xref: /freebsd/sys/kern/kern_descrip.c (revision 68d75eff68281c1b445e3010bb975eae07aac225)
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, NULL);
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, NULL);
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, NULL);
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, NULL);
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, NULL);
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, NULL);
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, NULL);
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 (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 	if (error == 0) {
1449 		sbp->st_atim_ext = 0;
1450 		sbp->st_mtim_ext = 0;
1451 		sbp->st_ctim_ext = 0;
1452 		sbp->st_btim_ext = 0;
1453 	}
1454 #endif
1455 #ifdef KTRACE
1456 	if (error == 0 && KTRPOINT(td, KTR_STRUCT))
1457 		ktrstat(sbp);
1458 #endif
1459 	return (error);
1460 }
1461 
1462 #if defined(COMPAT_FREEBSD11)
1463 /*
1464  * Return status information about a file descriptor.
1465  */
1466 #ifndef _SYS_SYSPROTO_H_
1467 struct freebsd11_nfstat_args {
1468 	int	fd;
1469 	struct	nstat *sb;
1470 };
1471 #endif
1472 /* ARGSUSED */
1473 int
1474 freebsd11_nfstat(struct thread *td, struct freebsd11_nfstat_args *uap)
1475 {
1476 	struct nstat nub;
1477 	struct stat ub;
1478 	int error;
1479 
1480 	error = kern_fstat(td, uap->fd, &ub);
1481 	if (error == 0) {
1482 		freebsd11_cvtnstat(&ub, &nub);
1483 		error = copyout(&nub, uap->sb, sizeof(nub));
1484 	}
1485 	return (error);
1486 }
1487 #endif /* COMPAT_FREEBSD11 */
1488 
1489 /*
1490  * Return pathconf information about a file descriptor.
1491  */
1492 #ifndef _SYS_SYSPROTO_H_
1493 struct fpathconf_args {
1494 	int	fd;
1495 	int	name;
1496 };
1497 #endif
1498 /* ARGSUSED */
1499 int
1500 sys_fpathconf(struct thread *td, struct fpathconf_args *uap)
1501 {
1502 	long value;
1503 	int error;
1504 
1505 	error = kern_fpathconf(td, uap->fd, uap->name, &value);
1506 	if (error == 0)
1507 		td->td_retval[0] = value;
1508 	return (error);
1509 }
1510 
1511 int
1512 kern_fpathconf(struct thread *td, int fd, int name, long *valuep)
1513 {
1514 	struct file *fp;
1515 	struct vnode *vp;
1516 	int error;
1517 
1518 	error = fget(td, fd, &cap_fpathconf_rights, &fp);
1519 	if (error != 0)
1520 		return (error);
1521 
1522 	if (name == _PC_ASYNC_IO) {
1523 		*valuep = _POSIX_ASYNCHRONOUS_IO;
1524 		goto out;
1525 	}
1526 	vp = fp->f_vnode;
1527 	if (vp != NULL) {
1528 		vn_lock(vp, LK_SHARED | LK_RETRY);
1529 		error = VOP_PATHCONF(vp, name, valuep);
1530 		VOP_UNLOCK(vp);
1531 	} else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) {
1532 		if (name != _PC_PIPE_BUF) {
1533 			error = EINVAL;
1534 		} else {
1535 			*valuep = PIPE_BUF;
1536 			error = 0;
1537 		}
1538 	} else {
1539 		error = EOPNOTSUPP;
1540 	}
1541 out:
1542 	fdrop(fp, td);
1543 	return (error);
1544 }
1545 
1546 /*
1547  * Copy filecaps structure allocating memory for ioctls array if needed.
1548  *
1549  * The last parameter indicates whether the fdtable is locked. If it is not and
1550  * ioctls are encountered, copying fails and the caller must lock the table.
1551  *
1552  * Note that if the table was not locked, the caller has to check the relevant
1553  * sequence counter to determine whether the operation was successful.
1554  */
1555 bool
1556 filecaps_copy(const struct filecaps *src, struct filecaps *dst, bool locked)
1557 {
1558 	size_t size;
1559 
1560 	if (src->fc_ioctls != NULL && !locked)
1561 		return (false);
1562 	memcpy(dst, src, sizeof(*src));
1563 	if (src->fc_ioctls == NULL)
1564 		return (true);
1565 
1566 	KASSERT(src->fc_nioctls > 0,
1567 	    ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1568 
1569 	size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1570 	dst->fc_ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1571 	memcpy(dst->fc_ioctls, src->fc_ioctls, size);
1572 	return (true);
1573 }
1574 
1575 static u_long *
1576 filecaps_copy_prep(const struct filecaps *src)
1577 {
1578 	u_long *ioctls;
1579 	size_t size;
1580 
1581 	if (__predict_true(src->fc_ioctls == NULL))
1582 		return (NULL);
1583 
1584 	KASSERT(src->fc_nioctls > 0,
1585 	    ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1586 
1587 	size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1588 	ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1589 	return (ioctls);
1590 }
1591 
1592 static void
1593 filecaps_copy_finish(const struct filecaps *src, struct filecaps *dst,
1594     u_long *ioctls)
1595 {
1596 	size_t size;
1597 
1598 	*dst = *src;
1599 	if (__predict_true(src->fc_ioctls == NULL)) {
1600 		MPASS(ioctls == NULL);
1601 		return;
1602 	}
1603 
1604 	size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1605 	dst->fc_ioctls = ioctls;
1606 	bcopy(src->fc_ioctls, dst->fc_ioctls, size);
1607 }
1608 
1609 /*
1610  * Move filecaps structure to the new place and clear the old place.
1611  */
1612 void
1613 filecaps_move(struct filecaps *src, struct filecaps *dst)
1614 {
1615 
1616 	*dst = *src;
1617 	bzero(src, sizeof(*src));
1618 }
1619 
1620 /*
1621  * Fill the given filecaps structure with full rights.
1622  */
1623 static void
1624 filecaps_fill(struct filecaps *fcaps)
1625 {
1626 
1627 	CAP_ALL(&fcaps->fc_rights);
1628 	fcaps->fc_ioctls = NULL;
1629 	fcaps->fc_nioctls = -1;
1630 	fcaps->fc_fcntls = CAP_FCNTL_ALL;
1631 }
1632 
1633 /*
1634  * Free memory allocated within filecaps structure.
1635  */
1636 void
1637 filecaps_free(struct filecaps *fcaps)
1638 {
1639 
1640 	free(fcaps->fc_ioctls, M_FILECAPS);
1641 	bzero(fcaps, sizeof(*fcaps));
1642 }
1643 
1644 static u_long *
1645 filecaps_free_prep(struct filecaps *fcaps)
1646 {
1647 	u_long *ioctls;
1648 
1649 	ioctls = fcaps->fc_ioctls;
1650 	bzero(fcaps, sizeof(*fcaps));
1651 	return (ioctls);
1652 }
1653 
1654 static void
1655 filecaps_free_finish(u_long *ioctls)
1656 {
1657 
1658 	free(ioctls, M_FILECAPS);
1659 }
1660 
1661 /*
1662  * Validate the given filecaps structure.
1663  */
1664 static void
1665 filecaps_validate(const struct filecaps *fcaps, const char *func)
1666 {
1667 
1668 	KASSERT(cap_rights_is_valid(&fcaps->fc_rights),
1669 	    ("%s: invalid rights", func));
1670 	KASSERT((fcaps->fc_fcntls & ~CAP_FCNTL_ALL) == 0,
1671 	    ("%s: invalid fcntls", func));
1672 	KASSERT(fcaps->fc_fcntls == 0 ||
1673 	    cap_rights_is_set(&fcaps->fc_rights, CAP_FCNTL),
1674 	    ("%s: fcntls without CAP_FCNTL", func));
1675 	KASSERT(fcaps->fc_ioctls != NULL ? fcaps->fc_nioctls > 0 :
1676 	    (fcaps->fc_nioctls == -1 || fcaps->fc_nioctls == 0),
1677 	    ("%s: invalid ioctls", func));
1678 	KASSERT(fcaps->fc_nioctls == 0 ||
1679 	    cap_rights_is_set(&fcaps->fc_rights, CAP_IOCTL),
1680 	    ("%s: ioctls without CAP_IOCTL", func));
1681 }
1682 
1683 static void
1684 fdgrowtable_exp(struct filedesc *fdp, int nfd)
1685 {
1686 	int nfd1;
1687 
1688 	FILEDESC_XLOCK_ASSERT(fdp);
1689 
1690 	nfd1 = fdp->fd_nfiles * 2;
1691 	if (nfd1 < nfd)
1692 		nfd1 = nfd;
1693 	fdgrowtable(fdp, nfd1);
1694 }
1695 
1696 /*
1697  * Grow the file table to accommodate (at least) nfd descriptors.
1698  */
1699 static void
1700 fdgrowtable(struct filedesc *fdp, int nfd)
1701 {
1702 	struct filedesc0 *fdp0;
1703 	struct freetable *ft;
1704 	struct fdescenttbl *ntable;
1705 	struct fdescenttbl *otable;
1706 	int nnfiles, onfiles;
1707 	NDSLOTTYPE *nmap, *omap;
1708 
1709 	/*
1710 	 * If lastfile is -1 this struct filedesc was just allocated and we are
1711 	 * growing it to accommodate for the one we are going to copy from. There
1712 	 * is no need to have a lock on this one as it's not visible to anyone.
1713 	 */
1714 	if (fdp->fd_lastfile != -1)
1715 		FILEDESC_XLOCK_ASSERT(fdp);
1716 
1717 	KASSERT(fdp->fd_nfiles > 0, ("zero-length file table"));
1718 
1719 	/* save old values */
1720 	onfiles = fdp->fd_nfiles;
1721 	otable = fdp->fd_files;
1722 	omap = fdp->fd_map;
1723 
1724 	/* compute the size of the new table */
1725 	nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */
1726 	if (nnfiles <= onfiles)
1727 		/* the table is already large enough */
1728 		return;
1729 
1730 	/*
1731 	 * Allocate a new table.  We need enough space for the number of
1732 	 * entries, file entries themselves and the struct freetable we will use
1733 	 * when we decommission the table and place it on the freelist.
1734 	 * We place the struct freetable in the middle so we don't have
1735 	 * to worry about padding.
1736 	 */
1737 	ntable = malloc(offsetof(struct fdescenttbl, fdt_ofiles) +
1738 	    nnfiles * sizeof(ntable->fdt_ofiles[0]) +
1739 	    sizeof(struct freetable),
1740 	    M_FILEDESC, M_ZERO | M_WAITOK);
1741 	/* copy the old data */
1742 	ntable->fdt_nfiles = nnfiles;
1743 	memcpy(ntable->fdt_ofiles, otable->fdt_ofiles,
1744 	    onfiles * sizeof(ntable->fdt_ofiles[0]));
1745 
1746 	/*
1747 	 * Allocate a new map only if the old is not large enough.  It will
1748 	 * grow at a slower rate than the table as it can map more
1749 	 * entries than the table can hold.
1750 	 */
1751 	if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) {
1752 		nmap = malloc(NDSLOTS(nnfiles) * NDSLOTSIZE, M_FILEDESC,
1753 		    M_ZERO | M_WAITOK);
1754 		/* copy over the old data and update the pointer */
1755 		memcpy(nmap, omap, NDSLOTS(onfiles) * sizeof(*omap));
1756 		fdp->fd_map = nmap;
1757 	}
1758 
1759 	/*
1760 	 * Make sure that ntable is correctly initialized before we replace
1761 	 * fd_files poiner. Otherwise fget_unlocked() may see inconsistent
1762 	 * data.
1763 	 */
1764 	atomic_store_rel_ptr((volatile void *)&fdp->fd_files, (uintptr_t)ntable);
1765 
1766 	/*
1767 	 * Do not free the old file table, as some threads may still
1768 	 * reference entries within it.  Instead, place it on a freelist
1769 	 * which will be processed when the struct filedesc is released.
1770 	 *
1771 	 * Note that if onfiles == NDFILE, we're dealing with the original
1772 	 * static allocation contained within (struct filedesc0 *)fdp,
1773 	 * which must not be freed.
1774 	 */
1775 	if (onfiles > NDFILE) {
1776 		ft = (struct freetable *)&otable->fdt_ofiles[onfiles];
1777 		fdp0 = (struct filedesc0 *)fdp;
1778 		ft->ft_table = otable;
1779 		SLIST_INSERT_HEAD(&fdp0->fd_free, ft, ft_next);
1780 	}
1781 	/*
1782 	 * The map does not have the same possibility of threads still
1783 	 * holding references to it.  So always free it as long as it
1784 	 * does not reference the original static allocation.
1785 	 */
1786 	if (NDSLOTS(onfiles) > NDSLOTS(NDFILE))
1787 		free(omap, M_FILEDESC);
1788 }
1789 
1790 /*
1791  * Allocate a file descriptor for the process.
1792  */
1793 int
1794 fdalloc(struct thread *td, int minfd, int *result)
1795 {
1796 	struct proc *p = td->td_proc;
1797 	struct filedesc *fdp = p->p_fd;
1798 	int fd, maxfd, allocfd;
1799 #ifdef RACCT
1800 	int error;
1801 #endif
1802 
1803 	FILEDESC_XLOCK_ASSERT(fdp);
1804 
1805 	if (fdp->fd_freefile > minfd)
1806 		minfd = fdp->fd_freefile;
1807 
1808 	maxfd = getmaxfd(td);
1809 
1810 	/*
1811 	 * Search the bitmap for a free descriptor starting at minfd.
1812 	 * If none is found, grow the file table.
1813 	 */
1814 	fd = fd_first_free(fdp, minfd, fdp->fd_nfiles);
1815 	if (fd >= maxfd)
1816 		return (EMFILE);
1817 	if (fd >= fdp->fd_nfiles) {
1818 		allocfd = min(fd * 2, maxfd);
1819 #ifdef RACCT
1820 		if (RACCT_ENABLED()) {
1821 			error = racct_set_unlocked(p, RACCT_NOFILE, allocfd);
1822 			if (error != 0)
1823 				return (EMFILE);
1824 		}
1825 #endif
1826 		/*
1827 		 * fd is already equal to first free descriptor >= minfd, so
1828 		 * we only need to grow the table and we are done.
1829 		 */
1830 		fdgrowtable_exp(fdp, allocfd);
1831 	}
1832 
1833 	/*
1834 	 * Perform some sanity checks, then mark the file descriptor as
1835 	 * used and return it to the caller.
1836 	 */
1837 	KASSERT(fd >= 0 && fd < min(maxfd, fdp->fd_nfiles),
1838 	    ("invalid descriptor %d", fd));
1839 	KASSERT(!fdisused(fdp, fd),
1840 	    ("fd_first_free() returned non-free descriptor"));
1841 	KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
1842 	    ("file descriptor isn't free"));
1843 	fdused(fdp, fd);
1844 	*result = fd;
1845 	return (0);
1846 }
1847 
1848 /*
1849  * Allocate n file descriptors for the process.
1850  */
1851 int
1852 fdallocn(struct thread *td, int minfd, int *fds, int n)
1853 {
1854 	struct proc *p = td->td_proc;
1855 	struct filedesc *fdp = p->p_fd;
1856 	int i;
1857 
1858 	FILEDESC_XLOCK_ASSERT(fdp);
1859 
1860 	for (i = 0; i < n; i++)
1861 		if (fdalloc(td, 0, &fds[i]) != 0)
1862 			break;
1863 
1864 	if (i < n) {
1865 		for (i--; i >= 0; i--)
1866 			fdunused(fdp, fds[i]);
1867 		return (EMFILE);
1868 	}
1869 
1870 	return (0);
1871 }
1872 
1873 /*
1874  * Create a new open file structure and allocate a file descriptor for the
1875  * process that refers to it.  We add one reference to the file for the
1876  * descriptor table and one reference for resultfp. This is to prevent us
1877  * being preempted and the entry in the descriptor table closed after we
1878  * release the FILEDESC lock.
1879  */
1880 int
1881 falloc_caps(struct thread *td, struct file **resultfp, int *resultfd, int flags,
1882     struct filecaps *fcaps)
1883 {
1884 	struct file *fp;
1885 	int error, fd;
1886 
1887 	error = falloc_noinstall(td, &fp);
1888 	if (error)
1889 		return (error);		/* no reference held on error */
1890 
1891 	error = finstall(td, fp, &fd, flags, fcaps);
1892 	if (error) {
1893 		fdrop(fp, td);		/* one reference (fp only) */
1894 		return (error);
1895 	}
1896 
1897 	if (resultfp != NULL)
1898 		*resultfp = fp;		/* copy out result */
1899 	else
1900 		fdrop(fp, td);		/* release local reference */
1901 
1902 	if (resultfd != NULL)
1903 		*resultfd = fd;
1904 
1905 	return (0);
1906 }
1907 
1908 /*
1909  * Create a new open file structure without allocating a file descriptor.
1910  */
1911 int
1912 falloc_noinstall(struct thread *td, struct file **resultfp)
1913 {
1914 	struct file *fp;
1915 	int maxuserfiles = maxfiles - (maxfiles / 20);
1916 	int openfiles_new;
1917 	static struct timeval lastfail;
1918 	static int curfail;
1919 
1920 	KASSERT(resultfp != NULL, ("%s: resultfp == NULL", __func__));
1921 
1922 	openfiles_new = atomic_fetchadd_int(&openfiles, 1) + 1;
1923 	if ((openfiles_new >= maxuserfiles &&
1924 	    priv_check(td, PRIV_MAXFILES) != 0) ||
1925 	    openfiles_new >= maxfiles) {
1926 		atomic_subtract_int(&openfiles, 1);
1927 		if (ppsratecheck(&lastfail, &curfail, 1)) {
1928 			printf("kern.maxfiles limit exceeded by uid %i, (%s) "
1929 			    "please see tuning(7).\n", td->td_ucred->cr_ruid, td->td_proc->p_comm);
1930 		}
1931 		return (ENFILE);
1932 	}
1933 	fp = uma_zalloc(file_zone, M_WAITOK);
1934 	bzero(fp, sizeof(*fp));
1935 	refcount_init(&fp->f_count, 1);
1936 	fp->f_cred = crhold(td->td_ucred);
1937 	fp->f_ops = &badfileops;
1938 	*resultfp = fp;
1939 	return (0);
1940 }
1941 
1942 /*
1943  * Install a file in a file descriptor table.
1944  */
1945 void
1946 _finstall(struct filedesc *fdp, struct file *fp, int fd, int flags,
1947     struct filecaps *fcaps)
1948 {
1949 	struct filedescent *fde;
1950 
1951 	MPASS(fp != NULL);
1952 	if (fcaps != NULL)
1953 		filecaps_validate(fcaps, __func__);
1954 	FILEDESC_XLOCK_ASSERT(fdp);
1955 
1956 	fde = &fdp->fd_ofiles[fd];
1957 #ifdef CAPABILITIES
1958 	seqc_write_begin(&fde->fde_seqc);
1959 #endif
1960 	fde->fde_file = fp;
1961 	fde->fde_flags = (flags & O_CLOEXEC) != 0 ? UF_EXCLOSE : 0;
1962 	if (fcaps != NULL)
1963 		filecaps_move(fcaps, &fde->fde_caps);
1964 	else
1965 		filecaps_fill(&fde->fde_caps);
1966 #ifdef CAPABILITIES
1967 	seqc_write_end(&fde->fde_seqc);
1968 #endif
1969 }
1970 
1971 int
1972 finstall(struct thread *td, struct file *fp, int *fd, int flags,
1973     struct filecaps *fcaps)
1974 {
1975 	struct filedesc *fdp = td->td_proc->p_fd;
1976 	int error;
1977 
1978 	MPASS(fd != NULL);
1979 
1980 	if (!fhold(fp))
1981 		return (EBADF);
1982 	FILEDESC_XLOCK(fdp);
1983 	if ((error = fdalloc(td, 0, fd))) {
1984 		FILEDESC_XUNLOCK(fdp);
1985 		fdrop(fp, td);
1986 		return (error);
1987 	}
1988 	_finstall(fdp, fp, *fd, flags, fcaps);
1989 	FILEDESC_XUNLOCK(fdp);
1990 	return (0);
1991 }
1992 
1993 /*
1994  * Build a new filedesc structure from another.
1995  * Copy the current, root, and jail root vnode references.
1996  *
1997  * If fdp is not NULL, return with it shared locked.
1998  */
1999 struct filedesc *
2000 fdinit(struct filedesc *fdp, bool prepfiles)
2001 {
2002 	struct filedesc0 *newfdp0;
2003 	struct filedesc *newfdp;
2004 
2005 	newfdp0 = uma_zalloc(filedesc0_zone, M_WAITOK | M_ZERO);
2006 	newfdp = &newfdp0->fd_fd;
2007 
2008 	/* Create the file descriptor table. */
2009 	FILEDESC_LOCK_INIT(newfdp);
2010 	refcount_init(&newfdp->fd_refcnt, 1);
2011 	refcount_init(&newfdp->fd_holdcnt, 1);
2012 	newfdp->fd_cmask = CMASK;
2013 	newfdp->fd_map = newfdp0->fd_dmap;
2014 	newfdp->fd_lastfile = -1;
2015 	newfdp->fd_files = (struct fdescenttbl *)&newfdp0->fd_dfiles;
2016 	newfdp->fd_files->fdt_nfiles = NDFILE;
2017 
2018 	if (fdp == NULL)
2019 		return (newfdp);
2020 
2021 	if (prepfiles && fdp->fd_lastfile >= newfdp->fd_nfiles)
2022 		fdgrowtable(newfdp, fdp->fd_lastfile + 1);
2023 
2024 	FILEDESC_SLOCK(fdp);
2025 	newfdp->fd_cdir = fdp->fd_cdir;
2026 	if (newfdp->fd_cdir)
2027 		vrefact(newfdp->fd_cdir);
2028 	newfdp->fd_rdir = fdp->fd_rdir;
2029 	if (newfdp->fd_rdir)
2030 		vrefact(newfdp->fd_rdir);
2031 	newfdp->fd_jdir = fdp->fd_jdir;
2032 	if (newfdp->fd_jdir)
2033 		vrefact(newfdp->fd_jdir);
2034 
2035 	if (!prepfiles) {
2036 		FILEDESC_SUNLOCK(fdp);
2037 	} else {
2038 		while (fdp->fd_lastfile >= newfdp->fd_nfiles) {
2039 			FILEDESC_SUNLOCK(fdp);
2040 			fdgrowtable(newfdp, fdp->fd_lastfile + 1);
2041 			FILEDESC_SLOCK(fdp);
2042 		}
2043 	}
2044 
2045 	return (newfdp);
2046 }
2047 
2048 static struct filedesc *
2049 fdhold(struct proc *p)
2050 {
2051 	struct filedesc *fdp;
2052 
2053 	PROC_LOCK_ASSERT(p, MA_OWNED);
2054 	fdp = p->p_fd;
2055 	if (fdp != NULL)
2056 		refcount_acquire(&fdp->fd_holdcnt);
2057 	return (fdp);
2058 }
2059 
2060 static void
2061 fddrop(struct filedesc *fdp)
2062 {
2063 
2064 	if (fdp->fd_holdcnt > 1) {
2065 		if (refcount_release(&fdp->fd_holdcnt) == 0)
2066 			return;
2067 	}
2068 
2069 	FILEDESC_LOCK_DESTROY(fdp);
2070 	uma_zfree(filedesc0_zone, fdp);
2071 }
2072 
2073 /*
2074  * Share a filedesc structure.
2075  */
2076 struct filedesc *
2077 fdshare(struct filedesc *fdp)
2078 {
2079 
2080 	refcount_acquire(&fdp->fd_refcnt);
2081 	return (fdp);
2082 }
2083 
2084 /*
2085  * Unshare a filedesc structure, if necessary by making a copy
2086  */
2087 void
2088 fdunshare(struct thread *td)
2089 {
2090 	struct filedesc *tmp;
2091 	struct proc *p = td->td_proc;
2092 
2093 	if (p->p_fd->fd_refcnt == 1)
2094 		return;
2095 
2096 	tmp = fdcopy(p->p_fd);
2097 	fdescfree(td);
2098 	p->p_fd = tmp;
2099 }
2100 
2101 void
2102 fdinstall_remapped(struct thread *td, struct filedesc *fdp)
2103 {
2104 
2105 	fdescfree(td);
2106 	td->td_proc->p_fd = fdp;
2107 }
2108 
2109 /*
2110  * Copy a filedesc structure.  A NULL pointer in returns a NULL reference,
2111  * this is to ease callers, not catch errors.
2112  */
2113 struct filedesc *
2114 fdcopy(struct filedesc *fdp)
2115 {
2116 	struct filedesc *newfdp;
2117 	struct filedescent *nfde, *ofde;
2118 	int i;
2119 
2120 	MPASS(fdp != NULL);
2121 
2122 	newfdp = fdinit(fdp, true);
2123 	/* copy all passable descriptors (i.e. not kqueue) */
2124 	newfdp->fd_freefile = -1;
2125 	for (i = 0; i <= fdp->fd_lastfile; ++i) {
2126 		ofde = &fdp->fd_ofiles[i];
2127 		if (ofde->fde_file == NULL ||
2128 		    (ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0 ||
2129 		    !fhold(ofde->fde_file)) {
2130 			if (newfdp->fd_freefile == -1)
2131 				newfdp->fd_freefile = i;
2132 			continue;
2133 		}
2134 		nfde = &newfdp->fd_ofiles[i];
2135 		*nfde = *ofde;
2136 		filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true);
2137 		fdused_init(newfdp, i);
2138 		newfdp->fd_lastfile = i;
2139 	}
2140 	if (newfdp->fd_freefile == -1)
2141 		newfdp->fd_freefile = i;
2142 	newfdp->fd_cmask = fdp->fd_cmask;
2143 	FILEDESC_SUNLOCK(fdp);
2144 	return (newfdp);
2145 }
2146 
2147 /*
2148  * Copies a filedesc structure, while remapping all file descriptors
2149  * stored inside using a translation table.
2150  *
2151  * File descriptors are copied over to the new file descriptor table,
2152  * regardless of whether the close-on-exec flag is set.
2153  */
2154 int
2155 fdcopy_remapped(struct filedesc *fdp, const int *fds, size_t nfds,
2156     struct filedesc **ret)
2157 {
2158 	struct filedesc *newfdp;
2159 	struct filedescent *nfde, *ofde;
2160 	int error, i;
2161 
2162 	MPASS(fdp != NULL);
2163 
2164 	newfdp = fdinit(fdp, true);
2165 	if (nfds > fdp->fd_lastfile + 1) {
2166 		/* New table cannot be larger than the old one. */
2167 		error = E2BIG;
2168 		goto bad;
2169 	}
2170 	/* Copy all passable descriptors (i.e. not kqueue). */
2171 	newfdp->fd_freefile = nfds;
2172 	for (i = 0; i < nfds; ++i) {
2173 		if (fds[i] < 0 || fds[i] > fdp->fd_lastfile) {
2174 			/* File descriptor out of bounds. */
2175 			error = EBADF;
2176 			goto bad;
2177 		}
2178 		ofde = &fdp->fd_ofiles[fds[i]];
2179 		if (ofde->fde_file == NULL) {
2180 			/* Unused file descriptor. */
2181 			error = EBADF;
2182 			goto bad;
2183 		}
2184 		if ((ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0) {
2185 			/* File descriptor cannot be passed. */
2186 			error = EINVAL;
2187 			goto bad;
2188 		}
2189 		if (!fhold(nfde->fde_file)) {
2190 			error = EBADF;
2191 			goto bad;
2192 		}
2193 		nfde = &newfdp->fd_ofiles[i];
2194 		*nfde = *ofde;
2195 		filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true);
2196 		fdused_init(newfdp, i);
2197 		newfdp->fd_lastfile = i;
2198 	}
2199 	newfdp->fd_cmask = fdp->fd_cmask;
2200 	FILEDESC_SUNLOCK(fdp);
2201 	*ret = newfdp;
2202 	return (0);
2203 bad:
2204 	FILEDESC_SUNLOCK(fdp);
2205 	fdescfree_remapped(newfdp);
2206 	return (error);
2207 }
2208 
2209 /*
2210  * Clear POSIX style locks. This is only used when fdp looses a reference (i.e.
2211  * one of processes using it exits) and the table used to be shared.
2212  */
2213 static void
2214 fdclearlocks(struct thread *td)
2215 {
2216 	struct filedesc *fdp;
2217 	struct filedesc_to_leader *fdtol;
2218 	struct flock lf;
2219 	struct file *fp;
2220 	struct proc *p;
2221 	struct vnode *vp;
2222 	int i;
2223 
2224 	p = td->td_proc;
2225 	fdp = p->p_fd;
2226 	fdtol = p->p_fdtol;
2227 	MPASS(fdtol != NULL);
2228 
2229 	FILEDESC_XLOCK(fdp);
2230 	KASSERT(fdtol->fdl_refcount > 0,
2231 	    ("filedesc_to_refcount botch: fdl_refcount=%d",
2232 	    fdtol->fdl_refcount));
2233 	if (fdtol->fdl_refcount == 1 &&
2234 	    (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2235 		for (i = 0; i <= fdp->fd_lastfile; i++) {
2236 			fp = fdp->fd_ofiles[i].fde_file;
2237 			if (fp == NULL || fp->f_type != DTYPE_VNODE ||
2238 			    !fhold(fp))
2239 				continue;
2240 			FILEDESC_XUNLOCK(fdp);
2241 			lf.l_whence = SEEK_SET;
2242 			lf.l_start = 0;
2243 			lf.l_len = 0;
2244 			lf.l_type = F_UNLCK;
2245 			vp = fp->f_vnode;
2246 			(void) VOP_ADVLOCK(vp,
2247 			    (caddr_t)p->p_leader, F_UNLCK,
2248 			    &lf, F_POSIX);
2249 			FILEDESC_XLOCK(fdp);
2250 			fdrop(fp, td);
2251 		}
2252 	}
2253 retry:
2254 	if (fdtol->fdl_refcount == 1) {
2255 		if (fdp->fd_holdleaderscount > 0 &&
2256 		    (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2257 			/*
2258 			 * close() or kern_dup() has cleared a reference
2259 			 * in a shared file descriptor table.
2260 			 */
2261 			fdp->fd_holdleaderswakeup = 1;
2262 			sx_sleep(&fdp->fd_holdleaderscount,
2263 			    FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0);
2264 			goto retry;
2265 		}
2266 		if (fdtol->fdl_holdcount > 0) {
2267 			/*
2268 			 * Ensure that fdtol->fdl_leader remains
2269 			 * valid in closef().
2270 			 */
2271 			fdtol->fdl_wakeup = 1;
2272 			sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK,
2273 			    "fdlhold", 0);
2274 			goto retry;
2275 		}
2276 	}
2277 	fdtol->fdl_refcount--;
2278 	if (fdtol->fdl_refcount == 0 &&
2279 	    fdtol->fdl_holdcount == 0) {
2280 		fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
2281 		fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
2282 	} else
2283 		fdtol = NULL;
2284 	p->p_fdtol = NULL;
2285 	FILEDESC_XUNLOCK(fdp);
2286 	if (fdtol != NULL)
2287 		free(fdtol, M_FILEDESC_TO_LEADER);
2288 }
2289 
2290 /*
2291  * Release a filedesc structure.
2292  */
2293 static void
2294 fdescfree_fds(struct thread *td, struct filedesc *fdp, bool needclose)
2295 {
2296 	struct filedesc0 *fdp0;
2297 	struct freetable *ft, *tft;
2298 	struct filedescent *fde;
2299 	struct file *fp;
2300 	int i;
2301 
2302 	for (i = 0; i <= fdp->fd_lastfile; i++) {
2303 		fde = &fdp->fd_ofiles[i];
2304 		fp = fde->fde_file;
2305 		if (fp != NULL) {
2306 			fdefree_last(fde);
2307 			if (needclose)
2308 				(void) closef(fp, td);
2309 			else
2310 				fdrop(fp, td);
2311 		}
2312 	}
2313 
2314 	if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE))
2315 		free(fdp->fd_map, M_FILEDESC);
2316 	if (fdp->fd_nfiles > NDFILE)
2317 		free(fdp->fd_files, M_FILEDESC);
2318 
2319 	fdp0 = (struct filedesc0 *)fdp;
2320 	SLIST_FOREACH_SAFE(ft, &fdp0->fd_free, ft_next, tft)
2321 		free(ft->ft_table, M_FILEDESC);
2322 
2323 	fddrop(fdp);
2324 }
2325 
2326 void
2327 fdescfree(struct thread *td)
2328 {
2329 	struct proc *p;
2330 	struct filedesc *fdp;
2331 	struct vnode *cdir, *jdir, *rdir;
2332 
2333 	p = td->td_proc;
2334 	fdp = p->p_fd;
2335 	MPASS(fdp != NULL);
2336 
2337 #ifdef RACCT
2338 	if (RACCT_ENABLED())
2339 		racct_set_unlocked(p, RACCT_NOFILE, 0);
2340 #endif
2341 
2342 	if (p->p_fdtol != NULL)
2343 		fdclearlocks(td);
2344 
2345 	PROC_LOCK(p);
2346 	p->p_fd = NULL;
2347 	PROC_UNLOCK(p);
2348 
2349 	if (refcount_release(&fdp->fd_refcnt) == 0)
2350 		return;
2351 
2352 	FILEDESC_XLOCK(fdp);
2353 	cdir = fdp->fd_cdir;
2354 	fdp->fd_cdir = NULL;
2355 	rdir = fdp->fd_rdir;
2356 	fdp->fd_rdir = NULL;
2357 	jdir = fdp->fd_jdir;
2358 	fdp->fd_jdir = NULL;
2359 	FILEDESC_XUNLOCK(fdp);
2360 
2361 	if (cdir != NULL)
2362 		vrele(cdir);
2363 	if (rdir != NULL)
2364 		vrele(rdir);
2365 	if (jdir != NULL)
2366 		vrele(jdir);
2367 
2368 	fdescfree_fds(td, fdp, 1);
2369 }
2370 
2371 void
2372 fdescfree_remapped(struct filedesc *fdp)
2373 {
2374 
2375 	if (fdp->fd_cdir != NULL)
2376 		vrele(fdp->fd_cdir);
2377 	if (fdp->fd_rdir != NULL)
2378 		vrele(fdp->fd_rdir);
2379 	if (fdp->fd_jdir != NULL)
2380 		vrele(fdp->fd_jdir);
2381 
2382 	fdescfree_fds(curthread, fdp, 0);
2383 }
2384 
2385 /*
2386  * For setugid programs, we don't want to people to use that setugidness
2387  * to generate error messages which write to a file which otherwise would
2388  * otherwise be off-limits to the process.  We check for filesystems where
2389  * the vnode can change out from under us after execve (like [lin]procfs).
2390  *
2391  * Since fdsetugidsafety calls this only for fd 0, 1 and 2, this check is
2392  * sufficient.  We also don't check for setugidness since we know we are.
2393  */
2394 static bool
2395 is_unsafe(struct file *fp)
2396 {
2397 	struct vnode *vp;
2398 
2399 	if (fp->f_type != DTYPE_VNODE)
2400 		return (false);
2401 
2402 	vp = fp->f_vnode;
2403 	return ((vp->v_vflag & VV_PROCDEP) != 0);
2404 }
2405 
2406 /*
2407  * Make this setguid thing safe, if at all possible.
2408  */
2409 void
2410 fdsetugidsafety(struct thread *td)
2411 {
2412 	struct filedesc *fdp;
2413 	struct file *fp;
2414 	int i;
2415 
2416 	fdp = td->td_proc->p_fd;
2417 	KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2418 	MPASS(fdp->fd_nfiles >= 3);
2419 	for (i = 0; i <= 2; i++) {
2420 		fp = fdp->fd_ofiles[i].fde_file;
2421 		if (fp != NULL && is_unsafe(fp)) {
2422 			FILEDESC_XLOCK(fdp);
2423 			knote_fdclose(td, i);
2424 			/*
2425 			 * NULL-out descriptor prior to close to avoid
2426 			 * a race while close blocks.
2427 			 */
2428 			fdfree(fdp, i);
2429 			FILEDESC_XUNLOCK(fdp);
2430 			(void) closef(fp, td);
2431 		}
2432 	}
2433 }
2434 
2435 /*
2436  * If a specific file object occupies a specific file descriptor, close the
2437  * file descriptor entry and drop a reference on the file object.  This is a
2438  * convenience function to handle a subsequent error in a function that calls
2439  * falloc() that handles the race that another thread might have closed the
2440  * file descriptor out from under the thread creating the file object.
2441  */
2442 void
2443 fdclose(struct thread *td, struct file *fp, int idx)
2444 {
2445 	struct filedesc *fdp = td->td_proc->p_fd;
2446 
2447 	FILEDESC_XLOCK(fdp);
2448 	if (fdp->fd_ofiles[idx].fde_file == fp) {
2449 		fdfree(fdp, idx);
2450 		FILEDESC_XUNLOCK(fdp);
2451 		fdrop(fp, td);
2452 	} else
2453 		FILEDESC_XUNLOCK(fdp);
2454 }
2455 
2456 /*
2457  * Close any files on exec?
2458  */
2459 void
2460 fdcloseexec(struct thread *td)
2461 {
2462 	struct filedesc *fdp;
2463 	struct filedescent *fde;
2464 	struct file *fp;
2465 	int i;
2466 
2467 	fdp = td->td_proc->p_fd;
2468 	KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2469 	for (i = 0; i <= fdp->fd_lastfile; i++) {
2470 		fde = &fdp->fd_ofiles[i];
2471 		fp = fde->fde_file;
2472 		if (fp != NULL && (fp->f_type == DTYPE_MQUEUE ||
2473 		    (fde->fde_flags & UF_EXCLOSE))) {
2474 			FILEDESC_XLOCK(fdp);
2475 			fdfree(fdp, i);
2476 			(void) closefp(fdp, i, fp, td, 0);
2477 			FILEDESC_UNLOCK_ASSERT(fdp);
2478 		}
2479 	}
2480 }
2481 
2482 /*
2483  * It is unsafe for set[ug]id processes to be started with file
2484  * descriptors 0..2 closed, as these descriptors are given implicit
2485  * significance in the Standard C library.  fdcheckstd() will create a
2486  * descriptor referencing /dev/null for each of stdin, stdout, and
2487  * stderr that is not already open.
2488  */
2489 int
2490 fdcheckstd(struct thread *td)
2491 {
2492 	struct filedesc *fdp;
2493 	register_t save;
2494 	int i, error, devnull;
2495 
2496 	fdp = td->td_proc->p_fd;
2497 	KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2498 	MPASS(fdp->fd_nfiles >= 3);
2499 	devnull = -1;
2500 	for (i = 0; i <= 2; i++) {
2501 		if (fdp->fd_ofiles[i].fde_file != NULL)
2502 			continue;
2503 
2504 		save = td->td_retval[0];
2505 		if (devnull != -1) {
2506 			error = kern_dup(td, FDDUP_FIXED, 0, devnull, i);
2507 		} else {
2508 			error = kern_openat(td, AT_FDCWD, "/dev/null",
2509 			    UIO_SYSSPACE, O_RDWR, 0);
2510 			if (error == 0) {
2511 				devnull = td->td_retval[0];
2512 				KASSERT(devnull == i, ("we didn't get our fd"));
2513 			}
2514 		}
2515 		td->td_retval[0] = save;
2516 		if (error != 0)
2517 			return (error);
2518 	}
2519 	return (0);
2520 }
2521 
2522 /*
2523  * Internal form of close.  Decrement reference count on file structure.
2524  * Note: td may be NULL when closing a file that was being passed in a
2525  * message.
2526  */
2527 int
2528 closef(struct file *fp, struct thread *td)
2529 {
2530 	struct vnode *vp;
2531 	struct flock lf;
2532 	struct filedesc_to_leader *fdtol;
2533 	struct filedesc *fdp;
2534 
2535 	/*
2536 	 * POSIX record locking dictates that any close releases ALL
2537 	 * locks owned by this process.  This is handled by setting
2538 	 * a flag in the unlock to free ONLY locks obeying POSIX
2539 	 * semantics, and not to free BSD-style file locks.
2540 	 * If the descriptor was in a message, POSIX-style locks
2541 	 * aren't passed with the descriptor, and the thread pointer
2542 	 * will be NULL.  Callers should be careful only to pass a
2543 	 * NULL thread pointer when there really is no owning
2544 	 * context that might have locks, or the locks will be
2545 	 * leaked.
2546 	 */
2547 	if (fp->f_type == DTYPE_VNODE && td != NULL) {
2548 		vp = fp->f_vnode;
2549 		if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
2550 			lf.l_whence = SEEK_SET;
2551 			lf.l_start = 0;
2552 			lf.l_len = 0;
2553 			lf.l_type = F_UNLCK;
2554 			(void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader,
2555 			    F_UNLCK, &lf, F_POSIX);
2556 		}
2557 		fdtol = td->td_proc->p_fdtol;
2558 		if (fdtol != NULL) {
2559 			/*
2560 			 * Handle special case where file descriptor table is
2561 			 * shared between multiple process leaders.
2562 			 */
2563 			fdp = td->td_proc->p_fd;
2564 			FILEDESC_XLOCK(fdp);
2565 			for (fdtol = fdtol->fdl_next;
2566 			    fdtol != td->td_proc->p_fdtol;
2567 			    fdtol = fdtol->fdl_next) {
2568 				if ((fdtol->fdl_leader->p_flag &
2569 				    P_ADVLOCK) == 0)
2570 					continue;
2571 				fdtol->fdl_holdcount++;
2572 				FILEDESC_XUNLOCK(fdp);
2573 				lf.l_whence = SEEK_SET;
2574 				lf.l_start = 0;
2575 				lf.l_len = 0;
2576 				lf.l_type = F_UNLCK;
2577 				vp = fp->f_vnode;
2578 				(void) VOP_ADVLOCK(vp,
2579 				    (caddr_t)fdtol->fdl_leader, F_UNLCK, &lf,
2580 				    F_POSIX);
2581 				FILEDESC_XLOCK(fdp);
2582 				fdtol->fdl_holdcount--;
2583 				if (fdtol->fdl_holdcount == 0 &&
2584 				    fdtol->fdl_wakeup != 0) {
2585 					fdtol->fdl_wakeup = 0;
2586 					wakeup(fdtol);
2587 				}
2588 			}
2589 			FILEDESC_XUNLOCK(fdp);
2590 		}
2591 	}
2592 	return (fdrop(fp, td));
2593 }
2594 
2595 /*
2596  * Initialize the file pointer with the specified properties.
2597  *
2598  * The ops are set with release semantics to be certain that the flags, type,
2599  * and data are visible when ops is.  This is to prevent ops methods from being
2600  * called with bad data.
2601  */
2602 void
2603 finit(struct file *fp, u_int flag, short type, void *data, struct fileops *ops)
2604 {
2605 	fp->f_data = data;
2606 	fp->f_flag = flag;
2607 	fp->f_type = type;
2608 	atomic_store_rel_ptr((volatile uintptr_t *)&fp->f_ops, (uintptr_t)ops);
2609 }
2610 
2611 int
2612 fget_cap_locked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
2613     struct file **fpp, struct filecaps *havecapsp)
2614 {
2615 	struct filedescent *fde;
2616 	int error;
2617 
2618 	FILEDESC_LOCK_ASSERT(fdp);
2619 
2620 	fde = fdeget_locked(fdp, fd);
2621 	if (fde == NULL) {
2622 		error = EBADF;
2623 		goto out;
2624 	}
2625 
2626 #ifdef CAPABILITIES
2627 	error = cap_check(cap_rights_fde_inline(fde), needrightsp);
2628 	if (error != 0)
2629 		goto out;
2630 #endif
2631 
2632 	if (havecapsp != NULL)
2633 		filecaps_copy(&fde->fde_caps, havecapsp, true);
2634 
2635 	*fpp = fde->fde_file;
2636 
2637 	error = 0;
2638 out:
2639 	return (error);
2640 }
2641 
2642 int
2643 fget_cap(struct thread *td, int fd, cap_rights_t *needrightsp,
2644     struct file **fpp, struct filecaps *havecapsp)
2645 {
2646 	struct filedesc *fdp = td->td_proc->p_fd;
2647 	int error;
2648 #ifndef CAPABILITIES
2649 	error = fget_unlocked(fdp, fd, needrightsp, fpp, NULL);
2650 	if (error == 0 && havecapsp != NULL)
2651 		filecaps_fill(havecapsp);
2652 #else
2653 	struct file *fp;
2654 	seqc_t seq;
2655 
2656 	for (;;) {
2657 		error = fget_unlocked(fdp, fd, needrightsp, &fp, &seq);
2658 		if (error != 0)
2659 			return (error);
2660 
2661 		if (havecapsp != NULL) {
2662 			if (!filecaps_copy(&fdp->fd_ofiles[fd].fde_caps,
2663 			    havecapsp, false)) {
2664 				fdrop(fp, td);
2665 				goto get_locked;
2666 			}
2667 		}
2668 
2669 		if (!fd_modified(fdp, fd, seq))
2670 			break;
2671 		fdrop(fp, td);
2672 	}
2673 
2674 	*fpp = fp;
2675 	return (0);
2676 
2677 get_locked:
2678 	FILEDESC_SLOCK(fdp);
2679 	error = fget_cap_locked(fdp, fd, needrightsp, fpp, havecapsp);
2680 	if (error == 0 && !fhold(*fpp))
2681 		error = EBADF;
2682 	FILEDESC_SUNLOCK(fdp);
2683 #endif
2684 	return (error);
2685 }
2686 
2687 int
2688 fget_unlocked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
2689     struct file **fpp, seqc_t *seqp)
2690 {
2691 #ifdef CAPABILITIES
2692 	const struct filedescent *fde;
2693 #endif
2694 	const struct fdescenttbl *fdt;
2695 	struct file *fp;
2696 	u_int count;
2697 #ifdef CAPABILITIES
2698 	seqc_t seq;
2699 	cap_rights_t haverights;
2700 	int error;
2701 #endif
2702 
2703 	fdt = fdp->fd_files;
2704 	if (__predict_false((u_int)fd >= fdt->fdt_nfiles))
2705 		return (EBADF);
2706 	/*
2707 	 * Fetch the descriptor locklessly.  We avoid fdrop() races by
2708 	 * never raising a refcount above 0.  To accomplish this we have
2709 	 * to use a cmpset loop rather than an atomic_add.  The descriptor
2710 	 * must be re-verified once we acquire a reference to be certain
2711 	 * that the identity is still correct and we did not lose a race
2712 	 * due to preemption.
2713 	 */
2714 	for (;;) {
2715 #ifdef CAPABILITIES
2716 		seq = seqc_read(fd_seqc(fdt, fd));
2717 		fde = &fdt->fdt_ofiles[fd];
2718 		haverights = *cap_rights_fde_inline(fde);
2719 		fp = fde->fde_file;
2720 		if (!seqc_consistent(fd_seqc(fdt, fd), seq))
2721 			continue;
2722 #else
2723 		fp = fdt->fdt_ofiles[fd].fde_file;
2724 #endif
2725 		if (fp == NULL)
2726 			return (EBADF);
2727 #ifdef CAPABILITIES
2728 		error = cap_check(&haverights, needrightsp);
2729 		if (error != 0)
2730 			return (error);
2731 #endif
2732 		count = fp->f_count;
2733 	retry:
2734 		if (count == 0) {
2735 			/*
2736 			 * Force a reload. Other thread could reallocate the
2737 			 * table before this fd was closed, so it possible that
2738 			 * there is a stale fp pointer in cached version.
2739 			 */
2740 			fdt = *(const struct fdescenttbl * const volatile *)
2741 			    &(fdp->fd_files);
2742 			continue;
2743 		}
2744 		if (__predict_false(count + 1 < count))
2745 			return (EBADF);
2746 
2747 		/*
2748 		 * Use an acquire barrier to force re-reading of fdt so it is
2749 		 * refreshed for verification.
2750 		 */
2751 		if (__predict_false(atomic_fcmpset_acq_int(&fp->f_count,
2752 		    &count, count + 1) == 0))
2753 			goto retry;
2754 		fdt = fdp->fd_files;
2755 #ifdef	CAPABILITIES
2756 		if (seqc_consistent_nomb(fd_seqc(fdt, fd), seq))
2757 #else
2758 		if (fp == fdt->fdt_ofiles[fd].fde_file)
2759 #endif
2760 			break;
2761 		fdrop(fp, curthread);
2762 	}
2763 	*fpp = fp;
2764 	if (seqp != NULL) {
2765 #ifdef CAPABILITIES
2766 		*seqp = seq;
2767 #endif
2768 	}
2769 	return (0);
2770 }
2771 
2772 /*
2773  * Extract the file pointer associated with the specified descriptor for the
2774  * current user process.
2775  *
2776  * If the descriptor doesn't exist or doesn't match 'flags', EBADF is
2777  * returned.
2778  *
2779  * File's rights will be checked against the capability rights mask.
2780  *
2781  * If an error occurred the non-zero error is returned and *fpp is set to
2782  * NULL.  Otherwise *fpp is held and set and zero is returned.  Caller is
2783  * responsible for fdrop().
2784  */
2785 static __inline int
2786 _fget(struct thread *td, int fd, struct file **fpp, int flags,
2787     cap_rights_t *needrightsp, seqc_t *seqp)
2788 {
2789 	struct filedesc *fdp;
2790 	struct file *fp;
2791 	int error;
2792 
2793 	*fpp = NULL;
2794 	fdp = td->td_proc->p_fd;
2795 	error = fget_unlocked(fdp, fd, needrightsp, &fp, seqp);
2796 	if (error != 0)
2797 		return (error);
2798 	if (fp->f_ops == &badfileops) {
2799 		fdrop(fp, td);
2800 		return (EBADF);
2801 	}
2802 
2803 	/*
2804 	 * FREAD and FWRITE failure return EBADF as per POSIX.
2805 	 */
2806 	error = 0;
2807 	switch (flags) {
2808 	case FREAD:
2809 	case FWRITE:
2810 		if ((fp->f_flag & flags) == 0)
2811 			error = EBADF;
2812 		break;
2813 	case FEXEC:
2814 	    	if ((fp->f_flag & (FREAD | FEXEC)) == 0 ||
2815 		    ((fp->f_flag & FWRITE) != 0))
2816 			error = EBADF;
2817 		break;
2818 	case 0:
2819 		break;
2820 	default:
2821 		KASSERT(0, ("wrong flags"));
2822 	}
2823 
2824 	if (error != 0) {
2825 		fdrop(fp, td);
2826 		return (error);
2827 	}
2828 
2829 	*fpp = fp;
2830 	return (0);
2831 }
2832 
2833 int
2834 fget(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
2835 {
2836 
2837 	return (_fget(td, fd, fpp, 0, rightsp, NULL));
2838 }
2839 
2840 int
2841 fget_mmap(struct thread *td, int fd, cap_rights_t *rightsp, u_char *maxprotp,
2842     struct file **fpp)
2843 {
2844 	int error;
2845 #ifndef CAPABILITIES
2846 	error = _fget(td, fd, fpp, 0, rightsp, NULL);
2847 	if (maxprotp != NULL)
2848 		*maxprotp = VM_PROT_ALL;
2849 #else
2850 	cap_rights_t fdrights;
2851 	struct filedesc *fdp = td->td_proc->p_fd;
2852 	seqc_t seq;
2853 
2854 	MPASS(cap_rights_is_set(rightsp, CAP_MMAP));
2855 	for (;;) {
2856 		error = _fget(td, fd, fpp, 0, rightsp, &seq);
2857 		if (error != 0)
2858 			return (error);
2859 		if (maxprotp != NULL)
2860 			fdrights = *cap_rights(fdp, fd);
2861 		if (!fd_modified(fdp, fd, seq))
2862 			break;
2863 		fdrop(*fpp, td);
2864 	}
2865 
2866 	/*
2867 	 * If requested, convert capability rights to access flags.
2868 	 */
2869 	if (maxprotp != NULL)
2870 		*maxprotp = cap_rights_to_vmprot(&fdrights);
2871 #endif
2872 	return (error);
2873 }
2874 
2875 int
2876 fget_read(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
2877 {
2878 
2879 	return (_fget(td, fd, fpp, FREAD, rightsp, NULL));
2880 }
2881 
2882 int
2883 fget_write(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
2884 {
2885 
2886 	return (_fget(td, fd, fpp, FWRITE, rightsp, NULL));
2887 }
2888 
2889 int
2890 fget_fcntl(struct thread *td, int fd, cap_rights_t *rightsp, int needfcntl,
2891     struct file **fpp)
2892 {
2893 	struct filedesc *fdp = td->td_proc->p_fd;
2894 #ifndef CAPABILITIES
2895 	return (fget_unlocked(fdp, fd, rightsp, fpp, NULL));
2896 #else
2897 	int error;
2898 	seqc_t seq;
2899 
2900 	MPASS(cap_rights_is_set(rightsp, CAP_FCNTL));
2901 	for (;;) {
2902 		error = fget_unlocked(fdp, fd, rightsp, fpp, &seq);
2903 		if (error != 0)
2904 			return (error);
2905 		error = cap_fcntl_check(fdp, fd, needfcntl);
2906 		if (!fd_modified(fdp, fd, seq))
2907 			break;
2908 		fdrop(*fpp, td);
2909 	}
2910 	if (error != 0) {
2911 		fdrop(*fpp, td);
2912 		*fpp = NULL;
2913 	}
2914 	return (error);
2915 #endif
2916 }
2917 
2918 /*
2919  * Like fget() but loads the underlying vnode, or returns an error if the
2920  * descriptor does not represent a vnode.  Note that pipes use vnodes but
2921  * never have VM objects.  The returned vnode will be vref()'d.
2922  *
2923  * XXX: what about the unused flags ?
2924  */
2925 static __inline int
2926 _fgetvp(struct thread *td, int fd, int flags, cap_rights_t *needrightsp,
2927     struct vnode **vpp)
2928 {
2929 	struct file *fp;
2930 	int error;
2931 
2932 	*vpp = NULL;
2933 	error = _fget(td, fd, &fp, flags, needrightsp, NULL);
2934 	if (error != 0)
2935 		return (error);
2936 	if (fp->f_vnode == NULL) {
2937 		error = EINVAL;
2938 	} else {
2939 		*vpp = fp->f_vnode;
2940 		vrefact(*vpp);
2941 	}
2942 	fdrop(fp, td);
2943 
2944 	return (error);
2945 }
2946 
2947 int
2948 fgetvp(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
2949 {
2950 
2951 	return (_fgetvp(td, fd, 0, rightsp, vpp));
2952 }
2953 
2954 int
2955 fgetvp_rights(struct thread *td, int fd, cap_rights_t *needrightsp,
2956     struct filecaps *havecaps, struct vnode **vpp)
2957 {
2958 	struct filedesc *fdp;
2959 	struct filecaps caps;
2960 	struct file *fp;
2961 	int error;
2962 
2963 	fdp = td->td_proc->p_fd;
2964 	error = fget_cap_locked(fdp, fd, needrightsp, &fp, &caps);
2965 	if (error != 0)
2966 		return (error);
2967 	if (fp->f_ops == &badfileops) {
2968 		error = EBADF;
2969 		goto out;
2970 	}
2971 	if (fp->f_vnode == NULL) {
2972 		error = EINVAL;
2973 		goto out;
2974 	}
2975 
2976 	*havecaps = caps;
2977 	*vpp = fp->f_vnode;
2978 	vrefact(*vpp);
2979 
2980 	return (0);
2981 out:
2982 	filecaps_free(&caps);
2983 	return (error);
2984 }
2985 
2986 int
2987 fgetvp_read(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
2988 {
2989 
2990 	return (_fgetvp(td, fd, FREAD, rightsp, vpp));
2991 }
2992 
2993 int
2994 fgetvp_exec(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
2995 {
2996 
2997 	return (_fgetvp(td, fd, FEXEC, rightsp, vpp));
2998 }
2999 
3000 #ifdef notyet
3001 int
3002 fgetvp_write(struct thread *td, int fd, cap_rights_t *rightsp,
3003     struct vnode **vpp)
3004 {
3005 
3006 	return (_fgetvp(td, fd, FWRITE, rightsp, vpp));
3007 }
3008 #endif
3009 
3010 /*
3011  * Handle the last reference to a file being closed.
3012  *
3013  * Without the noinline attribute clang keeps inlining the func thorough this
3014  * file when fdrop is used.
3015  */
3016 int __noinline
3017 _fdrop(struct file *fp, struct thread *td)
3018 {
3019 	int error;
3020 
3021 	if (fp->f_count != 0)
3022 		panic("fdrop: count %d", fp->f_count);
3023 	error = fo_close(fp, td);
3024 	atomic_subtract_int(&openfiles, 1);
3025 	crfree(fp->f_cred);
3026 	free(fp->f_advice, M_FADVISE);
3027 	uma_zfree(file_zone, fp);
3028 
3029 	return (error);
3030 }
3031 
3032 /*
3033  * Apply an advisory lock on a file descriptor.
3034  *
3035  * Just attempt to get a record lock of the requested type on the entire file
3036  * (l_whence = SEEK_SET, l_start = 0, l_len = 0).
3037  */
3038 #ifndef _SYS_SYSPROTO_H_
3039 struct flock_args {
3040 	int	fd;
3041 	int	how;
3042 };
3043 #endif
3044 /* ARGSUSED */
3045 int
3046 sys_flock(struct thread *td, struct flock_args *uap)
3047 {
3048 	struct file *fp;
3049 	struct vnode *vp;
3050 	struct flock lf;
3051 	int error;
3052 
3053 	error = fget(td, uap->fd, &cap_flock_rights, &fp);
3054 	if (error != 0)
3055 		return (error);
3056 	if (fp->f_type != DTYPE_VNODE) {
3057 		fdrop(fp, td);
3058 		return (EOPNOTSUPP);
3059 	}
3060 
3061 	vp = fp->f_vnode;
3062 	lf.l_whence = SEEK_SET;
3063 	lf.l_start = 0;
3064 	lf.l_len = 0;
3065 	if (uap->how & LOCK_UN) {
3066 		lf.l_type = F_UNLCK;
3067 		atomic_clear_int(&fp->f_flag, FHASLOCK);
3068 		error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK);
3069 		goto done2;
3070 	}
3071 	if (uap->how & LOCK_EX)
3072 		lf.l_type = F_WRLCK;
3073 	else if (uap->how & LOCK_SH)
3074 		lf.l_type = F_RDLCK;
3075 	else {
3076 		error = EBADF;
3077 		goto done2;
3078 	}
3079 	atomic_set_int(&fp->f_flag, FHASLOCK);
3080 	error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf,
3081 	    (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT);
3082 done2:
3083 	fdrop(fp, td);
3084 	return (error);
3085 }
3086 /*
3087  * Duplicate the specified descriptor to a free descriptor.
3088  */
3089 int
3090 dupfdopen(struct thread *td, struct filedesc *fdp, int dfd, int mode,
3091     int openerror, int *indxp)
3092 {
3093 	struct filedescent *newfde, *oldfde;
3094 	struct file *fp;
3095 	u_long *ioctls;
3096 	int error, indx;
3097 
3098 	KASSERT(openerror == ENODEV || openerror == ENXIO,
3099 	    ("unexpected error %d in %s", openerror, __func__));
3100 
3101 	/*
3102 	 * If the to-be-dup'd fd number is greater than the allowed number
3103 	 * of file descriptors, or the fd to be dup'd has already been
3104 	 * closed, then reject.
3105 	 */
3106 	FILEDESC_XLOCK(fdp);
3107 	if ((fp = fget_locked(fdp, dfd)) == NULL) {
3108 		FILEDESC_XUNLOCK(fdp);
3109 		return (EBADF);
3110 	}
3111 
3112 	error = fdalloc(td, 0, &indx);
3113 	if (error != 0) {
3114 		FILEDESC_XUNLOCK(fdp);
3115 		return (error);
3116 	}
3117 
3118 	/*
3119 	 * There are two cases of interest here.
3120 	 *
3121 	 * For ENODEV simply dup (dfd) to file descriptor (indx) and return.
3122 	 *
3123 	 * For ENXIO steal away the file structure from (dfd) and store it in
3124 	 * (indx).  (dfd) is effectively closed by this operation.
3125 	 */
3126 	switch (openerror) {
3127 	case ENODEV:
3128 		/*
3129 		 * Check that the mode the file is being opened for is a
3130 		 * subset of the mode of the existing descriptor.
3131 		 */
3132 		if (((mode & (FREAD|FWRITE)) | fp->f_flag) != fp->f_flag) {
3133 			fdunused(fdp, indx);
3134 			FILEDESC_XUNLOCK(fdp);
3135 			return (EACCES);
3136 		}
3137 		if (!fhold(fp)) {
3138 			fdunused(fdp, indx);
3139 			FILEDESC_XUNLOCK(fdp);
3140 			return (EBADF);
3141 		}
3142 		newfde = &fdp->fd_ofiles[indx];
3143 		oldfde = &fdp->fd_ofiles[dfd];
3144 		ioctls = filecaps_copy_prep(&oldfde->fde_caps);
3145 #ifdef CAPABILITIES
3146 		seqc_write_begin(&newfde->fde_seqc);
3147 #endif
3148 		memcpy(newfde, oldfde, fde_change_size);
3149 		filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps,
3150 		    ioctls);
3151 #ifdef CAPABILITIES
3152 		seqc_write_end(&newfde->fde_seqc);
3153 #endif
3154 		break;
3155 	case ENXIO:
3156 		/*
3157 		 * Steal away the file pointer from dfd and stuff it into indx.
3158 		 */
3159 		newfde = &fdp->fd_ofiles[indx];
3160 		oldfde = &fdp->fd_ofiles[dfd];
3161 #ifdef CAPABILITIES
3162 		seqc_write_begin(&newfde->fde_seqc);
3163 #endif
3164 		memcpy(newfde, oldfde, fde_change_size);
3165 		oldfde->fde_file = NULL;
3166 		fdunused(fdp, dfd);
3167 #ifdef CAPABILITIES
3168 		seqc_write_end(&newfde->fde_seqc);
3169 #endif
3170 		break;
3171 	}
3172 	FILEDESC_XUNLOCK(fdp);
3173 	*indxp = indx;
3174 	return (0);
3175 }
3176 
3177 /*
3178  * This sysctl determines if we will allow a process to chroot(2) if it
3179  * has a directory open:
3180  *	0: disallowed for all processes.
3181  *	1: allowed for processes that were not already chroot(2)'ed.
3182  *	2: allowed for all processes.
3183  */
3184 
3185 static int chroot_allow_open_directories = 1;
3186 
3187 SYSCTL_INT(_kern, OID_AUTO, chroot_allow_open_directories, CTLFLAG_RW,
3188     &chroot_allow_open_directories, 0,
3189     "Allow a process to chroot(2) if it has a directory open");
3190 
3191 /*
3192  * Helper function for raised chroot(2) security function:  Refuse if
3193  * any filedescriptors are open directories.
3194  */
3195 static int
3196 chroot_refuse_vdir_fds(struct filedesc *fdp)
3197 {
3198 	struct vnode *vp;
3199 	struct file *fp;
3200 	int fd;
3201 
3202 	FILEDESC_LOCK_ASSERT(fdp);
3203 
3204 	for (fd = 0; fd <= fdp->fd_lastfile; fd++) {
3205 		fp = fget_locked(fdp, fd);
3206 		if (fp == NULL)
3207 			continue;
3208 		if (fp->f_type == DTYPE_VNODE) {
3209 			vp = fp->f_vnode;
3210 			if (vp->v_type == VDIR)
3211 				return (EPERM);
3212 		}
3213 	}
3214 	return (0);
3215 }
3216 
3217 /*
3218  * Common routine for kern_chroot() and jail_attach().  The caller is
3219  * responsible for invoking priv_check() and mac_vnode_check_chroot() to
3220  * authorize this operation.
3221  */
3222 int
3223 pwd_chroot(struct thread *td, struct vnode *vp)
3224 {
3225 	struct filedesc *fdp;
3226 	struct vnode *oldvp;
3227 	int error;
3228 
3229 	fdp = td->td_proc->p_fd;
3230 	FILEDESC_XLOCK(fdp);
3231 	if (chroot_allow_open_directories == 0 ||
3232 	    (chroot_allow_open_directories == 1 && fdp->fd_rdir != rootvnode)) {
3233 		error = chroot_refuse_vdir_fds(fdp);
3234 		if (error != 0) {
3235 			FILEDESC_XUNLOCK(fdp);
3236 			return (error);
3237 		}
3238 	}
3239 	oldvp = fdp->fd_rdir;
3240 	vrefact(vp);
3241 	fdp->fd_rdir = vp;
3242 	if (fdp->fd_jdir == NULL) {
3243 		vrefact(vp);
3244 		fdp->fd_jdir = vp;
3245 	}
3246 	FILEDESC_XUNLOCK(fdp);
3247 	vrele(oldvp);
3248 	return (0);
3249 }
3250 
3251 void
3252 pwd_chdir(struct thread *td, struct vnode *vp)
3253 {
3254 	struct filedesc *fdp;
3255 	struct vnode *oldvp;
3256 
3257 	fdp = td->td_proc->p_fd;
3258 	FILEDESC_XLOCK(fdp);
3259 	VNASSERT(vp->v_usecount > 0, vp,
3260 	    ("chdir to a vnode with zero usecount"));
3261 	oldvp = fdp->fd_cdir;
3262 	fdp->fd_cdir = vp;
3263 	FILEDESC_XUNLOCK(fdp);
3264 	vrele(oldvp);
3265 }
3266 
3267 /*
3268  * Scan all active processes and prisons to see if any of them have a current
3269  * or root directory of `olddp'. If so, replace them with the new mount point.
3270  */
3271 void
3272 mountcheckdirs(struct vnode *olddp, struct vnode *newdp)
3273 {
3274 	struct filedesc *fdp;
3275 	struct prison *pr;
3276 	struct proc *p;
3277 	int nrele;
3278 
3279 	if (vrefcnt(olddp) == 1)
3280 		return;
3281 	nrele = 0;
3282 	sx_slock(&allproc_lock);
3283 	FOREACH_PROC_IN_SYSTEM(p) {
3284 		PROC_LOCK(p);
3285 		fdp = fdhold(p);
3286 		PROC_UNLOCK(p);
3287 		if (fdp == NULL)
3288 			continue;
3289 		FILEDESC_XLOCK(fdp);
3290 		if (fdp->fd_cdir == olddp) {
3291 			vrefact(newdp);
3292 			fdp->fd_cdir = newdp;
3293 			nrele++;
3294 		}
3295 		if (fdp->fd_rdir == olddp) {
3296 			vrefact(newdp);
3297 			fdp->fd_rdir = newdp;
3298 			nrele++;
3299 		}
3300 		if (fdp->fd_jdir == olddp) {
3301 			vrefact(newdp);
3302 			fdp->fd_jdir = newdp;
3303 			nrele++;
3304 		}
3305 		FILEDESC_XUNLOCK(fdp);
3306 		fddrop(fdp);
3307 	}
3308 	sx_sunlock(&allproc_lock);
3309 	if (rootvnode == olddp) {
3310 		vrefact(newdp);
3311 		rootvnode = newdp;
3312 		nrele++;
3313 	}
3314 	mtx_lock(&prison0.pr_mtx);
3315 	if (prison0.pr_root == olddp) {
3316 		vrefact(newdp);
3317 		prison0.pr_root = newdp;
3318 		nrele++;
3319 	}
3320 	mtx_unlock(&prison0.pr_mtx);
3321 	sx_slock(&allprison_lock);
3322 	TAILQ_FOREACH(pr, &allprison, pr_list) {
3323 		mtx_lock(&pr->pr_mtx);
3324 		if (pr->pr_root == olddp) {
3325 			vrefact(newdp);
3326 			pr->pr_root = newdp;
3327 			nrele++;
3328 		}
3329 		mtx_unlock(&pr->pr_mtx);
3330 	}
3331 	sx_sunlock(&allprison_lock);
3332 	while (nrele--)
3333 		vrele(olddp);
3334 }
3335 
3336 struct filedesc_to_leader *
3337 filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader)
3338 {
3339 	struct filedesc_to_leader *fdtol;
3340 
3341 	fdtol = malloc(sizeof(struct filedesc_to_leader),
3342 	    M_FILEDESC_TO_LEADER, M_WAITOK);
3343 	fdtol->fdl_refcount = 1;
3344 	fdtol->fdl_holdcount = 0;
3345 	fdtol->fdl_wakeup = 0;
3346 	fdtol->fdl_leader = leader;
3347 	if (old != NULL) {
3348 		FILEDESC_XLOCK(fdp);
3349 		fdtol->fdl_next = old->fdl_next;
3350 		fdtol->fdl_prev = old;
3351 		old->fdl_next = fdtol;
3352 		fdtol->fdl_next->fdl_prev = fdtol;
3353 		FILEDESC_XUNLOCK(fdp);
3354 	} else {
3355 		fdtol->fdl_next = fdtol;
3356 		fdtol->fdl_prev = fdtol;
3357 	}
3358 	return (fdtol);
3359 }
3360 
3361 static int
3362 sysctl_kern_proc_nfds(SYSCTL_HANDLER_ARGS)
3363 {
3364 	struct filedesc *fdp;
3365 	int i, count, slots;
3366 
3367 	if (*(int *)arg1 != 0)
3368 		return (EINVAL);
3369 
3370 	fdp = curproc->p_fd;
3371 	count = 0;
3372 	FILEDESC_SLOCK(fdp);
3373 	slots = NDSLOTS(fdp->fd_lastfile + 1);
3374 	for (i = 0; i < slots; i++)
3375 		count += bitcountl(fdp->fd_map[i]);
3376 	FILEDESC_SUNLOCK(fdp);
3377 
3378 	return (SYSCTL_OUT(req, &count, sizeof(count)));
3379 }
3380 
3381 static SYSCTL_NODE(_kern_proc, KERN_PROC_NFDS, nfds,
3382     CTLFLAG_RD|CTLFLAG_CAPRD|CTLFLAG_MPSAFE, sysctl_kern_proc_nfds,
3383     "Number of open file descriptors");
3384 
3385 /*
3386  * Get file structures globally.
3387  */
3388 static int
3389 sysctl_kern_file(SYSCTL_HANDLER_ARGS)
3390 {
3391 	struct xfile xf;
3392 	struct filedesc *fdp;
3393 	struct file *fp;
3394 	struct proc *p;
3395 	int error, n;
3396 
3397 	error = sysctl_wire_old_buffer(req, 0);
3398 	if (error != 0)
3399 		return (error);
3400 	if (req->oldptr == NULL) {
3401 		n = 0;
3402 		sx_slock(&allproc_lock);
3403 		FOREACH_PROC_IN_SYSTEM(p) {
3404 			PROC_LOCK(p);
3405 			if (p->p_state == PRS_NEW) {
3406 				PROC_UNLOCK(p);
3407 				continue;
3408 			}
3409 			fdp = fdhold(p);
3410 			PROC_UNLOCK(p);
3411 			if (fdp == NULL)
3412 				continue;
3413 			/* overestimates sparse tables. */
3414 			if (fdp->fd_lastfile > 0)
3415 				n += fdp->fd_lastfile;
3416 			fddrop(fdp);
3417 		}
3418 		sx_sunlock(&allproc_lock);
3419 		return (SYSCTL_OUT(req, 0, n * sizeof(xf)));
3420 	}
3421 	error = 0;
3422 	bzero(&xf, sizeof(xf));
3423 	xf.xf_size = sizeof(xf);
3424 	sx_slock(&allproc_lock);
3425 	FOREACH_PROC_IN_SYSTEM(p) {
3426 		PROC_LOCK(p);
3427 		if (p->p_state == PRS_NEW) {
3428 			PROC_UNLOCK(p);
3429 			continue;
3430 		}
3431 		if (p_cansee(req->td, p) != 0) {
3432 			PROC_UNLOCK(p);
3433 			continue;
3434 		}
3435 		xf.xf_pid = p->p_pid;
3436 		xf.xf_uid = p->p_ucred->cr_uid;
3437 		fdp = fdhold(p);
3438 		PROC_UNLOCK(p);
3439 		if (fdp == NULL)
3440 			continue;
3441 		FILEDESC_SLOCK(fdp);
3442 		for (n = 0; fdp->fd_refcnt > 0 && n <= fdp->fd_lastfile; ++n) {
3443 			if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
3444 				continue;
3445 			xf.xf_fd = n;
3446 			xf.xf_file = (uintptr_t)fp;
3447 			xf.xf_data = (uintptr_t)fp->f_data;
3448 			xf.xf_vnode = (uintptr_t)fp->f_vnode;
3449 			xf.xf_type = (uintptr_t)fp->f_type;
3450 			xf.xf_count = fp->f_count;
3451 			xf.xf_msgcount = 0;
3452 			xf.xf_offset = foffset_get(fp);
3453 			xf.xf_flag = fp->f_flag;
3454 			error = SYSCTL_OUT(req, &xf, sizeof(xf));
3455 			if (error)
3456 				break;
3457 		}
3458 		FILEDESC_SUNLOCK(fdp);
3459 		fddrop(fdp);
3460 		if (error)
3461 			break;
3462 	}
3463 	sx_sunlock(&allproc_lock);
3464 	return (error);
3465 }
3466 
3467 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD|CTLFLAG_MPSAFE,
3468     0, 0, sysctl_kern_file, "S,xfile", "Entire file table");
3469 
3470 #ifdef KINFO_FILE_SIZE
3471 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
3472 #endif
3473 
3474 static int
3475 xlate_fflags(int fflags)
3476 {
3477 	static const struct {
3478 		int	fflag;
3479 		int	kf_fflag;
3480 	} fflags_table[] = {
3481 		{ FAPPEND, KF_FLAG_APPEND },
3482 		{ FASYNC, KF_FLAG_ASYNC },
3483 		{ FFSYNC, KF_FLAG_FSYNC },
3484 		{ FHASLOCK, KF_FLAG_HASLOCK },
3485 		{ FNONBLOCK, KF_FLAG_NONBLOCK },
3486 		{ FREAD, KF_FLAG_READ },
3487 		{ FWRITE, KF_FLAG_WRITE },
3488 		{ O_CREAT, KF_FLAG_CREAT },
3489 		{ O_DIRECT, KF_FLAG_DIRECT },
3490 		{ O_EXCL, KF_FLAG_EXCL },
3491 		{ O_EXEC, KF_FLAG_EXEC },
3492 		{ O_EXLOCK, KF_FLAG_EXLOCK },
3493 		{ O_NOFOLLOW, KF_FLAG_NOFOLLOW },
3494 		{ O_SHLOCK, KF_FLAG_SHLOCK },
3495 		{ O_TRUNC, KF_FLAG_TRUNC }
3496 	};
3497 	unsigned int i;
3498 	int kflags;
3499 
3500 	kflags = 0;
3501 	for (i = 0; i < nitems(fflags_table); i++)
3502 		if (fflags & fflags_table[i].fflag)
3503 			kflags |=  fflags_table[i].kf_fflag;
3504 	return (kflags);
3505 }
3506 
3507 /* Trim unused data from kf_path by truncating the structure size. */
3508 void
3509 pack_kinfo(struct kinfo_file *kif)
3510 {
3511 
3512 	kif->kf_structsize = offsetof(struct kinfo_file, kf_path) +
3513 	    strlen(kif->kf_path) + 1;
3514 	kif->kf_structsize = roundup(kif->kf_structsize, sizeof(uint64_t));
3515 }
3516 
3517 static void
3518 export_file_to_kinfo(struct file *fp, int fd, cap_rights_t *rightsp,
3519     struct kinfo_file *kif, struct filedesc *fdp, int flags)
3520 {
3521 	int error;
3522 
3523 	bzero(kif, sizeof(*kif));
3524 
3525 	/* Set a default type to allow for empty fill_kinfo() methods. */
3526 	kif->kf_type = KF_TYPE_UNKNOWN;
3527 	kif->kf_flags = xlate_fflags(fp->f_flag);
3528 	if (rightsp != NULL)
3529 		kif->kf_cap_rights = *rightsp;
3530 	else
3531 		cap_rights_init(&kif->kf_cap_rights);
3532 	kif->kf_fd = fd;
3533 	kif->kf_ref_count = fp->f_count;
3534 	kif->kf_offset = foffset_get(fp);
3535 
3536 	/*
3537 	 * This may drop the filedesc lock, so the 'fp' cannot be
3538 	 * accessed after this call.
3539 	 */
3540 	error = fo_fill_kinfo(fp, kif, fdp);
3541 	if (error == 0)
3542 		kif->kf_status |= KF_ATTR_VALID;
3543 	if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
3544 		pack_kinfo(kif);
3545 	else
3546 		kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
3547 }
3548 
3549 static void
3550 export_vnode_to_kinfo(struct vnode *vp, int fd, int fflags,
3551     struct kinfo_file *kif, int flags)
3552 {
3553 	int error;
3554 
3555 	bzero(kif, sizeof(*kif));
3556 
3557 	kif->kf_type = KF_TYPE_VNODE;
3558 	error = vn_fill_kinfo_vnode(vp, kif);
3559 	if (error == 0)
3560 		kif->kf_status |= KF_ATTR_VALID;
3561 	kif->kf_flags = xlate_fflags(fflags);
3562 	cap_rights_init(&kif->kf_cap_rights);
3563 	kif->kf_fd = fd;
3564 	kif->kf_ref_count = -1;
3565 	kif->kf_offset = -1;
3566 	if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
3567 		pack_kinfo(kif);
3568 	else
3569 		kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
3570 	vrele(vp);
3571 }
3572 
3573 struct export_fd_buf {
3574 	struct filedesc		*fdp;
3575 	struct sbuf 		*sb;
3576 	ssize_t			remainder;
3577 	struct kinfo_file	kif;
3578 	int			flags;
3579 };
3580 
3581 static int
3582 export_kinfo_to_sb(struct export_fd_buf *efbuf)
3583 {
3584 	struct kinfo_file *kif;
3585 
3586 	kif = &efbuf->kif;
3587 	if (efbuf->remainder != -1) {
3588 		if (efbuf->remainder < kif->kf_structsize) {
3589 			/* Terminate export. */
3590 			efbuf->remainder = 0;
3591 			return (0);
3592 		}
3593 		efbuf->remainder -= kif->kf_structsize;
3594 	}
3595 	return (sbuf_bcat(efbuf->sb, kif, kif->kf_structsize) == 0 ? 0 : ENOMEM);
3596 }
3597 
3598 static int
3599 export_file_to_sb(struct file *fp, int fd, cap_rights_t *rightsp,
3600     struct export_fd_buf *efbuf)
3601 {
3602 	int error;
3603 
3604 	if (efbuf->remainder == 0)
3605 		return (0);
3606 	export_file_to_kinfo(fp, fd, rightsp, &efbuf->kif, efbuf->fdp,
3607 	    efbuf->flags);
3608 	FILEDESC_SUNLOCK(efbuf->fdp);
3609 	error = export_kinfo_to_sb(efbuf);
3610 	FILEDESC_SLOCK(efbuf->fdp);
3611 	return (error);
3612 }
3613 
3614 static int
3615 export_vnode_to_sb(struct vnode *vp, int fd, int fflags,
3616     struct export_fd_buf *efbuf)
3617 {
3618 	int error;
3619 
3620 	if (efbuf->remainder == 0)
3621 		return (0);
3622 	if (efbuf->fdp != NULL)
3623 		FILEDESC_SUNLOCK(efbuf->fdp);
3624 	export_vnode_to_kinfo(vp, fd, fflags, &efbuf->kif, efbuf->flags);
3625 	error = export_kinfo_to_sb(efbuf);
3626 	if (efbuf->fdp != NULL)
3627 		FILEDESC_SLOCK(efbuf->fdp);
3628 	return (error);
3629 }
3630 
3631 /*
3632  * Store a process file descriptor information to sbuf.
3633  *
3634  * Takes a locked proc as argument, and returns with the proc unlocked.
3635  */
3636 int
3637 kern_proc_filedesc_out(struct proc *p,  struct sbuf *sb, ssize_t maxlen,
3638     int flags)
3639 {
3640 	struct file *fp;
3641 	struct filedesc *fdp;
3642 	struct export_fd_buf *efbuf;
3643 	struct vnode *cttyvp, *textvp, *tracevp;
3644 	int error, i;
3645 	cap_rights_t rights;
3646 
3647 	PROC_LOCK_ASSERT(p, MA_OWNED);
3648 
3649 	/* ktrace vnode */
3650 	tracevp = p->p_tracevp;
3651 	if (tracevp != NULL)
3652 		vrefact(tracevp);
3653 	/* text vnode */
3654 	textvp = p->p_textvp;
3655 	if (textvp != NULL)
3656 		vrefact(textvp);
3657 	/* Controlling tty. */
3658 	cttyvp = NULL;
3659 	if (p->p_pgrp != NULL && p->p_pgrp->pg_session != NULL) {
3660 		cttyvp = p->p_pgrp->pg_session->s_ttyvp;
3661 		if (cttyvp != NULL)
3662 			vrefact(cttyvp);
3663 	}
3664 	fdp = fdhold(p);
3665 	PROC_UNLOCK(p);
3666 	efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
3667 	efbuf->fdp = NULL;
3668 	efbuf->sb = sb;
3669 	efbuf->remainder = maxlen;
3670 	efbuf->flags = flags;
3671 	if (tracevp != NULL)
3672 		export_vnode_to_sb(tracevp, KF_FD_TYPE_TRACE, FREAD | FWRITE,
3673 		    efbuf);
3674 	if (textvp != NULL)
3675 		export_vnode_to_sb(textvp, KF_FD_TYPE_TEXT, FREAD, efbuf);
3676 	if (cttyvp != NULL)
3677 		export_vnode_to_sb(cttyvp, KF_FD_TYPE_CTTY, FREAD | FWRITE,
3678 		    efbuf);
3679 	error = 0;
3680 	if (fdp == NULL)
3681 		goto fail;
3682 	efbuf->fdp = fdp;
3683 	FILEDESC_SLOCK(fdp);
3684 	/* working directory */
3685 	if (fdp->fd_cdir != NULL) {
3686 		vrefact(fdp->fd_cdir);
3687 		export_vnode_to_sb(fdp->fd_cdir, KF_FD_TYPE_CWD, FREAD, efbuf);
3688 	}
3689 	/* root directory */
3690 	if (fdp->fd_rdir != NULL) {
3691 		vrefact(fdp->fd_rdir);
3692 		export_vnode_to_sb(fdp->fd_rdir, KF_FD_TYPE_ROOT, FREAD, efbuf);
3693 	}
3694 	/* jail directory */
3695 	if (fdp->fd_jdir != NULL) {
3696 		vrefact(fdp->fd_jdir);
3697 		export_vnode_to_sb(fdp->fd_jdir, KF_FD_TYPE_JAIL, FREAD, efbuf);
3698 	}
3699 	for (i = 0; fdp->fd_refcnt > 0 && i <= fdp->fd_lastfile; i++) {
3700 		if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
3701 			continue;
3702 #ifdef CAPABILITIES
3703 		rights = *cap_rights(fdp, i);
3704 #else /* !CAPABILITIES */
3705 		rights = cap_no_rights;
3706 #endif
3707 		/*
3708 		 * Create sysctl entry.  It is OK to drop the filedesc
3709 		 * lock inside of export_file_to_sb() as we will
3710 		 * re-validate and re-evaluate its properties when the
3711 		 * loop continues.
3712 		 */
3713 		error = export_file_to_sb(fp, i, &rights, efbuf);
3714 		if (error != 0 || efbuf->remainder == 0)
3715 			break;
3716 	}
3717 	FILEDESC_SUNLOCK(fdp);
3718 	fddrop(fdp);
3719 fail:
3720 	free(efbuf, M_TEMP);
3721 	return (error);
3722 }
3723 
3724 #define FILEDESC_SBUF_SIZE	(sizeof(struct kinfo_file) * 5)
3725 
3726 /*
3727  * Get per-process file descriptors for use by procstat(1), et al.
3728  */
3729 static int
3730 sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS)
3731 {
3732 	struct sbuf sb;
3733 	struct proc *p;
3734 	ssize_t maxlen;
3735 	int error, error2, *name;
3736 
3737 	name = (int *)arg1;
3738 
3739 	sbuf_new_for_sysctl(&sb, NULL, FILEDESC_SBUF_SIZE, req);
3740 	sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
3741 	error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
3742 	if (error != 0) {
3743 		sbuf_delete(&sb);
3744 		return (error);
3745 	}
3746 	maxlen = req->oldptr != NULL ? req->oldlen : -1;
3747 	error = kern_proc_filedesc_out(p, &sb, maxlen,
3748 	    KERN_FILEDESC_PACK_KINFO);
3749 	error2 = sbuf_finish(&sb);
3750 	sbuf_delete(&sb);
3751 	return (error != 0 ? error : error2);
3752 }
3753 
3754 #ifdef COMPAT_FREEBSD7
3755 #ifdef KINFO_OFILE_SIZE
3756 CTASSERT(sizeof(struct kinfo_ofile) == KINFO_OFILE_SIZE);
3757 #endif
3758 
3759 static void
3760 kinfo_to_okinfo(struct kinfo_file *kif, struct kinfo_ofile *okif)
3761 {
3762 
3763 	okif->kf_structsize = sizeof(*okif);
3764 	okif->kf_type = kif->kf_type;
3765 	okif->kf_fd = kif->kf_fd;
3766 	okif->kf_ref_count = kif->kf_ref_count;
3767 	okif->kf_flags = kif->kf_flags & (KF_FLAG_READ | KF_FLAG_WRITE |
3768 	    KF_FLAG_APPEND | KF_FLAG_ASYNC | KF_FLAG_FSYNC | KF_FLAG_NONBLOCK |
3769 	    KF_FLAG_DIRECT | KF_FLAG_HASLOCK);
3770 	okif->kf_offset = kif->kf_offset;
3771 	if (kif->kf_type == KF_TYPE_VNODE)
3772 		okif->kf_vnode_type = kif->kf_un.kf_file.kf_file_type;
3773 	else
3774 		okif->kf_vnode_type = KF_VTYPE_VNON;
3775 	strlcpy(okif->kf_path, kif->kf_path, sizeof(okif->kf_path));
3776 	if (kif->kf_type == KF_TYPE_SOCKET) {
3777 		okif->kf_sock_domain = kif->kf_un.kf_sock.kf_sock_domain0;
3778 		okif->kf_sock_type = kif->kf_un.kf_sock.kf_sock_type0;
3779 		okif->kf_sock_protocol = kif->kf_un.kf_sock.kf_sock_protocol0;
3780 		okif->kf_sa_local = kif->kf_un.kf_sock.kf_sa_local;
3781 		okif->kf_sa_peer = kif->kf_un.kf_sock.kf_sa_peer;
3782 	} else {
3783 		okif->kf_sa_local.ss_family = AF_UNSPEC;
3784 		okif->kf_sa_peer.ss_family = AF_UNSPEC;
3785 	}
3786 }
3787 
3788 static int
3789 export_vnode_for_osysctl(struct vnode *vp, int type, struct kinfo_file *kif,
3790     struct kinfo_ofile *okif, struct filedesc *fdp, struct sysctl_req *req)
3791 {
3792 	int error;
3793 
3794 	vrefact(vp);
3795 	FILEDESC_SUNLOCK(fdp);
3796 	export_vnode_to_kinfo(vp, type, 0, kif, KERN_FILEDESC_PACK_KINFO);
3797 	kinfo_to_okinfo(kif, okif);
3798 	error = SYSCTL_OUT(req, okif, sizeof(*okif));
3799 	FILEDESC_SLOCK(fdp);
3800 	return (error);
3801 }
3802 
3803 /*
3804  * Get per-process file descriptors for use by procstat(1), et al.
3805  */
3806 static int
3807 sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS)
3808 {
3809 	struct kinfo_ofile *okif;
3810 	struct kinfo_file *kif;
3811 	struct filedesc *fdp;
3812 	int error, i, *name;
3813 	struct file *fp;
3814 	struct proc *p;
3815 
3816 	name = (int *)arg1;
3817 	error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
3818 	if (error != 0)
3819 		return (error);
3820 	fdp = fdhold(p);
3821 	PROC_UNLOCK(p);
3822 	if (fdp == NULL)
3823 		return (ENOENT);
3824 	kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK);
3825 	okif = malloc(sizeof(*okif), M_TEMP, M_WAITOK);
3826 	FILEDESC_SLOCK(fdp);
3827 	if (fdp->fd_cdir != NULL)
3828 		export_vnode_for_osysctl(fdp->fd_cdir, KF_FD_TYPE_CWD, kif,
3829 		    okif, fdp, req);
3830 	if (fdp->fd_rdir != NULL)
3831 		export_vnode_for_osysctl(fdp->fd_rdir, KF_FD_TYPE_ROOT, kif,
3832 		    okif, fdp, req);
3833 	if (fdp->fd_jdir != NULL)
3834 		export_vnode_for_osysctl(fdp->fd_jdir, KF_FD_TYPE_JAIL, kif,
3835 		    okif, fdp, req);
3836 	for (i = 0; fdp->fd_refcnt > 0 && i <= fdp->fd_lastfile; i++) {
3837 		if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
3838 			continue;
3839 		export_file_to_kinfo(fp, i, NULL, kif, fdp,
3840 		    KERN_FILEDESC_PACK_KINFO);
3841 		FILEDESC_SUNLOCK(fdp);
3842 		kinfo_to_okinfo(kif, okif);
3843 		error = SYSCTL_OUT(req, okif, sizeof(*okif));
3844 		FILEDESC_SLOCK(fdp);
3845 		if (error)
3846 			break;
3847 	}
3848 	FILEDESC_SUNLOCK(fdp);
3849 	fddrop(fdp);
3850 	free(kif, M_TEMP);
3851 	free(okif, M_TEMP);
3852 	return (0);
3853 }
3854 
3855 static SYSCTL_NODE(_kern_proc, KERN_PROC_OFILEDESC, ofiledesc,
3856     CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_ofiledesc,
3857     "Process ofiledesc entries");
3858 #endif	/* COMPAT_FREEBSD7 */
3859 
3860 int
3861 vntype_to_kinfo(int vtype)
3862 {
3863 	struct {
3864 		int	vtype;
3865 		int	kf_vtype;
3866 	} vtypes_table[] = {
3867 		{ VBAD, KF_VTYPE_VBAD },
3868 		{ VBLK, KF_VTYPE_VBLK },
3869 		{ VCHR, KF_VTYPE_VCHR },
3870 		{ VDIR, KF_VTYPE_VDIR },
3871 		{ VFIFO, KF_VTYPE_VFIFO },
3872 		{ VLNK, KF_VTYPE_VLNK },
3873 		{ VNON, KF_VTYPE_VNON },
3874 		{ VREG, KF_VTYPE_VREG },
3875 		{ VSOCK, KF_VTYPE_VSOCK }
3876 	};
3877 	unsigned int i;
3878 
3879 	/*
3880 	 * Perform vtype translation.
3881 	 */
3882 	for (i = 0; i < nitems(vtypes_table); i++)
3883 		if (vtypes_table[i].vtype == vtype)
3884 			return (vtypes_table[i].kf_vtype);
3885 
3886 	return (KF_VTYPE_UNKNOWN);
3887 }
3888 
3889 static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc,
3890     CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_filedesc,
3891     "Process filedesc entries");
3892 
3893 /*
3894  * Store a process current working directory information to sbuf.
3895  *
3896  * Takes a locked proc as argument, and returns with the proc unlocked.
3897  */
3898 int
3899 kern_proc_cwd_out(struct proc *p,  struct sbuf *sb, ssize_t maxlen)
3900 {
3901 	struct filedesc *fdp;
3902 	struct export_fd_buf *efbuf;
3903 	int error;
3904 
3905 	PROC_LOCK_ASSERT(p, MA_OWNED);
3906 
3907 	fdp = fdhold(p);
3908 	PROC_UNLOCK(p);
3909 	if (fdp == NULL)
3910 		return (EINVAL);
3911 
3912 	efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
3913 	efbuf->fdp = fdp;
3914 	efbuf->sb = sb;
3915 	efbuf->remainder = maxlen;
3916 
3917 	FILEDESC_SLOCK(fdp);
3918 	if (fdp->fd_cdir == NULL)
3919 		error = EINVAL;
3920 	else {
3921 		vrefact(fdp->fd_cdir);
3922 		error = export_vnode_to_sb(fdp->fd_cdir, KF_FD_TYPE_CWD,
3923 		    FREAD, efbuf);
3924 	}
3925 	FILEDESC_SUNLOCK(fdp);
3926 	fddrop(fdp);
3927 	free(efbuf, M_TEMP);
3928 	return (error);
3929 }
3930 
3931 /*
3932  * Get per-process current working directory.
3933  */
3934 static int
3935 sysctl_kern_proc_cwd(SYSCTL_HANDLER_ARGS)
3936 {
3937 	struct sbuf sb;
3938 	struct proc *p;
3939 	ssize_t maxlen;
3940 	int error, error2, *name;
3941 
3942 	name = (int *)arg1;
3943 
3944 	sbuf_new_for_sysctl(&sb, NULL, sizeof(struct kinfo_file), req);
3945 	sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
3946 	error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
3947 	if (error != 0) {
3948 		sbuf_delete(&sb);
3949 		return (error);
3950 	}
3951 	maxlen = req->oldptr != NULL ? req->oldlen : -1;
3952 	error = kern_proc_cwd_out(p, &sb, maxlen);
3953 	error2 = sbuf_finish(&sb);
3954 	sbuf_delete(&sb);
3955 	return (error != 0 ? error : error2);
3956 }
3957 
3958 static SYSCTL_NODE(_kern_proc, KERN_PROC_CWD, cwd, CTLFLAG_RD|CTLFLAG_MPSAFE,
3959     sysctl_kern_proc_cwd, "Process current working directory");
3960 
3961 #ifdef DDB
3962 /*
3963  * For the purposes of debugging, generate a human-readable string for the
3964  * file type.
3965  */
3966 static const char *
3967 file_type_to_name(short type)
3968 {
3969 
3970 	switch (type) {
3971 	case 0:
3972 		return ("zero");
3973 	case DTYPE_VNODE:
3974 		return ("vnode");
3975 	case DTYPE_SOCKET:
3976 		return ("socket");
3977 	case DTYPE_PIPE:
3978 		return ("pipe");
3979 	case DTYPE_FIFO:
3980 		return ("fifo");
3981 	case DTYPE_KQUEUE:
3982 		return ("kqueue");
3983 	case DTYPE_CRYPTO:
3984 		return ("crypto");
3985 	case DTYPE_MQUEUE:
3986 		return ("mqueue");
3987 	case DTYPE_SHM:
3988 		return ("shm");
3989 	case DTYPE_SEM:
3990 		return ("ksem");
3991 	case DTYPE_PTS:
3992 		return ("pts");
3993 	case DTYPE_DEV:
3994 		return ("dev");
3995 	case DTYPE_PROCDESC:
3996 		return ("proc");
3997 	case DTYPE_LINUXEFD:
3998 		return ("levent");
3999 	case DTYPE_LINUXTFD:
4000 		return ("ltimer");
4001 	default:
4002 		return ("unkn");
4003 	}
4004 }
4005 
4006 /*
4007  * For the purposes of debugging, identify a process (if any, perhaps one of
4008  * many) that references the passed file in its file descriptor array. Return
4009  * NULL if none.
4010  */
4011 static struct proc *
4012 file_to_first_proc(struct file *fp)
4013 {
4014 	struct filedesc *fdp;
4015 	struct proc *p;
4016 	int n;
4017 
4018 	FOREACH_PROC_IN_SYSTEM(p) {
4019 		if (p->p_state == PRS_NEW)
4020 			continue;
4021 		fdp = p->p_fd;
4022 		if (fdp == NULL)
4023 			continue;
4024 		for (n = 0; n <= fdp->fd_lastfile; n++) {
4025 			if (fp == fdp->fd_ofiles[n].fde_file)
4026 				return (p);
4027 		}
4028 	}
4029 	return (NULL);
4030 }
4031 
4032 static void
4033 db_print_file(struct file *fp, int header)
4034 {
4035 #define XPTRWIDTH ((int)howmany(sizeof(void *) * NBBY, 4))
4036 	struct proc *p;
4037 
4038 	if (header)
4039 		db_printf("%*s %6s %*s %8s %4s %5s %6s %*s %5s %s\n",
4040 		    XPTRWIDTH, "File", "Type", XPTRWIDTH, "Data", "Flag",
4041 		    "GCFl", "Count", "MCount", XPTRWIDTH, "Vnode", "FPID",
4042 		    "FCmd");
4043 	p = file_to_first_proc(fp);
4044 	db_printf("%*p %6s %*p %08x %04x %5d %6d %*p %5d %s\n", XPTRWIDTH,
4045 	    fp, file_type_to_name(fp->f_type), XPTRWIDTH, fp->f_data,
4046 	    fp->f_flag, 0, fp->f_count, 0, XPTRWIDTH, fp->f_vnode,
4047 	    p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-");
4048 
4049 #undef XPTRWIDTH
4050 }
4051 
4052 DB_SHOW_COMMAND(file, db_show_file)
4053 {
4054 	struct file *fp;
4055 
4056 	if (!have_addr) {
4057 		db_printf("usage: show file <addr>\n");
4058 		return;
4059 	}
4060 	fp = (struct file *)addr;
4061 	db_print_file(fp, 1);
4062 }
4063 
4064 DB_SHOW_COMMAND(files, db_show_files)
4065 {
4066 	struct filedesc *fdp;
4067 	struct file *fp;
4068 	struct proc *p;
4069 	int header;
4070 	int n;
4071 
4072 	header = 1;
4073 	FOREACH_PROC_IN_SYSTEM(p) {
4074 		if (p->p_state == PRS_NEW)
4075 			continue;
4076 		if ((fdp = p->p_fd) == NULL)
4077 			continue;
4078 		for (n = 0; n <= fdp->fd_lastfile; ++n) {
4079 			if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
4080 				continue;
4081 			db_print_file(fp, header);
4082 			header = 0;
4083 		}
4084 	}
4085 }
4086 #endif
4087 
4088 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
4089     &maxfilesperproc, 0, "Maximum files allowed open per process");
4090 
4091 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
4092     &maxfiles, 0, "Maximum number of files");
4093 
4094 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
4095     &openfiles, 0, "System-wide number of open files");
4096 
4097 /* ARGSUSED*/
4098 static void
4099 filelistinit(void *dummy)
4100 {
4101 
4102 	file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL,
4103 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
4104 	filedesc0_zone = uma_zcreate("filedesc0", sizeof(struct filedesc0),
4105 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
4106 	mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF);
4107 }
4108 SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL);
4109 
4110 /*-------------------------------------------------------------------*/
4111 
4112 static int
4113 badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred,
4114     int flags, struct thread *td)
4115 {
4116 
4117 	return (EBADF);
4118 }
4119 
4120 static int
4121 badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
4122     struct thread *td)
4123 {
4124 
4125 	return (EINVAL);
4126 }
4127 
4128 static int
4129 badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred,
4130     struct thread *td)
4131 {
4132 
4133 	return (EBADF);
4134 }
4135 
4136 static int
4137 badfo_poll(struct file *fp, int events, struct ucred *active_cred,
4138     struct thread *td)
4139 {
4140 
4141 	return (0);
4142 }
4143 
4144 static int
4145 badfo_kqfilter(struct file *fp, struct knote *kn)
4146 {
4147 
4148 	return (EBADF);
4149 }
4150 
4151 static int
4152 badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred,
4153     struct thread *td)
4154 {
4155 
4156 	return (EBADF);
4157 }
4158 
4159 static int
4160 badfo_close(struct file *fp, struct thread *td)
4161 {
4162 
4163 	return (0);
4164 }
4165 
4166 static int
4167 badfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
4168     struct thread *td)
4169 {
4170 
4171 	return (EBADF);
4172 }
4173 
4174 static int
4175 badfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
4176     struct thread *td)
4177 {
4178 
4179 	return (EBADF);
4180 }
4181 
4182 static int
4183 badfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
4184     struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
4185     struct thread *td)
4186 {
4187 
4188 	return (EBADF);
4189 }
4190 
4191 static int
4192 badfo_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
4193 {
4194 
4195 	return (0);
4196 }
4197 
4198 struct fileops badfileops = {
4199 	.fo_read = badfo_readwrite,
4200 	.fo_write = badfo_readwrite,
4201 	.fo_truncate = badfo_truncate,
4202 	.fo_ioctl = badfo_ioctl,
4203 	.fo_poll = badfo_poll,
4204 	.fo_kqfilter = badfo_kqfilter,
4205 	.fo_stat = badfo_stat,
4206 	.fo_close = badfo_close,
4207 	.fo_chmod = badfo_chmod,
4208 	.fo_chown = badfo_chown,
4209 	.fo_sendfile = badfo_sendfile,
4210 	.fo_fill_kinfo = badfo_fill_kinfo,
4211 };
4212 
4213 int
4214 invfo_rdwr(struct file *fp, struct uio *uio, struct ucred *active_cred,
4215     int flags, struct thread *td)
4216 {
4217 
4218 	return (EOPNOTSUPP);
4219 }
4220 
4221 int
4222 invfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
4223     struct thread *td)
4224 {
4225 
4226 	return (EINVAL);
4227 }
4228 
4229 int
4230 invfo_ioctl(struct file *fp, u_long com, void *data,
4231     struct ucred *active_cred, struct thread *td)
4232 {
4233 
4234 	return (ENOTTY);
4235 }
4236 
4237 int
4238 invfo_poll(struct file *fp, int events, struct ucred *active_cred,
4239     struct thread *td)
4240 {
4241 
4242 	return (poll_no_poll(events));
4243 }
4244 
4245 int
4246 invfo_kqfilter(struct file *fp, struct knote *kn)
4247 {
4248 
4249 	return (EINVAL);
4250 }
4251 
4252 int
4253 invfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
4254     struct thread *td)
4255 {
4256 
4257 	return (EINVAL);
4258 }
4259 
4260 int
4261 invfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
4262     struct thread *td)
4263 {
4264 
4265 	return (EINVAL);
4266 }
4267 
4268 int
4269 invfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
4270     struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
4271     struct thread *td)
4272 {
4273 
4274 	return (EINVAL);
4275 }
4276 
4277 /*-------------------------------------------------------------------*/
4278 
4279 /*
4280  * File Descriptor pseudo-device driver (/dev/fd/).
4281  *
4282  * Opening minor device N dup()s the file (if any) connected to file
4283  * descriptor N belonging to the calling process.  Note that this driver
4284  * consists of only the ``open()'' routine, because all subsequent
4285  * references to this file will be direct to the other driver.
4286  *
4287  * XXX: we could give this one a cloning event handler if necessary.
4288  */
4289 
4290 /* ARGSUSED */
4291 static int
4292 fdopen(struct cdev *dev, int mode, int type, struct thread *td)
4293 {
4294 
4295 	/*
4296 	 * XXX Kludge: set curthread->td_dupfd to contain the value of the
4297 	 * the file descriptor being sought for duplication. The error
4298 	 * return ensures that the vnode for this device will be released
4299 	 * by vn_open. Open will detect this special error and take the
4300 	 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
4301 	 * will simply report the error.
4302 	 */
4303 	td->td_dupfd = dev2unit(dev);
4304 	return (ENODEV);
4305 }
4306 
4307 static struct cdevsw fildesc_cdevsw = {
4308 	.d_version =	D_VERSION,
4309 	.d_open =	fdopen,
4310 	.d_name =	"FD",
4311 };
4312 
4313 static void
4314 fildesc_drvinit(void *unused)
4315 {
4316 	struct cdev *dev;
4317 
4318 	dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 0, NULL,
4319 	    UID_ROOT, GID_WHEEL, 0666, "fd/0");
4320 	make_dev_alias(dev, "stdin");
4321 	dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 1, NULL,
4322 	    UID_ROOT, GID_WHEEL, 0666, "fd/1");
4323 	make_dev_alias(dev, "stdout");
4324 	dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 2, NULL,
4325 	    UID_ROOT, GID_WHEEL, 0666, "fd/2");
4326 	make_dev_alias(dev, "stderr");
4327 }
4328 
4329 SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL);
4330