xref: /openbsd-src/sys/kern/kern_exit.c (revision 8445c53715e7030056b779e8ab40efb7820981f2)
1 /*	$OpenBSD: kern_exit.c,v 1.35 2001/09/11 20:05:25 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. All advertising materials mentioning features or use of this software
22  *    must display the following acknowledgement:
23  *	This product includes software developed by the University of
24  *	California, Berkeley and its contributors.
25  * 4. Neither the name of the University nor the names of its contributors
26  *    may be used to endorse or promote products derived from this software
27  *    without specific prior written permission.
28  *
29  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
30  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
31  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
32  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
33  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
34  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
35  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
36  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
37  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
38  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
39  * SUCH DAMAGE.
40  *
41  *	@(#)kern_exit.c	8.7 (Berkeley) 2/12/94
42  */
43 
44 #include <sys/param.h>
45 #include <sys/systm.h>
46 #include <sys/map.h>
47 #include <sys/ioctl.h>
48 #include <sys/proc.h>
49 #include <sys/tty.h>
50 #include <sys/time.h>
51 #include <sys/resource.h>
52 #include <sys/kernel.h>
53 #include <sys/proc.h>
54 #include <sys/buf.h>
55 #include <sys/wait.h>
56 #include <sys/file.h>
57 #include <sys/vnode.h>
58 #include <sys/syslog.h>
59 #include <sys/malloc.h>
60 #include <sys/resourcevar.h>
61 #include <sys/ptrace.h>
62 #include <sys/acct.h>
63 #include <sys/filedesc.h>
64 #include <sys/signalvar.h>
65 #include <sys/sched.h>
66 #include <sys/ktrace.h>
67 #include <sys/pool.h>
68 #ifdef SYSVSHM
69 #include <sys/shm.h>
70 #endif
71 #ifdef SYSVSEM
72 #include <sys/sem.h>
73 #endif
74 
75 #include <sys/mount.h>
76 #include <sys/syscallargs.h>
77 
78 #include <machine/cpu.h>
79 
80 #include <vm/vm.h>
81 
82 #include <uvm/uvm_extern.h>
83 
84 void proc_zap __P((struct proc *));
85 
86 /*
87  * exit --
88  *	Death of process.
89  */
90 int
91 sys_exit(p, v, retval)
92 	struct proc *p;
93 	void *v;
94 	register_t *retval;
95 {
96 	struct sys_exit_args /* {
97 		syscallarg(int) rval;
98 	} */ *uap = v;
99 
100 	exit1(p, W_EXITCODE(SCARG(uap, rval), 0));
101 	/* NOTREACHED */
102 	return (0);
103 }
104 
105 /*
106  * Exit: deallocate address space and other resources, change proc state
107  * to zombie, and unlink proc from allproc and parent's lists.  Save exit
108  * status and rusage for wait().  Check for child processes and orphan them.
109  */
110 void
111 exit1(p, rv)
112 	struct proc *p;
113 	int rv;
114 {
115 	struct proc *q, *nq;
116 	struct vmspace *vm;
117 
118 	if (p->p_pid == 1)
119 		panic("init died (signal %d, exit %d)",
120 		    WTERMSIG(rv), WEXITSTATUS(rv));
121 
122 	if (p->p_flag & P_PROFIL)
123 		stopprofclock(p);
124 	MALLOC(p->p_ru, struct rusage *, sizeof(struct rusage),
125 		M_ZOMBIE, M_WAITOK);
126 	/*
127 	 * If parent is waiting for us to exit or exec, P_PPWAIT is set; we
128 	 * wake up the parent early to avoid deadlock.
129 	 */
130 	p->p_flag |= P_WEXIT;
131 	p->p_flag &= ~P_TRACED;
132 	if (p->p_flag & P_PPWAIT) {
133 		p->p_flag &= ~P_PPWAIT;
134 		wakeup((caddr_t)p->p_pptr);
135 	}
136 	p->p_sigignore = ~0;
137 	p->p_siglist = 0;
138 	timeout_del(&p->p_realit_to);
139 
140 	/*
141 	 * Close open files and release open-file table.
142 	 * This may block!
143 	 */
144 	fdfree(p);
145 
146 	/* The next three chunks should probably be moved to vmspace_exit. */
147 	vm = p->p_vmspace;
148 #ifdef SYSVSHM
149 	if (vm->vm_shm && vm->vm_refcnt == 1)
150 		shmexit(vm);
151 #endif
152 #ifdef SYSVSEM
153 	semexit(p);
154 #endif
155 	/*
156 	 * Release user portion of address space.
157 	 * This releases references to vnodes,
158 	 * which could cause I/O if the file has been unlinked.
159 	 * Need to do this early enough that we can still sleep.
160 	 * Can't free the entire vmspace as the kernel stack
161 	 * may be mapped within that space also.
162 	 */
163 	if (vm->vm_refcnt == 1)
164 		(void) uvm_deallocate(&vm->vm_map, VM_MIN_ADDRESS,
165 		    VM_MAXUSER_ADDRESS - VM_MIN_ADDRESS);
166 
167 	if (SESS_LEADER(p)) {
168 		register struct session *sp = p->p_session;
169 
170 		if (sp->s_ttyvp) {
171 			/*
172 			 * Controlling process.
173 			 * Signal foreground pgrp,
174 			 * drain controlling terminal
175 			 * and revoke access to controlling terminal.
176 			 */
177 			if (sp->s_ttyp->t_session == sp) {
178 				if (sp->s_ttyp->t_pgrp)
179 					pgsignal(sp->s_ttyp->t_pgrp, SIGHUP, 1);
180 				(void) ttywait(sp->s_ttyp);
181 				/*
182 				 * The tty could have been revoked
183 				 * if we blocked.
184 				 */
185 				if (sp->s_ttyvp)
186 					VOP_REVOKE(sp->s_ttyvp, REVOKEALL);
187 			}
188 			if (sp->s_ttyvp)
189 				vrele(sp->s_ttyvp);
190 			sp->s_ttyvp = NULL;
191 			/*
192 			 * s_ttyp is not zero'd; we use this to indicate
193 			 * that the session once had a controlling terminal.
194 			 * (for logging and informational purposes)
195 			 */
196 		}
197 		sp->s_leader = NULL;
198 	}
199 	fixjobc(p, p->p_pgrp, 0);
200 	(void)acct_process(p);
201 #ifdef KTRACE
202 	/*
203 	 * release trace file
204 	 */
205 	p->p_traceflag = 0;	/* don't trace the vrele() */
206 	if (p->p_tracep)
207 		ktrsettracevnode(p, NULL);
208 #endif
209 	/*
210 	 * NOTE: WE ARE NO LONGER ALLOWED TO SLEEP!
211 	 */
212 	p->p_stat = SDEAD;
213 
214         /*
215          * Remove proc from pidhash chain so looking it up won't
216          * work.  Move it from allproc to zombproc, but do not yet
217          * wake up the reaper.  We will put the proc on the
218          * deadproc list later (using the p_hash member), and
219          * wake up the reaper when we do.
220          */
221 	LIST_REMOVE(p, p_hash);
222 	LIST_REMOVE(p, p_list);
223 	LIST_INSERT_HEAD(&zombproc, p, p_list);
224 
225 	/*
226 	 * Give orphaned children to init(8).
227 	 */
228 	q = p->p_children.lh_first;
229 	if (q)		/* only need this if any child is S_ZOMB */
230 		wakeup((caddr_t)initproc);
231 	for (; q != 0; q = nq) {
232 		nq = q->p_sibling.le_next;
233 		proc_reparent(q, initproc);
234 		/*
235 		 * Traced processes are killed
236 		 * since their existence means someone is screwing up.
237 		 */
238 		if (q->p_flag & P_TRACED) {
239 			q->p_flag &= ~P_TRACED;
240 			psignal(q, SIGKILL);
241 		}
242 	}
243 
244 	/*
245 	 * Save exit status and final rusage info, adding in child rusage
246 	 * info and self times.
247 	 */
248 	p->p_xstat = rv;
249 	*p->p_ru = p->p_stats->p_ru;
250 	calcru(p, &p->p_ru->ru_utime, &p->p_ru->ru_stime, NULL);
251 	ruadd(p->p_ru, &p->p_stats->p_cru);
252 
253 	/*
254 	 * clear %cpu usage during swap
255 	 */
256 	p->p_pctcpu = 0;
257 
258 	/*
259 	 * notify interested parties of our demise.
260 	 */
261 	KNOTE(&p->p_klist, NOTE_EXIT);
262 
263 	/*
264 	 * Notify parent that we're gone.  If we have P_NOZOMBIE or parent has
265 	 * the P_NOCLDWAIT flag set, notify process 1 instead (and hope it
266 	 * will handle this situation).
267 	 */
268 	if ((p->p_flag & P_NOZOMBIE) || (p->p_pptr->p_flag & P_NOCLDWAIT)) {
269 		struct proc *pp = p->p_pptr;
270 		proc_reparent(p, initproc);
271 		/*
272 		 * If this was the last child of our parent, notify
273 		 * parent, so in case he was wait(2)ing, he will
274 		 * continue.
275 		 */
276 		if (pp->p_children.lh_first == NULL)
277 			wakeup((caddr_t)pp);
278 	}
279 
280 	if ((p->p_flag & P_FSTRACE) == 0 && p->p_exitsig != 0)
281 		psignal(p->p_pptr, P_EXITSIG(p));
282 	wakeup((caddr_t)p->p_pptr);
283 
284 	/*
285 	 * Notify procfs debugger
286 	 */
287 	if (p->p_flag & P_FSTRACE)
288 		wakeup((caddr_t)p);
289 
290 	/*
291 	 * Release the process's signal state.
292 	 */
293 	sigactsfree(p);
294 
295 	/*
296 	 * Clear curproc after we've done all operations
297 	 * that could block, and before tearing down the rest
298 	 * of the process state that might be used from clock, etc.
299 	 * Also, can't clear curproc while we're still runnable,
300 	 * as we're not on a run queue (we are current, just not
301 	 * a proper proc any longer!).
302 	 *
303 	 * Other substructures are freed from wait().
304 	 */
305 	curproc = NULL;
306 	limfree(p->p_limit);
307 	p->p_limit = NULL;
308 
309 	/*
310 	 * Finally, call machine-dependent code to switch to a new
311 	 * context (possibly the idle context).  Once we are no longer
312 	 * using the dead process's vmspace and stack, exit2() will be
313 	 * called to schedule those resources to be released by the
314 	 * reaper thread.
315 	 *
316 	 * Note that cpu_exit() will end with a call equivalent to
317 	 * cpu_switch(), finishing our execution (pun intended).
318 	 */
319 	cpu_exit(p);
320 }
321 
322 /*
323  * We are called from cpu_exit() once it is safe to schedule the
324  * dead process's resources to be freed.
325  *
326  * NOTE: One must be careful with locking in this routine.  It's
327  * called from a critical section in machine-dependent code, so
328  * we should refrain from changing any interrupt state.
329  *
330  * We lock the deadproc list (a spin lock), place the proc on that
331  * list (using the p_hash member), and wake up the reaper.
332  */
333 void
334 exit2(p)
335 	struct proc *p;
336 {
337 
338 	simple_lock(&deadproc_slock);
339 	LIST_INSERT_HEAD(&deadproc, p, p_hash);
340 	simple_unlock(&deadproc_slock);
341 
342 	wakeup(&deadproc);
343 }
344 
345 /*
346  * Process reaper.  This is run by a kernel thread to free the resources
347  * of a dead process.  Once the resources are free, the process becomes
348  * a zombie, and the parent is allowed to read the undead's status.
349  */
350 void
351 reaper()
352 {
353 	struct proc *p;
354 
355 	for (;;) {
356 		simple_lock(&deadproc_slock);
357 		p = LIST_FIRST(&deadproc);
358 		if (p == NULL) {
359 			/* No work for us; go to sleep until someone exits. */
360 			simple_unlock(&deadproc_slock);
361 			(void) tsleep(&deadproc, PVM, "reaper", 0);
362 			continue;
363 		}
364 
365 		/* Remove us from the deadproc list. */
366 		LIST_REMOVE(p, p_hash);
367 		simple_unlock(&deadproc_slock);
368 
369 		/*
370 		 * Give machine-dependent code a chance to free any
371 		 * resources it couldn't free while still running on
372 		 * that process's context.  This must be done before
373 		 * uvm_exit(), in case these resources are in the PCB.
374 		 */
375 		cpu_wait(p);
376 
377 		/*
378 		 * Free the VM resources we're still holding on to.
379 		 * We must do this from a valid thread because doing
380 		 * so may block.
381 		 */
382 		uvm_exit(p);
383 
384 		/* Process is now a true zombie. */
385 		if ((p->p_flag & P_NOZOMBIE) == 0) {
386 			p->p_stat = SZOMB;
387 
388 			/* Wake up the parent so it can get exit status. */
389 			psignal(p->p_pptr, SIGCHLD);
390 			wakeup((caddr_t)p->p_pptr);
391 		} else {
392 			/* Noone will wait for us. Just zap the process now */
393 			proc_zap(p);
394 		}
395 	}
396 }
397 
398 int
399 sys_wait4(q, v, retval)
400 	register struct proc *q;
401 	void *v;
402 	register_t *retval;
403 {
404 	register struct sys_wait4_args /* {
405 		syscallarg(int) pid;
406 		syscallarg(int *) status;
407 		syscallarg(int) options;
408 		syscallarg(struct rusage *) rusage;
409 	} */ *uap = v;
410 	register int nfound;
411 	register struct proc *p, *t;
412 	int status, error;
413 
414 	if (SCARG(uap, pid) == 0)
415 		SCARG(uap, pid) = -q->p_pgid;
416 	if (SCARG(uap, options) &~ (WUNTRACED|WNOHANG|WALTSIG))
417 		return (EINVAL);
418 
419 loop:
420 	nfound = 0;
421 	for (p = q->p_children.lh_first; p != 0; p = p->p_sibling.le_next) {
422 		if ((p->p_flag & P_NOZOMBIE) ||
423 		    (SCARG(uap, pid) != WAIT_ANY &&
424 		    p->p_pid != SCARG(uap, pid) &&
425 		    p->p_pgid != -SCARG(uap, pid)))
426 			continue;
427 
428 		/*
429 		 * Wait for processes with p_exitsig != SIGCHLD processes only
430 		 * if WALTSIG is set; wait for processes with pexitsig ==
431 		 * SIGCHLD only if WALTSIG is clear.
432 		 */
433 		if ((SCARG(uap, options) & WALTSIG) ?
434 		    (p->p_exitsig == SIGCHLD) : (P_EXITSIG(p) != SIGCHLD))
435 			continue;
436 
437 		nfound++;
438 		if (p->p_stat == SZOMB) {
439 			retval[0] = p->p_pid;
440 
441 			if (SCARG(uap, status)) {
442 				status = p->p_xstat;	/* convert to int */
443 				error = copyout((caddr_t)&status,
444 						(caddr_t)SCARG(uap, status),
445 						sizeof(status));
446 				if (error)
447 					return (error);
448 			}
449 			if (SCARG(uap, rusage) &&
450 			    (error = copyout((caddr_t)p->p_ru,
451 			    (caddr_t)SCARG(uap, rusage),
452 			    sizeof(struct rusage))))
453 				return (error);
454 
455 			/*
456 			 * If we got the child via a ptrace 'attach',
457 			 * we need to give it back to the old parent.
458 			 */
459 			if (p->p_oppid && (t = pfind(p->p_oppid))) {
460 				p->p_oppid = 0;
461 				proc_reparent(p, t);
462 				if (p->p_exitsig != 0)
463 					psignal(t, P_EXITSIG(p));
464 				wakeup((caddr_t)t);
465 				return (0);
466 			}
467 
468 			scheduler_wait_hook(q, p);
469 			p->p_xstat = 0;
470 			ruadd(&q->p_stats->p_cru, p->p_ru);
471 
472 			proc_zap(p);
473 
474 			return (0);
475 		}
476 		if (p->p_stat == SSTOP && (p->p_flag & P_WAITED) == 0 &&
477 		    (p->p_flag & P_TRACED || SCARG(uap, options) & WUNTRACED)) {
478 			p->p_flag |= P_WAITED;
479 			retval[0] = p->p_pid;
480 
481 			if (SCARG(uap, status)) {
482 				status = W_STOPCODE(p->p_xstat);
483 				error = copyout((caddr_t)&status,
484 				    (caddr_t)SCARG(uap, status),
485 				    sizeof(status));
486 			} else
487 				error = 0;
488 			return (error);
489 		}
490 	}
491 	if (nfound == 0)
492 		return (ECHILD);
493 	if (SCARG(uap, options) & WNOHANG) {
494 		retval[0] = 0;
495 		return (0);
496 	}
497 	if ((error = tsleep((caddr_t)q, PWAIT | PCATCH, "wait", 0)) != 0)
498 		return (error);
499 	goto loop;
500 }
501 
502 /*
503  * make process 'parent' the new parent of process 'child'.
504  */
505 void
506 proc_reparent(child, parent)
507 	register struct proc *child;
508 	register struct proc *parent;
509 {
510 
511 	if (child->p_pptr == parent)
512 		return;
513 
514 	if (parent == initproc)
515 		child->p_exitsig = SIGCHLD;
516 
517 	LIST_REMOVE(child, p_sibling);
518 	LIST_INSERT_HEAD(&parent->p_children, child, p_sibling);
519 	child->p_pptr = parent;
520 }
521 
522 void
523 proc_zap(p)
524 	struct proc *p;
525 {
526 
527 	FREE(p->p_ru, M_ZOMBIE);
528 
529 	/*
530 	 * Finally finished with old proc entry.
531 	 * Unlink it from its process group and free it.
532 	 */
533 	leavepgrp(p);
534 	LIST_REMOVE(p, p_list);	/* off zombproc */
535 	LIST_REMOVE(p, p_sibling);
536 
537 	/*
538 	 * Decrement the count of procs running with this uid.
539 	 */
540 	(void)chgproccnt(p->p_cred->p_ruid, -1);
541 
542 	/*
543 	 * Free up credentials.
544 	 */
545 	if (--p->p_cred->p_refcnt == 0) {
546 		crfree(p->p_cred->pc_ucred);
547 		FREE(p->p_cred, M_SUBPROC);
548 	}
549 
550 	/*
551 	 * Release reference to text vnode
552 	 */
553 	if (p->p_textvp)
554 		vrele(p->p_textvp);
555 
556 	pool_put(&proc_pool, p);
557 	nprocs--;
558 }
559 
560