xref: /freebsd/sys/kern/kern_descrip.c (revision d056fa046c6a91b90cd98165face0e42a33a5173)
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 <security/audit/audit.h>
73 
74 #include <vm/uma.h>
75 
76 #include <ddb/ddb.h>
77 
78 static MALLOC_DEFINE(M_FILEDESC, "filedesc", "Open file descriptor table");
79 static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "filedesc_to_leader",
80 		     "file desc to leader structures");
81 static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures");
82 
83 static uma_zone_t file_zone;
84 
85 
86 /* How to treat 'new' parameter when allocating a fd for do_dup(). */
87 enum dup_type { DUP_VARIABLE, DUP_FIXED };
88 
89 static int do_dup(struct thread *td, enum dup_type type, int old, int new,
90     register_t *retval);
91 static int	fd_first_free(struct filedesc *, int, int);
92 static int	fd_last_used(struct filedesc *, int, int);
93 static void	fdgrowtable(struct filedesc *, int);
94 static int	fdrop_locked(struct file *fp, struct thread *td);
95 static void	fdunused(struct filedesc *fdp, int fd);
96 static void	fdused(struct filedesc *fdp, int fd);
97 
98 /*
99  * A process is initially started out with NDFILE descriptors stored within
100  * this structure, selected to be enough for typical applications based on
101  * the historical limit of 20 open files (and the usage of descriptors by
102  * shells).  If these descriptors are exhausted, a larger descriptor table
103  * may be allocated, up to a process' resource limit; the internal arrays
104  * are then unused.
105  */
106 #define NDFILE		20
107 #define NDSLOTSIZE	sizeof(NDSLOTTYPE)
108 #define	NDENTRIES	(NDSLOTSIZE * __CHAR_BIT)
109 #define NDSLOT(x)	((x) / NDENTRIES)
110 #define NDBIT(x)	((NDSLOTTYPE)1 << ((x) % NDENTRIES))
111 #define	NDSLOTS(x)	(((x) + NDENTRIES - 1) / NDENTRIES)
112 
113 /*
114  * Storage required per open file descriptor.
115  */
116 #define OFILESIZE (sizeof(struct file *) + sizeof(char))
117 
118 /*
119  * Basic allocation of descriptors:
120  * one of the above, plus arrays for NDFILE descriptors.
121  */
122 struct filedesc0 {
123 	struct	filedesc fd_fd;
124 	/*
125 	 * These arrays are used when the number of open files is
126 	 * <= NDFILE, and are then pointed to by the pointers above.
127 	 */
128 	struct	file *fd_dfiles[NDFILE];
129 	char	fd_dfileflags[NDFILE];
130 	NDSLOTTYPE fd_dmap[NDSLOTS(NDFILE)];
131 };
132 
133 /*
134  * Descriptor management.
135  */
136 struct filelist filehead;	/* head of list of open files */
137 int openfiles;			/* actual number of open files */
138 struct sx filelist_lock;	/* sx to protect filelist */
139 struct mtx sigio_lock;		/* mtx to protect pointers to sigio */
140 void	(*mq_fdclose)(struct thread *td, int fd, struct file *fp);
141 
142 /* A mutex to protect the association between a proc and filedesc. */
143 static struct mtx	fdesc_mtx;
144 
145 /*
146  * Find the first zero bit in the given bitmap, starting at low and not
147  * exceeding size - 1.
148  */
149 static int
150 fd_first_free(struct filedesc *fdp, int low, int size)
151 {
152 	NDSLOTTYPE *map = fdp->fd_map;
153 	NDSLOTTYPE mask;
154 	int off, maxoff;
155 
156 	if (low >= size)
157 		return (low);
158 
159 	off = NDSLOT(low);
160 	if (low % NDENTRIES) {
161 		mask = ~(~(NDSLOTTYPE)0 >> (NDENTRIES - (low % NDENTRIES)));
162 		if ((mask &= ~map[off]) != 0UL)
163 			return (off * NDENTRIES + ffsl(mask) - 1);
164 		++off;
165 	}
166 	for (maxoff = NDSLOTS(size); off < maxoff; ++off)
167 		if (map[off] != ~0UL)
168 			return (off * NDENTRIES + ffsl(~map[off]) - 1);
169 	return (size);
170 }
171 
172 /*
173  * Find the highest non-zero bit in the given bitmap, starting at low and
174  * not exceeding size - 1.
175  */
176 static int
177 fd_last_used(struct filedesc *fdp, int low, int size)
178 {
179 	NDSLOTTYPE *map = fdp->fd_map;
180 	NDSLOTTYPE mask;
181 	int off, minoff;
182 
183 	if (low >= size)
184 		return (-1);
185 
186 	off = NDSLOT(size);
187 	if (size % NDENTRIES) {
188 		mask = ~(~(NDSLOTTYPE)0 << (size % NDENTRIES));
189 		if ((mask &= map[off]) != 0)
190 			return (off * NDENTRIES + flsl(mask) - 1);
191 		--off;
192 	}
193 	for (minoff = NDSLOT(low); off >= minoff; --off)
194 		if (map[off] != 0)
195 			return (off * NDENTRIES + flsl(map[off]) - 1);
196 	return (low - 1);
197 }
198 
199 static int
200 fdisused(struct filedesc *fdp, int fd)
201 {
202         KASSERT(fd >= 0 && fd < fdp->fd_nfiles,
203             ("file descriptor %d out of range (0, %d)", fd, fdp->fd_nfiles));
204 	return ((fdp->fd_map[NDSLOT(fd)] & NDBIT(fd)) != 0);
205 }
206 
207 /*
208  * Mark a file descriptor as used.
209  */
210 static void
211 fdused(struct filedesc *fdp, int fd)
212 {
213 	FILEDESC_LOCK_ASSERT(fdp, MA_OWNED);
214 	KASSERT(!fdisused(fdp, fd),
215 	    ("fd already used"));
216 	fdp->fd_map[NDSLOT(fd)] |= NDBIT(fd);
217 	if (fd > fdp->fd_lastfile)
218 		fdp->fd_lastfile = fd;
219 	if (fd == fdp->fd_freefile)
220 		fdp->fd_freefile = fd_first_free(fdp, fd, fdp->fd_nfiles);
221 }
222 
223 /*
224  * Mark a file descriptor as unused.
225  */
226 static void
227 fdunused(struct filedesc *fdp, int fd)
228 {
229 	FILEDESC_LOCK_ASSERT(fdp, MA_OWNED);
230 	KASSERT(fdisused(fdp, fd),
231 	    ("fd is already unused"));
232 	KASSERT(fdp->fd_ofiles[fd] == NULL,
233 	    ("fd is still in use"));
234 	fdp->fd_map[NDSLOT(fd)] &= ~NDBIT(fd);
235 	if (fd < fdp->fd_freefile)
236 		fdp->fd_freefile = fd;
237 	if (fd == fdp->fd_lastfile)
238 		fdp->fd_lastfile = fd_last_used(fdp, 0, fd);
239 }
240 
241 /*
242  * System calls on descriptors.
243  */
244 #ifndef _SYS_SYSPROTO_H_
245 struct getdtablesize_args {
246 	int	dummy;
247 };
248 #endif
249 /*
250  * MPSAFE
251  */
252 /* ARGSUSED */
253 int
254 getdtablesize(struct thread *td, struct getdtablesize_args *uap)
255 {
256 	struct proc *p = td->td_proc;
257 
258 	PROC_LOCK(p);
259 	td->td_retval[0] =
260 	    min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc);
261 	PROC_UNLOCK(p);
262 	return (0);
263 }
264 
265 /*
266  * Duplicate a file descriptor to a particular value.
267  *
268  * note: keep in mind that a potential race condition exists when closing
269  * descriptors from a shared descriptor table (via rfork).
270  */
271 #ifndef _SYS_SYSPROTO_H_
272 struct dup2_args {
273 	u_int	from;
274 	u_int	to;
275 };
276 #endif
277 /*
278  * MPSAFE
279  */
280 /* ARGSUSED */
281 int
282 dup2(struct thread *td, struct dup2_args *uap)
283 {
284 
285 	return (do_dup(td, DUP_FIXED, (int)uap->from, (int)uap->to,
286 		    td->td_retval));
287 }
288 
289 /*
290  * Duplicate a file descriptor.
291  */
292 #ifndef _SYS_SYSPROTO_H_
293 struct dup_args {
294 	u_int	fd;
295 };
296 #endif
297 /*
298  * MPSAFE
299  */
300 /* ARGSUSED */
301 int
302 dup(struct thread *td, struct dup_args *uap)
303 {
304 
305 	return (do_dup(td, DUP_VARIABLE, (int)uap->fd, 0, td->td_retval));
306 }
307 
308 /*
309  * The file control system call.
310  */
311 #ifndef _SYS_SYSPROTO_H_
312 struct fcntl_args {
313 	int	fd;
314 	int	cmd;
315 	long	arg;
316 };
317 #endif
318 /*
319  * MPSAFE
320  */
321 /* ARGSUSED */
322 int
323 fcntl(struct thread *td, struct fcntl_args *uap)
324 {
325 	struct flock fl;
326 	intptr_t arg;
327 	int error;
328 
329 	error = 0;
330 	switch (uap->cmd) {
331 	case F_GETLK:
332 	case F_SETLK:
333 	case F_SETLKW:
334 		error = copyin((void *)(intptr_t)uap->arg, &fl, sizeof(fl));
335 		arg = (intptr_t)&fl;
336 		break;
337 	default:
338 		arg = uap->arg;
339 		break;
340 	}
341 	if (error)
342 		return (error);
343 	error = kern_fcntl(td, uap->fd, uap->cmd, arg);
344 	if (error)
345 		return (error);
346 	if (uap->cmd == F_GETLK)
347 		error = copyout(&fl, (void *)(intptr_t)uap->arg, sizeof(fl));
348 	return (error);
349 }
350 
351 int
352 kern_fcntl(struct thread *td, int fd, int cmd, intptr_t arg)
353 {
354 	struct filedesc *fdp;
355 	struct flock *flp;
356 	struct file *fp;
357 	struct proc *p;
358 	char *pop;
359 	struct vnode *vp;
360 	u_int newmin;
361 	int error, flg, tmp;
362 	int giant_locked;
363 
364 	/*
365 	 * XXXRW: Some fcntl() calls require Giant -- others don't.  Try to
366 	 * avoid grabbing Giant for calls we know don't need it.
367 	 */
368 	switch (cmd) {
369 	case F_DUPFD:
370 	case F_GETFD:
371 	case F_SETFD:
372 	case F_GETFL:
373 		giant_locked = 0;
374 		break;
375 
376 	default:
377 		giant_locked = 1;
378 		mtx_lock(&Giant);
379 	}
380 
381 	error = 0;
382 	flg = F_POSIX;
383 	p = td->td_proc;
384 	fdp = p->p_fd;
385 	FILEDESC_LOCK(fdp);
386 	if ((unsigned)fd >= fdp->fd_nfiles ||
387 	    (fp = fdp->fd_ofiles[fd]) == NULL) {
388 		FILEDESC_UNLOCK(fdp);
389 		error = EBADF;
390 		goto done2;
391 	}
392 	pop = &fdp->fd_ofileflags[fd];
393 
394 	switch (cmd) {
395 	case F_DUPFD:
396 		/* mtx_assert(&Giant, MA_NOTOWNED); */
397 		FILEDESC_UNLOCK(fdp);
398 		newmin = arg;
399 		PROC_LOCK(p);
400 		if (newmin >= lim_cur(p, RLIMIT_NOFILE) ||
401 		    newmin >= maxfilesperproc) {
402 			PROC_UNLOCK(p);
403 			error = EINVAL;
404 			break;
405 		}
406 		PROC_UNLOCK(p);
407 		error = do_dup(td, DUP_VARIABLE, fd, newmin, td->td_retval);
408 		break;
409 
410 	case F_GETFD:
411 		/* mtx_assert(&Giant, MA_NOTOWNED); */
412 		td->td_retval[0] = (*pop & UF_EXCLOSE) ? FD_CLOEXEC : 0;
413 		FILEDESC_UNLOCK(fdp);
414 		break;
415 
416 	case F_SETFD:
417 		/* mtx_assert(&Giant, MA_NOTOWNED); */
418 		*pop = (*pop &~ UF_EXCLOSE) |
419 		    (arg & FD_CLOEXEC ? UF_EXCLOSE : 0);
420 		FILEDESC_UNLOCK(fdp);
421 		break;
422 
423 	case F_GETFL:
424 		/* mtx_assert(&Giant, MA_NOTOWNED); */
425 		FILE_LOCK(fp);
426 		td->td_retval[0] = OFLAGS(fp->f_flag);
427 		FILE_UNLOCK(fp);
428 		FILEDESC_UNLOCK(fdp);
429 		break;
430 
431 	case F_SETFL:
432 		mtx_assert(&Giant, MA_OWNED);
433 		FILE_LOCK(fp);
434 		fhold_locked(fp);
435 		fp->f_flag &= ~FCNTLFLAGS;
436 		fp->f_flag |= FFLAGS(arg & ~O_ACCMODE) & FCNTLFLAGS;
437 		FILE_UNLOCK(fp);
438 		FILEDESC_UNLOCK(fdp);
439 		tmp = fp->f_flag & FNONBLOCK;
440 		error = fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
441 		if (error) {
442 			fdrop(fp, td);
443 			break;
444 		}
445 		tmp = fp->f_flag & FASYNC;
446 		error = fo_ioctl(fp, FIOASYNC, &tmp, td->td_ucred, td);
447 		if (error == 0) {
448 			fdrop(fp, td);
449 			break;
450 		}
451 		FILE_LOCK(fp);
452 		fp->f_flag &= ~FNONBLOCK;
453 		FILE_UNLOCK(fp);
454 		tmp = 0;
455 		(void)fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
456 		fdrop(fp, td);
457 		break;
458 
459 	case F_GETOWN:
460 		mtx_assert(&Giant, MA_OWNED);
461 		fhold(fp);
462 		FILEDESC_UNLOCK(fdp);
463 		error = fo_ioctl(fp, FIOGETOWN, &tmp, td->td_ucred, td);
464 		if (error == 0)
465 			td->td_retval[0] = tmp;
466 		fdrop(fp, td);
467 		break;
468 
469 	case F_SETOWN:
470 		mtx_assert(&Giant, MA_OWNED);
471 		fhold(fp);
472 		FILEDESC_UNLOCK(fdp);
473 		tmp = arg;
474 		error = fo_ioctl(fp, FIOSETOWN, &tmp, td->td_ucred, td);
475 		fdrop(fp, td);
476 		break;
477 
478 	case F_SETLKW:
479 		mtx_assert(&Giant, MA_OWNED);
480 		flg |= F_WAIT;
481 		/* FALLTHROUGH F_SETLK */
482 
483 	case F_SETLK:
484 		mtx_assert(&Giant, MA_OWNED);
485 		if (fp->f_type != DTYPE_VNODE) {
486 			FILEDESC_UNLOCK(fdp);
487 			error = EBADF;
488 			break;
489 		}
490 
491 		flp = (struct flock *)arg;
492 		if (flp->l_whence == SEEK_CUR) {
493 			if (fp->f_offset < 0 ||
494 			    (flp->l_start > 0 &&
495 			     fp->f_offset > OFF_MAX - flp->l_start)) {
496 				FILEDESC_UNLOCK(fdp);
497 				error = EOVERFLOW;
498 				break;
499 			}
500 			flp->l_start += fp->f_offset;
501 		}
502 
503 		/*
504 		 * VOP_ADVLOCK() may block.
505 		 */
506 		fhold(fp);
507 		FILEDESC_UNLOCK(fdp);
508 		vp = fp->f_vnode;
509 
510 		switch (flp->l_type) {
511 		case F_RDLCK:
512 			if ((fp->f_flag & FREAD) == 0) {
513 				error = EBADF;
514 				break;
515 			}
516 			PROC_LOCK(p->p_leader);
517 			p->p_leader->p_flag |= P_ADVLOCK;
518 			PROC_UNLOCK(p->p_leader);
519 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
520 			    flp, flg);
521 			break;
522 		case F_WRLCK:
523 			if ((fp->f_flag & FWRITE) == 0) {
524 				error = EBADF;
525 				break;
526 			}
527 			PROC_LOCK(p->p_leader);
528 			p->p_leader->p_flag |= P_ADVLOCK;
529 			PROC_UNLOCK(p->p_leader);
530 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
531 			    flp, flg);
532 			break;
533 		case F_UNLCK:
534 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
535 			    flp, F_POSIX);
536 			break;
537 		default:
538 			error = EINVAL;
539 			break;
540 		}
541 		/* Check for race with close */
542 		FILEDESC_LOCK_FAST(fdp);
543 		if ((unsigned) fd >= fdp->fd_nfiles ||
544 		    fp != fdp->fd_ofiles[fd]) {
545 			FILEDESC_UNLOCK_FAST(fdp);
546 			flp->l_whence = SEEK_SET;
547 			flp->l_start = 0;
548 			flp->l_len = 0;
549 			flp->l_type = F_UNLCK;
550 			(void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
551 					   F_UNLCK, flp, F_POSIX);
552 		} else
553 			FILEDESC_UNLOCK_FAST(fdp);
554 		fdrop(fp, td);
555 		break;
556 
557 	case F_GETLK:
558 		mtx_assert(&Giant, MA_OWNED);
559 		if (fp->f_type != DTYPE_VNODE) {
560 			FILEDESC_UNLOCK(fdp);
561 			error = EBADF;
562 			break;
563 		}
564 		flp = (struct flock *)arg;
565 		if (flp->l_type != F_RDLCK && flp->l_type != F_WRLCK &&
566 		    flp->l_type != F_UNLCK) {
567 			FILEDESC_UNLOCK(fdp);
568 			error = EINVAL;
569 			break;
570 		}
571 		if (flp->l_whence == SEEK_CUR) {
572 			if ((flp->l_start > 0 &&
573 			    fp->f_offset > OFF_MAX - flp->l_start) ||
574 			    (flp->l_start < 0 &&
575 			     fp->f_offset < OFF_MIN - flp->l_start)) {
576 				FILEDESC_UNLOCK(fdp);
577 				error = EOVERFLOW;
578 				break;
579 			}
580 			flp->l_start += fp->f_offset;
581 		}
582 		/*
583 		 * VOP_ADVLOCK() may block.
584 		 */
585 		fhold(fp);
586 		FILEDESC_UNLOCK(fdp);
587 		vp = fp->f_vnode;
588 		error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, flp,
589 		    F_POSIX);
590 		fdrop(fp, td);
591 		break;
592 	default:
593 		FILEDESC_UNLOCK(fdp);
594 		error = EINVAL;
595 		break;
596 	}
597 done2:
598 	if (giant_locked)
599 		mtx_unlock(&Giant);
600 	return (error);
601 }
602 
603 /*
604  * Common code for dup, dup2, and fcntl(F_DUPFD).
605  */
606 static int
607 do_dup(struct thread *td, enum dup_type type, int old, int new, register_t *retval)
608 {
609 	struct filedesc *fdp;
610 	struct proc *p;
611 	struct file *fp;
612 	struct file *delfp;
613 	int error, holdleaders, maxfd;
614 
615 	KASSERT((type == DUP_VARIABLE || type == DUP_FIXED),
616 	    ("invalid dup type %d", type));
617 
618 	p = td->td_proc;
619 	fdp = p->p_fd;
620 
621 	/*
622 	 * Verify we have a valid descriptor to dup from and possibly to
623 	 * dup to.
624 	 */
625 	if (old < 0 || new < 0)
626 		return (EBADF);
627 	PROC_LOCK(p);
628 	maxfd = min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc);
629 	PROC_UNLOCK(p);
630 	if (new >= maxfd)
631 		return (EMFILE);
632 
633 	FILEDESC_LOCK(fdp);
634 	if (old >= fdp->fd_nfiles || fdp->fd_ofiles[old] == NULL) {
635 		FILEDESC_UNLOCK(fdp);
636 		return (EBADF);
637 	}
638 	if (type == DUP_FIXED && old == new) {
639 		*retval = new;
640 		FILEDESC_UNLOCK(fdp);
641 		return (0);
642 	}
643 	fp = fdp->fd_ofiles[old];
644 	fhold(fp);
645 
646 	/*
647 	 * If the caller specified a file descriptor, make sure the file
648 	 * table is large enough to hold it, and grab it.  Otherwise, just
649 	 * allocate a new descriptor the usual way.  Since the filedesc
650 	 * lock may be temporarily dropped in the process, we have to look
651 	 * out for a race.
652 	 */
653 	if (type == DUP_FIXED) {
654 		if (new >= fdp->fd_nfiles)
655 			fdgrowtable(fdp, new + 1);
656 		if (fdp->fd_ofiles[new] == NULL)
657 			fdused(fdp, new);
658 	} else {
659 		if ((error = fdalloc(td, new, &new)) != 0) {
660 			FILEDESC_UNLOCK(fdp);
661 			fdrop(fp, td);
662 			return (error);
663 		}
664 	}
665 
666 	/*
667 	 * If the old file changed out from under us then treat it as a
668 	 * bad file descriptor.  Userland should do its own locking to
669 	 * avoid this case.
670 	 */
671 	if (fdp->fd_ofiles[old] != fp) {
672 		/* we've allocated a descriptor which we won't use */
673 		if (fdp->fd_ofiles[new] == NULL)
674 			fdunused(fdp, new);
675 		FILEDESC_UNLOCK(fdp);
676 		fdrop(fp, td);
677 		return (EBADF);
678 	}
679 	KASSERT(old != new,
680 	    ("new fd is same as old"));
681 
682 	/*
683 	 * Save info on the descriptor being overwritten.  We cannot close
684 	 * it without introducing an ownership race for the slot, since we
685 	 * need to drop the filedesc lock to call closef().
686 	 *
687 	 * XXX this duplicates parts of close().
688 	 */
689 	delfp = fdp->fd_ofiles[new];
690 	holdleaders = 0;
691 	if (delfp != NULL) {
692 		if (td->td_proc->p_fdtol != NULL) {
693 			/*
694 			 * Ask fdfree() to sleep to ensure that all relevant
695 			 * process leaders can be traversed in closef().
696 			 */
697 			fdp->fd_holdleaderscount++;
698 			holdleaders = 1;
699 		}
700 	}
701 
702 	/*
703 	 * Duplicate the source descriptor
704 	 */
705 	fdp->fd_ofiles[new] = fp;
706 	fdp->fd_ofileflags[new] = fdp->fd_ofileflags[old] &~ UF_EXCLOSE;
707 	if (new > fdp->fd_lastfile)
708 		fdp->fd_lastfile = new;
709 	*retval = new;
710 
711 	/*
712 	 * If we dup'd over a valid file, we now own the reference to it
713 	 * and must dispose of it using closef() semantics (as if a
714 	 * close() were performed on it).
715 	 *
716 	 * XXX this duplicates parts of close().
717 	 */
718 	if (delfp != NULL) {
719 		knote_fdclose(td, new);
720 		if (delfp->f_type == DTYPE_MQUEUE)
721 			mq_fdclose(td, new, delfp);
722 		FILEDESC_UNLOCK(fdp);
723 		(void) closef(delfp, td);
724 		if (holdleaders) {
725 			FILEDESC_LOCK_FAST(fdp);
726 			fdp->fd_holdleaderscount--;
727 			if (fdp->fd_holdleaderscount == 0 &&
728 			    fdp->fd_holdleaderswakeup != 0) {
729 				fdp->fd_holdleaderswakeup = 0;
730 				wakeup(&fdp->fd_holdleaderscount);
731 			}
732 			FILEDESC_UNLOCK_FAST(fdp);
733 		}
734 	} else {
735 		FILEDESC_UNLOCK(fdp);
736 	}
737 	return (0);
738 }
739 
740 /*
741  * If sigio is on the list associated with a process or process group,
742  * disable signalling from the device, remove sigio from the list and
743  * free sigio.
744  */
745 void
746 funsetown(struct sigio **sigiop)
747 {
748 	struct sigio *sigio;
749 
750 	SIGIO_LOCK();
751 	sigio = *sigiop;
752 	if (sigio == NULL) {
753 		SIGIO_UNLOCK();
754 		return;
755 	}
756 	*(sigio->sio_myref) = NULL;
757 	if ((sigio)->sio_pgid < 0) {
758 		struct pgrp *pg = (sigio)->sio_pgrp;
759 		PGRP_LOCK(pg);
760 		SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio,
761 			     sigio, sio_pgsigio);
762 		PGRP_UNLOCK(pg);
763 	} else {
764 		struct proc *p = (sigio)->sio_proc;
765 		PROC_LOCK(p);
766 		SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio,
767 			     sigio, sio_pgsigio);
768 		PROC_UNLOCK(p);
769 	}
770 	SIGIO_UNLOCK();
771 	crfree(sigio->sio_ucred);
772 	FREE(sigio, M_SIGIO);
773 }
774 
775 /*
776  * Free a list of sigio structures.
777  * We only need to lock the SIGIO_LOCK because we have made ourselves
778  * inaccessible to callers of fsetown and therefore do not need to lock
779  * the proc or pgrp struct for the list manipulation.
780  */
781 void
782 funsetownlst(struct sigiolst *sigiolst)
783 {
784 	struct proc *p;
785 	struct pgrp *pg;
786 	struct sigio *sigio;
787 
788 	sigio = SLIST_FIRST(sigiolst);
789 	if (sigio == NULL)
790 		return;
791 	p = NULL;
792 	pg = NULL;
793 
794 	/*
795 	 * Every entry of the list should belong
796 	 * to a single proc or pgrp.
797 	 */
798 	if (sigio->sio_pgid < 0) {
799 		pg = sigio->sio_pgrp;
800 		PGRP_LOCK_ASSERT(pg, MA_NOTOWNED);
801 	} else /* if (sigio->sio_pgid > 0) */ {
802 		p = sigio->sio_proc;
803 		PROC_LOCK_ASSERT(p, MA_NOTOWNED);
804 	}
805 
806 	SIGIO_LOCK();
807 	while ((sigio = SLIST_FIRST(sigiolst)) != NULL) {
808 		*(sigio->sio_myref) = NULL;
809 		if (pg != NULL) {
810 			KASSERT(sigio->sio_pgid < 0,
811 			    ("Proc sigio in pgrp sigio list"));
812 			KASSERT(sigio->sio_pgrp == pg,
813 			    ("Bogus pgrp in sigio list"));
814 			PGRP_LOCK(pg);
815 			SLIST_REMOVE(&pg->pg_sigiolst, sigio, sigio,
816 			    sio_pgsigio);
817 			PGRP_UNLOCK(pg);
818 		} else /* if (p != NULL) */ {
819 			KASSERT(sigio->sio_pgid > 0,
820 			    ("Pgrp sigio in proc sigio list"));
821 			KASSERT(sigio->sio_proc == p,
822 			    ("Bogus proc in sigio list"));
823 			PROC_LOCK(p);
824 			SLIST_REMOVE(&p->p_sigiolst, sigio, sigio,
825 			    sio_pgsigio);
826 			PROC_UNLOCK(p);
827 		}
828 		SIGIO_UNLOCK();
829 		crfree(sigio->sio_ucred);
830 		FREE(sigio, M_SIGIO);
831 		SIGIO_LOCK();
832 	}
833 	SIGIO_UNLOCK();
834 }
835 
836 /*
837  * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg).
838  *
839  * After permission checking, add a sigio structure to the sigio list for
840  * the process or process group.
841  */
842 int
843 fsetown(pid_t pgid, struct sigio **sigiop)
844 {
845 	struct proc *proc;
846 	struct pgrp *pgrp;
847 	struct sigio *sigio;
848 	int ret;
849 
850 	if (pgid == 0) {
851 		funsetown(sigiop);
852 		return (0);
853 	}
854 
855 	ret = 0;
856 
857 	/* Allocate and fill in the new sigio out of locks. */
858 	MALLOC(sigio, struct sigio *, sizeof(struct sigio), M_SIGIO, M_WAITOK);
859 	sigio->sio_pgid = pgid;
860 	sigio->sio_ucred = crhold(curthread->td_ucred);
861 	sigio->sio_myref = sigiop;
862 
863 	sx_slock(&proctree_lock);
864 	if (pgid > 0) {
865 		proc = pfind(pgid);
866 		if (proc == NULL) {
867 			ret = ESRCH;
868 			goto fail;
869 		}
870 
871 		/*
872 		 * Policy - Don't allow a process to FSETOWN a process
873 		 * in another session.
874 		 *
875 		 * Remove this test to allow maximum flexibility or
876 		 * restrict FSETOWN to the current process or process
877 		 * group for maximum safety.
878 		 */
879 		PROC_UNLOCK(proc);
880 		if (proc->p_session != curthread->td_proc->p_session) {
881 			ret = EPERM;
882 			goto fail;
883 		}
884 
885 		pgrp = NULL;
886 	} else /* if (pgid < 0) */ {
887 		pgrp = pgfind(-pgid);
888 		if (pgrp == NULL) {
889 			ret = ESRCH;
890 			goto fail;
891 		}
892 		PGRP_UNLOCK(pgrp);
893 
894 		/*
895 		 * Policy - Don't allow a process to FSETOWN a process
896 		 * in another session.
897 		 *
898 		 * Remove this test to allow maximum flexibility or
899 		 * restrict FSETOWN to the current process or process
900 		 * group for maximum safety.
901 		 */
902 		if (pgrp->pg_session != curthread->td_proc->p_session) {
903 			ret = EPERM;
904 			goto fail;
905 		}
906 
907 		proc = NULL;
908 	}
909 	funsetown(sigiop);
910 	if (pgid > 0) {
911 		PROC_LOCK(proc);
912 		/*
913 		 * Since funsetownlst() is called without the proctree
914 		 * locked, we need to check for P_WEXIT.
915 		 * XXX: is ESRCH correct?
916 		 */
917 		if ((proc->p_flag & P_WEXIT) != 0) {
918 			PROC_UNLOCK(proc);
919 			ret = ESRCH;
920 			goto fail;
921 		}
922 		SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio);
923 		sigio->sio_proc = proc;
924 		PROC_UNLOCK(proc);
925 	} else {
926 		PGRP_LOCK(pgrp);
927 		SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio);
928 		sigio->sio_pgrp = pgrp;
929 		PGRP_UNLOCK(pgrp);
930 	}
931 	sx_sunlock(&proctree_lock);
932 	SIGIO_LOCK();
933 	*sigiop = sigio;
934 	SIGIO_UNLOCK();
935 	return (0);
936 
937 fail:
938 	sx_sunlock(&proctree_lock);
939 	crfree(sigio->sio_ucred);
940 	FREE(sigio, M_SIGIO);
941 	return (ret);
942 }
943 
944 /*
945  * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg).
946  */
947 pid_t
948 fgetown(sigiop)
949 	struct sigio **sigiop;
950 {
951 	pid_t pgid;
952 
953 	SIGIO_LOCK();
954 	pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0;
955 	SIGIO_UNLOCK();
956 	return (pgid);
957 }
958 
959 /*
960  * Close a file descriptor.
961  */
962 #ifndef _SYS_SYSPROTO_H_
963 struct close_args {
964 	int     fd;
965 };
966 #endif
967 /*
968  * MPSAFE
969  */
970 /* ARGSUSED */
971 int
972 close(td, uap)
973 	struct thread *td;
974 	struct close_args *uap;
975 {
976 
977 	return (kern_close(td, uap->fd));
978 }
979 
980 int
981 kern_close(td, fd)
982 	struct thread *td;
983 	int fd;
984 {
985 	struct filedesc *fdp;
986 	struct file *fp;
987 	int error;
988 	int holdleaders;
989 
990 	error = 0;
991 	holdleaders = 0;
992 	fdp = td->td_proc->p_fd;
993 
994 	AUDIT_SYSCLOSE(td, fd);
995 
996 	FILEDESC_LOCK(fdp);
997 	if ((unsigned)fd >= fdp->fd_nfiles ||
998 	    (fp = fdp->fd_ofiles[fd]) == NULL) {
999 		FILEDESC_UNLOCK(fdp);
1000 		return (EBADF);
1001 	}
1002 	fdp->fd_ofiles[fd] = NULL;
1003 	fdp->fd_ofileflags[fd] = 0;
1004 	fdunused(fdp, fd);
1005 	if (td->td_proc->p_fdtol != NULL) {
1006 		/*
1007 		 * Ask fdfree() to sleep to ensure that all relevant
1008 		 * process leaders can be traversed in closef().
1009 		 */
1010 		fdp->fd_holdleaderscount++;
1011 		holdleaders = 1;
1012 	}
1013 
1014 	/*
1015 	 * We now hold the fp reference that used to be owned by the descriptor
1016 	 * array.
1017 	 * We have to unlock the FILEDESC *AFTER* knote_fdclose to prevent a
1018 	 * race of the fd getting opened, a knote added, and deleteing a knote
1019 	 * for the new fd.
1020 	 */
1021 	knote_fdclose(td, fd);
1022 	if (fp->f_type == DTYPE_MQUEUE)
1023 		mq_fdclose(td, fd, fp);
1024 	FILEDESC_UNLOCK(fdp);
1025 
1026 	error = closef(fp, td);
1027 	if (holdleaders) {
1028 		FILEDESC_LOCK_FAST(fdp);
1029 		fdp->fd_holdleaderscount--;
1030 		if (fdp->fd_holdleaderscount == 0 &&
1031 		    fdp->fd_holdleaderswakeup != 0) {
1032 			fdp->fd_holdleaderswakeup = 0;
1033 			wakeup(&fdp->fd_holdleaderscount);
1034 		}
1035 		FILEDESC_UNLOCK_FAST(fdp);
1036 	}
1037 	return (error);
1038 }
1039 
1040 #if defined(COMPAT_43)
1041 /*
1042  * Return status information about a file descriptor.
1043  */
1044 #ifndef _SYS_SYSPROTO_H_
1045 struct ofstat_args {
1046 	int	fd;
1047 	struct	ostat *sb;
1048 };
1049 #endif
1050 /*
1051  * MPSAFE
1052  */
1053 /* ARGSUSED */
1054 int
1055 ofstat(struct thread *td, struct ofstat_args *uap)
1056 {
1057 	struct ostat oub;
1058 	struct stat ub;
1059 	int error;
1060 
1061 	error = kern_fstat(td, uap->fd, &ub);
1062 	if (error == 0) {
1063 		cvtstat(&ub, &oub);
1064 		error = copyout(&oub, uap->sb, sizeof(oub));
1065 	}
1066 	return (error);
1067 }
1068 #endif /* COMPAT_43 */
1069 
1070 /*
1071  * Return status information about a file descriptor.
1072  */
1073 #ifndef _SYS_SYSPROTO_H_
1074 struct fstat_args {
1075 	int	fd;
1076 	struct	stat *sb;
1077 };
1078 #endif
1079 /*
1080  * MPSAFE
1081  */
1082 /* ARGSUSED */
1083 int
1084 fstat(struct thread *td, struct fstat_args *uap)
1085 {
1086 	struct stat ub;
1087 	int error;
1088 
1089 	error = kern_fstat(td, uap->fd, &ub);
1090 	if (error == 0)
1091 		error = copyout(&ub, uap->sb, sizeof(ub));
1092 	return (error);
1093 }
1094 
1095 int
1096 kern_fstat(struct thread *td, int fd, struct stat *sbp)
1097 {
1098 	struct file *fp;
1099 	int error;
1100 
1101 	AUDIT_ARG(fd, fd);
1102 
1103 	if ((error = fget(td, fd, &fp)) != 0)
1104 		return (error);
1105 
1106 	AUDIT_ARG(file, td->td_proc, fp);
1107 
1108 	error = fo_stat(fp, sbp, td->td_ucred, td);
1109 	fdrop(fp, td);
1110 	return (error);
1111 }
1112 
1113 /*
1114  * Return status information about a file descriptor.
1115  */
1116 #ifndef _SYS_SYSPROTO_H_
1117 struct nfstat_args {
1118 	int	fd;
1119 	struct	nstat *sb;
1120 };
1121 #endif
1122 /*
1123  * MPSAFE
1124  */
1125 /* ARGSUSED */
1126 int
1127 nfstat(struct thread *td, struct nfstat_args *uap)
1128 {
1129 	struct nstat nub;
1130 	struct stat ub;
1131 	int error;
1132 
1133 	error = kern_fstat(td, uap->fd, &ub);
1134 	if (error == 0) {
1135 		cvtnstat(&ub, &nub);
1136 		error = copyout(&nub, uap->sb, sizeof(nub));
1137 	}
1138 	return (error);
1139 }
1140 
1141 /*
1142  * Return pathconf information about a file descriptor.
1143  */
1144 #ifndef _SYS_SYSPROTO_H_
1145 struct fpathconf_args {
1146 	int	fd;
1147 	int	name;
1148 };
1149 #endif
1150 /*
1151  * MPSAFE
1152  */
1153 /* ARGSUSED */
1154 int
1155 fpathconf(struct thread *td, struct fpathconf_args *uap)
1156 {
1157 	struct file *fp;
1158 	struct vnode *vp;
1159 	int error;
1160 
1161 	if ((error = fget(td, uap->fd, &fp)) != 0)
1162 		return (error);
1163 
1164 	/* If asynchronous I/O is available, it works for all descriptors. */
1165 	if (uap->name == _PC_ASYNC_IO) {
1166 		td->td_retval[0] = async_io_version;
1167 		goto out;
1168 	}
1169 	vp = fp->f_vnode;
1170 	if (vp != NULL) {
1171 		int vfslocked;
1172 		vfslocked = VFS_LOCK_GIANT(vp->v_mount);
1173 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
1174 		error = VOP_PATHCONF(vp, uap->name, td->td_retval);
1175 		VOP_UNLOCK(vp, 0, td);
1176 		VFS_UNLOCK_GIANT(vfslocked);
1177 	} else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) {
1178 		if (uap->name != _PC_PIPE_BUF) {
1179 			error = EINVAL;
1180 		} else {
1181 			td->td_retval[0] = PIPE_BUF;
1182 		error = 0;
1183 		}
1184 	} else {
1185 		error = EOPNOTSUPP;
1186 	}
1187 out:
1188 	fdrop(fp, td);
1189 	return (error);
1190 }
1191 
1192 /*
1193  * Grow the file table to accomodate (at least) nfd descriptors.  This may
1194  * block and drop the filedesc lock, but it will reacquire it before
1195  * returning.
1196  */
1197 static void
1198 fdgrowtable(struct filedesc *fdp, int nfd)
1199 {
1200 	struct file **ntable;
1201 	char *nfileflags;
1202 	int nnfiles, onfiles;
1203 	NDSLOTTYPE *nmap;
1204 
1205 	FILEDESC_LOCK_ASSERT(fdp, MA_OWNED);
1206 
1207 	KASSERT(fdp->fd_nfiles > 0,
1208 	    ("zero-length file table"));
1209 
1210 	/* compute the size of the new table */
1211 	onfiles = fdp->fd_nfiles;
1212 	nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */
1213 	if (nnfiles <= onfiles)
1214 		/* the table is already large enough */
1215 		return;
1216 
1217 	/* allocate a new table and (if required) new bitmaps */
1218 	FILEDESC_UNLOCK(fdp);
1219 	MALLOC(ntable, struct file **, nnfiles * OFILESIZE,
1220 	    M_FILEDESC, M_ZERO | M_WAITOK);
1221 	nfileflags = (char *)&ntable[nnfiles];
1222 	if (NDSLOTS(nnfiles) > NDSLOTS(onfiles))
1223 		MALLOC(nmap, NDSLOTTYPE *, NDSLOTS(nnfiles) * NDSLOTSIZE,
1224 		    M_FILEDESC, M_ZERO | M_WAITOK);
1225 	else
1226 		nmap = NULL;
1227 	FILEDESC_LOCK(fdp);
1228 
1229 	/*
1230 	 * We now have new tables ready to go.  Since we dropped the
1231 	 * filedesc lock to call malloc(), watch out for a race.
1232 	 */
1233 	onfiles = fdp->fd_nfiles;
1234 	if (onfiles >= nnfiles) {
1235 		/* we lost the race, but that's OK */
1236 		free(ntable, M_FILEDESC);
1237 		if (nmap != NULL)
1238 			free(nmap, M_FILEDESC);
1239 		return;
1240 	}
1241 	bcopy(fdp->fd_ofiles, ntable, onfiles * sizeof(*ntable));
1242 	bcopy(fdp->fd_ofileflags, nfileflags, onfiles);
1243 	if (onfiles > NDFILE)
1244 		free(fdp->fd_ofiles, M_FILEDESC);
1245 	fdp->fd_ofiles = ntable;
1246 	fdp->fd_ofileflags = nfileflags;
1247 	if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) {
1248 		bcopy(fdp->fd_map, nmap, NDSLOTS(onfiles) * sizeof(*nmap));
1249 		if (NDSLOTS(onfiles) > NDSLOTS(NDFILE))
1250 			free(fdp->fd_map, M_FILEDESC);
1251 		fdp->fd_map = nmap;
1252 	}
1253 	fdp->fd_nfiles = nnfiles;
1254 }
1255 
1256 /*
1257  * Allocate a file descriptor for the process.
1258  */
1259 int
1260 fdalloc(struct thread *td, int minfd, int *result)
1261 {
1262 	struct proc *p = td->td_proc;
1263 	struct filedesc *fdp = p->p_fd;
1264 	int fd = -1, maxfd;
1265 
1266 	FILEDESC_LOCK_ASSERT(fdp, MA_OWNED);
1267 
1268 	if (fdp->fd_freefile > minfd)
1269 		minfd = fdp->fd_freefile;
1270 
1271 	PROC_LOCK(p);
1272 	maxfd = min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc);
1273 	PROC_UNLOCK(p);
1274 
1275 	/*
1276 	 * Search the bitmap for a free descriptor.  If none is found, try
1277 	 * to grow the file table.  Keep at it until we either get a file
1278 	 * descriptor or run into process or system limits; fdgrowtable()
1279 	 * may drop the filedesc lock, so we're in a race.
1280 	 */
1281 	for (;;) {
1282 		fd = fd_first_free(fdp, minfd, fdp->fd_nfiles);
1283 		if (fd >= maxfd)
1284 			return (EMFILE);
1285 		if (fd < fdp->fd_nfiles)
1286 			break;
1287 		fdgrowtable(fdp, min(fdp->fd_nfiles * 2, maxfd));
1288 	}
1289 
1290 	/*
1291 	 * Perform some sanity checks, then mark the file descriptor as
1292 	 * used and return it to the caller.
1293 	 */
1294 	KASSERT(!fdisused(fdp, fd),
1295 	    ("fd_first_free() returned non-free descriptor"));
1296 	KASSERT(fdp->fd_ofiles[fd] == NULL,
1297 	    ("free descriptor isn't"));
1298 	fdp->fd_ofileflags[fd] = 0; /* XXX needed? */
1299 	fdused(fdp, fd);
1300 	*result = fd;
1301 	return (0);
1302 }
1303 
1304 /*
1305  * Check to see whether n user file descriptors
1306  * are available to the process p.
1307  */
1308 int
1309 fdavail(struct thread *td, int n)
1310 {
1311 	struct proc *p = td->td_proc;
1312 	struct filedesc *fdp = td->td_proc->p_fd;
1313 	struct file **fpp;
1314 	int i, lim, last;
1315 
1316 	FILEDESC_LOCK_ASSERT(fdp, MA_OWNED);
1317 
1318 	PROC_LOCK(p);
1319 	lim = min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc);
1320 	PROC_UNLOCK(p);
1321 	if ((i = lim - fdp->fd_nfiles) > 0 && (n -= i) <= 0)
1322 		return (1);
1323 	last = min(fdp->fd_nfiles, lim);
1324 	fpp = &fdp->fd_ofiles[fdp->fd_freefile];
1325 	for (i = last - fdp->fd_freefile; --i >= 0; fpp++) {
1326 		if (*fpp == NULL && --n <= 0)
1327 			return (1);
1328 	}
1329 	return (0);
1330 }
1331 
1332 /*
1333  * Create a new open file structure and allocate
1334  * a file decriptor for the process that refers to it.
1335  * We add one reference to the file for the descriptor table
1336  * and one reference for resultfp. This is to prevent us being
1337  * preempted and the entry in the descriptor table closed after
1338  * we release the FILEDESC lock.
1339  */
1340 int
1341 falloc(struct thread *td, struct file **resultfp, int *resultfd)
1342 {
1343 	struct proc *p = td->td_proc;
1344 	struct file *fp, *fq;
1345 	int error, i;
1346 	int maxuserfiles = maxfiles - (maxfiles / 20);
1347 	static struct timeval lastfail;
1348 	static int curfail;
1349 
1350 	fp = uma_zalloc(file_zone, M_WAITOK | M_ZERO);
1351 	sx_xlock(&filelist_lock);
1352 
1353 	if ((openfiles >= maxuserfiles &&
1354 	     suser_cred(td->td_ucred, SUSER_RUID) != 0) ||
1355 	    openfiles >= maxfiles) {
1356 		if (ppsratecheck(&lastfail, &curfail, 1)) {
1357 			printf("kern.maxfiles limit exceeded by uid %i, please see tuning(7).\n",
1358 				td->td_ucred->cr_ruid);
1359 		}
1360 		sx_xunlock(&filelist_lock);
1361 		uma_zfree(file_zone, fp);
1362 		return (ENFILE);
1363 	}
1364 	openfiles++;
1365 
1366 	/*
1367 	 * If the process has file descriptor zero open, add the new file
1368 	 * descriptor to the list of open files at that point, otherwise
1369 	 * put it at the front of the list of open files.
1370 	 */
1371 	fp->f_mtxp = mtx_pool_alloc(mtxpool_sleep);
1372 	fp->f_count = 1;
1373 	if (resultfp)
1374 		fp->f_count++;
1375 	fp->f_cred = crhold(td->td_ucred);
1376 	fp->f_ops = &badfileops;
1377 	fp->f_data = NULL;
1378 	fp->f_vnode = NULL;
1379 	FILEDESC_LOCK(p->p_fd);
1380 	if ((fq = p->p_fd->fd_ofiles[0])) {
1381 		LIST_INSERT_AFTER(fq, fp, f_list);
1382 	} else {
1383 		LIST_INSERT_HEAD(&filehead, fp, f_list);
1384 	}
1385 	sx_xunlock(&filelist_lock);
1386 	if ((error = fdalloc(td, 0, &i))) {
1387 		FILEDESC_UNLOCK(p->p_fd);
1388 		fdrop(fp, td);
1389 		if (resultfp)
1390 			fdrop(fp, td);
1391 		return (error);
1392 	}
1393 	p->p_fd->fd_ofiles[i] = fp;
1394 	FILEDESC_UNLOCK(p->p_fd);
1395 	if (resultfp)
1396 		*resultfp = fp;
1397 	if (resultfd)
1398 		*resultfd = i;
1399 	return (0);
1400 }
1401 
1402 /*
1403  * Build a new filedesc structure from another.
1404  * Copy the current, root, and jail root vnode references.
1405  */
1406 struct filedesc *
1407 fdinit(struct filedesc *fdp)
1408 {
1409 	struct filedesc0 *newfdp;
1410 
1411 	newfdp = malloc(sizeof *newfdp, M_FILEDESC, M_WAITOK | M_ZERO);
1412 	mtx_init(&newfdp->fd_fd.fd_mtx, FILEDESC_LOCK_DESC, NULL, MTX_DEF);
1413 	if (fdp != NULL) {
1414 		FILEDESC_LOCK(fdp);
1415 		newfdp->fd_fd.fd_cdir = fdp->fd_cdir;
1416 		if (newfdp->fd_fd.fd_cdir)
1417 			VREF(newfdp->fd_fd.fd_cdir);
1418 		newfdp->fd_fd.fd_rdir = fdp->fd_rdir;
1419 		if (newfdp->fd_fd.fd_rdir)
1420 			VREF(newfdp->fd_fd.fd_rdir);
1421 		newfdp->fd_fd.fd_jdir = fdp->fd_jdir;
1422 		if (newfdp->fd_fd.fd_jdir)
1423 			VREF(newfdp->fd_fd.fd_jdir);
1424 		FILEDESC_UNLOCK(fdp);
1425 	}
1426 
1427 	/* Create the file descriptor table. */
1428 	newfdp->fd_fd.fd_refcnt = 1;
1429 	newfdp->fd_fd.fd_holdcnt = 1;
1430 	newfdp->fd_fd.fd_cmask = CMASK;
1431 	newfdp->fd_fd.fd_ofiles = newfdp->fd_dfiles;
1432 	newfdp->fd_fd.fd_ofileflags = newfdp->fd_dfileflags;
1433 	newfdp->fd_fd.fd_nfiles = NDFILE;
1434 	newfdp->fd_fd.fd_map = newfdp->fd_dmap;
1435 	newfdp->fd_fd.fd_lastfile = -1;
1436 	return (&newfdp->fd_fd);
1437 }
1438 
1439 static struct filedesc *
1440 fdhold(struct proc *p)
1441 {
1442 	struct filedesc *fdp;
1443 
1444 	mtx_lock(&fdesc_mtx);
1445 	fdp = p->p_fd;
1446 	if (fdp != NULL)
1447 		fdp->fd_holdcnt++;
1448 	mtx_unlock(&fdesc_mtx);
1449 	return (fdp);
1450 }
1451 
1452 static void
1453 fddrop(struct filedesc *fdp)
1454 {
1455 	int i;
1456 
1457 	mtx_lock(&fdesc_mtx);
1458 	i = --fdp->fd_holdcnt;
1459 	mtx_unlock(&fdesc_mtx);
1460 	if (i > 0)
1461 		return;
1462 
1463 	mtx_destroy(&fdp->fd_mtx);
1464 	FREE(fdp, M_FILEDESC);
1465 }
1466 
1467 /*
1468  * Share a filedesc structure.
1469  */
1470 struct filedesc *
1471 fdshare(struct filedesc *fdp)
1472 {
1473 	FILEDESC_LOCK_FAST(fdp);
1474 	fdp->fd_refcnt++;
1475 	FILEDESC_UNLOCK_FAST(fdp);
1476 	return (fdp);
1477 }
1478 
1479 /*
1480  * Unshare a filedesc structure, if necessary by making a copy
1481  */
1482 void
1483 fdunshare(struct proc *p, struct thread *td)
1484 {
1485 
1486 	FILEDESC_LOCK_FAST(p->p_fd);
1487 	if (p->p_fd->fd_refcnt > 1) {
1488 		struct filedesc *tmp;
1489 
1490 		FILEDESC_UNLOCK_FAST(p->p_fd);
1491 		tmp = fdcopy(p->p_fd);
1492 		fdfree(td);
1493 		p->p_fd = tmp;
1494 	} else
1495 		FILEDESC_UNLOCK_FAST(p->p_fd);
1496 }
1497 
1498 /*
1499  * Copy a filedesc structure.
1500  * A NULL pointer in returns a NULL reference, this is to ease callers,
1501  * not catch errors.
1502  */
1503 struct filedesc *
1504 fdcopy(struct filedesc *fdp)
1505 {
1506 	struct filedesc *newfdp;
1507 	int i;
1508 
1509 	/* Certain daemons might not have file descriptors. */
1510 	if (fdp == NULL)
1511 		return (NULL);
1512 
1513 	newfdp = fdinit(fdp);
1514 	FILEDESC_LOCK_FAST(fdp);
1515 	while (fdp->fd_lastfile >= newfdp->fd_nfiles) {
1516 		FILEDESC_UNLOCK_FAST(fdp);
1517 		FILEDESC_LOCK(newfdp);
1518 		fdgrowtable(newfdp, fdp->fd_lastfile + 1);
1519 		FILEDESC_UNLOCK(newfdp);
1520 		FILEDESC_LOCK_FAST(fdp);
1521 	}
1522 	/* copy everything except kqueue descriptors */
1523 	newfdp->fd_freefile = -1;
1524 	for (i = 0; i <= fdp->fd_lastfile; ++i) {
1525 		if (fdisused(fdp, i) &&
1526 		    fdp->fd_ofiles[i]->f_type != DTYPE_KQUEUE) {
1527 			newfdp->fd_ofiles[i] = fdp->fd_ofiles[i];
1528 			newfdp->fd_ofileflags[i] = fdp->fd_ofileflags[i];
1529 			fhold(newfdp->fd_ofiles[i]);
1530 			newfdp->fd_lastfile = i;
1531 		} else {
1532 			if (newfdp->fd_freefile == -1)
1533 				newfdp->fd_freefile = i;
1534 		}
1535 	}
1536 	FILEDESC_UNLOCK_FAST(fdp);
1537 	FILEDESC_LOCK(newfdp);
1538 	for (i = 0; i <= newfdp->fd_lastfile; ++i)
1539 		if (newfdp->fd_ofiles[i] != NULL)
1540 			fdused(newfdp, i);
1541 	FILEDESC_UNLOCK(newfdp);
1542 	FILEDESC_LOCK_FAST(fdp);
1543 	if (newfdp->fd_freefile == -1)
1544 		newfdp->fd_freefile = i;
1545 	newfdp->fd_cmask = fdp->fd_cmask;
1546 	FILEDESC_UNLOCK_FAST(fdp);
1547 	return (newfdp);
1548 }
1549 
1550 /*
1551  * Release a filedesc structure.
1552  */
1553 void
1554 fdfree(struct thread *td)
1555 {
1556 	struct filedesc *fdp;
1557 	struct file **fpp;
1558 	int i, locked;
1559 	struct filedesc_to_leader *fdtol;
1560 	struct file *fp;
1561 	struct vnode *cdir, *jdir, *rdir, *vp;
1562 	struct flock lf;
1563 
1564 	/* Certain daemons might not have file descriptors. */
1565 	fdp = td->td_proc->p_fd;
1566 	if (fdp == NULL)
1567 		return;
1568 
1569 	/* Check for special need to clear POSIX style locks */
1570 	fdtol = td->td_proc->p_fdtol;
1571 	if (fdtol != NULL) {
1572 		FILEDESC_LOCK(fdp);
1573 		KASSERT(fdtol->fdl_refcount > 0,
1574 			("filedesc_to_refcount botch: fdl_refcount=%d",
1575 			 fdtol->fdl_refcount));
1576 		if (fdtol->fdl_refcount == 1 &&
1577 		    (td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
1578 			for (i = 0, fpp = fdp->fd_ofiles;
1579 			     i <= fdp->fd_lastfile;
1580 			     i++, fpp++) {
1581 				if (*fpp == NULL ||
1582 				    (*fpp)->f_type != DTYPE_VNODE)
1583 					continue;
1584 				fp = *fpp;
1585 				fhold(fp);
1586 				FILEDESC_UNLOCK(fdp);
1587 				lf.l_whence = SEEK_SET;
1588 				lf.l_start = 0;
1589 				lf.l_len = 0;
1590 				lf.l_type = F_UNLCK;
1591 				vp = fp->f_vnode;
1592 				locked = VFS_LOCK_GIANT(vp->v_mount);
1593 				(void) VOP_ADVLOCK(vp,
1594 						   (caddr_t)td->td_proc->
1595 						   p_leader,
1596 						   F_UNLCK,
1597 						   &lf,
1598 						   F_POSIX);
1599 				VFS_UNLOCK_GIANT(locked);
1600 				FILEDESC_LOCK(fdp);
1601 				fdrop(fp, td);
1602 				fpp = fdp->fd_ofiles + i;
1603 			}
1604 		}
1605 	retry:
1606 		if (fdtol->fdl_refcount == 1) {
1607 			if (fdp->fd_holdleaderscount > 0 &&
1608 			    (td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
1609 				/*
1610 				 * close() or do_dup() has cleared a reference
1611 				 * in a shared file descriptor table.
1612 				 */
1613 				fdp->fd_holdleaderswakeup = 1;
1614 				msleep(&fdp->fd_holdleaderscount, &fdp->fd_mtx,
1615 				       PLOCK, "fdlhold", 0);
1616 				goto retry;
1617 			}
1618 			if (fdtol->fdl_holdcount > 0) {
1619 				/*
1620 				 * Ensure that fdtol->fdl_leader
1621 				 * remains valid in closef().
1622 				 */
1623 				fdtol->fdl_wakeup = 1;
1624 				msleep(fdtol, &fdp->fd_mtx,
1625 				       PLOCK, "fdlhold", 0);
1626 				goto retry;
1627 			}
1628 		}
1629 		fdtol->fdl_refcount--;
1630 		if (fdtol->fdl_refcount == 0 &&
1631 		    fdtol->fdl_holdcount == 0) {
1632 			fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
1633 			fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
1634 		} else
1635 			fdtol = NULL;
1636 		td->td_proc->p_fdtol = NULL;
1637 		FILEDESC_UNLOCK(fdp);
1638 		if (fdtol != NULL)
1639 			FREE(fdtol, M_FILEDESC_TO_LEADER);
1640 	}
1641 	FILEDESC_LOCK_FAST(fdp);
1642 	i = --fdp->fd_refcnt;
1643 	FILEDESC_UNLOCK_FAST(fdp);
1644 	if (i > 0)
1645 		return;
1646 	/*
1647 	 * We are the last reference to the structure, so we can
1648 	 * safely assume it will not change out from under us.
1649 	 */
1650 	fpp = fdp->fd_ofiles;
1651 	for (i = fdp->fd_lastfile; i-- >= 0; fpp++) {
1652 		if (*fpp)
1653 			(void) closef(*fpp, td);
1654 	}
1655 	FILEDESC_LOCK(fdp);
1656 
1657 	/* XXX This should happen earlier. */
1658 	mtx_lock(&fdesc_mtx);
1659 	td->td_proc->p_fd = NULL;
1660 	mtx_unlock(&fdesc_mtx);
1661 
1662 	if (fdp->fd_nfiles > NDFILE)
1663 		FREE(fdp->fd_ofiles, M_FILEDESC);
1664 	if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE))
1665 		FREE(fdp->fd_map, M_FILEDESC);
1666 
1667 	fdp->fd_nfiles = 0;
1668 
1669 	cdir = fdp->fd_cdir;
1670 	fdp->fd_cdir = NULL;
1671 	rdir = fdp->fd_rdir;
1672 	fdp->fd_rdir = NULL;
1673 	jdir = fdp->fd_jdir;
1674 	fdp->fd_jdir = NULL;
1675 	FILEDESC_UNLOCK(fdp);
1676 
1677 	if (cdir) {
1678 		locked = VFS_LOCK_GIANT(cdir->v_mount);
1679 		vrele(cdir);
1680 		VFS_UNLOCK_GIANT(locked);
1681 	}
1682 	if (rdir) {
1683 		locked = VFS_LOCK_GIANT(rdir->v_mount);
1684 		vrele(rdir);
1685 		VFS_UNLOCK_GIANT(locked);
1686 	}
1687 	if (jdir) {
1688 		locked = VFS_LOCK_GIANT(jdir->v_mount);
1689 		vrele(jdir);
1690 		VFS_UNLOCK_GIANT(locked);
1691 	}
1692 
1693 	fddrop(fdp);
1694 }
1695 
1696 /*
1697  * For setugid programs, we don't want to people to use that setugidness
1698  * to generate error messages which write to a file which otherwise would
1699  * otherwise be off-limits to the process.  We check for filesystems where
1700  * the vnode can change out from under us after execve (like [lin]procfs).
1701  *
1702  * Since setugidsafety calls this only for fd 0, 1 and 2, this check is
1703  * sufficient.  We also don't check for setugidness since we know we are.
1704  */
1705 static int
1706 is_unsafe(struct file *fp)
1707 {
1708 	if (fp->f_type == DTYPE_VNODE) {
1709 		struct vnode *vp = fp->f_vnode;
1710 
1711 		if ((vp->v_vflag & VV_PROCDEP) != 0)
1712 			return (1);
1713 	}
1714 	return (0);
1715 }
1716 
1717 /*
1718  * Make this setguid thing safe, if at all possible.
1719  */
1720 void
1721 setugidsafety(struct thread *td)
1722 {
1723 	struct filedesc *fdp;
1724 	int i;
1725 
1726 	/* Certain daemons might not have file descriptors. */
1727 	fdp = td->td_proc->p_fd;
1728 	if (fdp == NULL)
1729 		return;
1730 
1731 	/*
1732 	 * Note: fdp->fd_ofiles may be reallocated out from under us while
1733 	 * we are blocked in a close.  Be careful!
1734 	 */
1735 	FILEDESC_LOCK(fdp);
1736 	for (i = 0; i <= fdp->fd_lastfile; i++) {
1737 		if (i > 2)
1738 			break;
1739 		if (fdp->fd_ofiles[i] && is_unsafe(fdp->fd_ofiles[i])) {
1740 			struct file *fp;
1741 
1742 			knote_fdclose(td, i);
1743 			/*
1744 			 * NULL-out descriptor prior to close to avoid
1745 			 * a race while close blocks.
1746 			 */
1747 			fp = fdp->fd_ofiles[i];
1748 			fdp->fd_ofiles[i] = NULL;
1749 			fdp->fd_ofileflags[i] = 0;
1750 			fdunused(fdp, i);
1751 			FILEDESC_UNLOCK(fdp);
1752 			(void) closef(fp, td);
1753 			FILEDESC_LOCK(fdp);
1754 		}
1755 	}
1756 	FILEDESC_UNLOCK(fdp);
1757 }
1758 
1759 void
1760 fdclose(struct filedesc *fdp, struct file *fp, int idx, struct thread *td)
1761 {
1762 
1763 	FILEDESC_LOCK(fdp);
1764 	if (fdp->fd_ofiles[idx] == fp) {
1765 		fdp->fd_ofiles[idx] = NULL;
1766 		fdunused(fdp, idx);
1767 		FILEDESC_UNLOCK(fdp);
1768 		fdrop(fp, td);
1769 	} else {
1770 		FILEDESC_UNLOCK(fdp);
1771 	}
1772 }
1773 
1774 /*
1775  * Close any files on exec?
1776  */
1777 void
1778 fdcloseexec(struct thread *td)
1779 {
1780 	struct filedesc *fdp;
1781 	int i;
1782 
1783 	/* Certain daemons might not have file descriptors. */
1784 	fdp = td->td_proc->p_fd;
1785 	if (fdp == NULL)
1786 		return;
1787 
1788 	FILEDESC_LOCK(fdp);
1789 
1790 	/*
1791 	 * We cannot cache fd_ofiles or fd_ofileflags since operations
1792 	 * may block and rip them out from under us.
1793 	 */
1794 	for (i = 0; i <= fdp->fd_lastfile; i++) {
1795 		if (fdp->fd_ofiles[i] != NULL &&
1796 		    (fdp->fd_ofiles[i]->f_type == DTYPE_MQUEUE ||
1797 		    (fdp->fd_ofileflags[i] & UF_EXCLOSE))) {
1798 			struct file *fp;
1799 
1800 			knote_fdclose(td, i);
1801 			/*
1802 			 * NULL-out descriptor prior to close to avoid
1803 			 * a race while close blocks.
1804 			 */
1805 			fp = fdp->fd_ofiles[i];
1806 			fdp->fd_ofiles[i] = NULL;
1807 			fdp->fd_ofileflags[i] = 0;
1808 			fdunused(fdp, i);
1809 			if (fp->f_type == DTYPE_MQUEUE)
1810 				mq_fdclose(td, i, fp);
1811 			FILEDESC_UNLOCK(fdp);
1812 			(void) closef(fp, td);
1813 			FILEDESC_LOCK(fdp);
1814 		}
1815 	}
1816 	FILEDESC_UNLOCK(fdp);
1817 }
1818 
1819 /*
1820  * It is unsafe for set[ug]id processes to be started with file
1821  * descriptors 0..2 closed, as these descriptors are given implicit
1822  * significance in the Standard C library.  fdcheckstd() will create a
1823  * descriptor referencing /dev/null for each of stdin, stdout, and
1824  * stderr that is not already open.
1825  */
1826 int
1827 fdcheckstd(struct thread *td)
1828 {
1829 	struct nameidata nd;
1830 	struct filedesc *fdp;
1831 	struct file *fp;
1832 	register_t retval;
1833 	int fd, i, error, flags, devnull;
1834 
1835 	fdp = td->td_proc->p_fd;
1836 	if (fdp == NULL)
1837 		return (0);
1838 	KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
1839 	devnull = -1;
1840 	error = 0;
1841 	for (i = 0; i < 3; i++) {
1842 		if (fdp->fd_ofiles[i] != NULL)
1843 			continue;
1844 		if (devnull < 0) {
1845 			int vfslocked;
1846 			error = falloc(td, &fp, &fd);
1847 			if (error != 0)
1848 				break;
1849 			/* Note extra ref on `fp' held for us by falloc(). */
1850 			KASSERT(fd == i, ("oof, we didn't get our fd"));
1851 			NDINIT(&nd, LOOKUP, FOLLOW | MPSAFE, UIO_SYSSPACE,
1852 			    "/dev/null", td);
1853 			flags = FREAD | FWRITE;
1854 			error = vn_open(&nd, &flags, 0, fd);
1855 			if (error != 0) {
1856 				/*
1857 				 * Someone may have closed the entry in the
1858 				 * file descriptor table, so check it hasn't
1859 				 * changed before dropping the reference count.
1860 				 */
1861 				FILEDESC_LOCK(fdp);
1862 				KASSERT(fdp->fd_ofiles[fd] == fp,
1863 				    ("table not shared, how did it change?"));
1864 				fdp->fd_ofiles[fd] = NULL;
1865 				fdunused(fdp, fd);
1866 				FILEDESC_UNLOCK(fdp);
1867 				fdrop(fp, td);
1868 				fdrop(fp, td);
1869 				break;
1870 			}
1871 			vfslocked = NDHASGIANT(&nd);
1872 			NDFREE(&nd, NDF_ONLY_PNBUF);
1873 			fp->f_flag = flags;
1874 			fp->f_vnode = nd.ni_vp;
1875 			if (fp->f_data == NULL)
1876 				fp->f_data = nd.ni_vp;
1877 			if (fp->f_ops == &badfileops)
1878 				fp->f_ops = &vnops;
1879 			fp->f_type = DTYPE_VNODE;
1880 			VOP_UNLOCK(nd.ni_vp, 0, td);
1881 			VFS_UNLOCK_GIANT(vfslocked);
1882 			devnull = fd;
1883 			fdrop(fp, td);
1884 		} else {
1885 			error = do_dup(td, DUP_FIXED, devnull, i, &retval);
1886 			if (error != 0)
1887 				break;
1888 		}
1889 	}
1890 	return (error);
1891 }
1892 
1893 /*
1894  * Internal form of close.
1895  * Decrement reference count on file structure.
1896  * Note: td may be NULL when closing a file that was being passed in a
1897  * message.
1898  *
1899  * XXXRW: Giant is not required for the caller, but often will be held; this
1900  * makes it moderately likely the Giant will be recursed in the VFS case.
1901  */
1902 int
1903 closef(struct file *fp, struct thread *td)
1904 {
1905 	struct vnode *vp;
1906 	struct flock lf;
1907 	struct filedesc_to_leader *fdtol;
1908 	struct filedesc *fdp;
1909 
1910 	/*
1911 	 * POSIX record locking dictates that any close releases ALL
1912 	 * locks owned by this process.  This is handled by setting
1913 	 * a flag in the unlock to free ONLY locks obeying POSIX
1914 	 * semantics, and not to free BSD-style file locks.
1915 	 * If the descriptor was in a message, POSIX-style locks
1916 	 * aren't passed with the descriptor, and the thread pointer
1917 	 * will be NULL.  Callers should be careful only to pass a
1918 	 * NULL thread pointer when there really is no owning
1919 	 * context that might have locks, or the locks will be
1920 	 * leaked.
1921 	 */
1922 	if (fp->f_type == DTYPE_VNODE && td != NULL) {
1923 		int vfslocked;
1924 
1925 		vp = fp->f_vnode;
1926 		vfslocked = VFS_LOCK_GIANT(vp->v_mount);
1927 		if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
1928 			lf.l_whence = SEEK_SET;
1929 			lf.l_start = 0;
1930 			lf.l_len = 0;
1931 			lf.l_type = F_UNLCK;
1932 			(void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader,
1933 					   F_UNLCK, &lf, F_POSIX);
1934 		}
1935 		fdtol = td->td_proc->p_fdtol;
1936 		if (fdtol != NULL) {
1937 			/*
1938 			 * Handle special case where file descriptor table
1939 			 * is shared between multiple process leaders.
1940 			 */
1941 			fdp = td->td_proc->p_fd;
1942 			FILEDESC_LOCK(fdp);
1943 			for (fdtol = fdtol->fdl_next;
1944 			     fdtol != td->td_proc->p_fdtol;
1945 			     fdtol = fdtol->fdl_next) {
1946 				if ((fdtol->fdl_leader->p_flag &
1947 				     P_ADVLOCK) == 0)
1948 					continue;
1949 				fdtol->fdl_holdcount++;
1950 				FILEDESC_UNLOCK(fdp);
1951 				lf.l_whence = SEEK_SET;
1952 				lf.l_start = 0;
1953 				lf.l_len = 0;
1954 				lf.l_type = F_UNLCK;
1955 				vp = fp->f_vnode;
1956 				(void) VOP_ADVLOCK(vp,
1957 						   (caddr_t)fdtol->fdl_leader,
1958 						   F_UNLCK, &lf, F_POSIX);
1959 				FILEDESC_LOCK(fdp);
1960 				fdtol->fdl_holdcount--;
1961 				if (fdtol->fdl_holdcount == 0 &&
1962 				    fdtol->fdl_wakeup != 0) {
1963 					fdtol->fdl_wakeup = 0;
1964 					wakeup(fdtol);
1965 				}
1966 			}
1967 			FILEDESC_UNLOCK(fdp);
1968 		}
1969 		VFS_UNLOCK_GIANT(vfslocked);
1970 	}
1971 	return (fdrop(fp, td));
1972 }
1973 
1974 /*
1975  * Extract the file pointer associated with the specified descriptor for
1976  * the current user process.
1977  *
1978  * If the descriptor doesn't exist, EBADF is returned.
1979  *
1980  * If the descriptor exists but doesn't match 'flags' then
1981  * return EBADF for read attempts and EINVAL for write attempts.
1982  *
1983  * If 'hold' is set (non-zero) the file's refcount will be bumped on return.
1984  * It should be dropped with fdrop().
1985  * If it is not set, then the refcount will not be bumped however the
1986  * thread's filedesc struct will be returned locked (for fgetsock).
1987  *
1988  * If an error occured the non-zero error is returned and *fpp is set to NULL.
1989  * Otherwise *fpp is set and zero is returned.
1990  */
1991 static __inline int
1992 _fget(struct thread *td, int fd, struct file **fpp, int flags, int hold)
1993 {
1994 	struct filedesc *fdp;
1995 	struct file *fp;
1996 
1997 	*fpp = NULL;
1998 	if (td == NULL || (fdp = td->td_proc->p_fd) == NULL)
1999 		return (EBADF);
2000 	FILEDESC_LOCK(fdp);
2001 	if ((fp = fget_locked(fdp, fd)) == NULL || fp->f_ops == &badfileops) {
2002 		FILEDESC_UNLOCK(fdp);
2003 		return (EBADF);
2004 	}
2005 
2006 	/*
2007 	 * FREAD and FWRITE failure return EBADF as per POSIX.
2008 	 *
2009 	 * Only one flag, or 0, may be specified.
2010 	 */
2011 	if (flags == FREAD && (fp->f_flag & FREAD) == 0) {
2012 		FILEDESC_UNLOCK(fdp);
2013 		return (EBADF);
2014 	}
2015 	if (flags == FWRITE && (fp->f_flag & FWRITE) == 0) {
2016 		FILEDESC_UNLOCK(fdp);
2017 		return (EBADF);
2018 	}
2019 	if (hold) {
2020 		fhold(fp);
2021 		FILEDESC_UNLOCK(fdp);
2022 	}
2023 	*fpp = fp;
2024 	return (0);
2025 }
2026 
2027 int
2028 fget(struct thread *td, int fd, struct file **fpp)
2029 {
2030 
2031 	return(_fget(td, fd, fpp, 0, 1));
2032 }
2033 
2034 int
2035 fget_read(struct thread *td, int fd, struct file **fpp)
2036 {
2037 
2038 	return(_fget(td, fd, fpp, FREAD, 1));
2039 }
2040 
2041 int
2042 fget_write(struct thread *td, int fd, struct file **fpp)
2043 {
2044 
2045 	return(_fget(td, fd, fpp, FWRITE, 1));
2046 }
2047 
2048 /*
2049  * Like fget() but loads the underlying vnode, or returns an error if
2050  * the descriptor does not represent a vnode.  Note that pipes use vnodes
2051  * but never have VM objects.  The returned vnode will be vref()d.
2052  *
2053  * XXX: what about the unused flags ?
2054  */
2055 static __inline int
2056 _fgetvp(struct thread *td, int fd, struct vnode **vpp, int flags)
2057 {
2058 	struct file *fp;
2059 	int error;
2060 
2061 	*vpp = NULL;
2062 	if ((error = _fget(td, fd, &fp, 0, 0)) != 0)
2063 		return (error);
2064 	if (fp->f_vnode == NULL) {
2065 		error = EINVAL;
2066 	} else {
2067 		*vpp = fp->f_vnode;
2068 		vref(*vpp);
2069 	}
2070 	FILEDESC_UNLOCK(td->td_proc->p_fd);
2071 	return (error);
2072 }
2073 
2074 int
2075 fgetvp(struct thread *td, int fd, struct vnode **vpp)
2076 {
2077 
2078 	return (_fgetvp(td, fd, vpp, 0));
2079 }
2080 
2081 int
2082 fgetvp_read(struct thread *td, int fd, struct vnode **vpp)
2083 {
2084 
2085 	return (_fgetvp(td, fd, vpp, FREAD));
2086 }
2087 
2088 #ifdef notyet
2089 int
2090 fgetvp_write(struct thread *td, int fd, struct vnode **vpp)
2091 {
2092 
2093 	return (_fgetvp(td, fd, vpp, FWRITE));
2094 }
2095 #endif
2096 
2097 /*
2098  * Like fget() but loads the underlying socket, or returns an error if
2099  * the descriptor does not represent a socket.
2100  *
2101  * We bump the ref count on the returned socket.  XXX Also obtain the SX
2102  * lock in the future.
2103  *
2104  * XXXRW: fgetsock() and fputsock() are deprecated, as consumers should rely
2105  * on their file descriptor reference to prevent the socket from being
2106  * freed during use.
2107  */
2108 int
2109 fgetsock(struct thread *td, int fd, struct socket **spp, u_int *fflagp)
2110 {
2111 	struct file *fp;
2112 	int error;
2113 
2114 	NET_ASSERT_GIANT();
2115 
2116 	*spp = NULL;
2117 	if (fflagp != NULL)
2118 		*fflagp = 0;
2119 	if ((error = _fget(td, fd, &fp, 0, 0)) != 0)
2120 		return (error);
2121 	if (fp->f_type != DTYPE_SOCKET) {
2122 		error = ENOTSOCK;
2123 	} else {
2124 		*spp = fp->f_data;
2125 		if (fflagp)
2126 			*fflagp = fp->f_flag;
2127 		SOCK_LOCK(*spp);
2128 		soref(*spp);
2129 		SOCK_UNLOCK(*spp);
2130 	}
2131 	FILEDESC_UNLOCK(td->td_proc->p_fd);
2132 	return (error);
2133 }
2134 
2135 /*
2136  * Drop the reference count on the socket and XXX release the SX lock in the
2137  * future.  The last reference closes the socket.
2138  *
2139  * XXXRW: fputsock() is deprecated, see comment for fgetsock().
2140  */
2141 void
2142 fputsock(struct socket *so)
2143 {
2144 
2145 	NET_ASSERT_GIANT();
2146 	ACCEPT_LOCK();
2147 	SOCK_LOCK(so);
2148 	sorele(so);
2149 }
2150 
2151 int
2152 fdrop(struct file *fp, struct thread *td)
2153 {
2154 
2155 	FILE_LOCK(fp);
2156 	return (fdrop_locked(fp, td));
2157 }
2158 
2159 /*
2160  * Drop reference on struct file passed in, may call closef if the
2161  * reference hits zero.
2162  * Expects struct file locked, and will unlock it.
2163  */
2164 static int
2165 fdrop_locked(struct file *fp, struct thread *td)
2166 {
2167 	int error;
2168 
2169 	FILE_LOCK_ASSERT(fp, MA_OWNED);
2170 
2171 	if (--fp->f_count > 0) {
2172 		FILE_UNLOCK(fp);
2173 		return (0);
2174 	}
2175 	/* We have the last ref so we can proceed without the file lock. */
2176 	FILE_UNLOCK(fp);
2177 	if (fp->f_count < 0)
2178 		panic("fdrop: count < 0");
2179 	if (fp->f_ops != &badfileops)
2180 		error = fo_close(fp, td);
2181 	else
2182 		error = 0;
2183 
2184 	sx_xlock(&filelist_lock);
2185 	LIST_REMOVE(fp, f_list);
2186 	openfiles--;
2187 	sx_xunlock(&filelist_lock);
2188 	crfree(fp->f_cred);
2189 	uma_zfree(file_zone, fp);
2190 
2191 	return (error);
2192 }
2193 
2194 /*
2195  * Apply an advisory lock on a file descriptor.
2196  *
2197  * Just attempt to get a record lock of the requested type on
2198  * the entire file (l_whence = SEEK_SET, l_start = 0, l_len = 0).
2199  */
2200 #ifndef _SYS_SYSPROTO_H_
2201 struct flock_args {
2202 	int	fd;
2203 	int	how;
2204 };
2205 #endif
2206 /*
2207  * MPSAFE
2208  */
2209 /* ARGSUSED */
2210 int
2211 flock(struct thread *td, struct flock_args *uap)
2212 {
2213 	struct file *fp;
2214 	struct vnode *vp;
2215 	struct flock lf;
2216 	int error;
2217 
2218 	if ((error = fget(td, uap->fd, &fp)) != 0)
2219 		return (error);
2220 	if (fp->f_type != DTYPE_VNODE) {
2221 		fdrop(fp, td);
2222 		return (EOPNOTSUPP);
2223 	}
2224 
2225 	mtx_lock(&Giant);
2226 	vp = fp->f_vnode;
2227 	lf.l_whence = SEEK_SET;
2228 	lf.l_start = 0;
2229 	lf.l_len = 0;
2230 	if (uap->how & LOCK_UN) {
2231 		lf.l_type = F_UNLCK;
2232 		FILE_LOCK(fp);
2233 		fp->f_flag &= ~FHASLOCK;
2234 		FILE_UNLOCK(fp);
2235 		error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK);
2236 		goto done2;
2237 	}
2238 	if (uap->how & LOCK_EX)
2239 		lf.l_type = F_WRLCK;
2240 	else if (uap->how & LOCK_SH)
2241 		lf.l_type = F_RDLCK;
2242 	else {
2243 		error = EBADF;
2244 		goto done2;
2245 	}
2246 	FILE_LOCK(fp);
2247 	fp->f_flag |= FHASLOCK;
2248 	FILE_UNLOCK(fp);
2249 	error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf,
2250 	    (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT);
2251 done2:
2252 	fdrop(fp, td);
2253 	mtx_unlock(&Giant);
2254 	return (error);
2255 }
2256 /*
2257  * Duplicate the specified descriptor to a free descriptor.
2258  */
2259 int
2260 dupfdopen(struct thread *td, struct filedesc *fdp, int indx, int dfd, int mode, int error)
2261 {
2262 	struct file *wfp;
2263 	struct file *fp;
2264 
2265 	/*
2266 	 * If the to-be-dup'd fd number is greater than the allowed number
2267 	 * of file descriptors, or the fd to be dup'd has already been
2268 	 * closed, then reject.
2269 	 */
2270 	FILEDESC_LOCK(fdp);
2271 	if (dfd < 0 || dfd >= fdp->fd_nfiles ||
2272 	    (wfp = fdp->fd_ofiles[dfd]) == NULL) {
2273 		FILEDESC_UNLOCK(fdp);
2274 		return (EBADF);
2275 	}
2276 
2277 	/*
2278 	 * There are two cases of interest here.
2279 	 *
2280 	 * For ENODEV simply dup (dfd) to file descriptor
2281 	 * (indx) and return.
2282 	 *
2283 	 * For ENXIO steal away the file structure from (dfd) and
2284 	 * store it in (indx).  (dfd) is effectively closed by
2285 	 * this operation.
2286 	 *
2287 	 * Any other error code is just returned.
2288 	 */
2289 	switch (error) {
2290 	case ENODEV:
2291 		/*
2292 		 * Check that the mode the file is being opened for is a
2293 		 * subset of the mode of the existing descriptor.
2294 		 */
2295 		FILE_LOCK(wfp);
2296 		if (((mode & (FREAD|FWRITE)) | wfp->f_flag) != wfp->f_flag) {
2297 			FILE_UNLOCK(wfp);
2298 			FILEDESC_UNLOCK(fdp);
2299 			return (EACCES);
2300 		}
2301 		fp = fdp->fd_ofiles[indx];
2302 		fdp->fd_ofiles[indx] = wfp;
2303 		fdp->fd_ofileflags[indx] = fdp->fd_ofileflags[dfd];
2304 		if (fp == NULL)
2305 			fdused(fdp, indx);
2306 		fhold_locked(wfp);
2307 		FILE_UNLOCK(wfp);
2308 		FILEDESC_UNLOCK(fdp);
2309 		if (fp != NULL) {
2310 			/*
2311 			 * We now own the reference to fp that the ofiles[]
2312 			 * array used to own.  Release it.
2313 			 */
2314 			FILE_LOCK(fp);
2315 			fdrop_locked(fp, td);
2316 		}
2317 		return (0);
2318 
2319 	case ENXIO:
2320 		/*
2321 		 * Steal away the file pointer from dfd and stuff it into indx.
2322 		 */
2323 		fp = fdp->fd_ofiles[indx];
2324 		fdp->fd_ofiles[indx] = fdp->fd_ofiles[dfd];
2325 		fdp->fd_ofiles[dfd] = NULL;
2326 		fdp->fd_ofileflags[indx] = fdp->fd_ofileflags[dfd];
2327 		fdp->fd_ofileflags[dfd] = 0;
2328 		fdunused(fdp, dfd);
2329 		if (fp == NULL)
2330 			fdused(fdp, indx);
2331 		if (fp != NULL)
2332 			FILE_LOCK(fp);
2333 
2334 		/*
2335 		 * We now own the reference to fp that the ofiles[] array
2336 		 * used to own.  Release it.
2337 		 */
2338 		if (fp != NULL)
2339 			fdrop_locked(fp, td);
2340 
2341 		FILEDESC_UNLOCK(fdp);
2342 
2343 		return (0);
2344 
2345 	default:
2346 		FILEDESC_UNLOCK(fdp);
2347 		return (error);
2348 	}
2349 	/* NOTREACHED */
2350 }
2351 
2352 /*
2353  * Scan all active processes to see if any of them have a current
2354  * or root directory of `olddp'. If so, replace them with the new
2355  * mount point.
2356  */
2357 void
2358 mountcheckdirs(struct vnode *olddp, struct vnode *newdp)
2359 {
2360 	struct filedesc *fdp;
2361 	struct proc *p;
2362 	int nrele;
2363 
2364 	if (vrefcnt(olddp) == 1)
2365 		return;
2366 	sx_slock(&allproc_lock);
2367 	LIST_FOREACH(p, &allproc, p_list) {
2368 		fdp = fdhold(p);
2369 		if (fdp == NULL)
2370 			continue;
2371 		nrele = 0;
2372 		FILEDESC_LOCK_FAST(fdp);
2373 		if (fdp->fd_cdir == olddp) {
2374 			vref(newdp);
2375 			fdp->fd_cdir = newdp;
2376 			nrele++;
2377 		}
2378 		if (fdp->fd_rdir == olddp) {
2379 			vref(newdp);
2380 			fdp->fd_rdir = newdp;
2381 			nrele++;
2382 		}
2383 		FILEDESC_UNLOCK_FAST(fdp);
2384 		fddrop(fdp);
2385 		while (nrele--)
2386 			vrele(olddp);
2387 	}
2388 	sx_sunlock(&allproc_lock);
2389 	if (rootvnode == olddp) {
2390 		vrele(rootvnode);
2391 		vref(newdp);
2392 		rootvnode = newdp;
2393 	}
2394 }
2395 
2396 struct filedesc_to_leader *
2397 filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader)
2398 {
2399 	struct filedesc_to_leader *fdtol;
2400 
2401 	MALLOC(fdtol, struct filedesc_to_leader *,
2402 	       sizeof(struct filedesc_to_leader),
2403 	       M_FILEDESC_TO_LEADER,
2404 	       M_WAITOK);
2405 	fdtol->fdl_refcount = 1;
2406 	fdtol->fdl_holdcount = 0;
2407 	fdtol->fdl_wakeup = 0;
2408 	fdtol->fdl_leader = leader;
2409 	if (old != NULL) {
2410 		FILEDESC_LOCK(fdp);
2411 		fdtol->fdl_next = old->fdl_next;
2412 		fdtol->fdl_prev = old;
2413 		old->fdl_next = fdtol;
2414 		fdtol->fdl_next->fdl_prev = fdtol;
2415 		FILEDESC_UNLOCK(fdp);
2416 	} else {
2417 		fdtol->fdl_next = fdtol;
2418 		fdtol->fdl_prev = fdtol;
2419 	}
2420 	return (fdtol);
2421 }
2422 
2423 /*
2424  * Get file structures.
2425  */
2426 static int
2427 sysctl_kern_file(SYSCTL_HANDLER_ARGS)
2428 {
2429 	struct xfile xf;
2430 	struct filedesc *fdp;
2431 	struct file *fp;
2432 	struct proc *p;
2433 	int error, n;
2434 
2435 	/*
2436 	 * Note: because the number of file descriptors is calculated
2437 	 * in different ways for sizing vs returning the data,
2438 	 * there is information leakage from the first loop.  However,
2439 	 * it is of a similar order of magnitude to the leakage from
2440 	 * global system statistics such as kern.openfiles.
2441 	 */
2442 	error = sysctl_wire_old_buffer(req, 0);
2443 	if (error != 0)
2444 		return (error);
2445 	if (req->oldptr == NULL) {
2446 		n = 16;		/* A slight overestimate. */
2447 		sx_slock(&filelist_lock);
2448 		LIST_FOREACH(fp, &filehead, f_list) {
2449 			/*
2450 			 * We should grab the lock, but this is an
2451 			 * estimate, so does it really matter?
2452 			 */
2453 			/* mtx_lock(fp->f_mtxp); */
2454 			n += fp->f_count;
2455 			/* mtx_unlock(f->f_mtxp); */
2456 		}
2457 		sx_sunlock(&filelist_lock);
2458 		return (SYSCTL_OUT(req, 0, n * sizeof(xf)));
2459 	}
2460 	error = 0;
2461 	bzero(&xf, sizeof(xf));
2462 	xf.xf_size = sizeof(xf);
2463 	sx_slock(&allproc_lock);
2464 	LIST_FOREACH(p, &allproc, p_list) {
2465 		if (p->p_state == PRS_NEW)
2466 			continue;
2467 		PROC_LOCK(p);
2468 		if (p_cansee(req->td, p) != 0) {
2469 			PROC_UNLOCK(p);
2470 			continue;
2471 		}
2472 		xf.xf_pid = p->p_pid;
2473 		xf.xf_uid = p->p_ucred->cr_uid;
2474 		PROC_UNLOCK(p);
2475 		fdp = fdhold(p);
2476 		if (fdp == NULL)
2477 			continue;
2478 		FILEDESC_LOCK_FAST(fdp);
2479 		for (n = 0; fdp->fd_refcnt > 0 && n < fdp->fd_nfiles; ++n) {
2480 			if ((fp = fdp->fd_ofiles[n]) == NULL)
2481 				continue;
2482 			xf.xf_fd = n;
2483 			xf.xf_file = fp;
2484 			xf.xf_data = fp->f_data;
2485 			xf.xf_vnode = fp->f_vnode;
2486 			xf.xf_type = fp->f_type;
2487 			xf.xf_count = fp->f_count;
2488 			xf.xf_msgcount = fp->f_msgcount;
2489 			xf.xf_offset = fp->f_offset;
2490 			xf.xf_flag = fp->f_flag;
2491 			error = SYSCTL_OUT(req, &xf, sizeof(xf));
2492 			if (error)
2493 				break;
2494 		}
2495 		FILEDESC_UNLOCK_FAST(fdp);
2496 		fddrop(fdp);
2497 		if (error)
2498 			break;
2499 	}
2500 	sx_sunlock(&allproc_lock);
2501 	return (error);
2502 }
2503 
2504 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD,
2505     0, 0, sysctl_kern_file, "S,xfile", "Entire file table");
2506 
2507 #ifdef DDB
2508 /*
2509  * For the purposes of debugging, generate a human-readable string for the
2510  * file type.
2511  */
2512 static const char *
2513 file_type_to_name(short type)
2514 {
2515 
2516 	switch (type) {
2517 	case 0:
2518 		return ("zero");
2519 	case DTYPE_VNODE:
2520 		return ("vnod");
2521 	case DTYPE_SOCKET:
2522 		return ("sock");
2523 	case DTYPE_PIPE:
2524 		return ("pipe");
2525 	case DTYPE_FIFO:
2526 		return ("fifo");
2527 	case DTYPE_CRYPTO:
2528 		return ("crpt");
2529 	default:
2530 		return ("unkn");
2531 	}
2532 }
2533 
2534 /*
2535  * For the purposes of debugging, identify a process (if any, perhaps one of
2536  * many) that references the passed file in its file descriptor array. Return
2537  * NULL if none.
2538  */
2539 static struct proc *
2540 file_to_first_proc(struct file *fp)
2541 {
2542 	struct filedesc *fdp;
2543 	struct proc *p;
2544 	int n;
2545 
2546 	LIST_FOREACH(p, &allproc, p_list) {
2547 		if (p->p_state == PRS_NEW)
2548 			continue;
2549 		fdp = p->p_fd;
2550 		if (fdp == NULL)
2551 			continue;
2552 		for (n = 0; n < fdp->fd_nfiles; n++) {
2553 			if (fp == fdp->fd_ofiles[n])
2554 				return (p);
2555 		}
2556 	}
2557 	return (NULL);
2558 }
2559 
2560 DB_SHOW_COMMAND(files, db_show_files)
2561 {
2562 	struct file *fp;
2563 	struct proc *p;
2564 
2565 	db_printf("%8s %4s %8s %8s %4s %5s %6s %8s %5s %12s\n", "File",
2566 	    "Type", "Data", "Flag", "GCFl", "Count", "MCount", "Vnode",
2567 	    "FPID", "FCmd");
2568 	LIST_FOREACH(fp, &filehead, f_list) {
2569 		p = file_to_first_proc(fp);
2570 		db_printf("%8p %4s %8p %08x %04x %5d %6d %8p %5d %12s\n", fp,
2571 		    file_type_to_name(fp->f_type), fp->f_data, fp->f_flag,
2572 		    fp->f_gcflag, fp->f_count, fp->f_msgcount, fp->f_vnode,
2573 		    p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-");
2574 	}
2575 }
2576 #endif
2577 
2578 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
2579     &maxfilesperproc, 0, "Maximum files allowed open per process");
2580 
2581 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
2582     &maxfiles, 0, "Maximum number of files");
2583 
2584 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
2585     &openfiles, 0, "System-wide number of open files");
2586 
2587 /* ARGSUSED*/
2588 static void
2589 filelistinit(void *dummy)
2590 {
2591 
2592 	file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL,
2593 	    NULL, NULL, UMA_ALIGN_PTR, 0);
2594 	sx_init(&filelist_lock, "filelist lock");
2595 	mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF);
2596 	mtx_init(&fdesc_mtx, "fdesc", NULL, MTX_DEF);
2597 }
2598 SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL)
2599 
2600 /*-------------------------------------------------------------------*/
2601 
2602 static int
2603 badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred, int flags, struct thread *td)
2604 {
2605 
2606 	return (EBADF);
2607 }
2608 
2609 static int
2610 badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred, struct thread *td)
2611 {
2612 
2613 	return (EBADF);
2614 }
2615 
2616 static int
2617 badfo_poll(struct file *fp, int events, struct ucred *active_cred, struct thread *td)
2618 {
2619 
2620 	return (0);
2621 }
2622 
2623 static int
2624 badfo_kqfilter(struct file *fp, struct knote *kn)
2625 {
2626 
2627 	return (0);
2628 }
2629 
2630 static int
2631 badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred, struct thread *td)
2632 {
2633 
2634 	return (EBADF);
2635 }
2636 
2637 static int
2638 badfo_close(struct file *fp, struct thread *td)
2639 {
2640 
2641 	return (EBADF);
2642 }
2643 
2644 struct fileops badfileops = {
2645 	.fo_read = badfo_readwrite,
2646 	.fo_write = badfo_readwrite,
2647 	.fo_ioctl = badfo_ioctl,
2648 	.fo_poll = badfo_poll,
2649 	.fo_kqfilter = badfo_kqfilter,
2650 	.fo_stat = badfo_stat,
2651 	.fo_close = badfo_close,
2652 };
2653 
2654 
2655 /*-------------------------------------------------------------------*/
2656 
2657 /*
2658  * File Descriptor pseudo-device driver (/dev/fd/).
2659  *
2660  * Opening minor device N dup()s the file (if any) connected to file
2661  * descriptor N belonging to the calling process.  Note that this driver
2662  * consists of only the ``open()'' routine, because all subsequent
2663  * references to this file will be direct to the other driver.
2664  *
2665  * XXX: we could give this one a cloning event handler if necessary.
2666  */
2667 
2668 /* ARGSUSED */
2669 static int
2670 fdopen(struct cdev *dev, int mode, int type, struct thread *td)
2671 {
2672 
2673 	/*
2674 	 * XXX Kludge: set curthread->td_dupfd to contain the value of the
2675 	 * the file descriptor being sought for duplication. The error
2676 	 * return ensures that the vnode for this device will be released
2677 	 * by vn_open. Open will detect this special error and take the
2678 	 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
2679 	 * will simply report the error.
2680 	 */
2681 	td->td_dupfd = dev2unit(dev);
2682 	return (ENODEV);
2683 }
2684 
2685 static struct cdevsw fildesc_cdevsw = {
2686 	.d_version =	D_VERSION,
2687 	.d_flags =	D_NEEDGIANT,
2688 	.d_open =	fdopen,
2689 	.d_name =	"FD",
2690 };
2691 
2692 static void
2693 fildesc_drvinit(void *unused)
2694 {
2695 	struct cdev *dev;
2696 
2697 	dev = make_dev(&fildesc_cdevsw, 0, UID_ROOT, GID_WHEEL, 0666, "fd/0");
2698 	make_dev_alias(dev, "stdin");
2699 	dev = make_dev(&fildesc_cdevsw, 1, UID_ROOT, GID_WHEEL, 0666, "fd/1");
2700 	make_dev_alias(dev, "stdout");
2701 	dev = make_dev(&fildesc_cdevsw, 2, UID_ROOT, GID_WHEEL, 0666, "fd/2");
2702 	make_dev_alias(dev, "stderr");
2703 }
2704 
2705 SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL)
2706