xref: /netbsd-src/sys/kern/kern_exit.c (revision 3feacbccbd4c8b6fd5f19c8e34474c4c3f8923d7)
1 /*	$NetBSD: kern_exit.c,v 1.248 2015/10/13 06:47:21 pgoyette 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 1.248 2015/10/13 06:47:21 pgoyette Exp $");
71 
72 #include "opt_ktrace.h"
73 #include "opt_dtrace.h"
74 #include "opt_perfctrs.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/mount.h>
102 #include <sys/syscallargs.h>
103 #include <sys/kauth.h>
104 #include <sys/sleepq.h>
105 #include <sys/lockdebug.h>
106 #include <sys/ktrace.h>
107 #include <sys/cpu.h>
108 #include <sys/lwpctl.h>
109 #include <sys/atomic.h>
110 #include <sys/sdt.h>
111 
112 #include <uvm/uvm_extern.h>
113 
114 #ifdef DEBUG_EXIT
115 int debug_exit = 0;
116 #define DPRINTF(x) if (debug_exit) printf x
117 #else
118 #define DPRINTF(x)
119 #endif
120 
121 static int find_stopped_child(struct proc *, pid_t, int, struct proc **, int *);
122 static void proc_free(struct proc *, struct rusage *);
123 
124 /*
125  * DTrace SDT provider definitions
126  */
127 SDT_PROVIDER_DECLARE(proc);
128 SDT_PROBE_DEFINE1(proc, kernel, , exit, "int");
129 
130 /*
131  * Fill in the appropriate signal information, and signal the parent.
132  */
133 /* XXX noclone works around a gcc 4.5 bug on arm */
134 static void __noclone
135 exit_psignal(struct proc *p, struct proc *pp, ksiginfo_t *ksi)
136 {
137 
138 	KSI_INIT(ksi);
139 	if ((ksi->ksi_signo = P_EXITSIG(p)) == SIGCHLD) {
140 		if (WIFSIGNALED(p->p_xstat)) {
141 			if (WCOREDUMP(p->p_xstat))
142 				ksi->ksi_code = CLD_DUMPED;
143 			else
144 				ksi->ksi_code = CLD_KILLED;
145 		} else {
146 			ksi->ksi_code = CLD_EXITED;
147 		}
148 	}
149 	/*
150 	 * We fill those in, even for non-SIGCHLD.
151 	 * It's safe to access p->p_cred unlocked here.
152 	 */
153 	ksi->ksi_pid = p->p_pid;
154 	ksi->ksi_uid = kauth_cred_geteuid(p->p_cred);
155 	ksi->ksi_status = p->p_xstat;
156 	/* XXX: is this still valid? */
157 	ksi->ksi_utime = p->p_stats->p_ru.ru_utime.tv_sec;
158 	ksi->ksi_stime = p->p_stats->p_ru.ru_stime.tv_sec;
159 }
160 
161 /*
162  * exit --
163  *	Death of process.
164  */
165 int
166 sys_exit(struct lwp *l, const struct sys_exit_args *uap, register_t *retval)
167 {
168 	/* {
169 		syscallarg(int)	rval;
170 	} */
171 	struct proc *p = l->l_proc;
172 
173 	/* Don't call exit1() multiple times in the same process. */
174 	mutex_enter(p->p_lock);
175 	if (p->p_sflag & PS_WEXIT) {
176 		mutex_exit(p->p_lock);
177 		lwp_exit(l);
178 	}
179 
180 	/* exit1() will release the mutex. */
181 	exit1(l, W_EXITCODE(SCARG(uap, rval), 0));
182 	/* NOTREACHED */
183 	return (0);
184 }
185 
186 /*
187  * Exit: deallocate address space and other resources, change proc state
188  * to zombie, and unlink proc from allproc and parent's lists.  Save exit
189  * status and rusage for wait().  Check for child processes and orphan them.
190  *
191  * Must be called with p->p_lock held.  Does not return.
192  */
193 void
194 exit1(struct lwp *l, int rv)
195 {
196 	struct proc	*p, *child, *next_child, *old_parent, *new_parent;
197 	struct pgrp	*pgrp;
198 	ksiginfo_t	ksi;
199 	ksiginfoq_t	kq;
200 	int		wakeinit;
201 
202 	p = l->l_proc;
203 
204 	KASSERT(mutex_owned(p->p_lock));
205 	KASSERT(p->p_vmspace != NULL);
206 
207 	if (__predict_false(p == initproc)) {
208 		panic("init died (signal %d, exit %d)",
209 		    WTERMSIG(rv), WEXITSTATUS(rv));
210 	}
211 
212 	p->p_sflag |= PS_WEXIT;
213 
214 	/*
215 	 * Force all other LWPs to exit before we do.  Only then can we
216 	 * begin to tear down the rest of the process state.
217 	 */
218 	if (p->p_nlwps > 1) {
219 		exit_lwps(l);
220 	}
221 
222 	ksiginfo_queue_init(&kq);
223 
224 	/*
225 	 * If we have been asked to stop on exit, do so now.
226 	 */
227 	if (__predict_false(p->p_sflag & PS_STOPEXIT)) {
228 		KERNEL_UNLOCK_ALL(l, &l->l_biglocks);
229 		sigclearall(p, &contsigmask, &kq);
230 
231 		if (!mutex_tryenter(proc_lock)) {
232 			mutex_exit(p->p_lock);
233 			mutex_enter(proc_lock);
234 			mutex_enter(p->p_lock);
235 		}
236 		p->p_waited = 0;
237 		p->p_pptr->p_nstopchild++;
238 		p->p_stat = SSTOP;
239 		mutex_exit(proc_lock);
240 		lwp_lock(l);
241 		p->p_nrlwps--;
242 		l->l_stat = LSSTOP;
243 		lwp_unlock(l);
244 		mutex_exit(p->p_lock);
245 		lwp_lock(l);
246 		mi_switch(l);
247 		KERNEL_LOCK(l->l_biglocks, l);
248 		mutex_enter(p->p_lock);
249 	}
250 
251 	/*
252 	 * Bin any remaining signals and mark the process as dying so it will
253 	 * not be found for, e.g. signals.
254 	 */
255 	sigfillset(&p->p_sigctx.ps_sigignore);
256 	sigclearall(p, NULL, &kq);
257 	p->p_stat = SDYING;
258 	mutex_exit(p->p_lock);
259 	ksiginfo_queue_drain(&kq);
260 
261 	/* Destroy any lwpctl info. */
262 	if (p->p_lwpctl != NULL)
263 		lwp_ctl_exit();
264 
265 	/*
266 	 * Drain all remaining references that procfs, ptrace and others may
267 	 * have on the process.
268 	 */
269 	rw_enter(&p->p_reflock, RW_WRITER);
270 
271 	DPRINTF(("exit1: %d.%d exiting.\n", p->p_pid, l->l_lid));
272 
273 	timers_free(p, TIMERS_ALL);
274 #if defined(__HAVE_RAS)
275 	ras_purgeall();
276 #endif
277 
278 	/*
279 	 * Close open files, release open-file table and free signal
280 	 * actions.  This may block!
281 	 */
282 	fd_free();
283 	cwdfree(p->p_cwdi);
284 	p->p_cwdi = NULL;
285 	doexithooks(p);
286 	sigactsfree(p->p_sigacts);
287 
288 	/*
289 	 * Write out accounting data.
290 	 */
291 	(void)acct_process(l);
292 
293 #ifdef KTRACE
294 	/*
295 	 * Release trace file.
296 	 */
297 	if (p->p_tracep != NULL) {
298 		mutex_enter(&ktrace_lock);
299 		ktrderef(p);
300 		mutex_exit(&ktrace_lock);
301 	}
302 #endif
303 
304 	/*
305 	 * If emulation has process exit hook, call it now.
306 	 * Set the exit status now so that the exit hook has
307 	 * an opportunity to tweak it (COMPAT_LINUX requires
308 	 * this for thread group emulation)
309 	 */
310 	p->p_xstat = rv;
311 	if (p->p_emul->e_proc_exit)
312 		(*p->p_emul->e_proc_exit)(p);
313 
314 	/*
315 	 * Free the VM resources we're still holding on to.
316 	 * We must do this from a valid thread because doing
317 	 * so may block. This frees vmspace, which we don't
318 	 * need anymore. The only remaining lwp is the one
319 	 * we run at this moment, nothing runs in userland
320 	 * anymore.
321 	 */
322 	uvm_proc_exit(p);
323 
324 	/*
325 	 * Stop profiling.
326 	 */
327 	if (__predict_false((p->p_stflag & PST_PROFIL) != 0)) {
328 		mutex_spin_enter(&p->p_stmutex);
329 		stopprofclock(p);
330 		mutex_spin_exit(&p->p_stmutex);
331 	}
332 
333 	/*
334 	 * If parent is waiting for us to exit or exec, PL_PPWAIT is set; we
335 	 * wake up the parent early to avoid deadlock.  We can do this once
336 	 * the VM resources are released.
337 	 */
338 	mutex_enter(proc_lock);
339 	if (p->p_lflag & PL_PPWAIT) {
340 #if 0
341 		lwp_t *lp;
342 
343 		l->l_lwpctl = NULL; /* was on loan from blocked parent */
344 		p->p_lflag &= ~PL_PPWAIT;
345 
346 		lp = p->p_vforklwp;
347 		p->p_vforklwp = NULL;
348 		lp->l_pflag &= ~LP_VFORKWAIT; /* XXX */
349 		cv_broadcast(&lp->l_waitcv);
350 #else
351 		l->l_lwpctl = NULL; /* was on loan from blocked parent */
352 		p->p_lflag &= ~PL_PPWAIT;
353 		cv_broadcast(&p->p_pptr->p_waitcv);
354 #endif
355 	}
356 
357 	if (SESS_LEADER(p)) {
358 		struct vnode *vprele = NULL, *vprevoke = NULL;
359 		struct session *sp = p->p_session;
360 		struct tty *tp;
361 
362 		if (sp->s_ttyvp) {
363 			/*
364 			 * Controlling process.
365 			 * Signal foreground pgrp,
366 			 * drain controlling terminal
367 			 * and revoke access to controlling terminal.
368 			 */
369 			tp = sp->s_ttyp;
370 			mutex_spin_enter(&tty_lock);
371 			if (tp->t_session == sp) {
372 				/* we can't guarantee the revoke will do this */
373 				pgrp = tp->t_pgrp;
374 				tp->t_pgrp = NULL;
375 				tp->t_session = NULL;
376 				mutex_spin_exit(&tty_lock);
377 				if (pgrp != NULL) {
378 					pgsignal(pgrp, SIGHUP, 1);
379 				}
380 				mutex_exit(proc_lock);
381 				(void) ttywait(tp);
382 				mutex_enter(proc_lock);
383 
384 				/* The tty could have been revoked. */
385 				vprevoke = sp->s_ttyvp;
386 			} else
387 				mutex_spin_exit(&tty_lock);
388 			vprele = sp->s_ttyvp;
389 			sp->s_ttyvp = NULL;
390 			/*
391 			 * s_ttyp is not zero'd; we use this to indicate
392 			 * that the session once had a controlling terminal.
393 			 * (for logging and informational purposes)
394 			 */
395 		}
396 		sp->s_leader = NULL;
397 
398 		if (vprevoke != NULL || vprele != NULL) {
399 			if (vprevoke != NULL) {
400 				/* Releases proc_lock. */
401 				proc_sessrele(sp);
402 				VOP_REVOKE(vprevoke, REVOKEALL);
403 			} else
404 				mutex_exit(proc_lock);
405 			if (vprele != NULL)
406 				vrele(vprele);
407 			mutex_enter(proc_lock);
408 		}
409 	}
410 	fixjobc(p, p->p_pgrp, 0);
411 
412 	/*
413 	 * Finalize the last LWP's specificdata, as well as the
414 	 * specificdata for the proc itself.
415 	 */
416 	lwp_finispecific(l);
417 	proc_finispecific(p);
418 
419 	/*
420 	 * Notify interested parties of our demise.
421 	 */
422 	KNOTE(&p->p_klist, NOTE_EXIT);
423 
424 	SDT_PROBE(proc, kernel, , exit,
425 		(WCOREDUMP(rv) ? CLD_DUMPED :
426 		 (WIFSIGNALED(rv) ? CLD_KILLED : CLD_EXITED)),
427 		0,0,0,0);
428 
429 #if PERFCTRS
430 	/*
431 	 * Save final PMC information in parent process & clean up.
432 	 */
433 	if (PMC_ENABLED(p)) {
434 		pmc_save_context(p);
435 		pmc_accumulate(p->p_pptr, p);
436 		pmc_process_exit(p);
437 	}
438 #endif
439 
440 	/*
441 	 * Reset p_opptr pointer of all former children which got
442 	 * traced by another process and were reparented. We reset
443 	 * it to NULL here; the trace detach code then reparents
444 	 * the child to initproc. We only check allproc list, since
445 	 * eventual former children on zombproc list won't reference
446 	 * p_opptr anymore.
447 	 */
448 	if (__predict_false(p->p_slflag & PSL_CHTRACED)) {
449 		struct proc *q;
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 	child = LIST_FIRST(&p->p_children);
460 	wakeinit = (child != NULL);
461 	for (; child != NULL; child = next_child) {
462 		next_child = LIST_NEXT(child, 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 (__predict_false(child->p_slflag & PSL_TRACED)) {
472 			mutex_enter(p->p_lock);
473 			child->p_slflag &=
474 			    ~(PSL_TRACED|PSL_FSTRACE|PSL_SYSCALL);
475 			mutex_exit(p->p_lock);
476 			if (child->p_opptr != child->p_pptr) {
477 				struct proc *t = child->p_opptr;
478 				proc_reparent(child, t ? t : initproc);
479 				child->p_opptr = NULL;
480 			} else
481 				proc_reparent(child, initproc);
482 			killproc(child, "orphaned traced process");
483 		} else
484 			proc_reparent(child, initproc);
485 	}
486 
487 	/*
488 	 * Move proc from allproc to zombproc, it's now nearly ready to be
489 	 * collected by parent.
490 	 */
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 	old_parent = p->p_pptr;
503 	old_parent->p_nstopchild++;
504 	if (LIST_FIRST(&old_parent->p_children) != p) {
505 		/* Put child where it can be found quickly */
506 		LIST_REMOVE(p, p_sibling);
507 		LIST_INSERT_HEAD(&old_parent->p_children, p, p_sibling);
508 	}
509 
510 	/*
511 	 * Notify parent that we're gone.  If parent has the P_NOCLDWAIT
512 	 * flag set, notify init instead (and hope it will handle
513 	 * this situation).
514 	 */
515 	if (old_parent->p_flag & (PK_NOCLDWAIT|PK_CLDSIGIGN)) {
516 		proc_reparent(p, initproc);
517 		wakeinit = 1;
518 
519 		/*
520 		 * If this was the last child of our parent, notify
521 		 * parent, so in case he was wait(2)ing, he will
522 		 * continue.
523 		 */
524 		if (LIST_FIRST(&old_parent->p_children) == NULL)
525 			cv_broadcast(&old_parent->p_waitcv);
526 	}
527 
528 	/* Reload parent pointer, since p may have been reparented above */
529 	new_parent = p->p_pptr;
530 
531 	if (__predict_false((p->p_slflag & PSL_FSTRACE) == 0 &&
532 	    p->p_exitsig != 0)) {
533 		exit_psignal(p, new_parent, &ksi);
534 		kpsignal(new_parent, &ksi, NULL);
535 	}
536 
537 	/* Calculate the final rusage info.  */
538 	calcru(p, &p->p_stats->p_ru.ru_utime, &p->p_stats->p_ru.ru_stime,
539 	    NULL, NULL);
540 
541 	if (wakeinit)
542 		cv_broadcast(&initproc->p_waitcv);
543 
544 	callout_destroy(&l->l_timeout_ch);
545 
546 	/*
547 	 * Release any PCU resources before becoming a zombie.
548 	 */
549 	pcu_discard_all(l);
550 
551 	mutex_enter(p->p_lock);
552 	/* Free the linux lwp id */
553 	if ((l->l_pflag & LP_PIDLID) != 0 && l->l_lid != p->p_pid)
554 		proc_free_pid(l->l_lid);
555 	lwp_drainrefs(l);
556 	lwp_lock(l);
557 	l->l_prflag &= ~LPR_DETACHED;
558 	l->l_stat = LSZOMB;
559 	lwp_unlock(l);
560 	KASSERT(curlwp == l);
561 	KASSERT(p->p_nrlwps == 1);
562 	KASSERT(p->p_nlwps == 1);
563 	p->p_stat = SZOMB;
564 	p->p_nrlwps--;
565 	p->p_nzlwps++;
566 	p->p_ndlwps = 0;
567 	mutex_exit(p->p_lock);
568 
569 	/*
570 	 * Signal the parent to collect us, and drop the proclist lock.
571 	 * Drop debugger/procfs lock; no new references can be gained.
572 	 */
573 	cv_broadcast(&p->p_pptr->p_waitcv);
574 	rw_exit(&p->p_reflock);
575 	mutex_exit(proc_lock);
576 
577 	/* Verify that we hold no locks other than the kernel lock. */
578 	LOCKDEBUG_BARRIER(&kernel_lock, 0);
579 
580 	/*
581 	 * NOTE: WE ARE NO LONGER ALLOWED TO SLEEP!
582 	 */
583 
584 	/*
585 	 * Give machine-dependent code a chance to free any MD LWP
586 	 * resources.  This must be done before uvm_lwp_exit(), in
587 	 * case these resources are in the PCB.
588 	 */
589 	cpu_lwp_free(l, 1);
590 
591 	pmap_deactivate(l);
592 
593 	/* This process no longer needs to hold the kernel lock. */
594 #ifdef notyet
595 	/* XXXSMP hold in lwp_userret() */
596 	KERNEL_UNLOCK_LAST(l);
597 #else
598 	KERNEL_UNLOCK_ALL(l, NULL);
599 #endif
600 
601 	lwp_exit_switchaway(l);
602 }
603 
604 void
605 exit_lwps(struct lwp *l)
606 {
607 	proc_t *p = l->l_proc;
608 	lwp_t *l2;
609 	int nlocks;
610 
611 	KERNEL_UNLOCK_ALL(l, &nlocks);
612 retry:
613 	KASSERT(mutex_owned(p->p_lock));
614 
615 	/*
616 	 * Interrupt LWPs in interruptable sleep, unsuspend suspended
617 	 * LWPs and then wait for everyone else to finish.
618 	 */
619 	LIST_FOREACH(l2, &p->p_lwps, l_sibling) {
620 		if (l2 == l)
621 			continue;
622 		lwp_lock(l2);
623 		l2->l_flag |= LW_WEXIT;
624 		if ((l2->l_stat == LSSLEEP && (l2->l_flag & LW_SINTR)) ||
625 		    l2->l_stat == LSSUSPENDED || l2->l_stat == LSSTOP) {
626 		    	/* setrunnable() will release the lock. */
627 			setrunnable(l2);
628 			continue;
629 		}
630 		lwp_unlock(l2);
631 	}
632 
633 	/*
634 	 * Wait for every LWP to exit.  Note: LWPs can get suspended/slept
635 	 * behind us or there may even be new LWPs created.  Therefore, a
636 	 * full retry is required on error.
637 	 */
638 	while (p->p_nlwps > 1) {
639 		if (lwp_wait(l, 0, NULL, true)) {
640 			goto retry;
641 		}
642 	}
643 
644 	KERNEL_LOCK(nlocks, l);
645 	KASSERT(p->p_nlwps == 1);
646 }
647 
648 int
649 do_sys_wait(int *pid, int *status, int options, struct rusage *ru)
650 {
651 	proc_t *child;
652 	int error;
653 
654 	if (ru != NULL) {
655 		memset(ru, 0, sizeof(*ru));
656 	}
657 	mutex_enter(proc_lock);
658 	error = find_stopped_child(curproc, *pid, options, &child, status);
659 	if (child == NULL) {
660 		mutex_exit(proc_lock);
661 		*pid = 0;
662 		return error;
663 	}
664 	*pid = child->p_pid;
665 
666 	if (child->p_stat == SZOMB) {
667 		/* proc_free() will release the proc_lock. */
668 		if (options & WNOWAIT) {
669 			mutex_exit(proc_lock);
670 		} else {
671 			proc_free(child, ru);
672 		}
673 	} else {
674 		/* Child state must have been SSTOP. */
675 		mutex_exit(proc_lock);
676 		*status = W_STOPCODE(*status);
677 	}
678 	return 0;
679 }
680 
681 int
682 sys___wait450(struct lwp *l, const struct sys___wait450_args *uap,
683     register_t *retval)
684 {
685 	/* {
686 		syscallarg(int)			pid;
687 		syscallarg(int *)		status;
688 		syscallarg(int)			options;
689 		syscallarg(struct rusage *)	rusage;
690 	} */
691 	int error, status, pid = SCARG(uap, pid);
692 	struct rusage ru;
693 
694 	error = do_sys_wait(&pid, &status, SCARG(uap, options),
695 	    SCARG(uap, rusage) != NULL ? &ru : NULL);
696 
697 	retval[0] = pid;
698 	if (pid == 0) {
699 		return error;
700 	}
701 	if (SCARG(uap, status)) {
702 		error = copyout(&status, SCARG(uap, status), sizeof(status));
703 	}
704 	if (SCARG(uap, rusage) && error == 0) {
705 		error = copyout(&ru, SCARG(uap, rusage), sizeof(ru));
706 	}
707 	return error;
708 }
709 
710 /*
711  * Scan list of child processes for a child process that has stopped or
712  * exited.  Used by sys_wait4 and 'compat' equivalents.
713  *
714  * Must be called with the proc_lock held, and may release while waiting.
715  */
716 static int
717 find_stopped_child(struct proc *parent, pid_t pid, int options,
718 		   struct proc **child_p, int *status_p)
719 {
720 	struct proc *child, *dead;
721 	int error;
722 
723 	KASSERT(mutex_owned(proc_lock));
724 
725 	if (options & ~(WUNTRACED|WNOHANG|WALTSIG|WALLSIG)
726 	    && !(options & WOPTSCHECKED)) {
727 		*child_p = NULL;
728 		return EINVAL;
729 	}
730 
731 	if (pid == 0 && !(options & WOPTSCHECKED))
732 		pid = -parent->p_pgid;
733 
734 	for (;;) {
735 		error = ECHILD;
736 		dead = NULL;
737 
738 		LIST_FOREACH(child, &parent->p_children, p_sibling) {
739 			if (pid >= 0) {
740 				if (child->p_pid != pid) {
741 					child = proc_find_raw(pid);
742 					if (child == NULL ||
743 					    child->p_stat == SIDL ||
744 					    child->p_pptr != parent) {
745 						child = NULL;
746 						break;
747 					}
748 				}
749 			} else if (pid != WAIT_ANY && child->p_pgid != -pid) {
750 				/* Child not in correct pgrp */
751 				continue;
752 			}
753 
754 			/*
755 			 * Wait for processes with p_exitsig != SIGCHLD
756 			 * processes only if WALTSIG is set; wait for
757 			 * processes with p_exitsig == SIGCHLD only
758 			 * if WALTSIG is clear.
759 			 */
760 			if (((options & WALLSIG) == 0) &&
761 			    (options & WALTSIG ? child->p_exitsig == SIGCHLD
762 						: P_EXITSIG(child) != SIGCHLD)){
763 				if (child->p_pid == pid) {
764 					child = NULL;
765 					break;
766 				}
767 				continue;
768 			}
769 
770 			error = 0;
771 			if ((options & WNOZOMBIE) == 0) {
772 				if (child->p_stat == SZOMB)
773 					break;
774 				if (child->p_stat == SDEAD) {
775 					/*
776 					 * We may occasionally arrive here
777 					 * after receiving a signal, but
778 					 * immediately before the child
779 					 * process is zombified.  The wait
780 					 * will be short, so avoid returning
781 					 * to userspace.
782 					 */
783 					dead = child;
784 				}
785 			}
786 
787 			if (child->p_stat == SSTOP &&
788 			    child->p_waited == 0 &&
789 			    (child->p_slflag & PSL_TRACED ||
790 			    options & WUNTRACED)) {
791 				if ((options & WNOWAIT) == 0) {
792 					child->p_waited = 1;
793 					parent->p_nstopchild--;
794 				}
795 				break;
796 			}
797 			if (parent->p_nstopchild == 0 || child->p_pid == pid) {
798 				child = NULL;
799 				break;
800 			}
801 		}
802 
803 		if (child != NULL || error != 0 ||
804 		    ((options & WNOHANG) != 0 && dead == NULL)) {
805 		    	if (child != NULL) {
806 			    	*status_p = child->p_xstat;
807 			}
808 			*child_p = child;
809 			return error;
810 		}
811 
812 		/*
813 		 * Wait for another child process to stop.
814 		 */
815 		error = cv_wait_sig(&parent->p_waitcv, proc_lock);
816 
817 		if (error != 0) {
818 			*child_p = NULL;
819 			return error;
820 		}
821 	}
822 }
823 
824 /*
825  * Free a process after parent has taken all the state info.  Must be called
826  * with the proclist lock held, and will release before returning.
827  *
828  * *ru is returned to the caller, and must be freed by the caller.
829  */
830 static void
831 proc_free(struct proc *p, struct rusage *ru)
832 {
833 	struct proc *parent = p->p_pptr;
834 	struct lwp *l;
835 	ksiginfo_t ksi;
836 	kauth_cred_t cred1, cred2;
837 	uid_t uid;
838 
839 	KASSERT(mutex_owned(proc_lock));
840 	KASSERT(p->p_nlwps == 1);
841 	KASSERT(p->p_nzlwps == 1);
842 	KASSERT(p->p_nrlwps == 0);
843 	KASSERT(p->p_stat == SZOMB);
844 
845 	/*
846 	 * If we got the child via ptrace(2) or procfs, and
847 	 * the parent is different (meaning the process was
848 	 * attached, rather than run as a child), then we need
849 	 * to give it back to the old parent, and send the
850 	 * parent the exit signal.  The rest of the cleanup
851 	 * will be done when the old parent waits on the child.
852 	 */
853 	if ((p->p_slflag & PSL_TRACED) != 0 && p->p_opptr != parent) {
854 		mutex_enter(p->p_lock);
855 		p->p_slflag &= ~(PSL_TRACED|PSL_FSTRACE|PSL_SYSCALL);
856 		mutex_exit(p->p_lock);
857 		parent = (p->p_opptr == NULL) ? initproc : p->p_opptr;
858 		proc_reparent(p, parent);
859 		p->p_opptr = NULL;
860 		if (p->p_exitsig != 0) {
861 			exit_psignal(p, parent, &ksi);
862 			kpsignal(parent, &ksi, NULL);
863 		}
864 		cv_broadcast(&parent->p_waitcv);
865 		mutex_exit(proc_lock);
866 		return;
867 	}
868 
869 	sched_proc_exit(parent, p);
870 
871 	/*
872 	 * Add child times of exiting process onto its own times.
873 	 * This cannot be done any earlier else it might get done twice.
874 	 */
875 	l = LIST_FIRST(&p->p_lwps);
876 	p->p_stats->p_ru.ru_nvcsw += (l->l_ncsw - l->l_nivcsw);
877 	p->p_stats->p_ru.ru_nivcsw += l->l_nivcsw;
878 	ruadd(&p->p_stats->p_ru, &l->l_ru);
879 	ruadd(&p->p_stats->p_ru, &p->p_stats->p_cru);
880 	ruadd(&parent->p_stats->p_cru, &p->p_stats->p_ru);
881 	if (ru != NULL)
882 		*ru = p->p_stats->p_ru;
883 	p->p_xstat = 0;
884 
885 	/*
886 	 * At this point we are going to start freeing the final resources.
887 	 * If anyone tries to access the proc structure after here they will
888 	 * get a shock - bits are missing.  Attempt to make it hard!  We
889 	 * don't bother with any further locking past this point.
890 	 */
891 	p->p_stat = SIDL;		/* not even a zombie any more */
892 	LIST_REMOVE(p, p_list);	/* off zombproc */
893 	parent->p_nstopchild--;
894 	LIST_REMOVE(p, p_sibling);
895 
896 	/*
897 	 * Let pid be reallocated.
898 	 */
899 	proc_free_pid(p->p_pid);
900 
901 	/*
902 	 * Unlink process from its process group.
903 	 * Releases the proc_lock.
904 	 */
905 	proc_leavepgrp(p);
906 
907 	/*
908 	 * Delay release until after lwp_free.
909 	 */
910 	cred2 = l->l_cred;
911 
912 	/*
913 	 * Free the last LWP's resources.
914 	 *
915 	 * lwp_free ensures the LWP is no longer running on another CPU.
916 	 */
917 	lwp_free(l, false, true);
918 
919 	/*
920 	 * Now no one except us can reach the process p.
921 	 */
922 
923 	/*
924 	 * Decrement the count of procs running with this uid.
925 	 */
926 	cred1 = p->p_cred;
927 	uid = kauth_cred_getuid(cred1);
928 	(void)chgproccnt(uid, -1);
929 
930 	/*
931 	 * Release substructures.
932 	 */
933 
934 	lim_free(p->p_limit);
935 	pstatsfree(p->p_stats);
936 	kauth_cred_free(cred1);
937 	kauth_cred_free(cred2);
938 
939 	/*
940 	 * Release reference to text vnode
941 	 */
942 	if (p->p_textvp)
943 		vrele(p->p_textvp);
944 
945 	mutex_destroy(&p->p_auxlock);
946 	mutex_obj_free(p->p_lock);
947 	mutex_destroy(&p->p_stmutex);
948 	cv_destroy(&p->p_waitcv);
949 	cv_destroy(&p->p_lwpcv);
950 	rw_destroy(&p->p_reflock);
951 
952 	proc_free_mem(p);
953 }
954 
955 /*
956  * make process 'parent' the new parent of process 'child'.
957  *
958  * Must be called with proc_lock held.
959  */
960 void
961 proc_reparent(struct proc *child, struct proc *parent)
962 {
963 
964 	KASSERT(mutex_owned(proc_lock));
965 
966 	if (child->p_pptr == parent)
967 		return;
968 
969 	if (child->p_stat == SZOMB || child->p_stat == SDEAD ||
970 	    (child->p_stat == SSTOP && !child->p_waited)) {
971 		child->p_pptr->p_nstopchild--;
972 		parent->p_nstopchild++;
973 	}
974 	if (parent == initproc)
975 		child->p_exitsig = SIGCHLD;
976 
977 	LIST_REMOVE(child, p_sibling);
978 	LIST_INSERT_HEAD(&parent->p_children, child, p_sibling);
979 	child->p_pptr = parent;
980 	child->p_ppid = parent->p_pid;
981 }
982