xref: /openbsd-src/sys/kern/kern_exit.c (revision 458bf05c8809c4e815fe1c53afb84eec1ec811a6)
1 /*	$OpenBSD: kern_exit.c,v 1.97 2010/08/02 19:54:07 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, P_PPWAIT
166 		 * is set; we wake up the parent early to avoid deadlock.
167 		 */
168 		if (p->p_flag & P_PPWAIT) {
169 			atomic_clearbits_int(&p->p_flag, P_PPWAIT);
170 			wakeup(pr->ps_pptr);
171 		}
172 	}
173 
174 	if (p->p_flag & P_PROFIL)
175 		stopprofclock(p);
176 	p->p_ru = pool_get(&rusage_pool, PR_WAITOK);
177 	p->p_sigignore = ~0;
178 	p->p_siglist = 0;
179 	timeout_del(&p->p_realit_to);
180 	timeout_del(&p->p_stats->p_virt_to);
181 	timeout_del(&p->p_stats->p_prof_to);
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 #ifdef SYSVSEM
190 		semexit(pr);
191 #endif
192 		if (SESS_LEADER(pr)) {
193 			struct session *sp = pr->ps_session;
194 
195 			if (sp->s_ttyvp) {
196 				/*
197 				 * Controlling process.
198 				 * Signal foreground pgrp,
199 				 * drain controlling terminal
200 				 * and revoke access to controlling terminal.
201 				 */
202 				if (sp->s_ttyp->t_session == sp) {
203 					if (sp->s_ttyp->t_pgrp)
204 						pgsignal(sp->s_ttyp->t_pgrp,
205 						    SIGHUP, 1);
206 					(void) ttywait(sp->s_ttyp);
207 					/*
208 					 * The tty could have been revoked
209 					 * if we blocked.
210 					 */
211 					if (sp->s_ttyvp)
212 						VOP_REVOKE(sp->s_ttyvp,
213 						    REVOKEALL);
214 				}
215 				if (sp->s_ttyvp)
216 					vrele(sp->s_ttyvp);
217 				sp->s_ttyvp = NULL;
218 				/*
219 				 * s_ttyp is not zero'd; we use this to
220 				 * indicate that the session once had a
221 				 * controlling terminal.  (for logging and
222 				 * informational purposes)
223 				 */
224 			}
225 			sp->s_leader = NULL;
226 		}
227 		fixjobc(pr, pr->ps_pgrp, 0);
228 
229 #ifdef ACCOUNTING
230 		(void)acct_process(p);
231 #endif
232 	}
233 
234 #ifdef KTRACE
235 	/*
236 	 * release trace file
237 	 */
238 	p->p_traceflag = 0;	/* don't trace the vrele() */
239 	if (p->p_tracep)
240 		ktrsettracevnode(p, NULL);
241 #endif
242 #if NSYSTRACE > 0
243 	if (ISSET(p->p_flag, P_SYSTRACE))
244 		systrace_exit(p);
245 #endif
246         /*
247          * Remove proc from pidhash chain so looking it up won't
248          * work.  Move it from allproc to zombproc, but do not yet
249          * wake up the reaper.  We will put the proc on the
250          * deadproc list later (using the p_hash member), and
251          * wake up the reaper when we do.
252          */
253 	/*
254 	 * NOTE: WE ARE NO LONGER ALLOWED TO SLEEP!
255 	 */
256 	p->p_stat = SDEAD;
257 
258 	LIST_REMOVE(p, p_hash);
259 	LIST_REMOVE(p, p_list);
260 	LIST_INSERT_HEAD(&zombproc, p, p_list);
261 
262 	/*
263 	 * Give orphaned children to init(8).
264 	 */
265 	if ((p->p_flag & P_THREAD) == 0) {
266 		qr = LIST_FIRST(&pr->ps_children);
267 		if (qr)		/* only need this if any child is S_ZOMB */
268 			wakeup(initproc->p_p);
269 		for (; qr != 0; qr = nqr) {
270 			nqr = LIST_NEXT(qr, ps_sibling);
271 			proc_reparent(qr, initproc->p_p);
272 			/*
273 			 * Traced processes are killed
274 			 * since their existence means someone is screwing up.
275 			 */
276 			if (qr->ps_mainproc->p_flag & P_TRACED) {
277 				atomic_clearbits_int(&qr->ps_mainproc->p_flag,
278 				    P_TRACED);
279 				prsignal(qr, SIGKILL);
280 			}
281 		}
282 	}
283 
284 
285 	/*
286 	 * Save exit status and final rusage info, adding in child rusage
287 	 * info and self times.
288 	 */
289 	if (!(p->p_flag & P_IGNEXITRV))
290 		p->p_xstat = rv;
291 	*p->p_ru = p->p_stats->p_ru;
292 	calcru(p, &p->p_ru->ru_utime, &p->p_ru->ru_stime, NULL);
293 	ruadd(p->p_ru, &p->p_stats->p_cru);
294 
295 	/*
296 	 * clear %cpu usage during swap
297 	 */
298 	p->p_pctcpu = 0;
299 
300 	if ((p->p_flag & P_THREAD) == 0) {
301 		/* notify interested parties of our demise and clean up */
302 		knote_processexit(pr);
303 
304 		/*
305 		 * Notify parent that we're gone.  If we have P_NOZOMBIE
306 		 * or parent has the P_NOCLDWAIT flag set, notify process 1
307 		 * instead (and hope it will handle this situation).
308 		 */
309 		if ((p->p_flag & P_NOZOMBIE) ||
310 		    (pr->ps_pptr->ps_mainproc->p_flag & P_NOCLDWAIT)) {
311 			struct process *ppr = pr->ps_pptr;
312 			proc_reparent(pr, initproc->p_p);
313 			/*
314 			 * If this was the last child of our parent, notify
315 			 * parent, so in case he was wait(2)ing, he will
316 			 * continue.
317 			 */
318 			if (LIST_EMPTY(&ppr->ps_children))
319 				wakeup(ppr);
320 		}
321 	}
322 
323 	/*
324 	 * Release the process's signal state.
325 	 */
326 	sigactsfree(p);
327 
328 	/*
329 	 * Other substructures are freed from reaper and wait().
330 	 */
331 
332 	/*
333 	 * If emulation has process exit hook, call it now.
334 	 */
335 	if (p->p_emul->e_proc_exit)
336 		(*p->p_emul->e_proc_exit)(p);
337 
338 	/*
339 	 * Finally, call machine-dependent code to switch to a new
340 	 * context (possibly the idle context).  Once we are no longer
341 	 * using the dead process's vmspace and stack, exit2() will be
342 	 * called to schedule those resources to be released by the
343 	 * reaper thread.
344 	 *
345 	 * Note that cpu_exit() will end with a call equivalent to
346 	 * cpu_switch(), finishing our execution (pun intended).
347 	 */
348 	uvmexp.swtch++;
349 	cpu_exit(p);
350 	panic("cpu_exit returned");
351 }
352 
353 /*
354  * Locking of this proclist is special; it's accessed in a
355  * critical section of process exit, and thus locking it can't
356  * modify interrupt state.  We use a simple spin lock for this
357  * proclist.  Processes on this proclist are also on zombproc;
358  * we use the p_hash member to linkup to deadproc.
359  */
360 struct mutex deadproc_mutex = MUTEX_INITIALIZER(IPL_NONE);
361 struct proclist deadproc = LIST_HEAD_INITIALIZER(deadproc);
362 
363 /*
364  * We are called from cpu_exit() once it is safe to schedule the
365  * dead process's resources to be freed.
366  *
367  * NOTE: One must be careful with locking in this routine.  It's
368  * called from a critical section in machine-dependent code, so
369  * we should refrain from changing any interrupt state.
370  *
371  * We lock the deadproc list, place the proc on that list (using
372  * the p_hash member), and wake up the reaper.
373  */
374 void
375 exit2(struct proc *p)
376 {
377 	mtx_enter(&deadproc_mutex);
378 	LIST_INSERT_HEAD(&deadproc, p, p_hash);
379 	mtx_leave(&deadproc_mutex);
380 
381 	wakeup(&deadproc);
382 }
383 
384 /*
385  * Process reaper.  This is run by a kernel thread to free the resources
386  * of a dead process.  Once the resources are free, the process becomes
387  * a zombie, and the parent is allowed to read the undead's status.
388  */
389 void
390 reaper(void)
391 {
392 	struct proc *p;
393 
394 	KERNEL_PROC_UNLOCK(curproc);
395 
396 	SCHED_ASSERT_UNLOCKED();
397 
398 	for (;;) {
399 		mtx_enter(&deadproc_mutex);
400 		while ((p = LIST_FIRST(&deadproc)) == NULL)
401 			msleep(&deadproc, &deadproc_mutex, PVM, "reaper", 0);
402 
403 		/* Remove us from the deadproc list. */
404 		LIST_REMOVE(p, p_hash);
405 		mtx_leave(&deadproc_mutex);
406 
407 		KERNEL_PROC_LOCK(curproc);
408 
409 		/*
410 		 * Free the VM resources we're still holding on to.
411 		 * We must do this from a valid thread because doing
412 		 * so may block.
413 		 */
414 		uvm_exit(p);
415 
416 		/* Process is now a true zombie. */
417 		if ((p->p_flag & P_NOZOMBIE) == 0) {
418 			p->p_stat = SZOMB;
419 
420 			if (P_EXITSIG(p) != 0)
421 				prsignal(p->p_p->ps_pptr, P_EXITSIG(p));
422 			/* Wake up the parent so it can get exit status. */
423 			wakeup(p->p_p->ps_pptr);
424 		} else {
425 			/* Noone will wait for us. Just zap the process now */
426 			proc_zap(p);
427 		}
428 
429 		KERNEL_PROC_UNLOCK(curproc);
430 	}
431 }
432 
433 pid_t
434 sys_wait4(struct proc *q, void *v, register_t *retval)
435 {
436 	struct sys_wait4_args /* {
437 		syscallarg(pid_t) pid;
438 		syscallarg(int *) status;
439 		syscallarg(int) options;
440 		syscallarg(struct rusage *) rusage;
441 	} */ *uap = v;
442 	int nfound;
443 	struct process *pr;
444 	struct proc *p;
445 	int status, error;
446 
447 	if (SCARG(uap, pid) == 0)
448 		SCARG(uap, pid) = -q->p_p->ps_pgid;
449 	if (SCARG(uap, options) &~ (WUNTRACED|WNOHANG|WALTSIG|WCONTINUED))
450 		return (EINVAL);
451 
452 loop:
453 	nfound = 0;
454 	LIST_FOREACH(pr, &q->p_p->ps_children, ps_sibling) {
455 		p = pr->ps_mainproc;
456 		if ((p->p_flag & P_NOZOMBIE) ||
457 		    (SCARG(uap, pid) != WAIT_ANY &&
458 		    p->p_pid != SCARG(uap, pid) &&
459 		    pr->ps_pgid != -SCARG(uap, pid)))
460 			continue;
461 
462 		/*
463 		 * Wait for processes with p_exitsig != SIGCHLD processes only
464 		 * if WALTSIG is set; wait for processes with pexitsig ==
465 		 * SIGCHLD only if WALTSIG is clear.
466 		 */
467 		if ((SCARG(uap, options) & WALTSIG) ?
468 		    (p->p_exitsig == SIGCHLD) : (P_EXITSIG(p) != SIGCHLD))
469 			continue;
470 
471 		nfound++;
472 		if (p->p_stat == SZOMB) {
473 			retval[0] = p->p_pid;
474 
475 			if (SCARG(uap, status)) {
476 				status = p->p_xstat;	/* convert to int */
477 				error = copyout(&status,
478 				    SCARG(uap, status), sizeof(status));
479 				if (error)
480 					return (error);
481 			}
482 			if (SCARG(uap, rusage) &&
483 			    (error = copyout(p->p_ru,
484 			    SCARG(uap, rusage), sizeof(struct rusage))))
485 				return (error);
486 			proc_finish_wait(q, p);
487 			return (0);
488 		}
489 		if (p->p_stat == SSTOP && (p->p_flag & P_WAITED) == 0 &&
490 		    (p->p_flag & P_TRACED || SCARG(uap, options) & WUNTRACED)) {
491 			atomic_setbits_int(&p->p_flag, P_WAITED);
492 			retval[0] = p->p_pid;
493 
494 			if (SCARG(uap, status)) {
495 				status = W_STOPCODE(p->p_xstat);
496 				error = copyout(&status, SCARG(uap, status),
497 				    sizeof(status));
498 			} else
499 				error = 0;
500 			return (error);
501 		}
502 		if ((SCARG(uap, options) & WCONTINUED) && (p->p_flag & P_CONTINUED)) {
503 			atomic_clearbits_int(&p->p_flag, P_CONTINUED);
504 			retval[0] = p->p_pid;
505 
506 			if (SCARG(uap, status)) {
507 				status = _WCONTINUED;
508 				error = copyout(&status, SCARG(uap, status),
509 				    sizeof(status));
510 			} else
511 				error = 0;
512 			return (error);
513 		}
514 	}
515 	if (nfound == 0)
516 		return (ECHILD);
517 	if (SCARG(uap, options) & WNOHANG) {
518 		retval[0] = 0;
519 		return (0);
520 	}
521 	if ((error = tsleep(q->p_p, PWAIT | PCATCH, "wait", 0)) != 0)
522 		return (error);
523 	goto loop;
524 }
525 
526 void
527 proc_finish_wait(struct proc *waiter, struct proc *p)
528 {
529 	struct process *tr;
530 
531 	/*
532 	 * If we got the child via a ptrace 'attach',
533 	 * we need to give it back to the old parent.
534 	 */
535 	if (p->p_oppid && (tr = prfind(p->p_oppid))) {
536 		atomic_clearbits_int(&p->p_flag, P_TRACED);
537 		p->p_oppid = 0;
538 		proc_reparent(p->p_p, tr);
539 		if (p->p_exitsig != 0)
540 			prsignal(tr, p->p_exitsig);
541 		wakeup(tr);
542 	} else {
543 		scheduler_wait_hook(waiter, p);
544 		p->p_xstat = 0;
545 		ruadd(&waiter->p_stats->p_cru, p->p_ru);
546 		proc_zap(p);
547 	}
548 }
549 
550 /*
551  * make process 'parent' the new parent of process 'child'.
552  */
553 void
554 proc_reparent(struct process *child, struct process *parent)
555 {
556 
557 	if (child->ps_pptr == parent)
558 		return;
559 
560 	if (parent == initproc->p_p)
561 		child->ps_mainproc->p_exitsig = SIGCHLD;
562 
563 	LIST_REMOVE(child, ps_sibling);
564 	LIST_INSERT_HEAD(&parent->ps_children, child, ps_sibling);
565 	child->ps_pptr = parent;
566 }
567 
568 void
569 proc_zap(struct proc *p)
570 {
571 	struct process *pr = p->p_p;
572 
573 	pool_put(&rusage_pool, p->p_ru);
574 	if (p->p_ptstat)
575 		free(p->p_ptstat, M_SUBPROC);
576 
577 	/*
578 	 * Finally finished with old proc entry.
579 	 * Unlink it from its process group and free it.
580 	 */
581 	if ((p->p_flag & P_THREAD) == 0)
582 		leavepgrp(pr);
583 	LIST_REMOVE(p, p_list);	/* off zombproc */
584 	if ((p->p_flag & P_THREAD) == 0)
585 		LIST_REMOVE(pr, ps_sibling);
586 
587 	/*
588 	 * Decrement the count of procs running with this uid.
589 	 */
590 	(void)chgproccnt(p->p_cred->p_ruid, -1);
591 
592 	/*
593 	 * Release reference to text vnode
594 	 */
595 	if (p->p_textvp)
596 		vrele(p->p_textvp);
597 
598 	/*
599 	 * Remove us from our process list, possibly killing the process
600 	 * in the process (pun intended).
601 	 */
602 	if (--pr->ps_refcnt == 0) {
603 		KASSERT(TAILQ_EMPTY(&pr->ps_threads));
604 		limfree(pr->ps_limit);
605 		crfree(pr->ps_cred->pc_ucred);
606 		pool_put(&pcred_pool, pr->ps_cred);
607 		pool_put(&process_pool, pr);
608 	}
609 
610 	pool_put(&proc_pool, p);
611 	nprocs--;
612 }
613