xref: /freebsd/sys/kern/kern_descrip.c (revision d38c30c092828f4882ce13b08d0bd3fd6dc7afb5)
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/stat.h>
73 #include <sys/sx.h>
74 #include <sys/syscallsubr.h>
75 #include <sys/sysctl.h>
76 #include <sys/sysproto.h>
77 #include <sys/unistd.h>
78 #include <sys/user.h>
79 #include <sys/vnode.h>
80 #ifdef KTRACE
81 #include <sys/ktrace.h>
82 #endif
83 
84 #include <net/vnet.h>
85 
86 #include <security/audit/audit.h>
87 
88 #include <vm/uma.h>
89 #include <vm/vm.h>
90 
91 #include <ddb/ddb.h>
92 
93 static MALLOC_DEFINE(M_FILEDESC, "filedesc", "Open file descriptor table");
94 static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "filedesc_to_leader",
95     "file desc to leader structures");
96 static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures");
97 MALLOC_DEFINE(M_FILECAPS, "filecaps", "descriptor capabilities");
98 
99 MALLOC_DECLARE(M_FADVISE);
100 
101 static __read_mostly uma_zone_t file_zone;
102 static __read_mostly uma_zone_t filedesc0_zone;
103 
104 static int	closefp(struct filedesc *fdp, int fd, struct file *fp,
105 		    struct thread *td, int holdleaders);
106 static int	fd_first_free(struct filedesc *fdp, int low, int size);
107 static int	fd_last_used(struct filedesc *fdp, int size);
108 static void	fdgrowtable(struct filedesc *fdp, int nfd);
109 static void	fdgrowtable_exp(struct filedesc *fdp, int nfd);
110 static void	fdunused(struct filedesc *fdp, int fd);
111 static void	fdused(struct filedesc *fdp, int fd);
112 static int	getmaxfd(struct thread *td);
113 static u_long	*filecaps_copy_prep(const struct filecaps *src);
114 static void	filecaps_copy_finish(const struct filecaps *src,
115 		    struct filecaps *dst, u_long *ioctls);
116 static u_long 	*filecaps_free_prep(struct filecaps *fcaps);
117 static void	filecaps_free_finish(u_long *ioctls);
118 
119 /*
120  * Each process has:
121  *
122  * - An array of open file descriptors (fd_ofiles)
123  * - An array of file flags (fd_ofileflags)
124  * - A bitmap recording which descriptors are in use (fd_map)
125  *
126  * A process starts out with NDFILE descriptors.  The value of NDFILE has
127  * been selected based the historical limit of 20 open files, and an
128  * assumption that the majority of processes, especially short-lived
129  * processes like shells, will never need more.
130  *
131  * If this initial allocation is exhausted, a larger descriptor table and
132  * map are allocated dynamically, and the pointers in the process's struct
133  * filedesc are updated to point to those.  This is repeated every time
134  * the process runs out of file descriptors (provided it hasn't hit its
135  * resource limit).
136  *
137  * Since threads may hold references to individual descriptor table
138  * entries, the tables are never freed.  Instead, they are placed on a
139  * linked list and freed only when the struct filedesc is released.
140  */
141 #define NDFILE		20
142 #define NDSLOTSIZE	sizeof(NDSLOTTYPE)
143 #define	NDENTRIES	(NDSLOTSIZE * __CHAR_BIT)
144 #define NDSLOT(x)	((x) / NDENTRIES)
145 #define NDBIT(x)	((NDSLOTTYPE)1 << ((x) % NDENTRIES))
146 #define	NDSLOTS(x)	(((x) + NDENTRIES - 1) / NDENTRIES)
147 
148 /*
149  * SLIST entry used to keep track of ofiles which must be reclaimed when
150  * the process exits.
151  */
152 struct freetable {
153 	struct fdescenttbl *ft_table;
154 	SLIST_ENTRY(freetable) ft_next;
155 };
156 
157 /*
158  * Initial allocation: a filedesc structure + the head of SLIST used to
159  * keep track of old ofiles + enough space for NDFILE descriptors.
160  */
161 
162 struct fdescenttbl0 {
163 	int	fdt_nfiles;
164 	struct	filedescent fdt_ofiles[NDFILE];
165 };
166 
167 struct filedesc0 {
168 	struct filedesc fd_fd;
169 	SLIST_HEAD(, freetable) fd_free;
170 	struct	fdescenttbl0 fd_dfiles;
171 	NDSLOTTYPE fd_dmap[NDSLOTS(NDFILE)];
172 };
173 
174 /*
175  * Descriptor management.
176  */
177 static int __exclusive_cache_line openfiles; /* actual number of open files */
178 struct mtx sigio_lock;		/* mtx to protect pointers to sigio */
179 void __read_mostly (*mq_fdclose)(struct thread *td, int fd, struct file *fp);
180 
181 /*
182  * If low >= size, just return low. Otherwise find the first zero bit in the
183  * given bitmap, starting at low and not exceeding size - 1. Return size if
184  * not found.
185  */
186 static int
187 fd_first_free(struct filedesc *fdp, int low, int size)
188 {
189 	NDSLOTTYPE *map = fdp->fd_map;
190 	NDSLOTTYPE mask;
191 	int off, maxoff;
192 
193 	if (low >= size)
194 		return (low);
195 
196 	off = NDSLOT(low);
197 	if (low % NDENTRIES) {
198 		mask = ~(~(NDSLOTTYPE)0 >> (NDENTRIES - (low % NDENTRIES)));
199 		if ((mask &= ~map[off]) != 0UL)
200 			return (off * NDENTRIES + ffsl(mask) - 1);
201 		++off;
202 	}
203 	for (maxoff = NDSLOTS(size); off < maxoff; ++off)
204 		if (map[off] != ~0UL)
205 			return (off * NDENTRIES + ffsl(~map[off]) - 1);
206 	return (size);
207 }
208 
209 /*
210  * Find the highest non-zero bit in the given bitmap, starting at 0 and
211  * not exceeding size - 1. Return -1 if not found.
212  */
213 static int
214 fd_last_used(struct filedesc *fdp, int size)
215 {
216 	NDSLOTTYPE *map = fdp->fd_map;
217 	NDSLOTTYPE mask;
218 	int off, minoff;
219 
220 	off = NDSLOT(size);
221 	if (size % NDENTRIES) {
222 		mask = ~(~(NDSLOTTYPE)0 << (size % NDENTRIES));
223 		if ((mask &= map[off]) != 0)
224 			return (off * NDENTRIES + flsl(mask) - 1);
225 		--off;
226 	}
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 static int
234 fdisused(struct filedesc *fdp, int fd)
235 {
236 
237 	KASSERT(fd >= 0 && fd < fdp->fd_nfiles,
238 	    ("file descriptor %d out of range (0, %d)", fd, fdp->fd_nfiles));
239 
240 	return ((fdp->fd_map[NDSLOT(fd)] & NDBIT(fd)) != 0);
241 }
242 
243 /*
244  * Mark a file descriptor as used.
245  */
246 static void
247 fdused_init(struct filedesc *fdp, int fd)
248 {
249 
250 	KASSERT(!fdisused(fdp, fd), ("fd=%d is already used", fd));
251 
252 	fdp->fd_map[NDSLOT(fd)] |= NDBIT(fd);
253 }
254 
255 static void
256 fdused(struct filedesc *fdp, int fd)
257 {
258 
259 	FILEDESC_XLOCK_ASSERT(fdp);
260 
261 	fdused_init(fdp, fd);
262 	if (fd > fdp->fd_lastfile)
263 		fdp->fd_lastfile = fd;
264 	if (fd == fdp->fd_freefile)
265 		fdp->fd_freefile++;
266 }
267 
268 /*
269  * Mark a file descriptor as unused.
270  */
271 static void
272 fdunused(struct filedesc *fdp, int fd)
273 {
274 
275 	FILEDESC_XLOCK_ASSERT(fdp);
276 
277 	KASSERT(fdisused(fdp, fd), ("fd=%d is already unused", fd));
278 	KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
279 	    ("fd=%d is still in use", fd));
280 
281 	fdp->fd_map[NDSLOT(fd)] &= ~NDBIT(fd);
282 	if (fd < fdp->fd_freefile)
283 		fdp->fd_freefile = fd;
284 	if (fd == fdp->fd_lastfile)
285 		fdp->fd_lastfile = fd_last_used(fdp, fd);
286 }
287 
288 /*
289  * Free a file descriptor.
290  *
291  * Avoid some work if fdp is about to be destroyed.
292  */
293 static inline void
294 fdefree_last(struct filedescent *fde)
295 {
296 
297 	filecaps_free(&fde->fde_caps);
298 }
299 
300 static inline void
301 fdfree(struct filedesc *fdp, int fd)
302 {
303 	struct filedescent *fde;
304 
305 	fde = &fdp->fd_ofiles[fd];
306 #ifdef CAPABILITIES
307 	seqc_write_begin(&fde->fde_seqc);
308 #endif
309 	fde->fde_file = NULL;
310 #ifdef CAPABILITIES
311 	seqc_write_end(&fde->fde_seqc);
312 #endif
313 	fdefree_last(fde);
314 	fdunused(fdp, fd);
315 }
316 
317 void
318 pwd_ensure_dirs(void)
319 {
320 	struct filedesc *fdp;
321 
322 	fdp = curproc->p_fd;
323 	FILEDESC_XLOCK(fdp);
324 	if (fdp->fd_cdir == NULL) {
325 		fdp->fd_cdir = rootvnode;
326 		vrefact(rootvnode);
327 	}
328 	if (fdp->fd_rdir == NULL) {
329 		fdp->fd_rdir = rootvnode;
330 		vrefact(rootvnode);
331 	}
332 	FILEDESC_XUNLOCK(fdp);
333 }
334 
335 /*
336  * System calls on descriptors.
337  */
338 #ifndef _SYS_SYSPROTO_H_
339 struct getdtablesize_args {
340 	int	dummy;
341 };
342 #endif
343 /* ARGSUSED */
344 int
345 sys_getdtablesize(struct thread *td, struct getdtablesize_args *uap)
346 {
347 #ifdef	RACCT
348 	uint64_t lim;
349 #endif
350 
351 	td->td_retval[0] = getmaxfd(td);
352 #ifdef	RACCT
353 	PROC_LOCK(td->td_proc);
354 	lim = racct_get_limit(td->td_proc, RACCT_NOFILE);
355 	PROC_UNLOCK(td->td_proc);
356 	if (lim < td->td_retval[0])
357 		td->td_retval[0] = lim;
358 #endif
359 	return (0);
360 }
361 
362 /*
363  * Duplicate a file descriptor to a particular value.
364  *
365  * Note: keep in mind that a potential race condition exists when closing
366  * descriptors from a shared descriptor table (via rfork).
367  */
368 #ifndef _SYS_SYSPROTO_H_
369 struct dup2_args {
370 	u_int	from;
371 	u_int	to;
372 };
373 #endif
374 /* ARGSUSED */
375 int
376 sys_dup2(struct thread *td, struct dup2_args *uap)
377 {
378 
379 	return (kern_dup(td, FDDUP_FIXED, 0, (int)uap->from, (int)uap->to));
380 }
381 
382 /*
383  * Duplicate a file descriptor.
384  */
385 #ifndef _SYS_SYSPROTO_H_
386 struct dup_args {
387 	u_int	fd;
388 };
389 #endif
390 /* ARGSUSED */
391 int
392 sys_dup(struct thread *td, struct dup_args *uap)
393 {
394 
395 	return (kern_dup(td, FDDUP_NORMAL, 0, (int)uap->fd, 0));
396 }
397 
398 /*
399  * The file control system call.
400  */
401 #ifndef _SYS_SYSPROTO_H_
402 struct fcntl_args {
403 	int	fd;
404 	int	cmd;
405 	long	arg;
406 };
407 #endif
408 /* ARGSUSED */
409 int
410 sys_fcntl(struct thread *td, struct fcntl_args *uap)
411 {
412 
413 	return (kern_fcntl_freebsd(td, uap->fd, uap->cmd, uap->arg));
414 }
415 
416 int
417 kern_fcntl_freebsd(struct thread *td, int fd, int cmd, long arg)
418 {
419 	struct flock fl;
420 	struct __oflock ofl;
421 	intptr_t arg1;
422 	int error, newcmd;
423 
424 	error = 0;
425 	newcmd = cmd;
426 	switch (cmd) {
427 	case F_OGETLK:
428 	case F_OSETLK:
429 	case F_OSETLKW:
430 		/*
431 		 * Convert old flock structure to new.
432 		 */
433 		error = copyin((void *)(intptr_t)arg, &ofl, sizeof(ofl));
434 		fl.l_start = ofl.l_start;
435 		fl.l_len = ofl.l_len;
436 		fl.l_pid = ofl.l_pid;
437 		fl.l_type = ofl.l_type;
438 		fl.l_whence = ofl.l_whence;
439 		fl.l_sysid = 0;
440 
441 		switch (cmd) {
442 		case F_OGETLK:
443 			newcmd = F_GETLK;
444 			break;
445 		case F_OSETLK:
446 			newcmd = F_SETLK;
447 			break;
448 		case F_OSETLKW:
449 			newcmd = F_SETLKW;
450 			break;
451 		}
452 		arg1 = (intptr_t)&fl;
453 		break;
454 	case F_GETLK:
455 	case F_SETLK:
456 	case F_SETLKW:
457 	case F_SETLK_REMOTE:
458 		error = copyin((void *)(intptr_t)arg, &fl, sizeof(fl));
459 		arg1 = (intptr_t)&fl;
460 		break;
461 	default:
462 		arg1 = arg;
463 		break;
464 	}
465 	if (error)
466 		return (error);
467 	error = kern_fcntl(td, fd, newcmd, arg1);
468 	if (error)
469 		return (error);
470 	if (cmd == F_OGETLK) {
471 		ofl.l_start = fl.l_start;
472 		ofl.l_len = fl.l_len;
473 		ofl.l_pid = fl.l_pid;
474 		ofl.l_type = fl.l_type;
475 		ofl.l_whence = fl.l_whence;
476 		error = copyout(&ofl, (void *)(intptr_t)arg, sizeof(ofl));
477 	} else if (cmd == F_GETLK) {
478 		error = copyout(&fl, (void *)(intptr_t)arg, sizeof(fl));
479 	}
480 	return (error);
481 }
482 
483 int
484 kern_fcntl(struct thread *td, int fd, int cmd, intptr_t arg)
485 {
486 	struct filedesc *fdp;
487 	struct flock *flp;
488 	struct file *fp, *fp2;
489 	struct filedescent *fde;
490 	struct proc *p;
491 	struct vnode *vp;
492 	struct mount *mp;
493 	int error, flg, seals, tmp;
494 	uint64_t bsize;
495 	off_t foffset;
496 
497 	error = 0;
498 	flg = F_POSIX;
499 	p = td->td_proc;
500 	fdp = p->p_fd;
501 
502 	AUDIT_ARG_FD(cmd);
503 	AUDIT_ARG_CMD(cmd);
504 	switch (cmd) {
505 	case F_DUPFD:
506 		tmp = arg;
507 		error = kern_dup(td, FDDUP_FCNTL, 0, fd, tmp);
508 		break;
509 
510 	case F_DUPFD_CLOEXEC:
511 		tmp = arg;
512 		error = kern_dup(td, FDDUP_FCNTL, FDDUP_FLAG_CLOEXEC, fd, tmp);
513 		break;
514 
515 	case F_DUP2FD:
516 		tmp = arg;
517 		error = kern_dup(td, FDDUP_FIXED, 0, fd, tmp);
518 		break;
519 
520 	case F_DUP2FD_CLOEXEC:
521 		tmp = arg;
522 		error = kern_dup(td, FDDUP_FIXED, FDDUP_FLAG_CLOEXEC, fd, tmp);
523 		break;
524 
525 	case F_GETFD:
526 		error = EBADF;
527 		FILEDESC_SLOCK(fdp);
528 		fde = fdeget_locked(fdp, fd);
529 		if (fde != NULL) {
530 			td->td_retval[0] =
531 			    (fde->fde_flags & UF_EXCLOSE) ? FD_CLOEXEC : 0;
532 			error = 0;
533 		}
534 		FILEDESC_SUNLOCK(fdp);
535 		break;
536 
537 	case F_SETFD:
538 		error = EBADF;
539 		FILEDESC_XLOCK(fdp);
540 		fde = fdeget_locked(fdp, fd);
541 		if (fde != NULL) {
542 			fde->fde_flags = (fde->fde_flags & ~UF_EXCLOSE) |
543 			    (arg & FD_CLOEXEC ? UF_EXCLOSE : 0);
544 			error = 0;
545 		}
546 		FILEDESC_XUNLOCK(fdp);
547 		break;
548 
549 	case F_GETFL:
550 		error = fget_fcntl(td, fd, &cap_fcntl_rights, F_GETFL, &fp);
551 		if (error != 0)
552 			break;
553 		td->td_retval[0] = OFLAGS(fp->f_flag);
554 		fdrop(fp, td);
555 		break;
556 
557 	case F_SETFL:
558 		error = fget_fcntl(td, fd, &cap_fcntl_rights, F_SETFL, &fp);
559 		if (error != 0)
560 			break;
561 		do {
562 			tmp = flg = fp->f_flag;
563 			tmp &= ~FCNTLFLAGS;
564 			tmp |= FFLAGS(arg & ~O_ACCMODE) & FCNTLFLAGS;
565 		} while(atomic_cmpset_int(&fp->f_flag, flg, tmp) == 0);
566 		tmp = fp->f_flag & FNONBLOCK;
567 		error = fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
568 		if (error != 0) {
569 			fdrop(fp, td);
570 			break;
571 		}
572 		tmp = fp->f_flag & FASYNC;
573 		error = fo_ioctl(fp, FIOASYNC, &tmp, td->td_ucred, td);
574 		if (error == 0) {
575 			fdrop(fp, td);
576 			break;
577 		}
578 		atomic_clear_int(&fp->f_flag, FNONBLOCK);
579 		tmp = 0;
580 		(void)fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
581 		fdrop(fp, td);
582 		break;
583 
584 	case F_GETOWN:
585 		error = fget_fcntl(td, fd, &cap_fcntl_rights, F_GETOWN, &fp);
586 		if (error != 0)
587 			break;
588 		error = fo_ioctl(fp, FIOGETOWN, &tmp, td->td_ucred, td);
589 		if (error == 0)
590 			td->td_retval[0] = tmp;
591 		fdrop(fp, td);
592 		break;
593 
594 	case F_SETOWN:
595 		error = fget_fcntl(td, fd, &cap_fcntl_rights, F_SETOWN, &fp);
596 		if (error != 0)
597 			break;
598 		tmp = arg;
599 		error = fo_ioctl(fp, FIOSETOWN, &tmp, td->td_ucred, td);
600 		fdrop(fp, td);
601 		break;
602 
603 	case F_SETLK_REMOTE:
604 		error = priv_check(td, PRIV_NFS_LOCKD);
605 		if (error != 0)
606 			return (error);
607 		flg = F_REMOTE;
608 		goto do_setlk;
609 
610 	case F_SETLKW:
611 		flg |= F_WAIT;
612 		/* FALLTHROUGH F_SETLK */
613 
614 	case F_SETLK:
615 	do_setlk:
616 		flp = (struct flock *)arg;
617 		if ((flg & F_REMOTE) != 0 && flp->l_sysid == 0) {
618 			error = EINVAL;
619 			break;
620 		}
621 
622 		error = fget_unlocked(fdp, fd, &cap_flock_rights, &fp);
623 		if (error != 0)
624 			break;
625 		if (fp->f_type != DTYPE_VNODE) {
626 			error = EBADF;
627 			fdrop(fp, td);
628 			break;
629 		}
630 
631 		if (flp->l_whence == SEEK_CUR) {
632 			foffset = foffset_get(fp);
633 			if (foffset < 0 ||
634 			    (flp->l_start > 0 &&
635 			     foffset > OFF_MAX - flp->l_start)) {
636 				error = EOVERFLOW;
637 				fdrop(fp, td);
638 				break;
639 			}
640 			flp->l_start += foffset;
641 		}
642 
643 		vp = fp->f_vnode;
644 		switch (flp->l_type) {
645 		case F_RDLCK:
646 			if ((fp->f_flag & FREAD) == 0) {
647 				error = EBADF;
648 				break;
649 			}
650 			if ((p->p_leader->p_flag & P_ADVLOCK) == 0) {
651 				PROC_LOCK(p->p_leader);
652 				p->p_leader->p_flag |= P_ADVLOCK;
653 				PROC_UNLOCK(p->p_leader);
654 			}
655 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
656 			    flp, flg);
657 			break;
658 		case F_WRLCK:
659 			if ((fp->f_flag & FWRITE) == 0) {
660 				error = EBADF;
661 				break;
662 			}
663 			if ((p->p_leader->p_flag & P_ADVLOCK) == 0) {
664 				PROC_LOCK(p->p_leader);
665 				p->p_leader->p_flag |= P_ADVLOCK;
666 				PROC_UNLOCK(p->p_leader);
667 			}
668 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
669 			    flp, flg);
670 			break;
671 		case F_UNLCK:
672 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
673 			    flp, flg);
674 			break;
675 		case F_UNLCKSYS:
676 			if (flg != F_REMOTE) {
677 				error = EINVAL;
678 				break;
679 			}
680 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
681 			    F_UNLCKSYS, flp, flg);
682 			break;
683 		default:
684 			error = EINVAL;
685 			break;
686 		}
687 		if (error != 0 || flp->l_type == F_UNLCK ||
688 		    flp->l_type == F_UNLCKSYS) {
689 			fdrop(fp, td);
690 			break;
691 		}
692 
693 		/*
694 		 * Check for a race with close.
695 		 *
696 		 * The vnode is now advisory locked (or unlocked, but this case
697 		 * is not really important) as the caller requested.
698 		 * We had to drop the filedesc lock, so we need to recheck if
699 		 * the descriptor is still valid, because if it was closed
700 		 * in the meantime we need to remove advisory lock from the
701 		 * vnode - close on any descriptor leading to an advisory
702 		 * locked vnode, removes that lock.
703 		 * We will return 0 on purpose in that case, as the result of
704 		 * successful advisory lock might have been externally visible
705 		 * already. This is fine - effectively we pretend to the caller
706 		 * that the closing thread was a bit slower and that the
707 		 * advisory lock succeeded before the close.
708 		 */
709 		error = fget_unlocked(fdp, fd, &cap_no_rights, &fp2);
710 		if (error != 0) {
711 			fdrop(fp, td);
712 			break;
713 		}
714 		if (fp != fp2) {
715 			flp->l_whence = SEEK_SET;
716 			flp->l_start = 0;
717 			flp->l_len = 0;
718 			flp->l_type = F_UNLCK;
719 			(void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
720 			    F_UNLCK, flp, F_POSIX);
721 		}
722 		fdrop(fp, td);
723 		fdrop(fp2, td);
724 		break;
725 
726 	case F_GETLK:
727 		error = fget_unlocked(fdp, fd, &cap_flock_rights, &fp);
728 		if (error != 0)
729 			break;
730 		if (fp->f_type != DTYPE_VNODE) {
731 			error = EBADF;
732 			fdrop(fp, td);
733 			break;
734 		}
735 		flp = (struct flock *)arg;
736 		if (flp->l_type != F_RDLCK && flp->l_type != F_WRLCK &&
737 		    flp->l_type != F_UNLCK) {
738 			error = EINVAL;
739 			fdrop(fp, td);
740 			break;
741 		}
742 		if (flp->l_whence == SEEK_CUR) {
743 			foffset = foffset_get(fp);
744 			if ((flp->l_start > 0 &&
745 			    foffset > OFF_MAX - flp->l_start) ||
746 			    (flp->l_start < 0 &&
747 			    foffset < OFF_MIN - flp->l_start)) {
748 				error = EOVERFLOW;
749 				fdrop(fp, td);
750 				break;
751 			}
752 			flp->l_start += foffset;
753 		}
754 		vp = fp->f_vnode;
755 		error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, flp,
756 		    F_POSIX);
757 		fdrop(fp, td);
758 		break;
759 
760 	case F_ADD_SEALS:
761 		error = fget_unlocked(fdp, fd, &cap_no_rights, &fp);
762 		if (error != 0)
763 			break;
764 		error = fo_add_seals(fp, arg);
765 		fdrop(fp, td);
766 		break;
767 
768 	case F_GET_SEALS:
769 		error = fget_unlocked(fdp, fd, &cap_no_rights, &fp);
770 		if (error != 0)
771 			break;
772 		if (fo_get_seals(fp, &seals) == 0)
773 			td->td_retval[0] = seals;
774 		else
775 			error = EINVAL;
776 		fdrop(fp, td);
777 		break;
778 
779 	case F_RDAHEAD:
780 		arg = arg ? 128 * 1024: 0;
781 		/* FALLTHROUGH */
782 	case F_READAHEAD:
783 		error = fget_unlocked(fdp, fd, &cap_no_rights, &fp);
784 		if (error != 0)
785 			break;
786 		if (fp->f_type != DTYPE_VNODE) {
787 			fdrop(fp, td);
788 			error = EBADF;
789 			break;
790 		}
791 		vp = fp->f_vnode;
792 		if (vp->v_type != VREG) {
793 			fdrop(fp, td);
794 			error = ENOTTY;
795 			break;
796 		}
797 
798 		/*
799 		 * Exclusive lock synchronizes against f_seqcount reads and
800 		 * writes in sequential_heuristic().
801 		 */
802 		error = vn_lock(vp, LK_EXCLUSIVE);
803 		if (error != 0) {
804 			fdrop(fp, td);
805 			break;
806 		}
807 		if (arg >= 0) {
808 			bsize = fp->f_vnode->v_mount->mnt_stat.f_iosize;
809 			arg = MIN(arg, INT_MAX - bsize + 1);
810 			fp->f_seqcount = MIN(IO_SEQMAX,
811 			    (arg + bsize - 1) / bsize);
812 			atomic_set_int(&fp->f_flag, FRDAHEAD);
813 		} else {
814 			atomic_clear_int(&fp->f_flag, FRDAHEAD);
815 		}
816 		VOP_UNLOCK(vp);
817 		fdrop(fp, td);
818 		break;
819 
820 	case F_ISUNIONSTACK:
821 		/*
822 		 * Check if the vnode is part of a union stack (either the
823 		 * "union" flag from mount(2) or unionfs).
824 		 *
825 		 * Prior to introduction of this op libc's readdir would call
826 		 * fstatfs(2), in effect unnecessarily copying kilobytes of
827 		 * data just to check fs name and a mount flag.
828 		 *
829 		 * Fixing the code to handle everything in the kernel instead
830 		 * is a non-trivial endeavor and has low priority, thus this
831 		 * horrible kludge facilitates the current behavior in a much
832 		 * cheaper manner until someone(tm) sorts this out.
833 		 */
834 		error = fget_unlocked(fdp, fd, &cap_no_rights, &fp);
835 		if (error != 0)
836 			break;
837 		if (fp->f_type != DTYPE_VNODE) {
838 			fdrop(fp, td);
839 			error = EBADF;
840 			break;
841 		}
842 		vp = fp->f_vnode;
843 		/*
844 		 * Since we don't prevent dooming the vnode even non-null mp
845 		 * found can become immediately stale. This is tolerable since
846 		 * mount points are type-stable (providing safe memory access)
847 		 * and any vfs op on this vnode going forward will return an
848 		 * error (meaning return value in this case is meaningless).
849 		 */
850 		mp = (struct mount *)atomic_load_ptr(&vp->v_mount);
851 		if (__predict_false(mp == NULL)) {
852 			fdrop(fp, td);
853 			error = EBADF;
854 			break;
855 		}
856 		td->td_retval[0] = 0;
857 		if (mp->mnt_kern_flag & MNTK_UNIONFS ||
858 		    mp->mnt_flag & MNT_UNION)
859 			td->td_retval[0] = 1;
860 		fdrop(fp, td);
861 		break;
862 
863 	default:
864 		error = EINVAL;
865 		break;
866 	}
867 	return (error);
868 }
869 
870 static int
871 getmaxfd(struct thread *td)
872 {
873 
874 	return (min((int)lim_cur(td, RLIMIT_NOFILE), maxfilesperproc));
875 }
876 
877 /*
878  * Common code for dup, dup2, fcntl(F_DUPFD) and fcntl(F_DUP2FD).
879  */
880 int
881 kern_dup(struct thread *td, u_int mode, int flags, int old, int new)
882 {
883 	struct filedesc *fdp;
884 	struct filedescent *oldfde, *newfde;
885 	struct proc *p;
886 	struct file *delfp;
887 	u_long *oioctls, *nioctls;
888 	int error, maxfd;
889 
890 	p = td->td_proc;
891 	fdp = p->p_fd;
892 	oioctls = NULL;
893 
894 	MPASS((flags & ~(FDDUP_FLAG_CLOEXEC)) == 0);
895 	MPASS(mode < FDDUP_LASTMODE);
896 
897 	AUDIT_ARG_FD(old);
898 	/* XXXRW: if (flags & FDDUP_FIXED) AUDIT_ARG_FD2(new); */
899 
900 	/*
901 	 * Verify we have a valid descriptor to dup from and possibly to
902 	 * dup to. Unlike dup() and dup2(), fcntl()'s F_DUPFD should
903 	 * return EINVAL when the new descriptor is out of bounds.
904 	 */
905 	if (old < 0)
906 		return (EBADF);
907 	if (new < 0)
908 		return (mode == FDDUP_FCNTL ? EINVAL : EBADF);
909 	maxfd = getmaxfd(td);
910 	if (new >= maxfd)
911 		return (mode == FDDUP_FCNTL ? EINVAL : EBADF);
912 
913 	error = EBADF;
914 	FILEDESC_XLOCK(fdp);
915 	if (fget_locked(fdp, old) == NULL)
916 		goto unlock;
917 	if ((mode == FDDUP_FIXED || mode == FDDUP_MUSTREPLACE) && old == new) {
918 		td->td_retval[0] = new;
919 		if (flags & FDDUP_FLAG_CLOEXEC)
920 			fdp->fd_ofiles[new].fde_flags |= UF_EXCLOSE;
921 		error = 0;
922 		goto unlock;
923 	}
924 
925 	oldfde = &fdp->fd_ofiles[old];
926 	if (!fhold(oldfde->fde_file))
927 		goto unlock;
928 
929 	/*
930 	 * If the caller specified a file descriptor, make sure the file
931 	 * table is large enough to hold it, and grab it.  Otherwise, just
932 	 * allocate a new descriptor the usual way.
933 	 */
934 	switch (mode) {
935 	case FDDUP_NORMAL:
936 	case FDDUP_FCNTL:
937 		if ((error = fdalloc(td, new, &new)) != 0) {
938 			fdrop(oldfde->fde_file, td);
939 			goto unlock;
940 		}
941 		break;
942 	case FDDUP_MUSTREPLACE:
943 		/* Target file descriptor must exist. */
944 		if (fget_locked(fdp, new) == NULL) {
945 			fdrop(oldfde->fde_file, td);
946 			goto unlock;
947 		}
948 		break;
949 	case FDDUP_FIXED:
950 		if (new >= fdp->fd_nfiles) {
951 			/*
952 			 * The resource limits are here instead of e.g.
953 			 * fdalloc(), because the file descriptor table may be
954 			 * shared between processes, so we can't really use
955 			 * racct_add()/racct_sub().  Instead of counting the
956 			 * number of actually allocated descriptors, just put
957 			 * the limit on the size of the file descriptor table.
958 			 */
959 #ifdef RACCT
960 			if (RACCT_ENABLED()) {
961 				error = racct_set_unlocked(p, RACCT_NOFILE, new + 1);
962 				if (error != 0) {
963 					error = EMFILE;
964 					fdrop(oldfde->fde_file, td);
965 					goto unlock;
966 				}
967 			}
968 #endif
969 			fdgrowtable_exp(fdp, new + 1);
970 		}
971 		if (!fdisused(fdp, new))
972 			fdused(fdp, new);
973 		break;
974 	default:
975 		KASSERT(0, ("%s unsupported mode %d", __func__, mode));
976 	}
977 
978 	KASSERT(old != new, ("new fd is same as old"));
979 
980 	newfde = &fdp->fd_ofiles[new];
981 	delfp = newfde->fde_file;
982 
983 	oioctls = filecaps_free_prep(&newfde->fde_caps);
984 	nioctls = filecaps_copy_prep(&oldfde->fde_caps);
985 
986 	/*
987 	 * Duplicate the source descriptor.
988 	 */
989 #ifdef CAPABILITIES
990 	seqc_write_begin(&newfde->fde_seqc);
991 #endif
992 	memcpy(newfde, oldfde, fde_change_size);
993 	filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps,
994 	    nioctls);
995 	if ((flags & FDDUP_FLAG_CLOEXEC) != 0)
996 		newfde->fde_flags = oldfde->fde_flags | UF_EXCLOSE;
997 	else
998 		newfde->fde_flags = oldfde->fde_flags & ~UF_EXCLOSE;
999 #ifdef CAPABILITIES
1000 	seqc_write_end(&newfde->fde_seqc);
1001 #endif
1002 	td->td_retval[0] = new;
1003 
1004 	error = 0;
1005 
1006 	if (delfp != NULL) {
1007 		(void) closefp(fdp, new, delfp, td, 1);
1008 		FILEDESC_UNLOCK_ASSERT(fdp);
1009 	} else {
1010 unlock:
1011 		FILEDESC_XUNLOCK(fdp);
1012 	}
1013 
1014 	filecaps_free_finish(oioctls);
1015 	return (error);
1016 }
1017 
1018 /*
1019  * If sigio is on the list associated with a process or process group,
1020  * disable signalling from the device, remove sigio from the list and
1021  * free sigio.
1022  */
1023 void
1024 funsetown(struct sigio **sigiop)
1025 {
1026 	struct sigio *sigio;
1027 
1028 	if (*sigiop == NULL)
1029 		return;
1030 	SIGIO_LOCK();
1031 	sigio = *sigiop;
1032 	if (sigio == NULL) {
1033 		SIGIO_UNLOCK();
1034 		return;
1035 	}
1036 	*(sigio->sio_myref) = NULL;
1037 	if ((sigio)->sio_pgid < 0) {
1038 		struct pgrp *pg = (sigio)->sio_pgrp;
1039 		PGRP_LOCK(pg);
1040 		SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio,
1041 			    sigio, sio_pgsigio);
1042 		PGRP_UNLOCK(pg);
1043 	} else {
1044 		struct proc *p = (sigio)->sio_proc;
1045 		PROC_LOCK(p);
1046 		SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio,
1047 			    sigio, sio_pgsigio);
1048 		PROC_UNLOCK(p);
1049 	}
1050 	SIGIO_UNLOCK();
1051 	crfree(sigio->sio_ucred);
1052 	free(sigio, M_SIGIO);
1053 }
1054 
1055 /*
1056  * Free a list of sigio structures.
1057  * We only need to lock the SIGIO_LOCK because we have made ourselves
1058  * inaccessible to callers of fsetown and therefore do not need to lock
1059  * the proc or pgrp struct for the list manipulation.
1060  */
1061 void
1062 funsetownlst(struct sigiolst *sigiolst)
1063 {
1064 	struct proc *p;
1065 	struct pgrp *pg;
1066 	struct sigio *sigio;
1067 
1068 	sigio = SLIST_FIRST(sigiolst);
1069 	if (sigio == NULL)
1070 		return;
1071 	p = NULL;
1072 	pg = NULL;
1073 
1074 	/*
1075 	 * Every entry of the list should belong
1076 	 * to a single proc or pgrp.
1077 	 */
1078 	if (sigio->sio_pgid < 0) {
1079 		pg = sigio->sio_pgrp;
1080 		PGRP_LOCK_ASSERT(pg, MA_NOTOWNED);
1081 	} else /* if (sigio->sio_pgid > 0) */ {
1082 		p = sigio->sio_proc;
1083 		PROC_LOCK_ASSERT(p, MA_NOTOWNED);
1084 	}
1085 
1086 	SIGIO_LOCK();
1087 	while ((sigio = SLIST_FIRST(sigiolst)) != NULL) {
1088 		*(sigio->sio_myref) = NULL;
1089 		if (pg != NULL) {
1090 			KASSERT(sigio->sio_pgid < 0,
1091 			    ("Proc sigio in pgrp sigio list"));
1092 			KASSERT(sigio->sio_pgrp == pg,
1093 			    ("Bogus pgrp in sigio list"));
1094 			PGRP_LOCK(pg);
1095 			SLIST_REMOVE(&pg->pg_sigiolst, sigio, sigio,
1096 			    sio_pgsigio);
1097 			PGRP_UNLOCK(pg);
1098 		} else /* if (p != NULL) */ {
1099 			KASSERT(sigio->sio_pgid > 0,
1100 			    ("Pgrp sigio in proc sigio list"));
1101 			KASSERT(sigio->sio_proc == p,
1102 			    ("Bogus proc in sigio list"));
1103 			PROC_LOCK(p);
1104 			SLIST_REMOVE(&p->p_sigiolst, sigio, sigio,
1105 			    sio_pgsigio);
1106 			PROC_UNLOCK(p);
1107 		}
1108 		SIGIO_UNLOCK();
1109 		crfree(sigio->sio_ucred);
1110 		free(sigio, M_SIGIO);
1111 		SIGIO_LOCK();
1112 	}
1113 	SIGIO_UNLOCK();
1114 }
1115 
1116 /*
1117  * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg).
1118  *
1119  * After permission checking, add a sigio structure to the sigio list for
1120  * the process or process group.
1121  */
1122 int
1123 fsetown(pid_t pgid, struct sigio **sigiop)
1124 {
1125 	struct proc *proc;
1126 	struct pgrp *pgrp;
1127 	struct sigio *sigio;
1128 	int ret;
1129 
1130 	if (pgid == 0) {
1131 		funsetown(sigiop);
1132 		return (0);
1133 	}
1134 
1135 	ret = 0;
1136 
1137 	/* Allocate and fill in the new sigio out of locks. */
1138 	sigio = malloc(sizeof(struct sigio), M_SIGIO, M_WAITOK);
1139 	sigio->sio_pgid = pgid;
1140 	sigio->sio_ucred = crhold(curthread->td_ucred);
1141 	sigio->sio_myref = sigiop;
1142 
1143 	sx_slock(&proctree_lock);
1144 	if (pgid > 0) {
1145 		proc = pfind(pgid);
1146 		if (proc == NULL) {
1147 			ret = ESRCH;
1148 			goto fail;
1149 		}
1150 
1151 		/*
1152 		 * Policy - Don't allow a process to FSETOWN a process
1153 		 * in another session.
1154 		 *
1155 		 * Remove this test to allow maximum flexibility or
1156 		 * restrict FSETOWN to the current process or process
1157 		 * group for maximum safety.
1158 		 */
1159 		PROC_UNLOCK(proc);
1160 		if (proc->p_session != curthread->td_proc->p_session) {
1161 			ret = EPERM;
1162 			goto fail;
1163 		}
1164 
1165 		pgrp = NULL;
1166 	} else /* if (pgid < 0) */ {
1167 		pgrp = pgfind(-pgid);
1168 		if (pgrp == NULL) {
1169 			ret = ESRCH;
1170 			goto fail;
1171 		}
1172 		PGRP_UNLOCK(pgrp);
1173 
1174 		/*
1175 		 * Policy - Don't allow a process to FSETOWN a process
1176 		 * in another session.
1177 		 *
1178 		 * Remove this test to allow maximum flexibility or
1179 		 * restrict FSETOWN to the current process or process
1180 		 * group for maximum safety.
1181 		 */
1182 		if (pgrp->pg_session != curthread->td_proc->p_session) {
1183 			ret = EPERM;
1184 			goto fail;
1185 		}
1186 
1187 		proc = NULL;
1188 	}
1189 	funsetown(sigiop);
1190 	if (pgid > 0) {
1191 		PROC_LOCK(proc);
1192 		/*
1193 		 * Since funsetownlst() is called without the proctree
1194 		 * locked, we need to check for P_WEXIT.
1195 		 * XXX: is ESRCH correct?
1196 		 */
1197 		if ((proc->p_flag & P_WEXIT) != 0) {
1198 			PROC_UNLOCK(proc);
1199 			ret = ESRCH;
1200 			goto fail;
1201 		}
1202 		SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio);
1203 		sigio->sio_proc = proc;
1204 		PROC_UNLOCK(proc);
1205 	} else {
1206 		PGRP_LOCK(pgrp);
1207 		SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio);
1208 		sigio->sio_pgrp = pgrp;
1209 		PGRP_UNLOCK(pgrp);
1210 	}
1211 	sx_sunlock(&proctree_lock);
1212 	SIGIO_LOCK();
1213 	*sigiop = sigio;
1214 	SIGIO_UNLOCK();
1215 	return (0);
1216 
1217 fail:
1218 	sx_sunlock(&proctree_lock);
1219 	crfree(sigio->sio_ucred);
1220 	free(sigio, M_SIGIO);
1221 	return (ret);
1222 }
1223 
1224 /*
1225  * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg).
1226  */
1227 pid_t
1228 fgetown(struct sigio **sigiop)
1229 {
1230 	pid_t pgid;
1231 
1232 	SIGIO_LOCK();
1233 	pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0;
1234 	SIGIO_UNLOCK();
1235 	return (pgid);
1236 }
1237 
1238 /*
1239  * Function drops the filedesc lock on return.
1240  */
1241 static int
1242 closefp(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1243     int holdleaders)
1244 {
1245 	int error;
1246 
1247 	FILEDESC_XLOCK_ASSERT(fdp);
1248 
1249 	if (holdleaders) {
1250 		if (td->td_proc->p_fdtol != NULL) {
1251 			/*
1252 			 * Ask fdfree() to sleep to ensure that all relevant
1253 			 * process leaders can be traversed in closef().
1254 			 */
1255 			fdp->fd_holdleaderscount++;
1256 		} else {
1257 			holdleaders = 0;
1258 		}
1259 	}
1260 
1261 	/*
1262 	 * We now hold the fp reference that used to be owned by the
1263 	 * descriptor array.  We have to unlock the FILEDESC *AFTER*
1264 	 * knote_fdclose to prevent a race of the fd getting opened, a knote
1265 	 * added, and deleteing a knote for the new fd.
1266 	 */
1267 	if (__predict_false(!TAILQ_EMPTY(&fdp->fd_kqlist)))
1268 		knote_fdclose(td, fd);
1269 
1270 	/*
1271 	 * We need to notify mqueue if the object is of type mqueue.
1272 	 */
1273 	if (__predict_false(fp->f_type == DTYPE_MQUEUE))
1274 		mq_fdclose(td, fd, fp);
1275 	FILEDESC_XUNLOCK(fdp);
1276 
1277 	error = closef(fp, td);
1278 	if (holdleaders) {
1279 		FILEDESC_XLOCK(fdp);
1280 		fdp->fd_holdleaderscount--;
1281 		if (fdp->fd_holdleaderscount == 0 &&
1282 		    fdp->fd_holdleaderswakeup != 0) {
1283 			fdp->fd_holdleaderswakeup = 0;
1284 			wakeup(&fdp->fd_holdleaderscount);
1285 		}
1286 		FILEDESC_XUNLOCK(fdp);
1287 	}
1288 	return (error);
1289 }
1290 
1291 /*
1292  * Close a file descriptor.
1293  */
1294 #ifndef _SYS_SYSPROTO_H_
1295 struct close_args {
1296 	int     fd;
1297 };
1298 #endif
1299 /* ARGSUSED */
1300 int
1301 sys_close(struct thread *td, struct close_args *uap)
1302 {
1303 
1304 	return (kern_close(td, uap->fd));
1305 }
1306 
1307 int
1308 kern_close(struct thread *td, int fd)
1309 {
1310 	struct filedesc *fdp;
1311 	struct file *fp;
1312 
1313 	fdp = td->td_proc->p_fd;
1314 
1315 	AUDIT_SYSCLOSE(td, fd);
1316 
1317 	FILEDESC_XLOCK(fdp);
1318 	if ((fp = fget_locked(fdp, fd)) == NULL) {
1319 		FILEDESC_XUNLOCK(fdp);
1320 		return (EBADF);
1321 	}
1322 	fdfree(fdp, fd);
1323 
1324 	/* closefp() drops the FILEDESC lock for us. */
1325 	return (closefp(fdp, fd, fp, td, 1));
1326 }
1327 
1328 /*
1329  * Close open file descriptors.
1330  */
1331 #ifndef _SYS_SYSPROTO_H_
1332 struct closefrom_args {
1333 	int	lowfd;
1334 };
1335 #endif
1336 /* ARGSUSED */
1337 int
1338 sys_closefrom(struct thread *td, struct closefrom_args *uap)
1339 {
1340 	struct filedesc *fdp;
1341 	int fd;
1342 
1343 	fdp = td->td_proc->p_fd;
1344 	AUDIT_ARG_FD(uap->lowfd);
1345 
1346 	/*
1347 	 * Treat negative starting file descriptor values identical to
1348 	 * closefrom(0) which closes all files.
1349 	 */
1350 	if (uap->lowfd < 0)
1351 		uap->lowfd = 0;
1352 	FILEDESC_SLOCK(fdp);
1353 	for (fd = uap->lowfd; fd <= fdp->fd_lastfile; fd++) {
1354 		if (fdp->fd_ofiles[fd].fde_file != NULL) {
1355 			FILEDESC_SUNLOCK(fdp);
1356 			(void)kern_close(td, fd);
1357 			FILEDESC_SLOCK(fdp);
1358 		}
1359 	}
1360 	FILEDESC_SUNLOCK(fdp);
1361 	return (0);
1362 }
1363 
1364 #if defined(COMPAT_43)
1365 /*
1366  * Return status information about a file descriptor.
1367  */
1368 #ifndef _SYS_SYSPROTO_H_
1369 struct ofstat_args {
1370 	int	fd;
1371 	struct	ostat *sb;
1372 };
1373 #endif
1374 /* ARGSUSED */
1375 int
1376 ofstat(struct thread *td, struct ofstat_args *uap)
1377 {
1378 	struct ostat oub;
1379 	struct stat ub;
1380 	int error;
1381 
1382 	error = kern_fstat(td, uap->fd, &ub);
1383 	if (error == 0) {
1384 		cvtstat(&ub, &oub);
1385 		error = copyout(&oub, uap->sb, sizeof(oub));
1386 	}
1387 	return (error);
1388 }
1389 #endif /* COMPAT_43 */
1390 
1391 #if defined(COMPAT_FREEBSD11)
1392 int
1393 freebsd11_fstat(struct thread *td, struct freebsd11_fstat_args *uap)
1394 {
1395 	struct stat sb;
1396 	struct freebsd11_stat osb;
1397 	int error;
1398 
1399 	error = kern_fstat(td, uap->fd, &sb);
1400 	if (error != 0)
1401 		return (error);
1402 	error = freebsd11_cvtstat(&sb, &osb);
1403 	if (error == 0)
1404 		error = copyout(&osb, uap->sb, sizeof(osb));
1405 	return (error);
1406 }
1407 #endif	/* COMPAT_FREEBSD11 */
1408 
1409 /*
1410  * Return status information about a file descriptor.
1411  */
1412 #ifndef _SYS_SYSPROTO_H_
1413 struct fstat_args {
1414 	int	fd;
1415 	struct	stat *sb;
1416 };
1417 #endif
1418 /* ARGSUSED */
1419 int
1420 sys_fstat(struct thread *td, struct fstat_args *uap)
1421 {
1422 	struct stat ub;
1423 	int error;
1424 
1425 	error = kern_fstat(td, uap->fd, &ub);
1426 	if (error == 0)
1427 		error = copyout(&ub, uap->sb, sizeof(ub));
1428 	return (error);
1429 }
1430 
1431 int
1432 kern_fstat(struct thread *td, int fd, struct stat *sbp)
1433 {
1434 	struct file *fp;
1435 	int error;
1436 
1437 	AUDIT_ARG_FD(fd);
1438 
1439 	error = fget(td, fd, &cap_fstat_rights, &fp);
1440 	if (__predict_false(error != 0))
1441 		return (error);
1442 
1443 	AUDIT_ARG_FILE(td->td_proc, fp);
1444 
1445 	error = fo_stat(fp, sbp, td->td_ucred, td);
1446 	fdrop(fp, td);
1447 #ifdef __STAT_TIME_T_EXT
1448 	sbp->st_atim_ext = 0;
1449 	sbp->st_mtim_ext = 0;
1450 	sbp->st_ctim_ext = 0;
1451 	sbp->st_btim_ext = 0;
1452 #endif
1453 #ifdef KTRACE
1454 	if (KTRPOINT(td, KTR_STRUCT))
1455 		ktrstat_error(sbp, error);
1456 #endif
1457 	return (error);
1458 }
1459 
1460 #if defined(COMPAT_FREEBSD11)
1461 /*
1462  * Return status information about a file descriptor.
1463  */
1464 #ifndef _SYS_SYSPROTO_H_
1465 struct freebsd11_nfstat_args {
1466 	int	fd;
1467 	struct	nstat *sb;
1468 };
1469 #endif
1470 /* ARGSUSED */
1471 int
1472 freebsd11_nfstat(struct thread *td, struct freebsd11_nfstat_args *uap)
1473 {
1474 	struct nstat nub;
1475 	struct stat ub;
1476 	int error;
1477 
1478 	error = kern_fstat(td, uap->fd, &ub);
1479 	if (error == 0) {
1480 		freebsd11_cvtnstat(&ub, &nub);
1481 		error = copyout(&nub, uap->sb, sizeof(nub));
1482 	}
1483 	return (error);
1484 }
1485 #endif /* COMPAT_FREEBSD11 */
1486 
1487 /*
1488  * Return pathconf information about a file descriptor.
1489  */
1490 #ifndef _SYS_SYSPROTO_H_
1491 struct fpathconf_args {
1492 	int	fd;
1493 	int	name;
1494 };
1495 #endif
1496 /* ARGSUSED */
1497 int
1498 sys_fpathconf(struct thread *td, struct fpathconf_args *uap)
1499 {
1500 	long value;
1501 	int error;
1502 
1503 	error = kern_fpathconf(td, uap->fd, uap->name, &value);
1504 	if (error == 0)
1505 		td->td_retval[0] = value;
1506 	return (error);
1507 }
1508 
1509 int
1510 kern_fpathconf(struct thread *td, int fd, int name, long *valuep)
1511 {
1512 	struct file *fp;
1513 	struct vnode *vp;
1514 	int error;
1515 
1516 	error = fget(td, fd, &cap_fpathconf_rights, &fp);
1517 	if (error != 0)
1518 		return (error);
1519 
1520 	if (name == _PC_ASYNC_IO) {
1521 		*valuep = _POSIX_ASYNCHRONOUS_IO;
1522 		goto out;
1523 	}
1524 	vp = fp->f_vnode;
1525 	if (vp != NULL) {
1526 		vn_lock(vp, LK_SHARED | LK_RETRY);
1527 		error = VOP_PATHCONF(vp, name, valuep);
1528 		VOP_UNLOCK(vp);
1529 	} else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) {
1530 		if (name != _PC_PIPE_BUF) {
1531 			error = EINVAL;
1532 		} else {
1533 			*valuep = PIPE_BUF;
1534 			error = 0;
1535 		}
1536 	} else {
1537 		error = EOPNOTSUPP;
1538 	}
1539 out:
1540 	fdrop(fp, td);
1541 	return (error);
1542 }
1543 
1544 /*
1545  * Copy filecaps structure allocating memory for ioctls array if needed.
1546  *
1547  * The last parameter indicates whether the fdtable is locked. If it is not and
1548  * ioctls are encountered, copying fails and the caller must lock the table.
1549  *
1550  * Note that if the table was not locked, the caller has to check the relevant
1551  * sequence counter to determine whether the operation was successful.
1552  */
1553 bool
1554 filecaps_copy(const struct filecaps *src, struct filecaps *dst, bool locked)
1555 {
1556 	size_t size;
1557 
1558 	if (src->fc_ioctls != NULL && !locked)
1559 		return (false);
1560 	memcpy(dst, src, sizeof(*src));
1561 	if (src->fc_ioctls == NULL)
1562 		return (true);
1563 
1564 	KASSERT(src->fc_nioctls > 0,
1565 	    ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1566 
1567 	size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1568 	dst->fc_ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1569 	memcpy(dst->fc_ioctls, src->fc_ioctls, size);
1570 	return (true);
1571 }
1572 
1573 static u_long *
1574 filecaps_copy_prep(const struct filecaps *src)
1575 {
1576 	u_long *ioctls;
1577 	size_t size;
1578 
1579 	if (__predict_true(src->fc_ioctls == NULL))
1580 		return (NULL);
1581 
1582 	KASSERT(src->fc_nioctls > 0,
1583 	    ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1584 
1585 	size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1586 	ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1587 	return (ioctls);
1588 }
1589 
1590 static void
1591 filecaps_copy_finish(const struct filecaps *src, struct filecaps *dst,
1592     u_long *ioctls)
1593 {
1594 	size_t size;
1595 
1596 	*dst = *src;
1597 	if (__predict_true(src->fc_ioctls == NULL)) {
1598 		MPASS(ioctls == NULL);
1599 		return;
1600 	}
1601 
1602 	size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1603 	dst->fc_ioctls = ioctls;
1604 	bcopy(src->fc_ioctls, dst->fc_ioctls, size);
1605 }
1606 
1607 /*
1608  * Move filecaps structure to the new place and clear the old place.
1609  */
1610 void
1611 filecaps_move(struct filecaps *src, struct filecaps *dst)
1612 {
1613 
1614 	*dst = *src;
1615 	bzero(src, sizeof(*src));
1616 }
1617 
1618 /*
1619  * Fill the given filecaps structure with full rights.
1620  */
1621 static void
1622 filecaps_fill(struct filecaps *fcaps)
1623 {
1624 
1625 	CAP_ALL(&fcaps->fc_rights);
1626 	fcaps->fc_ioctls = NULL;
1627 	fcaps->fc_nioctls = -1;
1628 	fcaps->fc_fcntls = CAP_FCNTL_ALL;
1629 }
1630 
1631 /*
1632  * Free memory allocated within filecaps structure.
1633  */
1634 void
1635 filecaps_free(struct filecaps *fcaps)
1636 {
1637 
1638 	free(fcaps->fc_ioctls, M_FILECAPS);
1639 	bzero(fcaps, sizeof(*fcaps));
1640 }
1641 
1642 static u_long *
1643 filecaps_free_prep(struct filecaps *fcaps)
1644 {
1645 	u_long *ioctls;
1646 
1647 	ioctls = fcaps->fc_ioctls;
1648 	bzero(fcaps, sizeof(*fcaps));
1649 	return (ioctls);
1650 }
1651 
1652 static void
1653 filecaps_free_finish(u_long *ioctls)
1654 {
1655 
1656 	free(ioctls, M_FILECAPS);
1657 }
1658 
1659 /*
1660  * Validate the given filecaps structure.
1661  */
1662 static void
1663 filecaps_validate(const struct filecaps *fcaps, const char *func)
1664 {
1665 
1666 	KASSERT(cap_rights_is_valid(&fcaps->fc_rights),
1667 	    ("%s: invalid rights", func));
1668 	KASSERT((fcaps->fc_fcntls & ~CAP_FCNTL_ALL) == 0,
1669 	    ("%s: invalid fcntls", func));
1670 	KASSERT(fcaps->fc_fcntls == 0 ||
1671 	    cap_rights_is_set(&fcaps->fc_rights, CAP_FCNTL),
1672 	    ("%s: fcntls without CAP_FCNTL", func));
1673 	KASSERT(fcaps->fc_ioctls != NULL ? fcaps->fc_nioctls > 0 :
1674 	    (fcaps->fc_nioctls == -1 || fcaps->fc_nioctls == 0),
1675 	    ("%s: invalid ioctls", func));
1676 	KASSERT(fcaps->fc_nioctls == 0 ||
1677 	    cap_rights_is_set(&fcaps->fc_rights, CAP_IOCTL),
1678 	    ("%s: ioctls without CAP_IOCTL", func));
1679 }
1680 
1681 static void
1682 fdgrowtable_exp(struct filedesc *fdp, int nfd)
1683 {
1684 	int nfd1;
1685 
1686 	FILEDESC_XLOCK_ASSERT(fdp);
1687 
1688 	nfd1 = fdp->fd_nfiles * 2;
1689 	if (nfd1 < nfd)
1690 		nfd1 = nfd;
1691 	fdgrowtable(fdp, nfd1);
1692 }
1693 
1694 /*
1695  * Grow the file table to accommodate (at least) nfd descriptors.
1696  */
1697 static void
1698 fdgrowtable(struct filedesc *fdp, int nfd)
1699 {
1700 	struct filedesc0 *fdp0;
1701 	struct freetable *ft;
1702 	struct fdescenttbl *ntable;
1703 	struct fdescenttbl *otable;
1704 	int nnfiles, onfiles;
1705 	NDSLOTTYPE *nmap, *omap;
1706 
1707 	/*
1708 	 * If lastfile is -1 this struct filedesc was just allocated and we are
1709 	 * growing it to accommodate for the one we are going to copy from. There
1710 	 * is no need to have a lock on this one as it's not visible to anyone.
1711 	 */
1712 	if (fdp->fd_lastfile != -1)
1713 		FILEDESC_XLOCK_ASSERT(fdp);
1714 
1715 	KASSERT(fdp->fd_nfiles > 0, ("zero-length file table"));
1716 
1717 	/* save old values */
1718 	onfiles = fdp->fd_nfiles;
1719 	otable = fdp->fd_files;
1720 	omap = fdp->fd_map;
1721 
1722 	/* compute the size of the new table */
1723 	nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */
1724 	if (nnfiles <= onfiles)
1725 		/* the table is already large enough */
1726 		return;
1727 
1728 	/*
1729 	 * Allocate a new table.  We need enough space for the number of
1730 	 * entries, file entries themselves and the struct freetable we will use
1731 	 * when we decommission the table and place it on the freelist.
1732 	 * We place the struct freetable in the middle so we don't have
1733 	 * to worry about padding.
1734 	 */
1735 	ntable = malloc(offsetof(struct fdescenttbl, fdt_ofiles) +
1736 	    nnfiles * sizeof(ntable->fdt_ofiles[0]) +
1737 	    sizeof(struct freetable),
1738 	    M_FILEDESC, M_ZERO | M_WAITOK);
1739 	/* copy the old data */
1740 	ntable->fdt_nfiles = nnfiles;
1741 	memcpy(ntable->fdt_ofiles, otable->fdt_ofiles,
1742 	    onfiles * sizeof(ntable->fdt_ofiles[0]));
1743 
1744 	/*
1745 	 * Allocate a new map only if the old is not large enough.  It will
1746 	 * grow at a slower rate than the table as it can map more
1747 	 * entries than the table can hold.
1748 	 */
1749 	if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) {
1750 		nmap = malloc(NDSLOTS(nnfiles) * NDSLOTSIZE, M_FILEDESC,
1751 		    M_ZERO | M_WAITOK);
1752 		/* copy over the old data and update the pointer */
1753 		memcpy(nmap, omap, NDSLOTS(onfiles) * sizeof(*omap));
1754 		fdp->fd_map = nmap;
1755 	}
1756 
1757 	/*
1758 	 * Make sure that ntable is correctly initialized before we replace
1759 	 * fd_files poiner. Otherwise fget_unlocked() may see inconsistent
1760 	 * data.
1761 	 */
1762 	atomic_store_rel_ptr((volatile void *)&fdp->fd_files, (uintptr_t)ntable);
1763 
1764 	/*
1765 	 * Do not free the old file table, as some threads may still
1766 	 * reference entries within it.  Instead, place it on a freelist
1767 	 * which will be processed when the struct filedesc is released.
1768 	 *
1769 	 * Note that if onfiles == NDFILE, we're dealing with the original
1770 	 * static allocation contained within (struct filedesc0 *)fdp,
1771 	 * which must not be freed.
1772 	 */
1773 	if (onfiles > NDFILE) {
1774 		ft = (struct freetable *)&otable->fdt_ofiles[onfiles];
1775 		fdp0 = (struct filedesc0 *)fdp;
1776 		ft->ft_table = otable;
1777 		SLIST_INSERT_HEAD(&fdp0->fd_free, ft, ft_next);
1778 	}
1779 	/*
1780 	 * The map does not have the same possibility of threads still
1781 	 * holding references to it.  So always free it as long as it
1782 	 * does not reference the original static allocation.
1783 	 */
1784 	if (NDSLOTS(onfiles) > NDSLOTS(NDFILE))
1785 		free(omap, M_FILEDESC);
1786 }
1787 
1788 /*
1789  * Allocate a file descriptor for the process.
1790  */
1791 int
1792 fdalloc(struct thread *td, int minfd, int *result)
1793 {
1794 	struct proc *p = td->td_proc;
1795 	struct filedesc *fdp = p->p_fd;
1796 	int fd, maxfd, allocfd;
1797 #ifdef RACCT
1798 	int error;
1799 #endif
1800 
1801 	FILEDESC_XLOCK_ASSERT(fdp);
1802 
1803 	if (fdp->fd_freefile > minfd)
1804 		minfd = fdp->fd_freefile;
1805 
1806 	maxfd = getmaxfd(td);
1807 
1808 	/*
1809 	 * Search the bitmap for a free descriptor starting at minfd.
1810 	 * If none is found, grow the file table.
1811 	 */
1812 	fd = fd_first_free(fdp, minfd, fdp->fd_nfiles);
1813 	if (fd >= maxfd)
1814 		return (EMFILE);
1815 	if (fd >= fdp->fd_nfiles) {
1816 		allocfd = min(fd * 2, maxfd);
1817 #ifdef RACCT
1818 		if (RACCT_ENABLED()) {
1819 			error = racct_set_unlocked(p, RACCT_NOFILE, allocfd);
1820 			if (error != 0)
1821 				return (EMFILE);
1822 		}
1823 #endif
1824 		/*
1825 		 * fd is already equal to first free descriptor >= minfd, so
1826 		 * we only need to grow the table and we are done.
1827 		 */
1828 		fdgrowtable_exp(fdp, allocfd);
1829 	}
1830 
1831 	/*
1832 	 * Perform some sanity checks, then mark the file descriptor as
1833 	 * used and return it to the caller.
1834 	 */
1835 	KASSERT(fd >= 0 && fd < min(maxfd, fdp->fd_nfiles),
1836 	    ("invalid descriptor %d", fd));
1837 	KASSERT(!fdisused(fdp, fd),
1838 	    ("fd_first_free() returned non-free descriptor"));
1839 	KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
1840 	    ("file descriptor isn't free"));
1841 	fdused(fdp, fd);
1842 	*result = fd;
1843 	return (0);
1844 }
1845 
1846 /*
1847  * Allocate n file descriptors for the process.
1848  */
1849 int
1850 fdallocn(struct thread *td, int minfd, int *fds, int n)
1851 {
1852 	struct proc *p = td->td_proc;
1853 	struct filedesc *fdp = p->p_fd;
1854 	int i;
1855 
1856 	FILEDESC_XLOCK_ASSERT(fdp);
1857 
1858 	for (i = 0; i < n; i++)
1859 		if (fdalloc(td, 0, &fds[i]) != 0)
1860 			break;
1861 
1862 	if (i < n) {
1863 		for (i--; i >= 0; i--)
1864 			fdunused(fdp, fds[i]);
1865 		return (EMFILE);
1866 	}
1867 
1868 	return (0);
1869 }
1870 
1871 /*
1872  * Create a new open file structure and allocate a file descriptor for the
1873  * process that refers to it.  We add one reference to the file for the
1874  * descriptor table and one reference for resultfp. This is to prevent us
1875  * being preempted and the entry in the descriptor table closed after we
1876  * release the FILEDESC lock.
1877  */
1878 int
1879 falloc_caps(struct thread *td, struct file **resultfp, int *resultfd, int flags,
1880     struct filecaps *fcaps)
1881 {
1882 	struct file *fp;
1883 	int error, fd;
1884 
1885 	error = falloc_noinstall(td, &fp);
1886 	if (error)
1887 		return (error);		/* no reference held on error */
1888 
1889 	error = finstall(td, fp, &fd, flags, fcaps);
1890 	if (error) {
1891 		fdrop(fp, td);		/* one reference (fp only) */
1892 		return (error);
1893 	}
1894 
1895 	if (resultfp != NULL)
1896 		*resultfp = fp;		/* copy out result */
1897 	else
1898 		fdrop(fp, td);		/* release local reference */
1899 
1900 	if (resultfd != NULL)
1901 		*resultfd = fd;
1902 
1903 	return (0);
1904 }
1905 
1906 /*
1907  * Create a new open file structure without allocating a file descriptor.
1908  */
1909 int
1910 falloc_noinstall(struct thread *td, struct file **resultfp)
1911 {
1912 	struct file *fp;
1913 	int maxuserfiles = maxfiles - (maxfiles / 20);
1914 	int openfiles_new;
1915 	static struct timeval lastfail;
1916 	static int curfail;
1917 
1918 	KASSERT(resultfp != NULL, ("%s: resultfp == NULL", __func__));
1919 
1920 	openfiles_new = atomic_fetchadd_int(&openfiles, 1) + 1;
1921 	if ((openfiles_new >= maxuserfiles &&
1922 	    priv_check(td, PRIV_MAXFILES) != 0) ||
1923 	    openfiles_new >= maxfiles) {
1924 		atomic_subtract_int(&openfiles, 1);
1925 		if (ppsratecheck(&lastfail, &curfail, 1)) {
1926 			printf("kern.maxfiles limit exceeded by uid %i, (%s) "
1927 			    "please see tuning(7).\n", td->td_ucred->cr_ruid, td->td_proc->p_comm);
1928 		}
1929 		return (ENFILE);
1930 	}
1931 	fp = uma_zalloc(file_zone, M_WAITOK);
1932 	bzero(fp, sizeof(*fp));
1933 	refcount_init(&fp->f_count, 1);
1934 	fp->f_cred = crhold(td->td_ucred);
1935 	fp->f_ops = &badfileops;
1936 	*resultfp = fp;
1937 	return (0);
1938 }
1939 
1940 /*
1941  * Install a file in a file descriptor table.
1942  */
1943 void
1944 _finstall(struct filedesc *fdp, struct file *fp, int fd, int flags,
1945     struct filecaps *fcaps)
1946 {
1947 	struct filedescent *fde;
1948 
1949 	MPASS(fp != NULL);
1950 	if (fcaps != NULL)
1951 		filecaps_validate(fcaps, __func__);
1952 	FILEDESC_XLOCK_ASSERT(fdp);
1953 
1954 	fde = &fdp->fd_ofiles[fd];
1955 #ifdef CAPABILITIES
1956 	seqc_write_begin(&fde->fde_seqc);
1957 #endif
1958 	fde->fde_file = fp;
1959 	fde->fde_flags = (flags & O_CLOEXEC) != 0 ? UF_EXCLOSE : 0;
1960 	if (fcaps != NULL)
1961 		filecaps_move(fcaps, &fde->fde_caps);
1962 	else
1963 		filecaps_fill(&fde->fde_caps);
1964 #ifdef CAPABILITIES
1965 	seqc_write_end(&fde->fde_seqc);
1966 #endif
1967 }
1968 
1969 int
1970 finstall(struct thread *td, struct file *fp, int *fd, int flags,
1971     struct filecaps *fcaps)
1972 {
1973 	struct filedesc *fdp = td->td_proc->p_fd;
1974 	int error;
1975 
1976 	MPASS(fd != NULL);
1977 
1978 	if (!fhold(fp))
1979 		return (EBADF);
1980 	FILEDESC_XLOCK(fdp);
1981 	if ((error = fdalloc(td, 0, fd))) {
1982 		FILEDESC_XUNLOCK(fdp);
1983 		fdrop(fp, td);
1984 		return (error);
1985 	}
1986 	_finstall(fdp, fp, *fd, flags, fcaps);
1987 	FILEDESC_XUNLOCK(fdp);
1988 	return (0);
1989 }
1990 
1991 /*
1992  * Build a new filedesc structure from another.
1993  * Copy the current, root, and jail root vnode references.
1994  *
1995  * If fdp is not NULL, return with it shared locked.
1996  */
1997 struct filedesc *
1998 fdinit(struct filedesc *fdp, bool prepfiles)
1999 {
2000 	struct filedesc0 *newfdp0;
2001 	struct filedesc *newfdp;
2002 
2003 	newfdp0 = uma_zalloc(filedesc0_zone, M_WAITOK | M_ZERO);
2004 	newfdp = &newfdp0->fd_fd;
2005 
2006 	/* Create the file descriptor table. */
2007 	FILEDESC_LOCK_INIT(newfdp);
2008 	refcount_init(&newfdp->fd_refcnt, 1);
2009 	refcount_init(&newfdp->fd_holdcnt, 1);
2010 	newfdp->fd_cmask = CMASK;
2011 	newfdp->fd_map = newfdp0->fd_dmap;
2012 	newfdp->fd_lastfile = -1;
2013 	newfdp->fd_files = (struct fdescenttbl *)&newfdp0->fd_dfiles;
2014 	newfdp->fd_files->fdt_nfiles = NDFILE;
2015 
2016 	if (fdp == NULL)
2017 		return (newfdp);
2018 
2019 	if (prepfiles && fdp->fd_lastfile >= newfdp->fd_nfiles)
2020 		fdgrowtable(newfdp, fdp->fd_lastfile + 1);
2021 
2022 	FILEDESC_SLOCK(fdp);
2023 	newfdp->fd_cdir = fdp->fd_cdir;
2024 	if (newfdp->fd_cdir)
2025 		vrefact(newfdp->fd_cdir);
2026 	newfdp->fd_rdir = fdp->fd_rdir;
2027 	if (newfdp->fd_rdir)
2028 		vrefact(newfdp->fd_rdir);
2029 	newfdp->fd_jdir = fdp->fd_jdir;
2030 	if (newfdp->fd_jdir)
2031 		vrefact(newfdp->fd_jdir);
2032 
2033 	if (!prepfiles) {
2034 		FILEDESC_SUNLOCK(fdp);
2035 	} else {
2036 		while (fdp->fd_lastfile >= newfdp->fd_nfiles) {
2037 			FILEDESC_SUNLOCK(fdp);
2038 			fdgrowtable(newfdp, fdp->fd_lastfile + 1);
2039 			FILEDESC_SLOCK(fdp);
2040 		}
2041 	}
2042 
2043 	return (newfdp);
2044 }
2045 
2046 static struct filedesc *
2047 fdhold(struct proc *p)
2048 {
2049 	struct filedesc *fdp;
2050 
2051 	PROC_LOCK_ASSERT(p, MA_OWNED);
2052 	fdp = p->p_fd;
2053 	if (fdp != NULL)
2054 		refcount_acquire(&fdp->fd_holdcnt);
2055 	return (fdp);
2056 }
2057 
2058 static void
2059 fddrop(struct filedesc *fdp)
2060 {
2061 
2062 	if (fdp->fd_holdcnt > 1) {
2063 		if (refcount_release(&fdp->fd_holdcnt) == 0)
2064 			return;
2065 	}
2066 
2067 	FILEDESC_LOCK_DESTROY(fdp);
2068 	uma_zfree(filedesc0_zone, fdp);
2069 }
2070 
2071 /*
2072  * Share a filedesc structure.
2073  */
2074 struct filedesc *
2075 fdshare(struct filedesc *fdp)
2076 {
2077 
2078 	refcount_acquire(&fdp->fd_refcnt);
2079 	return (fdp);
2080 }
2081 
2082 /*
2083  * Unshare a filedesc structure, if necessary by making a copy
2084  */
2085 void
2086 fdunshare(struct thread *td)
2087 {
2088 	struct filedesc *tmp;
2089 	struct proc *p = td->td_proc;
2090 
2091 	if (p->p_fd->fd_refcnt == 1)
2092 		return;
2093 
2094 	tmp = fdcopy(p->p_fd);
2095 	fdescfree(td);
2096 	p->p_fd = tmp;
2097 }
2098 
2099 void
2100 fdinstall_remapped(struct thread *td, struct filedesc *fdp)
2101 {
2102 
2103 	fdescfree(td);
2104 	td->td_proc->p_fd = fdp;
2105 }
2106 
2107 /*
2108  * Copy a filedesc structure.  A NULL pointer in returns a NULL reference,
2109  * this is to ease callers, not catch errors.
2110  */
2111 struct filedesc *
2112 fdcopy(struct filedesc *fdp)
2113 {
2114 	struct filedesc *newfdp;
2115 	struct filedescent *nfde, *ofde;
2116 	int i;
2117 
2118 	MPASS(fdp != NULL);
2119 
2120 	newfdp = fdinit(fdp, true);
2121 	/* copy all passable descriptors (i.e. not kqueue) */
2122 	newfdp->fd_freefile = -1;
2123 	for (i = 0; i <= fdp->fd_lastfile; ++i) {
2124 		ofde = &fdp->fd_ofiles[i];
2125 		if (ofde->fde_file == NULL ||
2126 		    (ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0 ||
2127 		    !fhold(ofde->fde_file)) {
2128 			if (newfdp->fd_freefile == -1)
2129 				newfdp->fd_freefile = i;
2130 			continue;
2131 		}
2132 		nfde = &newfdp->fd_ofiles[i];
2133 		*nfde = *ofde;
2134 		filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true);
2135 		fdused_init(newfdp, i);
2136 		newfdp->fd_lastfile = i;
2137 	}
2138 	if (newfdp->fd_freefile == -1)
2139 		newfdp->fd_freefile = i;
2140 	newfdp->fd_cmask = fdp->fd_cmask;
2141 	FILEDESC_SUNLOCK(fdp);
2142 	return (newfdp);
2143 }
2144 
2145 /*
2146  * Copies a filedesc structure, while remapping all file descriptors
2147  * stored inside using a translation table.
2148  *
2149  * File descriptors are copied over to the new file descriptor table,
2150  * regardless of whether the close-on-exec flag is set.
2151  */
2152 int
2153 fdcopy_remapped(struct filedesc *fdp, const int *fds, size_t nfds,
2154     struct filedesc **ret)
2155 {
2156 	struct filedesc *newfdp;
2157 	struct filedescent *nfde, *ofde;
2158 	int error, i;
2159 
2160 	MPASS(fdp != NULL);
2161 
2162 	newfdp = fdinit(fdp, true);
2163 	if (nfds > fdp->fd_lastfile + 1) {
2164 		/* New table cannot be larger than the old one. */
2165 		error = E2BIG;
2166 		goto bad;
2167 	}
2168 	/* Copy all passable descriptors (i.e. not kqueue). */
2169 	newfdp->fd_freefile = nfds;
2170 	for (i = 0; i < nfds; ++i) {
2171 		if (fds[i] < 0 || fds[i] > fdp->fd_lastfile) {
2172 			/* File descriptor out of bounds. */
2173 			error = EBADF;
2174 			goto bad;
2175 		}
2176 		ofde = &fdp->fd_ofiles[fds[i]];
2177 		if (ofde->fde_file == NULL) {
2178 			/* Unused file descriptor. */
2179 			error = EBADF;
2180 			goto bad;
2181 		}
2182 		if ((ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0) {
2183 			/* File descriptor cannot be passed. */
2184 			error = EINVAL;
2185 			goto bad;
2186 		}
2187 		if (!fhold(nfde->fde_file)) {
2188 			error = EBADF;
2189 			goto bad;
2190 		}
2191 		nfde = &newfdp->fd_ofiles[i];
2192 		*nfde = *ofde;
2193 		filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true);
2194 		fdused_init(newfdp, i);
2195 		newfdp->fd_lastfile = i;
2196 	}
2197 	newfdp->fd_cmask = fdp->fd_cmask;
2198 	FILEDESC_SUNLOCK(fdp);
2199 	*ret = newfdp;
2200 	return (0);
2201 bad:
2202 	FILEDESC_SUNLOCK(fdp);
2203 	fdescfree_remapped(newfdp);
2204 	return (error);
2205 }
2206 
2207 /*
2208  * Clear POSIX style locks. This is only used when fdp looses a reference (i.e.
2209  * one of processes using it exits) and the table used to be shared.
2210  */
2211 static void
2212 fdclearlocks(struct thread *td)
2213 {
2214 	struct filedesc *fdp;
2215 	struct filedesc_to_leader *fdtol;
2216 	struct flock lf;
2217 	struct file *fp;
2218 	struct proc *p;
2219 	struct vnode *vp;
2220 	int i;
2221 
2222 	p = td->td_proc;
2223 	fdp = p->p_fd;
2224 	fdtol = p->p_fdtol;
2225 	MPASS(fdtol != NULL);
2226 
2227 	FILEDESC_XLOCK(fdp);
2228 	KASSERT(fdtol->fdl_refcount > 0,
2229 	    ("filedesc_to_refcount botch: fdl_refcount=%d",
2230 	    fdtol->fdl_refcount));
2231 	if (fdtol->fdl_refcount == 1 &&
2232 	    (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2233 		for (i = 0; i <= fdp->fd_lastfile; i++) {
2234 			fp = fdp->fd_ofiles[i].fde_file;
2235 			if (fp == NULL || fp->f_type != DTYPE_VNODE ||
2236 			    !fhold(fp))
2237 				continue;
2238 			FILEDESC_XUNLOCK(fdp);
2239 			lf.l_whence = SEEK_SET;
2240 			lf.l_start = 0;
2241 			lf.l_len = 0;
2242 			lf.l_type = F_UNLCK;
2243 			vp = fp->f_vnode;
2244 			(void) VOP_ADVLOCK(vp,
2245 			    (caddr_t)p->p_leader, F_UNLCK,
2246 			    &lf, F_POSIX);
2247 			FILEDESC_XLOCK(fdp);
2248 			fdrop(fp, td);
2249 		}
2250 	}
2251 retry:
2252 	if (fdtol->fdl_refcount == 1) {
2253 		if (fdp->fd_holdleaderscount > 0 &&
2254 		    (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2255 			/*
2256 			 * close() or kern_dup() has cleared a reference
2257 			 * in a shared file descriptor table.
2258 			 */
2259 			fdp->fd_holdleaderswakeup = 1;
2260 			sx_sleep(&fdp->fd_holdleaderscount,
2261 			    FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0);
2262 			goto retry;
2263 		}
2264 		if (fdtol->fdl_holdcount > 0) {
2265 			/*
2266 			 * Ensure that fdtol->fdl_leader remains
2267 			 * valid in closef().
2268 			 */
2269 			fdtol->fdl_wakeup = 1;
2270 			sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK,
2271 			    "fdlhold", 0);
2272 			goto retry;
2273 		}
2274 	}
2275 	fdtol->fdl_refcount--;
2276 	if (fdtol->fdl_refcount == 0 &&
2277 	    fdtol->fdl_holdcount == 0) {
2278 		fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
2279 		fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
2280 	} else
2281 		fdtol = NULL;
2282 	p->p_fdtol = NULL;
2283 	FILEDESC_XUNLOCK(fdp);
2284 	if (fdtol != NULL)
2285 		free(fdtol, M_FILEDESC_TO_LEADER);
2286 }
2287 
2288 /*
2289  * Release a filedesc structure.
2290  */
2291 static void
2292 fdescfree_fds(struct thread *td, struct filedesc *fdp, bool needclose)
2293 {
2294 	struct filedesc0 *fdp0;
2295 	struct freetable *ft, *tft;
2296 	struct filedescent *fde;
2297 	struct file *fp;
2298 	int i;
2299 
2300 	for (i = 0; i <= fdp->fd_lastfile; i++) {
2301 		fde = &fdp->fd_ofiles[i];
2302 		fp = fde->fde_file;
2303 		if (fp != NULL) {
2304 			fdefree_last(fde);
2305 			if (needclose)
2306 				(void) closef(fp, td);
2307 			else
2308 				fdrop(fp, td);
2309 		}
2310 	}
2311 
2312 	if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE))
2313 		free(fdp->fd_map, M_FILEDESC);
2314 	if (fdp->fd_nfiles > NDFILE)
2315 		free(fdp->fd_files, M_FILEDESC);
2316 
2317 	fdp0 = (struct filedesc0 *)fdp;
2318 	SLIST_FOREACH_SAFE(ft, &fdp0->fd_free, ft_next, tft)
2319 		free(ft->ft_table, M_FILEDESC);
2320 
2321 	fddrop(fdp);
2322 }
2323 
2324 void
2325 fdescfree(struct thread *td)
2326 {
2327 	struct proc *p;
2328 	struct filedesc *fdp;
2329 	struct vnode *cdir, *jdir, *rdir;
2330 
2331 	p = td->td_proc;
2332 	fdp = p->p_fd;
2333 	MPASS(fdp != NULL);
2334 
2335 #ifdef RACCT
2336 	if (RACCT_ENABLED())
2337 		racct_set_unlocked(p, RACCT_NOFILE, 0);
2338 #endif
2339 
2340 	if (p->p_fdtol != NULL)
2341 		fdclearlocks(td);
2342 
2343 	PROC_LOCK(p);
2344 	p->p_fd = NULL;
2345 	PROC_UNLOCK(p);
2346 
2347 	if (refcount_release(&fdp->fd_refcnt) == 0)
2348 		return;
2349 
2350 	FILEDESC_XLOCK(fdp);
2351 	cdir = fdp->fd_cdir;
2352 	fdp->fd_cdir = NULL;
2353 	rdir = fdp->fd_rdir;
2354 	fdp->fd_rdir = NULL;
2355 	jdir = fdp->fd_jdir;
2356 	fdp->fd_jdir = NULL;
2357 	FILEDESC_XUNLOCK(fdp);
2358 
2359 	if (cdir != NULL)
2360 		vrele(cdir);
2361 	if (rdir != NULL)
2362 		vrele(rdir);
2363 	if (jdir != NULL)
2364 		vrele(jdir);
2365 
2366 	fdescfree_fds(td, fdp, 1);
2367 }
2368 
2369 void
2370 fdescfree_remapped(struct filedesc *fdp)
2371 {
2372 
2373 	if (fdp->fd_cdir != NULL)
2374 		vrele(fdp->fd_cdir);
2375 	if (fdp->fd_rdir != NULL)
2376 		vrele(fdp->fd_rdir);
2377 	if (fdp->fd_jdir != NULL)
2378 		vrele(fdp->fd_jdir);
2379 
2380 	fdescfree_fds(curthread, fdp, 0);
2381 }
2382 
2383 /*
2384  * For setugid programs, we don't want to people to use that setugidness
2385  * to generate error messages which write to a file which otherwise would
2386  * otherwise be off-limits to the process.  We check for filesystems where
2387  * the vnode can change out from under us after execve (like [lin]procfs).
2388  *
2389  * Since fdsetugidsafety calls this only for fd 0, 1 and 2, this check is
2390  * sufficient.  We also don't check for setugidness since we know we are.
2391  */
2392 static bool
2393 is_unsafe(struct file *fp)
2394 {
2395 	struct vnode *vp;
2396 
2397 	if (fp->f_type != DTYPE_VNODE)
2398 		return (false);
2399 
2400 	vp = fp->f_vnode;
2401 	return ((vp->v_vflag & VV_PROCDEP) != 0);
2402 }
2403 
2404 /*
2405  * Make this setguid thing safe, if at all possible.
2406  */
2407 void
2408 fdsetugidsafety(struct thread *td)
2409 {
2410 	struct filedesc *fdp;
2411 	struct file *fp;
2412 	int i;
2413 
2414 	fdp = td->td_proc->p_fd;
2415 	KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2416 	MPASS(fdp->fd_nfiles >= 3);
2417 	for (i = 0; i <= 2; i++) {
2418 		fp = fdp->fd_ofiles[i].fde_file;
2419 		if (fp != NULL && is_unsafe(fp)) {
2420 			FILEDESC_XLOCK(fdp);
2421 			knote_fdclose(td, i);
2422 			/*
2423 			 * NULL-out descriptor prior to close to avoid
2424 			 * a race while close blocks.
2425 			 */
2426 			fdfree(fdp, i);
2427 			FILEDESC_XUNLOCK(fdp);
2428 			(void) closef(fp, td);
2429 		}
2430 	}
2431 }
2432 
2433 /*
2434  * If a specific file object occupies a specific file descriptor, close the
2435  * file descriptor entry and drop a reference on the file object.  This is a
2436  * convenience function to handle a subsequent error in a function that calls
2437  * falloc() that handles the race that another thread might have closed the
2438  * file descriptor out from under the thread creating the file object.
2439  */
2440 void
2441 fdclose(struct thread *td, struct file *fp, int idx)
2442 {
2443 	struct filedesc *fdp = td->td_proc->p_fd;
2444 
2445 	FILEDESC_XLOCK(fdp);
2446 	if (fdp->fd_ofiles[idx].fde_file == fp) {
2447 		fdfree(fdp, idx);
2448 		FILEDESC_XUNLOCK(fdp);
2449 		fdrop(fp, td);
2450 	} else
2451 		FILEDESC_XUNLOCK(fdp);
2452 }
2453 
2454 /*
2455  * Close any files on exec?
2456  */
2457 void
2458 fdcloseexec(struct thread *td)
2459 {
2460 	struct filedesc *fdp;
2461 	struct filedescent *fde;
2462 	struct file *fp;
2463 	int i;
2464 
2465 	fdp = td->td_proc->p_fd;
2466 	KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2467 	for (i = 0; i <= fdp->fd_lastfile; i++) {
2468 		fde = &fdp->fd_ofiles[i];
2469 		fp = fde->fde_file;
2470 		if (fp != NULL && (fp->f_type == DTYPE_MQUEUE ||
2471 		    (fde->fde_flags & UF_EXCLOSE))) {
2472 			FILEDESC_XLOCK(fdp);
2473 			fdfree(fdp, i);
2474 			(void) closefp(fdp, i, fp, td, 0);
2475 			FILEDESC_UNLOCK_ASSERT(fdp);
2476 		}
2477 	}
2478 }
2479 
2480 /*
2481  * It is unsafe for set[ug]id processes to be started with file
2482  * descriptors 0..2 closed, as these descriptors are given implicit
2483  * significance in the Standard C library.  fdcheckstd() will create a
2484  * descriptor referencing /dev/null for each of stdin, stdout, and
2485  * stderr that is not already open.
2486  */
2487 int
2488 fdcheckstd(struct thread *td)
2489 {
2490 	struct filedesc *fdp;
2491 	register_t save;
2492 	int i, error, devnull;
2493 
2494 	fdp = td->td_proc->p_fd;
2495 	KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2496 	MPASS(fdp->fd_nfiles >= 3);
2497 	devnull = -1;
2498 	for (i = 0; i <= 2; i++) {
2499 		if (fdp->fd_ofiles[i].fde_file != NULL)
2500 			continue;
2501 
2502 		save = td->td_retval[0];
2503 		if (devnull != -1) {
2504 			error = kern_dup(td, FDDUP_FIXED, 0, devnull, i);
2505 		} else {
2506 			error = kern_openat(td, AT_FDCWD, "/dev/null",
2507 			    UIO_SYSSPACE, O_RDWR, 0);
2508 			if (error == 0) {
2509 				devnull = td->td_retval[0];
2510 				KASSERT(devnull == i, ("we didn't get our fd"));
2511 			}
2512 		}
2513 		td->td_retval[0] = save;
2514 		if (error != 0)
2515 			return (error);
2516 	}
2517 	return (0);
2518 }
2519 
2520 /*
2521  * Internal form of close.  Decrement reference count on file structure.
2522  * Note: td may be NULL when closing a file that was being passed in a
2523  * message.
2524  */
2525 int
2526 closef(struct file *fp, struct thread *td)
2527 {
2528 	struct vnode *vp;
2529 	struct flock lf;
2530 	struct filedesc_to_leader *fdtol;
2531 	struct filedesc *fdp;
2532 
2533 	/*
2534 	 * POSIX record locking dictates that any close releases ALL
2535 	 * locks owned by this process.  This is handled by setting
2536 	 * a flag in the unlock to free ONLY locks obeying POSIX
2537 	 * semantics, and not to free BSD-style file locks.
2538 	 * If the descriptor was in a message, POSIX-style locks
2539 	 * aren't passed with the descriptor, and the thread pointer
2540 	 * will be NULL.  Callers should be careful only to pass a
2541 	 * NULL thread pointer when there really is no owning
2542 	 * context that might have locks, or the locks will be
2543 	 * leaked.
2544 	 */
2545 	if (fp->f_type == DTYPE_VNODE && td != NULL) {
2546 		vp = fp->f_vnode;
2547 		if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
2548 			lf.l_whence = SEEK_SET;
2549 			lf.l_start = 0;
2550 			lf.l_len = 0;
2551 			lf.l_type = F_UNLCK;
2552 			(void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader,
2553 			    F_UNLCK, &lf, F_POSIX);
2554 		}
2555 		fdtol = td->td_proc->p_fdtol;
2556 		if (fdtol != NULL) {
2557 			/*
2558 			 * Handle special case where file descriptor table is
2559 			 * shared between multiple process leaders.
2560 			 */
2561 			fdp = td->td_proc->p_fd;
2562 			FILEDESC_XLOCK(fdp);
2563 			for (fdtol = fdtol->fdl_next;
2564 			    fdtol != td->td_proc->p_fdtol;
2565 			    fdtol = fdtol->fdl_next) {
2566 				if ((fdtol->fdl_leader->p_flag &
2567 				    P_ADVLOCK) == 0)
2568 					continue;
2569 				fdtol->fdl_holdcount++;
2570 				FILEDESC_XUNLOCK(fdp);
2571 				lf.l_whence = SEEK_SET;
2572 				lf.l_start = 0;
2573 				lf.l_len = 0;
2574 				lf.l_type = F_UNLCK;
2575 				vp = fp->f_vnode;
2576 				(void) VOP_ADVLOCK(vp,
2577 				    (caddr_t)fdtol->fdl_leader, F_UNLCK, &lf,
2578 				    F_POSIX);
2579 				FILEDESC_XLOCK(fdp);
2580 				fdtol->fdl_holdcount--;
2581 				if (fdtol->fdl_holdcount == 0 &&
2582 				    fdtol->fdl_wakeup != 0) {
2583 					fdtol->fdl_wakeup = 0;
2584 					wakeup(fdtol);
2585 				}
2586 			}
2587 			FILEDESC_XUNLOCK(fdp);
2588 		}
2589 	}
2590 	return (fdrop(fp, td));
2591 }
2592 
2593 /*
2594  * Initialize the file pointer with the specified properties.
2595  *
2596  * The ops are set with release semantics to be certain that the flags, type,
2597  * and data are visible when ops is.  This is to prevent ops methods from being
2598  * called with bad data.
2599  */
2600 void
2601 finit(struct file *fp, u_int flag, short type, void *data, struct fileops *ops)
2602 {
2603 	fp->f_data = data;
2604 	fp->f_flag = flag;
2605 	fp->f_type = type;
2606 	atomic_store_rel_ptr((volatile uintptr_t *)&fp->f_ops, (uintptr_t)ops);
2607 }
2608 
2609 int
2610 fget_cap_locked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
2611     struct file **fpp, struct filecaps *havecapsp)
2612 {
2613 	struct filedescent *fde;
2614 	int error;
2615 
2616 	FILEDESC_LOCK_ASSERT(fdp);
2617 
2618 	fde = fdeget_locked(fdp, fd);
2619 	if (fde == NULL) {
2620 		error = EBADF;
2621 		goto out;
2622 	}
2623 
2624 #ifdef CAPABILITIES
2625 	error = cap_check(cap_rights_fde_inline(fde), needrightsp);
2626 	if (error != 0)
2627 		goto out;
2628 #endif
2629 
2630 	if (havecapsp != NULL)
2631 		filecaps_copy(&fde->fde_caps, havecapsp, true);
2632 
2633 	*fpp = fde->fde_file;
2634 
2635 	error = 0;
2636 out:
2637 	return (error);
2638 }
2639 
2640 int
2641 fget_cap(struct thread *td, int fd, cap_rights_t *needrightsp,
2642     struct file **fpp, struct filecaps *havecapsp)
2643 {
2644 	struct filedesc *fdp = td->td_proc->p_fd;
2645 	int error;
2646 #ifndef CAPABILITIES
2647 	error = fget_unlocked(fdp, fd, needrightsp, fpp);
2648 	if (havecapsp != NULL && error == 0)
2649 		filecaps_fill(havecapsp);
2650 #else
2651 	struct file *fp;
2652 	seqc_t seq;
2653 
2654 	*fpp = NULL;
2655 	for (;;) {
2656 		error = fget_unlocked_seq(fdp, fd, needrightsp, &fp, &seq);
2657 		if (error != 0)
2658 			return (error);
2659 
2660 		if (havecapsp != NULL) {
2661 			if (!filecaps_copy(&fdp->fd_ofiles[fd].fde_caps,
2662 			    havecapsp, false)) {
2663 				fdrop(fp, td);
2664 				goto get_locked;
2665 			}
2666 		}
2667 
2668 		if (!fd_modified(fdp, fd, seq))
2669 			break;
2670 		fdrop(fp, td);
2671 	}
2672 
2673 	*fpp = fp;
2674 	return (0);
2675 
2676 get_locked:
2677 	FILEDESC_SLOCK(fdp);
2678 	error = fget_cap_locked(fdp, fd, needrightsp, fpp, havecapsp);
2679 	if (error == 0 && !fhold(*fpp))
2680 		error = EBADF;
2681 	FILEDESC_SUNLOCK(fdp);
2682 #endif
2683 	return (error);
2684 }
2685 
2686 int
2687 fget_unlocked_seq(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
2688     struct file **fpp, seqc_t *seqp)
2689 {
2690 #ifdef CAPABILITIES
2691 	const struct filedescent *fde;
2692 #endif
2693 	const struct fdescenttbl *fdt;
2694 	struct file *fp;
2695 #ifdef CAPABILITIES
2696 	seqc_t seq;
2697 	cap_rights_t haverights;
2698 	int error;
2699 #endif
2700 
2701 	fdt = fdp->fd_files;
2702 	if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
2703 		return (EBADF);
2704 	/*
2705 	 * Fetch the descriptor locklessly.  We avoid fdrop() races by
2706 	 * never raising a refcount above 0.  To accomplish this we have
2707 	 * to use a cmpset loop rather than an atomic_add.  The descriptor
2708 	 * must be re-verified once we acquire a reference to be certain
2709 	 * that the identity is still correct and we did not lose a race
2710 	 * due to preemption.
2711 	 */
2712 	for (;;) {
2713 #ifdef CAPABILITIES
2714 		seq = seqc_read(fd_seqc(fdt, fd));
2715 		fde = &fdt->fdt_ofiles[fd];
2716 		haverights = *cap_rights_fde_inline(fde);
2717 		fp = fde->fde_file;
2718 		if (!seqc_consistent(fd_seqc(fdt, fd), seq))
2719 			continue;
2720 #else
2721 		fp = fdt->fdt_ofiles[fd].fde_file;
2722 #endif
2723 		if (fp == NULL)
2724 			return (EBADF);
2725 #ifdef CAPABILITIES
2726 		error = cap_check_inline(&haverights, needrightsp);
2727 		if (error != 0)
2728 			return (error);
2729 #endif
2730 		if (__predict_false(!refcount_acquire_if_not_zero(&fp->f_count))) {
2731 			/*
2732 			 * The count was found either saturated or zero.
2733 			 * This re-read is not any more racy than using the
2734 			 * return value from fcmpset.
2735 			 */
2736 			if (fp->f_count != 0)
2737 				return (EBADF);
2738 			/*
2739 			 * Force a reload. Other thread could reallocate the
2740 			 * table before this fd was closed, so it is possible
2741 			 * that there is a stale fp pointer in cached version.
2742 			 */
2743 			fdt = (struct fdescenttbl *)atomic_load_ptr(&fdp->fd_files);
2744 			continue;
2745 		}
2746 		/*
2747 		 * Use an acquire barrier to force re-reading of fdt so it is
2748 		 * refreshed for verification.
2749 		 */
2750 		atomic_thread_fence_acq();
2751 		fdt = fdp->fd_files;
2752 #ifdef	CAPABILITIES
2753 		if (seqc_consistent_nomb(fd_seqc(fdt, fd), seq))
2754 #else
2755 		if (fp == fdt->fdt_ofiles[fd].fde_file)
2756 #endif
2757 			break;
2758 		fdrop(fp, curthread);
2759 	}
2760 	*fpp = fp;
2761 	if (seqp != NULL) {
2762 #ifdef CAPABILITIES
2763 		*seqp = seq;
2764 #endif
2765 	}
2766 	return (0);
2767 }
2768 
2769 /*
2770  * See the comments in fget_unlocked_seq for an explanation of how this works.
2771  *
2772  * This is a simplified variant which bails out to the aforementioned routine
2773  * if anything goes wrong. In practice this only happens when userspace is
2774  * racing with itself.
2775  */
2776 int
2777 fget_unlocked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
2778     struct file **fpp)
2779 {
2780 #ifdef CAPABILITIES
2781 	const struct filedescent *fde;
2782 #endif
2783 	const struct fdescenttbl *fdt;
2784 	struct file *fp;
2785 #ifdef CAPABILITIES
2786 	seqc_t seq;
2787 	const cap_rights_t *haverights;
2788 #endif
2789 
2790 	fdt = fdp->fd_files;
2791 	if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
2792 		return (EBADF);
2793 #ifdef CAPABILITIES
2794 	seq = seqc_read_any(fd_seqc(fdt, fd));
2795 	if (__predict_false(seqc_in_modify(seq)))
2796 		goto out_fallback;
2797 	fde = &fdt->fdt_ofiles[fd];
2798 	haverights = cap_rights_fde_inline(fde);
2799 	fp = fde->fde_file;
2800 #else
2801 	fp = fdt->fdt_ofiles[fd].fde_file;
2802 #endif
2803 	if (__predict_false(fp == NULL))
2804 		goto out_fallback;
2805 #ifdef CAPABILITIES
2806 	if (__predict_false(cap_check_inline_transient(haverights, needrightsp)))
2807 		goto out_fallback;
2808 #endif
2809 	if (__predict_false(!refcount_acquire_if_not_zero(&fp->f_count)))
2810 		goto out_fallback;
2811 
2812 	/*
2813 	 * Use an acquire barrier to force re-reading of fdt so it is
2814 	 * refreshed for verification.
2815 	 */
2816 	atomic_thread_fence_acq();
2817 	fdt = fdp->fd_files;
2818 #ifdef	CAPABILITIES
2819 	if (__predict_false(!seqc_consistent_nomb(fd_seqc(fdt, fd), seq)))
2820 #else
2821 	if (__predict_false(fp != fdt->fdt_ofiles[fd].fde_file))
2822 #endif
2823 		goto out_fdrop;
2824 	*fpp = fp;
2825 	return (0);
2826 out_fdrop:
2827 	fdrop(fp, curthread);
2828 out_fallback:
2829 	return (fget_unlocked_seq(fdp, fd, needrightsp, fpp, NULL));
2830 }
2831 
2832 /*
2833  * Extract the file pointer associated with the specified descriptor for the
2834  * current user process.
2835  *
2836  * If the descriptor doesn't exist or doesn't match 'flags', EBADF is
2837  * returned.
2838  *
2839  * File's rights will be checked against the capability rights mask.
2840  *
2841  * If an error occurred the non-zero error is returned and *fpp is set to
2842  * NULL.  Otherwise *fpp is held and set and zero is returned.  Caller is
2843  * responsible for fdrop().
2844  */
2845 static __inline int
2846 _fget(struct thread *td, int fd, struct file **fpp, int flags,
2847     cap_rights_t *needrightsp)
2848 {
2849 	struct filedesc *fdp;
2850 	struct file *fp;
2851 	int error;
2852 
2853 	*fpp = NULL;
2854 	fdp = td->td_proc->p_fd;
2855 	error = fget_unlocked(fdp, fd, needrightsp, &fp);
2856 	if (__predict_false(error != 0))
2857 		return (error);
2858 	if (__predict_false(fp->f_ops == &badfileops)) {
2859 		fdrop(fp, td);
2860 		return (EBADF);
2861 	}
2862 
2863 	/*
2864 	 * FREAD and FWRITE failure return EBADF as per POSIX.
2865 	 */
2866 	error = 0;
2867 	switch (flags) {
2868 	case FREAD:
2869 	case FWRITE:
2870 		if ((fp->f_flag & flags) == 0)
2871 			error = EBADF;
2872 		break;
2873 	case FEXEC:
2874 	    	if ((fp->f_flag & (FREAD | FEXEC)) == 0 ||
2875 		    ((fp->f_flag & FWRITE) != 0))
2876 			error = EBADF;
2877 		break;
2878 	case 0:
2879 		break;
2880 	default:
2881 		KASSERT(0, ("wrong flags"));
2882 	}
2883 
2884 	if (error != 0) {
2885 		fdrop(fp, td);
2886 		return (error);
2887 	}
2888 
2889 	*fpp = fp;
2890 	return (0);
2891 }
2892 
2893 int
2894 fget(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
2895 {
2896 
2897 	return (_fget(td, fd, fpp, 0, rightsp));
2898 }
2899 
2900 int
2901 fget_mmap(struct thread *td, int fd, cap_rights_t *rightsp, u_char *maxprotp,
2902     struct file **fpp)
2903 {
2904 	int error;
2905 #ifndef CAPABILITIES
2906 	error = _fget(td, fd, fpp, 0, rightsp);
2907 	if (maxprotp != NULL)
2908 		*maxprotp = VM_PROT_ALL;
2909 	return (error);
2910 #else
2911 	cap_rights_t fdrights;
2912 	struct filedesc *fdp;
2913 	struct file *fp;
2914 	seqc_t seq;
2915 
2916 	*fpp = NULL;
2917 	fdp = td->td_proc->p_fd;
2918 	MPASS(cap_rights_is_set(rightsp, CAP_MMAP));
2919 	for (;;) {
2920 		error = fget_unlocked_seq(fdp, fd, rightsp, &fp, &seq);
2921 		if (__predict_false(error != 0))
2922 			return (error);
2923 		if (__predict_false(fp->f_ops == &badfileops)) {
2924 			fdrop(fp, td);
2925 			return (EBADF);
2926 		}
2927 		if (maxprotp != NULL)
2928 			fdrights = *cap_rights(fdp, fd);
2929 		if (!fd_modified(fdp, fd, seq))
2930 			break;
2931 		fdrop(fp, td);
2932 	}
2933 
2934 	/*
2935 	 * If requested, convert capability rights to access flags.
2936 	 */
2937 	if (maxprotp != NULL)
2938 		*maxprotp = cap_rights_to_vmprot(&fdrights);
2939 	*fpp = fp;
2940 	return (0);
2941 #endif
2942 }
2943 
2944 int
2945 fget_read(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
2946 {
2947 
2948 	return (_fget(td, fd, fpp, FREAD, rightsp));
2949 }
2950 
2951 int
2952 fget_write(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
2953 {
2954 
2955 	return (_fget(td, fd, fpp, FWRITE, rightsp));
2956 }
2957 
2958 int
2959 fget_fcntl(struct thread *td, int fd, cap_rights_t *rightsp, int needfcntl,
2960     struct file **fpp)
2961 {
2962 	struct filedesc *fdp = td->td_proc->p_fd;
2963 #ifndef CAPABILITIES
2964 	return (fget_unlocked(fdp, fd, rightsp, fpp));
2965 #else
2966 	struct file *fp;
2967 	int error;
2968 	seqc_t seq;
2969 
2970 	*fpp = NULL;
2971 	MPASS(cap_rights_is_set(rightsp, CAP_FCNTL));
2972 	for (;;) {
2973 		error = fget_unlocked_seq(fdp, fd, rightsp, &fp, &seq);
2974 		if (error != 0)
2975 			return (error);
2976 		error = cap_fcntl_check(fdp, fd, needfcntl);
2977 		if (!fd_modified(fdp, fd, seq))
2978 			break;
2979 		fdrop(fp, td);
2980 	}
2981 	if (error != 0) {
2982 		fdrop(fp, td);
2983 		return (error);
2984 	}
2985 	*fpp = fp;
2986 	return (0);
2987 #endif
2988 }
2989 
2990 /*
2991  * Like fget() but loads the underlying vnode, or returns an error if the
2992  * descriptor does not represent a vnode.  Note that pipes use vnodes but
2993  * never have VM objects.  The returned vnode will be vref()'d.
2994  *
2995  * XXX: what about the unused flags ?
2996  */
2997 static __inline int
2998 _fgetvp(struct thread *td, int fd, int flags, cap_rights_t *needrightsp,
2999     struct vnode **vpp)
3000 {
3001 	struct file *fp;
3002 	int error;
3003 
3004 	*vpp = NULL;
3005 	error = _fget(td, fd, &fp, flags, needrightsp);
3006 	if (error != 0)
3007 		return (error);
3008 	if (fp->f_vnode == NULL) {
3009 		error = EINVAL;
3010 	} else {
3011 		*vpp = fp->f_vnode;
3012 		vrefact(*vpp);
3013 	}
3014 	fdrop(fp, td);
3015 
3016 	return (error);
3017 }
3018 
3019 int
3020 fgetvp(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
3021 {
3022 
3023 	return (_fgetvp(td, fd, 0, rightsp, vpp));
3024 }
3025 
3026 int
3027 fgetvp_rights(struct thread *td, int fd, cap_rights_t *needrightsp,
3028     struct filecaps *havecaps, struct vnode **vpp)
3029 {
3030 	struct filedesc *fdp;
3031 	struct filecaps caps;
3032 	struct file *fp;
3033 	int error;
3034 
3035 	fdp = td->td_proc->p_fd;
3036 	error = fget_cap_locked(fdp, fd, needrightsp, &fp, &caps);
3037 	if (error != 0)
3038 		return (error);
3039 	if (fp->f_ops == &badfileops) {
3040 		error = EBADF;
3041 		goto out;
3042 	}
3043 	if (fp->f_vnode == NULL) {
3044 		error = EINVAL;
3045 		goto out;
3046 	}
3047 
3048 	*havecaps = caps;
3049 	*vpp = fp->f_vnode;
3050 	vrefact(*vpp);
3051 
3052 	return (0);
3053 out:
3054 	filecaps_free(&caps);
3055 	return (error);
3056 }
3057 
3058 int
3059 fgetvp_read(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
3060 {
3061 
3062 	return (_fgetvp(td, fd, FREAD, rightsp, vpp));
3063 }
3064 
3065 int
3066 fgetvp_exec(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
3067 {
3068 
3069 	return (_fgetvp(td, fd, FEXEC, rightsp, vpp));
3070 }
3071 
3072 #ifdef notyet
3073 int
3074 fgetvp_write(struct thread *td, int fd, cap_rights_t *rightsp,
3075     struct vnode **vpp)
3076 {
3077 
3078 	return (_fgetvp(td, fd, FWRITE, rightsp, vpp));
3079 }
3080 #endif
3081 
3082 /*
3083  * Handle the last reference to a file being closed.
3084  *
3085  * Without the noinline attribute clang keeps inlining the func thorough this
3086  * file when fdrop is used.
3087  */
3088 int __noinline
3089 _fdrop(struct file *fp, struct thread *td)
3090 {
3091 	int error;
3092 
3093 	if (fp->f_count != 0)
3094 		panic("fdrop: count %d", fp->f_count);
3095 	error = fo_close(fp, td);
3096 	atomic_subtract_int(&openfiles, 1);
3097 	crfree(fp->f_cred);
3098 	free(fp->f_advice, M_FADVISE);
3099 	uma_zfree(file_zone, fp);
3100 
3101 	return (error);
3102 }
3103 
3104 /*
3105  * Apply an advisory lock on a file descriptor.
3106  *
3107  * Just attempt to get a record lock of the requested type on the entire file
3108  * (l_whence = SEEK_SET, l_start = 0, l_len = 0).
3109  */
3110 #ifndef _SYS_SYSPROTO_H_
3111 struct flock_args {
3112 	int	fd;
3113 	int	how;
3114 };
3115 #endif
3116 /* ARGSUSED */
3117 int
3118 sys_flock(struct thread *td, struct flock_args *uap)
3119 {
3120 	struct file *fp;
3121 	struct vnode *vp;
3122 	struct flock lf;
3123 	int error;
3124 
3125 	error = fget(td, uap->fd, &cap_flock_rights, &fp);
3126 	if (error != 0)
3127 		return (error);
3128 	if (fp->f_type != DTYPE_VNODE) {
3129 		fdrop(fp, td);
3130 		return (EOPNOTSUPP);
3131 	}
3132 
3133 	vp = fp->f_vnode;
3134 	lf.l_whence = SEEK_SET;
3135 	lf.l_start = 0;
3136 	lf.l_len = 0;
3137 	if (uap->how & LOCK_UN) {
3138 		lf.l_type = F_UNLCK;
3139 		atomic_clear_int(&fp->f_flag, FHASLOCK);
3140 		error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK);
3141 		goto done2;
3142 	}
3143 	if (uap->how & LOCK_EX)
3144 		lf.l_type = F_WRLCK;
3145 	else if (uap->how & LOCK_SH)
3146 		lf.l_type = F_RDLCK;
3147 	else {
3148 		error = EBADF;
3149 		goto done2;
3150 	}
3151 	atomic_set_int(&fp->f_flag, FHASLOCK);
3152 	error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf,
3153 	    (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT);
3154 done2:
3155 	fdrop(fp, td);
3156 	return (error);
3157 }
3158 /*
3159  * Duplicate the specified descriptor to a free descriptor.
3160  */
3161 int
3162 dupfdopen(struct thread *td, struct filedesc *fdp, int dfd, int mode,
3163     int openerror, int *indxp)
3164 {
3165 	struct filedescent *newfde, *oldfde;
3166 	struct file *fp;
3167 	u_long *ioctls;
3168 	int error, indx;
3169 
3170 	KASSERT(openerror == ENODEV || openerror == ENXIO,
3171 	    ("unexpected error %d in %s", openerror, __func__));
3172 
3173 	/*
3174 	 * If the to-be-dup'd fd number is greater than the allowed number
3175 	 * of file descriptors, or the fd to be dup'd has already been
3176 	 * closed, then reject.
3177 	 */
3178 	FILEDESC_XLOCK(fdp);
3179 	if ((fp = fget_locked(fdp, dfd)) == NULL) {
3180 		FILEDESC_XUNLOCK(fdp);
3181 		return (EBADF);
3182 	}
3183 
3184 	error = fdalloc(td, 0, &indx);
3185 	if (error != 0) {
3186 		FILEDESC_XUNLOCK(fdp);
3187 		return (error);
3188 	}
3189 
3190 	/*
3191 	 * There are two cases of interest here.
3192 	 *
3193 	 * For ENODEV simply dup (dfd) to file descriptor (indx) and return.
3194 	 *
3195 	 * For ENXIO steal away the file structure from (dfd) and store it in
3196 	 * (indx).  (dfd) is effectively closed by this operation.
3197 	 */
3198 	switch (openerror) {
3199 	case ENODEV:
3200 		/*
3201 		 * Check that the mode the file is being opened for is a
3202 		 * subset of the mode of the existing descriptor.
3203 		 */
3204 		if (((mode & (FREAD|FWRITE)) | fp->f_flag) != fp->f_flag) {
3205 			fdunused(fdp, indx);
3206 			FILEDESC_XUNLOCK(fdp);
3207 			return (EACCES);
3208 		}
3209 		if (!fhold(fp)) {
3210 			fdunused(fdp, indx);
3211 			FILEDESC_XUNLOCK(fdp);
3212 			return (EBADF);
3213 		}
3214 		newfde = &fdp->fd_ofiles[indx];
3215 		oldfde = &fdp->fd_ofiles[dfd];
3216 		ioctls = filecaps_copy_prep(&oldfde->fde_caps);
3217 #ifdef CAPABILITIES
3218 		seqc_write_begin(&newfde->fde_seqc);
3219 #endif
3220 		memcpy(newfde, oldfde, fde_change_size);
3221 		filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps,
3222 		    ioctls);
3223 #ifdef CAPABILITIES
3224 		seqc_write_end(&newfde->fde_seqc);
3225 #endif
3226 		break;
3227 	case ENXIO:
3228 		/*
3229 		 * Steal away the file pointer from dfd and stuff it into indx.
3230 		 */
3231 		newfde = &fdp->fd_ofiles[indx];
3232 		oldfde = &fdp->fd_ofiles[dfd];
3233 #ifdef CAPABILITIES
3234 		seqc_write_begin(&newfde->fde_seqc);
3235 #endif
3236 		memcpy(newfde, oldfde, fde_change_size);
3237 		oldfde->fde_file = NULL;
3238 		fdunused(fdp, dfd);
3239 #ifdef CAPABILITIES
3240 		seqc_write_end(&newfde->fde_seqc);
3241 #endif
3242 		break;
3243 	}
3244 	FILEDESC_XUNLOCK(fdp);
3245 	*indxp = indx;
3246 	return (0);
3247 }
3248 
3249 /*
3250  * This sysctl determines if we will allow a process to chroot(2) if it
3251  * has a directory open:
3252  *	0: disallowed for all processes.
3253  *	1: allowed for processes that were not already chroot(2)'ed.
3254  *	2: allowed for all processes.
3255  */
3256 
3257 static int chroot_allow_open_directories = 1;
3258 
3259 SYSCTL_INT(_kern, OID_AUTO, chroot_allow_open_directories, CTLFLAG_RW,
3260     &chroot_allow_open_directories, 0,
3261     "Allow a process to chroot(2) if it has a directory open");
3262 
3263 /*
3264  * Helper function for raised chroot(2) security function:  Refuse if
3265  * any filedescriptors are open directories.
3266  */
3267 static int
3268 chroot_refuse_vdir_fds(struct filedesc *fdp)
3269 {
3270 	struct vnode *vp;
3271 	struct file *fp;
3272 	int fd;
3273 
3274 	FILEDESC_LOCK_ASSERT(fdp);
3275 
3276 	for (fd = 0; fd <= fdp->fd_lastfile; fd++) {
3277 		fp = fget_locked(fdp, fd);
3278 		if (fp == NULL)
3279 			continue;
3280 		if (fp->f_type == DTYPE_VNODE) {
3281 			vp = fp->f_vnode;
3282 			if (vp->v_type == VDIR)
3283 				return (EPERM);
3284 		}
3285 	}
3286 	return (0);
3287 }
3288 
3289 /*
3290  * Common routine for kern_chroot() and jail_attach().  The caller is
3291  * responsible for invoking priv_check() and mac_vnode_check_chroot() to
3292  * authorize this operation.
3293  */
3294 int
3295 pwd_chroot(struct thread *td, struct vnode *vp)
3296 {
3297 	struct filedesc *fdp;
3298 	struct vnode *oldvp;
3299 	int error;
3300 
3301 	fdp = td->td_proc->p_fd;
3302 	FILEDESC_XLOCK(fdp);
3303 	if (chroot_allow_open_directories == 0 ||
3304 	    (chroot_allow_open_directories == 1 && fdp->fd_rdir != rootvnode)) {
3305 		error = chroot_refuse_vdir_fds(fdp);
3306 		if (error != 0) {
3307 			FILEDESC_XUNLOCK(fdp);
3308 			return (error);
3309 		}
3310 	}
3311 	oldvp = fdp->fd_rdir;
3312 	vrefact(vp);
3313 	fdp->fd_rdir = vp;
3314 	if (fdp->fd_jdir == NULL) {
3315 		vrefact(vp);
3316 		fdp->fd_jdir = vp;
3317 	}
3318 	FILEDESC_XUNLOCK(fdp);
3319 	vrele(oldvp);
3320 	return (0);
3321 }
3322 
3323 void
3324 pwd_chdir(struct thread *td, struct vnode *vp)
3325 {
3326 	struct filedesc *fdp;
3327 	struct vnode *oldvp;
3328 
3329 	fdp = td->td_proc->p_fd;
3330 	FILEDESC_XLOCK(fdp);
3331 	VNASSERT(vp->v_usecount > 0, vp,
3332 	    ("chdir to a vnode with zero usecount"));
3333 	oldvp = fdp->fd_cdir;
3334 	fdp->fd_cdir = vp;
3335 	FILEDESC_XUNLOCK(fdp);
3336 	vrele(oldvp);
3337 }
3338 
3339 /*
3340  * Scan all active processes and prisons to see if any of them have a current
3341  * or root directory of `olddp'. If so, replace them with the new mount point.
3342  */
3343 void
3344 mountcheckdirs(struct vnode *olddp, struct vnode *newdp)
3345 {
3346 	struct filedesc *fdp;
3347 	struct prison *pr;
3348 	struct proc *p;
3349 	int nrele;
3350 
3351 	if (vrefcnt(olddp) == 1)
3352 		return;
3353 	nrele = 0;
3354 	sx_slock(&allproc_lock);
3355 	FOREACH_PROC_IN_SYSTEM(p) {
3356 		PROC_LOCK(p);
3357 		fdp = fdhold(p);
3358 		PROC_UNLOCK(p);
3359 		if (fdp == NULL)
3360 			continue;
3361 		FILEDESC_XLOCK(fdp);
3362 		if (fdp->fd_cdir == olddp) {
3363 			vrefact(newdp);
3364 			fdp->fd_cdir = newdp;
3365 			nrele++;
3366 		}
3367 		if (fdp->fd_rdir == olddp) {
3368 			vrefact(newdp);
3369 			fdp->fd_rdir = newdp;
3370 			nrele++;
3371 		}
3372 		if (fdp->fd_jdir == olddp) {
3373 			vrefact(newdp);
3374 			fdp->fd_jdir = newdp;
3375 			nrele++;
3376 		}
3377 		FILEDESC_XUNLOCK(fdp);
3378 		fddrop(fdp);
3379 	}
3380 	sx_sunlock(&allproc_lock);
3381 	if (rootvnode == olddp) {
3382 		vrefact(newdp);
3383 		rootvnode = newdp;
3384 		nrele++;
3385 	}
3386 	mtx_lock(&prison0.pr_mtx);
3387 	if (prison0.pr_root == olddp) {
3388 		vrefact(newdp);
3389 		prison0.pr_root = newdp;
3390 		nrele++;
3391 	}
3392 	mtx_unlock(&prison0.pr_mtx);
3393 	sx_slock(&allprison_lock);
3394 	TAILQ_FOREACH(pr, &allprison, pr_list) {
3395 		mtx_lock(&pr->pr_mtx);
3396 		if (pr->pr_root == olddp) {
3397 			vrefact(newdp);
3398 			pr->pr_root = newdp;
3399 			nrele++;
3400 		}
3401 		mtx_unlock(&pr->pr_mtx);
3402 	}
3403 	sx_sunlock(&allprison_lock);
3404 	while (nrele--)
3405 		vrele(olddp);
3406 }
3407 
3408 struct filedesc_to_leader *
3409 filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader)
3410 {
3411 	struct filedesc_to_leader *fdtol;
3412 
3413 	fdtol = malloc(sizeof(struct filedesc_to_leader),
3414 	    M_FILEDESC_TO_LEADER, M_WAITOK);
3415 	fdtol->fdl_refcount = 1;
3416 	fdtol->fdl_holdcount = 0;
3417 	fdtol->fdl_wakeup = 0;
3418 	fdtol->fdl_leader = leader;
3419 	if (old != NULL) {
3420 		FILEDESC_XLOCK(fdp);
3421 		fdtol->fdl_next = old->fdl_next;
3422 		fdtol->fdl_prev = old;
3423 		old->fdl_next = fdtol;
3424 		fdtol->fdl_next->fdl_prev = fdtol;
3425 		FILEDESC_XUNLOCK(fdp);
3426 	} else {
3427 		fdtol->fdl_next = fdtol;
3428 		fdtol->fdl_prev = fdtol;
3429 	}
3430 	return (fdtol);
3431 }
3432 
3433 static int
3434 sysctl_kern_proc_nfds(SYSCTL_HANDLER_ARGS)
3435 {
3436 	struct filedesc *fdp;
3437 	int i, count, slots;
3438 
3439 	if (*(int *)arg1 != 0)
3440 		return (EINVAL);
3441 
3442 	fdp = curproc->p_fd;
3443 	count = 0;
3444 	FILEDESC_SLOCK(fdp);
3445 	slots = NDSLOTS(fdp->fd_lastfile + 1);
3446 	for (i = 0; i < slots; i++)
3447 		count += bitcountl(fdp->fd_map[i]);
3448 	FILEDESC_SUNLOCK(fdp);
3449 
3450 	return (SYSCTL_OUT(req, &count, sizeof(count)));
3451 }
3452 
3453 static SYSCTL_NODE(_kern_proc, KERN_PROC_NFDS, nfds,
3454     CTLFLAG_RD|CTLFLAG_CAPRD|CTLFLAG_MPSAFE, sysctl_kern_proc_nfds,
3455     "Number of open file descriptors");
3456 
3457 /*
3458  * Get file structures globally.
3459  */
3460 static int
3461 sysctl_kern_file(SYSCTL_HANDLER_ARGS)
3462 {
3463 	struct xfile xf;
3464 	struct filedesc *fdp;
3465 	struct file *fp;
3466 	struct proc *p;
3467 	int error, n;
3468 
3469 	error = sysctl_wire_old_buffer(req, 0);
3470 	if (error != 0)
3471 		return (error);
3472 	if (req->oldptr == NULL) {
3473 		n = 0;
3474 		sx_slock(&allproc_lock);
3475 		FOREACH_PROC_IN_SYSTEM(p) {
3476 			PROC_LOCK(p);
3477 			if (p->p_state == PRS_NEW) {
3478 				PROC_UNLOCK(p);
3479 				continue;
3480 			}
3481 			fdp = fdhold(p);
3482 			PROC_UNLOCK(p);
3483 			if (fdp == NULL)
3484 				continue;
3485 			/* overestimates sparse tables. */
3486 			if (fdp->fd_lastfile > 0)
3487 				n += fdp->fd_lastfile;
3488 			fddrop(fdp);
3489 		}
3490 		sx_sunlock(&allproc_lock);
3491 		return (SYSCTL_OUT(req, 0, n * sizeof(xf)));
3492 	}
3493 	error = 0;
3494 	bzero(&xf, sizeof(xf));
3495 	xf.xf_size = sizeof(xf);
3496 	sx_slock(&allproc_lock);
3497 	FOREACH_PROC_IN_SYSTEM(p) {
3498 		PROC_LOCK(p);
3499 		if (p->p_state == PRS_NEW) {
3500 			PROC_UNLOCK(p);
3501 			continue;
3502 		}
3503 		if (p_cansee(req->td, p) != 0) {
3504 			PROC_UNLOCK(p);
3505 			continue;
3506 		}
3507 		xf.xf_pid = p->p_pid;
3508 		xf.xf_uid = p->p_ucred->cr_uid;
3509 		fdp = fdhold(p);
3510 		PROC_UNLOCK(p);
3511 		if (fdp == NULL)
3512 			continue;
3513 		FILEDESC_SLOCK(fdp);
3514 		for (n = 0; fdp->fd_refcnt > 0 && n <= fdp->fd_lastfile; ++n) {
3515 			if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
3516 				continue;
3517 			xf.xf_fd = n;
3518 			xf.xf_file = (uintptr_t)fp;
3519 			xf.xf_data = (uintptr_t)fp->f_data;
3520 			xf.xf_vnode = (uintptr_t)fp->f_vnode;
3521 			xf.xf_type = (uintptr_t)fp->f_type;
3522 			xf.xf_count = fp->f_count;
3523 			xf.xf_msgcount = 0;
3524 			xf.xf_offset = foffset_get(fp);
3525 			xf.xf_flag = fp->f_flag;
3526 			error = SYSCTL_OUT(req, &xf, sizeof(xf));
3527 			if (error)
3528 				break;
3529 		}
3530 		FILEDESC_SUNLOCK(fdp);
3531 		fddrop(fdp);
3532 		if (error)
3533 			break;
3534 	}
3535 	sx_sunlock(&allproc_lock);
3536 	return (error);
3537 }
3538 
3539 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD|CTLFLAG_MPSAFE,
3540     0, 0, sysctl_kern_file, "S,xfile", "Entire file table");
3541 
3542 #ifdef KINFO_FILE_SIZE
3543 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
3544 #endif
3545 
3546 static int
3547 xlate_fflags(int fflags)
3548 {
3549 	static const struct {
3550 		int	fflag;
3551 		int	kf_fflag;
3552 	} fflags_table[] = {
3553 		{ FAPPEND, KF_FLAG_APPEND },
3554 		{ FASYNC, KF_FLAG_ASYNC },
3555 		{ FFSYNC, KF_FLAG_FSYNC },
3556 		{ FHASLOCK, KF_FLAG_HASLOCK },
3557 		{ FNONBLOCK, KF_FLAG_NONBLOCK },
3558 		{ FREAD, KF_FLAG_READ },
3559 		{ FWRITE, KF_FLAG_WRITE },
3560 		{ O_CREAT, KF_FLAG_CREAT },
3561 		{ O_DIRECT, KF_FLAG_DIRECT },
3562 		{ O_EXCL, KF_FLAG_EXCL },
3563 		{ O_EXEC, KF_FLAG_EXEC },
3564 		{ O_EXLOCK, KF_FLAG_EXLOCK },
3565 		{ O_NOFOLLOW, KF_FLAG_NOFOLLOW },
3566 		{ O_SHLOCK, KF_FLAG_SHLOCK },
3567 		{ O_TRUNC, KF_FLAG_TRUNC }
3568 	};
3569 	unsigned int i;
3570 	int kflags;
3571 
3572 	kflags = 0;
3573 	for (i = 0; i < nitems(fflags_table); i++)
3574 		if (fflags & fflags_table[i].fflag)
3575 			kflags |=  fflags_table[i].kf_fflag;
3576 	return (kflags);
3577 }
3578 
3579 /* Trim unused data from kf_path by truncating the structure size. */
3580 void
3581 pack_kinfo(struct kinfo_file *kif)
3582 {
3583 
3584 	kif->kf_structsize = offsetof(struct kinfo_file, kf_path) +
3585 	    strlen(kif->kf_path) + 1;
3586 	kif->kf_structsize = roundup(kif->kf_structsize, sizeof(uint64_t));
3587 }
3588 
3589 static void
3590 export_file_to_kinfo(struct file *fp, int fd, cap_rights_t *rightsp,
3591     struct kinfo_file *kif, struct filedesc *fdp, int flags)
3592 {
3593 	int error;
3594 
3595 	bzero(kif, sizeof(*kif));
3596 
3597 	/* Set a default type to allow for empty fill_kinfo() methods. */
3598 	kif->kf_type = KF_TYPE_UNKNOWN;
3599 	kif->kf_flags = xlate_fflags(fp->f_flag);
3600 	if (rightsp != NULL)
3601 		kif->kf_cap_rights = *rightsp;
3602 	else
3603 		cap_rights_init(&kif->kf_cap_rights);
3604 	kif->kf_fd = fd;
3605 	kif->kf_ref_count = fp->f_count;
3606 	kif->kf_offset = foffset_get(fp);
3607 
3608 	/*
3609 	 * This may drop the filedesc lock, so the 'fp' cannot be
3610 	 * accessed after this call.
3611 	 */
3612 	error = fo_fill_kinfo(fp, kif, fdp);
3613 	if (error == 0)
3614 		kif->kf_status |= KF_ATTR_VALID;
3615 	if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
3616 		pack_kinfo(kif);
3617 	else
3618 		kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
3619 }
3620 
3621 static void
3622 export_vnode_to_kinfo(struct vnode *vp, int fd, int fflags,
3623     struct kinfo_file *kif, int flags)
3624 {
3625 	int error;
3626 
3627 	bzero(kif, sizeof(*kif));
3628 
3629 	kif->kf_type = KF_TYPE_VNODE;
3630 	error = vn_fill_kinfo_vnode(vp, kif);
3631 	if (error == 0)
3632 		kif->kf_status |= KF_ATTR_VALID;
3633 	kif->kf_flags = xlate_fflags(fflags);
3634 	cap_rights_init(&kif->kf_cap_rights);
3635 	kif->kf_fd = fd;
3636 	kif->kf_ref_count = -1;
3637 	kif->kf_offset = -1;
3638 	if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
3639 		pack_kinfo(kif);
3640 	else
3641 		kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
3642 	vrele(vp);
3643 }
3644 
3645 struct export_fd_buf {
3646 	struct filedesc		*fdp;
3647 	struct sbuf 		*sb;
3648 	ssize_t			remainder;
3649 	struct kinfo_file	kif;
3650 	int			flags;
3651 };
3652 
3653 static int
3654 export_kinfo_to_sb(struct export_fd_buf *efbuf)
3655 {
3656 	struct kinfo_file *kif;
3657 
3658 	kif = &efbuf->kif;
3659 	if (efbuf->remainder != -1) {
3660 		if (efbuf->remainder < kif->kf_structsize) {
3661 			/* Terminate export. */
3662 			efbuf->remainder = 0;
3663 			return (0);
3664 		}
3665 		efbuf->remainder -= kif->kf_structsize;
3666 	}
3667 	return (sbuf_bcat(efbuf->sb, kif, kif->kf_structsize) == 0 ? 0 : ENOMEM);
3668 }
3669 
3670 static int
3671 export_file_to_sb(struct file *fp, int fd, cap_rights_t *rightsp,
3672     struct export_fd_buf *efbuf)
3673 {
3674 	int error;
3675 
3676 	if (efbuf->remainder == 0)
3677 		return (0);
3678 	export_file_to_kinfo(fp, fd, rightsp, &efbuf->kif, efbuf->fdp,
3679 	    efbuf->flags);
3680 	FILEDESC_SUNLOCK(efbuf->fdp);
3681 	error = export_kinfo_to_sb(efbuf);
3682 	FILEDESC_SLOCK(efbuf->fdp);
3683 	return (error);
3684 }
3685 
3686 static int
3687 export_vnode_to_sb(struct vnode *vp, int fd, int fflags,
3688     struct export_fd_buf *efbuf)
3689 {
3690 	int error;
3691 
3692 	if (efbuf->remainder == 0)
3693 		return (0);
3694 	if (efbuf->fdp != NULL)
3695 		FILEDESC_SUNLOCK(efbuf->fdp);
3696 	export_vnode_to_kinfo(vp, fd, fflags, &efbuf->kif, efbuf->flags);
3697 	error = export_kinfo_to_sb(efbuf);
3698 	if (efbuf->fdp != NULL)
3699 		FILEDESC_SLOCK(efbuf->fdp);
3700 	return (error);
3701 }
3702 
3703 /*
3704  * Store a process file descriptor information to sbuf.
3705  *
3706  * Takes a locked proc as argument, and returns with the proc unlocked.
3707  */
3708 int
3709 kern_proc_filedesc_out(struct proc *p,  struct sbuf *sb, ssize_t maxlen,
3710     int flags)
3711 {
3712 	struct file *fp;
3713 	struct filedesc *fdp;
3714 	struct export_fd_buf *efbuf;
3715 	struct vnode *cttyvp, *textvp, *tracevp;
3716 	int error, i;
3717 	cap_rights_t rights;
3718 
3719 	PROC_LOCK_ASSERT(p, MA_OWNED);
3720 
3721 	/* ktrace vnode */
3722 	tracevp = p->p_tracevp;
3723 	if (tracevp != NULL)
3724 		vrefact(tracevp);
3725 	/* text vnode */
3726 	textvp = p->p_textvp;
3727 	if (textvp != NULL)
3728 		vrefact(textvp);
3729 	/* Controlling tty. */
3730 	cttyvp = NULL;
3731 	if (p->p_pgrp != NULL && p->p_pgrp->pg_session != NULL) {
3732 		cttyvp = p->p_pgrp->pg_session->s_ttyvp;
3733 		if (cttyvp != NULL)
3734 			vrefact(cttyvp);
3735 	}
3736 	fdp = fdhold(p);
3737 	PROC_UNLOCK(p);
3738 	efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
3739 	efbuf->fdp = NULL;
3740 	efbuf->sb = sb;
3741 	efbuf->remainder = maxlen;
3742 	efbuf->flags = flags;
3743 	if (tracevp != NULL)
3744 		export_vnode_to_sb(tracevp, KF_FD_TYPE_TRACE, FREAD | FWRITE,
3745 		    efbuf);
3746 	if (textvp != NULL)
3747 		export_vnode_to_sb(textvp, KF_FD_TYPE_TEXT, FREAD, efbuf);
3748 	if (cttyvp != NULL)
3749 		export_vnode_to_sb(cttyvp, KF_FD_TYPE_CTTY, FREAD | FWRITE,
3750 		    efbuf);
3751 	error = 0;
3752 	if (fdp == NULL)
3753 		goto fail;
3754 	efbuf->fdp = fdp;
3755 	FILEDESC_SLOCK(fdp);
3756 	/* working directory */
3757 	if (fdp->fd_cdir != NULL) {
3758 		vrefact(fdp->fd_cdir);
3759 		export_vnode_to_sb(fdp->fd_cdir, KF_FD_TYPE_CWD, FREAD, efbuf);
3760 	}
3761 	/* root directory */
3762 	if (fdp->fd_rdir != NULL) {
3763 		vrefact(fdp->fd_rdir);
3764 		export_vnode_to_sb(fdp->fd_rdir, KF_FD_TYPE_ROOT, FREAD, efbuf);
3765 	}
3766 	/* jail directory */
3767 	if (fdp->fd_jdir != NULL) {
3768 		vrefact(fdp->fd_jdir);
3769 		export_vnode_to_sb(fdp->fd_jdir, KF_FD_TYPE_JAIL, FREAD, efbuf);
3770 	}
3771 	for (i = 0; fdp->fd_refcnt > 0 && i <= fdp->fd_lastfile; i++) {
3772 		if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
3773 			continue;
3774 #ifdef CAPABILITIES
3775 		rights = *cap_rights(fdp, i);
3776 #else /* !CAPABILITIES */
3777 		rights = cap_no_rights;
3778 #endif
3779 		/*
3780 		 * Create sysctl entry.  It is OK to drop the filedesc
3781 		 * lock inside of export_file_to_sb() as we will
3782 		 * re-validate and re-evaluate its properties when the
3783 		 * loop continues.
3784 		 */
3785 		error = export_file_to_sb(fp, i, &rights, efbuf);
3786 		if (error != 0 || efbuf->remainder == 0)
3787 			break;
3788 	}
3789 	FILEDESC_SUNLOCK(fdp);
3790 	fddrop(fdp);
3791 fail:
3792 	free(efbuf, M_TEMP);
3793 	return (error);
3794 }
3795 
3796 #define FILEDESC_SBUF_SIZE	(sizeof(struct kinfo_file) * 5)
3797 
3798 /*
3799  * Get per-process file descriptors for use by procstat(1), et al.
3800  */
3801 static int
3802 sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS)
3803 {
3804 	struct sbuf sb;
3805 	struct proc *p;
3806 	ssize_t maxlen;
3807 	int error, error2, *name;
3808 
3809 	name = (int *)arg1;
3810 
3811 	sbuf_new_for_sysctl(&sb, NULL, FILEDESC_SBUF_SIZE, req);
3812 	sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
3813 	error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
3814 	if (error != 0) {
3815 		sbuf_delete(&sb);
3816 		return (error);
3817 	}
3818 	maxlen = req->oldptr != NULL ? req->oldlen : -1;
3819 	error = kern_proc_filedesc_out(p, &sb, maxlen,
3820 	    KERN_FILEDESC_PACK_KINFO);
3821 	error2 = sbuf_finish(&sb);
3822 	sbuf_delete(&sb);
3823 	return (error != 0 ? error : error2);
3824 }
3825 
3826 #ifdef COMPAT_FREEBSD7
3827 #ifdef KINFO_OFILE_SIZE
3828 CTASSERT(sizeof(struct kinfo_ofile) == KINFO_OFILE_SIZE);
3829 #endif
3830 
3831 static void
3832 kinfo_to_okinfo(struct kinfo_file *kif, struct kinfo_ofile *okif)
3833 {
3834 
3835 	okif->kf_structsize = sizeof(*okif);
3836 	okif->kf_type = kif->kf_type;
3837 	okif->kf_fd = kif->kf_fd;
3838 	okif->kf_ref_count = kif->kf_ref_count;
3839 	okif->kf_flags = kif->kf_flags & (KF_FLAG_READ | KF_FLAG_WRITE |
3840 	    KF_FLAG_APPEND | KF_FLAG_ASYNC | KF_FLAG_FSYNC | KF_FLAG_NONBLOCK |
3841 	    KF_FLAG_DIRECT | KF_FLAG_HASLOCK);
3842 	okif->kf_offset = kif->kf_offset;
3843 	if (kif->kf_type == KF_TYPE_VNODE)
3844 		okif->kf_vnode_type = kif->kf_un.kf_file.kf_file_type;
3845 	else
3846 		okif->kf_vnode_type = KF_VTYPE_VNON;
3847 	strlcpy(okif->kf_path, kif->kf_path, sizeof(okif->kf_path));
3848 	if (kif->kf_type == KF_TYPE_SOCKET) {
3849 		okif->kf_sock_domain = kif->kf_un.kf_sock.kf_sock_domain0;
3850 		okif->kf_sock_type = kif->kf_un.kf_sock.kf_sock_type0;
3851 		okif->kf_sock_protocol = kif->kf_un.kf_sock.kf_sock_protocol0;
3852 		okif->kf_sa_local = kif->kf_un.kf_sock.kf_sa_local;
3853 		okif->kf_sa_peer = kif->kf_un.kf_sock.kf_sa_peer;
3854 	} else {
3855 		okif->kf_sa_local.ss_family = AF_UNSPEC;
3856 		okif->kf_sa_peer.ss_family = AF_UNSPEC;
3857 	}
3858 }
3859 
3860 static int
3861 export_vnode_for_osysctl(struct vnode *vp, int type, struct kinfo_file *kif,
3862     struct kinfo_ofile *okif, struct filedesc *fdp, struct sysctl_req *req)
3863 {
3864 	int error;
3865 
3866 	vrefact(vp);
3867 	FILEDESC_SUNLOCK(fdp);
3868 	export_vnode_to_kinfo(vp, type, 0, kif, KERN_FILEDESC_PACK_KINFO);
3869 	kinfo_to_okinfo(kif, okif);
3870 	error = SYSCTL_OUT(req, okif, sizeof(*okif));
3871 	FILEDESC_SLOCK(fdp);
3872 	return (error);
3873 }
3874 
3875 /*
3876  * Get per-process file descriptors for use by procstat(1), et al.
3877  */
3878 static int
3879 sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS)
3880 {
3881 	struct kinfo_ofile *okif;
3882 	struct kinfo_file *kif;
3883 	struct filedesc *fdp;
3884 	int error, i, *name;
3885 	struct file *fp;
3886 	struct proc *p;
3887 
3888 	name = (int *)arg1;
3889 	error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
3890 	if (error != 0)
3891 		return (error);
3892 	fdp = fdhold(p);
3893 	PROC_UNLOCK(p);
3894 	if (fdp == NULL)
3895 		return (ENOENT);
3896 	kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK);
3897 	okif = malloc(sizeof(*okif), M_TEMP, M_WAITOK);
3898 	FILEDESC_SLOCK(fdp);
3899 	if (fdp->fd_cdir != NULL)
3900 		export_vnode_for_osysctl(fdp->fd_cdir, KF_FD_TYPE_CWD, kif,
3901 		    okif, fdp, req);
3902 	if (fdp->fd_rdir != NULL)
3903 		export_vnode_for_osysctl(fdp->fd_rdir, KF_FD_TYPE_ROOT, kif,
3904 		    okif, fdp, req);
3905 	if (fdp->fd_jdir != NULL)
3906 		export_vnode_for_osysctl(fdp->fd_jdir, KF_FD_TYPE_JAIL, kif,
3907 		    okif, fdp, req);
3908 	for (i = 0; fdp->fd_refcnt > 0 && i <= fdp->fd_lastfile; i++) {
3909 		if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
3910 			continue;
3911 		export_file_to_kinfo(fp, i, NULL, kif, fdp,
3912 		    KERN_FILEDESC_PACK_KINFO);
3913 		FILEDESC_SUNLOCK(fdp);
3914 		kinfo_to_okinfo(kif, okif);
3915 		error = SYSCTL_OUT(req, okif, sizeof(*okif));
3916 		FILEDESC_SLOCK(fdp);
3917 		if (error)
3918 			break;
3919 	}
3920 	FILEDESC_SUNLOCK(fdp);
3921 	fddrop(fdp);
3922 	free(kif, M_TEMP);
3923 	free(okif, M_TEMP);
3924 	return (0);
3925 }
3926 
3927 static SYSCTL_NODE(_kern_proc, KERN_PROC_OFILEDESC, ofiledesc,
3928     CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_ofiledesc,
3929     "Process ofiledesc entries");
3930 #endif	/* COMPAT_FREEBSD7 */
3931 
3932 int
3933 vntype_to_kinfo(int vtype)
3934 {
3935 	struct {
3936 		int	vtype;
3937 		int	kf_vtype;
3938 	} vtypes_table[] = {
3939 		{ VBAD, KF_VTYPE_VBAD },
3940 		{ VBLK, KF_VTYPE_VBLK },
3941 		{ VCHR, KF_VTYPE_VCHR },
3942 		{ VDIR, KF_VTYPE_VDIR },
3943 		{ VFIFO, KF_VTYPE_VFIFO },
3944 		{ VLNK, KF_VTYPE_VLNK },
3945 		{ VNON, KF_VTYPE_VNON },
3946 		{ VREG, KF_VTYPE_VREG },
3947 		{ VSOCK, KF_VTYPE_VSOCK }
3948 	};
3949 	unsigned int i;
3950 
3951 	/*
3952 	 * Perform vtype translation.
3953 	 */
3954 	for (i = 0; i < nitems(vtypes_table); i++)
3955 		if (vtypes_table[i].vtype == vtype)
3956 			return (vtypes_table[i].kf_vtype);
3957 
3958 	return (KF_VTYPE_UNKNOWN);
3959 }
3960 
3961 static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc,
3962     CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_filedesc,
3963     "Process filedesc entries");
3964 
3965 /*
3966  * Store a process current working directory information to sbuf.
3967  *
3968  * Takes a locked proc as argument, and returns with the proc unlocked.
3969  */
3970 int
3971 kern_proc_cwd_out(struct proc *p,  struct sbuf *sb, ssize_t maxlen)
3972 {
3973 	struct filedesc *fdp;
3974 	struct export_fd_buf *efbuf;
3975 	int error;
3976 
3977 	PROC_LOCK_ASSERT(p, MA_OWNED);
3978 
3979 	fdp = fdhold(p);
3980 	PROC_UNLOCK(p);
3981 	if (fdp == NULL)
3982 		return (EINVAL);
3983 
3984 	efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
3985 	efbuf->fdp = fdp;
3986 	efbuf->sb = sb;
3987 	efbuf->remainder = maxlen;
3988 
3989 	FILEDESC_SLOCK(fdp);
3990 	if (fdp->fd_cdir == NULL)
3991 		error = EINVAL;
3992 	else {
3993 		vrefact(fdp->fd_cdir);
3994 		error = export_vnode_to_sb(fdp->fd_cdir, KF_FD_TYPE_CWD,
3995 		    FREAD, efbuf);
3996 	}
3997 	FILEDESC_SUNLOCK(fdp);
3998 	fddrop(fdp);
3999 	free(efbuf, M_TEMP);
4000 	return (error);
4001 }
4002 
4003 /*
4004  * Get per-process current working directory.
4005  */
4006 static int
4007 sysctl_kern_proc_cwd(SYSCTL_HANDLER_ARGS)
4008 {
4009 	struct sbuf sb;
4010 	struct proc *p;
4011 	ssize_t maxlen;
4012 	int error, error2, *name;
4013 
4014 	name = (int *)arg1;
4015 
4016 	sbuf_new_for_sysctl(&sb, NULL, sizeof(struct kinfo_file), req);
4017 	sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
4018 	error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
4019 	if (error != 0) {
4020 		sbuf_delete(&sb);
4021 		return (error);
4022 	}
4023 	maxlen = req->oldptr != NULL ? req->oldlen : -1;
4024 	error = kern_proc_cwd_out(p, &sb, maxlen);
4025 	error2 = sbuf_finish(&sb);
4026 	sbuf_delete(&sb);
4027 	return (error != 0 ? error : error2);
4028 }
4029 
4030 static SYSCTL_NODE(_kern_proc, KERN_PROC_CWD, cwd, CTLFLAG_RD|CTLFLAG_MPSAFE,
4031     sysctl_kern_proc_cwd, "Process current working directory");
4032 
4033 #ifdef DDB
4034 /*
4035  * For the purposes of debugging, generate a human-readable string for the
4036  * file type.
4037  */
4038 static const char *
4039 file_type_to_name(short type)
4040 {
4041 
4042 	switch (type) {
4043 	case 0:
4044 		return ("zero");
4045 	case DTYPE_VNODE:
4046 		return ("vnode");
4047 	case DTYPE_SOCKET:
4048 		return ("socket");
4049 	case DTYPE_PIPE:
4050 		return ("pipe");
4051 	case DTYPE_FIFO:
4052 		return ("fifo");
4053 	case DTYPE_KQUEUE:
4054 		return ("kqueue");
4055 	case DTYPE_CRYPTO:
4056 		return ("crypto");
4057 	case DTYPE_MQUEUE:
4058 		return ("mqueue");
4059 	case DTYPE_SHM:
4060 		return ("shm");
4061 	case DTYPE_SEM:
4062 		return ("ksem");
4063 	case DTYPE_PTS:
4064 		return ("pts");
4065 	case DTYPE_DEV:
4066 		return ("dev");
4067 	case DTYPE_PROCDESC:
4068 		return ("proc");
4069 	case DTYPE_LINUXEFD:
4070 		return ("levent");
4071 	case DTYPE_LINUXTFD:
4072 		return ("ltimer");
4073 	default:
4074 		return ("unkn");
4075 	}
4076 }
4077 
4078 /*
4079  * For the purposes of debugging, identify a process (if any, perhaps one of
4080  * many) that references the passed file in its file descriptor array. Return
4081  * NULL if none.
4082  */
4083 static struct proc *
4084 file_to_first_proc(struct file *fp)
4085 {
4086 	struct filedesc *fdp;
4087 	struct proc *p;
4088 	int n;
4089 
4090 	FOREACH_PROC_IN_SYSTEM(p) {
4091 		if (p->p_state == PRS_NEW)
4092 			continue;
4093 		fdp = p->p_fd;
4094 		if (fdp == NULL)
4095 			continue;
4096 		for (n = 0; n <= fdp->fd_lastfile; n++) {
4097 			if (fp == fdp->fd_ofiles[n].fde_file)
4098 				return (p);
4099 		}
4100 	}
4101 	return (NULL);
4102 }
4103 
4104 static void
4105 db_print_file(struct file *fp, int header)
4106 {
4107 #define XPTRWIDTH ((int)howmany(sizeof(void *) * NBBY, 4))
4108 	struct proc *p;
4109 
4110 	if (header)
4111 		db_printf("%*s %6s %*s %8s %4s %5s %6s %*s %5s %s\n",
4112 		    XPTRWIDTH, "File", "Type", XPTRWIDTH, "Data", "Flag",
4113 		    "GCFl", "Count", "MCount", XPTRWIDTH, "Vnode", "FPID",
4114 		    "FCmd");
4115 	p = file_to_first_proc(fp);
4116 	db_printf("%*p %6s %*p %08x %04x %5d %6d %*p %5d %s\n", XPTRWIDTH,
4117 	    fp, file_type_to_name(fp->f_type), XPTRWIDTH, fp->f_data,
4118 	    fp->f_flag, 0, fp->f_count, 0, XPTRWIDTH, fp->f_vnode,
4119 	    p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-");
4120 
4121 #undef XPTRWIDTH
4122 }
4123 
4124 DB_SHOW_COMMAND(file, db_show_file)
4125 {
4126 	struct file *fp;
4127 
4128 	if (!have_addr) {
4129 		db_printf("usage: show file <addr>\n");
4130 		return;
4131 	}
4132 	fp = (struct file *)addr;
4133 	db_print_file(fp, 1);
4134 }
4135 
4136 DB_SHOW_COMMAND(files, db_show_files)
4137 {
4138 	struct filedesc *fdp;
4139 	struct file *fp;
4140 	struct proc *p;
4141 	int header;
4142 	int n;
4143 
4144 	header = 1;
4145 	FOREACH_PROC_IN_SYSTEM(p) {
4146 		if (p->p_state == PRS_NEW)
4147 			continue;
4148 		if ((fdp = p->p_fd) == NULL)
4149 			continue;
4150 		for (n = 0; n <= fdp->fd_lastfile; ++n) {
4151 			if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
4152 				continue;
4153 			db_print_file(fp, header);
4154 			header = 0;
4155 		}
4156 	}
4157 }
4158 #endif
4159 
4160 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
4161     &maxfilesperproc, 0, "Maximum files allowed open per process");
4162 
4163 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
4164     &maxfiles, 0, "Maximum number of files");
4165 
4166 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
4167     &openfiles, 0, "System-wide number of open files");
4168 
4169 /* ARGSUSED*/
4170 static void
4171 filelistinit(void *dummy)
4172 {
4173 
4174 	file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL,
4175 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
4176 	filedesc0_zone = uma_zcreate("filedesc0", sizeof(struct filedesc0),
4177 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
4178 	mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF);
4179 }
4180 SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL);
4181 
4182 /*-------------------------------------------------------------------*/
4183 
4184 static int
4185 badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred,
4186     int flags, struct thread *td)
4187 {
4188 
4189 	return (EBADF);
4190 }
4191 
4192 static int
4193 badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
4194     struct thread *td)
4195 {
4196 
4197 	return (EINVAL);
4198 }
4199 
4200 static int
4201 badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred,
4202     struct thread *td)
4203 {
4204 
4205 	return (EBADF);
4206 }
4207 
4208 static int
4209 badfo_poll(struct file *fp, int events, struct ucred *active_cred,
4210     struct thread *td)
4211 {
4212 
4213 	return (0);
4214 }
4215 
4216 static int
4217 badfo_kqfilter(struct file *fp, struct knote *kn)
4218 {
4219 
4220 	return (EBADF);
4221 }
4222 
4223 static int
4224 badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred,
4225     struct thread *td)
4226 {
4227 
4228 	return (EBADF);
4229 }
4230 
4231 static int
4232 badfo_close(struct file *fp, struct thread *td)
4233 {
4234 
4235 	return (0);
4236 }
4237 
4238 static int
4239 badfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
4240     struct thread *td)
4241 {
4242 
4243 	return (EBADF);
4244 }
4245 
4246 static int
4247 badfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
4248     struct thread *td)
4249 {
4250 
4251 	return (EBADF);
4252 }
4253 
4254 static int
4255 badfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
4256     struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
4257     struct thread *td)
4258 {
4259 
4260 	return (EBADF);
4261 }
4262 
4263 static int
4264 badfo_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
4265 {
4266 
4267 	return (0);
4268 }
4269 
4270 struct fileops badfileops = {
4271 	.fo_read = badfo_readwrite,
4272 	.fo_write = badfo_readwrite,
4273 	.fo_truncate = badfo_truncate,
4274 	.fo_ioctl = badfo_ioctl,
4275 	.fo_poll = badfo_poll,
4276 	.fo_kqfilter = badfo_kqfilter,
4277 	.fo_stat = badfo_stat,
4278 	.fo_close = badfo_close,
4279 	.fo_chmod = badfo_chmod,
4280 	.fo_chown = badfo_chown,
4281 	.fo_sendfile = badfo_sendfile,
4282 	.fo_fill_kinfo = badfo_fill_kinfo,
4283 };
4284 
4285 int
4286 invfo_rdwr(struct file *fp, struct uio *uio, struct ucred *active_cred,
4287     int flags, struct thread *td)
4288 {
4289 
4290 	return (EOPNOTSUPP);
4291 }
4292 
4293 int
4294 invfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
4295     struct thread *td)
4296 {
4297 
4298 	return (EINVAL);
4299 }
4300 
4301 int
4302 invfo_ioctl(struct file *fp, u_long com, void *data,
4303     struct ucred *active_cred, struct thread *td)
4304 {
4305 
4306 	return (ENOTTY);
4307 }
4308 
4309 int
4310 invfo_poll(struct file *fp, int events, struct ucred *active_cred,
4311     struct thread *td)
4312 {
4313 
4314 	return (poll_no_poll(events));
4315 }
4316 
4317 int
4318 invfo_kqfilter(struct file *fp, struct knote *kn)
4319 {
4320 
4321 	return (EINVAL);
4322 }
4323 
4324 int
4325 invfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
4326     struct thread *td)
4327 {
4328 
4329 	return (EINVAL);
4330 }
4331 
4332 int
4333 invfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
4334     struct thread *td)
4335 {
4336 
4337 	return (EINVAL);
4338 }
4339 
4340 int
4341 invfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
4342     struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
4343     struct thread *td)
4344 {
4345 
4346 	return (EINVAL);
4347 }
4348 
4349 /*-------------------------------------------------------------------*/
4350 
4351 /*
4352  * File Descriptor pseudo-device driver (/dev/fd/).
4353  *
4354  * Opening minor device N dup()s the file (if any) connected to file
4355  * descriptor N belonging to the calling process.  Note that this driver
4356  * consists of only the ``open()'' routine, because all subsequent
4357  * references to this file will be direct to the other driver.
4358  *
4359  * XXX: we could give this one a cloning event handler if necessary.
4360  */
4361 
4362 /* ARGSUSED */
4363 static int
4364 fdopen(struct cdev *dev, int mode, int type, struct thread *td)
4365 {
4366 
4367 	/*
4368 	 * XXX Kludge: set curthread->td_dupfd to contain the value of the
4369 	 * the file descriptor being sought for duplication. The error
4370 	 * return ensures that the vnode for this device will be released
4371 	 * by vn_open. Open will detect this special error and take the
4372 	 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
4373 	 * will simply report the error.
4374 	 */
4375 	td->td_dupfd = dev2unit(dev);
4376 	return (ENODEV);
4377 }
4378 
4379 static struct cdevsw fildesc_cdevsw = {
4380 	.d_version =	D_VERSION,
4381 	.d_open =	fdopen,
4382 	.d_name =	"FD",
4383 };
4384 
4385 static void
4386 fildesc_drvinit(void *unused)
4387 {
4388 	struct cdev *dev;
4389 
4390 	dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 0, NULL,
4391 	    UID_ROOT, GID_WHEEL, 0666, "fd/0");
4392 	make_dev_alias(dev, "stdin");
4393 	dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 1, NULL,
4394 	    UID_ROOT, GID_WHEEL, 0666, "fd/1");
4395 	make_dev_alias(dev, "stdout");
4396 	dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 2, NULL,
4397 	    UID_ROOT, GID_WHEEL, 0666, "fd/2");
4398 	make_dev_alias(dev, "stderr");
4399 }
4400 
4401 SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL);
4402