xref: /netbsd-src/sys/kern/kern_exit.c (revision d708163f7f0abab4c3b88f79c8cf7ec8bc530426)
1 /*	$NetBSD: kern_exit.c,v 1.127 2003/11/06 09:30:13 dsl Exp $	*/
2 
3 /*-
4  * Copyright (c) 1998, 1999 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.
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  * 3. All advertising materials mentioning features or use of this software
20  *    must display the following acknowledgement:
21  *	This product includes software developed by the NetBSD
22  *	Foundation, Inc. and its contributors.
23  * 4. Neither the name of The NetBSD Foundation nor the names of its
24  *    contributors may be used to endorse or promote products derived
25  *    from this software without specific prior written permission.
26  *
27  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
28  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
29  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
30  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
31  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
32  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
33  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
34  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
35  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
36  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
37  * POSSIBILITY OF SUCH DAMAGE.
38  */
39 
40 /*
41  * Copyright (c) 1982, 1986, 1989, 1991, 1993
42  *	The Regents of the University of California.  All rights reserved.
43  * (c) UNIX System Laboratories, Inc.
44  * All or some portions of this file are derived from material licensed
45  * to the University of California by American Telephone and Telegraph
46  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
47  * the permission of UNIX System Laboratories, Inc.
48  *
49  * Redistribution and use in source and binary forms, with or without
50  * modification, are permitted provided that the following conditions
51  * are met:
52  * 1. Redistributions of source code must retain the above copyright
53  *    notice, this list of conditions and the following disclaimer.
54  * 2. Redistributions in binary form must reproduce the above copyright
55  *    notice, this list of conditions and the following disclaimer in the
56  *    documentation and/or other materials provided with the distribution.
57  * 3. Neither the name of the University nor the names of its contributors
58  *    may be used to endorse or promote products derived from this software
59  *    without specific prior written permission.
60  *
61  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
62  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
63  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
64  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
65  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
66  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
67  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
68  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
69  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
70  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
71  * SUCH DAMAGE.
72  *
73  *	@(#)kern_exit.c	8.10 (Berkeley) 2/23/95
74  */
75 
76 #include <sys/cdefs.h>
77 __KERNEL_RCSID(0, "$NetBSD: kern_exit.c,v 1.127 2003/11/06 09:30:13 dsl Exp $");
78 
79 #include "opt_ktrace.h"
80 #include "opt_perfctrs.h"
81 #include "opt_systrace.h"
82 #include "opt_sysv.h"
83 
84 #include <sys/param.h>
85 #include <sys/systm.h>
86 #include <sys/ioctl.h>
87 #include <sys/proc.h>
88 #include <sys/tty.h>
89 #include <sys/time.h>
90 #include <sys/resource.h>
91 #include <sys/kernel.h>
92 #include <sys/ktrace.h>
93 #include <sys/proc.h>
94 #include <sys/buf.h>
95 #include <sys/wait.h>
96 #include <sys/file.h>
97 #include <sys/vnode.h>
98 #include <sys/syslog.h>
99 #include <sys/malloc.h>
100 #include <sys/pool.h>
101 #include <sys/resourcevar.h>
102 #if defined(PERFCTRS)
103 #include <sys/pmc.h>
104 #endif
105 #include <sys/ptrace.h>
106 #include <sys/acct.h>
107 #include <sys/filedesc.h>
108 #include <sys/ras.h>
109 #include <sys/signalvar.h>
110 #include <sys/sched.h>
111 #include <sys/sa.h>
112 #include <sys/savar.h>
113 #include <sys/mount.h>
114 #include <sys/syscallargs.h>
115 #include <sys/systrace.h>
116 
117 #include <machine/cpu.h>
118 
119 #include <uvm/uvm_extern.h>
120 
121 #define DEBUG_EXIT
122 
123 #ifdef DEBUG_EXIT
124 int debug_exit = 0;
125 #define DPRINTF(x) if (debug_exit) printf x
126 #else
127 #define DPRINTF(x)
128 #endif
129 
130 static void lwp_exit_hook(struct lwp *, void *);
131 static void exit_psignal(struct proc *, struct proc *);
132 
133 /*
134  * Fill in the appropriate signal information, and kill the parent.
135  */
136 static void
137 exit_psignal(struct proc *p, struct proc *pp)
138 {
139 	ksiginfo_t ksi;
140 
141 	(void)memset(&ksi, 0, sizeof(ksi));
142 	if ((ksi.ksi_signo = P_EXITSIG(p)) == SIGCHLD) {
143 		if (WIFSIGNALED(p->p_xstat)) {
144 			if (WCOREDUMP(p->p_xstat))
145 				ksi.ksi_code = CLD_DUMPED;
146 			else
147 				ksi.ksi_code = CLD_KILLED;
148 		} else {
149 			ksi.ksi_code = CLD_EXITED;
150 		}
151 	}
152 	/*
153 	 * we fill those in, even for non-SIGCHLD.
154 	 */
155 	ksi.ksi_pid = p->p_pid;
156 	ksi.ksi_uid = p->p_ucred->cr_uid;
157 	ksi.ksi_status = p->p_xstat;
158 	/* XXX: is this still valid? */
159 	ksi.ksi_utime = p->p_ru->ru_utime.tv_sec;
160 	ksi.ksi_stime = p->p_ru->ru_stime.tv_sec;
161 	kpsignal(pp, &ksi, NULL);
162 }
163 
164 /*
165  * exit --
166  *	Death of process.
167  */
168 int
169 sys_exit(struct lwp *l, void *v, register_t *retval)
170 {
171 	struct sys_exit_args /* {
172 		syscallarg(int)	rval;
173 	} */ *uap = v;
174 
175 	/* Don't call exit1() multiple times in the same process.*/
176 	if (l->l_proc->p_flag & P_WEXIT)
177 		lwp_exit(l);
178 
179 	exit1(l, W_EXITCODE(SCARG(uap, rval), 0));
180 	/* NOTREACHED */
181 	return (0);
182 }
183 
184 /*
185  * Exit: deallocate address space and other resources, change proc state
186  * to zombie, and unlink proc from allproc and parent's lists.  Save exit
187  * status and rusage for wait().  Check for child processes and orphan them.
188  */
189 void
190 exit1(struct lwp *l, int rv)
191 {
192 	struct proc	*p, *q, *nq;
193 	int		s, sa;
194 
195 	p = l->l_proc;
196 
197 	if (__predict_false(p == initproc))
198 		panic("init died (signal %d, exit %d)",
199 		    WTERMSIG(rv), WEXITSTATUS(rv));
200 
201 	p->p_flag |= P_WEXIT;
202 
203 	DPRINTF(("exit1: %d.%d exiting.\n", p->p_pid, l->l_lid));
204 	/*
205 	 * Disable scheduler activation upcalls.
206 	 * We're trying to get out of here.
207 	 */
208 	sa = 0;
209 	if (p->p_sa != NULL) {
210 		l->l_flag &= ~L_SA;
211 #if 0
212 		p->p_flag &= ~P_SA;
213 #endif
214 		sa = 1;
215 	}
216 
217 #ifdef PGINPROF
218 	vmsizmon();
219 #endif
220 	if (p->p_flag & P_PROFIL)
221 		stopprofclock(p);
222 	p->p_ru = pool_get(&rusage_pool, PR_WAITOK);
223 	/*
224 	 * If parent is waiting for us to exit or exec, P_PPWAIT is set; we
225 	 * wake up the parent early to avoid deadlock.
226 	 */
227 	if (p->p_flag & P_PPWAIT) {
228 		p->p_flag &= ~P_PPWAIT;
229 		wakeup(p->p_pptr);
230 	}
231 	sigfillset(&p->p_sigctx.ps_sigignore);
232 	sigemptyset(&p->p_sigctx.ps_siglist);
233 	p->p_sigctx.ps_sigcheck = 0;
234 	timers_free(p, TIMERS_ALL);
235 
236 	if (sa || (p->p_nlwps > 1))
237 		exit_lwps(l);
238 
239 #if defined(__HAVE_RAS)
240 	ras_purgeall(p);
241 #endif
242 
243 	/*
244 	 * Close open files and release open-file table.
245 	 * This may block!
246 	 */
247 	fdfree(p);
248 	cwdfree(p);
249 
250 	doexithooks(p);
251 
252 	if (SESS_LEADER(p)) {
253 		struct session *sp = p->p_session;
254 		struct tty *tp;
255 
256 		if (sp->s_ttyvp) {
257 			/*
258 			 * Controlling process.
259 			 * Signal foreground pgrp,
260 			 * drain controlling terminal
261 			 * and revoke access to controlling terminal.
262 			 */
263 			tp = sp->s_ttyp;
264 			TTY_LOCK(tp);
265 			if (tp->t_session == sp) {
266 				if (tp->t_pgrp)
267 					pgsignal(tp->t_pgrp, SIGHUP, 1);
268 				/* we can't guarantee the revoke will do this */
269 				tp->t_pgrp = NULL;
270 				tp->t_session = NULL;
271 				TTY_UNLOCK(tp);
272 				SESSRELE(sp);
273 				(void) ttywait(tp);
274 				/*
275 				 * The tty could have been revoked
276 				 * if we blocked.
277 				 */
278 				if (sp->s_ttyvp)
279 					VOP_REVOKE(sp->s_ttyvp, REVOKEALL);
280 			} else
281 				TTY_UNLOCK(tp);
282 			if (sp->s_ttyvp)
283 				vrele(sp->s_ttyvp);
284 			sp->s_ttyvp = NULL;
285 			/*
286 			 * s_ttyp is not zero'd; we use this to indicate
287 			 * that the session once had a controlling terminal.
288 			 * (for logging and informational purposes)
289 			 */
290 		}
291 		sp->s_leader = NULL;
292 	}
293 	fixjobc(p, p->p_pgrp, 0);
294 	(void)acct_process(p);
295 #ifdef KTRACE
296 	/*
297 	 * release trace file
298 	 */
299 	ktrderef(p);
300 #endif
301 #ifdef SYSTRACE
302 	systrace_sys_exit(p);
303 #endif
304 	/*
305 	 * If emulation has process exit hook, call it now.
306 	 */
307 	if (p->p_emul->e_proc_exit)
308 		(*p->p_emul->e_proc_exit)(p);
309 
310 	/*
311 	 * Give orphaned children to init(8).
312 	 */
313 	q = LIST_FIRST(&p->p_children);
314 	if (q)		/* only need this if any child is S_ZOMB */
315 		wakeup(initproc);
316 	for (; q != 0; q = nq) {
317 		nq = LIST_NEXT(q, p_sibling);
318 
319 		/*
320 		 * Traced processes are killed since their existence
321 		 * means someone is screwing up. Since we reset the
322 		 * trace flags, the logic in sys_wait4() would not be
323 		 * triggered to reparent the process to its
324 		 * original parent, so we must do this here.
325 		 */
326 		if (q->p_flag & P_TRACED) {
327 			if (q->p_opptr != q->p_pptr) {
328 				struct proc *t = q->p_opptr;
329 				proc_reparent(q, t ? t : initproc);
330 				q->p_opptr = NULL;
331 			} else
332 				proc_reparent(q, initproc);
333 			q->p_flag &= ~(P_TRACED|P_WAITED|P_FSTRACE);
334 			psignal(q, SIGKILL);
335 		} else {
336 			proc_reparent(q, initproc);
337 		}
338 	}
339 
340 	/*
341 	 * Reset p_opptr pointer of all former children which got
342 	 * traced by another process and were reparented. We reset
343 	 * it to NULL here; the trace detach code then reparents
344 	 * the child to initproc. We only check allproc list, since
345 	 * eventual former children on zombproc list won't reference
346 	 * p_opptr anymore.
347 	 */
348 	if (p->p_flag & P_CHTRACED) {
349 		struct proc *t;
350 
351 		proclist_lock_read();
352 
353 		LIST_FOREACH(t, &allproc, p_list) {
354 			if (t->p_opptr == p)
355 				t->p_opptr = NULL;
356 		}
357 
358 		proclist_unlock_read();
359 	}
360 
361 	/*
362 	 * Save exit status and final rusage info, adding in child rusage
363 	 * info and self times.
364 	 * In order to pick up the time for the current execution, we must
365 	 * do this before unlinking the lwp from l_list.
366 	 */
367 	p->p_xstat = rv;
368 	*p->p_ru = p->p_stats->p_ru;
369 	calcru(p, &p->p_ru->ru_utime, &p->p_ru->ru_stime, NULL);
370 	ruadd(p->p_ru, &p->p_stats->p_cru);
371 
372 	/*
373 	 * NOTE: WE ARE NO LONGER ALLOWED TO SLEEP!
374 	 */
375 
376 	/*
377 	 * Move proc from allproc to zombproc, but do not yet
378 	 * wake up the reaper.  We will put the proc on the
379 	 * deadproc list later (using the p_dead member), and
380 	 * wake up the reaper when we do.
381 	 * Changing the state to SDEAD stops it being found by pfind().
382 	 */
383 	s = proclist_lock_write();
384 	p->p_stat = SDEAD;
385 	p->p_nrlwps--;
386 	l->l_stat = LSDEAD;
387 	LIST_REMOVE(p, p_list);
388 	LIST_INSERT_HEAD(&zombproc, p, p_list);
389 	LIST_REMOVE(l, l_list);
390 	l->l_flag |= L_DETACHED;
391 	proclist_unlock_write(s);
392 
393 	/*
394 	 * Notify interested parties of our demise.
395 	 */
396 	KNOTE(&p->p_klist, NOTE_EXIT);
397 
398 #if PERFCTRS
399 	/*
400 	 * Save final PMC information in parent process & clean up.
401 	 */
402 	if (PMC_ENABLED(p)) {
403 		pmc_save_context(p);
404 		pmc_accumulate(p->p_pptr, p);
405 		pmc_process_exit(p);
406 	}
407 #endif
408 
409 	/*
410 	 * Notify parent that we're gone.  If parent has the P_NOCLDWAIT
411 	 * flag set, notify init instead (and hope it will handle
412 	 * this situation).
413 	 */
414 	if (p->p_pptr->p_flag & P_NOCLDWAIT) {
415 		struct proc *pp = p->p_pptr;
416 		proc_reparent(p, initproc);
417 		/*
418 		 * If this was the last child of our parent, notify
419 		 * parent, so in case he was wait(2)ing, he will
420 		 * continue.
421 		 */
422 		if (LIST_FIRST(&pp->p_children) == NULL)
423 			wakeup(pp);
424 	}
425 
426 	/*
427 	 * Release the process's signal state.
428 	 */
429 	sigactsfree(p);
430 
431 	/*
432 	 * Clear curlwp after we've done all operations
433 	 * that could block, and before tearing down the rest
434 	 * of the process state that might be used from clock, etc.
435 	 * Also, can't clear curlwp while we're still runnable,
436 	 * as we're not on a run queue (we are current, just not
437 	 * a proper proc any longer!).
438 	 *
439 	 * Other substructures are freed from wait().
440 	 */
441 	curlwp = NULL;
442 	limfree(p->p_limit);
443 	pstatsfree(p->p_stats);
444 	p->p_limit = NULL;
445 
446 	/* This process no longer needs to hold the kernel lock. */
447 	KERNEL_PROC_UNLOCK(l);
448 
449 	/*
450 	 * Finally, call machine-dependent code to switch to a new
451 	 * context (possibly the idle context).  Once we are no longer
452 	 * using the dead process's vmspace and stack, exit2() will be
453 	 * called to schedule those resources to be released by the
454 	 * reaper thread.
455 	 *
456 	 * Note that cpu_exit() will end with a call equivalent to
457 	 * cpu_switch(), finishing our execution (pun intended).
458 	 */
459 
460 	cpu_exit(l, 1);
461 }
462 
463 void
464 exit_lwps(struct lwp *l)
465 {
466 	struct proc *p;
467 	struct lwp *l2;
468 	int s, error;
469 	lwpid_t		waited;
470 
471 	p = l->l_proc;
472 
473 	/* XXX SMP
474 	 * This would be the right place to IPI any LWPs running on
475 	 * other processors so that they can notice the userret exit hook.
476 	 */
477 	p->p_userret = lwp_exit_hook;
478 	p->p_userret_arg = NULL;
479 
480 	/*
481 	 * Make SA-cached LWPs normal process runnable LWPs so that
482 	 * they'll also self-destruct.
483 	 */
484 	if (p->p_sa && p->p_sa->sa_ncached > 0) {
485 		DPRINTF(("exit_lwps: Making cached LWPs of %d runnable: ",
486 		    p->p_pid));
487 		SCHED_LOCK(s);
488 		while ((l2 = sa_getcachelwp(p)) != 0) {
489 			l2->l_priority = l2->l_usrpri;
490 			setrunnable(l2);
491 			DPRINTF(("%d ", l2->l_lid));
492 		}
493 		DPRINTF(("\n"));
494 		SCHED_UNLOCK(s);
495 	}
496 
497 	/*
498 	 * Interrupt LWPs in interruptable sleep, unsuspend suspended
499 	 * LWPs, make detached LWPs undetached (so we can wait for
500 	 * them) and then wait for everyone else to finish.
501 	 */
502 	LIST_FOREACH(l2, &p->p_lwps, l_sibling) {
503 		l2->l_flag &= ~(L_DETACHED|L_SA);
504 
505 		if (l2->l_wchan == &l2->l_upcallstack)
506 			wakeup(&l2->l_upcallstack);
507 
508 		if ((l2->l_stat == LSSLEEP && (l2->l_flag & L_SINTR)) ||
509 		    l2->l_stat == LSSUSPENDED || l2->l_stat == LSSTOP) {
510 			SCHED_LOCK(s);
511 			setrunnable(l2);
512 			SCHED_UNLOCK(s);
513 			DPRINTF(("exit_lwps: Made %d.%d runnable\n",
514 			    p->p_pid, l2->l_lid));
515 		}
516 	}
517 
518 
519 	while (p->p_nlwps > 1) {
520 		DPRINTF(("exit_lwps: waiting for %d LWPs (%d runnable, %d zombies)\n",
521 		    p->p_nlwps, p->p_nrlwps, p->p_nzlwps));
522 		error = lwp_wait1(l, 0, &waited, LWPWAIT_EXITCONTROL);
523 		if (error)
524 			panic("exit_lwps: lwp_wait1 failed with error %d\n",
525 			    error);
526 		DPRINTF(("exit_lwps: Got LWP %d from lwp_wait1()\n", waited));
527 	}
528 
529 	p->p_userret = NULL;
530 }
531 
532 /* Wrapper function for use in p_userret */
533 static void
534 lwp_exit_hook(struct lwp *l, void *arg)
535 {
536 	KERNEL_PROC_LOCK(l);
537 	lwp_exit(l);
538 }
539 
540 /*
541  * We are called from cpu_exit() once it is safe to schedule the
542  * dead process's resources to be freed (i.e., once we've switched to
543  * the idle PCB for the current CPU).
544  *
545  * NOTE: One must be careful with locking in this routine.  It's
546  * called from a critical section in machine-dependent code, so
547  * we should refrain from changing any interrupt state.
548  *
549  * We lock the deadproc list (a spin lock), place the proc on that
550  * list (using the p_dead member), and wake up the reaper.
551  */
552 void
553 exit2(struct lwp *l)
554 {
555 	struct proc *p = l->l_proc;
556 
557 	simple_lock(&deadproc_slock);
558 	SLIST_INSERT_HEAD(&deadprocs, p, p_dead);
559 	simple_unlock(&deadproc_slock);
560 
561 	/* lwp_exit2() will wake up deadproc for us. */
562 	lwp_exit2(l);
563 }
564 
565 /*
566  * Process reaper.  This is run by a kernel thread to free the resources
567  * of a dead process.  Once the resources are free, the process becomes
568  * a zombie, and the parent is allowed to read the undead's status.
569  */
570 void
571 reaper(void *arg)
572 {
573 	struct proc *p;
574 	struct lwp *l;
575 
576 	KERNEL_PROC_UNLOCK(curlwp);
577 
578 	for (;;) {
579 		simple_lock(&deadproc_slock);
580 		p = SLIST_FIRST(&deadprocs);
581 		l = LIST_FIRST(&deadlwp);
582 		if (p == NULL && l == NULL) {
583 			/* No work for us; go to sleep until someone exits. */
584 			(void) ltsleep(&deadprocs, PVM|PNORELOCK,
585 			    "reaper", 0, &deadproc_slock);
586 			continue;
587 		}
588 
589 		if (l != NULL ) {
590 			p = l->l_proc;
591 
592 			/* Remove lwp from the deadlwp list. */
593 			LIST_REMOVE(l, l_list);
594 			simple_unlock(&deadproc_slock);
595 			KERNEL_PROC_LOCK(curlwp);
596 
597 			/*
598 			 * Give machine-dependent code a chance to free any
599 			 * resources it couldn't free while still running on
600 			 * that process's context.  This must be done before
601 			 * uvm_lwp_exit(), in case these resources are in the
602 			 * PCB.
603 			 */
604 			cpu_wait(l);
605 
606 			/*
607 			 * Free the VM resources we're still holding on to.
608 			 */
609 			uvm_lwp_exit(l);
610 
611 			l->l_stat = LSZOMB;
612 			if (l->l_flag & L_DETACHED) {
613 				/* Nobody waits for detached LWPs. */
614 				LIST_REMOVE(l, l_sibling);
615 				p->p_nlwps--;
616 				pool_put(&lwp_pool, l);
617 			} else {
618 				p->p_nzlwps++;
619 				wakeup(&p->p_nlwps);
620 			}
621 			/* XXXNJW where should this be with respect to
622 			 * the wakeup() above? */
623 			KERNEL_PROC_UNLOCK(curlwp);
624 		} else {
625 			/* Remove proc from the deadproc list. */
626 			SLIST_REMOVE_HEAD(&deadprocs, p_dead);
627 			simple_unlock(&deadproc_slock);
628 			KERNEL_PROC_LOCK(curlwp);
629 
630 			/*
631 			 * Free the VM resources we're still holding on to.
632 			 * We must do this from a valid thread because doing
633 			 * so may block.
634 			 */
635 			uvm_proc_exit(p);
636 
637 			/* Process is now a true zombie. */
638 			p->p_stat = SZOMB;
639 
640 			/* Wake up the parent so it can get exit status. */
641 			if ((p->p_flag & P_FSTRACE) == 0 && p->p_exitsig != 0)
642 				exit_psignal(p, p->p_pptr);
643 			KERNEL_PROC_UNLOCK(curlwp);
644 			wakeup(p->p_pptr);
645 		}
646 	}
647 }
648 
649 int
650 sys_wait4(struct lwp *l, void *v, register_t *retval)
651 {
652 	struct sys_wait4_args /* {
653 		syscallarg(int)			pid;
654 		syscallarg(int *)		status;
655 		syscallarg(int)			options;
656 		syscallarg(struct rusage *)	rusage;
657 	} */ *uap = v;
658 	struct proc	*child, *parent;
659 	int		status, error;
660 
661 	parent = l->l_proc;
662 
663 	if (SCARG(uap, pid) == 0)
664 		SCARG(uap, pid) = -parent->p_pgid;
665 	if (SCARG(uap, options) & ~(WUNTRACED|WNOHANG|WALTSIG|WALLSIG))
666 		return (EINVAL);
667 
668 	error = find_stopped_child(parent, SCARG(uap,pid), SCARG(uap,options),
669 				&child);
670 	if (error != 0)
671 		return error;
672 	if (child == NULL) {
673 		*retval = 0;
674 		return 0;
675 	}
676 
677 	retval[0] = child->p_pid;
678 
679 	if (child->p_stat == SZOMB) {
680 		if (SCARG(uap, status)) {
681 			status = child->p_xstat;	/* convert to int */
682 			error = copyout(&status, SCARG(uap, status),
683 					sizeof(status));
684 			if (error)
685 				return (error);
686 		}
687 		if (SCARG(uap, rusage)) {
688 			error = copyout(child->p_ru, SCARG(uap, rusage),
689 				    sizeof(struct rusage));
690 			if (error)
691 				return (error);
692 		}
693 
694 		proc_free(child);
695 		return 0;
696 	}
697 
698 	/* child state must be SSTOP */
699 	if (SCARG(uap, status)) {
700 		status = W_STOPCODE(child->p_xstat);
701 		return copyout(&status, SCARG(uap, status), sizeof (status));
702 	}
703 	return 0;
704 }
705 
706 /*
707  * Scan list of child processes for a child process that has stopped or
708  * exited.  Used by sys_wait4 and 'compat' equivalents.
709  */
710 int
711 find_stopped_child(struct proc *parent, pid_t pid, int options,
712 	struct proc **child_p)
713 {
714 	struct proc *child;
715 	int c_found, error;
716 
717 	for (;;) {
718 		c_found = 0;
719 		LIST_FOREACH(child, &parent->p_children, p_sibling) {
720 			if (pid != WAIT_ANY &&
721 			    child->p_pid != pid &&
722 			    child->p_pgid != -pid)
723 				continue;
724 			/*
725 			 * Wait for processes with p_exitsig != SIGCHLD
726 			 * processes only if WALTSIG is set; wait for
727 			 * processes with p_exitsig == SIGCHLD only
728 			 * if WALTSIG is clear.
729 			 */
730 			if (((options & WALLSIG) == 0) &&
731 			    (options & WALTSIG ? child->p_exitsig == SIGCHLD
732 						: P_EXITSIG(child) != SIGCHLD))
733 				continue;
734 
735 			c_found = 1;
736 			if (child->p_stat == SZOMB &&
737 			    (options & WNOZOMBIE) == 0) {
738 				*child_p = child;
739 				return 0;
740 			}
741 
742 			if (child->p_stat == SSTOP &&
743 			    (child->p_flag & P_WAITED) == 0 &&
744 			    (child->p_flag & P_TRACED || options & WUNTRACED)) {
745 				if ((options & WNOWAIT) == 0)
746 					child->p_flag |= P_WAITED;
747 				*child_p = child;
748 				return 0;
749 			}
750 		}
751 		if (c_found == 0)
752 			return ECHILD;
753 		if (options & WNOHANG) {
754 			*child_p = NULL;
755 			return 0;
756 		}
757 		error = tsleep(parent, PWAIT | PCATCH, "wait", 0);
758 		if (error != 0)
759 			return error;
760 	}
761 }
762 
763 /*
764  * Free a process after parent has taken all the state info.
765  */
766 void
767 proc_free(struct proc *p)
768 {
769 	struct proc *parent = p->p_pptr;
770 	int s;
771 
772 	/*
773 	 * If we got the child via ptrace(2) or procfs, and
774 	 * the parent is different (meaning the process was
775 	 * attached, rather than run as a child), then we need
776 	 * to give it back to the old parent, and send the
777 	 * parent the exit signal.  The rest of the cleanup
778 	 * will be done when the old parent waits on the child.
779 	 */
780 	if ((p->p_flag & P_TRACED) && p->p_opptr != parent){
781 		parent = p->p_opptr;
782 		if (parent == NULL)
783 			parent = initproc;
784 		proc_reparent(p, parent);
785 		p->p_opptr = NULL;
786 		p->p_flag &= ~(P_TRACED|P_WAITED|P_FSTRACE);
787 		if (p->p_exitsig != 0)
788 			exit_psignal(p, parent);
789 		wakeup(parent);
790 		return;
791 	}
792 
793 	scheduler_wait_hook(parent, p);
794 	p->p_xstat = 0;
795 
796 	ruadd(&parent->p_stats->p_cru, p->p_ru);
797 
798 	/*
799 	 * At this point we are going to start freeing the final resources.
800 	 * If anyone tries to access the proc structure after here they
801 	 * will get a shock - bits are missing.
802 	 * Attempt to make it hard!
803 	 */
804 
805 	p->p_stat = SIDL;		/* not even a zombie any more */
806 
807 	pool_put(&rusage_pool, p->p_ru);
808 
809 	/*
810 	 * Finally finished with old proc entry.
811 	 * Unlink it from its process group and free it.
812 	 */
813 	leavepgrp(p);
814 
815 	s = proclist_lock_write();
816 	LIST_REMOVE(p, p_list);	/* off zombproc */
817 	LIST_REMOVE(p, p_sibling);
818 	proclist_unlock_write(s);
819 
820 	/*
821 	 * Decrement the count of procs running with this uid.
822 	 */
823 	(void)chgproccnt(p->p_cred->p_ruid, -1);
824 
825 	/*
826 	 * Free up credentials.
827 	 */
828 	if (--p->p_cred->p_refcnt == 0) {
829 		crfree(p->p_cred->pc_ucred);
830 		pool_put(&pcred_pool, p->p_cred);
831 	}
832 
833 	/*
834 	 * Release reference to text vnode
835 	 */
836 	if (p->p_textvp)
837 		vrele(p->p_textvp);
838 
839 	/*
840 	 * Release any SA state
841 	 */
842 	if (p->p_sa) {
843 		free(p->p_sa->sa_stacks, M_SA);
844 		pool_put(&sadata_pool, p->p_sa);
845 	}
846 
847 	/* Free proc structure and let pid be reallocated */
848 	proc_free_mem(p);
849 }
850 
851 /*
852  * make process 'parent' the new parent of process 'child'.
853  */
854 void
855 proc_reparent(struct proc *child, struct proc *parent)
856 {
857 
858 	if (child->p_pptr == parent)
859 		return;
860 
861 	if (parent == initproc)
862 		child->p_exitsig = SIGCHLD;
863 
864 	LIST_REMOVE(child, p_sibling);
865 	LIST_INSERT_HEAD(&parent->p_children, child, p_sibling);
866 	child->p_pptr = parent;
867 }
868