xref: /freebsd/sys/kern/kern_descrip.c (revision 81ad626541db97eb356e2c1d4a20eb2a26a766ab)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1982, 1986, 1989, 1991, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  * (c) UNIX System Laboratories, Inc.
7  * All or some portions of this file are derived from material licensed
8  * to the University of California by American Telephone and Telegraph
9  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
10  * the permission of UNIX System Laboratories, Inc.
11  *
12  * Redistribution and use in source and binary forms, with or without
13  * modification, are permitted provided that the following conditions
14  * are met:
15  * 1. Redistributions of source code must retain the above copyright
16  *    notice, this list of conditions and the following disclaimer.
17  * 2. Redistributions in binary form must reproduce the above copyright
18  *    notice, this list of conditions and the following disclaimer in the
19  *    documentation and/or other materials provided with the distribution.
20  * 3. Neither the name of the University nor the names of its contributors
21  *    may be used to endorse or promote products derived from this software
22  *    without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  *
36  *	@(#)kern_descrip.c	8.6 (Berkeley) 4/19/94
37  */
38 
39 #include <sys/cdefs.h>
40 __FBSDID("$FreeBSD$");
41 
42 #include "opt_capsicum.h"
43 #include "opt_ddb.h"
44 #include "opt_ktrace.h"
45 
46 #include <sys/param.h>
47 #include <sys/systm.h>
48 
49 #include <sys/capsicum.h>
50 #include <sys/conf.h>
51 #include <sys/fcntl.h>
52 #include <sys/file.h>
53 #include <sys/filedesc.h>
54 #include <sys/filio.h>
55 #include <sys/jail.h>
56 #include <sys/kernel.h>
57 #include <sys/limits.h>
58 #include <sys/lock.h>
59 #include <sys/malloc.h>
60 #include <sys/mount.h>
61 #include <sys/mutex.h>
62 #include <sys/namei.h>
63 #include <sys/selinfo.h>
64 #include <sys/poll.h>
65 #include <sys/priv.h>
66 #include <sys/proc.h>
67 #include <sys/protosw.h>
68 #include <sys/racct.h>
69 #include <sys/resourcevar.h>
70 #include <sys/sbuf.h>
71 #include <sys/signalvar.h>
72 #include <sys/kdb.h>
73 #include <sys/smr.h>
74 #include <sys/stat.h>
75 #include <sys/sx.h>
76 #include <sys/syscallsubr.h>
77 #include <sys/sysctl.h>
78 #include <sys/sysproto.h>
79 #include <sys/unistd.h>
80 #include <sys/user.h>
81 #include <sys/vnode.h>
82 #include <sys/ktrace.h>
83 
84 #include <net/vnet.h>
85 
86 #include <security/audit/audit.h>
87 
88 #include <vm/uma.h>
89 #include <vm/vm.h>
90 
91 #include <ddb/ddb.h>
92 
93 static MALLOC_DEFINE(M_FILEDESC, "filedesc", "Open file descriptor table");
94 static MALLOC_DEFINE(M_PWD, "pwd", "Descriptor table vnodes");
95 static MALLOC_DEFINE(M_PWDDESC, "pwddesc", "Pwd descriptors");
96 static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "filedesc_to_leader",
97     "file desc to leader structures");
98 static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures");
99 MALLOC_DEFINE(M_FILECAPS, "filecaps", "descriptor capabilities");
100 
101 MALLOC_DECLARE(M_FADVISE);
102 
103 static __read_mostly uma_zone_t file_zone;
104 static __read_mostly uma_zone_t filedesc0_zone;
105 __read_mostly uma_zone_t pwd_zone;
106 VFS_SMR_DECLARE;
107 
108 static int	closefp(struct filedesc *fdp, int fd, struct file *fp,
109 		    struct thread *td, bool holdleaders, bool audit);
110 static void	export_file_to_kinfo(struct file *fp, int fd,
111 		    cap_rights_t *rightsp, struct kinfo_file *kif,
112 		    struct filedesc *fdp, int flags);
113 static int	fd_first_free(struct filedesc *fdp, int low, int size);
114 static void	fdgrowtable(struct filedesc *fdp, int nfd);
115 static void	fdgrowtable_exp(struct filedesc *fdp, int nfd);
116 static void	fdunused(struct filedesc *fdp, int fd);
117 static void	fdused(struct filedesc *fdp, int fd);
118 static int	fget_unlocked_seq(struct thread *td, int fd,
119 		    cap_rights_t *needrightsp, struct file **fpp, seqc_t *seqp);
120 static int	getmaxfd(struct thread *td);
121 static u_long	*filecaps_copy_prep(const struct filecaps *src);
122 static void	filecaps_copy_finish(const struct filecaps *src,
123 		    struct filecaps *dst, u_long *ioctls);
124 static u_long 	*filecaps_free_prep(struct filecaps *fcaps);
125 static void	filecaps_free_finish(u_long *ioctls);
126 
127 static struct pwd *pwd_alloc(void);
128 
129 /*
130  * Each process has:
131  *
132  * - An array of open file descriptors (fd_ofiles)
133  * - An array of file flags (fd_ofileflags)
134  * - A bitmap recording which descriptors are in use (fd_map)
135  *
136  * A process starts out with NDFILE descriptors.  The value of NDFILE has
137  * been selected based the historical limit of 20 open files, and an
138  * assumption that the majority of processes, especially short-lived
139  * processes like shells, will never need more.
140  *
141  * If this initial allocation is exhausted, a larger descriptor table and
142  * map are allocated dynamically, and the pointers in the process's struct
143  * filedesc are updated to point to those.  This is repeated every time
144  * the process runs out of file descriptors (provided it hasn't hit its
145  * resource limit).
146  *
147  * Since threads may hold references to individual descriptor table
148  * entries, the tables are never freed.  Instead, they are placed on a
149  * linked list and freed only when the struct filedesc is released.
150  */
151 #define NDFILE		20
152 #define NDSLOTSIZE	sizeof(NDSLOTTYPE)
153 #define	NDENTRIES	(NDSLOTSIZE * __CHAR_BIT)
154 #define NDSLOT(x)	((x) / NDENTRIES)
155 #define NDBIT(x)	((NDSLOTTYPE)1 << ((x) % NDENTRIES))
156 #define	NDSLOTS(x)	(((x) + NDENTRIES - 1) / NDENTRIES)
157 
158 #define	FILEDESC_FOREACH_FDE(fdp, _iterator, _fde)				\
159 	struct filedesc *_fdp = (fdp);						\
160 	int _lastfile = fdlastfile_single(_fdp);				\
161 	for (_iterator = 0; _iterator <= _lastfile; _iterator++)		\
162 		if ((_fde = &_fdp->fd_ofiles[_iterator])->fde_file != NULL)
163 
164 #define	FILEDESC_FOREACH_FP(fdp, _iterator, _fp)				\
165 	struct filedesc *_fdp = (fdp);						\
166 	int _lastfile = fdlastfile_single(_fdp);				\
167 	for (_iterator = 0; _iterator <= _lastfile; _iterator++)		\
168 		if ((_fp = _fdp->fd_ofiles[_iterator].fde_file) != NULL)
169 
170 /*
171  * SLIST entry used to keep track of ofiles which must be reclaimed when
172  * the process exits.
173  */
174 struct freetable {
175 	struct fdescenttbl *ft_table;
176 	SLIST_ENTRY(freetable) ft_next;
177 };
178 
179 /*
180  * Initial allocation: a filedesc structure + the head of SLIST used to
181  * keep track of old ofiles + enough space for NDFILE descriptors.
182  */
183 
184 struct fdescenttbl0 {
185 	int	fdt_nfiles;
186 	struct	filedescent fdt_ofiles[NDFILE];
187 };
188 
189 struct filedesc0 {
190 	struct filedesc fd_fd;
191 	SLIST_HEAD(, freetable) fd_free;
192 	struct	fdescenttbl0 fd_dfiles;
193 	NDSLOTTYPE fd_dmap[NDSLOTS(NDFILE)];
194 };
195 
196 /*
197  * Descriptor management.
198  */
199 static int __exclusive_cache_line openfiles; /* actual number of open files */
200 struct mtx sigio_lock;		/* mtx to protect pointers to sigio */
201 void __read_mostly (*mq_fdclose)(struct thread *td, int fd, struct file *fp);
202 
203 /*
204  * If low >= size, just return low. Otherwise find the first zero bit in the
205  * given bitmap, starting at low and not exceeding size - 1. Return size if
206  * not found.
207  */
208 static int
209 fd_first_free(struct filedesc *fdp, int low, int size)
210 {
211 	NDSLOTTYPE *map = fdp->fd_map;
212 	NDSLOTTYPE mask;
213 	int off, maxoff;
214 
215 	if (low >= size)
216 		return (low);
217 
218 	off = NDSLOT(low);
219 	if (low % NDENTRIES) {
220 		mask = ~(~(NDSLOTTYPE)0 >> (NDENTRIES - (low % NDENTRIES)));
221 		if ((mask &= ~map[off]) != 0UL)
222 			return (off * NDENTRIES + ffsl(mask) - 1);
223 		++off;
224 	}
225 	for (maxoff = NDSLOTS(size); off < maxoff; ++off)
226 		if (map[off] != ~0UL)
227 			return (off * NDENTRIES + ffsl(~map[off]) - 1);
228 	return (size);
229 }
230 
231 /*
232  * Find the last used fd.
233  *
234  * Call this variant if fdp can't be modified by anyone else (e.g, during exec).
235  * Otherwise use fdlastfile.
236  */
237 int
238 fdlastfile_single(struct filedesc *fdp)
239 {
240 	NDSLOTTYPE *map = fdp->fd_map;
241 	int off, minoff;
242 
243 	off = NDSLOT(fdp->fd_nfiles - 1);
244 	for (minoff = NDSLOT(0); off >= minoff; --off)
245 		if (map[off] != 0)
246 			return (off * NDENTRIES + flsl(map[off]) - 1);
247 	return (-1);
248 }
249 
250 int
251 fdlastfile(struct filedesc *fdp)
252 {
253 
254 	FILEDESC_LOCK_ASSERT(fdp);
255 	return (fdlastfile_single(fdp));
256 }
257 
258 static int
259 fdisused(struct filedesc *fdp, int fd)
260 {
261 
262 	KASSERT(fd >= 0 && fd < fdp->fd_nfiles,
263 	    ("file descriptor %d out of range (0, %d)", fd, fdp->fd_nfiles));
264 
265 	return ((fdp->fd_map[NDSLOT(fd)] & NDBIT(fd)) != 0);
266 }
267 
268 /*
269  * Mark a file descriptor as used.
270  */
271 static void
272 fdused_init(struct filedesc *fdp, int fd)
273 {
274 
275 	KASSERT(!fdisused(fdp, fd), ("fd=%d is already used", fd));
276 
277 	fdp->fd_map[NDSLOT(fd)] |= NDBIT(fd);
278 }
279 
280 static void
281 fdused(struct filedesc *fdp, int fd)
282 {
283 
284 	FILEDESC_XLOCK_ASSERT(fdp);
285 
286 	fdused_init(fdp, fd);
287 	if (fd == fdp->fd_freefile)
288 		fdp->fd_freefile++;
289 }
290 
291 /*
292  * Mark a file descriptor as unused.
293  */
294 static void
295 fdunused(struct filedesc *fdp, int fd)
296 {
297 
298 	FILEDESC_XLOCK_ASSERT(fdp);
299 
300 	KASSERT(fdisused(fdp, fd), ("fd=%d is already unused", fd));
301 	KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
302 	    ("fd=%d is still in use", fd));
303 
304 	fdp->fd_map[NDSLOT(fd)] &= ~NDBIT(fd);
305 	if (fd < fdp->fd_freefile)
306 		fdp->fd_freefile = fd;
307 }
308 
309 /*
310  * Free a file descriptor.
311  *
312  * Avoid some work if fdp is about to be destroyed.
313  */
314 static inline void
315 fdefree_last(struct filedescent *fde)
316 {
317 
318 	filecaps_free(&fde->fde_caps);
319 }
320 
321 static inline void
322 fdfree(struct filedesc *fdp, int fd)
323 {
324 	struct filedescent *fde;
325 
326 	FILEDESC_XLOCK_ASSERT(fdp);
327 	fde = &fdp->fd_ofiles[fd];
328 #ifdef CAPABILITIES
329 	seqc_write_begin(&fde->fde_seqc);
330 #endif
331 	fde->fde_file = NULL;
332 #ifdef CAPABILITIES
333 	seqc_write_end(&fde->fde_seqc);
334 #endif
335 	fdefree_last(fde);
336 	fdunused(fdp, fd);
337 }
338 
339 /*
340  * System calls on descriptors.
341  */
342 #ifndef _SYS_SYSPROTO_H_
343 struct getdtablesize_args {
344 	int	dummy;
345 };
346 #endif
347 /* ARGSUSED */
348 int
349 sys_getdtablesize(struct thread *td, struct getdtablesize_args *uap)
350 {
351 #ifdef	RACCT
352 	uint64_t lim;
353 #endif
354 
355 	td->td_retval[0] = getmaxfd(td);
356 #ifdef	RACCT
357 	PROC_LOCK(td->td_proc);
358 	lim = racct_get_limit(td->td_proc, RACCT_NOFILE);
359 	PROC_UNLOCK(td->td_proc);
360 	if (lim < td->td_retval[0])
361 		td->td_retval[0] = lim;
362 #endif
363 	return (0);
364 }
365 
366 /*
367  * Duplicate a file descriptor to a particular value.
368  *
369  * Note: keep in mind that a potential race condition exists when closing
370  * descriptors from a shared descriptor table (via rfork).
371  */
372 #ifndef _SYS_SYSPROTO_H_
373 struct dup2_args {
374 	u_int	from;
375 	u_int	to;
376 };
377 #endif
378 /* ARGSUSED */
379 int
380 sys_dup2(struct thread *td, struct dup2_args *uap)
381 {
382 
383 	return (kern_dup(td, FDDUP_FIXED, 0, (int)uap->from, (int)uap->to));
384 }
385 
386 /*
387  * Duplicate a file descriptor.
388  */
389 #ifndef _SYS_SYSPROTO_H_
390 struct dup_args {
391 	u_int	fd;
392 };
393 #endif
394 /* ARGSUSED */
395 int
396 sys_dup(struct thread *td, struct dup_args *uap)
397 {
398 
399 	return (kern_dup(td, FDDUP_NORMAL, 0, (int)uap->fd, 0));
400 }
401 
402 /*
403  * The file control system call.
404  */
405 #ifndef _SYS_SYSPROTO_H_
406 struct fcntl_args {
407 	int	fd;
408 	int	cmd;
409 	long	arg;
410 };
411 #endif
412 /* ARGSUSED */
413 int
414 sys_fcntl(struct thread *td, struct fcntl_args *uap)
415 {
416 
417 	return (kern_fcntl_freebsd(td, uap->fd, uap->cmd, uap->arg));
418 }
419 
420 int
421 kern_fcntl_freebsd(struct thread *td, int fd, int cmd, long arg)
422 {
423 	struct flock fl;
424 	struct __oflock ofl;
425 	intptr_t arg1;
426 	int error, newcmd;
427 
428 	error = 0;
429 	newcmd = cmd;
430 	switch (cmd) {
431 	case F_OGETLK:
432 	case F_OSETLK:
433 	case F_OSETLKW:
434 		/*
435 		 * Convert old flock structure to new.
436 		 */
437 		error = copyin((void *)(intptr_t)arg, &ofl, sizeof(ofl));
438 		fl.l_start = ofl.l_start;
439 		fl.l_len = ofl.l_len;
440 		fl.l_pid = ofl.l_pid;
441 		fl.l_type = ofl.l_type;
442 		fl.l_whence = ofl.l_whence;
443 		fl.l_sysid = 0;
444 
445 		switch (cmd) {
446 		case F_OGETLK:
447 			newcmd = F_GETLK;
448 			break;
449 		case F_OSETLK:
450 			newcmd = F_SETLK;
451 			break;
452 		case F_OSETLKW:
453 			newcmd = F_SETLKW;
454 			break;
455 		}
456 		arg1 = (intptr_t)&fl;
457 		break;
458 	case F_GETLK:
459 	case F_SETLK:
460 	case F_SETLKW:
461 	case F_SETLK_REMOTE:
462 		error = copyin((void *)(intptr_t)arg, &fl, sizeof(fl));
463 		arg1 = (intptr_t)&fl;
464 		break;
465 	default:
466 		arg1 = arg;
467 		break;
468 	}
469 	if (error)
470 		return (error);
471 	error = kern_fcntl(td, fd, newcmd, arg1);
472 	if (error)
473 		return (error);
474 	if (cmd == F_OGETLK) {
475 		ofl.l_start = fl.l_start;
476 		ofl.l_len = fl.l_len;
477 		ofl.l_pid = fl.l_pid;
478 		ofl.l_type = fl.l_type;
479 		ofl.l_whence = fl.l_whence;
480 		error = copyout(&ofl, (void *)(intptr_t)arg, sizeof(ofl));
481 	} else if (cmd == F_GETLK) {
482 		error = copyout(&fl, (void *)(intptr_t)arg, sizeof(fl));
483 	}
484 	return (error);
485 }
486 
487 int
488 kern_fcntl(struct thread *td, int fd, int cmd, intptr_t arg)
489 {
490 	struct filedesc *fdp;
491 	struct flock *flp;
492 	struct file *fp, *fp2;
493 	struct filedescent *fde;
494 	struct proc *p;
495 	struct vnode *vp;
496 	struct mount *mp;
497 	struct kinfo_file *kif;
498 	int error, flg, kif_sz, seals, tmp;
499 	uint64_t bsize;
500 	off_t foffset;
501 
502 	error = 0;
503 	flg = F_POSIX;
504 	p = td->td_proc;
505 	fdp = p->p_fd;
506 
507 	AUDIT_ARG_FD(cmd);
508 	AUDIT_ARG_CMD(cmd);
509 	switch (cmd) {
510 	case F_DUPFD:
511 		tmp = arg;
512 		error = kern_dup(td, FDDUP_FCNTL, 0, fd, tmp);
513 		break;
514 
515 	case F_DUPFD_CLOEXEC:
516 		tmp = arg;
517 		error = kern_dup(td, FDDUP_FCNTL, FDDUP_FLAG_CLOEXEC, fd, tmp);
518 		break;
519 
520 	case F_DUP2FD:
521 		tmp = arg;
522 		error = kern_dup(td, FDDUP_FIXED, 0, fd, tmp);
523 		break;
524 
525 	case F_DUP2FD_CLOEXEC:
526 		tmp = arg;
527 		error = kern_dup(td, FDDUP_FIXED, FDDUP_FLAG_CLOEXEC, fd, tmp);
528 		break;
529 
530 	case F_GETFD:
531 		error = EBADF;
532 		FILEDESC_SLOCK(fdp);
533 		fde = fdeget_noref(fdp, fd);
534 		if (fde != NULL) {
535 			td->td_retval[0] =
536 			    (fde->fde_flags & UF_EXCLOSE) ? FD_CLOEXEC : 0;
537 			error = 0;
538 		}
539 		FILEDESC_SUNLOCK(fdp);
540 		break;
541 
542 	case F_SETFD:
543 		error = EBADF;
544 		FILEDESC_XLOCK(fdp);
545 		fde = fdeget_noref(fdp, fd);
546 		if (fde != NULL) {
547 			fde->fde_flags = (fde->fde_flags & ~UF_EXCLOSE) |
548 			    (arg & FD_CLOEXEC ? UF_EXCLOSE : 0);
549 			error = 0;
550 		}
551 		FILEDESC_XUNLOCK(fdp);
552 		break;
553 
554 	case F_GETFL:
555 		error = fget_fcntl(td, fd, &cap_fcntl_rights, F_GETFL, &fp);
556 		if (error != 0)
557 			break;
558 		td->td_retval[0] = OFLAGS(fp->f_flag);
559 		fdrop(fp, td);
560 		break;
561 
562 	case F_SETFL:
563 		error = fget_fcntl(td, fd, &cap_fcntl_rights, F_SETFL, &fp);
564 		if (error != 0)
565 			break;
566 		if (fp->f_ops == &path_fileops) {
567 			fdrop(fp, td);
568 			error = EBADF;
569 			break;
570 		}
571 		do {
572 			tmp = flg = fp->f_flag;
573 			tmp &= ~FCNTLFLAGS;
574 			tmp |= FFLAGS(arg & ~O_ACCMODE) & FCNTLFLAGS;
575 		} while (atomic_cmpset_int(&fp->f_flag, flg, tmp) == 0);
576 		tmp = fp->f_flag & FNONBLOCK;
577 		error = fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
578 		if (error != 0) {
579 			fdrop(fp, td);
580 			break;
581 		}
582 		tmp = fp->f_flag & FASYNC;
583 		error = fo_ioctl(fp, FIOASYNC, &tmp, td->td_ucred, td);
584 		if (error == 0) {
585 			fdrop(fp, td);
586 			break;
587 		}
588 		atomic_clear_int(&fp->f_flag, FNONBLOCK);
589 		tmp = 0;
590 		(void)fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
591 		fdrop(fp, td);
592 		break;
593 
594 	case F_GETOWN:
595 		error = fget_fcntl(td, fd, &cap_fcntl_rights, F_GETOWN, &fp);
596 		if (error != 0)
597 			break;
598 		error = fo_ioctl(fp, FIOGETOWN, &tmp, td->td_ucred, td);
599 		if (error == 0)
600 			td->td_retval[0] = tmp;
601 		fdrop(fp, td);
602 		break;
603 
604 	case F_SETOWN:
605 		error = fget_fcntl(td, fd, &cap_fcntl_rights, F_SETOWN, &fp);
606 		if (error != 0)
607 			break;
608 		tmp = arg;
609 		error = fo_ioctl(fp, FIOSETOWN, &tmp, td->td_ucred, td);
610 		fdrop(fp, td);
611 		break;
612 
613 	case F_SETLK_REMOTE:
614 		error = priv_check(td, PRIV_NFS_LOCKD);
615 		if (error != 0)
616 			return (error);
617 		flg = F_REMOTE;
618 		goto do_setlk;
619 
620 	case F_SETLKW:
621 		flg |= F_WAIT;
622 		/* FALLTHROUGH F_SETLK */
623 
624 	case F_SETLK:
625 	do_setlk:
626 		flp = (struct flock *)arg;
627 		if ((flg & F_REMOTE) != 0 && flp->l_sysid == 0) {
628 			error = EINVAL;
629 			break;
630 		}
631 
632 		error = fget_unlocked(td, fd, &cap_flock_rights, &fp);
633 		if (error != 0)
634 			break;
635 		if (fp->f_type != DTYPE_VNODE || fp->f_ops == &path_fileops) {
636 			error = EBADF;
637 			fdrop(fp, td);
638 			break;
639 		}
640 
641 		if (flp->l_whence == SEEK_CUR) {
642 			foffset = foffset_get(fp);
643 			if (foffset < 0 ||
644 			    (flp->l_start > 0 &&
645 			     foffset > OFF_MAX - flp->l_start)) {
646 				error = EOVERFLOW;
647 				fdrop(fp, td);
648 				break;
649 			}
650 			flp->l_start += foffset;
651 		}
652 
653 		vp = fp->f_vnode;
654 		switch (flp->l_type) {
655 		case F_RDLCK:
656 			if ((fp->f_flag & FREAD) == 0) {
657 				error = EBADF;
658 				break;
659 			}
660 			if ((p->p_leader->p_flag & P_ADVLOCK) == 0) {
661 				PROC_LOCK(p->p_leader);
662 				p->p_leader->p_flag |= P_ADVLOCK;
663 				PROC_UNLOCK(p->p_leader);
664 			}
665 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
666 			    flp, flg);
667 			break;
668 		case F_WRLCK:
669 			if ((fp->f_flag & FWRITE) == 0) {
670 				error = EBADF;
671 				break;
672 			}
673 			if ((p->p_leader->p_flag & P_ADVLOCK) == 0) {
674 				PROC_LOCK(p->p_leader);
675 				p->p_leader->p_flag |= P_ADVLOCK;
676 				PROC_UNLOCK(p->p_leader);
677 			}
678 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
679 			    flp, flg);
680 			break;
681 		case F_UNLCK:
682 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
683 			    flp, flg);
684 			break;
685 		case F_UNLCKSYS:
686 			if (flg != F_REMOTE) {
687 				error = EINVAL;
688 				break;
689 			}
690 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
691 			    F_UNLCKSYS, flp, flg);
692 			break;
693 		default:
694 			error = EINVAL;
695 			break;
696 		}
697 		if (error != 0 || flp->l_type == F_UNLCK ||
698 		    flp->l_type == F_UNLCKSYS) {
699 			fdrop(fp, td);
700 			break;
701 		}
702 
703 		/*
704 		 * Check for a race with close.
705 		 *
706 		 * The vnode is now advisory locked (or unlocked, but this case
707 		 * is not really important) as the caller requested.
708 		 * We had to drop the filedesc lock, so we need to recheck if
709 		 * the descriptor is still valid, because if it was closed
710 		 * in the meantime we need to remove advisory lock from the
711 		 * vnode - close on any descriptor leading to an advisory
712 		 * locked vnode, removes that lock.
713 		 * We will return 0 on purpose in that case, as the result of
714 		 * successful advisory lock might have been externally visible
715 		 * already. This is fine - effectively we pretend to the caller
716 		 * that the closing thread was a bit slower and that the
717 		 * advisory lock succeeded before the close.
718 		 */
719 		error = fget_unlocked(td, fd, &cap_no_rights, &fp2);
720 		if (error != 0) {
721 			fdrop(fp, td);
722 			break;
723 		}
724 		if (fp != fp2) {
725 			flp->l_whence = SEEK_SET;
726 			flp->l_start = 0;
727 			flp->l_len = 0;
728 			flp->l_type = F_UNLCK;
729 			(void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
730 			    F_UNLCK, flp, F_POSIX);
731 		}
732 		fdrop(fp, td);
733 		fdrop(fp2, td);
734 		break;
735 
736 	case F_GETLK:
737 		error = fget_unlocked(td, fd, &cap_flock_rights, &fp);
738 		if (error != 0)
739 			break;
740 		if (fp->f_type != DTYPE_VNODE || fp->f_ops == &path_fileops) {
741 			error = EBADF;
742 			fdrop(fp, td);
743 			break;
744 		}
745 		flp = (struct flock *)arg;
746 		if (flp->l_type != F_RDLCK && flp->l_type != F_WRLCK &&
747 		    flp->l_type != F_UNLCK) {
748 			error = EINVAL;
749 			fdrop(fp, td);
750 			break;
751 		}
752 		if (flp->l_whence == SEEK_CUR) {
753 			foffset = foffset_get(fp);
754 			if ((flp->l_start > 0 &&
755 			    foffset > OFF_MAX - flp->l_start) ||
756 			    (flp->l_start < 0 &&
757 			    foffset < OFF_MIN - flp->l_start)) {
758 				error = EOVERFLOW;
759 				fdrop(fp, td);
760 				break;
761 			}
762 			flp->l_start += foffset;
763 		}
764 		vp = fp->f_vnode;
765 		error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, flp,
766 		    F_POSIX);
767 		fdrop(fp, td);
768 		break;
769 
770 	case F_ADD_SEALS:
771 		error = fget_unlocked(td, fd, &cap_no_rights, &fp);
772 		if (error != 0)
773 			break;
774 		error = fo_add_seals(fp, arg);
775 		fdrop(fp, td);
776 		break;
777 
778 	case F_GET_SEALS:
779 		error = fget_unlocked(td, fd, &cap_no_rights, &fp);
780 		if (error != 0)
781 			break;
782 		if (fo_get_seals(fp, &seals) == 0)
783 			td->td_retval[0] = seals;
784 		else
785 			error = EINVAL;
786 		fdrop(fp, td);
787 		break;
788 
789 	case F_RDAHEAD:
790 		arg = arg ? 128 * 1024: 0;
791 		/* FALLTHROUGH */
792 	case F_READAHEAD:
793 		error = fget_unlocked(td, fd, &cap_no_rights, &fp);
794 		if (error != 0)
795 			break;
796 		if (fp->f_type != DTYPE_VNODE || fp->f_ops == &path_fileops) {
797 			fdrop(fp, td);
798 			error = EBADF;
799 			break;
800 		}
801 		vp = fp->f_vnode;
802 		if (vp->v_type != VREG) {
803 			fdrop(fp, td);
804 			error = ENOTTY;
805 			break;
806 		}
807 
808 		/*
809 		 * Exclusive lock synchronizes against f_seqcount reads and
810 		 * writes in sequential_heuristic().
811 		 */
812 		error = vn_lock(vp, LK_EXCLUSIVE);
813 		if (error != 0) {
814 			fdrop(fp, td);
815 			break;
816 		}
817 		if (arg >= 0) {
818 			bsize = fp->f_vnode->v_mount->mnt_stat.f_iosize;
819 			arg = MIN(arg, INT_MAX - bsize + 1);
820 			fp->f_seqcount[UIO_READ] = MIN(IO_SEQMAX,
821 			    (arg + bsize - 1) / bsize);
822 			atomic_set_int(&fp->f_flag, FRDAHEAD);
823 		} else {
824 			atomic_clear_int(&fp->f_flag, FRDAHEAD);
825 		}
826 		VOP_UNLOCK(vp);
827 		fdrop(fp, td);
828 		break;
829 
830 	case F_ISUNIONSTACK:
831 		/*
832 		 * Check if the vnode is part of a union stack (either the
833 		 * "union" flag from mount(2) or unionfs).
834 		 *
835 		 * Prior to introduction of this op libc's readdir would call
836 		 * fstatfs(2), in effect unnecessarily copying kilobytes of
837 		 * data just to check fs name and a mount flag.
838 		 *
839 		 * Fixing the code to handle everything in the kernel instead
840 		 * is a non-trivial endeavor and has low priority, thus this
841 		 * horrible kludge facilitates the current behavior in a much
842 		 * cheaper manner until someone(tm) sorts this out.
843 		 */
844 		error = fget_unlocked(td, fd, &cap_no_rights, &fp);
845 		if (error != 0)
846 			break;
847 		if (fp->f_type != DTYPE_VNODE) {
848 			fdrop(fp, td);
849 			error = EBADF;
850 			break;
851 		}
852 		vp = fp->f_vnode;
853 		/*
854 		 * Since we don't prevent dooming the vnode even non-null mp
855 		 * found can become immediately stale. This is tolerable since
856 		 * mount points are type-stable (providing safe memory access)
857 		 * and any vfs op on this vnode going forward will return an
858 		 * error (meaning return value in this case is meaningless).
859 		 */
860 		mp = atomic_load_ptr(&vp->v_mount);
861 		if (__predict_false(mp == NULL)) {
862 			fdrop(fp, td);
863 			error = EBADF;
864 			break;
865 		}
866 		td->td_retval[0] = 0;
867 		if (mp->mnt_kern_flag & MNTK_UNIONFS ||
868 		    mp->mnt_flag & MNT_UNION)
869 			td->td_retval[0] = 1;
870 		fdrop(fp, td);
871 		break;
872 
873 	case F_KINFO:
874 #ifdef CAPABILITY_MODE
875 		if (IN_CAPABILITY_MODE(td)) {
876 			error = ECAPMODE;
877 			break;
878 		}
879 #endif
880 		error = copyin((void *)arg, &kif_sz, sizeof(kif_sz));
881 		if (error != 0)
882 			break;
883 		if (kif_sz != sizeof(*kif)) {
884 			error = EINVAL;
885 			break;
886 		}
887 		kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK | M_ZERO);
888 		FILEDESC_SLOCK(fdp);
889 		error = fget_cap_noref(fdp, fd, &cap_fcntl_rights, &fp, NULL);
890 		if (error == 0 && fhold(fp)) {
891 			export_file_to_kinfo(fp, fd, NULL, kif, fdp, 0);
892 			FILEDESC_SUNLOCK(fdp);
893 			fdrop(fp, td);
894 			if ((kif->kf_status & KF_ATTR_VALID) != 0) {
895 				kif->kf_structsize = sizeof(*kif);
896 				error = copyout(kif, (void *)arg, sizeof(*kif));
897 			} else {
898 				error = EBADF;
899 			}
900 		} else {
901 			FILEDESC_SUNLOCK(fdp);
902 			if (error == 0)
903 				error = EBADF;
904 		}
905 		free(kif, M_TEMP);
906 		break;
907 
908 	default:
909 		error = EINVAL;
910 		break;
911 	}
912 	return (error);
913 }
914 
915 static int
916 getmaxfd(struct thread *td)
917 {
918 
919 	return (min((int)lim_cur(td, RLIMIT_NOFILE), maxfilesperproc));
920 }
921 
922 /*
923  * Common code for dup, dup2, fcntl(F_DUPFD) and fcntl(F_DUP2FD).
924  */
925 int
926 kern_dup(struct thread *td, u_int mode, int flags, int old, int new)
927 {
928 	struct filedesc *fdp;
929 	struct filedescent *oldfde, *newfde;
930 	struct proc *p;
931 	struct file *delfp, *oldfp;
932 	u_long *oioctls, *nioctls;
933 	int error, maxfd;
934 
935 	p = td->td_proc;
936 	fdp = p->p_fd;
937 	oioctls = NULL;
938 
939 	MPASS((flags & ~(FDDUP_FLAG_CLOEXEC)) == 0);
940 	MPASS(mode < FDDUP_LASTMODE);
941 
942 	AUDIT_ARG_FD(old);
943 	/* XXXRW: if (flags & FDDUP_FIXED) AUDIT_ARG_FD2(new); */
944 
945 	/*
946 	 * Verify we have a valid descriptor to dup from and possibly to
947 	 * dup to. Unlike dup() and dup2(), fcntl()'s F_DUPFD should
948 	 * return EINVAL when the new descriptor is out of bounds.
949 	 */
950 	if (old < 0)
951 		return (EBADF);
952 	if (new < 0)
953 		return (mode == FDDUP_FCNTL ? EINVAL : EBADF);
954 	maxfd = getmaxfd(td);
955 	if (new >= maxfd)
956 		return (mode == FDDUP_FCNTL ? EINVAL : EBADF);
957 
958 	error = EBADF;
959 	FILEDESC_XLOCK(fdp);
960 	if (fget_noref(fdp, old) == NULL)
961 		goto unlock;
962 	if (mode == FDDUP_FIXED && old == new) {
963 		td->td_retval[0] = new;
964 		if (flags & FDDUP_FLAG_CLOEXEC)
965 			fdp->fd_ofiles[new].fde_flags |= UF_EXCLOSE;
966 		error = 0;
967 		goto unlock;
968 	}
969 
970 	oldfde = &fdp->fd_ofiles[old];
971 	oldfp = oldfde->fde_file;
972 	if (!fhold(oldfp))
973 		goto unlock;
974 
975 	/*
976 	 * If the caller specified a file descriptor, make sure the file
977 	 * table is large enough to hold it, and grab it.  Otherwise, just
978 	 * allocate a new descriptor the usual way.
979 	 */
980 	switch (mode) {
981 	case FDDUP_NORMAL:
982 	case FDDUP_FCNTL:
983 		if ((error = fdalloc(td, new, &new)) != 0) {
984 			fdrop(oldfp, td);
985 			goto unlock;
986 		}
987 		break;
988 	case FDDUP_FIXED:
989 		if (new >= fdp->fd_nfiles) {
990 			/*
991 			 * The resource limits are here instead of e.g.
992 			 * fdalloc(), because the file descriptor table may be
993 			 * shared between processes, so we can't really use
994 			 * racct_add()/racct_sub().  Instead of counting the
995 			 * number of actually allocated descriptors, just put
996 			 * the limit on the size of the file descriptor table.
997 			 */
998 #ifdef RACCT
999 			if (RACCT_ENABLED()) {
1000 				error = racct_set_unlocked(p, RACCT_NOFILE, new + 1);
1001 				if (error != 0) {
1002 					error = EMFILE;
1003 					fdrop(oldfp, td);
1004 					goto unlock;
1005 				}
1006 			}
1007 #endif
1008 			fdgrowtable_exp(fdp, new + 1);
1009 		}
1010 		if (!fdisused(fdp, new))
1011 			fdused(fdp, new);
1012 		break;
1013 	default:
1014 		KASSERT(0, ("%s unsupported mode %d", __func__, mode));
1015 	}
1016 
1017 	KASSERT(old != new, ("new fd is same as old"));
1018 
1019 	/* Refetch oldfde because the table may have grown and old one freed. */
1020 	oldfde = &fdp->fd_ofiles[old];
1021 	KASSERT(oldfp == oldfde->fde_file,
1022 	    ("fdt_ofiles shift from growth observed at fd %d",
1023 	    old));
1024 
1025 	newfde = &fdp->fd_ofiles[new];
1026 	delfp = newfde->fde_file;
1027 
1028 	nioctls = filecaps_copy_prep(&oldfde->fde_caps);
1029 
1030 	/*
1031 	 * Duplicate the source descriptor.
1032 	 */
1033 #ifdef CAPABILITIES
1034 	seqc_write_begin(&newfde->fde_seqc);
1035 #endif
1036 	oioctls = filecaps_free_prep(&newfde->fde_caps);
1037 	fde_copy(oldfde, newfde);
1038 	filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps,
1039 	    nioctls);
1040 	if ((flags & FDDUP_FLAG_CLOEXEC) != 0)
1041 		newfde->fde_flags = oldfde->fde_flags | UF_EXCLOSE;
1042 	else
1043 		newfde->fde_flags = oldfde->fde_flags & ~UF_EXCLOSE;
1044 #ifdef CAPABILITIES
1045 	seqc_write_end(&newfde->fde_seqc);
1046 #endif
1047 	td->td_retval[0] = new;
1048 
1049 	error = 0;
1050 
1051 	if (delfp != NULL) {
1052 		(void) closefp(fdp, new, delfp, td, true, false);
1053 		FILEDESC_UNLOCK_ASSERT(fdp);
1054 	} else {
1055 unlock:
1056 		FILEDESC_XUNLOCK(fdp);
1057 	}
1058 
1059 	filecaps_free_finish(oioctls);
1060 	return (error);
1061 }
1062 
1063 static void
1064 sigiofree(struct sigio *sigio)
1065 {
1066 	crfree(sigio->sio_ucred);
1067 	free(sigio, M_SIGIO);
1068 }
1069 
1070 static struct sigio *
1071 funsetown_locked(struct sigio *sigio)
1072 {
1073 	struct proc *p;
1074 	struct pgrp *pg;
1075 
1076 	SIGIO_ASSERT_LOCKED();
1077 
1078 	if (sigio == NULL)
1079 		return (NULL);
1080 	*sigio->sio_myref = NULL;
1081 	if (sigio->sio_pgid < 0) {
1082 		pg = sigio->sio_pgrp;
1083 		PGRP_LOCK(pg);
1084 		SLIST_REMOVE(&pg->pg_sigiolst, sigio, sigio, sio_pgsigio);
1085 		PGRP_UNLOCK(pg);
1086 	} else {
1087 		p = sigio->sio_proc;
1088 		PROC_LOCK(p);
1089 		SLIST_REMOVE(&p->p_sigiolst, sigio, sigio, sio_pgsigio);
1090 		PROC_UNLOCK(p);
1091 	}
1092 	return (sigio);
1093 }
1094 
1095 /*
1096  * If sigio is on the list associated with a process or process group,
1097  * disable signalling from the device, remove sigio from the list and
1098  * free sigio.
1099  */
1100 void
1101 funsetown(struct sigio **sigiop)
1102 {
1103 	struct sigio *sigio;
1104 
1105 	/* Racy check, consumers must provide synchronization. */
1106 	if (*sigiop == NULL)
1107 		return;
1108 
1109 	SIGIO_LOCK();
1110 	sigio = funsetown_locked(*sigiop);
1111 	SIGIO_UNLOCK();
1112 	if (sigio != NULL)
1113 		sigiofree(sigio);
1114 }
1115 
1116 /*
1117  * Free a list of sigio structures.  The caller must ensure that new sigio
1118  * structures cannot be added after this point.  For process groups this is
1119  * guaranteed using the proctree lock; for processes, the P_WEXIT flag serves
1120  * as an interlock.
1121  */
1122 void
1123 funsetownlst(struct sigiolst *sigiolst)
1124 {
1125 	struct proc *p;
1126 	struct pgrp *pg;
1127 	struct sigio *sigio, *tmp;
1128 
1129 	/* Racy check. */
1130 	sigio = SLIST_FIRST(sigiolst);
1131 	if (sigio == NULL)
1132 		return;
1133 
1134 	p = NULL;
1135 	pg = NULL;
1136 
1137 	SIGIO_LOCK();
1138 	sigio = SLIST_FIRST(sigiolst);
1139 	if (sigio == NULL) {
1140 		SIGIO_UNLOCK();
1141 		return;
1142 	}
1143 
1144 	/*
1145 	 * Every entry of the list should belong to a single proc or pgrp.
1146 	 */
1147 	if (sigio->sio_pgid < 0) {
1148 		pg = sigio->sio_pgrp;
1149 		sx_assert(&proctree_lock, SX_XLOCKED);
1150 		PGRP_LOCK(pg);
1151 	} else /* if (sigio->sio_pgid > 0) */ {
1152 		p = sigio->sio_proc;
1153 		PROC_LOCK(p);
1154 		KASSERT((p->p_flag & P_WEXIT) != 0,
1155 		    ("%s: process %p is not exiting", __func__, p));
1156 	}
1157 
1158 	SLIST_FOREACH(sigio, sigiolst, sio_pgsigio) {
1159 		*sigio->sio_myref = NULL;
1160 		if (pg != NULL) {
1161 			KASSERT(sigio->sio_pgid < 0,
1162 			    ("Proc sigio in pgrp sigio list"));
1163 			KASSERT(sigio->sio_pgrp == pg,
1164 			    ("Bogus pgrp in sigio list"));
1165 		} else /* if (p != NULL) */ {
1166 			KASSERT(sigio->sio_pgid > 0,
1167 			    ("Pgrp sigio in proc sigio list"));
1168 			KASSERT(sigio->sio_proc == p,
1169 			    ("Bogus proc in sigio list"));
1170 		}
1171 	}
1172 
1173 	if (pg != NULL)
1174 		PGRP_UNLOCK(pg);
1175 	else
1176 		PROC_UNLOCK(p);
1177 	SIGIO_UNLOCK();
1178 
1179 	SLIST_FOREACH_SAFE(sigio, sigiolst, sio_pgsigio, tmp)
1180 		sigiofree(sigio);
1181 }
1182 
1183 /*
1184  * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg).
1185  *
1186  * After permission checking, add a sigio structure to the sigio list for
1187  * the process or process group.
1188  */
1189 int
1190 fsetown(pid_t pgid, struct sigio **sigiop)
1191 {
1192 	struct proc *proc;
1193 	struct pgrp *pgrp;
1194 	struct sigio *osigio, *sigio;
1195 	int ret;
1196 
1197 	if (pgid == 0) {
1198 		funsetown(sigiop);
1199 		return (0);
1200 	}
1201 
1202 	sigio = malloc(sizeof(struct sigio), M_SIGIO, M_WAITOK);
1203 	sigio->sio_pgid = pgid;
1204 	sigio->sio_ucred = crhold(curthread->td_ucred);
1205 	sigio->sio_myref = sigiop;
1206 
1207 	ret = 0;
1208 	if (pgid > 0) {
1209 		ret = pget(pgid, PGET_NOTWEXIT | PGET_NOTID | PGET_HOLD, &proc);
1210 		SIGIO_LOCK();
1211 		osigio = funsetown_locked(*sigiop);
1212 		if (ret == 0) {
1213 			PROC_LOCK(proc);
1214 			_PRELE(proc);
1215 			if ((proc->p_flag & P_WEXIT) != 0) {
1216 				ret = ESRCH;
1217 			} else if (proc->p_session !=
1218 			    curthread->td_proc->p_session) {
1219 				/*
1220 				 * Policy - Don't allow a process to FSETOWN a
1221 				 * process in another session.
1222 				 *
1223 				 * Remove this test to allow maximum flexibility
1224 				 * or restrict FSETOWN to the current process or
1225 				 * process group for maximum safety.
1226 				 */
1227 				ret = EPERM;
1228 			} else {
1229 				sigio->sio_proc = proc;
1230 				SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio,
1231 				    sio_pgsigio);
1232 			}
1233 			PROC_UNLOCK(proc);
1234 		}
1235 	} else /* if (pgid < 0) */ {
1236 		sx_slock(&proctree_lock);
1237 		SIGIO_LOCK();
1238 		osigio = funsetown_locked(*sigiop);
1239 		pgrp = pgfind(-pgid);
1240 		if (pgrp == NULL) {
1241 			ret = ESRCH;
1242 		} else {
1243 			if (pgrp->pg_session != curthread->td_proc->p_session) {
1244 				/*
1245 				 * Policy - Don't allow a process to FSETOWN a
1246 				 * process in another session.
1247 				 *
1248 				 * Remove this test to allow maximum flexibility
1249 				 * or restrict FSETOWN to the current process or
1250 				 * process group for maximum safety.
1251 				 */
1252 				ret = EPERM;
1253 			} else {
1254 				sigio->sio_pgrp = pgrp;
1255 				SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio,
1256 				    sio_pgsigio);
1257 			}
1258 			PGRP_UNLOCK(pgrp);
1259 		}
1260 		sx_sunlock(&proctree_lock);
1261 	}
1262 	if (ret == 0)
1263 		*sigiop = sigio;
1264 	SIGIO_UNLOCK();
1265 	if (osigio != NULL)
1266 		sigiofree(osigio);
1267 	return (ret);
1268 }
1269 
1270 /*
1271  * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg).
1272  */
1273 pid_t
1274 fgetown(struct sigio **sigiop)
1275 {
1276 	pid_t pgid;
1277 
1278 	SIGIO_LOCK();
1279 	pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0;
1280 	SIGIO_UNLOCK();
1281 	return (pgid);
1282 }
1283 
1284 static int
1285 closefp_impl(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1286     bool audit)
1287 {
1288 	int error;
1289 
1290 	FILEDESC_XLOCK_ASSERT(fdp);
1291 
1292 	/*
1293 	 * We now hold the fp reference that used to be owned by the
1294 	 * descriptor array.  We have to unlock the FILEDESC *AFTER*
1295 	 * knote_fdclose to prevent a race of the fd getting opened, a knote
1296 	 * added, and deleteing a knote for the new fd.
1297 	 */
1298 	if (__predict_false(!TAILQ_EMPTY(&fdp->fd_kqlist)))
1299 		knote_fdclose(td, fd);
1300 
1301 	/*
1302 	 * We need to notify mqueue if the object is of type mqueue.
1303 	 */
1304 	if (__predict_false(fp->f_type == DTYPE_MQUEUE))
1305 		mq_fdclose(td, fd, fp);
1306 	FILEDESC_XUNLOCK(fdp);
1307 
1308 #ifdef AUDIT
1309 	if (AUDITING_TD(td) && audit)
1310 		audit_sysclose(td, fd, fp);
1311 #endif
1312 	error = closef(fp, td);
1313 
1314 	/*
1315 	 * All paths leading up to closefp() will have already removed or
1316 	 * replaced the fd in the filedesc table, so a restart would not
1317 	 * operate on the same file.
1318 	 */
1319 	if (error == ERESTART)
1320 		error = EINTR;
1321 
1322 	return (error);
1323 }
1324 
1325 static int
1326 closefp_hl(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1327     bool holdleaders, bool audit)
1328 {
1329 	int error;
1330 
1331 	FILEDESC_XLOCK_ASSERT(fdp);
1332 
1333 	if (holdleaders) {
1334 		if (td->td_proc->p_fdtol != NULL) {
1335 			/*
1336 			 * Ask fdfree() to sleep to ensure that all relevant
1337 			 * process leaders can be traversed in closef().
1338 			 */
1339 			fdp->fd_holdleaderscount++;
1340 		} else {
1341 			holdleaders = false;
1342 		}
1343 	}
1344 
1345 	error = closefp_impl(fdp, fd, fp, td, audit);
1346 	if (holdleaders) {
1347 		FILEDESC_XLOCK(fdp);
1348 		fdp->fd_holdleaderscount--;
1349 		if (fdp->fd_holdleaderscount == 0 &&
1350 		    fdp->fd_holdleaderswakeup != 0) {
1351 			fdp->fd_holdleaderswakeup = 0;
1352 			wakeup(&fdp->fd_holdleaderscount);
1353 		}
1354 		FILEDESC_XUNLOCK(fdp);
1355 	}
1356 	return (error);
1357 }
1358 
1359 static int
1360 closefp(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1361     bool holdleaders, bool audit)
1362 {
1363 
1364 	FILEDESC_XLOCK_ASSERT(fdp);
1365 
1366 	if (__predict_false(td->td_proc->p_fdtol != NULL)) {
1367 		return (closefp_hl(fdp, fd, fp, td, holdleaders, audit));
1368 	} else {
1369 		return (closefp_impl(fdp, fd, fp, td, audit));
1370 	}
1371 }
1372 
1373 /*
1374  * Close a file descriptor.
1375  */
1376 #ifndef _SYS_SYSPROTO_H_
1377 struct close_args {
1378 	int     fd;
1379 };
1380 #endif
1381 /* ARGSUSED */
1382 int
1383 sys_close(struct thread *td, struct close_args *uap)
1384 {
1385 
1386 	return (kern_close(td, uap->fd));
1387 }
1388 
1389 int
1390 kern_close(struct thread *td, int fd)
1391 {
1392 	struct filedesc *fdp;
1393 	struct file *fp;
1394 
1395 	fdp = td->td_proc->p_fd;
1396 
1397 	FILEDESC_XLOCK(fdp);
1398 	if ((fp = fget_noref(fdp, fd)) == NULL) {
1399 		FILEDESC_XUNLOCK(fdp);
1400 		return (EBADF);
1401 	}
1402 	fdfree(fdp, fd);
1403 
1404 	/* closefp() drops the FILEDESC lock for us. */
1405 	return (closefp(fdp, fd, fp, td, true, true));
1406 }
1407 
1408 static int
1409 close_range_cloexec(struct thread *td, u_int lowfd, u_int highfd)
1410 {
1411 	struct filedesc *fdp;
1412 	struct fdescenttbl *fdt;
1413 	struct filedescent *fde;
1414 	int fd;
1415 
1416 	fdp = td->td_proc->p_fd;
1417 	FILEDESC_XLOCK(fdp);
1418 	fdt = atomic_load_ptr(&fdp->fd_files);
1419 	highfd = MIN(highfd, fdt->fdt_nfiles - 1);
1420 	fd = lowfd;
1421 	if (__predict_false(fd > highfd)) {
1422 		goto out_locked;
1423 	}
1424 	for (; fd <= highfd; fd++) {
1425 		fde = &fdt->fdt_ofiles[fd];
1426 		if (fde->fde_file != NULL)
1427 			fde->fde_flags |= UF_EXCLOSE;
1428 	}
1429 out_locked:
1430 	FILEDESC_XUNLOCK(fdp);
1431 	return (0);
1432 }
1433 
1434 static int
1435 close_range_impl(struct thread *td, u_int lowfd, u_int highfd)
1436 {
1437 	struct filedesc *fdp;
1438 	const struct fdescenttbl *fdt;
1439 	struct file *fp;
1440 	int fd;
1441 
1442 	fdp = td->td_proc->p_fd;
1443 	FILEDESC_XLOCK(fdp);
1444 	fdt = atomic_load_ptr(&fdp->fd_files);
1445 	highfd = MIN(highfd, fdt->fdt_nfiles - 1);
1446 	fd = lowfd;
1447 	if (__predict_false(fd > highfd)) {
1448 		goto out_locked;
1449 	}
1450 	for (;;) {
1451 		fp = fdt->fdt_ofiles[fd].fde_file;
1452 		if (fp == NULL) {
1453 			if (fd == highfd)
1454 				goto out_locked;
1455 		} else {
1456 			fdfree(fdp, fd);
1457 			(void) closefp(fdp, fd, fp, td, true, true);
1458 			if (fd == highfd)
1459 				goto out_unlocked;
1460 			FILEDESC_XLOCK(fdp);
1461 			fdt = atomic_load_ptr(&fdp->fd_files);
1462 		}
1463 		fd++;
1464 	}
1465 out_locked:
1466 	FILEDESC_XUNLOCK(fdp);
1467 out_unlocked:
1468 	return (0);
1469 }
1470 
1471 int
1472 kern_close_range(struct thread *td, int flags, u_int lowfd, u_int highfd)
1473 {
1474 
1475 	/*
1476 	 * Check this prior to clamping; closefrom(3) with only fd 0, 1, and 2
1477 	 * open should not be a usage error.  From a close_range() perspective,
1478 	 * close_range(3, ~0U, 0) in the same scenario should also likely not
1479 	 * be a usage error as all fd above 3 are in-fact already closed.
1480 	 */
1481 	if (highfd < lowfd) {
1482 		return (EINVAL);
1483 	}
1484 
1485 	if ((flags & CLOSE_RANGE_CLOEXEC) != 0)
1486 		return (close_range_cloexec(td, lowfd, highfd));
1487 
1488 	return (close_range_impl(td, lowfd, highfd));
1489 }
1490 
1491 #ifndef _SYS_SYSPROTO_H_
1492 struct close_range_args {
1493 	u_int	lowfd;
1494 	u_int	highfd;
1495 	int	flags;
1496 };
1497 #endif
1498 int
1499 sys_close_range(struct thread *td, struct close_range_args *uap)
1500 {
1501 
1502 	AUDIT_ARG_FD(uap->lowfd);
1503 	AUDIT_ARG_CMD(uap->highfd);
1504 	AUDIT_ARG_FFLAGS(uap->flags);
1505 
1506 	if ((uap->flags & ~(CLOSE_RANGE_CLOEXEC)) != 0)
1507 		return (EINVAL);
1508 	return (kern_close_range(td, uap->flags, uap->lowfd, uap->highfd));
1509 }
1510 
1511 #ifdef COMPAT_FREEBSD12
1512 /*
1513  * Close open file descriptors.
1514  */
1515 #ifndef _SYS_SYSPROTO_H_
1516 struct freebsd12_closefrom_args {
1517 	int	lowfd;
1518 };
1519 #endif
1520 /* ARGSUSED */
1521 int
1522 freebsd12_closefrom(struct thread *td, struct freebsd12_closefrom_args *uap)
1523 {
1524 	u_int lowfd;
1525 
1526 	AUDIT_ARG_FD(uap->lowfd);
1527 
1528 	/*
1529 	 * Treat negative starting file descriptor values identical to
1530 	 * closefrom(0) which closes all files.
1531 	 */
1532 	lowfd = MAX(0, uap->lowfd);
1533 	return (kern_close_range(td, 0, lowfd, ~0U));
1534 }
1535 #endif	/* COMPAT_FREEBSD12 */
1536 
1537 #if defined(COMPAT_43)
1538 /*
1539  * Return status information about a file descriptor.
1540  */
1541 #ifndef _SYS_SYSPROTO_H_
1542 struct ofstat_args {
1543 	int	fd;
1544 	struct	ostat *sb;
1545 };
1546 #endif
1547 /* ARGSUSED */
1548 int
1549 ofstat(struct thread *td, struct ofstat_args *uap)
1550 {
1551 	struct ostat oub;
1552 	struct stat ub;
1553 	int error;
1554 
1555 	error = kern_fstat(td, uap->fd, &ub);
1556 	if (error == 0) {
1557 		cvtstat(&ub, &oub);
1558 		error = copyout(&oub, uap->sb, sizeof(oub));
1559 	}
1560 	return (error);
1561 }
1562 #endif /* COMPAT_43 */
1563 
1564 #if defined(COMPAT_FREEBSD11)
1565 int
1566 freebsd11_fstat(struct thread *td, struct freebsd11_fstat_args *uap)
1567 {
1568 	struct stat sb;
1569 	struct freebsd11_stat osb;
1570 	int error;
1571 
1572 	error = kern_fstat(td, uap->fd, &sb);
1573 	if (error != 0)
1574 		return (error);
1575 	error = freebsd11_cvtstat(&sb, &osb);
1576 	if (error == 0)
1577 		error = copyout(&osb, uap->sb, sizeof(osb));
1578 	return (error);
1579 }
1580 #endif	/* COMPAT_FREEBSD11 */
1581 
1582 /*
1583  * Return status information about a file descriptor.
1584  */
1585 #ifndef _SYS_SYSPROTO_H_
1586 struct fstat_args {
1587 	int	fd;
1588 	struct	stat *sb;
1589 };
1590 #endif
1591 /* ARGSUSED */
1592 int
1593 sys_fstat(struct thread *td, struct fstat_args *uap)
1594 {
1595 	struct stat ub;
1596 	int error;
1597 
1598 	error = kern_fstat(td, uap->fd, &ub);
1599 	if (error == 0)
1600 		error = copyout(&ub, uap->sb, sizeof(ub));
1601 	return (error);
1602 }
1603 
1604 int
1605 kern_fstat(struct thread *td, int fd, struct stat *sbp)
1606 {
1607 	struct file *fp;
1608 	int error;
1609 
1610 	AUDIT_ARG_FD(fd);
1611 
1612 	error = fget(td, fd, &cap_fstat_rights, &fp);
1613 	if (__predict_false(error != 0))
1614 		return (error);
1615 
1616 	AUDIT_ARG_FILE(td->td_proc, fp);
1617 
1618 	error = fo_stat(fp, sbp, td->td_ucred);
1619 	fdrop(fp, td);
1620 #ifdef __STAT_TIME_T_EXT
1621 	sbp->st_atim_ext = 0;
1622 	sbp->st_mtim_ext = 0;
1623 	sbp->st_ctim_ext = 0;
1624 	sbp->st_btim_ext = 0;
1625 #endif
1626 #ifdef KTRACE
1627 	if (KTRPOINT(td, KTR_STRUCT))
1628 		ktrstat_error(sbp, error);
1629 #endif
1630 	return (error);
1631 }
1632 
1633 #if defined(COMPAT_FREEBSD11)
1634 /*
1635  * Return status information about a file descriptor.
1636  */
1637 #ifndef _SYS_SYSPROTO_H_
1638 struct freebsd11_nfstat_args {
1639 	int	fd;
1640 	struct	nstat *sb;
1641 };
1642 #endif
1643 /* ARGSUSED */
1644 int
1645 freebsd11_nfstat(struct thread *td, struct freebsd11_nfstat_args *uap)
1646 {
1647 	struct nstat nub;
1648 	struct stat ub;
1649 	int error;
1650 
1651 	error = kern_fstat(td, uap->fd, &ub);
1652 	if (error != 0)
1653 		return (error);
1654 	error = freebsd11_cvtnstat(&ub, &nub);
1655 	if (error != 0)
1656 		error = copyout(&nub, uap->sb, sizeof(nub));
1657 	return (error);
1658 }
1659 #endif /* COMPAT_FREEBSD11 */
1660 
1661 /*
1662  * Return pathconf information about a file descriptor.
1663  */
1664 #ifndef _SYS_SYSPROTO_H_
1665 struct fpathconf_args {
1666 	int	fd;
1667 	int	name;
1668 };
1669 #endif
1670 /* ARGSUSED */
1671 int
1672 sys_fpathconf(struct thread *td, struct fpathconf_args *uap)
1673 {
1674 	long value;
1675 	int error;
1676 
1677 	error = kern_fpathconf(td, uap->fd, uap->name, &value);
1678 	if (error == 0)
1679 		td->td_retval[0] = value;
1680 	return (error);
1681 }
1682 
1683 int
1684 kern_fpathconf(struct thread *td, int fd, int name, long *valuep)
1685 {
1686 	struct file *fp;
1687 	struct vnode *vp;
1688 	int error;
1689 
1690 	error = fget(td, fd, &cap_fpathconf_rights, &fp);
1691 	if (error != 0)
1692 		return (error);
1693 
1694 	if (name == _PC_ASYNC_IO) {
1695 		*valuep = _POSIX_ASYNCHRONOUS_IO;
1696 		goto out;
1697 	}
1698 	vp = fp->f_vnode;
1699 	if (vp != NULL) {
1700 		vn_lock(vp, LK_SHARED | LK_RETRY);
1701 		error = VOP_PATHCONF(vp, name, valuep);
1702 		VOP_UNLOCK(vp);
1703 	} else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) {
1704 		if (name != _PC_PIPE_BUF) {
1705 			error = EINVAL;
1706 		} else {
1707 			*valuep = PIPE_BUF;
1708 			error = 0;
1709 		}
1710 	} else {
1711 		error = EOPNOTSUPP;
1712 	}
1713 out:
1714 	fdrop(fp, td);
1715 	return (error);
1716 }
1717 
1718 /*
1719  * Copy filecaps structure allocating memory for ioctls array if needed.
1720  *
1721  * The last parameter indicates whether the fdtable is locked. If it is not and
1722  * ioctls are encountered, copying fails and the caller must lock the table.
1723  *
1724  * Note that if the table was not locked, the caller has to check the relevant
1725  * sequence counter to determine whether the operation was successful.
1726  */
1727 bool
1728 filecaps_copy(const struct filecaps *src, struct filecaps *dst, bool locked)
1729 {
1730 	size_t size;
1731 
1732 	if (src->fc_ioctls != NULL && !locked)
1733 		return (false);
1734 	memcpy(dst, src, sizeof(*src));
1735 	if (src->fc_ioctls == NULL)
1736 		return (true);
1737 
1738 	KASSERT(src->fc_nioctls > 0,
1739 	    ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1740 
1741 	size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1742 	dst->fc_ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1743 	memcpy(dst->fc_ioctls, src->fc_ioctls, size);
1744 	return (true);
1745 }
1746 
1747 static u_long *
1748 filecaps_copy_prep(const struct filecaps *src)
1749 {
1750 	u_long *ioctls;
1751 	size_t size;
1752 
1753 	if (__predict_true(src->fc_ioctls == NULL))
1754 		return (NULL);
1755 
1756 	KASSERT(src->fc_nioctls > 0,
1757 	    ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1758 
1759 	size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1760 	ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1761 	return (ioctls);
1762 }
1763 
1764 static void
1765 filecaps_copy_finish(const struct filecaps *src, struct filecaps *dst,
1766     u_long *ioctls)
1767 {
1768 	size_t size;
1769 
1770 	*dst = *src;
1771 	if (__predict_true(src->fc_ioctls == NULL)) {
1772 		MPASS(ioctls == NULL);
1773 		return;
1774 	}
1775 
1776 	size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1777 	dst->fc_ioctls = ioctls;
1778 	bcopy(src->fc_ioctls, dst->fc_ioctls, size);
1779 }
1780 
1781 /*
1782  * Move filecaps structure to the new place and clear the old place.
1783  */
1784 void
1785 filecaps_move(struct filecaps *src, struct filecaps *dst)
1786 {
1787 
1788 	*dst = *src;
1789 	bzero(src, sizeof(*src));
1790 }
1791 
1792 /*
1793  * Fill the given filecaps structure with full rights.
1794  */
1795 static void
1796 filecaps_fill(struct filecaps *fcaps)
1797 {
1798 
1799 	CAP_ALL(&fcaps->fc_rights);
1800 	fcaps->fc_ioctls = NULL;
1801 	fcaps->fc_nioctls = -1;
1802 	fcaps->fc_fcntls = CAP_FCNTL_ALL;
1803 }
1804 
1805 /*
1806  * Free memory allocated within filecaps structure.
1807  */
1808 static void
1809 filecaps_free_ioctl(struct filecaps *fcaps)
1810 {
1811 
1812 	free(fcaps->fc_ioctls, M_FILECAPS);
1813 	fcaps->fc_ioctls = NULL;
1814 }
1815 
1816 void
1817 filecaps_free(struct filecaps *fcaps)
1818 {
1819 
1820 	filecaps_free_ioctl(fcaps);
1821 	bzero(fcaps, sizeof(*fcaps));
1822 }
1823 
1824 static u_long *
1825 filecaps_free_prep(struct filecaps *fcaps)
1826 {
1827 	u_long *ioctls;
1828 
1829 	ioctls = fcaps->fc_ioctls;
1830 	bzero(fcaps, sizeof(*fcaps));
1831 	return (ioctls);
1832 }
1833 
1834 static void
1835 filecaps_free_finish(u_long *ioctls)
1836 {
1837 
1838 	free(ioctls, M_FILECAPS);
1839 }
1840 
1841 /*
1842  * Validate the given filecaps structure.
1843  */
1844 static void
1845 filecaps_validate(const struct filecaps *fcaps, const char *func)
1846 {
1847 
1848 	KASSERT(cap_rights_is_valid(&fcaps->fc_rights),
1849 	    ("%s: invalid rights", func));
1850 	KASSERT((fcaps->fc_fcntls & ~CAP_FCNTL_ALL) == 0,
1851 	    ("%s: invalid fcntls", func));
1852 	KASSERT(fcaps->fc_fcntls == 0 ||
1853 	    cap_rights_is_set(&fcaps->fc_rights, CAP_FCNTL),
1854 	    ("%s: fcntls without CAP_FCNTL", func));
1855 	/*
1856 	 * open calls without WANTIOCTLCAPS free caps but leave the counter
1857 	 */
1858 #if 0
1859 	KASSERT(fcaps->fc_ioctls != NULL ? fcaps->fc_nioctls > 0 :
1860 	    (fcaps->fc_nioctls == -1 || fcaps->fc_nioctls == 0),
1861 	    ("%s: invalid ioctls", func));
1862 #endif
1863 	KASSERT(fcaps->fc_nioctls == 0 ||
1864 	    cap_rights_is_set(&fcaps->fc_rights, CAP_IOCTL),
1865 	    ("%s: ioctls without CAP_IOCTL", func));
1866 }
1867 
1868 static void
1869 fdgrowtable_exp(struct filedesc *fdp, int nfd)
1870 {
1871 	int nfd1;
1872 
1873 	FILEDESC_XLOCK_ASSERT(fdp);
1874 
1875 	nfd1 = fdp->fd_nfiles * 2;
1876 	if (nfd1 < nfd)
1877 		nfd1 = nfd;
1878 	fdgrowtable(fdp, nfd1);
1879 }
1880 
1881 /*
1882  * Grow the file table to accommodate (at least) nfd descriptors.
1883  */
1884 static void
1885 fdgrowtable(struct filedesc *fdp, int nfd)
1886 {
1887 	struct filedesc0 *fdp0;
1888 	struct freetable *ft;
1889 	struct fdescenttbl *ntable;
1890 	struct fdescenttbl *otable;
1891 	int nnfiles, onfiles;
1892 	NDSLOTTYPE *nmap, *omap;
1893 
1894 	KASSERT(fdp->fd_nfiles > 0, ("zero-length file table"));
1895 
1896 	/* save old values */
1897 	onfiles = fdp->fd_nfiles;
1898 	otable = fdp->fd_files;
1899 	omap = fdp->fd_map;
1900 
1901 	/* compute the size of the new table */
1902 	nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */
1903 	if (nnfiles <= onfiles)
1904 		/* the table is already large enough */
1905 		return;
1906 
1907 	/*
1908 	 * Allocate a new table.  We need enough space for the number of
1909 	 * entries, file entries themselves and the struct freetable we will use
1910 	 * when we decommission the table and place it on the freelist.
1911 	 * We place the struct freetable in the middle so we don't have
1912 	 * to worry about padding.
1913 	 */
1914 	ntable = malloc(offsetof(struct fdescenttbl, fdt_ofiles) +
1915 	    nnfiles * sizeof(ntable->fdt_ofiles[0]) +
1916 	    sizeof(struct freetable),
1917 	    M_FILEDESC, M_ZERO | M_WAITOK);
1918 	/* copy the old data */
1919 	ntable->fdt_nfiles = nnfiles;
1920 	memcpy(ntable->fdt_ofiles, otable->fdt_ofiles,
1921 	    onfiles * sizeof(ntable->fdt_ofiles[0]));
1922 
1923 	/*
1924 	 * Allocate a new map only if the old is not large enough.  It will
1925 	 * grow at a slower rate than the table as it can map more
1926 	 * entries than the table can hold.
1927 	 */
1928 	if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) {
1929 		nmap = malloc(NDSLOTS(nnfiles) * NDSLOTSIZE, M_FILEDESC,
1930 		    M_ZERO | M_WAITOK);
1931 		/* copy over the old data and update the pointer */
1932 		memcpy(nmap, omap, NDSLOTS(onfiles) * sizeof(*omap));
1933 		fdp->fd_map = nmap;
1934 	}
1935 
1936 	/*
1937 	 * Make sure that ntable is correctly initialized before we replace
1938 	 * fd_files poiner. Otherwise fget_unlocked() may see inconsistent
1939 	 * data.
1940 	 */
1941 	atomic_store_rel_ptr((volatile void *)&fdp->fd_files, (uintptr_t)ntable);
1942 
1943 	/*
1944 	 * Free the old file table when not shared by other threads or processes.
1945 	 * The old file table is considered to be shared when either are true:
1946 	 * - The process has more than one thread.
1947 	 * - The file descriptor table has been shared via fdshare().
1948 	 *
1949 	 * When shared, the old file table will be placed on a freelist
1950 	 * which will be processed when the struct filedesc is released.
1951 	 *
1952 	 * Note that if onfiles == NDFILE, we're dealing with the original
1953 	 * static allocation contained within (struct filedesc0 *)fdp,
1954 	 * which must not be freed.
1955 	 */
1956 	if (onfiles > NDFILE) {
1957 		/*
1958 		 * Note we may be called here from fdinit while allocating a
1959 		 * table for a new process in which case ->p_fd points
1960 		 * elsewhere.
1961 		 */
1962 		if (curproc->p_fd != fdp || FILEDESC_IS_ONLY_USER(fdp)) {
1963 			free(otable, M_FILEDESC);
1964 		} else {
1965 			ft = (struct freetable *)&otable->fdt_ofiles[onfiles];
1966 			fdp0 = (struct filedesc0 *)fdp;
1967 			ft->ft_table = otable;
1968 			SLIST_INSERT_HEAD(&fdp0->fd_free, ft, ft_next);
1969 		}
1970 	}
1971 	/*
1972 	 * The map does not have the same possibility of threads still
1973 	 * holding references to it.  So always free it as long as it
1974 	 * does not reference the original static allocation.
1975 	 */
1976 	if (NDSLOTS(onfiles) > NDSLOTS(NDFILE))
1977 		free(omap, M_FILEDESC);
1978 }
1979 
1980 /*
1981  * Allocate a file descriptor for the process.
1982  */
1983 int
1984 fdalloc(struct thread *td, int minfd, int *result)
1985 {
1986 	struct proc *p = td->td_proc;
1987 	struct filedesc *fdp = p->p_fd;
1988 	int fd, maxfd, allocfd;
1989 #ifdef RACCT
1990 	int error;
1991 #endif
1992 
1993 	FILEDESC_XLOCK_ASSERT(fdp);
1994 
1995 	if (fdp->fd_freefile > minfd)
1996 		minfd = fdp->fd_freefile;
1997 
1998 	maxfd = getmaxfd(td);
1999 
2000 	/*
2001 	 * Search the bitmap for a free descriptor starting at minfd.
2002 	 * If none is found, grow the file table.
2003 	 */
2004 	fd = fd_first_free(fdp, minfd, fdp->fd_nfiles);
2005 	if (__predict_false(fd >= maxfd))
2006 		return (EMFILE);
2007 	if (__predict_false(fd >= fdp->fd_nfiles)) {
2008 		allocfd = min(fd * 2, maxfd);
2009 #ifdef RACCT
2010 		if (RACCT_ENABLED()) {
2011 			error = racct_set_unlocked(p, RACCT_NOFILE, allocfd);
2012 			if (error != 0)
2013 				return (EMFILE);
2014 		}
2015 #endif
2016 		/*
2017 		 * fd is already equal to first free descriptor >= minfd, so
2018 		 * we only need to grow the table and we are done.
2019 		 */
2020 		fdgrowtable_exp(fdp, allocfd);
2021 	}
2022 
2023 	/*
2024 	 * Perform some sanity checks, then mark the file descriptor as
2025 	 * used and return it to the caller.
2026 	 */
2027 	KASSERT(fd >= 0 && fd < min(maxfd, fdp->fd_nfiles),
2028 	    ("invalid descriptor %d", fd));
2029 	KASSERT(!fdisused(fdp, fd),
2030 	    ("fd_first_free() returned non-free descriptor"));
2031 	KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
2032 	    ("file descriptor isn't free"));
2033 	fdused(fdp, fd);
2034 	*result = fd;
2035 	return (0);
2036 }
2037 
2038 /*
2039  * Allocate n file descriptors for the process.
2040  */
2041 int
2042 fdallocn(struct thread *td, int minfd, int *fds, int n)
2043 {
2044 	struct proc *p = td->td_proc;
2045 	struct filedesc *fdp = p->p_fd;
2046 	int i;
2047 
2048 	FILEDESC_XLOCK_ASSERT(fdp);
2049 
2050 	for (i = 0; i < n; i++)
2051 		if (fdalloc(td, 0, &fds[i]) != 0)
2052 			break;
2053 
2054 	if (i < n) {
2055 		for (i--; i >= 0; i--)
2056 			fdunused(fdp, fds[i]);
2057 		return (EMFILE);
2058 	}
2059 
2060 	return (0);
2061 }
2062 
2063 /*
2064  * Create a new open file structure and allocate a file descriptor for the
2065  * process that refers to it.  We add one reference to the file for the
2066  * descriptor table and one reference for resultfp. This is to prevent us
2067  * being preempted and the entry in the descriptor table closed after we
2068  * release the FILEDESC lock.
2069  */
2070 int
2071 falloc_caps(struct thread *td, struct file **resultfp, int *resultfd, int flags,
2072     struct filecaps *fcaps)
2073 {
2074 	struct file *fp;
2075 	int error, fd;
2076 
2077 	MPASS(resultfp != NULL);
2078 	MPASS(resultfd != NULL);
2079 
2080 	error = _falloc_noinstall(td, &fp, 2);
2081 	if (__predict_false(error != 0)) {
2082 		return (error);
2083 	}
2084 
2085 	error = finstall_refed(td, fp, &fd, flags, fcaps);
2086 	if (__predict_false(error != 0)) {
2087 		falloc_abort(td, fp);
2088 		return (error);
2089 	}
2090 
2091 	*resultfp = fp;
2092 	*resultfd = fd;
2093 
2094 	return (0);
2095 }
2096 
2097 /*
2098  * Create a new open file structure without allocating a file descriptor.
2099  */
2100 int
2101 _falloc_noinstall(struct thread *td, struct file **resultfp, u_int n)
2102 {
2103 	struct file *fp;
2104 	int maxuserfiles = maxfiles - (maxfiles / 20);
2105 	int openfiles_new;
2106 	static struct timeval lastfail;
2107 	static int curfail;
2108 
2109 	KASSERT(resultfp != NULL, ("%s: resultfp == NULL", __func__));
2110 	MPASS(n > 0);
2111 
2112 	openfiles_new = atomic_fetchadd_int(&openfiles, 1) + 1;
2113 	if ((openfiles_new >= maxuserfiles &&
2114 	    priv_check(td, PRIV_MAXFILES) != 0) ||
2115 	    openfiles_new >= maxfiles) {
2116 		atomic_subtract_int(&openfiles, 1);
2117 		if (ppsratecheck(&lastfail, &curfail, 1)) {
2118 			printf("kern.maxfiles limit exceeded by uid %i, (%s) "
2119 			    "please see tuning(7).\n", td->td_ucred->cr_ruid, td->td_proc->p_comm);
2120 		}
2121 		return (ENFILE);
2122 	}
2123 	fp = uma_zalloc(file_zone, M_WAITOK);
2124 	bzero(fp, sizeof(*fp));
2125 	refcount_init(&fp->f_count, n);
2126 	fp->f_cred = crhold(td->td_ucred);
2127 	fp->f_ops = &badfileops;
2128 	*resultfp = fp;
2129 	return (0);
2130 }
2131 
2132 void
2133 falloc_abort(struct thread *td, struct file *fp)
2134 {
2135 
2136 	/*
2137 	 * For assertion purposes.
2138 	 */
2139 	refcount_init(&fp->f_count, 0);
2140 	_fdrop(fp, td);
2141 }
2142 
2143 /*
2144  * Install a file in a file descriptor table.
2145  */
2146 void
2147 _finstall(struct filedesc *fdp, struct file *fp, int fd, int flags,
2148     struct filecaps *fcaps)
2149 {
2150 	struct filedescent *fde;
2151 
2152 	MPASS(fp != NULL);
2153 	if (fcaps != NULL)
2154 		filecaps_validate(fcaps, __func__);
2155 	FILEDESC_XLOCK_ASSERT(fdp);
2156 
2157 	fde = &fdp->fd_ofiles[fd];
2158 #ifdef CAPABILITIES
2159 	seqc_write_begin(&fde->fde_seqc);
2160 #endif
2161 	fde->fde_file = fp;
2162 	fde->fde_flags = (flags & O_CLOEXEC) != 0 ? UF_EXCLOSE : 0;
2163 	if (fcaps != NULL)
2164 		filecaps_move(fcaps, &fde->fde_caps);
2165 	else
2166 		filecaps_fill(&fde->fde_caps);
2167 #ifdef CAPABILITIES
2168 	seqc_write_end(&fde->fde_seqc);
2169 #endif
2170 }
2171 
2172 int
2173 finstall_refed(struct thread *td, struct file *fp, int *fd, int flags,
2174     struct filecaps *fcaps)
2175 {
2176 	struct filedesc *fdp = td->td_proc->p_fd;
2177 	int error;
2178 
2179 	MPASS(fd != NULL);
2180 
2181 	FILEDESC_XLOCK(fdp);
2182 	error = fdalloc(td, 0, fd);
2183 	if (__predict_true(error == 0)) {
2184 		_finstall(fdp, fp, *fd, flags, fcaps);
2185 	}
2186 	FILEDESC_XUNLOCK(fdp);
2187 	return (error);
2188 }
2189 
2190 int
2191 finstall(struct thread *td, struct file *fp, int *fd, int flags,
2192     struct filecaps *fcaps)
2193 {
2194 	int error;
2195 
2196 	MPASS(fd != NULL);
2197 
2198 	if (!fhold(fp))
2199 		return (EBADF);
2200 	error = finstall_refed(td, fp, fd, flags, fcaps);
2201 	if (__predict_false(error != 0)) {
2202 		fdrop(fp, td);
2203 	}
2204 	return (error);
2205 }
2206 
2207 /*
2208  * Build a new filedesc structure from another.
2209  *
2210  * If fdp is not NULL, return with it shared locked.
2211  */
2212 struct filedesc *
2213 fdinit(void)
2214 {
2215 	struct filedesc0 *newfdp0;
2216 	struct filedesc *newfdp;
2217 
2218 	newfdp0 = uma_zalloc(filedesc0_zone, M_WAITOK | M_ZERO);
2219 	newfdp = &newfdp0->fd_fd;
2220 
2221 	/* Create the file descriptor table. */
2222 	FILEDESC_LOCK_INIT(newfdp);
2223 	refcount_init(&newfdp->fd_refcnt, 1);
2224 	refcount_init(&newfdp->fd_holdcnt, 1);
2225 	newfdp->fd_map = newfdp0->fd_dmap;
2226 	newfdp->fd_files = (struct fdescenttbl *)&newfdp0->fd_dfiles;
2227 	newfdp->fd_files->fdt_nfiles = NDFILE;
2228 
2229 	return (newfdp);
2230 }
2231 
2232 /*
2233  * Build a pwddesc structure from another.
2234  * Copy the current, root, and jail root vnode references.
2235  *
2236  * If pdp is not NULL, return with it shared locked.
2237  */
2238 struct pwddesc *
2239 pdinit(struct pwddesc *pdp, bool keeplock)
2240 {
2241 	struct pwddesc *newpdp;
2242 	struct pwd *newpwd;
2243 
2244 	newpdp = malloc(sizeof(*newpdp), M_PWDDESC, M_WAITOK | M_ZERO);
2245 
2246 	PWDDESC_LOCK_INIT(newpdp);
2247 	refcount_init(&newpdp->pd_refcount, 1);
2248 	newpdp->pd_cmask = CMASK;
2249 
2250 	if (pdp == NULL) {
2251 		newpwd = pwd_alloc();
2252 		smr_serialized_store(&newpdp->pd_pwd, newpwd, true);
2253 		return (newpdp);
2254 	}
2255 
2256 	PWDDESC_XLOCK(pdp);
2257 	newpwd = pwd_hold_pwddesc(pdp);
2258 	smr_serialized_store(&newpdp->pd_pwd, newpwd, true);
2259 	if (!keeplock)
2260 		PWDDESC_XUNLOCK(pdp);
2261 	return (newpdp);
2262 }
2263 
2264 /*
2265  * Hold either filedesc or pwddesc of the passed process.
2266  *
2267  * The process lock is used to synchronize against the target exiting and
2268  * freeing the data.
2269  *
2270  * Clearing can be ilustrated in 3 steps:
2271  * 1. set the pointer to NULL. Either routine can race against it, hence
2272  *   atomic_load_ptr.
2273  * 2. observe the process lock as not taken. Until then fdhold/pdhold can
2274  *   race to either still see the pointer or find NULL. It is still safe to
2275  *   grab a reference as clearing is stalled.
2276  * 3. after the lock is observed as not taken, any fdhold/pdhold calls are
2277  *   guaranteed to see NULL, making it safe to finish clearing
2278  */
2279 static struct filedesc *
2280 fdhold(struct proc *p)
2281 {
2282 	struct filedesc *fdp;
2283 
2284 	PROC_LOCK_ASSERT(p, MA_OWNED);
2285 	fdp = atomic_load_ptr(&p->p_fd);
2286 	if (fdp != NULL)
2287 		refcount_acquire(&fdp->fd_holdcnt);
2288 	return (fdp);
2289 }
2290 
2291 static struct pwddesc *
2292 pdhold(struct proc *p)
2293 {
2294 	struct pwddesc *pdp;
2295 
2296 	PROC_LOCK_ASSERT(p, MA_OWNED);
2297 	pdp = atomic_load_ptr(&p->p_pd);
2298 	if (pdp != NULL)
2299 		refcount_acquire(&pdp->pd_refcount);
2300 	return (pdp);
2301 }
2302 
2303 static void
2304 fddrop(struct filedesc *fdp)
2305 {
2306 
2307 	if (refcount_load(&fdp->fd_holdcnt) > 1) {
2308 		if (refcount_release(&fdp->fd_holdcnt) == 0)
2309 			return;
2310 	}
2311 
2312 	FILEDESC_LOCK_DESTROY(fdp);
2313 	uma_zfree(filedesc0_zone, fdp);
2314 }
2315 
2316 static void
2317 pddrop(struct pwddesc *pdp)
2318 {
2319 	struct pwd *pwd;
2320 
2321 	if (refcount_release_if_not_last(&pdp->pd_refcount))
2322 		return;
2323 
2324 	PWDDESC_XLOCK(pdp);
2325 	if (refcount_release(&pdp->pd_refcount) == 0) {
2326 		PWDDESC_XUNLOCK(pdp);
2327 		return;
2328 	}
2329 	pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
2330 	pwd_set(pdp, NULL);
2331 	PWDDESC_XUNLOCK(pdp);
2332 	pwd_drop(pwd);
2333 
2334 	PWDDESC_LOCK_DESTROY(pdp);
2335 	free(pdp, M_PWDDESC);
2336 }
2337 
2338 /*
2339  * Share a filedesc structure.
2340  */
2341 struct filedesc *
2342 fdshare(struct filedesc *fdp)
2343 {
2344 
2345 	refcount_acquire(&fdp->fd_refcnt);
2346 	return (fdp);
2347 }
2348 
2349 /*
2350  * Share a pwddesc structure.
2351  */
2352 struct pwddesc *
2353 pdshare(struct pwddesc *pdp)
2354 {
2355 	refcount_acquire(&pdp->pd_refcount);
2356 	return (pdp);
2357 }
2358 
2359 /*
2360  * Unshare a filedesc structure, if necessary by making a copy
2361  */
2362 void
2363 fdunshare(struct thread *td)
2364 {
2365 	struct filedesc *tmp;
2366 	struct proc *p = td->td_proc;
2367 
2368 	if (refcount_load(&p->p_fd->fd_refcnt) == 1)
2369 		return;
2370 
2371 	tmp = fdcopy(p->p_fd);
2372 	fdescfree(td);
2373 	p->p_fd = tmp;
2374 }
2375 
2376 /*
2377  * Unshare a pwddesc structure.
2378  */
2379 void
2380 pdunshare(struct thread *td)
2381 {
2382 	struct pwddesc *pdp;
2383 	struct proc *p;
2384 
2385 	p = td->td_proc;
2386 	/* Not shared. */
2387 	if (refcount_load(&p->p_pd->pd_refcount) == 1)
2388 		return;
2389 
2390 	pdp = pdcopy(p->p_pd);
2391 	pdescfree(td);
2392 	p->p_pd = pdp;
2393 }
2394 
2395 /*
2396  * Copy a filedesc structure.  A NULL pointer in returns a NULL reference,
2397  * this is to ease callers, not catch errors.
2398  */
2399 struct filedesc *
2400 fdcopy(struct filedesc *fdp)
2401 {
2402 	struct filedesc *newfdp;
2403 	struct filedescent *nfde, *ofde;
2404 	int i, lastfile;
2405 
2406 	MPASS(fdp != NULL);
2407 
2408 	newfdp = fdinit();
2409 	FILEDESC_SLOCK(fdp);
2410 	for (;;) {
2411 		lastfile = fdlastfile(fdp);
2412 		if (lastfile < newfdp->fd_nfiles)
2413 			break;
2414 		FILEDESC_SUNLOCK(fdp);
2415 		fdgrowtable(newfdp, lastfile + 1);
2416 		FILEDESC_SLOCK(fdp);
2417 	}
2418 	/* copy all passable descriptors (i.e. not kqueue) */
2419 	newfdp->fd_freefile = fdp->fd_freefile;
2420 	FILEDESC_FOREACH_FDE(fdp, i, ofde) {
2421 		if ((ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0 ||
2422 		    !fhold(ofde->fde_file)) {
2423 			if (newfdp->fd_freefile == fdp->fd_freefile)
2424 				newfdp->fd_freefile = i;
2425 			continue;
2426 		}
2427 		nfde = &newfdp->fd_ofiles[i];
2428 		*nfde = *ofde;
2429 		filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true);
2430 		fdused_init(newfdp, i);
2431 	}
2432 	MPASS(newfdp->fd_freefile != -1);
2433 	FILEDESC_SUNLOCK(fdp);
2434 	return (newfdp);
2435 }
2436 
2437 /*
2438  * Copy a pwddesc structure.
2439  */
2440 struct pwddesc *
2441 pdcopy(struct pwddesc *pdp)
2442 {
2443 	struct pwddesc *newpdp;
2444 
2445 	MPASS(pdp != NULL);
2446 
2447 	newpdp = pdinit(pdp, true);
2448 	newpdp->pd_cmask = pdp->pd_cmask;
2449 	PWDDESC_XUNLOCK(pdp);
2450 	return (newpdp);
2451 }
2452 
2453 /*
2454  * Clear POSIX style locks. This is only used when fdp looses a reference (i.e.
2455  * one of processes using it exits) and the table used to be shared.
2456  */
2457 static void
2458 fdclearlocks(struct thread *td)
2459 {
2460 	struct filedesc *fdp;
2461 	struct filedesc_to_leader *fdtol;
2462 	struct flock lf;
2463 	struct file *fp;
2464 	struct proc *p;
2465 	struct vnode *vp;
2466 	int i;
2467 
2468 	p = td->td_proc;
2469 	fdp = p->p_fd;
2470 	fdtol = p->p_fdtol;
2471 	MPASS(fdtol != NULL);
2472 
2473 	FILEDESC_XLOCK(fdp);
2474 	KASSERT(fdtol->fdl_refcount > 0,
2475 	    ("filedesc_to_refcount botch: fdl_refcount=%d",
2476 	    fdtol->fdl_refcount));
2477 	if (fdtol->fdl_refcount == 1 &&
2478 	    (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2479 		FILEDESC_FOREACH_FP(fdp, i, fp) {
2480 			if (fp->f_type != DTYPE_VNODE ||
2481 			    !fhold(fp))
2482 				continue;
2483 			FILEDESC_XUNLOCK(fdp);
2484 			lf.l_whence = SEEK_SET;
2485 			lf.l_start = 0;
2486 			lf.l_len = 0;
2487 			lf.l_type = F_UNLCK;
2488 			vp = fp->f_vnode;
2489 			(void) VOP_ADVLOCK(vp,
2490 			    (caddr_t)p->p_leader, F_UNLCK,
2491 			    &lf, F_POSIX);
2492 			FILEDESC_XLOCK(fdp);
2493 			fdrop(fp, td);
2494 		}
2495 	}
2496 retry:
2497 	if (fdtol->fdl_refcount == 1) {
2498 		if (fdp->fd_holdleaderscount > 0 &&
2499 		    (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2500 			/*
2501 			 * close() or kern_dup() has cleared a reference
2502 			 * in a shared file descriptor table.
2503 			 */
2504 			fdp->fd_holdleaderswakeup = 1;
2505 			sx_sleep(&fdp->fd_holdleaderscount,
2506 			    FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0);
2507 			goto retry;
2508 		}
2509 		if (fdtol->fdl_holdcount > 0) {
2510 			/*
2511 			 * Ensure that fdtol->fdl_leader remains
2512 			 * valid in closef().
2513 			 */
2514 			fdtol->fdl_wakeup = 1;
2515 			sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK,
2516 			    "fdlhold", 0);
2517 			goto retry;
2518 		}
2519 	}
2520 	fdtol->fdl_refcount--;
2521 	if (fdtol->fdl_refcount == 0 &&
2522 	    fdtol->fdl_holdcount == 0) {
2523 		fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
2524 		fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
2525 	} else
2526 		fdtol = NULL;
2527 	p->p_fdtol = NULL;
2528 	FILEDESC_XUNLOCK(fdp);
2529 	if (fdtol != NULL)
2530 		free(fdtol, M_FILEDESC_TO_LEADER);
2531 }
2532 
2533 /*
2534  * Release a filedesc structure.
2535  */
2536 static void
2537 fdescfree_fds(struct thread *td, struct filedesc *fdp)
2538 {
2539 	struct filedesc0 *fdp0;
2540 	struct freetable *ft, *tft;
2541 	struct filedescent *fde;
2542 	struct file *fp;
2543 	int i;
2544 
2545 	KASSERT(refcount_load(&fdp->fd_refcnt) == 0,
2546 	    ("%s: fd table %p carries references", __func__, fdp));
2547 
2548 	/*
2549 	 * Serialize with threads iterating over the table, if any.
2550 	 */
2551 	if (refcount_load(&fdp->fd_holdcnt) > 1) {
2552 		FILEDESC_XLOCK(fdp);
2553 		FILEDESC_XUNLOCK(fdp);
2554 	}
2555 
2556 	FILEDESC_FOREACH_FDE(fdp, i, fde) {
2557 		fp = fde->fde_file;
2558 		fdefree_last(fde);
2559 		(void) closef(fp, td);
2560 	}
2561 
2562 	if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE))
2563 		free(fdp->fd_map, M_FILEDESC);
2564 	if (fdp->fd_nfiles > NDFILE)
2565 		free(fdp->fd_files, M_FILEDESC);
2566 
2567 	fdp0 = (struct filedesc0 *)fdp;
2568 	SLIST_FOREACH_SAFE(ft, &fdp0->fd_free, ft_next, tft)
2569 		free(ft->ft_table, M_FILEDESC);
2570 
2571 	fddrop(fdp);
2572 }
2573 
2574 void
2575 fdescfree(struct thread *td)
2576 {
2577 	struct proc *p;
2578 	struct filedesc *fdp;
2579 
2580 	p = td->td_proc;
2581 	fdp = p->p_fd;
2582 	MPASS(fdp != NULL);
2583 
2584 #ifdef RACCT
2585 	if (RACCT_ENABLED())
2586 		racct_set_unlocked(p, RACCT_NOFILE, 0);
2587 #endif
2588 
2589 	if (p->p_fdtol != NULL)
2590 		fdclearlocks(td);
2591 
2592 	/*
2593 	 * Check fdhold for an explanation.
2594 	 */
2595 	atomic_store_ptr(&p->p_fd, NULL);
2596 	atomic_thread_fence_seq_cst();
2597 	PROC_WAIT_UNLOCKED(p);
2598 
2599 	if (refcount_release(&fdp->fd_refcnt) == 0)
2600 		return;
2601 
2602 	fdescfree_fds(td, fdp);
2603 }
2604 
2605 void
2606 pdescfree(struct thread *td)
2607 {
2608 	struct proc *p;
2609 	struct pwddesc *pdp;
2610 
2611 	p = td->td_proc;
2612 	pdp = p->p_pd;
2613 	MPASS(pdp != NULL);
2614 
2615 	/*
2616 	 * Check pdhold for an explanation.
2617 	 */
2618 	atomic_store_ptr(&p->p_pd, NULL);
2619 	atomic_thread_fence_seq_cst();
2620 	PROC_WAIT_UNLOCKED(p);
2621 
2622 	pddrop(pdp);
2623 }
2624 
2625 /*
2626  * For setugid programs, we don't want to people to use that setugidness
2627  * to generate error messages which write to a file which otherwise would
2628  * otherwise be off-limits to the process.  We check for filesystems where
2629  * the vnode can change out from under us after execve (like [lin]procfs).
2630  *
2631  * Since fdsetugidsafety calls this only for fd 0, 1 and 2, this check is
2632  * sufficient.  We also don't check for setugidness since we know we are.
2633  */
2634 static bool
2635 is_unsafe(struct file *fp)
2636 {
2637 	struct vnode *vp;
2638 
2639 	if (fp->f_type != DTYPE_VNODE)
2640 		return (false);
2641 
2642 	vp = fp->f_vnode;
2643 	return ((vp->v_vflag & VV_PROCDEP) != 0);
2644 }
2645 
2646 /*
2647  * Make this setguid thing safe, if at all possible.
2648  */
2649 void
2650 fdsetugidsafety(struct thread *td)
2651 {
2652 	struct filedesc *fdp;
2653 	struct file *fp;
2654 	int i;
2655 
2656 	fdp = td->td_proc->p_fd;
2657 	KASSERT(refcount_load(&fdp->fd_refcnt) == 1,
2658 	    ("the fdtable should not be shared"));
2659 	MPASS(fdp->fd_nfiles >= 3);
2660 	for (i = 0; i <= 2; i++) {
2661 		fp = fdp->fd_ofiles[i].fde_file;
2662 		if (fp != NULL && is_unsafe(fp)) {
2663 			FILEDESC_XLOCK(fdp);
2664 			knote_fdclose(td, i);
2665 			/*
2666 			 * NULL-out descriptor prior to close to avoid
2667 			 * a race while close blocks.
2668 			 */
2669 			fdfree(fdp, i);
2670 			FILEDESC_XUNLOCK(fdp);
2671 			(void) closef(fp, td);
2672 		}
2673 	}
2674 }
2675 
2676 /*
2677  * If a specific file object occupies a specific file descriptor, close the
2678  * file descriptor entry and drop a reference on the file object.  This is a
2679  * convenience function to handle a subsequent error in a function that calls
2680  * falloc() that handles the race that another thread might have closed the
2681  * file descriptor out from under the thread creating the file object.
2682  */
2683 void
2684 fdclose(struct thread *td, struct file *fp, int idx)
2685 {
2686 	struct filedesc *fdp = td->td_proc->p_fd;
2687 
2688 	FILEDESC_XLOCK(fdp);
2689 	if (fdp->fd_ofiles[idx].fde_file == fp) {
2690 		fdfree(fdp, idx);
2691 		FILEDESC_XUNLOCK(fdp);
2692 		fdrop(fp, td);
2693 	} else
2694 		FILEDESC_XUNLOCK(fdp);
2695 }
2696 
2697 /*
2698  * Close any files on exec?
2699  */
2700 void
2701 fdcloseexec(struct thread *td)
2702 {
2703 	struct filedesc *fdp;
2704 	struct filedescent *fde;
2705 	struct file *fp;
2706 	int i;
2707 
2708 	fdp = td->td_proc->p_fd;
2709 	KASSERT(refcount_load(&fdp->fd_refcnt) == 1,
2710 	    ("the fdtable should not be shared"));
2711 	FILEDESC_FOREACH_FDE(fdp, i, fde) {
2712 		fp = fde->fde_file;
2713 		if (fp->f_type == DTYPE_MQUEUE ||
2714 		    (fde->fde_flags & UF_EXCLOSE)) {
2715 			FILEDESC_XLOCK(fdp);
2716 			fdfree(fdp, i);
2717 			(void) closefp(fdp, i, fp, td, false, false);
2718 			FILEDESC_UNLOCK_ASSERT(fdp);
2719 		}
2720 	}
2721 }
2722 
2723 /*
2724  * It is unsafe for set[ug]id processes to be started with file
2725  * descriptors 0..2 closed, as these descriptors are given implicit
2726  * significance in the Standard C library.  fdcheckstd() will create a
2727  * descriptor referencing /dev/null for each of stdin, stdout, and
2728  * stderr that is not already open.
2729  */
2730 int
2731 fdcheckstd(struct thread *td)
2732 {
2733 	struct filedesc *fdp;
2734 	register_t save;
2735 	int i, error, devnull;
2736 
2737 	fdp = td->td_proc->p_fd;
2738 	KASSERT(refcount_load(&fdp->fd_refcnt) == 1,
2739 	    ("the fdtable should not be shared"));
2740 	MPASS(fdp->fd_nfiles >= 3);
2741 	devnull = -1;
2742 	for (i = 0; i <= 2; i++) {
2743 		if (fdp->fd_ofiles[i].fde_file != NULL)
2744 			continue;
2745 
2746 		save = td->td_retval[0];
2747 		if (devnull != -1) {
2748 			error = kern_dup(td, FDDUP_FIXED, 0, devnull, i);
2749 		} else {
2750 			error = kern_openat(td, AT_FDCWD, "/dev/null",
2751 			    UIO_SYSSPACE, O_RDWR, 0);
2752 			if (error == 0) {
2753 				devnull = td->td_retval[0];
2754 				KASSERT(devnull == i, ("we didn't get our fd"));
2755 			}
2756 		}
2757 		td->td_retval[0] = save;
2758 		if (error != 0)
2759 			return (error);
2760 	}
2761 	return (0);
2762 }
2763 
2764 /*
2765  * Internal form of close.  Decrement reference count on file structure.
2766  * Note: td may be NULL when closing a file that was being passed in a
2767  * message.
2768  */
2769 int
2770 closef(struct file *fp, struct thread *td)
2771 {
2772 	struct vnode *vp;
2773 	struct flock lf;
2774 	struct filedesc_to_leader *fdtol;
2775 	struct filedesc *fdp;
2776 
2777 	MPASS(td != NULL);
2778 
2779 	/*
2780 	 * POSIX record locking dictates that any close releases ALL
2781 	 * locks owned by this process.  This is handled by setting
2782 	 * a flag in the unlock to free ONLY locks obeying POSIX
2783 	 * semantics, and not to free BSD-style file locks.
2784 	 * If the descriptor was in a message, POSIX-style locks
2785 	 * aren't passed with the descriptor, and the thread pointer
2786 	 * will be NULL.  Callers should be careful only to pass a
2787 	 * NULL thread pointer when there really is no owning
2788 	 * context that might have locks, or the locks will be
2789 	 * leaked.
2790 	 */
2791 	if (fp->f_type == DTYPE_VNODE) {
2792 		vp = fp->f_vnode;
2793 		if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
2794 			lf.l_whence = SEEK_SET;
2795 			lf.l_start = 0;
2796 			lf.l_len = 0;
2797 			lf.l_type = F_UNLCK;
2798 			(void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader,
2799 			    F_UNLCK, &lf, F_POSIX);
2800 		}
2801 		fdtol = td->td_proc->p_fdtol;
2802 		if (fdtol != NULL) {
2803 			/*
2804 			 * Handle special case where file descriptor table is
2805 			 * shared between multiple process leaders.
2806 			 */
2807 			fdp = td->td_proc->p_fd;
2808 			FILEDESC_XLOCK(fdp);
2809 			for (fdtol = fdtol->fdl_next;
2810 			    fdtol != td->td_proc->p_fdtol;
2811 			    fdtol = fdtol->fdl_next) {
2812 				if ((fdtol->fdl_leader->p_flag &
2813 				    P_ADVLOCK) == 0)
2814 					continue;
2815 				fdtol->fdl_holdcount++;
2816 				FILEDESC_XUNLOCK(fdp);
2817 				lf.l_whence = SEEK_SET;
2818 				lf.l_start = 0;
2819 				lf.l_len = 0;
2820 				lf.l_type = F_UNLCK;
2821 				vp = fp->f_vnode;
2822 				(void) VOP_ADVLOCK(vp,
2823 				    (caddr_t)fdtol->fdl_leader, F_UNLCK, &lf,
2824 				    F_POSIX);
2825 				FILEDESC_XLOCK(fdp);
2826 				fdtol->fdl_holdcount--;
2827 				if (fdtol->fdl_holdcount == 0 &&
2828 				    fdtol->fdl_wakeup != 0) {
2829 					fdtol->fdl_wakeup = 0;
2830 					wakeup(fdtol);
2831 				}
2832 			}
2833 			FILEDESC_XUNLOCK(fdp);
2834 		}
2835 	}
2836 	return (fdrop_close(fp, td));
2837 }
2838 
2839 /*
2840  * Hack for file descriptor passing code.
2841  */
2842 void
2843 closef_nothread(struct file *fp)
2844 {
2845 
2846 	fdrop(fp, NULL);
2847 }
2848 
2849 /*
2850  * Initialize the file pointer with the specified properties.
2851  *
2852  * The ops are set with release semantics to be certain that the flags, type,
2853  * and data are visible when ops is.  This is to prevent ops methods from being
2854  * called with bad data.
2855  */
2856 void
2857 finit(struct file *fp, u_int flag, short type, void *data, struct fileops *ops)
2858 {
2859 	fp->f_data = data;
2860 	fp->f_flag = flag;
2861 	fp->f_type = type;
2862 	atomic_store_rel_ptr((volatile uintptr_t *)&fp->f_ops, (uintptr_t)ops);
2863 }
2864 
2865 void
2866 finit_vnode(struct file *fp, u_int flag, void *data, struct fileops *ops)
2867 {
2868 	fp->f_seqcount[UIO_READ] = 1;
2869 	fp->f_seqcount[UIO_WRITE] = 1;
2870 	finit(fp, (flag & FMASK) | (fp->f_flag & FHASLOCK), DTYPE_VNODE,
2871 	    data, ops);
2872 }
2873 
2874 int
2875 fget_cap_noref(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
2876     struct file **fpp, struct filecaps *havecapsp)
2877 {
2878 	struct filedescent *fde;
2879 	int error;
2880 
2881 	FILEDESC_LOCK_ASSERT(fdp);
2882 
2883 	*fpp = NULL;
2884 	fde = fdeget_noref(fdp, fd);
2885 	if (fde == NULL) {
2886 		error = EBADF;
2887 		goto out;
2888 	}
2889 
2890 #ifdef CAPABILITIES
2891 	error = cap_check(cap_rights_fde_inline(fde), needrightsp);
2892 	if (error != 0)
2893 		goto out;
2894 #endif
2895 
2896 	if (havecapsp != NULL)
2897 		filecaps_copy(&fde->fde_caps, havecapsp, true);
2898 
2899 	*fpp = fde->fde_file;
2900 
2901 	error = 0;
2902 out:
2903 	return (error);
2904 }
2905 
2906 #ifdef CAPABILITIES
2907 int
2908 fget_cap(struct thread *td, int fd, cap_rights_t *needrightsp,
2909     struct file **fpp, struct filecaps *havecapsp)
2910 {
2911 	struct filedesc *fdp = td->td_proc->p_fd;
2912 	int error;
2913 	struct file *fp;
2914 	seqc_t seq;
2915 
2916 	*fpp = NULL;
2917 	for (;;) {
2918 		error = fget_unlocked_seq(td, fd, needrightsp, &fp, &seq);
2919 		if (error != 0)
2920 			return (error);
2921 
2922 		if (havecapsp != NULL) {
2923 			if (!filecaps_copy(&fdp->fd_ofiles[fd].fde_caps,
2924 			    havecapsp, false)) {
2925 				fdrop(fp, td);
2926 				goto get_locked;
2927 			}
2928 		}
2929 
2930 		if (!fd_modified(fdp, fd, seq))
2931 			break;
2932 		fdrop(fp, td);
2933 	}
2934 
2935 	*fpp = fp;
2936 	return (0);
2937 
2938 get_locked:
2939 	FILEDESC_SLOCK(fdp);
2940 	error = fget_cap_noref(fdp, fd, needrightsp, fpp, havecapsp);
2941 	if (error == 0 && !fhold(*fpp))
2942 		error = EBADF;
2943 	FILEDESC_SUNLOCK(fdp);
2944 	return (error);
2945 }
2946 #else
2947 int
2948 fget_cap(struct thread *td, int fd, cap_rights_t *needrightsp,
2949     struct file **fpp, struct filecaps *havecapsp)
2950 {
2951 	int error;
2952 	error = fget_unlocked(td, fd, needrightsp, fpp);
2953 	if (havecapsp != NULL && error == 0)
2954 		filecaps_fill(havecapsp);
2955 
2956 	return (error);
2957 }
2958 #endif
2959 
2960 #ifdef CAPABILITIES
2961 int
2962 fgetvp_lookup_smr(int fd, struct nameidata *ndp, struct vnode **vpp, bool *fsearch)
2963 {
2964 	const struct filedescent *fde;
2965 	const struct fdescenttbl *fdt;
2966 	struct filedesc *fdp;
2967 	struct file *fp;
2968 	struct vnode *vp;
2969 	const cap_rights_t *haverights;
2970 	cap_rights_t rights;
2971 	seqc_t seq;
2972 
2973 	VFS_SMR_ASSERT_ENTERED();
2974 
2975 	rights = *ndp->ni_rightsneeded;
2976 	cap_rights_set_one(&rights, CAP_LOOKUP);
2977 
2978 	fdp = curproc->p_fd;
2979 	fdt = fdp->fd_files;
2980 	if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
2981 		return (EBADF);
2982 	seq = seqc_read_notmodify(fd_seqc(fdt, fd));
2983 	fde = &fdt->fdt_ofiles[fd];
2984 	haverights = cap_rights_fde_inline(fde);
2985 	fp = fde->fde_file;
2986 	if (__predict_false(fp == NULL))
2987 		return (EAGAIN);
2988 	if (__predict_false(cap_check_inline_transient(haverights, &rights)))
2989 		return (EAGAIN);
2990 	*fsearch = ((fp->f_flag & FSEARCH) != 0);
2991 	vp = fp->f_vnode;
2992 	if (__predict_false(vp == NULL)) {
2993 		return (EAGAIN);
2994 	}
2995 	if (!filecaps_copy(&fde->fde_caps, &ndp->ni_filecaps, false)) {
2996 		return (EAGAIN);
2997 	}
2998 	/*
2999 	 * Use an acquire barrier to force re-reading of fdt so it is
3000 	 * refreshed for verification.
3001 	 */
3002 	atomic_thread_fence_acq();
3003 	fdt = fdp->fd_files;
3004 	if (__predict_false(!seqc_consistent_no_fence(fd_seqc(fdt, fd), seq)))
3005 		return (EAGAIN);
3006 	/*
3007 	 * If file descriptor doesn't have all rights,
3008 	 * all lookups relative to it must also be
3009 	 * strictly relative.
3010 	 *
3011 	 * Not yet supported by fast path.
3012 	 */
3013 	CAP_ALL(&rights);
3014 	if (!cap_rights_contains(&ndp->ni_filecaps.fc_rights, &rights) ||
3015 	    ndp->ni_filecaps.fc_fcntls != CAP_FCNTL_ALL ||
3016 	    ndp->ni_filecaps.fc_nioctls != -1) {
3017 #ifdef notyet
3018 		ndp->ni_lcf |= NI_LCF_STRICTRELATIVE;
3019 #else
3020 		return (EAGAIN);
3021 #endif
3022 	}
3023 	*vpp = vp;
3024 	return (0);
3025 }
3026 #else
3027 int
3028 fgetvp_lookup_smr(int fd, struct nameidata *ndp, struct vnode **vpp, bool *fsearch)
3029 {
3030 	const struct fdescenttbl *fdt;
3031 	struct filedesc *fdp;
3032 	struct file *fp;
3033 	struct vnode *vp;
3034 
3035 	VFS_SMR_ASSERT_ENTERED();
3036 
3037 	fdp = curproc->p_fd;
3038 	fdt = fdp->fd_files;
3039 	if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
3040 		return (EBADF);
3041 	fp = fdt->fdt_ofiles[fd].fde_file;
3042 	if (__predict_false(fp == NULL))
3043 		return (EAGAIN);
3044 	*fsearch = ((fp->f_flag & FSEARCH) != 0);
3045 	vp = fp->f_vnode;
3046 	if (__predict_false(vp == NULL || vp->v_type != VDIR)) {
3047 		return (EAGAIN);
3048 	}
3049 	/*
3050 	 * Use an acquire barrier to force re-reading of fdt so it is
3051 	 * refreshed for verification.
3052 	 */
3053 	atomic_thread_fence_acq();
3054 	fdt = fdp->fd_files;
3055 	if (__predict_false(fp != fdt->fdt_ofiles[fd].fde_file))
3056 		return (EAGAIN);
3057 	filecaps_fill(&ndp->ni_filecaps);
3058 	*vpp = vp;
3059 	return (0);
3060 }
3061 #endif
3062 
3063 int
3064 fgetvp_lookup(int fd, struct nameidata *ndp, struct vnode **vpp)
3065 {
3066 	struct thread *td;
3067 	struct file *fp;
3068 	struct vnode *vp;
3069 	struct componentname *cnp;
3070 	cap_rights_t rights;
3071 	int error;
3072 
3073 	td = curthread;
3074 	rights = *ndp->ni_rightsneeded;
3075 	cap_rights_set_one(&rights, CAP_LOOKUP);
3076 	cnp = &ndp->ni_cnd;
3077 
3078 	error = fget_cap(td, ndp->ni_dirfd, &rights, &fp, &ndp->ni_filecaps);
3079 	if (__predict_false(error != 0))
3080 		return (error);
3081 	if (__predict_false(fp->f_ops == &badfileops)) {
3082 		error = EBADF;
3083 		goto out_free;
3084 	}
3085 	vp = fp->f_vnode;
3086 	if (__predict_false(vp == NULL)) {
3087 		error = ENOTDIR;
3088 		goto out_free;
3089 	}
3090 	vrefact(vp);
3091 	/*
3092 	 * XXX does not check for VDIR, handled by namei_setup
3093 	 */
3094 	if ((fp->f_flag & FSEARCH) != 0)
3095 		cnp->cn_flags |= NOEXECCHECK;
3096 	fdrop(fp, td);
3097 
3098 #ifdef CAPABILITIES
3099 	/*
3100 	 * If file descriptor doesn't have all rights,
3101 	 * all lookups relative to it must also be
3102 	 * strictly relative.
3103 	 */
3104 	CAP_ALL(&rights);
3105 	if (!cap_rights_contains(&ndp->ni_filecaps.fc_rights, &rights) ||
3106 	    ndp->ni_filecaps.fc_fcntls != CAP_FCNTL_ALL ||
3107 	    ndp->ni_filecaps.fc_nioctls != -1) {
3108 		ndp->ni_lcf |= NI_LCF_STRICTRELATIVE;
3109 		ndp->ni_resflags |= NIRES_STRICTREL;
3110 	}
3111 #endif
3112 
3113 	/*
3114 	 * TODO: avoid copying ioctl caps if it can be helped to begin with
3115 	 */
3116 	if ((cnp->cn_flags & WANTIOCTLCAPS) == 0)
3117 		filecaps_free_ioctl(&ndp->ni_filecaps);
3118 
3119 	*vpp = vp;
3120 	return (0);
3121 
3122 out_free:
3123 	filecaps_free(&ndp->ni_filecaps);
3124 	fdrop(fp, td);
3125 	return (error);
3126 }
3127 
3128 /*
3129  * Fetch the descriptor locklessly.
3130  *
3131  * We avoid fdrop() races by never raising a refcount above 0.  To accomplish
3132  * this we have to use a cmpset loop rather than an atomic_add.  The descriptor
3133  * must be re-verified once we acquire a reference to be certain that the
3134  * identity is still correct and we did not lose a race due to preemption.
3135  *
3136  * Force a reload of fdt when looping. Another thread could reallocate
3137  * the table before this fd was closed, so it is possible that there is
3138  * a stale fp pointer in cached version.
3139  */
3140 #ifdef CAPABILITIES
3141 static int
3142 fget_unlocked_seq(struct thread *td, int fd, cap_rights_t *needrightsp,
3143     struct file **fpp, seqc_t *seqp)
3144 {
3145 	struct filedesc *fdp;
3146 	const struct filedescent *fde;
3147 	const struct fdescenttbl *fdt;
3148 	struct file *fp;
3149 	seqc_t seq;
3150 	cap_rights_t haverights;
3151 	int error;
3152 
3153 	fdp = td->td_proc->p_fd;
3154 	fdt = fdp->fd_files;
3155 	if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
3156 		return (EBADF);
3157 
3158 	for (;;) {
3159 		seq = seqc_read_notmodify(fd_seqc(fdt, fd));
3160 		fde = &fdt->fdt_ofiles[fd];
3161 		haverights = *cap_rights_fde_inline(fde);
3162 		fp = fde->fde_file;
3163 		if (__predict_false(fp == NULL)) {
3164 			if (seqc_consistent(fd_seqc(fdt, fd), seq))
3165 				return (EBADF);
3166 			fdt = atomic_load_ptr(&fdp->fd_files);
3167 			continue;
3168 		}
3169 		error = cap_check_inline(&haverights, needrightsp);
3170 		if (__predict_false(error != 0)) {
3171 			if (seqc_consistent(fd_seqc(fdt, fd), seq))
3172 				return (error);
3173 			fdt = atomic_load_ptr(&fdp->fd_files);
3174 			continue;
3175 		}
3176 		if (__predict_false(!refcount_acquire_if_not_zero(&fp->f_count))) {
3177 			fdt = atomic_load_ptr(&fdp->fd_files);
3178 			continue;
3179 		}
3180 		/*
3181 		 * Use an acquire barrier to force re-reading of fdt so it is
3182 		 * refreshed for verification.
3183 		 */
3184 		atomic_thread_fence_acq();
3185 		fdt = fdp->fd_files;
3186 		if (seqc_consistent_no_fence(fd_seqc(fdt, fd), seq))
3187 			break;
3188 		fdrop(fp, td);
3189 	}
3190 	*fpp = fp;
3191 	if (seqp != NULL) {
3192 		*seqp = seq;
3193 	}
3194 	return (0);
3195 }
3196 #else
3197 static int
3198 fget_unlocked_seq(struct thread *td, int fd, cap_rights_t *needrightsp,
3199     struct file **fpp, seqc_t *seqp __unused)
3200 {
3201 	struct filedesc *fdp;
3202 	const struct fdescenttbl *fdt;
3203 	struct file *fp;
3204 
3205 	fdp = td->td_proc->p_fd;
3206 	fdt = fdp->fd_files;
3207 	if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
3208 		return (EBADF);
3209 
3210 	for (;;) {
3211 		fp = fdt->fdt_ofiles[fd].fde_file;
3212 		if (__predict_false(fp == NULL))
3213 			return (EBADF);
3214 		if (__predict_false(!refcount_acquire_if_not_zero(&fp->f_count))) {
3215 			fdt = atomic_load_ptr(&fdp->fd_files);
3216 			continue;
3217 		}
3218 		/*
3219 		 * Use an acquire barrier to force re-reading of fdt so it is
3220 		 * refreshed for verification.
3221 		 */
3222 		atomic_thread_fence_acq();
3223 		fdt = fdp->fd_files;
3224 		if (__predict_true(fp == fdt->fdt_ofiles[fd].fde_file))
3225 			break;
3226 		fdrop(fp, td);
3227 	}
3228 	*fpp = fp;
3229 	return (0);
3230 }
3231 #endif
3232 
3233 /*
3234  * See the comments in fget_unlocked_seq for an explanation of how this works.
3235  *
3236  * This is a simplified variant which bails out to the aforementioned routine
3237  * if anything goes wrong. In practice this only happens when userspace is
3238  * racing with itself.
3239  */
3240 int
3241 fget_unlocked(struct thread *td, int fd, cap_rights_t *needrightsp,
3242     struct file **fpp)
3243 {
3244 	struct filedesc *fdp;
3245 #ifdef CAPABILITIES
3246 	const struct filedescent *fde;
3247 #endif
3248 	const struct fdescenttbl *fdt;
3249 	struct file *fp;
3250 #ifdef CAPABILITIES
3251 	seqc_t seq;
3252 	const cap_rights_t *haverights;
3253 #endif
3254 
3255 	fdp = td->td_proc->p_fd;
3256 	fdt = fdp->fd_files;
3257 	if (__predict_false((u_int)fd >= fdt->fdt_nfiles)) {
3258 		*fpp = NULL;
3259 		return (EBADF);
3260 	}
3261 #ifdef CAPABILITIES
3262 	seq = seqc_read_notmodify(fd_seqc(fdt, fd));
3263 	fde = &fdt->fdt_ofiles[fd];
3264 	haverights = cap_rights_fde_inline(fde);
3265 	fp = fde->fde_file;
3266 #else
3267 	fp = fdt->fdt_ofiles[fd].fde_file;
3268 #endif
3269 	if (__predict_false(fp == NULL))
3270 		goto out_fallback;
3271 #ifdef CAPABILITIES
3272 	if (__predict_false(cap_check_inline_transient(haverights, needrightsp)))
3273 		goto out_fallback;
3274 #endif
3275 	if (__predict_false(!refcount_acquire_if_not_zero(&fp->f_count)))
3276 		goto out_fallback;
3277 
3278 	/*
3279 	 * Use an acquire barrier to force re-reading of fdt so it is
3280 	 * refreshed for verification.
3281 	 */
3282 	atomic_thread_fence_acq();
3283 	fdt = fdp->fd_files;
3284 #ifdef	CAPABILITIES
3285 	if (__predict_false(!seqc_consistent_no_fence(fd_seqc(fdt, fd), seq)))
3286 #else
3287 	if (__predict_false(fp != fdt->fdt_ofiles[fd].fde_file))
3288 #endif
3289 		goto out_fdrop;
3290 	*fpp = fp;
3291 	return (0);
3292 out_fdrop:
3293 	fdrop(fp, td);
3294 out_fallback:
3295 	*fpp = NULL;
3296 	return (fget_unlocked_seq(td, fd, needrightsp, fpp, NULL));
3297 }
3298 
3299 /*
3300  * Translate fd -> file when the caller guarantees the file descriptor table
3301  * can't be changed by others.
3302  *
3303  * Note this does not mean the file object itself is only visible to the caller,
3304  * merely that it wont disappear without having to be referenced.
3305  *
3306  * Must be paired with fput_only_user.
3307  */
3308 #ifdef	CAPABILITIES
3309 int
3310 fget_only_user(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
3311     struct file **fpp)
3312 {
3313 	const struct filedescent *fde;
3314 	const struct fdescenttbl *fdt;
3315 	const cap_rights_t *haverights;
3316 	struct file *fp;
3317 	int error;
3318 
3319 	MPASS(FILEDESC_IS_ONLY_USER(fdp));
3320 
3321 	*fpp = NULL;
3322 	if (__predict_false(fd >= fdp->fd_nfiles))
3323 		return (EBADF);
3324 
3325 	fdt = fdp->fd_files;
3326 	fde = &fdt->fdt_ofiles[fd];
3327 	fp = fde->fde_file;
3328 	if (__predict_false(fp == NULL))
3329 		return (EBADF);
3330 	MPASS(refcount_load(&fp->f_count) > 0);
3331 	haverights = cap_rights_fde_inline(fde);
3332 	error = cap_check_inline(haverights, needrightsp);
3333 	if (__predict_false(error != 0))
3334 		return (error);
3335 	*fpp = fp;
3336 	return (0);
3337 }
3338 #else
3339 int
3340 fget_only_user(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
3341     struct file **fpp)
3342 {
3343 	struct file *fp;
3344 
3345 	MPASS(FILEDESC_IS_ONLY_USER(fdp));
3346 
3347 	*fpp = NULL;
3348 	if (__predict_false(fd >= fdp->fd_nfiles))
3349 		return (EBADF);
3350 
3351 	fp = fdp->fd_ofiles[fd].fde_file;
3352 	if (__predict_false(fp == NULL))
3353 		return (EBADF);
3354 
3355 	MPASS(refcount_load(&fp->f_count) > 0);
3356 	*fpp = fp;
3357 	return (0);
3358 }
3359 #endif
3360 
3361 /*
3362  * Extract the file pointer associated with the specified descriptor for the
3363  * current user process.
3364  *
3365  * If the descriptor doesn't exist or doesn't match 'flags', EBADF is
3366  * returned.
3367  *
3368  * File's rights will be checked against the capability rights mask.
3369  *
3370  * If an error occurred the non-zero error is returned and *fpp is set to
3371  * NULL.  Otherwise *fpp is held and set and zero is returned.  Caller is
3372  * responsible for fdrop().
3373  */
3374 static __inline int
3375 _fget(struct thread *td, int fd, struct file **fpp, int flags,
3376     cap_rights_t *needrightsp)
3377 {
3378 	struct file *fp;
3379 	int error;
3380 
3381 	*fpp = NULL;
3382 	error = fget_unlocked(td, fd, needrightsp, &fp);
3383 	if (__predict_false(error != 0))
3384 		return (error);
3385 	if (__predict_false(fp->f_ops == &badfileops)) {
3386 		fdrop(fp, td);
3387 		return (EBADF);
3388 	}
3389 
3390 	/*
3391 	 * FREAD and FWRITE failure return EBADF as per POSIX.
3392 	 */
3393 	error = 0;
3394 	switch (flags) {
3395 	case FREAD:
3396 	case FWRITE:
3397 		if ((fp->f_flag & flags) == 0)
3398 			error = EBADF;
3399 		break;
3400 	case FEXEC:
3401 		if (fp->f_ops != &path_fileops &&
3402 		    ((fp->f_flag & (FREAD | FEXEC)) == 0 ||
3403 		    (fp->f_flag & FWRITE) != 0))
3404 			error = EBADF;
3405 		break;
3406 	case 0:
3407 		break;
3408 	default:
3409 		KASSERT(0, ("wrong flags"));
3410 	}
3411 
3412 	if (error != 0) {
3413 		fdrop(fp, td);
3414 		return (error);
3415 	}
3416 
3417 	*fpp = fp;
3418 	return (0);
3419 }
3420 
3421 int
3422 fget(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
3423 {
3424 
3425 	return (_fget(td, fd, fpp, 0, rightsp));
3426 }
3427 
3428 int
3429 fget_mmap(struct thread *td, int fd, cap_rights_t *rightsp, vm_prot_t *maxprotp,
3430     struct file **fpp)
3431 {
3432 	int error;
3433 #ifndef CAPABILITIES
3434 	error = _fget(td, fd, fpp, 0, rightsp);
3435 	if (maxprotp != NULL)
3436 		*maxprotp = VM_PROT_ALL;
3437 	return (error);
3438 #else
3439 	cap_rights_t fdrights;
3440 	struct filedesc *fdp;
3441 	struct file *fp;
3442 	seqc_t seq;
3443 
3444 	*fpp = NULL;
3445 	fdp = td->td_proc->p_fd;
3446 	MPASS(cap_rights_is_set(rightsp, CAP_MMAP));
3447 	for (;;) {
3448 		error = fget_unlocked_seq(td, fd, rightsp, &fp, &seq);
3449 		if (__predict_false(error != 0))
3450 			return (error);
3451 		if (__predict_false(fp->f_ops == &badfileops)) {
3452 			fdrop(fp, td);
3453 			return (EBADF);
3454 		}
3455 		if (maxprotp != NULL)
3456 			fdrights = *cap_rights(fdp, fd);
3457 		if (!fd_modified(fdp, fd, seq))
3458 			break;
3459 		fdrop(fp, td);
3460 	}
3461 
3462 	/*
3463 	 * If requested, convert capability rights to access flags.
3464 	 */
3465 	if (maxprotp != NULL)
3466 		*maxprotp = cap_rights_to_vmprot(&fdrights);
3467 	*fpp = fp;
3468 	return (0);
3469 #endif
3470 }
3471 
3472 int
3473 fget_read(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
3474 {
3475 
3476 	return (_fget(td, fd, fpp, FREAD, rightsp));
3477 }
3478 
3479 int
3480 fget_write(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
3481 {
3482 
3483 	return (_fget(td, fd, fpp, FWRITE, rightsp));
3484 }
3485 
3486 int
3487 fget_fcntl(struct thread *td, int fd, cap_rights_t *rightsp, int needfcntl,
3488     struct file **fpp)
3489 {
3490 #ifndef CAPABILITIES
3491 	return (fget_unlocked(td, fd, rightsp, fpp));
3492 #else
3493 	struct filedesc *fdp = td->td_proc->p_fd;
3494 	struct file *fp;
3495 	int error;
3496 	seqc_t seq;
3497 
3498 	*fpp = NULL;
3499 	MPASS(cap_rights_is_set(rightsp, CAP_FCNTL));
3500 	for (;;) {
3501 		error = fget_unlocked_seq(td, fd, rightsp, &fp, &seq);
3502 		if (error != 0)
3503 			return (error);
3504 		error = cap_fcntl_check(fdp, fd, needfcntl);
3505 		if (!fd_modified(fdp, fd, seq))
3506 			break;
3507 		fdrop(fp, td);
3508 	}
3509 	if (error != 0) {
3510 		fdrop(fp, td);
3511 		return (error);
3512 	}
3513 	*fpp = fp;
3514 	return (0);
3515 #endif
3516 }
3517 
3518 /*
3519  * Like fget() but loads the underlying vnode, or returns an error if the
3520  * descriptor does not represent a vnode.  Note that pipes use vnodes but
3521  * never have VM objects.  The returned vnode will be vref()'d.
3522  *
3523  * XXX: what about the unused flags ?
3524  */
3525 static __inline int
3526 _fgetvp(struct thread *td, int fd, int flags, cap_rights_t *needrightsp,
3527     struct vnode **vpp)
3528 {
3529 	struct file *fp;
3530 	int error;
3531 
3532 	*vpp = NULL;
3533 	error = _fget(td, fd, &fp, flags, needrightsp);
3534 	if (error != 0)
3535 		return (error);
3536 	if (fp->f_vnode == NULL) {
3537 		error = EINVAL;
3538 	} else {
3539 		*vpp = fp->f_vnode;
3540 		vrefact(*vpp);
3541 	}
3542 	fdrop(fp, td);
3543 
3544 	return (error);
3545 }
3546 
3547 int
3548 fgetvp(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
3549 {
3550 
3551 	return (_fgetvp(td, fd, 0, rightsp, vpp));
3552 }
3553 
3554 int
3555 fgetvp_rights(struct thread *td, int fd, cap_rights_t *needrightsp,
3556     struct filecaps *havecaps, struct vnode **vpp)
3557 {
3558 	struct filecaps caps;
3559 	struct file *fp;
3560 	int error;
3561 
3562 	error = fget_cap(td, fd, needrightsp, &fp, &caps);
3563 	if (error != 0)
3564 		return (error);
3565 	if (fp->f_ops == &badfileops) {
3566 		error = EBADF;
3567 		goto out;
3568 	}
3569 	if (fp->f_vnode == NULL) {
3570 		error = EINVAL;
3571 		goto out;
3572 	}
3573 
3574 	*havecaps = caps;
3575 	*vpp = fp->f_vnode;
3576 	vrefact(*vpp);
3577 	fdrop(fp, td);
3578 
3579 	return (0);
3580 out:
3581 	filecaps_free(&caps);
3582 	fdrop(fp, td);
3583 	return (error);
3584 }
3585 
3586 int
3587 fgetvp_read(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
3588 {
3589 
3590 	return (_fgetvp(td, fd, FREAD, rightsp, vpp));
3591 }
3592 
3593 int
3594 fgetvp_exec(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
3595 {
3596 
3597 	return (_fgetvp(td, fd, FEXEC, rightsp, vpp));
3598 }
3599 
3600 #ifdef notyet
3601 int
3602 fgetvp_write(struct thread *td, int fd, cap_rights_t *rightsp,
3603     struct vnode **vpp)
3604 {
3605 
3606 	return (_fgetvp(td, fd, FWRITE, rightsp, vpp));
3607 }
3608 #endif
3609 
3610 /*
3611  * Handle the last reference to a file being closed.
3612  *
3613  * Without the noinline attribute clang keeps inlining the func thorough this
3614  * file when fdrop is used.
3615  */
3616 int __noinline
3617 _fdrop(struct file *fp, struct thread *td)
3618 {
3619 	int error;
3620 #ifdef INVARIANTS
3621 	int count;
3622 
3623 	count = refcount_load(&fp->f_count);
3624 	if (count != 0)
3625 		panic("fdrop: fp %p count %d", fp, count);
3626 #endif
3627 	error = fo_close(fp, td);
3628 	atomic_subtract_int(&openfiles, 1);
3629 	crfree(fp->f_cred);
3630 	free(fp->f_advice, M_FADVISE);
3631 	uma_zfree(file_zone, fp);
3632 
3633 	return (error);
3634 }
3635 
3636 /*
3637  * Apply an advisory lock on a file descriptor.
3638  *
3639  * Just attempt to get a record lock of the requested type on the entire file
3640  * (l_whence = SEEK_SET, l_start = 0, l_len = 0).
3641  */
3642 #ifndef _SYS_SYSPROTO_H_
3643 struct flock_args {
3644 	int	fd;
3645 	int	how;
3646 };
3647 #endif
3648 /* ARGSUSED */
3649 int
3650 sys_flock(struct thread *td, struct flock_args *uap)
3651 {
3652 	struct file *fp;
3653 	struct vnode *vp;
3654 	struct flock lf;
3655 	int error;
3656 
3657 	error = fget(td, uap->fd, &cap_flock_rights, &fp);
3658 	if (error != 0)
3659 		return (error);
3660 	error = EOPNOTSUPP;
3661 	if (fp->f_type != DTYPE_VNODE && fp->f_type != DTYPE_FIFO) {
3662 		goto done;
3663 	}
3664 	if (fp->f_ops == &path_fileops) {
3665 		goto done;
3666 	}
3667 
3668 	error = 0;
3669 	vp = fp->f_vnode;
3670 	lf.l_whence = SEEK_SET;
3671 	lf.l_start = 0;
3672 	lf.l_len = 0;
3673 	if (uap->how & LOCK_UN) {
3674 		lf.l_type = F_UNLCK;
3675 		atomic_clear_int(&fp->f_flag, FHASLOCK);
3676 		error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK);
3677 		goto done;
3678 	}
3679 	if (uap->how & LOCK_EX)
3680 		lf.l_type = F_WRLCK;
3681 	else if (uap->how & LOCK_SH)
3682 		lf.l_type = F_RDLCK;
3683 	else {
3684 		error = EBADF;
3685 		goto done;
3686 	}
3687 	atomic_set_int(&fp->f_flag, FHASLOCK);
3688 	error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf,
3689 	    (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT);
3690 done:
3691 	fdrop(fp, td);
3692 	return (error);
3693 }
3694 /*
3695  * Duplicate the specified descriptor to a free descriptor.
3696  */
3697 int
3698 dupfdopen(struct thread *td, struct filedesc *fdp, int dfd, int mode,
3699     int openerror, int *indxp)
3700 {
3701 	struct filedescent *newfde, *oldfde;
3702 	struct file *fp;
3703 	u_long *ioctls;
3704 	int error, indx;
3705 
3706 	KASSERT(openerror == ENODEV || openerror == ENXIO,
3707 	    ("unexpected error %d in %s", openerror, __func__));
3708 
3709 	/*
3710 	 * If the to-be-dup'd fd number is greater than the allowed number
3711 	 * of file descriptors, or the fd to be dup'd has already been
3712 	 * closed, then reject.
3713 	 */
3714 	FILEDESC_XLOCK(fdp);
3715 	if ((fp = fget_noref(fdp, dfd)) == NULL) {
3716 		FILEDESC_XUNLOCK(fdp);
3717 		return (EBADF);
3718 	}
3719 
3720 	error = fdalloc(td, 0, &indx);
3721 	if (error != 0) {
3722 		FILEDESC_XUNLOCK(fdp);
3723 		return (error);
3724 	}
3725 
3726 	/*
3727 	 * There are two cases of interest here.
3728 	 *
3729 	 * For ENODEV simply dup (dfd) to file descriptor (indx) and return.
3730 	 *
3731 	 * For ENXIO steal away the file structure from (dfd) and store it in
3732 	 * (indx).  (dfd) is effectively closed by this operation.
3733 	 */
3734 	switch (openerror) {
3735 	case ENODEV:
3736 		/*
3737 		 * Check that the mode the file is being opened for is a
3738 		 * subset of the mode of the existing descriptor.
3739 		 */
3740 		if (((mode & (FREAD|FWRITE)) | fp->f_flag) != fp->f_flag) {
3741 			fdunused(fdp, indx);
3742 			FILEDESC_XUNLOCK(fdp);
3743 			return (EACCES);
3744 		}
3745 		if (!fhold(fp)) {
3746 			fdunused(fdp, indx);
3747 			FILEDESC_XUNLOCK(fdp);
3748 			return (EBADF);
3749 		}
3750 		newfde = &fdp->fd_ofiles[indx];
3751 		oldfde = &fdp->fd_ofiles[dfd];
3752 		ioctls = filecaps_copy_prep(&oldfde->fde_caps);
3753 #ifdef CAPABILITIES
3754 		seqc_write_begin(&newfde->fde_seqc);
3755 #endif
3756 		fde_copy(oldfde, newfde);
3757 		filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps,
3758 		    ioctls);
3759 #ifdef CAPABILITIES
3760 		seqc_write_end(&newfde->fde_seqc);
3761 #endif
3762 		break;
3763 	case ENXIO:
3764 		/*
3765 		 * Steal away the file pointer from dfd and stuff it into indx.
3766 		 */
3767 		newfde = &fdp->fd_ofiles[indx];
3768 		oldfde = &fdp->fd_ofiles[dfd];
3769 #ifdef CAPABILITIES
3770 		seqc_write_begin(&oldfde->fde_seqc);
3771 		seqc_write_begin(&newfde->fde_seqc);
3772 #endif
3773 		fde_copy(oldfde, newfde);
3774 		oldfde->fde_file = NULL;
3775 		fdunused(fdp, dfd);
3776 #ifdef CAPABILITIES
3777 		seqc_write_end(&newfde->fde_seqc);
3778 		seqc_write_end(&oldfde->fde_seqc);
3779 #endif
3780 		break;
3781 	}
3782 	FILEDESC_XUNLOCK(fdp);
3783 	*indxp = indx;
3784 	return (0);
3785 }
3786 
3787 /*
3788  * This sysctl determines if we will allow a process to chroot(2) if it
3789  * has a directory open:
3790  *	0: disallowed for all processes.
3791  *	1: allowed for processes that were not already chroot(2)'ed.
3792  *	2: allowed for all processes.
3793  */
3794 
3795 static int chroot_allow_open_directories = 1;
3796 
3797 SYSCTL_INT(_kern, OID_AUTO, chroot_allow_open_directories, CTLFLAG_RW,
3798     &chroot_allow_open_directories, 0,
3799     "Allow a process to chroot(2) if it has a directory open");
3800 
3801 /*
3802  * Helper function for raised chroot(2) security function:  Refuse if
3803  * any filedescriptors are open directories.
3804  */
3805 static int
3806 chroot_refuse_vdir_fds(struct filedesc *fdp)
3807 {
3808 	struct vnode *vp;
3809 	struct file *fp;
3810 	int i;
3811 
3812 	FILEDESC_LOCK_ASSERT(fdp);
3813 
3814 	FILEDESC_FOREACH_FP(fdp, i, fp) {
3815 		if (fp->f_type == DTYPE_VNODE) {
3816 			vp = fp->f_vnode;
3817 			if (vp->v_type == VDIR)
3818 				return (EPERM);
3819 		}
3820 	}
3821 	return (0);
3822 }
3823 
3824 static void
3825 pwd_fill(struct pwd *oldpwd, struct pwd *newpwd)
3826 {
3827 
3828 	if (newpwd->pwd_cdir == NULL && oldpwd->pwd_cdir != NULL) {
3829 		vrefact(oldpwd->pwd_cdir);
3830 		newpwd->pwd_cdir = oldpwd->pwd_cdir;
3831 	}
3832 
3833 	if (newpwd->pwd_rdir == NULL && oldpwd->pwd_rdir != NULL) {
3834 		vrefact(oldpwd->pwd_rdir);
3835 		newpwd->pwd_rdir = oldpwd->pwd_rdir;
3836 	}
3837 
3838 	if (newpwd->pwd_jdir == NULL && oldpwd->pwd_jdir != NULL) {
3839 		vrefact(oldpwd->pwd_jdir);
3840 		newpwd->pwd_jdir = oldpwd->pwd_jdir;
3841 	}
3842 }
3843 
3844 struct pwd *
3845 pwd_hold_pwddesc(struct pwddesc *pdp)
3846 {
3847 	struct pwd *pwd;
3848 
3849 	PWDDESC_ASSERT_XLOCKED(pdp);
3850 	pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3851 	if (pwd != NULL)
3852 		refcount_acquire(&pwd->pwd_refcount);
3853 	return (pwd);
3854 }
3855 
3856 bool
3857 pwd_hold_smr(struct pwd *pwd)
3858 {
3859 
3860 	MPASS(pwd != NULL);
3861 	if (__predict_true(refcount_acquire_if_not_zero(&pwd->pwd_refcount))) {
3862 		return (true);
3863 	}
3864 	return (false);
3865 }
3866 
3867 struct pwd *
3868 pwd_hold(struct thread *td)
3869 {
3870 	struct pwddesc *pdp;
3871 	struct pwd *pwd;
3872 
3873 	pdp = td->td_proc->p_pd;
3874 
3875 	vfs_smr_enter();
3876 	pwd = vfs_smr_entered_load(&pdp->pd_pwd);
3877 	if (pwd_hold_smr(pwd)) {
3878 		vfs_smr_exit();
3879 		return (pwd);
3880 	}
3881 	vfs_smr_exit();
3882 	PWDDESC_XLOCK(pdp);
3883 	pwd = pwd_hold_pwddesc(pdp);
3884 	MPASS(pwd != NULL);
3885 	PWDDESC_XUNLOCK(pdp);
3886 	return (pwd);
3887 }
3888 
3889 struct pwd *
3890 pwd_hold_proc(struct proc *p)
3891 {
3892 	struct pwddesc *pdp;
3893 	struct pwd *pwd;
3894 
3895 	PROC_ASSERT_HELD(p);
3896 	PROC_LOCK(p);
3897 	pdp = pdhold(p);
3898 	MPASS(pdp != NULL);
3899 	PROC_UNLOCK(p);
3900 
3901 	PWDDESC_XLOCK(pdp);
3902 	pwd = pwd_hold_pwddesc(pdp);
3903 	MPASS(pwd != NULL);
3904 	PWDDESC_XUNLOCK(pdp);
3905 	pddrop(pdp);
3906 	return (pwd);
3907 }
3908 
3909 static struct pwd *
3910 pwd_alloc(void)
3911 {
3912 	struct pwd *pwd;
3913 
3914 	pwd = uma_zalloc_smr(pwd_zone, M_WAITOK);
3915 	bzero(pwd, sizeof(*pwd));
3916 	refcount_init(&pwd->pwd_refcount, 1);
3917 	return (pwd);
3918 }
3919 
3920 void
3921 pwd_drop(struct pwd *pwd)
3922 {
3923 
3924 	if (!refcount_release(&pwd->pwd_refcount))
3925 		return;
3926 
3927 	if (pwd->pwd_cdir != NULL)
3928 		vrele(pwd->pwd_cdir);
3929 	if (pwd->pwd_rdir != NULL)
3930 		vrele(pwd->pwd_rdir);
3931 	if (pwd->pwd_jdir != NULL)
3932 		vrele(pwd->pwd_jdir);
3933 	uma_zfree_smr(pwd_zone, pwd);
3934 }
3935 
3936 /*
3937 * The caller is responsible for invoking priv_check() and
3938 * mac_vnode_check_chroot() to authorize this operation.
3939 */
3940 int
3941 pwd_chroot(struct thread *td, struct vnode *vp)
3942 {
3943 	struct pwddesc *pdp;
3944 	struct filedesc *fdp;
3945 	struct pwd *newpwd, *oldpwd;
3946 	int error;
3947 
3948 	fdp = td->td_proc->p_fd;
3949 	pdp = td->td_proc->p_pd;
3950 	newpwd = pwd_alloc();
3951 	FILEDESC_SLOCK(fdp);
3952 	PWDDESC_XLOCK(pdp);
3953 	oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3954 	if (chroot_allow_open_directories == 0 ||
3955 	    (chroot_allow_open_directories == 1 &&
3956 	    oldpwd->pwd_rdir != rootvnode)) {
3957 		error = chroot_refuse_vdir_fds(fdp);
3958 		FILEDESC_SUNLOCK(fdp);
3959 		if (error != 0) {
3960 			PWDDESC_XUNLOCK(pdp);
3961 			pwd_drop(newpwd);
3962 			return (error);
3963 		}
3964 	} else {
3965 		FILEDESC_SUNLOCK(fdp);
3966 	}
3967 
3968 	vrefact(vp);
3969 	newpwd->pwd_rdir = vp;
3970 	if (oldpwd->pwd_jdir == NULL) {
3971 		vrefact(vp);
3972 		newpwd->pwd_jdir = vp;
3973 	}
3974 	pwd_fill(oldpwd, newpwd);
3975 	pwd_set(pdp, newpwd);
3976 	PWDDESC_XUNLOCK(pdp);
3977 	pwd_drop(oldpwd);
3978 	return (0);
3979 }
3980 
3981 void
3982 pwd_chdir(struct thread *td, struct vnode *vp)
3983 {
3984 	struct pwddesc *pdp;
3985 	struct pwd *newpwd, *oldpwd;
3986 
3987 	VNPASS(vp->v_usecount > 0, vp);
3988 
3989 	newpwd = pwd_alloc();
3990 	pdp = td->td_proc->p_pd;
3991 	PWDDESC_XLOCK(pdp);
3992 	oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
3993 	newpwd->pwd_cdir = vp;
3994 	pwd_fill(oldpwd, newpwd);
3995 	pwd_set(pdp, newpwd);
3996 	PWDDESC_XUNLOCK(pdp);
3997 	pwd_drop(oldpwd);
3998 }
3999 
4000 /*
4001  * jail_attach(2) changes both root and working directories.
4002  */
4003 int
4004 pwd_chroot_chdir(struct thread *td, struct vnode *vp)
4005 {
4006 	struct pwddesc *pdp;
4007 	struct filedesc *fdp;
4008 	struct pwd *newpwd, *oldpwd;
4009 	int error;
4010 
4011 	fdp = td->td_proc->p_fd;
4012 	pdp = td->td_proc->p_pd;
4013 	newpwd = pwd_alloc();
4014 	FILEDESC_SLOCK(fdp);
4015 	PWDDESC_XLOCK(pdp);
4016 	oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
4017 	error = chroot_refuse_vdir_fds(fdp);
4018 	FILEDESC_SUNLOCK(fdp);
4019 	if (error != 0) {
4020 		PWDDESC_XUNLOCK(pdp);
4021 		pwd_drop(newpwd);
4022 		return (error);
4023 	}
4024 
4025 	vrefact(vp);
4026 	newpwd->pwd_rdir = vp;
4027 	vrefact(vp);
4028 	newpwd->pwd_cdir = vp;
4029 	if (oldpwd->pwd_jdir == NULL) {
4030 		vrefact(vp);
4031 		newpwd->pwd_jdir = vp;
4032 	}
4033 	pwd_fill(oldpwd, newpwd);
4034 	pwd_set(pdp, newpwd);
4035 	PWDDESC_XUNLOCK(pdp);
4036 	pwd_drop(oldpwd);
4037 	return (0);
4038 }
4039 
4040 void
4041 pwd_ensure_dirs(void)
4042 {
4043 	struct pwddesc *pdp;
4044 	struct pwd *oldpwd, *newpwd;
4045 
4046 	pdp = curproc->p_pd;
4047 	PWDDESC_XLOCK(pdp);
4048 	oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
4049 	if (oldpwd->pwd_cdir != NULL && oldpwd->pwd_rdir != NULL) {
4050 		PWDDESC_XUNLOCK(pdp);
4051 		return;
4052 	}
4053 	PWDDESC_XUNLOCK(pdp);
4054 
4055 	newpwd = pwd_alloc();
4056 	PWDDESC_XLOCK(pdp);
4057 	oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
4058 	pwd_fill(oldpwd, newpwd);
4059 	if (newpwd->pwd_cdir == NULL) {
4060 		vrefact(rootvnode);
4061 		newpwd->pwd_cdir = rootvnode;
4062 	}
4063 	if (newpwd->pwd_rdir == NULL) {
4064 		vrefact(rootvnode);
4065 		newpwd->pwd_rdir = rootvnode;
4066 	}
4067 	pwd_set(pdp, newpwd);
4068 	PWDDESC_XUNLOCK(pdp);
4069 	pwd_drop(oldpwd);
4070 }
4071 
4072 void
4073 pwd_set_rootvnode(void)
4074 {
4075 	struct pwddesc *pdp;
4076 	struct pwd *oldpwd, *newpwd;
4077 
4078 	pdp = curproc->p_pd;
4079 
4080 	newpwd = pwd_alloc();
4081 	PWDDESC_XLOCK(pdp);
4082 	oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
4083 	vrefact(rootvnode);
4084 	newpwd->pwd_cdir = rootvnode;
4085 	vrefact(rootvnode);
4086 	newpwd->pwd_rdir = rootvnode;
4087 	pwd_fill(oldpwd, newpwd);
4088 	pwd_set(pdp, newpwd);
4089 	PWDDESC_XUNLOCK(pdp);
4090 	pwd_drop(oldpwd);
4091 }
4092 
4093 /*
4094  * Scan all active processes and prisons to see if any of them have a current
4095  * or root directory of `olddp'. If so, replace them with the new mount point.
4096  */
4097 void
4098 mountcheckdirs(struct vnode *olddp, struct vnode *newdp)
4099 {
4100 	struct pwddesc *pdp;
4101 	struct pwd *newpwd, *oldpwd;
4102 	struct prison *pr;
4103 	struct proc *p;
4104 	int nrele;
4105 
4106 	if (vrefcnt(olddp) == 1)
4107 		return;
4108 	nrele = 0;
4109 	newpwd = pwd_alloc();
4110 	sx_slock(&allproc_lock);
4111 	FOREACH_PROC_IN_SYSTEM(p) {
4112 		PROC_LOCK(p);
4113 		pdp = pdhold(p);
4114 		PROC_UNLOCK(p);
4115 		if (pdp == NULL)
4116 			continue;
4117 		PWDDESC_XLOCK(pdp);
4118 		oldpwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
4119 		if (oldpwd == NULL ||
4120 		    (oldpwd->pwd_cdir != olddp &&
4121 		    oldpwd->pwd_rdir != olddp &&
4122 		    oldpwd->pwd_jdir != olddp)) {
4123 			PWDDESC_XUNLOCK(pdp);
4124 			pddrop(pdp);
4125 			continue;
4126 		}
4127 		if (oldpwd->pwd_cdir == olddp) {
4128 			vrefact(newdp);
4129 			newpwd->pwd_cdir = newdp;
4130 		}
4131 		if (oldpwd->pwd_rdir == olddp) {
4132 			vrefact(newdp);
4133 			newpwd->pwd_rdir = newdp;
4134 		}
4135 		if (oldpwd->pwd_jdir == olddp) {
4136 			vrefact(newdp);
4137 			newpwd->pwd_jdir = newdp;
4138 		}
4139 		pwd_fill(oldpwd, newpwd);
4140 		pwd_set(pdp, newpwd);
4141 		PWDDESC_XUNLOCK(pdp);
4142 		pwd_drop(oldpwd);
4143 		pddrop(pdp);
4144 		newpwd = pwd_alloc();
4145 	}
4146 	sx_sunlock(&allproc_lock);
4147 	pwd_drop(newpwd);
4148 	if (rootvnode == olddp) {
4149 		vrefact(newdp);
4150 		rootvnode = newdp;
4151 		nrele++;
4152 	}
4153 	mtx_lock(&prison0.pr_mtx);
4154 	if (prison0.pr_root == olddp) {
4155 		vrefact(newdp);
4156 		prison0.pr_root = newdp;
4157 		nrele++;
4158 	}
4159 	mtx_unlock(&prison0.pr_mtx);
4160 	sx_slock(&allprison_lock);
4161 	TAILQ_FOREACH(pr, &allprison, pr_list) {
4162 		mtx_lock(&pr->pr_mtx);
4163 		if (pr->pr_root == olddp) {
4164 			vrefact(newdp);
4165 			pr->pr_root = newdp;
4166 			nrele++;
4167 		}
4168 		mtx_unlock(&pr->pr_mtx);
4169 	}
4170 	sx_sunlock(&allprison_lock);
4171 	while (nrele--)
4172 		vrele(olddp);
4173 }
4174 
4175 int
4176 descrip_check_write_mp(struct filedesc *fdp, struct mount *mp)
4177 {
4178 	struct file *fp;
4179 	struct vnode *vp;
4180 	int error, i;
4181 
4182 	error = 0;
4183 	FILEDESC_SLOCK(fdp);
4184 	FILEDESC_FOREACH_FP(fdp, i, fp) {
4185 		if (fp->f_type != DTYPE_VNODE ||
4186 		    (atomic_load_int(&fp->f_flag) & FWRITE) == 0)
4187 			continue;
4188 		vp = fp->f_vnode;
4189 		if (vp->v_mount == mp) {
4190 			error = EDEADLK;
4191 			break;
4192 		}
4193 	}
4194 	FILEDESC_SUNLOCK(fdp);
4195 	return (error);
4196 }
4197 
4198 struct filedesc_to_leader *
4199 filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp,
4200     struct proc *leader)
4201 {
4202 	struct filedesc_to_leader *fdtol;
4203 
4204 	fdtol = malloc(sizeof(struct filedesc_to_leader),
4205 	    M_FILEDESC_TO_LEADER, M_WAITOK);
4206 	fdtol->fdl_refcount = 1;
4207 	fdtol->fdl_holdcount = 0;
4208 	fdtol->fdl_wakeup = 0;
4209 	fdtol->fdl_leader = leader;
4210 	if (old != NULL) {
4211 		FILEDESC_XLOCK(fdp);
4212 		fdtol->fdl_next = old->fdl_next;
4213 		fdtol->fdl_prev = old;
4214 		old->fdl_next = fdtol;
4215 		fdtol->fdl_next->fdl_prev = fdtol;
4216 		FILEDESC_XUNLOCK(fdp);
4217 	} else {
4218 		fdtol->fdl_next = fdtol;
4219 		fdtol->fdl_prev = fdtol;
4220 	}
4221 	return (fdtol);
4222 }
4223 
4224 struct filedesc_to_leader *
4225 filedesc_to_leader_share(struct filedesc_to_leader *fdtol, struct filedesc *fdp)
4226 {
4227 	FILEDESC_XLOCK(fdp);
4228 	fdtol->fdl_refcount++;
4229 	FILEDESC_XUNLOCK(fdp);
4230 	return (fdtol);
4231 }
4232 
4233 static int
4234 sysctl_kern_proc_nfds(SYSCTL_HANDLER_ARGS)
4235 {
4236 	NDSLOTTYPE *map;
4237 	struct filedesc *fdp;
4238 	u_int namelen;
4239 	int count, off, minoff;
4240 
4241 	namelen = arg2;
4242 	if (namelen != 1)
4243 		return (EINVAL);
4244 
4245 	if (*(int *)arg1 != 0)
4246 		return (EINVAL);
4247 
4248 	fdp = curproc->p_fd;
4249 	count = 0;
4250 	FILEDESC_SLOCK(fdp);
4251 	map = fdp->fd_map;
4252 	off = NDSLOT(fdp->fd_nfiles - 1);
4253 	for (minoff = NDSLOT(0); off >= minoff; --off)
4254 		count += bitcountl(map[off]);
4255 	FILEDESC_SUNLOCK(fdp);
4256 
4257 	return (SYSCTL_OUT(req, &count, sizeof(count)));
4258 }
4259 
4260 static SYSCTL_NODE(_kern_proc, KERN_PROC_NFDS, nfds,
4261     CTLFLAG_RD|CTLFLAG_CAPRD|CTLFLAG_MPSAFE, sysctl_kern_proc_nfds,
4262     "Number of open file descriptors");
4263 
4264 /*
4265  * Get file structures globally.
4266  */
4267 static int
4268 sysctl_kern_file(SYSCTL_HANDLER_ARGS)
4269 {
4270 	struct xfile xf;
4271 	struct filedesc *fdp;
4272 	struct file *fp;
4273 	struct proc *p;
4274 	int error, n;
4275 
4276 	error = sysctl_wire_old_buffer(req, 0);
4277 	if (error != 0)
4278 		return (error);
4279 	if (req->oldptr == NULL) {
4280 		n = 0;
4281 		sx_slock(&allproc_lock);
4282 		FOREACH_PROC_IN_SYSTEM(p) {
4283 			PROC_LOCK(p);
4284 			if (p->p_state == PRS_NEW) {
4285 				PROC_UNLOCK(p);
4286 				continue;
4287 			}
4288 			fdp = fdhold(p);
4289 			PROC_UNLOCK(p);
4290 			if (fdp == NULL)
4291 				continue;
4292 			/* overestimates sparse tables. */
4293 			n += fdp->fd_nfiles;
4294 			fddrop(fdp);
4295 		}
4296 		sx_sunlock(&allproc_lock);
4297 		return (SYSCTL_OUT(req, 0, n * sizeof(xf)));
4298 	}
4299 	error = 0;
4300 	bzero(&xf, sizeof(xf));
4301 	xf.xf_size = sizeof(xf);
4302 	sx_slock(&allproc_lock);
4303 	FOREACH_PROC_IN_SYSTEM(p) {
4304 		PROC_LOCK(p);
4305 		if (p->p_state == PRS_NEW) {
4306 			PROC_UNLOCK(p);
4307 			continue;
4308 		}
4309 		if (p_cansee(req->td, p) != 0) {
4310 			PROC_UNLOCK(p);
4311 			continue;
4312 		}
4313 		xf.xf_pid = p->p_pid;
4314 		xf.xf_uid = p->p_ucred->cr_uid;
4315 		fdp = fdhold(p);
4316 		PROC_UNLOCK(p);
4317 		if (fdp == NULL)
4318 			continue;
4319 		FILEDESC_SLOCK(fdp);
4320 		if (refcount_load(&fdp->fd_refcnt) == 0)
4321 			goto nextproc;
4322 		FILEDESC_FOREACH_FP(fdp, n, fp) {
4323 			xf.xf_fd = n;
4324 			xf.xf_file = (uintptr_t)fp;
4325 			xf.xf_data = (uintptr_t)fp->f_data;
4326 			xf.xf_vnode = (uintptr_t)fp->f_vnode;
4327 			xf.xf_type = (uintptr_t)fp->f_type;
4328 			xf.xf_count = refcount_load(&fp->f_count);
4329 			xf.xf_msgcount = 0;
4330 			xf.xf_offset = foffset_get(fp);
4331 			xf.xf_flag = fp->f_flag;
4332 			error = SYSCTL_OUT(req, &xf, sizeof(xf));
4333 
4334 			/*
4335 			 * There is no need to re-check the fdtable refcount
4336 			 * here since the filedesc lock is not dropped in the
4337 			 * loop body.
4338 			 */
4339 			if (error != 0)
4340 				break;
4341 		}
4342 nextproc:
4343 		FILEDESC_SUNLOCK(fdp);
4344 		fddrop(fdp);
4345 		if (error)
4346 			break;
4347 	}
4348 	sx_sunlock(&allproc_lock);
4349 	return (error);
4350 }
4351 
4352 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD|CTLFLAG_MPSAFE,
4353     0, 0, sysctl_kern_file, "S,xfile", "Entire file table");
4354 
4355 #ifdef KINFO_FILE_SIZE
4356 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
4357 #endif
4358 
4359 static int
4360 xlate_fflags(int fflags)
4361 {
4362 	static const struct {
4363 		int	fflag;
4364 		int	kf_fflag;
4365 	} fflags_table[] = {
4366 		{ FAPPEND, KF_FLAG_APPEND },
4367 		{ FASYNC, KF_FLAG_ASYNC },
4368 		{ FFSYNC, KF_FLAG_FSYNC },
4369 		{ FHASLOCK, KF_FLAG_HASLOCK },
4370 		{ FNONBLOCK, KF_FLAG_NONBLOCK },
4371 		{ FREAD, KF_FLAG_READ },
4372 		{ FWRITE, KF_FLAG_WRITE },
4373 		{ O_CREAT, KF_FLAG_CREAT },
4374 		{ O_DIRECT, KF_FLAG_DIRECT },
4375 		{ O_EXCL, KF_FLAG_EXCL },
4376 		{ O_EXEC, KF_FLAG_EXEC },
4377 		{ O_EXLOCK, KF_FLAG_EXLOCK },
4378 		{ O_NOFOLLOW, KF_FLAG_NOFOLLOW },
4379 		{ O_SHLOCK, KF_FLAG_SHLOCK },
4380 		{ O_TRUNC, KF_FLAG_TRUNC }
4381 	};
4382 	unsigned int i;
4383 	int kflags;
4384 
4385 	kflags = 0;
4386 	for (i = 0; i < nitems(fflags_table); i++)
4387 		if (fflags & fflags_table[i].fflag)
4388 			kflags |=  fflags_table[i].kf_fflag;
4389 	return (kflags);
4390 }
4391 
4392 /* Trim unused data from kf_path by truncating the structure size. */
4393 void
4394 pack_kinfo(struct kinfo_file *kif)
4395 {
4396 
4397 	kif->kf_structsize = offsetof(struct kinfo_file, kf_path) +
4398 	    strlen(kif->kf_path) + 1;
4399 	kif->kf_structsize = roundup(kif->kf_structsize, sizeof(uint64_t));
4400 }
4401 
4402 static void
4403 export_file_to_kinfo(struct file *fp, int fd, cap_rights_t *rightsp,
4404     struct kinfo_file *kif, struct filedesc *fdp, int flags)
4405 {
4406 	int error;
4407 
4408 	bzero(kif, sizeof(*kif));
4409 
4410 	/* Set a default type to allow for empty fill_kinfo() methods. */
4411 	kif->kf_type = KF_TYPE_UNKNOWN;
4412 	kif->kf_flags = xlate_fflags(fp->f_flag);
4413 	if (rightsp != NULL)
4414 		kif->kf_cap_rights = *rightsp;
4415 	else
4416 		cap_rights_init_zero(&kif->kf_cap_rights);
4417 	kif->kf_fd = fd;
4418 	kif->kf_ref_count = refcount_load(&fp->f_count);
4419 	kif->kf_offset = foffset_get(fp);
4420 
4421 	/*
4422 	 * This may drop the filedesc lock, so the 'fp' cannot be
4423 	 * accessed after this call.
4424 	 */
4425 	error = fo_fill_kinfo(fp, kif, fdp);
4426 	if (error == 0)
4427 		kif->kf_status |= KF_ATTR_VALID;
4428 	if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
4429 		pack_kinfo(kif);
4430 	else
4431 		kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
4432 }
4433 
4434 static void
4435 export_vnode_to_kinfo(struct vnode *vp, int fd, int fflags,
4436     struct kinfo_file *kif, int flags)
4437 {
4438 	int error;
4439 
4440 	bzero(kif, sizeof(*kif));
4441 
4442 	kif->kf_type = KF_TYPE_VNODE;
4443 	error = vn_fill_kinfo_vnode(vp, kif);
4444 	if (error == 0)
4445 		kif->kf_status |= KF_ATTR_VALID;
4446 	kif->kf_flags = xlate_fflags(fflags);
4447 	cap_rights_init_zero(&kif->kf_cap_rights);
4448 	kif->kf_fd = fd;
4449 	kif->kf_ref_count = -1;
4450 	kif->kf_offset = -1;
4451 	if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
4452 		pack_kinfo(kif);
4453 	else
4454 		kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
4455 	vrele(vp);
4456 }
4457 
4458 struct export_fd_buf {
4459 	struct filedesc		*fdp;
4460 	struct pwddesc	*pdp;
4461 	struct sbuf 		*sb;
4462 	ssize_t			remainder;
4463 	struct kinfo_file	kif;
4464 	int			flags;
4465 };
4466 
4467 static int
4468 export_kinfo_to_sb(struct export_fd_buf *efbuf)
4469 {
4470 	struct kinfo_file *kif;
4471 
4472 	kif = &efbuf->kif;
4473 	if (efbuf->remainder != -1) {
4474 		if (efbuf->remainder < kif->kf_structsize)
4475 			return (ENOMEM);
4476 		efbuf->remainder -= kif->kf_structsize;
4477 	}
4478 	if (sbuf_bcat(efbuf->sb, kif, kif->kf_structsize) != 0)
4479 		return (sbuf_error(efbuf->sb));
4480 	return (0);
4481 }
4482 
4483 static int
4484 export_file_to_sb(struct file *fp, int fd, cap_rights_t *rightsp,
4485     struct export_fd_buf *efbuf)
4486 {
4487 	int error;
4488 
4489 	if (efbuf->remainder == 0)
4490 		return (ENOMEM);
4491 	export_file_to_kinfo(fp, fd, rightsp, &efbuf->kif, efbuf->fdp,
4492 	    efbuf->flags);
4493 	FILEDESC_SUNLOCK(efbuf->fdp);
4494 	error = export_kinfo_to_sb(efbuf);
4495 	FILEDESC_SLOCK(efbuf->fdp);
4496 	return (error);
4497 }
4498 
4499 static int
4500 export_vnode_to_sb(struct vnode *vp, int fd, int fflags,
4501     struct export_fd_buf *efbuf)
4502 {
4503 	int error;
4504 
4505 	if (efbuf->remainder == 0)
4506 		return (ENOMEM);
4507 	if (efbuf->pdp != NULL)
4508 		PWDDESC_XUNLOCK(efbuf->pdp);
4509 	export_vnode_to_kinfo(vp, fd, fflags, &efbuf->kif, efbuf->flags);
4510 	error = export_kinfo_to_sb(efbuf);
4511 	if (efbuf->pdp != NULL)
4512 		PWDDESC_XLOCK(efbuf->pdp);
4513 	return (error);
4514 }
4515 
4516 /*
4517  * Store a process file descriptor information to sbuf.
4518  *
4519  * Takes a locked proc as argument, and returns with the proc unlocked.
4520  */
4521 int
4522 kern_proc_filedesc_out(struct proc *p,  struct sbuf *sb, ssize_t maxlen,
4523     int flags)
4524 {
4525 	struct file *fp;
4526 	struct filedesc *fdp;
4527 	struct pwddesc *pdp;
4528 	struct export_fd_buf *efbuf;
4529 	struct vnode *cttyvp, *textvp, *tracevp;
4530 	struct pwd *pwd;
4531 	int error, i;
4532 	cap_rights_t rights;
4533 
4534 	PROC_LOCK_ASSERT(p, MA_OWNED);
4535 
4536 	/* ktrace vnode */
4537 	tracevp = ktr_get_tracevp(p, true);
4538 	/* text vnode */
4539 	textvp = p->p_textvp;
4540 	if (textvp != NULL)
4541 		vrefact(textvp);
4542 	/* Controlling tty. */
4543 	cttyvp = NULL;
4544 	if (p->p_pgrp != NULL && p->p_pgrp->pg_session != NULL) {
4545 		cttyvp = p->p_pgrp->pg_session->s_ttyvp;
4546 		if (cttyvp != NULL)
4547 			vrefact(cttyvp);
4548 	}
4549 	fdp = fdhold(p);
4550 	pdp = pdhold(p);
4551 	PROC_UNLOCK(p);
4552 
4553 	efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
4554 	efbuf->fdp = NULL;
4555 	efbuf->pdp = NULL;
4556 	efbuf->sb = sb;
4557 	efbuf->remainder = maxlen;
4558 	efbuf->flags = flags;
4559 
4560 	error = 0;
4561 	if (tracevp != NULL)
4562 		error = export_vnode_to_sb(tracevp, KF_FD_TYPE_TRACE,
4563 		    FREAD | FWRITE, efbuf);
4564 	if (error == 0 && textvp != NULL)
4565 		error = export_vnode_to_sb(textvp, KF_FD_TYPE_TEXT, FREAD,
4566 		    efbuf);
4567 	if (error == 0 && cttyvp != NULL)
4568 		error = export_vnode_to_sb(cttyvp, KF_FD_TYPE_CTTY,
4569 		    FREAD | FWRITE, efbuf);
4570 	if (error != 0 || pdp == NULL || fdp == NULL)
4571 		goto fail;
4572 	efbuf->fdp = fdp;
4573 	efbuf->pdp = pdp;
4574 	PWDDESC_XLOCK(pdp);
4575 	pwd = pwd_hold_pwddesc(pdp);
4576 	if (pwd != NULL) {
4577 		/* working directory */
4578 		if (pwd->pwd_cdir != NULL) {
4579 			vrefact(pwd->pwd_cdir);
4580 			error = export_vnode_to_sb(pwd->pwd_cdir,
4581 			    KF_FD_TYPE_CWD, FREAD, efbuf);
4582 		}
4583 		/* root directory */
4584 		if (error == 0 && pwd->pwd_rdir != NULL) {
4585 			vrefact(pwd->pwd_rdir);
4586 			error = export_vnode_to_sb(pwd->pwd_rdir,
4587 			    KF_FD_TYPE_ROOT, FREAD, efbuf);
4588 		}
4589 		/* jail directory */
4590 		if (error == 0 && pwd->pwd_jdir != NULL) {
4591 			vrefact(pwd->pwd_jdir);
4592 			error = export_vnode_to_sb(pwd->pwd_jdir,
4593 			    KF_FD_TYPE_JAIL, FREAD, efbuf);
4594 		}
4595 	}
4596 	PWDDESC_XUNLOCK(pdp);
4597 	if (error != 0)
4598 		goto fail;
4599 	if (pwd != NULL)
4600 		pwd_drop(pwd);
4601 	FILEDESC_SLOCK(fdp);
4602 	if (refcount_load(&fdp->fd_refcnt) == 0)
4603 		goto skip;
4604 	FILEDESC_FOREACH_FP(fdp, i, fp) {
4605 #ifdef CAPABILITIES
4606 		rights = *cap_rights(fdp, i);
4607 #else /* !CAPABILITIES */
4608 		rights = cap_no_rights;
4609 #endif
4610 		/*
4611 		 * Create sysctl entry.  It is OK to drop the filedesc
4612 		 * lock inside of export_file_to_sb() as we will
4613 		 * re-validate and re-evaluate its properties when the
4614 		 * loop continues.
4615 		 */
4616 		error = export_file_to_sb(fp, i, &rights, efbuf);
4617 		if (error != 0 || refcount_load(&fdp->fd_refcnt) == 0)
4618 			break;
4619 	}
4620 skip:
4621 	FILEDESC_SUNLOCK(fdp);
4622 fail:
4623 	if (fdp != NULL)
4624 		fddrop(fdp);
4625 	if (pdp != NULL)
4626 		pddrop(pdp);
4627 	free(efbuf, M_TEMP);
4628 	return (error);
4629 }
4630 
4631 #define FILEDESC_SBUF_SIZE	(sizeof(struct kinfo_file) * 5)
4632 
4633 /*
4634  * Get per-process file descriptors for use by procstat(1), et al.
4635  */
4636 static int
4637 sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS)
4638 {
4639 	struct sbuf sb;
4640 	struct proc *p;
4641 	ssize_t maxlen;
4642 	u_int namelen;
4643 	int error, error2, *name;
4644 
4645 	namelen = arg2;
4646 	if (namelen != 1)
4647 		return (EINVAL);
4648 
4649 	name = (int *)arg1;
4650 
4651 	sbuf_new_for_sysctl(&sb, NULL, FILEDESC_SBUF_SIZE, req);
4652 	sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
4653 	error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
4654 	if (error != 0) {
4655 		sbuf_delete(&sb);
4656 		return (error);
4657 	}
4658 	maxlen = req->oldptr != NULL ? req->oldlen : -1;
4659 	error = kern_proc_filedesc_out(p, &sb, maxlen,
4660 	    KERN_FILEDESC_PACK_KINFO);
4661 	error2 = sbuf_finish(&sb);
4662 	sbuf_delete(&sb);
4663 	return (error != 0 ? error : error2);
4664 }
4665 
4666 #ifdef COMPAT_FREEBSD7
4667 #ifdef KINFO_OFILE_SIZE
4668 CTASSERT(sizeof(struct kinfo_ofile) == KINFO_OFILE_SIZE);
4669 #endif
4670 
4671 static void
4672 kinfo_to_okinfo(struct kinfo_file *kif, struct kinfo_ofile *okif)
4673 {
4674 
4675 	okif->kf_structsize = sizeof(*okif);
4676 	okif->kf_type = kif->kf_type;
4677 	okif->kf_fd = kif->kf_fd;
4678 	okif->kf_ref_count = kif->kf_ref_count;
4679 	okif->kf_flags = kif->kf_flags & (KF_FLAG_READ | KF_FLAG_WRITE |
4680 	    KF_FLAG_APPEND | KF_FLAG_ASYNC | KF_FLAG_FSYNC | KF_FLAG_NONBLOCK |
4681 	    KF_FLAG_DIRECT | KF_FLAG_HASLOCK);
4682 	okif->kf_offset = kif->kf_offset;
4683 	if (kif->kf_type == KF_TYPE_VNODE)
4684 		okif->kf_vnode_type = kif->kf_un.kf_file.kf_file_type;
4685 	else
4686 		okif->kf_vnode_type = KF_VTYPE_VNON;
4687 	strlcpy(okif->kf_path, kif->kf_path, sizeof(okif->kf_path));
4688 	if (kif->kf_type == KF_TYPE_SOCKET) {
4689 		okif->kf_sock_domain = kif->kf_un.kf_sock.kf_sock_domain0;
4690 		okif->kf_sock_type = kif->kf_un.kf_sock.kf_sock_type0;
4691 		okif->kf_sock_protocol = kif->kf_un.kf_sock.kf_sock_protocol0;
4692 		okif->kf_sa_local = kif->kf_un.kf_sock.kf_sa_local;
4693 		okif->kf_sa_peer = kif->kf_un.kf_sock.kf_sa_peer;
4694 	} else {
4695 		okif->kf_sa_local.ss_family = AF_UNSPEC;
4696 		okif->kf_sa_peer.ss_family = AF_UNSPEC;
4697 	}
4698 }
4699 
4700 static int
4701 export_vnode_for_osysctl(struct vnode *vp, int type, struct kinfo_file *kif,
4702     struct kinfo_ofile *okif, struct pwddesc *pdp, struct sysctl_req *req)
4703 {
4704 	int error;
4705 
4706 	vrefact(vp);
4707 	PWDDESC_XUNLOCK(pdp);
4708 	export_vnode_to_kinfo(vp, type, 0, kif, KERN_FILEDESC_PACK_KINFO);
4709 	kinfo_to_okinfo(kif, okif);
4710 	error = SYSCTL_OUT(req, okif, sizeof(*okif));
4711 	PWDDESC_XLOCK(pdp);
4712 	return (error);
4713 }
4714 
4715 /*
4716  * Get per-process file descriptors for use by procstat(1), et al.
4717  */
4718 static int
4719 sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS)
4720 {
4721 	struct kinfo_ofile *okif;
4722 	struct kinfo_file *kif;
4723 	struct filedesc *fdp;
4724 	struct pwddesc *pdp;
4725 	struct pwd *pwd;
4726 	u_int namelen;
4727 	int error, i, *name;
4728 	struct file *fp;
4729 	struct proc *p;
4730 
4731 	namelen = arg2;
4732 	if (namelen != 1)
4733 		return (EINVAL);
4734 
4735 	name = (int *)arg1;
4736 	error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
4737 	if (error != 0)
4738 		return (error);
4739 	fdp = fdhold(p);
4740 	if (fdp != NULL)
4741 		pdp = pdhold(p);
4742 	PROC_UNLOCK(p);
4743 	if (fdp == NULL || pdp == NULL) {
4744 		if (fdp != NULL)
4745 			fddrop(fdp);
4746 		return (ENOENT);
4747 	}
4748 	kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK);
4749 	okif = malloc(sizeof(*okif), M_TEMP, M_WAITOK);
4750 	PWDDESC_XLOCK(pdp);
4751 	pwd = pwd_hold_pwddesc(pdp);
4752 	if (pwd != NULL) {
4753 		if (pwd->pwd_cdir != NULL)
4754 			export_vnode_for_osysctl(pwd->pwd_cdir, KF_FD_TYPE_CWD, kif,
4755 			    okif, pdp, req);
4756 		if (pwd->pwd_rdir != NULL)
4757 			export_vnode_for_osysctl(pwd->pwd_rdir, KF_FD_TYPE_ROOT, kif,
4758 			    okif, pdp, req);
4759 		if (pwd->pwd_jdir != NULL)
4760 			export_vnode_for_osysctl(pwd->pwd_jdir, KF_FD_TYPE_JAIL, kif,
4761 			    okif, pdp, req);
4762 	}
4763 	PWDDESC_XUNLOCK(pdp);
4764 	if (pwd != NULL)
4765 		pwd_drop(pwd);
4766 	FILEDESC_SLOCK(fdp);
4767 	if (refcount_load(&fdp->fd_refcnt) == 0)
4768 		goto skip;
4769 	FILEDESC_FOREACH_FP(fdp, i, fp) {
4770 		export_file_to_kinfo(fp, i, NULL, kif, fdp,
4771 		    KERN_FILEDESC_PACK_KINFO);
4772 		FILEDESC_SUNLOCK(fdp);
4773 		kinfo_to_okinfo(kif, okif);
4774 		error = SYSCTL_OUT(req, okif, sizeof(*okif));
4775 		FILEDESC_SLOCK(fdp);
4776 		if (error != 0 || refcount_load(&fdp->fd_refcnt) == 0)
4777 			break;
4778 	}
4779 skip:
4780 	FILEDESC_SUNLOCK(fdp);
4781 	fddrop(fdp);
4782 	pddrop(pdp);
4783 	free(kif, M_TEMP);
4784 	free(okif, M_TEMP);
4785 	return (0);
4786 }
4787 
4788 static SYSCTL_NODE(_kern_proc, KERN_PROC_OFILEDESC, ofiledesc,
4789     CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_ofiledesc,
4790     "Process ofiledesc entries");
4791 #endif	/* COMPAT_FREEBSD7 */
4792 
4793 int
4794 vntype_to_kinfo(int vtype)
4795 {
4796 	struct {
4797 		int	vtype;
4798 		int	kf_vtype;
4799 	} vtypes_table[] = {
4800 		{ VBAD, KF_VTYPE_VBAD },
4801 		{ VBLK, KF_VTYPE_VBLK },
4802 		{ VCHR, KF_VTYPE_VCHR },
4803 		{ VDIR, KF_VTYPE_VDIR },
4804 		{ VFIFO, KF_VTYPE_VFIFO },
4805 		{ VLNK, KF_VTYPE_VLNK },
4806 		{ VNON, KF_VTYPE_VNON },
4807 		{ VREG, KF_VTYPE_VREG },
4808 		{ VSOCK, KF_VTYPE_VSOCK }
4809 	};
4810 	unsigned int i;
4811 
4812 	/*
4813 	 * Perform vtype translation.
4814 	 */
4815 	for (i = 0; i < nitems(vtypes_table); i++)
4816 		if (vtypes_table[i].vtype == vtype)
4817 			return (vtypes_table[i].kf_vtype);
4818 
4819 	return (KF_VTYPE_UNKNOWN);
4820 }
4821 
4822 static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc,
4823     CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_filedesc,
4824     "Process filedesc entries");
4825 
4826 /*
4827  * Store a process current working directory information to sbuf.
4828  *
4829  * Takes a locked proc as argument, and returns with the proc unlocked.
4830  */
4831 int
4832 kern_proc_cwd_out(struct proc *p,  struct sbuf *sb, ssize_t maxlen)
4833 {
4834 	struct pwddesc *pdp;
4835 	struct pwd *pwd;
4836 	struct export_fd_buf *efbuf;
4837 	struct vnode *cdir;
4838 	int error;
4839 
4840 	PROC_LOCK_ASSERT(p, MA_OWNED);
4841 
4842 	pdp = pdhold(p);
4843 	PROC_UNLOCK(p);
4844 	if (pdp == NULL)
4845 		return (EINVAL);
4846 
4847 	efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
4848 	efbuf->fdp = NULL;
4849 	efbuf->pdp = pdp;
4850 	efbuf->sb = sb;
4851 	efbuf->remainder = maxlen;
4852 	efbuf->flags = 0;
4853 
4854 	PWDDESC_XLOCK(pdp);
4855 	pwd = PWDDESC_XLOCKED_LOAD_PWD(pdp);
4856 	cdir = pwd->pwd_cdir;
4857 	if (cdir == NULL) {
4858 		error = EINVAL;
4859 	} else {
4860 		vrefact(cdir);
4861 		error = export_vnode_to_sb(cdir, KF_FD_TYPE_CWD, FREAD, efbuf);
4862 	}
4863 	PWDDESC_XUNLOCK(pdp);
4864 	pddrop(pdp);
4865 	free(efbuf, M_TEMP);
4866 	return (error);
4867 }
4868 
4869 /*
4870  * Get per-process current working directory.
4871  */
4872 static int
4873 sysctl_kern_proc_cwd(SYSCTL_HANDLER_ARGS)
4874 {
4875 	struct sbuf sb;
4876 	struct proc *p;
4877 	ssize_t maxlen;
4878 	u_int namelen;
4879 	int error, error2, *name;
4880 
4881 	namelen = arg2;
4882 	if (namelen != 1)
4883 		return (EINVAL);
4884 
4885 	name = (int *)arg1;
4886 
4887 	sbuf_new_for_sysctl(&sb, NULL, sizeof(struct kinfo_file), req);
4888 	sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
4889 	error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
4890 	if (error != 0) {
4891 		sbuf_delete(&sb);
4892 		return (error);
4893 	}
4894 	maxlen = req->oldptr != NULL ? req->oldlen : -1;
4895 	error = kern_proc_cwd_out(p, &sb, maxlen);
4896 	error2 = sbuf_finish(&sb);
4897 	sbuf_delete(&sb);
4898 	return (error != 0 ? error : error2);
4899 }
4900 
4901 static SYSCTL_NODE(_kern_proc, KERN_PROC_CWD, cwd, CTLFLAG_RD|CTLFLAG_MPSAFE,
4902     sysctl_kern_proc_cwd, "Process current working directory");
4903 
4904 #ifdef DDB
4905 /*
4906  * For the purposes of debugging, generate a human-readable string for the
4907  * file type.
4908  */
4909 static const char *
4910 file_type_to_name(short type)
4911 {
4912 
4913 	switch (type) {
4914 	case 0:
4915 		return ("zero");
4916 	case DTYPE_VNODE:
4917 		return ("vnode");
4918 	case DTYPE_SOCKET:
4919 		return ("socket");
4920 	case DTYPE_PIPE:
4921 		return ("pipe");
4922 	case DTYPE_FIFO:
4923 		return ("fifo");
4924 	case DTYPE_KQUEUE:
4925 		return ("kqueue");
4926 	case DTYPE_CRYPTO:
4927 		return ("crypto");
4928 	case DTYPE_MQUEUE:
4929 		return ("mqueue");
4930 	case DTYPE_SHM:
4931 		return ("shm");
4932 	case DTYPE_SEM:
4933 		return ("ksem");
4934 	case DTYPE_PTS:
4935 		return ("pts");
4936 	case DTYPE_DEV:
4937 		return ("dev");
4938 	case DTYPE_PROCDESC:
4939 		return ("proc");
4940 	case DTYPE_EVENTFD:
4941 		return ("eventfd");
4942 	case DTYPE_LINUXTFD:
4943 		return ("ltimer");
4944 	default:
4945 		return ("unkn");
4946 	}
4947 }
4948 
4949 /*
4950  * For the purposes of debugging, identify a process (if any, perhaps one of
4951  * many) that references the passed file in its file descriptor array. Return
4952  * NULL if none.
4953  */
4954 static struct proc *
4955 file_to_first_proc(struct file *fp)
4956 {
4957 	struct filedesc *fdp;
4958 	struct proc *p;
4959 	int n;
4960 
4961 	FOREACH_PROC_IN_SYSTEM(p) {
4962 		if (p->p_state == PRS_NEW)
4963 			continue;
4964 		fdp = p->p_fd;
4965 		if (fdp == NULL)
4966 			continue;
4967 		for (n = 0; n < fdp->fd_nfiles; n++) {
4968 			if (fp == fdp->fd_ofiles[n].fde_file)
4969 				return (p);
4970 		}
4971 	}
4972 	return (NULL);
4973 }
4974 
4975 static void
4976 db_print_file(struct file *fp, int header)
4977 {
4978 #define XPTRWIDTH ((int)howmany(sizeof(void *) * NBBY, 4))
4979 	struct proc *p;
4980 
4981 	if (header)
4982 		db_printf("%*s %6s %*s %8s %4s %5s %6s %*s %5s %s\n",
4983 		    XPTRWIDTH, "File", "Type", XPTRWIDTH, "Data", "Flag",
4984 		    "GCFl", "Count", "MCount", XPTRWIDTH, "Vnode", "FPID",
4985 		    "FCmd");
4986 	p = file_to_first_proc(fp);
4987 	db_printf("%*p %6s %*p %08x %04x %5d %6d %*p %5d %s\n", XPTRWIDTH,
4988 	    fp, file_type_to_name(fp->f_type), XPTRWIDTH, fp->f_data,
4989 	    fp->f_flag, 0, refcount_load(&fp->f_count), 0, XPTRWIDTH, fp->f_vnode,
4990 	    p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-");
4991 
4992 #undef XPTRWIDTH
4993 }
4994 
4995 DB_SHOW_COMMAND(file, db_show_file)
4996 {
4997 	struct file *fp;
4998 
4999 	if (!have_addr) {
5000 		db_printf("usage: show file <addr>\n");
5001 		return;
5002 	}
5003 	fp = (struct file *)addr;
5004 	db_print_file(fp, 1);
5005 }
5006 
5007 DB_SHOW_COMMAND_FLAGS(files, db_show_files, DB_CMD_MEMSAFE)
5008 {
5009 	struct filedesc *fdp;
5010 	struct file *fp;
5011 	struct proc *p;
5012 	int header;
5013 	int n;
5014 
5015 	header = 1;
5016 	FOREACH_PROC_IN_SYSTEM(p) {
5017 		if (p->p_state == PRS_NEW)
5018 			continue;
5019 		if ((fdp = p->p_fd) == NULL)
5020 			continue;
5021 		for (n = 0; n < fdp->fd_nfiles; ++n) {
5022 			if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
5023 				continue;
5024 			db_print_file(fp, header);
5025 			header = 0;
5026 		}
5027 	}
5028 }
5029 #endif
5030 
5031 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
5032     &maxfilesperproc, 0, "Maximum files allowed open per process");
5033 
5034 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
5035     &maxfiles, 0, "Maximum number of files");
5036 
5037 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
5038     &openfiles, 0, "System-wide number of open files");
5039 
5040 /* ARGSUSED*/
5041 static void
5042 filelistinit(void *dummy)
5043 {
5044 
5045 	file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL,
5046 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
5047 	filedesc0_zone = uma_zcreate("filedesc0", sizeof(struct filedesc0),
5048 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
5049 	pwd_zone = uma_zcreate("PWD", sizeof(struct pwd), NULL, NULL,
5050 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_SMR);
5051 	/*
5052 	 * XXXMJG this is a temporary hack due to boot ordering issues against
5053 	 * the vnode zone.
5054 	 */
5055 	vfs_smr = uma_zone_get_smr(pwd_zone);
5056 	mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF);
5057 }
5058 SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL);
5059 
5060 /*-------------------------------------------------------------------*/
5061 
5062 static int
5063 badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred,
5064     int flags, struct thread *td)
5065 {
5066 
5067 	return (EBADF);
5068 }
5069 
5070 static int
5071 badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
5072     struct thread *td)
5073 {
5074 
5075 	return (EINVAL);
5076 }
5077 
5078 static int
5079 badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred,
5080     struct thread *td)
5081 {
5082 
5083 	return (EBADF);
5084 }
5085 
5086 static int
5087 badfo_poll(struct file *fp, int events, struct ucred *active_cred,
5088     struct thread *td)
5089 {
5090 
5091 	return (0);
5092 }
5093 
5094 static int
5095 badfo_kqfilter(struct file *fp, struct knote *kn)
5096 {
5097 
5098 	return (EBADF);
5099 }
5100 
5101 static int
5102 badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred)
5103 {
5104 
5105 	return (EBADF);
5106 }
5107 
5108 static int
5109 badfo_close(struct file *fp, struct thread *td)
5110 {
5111 
5112 	return (0);
5113 }
5114 
5115 static int
5116 badfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
5117     struct thread *td)
5118 {
5119 
5120 	return (EBADF);
5121 }
5122 
5123 static int
5124 badfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
5125     struct thread *td)
5126 {
5127 
5128 	return (EBADF);
5129 }
5130 
5131 static int
5132 badfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
5133     struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
5134     struct thread *td)
5135 {
5136 
5137 	return (EBADF);
5138 }
5139 
5140 static int
5141 badfo_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
5142 {
5143 
5144 	return (0);
5145 }
5146 
5147 struct fileops badfileops = {
5148 	.fo_read = badfo_readwrite,
5149 	.fo_write = badfo_readwrite,
5150 	.fo_truncate = badfo_truncate,
5151 	.fo_ioctl = badfo_ioctl,
5152 	.fo_poll = badfo_poll,
5153 	.fo_kqfilter = badfo_kqfilter,
5154 	.fo_stat = badfo_stat,
5155 	.fo_close = badfo_close,
5156 	.fo_chmod = badfo_chmod,
5157 	.fo_chown = badfo_chown,
5158 	.fo_sendfile = badfo_sendfile,
5159 	.fo_fill_kinfo = badfo_fill_kinfo,
5160 };
5161 
5162 static int
5163 path_poll(struct file *fp, int events, struct ucred *active_cred,
5164     struct thread *td)
5165 {
5166 	return (POLLNVAL);
5167 }
5168 
5169 static int
5170 path_close(struct file *fp, struct thread *td)
5171 {
5172 	MPASS(fp->f_type == DTYPE_VNODE);
5173 	fp->f_ops = &badfileops;
5174 	vrele(fp->f_vnode);
5175 	return (0);
5176 }
5177 
5178 struct fileops path_fileops = {
5179 	.fo_read = badfo_readwrite,
5180 	.fo_write = badfo_readwrite,
5181 	.fo_truncate = badfo_truncate,
5182 	.fo_ioctl = badfo_ioctl,
5183 	.fo_poll = path_poll,
5184 	.fo_kqfilter = vn_kqfilter_opath,
5185 	.fo_stat = vn_statfile,
5186 	.fo_close = path_close,
5187 	.fo_chmod = badfo_chmod,
5188 	.fo_chown = badfo_chown,
5189 	.fo_sendfile = badfo_sendfile,
5190 	.fo_fill_kinfo = vn_fill_kinfo,
5191 	.fo_flags = DFLAG_PASSABLE,
5192 };
5193 
5194 int
5195 invfo_rdwr(struct file *fp, struct uio *uio, struct ucred *active_cred,
5196     int flags, struct thread *td)
5197 {
5198 
5199 	return (EOPNOTSUPP);
5200 }
5201 
5202 int
5203 invfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
5204     struct thread *td)
5205 {
5206 
5207 	return (EINVAL);
5208 }
5209 
5210 int
5211 invfo_ioctl(struct file *fp, u_long com, void *data,
5212     struct ucred *active_cred, struct thread *td)
5213 {
5214 
5215 	return (ENOTTY);
5216 }
5217 
5218 int
5219 invfo_poll(struct file *fp, int events, struct ucred *active_cred,
5220     struct thread *td)
5221 {
5222 
5223 	return (poll_no_poll(events));
5224 }
5225 
5226 int
5227 invfo_kqfilter(struct file *fp, struct knote *kn)
5228 {
5229 
5230 	return (EINVAL);
5231 }
5232 
5233 int
5234 invfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
5235     struct thread *td)
5236 {
5237 
5238 	return (EINVAL);
5239 }
5240 
5241 int
5242 invfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
5243     struct thread *td)
5244 {
5245 
5246 	return (EINVAL);
5247 }
5248 
5249 int
5250 invfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
5251     struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
5252     struct thread *td)
5253 {
5254 
5255 	return (EINVAL);
5256 }
5257 
5258 /*-------------------------------------------------------------------*/
5259 
5260 /*
5261  * File Descriptor pseudo-device driver (/dev/fd/).
5262  *
5263  * Opening minor device N dup()s the file (if any) connected to file
5264  * descriptor N belonging to the calling process.  Note that this driver
5265  * consists of only the ``open()'' routine, because all subsequent
5266  * references to this file will be direct to the other driver.
5267  *
5268  * XXX: we could give this one a cloning event handler if necessary.
5269  */
5270 
5271 /* ARGSUSED */
5272 static int
5273 fdopen(struct cdev *dev, int mode, int type, struct thread *td)
5274 {
5275 
5276 	/*
5277 	 * XXX Kludge: set curthread->td_dupfd to contain the value of the
5278 	 * the file descriptor being sought for duplication. The error
5279 	 * return ensures that the vnode for this device will be released
5280 	 * by vn_open. Open will detect this special error and take the
5281 	 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
5282 	 * will simply report the error.
5283 	 */
5284 	td->td_dupfd = dev2unit(dev);
5285 	return (ENODEV);
5286 }
5287 
5288 static struct cdevsw fildesc_cdevsw = {
5289 	.d_version =	D_VERSION,
5290 	.d_open =	fdopen,
5291 	.d_name =	"FD",
5292 };
5293 
5294 static void
5295 fildesc_drvinit(void *unused)
5296 {
5297 	struct cdev *dev;
5298 
5299 	dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 0, NULL,
5300 	    UID_ROOT, GID_WHEEL, 0666, "fd/0");
5301 	make_dev_alias(dev, "stdin");
5302 	dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 1, NULL,
5303 	    UID_ROOT, GID_WHEEL, 0666, "fd/1");
5304 	make_dev_alias(dev, "stdout");
5305 	dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 2, NULL,
5306 	    UID_ROOT, GID_WHEEL, 0666, "fd/2");
5307 	make_dev_alias(dev, "stderr");
5308 }
5309 
5310 SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL);
5311