xref: /netbsd-src/sys/kern/kern_exit.c (revision b83ebeba7f767758d2778bb0f9d7a76534253621)
1 /*	$NetBSD: kern_exit.c,v 1.258 2016/04/27 21:15:40 christos 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 1.258 2016/04/27 21:15:40 christos 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 *, idtype_t, id_t, int,
122     struct proc **, struct wrusage *, siginfo_t *);
123 static void proc_free(struct proc *, struct wrusage *);
124 
125 /*
126  * DTrace SDT provider definitions
127  */
128 SDT_PROVIDER_DECLARE(proc);
129 SDT_PROBE_DEFINE1(proc, kernel, , exit, "int");
130 
131 /*
132  * Fill in the appropriate signal information, and signal the parent.
133  */
134 /* XXX noclone works around a gcc 4.5 bug on arm */
135 static void __noclone
136 exit_psignal(struct proc *p, struct proc *pp, ksiginfo_t *ksi)
137 {
138 
139 	KSI_INIT(ksi);
140 	if ((ksi->ksi_signo = P_EXITSIG(p)) == SIGCHLD) {
141 		if (p->p_xsig) {
142 			if (p->p_sflag & PS_COREDUMP)
143 				ksi->ksi_code = CLD_DUMPED;
144 			else
145 				ksi->ksi_code = CLD_KILLED;
146 			ksi->ksi_status = p->p_xsig;
147 		} else {
148 			ksi->ksi_code = CLD_EXITED;
149 			ksi->ksi_status = p->p_xexit;
150 		}
151 	} else {
152 		ksi->ksi_code = SI_USER;
153 		ksi->ksi_status = p->p_xsig;
154 	}
155 	/*
156 	 * We fill those in, even for non-SIGCHLD.
157 	 * It's safe to access p->p_cred unlocked here.
158 	 */
159 	ksi->ksi_pid = p->p_pid;
160 	ksi->ksi_uid = kauth_cred_geteuid(p->p_cred);
161 	/* XXX: is this still valid? */
162 	ksi->ksi_utime = p->p_stats->p_ru.ru_utime.tv_sec;
163 	ksi->ksi_stime = p->p_stats->p_ru.ru_stime.tv_sec;
164 }
165 
166 /*
167  * exit --
168  *	Death of process.
169  */
170 int
171 sys_exit(struct lwp *l, const struct sys_exit_args *uap, register_t *retval)
172 {
173 	/* {
174 		syscallarg(int)	rval;
175 	} */
176 	struct proc *p = l->l_proc;
177 
178 	/* Don't call exit1() multiple times in the same process. */
179 	mutex_enter(p->p_lock);
180 	if (p->p_sflag & PS_WEXIT) {
181 		mutex_exit(p->p_lock);
182 		lwp_exit(l);
183 	}
184 
185 	/* exit1() will release the mutex. */
186 	exit1(l, SCARG(uap, rval), 0);
187 	/* NOTREACHED */
188 	return (0);
189 }
190 
191 /*
192  * Exit: deallocate address space and other resources, change proc state
193  * to zombie, and unlink proc from allproc and parent's lists.  Save exit
194  * status and rusage for wait().  Check for child processes and orphan them.
195  *
196  * Must be called with p->p_lock held.  Does not return.
197  */
198 void
199 exit1(struct lwp *l, int exitcode, int signo)
200 {
201 	struct proc	*p, *child, *next_child, *old_parent, *new_parent;
202 	struct pgrp	*pgrp;
203 	ksiginfo_t	ksi;
204 	ksiginfoq_t	kq;
205 	int		wakeinit;
206 
207 	p = l->l_proc;
208 
209 	KASSERT(mutex_owned(p->p_lock));
210 	KASSERT(p->p_vmspace != NULL);
211 
212 	if (__predict_false(p == initproc)) {
213 		panic("init died (signal %d, exit %d)", signo, exitcode);
214 	}
215 
216 	p->p_sflag |= PS_WEXIT;
217 
218 	/*
219 	 * Force all other LWPs to exit before we do.  Only then can we
220 	 * begin to tear down the rest of the process state.
221 	 */
222 	if (p->p_nlwps > 1) {
223 		exit_lwps(l);
224 	}
225 
226 	ksiginfo_queue_init(&kq);
227 
228 	/*
229 	 * If we have been asked to stop on exit, do so now.
230 	 */
231 	if (__predict_false(p->p_sflag & PS_STOPEXIT)) {
232 		KERNEL_UNLOCK_ALL(l, &l->l_biglocks);
233 		sigclearall(p, &contsigmask, &kq);
234 
235 		if (!mutex_tryenter(proc_lock)) {
236 			mutex_exit(p->p_lock);
237 			mutex_enter(proc_lock);
238 			mutex_enter(p->p_lock);
239 		}
240 		p->p_waited = 0;
241 		p->p_pptr->p_nstopchild++;
242 		p->p_stat = SSTOP;
243 		mutex_exit(proc_lock);
244 		lwp_lock(l);
245 		p->p_nrlwps--;
246 		l->l_stat = LSSTOP;
247 		lwp_unlock(l);
248 		mutex_exit(p->p_lock);
249 		lwp_lock(l);
250 		mi_switch(l);
251 		KERNEL_LOCK(l->l_biglocks, l);
252 		mutex_enter(p->p_lock);
253 	}
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 	sigfillset(&p->p_sigctx.ps_sigignore);
260 	sigclearall(p, NULL, &kq);
261 	p->p_stat = SDYING;
262 	mutex_exit(p->p_lock);
263 	ksiginfo_queue_drain(&kq);
264 
265 	/* Destroy any lwpctl info. */
266 	if (p->p_lwpctl != NULL)
267 		lwp_ctl_exit();
268 
269 	/*
270 	 * Drain all remaining references that procfs, ptrace and others may
271 	 * have on the process.
272 	 */
273 	rw_enter(&p->p_reflock, RW_WRITER);
274 
275 	DPRINTF(("%s: %d.%d exiting.\n", __func__, p->p_pid, l->l_lid));
276 
277 	timers_free(p, TIMERS_ALL);
278 #if defined(__HAVE_RAS)
279 	ras_purgeall();
280 #endif
281 
282 	/*
283 	 * Close open files, release open-file table and free signal
284 	 * actions.  This may block!
285 	 */
286 	fd_free();
287 	cwdfree(p->p_cwdi);
288 	p->p_cwdi = NULL;
289 	doexithooks(p);
290 	sigactsfree(p->p_sigacts);
291 
292 	/*
293 	 * Write out accounting data.
294 	 */
295 	(void)acct_process(l);
296 
297 #ifdef KTRACE
298 	/*
299 	 * Release trace file.
300 	 */
301 	if (p->p_tracep != NULL) {
302 		mutex_enter(&ktrace_lock);
303 		ktrderef(p);
304 		mutex_exit(&ktrace_lock);
305 	}
306 #endif
307 
308 	p->p_xexit = exitcode;
309 	p->p_xsig = signo;
310 
311 	/*
312 	 * If emulation has process exit hook, call it now.
313 	 * Set the exit status now so that the exit hook has
314 	 * an opportunity to tweak it (COMPAT_LINUX requires
315 	 * this for thread group emulation)
316 	 */
317 	if (p->p_emul->e_proc_exit)
318 		(*p->p_emul->e_proc_exit)(p);
319 
320 	/*
321 	 * Free the VM resources we're still holding on to.
322 	 * We must do this from a valid thread because doing
323 	 * so may block. This frees vmspace, which we don't
324 	 * need anymore. The only remaining lwp is the one
325 	 * we run at this moment, nothing runs in userland
326 	 * anymore.
327 	 */
328 	uvm_proc_exit(p);
329 
330 	/*
331 	 * Stop profiling.
332 	 */
333 	if (__predict_false((p->p_stflag & PST_PROFIL) != 0)) {
334 		mutex_spin_enter(&p->p_stmutex);
335 		stopprofclock(p);
336 		mutex_spin_exit(&p->p_stmutex);
337 	}
338 
339 	/*
340 	 * If parent is waiting for us to exit or exec, PL_PPWAIT is set; we
341 	 * wake up the parent early to avoid deadlock.  We can do this once
342 	 * the VM resources are released.
343 	 */
344 	mutex_enter(proc_lock);
345 	if (p->p_lflag & PL_PPWAIT) {
346 #if 0
347 		lwp_t *lp;
348 
349 		l->l_lwpctl = NULL; /* was on loan from blocked parent */
350 		p->p_lflag &= ~PL_PPWAIT;
351 
352 		lp = p->p_vforklwp;
353 		p->p_vforklwp = NULL;
354 		lp->l_pflag &= ~LP_VFORKWAIT; /* XXX */
355 		cv_broadcast(&lp->l_waitcv);
356 #else
357 		l->l_lwpctl = NULL; /* was on loan from blocked parent */
358 		p->p_lflag &= ~PL_PPWAIT;
359 		cv_broadcast(&p->p_pptr->p_waitcv);
360 #endif
361 	}
362 
363 	if (SESS_LEADER(p)) {
364 		struct vnode *vprele = NULL, *vprevoke = NULL;
365 		struct session *sp = p->p_session;
366 		struct tty *tp;
367 
368 		if (sp->s_ttyvp) {
369 			/*
370 			 * Controlling process.
371 			 * Signal foreground pgrp,
372 			 * drain controlling terminal
373 			 * and revoke access to controlling terminal.
374 			 */
375 			tp = sp->s_ttyp;
376 			mutex_spin_enter(&tty_lock);
377 			if (tp->t_session == sp) {
378 				/* we can't guarantee the revoke will do this */
379 				pgrp = tp->t_pgrp;
380 				tp->t_pgrp = NULL;
381 				tp->t_session = NULL;
382 				mutex_spin_exit(&tty_lock);
383 				if (pgrp != NULL) {
384 					pgsignal(pgrp, SIGHUP, 1);
385 				}
386 				mutex_exit(proc_lock);
387 				(void) ttywait(tp);
388 				mutex_enter(proc_lock);
389 
390 				/* The tty could have been revoked. */
391 				vprevoke = sp->s_ttyvp;
392 			} else
393 				mutex_spin_exit(&tty_lock);
394 			vprele = sp->s_ttyvp;
395 			sp->s_ttyvp = NULL;
396 			/*
397 			 * s_ttyp is not zero'd; we use this to indicate
398 			 * that the session once had a controlling terminal.
399 			 * (for logging and informational purposes)
400 			 */
401 		}
402 		sp->s_leader = NULL;
403 
404 		if (vprevoke != NULL || vprele != NULL) {
405 			if (vprevoke != NULL) {
406 				/* Releases proc_lock. */
407 				proc_sessrele(sp);
408 				VOP_REVOKE(vprevoke, REVOKEALL);
409 			} else
410 				mutex_exit(proc_lock);
411 			if (vprele != NULL)
412 				vrele(vprele);
413 			mutex_enter(proc_lock);
414 		}
415 	}
416 	fixjobc(p, p->p_pgrp, 0);
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 	SDT_PROBE(proc, kernel, , exit,
431 		((p->p_sflag & PS_COREDUMP) ? CLD_DUMPED :
432 		 (p->p_xsig ? CLD_KILLED : CLD_EXITED)),
433 		0,0,0,0);
434 
435 #if PERFCTRS
436 	/*
437 	 * Save final PMC information in parent process & clean up.
438 	 */
439 	if (PMC_ENABLED(p)) {
440 		pmc_save_context(p);
441 		pmc_accumulate(p->p_pptr, p);
442 		pmc_process_exit(p);
443 	}
444 #endif
445 
446 	/*
447 	 * Reset p_opptr pointer of all former children which got
448 	 * traced by another process and were reparented. We reset
449 	 * it to NULL here; the trace detach code then reparents
450 	 * the child to initproc. We only check allproc list, since
451 	 * eventual former children on zombproc list won't reference
452 	 * p_opptr anymore.
453 	 */
454 	if (__predict_false(p->p_slflag & PSL_CHTRACED)) {
455 		struct proc *q;
456 		PROCLIST_FOREACH(q, &allproc) {
457 			if (q->p_opptr == p)
458 				q->p_opptr = NULL;
459 		}
460 	}
461 
462 	/*
463 	 * Give orphaned children to init(8).
464 	 */
465 	child = LIST_FIRST(&p->p_children);
466 	wakeinit = (child != NULL);
467 	for (; child != NULL; child = next_child) {
468 		next_child = LIST_NEXT(child, p_sibling);
469 
470 		/*
471 		 * Traced processes are killed since their existence
472 		 * means someone is screwing up. Since we reset the
473 		 * trace flags, the logic in sys_wait4() would not be
474 		 * triggered to reparent the process to its
475 		 * original parent, so we must do this here.
476 		 */
477 		if (__predict_false(child->p_slflag & PSL_TRACED)) {
478 			mutex_enter(p->p_lock);
479 			child->p_slflag &=
480 			    ~(PSL_TRACED|PSL_FSTRACE|PSL_SYSCALL);
481 			mutex_exit(p->p_lock);
482 			if (child->p_opptr != child->p_pptr) {
483 				struct proc *t = child->p_opptr;
484 				proc_reparent(child, t ? t : initproc);
485 				child->p_opptr = NULL;
486 			} else
487 				proc_reparent(child, initproc);
488 			killproc(child, "orphaned traced process");
489 		} else
490 			proc_reparent(child, initproc);
491 	}
492 
493 	/*
494 	 * Move proc from allproc to zombproc, it's now nearly ready to be
495 	 * collected by parent.
496 	 */
497 	LIST_REMOVE(l, l_list);
498 	LIST_REMOVE(p, p_list);
499 	LIST_INSERT_HEAD(&zombproc, p, p_list);
500 
501 	/*
502 	 * Mark the process as dead.  We must do this before we signal
503 	 * the parent.
504 	 */
505 	p->p_stat = SDEAD;
506 
507 	/* Put in front of parent's sibling list for parent to collect it */
508 	old_parent = p->p_pptr;
509 	old_parent->p_nstopchild++;
510 	if (LIST_FIRST(&old_parent->p_children) != p) {
511 		/* Put child where it can be found quickly */
512 		LIST_REMOVE(p, p_sibling);
513 		LIST_INSERT_HEAD(&old_parent->p_children, p, p_sibling);
514 	}
515 
516 	/*
517 	 * Notify parent that we're gone.  If parent has the P_NOCLDWAIT
518 	 * flag set, notify init instead (and hope it will handle
519 	 * this situation).
520 	 */
521 	if (old_parent->p_flag & (PK_NOCLDWAIT|PK_CLDSIGIGN)) {
522 		proc_reparent(p, initproc);
523 		wakeinit = 1;
524 
525 		/*
526 		 * If this was the last child of our parent, notify
527 		 * parent, so in case he was wait(2)ing, he will
528 		 * continue.
529 		 */
530 		if (LIST_FIRST(&old_parent->p_children) == NULL)
531 			cv_broadcast(&old_parent->p_waitcv);
532 	}
533 
534 	/* Reload parent pointer, since p may have been reparented above */
535 	new_parent = p->p_pptr;
536 
537 	if (__predict_false((p->p_slflag & PSL_FSTRACE) == 0 &&
538 	    p->p_exitsig != 0)) {
539 		exit_psignal(p, new_parent, &ksi);
540 		kpsignal(new_parent, &ksi, NULL);
541 	}
542 
543 	/* Calculate the final rusage info.  */
544 	calcru(p, &p->p_stats->p_ru.ru_utime, &p->p_stats->p_ru.ru_stime,
545 	    NULL, NULL);
546 
547 	if (wakeinit)
548 		cv_broadcast(&initproc->p_waitcv);
549 
550 	callout_destroy(&l->l_timeout_ch);
551 
552 	/*
553 	 * Release any PCU resources before becoming a zombie.
554 	 */
555 	pcu_discard_all(l);
556 
557 	mutex_enter(p->p_lock);
558 	/* Free the linux lwp id */
559 	if ((l->l_pflag & LP_PIDLID) != 0 && l->l_lid != p->p_pid)
560 		proc_free_pid(l->l_lid);
561 	lwp_drainrefs(l);
562 	lwp_lock(l);
563 	l->l_prflag &= ~LPR_DETACHED;
564 	l->l_stat = LSZOMB;
565 	lwp_unlock(l);
566 	KASSERT(curlwp == l);
567 	KASSERT(p->p_nrlwps == 1);
568 	KASSERT(p->p_nlwps == 1);
569 	p->p_stat = SZOMB;
570 	p->p_nrlwps--;
571 	p->p_nzlwps++;
572 	p->p_ndlwps = 0;
573 	mutex_exit(p->p_lock);
574 
575 	/*
576 	 * Signal the parent to collect us, and drop the proclist lock.
577 	 * Drop debugger/procfs lock; no new references can be gained.
578 	 */
579 	cv_broadcast(&p->p_pptr->p_waitcv);
580 	rw_exit(&p->p_reflock);
581 	mutex_exit(proc_lock);
582 
583 	/* Verify that we hold no locks other than the kernel lock. */
584 	LOCKDEBUG_BARRIER(&kernel_lock, 0);
585 
586 	/*
587 	 * NOTE: WE ARE NO LONGER ALLOWED TO SLEEP!
588 	 */
589 
590 	/*
591 	 * Give machine-dependent code a chance to free any MD LWP
592 	 * resources.  This must be done before uvm_lwp_exit(), in
593 	 * case these resources are in the PCB.
594 	 */
595 	cpu_lwp_free(l, 1);
596 
597 	pmap_deactivate(l);
598 
599 	/* This process no longer needs to hold the kernel lock. */
600 #ifdef notyet
601 	/* XXXSMP hold in lwp_userret() */
602 	KERNEL_UNLOCK_LAST(l);
603 #else
604 	KERNEL_UNLOCK_ALL(l, NULL);
605 #endif
606 
607 	lwp_exit_switchaway(l);
608 }
609 
610 void
611 exit_lwps(struct lwp *l)
612 {
613 	proc_t *p = l->l_proc;
614 	lwp_t *l2;
615 	int nlocks;
616 
617 	KERNEL_UNLOCK_ALL(l, &nlocks);
618 retry:
619 	KASSERT(mutex_owned(p->p_lock));
620 
621 	/*
622 	 * Interrupt LWPs in interruptable sleep, unsuspend suspended
623 	 * LWPs and then wait for everyone else to finish.
624 	 */
625 	LIST_FOREACH(l2, &p->p_lwps, l_sibling) {
626 		if (l2 == l)
627 			continue;
628 		lwp_lock(l2);
629 		l2->l_flag |= LW_WEXIT;
630 		if ((l2->l_stat == LSSLEEP && (l2->l_flag & LW_SINTR)) ||
631 		    l2->l_stat == LSSUSPENDED || l2->l_stat == LSSTOP) {
632 		    	/* setrunnable() will release the lock. */
633 			setrunnable(l2);
634 			continue;
635 		}
636 		lwp_unlock(l2);
637 	}
638 
639 	/*
640 	 * Wait for every LWP to exit.  Note: LWPs can get suspended/slept
641 	 * behind us or there may even be new LWPs created.  Therefore, a
642 	 * full retry is required on error.
643 	 */
644 	while (p->p_nlwps > 1) {
645 		if (lwp_wait(l, 0, NULL, true)) {
646 			goto retry;
647 		}
648 	}
649 
650 	KERNEL_LOCK(nlocks, l);
651 	KASSERT(p->p_nlwps == 1);
652 }
653 
654 static int
655 do_sys_waitid(idtype_t idtype, id_t id, int *pid, int *status, int options,
656     struct wrusage *wru, siginfo_t *si)
657 {
658 	proc_t *child;
659 	int error;
660 
661 
662 	if (wru != NULL)
663 		memset(wru, 0, sizeof(*wru));
664 	if (si != NULL)
665 		memset(si, 0, sizeof(*si));
666 
667 	mutex_enter(proc_lock);
668 	error = find_stopped_child(curproc, idtype, id, options, &child,
669 	    wru, si);
670 	if (child == NULL) {
671 		mutex_exit(proc_lock);
672 		*pid = 0;
673 		return error;
674 	}
675 	*pid = child->p_pid;
676 
677 	if (child->p_stat == SZOMB) {
678 		/* Child is exiting */
679 		*status = P_WAITSTATUS(child);
680 		/* proc_free() will release the proc_lock. */
681 		if (options & WNOWAIT) {
682 			mutex_exit(proc_lock);
683 		} else {
684 			proc_free(child, wru);
685 		}
686 	} else {
687 		/* Child state must have been SSTOP. */
688 		*status = child->p_xsig == SIGCONT ? W_CONTCODE() :
689 		    W_STOPCODE(child->p_xsig);
690 		mutex_exit(proc_lock);
691 	}
692 	return 0;
693 }
694 
695 int
696 do_sys_wait(int *pid, int *status, int options, struct rusage *ru)
697 {
698 	idtype_t idtype;
699 	id_t id;
700 	int ret;
701 	struct wrusage wru;
702 
703 	/*
704 	 * Translate the special pid values into the (idtype, pid)
705 	 * pair for wait6. The WAIT_MYPGRP case is handled by
706 	 * find_stopped_child() on its own.
707 	 */
708 	if (*pid == WAIT_ANY) {
709 		idtype = P_ALL;
710 		id = 0;
711 	} else if (*pid < 0) {
712 		idtype = P_PGID;
713 		id = (id_t)-*pid;
714 	} else {
715 		idtype = P_PID;
716 		id = (id_t)*pid;
717 	}
718 	options |= WEXITED | WTRAPPED;
719 	ret = do_sys_waitid(idtype, id, pid, status, options, ru ? &wru : NULL,
720 	    NULL);
721 	if (ru)
722 		*ru = wru.wru_self;
723 	return ret;
724 }
725 
726 int
727 sys___wait450(struct lwp *l, const struct sys___wait450_args *uap,
728     register_t *retval)
729 {
730 	/* {
731 		syscallarg(int)			pid;
732 		syscallarg(int *)		status;
733 		syscallarg(int)			options;
734 		syscallarg(struct rusage *)	rusage;
735 	} */
736 	int error, status, pid = SCARG(uap, pid);
737 	struct rusage ru;
738 
739 	error = do_sys_wait(&pid, &status, SCARG(uap, options),
740 	    SCARG(uap, rusage) != NULL ? &ru : NULL);
741 
742 	retval[0] = pid;
743 	if (pid == 0) {
744 		return error;
745 	}
746 	if (SCARG(uap, status)) {
747 		error = copyout(&status, SCARG(uap, status), sizeof(status));
748 	}
749 	if (SCARG(uap, rusage) && error == 0) {
750 		error = copyout(&ru, SCARG(uap, rusage), sizeof(ru));
751 	}
752 	return error;
753 }
754 
755 int
756 sys_wait6(struct lwp *l, const struct sys_wait6_args *uap, register_t *retval)
757 {
758 	/* {
759 		syscallarg(idtype_t)		idtype;
760 		syscallarg(id_t)		id;
761 		syscallarg(int *)		status;
762 		syscallarg(int)			options;
763 		syscallarg(struct wrusage *)	wru;
764 		syscallarg(siginfo_t *)		si;
765 	} */
766 	struct wrusage wru, *wrup;
767 	siginfo_t si, *sip;
768 	idtype_t idtype;
769 	int pid;
770 	id_t id;
771 	int error, status;
772 
773 	idtype = SCARG(uap, idtype);
774 	id = SCARG(uap, id);
775 
776 	if (SCARG(uap, wru) != NULL)
777 		wrup = &wru;
778 	else
779 		wrup = NULL;
780 
781 	if (SCARG(uap, info) != NULL)
782 		sip = &si;
783 	else
784 		sip = NULL;
785 
786 	/*
787 	 *  We expect all callers of wait6() to know about WEXITED and
788 	 *  WTRAPPED.
789 	 */
790 	error = do_sys_waitid(idtype, id, &pid, &status, SCARG(uap, options),
791 	    wrup, sip);
792 
793 	retval[0] = pid; 	/* tell userland who it was */
794 
795 #if 0
796 	/*
797 	 * should we copyout if there was no process, hence no useful data?
798 	 * We don't for an old sytle wait4() (etc) but I believe
799 	 * FreeBSD does for wait6(), so a tossup...  Go with FreeBSD for now.
800 	 */
801 	if (pid == 0)
802 		return error;
803 #endif
804 
805 	if (SCARG(uap, status) != NULL && error == 0)
806 		error = copyout(&status, SCARG(uap, status), sizeof(status));
807 	if (SCARG(uap, wru) != NULL && error == 0)
808 		error = copyout(&wru, SCARG(uap, wru), sizeof(wru));
809 	if (SCARG(uap, info) != NULL && error == 0)
810 		error = copyout(&si, SCARG(uap, info), sizeof(si));
811 	return error;
812 }
813 
814 
815 /*
816  * Find a process that matches the provided criteria, and fill siginfo
817  * and resources if found.
818  * Returns:
819  *	-1: 	Not found, abort early
820  *	 0:	Not matched
821  *	 1:	Matched, there might be more matches
822  *	 2:	This is the only match
823  */
824 static int
825 match_process(struct proc *pp, struct proc **q, idtype_t idtype, id_t id,
826     int options, struct wrusage *wrusage, siginfo_t *siginfo)
827 {
828 	struct rusage *rup;
829 	struct proc *p = *q;
830 	int rv = 1;
831 
832 	mutex_enter(p->p_lock);
833 	switch (idtype) {
834 	case P_ALL:
835 		break;
836 	case P_PID:
837 		if (p->p_pid != (pid_t)id) {
838 			mutex_exit(p->p_lock);
839 			p = *q = proc_find_raw((pid_t)id);
840 			if (p == NULL || p->p_stat == SIDL || p->p_pptr != pp) {
841 				*q = NULL;
842 				return -1;
843 			}
844 			mutex_enter(p->p_lock);
845 		}
846 		rv++;
847 		break;
848 	case P_PGID:
849 		if (p->p_pgid != (pid_t)id)
850 			goto out;
851 		break;
852 	case P_SID:
853 		if (p->p_session->s_sid != (pid_t)id)
854 			goto out;
855 		break;
856 	case P_UID:
857 		if (kauth_cred_geteuid(p->p_cred) != (uid_t)id)
858 			goto out;
859 		break;
860 	case P_GID:
861 		if (kauth_cred_getegid(p->p_cred) != (gid_t)id)
862 			goto out;
863 		break;
864 	case P_CID:
865 	case P_PSETID:
866 	case P_CPUID:
867 		/* XXX: Implement me */
868 	default:
869 	out:
870 		mutex_exit(p->p_lock);
871 		return 0;
872 	}
873 
874 	if ((options & WEXITED) == 0 && p->p_stat == SZOMB)
875 		goto out;
876 
877 	if (siginfo != NULL) {
878 		siginfo->si_errno = 0;
879 
880 		/*
881 		 * SUSv4 requires that the si_signo value is always
882 		 * SIGCHLD. Obey it despite the rfork(2) interface
883 		 * allows to request other signal for child exit
884 		 * notification.
885 		 */
886 		siginfo->si_signo = SIGCHLD;
887 
888 		/*
889 		 *  This is still a rough estimate.  We will fix the
890 		 *  cases TRAPPED, STOPPED, and CONTINUED later.
891 		 */
892 		if (p->p_sflag & PS_COREDUMP) {
893 			siginfo->si_code = CLD_DUMPED;
894 			siginfo->si_status = p->p_xsig;
895 		} else if (p->p_xsig) {
896 			siginfo->si_code = CLD_KILLED;
897 			siginfo->si_status = p->p_xsig;
898 		} else {
899 			siginfo->si_code = CLD_EXITED;
900 			siginfo->si_status = p->p_xexit;
901 		}
902 
903 		siginfo->si_pid = p->p_pid;
904 		siginfo->si_uid = kauth_cred_geteuid(p->p_cred);
905 		siginfo->si_utime = p->p_stats->p_ru.ru_utime.tv_sec;
906 		siginfo->si_stime = p->p_stats->p_ru.ru_stime.tv_sec;
907 	}
908 
909 	/*
910 	 * There should be no reason to limit resources usage info to
911 	 * exited processes only.  A snapshot about any resources used
912 	 * by a stopped process may be exactly what is needed.
913 	 */
914 	if (wrusage != NULL) {
915 		rup = &wrusage->wru_self;
916 		*rup = p->p_stats->p_ru;
917 		calcru(p, &rup->ru_utime, &rup->ru_stime, NULL, NULL);
918 
919 		rup = &wrusage->wru_children;
920 		*rup = p->p_stats->p_cru;
921 		calcru(p, &rup->ru_utime, &rup->ru_stime, NULL, NULL);
922 	}
923 
924 	mutex_exit(p->p_lock);
925 	return rv;
926 }
927 
928 /*
929  * Scan list of child processes for a child process that has stopped or
930  * exited.  Used by sys_wait4 and 'compat' equivalents.
931  *
932  * Must be called with the proc_lock held, and may release while waiting.
933  */
934 static int
935 find_stopped_child(struct proc *parent, idtype_t idtype, id_t id, int options,
936     struct proc **child_p, struct wrusage *wru, siginfo_t *si)
937 {
938 	struct proc *child, *dead;
939 	int error;
940 
941 	KASSERT(mutex_owned(proc_lock));
942 
943 	if (options & ~(WUNTRACED|WNOHANG|WALTSIG|WALLSIG|WTRAPPED|WEXITED|
944 	    WNOWAIT|WCONTINUED)
945 	    && !(options & WOPTSCHECKED)) {
946 		*child_p = NULL;
947 		return EINVAL;
948 	}
949 
950 	if ((options & (WEXITED|WUNTRACED|WCONTINUED|WTRAPPED)) == 0) {
951 		/*
952 		 * We will be unable to find any matching processes,
953 		 * because there are no known events to look for.
954 		 * Prefer to return error instead of blocking
955 		 * indefinitely.
956 		 */
957 		*child_p = NULL;
958 		return EINVAL;
959 	}
960 
961 	if ((pid_t)id == WAIT_MYPGRP && (idtype == P_PID || idtype == P_PGID)) {
962 		mutex_enter(parent->p_lock);
963 		id = (id_t)parent->p_pgid;
964 		mutex_exit(parent->p_lock);
965 		idtype = P_PGID;
966 	}
967 
968 	for (;;) {
969 		error = ECHILD;
970 		dead = NULL;
971 
972 		LIST_FOREACH(child, &parent->p_children, p_sibling) {
973 			int rv = match_process(parent, &child, idtype, id,
974 			    options, wru, si);
975 			if (rv == -1)
976 				break;
977 			if (rv == 0)
978 				continue;
979 
980 			/*
981 			 * Wait for processes with p_exitsig != SIGCHLD
982 			 * processes only if WALTSIG is set; wait for
983 			 * processes with p_exitsig == SIGCHLD only
984 			 * if WALTSIG is clear.
985 			 */
986 			if (((options & WALLSIG) == 0) &&
987 			    (options & WALTSIG ? child->p_exitsig == SIGCHLD
988 						: P_EXITSIG(child) != SIGCHLD)){
989 				if (rv == 2) {
990 					child = NULL;
991 					break;
992 				}
993 				continue;
994 			}
995 
996 			error = 0;
997 			if ((options & WNOZOMBIE) == 0) {
998 				if (child->p_stat == SZOMB)
999 					break;
1000 				if (child->p_stat == SDEAD) {
1001 					/*
1002 					 * We may occasionally arrive here
1003 					 * after receiving a signal, but
1004 					 * immediately before the child
1005 					 * process is zombified.  The wait
1006 					 * will be short, so avoid returning
1007 					 * to userspace.
1008 					 */
1009 					dead = child;
1010 				}
1011 			}
1012 
1013 			if ((options & WCONTINUED) != 0 &&
1014 			    child->p_xsig == SIGCONT &&
1015 			    (child->p_sflag & PS_CONTINUED)) {
1016 				if ((options & WNOWAIT) == 0) {
1017 					child->p_sflag &= ~PS_CONTINUED;
1018 					child->p_waited = 1;
1019 					parent->p_nstopchild--;
1020 				}
1021 				if (si) {
1022 					si->si_status = child->p_xsig;
1023 					si->si_code = CLD_CONTINUED;
1024 				}
1025 				break;
1026 			}
1027 
1028 			if ((options & (WTRAPPED|WSTOPPED)) != 0 &&
1029 			    child->p_stat == SSTOP &&
1030 			    child->p_waited == 0 &&
1031 			    ((child->p_slflag & PSL_TRACED) ||
1032 			    options & (WUNTRACED|WSTOPPED))) {
1033 				if ((options & WNOWAIT) == 0) {
1034 					child->p_waited = 1;
1035 					parent->p_nstopchild--;
1036 				}
1037 				if (si) {
1038 					si->si_status = child->p_xsig;
1039 					si->si_code =
1040 					    (child->p_slflag & PSL_TRACED) ?
1041 					    CLD_TRAPPED : CLD_STOPPED;
1042 				}
1043 				break;
1044 			}
1045 			if (parent->p_nstopchild == 0 || rv == 2) {
1046 				child = NULL;
1047 				break;
1048 			}
1049 		}
1050 
1051 		if (child != NULL || error != 0 ||
1052 		    ((options & WNOHANG) != 0 && dead == NULL)) {
1053 			*child_p = child;
1054 			return error;
1055 		}
1056 
1057 		/*
1058 		 * Wait for another child process to stop.
1059 		 */
1060 		error = cv_wait_sig(&parent->p_waitcv, proc_lock);
1061 
1062 		if (error != 0) {
1063 			*child_p = NULL;
1064 			return error;
1065 		}
1066 	}
1067 }
1068 
1069 /*
1070  * Free a process after parent has taken all the state info.  Must be called
1071  * with the proclist lock held, and will release before returning.
1072  *
1073  * *ru is returned to the caller, and must be freed by the caller.
1074  */
1075 static void
1076 proc_free(struct proc *p, struct wrusage *wru)
1077 {
1078 	struct proc *parent = p->p_pptr;
1079 	struct lwp *l;
1080 	ksiginfo_t ksi;
1081 	kauth_cred_t cred1, cred2;
1082 	uid_t uid;
1083 
1084 	KASSERT(mutex_owned(proc_lock));
1085 	KASSERT(p->p_nlwps == 1);
1086 	KASSERT(p->p_nzlwps == 1);
1087 	KASSERT(p->p_nrlwps == 0);
1088 	KASSERT(p->p_stat == SZOMB);
1089 
1090 	/*
1091 	 * If we got the child via ptrace(2) or procfs, and
1092 	 * the parent is different (meaning the process was
1093 	 * attached, rather than run as a child), then we need
1094 	 * to give it back to the old parent, and send the
1095 	 * parent the exit signal.  The rest of the cleanup
1096 	 * will be done when the old parent waits on the child.
1097 	 */
1098 	if ((p->p_slflag & PSL_TRACED) != 0 && p->p_opptr != parent) {
1099 		mutex_enter(p->p_lock);
1100 		p->p_slflag &= ~(PSL_TRACED|PSL_FSTRACE|PSL_SYSCALL);
1101 		mutex_exit(p->p_lock);
1102 		parent = (p->p_opptr == NULL) ? initproc : p->p_opptr;
1103 		proc_reparent(p, parent);
1104 		p->p_opptr = NULL;
1105 		if (p->p_exitsig != 0) {
1106 			exit_psignal(p, parent, &ksi);
1107 			kpsignal(parent, &ksi, NULL);
1108 		}
1109 		cv_broadcast(&parent->p_waitcv);
1110 		mutex_exit(proc_lock);
1111 		return;
1112 	}
1113 
1114 	sched_proc_exit(parent, p);
1115 
1116 	/*
1117 	 * Add child times of exiting process onto its own times.
1118 	 * This cannot be done any earlier else it might get done twice.
1119 	 */
1120 	l = LIST_FIRST(&p->p_lwps);
1121 	p->p_stats->p_ru.ru_nvcsw += (l->l_ncsw - l->l_nivcsw);
1122 	p->p_stats->p_ru.ru_nivcsw += l->l_nivcsw;
1123 	ruadd(&p->p_stats->p_ru, &l->l_ru);
1124 	ruadd(&p->p_stats->p_ru, &p->p_stats->p_cru);
1125 	ruadd(&parent->p_stats->p_cru, &p->p_stats->p_ru);
1126 	if (wru != NULL) {
1127 		wru->wru_self = p->p_stats->p_ru;
1128 		wru->wru_children = p->p_stats->p_cru;
1129 	}
1130 	p->p_xsig = 0;
1131 	p->p_xexit = 0;
1132 
1133 	/*
1134 	 * At this point we are going to start freeing the final resources.
1135 	 * If anyone tries to access the proc structure after here they will
1136 	 * get a shock - bits are missing.  Attempt to make it hard!  We
1137 	 * don't bother with any further locking past this point.
1138 	 */
1139 	p->p_stat = SIDL;		/* not even a zombie any more */
1140 	LIST_REMOVE(p, p_list);	/* off zombproc */
1141 	parent->p_nstopchild--;
1142 	LIST_REMOVE(p, p_sibling);
1143 
1144 	/*
1145 	 * Let pid be reallocated.
1146 	 */
1147 	proc_free_pid(p->p_pid);
1148 
1149 	/*
1150 	 * Unlink process from its process group.
1151 	 * Releases the proc_lock.
1152 	 */
1153 	proc_leavepgrp(p);
1154 
1155 	/*
1156 	 * Delay release until after lwp_free.
1157 	 */
1158 	cred2 = l->l_cred;
1159 
1160 	/*
1161 	 * Free the last LWP's resources.
1162 	 *
1163 	 * lwp_free ensures the LWP is no longer running on another CPU.
1164 	 */
1165 	lwp_free(l, false, true);
1166 
1167 	/*
1168 	 * Now no one except us can reach the process p.
1169 	 */
1170 
1171 	/*
1172 	 * Decrement the count of procs running with this uid.
1173 	 */
1174 	cred1 = p->p_cred;
1175 	uid = kauth_cred_getuid(cred1);
1176 	(void)chgproccnt(uid, -1);
1177 
1178 	/*
1179 	 * Release substructures.
1180 	 */
1181 
1182 	lim_free(p->p_limit);
1183 	pstatsfree(p->p_stats);
1184 	kauth_cred_free(cred1);
1185 	kauth_cred_free(cred2);
1186 
1187 	/*
1188 	 * Release reference to text vnode
1189 	 */
1190 	if (p->p_textvp)
1191 		vrele(p->p_textvp);
1192 
1193 	mutex_destroy(&p->p_auxlock);
1194 	mutex_obj_free(p->p_lock);
1195 	mutex_destroy(&p->p_stmutex);
1196 	cv_destroy(&p->p_waitcv);
1197 	cv_destroy(&p->p_lwpcv);
1198 	rw_destroy(&p->p_reflock);
1199 
1200 	proc_free_mem(p);
1201 }
1202 
1203 /*
1204  * make process 'parent' the new parent of process 'child'.
1205  *
1206  * Must be called with proc_lock held.
1207  */
1208 void
1209 proc_reparent(struct proc *child, struct proc *parent)
1210 {
1211 
1212 	KASSERT(mutex_owned(proc_lock));
1213 
1214 	if (child->p_pptr == parent)
1215 		return;
1216 
1217 	if (child->p_stat == SZOMB || child->p_stat == SDEAD ||
1218 	    (child->p_stat == SSTOP && !child->p_waited)) {
1219 		child->p_pptr->p_nstopchild--;
1220 		parent->p_nstopchild++;
1221 	}
1222 	if (parent == initproc)
1223 		child->p_exitsig = SIGCHLD;
1224 
1225 	LIST_REMOVE(child, p_sibling);
1226 	LIST_INSERT_HEAD(&parent->p_children, child, p_sibling);
1227 	child->p_pptr = parent;
1228 	child->p_ppid = parent->p_pid;
1229 }
1230