xref: /dflybsd-src/sys/kern/kern_exit.c (revision 90ea502b8c5d21f908cedff6680ee2bc9e74ce74)
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 
72 #include <vm/vm.h>
73 #include <vm/vm_param.h>
74 #include <sys/lock.h>
75 #include <vm/pmap.h>
76 #include <vm/vm_map.h>
77 #include <vm/vm_extern.h>
78 #include <sys/user.h>
79 
80 #include <sys/thread2.h>
81 #include <sys/sysref2.h>
82 
83 static void reaplwps(void *context, int dummy);
84 static void reaplwp(struct lwp *lp);
85 static void killlwps(struct lwp *lp);
86 
87 static MALLOC_DEFINE(M_ATEXIT, "atexit", "atexit callback");
88 static MALLOC_DEFINE(M_ZOMBIE, "zombie", "zombie proc status");
89 
90 /*
91  * callout list for things to do at exit time
92  */
93 struct exitlist {
94 	exitlist_fn function;
95 	TAILQ_ENTRY(exitlist) next;
96 };
97 
98 TAILQ_HEAD(exit_list_head, exitlist);
99 static struct exit_list_head exit_list = TAILQ_HEAD_INITIALIZER(exit_list);
100 
101 /*
102  * LWP reaper data
103  */
104 struct task *deadlwp_task[MAXCPU];
105 struct lwplist deadlwp_list[MAXCPU];
106 
107 /*
108  * exit --
109  *	Death of process.
110  *
111  * SYS_EXIT_ARGS(int rval)
112  *
113  * MPALMOSTSAFE
114  */
115 int
116 sys_exit(struct exit_args *uap)
117 {
118 	get_mplock();
119 	exit1(W_EXITCODE(uap->rval, 0));
120 	/* NOTREACHED */
121 	rel_mplock();
122 }
123 
124 /*
125  * Extended exit --
126  *	Death of a lwp or process with optional bells and whistles.
127  *
128  * MPALMOSTSAFE
129  */
130 int
131 sys_extexit(struct extexit_args *uap)
132 {
133 	int action, who;
134 	int error;
135 
136 	action = EXTEXIT_ACTION(uap->how);
137 	who = EXTEXIT_WHO(uap->how);
138 
139 	/* Check parameters before we might perform some action */
140 	switch (who) {
141 	case EXTEXIT_PROC:
142 	case EXTEXIT_LWP:
143 		break;
144 	default:
145 		return (EINVAL);
146 	}
147 
148 	switch (action) {
149 	case EXTEXIT_SIMPLE:
150 		break;
151 	case EXTEXIT_SETINT:
152 		error = copyout(&uap->status, uap->addr, sizeof(uap->status));
153 		if (error)
154 			return (error);
155 		break;
156 	default:
157 		return (EINVAL);
158 	}
159 
160 	get_mplock();
161 
162 	switch (who) {
163 	case EXTEXIT_LWP:
164 		/*
165 		 * Be sure only to perform a simple lwp exit if there is at
166 		 * least one more lwp in the proc, which will call exit1()
167 		 * later, otherwise the proc will be an UNDEAD and not even a
168 		 * SZOMB!
169 		 */
170 		if (curproc->p_nthreads > 1) {
171 			lwp_exit(0);
172 			/* NOT REACHED */
173 		}
174 		/* else last lwp in proc:  do the real thing */
175 		/* FALLTHROUGH */
176 	default:	/* to help gcc */
177 	case EXTEXIT_PROC:
178 		exit1(W_EXITCODE(uap->status, 0));
179 		/* NOTREACHED */
180 	}
181 
182 	/* NOTREACHED */
183 	rel_mplock(); /* safety */
184 }
185 
186 /*
187  * Kill all lwps associated with the current process except the
188  * current lwp.   Return an error if we race another thread trying to
189  * do the same thing and lose the race.
190  *
191  * If forexec is non-zero the current thread and process flags are
192  * cleaned up so they can be reused.
193  */
194 int
195 killalllwps(int forexec)
196 {
197 	struct lwp *lp = curthread->td_lwp;
198 	struct proc *p = lp->lwp_proc;
199 
200 	/*
201 	 * Interlock against P_WEXIT.  Only one of the process's thread
202 	 * is allowed to do the master exit.
203 	 */
204 	if (p->p_flag & P_WEXIT)
205 		return (EALREADY);
206 	p->p_flag |= P_WEXIT;
207 
208 	/*
209 	 * Interlock with LWP_WEXIT and kill any remaining LWPs
210 	 */
211 	lp->lwp_flag |= LWP_WEXIT;
212 	if (p->p_nthreads > 1)
213 		killlwps(lp);
214 
215 	/*
216 	 * If doing this for an exec, clean up the remaining thread
217 	 * (us) for continuing operation after all the other threads
218 	 * have been killed.
219 	 */
220 	if (forexec) {
221 		lp->lwp_flag &= ~LWP_WEXIT;
222 		p->p_flag &= ~P_WEXIT;
223 	}
224 	return(0);
225 }
226 
227 /*
228  * Kill all LWPs except the current one.  Do not try to signal
229  * LWPs which have exited on their own or have already been
230  * signaled.
231  */
232 static void
233 killlwps(struct lwp *lp)
234 {
235 	struct proc *p = lp->lwp_proc;
236 	struct lwp *tlp;
237 
238 	/*
239 	 * Kill the remaining LWPs.  We must send the signal before setting
240 	 * LWP_WEXIT.  The setting of WEXIT is optional but helps reduce
241 	 * races.  tlp must be held across the call as it might block and
242 	 * allow the target lwp to rip itself out from under our loop.
243 	 */
244 	FOREACH_LWP_IN_PROC(tlp, p) {
245 		LWPHOLD(tlp);
246 		if ((tlp->lwp_flag & LWP_WEXIT) == 0) {
247 			lwpsignal(p, tlp, SIGKILL);
248 			tlp->lwp_flag |= LWP_WEXIT;
249 		}
250 		LWPRELE(tlp);
251 	}
252 
253 	/*
254 	 * Wait for everything to clear out.
255 	 */
256 	while (p->p_nthreads > 1) {
257 		tsleep(&p->p_nthreads, 0, "killlwps", 0);
258 	}
259 }
260 
261 /*
262  * Exit: deallocate address space and other resources, change proc state
263  * to zombie, and unlink proc from allproc and parent's lists.  Save exit
264  * status and rusage for wait().  Check for child processes and orphan them.
265  */
266 void
267 exit1(int rv)
268 {
269 	struct thread *td = curthread;
270 	struct proc *p = td->td_proc;
271 	struct lwp *lp = td->td_lwp;
272 	struct proc *q, *nq;
273 	struct vmspace *vm;
274 	struct vnode *vtmp;
275 	struct exitlist *ep;
276 	int error;
277 
278 	if (p->p_pid == 1) {
279 		kprintf("init died (signal %d, exit %d)\n",
280 		    WTERMSIG(rv), WEXITSTATUS(rv));
281 		panic("Going nowhere without my init!");
282 	}
283 
284 	varsymset_clean(&p->p_varsymset);
285 	lockuninit(&p->p_varsymset.vx_lock);
286 	/*
287 	 * Kill all lwps associated with the current process, return an
288 	 * error if we race another thread trying to do the same thing
289 	 * and lose the race.
290 	 */
291 	error = killalllwps(0);
292 	if (error) {
293 		lwp_exit(0);
294 		/* NOT REACHED */
295 	}
296 
297 	caps_exit(lp->lwp_thread);
298 	aio_proc_rundown(p);
299 
300 	/* are we a task leader? */
301 	if (p == p->p_leader) {
302         	struct kill_args killArgs;
303 		killArgs.signum = SIGKILL;
304 		q = p->p_peers;
305 		while(q) {
306 			killArgs.pid = q->p_pid;
307 			/*
308 		         * The interface for kill is better
309 			 * than the internal signal
310 			 */
311 			sys_kill(&killArgs);
312 			nq = q;
313 			q = q->p_peers;
314 		}
315 		while (p->p_peers)
316 			tsleep((caddr_t)p, 0, "exit1", 0);
317 	}
318 
319 #ifdef PGINPROF
320 	vmsizmon();
321 #endif
322 	STOPEVENT(p, S_EXIT, rv);
323 	wakeup(&p->p_stype);	/* Wakeup anyone in procfs' PIOCWAIT */
324 
325 	/*
326 	 * Check if any loadable modules need anything done at process exit.
327 	 * e.g. SYSV IPC stuff
328 	 * XXX what if one of these generates an error?
329 	 */
330 	TAILQ_FOREACH(ep, &exit_list, next)
331 		(*ep->function)(td);
332 
333 	if (p->p_flag & P_PROFIL)
334 		stopprofclock(p);
335 	/*
336 	 * If parent is waiting for us to exit or exec,
337 	 * P_PPWAIT is set; we will wakeup the parent below.
338 	 */
339 	p->p_flag &= ~(P_TRACED | P_PPWAIT);
340 	SIGEMPTYSET(p->p_siglist);
341 	SIGEMPTYSET(lp->lwp_siglist);
342 	if (timevalisset(&p->p_realtimer.it_value))
343 		callout_stop(&p->p_ithandle);
344 
345 	/*
346 	 * Reset any sigio structures pointing to us as a result of
347 	 * F_SETOWN with our pid.
348 	 */
349 	funsetownlst(&p->p_sigiolst);
350 
351 	/*
352 	 * Close open files and release open-file table.
353 	 * This may block!
354 	 */
355 	fdfree(p, NULL);
356 
357 	if(p->p_leader->p_peers) {
358 		q = p->p_leader;
359 		while(q->p_peers != p)
360 			q = q->p_peers;
361 		q->p_peers = p->p_peers;
362 		wakeup((caddr_t)p->p_leader);
363 	}
364 
365 	/*
366 	 * XXX Shutdown SYSV semaphores
367 	 */
368 	semexit(p);
369 
370 	KKASSERT(p->p_numposixlocks == 0);
371 
372 	/* The next two chunks should probably be moved to vmspace_exit. */
373 	vm = p->p_vmspace;
374 
375 	/*
376 	 * Release upcalls associated with this process
377 	 */
378 	if (vm->vm_upcalls)
379 		upc_release(vm, lp);
380 
381 	/*
382 	 * Clean up data related to virtual kernel operation.  Clean up
383 	 * any vkernel context related to the current lwp now so we can
384 	 * destroy p_vkernel.
385 	 */
386 	if (p->p_vkernel) {
387 		vkernel_lwp_exit(lp);
388 		vkernel_exit(p);
389 	}
390 
391 	/*
392 	 * Release user portion of address space.
393 	 * This releases references to vnodes,
394 	 * which could cause I/O if the file has been unlinked.
395 	 * Need to do this early enough that we can still sleep.
396 	 * Can't free the entire vmspace as the kernel stack
397 	 * may be mapped within that space also.
398 	 *
399 	 * Processes sharing the same vmspace may exit in one order, and
400 	 * get cleaned up by vmspace_exit() in a different order.  The
401 	 * last exiting process to reach this point releases as much of
402 	 * the environment as it can, and the last process cleaned up
403 	 * by vmspace_exit() (which decrements exitingcnt) cleans up the
404 	 * remainder.
405 	 */
406 	++vm->vm_exitingcnt;
407 	sysref_put(&vm->vm_sysref);
408 
409 	if (SESS_LEADER(p)) {
410 		struct session *sp = p->p_session;
411 
412 		if (sp->s_ttyvp) {
413 			/*
414 			 * We are the controlling process.  Signal the
415 			 * foreground process group, drain the controlling
416 			 * terminal, and revoke access to the controlling
417 			 * terminal.
418 			 *
419 			 * NOTE: while waiting for the process group to exit
420 			 * it is possible that one of the processes in the
421 			 * group will revoke the tty, so the ttyclosesession()
422 			 * function will re-check sp->s_ttyvp.
423 			 */
424 			if (sp->s_ttyp && (sp->s_ttyp->t_session == sp)) {
425 				if (sp->s_ttyp->t_pgrp)
426 					pgsignal(sp->s_ttyp->t_pgrp, SIGHUP, 1);
427 				ttywait(sp->s_ttyp);
428 				ttyclosesession(sp, 1); /* also revoke */
429 			}
430 			/*
431 			 * Release the tty.  If someone has it open via
432 			 * /dev/tty then close it (since they no longer can
433 			 * once we've NULL'd it out).
434 			 */
435 			ttyclosesession(sp, 0);
436 
437 			/*
438 			 * s_ttyp is not zero'd; we use this to indicate
439 			 * that the session once had a controlling terminal.
440 			 * (for logging and informational purposes)
441 			 */
442 		}
443 		sp->s_leader = NULL;
444 	}
445 	fixjobc(p, p->p_pgrp, 0);
446 	(void)acct_process(p);
447 #ifdef KTRACE
448 	/*
449 	 * release trace file
450 	 */
451 	if (p->p_tracenode)
452 		ktrdestroy(&p->p_tracenode);
453 	p->p_traceflag = 0;
454 #endif
455 	/*
456 	 * Release reference to text vnode
457 	 */
458 	if ((vtmp = p->p_textvp) != NULL) {
459 		p->p_textvp = NULL;
460 		vrele(vtmp);
461 	}
462 
463 	/*
464 	 * Move the process to the zombie list.  This will block
465 	 * until the process p_lock count reaches 0.  The process will
466 	 * not be reaped until TDF_EXITING is set by cpu_thread_exit(),
467 	 * which is called from cpu_proc_exit().
468 	 */
469 	proc_move_allproc_zombie(p);
470 
471 	q = LIST_FIRST(&p->p_children);
472 	if (q)		/* only need this if any child is S_ZOMB */
473 		wakeup((caddr_t) initproc);
474 	for (; q != 0; q = nq) {
475 		nq = LIST_NEXT(q, p_sibling);
476 		LIST_REMOVE(q, p_sibling);
477 		LIST_INSERT_HEAD(&initproc->p_children, q, p_sibling);
478 		q->p_pptr = initproc;
479 		q->p_sigparent = SIGCHLD;
480 		/*
481 		 * Traced processes are killed
482 		 * since their existence means someone is screwing up.
483 		 */
484 		if (q->p_flag & P_TRACED) {
485 			q->p_flag &= ~P_TRACED;
486 			ksignal(q, SIGKILL);
487 		}
488 	}
489 
490 	/*
491 	 * Save exit status and final rusage info, adding in child rusage
492 	 * info and self times.
493 	 */
494 	p->p_xstat = rv;
495 	calcru_proc(p, &p->p_ru);
496 	ruadd(&p->p_ru, &p->p_cru);
497 
498 	/*
499 	 * notify interested parties of our demise.
500 	 */
501 	KNOTE(&p->p_klist, NOTE_EXIT);
502 
503 	/*
504 	 * Notify parent that we're gone.  If parent has the PS_NOCLDWAIT
505 	 * flag set, notify process 1 instead (and hope it will handle
506 	 * this situation).
507 	 */
508 	if (p->p_pptr->p_sigacts->ps_flag & PS_NOCLDWAIT) {
509 		struct proc *pp = p->p_pptr;
510 		proc_reparent(p, initproc);
511 		/*
512 		 * If this was the last child of our parent, notify
513 		 * parent, so in case he was wait(2)ing, he will
514 		 * continue.
515 		 */
516 		if (LIST_EMPTY(&pp->p_children))
517 			wakeup((caddr_t)pp);
518 	}
519 
520 	if (p->p_sigparent && p->p_pptr != initproc) {
521 	        ksignal(p->p_pptr, p->p_sigparent);
522 	} else {
523 	        ksignal(p->p_pptr, SIGCHLD);
524 	}
525 
526 	wakeup((caddr_t)p->p_pptr);
527 	/*
528 	 * cpu_exit is responsible for clearing curproc, since
529 	 * it is heavily integrated with the thread/switching sequence.
530 	 *
531 	 * Other substructures are freed from wait().
532 	 */
533 	plimit_free(p);
534 
535 	/*
536 	 * Release the current user process designation on the process so
537 	 * the userland scheduler can work in someone else.
538 	 */
539 	p->p_usched->release_curproc(lp);
540 
541 	/*
542 	 * Finally, call machine-dependent code to release as many of the
543 	 * lwp's resources as we can and halt execution of this thread.
544 	 */
545 	lwp_exit(1);
546 }
547 
548 /*
549  * Eventually called by every exiting LWP
550  */
551 void
552 lwp_exit(int masterexit)
553 {
554 	struct thread *td = curthread;
555 	struct lwp *lp = td->td_lwp;
556 	struct proc *p = lp->lwp_proc;
557 
558 	/*
559 	 * lwp_exit() may be called without setting LWP_WEXIT, so
560 	 * make sure it is set here.
561 	 */
562 	lp->lwp_flag |= LWP_WEXIT;
563 
564 	/*
565 	 * Clean up any virtualization
566 	 */
567 	if (lp->lwp_vkernel)
568 		vkernel_lwp_exit(lp);
569 
570 	/*
571 	 * Clean up any syscall-cached ucred
572 	 */
573 	if (td->td_ucred) {
574 		crfree(td->td_ucred);
575 		td->td_ucred = NULL;
576 	}
577 
578 	/*
579 	 * Nobody actually wakes us when the lock
580 	 * count reaches zero, so just wait one tick.
581 	 */
582 	while (lp->lwp_lock > 0)
583 		tsleep(lp, 0, "lwpexit", 1);
584 
585 	/* Hand down resource usage to our proc */
586 	ruadd(&p->p_ru, &lp->lwp_ru);
587 
588 	/*
589 	 * If we don't hold the process until the LWP is reaped wait*()
590 	 * may try to dispose of its vmspace before all the LWPs have
591 	 * actually terminated.
592 	 */
593 	PHOLD(p);
594 
595 	/*
596 	 * We have to use the reaper for all the LWPs except the one doing
597 	 * the master exit.  The LWP doing the master exit can just be
598 	 * left on p_lwps and the process reaper will deal with it
599 	 * synchronously, which is much faster.
600 	 */
601 	if (masterexit == 0) {
602 		lwp_rb_tree_RB_REMOVE(&p->p_lwp_tree, lp);
603 		--p->p_nthreads;
604 		wakeup(&p->p_nthreads);
605 		LIST_INSERT_HEAD(&deadlwp_list[mycpuid], lp, u.lwp_reap_entry);
606 		taskqueue_enqueue(taskqueue_thread[mycpuid], deadlwp_task[mycpuid]);
607 	} else {
608 		--p->p_nthreads;
609 	}
610 	biosched_done(curthread);
611 	cpu_lwp_exit();
612 }
613 
614 /*
615  * Wait until a lwp is completely dead.
616  *
617  * If the thread is still executing, which can't be waited upon,
618  * return failure.  The caller is responsible of waiting a little
619  * bit and checking again.
620  *
621  * Suggested use:
622  * while (!lwp_wait(lp))
623  *	tsleep(lp, 0, "lwpwait", 1);
624  */
625 static int
626 lwp_wait(struct lwp *lp)
627 {
628 	struct thread *td = lp->lwp_thread;;
629 
630 	KKASSERT(lwkt_preempted_proc() != lp);
631 
632 	while (lp->lwp_lock > 0)
633 		tsleep(lp, 0, "lwpwait1", 1);
634 
635 	lwkt_wait_free(td);
636 
637 	/*
638 	 * The lwp's thread may still be in the middle
639 	 * of switching away, we can't rip its stack out from
640 	 * under it until TDF_EXITING is set and both
641 	 * TDF_RUNNING and TDF_PREEMPT_LOCK are clear.
642 	 * TDF_PREEMPT_LOCK must be checked because TDF_RUNNING
643 	 * will be cleared temporarily if a thread gets
644 	 * preempted.
645 	 *
646 	 * YYY no wakeup occurs, so we simply return failure
647 	 * and let the caller deal with sleeping and calling
648 	 * us again.
649 	 */
650 	if ((td->td_flags & (TDF_RUNNING|TDF_PREEMPT_LOCK|TDF_EXITING)) !=
651 	    TDF_EXITING)
652 		return (0);
653 
654 	return (1);
655 }
656 
657 /*
658  * Release the resources associated with a lwp.
659  * The lwp must be completely dead.
660  */
661 void
662 lwp_dispose(struct lwp *lp)
663 {
664 	struct thread *td = lp->lwp_thread;;
665 
666 	KKASSERT(lwkt_preempted_proc() != lp);
667 	KKASSERT(td->td_refs == 0);
668 	KKASSERT((td->td_flags & (TDF_RUNNING|TDF_PREEMPT_LOCK|TDF_EXITING)) ==
669 		 TDF_EXITING);
670 
671 	PRELE(lp->lwp_proc);
672 	lp->lwp_proc = NULL;
673 	if (td != NULL) {
674 		td->td_proc = NULL;
675 		td->td_lwp = NULL;
676 		lp->lwp_thread = NULL;
677 		lwkt_free_thread(td);
678 	}
679 	kfree(lp, M_LWP);
680 }
681 
682 /*
683  * MPSAFE
684  */
685 int
686 sys_wait4(struct wait_args *uap)
687 {
688 	struct rusage rusage;
689 	int error, status;
690 
691 	error = kern_wait(uap->pid, (uap->status ? &status : NULL),
692 			  uap->options, (uap->rusage ? &rusage : NULL),
693 			  &uap->sysmsg_result);
694 
695 	if (error == 0 && uap->status)
696 		error = copyout(&status, uap->status, sizeof(*uap->status));
697 	if (error == 0 && uap->rusage)
698 		error = copyout(&rusage, uap->rusage, sizeof(*uap->rusage));
699 	return (error);
700 }
701 
702 /*
703  * wait1()
704  *
705  * wait_args(int pid, int *status, int options, struct rusage *rusage)
706  *
707  * MPALMOSTSAFE
708  */
709 int
710 kern_wait(pid_t pid, int *status, int options, struct rusage *rusage, int *res)
711 {
712 	struct thread *td = curthread;
713 	struct lwp *lp;
714 	struct proc *q = td->td_proc;
715 	struct proc *p, *t;
716 	int nfound, error;
717 
718 	if (pid == 0)
719 		pid = -q->p_pgid;
720 	if (options &~ (WUNTRACED|WNOHANG|WCONTINUED|WLINUXCLONE))
721 		return (EINVAL);
722 	get_mplock();
723 loop:
724 	/*
725 	 * Hack for backwards compatibility with badly written user code.
726 	 * Or perhaps we have to do this anyway, it is unclear. XXX
727 	 *
728 	 * The problem is that if a process group is stopped and the parent
729 	 * is doing a wait*(..., WUNTRACED, ...), it will see the STOP
730 	 * of the child and then stop itself when it tries to return from the
731 	 * system call.  When the process group is resumed the parent will
732 	 * then get the STOP status even though the child has now resumed
733 	 * (a followup wait*() will get the CONT status).
734 	 *
735 	 * Previously the CONT would overwrite the STOP because the tstop
736 	 * was handled within tsleep(), and the parent would only see
737 	 * the CONT when both are stopped and continued together.  This litte
738 	 * two-line hack restores this effect.
739 	 */
740 	while (q->p_stat == SSTOP)
741             tstop();
742 
743 	nfound = 0;
744 	LIST_FOREACH(p, &q->p_children, p_sibling) {
745 		if (pid != WAIT_ANY &&
746 		    p->p_pid != pid && p->p_pgid != -pid)
747 			continue;
748 
749 		/* This special case handles a kthread spawned by linux_clone
750 		 * (see linux_misc.c).  The linux_wait4 and linux_waitpid
751 		 * functions need to be able to distinguish between waiting
752 		 * on a process and waiting on a thread.  It is a thread if
753 		 * p_sigparent is not SIGCHLD, and the WLINUXCLONE option
754 		 * signifies we want to wait for threads and not processes.
755 		 */
756 		if ((p->p_sigparent != SIGCHLD) ^
757 		    ((options & WLINUXCLONE) != 0)) {
758 			continue;
759 		}
760 
761 		nfound++;
762 		if (p->p_stat == SZOMB) {
763 			/*
764 			 * We may go into SZOMB with threads still present.
765 			 * We must wait for them to exit before we can reap
766 			 * the master thread, otherwise we may race reaping
767 			 * non-master threads.
768 			 */
769 			while (p->p_nthreads > 0) {
770 				tsleep(&p->p_nthreads, 0, "lwpzomb", hz);
771 			}
772 
773 			/*
774 			 * Reap any LWPs left in p->p_lwps.  This is usually
775 			 * just the last LWP.  This must be done before
776 			 * we loop on p_lock since the lwps hold a ref on
777 			 * it as a vmspace interlock.
778 			 *
779 			 * Once that is accomplished p_nthreads had better
780 			 * be zero.
781 			 */
782 			while ((lp = RB_ROOT(&p->p_lwp_tree)) != NULL) {
783 				lwp_rb_tree_RB_REMOVE(&p->p_lwp_tree, lp);
784 				reaplwp(lp);
785 			}
786 			KKASSERT(p->p_nthreads == 0);
787 
788 			/*
789 			 * Don't do anything really bad until all references
790 			 * to the process go away.  This may include other
791 			 * LWPs which are still in the process of being
792 			 * reaped.  We can't just pull the rug out from under
793 			 * them because they may still be using the VM space.
794 			 *
795 			 * Certain kernel facilities such as /proc will also
796 			 * put a hold on the process for short periods of
797 			 * time.
798 			 */
799 			while (p->p_lock)
800 				tsleep(p, 0, "reap3", hz);
801 
802 			/* scheduling hook for heuristic */
803 			/* XXX no lwp available, we need a different heuristic */
804 			/*
805 			p->p_usched->heuristic_exiting(td->td_lwp, deadlp);
806 			*/
807 
808 			/* Take care of our return values. */
809 			*res = p->p_pid;
810 			if (status)
811 				*status = p->p_xstat;
812 			if (rusage)
813 				*rusage = p->p_ru;
814 			/*
815 			 * If we got the child via a ptrace 'attach',
816 			 * we need to give it back to the old parent.
817 			 */
818 			if (p->p_oppid && (t = pfind(p->p_oppid))) {
819 				p->p_oppid = 0;
820 				proc_reparent(p, t);
821 				ksignal(t, SIGCHLD);
822 				wakeup((caddr_t)t);
823 				error = 0;
824 				goto done;
825 			}
826 
827 			/*
828 			 * Unlink the proc from its process group so that
829 			 * the following operations won't lead to an
830 			 * inconsistent state for processes running down
831 			 * the zombie list.
832 			 */
833 			KKASSERT(p->p_lock == 0);
834 			proc_remove_zombie(p);
835 			leavepgrp(p);
836 
837 			p->p_xstat = 0;
838 			ruadd(&q->p_cru, &p->p_ru);
839 
840 			/*
841 			 * Decrement the count of procs running with this uid.
842 			 */
843 			chgproccnt(p->p_ucred->cr_ruidinfo, -1, 0);
844 
845 			/*
846 			 * Free up credentials.
847 			 */
848 			crfree(p->p_ucred);
849 			p->p_ucred = NULL;
850 
851 			/*
852 			 * Remove unused arguments
853 			 */
854 			if (p->p_args && --p->p_args->ar_ref == 0)
855 				FREE(p->p_args, M_PARGS);
856 
857 			if (--p->p_sigacts->ps_refcnt == 0) {
858 				kfree(p->p_sigacts, M_SUBPROC);
859 				p->p_sigacts = NULL;
860 			}
861 
862 			vm_waitproc(p);
863 			kfree(p, M_PROC);
864 			nprocs--;
865 			error = 0;
866 			goto done;
867 		}
868 		if (p->p_stat == SSTOP && (p->p_flag & P_WAITED) == 0 &&
869 		    (p->p_flag & P_TRACED || options & WUNTRACED)) {
870 			p->p_flag |= P_WAITED;
871 
872 			*res = p->p_pid;
873 			if (status)
874 				*status = W_STOPCODE(p->p_xstat);
875 			/* Zero rusage so we get something consistent. */
876 			if (rusage)
877 				bzero(rusage, sizeof(rusage));
878 			error = 0;
879 			goto done;
880 		}
881 		if (options & WCONTINUED && (p->p_flag & P_CONTINUED)) {
882 			*res = p->p_pid;
883 			p->p_flag &= ~P_CONTINUED;
884 
885 			if (status)
886 				*status = SIGCONT;
887 			error = 0;
888 			goto done;
889 		}
890 	}
891 	if (nfound == 0) {
892 		error = ECHILD;
893 		goto done;
894 	}
895 	if (options & WNOHANG) {
896 		*res = 0;
897 		error = 0;
898 		goto done;
899 	}
900 	error = tsleep((caddr_t)q, PCATCH, "wait", 0);
901 	if (error) {
902 done:
903 		rel_mplock();
904 		return (error);
905 	}
906 	goto loop;
907 }
908 
909 /*
910  * make process 'parent' the new parent of process 'child'.
911  */
912 void
913 proc_reparent(struct proc *child, struct proc *parent)
914 {
915 
916 	if (child->p_pptr == parent)
917 		return;
918 
919 	LIST_REMOVE(child, p_sibling);
920 	LIST_INSERT_HEAD(&parent->p_children, child, p_sibling);
921 	child->p_pptr = parent;
922 }
923 
924 /*
925  * The next two functions are to handle adding/deleting items on the
926  * exit callout list
927  *
928  * at_exit():
929  * Take the arguments given and put them onto the exit callout list,
930  * However first make sure that it's not already there.
931  * returns 0 on success.
932  */
933 
934 int
935 at_exit(exitlist_fn function)
936 {
937 	struct exitlist *ep;
938 
939 #ifdef INVARIANTS
940 	/* Be noisy if the programmer has lost track of things */
941 	if (rm_at_exit(function))
942 		kprintf("WARNING: exit callout entry (%p) already present\n",
943 		    function);
944 #endif
945 	ep = kmalloc(sizeof(*ep), M_ATEXIT, M_NOWAIT);
946 	if (ep == NULL)
947 		return (ENOMEM);
948 	ep->function = function;
949 	TAILQ_INSERT_TAIL(&exit_list, ep, next);
950 	return (0);
951 }
952 
953 /*
954  * Scan the exit callout list for the given item and remove it.
955  * Returns the number of items removed (0 or 1)
956  */
957 int
958 rm_at_exit(exitlist_fn function)
959 {
960 	struct exitlist *ep;
961 
962 	TAILQ_FOREACH(ep, &exit_list, next) {
963 		if (ep->function == function) {
964 			TAILQ_REMOVE(&exit_list, ep, next);
965 			kfree(ep, M_ATEXIT);
966 			return(1);
967 		}
968 	}
969 	return (0);
970 }
971 
972 /*
973  * LWP reaper related code.
974  */
975 static void
976 reaplwps(void *context, int dummy)
977 {
978 	struct lwplist *lwplist = context;
979 	struct lwp *lp;
980 
981 	get_mplock();
982 	while ((lp = LIST_FIRST(lwplist))) {
983 		LIST_REMOVE(lp, u.lwp_reap_entry);
984 		reaplwp(lp);
985 	}
986 	rel_mplock();
987 }
988 
989 static void
990 reaplwp(struct lwp *lp)
991 {
992 	while (lwp_wait(lp) == 0)
993 		tsleep(lp, 0, "lwpreap", 1);
994 	lwp_dispose(lp);
995 }
996 
997 static void
998 deadlwp_init(void)
999 {
1000 	int cpu;
1001 
1002 	for (cpu = 0; cpu < ncpus; cpu++) {
1003 		LIST_INIT(&deadlwp_list[cpu]);
1004 		deadlwp_task[cpu] = kmalloc(sizeof(*deadlwp_task[cpu]), M_DEVBUF, M_WAITOK);
1005 		TASK_INIT(deadlwp_task[cpu], 0, reaplwps, &deadlwp_list[cpu]);
1006 	}
1007 }
1008 
1009 SYSINIT(deadlwpinit, SI_SUB_CONFIGURE, SI_ORDER_ANY, deadlwp_init, NULL);
1010