xref: /openbsd-src/sys/kern/kern_exit.c (revision 802bb6ac573f1028eee6b171a32fb0694b0dd4eb)
1 /*	$OpenBSD: kern_exit.c,v 1.57 2005/12/03 18:09:08 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 	/*
123 	 * if one thread calls exit, we take down everybody.
124 	 * we have to be careful not to get recursively caught.
125 	 * this is kinda sick.
126 	 */
127 	if (flags == EXIT_NORMAL && p != p->p_thrparent &&
128 	    (p->p_thrparent->p_flag & P_WEXIT) == 0) {
129 		printf("thread exiting normally %d\n", p->p_pid);
130 		/*
131 		 * we are one of the threads.  we SIGKILL the parent,
132 		 * then wait for it to kill us back.  as soon as we return,
133 		 * we'll exit again.
134 		 */
135 		p->p_thrparent->p_flag |= P_IGNEXITRV;
136 		p->p_thrparent->p_xstat = rv;
137 		psignal(p->p_thrparent, SIGKILL);
138 		tsleep(&p->p_thrparent->p_thrchildren, PUSER | PCATCH, "dying",
139 		    0);
140 		printf("thread got sig %d\n", p->p_pid);
141 		return;
142 	} else if (p == p->p_thrparent) {
143 		p->p_flag |= P_WEXIT;
144 		if (flags == EXIT_NORMAL) {
145 			q = LIST_FIRST(&p->p_thrchildren);
146 			for (; q != 0; q = nq) {
147 				nq = LIST_NEXT(q, p_thrsib);
148 				q->p_flag |= P_IGNEXITRV;
149 				q->p_xstat = rv;
150 				printf("parent killing child %d\n", q->p_pid);
151 				psignal(q, SIGKILL);
152 			}
153 		}
154 		while (!LIST_EMPTY(&p->p_thrchildren))
155 			tsleep(&p->p_thrchildren, PUSER, "thrdeath", 0);
156 	}
157 
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 	p->p_flag |= P_WEXIT;
167 	p->p_flag &= ~P_TRACED;
168 	if (p->p_flag & P_PPWAIT) {
169 		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 			q->p_flag &= ~P_TRACED;
266 			psignal(q, SIGKILL);
267 		}
268 	}
269 
270 	/* unlink oursleves from the active threads */
271 	if (p != p->p_thrparent) {
272 		LIST_REMOVE(p, p_thrsib);
273 		if (LIST_EMPTY(&p->p_thrparent->p_thrchildren))
274 			wakeup(&p->p_thrparent->p_thrchildren);
275 	}
276 
277 
278 	/*
279 	 * Save exit status and final rusage info, adding in child rusage
280 	 * info and self times.
281 	 */
282 	if (!(p->p_flag & P_IGNEXITRV))
283 		p->p_xstat = rv;
284 	*p->p_ru = p->p_stats->p_ru;
285 	calcru(p, &p->p_ru->ru_utime, &p->p_ru->ru_stime, NULL);
286 	ruadd(p->p_ru, &p->p_stats->p_cru);
287 
288 	/*
289 	 * clear %cpu usage during swap
290 	 */
291 	p->p_pctcpu = 0;
292 
293 	/*
294 	 * notify interested parties of our demise.
295 	 */
296 	KNOTE(&p->p_klist, NOTE_EXIT);
297 
298 	/*
299 	 * Notify parent that we're gone.  If we have P_NOZOMBIE or parent has
300 	 * the P_NOCLDWAIT flag set, notify process 1 instead (and hope it
301 	 * will handle this situation).
302 	 */
303 	if ((p->p_flag & P_NOZOMBIE) || (p->p_pptr->p_flag & P_NOCLDWAIT)) {
304 		struct proc *pp = p->p_pptr;
305 		proc_reparent(p, initproc);
306 		/*
307 		 * If this was the last child of our parent, notify
308 		 * parent, so in case he was wait(2)ing, he will
309 		 * continue.
310 		 */
311 		if (LIST_EMPTY(&pp->p_children))
312 			wakeup(pp);
313 	}
314 
315 	if ((p->p_flag & P_FSTRACE) == 0 && p->p_exitsig != 0)
316 		psignal(p->p_pptr, P_EXITSIG(p));
317 	wakeup(p->p_pptr);
318 
319 	/*
320 	 * Notify procfs debugger
321 	 */
322 	if (p->p_flag & P_FSTRACE)
323 		wakeup(p);
324 
325 	/*
326 	 * Release the process's signal state.
327 	 */
328 	sigactsfree(p);
329 
330 	/*
331 	 * Clear curproc after we've done all operations
332 	 * that could block, and before tearing down the rest
333 	 * of the process state that might be used from clock, etc.
334 	 * Also, can't clear curproc while we're still runnable,
335 	 * as we're not on a run queue (we are current, just not
336 	 * a proper proc any longer!).
337 	 *
338 	 * Other substructures are freed from wait().
339 	 */
340 	curproc = NULL;
341 	limfree(p->p_limit);
342 	p->p_limit = NULL;
343 
344 	/* This process no longer needs to hold the kernel lock. */
345 	KERNEL_PROC_UNLOCK(p);
346 
347 	/*
348 	 * If emulation has process exit hook, call it now.
349 	 */
350 	if (p->p_emul->e_proc_exit)
351 		(*p->p_emul->e_proc_exit)(p);
352 
353 	/*
354 	 * Finally, call machine-dependent code to switch to a new
355 	 * context (possibly the idle context).  Once we are no longer
356 	 * using the dead process's vmspace and stack, exit2() will be
357 	 * called to schedule those resources to be released by the
358 	 * reaper thread.
359 	 *
360 	 * Note that cpu_exit() will end with a call equivalent to
361 	 * cpu_switch(), finishing our execution (pun intended).
362 	 */
363 	cpu_exit(p);
364 }
365 
366 /*
367  * Locking of this proclist is special; it's accessed in a
368  * critical section of process exit, and thus locking it can't
369  * modify interrupt state.  We use a simple spin lock for this
370  * proclist.  Processes on this proclist are also on zombproc;
371  * we use the p_hash member to linkup to deadproc.
372  */
373 struct mutex deadproc_mutex = MUTEX_INITIALIZER(IPL_NONE);
374 struct proclist deadproc = LIST_HEAD_INITIALIZER(deadproc);
375 
376 /*
377  * We are called from cpu_exit() once it is safe to schedule the
378  * dead process's resources to be freed.
379  *
380  * NOTE: One must be careful with locking in this routine.  It's
381  * called from a critical section in machine-dependent code, so
382  * we should refrain from changing any interrupt state.
383  *
384  * We lock the deadproc list, place the proc on that list (using
385  * the p_hash member), and wake up the reaper.
386  */
387 void
388 exit2(struct proc *p)
389 {
390 	int s;
391 
392 	mtx_enter(&deadproc_mutex);
393 	LIST_INSERT_HEAD(&deadproc, p, p_hash);
394 	mtx_leave(&deadproc_mutex);
395 
396 	wakeup(&deadproc);
397 
398 	SCHED_LOCK(s);
399 }
400 
401 /*
402  * Process reaper.  This is run by a kernel thread to free the resources
403  * of a dead process.  Once the resources are free, the process becomes
404  * a zombie, and the parent is allowed to read the undead's status.
405  */
406 void
407 reaper(void)
408 {
409 	struct proc *p;
410 
411 	KERNEL_PROC_UNLOCK(curproc);
412 
413 	for (;;) {
414 		mtx_enter(&deadproc_mutex);
415 		p = LIST_FIRST(&deadproc);
416 		if (p == NULL) {
417 			/* No work for us; go to sleep until someone exits. */
418 			mtx_leave(&deadproc_mutex);
419 			(void) tsleep(&deadproc, PVM, "reaper", 0);
420 			continue;
421 		}
422 
423 		/* Remove us from the deadproc list. */
424 		LIST_REMOVE(p, p_hash);
425 		mtx_leave(&deadproc_mutex);
426 		KERNEL_PROC_LOCK(curproc);
427 
428 		/*
429 		 * Give machine-dependent code a chance to free any
430 		 * resources it couldn't free while still running on
431 		 * that process's context.  This must be done before
432 		 * uvm_exit(), in case these resources are in the PCB.
433 		 */
434 		cpu_wait(p);
435 
436 		/*
437 		 * Free the VM resources we're still holding on to.
438 		 * We must do this from a valid thread because doing
439 		 * so may block.
440 		 */
441 		uvm_exit(p);
442 
443 		/* Process is now a true zombie. */
444 		if ((p->p_flag & P_NOZOMBIE) == 0) {
445 			p->p_stat = SZOMB;
446 
447 			/* Wake up the parent so it can get exit status. */
448 			psignal(p->p_pptr, SIGCHLD);
449 			wakeup(p->p_pptr);
450 		} else {
451 			/* Noone will wait for us. Just zap the process now */
452 			proc_zap(p);
453 		}
454 
455 		KERNEL_PROC_UNLOCK(curproc);
456 	}
457 }
458 
459 pid_t
460 sys_wait4(struct proc *q, void *v, register_t *retval)
461 {
462 	struct sys_wait4_args /* {
463 		syscallarg(pid_t) pid;
464 		syscallarg(int *) status;
465 		syscallarg(int) options;
466 		syscallarg(struct rusage *) rusage;
467 	} */ *uap = v;
468 	int nfound;
469 	struct proc *p, *t;
470 	int status, error;
471 
472 	if (SCARG(uap, pid) == 0)
473 		SCARG(uap, pid) = -q->p_pgid;
474 	if (SCARG(uap, options) &~ (WUNTRACED|WNOHANG|WALTSIG|WCONTINUED))
475 		return (EINVAL);
476 
477 loop:
478 	nfound = 0;
479 	LIST_FOREACH(p, &q->p_children, p_sibling) {
480 		if ((p->p_flag & P_NOZOMBIE) ||
481 		    (SCARG(uap, pid) != WAIT_ANY &&
482 		    p->p_pid != SCARG(uap, pid) &&
483 		    p->p_pgid != -SCARG(uap, pid)))
484 			continue;
485 
486 		/*
487 		 * Wait for processes with p_exitsig != SIGCHLD processes only
488 		 * if WALTSIG is set; wait for processes with pexitsig ==
489 		 * SIGCHLD only if WALTSIG is clear.
490 		 */
491 		if ((SCARG(uap, options) & WALTSIG) ?
492 		    (p->p_exitsig == SIGCHLD) : (P_EXITSIG(p) != SIGCHLD))
493 			continue;
494 
495 		nfound++;
496 		if (p->p_stat == SZOMB) {
497 			retval[0] = p->p_pid;
498 
499 			if (SCARG(uap, status)) {
500 				status = p->p_xstat;	/* convert to int */
501 				error = copyout(&status,
502 				    SCARG(uap, status), sizeof(status));
503 				if (error)
504 					return (error);
505 			}
506 			if (SCARG(uap, rusage) &&
507 			    (error = copyout(p->p_ru,
508 			    SCARG(uap, rusage), sizeof(struct rusage))))
509 				return (error);
510 
511 			/*
512 			 * If we got the child via a ptrace 'attach',
513 			 * we need to give it back to the old parent.
514 			 */
515 			if (p->p_oppid && (t = pfind(p->p_oppid))) {
516 				p->p_oppid = 0;
517 				proc_reparent(p, t);
518 				if (p->p_exitsig != 0)
519 					psignal(t, P_EXITSIG(p));
520 				wakeup(t);
521 				return (0);
522 			}
523 
524 			scheduler_wait_hook(q, p);
525 			p->p_xstat = 0;
526 			ruadd(&q->p_stats->p_cru, p->p_ru);
527 
528 			proc_zap(p);
529 
530 			return (0);
531 		}
532 		if (p->p_stat == SSTOP && (p->p_flag & P_WAITED) == 0 &&
533 		    (p->p_flag & P_TRACED || SCARG(uap, options) & WUNTRACED)) {
534 			p->p_flag |= P_WAITED;
535 			retval[0] = p->p_pid;
536 
537 			if (SCARG(uap, status)) {
538 				status = W_STOPCODE(p->p_xstat);
539 				error = copyout(&status, SCARG(uap, status),
540 				    sizeof(status));
541 			} else
542 				error = 0;
543 			return (error);
544 		}
545 		if ((SCARG(uap, options) & WCONTINUED) && (p->p_flag & P_CONTINUED)) {
546 			p->p_flag &= ~P_CONTINUED;
547 			retval[0] = p->p_pid;
548 
549 			if (SCARG(uap, status)) {
550 				status = _WCONTINUED;
551 				error = copyout(&status, SCARG(uap, status),
552 				    sizeof(status));
553 			} else
554 				error = 0;
555 			return (error);
556 		}
557 	}
558 	if (nfound == 0)
559 		return (ECHILD);
560 	if (SCARG(uap, options) & WNOHANG) {
561 		retval[0] = 0;
562 		return (0);
563 	}
564 	if ((error = tsleep(q, PWAIT | PCATCH, "wait", 0)) != 0)
565 		return (error);
566 	goto loop;
567 }
568 
569 /*
570  * make process 'parent' the new parent of process 'child'.
571  */
572 void
573 proc_reparent(struct proc *child, struct proc *parent)
574 {
575 
576 	if (child->p_pptr == parent)
577 		return;
578 
579 	if (parent == initproc)
580 		child->p_exitsig = SIGCHLD;
581 
582 	LIST_REMOVE(child, p_sibling);
583 	LIST_INSERT_HEAD(&parent->p_children, child, p_sibling);
584 	child->p_pptr = parent;
585 }
586 
587 void
588 proc_zap(struct proc *p)
589 {
590 	pool_put(&rusage_pool, p->p_ru);
591 	if (p->p_ptstat)
592 		free(p->p_ptstat, M_SUBPROC);
593 
594 	/*
595 	 * Finally finished with old proc entry.
596 	 * Unlink it from its process group and free it.
597 	 */
598 	leavepgrp(p);
599 	LIST_REMOVE(p, p_list);	/* off zombproc */
600 	LIST_REMOVE(p, p_sibling);
601 
602 	/*
603 	 * Decrement the count of procs running with this uid.
604 	 */
605 	(void)chgproccnt(p->p_cred->p_ruid, -1);
606 
607 	/*
608 	 * Free up credentials.
609 	 */
610 	if (--p->p_cred->p_refcnt == 0) {
611 		crfree(p->p_cred->pc_ucred);
612 		pool_put(&pcred_pool, p->p_cred);
613 	}
614 
615 	/*
616 	 * Release reference to text vnode
617 	 */
618 	if (p->p_textvp)
619 		vrele(p->p_textvp);
620 
621 	pool_put(&proc_pool, p);
622 	nprocs--;
623 }
624 
625