xref: /openbsd-src/sys/kern/kern_exit.c (revision 7a825137f839261bb446ce207cdf17f8580bf3f6)
1 /*	$OpenBSD: kern_exit.c,v 1.116 2012/07/09 23:06:07 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 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 	/*
249 	 * If emulation has process exit hook, call it now.
250 	 */
251 	if (p->p_emul->e_proc_exit)
252 		(*p->p_emul->e_proc_exit)(p);
253 
254         /*
255          * Remove proc from pidhash chain so looking it up won't
256          * work.  Move it from allproc to zombproc, but do not yet
257          * wake up the reaper.  We will put the proc on the
258          * deadproc list later (using the p_hash member), and
259          * wake up the reaper when we do.
260          */
261 	/*
262 	 * NOTE: WE ARE NO LONGER ALLOWED TO SLEEP!
263 	 */
264 	p->p_stat = SDEAD;
265 
266 	LIST_REMOVE(p, p_hash);
267 	LIST_REMOVE(p, p_list);
268 	LIST_INSERT_HEAD(&zombproc, p, p_list);
269 
270 	/*
271 	 * Give orphaned children to init(8).
272 	 */
273 	if ((p->p_flag & P_THREAD) == 0) {
274 		qr = LIST_FIRST(&pr->ps_children);
275 		if (qr)		/* only need this if any child is S_ZOMB */
276 			wakeup(initproc->p_p);
277 		for (; qr != 0; qr = nqr) {
278 			nqr = LIST_NEXT(qr, ps_sibling);
279 			proc_reparent(qr, initproc->p_p);
280 			/*
281 			 * Traced processes are killed
282 			 * since their existence means someone is screwing up.
283 			 */
284 			if (qr->ps_flags & PS_TRACED) {
285 				atomic_clearbits_int(&qr->ps_flags, PS_TRACED);
286 				/*
287 				 * If single threading is active,
288 				 * direct the signal to the active
289 				 * thread to avoid deadlock.
290 				 */
291 				if (qr->ps_single)
292 					ptsignal(qr->ps_single, SIGKILL,
293 					    STHREAD);
294 				else
295 					prsignal(qr, SIGKILL);
296 			}
297 		}
298 	}
299 
300 
301 	/* add thread's accumulated rusage into the process's total */
302 	ruadd(rup, &p->p_ru);
303 
304 	/*
305 	 * clear %cpu usage during swap
306 	 */
307 	p->p_pctcpu = 0;
308 
309 	if ((p->p_flag & P_THREAD) == 0) {
310 		/*
311 		 * Final thread has died, so add on our children's rusage
312 		 * and calculate the total times
313 		 */
314 		calcru(&pr->ps_tu, &rup->ru_utime, &rup->ru_stime, NULL);
315 		ruadd(rup, &pr->ps_cru);
316 
317 		/* notify interested parties of our demise and clean up */
318 		knote_processexit(pr);
319 
320 		/*
321 		 * Notify parent that we're gone.  If we have P_NOZOMBIE
322 		 * or parent has the SAS_NOCLDWAIT flag set, notify process 1
323 		 * instead (and hope it will handle this situation).
324 		 */
325 		if ((p->p_flag & P_NOZOMBIE) ||
326 		    (pr->ps_pptr->ps_mainproc->p_sigacts->ps_flags &
327 		    SAS_NOCLDWAIT)) {
328 			struct process *ppr = pr->ps_pptr;
329 			proc_reparent(pr, initproc->p_p);
330 			/*
331 			 * If this was the last child of our parent, notify
332 			 * parent, so in case he was wait(2)ing, he will
333 			 * continue.
334 			 */
335 			if (LIST_EMPTY(&ppr->ps_children))
336 				wakeup(ppr);
337 		}
338 	}
339 
340 	/*
341 	 * Release the process's signal state.
342 	 */
343 	sigactsfree(p);
344 
345 	/*
346 	 * Other substructures are freed from reaper and wait().
347 	 */
348 
349 	/*
350 	 * Finally, call machine-dependent code to switch to a new
351 	 * context (possibly the idle context).  Once we are no longer
352 	 * using the dead process's vmspace and stack, exit2() will be
353 	 * called to schedule those resources to be released by the
354 	 * reaper thread.
355 	 *
356 	 * Note that cpu_exit() will end with a call equivalent to
357 	 * cpu_switch(), finishing our execution (pun intended).
358 	 */
359 	uvmexp.swtch++;
360 	cpu_exit(p);
361 	panic("cpu_exit returned");
362 }
363 
364 /*
365  * Locking of this proclist is special; it's accessed in a
366  * critical section of process exit, and thus locking it can't
367  * modify interrupt state.  We use a simple spin lock for this
368  * proclist.  Processes on this proclist are also on zombproc;
369  * we use the p_hash member to linkup to deadproc.
370  */
371 struct mutex deadproc_mutex = MUTEX_INITIALIZER(IPL_NONE);
372 struct proclist deadproc = LIST_HEAD_INITIALIZER(deadproc);
373 
374 /*
375  * We are called from cpu_exit() once it is safe to schedule the
376  * dead process's resources to be freed.
377  *
378  * NOTE: One must be careful with locking in this routine.  It's
379  * called from a critical section in machine-dependent code, so
380  * we should refrain from changing any interrupt state.
381  *
382  * We lock the deadproc list, place the proc on that list (using
383  * the p_hash member), and wake up the reaper.
384  */
385 void
386 exit2(struct proc *p)
387 {
388 	mtx_enter(&deadproc_mutex);
389 	LIST_INSERT_HEAD(&deadproc, p, p_hash);
390 	mtx_leave(&deadproc_mutex);
391 
392 	wakeup(&deadproc);
393 }
394 
395 /*
396  * Process reaper.  This is run by a kernel thread to free the resources
397  * of a dead process.  Once the resources are free, the process becomes
398  * a zombie, and the parent is allowed to read the undead's status.
399  */
400 void
401 reaper(void)
402 {
403 	struct proc *p;
404 
405 	KERNEL_UNLOCK();
406 
407 	SCHED_ASSERT_UNLOCKED();
408 
409 	for (;;) {
410 		mtx_enter(&deadproc_mutex);
411 		while ((p = LIST_FIRST(&deadproc)) == NULL)
412 			msleep(&deadproc, &deadproc_mutex, PVM, "reaper", 0);
413 
414 		/* Remove us from the deadproc list. */
415 		LIST_REMOVE(p, p_hash);
416 		mtx_leave(&deadproc_mutex);
417 
418 		KERNEL_LOCK();
419 
420 		/*
421 		 * Free the VM resources we're still holding on to.
422 		 * We must do this from a valid thread because doing
423 		 * so may block.
424 		 */
425 		uvm_exit(p);
426 
427 		/* Process is now a true zombie. */
428 		if ((p->p_flag & P_NOZOMBIE) == 0) {
429 			p->p_stat = SZOMB;
430 
431 			if (P_EXITSIG(p) != 0)
432 				prsignal(p->p_p->ps_pptr, P_EXITSIG(p));
433 			/* Wake up the parent so it can get exit status. */
434 			wakeup(p->p_p->ps_pptr);
435 		} else {
436 			/* Noone will wait for us. Just zap the process now */
437 			proc_zap(p);
438 		}
439 
440 		KERNEL_UNLOCK();
441 	}
442 }
443 
444 int
445 sys_wait4(struct proc *q, void *v, register_t *retval)
446 {
447 	struct sys_wait4_args /* {
448 		syscallarg(pid_t) pid;
449 		syscallarg(int *) status;
450 		syscallarg(int) options;
451 		syscallarg(struct rusage *) rusage;
452 	} */ *uap = v;
453 	int nfound;
454 	struct process *pr;
455 	struct proc *p;
456 	int status, error;
457 
458 	if (SCARG(uap, pid) == 0)
459 		SCARG(uap, pid) = -q->p_p->ps_pgid;
460 	if (SCARG(uap, options) &~ (WUNTRACED|WNOHANG|WALTSIG|WCONTINUED))
461 		return (EINVAL);
462 
463 loop:
464 	nfound = 0;
465 	LIST_FOREACH(pr, &q->p_p->ps_children, ps_sibling) {
466 		p = pr->ps_mainproc;
467 		if ((p->p_flag & P_NOZOMBIE) ||
468 		    (SCARG(uap, pid) != WAIT_ANY &&
469 		    p->p_pid != SCARG(uap, pid) &&
470 		    pr->ps_pgid != -SCARG(uap, pid)))
471 			continue;
472 
473 		/*
474 		 * Wait for processes with p_exitsig != SIGCHLD processes only
475 		 * if WALTSIG is set; wait for processes with pexitsig ==
476 		 * SIGCHLD only if WALTSIG is clear.
477 		 */
478 		if ((SCARG(uap, options) & WALTSIG) ?
479 		    (p->p_exitsig == SIGCHLD) : (P_EXITSIG(p) != SIGCHLD))
480 			continue;
481 
482 		nfound++;
483 		if (p->p_stat == SZOMB) {
484 			retval[0] = p->p_pid;
485 
486 			if (SCARG(uap, status)) {
487 				status = p->p_xstat;	/* convert to int */
488 				error = copyout(&status,
489 				    SCARG(uap, status), sizeof(status));
490 				if (error)
491 					return (error);
492 			}
493 			if (SCARG(uap, rusage) &&
494 			    (error = copyout(pr->ps_ru,
495 			    SCARG(uap, rusage), sizeof(struct rusage))))
496 				return (error);
497 			proc_finish_wait(q, p);
498 			return (0);
499 		}
500 		if (pr->ps_flags & PS_TRACED &&
501 		    (pr->ps_flags & PS_WAITED) == 0 && pr->ps_single &&
502 		    pr->ps_single->p_stat == SSTOP &&
503 		    (pr->ps_single->p_flag & P_SUSPSINGLE) == 0) {
504 			atomic_setbits_int(&pr->ps_flags, PS_WAITED);
505 			retval[0] = p->p_pid;
506 
507 			if (SCARG(uap, status)) {
508 				status = W_STOPCODE(pr->ps_single->p_xstat);
509 				error = copyout(&status, SCARG(uap, status),
510 				    sizeof(status));
511 			} else
512 				error = 0;
513 			return (error);
514 		}
515 		if (p->p_stat == SSTOP &&
516 		    (pr->ps_flags & PS_WAITED) == 0 &&
517 		    (p->p_flag & P_SUSPSINGLE) == 0 &&
518 		    (pr->ps_flags & PS_TRACED ||
519 		    SCARG(uap, options) & WUNTRACED)) {
520 			atomic_setbits_int(&pr->ps_flags, PS_WAITED);
521 			retval[0] = p->p_pid;
522 
523 			if (SCARG(uap, status)) {
524 				status = W_STOPCODE(p->p_xstat);
525 				error = copyout(&status, SCARG(uap, status),
526 				    sizeof(status));
527 			} else
528 				error = 0;
529 			return (error);
530 		}
531 		if ((SCARG(uap, options) & WCONTINUED) && (p->p_flag & P_CONTINUED)) {
532 			atomic_clearbits_int(&p->p_flag, P_CONTINUED);
533 			retval[0] = p->p_pid;
534 
535 			if (SCARG(uap, status)) {
536 				status = _WCONTINUED;
537 				error = copyout(&status, SCARG(uap, status),
538 				    sizeof(status));
539 			} else
540 				error = 0;
541 			return (error);
542 		}
543 	}
544 	if (nfound == 0)
545 		return (ECHILD);
546 	if (SCARG(uap, options) & WNOHANG) {
547 		retval[0] = 0;
548 		return (0);
549 	}
550 	if ((error = tsleep(q->p_p, PWAIT | PCATCH, "wait", 0)) != 0)
551 		return (error);
552 	goto loop;
553 }
554 
555 void
556 proc_finish_wait(struct proc *waiter, struct proc *p)
557 {
558 	struct process *pr, *tr;
559 	struct rusage *rup;
560 
561 	/*
562 	 * If we got the child via a ptrace 'attach',
563 	 * we need to give it back to the old parent.
564 	 */
565 	pr = p->p_p;
566 	if ((p->p_flag & P_THREAD) == 0 && pr->ps_oppid &&
567 	    (tr = prfind(pr->ps_oppid))) {
568 		atomic_clearbits_int(&pr->ps_flags, PS_TRACED);
569 		pr->ps_oppid = 0;
570 		proc_reparent(pr, tr);
571 		if (p->p_exitsig != 0)
572 			prsignal(tr, p->p_exitsig);
573 		wakeup(tr);
574 	} else {
575 		scheduler_wait_hook(waiter, p);
576 		p->p_xstat = 0;
577 		rup = &waiter->p_p->ps_cru;
578 		ruadd(rup, pr->ps_ru);
579 		proc_zap(p);
580 	}
581 }
582 
583 /*
584  * make process 'parent' the new parent of process 'child'.
585  */
586 void
587 proc_reparent(struct process *child, struct process *parent)
588 {
589 
590 	if (child->ps_pptr == parent)
591 		return;
592 
593 	if (parent == initproc->p_p)
594 		child->ps_mainproc->p_exitsig = SIGCHLD;
595 
596 	LIST_REMOVE(child, ps_sibling);
597 	LIST_INSERT_HEAD(&parent->ps_children, child, ps_sibling);
598 	child->ps_pptr = parent;
599 }
600 
601 void
602 proc_zap(struct proc *p)
603 {
604 	struct process *pr = p->p_p;
605 
606 	/*
607 	 * Finally finished with old proc entry.
608 	 * Unlink it from its process group and free it.
609 	 */
610 	if ((p->p_flag & P_THREAD) == 0)
611 		leavepgrp(pr);
612 	LIST_REMOVE(p, p_list);	/* off zombproc */
613 	if ((p->p_flag & P_THREAD) == 0) {
614 		LIST_REMOVE(pr, ps_sibling);
615 
616 		/*
617 		 * Decrement the count of procs running with this uid.
618 		 */
619 		(void)chgproccnt(p->p_cred->p_ruid, -1);
620 	}
621 
622 	/*
623 	 * Release reference to text vnode
624 	 */
625 	if (p->p_textvp)
626 		vrele(p->p_textvp);
627 
628 	/*
629 	 * Remove us from our process list, possibly killing the process
630 	 * in the process (pun intended).
631 	 */
632 	if (--pr->ps_refcnt == 0) {
633 		if (pr->ps_ptstat != NULL)
634 			free(pr->ps_ptstat, M_SUBPROC);
635 		pool_put(&rusage_pool, pr->ps_ru);
636 		KASSERT(TAILQ_EMPTY(&pr->ps_threads));
637 		limfree(pr->ps_limit);
638 		crfree(pr->ps_cred->pc_ucred);
639 		pool_put(&pcred_pool, pr->ps_cred);
640 		pool_put(&process_pool, pr);
641 		nprocesses--;
642 	}
643 
644 	pool_put(&proc_pool, p);
645 	nthreads--;
646 }
647