xref: /dflybsd-src/sys/kern/kern_exit.c (revision bd4539cc23771f3c0b3fae4ecf80e725b613b305)
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: src/sys/kern/kern_exit.c,v 1.92.2.11 2003/01/13 22:51:16 dillon Exp $
40  * $DragonFly: src/sys/kern/kern_exit.c,v 1.32 2004/03/20 23:35:18 dillon Exp $
41  */
42 
43 #include "opt_compat.h"
44 #include "opt_ktrace.h"
45 
46 #include <sys/param.h>
47 #include <sys/systm.h>
48 #include <sys/sysproto.h>
49 #include <sys/kernel.h>
50 #include <sys/malloc.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 #include <sys/kern_syscall.h>
66 #include <sys/upcall.h>
67 #include <sys/caps.h>
68 
69 #include <vm/vm.h>
70 #include <vm/vm_param.h>
71 #include <sys/lock.h>
72 #include <vm/pmap.h>
73 #include <vm/vm_map.h>
74 #include <vm/vm_zone.h>
75 #include <vm/vm_extern.h>
76 #include <sys/user.h>
77 
78 /* Required to be non-static for SysVR4 emulator */
79 MALLOC_DEFINE(M_ZOMBIE, "zombie", "zombie proc status");
80 
81 static MALLOC_DEFINE(M_ATEXIT, "atexit", "atexit callback");
82 
83 /*
84  * callout list for things to do at exit time
85  */
86 struct exitlist {
87 	exitlist_fn function;
88 	TAILQ_ENTRY(exitlist) next;
89 };
90 
91 TAILQ_HEAD(exit_list_head, exitlist);
92 static struct exit_list_head exit_list = TAILQ_HEAD_INITIALIZER(exit_list);
93 
94 /*
95  * exit --
96  *	Death of process.
97  *
98  * SYS_EXIT_ARGS(int rval)
99  */
100 void
101 sys_exit(struct sys_exit_args *uap)
102 {
103 	exit1(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(int rv)
114 {
115 	struct proc *p = curproc;
116 	struct proc *q, *nq;
117 	struct vmspace *vm;
118 	struct vnode *vtmp;
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 	caps_exit(p->p_thread);
128 
129 	aio_proc_rundown(p);
130 
131 	/* are we a task leader? */
132 	if(p == p->p_leader) {
133         	struct kill_args killArgs;
134 		killArgs.signum = SIGKILL;
135 		q = p->p_peers;
136 		while(q) {
137 			killArgs.pid = q->p_pid;
138 			/*
139 		         * The interface for kill is better
140 			 * than the internal signal
141 			 */
142 			kill(&killArgs);
143 			nq = q;
144 			q = q->p_peers;
145 		}
146 		while (p->p_peers)
147 		  tsleep((caddr_t)p, 0, "exit1", 0);
148 	}
149 
150 #ifdef PGINPROF
151 	vmsizmon();
152 #endif
153 	STOPEVENT(p, S_EXIT, rv);
154 	wakeup(&p->p_stype);	/* Wakeup anyone in procfs' PIOCWAIT */
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 	TAILQ_FOREACH(ep, &exit_list, next)
162 		(*ep->function)(p->p_thread);
163 
164 	if (p->p_flag & P_PROFIL)
165 		stopprofclock(p);
166 	MALLOC(p->p_ru, struct rusage *, sizeof(struct rusage),
167 		M_ZOMBIE, M_WAITOK);
168 	/*
169 	 * If parent is waiting for us to exit or exec,
170 	 * P_PPWAIT is set; we will wakeup the parent below.
171 	 */
172 	p->p_flag &= ~(P_TRACED | P_PPWAIT);
173 	p->p_flag |= P_WEXIT;
174 	SIGEMPTYSET(p->p_siglist);
175 	if (timevalisset(&p->p_realtimer.it_value))
176 		untimeout(realitexpire, (caddr_t)p, p->p_ithandle);
177 
178 	/*
179 	 * Reset any sigio structures pointing to us as a result of
180 	 * F_SETOWN with our pid.
181 	 */
182 	funsetownlst(&p->p_sigiolst);
183 
184 	/*
185 	 * Close open files and release open-file table.
186 	 * This may block!
187 	 */
188 	fdfree(p);
189 
190 	if(p->p_leader->p_peers) {
191 		q = p->p_leader;
192 		while(q->p_peers != p)
193 			q = q->p_peers;
194 		q->p_peers = p->p_peers;
195 		wakeup((caddr_t)p->p_leader);
196 	}
197 
198 	/*
199 	 * XXX Shutdown SYSV semaphores
200 	 */
201 	semexit(p);
202 
203 	/* The next two chunks should probably be moved to vmspace_exit. */
204 	vm = p->p_vmspace;
205 
206 	/*
207 	 * Release upcalls associated with this process
208 	 */
209 	if (vm->vm_upcalls)
210 		upc_release(vm, p);
211 
212 	/*
213 	 * Release user portion of address space.
214 	 * This releases references to vnodes,
215 	 * which could cause I/O if the file has been unlinked.
216 	 * Need to do this early enough that we can still sleep.
217 	 * Can't free the entire vmspace as the kernel stack
218 	 * may be mapped within that space also.
219 	 *
220 	 * Processes sharing the same vmspace may exit in one order, and
221 	 * get cleaned up by vmspace_exit() in a different order.  The
222 	 * last exiting process to reach this point releases as much of
223 	 * the environment as it can, and the last process cleaned up
224 	 * by vmspace_exit() (which decrements exitingcnt) cleans up the
225 	 * remainder.
226 	 */
227 	++vm->vm_exitingcnt;
228 	if (--vm->vm_refcnt == 0) {
229 		shmexit(vm);
230 		pmap_remove_pages(vmspace_pmap(vm), VM_MIN_ADDRESS,
231 		    VM_MAXUSER_ADDRESS);
232 		(void) vm_map_remove(&vm->vm_map, VM_MIN_ADDRESS,
233 		    VM_MAXUSER_ADDRESS);
234 	}
235 
236 	if (SESS_LEADER(p)) {
237 		struct session *sp = p->p_session;
238 
239 		if (sp->s_ttyvp) {
240 			/*
241 			 * Controlling process.
242 			 * Signal foreground pgrp,
243 			 * drain controlling terminal
244 			 * and revoke access to controlling terminal.
245 			 */
246 			if (sp->s_ttyp && (sp->s_ttyp->t_session == sp)) {
247 				if (sp->s_ttyp->t_pgrp)
248 					pgsignal(sp->s_ttyp->t_pgrp, SIGHUP, 1);
249 				(void) ttywait(sp->s_ttyp);
250 				/*
251 				 * The tty could have been revoked
252 				 * if we blocked.
253 				 */
254 				if (sp->s_ttyvp)
255 					VOP_REVOKE(sp->s_ttyvp, REVOKEALL);
256 			}
257 			if (sp->s_ttyvp)
258 				vrele(sp->s_ttyvp);
259 			sp->s_ttyvp = NULL;
260 			/*
261 			 * s_ttyp is not zero'd; we use this to indicate
262 			 * that the session once had a controlling terminal.
263 			 * (for logging and informational purposes)
264 			 */
265 		}
266 		sp->s_leader = NULL;
267 	}
268 	fixjobc(p, p->p_pgrp, 0);
269 	(void)acct_process(p);
270 #ifdef KTRACE
271 	/*
272 	 * release trace file
273 	 */
274 	p->p_traceflag = 0;	/* don't trace the vrele() */
275 	if ((vtmp = p->p_tracep) != NULL) {
276 		p->p_tracep = NULL;
277 		vrele(vtmp);
278 	}
279 #endif
280 	/*
281 	 * Release reference to text vnode
282 	 */
283 	if ((vtmp = p->p_textvp) != NULL) {
284 		p->p_textvp = NULL;
285 		vrele(vtmp);
286 	}
287 
288 	/*
289 	 * Once we set SZOMB the process can get reaped.  The wait1 code
290 	 * will also wait for TDF_RUNNING to be cleared in the thread's flags,
291 	 * indicating that it has been completely switched out.
292 	 */
293 
294 	/*
295 	 * Remove proc from allproc queue and pidhash chain.
296 	 * Place onto zombproc.  Unlink from parent's child list.
297 	 */
298 	LIST_REMOVE(p, p_list);
299 	LIST_INSERT_HEAD(&zombproc, p, p_list);
300 	p->p_stat = SZOMB;
301 
302 	LIST_REMOVE(p, p_hash);
303 
304 	q = LIST_FIRST(&p->p_children);
305 	if (q)		/* only need this if any child is S_ZOMB */
306 		wakeup((caddr_t) initproc);
307 	for (; q != 0; q = nq) {
308 		nq = LIST_NEXT(q, p_sibling);
309 		LIST_REMOVE(q, p_sibling);
310 		LIST_INSERT_HEAD(&initproc->p_children, q, p_sibling);
311 		q->p_pptr = initproc;
312 		q->p_sigparent = SIGCHLD;
313 		/*
314 		 * Traced processes are killed
315 		 * since their existence means someone is screwing up.
316 		 */
317 		if (q->p_flag & P_TRACED) {
318 			q->p_flag &= ~P_TRACED;
319 			psignal(q, SIGKILL);
320 		}
321 	}
322 
323 	/*
324 	 * Save exit status and final rusage info, adding in child rusage
325 	 * info and self times.
326 	 */
327 	p->p_xstat = rv;
328 	*p->p_ru = p->p_stats->p_ru;
329 	calcru(p, &p->p_ru->ru_utime, &p->p_ru->ru_stime, NULL);
330 	ruadd(p->p_ru, &p->p_stats->p_cru);
331 
332 	/*
333 	 * notify interested parties of our demise.
334 	 */
335 	KNOTE(&p->p_klist, NOTE_EXIT);
336 
337 	/*
338 	 * Notify parent that we're gone.  If parent has the PS_NOCLDWAIT
339 	 * flag set, notify process 1 instead (and hope it will handle
340 	 * this situation).
341 	 */
342 	if (p->p_pptr->p_procsig->ps_flag & PS_NOCLDWAIT) {
343 		struct proc *pp = p->p_pptr;
344 		proc_reparent(p, initproc);
345 		/*
346 		 * If this was the last child of our parent, notify
347 		 * parent, so in case he was wait(2)ing, he will
348 		 * continue.
349 		 */
350 		if (LIST_EMPTY(&pp->p_children))
351 			wakeup((caddr_t)pp);
352 	}
353 
354 	if (p->p_sigparent && p->p_pptr != initproc) {
355 	        psignal(p->p_pptr, p->p_sigparent);
356 	} else {
357 	        psignal(p->p_pptr, SIGCHLD);
358 	}
359 
360 	wakeup((caddr_t)p->p_pptr);
361 #if defined(tahoe)
362 	/* move this to cpu_exit */
363 	p->p_thread->td_pcb->pcb_saveacc.faddr = (float *)NULL;
364 #endif
365 	/*
366 	 * cpu_exit is responsible for clearing curproc, since
367 	 * it is heavily integrated with the thread/switching sequence.
368 	 *
369 	 * Other substructures are freed from wait().
370 	 */
371 	if (--p->p_limit->p_refcnt == 0) {
372 		FREE(p->p_limit, M_SUBPROC);
373 		p->p_limit = NULL;
374 	}
375 
376 	/*
377 	 * Release the P_CURPROC designation on the process so the userland
378 	 * scheduler can work in someone else.
379 	 */
380 	release_curproc(p);
381 
382 	/*
383 	 * Finally, call machine-dependent code to release the remaining
384 	 * resources including address space, the kernel stack and pcb.
385 	 * The address space is released by "vmspace_free(p->p_vmspace)";
386 	 * This is machine-dependent, as we may have to change stacks
387 	 * or ensure that the current one isn't reallocated before we
388 	 * finish.  cpu_exit will end with a call to cpu_switch(), finishing
389 	 * our execution (pun intended).
390 	 */
391 	cpu_proc_exit();
392 }
393 
394 int
395 wait4(struct wait_args *uap)
396 {
397 	struct rusage rusage;
398 	int error, status;
399 
400 	error = kern_wait(uap->pid, uap->status ? &status : NULL,
401 	    uap->options, uap->rusage ? &rusage : NULL, &uap->sysmsg_fds[0]);
402 
403 	if (error == 0 && uap->status)
404 		error = copyout(&status, uap->status, sizeof(*uap->status));
405 	if (error == 0 && uap->rusage)
406 		error = copyout(&rusage, uap->rusage, sizeof(*uap->rusage));
407 	return (error);
408 }
409 
410 /*
411  * wait1()
412  *
413  * wait_args(int pid, int *status, int options, struct rusage *rusage)
414  */
415 int
416 kern_wait(pid_t pid, int *status, int options, struct rusage *rusage, int *res)
417 {
418 	struct thread *td = curthread;
419 	struct proc *q = td->td_proc;
420 	struct proc *p, *t;
421 	int nfound, error;
422 
423 	if (pid == 0)
424 		pid = -q->p_pgid;
425 	if (options &~ (WUNTRACED|WNOHANG|WLINUXCLONE))
426 		return (EINVAL);
427 loop:
428 	nfound = 0;
429 	LIST_FOREACH(p, &q->p_children, p_sibling) {
430 		if (pid != WAIT_ANY &&
431 		    p->p_pid != pid && p->p_pgid != -pid)
432 			continue;
433 
434 		/* This special case handles a kthread spawned by linux_clone
435 		 * (see linux_misc.c).  The linux_wait4 and linux_waitpid functions
436 		 * need to be able to distinguish between waiting on a process and
437 		 * waiting on a thread.  It is a thread if p_sigparent is not SIGCHLD,
438 		 * and the WLINUXCLONE option signifies we want to wait for threads
439 		 * and not processes.
440 		 */
441 		if ((p->p_sigparent != SIGCHLD) ^ ((options & WLINUXCLONE) != 0))
442 			continue;
443 
444 		nfound++;
445 		if (p->p_stat == SZOMB) {
446 			/*
447 			 * The process's thread may still be in the middle
448 			 * of switching away, we can't rip its stack out from
449 			 * under it until TDF_RUNNING clears!
450 			 *
451 			 * YYY no wakeup occurs so we depend on the timeout.
452 			 */
453 			if ((p->p_thread->td_flags & TDF_RUNNING) != 0) {
454 				tsleep(p->p_thread, 0, "reap", 1);
455 				goto loop;
456 			}
457 
458 			/*
459 			 * Other kernel threads may be in the middle of
460 			 * accessing the proc.  For example, kern/kern_proc.c
461 			 * could be blocked writing proc data to a sysctl.
462 			 * At the moment, if this occurs, we are not woken
463 			 * up and rely on a one-second retry.
464 			 */
465 			if (p->p_lock) {
466 				while (p->p_lock)
467 					tsleep(p, 0, "reap2", hz);
468 			}
469 			lwkt_wait_free(p->p_thread);
470 
471 			/*
472 			 * Charge the parent for the child's change in
473 			 * estimated cpu as of when the child exits to
474 			 * account for batch scripts, large make's, etc.
475 			 */
476 			if (q->p_pid != 1) {
477 			    if (p->p_estcpu > p->p_estcpu_fork) {
478 				q->p_estcpu = ESTCPULIM(q->p_estcpu +
479 						p->p_estcpu - p->p_estcpu_fork);
480 			    }
481 			}
482 
483 			/* Take care of our return values. */
484 			*res = p->p_pid;
485 			if (status)
486 				*status = p->p_xstat;
487 			if (rusage)
488 				*rusage = *p->p_ru;
489 			/*
490 			 * If we got the child via a ptrace 'attach',
491 			 * we need to give it back to the old parent.
492 			 */
493 			if (p->p_oppid && (t = pfind(p->p_oppid))) {
494 				p->p_oppid = 0;
495 				proc_reparent(p, t);
496 				psignal(t, SIGCHLD);
497 				wakeup((caddr_t)t);
498 				return (0);
499 			}
500 			p->p_xstat = 0;
501 			ruadd(&q->p_stats->p_cru, p->p_ru);
502 			FREE(p->p_ru, M_ZOMBIE);
503 			p->p_ru = NULL;
504 
505 			/*
506 			 * Decrement the count of procs running with this uid.
507 			 */
508 			chgproccnt(p->p_ucred->cr_ruidinfo, -1, 0);
509 
510 			/*
511 			 * Free up credentials.
512 			 */
513 			crfree(p->p_ucred);
514 			p->p_ucred = NULL;
515 
516 			/*
517 			 * Remove unused arguments
518 			 */
519 			if (p->p_args && --p->p_args->ar_ref == 0)
520 				FREE(p->p_args, M_PARGS);
521 
522 			/*
523 			 * Finally finished with old proc entry.
524 			 * Unlink it from its process group and free it.
525 			 */
526 			leavepgrp(p);
527 			LIST_REMOVE(p, p_list);	/* off zombproc */
528 			LIST_REMOVE(p, p_sibling);
529 
530 			if (--p->p_procsig->ps_refcnt == 0) {
531 				if (p->p_sigacts != &p->p_addr->u_sigacts)
532 					FREE(p->p_sigacts, M_SUBPROC);
533 			        FREE(p->p_procsig, M_SUBPROC);
534 				p->p_procsig = NULL;
535 			}
536 
537 			vm_waitproc(p);
538 			zfree(proc_zone, p);
539 			nprocs--;
540 			return (0);
541 		}
542 		if (p->p_stat == SSTOP && (p->p_flag & P_WAITED) == 0 &&
543 		    (p->p_flag & P_TRACED || options & WUNTRACED)) {
544 			p->p_flag |= P_WAITED;
545 
546 			*res = p->p_pid;
547 			if (status)
548 				*status = W_STOPCODE(p->p_xstat);
549 			/* Zero rusage so we get something consistent. */
550 			if (rusage)
551 				bzero(rusage, sizeof(rusage));
552 			return (0);
553 		}
554 	}
555 	if (nfound == 0)
556 		return (ECHILD);
557 	if (options & WNOHANG) {
558 		*res = 0;
559 		return (0);
560 	}
561 	error = tsleep((caddr_t)q, PCATCH, "wait", 0);
562 	if (error)
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 	struct proc *child;
573 	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