xref: /freebsd/sys/kern/kern_descrip.c (revision f4f33ea0c752ff0f9bfad34991d5bbb54e71133d)
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 volatile 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 = fd_first_free(fdp, fd, fdp->fd_nfiles);
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 	seq_write_begin(&fde->fde_seq);
308 #endif
309 	fde->fde_file = NULL;
310 #ifdef CAPABILITIES
311 	seq_write_end(&fde->fde_seq);
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] =
352 	    min((int)lim_cur(td, RLIMIT_NOFILE), maxfilesperproc);
353 #ifdef	RACCT
354 	PROC_LOCK(td->td_proc);
355 	lim = racct_get_limit(td->td_proc, RACCT_NOFILE);
356 	PROC_UNLOCK(td->td_proc);
357 	if (lim < td->td_retval[0])
358 		td->td_retval[0] = lim;
359 #endif
360 	return (0);
361 }
362 
363 /*
364  * Duplicate a file descriptor to a particular value.
365  *
366  * Note: keep in mind that a potential race condition exists when closing
367  * descriptors from a shared descriptor table (via rfork).
368  */
369 #ifndef _SYS_SYSPROTO_H_
370 struct dup2_args {
371 	u_int	from;
372 	u_int	to;
373 };
374 #endif
375 /* ARGSUSED */
376 int
377 sys_dup2(struct thread *td, struct dup2_args *uap)
378 {
379 
380 	return (kern_dup(td, FDDUP_FIXED, 0, (int)uap->from, (int)uap->to));
381 }
382 
383 /*
384  * Duplicate a file descriptor.
385  */
386 #ifndef _SYS_SYSPROTO_H_
387 struct dup_args {
388 	u_int	fd;
389 };
390 #endif
391 /* ARGSUSED */
392 int
393 sys_dup(struct thread *td, struct dup_args *uap)
394 {
395 
396 	return (kern_dup(td, FDDUP_NORMAL, 0, (int)uap->fd, 0));
397 }
398 
399 /*
400  * The file control system call.
401  */
402 #ifndef _SYS_SYSPROTO_H_
403 struct fcntl_args {
404 	int	fd;
405 	int	cmd;
406 	long	arg;
407 };
408 #endif
409 /* ARGSUSED */
410 int
411 sys_fcntl(struct thread *td, struct fcntl_args *uap)
412 {
413 
414 	return (kern_fcntl_freebsd(td, uap->fd, uap->cmd, uap->arg));
415 }
416 
417 int
418 kern_fcntl_freebsd(struct thread *td, int fd, int cmd, long arg)
419 {
420 	struct flock fl;
421 	struct __oflock ofl;
422 	intptr_t arg1;
423 	int error, newcmd;
424 
425 	error = 0;
426 	newcmd = cmd;
427 	switch (cmd) {
428 	case F_OGETLK:
429 	case F_OSETLK:
430 	case F_OSETLKW:
431 		/*
432 		 * Convert old flock structure to new.
433 		 */
434 		error = copyin((void *)(intptr_t)arg, &ofl, sizeof(ofl));
435 		fl.l_start = ofl.l_start;
436 		fl.l_len = ofl.l_len;
437 		fl.l_pid = ofl.l_pid;
438 		fl.l_type = ofl.l_type;
439 		fl.l_whence = ofl.l_whence;
440 		fl.l_sysid = 0;
441 
442 		switch (cmd) {
443 		case F_OGETLK:
444 			newcmd = F_GETLK;
445 			break;
446 		case F_OSETLK:
447 			newcmd = F_SETLK;
448 			break;
449 		case F_OSETLKW:
450 			newcmd = F_SETLKW;
451 			break;
452 		}
453 		arg1 = (intptr_t)&fl;
454 		break;
455 	case F_GETLK:
456 	case F_SETLK:
457 	case F_SETLKW:
458 	case F_SETLK_REMOTE:
459 		error = copyin((void *)(intptr_t)arg, &fl, sizeof(fl));
460 		arg1 = (intptr_t)&fl;
461 		break;
462 	default:
463 		arg1 = arg;
464 		break;
465 	}
466 	if (error)
467 		return (error);
468 	error = kern_fcntl(td, fd, newcmd, arg1);
469 	if (error)
470 		return (error);
471 	if (cmd == F_OGETLK) {
472 		ofl.l_start = fl.l_start;
473 		ofl.l_len = fl.l_len;
474 		ofl.l_pid = fl.l_pid;
475 		ofl.l_type = fl.l_type;
476 		ofl.l_whence = fl.l_whence;
477 		error = copyout(&ofl, (void *)(intptr_t)arg, sizeof(ofl));
478 	} else if (cmd == F_GETLK) {
479 		error = copyout(&fl, (void *)(intptr_t)arg, sizeof(fl));
480 	}
481 	return (error);
482 }
483 
484 int
485 kern_fcntl(struct thread *td, int fd, int cmd, intptr_t arg)
486 {
487 	struct filedesc *fdp;
488 	struct flock *flp;
489 	struct file *fp, *fp2;
490 	struct filedescent *fde;
491 	struct proc *p;
492 	struct vnode *vp;
493 	int error, flg, 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)
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 		error = fget_unlocked(fdp, fd, &cap_flock_rights, &fp, NULL);
617 		if (error != 0)
618 			break;
619 		if (fp->f_type != DTYPE_VNODE) {
620 			error = EBADF;
621 			fdrop(fp, td);
622 			break;
623 		}
624 
625 		flp = (struct flock *)arg;
626 		if (flp->l_whence == SEEK_CUR) {
627 			foffset = foffset_get(fp);
628 			if (foffset < 0 ||
629 			    (flp->l_start > 0 &&
630 			     foffset > OFF_MAX - flp->l_start)) {
631 				error = EOVERFLOW;
632 				fdrop(fp, td);
633 				break;
634 			}
635 			flp->l_start += foffset;
636 		}
637 
638 		vp = fp->f_vnode;
639 		switch (flp->l_type) {
640 		case F_RDLCK:
641 			if ((fp->f_flag & FREAD) == 0) {
642 				error = EBADF;
643 				break;
644 			}
645 			if ((p->p_leader->p_flag & P_ADVLOCK) == 0) {
646 				PROC_LOCK(p->p_leader);
647 				p->p_leader->p_flag |= P_ADVLOCK;
648 				PROC_UNLOCK(p->p_leader);
649 			}
650 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
651 			    flp, flg);
652 			break;
653 		case F_WRLCK:
654 			if ((fp->f_flag & FWRITE) == 0) {
655 				error = EBADF;
656 				break;
657 			}
658 			if ((p->p_leader->p_flag & P_ADVLOCK) == 0) {
659 				PROC_LOCK(p->p_leader);
660 				p->p_leader->p_flag |= P_ADVLOCK;
661 				PROC_UNLOCK(p->p_leader);
662 			}
663 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
664 			    flp, flg);
665 			break;
666 		case F_UNLCK:
667 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
668 			    flp, flg);
669 			break;
670 		case F_UNLCKSYS:
671 			/*
672 			 * Temporary api for testing remote lock
673 			 * infrastructure.
674 			 */
675 			if (flg != F_REMOTE) {
676 				error = EINVAL;
677 				break;
678 			}
679 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
680 			    F_UNLCKSYS, flp, flg);
681 			break;
682 		default:
683 			error = EINVAL;
684 			break;
685 		}
686 		if (error != 0 || flp->l_type == F_UNLCK ||
687 		    flp->l_type == F_UNLCKSYS) {
688 			fdrop(fp, td);
689 			break;
690 		}
691 
692 		/*
693 		 * Check for a race with close.
694 		 *
695 		 * The vnode is now advisory locked (or unlocked, but this case
696 		 * is not really important) as the caller requested.
697 		 * We had to drop the filedesc lock, so we need to recheck if
698 		 * the descriptor is still valid, because if it was closed
699 		 * in the meantime we need to remove advisory lock from the
700 		 * vnode - close on any descriptor leading to an advisory
701 		 * locked vnode, removes that lock.
702 		 * We will return 0 on purpose in that case, as the result of
703 		 * successful advisory lock might have been externally visible
704 		 * already. This is fine - effectively we pretend to the caller
705 		 * that the closing thread was a bit slower and that the
706 		 * advisory lock succeeded before the close.
707 		 */
708 		error = fget_unlocked(fdp, fd, &cap_no_rights, &fp2, NULL);
709 		if (error != 0) {
710 			fdrop(fp, td);
711 			break;
712 		}
713 		if (fp != fp2) {
714 			flp->l_whence = SEEK_SET;
715 			flp->l_start = 0;
716 			flp->l_len = 0;
717 			flp->l_type = F_UNLCK;
718 			(void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
719 			    F_UNLCK, flp, F_POSIX);
720 		}
721 		fdrop(fp, td);
722 		fdrop(fp2, td);
723 		break;
724 
725 	case F_GETLK:
726 		error = fget_unlocked(fdp, fd, &cap_flock_rights, &fp, NULL);
727 		if (error != 0)
728 			break;
729 		if (fp->f_type != DTYPE_VNODE) {
730 			error = EBADF;
731 			fdrop(fp, td);
732 			break;
733 		}
734 		flp = (struct flock *)arg;
735 		if (flp->l_type != F_RDLCK && flp->l_type != F_WRLCK &&
736 		    flp->l_type != F_UNLCK) {
737 			error = EINVAL;
738 			fdrop(fp, td);
739 			break;
740 		}
741 		if (flp->l_whence == SEEK_CUR) {
742 			foffset = foffset_get(fp);
743 			if ((flp->l_start > 0 &&
744 			    foffset > OFF_MAX - flp->l_start) ||
745 			    (flp->l_start < 0 &&
746 			    foffset < OFF_MIN - flp->l_start)) {
747 				error = EOVERFLOW;
748 				fdrop(fp, td);
749 				break;
750 			}
751 			flp->l_start += foffset;
752 		}
753 		vp = fp->f_vnode;
754 		error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, flp,
755 		    F_POSIX);
756 		fdrop(fp, td);
757 		break;
758 
759 	case F_RDAHEAD:
760 		arg = arg ? 128 * 1024: 0;
761 		/* FALLTHROUGH */
762 	case F_READAHEAD:
763 		error = fget_unlocked(fdp, fd, &cap_no_rights, &fp, NULL);
764 		if (error != 0)
765 			break;
766 		if (fp->f_type != DTYPE_VNODE) {
767 			fdrop(fp, td);
768 			error = EBADF;
769 			break;
770 		}
771 		vp = fp->f_vnode;
772 		/*
773 		 * Exclusive lock synchronizes against f_seqcount reads and
774 		 * writes in sequential_heuristic().
775 		 */
776 		error = vn_lock(vp, LK_EXCLUSIVE);
777 		if (error != 0) {
778 			fdrop(fp, td);
779 			break;
780 		}
781 		if (arg >= 0) {
782 			bsize = fp->f_vnode->v_mount->mnt_stat.f_iosize;
783 			fp->f_seqcount = (arg + bsize - 1) / bsize;
784 			atomic_set_int(&fp->f_flag, FRDAHEAD);
785 		} else {
786 			atomic_clear_int(&fp->f_flag, FRDAHEAD);
787 		}
788 		VOP_UNLOCK(vp, 0);
789 		fdrop(fp, td);
790 		break;
791 
792 	default:
793 		error = EINVAL;
794 		break;
795 	}
796 	return (error);
797 }
798 
799 static int
800 getmaxfd(struct thread *td)
801 {
802 
803 	return (min((int)lim_cur(td, RLIMIT_NOFILE), maxfilesperproc));
804 }
805 
806 /*
807  * Common code for dup, dup2, fcntl(F_DUPFD) and fcntl(F_DUP2FD).
808  */
809 int
810 kern_dup(struct thread *td, u_int mode, int flags, int old, int new)
811 {
812 	struct filedesc *fdp;
813 	struct filedescent *oldfde, *newfde;
814 	struct proc *p;
815 	struct file *delfp;
816 	u_long *oioctls, *nioctls;
817 	int error, maxfd;
818 
819 	p = td->td_proc;
820 	fdp = p->p_fd;
821 
822 	MPASS((flags & ~(FDDUP_FLAG_CLOEXEC)) == 0);
823 	MPASS(mode < FDDUP_LASTMODE);
824 
825 	AUDIT_ARG_FD(old);
826 	/* XXXRW: if (flags & FDDUP_FIXED) AUDIT_ARG_FD2(new); */
827 
828 	/*
829 	 * Verify we have a valid descriptor to dup from and possibly to
830 	 * dup to. Unlike dup() and dup2(), fcntl()'s F_DUPFD should
831 	 * return EINVAL when the new descriptor is out of bounds.
832 	 */
833 	if (old < 0)
834 		return (EBADF);
835 	if (new < 0)
836 		return (mode == FDDUP_FCNTL ? EINVAL : EBADF);
837 	maxfd = getmaxfd(td);
838 	if (new >= maxfd)
839 		return (mode == FDDUP_FCNTL ? EINVAL : EBADF);
840 
841 	error = EBADF;
842 	FILEDESC_XLOCK(fdp);
843 	if (fget_locked(fdp, old) == NULL)
844 		goto unlock;
845 	if ((mode == FDDUP_FIXED || mode == FDDUP_MUSTREPLACE) && old == new) {
846 		td->td_retval[0] = new;
847 		if (flags & FDDUP_FLAG_CLOEXEC)
848 			fdp->fd_ofiles[new].fde_flags |= UF_EXCLOSE;
849 		error = 0;
850 		goto unlock;
851 	}
852 
853 	/*
854 	 * If the caller specified a file descriptor, make sure the file
855 	 * table is large enough to hold it, and grab it.  Otherwise, just
856 	 * allocate a new descriptor the usual way.
857 	 */
858 	switch (mode) {
859 	case FDDUP_NORMAL:
860 	case FDDUP_FCNTL:
861 		if ((error = fdalloc(td, new, &new)) != 0)
862 			goto unlock;
863 		break;
864 	case FDDUP_MUSTREPLACE:
865 		/* Target file descriptor must exist. */
866 		if (fget_locked(fdp, new) == NULL)
867 			goto unlock;
868 		break;
869 	case FDDUP_FIXED:
870 		if (new >= fdp->fd_nfiles) {
871 			/*
872 			 * The resource limits are here instead of e.g.
873 			 * fdalloc(), because the file descriptor table may be
874 			 * shared between processes, so we can't really use
875 			 * racct_add()/racct_sub().  Instead of counting the
876 			 * number of actually allocated descriptors, just put
877 			 * the limit on the size of the file descriptor table.
878 			 */
879 #ifdef RACCT
880 			if (racct_enable) {
881 				PROC_LOCK(p);
882 				error = racct_set(p, RACCT_NOFILE, new + 1);
883 				PROC_UNLOCK(p);
884 				if (error != 0) {
885 					error = EMFILE;
886 					goto unlock;
887 				}
888 			}
889 #endif
890 			fdgrowtable_exp(fdp, new + 1);
891 		}
892 		if (!fdisused(fdp, new))
893 			fdused(fdp, new);
894 		break;
895 	default:
896 		KASSERT(0, ("%s unsupported mode %d", __func__, mode));
897 	}
898 
899 	KASSERT(old != new, ("new fd is same as old"));
900 
901 	oldfde = &fdp->fd_ofiles[old];
902 	fhold(oldfde->fde_file);
903 	newfde = &fdp->fd_ofiles[new];
904 	delfp = newfde->fde_file;
905 
906 	oioctls = filecaps_free_prep(&newfde->fde_caps);
907 	nioctls = filecaps_copy_prep(&oldfde->fde_caps);
908 
909 	/*
910 	 * Duplicate the source descriptor.
911 	 */
912 #ifdef CAPABILITIES
913 	seq_write_begin(&newfde->fde_seq);
914 #endif
915 	memcpy(newfde, oldfde, fde_change_size);
916 	filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps,
917 	    nioctls);
918 	if ((flags & FDDUP_FLAG_CLOEXEC) != 0)
919 		newfde->fde_flags = oldfde->fde_flags | UF_EXCLOSE;
920 	else
921 		newfde->fde_flags = oldfde->fde_flags & ~UF_EXCLOSE;
922 #ifdef CAPABILITIES
923 	seq_write_end(&newfde->fde_seq);
924 #endif
925 	filecaps_free_finish(oioctls);
926 	td->td_retval[0] = new;
927 
928 	error = 0;
929 
930 	if (delfp != NULL) {
931 		(void) closefp(fdp, new, delfp, td, 1);
932 		FILEDESC_UNLOCK_ASSERT(fdp);
933 	} else {
934 unlock:
935 		FILEDESC_XUNLOCK(fdp);
936 	}
937 
938 	return (error);
939 }
940 
941 /*
942  * If sigio is on the list associated with a process or process group,
943  * disable signalling from the device, remove sigio from the list and
944  * free sigio.
945  */
946 void
947 funsetown(struct sigio **sigiop)
948 {
949 	struct sigio *sigio;
950 
951 	if (*sigiop == NULL)
952 		return;
953 	SIGIO_LOCK();
954 	sigio = *sigiop;
955 	if (sigio == NULL) {
956 		SIGIO_UNLOCK();
957 		return;
958 	}
959 	*(sigio->sio_myref) = NULL;
960 	if ((sigio)->sio_pgid < 0) {
961 		struct pgrp *pg = (sigio)->sio_pgrp;
962 		PGRP_LOCK(pg);
963 		SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio,
964 			    sigio, sio_pgsigio);
965 		PGRP_UNLOCK(pg);
966 	} else {
967 		struct proc *p = (sigio)->sio_proc;
968 		PROC_LOCK(p);
969 		SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio,
970 			    sigio, sio_pgsigio);
971 		PROC_UNLOCK(p);
972 	}
973 	SIGIO_UNLOCK();
974 	crfree(sigio->sio_ucred);
975 	free(sigio, M_SIGIO);
976 }
977 
978 /*
979  * Free a list of sigio structures.
980  * We only need to lock the SIGIO_LOCK because we have made ourselves
981  * inaccessible to callers of fsetown and therefore do not need to lock
982  * the proc or pgrp struct for the list manipulation.
983  */
984 void
985 funsetownlst(struct sigiolst *sigiolst)
986 {
987 	struct proc *p;
988 	struct pgrp *pg;
989 	struct sigio *sigio;
990 
991 	sigio = SLIST_FIRST(sigiolst);
992 	if (sigio == NULL)
993 		return;
994 	p = NULL;
995 	pg = NULL;
996 
997 	/*
998 	 * Every entry of the list should belong
999 	 * to a single proc or pgrp.
1000 	 */
1001 	if (sigio->sio_pgid < 0) {
1002 		pg = sigio->sio_pgrp;
1003 		PGRP_LOCK_ASSERT(pg, MA_NOTOWNED);
1004 	} else /* if (sigio->sio_pgid > 0) */ {
1005 		p = sigio->sio_proc;
1006 		PROC_LOCK_ASSERT(p, MA_NOTOWNED);
1007 	}
1008 
1009 	SIGIO_LOCK();
1010 	while ((sigio = SLIST_FIRST(sigiolst)) != NULL) {
1011 		*(sigio->sio_myref) = NULL;
1012 		if (pg != NULL) {
1013 			KASSERT(sigio->sio_pgid < 0,
1014 			    ("Proc sigio in pgrp sigio list"));
1015 			KASSERT(sigio->sio_pgrp == pg,
1016 			    ("Bogus pgrp in sigio list"));
1017 			PGRP_LOCK(pg);
1018 			SLIST_REMOVE(&pg->pg_sigiolst, sigio, sigio,
1019 			    sio_pgsigio);
1020 			PGRP_UNLOCK(pg);
1021 		} else /* if (p != NULL) */ {
1022 			KASSERT(sigio->sio_pgid > 0,
1023 			    ("Pgrp sigio in proc sigio list"));
1024 			KASSERT(sigio->sio_proc == p,
1025 			    ("Bogus proc in sigio list"));
1026 			PROC_LOCK(p);
1027 			SLIST_REMOVE(&p->p_sigiolst, sigio, sigio,
1028 			    sio_pgsigio);
1029 			PROC_UNLOCK(p);
1030 		}
1031 		SIGIO_UNLOCK();
1032 		crfree(sigio->sio_ucred);
1033 		free(sigio, M_SIGIO);
1034 		SIGIO_LOCK();
1035 	}
1036 	SIGIO_UNLOCK();
1037 }
1038 
1039 /*
1040  * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg).
1041  *
1042  * After permission checking, add a sigio structure to the sigio list for
1043  * the process or process group.
1044  */
1045 int
1046 fsetown(pid_t pgid, struct sigio **sigiop)
1047 {
1048 	struct proc *proc;
1049 	struct pgrp *pgrp;
1050 	struct sigio *sigio;
1051 	int ret;
1052 
1053 	if (pgid == 0) {
1054 		funsetown(sigiop);
1055 		return (0);
1056 	}
1057 
1058 	ret = 0;
1059 
1060 	/* Allocate and fill in the new sigio out of locks. */
1061 	sigio = malloc(sizeof(struct sigio), M_SIGIO, M_WAITOK);
1062 	sigio->sio_pgid = pgid;
1063 	sigio->sio_ucred = crhold(curthread->td_ucred);
1064 	sigio->sio_myref = sigiop;
1065 
1066 	sx_slock(&proctree_lock);
1067 	if (pgid > 0) {
1068 		proc = pfind(pgid);
1069 		if (proc == NULL) {
1070 			ret = ESRCH;
1071 			goto fail;
1072 		}
1073 
1074 		/*
1075 		 * Policy - Don't allow a process to FSETOWN a process
1076 		 * in another session.
1077 		 *
1078 		 * Remove this test to allow maximum flexibility or
1079 		 * restrict FSETOWN to the current process or process
1080 		 * group for maximum safety.
1081 		 */
1082 		PROC_UNLOCK(proc);
1083 		if (proc->p_session != curthread->td_proc->p_session) {
1084 			ret = EPERM;
1085 			goto fail;
1086 		}
1087 
1088 		pgrp = NULL;
1089 	} else /* if (pgid < 0) */ {
1090 		pgrp = pgfind(-pgid);
1091 		if (pgrp == NULL) {
1092 			ret = ESRCH;
1093 			goto fail;
1094 		}
1095 		PGRP_UNLOCK(pgrp);
1096 
1097 		/*
1098 		 * Policy - Don't allow a process to FSETOWN a process
1099 		 * in another session.
1100 		 *
1101 		 * Remove this test to allow maximum flexibility or
1102 		 * restrict FSETOWN to the current process or process
1103 		 * group for maximum safety.
1104 		 */
1105 		if (pgrp->pg_session != curthread->td_proc->p_session) {
1106 			ret = EPERM;
1107 			goto fail;
1108 		}
1109 
1110 		proc = NULL;
1111 	}
1112 	funsetown(sigiop);
1113 	if (pgid > 0) {
1114 		PROC_LOCK(proc);
1115 		/*
1116 		 * Since funsetownlst() is called without the proctree
1117 		 * locked, we need to check for P_WEXIT.
1118 		 * XXX: is ESRCH correct?
1119 		 */
1120 		if ((proc->p_flag & P_WEXIT) != 0) {
1121 			PROC_UNLOCK(proc);
1122 			ret = ESRCH;
1123 			goto fail;
1124 		}
1125 		SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio);
1126 		sigio->sio_proc = proc;
1127 		PROC_UNLOCK(proc);
1128 	} else {
1129 		PGRP_LOCK(pgrp);
1130 		SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio);
1131 		sigio->sio_pgrp = pgrp;
1132 		PGRP_UNLOCK(pgrp);
1133 	}
1134 	sx_sunlock(&proctree_lock);
1135 	SIGIO_LOCK();
1136 	*sigiop = sigio;
1137 	SIGIO_UNLOCK();
1138 	return (0);
1139 
1140 fail:
1141 	sx_sunlock(&proctree_lock);
1142 	crfree(sigio->sio_ucred);
1143 	free(sigio, M_SIGIO);
1144 	return (ret);
1145 }
1146 
1147 /*
1148  * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg).
1149  */
1150 pid_t
1151 fgetown(struct sigio **sigiop)
1152 {
1153 	pid_t pgid;
1154 
1155 	SIGIO_LOCK();
1156 	pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0;
1157 	SIGIO_UNLOCK();
1158 	return (pgid);
1159 }
1160 
1161 /*
1162  * Function drops the filedesc lock on return.
1163  */
1164 static int
1165 closefp(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1166     int holdleaders)
1167 {
1168 	int error;
1169 
1170 	FILEDESC_XLOCK_ASSERT(fdp);
1171 
1172 	if (holdleaders) {
1173 		if (td->td_proc->p_fdtol != NULL) {
1174 			/*
1175 			 * Ask fdfree() to sleep to ensure that all relevant
1176 			 * process leaders can be traversed in closef().
1177 			 */
1178 			fdp->fd_holdleaderscount++;
1179 		} else {
1180 			holdleaders = 0;
1181 		}
1182 	}
1183 
1184 	/*
1185 	 * We now hold the fp reference that used to be owned by the
1186 	 * descriptor array.  We have to unlock the FILEDESC *AFTER*
1187 	 * knote_fdclose to prevent a race of the fd getting opened, a knote
1188 	 * added, and deleteing a knote for the new fd.
1189 	 */
1190 	knote_fdclose(td, fd);
1191 
1192 	/*
1193 	 * We need to notify mqueue if the object is of type mqueue.
1194 	 */
1195 	if (fp->f_type == DTYPE_MQUEUE)
1196 		mq_fdclose(td, fd, fp);
1197 	FILEDESC_XUNLOCK(fdp);
1198 
1199 	error = closef(fp, td);
1200 	if (holdleaders) {
1201 		FILEDESC_XLOCK(fdp);
1202 		fdp->fd_holdleaderscount--;
1203 		if (fdp->fd_holdleaderscount == 0 &&
1204 		    fdp->fd_holdleaderswakeup != 0) {
1205 			fdp->fd_holdleaderswakeup = 0;
1206 			wakeup(&fdp->fd_holdleaderscount);
1207 		}
1208 		FILEDESC_XUNLOCK(fdp);
1209 	}
1210 	return (error);
1211 }
1212 
1213 /*
1214  * Close a file descriptor.
1215  */
1216 #ifndef _SYS_SYSPROTO_H_
1217 struct close_args {
1218 	int     fd;
1219 };
1220 #endif
1221 /* ARGSUSED */
1222 int
1223 sys_close(struct thread *td, struct close_args *uap)
1224 {
1225 
1226 	return (kern_close(td, uap->fd));
1227 }
1228 
1229 int
1230 kern_close(struct thread *td, int fd)
1231 {
1232 	struct filedesc *fdp;
1233 	struct file *fp;
1234 
1235 	fdp = td->td_proc->p_fd;
1236 
1237 	AUDIT_SYSCLOSE(td, fd);
1238 
1239 	FILEDESC_XLOCK(fdp);
1240 	if ((fp = fget_locked(fdp, fd)) == NULL) {
1241 		FILEDESC_XUNLOCK(fdp);
1242 		return (EBADF);
1243 	}
1244 	fdfree(fdp, fd);
1245 
1246 	/* closefp() drops the FILEDESC lock for us. */
1247 	return (closefp(fdp, fd, fp, td, 1));
1248 }
1249 
1250 /*
1251  * Close open file descriptors.
1252  */
1253 #ifndef _SYS_SYSPROTO_H_
1254 struct closefrom_args {
1255 	int	lowfd;
1256 };
1257 #endif
1258 /* ARGSUSED */
1259 int
1260 sys_closefrom(struct thread *td, struct closefrom_args *uap)
1261 {
1262 	struct filedesc *fdp;
1263 	int fd;
1264 
1265 	fdp = td->td_proc->p_fd;
1266 	AUDIT_ARG_FD(uap->lowfd);
1267 
1268 	/*
1269 	 * Treat negative starting file descriptor values identical to
1270 	 * closefrom(0) which closes all files.
1271 	 */
1272 	if (uap->lowfd < 0)
1273 		uap->lowfd = 0;
1274 	FILEDESC_SLOCK(fdp);
1275 	for (fd = uap->lowfd; fd <= fdp->fd_lastfile; fd++) {
1276 		if (fdp->fd_ofiles[fd].fde_file != NULL) {
1277 			FILEDESC_SUNLOCK(fdp);
1278 			(void)kern_close(td, fd);
1279 			FILEDESC_SLOCK(fdp);
1280 		}
1281 	}
1282 	FILEDESC_SUNLOCK(fdp);
1283 	return (0);
1284 }
1285 
1286 #if defined(COMPAT_43)
1287 /*
1288  * Return status information about a file descriptor.
1289  */
1290 #ifndef _SYS_SYSPROTO_H_
1291 struct ofstat_args {
1292 	int	fd;
1293 	struct	ostat *sb;
1294 };
1295 #endif
1296 /* ARGSUSED */
1297 int
1298 ofstat(struct thread *td, struct ofstat_args *uap)
1299 {
1300 	struct ostat oub;
1301 	struct stat ub;
1302 	int error;
1303 
1304 	error = kern_fstat(td, uap->fd, &ub);
1305 	if (error == 0) {
1306 		cvtstat(&ub, &oub);
1307 		error = copyout(&oub, uap->sb, sizeof(oub));
1308 	}
1309 	return (error);
1310 }
1311 #endif /* COMPAT_43 */
1312 
1313 #if defined(COMPAT_FREEBSD11)
1314 int
1315 freebsd11_fstat(struct thread *td, struct freebsd11_fstat_args *uap)
1316 {
1317 	struct stat sb;
1318 	struct freebsd11_stat osb;
1319 	int error;
1320 
1321 	error = kern_fstat(td, uap->fd, &sb);
1322 	if (error != 0)
1323 		return (error);
1324 	error = freebsd11_cvtstat(&sb, &osb);
1325 	if (error == 0)
1326 		error = copyout(&osb, uap->sb, sizeof(osb));
1327 	return (error);
1328 }
1329 #endif	/* COMPAT_FREEBSD11 */
1330 
1331 /*
1332  * Return status information about a file descriptor.
1333  */
1334 #ifndef _SYS_SYSPROTO_H_
1335 struct fstat_args {
1336 	int	fd;
1337 	struct	stat *sb;
1338 };
1339 #endif
1340 /* ARGSUSED */
1341 int
1342 sys_fstat(struct thread *td, struct fstat_args *uap)
1343 {
1344 	struct stat ub;
1345 	int error;
1346 
1347 	error = kern_fstat(td, uap->fd, &ub);
1348 	if (error == 0)
1349 		error = copyout(&ub, uap->sb, sizeof(ub));
1350 	return (error);
1351 }
1352 
1353 int
1354 kern_fstat(struct thread *td, int fd, struct stat *sbp)
1355 {
1356 	struct file *fp;
1357 	int error;
1358 
1359 	AUDIT_ARG_FD(fd);
1360 
1361 	error = fget(td, fd, &cap_fstat_rights, &fp);
1362 	if (error != 0)
1363 		return (error);
1364 
1365 	AUDIT_ARG_FILE(td->td_proc, fp);
1366 
1367 	error = fo_stat(fp, sbp, td->td_ucred, td);
1368 	fdrop(fp, td);
1369 #ifdef __STAT_TIME_T_EXT
1370 	if (error == 0) {
1371 		sbp->st_atim_ext = 0;
1372 		sbp->st_mtim_ext = 0;
1373 		sbp->st_ctim_ext = 0;
1374 		sbp->st_btim_ext = 0;
1375 	}
1376 #endif
1377 #ifdef KTRACE
1378 	if (error == 0 && KTRPOINT(td, KTR_STRUCT))
1379 		ktrstat(sbp);
1380 #endif
1381 	return (error);
1382 }
1383 
1384 #if defined(COMPAT_FREEBSD11)
1385 /*
1386  * Return status information about a file descriptor.
1387  */
1388 #ifndef _SYS_SYSPROTO_H_
1389 struct freebsd11_nfstat_args {
1390 	int	fd;
1391 	struct	nstat *sb;
1392 };
1393 #endif
1394 /* ARGSUSED */
1395 int
1396 freebsd11_nfstat(struct thread *td, struct freebsd11_nfstat_args *uap)
1397 {
1398 	struct nstat nub;
1399 	struct stat ub;
1400 	int error;
1401 
1402 	error = kern_fstat(td, uap->fd, &ub);
1403 	if (error == 0) {
1404 		freebsd11_cvtnstat(&ub, &nub);
1405 		error = copyout(&nub, uap->sb, sizeof(nub));
1406 	}
1407 	return (error);
1408 }
1409 #endif /* COMPAT_FREEBSD11 */
1410 
1411 /*
1412  * Return pathconf information about a file descriptor.
1413  */
1414 #ifndef _SYS_SYSPROTO_H_
1415 struct fpathconf_args {
1416 	int	fd;
1417 	int	name;
1418 };
1419 #endif
1420 /* ARGSUSED */
1421 int
1422 sys_fpathconf(struct thread *td, struct fpathconf_args *uap)
1423 {
1424 	long value;
1425 	int error;
1426 
1427 	error = kern_fpathconf(td, uap->fd, uap->name, &value);
1428 	if (error == 0)
1429 		td->td_retval[0] = value;
1430 	return (error);
1431 }
1432 
1433 int
1434 kern_fpathconf(struct thread *td, int fd, int name, long *valuep)
1435 {
1436 	struct file *fp;
1437 	struct vnode *vp;
1438 	int error;
1439 
1440 	error = fget(td, fd, &cap_fpathconf_rights, &fp);
1441 	if (error != 0)
1442 		return (error);
1443 
1444 	if (name == _PC_ASYNC_IO) {
1445 		*valuep = _POSIX_ASYNCHRONOUS_IO;
1446 		goto out;
1447 	}
1448 	vp = fp->f_vnode;
1449 	if (vp != NULL) {
1450 		vn_lock(vp, LK_SHARED | LK_RETRY);
1451 		error = VOP_PATHCONF(vp, name, valuep);
1452 		VOP_UNLOCK(vp, 0);
1453 	} else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) {
1454 		if (name != _PC_PIPE_BUF) {
1455 			error = EINVAL;
1456 		} else {
1457 			*valuep = PIPE_BUF;
1458 			error = 0;
1459 		}
1460 	} else {
1461 		error = EOPNOTSUPP;
1462 	}
1463 out:
1464 	fdrop(fp, td);
1465 	return (error);
1466 }
1467 
1468 /*
1469  * Initialize filecaps structure.
1470  */
1471 void
1472 filecaps_init(struct filecaps *fcaps)
1473 {
1474 
1475 	bzero(fcaps, sizeof(*fcaps));
1476 	fcaps->fc_nioctls = -1;
1477 }
1478 
1479 /*
1480  * Copy filecaps structure allocating memory for ioctls array if needed.
1481  *
1482  * The last parameter indicates whether the fdtable is locked. If it is not and
1483  * ioctls are encountered, copying fails and the caller must lock the table.
1484  *
1485  * Note that if the table was not locked, the caller has to check the relevant
1486  * sequence counter to determine whether the operation was successful.
1487  */
1488 bool
1489 filecaps_copy(const struct filecaps *src, struct filecaps *dst, bool locked)
1490 {
1491 	size_t size;
1492 
1493 	if (src->fc_ioctls != NULL && !locked)
1494 		return (false);
1495 	memcpy(dst, src, sizeof(*src));
1496 	if (src->fc_ioctls == NULL)
1497 		return (true);
1498 
1499 	KASSERT(src->fc_nioctls > 0,
1500 	    ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1501 
1502 	size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1503 	dst->fc_ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1504 	memcpy(dst->fc_ioctls, src->fc_ioctls, size);
1505 	return (true);
1506 }
1507 
1508 static u_long *
1509 filecaps_copy_prep(const struct filecaps *src)
1510 {
1511 	u_long *ioctls;
1512 	size_t size;
1513 
1514 	if (src->fc_ioctls == NULL)
1515 		return (NULL);
1516 
1517 	KASSERT(src->fc_nioctls > 0,
1518 	    ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1519 
1520 	size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1521 	ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1522 	return (ioctls);
1523 }
1524 
1525 static void
1526 filecaps_copy_finish(const struct filecaps *src, struct filecaps *dst,
1527     u_long *ioctls)
1528 {
1529 	size_t size;
1530 
1531 	*dst = *src;
1532 	if (src->fc_ioctls == NULL) {
1533 		MPASS(ioctls == NULL);
1534 		return;
1535 	}
1536 
1537 	size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1538 	dst->fc_ioctls = ioctls;
1539 	bcopy(src->fc_ioctls, dst->fc_ioctls, size);
1540 }
1541 
1542 /*
1543  * Move filecaps structure to the new place and clear the old place.
1544  */
1545 void
1546 filecaps_move(struct filecaps *src, struct filecaps *dst)
1547 {
1548 
1549 	*dst = *src;
1550 	bzero(src, sizeof(*src));
1551 }
1552 
1553 /*
1554  * Fill the given filecaps structure with full rights.
1555  */
1556 static void
1557 filecaps_fill(struct filecaps *fcaps)
1558 {
1559 
1560 	CAP_ALL(&fcaps->fc_rights);
1561 	fcaps->fc_ioctls = NULL;
1562 	fcaps->fc_nioctls = -1;
1563 	fcaps->fc_fcntls = CAP_FCNTL_ALL;
1564 }
1565 
1566 /*
1567  * Free memory allocated within filecaps structure.
1568  */
1569 void
1570 filecaps_free(struct filecaps *fcaps)
1571 {
1572 
1573 	free(fcaps->fc_ioctls, M_FILECAPS);
1574 	bzero(fcaps, sizeof(*fcaps));
1575 }
1576 
1577 static u_long *
1578 filecaps_free_prep(struct filecaps *fcaps)
1579 {
1580 	u_long *ioctls;
1581 
1582 	ioctls = fcaps->fc_ioctls;
1583 	bzero(fcaps, sizeof(*fcaps));
1584 	return (ioctls);
1585 }
1586 
1587 static void
1588 filecaps_free_finish(u_long *ioctls)
1589 {
1590 
1591 	free(ioctls, M_FILECAPS);
1592 }
1593 
1594 /*
1595  * Validate the given filecaps structure.
1596  */
1597 static void
1598 filecaps_validate(const struct filecaps *fcaps, const char *func)
1599 {
1600 
1601 	KASSERT(cap_rights_is_valid(&fcaps->fc_rights),
1602 	    ("%s: invalid rights", func));
1603 	KASSERT((fcaps->fc_fcntls & ~CAP_FCNTL_ALL) == 0,
1604 	    ("%s: invalid fcntls", func));
1605 	KASSERT(fcaps->fc_fcntls == 0 ||
1606 	    cap_rights_is_set(&fcaps->fc_rights, CAP_FCNTL),
1607 	    ("%s: fcntls without CAP_FCNTL", func));
1608 	KASSERT(fcaps->fc_ioctls != NULL ? fcaps->fc_nioctls > 0 :
1609 	    (fcaps->fc_nioctls == -1 || fcaps->fc_nioctls == 0),
1610 	    ("%s: invalid ioctls", func));
1611 	KASSERT(fcaps->fc_nioctls == 0 ||
1612 	    cap_rights_is_set(&fcaps->fc_rights, CAP_IOCTL),
1613 	    ("%s: ioctls without CAP_IOCTL", func));
1614 }
1615 
1616 static void
1617 fdgrowtable_exp(struct filedesc *fdp, int nfd)
1618 {
1619 	int nfd1;
1620 
1621 	FILEDESC_XLOCK_ASSERT(fdp);
1622 
1623 	nfd1 = fdp->fd_nfiles * 2;
1624 	if (nfd1 < nfd)
1625 		nfd1 = nfd;
1626 	fdgrowtable(fdp, nfd1);
1627 }
1628 
1629 /*
1630  * Grow the file table to accommodate (at least) nfd descriptors.
1631  */
1632 static void
1633 fdgrowtable(struct filedesc *fdp, int nfd)
1634 {
1635 	struct filedesc0 *fdp0;
1636 	struct freetable *ft;
1637 	struct fdescenttbl *ntable;
1638 	struct fdescenttbl *otable;
1639 	int nnfiles, onfiles;
1640 	NDSLOTTYPE *nmap, *omap;
1641 
1642 	/*
1643 	 * If lastfile is -1 this struct filedesc was just allocated and we are
1644 	 * growing it to accommodate for the one we are going to copy from. There
1645 	 * is no need to have a lock on this one as it's not visible to anyone.
1646 	 */
1647 	if (fdp->fd_lastfile != -1)
1648 		FILEDESC_XLOCK_ASSERT(fdp);
1649 
1650 	KASSERT(fdp->fd_nfiles > 0, ("zero-length file table"));
1651 
1652 	/* save old values */
1653 	onfiles = fdp->fd_nfiles;
1654 	otable = fdp->fd_files;
1655 	omap = fdp->fd_map;
1656 
1657 	/* compute the size of the new table */
1658 	nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */
1659 	if (nnfiles <= onfiles)
1660 		/* the table is already large enough */
1661 		return;
1662 
1663 	/*
1664 	 * Allocate a new table.  We need enough space for the number of
1665 	 * entries, file entries themselves and the struct freetable we will use
1666 	 * when we decommission the table and place it on the freelist.
1667 	 * We place the struct freetable in the middle so we don't have
1668 	 * to worry about padding.
1669 	 */
1670 	ntable = malloc(offsetof(struct fdescenttbl, fdt_ofiles) +
1671 	    nnfiles * sizeof(ntable->fdt_ofiles[0]) +
1672 	    sizeof(struct freetable),
1673 	    M_FILEDESC, M_ZERO | M_WAITOK);
1674 	/* copy the old data */
1675 	ntable->fdt_nfiles = nnfiles;
1676 	memcpy(ntable->fdt_ofiles, otable->fdt_ofiles,
1677 	    onfiles * sizeof(ntable->fdt_ofiles[0]));
1678 
1679 	/*
1680 	 * Allocate a new map only if the old is not large enough.  It will
1681 	 * grow at a slower rate than the table as it can map more
1682 	 * entries than the table can hold.
1683 	 */
1684 	if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) {
1685 		nmap = malloc(NDSLOTS(nnfiles) * NDSLOTSIZE, M_FILEDESC,
1686 		    M_ZERO | M_WAITOK);
1687 		/* copy over the old data and update the pointer */
1688 		memcpy(nmap, omap, NDSLOTS(onfiles) * sizeof(*omap));
1689 		fdp->fd_map = nmap;
1690 	}
1691 
1692 	/*
1693 	 * Make sure that ntable is correctly initialized before we replace
1694 	 * fd_files poiner. Otherwise fget_unlocked() may see inconsistent
1695 	 * data.
1696 	 */
1697 	atomic_store_rel_ptr((volatile void *)&fdp->fd_files, (uintptr_t)ntable);
1698 
1699 	/*
1700 	 * Do not free the old file table, as some threads may still
1701 	 * reference entries within it.  Instead, place it on a freelist
1702 	 * which will be processed when the struct filedesc is released.
1703 	 *
1704 	 * Note that if onfiles == NDFILE, we're dealing with the original
1705 	 * static allocation contained within (struct filedesc0 *)fdp,
1706 	 * which must not be freed.
1707 	 */
1708 	if (onfiles > NDFILE) {
1709 		ft = (struct freetable *)&otable->fdt_ofiles[onfiles];
1710 		fdp0 = (struct filedesc0 *)fdp;
1711 		ft->ft_table = otable;
1712 		SLIST_INSERT_HEAD(&fdp0->fd_free, ft, ft_next);
1713 	}
1714 	/*
1715 	 * The map does not have the same possibility of threads still
1716 	 * holding references to it.  So always free it as long as it
1717 	 * does not reference the original static allocation.
1718 	 */
1719 	if (NDSLOTS(onfiles) > NDSLOTS(NDFILE))
1720 		free(omap, M_FILEDESC);
1721 }
1722 
1723 /*
1724  * Allocate a file descriptor for the process.
1725  */
1726 int
1727 fdalloc(struct thread *td, int minfd, int *result)
1728 {
1729 	struct proc *p = td->td_proc;
1730 	struct filedesc *fdp = p->p_fd;
1731 	int fd, maxfd, allocfd;
1732 #ifdef RACCT
1733 	int error;
1734 #endif
1735 
1736 	FILEDESC_XLOCK_ASSERT(fdp);
1737 
1738 	if (fdp->fd_freefile > minfd)
1739 		minfd = fdp->fd_freefile;
1740 
1741 	maxfd = getmaxfd(td);
1742 
1743 	/*
1744 	 * Search the bitmap for a free descriptor starting at minfd.
1745 	 * If none is found, grow the file table.
1746 	 */
1747 	fd = fd_first_free(fdp, minfd, fdp->fd_nfiles);
1748 	if (fd >= maxfd)
1749 		return (EMFILE);
1750 	if (fd >= fdp->fd_nfiles) {
1751 		allocfd = min(fd * 2, maxfd);
1752 #ifdef RACCT
1753 		if (racct_enable) {
1754 			PROC_LOCK(p);
1755 			error = racct_set(p, RACCT_NOFILE, allocfd);
1756 			PROC_UNLOCK(p);
1757 			if (error != 0)
1758 				return (EMFILE);
1759 		}
1760 #endif
1761 		/*
1762 		 * fd is already equal to first free descriptor >= minfd, so
1763 		 * we only need to grow the table and we are done.
1764 		 */
1765 		fdgrowtable_exp(fdp, allocfd);
1766 	}
1767 
1768 	/*
1769 	 * Perform some sanity checks, then mark the file descriptor as
1770 	 * used and return it to the caller.
1771 	 */
1772 	KASSERT(fd >= 0 && fd < min(maxfd, fdp->fd_nfiles),
1773 	    ("invalid descriptor %d", fd));
1774 	KASSERT(!fdisused(fdp, fd),
1775 	    ("fd_first_free() returned non-free descriptor"));
1776 	KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
1777 	    ("file descriptor isn't free"));
1778 	fdused(fdp, fd);
1779 	*result = fd;
1780 	return (0);
1781 }
1782 
1783 /*
1784  * Allocate n file descriptors for the process.
1785  */
1786 int
1787 fdallocn(struct thread *td, int minfd, int *fds, int n)
1788 {
1789 	struct proc *p = td->td_proc;
1790 	struct filedesc *fdp = p->p_fd;
1791 	int i;
1792 
1793 	FILEDESC_XLOCK_ASSERT(fdp);
1794 
1795 	for (i = 0; i < n; i++)
1796 		if (fdalloc(td, 0, &fds[i]) != 0)
1797 			break;
1798 
1799 	if (i < n) {
1800 		for (i--; i >= 0; i--)
1801 			fdunused(fdp, fds[i]);
1802 		return (EMFILE);
1803 	}
1804 
1805 	return (0);
1806 }
1807 
1808 /*
1809  * Create a new open file structure and allocate a file descriptor for the
1810  * process that refers to it.  We add one reference to the file for the
1811  * descriptor table and one reference for resultfp. This is to prevent us
1812  * being preempted and the entry in the descriptor table closed after we
1813  * release the FILEDESC lock.
1814  */
1815 int
1816 falloc_caps(struct thread *td, struct file **resultfp, int *resultfd, int flags,
1817     struct filecaps *fcaps)
1818 {
1819 	struct file *fp;
1820 	int error, fd;
1821 
1822 	error = falloc_noinstall(td, &fp);
1823 	if (error)
1824 		return (error);		/* no reference held on error */
1825 
1826 	error = finstall(td, fp, &fd, flags, fcaps);
1827 	if (error) {
1828 		fdrop(fp, td);		/* one reference (fp only) */
1829 		return (error);
1830 	}
1831 
1832 	if (resultfp != NULL)
1833 		*resultfp = fp;		/* copy out result */
1834 	else
1835 		fdrop(fp, td);		/* release local reference */
1836 
1837 	if (resultfd != NULL)
1838 		*resultfd = fd;
1839 
1840 	return (0);
1841 }
1842 
1843 /*
1844  * Create a new open file structure without allocating a file descriptor.
1845  */
1846 int
1847 falloc_noinstall(struct thread *td, struct file **resultfp)
1848 {
1849 	struct file *fp;
1850 	int maxuserfiles = maxfiles - (maxfiles / 20);
1851 	int openfiles_new;
1852 	static struct timeval lastfail;
1853 	static int curfail;
1854 
1855 	KASSERT(resultfp != NULL, ("%s: resultfp == NULL", __func__));
1856 
1857 	openfiles_new = atomic_fetchadd_int(&openfiles, 1) + 1;
1858 	if ((openfiles_new >= maxuserfiles &&
1859 	    priv_check(td, PRIV_MAXFILES) != 0) ||
1860 	    openfiles_new >= maxfiles) {
1861 		atomic_subtract_int(&openfiles, 1);
1862 		if (ppsratecheck(&lastfail, &curfail, 1)) {
1863 			printf("kern.maxfiles limit exceeded by uid %i, (%s) "
1864 			    "please see tuning(7).\n", td->td_ucred->cr_ruid, td->td_proc->p_comm);
1865 		}
1866 		return (ENFILE);
1867 	}
1868 	fp = uma_zalloc(file_zone, M_WAITOK);
1869 	bzero(fp, sizeof(*fp));
1870 	refcount_init(&fp->f_count, 1);
1871 	fp->f_cred = crhold(td->td_ucred);
1872 	fp->f_ops = &badfileops;
1873 	*resultfp = fp;
1874 	return (0);
1875 }
1876 
1877 /*
1878  * Install a file in a file descriptor table.
1879  */
1880 void
1881 _finstall(struct filedesc *fdp, struct file *fp, int fd, int flags,
1882     struct filecaps *fcaps)
1883 {
1884 	struct filedescent *fde;
1885 
1886 	MPASS(fp != NULL);
1887 	if (fcaps != NULL)
1888 		filecaps_validate(fcaps, __func__);
1889 	FILEDESC_XLOCK_ASSERT(fdp);
1890 
1891 	fde = &fdp->fd_ofiles[fd];
1892 #ifdef CAPABILITIES
1893 	seq_write_begin(&fde->fde_seq);
1894 #endif
1895 	fde->fde_file = fp;
1896 	fde->fde_flags = (flags & O_CLOEXEC) != 0 ? UF_EXCLOSE : 0;
1897 	if (fcaps != NULL)
1898 		filecaps_move(fcaps, &fde->fde_caps);
1899 	else
1900 		filecaps_fill(&fde->fde_caps);
1901 #ifdef CAPABILITIES
1902 	seq_write_end(&fde->fde_seq);
1903 #endif
1904 }
1905 
1906 int
1907 finstall(struct thread *td, struct file *fp, int *fd, int flags,
1908     struct filecaps *fcaps)
1909 {
1910 	struct filedesc *fdp = td->td_proc->p_fd;
1911 	int error;
1912 
1913 	MPASS(fd != NULL);
1914 
1915 	FILEDESC_XLOCK(fdp);
1916 	if ((error = fdalloc(td, 0, fd))) {
1917 		FILEDESC_XUNLOCK(fdp);
1918 		return (error);
1919 	}
1920 	fhold(fp);
1921 	_finstall(fdp, fp, *fd, flags, fcaps);
1922 	FILEDESC_XUNLOCK(fdp);
1923 	return (0);
1924 }
1925 
1926 /*
1927  * Build a new filedesc structure from another.
1928  * Copy the current, root, and jail root vnode references.
1929  *
1930  * If fdp is not NULL, return with it shared locked.
1931  */
1932 struct filedesc *
1933 fdinit(struct filedesc *fdp, bool prepfiles)
1934 {
1935 	struct filedesc0 *newfdp0;
1936 	struct filedesc *newfdp;
1937 
1938 	newfdp0 = uma_zalloc(filedesc0_zone, M_WAITOK | M_ZERO);
1939 	newfdp = &newfdp0->fd_fd;
1940 
1941 	/* Create the file descriptor table. */
1942 	FILEDESC_LOCK_INIT(newfdp);
1943 	refcount_init(&newfdp->fd_refcnt, 1);
1944 	refcount_init(&newfdp->fd_holdcnt, 1);
1945 	newfdp->fd_cmask = CMASK;
1946 	newfdp->fd_map = newfdp0->fd_dmap;
1947 	newfdp->fd_lastfile = -1;
1948 	newfdp->fd_files = (struct fdescenttbl *)&newfdp0->fd_dfiles;
1949 	newfdp->fd_files->fdt_nfiles = NDFILE;
1950 
1951 	if (fdp == NULL)
1952 		return (newfdp);
1953 
1954 	if (prepfiles && fdp->fd_lastfile >= newfdp->fd_nfiles)
1955 		fdgrowtable(newfdp, fdp->fd_lastfile + 1);
1956 
1957 	FILEDESC_SLOCK(fdp);
1958 	newfdp->fd_cdir = fdp->fd_cdir;
1959 	if (newfdp->fd_cdir)
1960 		vrefact(newfdp->fd_cdir);
1961 	newfdp->fd_rdir = fdp->fd_rdir;
1962 	if (newfdp->fd_rdir)
1963 		vrefact(newfdp->fd_rdir);
1964 	newfdp->fd_jdir = fdp->fd_jdir;
1965 	if (newfdp->fd_jdir)
1966 		vrefact(newfdp->fd_jdir);
1967 
1968 	if (!prepfiles) {
1969 		FILEDESC_SUNLOCK(fdp);
1970 	} else {
1971 		while (fdp->fd_lastfile >= newfdp->fd_nfiles) {
1972 			FILEDESC_SUNLOCK(fdp);
1973 			fdgrowtable(newfdp, fdp->fd_lastfile + 1);
1974 			FILEDESC_SLOCK(fdp);
1975 		}
1976 	}
1977 
1978 	return (newfdp);
1979 }
1980 
1981 static struct filedesc *
1982 fdhold(struct proc *p)
1983 {
1984 	struct filedesc *fdp;
1985 
1986 	PROC_LOCK_ASSERT(p, MA_OWNED);
1987 	fdp = p->p_fd;
1988 	if (fdp != NULL)
1989 		refcount_acquire(&fdp->fd_holdcnt);
1990 	return (fdp);
1991 }
1992 
1993 static void
1994 fddrop(struct filedesc *fdp)
1995 {
1996 
1997 	if (fdp->fd_holdcnt > 1) {
1998 		if (refcount_release(&fdp->fd_holdcnt) == 0)
1999 			return;
2000 	}
2001 
2002 	FILEDESC_LOCK_DESTROY(fdp);
2003 	uma_zfree(filedesc0_zone, fdp);
2004 }
2005 
2006 /*
2007  * Share a filedesc structure.
2008  */
2009 struct filedesc *
2010 fdshare(struct filedesc *fdp)
2011 {
2012 
2013 	refcount_acquire(&fdp->fd_refcnt);
2014 	return (fdp);
2015 }
2016 
2017 /*
2018  * Unshare a filedesc structure, if necessary by making a copy
2019  */
2020 void
2021 fdunshare(struct thread *td)
2022 {
2023 	struct filedesc *tmp;
2024 	struct proc *p = td->td_proc;
2025 
2026 	if (p->p_fd->fd_refcnt == 1)
2027 		return;
2028 
2029 	tmp = fdcopy(p->p_fd);
2030 	fdescfree(td);
2031 	p->p_fd = tmp;
2032 }
2033 
2034 void
2035 fdinstall_remapped(struct thread *td, struct filedesc *fdp)
2036 {
2037 
2038 	fdescfree(td);
2039 	td->td_proc->p_fd = fdp;
2040 }
2041 
2042 /*
2043  * Copy a filedesc structure.  A NULL pointer in returns a NULL reference,
2044  * this is to ease callers, not catch errors.
2045  */
2046 struct filedesc *
2047 fdcopy(struct filedesc *fdp)
2048 {
2049 	struct filedesc *newfdp;
2050 	struct filedescent *nfde, *ofde;
2051 	int i;
2052 
2053 	MPASS(fdp != NULL);
2054 
2055 	newfdp = fdinit(fdp, true);
2056 	/* copy all passable descriptors (i.e. not kqueue) */
2057 	newfdp->fd_freefile = -1;
2058 	for (i = 0; i <= fdp->fd_lastfile; ++i) {
2059 		ofde = &fdp->fd_ofiles[i];
2060 		if (ofde->fde_file == NULL ||
2061 		    (ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0) {
2062 			if (newfdp->fd_freefile == -1)
2063 				newfdp->fd_freefile = i;
2064 			continue;
2065 		}
2066 		nfde = &newfdp->fd_ofiles[i];
2067 		*nfde = *ofde;
2068 		filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true);
2069 		fhold(nfde->fde_file);
2070 		fdused_init(newfdp, i);
2071 		newfdp->fd_lastfile = i;
2072 	}
2073 	if (newfdp->fd_freefile == -1)
2074 		newfdp->fd_freefile = i;
2075 	newfdp->fd_cmask = fdp->fd_cmask;
2076 	FILEDESC_SUNLOCK(fdp);
2077 	return (newfdp);
2078 }
2079 
2080 /*
2081  * Copies a filedesc structure, while remapping all file descriptors
2082  * stored inside using a translation table.
2083  *
2084  * File descriptors are copied over to the new file descriptor table,
2085  * regardless of whether the close-on-exec flag is set.
2086  */
2087 int
2088 fdcopy_remapped(struct filedesc *fdp, const int *fds, size_t nfds,
2089     struct filedesc **ret)
2090 {
2091 	struct filedesc *newfdp;
2092 	struct filedescent *nfde, *ofde;
2093 	int error, i;
2094 
2095 	MPASS(fdp != NULL);
2096 
2097 	newfdp = fdinit(fdp, true);
2098 	if (nfds > fdp->fd_lastfile + 1) {
2099 		/* New table cannot be larger than the old one. */
2100 		error = E2BIG;
2101 		goto bad;
2102 	}
2103 	/* Copy all passable descriptors (i.e. not kqueue). */
2104 	newfdp->fd_freefile = nfds;
2105 	for (i = 0; i < nfds; ++i) {
2106 		if (fds[i] < 0 || fds[i] > fdp->fd_lastfile) {
2107 			/* File descriptor out of bounds. */
2108 			error = EBADF;
2109 			goto bad;
2110 		}
2111 		ofde = &fdp->fd_ofiles[fds[i]];
2112 		if (ofde->fde_file == NULL) {
2113 			/* Unused file descriptor. */
2114 			error = EBADF;
2115 			goto bad;
2116 		}
2117 		if ((ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0) {
2118 			/* File descriptor cannot be passed. */
2119 			error = EINVAL;
2120 			goto bad;
2121 		}
2122 		nfde = &newfdp->fd_ofiles[i];
2123 		*nfde = *ofde;
2124 		filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true);
2125 		fhold(nfde->fde_file);
2126 		fdused_init(newfdp, i);
2127 		newfdp->fd_lastfile = i;
2128 	}
2129 	newfdp->fd_cmask = fdp->fd_cmask;
2130 	FILEDESC_SUNLOCK(fdp);
2131 	*ret = newfdp;
2132 	return (0);
2133 bad:
2134 	FILEDESC_SUNLOCK(fdp);
2135 	fdescfree_remapped(newfdp);
2136 	return (error);
2137 }
2138 
2139 /*
2140  * Clear POSIX style locks. This is only used when fdp looses a reference (i.e.
2141  * one of processes using it exits) and the table used to be shared.
2142  */
2143 static void
2144 fdclearlocks(struct thread *td)
2145 {
2146 	struct filedesc *fdp;
2147 	struct filedesc_to_leader *fdtol;
2148 	struct flock lf;
2149 	struct file *fp;
2150 	struct proc *p;
2151 	struct vnode *vp;
2152 	int i;
2153 
2154 	p = td->td_proc;
2155 	fdp = p->p_fd;
2156 	fdtol = p->p_fdtol;
2157 	MPASS(fdtol != NULL);
2158 
2159 	FILEDESC_XLOCK(fdp);
2160 	KASSERT(fdtol->fdl_refcount > 0,
2161 	    ("filedesc_to_refcount botch: fdl_refcount=%d",
2162 	    fdtol->fdl_refcount));
2163 	if (fdtol->fdl_refcount == 1 &&
2164 	    (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2165 		for (i = 0; i <= fdp->fd_lastfile; i++) {
2166 			fp = fdp->fd_ofiles[i].fde_file;
2167 			if (fp == NULL || fp->f_type != DTYPE_VNODE)
2168 				continue;
2169 			fhold(fp);
2170 			FILEDESC_XUNLOCK(fdp);
2171 			lf.l_whence = SEEK_SET;
2172 			lf.l_start = 0;
2173 			lf.l_len = 0;
2174 			lf.l_type = F_UNLCK;
2175 			vp = fp->f_vnode;
2176 			(void) VOP_ADVLOCK(vp,
2177 			    (caddr_t)p->p_leader, F_UNLCK,
2178 			    &lf, F_POSIX);
2179 			FILEDESC_XLOCK(fdp);
2180 			fdrop(fp, td);
2181 		}
2182 	}
2183 retry:
2184 	if (fdtol->fdl_refcount == 1) {
2185 		if (fdp->fd_holdleaderscount > 0 &&
2186 		    (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2187 			/*
2188 			 * close() or kern_dup() has cleared a reference
2189 			 * in a shared file descriptor table.
2190 			 */
2191 			fdp->fd_holdleaderswakeup = 1;
2192 			sx_sleep(&fdp->fd_holdleaderscount,
2193 			    FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0);
2194 			goto retry;
2195 		}
2196 		if (fdtol->fdl_holdcount > 0) {
2197 			/*
2198 			 * Ensure that fdtol->fdl_leader remains
2199 			 * valid in closef().
2200 			 */
2201 			fdtol->fdl_wakeup = 1;
2202 			sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK,
2203 			    "fdlhold", 0);
2204 			goto retry;
2205 		}
2206 	}
2207 	fdtol->fdl_refcount--;
2208 	if (fdtol->fdl_refcount == 0 &&
2209 	    fdtol->fdl_holdcount == 0) {
2210 		fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
2211 		fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
2212 	} else
2213 		fdtol = NULL;
2214 	p->p_fdtol = NULL;
2215 	FILEDESC_XUNLOCK(fdp);
2216 	if (fdtol != NULL)
2217 		free(fdtol, M_FILEDESC_TO_LEADER);
2218 }
2219 
2220 /*
2221  * Release a filedesc structure.
2222  */
2223 static void
2224 fdescfree_fds(struct thread *td, struct filedesc *fdp, bool needclose)
2225 {
2226 	struct filedesc0 *fdp0;
2227 	struct freetable *ft, *tft;
2228 	struct filedescent *fde;
2229 	struct file *fp;
2230 	int i;
2231 
2232 	for (i = 0; i <= fdp->fd_lastfile; i++) {
2233 		fde = &fdp->fd_ofiles[i];
2234 		fp = fde->fde_file;
2235 		if (fp != NULL) {
2236 			fdefree_last(fde);
2237 			if (needclose)
2238 				(void) closef(fp, td);
2239 			else
2240 				fdrop(fp, td);
2241 		}
2242 	}
2243 
2244 	if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE))
2245 		free(fdp->fd_map, M_FILEDESC);
2246 	if (fdp->fd_nfiles > NDFILE)
2247 		free(fdp->fd_files, M_FILEDESC);
2248 
2249 	fdp0 = (struct filedesc0 *)fdp;
2250 	SLIST_FOREACH_SAFE(ft, &fdp0->fd_free, ft_next, tft)
2251 		free(ft->ft_table, M_FILEDESC);
2252 
2253 	fddrop(fdp);
2254 }
2255 
2256 void
2257 fdescfree(struct thread *td)
2258 {
2259 	struct proc *p;
2260 	struct filedesc *fdp;
2261 	struct vnode *cdir, *jdir, *rdir;
2262 
2263 	p = td->td_proc;
2264 	fdp = p->p_fd;
2265 	MPASS(fdp != NULL);
2266 
2267 #ifdef RACCT
2268 	if (racct_enable) {
2269 		PROC_LOCK(p);
2270 		racct_set(p, RACCT_NOFILE, 0);
2271 		PROC_UNLOCK(p);
2272 	}
2273 #endif
2274 
2275 	if (p->p_fdtol != NULL)
2276 		fdclearlocks(td);
2277 
2278 	PROC_LOCK(p);
2279 	p->p_fd = NULL;
2280 	PROC_UNLOCK(p);
2281 
2282 	if (refcount_release(&fdp->fd_refcnt) == 0)
2283 		return;
2284 
2285 	FILEDESC_XLOCK(fdp);
2286 	cdir = fdp->fd_cdir;
2287 	fdp->fd_cdir = NULL;
2288 	rdir = fdp->fd_rdir;
2289 	fdp->fd_rdir = NULL;
2290 	jdir = fdp->fd_jdir;
2291 	fdp->fd_jdir = NULL;
2292 	FILEDESC_XUNLOCK(fdp);
2293 
2294 	if (cdir != NULL)
2295 		vrele(cdir);
2296 	if (rdir != NULL)
2297 		vrele(rdir);
2298 	if (jdir != NULL)
2299 		vrele(jdir);
2300 
2301 	fdescfree_fds(td, fdp, 1);
2302 }
2303 
2304 void
2305 fdescfree_remapped(struct filedesc *fdp)
2306 {
2307 
2308 	if (fdp->fd_cdir != NULL)
2309 		vrele(fdp->fd_cdir);
2310 	if (fdp->fd_rdir != NULL)
2311 		vrele(fdp->fd_rdir);
2312 	if (fdp->fd_jdir != NULL)
2313 		vrele(fdp->fd_jdir);
2314 
2315 	fdescfree_fds(curthread, fdp, 0);
2316 }
2317 
2318 /*
2319  * For setugid programs, we don't want to people to use that setugidness
2320  * to generate error messages which write to a file which otherwise would
2321  * otherwise be off-limits to the process.  We check for filesystems where
2322  * the vnode can change out from under us after execve (like [lin]procfs).
2323  *
2324  * Since fdsetugidsafety calls this only for fd 0, 1 and 2, this check is
2325  * sufficient.  We also don't check for setugidness since we know we are.
2326  */
2327 static bool
2328 is_unsafe(struct file *fp)
2329 {
2330 	struct vnode *vp;
2331 
2332 	if (fp->f_type != DTYPE_VNODE)
2333 		return (false);
2334 
2335 	vp = fp->f_vnode;
2336 	return ((vp->v_vflag & VV_PROCDEP) != 0);
2337 }
2338 
2339 /*
2340  * Make this setguid thing safe, if at all possible.
2341  */
2342 void
2343 fdsetugidsafety(struct thread *td)
2344 {
2345 	struct filedesc *fdp;
2346 	struct file *fp;
2347 	int i;
2348 
2349 	fdp = td->td_proc->p_fd;
2350 	KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2351 	MPASS(fdp->fd_nfiles >= 3);
2352 	for (i = 0; i <= 2; i++) {
2353 		fp = fdp->fd_ofiles[i].fde_file;
2354 		if (fp != NULL && is_unsafe(fp)) {
2355 			FILEDESC_XLOCK(fdp);
2356 			knote_fdclose(td, i);
2357 			/*
2358 			 * NULL-out descriptor prior to close to avoid
2359 			 * a race while close blocks.
2360 			 */
2361 			fdfree(fdp, i);
2362 			FILEDESC_XUNLOCK(fdp);
2363 			(void) closef(fp, td);
2364 		}
2365 	}
2366 }
2367 
2368 /*
2369  * If a specific file object occupies a specific file descriptor, close the
2370  * file descriptor entry and drop a reference on the file object.  This is a
2371  * convenience function to handle a subsequent error in a function that calls
2372  * falloc() that handles the race that another thread might have closed the
2373  * file descriptor out from under the thread creating the file object.
2374  */
2375 void
2376 fdclose(struct thread *td, struct file *fp, int idx)
2377 {
2378 	struct filedesc *fdp = td->td_proc->p_fd;
2379 
2380 	FILEDESC_XLOCK(fdp);
2381 	if (fdp->fd_ofiles[idx].fde_file == fp) {
2382 		fdfree(fdp, idx);
2383 		FILEDESC_XUNLOCK(fdp);
2384 		fdrop(fp, td);
2385 	} else
2386 		FILEDESC_XUNLOCK(fdp);
2387 }
2388 
2389 /*
2390  * Close any files on exec?
2391  */
2392 void
2393 fdcloseexec(struct thread *td)
2394 {
2395 	struct filedesc *fdp;
2396 	struct filedescent *fde;
2397 	struct file *fp;
2398 	int i;
2399 
2400 	fdp = td->td_proc->p_fd;
2401 	KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2402 	for (i = 0; i <= fdp->fd_lastfile; i++) {
2403 		fde = &fdp->fd_ofiles[i];
2404 		fp = fde->fde_file;
2405 		if (fp != NULL && (fp->f_type == DTYPE_MQUEUE ||
2406 		    (fde->fde_flags & UF_EXCLOSE))) {
2407 			FILEDESC_XLOCK(fdp);
2408 			fdfree(fdp, i);
2409 			(void) closefp(fdp, i, fp, td, 0);
2410 			FILEDESC_UNLOCK_ASSERT(fdp);
2411 		}
2412 	}
2413 }
2414 
2415 /*
2416  * It is unsafe for set[ug]id processes to be started with file
2417  * descriptors 0..2 closed, as these descriptors are given implicit
2418  * significance in the Standard C library.  fdcheckstd() will create a
2419  * descriptor referencing /dev/null for each of stdin, stdout, and
2420  * stderr that is not already open.
2421  */
2422 int
2423 fdcheckstd(struct thread *td)
2424 {
2425 	struct filedesc *fdp;
2426 	register_t save;
2427 	int i, error, devnull;
2428 
2429 	fdp = td->td_proc->p_fd;
2430 	KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2431 	MPASS(fdp->fd_nfiles >= 3);
2432 	devnull = -1;
2433 	for (i = 0; i <= 2; i++) {
2434 		if (fdp->fd_ofiles[i].fde_file != NULL)
2435 			continue;
2436 
2437 		save = td->td_retval[0];
2438 		if (devnull != -1) {
2439 			error = kern_dup(td, FDDUP_FIXED, 0, devnull, i);
2440 		} else {
2441 			error = kern_openat(td, AT_FDCWD, "/dev/null",
2442 			    UIO_SYSSPACE, O_RDWR, 0);
2443 			if (error == 0) {
2444 				devnull = td->td_retval[0];
2445 				KASSERT(devnull == i, ("we didn't get our fd"));
2446 			}
2447 		}
2448 		td->td_retval[0] = save;
2449 		if (error != 0)
2450 			return (error);
2451 	}
2452 	return (0);
2453 }
2454 
2455 /*
2456  * Internal form of close.  Decrement reference count on file structure.
2457  * Note: td may be NULL when closing a file that was being passed in a
2458  * message.
2459  *
2460  * XXXRW: Giant is not required for the caller, but often will be held; this
2461  * makes it moderately likely the Giant will be recursed in the VFS case.
2462  */
2463 int
2464 closef(struct file *fp, struct thread *td)
2465 {
2466 	struct vnode *vp;
2467 	struct flock lf;
2468 	struct filedesc_to_leader *fdtol;
2469 	struct filedesc *fdp;
2470 
2471 	/*
2472 	 * POSIX record locking dictates that any close releases ALL
2473 	 * locks owned by this process.  This is handled by setting
2474 	 * a flag in the unlock to free ONLY locks obeying POSIX
2475 	 * semantics, and not to free BSD-style file locks.
2476 	 * If the descriptor was in a message, POSIX-style locks
2477 	 * aren't passed with the descriptor, and the thread pointer
2478 	 * will be NULL.  Callers should be careful only to pass a
2479 	 * NULL thread pointer when there really is no owning
2480 	 * context that might have locks, or the locks will be
2481 	 * leaked.
2482 	 */
2483 	if (fp->f_type == DTYPE_VNODE && td != NULL) {
2484 		vp = fp->f_vnode;
2485 		if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
2486 			lf.l_whence = SEEK_SET;
2487 			lf.l_start = 0;
2488 			lf.l_len = 0;
2489 			lf.l_type = F_UNLCK;
2490 			(void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader,
2491 			    F_UNLCK, &lf, F_POSIX);
2492 		}
2493 		fdtol = td->td_proc->p_fdtol;
2494 		if (fdtol != NULL) {
2495 			/*
2496 			 * Handle special case where file descriptor table is
2497 			 * shared between multiple process leaders.
2498 			 */
2499 			fdp = td->td_proc->p_fd;
2500 			FILEDESC_XLOCK(fdp);
2501 			for (fdtol = fdtol->fdl_next;
2502 			    fdtol != td->td_proc->p_fdtol;
2503 			    fdtol = fdtol->fdl_next) {
2504 				if ((fdtol->fdl_leader->p_flag &
2505 				    P_ADVLOCK) == 0)
2506 					continue;
2507 				fdtol->fdl_holdcount++;
2508 				FILEDESC_XUNLOCK(fdp);
2509 				lf.l_whence = SEEK_SET;
2510 				lf.l_start = 0;
2511 				lf.l_len = 0;
2512 				lf.l_type = F_UNLCK;
2513 				vp = fp->f_vnode;
2514 				(void) VOP_ADVLOCK(vp,
2515 				    (caddr_t)fdtol->fdl_leader, F_UNLCK, &lf,
2516 				    F_POSIX);
2517 				FILEDESC_XLOCK(fdp);
2518 				fdtol->fdl_holdcount--;
2519 				if (fdtol->fdl_holdcount == 0 &&
2520 				    fdtol->fdl_wakeup != 0) {
2521 					fdtol->fdl_wakeup = 0;
2522 					wakeup(fdtol);
2523 				}
2524 			}
2525 			FILEDESC_XUNLOCK(fdp);
2526 		}
2527 	}
2528 	return (fdrop(fp, td));
2529 }
2530 
2531 /*
2532  * Initialize the file pointer with the specified properties.
2533  *
2534  * The ops are set with release semantics to be certain that the flags, type,
2535  * and data are visible when ops is.  This is to prevent ops methods from being
2536  * called with bad data.
2537  */
2538 void
2539 finit(struct file *fp, u_int flag, short type, void *data, struct fileops *ops)
2540 {
2541 	fp->f_data = data;
2542 	fp->f_flag = flag;
2543 	fp->f_type = type;
2544 	atomic_store_rel_ptr((volatile uintptr_t *)&fp->f_ops, (uintptr_t)ops);
2545 }
2546 
2547 int
2548 fget_cap_locked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
2549     struct file **fpp, struct filecaps *havecapsp)
2550 {
2551 	struct filedescent *fde;
2552 	int error;
2553 
2554 	FILEDESC_LOCK_ASSERT(fdp);
2555 
2556 	fde = fdeget_locked(fdp, fd);
2557 	if (fde == NULL) {
2558 		error = EBADF;
2559 		goto out;
2560 	}
2561 
2562 #ifdef CAPABILITIES
2563 	error = cap_check(cap_rights_fde_inline(fde), needrightsp);
2564 	if (error != 0)
2565 		goto out;
2566 #endif
2567 
2568 	if (havecapsp != NULL)
2569 		filecaps_copy(&fde->fde_caps, havecapsp, true);
2570 
2571 	*fpp = fde->fde_file;
2572 
2573 	error = 0;
2574 out:
2575 	return (error);
2576 }
2577 
2578 int
2579 fget_cap(struct thread *td, int fd, cap_rights_t *needrightsp,
2580     struct file **fpp, struct filecaps *havecapsp)
2581 {
2582 	struct filedesc *fdp = td->td_proc->p_fd;
2583 	int error;
2584 #ifndef CAPABILITIES
2585 	error = fget_unlocked(fdp, fd, needrightsp, fpp, NULL);
2586 	if (error == 0 && havecapsp != NULL)
2587 		filecaps_fill(havecapsp);
2588 #else
2589 	struct file *fp;
2590 	seq_t seq;
2591 
2592 	for (;;) {
2593 		error = fget_unlocked(fdp, fd, needrightsp, &fp, &seq);
2594 		if (error != 0)
2595 			return (error);
2596 
2597 		if (havecapsp != NULL) {
2598 			if (!filecaps_copy(&fdp->fd_ofiles[fd].fde_caps,
2599 			    havecapsp, false)) {
2600 				fdrop(fp, td);
2601 				goto get_locked;
2602 			}
2603 		}
2604 
2605 		if (!fd_modified(fdp, fd, seq))
2606 			break;
2607 		fdrop(fp, td);
2608 	}
2609 
2610 	*fpp = fp;
2611 	return (0);
2612 
2613 get_locked:
2614 	FILEDESC_SLOCK(fdp);
2615 	error = fget_cap_locked(fdp, fd, needrightsp, fpp, havecapsp);
2616 	if (error == 0)
2617 		fhold(*fpp);
2618 	FILEDESC_SUNLOCK(fdp);
2619 #endif
2620 	return (error);
2621 }
2622 
2623 int
2624 fget_unlocked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
2625     struct file **fpp, seq_t *seqp)
2626 {
2627 #ifdef CAPABILITIES
2628 	const struct filedescent *fde;
2629 #endif
2630 	const struct fdescenttbl *fdt;
2631 	struct file *fp;
2632 	u_int count;
2633 #ifdef CAPABILITIES
2634 	seq_t seq;
2635 	cap_rights_t haverights;
2636 	int error;
2637 #endif
2638 
2639 	fdt = fdp->fd_files;
2640 	if ((u_int)fd >= fdt->fdt_nfiles)
2641 		return (EBADF);
2642 	/*
2643 	 * Fetch the descriptor locklessly.  We avoid fdrop() races by
2644 	 * never raising a refcount above 0.  To accomplish this we have
2645 	 * to use a cmpset loop rather than an atomic_add.  The descriptor
2646 	 * must be re-verified once we acquire a reference to be certain
2647 	 * that the identity is still correct and we did not lose a race
2648 	 * due to preemption.
2649 	 */
2650 	for (;;) {
2651 #ifdef CAPABILITIES
2652 		seq = seq_read(fd_seq(fdt, fd));
2653 		fde = &fdt->fdt_ofiles[fd];
2654 		haverights = *cap_rights_fde_inline(fde);
2655 		fp = fde->fde_file;
2656 		if (!seq_consistent(fd_seq(fdt, fd), seq))
2657 			continue;
2658 #else
2659 		fp = fdt->fdt_ofiles[fd].fde_file;
2660 #endif
2661 		if (fp == NULL)
2662 			return (EBADF);
2663 #ifdef CAPABILITIES
2664 		error = cap_check(&haverights, needrightsp);
2665 		if (error != 0)
2666 			return (error);
2667 #endif
2668 		count = fp->f_count;
2669 	retry:
2670 		if (count == 0) {
2671 			/*
2672 			 * Force a reload. Other thread could reallocate the
2673 			 * table before this fd was closed, so it possible that
2674 			 * there is a stale fp pointer in cached version.
2675 			 */
2676 			fdt = *(const struct fdescenttbl * const volatile *)&(fdp->fd_files);
2677 			continue;
2678 		}
2679 		/*
2680 		 * Use an acquire barrier to force re-reading of fdt so it is
2681 		 * refreshed for verification.
2682 		 */
2683 		if (atomic_fcmpset_acq_int(&fp->f_count, &count, count + 1) == 0)
2684 			goto retry;
2685 		fdt = fdp->fd_files;
2686 #ifdef	CAPABILITIES
2687 		if (seq_consistent_nomb(fd_seq(fdt, fd), seq))
2688 #else
2689 		if (fp == fdt->fdt_ofiles[fd].fde_file)
2690 #endif
2691 			break;
2692 		fdrop(fp, curthread);
2693 	}
2694 	*fpp = fp;
2695 	if (seqp != NULL) {
2696 #ifdef CAPABILITIES
2697 		*seqp = seq;
2698 #endif
2699 	}
2700 	return (0);
2701 }
2702 
2703 /*
2704  * Extract the file pointer associated with the specified descriptor for the
2705  * current user process.
2706  *
2707  * If the descriptor doesn't exist or doesn't match 'flags', EBADF is
2708  * returned.
2709  *
2710  * File's rights will be checked against the capability rights mask.
2711  *
2712  * If an error occurred the non-zero error is returned and *fpp is set to
2713  * NULL.  Otherwise *fpp is held and set and zero is returned.  Caller is
2714  * responsible for fdrop().
2715  */
2716 static __inline int
2717 _fget(struct thread *td, int fd, struct file **fpp, int flags,
2718     cap_rights_t *needrightsp, seq_t *seqp)
2719 {
2720 	struct filedesc *fdp;
2721 	struct file *fp;
2722 	int error;
2723 
2724 	*fpp = NULL;
2725 	fdp = td->td_proc->p_fd;
2726 	error = fget_unlocked(fdp, fd, needrightsp, &fp, seqp);
2727 	if (error != 0)
2728 		return (error);
2729 	if (fp->f_ops == &badfileops) {
2730 		fdrop(fp, td);
2731 		return (EBADF);
2732 	}
2733 
2734 	/*
2735 	 * FREAD and FWRITE failure return EBADF as per POSIX.
2736 	 */
2737 	error = 0;
2738 	switch (flags) {
2739 	case FREAD:
2740 	case FWRITE:
2741 		if ((fp->f_flag & flags) == 0)
2742 			error = EBADF;
2743 		break;
2744 	case FEXEC:
2745 	    	if ((fp->f_flag & (FREAD | FEXEC)) == 0 ||
2746 		    ((fp->f_flag & FWRITE) != 0))
2747 			error = EBADF;
2748 		break;
2749 	case 0:
2750 		break;
2751 	default:
2752 		KASSERT(0, ("wrong flags"));
2753 	}
2754 
2755 	if (error != 0) {
2756 		fdrop(fp, td);
2757 		return (error);
2758 	}
2759 
2760 	*fpp = fp;
2761 	return (0);
2762 }
2763 
2764 int
2765 fget(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
2766 {
2767 
2768 	return (_fget(td, fd, fpp, 0, rightsp, NULL));
2769 }
2770 
2771 int
2772 fget_mmap(struct thread *td, int fd, cap_rights_t *rightsp, u_char *maxprotp,
2773     struct file **fpp)
2774 {
2775 	int error;
2776 #ifndef CAPABILITIES
2777 	error = _fget(td, fd, fpp, 0, rightsp, NULL);
2778 	if (maxprotp != NULL)
2779 		*maxprotp = VM_PROT_ALL;
2780 #else
2781 	struct filedesc *fdp = td->td_proc->p_fd;
2782 	seq_t seq;
2783 
2784 	MPASS(cap_rights_is_set(rightsp, CAP_MMAP));
2785 	for (;;) {
2786 		error = _fget(td, fd, fpp, 0, rightsp, &seq);
2787 		if (error != 0)
2788 			return (error);
2789 		/*
2790 		 * If requested, convert capability rights to access flags.
2791 		 */
2792 		if (maxprotp != NULL)
2793 			*maxprotp = cap_rights_to_vmprot(cap_rights(fdp, fd));
2794 		if (!fd_modified(fdp, fd, seq))
2795 			break;
2796 		fdrop(*fpp, td);
2797 	}
2798 #endif
2799 	return (error);
2800 }
2801 
2802 int
2803 fget_read(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
2804 {
2805 
2806 	return (_fget(td, fd, fpp, FREAD, rightsp, NULL));
2807 }
2808 
2809 int
2810 fget_write(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
2811 {
2812 
2813 	return (_fget(td, fd, fpp, FWRITE, rightsp, NULL));
2814 }
2815 
2816 int
2817 fget_fcntl(struct thread *td, int fd, cap_rights_t *rightsp, int needfcntl,
2818     struct file **fpp)
2819 {
2820 	struct filedesc *fdp = td->td_proc->p_fd;
2821 #ifndef CAPABILITIES
2822 	return (fget_unlocked(fdp, fd, rightsp, fpp, NULL));
2823 #else
2824 	int error;
2825 	seq_t seq;
2826 
2827 	MPASS(cap_rights_is_set(rightsp, CAP_FCNTL));
2828 	for (;;) {
2829 		error = fget_unlocked(fdp, fd, rightsp, fpp, &seq);
2830 		if (error != 0)
2831 			return (error);
2832 		error = cap_fcntl_check(fdp, fd, needfcntl);
2833 		if (!fd_modified(fdp, fd, seq))
2834 			break;
2835 		fdrop(*fpp, td);
2836 	}
2837 	if (error != 0) {
2838 		fdrop(*fpp, td);
2839 		*fpp = NULL;
2840 	}
2841 	return (error);
2842 #endif
2843 }
2844 
2845 /*
2846  * Like fget() but loads the underlying vnode, or returns an error if the
2847  * descriptor does not represent a vnode.  Note that pipes use vnodes but
2848  * never have VM objects.  The returned vnode will be vref()'d.
2849  *
2850  * XXX: what about the unused flags ?
2851  */
2852 static __inline int
2853 _fgetvp(struct thread *td, int fd, int flags, cap_rights_t *needrightsp,
2854     struct vnode **vpp)
2855 {
2856 	struct file *fp;
2857 	int error;
2858 
2859 	*vpp = NULL;
2860 	error = _fget(td, fd, &fp, flags, needrightsp, NULL);
2861 	if (error != 0)
2862 		return (error);
2863 	if (fp->f_vnode == NULL) {
2864 		error = EINVAL;
2865 	} else {
2866 		*vpp = fp->f_vnode;
2867 		vrefact(*vpp);
2868 	}
2869 	fdrop(fp, td);
2870 
2871 	return (error);
2872 }
2873 
2874 int
2875 fgetvp(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
2876 {
2877 
2878 	return (_fgetvp(td, fd, 0, rightsp, vpp));
2879 }
2880 
2881 int
2882 fgetvp_rights(struct thread *td, int fd, cap_rights_t *needrightsp,
2883     struct filecaps *havecaps, struct vnode **vpp)
2884 {
2885 	struct filedesc *fdp;
2886 	struct filecaps caps;
2887 	struct file *fp;
2888 	int error;
2889 
2890 	fdp = td->td_proc->p_fd;
2891 	error = fget_cap_locked(fdp, fd, needrightsp, &fp, &caps);
2892 	if (error != 0)
2893 		return (error);
2894 	if (fp->f_ops == &badfileops) {
2895 		error = EBADF;
2896 		goto out;
2897 	}
2898 	if (fp->f_vnode == NULL) {
2899 		error = EINVAL;
2900 		goto out;
2901 	}
2902 
2903 	*havecaps = caps;
2904 	*vpp = fp->f_vnode;
2905 	vrefact(*vpp);
2906 
2907 	return (0);
2908 out:
2909 	filecaps_free(&caps);
2910 	return (error);
2911 }
2912 
2913 int
2914 fgetvp_read(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
2915 {
2916 
2917 	return (_fgetvp(td, fd, FREAD, rightsp, vpp));
2918 }
2919 
2920 int
2921 fgetvp_exec(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
2922 {
2923 
2924 	return (_fgetvp(td, fd, FEXEC, rightsp, vpp));
2925 }
2926 
2927 #ifdef notyet
2928 int
2929 fgetvp_write(struct thread *td, int fd, cap_rights_t *rightsp,
2930     struct vnode **vpp)
2931 {
2932 
2933 	return (_fgetvp(td, fd, FWRITE, rightsp, vpp));
2934 }
2935 #endif
2936 
2937 /*
2938  * Handle the last reference to a file being closed.
2939  *
2940  * Without the noinline attribute clang keeps inlining the func thorough this
2941  * file when fdrop is used.
2942  */
2943 int __noinline
2944 _fdrop(struct file *fp, struct thread *td)
2945 {
2946 	int error;
2947 
2948 	if (fp->f_count != 0)
2949 		panic("fdrop: count %d", fp->f_count);
2950 	error = fo_close(fp, td);
2951 	atomic_subtract_int(&openfiles, 1);
2952 	crfree(fp->f_cred);
2953 	free(fp->f_advice, M_FADVISE);
2954 	uma_zfree(file_zone, fp);
2955 
2956 	return (error);
2957 }
2958 
2959 /*
2960  * Apply an advisory lock on a file descriptor.
2961  *
2962  * Just attempt to get a record lock of the requested type on the entire file
2963  * (l_whence = SEEK_SET, l_start = 0, l_len = 0).
2964  */
2965 #ifndef _SYS_SYSPROTO_H_
2966 struct flock_args {
2967 	int	fd;
2968 	int	how;
2969 };
2970 #endif
2971 /* ARGSUSED */
2972 int
2973 sys_flock(struct thread *td, struct flock_args *uap)
2974 {
2975 	struct file *fp;
2976 	struct vnode *vp;
2977 	struct flock lf;
2978 	int error;
2979 
2980 	error = fget(td, uap->fd, &cap_flock_rights, &fp);
2981 	if (error != 0)
2982 		return (error);
2983 	if (fp->f_type != DTYPE_VNODE) {
2984 		fdrop(fp, td);
2985 		return (EOPNOTSUPP);
2986 	}
2987 
2988 	vp = fp->f_vnode;
2989 	lf.l_whence = SEEK_SET;
2990 	lf.l_start = 0;
2991 	lf.l_len = 0;
2992 	if (uap->how & LOCK_UN) {
2993 		lf.l_type = F_UNLCK;
2994 		atomic_clear_int(&fp->f_flag, FHASLOCK);
2995 		error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK);
2996 		goto done2;
2997 	}
2998 	if (uap->how & LOCK_EX)
2999 		lf.l_type = F_WRLCK;
3000 	else if (uap->how & LOCK_SH)
3001 		lf.l_type = F_RDLCK;
3002 	else {
3003 		error = EBADF;
3004 		goto done2;
3005 	}
3006 	atomic_set_int(&fp->f_flag, FHASLOCK);
3007 	error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf,
3008 	    (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT);
3009 done2:
3010 	fdrop(fp, td);
3011 	return (error);
3012 }
3013 /*
3014  * Duplicate the specified descriptor to a free descriptor.
3015  */
3016 int
3017 dupfdopen(struct thread *td, struct filedesc *fdp, int dfd, int mode,
3018     int openerror, int *indxp)
3019 {
3020 	struct filedescent *newfde, *oldfde;
3021 	struct file *fp;
3022 	u_long *ioctls;
3023 	int error, indx;
3024 
3025 	KASSERT(openerror == ENODEV || openerror == ENXIO,
3026 	    ("unexpected error %d in %s", openerror, __func__));
3027 
3028 	/*
3029 	 * If the to-be-dup'd fd number is greater than the allowed number
3030 	 * of file descriptors, or the fd to be dup'd has already been
3031 	 * closed, then reject.
3032 	 */
3033 	FILEDESC_XLOCK(fdp);
3034 	if ((fp = fget_locked(fdp, dfd)) == NULL) {
3035 		FILEDESC_XUNLOCK(fdp);
3036 		return (EBADF);
3037 	}
3038 
3039 	error = fdalloc(td, 0, &indx);
3040 	if (error != 0) {
3041 		FILEDESC_XUNLOCK(fdp);
3042 		return (error);
3043 	}
3044 
3045 	/*
3046 	 * There are two cases of interest here.
3047 	 *
3048 	 * For ENODEV simply dup (dfd) to file descriptor (indx) and return.
3049 	 *
3050 	 * For ENXIO steal away the file structure from (dfd) and store it in
3051 	 * (indx).  (dfd) is effectively closed by this operation.
3052 	 */
3053 	switch (openerror) {
3054 	case ENODEV:
3055 		/*
3056 		 * Check that the mode the file is being opened for is a
3057 		 * subset of the mode of the existing descriptor.
3058 		 */
3059 		if (((mode & (FREAD|FWRITE)) | fp->f_flag) != fp->f_flag) {
3060 			fdunused(fdp, indx);
3061 			FILEDESC_XUNLOCK(fdp);
3062 			return (EACCES);
3063 		}
3064 		fhold(fp);
3065 		newfde = &fdp->fd_ofiles[indx];
3066 		oldfde = &fdp->fd_ofiles[dfd];
3067 		ioctls = filecaps_copy_prep(&oldfde->fde_caps);
3068 #ifdef CAPABILITIES
3069 		seq_write_begin(&newfde->fde_seq);
3070 #endif
3071 		memcpy(newfde, oldfde, fde_change_size);
3072 		filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps,
3073 		    ioctls);
3074 #ifdef CAPABILITIES
3075 		seq_write_end(&newfde->fde_seq);
3076 #endif
3077 		break;
3078 	case ENXIO:
3079 		/*
3080 		 * Steal away the file pointer from dfd and stuff it into indx.
3081 		 */
3082 		newfde = &fdp->fd_ofiles[indx];
3083 		oldfde = &fdp->fd_ofiles[dfd];
3084 #ifdef CAPABILITIES
3085 		seq_write_begin(&newfde->fde_seq);
3086 #endif
3087 		memcpy(newfde, oldfde, fde_change_size);
3088 		oldfde->fde_file = NULL;
3089 		fdunused(fdp, dfd);
3090 #ifdef CAPABILITIES
3091 		seq_write_end(&newfde->fde_seq);
3092 #endif
3093 		break;
3094 	}
3095 	FILEDESC_XUNLOCK(fdp);
3096 	*indxp = indx;
3097 	return (0);
3098 }
3099 
3100 /*
3101  * This sysctl determines if we will allow a process to chroot(2) if it
3102  * has a directory open:
3103  *	0: disallowed for all processes.
3104  *	1: allowed for processes that were not already chroot(2)'ed.
3105  *	2: allowed for all processes.
3106  */
3107 
3108 static int chroot_allow_open_directories = 1;
3109 
3110 SYSCTL_INT(_kern, OID_AUTO, chroot_allow_open_directories, CTLFLAG_RW,
3111     &chroot_allow_open_directories, 0,
3112     "Allow a process to chroot(2) if it has a directory open");
3113 
3114 /*
3115  * Helper function for raised chroot(2) security function:  Refuse if
3116  * any filedescriptors are open directories.
3117  */
3118 static int
3119 chroot_refuse_vdir_fds(struct filedesc *fdp)
3120 {
3121 	struct vnode *vp;
3122 	struct file *fp;
3123 	int fd;
3124 
3125 	FILEDESC_LOCK_ASSERT(fdp);
3126 
3127 	for (fd = 0; fd <= fdp->fd_lastfile; fd++) {
3128 		fp = fget_locked(fdp, fd);
3129 		if (fp == NULL)
3130 			continue;
3131 		if (fp->f_type == DTYPE_VNODE) {
3132 			vp = fp->f_vnode;
3133 			if (vp->v_type == VDIR)
3134 				return (EPERM);
3135 		}
3136 	}
3137 	return (0);
3138 }
3139 
3140 /*
3141  * Common routine for kern_chroot() and jail_attach().  The caller is
3142  * responsible for invoking priv_check() and mac_vnode_check_chroot() to
3143  * authorize this operation.
3144  */
3145 int
3146 pwd_chroot(struct thread *td, struct vnode *vp)
3147 {
3148 	struct filedesc *fdp;
3149 	struct vnode *oldvp;
3150 	int error;
3151 
3152 	fdp = td->td_proc->p_fd;
3153 	FILEDESC_XLOCK(fdp);
3154 	if (chroot_allow_open_directories == 0 ||
3155 	    (chroot_allow_open_directories == 1 && fdp->fd_rdir != rootvnode)) {
3156 		error = chroot_refuse_vdir_fds(fdp);
3157 		if (error != 0) {
3158 			FILEDESC_XUNLOCK(fdp);
3159 			return (error);
3160 		}
3161 	}
3162 	oldvp = fdp->fd_rdir;
3163 	vrefact(vp);
3164 	fdp->fd_rdir = vp;
3165 	if (fdp->fd_jdir == NULL) {
3166 		vrefact(vp);
3167 		fdp->fd_jdir = vp;
3168 	}
3169 	FILEDESC_XUNLOCK(fdp);
3170 	vrele(oldvp);
3171 	return (0);
3172 }
3173 
3174 void
3175 pwd_chdir(struct thread *td, struct vnode *vp)
3176 {
3177 	struct filedesc *fdp;
3178 	struct vnode *oldvp;
3179 
3180 	fdp = td->td_proc->p_fd;
3181 	FILEDESC_XLOCK(fdp);
3182 	VNASSERT(vp->v_usecount > 0, vp,
3183 	    ("chdir to a vnode with zero usecount"));
3184 	oldvp = fdp->fd_cdir;
3185 	fdp->fd_cdir = vp;
3186 	FILEDESC_XUNLOCK(fdp);
3187 	vrele(oldvp);
3188 }
3189 
3190 /*
3191  * Scan all active processes and prisons to see if any of them have a current
3192  * or root directory of `olddp'. If so, replace them with the new mount point.
3193  */
3194 void
3195 mountcheckdirs(struct vnode *olddp, struct vnode *newdp)
3196 {
3197 	struct filedesc *fdp;
3198 	struct prison *pr;
3199 	struct proc *p;
3200 	int nrele;
3201 
3202 	if (vrefcnt(olddp) == 1)
3203 		return;
3204 	nrele = 0;
3205 	sx_slock(&allproc_lock);
3206 	FOREACH_PROC_IN_SYSTEM(p) {
3207 		PROC_LOCK(p);
3208 		fdp = fdhold(p);
3209 		PROC_UNLOCK(p);
3210 		if (fdp == NULL)
3211 			continue;
3212 		FILEDESC_XLOCK(fdp);
3213 		if (fdp->fd_cdir == olddp) {
3214 			vrefact(newdp);
3215 			fdp->fd_cdir = newdp;
3216 			nrele++;
3217 		}
3218 		if (fdp->fd_rdir == olddp) {
3219 			vrefact(newdp);
3220 			fdp->fd_rdir = newdp;
3221 			nrele++;
3222 		}
3223 		if (fdp->fd_jdir == olddp) {
3224 			vrefact(newdp);
3225 			fdp->fd_jdir = newdp;
3226 			nrele++;
3227 		}
3228 		FILEDESC_XUNLOCK(fdp);
3229 		fddrop(fdp);
3230 	}
3231 	sx_sunlock(&allproc_lock);
3232 	if (rootvnode == olddp) {
3233 		vrefact(newdp);
3234 		rootvnode = newdp;
3235 		nrele++;
3236 	}
3237 	mtx_lock(&prison0.pr_mtx);
3238 	if (prison0.pr_root == olddp) {
3239 		vrefact(newdp);
3240 		prison0.pr_root = newdp;
3241 		nrele++;
3242 	}
3243 	mtx_unlock(&prison0.pr_mtx);
3244 	sx_slock(&allprison_lock);
3245 	TAILQ_FOREACH(pr, &allprison, pr_list) {
3246 		mtx_lock(&pr->pr_mtx);
3247 		if (pr->pr_root == olddp) {
3248 			vrefact(newdp);
3249 			pr->pr_root = newdp;
3250 			nrele++;
3251 		}
3252 		mtx_unlock(&pr->pr_mtx);
3253 	}
3254 	sx_sunlock(&allprison_lock);
3255 	while (nrele--)
3256 		vrele(olddp);
3257 }
3258 
3259 struct filedesc_to_leader *
3260 filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader)
3261 {
3262 	struct filedesc_to_leader *fdtol;
3263 
3264 	fdtol = malloc(sizeof(struct filedesc_to_leader),
3265 	    M_FILEDESC_TO_LEADER, M_WAITOK);
3266 	fdtol->fdl_refcount = 1;
3267 	fdtol->fdl_holdcount = 0;
3268 	fdtol->fdl_wakeup = 0;
3269 	fdtol->fdl_leader = leader;
3270 	if (old != NULL) {
3271 		FILEDESC_XLOCK(fdp);
3272 		fdtol->fdl_next = old->fdl_next;
3273 		fdtol->fdl_prev = old;
3274 		old->fdl_next = fdtol;
3275 		fdtol->fdl_next->fdl_prev = fdtol;
3276 		FILEDESC_XUNLOCK(fdp);
3277 	} else {
3278 		fdtol->fdl_next = fdtol;
3279 		fdtol->fdl_prev = fdtol;
3280 	}
3281 	return (fdtol);
3282 }
3283 
3284 static int
3285 sysctl_kern_proc_nfds(SYSCTL_HANDLER_ARGS)
3286 {
3287 	struct filedesc *fdp;
3288 	int i, count, slots;
3289 
3290 	if (*(int *)arg1 != 0)
3291 		return (EINVAL);
3292 
3293 	fdp = curproc->p_fd;
3294 	count = 0;
3295 	FILEDESC_SLOCK(fdp);
3296 	slots = NDSLOTS(fdp->fd_lastfile + 1);
3297 	for (i = 0; i < slots; i++)
3298 		count += bitcountl(fdp->fd_map[i]);
3299 	FILEDESC_SUNLOCK(fdp);
3300 
3301 	return (SYSCTL_OUT(req, &count, sizeof(count)));
3302 }
3303 
3304 static SYSCTL_NODE(_kern_proc, KERN_PROC_NFDS, nfds,
3305     CTLFLAG_RD|CTLFLAG_CAPRD|CTLFLAG_MPSAFE, sysctl_kern_proc_nfds,
3306     "Number of open file descriptors");
3307 
3308 /*
3309  * Get file structures globally.
3310  */
3311 static int
3312 sysctl_kern_file(SYSCTL_HANDLER_ARGS)
3313 {
3314 	struct xfile xf;
3315 	struct filedesc *fdp;
3316 	struct file *fp;
3317 	struct proc *p;
3318 	int error, n;
3319 
3320 	error = sysctl_wire_old_buffer(req, 0);
3321 	if (error != 0)
3322 		return (error);
3323 	if (req->oldptr == NULL) {
3324 		n = 0;
3325 		sx_slock(&allproc_lock);
3326 		FOREACH_PROC_IN_SYSTEM(p) {
3327 			PROC_LOCK(p);
3328 			if (p->p_state == PRS_NEW) {
3329 				PROC_UNLOCK(p);
3330 				continue;
3331 			}
3332 			fdp = fdhold(p);
3333 			PROC_UNLOCK(p);
3334 			if (fdp == NULL)
3335 				continue;
3336 			/* overestimates sparse tables. */
3337 			if (fdp->fd_lastfile > 0)
3338 				n += fdp->fd_lastfile;
3339 			fddrop(fdp);
3340 		}
3341 		sx_sunlock(&allproc_lock);
3342 		return (SYSCTL_OUT(req, 0, n * sizeof(xf)));
3343 	}
3344 	error = 0;
3345 	bzero(&xf, sizeof(xf));
3346 	xf.xf_size = sizeof(xf);
3347 	sx_slock(&allproc_lock);
3348 	FOREACH_PROC_IN_SYSTEM(p) {
3349 		PROC_LOCK(p);
3350 		if (p->p_state == PRS_NEW) {
3351 			PROC_UNLOCK(p);
3352 			continue;
3353 		}
3354 		if (p_cansee(req->td, p) != 0) {
3355 			PROC_UNLOCK(p);
3356 			continue;
3357 		}
3358 		xf.xf_pid = p->p_pid;
3359 		xf.xf_uid = p->p_ucred->cr_uid;
3360 		fdp = fdhold(p);
3361 		PROC_UNLOCK(p);
3362 		if (fdp == NULL)
3363 			continue;
3364 		FILEDESC_SLOCK(fdp);
3365 		for (n = 0; fdp->fd_refcnt > 0 && n <= fdp->fd_lastfile; ++n) {
3366 			if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
3367 				continue;
3368 			xf.xf_fd = n;
3369 			xf.xf_file = (uintptr_t)fp;
3370 			xf.xf_data = (uintptr_t)fp->f_data;
3371 			xf.xf_vnode = (uintptr_t)fp->f_vnode;
3372 			xf.xf_type = (uintptr_t)fp->f_type;
3373 			xf.xf_count = fp->f_count;
3374 			xf.xf_msgcount = 0;
3375 			xf.xf_offset = foffset_get(fp);
3376 			xf.xf_flag = fp->f_flag;
3377 			error = SYSCTL_OUT(req, &xf, sizeof(xf));
3378 			if (error)
3379 				break;
3380 		}
3381 		FILEDESC_SUNLOCK(fdp);
3382 		fddrop(fdp);
3383 		if (error)
3384 			break;
3385 	}
3386 	sx_sunlock(&allproc_lock);
3387 	return (error);
3388 }
3389 
3390 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD|CTLFLAG_MPSAFE,
3391     0, 0, sysctl_kern_file, "S,xfile", "Entire file table");
3392 
3393 #ifdef KINFO_FILE_SIZE
3394 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
3395 #endif
3396 
3397 static int
3398 xlate_fflags(int fflags)
3399 {
3400 	static const struct {
3401 		int	fflag;
3402 		int	kf_fflag;
3403 	} fflags_table[] = {
3404 		{ FAPPEND, KF_FLAG_APPEND },
3405 		{ FASYNC, KF_FLAG_ASYNC },
3406 		{ FFSYNC, KF_FLAG_FSYNC },
3407 		{ FHASLOCK, KF_FLAG_HASLOCK },
3408 		{ FNONBLOCK, KF_FLAG_NONBLOCK },
3409 		{ FREAD, KF_FLAG_READ },
3410 		{ FWRITE, KF_FLAG_WRITE },
3411 		{ O_CREAT, KF_FLAG_CREAT },
3412 		{ O_DIRECT, KF_FLAG_DIRECT },
3413 		{ O_EXCL, KF_FLAG_EXCL },
3414 		{ O_EXEC, KF_FLAG_EXEC },
3415 		{ O_EXLOCK, KF_FLAG_EXLOCK },
3416 		{ O_NOFOLLOW, KF_FLAG_NOFOLLOW },
3417 		{ O_SHLOCK, KF_FLAG_SHLOCK },
3418 		{ O_TRUNC, KF_FLAG_TRUNC }
3419 	};
3420 	unsigned int i;
3421 	int kflags;
3422 
3423 	kflags = 0;
3424 	for (i = 0; i < nitems(fflags_table); i++)
3425 		if (fflags & fflags_table[i].fflag)
3426 			kflags |=  fflags_table[i].kf_fflag;
3427 	return (kflags);
3428 }
3429 
3430 /* Trim unused data from kf_path by truncating the structure size. */
3431 static void
3432 pack_kinfo(struct kinfo_file *kif)
3433 {
3434 
3435 	kif->kf_structsize = offsetof(struct kinfo_file, kf_path) +
3436 	    strlen(kif->kf_path) + 1;
3437 	kif->kf_structsize = roundup(kif->kf_structsize, sizeof(uint64_t));
3438 }
3439 
3440 static void
3441 export_file_to_kinfo(struct file *fp, int fd, cap_rights_t *rightsp,
3442     struct kinfo_file *kif, struct filedesc *fdp, int flags)
3443 {
3444 	int error;
3445 
3446 	bzero(kif, sizeof(*kif));
3447 
3448 	/* Set a default type to allow for empty fill_kinfo() methods. */
3449 	kif->kf_type = KF_TYPE_UNKNOWN;
3450 	kif->kf_flags = xlate_fflags(fp->f_flag);
3451 	if (rightsp != NULL)
3452 		kif->kf_cap_rights = *rightsp;
3453 	else
3454 		cap_rights_init(&kif->kf_cap_rights);
3455 	kif->kf_fd = fd;
3456 	kif->kf_ref_count = fp->f_count;
3457 	kif->kf_offset = foffset_get(fp);
3458 
3459 	/*
3460 	 * This may drop the filedesc lock, so the 'fp' cannot be
3461 	 * accessed after this call.
3462 	 */
3463 	error = fo_fill_kinfo(fp, kif, fdp);
3464 	if (error == 0)
3465 		kif->kf_status |= KF_ATTR_VALID;
3466 	if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
3467 		pack_kinfo(kif);
3468 	else
3469 		kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
3470 }
3471 
3472 static void
3473 export_vnode_to_kinfo(struct vnode *vp, int fd, int fflags,
3474     struct kinfo_file *kif, int flags)
3475 {
3476 	int error;
3477 
3478 	bzero(kif, sizeof(*kif));
3479 
3480 	kif->kf_type = KF_TYPE_VNODE;
3481 	error = vn_fill_kinfo_vnode(vp, kif);
3482 	if (error == 0)
3483 		kif->kf_status |= KF_ATTR_VALID;
3484 	kif->kf_flags = xlate_fflags(fflags);
3485 	cap_rights_init(&kif->kf_cap_rights);
3486 	kif->kf_fd = fd;
3487 	kif->kf_ref_count = -1;
3488 	kif->kf_offset = -1;
3489 	if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
3490 		pack_kinfo(kif);
3491 	else
3492 		kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
3493 	vrele(vp);
3494 }
3495 
3496 struct export_fd_buf {
3497 	struct filedesc		*fdp;
3498 	struct sbuf 		*sb;
3499 	ssize_t			remainder;
3500 	struct kinfo_file	kif;
3501 	int			flags;
3502 };
3503 
3504 static int
3505 export_kinfo_to_sb(struct export_fd_buf *efbuf)
3506 {
3507 	struct kinfo_file *kif;
3508 
3509 	kif = &efbuf->kif;
3510 	if (efbuf->remainder != -1) {
3511 		if (efbuf->remainder < kif->kf_structsize) {
3512 			/* Terminate export. */
3513 			efbuf->remainder = 0;
3514 			return (0);
3515 		}
3516 		efbuf->remainder -= kif->kf_structsize;
3517 	}
3518 	return (sbuf_bcat(efbuf->sb, kif, kif->kf_structsize) == 0 ? 0 : ENOMEM);
3519 }
3520 
3521 static int
3522 export_file_to_sb(struct file *fp, int fd, cap_rights_t *rightsp,
3523     struct export_fd_buf *efbuf)
3524 {
3525 	int error;
3526 
3527 	if (efbuf->remainder == 0)
3528 		return (0);
3529 	export_file_to_kinfo(fp, fd, rightsp, &efbuf->kif, efbuf->fdp,
3530 	    efbuf->flags);
3531 	FILEDESC_SUNLOCK(efbuf->fdp);
3532 	error = export_kinfo_to_sb(efbuf);
3533 	FILEDESC_SLOCK(efbuf->fdp);
3534 	return (error);
3535 }
3536 
3537 static int
3538 export_vnode_to_sb(struct vnode *vp, int fd, int fflags,
3539     struct export_fd_buf *efbuf)
3540 {
3541 	int error;
3542 
3543 	if (efbuf->remainder == 0)
3544 		return (0);
3545 	if (efbuf->fdp != NULL)
3546 		FILEDESC_SUNLOCK(efbuf->fdp);
3547 	export_vnode_to_kinfo(vp, fd, fflags, &efbuf->kif, efbuf->flags);
3548 	error = export_kinfo_to_sb(efbuf);
3549 	if (efbuf->fdp != NULL)
3550 		FILEDESC_SLOCK(efbuf->fdp);
3551 	return (error);
3552 }
3553 
3554 /*
3555  * Store a process file descriptor information to sbuf.
3556  *
3557  * Takes a locked proc as argument, and returns with the proc unlocked.
3558  */
3559 int
3560 kern_proc_filedesc_out(struct proc *p,  struct sbuf *sb, ssize_t maxlen,
3561     int flags)
3562 {
3563 	struct file *fp;
3564 	struct filedesc *fdp;
3565 	struct export_fd_buf *efbuf;
3566 	struct vnode *cttyvp, *textvp, *tracevp;
3567 	int error, i;
3568 	cap_rights_t rights;
3569 
3570 	PROC_LOCK_ASSERT(p, MA_OWNED);
3571 
3572 	/* ktrace vnode */
3573 	tracevp = p->p_tracevp;
3574 	if (tracevp != NULL)
3575 		vrefact(tracevp);
3576 	/* text vnode */
3577 	textvp = p->p_textvp;
3578 	if (textvp != NULL)
3579 		vrefact(textvp);
3580 	/* Controlling tty. */
3581 	cttyvp = NULL;
3582 	if (p->p_pgrp != NULL && p->p_pgrp->pg_session != NULL) {
3583 		cttyvp = p->p_pgrp->pg_session->s_ttyvp;
3584 		if (cttyvp != NULL)
3585 			vrefact(cttyvp);
3586 	}
3587 	fdp = fdhold(p);
3588 	PROC_UNLOCK(p);
3589 	efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
3590 	efbuf->fdp = NULL;
3591 	efbuf->sb = sb;
3592 	efbuf->remainder = maxlen;
3593 	efbuf->flags = flags;
3594 	if (tracevp != NULL)
3595 		export_vnode_to_sb(tracevp, KF_FD_TYPE_TRACE, FREAD | FWRITE,
3596 		    efbuf);
3597 	if (textvp != NULL)
3598 		export_vnode_to_sb(textvp, KF_FD_TYPE_TEXT, FREAD, efbuf);
3599 	if (cttyvp != NULL)
3600 		export_vnode_to_sb(cttyvp, KF_FD_TYPE_CTTY, FREAD | FWRITE,
3601 		    efbuf);
3602 	error = 0;
3603 	if (fdp == NULL)
3604 		goto fail;
3605 	efbuf->fdp = fdp;
3606 	FILEDESC_SLOCK(fdp);
3607 	/* working directory */
3608 	if (fdp->fd_cdir != NULL) {
3609 		vrefact(fdp->fd_cdir);
3610 		export_vnode_to_sb(fdp->fd_cdir, KF_FD_TYPE_CWD, FREAD, efbuf);
3611 	}
3612 	/* root directory */
3613 	if (fdp->fd_rdir != NULL) {
3614 		vrefact(fdp->fd_rdir);
3615 		export_vnode_to_sb(fdp->fd_rdir, KF_FD_TYPE_ROOT, FREAD, efbuf);
3616 	}
3617 	/* jail directory */
3618 	if (fdp->fd_jdir != NULL) {
3619 		vrefact(fdp->fd_jdir);
3620 		export_vnode_to_sb(fdp->fd_jdir, KF_FD_TYPE_JAIL, FREAD, efbuf);
3621 	}
3622 	for (i = 0; fdp->fd_refcnt > 0 && i <= fdp->fd_lastfile; i++) {
3623 		if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
3624 			continue;
3625 #ifdef CAPABILITIES
3626 		rights = *cap_rights(fdp, i);
3627 #else /* !CAPABILITIES */
3628 		rights = cap_no_rights;
3629 #endif
3630 		/*
3631 		 * Create sysctl entry.  It is OK to drop the filedesc
3632 		 * lock inside of export_file_to_sb() as we will
3633 		 * re-validate and re-evaluate its properties when the
3634 		 * loop continues.
3635 		 */
3636 		error = export_file_to_sb(fp, i, &rights, efbuf);
3637 		if (error != 0 || efbuf->remainder == 0)
3638 			break;
3639 	}
3640 	FILEDESC_SUNLOCK(fdp);
3641 	fddrop(fdp);
3642 fail:
3643 	free(efbuf, M_TEMP);
3644 	return (error);
3645 }
3646 
3647 #define FILEDESC_SBUF_SIZE	(sizeof(struct kinfo_file) * 5)
3648 
3649 /*
3650  * Get per-process file descriptors for use by procstat(1), et al.
3651  */
3652 static int
3653 sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS)
3654 {
3655 	struct sbuf sb;
3656 	struct proc *p;
3657 	ssize_t maxlen;
3658 	int error, error2, *name;
3659 
3660 	name = (int *)arg1;
3661 
3662 	sbuf_new_for_sysctl(&sb, NULL, FILEDESC_SBUF_SIZE, req);
3663 	sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
3664 	error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
3665 	if (error != 0) {
3666 		sbuf_delete(&sb);
3667 		return (error);
3668 	}
3669 	maxlen = req->oldptr != NULL ? req->oldlen : -1;
3670 	error = kern_proc_filedesc_out(p, &sb, maxlen,
3671 	    KERN_FILEDESC_PACK_KINFO);
3672 	error2 = sbuf_finish(&sb);
3673 	sbuf_delete(&sb);
3674 	return (error != 0 ? error : error2);
3675 }
3676 
3677 #ifdef COMPAT_FREEBSD7
3678 #ifdef KINFO_OFILE_SIZE
3679 CTASSERT(sizeof(struct kinfo_ofile) == KINFO_OFILE_SIZE);
3680 #endif
3681 
3682 static void
3683 kinfo_to_okinfo(struct kinfo_file *kif, struct kinfo_ofile *okif)
3684 {
3685 
3686 	okif->kf_structsize = sizeof(*okif);
3687 	okif->kf_type = kif->kf_type;
3688 	okif->kf_fd = kif->kf_fd;
3689 	okif->kf_ref_count = kif->kf_ref_count;
3690 	okif->kf_flags = kif->kf_flags & (KF_FLAG_READ | KF_FLAG_WRITE |
3691 	    KF_FLAG_APPEND | KF_FLAG_ASYNC | KF_FLAG_FSYNC | KF_FLAG_NONBLOCK |
3692 	    KF_FLAG_DIRECT | KF_FLAG_HASLOCK);
3693 	okif->kf_offset = kif->kf_offset;
3694 	if (kif->kf_type == KF_TYPE_VNODE)
3695 		okif->kf_vnode_type = kif->kf_un.kf_file.kf_file_type;
3696 	else
3697 		okif->kf_vnode_type = KF_VTYPE_VNON;
3698 	strlcpy(okif->kf_path, kif->kf_path, sizeof(okif->kf_path));
3699 	if (kif->kf_type == KF_TYPE_SOCKET) {
3700 		okif->kf_sock_domain = kif->kf_un.kf_sock.kf_sock_domain0;
3701 		okif->kf_sock_type = kif->kf_un.kf_sock.kf_sock_type0;
3702 		okif->kf_sock_protocol = kif->kf_un.kf_sock.kf_sock_protocol0;
3703 		okif->kf_sa_local = kif->kf_un.kf_sock.kf_sa_local;
3704 		okif->kf_sa_peer = kif->kf_un.kf_sock.kf_sa_peer;
3705 	} else {
3706 		okif->kf_sa_local.ss_family = AF_UNSPEC;
3707 		okif->kf_sa_peer.ss_family = AF_UNSPEC;
3708 	}
3709 }
3710 
3711 static int
3712 export_vnode_for_osysctl(struct vnode *vp, int type, struct kinfo_file *kif,
3713     struct kinfo_ofile *okif, struct filedesc *fdp, struct sysctl_req *req)
3714 {
3715 	int error;
3716 
3717 	vrefact(vp);
3718 	FILEDESC_SUNLOCK(fdp);
3719 	export_vnode_to_kinfo(vp, type, 0, kif, KERN_FILEDESC_PACK_KINFO);
3720 	kinfo_to_okinfo(kif, okif);
3721 	error = SYSCTL_OUT(req, okif, sizeof(*okif));
3722 	FILEDESC_SLOCK(fdp);
3723 	return (error);
3724 }
3725 
3726 /*
3727  * Get per-process file descriptors for use by procstat(1), et al.
3728  */
3729 static int
3730 sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS)
3731 {
3732 	struct kinfo_ofile *okif;
3733 	struct kinfo_file *kif;
3734 	struct filedesc *fdp;
3735 	int error, i, *name;
3736 	struct file *fp;
3737 	struct proc *p;
3738 
3739 	name = (int *)arg1;
3740 	error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
3741 	if (error != 0)
3742 		return (error);
3743 	fdp = fdhold(p);
3744 	PROC_UNLOCK(p);
3745 	if (fdp == NULL)
3746 		return (ENOENT);
3747 	kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK);
3748 	okif = malloc(sizeof(*okif), M_TEMP, M_WAITOK);
3749 	FILEDESC_SLOCK(fdp);
3750 	if (fdp->fd_cdir != NULL)
3751 		export_vnode_for_osysctl(fdp->fd_cdir, KF_FD_TYPE_CWD, kif,
3752 		    okif, fdp, req);
3753 	if (fdp->fd_rdir != NULL)
3754 		export_vnode_for_osysctl(fdp->fd_rdir, KF_FD_TYPE_ROOT, kif,
3755 		    okif, fdp, req);
3756 	if (fdp->fd_jdir != NULL)
3757 		export_vnode_for_osysctl(fdp->fd_jdir, KF_FD_TYPE_JAIL, kif,
3758 		    okif, fdp, req);
3759 	for (i = 0; fdp->fd_refcnt > 0 && i <= fdp->fd_lastfile; i++) {
3760 		if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
3761 			continue;
3762 		export_file_to_kinfo(fp, i, NULL, kif, fdp,
3763 		    KERN_FILEDESC_PACK_KINFO);
3764 		FILEDESC_SUNLOCK(fdp);
3765 		kinfo_to_okinfo(kif, okif);
3766 		error = SYSCTL_OUT(req, okif, sizeof(*okif));
3767 		FILEDESC_SLOCK(fdp);
3768 		if (error)
3769 			break;
3770 	}
3771 	FILEDESC_SUNLOCK(fdp);
3772 	fddrop(fdp);
3773 	free(kif, M_TEMP);
3774 	free(okif, M_TEMP);
3775 	return (0);
3776 }
3777 
3778 static SYSCTL_NODE(_kern_proc, KERN_PROC_OFILEDESC, ofiledesc,
3779     CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_ofiledesc,
3780     "Process ofiledesc entries");
3781 #endif	/* COMPAT_FREEBSD7 */
3782 
3783 int
3784 vntype_to_kinfo(int vtype)
3785 {
3786 	struct {
3787 		int	vtype;
3788 		int	kf_vtype;
3789 	} vtypes_table[] = {
3790 		{ VBAD, KF_VTYPE_VBAD },
3791 		{ VBLK, KF_VTYPE_VBLK },
3792 		{ VCHR, KF_VTYPE_VCHR },
3793 		{ VDIR, KF_VTYPE_VDIR },
3794 		{ VFIFO, KF_VTYPE_VFIFO },
3795 		{ VLNK, KF_VTYPE_VLNK },
3796 		{ VNON, KF_VTYPE_VNON },
3797 		{ VREG, KF_VTYPE_VREG },
3798 		{ VSOCK, KF_VTYPE_VSOCK }
3799 	};
3800 	unsigned int i;
3801 
3802 	/*
3803 	 * Perform vtype translation.
3804 	 */
3805 	for (i = 0; i < nitems(vtypes_table); i++)
3806 		if (vtypes_table[i].vtype == vtype)
3807 			return (vtypes_table[i].kf_vtype);
3808 
3809 	return (KF_VTYPE_UNKNOWN);
3810 }
3811 
3812 static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc,
3813     CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_filedesc,
3814     "Process filedesc entries");
3815 
3816 /*
3817  * Store a process current working directory information to sbuf.
3818  *
3819  * Takes a locked proc as argument, and returns with the proc unlocked.
3820  */
3821 int
3822 kern_proc_cwd_out(struct proc *p,  struct sbuf *sb, ssize_t maxlen)
3823 {
3824 	struct filedesc *fdp;
3825 	struct export_fd_buf *efbuf;
3826 	int error;
3827 
3828 	PROC_LOCK_ASSERT(p, MA_OWNED);
3829 
3830 	fdp = fdhold(p);
3831 	PROC_UNLOCK(p);
3832 	if (fdp == NULL)
3833 		return (EINVAL);
3834 
3835 	efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
3836 	efbuf->fdp = fdp;
3837 	efbuf->sb = sb;
3838 	efbuf->remainder = maxlen;
3839 
3840 	FILEDESC_SLOCK(fdp);
3841 	if (fdp->fd_cdir == NULL)
3842 		error = EINVAL;
3843 	else {
3844 		vrefact(fdp->fd_cdir);
3845 		error = export_vnode_to_sb(fdp->fd_cdir, KF_FD_TYPE_CWD,
3846 		    FREAD, efbuf);
3847 	}
3848 	FILEDESC_SUNLOCK(fdp);
3849 	fddrop(fdp);
3850 	free(efbuf, M_TEMP);
3851 	return (error);
3852 }
3853 
3854 /*
3855  * Get per-process current working directory.
3856  */
3857 static int
3858 sysctl_kern_proc_cwd(SYSCTL_HANDLER_ARGS)
3859 {
3860 	struct sbuf sb;
3861 	struct proc *p;
3862 	ssize_t maxlen;
3863 	int error, error2, *name;
3864 
3865 	name = (int *)arg1;
3866 
3867 	sbuf_new_for_sysctl(&sb, NULL, sizeof(struct kinfo_file), req);
3868 	sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
3869 	error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
3870 	if (error != 0) {
3871 		sbuf_delete(&sb);
3872 		return (error);
3873 	}
3874 	maxlen = req->oldptr != NULL ? req->oldlen : -1;
3875 	error = kern_proc_cwd_out(p, &sb, maxlen);
3876 	error2 = sbuf_finish(&sb);
3877 	sbuf_delete(&sb);
3878 	return (error != 0 ? error : error2);
3879 }
3880 
3881 static SYSCTL_NODE(_kern_proc, KERN_PROC_CWD, cwd, CTLFLAG_RD|CTLFLAG_MPSAFE,
3882     sysctl_kern_proc_cwd, "Process current working directory");
3883 
3884 #ifdef DDB
3885 /*
3886  * For the purposes of debugging, generate a human-readable string for the
3887  * file type.
3888  */
3889 static const char *
3890 file_type_to_name(short type)
3891 {
3892 
3893 	switch (type) {
3894 	case 0:
3895 		return ("zero");
3896 	case DTYPE_VNODE:
3897 		return ("vnode");
3898 	case DTYPE_SOCKET:
3899 		return ("socket");
3900 	case DTYPE_PIPE:
3901 		return ("pipe");
3902 	case DTYPE_FIFO:
3903 		return ("fifo");
3904 	case DTYPE_KQUEUE:
3905 		return ("kqueue");
3906 	case DTYPE_CRYPTO:
3907 		return ("crypto");
3908 	case DTYPE_MQUEUE:
3909 		return ("mqueue");
3910 	case DTYPE_SHM:
3911 		return ("shm");
3912 	case DTYPE_SEM:
3913 		return ("ksem");
3914 	case DTYPE_PTS:
3915 		return ("pts");
3916 	case DTYPE_DEV:
3917 		return ("dev");
3918 	case DTYPE_PROCDESC:
3919 		return ("proc");
3920 	case DTYPE_LINUXEFD:
3921 		return ("levent");
3922 	case DTYPE_LINUXTFD:
3923 		return ("ltimer");
3924 	default:
3925 		return ("unkn");
3926 	}
3927 }
3928 
3929 /*
3930  * For the purposes of debugging, identify a process (if any, perhaps one of
3931  * many) that references the passed file in its file descriptor array. Return
3932  * NULL if none.
3933  */
3934 static struct proc *
3935 file_to_first_proc(struct file *fp)
3936 {
3937 	struct filedesc *fdp;
3938 	struct proc *p;
3939 	int n;
3940 
3941 	FOREACH_PROC_IN_SYSTEM(p) {
3942 		if (p->p_state == PRS_NEW)
3943 			continue;
3944 		fdp = p->p_fd;
3945 		if (fdp == NULL)
3946 			continue;
3947 		for (n = 0; n <= fdp->fd_lastfile; n++) {
3948 			if (fp == fdp->fd_ofiles[n].fde_file)
3949 				return (p);
3950 		}
3951 	}
3952 	return (NULL);
3953 }
3954 
3955 static void
3956 db_print_file(struct file *fp, int header)
3957 {
3958 #define XPTRWIDTH ((int)howmany(sizeof(void *) * NBBY, 4))
3959 	struct proc *p;
3960 
3961 	if (header)
3962 		db_printf("%*s %6s %*s %8s %4s %5s %6s %*s %5s %s\n",
3963 		    XPTRWIDTH, "File", "Type", XPTRWIDTH, "Data", "Flag",
3964 		    "GCFl", "Count", "MCount", XPTRWIDTH, "Vnode", "FPID",
3965 		    "FCmd");
3966 	p = file_to_first_proc(fp);
3967 	db_printf("%*p %6s %*p %08x %04x %5d %6d %*p %5d %s\n", XPTRWIDTH,
3968 	    fp, file_type_to_name(fp->f_type), XPTRWIDTH, fp->f_data,
3969 	    fp->f_flag, 0, fp->f_count, 0, XPTRWIDTH, fp->f_vnode,
3970 	    p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-");
3971 
3972 #undef XPTRWIDTH
3973 }
3974 
3975 DB_SHOW_COMMAND(file, db_show_file)
3976 {
3977 	struct file *fp;
3978 
3979 	if (!have_addr) {
3980 		db_printf("usage: show file <addr>\n");
3981 		return;
3982 	}
3983 	fp = (struct file *)addr;
3984 	db_print_file(fp, 1);
3985 }
3986 
3987 DB_SHOW_COMMAND(files, db_show_files)
3988 {
3989 	struct filedesc *fdp;
3990 	struct file *fp;
3991 	struct proc *p;
3992 	int header;
3993 	int n;
3994 
3995 	header = 1;
3996 	FOREACH_PROC_IN_SYSTEM(p) {
3997 		if (p->p_state == PRS_NEW)
3998 			continue;
3999 		if ((fdp = p->p_fd) == NULL)
4000 			continue;
4001 		for (n = 0; n <= fdp->fd_lastfile; ++n) {
4002 			if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
4003 				continue;
4004 			db_print_file(fp, header);
4005 			header = 0;
4006 		}
4007 	}
4008 }
4009 #endif
4010 
4011 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
4012     &maxfilesperproc, 0, "Maximum files allowed open per process");
4013 
4014 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
4015     &maxfiles, 0, "Maximum number of files");
4016 
4017 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
4018     __DEVOLATILE(int *, &openfiles), 0, "System-wide number of open files");
4019 
4020 /* ARGSUSED*/
4021 static void
4022 filelistinit(void *dummy)
4023 {
4024 
4025 	file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL,
4026 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
4027 	filedesc0_zone = uma_zcreate("filedesc0", sizeof(struct filedesc0),
4028 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
4029 	mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF);
4030 }
4031 SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL);
4032 
4033 /*-------------------------------------------------------------------*/
4034 
4035 static int
4036 badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred,
4037     int flags, struct thread *td)
4038 {
4039 
4040 	return (EBADF);
4041 }
4042 
4043 static int
4044 badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
4045     struct thread *td)
4046 {
4047 
4048 	return (EINVAL);
4049 }
4050 
4051 static int
4052 badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred,
4053     struct thread *td)
4054 {
4055 
4056 	return (EBADF);
4057 }
4058 
4059 static int
4060 badfo_poll(struct file *fp, int events, struct ucred *active_cred,
4061     struct thread *td)
4062 {
4063 
4064 	return (0);
4065 }
4066 
4067 static int
4068 badfo_kqfilter(struct file *fp, struct knote *kn)
4069 {
4070 
4071 	return (EBADF);
4072 }
4073 
4074 static int
4075 badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred,
4076     struct thread *td)
4077 {
4078 
4079 	return (EBADF);
4080 }
4081 
4082 static int
4083 badfo_close(struct file *fp, struct thread *td)
4084 {
4085 
4086 	return (0);
4087 }
4088 
4089 static int
4090 badfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
4091     struct thread *td)
4092 {
4093 
4094 	return (EBADF);
4095 }
4096 
4097 static int
4098 badfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
4099     struct thread *td)
4100 {
4101 
4102 	return (EBADF);
4103 }
4104 
4105 static int
4106 badfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
4107     struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
4108     struct thread *td)
4109 {
4110 
4111 	return (EBADF);
4112 }
4113 
4114 static int
4115 badfo_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
4116 {
4117 
4118 	return (0);
4119 }
4120 
4121 struct fileops badfileops = {
4122 	.fo_read = badfo_readwrite,
4123 	.fo_write = badfo_readwrite,
4124 	.fo_truncate = badfo_truncate,
4125 	.fo_ioctl = badfo_ioctl,
4126 	.fo_poll = badfo_poll,
4127 	.fo_kqfilter = badfo_kqfilter,
4128 	.fo_stat = badfo_stat,
4129 	.fo_close = badfo_close,
4130 	.fo_chmod = badfo_chmod,
4131 	.fo_chown = badfo_chown,
4132 	.fo_sendfile = badfo_sendfile,
4133 	.fo_fill_kinfo = badfo_fill_kinfo,
4134 };
4135 
4136 int
4137 invfo_rdwr(struct file *fp, struct uio *uio, struct ucred *active_cred,
4138     int flags, struct thread *td)
4139 {
4140 
4141 	return (EOPNOTSUPP);
4142 }
4143 
4144 int
4145 invfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
4146     struct thread *td)
4147 {
4148 
4149 	return (EINVAL);
4150 }
4151 
4152 int
4153 invfo_ioctl(struct file *fp, u_long com, void *data,
4154     struct ucred *active_cred, struct thread *td)
4155 {
4156 
4157 	return (ENOTTY);
4158 }
4159 
4160 int
4161 invfo_poll(struct file *fp, int events, struct ucred *active_cred,
4162     struct thread *td)
4163 {
4164 
4165 	return (poll_no_poll(events));
4166 }
4167 
4168 int
4169 invfo_kqfilter(struct file *fp, struct knote *kn)
4170 {
4171 
4172 	return (EINVAL);
4173 }
4174 
4175 int
4176 invfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
4177     struct thread *td)
4178 {
4179 
4180 	return (EINVAL);
4181 }
4182 
4183 int
4184 invfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
4185     struct thread *td)
4186 {
4187 
4188 	return (EINVAL);
4189 }
4190 
4191 int
4192 invfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
4193     struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
4194     struct thread *td)
4195 {
4196 
4197 	return (EINVAL);
4198 }
4199 
4200 /*-------------------------------------------------------------------*/
4201 
4202 /*
4203  * File Descriptor pseudo-device driver (/dev/fd/).
4204  *
4205  * Opening minor device N dup()s the file (if any) connected to file
4206  * descriptor N belonging to the calling process.  Note that this driver
4207  * consists of only the ``open()'' routine, because all subsequent
4208  * references to this file will be direct to the other driver.
4209  *
4210  * XXX: we could give this one a cloning event handler if necessary.
4211  */
4212 
4213 /* ARGSUSED */
4214 static int
4215 fdopen(struct cdev *dev, int mode, int type, struct thread *td)
4216 {
4217 
4218 	/*
4219 	 * XXX Kludge: set curthread->td_dupfd to contain the value of the
4220 	 * the file descriptor being sought for duplication. The error
4221 	 * return ensures that the vnode for this device will be released
4222 	 * by vn_open. Open will detect this special error and take the
4223 	 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
4224 	 * will simply report the error.
4225 	 */
4226 	td->td_dupfd = dev2unit(dev);
4227 	return (ENODEV);
4228 }
4229 
4230 static struct cdevsw fildesc_cdevsw = {
4231 	.d_version =	D_VERSION,
4232 	.d_open =	fdopen,
4233 	.d_name =	"FD",
4234 };
4235 
4236 static void
4237 fildesc_drvinit(void *unused)
4238 {
4239 	struct cdev *dev;
4240 
4241 	dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 0, NULL,
4242 	    UID_ROOT, GID_WHEEL, 0666, "fd/0");
4243 	make_dev_alias(dev, "stdin");
4244 	dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 1, NULL,
4245 	    UID_ROOT, GID_WHEEL, 0666, "fd/1");
4246 	make_dev_alias(dev, "stdout");
4247 	dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 2, NULL,
4248 	    UID_ROOT, GID_WHEEL, 0666, "fd/2");
4249 	make_dev_alias(dev, "stderr");
4250 }
4251 
4252 SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL);
4253