xref: /openbsd-src/sys/kern/kern_exit.c (revision 9d1260f9afdb43a2c90d0fa4a67b72cbe2efdd87)
1 /*	$OpenBSD: kern_exit.c,v 1.50 2004/05/27 20:48:46 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 #ifdef SYSVSHM
63 #include <sys/shm.h>
64 #endif
65 #ifdef SYSVSEM
66 #include <sys/sem.h>
67 #endif
68 
69 #include "systrace.h"
70 #include <dev/systrace.h>
71 
72 #include <sys/mount.h>
73 #include <sys/syscallargs.h>
74 
75 #include <machine/cpu.h>
76 
77 #include <uvm/uvm_extern.h>
78 
79 /*
80  * exit --
81  *	Death of process.
82  */
83 int
84 sys_exit(p, v, retval)
85 	struct proc *p;
86 	void *v;
87 	register_t *retval;
88 {
89 	struct sys_exit_args /* {
90 		syscallarg(int) rval;
91 	} */ *uap = v;
92 
93 	exit1(p, W_EXITCODE(SCARG(uap, rval), 0));
94 	/* NOTREACHED */
95 	return (0);
96 }
97 
98 /*
99  * Exit: deallocate address space and other resources, change proc state
100  * to zombie, and unlink proc from allproc and parent's lists.  Save exit
101  * status and rusage for wait().  Check for child processes and orphan them.
102  */
103 void
104 exit1(p, rv)
105 	struct proc *p;
106 	int rv;
107 {
108 	struct proc *q, *nq;
109 
110 	if (p->p_pid == 1)
111 		panic("init died (signal %d, exit %d)",
112 		    WTERMSIG(rv), WEXITSTATUS(rv));
113 
114 	if (p->p_flag & P_PROFIL)
115 		stopprofclock(p);
116 	p->p_ru = pool_get(&rusage_pool, PR_WAITOK);
117 	/*
118 	 * If parent is waiting for us to exit or exec, P_PPWAIT is set; we
119 	 * wake up the parent early to avoid deadlock.
120 	 */
121 	p->p_flag |= P_WEXIT;
122 	p->p_flag &= ~P_TRACED;
123 	if (p->p_flag & P_PPWAIT) {
124 		p->p_flag &= ~P_PPWAIT;
125 		wakeup(p->p_pptr);
126 	}
127 	p->p_sigignore = ~0;
128 	p->p_siglist = 0;
129 	timeout_del(&p->p_realit_to);
130 
131 	/*
132 	 * Close open files and release open-file table.
133 	 * This may block!
134 	 */
135 	fdfree(p);
136 
137 #ifdef SYSVSEM
138 	semexit(p);
139 #endif
140 	if (SESS_LEADER(p)) {
141 		register struct session *sp = p->p_session;
142 
143 		if (sp->s_ttyvp) {
144 			/*
145 			 * Controlling process.
146 			 * Signal foreground pgrp,
147 			 * drain controlling terminal
148 			 * and revoke access to controlling terminal.
149 			 */
150 			if (sp->s_ttyp->t_session == sp) {
151 				if (sp->s_ttyp->t_pgrp)
152 					pgsignal(sp->s_ttyp->t_pgrp, SIGHUP, 1);
153 				(void) ttywait(sp->s_ttyp);
154 				/*
155 				 * The tty could have been revoked
156 				 * if we blocked.
157 				 */
158 				if (sp->s_ttyvp)
159 					VOP_REVOKE(sp->s_ttyvp, REVOKEALL);
160 			}
161 			if (sp->s_ttyvp)
162 				vrele(sp->s_ttyvp);
163 			sp->s_ttyvp = NULL;
164 			/*
165 			 * s_ttyp is not zero'd; we use this to indicate
166 			 * that the session once had a controlling terminal.
167 			 * (for logging and informational purposes)
168 			 */
169 		}
170 		sp->s_leader = NULL;
171 	}
172 	fixjobc(p, p->p_pgrp, 0);
173 #ifdef ACCOUNTING
174 	(void)acct_process(p);
175 #endif
176 #ifdef KTRACE
177 	/*
178 	 * release trace file
179 	 */
180 	p->p_traceflag = 0;	/* don't trace the vrele() */
181 	if (p->p_tracep)
182 		ktrsettracevnode(p, NULL);
183 #endif
184 #if NSYSTRACE > 0
185 	if (ISSET(p->p_flag, P_SYSTRACE))
186 		systrace_exit(p);
187 #endif
188 	/*
189 	 * NOTE: WE ARE NO LONGER ALLOWED TO SLEEP!
190 	 */
191 	p->p_stat = SDEAD;
192 
193         /*
194          * Remove proc from pidhash chain so looking it up won't
195          * work.  Move it from allproc to zombproc, but do not yet
196          * wake up the reaper.  We will put the proc on the
197          * deadproc list later (using the p_hash member), and
198          * wake up the reaper when we do.
199          */
200 	LIST_REMOVE(p, p_hash);
201 	LIST_REMOVE(p, p_list);
202 	LIST_INSERT_HEAD(&zombproc, p, p_list);
203 
204 	/*
205 	 * Give orphaned children to init(8).
206 	 */
207 	q = p->p_children.lh_first;
208 	if (q)		/* only need this if any child is S_ZOMB */
209 		wakeup(initproc);
210 	for (; q != 0; q = nq) {
211 		nq = q->p_sibling.le_next;
212 		proc_reparent(q, initproc);
213 		/*
214 		 * Traced processes are killed
215 		 * since their existence means someone is screwing up.
216 		 */
217 		if (q->p_flag & P_TRACED) {
218 			q->p_flag &= ~P_TRACED;
219 			psignal(q, SIGKILL);
220 		}
221 	}
222 
223 	/*
224 	 * Save exit status and final rusage info, adding in child rusage
225 	 * info and self times.
226 	 */
227 	p->p_xstat = rv;
228 	*p->p_ru = p->p_stats->p_ru;
229 	calcru(p, &p->p_ru->ru_utime, &p->p_ru->ru_stime, NULL);
230 	ruadd(p->p_ru, &p->p_stats->p_cru);
231 
232 	/*
233 	 * clear %cpu usage during swap
234 	 */
235 	p->p_pctcpu = 0;
236 
237 	/*
238 	 * notify interested parties of our demise.
239 	 */
240 	KNOTE(&p->p_klist, NOTE_EXIT);
241 
242 	/*
243 	 * Notify parent that we're gone.  If we have P_NOZOMBIE or parent has
244 	 * the P_NOCLDWAIT flag set, notify process 1 instead (and hope it
245 	 * will handle this situation).
246 	 */
247 	if ((p->p_flag & P_NOZOMBIE) || (p->p_pptr->p_flag & P_NOCLDWAIT)) {
248 		struct proc *pp = p->p_pptr;
249 		proc_reparent(p, initproc);
250 		/*
251 		 * If this was the last child of our parent, notify
252 		 * parent, so in case he was wait(2)ing, he will
253 		 * continue.
254 		 */
255 		if (pp->p_children.lh_first == NULL)
256 			wakeup(pp);
257 	}
258 
259 	if ((p->p_flag & P_FSTRACE) == 0 && p->p_exitsig != 0)
260 		psignal(p->p_pptr, P_EXITSIG(p));
261 	wakeup(p->p_pptr);
262 
263 	/*
264 	 * Notify procfs debugger
265 	 */
266 	if (p->p_flag & P_FSTRACE)
267 		wakeup(p);
268 
269 	/*
270 	 * Release the process's signal state.
271 	 */
272 	sigactsfree(p);
273 
274 	/*
275 	 * Clear curproc after we've done all operations
276 	 * that could block, and before tearing down the rest
277 	 * of the process state that might be used from clock, etc.
278 	 * Also, can't clear curproc while we're still runnable,
279 	 * as we're not on a run queue (we are current, just not
280 	 * a proper proc any longer!).
281 	 *
282 	 * Other substructures are freed from wait().
283 	 */
284 	curproc = NULL;
285 	limfree(p->p_limit);
286 	p->p_limit = NULL;
287 
288 	/*
289 	 * If emulation has process exit hook, call it now.
290 	 */
291 	if (p->p_emul->e_proc_exit)
292 		(*p->p_emul->e_proc_exit)(p);
293 
294 	/*
295 	 * Finally, call machine-dependent code to switch to a new
296 	 * context (possibly the idle context).  Once we are no longer
297 	 * using the dead process's vmspace and stack, exit2() will be
298 	 * called to schedule those resources to be released by the
299 	 * reaper thread.
300 	 *
301 	 * Note that cpu_exit() will end with a call equivalent to
302 	 * cpu_switch(), finishing our execution (pun intended).
303 	 */
304 	cpu_exit(p);
305 }
306 
307 /*
308  * We are called from cpu_exit() once it is safe to schedule the
309  * dead process's resources to be freed.
310  *
311  * NOTE: One must be careful with locking in this routine.  It's
312  * called from a critical section in machine-dependent code, so
313  * we should refrain from changing any interrupt state.
314  *
315  * We lock the deadproc list (a spin lock), place the proc on that
316  * list (using the p_hash member), and wake up the reaper.
317  */
318 void
319 exit2(p)
320 	struct proc *p;
321 {
322 
323 	simple_lock(&deadproc_slock);
324 	LIST_INSERT_HEAD(&deadproc, p, p_hash);
325 	simple_unlock(&deadproc_slock);
326 
327 	wakeup(&deadproc);
328 }
329 
330 /*
331  * Process reaper.  This is run by a kernel thread to free the resources
332  * of a dead process.  Once the resources are free, the process becomes
333  * a zombie, and the parent is allowed to read the undead's status.
334  */
335 void
336 reaper(void)
337 {
338 	struct proc *p;
339 
340 	for (;;) {
341 		simple_lock(&deadproc_slock);
342 		p = LIST_FIRST(&deadproc);
343 		if (p == NULL) {
344 			/* No work for us; go to sleep until someone exits. */
345 			simple_unlock(&deadproc_slock);
346 			(void) tsleep(&deadproc, PVM, "reaper", 0);
347 			continue;
348 		}
349 
350 		/* Remove us from the deadproc list. */
351 		LIST_REMOVE(p, p_hash);
352 		simple_unlock(&deadproc_slock);
353 
354 		/*
355 		 * Give machine-dependent code a chance to free any
356 		 * resources it couldn't free while still running on
357 		 * that process's context.  This must be done before
358 		 * uvm_exit(), in case these resources are in the PCB.
359 		 */
360 		cpu_wait(p);
361 
362 		/*
363 		 * Free the VM resources we're still holding on to.
364 		 * We must do this from a valid thread because doing
365 		 * so may block.
366 		 */
367 		uvm_exit(p);
368 
369 		/* Process is now a true zombie. */
370 		if ((p->p_flag & P_NOZOMBIE) == 0) {
371 			p->p_stat = SZOMB;
372 
373 			/* Wake up the parent so it can get exit status. */
374 			psignal(p->p_pptr, SIGCHLD);
375 			wakeup(p->p_pptr);
376 		} else {
377 			/* Noone will wait for us. Just zap the process now */
378 			proc_zap(p);
379 		}
380 	}
381 }
382 
383 pid_t
384 sys_wait4(q, v, retval)
385 	register struct proc *q;
386 	void *v;
387 	register_t *retval;
388 {
389 	register struct sys_wait4_args /* {
390 		syscallarg(pid_t) pid;
391 		syscallarg(int *) status;
392 		syscallarg(int) options;
393 		syscallarg(struct rusage *) rusage;
394 	} */ *uap = v;
395 	register int nfound;
396 	register struct proc *p, *t;
397 	int status, error;
398 
399 	if (SCARG(uap, pid) == 0)
400 		SCARG(uap, pid) = -q->p_pgid;
401 	if (SCARG(uap, options) &~ (WUNTRACED|WNOHANG|WALTSIG|WCONTINUED))
402 		return (EINVAL);
403 
404 loop:
405 	nfound = 0;
406 	for (p = q->p_children.lh_first; p != 0; p = p->p_sibling.le_next) {
407 		if ((p->p_flag & P_NOZOMBIE) ||
408 		    (SCARG(uap, pid) != WAIT_ANY &&
409 		    p->p_pid != SCARG(uap, pid) &&
410 		    p->p_pgid != -SCARG(uap, pid)))
411 			continue;
412 
413 		/*
414 		 * Wait for processes with p_exitsig != SIGCHLD processes only
415 		 * if WALTSIG is set; wait for processes with pexitsig ==
416 		 * SIGCHLD only if WALTSIG is clear.
417 		 */
418 		if ((SCARG(uap, options) & WALTSIG) ?
419 		    (p->p_exitsig == SIGCHLD) : (P_EXITSIG(p) != SIGCHLD))
420 			continue;
421 
422 		nfound++;
423 		if (p->p_stat == SZOMB) {
424 			retval[0] = p->p_pid;
425 
426 			if (SCARG(uap, status)) {
427 				status = p->p_xstat;	/* convert to int */
428 				error = copyout(&status,
429 				    SCARG(uap, status), sizeof(status));
430 				if (error)
431 					return (error);
432 			}
433 			if (SCARG(uap, rusage) &&
434 			    (error = copyout(p->p_ru,
435 			    SCARG(uap, rusage), sizeof(struct rusage))))
436 				return (error);
437 
438 			/*
439 			 * If we got the child via a ptrace 'attach',
440 			 * we need to give it back to the old parent.
441 			 */
442 			if (p->p_oppid && (t = pfind(p->p_oppid))) {
443 				p->p_oppid = 0;
444 				proc_reparent(p, t);
445 				if (p->p_exitsig != 0)
446 					psignal(t, P_EXITSIG(p));
447 				wakeup(t);
448 				return (0);
449 			}
450 
451 			scheduler_wait_hook(q, p);
452 			p->p_xstat = 0;
453 			ruadd(&q->p_stats->p_cru, p->p_ru);
454 
455 			proc_zap(p);
456 
457 			return (0);
458 		}
459 		if (p->p_stat == SSTOP && (p->p_flag & P_WAITED) == 0 &&
460 		    (p->p_flag & P_TRACED || SCARG(uap, options) & WUNTRACED)) {
461 			p->p_flag |= P_WAITED;
462 			retval[0] = p->p_pid;
463 
464 			if (SCARG(uap, status)) {
465 				status = W_STOPCODE(p->p_xstat);
466 				error = copyout(&status, SCARG(uap, status),
467 				    sizeof(status));
468 			} else
469 				error = 0;
470 			return (error);
471 		}
472 		if ((SCARG(uap, options) & WCONTINUED) && (p->p_flag & P_CONTINUED)) {
473 			p->p_flag &= ~P_CONTINUED;
474 			retval[0] = p->p_pid;
475 
476 			if (SCARG(uap, status)) {
477 				status = _WCONTINUED;
478 				error = copyout(&status, SCARG(uap, status),
479 				    sizeof(status));
480 			} else
481 				error = 0;
482 			return (error);
483 		}
484 	}
485 	if (nfound == 0)
486 		return (ECHILD);
487 	if (SCARG(uap, options) & WNOHANG) {
488 		retval[0] = 0;
489 		return (0);
490 	}
491 	if ((error = tsleep(q, PWAIT | PCATCH, "wait", 0)) != 0)
492 		return (error);
493 	goto loop;
494 }
495 
496 /*
497  * make process 'parent' the new parent of process 'child'.
498  */
499 void
500 proc_reparent(child, parent)
501 	register struct proc *child;
502 	register struct proc *parent;
503 {
504 
505 	if (child->p_pptr == parent)
506 		return;
507 
508 	if (parent == initproc)
509 		child->p_exitsig = SIGCHLD;
510 
511 	LIST_REMOVE(child, p_sibling);
512 	LIST_INSERT_HEAD(&parent->p_children, child, p_sibling);
513 	child->p_pptr = parent;
514 }
515 
516 void
517 proc_zap(struct proc *p)
518 {
519 	pool_put(&rusage_pool, p->p_ru);
520 
521 	/*
522 	 * Finally finished with old proc entry.
523 	 * Unlink it from its process group and free it.
524 	 */
525 	leavepgrp(p);
526 	LIST_REMOVE(p, p_list);	/* off zombproc */
527 	LIST_REMOVE(p, p_sibling);
528 
529 	/*
530 	 * Decrement the count of procs running with this uid.
531 	 */
532 	(void)chgproccnt(p->p_cred->p_ruid, -1);
533 
534 	/*
535 	 * Free up credentials.
536 	 */
537 	if (--p->p_cred->p_refcnt == 0) {
538 		crfree(p->p_cred->pc_ucred);
539 		pool_put(&pcred_pool, p->p_cred);
540 	}
541 
542 	/*
543 	 * Release reference to text vnode
544 	 */
545 	if (p->p_textvp)
546 		vrele(p->p_textvp);
547 
548 	pool_put(&proc_pool, p);
549 	nprocs--;
550 }
551 
552