xref: /freebsd/sys/kern/kern_descrip.c (revision 3416500aef140042c64bc149cb1ec6620483bc44)
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 #if defined(COMPAT_FREEBSD11)
1308 int
1309 freebsd11_fstat(struct thread *td, struct freebsd11_fstat_args *uap)
1310 {
1311 	struct stat sb;
1312 	struct freebsd11_stat osb;
1313 	int error;
1314 
1315 	error = kern_fstat(td, uap->fd, &sb);
1316 	if (error != 0)
1317 		return (error);
1318 	error = freebsd11_cvtstat(&sb, &osb);
1319 	if (error == 0)
1320 		error = copyout(&osb, uap->sb, sizeof(osb));
1321 	return (error);
1322 }
1323 #endif	/* COMPAT_FREEBSD11 */
1324 
1325 /*
1326  * Return status information about a file descriptor.
1327  */
1328 #ifndef _SYS_SYSPROTO_H_
1329 struct fstat_args {
1330 	int	fd;
1331 	struct	stat *sb;
1332 };
1333 #endif
1334 /* ARGSUSED */
1335 int
1336 sys_fstat(struct thread *td, struct fstat_args *uap)
1337 {
1338 	struct stat ub;
1339 	int error;
1340 
1341 	error = kern_fstat(td, uap->fd, &ub);
1342 	if (error == 0)
1343 		error = copyout(&ub, uap->sb, sizeof(ub));
1344 	return (error);
1345 }
1346 
1347 int
1348 kern_fstat(struct thread *td, int fd, struct stat *sbp)
1349 {
1350 	struct file *fp;
1351 	cap_rights_t rights;
1352 	int error;
1353 
1354 	AUDIT_ARG_FD(fd);
1355 
1356 	error = fget(td, fd, cap_rights_init(&rights, CAP_FSTAT), &fp);
1357 	if (error != 0)
1358 		return (error);
1359 
1360 	AUDIT_ARG_FILE(td->td_proc, fp);
1361 
1362 	error = fo_stat(fp, sbp, td->td_ucred, td);
1363 	fdrop(fp, td);
1364 #ifdef __STAT_TIME_T_EXT
1365 	if (error == 0) {
1366 		sbp->st_atim_ext = 0;
1367 		sbp->st_mtim_ext = 0;
1368 		sbp->st_ctim_ext = 0;
1369 		sbp->st_btim_ext = 0;
1370 	}
1371 #endif
1372 #ifdef KTRACE
1373 	if (error == 0 && KTRPOINT(td, KTR_STRUCT))
1374 		ktrstat(sbp);
1375 #endif
1376 	return (error);
1377 }
1378 
1379 #if defined(COMPAT_FREEBSD11)
1380 /*
1381  * Return status information about a file descriptor.
1382  */
1383 #ifndef _SYS_SYSPROTO_H_
1384 struct freebsd11_nfstat_args {
1385 	int	fd;
1386 	struct	nstat *sb;
1387 };
1388 #endif
1389 /* ARGSUSED */
1390 int
1391 freebsd11_nfstat(struct thread *td, struct freebsd11_nfstat_args *uap)
1392 {
1393 	struct nstat nub;
1394 	struct stat ub;
1395 	int error;
1396 
1397 	error = kern_fstat(td, uap->fd, &ub);
1398 	if (error == 0) {
1399 		freebsd11_cvtnstat(&ub, &nub);
1400 		error = copyout(&nub, uap->sb, sizeof(nub));
1401 	}
1402 	return (error);
1403 }
1404 #endif /* COMPAT_FREEBSD11 */
1405 
1406 /*
1407  * Return pathconf information about a file descriptor.
1408  */
1409 #ifndef _SYS_SYSPROTO_H_
1410 struct fpathconf_args {
1411 	int	fd;
1412 	int	name;
1413 };
1414 #endif
1415 /* ARGSUSED */
1416 int
1417 sys_fpathconf(struct thread *td, struct fpathconf_args *uap)
1418 {
1419 	struct file *fp;
1420 	struct vnode *vp;
1421 	cap_rights_t rights;
1422 	int error;
1423 
1424 	error = fget(td, uap->fd, cap_rights_init(&rights, CAP_FPATHCONF), &fp);
1425 	if (error != 0)
1426 		return (error);
1427 
1428 	if (uap->name == _PC_ASYNC_IO) {
1429 		td->td_retval[0] = _POSIX_ASYNCHRONOUS_IO;
1430 		goto out;
1431 	}
1432 	vp = fp->f_vnode;
1433 	if (vp != NULL) {
1434 		vn_lock(vp, LK_SHARED | LK_RETRY);
1435 		error = VOP_PATHCONF(vp, uap->name, td->td_retval);
1436 		VOP_UNLOCK(vp, 0);
1437 	} else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) {
1438 		if (uap->name != _PC_PIPE_BUF) {
1439 			error = EINVAL;
1440 		} else {
1441 			td->td_retval[0] = PIPE_BUF;
1442 			error = 0;
1443 		}
1444 	} else {
1445 		error = EOPNOTSUPP;
1446 	}
1447 out:
1448 	fdrop(fp, td);
1449 	return (error);
1450 }
1451 
1452 /*
1453  * Initialize filecaps structure.
1454  */
1455 void
1456 filecaps_init(struct filecaps *fcaps)
1457 {
1458 
1459 	bzero(fcaps, sizeof(*fcaps));
1460 	fcaps->fc_nioctls = -1;
1461 }
1462 
1463 /*
1464  * Copy filecaps structure allocating memory for ioctls array if needed.
1465  *
1466  * The last parameter indicates whether the fdtable is locked. If it is not and
1467  * ioctls are encountered, copying fails and the caller must lock the table.
1468  *
1469  * Note that if the table was not locked, the caller has to check the relevant
1470  * sequence counter to determine whether the operation was successful.
1471  */
1472 int
1473 filecaps_copy(const struct filecaps *src, struct filecaps *dst, bool locked)
1474 {
1475 	size_t size;
1476 
1477 	*dst = *src;
1478 	if (src->fc_ioctls == NULL)
1479 		return (0);
1480 	if (!locked)
1481 		return (1);
1482 
1483 	KASSERT(src->fc_nioctls > 0,
1484 	    ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1485 
1486 	size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1487 	dst->fc_ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1488 	bcopy(src->fc_ioctls, dst->fc_ioctls, size);
1489 	return (0);
1490 }
1491 
1492 /*
1493  * Move filecaps structure to the new place and clear the old place.
1494  */
1495 void
1496 filecaps_move(struct filecaps *src, struct filecaps *dst)
1497 {
1498 
1499 	*dst = *src;
1500 	bzero(src, sizeof(*src));
1501 }
1502 
1503 /*
1504  * Fill the given filecaps structure with full rights.
1505  */
1506 static void
1507 filecaps_fill(struct filecaps *fcaps)
1508 {
1509 
1510 	CAP_ALL(&fcaps->fc_rights);
1511 	fcaps->fc_ioctls = NULL;
1512 	fcaps->fc_nioctls = -1;
1513 	fcaps->fc_fcntls = CAP_FCNTL_ALL;
1514 }
1515 
1516 /*
1517  * Free memory allocated within filecaps structure.
1518  */
1519 void
1520 filecaps_free(struct filecaps *fcaps)
1521 {
1522 
1523 	free(fcaps->fc_ioctls, M_FILECAPS);
1524 	bzero(fcaps, sizeof(*fcaps));
1525 }
1526 
1527 /*
1528  * Validate the given filecaps structure.
1529  */
1530 static void
1531 filecaps_validate(const struct filecaps *fcaps, const char *func)
1532 {
1533 
1534 	KASSERT(cap_rights_is_valid(&fcaps->fc_rights),
1535 	    ("%s: invalid rights", func));
1536 	KASSERT((fcaps->fc_fcntls & ~CAP_FCNTL_ALL) == 0,
1537 	    ("%s: invalid fcntls", func));
1538 	KASSERT(fcaps->fc_fcntls == 0 ||
1539 	    cap_rights_is_set(&fcaps->fc_rights, CAP_FCNTL),
1540 	    ("%s: fcntls without CAP_FCNTL", func));
1541 	KASSERT(fcaps->fc_ioctls != NULL ? fcaps->fc_nioctls > 0 :
1542 	    (fcaps->fc_nioctls == -1 || fcaps->fc_nioctls == 0),
1543 	    ("%s: invalid ioctls", func));
1544 	KASSERT(fcaps->fc_nioctls == 0 ||
1545 	    cap_rights_is_set(&fcaps->fc_rights, CAP_IOCTL),
1546 	    ("%s: ioctls without CAP_IOCTL", func));
1547 }
1548 
1549 static void
1550 fdgrowtable_exp(struct filedesc *fdp, int nfd)
1551 {
1552 	int nfd1;
1553 
1554 	FILEDESC_XLOCK_ASSERT(fdp);
1555 
1556 	nfd1 = fdp->fd_nfiles * 2;
1557 	if (nfd1 < nfd)
1558 		nfd1 = nfd;
1559 	fdgrowtable(fdp, nfd1);
1560 }
1561 
1562 /*
1563  * Grow the file table to accommodate (at least) nfd descriptors.
1564  */
1565 static void
1566 fdgrowtable(struct filedesc *fdp, int nfd)
1567 {
1568 	struct filedesc0 *fdp0;
1569 	struct freetable *ft;
1570 	struct fdescenttbl *ntable;
1571 	struct fdescenttbl *otable;
1572 	int nnfiles, onfiles;
1573 	NDSLOTTYPE *nmap, *omap;
1574 
1575 	/*
1576 	 * If lastfile is -1 this struct filedesc was just allocated and we are
1577 	 * growing it to accommodate for the one we are going to copy from. There
1578 	 * is no need to have a lock on this one as it's not visible to anyone.
1579 	 */
1580 	if (fdp->fd_lastfile != -1)
1581 		FILEDESC_XLOCK_ASSERT(fdp);
1582 
1583 	KASSERT(fdp->fd_nfiles > 0, ("zero-length file table"));
1584 
1585 	/* save old values */
1586 	onfiles = fdp->fd_nfiles;
1587 	otable = fdp->fd_files;
1588 	omap = fdp->fd_map;
1589 
1590 	/* compute the size of the new table */
1591 	nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */
1592 	if (nnfiles <= onfiles)
1593 		/* the table is already large enough */
1594 		return;
1595 
1596 	/*
1597 	 * Allocate a new table.  We need enough space for the number of
1598 	 * entries, file entries themselves and the struct freetable we will use
1599 	 * when we decommission the table and place it on the freelist.
1600 	 * We place the struct freetable in the middle so we don't have
1601 	 * to worry about padding.
1602 	 */
1603 	ntable = malloc(offsetof(struct fdescenttbl, fdt_ofiles) +
1604 	    nnfiles * sizeof(ntable->fdt_ofiles[0]) +
1605 	    sizeof(struct freetable),
1606 	    M_FILEDESC, M_ZERO | M_WAITOK);
1607 	/* copy the old data */
1608 	ntable->fdt_nfiles = nnfiles;
1609 	memcpy(ntable->fdt_ofiles, otable->fdt_ofiles,
1610 	    onfiles * sizeof(ntable->fdt_ofiles[0]));
1611 
1612 	/*
1613 	 * Allocate a new map only if the old is not large enough.  It will
1614 	 * grow at a slower rate than the table as it can map more
1615 	 * entries than the table can hold.
1616 	 */
1617 	if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) {
1618 		nmap = malloc(NDSLOTS(nnfiles) * NDSLOTSIZE, M_FILEDESC,
1619 		    M_ZERO | M_WAITOK);
1620 		/* copy over the old data and update the pointer */
1621 		memcpy(nmap, omap, NDSLOTS(onfiles) * sizeof(*omap));
1622 		fdp->fd_map = nmap;
1623 	}
1624 
1625 	/*
1626 	 * Make sure that ntable is correctly initialized before we replace
1627 	 * fd_files poiner. Otherwise fget_unlocked() may see inconsistent
1628 	 * data.
1629 	 */
1630 	atomic_store_rel_ptr((volatile void *)&fdp->fd_files, (uintptr_t)ntable);
1631 
1632 	/*
1633 	 * Do not free the old file table, as some threads may still
1634 	 * reference entries within it.  Instead, place it on a freelist
1635 	 * which will be processed when the struct filedesc is released.
1636 	 *
1637 	 * Note that if onfiles == NDFILE, we're dealing with the original
1638 	 * static allocation contained within (struct filedesc0 *)fdp,
1639 	 * which must not be freed.
1640 	 */
1641 	if (onfiles > NDFILE) {
1642 		ft = (struct freetable *)&otable->fdt_ofiles[onfiles];
1643 		fdp0 = (struct filedesc0 *)fdp;
1644 		ft->ft_table = otable;
1645 		SLIST_INSERT_HEAD(&fdp0->fd_free, ft, ft_next);
1646 	}
1647 	/*
1648 	 * The map does not have the same possibility of threads still
1649 	 * holding references to it.  So always free it as long as it
1650 	 * does not reference the original static allocation.
1651 	 */
1652 	if (NDSLOTS(onfiles) > NDSLOTS(NDFILE))
1653 		free(omap, M_FILEDESC);
1654 }
1655 
1656 /*
1657  * Allocate a file descriptor for the process.
1658  */
1659 int
1660 fdalloc(struct thread *td, int minfd, int *result)
1661 {
1662 	struct proc *p = td->td_proc;
1663 	struct filedesc *fdp = p->p_fd;
1664 	int fd, maxfd, allocfd;
1665 #ifdef RACCT
1666 	int error;
1667 #endif
1668 
1669 	FILEDESC_XLOCK_ASSERT(fdp);
1670 
1671 	if (fdp->fd_freefile > minfd)
1672 		minfd = fdp->fd_freefile;
1673 
1674 	maxfd = getmaxfd(td);
1675 
1676 	/*
1677 	 * Search the bitmap for a free descriptor starting at minfd.
1678 	 * If none is found, grow the file table.
1679 	 */
1680 	fd = fd_first_free(fdp, minfd, fdp->fd_nfiles);
1681 	if (fd >= maxfd)
1682 		return (EMFILE);
1683 	if (fd >= fdp->fd_nfiles) {
1684 		allocfd = min(fd * 2, maxfd);
1685 #ifdef RACCT
1686 		if (racct_enable) {
1687 			PROC_LOCK(p);
1688 			error = racct_set(p, RACCT_NOFILE, allocfd);
1689 			PROC_UNLOCK(p);
1690 			if (error != 0)
1691 				return (EMFILE);
1692 		}
1693 #endif
1694 		/*
1695 		 * fd is already equal to first free descriptor >= minfd, so
1696 		 * we only need to grow the table and we are done.
1697 		 */
1698 		fdgrowtable_exp(fdp, allocfd);
1699 	}
1700 
1701 	/*
1702 	 * Perform some sanity checks, then mark the file descriptor as
1703 	 * used and return it to the caller.
1704 	 */
1705 	KASSERT(fd >= 0 && fd < min(maxfd, fdp->fd_nfiles),
1706 	    ("invalid descriptor %d", fd));
1707 	KASSERT(!fdisused(fdp, fd),
1708 	    ("fd_first_free() returned non-free descriptor"));
1709 	KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
1710 	    ("file descriptor isn't free"));
1711 	fdused(fdp, fd);
1712 	*result = fd;
1713 	return (0);
1714 }
1715 
1716 /*
1717  * Allocate n file descriptors for the process.
1718  */
1719 int
1720 fdallocn(struct thread *td, int minfd, int *fds, int n)
1721 {
1722 	struct proc *p = td->td_proc;
1723 	struct filedesc *fdp = p->p_fd;
1724 	int i;
1725 
1726 	FILEDESC_XLOCK_ASSERT(fdp);
1727 
1728 	for (i = 0; i < n; i++)
1729 		if (fdalloc(td, 0, &fds[i]) != 0)
1730 			break;
1731 
1732 	if (i < n) {
1733 		for (i--; i >= 0; i--)
1734 			fdunused(fdp, fds[i]);
1735 		return (EMFILE);
1736 	}
1737 
1738 	return (0);
1739 }
1740 
1741 /*
1742  * Create a new open file structure and allocate a file descriptor for the
1743  * process that refers to it.  We add one reference to the file for the
1744  * descriptor table and one reference for resultfp. This is to prevent us
1745  * being preempted and the entry in the descriptor table closed after we
1746  * release the FILEDESC lock.
1747  */
1748 int
1749 falloc_caps(struct thread *td, struct file **resultfp, int *resultfd, int flags,
1750     struct filecaps *fcaps)
1751 {
1752 	struct file *fp;
1753 	int error, fd;
1754 
1755 	error = falloc_noinstall(td, &fp);
1756 	if (error)
1757 		return (error);		/* no reference held on error */
1758 
1759 	error = finstall(td, fp, &fd, flags, fcaps);
1760 	if (error) {
1761 		fdrop(fp, td);		/* one reference (fp only) */
1762 		return (error);
1763 	}
1764 
1765 	if (resultfp != NULL)
1766 		*resultfp = fp;		/* copy out result */
1767 	else
1768 		fdrop(fp, td);		/* release local reference */
1769 
1770 	if (resultfd != NULL)
1771 		*resultfd = fd;
1772 
1773 	return (0);
1774 }
1775 
1776 /*
1777  * Create a new open file structure without allocating a file descriptor.
1778  */
1779 int
1780 falloc_noinstall(struct thread *td, struct file **resultfp)
1781 {
1782 	struct file *fp;
1783 	int maxuserfiles = maxfiles - (maxfiles / 20);
1784 	int openfiles_new;
1785 	static struct timeval lastfail;
1786 	static int curfail;
1787 
1788 	KASSERT(resultfp != NULL, ("%s: resultfp == NULL", __func__));
1789 
1790 	openfiles_new = atomic_fetchadd_int(&openfiles, 1) + 1;
1791 	if ((openfiles_new >= maxuserfiles &&
1792 	    priv_check(td, PRIV_MAXFILES) != 0) ||
1793 	    openfiles_new >= maxfiles) {
1794 		atomic_subtract_int(&openfiles, 1);
1795 		if (ppsratecheck(&lastfail, &curfail, 1)) {
1796 			printf("kern.maxfiles limit exceeded by uid %i, (%s) "
1797 			    "please see tuning(7).\n", td->td_ucred->cr_ruid, td->td_proc->p_comm);
1798 		}
1799 		return (ENFILE);
1800 	}
1801 	fp = uma_zalloc(file_zone, M_WAITOK | M_ZERO);
1802 	refcount_init(&fp->f_count, 1);
1803 	fp->f_cred = crhold(td->td_ucred);
1804 	fp->f_ops = &badfileops;
1805 	*resultfp = fp;
1806 	return (0);
1807 }
1808 
1809 /*
1810  * Install a file in a file descriptor table.
1811  */
1812 void
1813 _finstall(struct filedesc *fdp, struct file *fp, int fd, int flags,
1814     struct filecaps *fcaps)
1815 {
1816 	struct filedescent *fde;
1817 
1818 	MPASS(fp != NULL);
1819 	if (fcaps != NULL)
1820 		filecaps_validate(fcaps, __func__);
1821 	FILEDESC_XLOCK_ASSERT(fdp);
1822 
1823 	fde = &fdp->fd_ofiles[fd];
1824 #ifdef CAPABILITIES
1825 	seq_write_begin(&fde->fde_seq);
1826 #endif
1827 	fde->fde_file = fp;
1828 	fde->fde_flags = (flags & O_CLOEXEC) != 0 ? UF_EXCLOSE : 0;
1829 	if (fcaps != NULL)
1830 		filecaps_move(fcaps, &fde->fde_caps);
1831 	else
1832 		filecaps_fill(&fde->fde_caps);
1833 #ifdef CAPABILITIES
1834 	seq_write_end(&fde->fde_seq);
1835 #endif
1836 }
1837 
1838 int
1839 finstall(struct thread *td, struct file *fp, int *fd, int flags,
1840     struct filecaps *fcaps)
1841 {
1842 	struct filedesc *fdp = td->td_proc->p_fd;
1843 	int error;
1844 
1845 	MPASS(fd != NULL);
1846 
1847 	FILEDESC_XLOCK(fdp);
1848 	if ((error = fdalloc(td, 0, fd))) {
1849 		FILEDESC_XUNLOCK(fdp);
1850 		return (error);
1851 	}
1852 	fhold(fp);
1853 	_finstall(fdp, fp, *fd, flags, fcaps);
1854 	FILEDESC_XUNLOCK(fdp);
1855 	return (0);
1856 }
1857 
1858 /*
1859  * Build a new filedesc structure from another.
1860  * Copy the current, root, and jail root vnode references.
1861  *
1862  * If fdp is not NULL, return with it shared locked.
1863  */
1864 struct filedesc *
1865 fdinit(struct filedesc *fdp, bool prepfiles)
1866 {
1867 	struct filedesc0 *newfdp0;
1868 	struct filedesc *newfdp;
1869 
1870 	newfdp0 = uma_zalloc(filedesc0_zone, M_WAITOK | M_ZERO);
1871 	newfdp = &newfdp0->fd_fd;
1872 
1873 	/* Create the file descriptor table. */
1874 	FILEDESC_LOCK_INIT(newfdp);
1875 	refcount_init(&newfdp->fd_refcnt, 1);
1876 	refcount_init(&newfdp->fd_holdcnt, 1);
1877 	newfdp->fd_cmask = CMASK;
1878 	newfdp->fd_map = newfdp0->fd_dmap;
1879 	newfdp->fd_lastfile = -1;
1880 	newfdp->fd_files = (struct fdescenttbl *)&newfdp0->fd_dfiles;
1881 	newfdp->fd_files->fdt_nfiles = NDFILE;
1882 
1883 	if (fdp == NULL)
1884 		return (newfdp);
1885 
1886 	if (prepfiles && fdp->fd_lastfile >= newfdp->fd_nfiles)
1887 		fdgrowtable(newfdp, fdp->fd_lastfile + 1);
1888 
1889 	FILEDESC_SLOCK(fdp);
1890 	newfdp->fd_cdir = fdp->fd_cdir;
1891 	if (newfdp->fd_cdir)
1892 		vrefact(newfdp->fd_cdir);
1893 	newfdp->fd_rdir = fdp->fd_rdir;
1894 	if (newfdp->fd_rdir)
1895 		vrefact(newfdp->fd_rdir);
1896 	newfdp->fd_jdir = fdp->fd_jdir;
1897 	if (newfdp->fd_jdir)
1898 		vrefact(newfdp->fd_jdir);
1899 
1900 	if (!prepfiles) {
1901 		FILEDESC_SUNLOCK(fdp);
1902 	} else {
1903 		while (fdp->fd_lastfile >= newfdp->fd_nfiles) {
1904 			FILEDESC_SUNLOCK(fdp);
1905 			fdgrowtable(newfdp, fdp->fd_lastfile + 1);
1906 			FILEDESC_SLOCK(fdp);
1907 		}
1908 	}
1909 
1910 	return (newfdp);
1911 }
1912 
1913 static struct filedesc *
1914 fdhold(struct proc *p)
1915 {
1916 	struct filedesc *fdp;
1917 
1918 	PROC_LOCK_ASSERT(p, MA_OWNED);
1919 	fdp = p->p_fd;
1920 	if (fdp != NULL)
1921 		refcount_acquire(&fdp->fd_holdcnt);
1922 	return (fdp);
1923 }
1924 
1925 static void
1926 fddrop(struct filedesc *fdp)
1927 {
1928 
1929 	if (fdp->fd_holdcnt > 1) {
1930 		if (refcount_release(&fdp->fd_holdcnt) == 0)
1931 			return;
1932 	}
1933 
1934 	FILEDESC_LOCK_DESTROY(fdp);
1935 	uma_zfree(filedesc0_zone, fdp);
1936 }
1937 
1938 /*
1939  * Share a filedesc structure.
1940  */
1941 struct filedesc *
1942 fdshare(struct filedesc *fdp)
1943 {
1944 
1945 	refcount_acquire(&fdp->fd_refcnt);
1946 	return (fdp);
1947 }
1948 
1949 /*
1950  * Unshare a filedesc structure, if necessary by making a copy
1951  */
1952 void
1953 fdunshare(struct thread *td)
1954 {
1955 	struct filedesc *tmp;
1956 	struct proc *p = td->td_proc;
1957 
1958 	if (p->p_fd->fd_refcnt == 1)
1959 		return;
1960 
1961 	tmp = fdcopy(p->p_fd);
1962 	fdescfree(td);
1963 	p->p_fd = tmp;
1964 }
1965 
1966 void
1967 fdinstall_remapped(struct thread *td, struct filedesc *fdp)
1968 {
1969 
1970 	fdescfree(td);
1971 	td->td_proc->p_fd = fdp;
1972 }
1973 
1974 /*
1975  * Copy a filedesc structure.  A NULL pointer in returns a NULL reference,
1976  * this is to ease callers, not catch errors.
1977  */
1978 struct filedesc *
1979 fdcopy(struct filedesc *fdp)
1980 {
1981 	struct filedesc *newfdp;
1982 	struct filedescent *nfde, *ofde;
1983 	int i;
1984 
1985 	MPASS(fdp != NULL);
1986 
1987 	newfdp = fdinit(fdp, true);
1988 	/* copy all passable descriptors (i.e. not kqueue) */
1989 	newfdp->fd_freefile = -1;
1990 	for (i = 0; i <= fdp->fd_lastfile; ++i) {
1991 		ofde = &fdp->fd_ofiles[i];
1992 		if (ofde->fde_file == NULL ||
1993 		    (ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0) {
1994 			if (newfdp->fd_freefile == -1)
1995 				newfdp->fd_freefile = i;
1996 			continue;
1997 		}
1998 		nfde = &newfdp->fd_ofiles[i];
1999 		*nfde = *ofde;
2000 		filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true);
2001 		fhold(nfde->fde_file);
2002 		fdused_init(newfdp, i);
2003 		newfdp->fd_lastfile = i;
2004 	}
2005 	if (newfdp->fd_freefile == -1)
2006 		newfdp->fd_freefile = i;
2007 	newfdp->fd_cmask = fdp->fd_cmask;
2008 	FILEDESC_SUNLOCK(fdp);
2009 	return (newfdp);
2010 }
2011 
2012 /*
2013  * Copies a filedesc structure, while remapping all file descriptors
2014  * stored inside using a translation table.
2015  *
2016  * File descriptors are copied over to the new file descriptor table,
2017  * regardless of whether the close-on-exec flag is set.
2018  */
2019 int
2020 fdcopy_remapped(struct filedesc *fdp, const int *fds, size_t nfds,
2021     struct filedesc **ret)
2022 {
2023 	struct filedesc *newfdp;
2024 	struct filedescent *nfde, *ofde;
2025 	int error, i;
2026 
2027 	MPASS(fdp != NULL);
2028 
2029 	newfdp = fdinit(fdp, true);
2030 	if (nfds > fdp->fd_lastfile + 1) {
2031 		/* New table cannot be larger than the old one. */
2032 		error = E2BIG;
2033 		goto bad;
2034 	}
2035 	/* Copy all passable descriptors (i.e. not kqueue). */
2036 	newfdp->fd_freefile = nfds;
2037 	for (i = 0; i < nfds; ++i) {
2038 		if (fds[i] < 0 || fds[i] > fdp->fd_lastfile) {
2039 			/* File descriptor out of bounds. */
2040 			error = EBADF;
2041 			goto bad;
2042 		}
2043 		ofde = &fdp->fd_ofiles[fds[i]];
2044 		if (ofde->fde_file == NULL) {
2045 			/* Unused file descriptor. */
2046 			error = EBADF;
2047 			goto bad;
2048 		}
2049 		if ((ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0) {
2050 			/* File descriptor cannot be passed. */
2051 			error = EINVAL;
2052 			goto bad;
2053 		}
2054 		nfde = &newfdp->fd_ofiles[i];
2055 		*nfde = *ofde;
2056 		filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true);
2057 		fhold(nfde->fde_file);
2058 		fdused_init(newfdp, i);
2059 		newfdp->fd_lastfile = i;
2060 	}
2061 	newfdp->fd_cmask = fdp->fd_cmask;
2062 	FILEDESC_SUNLOCK(fdp);
2063 	*ret = newfdp;
2064 	return (0);
2065 bad:
2066 	FILEDESC_SUNLOCK(fdp);
2067 	fdescfree_remapped(newfdp);
2068 	return (error);
2069 }
2070 
2071 /*
2072  * Clear POSIX style locks. This is only used when fdp looses a reference (i.e.
2073  * one of processes using it exits) and the table used to be shared.
2074  */
2075 static void
2076 fdclearlocks(struct thread *td)
2077 {
2078 	struct filedesc *fdp;
2079 	struct filedesc_to_leader *fdtol;
2080 	struct flock lf;
2081 	struct file *fp;
2082 	struct proc *p;
2083 	struct vnode *vp;
2084 	int i;
2085 
2086 	p = td->td_proc;
2087 	fdp = p->p_fd;
2088 	fdtol = p->p_fdtol;
2089 	MPASS(fdtol != NULL);
2090 
2091 	FILEDESC_XLOCK(fdp);
2092 	KASSERT(fdtol->fdl_refcount > 0,
2093 	    ("filedesc_to_refcount botch: fdl_refcount=%d",
2094 	    fdtol->fdl_refcount));
2095 	if (fdtol->fdl_refcount == 1 &&
2096 	    (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2097 		for (i = 0; i <= fdp->fd_lastfile; i++) {
2098 			fp = fdp->fd_ofiles[i].fde_file;
2099 			if (fp == NULL || fp->f_type != DTYPE_VNODE)
2100 				continue;
2101 			fhold(fp);
2102 			FILEDESC_XUNLOCK(fdp);
2103 			lf.l_whence = SEEK_SET;
2104 			lf.l_start = 0;
2105 			lf.l_len = 0;
2106 			lf.l_type = F_UNLCK;
2107 			vp = fp->f_vnode;
2108 			(void) VOP_ADVLOCK(vp,
2109 			    (caddr_t)p->p_leader, F_UNLCK,
2110 			    &lf, F_POSIX);
2111 			FILEDESC_XLOCK(fdp);
2112 			fdrop(fp, td);
2113 		}
2114 	}
2115 retry:
2116 	if (fdtol->fdl_refcount == 1) {
2117 		if (fdp->fd_holdleaderscount > 0 &&
2118 		    (p->p_leader->p_flag & P_ADVLOCK) != 0) {
2119 			/*
2120 			 * close() or kern_dup() has cleared a reference
2121 			 * in a shared file descriptor table.
2122 			 */
2123 			fdp->fd_holdleaderswakeup = 1;
2124 			sx_sleep(&fdp->fd_holdleaderscount,
2125 			    FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0);
2126 			goto retry;
2127 		}
2128 		if (fdtol->fdl_holdcount > 0) {
2129 			/*
2130 			 * Ensure that fdtol->fdl_leader remains
2131 			 * valid in closef().
2132 			 */
2133 			fdtol->fdl_wakeup = 1;
2134 			sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK,
2135 			    "fdlhold", 0);
2136 			goto retry;
2137 		}
2138 	}
2139 	fdtol->fdl_refcount--;
2140 	if (fdtol->fdl_refcount == 0 &&
2141 	    fdtol->fdl_holdcount == 0) {
2142 		fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
2143 		fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
2144 	} else
2145 		fdtol = NULL;
2146 	p->p_fdtol = NULL;
2147 	FILEDESC_XUNLOCK(fdp);
2148 	if (fdtol != NULL)
2149 		free(fdtol, M_FILEDESC_TO_LEADER);
2150 }
2151 
2152 /*
2153  * Release a filedesc structure.
2154  */
2155 static void
2156 fdescfree_fds(struct thread *td, struct filedesc *fdp, bool needclose)
2157 {
2158 	struct filedesc0 *fdp0;
2159 	struct freetable *ft, *tft;
2160 	struct filedescent *fde;
2161 	struct file *fp;
2162 	int i;
2163 
2164 	for (i = 0; i <= fdp->fd_lastfile; i++) {
2165 		fde = &fdp->fd_ofiles[i];
2166 		fp = fde->fde_file;
2167 		if (fp != NULL) {
2168 			fdefree_last(fde);
2169 			if (needclose)
2170 				(void) closef(fp, td);
2171 			else
2172 				fdrop(fp, td);
2173 		}
2174 	}
2175 
2176 	if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE))
2177 		free(fdp->fd_map, M_FILEDESC);
2178 	if (fdp->fd_nfiles > NDFILE)
2179 		free(fdp->fd_files, M_FILEDESC);
2180 
2181 	fdp0 = (struct filedesc0 *)fdp;
2182 	SLIST_FOREACH_SAFE(ft, &fdp0->fd_free, ft_next, tft)
2183 		free(ft->ft_table, M_FILEDESC);
2184 
2185 	fddrop(fdp);
2186 }
2187 
2188 void
2189 fdescfree(struct thread *td)
2190 {
2191 	struct proc *p;
2192 	struct filedesc *fdp;
2193 	struct vnode *cdir, *jdir, *rdir;
2194 
2195 	p = td->td_proc;
2196 	fdp = p->p_fd;
2197 	MPASS(fdp != NULL);
2198 
2199 #ifdef RACCT
2200 	if (racct_enable) {
2201 		PROC_LOCK(p);
2202 		racct_set(p, RACCT_NOFILE, 0);
2203 		PROC_UNLOCK(p);
2204 	}
2205 #endif
2206 
2207 	if (p->p_fdtol != NULL)
2208 		fdclearlocks(td);
2209 
2210 	PROC_LOCK(p);
2211 	p->p_fd = NULL;
2212 	PROC_UNLOCK(p);
2213 
2214 	if (refcount_release(&fdp->fd_refcnt) == 0)
2215 		return;
2216 
2217 	FILEDESC_XLOCK(fdp);
2218 	cdir = fdp->fd_cdir;
2219 	fdp->fd_cdir = NULL;
2220 	rdir = fdp->fd_rdir;
2221 	fdp->fd_rdir = NULL;
2222 	jdir = fdp->fd_jdir;
2223 	fdp->fd_jdir = NULL;
2224 	FILEDESC_XUNLOCK(fdp);
2225 
2226 	if (cdir != NULL)
2227 		vrele(cdir);
2228 	if (rdir != NULL)
2229 		vrele(rdir);
2230 	if (jdir != NULL)
2231 		vrele(jdir);
2232 
2233 	fdescfree_fds(td, fdp, 1);
2234 }
2235 
2236 void
2237 fdescfree_remapped(struct filedesc *fdp)
2238 {
2239 
2240 	if (fdp->fd_cdir != NULL)
2241 		vrele(fdp->fd_cdir);
2242 	if (fdp->fd_rdir != NULL)
2243 		vrele(fdp->fd_rdir);
2244 	if (fdp->fd_jdir != NULL)
2245 		vrele(fdp->fd_jdir);
2246 
2247 	fdescfree_fds(curthread, fdp, 0);
2248 }
2249 
2250 /*
2251  * For setugid programs, we don't want to people to use that setugidness
2252  * to generate error messages which write to a file which otherwise would
2253  * otherwise be off-limits to the process.  We check for filesystems where
2254  * the vnode can change out from under us after execve (like [lin]procfs).
2255  *
2256  * Since fdsetugidsafety calls this only for fd 0, 1 and 2, this check is
2257  * sufficient.  We also don't check for setugidness since we know we are.
2258  */
2259 static bool
2260 is_unsafe(struct file *fp)
2261 {
2262 	struct vnode *vp;
2263 
2264 	if (fp->f_type != DTYPE_VNODE)
2265 		return (false);
2266 
2267 	vp = fp->f_vnode;
2268 	return ((vp->v_vflag & VV_PROCDEP) != 0);
2269 }
2270 
2271 /*
2272  * Make this setguid thing safe, if at all possible.
2273  */
2274 void
2275 fdsetugidsafety(struct thread *td)
2276 {
2277 	struct filedesc *fdp;
2278 	struct file *fp;
2279 	int i;
2280 
2281 	fdp = td->td_proc->p_fd;
2282 	KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2283 	MPASS(fdp->fd_nfiles >= 3);
2284 	for (i = 0; i <= 2; i++) {
2285 		fp = fdp->fd_ofiles[i].fde_file;
2286 		if (fp != NULL && is_unsafe(fp)) {
2287 			FILEDESC_XLOCK(fdp);
2288 			knote_fdclose(td, i);
2289 			/*
2290 			 * NULL-out descriptor prior to close to avoid
2291 			 * a race while close blocks.
2292 			 */
2293 			fdfree(fdp, i);
2294 			FILEDESC_XUNLOCK(fdp);
2295 			(void) closef(fp, td);
2296 		}
2297 	}
2298 }
2299 
2300 /*
2301  * If a specific file object occupies a specific file descriptor, close the
2302  * file descriptor entry and drop a reference on the file object.  This is a
2303  * convenience function to handle a subsequent error in a function that calls
2304  * falloc() that handles the race that another thread might have closed the
2305  * file descriptor out from under the thread creating the file object.
2306  */
2307 void
2308 fdclose(struct thread *td, struct file *fp, int idx)
2309 {
2310 	struct filedesc *fdp = td->td_proc->p_fd;
2311 
2312 	FILEDESC_XLOCK(fdp);
2313 	if (fdp->fd_ofiles[idx].fde_file == fp) {
2314 		fdfree(fdp, idx);
2315 		FILEDESC_XUNLOCK(fdp);
2316 		fdrop(fp, td);
2317 	} else
2318 		FILEDESC_XUNLOCK(fdp);
2319 }
2320 
2321 /*
2322  * Close any files on exec?
2323  */
2324 void
2325 fdcloseexec(struct thread *td)
2326 {
2327 	struct filedesc *fdp;
2328 	struct filedescent *fde;
2329 	struct file *fp;
2330 	int i;
2331 
2332 	fdp = td->td_proc->p_fd;
2333 	KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2334 	for (i = 0; i <= fdp->fd_lastfile; i++) {
2335 		fde = &fdp->fd_ofiles[i];
2336 		fp = fde->fde_file;
2337 		if (fp != NULL && (fp->f_type == DTYPE_MQUEUE ||
2338 		    (fde->fde_flags & UF_EXCLOSE))) {
2339 			FILEDESC_XLOCK(fdp);
2340 			fdfree(fdp, i);
2341 			(void) closefp(fdp, i, fp, td, 0);
2342 			FILEDESC_UNLOCK_ASSERT(fdp);
2343 		}
2344 	}
2345 }
2346 
2347 /*
2348  * It is unsafe for set[ug]id processes to be started with file
2349  * descriptors 0..2 closed, as these descriptors are given implicit
2350  * significance in the Standard C library.  fdcheckstd() will create a
2351  * descriptor referencing /dev/null for each of stdin, stdout, and
2352  * stderr that is not already open.
2353  */
2354 int
2355 fdcheckstd(struct thread *td)
2356 {
2357 	struct filedesc *fdp;
2358 	register_t save;
2359 	int i, error, devnull;
2360 
2361 	fdp = td->td_proc->p_fd;
2362 	KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2363 	MPASS(fdp->fd_nfiles >= 3);
2364 	devnull = -1;
2365 	for (i = 0; i <= 2; i++) {
2366 		if (fdp->fd_ofiles[i].fde_file != NULL)
2367 			continue;
2368 
2369 		save = td->td_retval[0];
2370 		if (devnull != -1) {
2371 			error = kern_dup(td, FDDUP_FIXED, 0, devnull, i);
2372 		} else {
2373 			error = kern_openat(td, AT_FDCWD, "/dev/null",
2374 			    UIO_SYSSPACE, O_RDWR, 0);
2375 			if (error == 0) {
2376 				devnull = td->td_retval[0];
2377 				KASSERT(devnull == i, ("we didn't get our fd"));
2378 			}
2379 		}
2380 		td->td_retval[0] = save;
2381 		if (error != 0)
2382 			return (error);
2383 	}
2384 	return (0);
2385 }
2386 
2387 /*
2388  * Internal form of close.  Decrement reference count on file structure.
2389  * Note: td may be NULL when closing a file that was being passed in a
2390  * message.
2391  *
2392  * XXXRW: Giant is not required for the caller, but often will be held; this
2393  * makes it moderately likely the Giant will be recursed in the VFS case.
2394  */
2395 int
2396 closef(struct file *fp, struct thread *td)
2397 {
2398 	struct vnode *vp;
2399 	struct flock lf;
2400 	struct filedesc_to_leader *fdtol;
2401 	struct filedesc *fdp;
2402 
2403 	/*
2404 	 * POSIX record locking dictates that any close releases ALL
2405 	 * locks owned by this process.  This is handled by setting
2406 	 * a flag in the unlock to free ONLY locks obeying POSIX
2407 	 * semantics, and not to free BSD-style file locks.
2408 	 * If the descriptor was in a message, POSIX-style locks
2409 	 * aren't passed with the descriptor, and the thread pointer
2410 	 * will be NULL.  Callers should be careful only to pass a
2411 	 * NULL thread pointer when there really is no owning
2412 	 * context that might have locks, or the locks will be
2413 	 * leaked.
2414 	 */
2415 	if (fp->f_type == DTYPE_VNODE && td != NULL) {
2416 		vp = fp->f_vnode;
2417 		if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
2418 			lf.l_whence = SEEK_SET;
2419 			lf.l_start = 0;
2420 			lf.l_len = 0;
2421 			lf.l_type = F_UNLCK;
2422 			(void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader,
2423 			    F_UNLCK, &lf, F_POSIX);
2424 		}
2425 		fdtol = td->td_proc->p_fdtol;
2426 		if (fdtol != NULL) {
2427 			/*
2428 			 * Handle special case where file descriptor table is
2429 			 * shared between multiple process leaders.
2430 			 */
2431 			fdp = td->td_proc->p_fd;
2432 			FILEDESC_XLOCK(fdp);
2433 			for (fdtol = fdtol->fdl_next;
2434 			    fdtol != td->td_proc->p_fdtol;
2435 			    fdtol = fdtol->fdl_next) {
2436 				if ((fdtol->fdl_leader->p_flag &
2437 				    P_ADVLOCK) == 0)
2438 					continue;
2439 				fdtol->fdl_holdcount++;
2440 				FILEDESC_XUNLOCK(fdp);
2441 				lf.l_whence = SEEK_SET;
2442 				lf.l_start = 0;
2443 				lf.l_len = 0;
2444 				lf.l_type = F_UNLCK;
2445 				vp = fp->f_vnode;
2446 				(void) VOP_ADVLOCK(vp,
2447 				    (caddr_t)fdtol->fdl_leader, F_UNLCK, &lf,
2448 				    F_POSIX);
2449 				FILEDESC_XLOCK(fdp);
2450 				fdtol->fdl_holdcount--;
2451 				if (fdtol->fdl_holdcount == 0 &&
2452 				    fdtol->fdl_wakeup != 0) {
2453 					fdtol->fdl_wakeup = 0;
2454 					wakeup(fdtol);
2455 				}
2456 			}
2457 			FILEDESC_XUNLOCK(fdp);
2458 		}
2459 	}
2460 	return (fdrop(fp, td));
2461 }
2462 
2463 /*
2464  * Initialize the file pointer with the specified properties.
2465  *
2466  * The ops are set with release semantics to be certain that the flags, type,
2467  * and data are visible when ops is.  This is to prevent ops methods from being
2468  * called with bad data.
2469  */
2470 void
2471 finit(struct file *fp, u_int flag, short type, void *data, struct fileops *ops)
2472 {
2473 	fp->f_data = data;
2474 	fp->f_flag = flag;
2475 	fp->f_type = type;
2476 	atomic_store_rel_ptr((volatile uintptr_t *)&fp->f_ops, (uintptr_t)ops);
2477 }
2478 
2479 int
2480 fget_cap_locked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
2481     struct file **fpp, struct filecaps *havecapsp)
2482 {
2483 	struct filedescent *fde;
2484 	int error;
2485 
2486 	FILEDESC_LOCK_ASSERT(fdp);
2487 
2488 	fde = fdeget_locked(fdp, fd);
2489 	if (fde == NULL) {
2490 		error = EBADF;
2491 		goto out;
2492 	}
2493 
2494 #ifdef CAPABILITIES
2495 	error = cap_check(cap_rights_fde(fde), needrightsp);
2496 	if (error != 0)
2497 		goto out;
2498 #endif
2499 
2500 	if (havecapsp != NULL)
2501 		filecaps_copy(&fde->fde_caps, havecapsp, true);
2502 
2503 	*fpp = fde->fde_file;
2504 
2505 	error = 0;
2506 out:
2507 	return (error);
2508 }
2509 
2510 int
2511 fget_cap(struct thread *td, int fd, cap_rights_t *needrightsp,
2512     struct file **fpp, struct filecaps *havecapsp)
2513 {
2514 	struct filedesc *fdp = td->td_proc->p_fd;
2515 	int error;
2516 #ifndef CAPABILITIES
2517 	error = fget_unlocked(fdp, fd, needrightsp, fpp, NULL);
2518 	if (error == 0 && havecapsp != NULL)
2519 		filecaps_fill(havecapsp);
2520 #else
2521 	struct file *fp;
2522 	seq_t seq;
2523 
2524 	for (;;) {
2525 		error = fget_unlocked(fdp, fd, needrightsp, &fp, &seq);
2526 		if (error != 0)
2527 			return (error);
2528 
2529 		if (havecapsp != NULL) {
2530 			if (!filecaps_copy(&fdp->fd_ofiles[fd].fde_caps,
2531 			    havecapsp, false)) {
2532 				fdrop(fp, td);
2533 				goto get_locked;
2534 			}
2535 		}
2536 
2537 		if (!fd_modified(fdp, fd, seq))
2538 			break;
2539 		fdrop(fp, td);
2540 	}
2541 
2542 	*fpp = fp;
2543 	return (0);
2544 
2545 get_locked:
2546 	FILEDESC_SLOCK(fdp);
2547 	error = fget_cap_locked(fdp, fd, needrightsp, fpp, havecapsp);
2548 	if (error == 0)
2549 		fhold(*fpp);
2550 	FILEDESC_SUNLOCK(fdp);
2551 #endif
2552 	return (error);
2553 }
2554 
2555 int
2556 fget_unlocked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
2557     struct file **fpp, seq_t *seqp)
2558 {
2559 #ifdef CAPABILITIES
2560 	struct filedescent *fde;
2561 #endif
2562 	struct fdescenttbl *fdt;
2563 	struct file *fp;
2564 	u_int count;
2565 #ifdef CAPABILITIES
2566 	seq_t seq;
2567 	cap_rights_t haverights;
2568 	int error;
2569 #endif
2570 
2571 	fdt = fdp->fd_files;
2572 	if ((u_int)fd >= fdt->fdt_nfiles)
2573 		return (EBADF);
2574 	/*
2575 	 * Fetch the descriptor locklessly.  We avoid fdrop() races by
2576 	 * never raising a refcount above 0.  To accomplish this we have
2577 	 * to use a cmpset loop rather than an atomic_add.  The descriptor
2578 	 * must be re-verified once we acquire a reference to be certain
2579 	 * that the identity is still correct and we did not lose a race
2580 	 * due to preemption.
2581 	 */
2582 	for (;;) {
2583 #ifdef CAPABILITIES
2584 		seq = seq_read(fd_seq(fdt, fd));
2585 		fde = &fdt->fdt_ofiles[fd];
2586 		haverights = *cap_rights_fde(fde);
2587 		fp = fde->fde_file;
2588 		if (!seq_consistent(fd_seq(fdt, fd), seq))
2589 			continue;
2590 #else
2591 		fp = fdt->fdt_ofiles[fd].fde_file;
2592 #endif
2593 		if (fp == NULL)
2594 			return (EBADF);
2595 #ifdef CAPABILITIES
2596 		error = cap_check(&haverights, needrightsp);
2597 		if (error != 0)
2598 			return (error);
2599 #endif
2600 		count = fp->f_count;
2601 	retry:
2602 		if (count == 0) {
2603 			/*
2604 			 * Force a reload. Other thread could reallocate the
2605 			 * table before this fd was closed, so it possible that
2606 			 * there is a stale fp pointer in cached version.
2607 			 */
2608 			fdt = *(struct fdescenttbl * volatile *)&(fdp->fd_files);
2609 			continue;
2610 		}
2611 		/*
2612 		 * Use an acquire barrier to force re-reading of fdt so it is
2613 		 * refreshed for verification.
2614 		 */
2615 		if (atomic_fcmpset_acq_int(&fp->f_count, &count, count + 1) == 0)
2616 			goto retry;
2617 		fdt = fdp->fd_files;
2618 #ifdef	CAPABILITIES
2619 		if (seq_consistent_nomb(fd_seq(fdt, fd), seq))
2620 #else
2621 		if (fp == fdt->fdt_ofiles[fd].fde_file)
2622 #endif
2623 			break;
2624 		fdrop(fp, curthread);
2625 	}
2626 	*fpp = fp;
2627 	if (seqp != NULL) {
2628 #ifdef CAPABILITIES
2629 		*seqp = seq;
2630 #endif
2631 	}
2632 	return (0);
2633 }
2634 
2635 /*
2636  * Extract the file pointer associated with the specified descriptor for the
2637  * current user process.
2638  *
2639  * If the descriptor doesn't exist or doesn't match 'flags', EBADF is
2640  * returned.
2641  *
2642  * File's rights will be checked against the capability rights mask.
2643  *
2644  * If an error occurred the non-zero error is returned and *fpp is set to
2645  * NULL.  Otherwise *fpp is held and set and zero is returned.  Caller is
2646  * responsible for fdrop().
2647  */
2648 static __inline int
2649 _fget(struct thread *td, int fd, struct file **fpp, int flags,
2650     cap_rights_t *needrightsp, seq_t *seqp)
2651 {
2652 	struct filedesc *fdp;
2653 	struct file *fp;
2654 	int error;
2655 
2656 	*fpp = NULL;
2657 	fdp = td->td_proc->p_fd;
2658 	error = fget_unlocked(fdp, fd, needrightsp, &fp, seqp);
2659 	if (error != 0)
2660 		return (error);
2661 	if (fp->f_ops == &badfileops) {
2662 		fdrop(fp, td);
2663 		return (EBADF);
2664 	}
2665 
2666 	/*
2667 	 * FREAD and FWRITE failure return EBADF as per POSIX.
2668 	 */
2669 	error = 0;
2670 	switch (flags) {
2671 	case FREAD:
2672 	case FWRITE:
2673 		if ((fp->f_flag & flags) == 0)
2674 			error = EBADF;
2675 		break;
2676 	case FEXEC:
2677 	    	if ((fp->f_flag & (FREAD | FEXEC)) == 0 ||
2678 		    ((fp->f_flag & FWRITE) != 0))
2679 			error = EBADF;
2680 		break;
2681 	case 0:
2682 		break;
2683 	default:
2684 		KASSERT(0, ("wrong flags"));
2685 	}
2686 
2687 	if (error != 0) {
2688 		fdrop(fp, td);
2689 		return (error);
2690 	}
2691 
2692 	*fpp = fp;
2693 	return (0);
2694 }
2695 
2696 int
2697 fget(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
2698 {
2699 
2700 	return (_fget(td, fd, fpp, 0, rightsp, NULL));
2701 }
2702 
2703 int
2704 fget_mmap(struct thread *td, int fd, cap_rights_t *rightsp, u_char *maxprotp,
2705     struct file **fpp)
2706 {
2707 	int error;
2708 #ifndef CAPABILITIES
2709 	error = _fget(td, fd, fpp, 0, rightsp, NULL);
2710 	if (maxprotp != NULL)
2711 		*maxprotp = VM_PROT_ALL;
2712 #else
2713 	struct filedesc *fdp = td->td_proc->p_fd;
2714 	seq_t seq;
2715 
2716 	MPASS(cap_rights_is_set(rightsp, CAP_MMAP));
2717 	for (;;) {
2718 		error = _fget(td, fd, fpp, 0, rightsp, &seq);
2719 		if (error != 0)
2720 			return (error);
2721 		/*
2722 		 * If requested, convert capability rights to access flags.
2723 		 */
2724 		if (maxprotp != NULL)
2725 			*maxprotp = cap_rights_to_vmprot(cap_rights(fdp, fd));
2726 		if (!fd_modified(fdp, fd, seq))
2727 			break;
2728 		fdrop(*fpp, td);
2729 	}
2730 #endif
2731 	return (error);
2732 }
2733 
2734 int
2735 fget_read(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
2736 {
2737 
2738 	return (_fget(td, fd, fpp, FREAD, rightsp, NULL));
2739 }
2740 
2741 int
2742 fget_write(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
2743 {
2744 
2745 	return (_fget(td, fd, fpp, FWRITE, rightsp, NULL));
2746 }
2747 
2748 int
2749 fget_fcntl(struct thread *td, int fd, cap_rights_t *rightsp, int needfcntl,
2750     struct file **fpp)
2751 {
2752 	struct filedesc *fdp = td->td_proc->p_fd;
2753 #ifndef CAPABILITIES
2754 	return (fget_unlocked(fdp, fd, rightsp, fpp, NULL));
2755 #else
2756 	int error;
2757 	seq_t seq;
2758 
2759 	MPASS(cap_rights_is_set(rightsp, CAP_FCNTL));
2760 	for (;;) {
2761 		error = fget_unlocked(fdp, fd, rightsp, fpp, &seq);
2762 		if (error != 0)
2763 			return (error);
2764 		error = cap_fcntl_check(fdp, fd, needfcntl);
2765 		if (!fd_modified(fdp, fd, seq))
2766 			break;
2767 		fdrop(*fpp, td);
2768 	}
2769 	if (error != 0) {
2770 		fdrop(*fpp, td);
2771 		*fpp = NULL;
2772 	}
2773 	return (error);
2774 #endif
2775 }
2776 
2777 /*
2778  * Like fget() but loads the underlying vnode, or returns an error if the
2779  * descriptor does not represent a vnode.  Note that pipes use vnodes but
2780  * never have VM objects.  The returned vnode will be vref()'d.
2781  *
2782  * XXX: what about the unused flags ?
2783  */
2784 static __inline int
2785 _fgetvp(struct thread *td, int fd, int flags, cap_rights_t *needrightsp,
2786     struct vnode **vpp)
2787 {
2788 	struct file *fp;
2789 	int error;
2790 
2791 	*vpp = NULL;
2792 	error = _fget(td, fd, &fp, flags, needrightsp, NULL);
2793 	if (error != 0)
2794 		return (error);
2795 	if (fp->f_vnode == NULL) {
2796 		error = EINVAL;
2797 	} else {
2798 		*vpp = fp->f_vnode;
2799 		vrefact(*vpp);
2800 	}
2801 	fdrop(fp, td);
2802 
2803 	return (error);
2804 }
2805 
2806 int
2807 fgetvp(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
2808 {
2809 
2810 	return (_fgetvp(td, fd, 0, rightsp, vpp));
2811 }
2812 
2813 int
2814 fgetvp_rights(struct thread *td, int fd, cap_rights_t *needrightsp,
2815     struct filecaps *havecaps, struct vnode **vpp)
2816 {
2817 	struct filedesc *fdp;
2818 	struct filecaps caps;
2819 	struct file *fp;
2820 	int error;
2821 
2822 	fdp = td->td_proc->p_fd;
2823 	error = fget_cap_locked(fdp, fd, needrightsp, &fp, &caps);
2824 	if (error != 0)
2825 		return (error);
2826 	if (fp->f_ops == &badfileops) {
2827 		error = EBADF;
2828 		goto out;
2829 	}
2830 	if (fp->f_vnode == NULL) {
2831 		error = EINVAL;
2832 		goto out;
2833 	}
2834 
2835 	*havecaps = caps;
2836 	*vpp = fp->f_vnode;
2837 	vrefact(*vpp);
2838 
2839 	return (0);
2840 out:
2841 	filecaps_free(&caps);
2842 	return (error);
2843 }
2844 
2845 int
2846 fgetvp_read(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
2847 {
2848 
2849 	return (_fgetvp(td, fd, FREAD, rightsp, vpp));
2850 }
2851 
2852 int
2853 fgetvp_exec(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
2854 {
2855 
2856 	return (_fgetvp(td, fd, FEXEC, rightsp, vpp));
2857 }
2858 
2859 #ifdef notyet
2860 int
2861 fgetvp_write(struct thread *td, int fd, cap_rights_t *rightsp,
2862     struct vnode **vpp)
2863 {
2864 
2865 	return (_fgetvp(td, fd, FWRITE, rightsp, vpp));
2866 }
2867 #endif
2868 
2869 /*
2870  * Handle the last reference to a file being closed.
2871  */
2872 int
2873 _fdrop(struct file *fp, struct thread *td)
2874 {
2875 	int error;
2876 
2877 	if (fp->f_count != 0)
2878 		panic("fdrop: count %d", fp->f_count);
2879 	error = fo_close(fp, td);
2880 	atomic_subtract_int(&openfiles, 1);
2881 	crfree(fp->f_cred);
2882 	free(fp->f_advice, M_FADVISE);
2883 	uma_zfree(file_zone, fp);
2884 
2885 	return (error);
2886 }
2887 
2888 /*
2889  * Apply an advisory lock on a file descriptor.
2890  *
2891  * Just attempt to get a record lock of the requested type on the entire file
2892  * (l_whence = SEEK_SET, l_start = 0, l_len = 0).
2893  */
2894 #ifndef _SYS_SYSPROTO_H_
2895 struct flock_args {
2896 	int	fd;
2897 	int	how;
2898 };
2899 #endif
2900 /* ARGSUSED */
2901 int
2902 sys_flock(struct thread *td, struct flock_args *uap)
2903 {
2904 	struct file *fp;
2905 	struct vnode *vp;
2906 	struct flock lf;
2907 	cap_rights_t rights;
2908 	int error;
2909 
2910 	error = fget(td, uap->fd, cap_rights_init(&rights, CAP_FLOCK), &fp);
2911 	if (error != 0)
2912 		return (error);
2913 	if (fp->f_type != DTYPE_VNODE) {
2914 		fdrop(fp, td);
2915 		return (EOPNOTSUPP);
2916 	}
2917 
2918 	vp = fp->f_vnode;
2919 	lf.l_whence = SEEK_SET;
2920 	lf.l_start = 0;
2921 	lf.l_len = 0;
2922 	if (uap->how & LOCK_UN) {
2923 		lf.l_type = F_UNLCK;
2924 		atomic_clear_int(&fp->f_flag, FHASLOCK);
2925 		error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK);
2926 		goto done2;
2927 	}
2928 	if (uap->how & LOCK_EX)
2929 		lf.l_type = F_WRLCK;
2930 	else if (uap->how & LOCK_SH)
2931 		lf.l_type = F_RDLCK;
2932 	else {
2933 		error = EBADF;
2934 		goto done2;
2935 	}
2936 	atomic_set_int(&fp->f_flag, FHASLOCK);
2937 	error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf,
2938 	    (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT);
2939 done2:
2940 	fdrop(fp, td);
2941 	return (error);
2942 }
2943 /*
2944  * Duplicate the specified descriptor to a free descriptor.
2945  */
2946 int
2947 dupfdopen(struct thread *td, struct filedesc *fdp, int dfd, int mode,
2948     int openerror, int *indxp)
2949 {
2950 	struct filedescent *newfde, *oldfde;
2951 	struct file *fp;
2952 	int error, indx;
2953 
2954 	KASSERT(openerror == ENODEV || openerror == ENXIO,
2955 	    ("unexpected error %d in %s", openerror, __func__));
2956 
2957 	/*
2958 	 * If the to-be-dup'd fd number is greater than the allowed number
2959 	 * of file descriptors, or the fd to be dup'd has already been
2960 	 * closed, then reject.
2961 	 */
2962 	FILEDESC_XLOCK(fdp);
2963 	if ((fp = fget_locked(fdp, dfd)) == NULL) {
2964 		FILEDESC_XUNLOCK(fdp);
2965 		return (EBADF);
2966 	}
2967 
2968 	error = fdalloc(td, 0, &indx);
2969 	if (error != 0) {
2970 		FILEDESC_XUNLOCK(fdp);
2971 		return (error);
2972 	}
2973 
2974 	/*
2975 	 * There are two cases of interest here.
2976 	 *
2977 	 * For ENODEV simply dup (dfd) to file descriptor (indx) and return.
2978 	 *
2979 	 * For ENXIO steal away the file structure from (dfd) and store it in
2980 	 * (indx).  (dfd) is effectively closed by this operation.
2981 	 */
2982 	switch (openerror) {
2983 	case ENODEV:
2984 		/*
2985 		 * Check that the mode the file is being opened for is a
2986 		 * subset of the mode of the existing descriptor.
2987 		 */
2988 		if (((mode & (FREAD|FWRITE)) | fp->f_flag) != fp->f_flag) {
2989 			fdunused(fdp, indx);
2990 			FILEDESC_XUNLOCK(fdp);
2991 			return (EACCES);
2992 		}
2993 		fhold(fp);
2994 		newfde = &fdp->fd_ofiles[indx];
2995 		oldfde = &fdp->fd_ofiles[dfd];
2996 #ifdef CAPABILITIES
2997 		seq_write_begin(&newfde->fde_seq);
2998 #endif
2999 		memcpy(newfde, oldfde, fde_change_size);
3000 		filecaps_copy(&oldfde->fde_caps, &newfde->fde_caps, true);
3001 #ifdef CAPABILITIES
3002 		seq_write_end(&newfde->fde_seq);
3003 #endif
3004 		break;
3005 	case ENXIO:
3006 		/*
3007 		 * Steal away the file pointer from dfd and stuff it into indx.
3008 		 */
3009 		newfde = &fdp->fd_ofiles[indx];
3010 		oldfde = &fdp->fd_ofiles[dfd];
3011 #ifdef CAPABILITIES
3012 		seq_write_begin(&newfde->fde_seq);
3013 #endif
3014 		memcpy(newfde, oldfde, fde_change_size);
3015 		oldfde->fde_file = NULL;
3016 		fdunused(fdp, dfd);
3017 #ifdef CAPABILITIES
3018 		seq_write_end(&newfde->fde_seq);
3019 #endif
3020 		break;
3021 	}
3022 	FILEDESC_XUNLOCK(fdp);
3023 	*indxp = indx;
3024 	return (0);
3025 }
3026 
3027 /*
3028  * This sysctl determines if we will allow a process to chroot(2) if it
3029  * has a directory open:
3030  *	0: disallowed for all processes.
3031  *	1: allowed for processes that were not already chroot(2)'ed.
3032  *	2: allowed for all processes.
3033  */
3034 
3035 static int chroot_allow_open_directories = 1;
3036 
3037 SYSCTL_INT(_kern, OID_AUTO, chroot_allow_open_directories, CTLFLAG_RW,
3038     &chroot_allow_open_directories, 0,
3039     "Allow a process to chroot(2) if it has a directory open");
3040 
3041 /*
3042  * Helper function for raised chroot(2) security function:  Refuse if
3043  * any filedescriptors are open directories.
3044  */
3045 static int
3046 chroot_refuse_vdir_fds(struct filedesc *fdp)
3047 {
3048 	struct vnode *vp;
3049 	struct file *fp;
3050 	int fd;
3051 
3052 	FILEDESC_LOCK_ASSERT(fdp);
3053 
3054 	for (fd = 0; fd <= fdp->fd_lastfile; fd++) {
3055 		fp = fget_locked(fdp, fd);
3056 		if (fp == NULL)
3057 			continue;
3058 		if (fp->f_type == DTYPE_VNODE) {
3059 			vp = fp->f_vnode;
3060 			if (vp->v_type == VDIR)
3061 				return (EPERM);
3062 		}
3063 	}
3064 	return (0);
3065 }
3066 
3067 /*
3068  * Common routine for kern_chroot() and jail_attach().  The caller is
3069  * responsible for invoking priv_check() and mac_vnode_check_chroot() to
3070  * authorize this operation.
3071  */
3072 int
3073 pwd_chroot(struct thread *td, struct vnode *vp)
3074 {
3075 	struct filedesc *fdp;
3076 	struct vnode *oldvp;
3077 	int error;
3078 
3079 	fdp = td->td_proc->p_fd;
3080 	FILEDESC_XLOCK(fdp);
3081 	if (chroot_allow_open_directories == 0 ||
3082 	    (chroot_allow_open_directories == 1 && fdp->fd_rdir != rootvnode)) {
3083 		error = chroot_refuse_vdir_fds(fdp);
3084 		if (error != 0) {
3085 			FILEDESC_XUNLOCK(fdp);
3086 			return (error);
3087 		}
3088 	}
3089 	oldvp = fdp->fd_rdir;
3090 	vrefact(vp);
3091 	fdp->fd_rdir = vp;
3092 	if (fdp->fd_jdir == NULL) {
3093 		vrefact(vp);
3094 		fdp->fd_jdir = vp;
3095 	}
3096 	FILEDESC_XUNLOCK(fdp);
3097 	vrele(oldvp);
3098 	return (0);
3099 }
3100 
3101 void
3102 pwd_chdir(struct thread *td, struct vnode *vp)
3103 {
3104 	struct filedesc *fdp;
3105 	struct vnode *oldvp;
3106 
3107 	fdp = td->td_proc->p_fd;
3108 	FILEDESC_XLOCK(fdp);
3109 	VNASSERT(vp->v_usecount > 0, vp,
3110 	    ("chdir to a vnode with zero usecount"));
3111 	oldvp = fdp->fd_cdir;
3112 	fdp->fd_cdir = vp;
3113 	FILEDESC_XUNLOCK(fdp);
3114 	vrele(oldvp);
3115 }
3116 
3117 /*
3118  * Scan all active processes and prisons to see if any of them have a current
3119  * or root directory of `olddp'. If so, replace them with the new mount point.
3120  */
3121 void
3122 mountcheckdirs(struct vnode *olddp, struct vnode *newdp)
3123 {
3124 	struct filedesc *fdp;
3125 	struct prison *pr;
3126 	struct proc *p;
3127 	int nrele;
3128 
3129 	if (vrefcnt(olddp) == 1)
3130 		return;
3131 	nrele = 0;
3132 	sx_slock(&allproc_lock);
3133 	FOREACH_PROC_IN_SYSTEM(p) {
3134 		PROC_LOCK(p);
3135 		fdp = fdhold(p);
3136 		PROC_UNLOCK(p);
3137 		if (fdp == NULL)
3138 			continue;
3139 		FILEDESC_XLOCK(fdp);
3140 		if (fdp->fd_cdir == olddp) {
3141 			vrefact(newdp);
3142 			fdp->fd_cdir = newdp;
3143 			nrele++;
3144 		}
3145 		if (fdp->fd_rdir == olddp) {
3146 			vrefact(newdp);
3147 			fdp->fd_rdir = newdp;
3148 			nrele++;
3149 		}
3150 		if (fdp->fd_jdir == olddp) {
3151 			vrefact(newdp);
3152 			fdp->fd_jdir = newdp;
3153 			nrele++;
3154 		}
3155 		FILEDESC_XUNLOCK(fdp);
3156 		fddrop(fdp);
3157 	}
3158 	sx_sunlock(&allproc_lock);
3159 	if (rootvnode == olddp) {
3160 		vrefact(newdp);
3161 		rootvnode = newdp;
3162 		nrele++;
3163 	}
3164 	mtx_lock(&prison0.pr_mtx);
3165 	if (prison0.pr_root == olddp) {
3166 		vrefact(newdp);
3167 		prison0.pr_root = newdp;
3168 		nrele++;
3169 	}
3170 	mtx_unlock(&prison0.pr_mtx);
3171 	sx_slock(&allprison_lock);
3172 	TAILQ_FOREACH(pr, &allprison, pr_list) {
3173 		mtx_lock(&pr->pr_mtx);
3174 		if (pr->pr_root == olddp) {
3175 			vrefact(newdp);
3176 			pr->pr_root = newdp;
3177 			nrele++;
3178 		}
3179 		mtx_unlock(&pr->pr_mtx);
3180 	}
3181 	sx_sunlock(&allprison_lock);
3182 	while (nrele--)
3183 		vrele(olddp);
3184 }
3185 
3186 struct filedesc_to_leader *
3187 filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader)
3188 {
3189 	struct filedesc_to_leader *fdtol;
3190 
3191 	fdtol = malloc(sizeof(struct filedesc_to_leader),
3192 	    M_FILEDESC_TO_LEADER, M_WAITOK);
3193 	fdtol->fdl_refcount = 1;
3194 	fdtol->fdl_holdcount = 0;
3195 	fdtol->fdl_wakeup = 0;
3196 	fdtol->fdl_leader = leader;
3197 	if (old != NULL) {
3198 		FILEDESC_XLOCK(fdp);
3199 		fdtol->fdl_next = old->fdl_next;
3200 		fdtol->fdl_prev = old;
3201 		old->fdl_next = fdtol;
3202 		fdtol->fdl_next->fdl_prev = fdtol;
3203 		FILEDESC_XUNLOCK(fdp);
3204 	} else {
3205 		fdtol->fdl_next = fdtol;
3206 		fdtol->fdl_prev = fdtol;
3207 	}
3208 	return (fdtol);
3209 }
3210 
3211 static int
3212 sysctl_kern_proc_nfds(SYSCTL_HANDLER_ARGS)
3213 {
3214 	struct filedesc *fdp;
3215 	int i, count, slots;
3216 
3217 	if (*(int *)arg1 != 0)
3218 		return (EINVAL);
3219 
3220 	fdp = curproc->p_fd;
3221 	count = 0;
3222 	FILEDESC_SLOCK(fdp);
3223 	slots = NDSLOTS(fdp->fd_lastfile + 1);
3224 	for (i = 0; i < slots; i++)
3225 		count += bitcountl(fdp->fd_map[i]);
3226 	FILEDESC_SUNLOCK(fdp);
3227 
3228 	return (SYSCTL_OUT(req, &count, sizeof(count)));
3229 }
3230 
3231 static SYSCTL_NODE(_kern_proc, KERN_PROC_NFDS, nfds,
3232     CTLFLAG_RD|CTLFLAG_CAPRD|CTLFLAG_MPSAFE, sysctl_kern_proc_nfds,
3233     "Number of open file descriptors");
3234 
3235 /*
3236  * Get file structures globally.
3237  */
3238 static int
3239 sysctl_kern_file(SYSCTL_HANDLER_ARGS)
3240 {
3241 	struct xfile xf;
3242 	struct filedesc *fdp;
3243 	struct file *fp;
3244 	struct proc *p;
3245 	int error, n;
3246 
3247 	error = sysctl_wire_old_buffer(req, 0);
3248 	if (error != 0)
3249 		return (error);
3250 	if (req->oldptr == NULL) {
3251 		n = 0;
3252 		sx_slock(&allproc_lock);
3253 		FOREACH_PROC_IN_SYSTEM(p) {
3254 			PROC_LOCK(p);
3255 			if (p->p_state == PRS_NEW) {
3256 				PROC_UNLOCK(p);
3257 				continue;
3258 			}
3259 			fdp = fdhold(p);
3260 			PROC_UNLOCK(p);
3261 			if (fdp == NULL)
3262 				continue;
3263 			/* overestimates sparse tables. */
3264 			if (fdp->fd_lastfile > 0)
3265 				n += fdp->fd_lastfile;
3266 			fddrop(fdp);
3267 		}
3268 		sx_sunlock(&allproc_lock);
3269 		return (SYSCTL_OUT(req, 0, n * sizeof(xf)));
3270 	}
3271 	error = 0;
3272 	bzero(&xf, sizeof(xf));
3273 	xf.xf_size = sizeof(xf);
3274 	sx_slock(&allproc_lock);
3275 	FOREACH_PROC_IN_SYSTEM(p) {
3276 		PROC_LOCK(p);
3277 		if (p->p_state == PRS_NEW) {
3278 			PROC_UNLOCK(p);
3279 			continue;
3280 		}
3281 		if (p_cansee(req->td, p) != 0) {
3282 			PROC_UNLOCK(p);
3283 			continue;
3284 		}
3285 		xf.xf_pid = p->p_pid;
3286 		xf.xf_uid = p->p_ucred->cr_uid;
3287 		fdp = fdhold(p);
3288 		PROC_UNLOCK(p);
3289 		if (fdp == NULL)
3290 			continue;
3291 		FILEDESC_SLOCK(fdp);
3292 		for (n = 0; fdp->fd_refcnt > 0 && n <= fdp->fd_lastfile; ++n) {
3293 			if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
3294 				continue;
3295 			xf.xf_fd = n;
3296 			xf.xf_file = fp;
3297 			xf.xf_data = fp->f_data;
3298 			xf.xf_vnode = fp->f_vnode;
3299 			xf.xf_type = fp->f_type;
3300 			xf.xf_count = fp->f_count;
3301 			xf.xf_msgcount = 0;
3302 			xf.xf_offset = foffset_get(fp);
3303 			xf.xf_flag = fp->f_flag;
3304 			error = SYSCTL_OUT(req, &xf, sizeof(xf));
3305 			if (error)
3306 				break;
3307 		}
3308 		FILEDESC_SUNLOCK(fdp);
3309 		fddrop(fdp);
3310 		if (error)
3311 			break;
3312 	}
3313 	sx_sunlock(&allproc_lock);
3314 	return (error);
3315 }
3316 
3317 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD|CTLFLAG_MPSAFE,
3318     0, 0, sysctl_kern_file, "S,xfile", "Entire file table");
3319 
3320 #ifdef KINFO_FILE_SIZE
3321 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
3322 #endif
3323 
3324 static int
3325 xlate_fflags(int fflags)
3326 {
3327 	static const struct {
3328 		int	fflag;
3329 		int	kf_fflag;
3330 	} fflags_table[] = {
3331 		{ FAPPEND, KF_FLAG_APPEND },
3332 		{ FASYNC, KF_FLAG_ASYNC },
3333 		{ FFSYNC, KF_FLAG_FSYNC },
3334 		{ FHASLOCK, KF_FLAG_HASLOCK },
3335 		{ FNONBLOCK, KF_FLAG_NONBLOCK },
3336 		{ FREAD, KF_FLAG_READ },
3337 		{ FWRITE, KF_FLAG_WRITE },
3338 		{ O_CREAT, KF_FLAG_CREAT },
3339 		{ O_DIRECT, KF_FLAG_DIRECT },
3340 		{ O_EXCL, KF_FLAG_EXCL },
3341 		{ O_EXEC, KF_FLAG_EXEC },
3342 		{ O_EXLOCK, KF_FLAG_EXLOCK },
3343 		{ O_NOFOLLOW, KF_FLAG_NOFOLLOW },
3344 		{ O_SHLOCK, KF_FLAG_SHLOCK },
3345 		{ O_TRUNC, KF_FLAG_TRUNC }
3346 	};
3347 	unsigned int i;
3348 	int kflags;
3349 
3350 	kflags = 0;
3351 	for (i = 0; i < nitems(fflags_table); i++)
3352 		if (fflags & fflags_table[i].fflag)
3353 			kflags |=  fflags_table[i].kf_fflag;
3354 	return (kflags);
3355 }
3356 
3357 /* Trim unused data from kf_path by truncating the structure size. */
3358 static void
3359 pack_kinfo(struct kinfo_file *kif)
3360 {
3361 
3362 	kif->kf_structsize = offsetof(struct kinfo_file, kf_path) +
3363 	    strlen(kif->kf_path) + 1;
3364 	kif->kf_structsize = roundup(kif->kf_structsize, sizeof(uint64_t));
3365 }
3366 
3367 static void
3368 export_file_to_kinfo(struct file *fp, int fd, cap_rights_t *rightsp,
3369     struct kinfo_file *kif, struct filedesc *fdp, int flags)
3370 {
3371 	int error;
3372 
3373 	bzero(kif, sizeof(*kif));
3374 
3375 	/* Set a default type to allow for empty fill_kinfo() methods. */
3376 	kif->kf_type = KF_TYPE_UNKNOWN;
3377 	kif->kf_flags = xlate_fflags(fp->f_flag);
3378 	if (rightsp != NULL)
3379 		kif->kf_cap_rights = *rightsp;
3380 	else
3381 		cap_rights_init(&kif->kf_cap_rights);
3382 	kif->kf_fd = fd;
3383 	kif->kf_ref_count = fp->f_count;
3384 	kif->kf_offset = foffset_get(fp);
3385 
3386 	/*
3387 	 * This may drop the filedesc lock, so the 'fp' cannot be
3388 	 * accessed after this call.
3389 	 */
3390 	error = fo_fill_kinfo(fp, kif, fdp);
3391 	if (error == 0)
3392 		kif->kf_status |= KF_ATTR_VALID;
3393 	if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
3394 		pack_kinfo(kif);
3395 	else
3396 		kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
3397 }
3398 
3399 static void
3400 export_vnode_to_kinfo(struct vnode *vp, int fd, int fflags,
3401     struct kinfo_file *kif, int flags)
3402 {
3403 	int error;
3404 
3405 	bzero(kif, sizeof(*kif));
3406 
3407 	kif->kf_type = KF_TYPE_VNODE;
3408 	error = vn_fill_kinfo_vnode(vp, kif);
3409 	if (error == 0)
3410 		kif->kf_status |= KF_ATTR_VALID;
3411 	kif->kf_flags = xlate_fflags(fflags);
3412 	cap_rights_init(&kif->kf_cap_rights);
3413 	kif->kf_fd = fd;
3414 	kif->kf_ref_count = -1;
3415 	kif->kf_offset = -1;
3416 	if ((flags & KERN_FILEDESC_PACK_KINFO) != 0)
3417 		pack_kinfo(kif);
3418 	else
3419 		kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t));
3420 	vrele(vp);
3421 }
3422 
3423 struct export_fd_buf {
3424 	struct filedesc		*fdp;
3425 	struct sbuf 		*sb;
3426 	ssize_t			remainder;
3427 	struct kinfo_file	kif;
3428 	int			flags;
3429 };
3430 
3431 static int
3432 export_kinfo_to_sb(struct export_fd_buf *efbuf)
3433 {
3434 	struct kinfo_file *kif;
3435 
3436 	kif = &efbuf->kif;
3437 	if (efbuf->remainder != -1) {
3438 		if (efbuf->remainder < kif->kf_structsize) {
3439 			/* Terminate export. */
3440 			efbuf->remainder = 0;
3441 			return (0);
3442 		}
3443 		efbuf->remainder -= kif->kf_structsize;
3444 	}
3445 	return (sbuf_bcat(efbuf->sb, kif, kif->kf_structsize) == 0 ? 0 : ENOMEM);
3446 }
3447 
3448 static int
3449 export_file_to_sb(struct file *fp, int fd, cap_rights_t *rightsp,
3450     struct export_fd_buf *efbuf)
3451 {
3452 	int error;
3453 
3454 	if (efbuf->remainder == 0)
3455 		return (0);
3456 	export_file_to_kinfo(fp, fd, rightsp, &efbuf->kif, efbuf->fdp,
3457 	    efbuf->flags);
3458 	FILEDESC_SUNLOCK(efbuf->fdp);
3459 	error = export_kinfo_to_sb(efbuf);
3460 	FILEDESC_SLOCK(efbuf->fdp);
3461 	return (error);
3462 }
3463 
3464 static int
3465 export_vnode_to_sb(struct vnode *vp, int fd, int fflags,
3466     struct export_fd_buf *efbuf)
3467 {
3468 	int error;
3469 
3470 	if (efbuf->remainder == 0)
3471 		return (0);
3472 	if (efbuf->fdp != NULL)
3473 		FILEDESC_SUNLOCK(efbuf->fdp);
3474 	export_vnode_to_kinfo(vp, fd, fflags, &efbuf->kif, efbuf->flags);
3475 	error = export_kinfo_to_sb(efbuf);
3476 	if (efbuf->fdp != NULL)
3477 		FILEDESC_SLOCK(efbuf->fdp);
3478 	return (error);
3479 }
3480 
3481 /*
3482  * Store a process file descriptor information to sbuf.
3483  *
3484  * Takes a locked proc as argument, and returns with the proc unlocked.
3485  */
3486 int
3487 kern_proc_filedesc_out(struct proc *p,  struct sbuf *sb, ssize_t maxlen,
3488     int flags)
3489 {
3490 	struct file *fp;
3491 	struct filedesc *fdp;
3492 	struct export_fd_buf *efbuf;
3493 	struct vnode *cttyvp, *textvp, *tracevp;
3494 	int error, i;
3495 	cap_rights_t rights;
3496 
3497 	PROC_LOCK_ASSERT(p, MA_OWNED);
3498 
3499 	/* ktrace vnode */
3500 	tracevp = p->p_tracevp;
3501 	if (tracevp != NULL)
3502 		vrefact(tracevp);
3503 	/* text vnode */
3504 	textvp = p->p_textvp;
3505 	if (textvp != NULL)
3506 		vrefact(textvp);
3507 	/* Controlling tty. */
3508 	cttyvp = NULL;
3509 	if (p->p_pgrp != NULL && p->p_pgrp->pg_session != NULL) {
3510 		cttyvp = p->p_pgrp->pg_session->s_ttyvp;
3511 		if (cttyvp != NULL)
3512 			vrefact(cttyvp);
3513 	}
3514 	fdp = fdhold(p);
3515 	PROC_UNLOCK(p);
3516 	efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
3517 	efbuf->fdp = NULL;
3518 	efbuf->sb = sb;
3519 	efbuf->remainder = maxlen;
3520 	efbuf->flags = flags;
3521 	if (tracevp != NULL)
3522 		export_vnode_to_sb(tracevp, KF_FD_TYPE_TRACE, FREAD | FWRITE,
3523 		    efbuf);
3524 	if (textvp != NULL)
3525 		export_vnode_to_sb(textvp, KF_FD_TYPE_TEXT, FREAD, efbuf);
3526 	if (cttyvp != NULL)
3527 		export_vnode_to_sb(cttyvp, KF_FD_TYPE_CTTY, FREAD | FWRITE,
3528 		    efbuf);
3529 	error = 0;
3530 	if (fdp == NULL)
3531 		goto fail;
3532 	efbuf->fdp = fdp;
3533 	FILEDESC_SLOCK(fdp);
3534 	/* working directory */
3535 	if (fdp->fd_cdir != NULL) {
3536 		vrefact(fdp->fd_cdir);
3537 		export_vnode_to_sb(fdp->fd_cdir, KF_FD_TYPE_CWD, FREAD, efbuf);
3538 	}
3539 	/* root directory */
3540 	if (fdp->fd_rdir != NULL) {
3541 		vrefact(fdp->fd_rdir);
3542 		export_vnode_to_sb(fdp->fd_rdir, KF_FD_TYPE_ROOT, FREAD, efbuf);
3543 	}
3544 	/* jail directory */
3545 	if (fdp->fd_jdir != NULL) {
3546 		vrefact(fdp->fd_jdir);
3547 		export_vnode_to_sb(fdp->fd_jdir, KF_FD_TYPE_JAIL, FREAD, efbuf);
3548 	}
3549 	for (i = 0; fdp->fd_refcnt > 0 && i <= fdp->fd_lastfile; i++) {
3550 		if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
3551 			continue;
3552 #ifdef CAPABILITIES
3553 		rights = *cap_rights(fdp, i);
3554 #else /* !CAPABILITIES */
3555 		cap_rights_init(&rights);
3556 #endif
3557 		/*
3558 		 * Create sysctl entry.  It is OK to drop the filedesc
3559 		 * lock inside of export_file_to_sb() as we will
3560 		 * re-validate and re-evaluate its properties when the
3561 		 * loop continues.
3562 		 */
3563 		error = export_file_to_sb(fp, i, &rights, efbuf);
3564 		if (error != 0 || efbuf->remainder == 0)
3565 			break;
3566 	}
3567 	FILEDESC_SUNLOCK(fdp);
3568 	fddrop(fdp);
3569 fail:
3570 	free(efbuf, M_TEMP);
3571 	return (error);
3572 }
3573 
3574 #define FILEDESC_SBUF_SIZE	(sizeof(struct kinfo_file) * 5)
3575 
3576 /*
3577  * Get per-process file descriptors for use by procstat(1), et al.
3578  */
3579 static int
3580 sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS)
3581 {
3582 	struct sbuf sb;
3583 	struct proc *p;
3584 	ssize_t maxlen;
3585 	int error, error2, *name;
3586 
3587 	name = (int *)arg1;
3588 
3589 	sbuf_new_for_sysctl(&sb, NULL, FILEDESC_SBUF_SIZE, req);
3590 	sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
3591 	error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
3592 	if (error != 0) {
3593 		sbuf_delete(&sb);
3594 		return (error);
3595 	}
3596 	maxlen = req->oldptr != NULL ? req->oldlen : -1;
3597 	error = kern_proc_filedesc_out(p, &sb, maxlen,
3598 	    KERN_FILEDESC_PACK_KINFO);
3599 	error2 = sbuf_finish(&sb);
3600 	sbuf_delete(&sb);
3601 	return (error != 0 ? error : error2);
3602 }
3603 
3604 #ifdef COMPAT_FREEBSD7
3605 #ifdef KINFO_OFILE_SIZE
3606 CTASSERT(sizeof(struct kinfo_ofile) == KINFO_OFILE_SIZE);
3607 #endif
3608 
3609 static void
3610 kinfo_to_okinfo(struct kinfo_file *kif, struct kinfo_ofile *okif)
3611 {
3612 
3613 	okif->kf_structsize = sizeof(*okif);
3614 	okif->kf_type = kif->kf_type;
3615 	okif->kf_fd = kif->kf_fd;
3616 	okif->kf_ref_count = kif->kf_ref_count;
3617 	okif->kf_flags = kif->kf_flags & (KF_FLAG_READ | KF_FLAG_WRITE |
3618 	    KF_FLAG_APPEND | KF_FLAG_ASYNC | KF_FLAG_FSYNC | KF_FLAG_NONBLOCK |
3619 	    KF_FLAG_DIRECT | KF_FLAG_HASLOCK);
3620 	okif->kf_offset = kif->kf_offset;
3621 	if (kif->kf_type == KF_TYPE_VNODE)
3622 		okif->kf_vnode_type = kif->kf_un.kf_file.kf_file_type;
3623 	else
3624 		okif->kf_vnode_type = KF_VTYPE_VNON;
3625 	strlcpy(okif->kf_path, kif->kf_path, sizeof(okif->kf_path));
3626 	if (kif->kf_type == KF_TYPE_SOCKET) {
3627 		okif->kf_sock_domain = kif->kf_un.kf_sock.kf_sock_domain0;
3628 		okif->kf_sock_type = kif->kf_un.kf_sock.kf_sock_type0;
3629 		okif->kf_sock_protocol = kif->kf_un.kf_sock.kf_sock_protocol0;
3630 		okif->kf_sa_local = kif->kf_un.kf_sock.kf_sa_local;
3631 		okif->kf_sa_peer = kif->kf_un.kf_sock.kf_sa_peer;
3632 	} else {
3633 		okif->kf_sa_local.ss_family = AF_UNSPEC;
3634 		okif->kf_sa_peer.ss_family = AF_UNSPEC;
3635 	}
3636 }
3637 
3638 static int
3639 export_vnode_for_osysctl(struct vnode *vp, int type, struct kinfo_file *kif,
3640     struct kinfo_ofile *okif, struct filedesc *fdp, struct sysctl_req *req)
3641 {
3642 	int error;
3643 
3644 	vrefact(vp);
3645 	FILEDESC_SUNLOCK(fdp);
3646 	export_vnode_to_kinfo(vp, type, 0, kif, KERN_FILEDESC_PACK_KINFO);
3647 	kinfo_to_okinfo(kif, okif);
3648 	error = SYSCTL_OUT(req, okif, sizeof(*okif));
3649 	FILEDESC_SLOCK(fdp);
3650 	return (error);
3651 }
3652 
3653 /*
3654  * Get per-process file descriptors for use by procstat(1), et al.
3655  */
3656 static int
3657 sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS)
3658 {
3659 	struct kinfo_ofile *okif;
3660 	struct kinfo_file *kif;
3661 	struct filedesc *fdp;
3662 	int error, i, *name;
3663 	struct file *fp;
3664 	struct proc *p;
3665 
3666 	name = (int *)arg1;
3667 	error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
3668 	if (error != 0)
3669 		return (error);
3670 	fdp = fdhold(p);
3671 	PROC_UNLOCK(p);
3672 	if (fdp == NULL)
3673 		return (ENOENT);
3674 	kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK);
3675 	okif = malloc(sizeof(*okif), M_TEMP, M_WAITOK);
3676 	FILEDESC_SLOCK(fdp);
3677 	if (fdp->fd_cdir != NULL)
3678 		export_vnode_for_osysctl(fdp->fd_cdir, KF_FD_TYPE_CWD, kif,
3679 		    okif, fdp, req);
3680 	if (fdp->fd_rdir != NULL)
3681 		export_vnode_for_osysctl(fdp->fd_rdir, KF_FD_TYPE_ROOT, kif,
3682 		    okif, fdp, req);
3683 	if (fdp->fd_jdir != NULL)
3684 		export_vnode_for_osysctl(fdp->fd_jdir, KF_FD_TYPE_JAIL, kif,
3685 		    okif, fdp, req);
3686 	for (i = 0; fdp->fd_refcnt > 0 && i <= fdp->fd_lastfile; i++) {
3687 		if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
3688 			continue;
3689 		export_file_to_kinfo(fp, i, NULL, kif, fdp,
3690 		    KERN_FILEDESC_PACK_KINFO);
3691 		FILEDESC_SUNLOCK(fdp);
3692 		kinfo_to_okinfo(kif, okif);
3693 		error = SYSCTL_OUT(req, okif, sizeof(*okif));
3694 		FILEDESC_SLOCK(fdp);
3695 		if (error)
3696 			break;
3697 	}
3698 	FILEDESC_SUNLOCK(fdp);
3699 	fddrop(fdp);
3700 	free(kif, M_TEMP);
3701 	free(okif, M_TEMP);
3702 	return (0);
3703 }
3704 
3705 static SYSCTL_NODE(_kern_proc, KERN_PROC_OFILEDESC, ofiledesc,
3706     CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_ofiledesc,
3707     "Process ofiledesc entries");
3708 #endif	/* COMPAT_FREEBSD7 */
3709 
3710 int
3711 vntype_to_kinfo(int vtype)
3712 {
3713 	struct {
3714 		int	vtype;
3715 		int	kf_vtype;
3716 	} vtypes_table[] = {
3717 		{ VBAD, KF_VTYPE_VBAD },
3718 		{ VBLK, KF_VTYPE_VBLK },
3719 		{ VCHR, KF_VTYPE_VCHR },
3720 		{ VDIR, KF_VTYPE_VDIR },
3721 		{ VFIFO, KF_VTYPE_VFIFO },
3722 		{ VLNK, KF_VTYPE_VLNK },
3723 		{ VNON, KF_VTYPE_VNON },
3724 		{ VREG, KF_VTYPE_VREG },
3725 		{ VSOCK, KF_VTYPE_VSOCK }
3726 	};
3727 	unsigned int i;
3728 
3729 	/*
3730 	 * Perform vtype translation.
3731 	 */
3732 	for (i = 0; i < nitems(vtypes_table); i++)
3733 		if (vtypes_table[i].vtype == vtype)
3734 			return (vtypes_table[i].kf_vtype);
3735 
3736 	return (KF_VTYPE_UNKNOWN);
3737 }
3738 
3739 static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc,
3740     CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_filedesc,
3741     "Process filedesc entries");
3742 
3743 /*
3744  * Store a process current working directory information to sbuf.
3745  *
3746  * Takes a locked proc as argument, and returns with the proc unlocked.
3747  */
3748 int
3749 kern_proc_cwd_out(struct proc *p,  struct sbuf *sb, ssize_t maxlen)
3750 {
3751 	struct filedesc *fdp;
3752 	struct export_fd_buf *efbuf;
3753 	int error;
3754 
3755 	PROC_LOCK_ASSERT(p, MA_OWNED);
3756 
3757 	fdp = fdhold(p);
3758 	PROC_UNLOCK(p);
3759 	if (fdp == NULL)
3760 		return (EINVAL);
3761 
3762 	efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
3763 	efbuf->fdp = fdp;
3764 	efbuf->sb = sb;
3765 	efbuf->remainder = maxlen;
3766 
3767 	FILEDESC_SLOCK(fdp);
3768 	if (fdp->fd_cdir == NULL)
3769 		error = EINVAL;
3770 	else {
3771 		vrefact(fdp->fd_cdir);
3772 		error = export_vnode_to_sb(fdp->fd_cdir, KF_FD_TYPE_CWD,
3773 		    FREAD, efbuf);
3774 	}
3775 	FILEDESC_SUNLOCK(fdp);
3776 	fddrop(fdp);
3777 	free(efbuf, M_TEMP);
3778 	return (error);
3779 }
3780 
3781 /*
3782  * Get per-process current working directory.
3783  */
3784 static int
3785 sysctl_kern_proc_cwd(SYSCTL_HANDLER_ARGS)
3786 {
3787 	struct sbuf sb;
3788 	struct proc *p;
3789 	ssize_t maxlen;
3790 	int error, error2, *name;
3791 
3792 	name = (int *)arg1;
3793 
3794 	sbuf_new_for_sysctl(&sb, NULL, sizeof(struct kinfo_file), req);
3795 	sbuf_clear_flags(&sb, SBUF_INCLUDENUL);
3796 	error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p);
3797 	if (error != 0) {
3798 		sbuf_delete(&sb);
3799 		return (error);
3800 	}
3801 	maxlen = req->oldptr != NULL ? req->oldlen : -1;
3802 	error = kern_proc_cwd_out(p, &sb, maxlen);
3803 	error2 = sbuf_finish(&sb);
3804 	sbuf_delete(&sb);
3805 	return (error != 0 ? error : error2);
3806 }
3807 
3808 static SYSCTL_NODE(_kern_proc, KERN_PROC_CWD, cwd, CTLFLAG_RD|CTLFLAG_MPSAFE,
3809     sysctl_kern_proc_cwd, "Process current working directory");
3810 
3811 #ifdef DDB
3812 /*
3813  * For the purposes of debugging, generate a human-readable string for the
3814  * file type.
3815  */
3816 static const char *
3817 file_type_to_name(short type)
3818 {
3819 
3820 	switch (type) {
3821 	case 0:
3822 		return ("zero");
3823 	case DTYPE_VNODE:
3824 		return ("vnode");
3825 	case DTYPE_SOCKET:
3826 		return ("socket");
3827 	case DTYPE_PIPE:
3828 		return ("pipe");
3829 	case DTYPE_FIFO:
3830 		return ("fifo");
3831 	case DTYPE_KQUEUE:
3832 		return ("kqueue");
3833 	case DTYPE_CRYPTO:
3834 		return ("crypto");
3835 	case DTYPE_MQUEUE:
3836 		return ("mqueue");
3837 	case DTYPE_SHM:
3838 		return ("shm");
3839 	case DTYPE_SEM:
3840 		return ("ksem");
3841 	case DTYPE_PTS:
3842 		return ("pts");
3843 	case DTYPE_DEV:
3844 		return ("dev");
3845 	case DTYPE_PROCDESC:
3846 		return ("proc");
3847 	case DTYPE_LINUXEFD:
3848 		return ("levent");
3849 	case DTYPE_LINUXTFD:
3850 		return ("ltimer");
3851 	default:
3852 		return ("unkn");
3853 	}
3854 }
3855 
3856 /*
3857  * For the purposes of debugging, identify a process (if any, perhaps one of
3858  * many) that references the passed file in its file descriptor array. Return
3859  * NULL if none.
3860  */
3861 static struct proc *
3862 file_to_first_proc(struct file *fp)
3863 {
3864 	struct filedesc *fdp;
3865 	struct proc *p;
3866 	int n;
3867 
3868 	FOREACH_PROC_IN_SYSTEM(p) {
3869 		if (p->p_state == PRS_NEW)
3870 			continue;
3871 		fdp = p->p_fd;
3872 		if (fdp == NULL)
3873 			continue;
3874 		for (n = 0; n <= fdp->fd_lastfile; n++) {
3875 			if (fp == fdp->fd_ofiles[n].fde_file)
3876 				return (p);
3877 		}
3878 	}
3879 	return (NULL);
3880 }
3881 
3882 static void
3883 db_print_file(struct file *fp, int header)
3884 {
3885 #define XPTRWIDTH ((int)howmany(sizeof(void *) * NBBY, 4))
3886 	struct proc *p;
3887 
3888 	if (header)
3889 		db_printf("%*s %6s %*s %8s %4s %5s %6s %*s %5s %s\n",
3890 		    XPTRWIDTH, "File", "Type", XPTRWIDTH, "Data", "Flag",
3891 		    "GCFl", "Count", "MCount", XPTRWIDTH, "Vnode", "FPID",
3892 		    "FCmd");
3893 	p = file_to_first_proc(fp);
3894 	db_printf("%*p %6s %*p %08x %04x %5d %6d %*p %5d %s\n", XPTRWIDTH,
3895 	    fp, file_type_to_name(fp->f_type), XPTRWIDTH, fp->f_data,
3896 	    fp->f_flag, 0, fp->f_count, 0, XPTRWIDTH, fp->f_vnode,
3897 	    p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-");
3898 
3899 #undef XPTRWIDTH
3900 }
3901 
3902 DB_SHOW_COMMAND(file, db_show_file)
3903 {
3904 	struct file *fp;
3905 
3906 	if (!have_addr) {
3907 		db_printf("usage: show file <addr>\n");
3908 		return;
3909 	}
3910 	fp = (struct file *)addr;
3911 	db_print_file(fp, 1);
3912 }
3913 
3914 DB_SHOW_COMMAND(files, db_show_files)
3915 {
3916 	struct filedesc *fdp;
3917 	struct file *fp;
3918 	struct proc *p;
3919 	int header;
3920 	int n;
3921 
3922 	header = 1;
3923 	FOREACH_PROC_IN_SYSTEM(p) {
3924 		if (p->p_state == PRS_NEW)
3925 			continue;
3926 		if ((fdp = p->p_fd) == NULL)
3927 			continue;
3928 		for (n = 0; n <= fdp->fd_lastfile; ++n) {
3929 			if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
3930 				continue;
3931 			db_print_file(fp, header);
3932 			header = 0;
3933 		}
3934 	}
3935 }
3936 #endif
3937 
3938 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
3939     &maxfilesperproc, 0, "Maximum files allowed open per process");
3940 
3941 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
3942     &maxfiles, 0, "Maximum number of files");
3943 
3944 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
3945     __DEVOLATILE(int *, &openfiles), 0, "System-wide number of open files");
3946 
3947 /* ARGSUSED*/
3948 static void
3949 filelistinit(void *dummy)
3950 {
3951 
3952 	file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL,
3953 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
3954 	filedesc0_zone = uma_zcreate("filedesc0", sizeof(struct filedesc0),
3955 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
3956 	mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF);
3957 }
3958 SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL);
3959 
3960 /*-------------------------------------------------------------------*/
3961 
3962 static int
3963 badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred,
3964     int flags, struct thread *td)
3965 {
3966 
3967 	return (EBADF);
3968 }
3969 
3970 static int
3971 badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
3972     struct thread *td)
3973 {
3974 
3975 	return (EINVAL);
3976 }
3977 
3978 static int
3979 badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred,
3980     struct thread *td)
3981 {
3982 
3983 	return (EBADF);
3984 }
3985 
3986 static int
3987 badfo_poll(struct file *fp, int events, struct ucred *active_cred,
3988     struct thread *td)
3989 {
3990 
3991 	return (0);
3992 }
3993 
3994 static int
3995 badfo_kqfilter(struct file *fp, struct knote *kn)
3996 {
3997 
3998 	return (EBADF);
3999 }
4000 
4001 static int
4002 badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred,
4003     struct thread *td)
4004 {
4005 
4006 	return (EBADF);
4007 }
4008 
4009 static int
4010 badfo_close(struct file *fp, struct thread *td)
4011 {
4012 
4013 	return (0);
4014 }
4015 
4016 static int
4017 badfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
4018     struct thread *td)
4019 {
4020 
4021 	return (EBADF);
4022 }
4023 
4024 static int
4025 badfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
4026     struct thread *td)
4027 {
4028 
4029 	return (EBADF);
4030 }
4031 
4032 static int
4033 badfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
4034     struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
4035     struct thread *td)
4036 {
4037 
4038 	return (EBADF);
4039 }
4040 
4041 static int
4042 badfo_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
4043 {
4044 
4045 	return (0);
4046 }
4047 
4048 struct fileops badfileops = {
4049 	.fo_read = badfo_readwrite,
4050 	.fo_write = badfo_readwrite,
4051 	.fo_truncate = badfo_truncate,
4052 	.fo_ioctl = badfo_ioctl,
4053 	.fo_poll = badfo_poll,
4054 	.fo_kqfilter = badfo_kqfilter,
4055 	.fo_stat = badfo_stat,
4056 	.fo_close = badfo_close,
4057 	.fo_chmod = badfo_chmod,
4058 	.fo_chown = badfo_chown,
4059 	.fo_sendfile = badfo_sendfile,
4060 	.fo_fill_kinfo = badfo_fill_kinfo,
4061 };
4062 
4063 int
4064 invfo_rdwr(struct file *fp, struct uio *uio, struct ucred *active_cred,
4065     int flags, struct thread *td)
4066 {
4067 
4068 	return (EOPNOTSUPP);
4069 }
4070 
4071 int
4072 invfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
4073     struct thread *td)
4074 {
4075 
4076 	return (EINVAL);
4077 }
4078 
4079 int
4080 invfo_ioctl(struct file *fp, u_long com, void *data,
4081     struct ucred *active_cred, struct thread *td)
4082 {
4083 
4084 	return (ENOTTY);
4085 }
4086 
4087 int
4088 invfo_poll(struct file *fp, int events, struct ucred *active_cred,
4089     struct thread *td)
4090 {
4091 
4092 	return (poll_no_poll(events));
4093 }
4094 
4095 int
4096 invfo_kqfilter(struct file *fp, struct knote *kn)
4097 {
4098 
4099 	return (EINVAL);
4100 }
4101 
4102 int
4103 invfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
4104     struct thread *td)
4105 {
4106 
4107 	return (EINVAL);
4108 }
4109 
4110 int
4111 invfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
4112     struct thread *td)
4113 {
4114 
4115 	return (EINVAL);
4116 }
4117 
4118 int
4119 invfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
4120     struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
4121     struct thread *td)
4122 {
4123 
4124 	return (EINVAL);
4125 }
4126 
4127 /*-------------------------------------------------------------------*/
4128 
4129 /*
4130  * File Descriptor pseudo-device driver (/dev/fd/).
4131  *
4132  * Opening minor device N dup()s the file (if any) connected to file
4133  * descriptor N belonging to the calling process.  Note that this driver
4134  * consists of only the ``open()'' routine, because all subsequent
4135  * references to this file will be direct to the other driver.
4136  *
4137  * XXX: we could give this one a cloning event handler if necessary.
4138  */
4139 
4140 /* ARGSUSED */
4141 static int
4142 fdopen(struct cdev *dev, int mode, int type, struct thread *td)
4143 {
4144 
4145 	/*
4146 	 * XXX Kludge: set curthread->td_dupfd to contain the value of the
4147 	 * the file descriptor being sought for duplication. The error
4148 	 * return ensures that the vnode for this device will be released
4149 	 * by vn_open. Open will detect this special error and take the
4150 	 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
4151 	 * will simply report the error.
4152 	 */
4153 	td->td_dupfd = dev2unit(dev);
4154 	return (ENODEV);
4155 }
4156 
4157 static struct cdevsw fildesc_cdevsw = {
4158 	.d_version =	D_VERSION,
4159 	.d_open =	fdopen,
4160 	.d_name =	"FD",
4161 };
4162 
4163 static void
4164 fildesc_drvinit(void *unused)
4165 {
4166 	struct cdev *dev;
4167 
4168 	dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 0, NULL,
4169 	    UID_ROOT, GID_WHEEL, 0666, "fd/0");
4170 	make_dev_alias(dev, "stdin");
4171 	dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 1, NULL,
4172 	    UID_ROOT, GID_WHEEL, 0666, "fd/1");
4173 	make_dev_alias(dev, "stdout");
4174 	dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 2, NULL,
4175 	    UID_ROOT, GID_WHEEL, 0666, "fd/2");
4176 	make_dev_alias(dev, "stderr");
4177 }
4178 
4179 SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL);
4180