xref: /dflybsd-src/sys/kern/kern_exit.c (revision e90a7c45c3303ed54c0fde732b2ba32dc80ffd9b)
1 /*
2  * Copyright (c) 1982, 1986, 1989, 1991, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  * (c) UNIX System Laboratories, Inc.
5  * All or some portions of this file are derived from material licensed
6  * to the University of California by American Telephone and Telegraph
7  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8  * the permission of UNIX System Laboratories, Inc.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. All advertising materials mentioning features or use of this software
19  *    must display the following acknowledgement:
20  *	This product includes software developed by the University of
21  *	California, Berkeley and its contributors.
22  * 4. Neither the name of the University nor the names of its contributors
23  *    may be used to endorse or promote products derived from this software
24  *    without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36  * SUCH DAMAGE.
37  *
38  *	@(#)kern_exit.c	8.7 (Berkeley) 2/12/94
39  * $FreeBSD: src/sys/kern/kern_exit.c,v 1.92.2.11 2003/01/13 22:51:16 dillon Exp $
40  * $DragonFly: src/sys/kern/kern_exit.c,v 1.91 2008/05/18 20:02:02 nth Exp $
41  */
42 
43 #include "opt_compat.h"
44 #include "opt_ktrace.h"
45 
46 #include <sys/param.h>
47 #include <sys/systm.h>
48 #include <sys/sysproto.h>
49 #include <sys/kernel.h>
50 #include <sys/malloc.h>
51 #include <sys/proc.h>
52 #include <sys/ktrace.h>
53 #include <sys/pioctl.h>
54 #include <sys/tty.h>
55 #include <sys/wait.h>
56 #include <sys/vnode.h>
57 #include <sys/resourcevar.h>
58 #include <sys/signalvar.h>
59 #include <sys/taskqueue.h>
60 #include <sys/ptrace.h>
61 #include <sys/acct.h>		/* for acct_process() function prototype */
62 #include <sys/filedesc.h>
63 #include <sys/shm.h>
64 #include <sys/sem.h>
65 #include <sys/aio.h>
66 #include <sys/jail.h>
67 #include <sys/kern_syscall.h>
68 #include <sys/upcall.h>
69 #include <sys/caps.h>
70 #include <sys/unistd.h>
71 #include <sys/eventhandler.h>
72 #include <sys/dsched.h>
73 
74 #include <vm/vm.h>
75 #include <vm/vm_param.h>
76 #include <sys/lock.h>
77 #include <vm/pmap.h>
78 #include <vm/vm_map.h>
79 #include <vm/vm_extern.h>
80 #include <sys/user.h>
81 
82 #include <sys/thread2.h>
83 #include <sys/sysref2.h>
84 #include <sys/mplock2.h>
85 
86 static void reaplwps(void *context, int dummy);
87 static void reaplwp(struct lwp *lp);
88 static void killlwps(struct lwp *lp);
89 
90 static MALLOC_DEFINE(M_ATEXIT, "atexit", "atexit callback");
91 static MALLOC_DEFINE(M_ZOMBIE, "zombie", "zombie proc status");
92 
93 /*
94  * callout list for things to do at exit time
95  */
96 struct exitlist {
97 	exitlist_fn function;
98 	TAILQ_ENTRY(exitlist) next;
99 };
100 
101 TAILQ_HEAD(exit_list_head, exitlist);
102 static struct exit_list_head exit_list = TAILQ_HEAD_INITIALIZER(exit_list);
103 
104 /*
105  * LWP reaper data
106  */
107 struct task *deadlwp_task[MAXCPU];
108 struct lwplist deadlwp_list[MAXCPU];
109 
110 /*
111  * exit --
112  *	Death of process.
113  *
114  * SYS_EXIT_ARGS(int rval)
115  *
116  * MPALMOSTSAFE
117  */
118 int
119 sys_exit(struct exit_args *uap)
120 {
121 	exit1(W_EXITCODE(uap->rval, 0));
122 	/* NOTREACHED */
123 }
124 
125 /*
126  * Extended exit --
127  *	Death of a lwp or process with optional bells and whistles.
128  *
129  * MPALMOSTSAFE
130  */
131 int
132 sys_extexit(struct extexit_args *uap)
133 {
134 	int action, who;
135 	int error;
136 
137 	action = EXTEXIT_ACTION(uap->how);
138 	who = EXTEXIT_WHO(uap->how);
139 
140 	/* Check parameters before we might perform some action */
141 	switch (who) {
142 	case EXTEXIT_PROC:
143 	case EXTEXIT_LWP:
144 		break;
145 	default:
146 		return (EINVAL);
147 	}
148 
149 	switch (action) {
150 	case EXTEXIT_SIMPLE:
151 		break;
152 	case EXTEXIT_SETINT:
153 		error = copyout(&uap->status, uap->addr, sizeof(uap->status));
154 		if (error)
155 			return (error);
156 		break;
157 	default:
158 		return (EINVAL);
159 	}
160 
161 	get_mplock();
162 
163 	switch (who) {
164 	case EXTEXIT_LWP:
165 		/*
166 		 * Be sure only to perform a simple lwp exit if there is at
167 		 * least one more lwp in the proc, which will call exit1()
168 		 * later, otherwise the proc will be an UNDEAD and not even a
169 		 * SZOMB!
170 		 */
171 		if (curproc->p_nthreads > 1) {
172 			lwp_exit(0);
173 			/* NOT REACHED */
174 		}
175 		/* else last lwp in proc:  do the real thing */
176 		/* FALLTHROUGH */
177 	default:	/* to help gcc */
178 	case EXTEXIT_PROC:
179 		exit1(W_EXITCODE(uap->status, 0));
180 		/* NOTREACHED */
181 	}
182 
183 	/* NOTREACHED */
184 	rel_mplock(); /* safety */
185 }
186 
187 /*
188  * Kill all lwps associated with the current process except the
189  * current lwp.   Return an error if we race another thread trying to
190  * do the same thing and lose the race.
191  *
192  * If forexec is non-zero the current thread and process flags are
193  * cleaned up so they can be reused.
194  */
195 int
196 killalllwps(int forexec)
197 {
198 	struct lwp *lp = curthread->td_lwp;
199 	struct proc *p = lp->lwp_proc;
200 
201 	/*
202 	 * Interlock against P_WEXIT.  Only one of the process's thread
203 	 * is allowed to do the master exit.
204 	 */
205 	if (p->p_flag & P_WEXIT)
206 		return (EALREADY);
207 	p->p_flag |= P_WEXIT;
208 
209 	/*
210 	 * Interlock with LWP_WEXIT and kill any remaining LWPs
211 	 */
212 	lp->lwp_flag |= LWP_WEXIT;
213 	if (p->p_nthreads > 1)
214 		killlwps(lp);
215 
216 	/*
217 	 * If doing this for an exec, clean up the remaining thread
218 	 * (us) for continuing operation after all the other threads
219 	 * have been killed.
220 	 */
221 	if (forexec) {
222 		lp->lwp_flag &= ~LWP_WEXIT;
223 		p->p_flag &= ~P_WEXIT;
224 	}
225 	return(0);
226 }
227 
228 /*
229  * Kill all LWPs except the current one.  Do not try to signal
230  * LWPs which have exited on their own or have already been
231  * signaled.
232  */
233 static void
234 killlwps(struct lwp *lp)
235 {
236 	struct proc *p = lp->lwp_proc;
237 	struct lwp *tlp;
238 
239 	/*
240 	 * Kill the remaining LWPs.  We must send the signal before setting
241 	 * LWP_WEXIT.  The setting of WEXIT is optional but helps reduce
242 	 * races.  tlp must be held across the call as it might block and
243 	 * allow the target lwp to rip itself out from under our loop.
244 	 */
245 	FOREACH_LWP_IN_PROC(tlp, p) {
246 		LWPHOLD(tlp);
247 		if ((tlp->lwp_flag & LWP_WEXIT) == 0) {
248 			lwpsignal(p, tlp, SIGKILL);
249 			tlp->lwp_flag |= LWP_WEXIT;
250 		}
251 		LWPRELE(tlp);
252 	}
253 
254 	/*
255 	 * Wait for everything to clear out.
256 	 */
257 	while (p->p_nthreads > 1) {
258 		tsleep(&p->p_nthreads, 0, "killlwps", 0);
259 	}
260 }
261 
262 /*
263  * Exit: deallocate address space and other resources, change proc state
264  * to zombie, and unlink proc from allproc and parent's lists.  Save exit
265  * status and rusage for wait().  Check for child processes and orphan them.
266  */
267 void
268 exit1(int rv)
269 {
270 	struct thread *td = curthread;
271 	struct proc *p = td->td_proc;
272 	struct lwp *lp = td->td_lwp;
273 	struct proc *q, *nq;
274 	struct vmspace *vm;
275 	struct vnode *vtmp;
276 	struct exitlist *ep;
277 	int error;
278 
279 	if (p->p_pid == 1) {
280 		kprintf("init died (signal %d, exit %d)\n",
281 		    WTERMSIG(rv), WEXITSTATUS(rv));
282 		panic("Going nowhere without my init!");
283 	}
284 
285 	get_mplock();
286 
287 	varsymset_clean(&p->p_varsymset);
288 	lockuninit(&p->p_varsymset.vx_lock);
289 	/*
290 	 * Kill all lwps associated with the current process, return an
291 	 * error if we race another thread trying to do the same thing
292 	 * and lose the race.
293 	 */
294 	error = killalllwps(0);
295 	if (error) {
296 		lwp_exit(0);
297 		/* NOT REACHED */
298 	}
299 
300 	caps_exit(lp->lwp_thread);
301 	aio_proc_rundown(p);
302 
303 	/* are we a task leader? */
304 	if (p == p->p_leader) {
305         	struct kill_args killArgs;
306 		killArgs.signum = SIGKILL;
307 		q = p->p_peers;
308 		while(q) {
309 			killArgs.pid = q->p_pid;
310 			/*
311 		         * The interface for kill is better
312 			 * than the internal signal
313 			 */
314 			sys_kill(&killArgs);
315 			nq = q;
316 			q = q->p_peers;
317 		}
318 		while (p->p_peers)
319 			tsleep((caddr_t)p, 0, "exit1", 0);
320 	}
321 
322 #ifdef PGINPROF
323 	vmsizmon();
324 #endif
325 	STOPEVENT(p, S_EXIT, rv);
326 	wakeup(&p->p_stype);	/* Wakeup anyone in procfs' PIOCWAIT */
327 
328 	/*
329 	 * Check if any loadable modules need anything done at process exit.
330 	 * e.g. SYSV IPC stuff
331 	 * XXX what if one of these generates an error?
332 	 */
333 	p->p_xstat = rv;
334 	EVENTHANDLER_INVOKE(process_exit, p);
335 
336 	/*
337 	 * XXX: imho, the eventhandler stuff is much cleaner than this.
338 	 *	Maybe we should move everything to use eventhandler.
339 	 */
340 	TAILQ_FOREACH(ep, &exit_list, next)
341 		(*ep->function)(td);
342 
343 	if (p->p_flag & P_PROFIL)
344 		stopprofclock(p);
345 	/*
346 	 * If parent is waiting for us to exit or exec,
347 	 * P_PPWAIT is set; we will wakeup the parent below.
348 	 */
349 	p->p_flag &= ~(P_TRACED | P_PPWAIT);
350 	SIGEMPTYSET(p->p_siglist);
351 	SIGEMPTYSET(lp->lwp_siglist);
352 	if (timevalisset(&p->p_realtimer.it_value))
353 		callout_stop(&p->p_ithandle);
354 
355 	/*
356 	 * Reset any sigio structures pointing to us as a result of
357 	 * F_SETOWN with our pid.
358 	 */
359 	funsetownlst(&p->p_sigiolst);
360 
361 	/*
362 	 * Close open files and release open-file table.
363 	 * This may block!
364 	 */
365 	fdfree(p, NULL);
366 
367 	if(p->p_leader->p_peers) {
368 		q = p->p_leader;
369 		while(q->p_peers != p)
370 			q = q->p_peers;
371 		q->p_peers = p->p_peers;
372 		wakeup((caddr_t)p->p_leader);
373 	}
374 
375 	/*
376 	 * XXX Shutdown SYSV semaphores
377 	 */
378 	semexit(p);
379 
380 	KKASSERT(p->p_numposixlocks == 0);
381 
382 	/* The next two chunks should probably be moved to vmspace_exit. */
383 	vm = p->p_vmspace;
384 
385 	/*
386 	 * Release upcalls associated with this process
387 	 */
388 	if (vm->vm_upcalls)
389 		upc_release(vm, lp);
390 
391 	/*
392 	 * Clean up data related to virtual kernel operation.  Clean up
393 	 * any vkernel context related to the current lwp now so we can
394 	 * destroy p_vkernel.
395 	 */
396 	if (p->p_vkernel) {
397 		vkernel_lwp_exit(lp);
398 		vkernel_exit(p);
399 	}
400 
401 	/*
402 	 * Release user portion of address space.
403 	 * This releases references to vnodes,
404 	 * which could cause I/O if the file has been unlinked.
405 	 * Need to do this early enough that we can still sleep.
406 	 * Can't free the entire vmspace as the kernel stack
407 	 * may be mapped within that space also.
408 	 *
409 	 * Processes sharing the same vmspace may exit in one order, and
410 	 * get cleaned up by vmspace_exit() in a different order.  The
411 	 * last exiting process to reach this point releases as much of
412 	 * the environment as it can, and the last process cleaned up
413 	 * by vmspace_exit() (which decrements exitingcnt) cleans up the
414 	 * remainder.
415 	 */
416 	vmspace_exitbump(vm);
417 	sysref_put(&vm->vm_sysref);
418 
419 	if (SESS_LEADER(p)) {
420 		struct session *sp = p->p_session;
421 
422 		if (sp->s_ttyvp) {
423 			/*
424 			 * We are the controlling process.  Signal the
425 			 * foreground process group, drain the controlling
426 			 * terminal, and revoke access to the controlling
427 			 * terminal.
428 			 *
429 			 * NOTE: while waiting for the process group to exit
430 			 * it is possible that one of the processes in the
431 			 * group will revoke the tty, so the ttyclosesession()
432 			 * function will re-check sp->s_ttyvp.
433 			 */
434 			if (sp->s_ttyp && (sp->s_ttyp->t_session == sp)) {
435 				if (sp->s_ttyp->t_pgrp)
436 					pgsignal(sp->s_ttyp->t_pgrp, SIGHUP, 1);
437 				ttywait(sp->s_ttyp);
438 				ttyclosesession(sp, 1); /* also revoke */
439 			}
440 			/*
441 			 * Release the tty.  If someone has it open via
442 			 * /dev/tty then close it (since they no longer can
443 			 * once we've NULL'd it out).
444 			 */
445 			ttyclosesession(sp, 0);
446 
447 			/*
448 			 * s_ttyp is not zero'd; we use this to indicate
449 			 * that the session once had a controlling terminal.
450 			 * (for logging and informational purposes)
451 			 */
452 		}
453 		sp->s_leader = NULL;
454 	}
455 	fixjobc(p, p->p_pgrp, 0);
456 	(void)acct_process(p);
457 #ifdef KTRACE
458 	/*
459 	 * release trace file
460 	 */
461 	if (p->p_tracenode)
462 		ktrdestroy(&p->p_tracenode);
463 	p->p_traceflag = 0;
464 #endif
465 	/*
466 	 * Release reference to text vnode
467 	 */
468 	if ((vtmp = p->p_textvp) != NULL) {
469 		p->p_textvp = NULL;
470 		vrele(vtmp);
471 	}
472 
473 	/* Release namecache handle to text file */
474 	if (p->p_textnch.ncp)
475 		cache_drop(&p->p_textnch);
476 
477 	/*
478 	 * Move the process to the zombie list.  This will block
479 	 * until the process p_lock count reaches 0.  The process will
480 	 * not be reaped until TDF_EXITING is set by cpu_thread_exit(),
481 	 * which is called from cpu_proc_exit().
482 	 */
483 	proc_move_allproc_zombie(p);
484 
485 	q = LIST_FIRST(&p->p_children);
486 	if (q)		/* only need this if any child is S_ZOMB */
487 		wakeup((caddr_t) initproc);
488 	for (; q != 0; q = nq) {
489 		nq = LIST_NEXT(q, p_sibling);
490 		LIST_REMOVE(q, p_sibling);
491 		LIST_INSERT_HEAD(&initproc->p_children, q, p_sibling);
492 		q->p_pptr = initproc;
493 		q->p_sigparent = SIGCHLD;
494 		/*
495 		 * Traced processes are killed
496 		 * since their existence means someone is screwing up.
497 		 */
498 		if (q->p_flag & P_TRACED) {
499 			q->p_flag &= ~P_TRACED;
500 			ksignal(q, SIGKILL);
501 		}
502 	}
503 
504 	/*
505 	 * Save exit status and final rusage info, adding in child rusage
506 	 * info and self times.
507 	 */
508 	calcru_proc(p, &p->p_ru);
509 	ruadd(&p->p_ru, &p->p_cru);
510 
511 	/*
512 	 * notify interested parties of our demise.
513 	 */
514 	KNOTE(&p->p_klist, NOTE_EXIT);
515 
516 	/*
517 	 * Notify parent that we're gone.  If parent has the PS_NOCLDWAIT
518 	 * flag set, notify process 1 instead (and hope it will handle
519 	 * this situation).
520 	 */
521 	if (p->p_pptr->p_sigacts->ps_flag & PS_NOCLDWAIT) {
522 		struct proc *pp = p->p_pptr;
523 		proc_reparent(p, initproc);
524 		/*
525 		 * If this was the last child of our parent, notify
526 		 * parent, so in case he was wait(2)ing, he will
527 		 * continue.
528 		 */
529 		if (LIST_EMPTY(&pp->p_children))
530 			wakeup((caddr_t)pp);
531 	}
532 
533 	/* lwkt_gettoken(&proc_token); */
534 	q = p->p_pptr;
535 	if (p->p_sigparent && q != initproc) {
536 		PHOLD(q);
537 	        ksignal(q, p->p_sigparent);
538 		PRELE(q);
539 	} else {
540 	        ksignal(q, SIGCHLD);
541 	}
542 	/* lwkt_reltoken(&proc_token); */
543 	/* NOTE: p->p_pptr can get ripped out */
544 
545 	wakeup(p->p_pptr);
546 	/*
547 	 * cpu_exit is responsible for clearing curproc, since
548 	 * it is heavily integrated with the thread/switching sequence.
549 	 *
550 	 * Other substructures are freed from wait().
551 	 */
552 	plimit_free(p);
553 
554 	/*
555 	 * Release the current user process designation on the process so
556 	 * the userland scheduler can work in someone else.
557 	 */
558 	p->p_usched->release_curproc(lp);
559 
560 	/*
561 	 * Finally, call machine-dependent code to release as many of the
562 	 * lwp's resources as we can and halt execution of this thread.
563 	 */
564 	lwp_exit(1);
565 }
566 
567 /*
568  * Eventually called by every exiting LWP
569  */
570 void
571 lwp_exit(int masterexit)
572 {
573 	struct thread *td = curthread;
574 	struct lwp *lp = td->td_lwp;
575 	struct proc *p = lp->lwp_proc;
576 
577 	/*
578 	 * lwp_exit() may be called without setting LWP_WEXIT, so
579 	 * make sure it is set here.
580 	 */
581 	lp->lwp_flag |= LWP_WEXIT;
582 
583 	/*
584 	 * Clean up any virtualization
585 	 */
586 	if (lp->lwp_vkernel)
587 		vkernel_lwp_exit(lp);
588 
589 	/*
590 	 * Clean up select/poll support
591 	 */
592 	kqueue_terminate(&lp->lwp_kqueue);
593 
594 	/*
595 	 * Clean up any syscall-cached ucred
596 	 */
597 	if (td->td_ucred) {
598 		crfree(td->td_ucred);
599 		td->td_ucred = NULL;
600 	}
601 
602 	/*
603 	 * Nobody actually wakes us when the lock
604 	 * count reaches zero, so just wait one tick.
605 	 */
606 	while (lp->lwp_lock > 0)
607 		tsleep(lp, 0, "lwpexit", 1);
608 
609 	/* Hand down resource usage to our proc */
610 	ruadd(&p->p_ru, &lp->lwp_ru);
611 
612 	/*
613 	 * If we don't hold the process until the LWP is reaped wait*()
614 	 * may try to dispose of its vmspace before all the LWPs have
615 	 * actually terminated.
616 	 */
617 	PHOLD(p);
618 
619 	/*
620 	 * Do any remaining work that might block on us.  We should be
621 	 * coded such that further blocking is ok after decrementing
622 	 * p_nthreads but don't take the chance.
623 	 */
624 	dsched_exit_thread(td);
625 	biosched_done(curthread);
626 
627 	/*
628 	 * We have to use the reaper for all the LWPs except the one doing
629 	 * the master exit.  The LWP doing the master exit can just be
630 	 * left on p_lwps and the process reaper will deal with it
631 	 * synchronously, which is much faster.
632 	 *
633 	 * Wakeup anyone waiting on p_nthreads to drop to 1 or 0.
634 	 */
635 	if (masterexit == 0) {
636 		lwp_rb_tree_RB_REMOVE(&p->p_lwp_tree, lp);
637 		--p->p_nthreads;
638 		if (p->p_nthreads <= 1)
639 			wakeup(&p->p_nthreads);
640 		LIST_INSERT_HEAD(&deadlwp_list[mycpuid], lp, u.lwp_reap_entry);
641 		taskqueue_enqueue(taskqueue_thread[mycpuid],
642 				  deadlwp_task[mycpuid]);
643 	} else {
644 		--p->p_nthreads;
645 		if (p->p_nthreads <= 1)
646 			wakeup(&p->p_nthreads);
647 	}
648 	cpu_lwp_exit();
649 }
650 
651 /*
652  * Wait until a lwp is completely dead.
653  *
654  * If the thread is still executing, which can't be waited upon,
655  * return failure.  The caller is responsible of waiting a little
656  * bit and checking again.
657  *
658  * Suggested use:
659  * while (!lwp_wait(lp))
660  *	tsleep(lp, 0, "lwpwait", 1);
661  */
662 static int
663 lwp_wait(struct lwp *lp)
664 {
665 	struct thread *td = lp->lwp_thread;;
666 
667 	KKASSERT(lwkt_preempted_proc() != lp);
668 
669 	while (lp->lwp_lock > 0)
670 		tsleep(lp, 0, "lwpwait1", 1);
671 
672 	lwkt_wait_free(td);
673 
674 	/*
675 	 * The lwp's thread may still be in the middle
676 	 * of switching away, we can't rip its stack out from
677 	 * under it until TDF_EXITING is set and both
678 	 * TDF_RUNNING and TDF_PREEMPT_LOCK are clear.
679 	 * TDF_PREEMPT_LOCK must be checked because TDF_RUNNING
680 	 * will be cleared temporarily if a thread gets
681 	 * preempted.
682 	 *
683 	 * YYY no wakeup occurs, so we simply return failure
684 	 * and let the caller deal with sleeping and calling
685 	 * us again.
686 	 */
687 	if ((td->td_flags & (TDF_RUNNING|TDF_PREEMPT_LOCK|
688 			     TDF_EXITING|TDF_RUNQ)) != TDF_EXITING) {
689 		return (0);
690 	}
691 	KASSERT((td->td_flags & TDF_TSLEEPQ) == 0,
692 		("lwp_wait: td %p (%s) still on sleep queue", td, td->td_comm));
693 	return (1);
694 }
695 
696 /*
697  * Release the resources associated with a lwp.
698  * The lwp must be completely dead.
699  */
700 void
701 lwp_dispose(struct lwp *lp)
702 {
703 	struct thread *td = lp->lwp_thread;;
704 
705 	KKASSERT(lwkt_preempted_proc() != lp);
706 	KKASSERT(td->td_refs == 0);
707 	KKASSERT((td->td_flags & (TDF_RUNNING|TDF_PREEMPT_LOCK|TDF_EXITING)) ==
708 		 TDF_EXITING);
709 
710 	PRELE(lp->lwp_proc);
711 	lp->lwp_proc = NULL;
712 	if (td != NULL) {
713 		td->td_proc = NULL;
714 		td->td_lwp = NULL;
715 		lp->lwp_thread = NULL;
716 		lwkt_free_thread(td);
717 	}
718 	kfree(lp, M_LWP);
719 }
720 
721 /*
722  * MPSAFE
723  */
724 int
725 sys_wait4(struct wait_args *uap)
726 {
727 	struct rusage rusage;
728 	int error, status;
729 
730 	error = kern_wait(uap->pid, (uap->status ? &status : NULL),
731 			  uap->options, (uap->rusage ? &rusage : NULL),
732 			  &uap->sysmsg_result);
733 
734 	if (error == 0 && uap->status)
735 		error = copyout(&status, uap->status, sizeof(*uap->status));
736 	if (error == 0 && uap->rusage)
737 		error = copyout(&rusage, uap->rusage, sizeof(*uap->rusage));
738 	return (error);
739 }
740 
741 /*
742  * wait1()
743  *
744  * wait_args(int pid, int *status, int options, struct rusage *rusage)
745  *
746  * MPALMOSTSAFE
747  */
748 int
749 kern_wait(pid_t pid, int *status, int options, struct rusage *rusage, int *res)
750 {
751 	struct thread *td = curthread;
752 	struct lwp *lp;
753 	struct proc *q = td->td_proc;
754 	struct proc *p, *t;
755 	int nfound, error;
756 
757 	if (pid == 0)
758 		pid = -q->p_pgid;
759 	if (options &~ (WUNTRACED|WNOHANG|WCONTINUED|WLINUXCLONE))
760 		return (EINVAL);
761 	get_mplock();
762 loop:
763 	/*
764 	 * Hack for backwards compatibility with badly written user code.
765 	 * Or perhaps we have to do this anyway, it is unclear. XXX
766 	 *
767 	 * The problem is that if a process group is stopped and the parent
768 	 * is doing a wait*(..., WUNTRACED, ...), it will see the STOP
769 	 * of the child and then stop itself when it tries to return from the
770 	 * system call.  When the process group is resumed the parent will
771 	 * then get the STOP status even though the child has now resumed
772 	 * (a followup wait*() will get the CONT status).
773 	 *
774 	 * Previously the CONT would overwrite the STOP because the tstop
775 	 * was handled within tsleep(), and the parent would only see
776 	 * the CONT when both are stopped and continued together.  This litte
777 	 * two-line hack restores this effect.
778 	 */
779 	while (q->p_stat == SSTOP)
780             tstop();
781 
782 	nfound = 0;
783 	LIST_FOREACH(p, &q->p_children, p_sibling) {
784 		if (pid != WAIT_ANY &&
785 		    p->p_pid != pid && p->p_pgid != -pid)
786 			continue;
787 
788 		/* This special case handles a kthread spawned by linux_clone
789 		 * (see linux_misc.c).  The linux_wait4 and linux_waitpid
790 		 * functions need to be able to distinguish between waiting
791 		 * on a process and waiting on a thread.  It is a thread if
792 		 * p_sigparent is not SIGCHLD, and the WLINUXCLONE option
793 		 * signifies we want to wait for threads and not processes.
794 		 */
795 		if ((p->p_sigparent != SIGCHLD) ^
796 		    ((options & WLINUXCLONE) != 0)) {
797 			continue;
798 		}
799 
800 		nfound++;
801 		if (p->p_stat == SZOMB) {
802 			/*
803 			 * We may go into SZOMB with threads still present.
804 			 * We must wait for them to exit before we can reap
805 			 * the master thread, otherwise we may race reaping
806 			 * non-master threads.
807 			 */
808 			while (p->p_nthreads > 0) {
809 				tsleep(&p->p_nthreads, 0, "lwpzomb", hz);
810 			}
811 
812 			/*
813 			 * Reap any LWPs left in p->p_lwps.  This is usually
814 			 * just the last LWP.  This must be done before
815 			 * we loop on p_lock since the lwps hold a ref on
816 			 * it as a vmspace interlock.
817 			 *
818 			 * Once that is accomplished p_nthreads had better
819 			 * be zero.
820 			 */
821 			while ((lp = RB_ROOT(&p->p_lwp_tree)) != NULL) {
822 				lwp_rb_tree_RB_REMOVE(&p->p_lwp_tree, lp);
823 				reaplwp(lp);
824 			}
825 			KKASSERT(p->p_nthreads == 0);
826 
827 			/*
828 			 * Don't do anything really bad until all references
829 			 * to the process go away.  This may include other
830 			 * LWPs which are still in the process of being
831 			 * reaped.  We can't just pull the rug out from under
832 			 * them because they may still be using the VM space.
833 			 *
834 			 * Certain kernel facilities such as /proc will also
835 			 * put a hold on the process for short periods of
836 			 * time.
837 			 */
838 			while (p->p_lock)
839 				tsleep(p, 0, "reap3", hz);
840 
841 			/* scheduling hook for heuristic */
842 			/* XXX no lwp available, we need a different heuristic */
843 			/*
844 			p->p_usched->heuristic_exiting(td->td_lwp, deadlp);
845 			*/
846 
847 			/* Take care of our return values. */
848 			*res = p->p_pid;
849 			if (status)
850 				*status = p->p_xstat;
851 			if (rusage)
852 				*rusage = p->p_ru;
853 			/*
854 			 * If we got the child via a ptrace 'attach',
855 			 * we need to give it back to the old parent.
856 			 */
857 			if (p->p_oppid && (t = pfind(p->p_oppid))) {
858 				p->p_oppid = 0;
859 				proc_reparent(p, t);
860 				ksignal(t, SIGCHLD);
861 				wakeup((caddr_t)t);
862 				error = 0;
863 				goto done;
864 			}
865 
866 			/*
867 			 * Unlink the proc from its process group so that
868 			 * the following operations won't lead to an
869 			 * inconsistent state for processes running down
870 			 * the zombie list.
871 			 */
872 			KKASSERT(p->p_lock == 0);
873 			proc_remove_zombie(p);
874 			leavepgrp(p);
875 
876 			p->p_xstat = 0;
877 			ruadd(&q->p_cru, &p->p_ru);
878 
879 			/*
880 			 * Decrement the count of procs running with this uid.
881 			 */
882 			chgproccnt(p->p_ucred->cr_ruidinfo, -1, 0);
883 
884 			/*
885 			 * Free up credentials.
886 			 */
887 			crfree(p->p_ucred);
888 			p->p_ucred = NULL;
889 
890 			/*
891 			 * Remove unused arguments
892 			 */
893 			if (p->p_args && --p->p_args->ar_ref == 0)
894 				FREE(p->p_args, M_PARGS);
895 
896 			if (--p->p_sigacts->ps_refcnt == 0) {
897 				kfree(p->p_sigacts, M_SUBPROC);
898 				p->p_sigacts = NULL;
899 			}
900 
901 			vm_waitproc(p);
902 			kfree(p, M_PROC);
903 			nprocs--;
904 			error = 0;
905 			goto done;
906 		}
907 		if (p->p_stat == SSTOP && (p->p_flag & P_WAITED) == 0 &&
908 		    (p->p_flag & P_TRACED || options & WUNTRACED)) {
909 			p->p_flag |= P_WAITED;
910 
911 			*res = p->p_pid;
912 			if (status)
913 				*status = W_STOPCODE(p->p_xstat);
914 			/* Zero rusage so we get something consistent. */
915 			if (rusage)
916 				bzero(rusage, sizeof(rusage));
917 			error = 0;
918 			goto done;
919 		}
920 		if (options & WCONTINUED && (p->p_flag & P_CONTINUED)) {
921 			*res = p->p_pid;
922 			p->p_flag &= ~P_CONTINUED;
923 
924 			if (status)
925 				*status = SIGCONT;
926 			error = 0;
927 			goto done;
928 		}
929 	}
930 	if (nfound == 0) {
931 		error = ECHILD;
932 		goto done;
933 	}
934 	if (options & WNOHANG) {
935 		*res = 0;
936 		error = 0;
937 		goto done;
938 	}
939 	error = tsleep((caddr_t)q, PCATCH, "wait", 0);
940 	if (error) {
941 done:
942 		rel_mplock();
943 		return (error);
944 	}
945 	goto loop;
946 }
947 
948 /*
949  * make process 'parent' the new parent of process 'child'.
950  */
951 void
952 proc_reparent(struct proc *child, struct proc *parent)
953 {
954 
955 	if (child->p_pptr == parent)
956 		return;
957 
958 	LIST_REMOVE(child, p_sibling);
959 	LIST_INSERT_HEAD(&parent->p_children, child, p_sibling);
960 	child->p_pptr = parent;
961 }
962 
963 /*
964  * The next two functions are to handle adding/deleting items on the
965  * exit callout list
966  *
967  * at_exit():
968  * Take the arguments given and put them onto the exit callout list,
969  * However first make sure that it's not already there.
970  * returns 0 on success.
971  */
972 
973 int
974 at_exit(exitlist_fn function)
975 {
976 	struct exitlist *ep;
977 
978 #ifdef INVARIANTS
979 	/* Be noisy if the programmer has lost track of things */
980 	if (rm_at_exit(function))
981 		kprintf("WARNING: exit callout entry (%p) already present\n",
982 		    function);
983 #endif
984 	ep = kmalloc(sizeof(*ep), M_ATEXIT, M_NOWAIT);
985 	if (ep == NULL)
986 		return (ENOMEM);
987 	ep->function = function;
988 	TAILQ_INSERT_TAIL(&exit_list, ep, next);
989 	return (0);
990 }
991 
992 /*
993  * Scan the exit callout list for the given item and remove it.
994  * Returns the number of items removed (0 or 1)
995  */
996 int
997 rm_at_exit(exitlist_fn function)
998 {
999 	struct exitlist *ep;
1000 
1001 	TAILQ_FOREACH(ep, &exit_list, next) {
1002 		if (ep->function == function) {
1003 			TAILQ_REMOVE(&exit_list, ep, next);
1004 			kfree(ep, M_ATEXIT);
1005 			return(1);
1006 		}
1007 	}
1008 	return (0);
1009 }
1010 
1011 /*
1012  * LWP reaper related code.
1013  */
1014 static void
1015 reaplwps(void *context, int dummy)
1016 {
1017 	struct lwplist *lwplist = context;
1018 	struct lwp *lp;
1019 
1020 	get_mplock();
1021 	while ((lp = LIST_FIRST(lwplist))) {
1022 		LIST_REMOVE(lp, u.lwp_reap_entry);
1023 		reaplwp(lp);
1024 	}
1025 	rel_mplock();
1026 }
1027 
1028 static void
1029 reaplwp(struct lwp *lp)
1030 {
1031 	while (lwp_wait(lp) == 0)
1032 		tsleep(lp, 0, "lwpreap", 1);
1033 	lwp_dispose(lp);
1034 }
1035 
1036 static void
1037 deadlwp_init(void)
1038 {
1039 	int cpu;
1040 
1041 	for (cpu = 0; cpu < ncpus; cpu++) {
1042 		LIST_INIT(&deadlwp_list[cpu]);
1043 		deadlwp_task[cpu] = kmalloc(sizeof(*deadlwp_task[cpu]), M_DEVBUF, M_WAITOK);
1044 		TASK_INIT(deadlwp_task[cpu], 0, reaplwps, &deadlwp_list[cpu]);
1045 	}
1046 }
1047 
1048 SYSINIT(deadlwpinit, SI_SUB_CONFIGURE, SI_ORDER_ANY, deadlwp_init, NULL);
1049