xref: /openbsd-src/sys/kern/kern_exit.c (revision 9d08f8e56428da94db79b31da31f4e37cb4022b7)
1 /*	$OpenBSD: kern_exit.c,v 1.54 2004/12/26 21:22:13 miod 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(p, v, retval)
86 	struct proc *p;
87 	void *v;
88 	register_t *retval;
89 {
90 	struct sys_exit_args /* {
91 		syscallarg(int) rval;
92 	} */ *uap = v;
93 
94 	exit1(p, W_EXITCODE(SCARG(uap, rval), 0));
95 	/* NOTREACHED */
96 	return (0);
97 }
98 
99 /*
100  * Exit: deallocate address space and other resources, change proc state
101  * to zombie, and unlink proc from allproc and parent's lists.  Save exit
102  * status and rusage for wait().  Check for child processes and orphan them.
103  */
104 void
105 exit1(p, rv)
106 	struct proc *p;
107 	int rv;
108 {
109 	struct proc *q, *nq;
110 
111 	if (p->p_pid == 1)
112 		panic("init died (signal %d, exit %d)",
113 		    WTERMSIG(rv), WEXITSTATUS(rv));
114 
115 	if (p->p_flag & P_PROFIL)
116 		stopprofclock(p);
117 	p->p_ru = pool_get(&rusage_pool, PR_WAITOK);
118 	/*
119 	 * If parent is waiting for us to exit or exec, P_PPWAIT is set; we
120 	 * wake up the parent early to avoid deadlock.
121 	 */
122 	p->p_flag |= P_WEXIT;
123 	p->p_flag &= ~P_TRACED;
124 	if (p->p_flag & P_PPWAIT) {
125 		p->p_flag &= ~P_PPWAIT;
126 		wakeup(p->p_pptr);
127 	}
128 	p->p_sigignore = ~0;
129 	p->p_siglist = 0;
130 	timeout_del(&p->p_realit_to);
131 	timeout_del(&p->p_stats->p_virt_to);
132 	timeout_del(&p->p_stats->p_prof_to);
133 
134 	/*
135 	 * Close open files and release open-file table.
136 	 * This may block!
137 	 */
138 	fdfree(p);
139 
140 #ifdef SYSVSEM
141 	semexit(p);
142 #endif
143 	if (SESS_LEADER(p)) {
144 		register struct session *sp = p->p_session;
145 
146 		if (sp->s_ttyvp) {
147 			/*
148 			 * Controlling process.
149 			 * Signal foreground pgrp,
150 			 * drain controlling terminal
151 			 * and revoke access to controlling terminal.
152 			 */
153 			if (sp->s_ttyp->t_session == sp) {
154 				if (sp->s_ttyp->t_pgrp)
155 					pgsignal(sp->s_ttyp->t_pgrp, SIGHUP, 1);
156 				(void) ttywait(sp->s_ttyp);
157 				/*
158 				 * The tty could have been revoked
159 				 * if we blocked.
160 				 */
161 				if (sp->s_ttyvp)
162 					VOP_REVOKE(sp->s_ttyvp, REVOKEALL);
163 			}
164 			if (sp->s_ttyvp)
165 				vrele(sp->s_ttyvp);
166 			sp->s_ttyvp = NULL;
167 			/*
168 			 * s_ttyp is not zero'd; we use this to indicate
169 			 * that the session once had a controlling terminal.
170 			 * (for logging and informational purposes)
171 			 */
172 		}
173 		sp->s_leader = NULL;
174 	}
175 	fixjobc(p, p->p_pgrp, 0);
176 #ifdef ACCOUNTING
177 	(void)acct_process(p);
178 #endif
179 #ifdef KTRACE
180 	/*
181 	 * release trace file
182 	 */
183 	p->p_traceflag = 0;	/* don't trace the vrele() */
184 	if (p->p_tracep)
185 		ktrsettracevnode(p, NULL);
186 #endif
187 #if NSYSTRACE > 0
188 	if (ISSET(p->p_flag, P_SYSTRACE))
189 		systrace_exit(p);
190 #endif
191 	/*
192 	 * NOTE: WE ARE NO LONGER ALLOWED TO SLEEP!
193 	 */
194 	p->p_stat = SDEAD;
195 
196         /*
197          * Remove proc from pidhash chain so looking it up won't
198          * work.  Move it from allproc to zombproc, but do not yet
199          * wake up the reaper.  We will put the proc on the
200          * deadproc list later (using the p_hash member), and
201          * wake up the reaper when we do.
202          */
203 	LIST_REMOVE(p, p_hash);
204 	LIST_REMOVE(p, p_list);
205 	LIST_INSERT_HEAD(&zombproc, p, p_list);
206 
207 	/*
208 	 * Give orphaned children to init(8).
209 	 */
210 	q = LIST_FIRST(&p->p_children);
211 	if (q)		/* only need this if any child is S_ZOMB */
212 		wakeup(initproc);
213 	for (; q != 0; q = nq) {
214 		nq = LIST_NEXT(q, p_sibling);
215 		proc_reparent(q, initproc);
216 		/*
217 		 * Traced processes are killed
218 		 * since their existence means someone is screwing up.
219 		 */
220 		if (q->p_flag & P_TRACED) {
221 			q->p_flag &= ~P_TRACED;
222 			psignal(q, SIGKILL);
223 		}
224 	}
225 
226 	/*
227 	 * Save exit status and final rusage info, adding in child rusage
228 	 * info and self times.
229 	 */
230 	p->p_xstat = rv;
231 	*p->p_ru = p->p_stats->p_ru;
232 	calcru(p, &p->p_ru->ru_utime, &p->p_ru->ru_stime, NULL);
233 	ruadd(p->p_ru, &p->p_stats->p_cru);
234 
235 	/*
236 	 * clear %cpu usage during swap
237 	 */
238 	p->p_pctcpu = 0;
239 
240 	/*
241 	 * notify interested parties of our demise.
242 	 */
243 	KNOTE(&p->p_klist, NOTE_EXIT);
244 
245 	/*
246 	 * Notify parent that we're gone.  If we have P_NOZOMBIE or parent has
247 	 * the P_NOCLDWAIT flag set, notify process 1 instead (and hope it
248 	 * will handle this situation).
249 	 */
250 	if ((p->p_flag & P_NOZOMBIE) || (p->p_pptr->p_flag & P_NOCLDWAIT)) {
251 		struct proc *pp = p->p_pptr;
252 		proc_reparent(p, initproc);
253 		/*
254 		 * If this was the last child of our parent, notify
255 		 * parent, so in case he was wait(2)ing, he will
256 		 * continue.
257 		 */
258 		if (LIST_EMPTY(&pp->p_children))
259 			wakeup(pp);
260 	}
261 
262 	if ((p->p_flag & P_FSTRACE) == 0 && p->p_exitsig != 0)
263 		psignal(p->p_pptr, P_EXITSIG(p));
264 	wakeup(p->p_pptr);
265 
266 	/*
267 	 * Notify procfs debugger
268 	 */
269 	if (p->p_flag & P_FSTRACE)
270 		wakeup(p);
271 
272 	/*
273 	 * Release the process's signal state.
274 	 */
275 	sigactsfree(p);
276 
277 	/*
278 	 * Clear curproc after we've done all operations
279 	 * that could block, and before tearing down the rest
280 	 * of the process state that might be used from clock, etc.
281 	 * Also, can't clear curproc while we're still runnable,
282 	 * as we're not on a run queue (we are current, just not
283 	 * a proper proc any longer!).
284 	 *
285 	 * Other substructures are freed from wait().
286 	 */
287 	curproc = NULL;
288 	limfree(p->p_limit);
289 	p->p_limit = NULL;
290 
291 	/* This process no longer needs to hold the kernel lock. */
292 	KERNEL_PROC_UNLOCK(p);
293 
294 	/*
295 	 * If emulation has process exit hook, call it now.
296 	 */
297 	if (p->p_emul->e_proc_exit)
298 		(*p->p_emul->e_proc_exit)(p);
299 
300 	/*
301 	 * Finally, call machine-dependent code to switch to a new
302 	 * context (possibly the idle context).  Once we are no longer
303 	 * using the dead process's vmspace and stack, exit2() will be
304 	 * called to schedule those resources to be released by the
305 	 * reaper thread.
306 	 *
307 	 * Note that cpu_exit() will end with a call equivalent to
308 	 * cpu_switch(), finishing our execution (pun intended).
309 	 */
310 	cpu_exit(p);
311 }
312 
313 /*
314  * Locking of this proclist is special; it's accessed in a
315  * critical section of process exit, and thus locking it can't
316  * modify interrupt state.  We use a simple spin lock for this
317  * proclist.  Processes on this proclist are also on zombproc;
318  * we use the p_hash member to linkup to deadproc.
319  */
320 struct mutex deadproc_mutex = MUTEX_INITIALIZER(IPL_NONE);
321 struct proclist deadproc = LIST_HEAD_INITIALIZER(deadproc);
322 
323 /*
324  * We are called from cpu_exit() once it is safe to schedule the
325  * dead process's resources to be freed.
326  *
327  * NOTE: One must be careful with locking in this routine.  It's
328  * called from a critical section in machine-dependent code, so
329  * we should refrain from changing any interrupt state.
330  *
331  * We lock the deadproc list, place the proc on that list (using
332  * the p_hash member), and wake up the reaper.
333  */
334 void
335 exit2(struct proc *p)
336 {
337 	int s;
338 
339 	mtx_enter(&deadproc_mutex);
340 	LIST_INSERT_HEAD(&deadproc, p, p_hash);
341 	mtx_leave(&deadproc_mutex);
342 
343 	wakeup(&deadproc);
344 
345 	SCHED_LOCK(s);
346 }
347 
348 /*
349  * Process reaper.  This is run by a kernel thread to free the resources
350  * of a dead process.  Once the resources are free, the process becomes
351  * a zombie, and the parent is allowed to read the undead's status.
352  */
353 void
354 reaper(void)
355 {
356 	struct proc *p;
357 
358 	KERNEL_PROC_UNLOCK(curproc);
359 
360 	for (;;) {
361 		mtx_enter(&deadproc_mutex);
362 		p = LIST_FIRST(&deadproc);
363 		if (p == NULL) {
364 			/* No work for us; go to sleep until someone exits. */
365 			mtx_leave(&deadproc_mutex);
366 			(void) tsleep(&deadproc, PVM, "reaper", 0);
367 			continue;
368 		}
369 
370 		/* Remove us from the deadproc list. */
371 		LIST_REMOVE(p, p_hash);
372 		mtx_leave(&deadproc_mutex);
373 		KERNEL_PROC_LOCK(curproc);
374 
375 		/*
376 		 * Give machine-dependent code a chance to free any
377 		 * resources it couldn't free while still running on
378 		 * that process's context.  This must be done before
379 		 * uvm_exit(), in case these resources are in the PCB.
380 		 */
381 		cpu_wait(p);
382 
383 		/*
384 		 * Free the VM resources we're still holding on to.
385 		 * We must do this from a valid thread because doing
386 		 * so may block.
387 		 */
388 		uvm_exit(p);
389 
390 		/* Process is now a true zombie. */
391 		if ((p->p_flag & P_NOZOMBIE) == 0) {
392 			p->p_stat = SZOMB;
393 
394 			/* Wake up the parent so it can get exit status. */
395 			psignal(p->p_pptr, SIGCHLD);
396 			wakeup(p->p_pptr);
397 		} else {
398 			/* Noone will wait for us. Just zap the process now */
399 			proc_zap(p);
400 		}
401 
402 		KERNEL_PROC_UNLOCK(curproc);
403 	}
404 }
405 
406 pid_t
407 sys_wait4(q, v, retval)
408 	register struct proc *q;
409 	void *v;
410 	register_t *retval;
411 {
412 	register struct sys_wait4_args /* {
413 		syscallarg(pid_t) pid;
414 		syscallarg(int *) status;
415 		syscallarg(int) options;
416 		syscallarg(struct rusage *) rusage;
417 	} */ *uap = v;
418 	register int nfound;
419 	register struct proc *p, *t;
420 	int status, error;
421 
422 	if (SCARG(uap, pid) == 0)
423 		SCARG(uap, pid) = -q->p_pgid;
424 	if (SCARG(uap, options) &~ (WUNTRACED|WNOHANG|WALTSIG|WCONTINUED))
425 		return (EINVAL);
426 
427 loop:
428 	nfound = 0;
429 	LIST_FOREACH(p, &q->p_children, p_sibling) {
430 		if ((p->p_flag & P_NOZOMBIE) ||
431 		    (SCARG(uap, pid) != WAIT_ANY &&
432 		    p->p_pid != SCARG(uap, pid) &&
433 		    p->p_pgid != -SCARG(uap, pid)))
434 			continue;
435 
436 		/*
437 		 * Wait for processes with p_exitsig != SIGCHLD processes only
438 		 * if WALTSIG is set; wait for processes with pexitsig ==
439 		 * SIGCHLD only if WALTSIG is clear.
440 		 */
441 		if ((SCARG(uap, options) & WALTSIG) ?
442 		    (p->p_exitsig == SIGCHLD) : (P_EXITSIG(p) != SIGCHLD))
443 			continue;
444 
445 		nfound++;
446 		if (p->p_stat == SZOMB) {
447 			retval[0] = p->p_pid;
448 
449 			if (SCARG(uap, status)) {
450 				status = p->p_xstat;	/* convert to int */
451 				error = copyout(&status,
452 				    SCARG(uap, status), sizeof(status));
453 				if (error)
454 					return (error);
455 			}
456 			if (SCARG(uap, rusage) &&
457 			    (error = copyout(p->p_ru,
458 			    SCARG(uap, rusage), sizeof(struct rusage))))
459 				return (error);
460 
461 			/*
462 			 * If we got the child via a ptrace 'attach',
463 			 * we need to give it back to the old parent.
464 			 */
465 			if (p->p_oppid && (t = pfind(p->p_oppid))) {
466 				p->p_oppid = 0;
467 				proc_reparent(p, t);
468 				if (p->p_exitsig != 0)
469 					psignal(t, P_EXITSIG(p));
470 				wakeup(t);
471 				return (0);
472 			}
473 
474 			scheduler_wait_hook(q, p);
475 			p->p_xstat = 0;
476 			ruadd(&q->p_stats->p_cru, p->p_ru);
477 
478 			proc_zap(p);
479 
480 			return (0);
481 		}
482 		if (p->p_stat == SSTOP && (p->p_flag & P_WAITED) == 0 &&
483 		    (p->p_flag & P_TRACED || SCARG(uap, options) & WUNTRACED)) {
484 			p->p_flag |= P_WAITED;
485 			retval[0] = p->p_pid;
486 
487 			if (SCARG(uap, status)) {
488 				status = W_STOPCODE(p->p_xstat);
489 				error = copyout(&status, SCARG(uap, status),
490 				    sizeof(status));
491 			} else
492 				error = 0;
493 			return (error);
494 		}
495 		if ((SCARG(uap, options) & WCONTINUED) && (p->p_flag & P_CONTINUED)) {
496 			p->p_flag &= ~P_CONTINUED;
497 			retval[0] = p->p_pid;
498 
499 			if (SCARG(uap, status)) {
500 				status = _WCONTINUED;
501 				error = copyout(&status, SCARG(uap, status),
502 				    sizeof(status));
503 			} else
504 				error = 0;
505 			return (error);
506 		}
507 	}
508 	if (nfound == 0)
509 		return (ECHILD);
510 	if (SCARG(uap, options) & WNOHANG) {
511 		retval[0] = 0;
512 		return (0);
513 	}
514 	if ((error = tsleep(q, PWAIT | PCATCH, "wait", 0)) != 0)
515 		return (error);
516 	goto loop;
517 }
518 
519 /*
520  * make process 'parent' the new parent of process 'child'.
521  */
522 void
523 proc_reparent(child, parent)
524 	register struct proc *child;
525 	register struct proc *parent;
526 {
527 
528 	if (child->p_pptr == parent)
529 		return;
530 
531 	if (parent == initproc)
532 		child->p_exitsig = SIGCHLD;
533 
534 	LIST_REMOVE(child, p_sibling);
535 	LIST_INSERT_HEAD(&parent->p_children, child, p_sibling);
536 	child->p_pptr = parent;
537 }
538 
539 void
540 proc_zap(struct proc *p)
541 {
542 	pool_put(&rusage_pool, p->p_ru);
543 
544 	/*
545 	 * Finally finished with old proc entry.
546 	 * Unlink it from its process group and free it.
547 	 */
548 	leavepgrp(p);
549 	LIST_REMOVE(p, p_list);	/* off zombproc */
550 	LIST_REMOVE(p, p_sibling);
551 
552 	/*
553 	 * Decrement the count of procs running with this uid.
554 	 */
555 	(void)chgproccnt(p->p_cred->p_ruid, -1);
556 
557 	/*
558 	 * Free up credentials.
559 	 */
560 	if (--p->p_cred->p_refcnt == 0) {
561 		crfree(p->p_cred->pc_ucred);
562 		pool_put(&pcred_pool, p->p_cred);
563 	}
564 
565 	/*
566 	 * Release reference to text vnode
567 	 */
568 	if (p->p_textvp)
569 		vrele(p->p_textvp);
570 
571 	pool_put(&proc_pool, p);
572 	nprocs--;
573 }
574 
575