xref: /freebsd/sys/kern/kern_descrip.c (revision 6af83ee0d2941d18880b6aaa2b4facd1d30c6106)
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 
42 #include <sys/param.h>
43 #include <sys/systm.h>
44 
45 #include <sys/conf.h>
46 #include <sys/fcntl.h>
47 #include <sys/file.h>
48 #include <sys/filedesc.h>
49 #include <sys/filio.h>
50 #include <sys/jail.h>
51 #include <sys/kernel.h>
52 #include <sys/limits.h>
53 #include <sys/lock.h>
54 #include <sys/malloc.h>
55 #include <sys/mount.h>
56 #include <sys/mutex.h>
57 #include <sys/namei.h>
58 #include <sys/proc.h>
59 #include <sys/resourcevar.h>
60 #include <sys/signalvar.h>
61 #include <sys/socketvar.h>
62 #include <sys/stat.h>
63 #include <sys/sx.h>
64 #include <sys/syscallsubr.h>
65 #include <sys/sysctl.h>
66 #include <sys/sysproto.h>
67 #include <sys/unistd.h>
68 #include <sys/vnode.h>
69 
70 #include <vm/uma.h>
71 
72 static MALLOC_DEFINE(M_FILEDESC, "file desc", "Open file descriptor table");
73 static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "file desc to leader",
74 		     "file desc to leader structures");
75 static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures");
76 
77 static uma_zone_t file_zone;
78 
79 
80 /* How to treat 'new' parameter when allocating a fd for do_dup(). */
81 enum dup_type { DUP_VARIABLE, DUP_FIXED };
82 
83 static int do_dup(struct thread *td, enum dup_type type, int old, int new,
84     register_t *retval);
85 static int	fd_first_free(struct filedesc *, int, int);
86 static int	fd_last_used(struct filedesc *, int, int);
87 static void	fdgrowtable(struct filedesc *, int);
88 static int	fdrop_locked(struct file *fp, struct thread *td);
89 static void	fdunused(struct filedesc *fdp, int fd);
90 static void	fdused(struct filedesc *fdp, int fd);
91 
92 /*
93  * A process is initially started out with NDFILE descriptors stored within
94  * this structure, selected to be enough for typical applications based on
95  * the historical limit of 20 open files (and the usage of descriptors by
96  * shells).  If these descriptors are exhausted, a larger descriptor table
97  * may be allocated, up to a process' resource limit; the internal arrays
98  * are then unused.
99  */
100 #define NDFILE		20
101 #define NDSLOTSIZE	sizeof(NDSLOTTYPE)
102 #define	NDENTRIES	(NDSLOTSIZE * __CHAR_BIT)
103 #define NDSLOT(x)	((x) / NDENTRIES)
104 #define NDBIT(x)	((NDSLOTTYPE)1 << ((x) % NDENTRIES))
105 #define	NDSLOTS(x)	(((x) + NDENTRIES - 1) / NDENTRIES)
106 
107 /*
108  * Storage required per open file descriptor.
109  */
110 #define OFILESIZE (sizeof(struct file *) + sizeof(char))
111 
112 /*
113  * Basic allocation of descriptors:
114  * one of the above, plus arrays for NDFILE descriptors.
115  */
116 struct filedesc0 {
117 	struct	filedesc fd_fd;
118 	/*
119 	 * These arrays are used when the number of open files is
120 	 * <= NDFILE, and are then pointed to by the pointers above.
121 	 */
122 	struct	file *fd_dfiles[NDFILE];
123 	char	fd_dfileflags[NDFILE];
124 	NDSLOTTYPE fd_dmap[NDSLOTS(NDFILE)];
125 };
126 
127 /*
128  * Descriptor management.
129  */
130 struct filelist filehead;	/* head of list of open files */
131 int openfiles;			/* actual number of open files */
132 struct sx filelist_lock;	/* sx to protect filelist */
133 struct mtx sigio_lock;		/* mtx to protect pointers to sigio */
134 
135 /* A mutex to protect the association between a proc and filedesc. */
136 static struct mtx	fdesc_mtx;
137 
138 /*
139  * Find the first zero bit in the given bitmap, starting at low and not
140  * exceeding size - 1.
141  */
142 static int
143 fd_first_free(struct filedesc *fdp, int low, int size)
144 {
145 	NDSLOTTYPE *map = fdp->fd_map;
146 	NDSLOTTYPE mask;
147 	int off, maxoff;
148 
149 	if (low >= size)
150 		return (low);
151 
152 	off = NDSLOT(low);
153 	if (low % NDENTRIES) {
154 		mask = ~(~(NDSLOTTYPE)0 >> (NDENTRIES - (low % NDENTRIES)));
155 		if ((mask &= ~map[off]) != 0UL)
156 			return (off * NDENTRIES + ffsl(mask) - 1);
157 		++off;
158 	}
159 	for (maxoff = NDSLOTS(size); off < maxoff; ++off)
160 		if (map[off] != ~0UL)
161 			return (off * NDENTRIES + ffsl(~map[off]) - 1);
162 	return (size);
163 }
164 
165 /*
166  * Find the highest non-zero bit in the given bitmap, starting at low and
167  * not exceeding size - 1.
168  */
169 static int
170 fd_last_used(struct filedesc *fdp, int low, int size)
171 {
172 	NDSLOTTYPE *map = fdp->fd_map;
173 	NDSLOTTYPE mask;
174 	int off, minoff;
175 
176 	if (low >= size)
177 		return (-1);
178 
179 	off = NDSLOT(size);
180 	if (size % NDENTRIES) {
181 		mask = ~(~(NDSLOTTYPE)0 << (size % NDENTRIES));
182 		if ((mask &= map[off]) != 0)
183 			return (off * NDENTRIES + flsl(mask) - 1);
184 		--off;
185 	}
186 	for (minoff = NDSLOT(low); off >= minoff; --off)
187 		if (map[off] != 0)
188 			return (off * NDENTRIES + flsl(map[off]) - 1);
189 	return (size - 1);
190 }
191 
192 static int
193 fdisused(struct filedesc *fdp, int fd)
194 {
195         KASSERT(fd >= 0 && fd < fdp->fd_nfiles,
196             ("file descriptor %d out of range (0, %d)", fd, fdp->fd_nfiles));
197 	return ((fdp->fd_map[NDSLOT(fd)] & NDBIT(fd)) != 0);
198 }
199 
200 /*
201  * Mark a file descriptor as used.
202  */
203 static void
204 fdused(struct filedesc *fdp, int fd)
205 {
206 	FILEDESC_LOCK_ASSERT(fdp, MA_OWNED);
207 	KASSERT(!fdisused(fdp, fd),
208 	    ("fd already used"));
209 	fdp->fd_map[NDSLOT(fd)] |= NDBIT(fd);
210 	if (fd > fdp->fd_lastfile)
211 		fdp->fd_lastfile = fd;
212 	if (fd == fdp->fd_freefile)
213 		fdp->fd_freefile = fd_first_free(fdp, fd, fdp->fd_nfiles);
214 }
215 
216 /*
217  * Mark a file descriptor as unused.
218  */
219 static void
220 fdunused(struct filedesc *fdp, int fd)
221 {
222 	FILEDESC_LOCK_ASSERT(fdp, MA_OWNED);
223 	KASSERT(fdisused(fdp, fd),
224 	    ("fd is already unused"));
225 	KASSERT(fdp->fd_ofiles[fd] == NULL,
226 	    ("fd is still in use"));
227 	fdp->fd_map[NDSLOT(fd)] &= ~NDBIT(fd);
228 	if (fd < fdp->fd_freefile)
229 		fdp->fd_freefile = fd;
230 	if (fd == fdp->fd_lastfile)
231 		fdp->fd_lastfile = fd_last_used(fdp, 0, fd);
232 }
233 
234 /*
235  * System calls on descriptors.
236  */
237 #ifndef _SYS_SYSPROTO_H_
238 struct getdtablesize_args {
239 	int	dummy;
240 };
241 #endif
242 /*
243  * MPSAFE
244  */
245 /* ARGSUSED */
246 int
247 getdtablesize(struct thread *td, struct getdtablesize_args *uap)
248 {
249 	struct proc *p = td->td_proc;
250 
251 	PROC_LOCK(p);
252 	td->td_retval[0] =
253 	    min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc);
254 	PROC_UNLOCK(p);
255 	return (0);
256 }
257 
258 /*
259  * Duplicate a file descriptor to a particular value.
260  *
261  * note: keep in mind that a potential race condition exists when closing
262  * descriptors from a shared descriptor table (via rfork).
263  */
264 #ifndef _SYS_SYSPROTO_H_
265 struct dup2_args {
266 	u_int	from;
267 	u_int	to;
268 };
269 #endif
270 /*
271  * MPSAFE
272  */
273 /* ARGSUSED */
274 int
275 dup2(struct thread *td, struct dup2_args *uap)
276 {
277 
278 	return (do_dup(td, DUP_FIXED, (int)uap->from, (int)uap->to,
279 		    td->td_retval));
280 }
281 
282 /*
283  * Duplicate a file descriptor.
284  */
285 #ifndef _SYS_SYSPROTO_H_
286 struct dup_args {
287 	u_int	fd;
288 };
289 #endif
290 /*
291  * MPSAFE
292  */
293 /* ARGSUSED */
294 int
295 dup(struct thread *td, struct dup_args *uap)
296 {
297 
298 	return (do_dup(td, DUP_VARIABLE, (int)uap->fd, 0, td->td_retval));
299 }
300 
301 /*
302  * The file control system call.
303  */
304 #ifndef _SYS_SYSPROTO_H_
305 struct fcntl_args {
306 	int	fd;
307 	int	cmd;
308 	long	arg;
309 };
310 #endif
311 /*
312  * MPSAFE
313  */
314 /* ARGSUSED */
315 int
316 fcntl(struct thread *td, struct fcntl_args *uap)
317 {
318 	struct flock fl;
319 	intptr_t arg;
320 	int error;
321 
322 	error = 0;
323 	switch (uap->cmd) {
324 	case F_GETLK:
325 	case F_SETLK:
326 	case F_SETLKW:
327 		error = copyin((void *)(intptr_t)uap->arg, &fl, sizeof(fl));
328 		arg = (intptr_t)&fl;
329 		break;
330 	default:
331 		arg = uap->arg;
332 		break;
333 	}
334 	if (error)
335 		return (error);
336 	error = kern_fcntl(td, uap->fd, uap->cmd, arg);
337 	if (error)
338 		return (error);
339 	if (uap->cmd == F_GETLK)
340 		error = copyout(&fl, (void *)(intptr_t)uap->arg, sizeof(fl));
341 	return (error);
342 }
343 
344 int
345 kern_fcntl(struct thread *td, int fd, int cmd, intptr_t arg)
346 {
347 	struct filedesc *fdp;
348 	struct flock *flp;
349 	struct file *fp;
350 	struct proc *p;
351 	char *pop;
352 	struct vnode *vp;
353 	u_int newmin;
354 	int error, flg, tmp;
355 	int giant_locked;
356 
357 	/*
358 	 * XXXRW: Some fcntl() calls require Giant -- others don't.  Try to
359 	 * avoid grabbing Giant for calls we know don't need it.
360 	 */
361 	switch (cmd) {
362 	case F_DUPFD:
363 	case F_GETFD:
364 	case F_SETFD:
365 	case F_GETFL:
366 		giant_locked = 0;
367 		break;
368 
369 	default:
370 		giant_locked = 1;
371 		mtx_lock(&Giant);
372 	}
373 
374 	error = 0;
375 	flg = F_POSIX;
376 	p = td->td_proc;
377 	fdp = p->p_fd;
378 	FILEDESC_LOCK(fdp);
379 	if ((unsigned)fd >= fdp->fd_nfiles ||
380 	    (fp = fdp->fd_ofiles[fd]) == NULL) {
381 		FILEDESC_UNLOCK(fdp);
382 		error = EBADF;
383 		goto done2;
384 	}
385 	pop = &fdp->fd_ofileflags[fd];
386 
387 	switch (cmd) {
388 	case F_DUPFD:
389 		/* mtx_assert(&Giant, MA_NOTOWNED); */
390 		FILEDESC_UNLOCK(fdp);
391 		newmin = arg;
392 		PROC_LOCK(p);
393 		if (newmin >= lim_cur(p, RLIMIT_NOFILE) ||
394 		    newmin >= maxfilesperproc) {
395 			PROC_UNLOCK(p);
396 			error = EINVAL;
397 			break;
398 		}
399 		PROC_UNLOCK(p);
400 		error = do_dup(td, DUP_VARIABLE, fd, newmin, td->td_retval);
401 		break;
402 
403 	case F_GETFD:
404 		/* mtx_assert(&Giant, MA_NOTOWNED); */
405 		td->td_retval[0] = (*pop & UF_EXCLOSE) ? FD_CLOEXEC : 0;
406 		FILEDESC_UNLOCK(fdp);
407 		break;
408 
409 	case F_SETFD:
410 		/* mtx_assert(&Giant, MA_NOTOWNED); */
411 		*pop = (*pop &~ UF_EXCLOSE) |
412 		    (arg & FD_CLOEXEC ? UF_EXCLOSE : 0);
413 		FILEDESC_UNLOCK(fdp);
414 		break;
415 
416 	case F_GETFL:
417 		/* mtx_assert(&Giant, MA_NOTOWNED); */
418 		FILE_LOCK(fp);
419 		td->td_retval[0] = OFLAGS(fp->f_flag);
420 		FILE_UNLOCK(fp);
421 		FILEDESC_UNLOCK(fdp);
422 		break;
423 
424 	case F_SETFL:
425 		mtx_assert(&Giant, MA_OWNED);
426 		FILE_LOCK(fp);
427 		fhold_locked(fp);
428 		fp->f_flag &= ~FCNTLFLAGS;
429 		fp->f_flag |= FFLAGS(arg & ~O_ACCMODE) & FCNTLFLAGS;
430 		FILE_UNLOCK(fp);
431 		FILEDESC_UNLOCK(fdp);
432 		tmp = fp->f_flag & FNONBLOCK;
433 		error = fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
434 		if (error) {
435 			fdrop(fp, td);
436 			break;
437 		}
438 		tmp = fp->f_flag & FASYNC;
439 		error = fo_ioctl(fp, FIOASYNC, &tmp, td->td_ucred, td);
440 		if (error == 0) {
441 			fdrop(fp, td);
442 			break;
443 		}
444 		FILE_LOCK(fp);
445 		fp->f_flag &= ~FNONBLOCK;
446 		FILE_UNLOCK(fp);
447 		tmp = 0;
448 		(void)fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
449 		fdrop(fp, td);
450 		break;
451 
452 	case F_GETOWN:
453 		mtx_assert(&Giant, MA_OWNED);
454 		fhold(fp);
455 		FILEDESC_UNLOCK(fdp);
456 		error = fo_ioctl(fp, FIOGETOWN, &tmp, td->td_ucred, td);
457 		if (error == 0)
458 			td->td_retval[0] = tmp;
459 		fdrop(fp, td);
460 		break;
461 
462 	case F_SETOWN:
463 		mtx_assert(&Giant, MA_OWNED);
464 		fhold(fp);
465 		FILEDESC_UNLOCK(fdp);
466 		tmp = arg;
467 		error = fo_ioctl(fp, FIOSETOWN, &tmp, td->td_ucred, td);
468 		fdrop(fp, td);
469 		break;
470 
471 	case F_SETLKW:
472 		mtx_assert(&Giant, MA_OWNED);
473 		flg |= F_WAIT;
474 		/* FALLTHROUGH F_SETLK */
475 
476 	case F_SETLK:
477 		mtx_assert(&Giant, MA_OWNED);
478 		if (fp->f_type != DTYPE_VNODE) {
479 			FILEDESC_UNLOCK(fdp);
480 			error = EBADF;
481 			break;
482 		}
483 
484 		flp = (struct flock *)arg;
485 		if (flp->l_whence == SEEK_CUR) {
486 			if (fp->f_offset < 0 ||
487 			    (flp->l_start > 0 &&
488 			     fp->f_offset > OFF_MAX - flp->l_start)) {
489 				FILEDESC_UNLOCK(fdp);
490 				error = EOVERFLOW;
491 				break;
492 			}
493 			flp->l_start += fp->f_offset;
494 		}
495 
496 		/*
497 		 * VOP_ADVLOCK() may block.
498 		 */
499 		fhold(fp);
500 		FILEDESC_UNLOCK(fdp);
501 		vp = fp->f_vnode;
502 
503 		switch (flp->l_type) {
504 		case F_RDLCK:
505 			if ((fp->f_flag & FREAD) == 0) {
506 				error = EBADF;
507 				break;
508 			}
509 			PROC_LOCK(p->p_leader);
510 			p->p_leader->p_flag |= P_ADVLOCK;
511 			PROC_UNLOCK(p->p_leader);
512 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
513 			    flp, flg);
514 			break;
515 		case F_WRLCK:
516 			if ((fp->f_flag & FWRITE) == 0) {
517 				error = EBADF;
518 				break;
519 			}
520 			PROC_LOCK(p->p_leader);
521 			p->p_leader->p_flag |= P_ADVLOCK;
522 			PROC_UNLOCK(p->p_leader);
523 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
524 			    flp, flg);
525 			break;
526 		case F_UNLCK:
527 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
528 			    flp, F_POSIX);
529 			break;
530 		default:
531 			error = EINVAL;
532 			break;
533 		}
534 		/* Check for race with close */
535 		FILEDESC_LOCK_FAST(fdp);
536 		if ((unsigned) fd >= fdp->fd_nfiles ||
537 		    fp != fdp->fd_ofiles[fd]) {
538 			FILEDESC_UNLOCK_FAST(fdp);
539 			flp->l_whence = SEEK_SET;
540 			flp->l_start = 0;
541 			flp->l_len = 0;
542 			flp->l_type = F_UNLCK;
543 			(void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
544 					   F_UNLCK, flp, F_POSIX);
545 		} else
546 			FILEDESC_UNLOCK_FAST(fdp);
547 		fdrop(fp, td);
548 		break;
549 
550 	case F_GETLK:
551 		mtx_assert(&Giant, MA_OWNED);
552 		if (fp->f_type != DTYPE_VNODE) {
553 			FILEDESC_UNLOCK(fdp);
554 			error = EBADF;
555 			break;
556 		}
557 		flp = (struct flock *)arg;
558 		if (flp->l_type != F_RDLCK && flp->l_type != F_WRLCK &&
559 		    flp->l_type != F_UNLCK) {
560 			FILEDESC_UNLOCK(fdp);
561 			error = EINVAL;
562 			break;
563 		}
564 		if (flp->l_whence == SEEK_CUR) {
565 			if ((flp->l_start > 0 &&
566 			    fp->f_offset > OFF_MAX - flp->l_start) ||
567 			    (flp->l_start < 0 &&
568 			     fp->f_offset < OFF_MIN - flp->l_start)) {
569 				FILEDESC_UNLOCK(fdp);
570 				error = EOVERFLOW;
571 				break;
572 			}
573 			flp->l_start += fp->f_offset;
574 		}
575 		/*
576 		 * VOP_ADVLOCK() may block.
577 		 */
578 		fhold(fp);
579 		FILEDESC_UNLOCK(fdp);
580 		vp = fp->f_vnode;
581 		error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, flp,
582 		    F_POSIX);
583 		fdrop(fp, td);
584 		break;
585 	default:
586 		FILEDESC_UNLOCK(fdp);
587 		error = EINVAL;
588 		break;
589 	}
590 done2:
591 	if (giant_locked)
592 		mtx_unlock(&Giant);
593 	return (error);
594 }
595 
596 /*
597  * Common code for dup, dup2, and fcntl(F_DUPFD).
598  */
599 static int
600 do_dup(struct thread *td, enum dup_type type, int old, int new, register_t *retval)
601 {
602 	struct filedesc *fdp;
603 	struct proc *p;
604 	struct file *fp;
605 	struct file *delfp;
606 	int error, holdleaders, maxfd;
607 
608 	KASSERT((type == DUP_VARIABLE || type == DUP_FIXED),
609 	    ("invalid dup type %d", type));
610 
611 	p = td->td_proc;
612 	fdp = p->p_fd;
613 
614 	/*
615 	 * Verify we have a valid descriptor to dup from and possibly to
616 	 * dup to.
617 	 */
618 	if (old < 0 || new < 0)
619 		return (EBADF);
620 	PROC_LOCK(p);
621 	maxfd = min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc);
622 	PROC_UNLOCK(p);
623 	if (new >= maxfd)
624 		return (EMFILE);
625 
626 	FILEDESC_LOCK(fdp);
627 	if (old >= fdp->fd_nfiles || fdp->fd_ofiles[old] == NULL) {
628 		FILEDESC_UNLOCK(fdp);
629 		return (EBADF);
630 	}
631 	if (type == DUP_FIXED && old == new) {
632 		*retval = new;
633 		FILEDESC_UNLOCK(fdp);
634 		return (0);
635 	}
636 	fp = fdp->fd_ofiles[old];
637 	fhold(fp);
638 
639 	/*
640 	 * If the caller specified a file descriptor, make sure the file
641 	 * table is large enough to hold it, and grab it.  Otherwise, just
642 	 * allocate a new descriptor the usual way.  Since the filedesc
643 	 * lock may be temporarily dropped in the process, we have to look
644 	 * out for a race.
645 	 */
646 	if (type == DUP_FIXED) {
647 		if (new >= fdp->fd_nfiles)
648 			fdgrowtable(fdp, new + 1);
649 		if (fdp->fd_ofiles[new] == NULL)
650 			fdused(fdp, new);
651 	} else {
652 		if ((error = fdalloc(td, new, &new)) != 0) {
653 			FILEDESC_UNLOCK(fdp);
654 			fdrop(fp, td);
655 			return (error);
656 		}
657 	}
658 
659 	/*
660 	 * If the old file changed out from under us then treat it as a
661 	 * bad file descriptor.  Userland should do its own locking to
662 	 * avoid this case.
663 	 */
664 	if (fdp->fd_ofiles[old] != fp) {
665 		/* we've allocated a descriptor which we won't use */
666 		if (fdp->fd_ofiles[new] == NULL)
667 			fdunused(fdp, new);
668 		FILEDESC_UNLOCK(fdp);
669 		fdrop(fp, td);
670 		return (EBADF);
671 	}
672 	KASSERT(old != new,
673 	    ("new fd is same as old"));
674 
675 	/*
676 	 * Save info on the descriptor being overwritten.  We cannot close
677 	 * it without introducing an ownership race for the slot, since we
678 	 * need to drop the filedesc lock to call closef().
679 	 *
680 	 * XXX this duplicates parts of close().
681 	 */
682 	delfp = fdp->fd_ofiles[new];
683 	holdleaders = 0;
684 	if (delfp != NULL) {
685 		if (td->td_proc->p_fdtol != NULL) {
686 			/*
687 			 * Ask fdfree() to sleep to ensure that all relevant
688 			 * process leaders can be traversed in closef().
689 			 */
690 			fdp->fd_holdleaderscount++;
691 			holdleaders = 1;
692 		}
693 	}
694 
695 	/*
696 	 * Duplicate the source descriptor
697 	 */
698 	fdp->fd_ofiles[new] = fp;
699 	fdp->fd_ofileflags[new] = fdp->fd_ofileflags[old] &~ UF_EXCLOSE;
700 	if (new > fdp->fd_lastfile)
701 		fdp->fd_lastfile = new;
702 	*retval = new;
703 
704 	/*
705 	 * If we dup'd over a valid file, we now own the reference to it
706 	 * and must dispose of it using closef() semantics (as if a
707 	 * close() were performed on it).
708 	 *
709 	 * XXX this duplicates parts of close().
710 	 */
711 	if (delfp != NULL) {
712 		knote_fdclose(td, new);
713 		FILEDESC_UNLOCK(fdp);
714 		(void) closef(delfp, td);
715 		if (holdleaders) {
716 			FILEDESC_LOCK_FAST(fdp);
717 			fdp->fd_holdleaderscount--;
718 			if (fdp->fd_holdleaderscount == 0 &&
719 			    fdp->fd_holdleaderswakeup != 0) {
720 				fdp->fd_holdleaderswakeup = 0;
721 				wakeup(&fdp->fd_holdleaderscount);
722 			}
723 			FILEDESC_UNLOCK_FAST(fdp);
724 		}
725 	} else {
726 		FILEDESC_UNLOCK(fdp);
727 	}
728 	return (0);
729 }
730 
731 /*
732  * If sigio is on the list associated with a process or process group,
733  * disable signalling from the device, remove sigio from the list and
734  * free sigio.
735  */
736 void
737 funsetown(struct sigio **sigiop)
738 {
739 	struct sigio *sigio;
740 
741 	SIGIO_LOCK();
742 	sigio = *sigiop;
743 	if (sigio == NULL) {
744 		SIGIO_UNLOCK();
745 		return;
746 	}
747 	*(sigio->sio_myref) = NULL;
748 	if ((sigio)->sio_pgid < 0) {
749 		struct pgrp *pg = (sigio)->sio_pgrp;
750 		PGRP_LOCK(pg);
751 		SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio,
752 			     sigio, sio_pgsigio);
753 		PGRP_UNLOCK(pg);
754 	} else {
755 		struct proc *p = (sigio)->sio_proc;
756 		PROC_LOCK(p);
757 		SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio,
758 			     sigio, sio_pgsigio);
759 		PROC_UNLOCK(p);
760 	}
761 	SIGIO_UNLOCK();
762 	crfree(sigio->sio_ucred);
763 	FREE(sigio, M_SIGIO);
764 }
765 
766 /*
767  * Free a list of sigio structures.
768  * We only need to lock the SIGIO_LOCK because we have made ourselves
769  * inaccessable to callers of fsetown and therefore do not need to lock
770  * the proc or pgrp struct for the list manipulation.
771  */
772 void
773 funsetownlst(struct sigiolst *sigiolst)
774 {
775 	struct proc *p;
776 	struct pgrp *pg;
777 	struct sigio *sigio;
778 
779 	sigio = SLIST_FIRST(sigiolst);
780 	if (sigio == NULL)
781 		return;
782 	p = NULL;
783 	pg = NULL;
784 
785 	/*
786 	 * Every entry of the list should belong
787 	 * to a single proc or pgrp.
788 	 */
789 	if (sigio->sio_pgid < 0) {
790 		pg = sigio->sio_pgrp;
791 		PGRP_LOCK_ASSERT(pg, MA_NOTOWNED);
792 	} else /* if (sigio->sio_pgid > 0) */ {
793 		p = sigio->sio_proc;
794 		PROC_LOCK_ASSERT(p, MA_NOTOWNED);
795 	}
796 
797 	SIGIO_LOCK();
798 	while ((sigio = SLIST_FIRST(sigiolst)) != NULL) {
799 		*(sigio->sio_myref) = NULL;
800 		if (pg != NULL) {
801 			KASSERT(sigio->sio_pgid < 0,
802 			    ("Proc sigio in pgrp sigio list"));
803 			KASSERT(sigio->sio_pgrp == pg,
804 			    ("Bogus pgrp in sigio list"));
805 			PGRP_LOCK(pg);
806 			SLIST_REMOVE(&pg->pg_sigiolst, sigio, sigio,
807 			    sio_pgsigio);
808 			PGRP_UNLOCK(pg);
809 		} else /* if (p != NULL) */ {
810 			KASSERT(sigio->sio_pgid > 0,
811 			    ("Pgrp sigio in proc sigio list"));
812 			KASSERT(sigio->sio_proc == p,
813 			    ("Bogus proc in sigio list"));
814 			PROC_LOCK(p);
815 			SLIST_REMOVE(&p->p_sigiolst, sigio, sigio,
816 			    sio_pgsigio);
817 			PROC_UNLOCK(p);
818 		}
819 		SIGIO_UNLOCK();
820 		crfree(sigio->sio_ucred);
821 		FREE(sigio, M_SIGIO);
822 		SIGIO_LOCK();
823 	}
824 	SIGIO_UNLOCK();
825 }
826 
827 /*
828  * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg).
829  *
830  * After permission checking, add a sigio structure to the sigio list for
831  * the process or process group.
832  */
833 int
834 fsetown(pid_t pgid, struct sigio **sigiop)
835 {
836 	struct proc *proc;
837 	struct pgrp *pgrp;
838 	struct sigio *sigio;
839 	int ret;
840 
841 	if (pgid == 0) {
842 		funsetown(sigiop);
843 		return (0);
844 	}
845 
846 	ret = 0;
847 
848 	/* Allocate and fill in the new sigio out of locks. */
849 	MALLOC(sigio, struct sigio *, sizeof(struct sigio), M_SIGIO, M_WAITOK);
850 	sigio->sio_pgid = pgid;
851 	sigio->sio_ucred = crhold(curthread->td_ucred);
852 	sigio->sio_myref = sigiop;
853 
854 	sx_slock(&proctree_lock);
855 	if (pgid > 0) {
856 		proc = pfind(pgid);
857 		if (proc == NULL) {
858 			ret = ESRCH;
859 			goto fail;
860 		}
861 
862 		/*
863 		 * Policy - Don't allow a process to FSETOWN a process
864 		 * in another session.
865 		 *
866 		 * Remove this test to allow maximum flexibility or
867 		 * restrict FSETOWN to the current process or process
868 		 * group for maximum safety.
869 		 */
870 		PROC_UNLOCK(proc);
871 		if (proc->p_session != curthread->td_proc->p_session) {
872 			ret = EPERM;
873 			goto fail;
874 		}
875 
876 		pgrp = NULL;
877 	} else /* if (pgid < 0) */ {
878 		pgrp = pgfind(-pgid);
879 		if (pgrp == NULL) {
880 			ret = ESRCH;
881 			goto fail;
882 		}
883 		PGRP_UNLOCK(pgrp);
884 
885 		/*
886 		 * Policy - Don't allow a process to FSETOWN a process
887 		 * in another session.
888 		 *
889 		 * Remove this test to allow maximum flexibility or
890 		 * restrict FSETOWN to the current process or process
891 		 * group for maximum safety.
892 		 */
893 		if (pgrp->pg_session != curthread->td_proc->p_session) {
894 			ret = EPERM;
895 			goto fail;
896 		}
897 
898 		proc = NULL;
899 	}
900 	funsetown(sigiop);
901 	if (pgid > 0) {
902 		PROC_LOCK(proc);
903 		/*
904 		 * Since funsetownlst() is called without the proctree
905 		 * locked, we need to check for P_WEXIT.
906 		 * XXX: is ESRCH correct?
907 		 */
908 		if ((proc->p_flag & P_WEXIT) != 0) {
909 			PROC_UNLOCK(proc);
910 			ret = ESRCH;
911 			goto fail;
912 		}
913 		SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio);
914 		sigio->sio_proc = proc;
915 		PROC_UNLOCK(proc);
916 	} else {
917 		PGRP_LOCK(pgrp);
918 		SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio);
919 		sigio->sio_pgrp = pgrp;
920 		PGRP_UNLOCK(pgrp);
921 	}
922 	sx_sunlock(&proctree_lock);
923 	SIGIO_LOCK();
924 	*sigiop = sigio;
925 	SIGIO_UNLOCK();
926 	return (0);
927 
928 fail:
929 	sx_sunlock(&proctree_lock);
930 	crfree(sigio->sio_ucred);
931 	FREE(sigio, M_SIGIO);
932 	return (ret);
933 }
934 
935 /*
936  * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg).
937  */
938 pid_t
939 fgetown(sigiop)
940 	struct sigio **sigiop;
941 {
942 	pid_t pgid;
943 
944 	SIGIO_LOCK();
945 	pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0;
946 	SIGIO_UNLOCK();
947 	return (pgid);
948 }
949 
950 /*
951  * Close a file descriptor.
952  */
953 #ifndef _SYS_SYSPROTO_H_
954 struct close_args {
955 	int     fd;
956 };
957 #endif
958 /*
959  * MPSAFE
960  */
961 /* ARGSUSED */
962 int
963 close(td, uap)
964 	struct thread *td;
965 	struct close_args *uap;
966 {
967 	struct filedesc *fdp;
968 	struct file *fp;
969 	int fd, error;
970 	int holdleaders;
971 
972 	fd = uap->fd;
973 	error = 0;
974 	holdleaders = 0;
975 	fdp = td->td_proc->p_fd;
976 	FILEDESC_LOCK(fdp);
977 	if ((unsigned)fd >= fdp->fd_nfiles ||
978 	    (fp = fdp->fd_ofiles[fd]) == NULL) {
979 		FILEDESC_UNLOCK(fdp);
980 		return (EBADF);
981 	}
982 	fdp->fd_ofiles[fd] = NULL;
983 	fdp->fd_ofileflags[fd] = 0;
984 	fdunused(fdp, fd);
985 	if (td->td_proc->p_fdtol != NULL) {
986 		/*
987 		 * Ask fdfree() to sleep to ensure that all relevant
988 		 * process leaders can be traversed in closef().
989 		 */
990 		fdp->fd_holdleaderscount++;
991 		holdleaders = 1;
992 	}
993 
994 	/*
995 	 * we now hold the fp reference that used to be owned by the descriptor
996 	 * array.
997 	 * We have to unlock the FILEDESC *AFTER* knote_fdclose to prevent a
998 	 * race of the fd getting opened, a knote added, and deleteing a knote
999 	 * for the new fd.
1000 	 */
1001 	knote_fdclose(td, fd);
1002 	FILEDESC_UNLOCK(fdp);
1003 
1004 	error = closef(fp, td);
1005 	if (holdleaders) {
1006 		FILEDESC_LOCK_FAST(fdp);
1007 		fdp->fd_holdleaderscount--;
1008 		if (fdp->fd_holdleaderscount == 0 &&
1009 		    fdp->fd_holdleaderswakeup != 0) {
1010 			fdp->fd_holdleaderswakeup = 0;
1011 			wakeup(&fdp->fd_holdleaderscount);
1012 		}
1013 		FILEDESC_UNLOCK_FAST(fdp);
1014 	}
1015 	return (error);
1016 }
1017 
1018 #if defined(COMPAT_43)
1019 /*
1020  * Return status information about a file descriptor.
1021  */
1022 #ifndef _SYS_SYSPROTO_H_
1023 struct ofstat_args {
1024 	int	fd;
1025 	struct	ostat *sb;
1026 };
1027 #endif
1028 /*
1029  * MPSAFE
1030  */
1031 /* ARGSUSED */
1032 int
1033 ofstat(struct thread *td, struct ofstat_args *uap)
1034 {
1035 	struct ostat oub;
1036 	struct stat ub;
1037 	int error;
1038 
1039 	error = kern_fstat(td, uap->fd, &ub);
1040 	if (error == 0) {
1041 		cvtstat(&ub, &oub);
1042 		error = copyout(&oub, uap->sb, sizeof(oub));
1043 	}
1044 	return (error);
1045 }
1046 #endif /* COMPAT_43 */
1047 
1048 /*
1049  * Return status information about a file descriptor.
1050  */
1051 #ifndef _SYS_SYSPROTO_H_
1052 struct fstat_args {
1053 	int	fd;
1054 	struct	stat *sb;
1055 };
1056 #endif
1057 /*
1058  * MPSAFE
1059  */
1060 /* ARGSUSED */
1061 int
1062 fstat(struct thread *td, struct fstat_args *uap)
1063 {
1064 	struct stat ub;
1065 	int error;
1066 
1067 	error = kern_fstat(td, uap->fd, &ub);
1068 	if (error == 0)
1069 		error = copyout(&ub, uap->sb, sizeof(ub));
1070 	return (error);
1071 }
1072 
1073 int
1074 kern_fstat(struct thread *td, int fd, struct stat *sbp)
1075 {
1076 	struct file *fp;
1077 	int error;
1078 
1079 	if ((error = fget(td, fd, &fp)) != 0)
1080 		return (error);
1081 	error = fo_stat(fp, sbp, td->td_ucred, td);
1082 	fdrop(fp, td);
1083 	return (error);
1084 }
1085 
1086 /*
1087  * Return status information about a file descriptor.
1088  */
1089 #ifndef _SYS_SYSPROTO_H_
1090 struct nfstat_args {
1091 	int	fd;
1092 	struct	nstat *sb;
1093 };
1094 #endif
1095 /*
1096  * MPSAFE
1097  */
1098 /* ARGSUSED */
1099 int
1100 nfstat(struct thread *td, struct nfstat_args *uap)
1101 {
1102 	struct nstat nub;
1103 	struct stat ub;
1104 	int error;
1105 
1106 	error = kern_fstat(td, uap->fd, &ub);
1107 	if (error == 0) {
1108 		cvtnstat(&ub, &nub);
1109 		error = copyout(&nub, uap->sb, sizeof(nub));
1110 	}
1111 	return (error);
1112 }
1113 
1114 /*
1115  * Return pathconf information about a file descriptor.
1116  */
1117 #ifndef _SYS_SYSPROTO_H_
1118 struct fpathconf_args {
1119 	int	fd;
1120 	int	name;
1121 };
1122 #endif
1123 /*
1124  * MPSAFE
1125  */
1126 /* ARGSUSED */
1127 int
1128 fpathconf(struct thread *td, struct fpathconf_args *uap)
1129 {
1130 	struct file *fp;
1131 	struct vnode *vp;
1132 	int error;
1133 
1134 	if ((error = fget(td, uap->fd, &fp)) != 0)
1135 		return (error);
1136 
1137 	/* If asynchronous I/O is available, it works for all descriptors. */
1138 	if (uap->name == _PC_ASYNC_IO) {
1139 		td->td_retval[0] = async_io_version;
1140 		goto out;
1141 	}
1142 	vp = fp->f_vnode;
1143 	if (vp != NULL) {
1144 		int vfslocked;
1145 		vfslocked = VFS_LOCK_GIANT(vp->v_mount);
1146 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
1147 		error = VOP_PATHCONF(vp, uap->name, td->td_retval);
1148 		VOP_UNLOCK(vp, 0, td);
1149 		VFS_UNLOCK_GIANT(vfslocked);
1150 	} else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) {
1151 		if (uap->name != _PC_PIPE_BUF) {
1152 			error = EINVAL;
1153 		} else {
1154 			td->td_retval[0] = PIPE_BUF;
1155 		error = 0;
1156 		}
1157 	} else {
1158 		error = EOPNOTSUPP;
1159 	}
1160 out:
1161 	fdrop(fp, td);
1162 	return (error);
1163 }
1164 
1165 /*
1166  * Grow the file table to accomodate (at least) nfd descriptors.  This may
1167  * block and drop the filedesc lock, but it will reacquire it before
1168  * returing.
1169  */
1170 static void
1171 fdgrowtable(struct filedesc *fdp, int nfd)
1172 {
1173 	struct file **ntable;
1174 	char *nfileflags;
1175 	int nnfiles, onfiles;
1176 	NDSLOTTYPE *nmap;
1177 
1178 	FILEDESC_LOCK_ASSERT(fdp, MA_OWNED);
1179 
1180 	KASSERT(fdp->fd_nfiles > 0,
1181 	    ("zero-length file table"));
1182 
1183 	/* compute the size of the new table */
1184 	onfiles = fdp->fd_nfiles;
1185 	nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */
1186 	if (nnfiles <= onfiles)
1187 		/* the table is already large enough */
1188 		return;
1189 
1190 	/* allocate a new table and (if required) new bitmaps */
1191 	FILEDESC_UNLOCK(fdp);
1192 	MALLOC(ntable, struct file **, nnfiles * OFILESIZE,
1193 	    M_FILEDESC, M_ZERO | M_WAITOK);
1194 	nfileflags = (char *)&ntable[nnfiles];
1195 	if (NDSLOTS(nnfiles) > NDSLOTS(onfiles))
1196 		MALLOC(nmap, NDSLOTTYPE *, NDSLOTS(nnfiles) * NDSLOTSIZE,
1197 		    M_FILEDESC, M_ZERO | M_WAITOK);
1198 	else
1199 		nmap = NULL;
1200 	FILEDESC_LOCK(fdp);
1201 
1202 	/*
1203 	 * We now have new tables ready to go.  Since we dropped the
1204 	 * filedesc lock to call malloc(), watch out for a race.
1205 	 */
1206 	onfiles = fdp->fd_nfiles;
1207 	if (onfiles >= nnfiles) {
1208 		/* we lost the race, but that's OK */
1209 		free(ntable, M_FILEDESC);
1210 		if (nmap != NULL)
1211 			free(nmap, M_FILEDESC);
1212 		return;
1213 	}
1214 	bcopy(fdp->fd_ofiles, ntable, onfiles * sizeof(*ntable));
1215 	bcopy(fdp->fd_ofileflags, nfileflags, onfiles);
1216 	if (onfiles > NDFILE)
1217 		free(fdp->fd_ofiles, M_FILEDESC);
1218 	fdp->fd_ofiles = ntable;
1219 	fdp->fd_ofileflags = nfileflags;
1220 	if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) {
1221 		bcopy(fdp->fd_map, nmap, NDSLOTS(onfiles) * sizeof(*nmap));
1222 		if (NDSLOTS(onfiles) > NDSLOTS(NDFILE))
1223 			free(fdp->fd_map, M_FILEDESC);
1224 		fdp->fd_map = nmap;
1225 	}
1226 	fdp->fd_nfiles = nnfiles;
1227 }
1228 
1229 /*
1230  * Allocate a file descriptor for the process.
1231  */
1232 int
1233 fdalloc(struct thread *td, int minfd, int *result)
1234 {
1235 	struct proc *p = td->td_proc;
1236 	struct filedesc *fdp = p->p_fd;
1237 	int fd = -1, maxfd;
1238 
1239 	FILEDESC_LOCK_ASSERT(fdp, MA_OWNED);
1240 
1241 	PROC_LOCK(p);
1242 	maxfd = min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc);
1243 	PROC_UNLOCK(p);
1244 
1245 	/*
1246 	 * Search the bitmap for a free descriptor.  If none is found, try
1247 	 * to grow the file table.  Keep at it until we either get a file
1248 	 * descriptor or run into process or system limits; fdgrowtable()
1249 	 * may drop the filedesc lock, so we're in a race.
1250 	 */
1251 	for (;;) {
1252 		fd = fd_first_free(fdp, minfd, fdp->fd_nfiles);
1253 		if (fd >= maxfd)
1254 			return (EMFILE);
1255 		if (fd < fdp->fd_nfiles)
1256 			break;
1257 		fdgrowtable(fdp, min(fdp->fd_nfiles * 2, maxfd));
1258 	}
1259 
1260 	/*
1261 	 * Perform some sanity checks, then mark the file descriptor as
1262 	 * used and return it to the caller.
1263 	 */
1264 	KASSERT(!fdisused(fdp, fd),
1265 	    ("fd_first_free() returned non-free descriptor"));
1266 	KASSERT(fdp->fd_ofiles[fd] == NULL,
1267 	    ("free descriptor isn't"));
1268 	fdp->fd_ofileflags[fd] = 0; /* XXX needed? */
1269 	fdused(fdp, fd);
1270 	fdp->fd_freefile = fd_first_free(fdp, fd, fdp->fd_nfiles);
1271 	*result = fd;
1272 	return (0);
1273 }
1274 
1275 /*
1276  * Check to see whether n user file descriptors
1277  * are available to the process p.
1278  */
1279 int
1280 fdavail(struct thread *td, int n)
1281 {
1282 	struct proc *p = td->td_proc;
1283 	struct filedesc *fdp = td->td_proc->p_fd;
1284 	struct file **fpp;
1285 	int i, lim, last;
1286 
1287 	FILEDESC_LOCK_ASSERT(fdp, MA_OWNED);
1288 
1289 	PROC_LOCK(p);
1290 	lim = min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc);
1291 	PROC_UNLOCK(p);
1292 	if ((i = lim - fdp->fd_nfiles) > 0 && (n -= i) <= 0)
1293 		return (1);
1294 	last = min(fdp->fd_nfiles, lim);
1295 	fpp = &fdp->fd_ofiles[fdp->fd_freefile];
1296 	for (i = last - fdp->fd_freefile; --i >= 0; fpp++) {
1297 		if (*fpp == NULL && --n <= 0)
1298 			return (1);
1299 	}
1300 	return (0);
1301 }
1302 
1303 /*
1304  * Create a new open file structure and allocate
1305  * a file decriptor for the process that refers to it.
1306  * We add one reference to the file for the descriptor table
1307  * and one reference for resultfp. This is to prevent us being
1308  * prempted and the entry in the descriptor table closed after
1309  * we release the FILEDESC lock.
1310  */
1311 int
1312 falloc(struct thread *td, struct file **resultfp, int *resultfd)
1313 {
1314 	struct proc *p = td->td_proc;
1315 	struct file *fp, *fq;
1316 	int error, i;
1317 	int maxuserfiles = maxfiles - (maxfiles / 20);
1318 	static struct timeval lastfail;
1319 	static int curfail;
1320 
1321 	fp = uma_zalloc(file_zone, M_WAITOK | M_ZERO);
1322 	sx_xlock(&filelist_lock);
1323 	if ((openfiles >= maxuserfiles && (td->td_ucred->cr_ruid != 0 ||
1324 	   jailed(td->td_ucred))) || openfiles >= maxfiles) {
1325 		if (ppsratecheck(&lastfail, &curfail, 1)) {
1326 			printf("kern.maxfiles limit exceeded by uid %i, please see tuning(7).\n",
1327 				td->td_ucred->cr_ruid);
1328 		}
1329 		sx_xunlock(&filelist_lock);
1330 		uma_zfree(file_zone, fp);
1331 		return (ENFILE);
1332 	}
1333 	openfiles++;
1334 
1335 	/*
1336 	 * If the process has file descriptor zero open, add the new file
1337 	 * descriptor to the list of open files at that point, otherwise
1338 	 * put it at the front of the list of open files.
1339 	 */
1340 	fp->f_mtxp = mtx_pool_alloc(mtxpool_sleep);
1341 	fp->f_count = 1;
1342 	if (resultfp)
1343 		fp->f_count++;
1344 	fp->f_cred = crhold(td->td_ucred);
1345 	fp->f_ops = &badfileops;
1346 	fp->f_data = NULL;
1347 	fp->f_vnode = NULL;
1348 	FILEDESC_LOCK(p->p_fd);
1349 	if ((fq = p->p_fd->fd_ofiles[0])) {
1350 		LIST_INSERT_AFTER(fq, fp, f_list);
1351 	} else {
1352 		LIST_INSERT_HEAD(&filehead, fp, f_list);
1353 	}
1354 	sx_xunlock(&filelist_lock);
1355 	if ((error = fdalloc(td, 0, &i))) {
1356 		FILEDESC_UNLOCK(p->p_fd);
1357 		fdrop(fp, td);
1358 		if (resultfp)
1359 			fdrop(fp, td);
1360 		return (error);
1361 	}
1362 	p->p_fd->fd_ofiles[i] = fp;
1363 	FILEDESC_UNLOCK(p->p_fd);
1364 	if (resultfp)
1365 		*resultfp = fp;
1366 	if (resultfd)
1367 		*resultfd = i;
1368 	return (0);
1369 }
1370 
1371 /*
1372  * Build a new filedesc structure from another.
1373  * Copy the current, root, and jail root vnode references.
1374  */
1375 struct filedesc *
1376 fdinit(struct filedesc *fdp)
1377 {
1378 	struct filedesc0 *newfdp;
1379 
1380 	newfdp = malloc(sizeof *newfdp, M_FILEDESC, M_WAITOK | M_ZERO);
1381 	mtx_init(&newfdp->fd_fd.fd_mtx, FILEDESC_LOCK_DESC, NULL, MTX_DEF);
1382 	if (fdp != NULL) {
1383 		FILEDESC_LOCK(fdp);
1384 		newfdp->fd_fd.fd_cdir = fdp->fd_cdir;
1385 		if (newfdp->fd_fd.fd_cdir)
1386 			VREF(newfdp->fd_fd.fd_cdir);
1387 		newfdp->fd_fd.fd_rdir = fdp->fd_rdir;
1388 		if (newfdp->fd_fd.fd_rdir)
1389 			VREF(newfdp->fd_fd.fd_rdir);
1390 		newfdp->fd_fd.fd_jdir = fdp->fd_jdir;
1391 		if (newfdp->fd_fd.fd_jdir)
1392 			VREF(newfdp->fd_fd.fd_jdir);
1393 		FILEDESC_UNLOCK(fdp);
1394 	}
1395 
1396 	/* Create the file descriptor table. */
1397 	newfdp->fd_fd.fd_refcnt = 1;
1398 	newfdp->fd_fd.fd_holdcnt = 1;
1399 	newfdp->fd_fd.fd_cmask = CMASK;
1400 	newfdp->fd_fd.fd_ofiles = newfdp->fd_dfiles;
1401 	newfdp->fd_fd.fd_ofileflags = newfdp->fd_dfileflags;
1402 	newfdp->fd_fd.fd_nfiles = NDFILE;
1403 	newfdp->fd_fd.fd_map = newfdp->fd_dmap;
1404 	return (&newfdp->fd_fd);
1405 }
1406 
1407 static struct filedesc *
1408 fdhold(struct proc *p)
1409 {
1410 	struct filedesc *fdp;
1411 
1412 	mtx_lock(&fdesc_mtx);
1413 	fdp = p->p_fd;
1414 	if (fdp != NULL)
1415 		fdp->fd_holdcnt++;
1416 	mtx_unlock(&fdesc_mtx);
1417 	return (fdp);
1418 }
1419 
1420 static void
1421 fddrop(struct filedesc *fdp)
1422 {
1423 	int i;
1424 
1425 	mtx_lock(&fdesc_mtx);
1426 	i = --fdp->fd_holdcnt;
1427 	mtx_unlock(&fdesc_mtx);
1428 	if (i > 0)
1429 		return;
1430 
1431 	mtx_destroy(&fdp->fd_mtx);
1432 	FREE(fdp, M_FILEDESC);
1433 }
1434 
1435 /*
1436  * Share a filedesc structure.
1437  */
1438 struct filedesc *
1439 fdshare(struct filedesc *fdp)
1440 {
1441 	FILEDESC_LOCK_FAST(fdp);
1442 	fdp->fd_refcnt++;
1443 	FILEDESC_UNLOCK_FAST(fdp);
1444 	return (fdp);
1445 }
1446 
1447 /*
1448  * Unshare a filedesc structure, if necessary by making a copy
1449  */
1450 void
1451 fdunshare(struct proc *p, struct thread *td)
1452 {
1453 
1454 	FILEDESC_LOCK_FAST(p->p_fd);
1455 	if (p->p_fd->fd_refcnt > 1) {
1456 		struct filedesc *tmp;
1457 
1458 		FILEDESC_UNLOCK_FAST(p->p_fd);
1459 		tmp = fdcopy(p->p_fd);
1460 		fdfree(td);
1461 		p->p_fd = tmp;
1462 	} else
1463 		FILEDESC_UNLOCK_FAST(p->p_fd);
1464 }
1465 
1466 /*
1467  * Copy a filedesc structure.
1468  * A NULL pointer in returns a NULL reference, this is to ease callers,
1469  * not catch errors.
1470  */
1471 struct filedesc *
1472 fdcopy(struct filedesc *fdp)
1473 {
1474 	struct filedesc *newfdp;
1475 	int i;
1476 
1477 	/* Certain daemons might not have file descriptors. */
1478 	if (fdp == NULL)
1479 		return (NULL);
1480 
1481 	newfdp = fdinit(fdp);
1482 	FILEDESC_LOCK_FAST(fdp);
1483 	while (fdp->fd_lastfile >= newfdp->fd_nfiles) {
1484 		FILEDESC_UNLOCK_FAST(fdp);
1485 		FILEDESC_LOCK(newfdp);
1486 		fdgrowtable(newfdp, fdp->fd_lastfile + 1);
1487 		FILEDESC_UNLOCK(newfdp);
1488 		FILEDESC_LOCK_FAST(fdp);
1489 	}
1490 	/* copy everything except kqueue descriptors */
1491 	newfdp->fd_freefile = -1;
1492 	for (i = 0; i <= fdp->fd_lastfile; ++i) {
1493 		if (fdisused(fdp, i) &&
1494 		    fdp->fd_ofiles[i]->f_type != DTYPE_KQUEUE) {
1495 			newfdp->fd_ofiles[i] = fdp->fd_ofiles[i];
1496 			newfdp->fd_ofileflags[i] = fdp->fd_ofileflags[i];
1497 			fhold(newfdp->fd_ofiles[i]);
1498 			newfdp->fd_lastfile = i;
1499 		} else {
1500 			if (newfdp->fd_freefile == -1)
1501 				newfdp->fd_freefile = i;
1502 		}
1503 	}
1504 	FILEDESC_UNLOCK_FAST(fdp);
1505 	FILEDESC_LOCK(newfdp);
1506 	for (i = 0; i <= newfdp->fd_lastfile; ++i)
1507 		if (newfdp->fd_ofiles[i] != NULL)
1508 			fdused(newfdp, i);
1509 	FILEDESC_UNLOCK(newfdp);
1510 	FILEDESC_LOCK_FAST(fdp);
1511 	if (newfdp->fd_freefile == -1)
1512 		newfdp->fd_freefile = i;
1513 	newfdp->fd_cmask = fdp->fd_cmask;
1514 	FILEDESC_UNLOCK_FAST(fdp);
1515 	return (newfdp);
1516 }
1517 
1518 /*
1519  * Release a filedesc structure.
1520  */
1521 void
1522 fdfree(struct thread *td)
1523 {
1524 	struct filedesc *fdp;
1525 	struct file **fpp;
1526 	int i;
1527 	struct filedesc_to_leader *fdtol;
1528 	struct file *fp;
1529 	struct vnode *vp;
1530 	struct flock lf;
1531 
1532 	/* Certain daemons might not have file descriptors. */
1533 	fdp = td->td_proc->p_fd;
1534 	if (fdp == NULL)
1535 		return;
1536 
1537 	/* Check for special need to clear POSIX style locks */
1538 	fdtol = td->td_proc->p_fdtol;
1539 	if (fdtol != NULL) {
1540 		FILEDESC_LOCK(fdp);
1541 		KASSERT(fdtol->fdl_refcount > 0,
1542 			("filedesc_to_refcount botch: fdl_refcount=%d",
1543 			 fdtol->fdl_refcount));
1544 		if (fdtol->fdl_refcount == 1 &&
1545 		    (td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
1546 			i = 0;
1547 			fpp = fdp->fd_ofiles;
1548 			for (i = 0, fpp = fdp->fd_ofiles;
1549 			     i <= fdp->fd_lastfile;
1550 			     i++, fpp++) {
1551 				if (*fpp == NULL ||
1552 				    (*fpp)->f_type != DTYPE_VNODE)
1553 					continue;
1554 				fp = *fpp;
1555 				fhold(fp);
1556 				FILEDESC_UNLOCK(fdp);
1557 				lf.l_whence = SEEK_SET;
1558 				lf.l_start = 0;
1559 				lf.l_len = 0;
1560 				lf.l_type = F_UNLCK;
1561 				vp = fp->f_vnode;
1562 				VFS_ASSERT_GIANT(vp->v_mount);
1563 				(void) VOP_ADVLOCK(vp,
1564 						   (caddr_t)td->td_proc->
1565 						   p_leader,
1566 						   F_UNLCK,
1567 						   &lf,
1568 						   F_POSIX);
1569 				FILEDESC_LOCK(fdp);
1570 				fdrop(fp, td);
1571 				fpp = fdp->fd_ofiles + i;
1572 			}
1573 		}
1574 	retry:
1575 		if (fdtol->fdl_refcount == 1) {
1576 			if (fdp->fd_holdleaderscount > 0 &&
1577 			    (td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
1578 				/*
1579 				 * close() or do_dup() has cleared a reference
1580 				 * in a shared file descriptor table.
1581 				 */
1582 				fdp->fd_holdleaderswakeup = 1;
1583 				msleep(&fdp->fd_holdleaderscount, &fdp->fd_mtx,
1584 				       PLOCK, "fdlhold", 0);
1585 				goto retry;
1586 			}
1587 			if (fdtol->fdl_holdcount > 0) {
1588 				/*
1589 				 * Ensure that fdtol->fdl_leader
1590 				 * remains valid in closef().
1591 				 */
1592 				fdtol->fdl_wakeup = 1;
1593 				msleep(fdtol, &fdp->fd_mtx,
1594 				       PLOCK, "fdlhold", 0);
1595 				goto retry;
1596 			}
1597 		}
1598 		fdtol->fdl_refcount--;
1599 		if (fdtol->fdl_refcount == 0 &&
1600 		    fdtol->fdl_holdcount == 0) {
1601 			fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
1602 			fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
1603 		} else
1604 			fdtol = NULL;
1605 		td->td_proc->p_fdtol = NULL;
1606 		FILEDESC_UNLOCK(fdp);
1607 		if (fdtol != NULL)
1608 			FREE(fdtol, M_FILEDESC_TO_LEADER);
1609 	}
1610 	FILEDESC_LOCK_FAST(fdp);
1611 	i = --fdp->fd_refcnt;
1612 	FILEDESC_UNLOCK_FAST(fdp);
1613 	if (i > 0)
1614 		return;
1615 	/*
1616 	 * We are the last reference to the structure, so we can
1617 	 * safely assume it will not change out from under us.
1618 	 */
1619 	fpp = fdp->fd_ofiles;
1620 	for (i = fdp->fd_lastfile; i-- >= 0; fpp++) {
1621 		if (*fpp)
1622 			(void) closef(*fpp, td);
1623 	}
1624 	FILEDESC_LOCK(fdp);
1625 
1626 	/* XXX This should happen earlier. */
1627 	mtx_lock(&fdesc_mtx);
1628 	td->td_proc->p_fd = NULL;
1629 	mtx_unlock(&fdesc_mtx);
1630 
1631 	if (fdp->fd_nfiles > NDFILE)
1632 		FREE(fdp->fd_ofiles, M_FILEDESC);
1633 	if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE))
1634 		FREE(fdp->fd_map, M_FILEDESC);
1635 
1636 	fdp->fd_nfiles = 0;
1637 
1638 	if (fdp->fd_cdir)
1639 		vrele(fdp->fd_cdir);
1640 	fdp->fd_cdir = NULL;
1641 	if (fdp->fd_rdir)
1642 		vrele(fdp->fd_rdir);
1643 	fdp->fd_rdir = NULL;
1644 	if (fdp->fd_jdir)
1645 		vrele(fdp->fd_jdir);
1646 	fdp->fd_jdir = NULL;
1647 
1648 	FILEDESC_UNLOCK(fdp);
1649 
1650 	fddrop(fdp);
1651 }
1652 
1653 /*
1654  * For setugid programs, we don't want to people to use that setugidness
1655  * to generate error messages which write to a file which otherwise would
1656  * otherwise be off-limits to the process.  We check for filesystems where
1657  * the vnode can change out from under us after execve (like [lin]procfs).
1658  *
1659  * Since setugidsafety calls this only for fd 0, 1 and 2, this check is
1660  * sufficient.  We also don't for check setugidness since we know we are.
1661  */
1662 static int
1663 is_unsafe(struct file *fp)
1664 {
1665 	if (fp->f_type == DTYPE_VNODE) {
1666 		struct vnode *vp = fp->f_vnode;
1667 
1668 		if ((vp->v_vflag & VV_PROCDEP) != 0)
1669 			return (1);
1670 	}
1671 	return (0);
1672 }
1673 
1674 /*
1675  * Make this setguid thing safe, if at all possible.
1676  */
1677 void
1678 setugidsafety(struct thread *td)
1679 {
1680 	struct filedesc *fdp;
1681 	int i;
1682 
1683 	/* Certain daemons might not have file descriptors. */
1684 	fdp = td->td_proc->p_fd;
1685 	if (fdp == NULL)
1686 		return;
1687 
1688 	/*
1689 	 * Note: fdp->fd_ofiles may be reallocated out from under us while
1690 	 * we are blocked in a close.  Be careful!
1691 	 */
1692 	FILEDESC_LOCK(fdp);
1693 	for (i = 0; i <= fdp->fd_lastfile; i++) {
1694 		if (i > 2)
1695 			break;
1696 		if (fdp->fd_ofiles[i] && is_unsafe(fdp->fd_ofiles[i])) {
1697 			struct file *fp;
1698 
1699 			knote_fdclose(td, i);
1700 			/*
1701 			 * NULL-out descriptor prior to close to avoid
1702 			 * a race while close blocks.
1703 			 */
1704 			fp = fdp->fd_ofiles[i];
1705 			fdp->fd_ofiles[i] = NULL;
1706 			fdp->fd_ofileflags[i] = 0;
1707 			fdunused(fdp, i);
1708 			FILEDESC_UNLOCK(fdp);
1709 			(void) closef(fp, td);
1710 			FILEDESC_LOCK(fdp);
1711 		}
1712 	}
1713 	FILEDESC_UNLOCK(fdp);
1714 }
1715 
1716 void
1717 fdclose(struct filedesc *fdp, struct file *fp, int idx, struct thread *td)
1718 {
1719 
1720 	FILEDESC_LOCK(fdp);
1721 	if (fdp->fd_ofiles[idx] == fp) {
1722 		fdp->fd_ofiles[idx] = NULL;
1723 		fdunused(fdp, idx);
1724 		FILEDESC_UNLOCK(fdp);
1725 		fdrop(fp, td);
1726 	} else {
1727 		FILEDESC_UNLOCK(fdp);
1728 	}
1729 }
1730 
1731 /*
1732  * Close any files on exec?
1733  */
1734 void
1735 fdcloseexec(struct thread *td)
1736 {
1737 	struct filedesc *fdp;
1738 	int i;
1739 
1740 	/* Certain daemons might not have file descriptors. */
1741 	fdp = td->td_proc->p_fd;
1742 	if (fdp == NULL)
1743 		return;
1744 
1745 	FILEDESC_LOCK(fdp);
1746 
1747 	/*
1748 	 * We cannot cache fd_ofiles or fd_ofileflags since operations
1749 	 * may block and rip them out from under us.
1750 	 */
1751 	for (i = 0; i <= fdp->fd_lastfile; i++) {
1752 		if (fdp->fd_ofiles[i] != NULL &&
1753 		    (fdp->fd_ofileflags[i] & UF_EXCLOSE)) {
1754 			struct file *fp;
1755 
1756 			knote_fdclose(td, i);
1757 			/*
1758 			 * NULL-out descriptor prior to close to avoid
1759 			 * a race while close blocks.
1760 			 */
1761 			fp = fdp->fd_ofiles[i];
1762 			fdp->fd_ofiles[i] = NULL;
1763 			fdp->fd_ofileflags[i] = 0;
1764 			fdunused(fdp, i);
1765 			FILEDESC_UNLOCK(fdp);
1766 			(void) closef(fp, td);
1767 			FILEDESC_LOCK(fdp);
1768 		}
1769 	}
1770 	FILEDESC_UNLOCK(fdp);
1771 }
1772 
1773 /*
1774  * It is unsafe for set[ug]id processes to be started with file
1775  * descriptors 0..2 closed, as these descriptors are given implicit
1776  * significance in the Standard C library.  fdcheckstd() will create a
1777  * descriptor referencing /dev/null for each of stdin, stdout, and
1778  * stderr that is not already open.
1779  */
1780 int
1781 fdcheckstd(struct thread *td)
1782 {
1783 	struct nameidata nd;
1784 	struct filedesc *fdp;
1785 	struct file *fp;
1786 	register_t retval;
1787 	int fd, i, error, flags, devnull;
1788 
1789 	GIANT_REQUIRED;		/* VFS */
1790 
1791 	fdp = td->td_proc->p_fd;
1792 	if (fdp == NULL)
1793 		return (0);
1794 	KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
1795 	devnull = -1;
1796 	error = 0;
1797 	for (i = 0; i < 3; i++) {
1798 		if (fdp->fd_ofiles[i] != NULL)
1799 			continue;
1800 		if (devnull < 0) {
1801 			error = falloc(td, &fp, &fd);
1802 			if (error != 0)
1803 				break;
1804 			/* Note extra ref on `fp' held for us by falloc(). */
1805 			KASSERT(fd == i, ("oof, we didn't get our fd"));
1806 			NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, "/dev/null",
1807 			    td);
1808 			flags = FREAD | FWRITE;
1809 			error = vn_open(&nd, &flags, 0, -1);
1810 			if (error != 0) {
1811 				/*
1812 				 * Someone may have closed the entry in the
1813 				 * file descriptor table, so check it hasn't
1814 				 * changed before dropping the reference count.
1815 				 */
1816 				FILEDESC_LOCK(fdp);
1817 				KASSERT(fdp->fd_ofiles[fd] == fp,
1818 				    ("table not shared, how did it change?"));
1819 				fdp->fd_ofiles[fd] = NULL;
1820 				fdunused(fdp, fd);
1821 				FILEDESC_UNLOCK(fdp);
1822 				fdrop(fp, td);
1823 				fdrop(fp, td);
1824 				break;
1825 			}
1826 			NDFREE(&nd, NDF_ONLY_PNBUF);
1827 			fp->f_flag = flags;
1828 			fp->f_vnode = nd.ni_vp;
1829 			if (fp->f_data == NULL)
1830 				fp->f_data = nd.ni_vp;
1831 			if (fp->f_ops == &badfileops)
1832 				fp->f_ops = &vnops;
1833 			fp->f_type = DTYPE_VNODE;
1834 			VOP_UNLOCK(nd.ni_vp, 0, td);
1835 			devnull = fd;
1836 			fdrop(fp, td);
1837 		} else {
1838 			error = do_dup(td, DUP_FIXED, devnull, i, &retval);
1839 			if (error != 0)
1840 				break;
1841 		}
1842 	}
1843 	return (error);
1844 }
1845 
1846 /*
1847  * Internal form of close.
1848  * Decrement reference count on file structure.
1849  * Note: td may be NULL when closing a file that was being passed in a
1850  * message.
1851  *
1852  * XXXRW: Giant is not required for the caller, but often will be held; this
1853  * makes it moderately likely the Giant will be recursed in the VFS case.
1854  */
1855 int
1856 closef(struct file *fp, struct thread *td)
1857 {
1858 	struct vnode *vp;
1859 	struct flock lf;
1860 	struct filedesc_to_leader *fdtol;
1861 	struct filedesc *fdp;
1862 
1863 	/*
1864 	 * POSIX record locking dictates that any close releases ALL
1865 	 * locks owned by this process.  This is handled by setting
1866 	 * a flag in the unlock to free ONLY locks obeying POSIX
1867 	 * semantics, and not to free BSD-style file locks.
1868 	 * If the descriptor was in a message, POSIX-style locks
1869 	 * aren't passed with the descriptor.
1870 	 */
1871 	if (fp->f_type == DTYPE_VNODE) {
1872 		int vfslocked;
1873 
1874 		vp = fp->f_vnode;
1875 		vfslocked = VFS_LOCK_GIANT(vp->v_mount);
1876 		if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
1877 			lf.l_whence = SEEK_SET;
1878 			lf.l_start = 0;
1879 			lf.l_len = 0;
1880 			lf.l_type = F_UNLCK;
1881 			(void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader,
1882 					   F_UNLCK, &lf, F_POSIX);
1883 		}
1884 		fdtol = td->td_proc->p_fdtol;
1885 		if (fdtol != NULL) {
1886 			/*
1887 			 * Handle special case where file descriptor table
1888 			 * is shared between multiple process leaders.
1889 			 */
1890 			fdp = td->td_proc->p_fd;
1891 			FILEDESC_LOCK(fdp);
1892 			for (fdtol = fdtol->fdl_next;
1893 			     fdtol != td->td_proc->p_fdtol;
1894 			     fdtol = fdtol->fdl_next) {
1895 				if ((fdtol->fdl_leader->p_flag &
1896 				     P_ADVLOCK) == 0)
1897 					continue;
1898 				fdtol->fdl_holdcount++;
1899 				FILEDESC_UNLOCK(fdp);
1900 				lf.l_whence = SEEK_SET;
1901 				lf.l_start = 0;
1902 				lf.l_len = 0;
1903 				lf.l_type = F_UNLCK;
1904 				vp = fp->f_vnode;
1905 				(void) VOP_ADVLOCK(vp,
1906 						   (caddr_t)fdtol->fdl_leader,
1907 						   F_UNLCK, &lf, F_POSIX);
1908 				FILEDESC_LOCK(fdp);
1909 				fdtol->fdl_holdcount--;
1910 				if (fdtol->fdl_holdcount == 0 &&
1911 				    fdtol->fdl_wakeup != 0) {
1912 					fdtol->fdl_wakeup = 0;
1913 					wakeup(fdtol);
1914 				}
1915 			}
1916 			FILEDESC_UNLOCK(fdp);
1917 		}
1918 		VFS_UNLOCK_GIANT(vfslocked);
1919 	}
1920 	return (fdrop(fp, td));
1921 }
1922 
1923 /*
1924  * Extract the file pointer associated with the specified descriptor for
1925  * the current user process.
1926  *
1927  * If the descriptor doesn't exist, EBADF is returned.
1928  *
1929  * If the descriptor exists but doesn't match 'flags' then
1930  * return EBADF for read attempts and EINVAL for write attempts.
1931  *
1932  * If 'hold' is set (non-zero) the file's refcount will be bumped on return.
1933  * It should be droped with fdrop().
1934  * If it is not set, then the refcount will not be bumped however the
1935  * thread's filedesc struct will be returned locked (for fgetsock).
1936  *
1937  * If an error occured the non-zero error is returned and *fpp is set to NULL.
1938  * Otherwise *fpp is set and zero is returned.
1939  */
1940 static __inline int
1941 _fget(struct thread *td, int fd, struct file **fpp, int flags, int hold)
1942 {
1943 	struct filedesc *fdp;
1944 	struct file *fp;
1945 
1946 	*fpp = NULL;
1947 	if (td == NULL || (fdp = td->td_proc->p_fd) == NULL)
1948 		return (EBADF);
1949 	FILEDESC_LOCK(fdp);
1950 	if ((fp = fget_locked(fdp, fd)) == NULL || fp->f_ops == &badfileops) {
1951 		FILEDESC_UNLOCK(fdp);
1952 		return (EBADF);
1953 	}
1954 
1955 	/*
1956 	 * Note: FREAD failures returns EBADF to maintain backwards
1957 	 * compatibility with what routines returned before.
1958 	 *
1959 	 * Only one flag, or 0, may be specified.
1960 	 */
1961 	if (flags == FREAD && (fp->f_flag & FREAD) == 0) {
1962 		FILEDESC_UNLOCK(fdp);
1963 		return (EBADF);
1964 	}
1965 	if (flags == FWRITE && (fp->f_flag & FWRITE) == 0) {
1966 		FILEDESC_UNLOCK(fdp);
1967 		return (EINVAL);
1968 	}
1969 	if (hold) {
1970 		fhold(fp);
1971 		FILEDESC_UNLOCK(fdp);
1972 	}
1973 	*fpp = fp;
1974 	return (0);
1975 }
1976 
1977 int
1978 fget(struct thread *td, int fd, struct file **fpp)
1979 {
1980 
1981 	return(_fget(td, fd, fpp, 0, 1));
1982 }
1983 
1984 int
1985 fget_read(struct thread *td, int fd, struct file **fpp)
1986 {
1987 
1988 	return(_fget(td, fd, fpp, FREAD, 1));
1989 }
1990 
1991 int
1992 fget_write(struct thread *td, int fd, struct file **fpp)
1993 {
1994 
1995 	return(_fget(td, fd, fpp, FWRITE, 1));
1996 }
1997 
1998 /*
1999  * Like fget() but loads the underlying vnode, or returns an error if
2000  * the descriptor does not represent a vnode.  Note that pipes use vnodes
2001  * but never have VM objects.  The returned vnode will be vref()d.
2002  *
2003  * XXX: what about the unused flags ?
2004  */
2005 static __inline int
2006 _fgetvp(struct thread *td, int fd, struct vnode **vpp, int flags)
2007 {
2008 	struct file *fp;
2009 	int error;
2010 
2011 	*vpp = NULL;
2012 	if ((error = _fget(td, fd, &fp, 0, 0)) != 0)
2013 		return (error);
2014 	if (fp->f_vnode == NULL) {
2015 		error = EINVAL;
2016 	} else {
2017 		*vpp = fp->f_vnode;
2018 		vref(*vpp);
2019 	}
2020 	FILEDESC_UNLOCK(td->td_proc->p_fd);
2021 	return (error);
2022 }
2023 
2024 int
2025 fgetvp(struct thread *td, int fd, struct vnode **vpp)
2026 {
2027 
2028 	return (_fgetvp(td, fd, vpp, 0));
2029 }
2030 
2031 int
2032 fgetvp_read(struct thread *td, int fd, struct vnode **vpp)
2033 {
2034 
2035 	return (_fgetvp(td, fd, vpp, FREAD));
2036 }
2037 
2038 #ifdef notyet
2039 int
2040 fgetvp_write(struct thread *td, int fd, struct vnode **vpp)
2041 {
2042 
2043 	return (_fgetvp(td, fd, vpp, FWRITE));
2044 }
2045 #endif
2046 
2047 /*
2048  * Like fget() but loads the underlying socket, or returns an error if
2049  * the descriptor does not represent a socket.
2050  *
2051  * We bump the ref count on the returned socket.  XXX Also obtain the SX
2052  * lock in the future.
2053  */
2054 int
2055 fgetsock(struct thread *td, int fd, struct socket **spp, u_int *fflagp)
2056 {
2057 	struct file *fp;
2058 	int error;
2059 
2060 	NET_ASSERT_GIANT();
2061 
2062 	*spp = NULL;
2063 	if (fflagp != NULL)
2064 		*fflagp = 0;
2065 	if ((error = _fget(td, fd, &fp, 0, 0)) != 0)
2066 		return (error);
2067 	if (fp->f_type != DTYPE_SOCKET) {
2068 		error = ENOTSOCK;
2069 	} else {
2070 		*spp = fp->f_data;
2071 		if (fflagp)
2072 			*fflagp = fp->f_flag;
2073 		SOCK_LOCK(*spp);
2074 		soref(*spp);
2075 		SOCK_UNLOCK(*spp);
2076 	}
2077 	FILEDESC_UNLOCK(td->td_proc->p_fd);
2078 	return (error);
2079 }
2080 
2081 /*
2082  * Drop the reference count on the the socket and XXX release the SX lock in
2083  * the future.  The last reference closes the socket.
2084  */
2085 void
2086 fputsock(struct socket *so)
2087 {
2088 
2089 	NET_ASSERT_GIANT();
2090 	ACCEPT_LOCK();
2091 	SOCK_LOCK(so);
2092 	sorele(so);
2093 }
2094 
2095 int
2096 fdrop(struct file *fp, struct thread *td)
2097 {
2098 
2099 	FILE_LOCK(fp);
2100 	return (fdrop_locked(fp, td));
2101 }
2102 
2103 /*
2104  * Drop reference on struct file passed in, may call closef if the
2105  * reference hits zero.
2106  * Expects struct file locked, and will unlock it.
2107  */
2108 static int
2109 fdrop_locked(struct file *fp, struct thread *td)
2110 {
2111 	int error;
2112 
2113 	FILE_LOCK_ASSERT(fp, MA_OWNED);
2114 
2115 	if (--fp->f_count > 0) {
2116 		FILE_UNLOCK(fp);
2117 		return (0);
2118 	}
2119 	/* We have the last ref so we can proceed without the file lock. */
2120 	FILE_UNLOCK(fp);
2121 	if (fp->f_count < 0)
2122 		panic("fdrop: count < 0");
2123 	if (fp->f_ops != &badfileops)
2124 		error = fo_close(fp, td);
2125 	else
2126 		error = 0;
2127 
2128 	sx_xlock(&filelist_lock);
2129 	LIST_REMOVE(fp, f_list);
2130 	openfiles--;
2131 	sx_xunlock(&filelist_lock);
2132 	crfree(fp->f_cred);
2133 	uma_zfree(file_zone, fp);
2134 
2135 	return (error);
2136 }
2137 
2138 /*
2139  * Apply an advisory lock on a file descriptor.
2140  *
2141  * Just attempt to get a record lock of the requested type on
2142  * the entire file (l_whence = SEEK_SET, l_start = 0, l_len = 0).
2143  */
2144 #ifndef _SYS_SYSPROTO_H_
2145 struct flock_args {
2146 	int	fd;
2147 	int	how;
2148 };
2149 #endif
2150 /*
2151  * MPSAFE
2152  */
2153 /* ARGSUSED */
2154 int
2155 flock(struct thread *td, struct flock_args *uap)
2156 {
2157 	struct file *fp;
2158 	struct vnode *vp;
2159 	struct flock lf;
2160 	int error;
2161 
2162 	if ((error = fget(td, uap->fd, &fp)) != 0)
2163 		return (error);
2164 	if (fp->f_type != DTYPE_VNODE) {
2165 		fdrop(fp, td);
2166 		return (EOPNOTSUPP);
2167 	}
2168 
2169 	mtx_lock(&Giant);
2170 	vp = fp->f_vnode;
2171 	lf.l_whence = SEEK_SET;
2172 	lf.l_start = 0;
2173 	lf.l_len = 0;
2174 	if (uap->how & LOCK_UN) {
2175 		lf.l_type = F_UNLCK;
2176 		FILE_LOCK(fp);
2177 		fp->f_flag &= ~FHASLOCK;
2178 		FILE_UNLOCK(fp);
2179 		error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK);
2180 		goto done2;
2181 	}
2182 	if (uap->how & LOCK_EX)
2183 		lf.l_type = F_WRLCK;
2184 	else if (uap->how & LOCK_SH)
2185 		lf.l_type = F_RDLCK;
2186 	else {
2187 		error = EBADF;
2188 		goto done2;
2189 	}
2190 	FILE_LOCK(fp);
2191 	fp->f_flag |= FHASLOCK;
2192 	FILE_UNLOCK(fp);
2193 	error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf,
2194 	    (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT);
2195 done2:
2196 	fdrop(fp, td);
2197 	mtx_unlock(&Giant);
2198 	return (error);
2199 }
2200 /*
2201  * Duplicate the specified descriptor to a free descriptor.
2202  */
2203 int
2204 dupfdopen(struct thread *td, struct filedesc *fdp, int indx, int dfd, int mode, int error)
2205 {
2206 	struct file *wfp;
2207 	struct file *fp;
2208 
2209 	/*
2210 	 * If the to-be-dup'd fd number is greater than the allowed number
2211 	 * of file descriptors, or the fd to be dup'd has already been
2212 	 * closed, then reject.
2213 	 */
2214 	FILEDESC_LOCK(fdp);
2215 	if (dfd < 0 || dfd >= fdp->fd_nfiles ||
2216 	    (wfp = fdp->fd_ofiles[dfd]) == NULL) {
2217 		FILEDESC_UNLOCK(fdp);
2218 		return (EBADF);
2219 	}
2220 
2221 	/*
2222 	 * There are two cases of interest here.
2223 	 *
2224 	 * For ENODEV simply dup (dfd) to file descriptor
2225 	 * (indx) and return.
2226 	 *
2227 	 * For ENXIO steal away the file structure from (dfd) and
2228 	 * store it in (indx).  (dfd) is effectively closed by
2229 	 * this operation.
2230 	 *
2231 	 * Any other error code is just returned.
2232 	 */
2233 	switch (error) {
2234 	case ENODEV:
2235 		/*
2236 		 * Check that the mode the file is being opened for is a
2237 		 * subset of the mode of the existing descriptor.
2238 		 */
2239 		FILE_LOCK(wfp);
2240 		if (((mode & (FREAD|FWRITE)) | wfp->f_flag) != wfp->f_flag) {
2241 			FILE_UNLOCK(wfp);
2242 			FILEDESC_UNLOCK(fdp);
2243 			return (EACCES);
2244 		}
2245 		fp = fdp->fd_ofiles[indx];
2246 		fdp->fd_ofiles[indx] = wfp;
2247 		fdp->fd_ofileflags[indx] = fdp->fd_ofileflags[dfd];
2248 		if (fp == NULL)
2249 			fdused(fdp, indx);
2250 		fhold_locked(wfp);
2251 		FILE_UNLOCK(wfp);
2252 		FILEDESC_UNLOCK(fdp);
2253 		if (fp != NULL) {
2254 			/*
2255 			 * We now own the reference to fp that the ofiles[]
2256 			 * array used to own.  Release it.
2257 			 */
2258 			FILE_LOCK(fp);
2259 			fdrop_locked(fp, td);
2260 		}
2261 		return (0);
2262 
2263 	case ENXIO:
2264 		/*
2265 		 * Steal away the file pointer from dfd and stuff it into indx.
2266 		 */
2267 		fp = fdp->fd_ofiles[indx];
2268 		fdp->fd_ofiles[indx] = fdp->fd_ofiles[dfd];
2269 		fdp->fd_ofiles[dfd] = NULL;
2270 		fdp->fd_ofileflags[indx] = fdp->fd_ofileflags[dfd];
2271 		fdp->fd_ofileflags[dfd] = 0;
2272 		fdunused(fdp, dfd);
2273 		if (fp == NULL)
2274 			fdused(fdp, indx);
2275 		if (fp != NULL)
2276 			FILE_LOCK(fp);
2277 		FILEDESC_UNLOCK(fdp);
2278 
2279 		/*
2280 		 * we now own the reference to fp that the ofiles[] array
2281 		 * used to own.  Release it.
2282 		 */
2283 		if (fp != NULL)
2284 			fdrop_locked(fp, td);
2285 		return (0);
2286 
2287 	default:
2288 		FILEDESC_UNLOCK(fdp);
2289 		return (error);
2290 	}
2291 	/* NOTREACHED */
2292 }
2293 
2294 /*
2295  * Scan all active processes to see if any of them have a current
2296  * or root directory of `olddp'. If so, replace them with the new
2297  * mount point.
2298  */
2299 void
2300 mountcheckdirs(struct vnode *olddp, struct vnode *newdp)
2301 {
2302 	struct filedesc *fdp;
2303 	struct proc *p;
2304 	int nrele;
2305 
2306 	if (vrefcnt(olddp) == 1)
2307 		return;
2308 	sx_slock(&allproc_lock);
2309 	LIST_FOREACH(p, &allproc, p_list) {
2310 		fdp = fdhold(p);
2311 		if (fdp == NULL)
2312 			continue;
2313 		nrele = 0;
2314 		FILEDESC_LOCK_FAST(fdp);
2315 		if (fdp->fd_cdir == olddp) {
2316 			vref(newdp);
2317 			fdp->fd_cdir = newdp;
2318 			nrele++;
2319 		}
2320 		if (fdp->fd_rdir == olddp) {
2321 			vref(newdp);
2322 			fdp->fd_rdir = newdp;
2323 			nrele++;
2324 		}
2325 		FILEDESC_UNLOCK_FAST(fdp);
2326 		fddrop(fdp);
2327 		while (nrele--)
2328 			vrele(olddp);
2329 	}
2330 	sx_sunlock(&allproc_lock);
2331 	if (rootvnode == olddp) {
2332 		vrele(rootvnode);
2333 		vref(newdp);
2334 		rootvnode = newdp;
2335 	}
2336 }
2337 
2338 struct filedesc_to_leader *
2339 filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader)
2340 {
2341 	struct filedesc_to_leader *fdtol;
2342 
2343 	MALLOC(fdtol, struct filedesc_to_leader *,
2344 	       sizeof(struct filedesc_to_leader),
2345 	       M_FILEDESC_TO_LEADER,
2346 	       M_WAITOK);
2347 	fdtol->fdl_refcount = 1;
2348 	fdtol->fdl_holdcount = 0;
2349 	fdtol->fdl_wakeup = 0;
2350 	fdtol->fdl_leader = leader;
2351 	if (old != NULL) {
2352 		FILEDESC_LOCK(fdp);
2353 		fdtol->fdl_next = old->fdl_next;
2354 		fdtol->fdl_prev = old;
2355 		old->fdl_next = fdtol;
2356 		fdtol->fdl_next->fdl_prev = fdtol;
2357 		FILEDESC_UNLOCK(fdp);
2358 	} else {
2359 		fdtol->fdl_next = fdtol;
2360 		fdtol->fdl_prev = fdtol;
2361 	}
2362 	return (fdtol);
2363 }
2364 
2365 /*
2366  * Get file structures.
2367  */
2368 static int
2369 sysctl_kern_file(SYSCTL_HANDLER_ARGS)
2370 {
2371 	struct xfile xf;
2372 	struct filedesc *fdp;
2373 	struct file *fp;
2374 	struct proc *p;
2375 	int error, n;
2376 
2377 	/*
2378 	 * Note: because the number of file descriptors is calculated
2379 	 * in different ways for sizing vs returning the data,
2380 	 * there is information leakage from the first loop.  However,
2381 	 * it is of a similar order of magnitude to the leakage from
2382 	 * global system statistics such as kern.openfiles.
2383 	 */
2384 	error = sysctl_wire_old_buffer(req, 0);
2385 	if (error != 0)
2386 		return (error);
2387 	if (req->oldptr == NULL) {
2388 		n = 16;		/* A slight overestimate. */
2389 		sx_slock(&filelist_lock);
2390 		LIST_FOREACH(fp, &filehead, f_list) {
2391 			/*
2392 			 * We should grab the lock, but this is an
2393 			 * estimate, so does it really matter?
2394 			 */
2395 			/* mtx_lock(fp->f_mtxp); */
2396 			n += fp->f_count;
2397 			/* mtx_unlock(f->f_mtxp); */
2398 		}
2399 		sx_sunlock(&filelist_lock);
2400 		return (SYSCTL_OUT(req, 0, n * sizeof(xf)));
2401 	}
2402 	error = 0;
2403 	bzero(&xf, sizeof(xf));
2404 	xf.xf_size = sizeof(xf);
2405 	sx_slock(&allproc_lock);
2406 	LIST_FOREACH(p, &allproc, p_list) {
2407 		if (p->p_state == PRS_NEW)
2408 			continue;
2409 		PROC_LOCK(p);
2410 		if (p_cansee(req->td, p) != 0) {
2411 			PROC_UNLOCK(p);
2412 			continue;
2413 		}
2414 		xf.xf_pid = p->p_pid;
2415 		xf.xf_uid = p->p_ucred->cr_uid;
2416 		PROC_UNLOCK(p);
2417 		fdp = fdhold(p);
2418 		if (fdp == NULL)
2419 			continue;
2420 		FILEDESC_LOCK_FAST(fdp);
2421 		for (n = 0; fdp->fd_refcnt > 0 && n < fdp->fd_nfiles; ++n) {
2422 			if ((fp = fdp->fd_ofiles[n]) == NULL)
2423 				continue;
2424 			xf.xf_fd = n;
2425 			xf.xf_file = fp;
2426 			xf.xf_data = fp->f_data;
2427 			xf.xf_vnode = fp->f_vnode;
2428 			xf.xf_type = fp->f_type;
2429 			xf.xf_count = fp->f_count;
2430 			xf.xf_msgcount = fp->f_msgcount;
2431 			xf.xf_offset = fp->f_offset;
2432 			xf.xf_flag = fp->f_flag;
2433 			error = SYSCTL_OUT(req, &xf, sizeof(xf));
2434 			if (error)
2435 				break;
2436 		}
2437 		FILEDESC_UNLOCK_FAST(fdp);
2438 		fddrop(fdp);
2439 		if (error)
2440 			break;
2441 	}
2442 	sx_sunlock(&allproc_lock);
2443 	return (error);
2444 }
2445 
2446 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD,
2447     0, 0, sysctl_kern_file, "S,xfile", "Entire file table");
2448 
2449 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
2450     &maxfilesperproc, 0, "Maximum files allowed open per process");
2451 
2452 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
2453     &maxfiles, 0, "Maximum number of files");
2454 
2455 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
2456     &openfiles, 0, "System-wide number of open files");
2457 
2458 /* ARGSUSED*/
2459 static void
2460 filelistinit(void *dummy)
2461 {
2462 
2463 	file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL,
2464 	    NULL, NULL, UMA_ALIGN_PTR, 0);
2465 	sx_init(&filelist_lock, "filelist lock");
2466 	mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF);
2467 	mtx_init(&fdesc_mtx, "fdesc", NULL, MTX_DEF);
2468 }
2469 SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL)
2470 
2471 /*-------------------------------------------------------------------*/
2472 
2473 static int
2474 badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred, int flags, struct thread *td)
2475 {
2476 
2477 	return (EBADF);
2478 }
2479 
2480 static int
2481 badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred, struct thread *td)
2482 {
2483 
2484 	return (EBADF);
2485 }
2486 
2487 static int
2488 badfo_poll(struct file *fp, int events, struct ucred *active_cred, struct thread *td)
2489 {
2490 
2491 	return (0);
2492 }
2493 
2494 static int
2495 badfo_kqfilter(struct file *fp, struct knote *kn)
2496 {
2497 
2498 	return (0);
2499 }
2500 
2501 static int
2502 badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred, struct thread *td)
2503 {
2504 
2505 	return (EBADF);
2506 }
2507 
2508 static int
2509 badfo_close(struct file *fp, struct thread *td)
2510 {
2511 
2512 	return (EBADF);
2513 }
2514 
2515 struct fileops badfileops = {
2516 	.fo_read = badfo_readwrite,
2517 	.fo_write = badfo_readwrite,
2518 	.fo_ioctl = badfo_ioctl,
2519 	.fo_poll = badfo_poll,
2520 	.fo_kqfilter = badfo_kqfilter,
2521 	.fo_stat = badfo_stat,
2522 	.fo_close = badfo_close,
2523 };
2524 
2525 
2526 /*-------------------------------------------------------------------*/
2527 
2528 /*
2529  * File Descriptor pseudo-device driver (/dev/fd/).
2530  *
2531  * Opening minor device N dup()s the file (if any) connected to file
2532  * descriptor N belonging to the calling process.  Note that this driver
2533  * consists of only the ``open()'' routine, because all subsequent
2534  * references to this file will be direct to the other driver.
2535  *
2536  * XXX: we could give this one a cloning event handler if necessary.
2537  */
2538 
2539 /* ARGSUSED */
2540 static int
2541 fdopen(struct cdev *dev, int mode, int type, struct thread *td)
2542 {
2543 
2544 	/*
2545 	 * XXX Kludge: set curthread->td_dupfd to contain the value of the
2546 	 * the file descriptor being sought for duplication. The error
2547 	 * return ensures that the vnode for this device will be released
2548 	 * by vn_open. Open will detect this special error and take the
2549 	 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
2550 	 * will simply report the error.
2551 	 */
2552 	td->td_dupfd = dev2unit(dev);
2553 	return (ENODEV);
2554 }
2555 
2556 static struct cdevsw fildesc_cdevsw = {
2557 	.d_version =	D_VERSION,
2558 	.d_flags =	D_NEEDGIANT,
2559 	.d_open =	fdopen,
2560 	.d_name =	"FD",
2561 };
2562 
2563 static void
2564 fildesc_drvinit(void *unused)
2565 {
2566 	struct cdev *dev;
2567 
2568 	dev = make_dev(&fildesc_cdevsw, 0, UID_ROOT, GID_WHEEL, 0666, "fd/0");
2569 	make_dev_alias(dev, "stdin");
2570 	dev = make_dev(&fildesc_cdevsw, 1, UID_ROOT, GID_WHEEL, 0666, "fd/1");
2571 	make_dev_alias(dev, "stdout");
2572 	dev = make_dev(&fildesc_cdevsw, 2, UID_ROOT, GID_WHEEL, 0666, "fd/2");
2573 	make_dev_alias(dev, "stderr");
2574 }
2575 
2576 SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL)
2577