xref: /freebsd/sys/kern/kern_descrip.c (revision 3f0164abf32b9b761e0a2cb4bdca3a8b84f156d4)
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/ksem.h>
59 #include <sys/limits.h>
60 #include <sys/lock.h>
61 #include <sys/malloc.h>
62 #include <sys/mman.h>
63 #include <sys/mount.h>
64 #include <sys/mqueue.h>
65 #include <sys/mutex.h>
66 #include <sys/namei.h>
67 #include <sys/selinfo.h>
68 #include <sys/pipe.h>
69 #include <sys/priv.h>
70 #include <sys/proc.h>
71 #include <sys/procdesc.h>
72 #include <sys/protosw.h>
73 #include <sys/racct.h>
74 #include <sys/resourcevar.h>
75 #include <sys/sbuf.h>
76 #include <sys/signalvar.h>
77 #include <sys/socketvar.h>
78 #include <sys/stat.h>
79 #include <sys/sx.h>
80 #include <sys/syscallsubr.h>
81 #include <sys/sysctl.h>
82 #include <sys/sysproto.h>
83 #include <sys/tty.h>
84 #include <sys/unistd.h>
85 #include <sys/un.h>
86 #include <sys/unpcb.h>
87 #include <sys/user.h>
88 #include <sys/vnode.h>
89 #ifdef KTRACE
90 #include <sys/ktrace.h>
91 #endif
92 
93 #include <net/vnet.h>
94 
95 #include <netinet/in.h>
96 #include <netinet/in_pcb.h>
97 
98 #include <security/audit/audit.h>
99 
100 #include <vm/uma.h>
101 #include <vm/vm.h>
102 
103 #include <ddb/ddb.h>
104 
105 static MALLOC_DEFINE(M_FILEDESC, "filedesc", "Open file descriptor table");
106 static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "filedesc_to_leader",
107     "file desc to leader structures");
108 static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures");
109 MALLOC_DEFINE(M_FILECAPS, "filecaps", "descriptor capabilities");
110 
111 MALLOC_DECLARE(M_FADVISE);
112 
113 static uma_zone_t file_zone;
114 
115 void	(*ksem_info)(struct ksem *ks, char *path, size_t size, uint32_t *value);
116 
117 static int	closefp(struct filedesc *fdp, int fd, struct file *fp,
118 		    struct thread *td, int holdleaders);
119 static int	fd_first_free(struct filedesc *fdp, int low, int size);
120 static int	fd_last_used(struct filedesc *fdp, int size);
121 static void	fdgrowtable(struct filedesc *fdp, int nfd);
122 static void	fdunused(struct filedesc *fdp, int fd);
123 static void	fdused(struct filedesc *fdp, int fd);
124 static int	fill_pipe_info(struct pipe *pi, struct kinfo_file *kif);
125 static int	fill_procdesc_info(struct procdesc *pdp,
126 		    struct kinfo_file *kif);
127 static int	fill_pts_info(struct tty *tp, struct kinfo_file *kif);
128 static int	fill_sem_info(struct file *fp, struct kinfo_file *kif);
129 static int	fill_shm_info(struct file *fp, struct kinfo_file *kif);
130 static int	fill_socket_info(struct socket *so, struct kinfo_file *kif);
131 static int	fill_vnode_info(struct vnode *vp, struct kinfo_file *kif);
132 
133 /*
134  * Each process has:
135  *
136  * - An array of open file descriptors (fd_ofiles)
137  * - An array of file flags (fd_ofileflags)
138  * - A bitmap recording which descriptors are in use (fd_map)
139  *
140  * A process starts out with NDFILE descriptors.  The value of NDFILE has
141  * been selected based the historical limit of 20 open files, and an
142  * assumption that the majority of processes, especially short-lived
143  * processes like shells, will never need more.
144  *
145  * If this initial allocation is exhausted, a larger descriptor table and
146  * map are allocated dynamically, and the pointers in the process's struct
147  * filedesc are updated to point to those.  This is repeated every time
148  * the process runs out of file descriptors (provided it hasn't hit its
149  * resource limit).
150  *
151  * Since threads may hold references to individual descriptor table
152  * entries, the tables are never freed.  Instead, they are placed on a
153  * linked list and freed only when the struct filedesc is released.
154  */
155 #define NDFILE		20
156 #define NDSLOTSIZE	sizeof(NDSLOTTYPE)
157 #define	NDENTRIES	(NDSLOTSIZE * __CHAR_BIT)
158 #define NDSLOT(x)	((x) / NDENTRIES)
159 #define NDBIT(x)	((NDSLOTTYPE)1 << ((x) % NDENTRIES))
160 #define	NDSLOTS(x)	(((x) + NDENTRIES - 1) / NDENTRIES)
161 
162 /*
163  * SLIST entry used to keep track of ofiles which must be reclaimed when
164  * the process exits.
165  */
166 struct freetable {
167 	struct filedescent *ft_table;
168 	SLIST_ENTRY(freetable) ft_next;
169 };
170 
171 /*
172  * Initial allocation: a filedesc structure + the head of SLIST used to
173  * keep track of old ofiles + enough space for NDFILE descriptors.
174  */
175 struct filedesc0 {
176 	struct filedesc fd_fd;
177 	SLIST_HEAD(, freetable) fd_free;
178 	struct	filedescent fd_dfiles[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].fde_file == NULL,
285 	    ("fd=%d is still in use", fd));
286 
287 	fdp->fd_map[NDSLOT(fd)] &= ~NDBIT(fd);
288 	if (fd < fdp->fd_freefile)
289 		fdp->fd_freefile = fd;
290 	if (fd == fdp->fd_lastfile)
291 		fdp->fd_lastfile = fd_last_used(fdp, fd);
292 }
293 
294 /*
295  * Free a file descriptor.
296  */
297 static inline void
298 fdfree(struct filedesc *fdp, int fd)
299 {
300 	struct filedescent *fde;
301 
302 	fde = &fdp->fd_ofiles[fd];
303 	filecaps_free(&fde->fde_caps);
304 	bzero(fde, sizeof(*fde));
305 	fdunused(fdp, fd);
306 }
307 
308 /*
309  * System calls on descriptors.
310  */
311 #ifndef _SYS_SYSPROTO_H_
312 struct getdtablesize_args {
313 	int	dummy;
314 };
315 #endif
316 /* ARGSUSED */
317 int
318 sys_getdtablesize(struct thread *td, struct getdtablesize_args *uap)
319 {
320 	struct proc *p = td->td_proc;
321 	uint64_t lim;
322 
323 	PROC_LOCK(p);
324 	td->td_retval[0] =
325 	    min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc);
326 	lim = racct_get_limit(td->td_proc, RACCT_NOFILE);
327 	PROC_UNLOCK(p);
328 	if (lim < td->td_retval[0])
329 		td->td_retval[0] = lim;
330 	return (0);
331 }
332 
333 /*
334  * Duplicate a file descriptor to a particular value.
335  *
336  * Note: keep in mind that a potential race condition exists when closing
337  * descriptors from a shared descriptor table (via rfork).
338  */
339 #ifndef _SYS_SYSPROTO_H_
340 struct dup2_args {
341 	u_int	from;
342 	u_int	to;
343 };
344 #endif
345 /* ARGSUSED */
346 int
347 sys_dup2(struct thread *td, struct dup2_args *uap)
348 {
349 
350 	return (do_dup(td, DUP_FIXED, (int)uap->from, (int)uap->to,
351 		    td->td_retval));
352 }
353 
354 /*
355  * Duplicate a file descriptor.
356  */
357 #ifndef _SYS_SYSPROTO_H_
358 struct dup_args {
359 	u_int	fd;
360 };
361 #endif
362 /* ARGSUSED */
363 int
364 sys_dup(struct thread *td, struct dup_args *uap)
365 {
366 
367 	return (do_dup(td, 0, (int)uap->fd, 0, td->td_retval));
368 }
369 
370 /*
371  * The file control system call.
372  */
373 #ifndef _SYS_SYSPROTO_H_
374 struct fcntl_args {
375 	int	fd;
376 	int	cmd;
377 	long	arg;
378 };
379 #endif
380 /* ARGSUSED */
381 int
382 sys_fcntl(struct thread *td, struct fcntl_args *uap)
383 {
384 	struct flock fl;
385 	struct __oflock ofl;
386 	intptr_t arg;
387 	int error;
388 	int cmd;
389 
390 	error = 0;
391 	cmd = uap->cmd;
392 	switch (uap->cmd) {
393 	case F_OGETLK:
394 	case F_OSETLK:
395 	case F_OSETLKW:
396 		/*
397 		 * Convert old flock structure to new.
398 		 */
399 		error = copyin((void *)(intptr_t)uap->arg, &ofl, sizeof(ofl));
400 		fl.l_start = ofl.l_start;
401 		fl.l_len = ofl.l_len;
402 		fl.l_pid = ofl.l_pid;
403 		fl.l_type = ofl.l_type;
404 		fl.l_whence = ofl.l_whence;
405 		fl.l_sysid = 0;
406 
407 		switch (uap->cmd) {
408 		case F_OGETLK:
409 		    cmd = F_GETLK;
410 		    break;
411 		case F_OSETLK:
412 		    cmd = F_SETLK;
413 		    break;
414 		case F_OSETLKW:
415 		    cmd = F_SETLKW;
416 		    break;
417 		}
418 		arg = (intptr_t)&fl;
419 		break;
420         case F_GETLK:
421         case F_SETLK:
422         case F_SETLKW:
423 	case F_SETLK_REMOTE:
424                 error = copyin((void *)(intptr_t)uap->arg, &fl, sizeof(fl));
425                 arg = (intptr_t)&fl;
426                 break;
427 	default:
428 		arg = uap->arg;
429 		break;
430 	}
431 	if (error)
432 		return (error);
433 	error = kern_fcntl(td, uap->fd, cmd, arg);
434 	if (error)
435 		return (error);
436 	if (uap->cmd == F_OGETLK) {
437 		ofl.l_start = fl.l_start;
438 		ofl.l_len = fl.l_len;
439 		ofl.l_pid = fl.l_pid;
440 		ofl.l_type = fl.l_type;
441 		ofl.l_whence = fl.l_whence;
442 		error = copyout(&ofl, (void *)(intptr_t)uap->arg, sizeof(ofl));
443 	} else if (uap->cmd == F_GETLK) {
444 		error = copyout(&fl, (void *)(intptr_t)uap->arg, sizeof(fl));
445 	}
446 	return (error);
447 }
448 
449 int
450 kern_fcntl(struct thread *td, int fd, int cmd, intptr_t arg)
451 {
452 	struct filedesc *fdp;
453 	struct flock *flp;
454 	struct file *fp, *fp2;
455 	struct filedescent *fde;
456 	struct proc *p;
457 	struct vnode *vp;
458 	int error, flg, tmp;
459 	u_int old, new;
460 	uint64_t bsize;
461 	off_t foffset;
462 
463 	error = 0;
464 	flg = F_POSIX;
465 	p = td->td_proc;
466 	fdp = p->p_fd;
467 
468 	switch (cmd) {
469 	case F_DUPFD:
470 		tmp = arg;
471 		error = do_dup(td, DUP_FCNTL, fd, tmp, td->td_retval);
472 		break;
473 
474 	case F_DUPFD_CLOEXEC:
475 		tmp = arg;
476 		error = do_dup(td, DUP_FCNTL | DUP_CLOEXEC, fd, tmp,
477 		    td->td_retval);
478 		break;
479 
480 	case F_DUP2FD:
481 		tmp = arg;
482 		error = do_dup(td, DUP_FIXED, fd, tmp, td->td_retval);
483 		break;
484 
485 	case F_DUP2FD_CLOEXEC:
486 		tmp = arg;
487 		error = do_dup(td, DUP_FIXED | DUP_CLOEXEC, fd, tmp,
488 		    td->td_retval);
489 		break;
490 
491 	case F_GETFD:
492 		FILEDESC_SLOCK(fdp);
493 		if ((fp = fget_locked(fdp, fd)) == NULL) {
494 			FILEDESC_SUNLOCK(fdp);
495 			error = EBADF;
496 			break;
497 		}
498 		fde = &fdp->fd_ofiles[fd];
499 		td->td_retval[0] =
500 		    (fde->fde_flags & UF_EXCLOSE) ? FD_CLOEXEC : 0;
501 		FILEDESC_SUNLOCK(fdp);
502 		break;
503 
504 	case F_SETFD:
505 		FILEDESC_XLOCK(fdp);
506 		if ((fp = fget_locked(fdp, fd)) == NULL) {
507 			FILEDESC_XUNLOCK(fdp);
508 			error = EBADF;
509 			break;
510 		}
511 		fde = &fdp->fd_ofiles[fd];
512 		fde->fde_flags = (fde->fde_flags & ~UF_EXCLOSE) |
513 		    (arg & FD_CLOEXEC ? UF_EXCLOSE : 0);
514 		FILEDESC_XUNLOCK(fdp);
515 		break;
516 
517 	case F_GETFL:
518 		error = fget_unlocked(fdp, fd, CAP_FCNTL, F_GETFL, &fp, NULL);
519 		if (error != 0)
520 			break;
521 		td->td_retval[0] = OFLAGS(fp->f_flag);
522 		fdrop(fp, td);
523 		break;
524 
525 	case F_SETFL:
526 		error = fget_unlocked(fdp, fd, CAP_FCNTL, F_SETFL, &fp, NULL);
527 		if (error != 0)
528 			break;
529 		do {
530 			tmp = flg = fp->f_flag;
531 			tmp &= ~FCNTLFLAGS;
532 			tmp |= FFLAGS(arg & ~O_ACCMODE) & FCNTLFLAGS;
533 		} while(atomic_cmpset_int(&fp->f_flag, flg, tmp) == 0);
534 		tmp = fp->f_flag & FNONBLOCK;
535 		error = fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
536 		if (error != 0) {
537 			fdrop(fp, td);
538 			break;
539 		}
540 		tmp = fp->f_flag & FASYNC;
541 		error = fo_ioctl(fp, FIOASYNC, &tmp, td->td_ucred, td);
542 		if (error == 0) {
543 			fdrop(fp, td);
544 			break;
545 		}
546 		atomic_clear_int(&fp->f_flag, FNONBLOCK);
547 		tmp = 0;
548 		(void)fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
549 		fdrop(fp, td);
550 		break;
551 
552 	case F_GETOWN:
553 		error = fget_unlocked(fdp, fd, CAP_FCNTL, F_GETOWN, &fp, NULL);
554 		if (error != 0)
555 			break;
556 		error = fo_ioctl(fp, FIOGETOWN, &tmp, td->td_ucred, td);
557 		if (error == 0)
558 			td->td_retval[0] = tmp;
559 		fdrop(fp, td);
560 		break;
561 
562 	case F_SETOWN:
563 		error = fget_unlocked(fdp, fd, CAP_FCNTL, F_SETOWN, &fp, NULL);
564 		if (error != 0)
565 			break;
566 		tmp = arg;
567 		error = fo_ioctl(fp, FIOSETOWN, &tmp, td->td_ucred, td);
568 		fdrop(fp, td);
569 		break;
570 
571 	case F_SETLK_REMOTE:
572 		error = priv_check(td, PRIV_NFS_LOCKD);
573 		if (error)
574 			return (error);
575 		flg = F_REMOTE;
576 		goto do_setlk;
577 
578 	case F_SETLKW:
579 		flg |= F_WAIT;
580 		/* FALLTHROUGH F_SETLK */
581 
582 	case F_SETLK:
583 	do_setlk:
584 		error = fget_unlocked(fdp, fd, CAP_FLOCK, 0, &fp, NULL);
585 		if (error != 0)
586 			break;
587 		if (fp->f_type != DTYPE_VNODE) {
588 			error = EBADF;
589 			fdrop(fp, td);
590 			break;
591 		}
592 
593 		flp = (struct flock *)arg;
594 		if (flp->l_whence == SEEK_CUR) {
595 			foffset = foffset_get(fp);
596 			if (foffset < 0 ||
597 			    (flp->l_start > 0 &&
598 			     foffset > OFF_MAX - flp->l_start)) {
599 				FILEDESC_SUNLOCK(fdp);
600 				error = EOVERFLOW;
601 				fdrop(fp, td);
602 				break;
603 			}
604 			flp->l_start += foffset;
605 		}
606 
607 		vp = fp->f_vnode;
608 		switch (flp->l_type) {
609 		case F_RDLCK:
610 			if ((fp->f_flag & FREAD) == 0) {
611 				error = EBADF;
612 				break;
613 			}
614 			PROC_LOCK(p->p_leader);
615 			p->p_leader->p_flag |= P_ADVLOCK;
616 			PROC_UNLOCK(p->p_leader);
617 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
618 			    flp, flg);
619 			break;
620 		case F_WRLCK:
621 			if ((fp->f_flag & FWRITE) == 0) {
622 				error = EBADF;
623 				break;
624 			}
625 			PROC_LOCK(p->p_leader);
626 			p->p_leader->p_flag |= P_ADVLOCK;
627 			PROC_UNLOCK(p->p_leader);
628 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
629 			    flp, flg);
630 			break;
631 		case F_UNLCK:
632 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
633 			    flp, flg);
634 			break;
635 		case F_UNLCKSYS:
636 			/*
637 			 * Temporary api for testing remote lock
638 			 * infrastructure.
639 			 */
640 			if (flg != F_REMOTE) {
641 				error = EINVAL;
642 				break;
643 			}
644 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
645 			    F_UNLCKSYS, flp, flg);
646 			break;
647 		default:
648 			error = EINVAL;
649 			break;
650 		}
651 		if (error != 0 || flp->l_type == F_UNLCK ||
652 		    flp->l_type == F_UNLCKSYS) {
653 			fdrop(fp, td);
654 			break;
655 		}
656 
657 		/*
658 		 * Check for a race with close.
659 		 *
660 		 * The vnode is now advisory locked (or unlocked, but this case
661 		 * is not really important) as the caller requested.
662 		 * We had to drop the filedesc lock, so we need to recheck if
663 		 * the descriptor is still valid, because if it was closed
664 		 * in the meantime we need to remove advisory lock from the
665 		 * vnode - close on any descriptor leading to an advisory
666 		 * locked vnode, removes that lock.
667 		 * We will return 0 on purpose in that case, as the result of
668 		 * successful advisory lock might have been externally visible
669 		 * already. This is fine - effectively we pretend to the caller
670 		 * that the closing thread was a bit slower and that the
671 		 * advisory lock succeeded before the close.
672 		 */
673 		error = fget_unlocked(fdp, fd, 0, 0, &fp2, NULL);
674 		if (error != 0) {
675 			fdrop(fp, td);
676 			break;
677 		}
678 		if (fp != fp2) {
679 			flp->l_whence = SEEK_SET;
680 			flp->l_start = 0;
681 			flp->l_len = 0;
682 			flp->l_type = F_UNLCK;
683 			(void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
684 			    F_UNLCK, flp, F_POSIX);
685 		}
686 		fdrop(fp, td);
687 		fdrop(fp2, td);
688 		break;
689 
690 	case F_GETLK:
691 		error = fget_unlocked(fdp, fd, CAP_FLOCK, 0, &fp, NULL);
692 		if (error != 0)
693 			break;
694 		if (fp->f_type != DTYPE_VNODE) {
695 			error = EBADF;
696 			fdrop(fp, td);
697 			break;
698 		}
699 		flp = (struct flock *)arg;
700 		if (flp->l_type != F_RDLCK && flp->l_type != F_WRLCK &&
701 		    flp->l_type != F_UNLCK) {
702 			error = EINVAL;
703 			fdrop(fp, td);
704 			break;
705 		}
706 		if (flp->l_whence == SEEK_CUR) {
707 			foffset = foffset_get(fp);
708 			if ((flp->l_start > 0 &&
709 			    foffset > OFF_MAX - flp->l_start) ||
710 			    (flp->l_start < 0 &&
711 			     foffset < OFF_MIN - flp->l_start)) {
712 				FILEDESC_SUNLOCK(fdp);
713 				error = EOVERFLOW;
714 				fdrop(fp, td);
715 				break;
716 			}
717 			flp->l_start += foffset;
718 		}
719 		vp = fp->f_vnode;
720 		error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, flp,
721 		    F_POSIX);
722 		fdrop(fp, td);
723 		break;
724 
725 	case F_RDAHEAD:
726 		arg = arg ? 128 * 1024: 0;
727 		/* FALLTHROUGH */
728 	case F_READAHEAD:
729 		error = fget_unlocked(fdp, fd, 0, 0, &fp, NULL);
730 		if (error != 0)
731 			break;
732 		if (fp->f_type != DTYPE_VNODE) {
733 			fdrop(fp, td);
734 			error = EBADF;
735 			break;
736 		}
737 		if (arg >= 0) {
738 			vp = fp->f_vnode;
739 			error = vn_lock(vp, LK_SHARED);
740 			if (error != 0) {
741 				fdrop(fp, td);
742 				break;
743 			}
744 			bsize = fp->f_vnode->v_mount->mnt_stat.f_iosize;
745 			VOP_UNLOCK(vp, 0);
746 			fp->f_seqcount = (arg + bsize - 1) / bsize;
747 			do {
748 				new = old = fp->f_flag;
749 				new |= FRDAHEAD;
750 			} while (!atomic_cmpset_rel_int(&fp->f_flag, old, new));
751 		} else {
752 			do {
753 				new = old = fp->f_flag;
754 				new &= ~FRDAHEAD;
755 			} while (!atomic_cmpset_rel_int(&fp->f_flag, old, new));
756 		}
757 		fdrop(fp, td);
758 		break;
759 
760 	default:
761 		error = EINVAL;
762 		break;
763 	}
764 	return (error);
765 }
766 
767 /*
768  * Common code for dup, dup2, fcntl(F_DUPFD) and fcntl(F_DUP2FD).
769  */
770 int
771 do_dup(struct thread *td, int flags, int old, int new,
772     register_t *retval)
773 {
774 	struct filedesc *fdp;
775 	struct filedescent *oldfde, *newfde;
776 	struct proc *p;
777 	struct file *fp;
778 	struct file *delfp;
779 	int error, maxfd;
780 
781 	p = td->td_proc;
782 	fdp = p->p_fd;
783 
784 	/*
785 	 * Verify we have a valid descriptor to dup from and possibly to
786 	 * dup to. Unlike dup() and dup2(), fcntl()'s F_DUPFD should
787 	 * return EINVAL when the new descriptor is out of bounds.
788 	 */
789 	if (old < 0)
790 		return (EBADF);
791 	if (new < 0)
792 		return (flags & DUP_FCNTL ? EINVAL : EBADF);
793 	PROC_LOCK(p);
794 	maxfd = min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc);
795 	PROC_UNLOCK(p);
796 	if (new >= maxfd)
797 		return (flags & DUP_FCNTL ? EINVAL : EBADF);
798 
799 	FILEDESC_XLOCK(fdp);
800 	if (fget_locked(fdp, old) == NULL) {
801 		FILEDESC_XUNLOCK(fdp);
802 		return (EBADF);
803 	}
804 	oldfde = &fdp->fd_ofiles[old];
805 	if (flags & DUP_FIXED && old == new) {
806 		*retval = new;
807 		if (flags & DUP_CLOEXEC)
808 			fdp->fd_ofiles[new].fde_flags |= UF_EXCLOSE;
809 		FILEDESC_XUNLOCK(fdp);
810 		return (0);
811 	}
812 	fp = oldfde->fde_file;
813 	fhold(fp);
814 
815 	/*
816 	 * If the caller specified a file descriptor, make sure the file
817 	 * table is large enough to hold it, and grab it.  Otherwise, just
818 	 * allocate a new descriptor the usual way.
819 	 */
820 	if (flags & DUP_FIXED) {
821 		if (new >= fdp->fd_nfiles) {
822 			/*
823 			 * The resource limits are here instead of e.g.
824 			 * fdalloc(), because the file descriptor table may be
825 			 * shared between processes, so we can't really use
826 			 * racct_add()/racct_sub().  Instead of counting the
827 			 * number of actually allocated descriptors, just put
828 			 * the limit on the size of the file descriptor table.
829 			 */
830 #ifdef RACCT
831 			PROC_LOCK(p);
832 			error = racct_set(p, RACCT_NOFILE, new + 1);
833 			PROC_UNLOCK(p);
834 			if (error != 0) {
835 				FILEDESC_XUNLOCK(fdp);
836 				fdrop(fp, td);
837 				return (EMFILE);
838 			}
839 #endif
840 			fdgrowtable(fdp, new + 1);
841 			oldfde = &fdp->fd_ofiles[old];
842 		}
843 		newfde = &fdp->fd_ofiles[new];
844 		if (newfde->fde_file == NULL)
845 			fdused(fdp, new);
846 	} else {
847 		if ((error = fdalloc(td, new, &new)) != 0) {
848 			FILEDESC_XUNLOCK(fdp);
849 			fdrop(fp, td);
850 			return (error);
851 		}
852 		newfde = &fdp->fd_ofiles[new];
853 	}
854 
855 	KASSERT(fp == oldfde->fde_file, ("old fd has been modified"));
856 	KASSERT(old != new, ("new fd is same as old"));
857 
858 	delfp = newfde->fde_file;
859 
860 	/*
861 	 * Duplicate the source descriptor.
862 	 */
863 	*newfde = *oldfde;
864 	filecaps_copy(&oldfde->fde_caps, &newfde->fde_caps);
865 	if ((flags & DUP_CLOEXEC) != 0)
866 		newfde->fde_flags = oldfde->fde_flags | UF_EXCLOSE;
867 	else
868 		newfde->fde_flags = oldfde->fde_flags & ~UF_EXCLOSE;
869 	if (new > fdp->fd_lastfile)
870 		fdp->fd_lastfile = new;
871 	*retval = new;
872 
873 	if (delfp != NULL) {
874 		(void) closefp(fdp, new, delfp, td, 1);
875 		/* closefp() drops the FILEDESC lock for us. */
876 	} else {
877 		FILEDESC_XUNLOCK(fdp);
878 	}
879 
880 	return (0);
881 }
882 
883 /*
884  * If sigio is on the list associated with a process or process group,
885  * disable signalling from the device, remove sigio from the list and
886  * free sigio.
887  */
888 void
889 funsetown(struct sigio **sigiop)
890 {
891 	struct sigio *sigio;
892 
893 	SIGIO_LOCK();
894 	sigio = *sigiop;
895 	if (sigio == NULL) {
896 		SIGIO_UNLOCK();
897 		return;
898 	}
899 	*(sigio->sio_myref) = NULL;
900 	if ((sigio)->sio_pgid < 0) {
901 		struct pgrp *pg = (sigio)->sio_pgrp;
902 		PGRP_LOCK(pg);
903 		SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio,
904 			     sigio, sio_pgsigio);
905 		PGRP_UNLOCK(pg);
906 	} else {
907 		struct proc *p = (sigio)->sio_proc;
908 		PROC_LOCK(p);
909 		SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio,
910 			     sigio, sio_pgsigio);
911 		PROC_UNLOCK(p);
912 	}
913 	SIGIO_UNLOCK();
914 	crfree(sigio->sio_ucred);
915 	free(sigio, M_SIGIO);
916 }
917 
918 /*
919  * Free a list of sigio structures.
920  * We only need to lock the SIGIO_LOCK because we have made ourselves
921  * inaccessible to callers of fsetown and therefore do not need to lock
922  * the proc or pgrp struct for the list manipulation.
923  */
924 void
925 funsetownlst(struct sigiolst *sigiolst)
926 {
927 	struct proc *p;
928 	struct pgrp *pg;
929 	struct sigio *sigio;
930 
931 	sigio = SLIST_FIRST(sigiolst);
932 	if (sigio == NULL)
933 		return;
934 	p = NULL;
935 	pg = NULL;
936 
937 	/*
938 	 * Every entry of the list should belong
939 	 * to a single proc or pgrp.
940 	 */
941 	if (sigio->sio_pgid < 0) {
942 		pg = sigio->sio_pgrp;
943 		PGRP_LOCK_ASSERT(pg, MA_NOTOWNED);
944 	} else /* if (sigio->sio_pgid > 0) */ {
945 		p = sigio->sio_proc;
946 		PROC_LOCK_ASSERT(p, MA_NOTOWNED);
947 	}
948 
949 	SIGIO_LOCK();
950 	while ((sigio = SLIST_FIRST(sigiolst)) != NULL) {
951 		*(sigio->sio_myref) = NULL;
952 		if (pg != NULL) {
953 			KASSERT(sigio->sio_pgid < 0,
954 			    ("Proc sigio in pgrp sigio list"));
955 			KASSERT(sigio->sio_pgrp == pg,
956 			    ("Bogus pgrp in sigio list"));
957 			PGRP_LOCK(pg);
958 			SLIST_REMOVE(&pg->pg_sigiolst, sigio, sigio,
959 			    sio_pgsigio);
960 			PGRP_UNLOCK(pg);
961 		} else /* if (p != NULL) */ {
962 			KASSERT(sigio->sio_pgid > 0,
963 			    ("Pgrp sigio in proc sigio list"));
964 			KASSERT(sigio->sio_proc == p,
965 			    ("Bogus proc in sigio list"));
966 			PROC_LOCK(p);
967 			SLIST_REMOVE(&p->p_sigiolst, sigio, sigio,
968 			    sio_pgsigio);
969 			PROC_UNLOCK(p);
970 		}
971 		SIGIO_UNLOCK();
972 		crfree(sigio->sio_ucred);
973 		free(sigio, M_SIGIO);
974 		SIGIO_LOCK();
975 	}
976 	SIGIO_UNLOCK();
977 }
978 
979 /*
980  * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg).
981  *
982  * After permission checking, add a sigio structure to the sigio list for
983  * the process or process group.
984  */
985 int
986 fsetown(pid_t pgid, struct sigio **sigiop)
987 {
988 	struct proc *proc;
989 	struct pgrp *pgrp;
990 	struct sigio *sigio;
991 	int ret;
992 
993 	if (pgid == 0) {
994 		funsetown(sigiop);
995 		return (0);
996 	}
997 
998 	ret = 0;
999 
1000 	/* Allocate and fill in the new sigio out of locks. */
1001 	sigio = malloc(sizeof(struct sigio), M_SIGIO, M_WAITOK);
1002 	sigio->sio_pgid = pgid;
1003 	sigio->sio_ucred = crhold(curthread->td_ucred);
1004 	sigio->sio_myref = sigiop;
1005 
1006 	sx_slock(&proctree_lock);
1007 	if (pgid > 0) {
1008 		proc = pfind(pgid);
1009 		if (proc == NULL) {
1010 			ret = ESRCH;
1011 			goto fail;
1012 		}
1013 
1014 		/*
1015 		 * Policy - Don't allow a process to FSETOWN a process
1016 		 * in another session.
1017 		 *
1018 		 * Remove this test to allow maximum flexibility or
1019 		 * restrict FSETOWN to the current process or process
1020 		 * group for maximum safety.
1021 		 */
1022 		PROC_UNLOCK(proc);
1023 		if (proc->p_session != curthread->td_proc->p_session) {
1024 			ret = EPERM;
1025 			goto fail;
1026 		}
1027 
1028 		pgrp = NULL;
1029 	} else /* if (pgid < 0) */ {
1030 		pgrp = pgfind(-pgid);
1031 		if (pgrp == NULL) {
1032 			ret = ESRCH;
1033 			goto fail;
1034 		}
1035 		PGRP_UNLOCK(pgrp);
1036 
1037 		/*
1038 		 * Policy - Don't allow a process to FSETOWN a process
1039 		 * in another session.
1040 		 *
1041 		 * Remove this test to allow maximum flexibility or
1042 		 * restrict FSETOWN to the current process or process
1043 		 * group for maximum safety.
1044 		 */
1045 		if (pgrp->pg_session != curthread->td_proc->p_session) {
1046 			ret = EPERM;
1047 			goto fail;
1048 		}
1049 
1050 		proc = NULL;
1051 	}
1052 	funsetown(sigiop);
1053 	if (pgid > 0) {
1054 		PROC_LOCK(proc);
1055 		/*
1056 		 * Since funsetownlst() is called without the proctree
1057 		 * locked, we need to check for P_WEXIT.
1058 		 * XXX: is ESRCH correct?
1059 		 */
1060 		if ((proc->p_flag & P_WEXIT) != 0) {
1061 			PROC_UNLOCK(proc);
1062 			ret = ESRCH;
1063 			goto fail;
1064 		}
1065 		SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio);
1066 		sigio->sio_proc = proc;
1067 		PROC_UNLOCK(proc);
1068 	} else {
1069 		PGRP_LOCK(pgrp);
1070 		SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio);
1071 		sigio->sio_pgrp = pgrp;
1072 		PGRP_UNLOCK(pgrp);
1073 	}
1074 	sx_sunlock(&proctree_lock);
1075 	SIGIO_LOCK();
1076 	*sigiop = sigio;
1077 	SIGIO_UNLOCK();
1078 	return (0);
1079 
1080 fail:
1081 	sx_sunlock(&proctree_lock);
1082 	crfree(sigio->sio_ucred);
1083 	free(sigio, M_SIGIO);
1084 	return (ret);
1085 }
1086 
1087 /*
1088  * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg).
1089  */
1090 pid_t
1091 fgetown(sigiop)
1092 	struct sigio **sigiop;
1093 {
1094 	pid_t pgid;
1095 
1096 	SIGIO_LOCK();
1097 	pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0;
1098 	SIGIO_UNLOCK();
1099 	return (pgid);
1100 }
1101 
1102 /*
1103  * Function drops the filedesc lock on return.
1104  */
1105 static int
1106 closefp(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1107     int holdleaders)
1108 {
1109 	int error;
1110 
1111 	FILEDESC_XLOCK_ASSERT(fdp);
1112 
1113 	if (holdleaders) {
1114 		if (td->td_proc->p_fdtol != NULL) {
1115 			/*
1116 			 * Ask fdfree() to sleep to ensure that all relevant
1117 			 * process leaders can be traversed in closef().
1118 			 */
1119 			fdp->fd_holdleaderscount++;
1120 		} else {
1121 			holdleaders = 0;
1122 		}
1123 	}
1124 
1125 	/*
1126 	 * We now hold the fp reference that used to be owned by the
1127 	 * descriptor array.  We have to unlock the FILEDESC *AFTER*
1128 	 * knote_fdclose to prevent a race of the fd getting opened, a knote
1129 	 * added, and deleteing a knote for the new fd.
1130 	 */
1131 	knote_fdclose(td, fd);
1132 
1133 	/*
1134 	 * We need to notify mqueue if the object is of type mqueue.
1135 	 */
1136 	if (fp->f_type == DTYPE_MQUEUE)
1137 		mq_fdclose(td, fd, fp);
1138 	FILEDESC_XUNLOCK(fdp);
1139 
1140 	error = closef(fp, td);
1141 	if (holdleaders) {
1142 		FILEDESC_XLOCK(fdp);
1143 		fdp->fd_holdleaderscount--;
1144 		if (fdp->fd_holdleaderscount == 0 &&
1145 		    fdp->fd_holdleaderswakeup != 0) {
1146 			fdp->fd_holdleaderswakeup = 0;
1147 			wakeup(&fdp->fd_holdleaderscount);
1148 		}
1149 		FILEDESC_XUNLOCK(fdp);
1150 	}
1151 	return (error);
1152 }
1153 
1154 /*
1155  * Close a file descriptor.
1156  */
1157 #ifndef _SYS_SYSPROTO_H_
1158 struct close_args {
1159 	int     fd;
1160 };
1161 #endif
1162 /* ARGSUSED */
1163 int
1164 sys_close(td, uap)
1165 	struct thread *td;
1166 	struct close_args *uap;
1167 {
1168 
1169 	return (kern_close(td, uap->fd));
1170 }
1171 
1172 int
1173 kern_close(td, fd)
1174 	struct thread *td;
1175 	int fd;
1176 {
1177 	struct filedesc *fdp;
1178 	struct file *fp;
1179 
1180 	fdp = td->td_proc->p_fd;
1181 
1182 	AUDIT_SYSCLOSE(td, fd);
1183 
1184 	FILEDESC_XLOCK(fdp);
1185 	if ((fp = fget_locked(fdp, fd)) == NULL) {
1186 		FILEDESC_XUNLOCK(fdp);
1187 		return (EBADF);
1188 	}
1189 	fdfree(fdp, fd);
1190 
1191 	/* closefp() drops the FILEDESC lock for us. */
1192 	return (closefp(fdp, fd, fp, td, 1));
1193 }
1194 
1195 /*
1196  * Close open file descriptors.
1197  */
1198 #ifndef _SYS_SYSPROTO_H_
1199 struct closefrom_args {
1200 	int	lowfd;
1201 };
1202 #endif
1203 /* ARGSUSED */
1204 int
1205 sys_closefrom(struct thread *td, struct closefrom_args *uap)
1206 {
1207 	struct filedesc *fdp;
1208 	int fd;
1209 
1210 	fdp = td->td_proc->p_fd;
1211 	AUDIT_ARG_FD(uap->lowfd);
1212 
1213 	/*
1214 	 * Treat negative starting file descriptor values identical to
1215 	 * closefrom(0) which closes all files.
1216 	 */
1217 	if (uap->lowfd < 0)
1218 		uap->lowfd = 0;
1219 	FILEDESC_SLOCK(fdp);
1220 	for (fd = uap->lowfd; fd < fdp->fd_nfiles; fd++) {
1221 		if (fdp->fd_ofiles[fd].fde_file != NULL) {
1222 			FILEDESC_SUNLOCK(fdp);
1223 			(void)kern_close(td, fd);
1224 			FILEDESC_SLOCK(fdp);
1225 		}
1226 	}
1227 	FILEDESC_SUNLOCK(fdp);
1228 	return (0);
1229 }
1230 
1231 #if defined(COMPAT_43)
1232 /*
1233  * Return status information about a file descriptor.
1234  */
1235 #ifndef _SYS_SYSPROTO_H_
1236 struct ofstat_args {
1237 	int	fd;
1238 	struct	ostat *sb;
1239 };
1240 #endif
1241 /* ARGSUSED */
1242 int
1243 ofstat(struct thread *td, struct ofstat_args *uap)
1244 {
1245 	struct ostat oub;
1246 	struct stat ub;
1247 	int error;
1248 
1249 	error = kern_fstat(td, uap->fd, &ub);
1250 	if (error == 0) {
1251 		cvtstat(&ub, &oub);
1252 		error = copyout(&oub, uap->sb, sizeof(oub));
1253 	}
1254 	return (error);
1255 }
1256 #endif /* COMPAT_43 */
1257 
1258 /*
1259  * Return status information about a file descriptor.
1260  */
1261 #ifndef _SYS_SYSPROTO_H_
1262 struct fstat_args {
1263 	int	fd;
1264 	struct	stat *sb;
1265 };
1266 #endif
1267 /* ARGSUSED */
1268 int
1269 sys_fstat(struct thread *td, struct fstat_args *uap)
1270 {
1271 	struct stat ub;
1272 	int error;
1273 
1274 	error = kern_fstat(td, uap->fd, &ub);
1275 	if (error == 0)
1276 		error = copyout(&ub, uap->sb, sizeof(ub));
1277 	return (error);
1278 }
1279 
1280 int
1281 kern_fstat(struct thread *td, int fd, struct stat *sbp)
1282 {
1283 	struct file *fp;
1284 	int error;
1285 
1286 	AUDIT_ARG_FD(fd);
1287 
1288 	if ((error = fget(td, fd, CAP_FSTAT, &fp)) != 0)
1289 		return (error);
1290 
1291 	AUDIT_ARG_FILE(td->td_proc, fp);
1292 
1293 	error = fo_stat(fp, sbp, td->td_ucred, td);
1294 	fdrop(fp, td);
1295 #ifdef KTRACE
1296 	if (error == 0 && KTRPOINT(td, KTR_STRUCT))
1297 		ktrstat(sbp);
1298 #endif
1299 	return (error);
1300 }
1301 
1302 /*
1303  * Return status information about a file descriptor.
1304  */
1305 #ifndef _SYS_SYSPROTO_H_
1306 struct nfstat_args {
1307 	int	fd;
1308 	struct	nstat *sb;
1309 };
1310 #endif
1311 /* ARGSUSED */
1312 int
1313 sys_nfstat(struct thread *td, struct nfstat_args *uap)
1314 {
1315 	struct nstat nub;
1316 	struct stat ub;
1317 	int error;
1318 
1319 	error = kern_fstat(td, uap->fd, &ub);
1320 	if (error == 0) {
1321 		cvtnstat(&ub, &nub);
1322 		error = copyout(&nub, uap->sb, sizeof(nub));
1323 	}
1324 	return (error);
1325 }
1326 
1327 /*
1328  * Return pathconf information about a file descriptor.
1329  */
1330 #ifndef _SYS_SYSPROTO_H_
1331 struct fpathconf_args {
1332 	int	fd;
1333 	int	name;
1334 };
1335 #endif
1336 /* ARGSUSED */
1337 int
1338 sys_fpathconf(struct thread *td, struct fpathconf_args *uap)
1339 {
1340 	struct file *fp;
1341 	struct vnode *vp;
1342 	int error;
1343 
1344 	if ((error = fget(td, uap->fd, CAP_FPATHCONF, &fp)) != 0)
1345 		return (error);
1346 
1347 	/* If asynchronous I/O is available, it works for all descriptors. */
1348 	if (uap->name == _PC_ASYNC_IO) {
1349 		td->td_retval[0] = async_io_version;
1350 		goto out;
1351 	}
1352 	vp = fp->f_vnode;
1353 	if (vp != NULL) {
1354 		vn_lock(vp, LK_SHARED | LK_RETRY);
1355 		error = VOP_PATHCONF(vp, uap->name, td->td_retval);
1356 		VOP_UNLOCK(vp, 0);
1357 	} else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) {
1358 		if (uap->name != _PC_PIPE_BUF) {
1359 			error = EINVAL;
1360 		} else {
1361 			td->td_retval[0] = PIPE_BUF;
1362 			error = 0;
1363 		}
1364 	} else {
1365 		error = EOPNOTSUPP;
1366 	}
1367 out:
1368 	fdrop(fp, td);
1369 	return (error);
1370 }
1371 
1372 /*
1373  * Initialize filecaps structure.
1374  */
1375 void
1376 filecaps_init(struct filecaps *fcaps)
1377 {
1378 
1379 	bzero(fcaps, sizeof(*fcaps));
1380 	fcaps->fc_nioctls = -1;
1381 }
1382 
1383 /*
1384  * Copy filecaps structure allocating memory for ioctls array if needed.
1385  */
1386 void
1387 filecaps_copy(const struct filecaps *src, struct filecaps *dst)
1388 {
1389 	size_t size;
1390 
1391 	*dst = *src;
1392 	if (src->fc_ioctls != NULL) {
1393 		KASSERT(src->fc_nioctls > 0,
1394 		    ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1395 
1396 		size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1397 		dst->fc_ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1398 		bcopy(src->fc_ioctls, dst->fc_ioctls, size);
1399 	}
1400 }
1401 
1402 /*
1403  * Move filecaps structure to the new place and clear the old place.
1404  */
1405 void
1406 filecaps_move(struct filecaps *src, struct filecaps *dst)
1407 {
1408 
1409 	*dst = *src;
1410 	bzero(src, sizeof(*src));
1411 }
1412 
1413 /*
1414  * Fill the given filecaps structure with full rights.
1415  */
1416 static void
1417 filecaps_fill(struct filecaps *fcaps)
1418 {
1419 
1420 	fcaps->fc_rights = CAP_ALL;
1421 	fcaps->fc_ioctls = NULL;
1422 	fcaps->fc_nioctls = -1;
1423 	fcaps->fc_fcntls = CAP_FCNTL_ALL;
1424 }
1425 
1426 /*
1427  * Free memory allocated within filecaps structure.
1428  */
1429 void
1430 filecaps_free(struct filecaps *fcaps)
1431 {
1432 
1433 	free(fcaps->fc_ioctls, M_FILECAPS);
1434 	bzero(fcaps, sizeof(*fcaps));
1435 }
1436 
1437 /*
1438  * Validate the given filecaps structure.
1439  */
1440 static void
1441 filecaps_validate(const struct filecaps *fcaps, const char *func)
1442 {
1443 
1444 	KASSERT((fcaps->fc_rights & ~CAP_MASK_VALID) == 0,
1445 	    ("%s: invalid rights", func));
1446 	KASSERT((fcaps->fc_fcntls & ~CAP_FCNTL_ALL) == 0,
1447 	    ("%s: invalid fcntls", func));
1448 	KASSERT(fcaps->fc_fcntls == 0 || (fcaps->fc_rights & CAP_FCNTL) != 0,
1449 	    ("%s: fcntls without CAP_FCNTL", func));
1450 	KASSERT(fcaps->fc_ioctls != NULL ? fcaps->fc_nioctls > 0 :
1451 	    (fcaps->fc_nioctls == -1 || fcaps->fc_nioctls == 0),
1452 	    ("%s: invalid ioctls", func));
1453 	KASSERT(fcaps->fc_nioctls == 0 || (fcaps->fc_rights & CAP_IOCTL) != 0,
1454 	    ("%s: ioctls without CAP_IOCTL", func));
1455 }
1456 
1457 /*
1458  * Grow the file table to accomodate (at least) nfd descriptors.
1459  */
1460 static void
1461 fdgrowtable(struct filedesc *fdp, int nfd)
1462 {
1463 	struct filedesc0 *fdp0;
1464 	struct freetable *ft;
1465 	struct filedescent *ntable;
1466 	struct filedescent *otable;
1467 	int nnfiles, onfiles;
1468 	NDSLOTTYPE *nmap, *omap;
1469 
1470 	FILEDESC_XLOCK_ASSERT(fdp);
1471 
1472 	KASSERT(fdp->fd_nfiles > 0, ("zero-length file table"));
1473 
1474 	/* save old values */
1475 	onfiles = fdp->fd_nfiles;
1476 	otable = fdp->fd_ofiles;
1477 	omap = fdp->fd_map;
1478 
1479 	/* compute the size of the new table */
1480 	nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */
1481 	if (nnfiles <= onfiles)
1482 		/* the table is already large enough */
1483 		return;
1484 
1485 	/*
1486 	 * Allocate a new table and map.  We need enough space for the
1487 	 * file entries themselves and the struct freetable we will use
1488 	 * when we decommission the table and place it on the freelist.
1489 	 * We place the struct freetable in the middle so we don't have
1490 	 * to worry about padding.
1491 	 */
1492 	ntable = malloc(nnfiles * sizeof(ntable[0]) + sizeof(struct freetable),
1493 	    M_FILEDESC, M_ZERO | M_WAITOK);
1494 	nmap = malloc(NDSLOTS(nnfiles) * NDSLOTSIZE, M_FILEDESC,
1495 	    M_ZERO | M_WAITOK);
1496 
1497 	/* copy the old data over and point at the new tables */
1498 	memcpy(ntable, otable, onfiles * sizeof(*otable));
1499 	memcpy(nmap, omap, NDSLOTS(onfiles) * sizeof(*omap));
1500 
1501 	/* update the pointers and counters */
1502 	fdp->fd_nfiles = nnfiles;
1503 	memcpy(ntable, otable, onfiles * sizeof(ntable[0]));
1504 	fdp->fd_ofiles = ntable;
1505 	fdp->fd_map = nmap;
1506 
1507 	/*
1508 	 * Do not free the old file table, as some threads may still
1509 	 * reference entries within it.  Instead, place it on a freelist
1510 	 * which will be processed when the struct filedesc is released.
1511 	 *
1512 	 * Do, however, free the old map.
1513 	 *
1514 	 * Note that if onfiles == NDFILE, we're dealing with the original
1515 	 * static allocation contained within (struct filedesc0 *)fdp,
1516 	 * which must not be freed.
1517 	 */
1518 	if (onfiles > NDFILE) {
1519 		ft = (struct freetable *)&otable[onfiles];
1520 		fdp0 = (struct filedesc0 *)fdp;
1521 		ft->ft_table = otable;
1522 		SLIST_INSERT_HEAD(&fdp0->fd_free, ft, ft_next);
1523 		free(omap, M_FILEDESC);
1524 	}
1525 }
1526 
1527 /*
1528  * Allocate a file descriptor for the process.
1529  */
1530 int
1531 fdalloc(struct thread *td, int minfd, int *result)
1532 {
1533 	struct proc *p = td->td_proc;
1534 	struct filedesc *fdp = p->p_fd;
1535 	int fd = -1, maxfd, allocfd;
1536 #ifdef RACCT
1537 	int error;
1538 #endif
1539 
1540 	FILEDESC_XLOCK_ASSERT(fdp);
1541 
1542 	if (fdp->fd_freefile > minfd)
1543 		minfd = fdp->fd_freefile;
1544 
1545 	PROC_LOCK(p);
1546 	maxfd = min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc);
1547 	PROC_UNLOCK(p);
1548 
1549 	/*
1550 	 * Search the bitmap for a free descriptor starting at minfd.
1551 	 * If none is found, grow the file table.
1552 	 */
1553 	fd = fd_first_free(fdp, minfd, fdp->fd_nfiles);
1554 	if (fd >= maxfd)
1555 		return (EMFILE);
1556 	if (fd >= fdp->fd_nfiles) {
1557 		allocfd = min(fd * 2, maxfd);
1558 #ifdef RACCT
1559 		PROC_LOCK(p);
1560 		error = racct_set(p, RACCT_NOFILE, allocfd);
1561 		PROC_UNLOCK(p);
1562 		if (error != 0)
1563 			return (EMFILE);
1564 #endif
1565 		/*
1566 		 * fd is already equal to first free descriptor >= minfd, so
1567 		 * we only need to grow the table and we are done.
1568 		 */
1569 		fdgrowtable(fdp, allocfd);
1570 	}
1571 
1572 	/*
1573 	 * Perform some sanity checks, then mark the file descriptor as
1574 	 * used and return it to the caller.
1575 	 */
1576 	KASSERT(fd >= 0 && fd < min(maxfd, fdp->fd_nfiles),
1577 	    ("invalid descriptor %d", fd));
1578 	KASSERT(!fdisused(fdp, fd),
1579 	    ("fd_first_free() returned non-free descriptor"));
1580 	KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
1581 	    ("file descriptor isn't free"));
1582 	KASSERT(fdp->fd_ofiles[fd].fde_flags == 0, ("file flags are set"));
1583 	fdused(fdp, fd);
1584 	*result = fd;
1585 	return (0);
1586 }
1587 
1588 /*
1589  * Allocate n file descriptors for the process.
1590  */
1591 int
1592 fdallocn(struct thread *td, int minfd, int *fds, int n)
1593 {
1594 	struct proc *p = td->td_proc;
1595 	struct filedesc *fdp = p->p_fd;
1596 	int i;
1597 
1598 	FILEDESC_XLOCK_ASSERT(fdp);
1599 
1600 	if (!fdavail(td, n))
1601 		return (EMFILE);
1602 
1603 	for (i = 0; i < n; i++)
1604 		if (fdalloc(td, 0, &fds[i]) != 0)
1605 			break;
1606 
1607 	if (i < n) {
1608 		for (i--; i >= 0; i--)
1609 			fdunused(fdp, fds[i]);
1610 		return (EMFILE);
1611 	}
1612 
1613 	return (0);
1614 }
1615 
1616 /*
1617  * Check to see whether n user file descriptors are available to the process
1618  * p.
1619  */
1620 int
1621 fdavail(struct thread *td, int n)
1622 {
1623 	struct proc *p = td->td_proc;
1624 	struct filedesc *fdp = td->td_proc->p_fd;
1625 	int i, lim, last;
1626 
1627 	FILEDESC_LOCK_ASSERT(fdp);
1628 
1629 	/*
1630 	 * XXX: This is only called from uipc_usrreq.c:unp_externalize();
1631 	 *      call racct_add() from there instead of dealing with containers
1632 	 *      here.
1633 	 */
1634 	PROC_LOCK(p);
1635 	lim = min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc);
1636 	PROC_UNLOCK(p);
1637 	if ((i = lim - fdp->fd_nfiles) > 0 && (n -= i) <= 0)
1638 		return (1);
1639 	last = min(fdp->fd_nfiles, lim);
1640 	for (i = fdp->fd_freefile; i < last; i++) {
1641 		if (fdp->fd_ofiles[i].fde_file == NULL && --n <= 0)
1642 			return (1);
1643 	}
1644 	return (0);
1645 }
1646 
1647 /*
1648  * Create a new open file structure and allocate a file decriptor for the
1649  * process that refers to it.  We add one reference to the file for the
1650  * descriptor table and one reference for resultfp. This is to prevent us
1651  * being preempted and the entry in the descriptor table closed after we
1652  * release the FILEDESC lock.
1653  */
1654 int
1655 falloc(struct thread *td, struct file **resultfp, int *resultfd, int flags)
1656 {
1657 	struct file *fp;
1658 	int error, fd;
1659 
1660 	error = falloc_noinstall(td, &fp);
1661 	if (error)
1662 		return (error);		/* no reference held on error */
1663 
1664 	error = finstall(td, fp, &fd, flags, NULL);
1665 	if (error) {
1666 		fdrop(fp, td);		/* one reference (fp only) */
1667 		return (error);
1668 	}
1669 
1670 	if (resultfp != NULL)
1671 		*resultfp = fp;		/* copy out result */
1672 	else
1673 		fdrop(fp, td);		/* release local reference */
1674 
1675 	if (resultfd != NULL)
1676 		*resultfd = fd;
1677 
1678 	return (0);
1679 }
1680 
1681 /*
1682  * Create a new open file structure without allocating a file descriptor.
1683  */
1684 int
1685 falloc_noinstall(struct thread *td, struct file **resultfp)
1686 {
1687 	struct file *fp;
1688 	int maxuserfiles = maxfiles - (maxfiles / 20);
1689 	static struct timeval lastfail;
1690 	static int curfail;
1691 
1692 	KASSERT(resultfp != NULL, ("%s: resultfp == NULL", __func__));
1693 
1694 	if ((openfiles >= maxuserfiles &&
1695 	    priv_check(td, PRIV_MAXFILES) != 0) ||
1696 	    openfiles >= maxfiles) {
1697 		if (ppsratecheck(&lastfail, &curfail, 1)) {
1698 			printf("kern.maxfiles limit exceeded by uid %i, "
1699 			    "please see tuning(7).\n", td->td_ucred->cr_ruid);
1700 		}
1701 		return (ENFILE);
1702 	}
1703 	atomic_add_int(&openfiles, 1);
1704 	fp = uma_zalloc(file_zone, M_WAITOK | M_ZERO);
1705 	refcount_init(&fp->f_count, 1);
1706 	fp->f_cred = crhold(td->td_ucred);
1707 	fp->f_ops = &badfileops;
1708 	fp->f_data = NULL;
1709 	fp->f_vnode = NULL;
1710 	*resultfp = fp;
1711 	return (0);
1712 }
1713 
1714 /*
1715  * Install a file in a file descriptor table.
1716  */
1717 int
1718 finstall(struct thread *td, struct file *fp, int *fd, int flags,
1719     struct filecaps *fcaps)
1720 {
1721 	struct filedesc *fdp = td->td_proc->p_fd;
1722 	struct filedescent *fde;
1723 	int error;
1724 
1725 	KASSERT(fd != NULL, ("%s: fd == NULL", __func__));
1726 	KASSERT(fp != NULL, ("%s: fp == NULL", __func__));
1727 	if (fcaps != NULL)
1728 		filecaps_validate(fcaps, __func__);
1729 
1730 	FILEDESC_XLOCK(fdp);
1731 	if ((error = fdalloc(td, 0, fd))) {
1732 		FILEDESC_XUNLOCK(fdp);
1733 		return (error);
1734 	}
1735 	fhold(fp);
1736 	fde = &fdp->fd_ofiles[*fd];
1737 	fde->fde_file = fp;
1738 	if ((flags & O_CLOEXEC) != 0)
1739 		fde->fde_flags |= UF_EXCLOSE;
1740 	if (fcaps != NULL)
1741 		filecaps_move(fcaps, &fde->fde_caps);
1742 	else
1743 		filecaps_fill(&fde->fde_caps);
1744 	FILEDESC_XUNLOCK(fdp);
1745 	return (0);
1746 }
1747 
1748 /*
1749  * Build a new filedesc structure from another.
1750  * Copy the current, root, and jail root vnode references.
1751  */
1752 struct filedesc *
1753 fdinit(struct filedesc *fdp)
1754 {
1755 	struct filedesc0 *newfdp;
1756 
1757 	newfdp = malloc(sizeof *newfdp, M_FILEDESC, M_WAITOK | M_ZERO);
1758 	FILEDESC_LOCK_INIT(&newfdp->fd_fd);
1759 	if (fdp != NULL) {
1760 		FILEDESC_XLOCK(fdp);
1761 		newfdp->fd_fd.fd_cdir = fdp->fd_cdir;
1762 		if (newfdp->fd_fd.fd_cdir)
1763 			VREF(newfdp->fd_fd.fd_cdir);
1764 		newfdp->fd_fd.fd_rdir = fdp->fd_rdir;
1765 		if (newfdp->fd_fd.fd_rdir)
1766 			VREF(newfdp->fd_fd.fd_rdir);
1767 		newfdp->fd_fd.fd_jdir = fdp->fd_jdir;
1768 		if (newfdp->fd_fd.fd_jdir)
1769 			VREF(newfdp->fd_fd.fd_jdir);
1770 		FILEDESC_XUNLOCK(fdp);
1771 	}
1772 
1773 	/* Create the file descriptor table. */
1774 	newfdp->fd_fd.fd_refcnt = 1;
1775 	newfdp->fd_fd.fd_holdcnt = 1;
1776 	newfdp->fd_fd.fd_cmask = CMASK;
1777 	newfdp->fd_fd.fd_ofiles = newfdp->fd_dfiles;
1778 	newfdp->fd_fd.fd_nfiles = NDFILE;
1779 	newfdp->fd_fd.fd_map = newfdp->fd_dmap;
1780 	newfdp->fd_fd.fd_lastfile = -1;
1781 	return (&newfdp->fd_fd);
1782 }
1783 
1784 static struct filedesc *
1785 fdhold(struct proc *p)
1786 {
1787 	struct filedesc *fdp;
1788 
1789 	mtx_lock(&fdesc_mtx);
1790 	fdp = p->p_fd;
1791 	if (fdp != NULL)
1792 		fdp->fd_holdcnt++;
1793 	mtx_unlock(&fdesc_mtx);
1794 	return (fdp);
1795 }
1796 
1797 static void
1798 fddrop(struct filedesc *fdp)
1799 {
1800 	struct filedesc0 *fdp0;
1801 	struct freetable *ft;
1802 	int i;
1803 
1804 	mtx_lock(&fdesc_mtx);
1805 	i = --fdp->fd_holdcnt;
1806 	mtx_unlock(&fdesc_mtx);
1807 	if (i > 0)
1808 		return;
1809 
1810 	FILEDESC_LOCK_DESTROY(fdp);
1811 	fdp0 = (struct filedesc0 *)fdp;
1812 	while ((ft = SLIST_FIRST(&fdp0->fd_free)) != NULL) {
1813 		SLIST_REMOVE_HEAD(&fdp0->fd_free, ft_next);
1814 		free(ft->ft_table, M_FILEDESC);
1815 	}
1816 	free(fdp, M_FILEDESC);
1817 }
1818 
1819 /*
1820  * Share a filedesc structure.
1821  */
1822 struct filedesc *
1823 fdshare(struct filedesc *fdp)
1824 {
1825 
1826 	FILEDESC_XLOCK(fdp);
1827 	fdp->fd_refcnt++;
1828 	FILEDESC_XUNLOCK(fdp);
1829 	return (fdp);
1830 }
1831 
1832 /*
1833  * Unshare a filedesc structure, if necessary by making a copy
1834  */
1835 void
1836 fdunshare(struct proc *p, struct thread *td)
1837 {
1838 
1839 	FILEDESC_XLOCK(p->p_fd);
1840 	if (p->p_fd->fd_refcnt > 1) {
1841 		struct filedesc *tmp;
1842 
1843 		FILEDESC_XUNLOCK(p->p_fd);
1844 		tmp = fdcopy(p->p_fd);
1845 		fdescfree(td);
1846 		p->p_fd = tmp;
1847 	} else
1848 		FILEDESC_XUNLOCK(p->p_fd);
1849 }
1850 
1851 /*
1852  * Copy a filedesc structure.  A NULL pointer in returns a NULL reference,
1853  * this is to ease callers, not catch errors.
1854  */
1855 struct filedesc *
1856 fdcopy(struct filedesc *fdp)
1857 {
1858 	struct filedesc *newfdp;
1859 	struct filedescent *nfde, *ofde;
1860 	int i;
1861 
1862 	/* Certain daemons might not have file descriptors. */
1863 	if (fdp == NULL)
1864 		return (NULL);
1865 
1866 	newfdp = fdinit(fdp);
1867 	FILEDESC_SLOCK(fdp);
1868 	while (fdp->fd_lastfile >= newfdp->fd_nfiles) {
1869 		FILEDESC_SUNLOCK(fdp);
1870 		FILEDESC_XLOCK(newfdp);
1871 		fdgrowtable(newfdp, fdp->fd_lastfile + 1);
1872 		FILEDESC_XUNLOCK(newfdp);
1873 		FILEDESC_SLOCK(fdp);
1874 	}
1875 	/* copy all passable descriptors (i.e. not kqueue) */
1876 	newfdp->fd_freefile = -1;
1877 	for (i = 0; i <= fdp->fd_lastfile; ++i) {
1878 		ofde = &fdp->fd_ofiles[i];
1879 		if (fdisused(fdp, i) &&
1880 		    (ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) &&
1881 		    ofde->fde_file->f_ops != &badfileops) {
1882 			nfde = &newfdp->fd_ofiles[i];
1883 			*nfde = *ofde;
1884 			filecaps_copy(&ofde->fde_caps, &nfde->fde_caps);
1885 			fhold(nfde->fde_file);
1886 			newfdp->fd_lastfile = i;
1887 		} else {
1888 			if (newfdp->fd_freefile == -1)
1889 				newfdp->fd_freefile = i;
1890 		}
1891 	}
1892 	newfdp->fd_cmask = fdp->fd_cmask;
1893 	FILEDESC_SUNLOCK(fdp);
1894 	FILEDESC_XLOCK(newfdp);
1895 	for (i = 0; i <= newfdp->fd_lastfile; ++i) {
1896 		if (newfdp->fd_ofiles[i].fde_file != NULL)
1897 			fdused(newfdp, i);
1898 	}
1899 	if (newfdp->fd_freefile == -1)
1900 		newfdp->fd_freefile = i;
1901 	FILEDESC_XUNLOCK(newfdp);
1902 	return (newfdp);
1903 }
1904 
1905 /*
1906  * Release a filedesc structure.
1907  */
1908 void
1909 fdescfree(struct thread *td)
1910 {
1911 	struct filedesc *fdp;
1912 	int i;
1913 	struct filedesc_to_leader *fdtol;
1914 	struct file *fp;
1915 	struct vnode *cdir, *jdir, *rdir, *vp;
1916 	struct flock lf;
1917 
1918 	/* Certain daemons might not have file descriptors. */
1919 	fdp = td->td_proc->p_fd;
1920 	if (fdp == NULL)
1921 		return;
1922 
1923 #ifdef RACCT
1924 	PROC_LOCK(td->td_proc);
1925 	racct_set(td->td_proc, RACCT_NOFILE, 0);
1926 	PROC_UNLOCK(td->td_proc);
1927 #endif
1928 
1929 	/* Check for special need to clear POSIX style locks */
1930 	fdtol = td->td_proc->p_fdtol;
1931 	if (fdtol != NULL) {
1932 		FILEDESC_XLOCK(fdp);
1933 		KASSERT(fdtol->fdl_refcount > 0,
1934 		    ("filedesc_to_refcount botch: fdl_refcount=%d",
1935 		    fdtol->fdl_refcount));
1936 		if (fdtol->fdl_refcount == 1 &&
1937 		    (td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
1938 			for (i = 0; i <= fdp->fd_lastfile; i++) {
1939 				fp = fdp->fd_ofiles[i].fde_file;
1940 				if (fp == NULL || fp->f_type != DTYPE_VNODE)
1941 					continue;
1942 				fhold(fp);
1943 				FILEDESC_XUNLOCK(fdp);
1944 				lf.l_whence = SEEK_SET;
1945 				lf.l_start = 0;
1946 				lf.l_len = 0;
1947 				lf.l_type = F_UNLCK;
1948 				vp = fp->f_vnode;
1949 				(void) VOP_ADVLOCK(vp,
1950 				    (caddr_t)td->td_proc->p_leader, F_UNLCK,
1951 				    &lf, F_POSIX);
1952 				FILEDESC_XLOCK(fdp);
1953 				fdrop(fp, td);
1954 			}
1955 		}
1956 	retry:
1957 		if (fdtol->fdl_refcount == 1) {
1958 			if (fdp->fd_holdleaderscount > 0 &&
1959 			    (td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
1960 				/*
1961 				 * close() or do_dup() has cleared a reference
1962 				 * in a shared file descriptor table.
1963 				 */
1964 				fdp->fd_holdleaderswakeup = 1;
1965 				sx_sleep(&fdp->fd_holdleaderscount,
1966 				    FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0);
1967 				goto retry;
1968 			}
1969 			if (fdtol->fdl_holdcount > 0) {
1970 				/*
1971 				 * Ensure that fdtol->fdl_leader remains
1972 				 * valid in closef().
1973 				 */
1974 				fdtol->fdl_wakeup = 1;
1975 				sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK,
1976 				    "fdlhold", 0);
1977 				goto retry;
1978 			}
1979 		}
1980 		fdtol->fdl_refcount--;
1981 		if (fdtol->fdl_refcount == 0 &&
1982 		    fdtol->fdl_holdcount == 0) {
1983 			fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
1984 			fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
1985 		} else
1986 			fdtol = NULL;
1987 		td->td_proc->p_fdtol = NULL;
1988 		FILEDESC_XUNLOCK(fdp);
1989 		if (fdtol != NULL)
1990 			free(fdtol, M_FILEDESC_TO_LEADER);
1991 	}
1992 	FILEDESC_XLOCK(fdp);
1993 	i = --fdp->fd_refcnt;
1994 	FILEDESC_XUNLOCK(fdp);
1995 	if (i > 0)
1996 		return;
1997 
1998 	for (i = 0; i <= fdp->fd_lastfile; i++) {
1999 		fp = fdp->fd_ofiles[i].fde_file;
2000 		if (fp != NULL) {
2001 			FILEDESC_XLOCK(fdp);
2002 			fdfree(fdp, i);
2003 			FILEDESC_XUNLOCK(fdp);
2004 			(void) closef(fp, td);
2005 		}
2006 	}
2007 	FILEDESC_XLOCK(fdp);
2008 
2009 	/* XXX This should happen earlier. */
2010 	mtx_lock(&fdesc_mtx);
2011 	td->td_proc->p_fd = NULL;
2012 	mtx_unlock(&fdesc_mtx);
2013 
2014 	if (fdp->fd_nfiles > NDFILE)
2015 		free(fdp->fd_ofiles, M_FILEDESC);
2016 	if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE))
2017 		free(fdp->fd_map, M_FILEDESC);
2018 
2019 	fdp->fd_nfiles = 0;
2020 
2021 	cdir = fdp->fd_cdir;
2022 	fdp->fd_cdir = NULL;
2023 	rdir = fdp->fd_rdir;
2024 	fdp->fd_rdir = NULL;
2025 	jdir = fdp->fd_jdir;
2026 	fdp->fd_jdir = NULL;
2027 	FILEDESC_XUNLOCK(fdp);
2028 
2029 	if (cdir != NULL)
2030 		vrele(cdir);
2031 	if (rdir != NULL)
2032 		vrele(rdir);
2033 	if (jdir != NULL)
2034 		vrele(jdir);
2035 
2036 	fddrop(fdp);
2037 }
2038 
2039 /*
2040  * For setugid programs, we don't want to people to use that setugidness
2041  * to generate error messages which write to a file which otherwise would
2042  * otherwise be off-limits to the process.  We check for filesystems where
2043  * the vnode can change out from under us after execve (like [lin]procfs).
2044  *
2045  * Since setugidsafety calls this only for fd 0, 1 and 2, this check is
2046  * sufficient.  We also don't check for setugidness since we know we are.
2047  */
2048 static int
2049 is_unsafe(struct file *fp)
2050 {
2051 	if (fp->f_type == DTYPE_VNODE) {
2052 		struct vnode *vp = fp->f_vnode;
2053 
2054 		if ((vp->v_vflag & VV_PROCDEP) != 0)
2055 			return (1);
2056 	}
2057 	return (0);
2058 }
2059 
2060 /*
2061  * Make this setguid thing safe, if at all possible.
2062  */
2063 void
2064 setugidsafety(struct thread *td)
2065 {
2066 	struct filedesc *fdp;
2067 	struct file *fp;
2068 	int i;
2069 
2070 	/* Certain daemons might not have file descriptors. */
2071 	fdp = td->td_proc->p_fd;
2072 	if (fdp == NULL)
2073 		return;
2074 
2075 	/*
2076 	 * Note: fdp->fd_ofiles may be reallocated out from under us while
2077 	 * we are blocked in a close.  Be careful!
2078 	 */
2079 	FILEDESC_XLOCK(fdp);
2080 	for (i = 0; i <= fdp->fd_lastfile; i++) {
2081 		if (i > 2)
2082 			break;
2083 		fp = fdp->fd_ofiles[i].fde_file;
2084 		if (fp != NULL && is_unsafe(fp)) {
2085 			knote_fdclose(td, i);
2086 			/*
2087 			 * NULL-out descriptor prior to close to avoid
2088 			 * a race while close blocks.
2089 			 */
2090 			fdfree(fdp, i);
2091 			FILEDESC_XUNLOCK(fdp);
2092 			(void) closef(fp, td);
2093 			FILEDESC_XLOCK(fdp);
2094 		}
2095 	}
2096 	FILEDESC_XUNLOCK(fdp);
2097 }
2098 
2099 /*
2100  * If a specific file object occupies a specific file descriptor, close the
2101  * file descriptor entry and drop a reference on the file object.  This is a
2102  * convenience function to handle a subsequent error in a function that calls
2103  * falloc() that handles the race that another thread might have closed the
2104  * file descriptor out from under the thread creating the file object.
2105  */
2106 void
2107 fdclose(struct filedesc *fdp, struct file *fp, int idx, struct thread *td)
2108 {
2109 
2110 	FILEDESC_XLOCK(fdp);
2111 	if (fdp->fd_ofiles[idx].fde_file == fp) {
2112 		fdfree(fdp, idx);
2113 		FILEDESC_XUNLOCK(fdp);
2114 		fdrop(fp, td);
2115 	} else
2116 		FILEDESC_XUNLOCK(fdp);
2117 }
2118 
2119 /*
2120  * Close any files on exec?
2121  */
2122 void
2123 fdcloseexec(struct thread *td)
2124 {
2125 	struct filedesc *fdp;
2126 	struct filedescent *fde;
2127 	struct file *fp;
2128 	int i;
2129 
2130 	/* Certain daemons might not have file descriptors. */
2131 	fdp = td->td_proc->p_fd;
2132 	if (fdp == NULL)
2133 		return;
2134 
2135 	/*
2136 	 * We cannot cache fd_ofiles since operations
2137 	 * may block and rip them out from under us.
2138 	 */
2139 	FILEDESC_XLOCK(fdp);
2140 	for (i = 0; i <= fdp->fd_lastfile; i++) {
2141 		fde = &fdp->fd_ofiles[i];
2142 		fp = fde->fde_file;
2143 		if (fp != NULL && (fp->f_type == DTYPE_MQUEUE ||
2144 		    (fde->fde_flags & UF_EXCLOSE))) {
2145 			fdfree(fdp, i);
2146 			(void) closefp(fdp, i, fp, td, 0);
2147 			/* closefp() drops the FILEDESC lock. */
2148 			FILEDESC_XLOCK(fdp);
2149 		}
2150 	}
2151 	FILEDESC_XUNLOCK(fdp);
2152 }
2153 
2154 /*
2155  * It is unsafe for set[ug]id processes to be started with file
2156  * descriptors 0..2 closed, as these descriptors are given implicit
2157  * significance in the Standard C library.  fdcheckstd() will create a
2158  * descriptor referencing /dev/null for each of stdin, stdout, and
2159  * stderr that is not already open.
2160  */
2161 int
2162 fdcheckstd(struct thread *td)
2163 {
2164 	struct filedesc *fdp;
2165 	register_t retval, save;
2166 	int i, error, devnull;
2167 
2168 	fdp = td->td_proc->p_fd;
2169 	if (fdp == NULL)
2170 		return (0);
2171 	KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2172 	devnull = -1;
2173 	error = 0;
2174 	for (i = 0; i < 3; i++) {
2175 		if (fdp->fd_ofiles[i].fde_file != NULL)
2176 			continue;
2177 		if (devnull < 0) {
2178 			save = td->td_retval[0];
2179 			error = kern_open(td, "/dev/null", UIO_SYSSPACE,
2180 			    O_RDWR, 0);
2181 			devnull = td->td_retval[0];
2182 			td->td_retval[0] = save;
2183 			if (error)
2184 				break;
2185 			KASSERT(devnull == i, ("oof, we didn't get our fd"));
2186 		} else {
2187 			error = do_dup(td, DUP_FIXED, devnull, i, &retval);
2188 			if (error != 0)
2189 				break;
2190 		}
2191 	}
2192 	return (error);
2193 }
2194 
2195 /*
2196  * Internal form of close.  Decrement reference count on file structure.
2197  * Note: td may be NULL when closing a file that was being passed in a
2198  * message.
2199  *
2200  * XXXRW: Giant is not required for the caller, but often will be held; this
2201  * makes it moderately likely the Giant will be recursed in the VFS case.
2202  */
2203 int
2204 closef(struct file *fp, struct thread *td)
2205 {
2206 	struct vnode *vp;
2207 	struct flock lf;
2208 	struct filedesc_to_leader *fdtol;
2209 	struct filedesc *fdp;
2210 
2211 	/*
2212 	 * POSIX record locking dictates that any close releases ALL
2213 	 * locks owned by this process.  This is handled by setting
2214 	 * a flag in the unlock to free ONLY locks obeying POSIX
2215 	 * semantics, and not to free BSD-style file locks.
2216 	 * If the descriptor was in a message, POSIX-style locks
2217 	 * aren't passed with the descriptor, and the thread pointer
2218 	 * will be NULL.  Callers should be careful only to pass a
2219 	 * NULL thread pointer when there really is no owning
2220 	 * context that might have locks, or the locks will be
2221 	 * leaked.
2222 	 */
2223 	if (fp->f_type == DTYPE_VNODE && td != NULL) {
2224 		vp = fp->f_vnode;
2225 		if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
2226 			lf.l_whence = SEEK_SET;
2227 			lf.l_start = 0;
2228 			lf.l_len = 0;
2229 			lf.l_type = F_UNLCK;
2230 			(void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader,
2231 			    F_UNLCK, &lf, F_POSIX);
2232 		}
2233 		fdtol = td->td_proc->p_fdtol;
2234 		if (fdtol != NULL) {
2235 			/*
2236 			 * Handle special case where file descriptor table is
2237 			 * shared between multiple process leaders.
2238 			 */
2239 			fdp = td->td_proc->p_fd;
2240 			FILEDESC_XLOCK(fdp);
2241 			for (fdtol = fdtol->fdl_next;
2242 			     fdtol != td->td_proc->p_fdtol;
2243 			     fdtol = fdtol->fdl_next) {
2244 				if ((fdtol->fdl_leader->p_flag &
2245 				     P_ADVLOCK) == 0)
2246 					continue;
2247 				fdtol->fdl_holdcount++;
2248 				FILEDESC_XUNLOCK(fdp);
2249 				lf.l_whence = SEEK_SET;
2250 				lf.l_start = 0;
2251 				lf.l_len = 0;
2252 				lf.l_type = F_UNLCK;
2253 				vp = fp->f_vnode;
2254 				(void) VOP_ADVLOCK(vp,
2255 				    (caddr_t)fdtol->fdl_leader, F_UNLCK, &lf,
2256 				    F_POSIX);
2257 				FILEDESC_XLOCK(fdp);
2258 				fdtol->fdl_holdcount--;
2259 				if (fdtol->fdl_holdcount == 0 &&
2260 				    fdtol->fdl_wakeup != 0) {
2261 					fdtol->fdl_wakeup = 0;
2262 					wakeup(fdtol);
2263 				}
2264 			}
2265 			FILEDESC_XUNLOCK(fdp);
2266 		}
2267 	}
2268 	return (fdrop(fp, td));
2269 }
2270 
2271 /*
2272  * Initialize the file pointer with the specified properties.
2273  *
2274  * The ops are set with release semantics to be certain that the flags, type,
2275  * and data are visible when ops is.  This is to prevent ops methods from being
2276  * called with bad data.
2277  */
2278 void
2279 finit(struct file *fp, u_int flag, short type, void *data, struct fileops *ops)
2280 {
2281 	fp->f_data = data;
2282 	fp->f_flag = flag;
2283 	fp->f_type = type;
2284 	atomic_store_rel_ptr((volatile uintptr_t *)&fp->f_ops, (uintptr_t)ops);
2285 }
2286 
2287 int
2288 fget_unlocked(struct filedesc *fdp, int fd, cap_rights_t needrights,
2289     int needfcntl, struct file **fpp, cap_rights_t *haverightsp)
2290 {
2291 	struct file *fp;
2292 	u_int count;
2293 #ifdef CAPABILITIES
2294 	cap_rights_t haverights;
2295 	int error;
2296 #endif
2297 
2298 	if (fd < 0 || fd >= fdp->fd_nfiles)
2299 		return (EBADF);
2300 	/*
2301 	 * Fetch the descriptor locklessly.  We avoid fdrop() races by
2302 	 * never raising a refcount above 0.  To accomplish this we have
2303 	 * to use a cmpset loop rather than an atomic_add.  The descriptor
2304 	 * must be re-verified once we acquire a reference to be certain
2305 	 * that the identity is still correct and we did not lose a race
2306 	 * due to preemption.
2307 	 */
2308 	for (;;) {
2309 		fp = fdp->fd_ofiles[fd].fde_file;
2310 		if (fp == NULL)
2311 			return (EBADF);
2312 #ifdef CAPABILITIES
2313 		haverights = cap_rights(fdp, fd);
2314 		error = cap_check(haverights, needrights);
2315 		if (error != 0)
2316 			return (error);
2317 		if ((needrights & CAP_FCNTL) != 0) {
2318 			error = cap_fcntl_check(fdp, fd, needfcntl);
2319 			if (error != 0)
2320 				return (error);
2321 		}
2322 #endif
2323 		count = fp->f_count;
2324 		if (count == 0)
2325 			continue;
2326 		/*
2327 		 * Use an acquire barrier to prevent caching of fd_ofiles
2328 		 * so it is refreshed for verification.
2329 		 */
2330 		if (atomic_cmpset_acq_int(&fp->f_count, count, count + 1) != 1)
2331 			continue;
2332 		if (fp == fdp->fd_ofiles[fd].fde_file)
2333 			break;
2334 		fdrop(fp, curthread);
2335 	}
2336 	*fpp = fp;
2337 	if (haverightsp != NULL) {
2338 #ifdef CAPABILITIES
2339 		*haverightsp = haverights;
2340 #else
2341 		*haverightsp = CAP_ALL;
2342 #endif
2343 	}
2344 	return (0);
2345 }
2346 
2347 /*
2348  * Extract the file pointer associated with the specified descriptor for the
2349  * current user process.
2350  *
2351  * If the descriptor doesn't exist or doesn't match 'flags', EBADF is
2352  * returned.
2353  *
2354  * File's rights will be checked against the capability rights mask.
2355  *
2356  * If an error occured the non-zero error is returned and *fpp is set to
2357  * NULL.  Otherwise *fpp is held and set and zero is returned.  Caller is
2358  * responsible for fdrop().
2359  */
2360 static __inline int
2361 _fget(struct thread *td, int fd, struct file **fpp, int flags,
2362     cap_rights_t needrights, u_char *maxprotp)
2363 {
2364 	struct filedesc *fdp;
2365 	struct file *fp;
2366 	cap_rights_t haverights;
2367 	int error;
2368 
2369 	*fpp = NULL;
2370 	if (td == NULL || (fdp = td->td_proc->p_fd) == NULL)
2371 		return (EBADF);
2372 	if (maxprotp != NULL)
2373 		needrights |= CAP_MMAP;
2374 	error = fget_unlocked(fdp, fd, needrights, 0, &fp, &haverights);
2375 	if (error != 0)
2376 		return (error);
2377 	if (fp->f_ops == &badfileops) {
2378 		fdrop(fp, td);
2379 		return (EBADF);
2380 	}
2381 
2382 #ifdef CAPABILITIES
2383 	/*
2384 	 * If requested, convert capability rights to access flags.
2385 	 */
2386 	if (maxprotp != NULL)
2387 		*maxprotp = cap_rights_to_vmprot(haverights);
2388 #else /* !CAPABILITIES */
2389 	if (maxprotp != NULL)
2390 		*maxprotp = VM_PROT_ALL;
2391 #endif /* CAPABILITIES */
2392 
2393 	/*
2394 	 * FREAD and FWRITE failure return EBADF as per POSIX.
2395 	 */
2396 	error = 0;
2397 	switch (flags) {
2398 	case FREAD:
2399 	case FWRITE:
2400 		if ((fp->f_flag & flags) == 0)
2401 			error = EBADF;
2402 		break;
2403 	case FEXEC:
2404 	    	if ((fp->f_flag & (FREAD | FEXEC)) == 0 ||
2405 		    ((fp->f_flag & FWRITE) != 0))
2406 			error = EBADF;
2407 		break;
2408 	case 0:
2409 		break;
2410 	default:
2411 		KASSERT(0, ("wrong flags"));
2412 	}
2413 
2414 	if (error != 0) {
2415 		fdrop(fp, td);
2416 		return (error);
2417 	}
2418 
2419 	*fpp = fp;
2420 	return (0);
2421 }
2422 
2423 int
2424 fget(struct thread *td, int fd, cap_rights_t rights, struct file **fpp)
2425 {
2426 
2427 	return(_fget(td, fd, fpp, 0, rights, NULL));
2428 }
2429 
2430 int
2431 fget_mmap(struct thread *td, int fd, cap_rights_t rights, u_char *maxprotp,
2432     struct file **fpp)
2433 {
2434 
2435 	return (_fget(td, fd, fpp, 0, rights, maxprotp));
2436 }
2437 
2438 int
2439 fget_read(struct thread *td, int fd, cap_rights_t rights, struct file **fpp)
2440 {
2441 
2442 	return(_fget(td, fd, fpp, FREAD, rights, NULL));
2443 }
2444 
2445 int
2446 fget_write(struct thread *td, int fd, cap_rights_t rights, struct file **fpp)
2447 {
2448 
2449 	return (_fget(td, fd, fpp, FWRITE, rights, NULL));
2450 }
2451 
2452 /*
2453  * Like fget() but loads the underlying vnode, or returns an error if the
2454  * descriptor does not represent a vnode.  Note that pipes use vnodes but
2455  * never have VM objects.  The returned vnode will be vref()'d.
2456  *
2457  * XXX: what about the unused flags ?
2458  */
2459 static __inline int
2460 _fgetvp(struct thread *td, int fd, int flags, cap_rights_t needrights,
2461     struct vnode **vpp)
2462 {
2463 	struct file *fp;
2464 	int error;
2465 
2466 	*vpp = NULL;
2467 	error = _fget(td, fd, &fp, flags, needrights, NULL);
2468 	if (error)
2469 		return (error);
2470 	if (fp->f_vnode == NULL) {
2471 		error = EINVAL;
2472 	} else {
2473 		*vpp = fp->f_vnode;
2474 		vref(*vpp);
2475 	}
2476 	fdrop(fp, td);
2477 
2478 	return (error);
2479 }
2480 
2481 int
2482 fgetvp(struct thread *td, int fd, cap_rights_t rights, struct vnode **vpp)
2483 {
2484 
2485 	return (_fgetvp(td, fd, 0, rights, vpp));
2486 }
2487 
2488 int
2489 fgetvp_rights(struct thread *td, int fd, cap_rights_t need,
2490     struct filecaps *havecaps, struct vnode **vpp)
2491 {
2492 	struct filedesc *fdp;
2493 	struct file *fp;
2494 #ifdef CAPABILITIES
2495 	int error;
2496 #endif
2497 
2498 	if (td == NULL || (fdp = td->td_proc->p_fd) == NULL)
2499 		return (EBADF);
2500 
2501 	fp = fget_locked(fdp, fd);
2502 	if (fp == NULL || fp->f_ops == &badfileops)
2503 		return (EBADF);
2504 
2505 #ifdef CAPABILITIES
2506 	error = cap_check(cap_rights(fdp, fd), need);
2507 	if (error != 0)
2508 		return (error);
2509 #endif
2510 
2511 	if (fp->f_vnode == NULL)
2512 		return (EINVAL);
2513 
2514 	*vpp = fp->f_vnode;
2515 	vref(*vpp);
2516 	filecaps_copy(&fdp->fd_ofiles[fd].fde_caps, havecaps);
2517 
2518 	return (0);
2519 }
2520 
2521 int
2522 fgetvp_read(struct thread *td, int fd, cap_rights_t rights, struct vnode **vpp)
2523 {
2524 
2525 	return (_fgetvp(td, fd, FREAD, rights, vpp));
2526 }
2527 
2528 int
2529 fgetvp_exec(struct thread *td, int fd, cap_rights_t rights, struct vnode **vpp)
2530 {
2531 
2532 	return (_fgetvp(td, fd, FEXEC, rights, vpp));
2533 }
2534 
2535 #ifdef notyet
2536 int
2537 fgetvp_write(struct thread *td, int fd, cap_rights_t rights,
2538     struct vnode **vpp)
2539 {
2540 
2541 	return (_fgetvp(td, fd, FWRITE, rights, vpp));
2542 }
2543 #endif
2544 
2545 /*
2546  * Like fget() but loads the underlying socket, or returns an error if the
2547  * descriptor does not represent a socket.
2548  *
2549  * We bump the ref count on the returned socket.  XXX Also obtain the SX lock
2550  * in the future.
2551  *
2552  * Note: fgetsock() and fputsock() are deprecated, as consumers should rely
2553  * on their file descriptor reference to prevent the socket from being free'd
2554  * during use.
2555  */
2556 int
2557 fgetsock(struct thread *td, int fd, cap_rights_t rights, struct socket **spp,
2558     u_int *fflagp)
2559 {
2560 	struct file *fp;
2561 	int error;
2562 
2563 	*spp = NULL;
2564 	if (fflagp != NULL)
2565 		*fflagp = 0;
2566 	if ((error = _fget(td, fd, &fp, 0, rights, NULL)) != 0)
2567 		return (error);
2568 	if (fp->f_type != DTYPE_SOCKET) {
2569 		error = ENOTSOCK;
2570 	} else {
2571 		*spp = fp->f_data;
2572 		if (fflagp)
2573 			*fflagp = fp->f_flag;
2574 		SOCK_LOCK(*spp);
2575 		soref(*spp);
2576 		SOCK_UNLOCK(*spp);
2577 	}
2578 	fdrop(fp, td);
2579 
2580 	return (error);
2581 }
2582 
2583 /*
2584  * Drop the reference count on the socket and XXX release the SX lock in the
2585  * future.  The last reference closes the socket.
2586  *
2587  * Note: fputsock() is deprecated, see comment for fgetsock().
2588  */
2589 void
2590 fputsock(struct socket *so)
2591 {
2592 
2593 	ACCEPT_LOCK();
2594 	SOCK_LOCK(so);
2595 	CURVNET_SET(so->so_vnet);
2596 	sorele(so);
2597 	CURVNET_RESTORE();
2598 }
2599 
2600 /*
2601  * Handle the last reference to a file being closed.
2602  */
2603 int
2604 _fdrop(struct file *fp, struct thread *td)
2605 {
2606 	int error;
2607 
2608 	error = 0;
2609 	if (fp->f_count != 0)
2610 		panic("fdrop: count %d", fp->f_count);
2611 	if (fp->f_ops != &badfileops)
2612 		error = fo_close(fp, td);
2613 	atomic_subtract_int(&openfiles, 1);
2614 	crfree(fp->f_cred);
2615 	free(fp->f_advice, M_FADVISE);
2616 	uma_zfree(file_zone, fp);
2617 
2618 	return (error);
2619 }
2620 
2621 /*
2622  * Apply an advisory lock on a file descriptor.
2623  *
2624  * Just attempt to get a record lock of the requested type on the entire file
2625  * (l_whence = SEEK_SET, l_start = 0, l_len = 0).
2626  */
2627 #ifndef _SYS_SYSPROTO_H_
2628 struct flock_args {
2629 	int	fd;
2630 	int	how;
2631 };
2632 #endif
2633 /* ARGSUSED */
2634 int
2635 sys_flock(struct thread *td, struct flock_args *uap)
2636 {
2637 	struct file *fp;
2638 	struct vnode *vp;
2639 	struct flock lf;
2640 	int error;
2641 
2642 	if ((error = fget(td, uap->fd, CAP_FLOCK, &fp)) != 0)
2643 		return (error);
2644 	if (fp->f_type != DTYPE_VNODE) {
2645 		fdrop(fp, td);
2646 		return (EOPNOTSUPP);
2647 	}
2648 
2649 	vp = fp->f_vnode;
2650 	lf.l_whence = SEEK_SET;
2651 	lf.l_start = 0;
2652 	lf.l_len = 0;
2653 	if (uap->how & LOCK_UN) {
2654 		lf.l_type = F_UNLCK;
2655 		atomic_clear_int(&fp->f_flag, FHASLOCK);
2656 		error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK);
2657 		goto done2;
2658 	}
2659 	if (uap->how & LOCK_EX)
2660 		lf.l_type = F_WRLCK;
2661 	else if (uap->how & LOCK_SH)
2662 		lf.l_type = F_RDLCK;
2663 	else {
2664 		error = EBADF;
2665 		goto done2;
2666 	}
2667 	atomic_set_int(&fp->f_flag, FHASLOCK);
2668 	error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf,
2669 	    (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT);
2670 done2:
2671 	fdrop(fp, td);
2672 	return (error);
2673 }
2674 /*
2675  * Duplicate the specified descriptor to a free descriptor.
2676  */
2677 int
2678 dupfdopen(struct thread *td, struct filedesc *fdp, int dfd, int mode,
2679     int openerror, int *indxp)
2680 {
2681 	struct file *fp;
2682 	int error, indx;
2683 
2684 	KASSERT(openerror == ENODEV || openerror == ENXIO,
2685 	    ("unexpected error %d in %s", openerror, __func__));
2686 
2687 	/*
2688 	 * If the to-be-dup'd fd number is greater than the allowed number
2689 	 * of file descriptors, or the fd to be dup'd has already been
2690 	 * closed, then reject.
2691 	 */
2692 	FILEDESC_XLOCK(fdp);
2693 	if ((fp = fget_locked(fdp, dfd)) == NULL) {
2694 		FILEDESC_XUNLOCK(fdp);
2695 		return (EBADF);
2696 	}
2697 
2698 	error = fdalloc(td, 0, &indx);
2699 	if (error != 0) {
2700 		FILEDESC_XUNLOCK(fdp);
2701 		return (error);
2702 	}
2703 
2704 	/*
2705 	 * There are two cases of interest here.
2706 	 *
2707 	 * For ENODEV simply dup (dfd) to file descriptor (indx) and return.
2708 	 *
2709 	 * For ENXIO steal away the file structure from (dfd) and store it in
2710 	 * (indx).  (dfd) is effectively closed by this operation.
2711 	 */
2712 	switch (openerror) {
2713 	case ENODEV:
2714 		/*
2715 		 * Check that the mode the file is being opened for is a
2716 		 * subset of the mode of the existing descriptor.
2717 		 */
2718 		if (((mode & (FREAD|FWRITE)) | fp->f_flag) != fp->f_flag) {
2719 			fdunused(fdp, indx);
2720 			FILEDESC_XUNLOCK(fdp);
2721 			return (EACCES);
2722 		}
2723 		fhold(fp);
2724 		fdp->fd_ofiles[indx] = fdp->fd_ofiles[dfd];
2725 		filecaps_copy(&fdp->fd_ofiles[dfd].fde_caps,
2726 		    &fdp->fd_ofiles[indx].fde_caps);
2727 		break;
2728 	case ENXIO:
2729 		/*
2730 		 * Steal away the file pointer from dfd and stuff it into indx.
2731 		 */
2732 		fdp->fd_ofiles[indx] = fdp->fd_ofiles[dfd];
2733 		bzero(&fdp->fd_ofiles[dfd], sizeof(fdp->fd_ofiles[dfd]));
2734 		fdunused(fdp, dfd);
2735 		break;
2736 	}
2737 	FILEDESC_XUNLOCK(fdp);
2738 	*indxp = indx;
2739 	return (0);
2740 }
2741 
2742 /*
2743  * Scan all active processes and prisons to see if any of them have a current
2744  * or root directory of `olddp'. If so, replace them with the new mount point.
2745  */
2746 void
2747 mountcheckdirs(struct vnode *olddp, struct vnode *newdp)
2748 {
2749 	struct filedesc *fdp;
2750 	struct prison *pr;
2751 	struct proc *p;
2752 	int nrele;
2753 
2754 	if (vrefcnt(olddp) == 1)
2755 		return;
2756 	nrele = 0;
2757 	sx_slock(&allproc_lock);
2758 	FOREACH_PROC_IN_SYSTEM(p) {
2759 		fdp = fdhold(p);
2760 		if (fdp == NULL)
2761 			continue;
2762 		FILEDESC_XLOCK(fdp);
2763 		if (fdp->fd_cdir == olddp) {
2764 			vref(newdp);
2765 			fdp->fd_cdir = newdp;
2766 			nrele++;
2767 		}
2768 		if (fdp->fd_rdir == olddp) {
2769 			vref(newdp);
2770 			fdp->fd_rdir = newdp;
2771 			nrele++;
2772 		}
2773 		if (fdp->fd_jdir == olddp) {
2774 			vref(newdp);
2775 			fdp->fd_jdir = newdp;
2776 			nrele++;
2777 		}
2778 		FILEDESC_XUNLOCK(fdp);
2779 		fddrop(fdp);
2780 	}
2781 	sx_sunlock(&allproc_lock);
2782 	if (rootvnode == olddp) {
2783 		vref(newdp);
2784 		rootvnode = newdp;
2785 		nrele++;
2786 	}
2787 	mtx_lock(&prison0.pr_mtx);
2788 	if (prison0.pr_root == olddp) {
2789 		vref(newdp);
2790 		prison0.pr_root = newdp;
2791 		nrele++;
2792 	}
2793 	mtx_unlock(&prison0.pr_mtx);
2794 	sx_slock(&allprison_lock);
2795 	TAILQ_FOREACH(pr, &allprison, pr_list) {
2796 		mtx_lock(&pr->pr_mtx);
2797 		if (pr->pr_root == olddp) {
2798 			vref(newdp);
2799 			pr->pr_root = newdp;
2800 			nrele++;
2801 		}
2802 		mtx_unlock(&pr->pr_mtx);
2803 	}
2804 	sx_sunlock(&allprison_lock);
2805 	while (nrele--)
2806 		vrele(olddp);
2807 }
2808 
2809 struct filedesc_to_leader *
2810 filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader)
2811 {
2812 	struct filedesc_to_leader *fdtol;
2813 
2814 	fdtol = malloc(sizeof(struct filedesc_to_leader),
2815 	       M_FILEDESC_TO_LEADER,
2816 	       M_WAITOK);
2817 	fdtol->fdl_refcount = 1;
2818 	fdtol->fdl_holdcount = 0;
2819 	fdtol->fdl_wakeup = 0;
2820 	fdtol->fdl_leader = leader;
2821 	if (old != NULL) {
2822 		FILEDESC_XLOCK(fdp);
2823 		fdtol->fdl_next = old->fdl_next;
2824 		fdtol->fdl_prev = old;
2825 		old->fdl_next = fdtol;
2826 		fdtol->fdl_next->fdl_prev = fdtol;
2827 		FILEDESC_XUNLOCK(fdp);
2828 	} else {
2829 		fdtol->fdl_next = fdtol;
2830 		fdtol->fdl_prev = fdtol;
2831 	}
2832 	return (fdtol);
2833 }
2834 
2835 /*
2836  * Get file structures globally.
2837  */
2838 static int
2839 sysctl_kern_file(SYSCTL_HANDLER_ARGS)
2840 {
2841 	struct xfile xf;
2842 	struct filedesc *fdp;
2843 	struct file *fp;
2844 	struct proc *p;
2845 	int error, n;
2846 
2847 	error = sysctl_wire_old_buffer(req, 0);
2848 	if (error != 0)
2849 		return (error);
2850 	if (req->oldptr == NULL) {
2851 		n = 0;
2852 		sx_slock(&allproc_lock);
2853 		FOREACH_PROC_IN_SYSTEM(p) {
2854 			if (p->p_state == PRS_NEW)
2855 				continue;
2856 			fdp = fdhold(p);
2857 			if (fdp == NULL)
2858 				continue;
2859 			/* overestimates sparse tables. */
2860 			if (fdp->fd_lastfile > 0)
2861 				n += fdp->fd_lastfile;
2862 			fddrop(fdp);
2863 		}
2864 		sx_sunlock(&allproc_lock);
2865 		return (SYSCTL_OUT(req, 0, n * sizeof(xf)));
2866 	}
2867 	error = 0;
2868 	bzero(&xf, sizeof(xf));
2869 	xf.xf_size = sizeof(xf);
2870 	sx_slock(&allproc_lock);
2871 	FOREACH_PROC_IN_SYSTEM(p) {
2872 		PROC_LOCK(p);
2873 		if (p->p_state == PRS_NEW) {
2874 			PROC_UNLOCK(p);
2875 			continue;
2876 		}
2877 		if (p_cansee(req->td, p) != 0) {
2878 			PROC_UNLOCK(p);
2879 			continue;
2880 		}
2881 		xf.xf_pid = p->p_pid;
2882 		xf.xf_uid = p->p_ucred->cr_uid;
2883 		PROC_UNLOCK(p);
2884 		fdp = fdhold(p);
2885 		if (fdp == NULL)
2886 			continue;
2887 		FILEDESC_SLOCK(fdp);
2888 		for (n = 0; fdp->fd_refcnt > 0 && n < fdp->fd_nfiles; ++n) {
2889 			if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
2890 				continue;
2891 			xf.xf_fd = n;
2892 			xf.xf_file = fp;
2893 			xf.xf_data = fp->f_data;
2894 			xf.xf_vnode = fp->f_vnode;
2895 			xf.xf_type = fp->f_type;
2896 			xf.xf_count = fp->f_count;
2897 			xf.xf_msgcount = 0;
2898 			xf.xf_offset = foffset_get(fp);
2899 			xf.xf_flag = fp->f_flag;
2900 			error = SYSCTL_OUT(req, &xf, sizeof(xf));
2901 			if (error)
2902 				break;
2903 		}
2904 		FILEDESC_SUNLOCK(fdp);
2905 		fddrop(fdp);
2906 		if (error)
2907 			break;
2908 	}
2909 	sx_sunlock(&allproc_lock);
2910 	return (error);
2911 }
2912 
2913 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD,
2914     0, 0, sysctl_kern_file, "S,xfile", "Entire file table");
2915 
2916 #ifdef KINFO_OFILE_SIZE
2917 CTASSERT(sizeof(struct kinfo_ofile) == KINFO_OFILE_SIZE);
2918 #endif
2919 
2920 #ifdef COMPAT_FREEBSD7
2921 static int
2922 export_vnode_for_osysctl(struct vnode *vp, int type,
2923     struct kinfo_ofile *kif, struct filedesc *fdp, struct sysctl_req *req)
2924 {
2925 	int error;
2926 	char *fullpath, *freepath;
2927 
2928 	bzero(kif, sizeof(*kif));
2929 	kif->kf_structsize = sizeof(*kif);
2930 
2931 	vref(vp);
2932 	kif->kf_fd = type;
2933 	kif->kf_type = KF_TYPE_VNODE;
2934 	/* This function only handles directories. */
2935 	if (vp->v_type != VDIR) {
2936 		vrele(vp);
2937 		return (ENOTDIR);
2938 	}
2939 	kif->kf_vnode_type = KF_VTYPE_VDIR;
2940 
2941 	/*
2942 	 * This is not a true file descriptor, so we set a bogus refcount
2943 	 * and offset to indicate these fields should be ignored.
2944 	 */
2945 	kif->kf_ref_count = -1;
2946 	kif->kf_offset = -1;
2947 
2948 	freepath = NULL;
2949 	fullpath = "-";
2950 	FILEDESC_SUNLOCK(fdp);
2951 	vn_fullpath(curthread, vp, &fullpath, &freepath);
2952 	vrele(vp);
2953 	strlcpy(kif->kf_path, fullpath, sizeof(kif->kf_path));
2954 	if (freepath != NULL)
2955 		free(freepath, M_TEMP);
2956 	error = SYSCTL_OUT(req, kif, sizeof(*kif));
2957 	FILEDESC_SLOCK(fdp);
2958 	return (error);
2959 }
2960 
2961 /*
2962  * Get per-process file descriptors for use by procstat(1), et al.
2963  */
2964 static int
2965 sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS)
2966 {
2967 	char *fullpath, *freepath;
2968 	struct kinfo_ofile *kif;
2969 	struct filedesc *fdp;
2970 	int error, i, *name;
2971 	struct shmfd *shmfd;
2972 	struct socket *so;
2973 	struct vnode *vp;
2974 	struct ksem *ks;
2975 	struct file *fp;
2976 	struct proc *p;
2977 	struct tty *tp;
2978 
2979 	name = (int *)arg1;
2980 	error = pget((pid_t)name[0], PGET_CANDEBUG, &p);
2981 	if (error != 0)
2982 		return (error);
2983 	fdp = fdhold(p);
2984 	PROC_UNLOCK(p);
2985 	if (fdp == NULL)
2986 		return (ENOENT);
2987 	kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK);
2988 	FILEDESC_SLOCK(fdp);
2989 	if (fdp->fd_cdir != NULL)
2990 		export_vnode_for_osysctl(fdp->fd_cdir, KF_FD_TYPE_CWD, kif,
2991 				fdp, req);
2992 	if (fdp->fd_rdir != NULL)
2993 		export_vnode_for_osysctl(fdp->fd_rdir, KF_FD_TYPE_ROOT, kif,
2994 				fdp, req);
2995 	if (fdp->fd_jdir != NULL)
2996 		export_vnode_for_osysctl(fdp->fd_jdir, KF_FD_TYPE_JAIL, kif,
2997 				fdp, req);
2998 	for (i = 0; i < fdp->fd_nfiles; i++) {
2999 		if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
3000 			continue;
3001 		bzero(kif, sizeof(*kif));
3002 		kif->kf_structsize = sizeof(*kif);
3003 		ks = NULL;
3004 		vp = NULL;
3005 		so = NULL;
3006 		tp = NULL;
3007 		shmfd = NULL;
3008 		kif->kf_fd = i;
3009 
3010 		switch (fp->f_type) {
3011 		case DTYPE_VNODE:
3012 			kif->kf_type = KF_TYPE_VNODE;
3013 			vp = fp->f_vnode;
3014 			break;
3015 
3016 		case DTYPE_SOCKET:
3017 			kif->kf_type = KF_TYPE_SOCKET;
3018 			so = fp->f_data;
3019 			break;
3020 
3021 		case DTYPE_PIPE:
3022 			kif->kf_type = KF_TYPE_PIPE;
3023 			break;
3024 
3025 		case DTYPE_FIFO:
3026 			kif->kf_type = KF_TYPE_FIFO;
3027 			vp = fp->f_vnode;
3028 			break;
3029 
3030 		case DTYPE_KQUEUE:
3031 			kif->kf_type = KF_TYPE_KQUEUE;
3032 			break;
3033 
3034 		case DTYPE_CRYPTO:
3035 			kif->kf_type = KF_TYPE_CRYPTO;
3036 			break;
3037 
3038 		case DTYPE_MQUEUE:
3039 			kif->kf_type = KF_TYPE_MQUEUE;
3040 			break;
3041 
3042 		case DTYPE_SHM:
3043 			kif->kf_type = KF_TYPE_SHM;
3044 			shmfd = fp->f_data;
3045 			break;
3046 
3047 		case DTYPE_SEM:
3048 			kif->kf_type = KF_TYPE_SEM;
3049 			ks = fp->f_data;
3050 			break;
3051 
3052 		case DTYPE_PTS:
3053 			kif->kf_type = KF_TYPE_PTS;
3054 			tp = fp->f_data;
3055 			break;
3056 
3057 #ifdef PROCDESC
3058 		case DTYPE_PROCDESC:
3059 			kif->kf_type = KF_TYPE_PROCDESC;
3060 			break;
3061 #endif
3062 
3063 		default:
3064 			kif->kf_type = KF_TYPE_UNKNOWN;
3065 			break;
3066 		}
3067 		kif->kf_ref_count = fp->f_count;
3068 		if (fp->f_flag & FREAD)
3069 			kif->kf_flags |= KF_FLAG_READ;
3070 		if (fp->f_flag & FWRITE)
3071 			kif->kf_flags |= KF_FLAG_WRITE;
3072 		if (fp->f_flag & FAPPEND)
3073 			kif->kf_flags |= KF_FLAG_APPEND;
3074 		if (fp->f_flag & FASYNC)
3075 			kif->kf_flags |= KF_FLAG_ASYNC;
3076 		if (fp->f_flag & FFSYNC)
3077 			kif->kf_flags |= KF_FLAG_FSYNC;
3078 		if (fp->f_flag & FNONBLOCK)
3079 			kif->kf_flags |= KF_FLAG_NONBLOCK;
3080 		if (fp->f_flag & O_DIRECT)
3081 			kif->kf_flags |= KF_FLAG_DIRECT;
3082 		if (fp->f_flag & FHASLOCK)
3083 			kif->kf_flags |= KF_FLAG_HASLOCK;
3084 		kif->kf_offset = foffset_get(fp);
3085 		if (vp != NULL) {
3086 			vref(vp);
3087 			switch (vp->v_type) {
3088 			case VNON:
3089 				kif->kf_vnode_type = KF_VTYPE_VNON;
3090 				break;
3091 			case VREG:
3092 				kif->kf_vnode_type = KF_VTYPE_VREG;
3093 				break;
3094 			case VDIR:
3095 				kif->kf_vnode_type = KF_VTYPE_VDIR;
3096 				break;
3097 			case VBLK:
3098 				kif->kf_vnode_type = KF_VTYPE_VBLK;
3099 				break;
3100 			case VCHR:
3101 				kif->kf_vnode_type = KF_VTYPE_VCHR;
3102 				break;
3103 			case VLNK:
3104 				kif->kf_vnode_type = KF_VTYPE_VLNK;
3105 				break;
3106 			case VSOCK:
3107 				kif->kf_vnode_type = KF_VTYPE_VSOCK;
3108 				break;
3109 			case VFIFO:
3110 				kif->kf_vnode_type = KF_VTYPE_VFIFO;
3111 				break;
3112 			case VBAD:
3113 				kif->kf_vnode_type = KF_VTYPE_VBAD;
3114 				break;
3115 			default:
3116 				kif->kf_vnode_type = KF_VTYPE_UNKNOWN;
3117 				break;
3118 			}
3119 			/*
3120 			 * It is OK to drop the filedesc lock here as we will
3121 			 * re-validate and re-evaluate its properties when
3122 			 * the loop continues.
3123 			 */
3124 			freepath = NULL;
3125 			fullpath = "-";
3126 			FILEDESC_SUNLOCK(fdp);
3127 			vn_fullpath(curthread, vp, &fullpath, &freepath);
3128 			vrele(vp);
3129 			strlcpy(kif->kf_path, fullpath,
3130 			    sizeof(kif->kf_path));
3131 			if (freepath != NULL)
3132 				free(freepath, M_TEMP);
3133 			FILEDESC_SLOCK(fdp);
3134 		}
3135 		if (so != NULL) {
3136 			struct sockaddr *sa;
3137 
3138 			if (so->so_proto->pr_usrreqs->pru_sockaddr(so, &sa)
3139 			    == 0 && sa->sa_len <= sizeof(kif->kf_sa_local)) {
3140 				bcopy(sa, &kif->kf_sa_local, sa->sa_len);
3141 				free(sa, M_SONAME);
3142 			}
3143 			if (so->so_proto->pr_usrreqs->pru_peeraddr(so, &sa)
3144 			    == 0 && sa->sa_len <= sizeof(kif->kf_sa_peer)) {
3145 				bcopy(sa, &kif->kf_sa_peer, sa->sa_len);
3146 				free(sa, M_SONAME);
3147 			}
3148 			kif->kf_sock_domain =
3149 			    so->so_proto->pr_domain->dom_family;
3150 			kif->kf_sock_type = so->so_type;
3151 			kif->kf_sock_protocol = so->so_proto->pr_protocol;
3152 		}
3153 		if (tp != NULL) {
3154 			strlcpy(kif->kf_path, tty_devname(tp),
3155 			    sizeof(kif->kf_path));
3156 		}
3157 		if (shmfd != NULL)
3158 			shm_path(shmfd, kif->kf_path, sizeof(kif->kf_path));
3159 		if (ks != NULL && ksem_info != NULL)
3160 			ksem_info(ks, kif->kf_path, sizeof(kif->kf_path), NULL);
3161 		error = SYSCTL_OUT(req, kif, sizeof(*kif));
3162 		if (error)
3163 			break;
3164 	}
3165 	FILEDESC_SUNLOCK(fdp);
3166 	fddrop(fdp);
3167 	free(kif, M_TEMP);
3168 	return (0);
3169 }
3170 
3171 static SYSCTL_NODE(_kern_proc, KERN_PROC_OFILEDESC, ofiledesc, CTLFLAG_RD,
3172     sysctl_kern_proc_ofiledesc, "Process ofiledesc entries");
3173 #endif	/* COMPAT_FREEBSD7 */
3174 
3175 #ifdef KINFO_FILE_SIZE
3176 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
3177 #endif
3178 
3179 struct export_fd_buf {
3180 	struct filedesc		*fdp;
3181 	struct sbuf 		*sb;
3182 	ssize_t			remainder;
3183 	struct kinfo_file	kif;
3184 };
3185 
3186 static int
3187 export_fd_to_sb(void *data, int type, int fd, int fflags, int refcnt,
3188     int64_t offset, cap_rights_t fd_cap_rights, struct export_fd_buf *efbuf)
3189 {
3190 	struct {
3191 		int	fflag;
3192 		int	kf_fflag;
3193 	} fflags_table[] = {
3194 		{ FAPPEND, KF_FLAG_APPEND },
3195 		{ FASYNC, KF_FLAG_ASYNC },
3196 		{ FFSYNC, KF_FLAG_FSYNC },
3197 		{ FHASLOCK, KF_FLAG_HASLOCK },
3198 		{ FNONBLOCK, KF_FLAG_NONBLOCK },
3199 		{ FREAD, KF_FLAG_READ },
3200 		{ FWRITE, KF_FLAG_WRITE },
3201 		{ O_CREAT, KF_FLAG_CREAT },
3202 		{ O_DIRECT, KF_FLAG_DIRECT },
3203 		{ O_EXCL, KF_FLAG_EXCL },
3204 		{ O_EXEC, KF_FLAG_EXEC },
3205 		{ O_EXLOCK, KF_FLAG_EXLOCK },
3206 		{ O_NOFOLLOW, KF_FLAG_NOFOLLOW },
3207 		{ O_SHLOCK, KF_FLAG_SHLOCK },
3208 		{ O_TRUNC, KF_FLAG_TRUNC }
3209 	};
3210 #define	NFFLAGS	(sizeof(fflags_table) / sizeof(*fflags_table))
3211 	struct kinfo_file *kif;
3212 	struct vnode *vp;
3213 	int error, locked;
3214 	unsigned int i;
3215 
3216 	if (efbuf->remainder == 0)
3217 		return (0);
3218 	kif = &efbuf->kif;
3219 	bzero(kif, sizeof(*kif));
3220 	locked = efbuf->fdp != NULL;
3221 	switch (type) {
3222 	case KF_TYPE_FIFO:
3223 	case KF_TYPE_VNODE:
3224 		if (locked) {
3225 			FILEDESC_SUNLOCK(efbuf->fdp);
3226 			locked = 0;
3227 		}
3228 		vp = (struct vnode *)data;
3229 		error = fill_vnode_info(vp, kif);
3230 		vrele(vp);
3231 		break;
3232 	case KF_TYPE_SOCKET:
3233 		error = fill_socket_info((struct socket *)data, kif);
3234 		break;
3235 	case KF_TYPE_PIPE:
3236 		error = fill_pipe_info((struct pipe *)data, kif);
3237 		break;
3238 	case KF_TYPE_PTS:
3239 		error = fill_pts_info((struct tty *)data, kif);
3240 		break;
3241 	case KF_TYPE_PROCDESC:
3242 		error = fill_procdesc_info((struct procdesc *)data, kif);
3243 		break;
3244 	case KF_TYPE_SEM:
3245 		error = fill_sem_info((struct file *)data, kif);
3246 		break;
3247 	case KF_TYPE_SHM:
3248 		error = fill_shm_info((struct file *)data, kif);
3249 		break;
3250 	default:
3251 		error = 0;
3252 	}
3253 	if (error == 0)
3254 		kif->kf_status |= KF_ATTR_VALID;
3255 
3256 	/*
3257 	 * Translate file access flags.
3258 	 */
3259 	for (i = 0; i < NFFLAGS; i++)
3260 		if (fflags & fflags_table[i].fflag)
3261 			kif->kf_flags |=  fflags_table[i].kf_fflag;
3262 	kif->kf_cap_rights = fd_cap_rights;
3263 	kif->kf_fd = fd;
3264 	kif->kf_type = type;
3265 	kif->kf_ref_count = refcnt;
3266 	kif->kf_offset = offset;
3267 	/* Pack record size down */
3268 	kif->kf_structsize = offsetof(struct kinfo_file, kf_path) +
3269 	    strlen(kif->kf_path) + 1;
3270 	kif->kf_structsize = roundup(kif->kf_structsize, sizeof(uint64_t));
3271 	if (efbuf->remainder != -1) {
3272 		if (efbuf->remainder < kif->kf_structsize) {
3273 			/* Terminate export. */
3274 			efbuf->remainder = 0;
3275 			if (efbuf->fdp != NULL && !locked)
3276 				FILEDESC_SLOCK(efbuf->fdp);
3277 			return (0);
3278 		}
3279 		efbuf->remainder -= kif->kf_structsize;
3280 	}
3281 	if (locked)
3282 		FILEDESC_SUNLOCK(efbuf->fdp);
3283 	error = sbuf_bcat(efbuf->sb, kif, kif->kf_structsize);
3284 	if (efbuf->fdp != NULL)
3285 		FILEDESC_SLOCK(efbuf->fdp);
3286 	return (error);
3287 }
3288 
3289 /*
3290  * Store a process file descriptor information to sbuf.
3291  *
3292  * Takes a locked proc as argument, and returns with the proc unlocked.
3293  */
3294 int
3295 kern_proc_filedesc_out(struct proc *p,  struct sbuf *sb, ssize_t maxlen)
3296 {
3297 	struct file *fp;
3298 	struct filedesc *fdp;
3299 	struct export_fd_buf *efbuf;
3300 	struct vnode *cttyvp, *textvp, *tracevp;
3301 	int64_t offset;
3302 	void *data;
3303 	int error, i;
3304 	int type, refcnt, fflags;
3305 	cap_rights_t fd_cap_rights;
3306 
3307 	PROC_LOCK_ASSERT(p, MA_OWNED);
3308 
3309 	/* ktrace vnode */
3310 	tracevp = p->p_tracevp;
3311 	if (tracevp != NULL)
3312 		vref(tracevp);
3313 	/* text vnode */
3314 	textvp = p->p_textvp;
3315 	if (textvp != NULL)
3316 		vref(textvp);
3317 	/* Controlling tty. */
3318 	cttyvp = NULL;
3319 	if (p->p_pgrp != NULL && p->p_pgrp->pg_session != NULL) {
3320 		cttyvp = p->p_pgrp->pg_session->s_ttyvp;
3321 		if (cttyvp != NULL)
3322 			vref(cttyvp);
3323 	}
3324 	fdp = fdhold(p);
3325 	PROC_UNLOCK(p);
3326 	efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
3327 	efbuf->fdp = NULL;
3328 	efbuf->sb = sb;
3329 	efbuf->remainder = maxlen;
3330 	if (tracevp != NULL)
3331 		export_fd_to_sb(tracevp, KF_TYPE_VNODE, KF_FD_TYPE_TRACE,
3332 		    FREAD | FWRITE, -1, -1, 0, efbuf);
3333 	if (textvp != NULL)
3334 		export_fd_to_sb(textvp, KF_TYPE_VNODE, KF_FD_TYPE_TEXT,
3335 		    FREAD, -1, -1, 0, efbuf);
3336 	if (cttyvp != NULL)
3337 		export_fd_to_sb(cttyvp, KF_TYPE_VNODE, KF_FD_TYPE_CTTY,
3338 		    FREAD | FWRITE, -1, -1, 0, efbuf);
3339 	error = 0;
3340 	if (fdp == NULL)
3341 		goto fail;
3342 	efbuf->fdp = fdp;
3343 	FILEDESC_SLOCK(fdp);
3344 	/* working directory */
3345 	if (fdp->fd_cdir != NULL) {
3346 		vref(fdp->fd_cdir);
3347 		data = fdp->fd_cdir;
3348 		export_fd_to_sb(data, KF_TYPE_VNODE, KF_FD_TYPE_CWD,
3349 		    FREAD, -1, -1, 0, efbuf);
3350 	}
3351 	/* root directory */
3352 	if (fdp->fd_rdir != NULL) {
3353 		vref(fdp->fd_rdir);
3354 		data = fdp->fd_rdir;
3355 		export_fd_to_sb(data, KF_TYPE_VNODE, KF_FD_TYPE_ROOT,
3356 		    FREAD, -1, -1, 0, efbuf);
3357 	}
3358 	/* jail directory */
3359 	if (fdp->fd_jdir != NULL) {
3360 		vref(fdp->fd_jdir);
3361 		data = fdp->fd_jdir;
3362 		export_fd_to_sb(data, KF_TYPE_VNODE, KF_FD_TYPE_JAIL,
3363 		    FREAD, -1, -1, 0, efbuf);
3364 	}
3365 	for (i = 0; i < fdp->fd_nfiles; i++) {
3366 		if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
3367 			continue;
3368 		data = NULL;
3369 #ifdef CAPABILITIES
3370 		fd_cap_rights = cap_rights(fdp, i);
3371 #else /* !CAPABILITIES */
3372 		fd_cap_rights = 0;
3373 #endif
3374 		switch (fp->f_type) {
3375 		case DTYPE_VNODE:
3376 			type = KF_TYPE_VNODE;
3377 			vref(fp->f_vnode);
3378 			data = fp->f_vnode;
3379 			break;
3380 
3381 		case DTYPE_SOCKET:
3382 			type = KF_TYPE_SOCKET;
3383 			data = fp->f_data;
3384 			break;
3385 
3386 		case DTYPE_PIPE:
3387 			type = KF_TYPE_PIPE;
3388 			data = fp->f_data;
3389 			break;
3390 
3391 		case DTYPE_FIFO:
3392 			type = KF_TYPE_FIFO;
3393 			vref(fp->f_vnode);
3394 			data = fp->f_vnode;
3395 			break;
3396 
3397 		case DTYPE_KQUEUE:
3398 			type = KF_TYPE_KQUEUE;
3399 			break;
3400 
3401 		case DTYPE_CRYPTO:
3402 			type = KF_TYPE_CRYPTO;
3403 			break;
3404 
3405 		case DTYPE_MQUEUE:
3406 			type = KF_TYPE_MQUEUE;
3407 			break;
3408 
3409 		case DTYPE_SHM:
3410 			type = KF_TYPE_SHM;
3411 			data = fp;
3412 			break;
3413 
3414 		case DTYPE_SEM:
3415 			type = KF_TYPE_SEM;
3416 			data = fp;
3417 			break;
3418 
3419 		case DTYPE_PTS:
3420 			type = KF_TYPE_PTS;
3421 			data = fp->f_data;
3422 			break;
3423 
3424 #ifdef PROCDESC
3425 		case DTYPE_PROCDESC:
3426 			type = KF_TYPE_PROCDESC;
3427 			data = fp->f_data;
3428 			break;
3429 #endif
3430 
3431 		default:
3432 			type = KF_TYPE_UNKNOWN;
3433 			break;
3434 		}
3435 		refcnt = fp->f_count;
3436 		fflags = fp->f_flag;
3437 		offset = foffset_get(fp);
3438 
3439 		/*
3440 		 * Create sysctl entry.
3441 		 * It is OK to drop the filedesc lock here as we will
3442 		 * re-validate and re-evaluate its properties when
3443 		 * the loop continues.
3444 		 */
3445 		error = export_fd_to_sb(data, type, i, fflags, refcnt,
3446 		    offset, fd_cap_rights, efbuf);
3447 		if (error)
3448 			break;
3449 	}
3450 	FILEDESC_SUNLOCK(fdp);
3451 	fddrop(fdp);
3452 fail:
3453 	free(efbuf, M_TEMP);
3454 	return (error);
3455 }
3456 
3457 #define FILEDESC_SBUF_SIZE	(sizeof(struct kinfo_file) * 5)
3458 
3459 /*
3460  * Get per-process file descriptors for use by procstat(1), et al.
3461  */
3462 static int
3463 sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS)
3464 {
3465 	struct sbuf sb;
3466 	struct proc *p;
3467 	ssize_t maxlen;
3468 	int error, error2, *name;
3469 
3470 	name = (int *)arg1;
3471 
3472 	sbuf_new_for_sysctl(&sb, NULL, FILEDESC_SBUF_SIZE, req);
3473 	error = pget((pid_t)name[0], PGET_CANDEBUG, &p);
3474 	if (error != 0) {
3475 		sbuf_delete(&sb);
3476 		return (error);
3477 	}
3478 	maxlen = req->oldptr != NULL ? req->oldlen : -1;
3479 	error = kern_proc_filedesc_out(p, &sb, maxlen);
3480 	error2 = sbuf_finish(&sb);
3481 	sbuf_delete(&sb);
3482 	return (error != 0 ? error : error2);
3483 }
3484 
3485 int
3486 vntype_to_kinfo(int vtype)
3487 {
3488 	struct {
3489 		int	vtype;
3490 		int	kf_vtype;
3491 	} vtypes_table[] = {
3492 		{ VBAD, KF_VTYPE_VBAD },
3493 		{ VBLK, KF_VTYPE_VBLK },
3494 		{ VCHR, KF_VTYPE_VCHR },
3495 		{ VDIR, KF_VTYPE_VDIR },
3496 		{ VFIFO, KF_VTYPE_VFIFO },
3497 		{ VLNK, KF_VTYPE_VLNK },
3498 		{ VNON, KF_VTYPE_VNON },
3499 		{ VREG, KF_VTYPE_VREG },
3500 		{ VSOCK, KF_VTYPE_VSOCK }
3501 	};
3502 #define	NVTYPES	(sizeof(vtypes_table) / sizeof(*vtypes_table))
3503 	unsigned int i;
3504 
3505 	/*
3506 	 * Perform vtype translation.
3507 	 */
3508 	for (i = 0; i < NVTYPES; i++)
3509 		if (vtypes_table[i].vtype == vtype)
3510 			break;
3511 	if (i < NVTYPES)
3512 		return (vtypes_table[i].kf_vtype);
3513 
3514 	return (KF_VTYPE_UNKNOWN);
3515 }
3516 
3517 static int
3518 fill_vnode_info(struct vnode *vp, struct kinfo_file *kif)
3519 {
3520 	struct vattr va;
3521 	char *fullpath, *freepath;
3522 	int error;
3523 
3524 	if (vp == NULL)
3525 		return (1);
3526 	kif->kf_vnode_type = vntype_to_kinfo(vp->v_type);
3527 	freepath = NULL;
3528 	fullpath = "-";
3529 	error = vn_fullpath(curthread, vp, &fullpath, &freepath);
3530 	if (error == 0) {
3531 		strlcpy(kif->kf_path, fullpath, sizeof(kif->kf_path));
3532 	}
3533 	if (freepath != NULL)
3534 		free(freepath, M_TEMP);
3535 
3536 	/*
3537 	 * Retrieve vnode attributes.
3538 	 */
3539 	va.va_fsid = VNOVAL;
3540 	va.va_rdev = NODEV;
3541 	vn_lock(vp, LK_SHARED | LK_RETRY);
3542 	error = VOP_GETATTR(vp, &va, curthread->td_ucred);
3543 	VOP_UNLOCK(vp, 0);
3544 	if (error != 0)
3545 		return (error);
3546 	if (va.va_fsid != VNOVAL)
3547 		kif->kf_un.kf_file.kf_file_fsid = va.va_fsid;
3548 	else
3549 		kif->kf_un.kf_file.kf_file_fsid =
3550 		    vp->v_mount->mnt_stat.f_fsid.val[0];
3551 	kif->kf_un.kf_file.kf_file_fileid = va.va_fileid;
3552 	kif->kf_un.kf_file.kf_file_mode = MAKEIMODE(va.va_type, va.va_mode);
3553 	kif->kf_un.kf_file.kf_file_size = va.va_size;
3554 	kif->kf_un.kf_file.kf_file_rdev = va.va_rdev;
3555 	return (0);
3556 }
3557 
3558 static int
3559 fill_socket_info(struct socket *so, struct kinfo_file *kif)
3560 {
3561 	struct sockaddr *sa;
3562 	struct inpcb *inpcb;
3563 	struct unpcb *unpcb;
3564 	int error;
3565 
3566 	if (so == NULL)
3567 		return (1);
3568 	kif->kf_sock_domain = so->so_proto->pr_domain->dom_family;
3569 	kif->kf_sock_type = so->so_type;
3570 	kif->kf_sock_protocol = so->so_proto->pr_protocol;
3571 	kif->kf_un.kf_sock.kf_sock_pcb = (uintptr_t)so->so_pcb;
3572 	switch(kif->kf_sock_domain) {
3573 	case AF_INET:
3574 	case AF_INET6:
3575 		if (kif->kf_sock_protocol == IPPROTO_TCP) {
3576 			if (so->so_pcb != NULL) {
3577 				inpcb = (struct inpcb *)(so->so_pcb);
3578 				kif->kf_un.kf_sock.kf_sock_inpcb =
3579 				    (uintptr_t)inpcb->inp_ppcb;
3580 			}
3581 		}
3582 		break;
3583 	case AF_UNIX:
3584 		if (so->so_pcb != NULL) {
3585 			unpcb = (struct unpcb *)(so->so_pcb);
3586 			if (unpcb->unp_conn) {
3587 				kif->kf_un.kf_sock.kf_sock_unpconn =
3588 				    (uintptr_t)unpcb->unp_conn;
3589 				kif->kf_un.kf_sock.kf_sock_rcv_sb_state =
3590 				    so->so_rcv.sb_state;
3591 				kif->kf_un.kf_sock.kf_sock_snd_sb_state =
3592 				    so->so_snd.sb_state;
3593 			}
3594 		}
3595 		break;
3596 	}
3597 	error = so->so_proto->pr_usrreqs->pru_sockaddr(so, &sa);
3598 	if (error == 0 && sa->sa_len <= sizeof(kif->kf_sa_local)) {
3599 		bcopy(sa, &kif->kf_sa_local, sa->sa_len);
3600 		free(sa, M_SONAME);
3601 	}
3602 	error = so->so_proto->pr_usrreqs->pru_peeraddr(so, &sa);
3603 	if (error == 0 && sa->sa_len <= sizeof(kif->kf_sa_peer)) {
3604 		bcopy(sa, &kif->kf_sa_peer, sa->sa_len);
3605 		free(sa, M_SONAME);
3606 	}
3607 	strncpy(kif->kf_path, so->so_proto->pr_domain->dom_name,
3608 	    sizeof(kif->kf_path));
3609 	return (0);
3610 }
3611 
3612 static int
3613 fill_pts_info(struct tty *tp, struct kinfo_file *kif)
3614 {
3615 
3616 	if (tp == NULL)
3617 		return (1);
3618 	kif->kf_un.kf_pts.kf_pts_dev = tty_udev(tp);
3619 	strlcpy(kif->kf_path, tty_devname(tp), sizeof(kif->kf_path));
3620 	return (0);
3621 }
3622 
3623 static int
3624 fill_pipe_info(struct pipe *pi, struct kinfo_file *kif)
3625 {
3626 
3627 	if (pi == NULL)
3628 		return (1);
3629 	kif->kf_un.kf_pipe.kf_pipe_addr = (uintptr_t)pi;
3630 	kif->kf_un.kf_pipe.kf_pipe_peer = (uintptr_t)pi->pipe_peer;
3631 	kif->kf_un.kf_pipe.kf_pipe_buffer_cnt = pi->pipe_buffer.cnt;
3632 	return (0);
3633 }
3634 
3635 static int
3636 fill_procdesc_info(struct procdesc *pdp, struct kinfo_file *kif)
3637 {
3638 
3639 	if (pdp == NULL)
3640 		return (1);
3641 	kif->kf_un.kf_proc.kf_pid = pdp->pd_pid;
3642 	return (0);
3643 }
3644 
3645 static int
3646 fill_sem_info(struct file *fp, struct kinfo_file *kif)
3647 {
3648 	struct thread *td;
3649 	struct stat sb;
3650 
3651 	td = curthread;
3652 	if (fp->f_data == NULL)
3653 		return (1);
3654 	if (fo_stat(fp, &sb, td->td_ucred, td) != 0)
3655 		return (1);
3656 	if (ksem_info == NULL)
3657 		return (1);
3658 	ksem_info(fp->f_data, kif->kf_path, sizeof(kif->kf_path),
3659 	    &kif->kf_un.kf_sem.kf_sem_value);
3660 	kif->kf_un.kf_sem.kf_sem_mode = sb.st_mode;
3661 	return (0);
3662 }
3663 
3664 static int
3665 fill_shm_info(struct file *fp, struct kinfo_file *kif)
3666 {
3667 	struct thread *td;
3668 	struct stat sb;
3669 
3670 	td = curthread;
3671 	if (fp->f_data == NULL)
3672 		return (1);
3673 	if (fo_stat(fp, &sb, td->td_ucred, td) != 0)
3674 		return (1);
3675 	shm_path(fp->f_data, kif->kf_path, sizeof(kif->kf_path));
3676 	kif->kf_un.kf_file.kf_file_mode = sb.st_mode;
3677 	kif->kf_un.kf_file.kf_file_size = sb.st_size;
3678 	return (0);
3679 }
3680 
3681 static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc, CTLFLAG_RD,
3682     sysctl_kern_proc_filedesc, "Process filedesc entries");
3683 
3684 #ifdef DDB
3685 /*
3686  * For the purposes of debugging, generate a human-readable string for the
3687  * file type.
3688  */
3689 static const char *
3690 file_type_to_name(short type)
3691 {
3692 
3693 	switch (type) {
3694 	case 0:
3695 		return ("zero");
3696 	case DTYPE_VNODE:
3697 		return ("vnod");
3698 	case DTYPE_SOCKET:
3699 		return ("sock");
3700 	case DTYPE_PIPE:
3701 		return ("pipe");
3702 	case DTYPE_FIFO:
3703 		return ("fifo");
3704 	case DTYPE_KQUEUE:
3705 		return ("kque");
3706 	case DTYPE_CRYPTO:
3707 		return ("crpt");
3708 	case DTYPE_MQUEUE:
3709 		return ("mque");
3710 	case DTYPE_SHM:
3711 		return ("shm");
3712 	case DTYPE_SEM:
3713 		return ("ksem");
3714 	default:
3715 		return ("unkn");
3716 	}
3717 }
3718 
3719 /*
3720  * For the purposes of debugging, identify a process (if any, perhaps one of
3721  * many) that references the passed file in its file descriptor array. Return
3722  * NULL if none.
3723  */
3724 static struct proc *
3725 file_to_first_proc(struct file *fp)
3726 {
3727 	struct filedesc *fdp;
3728 	struct proc *p;
3729 	int n;
3730 
3731 	FOREACH_PROC_IN_SYSTEM(p) {
3732 		if (p->p_state == PRS_NEW)
3733 			continue;
3734 		fdp = p->p_fd;
3735 		if (fdp == NULL)
3736 			continue;
3737 		for (n = 0; n < fdp->fd_nfiles; n++) {
3738 			if (fp == fdp->fd_ofiles[n].fde_file)
3739 				return (p);
3740 		}
3741 	}
3742 	return (NULL);
3743 }
3744 
3745 static void
3746 db_print_file(struct file *fp, int header)
3747 {
3748 	struct proc *p;
3749 
3750 	if (header)
3751 		db_printf("%8s %4s %8s %8s %4s %5s %6s %8s %5s %12s\n",
3752 		    "File", "Type", "Data", "Flag", "GCFl", "Count",
3753 		    "MCount", "Vnode", "FPID", "FCmd");
3754 	p = file_to_first_proc(fp);
3755 	db_printf("%8p %4s %8p %08x %04x %5d %6d %8p %5d %12s\n", fp,
3756 	    file_type_to_name(fp->f_type), fp->f_data, fp->f_flag,
3757 	    0, fp->f_count, 0, fp->f_vnode,
3758 	    p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-");
3759 }
3760 
3761 DB_SHOW_COMMAND(file, db_show_file)
3762 {
3763 	struct file *fp;
3764 
3765 	if (!have_addr) {
3766 		db_printf("usage: show file <addr>\n");
3767 		return;
3768 	}
3769 	fp = (struct file *)addr;
3770 	db_print_file(fp, 1);
3771 }
3772 
3773 DB_SHOW_COMMAND(files, db_show_files)
3774 {
3775 	struct filedesc *fdp;
3776 	struct file *fp;
3777 	struct proc *p;
3778 	int header;
3779 	int n;
3780 
3781 	header = 1;
3782 	FOREACH_PROC_IN_SYSTEM(p) {
3783 		if (p->p_state == PRS_NEW)
3784 			continue;
3785 		if ((fdp = p->p_fd) == NULL)
3786 			continue;
3787 		for (n = 0; n < fdp->fd_nfiles; ++n) {
3788 			if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
3789 				continue;
3790 			db_print_file(fp, header);
3791 			header = 0;
3792 		}
3793 	}
3794 }
3795 #endif
3796 
3797 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
3798     &maxfilesperproc, 0, "Maximum files allowed open per process");
3799 
3800 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
3801     &maxfiles, 0, "Maximum number of files");
3802 
3803 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
3804     __DEVOLATILE(int *, &openfiles), 0, "System-wide number of open files");
3805 
3806 /* ARGSUSED*/
3807 static void
3808 filelistinit(void *dummy)
3809 {
3810 
3811 	file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL,
3812 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
3813 	mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF);
3814 	mtx_init(&fdesc_mtx, "fdesc", NULL, MTX_DEF);
3815 }
3816 SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL);
3817 
3818 /*-------------------------------------------------------------------*/
3819 
3820 static int
3821 badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred,
3822     int flags, struct thread *td)
3823 {
3824 
3825 	return (EBADF);
3826 }
3827 
3828 static int
3829 badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
3830     struct thread *td)
3831 {
3832 
3833 	return (EINVAL);
3834 }
3835 
3836 static int
3837 badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred,
3838     struct thread *td)
3839 {
3840 
3841 	return (EBADF);
3842 }
3843 
3844 static int
3845 badfo_poll(struct file *fp, int events, struct ucred *active_cred,
3846     struct thread *td)
3847 {
3848 
3849 	return (0);
3850 }
3851 
3852 static int
3853 badfo_kqfilter(struct file *fp, struct knote *kn)
3854 {
3855 
3856 	return (EBADF);
3857 }
3858 
3859 static int
3860 badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred,
3861     struct thread *td)
3862 {
3863 
3864 	return (EBADF);
3865 }
3866 
3867 static int
3868 badfo_close(struct file *fp, struct thread *td)
3869 {
3870 
3871 	return (EBADF);
3872 }
3873 
3874 static int
3875 badfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
3876     struct thread *td)
3877 {
3878 
3879 	return (EBADF);
3880 }
3881 
3882 static int
3883 badfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
3884     struct thread *td)
3885 {
3886 
3887 	return (EBADF);
3888 }
3889 
3890 static int
3891 badfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
3892     struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
3893     int kflags, struct thread *td)
3894 {
3895 
3896 	return (EBADF);
3897 }
3898 
3899 struct fileops badfileops = {
3900 	.fo_read = badfo_readwrite,
3901 	.fo_write = badfo_readwrite,
3902 	.fo_truncate = badfo_truncate,
3903 	.fo_ioctl = badfo_ioctl,
3904 	.fo_poll = badfo_poll,
3905 	.fo_kqfilter = badfo_kqfilter,
3906 	.fo_stat = badfo_stat,
3907 	.fo_close = badfo_close,
3908 	.fo_chmod = badfo_chmod,
3909 	.fo_chown = badfo_chown,
3910 	.fo_sendfile = badfo_sendfile,
3911 };
3912 
3913 int
3914 invfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
3915     struct thread *td)
3916 {
3917 
3918 	return (EINVAL);
3919 }
3920 
3921 int
3922 invfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
3923     struct thread *td)
3924 {
3925 
3926 	return (EINVAL);
3927 }
3928 
3929 int
3930 invfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
3931     struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
3932     int kflags, struct thread *td)
3933 {
3934 
3935 	return (EINVAL);
3936 }
3937 
3938 /*-------------------------------------------------------------------*/
3939 
3940 /*
3941  * File Descriptor pseudo-device driver (/dev/fd/).
3942  *
3943  * Opening minor device N dup()s the file (if any) connected to file
3944  * descriptor N belonging to the calling process.  Note that this driver
3945  * consists of only the ``open()'' routine, because all subsequent
3946  * references to this file will be direct to the other driver.
3947  *
3948  * XXX: we could give this one a cloning event handler if necessary.
3949  */
3950 
3951 /* ARGSUSED */
3952 static int
3953 fdopen(struct cdev *dev, int mode, int type, struct thread *td)
3954 {
3955 
3956 	/*
3957 	 * XXX Kludge: set curthread->td_dupfd to contain the value of the
3958 	 * the file descriptor being sought for duplication. The error
3959 	 * return ensures that the vnode for this device will be released
3960 	 * by vn_open. Open will detect this special error and take the
3961 	 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
3962 	 * will simply report the error.
3963 	 */
3964 	td->td_dupfd = dev2unit(dev);
3965 	return (ENODEV);
3966 }
3967 
3968 static struct cdevsw fildesc_cdevsw = {
3969 	.d_version =	D_VERSION,
3970 	.d_open =	fdopen,
3971 	.d_name =	"FD",
3972 };
3973 
3974 static void
3975 fildesc_drvinit(void *unused)
3976 {
3977 	struct cdev *dev;
3978 
3979 	dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 0, NULL,
3980 	    UID_ROOT, GID_WHEEL, 0666, "fd/0");
3981 	make_dev_alias(dev, "stdin");
3982 	dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 1, NULL,
3983 	    UID_ROOT, GID_WHEEL, 0666, "fd/1");
3984 	make_dev_alias(dev, "stdout");
3985 	dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 2, NULL,
3986 	    UID_ROOT, GID_WHEEL, 0666, "fd/2");
3987 	make_dev_alias(dev, "stderr");
3988 }
3989 
3990 SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL);
3991