xref: /dflybsd-src/sys/kern/kern_exit.c (revision 269ffd40655c28ffc04742b158e7500aec5b0166)
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 1.79 2007/03/12 21:05:48 corecode 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_zone.h>
78 #include <vm/vm_extern.h>
79 #include <sys/user.h>
80 
81 #include <sys/thread2.h>
82 
83 static MALLOC_DEFINE(M_ATEXIT, "atexit", "atexit callback");
84 static MALLOC_DEFINE(M_ZOMBIE, "zombie", "zombie proc status");
85 
86 /*
87  * callout list for things to do at exit time
88  */
89 struct exitlist {
90 	exitlist_fn function;
91 	TAILQ_ENTRY(exitlist) next;
92 };
93 
94 TAILQ_HEAD(exit_list_head, exitlist);
95 static struct exit_list_head exit_list = TAILQ_HEAD_INITIALIZER(exit_list);
96 
97 /*
98  * LWP reaper data
99  */
100 struct task *deadlwp_task[MAXCPU];
101 struct lwplist deadlwp_list[MAXCPU];
102 
103 /*
104  * exit --
105  *	Death of process.
106  *
107  * SYS_EXIT_ARGS(int rval)
108  */
109 int
110 sys_exit(struct exit_args *uap)
111 {
112 	exit1(W_EXITCODE(uap->rval, 0));
113 	/* NOTREACHED */
114 }
115 
116 /*
117  * Extended exit --
118  *	Death of a lwp or process with optional bells and whistles.
119  */
120 int
121 sys_extexit(struct extexit_args *uap)
122 {
123 	int action, who;
124 	int error;
125 
126 	action = EXTEXIT_ACTION(uap->how);
127 	who = EXTEXIT_WHO(uap->how);
128 
129 	/* Check parameters before we might perform some action */
130 	switch (who) {
131 	case EXTEXIT_PROC:
132 	case EXTEXIT_LWP:
133 		break;
134 
135 	default:
136 		return (EINVAL);
137 	}
138 
139 	switch (action) {
140 	case EXTEXIT_SIMPLE:
141 		break;
142 
143 	case EXTEXIT_SETINT:
144 		error = copyout(&uap->status, uap->addr, sizeof(uap->status));
145 		if (error)
146 			return (error);
147 		break;
148 
149 	default:
150 		return (EINVAL);
151 	}
152 
153 	switch (who) {
154 	case EXTEXIT_LWP:
155 		/*
156 		 * Be sure only to perform a simple lwp exit if there is at
157 		 * least one more lwp in the proc, which will call exit1()
158 		 * later, otherwise the proc will be an UNDEAD and not even a
159 		 * SZOMB!
160 		 */
161 		if (curproc->p_nthreads > 1)
162 			lwp_exit();
163 		/* else last lwp in proc:  do the real thing */
164 		/* FALLTHROUGH */
165 
166 	default:	/* to help gcc */
167 	case EXTEXIT_PROC:
168 		exit1(W_EXITCODE(uap->status, 0));
169 		/* NOTREACHED */
170 	}
171 
172 	/* NOTREACHED */
173 }
174 
175 void
176 killlwps(struct lwp *lp)
177 {
178 	struct proc *p = lp->lwp_proc;
179 	struct lwp *tlp;
180 
181 	KKASSERT((lp->lwp_flag & LWP_WEXIT) == 0);
182 
183 	FOREACH_LWP_IN_PROC(tlp, p) {
184 		if (tlp == lp)
185 			continue;	/* don't kill the current lwp */
186 		tlp->lwp_flag |= LWP_WEXIT;
187 		lwp_signotify(tlp);
188 	}
189 
190 	while (p->p_nthreads > 1) {
191 		if (bootverbose)
192 			kprintf("killlwps: waiting for %d lwps of pid %d to die\n",
193 				p->p_nthreads - 1, p->p_pid);
194 		tsleep(&p->p_nthreads, 0, "killlwps", hz);
195 	}
196 }
197 
198 /*
199  * Exit: deallocate address space and other resources, change proc state
200  * to zombie, and unlink proc from allproc and parent's lists.  Save exit
201  * status and rusage for wait().  Check for child processes and orphan them.
202  */
203 void
204 exit1(int rv)
205 {
206 	struct thread *td = curthread;
207 	struct proc *p = td->td_proc;
208 	struct lwp *lp = td->td_lwp;
209 	struct proc *q, *nq;
210 	struct vmspace *vm;
211 	struct vnode *vtmp;
212 	struct exitlist *ep;
213 
214 	if (p->p_pid == 1) {
215 		kprintf("init died (signal %d, exit %d)\n",
216 		    WTERMSIG(rv), WEXITSTATUS(rv));
217 		panic("Going nowhere without my init!");
218 	}
219 
220 	/*
221 	 * Kill all other threads if there are any.
222 	 *
223 	 * If some other thread initiated our exit, do so.
224 	 */
225 	if (lp->lwp_flag & LWP_WEXIT) {
226 		KKASSERT(p->p_nthreads > 1);
227 		lwp_exit();
228 	}
229 	if (p->p_nthreads > 1)
230 		killlwps(lp);
231 
232 	caps_exit(lp->lwp_thread);
233 	aio_proc_rundown(p);
234 
235 	/* are we a task leader? */
236 	if(p == p->p_leader) {
237         	struct kill_args killArgs;
238 		killArgs.signum = SIGKILL;
239 		q = p->p_peers;
240 		while(q) {
241 			killArgs.pid = q->p_pid;
242 			/*
243 		         * The interface for kill is better
244 			 * than the internal signal
245 			 */
246 			sys_kill(&killArgs);
247 			nq = q;
248 			q = q->p_peers;
249 		}
250 		while (p->p_peers)
251 		  tsleep((caddr_t)p, 0, "exit1", 0);
252 	}
253 
254 #ifdef PGINPROF
255 	vmsizmon();
256 #endif
257 	STOPEVENT(p, S_EXIT, rv);
258 	wakeup(&p->p_stype);	/* Wakeup anyone in procfs' PIOCWAIT */
259 
260 	/*
261 	 * Check if any loadable modules need anything done at process exit.
262 	 * e.g. SYSV IPC stuff
263 	 * XXX what if one of these generates an error?
264 	 */
265 	TAILQ_FOREACH(ep, &exit_list, next)
266 		(*ep->function)(td);
267 
268 	if (p->p_flag & P_PROFIL)
269 		stopprofclock(p);
270 	/*
271 	 * If parent is waiting for us to exit or exec,
272 	 * P_PPWAIT is set; we will wakeup the parent below.
273 	 */
274 	p->p_flag &= ~(P_TRACED | P_PPWAIT);
275 	p->p_flag |= P_WEXIT;
276 	SIGEMPTYSET(p->p_siglist);
277 	SIGEMPTYSET(lp->lwp_siglist);
278 	if (timevalisset(&p->p_realtimer.it_value))
279 		callout_stop(&p->p_ithandle);
280 
281 	/*
282 	 * Reset any sigio structures pointing to us as a result of
283 	 * F_SETOWN with our pid.
284 	 */
285 	funsetownlst(&p->p_sigiolst);
286 
287 	/*
288 	 * Close open files and release open-file table.
289 	 * This may block!
290 	 */
291 	fdfree(p);
292 	p->p_fd = NULL;
293 
294 	if(p->p_leader->p_peers) {
295 		q = p->p_leader;
296 		while(q->p_peers != p)
297 			q = q->p_peers;
298 		q->p_peers = p->p_peers;
299 		wakeup((caddr_t)p->p_leader);
300 	}
301 
302 	/*
303 	 * XXX Shutdown SYSV semaphores
304 	 */
305 	semexit(p);
306 
307 	KKASSERT(p->p_numposixlocks == 0);
308 
309 	/* The next two chunks should probably be moved to vmspace_exit. */
310 	vm = p->p_vmspace;
311 
312 	/*
313 	 * Release upcalls associated with this process
314 	 */
315 	if (vm->vm_upcalls)
316 		upc_release(vm, lp);
317 
318 	/* clean up data related to virtual kernel operation */
319 	if (p->p_vkernel)
320 		vkernel_exit(p);
321 
322 	/*
323 	 * Release user portion of address space.
324 	 * This releases references to vnodes,
325 	 * which could cause I/O if the file has been unlinked.
326 	 * Need to do this early enough that we can still sleep.
327 	 * Can't free the entire vmspace as the kernel stack
328 	 * may be mapped within that space also.
329 	 *
330 	 * Processes sharing the same vmspace may exit in one order, and
331 	 * get cleaned up by vmspace_exit() in a different order.  The
332 	 * last exiting process to reach this point releases as much of
333 	 * the environment as it can, and the last process cleaned up
334 	 * by vmspace_exit() (which decrements exitingcnt) cleans up the
335 	 * remainder.
336 	 */
337 	++vm->vm_exitingcnt;
338 	if (--vm->vm_refcnt == 0) {
339 		shmexit(vm);
340 		pmap_remove_pages(vmspace_pmap(vm), VM_MIN_USER_ADDRESS,
341 				  VM_MAX_USER_ADDRESS);
342 		vm_map_remove(&vm->vm_map, VM_MIN_USER_ADDRESS,
343 			      VM_MAX_USER_ADDRESS);
344 	}
345 
346 	if (SESS_LEADER(p)) {
347 		struct session *sp = p->p_session;
348 		struct vnode *vp;
349 
350 		if (sp->s_ttyvp) {
351 			/*
352 			 * We are the controlling process.  Signal the
353 			 * foreground process group, drain the controlling
354 			 * terminal, and revoke access to the controlling
355 			 * terminal.
356 			 *
357 			 * NOTE: while waiting for the process group to exit
358 			 * it is possible that one of the processes in the
359 			 * group will revoke the tty, so we have to recheck.
360 			 */
361 			if (sp->s_ttyp && (sp->s_ttyp->t_session == sp)) {
362 				if (sp->s_ttyp->t_pgrp)
363 					pgsignal(sp->s_ttyp->t_pgrp, SIGHUP, 1);
364 				(void) ttywait(sp->s_ttyp);
365 				/*
366 				 * The tty could have been revoked
367 				 * if we blocked.
368 				 */
369 				if ((vp = sp->s_ttyvp) != NULL) {
370 					ttyclosesession(sp, 0);
371 					vx_lock(vp);
372 					VOP_REVOKE(vp, REVOKEALL);
373 					vx_unlock(vp);
374 					vrele(vp);	/* s_ttyvp ref */
375 				}
376 			}
377 			/*
378 			 * Release the tty.  If someone has it open via
379 			 * /dev/tty then close it (since they no longer can
380 			 * once we've NULL'd it out).
381 			 */
382 			if (sp->s_ttyvp)
383 				ttyclosesession(sp, 1);
384 			/*
385 			 * s_ttyp is not zero'd; we use this to indicate
386 			 * that the session once had a controlling terminal.
387 			 * (for logging and informational purposes)
388 			 */
389 		}
390 		sp->s_leader = NULL;
391 	}
392 	fixjobc(p, p->p_pgrp, 0);
393 	(void)acct_process(p);
394 #ifdef KTRACE
395 	/*
396 	 * release trace file
397 	 */
398 	if (p->p_tracenode)
399 		ktrdestroy(&p->p_tracenode);
400 	p->p_traceflag = 0;
401 #endif
402 	/*
403 	 * Release reference to text vnode
404 	 */
405 	if ((vtmp = p->p_textvp) != NULL) {
406 		p->p_textvp = NULL;
407 		vrele(vtmp);
408 	}
409 
410 	/*
411 	 * Move the process to the zombie list.  This will block
412 	 * until the process p_lock count reaches 0.  The process will
413 	 * not be reaped until TDF_EXITING is set by cpu_thread_exit(),
414 	 * which is called from cpu_proc_exit().
415 	 */
416 	proc_move_allproc_zombie(p);
417 
418 	q = LIST_FIRST(&p->p_children);
419 	if (q)		/* only need this if any child is S_ZOMB */
420 		wakeup((caddr_t) initproc);
421 	for (; q != 0; q = nq) {
422 		nq = LIST_NEXT(q, p_sibling);
423 		LIST_REMOVE(q, p_sibling);
424 		LIST_INSERT_HEAD(&initproc->p_children, q, p_sibling);
425 		q->p_pptr = initproc;
426 		q->p_sigparent = SIGCHLD;
427 		/*
428 		 * Traced processes are killed
429 		 * since their existence means someone is screwing up.
430 		 */
431 		if (q->p_flag & P_TRACED) {
432 			q->p_flag &= ~P_TRACED;
433 			ksignal(q, SIGKILL);
434 		}
435 	}
436 
437 	/*
438 	 * Save exit status and final rusage info, adding in child rusage
439 	 * info and self times.
440 	 */
441 	p->p_xstat = rv;
442 	calcru_proc(p, &p->p_ru);
443 	ruadd(&p->p_ru, &p->p_cru);
444 
445 	/*
446 	 * notify interested parties of our demise.
447 	 */
448 	KNOTE(&p->p_klist, NOTE_EXIT);
449 
450 	/*
451 	 * Notify parent that we're gone.  If parent has the PS_NOCLDWAIT
452 	 * flag set, notify process 1 instead (and hope it will handle
453 	 * this situation).
454 	 */
455 	if (p->p_pptr->p_sigacts->ps_flag & PS_NOCLDWAIT) {
456 		struct proc *pp = p->p_pptr;
457 		proc_reparent(p, initproc);
458 		/*
459 		 * If this was the last child of our parent, notify
460 		 * parent, so in case he was wait(2)ing, he will
461 		 * continue.
462 		 */
463 		if (LIST_EMPTY(&pp->p_children))
464 			wakeup((caddr_t)pp);
465 	}
466 
467 	if (p->p_sigparent && p->p_pptr != initproc) {
468 	        ksignal(p->p_pptr, p->p_sigparent);
469 	} else {
470 	        ksignal(p->p_pptr, SIGCHLD);
471 	}
472 
473 	wakeup((caddr_t)p->p_pptr);
474 	/*
475 	 * cpu_exit is responsible for clearing curproc, since
476 	 * it is heavily integrated with the thread/switching sequence.
477 	 *
478 	 * Other substructures are freed from wait().
479 	 */
480 	plimit_free(&p->p_limit);
481 
482 	/*
483 	 * Release the current user process designation on the process so
484 	 * the userland scheduler can work in someone else.
485 	 */
486 	p->p_usched->release_curproc(lp);
487 
488 	/*
489 	 * Finally, call machine-dependent code to release the remaining
490 	 * resources including address space, the kernel stack and pcb.
491 	 * The address space is released by "vmspace_free(p->p_vmspace)";
492 	 * This is machine-dependent, as we may have to change stacks
493 	 * or ensure that the current one isn't reallocated before we
494 	 * finish.  cpu_exit will end with a call to cpu_switch(), finishing
495 	 * our execution (pun intended).
496 	 */
497 	lwp_exit();
498 }
499 
500 void
501 lwp_exit(void)
502 {
503 	struct lwp *lp = curthread->td_lwp;
504 	struct proc *p = lp->lwp_proc;
505 
506 	/*
507 	 * Nobody actually wakes us when the lock
508 	 * count reaches zero, so just wait one tick.
509 	 */
510 	while (lp->lwp_lock > 0)
511 		tsleep(lp, 0, "lwpexit", 1);
512 
513 	/* Hand down resource usage to our proc */
514 	ruadd(&p->p_ru, &lp->lwp_ru);
515 
516 	--p->p_nthreads;
517 	LIST_REMOVE(lp, lwp_list);
518 	wakeup(&p->p_nthreads);
519 	LIST_INSERT_HEAD(&deadlwp_list[mycpuid], lp, lwp_list);
520 	taskqueue_enqueue(taskqueue_thread[mycpuid], deadlwp_task[mycpuid]);
521 	cpu_lwp_exit();
522 }
523 
524 /*
525  * Wait until a lwp is completely dead.
526  *
527  * If the thread is still executing, which can't be waited upon,
528  * return failure.  The caller is responsible of waiting a little
529  * bit and checking again.
530  *
531  * Suggested use:
532  * while (!lwp_wait(lp))
533  *	tsleep(lp, 0, "lwpwait", 1);
534  */
535 static int
536 lwp_wait(struct lwp *lp)
537 {
538 	struct thread *td = lp->lwp_thread;;
539 
540 	KKASSERT(lwkt_preempted_proc() != lp);
541 
542 	while (lp->lwp_lock > 0)
543 		tsleep(lp, 0, "lwpwait1", 1);
544 
545 	lwkt_wait_free(td);
546 
547 	/*
548 	 * The lwp's thread may still be in the middle
549 	 * of switching away, we can't rip its stack out from
550 	 * under it until TDF_EXITING is set and both
551 	 * TDF_RUNNING and TDF_PREEMPT_LOCK are clear.
552 	 * TDF_PREEMPT_LOCK must be checked because TDF_RUNNING
553 	 * will be cleared temporarily if a thread gets
554 	 * preempted.
555 	 *
556 	 * YYY no wakeup occurs, so we simply return failure
557 	 * and let the caller deal with sleeping and calling
558 	 * us again.
559 	 */
560 	if ((td->td_flags & (TDF_RUNNING|TDF_PREEMPT_LOCK|TDF_EXITING)) !=
561 	    TDF_EXITING)
562 		return (0);
563 
564 	return (1);
565 }
566 
567 /*
568  * Release the resources associated with a lwp.
569  * The lwp must be completely dead.
570  */
571 void
572 lwp_dispose(struct lwp *lp)
573 {
574 	struct thread *td = lp->lwp_thread;;
575 
576 	KKASSERT(lwkt_preempted_proc() != lp);
577 	KKASSERT(td->td_refs == 0);
578 	KKASSERT((td->td_flags & (TDF_RUNNING|TDF_PREEMPT_LOCK|TDF_EXITING)) ==
579 		 TDF_EXITING);
580 
581 	if (td != NULL) {
582 		td->td_proc = NULL;
583 		td->td_lwp = NULL;
584 		lp->lwp_thread = NULL;
585 		lwkt_free_thread(td);
586 	}
587 	zfree(lwp_zone, lp);
588 }
589 
590 int
591 sys_wait4(struct wait_args *uap)
592 {
593 	struct rusage rusage;
594 	int error, status;
595 
596 	error = kern_wait(uap->pid, uap->status ? &status : NULL,
597 	    uap->options, uap->rusage ? &rusage : NULL, &uap->sysmsg_fds[0]);
598 
599 	if (error == 0 && uap->status)
600 		error = copyout(&status, uap->status, sizeof(*uap->status));
601 	if (error == 0 && uap->rusage)
602 		error = copyout(&rusage, uap->rusage, sizeof(*uap->rusage));
603 	return (error);
604 }
605 
606 /*
607  * wait1()
608  *
609  * wait_args(int pid, int *status, int options, struct rusage *rusage)
610  */
611 int
612 kern_wait(pid_t pid, int *status, int options, struct rusage *rusage, int *res)
613 {
614 	struct thread *td = curthread;
615 	struct proc *q = td->td_proc;
616 	struct proc *p, *t;
617 	int nfound, error;
618 
619 	if (pid == 0)
620 		pid = -q->p_pgid;
621 	if (options &~ (WUNTRACED|WNOHANG|WLINUXCLONE))
622 		return (EINVAL);
623 loop:
624 	/*
625 	 * Hack for backwards compatibility with badly written user code.
626 	 * Or perhaps we have to do this anyway, it is unclear. XXX
627 	 *
628 	 * The problem is that if a process group is stopped and the parent
629 	 * is doing a wait*(..., WUNTRACED, ...), it will see the STOP
630 	 * of the child and then stop itself when it tries to return from the
631 	 * system call.  When the process group is resumed the parent will
632 	 * then get the STOP status even though the child has now resumed
633 	 * (a followup wait*() will get the CONT status).
634 	 *
635 	 * Previously the CONT would overwrite the STOP because the tstop
636 	 * was handled within tsleep(), and the parent would only see
637 	 * the CONT when both are stopped and continued together.  This litte
638 	 * two-line hack restores this effect.
639 	 */
640 	while (q->p_stat == SSTOP)
641             tstop();
642 
643 	nfound = 0;
644 	LIST_FOREACH(p, &q->p_children, p_sibling) {
645 		if (pid != WAIT_ANY &&
646 		    p->p_pid != pid && p->p_pgid != -pid)
647 			continue;
648 
649 		/* This special case handles a kthread spawned by linux_clone
650 		 * (see linux_misc.c).  The linux_wait4 and linux_waitpid
651 		 * functions need to be able to distinguish between waiting
652 		 * on a process and waiting on a thread.  It is a thread if
653 		 * p_sigparent is not SIGCHLD, and the WLINUXCLONE option
654 		 * signifies we want to wait for threads and not processes.
655 		 */
656 		if ((p->p_sigparent != SIGCHLD) ^
657 		    ((options & WLINUXCLONE) != 0)) {
658 			continue;
659 		}
660 
661 		nfound++;
662 		if (p->p_stat == SZOMB) {
663 			/*
664 			 * Other kernel threads may be in the middle of
665 			 * accessing the proc.  For example, kern/kern_proc.c
666 			 * could be blocked writing proc data to a sysctl.
667 			 * At the moment, if this occurs, we are not woken
668 			 * up and rely on a one-second retry.
669 			 */
670 			while (p->p_lock)
671 				tsleep(p, 0, "reap3", hz);
672 
673 			/* scheduling hook for heuristic */
674 			/* XXX no lwp available, we need a different heuristic */
675 			/*
676 			p->p_usched->heuristic_exiting(td->td_lwp, deadlp);
677 			*/
678 
679 			/* Take care of our return values. */
680 			*res = p->p_pid;
681 			if (status)
682 				*status = p->p_xstat;
683 			if (rusage)
684 				*rusage = p->p_ru;
685 			/*
686 			 * If we got the child via a ptrace 'attach',
687 			 * we need to give it back to the old parent.
688 			 */
689 			if (p->p_oppid && (t = pfind(p->p_oppid))) {
690 				p->p_oppid = 0;
691 				proc_reparent(p, t);
692 				ksignal(t, SIGCHLD);
693 				wakeup((caddr_t)t);
694 				return (0);
695 			}
696 			p->p_xstat = 0;
697 			ruadd(&q->p_cru, &p->p_ru);
698 
699 			/*
700 			 * Decrement the count of procs running with this uid.
701 			 */
702 			chgproccnt(p->p_ucred->cr_ruidinfo, -1, 0);
703 
704 			/*
705 			 * Free up credentials.
706 			 */
707 			crfree(p->p_ucred);
708 			p->p_ucred = NULL;
709 
710 			/*
711 			 * Remove unused arguments
712 			 */
713 			if (p->p_args && --p->p_args->ar_ref == 0)
714 				FREE(p->p_args, M_PARGS);
715 
716 			/*
717 			 * Finally finished with old proc entry.
718 			 * Unlink it from its process group and free it.
719 			 */
720 			proc_remove_zombie(p);
721 			leavepgrp(p);
722 
723 			if (--p->p_sigacts->ps_refcnt == 0) {
724 				kfree(p->p_sigacts, M_SUBPROC);
725 				p->p_sigacts = NULL;
726 			}
727 
728 			vm_waitproc(p);
729 			zfree(proc_zone, p);
730 			nprocs--;
731 			return (0);
732 		}
733 		if (p->p_stat == SSTOP && (p->p_flag & P_WAITED) == 0 &&
734 		    (p->p_flag & P_TRACED || options & WUNTRACED)) {
735 			p->p_flag |= P_WAITED;
736 
737 			*res = p->p_pid;
738 			if (status)
739 				*status = W_STOPCODE(p->p_xstat);
740 			/* Zero rusage so we get something consistent. */
741 			if (rusage)
742 				bzero(rusage, sizeof(rusage));
743 			return (0);
744 		}
745 	}
746 	if (nfound == 0)
747 		return (ECHILD);
748 	if (options & WNOHANG) {
749 		*res = 0;
750 		return (0);
751 	}
752 	error = tsleep((caddr_t)q, PCATCH, "wait", 0);
753 	if (error)
754 		return (error);
755 	goto loop;
756 }
757 
758 /*
759  * make process 'parent' the new parent of process 'child'.
760  */
761 void
762 proc_reparent(struct proc *child, struct proc *parent)
763 {
764 
765 	if (child->p_pptr == parent)
766 		return;
767 
768 	LIST_REMOVE(child, p_sibling);
769 	LIST_INSERT_HEAD(&parent->p_children, child, p_sibling);
770 	child->p_pptr = parent;
771 }
772 
773 /*
774  * The next two functions are to handle adding/deleting items on the
775  * exit callout list
776  *
777  * at_exit():
778  * Take the arguments given and put them onto the exit callout list,
779  * However first make sure that it's not already there.
780  * returns 0 on success.
781  */
782 
783 int
784 at_exit(exitlist_fn function)
785 {
786 	struct exitlist *ep;
787 
788 #ifdef INVARIANTS
789 	/* Be noisy if the programmer has lost track of things */
790 	if (rm_at_exit(function))
791 		kprintf("WARNING: exit callout entry (%p) already present\n",
792 		    function);
793 #endif
794 	ep = kmalloc(sizeof(*ep), M_ATEXIT, M_NOWAIT);
795 	if (ep == NULL)
796 		return (ENOMEM);
797 	ep->function = function;
798 	TAILQ_INSERT_TAIL(&exit_list, ep, next);
799 	return (0);
800 }
801 
802 /*
803  * Scan the exit callout list for the given item and remove it.
804  * Returns the number of items removed (0 or 1)
805  */
806 int
807 rm_at_exit(exitlist_fn function)
808 {
809 	struct exitlist *ep;
810 
811 	TAILQ_FOREACH(ep, &exit_list, next) {
812 		if (ep->function == function) {
813 			TAILQ_REMOVE(&exit_list, ep, next);
814 			kfree(ep, M_ATEXIT);
815 			return(1);
816 		}
817 	}
818 	return (0);
819 }
820 
821 
822 /*
823  * LWP reaper related code.
824  */
825 
826 static void
827 reaplwps(void *context, int dummy)
828 {
829 	struct lwplist *lwplist = context;
830 	struct lwp *lp;
831 
832 	while ((lp = LIST_FIRST(lwplist))) {
833 		if (!lwp_wait(lp))
834 			tsleep(lp, 0, "lwpreap", 1);
835 		LIST_REMOVE(lp, lwp_list);
836 		lwp_dispose(lp);
837 	}
838 }
839 
840 static void
841 deadlwp_init(void)
842 {
843 	int cpu;
844 
845 	for (cpu = 0; cpu < ncpus; cpu++) {
846 		LIST_INIT(&deadlwp_list[cpu]);
847 		deadlwp_task[cpu] = kmalloc(sizeof(*deadlwp_task[cpu]), M_DEVBUF, M_WAITOK);
848 		TASK_INIT(deadlwp_task[cpu], 0, reaplwps, &deadlwp_list[cpu]);
849 	}
850 }
851 
852 SYSINIT(deadlwpinit, SI_SUB_CONFIGURE, SI_ORDER_ANY, deadlwp_init, NULL);
853