xref: /openbsd-src/sys/kern/kern_exit.c (revision 4ec195400d023537d947ee6557938900b7cc7161)
1 /*	$OpenBSD: kern_exit.c,v 1.113 2012/04/13 16:37:51 kettenis 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 (ISSET(p->p_flag, P_SUSPSINGLE)) {
162 		if (--pr->ps_singlecount == 0)
163 			wakeup(&pr->ps_singlecount);
164 	}
165 	if ((p->p_flag & P_THREAD) == 0) {
166 		/* main thread gotta wait because it has the pid, et al */
167 		while (! TAILQ_EMPTY(&pr->ps_threads))
168 			tsleep(&pr->ps_threads, PUSER, "thrdeath", 0);
169 	} else if (TAILQ_EMPTY(&pr->ps_threads))
170 		wakeup(&pr->ps_threads);
171 
172 	if (p->p_flag & P_PROFIL)
173 		stopprofclock(p);
174 	rup = pr->ps_ru;
175 	if (rup == NULL) {
176 		rup = pool_get(&rusage_pool, PR_WAITOK | PR_ZERO);
177 
178 		if (pr->ps_ru == NULL)
179 			pr->ps_ru = rup;
180 		else {
181 			pool_put(&rusage_pool, rup);
182 			rup = pr->ps_ru;
183 		}
184 	}
185 	p->p_siglist = 0;
186 
187 	/*
188 	 * Close open files and release open-file table.
189 	 */
190 	fdfree(p);
191 
192 	if ((p->p_flag & P_THREAD) == 0) {
193 		timeout_del(&pr->ps_realit_to);
194 		timeout_del(&pr->ps_virt_to);
195 		timeout_del(&pr->ps_prof_to);
196 #ifdef SYSVSEM
197 		semexit(pr);
198 #endif
199 		if (SESS_LEADER(pr)) {
200 			struct session *sp = pr->ps_session;
201 
202 			if (sp->s_ttyvp) {
203 				/*
204 				 * Controlling process.
205 				 * Signal foreground pgrp,
206 				 * drain controlling terminal
207 				 * and revoke access to controlling terminal.
208 				 */
209 				if (sp->s_ttyp->t_session == sp) {
210 					if (sp->s_ttyp->t_pgrp)
211 						pgsignal(sp->s_ttyp->t_pgrp,
212 						    SIGHUP, 1);
213 					(void) ttywait(sp->s_ttyp);
214 					/*
215 					 * The tty could have been revoked
216 					 * if we blocked.
217 					 */
218 					if (sp->s_ttyvp)
219 						VOP_REVOKE(sp->s_ttyvp,
220 						    REVOKEALL);
221 				}
222 				if (sp->s_ttyvp)
223 					vrele(sp->s_ttyvp);
224 				sp->s_ttyvp = NULL;
225 				/*
226 				 * s_ttyp is not zero'd; we use this to
227 				 * indicate that the session once had a
228 				 * controlling terminal.  (for logging and
229 				 * informational purposes)
230 				 */
231 			}
232 			sp->s_leader = NULL;
233 		}
234 		fixjobc(pr, pr->ps_pgrp, 0);
235 
236 #ifdef ACCOUNTING
237 		(void)acct_process(p);
238 #endif
239 
240 #ifdef KTRACE
241 		/* release trace file */
242 		if (pr->ps_tracevp)
243 			ktrcleartrace(pr);
244 #endif
245 	}
246 
247 #if NSYSTRACE > 0
248 	if (ISSET(p->p_flag, P_SYSTRACE))
249 		systrace_exit(p);
250 #endif
251         /*
252          * Remove proc from pidhash chain so looking it up won't
253          * work.  Move it from allproc to zombproc, but do not yet
254          * wake up the reaper.  We will put the proc on the
255          * deadproc list later (using the p_hash member), and
256          * wake up the reaper when we do.
257          */
258 	/*
259 	 * NOTE: WE ARE NO LONGER ALLOWED TO SLEEP!
260 	 */
261 	p->p_stat = SDEAD;
262 
263 	LIST_REMOVE(p, p_hash);
264 	LIST_REMOVE(p, p_list);
265 	LIST_INSERT_HEAD(&zombproc, p, p_list);
266 
267 	/*
268 	 * Give orphaned children to init(8).
269 	 */
270 	if ((p->p_flag & P_THREAD) == 0) {
271 		qr = LIST_FIRST(&pr->ps_children);
272 		if (qr)		/* only need this if any child is S_ZOMB */
273 			wakeup(initproc->p_p);
274 		for (; qr != 0; qr = nqr) {
275 			nqr = LIST_NEXT(qr, ps_sibling);
276 			proc_reparent(qr, initproc->p_p);
277 			/*
278 			 * Traced processes are killed
279 			 * since their existence means someone is screwing up.
280 			 */
281 			if (qr->ps_flags & PS_TRACED) {
282 				atomic_clearbits_int(&qr->ps_flags, PS_TRACED);
283 				prsignal(qr, SIGKILL);
284 			}
285 		}
286 	}
287 
288 
289 	/* add thread's accumulated rusage into the process's total */
290 	ruadd(rup, &p->p_ru);
291 
292 	/*
293 	 * clear %cpu usage during swap
294 	 */
295 	p->p_pctcpu = 0;
296 
297 	if ((p->p_flag & P_THREAD) == 0) {
298 		/*
299 		 * Final thread has died, so add on our children's rusage
300 		 * and calculate the total times
301 		 */
302 		calcru(&pr->ps_tu, &rup->ru_utime, &rup->ru_stime, NULL);
303 		ruadd(rup, &pr->ps_cru);
304 
305 		/* notify interested parties of our demise and clean up */
306 		knote_processexit(pr);
307 
308 		/*
309 		 * Notify parent that we're gone.  If we have P_NOZOMBIE
310 		 * or parent has the SAS_NOCLDWAIT flag set, notify process 1
311 		 * instead (and hope it will handle this situation).
312 		 */
313 		if ((p->p_flag & P_NOZOMBIE) ||
314 		    (pr->ps_pptr->ps_mainproc->p_sigacts->ps_flags &
315 		    SAS_NOCLDWAIT)) {
316 			struct process *ppr = pr->ps_pptr;
317 			proc_reparent(pr, initproc->p_p);
318 			/*
319 			 * If this was the last child of our parent, notify
320 			 * parent, so in case he was wait(2)ing, he will
321 			 * continue.
322 			 */
323 			if (LIST_EMPTY(&ppr->ps_children))
324 				wakeup(ppr);
325 		}
326 	}
327 
328 	/*
329 	 * Release the process's signal state.
330 	 */
331 	sigactsfree(p);
332 
333 	/*
334 	 * Other substructures are freed from reaper and wait().
335 	 */
336 
337 	/*
338 	 * If emulation has process exit hook, call it now.
339 	 */
340 	if (p->p_emul->e_proc_exit)
341 		(*p->p_emul->e_proc_exit)(p);
342 
343 	/*
344 	 * Finally, call machine-dependent code to switch to a new
345 	 * context (possibly the idle context).  Once we are no longer
346 	 * using the dead process's vmspace and stack, exit2() will be
347 	 * called to schedule those resources to be released by the
348 	 * reaper thread.
349 	 *
350 	 * Note that cpu_exit() will end with a call equivalent to
351 	 * cpu_switch(), finishing our execution (pun intended).
352 	 */
353 	uvmexp.swtch++;
354 	cpu_exit(p);
355 	panic("cpu_exit returned");
356 }
357 
358 /*
359  * Locking of this proclist is special; it's accessed in a
360  * critical section of process exit, and thus locking it can't
361  * modify interrupt state.  We use a simple spin lock for this
362  * proclist.  Processes on this proclist are also on zombproc;
363  * we use the p_hash member to linkup to deadproc.
364  */
365 struct mutex deadproc_mutex = MUTEX_INITIALIZER(IPL_NONE);
366 struct proclist deadproc = LIST_HEAD_INITIALIZER(deadproc);
367 
368 /*
369  * We are called from cpu_exit() once it is safe to schedule the
370  * dead process's resources to be freed.
371  *
372  * NOTE: One must be careful with locking in this routine.  It's
373  * called from a critical section in machine-dependent code, so
374  * we should refrain from changing any interrupt state.
375  *
376  * We lock the deadproc list, place the proc on that list (using
377  * the p_hash member), and wake up the reaper.
378  */
379 void
380 exit2(struct proc *p)
381 {
382 	mtx_enter(&deadproc_mutex);
383 	LIST_INSERT_HEAD(&deadproc, p, p_hash);
384 	mtx_leave(&deadproc_mutex);
385 
386 	wakeup(&deadproc);
387 }
388 
389 /*
390  * Process reaper.  This is run by a kernel thread to free the resources
391  * of a dead process.  Once the resources are free, the process becomes
392  * a zombie, and the parent is allowed to read the undead's status.
393  */
394 void
395 reaper(void)
396 {
397 	struct proc *p;
398 
399 	KERNEL_UNLOCK();
400 
401 	SCHED_ASSERT_UNLOCKED();
402 
403 	for (;;) {
404 		mtx_enter(&deadproc_mutex);
405 		while ((p = LIST_FIRST(&deadproc)) == NULL)
406 			msleep(&deadproc, &deadproc_mutex, PVM, "reaper", 0);
407 
408 		/* Remove us from the deadproc list. */
409 		LIST_REMOVE(p, p_hash);
410 		mtx_leave(&deadproc_mutex);
411 
412 		KERNEL_LOCK();
413 
414 		/*
415 		 * Free the VM resources we're still holding on to.
416 		 * We must do this from a valid thread because doing
417 		 * so may block.
418 		 */
419 		uvm_exit(p);
420 
421 		/* Process is now a true zombie. */
422 		if ((p->p_flag & P_NOZOMBIE) == 0) {
423 			p->p_stat = SZOMB;
424 
425 			if (P_EXITSIG(p) != 0)
426 				prsignal(p->p_p->ps_pptr, P_EXITSIG(p));
427 			/* Wake up the parent so it can get exit status. */
428 			wakeup(p->p_p->ps_pptr);
429 		} else {
430 			/* Noone will wait for us. Just zap the process now */
431 			proc_zap(p);
432 		}
433 
434 		KERNEL_UNLOCK();
435 	}
436 }
437 
438 int
439 sys_wait4(struct proc *q, void *v, register_t *retval)
440 {
441 	struct sys_wait4_args /* {
442 		syscallarg(pid_t) pid;
443 		syscallarg(int *) status;
444 		syscallarg(int) options;
445 		syscallarg(struct rusage *) rusage;
446 	} */ *uap = v;
447 	int nfound;
448 	struct process *pr;
449 	struct proc *p;
450 	int status, error;
451 
452 	if (SCARG(uap, pid) == 0)
453 		SCARG(uap, pid) = -q->p_p->ps_pgid;
454 	if (SCARG(uap, options) &~ (WUNTRACED|WNOHANG|WALTSIG|WCONTINUED))
455 		return (EINVAL);
456 
457 loop:
458 	nfound = 0;
459 	LIST_FOREACH(pr, &q->p_p->ps_children, ps_sibling) {
460 		p = pr->ps_mainproc;
461 		if ((p->p_flag & P_NOZOMBIE) ||
462 		    (SCARG(uap, pid) != WAIT_ANY &&
463 		    p->p_pid != SCARG(uap, pid) &&
464 		    pr->ps_pgid != -SCARG(uap, pid)))
465 			continue;
466 
467 		/*
468 		 * Wait for processes with p_exitsig != SIGCHLD processes only
469 		 * if WALTSIG is set; wait for processes with pexitsig ==
470 		 * SIGCHLD only if WALTSIG is clear.
471 		 */
472 		if ((SCARG(uap, options) & WALTSIG) ?
473 		    (p->p_exitsig == SIGCHLD) : (P_EXITSIG(p) != SIGCHLD))
474 			continue;
475 
476 		nfound++;
477 		if (p->p_stat == SZOMB) {
478 			retval[0] = p->p_pid;
479 
480 			if (SCARG(uap, status)) {
481 				status = p->p_xstat;	/* convert to int */
482 				error = copyout(&status,
483 				    SCARG(uap, status), sizeof(status));
484 				if (error)
485 					return (error);
486 			}
487 			if (SCARG(uap, rusage) &&
488 			    (error = copyout(pr->ps_ru,
489 			    SCARG(uap, rusage), sizeof(struct rusage))))
490 				return (error);
491 			proc_finish_wait(q, p);
492 			return (0);
493 		}
494 		if (pr->ps_flags & PS_TRACED &&
495 		    (pr->ps_flags & PS_WAITED) == 0 && pr->ps_single &&
496 		    pr->ps_single->p_stat == SSTOP &&
497 		    (pr->ps_single->p_flag & P_SUSPSINGLE) == 0) {
498 			atomic_setbits_int(&pr->ps_flags, PS_WAITED);
499 			retval[0] = p->p_pid;
500 
501 			if (SCARG(uap, status)) {
502 				status = W_STOPCODE(pr->ps_single->p_xstat);
503 				error = copyout(&status, SCARG(uap, status),
504 				    sizeof(status));
505 			} else
506 				error = 0;
507 			return (error);
508 		}
509 		if (p->p_stat == SSTOP &&
510 		    (pr->ps_flags & PS_WAITED) == 0 &&
511 		    (p->p_flag & P_SUSPSINGLE) == 0 &&
512 		    (pr->ps_flags & PS_TRACED ||
513 		    SCARG(uap, options) & WUNTRACED)) {
514 			atomic_setbits_int(&pr->ps_flags, PS_WAITED);
515 			retval[0] = p->p_pid;
516 
517 			if (SCARG(uap, status)) {
518 				status = W_STOPCODE(p->p_xstat);
519 				error = copyout(&status, SCARG(uap, status),
520 				    sizeof(status));
521 			} else
522 				error = 0;
523 			return (error);
524 		}
525 		if ((SCARG(uap, options) & WCONTINUED) && (p->p_flag & P_CONTINUED)) {
526 			atomic_clearbits_int(&p->p_flag, P_CONTINUED);
527 			retval[0] = p->p_pid;
528 
529 			if (SCARG(uap, status)) {
530 				status = _WCONTINUED;
531 				error = copyout(&status, SCARG(uap, status),
532 				    sizeof(status));
533 			} else
534 				error = 0;
535 			return (error);
536 		}
537 	}
538 	if (nfound == 0)
539 		return (ECHILD);
540 	if (SCARG(uap, options) & WNOHANG) {
541 		retval[0] = 0;
542 		return (0);
543 	}
544 	if ((error = tsleep(q->p_p, PWAIT | PCATCH, "wait", 0)) != 0)
545 		return (error);
546 	goto loop;
547 }
548 
549 void
550 proc_finish_wait(struct proc *waiter, struct proc *p)
551 {
552 	struct process *pr, *tr;
553 	struct rusage *rup;
554 
555 	/*
556 	 * If we got the child via a ptrace 'attach',
557 	 * we need to give it back to the old parent.
558 	 */
559 	pr = p->p_p;
560 	if ((p->p_flag & P_THREAD) == 0 && pr->ps_oppid &&
561 	    (tr = prfind(pr->ps_oppid))) {
562 		atomic_clearbits_int(&pr->ps_flags, PS_TRACED);
563 		pr->ps_oppid = 0;
564 		proc_reparent(pr, tr);
565 		if (p->p_exitsig != 0)
566 			prsignal(tr, p->p_exitsig);
567 		wakeup(tr);
568 	} else {
569 		scheduler_wait_hook(waiter, p);
570 		p->p_xstat = 0;
571 		rup = &waiter->p_p->ps_cru;
572 		ruadd(rup, pr->ps_ru);
573 		proc_zap(p);
574 	}
575 }
576 
577 /*
578  * make process 'parent' the new parent of process 'child'.
579  */
580 void
581 proc_reparent(struct process *child, struct process *parent)
582 {
583 
584 	if (child->ps_pptr == parent)
585 		return;
586 
587 	if (parent == initproc->p_p)
588 		child->ps_mainproc->p_exitsig = SIGCHLD;
589 
590 	LIST_REMOVE(child, ps_sibling);
591 	LIST_INSERT_HEAD(&parent->ps_children, child, ps_sibling);
592 	child->ps_pptr = parent;
593 }
594 
595 void
596 proc_zap(struct proc *p)
597 {
598 	struct process *pr = p->p_p;
599 
600 	/*
601 	 * Finally finished with old proc entry.
602 	 * Unlink it from its process group and free it.
603 	 */
604 	if ((p->p_flag & P_THREAD) == 0)
605 		leavepgrp(pr);
606 	LIST_REMOVE(p, p_list);	/* off zombproc */
607 	if ((p->p_flag & P_THREAD) == 0) {
608 		LIST_REMOVE(pr, ps_sibling);
609 
610 		/*
611 		 * Decrement the count of procs running with this uid.
612 		 */
613 		(void)chgproccnt(p->p_cred->p_ruid, -1);
614 	}
615 
616 	/*
617 	 * Release reference to text vnode
618 	 */
619 	if (p->p_textvp)
620 		vrele(p->p_textvp);
621 
622 	/*
623 	 * Remove us from our process list, possibly killing the process
624 	 * in the process (pun intended).
625 	 */
626 	if (--pr->ps_refcnt == 0) {
627 		if (pr->ps_ptstat != NULL)
628 			free(pr->ps_ptstat, M_SUBPROC);
629 		pool_put(&rusage_pool, pr->ps_ru);
630 		KASSERT(TAILQ_EMPTY(&pr->ps_threads));
631 		limfree(pr->ps_limit);
632 		crfree(pr->ps_cred->pc_ucred);
633 		pool_put(&pcred_pool, pr->ps_cred);
634 		pool_put(&process_pool, pr);
635 		nprocesses--;
636 	}
637 
638 	pool_put(&proc_pool, p);
639 	nthreads--;
640 }
641