xref: /openbsd-src/sys/kern/kern_exit.c (revision 158fb4f96134d3d8dd48810bf22ee46d27fcefa9)
1 /*	$OpenBSD: kern_exit.c,v 1.56 2005/11/28 00:14:28 jsg Exp $	*/
2 /*	$NetBSD: kern_exit.c,v 1.39 1996/04/22 01:38:25 christos Exp $	*/
3 
4 /*
5  * Copyright (c) 1982, 1986, 1989, 1991, 1993
6  *	The Regents of the University of California.  All rights reserved.
7  * (c) UNIX System Laboratories, Inc.
8  * All or some portions of this file are derived from material licensed
9  * to the University of California by American Telephone and Telegraph
10  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
11  * the permission of UNIX System Laboratories, Inc.
12  *
13  * Redistribution and use in source and binary forms, with or without
14  * modification, are permitted provided that the following conditions
15  * are met:
16  * 1. Redistributions of source code must retain the above copyright
17  *    notice, this list of conditions and the following disclaimer.
18  * 2. Redistributions in binary form must reproduce the above copyright
19  *    notice, this list of conditions and the following disclaimer in the
20  *    documentation and/or other materials provided with the distribution.
21  * 3. Neither the name of the University nor the names of its contributors
22  *    may be used to endorse or promote products derived from this software
23  *    without specific prior written permission.
24  *
25  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
26  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
27  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
28  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
29  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
30  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
31  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
32  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
33  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
34  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
35  * SUCH DAMAGE.
36  *
37  *	@(#)kern_exit.c	8.7 (Berkeley) 2/12/94
38  */
39 
40 #include <sys/param.h>
41 #include <sys/systm.h>
42 #include <sys/ioctl.h>
43 #include <sys/proc.h>
44 #include <sys/tty.h>
45 #include <sys/time.h>
46 #include <sys/resource.h>
47 #include <sys/kernel.h>
48 #include <sys/buf.h>
49 #include <sys/wait.h>
50 #include <sys/file.h>
51 #include <sys/vnode.h>
52 #include <sys/syslog.h>
53 #include <sys/malloc.h>
54 #include <sys/resourcevar.h>
55 #include <sys/ptrace.h>
56 #include <sys/acct.h>
57 #include <sys/filedesc.h>
58 #include <sys/signalvar.h>
59 #include <sys/sched.h>
60 #include <sys/ktrace.h>
61 #include <sys/pool.h>
62 #include <sys/mutex.h>
63 #ifdef SYSVSHM
64 #include <sys/shm.h>
65 #endif
66 #ifdef SYSVSEM
67 #include <sys/sem.h>
68 #endif
69 
70 #include "systrace.h"
71 #include <dev/systrace.h>
72 
73 #include <sys/mount.h>
74 #include <sys/syscallargs.h>
75 
76 #include <machine/cpu.h>
77 
78 #include <uvm/uvm_extern.h>
79 
80 /*
81  * exit --
82  *	Death of process.
83  */
84 int
85 sys_exit(struct proc *p, void *v, register_t *retval)
86 {
87 	struct sys_exit_args /* {
88 		syscallarg(int) rval;
89 	} */ *uap = v;
90 
91 	exit1(p, W_EXITCODE(SCARG(uap, rval), 0));
92 	/* NOTREACHED */
93 	return (0);
94 }
95 
96 /*
97  * Exit: deallocate address space and other resources, change proc state
98  * to zombie, and unlink proc from allproc and parent's lists.  Save exit
99  * status and rusage for wait().  Check for child processes and orphan them.
100  */
101 void
102 exit1(struct proc *p, int rv)
103 {
104 	struct proc *q, *nq;
105 
106 	if (p->p_pid == 1)
107 		panic("init died (signal %d, exit %d)",
108 		    WTERMSIG(rv), WEXITSTATUS(rv));
109 
110 	if (p->p_flag & P_PROFIL)
111 		stopprofclock(p);
112 	p->p_ru = pool_get(&rusage_pool, PR_WAITOK);
113 	/*
114 	 * If parent is waiting for us to exit or exec, P_PPWAIT is set; we
115 	 * wake up the parent early to avoid deadlock.
116 	 */
117 	p->p_flag |= P_WEXIT;
118 	p->p_flag &= ~P_TRACED;
119 	if (p->p_flag & P_PPWAIT) {
120 		p->p_flag &= ~P_PPWAIT;
121 		wakeup(p->p_pptr);
122 	}
123 	p->p_sigignore = ~0;
124 	p->p_siglist = 0;
125 	timeout_del(&p->p_realit_to);
126 	timeout_del(&p->p_stats->p_virt_to);
127 	timeout_del(&p->p_stats->p_prof_to);
128 
129 	/*
130 	 * Close open files and release open-file table.
131 	 * This may block!
132 	 */
133 	fdfree(p);
134 
135 #ifdef SYSVSEM
136 	semexit(p);
137 #endif
138 	if (SESS_LEADER(p)) {
139 		struct session *sp = p->p_session;
140 
141 		if (sp->s_ttyvp) {
142 			/*
143 			 * Controlling process.
144 			 * Signal foreground pgrp,
145 			 * drain controlling terminal
146 			 * and revoke access to controlling terminal.
147 			 */
148 			if (sp->s_ttyp->t_session == sp) {
149 				if (sp->s_ttyp->t_pgrp)
150 					pgsignal(sp->s_ttyp->t_pgrp, SIGHUP, 1);
151 				(void) ttywait(sp->s_ttyp);
152 				/*
153 				 * The tty could have been revoked
154 				 * if we blocked.
155 				 */
156 				if (sp->s_ttyvp)
157 					VOP_REVOKE(sp->s_ttyvp, REVOKEALL);
158 			}
159 			if (sp->s_ttyvp)
160 				vrele(sp->s_ttyvp);
161 			sp->s_ttyvp = NULL;
162 			/*
163 			 * s_ttyp is not zero'd; we use this to indicate
164 			 * that the session once had a controlling terminal.
165 			 * (for logging and informational purposes)
166 			 */
167 		}
168 		sp->s_leader = NULL;
169 	}
170 	fixjobc(p, p->p_pgrp, 0);
171 #ifdef ACCOUNTING
172 	(void)acct_process(p);
173 #endif
174 #ifdef KTRACE
175 	/*
176 	 * release trace file
177 	 */
178 	p->p_traceflag = 0;	/* don't trace the vrele() */
179 	if (p->p_tracep)
180 		ktrsettracevnode(p, NULL);
181 #endif
182 #if NSYSTRACE > 0
183 	if (ISSET(p->p_flag, P_SYSTRACE))
184 		systrace_exit(p);
185 #endif
186 	/*
187 	 * NOTE: WE ARE NO LONGER ALLOWED TO SLEEP!
188 	 */
189 	p->p_stat = SDEAD;
190 
191         /*
192          * Remove proc from pidhash chain so looking it up won't
193          * work.  Move it from allproc to zombproc, but do not yet
194          * wake up the reaper.  We will put the proc on the
195          * deadproc list later (using the p_hash member), and
196          * wake up the reaper when we do.
197          */
198 	LIST_REMOVE(p, p_hash);
199 	LIST_REMOVE(p, p_list);
200 	LIST_INSERT_HEAD(&zombproc, p, p_list);
201 
202 	/*
203 	 * Give orphaned children to init(8).
204 	 */
205 	q = LIST_FIRST(&p->p_children);
206 	if (q)		/* only need this if any child is S_ZOMB */
207 		wakeup(initproc);
208 	for (; q != 0; q = nq) {
209 		nq = LIST_NEXT(q, p_sibling);
210 		proc_reparent(q, initproc);
211 		/*
212 		 * Traced processes are killed
213 		 * since their existence means someone is screwing up.
214 		 */
215 		if (q->p_flag & P_TRACED) {
216 			q->p_flag &= ~P_TRACED;
217 			psignal(q, SIGKILL);
218 		}
219 	}
220 
221 	/*
222 	 * Save exit status and final rusage info, adding in child rusage
223 	 * info and self times.
224 	 */
225 	p->p_xstat = rv;
226 	*p->p_ru = p->p_stats->p_ru;
227 	calcru(p, &p->p_ru->ru_utime, &p->p_ru->ru_stime, NULL);
228 	ruadd(p->p_ru, &p->p_stats->p_cru);
229 
230 	/*
231 	 * clear %cpu usage during swap
232 	 */
233 	p->p_pctcpu = 0;
234 
235 	/*
236 	 * notify interested parties of our demise.
237 	 */
238 	KNOTE(&p->p_klist, NOTE_EXIT);
239 
240 	/*
241 	 * Notify parent that we're gone.  If we have P_NOZOMBIE or parent has
242 	 * the P_NOCLDWAIT flag set, notify process 1 instead (and hope it
243 	 * will handle this situation).
244 	 */
245 	if ((p->p_flag & P_NOZOMBIE) || (p->p_pptr->p_flag & P_NOCLDWAIT)) {
246 		struct proc *pp = p->p_pptr;
247 		proc_reparent(p, initproc);
248 		/*
249 		 * If this was the last child of our parent, notify
250 		 * parent, so in case he was wait(2)ing, he will
251 		 * continue.
252 		 */
253 		if (LIST_EMPTY(&pp->p_children))
254 			wakeup(pp);
255 	}
256 
257 	if ((p->p_flag & P_FSTRACE) == 0 && p->p_exitsig != 0)
258 		psignal(p->p_pptr, P_EXITSIG(p));
259 	wakeup(p->p_pptr);
260 
261 	/*
262 	 * Notify procfs debugger
263 	 */
264 	if (p->p_flag & P_FSTRACE)
265 		wakeup(p);
266 
267 	/*
268 	 * Release the process's signal state.
269 	 */
270 	sigactsfree(p);
271 
272 	/*
273 	 * Clear curproc after we've done all operations
274 	 * that could block, and before tearing down the rest
275 	 * of the process state that might be used from clock, etc.
276 	 * Also, can't clear curproc while we're still runnable,
277 	 * as we're not on a run queue (we are current, just not
278 	 * a proper proc any longer!).
279 	 *
280 	 * Other substructures are freed from wait().
281 	 */
282 	curproc = NULL;
283 	limfree(p->p_limit);
284 	p->p_limit = NULL;
285 
286 	/* This process no longer needs to hold the kernel lock. */
287 	KERNEL_PROC_UNLOCK(p);
288 
289 	/*
290 	 * If emulation has process exit hook, call it now.
291 	 */
292 	if (p->p_emul->e_proc_exit)
293 		(*p->p_emul->e_proc_exit)(p);
294 
295 	/*
296 	 * Finally, call machine-dependent code to switch to a new
297 	 * context (possibly the idle context).  Once we are no longer
298 	 * using the dead process's vmspace and stack, exit2() will be
299 	 * called to schedule those resources to be released by the
300 	 * reaper thread.
301 	 *
302 	 * Note that cpu_exit() will end with a call equivalent to
303 	 * cpu_switch(), finishing our execution (pun intended).
304 	 */
305 	cpu_exit(p);
306 }
307 
308 /*
309  * Locking of this proclist is special; it's accessed in a
310  * critical section of process exit, and thus locking it can't
311  * modify interrupt state.  We use a simple spin lock for this
312  * proclist.  Processes on this proclist are also on zombproc;
313  * we use the p_hash member to linkup to deadproc.
314  */
315 struct mutex deadproc_mutex = MUTEX_INITIALIZER(IPL_NONE);
316 struct proclist deadproc = LIST_HEAD_INITIALIZER(deadproc);
317 
318 /*
319  * We are called from cpu_exit() once it is safe to schedule the
320  * dead process's resources to be freed.
321  *
322  * NOTE: One must be careful with locking in this routine.  It's
323  * called from a critical section in machine-dependent code, so
324  * we should refrain from changing any interrupt state.
325  *
326  * We lock the deadproc list, place the proc on that list (using
327  * the p_hash member), and wake up the reaper.
328  */
329 void
330 exit2(struct proc *p)
331 {
332 	int s;
333 
334 	mtx_enter(&deadproc_mutex);
335 	LIST_INSERT_HEAD(&deadproc, p, p_hash);
336 	mtx_leave(&deadproc_mutex);
337 
338 	wakeup(&deadproc);
339 
340 	SCHED_LOCK(s);
341 }
342 
343 /*
344  * Process reaper.  This is run by a kernel thread to free the resources
345  * of a dead process.  Once the resources are free, the process becomes
346  * a zombie, and the parent is allowed to read the undead's status.
347  */
348 void
349 reaper(void)
350 {
351 	struct proc *p;
352 
353 	KERNEL_PROC_UNLOCK(curproc);
354 
355 	for (;;) {
356 		mtx_enter(&deadproc_mutex);
357 		p = LIST_FIRST(&deadproc);
358 		if (p == NULL) {
359 			/* No work for us; go to sleep until someone exits. */
360 			mtx_leave(&deadproc_mutex);
361 			(void) tsleep(&deadproc, PVM, "reaper", 0);
362 			continue;
363 		}
364 
365 		/* Remove us from the deadproc list. */
366 		LIST_REMOVE(p, p_hash);
367 		mtx_leave(&deadproc_mutex);
368 		KERNEL_PROC_LOCK(curproc);
369 
370 		/*
371 		 * Give machine-dependent code a chance to free any
372 		 * resources it couldn't free while still running on
373 		 * that process's context.  This must be done before
374 		 * uvm_exit(), in case these resources are in the PCB.
375 		 */
376 		cpu_wait(p);
377 
378 		/*
379 		 * Free the VM resources we're still holding on to.
380 		 * We must do this from a valid thread because doing
381 		 * so may block.
382 		 */
383 		uvm_exit(p);
384 
385 		/* Process is now a true zombie. */
386 		if ((p->p_flag & P_NOZOMBIE) == 0) {
387 			p->p_stat = SZOMB;
388 
389 			/* Wake up the parent so it can get exit status. */
390 			psignal(p->p_pptr, SIGCHLD);
391 			wakeup(p->p_pptr);
392 		} else {
393 			/* Noone will wait for us. Just zap the process now */
394 			proc_zap(p);
395 		}
396 
397 		KERNEL_PROC_UNLOCK(curproc);
398 	}
399 }
400 
401 pid_t
402 sys_wait4(struct proc *q, void *v, register_t *retval)
403 {
404 	struct sys_wait4_args /* {
405 		syscallarg(pid_t) pid;
406 		syscallarg(int *) status;
407 		syscallarg(int) options;
408 		syscallarg(struct rusage *) rusage;
409 	} */ *uap = v;
410 	int nfound;
411 	struct proc *p, *t;
412 	int status, error;
413 
414 	if (SCARG(uap, pid) == 0)
415 		SCARG(uap, pid) = -q->p_pgid;
416 	if (SCARG(uap, options) &~ (WUNTRACED|WNOHANG|WALTSIG|WCONTINUED))
417 		return (EINVAL);
418 
419 loop:
420 	nfound = 0;
421 	LIST_FOREACH(p, &q->p_children, p_sibling) {
422 		if ((p->p_flag & P_NOZOMBIE) ||
423 		    (SCARG(uap, pid) != WAIT_ANY &&
424 		    p->p_pid != SCARG(uap, pid) &&
425 		    p->p_pgid != -SCARG(uap, pid)))
426 			continue;
427 
428 		/*
429 		 * Wait for processes with p_exitsig != SIGCHLD processes only
430 		 * if WALTSIG is set; wait for processes with pexitsig ==
431 		 * SIGCHLD only if WALTSIG is clear.
432 		 */
433 		if ((SCARG(uap, options) & WALTSIG) ?
434 		    (p->p_exitsig == SIGCHLD) : (P_EXITSIG(p) != SIGCHLD))
435 			continue;
436 
437 		nfound++;
438 		if (p->p_stat == SZOMB) {
439 			retval[0] = p->p_pid;
440 
441 			if (SCARG(uap, status)) {
442 				status = p->p_xstat;	/* convert to int */
443 				error = copyout(&status,
444 				    SCARG(uap, status), sizeof(status));
445 				if (error)
446 					return (error);
447 			}
448 			if (SCARG(uap, rusage) &&
449 			    (error = copyout(p->p_ru,
450 			    SCARG(uap, rusage), sizeof(struct rusage))))
451 				return (error);
452 
453 			/*
454 			 * If we got the child via a ptrace 'attach',
455 			 * we need to give it back to the old parent.
456 			 */
457 			if (p->p_oppid && (t = pfind(p->p_oppid))) {
458 				p->p_oppid = 0;
459 				proc_reparent(p, t);
460 				if (p->p_exitsig != 0)
461 					psignal(t, P_EXITSIG(p));
462 				wakeup(t);
463 				return (0);
464 			}
465 
466 			scheduler_wait_hook(q, p);
467 			p->p_xstat = 0;
468 			ruadd(&q->p_stats->p_cru, p->p_ru);
469 
470 			proc_zap(p);
471 
472 			return (0);
473 		}
474 		if (p->p_stat == SSTOP && (p->p_flag & P_WAITED) == 0 &&
475 		    (p->p_flag & P_TRACED || SCARG(uap, options) & WUNTRACED)) {
476 			p->p_flag |= P_WAITED;
477 			retval[0] = p->p_pid;
478 
479 			if (SCARG(uap, status)) {
480 				status = W_STOPCODE(p->p_xstat);
481 				error = copyout(&status, SCARG(uap, status),
482 				    sizeof(status));
483 			} else
484 				error = 0;
485 			return (error);
486 		}
487 		if ((SCARG(uap, options) & WCONTINUED) && (p->p_flag & P_CONTINUED)) {
488 			p->p_flag &= ~P_CONTINUED;
489 			retval[0] = p->p_pid;
490 
491 			if (SCARG(uap, status)) {
492 				status = _WCONTINUED;
493 				error = copyout(&status, SCARG(uap, status),
494 				    sizeof(status));
495 			} else
496 				error = 0;
497 			return (error);
498 		}
499 	}
500 	if (nfound == 0)
501 		return (ECHILD);
502 	if (SCARG(uap, options) & WNOHANG) {
503 		retval[0] = 0;
504 		return (0);
505 	}
506 	if ((error = tsleep(q, PWAIT | PCATCH, "wait", 0)) != 0)
507 		return (error);
508 	goto loop;
509 }
510 
511 /*
512  * make process 'parent' the new parent of process 'child'.
513  */
514 void
515 proc_reparent(struct proc *child, struct proc *parent)
516 {
517 
518 	if (child->p_pptr == parent)
519 		return;
520 
521 	if (parent == initproc)
522 		child->p_exitsig = SIGCHLD;
523 
524 	LIST_REMOVE(child, p_sibling);
525 	LIST_INSERT_HEAD(&parent->p_children, child, p_sibling);
526 	child->p_pptr = parent;
527 }
528 
529 void
530 proc_zap(struct proc *p)
531 {
532 	pool_put(&rusage_pool, p->p_ru);
533 	if (p->p_ptstat)
534 		free(p->p_ptstat, M_SUBPROC);
535 
536 	/*
537 	 * Finally finished with old proc entry.
538 	 * Unlink it from its process group and free it.
539 	 */
540 	leavepgrp(p);
541 	LIST_REMOVE(p, p_list);	/* off zombproc */
542 	LIST_REMOVE(p, p_sibling);
543 
544 	/*
545 	 * Decrement the count of procs running with this uid.
546 	 */
547 	(void)chgproccnt(p->p_cred->p_ruid, -1);
548 
549 	/*
550 	 * Free up credentials.
551 	 */
552 	if (--p->p_cred->p_refcnt == 0) {
553 		crfree(p->p_cred->pc_ucred);
554 		pool_put(&pcred_pool, p->p_cred);
555 	}
556 
557 	/*
558 	 * Release reference to text vnode
559 	 */
560 	if (p->p_textvp)
561 		vrele(p->p_textvp);
562 
563 	pool_put(&proc_pool, p);
564 	nprocs--;
565 }
566 
567