children, p, p_sibling);
508 	}
509 	mutex_exit(&proclist_mutex);
510 
511 	/*
512 	 * Notify parent that we're gone.  If parent has the P_NOCLDWAIT
513 	 * flag set, notify init instead (and hope it will handle
514 	 * this situation).
515 	 */
516 	mutex_enter(&q->p_mutex);
517 	if (q->p_flag & (PK_NOCLDWAIT|PK_CLDSIGIGN)) {
518 		proc_reparent(p, initproc);
519 		wakeinit = 1;
520 
521 		/*
522 		 * If this was the last child of our parent, notify
523 		 * parent, so in case he was wait(2)ing, he will
524 		 * continue.
525 		 */
526 		if (LIST_FIRST(&q->p_children) == NULL)
527 			cv_broadcast(&q->p_waitcv);
528 	}
529 	mutex_exit(&q->p_mutex);
530 
531 	/* Reload parent pointer, since p may have been reparented above */
532 	q = p->p_pptr;
533 
534 	if ((p->p_slflag & PSL_FSTRACE) == 0 && p->p_exitsig != 0) {
535 		exit_psignal(p, q, &ksi);
536 		mutex_enter(&proclist_mutex);
537 		kpsignal(q, &ksi, NULL);
538 		mutex_exit(&proclist_mutex);
539 	}
540 
541 	/* Calculate the final rusage info.  */
542 	calcru(p, &p->p_stats->p_ru.ru_utime, &p->p_stats->p_ru.ru_stime,
543 	    NULL, NULL);
544 
545 	if (wakeinit)
546 		cv_broadcast(&initproc->p_waitcv);
547 
548 	callout_destroy(&l->l_timeout_ch);
549 
550 	/*
551 	 * Remaining lwp resources will be freed in lwp_exit2() once we've
552 	 * switch to idle context; at that point, we will be marked as a
553 	 * full blown zombie.
554 	 *
555 	 * XXXSMP disable preemption.
556 	 */
557 	mutex_enter(&p->p_smutex);
558 	lwp_drainrefs(l);
559 	lwp_lock(l);
560 	l->l_prflag &= ~LPR_DETACHED;
561 	l->l_stat = LSZOMB;
562 	lwp_unlock(l);
563 	KASSERT(curlwp == l);
564 	KASSERT(p->p_nrlwps == 1);
565 	KASSERT(p->p_nlwps == 1);
566 	p->p_stat = SZOMB;
567 	p->p_nrlwps--;
568 	p->p_nzlwps++;
569 	p->p_ndlwps = 0;
570 	mutex_exit(&p->p_smutex);
571 
572 	/*
573 	 * Signal the parent to collect us, and drop the proclist lock.
574 	 * Drop debugger/procfs lock; no new references can be gained.
575 	 */
576 	cv_broadcast(&p->p_pptr->p_waitcv);
577 	mutex_exit(&proclist_lock);
578 	rw_exit(&p->p_reflock);
579 
580 	/* Verify that we hold no locks other than the kernel lock. */
581 #ifdef MULTIPROCESSOR
582 	LOCKDEBUG_BARRIER(&kernel_lock, 0);
583 #else
584 	LOCKDEBUG_BARRIER(NULL, 0);
585 #endif
586 
587 	/*
588 	 * NOTE: WE ARE NO LONGER ALLOWED TO SLEEP!
589 	 */
590 
591 	/*
592 	 * Give machine-dependent code a chance to free any MD LWP
593 	 * resources.  This must be done before uvm_lwp_exit(), in
594 	 * case these resources are in the PCB.
595 	 */
596 #ifndef __NO_CPU_LWP_FREE
597 	cpu_lwp_free(l, 1);
598 #endif
599 	pmap_deactivate(l);
600 
601 	/* This process no longer needs to hold the kernel lock. */
602 #ifdef notyet
603 	/* XXXSMP hold in lwp_userret() */
604 	KERNEL_UNLOCK_LAST(l);
605 #else
606 	KERNEL_UNLOCK_ALL(l, NULL);
607 #endif
608 
609 	lwp_exit_switchaway(l);
610 }
611 
612 void
613 exit_lwps(struct lwp *l)
614 {
615 	struct proc *p;
616 	struct lwp *l2;
617 	int error;
618 	lwpid_t waited;
619 #if defined(MULTIPROCESSOR)
620 	int nlocks;
621 #endif
622 
623 	KERNEL_UNLOCK_ALL(l, &nlocks);
624 
625 	p = l->l_proc;
626 	KASSERT(mutex_owned(&p->p_smutex));
627 
628  retry:
629 	/*
630 	 * Interrupt LWPs in interruptable sleep, unsuspend suspended
631 	 * LWPs and then wait for everyone else to finish.
632 	 */
633 	LIST_FOREACH(l2, &p->p_lwps, l_sibling) {
634 		if (l2 == l)
635 			continue;
636 		lwp_lock(l2);
637 		l2->l_flag |= LW_WEXIT;
638 		if ((l2->l_stat == LSSLEEP && (l2->l_flag & LW_SINTR)) ||
639 		    l2->l_stat == LSSUSPENDED || l2->l_stat == LSSTOP) {
640 		    	/* setrunnable() will release the lock. */
641 			setrunnable(l2);
642 			DPRINTF(("exit_lwps: Made %d.%d runnable\n",
643 			    p->p_pid, l2->l_lid));
644 			continue;
645 		}
646 		lwp_unlock(l2);
647 	}
648 	while (p->p_nlwps > 1) {
649 		DPRINTF(("exit_lwps: waiting for %d LWPs (%d zombies)\n",
650 		    p->p_nlwps, p->p_nzlwps));
651 		error = lwp_wait1(l, 0, &waited, LWPWAIT_EXITCONTROL);
652 		if (p->p_nlwps == 1)
653 			break;
654 		if (error == EDEADLK) {
655 			/*
656 			 * LWPs can get suspended/slept behind us.
657 			 * (eg. sa_setwoken)
658 			 * kick them again and retry.
659 			 */
660 			goto retry;
661 		}
662 		if (error)
663 			panic("exit_lwps: lwp_wait1 failed with error %d",
664 			    error);
665 		DPRINTF(("exit_lwps: Got LWP %d from lwp_wait1()\n", waited));
666 	}
667 
668 #if defined(MULTIPROCESSOR)
669 	if (nlocks > 0) {
670 		mutex_exit(&p->p_smutex);
671 		KERNEL_LOCK(nlocks, l);
672 		mutex_enter(&p->p_smutex);
673 	}
674 #endif /* defined(MULTIPROCESSOR) */
675 	KASSERT(p->p_nlwps == 1);
676 }
677 
678 int
679 do_sys_wait(struct lwp *l, int *pid, int *status, int options,
680     struct rusage *ru, int *was_zombie)
681 {
682 	struct proc	*child;
683 	int		error;
684 
685 	mutex_enter(&proclist_lock);
686 
687 	error = find_stopped_child(l->l_proc, *pid, options, &child, status);
688 
689 	if (child == NULL) {
690 		mutex_exit(&proclist_lock);
691 		*pid = 0;
692 		return error;
693 	}
694 
695 	*pid = child->p_pid;
696 
697 	if (child->p_stat == SZOMB) {
698 		/* proc_free() will release the proclist_lock. */
699 		*was_zombie = 1;
700 		if (options & WNOWAIT)
701 			mutex_exit(&proclist_lock);
702 		else {
703 			KERNEL_LOCK(1, l);		/* XXXSMP */
704 			proc_free(child, ru);
705 			KERNEL_UNLOCK_ONE(l);		/* XXXSMP */
706 		}
707 	} else {
708 		/* Child state must have been SSTOP. */
709 		*was_zombie = 0;
710 		mutex_exit(&proclist_lock);
711 		*status = W_STOPCODE(*status);
712 	}
713 
714 	return 0;
715 }
716 
717 int
718 sys_wait4(struct lwp *l, void *v, register_t *retval)
719 {
720 	struct sys_wait4_args /* {
721 		syscallarg(int)			pid;
722 		syscallarg(int *)		status;
723 		syscallarg(int)			options;
724 		syscallarg(struct rusage *)	rusage;
725 	} */ *uap = v;
726 	int		status, error;
727 	int		was_zombie;
728 	struct rusage	ru;
729 
730 	error = do_sys_wait(l, &SCARG(uap, pid), &status, SCARG(uap, options),
731 	    SCARG(uap, rusage) != NULL ? &ru : NULL, &was_zombie);
732 
733 	retval[0] = SCARG(uap, pid);
734 	if (SCARG(uap, pid) == 0)
735 		return error;
736 
737 	if (SCARG(uap, rusage))
738 		error = copyout(&ru, SCARG(uap, rusage), sizeof(ru));
739 
740 	if (error == 0 && SCARG(uap, status))
741 		error = copyout(&status, SCARG(uap, status), sizeof(status));
742 
743 	return error;
744 }
745 
746 /*
747  * Scan list of child processes for a child process that has stopped or
748  * exited.  Used by sys_wait4 and 'compat' equivalents.
749  *
750  * Must be called with the proclist_lock held, and may release
751  * while waiting.
752  */
753 static int
754 find_stopped_child(struct proc *parent, pid_t pid, int options,
755 		   struct proc **child_p, int *status_p)
756 {
757 	struct proc *child, *dead;
758 	int error;
759 
760 	KASSERT(mutex_owned(&proclist_lock));
761 
762 	if (options & ~(WUNTRACED|WNOHANG|WALTSIG|WALLSIG)
763 	    && !(options & WOPTSCHECKED)) {
764 		*child_p = NULL;
765 		return EINVAL;
766 	}
767 
768 	if (pid == 0 && !(options & WOPTSCHECKED))
769 		pid = -parent->p_pgid;
770 
771 	for (;;) {
772 		error = ECHILD;
773 		dead = NULL;
774 
775 		mutex_enter(&proclist_mutex);
776 		LIST_FOREACH(child, &parent->p_children, p_sibling) {
777 			if (pid >= 0) {
778 				if (child->p_pid != pid) {
779 					child = p_find(pid, PFIND_ZOMBIE |
780 					    PFIND_LOCKED);
781 					if (child == NULL ||
782 					    child->p_pptr != parent) {
783 						child = NULL;
784 						break;
785 					}
786 				}
787 			} else if (pid != WAIT_ANY && child->p_pgid != -pid) {
788 				/* Child not in correct pgrp */
789 				continue;
790 			}
791 
792 			/*
793 			 * Wait for processes with p_exitsig != SIGCHLD
794 			 * processes only if WALTSIG is set; wait for
795 			 * processes with p_exitsig == SIGCHLD only
796 			 * if WALTSIG is clear.
797 			 */
798 			if (((options & WALLSIG) == 0) &&
799 			    (options & WALTSIG ? child->p_exitsig == SIGCHLD
800 						: P_EXITSIG(child) != SIGCHLD)){
801 				if (child->p_pid == pid) {
802 					child = NULL;
803 					break;
804 				}
805 				continue;
806 			}
807 
808 			error = 0;
809 			if ((options & WNOZOMBIE) == 0) {
810 				if (child->p_stat == SZOMB)
811 					break;
812 				if (child->p_stat == SDEAD) {
813 					/*
814 					 * We may occasionally arrive here
815 					 * after receiving a signal, but
816 					 * immediatley before the child
817 					 * process is zombified.  The wait
818 					 * will be short, so avoid returning
819 					 * to userspace.
820 					 */
821 					dead = child;
822 				}
823 			}
824 
825 			if (child->p_stat == SSTOP &&
826 			    child->p_waited == 0 &&
827 			    (child->p_slflag & PSL_TRACED ||
828 			    options & WUNTRACED)) {
829 				if ((options & WNOWAIT) == 0) {
830 					child->p_waited = 1;
831 					parent->p_nstopchild--;
832 				}
833 				break;
834 			}
835 			if (parent->p_nstopchild == 0 || child->p_pid == pid) {
836 				child = NULL;
837 				break;
838 			}
839 		}
840 
841 		if (child != NULL || error != 0 ||
842 		    ((options & WNOHANG) != 0 && dead == NULL)) {
843 		    	if (child != NULL) {
844 			    	*status_p = child->p_xstat;
845 			}
846 			mutex_exit(&proclist_mutex);
847 			*child_p = child;
848 			return error;
849 		}
850 
851 		/*
852 		 * Wait for another child process to stop.
853 		 */
854 		mutex_exit(&proclist_lock);
855 		error = cv_wait_sig(&parent->p_waitcv, &proclist_mutex);
856 		mutex_exit(&proclist_mutex);
857 		mutex_enter(&proclist_lock);
858 
859 		if (error != 0) {
860 			*child_p = NULL;
861 			return error;
862 		}
863 	}
864 }
865 
866 /*
867  * Free a process after parent has taken all the state info.  Must be called
868  * with the proclist lock held, and will release before returning.
869  *
870  * *ru is returned to the caller, and must be freed by the caller.
871  */
872 static void
873 proc_free(struct proc *p, struct rusage *ru)
874 {
875 	struct proc *parent;
876 	struct lwp *l;
877 	ksiginfo_t ksi;
878 	kauth_cred_t cred1, cred2;
879 	uid_t uid;
880 
881 	KASSERT(mutex_owned(&proclist_lock));
882 	KASSERT(p->p_nlwps == 1);
883 	KASSERT(p->p_nzlwps == 1);
884 	KASSERT(p->p_nrlwps == 0);
885 	KASSERT(p->p_stat == SZOMB);
886 
887 	/*
888 	 * If we got the child via ptrace(2) or procfs, and
889 	 * the parent is different (meaning the process was
890 	 * attached, rather than run as a child), then we need
891 	 * to give it back to the old parent, and send the
892 	 * parent the exit signal.  The rest of the cleanup
893 	 * will be done when the old parent waits on the child.
894 	 */
895 	if ((p->p_slflag & PSL_TRACED) != 0) {
896 		parent = p->p_pptr;
897 		if (p->p_opptr != parent){
898 			mutex_enter(&p->p_smutex);
899 			p->p_slflag &= ~(PSL_TRACED|PSL_FSTRACE|PSL_SYSCALL);
900 			mutex_exit(&p->p_smutex);
901 			parent = p->p_opptr;
902 			if (parent == NULL)
903 				parent = initproc;
904 			proc_reparent(p, parent);
905 			p->p_opptr = NULL;
906 			if (p->p_exitsig != 0) {
907 				exit_psignal(p, parent, &ksi);
908 				mutex_enter(&proclist_mutex);
909 				kpsignal(parent, &ksi, NULL);
910 				mutex_exit(&proclist_mutex);
911 			}
912 			cv_broadcast(&parent->p_waitcv);
913 			mutex_exit(&proclist_lock);
914 			return;
915 		}
916 	}
917 
918 	/*
919 	 * Finally finished with old proc entry.  Unlink it from its process
920 	 * group.
921 	 */
922 	leavepgrp(p);
923 
924 	parent = p->p_pptr;
925 	sched_proc_exit(parent, p);
926 	/*
927 	 * Add child times of exiting process onto its own times.
928 	 * This cannot be done any earlier else it might get done twice.
929 	 */
930 	ruadd(&p->p_stats->p_ru, &p->p_stats->p_cru);
931 	ruadd(&parent->p_stats->p_cru, &p->p_stats->p_ru);
932 	if (ru != NULL)
933 		*ru = p->p_stats->p_ru;
934 	p->p_xstat = 0;
935 
936 	/*
937 	 * At this point we are going to start freeing the final resources.
938 	 * If anyone tries to access the proc structure after here they will
939 	 * get a shock - bits are missing.  Attempt to make it hard!  We
940 	 * don't bother with any further locking past this point.
941 	 */
942 	mutex_enter(&proclist_mutex);
943 	p->p_stat = SIDL;		/* not even a zombie any more */
944 	LIST_REMOVE(p, p_list);	/* off zombproc */
945 	parent = p->p_pptr;
946 	p->p_pptr->p_nstopchild--;
947 	mutex_exit(&proclist_mutex);
948 	LIST_REMOVE(p, p_sibling);
949 
950 	/*
951 	 * Let pid be reallocated.
952 	 */
953 	proc_free_pid(p);
954 	mutex_exit(&proclist_lock);
955 
956 	l = LIST_FIRST(&p->p_lwps);
957 
958 	/*
959 	 * Delay release until after lwp_free.
960 	 */
961 	cred2 = l->l_cred;
962 
963 	/*
964 	 * Free the last LWP's resources.
965 	 *
966 	 * lwp_free ensures the LWP is no longer running on another CPU.
967 	 */
968 	lwp_free(l, false, true);
969 
970 	/*
971 	 * Now no one except us can reach the process p.
972 	 */
973 
974 	/*
975 	 * Decrement the count of procs running with this uid.
976 	 */
977 	cred1 = p->p_cred;
978 	uid = kauth_cred_getuid(cred1);
979 	(void)chgproccnt(uid, -1);
980 
981 	/*
982 	 * Release substructures.
983 	 */
984 
985 	limfree(p->p_limit);
986 	pstatsfree(p->p_stats);
987 	kauth_cred_free(cred1);
988 	kauth_cred_free(cred2);
989 
990 	/*
991 	 * Release reference to text vnode
992 	 */
993 	if (p->p_textvp)
994 		vrele(p->p_textvp);
995 
996 	mutex_destroy(&p->p_raslock);
997 	mutex_destroy(&p->p_mutex);
998 	mutex_destroy(&p->p_stmutex);
999 	mutex_destroy(&p->p_smutex);
1000 	cv_destroy(&p->p_waitcv);
1001 	cv_destroy(&p->p_lwpcv);
1002 	rw_destroy(&p->p_reflock);
1003 
1004 	pool_put(&proc_pool, p);
1005 
1006 	/*
1007 	 * Collect child u-areas.
1008 	 */
1009 	uvm_uarea_drain(false);
1010 }
1011 
1012 /*
1013  * make process 'parent' the new parent of process 'child'.
1014  *
1015  * Must be called with proclist_lock lock held.
1016  */
1017 void
1018 proc_reparent(struct proc *child, struct proc *parent)
1019 {
1020 
1021 	KASSERT(mutex_owned(&proclist_lock));
1022 
1023 	if (child->p_pptr == parent)
1024 		return;
1025 
1026 	mutex_enter(&proclist_mutex);
1027 	if (child->p_stat == SZOMB ||
1028 	    (child->p_stat == SSTOP && !child->p_waited)) {
1029 		child->p_pptr->p_nstopchild--;
1030 		parent->p_nstopchild++;
1031 	}
1032 	mutex_exit(&proclist_mutex);
1033 	if (parent == initproc)
1034 		child->p_exitsig = SIGCHLD;
1035 
1036 	LIST_REMOVE(child, p_sibling);
1037 	LIST_INSERT_HEAD(&parent->p_children, child, p_sibling);
1038 	child->p_pptr = parent;
1039 }
1040