xref: /openbsd-src/sys/kern/kern_exit.c (revision 121cb9296653d77a6eefbffed970253e2d58e97c)
1 /*	$OpenBSD: kern_exit.c,v 1.119 2012/09/08 14:52:00 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 		else if (flags == EXIT_THREAD)
142 			single_thread_check(p, 0);
143 	}
144 
145 	if (flags == EXIT_NORMAL) {
146 		atomic_setbits_int(&pr->ps_flags, PS_EXITING);
147 		pr->ps_mainproc->p_xstat = rv;
148 
149 		/*
150 		 * If parent is waiting for us to exit or exec, PS_PPWAIT
151 		 * is set; we wake up the parent early to avoid deadlock.
152 		 */
153 		if (pr->ps_flags & PS_PPWAIT) {
154 			atomic_clearbits_int(&pr->ps_flags, PS_PPWAIT);
155 			atomic_clearbits_int(&pr->ps_pptr->ps_flags,
156 			    PS_ISPWAIT);
157 			wakeup(pr->ps_pptr);
158 		}
159 	}
160 
161 	/* unlink ourselves from the active threads */
162 	TAILQ_REMOVE(&pr->ps_threads, p, p_thr_link);
163 	if ((p->p_flag & P_THREAD) == 0) {
164 		/* main thread gotta wait because it has the pid, et al */
165 		while (! TAILQ_EMPTY(&pr->ps_threads))
166 			tsleep(&pr->ps_threads, PUSER, "thrdeath", 0);
167 		if (pr->ps_flags & PS_PROFIL)
168 			stopprofclock(pr);
169 	} else if (TAILQ_EMPTY(&pr->ps_threads))
170 		wakeup(&pr->ps_threads);
171 
172 	rup = pr->ps_ru;
173 	if (rup == NULL) {
174 		rup = pool_get(&rusage_pool, PR_WAITOK | PR_ZERO);
175 
176 		if (pr->ps_ru == NULL)
177 			pr->ps_ru = rup;
178 		else {
179 			pool_put(&rusage_pool, rup);
180 			rup = pr->ps_ru;
181 		}
182 	}
183 	p->p_siglist = 0;
184 
185 	/*
186 	 * Close open files and release open-file table.
187 	 */
188 	fdfree(p);
189 
190 	if ((p->p_flag & P_THREAD) == 0) {
191 		timeout_del(&pr->ps_realit_to);
192 		timeout_del(&pr->ps_virt_to);
193 		timeout_del(&pr->ps_prof_to);
194 #ifdef SYSVSEM
195 		semexit(pr);
196 #endif
197 		if (SESS_LEADER(pr)) {
198 			struct session *sp = pr->ps_session;
199 
200 			if (sp->s_ttyvp) {
201 				/*
202 				 * Controlling process.
203 				 * Signal foreground pgrp,
204 				 * drain controlling terminal
205 				 * and revoke access to controlling terminal.
206 				 */
207 				if (sp->s_ttyp->t_session == sp) {
208 					if (sp->s_ttyp->t_pgrp)
209 						pgsignal(sp->s_ttyp->t_pgrp,
210 						    SIGHUP, 1);
211 					(void) ttywait(sp->s_ttyp);
212 					/*
213 					 * The tty could have been revoked
214 					 * if we blocked.
215 					 */
216 					if (sp->s_ttyvp)
217 						VOP_REVOKE(sp->s_ttyvp,
218 						    REVOKEALL);
219 				}
220 				if (sp->s_ttyvp)
221 					vrele(sp->s_ttyvp);
222 				sp->s_ttyvp = NULL;
223 				/*
224 				 * s_ttyp is not zero'd; we use this to
225 				 * indicate that the session once had a
226 				 * controlling terminal.  (for logging and
227 				 * informational purposes)
228 				 */
229 			}
230 			sp->s_leader = NULL;
231 		}
232 		fixjobc(pr, pr->ps_pgrp, 0);
233 
234 #ifdef ACCOUNTING
235 		(void)acct_process(p);
236 #endif
237 
238 #ifdef KTRACE
239 		/* release trace file */
240 		if (pr->ps_tracevp)
241 			ktrcleartrace(pr);
242 #endif
243 	}
244 
245 #if NSYSTRACE > 0
246 	if (ISSET(p->p_flag, P_SYSTRACE))
247 		systrace_exit(p);
248 #endif
249 
250 	/*
251 	 * If emulation has process exit hook, call it now.
252 	 */
253 	if (p->p_emul->e_proc_exit)
254 		(*p->p_emul->e_proc_exit)(p);
255 
256         /*
257          * Remove proc from pidhash chain so looking it up won't
258          * work.  Move it from allproc to zombproc, but do not yet
259          * wake up the reaper.  We will put the proc on the
260          * deadproc list later (using the p_hash member), and
261          * wake up the reaper when we do.
262          */
263 	/*
264 	 * NOTE: WE ARE NO LONGER ALLOWED TO SLEEP!
265 	 */
266 	p->p_stat = SDEAD;
267 
268 	LIST_REMOVE(p, p_hash);
269 	LIST_REMOVE(p, p_list);
270 	LIST_INSERT_HEAD(&zombproc, p, p_list);
271 
272 	/*
273 	 * Give orphaned children to init(8).
274 	 */
275 	if ((p->p_flag & P_THREAD) == 0) {
276 		qr = LIST_FIRST(&pr->ps_children);
277 		if (qr)		/* only need this if any child is S_ZOMB */
278 			wakeup(initproc->p_p);
279 		for (; qr != 0; qr = nqr) {
280 			nqr = LIST_NEXT(qr, ps_sibling);
281 			proc_reparent(qr, initproc->p_p);
282 			/*
283 			 * Traced processes are killed since their
284 			 * existence means someone is screwing up.
285 			 */
286 			if (qr->ps_flags & PS_TRACED &&
287 			    !(qr->ps_flags & PS_EXITING)) {
288 				atomic_clearbits_int(&qr->ps_flags, PS_TRACED);
289 				/*
290 				 * If single threading is active,
291 				 * direct the signal to the active
292 				 * thread to avoid deadlock.
293 				 */
294 				if (qr->ps_single)
295 					ptsignal(qr->ps_single, SIGKILL,
296 					    STHREAD);
297 				else
298 					prsignal(qr, SIGKILL);
299 			}
300 		}
301 	}
302 
303 
304 	/* add thread's accumulated rusage into the process's total */
305 	ruadd(rup, &p->p_ru);
306 
307 	/*
308 	 * clear %cpu usage during swap
309 	 */
310 	p->p_pctcpu = 0;
311 
312 	if ((p->p_flag & P_THREAD) == 0) {
313 		/*
314 		 * Final thread has died, so add on our children's rusage
315 		 * and calculate the total times
316 		 */
317 		calcru(&pr->ps_tu, &rup->ru_utime, &rup->ru_stime, NULL);
318 		ruadd(rup, &pr->ps_cru);
319 
320 		/* notify interested parties of our demise and clean up */
321 		knote_processexit(pr);
322 
323 		/*
324 		 * Notify parent that we're gone.  If we have P_NOZOMBIE
325 		 * or parent has the SAS_NOCLDWAIT flag set, notify process 1
326 		 * instead (and hope it will handle this situation).
327 		 */
328 		if ((p->p_flag & P_NOZOMBIE) ||
329 		    (pr->ps_pptr->ps_mainproc->p_sigacts->ps_flags &
330 		    SAS_NOCLDWAIT)) {
331 			struct process *ppr = pr->ps_pptr;
332 			proc_reparent(pr, initproc->p_p);
333 			/*
334 			 * If this was the last child of our parent, notify
335 			 * parent, so in case he was wait(2)ing, he will
336 			 * continue.
337 			 */
338 			if (LIST_EMPTY(&ppr->ps_children))
339 				wakeup(ppr);
340 		}
341 	}
342 
343 	/*
344 	 * Release the process's signal state.
345 	 */
346 	sigactsfree(p);
347 
348 	/*
349 	 * Other substructures are freed from reaper and wait().
350 	 */
351 
352 	/*
353 	 * Finally, call machine-dependent code to switch to a new
354 	 * context (possibly the idle context).  Once we are no longer
355 	 * using the dead process's vmspace and stack, exit2() will be
356 	 * called to schedule those resources to be released by the
357 	 * reaper thread.
358 	 *
359 	 * Note that cpu_exit() will end with a call equivalent to
360 	 * cpu_switch(), finishing our execution (pun intended).
361 	 */
362 	uvmexp.swtch++;
363 	cpu_exit(p);
364 	panic("cpu_exit returned");
365 }
366 
367 /*
368  * Locking of this proclist is special; it's accessed in a
369  * critical section of process exit, and thus locking it can't
370  * modify interrupt state.  We use a simple spin lock for this
371  * proclist.  Processes on this proclist are also on zombproc;
372  * we use the p_hash member to linkup to deadproc.
373  */
374 struct mutex deadproc_mutex = MUTEX_INITIALIZER(IPL_NONE);
375 struct proclist deadproc = LIST_HEAD_INITIALIZER(deadproc);
376 
377 /*
378  * We are called from cpu_exit() once it is safe to schedule the
379  * dead process's resources to be freed.
380  *
381  * NOTE: One must be careful with locking in this routine.  It's
382  * called from a critical section in machine-dependent code, so
383  * we should refrain from changing any interrupt state.
384  *
385  * We lock the deadproc list, place the proc on that list (using
386  * the p_hash member), and wake up the reaper.
387  */
388 void
389 exit2(struct proc *p)
390 {
391 	mtx_enter(&deadproc_mutex);
392 	LIST_INSERT_HEAD(&deadproc, p, p_hash);
393 	mtx_leave(&deadproc_mutex);
394 
395 	wakeup(&deadproc);
396 }
397 
398 /*
399  * Process reaper.  This is run by a kernel thread to free the resources
400  * of a dead process.  Once the resources are free, the process becomes
401  * a zombie, and the parent is allowed to read the undead's status.
402  */
403 void
404 reaper(void)
405 {
406 	struct proc *p;
407 
408 	KERNEL_UNLOCK();
409 
410 	SCHED_ASSERT_UNLOCKED();
411 
412 	for (;;) {
413 		mtx_enter(&deadproc_mutex);
414 		while ((p = LIST_FIRST(&deadproc)) == NULL)
415 			msleep(&deadproc, &deadproc_mutex, PVM, "reaper", 0);
416 
417 		/* Remove us from the deadproc list. */
418 		LIST_REMOVE(p, p_hash);
419 		mtx_leave(&deadproc_mutex);
420 
421 		KERNEL_LOCK();
422 
423 		/*
424 		 * Free the VM resources we're still holding on to.
425 		 * We must do this from a valid thread because doing
426 		 * so may block.
427 		 */
428 		uvm_exit(p);
429 
430 		/* Process is now a true zombie. */
431 		if ((p->p_flag & P_NOZOMBIE) == 0) {
432 			p->p_stat = SZOMB;
433 
434 			if (P_EXITSIG(p) != 0)
435 				prsignal(p->p_p->ps_pptr, P_EXITSIG(p));
436 			/* Wake up the parent so it can get exit status. */
437 			wakeup(p->p_p->ps_pptr);
438 		} else {
439 			/* Noone will wait for us. Just zap the process now */
440 			proc_zap(p);
441 		}
442 
443 		KERNEL_UNLOCK();
444 	}
445 }
446 
447 int
448 sys_wait4(struct proc *q, void *v, register_t *retval)
449 {
450 	struct sys_wait4_args /* {
451 		syscallarg(pid_t) pid;
452 		syscallarg(int *) status;
453 		syscallarg(int) options;
454 		syscallarg(struct rusage *) rusage;
455 	} */ *uap = v;
456 	int nfound;
457 	struct process *pr;
458 	struct proc *p;
459 	int status, error;
460 
461 	if (SCARG(uap, pid) == 0)
462 		SCARG(uap, pid) = -q->p_p->ps_pgid;
463 	if (SCARG(uap, options) &~ (WUNTRACED|WNOHANG|WALTSIG|WCONTINUED))
464 		return (EINVAL);
465 
466 loop:
467 	nfound = 0;
468 	LIST_FOREACH(pr, &q->p_p->ps_children, ps_sibling) {
469 		p = pr->ps_mainproc;
470 		if ((p->p_flag & P_NOZOMBIE) ||
471 		    (SCARG(uap, pid) != WAIT_ANY &&
472 		    p->p_pid != SCARG(uap, pid) &&
473 		    pr->ps_pgid != -SCARG(uap, pid)))
474 			continue;
475 
476 		/*
477 		 * Wait for processes with p_exitsig != SIGCHLD processes only
478 		 * if WALTSIG is set; wait for processes with pexitsig ==
479 		 * SIGCHLD only if WALTSIG is clear.
480 		 */
481 		if ((SCARG(uap, options) & WALTSIG) ?
482 		    (p->p_exitsig == SIGCHLD) : (P_EXITSIG(p) != SIGCHLD))
483 			continue;
484 
485 		nfound++;
486 		if (p->p_stat == SZOMB) {
487 			retval[0] = p->p_pid;
488 
489 			if (SCARG(uap, status)) {
490 				status = p->p_xstat;	/* convert to int */
491 				error = copyout(&status,
492 				    SCARG(uap, status), sizeof(status));
493 				if (error)
494 					return (error);
495 			}
496 			if (SCARG(uap, rusage) &&
497 			    (error = copyout(pr->ps_ru,
498 			    SCARG(uap, rusage), sizeof(struct rusage))))
499 				return (error);
500 			proc_finish_wait(q, p);
501 			return (0);
502 		}
503 		if (pr->ps_flags & PS_TRACED &&
504 		    (pr->ps_flags & PS_WAITED) == 0 && pr->ps_single &&
505 		    pr->ps_single->p_stat == SSTOP &&
506 		    (pr->ps_single->p_flag & P_SUSPSINGLE) == 0) {
507 			atomic_setbits_int(&pr->ps_flags, PS_WAITED);
508 			retval[0] = p->p_pid;
509 
510 			if (SCARG(uap, status)) {
511 				status = W_STOPCODE(pr->ps_single->p_xstat);
512 				error = copyout(&status, SCARG(uap, status),
513 				    sizeof(status));
514 			} else
515 				error = 0;
516 			return (error);
517 		}
518 		if (p->p_stat == SSTOP &&
519 		    (pr->ps_flags & PS_WAITED) == 0 &&
520 		    (p->p_flag & P_SUSPSINGLE) == 0 &&
521 		    (pr->ps_flags & PS_TRACED ||
522 		    SCARG(uap, options) & WUNTRACED)) {
523 			atomic_setbits_int(&pr->ps_flags, PS_WAITED);
524 			retval[0] = p->p_pid;
525 
526 			if (SCARG(uap, status)) {
527 				status = W_STOPCODE(p->p_xstat);
528 				error = copyout(&status, SCARG(uap, status),
529 				    sizeof(status));
530 			} else
531 				error = 0;
532 			return (error);
533 		}
534 		if ((SCARG(uap, options) & WCONTINUED) && (p->p_flag & P_CONTINUED)) {
535 			atomic_clearbits_int(&p->p_flag, P_CONTINUED);
536 			retval[0] = p->p_pid;
537 
538 			if (SCARG(uap, status)) {
539 				status = _WCONTINUED;
540 				error = copyout(&status, SCARG(uap, status),
541 				    sizeof(status));
542 			} else
543 				error = 0;
544 			return (error);
545 		}
546 	}
547 	if (nfound == 0)
548 		return (ECHILD);
549 	if (SCARG(uap, options) & WNOHANG) {
550 		retval[0] = 0;
551 		return (0);
552 	}
553 	if ((error = tsleep(q->p_p, PWAIT | PCATCH, "wait", 0)) != 0)
554 		return (error);
555 	goto loop;
556 }
557 
558 void
559 proc_finish_wait(struct proc *waiter, struct proc *p)
560 {
561 	struct process *pr, *tr;
562 	struct rusage *rup;
563 
564 	/*
565 	 * If we got the child via a ptrace 'attach',
566 	 * we need to give it back to the old parent.
567 	 */
568 	pr = p->p_p;
569 	if ((p->p_flag & P_THREAD) == 0 && pr->ps_oppid &&
570 	    (tr = prfind(pr->ps_oppid))) {
571 		atomic_clearbits_int(&pr->ps_flags, PS_TRACED);
572 		pr->ps_oppid = 0;
573 		proc_reparent(pr, tr);
574 		if (p->p_exitsig != 0)
575 			prsignal(tr, p->p_exitsig);
576 		wakeup(tr);
577 	} else {
578 		scheduler_wait_hook(waiter, p);
579 		p->p_xstat = 0;
580 		rup = &waiter->p_p->ps_cru;
581 		ruadd(rup, pr->ps_ru);
582 		proc_zap(p);
583 	}
584 }
585 
586 /*
587  * make process 'parent' the new parent of process 'child'.
588  */
589 void
590 proc_reparent(struct process *child, struct process *parent)
591 {
592 
593 	if (child->ps_pptr == parent)
594 		return;
595 
596 	if (parent == initproc->p_p)
597 		child->ps_mainproc->p_exitsig = SIGCHLD;
598 
599 	LIST_REMOVE(child, ps_sibling);
600 	LIST_INSERT_HEAD(&parent->ps_children, child, ps_sibling);
601 	child->ps_pptr = parent;
602 }
603 
604 void
605 proc_zap(struct proc *p)
606 {
607 	struct process *pr = p->p_p;
608 
609 	/*
610 	 * Finally finished with old proc entry.
611 	 * Unlink it from its process group and free it.
612 	 */
613 	if ((p->p_flag & P_THREAD) == 0)
614 		leavepgrp(pr);
615 	LIST_REMOVE(p, p_list);	/* off zombproc */
616 	if ((p->p_flag & P_THREAD) == 0) {
617 		LIST_REMOVE(pr, ps_sibling);
618 
619 		/*
620 		 * Decrement the count of procs running with this uid.
621 		 */
622 		(void)chgproccnt(p->p_cred->p_ruid, -1);
623 	}
624 
625 	/*
626 	 * Release reference to text vnode
627 	 */
628 	if (p->p_textvp)
629 		vrele(p->p_textvp);
630 
631 	/*
632 	 * Remove us from our process list, possibly killing the process
633 	 * in the process (pun intended).
634 	 */
635 	if (--pr->ps_refcnt == 0) {
636 		if (pr->ps_ptstat != NULL)
637 			free(pr->ps_ptstat, M_SUBPROC);
638 		pool_put(&rusage_pool, pr->ps_ru);
639 		KASSERT(TAILQ_EMPTY(&pr->ps_threads));
640 		limfree(pr->ps_limit);
641 		crfree(pr->ps_cred->pc_ucred);
642 		pool_put(&pcred_pool, pr->ps_cred);
643 		pool_put(&process_pool, pr);
644 		nprocesses--;
645 	}
646 
647 	pool_put(&proc_pool, p);
648 	nthreads--;
649 }
650