xref: /freebsd/sys/kern/kern_descrip.c (revision aa77200569e397d6ff1fdb4d255d0fa254d0a128)
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_capsicum.h"
41 #include "opt_compat.h"
42 #include "opt_ddb.h"
43 #include "opt_ktrace.h"
44 #include "opt_procdesc.h"
45 
46 #include <sys/param.h>
47 #include <sys/systm.h>
48 
49 #include <sys/capability.h>
50 #include <sys/conf.h>
51 #include <sys/domain.h>
52 #include <sys/fcntl.h>
53 #include <sys/file.h>
54 #include <sys/filedesc.h>
55 #include <sys/filio.h>
56 #include <sys/jail.h>
57 #include <sys/kernel.h>
58 #include <sys/limits.h>
59 #include <sys/lock.h>
60 #include <sys/malloc.h>
61 #include <sys/mman.h>
62 #include <sys/mount.h>
63 #include <sys/mqueue.h>
64 #include <sys/mutex.h>
65 #include <sys/namei.h>
66 #include <sys/selinfo.h>
67 #include <sys/pipe.h>
68 #include <sys/priv.h>
69 #include <sys/proc.h>
70 #include <sys/procdesc.h>
71 #include <sys/protosw.h>
72 #include <sys/racct.h>
73 #include <sys/resourcevar.h>
74 #include <sys/signalvar.h>
75 #include <sys/socketvar.h>
76 #include <sys/stat.h>
77 #include <sys/sx.h>
78 #include <sys/syscallsubr.h>
79 #include <sys/sysctl.h>
80 #include <sys/sysproto.h>
81 #include <sys/tty.h>
82 #include <sys/unistd.h>
83 #include <sys/un.h>
84 #include <sys/unpcb.h>
85 #include <sys/user.h>
86 #include <sys/vnode.h>
87 #ifdef KTRACE
88 #include <sys/ktrace.h>
89 #endif
90 
91 #include <net/vnet.h>
92 
93 #include <netinet/in.h>
94 #include <netinet/in_pcb.h>
95 
96 #include <security/audit/audit.h>
97 
98 #include <vm/uma.h>
99 #include <vm/vm.h>
100 
101 #include <ddb/ddb.h>
102 
103 static MALLOC_DEFINE(M_FILEDESC, "filedesc", "Open file descriptor table");
104 static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "filedesc_to_leader",
105     "file desc to leader structures");
106 static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures");
107 
108 MALLOC_DECLARE(M_FADVISE);
109 
110 static uma_zone_t file_zone;
111 
112 
113 /* Flags for do_dup() */
114 #define DUP_FIXED	0x1	/* Force fixed allocation */
115 #define DUP_FCNTL	0x2	/* fcntl()-style errors */
116 #define	DUP_CLOEXEC	0x4	/* Atomically set FD_CLOEXEC. */
117 
118 static int	closefp(struct filedesc *fdp, int fd, struct file *fp,
119 		    struct thread *td, int holdleaders);
120 static int	do_dup(struct thread *td, int flags, int old, int new,
121 		    register_t *retval);
122 static int	fd_first_free(struct filedesc *fdp, int low, int size);
123 static int	fd_last_used(struct filedesc *fdp, int size);
124 static void	fdgrowtable(struct filedesc *fdp, int nfd);
125 static void	fdunused(struct filedesc *fdp, int fd);
126 static void	fdused(struct filedesc *fdp, int fd);
127 static int	fill_pipe_info(struct pipe *pi, struct kinfo_file *kif);
128 static int	fill_procdesc_info(struct procdesc *pdp,
129 		    struct kinfo_file *kif);
130 static int	fill_pts_info(struct tty *tp, struct kinfo_file *kif);
131 static int	fill_shm_info(struct file *fp, struct kinfo_file *kif);
132 static int	fill_socket_info(struct socket *so, struct kinfo_file *kif);
133 static int	fill_vnode_info(struct vnode *vp, struct kinfo_file *kif);
134 
135 /*
136  * A process is initially started out with NDFILE descriptors stored within
137  * this structure, selected to be enough for typical applications based on
138  * the historical limit of 20 open files (and the usage of descriptors by
139  * shells).  If these descriptors are exhausted, a larger descriptor table
140  * may be allocated, up to a process' resource limit; the internal arrays
141  * are then unused.
142  */
143 #define NDFILE		20
144 #define NDSLOTSIZE	sizeof(NDSLOTTYPE)
145 #define	NDENTRIES	(NDSLOTSIZE * __CHAR_BIT)
146 #define NDSLOT(x)	((x) / NDENTRIES)
147 #define NDBIT(x)	((NDSLOTTYPE)1 << ((x) % NDENTRIES))
148 #define	NDSLOTS(x)	(((x) + NDENTRIES - 1) / NDENTRIES)
149 
150 /*
151  * Storage required per open file descriptor.
152  */
153 #define OFILESIZE (sizeof(struct file *) + sizeof(char))
154 
155 /*
156  * Storage to hold unused ofiles that need to be reclaimed.
157  */
158 struct freetable {
159 	struct file	**ft_table;
160 	SLIST_ENTRY(freetable) ft_next;
161 };
162 
163 /*
164  * Basic allocation of descriptors:
165  * one of the above, plus arrays for NDFILE descriptors.
166  */
167 struct filedesc0 {
168 	struct	filedesc fd_fd;
169 	/*
170 	 * ofiles which need to be reclaimed on free.
171 	 */
172 	SLIST_HEAD(,freetable) fd_free;
173 	/*
174 	 * These arrays are used when the number of open files is
175 	 * <= NDFILE, and are then pointed to by the pointers above.
176 	 */
177 	struct	file *fd_dfiles[NDFILE];
178 	char	fd_dfileflags[NDFILE];
179 	NDSLOTTYPE fd_dmap[NDSLOTS(NDFILE)];
180 };
181 
182 /*
183  * Descriptor management.
184  */
185 volatile int openfiles;			/* actual number of open files */
186 struct mtx sigio_lock;		/* mtx to protect pointers to sigio */
187 void (*mq_fdclose)(struct thread *td, int fd, struct file *fp);
188 
189 /* A mutex to protect the association between a proc and filedesc. */
190 static struct mtx fdesc_mtx;
191 
192 /*
193  * If low >= size, just return low. Otherwise find the first zero bit in the
194  * given bitmap, starting at low and not exceeding size - 1. Return size if
195  * not found.
196  */
197 static int
198 fd_first_free(struct filedesc *fdp, int low, int size)
199 {
200 	NDSLOTTYPE *map = fdp->fd_map;
201 	NDSLOTTYPE mask;
202 	int off, maxoff;
203 
204 	if (low >= size)
205 		return (low);
206 
207 	off = NDSLOT(low);
208 	if (low % NDENTRIES) {
209 		mask = ~(~(NDSLOTTYPE)0 >> (NDENTRIES - (low % NDENTRIES)));
210 		if ((mask &= ~map[off]) != 0UL)
211 			return (off * NDENTRIES + ffsl(mask) - 1);
212 		++off;
213 	}
214 	for (maxoff = NDSLOTS(size); off < maxoff; ++off)
215 		if (map[off] != ~0UL)
216 			return (off * NDENTRIES + ffsl(~map[off]) - 1);
217 	return (size);
218 }
219 
220 /*
221  * Find the highest non-zero bit in the given bitmap, starting at 0 and
222  * not exceeding size - 1. Return -1 if not found.
223  */
224 static int
225 fd_last_used(struct filedesc *fdp, int size)
226 {
227 	NDSLOTTYPE *map = fdp->fd_map;
228 	NDSLOTTYPE mask;
229 	int off, minoff;
230 
231 	off = NDSLOT(size);
232 	if (size % NDENTRIES) {
233 		mask = ~(~(NDSLOTTYPE)0 << (size % NDENTRIES));
234 		if ((mask &= map[off]) != 0)
235 			return (off * NDENTRIES + flsl(mask) - 1);
236 		--off;
237 	}
238 	for (minoff = NDSLOT(0); off >= minoff; --off)
239 		if (map[off] != 0)
240 			return (off * NDENTRIES + flsl(map[off]) - 1);
241 	return (-1);
242 }
243 
244 static int
245 fdisused(struct filedesc *fdp, int fd)
246 {
247 
248 	FILEDESC_LOCK_ASSERT(fdp);
249 
250 	KASSERT(fd >= 0 && fd < fdp->fd_nfiles,
251 	    ("file descriptor %d out of range (0, %d)", fd, fdp->fd_nfiles));
252 
253 	return ((fdp->fd_map[NDSLOT(fd)] & NDBIT(fd)) != 0);
254 }
255 
256 /*
257  * Mark a file descriptor as used.
258  */
259 static void
260 fdused(struct filedesc *fdp, int fd)
261 {
262 
263 	FILEDESC_XLOCK_ASSERT(fdp);
264 
265 	KASSERT(!fdisused(fdp, fd), ("fd=%d is already used", fd));
266 
267 	fdp->fd_map[NDSLOT(fd)] |= NDBIT(fd);
268 	if (fd > fdp->fd_lastfile)
269 		fdp->fd_lastfile = fd;
270 	if (fd == fdp->fd_freefile)
271 		fdp->fd_freefile = fd_first_free(fdp, fd, fdp->fd_nfiles);
272 }
273 
274 /*
275  * Mark a file descriptor as unused.
276  */
277 static void
278 fdunused(struct filedesc *fdp, int fd)
279 {
280 
281 	FILEDESC_XLOCK_ASSERT(fdp);
282 
283 	KASSERT(fdisused(fdp, fd), ("fd=%d is already unused", fd));
284 	KASSERT(fdp->fd_ofiles[fd] == NULL, ("fd=%d is still in use", fd));
285 
286 	fdp->fd_map[NDSLOT(fd)] &= ~NDBIT(fd);
287 	if (fd < fdp->fd_freefile)
288 		fdp->fd_freefile = fd;
289 	if (fd == fdp->fd_lastfile)
290 		fdp->fd_lastfile = fd_last_used(fdp, fd);
291 }
292 
293 /*
294  * System calls on descriptors.
295  */
296 #ifndef _SYS_SYSPROTO_H_
297 struct getdtablesize_args {
298 	int	dummy;
299 };
300 #endif
301 /* ARGSUSED */
302 int
303 sys_getdtablesize(struct thread *td, struct getdtablesize_args *uap)
304 {
305 	struct proc *p = td->td_proc;
306 	uint64_t lim;
307 
308 	PROC_LOCK(p);
309 	td->td_retval[0] =
310 	    min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc);
311 	lim = racct_get_limit(td->td_proc, RACCT_NOFILE);
312 	PROC_UNLOCK(p);
313 	if (lim < td->td_retval[0])
314 		td->td_retval[0] = lim;
315 	return (0);
316 }
317 
318 /*
319  * Duplicate a file descriptor to a particular value.
320  *
321  * Note: keep in mind that a potential race condition exists when closing
322  * descriptors from a shared descriptor table (via rfork).
323  */
324 #ifndef _SYS_SYSPROTO_H_
325 struct dup2_args {
326 	u_int	from;
327 	u_int	to;
328 };
329 #endif
330 /* ARGSUSED */
331 int
332 sys_dup2(struct thread *td, struct dup2_args *uap)
333 {
334 
335 	return (do_dup(td, DUP_FIXED, (int)uap->from, (int)uap->to,
336 		    td->td_retval));
337 }
338 
339 /*
340  * Duplicate a file descriptor.
341  */
342 #ifndef _SYS_SYSPROTO_H_
343 struct dup_args {
344 	u_int	fd;
345 };
346 #endif
347 /* ARGSUSED */
348 int
349 sys_dup(struct thread *td, struct dup_args *uap)
350 {
351 
352 	return (do_dup(td, 0, (int)uap->fd, 0, td->td_retval));
353 }
354 
355 /*
356  * The file control system call.
357  */
358 #ifndef _SYS_SYSPROTO_H_
359 struct fcntl_args {
360 	int	fd;
361 	int	cmd;
362 	long	arg;
363 };
364 #endif
365 /* ARGSUSED */
366 int
367 sys_fcntl(struct thread *td, struct fcntl_args *uap)
368 {
369 	struct flock fl;
370 	struct __oflock ofl;
371 	intptr_t arg;
372 	int error;
373 	int cmd;
374 
375 	error = 0;
376 	cmd = uap->cmd;
377 	switch (uap->cmd) {
378 	case F_OGETLK:
379 	case F_OSETLK:
380 	case F_OSETLKW:
381 		/*
382 		 * Convert old flock structure to new.
383 		 */
384 		error = copyin((void *)(intptr_t)uap->arg, &ofl, sizeof(ofl));
385 		fl.l_start = ofl.l_start;
386 		fl.l_len = ofl.l_len;
387 		fl.l_pid = ofl.l_pid;
388 		fl.l_type = ofl.l_type;
389 		fl.l_whence = ofl.l_whence;
390 		fl.l_sysid = 0;
391 
392 		switch (uap->cmd) {
393 		case F_OGETLK:
394 		    cmd = F_GETLK;
395 		    break;
396 		case F_OSETLK:
397 		    cmd = F_SETLK;
398 		    break;
399 		case F_OSETLKW:
400 		    cmd = F_SETLKW;
401 		    break;
402 		}
403 		arg = (intptr_t)&fl;
404 		break;
405         case F_GETLK:
406         case F_SETLK:
407         case F_SETLKW:
408 	case F_SETLK_REMOTE:
409                 error = copyin((void *)(intptr_t)uap->arg, &fl, sizeof(fl));
410                 arg = (intptr_t)&fl;
411                 break;
412 	default:
413 		arg = uap->arg;
414 		break;
415 	}
416 	if (error)
417 		return (error);
418 	error = kern_fcntl(td, uap->fd, cmd, arg);
419 	if (error)
420 		return (error);
421 	if (uap->cmd == F_OGETLK) {
422 		ofl.l_start = fl.l_start;
423 		ofl.l_len = fl.l_len;
424 		ofl.l_pid = fl.l_pid;
425 		ofl.l_type = fl.l_type;
426 		ofl.l_whence = fl.l_whence;
427 		error = copyout(&ofl, (void *)(intptr_t)uap->arg, sizeof(ofl));
428 	} else if (uap->cmd == F_GETLK) {
429 		error = copyout(&fl, (void *)(intptr_t)uap->arg, sizeof(fl));
430 	}
431 	return (error);
432 }
433 
434 static inline int
435 fdunwrap(int fd, cap_rights_t rights, struct filedesc *fdp, struct file **fpp)
436 {
437 
438 	FILEDESC_LOCK_ASSERT(fdp);
439 
440 	*fpp = fget_locked(fdp, fd);
441 	if (*fpp == NULL)
442 		return (EBADF);
443 
444 #ifdef CAPABILITIES
445 	if ((*fpp)->f_type == DTYPE_CAPABILITY) {
446 		int err = cap_funwrap(*fpp, rights, fpp);
447 		if (err != 0) {
448 			*fpp = NULL;
449 			return (err);
450 		}
451 	}
452 #endif /* CAPABILITIES */
453 	return (0);
454 }
455 
456 int
457 kern_fcntl(struct thread *td, int fd, int cmd, intptr_t arg)
458 {
459 	struct filedesc *fdp;
460 	struct flock *flp;
461 	struct file *fp;
462 	struct proc *p;
463 	char *pop;
464 	struct vnode *vp;
465 	int error, flg, tmp;
466 	u_int old, new;
467 	uint64_t bsize;
468 	off_t foffset;
469 
470 	error = 0;
471 	flg = F_POSIX;
472 	p = td->td_proc;
473 	fdp = p->p_fd;
474 
475 	switch (cmd) {
476 	case F_DUPFD:
477 		tmp = arg;
478 		error = do_dup(td, DUP_FCNTL, fd, tmp, td->td_retval);
479 		break;
480 
481 	case F_DUPFD_CLOEXEC:
482 		tmp = arg;
483 		error = do_dup(td, DUP_FCNTL | DUP_CLOEXEC, fd, tmp,
484 		    td->td_retval);
485 		break;
486 
487 	case F_DUP2FD:
488 		tmp = arg;
489 		error = do_dup(td, DUP_FIXED, fd, tmp, td->td_retval);
490 		break;
491 
492 	case F_DUP2FD_CLOEXEC:
493 		tmp = arg;
494 		error = do_dup(td, DUP_FIXED | DUP_CLOEXEC, fd, tmp,
495 		    td->td_retval);
496 		break;
497 
498 	case F_GETFD:
499 		FILEDESC_SLOCK(fdp);
500 		if ((fp = fget_locked(fdp, fd)) == NULL) {
501 			FILEDESC_SUNLOCK(fdp);
502 			error = EBADF;
503 			break;
504 		}
505 		pop = &fdp->fd_ofileflags[fd];
506 		td->td_retval[0] = (*pop & UF_EXCLOSE) ? FD_CLOEXEC : 0;
507 		FILEDESC_SUNLOCK(fdp);
508 		break;
509 
510 	case F_SETFD:
511 		FILEDESC_XLOCK(fdp);
512 		if ((fp = fget_locked(fdp, fd)) == NULL) {
513 			FILEDESC_XUNLOCK(fdp);
514 			error = EBADF;
515 			break;
516 		}
517 		pop = &fdp->fd_ofileflags[fd];
518 		*pop = (*pop &~ UF_EXCLOSE) |
519 		    (arg & FD_CLOEXEC ? UF_EXCLOSE : 0);
520 		FILEDESC_XUNLOCK(fdp);
521 		break;
522 
523 	case F_GETFL:
524 		FILEDESC_SLOCK(fdp);
525 		error = fdunwrap(fd, CAP_FCNTL, fdp, &fp);
526 		if (error != 0) {
527 			FILEDESC_SUNLOCK(fdp);
528 			break;
529 		}
530 		td->td_retval[0] = OFLAGS(fp->f_flag);
531 		FILEDESC_SUNLOCK(fdp);
532 		break;
533 
534 	case F_SETFL:
535 		FILEDESC_SLOCK(fdp);
536 		error = fdunwrap(fd, CAP_FCNTL, fdp, &fp);
537 		if (error != 0) {
538 			FILEDESC_SUNLOCK(fdp);
539 			break;
540 		}
541 		fhold(fp);
542 		FILEDESC_SUNLOCK(fdp);
543 		do {
544 			tmp = flg = fp->f_flag;
545 			tmp &= ~FCNTLFLAGS;
546 			tmp |= FFLAGS(arg & ~O_ACCMODE) & FCNTLFLAGS;
547 		} while(atomic_cmpset_int(&fp->f_flag, flg, tmp) == 0);
548 		tmp = fp->f_flag & FNONBLOCK;
549 		error = fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
550 		if (error) {
551 			fdrop(fp, td);
552 			break;
553 		}
554 		tmp = fp->f_flag & FASYNC;
555 		error = fo_ioctl(fp, FIOASYNC, &tmp, td->td_ucred, td);
556 		if (error == 0) {
557 			fdrop(fp, td);
558 			break;
559 		}
560 		atomic_clear_int(&fp->f_flag, FNONBLOCK);
561 		tmp = 0;
562 		(void)fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
563 		fdrop(fp, td);
564 		break;
565 
566 	case F_GETOWN:
567 		FILEDESC_SLOCK(fdp);
568 		error = fdunwrap(fd, CAP_FCNTL, fdp, &fp);
569 		if (error != 0) {
570 			FILEDESC_SUNLOCK(fdp);
571 			break;
572 		}
573 		fhold(fp);
574 		FILEDESC_SUNLOCK(fdp);
575 		error = fo_ioctl(fp, FIOGETOWN, &tmp, td->td_ucred, td);
576 		if (error == 0)
577 			td->td_retval[0] = tmp;
578 		fdrop(fp, td);
579 		break;
580 
581 	case F_SETOWN:
582 		FILEDESC_SLOCK(fdp);
583 		error = fdunwrap(fd, CAP_FCNTL, fdp, &fp);
584 		if (error != 0) {
585 			FILEDESC_SUNLOCK(fdp);
586 			break;
587 		}
588 		fhold(fp);
589 		FILEDESC_SUNLOCK(fdp);
590 		tmp = arg;
591 		error = fo_ioctl(fp, FIOSETOWN, &tmp, td->td_ucred, td);
592 		fdrop(fp, td);
593 		break;
594 
595 	case F_SETLK_REMOTE:
596 		error = priv_check(td, PRIV_NFS_LOCKD);
597 		if (error)
598 			return (error);
599 		flg = F_REMOTE;
600 		goto do_setlk;
601 
602 	case F_SETLKW:
603 		flg |= F_WAIT;
604 		/* FALLTHROUGH F_SETLK */
605 
606 	case F_SETLK:
607 	do_setlk:
608 		FILEDESC_SLOCK(fdp);
609 		error = fdunwrap(fd, CAP_FLOCK, fdp, &fp);
610 		if (error != 0) {
611 			FILEDESC_SUNLOCK(fdp);
612 			break;
613 		}
614 		if (fp->f_type != DTYPE_VNODE) {
615 			FILEDESC_SUNLOCK(fdp);
616 			error = EBADF;
617 			break;
618 		}
619 		flp = (struct flock *)arg;
620 		if (flp->l_whence == SEEK_CUR) {
621 			foffset = foffset_get(fp);
622 			if (foffset < 0 ||
623 			    (flp->l_start > 0 &&
624 			     foffset > OFF_MAX - flp->l_start)) {
625 				FILEDESC_SUNLOCK(fdp);
626 				error = EOVERFLOW;
627 				break;
628 			}
629 			flp->l_start += foffset;
630 		}
631 
632 		/*
633 		 * VOP_ADVLOCK() may block.
634 		 */
635 		fhold(fp);
636 		FILEDESC_SUNLOCK(fdp);
637 		vp = fp->f_vnode;
638 		switch (flp->l_type) {
639 		case F_RDLCK:
640 			if ((fp->f_flag & FREAD) == 0) {
641 				error = EBADF;
642 				break;
643 			}
644 			PROC_LOCK(p->p_leader);
645 			p->p_leader->p_flag |= P_ADVLOCK;
646 			PROC_UNLOCK(p->p_leader);
647 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
648 			    flp, flg);
649 			break;
650 		case F_WRLCK:
651 			if ((fp->f_flag & FWRITE) == 0) {
652 				error = EBADF;
653 				break;
654 			}
655 			PROC_LOCK(p->p_leader);
656 			p->p_leader->p_flag |= P_ADVLOCK;
657 			PROC_UNLOCK(p->p_leader);
658 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
659 			    flp, flg);
660 			break;
661 		case F_UNLCK:
662 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
663 			    flp, flg);
664 			break;
665 		case F_UNLCKSYS:
666 			/*
667 			 * Temporary api for testing remote lock
668 			 * infrastructure.
669 			 */
670 			if (flg != F_REMOTE) {
671 				error = EINVAL;
672 				break;
673 			}
674 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
675 			    F_UNLCKSYS, flp, flg);
676 			break;
677 		default:
678 			error = EINVAL;
679 			break;
680 		}
681 		if (error != 0 || flp->l_type == F_UNLCK ||
682 		    flp->l_type == F_UNLCKSYS) {
683 			fdrop(fp, td);
684 			break;
685 		}
686 
687 		/*
688 		 * Check for a race with close.
689 		 *
690 		 * The vnode is now advisory locked (or unlocked, but this case
691 		 * is not really important) as the caller requested.
692 		 * We had to drop the filedesc lock, so we need to recheck if
693 		 * the descriptor is still valid, because if it was closed
694 		 * in the meantime we need to remove advisory lock from the
695 		 * vnode - close on any descriptor leading to an advisory
696 		 * locked vnode, removes that lock.
697 		 * We will return 0 on purpose in that case, as the result of
698 		 * successful advisory lock might have been externally visible
699 		 * already. This is fine - effectively we pretend to the caller
700 		 * that the closing thread was a bit slower and that the
701 		 * advisory lock succeeded before the close.
702 		 */
703 		FILEDESC_SLOCK(fdp);
704 		if (fget_locked(fdp, fd) != fp) {
705 			FILEDESC_SUNLOCK(fdp);
706 			flp->l_whence = SEEK_SET;
707 			flp->l_start = 0;
708 			flp->l_len = 0;
709 			flp->l_type = F_UNLCK;
710 			(void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
711 			    F_UNLCK, flp, F_POSIX);
712 		} else
713 			FILEDESC_SUNLOCK(fdp);
714 		fdrop(fp, td);
715 		break;
716 
717 	case F_GETLK:
718 		FILEDESC_SLOCK(fdp);
719 		error = fdunwrap(fd, CAP_FLOCK, fdp, &fp);
720 		if (error != 0) {
721 			FILEDESC_SUNLOCK(fdp);
722 			break;
723 		}
724 		if (fp->f_type != DTYPE_VNODE) {
725 			FILEDESC_SUNLOCK(fdp);
726 			error = EBADF;
727 			break;
728 		}
729 		flp = (struct flock *)arg;
730 		if (flp->l_type != F_RDLCK && flp->l_type != F_WRLCK &&
731 		    flp->l_type != F_UNLCK) {
732 			FILEDESC_SUNLOCK(fdp);
733 			error = EINVAL;
734 			break;
735 		}
736 		if (flp->l_whence == SEEK_CUR) {
737 			foffset = foffset_get(fp);
738 			if ((flp->l_start > 0 &&
739 			    foffset > OFF_MAX - flp->l_start) ||
740 			    (flp->l_start < 0 &&
741 			     foffset < OFF_MIN - flp->l_start)) {
742 				FILEDESC_SUNLOCK(fdp);
743 				error = EOVERFLOW;
744 				break;
745 			}
746 			flp->l_start += foffset;
747 		}
748 		/*
749 		 * VOP_ADVLOCK() may block.
750 		 */
751 		fhold(fp);
752 		FILEDESC_SUNLOCK(fdp);
753 		vp = fp->f_vnode;
754 		error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, flp,
755 		    F_POSIX);
756 		fdrop(fp, td);
757 		break;
758 
759 	case F_RDAHEAD:
760 		arg = arg ? 128 * 1024: 0;
761 		/* FALLTHROUGH */
762 	case F_READAHEAD:
763 		FILEDESC_SLOCK(fdp);
764 		if ((fp = fget_locked(fdp, fd)) == NULL) {
765 			FILEDESC_SUNLOCK(fdp);
766 			error = EBADF;
767 			break;
768 		}
769 		if (fp->f_type != DTYPE_VNODE) {
770 			FILEDESC_SUNLOCK(fdp);
771 			error = EBADF;
772 			break;
773 		}
774 		fhold(fp);
775 		FILEDESC_SUNLOCK(fdp);
776 		if (arg != 0) {
777 			vp = fp->f_vnode;
778 			error = vn_lock(vp, LK_SHARED);
779 			if (error != 0)
780 				goto readahead_vnlock_fail;
781 			bsize = fp->f_vnode->v_mount->mnt_stat.f_iosize;
782 			VOP_UNLOCK(vp, 0);
783 			fp->f_seqcount = (arg + bsize - 1) / bsize;
784 			do {
785 				new = old = fp->f_flag;
786 				new |= FRDAHEAD;
787 			} while (!atomic_cmpset_rel_int(&fp->f_flag, old, new));
788 		readahead_vnlock_fail:;
789 		} else {
790 			do {
791 				new = old = fp->f_flag;
792 				new &= ~FRDAHEAD;
793 			} while (!atomic_cmpset_rel_int(&fp->f_flag, old, new));
794 		}
795 		fdrop(fp, td);
796 		break;
797 
798 	default:
799 		error = EINVAL;
800 		break;
801 	}
802 	return (error);
803 }
804 
805 /*
806  * Common code for dup, dup2, fcntl(F_DUPFD) and fcntl(F_DUP2FD).
807  */
808 static int
809 do_dup(struct thread *td, int flags, int old, int new,
810     register_t *retval)
811 {
812 	struct filedesc *fdp;
813 	struct proc *p;
814 	struct file *fp;
815 	struct file *delfp;
816 	int error, maxfd;
817 
818 	p = td->td_proc;
819 	fdp = p->p_fd;
820 
821 	/*
822 	 * Verify we have a valid descriptor to dup from and possibly to
823 	 * dup to. Unlike dup() and dup2(), fcntl()'s F_DUPFD should
824 	 * return EINVAL when the new descriptor is out of bounds.
825 	 */
826 	if (old < 0)
827 		return (EBADF);
828 	if (new < 0)
829 		return (flags & DUP_FCNTL ? EINVAL : EBADF);
830 	PROC_LOCK(p);
831 	maxfd = min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc);
832 	PROC_UNLOCK(p);
833 	if (new >= maxfd)
834 		return (flags & DUP_FCNTL ? EINVAL : EBADF);
835 
836 	FILEDESC_XLOCK(fdp);
837 	if (fget_locked(fdp, old) == NULL) {
838 		FILEDESC_XUNLOCK(fdp);
839 		return (EBADF);
840 	}
841 	if (flags & DUP_FIXED && old == new) {
842 		*retval = new;
843 		if (flags & DUP_CLOEXEC)
844 			fdp->fd_ofileflags[new] |= UF_EXCLOSE;
845 		FILEDESC_XUNLOCK(fdp);
846 		return (0);
847 	}
848 	fp = fdp->fd_ofiles[old];
849 	fhold(fp);
850 
851 	/*
852 	 * If the caller specified a file descriptor, make sure the file
853 	 * table is large enough to hold it, and grab it.  Otherwise, just
854 	 * allocate a new descriptor the usual way.  Since the filedesc
855 	 * lock may be temporarily dropped in the process, we have to look
856 	 * out for a race.
857 	 */
858 	if (flags & DUP_FIXED) {
859 		if (new >= fdp->fd_nfiles) {
860 			/*
861 			 * The resource limits are here instead of e.g.
862 			 * fdalloc(), because the file descriptor table may be
863 			 * shared between processes, so we can't really use
864 			 * racct_add()/racct_sub().  Instead of counting the
865 			 * number of actually allocated descriptors, just put
866 			 * the limit on the size of the file descriptor table.
867 			 */
868 #ifdef RACCT
869 			PROC_LOCK(p);
870 			error = racct_set(p, RACCT_NOFILE, new + 1);
871 			PROC_UNLOCK(p);
872 			if (error != 0) {
873 				FILEDESC_XUNLOCK(fdp);
874 				fdrop(fp, td);
875 				return (EMFILE);
876 			}
877 #endif
878 			fdgrowtable(fdp, new + 1);
879 		}
880 		if (fdp->fd_ofiles[new] == NULL)
881 			fdused(fdp, new);
882 	} else {
883 		if ((error = fdalloc(td, new, &new)) != 0) {
884 			FILEDESC_XUNLOCK(fdp);
885 			fdrop(fp, td);
886 			return (error);
887 		}
888 	}
889 
890 	KASSERT(fp == fdp->fd_ofiles[old], ("old fd has been modified"));
891 	KASSERT(old != new, ("new fd is same as old"));
892 
893 	delfp = fdp->fd_ofiles[new];
894 	/*
895 	 * Duplicate the source descriptor.
896 	 */
897 	fdp->fd_ofiles[new] = fp;
898 	if ((flags & DUP_CLOEXEC) != 0)
899 		fdp->fd_ofileflags[new] = fdp->fd_ofileflags[old] | UF_EXCLOSE;
900 	else
901 		fdp->fd_ofileflags[new] = fdp->fd_ofileflags[old] & ~UF_EXCLOSE;
902 	if (new > fdp->fd_lastfile)
903 		fdp->fd_lastfile = new;
904 	*retval = new;
905 
906 	if (delfp != NULL) {
907 		(void) closefp(fdp, new, delfp, td, 1);
908 		/* closefp() drops the FILEDESC lock for us. */
909 	} else {
910 		FILEDESC_XUNLOCK(fdp);
911 	}
912 
913 	return (0);
914 }
915 
916 /*
917  * If sigio is on the list associated with a process or process group,
918  * disable signalling from the device, remove sigio from the list and
919  * free sigio.
920  */
921 void
922 funsetown(struct sigio **sigiop)
923 {
924 	struct sigio *sigio;
925 
926 	SIGIO_LOCK();
927 	sigio = *sigiop;
928 	if (sigio == NULL) {
929 		SIGIO_UNLOCK();
930 		return;
931 	}
932 	*(sigio->sio_myref) = NULL;
933 	if ((sigio)->sio_pgid < 0) {
934 		struct pgrp *pg = (sigio)->sio_pgrp;
935 		PGRP_LOCK(pg);
936 		SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio,
937 			     sigio, sio_pgsigio);
938 		PGRP_UNLOCK(pg);
939 	} else {
940 		struct proc *p = (sigio)->sio_proc;
941 		PROC_LOCK(p);
942 		SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio,
943 			     sigio, sio_pgsigio);
944 		PROC_UNLOCK(p);
945 	}
946 	SIGIO_UNLOCK();
947 	crfree(sigio->sio_ucred);
948 	free(sigio, M_SIGIO);
949 }
950 
951 /*
952  * Free a list of sigio structures.
953  * We only need to lock the SIGIO_LOCK because we have made ourselves
954  * inaccessible to callers of fsetown and therefore do not need to lock
955  * the proc or pgrp struct for the list manipulation.
956  */
957 void
958 funsetownlst(struct sigiolst *sigiolst)
959 {
960 	struct proc *p;
961 	struct pgrp *pg;
962 	struct sigio *sigio;
963 
964 	sigio = SLIST_FIRST(sigiolst);
965 	if (sigio == NULL)
966 		return;
967 	p = NULL;
968 	pg = NULL;
969 
970 	/*
971 	 * Every entry of the list should belong
972 	 * to a single proc or pgrp.
973 	 */
974 	if (sigio->sio_pgid < 0) {
975 		pg = sigio->sio_pgrp;
976 		PGRP_LOCK_ASSERT(pg, MA_NOTOWNED);
977 	} else /* if (sigio->sio_pgid > 0) */ {
978 		p = sigio->sio_proc;
979 		PROC_LOCK_ASSERT(p, MA_NOTOWNED);
980 	}
981 
982 	SIGIO_LOCK();
983 	while ((sigio = SLIST_FIRST(sigiolst)) != NULL) {
984 		*(sigio->sio_myref) = NULL;
985 		if (pg != NULL) {
986 			KASSERT(sigio->sio_pgid < 0,
987 			    ("Proc sigio in pgrp sigio list"));
988 			KASSERT(sigio->sio_pgrp == pg,
989 			    ("Bogus pgrp in sigio list"));
990 			PGRP_LOCK(pg);
991 			SLIST_REMOVE(&pg->pg_sigiolst, sigio, sigio,
992 			    sio_pgsigio);
993 			PGRP_UNLOCK(pg);
994 		} else /* if (p != NULL) */ {
995 			KASSERT(sigio->sio_pgid > 0,
996 			    ("Pgrp sigio in proc sigio list"));
997 			KASSERT(sigio->sio_proc == p,
998 			    ("Bogus proc in sigio list"));
999 			PROC_LOCK(p);
1000 			SLIST_REMOVE(&p->p_sigiolst, sigio, sigio,
1001 			    sio_pgsigio);
1002 			PROC_UNLOCK(p);
1003 		}
1004 		SIGIO_UNLOCK();
1005 		crfree(sigio->sio_ucred);
1006 		free(sigio, M_SIGIO);
1007 		SIGIO_LOCK();
1008 	}
1009 	SIGIO_UNLOCK();
1010 }
1011 
1012 /*
1013  * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg).
1014  *
1015  * After permission checking, add a sigio structure to the sigio list for
1016  * the process or process group.
1017  */
1018 int
1019 fsetown(pid_t pgid, struct sigio **sigiop)
1020 {
1021 	struct proc *proc;
1022 	struct pgrp *pgrp;
1023 	struct sigio *sigio;
1024 	int ret;
1025 
1026 	if (pgid == 0) {
1027 		funsetown(sigiop);
1028 		return (0);
1029 	}
1030 
1031 	ret = 0;
1032 
1033 	/* Allocate and fill in the new sigio out of locks. */
1034 	sigio = malloc(sizeof(struct sigio), M_SIGIO, M_WAITOK);
1035 	sigio->sio_pgid = pgid;
1036 	sigio->sio_ucred = crhold(curthread->td_ucred);
1037 	sigio->sio_myref = sigiop;
1038 
1039 	sx_slock(&proctree_lock);
1040 	if (pgid > 0) {
1041 		proc = pfind(pgid);
1042 		if (proc == NULL) {
1043 			ret = ESRCH;
1044 			goto fail;
1045 		}
1046 
1047 		/*
1048 		 * Policy - Don't allow a process to FSETOWN a process
1049 		 * in another session.
1050 		 *
1051 		 * Remove this test to allow maximum flexibility or
1052 		 * restrict FSETOWN to the current process or process
1053 		 * group for maximum safety.
1054 		 */
1055 		PROC_UNLOCK(proc);
1056 		if (proc->p_session != curthread->td_proc->p_session) {
1057 			ret = EPERM;
1058 			goto fail;
1059 		}
1060 
1061 		pgrp = NULL;
1062 	} else /* if (pgid < 0) */ {
1063 		pgrp = pgfind(-pgid);
1064 		if (pgrp == NULL) {
1065 			ret = ESRCH;
1066 			goto fail;
1067 		}
1068 		PGRP_UNLOCK(pgrp);
1069 
1070 		/*
1071 		 * Policy - Don't allow a process to FSETOWN a process
1072 		 * in another session.
1073 		 *
1074 		 * Remove this test to allow maximum flexibility or
1075 		 * restrict FSETOWN to the current process or process
1076 		 * group for maximum safety.
1077 		 */
1078 		if (pgrp->pg_session != curthread->td_proc->p_session) {
1079 			ret = EPERM;
1080 			goto fail;
1081 		}
1082 
1083 		proc = NULL;
1084 	}
1085 	funsetown(sigiop);
1086 	if (pgid > 0) {
1087 		PROC_LOCK(proc);
1088 		/*
1089 		 * Since funsetownlst() is called without the proctree
1090 		 * locked, we need to check for P_WEXIT.
1091 		 * XXX: is ESRCH correct?
1092 		 */
1093 		if ((proc->p_flag & P_WEXIT) != 0) {
1094 			PROC_UNLOCK(proc);
1095 			ret = ESRCH;
1096 			goto fail;
1097 		}
1098 		SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio);
1099 		sigio->sio_proc = proc;
1100 		PROC_UNLOCK(proc);
1101 	} else {
1102 		PGRP_LOCK(pgrp);
1103 		SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio);
1104 		sigio->sio_pgrp = pgrp;
1105 		PGRP_UNLOCK(pgrp);
1106 	}
1107 	sx_sunlock(&proctree_lock);
1108 	SIGIO_LOCK();
1109 	*sigiop = sigio;
1110 	SIGIO_UNLOCK();
1111 	return (0);
1112 
1113 fail:
1114 	sx_sunlock(&proctree_lock);
1115 	crfree(sigio->sio_ucred);
1116 	free(sigio, M_SIGIO);
1117 	return (ret);
1118 }
1119 
1120 /*
1121  * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg).
1122  */
1123 pid_t
1124 fgetown(sigiop)
1125 	struct sigio **sigiop;
1126 {
1127 	pid_t pgid;
1128 
1129 	SIGIO_LOCK();
1130 	pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0;
1131 	SIGIO_UNLOCK();
1132 	return (pgid);
1133 }
1134 
1135 /*
1136  * Function drops the filedesc lock on return.
1137  */
1138 static int
1139 closefp(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1140     int holdleaders)
1141 {
1142 	struct file *fp_object;
1143 	int error;
1144 
1145 	FILEDESC_XLOCK_ASSERT(fdp);
1146 
1147 	if (holdleaders) {
1148 		if (td->td_proc->p_fdtol != NULL) {
1149 			/*
1150 			 * Ask fdfree() to sleep to ensure that all relevant
1151 			 * process leaders can be traversed in closef().
1152 			 */
1153 			fdp->fd_holdleaderscount++;
1154 		} else {
1155 			holdleaders = 0;
1156 		}
1157 	}
1158 
1159 	/*
1160 	 * We now hold the fp reference that used to be owned by the
1161 	 * descriptor array.  We have to unlock the FILEDESC *AFTER*
1162 	 * knote_fdclose to prevent a race of the fd getting opened, a knote
1163 	 * added, and deleteing a knote for the new fd.
1164 	 */
1165 	knote_fdclose(td, fd);
1166 
1167 	/*
1168 	 * When we're closing an fd with a capability, we need to notify
1169 	 * mqueue if the underlying object is of type mqueue.
1170 	 */
1171 	(void)cap_funwrap(fp, 0, &fp_object);
1172 	if (fp_object->f_type == DTYPE_MQUEUE)
1173 		mq_fdclose(td, fd, fp_object);
1174 	FILEDESC_XUNLOCK(fdp);
1175 
1176 	error = closef(fp, td);
1177 	if (holdleaders) {
1178 		FILEDESC_XLOCK(fdp);
1179 		fdp->fd_holdleaderscount--;
1180 		if (fdp->fd_holdleaderscount == 0 &&
1181 		    fdp->fd_holdleaderswakeup != 0) {
1182 			fdp->fd_holdleaderswakeup = 0;
1183 			wakeup(&fdp->fd_holdleaderscount);
1184 		}
1185 		FILEDESC_XUNLOCK(fdp);
1186 	}
1187 	return (error);
1188 }
1189 
1190 /*
1191  * Close a file descriptor.
1192  */
1193 #ifndef _SYS_SYSPROTO_H_
1194 struct close_args {
1195 	int     fd;
1196 };
1197 #endif
1198 /* ARGSUSED */
1199 int
1200 sys_close(td, uap)
1201 	struct thread *td;
1202 	struct close_args *uap;
1203 {
1204 
1205 	return (kern_close(td, uap->fd));
1206 }
1207 
1208 int
1209 kern_close(td, fd)
1210 	struct thread *td;
1211 	int fd;
1212 {
1213 	struct filedesc *fdp;
1214 	struct file *fp;
1215 
1216 	fdp = td->td_proc->p_fd;
1217 
1218 	AUDIT_SYSCLOSE(td, fd);
1219 
1220 	FILEDESC_XLOCK(fdp);
1221 	if ((fp = fget_locked(fdp, fd)) == NULL) {
1222 		FILEDESC_XUNLOCK(fdp);
1223 		return (EBADF);
1224 	}
1225 	fdp->fd_ofiles[fd] = NULL;
1226 	fdp->fd_ofileflags[fd] = 0;
1227 	fdunused(fdp, fd);
1228 
1229 	/* closefp() drops the FILEDESC lock for us. */
1230 	return (closefp(fdp, fd, fp, td, 1));
1231 }
1232 
1233 /*
1234  * Close open file descriptors.
1235  */
1236 #ifndef _SYS_SYSPROTO_H_
1237 struct closefrom_args {
1238 	int	lowfd;
1239 };
1240 #endif
1241 /* ARGSUSED */
1242 int
1243 sys_closefrom(struct thread *td, struct closefrom_args *uap)
1244 {
1245 	struct filedesc *fdp;
1246 	int fd;
1247 
1248 	fdp = td->td_proc->p_fd;
1249 	AUDIT_ARG_FD(uap->lowfd);
1250 
1251 	/*
1252 	 * Treat negative starting file descriptor values identical to
1253 	 * closefrom(0) which closes all files.
1254 	 */
1255 	if (uap->lowfd < 0)
1256 		uap->lowfd = 0;
1257 	FILEDESC_SLOCK(fdp);
1258 	for (fd = uap->lowfd; fd < fdp->fd_nfiles; fd++) {
1259 		if (fdp->fd_ofiles[fd] != NULL) {
1260 			FILEDESC_SUNLOCK(fdp);
1261 			(void)kern_close(td, fd);
1262 			FILEDESC_SLOCK(fdp);
1263 		}
1264 	}
1265 	FILEDESC_SUNLOCK(fdp);
1266 	return (0);
1267 }
1268 
1269 #if defined(COMPAT_43)
1270 /*
1271  * Return status information about a file descriptor.
1272  */
1273 #ifndef _SYS_SYSPROTO_H_
1274 struct ofstat_args {
1275 	int	fd;
1276 	struct	ostat *sb;
1277 };
1278 #endif
1279 /* ARGSUSED */
1280 int
1281 ofstat(struct thread *td, struct ofstat_args *uap)
1282 {
1283 	struct ostat oub;
1284 	struct stat ub;
1285 	int error;
1286 
1287 	error = kern_fstat(td, uap->fd, &ub);
1288 	if (error == 0) {
1289 		cvtstat(&ub, &oub);
1290 		error = copyout(&oub, uap->sb, sizeof(oub));
1291 	}
1292 	return (error);
1293 }
1294 #endif /* COMPAT_43 */
1295 
1296 /*
1297  * Return status information about a file descriptor.
1298  */
1299 #ifndef _SYS_SYSPROTO_H_
1300 struct fstat_args {
1301 	int	fd;
1302 	struct	stat *sb;
1303 };
1304 #endif
1305 /* ARGSUSED */
1306 int
1307 sys_fstat(struct thread *td, struct fstat_args *uap)
1308 {
1309 	struct stat ub;
1310 	int error;
1311 
1312 	error = kern_fstat(td, uap->fd, &ub);
1313 	if (error == 0)
1314 		error = copyout(&ub, uap->sb, sizeof(ub));
1315 	return (error);
1316 }
1317 
1318 int
1319 kern_fstat(struct thread *td, int fd, struct stat *sbp)
1320 {
1321 	struct file *fp;
1322 	int error;
1323 
1324 	AUDIT_ARG_FD(fd);
1325 
1326 	if ((error = fget(td, fd, CAP_FSTAT, &fp)) != 0)
1327 		return (error);
1328 
1329 	AUDIT_ARG_FILE(td->td_proc, fp);
1330 
1331 	error = fo_stat(fp, sbp, td->td_ucred, td);
1332 	fdrop(fp, td);
1333 #ifdef KTRACE
1334 	if (error == 0 && KTRPOINT(td, KTR_STRUCT))
1335 		ktrstat(sbp);
1336 #endif
1337 	return (error);
1338 }
1339 
1340 /*
1341  * Return status information about a file descriptor.
1342  */
1343 #ifndef _SYS_SYSPROTO_H_
1344 struct nfstat_args {
1345 	int	fd;
1346 	struct	nstat *sb;
1347 };
1348 #endif
1349 /* ARGSUSED */
1350 int
1351 sys_nfstat(struct thread *td, struct nfstat_args *uap)
1352 {
1353 	struct nstat nub;
1354 	struct stat ub;
1355 	int error;
1356 
1357 	error = kern_fstat(td, uap->fd, &ub);
1358 	if (error == 0) {
1359 		cvtnstat(&ub, &nub);
1360 		error = copyout(&nub, uap->sb, sizeof(nub));
1361 	}
1362 	return (error);
1363 }
1364 
1365 /*
1366  * Return pathconf information about a file descriptor.
1367  */
1368 #ifndef _SYS_SYSPROTO_H_
1369 struct fpathconf_args {
1370 	int	fd;
1371 	int	name;
1372 };
1373 #endif
1374 /* ARGSUSED */
1375 int
1376 sys_fpathconf(struct thread *td, struct fpathconf_args *uap)
1377 {
1378 	struct file *fp;
1379 	struct vnode *vp;
1380 	int error;
1381 
1382 	if ((error = fget(td, uap->fd, CAP_FPATHCONF, &fp)) != 0)
1383 		return (error);
1384 
1385 	/* If asynchronous I/O is available, it works for all descriptors. */
1386 	if (uap->name == _PC_ASYNC_IO) {
1387 		td->td_retval[0] = async_io_version;
1388 		goto out;
1389 	}
1390 	vp = fp->f_vnode;
1391 	if (vp != NULL) {
1392 		vn_lock(vp, LK_SHARED | LK_RETRY);
1393 		error = VOP_PATHCONF(vp, uap->name, td->td_retval);
1394 		VOP_UNLOCK(vp, 0);
1395 	} else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) {
1396 		if (uap->name != _PC_PIPE_BUF) {
1397 			error = EINVAL;
1398 		} else {
1399 			td->td_retval[0] = PIPE_BUF;
1400 			error = 0;
1401 		}
1402 	} else {
1403 		error = EOPNOTSUPP;
1404 	}
1405 out:
1406 	fdrop(fp, td);
1407 	return (error);
1408 }
1409 
1410 /*
1411  * Grow the file table to accomodate (at least) nfd descriptors.
1412  */
1413 static void
1414 fdgrowtable(struct filedesc *fdp, int nfd)
1415 {
1416 	struct filedesc0 *fdp0;
1417 	struct freetable *fo;
1418 	struct file **ntable;
1419 	struct file **otable;
1420 	char *nfileflags;
1421 	int nnfiles, onfiles;
1422 	NDSLOTTYPE *nmap;
1423 
1424 	FILEDESC_XLOCK_ASSERT(fdp);
1425 
1426 	KASSERT(fdp->fd_nfiles > 0,
1427 	    ("zero-length file table"));
1428 
1429 	/* compute the size of the new table */
1430 	onfiles = fdp->fd_nfiles;
1431 	nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */
1432 	if (nnfiles <= onfiles)
1433 		/* the table is already large enough */
1434 		return;
1435 
1436 	/* allocate a new table and (if required) new bitmaps */
1437 	ntable = malloc((nnfiles * OFILESIZE) + sizeof(struct freetable),
1438 	    M_FILEDESC, M_ZERO | M_WAITOK);
1439 	nfileflags = (char *)&ntable[nnfiles];
1440 	if (NDSLOTS(nnfiles) > NDSLOTS(onfiles))
1441 		nmap = malloc(NDSLOTS(nnfiles) * NDSLOTSIZE,
1442 		    M_FILEDESC, M_ZERO | M_WAITOK);
1443 	else
1444 		nmap = NULL;
1445 
1446 	bcopy(fdp->fd_ofiles, ntable, onfiles * sizeof(*ntable));
1447 	bcopy(fdp->fd_ofileflags, nfileflags, onfiles);
1448 	otable = fdp->fd_ofiles;
1449 	fdp->fd_ofileflags = nfileflags;
1450 	fdp->fd_ofiles = ntable;
1451 	/*
1452 	 * We must preserve ofiles until the process exits because we can't
1453 	 * be certain that no threads have references to the old table via
1454 	 * _fget().
1455 	 */
1456 	if (onfiles > NDFILE) {
1457 		fo = (struct freetable *)&otable[onfiles];
1458 		fdp0 = (struct filedesc0 *)fdp;
1459 		fo->ft_table = otable;
1460 		SLIST_INSERT_HEAD(&fdp0->fd_free, fo, ft_next);
1461 	}
1462 	if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) {
1463 		bcopy(fdp->fd_map, nmap, NDSLOTS(onfiles) * sizeof(*nmap));
1464 		if (NDSLOTS(onfiles) > NDSLOTS(NDFILE))
1465 			free(fdp->fd_map, M_FILEDESC);
1466 		fdp->fd_map = nmap;
1467 	}
1468 	fdp->fd_nfiles = nnfiles;
1469 }
1470 
1471 /*
1472  * Allocate a file descriptor for the process.
1473  */
1474 int
1475 fdalloc(struct thread *td, int minfd, int *result)
1476 {
1477 	struct proc *p = td->td_proc;
1478 	struct filedesc *fdp = p->p_fd;
1479 	int fd = -1, maxfd, allocfd;
1480 #ifdef RACCT
1481 	int error;
1482 #endif
1483 
1484 	FILEDESC_XLOCK_ASSERT(fdp);
1485 
1486 	if (fdp->fd_freefile > minfd)
1487 		minfd = fdp->fd_freefile;
1488 
1489 	PROC_LOCK(p);
1490 	maxfd = min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc);
1491 	PROC_UNLOCK(p);
1492 
1493 	/*
1494 	 * Search the bitmap for a free descriptor starting at minfd.
1495 	 * If none is found, grow the file table.
1496 	 */
1497 	fd = fd_first_free(fdp, minfd, fdp->fd_nfiles);
1498 	if (fd >= maxfd)
1499 		return (EMFILE);
1500 	if (fd >= fdp->fd_nfiles) {
1501 		allocfd = min(fd * 2, maxfd);
1502 #ifdef RACCT
1503 		PROC_LOCK(p);
1504 		error = racct_set(p, RACCT_NOFILE, allocfd);
1505 		PROC_UNLOCK(p);
1506 		if (error != 0)
1507 			return (EMFILE);
1508 #endif
1509 		/*
1510 		 * fd is already equal to first free descriptor >= minfd, so
1511 		 * we only need to grow the table and we are done.
1512 		 */
1513 		fdgrowtable(fdp, allocfd);
1514 	}
1515 
1516 	/*
1517 	 * Perform some sanity checks, then mark the file descriptor as
1518 	 * used and return it to the caller.
1519 	 */
1520 	KASSERT(fd >= 0 && fd < min(maxfd, fdp->fd_nfiles),
1521 	    ("invalid descriptor %d", fd));
1522 	KASSERT(!fdisused(fdp, fd),
1523 	    ("fd_first_free() returned non-free descriptor"));
1524 	KASSERT(fdp->fd_ofiles[fd] == NULL, ("file descriptor isn't free"));
1525 	KASSERT(fdp->fd_ofileflags[fd] == 0, ("file flags are set"));
1526 	fdused(fdp, fd);
1527 	*result = fd;
1528 	return (0);
1529 }
1530 
1531 /*
1532  * Check to see whether n user file descriptors are available to the process
1533  * p.
1534  */
1535 int
1536 fdavail(struct thread *td, int n)
1537 {
1538 	struct proc *p = td->td_proc;
1539 	struct filedesc *fdp = td->td_proc->p_fd;
1540 	int i, lim, last;
1541 
1542 	FILEDESC_LOCK_ASSERT(fdp);
1543 
1544 	/*
1545 	 * XXX: This is only called from uipc_usrreq.c:unp_externalize();
1546 	 *      call racct_add() from there instead of dealing with containers
1547 	 *      here.
1548 	 */
1549 	PROC_LOCK(p);
1550 	lim = min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc);
1551 	PROC_UNLOCK(p);
1552 	if ((i = lim - fdp->fd_nfiles) > 0 && (n -= i) <= 0)
1553 		return (1);
1554 	last = min(fdp->fd_nfiles, lim);
1555 	for (i = fdp->fd_freefile; i < last; i++) {
1556 		if (fdp->fd_ofiles[i] == NULL && --n <= 0)
1557 			return (1);
1558 	}
1559 	return (0);
1560 }
1561 
1562 /*
1563  * Create a new open file structure and allocate a file decriptor for the
1564  * process that refers to it.  We add one reference to the file for the
1565  * descriptor table and one reference for resultfp. This is to prevent us
1566  * being preempted and the entry in the descriptor table closed after we
1567  * release the FILEDESC lock.
1568  */
1569 int
1570 falloc(struct thread *td, struct file **resultfp, int *resultfd, int flags)
1571 {
1572 	struct file *fp;
1573 	int error, fd;
1574 
1575 	error = falloc_noinstall(td, &fp);
1576 	if (error)
1577 		return (error);		/* no reference held on error */
1578 
1579 	error = finstall(td, fp, &fd, flags);
1580 	if (error) {
1581 		fdrop(fp, td);		/* one reference (fp only) */
1582 		return (error);
1583 	}
1584 
1585 	if (resultfp != NULL)
1586 		*resultfp = fp;		/* copy out result */
1587 	else
1588 		fdrop(fp, td);		/* release local reference */
1589 
1590 	if (resultfd != NULL)
1591 		*resultfd = fd;
1592 
1593 	return (0);
1594 }
1595 
1596 /*
1597  * Create a new open file structure without allocating a file descriptor.
1598  */
1599 int
1600 falloc_noinstall(struct thread *td, struct file **resultfp)
1601 {
1602 	struct file *fp;
1603 	int maxuserfiles = maxfiles - (maxfiles / 20);
1604 	static struct timeval lastfail;
1605 	static int curfail;
1606 
1607 	KASSERT(resultfp != NULL, ("%s: resultfp == NULL", __func__));
1608 
1609 	if ((openfiles >= maxuserfiles &&
1610 	    priv_check(td, PRIV_MAXFILES) != 0) ||
1611 	    openfiles >= maxfiles) {
1612 		if (ppsratecheck(&lastfail, &curfail, 1)) {
1613 			printf("kern.maxfiles limit exceeded by uid %i, "
1614 			    "please see tuning(7).\n", td->td_ucred->cr_ruid);
1615 		}
1616 		return (ENFILE);
1617 	}
1618 	atomic_add_int(&openfiles, 1);
1619 	fp = uma_zalloc(file_zone, M_WAITOK | M_ZERO);
1620 	refcount_init(&fp->f_count, 1);
1621 	fp->f_cred = crhold(td->td_ucred);
1622 	fp->f_ops = &badfileops;
1623 	fp->f_data = NULL;
1624 	fp->f_vnode = NULL;
1625 	*resultfp = fp;
1626 	return (0);
1627 }
1628 
1629 /*
1630  * Install a file in a file descriptor table.
1631  */
1632 int
1633 finstall(struct thread *td, struct file *fp, int *fd, int flags)
1634 {
1635 	struct filedesc *fdp = td->td_proc->p_fd;
1636 	int error;
1637 
1638 	KASSERT(fd != NULL, ("%s: fd == NULL", __func__));
1639 	KASSERT(fp != NULL, ("%s: fp == NULL", __func__));
1640 
1641 	FILEDESC_XLOCK(fdp);
1642 	if ((error = fdalloc(td, 0, fd))) {
1643 		FILEDESC_XUNLOCK(fdp);
1644 		return (error);
1645 	}
1646 	fhold(fp);
1647 	fdp->fd_ofiles[*fd] = fp;
1648 	if ((flags & O_CLOEXEC) != 0)
1649 		fdp->fd_ofileflags[*fd] |= UF_EXCLOSE;
1650 	FILEDESC_XUNLOCK(fdp);
1651 	return (0);
1652 }
1653 
1654 /*
1655  * Build a new filedesc structure from another.
1656  * Copy the current, root, and jail root vnode references.
1657  */
1658 struct filedesc *
1659 fdinit(struct filedesc *fdp)
1660 {
1661 	struct filedesc0 *newfdp;
1662 
1663 	newfdp = malloc(sizeof *newfdp, M_FILEDESC, M_WAITOK | M_ZERO);
1664 	FILEDESC_LOCK_INIT(&newfdp->fd_fd);
1665 	if (fdp != NULL) {
1666 		FILEDESC_XLOCK(fdp);
1667 		newfdp->fd_fd.fd_cdir = fdp->fd_cdir;
1668 		if (newfdp->fd_fd.fd_cdir)
1669 			VREF(newfdp->fd_fd.fd_cdir);
1670 		newfdp->fd_fd.fd_rdir = fdp->fd_rdir;
1671 		if (newfdp->fd_fd.fd_rdir)
1672 			VREF(newfdp->fd_fd.fd_rdir);
1673 		newfdp->fd_fd.fd_jdir = fdp->fd_jdir;
1674 		if (newfdp->fd_fd.fd_jdir)
1675 			VREF(newfdp->fd_fd.fd_jdir);
1676 		FILEDESC_XUNLOCK(fdp);
1677 	}
1678 
1679 	/* Create the file descriptor table. */
1680 	newfdp->fd_fd.fd_refcnt = 1;
1681 	newfdp->fd_fd.fd_holdcnt = 1;
1682 	newfdp->fd_fd.fd_cmask = CMASK;
1683 	newfdp->fd_fd.fd_ofiles = newfdp->fd_dfiles;
1684 	newfdp->fd_fd.fd_ofileflags = newfdp->fd_dfileflags;
1685 	newfdp->fd_fd.fd_nfiles = NDFILE;
1686 	newfdp->fd_fd.fd_map = newfdp->fd_dmap;
1687 	newfdp->fd_fd.fd_lastfile = -1;
1688 	return (&newfdp->fd_fd);
1689 }
1690 
1691 static struct filedesc *
1692 fdhold(struct proc *p)
1693 {
1694 	struct filedesc *fdp;
1695 
1696 	mtx_lock(&fdesc_mtx);
1697 	fdp = p->p_fd;
1698 	if (fdp != NULL)
1699 		fdp->fd_holdcnt++;
1700 	mtx_unlock(&fdesc_mtx);
1701 	return (fdp);
1702 }
1703 
1704 static void
1705 fddrop(struct filedesc *fdp)
1706 {
1707 	struct filedesc0 *fdp0;
1708 	struct freetable *ft;
1709 	int i;
1710 
1711 	mtx_lock(&fdesc_mtx);
1712 	i = --fdp->fd_holdcnt;
1713 	mtx_unlock(&fdesc_mtx);
1714 	if (i > 0)
1715 		return;
1716 
1717 	FILEDESC_LOCK_DESTROY(fdp);
1718 	fdp0 = (struct filedesc0 *)fdp;
1719 	while ((ft = SLIST_FIRST(&fdp0->fd_free)) != NULL) {
1720 		SLIST_REMOVE_HEAD(&fdp0->fd_free, ft_next);
1721 		free(ft->ft_table, M_FILEDESC);
1722 	}
1723 	free(fdp, M_FILEDESC);
1724 }
1725 
1726 /*
1727  * Share a filedesc structure.
1728  */
1729 struct filedesc *
1730 fdshare(struct filedesc *fdp)
1731 {
1732 
1733 	FILEDESC_XLOCK(fdp);
1734 	fdp->fd_refcnt++;
1735 	FILEDESC_XUNLOCK(fdp);
1736 	return (fdp);
1737 }
1738 
1739 /*
1740  * Unshare a filedesc structure, if necessary by making a copy
1741  */
1742 void
1743 fdunshare(struct proc *p, struct thread *td)
1744 {
1745 
1746 	FILEDESC_XLOCK(p->p_fd);
1747 	if (p->p_fd->fd_refcnt > 1) {
1748 		struct filedesc *tmp;
1749 
1750 		FILEDESC_XUNLOCK(p->p_fd);
1751 		tmp = fdcopy(p->p_fd);
1752 		fdfree(td);
1753 		p->p_fd = tmp;
1754 	} else
1755 		FILEDESC_XUNLOCK(p->p_fd);
1756 }
1757 
1758 /*
1759  * Copy a filedesc structure.  A NULL pointer in returns a NULL reference,
1760  * this is to ease callers, not catch errors.
1761  */
1762 struct filedesc *
1763 fdcopy(struct filedesc *fdp)
1764 {
1765 	struct filedesc *newfdp;
1766 	int i;
1767 
1768 	/* Certain daemons might not have file descriptors. */
1769 	if (fdp == NULL)
1770 		return (NULL);
1771 
1772 	newfdp = fdinit(fdp);
1773 	FILEDESC_SLOCK(fdp);
1774 	while (fdp->fd_lastfile >= newfdp->fd_nfiles) {
1775 		FILEDESC_SUNLOCK(fdp);
1776 		FILEDESC_XLOCK(newfdp);
1777 		fdgrowtable(newfdp, fdp->fd_lastfile + 1);
1778 		FILEDESC_XUNLOCK(newfdp);
1779 		FILEDESC_SLOCK(fdp);
1780 	}
1781 	/* copy all passable descriptors (i.e. not kqueue) */
1782 	newfdp->fd_freefile = -1;
1783 	for (i = 0; i <= fdp->fd_lastfile; ++i) {
1784 		if (fdisused(fdp, i) &&
1785 		    (fdp->fd_ofiles[i]->f_ops->fo_flags & DFLAG_PASSABLE) &&
1786 		    fdp->fd_ofiles[i]->f_ops != &badfileops) {
1787 			newfdp->fd_ofiles[i] = fdp->fd_ofiles[i];
1788 			newfdp->fd_ofileflags[i] = fdp->fd_ofileflags[i];
1789 			fhold(newfdp->fd_ofiles[i]);
1790 			newfdp->fd_lastfile = i;
1791 		} else {
1792 			if (newfdp->fd_freefile == -1)
1793 				newfdp->fd_freefile = i;
1794 		}
1795 	}
1796 	newfdp->fd_cmask = fdp->fd_cmask;
1797 	FILEDESC_SUNLOCK(fdp);
1798 	FILEDESC_XLOCK(newfdp);
1799 	for (i = 0; i <= newfdp->fd_lastfile; ++i)
1800 		if (newfdp->fd_ofiles[i] != NULL)
1801 			fdused(newfdp, i);
1802 	if (newfdp->fd_freefile == -1)
1803 		newfdp->fd_freefile = i;
1804 	FILEDESC_XUNLOCK(newfdp);
1805 	return (newfdp);
1806 }
1807 
1808 /*
1809  * Release a filedesc structure.
1810  */
1811 void
1812 fdfree(struct thread *td)
1813 {
1814 	struct filedesc *fdp;
1815 	int i;
1816 	struct filedesc_to_leader *fdtol;
1817 	struct file *fp;
1818 	struct vnode *cdir, *jdir, *rdir, *vp;
1819 	struct flock lf;
1820 
1821 	/* Certain daemons might not have file descriptors. */
1822 	fdp = td->td_proc->p_fd;
1823 	if (fdp == NULL)
1824 		return;
1825 
1826 #ifdef RACCT
1827 	PROC_LOCK(td->td_proc);
1828 	racct_set(td->td_proc, RACCT_NOFILE, 0);
1829 	PROC_UNLOCK(td->td_proc);
1830 #endif
1831 
1832 	/* Check for special need to clear POSIX style locks */
1833 	fdtol = td->td_proc->p_fdtol;
1834 	if (fdtol != NULL) {
1835 		FILEDESC_XLOCK(fdp);
1836 		KASSERT(fdtol->fdl_refcount > 0,
1837 			("filedesc_to_refcount botch: fdl_refcount=%d",
1838 			 fdtol->fdl_refcount));
1839 		if (fdtol->fdl_refcount == 1 &&
1840 		    (td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
1841 			for (i = 0; i <= fdp->fd_lastfile; i++) {
1842 				fp = fdp->fd_ofiles[i];
1843 				if (fp == NULL || fp->f_type != DTYPE_VNODE)
1844 					continue;
1845 				fhold(fp);
1846 				FILEDESC_XUNLOCK(fdp);
1847 				lf.l_whence = SEEK_SET;
1848 				lf.l_start = 0;
1849 				lf.l_len = 0;
1850 				lf.l_type = F_UNLCK;
1851 				vp = fp->f_vnode;
1852 				(void) VOP_ADVLOCK(vp,
1853 				    (caddr_t)td->td_proc->p_leader, F_UNLCK,
1854 				    &lf, F_POSIX);
1855 				FILEDESC_XLOCK(fdp);
1856 				fdrop(fp, td);
1857 			}
1858 		}
1859 	retry:
1860 		if (fdtol->fdl_refcount == 1) {
1861 			if (fdp->fd_holdleaderscount > 0 &&
1862 			    (td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
1863 				/*
1864 				 * close() or do_dup() has cleared a reference
1865 				 * in a shared file descriptor table.
1866 				 */
1867 				fdp->fd_holdleaderswakeup = 1;
1868 				sx_sleep(&fdp->fd_holdleaderscount,
1869 				    FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0);
1870 				goto retry;
1871 			}
1872 			if (fdtol->fdl_holdcount > 0) {
1873 				/*
1874 				 * Ensure that fdtol->fdl_leader remains
1875 				 * valid in closef().
1876 				 */
1877 				fdtol->fdl_wakeup = 1;
1878 				sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK,
1879 				    "fdlhold", 0);
1880 				goto retry;
1881 			}
1882 		}
1883 		fdtol->fdl_refcount--;
1884 		if (fdtol->fdl_refcount == 0 &&
1885 		    fdtol->fdl_holdcount == 0) {
1886 			fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
1887 			fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
1888 		} else
1889 			fdtol = NULL;
1890 		td->td_proc->p_fdtol = NULL;
1891 		FILEDESC_XUNLOCK(fdp);
1892 		if (fdtol != NULL)
1893 			free(fdtol, M_FILEDESC_TO_LEADER);
1894 	}
1895 	FILEDESC_XLOCK(fdp);
1896 	i = --fdp->fd_refcnt;
1897 	FILEDESC_XUNLOCK(fdp);
1898 	if (i > 0)
1899 		return;
1900 
1901 	for (i = 0; i <= fdp->fd_lastfile; i++) {
1902 		fp = fdp->fd_ofiles[i];
1903 		if (fp != NULL) {
1904 			FILEDESC_XLOCK(fdp);
1905 			fdp->fd_ofiles[i] = NULL;
1906 			FILEDESC_XUNLOCK(fdp);
1907 			(void) closef(fp, td);
1908 		}
1909 	}
1910 	FILEDESC_XLOCK(fdp);
1911 
1912 	/* XXX This should happen earlier. */
1913 	mtx_lock(&fdesc_mtx);
1914 	td->td_proc->p_fd = NULL;
1915 	mtx_unlock(&fdesc_mtx);
1916 
1917 	if (fdp->fd_nfiles > NDFILE)
1918 		free(fdp->fd_ofiles, M_FILEDESC);
1919 	if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE))
1920 		free(fdp->fd_map, M_FILEDESC);
1921 
1922 	fdp->fd_nfiles = 0;
1923 
1924 	cdir = fdp->fd_cdir;
1925 	fdp->fd_cdir = NULL;
1926 	rdir = fdp->fd_rdir;
1927 	fdp->fd_rdir = NULL;
1928 	jdir = fdp->fd_jdir;
1929 	fdp->fd_jdir = NULL;
1930 	FILEDESC_XUNLOCK(fdp);
1931 
1932 	if (cdir)
1933 		vrele(cdir);
1934 	if (rdir)
1935 		vrele(rdir);
1936 	if (jdir)
1937 		vrele(jdir);
1938 
1939 	fddrop(fdp);
1940 }
1941 
1942 /*
1943  * For setugid programs, we don't want to people to use that setugidness
1944  * to generate error messages which write to a file which otherwise would
1945  * otherwise be off-limits to the process.  We check for filesystems where
1946  * the vnode can change out from under us after execve (like [lin]procfs).
1947  *
1948  * Since setugidsafety calls this only for fd 0, 1 and 2, this check is
1949  * sufficient.  We also don't check for setugidness since we know we are.
1950  */
1951 static int
1952 is_unsafe(struct file *fp)
1953 {
1954 	if (fp->f_type == DTYPE_VNODE) {
1955 		struct vnode *vp = fp->f_vnode;
1956 
1957 		if ((vp->v_vflag & VV_PROCDEP) != 0)
1958 			return (1);
1959 	}
1960 	return (0);
1961 }
1962 
1963 /*
1964  * Make this setguid thing safe, if at all possible.
1965  */
1966 void
1967 setugidsafety(struct thread *td)
1968 {
1969 	struct filedesc *fdp;
1970 	int i;
1971 
1972 	/* Certain daemons might not have file descriptors. */
1973 	fdp = td->td_proc->p_fd;
1974 	if (fdp == NULL)
1975 		return;
1976 
1977 	/*
1978 	 * Note: fdp->fd_ofiles may be reallocated out from under us while
1979 	 * we are blocked in a close.  Be careful!
1980 	 */
1981 	FILEDESC_XLOCK(fdp);
1982 	for (i = 0; i <= fdp->fd_lastfile; i++) {
1983 		if (i > 2)
1984 			break;
1985 		if (fdp->fd_ofiles[i] && is_unsafe(fdp->fd_ofiles[i])) {
1986 			struct file *fp;
1987 
1988 			knote_fdclose(td, i);
1989 			/*
1990 			 * NULL-out descriptor prior to close to avoid
1991 			 * a race while close blocks.
1992 			 */
1993 			fp = fdp->fd_ofiles[i];
1994 			fdp->fd_ofiles[i] = NULL;
1995 			fdp->fd_ofileflags[i] = 0;
1996 			fdunused(fdp, i);
1997 			FILEDESC_XUNLOCK(fdp);
1998 			(void) closef(fp, td);
1999 			FILEDESC_XLOCK(fdp);
2000 		}
2001 	}
2002 	FILEDESC_XUNLOCK(fdp);
2003 }
2004 
2005 /*
2006  * If a specific file object occupies a specific file descriptor, close the
2007  * file descriptor entry and drop a reference on the file object.  This is a
2008  * convenience function to handle a subsequent error in a function that calls
2009  * falloc() that handles the race that another thread might have closed the
2010  * file descriptor out from under the thread creating the file object.
2011  */
2012 void
2013 fdclose(struct filedesc *fdp, struct file *fp, int idx, struct thread *td)
2014 {
2015 
2016 	FILEDESC_XLOCK(fdp);
2017 	if (fdp->fd_ofiles[idx] == fp) {
2018 		fdp->fd_ofiles[idx] = NULL;
2019 		fdunused(fdp, idx);
2020 		FILEDESC_XUNLOCK(fdp);
2021 		fdrop(fp, td);
2022 	} else
2023 		FILEDESC_XUNLOCK(fdp);
2024 }
2025 
2026 /*
2027  * Close any files on exec?
2028  */
2029 void
2030 fdcloseexec(struct thread *td)
2031 {
2032 	struct filedesc *fdp;
2033 	struct file *fp;
2034 	int i;
2035 
2036 	/* Certain daemons might not have file descriptors. */
2037 	fdp = td->td_proc->p_fd;
2038 	if (fdp == NULL)
2039 		return;
2040 
2041 	/*
2042 	 * We cannot cache fd_ofiles or fd_ofileflags since operations
2043 	 * may block and rip them out from under us.
2044 	 */
2045 	FILEDESC_XLOCK(fdp);
2046 	for (i = 0; i <= fdp->fd_lastfile; i++) {
2047 		fp = fdp->fd_ofiles[i];
2048 		if (fp != NULL && (fp->f_type == DTYPE_MQUEUE ||
2049 		    (fdp->fd_ofileflags[i] & UF_EXCLOSE))) {
2050 			fdp->fd_ofiles[i] = NULL;
2051 			fdp->fd_ofileflags[i] = 0;
2052 			fdunused(fdp, i);
2053 			(void) closefp(fdp, i, fp, td, 0);
2054 			/* closefp() drops the FILEDESC lock. */
2055 			FILEDESC_XLOCK(fdp);
2056 		}
2057 	}
2058 	FILEDESC_XUNLOCK(fdp);
2059 }
2060 
2061 /*
2062  * It is unsafe for set[ug]id processes to be started with file
2063  * descriptors 0..2 closed, as these descriptors are given implicit
2064  * significance in the Standard C library.  fdcheckstd() will create a
2065  * descriptor referencing /dev/null for each of stdin, stdout, and
2066  * stderr that is not already open.
2067  */
2068 int
2069 fdcheckstd(struct thread *td)
2070 {
2071 	struct filedesc *fdp;
2072 	register_t retval, save;
2073 	int i, error, devnull;
2074 
2075 	fdp = td->td_proc->p_fd;
2076 	if (fdp == NULL)
2077 		return (0);
2078 	KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2079 	devnull = -1;
2080 	error = 0;
2081 	for (i = 0; i < 3; i++) {
2082 		if (fdp->fd_ofiles[i] != NULL)
2083 			continue;
2084 		if (devnull < 0) {
2085 			save = td->td_retval[0];
2086 			error = kern_open(td, "/dev/null", UIO_SYSSPACE,
2087 			    O_RDWR, 0);
2088 			devnull = td->td_retval[0];
2089 			td->td_retval[0] = save;
2090 			if (error)
2091 				break;
2092 			KASSERT(devnull == i, ("oof, we didn't get our fd"));
2093 		} else {
2094 			error = do_dup(td, DUP_FIXED, devnull, i, &retval);
2095 			if (error != 0)
2096 				break;
2097 		}
2098 	}
2099 	return (error);
2100 }
2101 
2102 /*
2103  * Internal form of close.  Decrement reference count on file structure.
2104  * Note: td may be NULL when closing a file that was being passed in a
2105  * message.
2106  *
2107  * XXXRW: Giant is not required for the caller, but often will be held; this
2108  * makes it moderately likely the Giant will be recursed in the VFS case.
2109  */
2110 int
2111 closef(struct file *fp, struct thread *td)
2112 {
2113 	struct vnode *vp;
2114 	struct flock lf;
2115 	struct filedesc_to_leader *fdtol;
2116 	struct filedesc *fdp;
2117 	struct file *fp_object;
2118 
2119 	/*
2120 	 * POSIX record locking dictates that any close releases ALL
2121 	 * locks owned by this process.  This is handled by setting
2122 	 * a flag in the unlock to free ONLY locks obeying POSIX
2123 	 * semantics, and not to free BSD-style file locks.
2124 	 * If the descriptor was in a message, POSIX-style locks
2125 	 * aren't passed with the descriptor, and the thread pointer
2126 	 * will be NULL.  Callers should be careful only to pass a
2127 	 * NULL thread pointer when there really is no owning
2128 	 * context that might have locks, or the locks will be
2129 	 * leaked.
2130 	 *
2131 	 * If this is a capability, we do lock processing under the underlying
2132 	 * node, not the capability itself.
2133 	 */
2134 	(void)cap_funwrap(fp, 0, &fp_object);
2135 	if (fp_object->f_type == DTYPE_VNODE && td != NULL) {
2136 		vp = fp_object->f_vnode;
2137 		if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
2138 			lf.l_whence = SEEK_SET;
2139 			lf.l_start = 0;
2140 			lf.l_len = 0;
2141 			lf.l_type = F_UNLCK;
2142 			(void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader,
2143 			    F_UNLCK, &lf, F_POSIX);
2144 		}
2145 		fdtol = td->td_proc->p_fdtol;
2146 		if (fdtol != NULL) {
2147 			/*
2148 			 * Handle special case where file descriptor table is
2149 			 * shared between multiple process leaders.
2150 			 */
2151 			fdp = td->td_proc->p_fd;
2152 			FILEDESC_XLOCK(fdp);
2153 			for (fdtol = fdtol->fdl_next;
2154 			     fdtol != td->td_proc->p_fdtol;
2155 			     fdtol = fdtol->fdl_next) {
2156 				if ((fdtol->fdl_leader->p_flag &
2157 				     P_ADVLOCK) == 0)
2158 					continue;
2159 				fdtol->fdl_holdcount++;
2160 				FILEDESC_XUNLOCK(fdp);
2161 				lf.l_whence = SEEK_SET;
2162 				lf.l_start = 0;
2163 				lf.l_len = 0;
2164 				lf.l_type = F_UNLCK;
2165 				vp = fp_object->f_vnode;
2166 				(void) VOP_ADVLOCK(vp,
2167 				    (caddr_t)fdtol->fdl_leader, F_UNLCK, &lf,
2168 				    F_POSIX);
2169 				FILEDESC_XLOCK(fdp);
2170 				fdtol->fdl_holdcount--;
2171 				if (fdtol->fdl_holdcount == 0 &&
2172 				    fdtol->fdl_wakeup != 0) {
2173 					fdtol->fdl_wakeup = 0;
2174 					wakeup(fdtol);
2175 				}
2176 			}
2177 			FILEDESC_XUNLOCK(fdp);
2178 		}
2179 	}
2180 	return (fdrop(fp, td));
2181 }
2182 
2183 /*
2184  * Initialize the file pointer with the specified properties.
2185  *
2186  * The ops are set with release semantics to be certain that the flags, type,
2187  * and data are visible when ops is.  This is to prevent ops methods from being
2188  * called with bad data.
2189  */
2190 void
2191 finit(struct file *fp, u_int flag, short type, void *data, struct fileops *ops)
2192 {
2193 	fp->f_data = data;
2194 	fp->f_flag = flag;
2195 	fp->f_type = type;
2196 	atomic_store_rel_ptr((volatile uintptr_t *)&fp->f_ops, (uintptr_t)ops);
2197 }
2198 
2199 struct file *
2200 fget_unlocked(struct filedesc *fdp, int fd)
2201 {
2202 	struct file *fp;
2203 	u_int count;
2204 
2205 	if (fd < 0 || fd >= fdp->fd_nfiles)
2206 		return (NULL);
2207 	/*
2208 	 * Fetch the descriptor locklessly.  We avoid fdrop() races by
2209 	 * never raising a refcount above 0.  To accomplish this we have
2210 	 * to use a cmpset loop rather than an atomic_add.  The descriptor
2211 	 * must be re-verified once we acquire a reference to be certain
2212 	 * that the identity is still correct and we did not lose a race
2213 	 * due to preemption.
2214 	 */
2215 	for (;;) {
2216 		fp = fdp->fd_ofiles[fd];
2217 		if (fp == NULL)
2218 			break;
2219 		count = fp->f_count;
2220 		if (count == 0)
2221 			continue;
2222 		/*
2223 		 * Use an acquire barrier to prevent caching of fd_ofiles
2224 		 * so it is refreshed for verification.
2225 		 */
2226 		if (atomic_cmpset_acq_int(&fp->f_count, count, count + 1) != 1)
2227 			continue;
2228 		if (fp == fdp->fd_ofiles[fd])
2229 			break;
2230 		fdrop(fp, curthread);
2231 	}
2232 
2233 	return (fp);
2234 }
2235 
2236 /*
2237  * Extract the file pointer associated with the specified descriptor for the
2238  * current user process.
2239  *
2240  * If the descriptor doesn't exist or doesn't match 'flags', EBADF is
2241  * returned.
2242  *
2243  * If the FGET_GETCAP flag is set, the capability itself will be returned.
2244  * Calling _fget() with FGET_GETCAP on a non-capability will return EINVAL.
2245  * Otherwise, if the file is a capability, its rights will be checked against
2246  * the capability rights mask, and if successful, the object will be unwrapped.
2247  *
2248  * If an error occured the non-zero error is returned and *fpp is set to
2249  * NULL.  Otherwise *fpp is held and set and zero is returned.  Caller is
2250  * responsible for fdrop().
2251  */
2252 #define	FGET_GETCAP	0x00000001
2253 static __inline int
2254 _fget(struct thread *td, int fd, struct file **fpp, int flags,
2255     cap_rights_t needrights, cap_rights_t *haverightsp, u_char *maxprotp,
2256     int fget_flags)
2257 {
2258 	struct filedesc *fdp;
2259 	struct file *fp;
2260 #ifdef CAPABILITIES
2261 	struct file *fp_fromcap;
2262 #endif
2263 	int error;
2264 
2265 	*fpp = NULL;
2266 	if (td == NULL || (fdp = td->td_proc->p_fd) == NULL)
2267 		return (EBADF);
2268 	if ((fp = fget_unlocked(fdp, fd)) == NULL)
2269 		return (EBADF);
2270 	if (fp->f_ops == &badfileops) {
2271 		fdrop(fp, td);
2272 		return (EBADF);
2273 	}
2274 
2275 #ifdef CAPABILITIES
2276 	/*
2277 	 * If this is a capability, what rights does it have?
2278 	 */
2279 	if (haverightsp != NULL) {
2280 		if (fp->f_type == DTYPE_CAPABILITY)
2281 			*haverightsp = cap_rights(fp);
2282 		else
2283 			*haverightsp = CAP_MASK_VALID;
2284 	}
2285 
2286 	/*
2287 	 * If a capability has been requested, return the capability directly.
2288 	 * Otherwise, check capability rights, extract the underlying object,
2289 	 * and check its access flags.
2290 	 */
2291 	if (fget_flags & FGET_GETCAP) {
2292 		if (fp->f_type != DTYPE_CAPABILITY) {
2293 			fdrop(fp, td);
2294 			return (EINVAL);
2295 		}
2296 	} else {
2297 		if (maxprotp == NULL)
2298 			error = cap_funwrap(fp, needrights, &fp_fromcap);
2299 		else
2300 			error = cap_funwrap_mmap(fp, needrights, maxprotp,
2301 			    &fp_fromcap);
2302 		if (error != 0) {
2303 			fdrop(fp, td);
2304 			return (error);
2305 		}
2306 
2307 		/*
2308 		 * If we've unwrapped a file, drop the original capability
2309 		 * and hold the new descriptor.  fp after this point refers to
2310 		 * the actual (unwrapped) object, not the capability.
2311 		 */
2312 		if (fp != fp_fromcap) {
2313 			fhold(fp_fromcap);
2314 			fdrop(fp, td);
2315 			fp = fp_fromcap;
2316 		}
2317 	}
2318 #else /* !CAPABILITIES */
2319 	KASSERT(fp->f_type != DTYPE_CAPABILITY,
2320 	    ("%s: saw capability", __func__));
2321 	if (maxprotp != NULL)
2322 		*maxprotp = VM_PROT_ALL;
2323 #endif /* CAPABILITIES */
2324 
2325 	/*
2326 	 * FREAD and FWRITE failure return EBADF as per POSIX.
2327 	 */
2328 	error = 0;
2329 	switch (flags) {
2330 	case FREAD:
2331 	case FWRITE:
2332 		if ((fp->f_flag & flags) == 0)
2333 			error = EBADF;
2334 		break;
2335 	case FEXEC:
2336 	    	if ((fp->f_flag & (FREAD | FEXEC)) == 0 ||
2337 		    ((fp->f_flag & FWRITE) != 0))
2338 			error = EBADF;
2339 		break;
2340 	case 0:
2341 		break;
2342 	default:
2343 		KASSERT(0, ("wrong flags"));
2344 	}
2345 
2346 	if (error != 0) {
2347 		fdrop(fp, td);
2348 		return (error);
2349 	}
2350 
2351 	*fpp = fp;
2352 	return (0);
2353 }
2354 
2355 int
2356 fget(struct thread *td, int fd, cap_rights_t rights, struct file **fpp)
2357 {
2358 
2359 	return(_fget(td, fd, fpp, 0, rights, NULL, NULL, 0));
2360 }
2361 
2362 int
2363 fget_mmap(struct thread *td, int fd, cap_rights_t rights, u_char *maxprotp,
2364     struct file **fpp)
2365 {
2366 
2367 	return (_fget(td, fd, fpp, 0, rights, NULL, maxprotp, 0));
2368 }
2369 
2370 int
2371 fget_read(struct thread *td, int fd, cap_rights_t rights, struct file **fpp)
2372 {
2373 
2374 	return(_fget(td, fd, fpp, FREAD, rights, NULL, NULL, 0));
2375 }
2376 
2377 int
2378 fget_write(struct thread *td, int fd, cap_rights_t rights, struct file **fpp)
2379 {
2380 
2381 	return (_fget(td, fd, fpp, FWRITE, rights, NULL, NULL, 0));
2382 }
2383 
2384 /*
2385  * Unlike the other fget() calls, which accept and check capability rights
2386  * but never return capabilities, fgetcap() returns the capability but doesn't
2387  * check capability rights.
2388  */
2389 int
2390 fgetcap(struct thread *td, int fd, struct file **fpp)
2391 {
2392 
2393 	return (_fget(td, fd, fpp, 0, 0, NULL, NULL, FGET_GETCAP));
2394 }
2395 
2396 
2397 /*
2398  * Like fget() but loads the underlying vnode, or returns an error if the
2399  * descriptor does not represent a vnode.  Note that pipes use vnodes but
2400  * never have VM objects.  The returned vnode will be vref()'d.
2401  *
2402  * XXX: what about the unused flags ?
2403  */
2404 static __inline int
2405 _fgetvp(struct thread *td, int fd, int flags, cap_rights_t needrights,
2406     cap_rights_t *haverightsp, struct vnode **vpp)
2407 {
2408 	struct file *fp;
2409 	int error;
2410 
2411 	*vpp = NULL;
2412 	if ((error = _fget(td, fd, &fp, flags, needrights, haverightsp,
2413 	    NULL, 0)) != 0)
2414 		return (error);
2415 	if (fp->f_vnode == NULL) {
2416 		error = EINVAL;
2417 	} else {
2418 		*vpp = fp->f_vnode;
2419 		vref(*vpp);
2420 	}
2421 	fdrop(fp, td);
2422 
2423 	return (error);
2424 }
2425 
2426 int
2427 fgetvp(struct thread *td, int fd, cap_rights_t rights, struct vnode **vpp)
2428 {
2429 
2430 	return (_fgetvp(td, fd, 0, rights, NULL, vpp));
2431 }
2432 
2433 int
2434 fgetvp_rights(struct thread *td, int fd, cap_rights_t need, cap_rights_t *have,
2435     struct vnode **vpp)
2436 {
2437 	return (_fgetvp(td, fd, 0, need, have, vpp));
2438 }
2439 
2440 int
2441 fgetvp_read(struct thread *td, int fd, cap_rights_t rights, struct vnode **vpp)
2442 {
2443 
2444 	return (_fgetvp(td, fd, FREAD, rights, NULL, vpp));
2445 }
2446 
2447 int
2448 fgetvp_exec(struct thread *td, int fd, cap_rights_t rights, struct vnode **vpp)
2449 {
2450 
2451 	return (_fgetvp(td, fd, FEXEC, rights, NULL, vpp));
2452 }
2453 
2454 #ifdef notyet
2455 int
2456 fgetvp_write(struct thread *td, int fd, cap_rights_t rights,
2457     struct vnode **vpp)
2458 {
2459 
2460 	return (_fgetvp(td, fd, FWRITE, rights, NULL, vpp));
2461 }
2462 #endif
2463 
2464 /*
2465  * Like fget() but loads the underlying socket, or returns an error if the
2466  * descriptor does not represent a socket.
2467  *
2468  * We bump the ref count on the returned socket.  XXX Also obtain the SX lock
2469  * in the future.
2470  *
2471  * Note: fgetsock() and fputsock() are deprecated, as consumers should rely
2472  * on their file descriptor reference to prevent the socket from being free'd
2473  * during use.
2474  */
2475 int
2476 fgetsock(struct thread *td, int fd, cap_rights_t rights, struct socket **spp,
2477     u_int *fflagp)
2478 {
2479 	struct file *fp;
2480 	int error;
2481 
2482 	*spp = NULL;
2483 	if (fflagp != NULL)
2484 		*fflagp = 0;
2485 	if ((error = _fget(td, fd, &fp, 0, rights, NULL, NULL, 0)) != 0)
2486 		return (error);
2487 	if (fp->f_type != DTYPE_SOCKET) {
2488 		error = ENOTSOCK;
2489 	} else {
2490 		*spp = fp->f_data;
2491 		if (fflagp)
2492 			*fflagp = fp->f_flag;
2493 		SOCK_LOCK(*spp);
2494 		soref(*spp);
2495 		SOCK_UNLOCK(*spp);
2496 	}
2497 	fdrop(fp, td);
2498 
2499 	return (error);
2500 }
2501 
2502 /*
2503  * Drop the reference count on the socket and XXX release the SX lock in the
2504  * future.  The last reference closes the socket.
2505  *
2506  * Note: fputsock() is deprecated, see comment for fgetsock().
2507  */
2508 void
2509 fputsock(struct socket *so)
2510 {
2511 
2512 	ACCEPT_LOCK();
2513 	SOCK_LOCK(so);
2514 	CURVNET_SET(so->so_vnet);
2515 	sorele(so);
2516 	CURVNET_RESTORE();
2517 }
2518 
2519 /*
2520  * Handle the last reference to a file being closed.
2521  *
2522  * No special capability handling here, as the capability's fo_close will run
2523  * instead of the object here, and perform any necessary drop on the object.
2524  */
2525 int
2526 _fdrop(struct file *fp, struct thread *td)
2527 {
2528 	int error;
2529 
2530 	error = 0;
2531 	if (fp->f_count != 0)
2532 		panic("fdrop: count %d", fp->f_count);
2533 	if (fp->f_ops != &badfileops)
2534 		error = fo_close(fp, td);
2535 	atomic_subtract_int(&openfiles, 1);
2536 	crfree(fp->f_cred);
2537 	free(fp->f_advice, M_FADVISE);
2538 	uma_zfree(file_zone, fp);
2539 
2540 	return (error);
2541 }
2542 
2543 /*
2544  * Apply an advisory lock on a file descriptor.
2545  *
2546  * Just attempt to get a record lock of the requested type on the entire file
2547  * (l_whence = SEEK_SET, l_start = 0, l_len = 0).
2548  */
2549 #ifndef _SYS_SYSPROTO_H_
2550 struct flock_args {
2551 	int	fd;
2552 	int	how;
2553 };
2554 #endif
2555 /* ARGSUSED */
2556 int
2557 sys_flock(struct thread *td, struct flock_args *uap)
2558 {
2559 	struct file *fp;
2560 	struct vnode *vp;
2561 	struct flock lf;
2562 	int error;
2563 
2564 	if ((error = fget(td, uap->fd, CAP_FLOCK, &fp)) != 0)
2565 		return (error);
2566 	if (fp->f_type != DTYPE_VNODE) {
2567 		fdrop(fp, td);
2568 		return (EOPNOTSUPP);
2569 	}
2570 
2571 	vp = fp->f_vnode;
2572 	lf.l_whence = SEEK_SET;
2573 	lf.l_start = 0;
2574 	lf.l_len = 0;
2575 	if (uap->how & LOCK_UN) {
2576 		lf.l_type = F_UNLCK;
2577 		atomic_clear_int(&fp->f_flag, FHASLOCK);
2578 		error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK);
2579 		goto done2;
2580 	}
2581 	if (uap->how & LOCK_EX)
2582 		lf.l_type = F_WRLCK;
2583 	else if (uap->how & LOCK_SH)
2584 		lf.l_type = F_RDLCK;
2585 	else {
2586 		error = EBADF;
2587 		goto done2;
2588 	}
2589 	atomic_set_int(&fp->f_flag, FHASLOCK);
2590 	error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf,
2591 	    (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT);
2592 done2:
2593 	fdrop(fp, td);
2594 	return (error);
2595 }
2596 /*
2597  * Duplicate the specified descriptor to a free descriptor.
2598  */
2599 int
2600 dupfdopen(struct thread *td, struct filedesc *fdp, int dfd, int mode, int openerror, int *indxp)
2601 {
2602 	struct file *fp;
2603 	int error, indx;
2604 
2605 	KASSERT(openerror == ENODEV || openerror == ENXIO,
2606 	    ("unexpected error %d in %s", openerror, __func__));
2607 
2608 	/*
2609 	 * If the to-be-dup'd fd number is greater than the allowed number
2610 	 * of file descriptors, or the fd to be dup'd has already been
2611 	 * closed, then reject.
2612 	 */
2613 	FILEDESC_XLOCK(fdp);
2614 	if ((fp = fget_locked(fdp, dfd)) == NULL) {
2615 		FILEDESC_XUNLOCK(fdp);
2616 		return (EBADF);
2617 	}
2618 
2619 	error = fdalloc(td, 0, &indx);
2620 	if (error != 0) {
2621 		FILEDESC_XUNLOCK(fdp);
2622 		return (error);
2623 	}
2624 
2625 	/*
2626 	 * There are two cases of interest here.
2627 	 *
2628 	 * For ENODEV simply dup (dfd) to file descriptor (indx) and return.
2629 	 *
2630 	 * For ENXIO steal away the file structure from (dfd) and store it in
2631 	 * (indx).  (dfd) is effectively closed by this operation.
2632 	 */
2633 	switch (openerror) {
2634 	case ENODEV:
2635 		/*
2636 		 * Check that the mode the file is being opened for is a
2637 		 * subset of the mode of the existing descriptor.
2638 		 */
2639 		if (((mode & (FREAD|FWRITE)) | fp->f_flag) != fp->f_flag) {
2640 			fdunused(fdp, indx);
2641 			FILEDESC_XUNLOCK(fdp);
2642 			return (EACCES);
2643 		}
2644 		fdp->fd_ofiles[indx] = fp;
2645 		fdp->fd_ofileflags[indx] = fdp->fd_ofileflags[dfd];
2646 		fhold(fp);
2647 		break;
2648 	case ENXIO:
2649 		/*
2650 		 * Steal away the file pointer from dfd and stuff it into indx.
2651 		 */
2652 		fdp->fd_ofiles[indx] = fp;
2653 		fdp->fd_ofiles[dfd] = NULL;
2654 		fdp->fd_ofileflags[indx] = fdp->fd_ofileflags[dfd];
2655 		fdp->fd_ofileflags[dfd] = 0;
2656 		fdunused(fdp, dfd);
2657 		break;
2658 	}
2659 	FILEDESC_XUNLOCK(fdp);
2660 	*indxp = indx;
2661 	return (0);
2662 }
2663 
2664 /*
2665  * Scan all active processes and prisons to see if any of them have a current
2666  * or root directory of `olddp'. If so, replace them with the new mount point.
2667  */
2668 void
2669 mountcheckdirs(struct vnode *olddp, struct vnode *newdp)
2670 {
2671 	struct filedesc *fdp;
2672 	struct prison *pr;
2673 	struct proc *p;
2674 	int nrele;
2675 
2676 	if (vrefcnt(olddp) == 1)
2677 		return;
2678 	nrele = 0;
2679 	sx_slock(&allproc_lock);
2680 	FOREACH_PROC_IN_SYSTEM(p) {
2681 		fdp = fdhold(p);
2682 		if (fdp == NULL)
2683 			continue;
2684 		FILEDESC_XLOCK(fdp);
2685 		if (fdp->fd_cdir == olddp) {
2686 			vref(newdp);
2687 			fdp->fd_cdir = newdp;
2688 			nrele++;
2689 		}
2690 		if (fdp->fd_rdir == olddp) {
2691 			vref(newdp);
2692 			fdp->fd_rdir = newdp;
2693 			nrele++;
2694 		}
2695 		if (fdp->fd_jdir == olddp) {
2696 			vref(newdp);
2697 			fdp->fd_jdir = newdp;
2698 			nrele++;
2699 		}
2700 		FILEDESC_XUNLOCK(fdp);
2701 		fddrop(fdp);
2702 	}
2703 	sx_sunlock(&allproc_lock);
2704 	if (rootvnode == olddp) {
2705 		vref(newdp);
2706 		rootvnode = newdp;
2707 		nrele++;
2708 	}
2709 	mtx_lock(&prison0.pr_mtx);
2710 	if (prison0.pr_root == olddp) {
2711 		vref(newdp);
2712 		prison0.pr_root = newdp;
2713 		nrele++;
2714 	}
2715 	mtx_unlock(&prison0.pr_mtx);
2716 	sx_slock(&allprison_lock);
2717 	TAILQ_FOREACH(pr, &allprison, pr_list) {
2718 		mtx_lock(&pr->pr_mtx);
2719 		if (pr->pr_root == olddp) {
2720 			vref(newdp);
2721 			pr->pr_root = newdp;
2722 			nrele++;
2723 		}
2724 		mtx_unlock(&pr->pr_mtx);
2725 	}
2726 	sx_sunlock(&allprison_lock);
2727 	while (nrele--)
2728 		vrele(olddp);
2729 }
2730 
2731 struct filedesc_to_leader *
2732 filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader)
2733 {
2734 	struct filedesc_to_leader *fdtol;
2735 
2736 	fdtol = malloc(sizeof(struct filedesc_to_leader),
2737 	       M_FILEDESC_TO_LEADER,
2738 	       M_WAITOK);
2739 	fdtol->fdl_refcount = 1;
2740 	fdtol->fdl_holdcount = 0;
2741 	fdtol->fdl_wakeup = 0;
2742 	fdtol->fdl_leader = leader;
2743 	if (old != NULL) {
2744 		FILEDESC_XLOCK(fdp);
2745 		fdtol->fdl_next = old->fdl_next;
2746 		fdtol->fdl_prev = old;
2747 		old->fdl_next = fdtol;
2748 		fdtol->fdl_next->fdl_prev = fdtol;
2749 		FILEDESC_XUNLOCK(fdp);
2750 	} else {
2751 		fdtol->fdl_next = fdtol;
2752 		fdtol->fdl_prev = fdtol;
2753 	}
2754 	return (fdtol);
2755 }
2756 
2757 /*
2758  * Get file structures globally.
2759  */
2760 static int
2761 sysctl_kern_file(SYSCTL_HANDLER_ARGS)
2762 {
2763 	struct xfile xf;
2764 	struct filedesc *fdp;
2765 	struct file *fp;
2766 	struct proc *p;
2767 	int error, n;
2768 
2769 	error = sysctl_wire_old_buffer(req, 0);
2770 	if (error != 0)
2771 		return (error);
2772 	if (req->oldptr == NULL) {
2773 		n = 0;
2774 		sx_slock(&allproc_lock);
2775 		FOREACH_PROC_IN_SYSTEM(p) {
2776 			if (p->p_state == PRS_NEW)
2777 				continue;
2778 			fdp = fdhold(p);
2779 			if (fdp == NULL)
2780 				continue;
2781 			/* overestimates sparse tables. */
2782 			if (fdp->fd_lastfile > 0)
2783 				n += fdp->fd_lastfile;
2784 			fddrop(fdp);
2785 		}
2786 		sx_sunlock(&allproc_lock);
2787 		return (SYSCTL_OUT(req, 0, n * sizeof(xf)));
2788 	}
2789 	error = 0;
2790 	bzero(&xf, sizeof(xf));
2791 	xf.xf_size = sizeof(xf);
2792 	sx_slock(&allproc_lock);
2793 	FOREACH_PROC_IN_SYSTEM(p) {
2794 		PROC_LOCK(p);
2795 		if (p->p_state == PRS_NEW) {
2796 			PROC_UNLOCK(p);
2797 			continue;
2798 		}
2799 		if (p_cansee(req->td, p) != 0) {
2800 			PROC_UNLOCK(p);
2801 			continue;
2802 		}
2803 		xf.xf_pid = p->p_pid;
2804 		xf.xf_uid = p->p_ucred->cr_uid;
2805 		PROC_UNLOCK(p);
2806 		fdp = fdhold(p);
2807 		if (fdp == NULL)
2808 			continue;
2809 		FILEDESC_SLOCK(fdp);
2810 		for (n = 0; fdp->fd_refcnt > 0 && n < fdp->fd_nfiles; ++n) {
2811 			if ((fp = fdp->fd_ofiles[n]) == NULL)
2812 				continue;
2813 			xf.xf_fd = n;
2814 			xf.xf_file = fp;
2815 			xf.xf_data = fp->f_data;
2816 			xf.xf_vnode = fp->f_vnode;
2817 			xf.xf_type = fp->f_type;
2818 			xf.xf_count = fp->f_count;
2819 			xf.xf_msgcount = 0;
2820 			xf.xf_offset = foffset_get(fp);
2821 			xf.xf_flag = fp->f_flag;
2822 			error = SYSCTL_OUT(req, &xf, sizeof(xf));
2823 			if (error)
2824 				break;
2825 		}
2826 		FILEDESC_SUNLOCK(fdp);
2827 		fddrop(fdp);
2828 		if (error)
2829 			break;
2830 	}
2831 	sx_sunlock(&allproc_lock);
2832 	return (error);
2833 }
2834 
2835 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD,
2836     0, 0, sysctl_kern_file, "S,xfile", "Entire file table");
2837 
2838 #ifdef KINFO_OFILE_SIZE
2839 CTASSERT(sizeof(struct kinfo_ofile) == KINFO_OFILE_SIZE);
2840 #endif
2841 
2842 #ifdef COMPAT_FREEBSD7
2843 static int
2844 export_vnode_for_osysctl(struct vnode *vp, int type,
2845     struct kinfo_ofile *kif, struct filedesc *fdp, struct sysctl_req *req)
2846 {
2847 	int error;
2848 	char *fullpath, *freepath;
2849 
2850 	bzero(kif, sizeof(*kif));
2851 	kif->kf_structsize = sizeof(*kif);
2852 
2853 	vref(vp);
2854 	kif->kf_fd = type;
2855 	kif->kf_type = KF_TYPE_VNODE;
2856 	/* This function only handles directories. */
2857 	if (vp->v_type != VDIR) {
2858 		vrele(vp);
2859 		return (ENOTDIR);
2860 	}
2861 	kif->kf_vnode_type = KF_VTYPE_VDIR;
2862 
2863 	/*
2864 	 * This is not a true file descriptor, so we set a bogus refcount
2865 	 * and offset to indicate these fields should be ignored.
2866 	 */
2867 	kif->kf_ref_count = -1;
2868 	kif->kf_offset = -1;
2869 
2870 	freepath = NULL;
2871 	fullpath = "-";
2872 	FILEDESC_SUNLOCK(fdp);
2873 	vn_fullpath(curthread, vp, &fullpath, &freepath);
2874 	vrele(vp);
2875 	strlcpy(kif->kf_path, fullpath, sizeof(kif->kf_path));
2876 	if (freepath != NULL)
2877 		free(freepath, M_TEMP);
2878 	error = SYSCTL_OUT(req, kif, sizeof(*kif));
2879 	FILEDESC_SLOCK(fdp);
2880 	return (error);
2881 }
2882 
2883 /*
2884  * Get per-process file descriptors for use by procstat(1), et al.
2885  */
2886 static int
2887 sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS)
2888 {
2889 	char *fullpath, *freepath;
2890 	struct kinfo_ofile *kif;
2891 	struct filedesc *fdp;
2892 	int error, i, *name;
2893 	struct shmfd *shmfd;
2894 	struct socket *so;
2895 	struct vnode *vp;
2896 	struct file *fp;
2897 	struct proc *p;
2898 	struct tty *tp;
2899 
2900 	name = (int *)arg1;
2901 	if ((p = pfind((pid_t)name[0])) == NULL)
2902 		return (ESRCH);
2903 	if ((error = p_candebug(curthread, p))) {
2904 		PROC_UNLOCK(p);
2905 		return (error);
2906 	}
2907 	fdp = fdhold(p);
2908 	PROC_UNLOCK(p);
2909 	if (fdp == NULL)
2910 		return (ENOENT);
2911 	kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK);
2912 	FILEDESC_SLOCK(fdp);
2913 	if (fdp->fd_cdir != NULL)
2914 		export_vnode_for_osysctl(fdp->fd_cdir, KF_FD_TYPE_CWD, kif,
2915 				fdp, req);
2916 	if (fdp->fd_rdir != NULL)
2917 		export_vnode_for_osysctl(fdp->fd_rdir, KF_FD_TYPE_ROOT, kif,
2918 				fdp, req);
2919 	if (fdp->fd_jdir != NULL)
2920 		export_vnode_for_osysctl(fdp->fd_jdir, KF_FD_TYPE_JAIL, kif,
2921 				fdp, req);
2922 	for (i = 0; i < fdp->fd_nfiles; i++) {
2923 		if ((fp = fdp->fd_ofiles[i]) == NULL)
2924 			continue;
2925 		bzero(kif, sizeof(*kif));
2926 		kif->kf_structsize = sizeof(*kif);
2927 		vp = NULL;
2928 		so = NULL;
2929 		tp = NULL;
2930 		shmfd = NULL;
2931 		kif->kf_fd = i;
2932 
2933 #ifdef CAPABILITIES
2934 		/*
2935 		 * When reporting a capability, most fields will be from the
2936 		 * underlying object, but do mark as a capability. With
2937 		 * ofiledesc, we don't have a field to export the cap_rights_t,
2938 		 * but we do with the new filedesc.
2939 		 */
2940 		if (fp->f_type == DTYPE_CAPABILITY) {
2941 			kif->kf_flags |= KF_FLAG_CAPABILITY;
2942 			(void)cap_funwrap(fp, 0, &fp);
2943 		}
2944 #else
2945 		KASSERT(fp->f_type != DTYPE_CAPABILITY,
2946 		    ("sysctl_kern_proc_ofiledesc: saw capability"));
2947 #endif
2948 		switch (fp->f_type) {
2949 		case DTYPE_VNODE:
2950 			kif->kf_type = KF_TYPE_VNODE;
2951 			vp = fp->f_vnode;
2952 			break;
2953 
2954 		case DTYPE_SOCKET:
2955 			kif->kf_type = KF_TYPE_SOCKET;
2956 			so = fp->f_data;
2957 			break;
2958 
2959 		case DTYPE_PIPE:
2960 			kif->kf_type = KF_TYPE_PIPE;
2961 			break;
2962 
2963 		case DTYPE_FIFO:
2964 			kif->kf_type = KF_TYPE_FIFO;
2965 			vp = fp->f_vnode;
2966 			break;
2967 
2968 		case DTYPE_KQUEUE:
2969 			kif->kf_type = KF_TYPE_KQUEUE;
2970 			break;
2971 
2972 		case DTYPE_CRYPTO:
2973 			kif->kf_type = KF_TYPE_CRYPTO;
2974 			break;
2975 
2976 		case DTYPE_MQUEUE:
2977 			kif->kf_type = KF_TYPE_MQUEUE;
2978 			break;
2979 
2980 		case DTYPE_SHM:
2981 			kif->kf_type = KF_TYPE_SHM;
2982 			shmfd = fp->f_data;
2983 			break;
2984 
2985 		case DTYPE_SEM:
2986 			kif->kf_type = KF_TYPE_SEM;
2987 			break;
2988 
2989 		case DTYPE_PTS:
2990 			kif->kf_type = KF_TYPE_PTS;
2991 			tp = fp->f_data;
2992 			break;
2993 
2994 #ifdef PROCDESC
2995 		case DTYPE_PROCDESC:
2996 			kif->kf_type = KF_TYPE_PROCDESC;
2997 			break;
2998 #endif
2999 
3000 		default:
3001 			kif->kf_type = KF_TYPE_UNKNOWN;
3002 			break;
3003 		}
3004 		kif->kf_ref_count = fp->f_count;
3005 		if (fp->f_flag & FREAD)
3006 			kif->kf_flags |= KF_FLAG_READ;
3007 		if (fp->f_flag & FWRITE)
3008 			kif->kf_flags |= KF_FLAG_WRITE;
3009 		if (fp->f_flag & FAPPEND)
3010 			kif->kf_flags |= KF_FLAG_APPEND;
3011 		if (fp->f_flag & FASYNC)
3012 			kif->kf_flags |= KF_FLAG_ASYNC;
3013 		if (fp->f_flag & FFSYNC)
3014 			kif->kf_flags |= KF_FLAG_FSYNC;
3015 		if (fp->f_flag & FNONBLOCK)
3016 			kif->kf_flags |= KF_FLAG_NONBLOCK;
3017 		if (fp->f_flag & O_DIRECT)
3018 			kif->kf_flags |= KF_FLAG_DIRECT;
3019 		if (fp->f_flag & FHASLOCK)
3020 			kif->kf_flags |= KF_FLAG_HASLOCK;
3021 		kif->kf_offset = foffset_get(fp);
3022 		if (vp != NULL) {
3023 			vref(vp);
3024 			switch (vp->v_type) {
3025 			case VNON:
3026 				kif->kf_vnode_type = KF_VTYPE_VNON;
3027 				break;
3028 			case VREG:
3029 				kif->kf_vnode_type = KF_VTYPE_VREG;
3030 				break;
3031 			case VDIR:
3032 				kif->kf_vnode_type = KF_VTYPE_VDIR;
3033 				break;
3034 			case VBLK:
3035 				kif->kf_vnode_type = KF_VTYPE_VBLK;
3036 				break;
3037 			case VCHR:
3038 				kif->kf_vnode_type = KF_VTYPE_VCHR;
3039 				break;
3040 			case VLNK:
3041 				kif->kf_vnode_type = KF_VTYPE_VLNK;
3042 				break;
3043 			case VSOCK:
3044 				kif->kf_vnode_type = KF_VTYPE_VSOCK;
3045 				break;
3046 			case VFIFO:
3047 				kif->kf_vnode_type = KF_VTYPE_VFIFO;
3048 				break;
3049 			case VBAD:
3050 				kif->kf_vnode_type = KF_VTYPE_VBAD;
3051 				break;
3052 			default:
3053 				kif->kf_vnode_type = KF_VTYPE_UNKNOWN;
3054 				break;
3055 			}
3056 			/*
3057 			 * It is OK to drop the filedesc lock here as we will
3058 			 * re-validate and re-evaluate its properties when
3059 			 * the loop continues.
3060 			 */
3061 			freepath = NULL;
3062 			fullpath = "-";
3063 			FILEDESC_SUNLOCK(fdp);
3064 			vn_fullpath(curthread, vp, &fullpath, &freepath);
3065 			vrele(vp);
3066 			strlcpy(kif->kf_path, fullpath,
3067 			    sizeof(kif->kf_path));
3068 			if (freepath != NULL)
3069 				free(freepath, M_TEMP);
3070 			FILEDESC_SLOCK(fdp);
3071 		}
3072 		if (so != NULL) {
3073 			struct sockaddr *sa;
3074 
3075 			if (so->so_proto->pr_usrreqs->pru_sockaddr(so, &sa)
3076 			    == 0 && sa->sa_len <= sizeof(kif->kf_sa_local)) {
3077 				bcopy(sa, &kif->kf_sa_local, sa->sa_len);
3078 				free(sa, M_SONAME);
3079 			}
3080 			if (so->so_proto->pr_usrreqs->pru_peeraddr(so, &sa)
3081 			    == 0 && sa->sa_len <= sizeof(kif->kf_sa_peer)) {
3082 				bcopy(sa, &kif->kf_sa_peer, sa->sa_len);
3083 				free(sa, M_SONAME);
3084 			}
3085 			kif->kf_sock_domain =
3086 			    so->so_proto->pr_domain->dom_family;
3087 			kif->kf_sock_type = so->so_type;
3088 			kif->kf_sock_protocol = so->so_proto->pr_protocol;
3089 		}
3090 		if (tp != NULL) {
3091 			strlcpy(kif->kf_path, tty_devname(tp),
3092 			    sizeof(kif->kf_path));
3093 		}
3094 		if (shmfd != NULL)
3095 			shm_path(shmfd, kif->kf_path, sizeof(kif->kf_path));
3096 		error = SYSCTL_OUT(req, kif, sizeof(*kif));
3097 		if (error)
3098 			break;
3099 	}
3100 	FILEDESC_SUNLOCK(fdp);
3101 	fddrop(fdp);
3102 	free(kif, M_TEMP);
3103 	return (0);
3104 }
3105 
3106 static SYSCTL_NODE(_kern_proc, KERN_PROC_OFILEDESC, ofiledesc, CTLFLAG_RD,
3107     sysctl_kern_proc_ofiledesc, "Process ofiledesc entries");
3108 #endif	/* COMPAT_FREEBSD7 */
3109 
3110 #ifdef KINFO_FILE_SIZE
3111 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
3112 #endif
3113 
3114 static int
3115 export_fd_for_sysctl(void *data, int type, int fd, int fflags, int refcnt,
3116     int64_t offset, int fd_is_cap, cap_rights_t fd_cap_rights,
3117     struct kinfo_file *kif, struct sysctl_req *req)
3118 {
3119 	struct {
3120 		int	fflag;
3121 		int	kf_fflag;
3122 	} fflags_table[] = {
3123 		{ FAPPEND, KF_FLAG_APPEND },
3124 		{ FASYNC, KF_FLAG_ASYNC },
3125 		{ FFSYNC, KF_FLAG_FSYNC },
3126 		{ FHASLOCK, KF_FLAG_HASLOCK },
3127 		{ FNONBLOCK, KF_FLAG_NONBLOCK },
3128 		{ FREAD, KF_FLAG_READ },
3129 		{ FWRITE, KF_FLAG_WRITE },
3130 		{ O_CREAT, KF_FLAG_CREAT },
3131 		{ O_DIRECT, KF_FLAG_DIRECT },
3132 		{ O_EXCL, KF_FLAG_EXCL },
3133 		{ O_EXEC, KF_FLAG_EXEC },
3134 		{ O_EXLOCK, KF_FLAG_EXLOCK },
3135 		{ O_NOFOLLOW, KF_FLAG_NOFOLLOW },
3136 		{ O_SHLOCK, KF_FLAG_SHLOCK },
3137 		{ O_TRUNC, KF_FLAG_TRUNC }
3138 	};
3139 #define	NFFLAGS	(sizeof(fflags_table) / sizeof(*fflags_table))
3140 	struct vnode *vp;
3141 	int error;
3142 	unsigned int i;
3143 
3144 	bzero(kif, sizeof(*kif));
3145 	switch (type) {
3146 	case KF_TYPE_FIFO:
3147 	case KF_TYPE_VNODE:
3148 		vp = (struct vnode *)data;
3149 		error = fill_vnode_info(vp, kif);
3150 		vrele(vp);
3151 		break;
3152 	case KF_TYPE_SOCKET:
3153 		error = fill_socket_info((struct socket *)data, kif);
3154 		break;
3155 	case KF_TYPE_PIPE:
3156 		error = fill_pipe_info((struct pipe *)data, kif);
3157 		break;
3158 	case KF_TYPE_PTS:
3159 		error = fill_pts_info((struct tty *)data, kif);
3160 		break;
3161 	case KF_TYPE_PROCDESC:
3162 		error = fill_procdesc_info((struct procdesc *)data, kif);
3163 		break;
3164 	case KF_TYPE_SHM:
3165 		error = fill_shm_info((struct file *)data, kif);
3166 		break;
3167 	default:
3168 		error = 0;
3169 	}
3170 	if (error == 0)
3171 		kif->kf_status |= KF_ATTR_VALID;
3172 
3173 	/*
3174 	 * Translate file access flags.
3175 	 */
3176 	for (i = 0; i < NFFLAGS; i++)
3177 		if (fflags & fflags_table[i].fflag)
3178 			kif->kf_flags |=  fflags_table[i].kf_fflag;
3179 	if (fd_is_cap)
3180 		kif->kf_flags |= KF_FLAG_CAPABILITY;
3181 	if (fd_is_cap)
3182 		kif->kf_cap_rights = fd_cap_rights;
3183 	kif->kf_fd = fd;
3184 	kif->kf_type = type;
3185 	kif->kf_ref_count = refcnt;
3186 	kif->kf_offset = offset;
3187 	/* Pack record size down */
3188 	kif->kf_structsize = offsetof(struct kinfo_file, kf_path) +
3189 	    strlen(kif->kf_path) + 1;
3190 	kif->kf_structsize = roundup(kif->kf_structsize, sizeof(uint64_t));
3191 	error = SYSCTL_OUT(req, kif, kif->kf_structsize);
3192 	return (error);
3193 }
3194 
3195 /*
3196  * Get per-process file descriptors for use by procstat(1), et al.
3197  */
3198 static int
3199 sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS)
3200 {
3201 	struct file *fp;
3202 	struct filedesc *fdp;
3203 	struct kinfo_file *kif;
3204 	struct proc *p;
3205 	struct vnode *cttyvp, *textvp, *tracevp;
3206 	size_t oldidx;
3207 	int64_t offset;
3208 	void *data;
3209 	int error, i, *name;
3210 	int fd_is_cap, type, refcnt, fflags;
3211 	cap_rights_t fd_cap_rights;
3212 
3213 	name = (int *)arg1;
3214 	if ((p = pfind((pid_t)name[0])) == NULL)
3215 		return (ESRCH);
3216 	if ((error = p_candebug(curthread, p))) {
3217 		PROC_UNLOCK(p);
3218 		return (error);
3219 	}
3220 	/* ktrace vnode */
3221 	tracevp = p->p_tracevp;
3222 	if (tracevp != NULL)
3223 		vref(tracevp);
3224 	/* text vnode */
3225 	textvp = p->p_textvp;
3226 	if (textvp != NULL)
3227 		vref(textvp);
3228 	/* Controlling tty. */
3229 	cttyvp = NULL;
3230 	if (p->p_pgrp != NULL && p->p_pgrp->pg_session != NULL) {
3231 		cttyvp = p->p_pgrp->pg_session->s_ttyvp;
3232 		if (cttyvp != NULL)
3233 			vref(cttyvp);
3234 	}
3235 	fdp = fdhold(p);
3236 	PROC_UNLOCK(p);
3237 	kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK);
3238 	if (tracevp != NULL)
3239 		export_fd_for_sysctl(tracevp, KF_TYPE_VNODE, KF_FD_TYPE_TRACE,
3240 		    FREAD | FWRITE, -1, -1, 0, 0, kif, req);
3241 	if (textvp != NULL)
3242 		export_fd_for_sysctl(textvp, KF_TYPE_VNODE, KF_FD_TYPE_TEXT,
3243 		    FREAD, -1, -1, 0, 0, kif, req);
3244 	if (cttyvp != NULL)
3245 		export_fd_for_sysctl(cttyvp, KF_TYPE_VNODE, KF_FD_TYPE_CTTY,
3246 		    FREAD | FWRITE, -1, -1, 0, 0, kif, req);
3247 	if (fdp == NULL)
3248 		goto fail;
3249 	FILEDESC_SLOCK(fdp);
3250 	/* working directory */
3251 	if (fdp->fd_cdir != NULL) {
3252 		vref(fdp->fd_cdir);
3253 		data = fdp->fd_cdir;
3254 		FILEDESC_SUNLOCK(fdp);
3255 		export_fd_for_sysctl(data, KF_TYPE_VNODE, KF_FD_TYPE_CWD,
3256 		    FREAD, -1, -1, 0, 0, kif, req);
3257 		FILEDESC_SLOCK(fdp);
3258 	}
3259 	/* root directory */
3260 	if (fdp->fd_rdir != NULL) {
3261 		vref(fdp->fd_rdir);
3262 		data = fdp->fd_rdir;
3263 		FILEDESC_SUNLOCK(fdp);
3264 		export_fd_for_sysctl(data, KF_TYPE_VNODE, KF_FD_TYPE_ROOT,
3265 		    FREAD, -1, -1, 0, 0, kif, req);
3266 		FILEDESC_SLOCK(fdp);
3267 	}
3268 	/* jail directory */
3269 	if (fdp->fd_jdir != NULL) {
3270 		vref(fdp->fd_jdir);
3271 		data = fdp->fd_jdir;
3272 		FILEDESC_SUNLOCK(fdp);
3273 		export_fd_for_sysctl(data, KF_TYPE_VNODE, KF_FD_TYPE_JAIL,
3274 		    FREAD, -1, -1, 0, 0, kif, req);
3275 		FILEDESC_SLOCK(fdp);
3276 	}
3277 	for (i = 0; i < fdp->fd_nfiles; i++) {
3278 		if ((fp = fdp->fd_ofiles[i]) == NULL)
3279 			continue;
3280 		data = NULL;
3281 		fd_is_cap = 0;
3282 		fd_cap_rights = 0;
3283 
3284 #ifdef CAPABILITIES
3285 		/*
3286 		 * When reporting a capability, most fields will be from the
3287 		 * underlying object, but do mark as a capability and export
3288 		 * the capability rights mask.
3289 		 */
3290 		if (fp->f_type == DTYPE_CAPABILITY) {
3291 			fd_is_cap = 1;
3292 			fd_cap_rights = cap_rights(fp);
3293 			(void)cap_funwrap(fp, 0, &fp);
3294 		}
3295 #else /* !CAPABILITIES */
3296 		KASSERT(fp->f_type != DTYPE_CAPABILITY,
3297 		    ("sysctl_kern_proc_filedesc: saw capability"));
3298 #endif
3299 		switch (fp->f_type) {
3300 		case DTYPE_VNODE:
3301 			type = KF_TYPE_VNODE;
3302 			vref(fp->f_vnode);
3303 			data = fp->f_vnode;
3304 			break;
3305 
3306 		case DTYPE_SOCKET:
3307 			type = KF_TYPE_SOCKET;
3308 			data = fp->f_data;
3309 			break;
3310 
3311 		case DTYPE_PIPE:
3312 			type = KF_TYPE_PIPE;
3313 			data = fp->f_data;
3314 			break;
3315 
3316 		case DTYPE_FIFO:
3317 			type = KF_TYPE_FIFO;
3318 			vref(fp->f_vnode);
3319 			data = fp->f_vnode;
3320 			break;
3321 
3322 		case DTYPE_KQUEUE:
3323 			type = KF_TYPE_KQUEUE;
3324 			break;
3325 
3326 		case DTYPE_CRYPTO:
3327 			type = KF_TYPE_CRYPTO;
3328 			break;
3329 
3330 		case DTYPE_MQUEUE:
3331 			type = KF_TYPE_MQUEUE;
3332 			break;
3333 
3334 		case DTYPE_SHM:
3335 			type = KF_TYPE_SHM;
3336 			data = fp;
3337 			break;
3338 
3339 		case DTYPE_SEM:
3340 			type = KF_TYPE_SEM;
3341 			break;
3342 
3343 		case DTYPE_PTS:
3344 			type = KF_TYPE_PTS;
3345 			data = fp->f_data;
3346 			break;
3347 
3348 #ifdef PROCDESC
3349 		case DTYPE_PROCDESC:
3350 			type = KF_TYPE_PROCDESC;
3351 			data = fp->f_data;
3352 			break;
3353 #endif
3354 
3355 		default:
3356 			type = KF_TYPE_UNKNOWN;
3357 			break;
3358 		}
3359 		refcnt = fp->f_count;
3360 		fflags = fp->f_flag;
3361 		offset = foffset_get(fp);
3362 
3363 		/*
3364 		 * Create sysctl entry.
3365 		 * It is OK to drop the filedesc lock here as we will
3366 		 * re-validate and re-evaluate its properties when
3367 		 * the loop continues.
3368 		 */
3369 		oldidx = req->oldidx;
3370 		if (type == KF_TYPE_VNODE || type == KF_TYPE_FIFO)
3371 			FILEDESC_SUNLOCK(fdp);
3372 		error = export_fd_for_sysctl(data, type, i, fflags, refcnt,
3373 		    offset, fd_is_cap, fd_cap_rights, kif, req);
3374 		if (type == KF_TYPE_VNODE || type == KF_TYPE_FIFO)
3375 			FILEDESC_SLOCK(fdp);
3376 		if (error) {
3377 			if (error == ENOMEM) {
3378 				/*
3379 				 * The hack to keep the ABI of sysctl
3380 				 * kern.proc.filedesc intact, but not
3381 				 * to account a partially copied
3382 				 * kinfo_file into the oldidx.
3383 				 */
3384 				req->oldidx = oldidx;
3385 				error = 0;
3386 			}
3387 			break;
3388 		}
3389 	}
3390 	FILEDESC_SUNLOCK(fdp);
3391 fail:
3392 	if (fdp != NULL)
3393 		fddrop(fdp);
3394 	free(kif, M_TEMP);
3395 	return (error);
3396 }
3397 
3398 int
3399 vntype_to_kinfo(int vtype)
3400 {
3401 	struct {
3402 		int	vtype;
3403 		int	kf_vtype;
3404 	} vtypes_table[] = {
3405 		{ VBAD, KF_VTYPE_VBAD },
3406 		{ VBLK, KF_VTYPE_VBLK },
3407 		{ VCHR, KF_VTYPE_VCHR },
3408 		{ VDIR, KF_VTYPE_VDIR },
3409 		{ VFIFO, KF_VTYPE_VFIFO },
3410 		{ VLNK, KF_VTYPE_VLNK },
3411 		{ VNON, KF_VTYPE_VNON },
3412 		{ VREG, KF_VTYPE_VREG },
3413 		{ VSOCK, KF_VTYPE_VSOCK }
3414 	};
3415 #define	NVTYPES	(sizeof(vtypes_table) / sizeof(*vtypes_table))
3416 	unsigned int i;
3417 
3418 	/*
3419 	 * Perform vtype translation.
3420 	 */
3421 	for (i = 0; i < NVTYPES; i++)
3422 		if (vtypes_table[i].vtype == vtype)
3423 			break;
3424 	if (i < NVTYPES)
3425 		return (vtypes_table[i].kf_vtype);
3426 
3427 	return (KF_VTYPE_UNKNOWN);
3428 }
3429 
3430 static int
3431 fill_vnode_info(struct vnode *vp, struct kinfo_file *kif)
3432 {
3433 	struct vattr va;
3434 	char *fullpath, *freepath;
3435 	int error;
3436 
3437 	if (vp == NULL)
3438 		return (1);
3439 	kif->kf_vnode_type = vntype_to_kinfo(vp->v_type);
3440 	freepath = NULL;
3441 	fullpath = "-";
3442 	error = vn_fullpath(curthread, vp, &fullpath, &freepath);
3443 	if (error == 0) {
3444 		strlcpy(kif->kf_path, fullpath, sizeof(kif->kf_path));
3445 	}
3446 	if (freepath != NULL)
3447 		free(freepath, M_TEMP);
3448 
3449 	/*
3450 	 * Retrieve vnode attributes.
3451 	 */
3452 	va.va_fsid = VNOVAL;
3453 	va.va_rdev = NODEV;
3454 	vn_lock(vp, LK_SHARED | LK_RETRY);
3455 	error = VOP_GETATTR(vp, &va, curthread->td_ucred);
3456 	VOP_UNLOCK(vp, 0);
3457 	if (error != 0)
3458 		return (error);
3459 	if (va.va_fsid != VNOVAL)
3460 		kif->kf_un.kf_file.kf_file_fsid = va.va_fsid;
3461 	else
3462 		kif->kf_un.kf_file.kf_file_fsid =
3463 		    vp->v_mount->mnt_stat.f_fsid.val[0];
3464 	kif->kf_un.kf_file.kf_file_fileid = va.va_fileid;
3465 	kif->kf_un.kf_file.kf_file_mode = MAKEIMODE(va.va_type, va.va_mode);
3466 	kif->kf_un.kf_file.kf_file_size = va.va_size;
3467 	kif->kf_un.kf_file.kf_file_rdev = va.va_rdev;
3468 	return (0);
3469 }
3470 
3471 static int
3472 fill_socket_info(struct socket *so, struct kinfo_file *kif)
3473 {
3474 	struct sockaddr *sa;
3475 	struct inpcb *inpcb;
3476 	struct unpcb *unpcb;
3477 	int error;
3478 
3479 	if (so == NULL)
3480 		return (1);
3481 	kif->kf_sock_domain = so->so_proto->pr_domain->dom_family;
3482 	kif->kf_sock_type = so->so_type;
3483 	kif->kf_sock_protocol = so->so_proto->pr_protocol;
3484 	kif->kf_un.kf_sock.kf_sock_pcb = (uintptr_t)so->so_pcb;
3485 	switch(kif->kf_sock_domain) {
3486 	case AF_INET:
3487 	case AF_INET6:
3488 		if (kif->kf_sock_protocol == IPPROTO_TCP) {
3489 			if (so->so_pcb != NULL) {
3490 				inpcb = (struct inpcb *)(so->so_pcb);
3491 				kif->kf_un.kf_sock.kf_sock_inpcb =
3492 				    (uintptr_t)inpcb->inp_ppcb;
3493 			}
3494 		}
3495 		break;
3496 	case AF_UNIX:
3497 		if (so->so_pcb != NULL) {
3498 			unpcb = (struct unpcb *)(so->so_pcb);
3499 			if (unpcb->unp_conn) {
3500 				kif->kf_un.kf_sock.kf_sock_unpconn =
3501 				    (uintptr_t)unpcb->unp_conn;
3502 				kif->kf_un.kf_sock.kf_sock_rcv_sb_state =
3503 				    so->so_rcv.sb_state;
3504 				kif->kf_un.kf_sock.kf_sock_snd_sb_state =
3505 				    so->so_snd.sb_state;
3506 			}
3507 		}
3508 		break;
3509 	}
3510 	error = so->so_proto->pr_usrreqs->pru_sockaddr(so, &sa);
3511 	if (error == 0 && sa->sa_len <= sizeof(kif->kf_sa_local)) {
3512 		bcopy(sa, &kif->kf_sa_local, sa->sa_len);
3513 		free(sa, M_SONAME);
3514 	}
3515 	error = so->so_proto->pr_usrreqs->pru_peeraddr(so, &sa);
3516 	if (error == 0 && sa->sa_len <= sizeof(kif->kf_sa_peer)) {
3517 		bcopy(sa, &kif->kf_sa_peer, sa->sa_len);
3518 		free(sa, M_SONAME);
3519 	}
3520 	strncpy(kif->kf_path, so->so_proto->pr_domain->dom_name,
3521 	    sizeof(kif->kf_path));
3522 	return (0);
3523 }
3524 
3525 static int
3526 fill_pts_info(struct tty *tp, struct kinfo_file *kif)
3527 {
3528 
3529 	if (tp == NULL)
3530 		return (1);
3531 	kif->kf_un.kf_pts.kf_pts_dev = tty_udev(tp);
3532 	strlcpy(kif->kf_path, tty_devname(tp), sizeof(kif->kf_path));
3533 	return (0);
3534 }
3535 
3536 static int
3537 fill_pipe_info(struct pipe *pi, struct kinfo_file *kif)
3538 {
3539 
3540 	if (pi == NULL)
3541 		return (1);
3542 	kif->kf_un.kf_pipe.kf_pipe_addr = (uintptr_t)pi;
3543 	kif->kf_un.kf_pipe.kf_pipe_peer = (uintptr_t)pi->pipe_peer;
3544 	kif->kf_un.kf_pipe.kf_pipe_buffer_cnt = pi->pipe_buffer.cnt;
3545 	return (0);
3546 }
3547 
3548 static int
3549 fill_procdesc_info(struct procdesc *pdp, struct kinfo_file *kif)
3550 {
3551 
3552 	if (pdp == NULL)
3553 		return (1);
3554 	kif->kf_un.kf_proc.kf_pid = pdp->pd_pid;
3555 	return (0);
3556 }
3557 
3558 static int
3559 fill_shm_info(struct file *fp, struct kinfo_file *kif)
3560 {
3561 	struct thread *td;
3562 	struct stat sb;
3563 
3564 	td = curthread;
3565 	if (fp->f_data == NULL)
3566 		return (1);
3567 	if (fo_stat(fp, &sb, td->td_ucred, td) != 0)
3568 		return (1);
3569 	shm_path(fp->f_data, kif->kf_path, sizeof(kif->kf_path));
3570 	kif->kf_un.kf_file.kf_file_mode = sb.st_mode;
3571 	kif->kf_un.kf_file.kf_file_size = sb.st_size;
3572 	return (0);
3573 }
3574 
3575 static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc, CTLFLAG_RD,
3576     sysctl_kern_proc_filedesc, "Process filedesc entries");
3577 
3578 #ifdef DDB
3579 /*
3580  * For the purposes of debugging, generate a human-readable string for the
3581  * file type.
3582  */
3583 static const char *
3584 file_type_to_name(short type)
3585 {
3586 
3587 	switch (type) {
3588 	case 0:
3589 		return ("zero");
3590 	case DTYPE_VNODE:
3591 		return ("vnod");
3592 	case DTYPE_SOCKET:
3593 		return ("sock");
3594 	case DTYPE_PIPE:
3595 		return ("pipe");
3596 	case DTYPE_FIFO:
3597 		return ("fifo");
3598 	case DTYPE_KQUEUE:
3599 		return ("kque");
3600 	case DTYPE_CRYPTO:
3601 		return ("crpt");
3602 	case DTYPE_MQUEUE:
3603 		return ("mque");
3604 	case DTYPE_SHM:
3605 		return ("shm");
3606 	case DTYPE_SEM:
3607 		return ("ksem");
3608 	default:
3609 		return ("unkn");
3610 	}
3611 }
3612 
3613 /*
3614  * For the purposes of debugging, identify a process (if any, perhaps one of
3615  * many) that references the passed file in its file descriptor array. Return
3616  * NULL if none.
3617  */
3618 static struct proc *
3619 file_to_first_proc(struct file *fp)
3620 {
3621 	struct filedesc *fdp;
3622 	struct proc *p;
3623 	int n;
3624 
3625 	FOREACH_PROC_IN_SYSTEM(p) {
3626 		if (p->p_state == PRS_NEW)
3627 			continue;
3628 		fdp = p->p_fd;
3629 		if (fdp == NULL)
3630 			continue;
3631 		for (n = 0; n < fdp->fd_nfiles; n++) {
3632 			if (fp == fdp->fd_ofiles[n])
3633 				return (p);
3634 		}
3635 	}
3636 	return (NULL);
3637 }
3638 
3639 static void
3640 db_print_file(struct file *fp, int header)
3641 {
3642 	struct proc *p;
3643 
3644 	if (header)
3645 		db_printf("%8s %4s %8s %8s %4s %5s %6s %8s %5s %12s\n",
3646 		    "File", "Type", "Data", "Flag", "GCFl", "Count",
3647 		    "MCount", "Vnode", "FPID", "FCmd");
3648 	p = file_to_first_proc(fp);
3649 	db_printf("%8p %4s %8p %08x %04x %5d %6d %8p %5d %12s\n", fp,
3650 	    file_type_to_name(fp->f_type), fp->f_data, fp->f_flag,
3651 	    0, fp->f_count, 0, fp->f_vnode,
3652 	    p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-");
3653 }
3654 
3655 DB_SHOW_COMMAND(file, db_show_file)
3656 {
3657 	struct file *fp;
3658 
3659 	if (!have_addr) {
3660 		db_printf("usage: show file <addr>\n");
3661 		return;
3662 	}
3663 	fp = (struct file *)addr;
3664 	db_print_file(fp, 1);
3665 }
3666 
3667 DB_SHOW_COMMAND(files, db_show_files)
3668 {
3669 	struct filedesc *fdp;
3670 	struct file *fp;
3671 	struct proc *p;
3672 	int header;
3673 	int n;
3674 
3675 	header = 1;
3676 	FOREACH_PROC_IN_SYSTEM(p) {
3677 		if (p->p_state == PRS_NEW)
3678 			continue;
3679 		if ((fdp = p->p_fd) == NULL)
3680 			continue;
3681 		for (n = 0; n < fdp->fd_nfiles; ++n) {
3682 			if ((fp = fdp->fd_ofiles[n]) == NULL)
3683 				continue;
3684 			db_print_file(fp, header);
3685 			header = 0;
3686 		}
3687 	}
3688 }
3689 #endif
3690 
3691 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
3692     &maxfilesperproc, 0, "Maximum files allowed open per process");
3693 
3694 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
3695     &maxfiles, 0, "Maximum number of files");
3696 
3697 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
3698     __DEVOLATILE(int *, &openfiles), 0, "System-wide number of open files");
3699 
3700 /* ARGSUSED*/
3701 static void
3702 filelistinit(void *dummy)
3703 {
3704 
3705 	file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL,
3706 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
3707 	mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF);
3708 	mtx_init(&fdesc_mtx, "fdesc", NULL, MTX_DEF);
3709 }
3710 SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL);
3711 
3712 /*-------------------------------------------------------------------*/
3713 
3714 static int
3715 badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred,
3716     int flags, struct thread *td)
3717 {
3718 
3719 	return (EBADF);
3720 }
3721 
3722 static int
3723 badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
3724     struct thread *td)
3725 {
3726 
3727 	return (EINVAL);
3728 }
3729 
3730 static int
3731 badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred,
3732     struct thread *td)
3733 {
3734 
3735 	return (EBADF);
3736 }
3737 
3738 static int
3739 badfo_poll(struct file *fp, int events, struct ucred *active_cred,
3740     struct thread *td)
3741 {
3742 
3743 	return (0);
3744 }
3745 
3746 static int
3747 badfo_kqfilter(struct file *fp, struct knote *kn)
3748 {
3749 
3750 	return (EBADF);
3751 }
3752 
3753 static int
3754 badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred,
3755     struct thread *td)
3756 {
3757 
3758 	return (EBADF);
3759 }
3760 
3761 static int
3762 badfo_close(struct file *fp, struct thread *td)
3763 {
3764 
3765 	return (EBADF);
3766 }
3767 
3768 static int
3769 badfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
3770     struct thread *td)
3771 {
3772 
3773 	return (EBADF);
3774 }
3775 
3776 static int
3777 badfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
3778     struct thread *td)
3779 {
3780 
3781 	return (EBADF);
3782 }
3783 
3784 struct fileops badfileops = {
3785 	.fo_read = badfo_readwrite,
3786 	.fo_write = badfo_readwrite,
3787 	.fo_truncate = badfo_truncate,
3788 	.fo_ioctl = badfo_ioctl,
3789 	.fo_poll = badfo_poll,
3790 	.fo_kqfilter = badfo_kqfilter,
3791 	.fo_stat = badfo_stat,
3792 	.fo_close = badfo_close,
3793 	.fo_chmod = badfo_chmod,
3794 	.fo_chown = badfo_chown,
3795 };
3796 
3797 int
3798 invfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
3799     struct thread *td)
3800 {
3801 
3802 	return (EINVAL);
3803 }
3804 
3805 int
3806 invfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
3807     struct thread *td)
3808 {
3809 
3810 	return (EINVAL);
3811 }
3812 
3813 /*-------------------------------------------------------------------*/
3814 
3815 /*
3816  * File Descriptor pseudo-device driver (/dev/fd/).
3817  *
3818  * Opening minor device N dup()s the file (if any) connected to file
3819  * descriptor N belonging to the calling process.  Note that this driver
3820  * consists of only the ``open()'' routine, because all subsequent
3821  * references to this file will be direct to the other driver.
3822  *
3823  * XXX: we could give this one a cloning event handler if necessary.
3824  */
3825 
3826 /* ARGSUSED */
3827 static int
3828 fdopen(struct cdev *dev, int mode, int type, struct thread *td)
3829 {
3830 
3831 	/*
3832 	 * XXX Kludge: set curthread->td_dupfd to contain the value of the
3833 	 * the file descriptor being sought for duplication. The error
3834 	 * return ensures that the vnode for this device will be released
3835 	 * by vn_open. Open will detect this special error and take the
3836 	 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
3837 	 * will simply report the error.
3838 	 */
3839 	td->td_dupfd = dev2unit(dev);
3840 	return (ENODEV);
3841 }
3842 
3843 static struct cdevsw fildesc_cdevsw = {
3844 	.d_version =	D_VERSION,
3845 	.d_open =	fdopen,
3846 	.d_name =	"FD",
3847 };
3848 
3849 static void
3850 fildesc_drvinit(void *unused)
3851 {
3852 	struct cdev *dev;
3853 
3854 	dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 0, NULL,
3855 	    UID_ROOT, GID_WHEEL, 0666, "fd/0");
3856 	make_dev_alias(dev, "stdin");
3857 	dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 1, NULL,
3858 	    UID_ROOT, GID_WHEEL, 0666, "fd/1");
3859 	make_dev_alias(dev, "stdout");
3860 	dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 2, NULL,
3861 	    UID_ROOT, GID_WHEEL, 0666, "fd/2");
3862 	make_dev_alias(dev, "stderr");
3863 }
3864 
3865 SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL);
3866