xref: /openbsd-src/sys/kern/kern_exit.c (revision a28daedfc357b214be5c701aa8ba8adb29a7f1c2)
1 /*	$OpenBSD: kern_exit.c,v 1.84 2009/04/03 04:22:49 guenther 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 SYSVSEM
64 #include <sys/sem.h>
65 #endif
66 
67 #include "systrace.h"
68 #include <dev/systrace.h>
69 
70 #include <sys/mount.h>
71 #include <sys/syscallargs.h>
72 
73 #include <machine/cpu.h>
74 
75 #include <uvm/uvm_extern.h>
76 
77 /*
78  * exit --
79  *	Death of process.
80  */
81 int
82 sys_exit(struct proc *p, void *v, register_t *retval)
83 {
84 	struct sys_exit_args /* {
85 		syscallarg(int) rval;
86 	} */ *uap = v;
87 
88 	exit1(p, W_EXITCODE(SCARG(uap, rval), 0), EXIT_NORMAL);
89 	/* NOTREACHED */
90 	return (0);
91 }
92 
93 #ifdef RTHREADS
94 int
95 sys_threxit(struct proc *p, void *v, register_t *retval)
96 {
97 	struct sys_threxit_args /* {
98 		syscallarg(int) rval;
99 	} */ *uap = v;
100 
101 	exit1(p, W_EXITCODE(SCARG(uap, rval), 0), EXIT_THREAD);
102 
103 	return (0);
104 }
105 #endif
106 
107 /*
108  * Exit: deallocate address space and other resources, change proc state
109  * to zombie, and unlink proc from allproc and parent's lists.  Save exit
110  * status and rusage for wait().  Check for child processes and orphan them.
111  */
112 void
113 exit1(struct proc *p, int rv, int flags)
114 {
115 	struct proc *q, *nq;
116 
117 	if (p->p_pid == 1)
118 		panic("init died (signal %d, exit %d)",
119 		    WTERMSIG(rv), WEXITSTATUS(rv));
120 
121 	/* unlink ourselves from the active threads */
122 	TAILQ_REMOVE(&p->p_p->ps_threads, p, p_thr_link);
123 #ifdef RTHREADS
124 	if (TAILQ_EMPTY(&p->p_p->ps_threads))
125 		wakeup(&p->p_p->ps_threads);
126 	/*
127 	 * if one thread calls exit, we take down everybody.
128 	 * we have to be careful not to get recursively caught.
129 	 * this is kinda sick.
130 	 */
131 	if (flags == EXIT_NORMAL && (p->p_flag & P_THREAD) &&
132 	    (p->p_p->ps_mainproc->p_flag & P_WEXIT) == 0) {
133 		/*
134 		 * we are one of the threads.  we SIGKILL the parent,
135 		 * it will wake us up again, then we proceed.
136 		 */
137 		atomic_setbits_int(&p->p_p->ps_mainproc->p_flag, P_IGNEXITRV);
138 		p->p_p->ps_mainproc->p_xstat = rv;
139 		ptsignal(p->p_p->ps_mainproc, SIGKILL, SPROPAGATED);
140 		tsleep(p->p_p, PUSER, "thrdying", 0);
141 	} else if ((p->p_flag & P_THREAD) == 0) {
142 		atomic_setbits_int(&p->p_flag, P_WEXIT);
143 		if (flags == EXIT_NORMAL) {
144 			q = TAILQ_FIRST(&p->p_p->ps_threads);
145 			for (; q != NULL; q = nq) {
146 				nq = TAILQ_NEXT(q, p_thr_link);
147 				atomic_setbits_int(&q->p_flag, P_IGNEXITRV);
148 				q->p_xstat = rv;
149 				ptsignal(q, SIGKILL, SPROPAGATED);
150 			}
151 		}
152 		wakeup(p->p_p);
153 		while (!TAILQ_EMPTY(&p->p_p->ps_threads))
154 			tsleep(&p->p_p->ps_threads, PUSER, "thrdeath", 0);
155 	}
156 #endif
157 
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 	atomic_setbits_int(&p->p_flag, P_WEXIT);
166 	atomic_clearbits_int(&p->p_flag, P_TRACED);
167 	if (p->p_flag & P_PPWAIT) {
168 		atomic_clearbits_int(&p->p_flag, P_PPWAIT);
169 		wakeup(p->p_pptr);
170 	}
171 	p->p_sigignore = ~0;
172 	p->p_siglist = 0;
173 	timeout_del(&p->p_realit_to);
174 	timeout_del(&p->p_stats->p_virt_to);
175 	timeout_del(&p->p_stats->p_prof_to);
176 
177 	/*
178 	 * Close open files and release open-file table.
179 	 */
180 	fdfree(p);
181 
182 #ifdef SYSVSEM
183 	if ((p->p_flag & P_THREAD) == 0)
184 		semexit(p->p_p);
185 #endif
186 	if (SESS_LEADER(p)) {
187 		struct session *sp = p->p_session;
188 
189 		if (sp->s_ttyvp) {
190 			/*
191 			 * Controlling process.
192 			 * Signal foreground pgrp,
193 			 * drain controlling terminal
194 			 * and revoke access to controlling terminal.
195 			 */
196 			if (sp->s_ttyp->t_session == sp) {
197 				if (sp->s_ttyp->t_pgrp)
198 					pgsignal(sp->s_ttyp->t_pgrp, SIGHUP, 1);
199 				(void) ttywait(sp->s_ttyp);
200 				/*
201 				 * The tty could have been revoked
202 				 * if we blocked.
203 				 */
204 				if (sp->s_ttyvp)
205 					VOP_REVOKE(sp->s_ttyvp, REVOKEALL);
206 			}
207 			if (sp->s_ttyvp)
208 				vrele(sp->s_ttyvp);
209 			sp->s_ttyvp = NULL;
210 			/*
211 			 * s_ttyp is not zero'd; we use this to indicate
212 			 * that the session once had a controlling terminal.
213 			 * (for logging and informational purposes)
214 			 */
215 		}
216 		sp->s_leader = NULL;
217 	}
218 	fixjobc(p, p->p_pgrp, 0);
219 #ifdef ACCOUNTING
220 	(void)acct_process(p);
221 #endif
222 #ifdef KTRACE
223 	/*
224 	 * release trace file
225 	 */
226 	p->p_traceflag = 0;	/* don't trace the vrele() */
227 	if (p->p_tracep)
228 		ktrsettracevnode(p, NULL);
229 #endif
230 #if NSYSTRACE > 0
231 	if (ISSET(p->p_flag, P_SYSTRACE))
232 		systrace_exit(p);
233 #endif
234 	/*
235 	 * NOTE: WE ARE NO LONGER ALLOWED TO SLEEP!
236 	 */
237 	p->p_stat = SDEAD;
238 
239         /*
240          * Remove proc from pidhash chain so looking it up won't
241          * work.  Move it from allproc to zombproc, but do not yet
242          * wake up the reaper.  We will put the proc on the
243          * deadproc list later (using the p_hash member), and
244          * wake up the reaper when we do.
245          */
246 	LIST_REMOVE(p, p_hash);
247 	LIST_REMOVE(p, p_list);
248 	LIST_INSERT_HEAD(&zombproc, p, p_list);
249 
250 	/*
251 	 * Give orphaned children to init(8).
252 	 */
253 	q = LIST_FIRST(&p->p_children);
254 	if (q)		/* only need this if any child is S_ZOMB */
255 		wakeup(initproc);
256 	for (; q != 0; q = nq) {
257 		nq = LIST_NEXT(q, p_sibling);
258 		proc_reparent(q, initproc);
259 		/*
260 		 * Traced processes are killed
261 		 * since their existence means someone is screwing up.
262 		 */
263 		if (q->p_flag & P_TRACED) {
264 			atomic_clearbits_int(&q->p_flag, P_TRACED);
265 			psignal(q, SIGKILL);
266 		}
267 	}
268 
269 
270 	/*
271 	 * Save exit status and final rusage info, adding in child rusage
272 	 * info and self times.
273 	 */
274 	if (!(p->p_flag & P_IGNEXITRV))
275 		p->p_xstat = rv;
276 	*p->p_ru = p->p_stats->p_ru;
277 	calcru(p, &p->p_ru->ru_utime, &p->p_ru->ru_stime, NULL);
278 	ruadd(p->p_ru, &p->p_stats->p_cru);
279 
280 	/*
281 	 * clear %cpu usage during swap
282 	 */
283 	p->p_pctcpu = 0;
284 
285 	/*
286 	 * notify interested parties of our demise.
287 	 */
288 	KNOTE(&p->p_klist, NOTE_EXIT);
289 
290 	/*
291 	 * Notify parent that we're gone.  If we have P_NOZOMBIE or parent has
292 	 * the P_NOCLDWAIT flag set, notify process 1 instead (and hope it
293 	 * will handle this situation).
294 	 */
295 	if ((p->p_flag & P_NOZOMBIE) || (p->p_pptr->p_flag & P_NOCLDWAIT)) {
296 		struct proc *pp = p->p_pptr;
297 		proc_reparent(p, initproc);
298 		/*
299 		 * If this was the last child of our parent, notify
300 		 * parent, so in case he was wait(2)ing, he will
301 		 * continue.
302 		 */
303 		if (LIST_EMPTY(&pp->p_children))
304 			wakeup(pp);
305 	}
306 
307 	if (p->p_exitsig != 0)
308 		psignal(p->p_pptr, P_EXITSIG(p));
309 	wakeup(p->p_pptr);
310 
311 	/*
312 	 * Release the process's signal state.
313 	 */
314 	sigactsfree(p);
315 
316 	/*
317 	 * Other substructures are freed from reaper and wait().
318 	 */
319 
320 	/*
321 	 * If emulation has process exit hook, call it now.
322 	 */
323 	if (p->p_emul->e_proc_exit)
324 		(*p->p_emul->e_proc_exit)(p);
325 
326 	/* This process no longer needs to hold the kernel lock. */
327 	KERNEL_PROC_UNLOCK(p);
328 
329 	/*
330 	 * Finally, call machine-dependent code to switch to a new
331 	 * context (possibly the idle context).  Once we are no longer
332 	 * using the dead process's vmspace and stack, exit2() will be
333 	 * called to schedule those resources to be released by the
334 	 * reaper thread.
335 	 *
336 	 * Note that cpu_exit() will end with a call equivalent to
337 	 * cpu_switch(), finishing our execution (pun intended).
338 	 */
339 	uvmexp.swtch++;
340 	cpu_exit(p);
341 	panic("cpu_exit returned");
342 }
343 
344 /*
345  * Locking of this proclist is special; it's accessed in a
346  * critical section of process exit, and thus locking it can't
347  * modify interrupt state.  We use a simple spin lock for this
348  * proclist.  Processes on this proclist are also on zombproc;
349  * we use the p_hash member to linkup to deadproc.
350  */
351 struct mutex deadproc_mutex = MUTEX_INITIALIZER(IPL_NONE);
352 struct proclist deadproc = LIST_HEAD_INITIALIZER(deadproc);
353 
354 /*
355  * We are called from cpu_exit() once it is safe to schedule the
356  * dead process's resources to be freed.
357  *
358  * NOTE: One must be careful with locking in this routine.  It's
359  * called from a critical section in machine-dependent code, so
360  * we should refrain from changing any interrupt state.
361  *
362  * We lock the deadproc list, place the proc on that list (using
363  * the p_hash member), and wake up the reaper.
364  */
365 void
366 exit2(struct proc *p)
367 {
368 	mtx_enter(&deadproc_mutex);
369 	LIST_INSERT_HEAD(&deadproc, p, p_hash);
370 	mtx_leave(&deadproc_mutex);
371 
372 	wakeup(&deadproc);
373 }
374 
375 /*
376  * Process reaper.  This is run by a kernel thread to free the resources
377  * of a dead process.  Once the resources are free, the process becomes
378  * a zombie, and the parent is allowed to read the undead's status.
379  */
380 void
381 reaper(void)
382 {
383 	struct proc *p;
384 
385 	KERNEL_PROC_UNLOCK(curproc);
386 
387 	SCHED_ASSERT_UNLOCKED();
388 
389 	for (;;) {
390 		mtx_enter(&deadproc_mutex);
391 		while ((p = LIST_FIRST(&deadproc)) == NULL)
392 			msleep(&deadproc, &deadproc_mutex, PVM, "reaper", 0);
393 
394 		/* Remove us from the deadproc list. */
395 		LIST_REMOVE(p, p_hash);
396 		mtx_leave(&deadproc_mutex);
397 
398 		KERNEL_PROC_LOCK(curproc);
399 
400 		/*
401 		 * Free the VM resources we're still holding on to.
402 		 * We must do this from a valid thread because doing
403 		 * so may block.
404 		 */
405 		uvm_exit(p);
406 
407 		/* Process is now a true zombie. */
408 		if ((p->p_flag & P_NOZOMBIE) == 0) {
409 			p->p_stat = SZOMB;
410 
411 			/* Wake up the parent so it can get exit status. */
412 			psignal(p->p_pptr, SIGCHLD);
413 			wakeup(p->p_pptr);
414 		} else {
415 			/* Noone will wait for us. Just zap the process now */
416 			proc_zap(p);
417 		}
418 
419 		KERNEL_PROC_UNLOCK(curproc);
420 	}
421 }
422 
423 pid_t
424 sys_wait4(struct proc *q, void *v, register_t *retval)
425 {
426 	struct sys_wait4_args /* {
427 		syscallarg(pid_t) pid;
428 		syscallarg(int *) status;
429 		syscallarg(int) options;
430 		syscallarg(struct rusage *) rusage;
431 	} */ *uap = v;
432 	int nfound;
433 	struct proc *p, *t;
434 	int status, error;
435 
436 	if (SCARG(uap, pid) == 0)
437 		SCARG(uap, pid) = -q->p_pgid;
438 	if (SCARG(uap, options) &~ (WUNTRACED|WNOHANG|WALTSIG|WCONTINUED))
439 		return (EINVAL);
440 
441 loop:
442 	nfound = 0;
443 	LIST_FOREACH(p, &q->p_children, p_sibling) {
444 		if ((p->p_flag & P_NOZOMBIE) ||
445 		    (SCARG(uap, pid) != WAIT_ANY &&
446 		    p->p_pid != SCARG(uap, pid) &&
447 		    p->p_pgid != -SCARG(uap, pid)))
448 			continue;
449 
450 		/*
451 		 * Wait for processes with p_exitsig != SIGCHLD processes only
452 		 * if WALTSIG is set; wait for processes with pexitsig ==
453 		 * SIGCHLD only if WALTSIG is clear.
454 		 */
455 		if ((SCARG(uap, options) & WALTSIG) ?
456 		    (p->p_exitsig == SIGCHLD) : (P_EXITSIG(p) != SIGCHLD))
457 			continue;
458 
459 		nfound++;
460 		if (p->p_stat == SZOMB) {
461 			retval[0] = p->p_pid;
462 
463 			if (SCARG(uap, status)) {
464 				status = p->p_xstat;	/* convert to int */
465 				error = copyout(&status,
466 				    SCARG(uap, status), sizeof(status));
467 				if (error)
468 					return (error);
469 			}
470 			if (SCARG(uap, rusage) &&
471 			    (error = copyout(p->p_ru,
472 			    SCARG(uap, rusage), sizeof(struct rusage))))
473 				return (error);
474 
475 			/*
476 			 * If we got the child via a ptrace 'attach',
477 			 * we need to give it back to the old parent.
478 			 */
479 			if (p->p_oppid && (t = pfind(p->p_oppid))) {
480 				p->p_oppid = 0;
481 				proc_reparent(p, t);
482 				if (p->p_exitsig != 0)
483 					psignal(t, P_EXITSIG(p));
484 				wakeup(t);
485 				return (0);
486 			}
487 
488 			scheduler_wait_hook(q, p);
489 			p->p_xstat = 0;
490 			ruadd(&q->p_stats->p_cru, p->p_ru);
491 
492 			proc_zap(p);
493 
494 			return (0);
495 		}
496 		if (p->p_stat == SSTOP && (p->p_flag & P_WAITED) == 0 &&
497 		    (p->p_flag & P_TRACED || SCARG(uap, options) & WUNTRACED)) {
498 			atomic_setbits_int(&p->p_flag, P_WAITED);
499 			retval[0] = p->p_pid;
500 
501 			if (SCARG(uap, status)) {
502 				status = W_STOPCODE(p->p_xstat);
503 				error = copyout(&status, SCARG(uap, status),
504 				    sizeof(status));
505 			} else
506 				error = 0;
507 			return (error);
508 		}
509 		if ((SCARG(uap, options) & WCONTINUED) && (p->p_flag & P_CONTINUED)) {
510 			atomic_clearbits_int(&p->p_flag, P_CONTINUED);
511 			retval[0] = p->p_pid;
512 
513 			if (SCARG(uap, status)) {
514 				status = _WCONTINUED;
515 				error = copyout(&status, SCARG(uap, status),
516 				    sizeof(status));
517 			} else
518 				error = 0;
519 			return (error);
520 		}
521 	}
522 	if (nfound == 0)
523 		return (ECHILD);
524 	if (SCARG(uap, options) & WNOHANG) {
525 		retval[0] = 0;
526 		return (0);
527 	}
528 	if ((error = tsleep(q, PWAIT | PCATCH, "wait", 0)) != 0)
529 		return (error);
530 	goto loop;
531 }
532 
533 /*
534  * make process 'parent' the new parent of process 'child'.
535  */
536 void
537 proc_reparent(struct proc *child, struct proc *parent)
538 {
539 
540 	if (child->p_pptr == parent)
541 		return;
542 
543 	if (parent == initproc)
544 		child->p_exitsig = SIGCHLD;
545 
546 	LIST_REMOVE(child, p_sibling);
547 	LIST_INSERT_HEAD(&parent->p_children, child, p_sibling);
548 	child->p_pptr = parent;
549 }
550 
551 void
552 proc_zap(struct proc *p)
553 {
554 	pool_put(&rusage_pool, p->p_ru);
555 	if (p->p_ptstat)
556 		free(p->p_ptstat, M_SUBPROC);
557 
558 	/*
559 	 * Finally finished with old proc entry.
560 	 * Unlink it from its process group and free it.
561 	 */
562 	leavepgrp(p);
563 	LIST_REMOVE(p, p_list);	/* off zombproc */
564 	LIST_REMOVE(p, p_sibling);
565 
566 	/*
567 	 * Decrement the count of procs running with this uid.
568 	 */
569 	(void)chgproccnt(p->p_cred->p_ruid, -1);
570 
571 	/*
572 	 * Release reference to text vnode
573 	 */
574 	if (p->p_textvp)
575 		vrele(p->p_textvp);
576 
577 	/*
578 	 * Remove us from our process list, possibly killing the process
579 	 * in the process (pun intended).
580 	 */
581 	if (--p->p_p->ps_refcnt == 0) {
582 		KASSERT(TAILQ_EMPTY(&p->p_p->ps_threads));
583 		limfree(p->p_p->ps_limit);
584 		if (--p->p_p->ps_cred->p_refcnt == 0) {
585 			crfree(p->p_p->ps_cred->pc_ucred);
586 			pool_put(&pcred_pool, p->p_p->ps_cred);
587 		}
588 		pool_put(&process_pool, p->p_p);
589 	}
590 
591 	pool_put(&proc_pool, p);
592 	nprocs--;
593 }
594 
595