xref: /freebsd/sys/kern/kern_descrip.c (revision 06e20d1babecec1f45ffda513f55a8db5f1c0f56)
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/priv.h>
65 #include <sys/proc.h>
66 #include <sys/protosw.h>
67 #include <sys/racct.h>
68 #include <sys/resourcevar.h>
69 #include <sys/sbuf.h>
70 #include <sys/signalvar.h>
71 #include <sys/kdb.h>
72 #include <sys/smr.h>
73 #include <sys/stat.h>
74 #include <sys/sx.h>
75 #include <sys/syscallsubr.h>
76 #include <sys/sysctl.h>
77 #include <sys/sysproto.h>
78 #include <sys/unistd.h>
79 #include <sys/user.h>
80 #include <sys/vnode.h>
81 #ifdef KTRACE
82 #include <sys/ktrace.h>
83 #endif
84 
85 #include <net/vnet.h>
86 
87 #include <security/audit/audit.h>
88 
89 #include <vm/uma.h>
90 #include <vm/vm.h>
91 
92 #include <ddb/ddb.h>
93 
94 static MALLOC_DEFINE(M_FILEDESC, "filedesc", "Open file descriptor table");
95 static MALLOC_DEFINE(M_PWD, "pwd", "Descriptor table vnodes");
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, int holdleaders);
110 static int	fd_first_free(struct filedesc *fdp, int low, int size);
111 static void	fdgrowtable(struct filedesc *fdp, int nfd);
112 static void	fdgrowtable_exp(struct filedesc *fdp, int nfd);
113 static void	fdunused(struct filedesc *fdp, int fd);
114 static void	fdused(struct filedesc *fdp, int fd);
115 static int	getmaxfd(struct thread *td);
116 static u_long	*filecaps_copy_prep(const struct filecaps *src);
117 static void	filecaps_copy_finish(const struct filecaps *src,
118 		    struct filecaps *dst, u_long *ioctls);
119 static u_long 	*filecaps_free_prep(struct filecaps *fcaps);
120 static void	filecaps_free_finish(u_long *ioctls);
121 
122 static struct pwd *pwd_alloc(void);
123 
124 /*
125  * Each process has:
126  *
127  * - An array of open file descriptors (fd_ofiles)
128  * - An array of file flags (fd_ofileflags)
129  * - A bitmap recording which descriptors are in use (fd_map)
130  *
131  * A process starts out with NDFILE descriptors.  The value of NDFILE has
132  * been selected based the historical limit of 20 open files, and an
133  * assumption that the majority of processes, especially short-lived
134  * processes like shells, will never need more.
135  *
136  * If this initial allocation is exhausted, a larger descriptor table and
137  * map are allocated dynamically, and the pointers in the process's struct
138  * filedesc are updated to point to those.  This is repeated every time
139  * the process runs out of file descriptors (provided it hasn't hit its
140  * resource limit).
141  *
142  * Since threads may hold references to individual descriptor table
143  * entries, the tables are never freed.  Instead, they are placed on a
144  * linked list and freed only when the struct filedesc is released.
145  */
146 #define NDFILE		20
147 #define NDSLOTSIZE	sizeof(NDSLOTTYPE)
148 #define	NDENTRIES	(NDSLOTSIZE * __CHAR_BIT)
149 #define NDSLOT(x)	((x) / NDENTRIES)
150 #define NDBIT(x)	((NDSLOTTYPE)1 << ((x) % NDENTRIES))
151 #define	NDSLOTS(x)	(((x) + NDENTRIES - 1) / NDENTRIES)
152 
153 /*
154  * SLIST entry used to keep track of ofiles which must be reclaimed when
155  * the process exits.
156  */
157 struct freetable {
158 	struct fdescenttbl *ft_table;
159 	SLIST_ENTRY(freetable) ft_next;
160 };
161 
162 /*
163  * Initial allocation: a filedesc structure + the head of SLIST used to
164  * keep track of old ofiles + enough space for NDFILE descriptors.
165  */
166 
167 struct fdescenttbl0 {
168 	int	fdt_nfiles;
169 	struct	filedescent fdt_ofiles[NDFILE];
170 };
171 
172 struct filedesc0 {
173 	struct filedesc fd_fd;
174 	SLIST_HEAD(, freetable) fd_free;
175 	struct	fdescenttbl0 fd_dfiles;
176 	NDSLOTTYPE fd_dmap[NDSLOTS(NDFILE)];
177 };
178 
179 /*
180  * Descriptor management.
181  */
182 static int __exclusive_cache_line openfiles; /* actual number of open files */
183 struct mtx sigio_lock;		/* mtx to protect pointers to sigio */
184 void __read_mostly (*mq_fdclose)(struct thread *td, int fd, struct file *fp);
185 
186 /*
187  * If low >= size, just return low. Otherwise find the first zero bit in the
188  * given bitmap, starting at low and not exceeding size - 1. Return size if
189  * not found.
190  */
191 static int
192 fd_first_free(struct filedesc *fdp, int low, int size)
193 {
194 	NDSLOTTYPE *map = fdp->fd_map;
195 	NDSLOTTYPE mask;
196 	int off, maxoff;
197 
198 	if (low >= size)
199 		return (low);
200 
201 	off = NDSLOT(low);
202 	if (low % NDENTRIES) {
203 		mask = ~(~(NDSLOTTYPE)0 >> (NDENTRIES - (low % NDENTRIES)));
204 		if ((mask &= ~map[off]) != 0UL)
205 			return (off * NDENTRIES + ffsl(mask) - 1);
206 		++off;
207 	}
208 	for (maxoff = NDSLOTS(size); off < maxoff; ++off)
209 		if (map[off] != ~0UL)
210 			return (off * NDENTRIES + ffsl(~map[off]) - 1);
211 	return (size);
212 }
213 
214 /*
215  * Find the last used fd.
216  *
217  * Call this variant if fdp can't be modified by anyone else (e.g, during exec).
218  * Otherwise use fdlastfile.
219  */
220 int
221 fdlastfile_single(struct filedesc *fdp)
222 {
223 	NDSLOTTYPE *map = fdp->fd_map;
224 	int off, minoff;
225 
226 	off = NDSLOT(fdp->fd_nfiles - 1);
227 	for (minoff = NDSLOT(0); off >= minoff; --off)
228 		if (map[off] != 0)
229 			return (off * NDENTRIES + flsl(map[off]) - 1);
230 	return (-1);
231 }
232 
233 int
234 fdlastfile(struct filedesc *fdp)
235 {
236 
237 	FILEDESC_LOCK_ASSERT(fdp);
238 	return (fdlastfile_single(fdp));
239 }
240 
241 static int
242 fdisused(struct filedesc *fdp, int fd)
243 {
244 
245 	KASSERT(fd >= 0 && fd < fdp->fd_nfiles,
246 	    ("file descriptor %d out of range (0, %d)", fd, fdp->fd_nfiles));
247 
248 	return ((fdp->fd_map[NDSLOT(fd)] & NDBIT(fd)) != 0);
249 }
250 
251 /*
252  * Mark a file descriptor as used.
253  */
254 static void
255 fdused_init(struct filedesc *fdp, int fd)
256 {
257 
258 	KASSERT(!fdisused(fdp, fd), ("fd=%d is already used", fd));
259 
260 	fdp->fd_map[NDSLOT(fd)] |= NDBIT(fd);
261 }
262 
263 static void
264 fdused(struct filedesc *fdp, int fd)
265 {
266 
267 	FILEDESC_XLOCK_ASSERT(fdp);
268 
269 	fdused_init(fdp, fd);
270 	if (fd == fdp->fd_freefile)
271 		fdp->fd_freefile++;
272 }
273 
274 /*
275  * Mark a file descriptor as unused.
276  */
277 static void
278 fdunused(struct filedesc *fdp, int fd)
279 {
280 
281 	FILEDESC_XLOCK_ASSERT(fdp);
282 
283 	KASSERT(fdisused(fdp, fd), ("fd=%d is already unused", fd));
284 	KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
285 	    ("fd=%d is still in use", fd));
286 
287 	fdp->fd_map[NDSLOT(fd)] &= ~NDBIT(fd);
288 	if (fd < fdp->fd_freefile)
289 		fdp->fd_freefile = fd;
290 }
291 
292 /*
293  * Free a file descriptor.
294  *
295  * Avoid some work if fdp is about to be destroyed.
296  */
297 static inline void
298 fdefree_last(struct filedescent *fde)
299 {
300 
301 	filecaps_free(&fde->fde_caps);
302 }
303 
304 static inline void
305 fdfree(struct filedesc *fdp, int fd)
306 {
307 	struct filedescent *fde;
308 
309 	fde = &fdp->fd_ofiles[fd];
310 #ifdef CAPABILITIES
311 	seqc_write_begin(&fde->fde_seqc);
312 #endif
313 	fde->fde_file = NULL;
314 #ifdef CAPABILITIES
315 	seqc_write_end(&fde->fde_seqc);
316 #endif
317 	fdefree_last(fde);
318 	fdunused(fdp, fd);
319 }
320 
321 /*
322  * System calls on descriptors.
323  */
324 #ifndef _SYS_SYSPROTO_H_
325 struct getdtablesize_args {
326 	int	dummy;
327 };
328 #endif
329 /* ARGSUSED */
330 int
331 sys_getdtablesize(struct thread *td, struct getdtablesize_args *uap)
332 {
333 #ifdef	RACCT
334 	uint64_t lim;
335 #endif
336 
337 	td->td_retval[0] = getmaxfd(td);
338 #ifdef	RACCT
339 	PROC_LOCK(td->td_proc);
340 	lim = racct_get_limit(td->td_proc, RACCT_NOFILE);
341 	PROC_UNLOCK(td->td_proc);
342 	if (lim < td->td_retval[0])
343 		td->td_retval[0] = lim;
344 #endif
345 	return (0);
346 }
347 
348 /*
349  * Duplicate a file descriptor to a particular value.
350  *
351  * Note: keep in mind that a potential race condition exists when closing
352  * descriptors from a shared descriptor table (via rfork).
353  */
354 #ifndef _SYS_SYSPROTO_H_
355 struct dup2_args {
356 	u_int	from;
357 	u_int	to;
358 };
359 #endif
360 /* ARGSUSED */
361 int
362 sys_dup2(struct thread *td, struct dup2_args *uap)
363 {
364 
365 	return (kern_dup(td, FDDUP_FIXED, 0, (int)uap->from, (int)uap->to));
366 }
367 
368 /*
369  * Duplicate a file descriptor.
370  */
371 #ifndef _SYS_SYSPROTO_H_
372 struct dup_args {
373 	u_int	fd;
374 };
375 #endif
376 /* ARGSUSED */
377 int
378 sys_dup(struct thread *td, struct dup_args *uap)
379 {
380 
381 	return (kern_dup(td, FDDUP_NORMAL, 0, (int)uap->fd, 0));
382 }
383 
384 /*
385  * The file control system call.
386  */
387 #ifndef _SYS_SYSPROTO_H_
388 struct fcntl_args {
389 	int	fd;
390 	int	cmd;
391 	long	arg;
392 };
393 #endif
394 /* ARGSUSED */
395 int
396 sys_fcntl(struct thread *td, struct fcntl_args *uap)
397 {
398 
399 	return (kern_fcntl_freebsd(td, uap->fd, uap->cmd, uap->arg));
400 }
401 
402 int
403 kern_fcntl_freebsd(struct thread *td, int fd, int cmd, long arg)
404 {
405 	struct flock fl;
406 	struct __oflock ofl;
407 	intptr_t arg1;
408 	int error, newcmd;
409 
410 	error = 0;
411 	newcmd = cmd;
412 	switch (cmd) {
413 	case F_OGETLK:
414 	case F_OSETLK:
415 	case F_OSETLKW:
416 		/*
417 		 * Convert old flock structure to new.
418 		 */
419 		error = copyin((void *)(intptr_t)arg, &ofl, sizeof(ofl));
420 		fl.l_start = ofl.l_start;
421 		fl.l_len = ofl.l_len;
422 		fl.l_pid = ofl.l_pid;
423 		fl.l_type = ofl.l_type;
424 		fl.l_whence = ofl.l_whence;
425 		fl.l_sysid = 0;
426 
427 		switch (cmd) {
428 		case F_OGETLK:
429 			newcmd = F_GETLK;
430 			break;
431 		case F_OSETLK:
432 			newcmd = F_SETLK;
433 			break;
434 		case F_OSETLKW:
435 			newcmd = F_SETLKW;
436 			break;
437 		}
438 		arg1 = (intptr_t)&fl;
439 		break;
440 	case F_GETLK:
441 	case F_SETLK:
442 	case F_SETLKW:
443 	case F_SETLK_REMOTE:
444 		error = copyin((void *)(intptr_t)arg, &fl, sizeof(fl));
445 		arg1 = (intptr_t)&fl;
446 		break;
447 	default:
448 		arg1 = arg;
449 		break;
450 	}
451 	if (error)
452 		return (error);
453 	error = kern_fcntl(td, fd, newcmd, arg1);
454 	if (error)
455 		return (error);
456 	if (cmd == F_OGETLK) {
457 		ofl.l_start = fl.l_start;
458 		ofl.l_len = fl.l_len;
459 		ofl.l_pid = fl.l_pid;
460 		ofl.l_type = fl.l_type;
461 		ofl.l_whence = fl.l_whence;
462 		error = copyout(&ofl, (void *)(intptr_t)arg, sizeof(ofl));
463 	} else if (cmd == F_GETLK) {
464 		error = copyout(&fl, (void *)(intptr_t)arg, sizeof(fl));
465 	}
466 	return (error);
467 }
468 
469 int
470 kern_fcntl(struct thread *td, int fd, int cmd, intptr_t arg)
471 {
472 	struct filedesc *fdp;
473 	struct flock *flp;
474 	struct file *fp, *fp2;
475 	struct filedescent *fde;
476 	struct proc *p;
477 	struct vnode *vp;
478 	struct mount *mp;
479 	int error, flg, seals, tmp;
480 	uint64_t bsize;
481 	off_t foffset;
482 
483 	error = 0;
484 	flg = F_POSIX;
485 	p = td->td_proc;
486 	fdp = p->p_fd;
487 
488 	AUDIT_ARG_FD(cmd);
489 	AUDIT_ARG_CMD(cmd);
490 	switch (cmd) {
491 	case F_DUPFD:
492 		tmp = arg;
493 		error = kern_dup(td, FDDUP_FCNTL, 0, fd, tmp);
494 		break;
495 
496 	case F_DUPFD_CLOEXEC:
497 		tmp = arg;
498 		error = kern_dup(td, FDDUP_FCNTL, FDDUP_FLAG_CLOEXEC, fd, tmp);
499 		break;
500 
501 	case F_DUP2FD:
502 		tmp = arg;
503 		error = kern_dup(td, FDDUP_FIXED, 0, fd, tmp);
504 		break;
505 
506 	case F_DUP2FD_CLOEXEC:
507 		tmp = arg;
508 		error = kern_dup(td, FDDUP_FIXED, FDDUP_FLAG_CLOEXEC, fd, tmp);
509 		break;
510 
511 	case F_GETFD:
512 		error = EBADF;
513 		FILEDESC_SLOCK(fdp);
514 		fde = fdeget_locked(fdp, fd);
515 		if (fde != NULL) {
516 			td->td_retval[0] =
517 			    (fde->fde_flags & UF_EXCLOSE) ? FD_CLOEXEC : 0;
518 			error = 0;
519 		}
520 		FILEDESC_SUNLOCK(fdp);
521 		break;
522 
523 	case F_SETFD:
524 		error = EBADF;
525 		FILEDESC_XLOCK(fdp);
526 		fde = fdeget_locked(fdp, fd);
527 		if (fde != NULL) {
528 			fde->fde_flags = (fde->fde_flags & ~UF_EXCLOSE) |
529 			    (arg & FD_CLOEXEC ? UF_EXCLOSE : 0);
530 			error = 0;
531 		}
532 		FILEDESC_XUNLOCK(fdp);
533 		break;
534 
535 	case F_GETFL:
536 		error = fget_fcntl(td, fd, &cap_fcntl_rights, F_GETFL, &fp);
537 		if (error != 0)
538 			break;
539 		td->td_retval[0] = OFLAGS(fp->f_flag);
540 		fdrop(fp, td);
541 		break;
542 
543 	case F_SETFL:
544 		error = fget_fcntl(td, fd, &cap_fcntl_rights, F_SETFL, &fp);
545 		if (error != 0)
546 			break;
547 		do {
548 			tmp = flg = fp->f_flag;
549 			tmp &= ~FCNTLFLAGS;
550 			tmp |= FFLAGS(arg & ~O_ACCMODE) & FCNTLFLAGS;
551 		} while(atomic_cmpset_int(&fp->f_flag, flg, tmp) == 0);
552 		tmp = fp->f_flag & FNONBLOCK;
553 		error = fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
554 		if (error != 0) {
555 			fdrop(fp, td);
556 			break;
557 		}
558 		tmp = fp->f_flag & FASYNC;
559 		error = fo_ioctl(fp, FIOASYNC, &tmp, td->td_ucred, td);
560 		if (error == 0) {
561 			fdrop(fp, td);
562 			break;
563 		}
564 		atomic_clear_int(&fp->f_flag, FNONBLOCK);
565 		tmp = 0;
566 		(void)fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
567 		fdrop(fp, td);
568 		break;
569 
570 	case F_GETOWN:
571 		error = fget_fcntl(td, fd, &cap_fcntl_rights, F_GETOWN, &fp);
572 		if (error != 0)
573 			break;
574 		error = fo_ioctl(fp, FIOGETOWN, &tmp, td->td_ucred, td);
575 		if (error == 0)
576 			td->td_retval[0] = tmp;
577 		fdrop(fp, td);
578 		break;
579 
580 	case F_SETOWN:
581 		error = fget_fcntl(td, fd, &cap_fcntl_rights, F_SETOWN, &fp);
582 		if (error != 0)
583 			break;
584 		tmp = arg;
585 		error = fo_ioctl(fp, FIOSETOWN, &tmp, td->td_ucred, td);
586 		fdrop(fp, td);
587 		break;
588 
589 	case F_SETLK_REMOTE:
590 		error = priv_check(td, PRIV_NFS_LOCKD);
591 		if (error != 0)
592 			return (error);
593 		flg = F_REMOTE;
594 		goto do_setlk;
595 
596 	case F_SETLKW:
597 		flg |= F_WAIT;
598 		/* FALLTHROUGH F_SETLK */
599 
600 	case F_SETLK:
601 	do_setlk:
602 		flp = (struct flock *)arg;
603 		if ((flg & F_REMOTE) != 0 && flp->l_sysid == 0) {
604 			error = EINVAL;
605 			break;
606 		}
607 
608 		error = fget_unlocked(fdp, fd, &cap_flock_rights, &fp);
609 		if (error != 0)
610 			break;
611 		if (fp->f_type != DTYPE_VNODE) {
612 			error = EBADF;
613 			fdrop(fp, td);
614 			break;
615 		}
616 
617 		if (flp->l_whence == SEEK_CUR) {
618 			foffset = foffset_get(fp);
619 			if (foffset < 0 ||
620 			    (flp->l_start > 0 &&
621 			     foffset > OFF_MAX - flp->l_start)) {
622 				error = EOVERFLOW;
623 				fdrop(fp, td);
624 				break;
625 			}
626 			flp->l_start += foffset;
627 		}
628 
629 		vp = fp->f_vnode;
630 		switch (flp->l_type) {
631 		case F_RDLCK:
632 			if ((fp->f_flag & FREAD) == 0) {
633 				error = EBADF;
634 				break;
635 			}
636 			if ((p->p_leader->p_flag & P_ADVLOCK) == 0) {
637 				PROC_LOCK(p->p_leader);
638 				p->p_leader->p_flag |= P_ADVLOCK;
639 				PROC_UNLOCK(p->p_leader);
640 			}
641 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
642 			    flp, flg);
643 			break;
644 		case F_WRLCK:
645 			if ((fp->f_flag & FWRITE) == 0) {
646 				error = EBADF;
647 				break;
648 			}
649 			if ((p->p_leader->p_flag & P_ADVLOCK) == 0) {
650 				PROC_LOCK(p->p_leader);
651 				p->p_leader->p_flag |= P_ADVLOCK;
652 				PROC_UNLOCK(p->p_leader);
653 			}
654 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
655 			    flp, flg);
656 			break;
657 		case F_UNLCK:
658 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
659 			    flp, flg);
660 			break;
661 		case F_UNLCKSYS:
662 			if (flg != F_REMOTE) {
663 				error = EINVAL;
664 				break;
665 			}
666 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
667 			    F_UNLCKSYS, flp, flg);
668 			break;
669 		default:
670 			error = EINVAL;
671 			break;
672 		}
673 		if (error != 0 || flp->l_type == F_UNLCK ||
674 		    flp->l_type == F_UNLCKSYS) {
675 			fdrop(fp, td);
676 			break;
677 		}
678 
679 		/*
680 		 * Check for a race with close.
681 		 *
682 		 * The vnode is now advisory locked (or unlocked, but this case
683 		 * is not really important) as the caller requested.
684 		 * We had to drop the filedesc lock, so we need to recheck if
685 		 * the descriptor is still valid, because if it was closed
686 		 * in the meantime we need to remove advisory lock from the
687 		 * vnode - close on any descriptor leading to an advisory
688 		 * locked vnode, removes that lock.
689 		 * We will return 0 on purpose in that case, as the result of
690 		 * successful advisory lock might have been externally visible
691 		 * already. This is fine - effectively we pretend to the caller
692 		 * that the closing thread was a bit slower and that the
693 		 * advisory lock succeeded before the close.
694 		 */
695 		error = fget_unlocked(fdp, fd, &cap_no_rights, &fp2);
696 		if (error != 0) {
697 			fdrop(fp, td);
698 			break;
699 		}
700 		if (fp != fp2) {
701 			flp->l_whence = SEEK_SET;
702 			flp->l_start = 0;
703 			flp->l_len = 0;
704 			flp->l_type = F_UNLCK;
705 			(void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
706 			    F_UNLCK, flp, F_POSIX);
707 		}
708 		fdrop(fp, td);
709 		fdrop(fp2, td);
710 		break;
711 
712 	case F_GETLK:
713 		error = fget_unlocked(fdp, fd, &cap_flock_rights, &fp);
714 		if (error != 0)
715 			break;
716 		if (fp->f_type != DTYPE_VNODE) {
717 			error = EBADF;
718 			fdrop(fp, td);
719 			break;
720 		}
721 		flp = (struct flock *)arg;
722 		if (flp->l_type != F_RDLCK && flp->l_type != F_WRLCK &&
723 		    flp->l_type != F_UNLCK) {
724 			error = EINVAL;
725 			fdrop(fp, td);
726 			break;
727 		}
728 		if (flp->l_whence == SEEK_CUR) {
729 			foffset = foffset_get(fp);
730 			if ((flp->l_start > 0 &&
731 			    foffset > OFF_MAX - flp->l_start) ||
732 			    (flp->l_start < 0 &&
733 			    foffset < OFF_MIN - flp->l_start)) {
734 				error = EOVERFLOW;
735 				fdrop(fp, td);
736 				break;
737 			}
738 			flp->l_start += foffset;
739 		}
740 		vp = fp->f_vnode;
741 		error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, flp,
742 		    F_POSIX);
743 		fdrop(fp, td);
744 		break;
745 
746 	case F_ADD_SEALS:
747 		error = fget_unlocked(fdp, fd, &cap_no_rights, &fp);
748 		if (error != 0)
749 			break;
750 		error = fo_add_seals(fp, arg);
751 		fdrop(fp, td);
752 		break;
753 
754 	case F_GET_SEALS:
755 		error = fget_unlocked(fdp, fd, &cap_no_rights, &fp);
756 		if (error != 0)
757 			break;
758 		if (fo_get_seals(fp, &seals) == 0)
759 			td->td_retval[0] = seals;
760 		else
761 			error = EINVAL;
762 		fdrop(fp, td);
763 		break;
764 
765 	case F_RDAHEAD:
766 		arg = arg ? 128 * 1024: 0;
767 		/* FALLTHROUGH */
768 	case F_READAHEAD:
769 		error = fget_unlocked(fdp, fd, &cap_no_rights, &fp);
770 		if (error != 0)
771 			break;
772 		if (fp->f_type != DTYPE_VNODE) {
773 			fdrop(fp, td);
774 			error = EBADF;
775 			break;
776 		}
777 		vp = fp->f_vnode;
778 		if (vp->v_type != VREG) {
779 			fdrop(fp, td);
780 			error = ENOTTY;
781 			break;
782 		}
783 
784 		/*
785 		 * Exclusive lock synchronizes against f_seqcount reads and
786 		 * writes in sequential_heuristic().
787 		 */
788 		error = vn_lock(vp, LK_EXCLUSIVE);
789 		if (error != 0) {
790 			fdrop(fp, td);
791 			break;
792 		}
793 		if (arg >= 0) {
794 			bsize = fp->f_vnode->v_mount->mnt_stat.f_iosize;
795 			arg = MIN(arg, INT_MAX - bsize + 1);
796 			fp->f_seqcount[UIO_READ] = MIN(IO_SEQMAX,
797 			    (arg + bsize - 1) / bsize);
798 			atomic_set_int(&fp->f_flag, FRDAHEAD);
799 		} else {
800 			atomic_clear_int(&fp->f_flag, FRDAHEAD);
801 		}
802 		VOP_UNLOCK(vp);
803 		fdrop(fp, td);
804 		break;
805 
806 	case F_ISUNIONSTACK:
807 		/*
808 		 * Check if the vnode is part of a union stack (either the
809 		 * "union" flag from mount(2) or unionfs).
810 		 *
811 		 * Prior to introduction of this op libc's readdir would call
812 		 * fstatfs(2), in effect unnecessarily copying kilobytes of
813 		 * data just to check fs name and a mount flag.
814 		 *
815 		 * Fixing the code to handle everything in the kernel instead
816 		 * is a non-trivial endeavor and has low priority, thus this
817 		 * horrible kludge facilitates the current behavior in a much
818 		 * cheaper manner until someone(tm) sorts this out.
819 		 */
820 		error = fget_unlocked(fdp, fd, &cap_no_rights, &fp);
821 		if (error != 0)
822 			break;
823 		if (fp->f_type != DTYPE_VNODE) {
824 			fdrop(fp, td);
825 			error = EBADF;
826 			break;
827 		}
828 		vp = fp->f_vnode;
829 		/*
830 		 * Since we don't prevent dooming the vnode even non-null mp
831 		 * found can become immediately stale. This is tolerable since
832 		 * mount points are type-stable (providing safe memory access)
833 		 * and any vfs op on this vnode going forward will return an
834 		 * error (meaning return value in this case is meaningless).
835 		 */
836 		mp = atomic_load_ptr(&vp->v_mount);
837 		if (__predict_false(mp == NULL)) {
838 			fdrop(fp, td);
839 			error = EBADF;
840 			break;
841 		}
842 		td->td_retval[0] = 0;
843 		if (mp->mnt_kern_flag & MNTK_UNIONFS ||
844 		    mp->mnt_flag & MNT_UNION)
845 			td->td_retval[0] = 1;
846 		fdrop(fp, td);
847 		break;
848 
849 	default:
850 		error = EINVAL;
851 		break;
852 	}
853 	return (error);
854 }
855 
856 static int
857 getmaxfd(struct thread *td)
858 {
859 
860 	return (min((int)lim_cur(td, RLIMIT_NOFILE), maxfilesperproc));
861 }
862 
863 /*
864  * Common code for dup, dup2, fcntl(F_DUPFD) and fcntl(F_DUP2FD).
865  */
866 int
867 kern_dup(struct thread *td, u_int mode, int flags, int old, int new)
868 {
869 	struct filedesc *fdp;
870 	struct filedescent *oldfde, *newfde;
871 	struct proc *p;
872 	struct file *delfp;
873 	u_long *oioctls, *nioctls;
874 	int error, maxfd;
875 
876 	p = td->td_proc;
877 	fdp = p->p_fd;
878 	oioctls = NULL;
879 
880 	MPASS((flags & ~(FDDUP_FLAG_CLOEXEC)) == 0);
881 	MPASS(mode < FDDUP_LASTMODE);
882 
883 	AUDIT_ARG_FD(old);
884 	/* XXXRW: if (flags & FDDUP_FIXED) AUDIT_ARG_FD2(new); */
885 
886 	/*
887 	 * Verify we have a valid descriptor to dup from and possibly to
888 	 * dup to. Unlike dup() and dup2(), fcntl()'s F_DUPFD should
889 	 * return EINVAL when the new descriptor is out of bounds.
890 	 */
891 	if (old < 0)
892 		return (EBADF);
893 	if (new < 0)
894 		return (mode == FDDUP_FCNTL ? EINVAL : EBADF);
895 	maxfd = getmaxfd(td);
896 	if (new >= maxfd)
897 		return (mode == FDDUP_FCNTL ? EINVAL : EBADF);
898 
899 	error = EBADF;
900 	FILEDESC_XLOCK(fdp);
901 	if (fget_locked(fdp, old) == NULL)
902 		goto unlock;
903 	if ((mode == FDDUP_FIXED || mode == FDDUP_MUSTREPLACE) && old == new) {
904 		td->td_retval[0] = new;
905 		if (flags & FDDUP_FLAG_CLOEXEC)
906 			fdp->fd_ofiles[new].fde_flags |= UF_EXCLOSE;
907 		error = 0;
908 		goto unlock;
909 	}
910 
911 	oldfde = &fdp->fd_ofiles[old];
912 	if (!fhold(oldfde->fde_file))
913 		goto unlock;
914 
915 	/*
916 	 * If the caller specified a file descriptor, make sure the file
917 	 * table is large enough to hold it, and grab it.  Otherwise, just
918 	 * allocate a new descriptor the usual way.
919 	 */
920 	switch (mode) {
921 	case FDDUP_NORMAL:
922 	case FDDUP_FCNTL:
923 		if ((error = fdalloc(td, new, &new)) != 0) {
924 			fdrop(oldfde->fde_file, td);
925 			goto unlock;
926 		}
927 		break;
928 	case FDDUP_MUSTREPLACE:
929 		/* Target file descriptor must exist. */
930 		if (fget_locked(fdp, new) == NULL) {
931 			fdrop(oldfde->fde_file, td);
932 			goto unlock;
933 		}
934 		break;
935 	case FDDUP_FIXED:
936 		if (new >= fdp->fd_nfiles) {
937 			/*
938 			 * The resource limits are here instead of e.g.
939 			 * fdalloc(), because the file descriptor table may be
940 			 * shared between processes, so we can't really use
941 			 * racct_add()/racct_sub().  Instead of counting the
942 			 * number of actually allocated descriptors, just put
943 			 * the limit on the size of the file descriptor table.
944 			 */
945 #ifdef RACCT
946 			if (RACCT_ENABLED()) {
947 				error = racct_set_unlocked(p, RACCT_NOFILE, new + 1);
948 				if (error != 0) {
949 					error = EMFILE;
950 					fdrop(oldfde->fde_file, td);
951 					goto unlock;
952 				}
953 			}
954 #endif
955 			fdgrowtable_exp(fdp, new + 1);
956 		}
957 		if (!fdisused(fdp, new))
958 			fdused(fdp, new);
959 		break;
960 	default:
961 		KASSERT(0, ("%s unsupported mode %d", __func__, mode));
962 	}
963 
964 	KASSERT(old != new, ("new fd is same as old"));
965 
966 	newfde = &fdp->fd_ofiles[new];
967 	delfp = newfde->fde_file;
968 
969 	nioctls = filecaps_copy_prep(&oldfde->fde_caps);
970 
971 	/*
972 	 * Duplicate the source descriptor.
973 	 */
974 #ifdef CAPABILITIES
975 	seqc_write_begin(&newfde->fde_seqc);
976 #endif
977 	oioctls = filecaps_free_prep(&newfde->fde_caps);
978 	memcpy(newfde, oldfde, fde_change_size);
979 	filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps,
980 	    nioctls);
981 	if ((flags & FDDUP_FLAG_CLOEXEC) != 0)
982 		newfde->fde_flags = oldfde->fde_flags | UF_EXCLOSE;
983 	else
984 		newfde->fde_flags = oldfde->fde_flags & ~UF_EXCLOSE;
985 #ifdef CAPABILITIES
986 	seqc_write_end(&newfde->fde_seqc);
987 #endif
988 	td->td_retval[0] = new;
989 
990 	error = 0;
991 
992 	if (delfp != NULL) {
993 		(void) closefp(fdp, new, delfp, td, 1);
994 		FILEDESC_UNLOCK_ASSERT(fdp);
995 	} else {
996 unlock:
997 		FILEDESC_XUNLOCK(fdp);
998 	}
999 
1000 	filecaps_free_finish(oioctls);
1001 	return (error);
1002 }
1003 
1004 /*
1005  * If sigio is on the list associated with a process or process group,
1006  * disable signalling from the device, remove sigio from the list and
1007  * free sigio.
1008  */
1009 void
1010 funsetown(struct sigio **sigiop)
1011 {
1012 	struct sigio *sigio;
1013 
1014 	if (*sigiop == NULL)
1015 		return;
1016 	SIGIO_LOCK();
1017 	sigio = *sigiop;
1018 	if (sigio == NULL) {
1019 		SIGIO_UNLOCK();
1020 		return;
1021 	}
1022 	*(sigio->sio_myref) = NULL;
1023 	if ((sigio)->sio_pgid < 0) {
1024 		struct pgrp *pg = (sigio)->sio_pgrp;
1025 		PGRP_LOCK(pg);
1026 		SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio,
1027 			    sigio, sio_pgsigio);
1028 		PGRP_UNLOCK(pg);
1029 	} else {
1030 		struct proc *p = (sigio)->sio_proc;
1031 		PROC_LOCK(p);
1032 		SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio,
1033 			    sigio, sio_pgsigio);
1034 		PROC_UNLOCK(p);
1035 	}
1036 	SIGIO_UNLOCK();
1037 	crfree(sigio->sio_ucred);
1038 	free(sigio, M_SIGIO);
1039 }
1040 
1041 /*
1042  * Free a list of sigio structures.
1043  * We only need to lock the SIGIO_LOCK because we have made ourselves
1044  * inaccessible to callers of fsetown and therefore do not need to lock
1045  * the proc or pgrp struct for the list manipulation.
1046  */
1047 void
1048 funsetownlst(struct sigiolst *sigiolst)
1049 {
1050 	struct proc *p;
1051 	struct pgrp *pg;
1052 	struct sigio *sigio;
1053 
1054 	sigio = SLIST_FIRST(sigiolst);
1055 	if (sigio == NULL)
1056 		return;
1057 	p = NULL;
1058 	pg = NULL;
1059 
1060 	/*
1061 	 * Every entry of the list should belong
1062 	 * to a single proc or pgrp.
1063 	 */
1064 	if (sigio->sio_pgid < 0) {
1065 		pg = sigio->sio_pgrp;
1066 		PGRP_LOCK_ASSERT(pg, MA_NOTOWNED);
1067 	} else /* if (sigio->sio_pgid > 0) */ {
1068 		p = sigio->sio_proc;
1069 		PROC_LOCK_ASSERT(p, MA_NOTOWNED);
1070 	}
1071 
1072 	SIGIO_LOCK();
1073 	while ((sigio = SLIST_FIRST(sigiolst)) != NULL) {
1074 		*(sigio->sio_myref) = NULL;
1075 		if (pg != NULL) {
1076 			KASSERT(sigio->sio_pgid < 0,
1077 			    ("Proc sigio in pgrp sigio list"));
1078 			KASSERT(sigio->sio_pgrp == pg,
1079 			    ("Bogus pgrp in sigio list"));
1080 			PGRP_LOCK(pg);
1081 			SLIST_REMOVE(&pg->pg_sigiolst, sigio, sigio,
1082 			    sio_pgsigio);
1083 			PGRP_UNLOCK(pg);
1084 		} else /* if (p != NULL) */ {
1085 			KASSERT(sigio->sio_pgid > 0,
1086 			    ("Pgrp sigio in proc sigio list"));
1087 			KASSERT(sigio->sio_proc == p,
1088 			    ("Bogus proc in sigio list"));
1089 			PROC_LOCK(p);
1090 			SLIST_REMOVE(&p->p_sigiolst, sigio, sigio,
1091 			    sio_pgsigio);
1092 			PROC_UNLOCK(p);
1093 		}
1094 		SIGIO_UNLOCK();
1095 		crfree(sigio->sio_ucred);
1096 		free(sigio, M_SIGIO);
1097 		SIGIO_LOCK();
1098 	}
1099 	SIGIO_UNLOCK();
1100 }
1101 
1102 /*
1103  * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg).
1104  *
1105  * After permission checking, add a sigio structure to the sigio list for
1106  * the process or process group.
1107  */
1108 int
1109 fsetown(pid_t pgid, struct sigio **sigiop)
1110 {
1111 	struct proc *proc;
1112 	struct pgrp *pgrp;
1113 	struct sigio *sigio;
1114 	int ret;
1115 
1116 	if (pgid == 0) {
1117 		funsetown(sigiop);
1118 		return (0);
1119 	}
1120 
1121 	ret = 0;
1122 
1123 	/* Allocate and fill in the new sigio out of locks. */
1124 	sigio = malloc(sizeof(struct sigio), M_SIGIO, M_WAITOK);
1125 	sigio->sio_pgid = pgid;
1126 	sigio->sio_ucred = crhold(curthread->td_ucred);
1127 	sigio->sio_myref = sigiop;
1128 
1129 	sx_slock(&proctree_lock);
1130 	if (pgid > 0) {
1131 		proc = pfind(pgid);
1132 		if (proc == NULL) {
1133 			ret = ESRCH;
1134 			goto fail;
1135 		}
1136 
1137 		/*
1138 		 * Policy - Don't allow a process to FSETOWN a process
1139 		 * in another session.
1140 		 *
1141 		 * Remove this test to allow maximum flexibility or
1142 		 * restrict FSETOWN to the current process or process
1143 		 * group for maximum safety.
1144 		 */
1145 		PROC_UNLOCK(proc);
1146 		if (proc->p_session != curthread->td_proc->p_session) {
1147 			ret = EPERM;
1148 			goto fail;
1149 		}
1150 
1151 		pgrp = NULL;
1152 	} else /* if (pgid < 0) */ {
1153 		pgrp = pgfind(-pgid);
1154 		if (pgrp == NULL) {
1155 			ret = ESRCH;
1156 			goto fail;
1157 		}
1158 		PGRP_UNLOCK(pgrp);
1159 
1160 		/*
1161 		 * Policy - Don't allow a process to FSETOWN a process
1162 		 * in another session.
1163 		 *
1164 		 * Remove this test to allow maximum flexibility or
1165 		 * restrict FSETOWN to the current process or process
1166 		 * group for maximum safety.
1167 		 */
1168 		if (pgrp->pg_session != curthread->td_proc->p_session) {
1169 			ret = EPERM;
1170 			goto fail;
1171 		}
1172 
1173 		proc = NULL;
1174 	}
1175 	funsetown(sigiop);
1176 	if (pgid > 0) {
1177 		PROC_LOCK(proc);
1178 		/*
1179 		 * Since funsetownlst() is called without the proctree
1180 		 * locked, we need to check for P_WEXIT.
1181 		 * XXX: is ESRCH correct?
1182 		 */
1183 		if ((proc->p_flag & P_WEXIT) != 0) {
1184 			PROC_UNLOCK(proc);
1185 			ret = ESRCH;
1186 			goto fail;
1187 		}
1188 		SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio);
1189 		sigio->sio_proc = proc;
1190 		PROC_UNLOCK(proc);
1191 	} else {
1192 		PGRP_LOCK(pgrp);
1193 		SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio);
1194 		sigio->sio_pgrp = pgrp;
1195 		PGRP_UNLOCK(pgrp);
1196 	}
1197 	sx_sunlock(&proctree_lock);
1198 	SIGIO_LOCK();
1199 	*sigiop = sigio;
1200 	SIGIO_UNLOCK();
1201 	return (0);
1202 
1203 fail:
1204 	sx_sunlock(&proctree_lock);
1205 	crfree(sigio->sio_ucred);
1206 	free(sigio, M_SIGIO);
1207 	return (ret);
1208 }
1209 
1210 /*
1211  * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg).
1212  */
1213 pid_t
1214 fgetown(struct sigio **sigiop)
1215 {
1216 	pid_t pgid;
1217 
1218 	SIGIO_LOCK();
1219 	pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0;
1220 	SIGIO_UNLOCK();
1221 	return (pgid);
1222 }
1223 
1224 /*
1225  * Function drops the filedesc lock on return.
1226  */
1227 static int
1228 closefp(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1229     int holdleaders)
1230 {
1231 	int error;
1232 
1233 	FILEDESC_XLOCK_ASSERT(fdp);
1234 
1235 	if (holdleaders) {
1236 		if (td->td_proc->p_fdtol != NULL) {
1237 			/*
1238 			 * Ask fdfree() to sleep to ensure that all relevant
1239 			 * process leaders can be traversed in closef().
1240 			 */
1241 			fdp->fd_holdleaderscount++;
1242 		} else {
1243 			holdleaders = 0;
1244 		}
1245 	}
1246 
1247 	/*
1248 	 * We now hold the fp reference that used to be owned by the
1249 	 * descriptor array.  We have to unlock the FILEDESC *AFTER*
1250 	 * knote_fdclose to prevent a race of the fd getting opened, a knote
1251 	 * added, and deleteing a knote for the new fd.
1252 	 */
1253 	if (__predict_false(!TAILQ_EMPTY(&fdp->fd_kqlist)))
1254 		knote_fdclose(td, fd);
1255 
1256 	/*
1257 	 * We need to notify mqueue if the object is of type mqueue.
1258 	 */
1259 	if (__predict_false(fp->f_type == DTYPE_MQUEUE))
1260 		mq_fdclose(td, fd, fp);
1261 	FILEDESC_XUNLOCK(fdp);
1262 
1263 	error = closef(fp, td);
1264 	if (holdleaders) {
1265 		FILEDESC_XLOCK(fdp);
1266 		fdp->fd_holdleaderscount--;
1267 		if (fdp->fd_holdleaderscount == 0 &&
1268 		    fdp->fd_holdleaderswakeup != 0) {
1269 			fdp->fd_holdleaderswakeup = 0;
1270 			wakeup(&fdp->fd_holdleaderscount);
1271 		}
1272 		FILEDESC_XUNLOCK(fdp);
1273 	}
1274 	return (error);
1275 }
1276 
1277 /*
1278  * Close a file descriptor.
1279  */
1280 #ifndef _SYS_SYSPROTO_H_
1281 struct close_args {
1282 	int     fd;
1283 };
1284 #endif
1285 /* ARGSUSED */
1286 int
1287 sys_close(struct thread *td, struct close_args *uap)
1288 {
1289 
1290 	return (kern_close(td, uap->fd));
1291 }
1292 
1293 int
1294 kern_close(struct thread *td, int fd)
1295 {
1296 	struct filedesc *fdp;
1297 	struct file *fp;
1298 
1299 	fdp = td->td_proc->p_fd;
1300 
1301 	AUDIT_SYSCLOSE(td, fd);
1302 
1303 	FILEDESC_XLOCK(fdp);
1304 	if ((fp = fget_locked(fdp, fd)) == NULL) {
1305 		FILEDESC_XUNLOCK(fdp);
1306 		return (EBADF);
1307 	}
1308 	fdfree(fdp, fd);
1309 
1310 	/* closefp() drops the FILEDESC lock for us. */
1311 	return (closefp(fdp, fd, fp, td, 1));
1312 }
1313 
1314 int
1315 kern_close_range(struct thread *td, u_int lowfd, u_int highfd)
1316 {
1317 	struct filedesc *fdp;
1318 	int fd, ret, lastfile;
1319 
1320 	ret = 0;
1321 	fdp = td->td_proc->p_fd;
1322 	FILEDESC_SLOCK(fdp);
1323 
1324 	/*
1325 	 * Check this prior to clamping; closefrom(3) with only fd 0, 1, and 2
1326 	 * open should not be a usage error.  From a close_range() perspective,
1327 	 * close_range(3, ~0U, 0) in the same scenario should also likely not
1328 	 * be a usage error as all fd above 3 are in-fact already closed.
1329 	 */
1330 	if (highfd < lowfd) {
1331 		ret = EINVAL;
1332 		goto out;
1333 	}
1334 
1335 	/*
1336 	 * If lastfile == -1, we're dealing with either a fresh file
1337 	 * table or one in which every fd has been closed.  Just return
1338 	 * successful; there's nothing left to do.
1339 	 */
1340 	lastfile = fdlastfile(fdp);
1341 	if (lastfile == -1)
1342 		goto out;
1343 	/* Clamped to [lowfd, lastfile] */
1344 	highfd = MIN(highfd, lastfile);
1345 	for (fd = lowfd; fd <= highfd; fd++) {
1346 		if (fdp->fd_ofiles[fd].fde_file != NULL) {
1347 			FILEDESC_SUNLOCK(fdp);
1348 			(void)kern_close(td, fd);
1349 			FILEDESC_SLOCK(fdp);
1350 		}
1351 	}
1352 out:
1353 	FILEDESC_SUNLOCK(fdp);
1354 	return (ret);
1355 }
1356 
1357 #ifndef _SYS_SYSPROTO_H_
1358 struct close_range_args {
1359 	u_int	lowfd;
1360 	u_int	highfd;
1361 	int	flags;
1362 };
1363 #endif
1364 int
1365 sys_close_range(struct thread *td, struct close_range_args *uap)
1366 {
1367 
1368 	/* No flags currently defined */
1369 	if (uap->flags != 0)
1370 		return (EINVAL);
1371 	return (kern_close_range(td, uap->lowfd, uap->highfd));
1372 }
1373 
1374 #ifdef COMPAT_FREEBSD12
1375 /*
1376  * Close open file descriptors.
1377  */
1378 #ifndef _SYS_SYSPROTO_H_
1379 struct freebsd12_closefrom_args {
1380 	int	lowfd;
1381 };
1382 #endif
1383 /* ARGSUSED */
1384 int
1385 freebsd12_closefrom(struct thread *td, struct freebsd12_closefrom_args *uap)
1386 {
1387 	u_int lowfd;
1388 
1389 	AUDIT_ARG_FD(uap->lowfd);
1390 
1391 	/*
1392 	 * Treat negative starting file descriptor values identical to
1393 	 * closefrom(0) which closes all files.
1394 	 */
1395 	lowfd = MAX(0, uap->lowfd);
1396 	return (kern_close_range(td, lowfd, ~0U));
1397 }
1398 #endif	/* COMPAT_FREEBSD12 */
1399 
1400 #if defined(COMPAT_43)
1401 /*
1402  * Return status information about a file descriptor.
1403  */
1404 #ifndef _SYS_SYSPROTO_H_
1405 struct ofstat_args {
1406 	int	fd;
1407 	struct	ostat *sb;
1408 };
1409 #endif
1410 /* ARGSUSED */
1411 int
1412 ofstat(struct thread *td, struct ofstat_args *uap)
1413 {
1414 	struct ostat oub;
1415 	struct stat ub;
1416 	int error;
1417 
1418 	error = kern_fstat(td, uap->fd, &ub);
1419 	if (error == 0) {
1420 		cvtstat(&ub, &oub);
1421 		error = copyout(&oub, uap->sb, sizeof(oub));
1422 	}
1423 	return (error);
1424 }
1425 #endif /* COMPAT_43 */
1426 
1427 #if defined(COMPAT_FREEBSD11)
1428 int
1429 freebsd11_fstat(struct thread *td, struct freebsd11_fstat_args *uap)
1430 {
1431 	struct stat sb;
1432 	struct freebsd11_stat osb;
1433 	int error;
1434 
1435 	error = kern_fstat(td, uap->fd, &sb);
1436 	if (error != 0)
1437 		return (error);
1438 	error = freebsd11_cvtstat(&sb, &osb);
1439 	if (error == 0)
1440 		error = copyout(&osb, uap->sb, sizeof(osb));
1441 	return (error);
1442 }
1443 #endif	/* COMPAT_FREEBSD11 */
1444 
1445 /*
1446  * Return status information about a file descriptor.
1447  */
1448 #ifndef _SYS_SYSPROTO_H_
1449 struct fstat_args {
1450 	int	fd;
1451 	struct	stat *sb;
1452 };
1453 #endif
1454 /* ARGSUSED */
1455 int
1456 sys_fstat(struct thread *td, struct fstat_args *uap)
1457 {
1458 	struct stat ub;
1459 	int error;
1460 
1461 	error = kern_fstat(td, uap->fd, &ub);
1462 	if (error == 0)
1463 		error = copyout(&ub, uap->sb, sizeof(ub));
1464 	return (error);
1465 }
1466 
1467 int
1468 kern_fstat(struct thread *td, int fd, struct stat *sbp)
1469 {
1470 	struct file *fp;
1471 	int error;
1472 
1473 	AUDIT_ARG_FD(fd);
1474 
1475 	error = fget(td, fd, &cap_fstat_rights, &fp);
1476 	if (__predict_false(error != 0))
1477 		return (error);
1478 
1479 	AUDIT_ARG_FILE(td->td_proc, fp);
1480 
1481 	error = fo_stat(fp, sbp, td->td_ucred, td);
1482 	fdrop(fp, td);
1483 #ifdef __STAT_TIME_T_EXT
1484 	sbp->st_atim_ext = 0;
1485 	sbp->st_mtim_ext = 0;
1486 	sbp->st_ctim_ext = 0;
1487 	sbp->st_btim_ext = 0;
1488 #endif
1489 #ifdef KTRACE
1490 	if (KTRPOINT(td, KTR_STRUCT))
1491 		ktrstat_error(sbp, error);
1492 #endif
1493 	return (error);
1494 }
1495 
1496 #if defined(COMPAT_FREEBSD11)
1497 /*
1498  * Return status information about a file descriptor.
1499  */
1500 #ifndef _SYS_SYSPROTO_H_
1501 struct freebsd11_nfstat_args {
1502 	int	fd;
1503 	struct	nstat *sb;
1504 };
1505 #endif
1506 /* ARGSUSED */
1507 int
1508 freebsd11_nfstat(struct thread *td, struct freebsd11_nfstat_args *uap)
1509 {
1510 	struct nstat nub;
1511 	struct stat ub;
1512 	int error;
1513 
1514 	error = kern_fstat(td, uap->fd, &ub);
1515 	if (error == 0) {
1516 		freebsd11_cvtnstat(&ub, &nub);
1517 		error = copyout(&nub, uap->sb, sizeof(nub));
1518 	}
1519 	return (error);
1520 }
1521 #endif /* COMPAT_FREEBSD11 */
1522 
1523 /*
1524  * Return pathconf information about a file descriptor.
1525  */
1526 #ifndef _SYS_SYSPROTO_H_
1527 struct fpathconf_args {
1528 	int	fd;
1529 	int	name;
1530 };
1531 #endif
1532 /* ARGSUSED */
1533 int
1534 sys_fpathconf(struct thread *td, struct fpathconf_args *uap)
1535 {
1536 	long value;
1537 	int error;
1538 
1539 	error = kern_fpathconf(td, uap->fd, uap->name, &value);
1540 	if (error == 0)
1541 		td->td_retval[0] = value;
1542 	return (error);
1543 }
1544 
1545 int
1546 kern_fpathconf(struct thread *td, int fd, int name, long *valuep)
1547 {
1548 	struct file *fp;
1549 	struct vnode *vp;
1550 	int error;
1551 
1552 	error = fget(td, fd, &cap_fpathconf_rights, &fp);
1553 	if (error != 0)
1554 		return (error);
1555 
1556 	if (name == _PC_ASYNC_IO) {
1557 		*valuep = _POSIX_ASYNCHRONOUS_IO;
1558 		goto out;
1559 	}
1560 	vp = fp->f_vnode;
1561 	if (vp != NULL) {
1562 		vn_lock(vp, LK_SHARED | LK_RETRY);
1563 		error = VOP_PATHCONF(vp, name, valuep);
1564 		VOP_UNLOCK(vp);
1565 	} else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) {
1566 		if (name != _PC_PIPE_BUF) {
1567 			error = EINVAL;
1568 		} else {
1569 			*valuep = PIPE_BUF;
1570 			error = 0;
1571 		}
1572 	} else {
1573 		error = EOPNOTSUPP;
1574 	}
1575 out:
1576 	fdrop(fp, td);
1577 	return (error);
1578 }
1579 
1580 /*
1581  * Copy filecaps structure allocating memory for ioctls array if needed.
1582  *
1583  * The last parameter indicates whether the fdtable is locked. If it is not and
1584  * ioctls are encountered, copying fails and the caller must lock the table.
1585  *
1586  * Note that if the table was not locked, the caller has to check the relevant
1587  * sequence counter to determine whether the operation was successful.
1588  */
1589 bool
1590 filecaps_copy(const struct filecaps *src, struct filecaps *dst, bool locked)
1591 {
1592 	size_t size;
1593 
1594 	if (src->fc_ioctls != NULL && !locked)
1595 		return (false);
1596 	memcpy(dst, src, sizeof(*src));
1597 	if (src->fc_ioctls == NULL)
1598 		return (true);
1599 
1600 	KASSERT(src->fc_nioctls > 0,
1601 	    ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1602 
1603 	size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1604 	dst->fc_ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1605 	memcpy(dst->fc_ioctls, src->fc_ioctls, size);
1606 	return (true);
1607 }
1608 
1609 static u_long *
1610 filecaps_copy_prep(const struct filecaps *src)
1611 {
1612 	u_long *ioctls;
1613 	size_t size;
1614 
1615 	if (__predict_true(src->fc_ioctls == NULL))
1616 		return (NULL);
1617 
1618 	KASSERT(src->fc_nioctls > 0,
1619 	    ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1620 
1621 	size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1622 	ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1623 	return (ioctls);
1624 }
1625 
1626 static void
1627 filecaps_copy_finish(const struct filecaps *src, struct filecaps *dst,
1628     u_long *ioctls)
1629 {
1630 	size_t size;
1631 
1632 	*dst = *src;
1633 	if (__predict_true(src->fc_ioctls == NULL)) {
1634 		MPASS(ioctls == NULL);
1635 		return;
1636 	}
1637 
1638 	size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1639 	dst->fc_ioctls = ioctls;
1640 	bcopy(src->fc_ioctls, dst->fc_ioctls, size);
1641 }
1642 
1643 /*
1644  * Move filecaps structure to the new place and clear the old place.
1645  */
1646 void
1647 filecaps_move(struct filecaps *src, struct filecaps *dst)
1648 {
1649 
1650 	*dst = *src;
1651 	bzero(src, sizeof(*src));
1652 }
1653 
1654 /*
1655  * Fill the given filecaps structure with full rights.
1656  */
1657 static void
1658 filecaps_fill(struct filecaps *fcaps)
1659 {
1660 
1661 	CAP_ALL(&fcaps->fc_rights);
1662 	fcaps->fc_ioctls = NULL;
1663 	fcaps->fc_nioctls = -1;
1664 	fcaps->fc_fcntls = CAP_FCNTL_ALL;
1665 }
1666 
1667 /*
1668  * Free memory allocated within filecaps structure.
1669  */
1670 void
1671 filecaps_free(struct filecaps *fcaps)
1672 {
1673 
1674 	free(fcaps->fc_ioctls, M_FILECAPS);
1675 	bzero(fcaps, sizeof(*fcaps));
1676 }
1677 
1678 static u_long *
1679 filecaps_free_prep(struct filecaps *fcaps)
1680 {
1681 	u_long *ioctls;
1682 
1683 	ioctls = fcaps->fc_ioctls;
1684 	bzero(fcaps, sizeof(*fcaps));
1685 	return (ioctls);
1686 }
1687 
1688 static void
1689 filecaps_free_finish(u_long *ioctls)
1690 {
1691 
1692 	free(ioctls, M_FILECAPS);
1693 }
1694 
1695 /*
1696  * Validate the given filecaps structure.
1697  */
1698 static void
1699 filecaps_validate(const struct filecaps *fcaps, const char *func)
1700 {
1701 
1702 	KASSERT(cap_rights_is_valid(&fcaps->fc_rights),
1703 	    ("%s: invalid rights", func));
1704 	KASSERT((fcaps->fc_fcntls & ~CAP_FCNTL_ALL) == 0,
1705 	    ("%s: invalid fcntls", func));
1706 	KASSERT(fcaps->fc_fcntls == 0 ||
1707 	    cap_rights_is_set(&fcaps->fc_rights, CAP_FCNTL),
1708 	    ("%s: fcntls without CAP_FCNTL", func));
1709 	KASSERT(fcaps->fc_ioctls != NULL ? fcaps->fc_nioctls > 0 :
1710 	    (fcaps->fc_nioctls == -1 || fcaps->fc_nioctls == 0),
1711 	    ("%s: invalid ioctls", func));
1712 	KASSERT(fcaps->fc_nioctls == 0 ||
1713 	    cap_rights_is_set(&fcaps->fc_rights, CAP_IOCTL),
1714 	    ("%s: ioctls without CAP_IOCTL", func));
1715 }
1716 
1717 static void
1718 fdgrowtable_exp(struct filedesc *fdp, int nfd)
1719 {
1720 	int nfd1;
1721 
1722 	FILEDESC_XLOCK_ASSERT(fdp);
1723 
1724 	nfd1 = fdp->fd_nfiles * 2;
1725 	if (nfd1 < nfd)
1726 		nfd1 = nfd;
1727 	fdgrowtable(fdp, nfd1);
1728 }
1729 
1730 /*
1731  * Grow the file table to accommodate (at least) nfd descriptors.
1732  */
1733 static void
1734 fdgrowtable(struct filedesc *fdp, int nfd)
1735 {
1736 	struct filedesc0 *fdp0;
1737 	struct freetable *ft;
1738 	struct fdescenttbl *ntable;
1739 	struct fdescenttbl *otable;
1740 	int nnfiles, onfiles;
1741 	NDSLOTTYPE *nmap, *omap;
1742 
1743 	KASSERT(fdp->fd_nfiles > 0, ("zero-length file table"));
1744 
1745 	/* save old values */
1746 	onfiles = fdp->fd_nfiles;
1747 	otable = fdp->fd_files;
1748 	omap = fdp->fd_map;
1749 
1750 	/* compute the size of the new table */
1751 	nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */
1752 	if (nnfiles <= onfiles)
1753 		/* the table is already large enough */
1754 		return;
1755 
1756 	/*
1757 	 * Allocate a new table.  We need enough space for the number of
1758 	 * entries, file entries themselves and the struct freetable we will use
1759 	 * when we decommission the table and place it on the freelist.
1760 	 * We place the struct freetable in the middle so we don't have
1761 	 * to worry about padding.
1762 	 */
1763 	ntable = malloc(offsetof(struct fdescenttbl, fdt_ofiles) +
1764 	    nnfiles * sizeof(ntable->fdt_ofiles[0]) +
1765 	    sizeof(struct freetable),
1766 	    M_FILEDESC, M_ZERO | M_WAITOK);
1767 	/* copy the old data */
1768 	ntable->fdt_nfiles = nnfiles;
1769 	memcpy(ntable->fdt_ofiles, otable->fdt_ofiles,
1770 	    onfiles * sizeof(ntable->fdt_ofiles[0]));
1771 
1772 	/*
1773 	 * Allocate a new map only if the old is not large enough.  It will
1774 	 * grow at a slower rate than the table as it can map more
1775 	 * entries than the table can hold.
1776 	 */
1777 	if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) {
1778 		nmap = malloc(NDSLOTS(nnfiles) * NDSLOTSIZE, M_FILEDESC,
1779 		    M_ZERO | M_WAITOK);
1780 		/* copy over the old data and update the pointer */
1781 		memcpy(nmap, omap, NDSLOTS(onfiles) * sizeof(*omap));
1782 		fdp->fd_map = nmap;
1783 	}
1784 
1785 	/*
1786 	 * Make sure that ntable is correctly initialized before we replace
1787 	 * fd_files poiner. Otherwise fget_unlocked() may see inconsistent
1788 	 * data.
1789 	 */
1790 	atomic_store_rel_ptr((volatile void *)&fdp->fd_files, (uintptr_t)ntable);
1791 
1792 	/*
1793 	 * Do not free the old file table, as some threads may still
1794 	 * reference entries within it.  Instead, place it on a freelist
1795 	 * which will be processed when the struct filedesc is released.
1796 	 *
1797 	 * Note that if onfiles == NDFILE, we're dealing with the original
1798 	 * static allocation contained within (struct filedesc0 *)fdp,
1799 	 * which must not be freed.
1800 	 */
1801 	if (onfiles > NDFILE) {
1802 		ft = (struct freetable *)&otable->fdt_ofiles[onfiles];
1803 		fdp0 = (struct filedesc0 *)fdp;
1804 		ft->ft_table = otable;
1805 		SLIST_INSERT_HEAD(&fdp0->fd_free, ft, ft_next);
1806 	}
1807 	/*
1808 	 * The map does not have the same possibility of threads still
1809 	 * holding references to it.  So always free it as long as it
1810 	 * does not reference the original static allocation.
1811 	 */
1812 	if (NDSLOTS(onfiles) > NDSLOTS(NDFILE))
1813 		free(omap, M_FILEDESC);
1814 }
1815 
1816 /*
1817  * Allocate a file descriptor for the process.
1818  */
1819 int
1820 fdalloc(struct thread *td, int minfd, int *result)
1821 {
1822 	struct proc *p = td->td_proc;
1823 	struct filedesc *fdp = p->p_fd;
1824 	int fd, maxfd, allocfd;
1825 #ifdef RACCT
1826 	int error;
1827 #endif
1828 
1829 	FILEDESC_XLOCK_ASSERT(fdp);
1830 
1831 	if (fdp->fd_freefile > minfd)
1832 		minfd = fdp->fd_freefile;
1833 
1834 	maxfd = getmaxfd(td);
1835 
1836 	/*
1837 	 * Search the bitmap for a free descriptor starting at minfd.
1838 	 * If none is found, grow the file table.
1839 	 */
1840 	fd = fd_first_free(fdp, minfd, fdp->fd_nfiles);
1841 	if (__predict_false(fd >= maxfd))
1842 		return (EMFILE);
1843 	if (__predict_false(fd >= fdp->fd_nfiles)) {
1844 		allocfd = min(fd * 2, maxfd);
1845 #ifdef RACCT
1846 		if (RACCT_ENABLED()) {
1847 			error = racct_set_unlocked(p, RACCT_NOFILE, allocfd);
1848 			if (error != 0)
1849 				return (EMFILE);
1850 		}
1851 #endif
1852 		/*
1853 		 * fd is already equal to first free descriptor >= minfd, so
1854 		 * we only need to grow the table and we are done.
1855 		 */
1856 		fdgrowtable_exp(fdp, allocfd);
1857 	}
1858 
1859 	/*
1860 	 * Perform some sanity checks, then mark the file descriptor as
1861 	 * used and return it to the caller.
1862 	 */
1863 	KASSERT(fd >= 0 && fd < min(maxfd, fdp->fd_nfiles),
1864 	    ("invalid descriptor %d", fd));
1865 	KASSERT(!fdisused(fdp, fd),
1866 	    ("fd_first_free() returned non-free descriptor"));
1867 	KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
1868 	    ("file descriptor isn't free"));
1869 	fdused(fdp, fd);
1870 	*result = fd;
1871 	return (0);
1872 }
1873 
1874 /*
1875  * Allocate n file descriptors for the process.
1876  */
1877 int
1878 fdallocn(struct thread *td, int minfd, int *fds, int n)
1879 {
1880 	struct proc *p = td->td_proc;
1881 	struct filedesc *fdp = p->p_fd;
1882 	int i;
1883 
1884 	FILEDESC_XLOCK_ASSERT(fdp);
1885 
1886 	for (i = 0; i < n; i++)
1887 		if (fdalloc(td, 0, &fds[i]) != 0)
1888 			break;
1889 
1890 	if (i < n) {
1891 		for (i--; i >= 0; i--)
1892 			fdunused(fdp, fds[i]);
1893 		return (EMFILE);
1894 	}
1895 
1896 	return (0);
1897 }
1898 
1899 /*
1900  * Create a new open file structure and allocate a file descriptor for the
1901  * process that refers to it.  We add one reference to the file for the
1902  * descriptor table and one reference for resultfp. This is to prevent us
1903  * being preempted and the entry in the descriptor table closed after we
1904  * release the FILEDESC lock.
1905  */
1906 int
1907 falloc_caps(struct thread *td, struct file **resultfp, int *resultfd, int flags,
1908     struct filecaps *fcaps)
1909 {
1910 	struct file *fp;
1911 	int error, fd;
1912 
1913 	error = falloc_noinstall(td, &fp);
1914 	if (error)
1915 		return (error);		/* no reference held on error */
1916 
1917 	error = finstall(td, fp, &fd, flags, fcaps);
1918 	if (error) {
1919 		fdrop(fp, td);		/* one reference (fp only) */
1920 		return (error);
1921 	}
1922 
1923 	if (resultfp != NULL)
1924 		*resultfp = fp;		/* copy out result */
1925 	else
1926 		fdrop(fp, td);		/* release local reference */
1927 
1928 	if (resultfd != NULL)
1929 		*resultfd = fd;
1930 
1931 	return (0);
1932 }
1933 
1934 /*
1935  * Create a new open file structure without allocating a file descriptor.
1936  */
1937 int
1938 falloc_noinstall(struct thread *td, struct file **resultfp)
1939 {
1940 	struct file *fp;
1941 	int maxuserfiles = maxfiles - (maxfiles / 20);
1942 	int openfiles_new;
1943 	static struct timeval lastfail;
1944 	static int curfail;
1945 
1946 	KASSERT(resultfp != NULL, ("%s: resultfp == NULL", __func__));
1947 
1948 	openfiles_new = atomic_fetchadd_int(&openfiles, 1) + 1;
1949 	if ((openfiles_new >= maxuserfiles &&
1950 	    priv_check(td, PRIV_MAXFILES) != 0) ||
1951 	    openfiles_new >= maxfiles) {
1952 		atomic_subtract_int(&openfiles, 1);
1953 		if (ppsratecheck(&lastfail, &curfail, 1)) {
1954 			printf("kern.maxfiles limit exceeded by uid %i, (%s) "
1955 			    "please see tuning(7).\n", td->td_ucred->cr_ruid, td->td_proc->p_comm);
1956 		}
1957 		return (ENFILE);
1958 	}
1959 	fp = uma_zalloc(file_zone, M_WAITOK);
1960 	bzero(fp, sizeof(*fp));
1961 	refcount_init(&fp->f_count, 1);
1962 	fp->f_cred = crhold(td->td_ucred);
1963 	fp->f_ops = &badfileops;
1964 	*resultfp = fp;
1965 	return (0);
1966 }
1967 
1968 /*
1969  * Install a file in a file descriptor table.
1970  */
1971 void
1972 _finstall(struct filedesc *fdp, struct file *fp, int fd, int flags,
1973     struct filecaps *fcaps)
1974 {
1975 	struct filedescent *fde;
1976 
1977 	MPASS(fp != NULL);
1978 	if (fcaps != NULL)
1979 		filecaps_validate(fcaps, __func__);
1980 	FILEDESC_XLOCK_ASSERT(fdp);
1981 
1982 	fde = &fdp->fd_ofiles[fd];
1983 #ifdef CAPABILITIES
1984 	seqc_write_begin(&fde->fde_seqc);
1985 #endif
1986 	fde->fde_file = fp;
1987 	fde->fde_flags = (flags & O_CLOEXEC) != 0 ? UF_EXCLOSE : 0;
1988 	if (fcaps != NULL)
1989 		filecaps_move(fcaps, &fde->fde_caps);
1990 	else
1991 		filecaps_fill(&fde->fde_caps);
1992 #ifdef CAPABILITIES
1993 	seqc_write_end(&fde->fde_seqc);
1994 #endif
1995 }
1996 
1997 int
1998 finstall(struct thread *td, struct file *fp, int *fd, int flags,
1999     struct filecaps *fcaps)
2000 {
2001 	struct filedesc *fdp = td->td_proc->p_fd;
2002 	int error;
2003 
2004 	MPASS(fd != NULL);
2005 
2006 	if (!fhold(fp))
2007 		return (EBADF);
2008 	FILEDESC_XLOCK(fdp);
2009 	error = fdalloc(td, 0, fd);
2010 	if (__predict_false(error != 0)) {
2011 		FILEDESC_XUNLOCK(fdp);
2012 		fdrop(fp, td);
2013 		return (error);
2014 	}
2015 	_finstall(fdp, fp, *fd, flags, fcaps);
2016 	FILEDESC_XUNLOCK(fdp);
2017 	return (0);
2018 }
2019 
2020 /*
2021  * Build a new filedesc structure from another.
2022  * Copy the current, root, and jail root vnode references.
2023  *
2024  * If fdp is not NULL, return with it shared locked.
2025  */
2026 struct filedesc *
2027 fdinit(struct filedesc *fdp, bool prepfiles, int *lastfile)
2028 {
2029 	struct filedesc0 *newfdp0;
2030 	struct filedesc *newfdp;
2031 	struct pwd *newpwd;
2032 
2033 	if (prepfiles)
2034 		MPASS(lastfile != NULL);
2035 	else
2036 		MPASS(lastfile == NULL);
2037 
2038 	newfdp0 = uma_zalloc(filedesc0_zone, M_WAITOK | M_ZERO);
2039 	newfdp = &newfdp0->fd_fd;
2040 
2041 	/* Create the file descriptor table. */
2042 	FILEDESC_LOCK_INIT(newfdp);
2043 	refcount_init(&newfdp->fd_refcnt, 1);
2044 	refcount_init(&newfdp->fd_holdcnt, 1);
2045 	newfdp->fd_cmask = CMASK;
2046 	newfdp->fd_map = newfdp0->fd_dmap;
2047 	newfdp->fd_files = (struct fdescenttbl *)&newfdp0->fd_dfiles;
2048 	newfdp->fd_files->fdt_nfiles = NDFILE;
2049 
2050 	if (fdp == NULL) {
2051 		newpwd = pwd_alloc();
2052 		smr_serialized_store(&newfdp->fd_pwd, newpwd, true);
2053 		return (newfdp);
2054 	}
2055 
2056 	FILEDESC_SLOCK(fdp);
2057 	newpwd = pwd_hold_filedesc(fdp);
2058 	smr_serialized_store(&newfdp->fd_pwd, newpwd, true);
2059 	if (!prepfiles) {
2060 		FILEDESC_SUNLOCK(fdp);
2061 		return (newfdp);
2062 	}
2063 
2064 	for (;;) {
2065 		*lastfile = fdlastfile(fdp);
2066 		if (*lastfile < newfdp->fd_nfiles)
2067 			break;
2068 		FILEDESC_SUNLOCK(fdp);
2069 		fdgrowtable(newfdp, *lastfile + 1);
2070 		FILEDESC_SLOCK(fdp);
2071 	}
2072 
2073 	return (newfdp);
2074 }
2075 
2076 static struct filedesc *
2077 fdhold(struct proc *p)
2078 {
2079 	struct filedesc *fdp;
2080 
2081 	PROC_LOCK_ASSERT(p, MA_OWNED);
2082 	fdp = p->p_fd;
2083 	if (fdp != NULL)
2084 		refcount_acquire(&fdp->fd_holdcnt);
2085 	return (fdp);
2086 }
2087 
2088 static void
2089 fddrop(struct filedesc *fdp)
2090 {
2091 
2092 	if (fdp->fd_holdcnt > 1) {
2093 		if (refcount_release(&fdp->fd_holdcnt) == 0)
2094 			return;
2095 	}
2096 
2097 	FILEDESC_LOCK_DESTROY(fdp);
2098 	uma_zfree(filedesc0_zone, fdp);
2099 }
2100 
2101 /*
2102  * Share a filedesc structure.
2103  */
2104 struct filedesc *
2105 fdshare(struct filedesc *fdp)
2106 {
2107 
2108 	refcount_acquire(&fdp->fd_refcnt);
2109 	return (fdp);
2110 }
2111 
2112 /*
2113  * Unshare a filedesc structure, if necessary by making a copy
2114  */
2115 void
2116 fdunshare(struct thread *td)
2117 {
2118 	struct filedesc *tmp;
2119 	struct proc *p = td->td_proc;
2120 
2121 	if (p->p_fd->fd_refcnt == 1)
2122 		return;
2123 
2124 	tmp = fdcopy(p->p_fd);
2125 	fdescfree(td);
2126 	p->p_fd = tmp;
2127 }
2128 
2129 void
2130 fdinstall_remapped(struct thread *td, struct filedesc *fdp)
2131 {
2132 
2133 	fdescfree(td);
2134 	td->td_proc->p_fd = fdp;
2135 }
2136 
2137 /*
2138  * Copy a filedesc structure.  A NULL pointer in returns a NULL reference,
2139  * this is to ease callers, not catch errors.
2140  */
2141 struct filedesc *
2142 fdcopy(struct filedesc *fdp)
2143 {
2144 	struct filedesc *newfdp;
2145 	struct filedescent *nfde, *ofde;
2146 	int i, lastfile;
2147 
2148 	MPASS(fdp != NULL);
2149 
2150 	newfdp = fdinit(fdp, true, &lastfile);
2151 	/* copy all passable descriptors (i.e. not kqueue) */
2152 	newfdp->fd_freefile = -1;
2153 	for (i = 0; i <= lastfile; ++i) {
2154 		ofde = &fdp->fd_ofiles[i];
2155 		if (ofde->fde_file == NULL ||
2156 		    (ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0 ||
2157 		    !fhold(ofde->fde_file)) {
2158 			if (newfdp->fd_freefile == -1)
2159 				newfdp->fd_freefile = i;
2160 			continue;
2161 		}
2162 		nfde = &newfdp->fd_ofiles[i];
2163 		*nfde = *ofde;
2164 		filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true);
2165 		fdused_init(newfdp, i);
2166 	}
2167 	if (newfdp->fd_freefile == -1)
2168 		newfdp->fd_freefile = i;
2169 	newfdp->fd_cmask = fdp->fd_cmask;
2170 	FILEDESC_SUNLOCK(fdp);
2171 	return (newfdp);
2172 }
2173 
2174 /*
2175  * Copies a filedesc structure, while remapping all file descriptors
2176  * stored inside using a translation table.
2177  *
2178  * File descriptors are copied over to the new file descriptor table,
2179  * regardless of whether the close-on-exec flag is set.
2180  */
2181 int
2182 fdcopy_remapped(struct filedesc *fdp, const int *fds, size_t nfds,
2183     struct filedesc **ret)
2184 {
2185 	struct filedesc *newfdp;
2186 	struct filedescent *nfde, *ofde;
2187 	int error, i, lastfile;
2188 
2189 	MPASS(fdp != NULL);
2190 
2191 	newfdp = fdinit(fdp, true, &lastfile);
2192 	if (nfds > lastfile + 1) {
2193 		/* New table cannot be larger than the old one. */
2194 		error = E2BIG;
2195 		goto bad;
2196 	}
2197 	/* Copy all passable descriptors (i.e. not kqueue). */
2198 	newfdp->fd_freefile = nfds;
2199 	for (i = 0; i < nfds; ++i) {
2200 		if (fds[i] < 0 || fds[i] > lastfile) {
2201 			/* File descriptor out of bounds. */
2202 			error = EBADF;
2203 			goto bad;
2204 		}
2205 		ofde = &fdp->fd_ofiles[fds[i]];
2206 		if (ofde->fde_file == NULL) {
2207 			/* Unused file descriptor. */
2208 			error = EBADF;
2209 			goto bad;
2210 		}
2211 		if ((ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0) {
2212 			/* File descriptor cannot be passed. */
2213 			error = EINVAL;
2214 			goto bad;
2215 		}
2216 		if (!fhold(nfde->fde_file)) {
2217 			error = EBADF;
2218 			goto bad;
2219 		}
2220 		nfde = &newfdp->fd_ofiles[i];
2221 		*nfde = *ofde;
2222 		filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true);
2223 		fdused_init(newfdp, i);
2224 	}
2225 	newfdp->fd_cmask = fdp->fd_cmask;
2226 	FILEDESC_SUNLOCK(fdp);
2227 	*ret = newfdp;
2228 	return (0);
2229 bad:
2230 	FILEDESC_SUNLOCK(fdp);
2231 	fdescfree_remapped(newfdp);
2232 	return (error);
2233 }
2234 
2235 /*
2236  * Clear POSIX style locks. This is only used when fdp looses a reference (i.e.
2237  * one of processes using it exits) and the table used to be shared.
2238  */
2239 static void
2240 fdclearlocks(struct thread *td)
2241 {
2242 	struct filedesc *fdp;
2243 	struct filedesc_to_leader *fdtol;
2244 	struct flock lf;
2245 	struct file *fp;
2246 	struct proc *p;
2247 	struct vnode *vp;
2248 	int i, lastfile;
2249 
2250 	p = td->td_proc;
2251 	fdp = p->p_fd;
2252 	fdtol = p->p_fdtol;
2253 	MPASS(fdtol != NULL);
2254 
2255 	FILEDESC_XLOCK(fdp);
2256 	KASSERT(fdtol->fdl_refcount > 0,
2257 	    ("filedesc_to_refcount botch: fdl_refcount=%d",
2258 	    fdtol->fdl_refcount));
2259 	if (fdtol->fdl_refcount == 1 &&
2260 	    (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2261 		lastfile = fdlastfile(fdp);
2262 		for (i = 0; i <= lastfile; i++) {
2263 			fp = fdp->fd_ofiles[i].fde_file;
2264 			if (fp == NULL || fp->f_type != DTYPE_VNODE ||
2265 			    !fhold(fp))
2266 				continue;
2267 			FILEDESC_XUNLOCK(fdp);
2268 			lf.l_whence = SEEK_SET;
2269 			lf.l_start = 0;
2270 			lf.l_len = 0;
2271 			lf.l_type = F_UNLCK;
2272 			vp = fp->f_vnode;
2273 			(void) VOP_ADVLOCK(vp,
2274 			    (caddr_t)p->p_leader, F_UNLCK,
2275 			    &lf, F_POSIX);
2276 			FILEDESC_XLOCK(fdp);
2277 			fdrop(fp, td);
2278 		}
2279 	}
2280 retry:
2281 	if (fdtol->fdl_refcount == 1) {
2282 		if (fdp->fd_holdleaderscount > 0 &&
2283 		    (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2284 			/*
2285 			 * close() or kern_dup() has cleared a reference
2286 			 * in a shared file descriptor table.
2287 			 */
2288 			fdp->fd_holdleaderswakeup = 1;
2289 			sx_sleep(&fdp->fd_holdleaderscount,
2290 			    FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0);
2291 			goto retry;
2292 		}
2293 		if (fdtol->fdl_holdcount > 0) {
2294 			/*
2295 			 * Ensure that fdtol->fdl_leader remains
2296 			 * valid in closef().
2297 			 */
2298 			fdtol->fdl_wakeup = 1;
2299 			sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK,
2300 			    "fdlhold", 0);
2301 			goto retry;
2302 		}
2303 	}
2304 	fdtol->fdl_refcount--;
2305 	if (fdtol->fdl_refcount == 0 &&
2306 	    fdtol->fdl_holdcount == 0) {
2307 		fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
2308 		fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
2309 	} else
2310 		fdtol = NULL;
2311 	p->p_fdtol = NULL;
2312 	FILEDESC_XUNLOCK(fdp);
2313 	if (fdtol != NULL)
2314 		free(fdtol, M_FILEDESC_TO_LEADER);
2315 }
2316 
2317 /*
2318  * Release a filedesc structure.
2319  */
2320 static void
2321 fdescfree_fds(struct thread *td, struct filedesc *fdp, bool needclose)
2322 {
2323 	struct filedesc0 *fdp0;
2324 	struct freetable *ft, *tft;
2325 	struct filedescent *fde;
2326 	struct file *fp;
2327 	int i, lastfile;
2328 
2329 	lastfile = fdlastfile_single(fdp);
2330 	for (i = 0; i <= lastfile; i++) {
2331 		fde = &fdp->fd_ofiles[i];
2332 		fp = fde->fde_file;
2333 		if (fp != NULL) {
2334 			fdefree_last(fde);
2335 			if (needclose)
2336 				(void) closef(fp, td);
2337 			else
2338 				fdrop(fp, td);
2339 		}
2340 	}
2341 
2342 	if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE))
2343 		free(fdp->fd_map, M_FILEDESC);
2344 	if (fdp->fd_nfiles > NDFILE)
2345 		free(fdp->fd_files, M_FILEDESC);
2346 
2347 	fdp0 = (struct filedesc0 *)fdp;
2348 	SLIST_FOREACH_SAFE(ft, &fdp0->fd_free, ft_next, tft)
2349 		free(ft->ft_table, M_FILEDESC);
2350 
2351 	fddrop(fdp);
2352 }
2353 
2354 void
2355 fdescfree(struct thread *td)
2356 {
2357 	struct proc *p;
2358 	struct filedesc *fdp;
2359 	struct pwd *pwd;
2360 
2361 	p = td->td_proc;
2362 	fdp = p->p_fd;
2363 	MPASS(fdp != NULL);
2364 
2365 #ifdef RACCT
2366 	if (RACCT_ENABLED())
2367 		racct_set_unlocked(p, RACCT_NOFILE, 0);
2368 #endif
2369 
2370 	if (p->p_fdtol != NULL)
2371 		fdclearlocks(td);
2372 
2373 	PROC_LOCK(p);
2374 	p->p_fd = NULL;
2375 	PROC_UNLOCK(p);
2376 
2377 	if (refcount_release(&fdp->fd_refcnt) == 0)
2378 		return;
2379 
2380 	FILEDESC_XLOCK(fdp);
2381 	pwd = FILEDESC_XLOCKED_LOAD_PWD(fdp);
2382 	pwd_set(fdp, NULL);
2383 	FILEDESC_XUNLOCK(fdp);
2384 
2385 	pwd_drop(pwd);
2386 
2387 	fdescfree_fds(td, fdp, 1);
2388 }
2389 
2390 void
2391 fdescfree_remapped(struct filedesc *fdp)
2392 {
2393 
2394 	pwd_drop(smr_serialized_load(&fdp->fd_pwd, true));
2395 	fdescfree_fds(curthread, fdp, 0);
2396 }
2397 
2398 /*
2399  * For setugid programs, we don't want to people to use that setugidness
2400  * to generate error messages which write to a file which otherwise would
2401  * otherwise be off-limits to the process.  We check for filesystems where
2402  * the vnode can change out from under us after execve (like [lin]procfs).
2403  *
2404  * Since fdsetugidsafety calls this only for fd 0, 1 and 2, this check is
2405  * sufficient.  We also don't check for setugidness since we know we are.
2406  */
2407 static bool
2408 is_unsafe(struct file *fp)
2409 {
2410 	struct vnode *vp;
2411 
2412 	if (fp->f_type != DTYPE_VNODE)
2413 		return (false);
2414 
2415 	vp = fp->f_vnode;
2416 	return ((vp->v_vflag & VV_PROCDEP) != 0);
2417 }
2418 
2419 /*
2420  * Make this setguid thing safe, if at all possible.
2421  */
2422 void
2423 fdsetugidsafety(struct thread *td)
2424 {
2425 	struct filedesc *fdp;
2426 	struct file *fp;
2427 	int i;
2428 
2429 	fdp = td->td_proc->p_fd;
2430 	KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2431 	MPASS(fdp->fd_nfiles >= 3);
2432 	for (i = 0; i <= 2; i++) {
2433 		fp = fdp->fd_ofiles[i].fde_file;
2434 		if (fp != NULL && is_unsafe(fp)) {
2435 			FILEDESC_XLOCK(fdp);
2436 			knote_fdclose(td, i);
2437 			/*
2438 			 * NULL-out descriptor prior to close to avoid
2439 			 * a race while close blocks.
2440 			 */
2441 			fdfree(fdp, i);
2442 			FILEDESC_XUNLOCK(fdp);
2443 			(void) closef(fp, td);
2444 		}
2445 	}
2446 }
2447 
2448 /*
2449  * If a specific file object occupies a specific file descriptor, close the
2450  * file descriptor entry and drop a reference on the file object.  This is a
2451  * convenience function to handle a subsequent error in a function that calls
2452  * falloc() that handles the race that another thread might have closed the
2453  * file descriptor out from under the thread creating the file object.
2454  */
2455 void
2456 fdclose(struct thread *td, struct file *fp, int idx)
2457 {
2458 	struct filedesc *fdp = td->td_proc->p_fd;
2459 
2460 	FILEDESC_XLOCK(fdp);
2461 	if (fdp->fd_ofiles[idx].fde_file == fp) {
2462 		fdfree(fdp, idx);
2463 		FILEDESC_XUNLOCK(fdp);
2464 		fdrop(fp, td);
2465 	} else
2466 		FILEDESC_XUNLOCK(fdp);
2467 }
2468 
2469 /*
2470  * Close any files on exec?
2471  */
2472 void
2473 fdcloseexec(struct thread *td)
2474 {
2475 	struct filedesc *fdp;
2476 	struct filedescent *fde;
2477 	struct file *fp;
2478 	int i, lastfile;
2479 
2480 	fdp = td->td_proc->p_fd;
2481 	KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2482 	lastfile = fdlastfile_single(fdp);
2483 	for (i = 0; i <= lastfile; i++) {
2484 		fde = &fdp->fd_ofiles[i];
2485 		fp = fde->fde_file;
2486 		if (fp != NULL && (fp->f_type == DTYPE_MQUEUE ||
2487 		    (fde->fde_flags & UF_EXCLOSE))) {
2488 			FILEDESC_XLOCK(fdp);
2489 			fdfree(fdp, i);
2490 			(void) closefp(fdp, i, fp, td, 0);
2491 			FILEDESC_UNLOCK_ASSERT(fdp);
2492 		}
2493 	}
2494 }
2495 
2496 /*
2497  * It is unsafe for set[ug]id processes to be started with file
2498  * descriptors 0..2 closed, as these descriptors are given implicit
2499  * significance in the Standard C library.  fdcheckstd() will create a
2500  * descriptor referencing /dev/null for each of stdin, stdout, and
2501  * stderr that is not already open.
2502  */
2503 int
2504 fdcheckstd(struct thread *td)
2505 {
2506 	struct filedesc *fdp;
2507 	register_t save;
2508 	int i, error, devnull;
2509 
2510 	fdp = td->td_proc->p_fd;
2511 	KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2512 	MPASS(fdp->fd_nfiles >= 3);
2513 	devnull = -1;
2514 	for (i = 0; i <= 2; i++) {
2515 		if (fdp->fd_ofiles[i].fde_file != NULL)
2516 			continue;
2517 
2518 		save = td->td_retval[0];
2519 		if (devnull != -1) {
2520 			error = kern_dup(td, FDDUP_FIXED, 0, devnull, i);
2521 		} else {
2522 			error = kern_openat(td, AT_FDCWD, "/dev/null",
2523 			    UIO_SYSSPACE, O_RDWR, 0);
2524 			if (error == 0) {
2525 				devnull = td->td_retval[0];
2526 				KASSERT(devnull == i, ("we didn't get our fd"));
2527 			}
2528 		}
2529 		td->td_retval[0] = save;
2530 		if (error != 0)
2531 			return (error);
2532 	}
2533 	return (0);
2534 }
2535 
2536 /*
2537  * Internal form of close.  Decrement reference count on file structure.
2538  * Note: td may be NULL when closing a file that was being passed in a
2539  * message.
2540  */
2541 int
2542 closef(struct file *fp, struct thread *td)
2543 {
2544 	struct vnode *vp;
2545 	struct flock lf;
2546 	struct filedesc_to_leader *fdtol;
2547 	struct filedesc *fdp;
2548 
2549 	/*
2550 	 * POSIX record locking dictates that any close releases ALL
2551 	 * locks owned by this process.  This is handled by setting
2552 	 * a flag in the unlock to free ONLY locks obeying POSIX
2553 	 * semantics, and not to free BSD-style file locks.
2554 	 * If the descriptor was in a message, POSIX-style locks
2555 	 * aren't passed with the descriptor, and the thread pointer
2556 	 * will be NULL.  Callers should be careful only to pass a
2557 	 * NULL thread pointer when there really is no owning
2558 	 * context that might have locks, or the locks will be
2559 	 * leaked.
2560 	 */
2561 	if (fp->f_type == DTYPE_VNODE && td != NULL) {
2562 		vp = fp->f_vnode;
2563 		if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
2564 			lf.l_whence = SEEK_SET;
2565 			lf.l_start = 0;
2566 			lf.l_len = 0;
2567 			lf.l_type = F_UNLCK;
2568 			(void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader,
2569 			    F_UNLCK, &lf, F_POSIX);
2570 		}
2571 		fdtol = td->td_proc->p_fdtol;
2572 		if (fdtol != NULL) {
2573 			/*
2574 			 * Handle special case where file descriptor table is
2575 			 * shared between multiple process leaders.
2576 			 */
2577 			fdp = td->td_proc->p_fd;
2578 			FILEDESC_XLOCK(fdp);
2579 			for (fdtol = fdtol->fdl_next;
2580 			    fdtol != td->td_proc->p_fdtol;
2581 			    fdtol = fdtol->fdl_next) {
2582 				if ((fdtol->fdl_leader->p_flag &
2583 				    P_ADVLOCK) == 0)
2584 					continue;
2585 				fdtol->fdl_holdcount++;
2586 				FILEDESC_XUNLOCK(fdp);
2587 				lf.l_whence = SEEK_SET;
2588 				lf.l_start = 0;
2589 				lf.l_len = 0;
2590 				lf.l_type = F_UNLCK;
2591 				vp = fp->f_vnode;
2592 				(void) VOP_ADVLOCK(vp,
2593 				    (caddr_t)fdtol->fdl_leader, F_UNLCK, &lf,
2594 				    F_POSIX);
2595 				FILEDESC_XLOCK(fdp);
2596 				fdtol->fdl_holdcount--;
2597 				if (fdtol->fdl_holdcount == 0 &&
2598 				    fdtol->fdl_wakeup != 0) {
2599 					fdtol->fdl_wakeup = 0;
2600 					wakeup(fdtol);
2601 				}
2602 			}
2603 			FILEDESC_XUNLOCK(fdp);
2604 		}
2605 	}
2606 	return (fdrop(fp, td));
2607 }
2608 
2609 /*
2610  * Initialize the file pointer with the specified properties.
2611  *
2612  * The ops are set with release semantics to be certain that the flags, type,
2613  * and data are visible when ops is.  This is to prevent ops methods from being
2614  * called with bad data.
2615  */
2616 void
2617 finit(struct file *fp, u_int flag, short type, void *data, struct fileops *ops)
2618 {
2619 	fp->f_data = data;
2620 	fp->f_flag = flag;
2621 	fp->f_type = type;
2622 	atomic_store_rel_ptr((volatile uintptr_t *)&fp->f_ops, (uintptr_t)ops);
2623 }
2624 
2625 int
2626 fget_cap_locked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
2627     struct file **fpp, struct filecaps *havecapsp)
2628 {
2629 	struct filedescent *fde;
2630 	int error;
2631 
2632 	FILEDESC_LOCK_ASSERT(fdp);
2633 
2634 	fde = fdeget_locked(fdp, fd);
2635 	if (fde == NULL) {
2636 		error = EBADF;
2637 		goto out;
2638 	}
2639 
2640 #ifdef CAPABILITIES
2641 	error = cap_check(cap_rights_fde_inline(fde), needrightsp);
2642 	if (error != 0)
2643 		goto out;
2644 #endif
2645 
2646 	if (havecapsp != NULL)
2647 		filecaps_copy(&fde->fde_caps, havecapsp, true);
2648 
2649 	*fpp = fde->fde_file;
2650 
2651 	error = 0;
2652 out:
2653 	return (error);
2654 }
2655 
2656 int
2657 fget_cap(struct thread *td, int fd, cap_rights_t *needrightsp,
2658     struct file **fpp, struct filecaps *havecapsp)
2659 {
2660 	struct filedesc *fdp = td->td_proc->p_fd;
2661 	int error;
2662 #ifndef CAPABILITIES
2663 	error = fget_unlocked(fdp, fd, needrightsp, fpp);
2664 	if (havecapsp != NULL && error == 0)
2665 		filecaps_fill(havecapsp);
2666 #else
2667 	struct file *fp;
2668 	seqc_t seq;
2669 
2670 	*fpp = NULL;
2671 	for (;;) {
2672 		error = fget_unlocked_seq(fdp, fd, needrightsp, &fp, &seq);
2673 		if (error != 0)
2674 			return (error);
2675 
2676 		if (havecapsp != NULL) {
2677 			if (!filecaps_copy(&fdp->fd_ofiles[fd].fde_caps,
2678 			    havecapsp, false)) {
2679 				fdrop(fp, td);
2680 				goto get_locked;
2681 			}
2682 		}
2683 
2684 		if (!fd_modified(fdp, fd, seq))
2685 			break;
2686 		fdrop(fp, td);
2687 	}
2688 
2689 	*fpp = fp;
2690 	return (0);
2691 
2692 get_locked:
2693 	FILEDESC_SLOCK(fdp);
2694 	error = fget_cap_locked(fdp, fd, needrightsp, fpp, havecapsp);
2695 	if (error == 0 && !fhold(*fpp))
2696 		error = EBADF;
2697 	FILEDESC_SUNLOCK(fdp);
2698 #endif
2699 	return (error);
2700 }
2701 
2702 int
2703 fget_unlocked_seq(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
2704     struct file **fpp, seqc_t *seqp)
2705 {
2706 #ifdef CAPABILITIES
2707 	const struct filedescent *fde;
2708 #endif
2709 	const struct fdescenttbl *fdt;
2710 	struct file *fp;
2711 #ifdef CAPABILITIES
2712 	seqc_t seq;
2713 	cap_rights_t haverights;
2714 	int error;
2715 #endif
2716 
2717 	fdt = fdp->fd_files;
2718 	if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
2719 		return (EBADF);
2720 	/*
2721 	 * Fetch the descriptor locklessly.  We avoid fdrop() races by
2722 	 * never raising a refcount above 0.  To accomplish this we have
2723 	 * to use a cmpset loop rather than an atomic_add.  The descriptor
2724 	 * must be re-verified once we acquire a reference to be certain
2725 	 * that the identity is still correct and we did not lose a race
2726 	 * due to preemption.
2727 	 */
2728 	for (;;) {
2729 #ifdef CAPABILITIES
2730 		seq = seqc_read(fd_seqc(fdt, fd));
2731 		fde = &fdt->fdt_ofiles[fd];
2732 		haverights = *cap_rights_fde_inline(fde);
2733 		fp = fde->fde_file;
2734 		if (!seqc_consistent(fd_seqc(fdt, fd), seq))
2735 			continue;
2736 #else
2737 		fp = fdt->fdt_ofiles[fd].fde_file;
2738 #endif
2739 		if (fp == NULL)
2740 			return (EBADF);
2741 #ifdef CAPABILITIES
2742 		error = cap_check_inline(&haverights, needrightsp);
2743 		if (error != 0)
2744 			return (error);
2745 #endif
2746 		if (__predict_false(!refcount_acquire_if_not_zero(&fp->f_count))) {
2747 			/*
2748 			 * The count was found either saturated or zero.
2749 			 * This re-read is not any more racy than using the
2750 			 * return value from fcmpset.
2751 			 */
2752 			if (fp->f_count != 0)
2753 				return (EBADF);
2754 			/*
2755 			 * Force a reload. Other thread could reallocate the
2756 			 * table before this fd was closed, so it is possible
2757 			 * that there is a stale fp pointer in cached version.
2758 			 */
2759 			fdt = atomic_load_ptr(&fdp->fd_files);
2760 			continue;
2761 		}
2762 		/*
2763 		 * Use an acquire barrier to force re-reading of fdt so it is
2764 		 * refreshed for verification.
2765 		 */
2766 		atomic_thread_fence_acq();
2767 		fdt = fdp->fd_files;
2768 #ifdef	CAPABILITIES
2769 		if (seqc_consistent_nomb(fd_seqc(fdt, fd), seq))
2770 #else
2771 		if (fp == fdt->fdt_ofiles[fd].fde_file)
2772 #endif
2773 			break;
2774 		fdrop(fp, curthread);
2775 	}
2776 	*fpp = fp;
2777 	if (seqp != NULL) {
2778 #ifdef CAPABILITIES
2779 		*seqp = seq;
2780 #endif
2781 	}
2782 	return (0);
2783 }
2784 
2785 /*
2786  * See the comments in fget_unlocked_seq for an explanation of how this works.
2787  *
2788  * This is a simplified variant which bails out to the aforementioned routine
2789  * if anything goes wrong. In practice this only happens when userspace is
2790  * racing with itself.
2791  */
2792 int
2793 fget_unlocked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
2794     struct file **fpp)
2795 {
2796 #ifdef CAPABILITIES
2797 	const struct filedescent *fde;
2798 #endif
2799 	const struct fdescenttbl *fdt;
2800 	struct file *fp;
2801 #ifdef CAPABILITIES
2802 	seqc_t seq;
2803 	const cap_rights_t *haverights;
2804 #endif
2805 
2806 	fdt = fdp->fd_files;
2807 	if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
2808 		return (EBADF);
2809 #ifdef CAPABILITIES
2810 	seq = seqc_read_any(fd_seqc(fdt, fd));
2811 	if (__predict_false(seqc_in_modify(seq)))
2812 		goto out_fallback;
2813 	fde = &fdt->fdt_ofiles[fd];
2814 	haverights = cap_rights_fde_inline(fde);
2815 	fp = fde->fde_file;
2816 #else
2817 	fp = fdt->fdt_ofiles[fd].fde_file;
2818 #endif
2819 	if (__predict_false(fp == NULL))
2820 		goto out_fallback;
2821 #ifdef CAPABILITIES
2822 	if (__predict_false(cap_check_inline_transient(haverights, needrightsp)))
2823 		goto out_fallback;
2824 #endif
2825 	if (__predict_false(!refcount_acquire_if_not_zero(&fp->f_count)))
2826 		goto out_fallback;
2827 
2828 	/*
2829 	 * Use an acquire barrier to force re-reading of fdt so it is
2830 	 * refreshed for verification.
2831 	 */
2832 	atomic_thread_fence_acq();
2833 	fdt = fdp->fd_files;
2834 #ifdef	CAPABILITIES
2835 	if (__predict_false(!seqc_consistent_nomb(fd_seqc(fdt, fd), seq)))
2836 #else
2837 	if (__predict_false(fp != fdt->fdt_ofiles[fd].fde_file))
2838 #endif
2839 		goto out_fdrop;
2840 	*fpp = fp;
2841 	return (0);
2842 out_fdrop:
2843 	fdrop(fp, curthread);
2844 out_fallback:
2845 	return (fget_unlocked_seq(fdp, fd, needrightsp, fpp, NULL));
2846 }
2847 
2848 /*
2849  * Extract the file pointer associated with the specified descriptor for the
2850  * current user process.
2851  *
2852  * If the descriptor doesn't exist or doesn't match 'flags', EBADF is
2853  * returned.
2854  *
2855  * File's rights will be checked against the capability rights mask.
2856  *
2857  * If an error occurred the non-zero error is returned and *fpp is set to
2858  * NULL.  Otherwise *fpp is held and set and zero is returned.  Caller is
2859  * responsible for fdrop().
2860  */
2861 static __inline int
2862 _fget(struct thread *td, int fd, struct file **fpp, int flags,
2863     cap_rights_t *needrightsp)
2864 {
2865 	struct filedesc *fdp;
2866 	struct file *fp;
2867 	int error;
2868 
2869 	*fpp = NULL;
2870 	fdp = td->td_proc->p_fd;
2871 	error = fget_unlocked(fdp, fd, needrightsp, &fp);
2872 	if (__predict_false(error != 0))
2873 		return (error);
2874 	if (__predict_false(fp->f_ops == &badfileops)) {
2875 		fdrop(fp, td);
2876 		return (EBADF);
2877 	}
2878 
2879 	/*
2880 	 * FREAD and FWRITE failure return EBADF as per POSIX.
2881 	 */
2882 	error = 0;
2883 	switch (flags) {
2884 	case FREAD:
2885 	case FWRITE:
2886 		if ((fp->f_flag & flags) == 0)
2887 			error = EBADF;
2888 		break;
2889 	case FEXEC:
2890 	    	if ((fp->f_flag & (FREAD | FEXEC)) == 0 ||
2891 		    ((fp->f_flag & FWRITE) != 0))
2892 			error = EBADF;
2893 		break;
2894 	case 0:
2895 		break;
2896 	default:
2897 		KASSERT(0, ("wrong flags"));
2898 	}
2899 
2900 	if (error != 0) {
2901 		fdrop(fp, td);
2902 		return (error);
2903 	}
2904 
2905 	*fpp = fp;
2906 	return (0);
2907 }
2908 
2909 int
2910 fget(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
2911 {
2912 
2913 	return (_fget(td, fd, fpp, 0, rightsp));
2914 }
2915 
2916 int
2917 fget_mmap(struct thread *td, int fd, cap_rights_t *rightsp, vm_prot_t *maxprotp,
2918     struct file **fpp)
2919 {
2920 	int error;
2921 #ifndef CAPABILITIES
2922 	error = _fget(td, fd, fpp, 0, rightsp);
2923 	if (maxprotp != NULL)
2924 		*maxprotp = VM_PROT_ALL;
2925 	return (error);
2926 #else
2927 	cap_rights_t fdrights;
2928 	struct filedesc *fdp;
2929 	struct file *fp;
2930 	seqc_t seq;
2931 
2932 	*fpp = NULL;
2933 	fdp = td->td_proc->p_fd;
2934 	MPASS(cap_rights_is_set(rightsp, CAP_MMAP));
2935 	for (;;) {
2936 		error = fget_unlocked_seq(fdp, fd, rightsp, &fp, &seq);
2937 		if (__predict_false(error != 0))
2938 			return (error);
2939 		if (__predict_false(fp->f_ops == &badfileops)) {
2940 			fdrop(fp, td);
2941 			return (EBADF);
2942 		}
2943 		if (maxprotp != NULL)
2944 			fdrights = *cap_rights(fdp, fd);
2945 		if (!fd_modified(fdp, fd, seq))
2946 			break;
2947 		fdrop(fp, td);
2948 	}
2949 
2950 	/*
2951 	 * If requested, convert capability rights to access flags.
2952 	 */
2953 	if (maxprotp != NULL)
2954 		*maxprotp = cap_rights_to_vmprot(&fdrights);
2955 	*fpp = fp;
2956 	return (0);
2957 #endif
2958 }
2959 
2960 int
2961 fget_read(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
2962 {
2963 
2964 	return (_fget(td, fd, fpp, FREAD, rightsp));
2965 }
2966 
2967 int
2968 fget_write(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
2969 {
2970 
2971 	return (_fget(td, fd, fpp, FWRITE, rightsp));
2972 }
2973 
2974 int
2975 fget_fcntl(struct thread *td, int fd, cap_rights_t *rightsp, int needfcntl,
2976     struct file **fpp)
2977 {
2978 	struct filedesc *fdp = td->td_proc->p_fd;
2979 #ifndef CAPABILITIES
2980 	return (fget_unlocked(fdp, fd, rightsp, fpp));
2981 #else
2982 	struct file *fp;
2983 	int error;
2984 	seqc_t seq;
2985 
2986 	*fpp = NULL;
2987 	MPASS(cap_rights_is_set(rightsp, CAP_FCNTL));
2988 	for (;;) {
2989 		error = fget_unlocked_seq(fdp, fd, rightsp, &fp, &seq);
2990 		if (error != 0)
2991 			return (error);
2992 		error = cap_fcntl_check(fdp, fd, needfcntl);
2993 		if (!fd_modified(fdp, fd, seq))
2994 			break;
2995 		fdrop(fp, td);
2996 	}
2997 	if (error != 0) {
2998 		fdrop(fp, td);
2999 		return (error);
3000 	}
3001 	*fpp = fp;
3002 	return (0);
3003 #endif
3004 }
3005 
3006 /*
3007  * Like fget() but loads the underlying vnode, or returns an error if the
3008  * descriptor does not represent a vnode.  Note that pipes use vnodes but
3009  * never have VM objects.  The returned vnode will be vref()'d.
3010  *
3011  * XXX: what about the unused flags ?
3012  */
3013 static __inline int
3014 _fgetvp(struct thread *td, int fd, int flags, cap_rights_t *needrightsp,
3015     struct vnode **vpp)
3016 {
3017 	struct file *fp;
3018 	int error;
3019 
3020 	*vpp = NULL;
3021 	error = _fget(td, fd, &fp, flags, needrightsp);
3022 	if (error != 0)
3023 		return (error);
3024 	if (fp->f_vnode == NULL) {
3025 		error = EINVAL;
3026 	} else {
3027 		*vpp = fp->f_vnode;
3028 		vrefact(*vpp);
3029 	}
3030 	fdrop(fp, td);
3031 
3032 	return (error);
3033 }
3034 
3035 int
3036 fgetvp(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
3037 {
3038 
3039 	return (_fgetvp(td, fd, 0, rightsp, vpp));
3040 }
3041 
3042 int
3043 fgetvp_rights(struct thread *td, int fd, cap_rights_t *needrightsp,
3044     struct filecaps *havecaps, struct vnode **vpp)
3045 {
3046 	struct filecaps caps;
3047 	struct file *fp;
3048 	int error;
3049 
3050 	error = fget_cap(td, fd, needrightsp, &fp, &caps);
3051 	if (error != 0)
3052 		return (error);
3053 	if (fp->f_ops == &badfileops) {
3054 		error = EBADF;
3055 		goto out;
3056 	}
3057 	if (fp->f_vnode == NULL) {
3058 		error = EINVAL;
3059 		goto out;
3060 	}
3061 
3062 	*havecaps = caps;
3063 	*vpp = fp->f_vnode;
3064 	vrefact(*vpp);
3065 	fdrop(fp, td);
3066 
3067 	return (0);
3068 out:
3069 	filecaps_free(&caps);
3070 	fdrop(fp, td);
3071 	return (error);
3072 }
3073 
3074 int
3075 fgetvp_read(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
3076 {
3077 
3078 	return (_fgetvp(td, fd, FREAD, rightsp, vpp));
3079 }
3080 
3081 int
3082 fgetvp_exec(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
3083 {
3084 
3085 	return (_fgetvp(td, fd, FEXEC, rightsp, vpp));
3086 }
3087 
3088 #ifdef notyet
3089 int
3090 fgetvp_write(struct thread *td, int fd, cap_rights_t *rightsp,
3091     struct vnode **vpp)
3092 {
3093 
3094 	return (_fgetvp(td, fd, FWRITE, rightsp, vpp));
3095 }
3096 #endif
3097 
3098 /*
3099  * Handle the last reference to a file being closed.
3100  *
3101  * Without the noinline attribute clang keeps inlining the func thorough this
3102  * file when fdrop is used.
3103  */
3104 int __noinline
3105 _fdrop(struct file *fp, struct thread *td)
3106 {
3107 	int error;
3108 
3109 	if (fp->f_count != 0)
3110 		panic("fdrop: count %d", fp->f_count);
3111 	error = fo_close(fp, td);
3112 	atomic_subtract_int(&openfiles, 1);
3113 	crfree(fp->f_cred);
3114 	free(fp->f_advice, M_FADVISE);
3115 	uma_zfree(file_zone, fp);
3116 
3117 	return (error);
3118 }
3119 
3120 /*
3121  * Apply an advisory lock on a file descriptor.
3122  *
3123  * Just attempt to get a record lock of the requested type on the entire file
3124  * (l_whence = SEEK_SET, l_start = 0, l_len = 0).
3125  */
3126 #ifndef _SYS_SYSPROTO_H_
3127 struct flock_args {
3128 	int	fd;
3129 	int	how;
3130 };
3131 #endif
3132 /* ARGSUSED */
3133 int
3134 sys_flock(struct thread *td, struct flock_args *uap)
3135 {
3136 	struct file *fp;
3137 	struct vnode *vp;
3138 	struct flock lf;
3139 	int error;
3140 
3141 	error = fget(td, uap->fd, &cap_flock_rights, &fp);
3142 	if (error != 0)
3143 		return (error);
3144 	if (fp->f_type != DTYPE_VNODE) {
3145 		fdrop(fp, td);
3146 		return (EOPNOTSUPP);
3147 	}
3148 
3149 	vp = fp->f_vnode;
3150 	lf.l_whence = SEEK_SET;
3151 	lf.l_start = 0;
3152 	lf.l_len = 0;
3153 	if (uap->how & LOCK_UN) {
3154 		lf.l_type = F_UNLCK;
3155 		atomic_clear_int(&fp->f_flag, FHASLOCK);
3156 		error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK);
3157 		goto done2;
3158 	}
3159 	if (uap->how & LOCK_EX)
3160 		lf.l_type = F_WRLCK;
3161 	else if (uap->how & LOCK_SH)
3162 		lf.l_type = F_RDLCK;
3163 	else {
3164 		error = EBADF;
3165 		goto done2;
3166 	}
3167 	atomic_set_int(&fp->f_flag, FHASLOCK);
3168 	error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf,
3169 	    (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT);
3170 done2:
3171 	fdrop(fp, td);
3172 	return (error);
3173 }
3174 /*
3175  * Duplicate the specified descriptor to a free descriptor.
3176  */
3177 int
3178 dupfdopen(struct thread *td, struct filedesc *fdp, int dfd, int mode,
3179     int openerror, int *indxp)
3180 {
3181 	struct filedescent *newfde, *oldfde;
3182 	struct file *fp;
3183 	u_long *ioctls;
3184 	int error, indx;
3185 
3186 	KASSERT(openerror == ENODEV || openerror == ENXIO,
3187 	    ("unexpected error %d in %s", openerror, __func__));
3188 
3189 	/*
3190 	 * If the to-be-dup'd fd number is greater than the allowed number
3191 	 * of file descriptors, or the fd to be dup'd has already been
3192 	 * closed, then reject.
3193 	 */
3194 	FILEDESC_XLOCK(fdp);
3195 	if ((fp = fget_locked(fdp, dfd)) == NULL) {
3196 		FILEDESC_XUNLOCK(fdp);
3197 		return (EBADF);
3198 	}
3199 
3200 	error = fdalloc(td, 0, &indx);
3201 	if (error != 0) {
3202 		FILEDESC_XUNLOCK(fdp);
3203 		return (error);
3204 	}
3205 
3206 	/*
3207 	 * There are two cases of interest here.
3208 	 *
3209 	 * For ENODEV simply dup (dfd) to file descriptor (indx) and return.
3210 	 *
3211 	 * For ENXIO steal away the file structure from (dfd) and store it in
3212 	 * (indx).  (dfd) is effectively closed by this operation.
3213 	 */
3214 	switch (openerror) {
3215 	case ENODEV:
3216 		/*
3217 		 * Check that the mode the file is being opened for is a
3218 		 * subset of the mode of the existing descriptor.
3219 		 */
3220 		if (((mode & (FREAD|FWRITE)) | fp->f_flag) != fp->f_flag) {
3221 			fdunused(fdp, indx);
3222 			FILEDESC_XUNLOCK(fdp);
3223 			return (EACCES);
3224 		}
3225 		if (!fhold(fp)) {
3226 			fdunused(fdp, indx);
3227 			FILEDESC_XUNLOCK(fdp);
3228 			return (EBADF);
3229 		}
3230 		newfde = &fdp->fd_ofiles[indx];
3231 		oldfde = &fdp->fd_ofiles[dfd];
3232 		ioctls = filecaps_copy_prep(&oldfde->fde_caps);
3233 #ifdef CAPABILITIES
3234 		seqc_write_begin(&newfde->fde_seqc);
3235 #endif
3236 		memcpy(newfde, oldfde, fde_change_size);
3237 		filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps,
3238 		    ioctls);
3239 #ifdef CAPABILITIES
3240 		seqc_write_end(&newfde->fde_seqc);
3241 #endif
3242 		break;
3243 	case ENXIO:
3244 		/*
3245 		 * Steal away the file pointer from dfd and stuff it into indx.
3246 		 */
3247 		newfde = &fdp->fd_ofiles[indx];
3248 		oldfde = &fdp->fd_ofiles[dfd];
3249 #ifdef CAPABILITIES
3250 		seqc_write_begin(&newfde->fde_seqc);
3251 #endif
3252 		memcpy(newfde, oldfde, fde_change_size);
3253 		oldfde->fde_file = NULL;
3254 		fdunused(fdp, dfd);
3255 #ifdef CAPABILITIES
3256 		seqc_write_end(&newfde->fde_seqc);
3257 #endif
3258 		break;
3259 	}
3260 	FILEDESC_XUNLOCK(fdp);
3261 	*indxp = indx;
3262 	return (0);
3263 }
3264 
3265 /*
3266  * This sysctl determines if we will allow a process to chroot(2) if it
3267  * has a directory open:
3268  *	0: disallowed for all processes.
3269  *	1: allowed for processes that were not already chroot(2)'ed.
3270  *	2: allowed for all processes.
3271  */
3272 
3273 static int chroot_allow_open_directories = 1;
3274 
3275 SYSCTL_INT(_kern, OID_AUTO, chroot_allow_open_directories, CTLFLAG_RW,
3276     &chroot_allow_open_directories, 0,
3277     "Allow a process to chroot(2) if it has a directory open");
3278 
3279 /*
3280  * Helper function for raised chroot(2) security function:  Refuse if
3281  * any filedescriptors are open directories.
3282  */
3283 static int
3284 chroot_refuse_vdir_fds(struct filedesc *fdp)
3285 {
3286 	struct vnode *vp;
3287 	struct file *fp;
3288 	int fd, lastfile;
3289 
3290 	FILEDESC_LOCK_ASSERT(fdp);
3291 
3292 	lastfile = fdlastfile(fdp);
3293 	for (fd = 0; fd <= lastfile; fd++) {
3294 		fp = fget_locked(fdp, fd);
3295 		if (fp == NULL)
3296 			continue;
3297 		if (fp->f_type == DTYPE_VNODE) {
3298 			vp = fp->f_vnode;
3299 			if (vp->v_type == VDIR)
3300 				return (EPERM);
3301 		}
3302 	}
3303 	return (0);
3304 }
3305 
3306 static void
3307 pwd_fill(struct pwd *oldpwd, struct pwd *newpwd)
3308 {
3309 
3310 	if (newpwd->pwd_cdir == NULL && oldpwd->pwd_cdir != NULL) {
3311 		vrefact(oldpwd->pwd_cdir);
3312 		newpwd->pwd_cdir = oldpwd->pwd_cdir;
3313 	}
3314 
3315 	if (newpwd->pwd_rdir == NULL && oldpwd->pwd_rdir != NULL) {
3316 		vrefact(oldpwd->pwd_rdir);
3317 		newpwd->pwd_rdir = oldpwd->pwd_rdir;
3318 	}
3319 
3320 	if (newpwd->pwd_jdir == NULL && oldpwd->pwd_jdir != NULL) {
3321 		vrefact(oldpwd->pwd_jdir);
3322 		newpwd->pwd_jdir = oldpwd->pwd_jdir;
3323 	}
3324 }
3325 
3326 struct pwd *
3327 pwd_hold_filedesc(struct filedesc *fdp)
3328 {
3329 	struct pwd *pwd;
3330 
3331 	FILEDESC_LOCK_ASSERT(fdp);
3332 	pwd = FILEDESC_LOCKED_LOAD_PWD(fdp);
3333 	if (pwd != NULL)
3334 		refcount_acquire(&pwd->pwd_refcount);
3335 	return (pwd);
3336 }
3337 
3338 struct pwd *
3339 pwd_hold(struct thread *td)
3340 {
3341 	struct filedesc *fdp;
3342 	struct pwd *pwd;
3343 
3344 	fdp = td->td_proc->p_fd;
3345 
3346 	vfs_smr_enter();
3347 	pwd = vfs_smr_entered_load(&fdp->fd_pwd);
3348 	MPASS(pwd != NULL);
3349 	if (__predict_true(refcount_acquire_if_not_zero(&pwd->pwd_refcount))) {
3350 		vfs_smr_exit();
3351 		return (pwd);
3352 	}
3353 	vfs_smr_exit();
3354 	FILEDESC_SLOCK(fdp);
3355 	pwd = pwd_hold_filedesc(fdp);
3356 	MPASS(pwd != NULL);
3357 	FILEDESC_SUNLOCK(fdp);
3358 	return (pwd);
3359 }
3360 
3361 struct pwd *
3362 pwd_get_smr(void)
3363 {
3364 	struct pwd *pwd;
3365 
3366 	pwd = vfs_smr_entered_load(&curproc->p_fd->fd_pwd);
3367 	MPASS(pwd != NULL);
3368 	return (pwd);
3369 }
3370 
3371 static struct pwd *
3372 pwd_alloc(void)
3373 {
3374 	struct pwd *pwd;
3375 
3376 	pwd = uma_zalloc_smr(pwd_zone, M_WAITOK);
3377 	bzero(pwd, sizeof(*pwd));
3378 	refcount_init(&pwd->pwd_refcount, 1);
3379 	return (pwd);
3380 }
3381 
3382 void
3383 pwd_drop(struct pwd *pwd)
3384 {
3385 
3386 	if (!refcount_release(&pwd->pwd_refcount))
3387 		return;
3388 
3389 	if (pwd->pwd_cdir != NULL)
3390 		vrele(pwd->pwd_cdir);
3391 	if (pwd->pwd_rdir != NULL)
3392 		vrele(pwd->pwd_rdir);
3393 	if (pwd->pwd_jdir != NULL)
3394 		vrele(pwd->pwd_jdir);
3395 	uma_zfree_smr(pwd_zone, pwd);
3396 }
3397 
3398 /*
3399 * Common routine for kern_chroot() and jail_attach().  The caller is
3400 * responsible for invoking priv_check() and mac_vnode_check_chroot() to
3401 * authorize this operation.
3402 */
3403 int
3404 pwd_chroot(struct thread *td, struct vnode *vp)
3405 {
3406 	struct filedesc *fdp;
3407 	struct pwd *newpwd, *oldpwd;
3408 	int error;
3409 
3410 	fdp = td->td_proc->p_fd;
3411 	newpwd = pwd_alloc();
3412 	FILEDESC_XLOCK(fdp);
3413 	oldpwd = FILEDESC_XLOCKED_LOAD_PWD(fdp);
3414 	if (chroot_allow_open_directories == 0 ||
3415 	    (chroot_allow_open_directories == 1 &&
3416 	    oldpwd->pwd_rdir != rootvnode)) {
3417 		error = chroot_refuse_vdir_fds(fdp);
3418 		if (error != 0) {
3419 			FILEDESC_XUNLOCK(fdp);
3420 			pwd_drop(newpwd);
3421 			return (error);
3422 		}
3423 	}
3424 
3425 	vrefact(vp);
3426 	newpwd->pwd_rdir = vp;
3427 	if (oldpwd->pwd_jdir == NULL) {
3428 		vrefact(vp);
3429 		newpwd->pwd_jdir = vp;
3430 	}
3431 	pwd_fill(oldpwd, newpwd);
3432 	pwd_set(fdp, newpwd);
3433 	FILEDESC_XUNLOCK(fdp);
3434 	pwd_drop(oldpwd);
3435 	return (0);
3436 }
3437 
3438 void
3439 pwd_chdir(struct thread *td, struct vnode *vp)
3440 {
3441 	struct filedesc *fdp;
3442 	struct pwd *newpwd, *oldpwd;
3443 
3444 	VNPASS(vp->v_usecount > 0, vp);
3445 
3446 	newpwd = pwd_alloc();
3447 	fdp = td->td_proc->p_fd;
3448 	FILEDESC_XLOCK(fdp);
3449 	oldpwd = FILEDESC_XLOCKED_LOAD_PWD(fdp);
3450 	newpwd->pwd_cdir = vp;
3451 	pwd_fill(oldpwd, newpwd);
3452 	pwd_set(fdp, newpwd);
3453 	FILEDESC_XUNLOCK(fdp);
3454 	pwd_drop(oldpwd);
3455 }
3456 
3457 void
3458 pwd_ensure_dirs(void)
3459 {
3460 	struct filedesc *fdp;
3461 	struct pwd *oldpwd, *newpwd;
3462 
3463 	fdp = curproc->p_fd;
3464 	FILEDESC_XLOCK(fdp);
3465 	oldpwd = FILEDESC_XLOCKED_LOAD_PWD(fdp);
3466 	if (oldpwd->pwd_cdir != NULL && oldpwd->pwd_rdir != NULL) {
3467 		FILEDESC_XUNLOCK(fdp);
3468 		return;
3469 	}
3470 	FILEDESC_XUNLOCK(fdp);
3471 
3472 	newpwd = pwd_alloc();
3473 	FILEDESC_XLOCK(fdp);
3474 	oldpwd = FILEDESC_XLOCKED_LOAD_PWD(fdp);
3475 	pwd_fill(oldpwd, newpwd);
3476 	if (newpwd->pwd_cdir == NULL) {
3477 		vrefact(rootvnode);
3478 		newpwd->pwd_cdir = rootvnode;
3479 	}
3480 	if (newpwd->pwd_rdir == NULL) {
3481 		vrefact(rootvnode);
3482 		newpwd->pwd_rdir = rootvnode;
3483 	}
3484 	pwd_set(fdp, newpwd);
3485 	FILEDESC_XUNLOCK(fdp);
3486 	pwd_drop(oldpwd);
3487 }
3488 
3489 void
3490 pwd_set_rootvnode(void)
3491 {
3492 	struct filedesc *fdp;
3493 	struct pwd *oldpwd, *newpwd;
3494 
3495 	fdp = curproc->p_fd;
3496 
3497 	newpwd = pwd_alloc();
3498 	FILEDESC_XLOCK(fdp);
3499 	oldpwd = FILEDESC_XLOCKED_LOAD_PWD(fdp);
3500 	vrefact(rootvnode);
3501 	newpwd->pwd_cdir = rootvnode;
3502 	vrefact(rootvnode);
3503 	newpwd->pwd_rdir = rootvnode;
3504 	pwd_fill(oldpwd, newpwd);
3505 	pwd_set(fdp, newpwd);
3506 	FILEDESC_XUNLOCK(fdp);
3507 	pwd_drop(oldpwd);
3508 }
3509 
3510 /*
3511  * Scan all active processes and prisons to see if any of them have a current
3512  * or root directory of `olddp'. If so, replace them with the new mount point.
3513  */
3514 void
3515 mountcheckdirs(struct vnode *olddp, struct vnode *newdp)
3516 {
3517 	struct filedesc *fdp;
3518 	struct pwd *newpwd, *oldpwd;
3519 	struct prison *pr;
3520 	struct proc *p;
3521 	int nrele;
3522 
3523 	if (vrefcnt(olddp) == 1)
3524 		return;
3525 	nrele = 0;
3526 	newpwd = pwd_alloc();
3527 	sx_slock(&allproc_lock);
3528 	FOREACH_PROC_IN_SYSTEM(p) {
3529 		PROC_LOCK(p);
3530 		fdp = fdhold(p);
3531 		PROC_UNLOCK(p);
3532 		if (fdp == NULL)
3533 			continue;
3534 		FILEDESC_XLOCK(fdp);
3535 		oldpwd = FILEDESC_XLOCKED_LOAD_PWD(fdp);
3536 		if (oldpwd == NULL ||
3537 		    (oldpwd->pwd_cdir != olddp &&
3538 		    oldpwd->pwd_rdir != olddp &&
3539 		    oldpwd->pwd_jdir != olddp)) {
3540 			FILEDESC_XUNLOCK(fdp);
3541 			fddrop(fdp);
3542 			continue;
3543 		}
3544 		if (oldpwd->pwd_cdir == olddp) {
3545 			vrefact(newdp);
3546 			newpwd->pwd_cdir = newdp;
3547 		}
3548 		if (oldpwd->pwd_rdir == olddp) {
3549 			vrefact(newdp);
3550 			newpwd->pwd_rdir = newdp;
3551 		}
3552 		if (oldpwd->pwd_jdir == olddp) {
3553 			vrefact(newdp);
3554 			newpwd->pwd_jdir = newdp;
3555 		}
3556 		pwd_fill(oldpwd, newpwd);
3557 		pwd_set(fdp, newpwd);
3558 		FILEDESC_XUNLOCK(fdp);
3559 		pwd_drop(oldpwd);
3560 		fddrop(fdp);
3561 		newpwd = pwd_alloc();
3562 	}
3563 	sx_sunlock(&allproc_lock);
3564 	pwd_drop(newpwd);
3565 	if (rootvnode == olddp) {
3566 		vrefact(newdp);
3567 		rootvnode = newdp;
3568 		nrele++;
3569 	}
3570 	mtx_lock(&prison0.pr_mtx);
3571 	if (prison0.pr_root == olddp) {
3572 		vrefact(newdp);
3573 		prison0.pr_root = newdp;
3574 		nrele++;
3575 	}
3576 	mtx_unlock(&prison0.pr_mtx);
3577 	sx_slock(&allprison_lock);
3578 	TAILQ_FOREACH(pr, &allprison, pr_list) {
3579 		mtx_lock(&pr->pr_mtx);
3580 		if (pr->pr_root == olddp) {
3581 			vrefact(newdp);
3582 			pr->pr_root = newdp;
3583 			nrele++;
3584 		}
3585 		mtx_unlock(&pr->pr_mtx);
3586 	}
3587 	sx_sunlock(&allprison_lock);
3588 	while (nrele--)
3589 		vrele(olddp);
3590 }
3591 
3592 struct filedesc_to_leader *
3593 filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader)
3594 {
3595 	struct filedesc_to_leader *fdtol;
3596 
3597 	fdtol = malloc(sizeof(struct filedesc_to_leader),
3598 	    M_FILEDESC_TO_LEADER, M_WAITOK);
3599 	fdtol->fdl_refcount = 1;
3600 	fdtol->fdl_holdcount = 0;
3601 	fdtol->fdl_wakeup = 0;
3602 	fdtol->fdl_leader = leader;
3603 	if (old != NULL) {
3604 		FILEDESC_XLOCK(fdp);
3605 		fdtol->fdl_next = old->fdl_next;
3606 		fdtol->fdl_prev = old;
3607 		old->fdl_next = fdtol;
3608 		fdtol->fdl_next->fdl_prev = fdtol;
3609 		FILEDESC_XUNLOCK(fdp);
3610 	} else {
3611 		fdtol->fdl_next = fdtol;
3612 		fdtol->fdl_prev = fdtol;
3613 	}
3614 	return (fdtol);
3615 }
3616 
3617 static int
3618 sysctl_kern_proc_nfds(SYSCTL_HANDLER_ARGS)
3619 {
3620 	NDSLOTTYPE *map;
3621 	struct filedesc *fdp;
3622 	int count, off, minoff;
3623 
3624 	if (*(int *)arg1 != 0)
3625 		return (EINVAL);
3626 
3627 	fdp = curproc->p_fd;
3628 	count = 0;
3629 	FILEDESC_SLOCK(fdp);
3630 	map = fdp->fd_map;
3631 	off = NDSLOT(fdp->fd_nfiles - 1);
3632 	for (minoff = NDSLOT(0); off >= minoff; --off)
3633 		count += bitcountl(map[off]);
3634 	FILEDESC_SUNLOCK(fdp);
3635 
3636 	return (SYSCTL_OUT(req, &count, sizeof(count)));
3637 }
3638 
3639 static SYSCTL_NODE(_kern_proc, KERN_PROC_NFDS, nfds,
3640     CTLFLAG_RD|CTLFLAG_CAPRD|CTLFLAG_MPSAFE, sysctl_kern_proc_nfds,
3641     "Number of open file descriptors");
3642 
3643 /*
3644  * Get file structures globally.
3645  */
3646 static int
3647 sysctl_kern_file(SYSCTL_HANDLER_ARGS)
3648 {
3649 	struct xfile xf;
3650 	struct filedesc *fdp;
3651 	struct file *fp;
3652 	struct proc *p;
3653 	int error, n, lastfile;
3654 
3655 	error = sysctl_wire_old_buffer(req, 0);
3656 	if (error != 0)
3657 		return (error);
3658 	if (req->oldptr == NULL) {
3659 		n = 0;
3660 		sx_slock(&allproc_lock);
3661 		FOREACH_PROC_IN_SYSTEM(p) {
3662 			PROC_LOCK(p);
3663 			if (p->p_state == PRS_NEW) {
3664 				PROC_UNLOCK(p);
3665 				continue;
3666 			}
3667 			fdp = fdhold(p);
3668 			PROC_UNLOCK(p);
3669 			if (fdp == NULL)
3670 				continue;
3671 			/* overestimates sparse tables. */
3672 			n += fdp->fd_nfiles;
3673 			fddrop(fdp);
3674 		}
3675 		sx_sunlock(&allproc_lock);
3676 		return (SYSCTL_OUT(req, 0, n * sizeof(xf)));
3677 	}
3678 	error = 0;
3679 	bzero(&xf, sizeof(xf));
3680 	xf.xf_size = sizeof(xf);
3681 	sx_slock(&allproc_lock);
3682 	FOREACH_PROC_IN_SYSTEM(p) {
3683 		PROC_LOCK(p);
3684 		if (p->p_state == PRS_NEW) {
3685 			PROC_UNLOCK(p);
3686 			continue;
3687 		}
3688 		if (p_cansee(req->td, p) != 0) {
3689 			PROC_UNLOCK(p);
3690 			continue;
3691 		}
3692 		xf.xf_pid = p->p_pid;
3693 		xf.xf_uid = p->p_ucred->cr_uid;
3694 		fdp = fdhold(p);
3695 		PROC_UNLOCK(p);
3696 		if (fdp == NULL)
3697 			continue;
3698 		FILEDESC_SLOCK(fdp);
3699 		lastfile = fdlastfile(fdp);
3700 		for (n = 0; fdp->fd_refcnt > 0 && n <= lastfile; ++n) {
3701 			if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
3702 				continue;
3703 			xf.xf_fd = n;
3704 			xf.xf_file = (uintptr_t)fp;
3705 			xf.xf_data = (uintptr_t)fp->f_data;
3706 			xf.xf_vnode = (uintptr_t)fp->f_vnode;
3707 			xf.xf_type = (uintptr_t)fp->f_type;
3708 			xf.xf_count = fp->f_count;
3709 			xf.xf_msgcount = 0;
3710 			xf.xf_offset = foffset_get(fp);
3711 			xf.xf_flag = fp->f_flag;
3712 			error = SYSCTL_OUT(req, &xf, sizeof(xf));
3713 			if (error)
3714 				break;
3715 		}
3716 		FILEDESC_SUNLOCK(fdp);
3717 		fddrop(fdp);
3718 		if (error)
3719 			break;
3720 	}
3721 	sx_sunlock(&allproc_lock);
3722 	return (error);
3723 }
3724 
3725 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD|CTLFLAG_MPSAFE,
3726     0, 0, sysctl_kern_file, "S,xfile", "Entire file table");
3727 
3728 #ifdef KINFO_FILE_SIZE
3729 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
3730 #endif
3731 
3732 static int
3733 xlate_fflags(int fflags)
3734 {
3735 	static const struct {
3736 		int	fflag;
3737 		int	kf_fflag;
3738 	} fflags_table[] = {
3739 		{ FAPPEND, KF_FLAG_APPEND },
3740 		{ FASYNC, KF_FLAG_ASYNC },
3741 		{ FFSYNC, KF_FLAG_FSYNC },
3742 		{ FHASLOCK, KF_FLAG_HASLOCK },
3743 		{ FNONBLOCK, KF_FLAG_NONBLOCK },
3744 		{ FREAD, KF_FLAG_READ },
3745 		{ FWRITE, KF_FLAG_WRITE },
3746 		{ O_CREAT, KF_FLAG_CREAT },
3747 		{ O_DIRECT, KF_FLAG_DIRECT },
3748 		{ O_EXCL, KF_FLAG_EXCL },
3749 		{ O_EXEC, KF_FLAG_EXEC },
3750 		{ O_EXLOCK, KF_FLAG_EXLOCK },
3751 		{ O_NOFOLLOW, KF_FLAG_NOFOLLOW },
3752 		{ O_SHLOCK, KF_FLAG_SHLOCK },
3753 		{ O_TRUNC, KF_FLAG_TRUNC }
3754 	};
3755 	unsigned int i;
3756 	int kflags;
3757 
3758 	kflags = 0;
3759 	for (i = 0; i < nitems(fflags_table); i++)
3760 		if (fflags & fflags_table[i].fflag)
3761 			kflags |=  fflags_table[i].kf_fflag;
3762 	return (kflags);
3763 }
3764 
3765 /* Trim unused data from kf_path by truncating the structure size. */
3766 void
3767 pack_kinfo(struct kinfo_file *kif)
3768 {
3769 
3770 	kif->kf_structsize = offsetof(struct kinfo_file, kf_path) +
3771 	    strlen(kif->kf_path) + 1;
3772 	kif->kf_structsize = roundup(kif->kf_structsize, sizeof(uint64_t));
3773 }
3774 
3775 static void
3776 export_file_to_kinfo(struct file *fp, int fd, cap_rights_t *rightsp,
3777     struct kinfo_file *kif, struct filedesc *fdp, int flags)
3778 {
3779 	int error;
3780 
3781 	bzero(kif, sizeof(*kif));
3782 
3783 	/* Set a default type to allow for empty fill_kinfo() methods. */
3784 	kif->kf_type = KF_TYPE_UNKNOWN;
3785 	kif->kf_flags = xlate_fflags(fp->f_flag);
3786 	if (rightsp != NULL)
3787 		kif->kf_cap_rights = *rightsp;
3788 	else
3789 		cap_rights_init_zero(&kif->kf_cap_rights);
3790 	kif->kf_fd = fd;
3791 	kif->kf_ref_count = fp->f_count;
3792 	kif->kf_offset = foffset_get(fp);
3793 
3794 	/*
3795 	 * This may drop the filedesc lock, so the 'fp' cannot be
3796 	 * accessed after this call.
3797 	 */
3798 	error = fo_fill_kinfo(fp, kif, fdp);
3799 	if (error == 0)
3800 		kif->kf_status |= KF_ATTR_VALID;
3801 	if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
3802 		pack_kinfo(kif);
3803 	else
3804 		kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
3805 }
3806 
3807 static void
3808 export_vnode_to_kinfo(struct vnode *vp, int fd, int fflags,
3809     struct kinfo_file *kif, int flags)
3810 {
3811 	int error;
3812 
3813 	bzero(kif, sizeof(*kif));
3814 
3815 	kif->kf_type = KF_TYPE_VNODE;
3816 	error = vn_fill_kinfo_vnode(vp, kif);
3817 	if (error == 0)
3818 		kif->kf_status |= KF_ATTR_VALID;
3819 	kif->kf_flags = xlate_fflags(fflags);
3820 	cap_rights_init_zero(&kif->kf_cap_rights);
3821 	kif->kf_fd = fd;
3822 	kif->kf_ref_count = -1;
3823 	kif->kf_offset = -1;
3824 	if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
3825 		pack_kinfo(kif);
3826 	else
3827 		kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
3828 	vrele(vp);
3829 }
3830 
3831 struct export_fd_buf {
3832 	struct filedesc		*fdp;
3833 	struct sbuf 		*sb;
3834 	ssize_t			remainder;
3835 	struct kinfo_file	kif;
3836 	int			flags;
3837 };
3838 
3839 static int
3840 export_kinfo_to_sb(struct export_fd_buf *efbuf)
3841 {
3842 	struct kinfo_file *kif;
3843 
3844 	kif = &efbuf->kif;
3845 	if (efbuf->remainder != -1) {
3846 		if (efbuf->remainder < kif->kf_structsize) {
3847 			/* Terminate export. */
3848 			efbuf->remainder = 0;
3849 			return (0);
3850 		}
3851 		efbuf->remainder -= kif->kf_structsize;
3852 	}
3853 	return (sbuf_bcat(efbuf->sb, kif, kif->kf_structsize) == 0 ? 0 : ENOMEM);
3854 }
3855 
3856 static int
3857 export_file_to_sb(struct file *fp, int fd, cap_rights_t *rightsp,
3858     struct export_fd_buf *efbuf)
3859 {
3860 	int error;
3861 
3862 	if (efbuf->remainder == 0)
3863 		return (0);
3864 	export_file_to_kinfo(fp, fd, rightsp, &efbuf->kif, efbuf->fdp,
3865 	    efbuf->flags);
3866 	FILEDESC_SUNLOCK(efbuf->fdp);
3867 	error = export_kinfo_to_sb(efbuf);
3868 	FILEDESC_SLOCK(efbuf->fdp);
3869 	return (error);
3870 }
3871 
3872 static int
3873 export_vnode_to_sb(struct vnode *vp, int fd, int fflags,
3874     struct export_fd_buf *efbuf)
3875 {
3876 	int error;
3877 
3878 	if (efbuf->remainder == 0)
3879 		return (0);
3880 	if (efbuf->fdp != NULL)
3881 		FILEDESC_SUNLOCK(efbuf->fdp);
3882 	export_vnode_to_kinfo(vp, fd, fflags, &efbuf->kif, efbuf->flags);
3883 	error = export_kinfo_to_sb(efbuf);
3884 	if (efbuf->fdp != NULL)
3885 		FILEDESC_SLOCK(efbuf->fdp);
3886 	return (error);
3887 }
3888 
3889 /*
3890  * Store a process file descriptor information to sbuf.
3891  *
3892  * Takes a locked proc as argument, and returns with the proc unlocked.
3893  */
3894 int
3895 kern_proc_filedesc_out(struct proc *p,  struct sbuf *sb, ssize_t maxlen,
3896     int flags)
3897 {
3898 	struct file *fp;
3899 	struct filedesc *fdp;
3900 	struct export_fd_buf *efbuf;
3901 	struct vnode *cttyvp, *textvp, *tracevp;
3902 	struct pwd *pwd;
3903 	int error, i, lastfile;
3904 	cap_rights_t rights;
3905 
3906 	PROC_LOCK_ASSERT(p, MA_OWNED);
3907 
3908 	/* ktrace vnode */
3909 	tracevp = p->p_tracevp;
3910 	if (tracevp != NULL)
3911 		vrefact(tracevp);
3912 	/* text vnode */
3913 	textvp = p->p_textvp;
3914 	if (textvp != NULL)
3915 		vrefact(textvp);
3916 	/* Controlling tty. */
3917 	cttyvp = NULL;
3918 	if (p->p_pgrp != NULL && p->p_pgrp->pg_session != NULL) {
3919 		cttyvp = p->p_pgrp->pg_session->s_ttyvp;
3920 		if (cttyvp != NULL)
3921 			vrefact(cttyvp);
3922 	}
3923 	fdp = fdhold(p);
3924 	PROC_UNLOCK(p);
3925 	efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
3926 	efbuf->fdp = NULL;
3927 	efbuf->sb = sb;
3928 	efbuf->remainder = maxlen;
3929 	efbuf->flags = flags;
3930 	if (tracevp != NULL)
3931 		export_vnode_to_sb(tracevp, KF_FD_TYPE_TRACE, FREAD | FWRITE,
3932 		    efbuf);
3933 	if (textvp != NULL)
3934 		export_vnode_to_sb(textvp, KF_FD_TYPE_TEXT, FREAD, efbuf);
3935 	if (cttyvp != NULL)
3936 		export_vnode_to_sb(cttyvp, KF_FD_TYPE_CTTY, FREAD | FWRITE,
3937 		    efbuf);
3938 	error = 0;
3939 	if (fdp == NULL)
3940 		goto fail;
3941 	efbuf->fdp = fdp;
3942 	FILEDESC_SLOCK(fdp);
3943 	pwd = pwd_hold_filedesc(fdp);
3944 	if (pwd != NULL) {
3945 		/* working directory */
3946 		if (pwd->pwd_cdir != NULL) {
3947 			vrefact(pwd->pwd_cdir);
3948 			export_vnode_to_sb(pwd->pwd_cdir, KF_FD_TYPE_CWD, FREAD, efbuf);
3949 		}
3950 		/* root directory */
3951 		if (pwd->pwd_rdir != NULL) {
3952 			vrefact(pwd->pwd_rdir);
3953 			export_vnode_to_sb(pwd->pwd_rdir, KF_FD_TYPE_ROOT, FREAD, efbuf);
3954 		}
3955 		/* jail directory */
3956 		if (pwd->pwd_jdir != NULL) {
3957 			vrefact(pwd->pwd_jdir);
3958 			export_vnode_to_sb(pwd->pwd_jdir, KF_FD_TYPE_JAIL, FREAD, efbuf);
3959 		}
3960 	}
3961 	lastfile = fdlastfile(fdp);
3962 	for (i = 0; fdp->fd_refcnt > 0 && i <= lastfile; i++) {
3963 		if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
3964 			continue;
3965 #ifdef CAPABILITIES
3966 		rights = *cap_rights(fdp, i);
3967 #else /* !CAPABILITIES */
3968 		rights = cap_no_rights;
3969 #endif
3970 		/*
3971 		 * Create sysctl entry.  It is OK to drop the filedesc
3972 		 * lock inside of export_file_to_sb() as we will
3973 		 * re-validate and re-evaluate its properties when the
3974 		 * loop continues.
3975 		 */
3976 		error = export_file_to_sb(fp, i, &rights, efbuf);
3977 		if (error != 0 || efbuf->remainder == 0)
3978 			break;
3979 	}
3980 	FILEDESC_SUNLOCK(fdp);
3981 	if (pwd != NULL)
3982 		pwd_drop(pwd);
3983 	fddrop(fdp);
3984 fail:
3985 	free(efbuf, M_TEMP);
3986 	return (error);
3987 }
3988 
3989 #define FILEDESC_SBUF_SIZE	(sizeof(struct kinfo_file) * 5)
3990 
3991 /*
3992  * Get per-process file descriptors for use by procstat(1), et al.
3993  */
3994 static int
3995 sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS)
3996 {
3997 	struct sbuf sb;
3998 	struct proc *p;
3999 	ssize_t maxlen;
4000 	int error, error2, *name;
4001 
4002 	name = (int *)arg1;
4003 
4004 	sbuf_new_for_sysctl(&sb, NULL, FILEDESC_SBUF_SIZE, req);
4005 	sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
4006 	error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
4007 	if (error != 0) {
4008 		sbuf_delete(&sb);
4009 		return (error);
4010 	}
4011 	maxlen = req->oldptr != NULL ? req->oldlen : -1;
4012 	error = kern_proc_filedesc_out(p, &sb, maxlen,
4013 	    KERN_FILEDESC_PACK_KINFO);
4014 	error2 = sbuf_finish(&sb);
4015 	sbuf_delete(&sb);
4016 	return (error != 0 ? error : error2);
4017 }
4018 
4019 #ifdef COMPAT_FREEBSD7
4020 #ifdef KINFO_OFILE_SIZE
4021 CTASSERT(sizeof(struct kinfo_ofile) == KINFO_OFILE_SIZE);
4022 #endif
4023 
4024 static void
4025 kinfo_to_okinfo(struct kinfo_file *kif, struct kinfo_ofile *okif)
4026 {
4027 
4028 	okif->kf_structsize = sizeof(*okif);
4029 	okif->kf_type = kif->kf_type;
4030 	okif->kf_fd = kif->kf_fd;
4031 	okif->kf_ref_count = kif->kf_ref_count;
4032 	okif->kf_flags = kif->kf_flags & (KF_FLAG_READ | KF_FLAG_WRITE |
4033 	    KF_FLAG_APPEND | KF_FLAG_ASYNC | KF_FLAG_FSYNC | KF_FLAG_NONBLOCK |
4034 	    KF_FLAG_DIRECT | KF_FLAG_HASLOCK);
4035 	okif->kf_offset = kif->kf_offset;
4036 	if (kif->kf_type == KF_TYPE_VNODE)
4037 		okif->kf_vnode_type = kif->kf_un.kf_file.kf_file_type;
4038 	else
4039 		okif->kf_vnode_type = KF_VTYPE_VNON;
4040 	strlcpy(okif->kf_path, kif->kf_path, sizeof(okif->kf_path));
4041 	if (kif->kf_type == KF_TYPE_SOCKET) {
4042 		okif->kf_sock_domain = kif->kf_un.kf_sock.kf_sock_domain0;
4043 		okif->kf_sock_type = kif->kf_un.kf_sock.kf_sock_type0;
4044 		okif->kf_sock_protocol = kif->kf_un.kf_sock.kf_sock_protocol0;
4045 		okif->kf_sa_local = kif->kf_un.kf_sock.kf_sa_local;
4046 		okif->kf_sa_peer = kif->kf_un.kf_sock.kf_sa_peer;
4047 	} else {
4048 		okif->kf_sa_local.ss_family = AF_UNSPEC;
4049 		okif->kf_sa_peer.ss_family = AF_UNSPEC;
4050 	}
4051 }
4052 
4053 static int
4054 export_vnode_for_osysctl(struct vnode *vp, int type, struct kinfo_file *kif,
4055     struct kinfo_ofile *okif, struct filedesc *fdp, struct sysctl_req *req)
4056 {
4057 	int error;
4058 
4059 	vrefact(vp);
4060 	FILEDESC_SUNLOCK(fdp);
4061 	export_vnode_to_kinfo(vp, type, 0, kif, KERN_FILEDESC_PACK_KINFO);
4062 	kinfo_to_okinfo(kif, okif);
4063 	error = SYSCTL_OUT(req, okif, sizeof(*okif));
4064 	FILEDESC_SLOCK(fdp);
4065 	return (error);
4066 }
4067 
4068 /*
4069  * Get per-process file descriptors for use by procstat(1), et al.
4070  */
4071 static int
4072 sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS)
4073 {
4074 	struct kinfo_ofile *okif;
4075 	struct kinfo_file *kif;
4076 	struct filedesc *fdp;
4077 	struct pwd *pwd;
4078 	int error, i, lastfile, *name;
4079 	struct file *fp;
4080 	struct proc *p;
4081 
4082 	name = (int *)arg1;
4083 	error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
4084 	if (error != 0)
4085 		return (error);
4086 	fdp = fdhold(p);
4087 	PROC_UNLOCK(p);
4088 	if (fdp == NULL)
4089 		return (ENOENT);
4090 	kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK);
4091 	okif = malloc(sizeof(*okif), M_TEMP, M_WAITOK);
4092 	FILEDESC_SLOCK(fdp);
4093 	pwd = pwd_hold_filedesc(fdp);
4094 	if (pwd != NULL) {
4095 		if (pwd->pwd_cdir != NULL)
4096 			export_vnode_for_osysctl(pwd->pwd_cdir, KF_FD_TYPE_CWD, kif,
4097 			    okif, fdp, req);
4098 		if (pwd->pwd_rdir != NULL)
4099 			export_vnode_for_osysctl(pwd->pwd_rdir, KF_FD_TYPE_ROOT, kif,
4100 			    okif, fdp, req);
4101 		if (pwd->pwd_jdir != NULL)
4102 			export_vnode_for_osysctl(pwd->pwd_jdir, KF_FD_TYPE_JAIL, kif,
4103 			    okif, fdp, req);
4104 	}
4105 	lastfile = fdlastfile(fdp);
4106 	for (i = 0; fdp->fd_refcnt > 0 && i <= lastfile; i++) {
4107 		if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
4108 			continue;
4109 		export_file_to_kinfo(fp, i, NULL, kif, fdp,
4110 		    KERN_FILEDESC_PACK_KINFO);
4111 		FILEDESC_SUNLOCK(fdp);
4112 		kinfo_to_okinfo(kif, okif);
4113 		error = SYSCTL_OUT(req, okif, sizeof(*okif));
4114 		FILEDESC_SLOCK(fdp);
4115 		if (error)
4116 			break;
4117 	}
4118 	FILEDESC_SUNLOCK(fdp);
4119 	if (pwd != NULL)
4120 		pwd_drop(pwd);
4121 	fddrop(fdp);
4122 	free(kif, M_TEMP);
4123 	free(okif, M_TEMP);
4124 	return (0);
4125 }
4126 
4127 static SYSCTL_NODE(_kern_proc, KERN_PROC_OFILEDESC, ofiledesc,
4128     CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_ofiledesc,
4129     "Process ofiledesc entries");
4130 #endif	/* COMPAT_FREEBSD7 */
4131 
4132 int
4133 vntype_to_kinfo(int vtype)
4134 {
4135 	struct {
4136 		int	vtype;
4137 		int	kf_vtype;
4138 	} vtypes_table[] = {
4139 		{ VBAD, KF_VTYPE_VBAD },
4140 		{ VBLK, KF_VTYPE_VBLK },
4141 		{ VCHR, KF_VTYPE_VCHR },
4142 		{ VDIR, KF_VTYPE_VDIR },
4143 		{ VFIFO, KF_VTYPE_VFIFO },
4144 		{ VLNK, KF_VTYPE_VLNK },
4145 		{ VNON, KF_VTYPE_VNON },
4146 		{ VREG, KF_VTYPE_VREG },
4147 		{ VSOCK, KF_VTYPE_VSOCK }
4148 	};
4149 	unsigned int i;
4150 
4151 	/*
4152 	 * Perform vtype translation.
4153 	 */
4154 	for (i = 0; i < nitems(vtypes_table); i++)
4155 		if (vtypes_table[i].vtype == vtype)
4156 			return (vtypes_table[i].kf_vtype);
4157 
4158 	return (KF_VTYPE_UNKNOWN);
4159 }
4160 
4161 static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc,
4162     CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_filedesc,
4163     "Process filedesc entries");
4164 
4165 /*
4166  * Store a process current working directory information to sbuf.
4167  *
4168  * Takes a locked proc as argument, and returns with the proc unlocked.
4169  */
4170 int
4171 kern_proc_cwd_out(struct proc *p,  struct sbuf *sb, ssize_t maxlen)
4172 {
4173 	struct filedesc *fdp;
4174 	struct pwd *pwd;
4175 	struct export_fd_buf *efbuf;
4176 	struct vnode *cdir;
4177 	int error;
4178 
4179 	PROC_LOCK_ASSERT(p, MA_OWNED);
4180 
4181 	fdp = fdhold(p);
4182 	PROC_UNLOCK(p);
4183 	if (fdp == NULL)
4184 		return (EINVAL);
4185 
4186 	efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
4187 	efbuf->fdp = fdp;
4188 	efbuf->sb = sb;
4189 	efbuf->remainder = maxlen;
4190 
4191 	FILEDESC_SLOCK(fdp);
4192 	pwd = FILEDESC_LOCKED_LOAD_PWD(fdp);
4193 	cdir = pwd->pwd_cdir;
4194 	if (cdir == NULL) {
4195 		error = EINVAL;
4196 	} else {
4197 		vrefact(cdir);
4198 		error = export_vnode_to_sb(cdir, KF_FD_TYPE_CWD, FREAD, efbuf);
4199 	}
4200 	FILEDESC_SUNLOCK(fdp);
4201 	fddrop(fdp);
4202 	free(efbuf, M_TEMP);
4203 	return (error);
4204 }
4205 
4206 /*
4207  * Get per-process current working directory.
4208  */
4209 static int
4210 sysctl_kern_proc_cwd(SYSCTL_HANDLER_ARGS)
4211 {
4212 	struct sbuf sb;
4213 	struct proc *p;
4214 	ssize_t maxlen;
4215 	int error, error2, *name;
4216 
4217 	name = (int *)arg1;
4218 
4219 	sbuf_new_for_sysctl(&sb, NULL, sizeof(struct kinfo_file), req);
4220 	sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
4221 	error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
4222 	if (error != 0) {
4223 		sbuf_delete(&sb);
4224 		return (error);
4225 	}
4226 	maxlen = req->oldptr != NULL ? req->oldlen : -1;
4227 	error = kern_proc_cwd_out(p, &sb, maxlen);
4228 	error2 = sbuf_finish(&sb);
4229 	sbuf_delete(&sb);
4230 	return (error != 0 ? error : error2);
4231 }
4232 
4233 static SYSCTL_NODE(_kern_proc, KERN_PROC_CWD, cwd, CTLFLAG_RD|CTLFLAG_MPSAFE,
4234     sysctl_kern_proc_cwd, "Process current working directory");
4235 
4236 #ifdef DDB
4237 /*
4238  * For the purposes of debugging, generate a human-readable string for the
4239  * file type.
4240  */
4241 static const char *
4242 file_type_to_name(short type)
4243 {
4244 
4245 	switch (type) {
4246 	case 0:
4247 		return ("zero");
4248 	case DTYPE_VNODE:
4249 		return ("vnode");
4250 	case DTYPE_SOCKET:
4251 		return ("socket");
4252 	case DTYPE_PIPE:
4253 		return ("pipe");
4254 	case DTYPE_FIFO:
4255 		return ("fifo");
4256 	case DTYPE_KQUEUE:
4257 		return ("kqueue");
4258 	case DTYPE_CRYPTO:
4259 		return ("crypto");
4260 	case DTYPE_MQUEUE:
4261 		return ("mqueue");
4262 	case DTYPE_SHM:
4263 		return ("shm");
4264 	case DTYPE_SEM:
4265 		return ("ksem");
4266 	case DTYPE_PTS:
4267 		return ("pts");
4268 	case DTYPE_DEV:
4269 		return ("dev");
4270 	case DTYPE_PROCDESC:
4271 		return ("proc");
4272 	case DTYPE_LINUXEFD:
4273 		return ("levent");
4274 	case DTYPE_LINUXTFD:
4275 		return ("ltimer");
4276 	default:
4277 		return ("unkn");
4278 	}
4279 }
4280 
4281 /*
4282  * For the purposes of debugging, identify a process (if any, perhaps one of
4283  * many) that references the passed file in its file descriptor array. Return
4284  * NULL if none.
4285  */
4286 static struct proc *
4287 file_to_first_proc(struct file *fp)
4288 {
4289 	struct filedesc *fdp;
4290 	struct proc *p;
4291 	int n;
4292 
4293 	FOREACH_PROC_IN_SYSTEM(p) {
4294 		if (p->p_state == PRS_NEW)
4295 			continue;
4296 		fdp = p->p_fd;
4297 		if (fdp == NULL)
4298 			continue;
4299 		for (n = 0; n < fdp->fd_nfiles; n++) {
4300 			if (fp == fdp->fd_ofiles[n].fde_file)
4301 				return (p);
4302 		}
4303 	}
4304 	return (NULL);
4305 }
4306 
4307 static void
4308 db_print_file(struct file *fp, int header)
4309 {
4310 #define XPTRWIDTH ((int)howmany(sizeof(void *) * NBBY, 4))
4311 	struct proc *p;
4312 
4313 	if (header)
4314 		db_printf("%*s %6s %*s %8s %4s %5s %6s %*s %5s %s\n",
4315 		    XPTRWIDTH, "File", "Type", XPTRWIDTH, "Data", "Flag",
4316 		    "GCFl", "Count", "MCount", XPTRWIDTH, "Vnode", "FPID",
4317 		    "FCmd");
4318 	p = file_to_first_proc(fp);
4319 	db_printf("%*p %6s %*p %08x %04x %5d %6d %*p %5d %s\n", XPTRWIDTH,
4320 	    fp, file_type_to_name(fp->f_type), XPTRWIDTH, fp->f_data,
4321 	    fp->f_flag, 0, fp->f_count, 0, XPTRWIDTH, fp->f_vnode,
4322 	    p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-");
4323 
4324 #undef XPTRWIDTH
4325 }
4326 
4327 DB_SHOW_COMMAND(file, db_show_file)
4328 {
4329 	struct file *fp;
4330 
4331 	if (!have_addr) {
4332 		db_printf("usage: show file <addr>\n");
4333 		return;
4334 	}
4335 	fp = (struct file *)addr;
4336 	db_print_file(fp, 1);
4337 }
4338 
4339 DB_SHOW_COMMAND(files, db_show_files)
4340 {
4341 	struct filedesc *fdp;
4342 	struct file *fp;
4343 	struct proc *p;
4344 	int header;
4345 	int n;
4346 
4347 	header = 1;
4348 	FOREACH_PROC_IN_SYSTEM(p) {
4349 		if (p->p_state == PRS_NEW)
4350 			continue;
4351 		if ((fdp = p->p_fd) == NULL)
4352 			continue;
4353 		for (n = 0; n < fdp->fd_nfiles; ++n) {
4354 			if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
4355 				continue;
4356 			db_print_file(fp, header);
4357 			header = 0;
4358 		}
4359 	}
4360 }
4361 #endif
4362 
4363 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
4364     &maxfilesperproc, 0, "Maximum files allowed open per process");
4365 
4366 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
4367     &maxfiles, 0, "Maximum number of files");
4368 
4369 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
4370     &openfiles, 0, "System-wide number of open files");
4371 
4372 /* ARGSUSED*/
4373 static void
4374 filelistinit(void *dummy)
4375 {
4376 
4377 	file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL,
4378 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
4379 	filedesc0_zone = uma_zcreate("filedesc0", sizeof(struct filedesc0),
4380 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
4381 	pwd_zone = uma_zcreate("PWD", sizeof(struct pwd), NULL, NULL,
4382 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_SMR);
4383 	/*
4384 	 * XXXMJG this is a temporary hack due to boot ordering issues against
4385 	 * the vnode zone.
4386 	 */
4387 	vfs_smr = uma_zone_get_smr(pwd_zone);
4388 	mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF);
4389 }
4390 SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL);
4391 
4392 /*-------------------------------------------------------------------*/
4393 
4394 static int
4395 badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred,
4396     int flags, struct thread *td)
4397 {
4398 
4399 	return (EBADF);
4400 }
4401 
4402 static int
4403 badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
4404     struct thread *td)
4405 {
4406 
4407 	return (EINVAL);
4408 }
4409 
4410 static int
4411 badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred,
4412     struct thread *td)
4413 {
4414 
4415 	return (EBADF);
4416 }
4417 
4418 static int
4419 badfo_poll(struct file *fp, int events, struct ucred *active_cred,
4420     struct thread *td)
4421 {
4422 
4423 	return (0);
4424 }
4425 
4426 static int
4427 badfo_kqfilter(struct file *fp, struct knote *kn)
4428 {
4429 
4430 	return (EBADF);
4431 }
4432 
4433 static int
4434 badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred,
4435     struct thread *td)
4436 {
4437 
4438 	return (EBADF);
4439 }
4440 
4441 static int
4442 badfo_close(struct file *fp, struct thread *td)
4443 {
4444 
4445 	return (0);
4446 }
4447 
4448 static int
4449 badfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
4450     struct thread *td)
4451 {
4452 
4453 	return (EBADF);
4454 }
4455 
4456 static int
4457 badfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
4458     struct thread *td)
4459 {
4460 
4461 	return (EBADF);
4462 }
4463 
4464 static int
4465 badfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
4466     struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
4467     struct thread *td)
4468 {
4469 
4470 	return (EBADF);
4471 }
4472 
4473 static int
4474 badfo_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
4475 {
4476 
4477 	return (0);
4478 }
4479 
4480 struct fileops badfileops = {
4481 	.fo_read = badfo_readwrite,
4482 	.fo_write = badfo_readwrite,
4483 	.fo_truncate = badfo_truncate,
4484 	.fo_ioctl = badfo_ioctl,
4485 	.fo_poll = badfo_poll,
4486 	.fo_kqfilter = badfo_kqfilter,
4487 	.fo_stat = badfo_stat,
4488 	.fo_close = badfo_close,
4489 	.fo_chmod = badfo_chmod,
4490 	.fo_chown = badfo_chown,
4491 	.fo_sendfile = badfo_sendfile,
4492 	.fo_fill_kinfo = badfo_fill_kinfo,
4493 };
4494 
4495 int
4496 invfo_rdwr(struct file *fp, struct uio *uio, struct ucred *active_cred,
4497     int flags, struct thread *td)
4498 {
4499 
4500 	return (EOPNOTSUPP);
4501 }
4502 
4503 int
4504 invfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
4505     struct thread *td)
4506 {
4507 
4508 	return (EINVAL);
4509 }
4510 
4511 int
4512 invfo_ioctl(struct file *fp, u_long com, void *data,
4513     struct ucred *active_cred, struct thread *td)
4514 {
4515 
4516 	return (ENOTTY);
4517 }
4518 
4519 int
4520 invfo_poll(struct file *fp, int events, struct ucred *active_cred,
4521     struct thread *td)
4522 {
4523 
4524 	return (poll_no_poll(events));
4525 }
4526 
4527 int
4528 invfo_kqfilter(struct file *fp, struct knote *kn)
4529 {
4530 
4531 	return (EINVAL);
4532 }
4533 
4534 int
4535 invfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
4536     struct thread *td)
4537 {
4538 
4539 	return (EINVAL);
4540 }
4541 
4542 int
4543 invfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
4544     struct thread *td)
4545 {
4546 
4547 	return (EINVAL);
4548 }
4549 
4550 int
4551 invfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
4552     struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
4553     struct thread *td)
4554 {
4555 
4556 	return (EINVAL);
4557 }
4558 
4559 /*-------------------------------------------------------------------*/
4560 
4561 /*
4562  * File Descriptor pseudo-device driver (/dev/fd/).
4563  *
4564  * Opening minor device N dup()s the file (if any) connected to file
4565  * descriptor N belonging to the calling process.  Note that this driver
4566  * consists of only the ``open()'' routine, because all subsequent
4567  * references to this file will be direct to the other driver.
4568  *
4569  * XXX: we could give this one a cloning event handler if necessary.
4570  */
4571 
4572 /* ARGSUSED */
4573 static int
4574 fdopen(struct cdev *dev, int mode, int type, struct thread *td)
4575 {
4576 
4577 	/*
4578 	 * XXX Kludge: set curthread->td_dupfd to contain the value of the
4579 	 * the file descriptor being sought for duplication. The error
4580 	 * return ensures that the vnode for this device will be released
4581 	 * by vn_open. Open will detect this special error and take the
4582 	 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
4583 	 * will simply report the error.
4584 	 */
4585 	td->td_dupfd = dev2unit(dev);
4586 	return (ENODEV);
4587 }
4588 
4589 static struct cdevsw fildesc_cdevsw = {
4590 	.d_version =	D_VERSION,
4591 	.d_open =	fdopen,
4592 	.d_name =	"FD",
4593 };
4594 
4595 static void
4596 fildesc_drvinit(void *unused)
4597 {
4598 	struct cdev *dev;
4599 
4600 	dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 0, NULL,
4601 	    UID_ROOT, GID_WHEEL, 0666, "fd/0");
4602 	make_dev_alias(dev, "stdin");
4603 	dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 1, NULL,
4604 	    UID_ROOT, GID_WHEEL, 0666, "fd/1");
4605 	make_dev_alias(dev, "stdout");
4606 	dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 2, NULL,
4607 	    UID_ROOT, GID_WHEEL, 0666, "fd/2");
4608 	make_dev_alias(dev, "stderr");
4609 }
4610 
4611 SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL);
4612