xref: /openbsd-src/sys/kern/kern_exit.c (revision 8f15e6a4ddcf78e1c1320bc27380937fb93b1657)
1 /*	$OpenBSD: kern_exit.c,v 1.109 2012/03/23 15:51:26 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 process *pr, *qr, *nqr;
123 	struct rusage *rup;
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 	pr = p->p_p;
132 
133 	/* single-threaded? */
134 	if (TAILQ_FIRST(&pr->ps_threads) == p &&
135 	    TAILQ_NEXT(p, p_thr_link) == NULL)
136 		flags = EXIT_NORMAL;
137 	else {
138 		/* nope, multi-threaded */
139 		if (flags == EXIT_NORMAL)
140 			single_thread_set(p, SINGLE_EXIT, 0);
141 	}
142 
143 	if (flags == EXIT_NORMAL) {
144 		atomic_setbits_int(&pr->ps_flags, PS_EXITING);
145 		pr->ps_mainproc->p_xstat = rv;
146 
147 		/*
148 		 * If parent is waiting for us to exit or exec, PS_PPWAIT
149 		 * is set; we wake up the parent early to avoid deadlock.
150 		 */
151 		if (pr->ps_flags & PS_PPWAIT) {
152 			atomic_clearbits_int(&pr->ps_flags, PS_PPWAIT);
153 			atomic_clearbits_int(&pr->ps_pptr->ps_flags,
154 			    PS_ISPWAIT);
155 			wakeup(pr->ps_pptr);
156 		}
157 	}
158 
159 	/* unlink ourselves from the active threads */
160 	TAILQ_REMOVE(&pr->ps_threads, p, p_thr_link);
161 	if ((p->p_flag & P_THREAD) == 0) {
162 		/* main thread gotta wait because it has the pid, et al */
163 		while (! TAILQ_EMPTY(&pr->ps_threads))
164 			tsleep(&pr->ps_threads, PUSER, "thrdeath", 0);
165 	} else if (TAILQ_EMPTY(&pr->ps_threads))
166 		wakeup(&pr->ps_threads);
167 
168 	if (p->p_flag & P_PROFIL)
169 		stopprofclock(p);
170 	rup = pr->ps_ru;
171 	if (rup == NULL) {
172 		rup = pool_get(&rusage_pool, PR_WAITOK | PR_ZERO);
173 
174 		if (pr->ps_ru == NULL)
175 			pr->ps_ru = rup;
176 		else {
177 			pool_put(&rusage_pool, rup);
178 			rup = pr->ps_ru;
179 		}
180 	}
181 	p->p_siglist = 0;
182 
183 	/*
184 	 * Close open files and release open-file table.
185 	 */
186 	fdfree(p);
187 
188 	if ((p->p_flag & P_THREAD) == 0) {
189 		timeout_del(&pr->ps_realit_to);
190 		timeout_del(&pr->ps_virt_to);
191 		timeout_del(&pr->ps_prof_to);
192 #ifdef SYSVSEM
193 		semexit(pr);
194 #endif
195 		if (SESS_LEADER(pr)) {
196 			struct session *sp = pr->ps_session;
197 
198 			if (sp->s_ttyvp) {
199 				/*
200 				 * Controlling process.
201 				 * Signal foreground pgrp,
202 				 * drain controlling terminal
203 				 * and revoke access to controlling terminal.
204 				 */
205 				if (sp->s_ttyp->t_session == sp) {
206 					if (sp->s_ttyp->t_pgrp)
207 						pgsignal(sp->s_ttyp->t_pgrp,
208 						    SIGHUP, 1);
209 					(void) ttywait(sp->s_ttyp);
210 					/*
211 					 * The tty could have been revoked
212 					 * if we blocked.
213 					 */
214 					if (sp->s_ttyvp)
215 						VOP_REVOKE(sp->s_ttyvp,
216 						    REVOKEALL);
217 				}
218 				if (sp->s_ttyvp)
219 					vrele(sp->s_ttyvp);
220 				sp->s_ttyvp = NULL;
221 				/*
222 				 * s_ttyp is not zero'd; we use this to
223 				 * indicate that the session once had a
224 				 * controlling terminal.  (for logging and
225 				 * informational purposes)
226 				 */
227 			}
228 			sp->s_leader = NULL;
229 		}
230 		fixjobc(pr, pr->ps_pgrp, 0);
231 
232 #ifdef ACCOUNTING
233 		(void)acct_process(p);
234 #endif
235 
236 #ifdef KTRACE
237 		/* release trace file */
238 		if (pr->ps_tracevp)
239 			ktrcleartrace(pr);
240 #endif
241 	}
242 
243 #if NSYSTRACE > 0
244 	if (ISSET(p->p_flag, P_SYSTRACE))
245 		systrace_exit(p);
246 #endif
247         /*
248          * Remove proc from pidhash chain so looking it up won't
249          * work.  Move it from allproc to zombproc, but do not yet
250          * wake up the reaper.  We will put the proc on the
251          * deadproc list later (using the p_hash member), and
252          * wake up the reaper when we do.
253          */
254 	/*
255 	 * NOTE: WE ARE NO LONGER ALLOWED TO SLEEP!
256 	 */
257 	p->p_stat = SDEAD;
258 
259 	LIST_REMOVE(p, p_hash);
260 	LIST_REMOVE(p, p_list);
261 	LIST_INSERT_HEAD(&zombproc, p, p_list);
262 
263 	/*
264 	 * Give orphaned children to init(8).
265 	 */
266 	if ((p->p_flag & P_THREAD) == 0) {
267 		qr = LIST_FIRST(&pr->ps_children);
268 		if (qr)		/* only need this if any child is S_ZOMB */
269 			wakeup(initproc->p_p);
270 		for (; qr != 0; qr = nqr) {
271 			nqr = LIST_NEXT(qr, ps_sibling);
272 			proc_reparent(qr, initproc->p_p);
273 			/*
274 			 * Traced processes are killed
275 			 * since their existence means someone is screwing up.
276 			 */
277 			if (qr->ps_flags & PS_TRACED) {
278 				atomic_clearbits_int(&qr->ps_flags, PS_TRACED);
279 				prsignal(qr, SIGKILL);
280 			}
281 		}
282 	}
283 
284 
285 	/* add thread's accumulated rusage into the process's total */
286 	ruadd(rup, &p->p_ru);
287 
288 	/*
289 	 * clear %cpu usage during swap
290 	 */
291 	p->p_pctcpu = 0;
292 
293 	if ((p->p_flag & P_THREAD) == 0) {
294 		/*
295 		 * Final thread has died, so add on our children's rusage
296 		 * and calculate the total times
297 		 */
298 		calcru(&pr->ps_tu, &rup->ru_utime, &rup->ru_stime, NULL);
299 		ruadd(rup, &pr->ps_cru);
300 		timeradd(&rup->ru_utime, &pr->ps_cru.ru_utime, &rup->ru_utime);
301 		timeradd(&rup->ru_stime, &pr->ps_cru.ru_stime, &rup->ru_stime);
302 
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 SAS_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_sigacts->ps_flags &
313 		    SAS_NOCLDWAIT)) {
314 			struct process *ppr = pr->ps_pptr;
315 			proc_reparent(pr, initproc->p_p);
316 			/*
317 			 * If this was the last child of our parent, notify
318 			 * parent, so in case he was wait(2)ing, he will
319 			 * continue.
320 			 */
321 			if (LIST_EMPTY(&ppr->ps_children))
322 				wakeup(ppr);
323 		}
324 	}
325 
326 	/*
327 	 * Release the process's signal state.
328 	 */
329 	sigactsfree(p);
330 
331 	/*
332 	 * Other substructures are freed from reaper and wait().
333 	 */
334 
335 	/*
336 	 * If emulation has process exit hook, call it now.
337 	 */
338 	if (p->p_emul->e_proc_exit)
339 		(*p->p_emul->e_proc_exit)(p);
340 
341 	/*
342 	 * Finally, call machine-dependent code to switch to a new
343 	 * context (possibly the idle context).  Once we are no longer
344 	 * using the dead process's vmspace and stack, exit2() will be
345 	 * called to schedule those resources to be released by the
346 	 * reaper thread.
347 	 *
348 	 * Note that cpu_exit() will end with a call equivalent to
349 	 * cpu_switch(), finishing our execution (pun intended).
350 	 */
351 	uvmexp.swtch++;
352 	cpu_exit(p);
353 	panic("cpu_exit returned");
354 }
355 
356 /*
357  * Locking of this proclist is special; it's accessed in a
358  * critical section of process exit, and thus locking it can't
359  * modify interrupt state.  We use a simple spin lock for this
360  * proclist.  Processes on this proclist are also on zombproc;
361  * we use the p_hash member to linkup to deadproc.
362  */
363 struct mutex deadproc_mutex = MUTEX_INITIALIZER(IPL_NONE);
364 struct proclist deadproc = LIST_HEAD_INITIALIZER(deadproc);
365 
366 /*
367  * We are called from cpu_exit() once it is safe to schedule the
368  * dead process's resources to be freed.
369  *
370  * NOTE: One must be careful with locking in this routine.  It's
371  * called from a critical section in machine-dependent code, so
372  * we should refrain from changing any interrupt state.
373  *
374  * We lock the deadproc list, place the proc on that list (using
375  * the p_hash member), and wake up the reaper.
376  */
377 void
378 exit2(struct proc *p)
379 {
380 	mtx_enter(&deadproc_mutex);
381 	LIST_INSERT_HEAD(&deadproc, p, p_hash);
382 	mtx_leave(&deadproc_mutex);
383 
384 	wakeup(&deadproc);
385 }
386 
387 /*
388  * Process reaper.  This is run by a kernel thread to free the resources
389  * of a dead process.  Once the resources are free, the process becomes
390  * a zombie, and the parent is allowed to read the undead's status.
391  */
392 void
393 reaper(void)
394 {
395 	struct proc *p;
396 
397 	KERNEL_UNLOCK();
398 
399 	SCHED_ASSERT_UNLOCKED();
400 
401 	for (;;) {
402 		mtx_enter(&deadproc_mutex);
403 		while ((p = LIST_FIRST(&deadproc)) == NULL)
404 			msleep(&deadproc, &deadproc_mutex, PVM, "reaper", 0);
405 
406 		/* Remove us from the deadproc list. */
407 		LIST_REMOVE(p, p_hash);
408 		mtx_leave(&deadproc_mutex);
409 
410 		KERNEL_LOCK();
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 			if (P_EXITSIG(p) != 0)
424 				prsignal(p->p_p->ps_pptr, P_EXITSIG(p));
425 			/* Wake up the parent so it can get exit status. */
426 			wakeup(p->p_p->ps_pptr);
427 		} else {
428 			/* Noone will wait for us. Just zap the process now */
429 			proc_zap(p);
430 		}
431 
432 		KERNEL_UNLOCK();
433 	}
434 }
435 
436 int
437 sys_wait4(struct proc *q, void *v, register_t *retval)
438 {
439 	struct sys_wait4_args /* {
440 		syscallarg(pid_t) pid;
441 		syscallarg(int *) status;
442 		syscallarg(int) options;
443 		syscallarg(struct rusage *) rusage;
444 	} */ *uap = v;
445 	int nfound;
446 	struct process *pr;
447 	struct proc *p;
448 	int status, error;
449 
450 	if (SCARG(uap, pid) == 0)
451 		SCARG(uap, pid) = -q->p_p->ps_pgid;
452 	if (SCARG(uap, options) &~ (WUNTRACED|WNOHANG|WALTSIG|WCONTINUED))
453 		return (EINVAL);
454 
455 loop:
456 	nfound = 0;
457 	LIST_FOREACH(pr, &q->p_p->ps_children, ps_sibling) {
458 		p = pr->ps_mainproc;
459 		if ((p->p_flag & P_NOZOMBIE) ||
460 		    (SCARG(uap, pid) != WAIT_ANY &&
461 		    p->p_pid != SCARG(uap, pid) &&
462 		    pr->ps_pgid != -SCARG(uap, pid)))
463 			continue;
464 
465 		/*
466 		 * Wait for processes with p_exitsig != SIGCHLD processes only
467 		 * if WALTSIG is set; wait for processes with pexitsig ==
468 		 * SIGCHLD only if WALTSIG is clear.
469 		 */
470 		if ((SCARG(uap, options) & WALTSIG) ?
471 		    (p->p_exitsig == SIGCHLD) : (P_EXITSIG(p) != SIGCHLD))
472 			continue;
473 
474 		nfound++;
475 		if (p->p_stat == SZOMB) {
476 			retval[0] = p->p_pid;
477 
478 			if (SCARG(uap, status)) {
479 				status = p->p_xstat;	/* convert to int */
480 				error = copyout(&status,
481 				    SCARG(uap, status), sizeof(status));
482 				if (error)
483 					return (error);
484 			}
485 			if (SCARG(uap, rusage) &&
486 			    (error = copyout(pr->ps_ru,
487 			    SCARG(uap, rusage), sizeof(struct rusage))))
488 				return (error);
489 			proc_finish_wait(q, p);
490 			return (0);
491 		}
492 		if (p->p_stat == SSTOP &&
493 		    (p->p_flag & (P_WAITED|P_SUSPSINGLE)) == 0 &&
494 		    (pr->ps_flags & PS_TRACED ||
495 		    SCARG(uap, options) & WUNTRACED)) {
496 			atomic_setbits_int(&p->p_flag, P_WAITED);
497 			retval[0] = p->p_pid;
498 
499 			if (SCARG(uap, status)) {
500 				status = W_STOPCODE(p->p_xstat);
501 				error = copyout(&status, SCARG(uap, status),
502 				    sizeof(status));
503 			} else
504 				error = 0;
505 			return (error);
506 		}
507 		if ((SCARG(uap, options) & WCONTINUED) && (p->p_flag & P_CONTINUED)) {
508 			atomic_clearbits_int(&p->p_flag, P_CONTINUED);
509 			retval[0] = p->p_pid;
510 
511 			if (SCARG(uap, status)) {
512 				status = _WCONTINUED;
513 				error = copyout(&status, SCARG(uap, status),
514 				    sizeof(status));
515 			} else
516 				error = 0;
517 			return (error);
518 		}
519 	}
520 	if (nfound == 0)
521 		return (ECHILD);
522 	if (SCARG(uap, options) & WNOHANG) {
523 		retval[0] = 0;
524 		return (0);
525 	}
526 	if ((error = tsleep(q->p_p, PWAIT | PCATCH, "wait", 0)) != 0)
527 		return (error);
528 	goto loop;
529 }
530 
531 void
532 proc_finish_wait(struct proc *waiter, struct proc *p)
533 {
534 	struct process *pr, *tr;
535 	struct rusage *rup;
536 
537 	/*
538 	 * If we got the child via a ptrace 'attach',
539 	 * we need to give it back to the old parent.
540 	 */
541 	pr = p->p_p;
542 	if ((p->p_flag & P_THREAD) == 0 && pr->ps_oppid &&
543 	    (tr = prfind(pr->ps_oppid))) {
544 		atomic_clearbits_int(&pr->ps_flags, PS_TRACED);
545 		pr->ps_oppid = 0;
546 		proc_reparent(pr, tr);
547 		if (p->p_exitsig != 0)
548 			prsignal(tr, p->p_exitsig);
549 		wakeup(tr);
550 	} else {
551 		scheduler_wait_hook(waiter, p);
552 		p->p_xstat = 0;
553 		rup = &waiter->p_p->ps_cru;
554 		ruadd(rup, pr->ps_ru);
555 		timeradd(&rup->ru_utime, &pr->ps_ru->ru_utime, &rup->ru_utime);
556 		timeradd(&rup->ru_stime, &pr->ps_ru->ru_stime, &rup->ru_stime);
557 		proc_zap(p);
558 	}
559 }
560 
561 /*
562  * make process 'parent' the new parent of process 'child'.
563  */
564 void
565 proc_reparent(struct process *child, struct process *parent)
566 {
567 
568 	if (child->ps_pptr == parent)
569 		return;
570 
571 	if (parent == initproc->p_p)
572 		child->ps_mainproc->p_exitsig = SIGCHLD;
573 
574 	LIST_REMOVE(child, ps_sibling);
575 	LIST_INSERT_HEAD(&parent->ps_children, child, ps_sibling);
576 	child->ps_pptr = parent;
577 }
578 
579 void
580 proc_zap(struct proc *p)
581 {
582 	struct process *pr = p->p_p;
583 
584 	/*
585 	 * Finally finished with old proc entry.
586 	 * Unlink it from its process group and free it.
587 	 */
588 	if ((p->p_flag & P_THREAD) == 0)
589 		leavepgrp(pr);
590 	LIST_REMOVE(p, p_list);	/* off zombproc */
591 	if ((p->p_flag & P_THREAD) == 0)
592 		LIST_REMOVE(pr, ps_sibling);
593 
594 	/*
595 	 * Decrement the count of procs running with this uid.
596 	 */
597 	(void)chgproccnt(p->p_cred->p_ruid, -1);
598 
599 	/*
600 	 * Release reference to text vnode
601 	 */
602 	if (p->p_textvp)
603 		vrele(p->p_textvp);
604 
605 	/*
606 	 * Remove us from our process list, possibly killing the process
607 	 * in the process (pun intended).
608 	 */
609 	if (--pr->ps_refcnt == 0) {
610 		if (pr->ps_ptstat != NULL)
611 			free(pr->ps_ptstat, M_SUBPROC);
612 		pool_put(&rusage_pool, pr->ps_ru);
613 		KASSERT(TAILQ_EMPTY(&pr->ps_threads));
614 		limfree(pr->ps_limit);
615 		crfree(pr->ps_cred->pc_ucred);
616 		pool_put(&pcred_pool, pr->ps_cred);
617 		pool_put(&process_pool, pr);
618 	}
619 
620 	pool_put(&proc_pool, p);
621 	nprocs--;
622 }
623