xref: /openbsd-src/sys/kern/kern_exit.c (revision 665e07d6cfe372a9cb7c136fce813ece1d8a4be8)
1 /*	$OpenBSD: kern_exit.c,v 1.139 2014/04/17 14:52:50 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 
74 #include <uvm/uvm_extern.h>
75 
76 /*
77  * exit --
78  *	Death of process.
79  */
80 int
81 sys_exit(struct proc *p, void *v, register_t *retval)
82 {
83 	struct sys_exit_args /* {
84 		syscallarg(int) rval;
85 	} */ *uap = v;
86 
87 	exit1(p, W_EXITCODE(SCARG(uap, rval), 0), EXIT_NORMAL);
88 	/* NOTREACHED */
89 	return (0);
90 }
91 
92 int
93 sys___threxit(struct proc *p, void *v, register_t *retval)
94 {
95 	struct sys___threxit_args /* {
96 		syscallarg(pid_t *) notdead;
97 	} */ *uap = v;
98 
99 	if (SCARG(uap, notdead) != NULL) {
100 		pid_t zero = 0;
101 		if (copyout(&zero, SCARG(uap, notdead), sizeof(zero)))
102 			psignal(p, SIGSEGV);
103 	}
104 	exit1(p, 0, EXIT_THREAD);
105 
106 	return (0);
107 }
108 
109 /*
110  * Exit: deallocate address space and other resources, change proc state
111  * to zombie, and unlink proc from allproc and parent's lists.  Save exit
112  * status and rusage for wait().  Check for child processes and orphan them.
113  */
114 void
115 exit1(struct proc *p, int rv, int flags)
116 {
117 	struct process *pr, *qr, *nqr;
118 	struct rusage *rup;
119 	struct vnode *ovp;
120 
121 	if (p->p_pid == 1)
122 		panic("init died (signal %d, exit %d)",
123 		    WTERMSIG(rv), WEXITSTATUS(rv));
124 
125 	atomic_setbits_int(&p->p_flag, P_WEXIT);
126 
127 	pr = p->p_p;
128 
129 	/* single-threaded? */
130 	if (TAILQ_FIRST(&pr->ps_threads) == p &&
131 	    TAILQ_NEXT(p, p_thr_link) == NULL) {
132 		flags = EXIT_NORMAL;
133 	} else {
134 		/* nope, multi-threaded */
135 		if (flags == EXIT_NORMAL)
136 			single_thread_set(p, SINGLE_EXIT, 0);
137 		else if (flags == EXIT_THREAD)
138 			single_thread_check(p, 0);
139 	}
140 
141 	if (flags == EXIT_NORMAL) {
142 		atomic_setbits_int(&pr->ps_flags, PS_EXITING);
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 (pr->ps_refcnt > 1)
162 			tsleep(&pr->ps_threads, PUSER, "thrdeath", 0);
163 		if (pr->ps_flags & PS_PROFIL)
164 			stopprofclock(pr);
165 	}
166 
167 	rup = pr->ps_ru;
168 	if (rup == NULL) {
169 		rup = pool_get(&rusage_pool, PR_WAITOK | PR_ZERO);
170 		if (pr->ps_ru == NULL) {
171 			pr->ps_ru = rup;
172 		} else {
173 			pool_put(&rusage_pool, rup);
174 			rup = pr->ps_ru;
175 		}
176 	}
177 	p->p_siglist = 0;
178 
179 	/*
180 	 * Close open files and release open-file table.
181 	 */
182 	fdfree(p);
183 
184 	if ((p->p_flag & P_THREAD) == 0) {
185 		timeout_del(&pr->ps_realit_to);
186 #ifdef SYSVSEM
187 		semexit(pr);
188 #endif
189 		if (SESS_LEADER(pr)) {
190 			struct session *sp = pr->ps_session;
191 
192 			if (sp->s_ttyvp) {
193 				/*
194 				 * Controlling process.
195 				 * Signal foreground pgrp,
196 				 * drain controlling terminal
197 				 * and revoke access to controlling terminal.
198 				 */
199 				if (sp->s_ttyp->t_session == sp) {
200 					if (sp->s_ttyp->t_pgrp)
201 						pgsignal(sp->s_ttyp->t_pgrp,
202 						    SIGHUP, 1);
203 					ttywait(sp->s_ttyp);
204 					/*
205 					 * The tty could have been revoked
206 					 * if we blocked.
207 					 */
208 					if (sp->s_ttyvp)
209 						VOP_REVOKE(sp->s_ttyvp,
210 						    REVOKEALL);
211 				}
212 				ovp = sp->s_ttyvp;
213 				sp->s_ttyvp = NULL;
214 				if (ovp)
215 					vrele(ovp);
216 				/*
217 				 * s_ttyp is not zero'd; we use this to
218 				 * indicate that the session once had a
219 				 * controlling terminal.  (for logging and
220 				 * informational purposes)
221 				 */
222 			}
223 			sp->s_leader = NULL;
224 		}
225 		fixjobc(pr, pr->ps_pgrp, 0);
226 
227 #ifdef ACCOUNTING
228 		acct_process(p);
229 #endif
230 
231 #ifdef KTRACE
232 		/* release trace file */
233 		if (pr->ps_tracevp)
234 			ktrcleartrace(pr);
235 #endif
236 
237 		/*
238 		 * If parent has the SAS_NOCLDWAIT flag set, we're not
239 		 * going to become a zombie.
240 		 */
241 		if (pr->ps_pptr->ps_sigacts->ps_flags & SAS_NOCLDWAIT)
242 			atomic_setbits_int(&pr->ps_flags, PS_NOZOMBIE);
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 thread exit hook, call it now.
252 	 */
253 	if (pr->ps_emul->e_proc_exit)
254 		(*pr->ps_emul->e_proc_exit)(p);
255 
256         /*
257 	 * Remove proc from pidhash chain and allproc so looking
258 	 * it up won't work.  We will put the proc on the
259 	 * deadproc list later (using the p_hash member), and
260 	 * wake up the reaper when we do.  If this is the last
261 	 * thread of a process that isn't PS_NOZOMBIE, we'll put
262 	 * the process on the zombprocess list below.
263 	 */
264 	/*
265 	 * NOTE: WE ARE NO LONGER ALLOWED TO SLEEP!
266 	 */
267 	p->p_stat = SDEAD;
268 
269 	LIST_REMOVE(p, p_hash);
270 	LIST_REMOVE(p, p_list);
271 
272 	if ((p->p_flag & P_THREAD) == 0) {
273 		LIST_REMOVE(pr, ps_list);
274 
275 		if ((pr->ps_flags & PS_NOZOMBIE) == 0)
276 			LIST_INSERT_HEAD(&zombprocess, pr, ps_list);
277 		else {
278 			/*
279 			 * Not going to be a zombie, so it's now off all
280 			 * the lists scanned by ispidtaken(), so block
281 			 * fast reuse of the pid now.
282 			 */
283 			freepid(p->p_pid);
284 		}
285 
286 		/*
287 		 * Give orphaned children to init(8).
288 		 */
289 		qr = LIST_FIRST(&pr->ps_children);
290 		if (qr)		/* only need this if any child is S_ZOMB */
291 			wakeup(initproc->p_p);
292 		for (; qr != 0; qr = nqr) {
293 			nqr = LIST_NEXT(qr, ps_sibling);
294 			proc_reparent(qr, initproc->p_p);
295 			/*
296 			 * Traced processes are killed since their
297 			 * existence means someone is screwing up.
298 			 */
299 			if (qr->ps_flags & PS_TRACED &&
300 			    !(qr->ps_flags & PS_EXITING)) {
301 				atomic_clearbits_int(&qr->ps_flags, PS_TRACED);
302 				/*
303 				 * If single threading is active,
304 				 * direct the signal to the active
305 				 * thread to avoid deadlock.
306 				 */
307 				if (qr->ps_single)
308 					ptsignal(qr->ps_single, SIGKILL,
309 					    STHREAD);
310 				else
311 					prsignal(qr, SIGKILL);
312 			}
313 		}
314 	}
315 
316 	/* add thread's accumulated rusage into the process's total */
317 	ruadd(rup, &p->p_ru);
318 	tuagg(pr, p);
319 
320 	/*
321 	 * clear %cpu usage during swap
322 	 */
323 	p->p_pctcpu = 0;
324 
325 	if ((p->p_flag & P_THREAD) == 0) {
326 		/*
327 		 * Final thread has died, so add on our children's rusage
328 		 * and calculate the total times
329 		 */
330 		calcru(&pr->ps_tu, &rup->ru_utime, &rup->ru_stime, NULL);
331 		ruadd(rup, &pr->ps_cru);
332 
333 		/* notify interested parties of our demise and clean up */
334 		knote_processexit(pr);
335 
336 		/*
337 		 * Notify parent that we're gone.  If we're not going to
338 		 * become a zombie, reparent to process 1 (init) so that
339 		 * we can wake our original parent to possibly unblock
340 		 * wait4() to return ECHILD.
341 		 */
342 		if (pr->ps_flags & PS_NOZOMBIE) {
343 			struct process *ppr = pr->ps_pptr;
344 			proc_reparent(pr, initproc->p_p);
345 			wakeup(ppr);
346 		}
347 
348 		/*
349 		 * Release the process's signal state.
350 		 */
351 		sigactsfree(pr);
352 	}
353 
354 	/* just a thread? detach it from its process */
355 	if (p->p_flag & P_THREAD) {
356 		/* scheduler_wait_hook(pr->ps_mainproc, p); XXX */
357 		if (--pr->ps_refcnt == 1)
358 			wakeup(&pr->ps_threads);
359 		KASSERT(pr->ps_refcnt > 0);
360 	}
361 
362 	/*
363 	 * Other substructures are freed from reaper and wait().
364 	 */
365 
366 	/*
367 	 * Finally, call machine-dependent code to switch to a new
368 	 * context (possibly the idle context).  Once we are no longer
369 	 * using the dead process's vmspace and stack, exit2() will be
370 	 * called to schedule those resources to be released by the
371 	 * reaper thread.
372 	 *
373 	 * Note that cpu_exit() will end with a call equivalent to
374 	 * cpu_switch(), finishing our execution (pun intended).
375 	 */
376 	uvmexp.swtch++;
377 	cpu_exit(p);
378 	panic("cpu_exit returned");
379 }
380 
381 /*
382  * Locking of this proclist is special; it's accessed in a
383  * critical section of process exit, and thus locking it can't
384  * modify interrupt state.  We use a simple spin lock for this
385  * proclist.  We use the p_hash member to linkup to deadproc.
386  */
387 struct mutex deadproc_mutex = MUTEX_INITIALIZER(IPL_NONE);
388 struct proclist deadproc = LIST_HEAD_INITIALIZER(deadproc);
389 
390 /*
391  * We are called from cpu_exit() once it is safe to schedule the
392  * dead process's resources to be freed.
393  *
394  * NOTE: One must be careful with locking in this routine.  It's
395  * called from a critical section in machine-dependent code, so
396  * we should refrain from changing any interrupt state.
397  *
398  * We lock the deadproc list, place the proc on that list (using
399  * the p_hash member), and wake up the reaper.
400  */
401 void
402 exit2(struct proc *p)
403 {
404 	mtx_enter(&deadproc_mutex);
405 	LIST_INSERT_HEAD(&deadproc, p, p_hash);
406 	mtx_leave(&deadproc_mutex);
407 
408 	wakeup(&deadproc);
409 }
410 
411 void
412 proc_free(struct proc *p)
413 {
414 	pool_put(&proc_pool, p);
415 	nthreads--;
416 }
417 
418 /*
419  * Process reaper.  This is run by a kernel thread to free the resources
420  * of a dead process.  Once the resources are free, the process becomes
421  * a zombie, and the parent is allowed to read the undead's status.
422  */
423 void
424 reaper(void)
425 {
426 	struct proc *p;
427 
428 	KERNEL_UNLOCK();
429 
430 	SCHED_ASSERT_UNLOCKED();
431 
432 	for (;;) {
433 		mtx_enter(&deadproc_mutex);
434 		while ((p = LIST_FIRST(&deadproc)) == NULL)
435 			msleep(&deadproc, &deadproc_mutex, PVM, "reaper", 0);
436 
437 		/* Remove us from the deadproc list. */
438 		LIST_REMOVE(p, p_hash);
439 		mtx_leave(&deadproc_mutex);
440 
441 		KERNEL_LOCK();
442 
443 		/*
444 		 * Free the VM resources we're still holding on to.
445 		 * We must do this from a valid thread because doing
446 		 * so may block.
447 		 */
448 		uvm_exit(p);
449 
450 		if (p->p_flag & P_THREAD) {
451 			/* Just a thread */
452 			proc_free(p);
453 		} else {
454 			struct process *pr = p->p_p;
455 
456 			if ((pr->ps_flags & PS_NOZOMBIE) == 0) {
457 				/* Process is now a true zombie. */
458 				p->p_stat = SZOMB;
459 				prsignal(pr->ps_pptr, SIGCHLD);
460 
461 				/* Wake up the parent so it can get exit status. */
462 				wakeup(pr->ps_pptr);
463 			} else {
464 				/* No one will wait for us. Just zap the process now */
465 				process_zap(pr);
466 			}
467 		}
468 
469 		KERNEL_UNLOCK();
470 	}
471 }
472 
473 int
474 sys_wait4(struct proc *q, void *v, register_t *retval)
475 {
476 	struct sys_wait4_args /* {
477 		syscallarg(pid_t) pid;
478 		syscallarg(int *) status;
479 		syscallarg(int) options;
480 		syscallarg(struct rusage *) rusage;
481 	} */ *uap = v;
482 	struct rusage ru;
483 	int status, error;
484 
485 	error = dowait4(q, SCARG(uap, pid),
486 	    SCARG(uap, status) ? &status : NULL,
487 	    SCARG(uap, options), SCARG(uap, rusage) ? &ru : NULL, retval);
488 	if (error == 0 && SCARG(uap, status)) {
489 		error = copyout(&status, SCARG(uap, status), sizeof(status));
490 	}
491 	if (error == 0 && SCARG(uap, rusage)) {
492 		error = copyout(&ru, SCARG(uap, rusage), sizeof(ru));
493 #ifdef KTRACE
494 		if (error == 0 && KTRPOINT(q, KTR_STRUCT))
495 			ktrrusage(q, &ru);
496 #endif
497 	}
498 	return (error);
499 }
500 
501 int
502 dowait4(struct proc *q, pid_t pid, int *statusp, int options,
503     struct rusage *rusage, register_t *retval)
504 {
505 	int nfound;
506 	struct process *pr;
507 	struct proc *p;
508 	int error;
509 
510 	if (pid == 0)
511 		pid = -q->p_p->ps_pgid;
512 	if (options &~ (WUNTRACED|WNOHANG|WCONTINUED))
513 		return (EINVAL);
514 
515 loop:
516 	nfound = 0;
517 	LIST_FOREACH(pr, &q->p_p->ps_children, ps_sibling) {
518 		p = pr->ps_mainproc;
519 		if ((pr->ps_flags & PS_NOZOMBIE) ||
520 		    (pid != WAIT_ANY &&
521 		    p->p_pid != pid &&
522 		    pr->ps_pgid != -pid))
523 			continue;
524 
525 		nfound++;
526 		if (p->p_stat == SZOMB) {
527 			retval[0] = p->p_pid;
528 
529 			if (statusp != NULL)
530 				*statusp = p->p_xstat;	/* convert to int */
531 			if (rusage != NULL)
532 				memcpy(rusage, pr->ps_ru, sizeof(*rusage));
533 			proc_finish_wait(q, p);
534 			return (0);
535 		}
536 		if (pr->ps_flags & PS_TRACED &&
537 		    (pr->ps_flags & PS_WAITED) == 0 && pr->ps_single &&
538 		    pr->ps_single->p_stat == SSTOP &&
539 		    (pr->ps_single->p_flag & P_SUSPSINGLE) == 0) {
540 			single_thread_wait(pr);
541 
542 			atomic_setbits_int(&pr->ps_flags, PS_WAITED);
543 			retval[0] = p->p_pid;
544 
545 			if (statusp != NULL)
546 				*statusp = W_STOPCODE(pr->ps_single->p_xstat);
547 			return (0);
548 		}
549 		if (p->p_stat == SSTOP &&
550 		    (pr->ps_flags & PS_WAITED) == 0 &&
551 		    (p->p_flag & P_SUSPSINGLE) == 0 &&
552 		    (pr->ps_flags & PS_TRACED ||
553 		    options & WUNTRACED)) {
554 			atomic_setbits_int(&pr->ps_flags, PS_WAITED);
555 			retval[0] = p->p_pid;
556 
557 			if (statusp != NULL)
558 				*statusp = W_STOPCODE(p->p_xstat);
559 			return (0);
560 		}
561 		if ((options & WCONTINUED) && (p->p_flag & P_CONTINUED)) {
562 			atomic_clearbits_int(&p->p_flag, P_CONTINUED);
563 			retval[0] = p->p_pid;
564 
565 			if (statusp != NULL)
566 				*statusp = _WCONTINUED;
567 			return (0);
568 		}
569 	}
570 	if (nfound == 0)
571 		return (ECHILD);
572 	if (options & WNOHANG) {
573 		retval[0] = 0;
574 		return (0);
575 	}
576 	if ((error = tsleep(q->p_p, PWAIT | PCATCH, "wait", 0)) != 0)
577 		return (error);
578 	goto loop;
579 }
580 
581 void
582 proc_finish_wait(struct proc *waiter, struct proc *p)
583 {
584 	struct process *pr, *tr;
585 	struct rusage *rup;
586 
587 	/*
588 	 * If we got the child via a ptrace 'attach',
589 	 * we need to give it back to the old parent.
590 	 */
591 	pr = p->p_p;
592 	if (pr->ps_oppid && (tr = prfind(pr->ps_oppid))) {
593 		atomic_clearbits_int(&pr->ps_flags, PS_TRACED);
594 		pr->ps_oppid = 0;
595 		proc_reparent(pr, tr);
596 		prsignal(tr, SIGCHLD);
597 		wakeup(tr);
598 	} else {
599 		scheduler_wait_hook(waiter, p);
600 		p->p_xstat = 0;
601 		rup = &waiter->p_p->ps_cru;
602 		ruadd(rup, pr->ps_ru);
603 		LIST_REMOVE(pr, ps_list);	/* off zombprocess */
604 		freepid(p->p_pid);
605 		process_zap(pr);
606 	}
607 }
608 
609 /*
610  * make process 'parent' the new parent of process 'child'.
611  */
612 void
613 proc_reparent(struct process *child, struct process *parent)
614 {
615 
616 	if (child->ps_pptr == parent)
617 		return;
618 
619 	LIST_REMOVE(child, ps_sibling);
620 	LIST_INSERT_HEAD(&parent->ps_children, child, ps_sibling);
621 	child->ps_pptr = parent;
622 }
623 
624 void
625 process_zap(struct process *pr)
626 {
627 	struct vnode *otvp;
628 	struct proc *p = pr->ps_mainproc;
629 
630 	/*
631 	 * Finally finished with old proc entry.
632 	 * Unlink it from its process group and free it.
633 	 */
634 	leavepgrp(pr);
635 	LIST_REMOVE(pr, ps_sibling);
636 
637 	/*
638 	 * Decrement the count of procs running with this uid.
639 	 */
640 	(void)chgproccnt(pr->ps_ucred->cr_ruid, -1);
641 
642 	/*
643 	 * Release reference to text vnode
644 	 */
645 	otvp = pr->ps_textvp;
646 	pr->ps_textvp = NULL;
647 	if (otvp)
648 		vrele(otvp);
649 
650 	KASSERT(pr->ps_refcnt == 1);
651 	if (pr->ps_ptstat != NULL)
652 		free(pr->ps_ptstat, M_SUBPROC);
653 	pool_put(&rusage_pool, pr->ps_ru);
654 	KASSERT(TAILQ_EMPTY(&pr->ps_threads));
655 	limfree(pr->ps_limit);
656 	crfree(pr->ps_ucred);
657 	pool_put(&process_pool, pr);
658 	nprocesses--;
659 
660 	proc_free(p);
661 }
662