xref: /openbsd-src/sys/kern/kern_exit.c (revision 2ea51a70232b4bd59dc6059ffd1585fd81f5b25c)
1 /*	$OpenBSD: kern_exit.c,v 1.104 2011/12/11 19:42:28 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 
124 	if (p->p_pid == 1)
125 		panic("init died (signal %d, exit %d)",
126 		    WTERMSIG(rv), WEXITSTATUS(rv));
127 
128 	atomic_setbits_int(&p->p_flag, P_WEXIT);
129 
130 	pr = p->p_p;
131 
132 	/* single-threaded? */
133 	if (TAILQ_FIRST(&pr->ps_threads) == p &&
134 	    TAILQ_NEXT(p, p_thr_link) == NULL)
135 		flags = EXIT_NORMAL;
136 	else {
137 		/* nope, multi-threaded */
138 		if (flags == EXIT_NORMAL)
139 			single_thread_set(p, SINGLE_EXIT, 0);
140 	}
141 
142 	if (flags == EXIT_NORMAL) {
143 		pr->ps_mainproc->p_xstat = rv;
144 
145 		/*
146 		 * If parent is waiting for us to exit or exec, PS_PPWAIT
147 		 * is set; we wake up the parent early to avoid deadlock.
148 		 */
149 		if (pr->ps_flags & PS_PPWAIT) {
150 			atomic_clearbits_int(&pr->ps_flags, PS_PPWAIT);
151 			atomic_clearbits_int(&pr->ps_pptr->ps_flags,
152 			    PS_ISPWAIT);
153 			wakeup(pr->ps_pptr);
154 		}
155 	}
156 
157 	/* unlink ourselves from the active threads */
158 	TAILQ_REMOVE(&pr->ps_threads, p, p_thr_link);
159 	if ((p->p_flag & P_THREAD) == 0) {
160 		/* main thread gotta wait because it has the pid, et al */
161 		while (! TAILQ_EMPTY(&pr->ps_threads))
162 			tsleep(&pr->ps_threads, PUSER, "thrdeath", 0);
163 	} else if (TAILQ_EMPTY(&pr->ps_threads))
164 		wakeup(&pr->ps_threads);
165 
166 	if (p->p_flag & P_PROFIL)
167 		stopprofclock(p);
168 	p->p_ru = pool_get(&rusage_pool, PR_WAITOK);
169 	p->p_siglist = 0;
170 	timeout_del(&p->p_realit_to);
171 	timeout_del(&p->p_stats->p_virt_to);
172 	timeout_del(&p->p_stats->p_prof_to);
173 
174 	/*
175 	 * Close open files and release open-file table.
176 	 */
177 	fdfree(p);
178 
179 	if ((p->p_flag & P_THREAD) == 0) {
180 #ifdef SYSVSEM
181 		semexit(pr);
182 #endif
183 		if (SESS_LEADER(pr)) {
184 			struct session *sp = pr->ps_session;
185 
186 			if (sp->s_ttyvp) {
187 				/*
188 				 * Controlling process.
189 				 * Signal foreground pgrp,
190 				 * drain controlling terminal
191 				 * and revoke access to controlling terminal.
192 				 */
193 				if (sp->s_ttyp->t_session == sp) {
194 					if (sp->s_ttyp->t_pgrp)
195 						pgsignal(sp->s_ttyp->t_pgrp,
196 						    SIGHUP, 1);
197 					(void) ttywait(sp->s_ttyp);
198 					/*
199 					 * The tty could have been revoked
200 					 * if we blocked.
201 					 */
202 					if (sp->s_ttyvp)
203 						VOP_REVOKE(sp->s_ttyvp,
204 						    REVOKEALL);
205 				}
206 				if (sp->s_ttyvp)
207 					vrele(sp->s_ttyvp);
208 				sp->s_ttyvp = NULL;
209 				/*
210 				 * s_ttyp is not zero'd; we use this to
211 				 * indicate that the session once had a
212 				 * controlling terminal.  (for logging and
213 				 * informational purposes)
214 				 */
215 			}
216 			sp->s_leader = NULL;
217 		}
218 		fixjobc(pr, pr->ps_pgrp, 0);
219 
220 #ifdef ACCOUNTING
221 		(void)acct_process(p);
222 #endif
223 	}
224 
225 #ifdef KTRACE
226 	/*
227 	 * release trace file
228 	 */
229 	p->p_traceflag = 0;	/* don't trace the vrele() */
230 	if (p->p_tracep)
231 		ktrsettracevnode(p, NULL);
232 #endif
233 #if NSYSTRACE > 0
234 	if (ISSET(p->p_flag, P_SYSTRACE))
235 		systrace_exit(p);
236 #endif
237         /*
238          * Remove proc from pidhash chain so looking it up won't
239          * work.  Move it from allproc to zombproc, but do not yet
240          * wake up the reaper.  We will put the proc on the
241          * deadproc list later (using the p_hash member), and
242          * wake up the reaper when we do.
243          */
244 	/*
245 	 * NOTE: WE ARE NO LONGER ALLOWED TO SLEEP!
246 	 */
247 	p->p_stat = SDEAD;
248 
249 	LIST_REMOVE(p, p_hash);
250 	LIST_REMOVE(p, p_list);
251 	LIST_INSERT_HEAD(&zombproc, p, p_list);
252 
253 	/*
254 	 * Give orphaned children to init(8).
255 	 */
256 	if ((p->p_flag & P_THREAD) == 0) {
257 		qr = LIST_FIRST(&pr->ps_children);
258 		if (qr)		/* only need this if any child is S_ZOMB */
259 			wakeup(initproc->p_p);
260 		for (; qr != 0; qr = nqr) {
261 			nqr = LIST_NEXT(qr, ps_sibling);
262 			proc_reparent(qr, initproc->p_p);
263 			/*
264 			 * Traced processes are killed
265 			 * since their existence means someone is screwing up.
266 			 */
267 			if (qr->ps_mainproc->p_flag & P_TRACED) {
268 				atomic_clearbits_int(&qr->ps_mainproc->p_flag,
269 				    P_TRACED);
270 				prsignal(qr, SIGKILL);
271 			}
272 		}
273 	}
274 
275 
276 	/*
277 	 * Save exit status and final rusage info, adding in child rusage
278 	 * info and self times.
279 	 */
280 	*p->p_ru = p->p_stats->p_ru;
281 	calcru(p, &p->p_ru->ru_utime, &p->p_ru->ru_stime, NULL);
282 	ruadd(p->p_ru, &p->p_stats->p_cru);
283 
284 	/*
285 	 * clear %cpu usage during swap
286 	 */
287 	p->p_pctcpu = 0;
288 
289 	if ((p->p_flag & P_THREAD) == 0) {
290 		/* notify interested parties of our demise and clean up */
291 		knote_processexit(pr);
292 
293 		/*
294 		 * Notify parent that we're gone.  If we have P_NOZOMBIE
295 		 * or parent has the SAS_NOCLDWAIT flag set, notify process 1
296 		 * instead (and hope it will handle this situation).
297 		 */
298 		if ((p->p_flag & P_NOZOMBIE) ||
299 		    (pr->ps_pptr->ps_mainproc->p_sigacts->ps_flags &
300 		    SAS_NOCLDWAIT)) {
301 			struct process *ppr = pr->ps_pptr;
302 			proc_reparent(pr, initproc->p_p);
303 			/*
304 			 * If this was the last child of our parent, notify
305 			 * parent, so in case he was wait(2)ing, he will
306 			 * continue.
307 			 */
308 			if (LIST_EMPTY(&ppr->ps_children))
309 				wakeup(ppr);
310 		}
311 	}
312 
313 	/*
314 	 * Release the process's signal state.
315 	 */
316 	sigactsfree(p);
317 
318 	/*
319 	 * Other substructures are freed from reaper and wait().
320 	 */
321 
322 	/*
323 	 * If emulation has process exit hook, call it now.
324 	 */
325 	if (p->p_emul->e_proc_exit)
326 		(*p->p_emul->e_proc_exit)(p);
327 
328 	/*
329 	 * Finally, call machine-dependent code to switch to a new
330 	 * context (possibly the idle context).  Once we are no longer
331 	 * using the dead process's vmspace and stack, exit2() will be
332 	 * called to schedule those resources to be released by the
333 	 * reaper thread.
334 	 *
335 	 * Note that cpu_exit() will end with a call equivalent to
336 	 * cpu_switch(), finishing our execution (pun intended).
337 	 */
338 	uvmexp.swtch++;
339 	cpu_exit(p);
340 	panic("cpu_exit returned");
341 }
342 
343 /*
344  * Locking of this proclist is special; it's accessed in a
345  * critical section of process exit, and thus locking it can't
346  * modify interrupt state.  We use a simple spin lock for this
347  * proclist.  Processes on this proclist are also on zombproc;
348  * we use the p_hash member to linkup to deadproc.
349  */
350 struct mutex deadproc_mutex = MUTEX_INITIALIZER(IPL_NONE);
351 struct proclist deadproc = LIST_HEAD_INITIALIZER(deadproc);
352 
353 /*
354  * We are called from cpu_exit() once it is safe to schedule the
355  * dead process's resources to be freed.
356  *
357  * NOTE: One must be careful with locking in this routine.  It's
358  * called from a critical section in machine-dependent code, so
359  * we should refrain from changing any interrupt state.
360  *
361  * We lock the deadproc list, place the proc on that list (using
362  * the p_hash member), and wake up the reaper.
363  */
364 void
365 exit2(struct proc *p)
366 {
367 	mtx_enter(&deadproc_mutex);
368 	LIST_INSERT_HEAD(&deadproc, p, p_hash);
369 	mtx_leave(&deadproc_mutex);
370 
371 	wakeup(&deadproc);
372 }
373 
374 /*
375  * Process reaper.  This is run by a kernel thread to free the resources
376  * of a dead process.  Once the resources are free, the process becomes
377  * a zombie, and the parent is allowed to read the undead's status.
378  */
379 void
380 reaper(void)
381 {
382 	struct proc *p;
383 
384 	KERNEL_UNLOCK();
385 
386 	SCHED_ASSERT_UNLOCKED();
387 
388 	for (;;) {
389 		mtx_enter(&deadproc_mutex);
390 		while ((p = LIST_FIRST(&deadproc)) == NULL)
391 			msleep(&deadproc, &deadproc_mutex, PVM, "reaper", 0);
392 
393 		/* Remove us from the deadproc list. */
394 		LIST_REMOVE(p, p_hash);
395 		mtx_leave(&deadproc_mutex);
396 
397 		KERNEL_LOCK();
398 
399 		/*
400 		 * Free the VM resources we're still holding on to.
401 		 * We must do this from a valid thread because doing
402 		 * so may block.
403 		 */
404 		uvm_exit(p);
405 
406 		/* Process is now a true zombie. */
407 		if ((p->p_flag & P_NOZOMBIE) == 0) {
408 			p->p_stat = SZOMB;
409 
410 			if (P_EXITSIG(p) != 0)
411 				prsignal(p->p_p->ps_pptr, P_EXITSIG(p));
412 			/* Wake up the parent so it can get exit status. */
413 			wakeup(p->p_p->ps_pptr);
414 		} else {
415 			/* Noone will wait for us. Just zap the process now */
416 			proc_zap(p);
417 		}
418 
419 		KERNEL_UNLOCK();
420 	}
421 }
422 
423 int
424 sys_wait4(struct proc *q, void *v, register_t *retval)
425 {
426 	struct sys_wait4_args /* {
427 		syscallarg(pid_t) pid;
428 		syscallarg(int *) status;
429 		syscallarg(int) options;
430 		syscallarg(struct rusage *) rusage;
431 	} */ *uap = v;
432 	int nfound;
433 	struct process *pr;
434 	struct proc *p;
435 	int status, error;
436 
437 	if (SCARG(uap, pid) == 0)
438 		SCARG(uap, pid) = -q->p_p->ps_pgid;
439 	if (SCARG(uap, options) &~ (WUNTRACED|WNOHANG|WALTSIG|WCONTINUED))
440 		return (EINVAL);
441 
442 loop:
443 	nfound = 0;
444 	LIST_FOREACH(pr, &q->p_p->ps_children, ps_sibling) {
445 		p = pr->ps_mainproc;
446 		if ((p->p_flag & P_NOZOMBIE) ||
447 		    (SCARG(uap, pid) != WAIT_ANY &&
448 		    p->p_pid != SCARG(uap, pid) &&
449 		    pr->ps_pgid != -SCARG(uap, pid)))
450 			continue;
451 
452 		/*
453 		 * Wait for processes with p_exitsig != SIGCHLD processes only
454 		 * if WALTSIG is set; wait for processes with pexitsig ==
455 		 * SIGCHLD only if WALTSIG is clear.
456 		 */
457 		if ((SCARG(uap, options) & WALTSIG) ?
458 		    (p->p_exitsig == SIGCHLD) : (P_EXITSIG(p) != SIGCHLD))
459 			continue;
460 
461 		nfound++;
462 		if (p->p_stat == SZOMB) {
463 			retval[0] = p->p_pid;
464 
465 			if (SCARG(uap, status)) {
466 				status = p->p_xstat;	/* convert to int */
467 				error = copyout(&status,
468 				    SCARG(uap, status), sizeof(status));
469 				if (error)
470 					return (error);
471 			}
472 			if (SCARG(uap, rusage) &&
473 			    (error = copyout(p->p_ru,
474 			    SCARG(uap, rusage), sizeof(struct rusage))))
475 				return (error);
476 			proc_finish_wait(q, p);
477 			return (0);
478 		}
479 		if (p->p_stat == SSTOP &&
480 		    (p->p_flag & (P_WAITED|P_SUSPSINGLE)) == 0 &&
481 		    (p->p_flag & P_TRACED || SCARG(uap, options) & WUNTRACED)) {
482 			atomic_setbits_int(&p->p_flag, P_WAITED);
483 			retval[0] = p->p_pid;
484 
485 			if (SCARG(uap, status)) {
486 				status = W_STOPCODE(p->p_xstat);
487 				error = copyout(&status, SCARG(uap, status),
488 				    sizeof(status));
489 			} else
490 				error = 0;
491 			return (error);
492 		}
493 		if ((SCARG(uap, options) & WCONTINUED) && (p->p_flag & P_CONTINUED)) {
494 			atomic_clearbits_int(&p->p_flag, P_CONTINUED);
495 			retval[0] = p->p_pid;
496 
497 			if (SCARG(uap, status)) {
498 				status = _WCONTINUED;
499 				error = copyout(&status, SCARG(uap, status),
500 				    sizeof(status));
501 			} else
502 				error = 0;
503 			return (error);
504 		}
505 	}
506 	if (nfound == 0)
507 		return (ECHILD);
508 	if (SCARG(uap, options) & WNOHANG) {
509 		retval[0] = 0;
510 		return (0);
511 	}
512 	if ((error = tsleep(q->p_p, PWAIT | PCATCH, "wait", 0)) != 0)
513 		return (error);
514 	goto loop;
515 }
516 
517 void
518 proc_finish_wait(struct proc *waiter, struct proc *p)
519 {
520 	struct process *tr;
521 
522 	/*
523 	 * If we got the child via a ptrace 'attach',
524 	 * we need to give it back to the old parent.
525 	 */
526 	if (p->p_oppid && (tr = prfind(p->p_oppid))) {
527 		atomic_clearbits_int(&p->p_flag, P_TRACED);
528 		p->p_oppid = 0;
529 		proc_reparent(p->p_p, tr);
530 		if (p->p_exitsig != 0)
531 			prsignal(tr, p->p_exitsig);
532 		wakeup(tr);
533 	} else {
534 		scheduler_wait_hook(waiter, p);
535 		p->p_xstat = 0;
536 		ruadd(&waiter->p_stats->p_cru, p->p_ru);
537 		proc_zap(p);
538 	}
539 }
540 
541 /*
542  * make process 'parent' the new parent of process 'child'.
543  */
544 void
545 proc_reparent(struct process *child, struct process *parent)
546 {
547 
548 	if (child->ps_pptr == parent)
549 		return;
550 
551 	if (parent == initproc->p_p)
552 		child->ps_mainproc->p_exitsig = SIGCHLD;
553 
554 	LIST_REMOVE(child, ps_sibling);
555 	LIST_INSERT_HEAD(&parent->ps_children, child, ps_sibling);
556 	child->ps_pptr = parent;
557 }
558 
559 void
560 proc_zap(struct proc *p)
561 {
562 	struct process *pr = p->p_p;
563 
564 	pool_put(&rusage_pool, p->p_ru);
565 	if (p->p_ptstat)
566 		free(p->p_ptstat, M_SUBPROC);
567 
568 	/*
569 	 * Finally finished with old proc entry.
570 	 * Unlink it from its process group and free it.
571 	 */
572 	if ((p->p_flag & P_THREAD) == 0)
573 		leavepgrp(pr);
574 	LIST_REMOVE(p, p_list);	/* off zombproc */
575 	if ((p->p_flag & P_THREAD) == 0)
576 		LIST_REMOVE(pr, ps_sibling);
577 
578 	/*
579 	 * Decrement the count of procs running with this uid.
580 	 */
581 	(void)chgproccnt(p->p_cred->p_ruid, -1);
582 
583 	/*
584 	 * Release reference to text vnode
585 	 */
586 	if (p->p_textvp)
587 		vrele(p->p_textvp);
588 
589 	/*
590 	 * Remove us from our process list, possibly killing the process
591 	 * in the process (pun intended).
592 	 */
593 	if (--pr->ps_refcnt == 0) {
594 		KASSERT(TAILQ_EMPTY(&pr->ps_threads));
595 		limfree(pr->ps_limit);
596 		crfree(pr->ps_cred->pc_ucred);
597 		pool_put(&pcred_pool, pr->ps_cred);
598 		pool_put(&process_pool, pr);
599 	}
600 
601 	pool_put(&proc_pool, p);
602 	nprocs--;
603 }
604