xref: /netbsd-src/sys/kern/kern_exit.c (revision 1ca5c1b28139779176bd5c13ad7c5f25c0bcd5f8)
1 /*	$NetBSD: kern_exit.c,v 1.92 2001/11/12 15:25:08 lukem 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 1.92 2001/11/12 15:25:08 lukem Exp $");
82 
83 #include "opt_ktrace.h"
84 #include "opt_sysv.h"
85 
86 #include <sys/param.h>
87 #include <sys/systm.h>
88 #include <sys/map.h>
89 #include <sys/ioctl.h>
90 #include <sys/proc.h>
91 #include <sys/tty.h>
92 #include <sys/time.h>
93 #include <sys/resource.h>
94 #include <sys/kernel.h>
95 #include <sys/ktrace.h>
96 #include <sys/proc.h>
97 #include <sys/buf.h>
98 #include <sys/wait.h>
99 #include <sys/file.h>
100 #include <sys/vnode.h>
101 #include <sys/syslog.h>
102 #include <sys/malloc.h>
103 #include <sys/pool.h>
104 #include <sys/resourcevar.h>
105 #include <sys/ptrace.h>
106 #include <sys/acct.h>
107 #include <sys/filedesc.h>
108 #include <sys/signalvar.h>
109 #include <sys/sched.h>
110 #ifdef SYSVSHM
111 #include <sys/shm.h>
112 #endif
113 #ifdef SYSVSEM
114 #include <sys/sem.h>
115 #endif
116 
117 #include <sys/mount.h>
118 #include <sys/syscallargs.h>
119 
120 #include <machine/cpu.h>
121 
122 #include <uvm/uvm_extern.h>
123 
124 /*
125  * exit --
126  *	Death of process.
127  */
128 int
129 sys_exit(struct proc *p, void *v, register_t *retval)
130 {
131 	struct sys_exit_args /* {
132 		syscallarg(int)	rval;
133 	} */ *uap = v;
134 
135 	exit1(p, W_EXITCODE(SCARG(uap, rval), 0));
136 	/* NOTREACHED */
137 	return (0);
138 }
139 
140 /*
141  * Exit: deallocate address space and other resources, change proc state
142  * to zombie, and unlink proc from allproc and parent's lists.  Save exit
143  * status and rusage for wait().  Check for child processes and orphan them.
144  */
145 void
146 exit1(struct proc *p, int rv)
147 {
148 	struct proc	*q, *nq;
149 	int		s;
150 
151 	if (__predict_false(p == initproc))
152 		panic("init died (signal %d, exit %d)",
153 		    WTERMSIG(rv), WEXITSTATUS(rv));
154 
155 #ifdef PGINPROF
156 	vmsizmon();
157 #endif
158 	if (p->p_flag & P_PROFIL)
159 		stopprofclock(p);
160 	p->p_ru = pool_get(&rusage_pool, PR_WAITOK);
161 	/*
162 	 * If parent is waiting for us to exit or exec, P_PPWAIT is set; we
163 	 * wake up the parent early to avoid deadlock.
164 	 */
165 	p->p_flag |= P_WEXIT;
166 	if (p->p_flag & P_PPWAIT) {
167 		p->p_flag &= ~P_PPWAIT;
168 		wakeup((caddr_t)p->p_pptr);
169 	}
170 	sigfillset(&p->p_sigctx.ps_sigignore);
171 	sigemptyset(&p->p_sigctx.ps_siglist);
172 	p->p_sigctx.ps_sigcheck = 0;
173 	callout_stop(&p->p_realit_ch);
174 
175 	/*
176 	 * Close open files and release open-file table.
177 	 * This may block!
178 	 */
179 	fdfree(p);
180 	cwdfree(p);
181 
182 #ifdef SYSVSEM
183 	semexit(p);
184 #endif
185 	if (SESS_LEADER(p)) {
186 		struct session *sp = p->p_session;
187 
188 		if (sp->s_ttyvp) {
189 			/*
190 			 * Controlling process.
191 			 * Signal foreground pgrp,
192 			 * drain controlling terminal
193 			 * and revoke access to controlling terminal.
194 			 */
195 			if (sp->s_ttyp->t_session == sp) {
196 				if (sp->s_ttyp->t_pgrp)
197 					pgsignal(sp->s_ttyp->t_pgrp, SIGHUP, 1);
198 				(void) ttywait(sp->s_ttyp);
199 				/*
200 				 * The tty could have been revoked
201 				 * if we blocked.
202 				 */
203 				if (sp->s_ttyvp)
204 					VOP_REVOKE(sp->s_ttyvp, REVOKEALL);
205 			}
206 			if (sp->s_ttyvp)
207 				vrele(sp->s_ttyvp);
208 			sp->s_ttyvp = NULL;
209 			/*
210 			 * s_ttyp is not zero'd; we use this to indicate
211 			 * that the session once had a controlling terminal.
212 			 * (for logging and informational purposes)
213 			 */
214 		}
215 		sp->s_leader = NULL;
216 	}
217 	fixjobc(p, p->p_pgrp, 0);
218 	(void)acct_process(p);
219 #ifdef KTRACE
220 	/*
221 	 * release trace file
222 	 */
223 	ktrderef(p);
224 #endif
225 	/*
226 	 * NOTE: WE ARE NO LONGER ALLOWED TO SLEEP!
227 	 */
228 	p->p_stat = SDEAD;
229 
230 	/*
231 	 * Remove proc from pidhash chain so looking it up won't
232 	 * work.  Move it from allproc to zombproc, but do not yet
233 	 * wake up the reaper.  We will put the proc on the
234 	 * deadproc list later (using the p_hash member), and
235 	 * wake up the reaper when we do.
236 	 */
237 	s = proclist_lock_write();
238 	LIST_REMOVE(p, p_hash);
239 	LIST_REMOVE(p, p_list);
240 	LIST_INSERT_HEAD(&zombproc, p, p_list);
241 	proclist_unlock_write(s);
242 
243 	/*
244 	 * Give orphaned children to init(8).
245 	 */
246 	q = p->p_children.lh_first;
247 	if (q)		/* only need this if any child is S_ZOMB */
248 		wakeup((caddr_t)initproc);
249 	for (; q != 0; q = nq) {
250 		nq = q->p_sibling.le_next;
251 		proc_reparent(q, initproc);
252 		/*
253 		 * Traced processes are killed
254 		 * since their existence means someone is screwing up.
255 		 */
256 		if (q->p_flag & P_TRACED) {
257 			q->p_flag &= ~(P_TRACED|P_WAITED|P_FSTRACE);
258 			psignal(q, SIGKILL);
259 		}
260 	}
261 
262 	/*
263 	 * Save exit status and final rusage info, adding in child rusage
264 	 * info and self times.
265 	 */
266 	p->p_xstat = rv;
267 	*p->p_ru = p->p_stats->p_ru;
268 	calcru(p, &p->p_ru->ru_utime, &p->p_ru->ru_stime, NULL);
269 	ruadd(p->p_ru, &p->p_stats->p_cru);
270 
271 	/*
272 	 * Notify parent that we're gone.  If parent has the P_NOCLDWAIT
273 	 * flag set, notify init instead (and hope it will handle
274 	 * this situation).
275 	 */
276 	if (p->p_pptr->p_flag & P_NOCLDWAIT) {
277 		struct proc *pp = p->p_pptr;
278 		proc_reparent(p, initproc);
279 		/*
280 		 * If this was the last child of our parent, notify
281 		 * parent, so in case he was wait(2)ing, he will
282 		 * continue.
283 		 */
284 		if (pp->p_children.lh_first == NULL)
285 			wakeup((caddr_t)pp);
286 	}
287 
288 	/*
289 	 * Release the process's signal state.
290 	 */
291 	sigactsfree(p);
292 
293 	/*
294 	 * Clear curproc after we've done all operations
295 	 * that could block, and before tearing down the rest
296 	 * of the process state that might be used from clock, etc.
297 	 * Also, can't clear curproc while we're still runnable,
298 	 * as we're not on a run queue (we are current, just not
299 	 * a proper proc any longer!).
300 	 *
301 	 * Other substructures are freed from wait().
302 	 */
303 	curproc = NULL;
304 	limfree(p->p_limit);
305 	p->p_limit = NULL;
306 
307 	/*
308 	 * If emulation has process exit hook, call it now.
309 	 */
310 	if (p->p_emul->e_proc_exit)
311 		(*p->p_emul->e_proc_exit)(p);
312 
313 	/* This process no longer needs to hold the kernel lock. */
314 	KERNEL_PROC_UNLOCK(p);
315 
316 	/*
317 	 * Finally, call machine-dependent code to switch to a new
318 	 * context (possibly the idle context).  Once we are no longer
319 	 * using the dead process's vmspace and stack, exit2() will be
320 	 * called to schedule those resources to be released by the
321 	 * reaper thread.
322 	 *
323 	 * Note that cpu_exit() will end with a call equivalent to
324 	 * cpu_switch(), finishing our execution (pun intended).
325 	 */
326 	cpu_exit(p);
327 }
328 
329 /*
330  * We are called from cpu_exit() once it is safe to schedule the
331  * dead process's resources to be freed (i.e., once we've switched to
332  * the idle PCB for the current CPU).
333  *
334  * NOTE: One must be careful with locking in this routine.  It's
335  * called from a critical section in machine-dependent code, so
336  * we should refrain from changing any interrupt state.
337  *
338  * We lock the deadproc list (a spin lock), place the proc on that
339  * list (using the p_hash member), and wake up the reaper.
340  */
341 void
342 exit2(struct proc *p)
343 {
344 
345 	simple_lock(&deadproc_slock);
346 	LIST_INSERT_HEAD(&deadproc, p, p_hash);
347 	simple_unlock(&deadproc_slock);
348 
349 	wakeup(&deadproc);
350 }
351 
352 /*
353  * Process reaper.  This is run by a kernel thread to free the resources
354  * of a dead process.  Once the resources are free, the process becomes
355  * a zombie, and the parent is allowed to read the undead's status.
356  */
357 void
358 reaper(void *arg)
359 {
360 	struct proc *p;
361 
362 	KERNEL_PROC_UNLOCK(curproc);
363 
364 	for (;;) {
365 		simple_lock(&deadproc_slock);
366 		p = LIST_FIRST(&deadproc);
367 		if (p == NULL) {
368 			/* No work for us; go to sleep until someone exits. */
369 			(void) ltsleep(&deadproc, PVM|PNORELOCK,
370 			    "reaper", 0, &deadproc_slock);
371 			continue;
372 		}
373 
374 		/* Remove us from the deadproc list. */
375 		LIST_REMOVE(p, p_hash);
376 		simple_unlock(&deadproc_slock);
377 		KERNEL_PROC_LOCK(curproc);
378 
379 		/*
380 		 * Give machine-dependent code a chance to free any
381 		 * resources it couldn't free while still running on
382 		 * that process's context.  This must be done before
383 		 * uvm_exit(), in case these resources are in the PCB.
384 		 */
385 		cpu_wait(p);
386 
387 		/*
388 		 * Free the VM resources we're still holding on to.
389 		 * We must do this from a valid thread because doing
390 		 * so may block.
391 		 */
392 		uvm_exit(p);
393 
394 		/* Process is now a true zombie. */
395 		p->p_stat = SZOMB;
396 
397 		/* Wake up the parent so it can get exit status. */
398 		if ((p->p_flag & P_FSTRACE) == 0 && p->p_exitsig != 0)
399 			psignal(p->p_pptr, P_EXITSIG(p));
400 		KERNEL_PROC_UNLOCK(curproc);
401 		wakeup((caddr_t)p->p_pptr);
402 	}
403 }
404 
405 int
406 sys_wait4(struct proc *q, void *v, register_t *retval)
407 {
408 	struct sys_wait4_args /* {
409 		syscallarg(int)			pid;
410 		syscallarg(int *)		status;
411 		syscallarg(int)			options;
412 		syscallarg(struct rusage *)	rusage;
413 	} */ *uap = v;
414 	struct proc	*p, *t;
415 	int		nfound, status, error, s;
416 
417 	if (SCARG(uap, pid) == 0)
418 		SCARG(uap, pid) = -q->p_pgid;
419 	if (SCARG(uap, options) &~ (WUNTRACED|WNOHANG|WALTSIG))
420 		return (EINVAL);
421 
422  loop:
423 	nfound = 0;
424 	for (p = q->p_children.lh_first; p != 0; p = p->p_sibling.le_next) {
425 		if (SCARG(uap, pid) != WAIT_ANY &&
426 		    p->p_pid != SCARG(uap, pid) &&
427 		    p->p_pgid != -SCARG(uap, pid))
428 			continue;
429 		/*
430 		 * Wait for processes with p_exitsig != SIGCHLD processes only
431 		 * if WALTSIG is set; wait for processes with p_exitsig ==
432 		 * SIGCHLD only if WALTSIG is clear.
433 		 */
434 		if (((SCARG(uap, options) & WALLSIG) == 0) &&
435 		    ((SCARG(uap, options) & WALTSIG) ?
436 		     (p->p_exitsig == SIGCHLD) : (P_EXITSIG(p) != SIGCHLD)))
437 			continue;
438 
439 		nfound++;
440 		if (p->p_stat == SZOMB) {
441 			retval[0] = p->p_pid;
442 
443 			if (SCARG(uap, status)) {
444 				status = p->p_xstat;	/* convert to int */
445 				error = copyout((caddr_t)&status,
446 						(caddr_t)SCARG(uap, status),
447 						sizeof(status));
448 				if (error)
449 					return (error);
450 			}
451 			if (SCARG(uap, rusage) &&
452 			    (error = copyout((caddr_t)p->p_ru,
453 			    (caddr_t)SCARG(uap, rusage),
454 			    sizeof(struct rusage))))
455 				return (error);
456 			/*
457 			 * If we got the child via ptrace(2) or procfs, and
458 			 * the parent is different (meaning the process was
459 			 * attached, rather than run as a child), then we need
460 			 * to give it back to the old parent, and send the
461 			 * parent the exit signal.  The rest of the cleanup
462 			 * will be done when the old parent waits on the child.
463 			 */
464 			if ((p->p_flag & P_TRACED) &&
465 			    p->p_oppid != p->p_pptr->p_pid) {
466 				t = pfind(p->p_oppid);
467 				proc_reparent(p, t ? t : initproc);
468 				p->p_oppid = 0;
469 				p->p_flag &= ~(P_TRACED|P_WAITED|P_FSTRACE);
470 				if (p->p_exitsig != 0)
471 					psignal(p->p_pptr, P_EXITSIG(p));
472 				wakeup((caddr_t)p->p_pptr);
473 				return (0);
474 			}
475 			scheduler_wait_hook(q, p);
476 			p->p_xstat = 0;
477 			ruadd(&q->p_stats->p_cru, p->p_ru);
478 			pool_put(&rusage_pool, p->p_ru);
479 
480 			/*
481 			 * Finally finished with old proc entry.
482 			 * Unlink it from its process group and free it.
483 			 */
484 			leavepgrp(p);
485 
486 			s = proclist_lock_write();
487 			LIST_REMOVE(p, p_list);	/* off zombproc */
488 			proclist_unlock_write(s);
489 
490 			LIST_REMOVE(p, p_sibling);
491 
492 			/*
493 			 * Decrement the count of procs running with this uid.
494 			 */
495 			(void)chgproccnt(p->p_cred->p_ruid, -1);
496 
497 			/*
498 			 * Free up credentials.
499 			 */
500 			if (--p->p_cred->p_refcnt == 0) {
501 				crfree(p->p_cred->pc_ucred);
502 				pool_put(&pcred_pool, p->p_cred);
503 			}
504 
505 			/*
506 			 * Release reference to text vnode
507 			 */
508 			if (p->p_textvp)
509 				vrele(p->p_textvp);
510 
511 			pool_put(&proc_pool, p);
512 			nprocs--;
513 			return (0);
514 		}
515 		if (p->p_stat == SSTOP && (p->p_flag & P_WAITED) == 0 &&
516 		    (p->p_flag & P_TRACED || SCARG(uap, options) & WUNTRACED)) {
517 			p->p_flag |= P_WAITED;
518 			retval[0] = p->p_pid;
519 
520 			if (SCARG(uap, status)) {
521 				status = W_STOPCODE(p->p_xstat);
522 				error = copyout((caddr_t)&status,
523 				    (caddr_t)SCARG(uap, status),
524 				    sizeof(status));
525 			} else
526 				error = 0;
527 			return (error);
528 		}
529 	}
530 	if (nfound == 0)
531 		return (ECHILD);
532 	if (SCARG(uap, options) & WNOHANG) {
533 		retval[0] = 0;
534 		return (0);
535 	}
536 	if ((error = tsleep((caddr_t)q, PWAIT | PCATCH, "wait", 0)) != 0)
537 		return (error);
538 	goto loop;
539 }
540 
541 /*
542  * make process 'parent' the new parent of process 'child'.
543  */
544 void
545 proc_reparent(struct proc *child, struct proc *parent)
546 {
547 
548 	if (child->p_pptr == parent)
549 		return;
550 
551 	if (parent == initproc)
552 		child->p_exitsig = SIGCHLD;
553 
554 	LIST_REMOVE(child, p_sibling);
555 	LIST_INSERT_HEAD(&parent->p_children, child, p_sibling);
556 	child->p_pptr = parent;
557 }
558