xref: /freebsd/sys/kern/kern_exit.c (revision 6e8394b8baa7d5d9153ab90de6824bcd19b3b4e1)
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  * $Id: kern_exit.c,v 1.80 1999/04/28 11:36:52 phk Exp $
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 #ifdef COMPAT_43
66 #include <machine/reg.h>
67 #include <machine/psl.h>
68 #endif
69 #include <machine/limits.h>	/* for UCHAR_MAX = typeof(p_priority)_MAX */
70 
71 #include <vm/vm.h>
72 #include <vm/vm_param.h>
73 #include <sys/lock.h>
74 #include <vm/pmap.h>
75 #include <vm/vm_map.h>
76 #include <vm/vm_zone.h>
77 #include <sys/user.h>
78 
79 /* Required to be non-static for SysVR4 emulator */
80 MALLOC_DEFINE(M_ZOMBIE, "zombie", "zombie proc status");
81 
82 static int wait1 __P((struct proc *, struct wait_args *, int));
83 
84 /*
85  * callout list for things to do at exit time
86  */
87 typedef struct exit_list_element {
88 	struct exit_list_element *next;
89 	exitlist_fn function;
90 } *ele_p;
91 
92 static ele_p exit_list;
93 
94 /*
95  * exit --
96  *	Death of process.
97  */
98 void
99 exit(p, uap)
100 	struct proc *p;
101 	struct rexit_args /* {
102 		int	rval;
103 	} */ *uap;
104 {
105 
106 	exit1(p, W_EXITCODE(uap->rval, 0));
107 	/* NOTREACHED */
108 }
109 
110 /*
111  * Exit: deallocate address space and other resources, change proc state
112  * to zombie, and unlink proc from allproc and parent's lists.  Save exit
113  * status and rusage for wait().  Check for child processes and orphan them.
114  */
115 void
116 exit1(p, rv)
117 	register struct proc *p;
118 	int rv;
119 {
120 	register struct proc *q, *nq;
121 	register struct vmspace *vm;
122 	ele_p ep = exit_list;
123 
124 	if (p->p_pid == 1) {
125 		printf("init died (signal %d, exit %d)\n",
126 		    WTERMSIG(rv), WEXITSTATUS(rv));
127 		panic("Going nowhere without my init!");
128 	}
129 
130 	aio_proc_rundown(p);
131 
132 	/* are we a task leader? */
133 	if(p == p->p_leader) {
134         	struct kill_args killArgs;
135 		killArgs.signum = SIGKILL;
136 		q = p->p_peers;
137 		while(q) {
138 			killArgs.pid = q->p_pid;
139 			/*
140 		         * The interface for kill is better
141 			 * than the internal signal
142 			 */
143 			kill(p, &killArgs);
144 			nq = q;
145 			q = q->p_peers;
146 		}
147 		while (p->p_peers)
148 		  tsleep((caddr_t)p, PWAIT, "exit1", 0);
149 	}
150 
151 #ifdef PGINPROF
152 	vmsizmon();
153 #endif
154 	STOPEVENT(p, S_EXIT, rv);
155 
156 	/*
157 	 * Check if any loadable modules need anything done at process exit.
158 	 * e.g. SYSV IPC stuff
159 	 * XXX what if one of these generates an error?
160 	 */
161 	while (ep) {
162 		(*ep->function)(p);
163 		ep = ep->next;
164 	}
165 
166 	if (p->p_flag & P_PROFIL)
167 		stopprofclock(p);
168 	MALLOC(p->p_ru, struct rusage *, sizeof(struct rusage),
169 		M_ZOMBIE, M_WAITOK);
170 	/*
171 	 * If parent is waiting for us to exit or exec,
172 	 * P_PPWAIT is set; we will wakeup the parent below.
173 	 */
174 	p->p_flag &= ~(P_TRACED | P_PPWAIT);
175 	p->p_flag |= P_WEXIT;
176 	p->p_siglist = 0;
177 	if (timevalisset(&p->p_realtimer.it_value))
178 		untimeout(realitexpire, (caddr_t)p, p->p_ithandle);
179 
180 	/*
181 	 * Reset any sigio structures pointing to us as a result of
182 	 * F_SETOWN with our pid.
183 	 */
184 	funsetownlst(&p->p_sigiolst);
185 
186 	/*
187 	 * Close open files and release open-file table.
188 	 * This may block!
189 	 */
190 	fdfree(p);
191 
192 	if(p->p_leader->p_peers) {
193 		q = p->p_leader;
194 		while(q->p_peers != p)
195 			q = q->p_peers;
196 		q->p_peers = p->p_peers;
197 		wakeup((caddr_t)p->p_leader);
198 	}
199 
200 	/*
201 	 * XXX Shutdown SYSV semaphores
202 	 */
203 	semexit(p);
204 
205 	/* The next two chunks should probably be moved to vmspace_exit. */
206 	vm = p->p_vmspace;
207 	/*
208 	 * Release user portion of address space.
209 	 * This releases references to vnodes,
210 	 * which could cause I/O if the file has been unlinked.
211 	 * Need to do this early enough that we can still sleep.
212 	 * Can't free the entire vmspace as the kernel stack
213 	 * may be mapped within that space also.
214 	 */
215 	if (vm->vm_refcnt == 1) {
216 		if (vm->vm_shm)
217 			shmexit(p);
218 		pmap_remove_pages(vmspace_pmap(vm), VM_MIN_ADDRESS,
219 		    VM_MAXUSER_ADDRESS);
220 		(void) vm_map_remove(&vm->vm_map, VM_MIN_ADDRESS,
221 		    VM_MAXUSER_ADDRESS);
222 	}
223 
224 	if (SESS_LEADER(p)) {
225 		register struct session *sp = p->p_session;
226 
227 		if (sp->s_ttyvp) {
228 			/*
229 			 * Controlling process.
230 			 * Signal foreground pgrp,
231 			 * drain controlling terminal
232 			 * and revoke access to controlling terminal.
233 			 */
234 			if (sp->s_ttyp && (sp->s_ttyp->t_session == sp)) {
235 				if (sp->s_ttyp->t_pgrp)
236 					pgsignal(sp->s_ttyp->t_pgrp, SIGHUP, 1);
237 				(void) ttywait(sp->s_ttyp);
238 				/*
239 				 * The tty could have been revoked
240 				 * if we blocked.
241 				 */
242 				if (sp->s_ttyvp)
243 					VOP_REVOKE(sp->s_ttyvp, REVOKEALL);
244 			}
245 			if (sp->s_ttyvp)
246 				vrele(sp->s_ttyvp);
247 			sp->s_ttyvp = NULL;
248 			/*
249 			 * s_ttyp is not zero'd; we use this to indicate
250 			 * that the session once had a controlling terminal.
251 			 * (for logging and informational purposes)
252 			 */
253 		}
254 		sp->s_leader = NULL;
255 	}
256 	fixjobc(p, p->p_pgrp, 0);
257 	(void)acct_process(p);
258 #ifdef KTRACE
259 	/*
260 	 * release trace file
261 	 */
262 	p->p_traceflag = 0;	/* don't trace the vrele() */
263 	if (p->p_tracep)
264 		vrele(p->p_tracep);
265 #endif
266 	/*
267 	 * Remove proc from allproc queue and pidhash chain.
268 	 * Place onto zombproc.  Unlink from parent's child list.
269 	 */
270 	LIST_REMOVE(p, p_list);
271 	LIST_INSERT_HEAD(&zombproc, p, p_list);
272 	p->p_stat = SZOMB;
273 
274 	LIST_REMOVE(p, p_hash);
275 
276 	q = p->p_children.lh_first;
277 	if (q)		/* only need this if any child is S_ZOMB */
278 		wakeup((caddr_t) initproc);
279 	for (; q != 0; q = nq) {
280 		nq = q->p_sibling.le_next;
281 		LIST_REMOVE(q, p_sibling);
282 		LIST_INSERT_HEAD(&initproc->p_children, q, p_sibling);
283 		q->p_pptr = initproc;
284 		q->p_sigparent = SIGCHLD;
285 		/*
286 		 * Traced processes are killed
287 		 * since their existence means someone is screwing up.
288 		 */
289 		if (q->p_flag & P_TRACED) {
290 			q->p_flag &= ~P_TRACED;
291 			psignal(q, SIGKILL);
292 		}
293 	}
294 
295 	/*
296 	 * Save exit status and final rusage info, adding in child rusage
297 	 * info and self times.
298 	 */
299 	p->p_xstat = rv;
300 	*p->p_ru = p->p_stats->p_ru;
301 	calcru(p, &p->p_ru->ru_utime, &p->p_ru->ru_stime, NULL);
302 	ruadd(p->p_ru, &p->p_stats->p_cru);
303 
304 	/*
305 	 * Pretend that an mi_switch() to the next process occurs now.  We
306 	 * must set `switchtime' directly since we will call cpu_switch()
307 	 * directly.  Set it now so that the rest of the exit time gets
308 	 * counted somewhere if possible.
309 	 */
310 	microuptime(&switchtime);
311 	switchticks = ticks;
312 
313 	/*
314 	 * Notify parent that we're gone.  If parent has the P_NOCLDWAIT
315 	 * flag set, notify process 1 instead (and hope it will handle
316 	 * this situation).
317 	 */
318 	if (p->p_pptr->p_procsig->ps_flag & P_NOCLDWAIT) {
319 		struct proc *pp = p->p_pptr;
320 		proc_reparent(p, initproc);
321 		/*
322 		 * If this was the last child of our parent, notify
323 		 * parent, so in case he was wait(2)ing, he will
324 		 * continue.
325 		 */
326 		if (LIST_EMPTY(&pp->p_children))
327 			wakeup((caddr_t)pp);
328 	}
329 
330 	if (p->p_sigparent && p->p_pptr != initproc) {
331 	        psignal(p->p_pptr, p->p_sigparent);
332 	} else {
333 	        psignal(p->p_pptr, SIGCHLD);
334 	}
335 
336 	wakeup((caddr_t)p->p_pptr);
337 #if defined(tahoe)
338 	/* move this to cpu_exit */
339 	p->p_addr->u_pcb.pcb_savacc.faddr = (float *)NULL;
340 #endif
341 	/*
342 	 * Clear curproc after we've done all operations
343 	 * that could block, and before tearing down the rest
344 	 * of the process state that might be used from clock, etc.
345 	 * Also, can't clear curproc while we're still runnable,
346 	 * as we're not on a run queue (we are current, just not
347 	 * a proper proc any longer!).
348 	 *
349 	 * Other substructures are freed from wait().
350 	 */
351 	SET_CURPROC(NULL);
352 	if (--p->p_limit->p_refcnt == 0) {
353 		FREE(p->p_limit, M_SUBPROC);
354 		p->p_limit = NULL;
355 	}
356 
357 	/*
358 	 * Finally, call machine-dependent code to release the remaining
359 	 * resources including address space, the kernel stack and pcb.
360 	 * The address space is released by "vmspace_free(p->p_vmspace)";
361 	 * This is machine-dependent, as we may have to change stacks
362 	 * or ensure that the current one isn't reallocated before we
363 	 * finish.  cpu_exit will end with a call to cpu_switch(), finishing
364 	 * our execution (pun intended).
365 	 */
366 	cpu_exit(p);
367 }
368 
369 #ifdef COMPAT_43
370 #if defined(hp300) || defined(luna68k)
371 #include <machine/frame.h>
372 #define GETPS(rp)	((struct frame *)(rp))->f_sr
373 #else
374 #define GETPS(rp)	(rp)[PS]
375 #endif
376 
377 int
378 owait(p, uap)
379 	struct proc *p;
380 	register struct owait_args /* {
381 		int     dummy;
382 	} */ *uap;
383 {
384 	struct wait_args w;
385 
386 #ifdef PSL_ALLCC
387 	if ((GETPS(p->p_md.md_regs) & PSL_ALLCC) != PSL_ALLCC) {
388 		w.options = 0;
389 		w.rusage = NULL;
390 	} else {
391 		w.options = p->p_md.md_regs[R0];
392 		w.rusage = (struct rusage *)p->p_md.md_regs[R1];
393 	}
394 #else
395 	w.options = 0;
396 	w.rusage = NULL;
397 #endif
398 	w.pid = WAIT_ANY;
399 	w.status = NULL;
400 	return (wait1(p, &w, 1));
401 }
402 #endif /* COMPAT_43 */
403 
404 int
405 wait4(p, uap)
406 	struct proc *p;
407 	struct wait_args *uap;
408 {
409 
410 	return (wait1(p, uap, 0));
411 }
412 
413 static int
414 wait1(q, uap, compat)
415 	register struct proc *q;
416 	register struct wait_args /* {
417 		int pid;
418 		int *status;
419 		int options;
420 		struct rusage *rusage;
421 	} */ *uap;
422 	int compat;
423 {
424 	register int nfound;
425 	register struct proc *p, *t;
426 	int status, error;
427 
428 	if (uap->pid == 0)
429 		uap->pid = -q->p_pgid;
430 	if (uap->options &~ (WUNTRACED|WNOHANG|WLINUXCLONE))
431 		return (EINVAL);
432 loop:
433 	nfound = 0;
434 	for (p = q->p_children.lh_first; p != 0; p = p->p_sibling.le_next) {
435 		if (uap->pid != WAIT_ANY &&
436 		    p->p_pid != uap->pid && p->p_pgid != -uap->pid)
437 			continue;
438 
439 		/* This special case handles a kthread spawned by linux_clone
440 		 * (see linux_misc.c).  The linux_wait4 and linux_waitpid functions
441 		 * need to be able to distinguish between waiting on a process and
442 		 * waiting on a thread.  It is a thread if p_sigparent is not SIGCHLD,
443 		 * and the WLINUXCLONE option signifies we want to wait for threads
444 		 * and not processes.
445 		 */
446 		if ((p->p_sigparent != SIGCHLD) ^ ((uap->options & WLINUXCLONE) != 0))
447 			continue;
448 
449 		nfound++;
450 		if (p->p_stat == SZOMB) {
451 			/* charge childs scheduling cpu usage to parent */
452 			if (curproc->p_pid != 1) {
453 				curproc->p_estcpu = min(curproc->p_estcpu +
454 				    p->p_estcpu, UCHAR_MAX);
455 			}
456 
457 			q->p_retval[0] = p->p_pid;
458 #ifdef COMPAT_43
459 			if (compat)
460 				q->p_retval[1] = p->p_xstat;
461 			else
462 #endif
463 			if (uap->status) {
464 				status = p->p_xstat;	/* convert to int */
465 				if ((error = copyout((caddr_t)&status,
466 				    (caddr_t)uap->status, sizeof(status))))
467 					return (error);
468 			}
469 			if (uap->rusage && (error = copyout((caddr_t)p->p_ru,
470 			    (caddr_t)uap->rusage, sizeof (struct rusage))))
471 				return (error);
472 			/*
473 			 * If we got the child via a ptrace 'attach',
474 			 * we need to give it back to the old parent.
475 			 */
476 			if (p->p_oppid && (t = pfind(p->p_oppid))) {
477 				p->p_oppid = 0;
478 				proc_reparent(p, t);
479 				psignal(t, SIGCHLD);
480 				wakeup((caddr_t)t);
481 				return (0);
482 			}
483 			p->p_xstat = 0;
484 			ruadd(&q->p_stats->p_cru, p->p_ru);
485 			FREE(p->p_ru, M_ZOMBIE);
486 			p->p_ru = NULL;
487 
488 			/*
489 			 * Decrement the count of procs running with this uid.
490 			 */
491 			(void)chgproccnt(p->p_cred->p_ruid, -1);
492 
493 			/*
494 			 * Release reference to text vnode
495 			 */
496 			if (p->p_textvp)
497 				vrele(p->p_textvp);
498 
499 			/*
500 			 * Free up credentials.
501 			 */
502 			if (--p->p_cred->p_refcnt == 0) {
503 				crfree(p->p_cred->pc_ucred);
504 				FREE(p->p_cred, M_SUBPROC);
505 				p->p_cred = NULL;
506 			}
507 
508 			/*
509 			 * Destroy empty prisons
510 			 */
511 			if (p->p_prison && !--p->p_prison->pr_ref)
512 				FREE(p->p_prison, M_PRISON);
513 
514 			/*
515 			 * Finally finished with old proc entry.
516 			 * Unlink it from its process group and free it.
517 			 */
518 			leavepgrp(p);
519 			LIST_REMOVE(p, p_list);	/* off zombproc */
520 			LIST_REMOVE(p, p_sibling);
521 
522 			if (--p->p_procsig->ps_refcnt == 0) {
523 				if (p->p_sigacts != &p->p_addr->u_sigacts)
524 					FREE(p->p_sigacts, M_SUBPROC);
525 			        FREE(p->p_procsig, M_SUBPROC);
526 				p->p_procsig = NULL;
527 			}
528 
529 			/*
530 			 * Give machine-dependent layer a chance
531 			 * to free anything that cpu_exit couldn't
532 			 * release while still running in process context.
533 			 */
534 			cpu_wait(p);
535 			zfree(proc_zone, p);
536 			nprocs--;
537 			return (0);
538 		}
539 		if (p->p_stat == SSTOP && (p->p_flag & P_WAITED) == 0 &&
540 		    (p->p_flag & P_TRACED || uap->options & WUNTRACED)) {
541 			p->p_flag |= P_WAITED;
542 			q->p_retval[0] = p->p_pid;
543 #ifdef COMPAT_43
544 			if (compat) {
545 				q->p_retval[1] = W_STOPCODE(p->p_xstat);
546 				error = 0;
547 			} else
548 #endif
549 			if (uap->status) {
550 				status = W_STOPCODE(p->p_xstat);
551 				error = copyout((caddr_t)&status,
552 					(caddr_t)uap->status, sizeof(status));
553 			} else
554 				error = 0;
555 			return (error);
556 		}
557 	}
558 	if (nfound == 0)
559 		return (ECHILD);
560 	if (uap->options & WNOHANG) {
561 		q->p_retval[0] = 0;
562 		return (0);
563 	}
564 	if ((error = tsleep((caddr_t)q, PWAIT | PCATCH, "wait", 0)))
565 		return (error);
566 	goto loop;
567 }
568 
569 /*
570  * make process 'parent' the new parent of process 'child'.
571  */
572 void
573 proc_reparent(child, parent)
574 	register struct proc *child;
575 	register struct proc *parent;
576 {
577 
578 	if (child->p_pptr == parent)
579 		return;
580 
581 	LIST_REMOVE(child, p_sibling);
582 	LIST_INSERT_HEAD(&parent->p_children, child, p_sibling);
583 	child->p_pptr = parent;
584 }
585 
586 /*
587  * The next two functions are to handle adding/deleting items on the
588  * exit callout list
589  *
590  * at_exit():
591  * Take the arguments given and put them onto the exit callout list,
592  * However first make sure that it's not already there.
593  * returns 0 on success.
594  */
595 int
596 at_exit(function)
597 	exitlist_fn function;
598 {
599 	ele_p ep;
600 
601 	/* Be noisy if the programmer has lost track of things */
602 	if (rm_at_exit(function))
603 		printf("exit callout entry already present\n");
604 	ep = malloc(sizeof(*ep), M_TEMP, M_NOWAIT);
605 	if (ep == NULL)
606 		return (ENOMEM);
607 	ep->next = exit_list;
608 	ep->function = function;
609 	exit_list = ep;
610 	return (0);
611 }
612 /*
613  * Scan the exit callout list for the given items and remove them.
614  * Returns the number of items removed.
615  * Logically this can only be 0 or 1.
616  */
617 int
618 rm_at_exit(function)
619 	exitlist_fn function;
620 {
621 	ele_p *epp, ep;
622 	int count;
623 
624 	count = 0;
625 	epp = &exit_list;
626 	ep = *epp;
627 	while (ep) {
628 		if (ep->function == function) {
629 			*epp = ep->next;
630 			free(ep, M_TEMP);
631 			count++;
632 		} else {
633 			epp = &ep->next;
634 		}
635 		ep = *epp;
636 	}
637 	return (count);
638 }
639 
640 void check_sigacts (void)
641 {
642 	struct proc *p = curproc;
643 	struct sigacts *pss;
644 	int s;
645 
646 	if (p->p_procsig->ps_refcnt == 1 &&
647 	    p->p_sigacts != &p->p_addr->u_sigacts) {
648 		pss = p->p_sigacts;
649 		s = splhigh();
650 		p->p_addr->u_sigacts = *pss;
651 		p->p_sigacts = &p->p_addr->u_sigacts;
652 		splx(s);
653 		FREE(pss, M_SUBPROC);
654 	}
655 }
656 
657