xref: /netbsd-src/sys/kern/kern_descrip.c (revision bf1e9b32e27832f0c493206710fb8b58a980838a)
1 /*	$NetBSD: kern_descrip.c,v 1.134 2005/06/23 23:15:12 thorpej Exp $	*/
2 
3 /*
4  * Copyright (c) 1982, 1986, 1989, 1991, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  * (c) UNIX System Laboratories, Inc.
7  * All or some portions of this file are derived from material licensed
8  * to the University of California by American Telephone and Telegraph
9  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
10  * the permission of UNIX System Laboratories, Inc.
11  *
12  * Redistribution and use in source and binary forms, with or without
13  * modification, are permitted provided that the following conditions
14  * are met:
15  * 1. Redistributions of source code must retain the above copyright
16  *    notice, this list of conditions and the following disclaimer.
17  * 2. Redistributions in binary form must reproduce the above copyright
18  *    notice, this list of conditions and the following disclaimer in the
19  *    documentation and/or other materials provided with the distribution.
20  * 3. Neither the name of the University nor the names of its contributors
21  *    may be used to endorse or promote products derived from this software
22  *    without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  *
36  *	@(#)kern_descrip.c	8.8 (Berkeley) 2/14/95
37  */
38 
39 #include <sys/cdefs.h>
40 __KERNEL_RCSID(0, "$NetBSD: kern_descrip.c,v 1.134 2005/06/23 23:15:12 thorpej Exp $");
41 
42 #include <sys/param.h>
43 #include <sys/systm.h>
44 #include <sys/filedesc.h>
45 #include <sys/kernel.h>
46 #include <sys/vnode.h>
47 #include <sys/proc.h>
48 #include <sys/file.h>
49 #include <sys/namei.h>
50 #include <sys/socket.h>
51 #include <sys/socketvar.h>
52 #include <sys/stat.h>
53 #include <sys/ioctl.h>
54 #include <sys/fcntl.h>
55 #include <sys/malloc.h>
56 #include <sys/pool.h>
57 #include <sys/syslog.h>
58 #include <sys/unistd.h>
59 #include <sys/resourcevar.h>
60 #include <sys/conf.h>
61 #include <sys/event.h>
62 
63 #include <sys/mount.h>
64 #include <sys/sa.h>
65 #include <sys/syscallargs.h>
66 
67 /*
68  * Descriptor management.
69  */
70 struct filelist	filehead;	/* head of list of open files */
71 int		nfiles;		/* actual number of open files */
72 POOL_INIT(file_pool, sizeof(struct file), 0, 0, 0, "filepl",
73     &pool_allocator_nointr);
74 POOL_INIT(cwdi_pool, sizeof(struct cwdinfo), 0, 0, 0, "cwdipl",
75     &pool_allocator_nointr);
76 POOL_INIT(filedesc0_pool, sizeof(struct filedesc0), 0, 0, 0, "fdescpl",
77     &pool_allocator_nointr);
78 
79 /* Global file list lock */
80 static struct simplelock filelist_slock = SIMPLELOCK_INITIALIZER;
81 
82 MALLOC_DEFINE(M_FILE, "file", "Open file structure");
83 MALLOC_DEFINE(M_FILEDESC, "file desc", "Open file descriptor table");
84 MALLOC_DEFINE(M_IOCTLOPS, "ioctlops", "ioctl data buffer");
85 
86 static __inline int
87 find_next_zero(uint32_t *bitmap, int want, u_int bits)
88 {
89 	int i, off, maxoff;
90 	uint32_t sub;
91 
92 	if (want > bits)
93 		return -1;
94 
95 	off = want >> NDENTRYSHIFT;
96 	i = want & NDENTRYMASK;
97 	if (i) {
98 		sub = bitmap[off] | ((u_int)~0 >> (NDENTRIES - i));
99 		if (sub != ~0)
100 			goto found;
101 		off++;
102 	}
103 
104 	maxoff = NDLOSLOTS(bits);
105 	while (off < maxoff) {
106 		if ((sub = bitmap[off]) != ~0)
107 			goto found;
108 		off++;
109 	}
110 
111 	return (-1);
112 
113  found:
114 	return (off << NDENTRYSHIFT) + ffs(~sub) - 1;
115 }
116 
117 static int
118 find_last_set(struct filedesc *fd, int last)
119 {
120 	int off, i;
121 	struct file **ofiles = fd->fd_ofiles;
122 	uint32_t *bitmap = fd->fd_lomap;
123 
124 	off = (last - 1) >> NDENTRYSHIFT;
125 
126 	while (off >= 0 && !bitmap[off])
127 		off--;
128 
129 	if (off < 0)
130 		return (-1);
131 
132 	i = ((off + 1) << NDENTRYSHIFT) - 1;
133 	if (i >= last)
134 		i = last - 1;
135 
136 	while (i > 0 && ofiles[i] == NULL)
137 		i--;
138 
139 	return (i);
140 }
141 
142 static __inline void
143 fd_used(struct filedesc *fdp, int fd)
144 {
145 	u_int off = fd >> NDENTRYSHIFT;
146 
147 	LOCK_ASSERT(simple_lock_held(&fdp->fd_slock));
148 	KDASSERT((fdp->fd_lomap[off] & (1 << (fd & NDENTRYMASK))) == 0);
149 
150 	fdp->fd_lomap[off] |= 1 << (fd & NDENTRYMASK);
151 	if (fdp->fd_lomap[off] == ~0) {
152 		KDASSERT((fdp->fd_himap[off >> NDENTRYSHIFT] &
153 		    (1 << (off & NDENTRYMASK))) == 0);
154 		fdp->fd_himap[off >> NDENTRYSHIFT] |= 1 << (off & NDENTRYMASK);
155 	}
156 
157 	if (fd > fdp->fd_lastfile)
158 		fdp->fd_lastfile = fd;
159 }
160 
161 static __inline void
162 fd_unused(struct filedesc *fdp, int fd)
163 {
164 	u_int off = fd >> NDENTRYSHIFT;
165 
166 	LOCK_ASSERT(simple_lock_held(&fdp->fd_slock));
167 	if (fd < fdp->fd_freefile)
168 		fdp->fd_freefile = fd;
169 
170 	if (fdp->fd_lomap[off] == ~0) {
171 		KDASSERT((fdp->fd_himap[off >> NDENTRYSHIFT] &
172 		    (1 << (off & NDENTRYMASK))) != 0);
173 		fdp->fd_himap[off >> NDENTRYSHIFT] &=
174 		    ~(1 << (off & NDENTRYMASK));
175 	}
176 	KDASSERT((fdp->fd_lomap[off] & (1 << (fd & NDENTRYMASK))) != 0);
177 	fdp->fd_lomap[off] &= ~(1 << (fd & NDENTRYMASK));
178 
179 #ifdef DIAGNOSTIC
180 	if (fd > fdp->fd_lastfile)
181 		panic("fd_unused: fd_lastfile inconsistent");
182 #endif
183 	if (fd == fdp->fd_lastfile)
184 		fdp->fd_lastfile = find_last_set(fdp, fd);
185 }
186 
187 /*
188  * Lookup the file structure corresponding to a file descriptor
189  * and return it locked.
190  * Note: typical usage is: `fp = fd_getfile(..); FILE_USE(fp);'
191  * The locking strategy has been optimised for this case, i.e.
192  * fd_getfile() returns the file locked while FILE_USE() will increment
193  * the file's use count and unlock.
194  */
195 struct file *
196 fd_getfile(struct filedesc *fdp, int fd)
197 {
198 	struct file *fp;
199 
200 	if ((u_int) fd >= fdp->fd_nfiles || (fp = fdp->fd_ofiles[fd]) == NULL)
201 		return (NULL);
202 
203 	simple_lock(&fp->f_slock);
204 	if (FILE_IS_USABLE(fp) == 0) {
205 		simple_unlock(&fp->f_slock);
206 		return (NULL);
207 	}
208 
209 	return (fp);
210 }
211 
212 /*
213  * Common code for dup, dup2, and fcntl(F_DUPFD).
214  */
215 static int
216 finishdup(struct proc *p, int old, int new, register_t *retval)
217 {
218 	struct filedesc	*fdp;
219 	struct file	*fp, *delfp;
220 
221 	fdp = p->p_fd;
222 
223 	/*
224 	 * If there is a file in the new slot, remember it so we
225 	 * can close it after we've finished the dup.  We need
226 	 * to do it after the dup is finished, since closing
227 	 * the file may block.
228 	 *
229 	 * Note: `old' is already used for us.
230 	 * Note: Caller already marked `new' slot "used".
231 	 */
232 	simple_lock(&fdp->fd_slock);
233 	delfp = fdp->fd_ofiles[new];
234 
235 	fp = fdp->fd_ofiles[old];
236 	KDASSERT(fp != NULL);
237 	fdp->fd_ofiles[new] = fp;
238 	fdp->fd_ofileflags[new] = fdp->fd_ofileflags[old] &~ UF_EXCLOSE;
239 	simple_unlock(&fdp->fd_slock);
240 
241 	*retval = new;
242 	simple_lock(&fp->f_slock);
243 	fp->f_count++;
244 	FILE_UNUSE_HAVELOCK(fp, p);
245 
246 	if (delfp != NULL) {
247 		simple_lock(&delfp->f_slock);
248 		FILE_USE(delfp);
249 		if (new < fdp->fd_knlistsize)
250 			knote_fdclose(p, new);
251 		(void) closef(delfp, p);
252 	}
253 	return (0);
254 }
255 
256 /*
257  * System calls on descriptors.
258  */
259 
260 /*
261  * Duplicate a file descriptor.
262  */
263 /* ARGSUSED */
264 int
265 sys_dup(struct lwp *l, void *v, register_t *retval)
266 {
267 	struct sys_dup_args /* {
268 		syscallarg(int)	fd;
269 	} */ *uap = v;
270 	struct file	*fp;
271 	struct filedesc	*fdp;
272 	struct proc	*p;
273 	int		old, new, error;
274 
275 	p = l->l_proc;
276 	fdp = p->p_fd;
277 	old = SCARG(uap, fd);
278 
279  restart:
280 	if ((fp = fd_getfile(fdp, old)) == NULL)
281 		return (EBADF);
282 
283 	FILE_USE(fp);
284 
285 	if ((error = fdalloc(p, 0, &new)) != 0) {
286 		if (error == ENOSPC) {
287 			fdexpand(p);
288 			FILE_UNUSE(fp, p);
289 			goto restart;
290 		}
291 		FILE_UNUSE(fp, p);
292 		return (error);
293 	}
294 
295 	/* finishdup() will unuse the descriptors for us */
296 	return (finishdup(p, old, new, retval));
297 }
298 
299 /*
300  * Duplicate a file descriptor to a particular value.
301  */
302 /* ARGSUSED */
303 int
304 sys_dup2(struct lwp *l, void *v, register_t *retval)
305 {
306 	struct sys_dup2_args /* {
307 		syscallarg(int)	from;
308 		syscallarg(int)	to;
309 	} */ *uap = v;
310 	struct file	*fp;
311 	struct filedesc	*fdp;
312 	struct proc	*p;
313 	int		old, new, i, error;
314 
315 	p = l->l_proc;
316 	fdp = p->p_fd;
317 	old = SCARG(uap, from);
318 	new = SCARG(uap, to);
319 
320  restart:
321 	if ((fp = fd_getfile(fdp, old)) == NULL)
322 		return (EBADF);
323 
324 	if ((u_int)new >= p->p_rlimit[RLIMIT_NOFILE].rlim_cur ||
325 	    (u_int)new >= maxfiles) {
326 		simple_unlock(&fp->f_slock);
327 		return (EBADF);
328 	}
329 
330 	if (old == new) {
331 		simple_unlock(&fp->f_slock);
332 		*retval = new;
333 		return (0);
334 	}
335 
336 	FILE_USE(fp);
337 
338 	if (new >= fdp->fd_nfiles) {
339 		if ((error = fdalloc(p, new, &i)) != 0) {
340 			if (error == ENOSPC) {
341 				fdexpand(p);
342 				FILE_UNUSE(fp, p);
343 				goto restart;
344 			}
345 			FILE_UNUSE(fp, p);
346 			return (error);
347 		}
348 		if (new != i)
349 			panic("dup2: fdalloc");
350 	} else {
351 		simple_lock(&fdp->fd_slock);
352 		/*
353 		 * Mark `new' slot "used" only if it was empty.
354 		 */
355 		if (fdp->fd_ofiles[new] == NULL)
356 			fd_used(fdp, new);
357 		simple_unlock(&fdp->fd_slock);
358 	}
359 
360 	/*
361 	 * finishdup() will close the file that's in the `new'
362 	 * slot, if there's one there.
363 	 */
364 
365 	/* finishdup() will unuse the descriptors for us */
366 	return (finishdup(p, old, new, retval));
367 }
368 
369 /*
370  * fcntl call which is being passed to the file's fs.
371  */
372 static int
373 fcntl_forfs(int fd, struct proc *p, int cmd, void *arg)
374 {
375 	struct file	*fp;
376 	struct filedesc	*fdp;
377 	int		error;
378 	u_int		size;
379 	void		*data, *memp;
380 #define STK_PARAMS	128
381 	char		stkbuf[STK_PARAMS];
382 
383 	/* fd's value was validated in sys_fcntl before calling this routine */
384 	fdp = p->p_fd;
385 	fp = fdp->fd_ofiles[fd];
386 
387 	if ((fp->f_flag & (FREAD | FWRITE)) == 0)
388 		return (EBADF);
389 
390 	/*
391 	 * Interpret high order word to find amount of data to be
392 	 * copied to/from the user's address space.
393 	 */
394 	size = (size_t)F_PARAM_LEN(cmd);
395 	if (size > F_PARAM_MAX)
396 		return (EINVAL);
397 	memp = NULL;
398 	if (size > sizeof(stkbuf)) {
399 		memp = malloc((u_long)size, M_IOCTLOPS, M_WAITOK);
400 		data = memp;
401 	} else
402 		data = stkbuf;
403 	if (cmd & F_FSIN) {
404 		if (size) {
405 			error = copyin(arg, data, size);
406 			if (error) {
407 				if (memp)
408 					free(memp, M_IOCTLOPS);
409 				return (error);
410 			}
411 		} else
412 			*(void **)data = arg;
413 	} else if ((cmd & F_FSOUT) && size)
414 		/*
415 		 * Zero the buffer so the user always
416 		 * gets back something deterministic.
417 		 */
418 		memset(data, 0, size);
419 	else if (cmd & F_FSVOID)
420 		*(void **)data = arg;
421 
422 
423 	error = (*fp->f_ops->fo_fcntl)(fp, cmd, data, p);
424 
425 	/*
426 	 * Copy any data to user, size was
427 	 * already set and checked above.
428 	 */
429 	if (error == 0 && (cmd & F_FSOUT) && size)
430 		error = copyout(data, arg, size);
431 	if (memp)
432 		free(memp, M_IOCTLOPS);
433 	return (error);
434 }
435 
436 /*
437  * The file control system call.
438  */
439 /* ARGSUSED */
440 int
441 sys_fcntl(struct lwp *l, void *v, register_t *retval)
442 {
443 	struct sys_fcntl_args /* {
444 		syscallarg(int)		fd;
445 		syscallarg(int)		cmd;
446 		syscallarg(void *)	arg;
447 	} */ *uap = v;
448 	struct filedesc *fdp;
449 	struct file	*fp;
450 	struct proc	*p;
451 	struct vnode	*vp;
452 	int		fd, i, tmp, error, flg, cmd, newmin;
453 	struct flock	fl;
454 
455 	p = l->l_proc;
456 	fd = SCARG(uap, fd);
457 	cmd = SCARG(uap, cmd);
458 	fdp = p->p_fd;
459 	error = 0;
460 	flg = F_POSIX;
461 
462 	switch (cmd) {
463 	case F_CLOSEM:
464 		if (fd < 0)
465 			return EBADF;
466 		while (fdp->fd_lastfile >= fd)
467 			fdrelease(p, fdp->fd_lastfile);
468 		return 0;
469 
470 	case F_MAXFD:
471 		*retval = fdp->fd_lastfile;
472 		return 0;
473 
474 	default:
475 		/* Handled below */
476 		break;
477 	}
478 
479  restart:
480 	if ((fp = fd_getfile(fdp, fd)) == NULL)
481 		return (EBADF);
482 
483 	FILE_USE(fp);
484 
485 	if ((cmd & F_FSCTL)) {
486 		error = fcntl_forfs(fd, p, cmd, SCARG(uap, arg));
487 		goto out;
488 	}
489 
490 	switch (cmd) {
491 
492 	case F_DUPFD:
493 		newmin = (long)SCARG(uap, arg);
494 		if ((u_int)newmin >= p->p_rlimit[RLIMIT_NOFILE].rlim_cur ||
495 		    (u_int)newmin >= maxfiles) {
496 			error = EINVAL;
497 			goto out;
498 		}
499 		if ((error = fdalloc(p, newmin, &i)) != 0) {
500 			if (error == ENOSPC) {
501 				fdexpand(p);
502 				FILE_UNUSE(fp, p);
503 				goto restart;
504 			}
505 			goto out;
506 		}
507 
508 		/* finishdup() will unuse the descriptors for us */
509 		return (finishdup(p, fd, i, retval));
510 
511 	case F_GETFD:
512 		*retval = fdp->fd_ofileflags[fd] & UF_EXCLOSE ? 1 : 0;
513 		break;
514 
515 	case F_SETFD:
516 		if ((long)SCARG(uap, arg) & 1)
517 			fdp->fd_ofileflags[fd] |= UF_EXCLOSE;
518 		else
519 			fdp->fd_ofileflags[fd] &= ~UF_EXCLOSE;
520 		break;
521 
522 	case F_GETFL:
523 		*retval = OFLAGS(fp->f_flag);
524 		break;
525 
526 	case F_SETFL:
527 		tmp = FFLAGS((long)SCARG(uap, arg)) & FCNTLFLAGS;
528 		error = (*fp->f_ops->fo_fcntl)(fp, F_SETFL, &tmp, p);
529 		if (error)
530 			break;
531 		i = tmp ^ fp->f_flag;
532 		if (i & FNONBLOCK) {
533 			int flgs = tmp & FNONBLOCK;
534 			error = (*fp->f_ops->fo_ioctl)(fp, FIONBIO, &flgs, p);
535 			if (error)
536 				goto reset_fcntl;
537 		}
538 		if (i & FASYNC) {
539 			int flgs = tmp & FASYNC;
540 			error = (*fp->f_ops->fo_ioctl)(fp, FIOASYNC, &flgs, p);
541 			if (error) {
542 				if (i & FNONBLOCK) {
543 					tmp = fp->f_flag & FNONBLOCK;
544 					(void)(*fp->f_ops->fo_ioctl)(fp,
545 						FIONBIO, &tmp, p);
546 				}
547 				goto reset_fcntl;
548 			}
549 		}
550 		fp->f_flag = (fp->f_flag & ~FCNTLFLAGS) | tmp;
551 		break;
552 	    reset_fcntl:
553 		(void)(*fp->f_ops->fo_fcntl)(fp, F_SETFL, &fp->f_flag, p);
554 		break;
555 
556 	case F_GETOWN:
557 		error = (*fp->f_ops->fo_ioctl)(fp, FIOGETOWN, retval, p);
558 		break;
559 
560 	case F_SETOWN:
561 		tmp = (int)(intptr_t) SCARG(uap, arg);
562 		error = (*fp->f_ops->fo_ioctl)(fp, FIOSETOWN, &tmp, p);
563 		break;
564 
565 	case F_SETLKW:
566 		flg |= F_WAIT;
567 		/* Fall into F_SETLK */
568 
569 	case F_SETLK:
570 		if (fp->f_type != DTYPE_VNODE) {
571 			error = EINVAL;
572 			goto out;
573 		}
574 		vp = (struct vnode *)fp->f_data;
575 		/* Copy in the lock structure */
576 		error = copyin(SCARG(uap, arg), &fl, sizeof(fl));
577 		if (error)
578 			goto out;
579 		if (fl.l_whence == SEEK_CUR)
580 			fl.l_start += fp->f_offset;
581 		switch (fl.l_type) {
582 		case F_RDLCK:
583 			if ((fp->f_flag & FREAD) == 0) {
584 				error = EBADF;
585 				goto out;
586 			}
587 			p->p_flag |= P_ADVLOCK;
588 			error = VOP_ADVLOCK(vp, p, F_SETLK, &fl, flg);
589 			goto out;
590 
591 		case F_WRLCK:
592 			if ((fp->f_flag & FWRITE) == 0) {
593 				error = EBADF;
594 				goto out;
595 			}
596 			p->p_flag |= P_ADVLOCK;
597 			error = VOP_ADVLOCK(vp, p, F_SETLK, &fl, flg);
598 			goto out;
599 
600 		case F_UNLCK:
601 			error = VOP_ADVLOCK(vp, p, F_UNLCK, &fl, F_POSIX);
602 			goto out;
603 
604 		default:
605 			error = EINVAL;
606 			goto out;
607 		}
608 
609 	case F_GETLK:
610 		if (fp->f_type != DTYPE_VNODE) {
611 			error = EINVAL;
612 			goto out;
613 		}
614 		vp = (struct vnode *)fp->f_data;
615 		/* Copy in the lock structure */
616 		error = copyin(SCARG(uap, arg), &fl, sizeof(fl));
617 		if (error)
618 			goto out;
619 		if (fl.l_whence == SEEK_CUR)
620 			fl.l_start += fp->f_offset;
621 		if (fl.l_type != F_RDLCK &&
622 		    fl.l_type != F_WRLCK &&
623 		    fl.l_type != F_UNLCK) {
624 			error = EINVAL;
625 			goto out;
626 		}
627 		error = VOP_ADVLOCK(vp, p, F_GETLK, &fl, F_POSIX);
628 		if (error)
629 			goto out;
630 		error = copyout(&fl, SCARG(uap, arg), sizeof(fl));
631 		break;
632 
633 	default:
634 		error = EINVAL;
635 	}
636 
637  out:
638 	FILE_UNUSE(fp, p);
639 	return (error);
640 }
641 
642 void
643 fdremove(struct filedesc *fdp, int fd)
644 {
645 
646 	simple_lock(&fdp->fd_slock);
647 	fdp->fd_ofiles[fd] = NULL;
648 	fd_unused(fdp, fd);
649 	simple_unlock(&fdp->fd_slock);
650 }
651 
652 int
653 fdrelease(struct proc *p, int fd)
654 {
655 	struct filedesc	*fdp;
656 	struct file	**fpp, *fp;
657 
658 	fdp = p->p_fd;
659 	simple_lock(&fdp->fd_slock);
660 	if (fd < 0 || fd > fdp->fd_lastfile)
661 		goto badf;
662 	fpp = &fdp->fd_ofiles[fd];
663 	fp = *fpp;
664 	if (fp == NULL)
665 		goto badf;
666 
667 	simple_lock(&fp->f_slock);
668 	if (!FILE_IS_USABLE(fp)) {
669 		simple_unlock(&fp->f_slock);
670 		goto badf;
671 	}
672 
673 	FILE_USE(fp);
674 
675 	*fpp = NULL;
676 	fdp->fd_ofileflags[fd] = 0;
677 	fd_unused(fdp, fd);
678 	simple_unlock(&fdp->fd_slock);
679 	if (fd < fdp->fd_knlistsize)
680 		knote_fdclose(p, fd);
681 	return (closef(fp, p));
682 
683 badf:
684 	simple_unlock(&fdp->fd_slock);
685 	return (EBADF);
686 }
687 
688 /*
689  * Close a file descriptor.
690  */
691 /* ARGSUSED */
692 int
693 sys_close(struct lwp *l, void *v, register_t *retval)
694 {
695 	struct sys_close_args /* {
696 		syscallarg(int)	fd;
697 	} */ *uap = v;
698 	int		fd;
699 	struct filedesc	*fdp;
700 	struct proc *p;
701 
702 	p = l->l_proc;
703 	fd = SCARG(uap, fd);
704 	fdp = p->p_fd;
705 
706 #if 0
707 	if (fd_getfile(fdp, fd) == NULL)
708 		return (EBADF);
709 #endif
710 
711 	return (fdrelease(p, fd));
712 }
713 
714 /*
715  * Return status information about a file descriptor.
716  */
717 /* ARGSUSED */
718 int
719 sys___fstat13(struct lwp *l, void *v, register_t *retval)
720 {
721 	struct sys___fstat13_args /* {
722 		syscallarg(int)			fd;
723 		syscallarg(struct stat *)	sb;
724 	} */ *uap = v;
725 	int		fd;
726 	struct filedesc	*fdp;
727 	struct file	*fp;
728 	struct proc	*p;
729 	struct stat	ub;
730 	int		error;
731 
732 	p = l->l_proc;
733 	fd = SCARG(uap, fd);
734 	fdp = p->p_fd;
735 
736 	if ((fp = fd_getfile(fdp, fd)) == NULL)
737 		return (EBADF);
738 
739 	FILE_USE(fp);
740 	error = (*fp->f_ops->fo_stat)(fp, &ub, p);
741 	FILE_UNUSE(fp, p);
742 
743 	if (error == 0)
744 		error = copyout(&ub, SCARG(uap, sb), sizeof(ub));
745 
746 	return (error);
747 }
748 
749 /*
750  * Return pathconf information about a file descriptor.
751  */
752 /* ARGSUSED */
753 int
754 sys_fpathconf(struct lwp *l, void *v, register_t *retval)
755 {
756 	struct sys_fpathconf_args /* {
757 		syscallarg(int)	fd;
758 		syscallarg(int)	name;
759 	} */ *uap = v;
760 	int		fd;
761 	struct filedesc	*fdp;
762 	struct file	*fp;
763 	struct proc 	*p;
764 	struct vnode	*vp;
765 	int		error;
766 
767 	p = l->l_proc;
768 	fd = SCARG(uap, fd);
769 	fdp = p->p_fd;
770 	error = 0;
771 
772 	if ((fp = fd_getfile(fdp, fd)) == NULL)
773 		return (EBADF);
774 
775 	FILE_USE(fp);
776 
777 	switch (fp->f_type) {
778 
779 	case DTYPE_SOCKET:
780 	case DTYPE_PIPE:
781 		if (SCARG(uap, name) != _PC_PIPE_BUF)
782 			error = EINVAL;
783 		else
784 			*retval = PIPE_BUF;
785 		break;
786 
787 	case DTYPE_VNODE:
788 		vp = (struct vnode *)fp->f_data;
789 		error = VOP_PATHCONF(vp, SCARG(uap, name), retval);
790 		break;
791 
792 	case DTYPE_KQUEUE:
793 		error = EINVAL;
794 		break;
795 
796 	default:
797 		error = EOPNOTSUPP;
798 		break;
799 	}
800 
801 	FILE_UNUSE(fp, p);
802 	return (error);
803 }
804 
805 /*
806  * Allocate a file descriptor for the process.
807  */
808 int	fdexpanded;		/* XXX: what else uses this? */
809 
810 int
811 fdalloc(struct proc *p, int want, int *result)
812 {
813 	struct filedesc	*fdp;
814 	int i, lim, last, error;
815 	u_int off, new;
816 
817 	fdp = p->p_fd;
818 	simple_lock(&fdp->fd_slock);
819 
820 	/*
821 	 * Search for a free descriptor starting at the higher
822 	 * of want or fd_freefile.  If that fails, consider
823 	 * expanding the ofile array.
824 	 */
825 	lim = min((int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur, maxfiles);
826 	last = min(fdp->fd_nfiles, lim);
827  again:
828 	if ((i = want) < fdp->fd_freefile)
829 		i = fdp->fd_freefile;
830 	off = i >> NDENTRYSHIFT;
831 	new = find_next_zero(fdp->fd_himap, off,
832 	    (last + NDENTRIES - 1) >> NDENTRYSHIFT);
833 	if (new != -1) {
834 		i = find_next_zero(&fdp->fd_lomap[new],
835 		    new > off ? 0 : i & NDENTRYMASK, NDENTRIES);
836 		if (i == -1) {
837 			/*
838 			 * free file descriptor in this block was
839 			 * below want, try again with higher want.
840 			 */
841 			want = (new + 1) << NDENTRYSHIFT;
842 			goto again;
843 		}
844 		i += (new << NDENTRYSHIFT);
845 		if (i < last) {
846 			if (fdp->fd_ofiles[i] == NULL) {
847 				fd_used(fdp, i);
848 				if (want <= fdp->fd_freefile)
849 					fdp->fd_freefile = i;
850 				*result = i;
851 				error = 0;
852 				goto out;
853 			}
854 		}
855 	}
856 
857 	/* No space in current array.  Expand or let the caller do it. */
858 	error = (fdp->fd_nfiles >= lim) ? EMFILE : ENOSPC;
859 
860 out:
861 	simple_unlock(&fdp->fd_slock);
862 	return (error);
863 }
864 
865 void
866 fdexpand(struct proc *p)
867 {
868 	struct filedesc	*fdp;
869 	int		i, numfiles, oldnfiles;
870 	struct file	**newofile;
871 	char		*newofileflags;
872 	uint32_t	*newhimap = NULL, *newlomap = NULL;
873 
874 	fdp = p->p_fd;
875 
876 restart:
877 	oldnfiles = fdp->fd_nfiles;
878 
879 	if (oldnfiles < NDEXTENT)
880 		numfiles = NDEXTENT;
881 	else
882 		numfiles = 2 * oldnfiles;
883 
884 	newofile = malloc(numfiles * OFILESIZE, M_FILEDESC, M_WAITOK);
885 	if (NDHISLOTS(numfiles) > NDHISLOTS(oldnfiles)) {
886 		newhimap = malloc(NDHISLOTS(numfiles) * sizeof(uint32_t),
887 		    M_FILEDESC, M_WAITOK);
888 		newlomap = malloc(NDLOSLOTS(numfiles) * sizeof(uint32_t),
889 		    M_FILEDESC, M_WAITOK);
890 	}
891 
892 	simple_lock(&fdp->fd_slock);
893 	/* lock fdp */
894 	if (fdp->fd_nfiles != oldnfiles) {
895 		/* fdp changed; retry */
896 		simple_unlock(&fdp->fd_slock);
897 		free(newofile, M_FILEDESC);
898 		if (newhimap != NULL) free(newhimap, M_FILEDESC);
899 		if (newlomap != NULL) free(newlomap, M_FILEDESC);
900 		goto restart;
901 	}
902 
903 	newofileflags = (char *) &newofile[numfiles];
904 	/*
905 	 * Copy the existing ofile and ofileflags arrays
906 	 * and zero the new portion of each array.
907 	 */
908 	memcpy(newofile, fdp->fd_ofiles,
909 	    (i = sizeof(struct file *) * fdp->fd_nfiles));
910 	memset((char *)newofile + i, 0,
911 	    numfiles * sizeof(struct file *) - i);
912 	memcpy(newofileflags, fdp->fd_ofileflags,
913 	    (i = sizeof(char) * fdp->fd_nfiles));
914 	memset(newofileflags + i, 0, numfiles * sizeof(char) - i);
915 	if (oldnfiles > NDFILE)
916 		free(fdp->fd_ofiles, M_FILEDESC);
917 
918 	if (NDHISLOTS(numfiles) > NDHISLOTS(oldnfiles)) {
919 		memcpy(newhimap, fdp->fd_himap,
920 		    (i = NDHISLOTS(oldnfiles) * sizeof(uint32_t)));
921 		memset((char *)newhimap + i, 0,
922 		    NDHISLOTS(numfiles) * sizeof(uint32_t) - i);
923 
924 		memcpy(newlomap, fdp->fd_lomap,
925 		    (i = NDLOSLOTS(oldnfiles) * sizeof(uint32_t)));
926 		memset((char *)newlomap + i, 0,
927 		    NDLOSLOTS(numfiles) * sizeof(uint32_t) - i);
928 
929 		if (NDHISLOTS(oldnfiles) > NDHISLOTS(NDFILE)) {
930 			free(fdp->fd_himap, M_FILEDESC);
931 			free(fdp->fd_lomap, M_FILEDESC);
932 		}
933 		fdp->fd_himap = newhimap;
934 		fdp->fd_lomap = newlomap;
935 	}
936 
937 	fdp->fd_ofiles = newofile;
938 	fdp->fd_ofileflags = newofileflags;
939 	fdp->fd_nfiles = numfiles;
940 
941 	simple_unlock(&fdp->fd_slock);
942 
943 	fdexpanded++;
944 }
945 
946 /*
947  * Check to see whether n user file descriptors
948  * are available to the process p.
949  */
950 int
951 fdavail(struct proc *p, int n)
952 {
953 	struct filedesc	*fdp;
954 	struct file	**fpp;
955 	int		i, lim;
956 
957 	fdp = p->p_fd;
958 	lim = min((int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur, maxfiles);
959 	if ((i = lim - fdp->fd_nfiles) > 0 && (n -= i) <= 0)
960 		return (1);
961 	fpp = &fdp->fd_ofiles[fdp->fd_freefile];
962 	for (i = min(lim,fdp->fd_nfiles) - fdp->fd_freefile; --i >= 0; fpp++)
963 		if (*fpp == NULL && --n <= 0)
964 			return (1);
965 	return (0);
966 }
967 
968 /*
969  * Create a new open file structure and allocate
970  * a file descriptor for the process that refers to it.
971  */
972 int
973 falloc(struct proc *p, struct file **resultfp, int *resultfd)
974 {
975 	struct file	*fp, *fq;
976 	int		error, i;
977 
978  restart:
979 	if ((error = fdalloc(p, 0, &i)) != 0) {
980 		if (error == ENOSPC) {
981 			fdexpand(p);
982 			goto restart;
983 		}
984 		return (error);
985 	}
986 
987 	fp = pool_get(&file_pool, PR_WAITOK);
988 	simple_lock(&filelist_slock);
989 	if (nfiles >= maxfiles) {
990 		tablefull("file", "increase kern.maxfiles or MAXFILES");
991 		simple_unlock(&filelist_slock);
992 		fd_unused(p->p_fd, i);
993 		pool_put(&file_pool, fp);
994 		return (ENFILE);
995 	}
996 	/*
997 	 * Allocate a new file descriptor.
998 	 * If the process has file descriptor zero open, add to the list
999 	 * of open files at that point, otherwise put it at the front of
1000 	 * the list of open files.
1001 	 */
1002 	nfiles++;
1003 	memset(fp, 0, sizeof(struct file));
1004 	fp->f_iflags = FIF_LARVAL;
1005 	if ((fq = p->p_fd->fd_ofiles[0]) != NULL) {
1006 		LIST_INSERT_AFTER(fq, fp, f_list);
1007 	} else {
1008 		LIST_INSERT_HEAD(&filehead, fp, f_list);
1009 	}
1010 	simple_unlock(&filelist_slock);
1011 	KDASSERT(p->p_fd->fd_ofiles[i] == NULL);
1012 	p->p_fd->fd_ofiles[i] = fp;
1013 	simple_lock_init(&fp->f_slock);
1014 	fp->f_count = 1;
1015 	fp->f_cred = p->p_ucred;
1016 	crhold(fp->f_cred);
1017 	if (resultfp) {
1018 		fp->f_usecount = 1;
1019 		*resultfp = fp;
1020 	}
1021 	if (resultfd)
1022 		*resultfd = i;
1023 	return (0);
1024 }
1025 
1026 /*
1027  * Free a file descriptor.
1028  */
1029 void
1030 ffree(struct file *fp)
1031 {
1032 
1033 #ifdef DIAGNOSTIC
1034 	if (fp->f_usecount)
1035 		panic("ffree");
1036 #endif
1037 
1038 	simple_lock(&filelist_slock);
1039 	LIST_REMOVE(fp, f_list);
1040 	crfree(fp->f_cred);
1041 #ifdef DIAGNOSTIC
1042 	fp->f_count = 0; /* What's the point? */
1043 #endif
1044 	nfiles--;
1045 	simple_unlock(&filelist_slock);
1046 	pool_put(&file_pool, fp);
1047 }
1048 
1049 /*
1050  * Create an initial cwdinfo structure, using the same current and root
1051  * directories as p.
1052  */
1053 struct cwdinfo *
1054 cwdinit(struct proc *p)
1055 {
1056 	struct cwdinfo *cwdi;
1057 
1058 	cwdi = pool_get(&cwdi_pool, PR_WAITOK);
1059 
1060 	simple_lock_init(&cwdi->cwdi_slock);
1061 	cwdi->cwdi_cdir = p->p_cwdi->cwdi_cdir;
1062 	if (cwdi->cwdi_cdir)
1063 		VREF(cwdi->cwdi_cdir);
1064 	cwdi->cwdi_rdir = p->p_cwdi->cwdi_rdir;
1065 	if (cwdi->cwdi_rdir)
1066 		VREF(cwdi->cwdi_rdir);
1067 	cwdi->cwdi_cmask =  p->p_cwdi->cwdi_cmask;
1068 	cwdi->cwdi_refcnt = 1;
1069 
1070 	return (cwdi);
1071 }
1072 
1073 /*
1074  * Make p2 share p1's cwdinfo.
1075  */
1076 void
1077 cwdshare(struct proc *p1, struct proc *p2)
1078 {
1079 	struct cwdinfo *cwdi = p1->p_cwdi;
1080 
1081 	simple_lock(&cwdi->cwdi_slock);
1082 	cwdi->cwdi_refcnt++;
1083 	simple_unlock(&cwdi->cwdi_slock);
1084 	p2->p_cwdi = cwdi;
1085 }
1086 
1087 /*
1088  * Make this process not share its cwdinfo structure, maintaining
1089  * all cwdinfo state.
1090  */
1091 void
1092 cwdunshare(struct proc *p)
1093 {
1094 	struct cwdinfo *oldcwdi, *newcwdi;
1095 
1096 	if (p->p_cwdi->cwdi_refcnt == 1)
1097 		return;
1098 
1099 	newcwdi = cwdinit(p);
1100 	oldcwdi = p->p_cwdi;
1101 	p->p_cwdi = newcwdi;
1102 	cwdfree(oldcwdi);
1103 }
1104 
1105 /*
1106  * Release a cwdinfo structure.
1107  */
1108 void
1109 cwdfree(struct cwdinfo *cwdi)
1110 {
1111 	int n;
1112 
1113 	simple_lock(&cwdi->cwdi_slock);
1114 	n = --cwdi->cwdi_refcnt;
1115 	simple_unlock(&cwdi->cwdi_slock);
1116 	if (n > 0)
1117 		return;
1118 
1119 	vrele(cwdi->cwdi_cdir);
1120 	if (cwdi->cwdi_rdir)
1121 		vrele(cwdi->cwdi_rdir);
1122 	pool_put(&cwdi_pool, cwdi);
1123 }
1124 
1125 /*
1126  * Create an initial filedesc structure, using the same current and root
1127  * directories as p.
1128  */
1129 struct filedesc *
1130 fdinit(struct proc *p)
1131 {
1132 	struct filedesc0 *newfdp;
1133 
1134 	newfdp = pool_get(&filedesc0_pool, PR_WAITOK);
1135 	memset(newfdp, 0, sizeof(struct filedesc0));
1136 
1137 	fdinit1(newfdp);
1138 
1139 	return (&newfdp->fd_fd);
1140 }
1141 
1142 /*
1143  * Initialize a file descriptor table.
1144  */
1145 void
1146 fdinit1(struct filedesc0 *newfdp)
1147 {
1148 
1149 	newfdp->fd_fd.fd_refcnt = 1;
1150 	newfdp->fd_fd.fd_ofiles = newfdp->fd_dfiles;
1151 	newfdp->fd_fd.fd_ofileflags = newfdp->fd_dfileflags;
1152 	newfdp->fd_fd.fd_nfiles = NDFILE;
1153 	newfdp->fd_fd.fd_knlistsize = -1;
1154 	newfdp->fd_fd.fd_himap = newfdp->fd_dhimap;
1155 	newfdp->fd_fd.fd_lomap = newfdp->fd_dlomap;
1156 	newfdp->fd_fd.fd_lastfile = -1;
1157 	simple_lock_init(&newfdp->fd_fd.fd_slock);
1158 }
1159 
1160 /*
1161  * Make p2 share p1's filedesc structure.
1162  */
1163 void
1164 fdshare(struct proc *p1, struct proc *p2)
1165 {
1166 	struct filedesc *fdp = p1->p_fd;
1167 
1168 	simple_lock(&fdp->fd_slock);
1169 	p2->p_fd = fdp;
1170 	fdp->fd_refcnt++;
1171 	simple_unlock(&fdp->fd_slock);
1172 }
1173 
1174 /*
1175  * Make this process not share its filedesc structure, maintaining
1176  * all file descriptor state.
1177  */
1178 void
1179 fdunshare(struct proc *p)
1180 {
1181 	struct filedesc *newfd;
1182 
1183 	if (p->p_fd->fd_refcnt == 1)
1184 		return;
1185 
1186 	newfd = fdcopy(p);
1187 	fdfree(p);
1188 	p->p_fd = newfd;
1189 }
1190 
1191 /*
1192  * Clear a process's fd table.
1193  */
1194 void
1195 fdclear(struct proc *p)
1196 {
1197 	struct filedesc *newfd;
1198 
1199 	newfd = fdinit(p);
1200 	fdfree(p);
1201 	p->p_fd = newfd;
1202 }
1203 
1204 /*
1205  * Copy a filedesc structure.
1206  */
1207 struct filedesc *
1208 fdcopy(struct proc *p)
1209 {
1210 	struct filedesc	*newfdp, *fdp;
1211 	struct file	**fpp, **nfpp;
1212 	int		i, numfiles, lastfile;
1213 
1214 	fdp = p->p_fd;
1215 	newfdp = pool_get(&filedesc0_pool, PR_WAITOK);
1216 	newfdp->fd_refcnt = 1;
1217 	simple_lock_init(&newfdp->fd_slock);
1218 
1219 restart:
1220 	numfiles = fdp->fd_nfiles;
1221 	lastfile = fdp->fd_lastfile;
1222 
1223 	/*
1224 	 * If the number of open files fits in the internal arrays
1225 	 * of the open file structure, use them, otherwise allocate
1226 	 * additional memory for the number of descriptors currently
1227 	 * in use.
1228 	 */
1229 	if (lastfile < NDFILE) {
1230 		i = NDFILE;
1231 	} else {
1232 		/*
1233 		 * Compute the smallest multiple of NDEXTENT needed
1234 		 * for the file descriptors currently in use,
1235 		 * allowing the table to shrink.
1236 		 */
1237 		i = numfiles;
1238 		while (i >= 2 * NDEXTENT && i > lastfile * 2)
1239 			i /= 2;
1240 		newfdp->fd_ofiles = malloc(i * OFILESIZE, M_FILEDESC, M_WAITOK);
1241 	}
1242 	if (NDHISLOTS(i) > NDHISLOTS(NDFILE)) {
1243 		newfdp->fd_himap = malloc(NDHISLOTS(i) * sizeof(uint32_t),
1244 		    M_FILEDESC, M_WAITOK);
1245 		newfdp->fd_lomap = malloc(NDLOSLOTS(i) * sizeof(uint32_t),
1246 		    M_FILEDESC, M_WAITOK);
1247 	}
1248 
1249 	simple_lock(&fdp->fd_slock);
1250 	if (numfiles != fdp->fd_nfiles || lastfile != fdp->fd_lastfile) {
1251 		simple_unlock(&fdp->fd_slock);
1252 		if (i > NDFILE)
1253 			free(newfdp->fd_ofiles, M_FILEDESC);
1254 		if (NDHISLOTS(i) > NDHISLOTS(NDFILE)) {
1255 			free(newfdp->fd_himap, M_FILEDESC);
1256 			free(newfdp->fd_lomap, M_FILEDESC);
1257 		}
1258 		goto restart;
1259 	}
1260 
1261 	if (lastfile < NDFILE) {
1262 		newfdp->fd_ofiles = ((struct filedesc0 *) newfdp)->fd_dfiles;
1263 		newfdp->fd_ofileflags =
1264 		    ((struct filedesc0 *) newfdp)->fd_dfileflags;
1265 	} else {
1266 		newfdp->fd_ofileflags = (char *) &newfdp->fd_ofiles[i];
1267 	}
1268 	if (NDHISLOTS(i) <= NDHISLOTS(NDFILE)) {
1269 		newfdp->fd_himap =
1270 		    ((struct filedesc0 *) newfdp)->fd_dhimap;
1271 		newfdp->fd_lomap =
1272 		    ((struct filedesc0 *) newfdp)->fd_dlomap;
1273 	}
1274 
1275 	newfdp->fd_nfiles = i;
1276 	newfdp->fd_lastfile = lastfile;
1277 	newfdp->fd_freefile = fdp->fd_freefile;
1278 
1279 	/* Clear the entries that will not be copied over.
1280 	 * Avoid calling memset with 0 size (i.e. when
1281 	 * lastfile == i-1 */
1282 	if (lastfile < (i-1))
1283 		memset(newfdp->fd_ofiles + lastfile + 1, 0,
1284 		    (i - lastfile - 1) * sizeof(struct file **));
1285 	memcpy(newfdp->fd_ofileflags, fdp->fd_ofileflags, i * sizeof(char));
1286 	if (i < NDENTRIES * NDENTRIES)
1287 		i = NDENTRIES * NDENTRIES; /* size of inlined bitmaps */
1288 	memcpy(newfdp->fd_himap, fdp->fd_himap, NDHISLOTS(i)*sizeof(uint32_t));
1289 	memcpy(newfdp->fd_lomap, fdp->fd_lomap, NDLOSLOTS(i)*sizeof(uint32_t));
1290 
1291 	fpp = fdp->fd_ofiles;
1292 	nfpp = newfdp->fd_ofiles;
1293 	for (i = 0; i <= lastfile; i++, fpp++, nfpp++) {
1294 		if ((*nfpp = *fpp) == NULL)
1295 			continue;
1296 
1297 		if ((*fpp)->f_type == DTYPE_KQUEUE)
1298 			/* kq descriptors cannot be copied. */
1299 			fdremove(newfdp, i);
1300 		else {
1301 			simple_lock(&(*fpp)->f_slock);
1302 			(*fpp)->f_count++;
1303 			simple_unlock(&(*fpp)->f_slock);
1304 		}
1305 	}
1306 
1307 	simple_unlock(&fdp->fd_slock);
1308 
1309 	newfdp->fd_knlist = NULL;
1310 	newfdp->fd_knlistsize = -1;
1311 	newfdp->fd_knhash = NULL;
1312 	newfdp->fd_knhashmask = 0;
1313 
1314 	return (newfdp);
1315 }
1316 
1317 /*
1318  * Release a filedesc structure.
1319  */
1320 void
1321 fdfree(struct proc *p)
1322 {
1323 	struct filedesc	*fdp;
1324 	struct file	**fpp, *fp;
1325 	int		i;
1326 
1327 	fdp = p->p_fd;
1328 	simple_lock(&fdp->fd_slock);
1329 	i = --fdp->fd_refcnt;
1330 	simple_unlock(&fdp->fd_slock);
1331 	if (i > 0)
1332 		return;
1333 
1334 	fpp = fdp->fd_ofiles;
1335 	for (i = fdp->fd_lastfile; i >= 0; i--, fpp++) {
1336 		fp = *fpp;
1337 		if (fp != NULL) {
1338 			*fpp = NULL;
1339 			simple_lock(&fp->f_slock);
1340 			FILE_USE(fp);
1341 			if ((fdp->fd_lastfile - i) < fdp->fd_knlistsize)
1342 				knote_fdclose(p, fdp->fd_lastfile - i);
1343 			(void) closef(fp, p);
1344 		}
1345 	}
1346 	p->p_fd = NULL;
1347 	if (fdp->fd_nfiles > NDFILE)
1348 		free(fdp->fd_ofiles, M_FILEDESC);
1349 	if (NDHISLOTS(fdp->fd_nfiles) > NDHISLOTS(NDFILE)) {
1350 		free(fdp->fd_himap, M_FILEDESC);
1351 		free(fdp->fd_lomap, M_FILEDESC);
1352 	}
1353 	if (fdp->fd_knlist)
1354 		free(fdp->fd_knlist, M_KEVENT);
1355 	if (fdp->fd_knhash)
1356 		hashdone(fdp->fd_knhash, M_KEVENT);
1357 	pool_put(&filedesc0_pool, fdp);
1358 }
1359 
1360 /*
1361  * Internal form of close.
1362  * Decrement reference count on file structure.
1363  * Note: p may be NULL when closing a file
1364  * that was being passed in a message.
1365  *
1366  * Note: we expect the caller is holding a usecount, and expects us
1367  * to drop it (the caller thinks the file is going away forever).
1368  */
1369 int
1370 closef(struct file *fp, struct proc *p)
1371 {
1372 	struct vnode	*vp;
1373 	struct flock	lf;
1374 	int		error;
1375 
1376 	if (fp == NULL)
1377 		return (0);
1378 
1379 	/*
1380 	 * POSIX record locking dictates that any close releases ALL
1381 	 * locks owned by this process.  This is handled by setting
1382 	 * a flag in the unlock to free ONLY locks obeying POSIX
1383 	 * semantics, and not to free BSD-style file locks.
1384 	 * If the descriptor was in a message, POSIX-style locks
1385 	 * aren't passed with the descriptor.
1386 	 */
1387 	if (p && (p->p_flag & P_ADVLOCK) && fp->f_type == DTYPE_VNODE) {
1388 		lf.l_whence = SEEK_SET;
1389 		lf.l_start = 0;
1390 		lf.l_len = 0;
1391 		lf.l_type = F_UNLCK;
1392 		vp = (struct vnode *)fp->f_data;
1393 		(void) VOP_ADVLOCK(vp, p, F_UNLCK, &lf, F_POSIX);
1394 	}
1395 
1396 	/*
1397 	 * If WANTCLOSE is set, then the reference count on the file
1398 	 * is 0, but there were multiple users of the file.  This can
1399 	 * happen if a filedesc structure is shared by multiple
1400 	 * processes.
1401 	 */
1402 	simple_lock(&fp->f_slock);
1403 	if (fp->f_iflags & FIF_WANTCLOSE) {
1404 		/*
1405 		 * Another user of the file is already closing, and is
1406 		 * simply waiting for other users of the file to drain.
1407 		 * Release our usecount, and wake up the closer if it
1408 		 * is the only remaining use.
1409 		 */
1410 #ifdef DIAGNOSTIC
1411 		if (fp->f_count != 0)
1412 			panic("closef: wantclose and count != 0");
1413 		if (fp->f_usecount < 2)
1414 			panic("closef: wantclose and usecount < 2");
1415 #endif
1416 		if (--fp->f_usecount == 1)
1417 			wakeup(&fp->f_usecount);
1418 		simple_unlock(&fp->f_slock);
1419 		return (0);
1420 	} else {
1421 		/*
1422 		 * Decrement the reference count.  If we were not the
1423 		 * last reference, then release our use and just
1424 		 * return.
1425 		 */
1426 		if (--fp->f_count > 0) {
1427 #ifdef DIAGNOSTIC
1428 			if (fp->f_usecount < 1)
1429 				panic("closef: no wantclose and usecount < 1");
1430 #endif
1431 			fp->f_usecount--;
1432 			simple_unlock(&fp->f_slock);
1433 			return (0);
1434 		}
1435 	}
1436 
1437 	/*
1438 	 * The reference count is now 0.  However, there may be
1439 	 * multiple potential users of this file.  This can happen
1440 	 * if multiple processes shared a single filedesc structure.
1441 	 *
1442 	 * Notify these potential users that the file is closing.
1443 	 * This will prevent them from adding additional uses to
1444 	 * the file.
1445 	 */
1446 	fp->f_iflags |= FIF_WANTCLOSE;
1447 
1448 	/*
1449 	 * We expect the caller to add a use to the file.  So, if we
1450 	 * are the last user, usecount will be 1.  If it is not, we
1451 	 * must wait for the usecount to drain.  When it drains back
1452 	 * to 1, we will be awakened so that we may proceed with the
1453 	 * close.
1454 	 */
1455 #ifdef DIAGNOSTIC
1456 	if (fp->f_usecount < 1)
1457 		panic("closef: usecount < 1");
1458 #endif
1459 	while (fp->f_usecount > 1)
1460 		(void) ltsleep(&fp->f_usecount, PRIBIO, "closef", 0,
1461 				&fp->f_slock);
1462 #ifdef DIAGNOSTIC
1463 	if (fp->f_usecount != 1)
1464 		panic("closef: usecount != 1");
1465 #endif
1466 
1467 	simple_unlock(&fp->f_slock);
1468 	if ((fp->f_flag & FHASLOCK) && fp->f_type == DTYPE_VNODE) {
1469 		lf.l_whence = SEEK_SET;
1470 		lf.l_start = 0;
1471 		lf.l_len = 0;
1472 		lf.l_type = F_UNLCK;
1473 		vp = (struct vnode *)fp->f_data;
1474 		(void) VOP_ADVLOCK(vp, fp, F_UNLCK, &lf, F_FLOCK);
1475 	}
1476 	if (fp->f_ops)
1477 		error = (*fp->f_ops->fo_close)(fp, p);
1478 	else
1479 		error = 0;
1480 
1481 	/* Nothing references the file now, drop the final use (us). */
1482 	fp->f_usecount--;
1483 
1484 	ffree(fp);
1485 	return (error);
1486 }
1487 
1488 /*
1489  * Apply an advisory lock on a file descriptor.
1490  *
1491  * Just attempt to get a record lock of the requested type on
1492  * the entire file (l_whence = SEEK_SET, l_start = 0, l_len = 0).
1493  */
1494 /* ARGSUSED */
1495 int
1496 sys_flock(struct lwp *l, void *v, register_t *retval)
1497 {
1498 	struct sys_flock_args /* {
1499 		syscallarg(int)	fd;
1500 		syscallarg(int)	how;
1501 	} */ *uap = v;
1502 	int		fd, how, error;
1503 	struct proc	*p;
1504 	struct filedesc	*fdp;
1505 	struct file	*fp;
1506 	struct vnode	*vp;
1507 	struct flock	lf;
1508 
1509 	p = l->l_proc;
1510 	fd = SCARG(uap, fd);
1511 	how = SCARG(uap, how);
1512 	fdp = p->p_fd;
1513 	error = 0;
1514 
1515 	if ((fp = fd_getfile(fdp, fd)) == NULL)
1516 		return (EBADF);
1517 
1518 	FILE_USE(fp);
1519 
1520 	if (fp->f_type != DTYPE_VNODE) {
1521 		error = EOPNOTSUPP;
1522 		goto out;
1523 	}
1524 
1525 	vp = (struct vnode *)fp->f_data;
1526 	lf.l_whence = SEEK_SET;
1527 	lf.l_start = 0;
1528 	lf.l_len = 0;
1529 	if (how & LOCK_UN) {
1530 		lf.l_type = F_UNLCK;
1531 		fp->f_flag &= ~FHASLOCK;
1532 		error = VOP_ADVLOCK(vp, fp, F_UNLCK, &lf, F_FLOCK);
1533 		goto out;
1534 	}
1535 	if (how & LOCK_EX)
1536 		lf.l_type = F_WRLCK;
1537 	else if (how & LOCK_SH)
1538 		lf.l_type = F_RDLCK;
1539 	else {
1540 		error = EINVAL;
1541 		goto out;
1542 	}
1543 	fp->f_flag |= FHASLOCK;
1544 	if (how & LOCK_NB)
1545 		error = VOP_ADVLOCK(vp, fp, F_SETLK, &lf, F_FLOCK);
1546 	else
1547 		error = VOP_ADVLOCK(vp, fp, F_SETLK, &lf,
1548 		    F_FLOCK|F_WAIT);
1549  out:
1550 	FILE_UNUSE(fp, p);
1551 	return (error);
1552 }
1553 
1554 /*
1555  * File Descriptor pseudo-device driver (/dev/fd/).
1556  *
1557  * Opening minor device N dup()s the file (if any) connected to file
1558  * descriptor N belonging to the calling process.  Note that this driver
1559  * consists of only the ``open()'' routine, because all subsequent
1560  * references to this file will be direct to the other driver.
1561  */
1562 /* ARGSUSED */
1563 static int
1564 filedescopen(dev_t dev, int mode, int type, struct proc *p)
1565 {
1566 
1567 	/*
1568 	 * XXX Kludge: set dupfd to contain the value of the
1569 	 * the file descriptor being sought for duplication. The error
1570 	 * return ensures that the vnode for this device will be released
1571 	 * by vn_open. Open will detect this special error and take the
1572 	 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
1573 	 * will simply report the error.
1574 	 */
1575 	curlwp->l_dupfd = minor(dev);	/* XXX */
1576 	return EDUPFD;
1577 }
1578 
1579 const struct cdevsw filedesc_cdevsw = {
1580 	filedescopen, noclose, noread, nowrite, noioctl,
1581 	    nostop, notty, nopoll, nommap, nokqfilter,
1582 };
1583 
1584 /*
1585  * Duplicate the specified descriptor to a free descriptor.
1586  *
1587  * 'indx' has been fdalloc'ed (and will be fdremove'ed on error) by the caller.
1588  */
1589 int
1590 dupfdopen(struct proc *p, int indx, int dfd, int mode, int error)
1591 {
1592 	struct filedesc	*fdp;
1593 	struct file	*wfp;
1594 
1595 	fdp = p->p_fd;
1596 
1597 	/* should be cleared by the caller */
1598 	KASSERT(fdp->fd_ofiles[indx] == NULL);
1599 
1600 	/*
1601 	 * If the to-be-dup'd fd number is greater than the allowed number
1602 	 * of file descriptors, or the fd to be dup'd has already been
1603 	 * closed, reject.
1604 	 */
1605 
1606 	/*
1607 	 * Note, in the case of indx == dfd, fd_getfile below returns NULL.
1608 	 */
1609 	if ((wfp = fd_getfile(fdp, dfd)) == NULL)
1610 		return (EBADF);
1611 
1612 	FILE_USE(wfp);
1613 
1614 	/*
1615 	 * There are two cases of interest here.
1616 	 *
1617 	 * For EDUPFD simply dup (dfd) to file descriptor
1618 	 * (indx) and return.
1619 	 *
1620 	 * For EMOVEFD steal away the file structure from (dfd) and
1621 	 * store it in (indx).  (dfd) is effectively closed by
1622 	 * this operation.
1623 	 *
1624 	 * Any other error code is just returned.
1625 	 */
1626 	switch (error) {
1627 	case EDUPFD:
1628 		/*
1629 		 * Check that the mode the file is being opened for is a
1630 		 * subset of the mode of the existing descriptor.
1631 		 */
1632 		if (((mode & (FREAD|FWRITE)) | wfp->f_flag) != wfp->f_flag) {
1633 			FILE_UNUSE(wfp, p);
1634 			return (EACCES);
1635 		}
1636 		simple_lock(&fdp->fd_slock);
1637 		fdp->fd_ofiles[indx] = wfp;
1638 		fdp->fd_ofileflags[indx] = fdp->fd_ofileflags[dfd];
1639 		simple_unlock(&fdp->fd_slock);
1640 		simple_lock(&wfp->f_slock);
1641 		wfp->f_count++;
1642 		/* 'indx' has been fd_used'ed by caller */
1643 		FILE_UNUSE_HAVELOCK(wfp, p);
1644 		return (0);
1645 
1646 	case EMOVEFD:
1647 		/*
1648 		 * Steal away the file pointer from dfd, and stuff it into indx.
1649 		 */
1650 		simple_lock(&fdp->fd_slock);
1651 		fdp->fd_ofiles[indx] = fdp->fd_ofiles[dfd];
1652 		fdp->fd_ofileflags[indx] = fdp->fd_ofileflags[dfd];
1653 		fdp->fd_ofiles[dfd] = NULL;
1654 		fdp->fd_ofileflags[dfd] = 0;
1655 		/*
1656 		 * Complete the clean up of the filedesc structure by
1657 		 * recomputing the various hints.
1658 		 */
1659 		/* 'indx' has been fd_used'ed by caller */
1660 		fd_unused(fdp, dfd);
1661 		simple_unlock(&fdp->fd_slock);
1662 		FILE_UNUSE(wfp, p);
1663 		return (0);
1664 
1665 	default:
1666 		FILE_UNUSE(wfp, p);
1667 		return (error);
1668 	}
1669 	/* NOTREACHED */
1670 }
1671 
1672 /*
1673  * Close any files on exec?
1674  */
1675 void
1676 fdcloseexec(struct proc *p)
1677 {
1678 	struct filedesc	*fdp;
1679 	int		fd;
1680 
1681 	fdunshare(p);
1682 	cwdunshare(p);
1683 
1684 	fdp = p->p_fd;
1685 	for (fd = 0; fd <= fdp->fd_lastfile; fd++)
1686 		if (fdp->fd_ofileflags[fd] & UF_EXCLOSE)
1687 			(void) fdrelease(p, fd);
1688 }
1689 
1690 /*
1691  * It is unsafe for set[ug]id processes to be started with file
1692  * descriptors 0..2 closed, as these descriptors are given implicit
1693  * significance in the Standard C library.  fdcheckstd() will create a
1694  * descriptor referencing /dev/null for each of stdin, stdout, and
1695  * stderr that is not already open.
1696  */
1697 #define CHECK_UPTO 3
1698 int
1699 fdcheckstd(p)
1700 	struct proc *p;
1701 {
1702 	struct nameidata nd;
1703 	struct filedesc *fdp;
1704 	struct file *fp;
1705 	struct file *devnullfp = NULL;	/* Quell compiler warning */
1706 	struct proc *pp;
1707 	register_t retval;
1708 	int fd, i, error, flags = FREAD|FWRITE, devnull = -1;
1709 	char closed[CHECK_UPTO * 3 + 1], which[3 + 1];
1710 
1711 	closed[0] = '\0';
1712 	if ((fdp = p->p_fd) == NULL)
1713 		return (0);
1714 	for (i = 0; i < CHECK_UPTO; i++) {
1715 		if (fdp->fd_ofiles[i] != NULL)
1716 			continue;
1717 		snprintf(which, sizeof(which), ",%d", i);
1718 		strlcat(closed, which, sizeof(closed));
1719 		if (devnull < 0) {
1720 			if ((error = falloc(p, &fp, &fd)) != 0)
1721 				return (error);
1722 			NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, "/dev/null",
1723 			    p);
1724 			if ((error = vn_open(&nd, flags, 0)) != 0) {
1725 				FILE_UNUSE(fp, p);
1726 				ffree(fp);
1727 				fdremove(p->p_fd, fd);
1728 				return (error);
1729 			}
1730 			fp->f_data = nd.ni_vp;
1731 			fp->f_flag = flags;
1732 			fp->f_ops = &vnops;
1733 			fp->f_type = DTYPE_VNODE;
1734 			VOP_UNLOCK(nd.ni_vp, 0);
1735 			devnull = fd;
1736 			devnullfp = fp;
1737 			FILE_SET_MATURE(fp);
1738 		} else {
1739 restart:
1740 			if ((error = fdalloc(p, 0, &fd)) != 0) {
1741 				if (error == ENOSPC) {
1742 					fdexpand(p);
1743 					goto restart;
1744 				}
1745 				return (error);
1746 			}
1747 
1748 			simple_lock(&devnullfp->f_slock);
1749 			FILE_USE(devnullfp);
1750 			/* finishdup() will unuse the descriptors for us */
1751 			if ((error = finishdup(p, devnull, fd, &retval)) != 0)
1752 				return (error);
1753 		}
1754 	}
1755 	if (devnullfp)
1756 		FILE_UNUSE(devnullfp, p);
1757 	if (closed[0] != '\0') {
1758 		pp = p->p_pptr;
1759 		log(LOG_WARNING, "set{u,g}id pid %d (%s) "
1760 		    "was invoked by uid %d ppid %d (%s) "
1761 		    "with fd %s closed\n",
1762 		    p->p_pid, p->p_comm, pp->p_ucred->cr_uid,
1763 		    pp->p_pid, pp->p_comm, &closed[1]);
1764 	}
1765 	return (0);
1766 }
1767 #undef CHECK_UPTO
1768 
1769 /*
1770  * Sets descriptor owner. If the owner is a process, 'pgid'
1771  * is set to positive value, process ID. If the owner is process group,
1772  * 'pgid' is set to -pg_id.
1773  */
1774 int
1775 fsetown(struct proc *p, pid_t *pgid, int cmd, const void *data)
1776 {
1777 	int id = *(const int *)data;
1778 	int error;
1779 
1780 	switch (cmd) {
1781 	case TIOCSPGRP:
1782 		if (id < 0)
1783 			return (EINVAL);
1784 		id = -id;
1785 		break;
1786 	default:
1787 		break;
1788 	}
1789 
1790 	if (id > 0 && !pfind(id))
1791 		return (ESRCH);
1792 	else if (id < 0 && (error = pgid_in_session(p, -id)))
1793 		return (error);
1794 
1795 	*pgid = id;
1796 	return (0);
1797 }
1798 
1799 /*
1800  * Return descriptor owner information. If the value is positive,
1801  * it's process ID. If it's negative, it's process group ID and
1802  * needs the sign removed before use.
1803  */
1804 int
1805 fgetown(struct proc *p, pid_t pgid, int cmd, void *data)
1806 {
1807 	switch (cmd) {
1808 	case TIOCGPGRP:
1809 		*(int *)data = -pgid;
1810 		break;
1811 	default:
1812 		*(int *)data = pgid;
1813 		break;
1814 	}
1815 	return (0);
1816 }
1817 
1818 /*
1819  * Send signal to descriptor owner, either process or process group.
1820  */
1821 void
1822 fownsignal(pid_t pgid, int signo, int code, int band, void *fdescdata)
1823 {
1824 	struct proc *p1;
1825 	ksiginfo_t ksi;
1826 
1827 	memset(&ksi, 0, sizeof(ksi));
1828 	ksi.ksi_signo = signo;
1829 	ksi.ksi_code = code;
1830 	ksi.ksi_band = band;
1831 
1832 	if (pgid > 0 && (p1 = pfind(pgid)))
1833 		kpsignal(p1, &ksi, fdescdata);
1834 	else if (pgid < 0)
1835 		kgsignal(-pgid, &ksi, fdescdata);
1836 }
1837 
1838 int
1839 fdclone(struct proc *p, struct file *fp, int fd, int flag,
1840     const struct fileops *fops, void *data)
1841 {
1842 	fp->f_flag = flag;
1843 	fp->f_type = DTYPE_MISC;
1844 	fp->f_ops = fops;
1845 	fp->f_data = data;
1846 
1847 	curlwp->l_dupfd = fd;
1848 
1849 	FILE_SET_MATURE(fp);
1850 	FILE_UNUSE(fp, p);
1851 	return EMOVEFD;
1852 }
1853 
1854 /* ARGSUSED */
1855 int
1856 fnullop_fcntl(struct file *fp, u_int cmd, void *data, struct proc *p)
1857 {
1858 	if (cmd == F_SETFL)
1859 		return 0;
1860 
1861 	return EOPNOTSUPP;
1862 }
1863 
1864 /* ARGSUSED */
1865 int
1866 fnullop_poll(struct file *fp, int which, struct proc *p)
1867 {
1868 	return 0;
1869 }
1870 
1871 
1872 /* ARGSUSED */
1873 int
1874 fnullop_kqfilter(struct file *fp, struct knote *kn)
1875 {
1876 
1877 	return 0;
1878 }
1879 
1880 /* ARGSUSED */
1881 int
1882 fbadop_stat(struct file *fp, struct stat *sb, struct proc *p)
1883 {
1884 	return EOPNOTSUPP;
1885 }
1886