xref: /openbsd-src/sys/kern/kern_exit.c (revision 80b87c503b3a5eaa2c4345d138a3da33607b5c6d)
1 /*	$OpenBSD: kern_exit.c,v 1.100 2011/04/18 21:44:56 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 int
94 sys_threxit(struct proc *p, void *v, register_t *retval)
95 {
96 	struct sys_threxit_args /* {
97 		syscallarg(pid_t *) notdead;
98 	} */ *uap = v;
99 
100 	if (!rthreads_enabled)
101 		return (EINVAL);
102 
103 	if (SCARG(uap, notdead) != NULL) {
104 		pid_t zero = 0;
105 		if (copyout(&zero, SCARG(uap, notdead), sizeof(zero))) {
106 			psignal(p, SIGSEGV);
107 		}
108 	}
109 	exit1(p, 0, EXIT_THREAD);
110 
111 	return (0);
112 }
113 
114 /*
115  * Exit: deallocate address space and other resources, change proc state
116  * to zombie, and unlink proc from allproc and parent's lists.  Save exit
117  * status and rusage for wait().  Check for child processes and orphan them.
118  */
119 void
120 exit1(struct proc *p, int rv, int flags)
121 {
122 	struct proc *q, *nq;
123 	struct process *pr, *qr, *nqr;
124 
125 	if (p->p_pid == 1)
126 		panic("init died (signal %d, exit %d)",
127 		    WTERMSIG(rv), WEXITSTATUS(rv));
128 
129 	atomic_setbits_int(&p->p_flag, P_WEXIT);
130 
131 	/* unlink ourselves from the active threads */
132 	pr = p->p_p;
133 	TAILQ_REMOVE(&pr->ps_threads, p, p_thr_link);
134 	if (TAILQ_EMPTY(&pr->ps_threads))
135 		wakeup(&pr->ps_threads);
136 	/*
137 	 * if one thread calls exit, we take down everybody.
138 	 * we have to be careful not to get recursively caught.
139 	 * this is kinda sick.
140 	 */
141 	if (flags == EXIT_NORMAL && (p->p_flag & P_THREAD) &&
142 	    (pr->ps_mainproc->p_flag & P_WEXIT) == 0) {
143 		/*
144 		 * we are one of the threads.  we SIGKILL the parent,
145 		 * it will wake us up again, then we proceed.
146 		 */
147 		atomic_setbits_int(&pr->ps_mainproc->p_flag, P_IGNEXITRV);
148 		pr->ps_mainproc->p_xstat = rv;
149 		ptsignal(pr->ps_mainproc, SIGKILL, SPROPAGATED);
150 		tsleep(pr, PUSER, "thrdying", 0);
151 	} else if ((p->p_flag & P_THREAD) == 0) {
152 		if (flags == EXIT_NORMAL) {
153 			q = TAILQ_FIRST(&pr->ps_threads);
154 			for (; q != NULL; q = nq) {
155 				nq = TAILQ_NEXT(q, p_thr_link);
156 				atomic_setbits_int(&q->p_flag, P_IGNEXITRV);
157 				q->p_xstat = rv;
158 				ptsignal(q, SIGKILL, SPROPAGATED);
159 			}
160 		}
161 		wakeup(pr);
162 		while (!TAILQ_EMPTY(&pr->ps_threads))
163 			tsleep(&pr->ps_threads, PUSER, "thrdeath", 0);
164 		/*
165 		 * If parent is waiting for us to exit or exec, PS_PPWAIT
166 		 * is set; we wake up the parent early to avoid deadlock.
167 		 */
168 		if (pr->ps_flags & PS_PPWAIT) {
169 			atomic_clearbits_int(&pr->ps_flags, PS_PPWAIT);
170 			atomic_clearbits_int(&pr->ps_pptr->ps_flags,
171 			    PS_ISPWAIT);
172 			wakeup(pr->ps_pptr);
173 		}
174 	}
175 
176 	if (p->p_flag & P_PROFIL)
177 		stopprofclock(p);
178 	p->p_ru = pool_get(&rusage_pool, PR_WAITOK);
179 	p->p_sigignore = ~0;
180 	p->p_siglist = 0;
181 	timeout_del(&p->p_realit_to);
182 	timeout_del(&p->p_stats->p_virt_to);
183 	timeout_del(&p->p_stats->p_prof_to);
184 
185 	/*
186 	 * Close open files and release open-file table.
187 	 */
188 	fdfree(p);
189 
190 	if ((p->p_flag & P_THREAD) == 0) {
191 #ifdef SYSVSEM
192 		semexit(pr);
193 #endif
194 		if (SESS_LEADER(pr)) {
195 			struct session *sp = pr->ps_session;
196 
197 			if (sp->s_ttyvp) {
198 				/*
199 				 * Controlling process.
200 				 * Signal foreground pgrp,
201 				 * drain controlling terminal
202 				 * and revoke access to controlling terminal.
203 				 */
204 				if (sp->s_ttyp->t_session == sp) {
205 					if (sp->s_ttyp->t_pgrp)
206 						pgsignal(sp->s_ttyp->t_pgrp,
207 						    SIGHUP, 1);
208 					(void) ttywait(sp->s_ttyp);
209 					/*
210 					 * The tty could have been revoked
211 					 * if we blocked.
212 					 */
213 					if (sp->s_ttyvp)
214 						VOP_REVOKE(sp->s_ttyvp,
215 						    REVOKEALL);
216 				}
217 				if (sp->s_ttyvp)
218 					vrele(sp->s_ttyvp);
219 				sp->s_ttyvp = NULL;
220 				/*
221 				 * s_ttyp is not zero'd; we use this to
222 				 * indicate that the session once had a
223 				 * controlling terminal.  (for logging and
224 				 * informational purposes)
225 				 */
226 			}
227 			sp->s_leader = NULL;
228 		}
229 		fixjobc(pr, pr->ps_pgrp, 0);
230 
231 #ifdef ACCOUNTING
232 		(void)acct_process(p);
233 #endif
234 	}
235 
236 #ifdef KTRACE
237 	/*
238 	 * release trace file
239 	 */
240 	p->p_traceflag = 0;	/* don't trace the vrele() */
241 	if (p->p_tracep)
242 		ktrsettracevnode(p, NULL);
243 #endif
244 #if NSYSTRACE > 0
245 	if (ISSET(p->p_flag, P_SYSTRACE))
246 		systrace_exit(p);
247 #endif
248         /*
249          * Remove proc from pidhash chain so looking it up won't
250          * work.  Move it from allproc to zombproc, but do not yet
251          * wake up the reaper.  We will put the proc on the
252          * deadproc list later (using the p_hash member), and
253          * wake up the reaper when we do.
254          */
255 	/*
256 	 * NOTE: WE ARE NO LONGER ALLOWED TO SLEEP!
257 	 */
258 	p->p_stat = SDEAD;
259 
260 	LIST_REMOVE(p, p_hash);
261 	LIST_REMOVE(p, p_list);
262 	LIST_INSERT_HEAD(&zombproc, p, p_list);
263 
264 	/*
265 	 * Give orphaned children to init(8).
266 	 */
267 	if ((p->p_flag & P_THREAD) == 0) {
268 		qr = LIST_FIRST(&pr->ps_children);
269 		if (qr)		/* only need this if any child is S_ZOMB */
270 			wakeup(initproc->p_p);
271 		for (; qr != 0; qr = nqr) {
272 			nqr = LIST_NEXT(qr, ps_sibling);
273 			proc_reparent(qr, initproc->p_p);
274 			/*
275 			 * Traced processes are killed
276 			 * since their existence means someone is screwing up.
277 			 */
278 			if (qr->ps_mainproc->p_flag & P_TRACED) {
279 				atomic_clearbits_int(&qr->ps_mainproc->p_flag,
280 				    P_TRACED);
281 				prsignal(qr, SIGKILL);
282 			}
283 		}
284 	}
285 
286 
287 	/*
288 	 * Save exit status and final rusage info, adding in child rusage
289 	 * info and self times.
290 	 */
291 	if (!(p->p_flag & P_IGNEXITRV))
292 		p->p_xstat = rv;
293 	*p->p_ru = p->p_stats->p_ru;
294 	calcru(p, &p->p_ru->ru_utime, &p->p_ru->ru_stime, NULL);
295 	ruadd(p->p_ru, &p->p_stats->p_cru);
296 
297 	/*
298 	 * clear %cpu usage during swap
299 	 */
300 	p->p_pctcpu = 0;
301 
302 	if ((p->p_flag & P_THREAD) == 0) {
303 		/* notify interested parties of our demise and clean up */
304 		knote_processexit(pr);
305 
306 		/*
307 		 * Notify parent that we're gone.  If we have P_NOZOMBIE
308 		 * or parent has the P_NOCLDWAIT flag set, notify process 1
309 		 * instead (and hope it will handle this situation).
310 		 */
311 		if ((p->p_flag & P_NOZOMBIE) ||
312 		    (pr->ps_pptr->ps_mainproc->p_flag & P_NOCLDWAIT)) {
313 			struct process *ppr = pr->ps_pptr;
314 			proc_reparent(pr, initproc->p_p);
315 			/*
316 			 * If this was the last child of our parent, notify
317 			 * parent, so in case he was wait(2)ing, he will
318 			 * continue.
319 			 */
320 			if (LIST_EMPTY(&ppr->ps_children))
321 				wakeup(ppr);
322 		}
323 	}
324 
325 	/*
326 	 * Release the process's signal state.
327 	 */
328 	sigactsfree(p);
329 
330 	/*
331 	 * Other substructures are freed from reaper and wait().
332 	 */
333 
334 	/*
335 	 * If emulation has process exit hook, call it now.
336 	 */
337 	if (p->p_emul->e_proc_exit)
338 		(*p->p_emul->e_proc_exit)(p);
339 
340 	/*
341 	 * Finally, call machine-dependent code to switch to a new
342 	 * context (possibly the idle context).  Once we are no longer
343 	 * using the dead process's vmspace and stack, exit2() will be
344 	 * called to schedule those resources to be released by the
345 	 * reaper thread.
346 	 *
347 	 * Note that cpu_exit() will end with a call equivalent to
348 	 * cpu_switch(), finishing our execution (pun intended).
349 	 */
350 	uvmexp.swtch++;
351 	cpu_exit(p);
352 	panic("cpu_exit returned");
353 }
354 
355 /*
356  * Locking of this proclist is special; it's accessed in a
357  * critical section of process exit, and thus locking it can't
358  * modify interrupt state.  We use a simple spin lock for this
359  * proclist.  Processes on this proclist are also on zombproc;
360  * we use the p_hash member to linkup to deadproc.
361  */
362 struct mutex deadproc_mutex = MUTEX_INITIALIZER(IPL_NONE);
363 struct proclist deadproc = LIST_HEAD_INITIALIZER(deadproc);
364 
365 /*
366  * We are called from cpu_exit() once it is safe to schedule the
367  * dead process's resources to be freed.
368  *
369  * NOTE: One must be careful with locking in this routine.  It's
370  * called from a critical section in machine-dependent code, so
371  * we should refrain from changing any interrupt state.
372  *
373  * We lock the deadproc list, place the proc on that list (using
374  * the p_hash member), and wake up the reaper.
375  */
376 void
377 exit2(struct proc *p)
378 {
379 	mtx_enter(&deadproc_mutex);
380 	LIST_INSERT_HEAD(&deadproc, p, p_hash);
381 	mtx_leave(&deadproc_mutex);
382 
383 	wakeup(&deadproc);
384 }
385 
386 /*
387  * Process reaper.  This is run by a kernel thread to free the resources
388  * of a dead process.  Once the resources are free, the process becomes
389  * a zombie, and the parent is allowed to read the undead's status.
390  */
391 void
392 reaper(void)
393 {
394 	struct proc *p;
395 
396 	KERNEL_PROC_UNLOCK(curproc);
397 
398 	SCHED_ASSERT_UNLOCKED();
399 
400 	for (;;) {
401 		mtx_enter(&deadproc_mutex);
402 		while ((p = LIST_FIRST(&deadproc)) == NULL)
403 			msleep(&deadproc, &deadproc_mutex, PVM, "reaper", 0);
404 
405 		/* Remove us from the deadproc list. */
406 		LIST_REMOVE(p, p_hash);
407 		mtx_leave(&deadproc_mutex);
408 
409 		KERNEL_PROC_LOCK(curproc);
410 
411 		/*
412 		 * Free the VM resources we're still holding on to.
413 		 * We must do this from a valid thread because doing
414 		 * so may block.
415 		 */
416 		uvm_exit(p);
417 
418 		/* Process is now a true zombie. */
419 		if ((p->p_flag & P_NOZOMBIE) == 0) {
420 			p->p_stat = SZOMB;
421 
422 			if (P_EXITSIG(p) != 0)
423 				prsignal(p->p_p->ps_pptr, P_EXITSIG(p));
424 			/* Wake up the parent so it can get exit status. */
425 			wakeup(p->p_p->ps_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 process *pr;
446 	struct proc *p;
447 	int status, error;
448 
449 	if (SCARG(uap, pid) == 0)
450 		SCARG(uap, pid) = -q->p_p->ps_pgid;
451 	if (SCARG(uap, options) &~ (WUNTRACED|WNOHANG|WALTSIG|WCONTINUED))
452 		return (EINVAL);
453 
454 loop:
455 	nfound = 0;
456 	LIST_FOREACH(pr, &q->p_p->ps_children, ps_sibling) {
457 		p = pr->ps_mainproc;
458 		if ((p->p_flag & P_NOZOMBIE) ||
459 		    (SCARG(uap, pid) != WAIT_ANY &&
460 		    p->p_pid != SCARG(uap, pid) &&
461 		    pr->ps_pgid != -SCARG(uap, pid)))
462 			continue;
463 
464 		/*
465 		 * Wait for processes with p_exitsig != SIGCHLD processes only
466 		 * if WALTSIG is set; wait for processes with pexitsig ==
467 		 * SIGCHLD only if WALTSIG is clear.
468 		 */
469 		if ((SCARG(uap, options) & WALTSIG) ?
470 		    (p->p_exitsig == SIGCHLD) : (P_EXITSIG(p) != SIGCHLD))
471 			continue;
472 
473 		nfound++;
474 		if (p->p_stat == SZOMB) {
475 			retval[0] = p->p_pid;
476 
477 			if (SCARG(uap, status)) {
478 				status = p->p_xstat;	/* convert to int */
479 				error = copyout(&status,
480 				    SCARG(uap, status), sizeof(status));
481 				if (error)
482 					return (error);
483 			}
484 			if (SCARG(uap, rusage) &&
485 			    (error = copyout(p->p_ru,
486 			    SCARG(uap, rusage), sizeof(struct rusage))))
487 				return (error);
488 			proc_finish_wait(q, p);
489 			return (0);
490 		}
491 		if (p->p_stat == SSTOP && (p->p_flag & P_WAITED) == 0 &&
492 		    (p->p_flag & P_TRACED || SCARG(uap, options) & WUNTRACED)) {
493 			atomic_setbits_int(&p->p_flag, P_WAITED);
494 			retval[0] = p->p_pid;
495 
496 			if (SCARG(uap, status)) {
497 				status = W_STOPCODE(p->p_xstat);
498 				error = copyout(&status, SCARG(uap, status),
499 				    sizeof(status));
500 			} else
501 				error = 0;
502 			return (error);
503 		}
504 		if ((SCARG(uap, options) & WCONTINUED) && (p->p_flag & P_CONTINUED)) {
505 			atomic_clearbits_int(&p->p_flag, P_CONTINUED);
506 			retval[0] = p->p_pid;
507 
508 			if (SCARG(uap, status)) {
509 				status = _WCONTINUED;
510 				error = copyout(&status, SCARG(uap, status),
511 				    sizeof(status));
512 			} else
513 				error = 0;
514 			return (error);
515 		}
516 	}
517 	if (nfound == 0)
518 		return (ECHILD);
519 	if (SCARG(uap, options) & WNOHANG) {
520 		retval[0] = 0;
521 		return (0);
522 	}
523 	if ((error = tsleep(q->p_p, PWAIT | PCATCH, "wait", 0)) != 0)
524 		return (error);
525 	goto loop;
526 }
527 
528 void
529 proc_finish_wait(struct proc *waiter, struct proc *p)
530 {
531 	struct process *tr;
532 
533 	/*
534 	 * If we got the child via a ptrace 'attach',
535 	 * we need to give it back to the old parent.
536 	 */
537 	if (p->p_oppid && (tr = prfind(p->p_oppid))) {
538 		atomic_clearbits_int(&p->p_flag, P_TRACED);
539 		p->p_oppid = 0;
540 		proc_reparent(p->p_p, tr);
541 		if (p->p_exitsig != 0)
542 			prsignal(tr, p->p_exitsig);
543 		wakeup(tr);
544 	} else {
545 		scheduler_wait_hook(waiter, p);
546 		p->p_xstat = 0;
547 		ruadd(&waiter->p_stats->p_cru, p->p_ru);
548 		proc_zap(p);
549 	}
550 }
551 
552 /*
553  * make process 'parent' the new parent of process 'child'.
554  */
555 void
556 proc_reparent(struct process *child, struct process *parent)
557 {
558 
559 	if (child->ps_pptr == parent)
560 		return;
561 
562 	if (parent == initproc->p_p)
563 		child->ps_mainproc->p_exitsig = SIGCHLD;
564 
565 	LIST_REMOVE(child, ps_sibling);
566 	LIST_INSERT_HEAD(&parent->ps_children, child, ps_sibling);
567 	child->ps_pptr = parent;
568 }
569 
570 void
571 proc_zap(struct proc *p)
572 {
573 	struct process *pr = p->p_p;
574 
575 	pool_put(&rusage_pool, p->p_ru);
576 	if (p->p_ptstat)
577 		free(p->p_ptstat, M_SUBPROC);
578 
579 	/*
580 	 * Finally finished with old proc entry.
581 	 * Unlink it from its process group and free it.
582 	 */
583 	if ((p->p_flag & P_THREAD) == 0)
584 		leavepgrp(pr);
585 	LIST_REMOVE(p, p_list);	/* off zombproc */
586 	if ((p->p_flag & P_THREAD) == 0)
587 		LIST_REMOVE(pr, ps_sibling);
588 
589 	/*
590 	 * Decrement the count of procs running with this uid.
591 	 */
592 	(void)chgproccnt(p->p_cred->p_ruid, -1);
593 
594 	/*
595 	 * Release reference to text vnode
596 	 */
597 	if (p->p_textvp)
598 		vrele(p->p_textvp);
599 
600 	/*
601 	 * Remove us from our process list, possibly killing the process
602 	 * in the process (pun intended).
603 	 */
604 	if (--pr->ps_refcnt == 0) {
605 		KASSERT(TAILQ_EMPTY(&pr->ps_threads));
606 		limfree(pr->ps_limit);
607 		crfree(pr->ps_cred->pc_ucred);
608 		pool_put(&pcred_pool, pr->ps_cred);
609 		pool_put(&process_pool, pr);
610 	}
611 
612 	pool_put(&proc_pool, p);
613 	nprocs--;
614 }
615