xref: /freebsd/sys/kern/kern_exit.c (revision 77a0943ded95b9e6438f7db70c4a28e4d93946d4)
1 /*
2  * Copyright (c) 1982, 1986, 1989, 1991, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  * (c) UNIX System Laboratories, Inc.
5  * All or some portions of this file are derived from material licensed
6  * to the University of California by American Telephone and Telegraph
7  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8  * the permission of UNIX System Laboratories, Inc.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. All advertising materials mentioning features or use of this software
19  *    must display the following acknowledgement:
20  *	This product includes software developed by the University of
21  *	California, Berkeley and its contributors.
22  * 4. Neither the name of the University nor the names of its contributors
23  *    may be used to endorse or promote products derived from this software
24  *    without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36  * SUCH DAMAGE.
37  *
38  *	@(#)kern_exit.c	8.7 (Berkeley) 2/12/94
39  * $FreeBSD$
40  */
41 
42 #include "opt_compat.h"
43 #include "opt_ktrace.h"
44 
45 #include <sys/param.h>
46 #include <sys/systm.h>
47 #include <sys/sysproto.h>
48 #include <sys/kernel.h>
49 #include <sys/malloc.h>
50 #include <sys/mutex.h>
51 #include <sys/proc.h>
52 #include <sys/pioctl.h>
53 #include <sys/tty.h>
54 #include <sys/wait.h>
55 #include <sys/vnode.h>
56 #include <sys/resourcevar.h>
57 #include <sys/signalvar.h>
58 #include <sys/ptrace.h>
59 #include <sys/acct.h>		/* for acct_process() function prototype */
60 #include <sys/filedesc.h>
61 #include <sys/shm.h>
62 #include <sys/sem.h>
63 #include <sys/aio.h>
64 #include <sys/jail.h>
65 
66 #include <vm/vm.h>
67 #include <vm/vm_param.h>
68 #include <sys/lock.h>
69 #include <vm/pmap.h>
70 #include <vm/vm_map.h>
71 #include <vm/vm_zone.h>
72 #include <sys/user.h>
73 
74 /* Required to be non-static for SysVR4 emulator */
75 MALLOC_DEFINE(M_ZOMBIE, "zombie", "zombie proc status");
76 
77 static MALLOC_DEFINE(M_ATEXIT, "atexit", "atexit callback");
78 
79 static int wait1 __P((struct proc *, struct wait_args *, int));
80 
81 /*
82  * callout list for things to do at exit time
83  */
84 struct exitlist {
85 	exitlist_fn function;
86 	TAILQ_ENTRY(exitlist) next;
87 };
88 
89 TAILQ_HEAD(exit_list_head, exitlist);
90 static struct exit_list_head exit_list = TAILQ_HEAD_INITIALIZER(exit_list);
91 
92 /*
93  * exit --
94  *	Death of process.
95  */
96 void
97 sys_exit(p, uap)
98 	struct proc *p;
99 	struct sys_exit_args /* {
100 		int	rval;
101 	} */ *uap;
102 {
103 
104 	exit1(p, W_EXITCODE(uap->rval, 0));
105 	/* NOTREACHED */
106 }
107 
108 /*
109  * Exit: deallocate address space and other resources, change proc state
110  * to zombie, and unlink proc from allproc and parent's lists.  Save exit
111  * status and rusage for wait().  Check for child processes and orphan them.
112  */
113 void
114 exit1(p, rv)
115 	register struct proc *p;
116 	int rv;
117 {
118 	register struct proc *q, *nq;
119 	register struct vmspace *vm;
120 	struct exitlist *ep;
121 
122 	if (p->p_pid == 1) {
123 		printf("init died (signal %d, exit %d)\n",
124 		    WTERMSIG(rv), WEXITSTATUS(rv));
125 		panic("Going nowhere without my init!");
126 	}
127 
128 	aio_proc_rundown(p);
129 
130 	/* are we a task leader? */
131 	if(p == p->p_leader) {
132         	struct kill_args killArgs;
133 		killArgs.signum = SIGKILL;
134 		q = p->p_peers;
135 		while(q) {
136 			killArgs.pid = q->p_pid;
137 			/*
138 		         * The interface for kill is better
139 			 * than the internal signal
140 			 */
141 			kill(p, &killArgs);
142 			nq = q;
143 			q = q->p_peers;
144 		}
145 		while (p->p_peers)
146 		  tsleep((caddr_t)p, PWAIT, "exit1", 0);
147 	}
148 
149 #ifdef PGINPROF
150 	vmsizmon();
151 #endif
152 	STOPEVENT(p, S_EXIT, rv);
153 	wakeup(&p->p_stype);	/* Wakeup anyone in procfs' PIOCWAIT */
154 
155 	/*
156 	 * Check if any loadable modules need anything done at process exit.
157 	 * e.g. SYSV IPC stuff
158 	 * XXX what if one of these generates an error?
159 	 */
160 	TAILQ_FOREACH(ep, &exit_list, next)
161 		(*ep->function)(p);
162 
163 	if (p->p_flag & P_PROFIL)
164 		stopprofclock(p);
165 	MALLOC(p->p_ru, struct rusage *, sizeof(struct rusage),
166 		M_ZOMBIE, M_WAITOK);
167 	/*
168 	 * If parent is waiting for us to exit or exec,
169 	 * P_PPWAIT is set; we will wakeup the parent below.
170 	 */
171 	p->p_flag &= ~(P_TRACED | P_PPWAIT);
172 	p->p_flag |= P_WEXIT;
173 	SIGEMPTYSET(p->p_siglist);
174 	if (timevalisset(&p->p_realtimer.it_value))
175 		untimeout(realitexpire, (caddr_t)p, p->p_ithandle);
176 
177 	/*
178 	 * Reset any sigio structures pointing to us as a result of
179 	 * F_SETOWN with our pid.
180 	 */
181 	funsetownlst(&p->p_sigiolst);
182 
183 	/*
184 	 * Close open files and release open-file table.
185 	 * This may block!
186 	 */
187 	fdfree(p);
188 
189 	if(p->p_leader->p_peers) {
190 		q = p->p_leader;
191 		while(q->p_peers != p)
192 			q = q->p_peers;
193 		q->p_peers = p->p_peers;
194 		wakeup((caddr_t)p->p_leader);
195 	}
196 
197 	/*
198 	 * XXX Shutdown SYSV semaphores
199 	 */
200 	semexit(p);
201 
202 	/* The next two chunks should probably be moved to vmspace_exit. */
203 	vm = p->p_vmspace;
204 	/*
205 	 * Release user portion of address space.
206 	 * This releases references to vnodes,
207 	 * which could cause I/O if the file has been unlinked.
208 	 * Need to do this early enough that we can still sleep.
209 	 * Can't free the entire vmspace as the kernel stack
210 	 * may be mapped within that space also.
211 	 */
212 	if (vm->vm_refcnt == 1) {
213 		if (vm->vm_shm)
214 			shmexit(p);
215 		pmap_remove_pages(vmspace_pmap(vm), VM_MIN_ADDRESS,
216 		    VM_MAXUSER_ADDRESS);
217 		(void) vm_map_remove(&vm->vm_map, VM_MIN_ADDRESS,
218 		    VM_MAXUSER_ADDRESS);
219 	}
220 
221 	if (SESS_LEADER(p)) {
222 		register struct session *sp = p->p_session;
223 
224 		if (sp->s_ttyvp) {
225 			/*
226 			 * Controlling process.
227 			 * Signal foreground pgrp,
228 			 * drain controlling terminal
229 			 * and revoke access to controlling terminal.
230 			 */
231 			if (sp->s_ttyp && (sp->s_ttyp->t_session == sp)) {
232 				if (sp->s_ttyp->t_pgrp)
233 					pgsignal(sp->s_ttyp->t_pgrp, SIGHUP, 1);
234 				(void) ttywait(sp->s_ttyp);
235 				/*
236 				 * The tty could have been revoked
237 				 * if we blocked.
238 				 */
239 				if (sp->s_ttyvp)
240 					VOP_REVOKE(sp->s_ttyvp, REVOKEALL);
241 			}
242 			if (sp->s_ttyvp)
243 				vrele(sp->s_ttyvp);
244 			sp->s_ttyvp = NULL;
245 			/*
246 			 * s_ttyp is not zero'd; we use this to indicate
247 			 * that the session once had a controlling terminal.
248 			 * (for logging and informational purposes)
249 			 */
250 		}
251 		sp->s_leader = NULL;
252 	}
253 	fixjobc(p, p->p_pgrp, 0);
254 	(void)acct_process(p);
255 #ifdef KTRACE
256 	/*
257 	 * release trace file
258 	 */
259 	p->p_traceflag = 0;	/* don't trace the vrele() */
260 	if (p->p_tracep)
261 		vrele(p->p_tracep);
262 #endif
263 	/*
264 	 * Remove proc from allproc queue and pidhash chain.
265 	 * Place onto zombproc.  Unlink from parent's child list.
266 	 */
267 	lockmgr(&allproc_lock, LK_EXCLUSIVE, NULL, CURPROC);
268 	LIST_REMOVE(p, p_list);
269 	LIST_INSERT_HEAD(&zombproc, p, p_list);
270 	p->p_stat = SZOMB;
271 
272 	LIST_REMOVE(p, p_hash);
273 	lockmgr(&allproc_lock, LK_RELEASE, NULL, CURPROC);
274 
275 	q = LIST_FIRST(&p->p_children);
276 	if (q)		/* only need this if any child is S_ZOMB */
277 		wakeup((caddr_t) initproc);
278 	for (; q != 0; q = nq) {
279 		nq = LIST_NEXT(q, p_sibling);
280 		LIST_REMOVE(q, p_sibling);
281 		LIST_INSERT_HEAD(&initproc->p_children, q, p_sibling);
282 		q->p_pptr = initproc;
283 		q->p_sigparent = SIGCHLD;
284 		/*
285 		 * Traced processes are killed
286 		 * since their existence means someone is screwing up.
287 		 */
288 		if (q->p_flag & P_TRACED) {
289 			q->p_flag &= ~P_TRACED;
290 			psignal(q, SIGKILL);
291 		}
292 	}
293 
294 	/*
295 	 * Save exit status and final rusage info, adding in child rusage
296 	 * info and self times.
297 	 */
298 	p->p_xstat = rv;
299 	*p->p_ru = p->p_stats->p_ru;
300 	calcru(p, &p->p_ru->ru_utime, &p->p_ru->ru_stime, NULL);
301 	ruadd(p->p_ru, &p->p_stats->p_cru);
302 
303 	/*
304 	 * Pretend that an mi_switch() to the next process occurs now.  We
305 	 * must set `switchtime' directly since we will call cpu_switch()
306 	 * directly.  Set it now so that the rest of the exit time gets
307 	 * counted somewhere if possible.
308 	 */
309 	microuptime(&switchtime);
310 	switchticks = ticks;
311 
312 	/*
313 	 * notify interested parties of our demise.
314 	 */
315 	KNOTE(&p->p_klist, NOTE_EXIT);
316 
317 	/*
318 	 * Notify parent that we're gone.  If parent has the PS_NOCLDWAIT
319 	 * flag set, notify process 1 instead (and hope it will handle
320 	 * this situation).
321 	 */
322 	if (p->p_pptr->p_procsig->ps_flag & PS_NOCLDWAIT) {
323 		struct proc *pp = p->p_pptr;
324 		proc_reparent(p, initproc);
325 		/*
326 		 * If this was the last child of our parent, notify
327 		 * parent, so in case he was wait(2)ing, he will
328 		 * continue.
329 		 */
330 		if (LIST_EMPTY(&pp->p_children))
331 			wakeup((caddr_t)pp);
332 	}
333 
334 	if (p->p_sigparent && p->p_pptr != initproc) {
335 	        psignal(p->p_pptr, p->p_sigparent);
336 	} else {
337 	        psignal(p->p_pptr, SIGCHLD);
338 	}
339 
340 	wakeup((caddr_t)p->p_pptr);
341 #if defined(tahoe)
342 	/* move this to cpu_exit */
343 	p->p_addr->u_pcb.pcb_savacc.faddr = (float *)NULL;
344 #endif
345 	/*
346 	 * Clear curproc after we've done all operations
347 	 * that could block, and before tearing down the rest
348 	 * of the process state that might be used from clock, etc.
349 	 * Also, can't clear curproc while we're still runnable,
350 	 * as we're not on a run queue (we are current, just not
351 	 * a proper proc any longer!).
352 	 *
353 	 * Other substructures are freed from wait().
354 	 */
355 	if (--p->p_limit->p_refcnt == 0) {
356 		FREE(p->p_limit, M_SUBPROC);
357 		p->p_limit = NULL;
358 	}
359 
360 	/*
361 	 * Finally, call machine-dependent code to release the remaining
362 	 * resources including address space, the kernel stack and pcb.
363 	 * The address space is released by "vmspace_free(p->p_vmspace)";
364 	 * This is machine-dependent, as we may have to change stacks
365 	 * or ensure that the current one isn't reallocated before we
366 	 * finish.  cpu_exit will end with a call to cpu_switch(), finishing
367 	 * our execution (pun intended).
368 	 */
369 	cpu_exit(p);
370 }
371 
372 #ifdef COMPAT_43
373 int
374 owait(p, uap)
375 	struct proc *p;
376 	register struct owait_args /* {
377 		int     dummy;
378 	} */ *uap;
379 {
380 	struct wait_args w;
381 
382 	w.options = 0;
383 	w.rusage = NULL;
384 	w.pid = WAIT_ANY;
385 	w.status = NULL;
386 	return (wait1(p, &w, 1));
387 }
388 #endif /* COMPAT_43 */
389 
390 int
391 wait4(p, uap)
392 	struct proc *p;
393 	struct wait_args *uap;
394 {
395 
396 	return (wait1(p, uap, 0));
397 }
398 
399 static int
400 wait1(q, uap, compat)
401 	register struct proc *q;
402 	register struct wait_args /* {
403 		int pid;
404 		int *status;
405 		int options;
406 		struct rusage *rusage;
407 	} */ *uap;
408 	int compat;
409 {
410 	register int nfound;
411 	register struct proc *p, *t;
412 	int status, error;
413 
414 	if (uap->pid == 0)
415 		uap->pid = -q->p_pgid;
416 	if (uap->options &~ (WUNTRACED|WNOHANG|WLINUXCLONE))
417 		return (EINVAL);
418 loop:
419 	nfound = 0;
420 	LIST_FOREACH(p, &q->p_children, p_sibling) {
421 		if (uap->pid != WAIT_ANY &&
422 		    p->p_pid != uap->pid && p->p_pgid != -uap->pid)
423 			continue;
424 
425 		/* This special case handles a kthread spawned by linux_clone
426 		 * (see linux_misc.c).  The linux_wait4 and linux_waitpid functions
427 		 * need to be able to distinguish between waiting on a process and
428 		 * waiting on a thread.  It is a thread if p_sigparent is not SIGCHLD,
429 		 * and the WLINUXCLONE option signifies we want to wait for threads
430 		 * and not processes.
431 		 */
432 		if ((p->p_sigparent != SIGCHLD) ^ ((uap->options & WLINUXCLONE) != 0))
433 			continue;
434 
435 		nfound++;
436 		if (p->p_stat == SZOMB) {
437 			/* charge childs scheduling cpu usage to parent */
438 			if (curproc->p_pid != 1) {
439 				curproc->p_estcpu =
440 				    ESTCPULIM(curproc->p_estcpu + p->p_estcpu);
441 			}
442 
443 			q->p_retval[0] = p->p_pid;
444 #ifdef COMPAT_43
445 			if (compat)
446 				q->p_retval[1] = p->p_xstat;
447 			else
448 #endif
449 			if (uap->status) {
450 				status = p->p_xstat;	/* convert to int */
451 				if ((error = copyout((caddr_t)&status,
452 				    (caddr_t)uap->status, sizeof(status))))
453 					return (error);
454 			}
455 			if (uap->rusage && (error = copyout((caddr_t)p->p_ru,
456 			    (caddr_t)uap->rusage, sizeof (struct rusage))))
457 				return (error);
458 			/*
459 			 * If we got the child via a ptrace 'attach',
460 			 * we need to give it back to the old parent.
461 			 */
462 			if (p->p_oppid && (t = pfind(p->p_oppid))) {
463 				p->p_oppid = 0;
464 				proc_reparent(p, t);
465 				psignal(t, SIGCHLD);
466 				wakeup((caddr_t)t);
467 				return (0);
468 			}
469 			p->p_xstat = 0;
470 			ruadd(&q->p_stats->p_cru, p->p_ru);
471 			FREE(p->p_ru, M_ZOMBIE);
472 			p->p_ru = NULL;
473 
474 			/*
475 			 * Decrement the count of procs running with this uid.
476 			 */
477 			(void)chgproccnt(p->p_cred->p_uidinfo, -1, 0);
478 
479 			/*
480 			 * Release reference to text vnode
481 			 */
482 			if (p->p_textvp)
483 				vrele(p->p_textvp);
484 
485 			/*
486 			 * Free up credentials.
487 			 */
488 			if (--p->p_cred->p_refcnt == 0) {
489 				crfree(p->p_ucred);
490 				uifree(p->p_cred->p_uidinfo);
491 				FREE(p->p_cred, M_SUBPROC);
492 				p->p_cred = NULL;
493 			}
494 
495 			/*
496 			 * Destroy empty prisons
497 			 */
498 			if (p->p_prison && !--p->p_prison->pr_ref) {
499 				if (p->p_prison->pr_linux != NULL)
500 					FREE(p->p_prison->pr_linux, M_PRISON);
501 				FREE(p->p_prison, M_PRISON);
502 			}
503 
504 			/*
505 			 * Remove unused arguments
506 			 */
507 			if (p->p_args && --p->p_args->ar_ref == 0)
508 				FREE(p->p_args, M_PARGS);
509 
510 			/*
511 			 * Finally finished with old proc entry.
512 			 * Unlink it from its process group and free it.
513 			 */
514 			leavepgrp(p);
515 			lockmgr(&allproc_lock, LK_EXCLUSIVE, NULL, CURPROC);
516 			LIST_REMOVE(p, p_list);	/* off zombproc */
517 			lockmgr(&allproc_lock, LK_RELEASE, NULL, CURPROC);
518 			LIST_REMOVE(p, p_sibling);
519 
520 			if (--p->p_procsig->ps_refcnt == 0) {
521 				if (p->p_sigacts != &p->p_addr->u_sigacts)
522 					FREE(p->p_sigacts, M_SUBPROC);
523 			        FREE(p->p_procsig, M_SUBPROC);
524 				p->p_procsig = NULL;
525 			}
526 
527 			/*
528 			 * Give machine-dependent layer a chance
529 			 * to free anything that cpu_exit couldn't
530 			 * release while still running in process context.
531 			 */
532 			cpu_wait(p);
533 			zfree(proc_zone, p);
534 			nprocs--;
535 			return (0);
536 		}
537 		if (p->p_stat == SSTOP && (p->p_flag & P_WAITED) == 0 &&
538 		    (p->p_flag & P_TRACED || uap->options & WUNTRACED)) {
539 			p->p_flag |= P_WAITED;
540 			q->p_retval[0] = p->p_pid;
541 #ifdef COMPAT_43
542 			if (compat) {
543 				q->p_retval[1] = W_STOPCODE(p->p_xstat);
544 				error = 0;
545 			} else
546 #endif
547 			if (uap->status) {
548 				status = W_STOPCODE(p->p_xstat);
549 				error = copyout((caddr_t)&status,
550 					(caddr_t)uap->status, sizeof(status));
551 			} else
552 				error = 0;
553 			return (error);
554 		}
555 	}
556 	if (nfound == 0)
557 		return (ECHILD);
558 	if (uap->options & WNOHANG) {
559 		q->p_retval[0] = 0;
560 		return (0);
561 	}
562 	if ((error = tsleep((caddr_t)q, PWAIT | PCATCH, "wait", 0)))
563 		return (error);
564 	goto loop;
565 }
566 
567 /*
568  * make process 'parent' the new parent of process 'child'.
569  */
570 void
571 proc_reparent(child, parent)
572 	register struct proc *child;
573 	register struct proc *parent;
574 {
575 
576 	if (child->p_pptr == parent)
577 		return;
578 
579 	LIST_REMOVE(child, p_sibling);
580 	LIST_INSERT_HEAD(&parent->p_children, child, p_sibling);
581 	child->p_pptr = parent;
582 }
583 
584 /*
585  * The next two functions are to handle adding/deleting items on the
586  * exit callout list
587  *
588  * at_exit():
589  * Take the arguments given and put them onto the exit callout list,
590  * However first make sure that it's not already there.
591  * returns 0 on success.
592  */
593 
594 int
595 at_exit(function)
596 	exitlist_fn function;
597 {
598 	struct exitlist *ep;
599 
600 #ifdef INVARIANTS
601 	/* Be noisy if the programmer has lost track of things */
602 	if (rm_at_exit(function))
603 		printf("WARNING: exit callout entry (%p) already present\n",
604 		    function);
605 #endif
606 	ep = malloc(sizeof(*ep), M_ATEXIT, M_NOWAIT);
607 	if (ep == NULL)
608 		return (ENOMEM);
609 	ep->function = function;
610 	TAILQ_INSERT_TAIL(&exit_list, ep, next);
611 	return (0);
612 }
613 
614 /*
615  * Scan the exit callout list for the given item and remove it.
616  * Returns the number of items removed (0 or 1)
617  */
618 int
619 rm_at_exit(function)
620 	exitlist_fn function;
621 {
622 	struct exitlist *ep;
623 
624 	TAILQ_FOREACH(ep, &exit_list, next) {
625 		if (ep->function == function) {
626 			TAILQ_REMOVE(&exit_list, ep, next);
627 			free(ep, M_ATEXIT);
628 			return(1);
629 		}
630 	}
631 	return (0);
632 }
633 
634 void check_sigacts (void)
635 {
636 	struct proc *p = curproc;
637 	struct sigacts *pss;
638 	int s;
639 
640 	if (p->p_procsig->ps_refcnt == 1 &&
641 	    p->p_sigacts != &p->p_addr->u_sigacts) {
642 		pss = p->p_sigacts;
643 		s = splhigh();
644 		p->p_addr->u_sigacts = *pss;
645 		p->p_sigacts = &p->p_addr->u_sigacts;
646 		splx(s);
647 		FREE(pss, M_SUBPROC);
648 	}
649 }
650 
651