xref: /freebsd/sys/kern/kern_exit.c (revision 5129159789cc9d7bc514e4546b88e3427695002d)
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/proc.h>
51 #include <sys/pioctl.h>
52 #include <sys/tty.h>
53 #include <sys/wait.h>
54 #include <sys/vnode.h>
55 #include <sys/resourcevar.h>
56 #include <sys/signalvar.h>
57 #include <sys/ptrace.h>
58 #include <sys/acct.h>		/* for acct_process() function prototype */
59 #include <sys/filedesc.h>
60 #include <sys/shm.h>
61 #include <sys/sem.h>
62 #include <sys/aio.h>
63 #include <sys/jail.h>
64 
65 #include <vm/vm.h>
66 #include <vm/vm_param.h>
67 #include <sys/lock.h>
68 #include <vm/pmap.h>
69 #include <vm/vm_map.h>
70 #include <vm/vm_zone.h>
71 #include <sys/user.h>
72 
73 /* Required to be non-static for SysVR4 emulator */
74 MALLOC_DEFINE(M_ZOMBIE, "zombie", "zombie proc status");
75 
76 static MALLOC_DEFINE(M_ATEXIT, "atexit", "atexit callback");
77 
78 static int wait1 __P((struct proc *, struct wait_args *, int));
79 
80 /*
81  * callout list for things to do at exit time
82  */
83 struct exitlist {
84 	exitlist_fn function;
85 	TAILQ_ENTRY(exitlist) next;
86 };
87 
88 TAILQ_HEAD(exit_list_head, exitlist);
89 static struct exit_list_head exit_list = TAILQ_HEAD_INITIALIZER(exit_list);
90 
91 /*
92  * exit --
93  *	Death of process.
94  */
95 void
96 exit(p, uap)
97 	struct proc *p;
98 	struct rexit_args /* {
99 		int	rval;
100 	} */ *uap;
101 {
102 
103 	exit1(p, W_EXITCODE(uap->rval, 0));
104 	/* NOTREACHED */
105 }
106 
107 /*
108  * Exit: deallocate address space and other resources, change proc state
109  * to zombie, and unlink proc from allproc and parent's lists.  Save exit
110  * status and rusage for wait().  Check for child processes and orphan them.
111  */
112 void
113 exit1(p, rv)
114 	register struct proc *p;
115 	int rv;
116 {
117 	register struct proc *q, *nq;
118 	register struct vmspace *vm;
119 	struct exitlist *ep;
120 
121 	if (p->p_pid == 1) {
122 		printf("init died (signal %d, exit %d)\n",
123 		    WTERMSIG(rv), WEXITSTATUS(rv));
124 		panic("Going nowhere without my init!");
125 	}
126 
127 	aio_proc_rundown(p);
128 
129 	/* are we a task leader? */
130 	if(p == p->p_leader) {
131         	struct kill_args killArgs;
132 		killArgs.signum = SIGKILL;
133 		q = p->p_peers;
134 		while(q) {
135 			killArgs.pid = q->p_pid;
136 			/*
137 		         * The interface for kill is better
138 			 * than the internal signal
139 			 */
140 			kill(p, &killArgs);
141 			nq = q;
142 			q = q->p_peers;
143 		}
144 		while (p->p_peers)
145 		  tsleep((caddr_t)p, PWAIT, "exit1", 0);
146 	}
147 
148 #ifdef PGINPROF
149 	vmsizmon();
150 #endif
151 	STOPEVENT(p, S_EXIT, rv);
152 
153 	/*
154 	 * Check if any loadable modules need anything done at process exit.
155 	 * e.g. SYSV IPC stuff
156 	 * XXX what if one of these generates an error?
157 	 */
158 	TAILQ_FOREACH(ep, &exit_list, next)
159 		(*ep->function)(p);
160 
161 	if (p->p_flag & P_PROFIL)
162 		stopprofclock(p);
163 	MALLOC(p->p_ru, struct rusage *, sizeof(struct rusage),
164 		M_ZOMBIE, M_WAITOK);
165 	/*
166 	 * If parent is waiting for us to exit or exec,
167 	 * P_PPWAIT is set; we will wakeup the parent below.
168 	 */
169 	p->p_flag &= ~(P_TRACED | P_PPWAIT);
170 	p->p_flag |= P_WEXIT;
171 	SIGEMPTYSET(p->p_siglist);
172 	if (timevalisset(&p->p_realtimer.it_value))
173 		untimeout(realitexpire, (caddr_t)p, p->p_ithandle);
174 
175 	/*
176 	 * Reset any sigio structures pointing to us as a result of
177 	 * F_SETOWN with our pid.
178 	 */
179 	funsetownlst(&p->p_sigiolst);
180 
181 	/*
182 	 * Close open files and release open-file table.
183 	 * This may block!
184 	 */
185 	fdfree(p);
186 
187 	if(p->p_leader->p_peers) {
188 		q = p->p_leader;
189 		while(q->p_peers != p)
190 			q = q->p_peers;
191 		q->p_peers = p->p_peers;
192 		wakeup((caddr_t)p->p_leader);
193 	}
194 
195 	/*
196 	 * XXX Shutdown SYSV semaphores
197 	 */
198 	semexit(p);
199 
200 	/* The next two chunks should probably be moved to vmspace_exit. */
201 	vm = p->p_vmspace;
202 	/*
203 	 * Release user portion of address space.
204 	 * This releases references to vnodes,
205 	 * which could cause I/O if the file has been unlinked.
206 	 * Need to do this early enough that we can still sleep.
207 	 * Can't free the entire vmspace as the kernel stack
208 	 * may be mapped within that space also.
209 	 */
210 	if (vm->vm_refcnt == 1) {
211 		if (vm->vm_shm)
212 			shmexit(p);
213 		pmap_remove_pages(vmspace_pmap(vm), VM_MIN_ADDRESS,
214 		    VM_MAXUSER_ADDRESS);
215 		(void) vm_map_remove(&vm->vm_map, VM_MIN_ADDRESS,
216 		    VM_MAXUSER_ADDRESS);
217 	}
218 
219 	if (SESS_LEADER(p)) {
220 		register struct session *sp = p->p_session;
221 
222 		if (sp->s_ttyvp) {
223 			/*
224 			 * Controlling process.
225 			 * Signal foreground pgrp,
226 			 * drain controlling terminal
227 			 * and revoke access to controlling terminal.
228 			 */
229 			if (sp->s_ttyp && (sp->s_ttyp->t_session == sp)) {
230 				if (sp->s_ttyp->t_pgrp)
231 					pgsignal(sp->s_ttyp->t_pgrp, SIGHUP, 1);
232 				(void) ttywait(sp->s_ttyp);
233 				/*
234 				 * The tty could have been revoked
235 				 * if we blocked.
236 				 */
237 				if (sp->s_ttyvp)
238 					VOP_REVOKE(sp->s_ttyvp, REVOKEALL);
239 			}
240 			if (sp->s_ttyvp)
241 				vrele(sp->s_ttyvp);
242 			sp->s_ttyvp = NULL;
243 			/*
244 			 * s_ttyp is not zero'd; we use this to indicate
245 			 * that the session once had a controlling terminal.
246 			 * (for logging and informational purposes)
247 			 */
248 		}
249 		sp->s_leader = NULL;
250 	}
251 	fixjobc(p, p->p_pgrp, 0);
252 	(void)acct_process(p);
253 #ifdef KTRACE
254 	/*
255 	 * release trace file
256 	 */
257 	p->p_traceflag = 0;	/* don't trace the vrele() */
258 	if (p->p_tracep)
259 		vrele(p->p_tracep);
260 #endif
261 	/*
262 	 * Remove proc from allproc queue and pidhash chain.
263 	 * Place onto zombproc.  Unlink from parent's child list.
264 	 */
265 	LIST_REMOVE(p, p_list);
266 	LIST_INSERT_HEAD(&zombproc, p, p_list);
267 	p->p_stat = SZOMB;
268 
269 	LIST_REMOVE(p, p_hash);
270 
271 	q = LIST_FIRST(&p->p_children);
272 	if (q)		/* only need this if any child is S_ZOMB */
273 		wakeup((caddr_t) initproc);
274 	for (; q != 0; q = nq) {
275 		nq = LIST_NEXT(q, p_sibling);
276 		LIST_REMOVE(q, p_sibling);
277 		LIST_INSERT_HEAD(&initproc->p_children, q, p_sibling);
278 		q->p_pptr = initproc;
279 		q->p_sigparent = SIGCHLD;
280 		/*
281 		 * Traced processes are killed
282 		 * since their existence means someone is screwing up.
283 		 */
284 		if (q->p_flag & P_TRACED) {
285 			q->p_flag &= ~P_TRACED;
286 			psignal(q, SIGKILL);
287 		}
288 	}
289 
290 	/*
291 	 * Save exit status and final rusage info, adding in child rusage
292 	 * info and self times.
293 	 */
294 	p->p_xstat = rv;
295 	*p->p_ru = p->p_stats->p_ru;
296 	calcru(p, &p->p_ru->ru_utime, &p->p_ru->ru_stime, NULL);
297 	ruadd(p->p_ru, &p->p_stats->p_cru);
298 
299 	/*
300 	 * Pretend that an mi_switch() to the next process occurs now.  We
301 	 * must set `switchtime' directly since we will call cpu_switch()
302 	 * directly.  Set it now so that the rest of the exit time gets
303 	 * counted somewhere if possible.
304 	 */
305 	microuptime(&switchtime);
306 	switchticks = ticks;
307 
308 	/*
309 	 * Notify parent that we're gone.  If parent has the PS_NOCLDWAIT
310 	 * flag set, notify process 1 instead (and hope it will handle
311 	 * this situation).
312 	 */
313 	if (p->p_pptr->p_procsig->ps_flag & PS_NOCLDWAIT) {
314 		struct proc *pp = p->p_pptr;
315 		proc_reparent(p, initproc);
316 		/*
317 		 * If this was the last child of our parent, notify
318 		 * parent, so in case he was wait(2)ing, he will
319 		 * continue.
320 		 */
321 		if (LIST_EMPTY(&pp->p_children))
322 			wakeup((caddr_t)pp);
323 	}
324 
325 	if (p->p_sigparent && p->p_pptr != initproc) {
326 	        psignal(p->p_pptr, p->p_sigparent);
327 	} else {
328 	        psignal(p->p_pptr, SIGCHLD);
329 	}
330 
331 	wakeup((caddr_t)p->p_pptr);
332 #if defined(tahoe)
333 	/* move this to cpu_exit */
334 	p->p_addr->u_pcb.pcb_savacc.faddr = (float *)NULL;
335 #endif
336 	/*
337 	 * Clear curproc after we've done all operations
338 	 * that could block, and before tearing down the rest
339 	 * of the process state that might be used from clock, etc.
340 	 * Also, can't clear curproc while we're still runnable,
341 	 * as we're not on a run queue (we are current, just not
342 	 * a proper proc any longer!).
343 	 *
344 	 * Other substructures are freed from wait().
345 	 */
346 	SET_CURPROC(NULL);
347 	if (--p->p_limit->p_refcnt == 0) {
348 		FREE(p->p_limit, M_SUBPROC);
349 		p->p_limit = NULL;
350 	}
351 
352 	/*
353 	 * Finally, call machine-dependent code to release the remaining
354 	 * resources including address space, the kernel stack and pcb.
355 	 * The address space is released by "vmspace_free(p->p_vmspace)";
356 	 * This is machine-dependent, as we may have to change stacks
357 	 * or ensure that the current one isn't reallocated before we
358 	 * finish.  cpu_exit will end with a call to cpu_switch(), finishing
359 	 * our execution (pun intended).
360 	 */
361 	cpu_exit(p);
362 }
363 
364 #ifdef COMPAT_43
365 int
366 owait(p, uap)
367 	struct proc *p;
368 	register struct owait_args /* {
369 		int     dummy;
370 	} */ *uap;
371 {
372 	struct wait_args w;
373 
374 	w.options = 0;
375 	w.rusage = NULL;
376 	w.pid = WAIT_ANY;
377 	w.status = NULL;
378 	return (wait1(p, &w, 1));
379 }
380 #endif /* COMPAT_43 */
381 
382 int
383 wait4(p, uap)
384 	struct proc *p;
385 	struct wait_args *uap;
386 {
387 
388 	return (wait1(p, uap, 0));
389 }
390 
391 static int
392 wait1(q, uap, compat)
393 	register struct proc *q;
394 	register struct wait_args /* {
395 		int pid;
396 		int *status;
397 		int options;
398 		struct rusage *rusage;
399 	} */ *uap;
400 	int compat;
401 {
402 	register int nfound;
403 	register struct proc *p, *t;
404 	int status, error;
405 
406 	if (uap->pid == 0)
407 		uap->pid = -q->p_pgid;
408 	if (uap->options &~ (WUNTRACED|WNOHANG|WLINUXCLONE))
409 		return (EINVAL);
410 loop:
411 	nfound = 0;
412 	LIST_FOREACH(p, &q->p_children, p_sibling) {
413 		if (uap->pid != WAIT_ANY &&
414 		    p->p_pid != uap->pid && p->p_pgid != -uap->pid)
415 			continue;
416 
417 		/* This special case handles a kthread spawned by linux_clone
418 		 * (see linux_misc.c).  The linux_wait4 and linux_waitpid functions
419 		 * need to be able to distinguish between waiting on a process and
420 		 * waiting on a thread.  It is a thread if p_sigparent is not SIGCHLD,
421 		 * and the WLINUXCLONE option signifies we want to wait for threads
422 		 * and not processes.
423 		 */
424 		if ((p->p_sigparent != SIGCHLD) ^ ((uap->options & WLINUXCLONE) != 0))
425 			continue;
426 
427 		nfound++;
428 		if (p->p_stat == SZOMB) {
429 			/* charge childs scheduling cpu usage to parent */
430 			if (curproc->p_pid != 1) {
431 				curproc->p_estcpu =
432 				    ESTCPULIM(curproc->p_estcpu + p->p_estcpu);
433 			}
434 
435 			q->p_retval[0] = p->p_pid;
436 #ifdef COMPAT_43
437 			if (compat)
438 				q->p_retval[1] = p->p_xstat;
439 			else
440 #endif
441 			if (uap->status) {
442 				status = p->p_xstat;	/* convert to int */
443 				if ((error = copyout((caddr_t)&status,
444 				    (caddr_t)uap->status, sizeof(status))))
445 					return (error);
446 			}
447 			if (uap->rusage && (error = copyout((caddr_t)p->p_ru,
448 			    (caddr_t)uap->rusage, sizeof (struct rusage))))
449 				return (error);
450 			/*
451 			 * If we got the child via a ptrace 'attach',
452 			 * we need to give it back to the old parent.
453 			 */
454 			if (p->p_oppid && (t = pfind(p->p_oppid))) {
455 				p->p_oppid = 0;
456 				proc_reparent(p, t);
457 				psignal(t, SIGCHLD);
458 				wakeup((caddr_t)t);
459 				return (0);
460 			}
461 			p->p_xstat = 0;
462 			ruadd(&q->p_stats->p_cru, p->p_ru);
463 			FREE(p->p_ru, M_ZOMBIE);
464 			p->p_ru = NULL;
465 
466 			/*
467 			 * Decrement the count of procs running with this uid.
468 			 */
469 			(void)chgproccnt(p->p_cred->p_ruid, -1);
470 
471 			/*
472 			 * Release reference to text vnode
473 			 */
474 			if (p->p_textvp)
475 				vrele(p->p_textvp);
476 
477 			/*
478 			 * Free up credentials.
479 			 */
480 			if (--p->p_cred->p_refcnt == 0) {
481 				crfree(p->p_ucred);
482 				FREE(p->p_cred, M_SUBPROC);
483 				p->p_cred = NULL;
484 			}
485 
486 			/*
487 			 * Destroy empty prisons
488 			 */
489 			if (p->p_prison && !--p->p_prison->pr_ref) {
490 				if (p->p_prison->pr_linux != NULL)
491 					FREE(p->p_prison->pr_linux, M_PRISON);
492 				FREE(p->p_prison, M_PRISON);
493 			}
494 
495 			/*
496 			 * Remove unused arguments
497 			 */
498 			if (p->p_args && --p->p_args->ar_ref == 0)
499 				FREE(p->p_args, M_PARGS);
500 
501 			/*
502 			 * Finally finished with old proc entry.
503 			 * Unlink it from its process group and free it.
504 			 */
505 			leavepgrp(p);
506 			LIST_REMOVE(p, p_list);	/* off zombproc */
507 			LIST_REMOVE(p, p_sibling);
508 
509 			if (--p->p_procsig->ps_refcnt == 0) {
510 				if (p->p_sigacts != &p->p_addr->u_sigacts)
511 					FREE(p->p_sigacts, M_SUBPROC);
512 			        FREE(p->p_procsig, M_SUBPROC);
513 				p->p_procsig = NULL;
514 			}
515 
516 			/*
517 			 * Give machine-dependent layer a chance
518 			 * to free anything that cpu_exit couldn't
519 			 * release while still running in process context.
520 			 */
521 			cpu_wait(p);
522 			zfree(proc_zone, p);
523 			nprocs--;
524 			return (0);
525 		}
526 		if (p->p_stat == SSTOP && (p->p_flag & P_WAITED) == 0 &&
527 		    (p->p_flag & P_TRACED || uap->options & WUNTRACED)) {
528 			p->p_flag |= P_WAITED;
529 			q->p_retval[0] = p->p_pid;
530 #ifdef COMPAT_43
531 			if (compat) {
532 				q->p_retval[1] = W_STOPCODE(p->p_xstat);
533 				error = 0;
534 			} else
535 #endif
536 			if (uap->status) {
537 				status = W_STOPCODE(p->p_xstat);
538 				error = copyout((caddr_t)&status,
539 					(caddr_t)uap->status, sizeof(status));
540 			} else
541 				error = 0;
542 			return (error);
543 		}
544 	}
545 	if (nfound == 0)
546 		return (ECHILD);
547 	if (uap->options & WNOHANG) {
548 		q->p_retval[0] = 0;
549 		return (0);
550 	}
551 	if ((error = tsleep((caddr_t)q, PWAIT | PCATCH, "wait", 0)))
552 		return (error);
553 	goto loop;
554 }
555 
556 /*
557  * make process 'parent' the new parent of process 'child'.
558  */
559 void
560 proc_reparent(child, parent)
561 	register struct proc *child;
562 	register struct proc *parent;
563 {
564 
565 	if (child->p_pptr == parent)
566 		return;
567 
568 	LIST_REMOVE(child, p_sibling);
569 	LIST_INSERT_HEAD(&parent->p_children, child, p_sibling);
570 	child->p_pptr = parent;
571 }
572 
573 /*
574  * The next two functions are to handle adding/deleting items on the
575  * exit callout list
576  *
577  * at_exit():
578  * Take the arguments given and put them onto the exit callout list,
579  * However first make sure that it's not already there.
580  * returns 0 on success.
581  */
582 
583 int
584 at_exit(function)
585 	exitlist_fn function;
586 {
587 	struct exitlist *ep;
588 
589 #ifdef INVARIANTS
590 	/* Be noisy if the programmer has lost track of things */
591 	if (rm_at_exit(function))
592 		printf("WARNING: exit callout entry (%p) already present\n",
593 		    function);
594 #endif
595 	ep = malloc(sizeof(*ep), M_ATEXIT, M_NOWAIT);
596 	if (ep == NULL)
597 		return (ENOMEM);
598 	ep->function = function;
599 	TAILQ_INSERT_TAIL(&exit_list, ep, next);
600 	return (0);
601 }
602 
603 /*
604  * Scan the exit callout list for the given item and remove it.
605  * Returns the number of items removed (0 or 1)
606  */
607 int
608 rm_at_exit(function)
609 	exitlist_fn function;
610 {
611 	struct exitlist *ep;
612 
613 	TAILQ_FOREACH(ep, &exit_list, next) {
614 		if (ep->function == function) {
615 			TAILQ_REMOVE(&exit_list, ep, next);
616 			free(ep, M_ATEXIT);
617 			return(1);
618 		}
619 	}
620 	return (0);
621 }
622 
623 void check_sigacts (void)
624 {
625 	struct proc *p = curproc;
626 	struct sigacts *pss;
627 	int s;
628 
629 	if (p->p_procsig->ps_refcnt == 1 &&
630 	    p->p_sigacts != &p->p_addr->u_sigacts) {
631 		pss = p->p_sigacts;
632 		s = splhigh();
633 		p->p_addr->u_sigacts = *pss;
634 		p->p_sigacts = &p->p_addr->u_sigacts;
635 		splx(s);
636 		FREE(pss, M_SUBPROC);
637 	}
638 }
639 
640