xref: /freebsd/sys/kern/kern_descrip.c (revision 262e143bd46171a6415a5b28af260a5efa2a3db8)
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  * 4. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  *	@(#)kern_descrip.c	8.6 (Berkeley) 4/19/94
35  */
36 
37 #include <sys/cdefs.h>
38 __FBSDID("$FreeBSD$");
39 
40 #include "opt_compat.h"
41 #include "opt_ddb.h"
42 
43 #include <sys/param.h>
44 #include <sys/systm.h>
45 
46 #include <sys/conf.h>
47 #include <sys/fcntl.h>
48 #include <sys/file.h>
49 #include <sys/filedesc.h>
50 #include <sys/filio.h>
51 #include <sys/jail.h>
52 #include <sys/kernel.h>
53 #include <sys/limits.h>
54 #include <sys/lock.h>
55 #include <sys/malloc.h>
56 #include <sys/mount.h>
57 #include <sys/mqueue.h>
58 #include <sys/mutex.h>
59 #include <sys/namei.h>
60 #include <sys/proc.h>
61 #include <sys/resourcevar.h>
62 #include <sys/signalvar.h>
63 #include <sys/socketvar.h>
64 #include <sys/stat.h>
65 #include <sys/sx.h>
66 #include <sys/syscallsubr.h>
67 #include <sys/sysctl.h>
68 #include <sys/sysproto.h>
69 #include <sys/unistd.h>
70 #include <sys/vnode.h>
71 
72 #include <vm/uma.h>
73 
74 #include <ddb/ddb.h>
75 
76 static MALLOC_DEFINE(M_FILEDESC, "filedesc", "Open file descriptor table");
77 static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "filedesc_to_leader",
78 		     "file desc to leader structures");
79 static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures");
80 
81 static uma_zone_t file_zone;
82 
83 
84 /* How to treat 'new' parameter when allocating a fd for do_dup(). */
85 enum dup_type { DUP_VARIABLE, DUP_FIXED };
86 
87 static int do_dup(struct thread *td, enum dup_type type, int old, int new,
88     register_t *retval);
89 static int	fd_first_free(struct filedesc *, int, int);
90 static int	fd_last_used(struct filedesc *, int, int);
91 static void	fdgrowtable(struct filedesc *, int);
92 static int	fdrop_locked(struct file *fp, struct thread *td);
93 static void	fdunused(struct filedesc *fdp, int fd);
94 static void	fdused(struct filedesc *fdp, int fd);
95 
96 /*
97  * A process is initially started out with NDFILE descriptors stored within
98  * this structure, selected to be enough for typical applications based on
99  * the historical limit of 20 open files (and the usage of descriptors by
100  * shells).  If these descriptors are exhausted, a larger descriptor table
101  * may be allocated, up to a process' resource limit; the internal arrays
102  * are then unused.
103  */
104 #define NDFILE		20
105 #define NDSLOTSIZE	sizeof(NDSLOTTYPE)
106 #define	NDENTRIES	(NDSLOTSIZE * __CHAR_BIT)
107 #define NDSLOT(x)	((x) / NDENTRIES)
108 #define NDBIT(x)	((NDSLOTTYPE)1 << ((x) % NDENTRIES))
109 #define	NDSLOTS(x)	(((x) + NDENTRIES - 1) / NDENTRIES)
110 
111 /*
112  * Storage required per open file descriptor.
113  */
114 #define OFILESIZE (sizeof(struct file *) + sizeof(char))
115 
116 /*
117  * Basic allocation of descriptors:
118  * one of the above, plus arrays for NDFILE descriptors.
119  */
120 struct filedesc0 {
121 	struct	filedesc fd_fd;
122 	/*
123 	 * These arrays are used when the number of open files is
124 	 * <= NDFILE, and are then pointed to by the pointers above.
125 	 */
126 	struct	file *fd_dfiles[NDFILE];
127 	char	fd_dfileflags[NDFILE];
128 	NDSLOTTYPE fd_dmap[NDSLOTS(NDFILE)];
129 };
130 
131 /*
132  * Descriptor management.
133  */
134 struct filelist filehead;	/* head of list of open files */
135 int openfiles;			/* actual number of open files */
136 struct sx filelist_lock;	/* sx to protect filelist */
137 struct mtx sigio_lock;		/* mtx to protect pointers to sigio */
138 void	(*mq_fdclose)(struct thread *td, int fd, struct file *fp);
139 
140 /* A mutex to protect the association between a proc and filedesc. */
141 static struct mtx	fdesc_mtx;
142 
143 /*
144  * Find the first zero bit in the given bitmap, starting at low and not
145  * exceeding size - 1.
146  */
147 static int
148 fd_first_free(struct filedesc *fdp, int low, int size)
149 {
150 	NDSLOTTYPE *map = fdp->fd_map;
151 	NDSLOTTYPE mask;
152 	int off, maxoff;
153 
154 	if (low >= size)
155 		return (low);
156 
157 	off = NDSLOT(low);
158 	if (low % NDENTRIES) {
159 		mask = ~(~(NDSLOTTYPE)0 >> (NDENTRIES - (low % NDENTRIES)));
160 		if ((mask &= ~map[off]) != 0UL)
161 			return (off * NDENTRIES + ffsl(mask) - 1);
162 		++off;
163 	}
164 	for (maxoff = NDSLOTS(size); off < maxoff; ++off)
165 		if (map[off] != ~0UL)
166 			return (off * NDENTRIES + ffsl(~map[off]) - 1);
167 	return (size);
168 }
169 
170 /*
171  * Find the highest non-zero bit in the given bitmap, starting at low and
172  * not exceeding size - 1.
173  */
174 static int
175 fd_last_used(struct filedesc *fdp, int low, int size)
176 {
177 	NDSLOTTYPE *map = fdp->fd_map;
178 	NDSLOTTYPE mask;
179 	int off, minoff;
180 
181 	if (low >= size)
182 		return (-1);
183 
184 	off = NDSLOT(size);
185 	if (size % NDENTRIES) {
186 		mask = ~(~(NDSLOTTYPE)0 << (size % NDENTRIES));
187 		if ((mask &= map[off]) != 0)
188 			return (off * NDENTRIES + flsl(mask) - 1);
189 		--off;
190 	}
191 	for (minoff = NDSLOT(low); off >= minoff; --off)
192 		if (map[off] != 0)
193 			return (off * NDENTRIES + flsl(map[off]) - 1);
194 	return (size - 1);
195 }
196 
197 static int
198 fdisused(struct filedesc *fdp, int fd)
199 {
200         KASSERT(fd >= 0 && fd < fdp->fd_nfiles,
201             ("file descriptor %d out of range (0, %d)", fd, fdp->fd_nfiles));
202 	return ((fdp->fd_map[NDSLOT(fd)] & NDBIT(fd)) != 0);
203 }
204 
205 /*
206  * Mark a file descriptor as used.
207  */
208 static void
209 fdused(struct filedesc *fdp, int fd)
210 {
211 	FILEDESC_LOCK_ASSERT(fdp, MA_OWNED);
212 	KASSERT(!fdisused(fdp, fd),
213 	    ("fd already used"));
214 	fdp->fd_map[NDSLOT(fd)] |= NDBIT(fd);
215 	if (fd > fdp->fd_lastfile)
216 		fdp->fd_lastfile = fd;
217 	if (fd == fdp->fd_freefile)
218 		fdp->fd_freefile = fd_first_free(fdp, fd, fdp->fd_nfiles);
219 }
220 
221 /*
222  * Mark a file descriptor as unused.
223  */
224 static void
225 fdunused(struct filedesc *fdp, int fd)
226 {
227 	FILEDESC_LOCK_ASSERT(fdp, MA_OWNED);
228 	KASSERT(fdisused(fdp, fd),
229 	    ("fd is already unused"));
230 	KASSERT(fdp->fd_ofiles[fd] == NULL,
231 	    ("fd is still in use"));
232 	fdp->fd_map[NDSLOT(fd)] &= ~NDBIT(fd);
233 	if (fd < fdp->fd_freefile)
234 		fdp->fd_freefile = fd;
235 	if (fd == fdp->fd_lastfile)
236 		fdp->fd_lastfile = fd_last_used(fdp, 0, fd);
237 }
238 
239 /*
240  * System calls on descriptors.
241  */
242 #ifndef _SYS_SYSPROTO_H_
243 struct getdtablesize_args {
244 	int	dummy;
245 };
246 #endif
247 /*
248  * MPSAFE
249  */
250 /* ARGSUSED */
251 int
252 getdtablesize(struct thread *td, struct getdtablesize_args *uap)
253 {
254 	struct proc *p = td->td_proc;
255 
256 	PROC_LOCK(p);
257 	td->td_retval[0] =
258 	    min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc);
259 	PROC_UNLOCK(p);
260 	return (0);
261 }
262 
263 /*
264  * Duplicate a file descriptor to a particular value.
265  *
266  * note: keep in mind that a potential race condition exists when closing
267  * descriptors from a shared descriptor table (via rfork).
268  */
269 #ifndef _SYS_SYSPROTO_H_
270 struct dup2_args {
271 	u_int	from;
272 	u_int	to;
273 };
274 #endif
275 /*
276  * MPSAFE
277  */
278 /* ARGSUSED */
279 int
280 dup2(struct thread *td, struct dup2_args *uap)
281 {
282 
283 	return (do_dup(td, DUP_FIXED, (int)uap->from, (int)uap->to,
284 		    td->td_retval));
285 }
286 
287 /*
288  * Duplicate a file descriptor.
289  */
290 #ifndef _SYS_SYSPROTO_H_
291 struct dup_args {
292 	u_int	fd;
293 };
294 #endif
295 /*
296  * MPSAFE
297  */
298 /* ARGSUSED */
299 int
300 dup(struct thread *td, struct dup_args *uap)
301 {
302 
303 	return (do_dup(td, DUP_VARIABLE, (int)uap->fd, 0, td->td_retval));
304 }
305 
306 /*
307  * The file control system call.
308  */
309 #ifndef _SYS_SYSPROTO_H_
310 struct fcntl_args {
311 	int	fd;
312 	int	cmd;
313 	long	arg;
314 };
315 #endif
316 /*
317  * MPSAFE
318  */
319 /* ARGSUSED */
320 int
321 fcntl(struct thread *td, struct fcntl_args *uap)
322 {
323 	struct flock fl;
324 	intptr_t arg;
325 	int error;
326 
327 	error = 0;
328 	switch (uap->cmd) {
329 	case F_GETLK:
330 	case F_SETLK:
331 	case F_SETLKW:
332 		error = copyin((void *)(intptr_t)uap->arg, &fl, sizeof(fl));
333 		arg = (intptr_t)&fl;
334 		break;
335 	default:
336 		arg = uap->arg;
337 		break;
338 	}
339 	if (error)
340 		return (error);
341 	error = kern_fcntl(td, uap->fd, uap->cmd, arg);
342 	if (error)
343 		return (error);
344 	if (uap->cmd == F_GETLK)
345 		error = copyout(&fl, (void *)(intptr_t)uap->arg, sizeof(fl));
346 	return (error);
347 }
348 
349 int
350 kern_fcntl(struct thread *td, int fd, int cmd, intptr_t arg)
351 {
352 	struct filedesc *fdp;
353 	struct flock *flp;
354 	struct file *fp;
355 	struct proc *p;
356 	char *pop;
357 	struct vnode *vp;
358 	u_int newmin;
359 	int error, flg, tmp;
360 	int giant_locked;
361 
362 	/*
363 	 * XXXRW: Some fcntl() calls require Giant -- others don't.  Try to
364 	 * avoid grabbing Giant for calls we know don't need it.
365 	 */
366 	switch (cmd) {
367 	case F_DUPFD:
368 	case F_GETFD:
369 	case F_SETFD:
370 	case F_GETFL:
371 		giant_locked = 0;
372 		break;
373 
374 	default:
375 		giant_locked = 1;
376 		mtx_lock(&Giant);
377 	}
378 
379 	error = 0;
380 	flg = F_POSIX;
381 	p = td->td_proc;
382 	fdp = p->p_fd;
383 	FILEDESC_LOCK(fdp);
384 	if ((unsigned)fd >= fdp->fd_nfiles ||
385 	    (fp = fdp->fd_ofiles[fd]) == NULL) {
386 		FILEDESC_UNLOCK(fdp);
387 		error = EBADF;
388 		goto done2;
389 	}
390 	pop = &fdp->fd_ofileflags[fd];
391 
392 	switch (cmd) {
393 	case F_DUPFD:
394 		/* mtx_assert(&Giant, MA_NOTOWNED); */
395 		FILEDESC_UNLOCK(fdp);
396 		newmin = arg;
397 		PROC_LOCK(p);
398 		if (newmin >= lim_cur(p, RLIMIT_NOFILE) ||
399 		    newmin >= maxfilesperproc) {
400 			PROC_UNLOCK(p);
401 			error = EINVAL;
402 			break;
403 		}
404 		PROC_UNLOCK(p);
405 		error = do_dup(td, DUP_VARIABLE, fd, newmin, td->td_retval);
406 		break;
407 
408 	case F_GETFD:
409 		/* mtx_assert(&Giant, MA_NOTOWNED); */
410 		td->td_retval[0] = (*pop & UF_EXCLOSE) ? FD_CLOEXEC : 0;
411 		FILEDESC_UNLOCK(fdp);
412 		break;
413 
414 	case F_SETFD:
415 		/* mtx_assert(&Giant, MA_NOTOWNED); */
416 		*pop = (*pop &~ UF_EXCLOSE) |
417 		    (arg & FD_CLOEXEC ? UF_EXCLOSE : 0);
418 		FILEDESC_UNLOCK(fdp);
419 		break;
420 
421 	case F_GETFL:
422 		/* mtx_assert(&Giant, MA_NOTOWNED); */
423 		FILE_LOCK(fp);
424 		td->td_retval[0] = OFLAGS(fp->f_flag);
425 		FILE_UNLOCK(fp);
426 		FILEDESC_UNLOCK(fdp);
427 		break;
428 
429 	case F_SETFL:
430 		mtx_assert(&Giant, MA_OWNED);
431 		FILE_LOCK(fp);
432 		fhold_locked(fp);
433 		fp->f_flag &= ~FCNTLFLAGS;
434 		fp->f_flag |= FFLAGS(arg & ~O_ACCMODE) & FCNTLFLAGS;
435 		FILE_UNLOCK(fp);
436 		FILEDESC_UNLOCK(fdp);
437 		tmp = fp->f_flag & FNONBLOCK;
438 		error = fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
439 		if (error) {
440 			fdrop(fp, td);
441 			break;
442 		}
443 		tmp = fp->f_flag & FASYNC;
444 		error = fo_ioctl(fp, FIOASYNC, &tmp, td->td_ucred, td);
445 		if (error == 0) {
446 			fdrop(fp, td);
447 			break;
448 		}
449 		FILE_LOCK(fp);
450 		fp->f_flag &= ~FNONBLOCK;
451 		FILE_UNLOCK(fp);
452 		tmp = 0;
453 		(void)fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
454 		fdrop(fp, td);
455 		break;
456 
457 	case F_GETOWN:
458 		mtx_assert(&Giant, MA_OWNED);
459 		fhold(fp);
460 		FILEDESC_UNLOCK(fdp);
461 		error = fo_ioctl(fp, FIOGETOWN, &tmp, td->td_ucred, td);
462 		if (error == 0)
463 			td->td_retval[0] = tmp;
464 		fdrop(fp, td);
465 		break;
466 
467 	case F_SETOWN:
468 		mtx_assert(&Giant, MA_OWNED);
469 		fhold(fp);
470 		FILEDESC_UNLOCK(fdp);
471 		tmp = arg;
472 		error = fo_ioctl(fp, FIOSETOWN, &tmp, td->td_ucred, td);
473 		fdrop(fp, td);
474 		break;
475 
476 	case F_SETLKW:
477 		mtx_assert(&Giant, MA_OWNED);
478 		flg |= F_WAIT;
479 		/* FALLTHROUGH F_SETLK */
480 
481 	case F_SETLK:
482 		mtx_assert(&Giant, MA_OWNED);
483 		if (fp->f_type != DTYPE_VNODE) {
484 			FILEDESC_UNLOCK(fdp);
485 			error = EBADF;
486 			break;
487 		}
488 
489 		flp = (struct flock *)arg;
490 		if (flp->l_whence == SEEK_CUR) {
491 			if (fp->f_offset < 0 ||
492 			    (flp->l_start > 0 &&
493 			     fp->f_offset > OFF_MAX - flp->l_start)) {
494 				FILEDESC_UNLOCK(fdp);
495 				error = EOVERFLOW;
496 				break;
497 			}
498 			flp->l_start += fp->f_offset;
499 		}
500 
501 		/*
502 		 * VOP_ADVLOCK() may block.
503 		 */
504 		fhold(fp);
505 		FILEDESC_UNLOCK(fdp);
506 		vp = fp->f_vnode;
507 
508 		switch (flp->l_type) {
509 		case F_RDLCK:
510 			if ((fp->f_flag & FREAD) == 0) {
511 				error = EBADF;
512 				break;
513 			}
514 			PROC_LOCK(p->p_leader);
515 			p->p_leader->p_flag |= P_ADVLOCK;
516 			PROC_UNLOCK(p->p_leader);
517 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
518 			    flp, flg);
519 			break;
520 		case F_WRLCK:
521 			if ((fp->f_flag & FWRITE) == 0) {
522 				error = EBADF;
523 				break;
524 			}
525 			PROC_LOCK(p->p_leader);
526 			p->p_leader->p_flag |= P_ADVLOCK;
527 			PROC_UNLOCK(p->p_leader);
528 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
529 			    flp, flg);
530 			break;
531 		case F_UNLCK:
532 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
533 			    flp, F_POSIX);
534 			break;
535 		default:
536 			error = EINVAL;
537 			break;
538 		}
539 		/* Check for race with close */
540 		FILEDESC_LOCK_FAST(fdp);
541 		if ((unsigned) fd >= fdp->fd_nfiles ||
542 		    fp != fdp->fd_ofiles[fd]) {
543 			FILEDESC_UNLOCK_FAST(fdp);
544 			flp->l_whence = SEEK_SET;
545 			flp->l_start = 0;
546 			flp->l_len = 0;
547 			flp->l_type = F_UNLCK;
548 			(void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
549 					   F_UNLCK, flp, F_POSIX);
550 		} else
551 			FILEDESC_UNLOCK_FAST(fdp);
552 		fdrop(fp, td);
553 		break;
554 
555 	case F_GETLK:
556 		mtx_assert(&Giant, MA_OWNED);
557 		if (fp->f_type != DTYPE_VNODE) {
558 			FILEDESC_UNLOCK(fdp);
559 			error = EBADF;
560 			break;
561 		}
562 		flp = (struct flock *)arg;
563 		if (flp->l_type != F_RDLCK && flp->l_type != F_WRLCK &&
564 		    flp->l_type != F_UNLCK) {
565 			FILEDESC_UNLOCK(fdp);
566 			error = EINVAL;
567 			break;
568 		}
569 		if (flp->l_whence == SEEK_CUR) {
570 			if ((flp->l_start > 0 &&
571 			    fp->f_offset > OFF_MAX - flp->l_start) ||
572 			    (flp->l_start < 0 &&
573 			     fp->f_offset < OFF_MIN - flp->l_start)) {
574 				FILEDESC_UNLOCK(fdp);
575 				error = EOVERFLOW;
576 				break;
577 			}
578 			flp->l_start += fp->f_offset;
579 		}
580 		/*
581 		 * VOP_ADVLOCK() may block.
582 		 */
583 		fhold(fp);
584 		FILEDESC_UNLOCK(fdp);
585 		vp = fp->f_vnode;
586 		error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, flp,
587 		    F_POSIX);
588 		fdrop(fp, td);
589 		break;
590 	default:
591 		FILEDESC_UNLOCK(fdp);
592 		error = EINVAL;
593 		break;
594 	}
595 done2:
596 	if (giant_locked)
597 		mtx_unlock(&Giant);
598 	return (error);
599 }
600 
601 /*
602  * Common code for dup, dup2, and fcntl(F_DUPFD).
603  */
604 static int
605 do_dup(struct thread *td, enum dup_type type, int old, int new, register_t *retval)
606 {
607 	struct filedesc *fdp;
608 	struct proc *p;
609 	struct file *fp;
610 	struct file *delfp;
611 	int error, holdleaders, maxfd;
612 
613 	KASSERT((type == DUP_VARIABLE || type == DUP_FIXED),
614 	    ("invalid dup type %d", type));
615 
616 	p = td->td_proc;
617 	fdp = p->p_fd;
618 
619 	/*
620 	 * Verify we have a valid descriptor to dup from and possibly to
621 	 * dup to.
622 	 */
623 	if (old < 0 || new < 0)
624 		return (EBADF);
625 	PROC_LOCK(p);
626 	maxfd = min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc);
627 	PROC_UNLOCK(p);
628 	if (new >= maxfd)
629 		return (EMFILE);
630 
631 	FILEDESC_LOCK(fdp);
632 	if (old >= fdp->fd_nfiles || fdp->fd_ofiles[old] == NULL) {
633 		FILEDESC_UNLOCK(fdp);
634 		return (EBADF);
635 	}
636 	if (type == DUP_FIXED && old == new) {
637 		*retval = new;
638 		FILEDESC_UNLOCK(fdp);
639 		return (0);
640 	}
641 	fp = fdp->fd_ofiles[old];
642 	fhold(fp);
643 
644 	/*
645 	 * If the caller specified a file descriptor, make sure the file
646 	 * table is large enough to hold it, and grab it.  Otherwise, just
647 	 * allocate a new descriptor the usual way.  Since the filedesc
648 	 * lock may be temporarily dropped in the process, we have to look
649 	 * out for a race.
650 	 */
651 	if (type == DUP_FIXED) {
652 		if (new >= fdp->fd_nfiles)
653 			fdgrowtable(fdp, new + 1);
654 		if (fdp->fd_ofiles[new] == NULL)
655 			fdused(fdp, new);
656 	} else {
657 		if ((error = fdalloc(td, new, &new)) != 0) {
658 			FILEDESC_UNLOCK(fdp);
659 			fdrop(fp, td);
660 			return (error);
661 		}
662 	}
663 
664 	/*
665 	 * If the old file changed out from under us then treat it as a
666 	 * bad file descriptor.  Userland should do its own locking to
667 	 * avoid this case.
668 	 */
669 	if (fdp->fd_ofiles[old] != fp) {
670 		/* we've allocated a descriptor which we won't use */
671 		if (fdp->fd_ofiles[new] == NULL)
672 			fdunused(fdp, new);
673 		FILEDESC_UNLOCK(fdp);
674 		fdrop(fp, td);
675 		return (EBADF);
676 	}
677 	KASSERT(old != new,
678 	    ("new fd is same as old"));
679 
680 	/*
681 	 * Save info on the descriptor being overwritten.  We cannot close
682 	 * it without introducing an ownership race for the slot, since we
683 	 * need to drop the filedesc lock to call closef().
684 	 *
685 	 * XXX this duplicates parts of close().
686 	 */
687 	delfp = fdp->fd_ofiles[new];
688 	holdleaders = 0;
689 	if (delfp != NULL) {
690 		if (td->td_proc->p_fdtol != NULL) {
691 			/*
692 			 * Ask fdfree() to sleep to ensure that all relevant
693 			 * process leaders can be traversed in closef().
694 			 */
695 			fdp->fd_holdleaderscount++;
696 			holdleaders = 1;
697 		}
698 	}
699 
700 	/*
701 	 * Duplicate the source descriptor
702 	 */
703 	fdp->fd_ofiles[new] = fp;
704 	fdp->fd_ofileflags[new] = fdp->fd_ofileflags[old] &~ UF_EXCLOSE;
705 	if (new > fdp->fd_lastfile)
706 		fdp->fd_lastfile = new;
707 	*retval = new;
708 
709 	/*
710 	 * If we dup'd over a valid file, we now own the reference to it
711 	 * and must dispose of it using closef() semantics (as if a
712 	 * close() were performed on it).
713 	 *
714 	 * XXX this duplicates parts of close().
715 	 */
716 	if (delfp != NULL) {
717 		knote_fdclose(td, new);
718 		if (delfp->f_type == DTYPE_MQUEUE)
719 			mq_fdclose(td, new, delfp);
720 		FILEDESC_UNLOCK(fdp);
721 		(void) closef(delfp, td);
722 		if (holdleaders) {
723 			FILEDESC_LOCK_FAST(fdp);
724 			fdp->fd_holdleaderscount--;
725 			if (fdp->fd_holdleaderscount == 0 &&
726 			    fdp->fd_holdleaderswakeup != 0) {
727 				fdp->fd_holdleaderswakeup = 0;
728 				wakeup(&fdp->fd_holdleaderscount);
729 			}
730 			FILEDESC_UNLOCK_FAST(fdp);
731 		}
732 	} else {
733 		FILEDESC_UNLOCK(fdp);
734 	}
735 	return (0);
736 }
737 
738 /*
739  * If sigio is on the list associated with a process or process group,
740  * disable signalling from the device, remove sigio from the list and
741  * free sigio.
742  */
743 void
744 funsetown(struct sigio **sigiop)
745 {
746 	struct sigio *sigio;
747 
748 	SIGIO_LOCK();
749 	sigio = *sigiop;
750 	if (sigio == NULL) {
751 		SIGIO_UNLOCK();
752 		return;
753 	}
754 	*(sigio->sio_myref) = NULL;
755 	if ((sigio)->sio_pgid < 0) {
756 		struct pgrp *pg = (sigio)->sio_pgrp;
757 		PGRP_LOCK(pg);
758 		SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio,
759 			     sigio, sio_pgsigio);
760 		PGRP_UNLOCK(pg);
761 	} else {
762 		struct proc *p = (sigio)->sio_proc;
763 		PROC_LOCK(p);
764 		SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio,
765 			     sigio, sio_pgsigio);
766 		PROC_UNLOCK(p);
767 	}
768 	SIGIO_UNLOCK();
769 	crfree(sigio->sio_ucred);
770 	FREE(sigio, M_SIGIO);
771 }
772 
773 /*
774  * Free a list of sigio structures.
775  * We only need to lock the SIGIO_LOCK because we have made ourselves
776  * inaccessible to callers of fsetown and therefore do not need to lock
777  * the proc or pgrp struct for the list manipulation.
778  */
779 void
780 funsetownlst(struct sigiolst *sigiolst)
781 {
782 	struct proc *p;
783 	struct pgrp *pg;
784 	struct sigio *sigio;
785 
786 	sigio = SLIST_FIRST(sigiolst);
787 	if (sigio == NULL)
788 		return;
789 	p = NULL;
790 	pg = NULL;
791 
792 	/*
793 	 * Every entry of the list should belong
794 	 * to a single proc or pgrp.
795 	 */
796 	if (sigio->sio_pgid < 0) {
797 		pg = sigio->sio_pgrp;
798 		PGRP_LOCK_ASSERT(pg, MA_NOTOWNED);
799 	} else /* if (sigio->sio_pgid > 0) */ {
800 		p = sigio->sio_proc;
801 		PROC_LOCK_ASSERT(p, MA_NOTOWNED);
802 	}
803 
804 	SIGIO_LOCK();
805 	while ((sigio = SLIST_FIRST(sigiolst)) != NULL) {
806 		*(sigio->sio_myref) = NULL;
807 		if (pg != NULL) {
808 			KASSERT(sigio->sio_pgid < 0,
809 			    ("Proc sigio in pgrp sigio list"));
810 			KASSERT(sigio->sio_pgrp == pg,
811 			    ("Bogus pgrp in sigio list"));
812 			PGRP_LOCK(pg);
813 			SLIST_REMOVE(&pg->pg_sigiolst, sigio, sigio,
814 			    sio_pgsigio);
815 			PGRP_UNLOCK(pg);
816 		} else /* if (p != NULL) */ {
817 			KASSERT(sigio->sio_pgid > 0,
818 			    ("Pgrp sigio in proc sigio list"));
819 			KASSERT(sigio->sio_proc == p,
820 			    ("Bogus proc in sigio list"));
821 			PROC_LOCK(p);
822 			SLIST_REMOVE(&p->p_sigiolst, sigio, sigio,
823 			    sio_pgsigio);
824 			PROC_UNLOCK(p);
825 		}
826 		SIGIO_UNLOCK();
827 		crfree(sigio->sio_ucred);
828 		FREE(sigio, M_SIGIO);
829 		SIGIO_LOCK();
830 	}
831 	SIGIO_UNLOCK();
832 }
833 
834 /*
835  * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg).
836  *
837  * After permission checking, add a sigio structure to the sigio list for
838  * the process or process group.
839  */
840 int
841 fsetown(pid_t pgid, struct sigio **sigiop)
842 {
843 	struct proc *proc;
844 	struct pgrp *pgrp;
845 	struct sigio *sigio;
846 	int ret;
847 
848 	if (pgid == 0) {
849 		funsetown(sigiop);
850 		return (0);
851 	}
852 
853 	ret = 0;
854 
855 	/* Allocate and fill in the new sigio out of locks. */
856 	MALLOC(sigio, struct sigio *, sizeof(struct sigio), M_SIGIO, M_WAITOK);
857 	sigio->sio_pgid = pgid;
858 	sigio->sio_ucred = crhold(curthread->td_ucred);
859 	sigio->sio_myref = sigiop;
860 
861 	sx_slock(&proctree_lock);
862 	if (pgid > 0) {
863 		proc = pfind(pgid);
864 		if (proc == NULL) {
865 			ret = ESRCH;
866 			goto fail;
867 		}
868 
869 		/*
870 		 * Policy - Don't allow a process to FSETOWN a process
871 		 * in another session.
872 		 *
873 		 * Remove this test to allow maximum flexibility or
874 		 * restrict FSETOWN to the current process or process
875 		 * group for maximum safety.
876 		 */
877 		PROC_UNLOCK(proc);
878 		if (proc->p_session != curthread->td_proc->p_session) {
879 			ret = EPERM;
880 			goto fail;
881 		}
882 
883 		pgrp = NULL;
884 	} else /* if (pgid < 0) */ {
885 		pgrp = pgfind(-pgid);
886 		if (pgrp == NULL) {
887 			ret = ESRCH;
888 			goto fail;
889 		}
890 		PGRP_UNLOCK(pgrp);
891 
892 		/*
893 		 * Policy - Don't allow a process to FSETOWN a process
894 		 * in another session.
895 		 *
896 		 * Remove this test to allow maximum flexibility or
897 		 * restrict FSETOWN to the current process or process
898 		 * group for maximum safety.
899 		 */
900 		if (pgrp->pg_session != curthread->td_proc->p_session) {
901 			ret = EPERM;
902 			goto fail;
903 		}
904 
905 		proc = NULL;
906 	}
907 	funsetown(sigiop);
908 	if (pgid > 0) {
909 		PROC_LOCK(proc);
910 		/*
911 		 * Since funsetownlst() is called without the proctree
912 		 * locked, we need to check for P_WEXIT.
913 		 * XXX: is ESRCH correct?
914 		 */
915 		if ((proc->p_flag & P_WEXIT) != 0) {
916 			PROC_UNLOCK(proc);
917 			ret = ESRCH;
918 			goto fail;
919 		}
920 		SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio);
921 		sigio->sio_proc = proc;
922 		PROC_UNLOCK(proc);
923 	} else {
924 		PGRP_LOCK(pgrp);
925 		SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio);
926 		sigio->sio_pgrp = pgrp;
927 		PGRP_UNLOCK(pgrp);
928 	}
929 	sx_sunlock(&proctree_lock);
930 	SIGIO_LOCK();
931 	*sigiop = sigio;
932 	SIGIO_UNLOCK();
933 	return (0);
934 
935 fail:
936 	sx_sunlock(&proctree_lock);
937 	crfree(sigio->sio_ucred);
938 	FREE(sigio, M_SIGIO);
939 	return (ret);
940 }
941 
942 /*
943  * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg).
944  */
945 pid_t
946 fgetown(sigiop)
947 	struct sigio **sigiop;
948 {
949 	pid_t pgid;
950 
951 	SIGIO_LOCK();
952 	pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0;
953 	SIGIO_UNLOCK();
954 	return (pgid);
955 }
956 
957 /*
958  * Close a file descriptor.
959  */
960 #ifndef _SYS_SYSPROTO_H_
961 struct close_args {
962 	int     fd;
963 };
964 #endif
965 /*
966  * MPSAFE
967  */
968 /* ARGSUSED */
969 int
970 close(td, uap)
971 	struct thread *td;
972 	struct close_args *uap;
973 {
974 	struct filedesc *fdp;
975 	struct file *fp;
976 	int fd, error;
977 	int holdleaders;
978 
979 	fd = uap->fd;
980 	error = 0;
981 	holdleaders = 0;
982 	fdp = td->td_proc->p_fd;
983 	FILEDESC_LOCK(fdp);
984 	if ((unsigned)fd >= fdp->fd_nfiles ||
985 	    (fp = fdp->fd_ofiles[fd]) == NULL) {
986 		FILEDESC_UNLOCK(fdp);
987 		return (EBADF);
988 	}
989 	fdp->fd_ofiles[fd] = NULL;
990 	fdp->fd_ofileflags[fd] = 0;
991 	fdunused(fdp, fd);
992 	if (td->td_proc->p_fdtol != NULL) {
993 		/*
994 		 * Ask fdfree() to sleep to ensure that all relevant
995 		 * process leaders can be traversed in closef().
996 		 */
997 		fdp->fd_holdleaderscount++;
998 		holdleaders = 1;
999 	}
1000 
1001 	/*
1002 	 * We now hold the fp reference that used to be owned by the descriptor
1003 	 * array.
1004 	 * We have to unlock the FILEDESC *AFTER* knote_fdclose to prevent a
1005 	 * race of the fd getting opened, a knote added, and deleteing a knote
1006 	 * for the new fd.
1007 	 */
1008 	knote_fdclose(td, fd);
1009 	if (fp->f_type == DTYPE_MQUEUE)
1010 		mq_fdclose(td, fd, fp);
1011 	FILEDESC_UNLOCK(fdp);
1012 
1013 	error = closef(fp, td);
1014 	if (holdleaders) {
1015 		FILEDESC_LOCK_FAST(fdp);
1016 		fdp->fd_holdleaderscount--;
1017 		if (fdp->fd_holdleaderscount == 0 &&
1018 		    fdp->fd_holdleaderswakeup != 0) {
1019 			fdp->fd_holdleaderswakeup = 0;
1020 			wakeup(&fdp->fd_holdleaderscount);
1021 		}
1022 		FILEDESC_UNLOCK_FAST(fdp);
1023 	}
1024 	return (error);
1025 }
1026 
1027 #if defined(COMPAT_43)
1028 /*
1029  * Return status information about a file descriptor.
1030  */
1031 #ifndef _SYS_SYSPROTO_H_
1032 struct ofstat_args {
1033 	int	fd;
1034 	struct	ostat *sb;
1035 };
1036 #endif
1037 /*
1038  * MPSAFE
1039  */
1040 /* ARGSUSED */
1041 int
1042 ofstat(struct thread *td, struct ofstat_args *uap)
1043 {
1044 	struct ostat oub;
1045 	struct stat ub;
1046 	int error;
1047 
1048 	error = kern_fstat(td, uap->fd, &ub);
1049 	if (error == 0) {
1050 		cvtstat(&ub, &oub);
1051 		error = copyout(&oub, uap->sb, sizeof(oub));
1052 	}
1053 	return (error);
1054 }
1055 #endif /* COMPAT_43 */
1056 
1057 /*
1058  * Return status information about a file descriptor.
1059  */
1060 #ifndef _SYS_SYSPROTO_H_
1061 struct fstat_args {
1062 	int	fd;
1063 	struct	stat *sb;
1064 };
1065 #endif
1066 /*
1067  * MPSAFE
1068  */
1069 /* ARGSUSED */
1070 int
1071 fstat(struct thread *td, struct fstat_args *uap)
1072 {
1073 	struct stat ub;
1074 	int error;
1075 
1076 	error = kern_fstat(td, uap->fd, &ub);
1077 	if (error == 0)
1078 		error = copyout(&ub, uap->sb, sizeof(ub));
1079 	return (error);
1080 }
1081 
1082 int
1083 kern_fstat(struct thread *td, int fd, struct stat *sbp)
1084 {
1085 	struct file *fp;
1086 	int error;
1087 
1088 	if ((error = fget(td, fd, &fp)) != 0)
1089 		return (error);
1090 	error = fo_stat(fp, sbp, td->td_ucred, td);
1091 	fdrop(fp, td);
1092 	return (error);
1093 }
1094 
1095 /*
1096  * Return status information about a file descriptor.
1097  */
1098 #ifndef _SYS_SYSPROTO_H_
1099 struct nfstat_args {
1100 	int	fd;
1101 	struct	nstat *sb;
1102 };
1103 #endif
1104 /*
1105  * MPSAFE
1106  */
1107 /* ARGSUSED */
1108 int
1109 nfstat(struct thread *td, struct nfstat_args *uap)
1110 {
1111 	struct nstat nub;
1112 	struct stat ub;
1113 	int error;
1114 
1115 	error = kern_fstat(td, uap->fd, &ub);
1116 	if (error == 0) {
1117 		cvtnstat(&ub, &nub);
1118 		error = copyout(&nub, uap->sb, sizeof(nub));
1119 	}
1120 	return (error);
1121 }
1122 
1123 /*
1124  * Return pathconf information about a file descriptor.
1125  */
1126 #ifndef _SYS_SYSPROTO_H_
1127 struct fpathconf_args {
1128 	int	fd;
1129 	int	name;
1130 };
1131 #endif
1132 /*
1133  * MPSAFE
1134  */
1135 /* ARGSUSED */
1136 int
1137 fpathconf(struct thread *td, struct fpathconf_args *uap)
1138 {
1139 	struct file *fp;
1140 	struct vnode *vp;
1141 	int error;
1142 
1143 	if ((error = fget(td, uap->fd, &fp)) != 0)
1144 		return (error);
1145 
1146 	/* If asynchronous I/O is available, it works for all descriptors. */
1147 	if (uap->name == _PC_ASYNC_IO) {
1148 		td->td_retval[0] = async_io_version;
1149 		goto out;
1150 	}
1151 	vp = fp->f_vnode;
1152 	if (vp != NULL) {
1153 		int vfslocked;
1154 		vfslocked = VFS_LOCK_GIANT(vp->v_mount);
1155 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
1156 		error = VOP_PATHCONF(vp, uap->name, td->td_retval);
1157 		VOP_UNLOCK(vp, 0, td);
1158 		VFS_UNLOCK_GIANT(vfslocked);
1159 	} else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) {
1160 		if (uap->name != _PC_PIPE_BUF) {
1161 			error = EINVAL;
1162 		} else {
1163 			td->td_retval[0] = PIPE_BUF;
1164 		error = 0;
1165 		}
1166 	} else {
1167 		error = EOPNOTSUPP;
1168 	}
1169 out:
1170 	fdrop(fp, td);
1171 	return (error);
1172 }
1173 
1174 /*
1175  * Grow the file table to accomodate (at least) nfd descriptors.  This may
1176  * block and drop the filedesc lock, but it will reacquire it before
1177  * returning.
1178  */
1179 static void
1180 fdgrowtable(struct filedesc *fdp, int nfd)
1181 {
1182 	struct file **ntable;
1183 	char *nfileflags;
1184 	int nnfiles, onfiles;
1185 	NDSLOTTYPE *nmap;
1186 
1187 	FILEDESC_LOCK_ASSERT(fdp, MA_OWNED);
1188 
1189 	KASSERT(fdp->fd_nfiles > 0,
1190 	    ("zero-length file table"));
1191 
1192 	/* compute the size of the new table */
1193 	onfiles = fdp->fd_nfiles;
1194 	nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */
1195 	if (nnfiles <= onfiles)
1196 		/* the table is already large enough */
1197 		return;
1198 
1199 	/* allocate a new table and (if required) new bitmaps */
1200 	FILEDESC_UNLOCK(fdp);
1201 	MALLOC(ntable, struct file **, nnfiles * OFILESIZE,
1202 	    M_FILEDESC, M_ZERO | M_WAITOK);
1203 	nfileflags = (char *)&ntable[nnfiles];
1204 	if (NDSLOTS(nnfiles) > NDSLOTS(onfiles))
1205 		MALLOC(nmap, NDSLOTTYPE *, NDSLOTS(nnfiles) * NDSLOTSIZE,
1206 		    M_FILEDESC, M_ZERO | M_WAITOK);
1207 	else
1208 		nmap = NULL;
1209 	FILEDESC_LOCK(fdp);
1210 
1211 	/*
1212 	 * We now have new tables ready to go.  Since we dropped the
1213 	 * filedesc lock to call malloc(), watch out for a race.
1214 	 */
1215 	onfiles = fdp->fd_nfiles;
1216 	if (onfiles >= nnfiles) {
1217 		/* we lost the race, but that's OK */
1218 		free(ntable, M_FILEDESC);
1219 		if (nmap != NULL)
1220 			free(nmap, M_FILEDESC);
1221 		return;
1222 	}
1223 	bcopy(fdp->fd_ofiles, ntable, onfiles * sizeof(*ntable));
1224 	bcopy(fdp->fd_ofileflags, nfileflags, onfiles);
1225 	if (onfiles > NDFILE)
1226 		free(fdp->fd_ofiles, M_FILEDESC);
1227 	fdp->fd_ofiles = ntable;
1228 	fdp->fd_ofileflags = nfileflags;
1229 	if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) {
1230 		bcopy(fdp->fd_map, nmap, NDSLOTS(onfiles) * sizeof(*nmap));
1231 		if (NDSLOTS(onfiles) > NDSLOTS(NDFILE))
1232 			free(fdp->fd_map, M_FILEDESC);
1233 		fdp->fd_map = nmap;
1234 	}
1235 	fdp->fd_nfiles = nnfiles;
1236 }
1237 
1238 /*
1239  * Allocate a file descriptor for the process.
1240  */
1241 int
1242 fdalloc(struct thread *td, int minfd, int *result)
1243 {
1244 	struct proc *p = td->td_proc;
1245 	struct filedesc *fdp = p->p_fd;
1246 	int fd = -1, maxfd;
1247 
1248 	FILEDESC_LOCK_ASSERT(fdp, MA_OWNED);
1249 
1250 	if (fdp->fd_freefile > minfd)
1251 		minfd = fdp->fd_freefile;
1252 
1253 	PROC_LOCK(p);
1254 	maxfd = min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc);
1255 	PROC_UNLOCK(p);
1256 
1257 	/*
1258 	 * Search the bitmap for a free descriptor.  If none is found, try
1259 	 * to grow the file table.  Keep at it until we either get a file
1260 	 * descriptor or run into process or system limits; fdgrowtable()
1261 	 * may drop the filedesc lock, so we're in a race.
1262 	 */
1263 	for (;;) {
1264 		fd = fd_first_free(fdp, minfd, fdp->fd_nfiles);
1265 		if (fd >= maxfd)
1266 			return (EMFILE);
1267 		if (fd < fdp->fd_nfiles)
1268 			break;
1269 		fdgrowtable(fdp, min(fdp->fd_nfiles * 2, maxfd));
1270 	}
1271 
1272 	/*
1273 	 * Perform some sanity checks, then mark the file descriptor as
1274 	 * used and return it to the caller.
1275 	 */
1276 	KASSERT(!fdisused(fdp, fd),
1277 	    ("fd_first_free() returned non-free descriptor"));
1278 	KASSERT(fdp->fd_ofiles[fd] == NULL,
1279 	    ("free descriptor isn't"));
1280 	fdp->fd_ofileflags[fd] = 0; /* XXX needed? */
1281 	fdused(fdp, fd);
1282 	*result = fd;
1283 	return (0);
1284 }
1285 
1286 /*
1287  * Check to see whether n user file descriptors
1288  * are available to the process p.
1289  */
1290 int
1291 fdavail(struct thread *td, int n)
1292 {
1293 	struct proc *p = td->td_proc;
1294 	struct filedesc *fdp = td->td_proc->p_fd;
1295 	struct file **fpp;
1296 	int i, lim, last;
1297 
1298 	FILEDESC_LOCK_ASSERT(fdp, MA_OWNED);
1299 
1300 	PROC_LOCK(p);
1301 	lim = min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc);
1302 	PROC_UNLOCK(p);
1303 	if ((i = lim - fdp->fd_nfiles) > 0 && (n -= i) <= 0)
1304 		return (1);
1305 	last = min(fdp->fd_nfiles, lim);
1306 	fpp = &fdp->fd_ofiles[fdp->fd_freefile];
1307 	for (i = last - fdp->fd_freefile; --i >= 0; fpp++) {
1308 		if (*fpp == NULL && --n <= 0)
1309 			return (1);
1310 	}
1311 	return (0);
1312 }
1313 
1314 /*
1315  * Create a new open file structure and allocate
1316  * a file decriptor for the process that refers to it.
1317  * We add one reference to the file for the descriptor table
1318  * and one reference for resultfp. This is to prevent us being
1319  * preempted and the entry in the descriptor table closed after
1320  * we release the FILEDESC lock.
1321  */
1322 int
1323 falloc(struct thread *td, struct file **resultfp, int *resultfd)
1324 {
1325 	struct proc *p = td->td_proc;
1326 	struct file *fp, *fq;
1327 	int error, i;
1328 	int maxuserfiles = maxfiles - (maxfiles / 20);
1329 	static struct timeval lastfail;
1330 	static int curfail;
1331 
1332 	fp = uma_zalloc(file_zone, M_WAITOK | M_ZERO);
1333 	sx_xlock(&filelist_lock);
1334 	if ((openfiles >= maxuserfiles && (td->td_ucred->cr_ruid != 0 ||
1335 	   jailed(td->td_ucred))) || openfiles >= maxfiles) {
1336 		if (ppsratecheck(&lastfail, &curfail, 1)) {
1337 			printf("kern.maxfiles limit exceeded by uid %i, please see tuning(7).\n",
1338 				td->td_ucred->cr_ruid);
1339 		}
1340 		sx_xunlock(&filelist_lock);
1341 		uma_zfree(file_zone, fp);
1342 		return (ENFILE);
1343 	}
1344 	openfiles++;
1345 
1346 	/*
1347 	 * If the process has file descriptor zero open, add the new file
1348 	 * descriptor to the list of open files at that point, otherwise
1349 	 * put it at the front of the list of open files.
1350 	 */
1351 	fp->f_mtxp = mtx_pool_alloc(mtxpool_sleep);
1352 	fp->f_count = 1;
1353 	if (resultfp)
1354 		fp->f_count++;
1355 	fp->f_cred = crhold(td->td_ucred);
1356 	fp->f_ops = &badfileops;
1357 	fp->f_data = NULL;
1358 	fp->f_vnode = NULL;
1359 	FILEDESC_LOCK(p->p_fd);
1360 	if ((fq = p->p_fd->fd_ofiles[0])) {
1361 		LIST_INSERT_AFTER(fq, fp, f_list);
1362 	} else {
1363 		LIST_INSERT_HEAD(&filehead, fp, f_list);
1364 	}
1365 	sx_xunlock(&filelist_lock);
1366 	if ((error = fdalloc(td, 0, &i))) {
1367 		FILEDESC_UNLOCK(p->p_fd);
1368 		fdrop(fp, td);
1369 		if (resultfp)
1370 			fdrop(fp, td);
1371 		return (error);
1372 	}
1373 	p->p_fd->fd_ofiles[i] = fp;
1374 	FILEDESC_UNLOCK(p->p_fd);
1375 	if (resultfp)
1376 		*resultfp = fp;
1377 	if (resultfd)
1378 		*resultfd = i;
1379 	return (0);
1380 }
1381 
1382 /*
1383  * Build a new filedesc structure from another.
1384  * Copy the current, root, and jail root vnode references.
1385  */
1386 struct filedesc *
1387 fdinit(struct filedesc *fdp)
1388 {
1389 	struct filedesc0 *newfdp;
1390 
1391 	newfdp = malloc(sizeof *newfdp, M_FILEDESC, M_WAITOK | M_ZERO);
1392 	mtx_init(&newfdp->fd_fd.fd_mtx, FILEDESC_LOCK_DESC, NULL, MTX_DEF);
1393 	if (fdp != NULL) {
1394 		FILEDESC_LOCK(fdp);
1395 		newfdp->fd_fd.fd_cdir = fdp->fd_cdir;
1396 		if (newfdp->fd_fd.fd_cdir)
1397 			VREF(newfdp->fd_fd.fd_cdir);
1398 		newfdp->fd_fd.fd_rdir = fdp->fd_rdir;
1399 		if (newfdp->fd_fd.fd_rdir)
1400 			VREF(newfdp->fd_fd.fd_rdir);
1401 		newfdp->fd_fd.fd_jdir = fdp->fd_jdir;
1402 		if (newfdp->fd_fd.fd_jdir)
1403 			VREF(newfdp->fd_fd.fd_jdir);
1404 		FILEDESC_UNLOCK(fdp);
1405 	}
1406 
1407 	/* Create the file descriptor table. */
1408 	newfdp->fd_fd.fd_refcnt = 1;
1409 	newfdp->fd_fd.fd_holdcnt = 1;
1410 	newfdp->fd_fd.fd_cmask = CMASK;
1411 	newfdp->fd_fd.fd_ofiles = newfdp->fd_dfiles;
1412 	newfdp->fd_fd.fd_ofileflags = newfdp->fd_dfileflags;
1413 	newfdp->fd_fd.fd_nfiles = NDFILE;
1414 	newfdp->fd_fd.fd_map = newfdp->fd_dmap;
1415 	return (&newfdp->fd_fd);
1416 }
1417 
1418 static struct filedesc *
1419 fdhold(struct proc *p)
1420 {
1421 	struct filedesc *fdp;
1422 
1423 	mtx_lock(&fdesc_mtx);
1424 	fdp = p->p_fd;
1425 	if (fdp != NULL)
1426 		fdp->fd_holdcnt++;
1427 	mtx_unlock(&fdesc_mtx);
1428 	return (fdp);
1429 }
1430 
1431 static void
1432 fddrop(struct filedesc *fdp)
1433 {
1434 	int i;
1435 
1436 	mtx_lock(&fdesc_mtx);
1437 	i = --fdp->fd_holdcnt;
1438 	mtx_unlock(&fdesc_mtx);
1439 	if (i > 0)
1440 		return;
1441 
1442 	mtx_destroy(&fdp->fd_mtx);
1443 	FREE(fdp, M_FILEDESC);
1444 }
1445 
1446 /*
1447  * Share a filedesc structure.
1448  */
1449 struct filedesc *
1450 fdshare(struct filedesc *fdp)
1451 {
1452 	FILEDESC_LOCK_FAST(fdp);
1453 	fdp->fd_refcnt++;
1454 	FILEDESC_UNLOCK_FAST(fdp);
1455 	return (fdp);
1456 }
1457 
1458 /*
1459  * Unshare a filedesc structure, if necessary by making a copy
1460  */
1461 void
1462 fdunshare(struct proc *p, struct thread *td)
1463 {
1464 
1465 	FILEDESC_LOCK_FAST(p->p_fd);
1466 	if (p->p_fd->fd_refcnt > 1) {
1467 		struct filedesc *tmp;
1468 
1469 		FILEDESC_UNLOCK_FAST(p->p_fd);
1470 		tmp = fdcopy(p->p_fd);
1471 		fdfree(td);
1472 		p->p_fd = tmp;
1473 	} else
1474 		FILEDESC_UNLOCK_FAST(p->p_fd);
1475 }
1476 
1477 /*
1478  * Copy a filedesc structure.
1479  * A NULL pointer in returns a NULL reference, this is to ease callers,
1480  * not catch errors.
1481  */
1482 struct filedesc *
1483 fdcopy(struct filedesc *fdp)
1484 {
1485 	struct filedesc *newfdp;
1486 	int i;
1487 
1488 	/* Certain daemons might not have file descriptors. */
1489 	if (fdp == NULL)
1490 		return (NULL);
1491 
1492 	newfdp = fdinit(fdp);
1493 	FILEDESC_LOCK_FAST(fdp);
1494 	while (fdp->fd_lastfile >= newfdp->fd_nfiles) {
1495 		FILEDESC_UNLOCK_FAST(fdp);
1496 		FILEDESC_LOCK(newfdp);
1497 		fdgrowtable(newfdp, fdp->fd_lastfile + 1);
1498 		FILEDESC_UNLOCK(newfdp);
1499 		FILEDESC_LOCK_FAST(fdp);
1500 	}
1501 	/* copy everything except kqueue descriptors */
1502 	newfdp->fd_freefile = -1;
1503 	for (i = 0; i <= fdp->fd_lastfile; ++i) {
1504 		if (fdisused(fdp, i) &&
1505 		    fdp->fd_ofiles[i]->f_type != DTYPE_KQUEUE) {
1506 			newfdp->fd_ofiles[i] = fdp->fd_ofiles[i];
1507 			newfdp->fd_ofileflags[i] = fdp->fd_ofileflags[i];
1508 			fhold(newfdp->fd_ofiles[i]);
1509 			newfdp->fd_lastfile = i;
1510 		} else {
1511 			if (newfdp->fd_freefile == -1)
1512 				newfdp->fd_freefile = i;
1513 		}
1514 	}
1515 	FILEDESC_UNLOCK_FAST(fdp);
1516 	FILEDESC_LOCK(newfdp);
1517 	for (i = 0; i <= newfdp->fd_lastfile; ++i)
1518 		if (newfdp->fd_ofiles[i] != NULL)
1519 			fdused(newfdp, i);
1520 	FILEDESC_UNLOCK(newfdp);
1521 	FILEDESC_LOCK_FAST(fdp);
1522 	if (newfdp->fd_freefile == -1)
1523 		newfdp->fd_freefile = i;
1524 	newfdp->fd_cmask = fdp->fd_cmask;
1525 	FILEDESC_UNLOCK_FAST(fdp);
1526 	return (newfdp);
1527 }
1528 
1529 /*
1530  * Release a filedesc structure.
1531  */
1532 void
1533 fdfree(struct thread *td)
1534 {
1535 	struct filedesc *fdp;
1536 	struct file **fpp;
1537 	int i, locked;
1538 	struct filedesc_to_leader *fdtol;
1539 	struct file *fp;
1540 	struct vnode *cdir, *jdir, *rdir, *vp;
1541 	struct flock lf;
1542 
1543 	/* Certain daemons might not have file descriptors. */
1544 	fdp = td->td_proc->p_fd;
1545 	if (fdp == NULL)
1546 		return;
1547 
1548 	/* Check for special need to clear POSIX style locks */
1549 	fdtol = td->td_proc->p_fdtol;
1550 	if (fdtol != NULL) {
1551 		FILEDESC_LOCK(fdp);
1552 		KASSERT(fdtol->fdl_refcount > 0,
1553 			("filedesc_to_refcount botch: fdl_refcount=%d",
1554 			 fdtol->fdl_refcount));
1555 		if (fdtol->fdl_refcount == 1 &&
1556 		    (td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
1557 			for (i = 0, fpp = fdp->fd_ofiles;
1558 			     i <= fdp->fd_lastfile;
1559 			     i++, fpp++) {
1560 				if (*fpp == NULL ||
1561 				    (*fpp)->f_type != DTYPE_VNODE)
1562 					continue;
1563 				fp = *fpp;
1564 				fhold(fp);
1565 				FILEDESC_UNLOCK(fdp);
1566 				lf.l_whence = SEEK_SET;
1567 				lf.l_start = 0;
1568 				lf.l_len = 0;
1569 				lf.l_type = F_UNLCK;
1570 				vp = fp->f_vnode;
1571 				locked = VFS_LOCK_GIANT(vp->v_mount);
1572 				(void) VOP_ADVLOCK(vp,
1573 						   (caddr_t)td->td_proc->
1574 						   p_leader,
1575 						   F_UNLCK,
1576 						   &lf,
1577 						   F_POSIX);
1578 				VFS_UNLOCK_GIANT(locked);
1579 				FILEDESC_LOCK(fdp);
1580 				fdrop(fp, td);
1581 				fpp = fdp->fd_ofiles + i;
1582 			}
1583 		}
1584 	retry:
1585 		if (fdtol->fdl_refcount == 1) {
1586 			if (fdp->fd_holdleaderscount > 0 &&
1587 			    (td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
1588 				/*
1589 				 * close() or do_dup() has cleared a reference
1590 				 * in a shared file descriptor table.
1591 				 */
1592 				fdp->fd_holdleaderswakeup = 1;
1593 				msleep(&fdp->fd_holdleaderscount, &fdp->fd_mtx,
1594 				       PLOCK, "fdlhold", 0);
1595 				goto retry;
1596 			}
1597 			if (fdtol->fdl_holdcount > 0) {
1598 				/*
1599 				 * Ensure that fdtol->fdl_leader
1600 				 * remains valid in closef().
1601 				 */
1602 				fdtol->fdl_wakeup = 1;
1603 				msleep(fdtol, &fdp->fd_mtx,
1604 				       PLOCK, "fdlhold", 0);
1605 				goto retry;
1606 			}
1607 		}
1608 		fdtol->fdl_refcount--;
1609 		if (fdtol->fdl_refcount == 0 &&
1610 		    fdtol->fdl_holdcount == 0) {
1611 			fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
1612 			fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
1613 		} else
1614 			fdtol = NULL;
1615 		td->td_proc->p_fdtol = NULL;
1616 		FILEDESC_UNLOCK(fdp);
1617 		if (fdtol != NULL)
1618 			FREE(fdtol, M_FILEDESC_TO_LEADER);
1619 	}
1620 	FILEDESC_LOCK_FAST(fdp);
1621 	i = --fdp->fd_refcnt;
1622 	FILEDESC_UNLOCK_FAST(fdp);
1623 	if (i > 0)
1624 		return;
1625 	/*
1626 	 * We are the last reference to the structure, so we can
1627 	 * safely assume it will not change out from under us.
1628 	 */
1629 	fpp = fdp->fd_ofiles;
1630 	for (i = fdp->fd_lastfile; i-- >= 0; fpp++) {
1631 		if (*fpp)
1632 			(void) closef(*fpp, td);
1633 	}
1634 	FILEDESC_LOCK(fdp);
1635 
1636 	/* XXX This should happen earlier. */
1637 	mtx_lock(&fdesc_mtx);
1638 	td->td_proc->p_fd = NULL;
1639 	mtx_unlock(&fdesc_mtx);
1640 
1641 	if (fdp->fd_nfiles > NDFILE)
1642 		FREE(fdp->fd_ofiles, M_FILEDESC);
1643 	if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE))
1644 		FREE(fdp->fd_map, M_FILEDESC);
1645 
1646 	fdp->fd_nfiles = 0;
1647 
1648 	cdir = fdp->fd_cdir;
1649 	fdp->fd_cdir = NULL;
1650 	rdir = fdp->fd_rdir;
1651 	fdp->fd_rdir = NULL;
1652 	jdir = fdp->fd_jdir;
1653 	fdp->fd_jdir = NULL;
1654 	FILEDESC_UNLOCK(fdp);
1655 
1656 	if (cdir) {
1657 		locked = VFS_LOCK_GIANT(cdir->v_mount);
1658 		vrele(cdir);
1659 		VFS_UNLOCK_GIANT(locked);
1660 	}
1661 	if (rdir) {
1662 		locked = VFS_LOCK_GIANT(rdir->v_mount);
1663 		vrele(rdir);
1664 		VFS_UNLOCK_GIANT(locked);
1665 	}
1666 	if (jdir) {
1667 		locked = VFS_LOCK_GIANT(jdir->v_mount);
1668 		vrele(jdir);
1669 		VFS_UNLOCK_GIANT(locked);
1670 	}
1671 
1672 	fddrop(fdp);
1673 }
1674 
1675 /*
1676  * For setugid programs, we don't want to people to use that setugidness
1677  * to generate error messages which write to a file which otherwise would
1678  * otherwise be off-limits to the process.  We check for filesystems where
1679  * the vnode can change out from under us after execve (like [lin]procfs).
1680  *
1681  * Since setugidsafety calls this only for fd 0, 1 and 2, this check is
1682  * sufficient.  We also don't check for setugidness since we know we are.
1683  */
1684 static int
1685 is_unsafe(struct file *fp)
1686 {
1687 	if (fp->f_type == DTYPE_VNODE) {
1688 		struct vnode *vp = fp->f_vnode;
1689 
1690 		if ((vp->v_vflag & VV_PROCDEP) != 0)
1691 			return (1);
1692 	}
1693 	return (0);
1694 }
1695 
1696 /*
1697  * Make this setguid thing safe, if at all possible.
1698  */
1699 void
1700 setugidsafety(struct thread *td)
1701 {
1702 	struct filedesc *fdp;
1703 	int i;
1704 
1705 	/* Certain daemons might not have file descriptors. */
1706 	fdp = td->td_proc->p_fd;
1707 	if (fdp == NULL)
1708 		return;
1709 
1710 	/*
1711 	 * Note: fdp->fd_ofiles may be reallocated out from under us while
1712 	 * we are blocked in a close.  Be careful!
1713 	 */
1714 	FILEDESC_LOCK(fdp);
1715 	for (i = 0; i <= fdp->fd_lastfile; i++) {
1716 		if (i > 2)
1717 			break;
1718 		if (fdp->fd_ofiles[i] && is_unsafe(fdp->fd_ofiles[i])) {
1719 			struct file *fp;
1720 
1721 			knote_fdclose(td, i);
1722 			/*
1723 			 * NULL-out descriptor prior to close to avoid
1724 			 * a race while close blocks.
1725 			 */
1726 			fp = fdp->fd_ofiles[i];
1727 			fdp->fd_ofiles[i] = NULL;
1728 			fdp->fd_ofileflags[i] = 0;
1729 			fdunused(fdp, i);
1730 			FILEDESC_UNLOCK(fdp);
1731 			(void) closef(fp, td);
1732 			FILEDESC_LOCK(fdp);
1733 		}
1734 	}
1735 	FILEDESC_UNLOCK(fdp);
1736 }
1737 
1738 void
1739 fdclose(struct filedesc *fdp, struct file *fp, int idx, struct thread *td)
1740 {
1741 
1742 	FILEDESC_LOCK(fdp);
1743 	if (fdp->fd_ofiles[idx] == fp) {
1744 		fdp->fd_ofiles[idx] = NULL;
1745 		fdunused(fdp, idx);
1746 		FILEDESC_UNLOCK(fdp);
1747 		fdrop(fp, td);
1748 	} else {
1749 		FILEDESC_UNLOCK(fdp);
1750 	}
1751 }
1752 
1753 /*
1754  * Close any files on exec?
1755  */
1756 void
1757 fdcloseexec(struct thread *td)
1758 {
1759 	struct filedesc *fdp;
1760 	int i;
1761 
1762 	/* Certain daemons might not have file descriptors. */
1763 	fdp = td->td_proc->p_fd;
1764 	if (fdp == NULL)
1765 		return;
1766 
1767 	FILEDESC_LOCK(fdp);
1768 
1769 	/*
1770 	 * We cannot cache fd_ofiles or fd_ofileflags since operations
1771 	 * may block and rip them out from under us.
1772 	 */
1773 	for (i = 0; i <= fdp->fd_lastfile; i++) {
1774 		if (fdp->fd_ofiles[i] != NULL &&
1775 		    (fdp->fd_ofiles[i]->f_type == DTYPE_MQUEUE ||
1776 		    (fdp->fd_ofileflags[i] & UF_EXCLOSE))) {
1777 			struct file *fp;
1778 
1779 			knote_fdclose(td, i);
1780 			/*
1781 			 * NULL-out descriptor prior to close to avoid
1782 			 * a race while close blocks.
1783 			 */
1784 			fp = fdp->fd_ofiles[i];
1785 			fdp->fd_ofiles[i] = NULL;
1786 			fdp->fd_ofileflags[i] = 0;
1787 			fdunused(fdp, i);
1788 			if (fp->f_type == DTYPE_MQUEUE)
1789 				mq_fdclose(td, i, fp);
1790 			FILEDESC_UNLOCK(fdp);
1791 			(void) closef(fp, td);
1792 			FILEDESC_LOCK(fdp);
1793 		}
1794 	}
1795 	FILEDESC_UNLOCK(fdp);
1796 }
1797 
1798 /*
1799  * It is unsafe for set[ug]id processes to be started with file
1800  * descriptors 0..2 closed, as these descriptors are given implicit
1801  * significance in the Standard C library.  fdcheckstd() will create a
1802  * descriptor referencing /dev/null for each of stdin, stdout, and
1803  * stderr that is not already open.
1804  */
1805 int
1806 fdcheckstd(struct thread *td)
1807 {
1808 	struct nameidata nd;
1809 	struct filedesc *fdp;
1810 	struct file *fp;
1811 	register_t retval;
1812 	int fd, i, error, flags, devnull;
1813 
1814 	fdp = td->td_proc->p_fd;
1815 	if (fdp == NULL)
1816 		return (0);
1817 	KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
1818 	devnull = -1;
1819 	error = 0;
1820 	for (i = 0; i < 3; i++) {
1821 		if (fdp->fd_ofiles[i] != NULL)
1822 			continue;
1823 		if (devnull < 0) {
1824 			int vfslocked;
1825 			error = falloc(td, &fp, &fd);
1826 			if (error != 0)
1827 				break;
1828 			/* Note extra ref on `fp' held for us by falloc(). */
1829 			KASSERT(fd == i, ("oof, we didn't get our fd"));
1830 			NDINIT(&nd, LOOKUP, FOLLOW | MPSAFE, UIO_SYSSPACE,
1831 			    "/dev/null", td);
1832 			flags = FREAD | FWRITE;
1833 			error = vn_open(&nd, &flags, 0, fd);
1834 			if (error != 0) {
1835 				/*
1836 				 * Someone may have closed the entry in the
1837 				 * file descriptor table, so check it hasn't
1838 				 * changed before dropping the reference count.
1839 				 */
1840 				FILEDESC_LOCK(fdp);
1841 				KASSERT(fdp->fd_ofiles[fd] == fp,
1842 				    ("table not shared, how did it change?"));
1843 				fdp->fd_ofiles[fd] = NULL;
1844 				fdunused(fdp, fd);
1845 				FILEDESC_UNLOCK(fdp);
1846 				fdrop(fp, td);
1847 				fdrop(fp, td);
1848 				break;
1849 			}
1850 			vfslocked = NDHASGIANT(&nd);
1851 			NDFREE(&nd, NDF_ONLY_PNBUF);
1852 			fp->f_flag = flags;
1853 			fp->f_vnode = nd.ni_vp;
1854 			if (fp->f_data == NULL)
1855 				fp->f_data = nd.ni_vp;
1856 			if (fp->f_ops == &badfileops)
1857 				fp->f_ops = &vnops;
1858 			fp->f_type = DTYPE_VNODE;
1859 			VOP_UNLOCK(nd.ni_vp, 0, td);
1860 			VFS_UNLOCK_GIANT(vfslocked);
1861 			devnull = fd;
1862 			fdrop(fp, td);
1863 		} else {
1864 			error = do_dup(td, DUP_FIXED, devnull, i, &retval);
1865 			if (error != 0)
1866 				break;
1867 		}
1868 	}
1869 	return (error);
1870 }
1871 
1872 /*
1873  * Internal form of close.
1874  * Decrement reference count on file structure.
1875  * Note: td may be NULL when closing a file that was being passed in a
1876  * message.
1877  *
1878  * XXXRW: Giant is not required for the caller, but often will be held; this
1879  * makes it moderately likely the Giant will be recursed in the VFS case.
1880  */
1881 int
1882 closef(struct file *fp, struct thread *td)
1883 {
1884 	struct vnode *vp;
1885 	struct flock lf;
1886 	struct filedesc_to_leader *fdtol;
1887 	struct filedesc *fdp;
1888 
1889 	/*
1890 	 * POSIX record locking dictates that any close releases ALL
1891 	 * locks owned by this process.  This is handled by setting
1892 	 * a flag in the unlock to free ONLY locks obeying POSIX
1893 	 * semantics, and not to free BSD-style file locks.
1894 	 * If the descriptor was in a message, POSIX-style locks
1895 	 * aren't passed with the descriptor, and the thread pointer
1896 	 * will be NULL.  Callers should be careful only to pass a
1897 	 * NULL thread pointer when there really is no owning
1898 	 * context that might have locks, or the locks will be
1899 	 * leaked.
1900 	 */
1901 	if (fp->f_type == DTYPE_VNODE && td != NULL) {
1902 		int vfslocked;
1903 
1904 		vp = fp->f_vnode;
1905 		vfslocked = VFS_LOCK_GIANT(vp->v_mount);
1906 		if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
1907 			lf.l_whence = SEEK_SET;
1908 			lf.l_start = 0;
1909 			lf.l_len = 0;
1910 			lf.l_type = F_UNLCK;
1911 			(void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader,
1912 					   F_UNLCK, &lf, F_POSIX);
1913 		}
1914 		fdtol = td->td_proc->p_fdtol;
1915 		if (fdtol != NULL) {
1916 			/*
1917 			 * Handle special case where file descriptor table
1918 			 * is shared between multiple process leaders.
1919 			 */
1920 			fdp = td->td_proc->p_fd;
1921 			FILEDESC_LOCK(fdp);
1922 			for (fdtol = fdtol->fdl_next;
1923 			     fdtol != td->td_proc->p_fdtol;
1924 			     fdtol = fdtol->fdl_next) {
1925 				if ((fdtol->fdl_leader->p_flag &
1926 				     P_ADVLOCK) == 0)
1927 					continue;
1928 				fdtol->fdl_holdcount++;
1929 				FILEDESC_UNLOCK(fdp);
1930 				lf.l_whence = SEEK_SET;
1931 				lf.l_start = 0;
1932 				lf.l_len = 0;
1933 				lf.l_type = F_UNLCK;
1934 				vp = fp->f_vnode;
1935 				(void) VOP_ADVLOCK(vp,
1936 						   (caddr_t)fdtol->fdl_leader,
1937 						   F_UNLCK, &lf, F_POSIX);
1938 				FILEDESC_LOCK(fdp);
1939 				fdtol->fdl_holdcount--;
1940 				if (fdtol->fdl_holdcount == 0 &&
1941 				    fdtol->fdl_wakeup != 0) {
1942 					fdtol->fdl_wakeup = 0;
1943 					wakeup(fdtol);
1944 				}
1945 			}
1946 			FILEDESC_UNLOCK(fdp);
1947 		}
1948 		VFS_UNLOCK_GIANT(vfslocked);
1949 	}
1950 	return (fdrop(fp, td));
1951 }
1952 
1953 /*
1954  * Extract the file pointer associated with the specified descriptor for
1955  * the current user process.
1956  *
1957  * If the descriptor doesn't exist, EBADF is returned.
1958  *
1959  * If the descriptor exists but doesn't match 'flags' then
1960  * return EBADF for read attempts and EINVAL for write attempts.
1961  *
1962  * If 'hold' is set (non-zero) the file's refcount will be bumped on return.
1963  * It should be dropped with fdrop().
1964  * If it is not set, then the refcount will not be bumped however the
1965  * thread's filedesc struct will be returned locked (for fgetsock).
1966  *
1967  * If an error occured the non-zero error is returned and *fpp is set to NULL.
1968  * Otherwise *fpp is set and zero is returned.
1969  */
1970 static __inline int
1971 _fget(struct thread *td, int fd, struct file **fpp, int flags, int hold)
1972 {
1973 	struct filedesc *fdp;
1974 	struct file *fp;
1975 
1976 	*fpp = NULL;
1977 	if (td == NULL || (fdp = td->td_proc->p_fd) == NULL)
1978 		return (EBADF);
1979 	FILEDESC_LOCK(fdp);
1980 	if ((fp = fget_locked(fdp, fd)) == NULL || fp->f_ops == &badfileops) {
1981 		FILEDESC_UNLOCK(fdp);
1982 		return (EBADF);
1983 	}
1984 
1985 	/*
1986 	 * Note: FREAD failure returns EBADF to maintain backwards
1987 	 * compatibility with what routines returned before.
1988 	 *
1989 	 * Only one flag, or 0, may be specified.
1990 	 */
1991 	if (flags == FREAD && (fp->f_flag & FREAD) == 0) {
1992 		FILEDESC_UNLOCK(fdp);
1993 		return (EBADF);
1994 	}
1995 	if (flags == FWRITE && (fp->f_flag & FWRITE) == 0) {
1996 		FILEDESC_UNLOCK(fdp);
1997 		return (EINVAL);
1998 	}
1999 	if (hold) {
2000 		fhold(fp);
2001 		FILEDESC_UNLOCK(fdp);
2002 	}
2003 	*fpp = fp;
2004 	return (0);
2005 }
2006 
2007 int
2008 fget(struct thread *td, int fd, struct file **fpp)
2009 {
2010 
2011 	return(_fget(td, fd, fpp, 0, 1));
2012 }
2013 
2014 int
2015 fget_read(struct thread *td, int fd, struct file **fpp)
2016 {
2017 
2018 	return(_fget(td, fd, fpp, FREAD, 1));
2019 }
2020 
2021 int
2022 fget_write(struct thread *td, int fd, struct file **fpp)
2023 {
2024 
2025 	return(_fget(td, fd, fpp, FWRITE, 1));
2026 }
2027 
2028 /*
2029  * Like fget() but loads the underlying vnode, or returns an error if
2030  * the descriptor does not represent a vnode.  Note that pipes use vnodes
2031  * but never have VM objects.  The returned vnode will be vref()d.
2032  *
2033  * XXX: what about the unused flags ?
2034  */
2035 static __inline int
2036 _fgetvp(struct thread *td, int fd, struct vnode **vpp, int flags)
2037 {
2038 	struct file *fp;
2039 	int error;
2040 
2041 	*vpp = NULL;
2042 	if ((error = _fget(td, fd, &fp, 0, 0)) != 0)
2043 		return (error);
2044 	if (fp->f_vnode == NULL) {
2045 		error = EINVAL;
2046 	} else {
2047 		*vpp = fp->f_vnode;
2048 		vref(*vpp);
2049 	}
2050 	FILEDESC_UNLOCK(td->td_proc->p_fd);
2051 	return (error);
2052 }
2053 
2054 int
2055 fgetvp(struct thread *td, int fd, struct vnode **vpp)
2056 {
2057 
2058 	return (_fgetvp(td, fd, vpp, 0));
2059 }
2060 
2061 int
2062 fgetvp_read(struct thread *td, int fd, struct vnode **vpp)
2063 {
2064 
2065 	return (_fgetvp(td, fd, vpp, FREAD));
2066 }
2067 
2068 #ifdef notyet
2069 int
2070 fgetvp_write(struct thread *td, int fd, struct vnode **vpp)
2071 {
2072 
2073 	return (_fgetvp(td, fd, vpp, FWRITE));
2074 }
2075 #endif
2076 
2077 /*
2078  * Like fget() but loads the underlying socket, or returns an error if
2079  * the descriptor does not represent a socket.
2080  *
2081  * We bump the ref count on the returned socket.  XXX Also obtain the SX
2082  * lock in the future.
2083  */
2084 int
2085 fgetsock(struct thread *td, int fd, struct socket **spp, u_int *fflagp)
2086 {
2087 	struct file *fp;
2088 	int error;
2089 
2090 	NET_ASSERT_GIANT();
2091 
2092 	*spp = NULL;
2093 	if (fflagp != NULL)
2094 		*fflagp = 0;
2095 	if ((error = _fget(td, fd, &fp, 0, 0)) != 0)
2096 		return (error);
2097 	if (fp->f_type != DTYPE_SOCKET) {
2098 		error = ENOTSOCK;
2099 	} else {
2100 		*spp = fp->f_data;
2101 		if (fflagp)
2102 			*fflagp = fp->f_flag;
2103 		SOCK_LOCK(*spp);
2104 		soref(*spp);
2105 		SOCK_UNLOCK(*spp);
2106 	}
2107 	FILEDESC_UNLOCK(td->td_proc->p_fd);
2108 	return (error);
2109 }
2110 
2111 /*
2112  * Drop the reference count on the socket and XXX release the SX lock in
2113  * the future.  The last reference closes the socket.
2114  */
2115 void
2116 fputsock(struct socket *so)
2117 {
2118 
2119 	NET_ASSERT_GIANT();
2120 	ACCEPT_LOCK();
2121 	SOCK_LOCK(so);
2122 	sorele(so);
2123 }
2124 
2125 int
2126 fdrop(struct file *fp, struct thread *td)
2127 {
2128 
2129 	FILE_LOCK(fp);
2130 	return (fdrop_locked(fp, td));
2131 }
2132 
2133 /*
2134  * Drop reference on struct file passed in, may call closef if the
2135  * reference hits zero.
2136  * Expects struct file locked, and will unlock it.
2137  */
2138 static int
2139 fdrop_locked(struct file *fp, struct thread *td)
2140 {
2141 	int error;
2142 
2143 	FILE_LOCK_ASSERT(fp, MA_OWNED);
2144 
2145 	if (--fp->f_count > 0) {
2146 		FILE_UNLOCK(fp);
2147 		return (0);
2148 	}
2149 	/* We have the last ref so we can proceed without the file lock. */
2150 	FILE_UNLOCK(fp);
2151 	if (fp->f_count < 0)
2152 		panic("fdrop: count < 0");
2153 	if (fp->f_ops != &badfileops)
2154 		error = fo_close(fp, td);
2155 	else
2156 		error = 0;
2157 
2158 	sx_xlock(&filelist_lock);
2159 	LIST_REMOVE(fp, f_list);
2160 	openfiles--;
2161 	sx_xunlock(&filelist_lock);
2162 	crfree(fp->f_cred);
2163 	uma_zfree(file_zone, fp);
2164 
2165 	return (error);
2166 }
2167 
2168 /*
2169  * Apply an advisory lock on a file descriptor.
2170  *
2171  * Just attempt to get a record lock of the requested type on
2172  * the entire file (l_whence = SEEK_SET, l_start = 0, l_len = 0).
2173  */
2174 #ifndef _SYS_SYSPROTO_H_
2175 struct flock_args {
2176 	int	fd;
2177 	int	how;
2178 };
2179 #endif
2180 /*
2181  * MPSAFE
2182  */
2183 /* ARGSUSED */
2184 int
2185 flock(struct thread *td, struct flock_args *uap)
2186 {
2187 	struct file *fp;
2188 	struct vnode *vp;
2189 	struct flock lf;
2190 	int error;
2191 
2192 	if ((error = fget(td, uap->fd, &fp)) != 0)
2193 		return (error);
2194 	if (fp->f_type != DTYPE_VNODE) {
2195 		fdrop(fp, td);
2196 		return (EOPNOTSUPP);
2197 	}
2198 
2199 	mtx_lock(&Giant);
2200 	vp = fp->f_vnode;
2201 	lf.l_whence = SEEK_SET;
2202 	lf.l_start = 0;
2203 	lf.l_len = 0;
2204 	if (uap->how & LOCK_UN) {
2205 		lf.l_type = F_UNLCK;
2206 		FILE_LOCK(fp);
2207 		fp->f_flag &= ~FHASLOCK;
2208 		FILE_UNLOCK(fp);
2209 		error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK);
2210 		goto done2;
2211 	}
2212 	if (uap->how & LOCK_EX)
2213 		lf.l_type = F_WRLCK;
2214 	else if (uap->how & LOCK_SH)
2215 		lf.l_type = F_RDLCK;
2216 	else {
2217 		error = EBADF;
2218 		goto done2;
2219 	}
2220 	FILE_LOCK(fp);
2221 	fp->f_flag |= FHASLOCK;
2222 	FILE_UNLOCK(fp);
2223 	error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf,
2224 	    (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT);
2225 done2:
2226 	fdrop(fp, td);
2227 	mtx_unlock(&Giant);
2228 	return (error);
2229 }
2230 /*
2231  * Duplicate the specified descriptor to a free descriptor.
2232  */
2233 int
2234 dupfdopen(struct thread *td, struct filedesc *fdp, int indx, int dfd, int mode, int error)
2235 {
2236 	struct file *wfp;
2237 	struct file *fp;
2238 
2239 	/*
2240 	 * If the to-be-dup'd fd number is greater than the allowed number
2241 	 * of file descriptors, or the fd to be dup'd has already been
2242 	 * closed, then reject.
2243 	 */
2244 	FILEDESC_LOCK(fdp);
2245 	if (dfd < 0 || dfd >= fdp->fd_nfiles ||
2246 	    (wfp = fdp->fd_ofiles[dfd]) == NULL) {
2247 		FILEDESC_UNLOCK(fdp);
2248 		return (EBADF);
2249 	}
2250 
2251 	/*
2252 	 * There are two cases of interest here.
2253 	 *
2254 	 * For ENODEV simply dup (dfd) to file descriptor
2255 	 * (indx) and return.
2256 	 *
2257 	 * For ENXIO steal away the file structure from (dfd) and
2258 	 * store it in (indx).  (dfd) is effectively closed by
2259 	 * this operation.
2260 	 *
2261 	 * Any other error code is just returned.
2262 	 */
2263 	switch (error) {
2264 	case ENODEV:
2265 		/*
2266 		 * Check that the mode the file is being opened for is a
2267 		 * subset of the mode of the existing descriptor.
2268 		 */
2269 		FILE_LOCK(wfp);
2270 		if (((mode & (FREAD|FWRITE)) | wfp->f_flag) != wfp->f_flag) {
2271 			FILE_UNLOCK(wfp);
2272 			FILEDESC_UNLOCK(fdp);
2273 			return (EACCES);
2274 		}
2275 		fp = fdp->fd_ofiles[indx];
2276 		fdp->fd_ofiles[indx] = wfp;
2277 		fdp->fd_ofileflags[indx] = fdp->fd_ofileflags[dfd];
2278 		if (fp == NULL)
2279 			fdused(fdp, indx);
2280 		fhold_locked(wfp);
2281 		FILE_UNLOCK(wfp);
2282 		FILEDESC_UNLOCK(fdp);
2283 		if (fp != NULL) {
2284 			/*
2285 			 * We now own the reference to fp that the ofiles[]
2286 			 * array used to own.  Release it.
2287 			 */
2288 			FILE_LOCK(fp);
2289 			fdrop_locked(fp, td);
2290 		}
2291 		return (0);
2292 
2293 	case ENXIO:
2294 		/*
2295 		 * Steal away the file pointer from dfd and stuff it into indx.
2296 		 */
2297 		fp = fdp->fd_ofiles[indx];
2298 		fdp->fd_ofiles[indx] = fdp->fd_ofiles[dfd];
2299 		fdp->fd_ofiles[dfd] = NULL;
2300 		fdp->fd_ofileflags[indx] = fdp->fd_ofileflags[dfd];
2301 		fdp->fd_ofileflags[dfd] = 0;
2302 		fdunused(fdp, dfd);
2303 		if (fp == NULL)
2304 			fdused(fdp, indx);
2305 		if (fp != NULL)
2306 			FILE_LOCK(fp);
2307 
2308 		/*
2309 		 * We now own the reference to fp that the ofiles[] array
2310 		 * used to own.  Release it.
2311 		 */
2312 		if (fp != NULL)
2313 			fdrop_locked(fp, td);
2314 
2315 		FILEDESC_UNLOCK(fdp);
2316 
2317 		return (0);
2318 
2319 	default:
2320 		FILEDESC_UNLOCK(fdp);
2321 		return (error);
2322 	}
2323 	/* NOTREACHED */
2324 }
2325 
2326 /*
2327  * Scan all active processes to see if any of them have a current
2328  * or root directory of `olddp'. If so, replace them with the new
2329  * mount point.
2330  */
2331 void
2332 mountcheckdirs(struct vnode *olddp, struct vnode *newdp)
2333 {
2334 	struct filedesc *fdp;
2335 	struct proc *p;
2336 	int nrele;
2337 
2338 	if (vrefcnt(olddp) == 1)
2339 		return;
2340 	sx_slock(&allproc_lock);
2341 	LIST_FOREACH(p, &allproc, p_list) {
2342 		fdp = fdhold(p);
2343 		if (fdp == NULL)
2344 			continue;
2345 		nrele = 0;
2346 		FILEDESC_LOCK_FAST(fdp);
2347 		if (fdp->fd_cdir == olddp) {
2348 			vref(newdp);
2349 			fdp->fd_cdir = newdp;
2350 			nrele++;
2351 		}
2352 		if (fdp->fd_rdir == olddp) {
2353 			vref(newdp);
2354 			fdp->fd_rdir = newdp;
2355 			nrele++;
2356 		}
2357 		FILEDESC_UNLOCK_FAST(fdp);
2358 		fddrop(fdp);
2359 		while (nrele--)
2360 			vrele(olddp);
2361 	}
2362 	sx_sunlock(&allproc_lock);
2363 	if (rootvnode == olddp) {
2364 		vrele(rootvnode);
2365 		vref(newdp);
2366 		rootvnode = newdp;
2367 	}
2368 }
2369 
2370 struct filedesc_to_leader *
2371 filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader)
2372 {
2373 	struct filedesc_to_leader *fdtol;
2374 
2375 	MALLOC(fdtol, struct filedesc_to_leader *,
2376 	       sizeof(struct filedesc_to_leader),
2377 	       M_FILEDESC_TO_LEADER,
2378 	       M_WAITOK);
2379 	fdtol->fdl_refcount = 1;
2380 	fdtol->fdl_holdcount = 0;
2381 	fdtol->fdl_wakeup = 0;
2382 	fdtol->fdl_leader = leader;
2383 	if (old != NULL) {
2384 		FILEDESC_LOCK(fdp);
2385 		fdtol->fdl_next = old->fdl_next;
2386 		fdtol->fdl_prev = old;
2387 		old->fdl_next = fdtol;
2388 		fdtol->fdl_next->fdl_prev = fdtol;
2389 		FILEDESC_UNLOCK(fdp);
2390 	} else {
2391 		fdtol->fdl_next = fdtol;
2392 		fdtol->fdl_prev = fdtol;
2393 	}
2394 	return (fdtol);
2395 }
2396 
2397 /*
2398  * Get file structures.
2399  */
2400 static int
2401 sysctl_kern_file(SYSCTL_HANDLER_ARGS)
2402 {
2403 	struct xfile xf;
2404 	struct filedesc *fdp;
2405 	struct file *fp;
2406 	struct proc *p;
2407 	int error, n;
2408 
2409 	/*
2410 	 * Note: because the number of file descriptors is calculated
2411 	 * in different ways for sizing vs returning the data,
2412 	 * there is information leakage from the first loop.  However,
2413 	 * it is of a similar order of magnitude to the leakage from
2414 	 * global system statistics such as kern.openfiles.
2415 	 */
2416 	error = sysctl_wire_old_buffer(req, 0);
2417 	if (error != 0)
2418 		return (error);
2419 	if (req->oldptr == NULL) {
2420 		n = 16;		/* A slight overestimate. */
2421 		sx_slock(&filelist_lock);
2422 		LIST_FOREACH(fp, &filehead, f_list) {
2423 			/*
2424 			 * We should grab the lock, but this is an
2425 			 * estimate, so does it really matter?
2426 			 */
2427 			/* mtx_lock(fp->f_mtxp); */
2428 			n += fp->f_count;
2429 			/* mtx_unlock(f->f_mtxp); */
2430 		}
2431 		sx_sunlock(&filelist_lock);
2432 		return (SYSCTL_OUT(req, 0, n * sizeof(xf)));
2433 	}
2434 	error = 0;
2435 	bzero(&xf, sizeof(xf));
2436 	xf.xf_size = sizeof(xf);
2437 	sx_slock(&allproc_lock);
2438 	LIST_FOREACH(p, &allproc, p_list) {
2439 		if (p->p_state == PRS_NEW)
2440 			continue;
2441 		PROC_LOCK(p);
2442 		if (p_cansee(req->td, p) != 0) {
2443 			PROC_UNLOCK(p);
2444 			continue;
2445 		}
2446 		xf.xf_pid = p->p_pid;
2447 		xf.xf_uid = p->p_ucred->cr_uid;
2448 		PROC_UNLOCK(p);
2449 		fdp = fdhold(p);
2450 		if (fdp == NULL)
2451 			continue;
2452 		FILEDESC_LOCK_FAST(fdp);
2453 		for (n = 0; fdp->fd_refcnt > 0 && n < fdp->fd_nfiles; ++n) {
2454 			if ((fp = fdp->fd_ofiles[n]) == NULL)
2455 				continue;
2456 			xf.xf_fd = n;
2457 			xf.xf_file = fp;
2458 			xf.xf_data = fp->f_data;
2459 			xf.xf_vnode = fp->f_vnode;
2460 			xf.xf_type = fp->f_type;
2461 			xf.xf_count = fp->f_count;
2462 			xf.xf_msgcount = fp->f_msgcount;
2463 			xf.xf_offset = fp->f_offset;
2464 			xf.xf_flag = fp->f_flag;
2465 			error = SYSCTL_OUT(req, &xf, sizeof(xf));
2466 			if (error)
2467 				break;
2468 		}
2469 		FILEDESC_UNLOCK_FAST(fdp);
2470 		fddrop(fdp);
2471 		if (error)
2472 			break;
2473 	}
2474 	sx_sunlock(&allproc_lock);
2475 	return (error);
2476 }
2477 
2478 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD,
2479     0, 0, sysctl_kern_file, "S,xfile", "Entire file table");
2480 
2481 #ifdef DDB
2482 /*
2483  * For the purposes of debugging, generate a human-readable string for the
2484  * file type.
2485  */
2486 static const char *
2487 file_type_to_name(short type)
2488 {
2489 
2490 	switch (type) {
2491 	case 0:
2492 		return ("zero");
2493 	case DTYPE_VNODE:
2494 		return ("vnod");
2495 	case DTYPE_SOCKET:
2496 		return ("sock");
2497 	case DTYPE_PIPE:
2498 		return ("pipe");
2499 	case DTYPE_FIFO:
2500 		return ("fifo");
2501 	case DTYPE_CRYPTO:
2502 		return ("crpt");
2503 	default:
2504 		return ("unkn");
2505 	}
2506 }
2507 
2508 /*
2509  * For the purposes of debugging, identify a process (if any, perhaps one of
2510  * many) that references the passed file in its file descriptor array. Return
2511  * NULL if none.
2512  */
2513 static struct proc *
2514 file_to_first_proc(struct file *fp)
2515 {
2516 	struct filedesc *fdp;
2517 	struct proc *p;
2518 	int n;
2519 
2520 	LIST_FOREACH(p, &allproc, p_list) {
2521 		if (p->p_state == PRS_NEW)
2522 			continue;
2523 		fdp = p->p_fd;
2524 		if (fdp == NULL)
2525 			continue;
2526 		for (n = 0; n < fdp->fd_nfiles; n++) {
2527 			if (fp == fdp->fd_ofiles[n])
2528 				return (p);
2529 		}
2530 	}
2531 	return (NULL);
2532 }
2533 
2534 DB_SHOW_COMMAND(files, db_show_files)
2535 {
2536 	struct file *fp;
2537 	struct proc *p;
2538 
2539 	db_printf("%8s %4s %8s %8s %4s %5s %6s %8s %5s %12s\n", "File",
2540 	    "Type", "Data", "Flag", "GCFl", "Count", "MCount", "Vnode",
2541 	    "FPID", "FCmd");
2542 	LIST_FOREACH(fp, &filehead, f_list) {
2543 		p = file_to_first_proc(fp);
2544 		db_printf("%8p %4s %8p %08x %04x %5d %6d %8p %5d %12s\n", fp,
2545 		    file_type_to_name(fp->f_type), fp->f_data, fp->f_flag,
2546 		    fp->f_gcflag, fp->f_count, fp->f_msgcount, fp->f_vnode,
2547 		    p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-");
2548 	}
2549 }
2550 #endif
2551 
2552 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
2553     &maxfilesperproc, 0, "Maximum files allowed open per process");
2554 
2555 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
2556     &maxfiles, 0, "Maximum number of files");
2557 
2558 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
2559     &openfiles, 0, "System-wide number of open files");
2560 
2561 /* ARGSUSED*/
2562 static void
2563 filelistinit(void *dummy)
2564 {
2565 
2566 	file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL,
2567 	    NULL, NULL, UMA_ALIGN_PTR, 0);
2568 	sx_init(&filelist_lock, "filelist lock");
2569 	mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF);
2570 	mtx_init(&fdesc_mtx, "fdesc", NULL, MTX_DEF);
2571 }
2572 SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL)
2573 
2574 /*-------------------------------------------------------------------*/
2575 
2576 static int
2577 badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred, int flags, struct thread *td)
2578 {
2579 
2580 	return (EBADF);
2581 }
2582 
2583 static int
2584 badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred, struct thread *td)
2585 {
2586 
2587 	return (EBADF);
2588 }
2589 
2590 static int
2591 badfo_poll(struct file *fp, int events, struct ucred *active_cred, struct thread *td)
2592 {
2593 
2594 	return (0);
2595 }
2596 
2597 static int
2598 badfo_kqfilter(struct file *fp, struct knote *kn)
2599 {
2600 
2601 	return (0);
2602 }
2603 
2604 static int
2605 badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred, struct thread *td)
2606 {
2607 
2608 	return (EBADF);
2609 }
2610 
2611 static int
2612 badfo_close(struct file *fp, struct thread *td)
2613 {
2614 
2615 	return (EBADF);
2616 }
2617 
2618 struct fileops badfileops = {
2619 	.fo_read = badfo_readwrite,
2620 	.fo_write = badfo_readwrite,
2621 	.fo_ioctl = badfo_ioctl,
2622 	.fo_poll = badfo_poll,
2623 	.fo_kqfilter = badfo_kqfilter,
2624 	.fo_stat = badfo_stat,
2625 	.fo_close = badfo_close,
2626 };
2627 
2628 
2629 /*-------------------------------------------------------------------*/
2630 
2631 /*
2632  * File Descriptor pseudo-device driver (/dev/fd/).
2633  *
2634  * Opening minor device N dup()s the file (if any) connected to file
2635  * descriptor N belonging to the calling process.  Note that this driver
2636  * consists of only the ``open()'' routine, because all subsequent
2637  * references to this file will be direct to the other driver.
2638  *
2639  * XXX: we could give this one a cloning event handler if necessary.
2640  */
2641 
2642 /* ARGSUSED */
2643 static int
2644 fdopen(struct cdev *dev, int mode, int type, struct thread *td)
2645 {
2646 
2647 	/*
2648 	 * XXX Kludge: set curthread->td_dupfd to contain the value of the
2649 	 * the file descriptor being sought for duplication. The error
2650 	 * return ensures that the vnode for this device will be released
2651 	 * by vn_open. Open will detect this special error and take the
2652 	 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
2653 	 * will simply report the error.
2654 	 */
2655 	td->td_dupfd = dev2unit(dev);
2656 	return (ENODEV);
2657 }
2658 
2659 static struct cdevsw fildesc_cdevsw = {
2660 	.d_version =	D_VERSION,
2661 	.d_flags =	D_NEEDGIANT,
2662 	.d_open =	fdopen,
2663 	.d_name =	"FD",
2664 };
2665 
2666 static void
2667 fildesc_drvinit(void *unused)
2668 {
2669 	struct cdev *dev;
2670 
2671 	dev = make_dev(&fildesc_cdevsw, 0, UID_ROOT, GID_WHEEL, 0666, "fd/0");
2672 	make_dev_alias(dev, "stdin");
2673 	dev = make_dev(&fildesc_cdevsw, 1, UID_ROOT, GID_WHEEL, 0666, "fd/1");
2674 	make_dev_alias(dev, "stdout");
2675 	dev = make_dev(&fildesc_cdevsw, 2, UID_ROOT, GID_WHEEL, 0666, "fd/2");
2676 	make_dev_alias(dev, "stderr");
2677 }
2678 
2679 SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL)
2680