xref: /netbsd-src/sys/kern/kern_exit.c (revision 08c81a9c2dc8c7300e893321eb65c0925d60871c)
1 /*	$NetBSD: kern_exit.c,v 1.101 2002/09/04 01:32:32 matt 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. All advertising materials mentioning features or use of this software
58  *    must display the following acknowledgement:
59  *	This product includes software developed by the University of
60  *	California, Berkeley and its contributors.
61  * 4. Neither the name of the University nor the names of its contributors
62  *    may be used to endorse or promote products derived from this software
63  *    without specific prior written permission.
64  *
65  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
66  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
67  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
68  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
69  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
70  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
71  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
72  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
73  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
74  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
75  * SUCH DAMAGE.
76  *
77  *	@(#)kern_exit.c	8.10 (Berkeley) 2/23/95
78  */
79 
80 #include <sys/cdefs.h>
81 __KERNEL_RCSID(0, "$NetBSD: kern_exit.c,v 1.101 2002/09/04 01:32:32 matt Exp $");
82 
83 #include "opt_ktrace.h"
84 #include "opt_perfctrs.h"
85 #include "opt_systrace.h"
86 #include "opt_sysv.h"
87 
88 #include <sys/param.h>
89 #include <sys/systm.h>
90 #include <sys/map.h>
91 #include <sys/ioctl.h>
92 #include <sys/proc.h>
93 #include <sys/tty.h>
94 #include <sys/time.h>
95 #include <sys/resource.h>
96 #include <sys/kernel.h>
97 #include <sys/ktrace.h>
98 #include <sys/proc.h>
99 #include <sys/buf.h>
100 #include <sys/wait.h>
101 #include <sys/file.h>
102 #include <sys/vnode.h>
103 #include <sys/syslog.h>
104 #include <sys/malloc.h>
105 #include <sys/pool.h>
106 #include <sys/resourcevar.h>
107 #if defined(PERFCTRS)
108 #include <sys/pmc.h>
109 #endif
110 #include <sys/ptrace.h>
111 #include <sys/acct.h>
112 #include <sys/filedesc.h>
113 #include <sys/ras.h>
114 #include <sys/signalvar.h>
115 #include <sys/sched.h>
116 #include <sys/mount.h>
117 #include <sys/syscallargs.h>
118 #include <sys/systrace.h>
119 
120 #include <machine/cpu.h>
121 
122 #include <uvm/uvm_extern.h>
123 
124 
125 /*
126  * exit --
127  *	Death of process.
128  */
129 int
130 sys_exit(struct proc *p, void *v, register_t *retval)
131 {
132 	struct sys_exit_args /* {
133 		syscallarg(int)	rval;
134 	} */ *uap = v;
135 
136 	exit1(p, W_EXITCODE(SCARG(uap, rval), 0));
137 	/* NOTREACHED */
138 	return (0);
139 }
140 
141 /*
142  * Exit: deallocate address space and other resources, change proc state
143  * to zombie, and unlink proc from allproc and parent's lists.  Save exit
144  * status and rusage for wait().  Check for child processes and orphan them.
145  */
146 void
147 exit1(struct proc *p, int rv)
148 {
149 	struct proc	*q, *nq;
150 	int		s;
151 
152 	if (__predict_false(p == initproc))
153 		panic("init died (signal %d, exit %d)",
154 		    WTERMSIG(rv), WEXITSTATUS(rv));
155 
156 #ifdef PGINPROF
157 	vmsizmon();
158 #endif
159 	if (p->p_flag & P_PROFIL)
160 		stopprofclock(p);
161 	p->p_ru = pool_get(&rusage_pool, PR_WAITOK);
162 	/*
163 	 * If parent is waiting for us to exit or exec, P_PPWAIT is set; we
164 	 * wake up the parent early to avoid deadlock.
165 	 */
166 	p->p_flag |= P_WEXIT;
167 	if (p->p_flag & P_PPWAIT) {
168 		p->p_flag &= ~P_PPWAIT;
169 		wakeup((caddr_t)p->p_pptr);
170 	}
171 	sigfillset(&p->p_sigctx.ps_sigignore);
172 	sigemptyset(&p->p_sigctx.ps_siglist);
173 	p->p_sigctx.ps_sigcheck = 0;
174 	callout_stop(&p->p_realit_ch);
175 
176 #if defined(__HAVE_RAS)
177 	ras_purgeall(p);
178 #endif
179 
180 	/*
181 	 * Close open files and release open-file table.
182 	 * This may block!
183 	 */
184 	fdfree(p);
185 	cwdfree(p);
186 
187 	doexithooks(p);
188 
189 	if (SESS_LEADER(p)) {
190 		struct session *sp = p->p_session;
191 
192 		if (sp->s_ttyvp) {
193 			/*
194 			 * Controlling process.
195 			 * Signal foreground pgrp,
196 			 * drain controlling terminal
197 			 * and revoke access to controlling terminal.
198 			 */
199 			if (sp->s_ttyp->t_session == sp) {
200 				if (sp->s_ttyp->t_pgrp)
201 					pgsignal(sp->s_ttyp->t_pgrp, SIGHUP, 1);
202 				(void) ttywait(sp->s_ttyp);
203 				/*
204 				 * The tty could have been revoked
205 				 * if we blocked.
206 				 */
207 				if (sp->s_ttyvp)
208 					VOP_REVOKE(sp->s_ttyvp, REVOKEALL);
209 			}
210 			if (sp->s_ttyvp)
211 				vrele(sp->s_ttyvp);
212 			sp->s_ttyvp = NULL;
213 			/*
214 			 * s_ttyp is not zero'd; we use this to indicate
215 			 * that the session once had a controlling terminal.
216 			 * (for logging and informational purposes)
217 			 */
218 		}
219 		sp->s_leader = NULL;
220 	}
221 	fixjobc(p, p->p_pgrp, 0);
222 	(void)acct_process(p);
223 #ifdef KTRACE
224 	/*
225 	 * release trace file
226 	 */
227 	ktrderef(p);
228 #endif
229 #ifdef SYSTRACE
230 	systrace_sys_exit(p);
231 #endif
232 	/*
233 	 * If emulation has process exit hook, call it now.
234 	 */
235 	if (p->p_emul->e_proc_exit)
236 		(*p->p_emul->e_proc_exit)(p);
237 
238 	/*
239 	 * NOTE: WE ARE NO LONGER ALLOWED TO SLEEP!
240 	 */
241 	p->p_stat = SDEAD;
242 
243 	/*
244 	 * Remove proc from pidhash chain so looking it up won't
245 	 * work.  Move it from allproc to zombproc, but do not yet
246 	 * wake up the reaper.  We will put the proc on the
247 	 * deadproc list later (using the p_hash member), and
248 	 * wake up the reaper when we do.
249 	 */
250 	s = proclist_lock_write();
251 	LIST_REMOVE(p, p_hash);
252 	LIST_REMOVE(p, p_list);
253 	LIST_INSERT_HEAD(&zombproc, p, p_list);
254 	proclist_unlock_write(s);
255 
256 	/*
257 	 * Give orphaned children to init(8).
258 	 */
259 	q = LIST_FIRST(&p->p_children);
260 	if (q)		/* only need this if any child is S_ZOMB */
261 		wakeup((caddr_t)initproc);
262 	for (; q != 0; q = nq) {
263 		nq = LIST_NEXT(q, p_sibling);
264 		proc_reparent(q, initproc);
265 		/*
266 		 * Traced processes are killed
267 		 * since their existence means someone is screwing up.
268 		 */
269 		if (q->p_flag & P_TRACED) {
270 			q->p_flag &= ~(P_TRACED|P_WAITED|P_FSTRACE);
271 			psignal(q, SIGKILL);
272 		}
273 	}
274 
275 	/*
276 	 * Reset p_opptr pointer of all former children which got
277 	 * traced by another process and were reparented. We reset
278 	 * it to NULL here; the trace detach code then reparents
279 	 * the child to initproc. We only check allproc list, since
280 	 * eventual former children on zombproc list won't reference
281 	 * p_opptr anymore.
282 	 */
283 	if (p->p_flag & P_CHTRACED) {
284 		struct proc *t;
285 
286 		proclist_lock_read();
287 
288 		LIST_FOREACH(t, &allproc, p_list) {
289 			if (t->p_opptr == p)
290 				t->p_opptr = NULL;
291 		}
292 
293 		proclist_unlock_read();
294 	}
295 
296 	/*
297 	 * Save exit status and final rusage info, adding in child rusage
298 	 * info and self times.
299 	 */
300 	p->p_xstat = rv;
301 	*p->p_ru = p->p_stats->p_ru;
302 	calcru(p, &p->p_ru->ru_utime, &p->p_ru->ru_stime, NULL);
303 	ruadd(p->p_ru, &p->p_stats->p_cru);
304 
305 #if PERFCTRS
306 	/*
307 	 * Save final PMC information in parent process & clean up.
308 	 */
309 	if (PMC_ENABLED(p)) {
310 		pmc_save_context(p);
311 		pmc_accumulate(p->p_pptr, p);
312 		pmc_process_exit(p);
313 	}
314 #endif
315 
316 	/*
317 	 * Notify parent that we're gone.  If parent has the P_NOCLDWAIT
318 	 * flag set, notify init instead (and hope it will handle
319 	 * this situation).
320 	 */
321 	if (p->p_pptr->p_flag & P_NOCLDWAIT) {
322 		struct proc *pp = p->p_pptr;
323 		proc_reparent(p, initproc);
324 		/*
325 		 * If this was the last child of our parent, notify
326 		 * parent, so in case he was wait(2)ing, he will
327 		 * continue.
328 		 */
329 		if (LIST_FIRST(&pp->p_children) == NULL)
330 			wakeup((caddr_t)pp);
331 	}
332 
333 	/*
334 	 * Release the process's signal state.
335 	 */
336 	sigactsfree(p);
337 
338 	/*
339 	 * Clear curproc after we've done all operations
340 	 * that could block, and before tearing down the rest
341 	 * of the process state that might be used from clock, etc.
342 	 * Also, can't clear curproc while we're still runnable,
343 	 * as we're not on a run queue (we are current, just not
344 	 * a proper proc any longer!).
345 	 *
346 	 * Other substructures are freed from wait().
347 	 */
348 	curproc = NULL;
349 	limfree(p->p_limit);
350 	p->p_limit = NULL;
351 
352 	/* This process no longer needs to hold the kernel lock. */
353 	KERNEL_PROC_UNLOCK(p);
354 
355 	/*
356 	 * Finally, call machine-dependent code to switch to a new
357 	 * context (possibly the idle context).  Once we are no longer
358 	 * using the dead process's vmspace and stack, exit2() will be
359 	 * called to schedule those resources to be released by the
360 	 * reaper thread.
361 	 *
362 	 * Note that cpu_exit() will end with a call equivalent to
363 	 * cpu_switch(), finishing our execution (pun intended).
364 	 */
365 	cpu_exit(p);
366 }
367 
368 /*
369  * We are called from cpu_exit() once it is safe to schedule the
370  * dead process's resources to be freed (i.e., once we've switched to
371  * the idle PCB for the current CPU).
372  *
373  * NOTE: One must be careful with locking in this routine.  It's
374  * called from a critical section in machine-dependent code, so
375  * we should refrain from changing any interrupt state.
376  *
377  * We lock the deadproc list (a spin lock), place the proc on that
378  * list (using the p_hash member), and wake up the reaper.
379  */
380 void
381 exit2(struct proc *p)
382 {
383 
384 	simple_lock(&deadproc_slock);
385 	LIST_INSERT_HEAD(&deadproc, p, p_hash);
386 	simple_unlock(&deadproc_slock);
387 
388 	wakeup(&deadproc);
389 }
390 
391 /*
392  * Process reaper.  This is run by a kernel thread to free the resources
393  * of a dead process.  Once the resources are free, the process becomes
394  * a zombie, and the parent is allowed to read the undead's status.
395  */
396 void
397 reaper(void *arg)
398 {
399 	struct proc *p;
400 
401 	KERNEL_PROC_UNLOCK(curproc);
402 
403 	for (;;) {
404 		simple_lock(&deadproc_slock);
405 		p = LIST_FIRST(&deadproc);
406 		if (p == NULL) {
407 			/* No work for us; go to sleep until someone exits. */
408 			(void) ltsleep(&deadproc, PVM|PNORELOCK,
409 			    "reaper", 0, &deadproc_slock);
410 			continue;
411 		}
412 
413 		/* Remove us from the deadproc list. */
414 		LIST_REMOVE(p, p_hash);
415 		simple_unlock(&deadproc_slock);
416 		KERNEL_PROC_LOCK(curproc);
417 
418 		/*
419 		 * Give machine-dependent code a chance to free any
420 		 * resources it couldn't free while still running on
421 		 * that process's context.  This must be done before
422 		 * uvm_exit(), in case these resources are in the PCB.
423 		 */
424 		cpu_wait(p);
425 
426 		/*
427 		 * Free the VM resources we're still holding on to.
428 		 * We must do this from a valid thread because doing
429 		 * so may block.
430 		 */
431 		uvm_exit(p);
432 
433 		/* Process is now a true zombie. */
434 		p->p_stat = SZOMB;
435 
436 		/* Wake up the parent so it can get exit status. */
437 		if ((p->p_flag & P_FSTRACE) == 0 && p->p_exitsig != 0)
438 			psignal(p->p_pptr, P_EXITSIG(p));
439 		KERNEL_PROC_UNLOCK(curproc);
440 		wakeup((caddr_t)p->p_pptr);
441 	}
442 }
443 
444 int
445 sys_wait4(struct proc *q, void *v, register_t *retval)
446 {
447 	struct sys_wait4_args /* {
448 		syscallarg(int)			pid;
449 		syscallarg(int *)		status;
450 		syscallarg(int)			options;
451 		syscallarg(struct rusage *)	rusage;
452 	} */ *uap = v;
453 	struct proc	*p, *t;
454 	int		nfound, status, error, s;
455 
456 	if (SCARG(uap, pid) == 0)
457 		SCARG(uap, pid) = -q->p_pgid;
458 	if (SCARG(uap, options) &~ (WUNTRACED|WNOHANG|WALTSIG))
459 		return (EINVAL);
460 
461  loop:
462 	nfound = 0;
463 	LIST_FOREACH(p, &q->p_children, p_sibling) {
464 		if (SCARG(uap, pid) != WAIT_ANY &&
465 		    p->p_pid != SCARG(uap, pid) &&
466 		    p->p_pgid != -SCARG(uap, pid))
467 			continue;
468 		/*
469 		 * Wait for processes with p_exitsig != SIGCHLD processes only
470 		 * if WALTSIG is set; wait for processes with p_exitsig ==
471 		 * SIGCHLD only if WALTSIG is clear.
472 		 */
473 		if (((SCARG(uap, options) & WALLSIG) == 0) &&
474 		    ((SCARG(uap, options) & WALTSIG) ?
475 		     (p->p_exitsig == SIGCHLD) : (P_EXITSIG(p) != SIGCHLD)))
476 			continue;
477 
478 		nfound++;
479 		if (p->p_stat == SZOMB) {
480 			retval[0] = p->p_pid;
481 
482 			if (SCARG(uap, status)) {
483 				status = p->p_xstat;	/* convert to int */
484 				error = copyout((caddr_t)&status,
485 						(caddr_t)SCARG(uap, status),
486 						sizeof(status));
487 				if (error)
488 					return (error);
489 			}
490 			if (SCARG(uap, rusage) &&
491 			    (error = copyout((caddr_t)p->p_ru,
492 			    (caddr_t)SCARG(uap, rusage),
493 			    sizeof(struct rusage))))
494 				return (error);
495 			/*
496 			 * If we got the child via ptrace(2) or procfs, and
497 			 * the parent is different (meaning the process was
498 			 * attached, rather than run as a child), then we need
499 			 * to give it back to the old parent, and send the
500 			 * parent the exit signal.  The rest of the cleanup
501 			 * will be done when the old parent waits on the child.
502 			 */
503 			if ((p->p_flag & P_TRACED) && p->p_opptr != p->p_pptr){
504 				t = p->p_opptr;
505 				proc_reparent(p, t ? t : initproc);
506 				p->p_opptr = NULL;
507 				p->p_flag &= ~(P_TRACED|P_WAITED|P_FSTRACE);
508 				if (p->p_exitsig != 0)
509 					psignal(p->p_pptr, P_EXITSIG(p));
510 				wakeup((caddr_t)p->p_pptr);
511 				return (0);
512 			}
513 			scheduler_wait_hook(q, p);
514 			p->p_xstat = 0;
515 			ruadd(&q->p_stats->p_cru, p->p_ru);
516 			pool_put(&rusage_pool, p->p_ru);
517 
518 			/*
519 			 * Finally finished with old proc entry.
520 			 * Unlink it from its process group and free it.
521 			 */
522 			leavepgrp(p);
523 
524 			s = proclist_lock_write();
525 			LIST_REMOVE(p, p_list);	/* off zombproc */
526 			proclist_unlock_write(s);
527 
528 			LIST_REMOVE(p, p_sibling);
529 
530 			/*
531 			 * Decrement the count of procs running with this uid.
532 			 */
533 			(void)chgproccnt(p->p_cred->p_ruid, -1);
534 
535 			/*
536 			 * Free up credentials.
537 			 */
538 			if (--p->p_cred->p_refcnt == 0) {
539 				crfree(p->p_cred->pc_ucred);
540 				pool_put(&pcred_pool, p->p_cred);
541 			}
542 
543 			/*
544 			 * Release reference to text vnode
545 			 */
546 			if (p->p_textvp)
547 				vrele(p->p_textvp);
548 
549 			pool_put(&proc_pool, p);
550 			nprocs--;
551 			return (0);
552 		}
553 		if (p->p_stat == SSTOP && (p->p_flag & P_WAITED) == 0 &&
554 		    (p->p_flag & P_TRACED || SCARG(uap, options) & WUNTRACED)) {
555 			p->p_flag |= P_WAITED;
556 			retval[0] = p->p_pid;
557 
558 			if (SCARG(uap, status)) {
559 				status = W_STOPCODE(p->p_xstat);
560 				error = copyout((caddr_t)&status,
561 				    (caddr_t)SCARG(uap, status),
562 				    sizeof(status));
563 			} else
564 				error = 0;
565 			return (error);
566 		}
567 	}
568 	if (nfound == 0)
569 		return (ECHILD);
570 	if (SCARG(uap, options) & WNOHANG) {
571 		retval[0] = 0;
572 		return (0);
573 	}
574 	if ((error = tsleep((caddr_t)q, PWAIT | PCATCH, "wait", 0)) != 0)
575 		return (error);
576 	goto loop;
577 }
578 
579 /*
580  * make process 'parent' the new parent of process 'child'.
581  */
582 void
583 proc_reparent(struct proc *child, struct proc *parent)
584 {
585 
586 	if (child->p_pptr == parent)
587 		return;
588 
589 	if (parent == initproc)
590 		child->p_exitsig = SIGCHLD;
591 
592 	LIST_REMOVE(child, p_sibling);
593 	LIST_INSERT_HEAD(&parent->p_children, child, p_sibling);
594 	child->p_pptr = parent;
595 }
596