xref: /freebsd/sys/kern/kern_descrip.c (revision af6a5351a1fdb1130f18be6c782c4d48916eb971)
1 /*-
2  * Copyright (c) 1982, 1986, 1989, 1991, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  * (c) UNIX System Laboratories, Inc.
5  * All or some portions of this file are derived from material licensed
6  * to the University of California by American Telephone and Telegraph
7  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8  * the permission of UNIX System Laboratories, Inc.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  *	@(#)kern_descrip.c	8.6 (Berkeley) 4/19/94
35  */
36 
37 #include <sys/cdefs.h>
38 __FBSDID("$FreeBSD$");
39 
40 #include "opt_capsicum.h"
41 #include "opt_compat.h"
42 #include "opt_ddb.h"
43 #include "opt_ktrace.h"
44 
45 #include <sys/param.h>
46 #include <sys/systm.h>
47 
48 #include <sys/capsicum.h>
49 #include <sys/conf.h>
50 #include <sys/fcntl.h>
51 #include <sys/file.h>
52 #include <sys/filedesc.h>
53 #include <sys/filio.h>
54 #include <sys/jail.h>
55 #include <sys/kernel.h>
56 #include <sys/limits.h>
57 #include <sys/lock.h>
58 #include <sys/malloc.h>
59 #include <sys/mount.h>
60 #include <sys/mutex.h>
61 #include <sys/namei.h>
62 #include <sys/selinfo.h>
63 #include <sys/priv.h>
64 #include <sys/proc.h>
65 #include <sys/protosw.h>
66 #include <sys/racct.h>
67 #include <sys/resourcevar.h>
68 #include <sys/sbuf.h>
69 #include <sys/signalvar.h>
70 #include <sys/kdb.h>
71 #include <sys/stat.h>
72 #include <sys/sx.h>
73 #include <sys/syscallsubr.h>
74 #include <sys/sysctl.h>
75 #include <sys/sysproto.h>
76 #include <sys/unistd.h>
77 #include <sys/user.h>
78 #include <sys/vnode.h>
79 #ifdef KTRACE
80 #include <sys/ktrace.h>
81 #endif
82 
83 #include <net/vnet.h>
84 
85 #include <security/audit/audit.h>
86 
87 #include <vm/uma.h>
88 #include <vm/vm.h>
89 
90 #include <ddb/ddb.h>
91 
92 static MALLOC_DEFINE(M_FILEDESC, "filedesc", "Open file descriptor table");
93 static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "filedesc_to_leader",
94     "file desc to leader structures");
95 static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures");
96 MALLOC_DEFINE(M_FILECAPS, "filecaps", "descriptor capabilities");
97 
98 MALLOC_DECLARE(M_FADVISE);
99 
100 static __read_mostly uma_zone_t file_zone;
101 static __read_mostly uma_zone_t filedesc0_zone;
102 
103 static int	closefp(struct filedesc *fdp, int fd, struct file *fp,
104 		    struct thread *td, int holdleaders);
105 static int	fd_first_free(struct filedesc *fdp, int low, int size);
106 static int	fd_last_used(struct filedesc *fdp, int size);
107 static void	fdgrowtable(struct filedesc *fdp, int nfd);
108 static void	fdgrowtable_exp(struct filedesc *fdp, int nfd);
109 static void	fdunused(struct filedesc *fdp, int fd);
110 static void	fdused(struct filedesc *fdp, int fd);
111 static int	getmaxfd(struct thread *td);
112 
113 /*
114  * Each process has:
115  *
116  * - An array of open file descriptors (fd_ofiles)
117  * - An array of file flags (fd_ofileflags)
118  * - A bitmap recording which descriptors are in use (fd_map)
119  *
120  * A process starts out with NDFILE descriptors.  The value of NDFILE has
121  * been selected based the historical limit of 20 open files, and an
122  * assumption that the majority of processes, especially short-lived
123  * processes like shells, will never need more.
124  *
125  * If this initial allocation is exhausted, a larger descriptor table and
126  * map are allocated dynamically, and the pointers in the process's struct
127  * filedesc are updated to point to those.  This is repeated every time
128  * the process runs out of file descriptors (provided it hasn't hit its
129  * resource limit).
130  *
131  * Since threads may hold references to individual descriptor table
132  * entries, the tables are never freed.  Instead, they are placed on a
133  * linked list and freed only when the struct filedesc is released.
134  */
135 #define NDFILE		20
136 #define NDSLOTSIZE	sizeof(NDSLOTTYPE)
137 #define	NDENTRIES	(NDSLOTSIZE * __CHAR_BIT)
138 #define NDSLOT(x)	((x) / NDENTRIES)
139 #define NDBIT(x)	((NDSLOTTYPE)1 << ((x) % NDENTRIES))
140 #define	NDSLOTS(x)	(((x) + NDENTRIES - 1) / NDENTRIES)
141 
142 /*
143  * SLIST entry used to keep track of ofiles which must be reclaimed when
144  * the process exits.
145  */
146 struct freetable {
147 	struct fdescenttbl *ft_table;
148 	SLIST_ENTRY(freetable) ft_next;
149 };
150 
151 /*
152  * Initial allocation: a filedesc structure + the head of SLIST used to
153  * keep track of old ofiles + enough space for NDFILE descriptors.
154  */
155 
156 struct fdescenttbl0 {
157 	int	fdt_nfiles;
158 	struct	filedescent fdt_ofiles[NDFILE];
159 };
160 
161 struct filedesc0 {
162 	struct filedesc fd_fd;
163 	SLIST_HEAD(, freetable) fd_free;
164 	struct	fdescenttbl0 fd_dfiles;
165 	NDSLOTTYPE fd_dmap[NDSLOTS(NDFILE)];
166 };
167 
168 /*
169  * Descriptor management.
170  */
171 volatile int __exclusive_cache_line openfiles; /* actual number of open files */
172 struct mtx sigio_lock;		/* mtx to protect pointers to sigio */
173 void __read_mostly (*mq_fdclose)(struct thread *td, int fd, struct file *fp);
174 
175 /*
176  * If low >= size, just return low. Otherwise find the first zero bit in the
177  * given bitmap, starting at low and not exceeding size - 1. Return size if
178  * not found.
179  */
180 static int
181 fd_first_free(struct filedesc *fdp, int low, int size)
182 {
183 	NDSLOTTYPE *map = fdp->fd_map;
184 	NDSLOTTYPE mask;
185 	int off, maxoff;
186 
187 	if (low >= size)
188 		return (low);
189 
190 	off = NDSLOT(low);
191 	if (low % NDENTRIES) {
192 		mask = ~(~(NDSLOTTYPE)0 >> (NDENTRIES - (low % NDENTRIES)));
193 		if ((mask &= ~map[off]) != 0UL)
194 			return (off * NDENTRIES + ffsl(mask) - 1);
195 		++off;
196 	}
197 	for (maxoff = NDSLOTS(size); off < maxoff; ++off)
198 		if (map[off] != ~0UL)
199 			return (off * NDENTRIES + ffsl(~map[off]) - 1);
200 	return (size);
201 }
202 
203 /*
204  * Find the highest non-zero bit in the given bitmap, starting at 0 and
205  * not exceeding size - 1. Return -1 if not found.
206  */
207 static int
208 fd_last_used(struct filedesc *fdp, int size)
209 {
210 	NDSLOTTYPE *map = fdp->fd_map;
211 	NDSLOTTYPE mask;
212 	int off, minoff;
213 
214 	off = NDSLOT(size);
215 	if (size % NDENTRIES) {
216 		mask = ~(~(NDSLOTTYPE)0 << (size % NDENTRIES));
217 		if ((mask &= map[off]) != 0)
218 			return (off * NDENTRIES + flsl(mask) - 1);
219 		--off;
220 	}
221 	for (minoff = NDSLOT(0); off >= minoff; --off)
222 		if (map[off] != 0)
223 			return (off * NDENTRIES + flsl(map[off]) - 1);
224 	return (-1);
225 }
226 
227 static int
228 fdisused(struct filedesc *fdp, int fd)
229 {
230 
231 	KASSERT(fd >= 0 && fd < fdp->fd_nfiles,
232 	    ("file descriptor %d out of range (0, %d)", fd, fdp->fd_nfiles));
233 
234 	return ((fdp->fd_map[NDSLOT(fd)] & NDBIT(fd)) != 0);
235 }
236 
237 /*
238  * Mark a file descriptor as used.
239  */
240 static void
241 fdused_init(struct filedesc *fdp, int fd)
242 {
243 
244 	KASSERT(!fdisused(fdp, fd), ("fd=%d is already used", fd));
245 
246 	fdp->fd_map[NDSLOT(fd)] |= NDBIT(fd);
247 }
248 
249 static void
250 fdused(struct filedesc *fdp, int fd)
251 {
252 
253 	FILEDESC_XLOCK_ASSERT(fdp);
254 
255 	fdused_init(fdp, fd);
256 	if (fd > fdp->fd_lastfile)
257 		fdp->fd_lastfile = fd;
258 	if (fd == fdp->fd_freefile)
259 		fdp->fd_freefile = fd_first_free(fdp, fd, fdp->fd_nfiles);
260 }
261 
262 /*
263  * Mark a file descriptor as unused.
264  */
265 static void
266 fdunused(struct filedesc *fdp, int fd)
267 {
268 
269 	FILEDESC_XLOCK_ASSERT(fdp);
270 
271 	KASSERT(fdisused(fdp, fd), ("fd=%d is already unused", fd));
272 	KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
273 	    ("fd=%d is still in use", fd));
274 
275 	fdp->fd_map[NDSLOT(fd)] &= ~NDBIT(fd);
276 	if (fd < fdp->fd_freefile)
277 		fdp->fd_freefile = fd;
278 	if (fd == fdp->fd_lastfile)
279 		fdp->fd_lastfile = fd_last_used(fdp, fd);
280 }
281 
282 /*
283  * Free a file descriptor.
284  *
285  * Avoid some work if fdp is about to be destroyed.
286  */
287 static inline void
288 fdefree_last(struct filedescent *fde)
289 {
290 
291 	filecaps_free(&fde->fde_caps);
292 }
293 
294 static inline void
295 fdfree(struct filedesc *fdp, int fd)
296 {
297 	struct filedescent *fde;
298 
299 	fde = &fdp->fd_ofiles[fd];
300 #ifdef CAPABILITIES
301 	seq_write_begin(&fde->fde_seq);
302 #endif
303 	fdefree_last(fde);
304 	fde->fde_file = NULL;
305 	fdunused(fdp, fd);
306 #ifdef CAPABILITIES
307 	seq_write_end(&fde->fde_seq);
308 #endif
309 }
310 
311 void
312 pwd_ensure_dirs(void)
313 {
314 	struct filedesc *fdp;
315 
316 	fdp = curproc->p_fd;
317 	FILEDESC_XLOCK(fdp);
318 	if (fdp->fd_cdir == NULL) {
319 		fdp->fd_cdir = rootvnode;
320 		vrefact(rootvnode);
321 	}
322 	if (fdp->fd_rdir == NULL) {
323 		fdp->fd_rdir = rootvnode;
324 		vrefact(rootvnode);
325 	}
326 	FILEDESC_XUNLOCK(fdp);
327 }
328 
329 /*
330  * System calls on descriptors.
331  */
332 #ifndef _SYS_SYSPROTO_H_
333 struct getdtablesize_args {
334 	int	dummy;
335 };
336 #endif
337 /* ARGSUSED */
338 int
339 sys_getdtablesize(struct thread *td, struct getdtablesize_args *uap)
340 {
341 #ifdef	RACCT
342 	uint64_t lim;
343 #endif
344 
345 	td->td_retval[0] =
346 	    min((int)lim_cur(td, RLIMIT_NOFILE), maxfilesperproc);
347 #ifdef	RACCT
348 	PROC_LOCK(td->td_proc);
349 	lim = racct_get_limit(td->td_proc, RACCT_NOFILE);
350 	PROC_UNLOCK(td->td_proc);
351 	if (lim < td->td_retval[0])
352 		td->td_retval[0] = lim;
353 #endif
354 	return (0);
355 }
356 
357 /*
358  * Duplicate a file descriptor to a particular value.
359  *
360  * Note: keep in mind that a potential race condition exists when closing
361  * descriptors from a shared descriptor table (via rfork).
362  */
363 #ifndef _SYS_SYSPROTO_H_
364 struct dup2_args {
365 	u_int	from;
366 	u_int	to;
367 };
368 #endif
369 /* ARGSUSED */
370 int
371 sys_dup2(struct thread *td, struct dup2_args *uap)
372 {
373 
374 	return (kern_dup(td, FDDUP_FIXED, 0, (int)uap->from, (int)uap->to));
375 }
376 
377 /*
378  * Duplicate a file descriptor.
379  */
380 #ifndef _SYS_SYSPROTO_H_
381 struct dup_args {
382 	u_int	fd;
383 };
384 #endif
385 /* ARGSUSED */
386 int
387 sys_dup(struct thread *td, struct dup_args *uap)
388 {
389 
390 	return (kern_dup(td, FDDUP_NORMAL, 0, (int)uap->fd, 0));
391 }
392 
393 /*
394  * The file control system call.
395  */
396 #ifndef _SYS_SYSPROTO_H_
397 struct fcntl_args {
398 	int	fd;
399 	int	cmd;
400 	long	arg;
401 };
402 #endif
403 /* ARGSUSED */
404 int
405 sys_fcntl(struct thread *td, struct fcntl_args *uap)
406 {
407 
408 	return (kern_fcntl_freebsd(td, uap->fd, uap->cmd, uap->arg));
409 }
410 
411 int
412 kern_fcntl_freebsd(struct thread *td, int fd, int cmd, long arg)
413 {
414 	struct flock fl;
415 	struct __oflock ofl;
416 	intptr_t arg1;
417 	int error, newcmd;
418 
419 	error = 0;
420 	newcmd = cmd;
421 	switch (cmd) {
422 	case F_OGETLK:
423 	case F_OSETLK:
424 	case F_OSETLKW:
425 		/*
426 		 * Convert old flock structure to new.
427 		 */
428 		error = copyin((void *)(intptr_t)arg, &ofl, sizeof(ofl));
429 		fl.l_start = ofl.l_start;
430 		fl.l_len = ofl.l_len;
431 		fl.l_pid = ofl.l_pid;
432 		fl.l_type = ofl.l_type;
433 		fl.l_whence = ofl.l_whence;
434 		fl.l_sysid = 0;
435 
436 		switch (cmd) {
437 		case F_OGETLK:
438 			newcmd = F_GETLK;
439 			break;
440 		case F_OSETLK:
441 			newcmd = F_SETLK;
442 			break;
443 		case F_OSETLKW:
444 			newcmd = F_SETLKW;
445 			break;
446 		}
447 		arg1 = (intptr_t)&fl;
448 		break;
449 	case F_GETLK:
450 	case F_SETLK:
451 	case F_SETLKW:
452 	case F_SETLK_REMOTE:
453 		error = copyin((void *)(intptr_t)arg, &fl, sizeof(fl));
454 		arg1 = (intptr_t)&fl;
455 		break;
456 	default:
457 		arg1 = arg;
458 		break;
459 	}
460 	if (error)
461 		return (error);
462 	error = kern_fcntl(td, fd, newcmd, arg1);
463 	if (error)
464 		return (error);
465 	if (cmd == F_OGETLK) {
466 		ofl.l_start = fl.l_start;
467 		ofl.l_len = fl.l_len;
468 		ofl.l_pid = fl.l_pid;
469 		ofl.l_type = fl.l_type;
470 		ofl.l_whence = fl.l_whence;
471 		error = copyout(&ofl, (void *)(intptr_t)arg, sizeof(ofl));
472 	} else if (cmd == F_GETLK) {
473 		error = copyout(&fl, (void *)(intptr_t)arg, sizeof(fl));
474 	}
475 	return (error);
476 }
477 
478 int
479 kern_fcntl(struct thread *td, int fd, int cmd, intptr_t arg)
480 {
481 	struct filedesc *fdp;
482 	struct flock *flp;
483 	struct file *fp, *fp2;
484 	struct filedescent *fde;
485 	struct proc *p;
486 	struct vnode *vp;
487 	cap_rights_t rights;
488 	int error, flg, tmp;
489 	uint64_t bsize;
490 	off_t foffset;
491 
492 	error = 0;
493 	flg = F_POSIX;
494 	p = td->td_proc;
495 	fdp = p->p_fd;
496 
497 	AUDIT_ARG_FD(cmd);
498 	AUDIT_ARG_CMD(cmd);
499 	switch (cmd) {
500 	case F_DUPFD:
501 		tmp = arg;
502 		error = kern_dup(td, FDDUP_FCNTL, 0, fd, tmp);
503 		break;
504 
505 	case F_DUPFD_CLOEXEC:
506 		tmp = arg;
507 		error = kern_dup(td, FDDUP_FCNTL, FDDUP_FLAG_CLOEXEC, fd, tmp);
508 		break;
509 
510 	case F_DUP2FD:
511 		tmp = arg;
512 		error = kern_dup(td, FDDUP_FIXED, 0, fd, tmp);
513 		break;
514 
515 	case F_DUP2FD_CLOEXEC:
516 		tmp = arg;
517 		error = kern_dup(td, FDDUP_FIXED, FDDUP_FLAG_CLOEXEC, fd, tmp);
518 		break;
519 
520 	case F_GETFD:
521 		error = EBADF;
522 		FILEDESC_SLOCK(fdp);
523 		fde = fdeget_locked(fdp, fd);
524 		if (fde != NULL) {
525 			td->td_retval[0] =
526 			    (fde->fde_flags & UF_EXCLOSE) ? FD_CLOEXEC : 0;
527 			error = 0;
528 		}
529 		FILEDESC_SUNLOCK(fdp);
530 		break;
531 
532 	case F_SETFD:
533 		error = EBADF;
534 		FILEDESC_XLOCK(fdp);
535 		fde = fdeget_locked(fdp, fd);
536 		if (fde != NULL) {
537 			fde->fde_flags = (fde->fde_flags & ~UF_EXCLOSE) |
538 			    (arg & FD_CLOEXEC ? UF_EXCLOSE : 0);
539 			error = 0;
540 		}
541 		FILEDESC_XUNLOCK(fdp);
542 		break;
543 
544 	case F_GETFL:
545 		error = fget_fcntl(td, fd,
546 		    cap_rights_init(&rights, CAP_FCNTL), F_GETFL, &fp);
547 		if (error != 0)
548 			break;
549 		td->td_retval[0] = OFLAGS(fp->f_flag);
550 		fdrop(fp, td);
551 		break;
552 
553 	case F_SETFL:
554 		error = fget_fcntl(td, fd,
555 		    cap_rights_init(&rights, CAP_FCNTL), F_SETFL, &fp);
556 		if (error != 0)
557 			break;
558 		do {
559 			tmp = flg = fp->f_flag;
560 			tmp &= ~FCNTLFLAGS;
561 			tmp |= FFLAGS(arg & ~O_ACCMODE) & FCNTLFLAGS;
562 		} while(atomic_cmpset_int(&fp->f_flag, flg, tmp) == 0);
563 		tmp = fp->f_flag & FNONBLOCK;
564 		error = fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
565 		if (error != 0) {
566 			fdrop(fp, td);
567 			break;
568 		}
569 		tmp = fp->f_flag & FASYNC;
570 		error = fo_ioctl(fp, FIOASYNC, &tmp, td->td_ucred, td);
571 		if (error == 0) {
572 			fdrop(fp, td);
573 			break;
574 		}
575 		atomic_clear_int(&fp->f_flag, FNONBLOCK);
576 		tmp = 0;
577 		(void)fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
578 		fdrop(fp, td);
579 		break;
580 
581 	case F_GETOWN:
582 		error = fget_fcntl(td, fd,
583 		    cap_rights_init(&rights, CAP_FCNTL), F_GETOWN, &fp);
584 		if (error != 0)
585 			break;
586 		error = fo_ioctl(fp, FIOGETOWN, &tmp, td->td_ucred, td);
587 		if (error == 0)
588 			td->td_retval[0] = tmp;
589 		fdrop(fp, td);
590 		break;
591 
592 	case F_SETOWN:
593 		error = fget_fcntl(td, fd,
594 		    cap_rights_init(&rights, CAP_FCNTL), F_SETOWN, &fp);
595 		if (error != 0)
596 			break;
597 		tmp = arg;
598 		error = fo_ioctl(fp, FIOSETOWN, &tmp, td->td_ucred, td);
599 		fdrop(fp, td);
600 		break;
601 
602 	case F_SETLK_REMOTE:
603 		error = priv_check(td, PRIV_NFS_LOCKD);
604 		if (error)
605 			return (error);
606 		flg = F_REMOTE;
607 		goto do_setlk;
608 
609 	case F_SETLKW:
610 		flg |= F_WAIT;
611 		/* FALLTHROUGH F_SETLK */
612 
613 	case F_SETLK:
614 	do_setlk:
615 		cap_rights_init(&rights, CAP_FLOCK);
616 		error = fget_unlocked(fdp, fd, &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 			PROC_LOCK(p->p_leader);
646 			p->p_leader->p_flag |= P_ADVLOCK;
647 			PROC_UNLOCK(p->p_leader);
648 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
649 			    flp, flg);
650 			break;
651 		case F_WRLCK:
652 			if ((fp->f_flag & FWRITE) == 0) {
653 				error = EBADF;
654 				break;
655 			}
656 			PROC_LOCK(p->p_leader);
657 			p->p_leader->p_flag |= P_ADVLOCK;
658 			PROC_UNLOCK(p->p_leader);
659 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
660 			    flp, flg);
661 			break;
662 		case F_UNLCK:
663 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
664 			    flp, flg);
665 			break;
666 		case F_UNLCKSYS:
667 			/*
668 			 * Temporary api for testing remote lock
669 			 * infrastructure.
670 			 */
671 			if (flg != F_REMOTE) {
672 				error = EINVAL;
673 				break;
674 			}
675 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
676 			    F_UNLCKSYS, flp, flg);
677 			break;
678 		default:
679 			error = EINVAL;
680 			break;
681 		}
682 		if (error != 0 || flp->l_type == F_UNLCK ||
683 		    flp->l_type == F_UNLCKSYS) {
684 			fdrop(fp, td);
685 			break;
686 		}
687 
688 		/*
689 		 * Check for a race with close.
690 		 *
691 		 * The vnode is now advisory locked (or unlocked, but this case
692 		 * is not really important) as the caller requested.
693 		 * We had to drop the filedesc lock, so we need to recheck if
694 		 * the descriptor is still valid, because if it was closed
695 		 * in the meantime we need to remove advisory lock from the
696 		 * vnode - close on any descriptor leading to an advisory
697 		 * locked vnode, removes that lock.
698 		 * We will return 0 on purpose in that case, as the result of
699 		 * successful advisory lock might have been externally visible
700 		 * already. This is fine - effectively we pretend to the caller
701 		 * that the closing thread was a bit slower and that the
702 		 * advisory lock succeeded before the close.
703 		 */
704 		error = fget_unlocked(fdp, fd, &rights, &fp2, NULL);
705 		if (error != 0) {
706 			fdrop(fp, td);
707 			break;
708 		}
709 		if (fp != fp2) {
710 			flp->l_whence = SEEK_SET;
711 			flp->l_start = 0;
712 			flp->l_len = 0;
713 			flp->l_type = F_UNLCK;
714 			(void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
715 			    F_UNLCK, flp, F_POSIX);
716 		}
717 		fdrop(fp, td);
718 		fdrop(fp2, td);
719 		break;
720 
721 	case F_GETLK:
722 		error = fget_unlocked(fdp, fd,
723 		    cap_rights_init(&rights, CAP_FLOCK), &fp, NULL);
724 		if (error != 0)
725 			break;
726 		if (fp->f_type != DTYPE_VNODE) {
727 			error = EBADF;
728 			fdrop(fp, td);
729 			break;
730 		}
731 		flp = (struct flock *)arg;
732 		if (flp->l_type != F_RDLCK && flp->l_type != F_WRLCK &&
733 		    flp->l_type != F_UNLCK) {
734 			error = EINVAL;
735 			fdrop(fp, td);
736 			break;
737 		}
738 		if (flp->l_whence == SEEK_CUR) {
739 			foffset = foffset_get(fp);
740 			if ((flp->l_start > 0 &&
741 			    foffset > OFF_MAX - flp->l_start) ||
742 			    (flp->l_start < 0 &&
743 			    foffset < OFF_MIN - flp->l_start)) {
744 				error = EOVERFLOW;
745 				fdrop(fp, td);
746 				break;
747 			}
748 			flp->l_start += foffset;
749 		}
750 		vp = fp->f_vnode;
751 		error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, flp,
752 		    F_POSIX);
753 		fdrop(fp, td);
754 		break;
755 
756 	case F_RDAHEAD:
757 		arg = arg ? 128 * 1024: 0;
758 		/* FALLTHROUGH */
759 	case F_READAHEAD:
760 		error = fget_unlocked(fdp, fd,
761 		    cap_rights_init(&rights), &fp, NULL);
762 		if (error != 0)
763 			break;
764 		if (fp->f_type != DTYPE_VNODE) {
765 			fdrop(fp, td);
766 			error = EBADF;
767 			break;
768 		}
769 		vp = fp->f_vnode;
770 		/*
771 		 * Exclusive lock synchronizes against f_seqcount reads and
772 		 * writes in sequential_heuristic().
773 		 */
774 		error = vn_lock(vp, LK_EXCLUSIVE);
775 		if (error != 0) {
776 			fdrop(fp, td);
777 			break;
778 		}
779 		if (arg >= 0) {
780 			bsize = fp->f_vnode->v_mount->mnt_stat.f_iosize;
781 			fp->f_seqcount = (arg + bsize - 1) / bsize;
782 			atomic_set_int(&fp->f_flag, FRDAHEAD);
783 		} else {
784 			atomic_clear_int(&fp->f_flag, FRDAHEAD);
785 		}
786 		VOP_UNLOCK(vp, 0);
787 		fdrop(fp, td);
788 		break;
789 
790 	default:
791 		error = EINVAL;
792 		break;
793 	}
794 	return (error);
795 }
796 
797 static int
798 getmaxfd(struct thread *td)
799 {
800 
801 	return (min((int)lim_cur(td, RLIMIT_NOFILE), maxfilesperproc));
802 }
803 
804 /*
805  * Common code for dup, dup2, fcntl(F_DUPFD) and fcntl(F_DUP2FD).
806  */
807 int
808 kern_dup(struct thread *td, u_int mode, int flags, int old, int new)
809 {
810 	struct filedesc *fdp;
811 	struct filedescent *oldfde, *newfde;
812 	struct proc *p;
813 	struct file *delfp;
814 	int error, maxfd;
815 
816 	p = td->td_proc;
817 	fdp = p->p_fd;
818 
819 	MPASS((flags & ~(FDDUP_FLAG_CLOEXEC)) == 0);
820 	MPASS(mode < FDDUP_LASTMODE);
821 
822 	AUDIT_ARG_FD(old);
823 	/* XXXRW: if (flags & FDDUP_FIXED) AUDIT_ARG_FD2(new); */
824 
825 	/*
826 	 * Verify we have a valid descriptor to dup from and possibly to
827 	 * dup to. Unlike dup() and dup2(), fcntl()'s F_DUPFD should
828 	 * return EINVAL when the new descriptor is out of bounds.
829 	 */
830 	if (old < 0)
831 		return (EBADF);
832 	if (new < 0)
833 		return (mode == FDDUP_FCNTL ? EINVAL : EBADF);
834 	maxfd = getmaxfd(td);
835 	if (new >= maxfd)
836 		return (mode == FDDUP_FCNTL ? EINVAL : EBADF);
837 
838 	error = EBADF;
839 	FILEDESC_XLOCK(fdp);
840 	if (fget_locked(fdp, old) == NULL)
841 		goto unlock;
842 	if ((mode == FDDUP_FIXED || mode == FDDUP_MUSTREPLACE) && old == new) {
843 		td->td_retval[0] = new;
844 		if (flags & FDDUP_FLAG_CLOEXEC)
845 			fdp->fd_ofiles[new].fde_flags |= UF_EXCLOSE;
846 		error = 0;
847 		goto unlock;
848 	}
849 
850 	/*
851 	 * If the caller specified a file descriptor, make sure the file
852 	 * table is large enough to hold it, and grab it.  Otherwise, just
853 	 * allocate a new descriptor the usual way.
854 	 */
855 	switch (mode) {
856 	case FDDUP_NORMAL:
857 	case FDDUP_FCNTL:
858 		if ((error = fdalloc(td, new, &new)) != 0)
859 			goto unlock;
860 		break;
861 	case FDDUP_MUSTREPLACE:
862 		/* Target file descriptor must exist. */
863 		if (fget_locked(fdp, new) == NULL)
864 			goto unlock;
865 		break;
866 	case FDDUP_FIXED:
867 		if (new >= fdp->fd_nfiles) {
868 			/*
869 			 * The resource limits are here instead of e.g.
870 			 * fdalloc(), because the file descriptor table may be
871 			 * shared between processes, so we can't really use
872 			 * racct_add()/racct_sub().  Instead of counting the
873 			 * number of actually allocated descriptors, just put
874 			 * the limit on the size of the file descriptor table.
875 			 */
876 #ifdef RACCT
877 			if (racct_enable) {
878 				PROC_LOCK(p);
879 				error = racct_set(p, RACCT_NOFILE, new + 1);
880 				PROC_UNLOCK(p);
881 				if (error != 0) {
882 					error = EMFILE;
883 					goto unlock;
884 				}
885 			}
886 #endif
887 			fdgrowtable_exp(fdp, new + 1);
888 		}
889 		if (!fdisused(fdp, new))
890 			fdused(fdp, new);
891 		break;
892 	default:
893 		KASSERT(0, ("%s unsupported mode %d", __func__, mode));
894 	}
895 
896 	KASSERT(old != new, ("new fd is same as old"));
897 
898 	oldfde = &fdp->fd_ofiles[old];
899 	fhold(oldfde->fde_file);
900 	newfde = &fdp->fd_ofiles[new];
901 	delfp = newfde->fde_file;
902 
903 	/*
904 	 * Duplicate the source descriptor.
905 	 */
906 #ifdef CAPABILITIES
907 	seq_write_begin(&newfde->fde_seq);
908 #endif
909 	filecaps_free(&newfde->fde_caps);
910 	memcpy(newfde, oldfde, fde_change_size);
911 	filecaps_copy(&oldfde->fde_caps, &newfde->fde_caps, true);
912 	if ((flags & FDDUP_FLAG_CLOEXEC) != 0)
913 		newfde->fde_flags = oldfde->fde_flags | UF_EXCLOSE;
914 	else
915 		newfde->fde_flags = oldfde->fde_flags & ~UF_EXCLOSE;
916 #ifdef CAPABILITIES
917 	seq_write_end(&newfde->fde_seq);
918 #endif
919 	td->td_retval[0] = new;
920 
921 	error = 0;
922 
923 	if (delfp != NULL) {
924 		(void) closefp(fdp, new, delfp, td, 1);
925 		FILEDESC_UNLOCK_ASSERT(fdp);
926 	} else {
927 unlock:
928 		FILEDESC_XUNLOCK(fdp);
929 	}
930 
931 	return (error);
932 }
933 
934 /*
935  * If sigio is on the list associated with a process or process group,
936  * disable signalling from the device, remove sigio from the list and
937  * free sigio.
938  */
939 void
940 funsetown(struct sigio **sigiop)
941 {
942 	struct sigio *sigio;
943 
944 	if (*sigiop == NULL)
945 		return;
946 	SIGIO_LOCK();
947 	sigio = *sigiop;
948 	if (sigio == NULL) {
949 		SIGIO_UNLOCK();
950 		return;
951 	}
952 	*(sigio->sio_myref) = NULL;
953 	if ((sigio)->sio_pgid < 0) {
954 		struct pgrp *pg = (sigio)->sio_pgrp;
955 		PGRP_LOCK(pg);
956 		SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio,
957 			    sigio, sio_pgsigio);
958 		PGRP_UNLOCK(pg);
959 	} else {
960 		struct proc *p = (sigio)->sio_proc;
961 		PROC_LOCK(p);
962 		SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio,
963 			    sigio, sio_pgsigio);
964 		PROC_UNLOCK(p);
965 	}
966 	SIGIO_UNLOCK();
967 	crfree(sigio->sio_ucred);
968 	free(sigio, M_SIGIO);
969 }
970 
971 /*
972  * Free a list of sigio structures.
973  * We only need to lock the SIGIO_LOCK because we have made ourselves
974  * inaccessible to callers of fsetown and therefore do not need to lock
975  * the proc or pgrp struct for the list manipulation.
976  */
977 void
978 funsetownlst(struct sigiolst *sigiolst)
979 {
980 	struct proc *p;
981 	struct pgrp *pg;
982 	struct sigio *sigio;
983 
984 	sigio = SLIST_FIRST(sigiolst);
985 	if (sigio == NULL)
986 		return;
987 	p = NULL;
988 	pg = NULL;
989 
990 	/*
991 	 * Every entry of the list should belong
992 	 * to a single proc or pgrp.
993 	 */
994 	if (sigio->sio_pgid < 0) {
995 		pg = sigio->sio_pgrp;
996 		PGRP_LOCK_ASSERT(pg, MA_NOTOWNED);
997 	} else /* if (sigio->sio_pgid > 0) */ {
998 		p = sigio->sio_proc;
999 		PROC_LOCK_ASSERT(p, MA_NOTOWNED);
1000 	}
1001 
1002 	SIGIO_LOCK();
1003 	while ((sigio = SLIST_FIRST(sigiolst)) != NULL) {
1004 		*(sigio->sio_myref) = NULL;
1005 		if (pg != NULL) {
1006 			KASSERT(sigio->sio_pgid < 0,
1007 			    ("Proc sigio in pgrp sigio list"));
1008 			KASSERT(sigio->sio_pgrp == pg,
1009 			    ("Bogus pgrp in sigio list"));
1010 			PGRP_LOCK(pg);
1011 			SLIST_REMOVE(&pg->pg_sigiolst, sigio, sigio,
1012 			    sio_pgsigio);
1013 			PGRP_UNLOCK(pg);
1014 		} else /* if (p != NULL) */ {
1015 			KASSERT(sigio->sio_pgid > 0,
1016 			    ("Pgrp sigio in proc sigio list"));
1017 			KASSERT(sigio->sio_proc == p,
1018 			    ("Bogus proc in sigio list"));
1019 			PROC_LOCK(p);
1020 			SLIST_REMOVE(&p->p_sigiolst, sigio, sigio,
1021 			    sio_pgsigio);
1022 			PROC_UNLOCK(p);
1023 		}
1024 		SIGIO_UNLOCK();
1025 		crfree(sigio->sio_ucred);
1026 		free(sigio, M_SIGIO);
1027 		SIGIO_LOCK();
1028 	}
1029 	SIGIO_UNLOCK();
1030 }
1031 
1032 /*
1033  * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg).
1034  *
1035  * After permission checking, add a sigio structure to the sigio list for
1036  * the process or process group.
1037  */
1038 int
1039 fsetown(pid_t pgid, struct sigio **sigiop)
1040 {
1041 	struct proc *proc;
1042 	struct pgrp *pgrp;
1043 	struct sigio *sigio;
1044 	int ret;
1045 
1046 	if (pgid == 0) {
1047 		funsetown(sigiop);
1048 		return (0);
1049 	}
1050 
1051 	ret = 0;
1052 
1053 	/* Allocate and fill in the new sigio out of locks. */
1054 	sigio = malloc(sizeof(struct sigio), M_SIGIO, M_WAITOK);
1055 	sigio->sio_pgid = pgid;
1056 	sigio->sio_ucred = crhold(curthread->td_ucred);
1057 	sigio->sio_myref = sigiop;
1058 
1059 	sx_slock(&proctree_lock);
1060 	if (pgid > 0) {
1061 		proc = pfind(pgid);
1062 		if (proc == NULL) {
1063 			ret = ESRCH;
1064 			goto fail;
1065 		}
1066 
1067 		/*
1068 		 * Policy - Don't allow a process to FSETOWN a process
1069 		 * in another session.
1070 		 *
1071 		 * Remove this test to allow maximum flexibility or
1072 		 * restrict FSETOWN to the current process or process
1073 		 * group for maximum safety.
1074 		 */
1075 		PROC_UNLOCK(proc);
1076 		if (proc->p_session != curthread->td_proc->p_session) {
1077 			ret = EPERM;
1078 			goto fail;
1079 		}
1080 
1081 		pgrp = NULL;
1082 	} else /* if (pgid < 0) */ {
1083 		pgrp = pgfind(-pgid);
1084 		if (pgrp == NULL) {
1085 			ret = ESRCH;
1086 			goto fail;
1087 		}
1088 		PGRP_UNLOCK(pgrp);
1089 
1090 		/*
1091 		 * Policy - Don't allow a process to FSETOWN a process
1092 		 * in another session.
1093 		 *
1094 		 * Remove this test to allow maximum flexibility or
1095 		 * restrict FSETOWN to the current process or process
1096 		 * group for maximum safety.
1097 		 */
1098 		if (pgrp->pg_session != curthread->td_proc->p_session) {
1099 			ret = EPERM;
1100 			goto fail;
1101 		}
1102 
1103 		proc = NULL;
1104 	}
1105 	funsetown(sigiop);
1106 	if (pgid > 0) {
1107 		PROC_LOCK(proc);
1108 		/*
1109 		 * Since funsetownlst() is called without the proctree
1110 		 * locked, we need to check for P_WEXIT.
1111 		 * XXX: is ESRCH correct?
1112 		 */
1113 		if ((proc->p_flag & P_WEXIT) != 0) {
1114 			PROC_UNLOCK(proc);
1115 			ret = ESRCH;
1116 			goto fail;
1117 		}
1118 		SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio);
1119 		sigio->sio_proc = proc;
1120 		PROC_UNLOCK(proc);
1121 	} else {
1122 		PGRP_LOCK(pgrp);
1123 		SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio);
1124 		sigio->sio_pgrp = pgrp;
1125 		PGRP_UNLOCK(pgrp);
1126 	}
1127 	sx_sunlock(&proctree_lock);
1128 	SIGIO_LOCK();
1129 	*sigiop = sigio;
1130 	SIGIO_UNLOCK();
1131 	return (0);
1132 
1133 fail:
1134 	sx_sunlock(&proctree_lock);
1135 	crfree(sigio->sio_ucred);
1136 	free(sigio, M_SIGIO);
1137 	return (ret);
1138 }
1139 
1140 /*
1141  * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg).
1142  */
1143 pid_t
1144 fgetown(sigiop)
1145 	struct sigio **sigiop;
1146 {
1147 	pid_t pgid;
1148 
1149 	SIGIO_LOCK();
1150 	pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0;
1151 	SIGIO_UNLOCK();
1152 	return (pgid);
1153 }
1154 
1155 /*
1156  * Function drops the filedesc lock on return.
1157  */
1158 static int
1159 closefp(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1160     int holdleaders)
1161 {
1162 	int error;
1163 
1164 	FILEDESC_XLOCK_ASSERT(fdp);
1165 
1166 	if (holdleaders) {
1167 		if (td->td_proc->p_fdtol != NULL) {
1168 			/*
1169 			 * Ask fdfree() to sleep to ensure that all relevant
1170 			 * process leaders can be traversed in closef().
1171 			 */
1172 			fdp->fd_holdleaderscount++;
1173 		} else {
1174 			holdleaders = 0;
1175 		}
1176 	}
1177 
1178 	/*
1179 	 * We now hold the fp reference that used to be owned by the
1180 	 * descriptor array.  We have to unlock the FILEDESC *AFTER*
1181 	 * knote_fdclose to prevent a race of the fd getting opened, a knote
1182 	 * added, and deleteing a knote for the new fd.
1183 	 */
1184 	knote_fdclose(td, fd);
1185 
1186 	/*
1187 	 * We need to notify mqueue if the object is of type mqueue.
1188 	 */
1189 	if (fp->f_type == DTYPE_MQUEUE)
1190 		mq_fdclose(td, fd, fp);
1191 	FILEDESC_XUNLOCK(fdp);
1192 
1193 	error = closef(fp, td);
1194 	if (holdleaders) {
1195 		FILEDESC_XLOCK(fdp);
1196 		fdp->fd_holdleaderscount--;
1197 		if (fdp->fd_holdleaderscount == 0 &&
1198 		    fdp->fd_holdleaderswakeup != 0) {
1199 			fdp->fd_holdleaderswakeup = 0;
1200 			wakeup(&fdp->fd_holdleaderscount);
1201 		}
1202 		FILEDESC_XUNLOCK(fdp);
1203 	}
1204 	return (error);
1205 }
1206 
1207 /*
1208  * Close a file descriptor.
1209  */
1210 #ifndef _SYS_SYSPROTO_H_
1211 struct close_args {
1212 	int     fd;
1213 };
1214 #endif
1215 /* ARGSUSED */
1216 int
1217 sys_close(struct thread *td, struct close_args *uap)
1218 {
1219 
1220 	return (kern_close(td, uap->fd));
1221 }
1222 
1223 int
1224 kern_close(struct thread *td, int fd)
1225 {
1226 	struct filedesc *fdp;
1227 	struct file *fp;
1228 
1229 	fdp = td->td_proc->p_fd;
1230 
1231 	AUDIT_SYSCLOSE(td, fd);
1232 
1233 	FILEDESC_XLOCK(fdp);
1234 	if ((fp = fget_locked(fdp, fd)) == NULL) {
1235 		FILEDESC_XUNLOCK(fdp);
1236 		return (EBADF);
1237 	}
1238 	fdfree(fdp, fd);
1239 
1240 	/* closefp() drops the FILEDESC lock for us. */
1241 	return (closefp(fdp, fd, fp, td, 1));
1242 }
1243 
1244 /*
1245  * Close open file descriptors.
1246  */
1247 #ifndef _SYS_SYSPROTO_H_
1248 struct closefrom_args {
1249 	int	lowfd;
1250 };
1251 #endif
1252 /* ARGSUSED */
1253 int
1254 sys_closefrom(struct thread *td, struct closefrom_args *uap)
1255 {
1256 	struct filedesc *fdp;
1257 	int fd;
1258 
1259 	fdp = td->td_proc->p_fd;
1260 	AUDIT_ARG_FD(uap->lowfd);
1261 
1262 	/*
1263 	 * Treat negative starting file descriptor values identical to
1264 	 * closefrom(0) which closes all files.
1265 	 */
1266 	if (uap->lowfd < 0)
1267 		uap->lowfd = 0;
1268 	FILEDESC_SLOCK(fdp);
1269 	for (fd = uap->lowfd; fd <= fdp->fd_lastfile; fd++) {
1270 		if (fdp->fd_ofiles[fd].fde_file != NULL) {
1271 			FILEDESC_SUNLOCK(fdp);
1272 			(void)kern_close(td, fd);
1273 			FILEDESC_SLOCK(fdp);
1274 		}
1275 	}
1276 	FILEDESC_SUNLOCK(fdp);
1277 	return (0);
1278 }
1279 
1280 #if defined(COMPAT_43)
1281 /*
1282  * Return status information about a file descriptor.
1283  */
1284 #ifndef _SYS_SYSPROTO_H_
1285 struct ofstat_args {
1286 	int	fd;
1287 	struct	ostat *sb;
1288 };
1289 #endif
1290 /* ARGSUSED */
1291 int
1292 ofstat(struct thread *td, struct ofstat_args *uap)
1293 {
1294 	struct ostat oub;
1295 	struct stat ub;
1296 	int error;
1297 
1298 	error = kern_fstat(td, uap->fd, &ub);
1299 	if (error == 0) {
1300 		cvtstat(&ub, &oub);
1301 		error = copyout(&oub, uap->sb, sizeof(oub));
1302 	}
1303 	return (error);
1304 }
1305 #endif /* COMPAT_43 */
1306 
1307 /*
1308  * Return status information about a file descriptor.
1309  */
1310 #ifndef _SYS_SYSPROTO_H_
1311 struct fstat_args {
1312 	int	fd;
1313 	struct	stat *sb;
1314 };
1315 #endif
1316 /* ARGSUSED */
1317 int
1318 sys_fstat(struct thread *td, struct fstat_args *uap)
1319 {
1320 	struct stat ub;
1321 	int error;
1322 
1323 	error = kern_fstat(td, uap->fd, &ub);
1324 	if (error == 0)
1325 		error = copyout(&ub, uap->sb, sizeof(ub));
1326 	return (error);
1327 }
1328 
1329 int
1330 kern_fstat(struct thread *td, int fd, struct stat *sbp)
1331 {
1332 	struct file *fp;
1333 	cap_rights_t rights;
1334 	int error;
1335 
1336 	AUDIT_ARG_FD(fd);
1337 
1338 	error = fget(td, fd, cap_rights_init(&rights, CAP_FSTAT), &fp);
1339 	if (error != 0)
1340 		return (error);
1341 
1342 	AUDIT_ARG_FILE(td->td_proc, fp);
1343 
1344 	error = fo_stat(fp, sbp, td->td_ucred, td);
1345 	fdrop(fp, td);
1346 #ifdef KTRACE
1347 	if (error == 0 && KTRPOINT(td, KTR_STRUCT))
1348 		ktrstat(sbp);
1349 #endif
1350 	return (error);
1351 }
1352 
1353 /*
1354  * Return status information about a file descriptor.
1355  */
1356 #ifndef _SYS_SYSPROTO_H_
1357 struct nfstat_args {
1358 	int	fd;
1359 	struct	nstat *sb;
1360 };
1361 #endif
1362 /* ARGSUSED */
1363 int
1364 sys_nfstat(struct thread *td, struct nfstat_args *uap)
1365 {
1366 	struct nstat nub;
1367 	struct stat ub;
1368 	int error;
1369 
1370 	error = kern_fstat(td, uap->fd, &ub);
1371 	if (error == 0) {
1372 		cvtnstat(&ub, &nub);
1373 		error = copyout(&nub, uap->sb, sizeof(nub));
1374 	}
1375 	return (error);
1376 }
1377 
1378 /*
1379  * Return pathconf information about a file descriptor.
1380  */
1381 #ifndef _SYS_SYSPROTO_H_
1382 struct fpathconf_args {
1383 	int	fd;
1384 	int	name;
1385 };
1386 #endif
1387 /* ARGSUSED */
1388 int
1389 sys_fpathconf(struct thread *td, struct fpathconf_args *uap)
1390 {
1391 	struct file *fp;
1392 	struct vnode *vp;
1393 	cap_rights_t rights;
1394 	int error;
1395 
1396 	error = fget(td, uap->fd, cap_rights_init(&rights, CAP_FPATHCONF), &fp);
1397 	if (error != 0)
1398 		return (error);
1399 
1400 	if (uap->name == _PC_ASYNC_IO) {
1401 		td->td_retval[0] = _POSIX_ASYNCHRONOUS_IO;
1402 		goto out;
1403 	}
1404 	vp = fp->f_vnode;
1405 	if (vp != NULL) {
1406 		vn_lock(vp, LK_SHARED | LK_RETRY);
1407 		error = VOP_PATHCONF(vp, uap->name, td->td_retval);
1408 		VOP_UNLOCK(vp, 0);
1409 	} else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) {
1410 		if (uap->name != _PC_PIPE_BUF) {
1411 			error = EINVAL;
1412 		} else {
1413 			td->td_retval[0] = PIPE_BUF;
1414 			error = 0;
1415 		}
1416 	} else {
1417 		error = EOPNOTSUPP;
1418 	}
1419 out:
1420 	fdrop(fp, td);
1421 	return (error);
1422 }
1423 
1424 /*
1425  * Initialize filecaps structure.
1426  */
1427 void
1428 filecaps_init(struct filecaps *fcaps)
1429 {
1430 
1431 	bzero(fcaps, sizeof(*fcaps));
1432 	fcaps->fc_nioctls = -1;
1433 }
1434 
1435 /*
1436  * Copy filecaps structure allocating memory for ioctls array if needed.
1437  *
1438  * The last parameter indicates whether the fdtable is locked. If it is not and
1439  * ioctls are encountered, copying fails and the caller must lock the table.
1440  *
1441  * Note that if the table was not locked, the caller has to check the relevant
1442  * sequence counter to determine whether the operation was successful.
1443  */
1444 int
1445 filecaps_copy(const struct filecaps *src, struct filecaps *dst, bool locked)
1446 {
1447 	size_t size;
1448 
1449 	*dst = *src;
1450 	if (src->fc_ioctls == NULL)
1451 		return (0);
1452 	if (!locked)
1453 		return (1);
1454 
1455 	KASSERT(src->fc_nioctls > 0,
1456 	    ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1457 
1458 	size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1459 	dst->fc_ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1460 	bcopy(src->fc_ioctls, dst->fc_ioctls, size);
1461 	return (0);
1462 }
1463 
1464 /*
1465  * Move filecaps structure to the new place and clear the old place.
1466  */
1467 void
1468 filecaps_move(struct filecaps *src, struct filecaps *dst)
1469 {
1470 
1471 	*dst = *src;
1472 	bzero(src, sizeof(*src));
1473 }
1474 
1475 /*
1476  * Fill the given filecaps structure with full rights.
1477  */
1478 static void
1479 filecaps_fill(struct filecaps *fcaps)
1480 {
1481 
1482 	CAP_ALL(&fcaps->fc_rights);
1483 	fcaps->fc_ioctls = NULL;
1484 	fcaps->fc_nioctls = -1;
1485 	fcaps->fc_fcntls = CAP_FCNTL_ALL;
1486 }
1487 
1488 /*
1489  * Free memory allocated within filecaps structure.
1490  */
1491 void
1492 filecaps_free(struct filecaps *fcaps)
1493 {
1494 
1495 	free(fcaps->fc_ioctls, M_FILECAPS);
1496 	bzero(fcaps, sizeof(*fcaps));
1497 }
1498 
1499 /*
1500  * Validate the given filecaps structure.
1501  */
1502 static void
1503 filecaps_validate(const struct filecaps *fcaps, const char *func)
1504 {
1505 
1506 	KASSERT(cap_rights_is_valid(&fcaps->fc_rights),
1507 	    ("%s: invalid rights", func));
1508 	KASSERT((fcaps->fc_fcntls & ~CAP_FCNTL_ALL) == 0,
1509 	    ("%s: invalid fcntls", func));
1510 	KASSERT(fcaps->fc_fcntls == 0 ||
1511 	    cap_rights_is_set(&fcaps->fc_rights, CAP_FCNTL),
1512 	    ("%s: fcntls without CAP_FCNTL", func));
1513 	KASSERT(fcaps->fc_ioctls != NULL ? fcaps->fc_nioctls > 0 :
1514 	    (fcaps->fc_nioctls == -1 || fcaps->fc_nioctls == 0),
1515 	    ("%s: invalid ioctls", func));
1516 	KASSERT(fcaps->fc_nioctls == 0 ||
1517 	    cap_rights_is_set(&fcaps->fc_rights, CAP_IOCTL),
1518 	    ("%s: ioctls without CAP_IOCTL", func));
1519 }
1520 
1521 static void
1522 fdgrowtable_exp(struct filedesc *fdp, int nfd)
1523 {
1524 	int nfd1;
1525 
1526 	FILEDESC_XLOCK_ASSERT(fdp);
1527 
1528 	nfd1 = fdp->fd_nfiles * 2;
1529 	if (nfd1 < nfd)
1530 		nfd1 = nfd;
1531 	fdgrowtable(fdp, nfd1);
1532 }
1533 
1534 /*
1535  * Grow the file table to accommodate (at least) nfd descriptors.
1536  */
1537 static void
1538 fdgrowtable(struct filedesc *fdp, int nfd)
1539 {
1540 	struct filedesc0 *fdp0;
1541 	struct freetable *ft;
1542 	struct fdescenttbl *ntable;
1543 	struct fdescenttbl *otable;
1544 	int nnfiles, onfiles;
1545 	NDSLOTTYPE *nmap, *omap;
1546 
1547 	/*
1548 	 * If lastfile is -1 this struct filedesc was just allocated and we are
1549 	 * growing it to accommodate for the one we are going to copy from. There
1550 	 * is no need to have a lock on this one as it's not visible to anyone.
1551 	 */
1552 	if (fdp->fd_lastfile != -1)
1553 		FILEDESC_XLOCK_ASSERT(fdp);
1554 
1555 	KASSERT(fdp->fd_nfiles > 0, ("zero-length file table"));
1556 
1557 	/* save old values */
1558 	onfiles = fdp->fd_nfiles;
1559 	otable = fdp->fd_files;
1560 	omap = fdp->fd_map;
1561 
1562 	/* compute the size of the new table */
1563 	nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */
1564 	if (nnfiles <= onfiles)
1565 		/* the table is already large enough */
1566 		return;
1567 
1568 	/*
1569 	 * Allocate a new table.  We need enough space for the number of
1570 	 * entries, file entries themselves and the struct freetable we will use
1571 	 * when we decommission the table and place it on the freelist.
1572 	 * We place the struct freetable in the middle so we don't have
1573 	 * to worry about padding.
1574 	 */
1575 	ntable = malloc(offsetof(struct fdescenttbl, fdt_ofiles) +
1576 	    nnfiles * sizeof(ntable->fdt_ofiles[0]) +
1577 	    sizeof(struct freetable),
1578 	    M_FILEDESC, M_ZERO | M_WAITOK);
1579 	/* copy the old data */
1580 	ntable->fdt_nfiles = nnfiles;
1581 	memcpy(ntable->fdt_ofiles, otable->fdt_ofiles,
1582 	    onfiles * sizeof(ntable->fdt_ofiles[0]));
1583 
1584 	/*
1585 	 * Allocate a new map only if the old is not large enough.  It will
1586 	 * grow at a slower rate than the table as it can map more
1587 	 * entries than the table can hold.
1588 	 */
1589 	if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) {
1590 		nmap = malloc(NDSLOTS(nnfiles) * NDSLOTSIZE, M_FILEDESC,
1591 		    M_ZERO | M_WAITOK);
1592 		/* copy over the old data and update the pointer */
1593 		memcpy(nmap, omap, NDSLOTS(onfiles) * sizeof(*omap));
1594 		fdp->fd_map = nmap;
1595 	}
1596 
1597 	/*
1598 	 * Make sure that ntable is correctly initialized before we replace
1599 	 * fd_files poiner. Otherwise fget_unlocked() may see inconsistent
1600 	 * data.
1601 	 */
1602 	atomic_store_rel_ptr((volatile void *)&fdp->fd_files, (uintptr_t)ntable);
1603 
1604 	/*
1605 	 * Do not free the old file table, as some threads may still
1606 	 * reference entries within it.  Instead, place it on a freelist
1607 	 * which will be processed when the struct filedesc is released.
1608 	 *
1609 	 * Note that if onfiles == NDFILE, we're dealing with the original
1610 	 * static allocation contained within (struct filedesc0 *)fdp,
1611 	 * which must not be freed.
1612 	 */
1613 	if (onfiles > NDFILE) {
1614 		ft = (struct freetable *)&otable->fdt_ofiles[onfiles];
1615 		fdp0 = (struct filedesc0 *)fdp;
1616 		ft->ft_table = otable;
1617 		SLIST_INSERT_HEAD(&fdp0->fd_free, ft, ft_next);
1618 	}
1619 	/*
1620 	 * The map does not have the same possibility of threads still
1621 	 * holding references to it.  So always free it as long as it
1622 	 * does not reference the original static allocation.
1623 	 */
1624 	if (NDSLOTS(onfiles) > NDSLOTS(NDFILE))
1625 		free(omap, M_FILEDESC);
1626 }
1627 
1628 /*
1629  * Allocate a file descriptor for the process.
1630  */
1631 int
1632 fdalloc(struct thread *td, int minfd, int *result)
1633 {
1634 	struct proc *p = td->td_proc;
1635 	struct filedesc *fdp = p->p_fd;
1636 	int fd, maxfd, allocfd;
1637 #ifdef RACCT
1638 	int error;
1639 #endif
1640 
1641 	FILEDESC_XLOCK_ASSERT(fdp);
1642 
1643 	if (fdp->fd_freefile > minfd)
1644 		minfd = fdp->fd_freefile;
1645 
1646 	maxfd = getmaxfd(td);
1647 
1648 	/*
1649 	 * Search the bitmap for a free descriptor starting at minfd.
1650 	 * If none is found, grow the file table.
1651 	 */
1652 	fd = fd_first_free(fdp, minfd, fdp->fd_nfiles);
1653 	if (fd >= maxfd)
1654 		return (EMFILE);
1655 	if (fd >= fdp->fd_nfiles) {
1656 		allocfd = min(fd * 2, maxfd);
1657 #ifdef RACCT
1658 		if (racct_enable) {
1659 			PROC_LOCK(p);
1660 			error = racct_set(p, RACCT_NOFILE, allocfd);
1661 			PROC_UNLOCK(p);
1662 			if (error != 0)
1663 				return (EMFILE);
1664 		}
1665 #endif
1666 		/*
1667 		 * fd is already equal to first free descriptor >= minfd, so
1668 		 * we only need to grow the table and we are done.
1669 		 */
1670 		fdgrowtable_exp(fdp, allocfd);
1671 	}
1672 
1673 	/*
1674 	 * Perform some sanity checks, then mark the file descriptor as
1675 	 * used and return it to the caller.
1676 	 */
1677 	KASSERT(fd >= 0 && fd < min(maxfd, fdp->fd_nfiles),
1678 	    ("invalid descriptor %d", fd));
1679 	KASSERT(!fdisused(fdp, fd),
1680 	    ("fd_first_free() returned non-free descriptor"));
1681 	KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
1682 	    ("file descriptor isn't free"));
1683 	fdused(fdp, fd);
1684 	*result = fd;
1685 	return (0);
1686 }
1687 
1688 /*
1689  * Allocate n file descriptors for the process.
1690  */
1691 int
1692 fdallocn(struct thread *td, int minfd, int *fds, int n)
1693 {
1694 	struct proc *p = td->td_proc;
1695 	struct filedesc *fdp = p->p_fd;
1696 	int i;
1697 
1698 	FILEDESC_XLOCK_ASSERT(fdp);
1699 
1700 	for (i = 0; i < n; i++)
1701 		if (fdalloc(td, 0, &fds[i]) != 0)
1702 			break;
1703 
1704 	if (i < n) {
1705 		for (i--; i >= 0; i--)
1706 			fdunused(fdp, fds[i]);
1707 		return (EMFILE);
1708 	}
1709 
1710 	return (0);
1711 }
1712 
1713 /*
1714  * Create a new open file structure and allocate a file descriptor for the
1715  * process that refers to it.  We add one reference to the file for the
1716  * descriptor table and one reference for resultfp. This is to prevent us
1717  * being preempted and the entry in the descriptor table closed after we
1718  * release the FILEDESC lock.
1719  */
1720 int
1721 falloc_caps(struct thread *td, struct file **resultfp, int *resultfd, int flags,
1722     struct filecaps *fcaps)
1723 {
1724 	struct file *fp;
1725 	int error, fd;
1726 
1727 	error = falloc_noinstall(td, &fp);
1728 	if (error)
1729 		return (error);		/* no reference held on error */
1730 
1731 	error = finstall(td, fp, &fd, flags, fcaps);
1732 	if (error) {
1733 		fdrop(fp, td);		/* one reference (fp only) */
1734 		return (error);
1735 	}
1736 
1737 	if (resultfp != NULL)
1738 		*resultfp = fp;		/* copy out result */
1739 	else
1740 		fdrop(fp, td);		/* release local reference */
1741 
1742 	if (resultfd != NULL)
1743 		*resultfd = fd;
1744 
1745 	return (0);
1746 }
1747 
1748 /*
1749  * Create a new open file structure without allocating a file descriptor.
1750  */
1751 int
1752 falloc_noinstall(struct thread *td, struct file **resultfp)
1753 {
1754 	struct file *fp;
1755 	int maxuserfiles = maxfiles - (maxfiles / 20);
1756 	int openfiles_new;
1757 	static struct timeval lastfail;
1758 	static int curfail;
1759 
1760 	KASSERT(resultfp != NULL, ("%s: resultfp == NULL", __func__));
1761 
1762 	openfiles_new = atomic_fetchadd_int(&openfiles, 1) + 1;
1763 	if ((openfiles_new >= maxuserfiles &&
1764 	    priv_check(td, PRIV_MAXFILES) != 0) ||
1765 	    openfiles_new >= maxfiles) {
1766 		atomic_subtract_int(&openfiles, 1);
1767 		if (ppsratecheck(&lastfail, &curfail, 1)) {
1768 			printf("kern.maxfiles limit exceeded by uid %i, (%s) "
1769 			    "please see tuning(7).\n", td->td_ucred->cr_ruid, td->td_proc->p_comm);
1770 		}
1771 		return (ENFILE);
1772 	}
1773 	fp = uma_zalloc(file_zone, M_WAITOK | M_ZERO);
1774 	refcount_init(&fp->f_count, 1);
1775 	fp->f_cred = crhold(td->td_ucred);
1776 	fp->f_ops = &badfileops;
1777 	*resultfp = fp;
1778 	return (0);
1779 }
1780 
1781 /*
1782  * Install a file in a file descriptor table.
1783  */
1784 void
1785 _finstall(struct filedesc *fdp, struct file *fp, int fd, int flags,
1786     struct filecaps *fcaps)
1787 {
1788 	struct filedescent *fde;
1789 
1790 	MPASS(fp != NULL);
1791 	if (fcaps != NULL)
1792 		filecaps_validate(fcaps, __func__);
1793 	FILEDESC_XLOCK_ASSERT(fdp);
1794 
1795 	fde = &fdp->fd_ofiles[fd];
1796 #ifdef CAPABILITIES
1797 	seq_write_begin(&fde->fde_seq);
1798 #endif
1799 	fde->fde_file = fp;
1800 	fde->fde_flags = (flags & O_CLOEXEC) != 0 ? UF_EXCLOSE : 0;
1801 	if (fcaps != NULL)
1802 		filecaps_move(fcaps, &fde->fde_caps);
1803 	else
1804 		filecaps_fill(&fde->fde_caps);
1805 #ifdef CAPABILITIES
1806 	seq_write_end(&fde->fde_seq);
1807 #endif
1808 }
1809 
1810 int
1811 finstall(struct thread *td, struct file *fp, int *fd, int flags,
1812     struct filecaps *fcaps)
1813 {
1814 	struct filedesc *fdp = td->td_proc->p_fd;
1815 	int error;
1816 
1817 	MPASS(fd != NULL);
1818 
1819 	FILEDESC_XLOCK(fdp);
1820 	if ((error = fdalloc(td, 0, fd))) {
1821 		FILEDESC_XUNLOCK(fdp);
1822 		return (error);
1823 	}
1824 	fhold(fp);
1825 	_finstall(fdp, fp, *fd, flags, fcaps);
1826 	FILEDESC_XUNLOCK(fdp);
1827 	return (0);
1828 }
1829 
1830 /*
1831  * Build a new filedesc structure from another.
1832  * Copy the current, root, and jail root vnode references.
1833  *
1834  * If fdp is not NULL, return with it shared locked.
1835  */
1836 struct filedesc *
1837 fdinit(struct filedesc *fdp, bool prepfiles)
1838 {
1839 	struct filedesc0 *newfdp0;
1840 	struct filedesc *newfdp;
1841 
1842 	newfdp0 = uma_zalloc(filedesc0_zone, M_WAITOK | M_ZERO);
1843 	newfdp = &newfdp0->fd_fd;
1844 
1845 	/* Create the file descriptor table. */
1846 	FILEDESC_LOCK_INIT(newfdp);
1847 	refcount_init(&newfdp->fd_refcnt, 1);
1848 	refcount_init(&newfdp->fd_holdcnt, 1);
1849 	newfdp->fd_cmask = CMASK;
1850 	newfdp->fd_map = newfdp0->fd_dmap;
1851 	newfdp->fd_lastfile = -1;
1852 	newfdp->fd_files = (struct fdescenttbl *)&newfdp0->fd_dfiles;
1853 	newfdp->fd_files->fdt_nfiles = NDFILE;
1854 
1855 	if (fdp == NULL)
1856 		return (newfdp);
1857 
1858 	if (prepfiles && fdp->fd_lastfile >= newfdp->fd_nfiles)
1859 		fdgrowtable(newfdp, fdp->fd_lastfile + 1);
1860 
1861 	FILEDESC_SLOCK(fdp);
1862 	newfdp->fd_cdir = fdp->fd_cdir;
1863 	if (newfdp->fd_cdir)
1864 		vrefact(newfdp->fd_cdir);
1865 	newfdp->fd_rdir = fdp->fd_rdir;
1866 	if (newfdp->fd_rdir)
1867 		vrefact(newfdp->fd_rdir);
1868 	newfdp->fd_jdir = fdp->fd_jdir;
1869 	if (newfdp->fd_jdir)
1870 		vrefact(newfdp->fd_jdir);
1871 
1872 	if (!prepfiles) {
1873 		FILEDESC_SUNLOCK(fdp);
1874 	} else {
1875 		while (fdp->fd_lastfile >= newfdp->fd_nfiles) {
1876 			FILEDESC_SUNLOCK(fdp);
1877 			fdgrowtable(newfdp, fdp->fd_lastfile + 1);
1878 			FILEDESC_SLOCK(fdp);
1879 		}
1880 	}
1881 
1882 	return (newfdp);
1883 }
1884 
1885 static struct filedesc *
1886 fdhold(struct proc *p)
1887 {
1888 	struct filedesc *fdp;
1889 
1890 	PROC_LOCK_ASSERT(p, MA_OWNED);
1891 	fdp = p->p_fd;
1892 	if (fdp != NULL)
1893 		refcount_acquire(&fdp->fd_holdcnt);
1894 	return (fdp);
1895 }
1896 
1897 static void
1898 fddrop(struct filedesc *fdp)
1899 {
1900 
1901 	if (fdp->fd_holdcnt > 1) {
1902 		if (refcount_release(&fdp->fd_holdcnt) == 0)
1903 			return;
1904 	}
1905 
1906 	FILEDESC_LOCK_DESTROY(fdp);
1907 	uma_zfree(filedesc0_zone, fdp);
1908 }
1909 
1910 /*
1911  * Share a filedesc structure.
1912  */
1913 struct filedesc *
1914 fdshare(struct filedesc *fdp)
1915 {
1916 
1917 	refcount_acquire(&fdp->fd_refcnt);
1918 	return (fdp);
1919 }
1920 
1921 /*
1922  * Unshare a filedesc structure, if necessary by making a copy
1923  */
1924 void
1925 fdunshare(struct thread *td)
1926 {
1927 	struct filedesc *tmp;
1928 	struct proc *p = td->td_proc;
1929 
1930 	if (p->p_fd->fd_refcnt == 1)
1931 		return;
1932 
1933 	tmp = fdcopy(p->p_fd);
1934 	fdescfree(td);
1935 	p->p_fd = tmp;
1936 }
1937 
1938 void
1939 fdinstall_remapped(struct thread *td, struct filedesc *fdp)
1940 {
1941 
1942 	fdescfree(td);
1943 	td->td_proc->p_fd = fdp;
1944 }
1945 
1946 /*
1947  * Copy a filedesc structure.  A NULL pointer in returns a NULL reference,
1948  * this is to ease callers, not catch errors.
1949  */
1950 struct filedesc *
1951 fdcopy(struct filedesc *fdp)
1952 {
1953 	struct filedesc *newfdp;
1954 	struct filedescent *nfde, *ofde;
1955 	int i;
1956 
1957 	MPASS(fdp != NULL);
1958 
1959 	newfdp = fdinit(fdp, true);
1960 	/* copy all passable descriptors (i.e. not kqueue) */
1961 	newfdp->fd_freefile = -1;
1962 	for (i = 0; i <= fdp->fd_lastfile; ++i) {
1963 		ofde = &fdp->fd_ofiles[i];
1964 		if (ofde->fde_file == NULL ||
1965 		    (ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0) {
1966 			if (newfdp->fd_freefile == -1)
1967 				newfdp->fd_freefile = i;
1968 			continue;
1969 		}
1970 		nfde = &newfdp->fd_ofiles[i];
1971 		*nfde = *ofde;
1972 		filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true);
1973 		fhold(nfde->fde_file);
1974 		fdused_init(newfdp, i);
1975 		newfdp->fd_lastfile = i;
1976 	}
1977 	if (newfdp->fd_freefile == -1)
1978 		newfdp->fd_freefile = i;
1979 	newfdp->fd_cmask = fdp->fd_cmask;
1980 	FILEDESC_SUNLOCK(fdp);
1981 	return (newfdp);
1982 }
1983 
1984 /*
1985  * Copies a filedesc structure, while remapping all file descriptors
1986  * stored inside using a translation table.
1987  *
1988  * File descriptors are copied over to the new file descriptor table,
1989  * regardless of whether the close-on-exec flag is set.
1990  */
1991 int
1992 fdcopy_remapped(struct filedesc *fdp, const int *fds, size_t nfds,
1993     struct filedesc **ret)
1994 {
1995 	struct filedesc *newfdp;
1996 	struct filedescent *nfde, *ofde;
1997 	int error, i;
1998 
1999 	MPASS(fdp != NULL);
2000 
2001 	newfdp = fdinit(fdp, true);
2002 	if (nfds > fdp->fd_lastfile + 1) {
2003 		/* New table cannot be larger than the old one. */
2004 		error = E2BIG;
2005 		goto bad;
2006 	}
2007 	/* Copy all passable descriptors (i.e. not kqueue). */
2008 	newfdp->fd_freefile = nfds;
2009 	for (i = 0; i < nfds; ++i) {
2010 		if (fds[i] < 0 || fds[i] > fdp->fd_lastfile) {
2011 			/* File descriptor out of bounds. */
2012 			error = EBADF;
2013 			goto bad;
2014 		}
2015 		ofde = &fdp->fd_ofiles[fds[i]];
2016 		if (ofde->fde_file == NULL) {
2017 			/* Unused file descriptor. */
2018 			error = EBADF;
2019 			goto bad;
2020 		}
2021 		if ((ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0) {
2022 			/* File descriptor cannot be passed. */
2023 			error = EINVAL;
2024 			goto bad;
2025 		}
2026 		nfde = &newfdp->fd_ofiles[i];
2027 		*nfde = *ofde;
2028 		filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true);
2029 		fhold(nfde->fde_file);
2030 		fdused_init(newfdp, i);
2031 		newfdp->fd_lastfile = i;
2032 	}
2033 	newfdp->fd_cmask = fdp->fd_cmask;
2034 	FILEDESC_SUNLOCK(fdp);
2035 	*ret = newfdp;
2036 	return (0);
2037 bad:
2038 	FILEDESC_SUNLOCK(fdp);
2039 	fdescfree_remapped(newfdp);
2040 	return (error);
2041 }
2042 
2043 /*
2044  * Clear POSIX style locks. This is only used when fdp looses a reference (i.e.
2045  * one of processes using it exits) and the table used to be shared.
2046  */
2047 static void
2048 fdclearlocks(struct thread *td)
2049 {
2050 	struct filedesc *fdp;
2051 	struct filedesc_to_leader *fdtol;
2052 	struct flock lf;
2053 	struct file *fp;
2054 	struct proc *p;
2055 	struct vnode *vp;
2056 	int i;
2057 
2058 	p = td->td_proc;
2059 	fdp = p->p_fd;
2060 	fdtol = p->p_fdtol;
2061 	MPASS(fdtol != NULL);
2062 
2063 	FILEDESC_XLOCK(fdp);
2064 	KASSERT(fdtol->fdl_refcount > 0,
2065 	    ("filedesc_to_refcount botch: fdl_refcount=%d",
2066 	    fdtol->fdl_refcount));
2067 	if (fdtol->fdl_refcount == 1 &&
2068 	    (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2069 		for (i = 0; i <= fdp->fd_lastfile; i++) {
2070 			fp = fdp->fd_ofiles[i].fde_file;
2071 			if (fp == NULL || fp->f_type != DTYPE_VNODE)
2072 				continue;
2073 			fhold(fp);
2074 			FILEDESC_XUNLOCK(fdp);
2075 			lf.l_whence = SEEK_SET;
2076 			lf.l_start = 0;
2077 			lf.l_len = 0;
2078 			lf.l_type = F_UNLCK;
2079 			vp = fp->f_vnode;
2080 			(void) VOP_ADVLOCK(vp,
2081 			    (caddr_t)p->p_leader, F_UNLCK,
2082 			    &lf, F_POSIX);
2083 			FILEDESC_XLOCK(fdp);
2084 			fdrop(fp, td);
2085 		}
2086 	}
2087 retry:
2088 	if (fdtol->fdl_refcount == 1) {
2089 		if (fdp->fd_holdleaderscount > 0 &&
2090 		    (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2091 			/*
2092 			 * close() or kern_dup() has cleared a reference
2093 			 * in a shared file descriptor table.
2094 			 */
2095 			fdp->fd_holdleaderswakeup = 1;
2096 			sx_sleep(&fdp->fd_holdleaderscount,
2097 			    FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0);
2098 			goto retry;
2099 		}
2100 		if (fdtol->fdl_holdcount > 0) {
2101 			/*
2102 			 * Ensure that fdtol->fdl_leader remains
2103 			 * valid in closef().
2104 			 */
2105 			fdtol->fdl_wakeup = 1;
2106 			sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK,
2107 			    "fdlhold", 0);
2108 			goto retry;
2109 		}
2110 	}
2111 	fdtol->fdl_refcount--;
2112 	if (fdtol->fdl_refcount == 0 &&
2113 	    fdtol->fdl_holdcount == 0) {
2114 		fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
2115 		fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
2116 	} else
2117 		fdtol = NULL;
2118 	p->p_fdtol = NULL;
2119 	FILEDESC_XUNLOCK(fdp);
2120 	if (fdtol != NULL)
2121 		free(fdtol, M_FILEDESC_TO_LEADER);
2122 }
2123 
2124 /*
2125  * Release a filedesc structure.
2126  */
2127 static void
2128 fdescfree_fds(struct thread *td, struct filedesc *fdp, bool needclose)
2129 {
2130 	struct filedesc0 *fdp0;
2131 	struct freetable *ft, *tft;
2132 	struct filedescent *fde;
2133 	struct file *fp;
2134 	int i;
2135 
2136 	for (i = 0; i <= fdp->fd_lastfile; i++) {
2137 		fde = &fdp->fd_ofiles[i];
2138 		fp = fde->fde_file;
2139 		if (fp != NULL) {
2140 			fdefree_last(fde);
2141 			if (needclose)
2142 				(void) closef(fp, td);
2143 			else
2144 				fdrop(fp, td);
2145 		}
2146 	}
2147 
2148 	if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE))
2149 		free(fdp->fd_map, M_FILEDESC);
2150 	if (fdp->fd_nfiles > NDFILE)
2151 		free(fdp->fd_files, M_FILEDESC);
2152 
2153 	fdp0 = (struct filedesc0 *)fdp;
2154 	SLIST_FOREACH_SAFE(ft, &fdp0->fd_free, ft_next, tft)
2155 		free(ft->ft_table, M_FILEDESC);
2156 
2157 	fddrop(fdp);
2158 }
2159 
2160 void
2161 fdescfree(struct thread *td)
2162 {
2163 	struct proc *p;
2164 	struct filedesc *fdp;
2165 	struct vnode *cdir, *jdir, *rdir;
2166 
2167 	p = td->td_proc;
2168 	fdp = p->p_fd;
2169 	MPASS(fdp != NULL);
2170 
2171 #ifdef RACCT
2172 	if (racct_enable) {
2173 		PROC_LOCK(p);
2174 		racct_set(p, RACCT_NOFILE, 0);
2175 		PROC_UNLOCK(p);
2176 	}
2177 #endif
2178 
2179 	if (p->p_fdtol != NULL)
2180 		fdclearlocks(td);
2181 
2182 	PROC_LOCK(p);
2183 	p->p_fd = NULL;
2184 	PROC_UNLOCK(p);
2185 
2186 	if (refcount_release(&fdp->fd_refcnt) == 0)
2187 		return;
2188 
2189 	FILEDESC_XLOCK(fdp);
2190 	cdir = fdp->fd_cdir;
2191 	fdp->fd_cdir = NULL;
2192 	rdir = fdp->fd_rdir;
2193 	fdp->fd_rdir = NULL;
2194 	jdir = fdp->fd_jdir;
2195 	fdp->fd_jdir = NULL;
2196 	FILEDESC_XUNLOCK(fdp);
2197 
2198 	if (cdir != NULL)
2199 		vrele(cdir);
2200 	if (rdir != NULL)
2201 		vrele(rdir);
2202 	if (jdir != NULL)
2203 		vrele(jdir);
2204 
2205 	fdescfree_fds(td, fdp, 1);
2206 }
2207 
2208 void
2209 fdescfree_remapped(struct filedesc *fdp)
2210 {
2211 
2212 	if (fdp->fd_cdir != NULL)
2213 		vrele(fdp->fd_cdir);
2214 	if (fdp->fd_rdir != NULL)
2215 		vrele(fdp->fd_rdir);
2216 	if (fdp->fd_jdir != NULL)
2217 		vrele(fdp->fd_jdir);
2218 
2219 	fdescfree_fds(curthread, fdp, 0);
2220 }
2221 
2222 /*
2223  * For setugid programs, we don't want to people to use that setugidness
2224  * to generate error messages which write to a file which otherwise would
2225  * otherwise be off-limits to the process.  We check for filesystems where
2226  * the vnode can change out from under us after execve (like [lin]procfs).
2227  *
2228  * Since fdsetugidsafety calls this only for fd 0, 1 and 2, this check is
2229  * sufficient.  We also don't check for setugidness since we know we are.
2230  */
2231 static bool
2232 is_unsafe(struct file *fp)
2233 {
2234 	struct vnode *vp;
2235 
2236 	if (fp->f_type != DTYPE_VNODE)
2237 		return (false);
2238 
2239 	vp = fp->f_vnode;
2240 	return ((vp->v_vflag & VV_PROCDEP) != 0);
2241 }
2242 
2243 /*
2244  * Make this setguid thing safe, if at all possible.
2245  */
2246 void
2247 fdsetugidsafety(struct thread *td)
2248 {
2249 	struct filedesc *fdp;
2250 	struct file *fp;
2251 	int i;
2252 
2253 	fdp = td->td_proc->p_fd;
2254 	KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2255 	MPASS(fdp->fd_nfiles >= 3);
2256 	for (i = 0; i <= 2; i++) {
2257 		fp = fdp->fd_ofiles[i].fde_file;
2258 		if (fp != NULL && is_unsafe(fp)) {
2259 			FILEDESC_XLOCK(fdp);
2260 			knote_fdclose(td, i);
2261 			/*
2262 			 * NULL-out descriptor prior to close to avoid
2263 			 * a race while close blocks.
2264 			 */
2265 			fdfree(fdp, i);
2266 			FILEDESC_XUNLOCK(fdp);
2267 			(void) closef(fp, td);
2268 		}
2269 	}
2270 }
2271 
2272 /*
2273  * If a specific file object occupies a specific file descriptor, close the
2274  * file descriptor entry and drop a reference on the file object.  This is a
2275  * convenience function to handle a subsequent error in a function that calls
2276  * falloc() that handles the race that another thread might have closed the
2277  * file descriptor out from under the thread creating the file object.
2278  */
2279 void
2280 fdclose(struct thread *td, struct file *fp, int idx)
2281 {
2282 	struct filedesc *fdp = td->td_proc->p_fd;
2283 
2284 	FILEDESC_XLOCK(fdp);
2285 	if (fdp->fd_ofiles[idx].fde_file == fp) {
2286 		fdfree(fdp, idx);
2287 		FILEDESC_XUNLOCK(fdp);
2288 		fdrop(fp, td);
2289 	} else
2290 		FILEDESC_XUNLOCK(fdp);
2291 }
2292 
2293 /*
2294  * Close any files on exec?
2295  */
2296 void
2297 fdcloseexec(struct thread *td)
2298 {
2299 	struct filedesc *fdp;
2300 	struct filedescent *fde;
2301 	struct file *fp;
2302 	int i;
2303 
2304 	fdp = td->td_proc->p_fd;
2305 	KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2306 	for (i = 0; i <= fdp->fd_lastfile; i++) {
2307 		fde = &fdp->fd_ofiles[i];
2308 		fp = fde->fde_file;
2309 		if (fp != NULL && (fp->f_type == DTYPE_MQUEUE ||
2310 		    (fde->fde_flags & UF_EXCLOSE))) {
2311 			FILEDESC_XLOCK(fdp);
2312 			fdfree(fdp, i);
2313 			(void) closefp(fdp, i, fp, td, 0);
2314 			FILEDESC_UNLOCK_ASSERT(fdp);
2315 		}
2316 	}
2317 }
2318 
2319 /*
2320  * It is unsafe for set[ug]id processes to be started with file
2321  * descriptors 0..2 closed, as these descriptors are given implicit
2322  * significance in the Standard C library.  fdcheckstd() will create a
2323  * descriptor referencing /dev/null for each of stdin, stdout, and
2324  * stderr that is not already open.
2325  */
2326 int
2327 fdcheckstd(struct thread *td)
2328 {
2329 	struct filedesc *fdp;
2330 	register_t save;
2331 	int i, error, devnull;
2332 
2333 	fdp = td->td_proc->p_fd;
2334 	KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2335 	MPASS(fdp->fd_nfiles >= 3);
2336 	devnull = -1;
2337 	for (i = 0; i <= 2; i++) {
2338 		if (fdp->fd_ofiles[i].fde_file != NULL)
2339 			continue;
2340 
2341 		save = td->td_retval[0];
2342 		if (devnull != -1) {
2343 			error = kern_dup(td, FDDUP_FIXED, 0, devnull, i);
2344 		} else {
2345 			error = kern_openat(td, AT_FDCWD, "/dev/null",
2346 			    UIO_SYSSPACE, O_RDWR, 0);
2347 			if (error == 0) {
2348 				devnull = td->td_retval[0];
2349 				KASSERT(devnull == i, ("we didn't get our fd"));
2350 			}
2351 		}
2352 		td->td_retval[0] = save;
2353 		if (error != 0)
2354 			return (error);
2355 	}
2356 	return (0);
2357 }
2358 
2359 /*
2360  * Internal form of close.  Decrement reference count on file structure.
2361  * Note: td may be NULL when closing a file that was being passed in a
2362  * message.
2363  *
2364  * XXXRW: Giant is not required for the caller, but often will be held; this
2365  * makes it moderately likely the Giant will be recursed in the VFS case.
2366  */
2367 int
2368 closef(struct file *fp, struct thread *td)
2369 {
2370 	struct vnode *vp;
2371 	struct flock lf;
2372 	struct filedesc_to_leader *fdtol;
2373 	struct filedesc *fdp;
2374 
2375 	/*
2376 	 * POSIX record locking dictates that any close releases ALL
2377 	 * locks owned by this process.  This is handled by setting
2378 	 * a flag in the unlock to free ONLY locks obeying POSIX
2379 	 * semantics, and not to free BSD-style file locks.
2380 	 * If the descriptor was in a message, POSIX-style locks
2381 	 * aren't passed with the descriptor, and the thread pointer
2382 	 * will be NULL.  Callers should be careful only to pass a
2383 	 * NULL thread pointer when there really is no owning
2384 	 * context that might have locks, or the locks will be
2385 	 * leaked.
2386 	 */
2387 	if (fp->f_type == DTYPE_VNODE && td != NULL) {
2388 		vp = fp->f_vnode;
2389 		if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
2390 			lf.l_whence = SEEK_SET;
2391 			lf.l_start = 0;
2392 			lf.l_len = 0;
2393 			lf.l_type = F_UNLCK;
2394 			(void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader,
2395 			    F_UNLCK, &lf, F_POSIX);
2396 		}
2397 		fdtol = td->td_proc->p_fdtol;
2398 		if (fdtol != NULL) {
2399 			/*
2400 			 * Handle special case where file descriptor table is
2401 			 * shared between multiple process leaders.
2402 			 */
2403 			fdp = td->td_proc->p_fd;
2404 			FILEDESC_XLOCK(fdp);
2405 			for (fdtol = fdtol->fdl_next;
2406 			    fdtol != td->td_proc->p_fdtol;
2407 			    fdtol = fdtol->fdl_next) {
2408 				if ((fdtol->fdl_leader->p_flag &
2409 				    P_ADVLOCK) == 0)
2410 					continue;
2411 				fdtol->fdl_holdcount++;
2412 				FILEDESC_XUNLOCK(fdp);
2413 				lf.l_whence = SEEK_SET;
2414 				lf.l_start = 0;
2415 				lf.l_len = 0;
2416 				lf.l_type = F_UNLCK;
2417 				vp = fp->f_vnode;
2418 				(void) VOP_ADVLOCK(vp,
2419 				    (caddr_t)fdtol->fdl_leader, F_UNLCK, &lf,
2420 				    F_POSIX);
2421 				FILEDESC_XLOCK(fdp);
2422 				fdtol->fdl_holdcount--;
2423 				if (fdtol->fdl_holdcount == 0 &&
2424 				    fdtol->fdl_wakeup != 0) {
2425 					fdtol->fdl_wakeup = 0;
2426 					wakeup(fdtol);
2427 				}
2428 			}
2429 			FILEDESC_XUNLOCK(fdp);
2430 		}
2431 	}
2432 	return (fdrop(fp, td));
2433 }
2434 
2435 /*
2436  * Initialize the file pointer with the specified properties.
2437  *
2438  * The ops are set with release semantics to be certain that the flags, type,
2439  * and data are visible when ops is.  This is to prevent ops methods from being
2440  * called with bad data.
2441  */
2442 void
2443 finit(struct file *fp, u_int flag, short type, void *data, struct fileops *ops)
2444 {
2445 	fp->f_data = data;
2446 	fp->f_flag = flag;
2447 	fp->f_type = type;
2448 	atomic_store_rel_ptr((volatile uintptr_t *)&fp->f_ops, (uintptr_t)ops);
2449 }
2450 
2451 int
2452 fget_cap_locked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
2453     struct file **fpp, struct filecaps *havecapsp)
2454 {
2455 	struct filedescent *fde;
2456 	int error;
2457 
2458 	FILEDESC_LOCK_ASSERT(fdp);
2459 
2460 	fde = fdeget_locked(fdp, fd);
2461 	if (fde == NULL) {
2462 		error = EBADF;
2463 		goto out;
2464 	}
2465 
2466 #ifdef CAPABILITIES
2467 	error = cap_check(cap_rights_fde(fde), needrightsp);
2468 	if (error != 0)
2469 		goto out;
2470 #endif
2471 
2472 	if (havecapsp != NULL)
2473 		filecaps_copy(&fde->fde_caps, havecapsp, true);
2474 
2475 	*fpp = fde->fde_file;
2476 
2477 	error = 0;
2478 out:
2479 	return (error);
2480 }
2481 
2482 int
2483 fget_cap(struct thread *td, int fd, cap_rights_t *needrightsp,
2484     struct file **fpp, struct filecaps *havecapsp)
2485 {
2486 	struct filedesc *fdp = td->td_proc->p_fd;
2487 	int error;
2488 #ifndef CAPABILITIES
2489 	error = fget_unlocked(fdp, fd, needrightsp, fpp, NULL);
2490 	if (error == 0 && havecapsp != NULL)
2491 		filecaps_fill(havecapsp);
2492 #else
2493 	struct file *fp;
2494 	seq_t seq;
2495 
2496 	for (;;) {
2497 		error = fget_unlocked(fdp, fd, needrightsp, &fp, &seq);
2498 		if (error != 0)
2499 			return (error);
2500 
2501 		if (havecapsp != NULL) {
2502 			if (!filecaps_copy(&fdp->fd_ofiles[fd].fde_caps,
2503 			    havecapsp, false)) {
2504 				fdrop(fp, td);
2505 				goto get_locked;
2506 			}
2507 		}
2508 
2509 		if (!fd_modified(fdp, fd, seq))
2510 			break;
2511 		fdrop(fp, td);
2512 	}
2513 
2514 	*fpp = fp;
2515 	return (0);
2516 
2517 get_locked:
2518 	FILEDESC_SLOCK(fdp);
2519 	error = fget_cap_locked(fdp, fd, needrightsp, fpp, havecapsp);
2520 	if (error == 0)
2521 		fhold(*fpp);
2522 	FILEDESC_SUNLOCK(fdp);
2523 #endif
2524 	return (error);
2525 }
2526 
2527 int
2528 fget_unlocked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
2529     struct file **fpp, seq_t *seqp)
2530 {
2531 #ifdef CAPABILITIES
2532 	struct filedescent *fde;
2533 #endif
2534 	struct fdescenttbl *fdt;
2535 	struct file *fp;
2536 	u_int count;
2537 #ifdef CAPABILITIES
2538 	seq_t seq;
2539 	cap_rights_t haverights;
2540 	int error;
2541 #endif
2542 
2543 	fdt = fdp->fd_files;
2544 	if ((u_int)fd >= fdt->fdt_nfiles)
2545 		return (EBADF);
2546 	/*
2547 	 * Fetch the descriptor locklessly.  We avoid fdrop() races by
2548 	 * never raising a refcount above 0.  To accomplish this we have
2549 	 * to use a cmpset loop rather than an atomic_add.  The descriptor
2550 	 * must be re-verified once we acquire a reference to be certain
2551 	 * that the identity is still correct and we did not lose a race
2552 	 * due to preemption.
2553 	 */
2554 	for (;;) {
2555 #ifdef CAPABILITIES
2556 		seq = seq_read(fd_seq(fdt, fd));
2557 		fde = &fdt->fdt_ofiles[fd];
2558 		haverights = *cap_rights_fde(fde);
2559 		fp = fde->fde_file;
2560 		if (!seq_consistent(fd_seq(fdt, fd), seq))
2561 			continue;
2562 #else
2563 		fp = fdt->fdt_ofiles[fd].fde_file;
2564 #endif
2565 		if (fp == NULL)
2566 			return (EBADF);
2567 #ifdef CAPABILITIES
2568 		error = cap_check(&haverights, needrightsp);
2569 		if (error != 0)
2570 			return (error);
2571 #endif
2572 		count = fp->f_count;
2573 	retry:
2574 		if (count == 0) {
2575 			/*
2576 			 * Force a reload. Other thread could reallocate the
2577 			 * table before this fd was closed, so it possible that
2578 			 * there is a stale fp pointer in cached version.
2579 			 */
2580 			fdt = *(struct fdescenttbl * volatile *)&(fdp->fd_files);
2581 			continue;
2582 		}
2583 		/*
2584 		 * Use an acquire barrier to force re-reading of fdt so it is
2585 		 * refreshed for verification.
2586 		 */
2587 		if (atomic_fcmpset_acq_int(&fp->f_count, &count, count + 1) == 0)
2588 			goto retry;
2589 		fdt = fdp->fd_files;
2590 #ifdef	CAPABILITIES
2591 		if (seq_consistent_nomb(fd_seq(fdt, fd), seq))
2592 #else
2593 		if (fp == fdt->fdt_ofiles[fd].fde_file)
2594 #endif
2595 			break;
2596 		fdrop(fp, curthread);
2597 	}
2598 	*fpp = fp;
2599 	if (seqp != NULL) {
2600 #ifdef CAPABILITIES
2601 		*seqp = seq;
2602 #endif
2603 	}
2604 	return (0);
2605 }
2606 
2607 /*
2608  * Extract the file pointer associated with the specified descriptor for the
2609  * current user process.
2610  *
2611  * If the descriptor doesn't exist or doesn't match 'flags', EBADF is
2612  * returned.
2613  *
2614  * File's rights will be checked against the capability rights mask.
2615  *
2616  * If an error occurred the non-zero error is returned and *fpp is set to
2617  * NULL.  Otherwise *fpp is held and set and zero is returned.  Caller is
2618  * responsible for fdrop().
2619  */
2620 static __inline int
2621 _fget(struct thread *td, int fd, struct file **fpp, int flags,
2622     cap_rights_t *needrightsp, seq_t *seqp)
2623 {
2624 	struct filedesc *fdp;
2625 	struct file *fp;
2626 	int error;
2627 
2628 	*fpp = NULL;
2629 	fdp = td->td_proc->p_fd;
2630 	error = fget_unlocked(fdp, fd, needrightsp, &fp, seqp);
2631 	if (error != 0)
2632 		return (error);
2633 	if (fp->f_ops == &badfileops) {
2634 		fdrop(fp, td);
2635 		return (EBADF);
2636 	}
2637 
2638 	/*
2639 	 * FREAD and FWRITE failure return EBADF as per POSIX.
2640 	 */
2641 	error = 0;
2642 	switch (flags) {
2643 	case FREAD:
2644 	case FWRITE:
2645 		if ((fp->f_flag & flags) == 0)
2646 			error = EBADF;
2647 		break;
2648 	case FEXEC:
2649 	    	if ((fp->f_flag & (FREAD | FEXEC)) == 0 ||
2650 		    ((fp->f_flag & FWRITE) != 0))
2651 			error = EBADF;
2652 		break;
2653 	case 0:
2654 		break;
2655 	default:
2656 		KASSERT(0, ("wrong flags"));
2657 	}
2658 
2659 	if (error != 0) {
2660 		fdrop(fp, td);
2661 		return (error);
2662 	}
2663 
2664 	*fpp = fp;
2665 	return (0);
2666 }
2667 
2668 int
2669 fget(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
2670 {
2671 
2672 	return (_fget(td, fd, fpp, 0, rightsp, NULL));
2673 }
2674 
2675 int
2676 fget_mmap(struct thread *td, int fd, cap_rights_t *rightsp, u_char *maxprotp,
2677     struct file **fpp)
2678 {
2679 	int error;
2680 #ifndef CAPABILITIES
2681 	error = _fget(td, fd, fpp, 0, rightsp, NULL);
2682 	if (maxprotp != NULL)
2683 		*maxprotp = VM_PROT_ALL;
2684 #else
2685 	struct filedesc *fdp = td->td_proc->p_fd;
2686 	seq_t seq;
2687 
2688 	MPASS(cap_rights_is_set(rightsp, CAP_MMAP));
2689 	for (;;) {
2690 		error = _fget(td, fd, fpp, 0, rightsp, &seq);
2691 		if (error != 0)
2692 			return (error);
2693 		/*
2694 		 * If requested, convert capability rights to access flags.
2695 		 */
2696 		if (maxprotp != NULL)
2697 			*maxprotp = cap_rights_to_vmprot(cap_rights(fdp, fd));
2698 		if (!fd_modified(fdp, fd, seq))
2699 			break;
2700 		fdrop(*fpp, td);
2701 	}
2702 #endif
2703 	return (error);
2704 }
2705 
2706 int
2707 fget_read(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
2708 {
2709 
2710 	return (_fget(td, fd, fpp, FREAD, rightsp, NULL));
2711 }
2712 
2713 int
2714 fget_write(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
2715 {
2716 
2717 	return (_fget(td, fd, fpp, FWRITE, rightsp, NULL));
2718 }
2719 
2720 int
2721 fget_fcntl(struct thread *td, int fd, cap_rights_t *rightsp, int needfcntl,
2722     struct file **fpp)
2723 {
2724 	struct filedesc *fdp = td->td_proc->p_fd;
2725 #ifndef CAPABILITIES
2726 	return (fget_unlocked(fdp, fd, rightsp, fpp, NULL));
2727 #else
2728 	int error;
2729 	seq_t seq;
2730 
2731 	MPASS(cap_rights_is_set(rightsp, CAP_FCNTL));
2732 	for (;;) {
2733 		error = fget_unlocked(fdp, fd, rightsp, fpp, &seq);
2734 		if (error != 0)
2735 			return (error);
2736 		error = cap_fcntl_check(fdp, fd, needfcntl);
2737 		if (!fd_modified(fdp, fd, seq))
2738 			break;
2739 		fdrop(*fpp, td);
2740 	}
2741 	if (error != 0) {
2742 		fdrop(*fpp, td);
2743 		*fpp = NULL;
2744 	}
2745 	return (error);
2746 #endif
2747 }
2748 
2749 /*
2750  * Like fget() but loads the underlying vnode, or returns an error if the
2751  * descriptor does not represent a vnode.  Note that pipes use vnodes but
2752  * never have VM objects.  The returned vnode will be vref()'d.
2753  *
2754  * XXX: what about the unused flags ?
2755  */
2756 static __inline int
2757 _fgetvp(struct thread *td, int fd, int flags, cap_rights_t *needrightsp,
2758     struct vnode **vpp)
2759 {
2760 	struct file *fp;
2761 	int error;
2762 
2763 	*vpp = NULL;
2764 	error = _fget(td, fd, &fp, flags, needrightsp, NULL);
2765 	if (error != 0)
2766 		return (error);
2767 	if (fp->f_vnode == NULL) {
2768 		error = EINVAL;
2769 	} else {
2770 		*vpp = fp->f_vnode;
2771 		vrefact(*vpp);
2772 	}
2773 	fdrop(fp, td);
2774 
2775 	return (error);
2776 }
2777 
2778 int
2779 fgetvp(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
2780 {
2781 
2782 	return (_fgetvp(td, fd, 0, rightsp, vpp));
2783 }
2784 
2785 int
2786 fgetvp_rights(struct thread *td, int fd, cap_rights_t *needrightsp,
2787     struct filecaps *havecaps, struct vnode **vpp)
2788 {
2789 	struct filedesc *fdp;
2790 	struct filecaps caps;
2791 	struct file *fp;
2792 	int error;
2793 
2794 	fdp = td->td_proc->p_fd;
2795 	error = fget_cap_locked(fdp, fd, needrightsp, &fp, &caps);
2796 	if (error != 0)
2797 		return (error);
2798 	if (fp->f_ops == &badfileops) {
2799 		error = EBADF;
2800 		goto out;
2801 	}
2802 	if (fp->f_vnode == NULL) {
2803 		error = EINVAL;
2804 		goto out;
2805 	}
2806 
2807 	*havecaps = caps;
2808 	*vpp = fp->f_vnode;
2809 	vrefact(*vpp);
2810 
2811 	return (0);
2812 out:
2813 	filecaps_free(&caps);
2814 	return (error);
2815 }
2816 
2817 int
2818 fgetvp_read(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
2819 {
2820 
2821 	return (_fgetvp(td, fd, FREAD, rightsp, vpp));
2822 }
2823 
2824 int
2825 fgetvp_exec(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
2826 {
2827 
2828 	return (_fgetvp(td, fd, FEXEC, rightsp, vpp));
2829 }
2830 
2831 #ifdef notyet
2832 int
2833 fgetvp_write(struct thread *td, int fd, cap_rights_t *rightsp,
2834     struct vnode **vpp)
2835 {
2836 
2837 	return (_fgetvp(td, fd, FWRITE, rightsp, vpp));
2838 }
2839 #endif
2840 
2841 /*
2842  * Handle the last reference to a file being closed.
2843  */
2844 int
2845 _fdrop(struct file *fp, struct thread *td)
2846 {
2847 	int error;
2848 
2849 	if (fp->f_count != 0)
2850 		panic("fdrop: count %d", fp->f_count);
2851 	error = fo_close(fp, td);
2852 	atomic_subtract_int(&openfiles, 1);
2853 	crfree(fp->f_cred);
2854 	free(fp->f_advice, M_FADVISE);
2855 	uma_zfree(file_zone, fp);
2856 
2857 	return (error);
2858 }
2859 
2860 /*
2861  * Apply an advisory lock on a file descriptor.
2862  *
2863  * Just attempt to get a record lock of the requested type on the entire file
2864  * (l_whence = SEEK_SET, l_start = 0, l_len = 0).
2865  */
2866 #ifndef _SYS_SYSPROTO_H_
2867 struct flock_args {
2868 	int	fd;
2869 	int	how;
2870 };
2871 #endif
2872 /* ARGSUSED */
2873 int
2874 sys_flock(struct thread *td, struct flock_args *uap)
2875 {
2876 	struct file *fp;
2877 	struct vnode *vp;
2878 	struct flock lf;
2879 	cap_rights_t rights;
2880 	int error;
2881 
2882 	error = fget(td, uap->fd, cap_rights_init(&rights, CAP_FLOCK), &fp);
2883 	if (error != 0)
2884 		return (error);
2885 	if (fp->f_type != DTYPE_VNODE) {
2886 		fdrop(fp, td);
2887 		return (EOPNOTSUPP);
2888 	}
2889 
2890 	vp = fp->f_vnode;
2891 	lf.l_whence = SEEK_SET;
2892 	lf.l_start = 0;
2893 	lf.l_len = 0;
2894 	if (uap->how & LOCK_UN) {
2895 		lf.l_type = F_UNLCK;
2896 		atomic_clear_int(&fp->f_flag, FHASLOCK);
2897 		error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK);
2898 		goto done2;
2899 	}
2900 	if (uap->how & LOCK_EX)
2901 		lf.l_type = F_WRLCK;
2902 	else if (uap->how & LOCK_SH)
2903 		lf.l_type = F_RDLCK;
2904 	else {
2905 		error = EBADF;
2906 		goto done2;
2907 	}
2908 	atomic_set_int(&fp->f_flag, FHASLOCK);
2909 	error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf,
2910 	    (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT);
2911 done2:
2912 	fdrop(fp, td);
2913 	return (error);
2914 }
2915 /*
2916  * Duplicate the specified descriptor to a free descriptor.
2917  */
2918 int
2919 dupfdopen(struct thread *td, struct filedesc *fdp, int dfd, int mode,
2920     int openerror, int *indxp)
2921 {
2922 	struct filedescent *newfde, *oldfde;
2923 	struct file *fp;
2924 	int error, indx;
2925 
2926 	KASSERT(openerror == ENODEV || openerror == ENXIO,
2927 	    ("unexpected error %d in %s", openerror, __func__));
2928 
2929 	/*
2930 	 * If the to-be-dup'd fd number is greater than the allowed number
2931 	 * of file descriptors, or the fd to be dup'd has already been
2932 	 * closed, then reject.
2933 	 */
2934 	FILEDESC_XLOCK(fdp);
2935 	if ((fp = fget_locked(fdp, dfd)) == NULL) {
2936 		FILEDESC_XUNLOCK(fdp);
2937 		return (EBADF);
2938 	}
2939 
2940 	error = fdalloc(td, 0, &indx);
2941 	if (error != 0) {
2942 		FILEDESC_XUNLOCK(fdp);
2943 		return (error);
2944 	}
2945 
2946 	/*
2947 	 * There are two cases of interest here.
2948 	 *
2949 	 * For ENODEV simply dup (dfd) to file descriptor (indx) and return.
2950 	 *
2951 	 * For ENXIO steal away the file structure from (dfd) and store it in
2952 	 * (indx).  (dfd) is effectively closed by this operation.
2953 	 */
2954 	switch (openerror) {
2955 	case ENODEV:
2956 		/*
2957 		 * Check that the mode the file is being opened for is a
2958 		 * subset of the mode of the existing descriptor.
2959 		 */
2960 		if (((mode & (FREAD|FWRITE)) | fp->f_flag) != fp->f_flag) {
2961 			fdunused(fdp, indx);
2962 			FILEDESC_XUNLOCK(fdp);
2963 			return (EACCES);
2964 		}
2965 		fhold(fp);
2966 		newfde = &fdp->fd_ofiles[indx];
2967 		oldfde = &fdp->fd_ofiles[dfd];
2968 #ifdef CAPABILITIES
2969 		seq_write_begin(&newfde->fde_seq);
2970 #endif
2971 		memcpy(newfde, oldfde, fde_change_size);
2972 		filecaps_copy(&oldfde->fde_caps, &newfde->fde_caps, true);
2973 #ifdef CAPABILITIES
2974 		seq_write_end(&newfde->fde_seq);
2975 #endif
2976 		break;
2977 	case ENXIO:
2978 		/*
2979 		 * Steal away the file pointer from dfd and stuff it into indx.
2980 		 */
2981 		newfde = &fdp->fd_ofiles[indx];
2982 		oldfde = &fdp->fd_ofiles[dfd];
2983 #ifdef CAPABILITIES
2984 		seq_write_begin(&newfde->fde_seq);
2985 #endif
2986 		memcpy(newfde, oldfde, fde_change_size);
2987 		oldfde->fde_file = NULL;
2988 		fdunused(fdp, dfd);
2989 #ifdef CAPABILITIES
2990 		seq_write_end(&newfde->fde_seq);
2991 #endif
2992 		break;
2993 	}
2994 	FILEDESC_XUNLOCK(fdp);
2995 	*indxp = indx;
2996 	return (0);
2997 }
2998 
2999 /*
3000  * This sysctl determines if we will allow a process to chroot(2) if it
3001  * has a directory open:
3002  *	0: disallowed for all processes.
3003  *	1: allowed for processes that were not already chroot(2)'ed.
3004  *	2: allowed for all processes.
3005  */
3006 
3007 static int chroot_allow_open_directories = 1;
3008 
3009 SYSCTL_INT(_kern, OID_AUTO, chroot_allow_open_directories, CTLFLAG_RW,
3010     &chroot_allow_open_directories, 0,
3011     "Allow a process to chroot(2) if it has a directory open");
3012 
3013 /*
3014  * Helper function for raised chroot(2) security function:  Refuse if
3015  * any filedescriptors are open directories.
3016  */
3017 static int
3018 chroot_refuse_vdir_fds(struct filedesc *fdp)
3019 {
3020 	struct vnode *vp;
3021 	struct file *fp;
3022 	int fd;
3023 
3024 	FILEDESC_LOCK_ASSERT(fdp);
3025 
3026 	for (fd = 0; fd <= fdp->fd_lastfile; fd++) {
3027 		fp = fget_locked(fdp, fd);
3028 		if (fp == NULL)
3029 			continue;
3030 		if (fp->f_type == DTYPE_VNODE) {
3031 			vp = fp->f_vnode;
3032 			if (vp->v_type == VDIR)
3033 				return (EPERM);
3034 		}
3035 	}
3036 	return (0);
3037 }
3038 
3039 /*
3040  * Common routine for kern_chroot() and jail_attach().  The caller is
3041  * responsible for invoking priv_check() and mac_vnode_check_chroot() to
3042  * authorize this operation.
3043  */
3044 int
3045 pwd_chroot(struct thread *td, struct vnode *vp)
3046 {
3047 	struct filedesc *fdp;
3048 	struct vnode *oldvp;
3049 	int error;
3050 
3051 	fdp = td->td_proc->p_fd;
3052 	FILEDESC_XLOCK(fdp);
3053 	if (chroot_allow_open_directories == 0 ||
3054 	    (chroot_allow_open_directories == 1 && fdp->fd_rdir != rootvnode)) {
3055 		error = chroot_refuse_vdir_fds(fdp);
3056 		if (error != 0) {
3057 			FILEDESC_XUNLOCK(fdp);
3058 			return (error);
3059 		}
3060 	}
3061 	oldvp = fdp->fd_rdir;
3062 	vrefact(vp);
3063 	fdp->fd_rdir = vp;
3064 	if (fdp->fd_jdir == NULL) {
3065 		vrefact(vp);
3066 		fdp->fd_jdir = vp;
3067 	}
3068 	FILEDESC_XUNLOCK(fdp);
3069 	vrele(oldvp);
3070 	return (0);
3071 }
3072 
3073 void
3074 pwd_chdir(struct thread *td, struct vnode *vp)
3075 {
3076 	struct filedesc *fdp;
3077 	struct vnode *oldvp;
3078 
3079 	fdp = td->td_proc->p_fd;
3080 	FILEDESC_XLOCK(fdp);
3081 	VNASSERT(vp->v_usecount > 0, vp,
3082 	    ("chdir to a vnode with zero usecount"));
3083 	oldvp = fdp->fd_cdir;
3084 	fdp->fd_cdir = vp;
3085 	FILEDESC_XUNLOCK(fdp);
3086 	vrele(oldvp);
3087 }
3088 
3089 /*
3090  * Scan all active processes and prisons to see if any of them have a current
3091  * or root directory of `olddp'. If so, replace them with the new mount point.
3092  */
3093 void
3094 mountcheckdirs(struct vnode *olddp, struct vnode *newdp)
3095 {
3096 	struct filedesc *fdp;
3097 	struct prison *pr;
3098 	struct proc *p;
3099 	int nrele;
3100 
3101 	if (vrefcnt(olddp) == 1)
3102 		return;
3103 	nrele = 0;
3104 	sx_slock(&allproc_lock);
3105 	FOREACH_PROC_IN_SYSTEM(p) {
3106 		PROC_LOCK(p);
3107 		fdp = fdhold(p);
3108 		PROC_UNLOCK(p);
3109 		if (fdp == NULL)
3110 			continue;
3111 		FILEDESC_XLOCK(fdp);
3112 		if (fdp->fd_cdir == olddp) {
3113 			vrefact(newdp);
3114 			fdp->fd_cdir = newdp;
3115 			nrele++;
3116 		}
3117 		if (fdp->fd_rdir == olddp) {
3118 			vrefact(newdp);
3119 			fdp->fd_rdir = newdp;
3120 			nrele++;
3121 		}
3122 		if (fdp->fd_jdir == olddp) {
3123 			vrefact(newdp);
3124 			fdp->fd_jdir = newdp;
3125 			nrele++;
3126 		}
3127 		FILEDESC_XUNLOCK(fdp);
3128 		fddrop(fdp);
3129 	}
3130 	sx_sunlock(&allproc_lock);
3131 	if (rootvnode == olddp) {
3132 		vrefact(newdp);
3133 		rootvnode = newdp;
3134 		nrele++;
3135 	}
3136 	mtx_lock(&prison0.pr_mtx);
3137 	if (prison0.pr_root == olddp) {
3138 		vrefact(newdp);
3139 		prison0.pr_root = newdp;
3140 		nrele++;
3141 	}
3142 	mtx_unlock(&prison0.pr_mtx);
3143 	sx_slock(&allprison_lock);
3144 	TAILQ_FOREACH(pr, &allprison, pr_list) {
3145 		mtx_lock(&pr->pr_mtx);
3146 		if (pr->pr_root == olddp) {
3147 			vrefact(newdp);
3148 			pr->pr_root = newdp;
3149 			nrele++;
3150 		}
3151 		mtx_unlock(&pr->pr_mtx);
3152 	}
3153 	sx_sunlock(&allprison_lock);
3154 	while (nrele--)
3155 		vrele(olddp);
3156 }
3157 
3158 struct filedesc_to_leader *
3159 filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader)
3160 {
3161 	struct filedesc_to_leader *fdtol;
3162 
3163 	fdtol = malloc(sizeof(struct filedesc_to_leader),
3164 	    M_FILEDESC_TO_LEADER, M_WAITOK);
3165 	fdtol->fdl_refcount = 1;
3166 	fdtol->fdl_holdcount = 0;
3167 	fdtol->fdl_wakeup = 0;
3168 	fdtol->fdl_leader = leader;
3169 	if (old != NULL) {
3170 		FILEDESC_XLOCK(fdp);
3171 		fdtol->fdl_next = old->fdl_next;
3172 		fdtol->fdl_prev = old;
3173 		old->fdl_next = fdtol;
3174 		fdtol->fdl_next->fdl_prev = fdtol;
3175 		FILEDESC_XUNLOCK(fdp);
3176 	} else {
3177 		fdtol->fdl_next = fdtol;
3178 		fdtol->fdl_prev = fdtol;
3179 	}
3180 	return (fdtol);
3181 }
3182 
3183 static int
3184 sysctl_kern_proc_nfds(SYSCTL_HANDLER_ARGS)
3185 {
3186 	struct filedesc *fdp;
3187 	int i, count, slots;
3188 
3189 	if (*(int *)arg1 != 0)
3190 		return (EINVAL);
3191 
3192 	fdp = curproc->p_fd;
3193 	count = 0;
3194 	FILEDESC_SLOCK(fdp);
3195 	slots = NDSLOTS(fdp->fd_lastfile + 1);
3196 	for (i = 0; i < slots; i++)
3197 		count += bitcountl(fdp->fd_map[i]);
3198 	FILEDESC_SUNLOCK(fdp);
3199 
3200 	return (SYSCTL_OUT(req, &count, sizeof(count)));
3201 }
3202 
3203 static SYSCTL_NODE(_kern_proc, KERN_PROC_NFDS, nfds,
3204     CTLFLAG_RD|CTLFLAG_CAPRD|CTLFLAG_MPSAFE, sysctl_kern_proc_nfds,
3205     "Number of open file descriptors");
3206 
3207 /*
3208  * Get file structures globally.
3209  */
3210 static int
3211 sysctl_kern_file(SYSCTL_HANDLER_ARGS)
3212 {
3213 	struct xfile xf;
3214 	struct filedesc *fdp;
3215 	struct file *fp;
3216 	struct proc *p;
3217 	int error, n;
3218 
3219 	error = sysctl_wire_old_buffer(req, 0);
3220 	if (error != 0)
3221 		return (error);
3222 	if (req->oldptr == NULL) {
3223 		n = 0;
3224 		sx_slock(&allproc_lock);
3225 		FOREACH_PROC_IN_SYSTEM(p) {
3226 			PROC_LOCK(p);
3227 			if (p->p_state == PRS_NEW) {
3228 				PROC_UNLOCK(p);
3229 				continue;
3230 			}
3231 			fdp = fdhold(p);
3232 			PROC_UNLOCK(p);
3233 			if (fdp == NULL)
3234 				continue;
3235 			/* overestimates sparse tables. */
3236 			if (fdp->fd_lastfile > 0)
3237 				n += fdp->fd_lastfile;
3238 			fddrop(fdp);
3239 		}
3240 		sx_sunlock(&allproc_lock);
3241 		return (SYSCTL_OUT(req, 0, n * sizeof(xf)));
3242 	}
3243 	error = 0;
3244 	bzero(&xf, sizeof(xf));
3245 	xf.xf_size = sizeof(xf);
3246 	sx_slock(&allproc_lock);
3247 	FOREACH_PROC_IN_SYSTEM(p) {
3248 		PROC_LOCK(p);
3249 		if (p->p_state == PRS_NEW) {
3250 			PROC_UNLOCK(p);
3251 			continue;
3252 		}
3253 		if (p_cansee(req->td, p) != 0) {
3254 			PROC_UNLOCK(p);
3255 			continue;
3256 		}
3257 		xf.xf_pid = p->p_pid;
3258 		xf.xf_uid = p->p_ucred->cr_uid;
3259 		fdp = fdhold(p);
3260 		PROC_UNLOCK(p);
3261 		if (fdp == NULL)
3262 			continue;
3263 		FILEDESC_SLOCK(fdp);
3264 		for (n = 0; fdp->fd_refcnt > 0 && n <= fdp->fd_lastfile; ++n) {
3265 			if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
3266 				continue;
3267 			xf.xf_fd = n;
3268 			xf.xf_file = fp;
3269 			xf.xf_data = fp->f_data;
3270 			xf.xf_vnode = fp->f_vnode;
3271 			xf.xf_type = fp->f_type;
3272 			xf.xf_count = fp->f_count;
3273 			xf.xf_msgcount = 0;
3274 			xf.xf_offset = foffset_get(fp);
3275 			xf.xf_flag = fp->f_flag;
3276 			error = SYSCTL_OUT(req, &xf, sizeof(xf));
3277 			if (error)
3278 				break;
3279 		}
3280 		FILEDESC_SUNLOCK(fdp);
3281 		fddrop(fdp);
3282 		if (error)
3283 			break;
3284 	}
3285 	sx_sunlock(&allproc_lock);
3286 	return (error);
3287 }
3288 
3289 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD|CTLFLAG_MPSAFE,
3290     0, 0, sysctl_kern_file, "S,xfile", "Entire file table");
3291 
3292 #ifdef KINFO_FILE_SIZE
3293 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
3294 #endif
3295 
3296 static int
3297 xlate_fflags(int fflags)
3298 {
3299 	static const struct {
3300 		int	fflag;
3301 		int	kf_fflag;
3302 	} fflags_table[] = {
3303 		{ FAPPEND, KF_FLAG_APPEND },
3304 		{ FASYNC, KF_FLAG_ASYNC },
3305 		{ FFSYNC, KF_FLAG_FSYNC },
3306 		{ FHASLOCK, KF_FLAG_HASLOCK },
3307 		{ FNONBLOCK, KF_FLAG_NONBLOCK },
3308 		{ FREAD, KF_FLAG_READ },
3309 		{ FWRITE, KF_FLAG_WRITE },
3310 		{ O_CREAT, KF_FLAG_CREAT },
3311 		{ O_DIRECT, KF_FLAG_DIRECT },
3312 		{ O_EXCL, KF_FLAG_EXCL },
3313 		{ O_EXEC, KF_FLAG_EXEC },
3314 		{ O_EXLOCK, KF_FLAG_EXLOCK },
3315 		{ O_NOFOLLOW, KF_FLAG_NOFOLLOW },
3316 		{ O_SHLOCK, KF_FLAG_SHLOCK },
3317 		{ O_TRUNC, KF_FLAG_TRUNC }
3318 	};
3319 	unsigned int i;
3320 	int kflags;
3321 
3322 	kflags = 0;
3323 	for (i = 0; i < nitems(fflags_table); i++)
3324 		if (fflags & fflags_table[i].fflag)
3325 			kflags |=  fflags_table[i].kf_fflag;
3326 	return (kflags);
3327 }
3328 
3329 /* Trim unused data from kf_path by truncating the structure size. */
3330 static void
3331 pack_kinfo(struct kinfo_file *kif)
3332 {
3333 
3334 	kif->kf_structsize = offsetof(struct kinfo_file, kf_path) +
3335 	    strlen(kif->kf_path) + 1;
3336 	kif->kf_structsize = roundup(kif->kf_structsize, sizeof(uint64_t));
3337 }
3338 
3339 static void
3340 export_file_to_kinfo(struct file *fp, int fd, cap_rights_t *rightsp,
3341     struct kinfo_file *kif, struct filedesc *fdp, int flags)
3342 {
3343 	int error;
3344 
3345 	bzero(kif, sizeof(*kif));
3346 
3347 	/* Set a default type to allow for empty fill_kinfo() methods. */
3348 	kif->kf_type = KF_TYPE_UNKNOWN;
3349 	kif->kf_flags = xlate_fflags(fp->f_flag);
3350 	if (rightsp != NULL)
3351 		kif->kf_cap_rights = *rightsp;
3352 	else
3353 		cap_rights_init(&kif->kf_cap_rights);
3354 	kif->kf_fd = fd;
3355 	kif->kf_ref_count = fp->f_count;
3356 	kif->kf_offset = foffset_get(fp);
3357 
3358 	/*
3359 	 * This may drop the filedesc lock, so the 'fp' cannot be
3360 	 * accessed after this call.
3361 	 */
3362 	error = fo_fill_kinfo(fp, kif, fdp);
3363 	if (error == 0)
3364 		kif->kf_status |= KF_ATTR_VALID;
3365 	if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
3366 		pack_kinfo(kif);
3367 	else
3368 		kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
3369 }
3370 
3371 static void
3372 export_vnode_to_kinfo(struct vnode *vp, int fd, int fflags,
3373     struct kinfo_file *kif, int flags)
3374 {
3375 	int error;
3376 
3377 	bzero(kif, sizeof(*kif));
3378 
3379 	kif->kf_type = KF_TYPE_VNODE;
3380 	error = vn_fill_kinfo_vnode(vp, kif);
3381 	if (error == 0)
3382 		kif->kf_status |= KF_ATTR_VALID;
3383 	kif->kf_flags = xlate_fflags(fflags);
3384 	cap_rights_init(&kif->kf_cap_rights);
3385 	kif->kf_fd = fd;
3386 	kif->kf_ref_count = -1;
3387 	kif->kf_offset = -1;
3388 	if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
3389 		pack_kinfo(kif);
3390 	else
3391 		kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
3392 	vrele(vp);
3393 }
3394 
3395 struct export_fd_buf {
3396 	struct filedesc		*fdp;
3397 	struct sbuf 		*sb;
3398 	ssize_t			remainder;
3399 	struct kinfo_file	kif;
3400 	int			flags;
3401 };
3402 
3403 static int
3404 export_kinfo_to_sb(struct export_fd_buf *efbuf)
3405 {
3406 	struct kinfo_file *kif;
3407 
3408 	kif = &efbuf->kif;
3409 	if (efbuf->remainder != -1) {
3410 		if (efbuf->remainder < kif->kf_structsize) {
3411 			/* Terminate export. */
3412 			efbuf->remainder = 0;
3413 			return (0);
3414 		}
3415 		efbuf->remainder -= kif->kf_structsize;
3416 	}
3417 	return (sbuf_bcat(efbuf->sb, kif, kif->kf_structsize) == 0 ? 0 : ENOMEM);
3418 }
3419 
3420 static int
3421 export_file_to_sb(struct file *fp, int fd, cap_rights_t *rightsp,
3422     struct export_fd_buf *efbuf)
3423 {
3424 	int error;
3425 
3426 	if (efbuf->remainder == 0)
3427 		return (0);
3428 	export_file_to_kinfo(fp, fd, rightsp, &efbuf->kif, efbuf->fdp,
3429 	    efbuf->flags);
3430 	FILEDESC_SUNLOCK(efbuf->fdp);
3431 	error = export_kinfo_to_sb(efbuf);
3432 	FILEDESC_SLOCK(efbuf->fdp);
3433 	return (error);
3434 }
3435 
3436 static int
3437 export_vnode_to_sb(struct vnode *vp, int fd, int fflags,
3438     struct export_fd_buf *efbuf)
3439 {
3440 	int error;
3441 
3442 	if (efbuf->remainder == 0)
3443 		return (0);
3444 	if (efbuf->fdp != NULL)
3445 		FILEDESC_SUNLOCK(efbuf->fdp);
3446 	export_vnode_to_kinfo(vp, fd, fflags, &efbuf->kif, efbuf->flags);
3447 	error = export_kinfo_to_sb(efbuf);
3448 	if (efbuf->fdp != NULL)
3449 		FILEDESC_SLOCK(efbuf->fdp);
3450 	return (error);
3451 }
3452 
3453 /*
3454  * Store a process file descriptor information to sbuf.
3455  *
3456  * Takes a locked proc as argument, and returns with the proc unlocked.
3457  */
3458 int
3459 kern_proc_filedesc_out(struct proc *p,  struct sbuf *sb, ssize_t maxlen,
3460     int flags)
3461 {
3462 	struct file *fp;
3463 	struct filedesc *fdp;
3464 	struct export_fd_buf *efbuf;
3465 	struct vnode *cttyvp, *textvp, *tracevp;
3466 	int error, i;
3467 	cap_rights_t rights;
3468 
3469 	PROC_LOCK_ASSERT(p, MA_OWNED);
3470 
3471 	/* ktrace vnode */
3472 	tracevp = p->p_tracevp;
3473 	if (tracevp != NULL)
3474 		vrefact(tracevp);
3475 	/* text vnode */
3476 	textvp = p->p_textvp;
3477 	if (textvp != NULL)
3478 		vrefact(textvp);
3479 	/* Controlling tty. */
3480 	cttyvp = NULL;
3481 	if (p->p_pgrp != NULL && p->p_pgrp->pg_session != NULL) {
3482 		cttyvp = p->p_pgrp->pg_session->s_ttyvp;
3483 		if (cttyvp != NULL)
3484 			vrefact(cttyvp);
3485 	}
3486 	fdp = fdhold(p);
3487 	PROC_UNLOCK(p);
3488 	efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
3489 	efbuf->fdp = NULL;
3490 	efbuf->sb = sb;
3491 	efbuf->remainder = maxlen;
3492 	efbuf->flags = flags;
3493 	if (tracevp != NULL)
3494 		export_vnode_to_sb(tracevp, KF_FD_TYPE_TRACE, FREAD | FWRITE,
3495 		    efbuf);
3496 	if (textvp != NULL)
3497 		export_vnode_to_sb(textvp, KF_FD_TYPE_TEXT, FREAD, efbuf);
3498 	if (cttyvp != NULL)
3499 		export_vnode_to_sb(cttyvp, KF_FD_TYPE_CTTY, FREAD | FWRITE,
3500 		    efbuf);
3501 	error = 0;
3502 	if (fdp == NULL)
3503 		goto fail;
3504 	efbuf->fdp = fdp;
3505 	FILEDESC_SLOCK(fdp);
3506 	/* working directory */
3507 	if (fdp->fd_cdir != NULL) {
3508 		vrefact(fdp->fd_cdir);
3509 		export_vnode_to_sb(fdp->fd_cdir, KF_FD_TYPE_CWD, FREAD, efbuf);
3510 	}
3511 	/* root directory */
3512 	if (fdp->fd_rdir != NULL) {
3513 		vrefact(fdp->fd_rdir);
3514 		export_vnode_to_sb(fdp->fd_rdir, KF_FD_TYPE_ROOT, FREAD, efbuf);
3515 	}
3516 	/* jail directory */
3517 	if (fdp->fd_jdir != NULL) {
3518 		vrefact(fdp->fd_jdir);
3519 		export_vnode_to_sb(fdp->fd_jdir, KF_FD_TYPE_JAIL, FREAD, efbuf);
3520 	}
3521 	for (i = 0; fdp->fd_refcnt > 0 && i <= fdp->fd_lastfile; i++) {
3522 		if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
3523 			continue;
3524 #ifdef CAPABILITIES
3525 		rights = *cap_rights(fdp, i);
3526 #else /* !CAPABILITIES */
3527 		cap_rights_init(&rights);
3528 #endif
3529 		/*
3530 		 * Create sysctl entry.  It is OK to drop the filedesc
3531 		 * lock inside of export_file_to_sb() as we will
3532 		 * re-validate and re-evaluate its properties when the
3533 		 * loop continues.
3534 		 */
3535 		error = export_file_to_sb(fp, i, &rights, efbuf);
3536 		if (error != 0 || efbuf->remainder == 0)
3537 			break;
3538 	}
3539 	FILEDESC_SUNLOCK(fdp);
3540 	fddrop(fdp);
3541 fail:
3542 	free(efbuf, M_TEMP);
3543 	return (error);
3544 }
3545 
3546 #define FILEDESC_SBUF_SIZE	(sizeof(struct kinfo_file) * 5)
3547 
3548 /*
3549  * Get per-process file descriptors for use by procstat(1), et al.
3550  */
3551 static int
3552 sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS)
3553 {
3554 	struct sbuf sb;
3555 	struct proc *p;
3556 	ssize_t maxlen;
3557 	int error, error2, *name;
3558 
3559 	name = (int *)arg1;
3560 
3561 	sbuf_new_for_sysctl(&sb, NULL, FILEDESC_SBUF_SIZE, req);
3562 	sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
3563 	error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
3564 	if (error != 0) {
3565 		sbuf_delete(&sb);
3566 		return (error);
3567 	}
3568 	maxlen = req->oldptr != NULL ? req->oldlen : -1;
3569 	error = kern_proc_filedesc_out(p, &sb, maxlen,
3570 	    KERN_FILEDESC_PACK_KINFO);
3571 	error2 = sbuf_finish(&sb);
3572 	sbuf_delete(&sb);
3573 	return (error != 0 ? error : error2);
3574 }
3575 
3576 #ifdef KINFO_OFILE_SIZE
3577 CTASSERT(sizeof(struct kinfo_ofile) == KINFO_OFILE_SIZE);
3578 #endif
3579 
3580 #ifdef COMPAT_FREEBSD7
3581 static void
3582 kinfo_to_okinfo(struct kinfo_file *kif, struct kinfo_ofile *okif)
3583 {
3584 
3585 	okif->kf_structsize = sizeof(*okif);
3586 	okif->kf_type = kif->kf_type;
3587 	okif->kf_fd = kif->kf_fd;
3588 	okif->kf_ref_count = kif->kf_ref_count;
3589 	okif->kf_flags = kif->kf_flags & (KF_FLAG_READ | KF_FLAG_WRITE |
3590 	    KF_FLAG_APPEND | KF_FLAG_ASYNC | KF_FLAG_FSYNC | KF_FLAG_NONBLOCK |
3591 	    KF_FLAG_DIRECT | KF_FLAG_HASLOCK);
3592 	okif->kf_offset = kif->kf_offset;
3593 	okif->kf_vnode_type = kif->kf_vnode_type;
3594 	okif->kf_sock_domain = kif->kf_sock_domain;
3595 	okif->kf_sock_type = kif->kf_sock_type;
3596 	okif->kf_sock_protocol = kif->kf_sock_protocol;
3597 	strlcpy(okif->kf_path, kif->kf_path, sizeof(okif->kf_path));
3598 	okif->kf_sa_local = kif->kf_sa_local;
3599 	okif->kf_sa_peer = kif->kf_sa_peer;
3600 }
3601 
3602 static int
3603 export_vnode_for_osysctl(struct vnode *vp, int type, struct kinfo_file *kif,
3604     struct kinfo_ofile *okif, struct filedesc *fdp, struct sysctl_req *req)
3605 {
3606 	int error;
3607 
3608 	vrefact(vp);
3609 	FILEDESC_SUNLOCK(fdp);
3610 	export_vnode_to_kinfo(vp, type, 0, kif, KERN_FILEDESC_PACK_KINFO);
3611 	kinfo_to_okinfo(kif, okif);
3612 	error = SYSCTL_OUT(req, okif, sizeof(*okif));
3613 	FILEDESC_SLOCK(fdp);
3614 	return (error);
3615 }
3616 
3617 /*
3618  * Get per-process file descriptors for use by procstat(1), et al.
3619  */
3620 static int
3621 sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS)
3622 {
3623 	struct kinfo_ofile *okif;
3624 	struct kinfo_file *kif;
3625 	struct filedesc *fdp;
3626 	int error, i, *name;
3627 	struct file *fp;
3628 	struct proc *p;
3629 
3630 	name = (int *)arg1;
3631 	error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
3632 	if (error != 0)
3633 		return (error);
3634 	fdp = fdhold(p);
3635 	PROC_UNLOCK(p);
3636 	if (fdp == NULL)
3637 		return (ENOENT);
3638 	kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK);
3639 	okif = malloc(sizeof(*okif), M_TEMP, M_WAITOK);
3640 	FILEDESC_SLOCK(fdp);
3641 	if (fdp->fd_cdir != NULL)
3642 		export_vnode_for_osysctl(fdp->fd_cdir, KF_FD_TYPE_CWD, kif,
3643 		    okif, fdp, req);
3644 	if (fdp->fd_rdir != NULL)
3645 		export_vnode_for_osysctl(fdp->fd_rdir, KF_FD_TYPE_ROOT, kif,
3646 		    okif, fdp, req);
3647 	if (fdp->fd_jdir != NULL)
3648 		export_vnode_for_osysctl(fdp->fd_jdir, KF_FD_TYPE_JAIL, kif,
3649 		    okif, fdp, req);
3650 	for (i = 0; fdp->fd_refcnt > 0 && i <= fdp->fd_lastfile; i++) {
3651 		if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
3652 			continue;
3653 		export_file_to_kinfo(fp, i, NULL, kif, fdp,
3654 		    KERN_FILEDESC_PACK_KINFO);
3655 		FILEDESC_SUNLOCK(fdp);
3656 		kinfo_to_okinfo(kif, okif);
3657 		error = SYSCTL_OUT(req, okif, sizeof(*okif));
3658 		FILEDESC_SLOCK(fdp);
3659 		if (error)
3660 			break;
3661 	}
3662 	FILEDESC_SUNLOCK(fdp);
3663 	fddrop(fdp);
3664 	free(kif, M_TEMP);
3665 	free(okif, M_TEMP);
3666 	return (0);
3667 }
3668 
3669 static SYSCTL_NODE(_kern_proc, KERN_PROC_OFILEDESC, ofiledesc,
3670     CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_ofiledesc,
3671     "Process ofiledesc entries");
3672 #endif	/* COMPAT_FREEBSD7 */
3673 
3674 int
3675 vntype_to_kinfo(int vtype)
3676 {
3677 	struct {
3678 		int	vtype;
3679 		int	kf_vtype;
3680 	} vtypes_table[] = {
3681 		{ VBAD, KF_VTYPE_VBAD },
3682 		{ VBLK, KF_VTYPE_VBLK },
3683 		{ VCHR, KF_VTYPE_VCHR },
3684 		{ VDIR, KF_VTYPE_VDIR },
3685 		{ VFIFO, KF_VTYPE_VFIFO },
3686 		{ VLNK, KF_VTYPE_VLNK },
3687 		{ VNON, KF_VTYPE_VNON },
3688 		{ VREG, KF_VTYPE_VREG },
3689 		{ VSOCK, KF_VTYPE_VSOCK }
3690 	};
3691 	unsigned int i;
3692 
3693 	/*
3694 	 * Perform vtype translation.
3695 	 */
3696 	for (i = 0; i < nitems(vtypes_table); i++)
3697 		if (vtypes_table[i].vtype == vtype)
3698 			return (vtypes_table[i].kf_vtype);
3699 
3700 	return (KF_VTYPE_UNKNOWN);
3701 }
3702 
3703 static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc,
3704     CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_filedesc,
3705     "Process filedesc entries");
3706 
3707 /*
3708  * Store a process current working directory information to sbuf.
3709  *
3710  * Takes a locked proc as argument, and returns with the proc unlocked.
3711  */
3712 int
3713 kern_proc_cwd_out(struct proc *p,  struct sbuf *sb, ssize_t maxlen)
3714 {
3715 	struct filedesc *fdp;
3716 	struct export_fd_buf *efbuf;
3717 	int error;
3718 
3719 	PROC_LOCK_ASSERT(p, MA_OWNED);
3720 
3721 	fdp = fdhold(p);
3722 	PROC_UNLOCK(p);
3723 	if (fdp == NULL)
3724 		return (EINVAL);
3725 
3726 	efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
3727 	efbuf->fdp = fdp;
3728 	efbuf->sb = sb;
3729 	efbuf->remainder = maxlen;
3730 
3731 	FILEDESC_SLOCK(fdp);
3732 	if (fdp->fd_cdir == NULL)
3733 		error = EINVAL;
3734 	else {
3735 		vrefact(fdp->fd_cdir);
3736 		error = export_vnode_to_sb(fdp->fd_cdir, KF_FD_TYPE_CWD,
3737 		    FREAD, efbuf);
3738 	}
3739 	FILEDESC_SUNLOCK(fdp);
3740 	fddrop(fdp);
3741 	free(efbuf, M_TEMP);
3742 	return (error);
3743 }
3744 
3745 /*
3746  * Get per-process current working directory.
3747  */
3748 static int
3749 sysctl_kern_proc_cwd(SYSCTL_HANDLER_ARGS)
3750 {
3751 	struct sbuf sb;
3752 	struct proc *p;
3753 	ssize_t maxlen;
3754 	int error, error2, *name;
3755 
3756 	name = (int *)arg1;
3757 
3758 	sbuf_new_for_sysctl(&sb, NULL, sizeof(struct kinfo_file), req);
3759 	sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
3760 	error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
3761 	if (error != 0) {
3762 		sbuf_delete(&sb);
3763 		return (error);
3764 	}
3765 	maxlen = req->oldptr != NULL ? req->oldlen : -1;
3766 	error = kern_proc_cwd_out(p, &sb, maxlen);
3767 	error2 = sbuf_finish(&sb);
3768 	sbuf_delete(&sb);
3769 	return (error != 0 ? error : error2);
3770 }
3771 
3772 static SYSCTL_NODE(_kern_proc, KERN_PROC_CWD, cwd, CTLFLAG_RD|CTLFLAG_MPSAFE,
3773     sysctl_kern_proc_cwd, "Process current working directory");
3774 
3775 #ifdef DDB
3776 /*
3777  * For the purposes of debugging, generate a human-readable string for the
3778  * file type.
3779  */
3780 static const char *
3781 file_type_to_name(short type)
3782 {
3783 
3784 	switch (type) {
3785 	case 0:
3786 		return ("zero");
3787 	case DTYPE_VNODE:
3788 		return ("vnod");
3789 	case DTYPE_SOCKET:
3790 		return ("sock");
3791 	case DTYPE_PIPE:
3792 		return ("pipe");
3793 	case DTYPE_FIFO:
3794 		return ("fifo");
3795 	case DTYPE_KQUEUE:
3796 		return ("kque");
3797 	case DTYPE_CRYPTO:
3798 		return ("crpt");
3799 	case DTYPE_MQUEUE:
3800 		return ("mque");
3801 	case DTYPE_SHM:
3802 		return ("shm");
3803 	case DTYPE_SEM:
3804 		return ("ksem");
3805 	default:
3806 		return ("unkn");
3807 	}
3808 }
3809 
3810 /*
3811  * For the purposes of debugging, identify a process (if any, perhaps one of
3812  * many) that references the passed file in its file descriptor array. Return
3813  * NULL if none.
3814  */
3815 static struct proc *
3816 file_to_first_proc(struct file *fp)
3817 {
3818 	struct filedesc *fdp;
3819 	struct proc *p;
3820 	int n;
3821 
3822 	FOREACH_PROC_IN_SYSTEM(p) {
3823 		if (p->p_state == PRS_NEW)
3824 			continue;
3825 		fdp = p->p_fd;
3826 		if (fdp == NULL)
3827 			continue;
3828 		for (n = 0; n <= fdp->fd_lastfile; n++) {
3829 			if (fp == fdp->fd_ofiles[n].fde_file)
3830 				return (p);
3831 		}
3832 	}
3833 	return (NULL);
3834 }
3835 
3836 static void
3837 db_print_file(struct file *fp, int header)
3838 {
3839 	struct proc *p;
3840 
3841 	if (header)
3842 		db_printf("%8s %4s %8s %8s %4s %5s %6s %8s %5s %12s\n",
3843 		    "File", "Type", "Data", "Flag", "GCFl", "Count",
3844 		    "MCount", "Vnode", "FPID", "FCmd");
3845 	p = file_to_first_proc(fp);
3846 	db_printf("%8p %4s %8p %08x %04x %5d %6d %8p %5d %12s\n", fp,
3847 	    file_type_to_name(fp->f_type), fp->f_data, fp->f_flag,
3848 	    0, fp->f_count, 0, fp->f_vnode,
3849 	    p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-");
3850 }
3851 
3852 DB_SHOW_COMMAND(file, db_show_file)
3853 {
3854 	struct file *fp;
3855 
3856 	if (!have_addr) {
3857 		db_printf("usage: show file <addr>\n");
3858 		return;
3859 	}
3860 	fp = (struct file *)addr;
3861 	db_print_file(fp, 1);
3862 }
3863 
3864 DB_SHOW_COMMAND(files, db_show_files)
3865 {
3866 	struct filedesc *fdp;
3867 	struct file *fp;
3868 	struct proc *p;
3869 	int header;
3870 	int n;
3871 
3872 	header = 1;
3873 	FOREACH_PROC_IN_SYSTEM(p) {
3874 		if (p->p_state == PRS_NEW)
3875 			continue;
3876 		if ((fdp = p->p_fd) == NULL)
3877 			continue;
3878 		for (n = 0; n <= fdp->fd_lastfile; ++n) {
3879 			if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
3880 				continue;
3881 			db_print_file(fp, header);
3882 			header = 0;
3883 		}
3884 	}
3885 }
3886 #endif
3887 
3888 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
3889     &maxfilesperproc, 0, "Maximum files allowed open per process");
3890 
3891 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
3892     &maxfiles, 0, "Maximum number of files");
3893 
3894 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
3895     __DEVOLATILE(int *, &openfiles), 0, "System-wide number of open files");
3896 
3897 /* ARGSUSED*/
3898 static void
3899 filelistinit(void *dummy)
3900 {
3901 
3902 	file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL,
3903 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
3904 	filedesc0_zone = uma_zcreate("filedesc0", sizeof(struct filedesc0),
3905 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
3906 	mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF);
3907 }
3908 SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL);
3909 
3910 /*-------------------------------------------------------------------*/
3911 
3912 static int
3913 badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred,
3914     int flags, struct thread *td)
3915 {
3916 
3917 	return (EBADF);
3918 }
3919 
3920 static int
3921 badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
3922     struct thread *td)
3923 {
3924 
3925 	return (EINVAL);
3926 }
3927 
3928 static int
3929 badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred,
3930     struct thread *td)
3931 {
3932 
3933 	return (EBADF);
3934 }
3935 
3936 static int
3937 badfo_poll(struct file *fp, int events, struct ucred *active_cred,
3938     struct thread *td)
3939 {
3940 
3941 	return (0);
3942 }
3943 
3944 static int
3945 badfo_kqfilter(struct file *fp, struct knote *kn)
3946 {
3947 
3948 	return (EBADF);
3949 }
3950 
3951 static int
3952 badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred,
3953     struct thread *td)
3954 {
3955 
3956 	return (EBADF);
3957 }
3958 
3959 static int
3960 badfo_close(struct file *fp, struct thread *td)
3961 {
3962 
3963 	return (0);
3964 }
3965 
3966 static int
3967 badfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
3968     struct thread *td)
3969 {
3970 
3971 	return (EBADF);
3972 }
3973 
3974 static int
3975 badfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
3976     struct thread *td)
3977 {
3978 
3979 	return (EBADF);
3980 }
3981 
3982 static int
3983 badfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
3984     struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
3985     struct thread *td)
3986 {
3987 
3988 	return (EBADF);
3989 }
3990 
3991 static int
3992 badfo_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
3993 {
3994 
3995 	return (0);
3996 }
3997 
3998 struct fileops badfileops = {
3999 	.fo_read = badfo_readwrite,
4000 	.fo_write = badfo_readwrite,
4001 	.fo_truncate = badfo_truncate,
4002 	.fo_ioctl = badfo_ioctl,
4003 	.fo_poll = badfo_poll,
4004 	.fo_kqfilter = badfo_kqfilter,
4005 	.fo_stat = badfo_stat,
4006 	.fo_close = badfo_close,
4007 	.fo_chmod = badfo_chmod,
4008 	.fo_chown = badfo_chown,
4009 	.fo_sendfile = badfo_sendfile,
4010 	.fo_fill_kinfo = badfo_fill_kinfo,
4011 };
4012 
4013 int
4014 invfo_rdwr(struct file *fp, struct uio *uio, struct ucred *active_cred,
4015     int flags, struct thread *td)
4016 {
4017 
4018 	return (EOPNOTSUPP);
4019 }
4020 
4021 int
4022 invfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
4023     struct thread *td)
4024 {
4025 
4026 	return (EINVAL);
4027 }
4028 
4029 int
4030 invfo_ioctl(struct file *fp, u_long com, void *data,
4031     struct ucred *active_cred, struct thread *td)
4032 {
4033 
4034 	return (ENOTTY);
4035 }
4036 
4037 int
4038 invfo_poll(struct file *fp, int events, struct ucred *active_cred,
4039     struct thread *td)
4040 {
4041 
4042 	return (poll_no_poll(events));
4043 }
4044 
4045 int
4046 invfo_kqfilter(struct file *fp, struct knote *kn)
4047 {
4048 
4049 	return (EINVAL);
4050 }
4051 
4052 int
4053 invfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
4054     struct thread *td)
4055 {
4056 
4057 	return (EINVAL);
4058 }
4059 
4060 int
4061 invfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
4062     struct thread *td)
4063 {
4064 
4065 	return (EINVAL);
4066 }
4067 
4068 int
4069 invfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
4070     struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
4071     struct thread *td)
4072 {
4073 
4074 	return (EINVAL);
4075 }
4076 
4077 /*-------------------------------------------------------------------*/
4078 
4079 /*
4080  * File Descriptor pseudo-device driver (/dev/fd/).
4081  *
4082  * Opening minor device N dup()s the file (if any) connected to file
4083  * descriptor N belonging to the calling process.  Note that this driver
4084  * consists of only the ``open()'' routine, because all subsequent
4085  * references to this file will be direct to the other driver.
4086  *
4087  * XXX: we could give this one a cloning event handler if necessary.
4088  */
4089 
4090 /* ARGSUSED */
4091 static int
4092 fdopen(struct cdev *dev, int mode, int type, struct thread *td)
4093 {
4094 
4095 	/*
4096 	 * XXX Kludge: set curthread->td_dupfd to contain the value of the
4097 	 * the file descriptor being sought for duplication. The error
4098 	 * return ensures that the vnode for this device will be released
4099 	 * by vn_open. Open will detect this special error and take the
4100 	 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
4101 	 * will simply report the error.
4102 	 */
4103 	td->td_dupfd = dev2unit(dev);
4104 	return (ENODEV);
4105 }
4106 
4107 static struct cdevsw fildesc_cdevsw = {
4108 	.d_version =	D_VERSION,
4109 	.d_open =	fdopen,
4110 	.d_name =	"FD",
4111 };
4112 
4113 static void
4114 fildesc_drvinit(void *unused)
4115 {
4116 	struct cdev *dev;
4117 
4118 	dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 0, NULL,
4119 	    UID_ROOT, GID_WHEEL, 0666, "fd/0");
4120 	make_dev_alias(dev, "stdin");
4121 	dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 1, NULL,
4122 	    UID_ROOT, GID_WHEEL, 0666, "fd/1");
4123 	make_dev_alias(dev, "stdout");
4124 	dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 2, NULL,
4125 	    UID_ROOT, GID_WHEEL, 0666, "fd/2");
4126 	make_dev_alias(dev, "stderr");
4127 }
4128 
4129 SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL);
4130