xref: /openbsd-src/sys/kern/kern_exit.c (revision 850e275390052b330d93020bf619a739a3c277ac)
1 /*	$OpenBSD: kern_exit.c,v 1.73 2008/05/11 23:54:40 tedu 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), EXIT_NORMAL);
92 	/* NOTREACHED */
93 	return (0);
94 }
95 
96 #ifdef RTHREADS
97 int
98 sys_threxit(struct proc *p, void *v, register_t *retval)
99 {
100 	struct sys_threxit_args *uap = v;
101 
102 	exit1(p, W_EXITCODE(SCARG(uap, rval), 0), EXIT_THREAD);
103 
104 	return (0);
105 }
106 #endif
107 
108 /*
109  * Exit: deallocate address space and other resources, change proc state
110  * to zombie, and unlink proc from allproc and parent's lists.  Save exit
111  * status and rusage for wait().  Check for child processes and orphan them.
112  */
113 void
114 exit1(struct proc *p, int rv, int flags)
115 {
116 	struct proc *q, *nq;
117 
118 	if (p->p_pid == 1)
119 		panic("init died (signal %d, exit %d)",
120 		    WTERMSIG(rv), WEXITSTATUS(rv));
121 
122 	/* unlink ourselves from the active threads */
123 	TAILQ_REMOVE(&p->p_p->ps_threads, p, p_thr_link);
124 #ifdef RTHREADS
125 	if (TAILQ_EMPTY(&p->p_p->ps_threads))
126 		wakeup(&p->p_p->ps_threads);
127 	/*
128 	 * if one thread calls exit, we take down everybody.
129 	 * we have to be careful not to get recursively caught.
130 	 * this is kinda sick.
131 	 */
132 	if (flags == EXIT_NORMAL && p->p_p->ps_mainproc != p &&
133 	    (p->p_p->ps_mainproc->p_flag & P_WEXIT) == 0) {
134 		/*
135 		 * we are one of the threads.  we SIGKILL the parent,
136 		 * it will wake us up again, then we proceed.
137 		 */
138 		atomic_setbits_int(&p->p_p->ps_mainproc->p_flag, P_IGNEXITRV);
139 		p->p_p->ps_mainproc->p_xstat = rv;
140 		psignal(p->p_p->ps_mainproc, SIGKILL);
141 		tsleep(p->p_p, PUSER, "thrdying", 0);
142 	} else if (p == p->p_p->ps_mainproc) {
143 		atomic_setbits_int(&p->p_flag, P_WEXIT);
144 		if (flags == EXIT_NORMAL) {
145 			q = TAILQ_FIRST(&p->p_p->ps_threads);
146 			for (; q != NULL; q = nq) {
147 				nq = TAILQ_NEXT(q, p_thr_link);
148 				atomic_setbits_int(&q->p_flag, P_IGNEXITRV);
149 				q->p_xstat = rv;
150 				psignal(q, SIGKILL);
151 			}
152 		}
153 		wakeup(p->p_p);
154 		while (!TAILQ_EMPTY(&p->p_p->ps_threads))
155 			tsleep(&p->p_p->ps_threads, PUSER, "thrdeath", 0);
156 	}
157 #endif
158 
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 	atomic_setbits_int(&p->p_flag, P_WEXIT);
167 	atomic_clearbits_int(&p->p_flag, P_TRACED);
168 	if (p->p_flag & P_PPWAIT) {
169 		atomic_clearbits_int(&p->p_flag, P_PPWAIT);
170 		wakeup(p->p_pptr);
171 	}
172 	p->p_sigignore = ~0;
173 	p->p_siglist = 0;
174 	timeout_del(&p->p_realit_to);
175 	timeout_del(&p->p_stats->p_virt_to);
176 	timeout_del(&p->p_stats->p_prof_to);
177 
178 	/*
179 	 * Close open files and release open-file table.
180 	 * This may block!
181 	 */
182 	fdfree(p);
183 
184 #ifdef SYSVSEM
185 	semexit(p);
186 #endif
187 	if (SESS_LEADER(p)) {
188 		struct session *sp = p->p_session;
189 
190 		if (sp->s_ttyvp) {
191 			/*
192 			 * Controlling process.
193 			 * Signal foreground pgrp,
194 			 * drain controlling terminal
195 			 * and revoke access to controlling terminal.
196 			 */
197 			if (sp->s_ttyp->t_session == sp) {
198 				if (sp->s_ttyp->t_pgrp)
199 					pgsignal(sp->s_ttyp->t_pgrp, SIGHUP, 1);
200 				(void) ttywait(sp->s_ttyp);
201 				/*
202 				 * The tty could have been revoked
203 				 * if we blocked.
204 				 */
205 				if (sp->s_ttyvp)
206 					VOP_REVOKE(sp->s_ttyvp, REVOKEALL);
207 			}
208 			if (sp->s_ttyvp)
209 				vrele(sp->s_ttyvp);
210 			sp->s_ttyvp = NULL;
211 			/*
212 			 * s_ttyp is not zero'd; we use this to indicate
213 			 * that the session once had a controlling terminal.
214 			 * (for logging and informational purposes)
215 			 */
216 		}
217 		sp->s_leader = NULL;
218 	}
219 	fixjobc(p, p->p_pgrp, 0);
220 #ifdef ACCOUNTING
221 	(void)acct_process(p);
222 #endif
223 #ifdef KTRACE
224 	/*
225 	 * release trace file
226 	 */
227 	p->p_traceflag = 0;	/* don't trace the vrele() */
228 	if (p->p_tracep)
229 		ktrsettracevnode(p, NULL);
230 #endif
231 #if NSYSTRACE > 0
232 	if (ISSET(p->p_flag, P_SYSTRACE))
233 		systrace_exit(p);
234 #endif
235 	/*
236 	 * NOTE: WE ARE NO LONGER ALLOWED TO SLEEP!
237 	 */
238 	p->p_stat = SDEAD;
239 
240         /*
241          * Remove proc from pidhash chain so looking it up won't
242          * work.  Move it from allproc to zombproc, but do not yet
243          * wake up the reaper.  We will put the proc on the
244          * deadproc list later (using the p_hash member), and
245          * wake up the reaper when we do.
246          */
247 	LIST_REMOVE(p, p_hash);
248 	LIST_REMOVE(p, p_list);
249 	LIST_INSERT_HEAD(&zombproc, p, p_list);
250 
251 	/*
252 	 * Give orphaned children to init(8).
253 	 */
254 	q = LIST_FIRST(&p->p_children);
255 	if (q)		/* only need this if any child is S_ZOMB */
256 		wakeup(initproc);
257 	for (; q != 0; q = nq) {
258 		nq = LIST_NEXT(q, p_sibling);
259 		proc_reparent(q, initproc);
260 		/*
261 		 * Traced processes are killed
262 		 * since their existence means someone is screwing up.
263 		 */
264 		if (q->p_flag & P_TRACED) {
265 			atomic_clearbits_int(&q->p_flag, P_TRACED);
266 			psignal(q, SIGKILL);
267 		}
268 	}
269 
270 
271 	/*
272 	 * Save exit status and final rusage info, adding in child rusage
273 	 * info and self times.
274 	 */
275 	if (!(p->p_flag & P_IGNEXITRV))
276 		p->p_xstat = rv;
277 	*p->p_ru = p->p_stats->p_ru;
278 	calcru(p, &p->p_ru->ru_utime, &p->p_ru->ru_stime, NULL);
279 	ruadd(p->p_ru, &p->p_stats->p_cru);
280 
281 	/*
282 	 * clear %cpu usage during swap
283 	 */
284 	p->p_pctcpu = 0;
285 
286 	/*
287 	 * notify interested parties of our demise.
288 	 */
289 	KNOTE(&p->p_klist, NOTE_EXIT);
290 
291 	/*
292 	 * Notify parent that we're gone.  If we have P_NOZOMBIE or parent has
293 	 * the P_NOCLDWAIT flag set, notify process 1 instead (and hope it
294 	 * will handle this situation).
295 	 */
296 	if ((p->p_flag & P_NOZOMBIE) || (p->p_pptr->p_flag & P_NOCLDWAIT)) {
297 		struct proc *pp = p->p_pptr;
298 		proc_reparent(p, initproc);
299 		/*
300 		 * If this was the last child of our parent, notify
301 		 * parent, so in case he was wait(2)ing, he will
302 		 * continue.
303 		 */
304 		if (LIST_EMPTY(&pp->p_children))
305 			wakeup(pp);
306 	}
307 
308 	if (p->p_exitsig != 0)
309 		psignal(p->p_pptr, P_EXITSIG(p));
310 	wakeup(p->p_pptr);
311 
312 	/*
313 	 * Release the process's signal state.
314 	 */
315 	sigactsfree(p);
316 
317 	/*
318 	 * Other substructures are freed from reaper and wait().
319 	 */
320 
321 	/*
322 	 * If emulation has process exit hook, call it now.
323 	 */
324 	if (p->p_emul->e_proc_exit)
325 		(*p->p_emul->e_proc_exit)(p);
326 
327 	/* This process no longer needs to hold the kernel lock. */
328 	KERNEL_PROC_UNLOCK(p);
329 
330 	/*
331 	 * Finally, call machine-dependent code to switch to a new
332 	 * context (possibly the idle context).  Once we are no longer
333 	 * using the dead process's vmspace and stack, exit2() will be
334 	 * called to schedule those resources to be released by the
335 	 * reaper thread.
336 	 *
337 	 * Note that cpu_exit() will end with a call equivalent to
338 	 * cpu_switch(), finishing our execution (pun intended).
339 	 */
340 	uvmexp.swtch++;
341 	cpu_exit(p);
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 		p = LIST_FIRST(&deadproc);
392 		if (p == NULL) {
393 			/* No work for us; go to sleep until someone exits. */
394 			mtx_leave(&deadproc_mutex);
395 			(void) tsleep(&deadproc, PVM, "reaper", 0);
396 			continue;
397 		}
398 
399 		/* Remove us from the deadproc list. */
400 		LIST_REMOVE(p, p_hash);
401 		mtx_leave(&deadproc_mutex);
402 		KERNEL_PROC_LOCK(curproc);
403 
404 		/*
405 		 * Give machine-dependent code a chance to free any
406 		 * resources it couldn't free while still running on
407 		 * that process's context.  This must be done before
408 		 * uvm_exit(), in case these resources are in the PCB.
409 		 */
410 		cpu_wait(p);
411 
412 		/*
413 		 * Free the VM resources we're still holding on to.
414 		 * We must do this from a valid thread because doing
415 		 * so may block.
416 		 */
417 		uvm_exit(p);
418 
419 		/* Process is now a true zombie. */
420 		if ((p->p_flag & P_NOZOMBIE) == 0) {
421 			p->p_stat = SZOMB;
422 
423 			/* Wake up the parent so it can get exit status. */
424 			psignal(p->p_pptr, SIGCHLD);
425 			wakeup(p->p_pptr);
426 		} else {
427 			/* Noone will wait for us. Just zap the process now */
428 			proc_zap(p);
429 		}
430 
431 		KERNEL_PROC_UNLOCK(curproc);
432 	}
433 }
434 
435 pid_t
436 sys_wait4(struct proc *q, void *v, register_t *retval)
437 {
438 	struct sys_wait4_args /* {
439 		syscallarg(pid_t) pid;
440 		syscallarg(int *) status;
441 		syscallarg(int) options;
442 		syscallarg(struct rusage *) rusage;
443 	} */ *uap = v;
444 	int nfound;
445 	struct proc *p, *t;
446 	int status, error;
447 
448 	if (SCARG(uap, pid) == 0)
449 		SCARG(uap, pid) = -q->p_pgid;
450 	if (SCARG(uap, options) &~ (WUNTRACED|WNOHANG|WALTSIG|WCONTINUED))
451 		return (EINVAL);
452 
453 loop:
454 	nfound = 0;
455 	LIST_FOREACH(p, &q->p_children, p_sibling) {
456 		if ((p->p_flag & P_NOZOMBIE) ||
457 		    (SCARG(uap, pid) != WAIT_ANY &&
458 		    p->p_pid != SCARG(uap, pid) &&
459 		    p->p_pgid != -SCARG(uap, pid)))
460 			continue;
461 
462 		/*
463 		 * Wait for processes with p_exitsig != SIGCHLD processes only
464 		 * if WALTSIG is set; wait for processes with pexitsig ==
465 		 * SIGCHLD only if WALTSIG is clear.
466 		 */
467 		if ((SCARG(uap, options) & WALTSIG) ?
468 		    (p->p_exitsig == SIGCHLD) : (P_EXITSIG(p) != SIGCHLD))
469 			continue;
470 
471 		nfound++;
472 		if (p->p_stat == SZOMB) {
473 			retval[0] = p->p_pid;
474 
475 			if (SCARG(uap, status)) {
476 				status = p->p_xstat;	/* convert to int */
477 				error = copyout(&status,
478 				    SCARG(uap, status), sizeof(status));
479 				if (error)
480 					return (error);
481 			}
482 			if (SCARG(uap, rusage) &&
483 			    (error = copyout(p->p_ru,
484 			    SCARG(uap, rusage), sizeof(struct rusage))))
485 				return (error);
486 
487 			/*
488 			 * If we got the child via a ptrace 'attach',
489 			 * we need to give it back to the old parent.
490 			 */
491 			if (p->p_oppid && (t = pfind(p->p_oppid))) {
492 				p->p_oppid = 0;
493 				proc_reparent(p, t);
494 				if (p->p_exitsig != 0)
495 					psignal(t, P_EXITSIG(p));
496 				wakeup(t);
497 				return (0);
498 			}
499 
500 			scheduler_wait_hook(q, p);
501 			p->p_xstat = 0;
502 			ruadd(&q->p_stats->p_cru, p->p_ru);
503 
504 			proc_zap(p);
505 
506 			return (0);
507 		}
508 		if (p->p_stat == SSTOP && (p->p_flag & P_WAITED) == 0 &&
509 		    (p->p_flag & P_TRACED || SCARG(uap, options) & WUNTRACED)) {
510 			atomic_setbits_int(&p->p_flag, P_WAITED);
511 			retval[0] = p->p_pid;
512 
513 			if (SCARG(uap, status)) {
514 				status = W_STOPCODE(p->p_xstat);
515 				error = copyout(&status, SCARG(uap, status),
516 				    sizeof(status));
517 			} else
518 				error = 0;
519 			return (error);
520 		}
521 		if ((SCARG(uap, options) & WCONTINUED) && (p->p_flag & P_CONTINUED)) {
522 			atomic_clearbits_int(&p->p_flag, P_CONTINUED);
523 			retval[0] = p->p_pid;
524 
525 			if (SCARG(uap, status)) {
526 				status = _WCONTINUED;
527 				error = copyout(&status, SCARG(uap, status),
528 				    sizeof(status));
529 			} else
530 				error = 0;
531 			return (error);
532 		}
533 	}
534 	if (nfound == 0)
535 		return (ECHILD);
536 	if (SCARG(uap, options) & WNOHANG) {
537 		retval[0] = 0;
538 		return (0);
539 	}
540 	if ((error = tsleep(q, PWAIT | PCATCH, "wait", 0)) != 0)
541 		return (error);
542 	goto loop;
543 }
544 
545 /*
546  * make process 'parent' the new parent of process 'child'.
547  */
548 void
549 proc_reparent(struct proc *child, struct proc *parent)
550 {
551 
552 	if (child->p_pptr == parent)
553 		return;
554 
555 	if (parent == initproc)
556 		child->p_exitsig = SIGCHLD;
557 
558 	LIST_REMOVE(child, p_sibling);
559 	LIST_INSERT_HEAD(&parent->p_children, child, p_sibling);
560 	child->p_pptr = parent;
561 }
562 
563 void
564 proc_zap(struct proc *p)
565 {
566 	pool_put(&rusage_pool, p->p_ru);
567 	if (p->p_ptstat)
568 		free(p->p_ptstat, M_SUBPROC);
569 
570 	/*
571 	 * Finally finished with old proc entry.
572 	 * Unlink it from its process group and free it.
573 	 */
574 	leavepgrp(p);
575 	LIST_REMOVE(p, p_list);	/* off zombproc */
576 	LIST_REMOVE(p, p_sibling);
577 
578 	/*
579 	 * Decrement the count of procs running with this uid.
580 	 */
581 	(void)chgproccnt(p->p_cred->p_ruid, -1);
582 
583 	/*
584 	 * Release reference to text vnode
585 	 */
586 	if (p->p_textvp)
587 		vrele(p->p_textvp);
588 
589 	/*
590 	 * Remove us from our process list, possibly killing the process
591 	 * in the process (pun intended).
592 	 */
593 	if ((p->p_flag & P_THREAD) == 0) {
594 		KASSERT(TAILQ_EMPTY(&p->p_p->ps_threads));
595 		limfree(p->p_p->ps_limit);
596 		if (--p->p_p->ps_cred->p_refcnt == 0) {
597 			crfree(p->p_p->ps_cred->pc_ucred);
598 			pool_put(&pcred_pool, p->p_p->ps_cred);
599 		}
600 		pool_put(&process_pool, p->p_p);
601 	}
602 
603 	pool_put(&proc_pool, p);
604 	nprocs--;
605 }
606 
607