xref: /freebsd/sys/kern/kern_descrip.c (revision 6d732c66bca5da4d261577aad2c8ea84519b0bea)
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  * 4. 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/capability.h>
49 #include <sys/conf.h>
50 #include <sys/domain.h>
51 #include <sys/fcntl.h>
52 #include <sys/file.h>
53 #include <sys/filedesc.h>
54 #include <sys/filio.h>
55 #include <sys/jail.h>
56 #include <sys/kernel.h>
57 #include <sys/ksem.h>
58 #include <sys/limits.h>
59 #include <sys/lock.h>
60 #include <sys/malloc.h>
61 #include <sys/mman.h>
62 #include <sys/mount.h>
63 #include <sys/mqueue.h>
64 #include <sys/mutex.h>
65 #include <sys/namei.h>
66 #include <sys/selinfo.h>
67 #include <sys/pipe.h>
68 #include <sys/priv.h>
69 #include <sys/proc.h>
70 #include <sys/procdesc.h>
71 #include <sys/protosw.h>
72 #include <sys/racct.h>
73 #include <sys/resourcevar.h>
74 #include <sys/sbuf.h>
75 #include <sys/signalvar.h>
76 #include <sys/socketvar.h>
77 #include <sys/stat.h>
78 #include <sys/sx.h>
79 #include <sys/syscallsubr.h>
80 #include <sys/sysctl.h>
81 #include <sys/sysproto.h>
82 #include <sys/tty.h>
83 #include <sys/unistd.h>
84 #include <sys/un.h>
85 #include <sys/unpcb.h>
86 #include <sys/user.h>
87 #include <sys/vnode.h>
88 #ifdef KTRACE
89 #include <sys/ktrace.h>
90 #endif
91 
92 #include <net/vnet.h>
93 
94 #include <netinet/in.h>
95 #include <netinet/in_pcb.h>
96 
97 #include <security/audit/audit.h>
98 
99 #include <vm/uma.h>
100 #include <vm/vm.h>
101 
102 #include <ddb/ddb.h>
103 
104 static MALLOC_DEFINE(M_FILEDESC, "filedesc", "Open file descriptor table");
105 static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "filedesc_to_leader",
106     "file desc to leader structures");
107 static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures");
108 MALLOC_DEFINE(M_FILECAPS, "filecaps", "descriptor capabilities");
109 
110 MALLOC_DECLARE(M_FADVISE);
111 
112 static uma_zone_t file_zone;
113 
114 void	(*ksem_info)(struct ksem *ks, char *path, size_t size, uint32_t *value);
115 
116 static int	closefp(struct filedesc *fdp, int fd, struct file *fp,
117 		    struct thread *td, int holdleaders);
118 static int	fd_first_free(struct filedesc *fdp, int low, int size);
119 static int	fd_last_used(struct filedesc *fdp, int size);
120 static void	fdgrowtable(struct filedesc *fdp, int nfd);
121 static void	fdgrowtable_exp(struct filedesc *fdp, int nfd);
122 static void	fdunused(struct filedesc *fdp, int fd);
123 static void	fdused(struct filedesc *fdp, int fd);
124 static int	fill_pipe_info(struct pipe *pi, struct kinfo_file *kif);
125 static int	fill_procdesc_info(struct procdesc *pdp,
126 		    struct kinfo_file *kif);
127 static int	fill_pts_info(struct tty *tp, struct kinfo_file *kif);
128 static int	fill_sem_info(struct file *fp, struct kinfo_file *kif);
129 static int	fill_shm_info(struct file *fp, struct kinfo_file *kif);
130 static int	fill_socket_info(struct socket *so, struct kinfo_file *kif);
131 static int	fill_vnode_info(struct vnode *vp, struct kinfo_file *kif);
132 static int	getmaxfd(struct proc *p);
133 
134 /*
135  * Each process has:
136  *
137  * - An array of open file descriptors (fd_ofiles)
138  * - An array of file flags (fd_ofileflags)
139  * - A bitmap recording which descriptors are in use (fd_map)
140  *
141  * A process starts out with NDFILE descriptors.  The value of NDFILE has
142  * been selected based the historical limit of 20 open files, and an
143  * assumption that the majority of processes, especially short-lived
144  * processes like shells, will never need more.
145  *
146  * If this initial allocation is exhausted, a larger descriptor table and
147  * map are allocated dynamically, and the pointers in the process's struct
148  * filedesc are updated to point to those.  This is repeated every time
149  * the process runs out of file descriptors (provided it hasn't hit its
150  * resource limit).
151  *
152  * Since threads may hold references to individual descriptor table
153  * entries, the tables are never freed.  Instead, they are placed on a
154  * linked list and freed only when the struct filedesc is released.
155  */
156 #define NDFILE		20
157 #define NDSLOTSIZE	sizeof(NDSLOTTYPE)
158 #define	NDENTRIES	(NDSLOTSIZE * __CHAR_BIT)
159 #define NDSLOT(x)	((x) / NDENTRIES)
160 #define NDBIT(x)	((NDSLOTTYPE)1 << ((x) % NDENTRIES))
161 #define	NDSLOTS(x)	(((x) + NDENTRIES - 1) / NDENTRIES)
162 
163 /*
164  * SLIST entry used to keep track of ofiles which must be reclaimed when
165  * the process exits.
166  */
167 struct freetable {
168 	struct filedescent *ft_table;
169 	SLIST_ENTRY(freetable) ft_next;
170 };
171 
172 /*
173  * Initial allocation: a filedesc structure + the head of SLIST used to
174  * keep track of old ofiles + enough space for NDFILE descriptors.
175  */
176 struct filedesc0 {
177 	struct filedesc fd_fd;
178 	SLIST_HEAD(, freetable) fd_free;
179 	struct	filedescent fd_dfiles[NDFILE];
180 	NDSLOTTYPE fd_dmap[NDSLOTS(NDFILE)];
181 };
182 
183 /*
184  * Descriptor management.
185  */
186 volatile int openfiles;			/* actual number of open files */
187 struct mtx sigio_lock;		/* mtx to protect pointers to sigio */
188 void (*mq_fdclose)(struct thread *td, int fd, struct file *fp);
189 
190 /* A mutex to protect the association between a proc and filedesc. */
191 static struct mtx fdesc_mtx;
192 
193 /*
194  * If low >= size, just return low. Otherwise find the first zero bit in the
195  * given bitmap, starting at low and not exceeding size - 1. Return size if
196  * not found.
197  */
198 static int
199 fd_first_free(struct filedesc *fdp, int low, int size)
200 {
201 	NDSLOTTYPE *map = fdp->fd_map;
202 	NDSLOTTYPE mask;
203 	int off, maxoff;
204 
205 	if (low >= size)
206 		return (low);
207 
208 	off = NDSLOT(low);
209 	if (low % NDENTRIES) {
210 		mask = ~(~(NDSLOTTYPE)0 >> (NDENTRIES - (low % NDENTRIES)));
211 		if ((mask &= ~map[off]) != 0UL)
212 			return (off * NDENTRIES + ffsl(mask) - 1);
213 		++off;
214 	}
215 	for (maxoff = NDSLOTS(size); off < maxoff; ++off)
216 		if (map[off] != ~0UL)
217 			return (off * NDENTRIES + ffsl(~map[off]) - 1);
218 	return (size);
219 }
220 
221 /*
222  * Find the highest non-zero bit in the given bitmap, starting at 0 and
223  * not exceeding size - 1. Return -1 if not found.
224  */
225 static int
226 fd_last_used(struct filedesc *fdp, int size)
227 {
228 	NDSLOTTYPE *map = fdp->fd_map;
229 	NDSLOTTYPE mask;
230 	int off, minoff;
231 
232 	off = NDSLOT(size);
233 	if (size % NDENTRIES) {
234 		mask = ~(~(NDSLOTTYPE)0 << (size % NDENTRIES));
235 		if ((mask &= map[off]) != 0)
236 			return (off * NDENTRIES + flsl(mask) - 1);
237 		--off;
238 	}
239 	for (minoff = NDSLOT(0); off >= minoff; --off)
240 		if (map[off] != 0)
241 			return (off * NDENTRIES + flsl(map[off]) - 1);
242 	return (-1);
243 }
244 
245 static int
246 fdisused(struct filedesc *fdp, int fd)
247 {
248 
249 	FILEDESC_LOCK_ASSERT(fdp);
250 
251 	KASSERT(fd >= 0 && fd < fdp->fd_nfiles,
252 	    ("file descriptor %d out of range (0, %d)", fd, fdp->fd_nfiles));
253 
254 	return ((fdp->fd_map[NDSLOT(fd)] & NDBIT(fd)) != 0);
255 }
256 
257 /*
258  * Mark a file descriptor as used.
259  */
260 static void
261 fdused(struct filedesc *fdp, int fd)
262 {
263 
264 	FILEDESC_XLOCK_ASSERT(fdp);
265 
266 	KASSERT(!fdisused(fdp, fd), ("fd=%d is already used", fd));
267 
268 	fdp->fd_map[NDSLOT(fd)] |= NDBIT(fd);
269 	if (fd > fdp->fd_lastfile)
270 		fdp->fd_lastfile = fd;
271 	if (fd == fdp->fd_freefile)
272 		fdp->fd_freefile = fd_first_free(fdp, fd, fdp->fd_nfiles);
273 }
274 
275 /*
276  * Mark a file descriptor as unused.
277  */
278 static void
279 fdunused(struct filedesc *fdp, int fd)
280 {
281 
282 	FILEDESC_XLOCK_ASSERT(fdp);
283 
284 	KASSERT(fdisused(fdp, fd), ("fd=%d is already unused", fd));
285 	KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
286 	    ("fd=%d is still in use", fd));
287 
288 	fdp->fd_map[NDSLOT(fd)] &= ~NDBIT(fd);
289 	if (fd < fdp->fd_freefile)
290 		fdp->fd_freefile = fd;
291 	if (fd == fdp->fd_lastfile)
292 		fdp->fd_lastfile = fd_last_used(fdp, fd);
293 }
294 
295 /*
296  * Free a file descriptor.
297  */
298 static inline void
299 fdfree(struct filedesc *fdp, int fd)
300 {
301 	struct filedescent *fde;
302 
303 	fde = &fdp->fd_ofiles[fd];
304 	filecaps_free(&fde->fde_caps);
305 	bzero(fde, sizeof(*fde));
306 	fdunused(fdp, fd);
307 }
308 
309 /*
310  * System calls on descriptors.
311  */
312 #ifndef _SYS_SYSPROTO_H_
313 struct getdtablesize_args {
314 	int	dummy;
315 };
316 #endif
317 /* ARGSUSED */
318 int
319 sys_getdtablesize(struct thread *td, struct getdtablesize_args *uap)
320 {
321 	struct proc *p = td->td_proc;
322 	uint64_t lim;
323 
324 	PROC_LOCK(p);
325 	td->td_retval[0] =
326 	    min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc);
327 	lim = racct_get_limit(td->td_proc, RACCT_NOFILE);
328 	PROC_UNLOCK(p);
329 	if (lim < td->td_retval[0])
330 		td->td_retval[0] = lim;
331 	return (0);
332 }
333 
334 /*
335  * Duplicate a file descriptor to a particular value.
336  *
337  * Note: keep in mind that a potential race condition exists when closing
338  * descriptors from a shared descriptor table (via rfork).
339  */
340 #ifndef _SYS_SYSPROTO_H_
341 struct dup2_args {
342 	u_int	from;
343 	u_int	to;
344 };
345 #endif
346 /* ARGSUSED */
347 int
348 sys_dup2(struct thread *td, struct dup2_args *uap)
349 {
350 
351 	return (do_dup(td, DUP_FIXED, (int)uap->from, (int)uap->to,
352 		    td->td_retval));
353 }
354 
355 /*
356  * Duplicate a file descriptor.
357  */
358 #ifndef _SYS_SYSPROTO_H_
359 struct dup_args {
360 	u_int	fd;
361 };
362 #endif
363 /* ARGSUSED */
364 int
365 sys_dup(struct thread *td, struct dup_args *uap)
366 {
367 
368 	return (do_dup(td, 0, (int)uap->fd, 0, td->td_retval));
369 }
370 
371 /*
372  * The file control system call.
373  */
374 #ifndef _SYS_SYSPROTO_H_
375 struct fcntl_args {
376 	int	fd;
377 	int	cmd;
378 	long	arg;
379 };
380 #endif
381 /* ARGSUSED */
382 int
383 sys_fcntl(struct thread *td, struct fcntl_args *uap)
384 {
385 	struct flock fl;
386 	struct __oflock ofl;
387 	intptr_t arg;
388 	int error;
389 	int cmd;
390 
391 	error = 0;
392 	cmd = uap->cmd;
393 	switch (uap->cmd) {
394 	case F_OGETLK:
395 	case F_OSETLK:
396 	case F_OSETLKW:
397 		/*
398 		 * Convert old flock structure to new.
399 		 */
400 		error = copyin((void *)(intptr_t)uap->arg, &ofl, sizeof(ofl));
401 		fl.l_start = ofl.l_start;
402 		fl.l_len = ofl.l_len;
403 		fl.l_pid = ofl.l_pid;
404 		fl.l_type = ofl.l_type;
405 		fl.l_whence = ofl.l_whence;
406 		fl.l_sysid = 0;
407 
408 		switch (uap->cmd) {
409 		case F_OGETLK:
410 		    cmd = F_GETLK;
411 		    break;
412 		case F_OSETLK:
413 		    cmd = F_SETLK;
414 		    break;
415 		case F_OSETLKW:
416 		    cmd = F_SETLKW;
417 		    break;
418 		}
419 		arg = (intptr_t)&fl;
420 		break;
421         case F_GETLK:
422         case F_SETLK:
423         case F_SETLKW:
424 	case F_SETLK_REMOTE:
425                 error = copyin((void *)(intptr_t)uap->arg, &fl, sizeof(fl));
426                 arg = (intptr_t)&fl;
427                 break;
428 	default:
429 		arg = uap->arg;
430 		break;
431 	}
432 	if (error)
433 		return (error);
434 	error = kern_fcntl(td, uap->fd, cmd, arg);
435 	if (error)
436 		return (error);
437 	if (uap->cmd == F_OGETLK) {
438 		ofl.l_start = fl.l_start;
439 		ofl.l_len = fl.l_len;
440 		ofl.l_pid = fl.l_pid;
441 		ofl.l_type = fl.l_type;
442 		ofl.l_whence = fl.l_whence;
443 		error = copyout(&ofl, (void *)(intptr_t)uap->arg, sizeof(ofl));
444 	} else if (uap->cmd == F_GETLK) {
445 		error = copyout(&fl, (void *)(intptr_t)uap->arg, sizeof(fl));
446 	}
447 	return (error);
448 }
449 
450 int
451 kern_fcntl(struct thread *td, int fd, int cmd, intptr_t arg)
452 {
453 	struct filedesc *fdp;
454 	struct flock *flp;
455 	struct file *fp, *fp2;
456 	struct filedescent *fde;
457 	struct proc *p;
458 	struct vnode *vp;
459 	cap_rights_t rights;
460 	int error, flg, tmp;
461 	u_int old, new;
462 	uint64_t bsize;
463 	off_t foffset;
464 
465 	error = 0;
466 	flg = F_POSIX;
467 	p = td->td_proc;
468 	fdp = p->p_fd;
469 
470 	switch (cmd) {
471 	case F_DUPFD:
472 		tmp = arg;
473 		error = do_dup(td, DUP_FCNTL, fd, tmp, td->td_retval);
474 		break;
475 
476 	case F_DUPFD_CLOEXEC:
477 		tmp = arg;
478 		error = do_dup(td, DUP_FCNTL | DUP_CLOEXEC, fd, tmp,
479 		    td->td_retval);
480 		break;
481 
482 	case F_DUP2FD:
483 		tmp = arg;
484 		error = do_dup(td, DUP_FIXED, fd, tmp, td->td_retval);
485 		break;
486 
487 	case F_DUP2FD_CLOEXEC:
488 		tmp = arg;
489 		error = do_dup(td, DUP_FIXED | DUP_CLOEXEC, fd, tmp,
490 		    td->td_retval);
491 		break;
492 
493 	case F_GETFD:
494 		FILEDESC_SLOCK(fdp);
495 		if ((fp = fget_locked(fdp, fd)) == NULL) {
496 			FILEDESC_SUNLOCK(fdp);
497 			error = EBADF;
498 			break;
499 		}
500 		fde = &fdp->fd_ofiles[fd];
501 		td->td_retval[0] =
502 		    (fde->fde_flags & UF_EXCLOSE) ? FD_CLOEXEC : 0;
503 		FILEDESC_SUNLOCK(fdp);
504 		break;
505 
506 	case F_SETFD:
507 		FILEDESC_XLOCK(fdp);
508 		if ((fp = fget_locked(fdp, fd)) == NULL) {
509 			FILEDESC_XUNLOCK(fdp);
510 			error = EBADF;
511 			break;
512 		}
513 		fde = &fdp->fd_ofiles[fd];
514 		fde->fde_flags = (fde->fde_flags & ~UF_EXCLOSE) |
515 		    (arg & FD_CLOEXEC ? UF_EXCLOSE : 0);
516 		FILEDESC_XUNLOCK(fdp);
517 		break;
518 
519 	case F_GETFL:
520 		error = fget_unlocked(fdp, fd,
521 		    cap_rights_init(&rights, CAP_FCNTL), F_GETFL, &fp, NULL);
522 		if (error != 0)
523 			break;
524 		td->td_retval[0] = OFLAGS(fp->f_flag);
525 		fdrop(fp, td);
526 		break;
527 
528 	case F_SETFL:
529 		error = fget_unlocked(fdp, fd,
530 		    cap_rights_init(&rights, CAP_FCNTL), F_SETFL, &fp, NULL);
531 		if (error != 0)
532 			break;
533 		do {
534 			tmp = flg = fp->f_flag;
535 			tmp &= ~FCNTLFLAGS;
536 			tmp |= FFLAGS(arg & ~O_ACCMODE) & FCNTLFLAGS;
537 		} while(atomic_cmpset_int(&fp->f_flag, flg, tmp) == 0);
538 		tmp = fp->f_flag & FNONBLOCK;
539 		error = fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
540 		if (error != 0) {
541 			fdrop(fp, td);
542 			break;
543 		}
544 		tmp = fp->f_flag & FASYNC;
545 		error = fo_ioctl(fp, FIOASYNC, &tmp, td->td_ucred, td);
546 		if (error == 0) {
547 			fdrop(fp, td);
548 			break;
549 		}
550 		atomic_clear_int(&fp->f_flag, FNONBLOCK);
551 		tmp = 0;
552 		(void)fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
553 		fdrop(fp, td);
554 		break;
555 
556 	case F_GETOWN:
557 		error = fget_unlocked(fdp, fd,
558 		    cap_rights_init(&rights, CAP_FCNTL), F_GETOWN, &fp, NULL);
559 		if (error != 0)
560 			break;
561 		error = fo_ioctl(fp, FIOGETOWN, &tmp, td->td_ucred, td);
562 		if (error == 0)
563 			td->td_retval[0] = tmp;
564 		fdrop(fp, td);
565 		break;
566 
567 	case F_SETOWN:
568 		error = fget_unlocked(fdp, fd,
569 		    cap_rights_init(&rights, CAP_FCNTL), F_SETOWN, &fp, NULL);
570 		if (error != 0)
571 			break;
572 		tmp = arg;
573 		error = fo_ioctl(fp, FIOSETOWN, &tmp, td->td_ucred, td);
574 		fdrop(fp, td);
575 		break;
576 
577 	case F_SETLK_REMOTE:
578 		error = priv_check(td, PRIV_NFS_LOCKD);
579 		if (error)
580 			return (error);
581 		flg = F_REMOTE;
582 		goto do_setlk;
583 
584 	case F_SETLKW:
585 		flg |= F_WAIT;
586 		/* FALLTHROUGH F_SETLK */
587 
588 	case F_SETLK:
589 	do_setlk:
590 		cap_rights_init(&rights, CAP_FLOCK);
591 		error = fget_unlocked(fdp, fd, &rights, 0, &fp, NULL);
592 		if (error != 0)
593 			break;
594 		if (fp->f_type != DTYPE_VNODE) {
595 			error = EBADF;
596 			fdrop(fp, td);
597 			break;
598 		}
599 
600 		flp = (struct flock *)arg;
601 		if (flp->l_whence == SEEK_CUR) {
602 			foffset = foffset_get(fp);
603 			if (foffset < 0 ||
604 			    (flp->l_start > 0 &&
605 			     foffset > OFF_MAX - flp->l_start)) {
606 				FILEDESC_SUNLOCK(fdp);
607 				error = EOVERFLOW;
608 				fdrop(fp, td);
609 				break;
610 			}
611 			flp->l_start += foffset;
612 		}
613 
614 		vp = fp->f_vnode;
615 		switch (flp->l_type) {
616 		case F_RDLCK:
617 			if ((fp->f_flag & FREAD) == 0) {
618 				error = EBADF;
619 				break;
620 			}
621 			PROC_LOCK(p->p_leader);
622 			p->p_leader->p_flag |= P_ADVLOCK;
623 			PROC_UNLOCK(p->p_leader);
624 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
625 			    flp, flg);
626 			break;
627 		case F_WRLCK:
628 			if ((fp->f_flag & FWRITE) == 0) {
629 				error = EBADF;
630 				break;
631 			}
632 			PROC_LOCK(p->p_leader);
633 			p->p_leader->p_flag |= P_ADVLOCK;
634 			PROC_UNLOCK(p->p_leader);
635 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
636 			    flp, flg);
637 			break;
638 		case F_UNLCK:
639 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
640 			    flp, flg);
641 			break;
642 		case F_UNLCKSYS:
643 			/*
644 			 * Temporary api for testing remote lock
645 			 * infrastructure.
646 			 */
647 			if (flg != F_REMOTE) {
648 				error = EINVAL;
649 				break;
650 			}
651 			error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
652 			    F_UNLCKSYS, flp, flg);
653 			break;
654 		default:
655 			error = EINVAL;
656 			break;
657 		}
658 		if (error != 0 || flp->l_type == F_UNLCK ||
659 		    flp->l_type == F_UNLCKSYS) {
660 			fdrop(fp, td);
661 			break;
662 		}
663 
664 		/*
665 		 * Check for a race with close.
666 		 *
667 		 * The vnode is now advisory locked (or unlocked, but this case
668 		 * is not really important) as the caller requested.
669 		 * We had to drop the filedesc lock, so we need to recheck if
670 		 * the descriptor is still valid, because if it was closed
671 		 * in the meantime we need to remove advisory lock from the
672 		 * vnode - close on any descriptor leading to an advisory
673 		 * locked vnode, removes that lock.
674 		 * We will return 0 on purpose in that case, as the result of
675 		 * successful advisory lock might have been externally visible
676 		 * already. This is fine - effectively we pretend to the caller
677 		 * that the closing thread was a bit slower and that the
678 		 * advisory lock succeeded before the close.
679 		 */
680 		error = fget_unlocked(fdp, fd, &rights, 0, &fp2, NULL);
681 		if (error != 0) {
682 			fdrop(fp, td);
683 			break;
684 		}
685 		if (fp != fp2) {
686 			flp->l_whence = SEEK_SET;
687 			flp->l_start = 0;
688 			flp->l_len = 0;
689 			flp->l_type = F_UNLCK;
690 			(void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
691 			    F_UNLCK, flp, F_POSIX);
692 		}
693 		fdrop(fp, td);
694 		fdrop(fp2, td);
695 		break;
696 
697 	case F_GETLK:
698 		error = fget_unlocked(fdp, fd,
699 		    cap_rights_init(&rights, CAP_FLOCK), 0, &fp, NULL);
700 		if (error != 0)
701 			break;
702 		if (fp->f_type != DTYPE_VNODE) {
703 			error = EBADF;
704 			fdrop(fp, td);
705 			break;
706 		}
707 		flp = (struct flock *)arg;
708 		if (flp->l_type != F_RDLCK && flp->l_type != F_WRLCK &&
709 		    flp->l_type != F_UNLCK) {
710 			error = EINVAL;
711 			fdrop(fp, td);
712 			break;
713 		}
714 		if (flp->l_whence == SEEK_CUR) {
715 			foffset = foffset_get(fp);
716 			if ((flp->l_start > 0 &&
717 			    foffset > OFF_MAX - flp->l_start) ||
718 			    (flp->l_start < 0 &&
719 			     foffset < OFF_MIN - flp->l_start)) {
720 				FILEDESC_SUNLOCK(fdp);
721 				error = EOVERFLOW;
722 				fdrop(fp, td);
723 				break;
724 			}
725 			flp->l_start += foffset;
726 		}
727 		vp = fp->f_vnode;
728 		error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, flp,
729 		    F_POSIX);
730 		fdrop(fp, td);
731 		break;
732 
733 	case F_RDAHEAD:
734 		arg = arg ? 128 * 1024: 0;
735 		/* FALLTHROUGH */
736 	case F_READAHEAD:
737 		error = fget_unlocked(fdp, fd, NULL, 0, &fp, NULL);
738 		if (error != 0)
739 			break;
740 		if (fp->f_type != DTYPE_VNODE) {
741 			fdrop(fp, td);
742 			error = EBADF;
743 			break;
744 		}
745 		if (arg >= 0) {
746 			vp = fp->f_vnode;
747 			error = vn_lock(vp, LK_SHARED);
748 			if (error != 0) {
749 				fdrop(fp, td);
750 				break;
751 			}
752 			bsize = fp->f_vnode->v_mount->mnt_stat.f_iosize;
753 			VOP_UNLOCK(vp, 0);
754 			fp->f_seqcount = (arg + bsize - 1) / bsize;
755 			do {
756 				new = old = fp->f_flag;
757 				new |= FRDAHEAD;
758 			} while (!atomic_cmpset_rel_int(&fp->f_flag, old, new));
759 		} else {
760 			do {
761 				new = old = fp->f_flag;
762 				new &= ~FRDAHEAD;
763 			} while (!atomic_cmpset_rel_int(&fp->f_flag, old, new));
764 		}
765 		fdrop(fp, td);
766 		break;
767 
768 	default:
769 		error = EINVAL;
770 		break;
771 	}
772 	return (error);
773 }
774 
775 static int
776 getmaxfd(struct proc *p)
777 {
778 	int maxfd;
779 
780 	PROC_LOCK(p);
781 	maxfd = min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc);
782 	PROC_UNLOCK(p);
783 
784 	return (maxfd);
785 }
786 
787 /*
788  * Common code for dup, dup2, fcntl(F_DUPFD) and fcntl(F_DUP2FD).
789  */
790 int
791 do_dup(struct thread *td, int flags, int old, int new,
792     register_t *retval)
793 {
794 	struct filedesc *fdp;
795 	struct filedescent *oldfde, *newfde;
796 	struct proc *p;
797 	struct file *fp;
798 	struct file *delfp;
799 	int error, maxfd;
800 
801 	p = td->td_proc;
802 	fdp = p->p_fd;
803 
804 	/*
805 	 * Verify we have a valid descriptor to dup from and possibly to
806 	 * dup to. Unlike dup() and dup2(), fcntl()'s F_DUPFD should
807 	 * return EINVAL when the new descriptor is out of bounds.
808 	 */
809 	if (old < 0)
810 		return (EBADF);
811 	if (new < 0)
812 		return (flags & DUP_FCNTL ? EINVAL : EBADF);
813 	maxfd = getmaxfd(p);
814 	if (new >= maxfd)
815 		return (flags & DUP_FCNTL ? EINVAL : EBADF);
816 
817 	FILEDESC_XLOCK(fdp);
818 	if (fget_locked(fdp, old) == NULL) {
819 		FILEDESC_XUNLOCK(fdp);
820 		return (EBADF);
821 	}
822 	oldfde = &fdp->fd_ofiles[old];
823 	if (flags & DUP_FIXED && old == new) {
824 		*retval = new;
825 		if (flags & DUP_CLOEXEC)
826 			fdp->fd_ofiles[new].fde_flags |= UF_EXCLOSE;
827 		FILEDESC_XUNLOCK(fdp);
828 		return (0);
829 	}
830 	fp = oldfde->fde_file;
831 	fhold(fp);
832 
833 	/*
834 	 * If the caller specified a file descriptor, make sure the file
835 	 * table is large enough to hold it, and grab it.  Otherwise, just
836 	 * allocate a new descriptor the usual way.
837 	 */
838 	if (flags & DUP_FIXED) {
839 		if (new >= fdp->fd_nfiles) {
840 			/*
841 			 * The resource limits are here instead of e.g.
842 			 * fdalloc(), because the file descriptor table may be
843 			 * shared between processes, so we can't really use
844 			 * racct_add()/racct_sub().  Instead of counting the
845 			 * number of actually allocated descriptors, just put
846 			 * the limit on the size of the file descriptor table.
847 			 */
848 #ifdef RACCT
849 			PROC_LOCK(p);
850 			error = racct_set(p, RACCT_NOFILE, new + 1);
851 			PROC_UNLOCK(p);
852 			if (error != 0) {
853 				FILEDESC_XUNLOCK(fdp);
854 				fdrop(fp, td);
855 				return (EMFILE);
856 			}
857 #endif
858 			fdgrowtable_exp(fdp, new + 1);
859 			oldfde = &fdp->fd_ofiles[old];
860 		}
861 		newfde = &fdp->fd_ofiles[new];
862 		if (newfde->fde_file == NULL)
863 			fdused(fdp, new);
864 	} else {
865 		if ((error = fdalloc(td, new, &new)) != 0) {
866 			FILEDESC_XUNLOCK(fdp);
867 			fdrop(fp, td);
868 			return (error);
869 		}
870 		newfde = &fdp->fd_ofiles[new];
871 	}
872 
873 	KASSERT(fp == oldfde->fde_file, ("old fd has been modified"));
874 	KASSERT(old != new, ("new fd is same as old"));
875 
876 	delfp = newfde->fde_file;
877 
878 	/*
879 	 * Duplicate the source descriptor.
880 	 */
881 	filecaps_free(&newfde->fde_caps);
882 	*newfde = *oldfde;
883 	filecaps_copy(&oldfde->fde_caps, &newfde->fde_caps);
884 	if ((flags & DUP_CLOEXEC) != 0)
885 		newfde->fde_flags = oldfde->fde_flags | UF_EXCLOSE;
886 	else
887 		newfde->fde_flags = oldfde->fde_flags & ~UF_EXCLOSE;
888 	if (new > fdp->fd_lastfile)
889 		fdp->fd_lastfile = new;
890 	*retval = new;
891 
892 	if (delfp != NULL) {
893 		(void) closefp(fdp, new, delfp, td, 1);
894 		/* closefp() drops the FILEDESC lock for us. */
895 	} else {
896 		FILEDESC_XUNLOCK(fdp);
897 	}
898 
899 	return (0);
900 }
901 
902 /*
903  * If sigio is on the list associated with a process or process group,
904  * disable signalling from the device, remove sigio from the list and
905  * free sigio.
906  */
907 void
908 funsetown(struct sigio **sigiop)
909 {
910 	struct sigio *sigio;
911 
912 	SIGIO_LOCK();
913 	sigio = *sigiop;
914 	if (sigio == NULL) {
915 		SIGIO_UNLOCK();
916 		return;
917 	}
918 	*(sigio->sio_myref) = NULL;
919 	if ((sigio)->sio_pgid < 0) {
920 		struct pgrp *pg = (sigio)->sio_pgrp;
921 		PGRP_LOCK(pg);
922 		SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio,
923 			     sigio, sio_pgsigio);
924 		PGRP_UNLOCK(pg);
925 	} else {
926 		struct proc *p = (sigio)->sio_proc;
927 		PROC_LOCK(p);
928 		SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio,
929 			     sigio, sio_pgsigio);
930 		PROC_UNLOCK(p);
931 	}
932 	SIGIO_UNLOCK();
933 	crfree(sigio->sio_ucred);
934 	free(sigio, M_SIGIO);
935 }
936 
937 /*
938  * Free a list of sigio structures.
939  * We only need to lock the SIGIO_LOCK because we have made ourselves
940  * inaccessible to callers of fsetown and therefore do not need to lock
941  * the proc or pgrp struct for the list manipulation.
942  */
943 void
944 funsetownlst(struct sigiolst *sigiolst)
945 {
946 	struct proc *p;
947 	struct pgrp *pg;
948 	struct sigio *sigio;
949 
950 	sigio = SLIST_FIRST(sigiolst);
951 	if (sigio == NULL)
952 		return;
953 	p = NULL;
954 	pg = NULL;
955 
956 	/*
957 	 * Every entry of the list should belong
958 	 * to a single proc or pgrp.
959 	 */
960 	if (sigio->sio_pgid < 0) {
961 		pg = sigio->sio_pgrp;
962 		PGRP_LOCK_ASSERT(pg, MA_NOTOWNED);
963 	} else /* if (sigio->sio_pgid > 0) */ {
964 		p = sigio->sio_proc;
965 		PROC_LOCK_ASSERT(p, MA_NOTOWNED);
966 	}
967 
968 	SIGIO_LOCK();
969 	while ((sigio = SLIST_FIRST(sigiolst)) != NULL) {
970 		*(sigio->sio_myref) = NULL;
971 		if (pg != NULL) {
972 			KASSERT(sigio->sio_pgid < 0,
973 			    ("Proc sigio in pgrp sigio list"));
974 			KASSERT(sigio->sio_pgrp == pg,
975 			    ("Bogus pgrp in sigio list"));
976 			PGRP_LOCK(pg);
977 			SLIST_REMOVE(&pg->pg_sigiolst, sigio, sigio,
978 			    sio_pgsigio);
979 			PGRP_UNLOCK(pg);
980 		} else /* if (p != NULL) */ {
981 			KASSERT(sigio->sio_pgid > 0,
982 			    ("Pgrp sigio in proc sigio list"));
983 			KASSERT(sigio->sio_proc == p,
984 			    ("Bogus proc in sigio list"));
985 			PROC_LOCK(p);
986 			SLIST_REMOVE(&p->p_sigiolst, sigio, sigio,
987 			    sio_pgsigio);
988 			PROC_UNLOCK(p);
989 		}
990 		SIGIO_UNLOCK();
991 		crfree(sigio->sio_ucred);
992 		free(sigio, M_SIGIO);
993 		SIGIO_LOCK();
994 	}
995 	SIGIO_UNLOCK();
996 }
997 
998 /*
999  * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg).
1000  *
1001  * After permission checking, add a sigio structure to the sigio list for
1002  * the process or process group.
1003  */
1004 int
1005 fsetown(pid_t pgid, struct sigio **sigiop)
1006 {
1007 	struct proc *proc;
1008 	struct pgrp *pgrp;
1009 	struct sigio *sigio;
1010 	int ret;
1011 
1012 	if (pgid == 0) {
1013 		funsetown(sigiop);
1014 		return (0);
1015 	}
1016 
1017 	ret = 0;
1018 
1019 	/* Allocate and fill in the new sigio out of locks. */
1020 	sigio = malloc(sizeof(struct sigio), M_SIGIO, M_WAITOK);
1021 	sigio->sio_pgid = pgid;
1022 	sigio->sio_ucred = crhold(curthread->td_ucred);
1023 	sigio->sio_myref = sigiop;
1024 
1025 	sx_slock(&proctree_lock);
1026 	if (pgid > 0) {
1027 		proc = pfind(pgid);
1028 		if (proc == NULL) {
1029 			ret = ESRCH;
1030 			goto fail;
1031 		}
1032 
1033 		/*
1034 		 * Policy - Don't allow a process to FSETOWN a process
1035 		 * in another session.
1036 		 *
1037 		 * Remove this test to allow maximum flexibility or
1038 		 * restrict FSETOWN to the current process or process
1039 		 * group for maximum safety.
1040 		 */
1041 		PROC_UNLOCK(proc);
1042 		if (proc->p_session != curthread->td_proc->p_session) {
1043 			ret = EPERM;
1044 			goto fail;
1045 		}
1046 
1047 		pgrp = NULL;
1048 	} else /* if (pgid < 0) */ {
1049 		pgrp = pgfind(-pgid);
1050 		if (pgrp == NULL) {
1051 			ret = ESRCH;
1052 			goto fail;
1053 		}
1054 		PGRP_UNLOCK(pgrp);
1055 
1056 		/*
1057 		 * Policy - Don't allow a process to FSETOWN a process
1058 		 * in another session.
1059 		 *
1060 		 * Remove this test to allow maximum flexibility or
1061 		 * restrict FSETOWN to the current process or process
1062 		 * group for maximum safety.
1063 		 */
1064 		if (pgrp->pg_session != curthread->td_proc->p_session) {
1065 			ret = EPERM;
1066 			goto fail;
1067 		}
1068 
1069 		proc = NULL;
1070 	}
1071 	funsetown(sigiop);
1072 	if (pgid > 0) {
1073 		PROC_LOCK(proc);
1074 		/*
1075 		 * Since funsetownlst() is called without the proctree
1076 		 * locked, we need to check for P_WEXIT.
1077 		 * XXX: is ESRCH correct?
1078 		 */
1079 		if ((proc->p_flag & P_WEXIT) != 0) {
1080 			PROC_UNLOCK(proc);
1081 			ret = ESRCH;
1082 			goto fail;
1083 		}
1084 		SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio);
1085 		sigio->sio_proc = proc;
1086 		PROC_UNLOCK(proc);
1087 	} else {
1088 		PGRP_LOCK(pgrp);
1089 		SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio);
1090 		sigio->sio_pgrp = pgrp;
1091 		PGRP_UNLOCK(pgrp);
1092 	}
1093 	sx_sunlock(&proctree_lock);
1094 	SIGIO_LOCK();
1095 	*sigiop = sigio;
1096 	SIGIO_UNLOCK();
1097 	return (0);
1098 
1099 fail:
1100 	sx_sunlock(&proctree_lock);
1101 	crfree(sigio->sio_ucred);
1102 	free(sigio, M_SIGIO);
1103 	return (ret);
1104 }
1105 
1106 /*
1107  * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg).
1108  */
1109 pid_t
1110 fgetown(sigiop)
1111 	struct sigio **sigiop;
1112 {
1113 	pid_t pgid;
1114 
1115 	SIGIO_LOCK();
1116 	pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0;
1117 	SIGIO_UNLOCK();
1118 	return (pgid);
1119 }
1120 
1121 /*
1122  * Function drops the filedesc lock on return.
1123  */
1124 static int
1125 closefp(struct filedesc *fdp, int fd, struct file *fp, struct thread *td,
1126     int holdleaders)
1127 {
1128 	int error;
1129 
1130 	FILEDESC_XLOCK_ASSERT(fdp);
1131 
1132 	if (holdleaders) {
1133 		if (td->td_proc->p_fdtol != NULL) {
1134 			/*
1135 			 * Ask fdfree() to sleep to ensure that all relevant
1136 			 * process leaders can be traversed in closef().
1137 			 */
1138 			fdp->fd_holdleaderscount++;
1139 		} else {
1140 			holdleaders = 0;
1141 		}
1142 	}
1143 
1144 	/*
1145 	 * We now hold the fp reference that used to be owned by the
1146 	 * descriptor array.  We have to unlock the FILEDESC *AFTER*
1147 	 * knote_fdclose to prevent a race of the fd getting opened, a knote
1148 	 * added, and deleteing a knote for the new fd.
1149 	 */
1150 	knote_fdclose(td, fd);
1151 
1152 	/*
1153 	 * We need to notify mqueue if the object is of type mqueue.
1154 	 */
1155 	if (fp->f_type == DTYPE_MQUEUE)
1156 		mq_fdclose(td, fd, fp);
1157 	FILEDESC_XUNLOCK(fdp);
1158 
1159 	error = closef(fp, td);
1160 	if (holdleaders) {
1161 		FILEDESC_XLOCK(fdp);
1162 		fdp->fd_holdleaderscount--;
1163 		if (fdp->fd_holdleaderscount == 0 &&
1164 		    fdp->fd_holdleaderswakeup != 0) {
1165 			fdp->fd_holdleaderswakeup = 0;
1166 			wakeup(&fdp->fd_holdleaderscount);
1167 		}
1168 		FILEDESC_XUNLOCK(fdp);
1169 	}
1170 	return (error);
1171 }
1172 
1173 /*
1174  * Close a file descriptor.
1175  */
1176 #ifndef _SYS_SYSPROTO_H_
1177 struct close_args {
1178 	int     fd;
1179 };
1180 #endif
1181 /* ARGSUSED */
1182 int
1183 sys_close(td, uap)
1184 	struct thread *td;
1185 	struct close_args *uap;
1186 {
1187 
1188 	return (kern_close(td, uap->fd));
1189 }
1190 
1191 int
1192 kern_close(td, fd)
1193 	struct thread *td;
1194 	int fd;
1195 {
1196 	struct filedesc *fdp;
1197 	struct file *fp;
1198 
1199 	fdp = td->td_proc->p_fd;
1200 
1201 	AUDIT_SYSCLOSE(td, fd);
1202 
1203 	FILEDESC_XLOCK(fdp);
1204 	if ((fp = fget_locked(fdp, fd)) == NULL) {
1205 		FILEDESC_XUNLOCK(fdp);
1206 		return (EBADF);
1207 	}
1208 	fdfree(fdp, fd);
1209 
1210 	/* closefp() drops the FILEDESC lock for us. */
1211 	return (closefp(fdp, fd, fp, td, 1));
1212 }
1213 
1214 /*
1215  * Close open file descriptors.
1216  */
1217 #ifndef _SYS_SYSPROTO_H_
1218 struct closefrom_args {
1219 	int	lowfd;
1220 };
1221 #endif
1222 /* ARGSUSED */
1223 int
1224 sys_closefrom(struct thread *td, struct closefrom_args *uap)
1225 {
1226 	struct filedesc *fdp;
1227 	int fd;
1228 
1229 	fdp = td->td_proc->p_fd;
1230 	AUDIT_ARG_FD(uap->lowfd);
1231 
1232 	/*
1233 	 * Treat negative starting file descriptor values identical to
1234 	 * closefrom(0) which closes all files.
1235 	 */
1236 	if (uap->lowfd < 0)
1237 		uap->lowfd = 0;
1238 	FILEDESC_SLOCK(fdp);
1239 	for (fd = uap->lowfd; fd < fdp->fd_nfiles; fd++) {
1240 		if (fdp->fd_ofiles[fd].fde_file != NULL) {
1241 			FILEDESC_SUNLOCK(fdp);
1242 			(void)kern_close(td, fd);
1243 			FILEDESC_SLOCK(fdp);
1244 		}
1245 	}
1246 	FILEDESC_SUNLOCK(fdp);
1247 	return (0);
1248 }
1249 
1250 #if defined(COMPAT_43)
1251 /*
1252  * Return status information about a file descriptor.
1253  */
1254 #ifndef _SYS_SYSPROTO_H_
1255 struct ofstat_args {
1256 	int	fd;
1257 	struct	ostat *sb;
1258 };
1259 #endif
1260 /* ARGSUSED */
1261 int
1262 ofstat(struct thread *td, struct ofstat_args *uap)
1263 {
1264 	struct ostat oub;
1265 	struct stat ub;
1266 	int error;
1267 
1268 	error = kern_fstat(td, uap->fd, &ub);
1269 	if (error == 0) {
1270 		cvtstat(&ub, &oub);
1271 		error = copyout(&oub, uap->sb, sizeof(oub));
1272 	}
1273 	return (error);
1274 }
1275 #endif /* COMPAT_43 */
1276 
1277 /*
1278  * Return status information about a file descriptor.
1279  */
1280 #ifndef _SYS_SYSPROTO_H_
1281 struct fstat_args {
1282 	int	fd;
1283 	struct	stat *sb;
1284 };
1285 #endif
1286 /* ARGSUSED */
1287 int
1288 sys_fstat(struct thread *td, struct fstat_args *uap)
1289 {
1290 	struct stat ub;
1291 	int error;
1292 
1293 	error = kern_fstat(td, uap->fd, &ub);
1294 	if (error == 0)
1295 		error = copyout(&ub, uap->sb, sizeof(ub));
1296 	return (error);
1297 }
1298 
1299 int
1300 kern_fstat(struct thread *td, int fd, struct stat *sbp)
1301 {
1302 	struct file *fp;
1303 	cap_rights_t rights;
1304 	int error;
1305 
1306 	AUDIT_ARG_FD(fd);
1307 
1308 	error = fget(td, fd, cap_rights_init(&rights, CAP_FSTAT), &fp);
1309 	if (error != 0)
1310 		return (error);
1311 
1312 	AUDIT_ARG_FILE(td->td_proc, fp);
1313 
1314 	error = fo_stat(fp, sbp, td->td_ucred, td);
1315 	fdrop(fp, td);
1316 #ifdef KTRACE
1317 	if (error == 0 && KTRPOINT(td, KTR_STRUCT))
1318 		ktrstat(sbp);
1319 #endif
1320 	return (error);
1321 }
1322 
1323 /*
1324  * Return status information about a file descriptor.
1325  */
1326 #ifndef _SYS_SYSPROTO_H_
1327 struct nfstat_args {
1328 	int	fd;
1329 	struct	nstat *sb;
1330 };
1331 #endif
1332 /* ARGSUSED */
1333 int
1334 sys_nfstat(struct thread *td, struct nfstat_args *uap)
1335 {
1336 	struct nstat nub;
1337 	struct stat ub;
1338 	int error;
1339 
1340 	error = kern_fstat(td, uap->fd, &ub);
1341 	if (error == 0) {
1342 		cvtnstat(&ub, &nub);
1343 		error = copyout(&nub, uap->sb, sizeof(nub));
1344 	}
1345 	return (error);
1346 }
1347 
1348 /*
1349  * Return pathconf information about a file descriptor.
1350  */
1351 #ifndef _SYS_SYSPROTO_H_
1352 struct fpathconf_args {
1353 	int	fd;
1354 	int	name;
1355 };
1356 #endif
1357 /* ARGSUSED */
1358 int
1359 sys_fpathconf(struct thread *td, struct fpathconf_args *uap)
1360 {
1361 	struct file *fp;
1362 	struct vnode *vp;
1363 	cap_rights_t rights;
1364 	int error;
1365 
1366 	error = fget(td, uap->fd, cap_rights_init(&rights, CAP_FPATHCONF), &fp);
1367 	if (error != 0)
1368 		return (error);
1369 
1370 	/* If asynchronous I/O is available, it works for all descriptors. */
1371 	if (uap->name == _PC_ASYNC_IO) {
1372 		td->td_retval[0] = async_io_version;
1373 		goto out;
1374 	}
1375 	vp = fp->f_vnode;
1376 	if (vp != NULL) {
1377 		vn_lock(vp, LK_SHARED | LK_RETRY);
1378 		error = VOP_PATHCONF(vp, uap->name, td->td_retval);
1379 		VOP_UNLOCK(vp, 0);
1380 	} else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) {
1381 		if (uap->name != _PC_PIPE_BUF) {
1382 			error = EINVAL;
1383 		} else {
1384 			td->td_retval[0] = PIPE_BUF;
1385 			error = 0;
1386 		}
1387 	} else {
1388 		error = EOPNOTSUPP;
1389 	}
1390 out:
1391 	fdrop(fp, td);
1392 	return (error);
1393 }
1394 
1395 /*
1396  * Initialize filecaps structure.
1397  */
1398 void
1399 filecaps_init(struct filecaps *fcaps)
1400 {
1401 
1402 	bzero(fcaps, sizeof(*fcaps));
1403 	fcaps->fc_nioctls = -1;
1404 }
1405 
1406 /*
1407  * Copy filecaps structure allocating memory for ioctls array if needed.
1408  */
1409 void
1410 filecaps_copy(const struct filecaps *src, struct filecaps *dst)
1411 {
1412 	size_t size;
1413 
1414 	*dst = *src;
1415 	if (src->fc_ioctls != NULL) {
1416 		KASSERT(src->fc_nioctls > 0,
1417 		    ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls));
1418 
1419 		size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls;
1420 		dst->fc_ioctls = malloc(size, M_FILECAPS, M_WAITOK);
1421 		bcopy(src->fc_ioctls, dst->fc_ioctls, size);
1422 	}
1423 }
1424 
1425 /*
1426  * Move filecaps structure to the new place and clear the old place.
1427  */
1428 void
1429 filecaps_move(struct filecaps *src, struct filecaps *dst)
1430 {
1431 
1432 	*dst = *src;
1433 	bzero(src, sizeof(*src));
1434 }
1435 
1436 /*
1437  * Fill the given filecaps structure with full rights.
1438  */
1439 static void
1440 filecaps_fill(struct filecaps *fcaps)
1441 {
1442 
1443 	CAP_ALL(&fcaps->fc_rights);
1444 	fcaps->fc_ioctls = NULL;
1445 	fcaps->fc_nioctls = -1;
1446 	fcaps->fc_fcntls = CAP_FCNTL_ALL;
1447 }
1448 
1449 /*
1450  * Free memory allocated within filecaps structure.
1451  */
1452 void
1453 filecaps_free(struct filecaps *fcaps)
1454 {
1455 
1456 	free(fcaps->fc_ioctls, M_FILECAPS);
1457 	bzero(fcaps, sizeof(*fcaps));
1458 }
1459 
1460 /*
1461  * Validate the given filecaps structure.
1462  */
1463 static void
1464 filecaps_validate(const struct filecaps *fcaps, const char *func)
1465 {
1466 
1467 	KASSERT(cap_rights_is_valid(&fcaps->fc_rights),
1468 	    ("%s: invalid rights", func));
1469 	KASSERT((fcaps->fc_fcntls & ~CAP_FCNTL_ALL) == 0,
1470 	    ("%s: invalid fcntls", func));
1471 	KASSERT(fcaps->fc_fcntls == 0 ||
1472 	    cap_rights_is_set(&fcaps->fc_rights, CAP_FCNTL),
1473 	    ("%s: fcntls without CAP_FCNTL", func));
1474 	KASSERT(fcaps->fc_ioctls != NULL ? fcaps->fc_nioctls > 0 :
1475 	    (fcaps->fc_nioctls == -1 || fcaps->fc_nioctls == 0),
1476 	    ("%s: invalid ioctls", func));
1477 	KASSERT(fcaps->fc_nioctls == 0 ||
1478 	    cap_rights_is_set(&fcaps->fc_rights, CAP_IOCTL),
1479 	    ("%s: ioctls without CAP_IOCTL", func));
1480 }
1481 
1482 static void
1483 fdgrowtable_exp(struct filedesc *fdp, int nfd)
1484 {
1485 	int nfd1;
1486 
1487 	FILEDESC_XLOCK_ASSERT(fdp);
1488 
1489 	nfd1 = fdp->fd_nfiles * 2;
1490 	if (nfd1 < nfd)
1491 		nfd1 = nfd;
1492 	fdgrowtable(fdp, nfd1);
1493 }
1494 
1495 /*
1496  * Grow the file table to accomodate (at least) nfd descriptors.
1497  */
1498 static void
1499 fdgrowtable(struct filedesc *fdp, int nfd)
1500 {
1501 	struct filedesc0 *fdp0;
1502 	struct freetable *ft;
1503 	struct filedescent *ntable;
1504 	struct filedescent *otable;
1505 	int nnfiles, onfiles;
1506 	NDSLOTTYPE *nmap, *omap;
1507 
1508 	FILEDESC_XLOCK_ASSERT(fdp);
1509 
1510 	KASSERT(fdp->fd_nfiles > 0, ("zero-length file table"));
1511 
1512 	/* save old values */
1513 	onfiles = fdp->fd_nfiles;
1514 	otable = fdp->fd_ofiles;
1515 	omap = fdp->fd_map;
1516 
1517 	/* compute the size of the new table */
1518 	nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */
1519 	if (nnfiles <= onfiles)
1520 		/* the table is already large enough */
1521 		return;
1522 
1523 	/*
1524 	 * Allocate a new table and map.  We need enough space for the
1525 	 * file entries themselves and the struct freetable we will use
1526 	 * when we decommission the table and place it on the freelist.
1527 	 * We place the struct freetable in the middle so we don't have
1528 	 * to worry about padding.
1529 	 */
1530 	ntable = malloc(nnfiles * sizeof(ntable[0]) + sizeof(struct freetable),
1531 	    M_FILEDESC, M_ZERO | M_WAITOK);
1532 	nmap = malloc(NDSLOTS(nnfiles) * NDSLOTSIZE, M_FILEDESC,
1533 	    M_ZERO | M_WAITOK);
1534 
1535 	/* copy the old data over and point at the new tables */
1536 	memcpy(ntable, otable, onfiles * sizeof(*otable));
1537 	memcpy(nmap, omap, NDSLOTS(onfiles) * sizeof(*omap));
1538 
1539 	/* update the pointers and counters */
1540 	memcpy(ntable, otable, onfiles * sizeof(ntable[0]));
1541 	fdp->fd_ofiles = ntable;
1542 	fdp->fd_map = nmap;
1543 
1544 	/*
1545 	 * In order to have a valid pattern for fget_unlocked()
1546 	 * fdp->fd_nfiles must be the last member to be updated, otherwise
1547 	 * fget_unlocked() consumers may reference a new, higher value for
1548 	 * fdp->fd_nfiles before to access the fdp->fd_ofiles array,
1549 	 * resulting in OOB accesses.
1550 	 */
1551 	atomic_store_rel_int(&fdp->fd_nfiles, nnfiles);
1552 
1553 	/*
1554 	 * Do not free the old file table, as some threads may still
1555 	 * reference entries within it.  Instead, place it on a freelist
1556 	 * which will be processed when the struct filedesc is released.
1557 	 *
1558 	 * Do, however, free the old map.
1559 	 *
1560 	 * Note that if onfiles == NDFILE, we're dealing with the original
1561 	 * static allocation contained within (struct filedesc0 *)fdp,
1562 	 * which must not be freed.
1563 	 */
1564 	if (onfiles > NDFILE) {
1565 		ft = (struct freetable *)&otable[onfiles];
1566 		fdp0 = (struct filedesc0 *)fdp;
1567 		ft->ft_table = otable;
1568 		SLIST_INSERT_HEAD(&fdp0->fd_free, ft, ft_next);
1569 		free(omap, M_FILEDESC);
1570 	}
1571 }
1572 
1573 /*
1574  * Allocate a file descriptor for the process.
1575  */
1576 int
1577 fdalloc(struct thread *td, int minfd, int *result)
1578 {
1579 	struct proc *p = td->td_proc;
1580 	struct filedesc *fdp = p->p_fd;
1581 	int fd = -1, maxfd, allocfd;
1582 #ifdef RACCT
1583 	int error;
1584 #endif
1585 
1586 	FILEDESC_XLOCK_ASSERT(fdp);
1587 
1588 	if (fdp->fd_freefile > minfd)
1589 		minfd = fdp->fd_freefile;
1590 
1591 	maxfd = getmaxfd(p);
1592 
1593 	/*
1594 	 * Search the bitmap for a free descriptor starting at minfd.
1595 	 * If none is found, grow the file table.
1596 	 */
1597 	fd = fd_first_free(fdp, minfd, fdp->fd_nfiles);
1598 	if (fd >= maxfd)
1599 		return (EMFILE);
1600 	if (fd >= fdp->fd_nfiles) {
1601 		allocfd = min(fd * 2, maxfd);
1602 #ifdef RACCT
1603 		PROC_LOCK(p);
1604 		error = racct_set(p, RACCT_NOFILE, allocfd);
1605 		PROC_UNLOCK(p);
1606 		if (error != 0)
1607 			return (EMFILE);
1608 #endif
1609 		/*
1610 		 * fd is already equal to first free descriptor >= minfd, so
1611 		 * we only need to grow the table and we are done.
1612 		 */
1613 		fdgrowtable_exp(fdp, allocfd);
1614 	}
1615 
1616 	/*
1617 	 * Perform some sanity checks, then mark the file descriptor as
1618 	 * used and return it to the caller.
1619 	 */
1620 	KASSERT(fd >= 0 && fd < min(maxfd, fdp->fd_nfiles),
1621 	    ("invalid descriptor %d", fd));
1622 	KASSERT(!fdisused(fdp, fd),
1623 	    ("fd_first_free() returned non-free descriptor"));
1624 	KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
1625 	    ("file descriptor isn't free"));
1626 	KASSERT(fdp->fd_ofiles[fd].fde_flags == 0, ("file flags are set"));
1627 	fdused(fdp, fd);
1628 	*result = fd;
1629 	return (0);
1630 }
1631 
1632 /*
1633  * Allocate n file descriptors for the process.
1634  */
1635 int
1636 fdallocn(struct thread *td, int minfd, int *fds, int n)
1637 {
1638 	struct proc *p = td->td_proc;
1639 	struct filedesc *fdp = p->p_fd;
1640 	int i;
1641 
1642 	FILEDESC_XLOCK_ASSERT(fdp);
1643 
1644 	if (!fdavail(td, n))
1645 		return (EMFILE);
1646 
1647 	for (i = 0; i < n; i++)
1648 		if (fdalloc(td, 0, &fds[i]) != 0)
1649 			break;
1650 
1651 	if (i < n) {
1652 		for (i--; i >= 0; i--)
1653 			fdunused(fdp, fds[i]);
1654 		return (EMFILE);
1655 	}
1656 
1657 	return (0);
1658 }
1659 
1660 /*
1661  * Check to see whether n user file descriptors are available to the process
1662  * p.
1663  */
1664 int
1665 fdavail(struct thread *td, int n)
1666 {
1667 	struct proc *p = td->td_proc;
1668 	struct filedesc *fdp = td->td_proc->p_fd;
1669 	int i, lim, last;
1670 
1671 	FILEDESC_LOCK_ASSERT(fdp);
1672 
1673 	/*
1674 	 * XXX: This is only called from uipc_usrreq.c:unp_externalize();
1675 	 *      call racct_add() from there instead of dealing with containers
1676 	 *      here.
1677 	 */
1678 	lim = getmaxfd(p);
1679 	if ((i = lim - fdp->fd_nfiles) > 0 && (n -= i) <= 0)
1680 		return (1);
1681 	last = min(fdp->fd_nfiles, lim);
1682 	for (i = fdp->fd_freefile; i < last; i++) {
1683 		if (fdp->fd_ofiles[i].fde_file == NULL && --n <= 0)
1684 			return (1);
1685 	}
1686 	return (0);
1687 }
1688 
1689 /*
1690  * Create a new open file structure and allocate a file decriptor for the
1691  * process that refers to it.  We add one reference to the file for the
1692  * descriptor table and one reference for resultfp. This is to prevent us
1693  * being preempted and the entry in the descriptor table closed after we
1694  * release the FILEDESC lock.
1695  */
1696 int
1697 falloc(struct thread *td, struct file **resultfp, int *resultfd, int flags)
1698 {
1699 	struct file *fp;
1700 	int error, fd;
1701 
1702 	error = falloc_noinstall(td, &fp);
1703 	if (error)
1704 		return (error);		/* no reference held on error */
1705 
1706 	error = finstall(td, fp, &fd, flags, NULL);
1707 	if (error) {
1708 		fdrop(fp, td);		/* one reference (fp only) */
1709 		return (error);
1710 	}
1711 
1712 	if (resultfp != NULL)
1713 		*resultfp = fp;		/* copy out result */
1714 	else
1715 		fdrop(fp, td);		/* release local reference */
1716 
1717 	if (resultfd != NULL)
1718 		*resultfd = fd;
1719 
1720 	return (0);
1721 }
1722 
1723 /*
1724  * Create a new open file structure without allocating a file descriptor.
1725  */
1726 int
1727 falloc_noinstall(struct thread *td, struct file **resultfp)
1728 {
1729 	struct file *fp;
1730 	int maxuserfiles = maxfiles - (maxfiles / 20);
1731 	static struct timeval lastfail;
1732 	static int curfail;
1733 
1734 	KASSERT(resultfp != NULL, ("%s: resultfp == NULL", __func__));
1735 
1736 	if ((openfiles >= maxuserfiles &&
1737 	    priv_check(td, PRIV_MAXFILES) != 0) ||
1738 	    openfiles >= maxfiles) {
1739 		if (ppsratecheck(&lastfail, &curfail, 1)) {
1740 			printf("kern.maxfiles limit exceeded by uid %i, "
1741 			    "please see tuning(7).\n", td->td_ucred->cr_ruid);
1742 		}
1743 		return (ENFILE);
1744 	}
1745 	atomic_add_int(&openfiles, 1);
1746 	fp = uma_zalloc(file_zone, M_WAITOK | M_ZERO);
1747 	refcount_init(&fp->f_count, 1);
1748 	fp->f_cred = crhold(td->td_ucred);
1749 	fp->f_ops = &badfileops;
1750 	fp->f_data = NULL;
1751 	fp->f_vnode = NULL;
1752 	*resultfp = fp;
1753 	return (0);
1754 }
1755 
1756 /*
1757  * Install a file in a file descriptor table.
1758  */
1759 int
1760 finstall(struct thread *td, struct file *fp, int *fd, int flags,
1761     struct filecaps *fcaps)
1762 {
1763 	struct filedesc *fdp = td->td_proc->p_fd;
1764 	struct filedescent *fde;
1765 	int error;
1766 
1767 	KASSERT(fd != NULL, ("%s: fd == NULL", __func__));
1768 	KASSERT(fp != NULL, ("%s: fp == NULL", __func__));
1769 	if (fcaps != NULL)
1770 		filecaps_validate(fcaps, __func__);
1771 
1772 	FILEDESC_XLOCK(fdp);
1773 	if ((error = fdalloc(td, 0, fd))) {
1774 		FILEDESC_XUNLOCK(fdp);
1775 		return (error);
1776 	}
1777 	fhold(fp);
1778 	fde = &fdp->fd_ofiles[*fd];
1779 	fde->fde_file = fp;
1780 	if ((flags & O_CLOEXEC) != 0)
1781 		fde->fde_flags |= UF_EXCLOSE;
1782 	if (fcaps != NULL)
1783 		filecaps_move(fcaps, &fde->fde_caps);
1784 	else
1785 		filecaps_fill(&fde->fde_caps);
1786 	FILEDESC_XUNLOCK(fdp);
1787 	return (0);
1788 }
1789 
1790 /*
1791  * Build a new filedesc structure from another.
1792  * Copy the current, root, and jail root vnode references.
1793  */
1794 struct filedesc *
1795 fdinit(struct filedesc *fdp)
1796 {
1797 	struct filedesc0 *newfdp;
1798 
1799 	newfdp = malloc(sizeof *newfdp, M_FILEDESC, M_WAITOK | M_ZERO);
1800 	FILEDESC_LOCK_INIT(&newfdp->fd_fd);
1801 	if (fdp != NULL) {
1802 		FILEDESC_XLOCK(fdp);
1803 		newfdp->fd_fd.fd_cdir = fdp->fd_cdir;
1804 		if (newfdp->fd_fd.fd_cdir)
1805 			VREF(newfdp->fd_fd.fd_cdir);
1806 		newfdp->fd_fd.fd_rdir = fdp->fd_rdir;
1807 		if (newfdp->fd_fd.fd_rdir)
1808 			VREF(newfdp->fd_fd.fd_rdir);
1809 		newfdp->fd_fd.fd_jdir = fdp->fd_jdir;
1810 		if (newfdp->fd_fd.fd_jdir)
1811 			VREF(newfdp->fd_fd.fd_jdir);
1812 		FILEDESC_XUNLOCK(fdp);
1813 	}
1814 
1815 	/* Create the file descriptor table. */
1816 	newfdp->fd_fd.fd_refcnt = 1;
1817 	newfdp->fd_fd.fd_holdcnt = 1;
1818 	newfdp->fd_fd.fd_cmask = CMASK;
1819 	newfdp->fd_fd.fd_ofiles = newfdp->fd_dfiles;
1820 	newfdp->fd_fd.fd_nfiles = NDFILE;
1821 	newfdp->fd_fd.fd_map = newfdp->fd_dmap;
1822 	newfdp->fd_fd.fd_lastfile = -1;
1823 	return (&newfdp->fd_fd);
1824 }
1825 
1826 static struct filedesc *
1827 fdhold(struct proc *p)
1828 {
1829 	struct filedesc *fdp;
1830 
1831 	mtx_lock(&fdesc_mtx);
1832 	fdp = p->p_fd;
1833 	if (fdp != NULL)
1834 		fdp->fd_holdcnt++;
1835 	mtx_unlock(&fdesc_mtx);
1836 	return (fdp);
1837 }
1838 
1839 static void
1840 fddrop(struct filedesc *fdp)
1841 {
1842 	struct filedesc0 *fdp0;
1843 	struct freetable *ft;
1844 	int i;
1845 
1846 	mtx_lock(&fdesc_mtx);
1847 	i = --fdp->fd_holdcnt;
1848 	mtx_unlock(&fdesc_mtx);
1849 	if (i > 0)
1850 		return;
1851 
1852 	FILEDESC_LOCK_DESTROY(fdp);
1853 	fdp0 = (struct filedesc0 *)fdp;
1854 	while ((ft = SLIST_FIRST(&fdp0->fd_free)) != NULL) {
1855 		SLIST_REMOVE_HEAD(&fdp0->fd_free, ft_next);
1856 		free(ft->ft_table, M_FILEDESC);
1857 	}
1858 	free(fdp, M_FILEDESC);
1859 }
1860 
1861 /*
1862  * Share a filedesc structure.
1863  */
1864 struct filedesc *
1865 fdshare(struct filedesc *fdp)
1866 {
1867 
1868 	FILEDESC_XLOCK(fdp);
1869 	fdp->fd_refcnt++;
1870 	FILEDESC_XUNLOCK(fdp);
1871 	return (fdp);
1872 }
1873 
1874 /*
1875  * Unshare a filedesc structure, if necessary by making a copy
1876  */
1877 void
1878 fdunshare(struct proc *p, struct thread *td)
1879 {
1880 
1881 	FILEDESC_XLOCK(p->p_fd);
1882 	if (p->p_fd->fd_refcnt > 1) {
1883 		struct filedesc *tmp;
1884 
1885 		FILEDESC_XUNLOCK(p->p_fd);
1886 		tmp = fdcopy(p->p_fd);
1887 		fdescfree(td);
1888 		p->p_fd = tmp;
1889 	} else
1890 		FILEDESC_XUNLOCK(p->p_fd);
1891 }
1892 
1893 /*
1894  * Copy a filedesc structure.  A NULL pointer in returns a NULL reference,
1895  * this is to ease callers, not catch errors.
1896  */
1897 struct filedesc *
1898 fdcopy(struct filedesc *fdp)
1899 {
1900 	struct filedesc *newfdp;
1901 	struct filedescent *nfde, *ofde;
1902 	int i;
1903 
1904 	/* Certain daemons might not have file descriptors. */
1905 	if (fdp == NULL)
1906 		return (NULL);
1907 
1908 	newfdp = fdinit(fdp);
1909 	FILEDESC_SLOCK(fdp);
1910 	while (fdp->fd_lastfile >= newfdp->fd_nfiles) {
1911 		FILEDESC_SUNLOCK(fdp);
1912 		FILEDESC_XLOCK(newfdp);
1913 		fdgrowtable(newfdp, fdp->fd_lastfile + 1);
1914 		FILEDESC_XUNLOCK(newfdp);
1915 		FILEDESC_SLOCK(fdp);
1916 	}
1917 	/* copy all passable descriptors (i.e. not kqueue) */
1918 	newfdp->fd_freefile = -1;
1919 	for (i = 0; i <= fdp->fd_lastfile; ++i) {
1920 		ofde = &fdp->fd_ofiles[i];
1921 		if (fdisused(fdp, i) &&
1922 		    (ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) &&
1923 		    ofde->fde_file->f_ops != &badfileops) {
1924 			nfde = &newfdp->fd_ofiles[i];
1925 			*nfde = *ofde;
1926 			filecaps_copy(&ofde->fde_caps, &nfde->fde_caps);
1927 			fhold(nfde->fde_file);
1928 			newfdp->fd_lastfile = i;
1929 		} else {
1930 			if (newfdp->fd_freefile == -1)
1931 				newfdp->fd_freefile = i;
1932 		}
1933 	}
1934 	newfdp->fd_cmask = fdp->fd_cmask;
1935 	FILEDESC_SUNLOCK(fdp);
1936 	FILEDESC_XLOCK(newfdp);
1937 	for (i = 0; i <= newfdp->fd_lastfile; ++i) {
1938 		if (newfdp->fd_ofiles[i].fde_file != NULL)
1939 			fdused(newfdp, i);
1940 	}
1941 	if (newfdp->fd_freefile == -1)
1942 		newfdp->fd_freefile = i;
1943 	FILEDESC_XUNLOCK(newfdp);
1944 	return (newfdp);
1945 }
1946 
1947 /*
1948  * Release a filedesc structure.
1949  */
1950 void
1951 fdescfree(struct thread *td)
1952 {
1953 	struct filedesc *fdp;
1954 	int i;
1955 	struct filedesc_to_leader *fdtol;
1956 	struct file *fp;
1957 	struct vnode *cdir, *jdir, *rdir, *vp;
1958 	struct flock lf;
1959 
1960 	/* Certain daemons might not have file descriptors. */
1961 	fdp = td->td_proc->p_fd;
1962 	if (fdp == NULL)
1963 		return;
1964 
1965 #ifdef RACCT
1966 	PROC_LOCK(td->td_proc);
1967 	racct_set(td->td_proc, RACCT_NOFILE, 0);
1968 	PROC_UNLOCK(td->td_proc);
1969 #endif
1970 
1971 	/* Check for special need to clear POSIX style locks */
1972 	fdtol = td->td_proc->p_fdtol;
1973 	if (fdtol != NULL) {
1974 		FILEDESC_XLOCK(fdp);
1975 		KASSERT(fdtol->fdl_refcount > 0,
1976 		    ("filedesc_to_refcount botch: fdl_refcount=%d",
1977 		    fdtol->fdl_refcount));
1978 		if (fdtol->fdl_refcount == 1 &&
1979 		    (td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
1980 			for (i = 0; i <= fdp->fd_lastfile; i++) {
1981 				fp = fdp->fd_ofiles[i].fde_file;
1982 				if (fp == NULL || fp->f_type != DTYPE_VNODE)
1983 					continue;
1984 				fhold(fp);
1985 				FILEDESC_XUNLOCK(fdp);
1986 				lf.l_whence = SEEK_SET;
1987 				lf.l_start = 0;
1988 				lf.l_len = 0;
1989 				lf.l_type = F_UNLCK;
1990 				vp = fp->f_vnode;
1991 				(void) VOP_ADVLOCK(vp,
1992 				    (caddr_t)td->td_proc->p_leader, F_UNLCK,
1993 				    &lf, F_POSIX);
1994 				FILEDESC_XLOCK(fdp);
1995 				fdrop(fp, td);
1996 			}
1997 		}
1998 	retry:
1999 		if (fdtol->fdl_refcount == 1) {
2000 			if (fdp->fd_holdleaderscount > 0 &&
2001 			    (td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
2002 				/*
2003 				 * close() or do_dup() has cleared a reference
2004 				 * in a shared file descriptor table.
2005 				 */
2006 				fdp->fd_holdleaderswakeup = 1;
2007 				sx_sleep(&fdp->fd_holdleaderscount,
2008 				    FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0);
2009 				goto retry;
2010 			}
2011 			if (fdtol->fdl_holdcount > 0) {
2012 				/*
2013 				 * Ensure that fdtol->fdl_leader remains
2014 				 * valid in closef().
2015 				 */
2016 				fdtol->fdl_wakeup = 1;
2017 				sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK,
2018 				    "fdlhold", 0);
2019 				goto retry;
2020 			}
2021 		}
2022 		fdtol->fdl_refcount--;
2023 		if (fdtol->fdl_refcount == 0 &&
2024 		    fdtol->fdl_holdcount == 0) {
2025 			fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
2026 			fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
2027 		} else
2028 			fdtol = NULL;
2029 		td->td_proc->p_fdtol = NULL;
2030 		FILEDESC_XUNLOCK(fdp);
2031 		if (fdtol != NULL)
2032 			free(fdtol, M_FILEDESC_TO_LEADER);
2033 	}
2034 	FILEDESC_XLOCK(fdp);
2035 	i = --fdp->fd_refcnt;
2036 	FILEDESC_XUNLOCK(fdp);
2037 	if (i > 0)
2038 		return;
2039 
2040 	for (i = 0; i <= fdp->fd_lastfile; i++) {
2041 		fp = fdp->fd_ofiles[i].fde_file;
2042 		if (fp != NULL) {
2043 			FILEDESC_XLOCK(fdp);
2044 			fdfree(fdp, i);
2045 			FILEDESC_XUNLOCK(fdp);
2046 			(void) closef(fp, td);
2047 		}
2048 	}
2049 	FILEDESC_XLOCK(fdp);
2050 
2051 	/* XXX This should happen earlier. */
2052 	mtx_lock(&fdesc_mtx);
2053 	td->td_proc->p_fd = NULL;
2054 	mtx_unlock(&fdesc_mtx);
2055 
2056 	if (fdp->fd_nfiles > NDFILE)
2057 		free(fdp->fd_ofiles, M_FILEDESC);
2058 	if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE))
2059 		free(fdp->fd_map, M_FILEDESC);
2060 
2061 	fdp->fd_nfiles = 0;
2062 
2063 	cdir = fdp->fd_cdir;
2064 	fdp->fd_cdir = NULL;
2065 	rdir = fdp->fd_rdir;
2066 	fdp->fd_rdir = NULL;
2067 	jdir = fdp->fd_jdir;
2068 	fdp->fd_jdir = NULL;
2069 	FILEDESC_XUNLOCK(fdp);
2070 
2071 	if (cdir != NULL)
2072 		vrele(cdir);
2073 	if (rdir != NULL)
2074 		vrele(rdir);
2075 	if (jdir != NULL)
2076 		vrele(jdir);
2077 
2078 	fddrop(fdp);
2079 }
2080 
2081 /*
2082  * For setugid programs, we don't want to people to use that setugidness
2083  * to generate error messages which write to a file which otherwise would
2084  * otherwise be off-limits to the process.  We check for filesystems where
2085  * the vnode can change out from under us after execve (like [lin]procfs).
2086  *
2087  * Since setugidsafety calls this only for fd 0, 1 and 2, this check is
2088  * sufficient.  We also don't check for setugidness since we know we are.
2089  */
2090 static int
2091 is_unsafe(struct file *fp)
2092 {
2093 	if (fp->f_type == DTYPE_VNODE) {
2094 		struct vnode *vp = fp->f_vnode;
2095 
2096 		if ((vp->v_vflag & VV_PROCDEP) != 0)
2097 			return (1);
2098 	}
2099 	return (0);
2100 }
2101 
2102 /*
2103  * Make this setguid thing safe, if at all possible.
2104  */
2105 void
2106 setugidsafety(struct thread *td)
2107 {
2108 	struct filedesc *fdp;
2109 	struct file *fp;
2110 	int i;
2111 
2112 	/* Certain daemons might not have file descriptors. */
2113 	fdp = td->td_proc->p_fd;
2114 	if (fdp == NULL)
2115 		return;
2116 
2117 	/*
2118 	 * Note: fdp->fd_ofiles may be reallocated out from under us while
2119 	 * we are blocked in a close.  Be careful!
2120 	 */
2121 	FILEDESC_XLOCK(fdp);
2122 	for (i = 0; i <= fdp->fd_lastfile; i++) {
2123 		if (i > 2)
2124 			break;
2125 		fp = fdp->fd_ofiles[i].fde_file;
2126 		if (fp != NULL && is_unsafe(fp)) {
2127 			knote_fdclose(td, i);
2128 			/*
2129 			 * NULL-out descriptor prior to close to avoid
2130 			 * a race while close blocks.
2131 			 */
2132 			fdfree(fdp, i);
2133 			FILEDESC_XUNLOCK(fdp);
2134 			(void) closef(fp, td);
2135 			FILEDESC_XLOCK(fdp);
2136 		}
2137 	}
2138 	FILEDESC_XUNLOCK(fdp);
2139 }
2140 
2141 /*
2142  * If a specific file object occupies a specific file descriptor, close the
2143  * file descriptor entry and drop a reference on the file object.  This is a
2144  * convenience function to handle a subsequent error in a function that calls
2145  * falloc() that handles the race that another thread might have closed the
2146  * file descriptor out from under the thread creating the file object.
2147  */
2148 void
2149 fdclose(struct filedesc *fdp, struct file *fp, int idx, struct thread *td)
2150 {
2151 
2152 	FILEDESC_XLOCK(fdp);
2153 	if (fdp->fd_ofiles[idx].fde_file == fp) {
2154 		fdfree(fdp, idx);
2155 		FILEDESC_XUNLOCK(fdp);
2156 		fdrop(fp, td);
2157 	} else
2158 		FILEDESC_XUNLOCK(fdp);
2159 }
2160 
2161 /*
2162  * Close any files on exec?
2163  */
2164 void
2165 fdcloseexec(struct thread *td)
2166 {
2167 	struct filedesc *fdp;
2168 	struct filedescent *fde;
2169 	struct file *fp;
2170 	int i;
2171 
2172 	/* Certain daemons might not have file descriptors. */
2173 	fdp = td->td_proc->p_fd;
2174 	if (fdp == NULL)
2175 		return;
2176 
2177 	/*
2178 	 * We cannot cache fd_ofiles since operations
2179 	 * may block and rip them out from under us.
2180 	 */
2181 	FILEDESC_XLOCK(fdp);
2182 	for (i = 0; i <= fdp->fd_lastfile; i++) {
2183 		fde = &fdp->fd_ofiles[i];
2184 		fp = fde->fde_file;
2185 		if (fp != NULL && (fp->f_type == DTYPE_MQUEUE ||
2186 		    (fde->fde_flags & UF_EXCLOSE))) {
2187 			fdfree(fdp, i);
2188 			(void) closefp(fdp, i, fp, td, 0);
2189 			/* closefp() drops the FILEDESC lock. */
2190 			FILEDESC_XLOCK(fdp);
2191 		}
2192 	}
2193 	FILEDESC_XUNLOCK(fdp);
2194 }
2195 
2196 /*
2197  * It is unsafe for set[ug]id processes to be started with file
2198  * descriptors 0..2 closed, as these descriptors are given implicit
2199  * significance in the Standard C library.  fdcheckstd() will create a
2200  * descriptor referencing /dev/null for each of stdin, stdout, and
2201  * stderr that is not already open.
2202  */
2203 int
2204 fdcheckstd(struct thread *td)
2205 {
2206 	struct filedesc *fdp;
2207 	register_t retval, save;
2208 	int i, error, devnull;
2209 
2210 	fdp = td->td_proc->p_fd;
2211 	if (fdp == NULL)
2212 		return (0);
2213 	KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2214 	devnull = -1;
2215 	error = 0;
2216 	for (i = 0; i < 3; i++) {
2217 		if (fdp->fd_ofiles[i].fde_file != NULL)
2218 			continue;
2219 		if (devnull < 0) {
2220 			save = td->td_retval[0];
2221 			error = kern_open(td, "/dev/null", UIO_SYSSPACE,
2222 			    O_RDWR, 0);
2223 			devnull = td->td_retval[0];
2224 			td->td_retval[0] = save;
2225 			if (error)
2226 				break;
2227 			KASSERT(devnull == i, ("oof, we didn't get our fd"));
2228 		} else {
2229 			error = do_dup(td, DUP_FIXED, devnull, i, &retval);
2230 			if (error != 0)
2231 				break;
2232 		}
2233 	}
2234 	return (error);
2235 }
2236 
2237 /*
2238  * Internal form of close.  Decrement reference count on file structure.
2239  * Note: td may be NULL when closing a file that was being passed in a
2240  * message.
2241  *
2242  * XXXRW: Giant is not required for the caller, but often will be held; this
2243  * makes it moderately likely the Giant will be recursed in the VFS case.
2244  */
2245 int
2246 closef(struct file *fp, struct thread *td)
2247 {
2248 	struct vnode *vp;
2249 	struct flock lf;
2250 	struct filedesc_to_leader *fdtol;
2251 	struct filedesc *fdp;
2252 
2253 	/*
2254 	 * POSIX record locking dictates that any close releases ALL
2255 	 * locks owned by this process.  This is handled by setting
2256 	 * a flag in the unlock to free ONLY locks obeying POSIX
2257 	 * semantics, and not to free BSD-style file locks.
2258 	 * If the descriptor was in a message, POSIX-style locks
2259 	 * aren't passed with the descriptor, and the thread pointer
2260 	 * will be NULL.  Callers should be careful only to pass a
2261 	 * NULL thread pointer when there really is no owning
2262 	 * context that might have locks, or the locks will be
2263 	 * leaked.
2264 	 */
2265 	if (fp->f_type == DTYPE_VNODE && td != NULL) {
2266 		vp = fp->f_vnode;
2267 		if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
2268 			lf.l_whence = SEEK_SET;
2269 			lf.l_start = 0;
2270 			lf.l_len = 0;
2271 			lf.l_type = F_UNLCK;
2272 			(void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader,
2273 			    F_UNLCK, &lf, F_POSIX);
2274 		}
2275 		fdtol = td->td_proc->p_fdtol;
2276 		if (fdtol != NULL) {
2277 			/*
2278 			 * Handle special case where file descriptor table is
2279 			 * shared between multiple process leaders.
2280 			 */
2281 			fdp = td->td_proc->p_fd;
2282 			FILEDESC_XLOCK(fdp);
2283 			for (fdtol = fdtol->fdl_next;
2284 			     fdtol != td->td_proc->p_fdtol;
2285 			     fdtol = fdtol->fdl_next) {
2286 				if ((fdtol->fdl_leader->p_flag &
2287 				     P_ADVLOCK) == 0)
2288 					continue;
2289 				fdtol->fdl_holdcount++;
2290 				FILEDESC_XUNLOCK(fdp);
2291 				lf.l_whence = SEEK_SET;
2292 				lf.l_start = 0;
2293 				lf.l_len = 0;
2294 				lf.l_type = F_UNLCK;
2295 				vp = fp->f_vnode;
2296 				(void) VOP_ADVLOCK(vp,
2297 				    (caddr_t)fdtol->fdl_leader, F_UNLCK, &lf,
2298 				    F_POSIX);
2299 				FILEDESC_XLOCK(fdp);
2300 				fdtol->fdl_holdcount--;
2301 				if (fdtol->fdl_holdcount == 0 &&
2302 				    fdtol->fdl_wakeup != 0) {
2303 					fdtol->fdl_wakeup = 0;
2304 					wakeup(fdtol);
2305 				}
2306 			}
2307 			FILEDESC_XUNLOCK(fdp);
2308 		}
2309 	}
2310 	return (fdrop(fp, td));
2311 }
2312 
2313 /*
2314  * Initialize the file pointer with the specified properties.
2315  *
2316  * The ops are set with release semantics to be certain that the flags, type,
2317  * and data are visible when ops is.  This is to prevent ops methods from being
2318  * called with bad data.
2319  */
2320 void
2321 finit(struct file *fp, u_int flag, short type, void *data, struct fileops *ops)
2322 {
2323 	fp->f_data = data;
2324 	fp->f_flag = flag;
2325 	fp->f_type = type;
2326 	atomic_store_rel_ptr((volatile uintptr_t *)&fp->f_ops, (uintptr_t)ops);
2327 }
2328 
2329 int
2330 fget_unlocked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
2331     int needfcntl, struct file **fpp, cap_rights_t *haverightsp)
2332 {
2333 	struct file *fp;
2334 	u_int count;
2335 #ifdef CAPABILITIES
2336 	cap_rights_t haverights;
2337 	int error;
2338 #endif
2339 
2340 	/*
2341 	 * Avoid reads reordering and then a first access to the
2342 	 * fdp->fd_ofiles table which could result in OOB operation.
2343 	 */
2344 	if (fd < 0 || fd >= atomic_load_acq_int(&fdp->fd_nfiles))
2345 		return (EBADF);
2346 	/*
2347 	 * Fetch the descriptor locklessly.  We avoid fdrop() races by
2348 	 * never raising a refcount above 0.  To accomplish this we have
2349 	 * to use a cmpset loop rather than an atomic_add.  The descriptor
2350 	 * must be re-verified once we acquire a reference to be certain
2351 	 * that the identity is still correct and we did not lose a race
2352 	 * due to preemption.
2353 	 */
2354 	for (;;) {
2355 		fp = fdp->fd_ofiles[fd].fde_file;
2356 		if (fp == NULL)
2357 			return (EBADF);
2358 #ifdef CAPABILITIES
2359 		haverights = *cap_rights(fdp, fd);
2360 		if (needrightsp != NULL) {
2361 			error = cap_check(&haverights, needrightsp);
2362 			if (error != 0)
2363 				return (error);
2364 			if (cap_rights_is_set(needrightsp, CAP_FCNTL)) {
2365 				error = cap_fcntl_check(fdp, fd, needfcntl);
2366 				if (error != 0)
2367 					return (error);
2368 			}
2369 		}
2370 #endif
2371 		count = fp->f_count;
2372 		if (count == 0)
2373 			continue;
2374 		/*
2375 		 * Use an acquire barrier to prevent caching of fd_ofiles
2376 		 * so it is refreshed for verification.
2377 		 */
2378 		if (atomic_cmpset_acq_int(&fp->f_count, count, count + 1) != 1)
2379 			continue;
2380 		if (fp == fdp->fd_ofiles[fd].fde_file)
2381 			break;
2382 		fdrop(fp, curthread);
2383 	}
2384 	*fpp = fp;
2385 	if (haverightsp != NULL) {
2386 #ifdef CAPABILITIES
2387 		*haverightsp = haverights;
2388 #else
2389 		CAP_ALL(haverightsp);
2390 #endif
2391 	}
2392 	return (0);
2393 }
2394 
2395 /*
2396  * Extract the file pointer associated with the specified descriptor for the
2397  * current user process.
2398  *
2399  * If the descriptor doesn't exist or doesn't match 'flags', EBADF is
2400  * returned.
2401  *
2402  * File's rights will be checked against the capability rights mask.
2403  *
2404  * If an error occured the non-zero error is returned and *fpp is set to
2405  * NULL.  Otherwise *fpp is held and set and zero is returned.  Caller is
2406  * responsible for fdrop().
2407  */
2408 static __inline int
2409 _fget(struct thread *td, int fd, struct file **fpp, int flags,
2410     cap_rights_t *needrightsp, u_char *maxprotp)
2411 {
2412 	struct filedesc *fdp;
2413 	struct file *fp;
2414 	cap_rights_t haverights, needrights;
2415 	int error;
2416 
2417 	*fpp = NULL;
2418 	if (td == NULL || (fdp = td->td_proc->p_fd) == NULL)
2419 		return (EBADF);
2420 	if (needrightsp != NULL)
2421 		needrights = *needrightsp;
2422 	else
2423 		cap_rights_init(&needrights);
2424 	if (maxprotp != NULL)
2425 		cap_rights_set(&needrights, CAP_MMAP);
2426 	error = fget_unlocked(fdp, fd, &needrights, 0, &fp, &haverights);
2427 	if (error != 0)
2428 		return (error);
2429 	if (fp->f_ops == &badfileops) {
2430 		fdrop(fp, td);
2431 		return (EBADF);
2432 	}
2433 
2434 #ifdef CAPABILITIES
2435 	/*
2436 	 * If requested, convert capability rights to access flags.
2437 	 */
2438 	if (maxprotp != NULL)
2439 		*maxprotp = cap_rights_to_vmprot(&haverights);
2440 #else /* !CAPABILITIES */
2441 	if (maxprotp != NULL)
2442 		*maxprotp = VM_PROT_ALL;
2443 #endif /* CAPABILITIES */
2444 
2445 	/*
2446 	 * FREAD and FWRITE failure return EBADF as per POSIX.
2447 	 */
2448 	error = 0;
2449 	switch (flags) {
2450 	case FREAD:
2451 	case FWRITE:
2452 		if ((fp->f_flag & flags) == 0)
2453 			error = EBADF;
2454 		break;
2455 	case FEXEC:
2456 	    	if ((fp->f_flag & (FREAD | FEXEC)) == 0 ||
2457 		    ((fp->f_flag & FWRITE) != 0))
2458 			error = EBADF;
2459 		break;
2460 	case 0:
2461 		break;
2462 	default:
2463 		KASSERT(0, ("wrong flags"));
2464 	}
2465 
2466 	if (error != 0) {
2467 		fdrop(fp, td);
2468 		return (error);
2469 	}
2470 
2471 	*fpp = fp;
2472 	return (0);
2473 }
2474 
2475 int
2476 fget(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
2477 {
2478 
2479 	return(_fget(td, fd, fpp, 0, rightsp, NULL));
2480 }
2481 
2482 int
2483 fget_mmap(struct thread *td, int fd, cap_rights_t *rightsp, u_char *maxprotp,
2484     struct file **fpp)
2485 {
2486 
2487 	return (_fget(td, fd, fpp, 0, rightsp, maxprotp));
2488 }
2489 
2490 int
2491 fget_read(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
2492 {
2493 
2494 	return(_fget(td, fd, fpp, FREAD, rightsp, NULL));
2495 }
2496 
2497 int
2498 fget_write(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
2499 {
2500 
2501 	return (_fget(td, fd, fpp, FWRITE, rightsp, NULL));
2502 }
2503 
2504 /*
2505  * Like fget() but loads the underlying vnode, or returns an error if the
2506  * descriptor does not represent a vnode.  Note that pipes use vnodes but
2507  * never have VM objects.  The returned vnode will be vref()'d.
2508  *
2509  * XXX: what about the unused flags ?
2510  */
2511 static __inline int
2512 _fgetvp(struct thread *td, int fd, int flags, cap_rights_t *needrightsp,
2513     struct vnode **vpp)
2514 {
2515 	struct file *fp;
2516 	int error;
2517 
2518 	*vpp = NULL;
2519 	error = _fget(td, fd, &fp, flags, needrightsp, NULL);
2520 	if (error != 0)
2521 		return (error);
2522 	if (fp->f_vnode == NULL) {
2523 		error = EINVAL;
2524 	} else {
2525 		*vpp = fp->f_vnode;
2526 		vref(*vpp);
2527 	}
2528 	fdrop(fp, td);
2529 
2530 	return (error);
2531 }
2532 
2533 int
2534 fgetvp(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
2535 {
2536 
2537 	return (_fgetvp(td, fd, 0, rightsp, vpp));
2538 }
2539 
2540 int
2541 fgetvp_rights(struct thread *td, int fd, cap_rights_t *needrightsp,
2542     struct filecaps *havecaps, struct vnode **vpp)
2543 {
2544 	struct filedesc *fdp;
2545 	struct file *fp;
2546 #ifdef CAPABILITIES
2547 	int error;
2548 #endif
2549 
2550 	if (td == NULL || (fdp = td->td_proc->p_fd) == NULL)
2551 		return (EBADF);
2552 
2553 	fp = fget_locked(fdp, fd);
2554 	if (fp == NULL || fp->f_ops == &badfileops)
2555 		return (EBADF);
2556 
2557 #ifdef CAPABILITIES
2558 	if (needrightsp != NULL) {
2559 		error = cap_check(cap_rights(fdp, fd), needrightsp);
2560 		if (error != 0)
2561 			return (error);
2562 	}
2563 #endif
2564 
2565 	if (fp->f_vnode == NULL)
2566 		return (EINVAL);
2567 
2568 	*vpp = fp->f_vnode;
2569 	vref(*vpp);
2570 	filecaps_copy(&fdp->fd_ofiles[fd].fde_caps, havecaps);
2571 
2572 	return (0);
2573 }
2574 
2575 int
2576 fgetvp_read(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
2577 {
2578 
2579 	return (_fgetvp(td, fd, FREAD, rightsp, vpp));
2580 }
2581 
2582 int
2583 fgetvp_exec(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
2584 {
2585 
2586 	return (_fgetvp(td, fd, FEXEC, rightsp, vpp));
2587 }
2588 
2589 #ifdef notyet
2590 int
2591 fgetvp_write(struct thread *td, int fd, cap_rights_t *rightsp,
2592     struct vnode **vpp)
2593 {
2594 
2595 	return (_fgetvp(td, fd, FWRITE, rightsp, vpp));
2596 }
2597 #endif
2598 
2599 /*
2600  * Like fget() but loads the underlying socket, or returns an error if the
2601  * descriptor does not represent a socket.
2602  *
2603  * We bump the ref count on the returned socket.  XXX Also obtain the SX lock
2604  * in the future.
2605  *
2606  * Note: fgetsock() and fputsock() are deprecated, as consumers should rely
2607  * on their file descriptor reference to prevent the socket from being free'd
2608  * during use.
2609  */
2610 int
2611 fgetsock(struct thread *td, int fd, cap_rights_t *rightsp, struct socket **spp,
2612     u_int *fflagp)
2613 {
2614 	struct file *fp;
2615 	int error;
2616 
2617 	*spp = NULL;
2618 	if (fflagp != NULL)
2619 		*fflagp = 0;
2620 	if ((error = _fget(td, fd, &fp, 0, rightsp, NULL)) != 0)
2621 		return (error);
2622 	if (fp->f_type != DTYPE_SOCKET) {
2623 		error = ENOTSOCK;
2624 	} else {
2625 		*spp = fp->f_data;
2626 		if (fflagp)
2627 			*fflagp = fp->f_flag;
2628 		SOCK_LOCK(*spp);
2629 		soref(*spp);
2630 		SOCK_UNLOCK(*spp);
2631 	}
2632 	fdrop(fp, td);
2633 
2634 	return (error);
2635 }
2636 
2637 /*
2638  * Drop the reference count on the socket and XXX release the SX lock in the
2639  * future.  The last reference closes the socket.
2640  *
2641  * Note: fputsock() is deprecated, see comment for fgetsock().
2642  */
2643 void
2644 fputsock(struct socket *so)
2645 {
2646 
2647 	ACCEPT_LOCK();
2648 	SOCK_LOCK(so);
2649 	CURVNET_SET(so->so_vnet);
2650 	sorele(so);
2651 	CURVNET_RESTORE();
2652 }
2653 
2654 /*
2655  * Handle the last reference to a file being closed.
2656  */
2657 int
2658 _fdrop(struct file *fp, struct thread *td)
2659 {
2660 	int error;
2661 
2662 	error = 0;
2663 	if (fp->f_count != 0)
2664 		panic("fdrop: count %d", fp->f_count);
2665 	if (fp->f_ops != &badfileops)
2666 		error = fo_close(fp, td);
2667 	atomic_subtract_int(&openfiles, 1);
2668 	crfree(fp->f_cred);
2669 	free(fp->f_advice, M_FADVISE);
2670 	uma_zfree(file_zone, fp);
2671 
2672 	return (error);
2673 }
2674 
2675 /*
2676  * Apply an advisory lock on a file descriptor.
2677  *
2678  * Just attempt to get a record lock of the requested type on the entire file
2679  * (l_whence = SEEK_SET, l_start = 0, l_len = 0).
2680  */
2681 #ifndef _SYS_SYSPROTO_H_
2682 struct flock_args {
2683 	int	fd;
2684 	int	how;
2685 };
2686 #endif
2687 /* ARGSUSED */
2688 int
2689 sys_flock(struct thread *td, struct flock_args *uap)
2690 {
2691 	struct file *fp;
2692 	struct vnode *vp;
2693 	struct flock lf;
2694 	cap_rights_t rights;
2695 	int error;
2696 
2697 	error = fget(td, uap->fd, cap_rights_init(&rights, CAP_FLOCK), &fp);
2698 	if (error != 0)
2699 		return (error);
2700 	if (fp->f_type != DTYPE_VNODE) {
2701 		fdrop(fp, td);
2702 		return (EOPNOTSUPP);
2703 	}
2704 
2705 	vp = fp->f_vnode;
2706 	lf.l_whence = SEEK_SET;
2707 	lf.l_start = 0;
2708 	lf.l_len = 0;
2709 	if (uap->how & LOCK_UN) {
2710 		lf.l_type = F_UNLCK;
2711 		atomic_clear_int(&fp->f_flag, FHASLOCK);
2712 		error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK);
2713 		goto done2;
2714 	}
2715 	if (uap->how & LOCK_EX)
2716 		lf.l_type = F_WRLCK;
2717 	else if (uap->how & LOCK_SH)
2718 		lf.l_type = F_RDLCK;
2719 	else {
2720 		error = EBADF;
2721 		goto done2;
2722 	}
2723 	atomic_set_int(&fp->f_flag, FHASLOCK);
2724 	error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf,
2725 	    (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT);
2726 done2:
2727 	fdrop(fp, td);
2728 	return (error);
2729 }
2730 /*
2731  * Duplicate the specified descriptor to a free descriptor.
2732  */
2733 int
2734 dupfdopen(struct thread *td, struct filedesc *fdp, int dfd, int mode,
2735     int openerror, int *indxp)
2736 {
2737 	struct file *fp;
2738 	int error, indx;
2739 
2740 	KASSERT(openerror == ENODEV || openerror == ENXIO,
2741 	    ("unexpected error %d in %s", openerror, __func__));
2742 
2743 	/*
2744 	 * If the to-be-dup'd fd number is greater than the allowed number
2745 	 * of file descriptors, or the fd to be dup'd has already been
2746 	 * closed, then reject.
2747 	 */
2748 	FILEDESC_XLOCK(fdp);
2749 	if ((fp = fget_locked(fdp, dfd)) == NULL) {
2750 		FILEDESC_XUNLOCK(fdp);
2751 		return (EBADF);
2752 	}
2753 
2754 	error = fdalloc(td, 0, &indx);
2755 	if (error != 0) {
2756 		FILEDESC_XUNLOCK(fdp);
2757 		return (error);
2758 	}
2759 
2760 	/*
2761 	 * There are two cases of interest here.
2762 	 *
2763 	 * For ENODEV simply dup (dfd) to file descriptor (indx) and return.
2764 	 *
2765 	 * For ENXIO steal away the file structure from (dfd) and store it in
2766 	 * (indx).  (dfd) is effectively closed by this operation.
2767 	 */
2768 	switch (openerror) {
2769 	case ENODEV:
2770 		/*
2771 		 * Check that the mode the file is being opened for is a
2772 		 * subset of the mode of the existing descriptor.
2773 		 */
2774 		if (((mode & (FREAD|FWRITE)) | fp->f_flag) != fp->f_flag) {
2775 			fdunused(fdp, indx);
2776 			FILEDESC_XUNLOCK(fdp);
2777 			return (EACCES);
2778 		}
2779 		fhold(fp);
2780 		fdp->fd_ofiles[indx] = fdp->fd_ofiles[dfd];
2781 		filecaps_copy(&fdp->fd_ofiles[dfd].fde_caps,
2782 		    &fdp->fd_ofiles[indx].fde_caps);
2783 		break;
2784 	case ENXIO:
2785 		/*
2786 		 * Steal away the file pointer from dfd and stuff it into indx.
2787 		 */
2788 		fdp->fd_ofiles[indx] = fdp->fd_ofiles[dfd];
2789 		bzero(&fdp->fd_ofiles[dfd], sizeof(fdp->fd_ofiles[dfd]));
2790 		fdunused(fdp, dfd);
2791 		break;
2792 	}
2793 	FILEDESC_XUNLOCK(fdp);
2794 	*indxp = indx;
2795 	return (0);
2796 }
2797 
2798 /*
2799  * Scan all active processes and prisons to see if any of them have a current
2800  * or root directory of `olddp'. If so, replace them with the new mount point.
2801  */
2802 void
2803 mountcheckdirs(struct vnode *olddp, struct vnode *newdp)
2804 {
2805 	struct filedesc *fdp;
2806 	struct prison *pr;
2807 	struct proc *p;
2808 	int nrele;
2809 
2810 	if (vrefcnt(olddp) == 1)
2811 		return;
2812 	nrele = 0;
2813 	sx_slock(&allproc_lock);
2814 	FOREACH_PROC_IN_SYSTEM(p) {
2815 		fdp = fdhold(p);
2816 		if (fdp == NULL)
2817 			continue;
2818 		FILEDESC_XLOCK(fdp);
2819 		if (fdp->fd_cdir == olddp) {
2820 			vref(newdp);
2821 			fdp->fd_cdir = newdp;
2822 			nrele++;
2823 		}
2824 		if (fdp->fd_rdir == olddp) {
2825 			vref(newdp);
2826 			fdp->fd_rdir = newdp;
2827 			nrele++;
2828 		}
2829 		if (fdp->fd_jdir == olddp) {
2830 			vref(newdp);
2831 			fdp->fd_jdir = newdp;
2832 			nrele++;
2833 		}
2834 		FILEDESC_XUNLOCK(fdp);
2835 		fddrop(fdp);
2836 	}
2837 	sx_sunlock(&allproc_lock);
2838 	if (rootvnode == olddp) {
2839 		vref(newdp);
2840 		rootvnode = newdp;
2841 		nrele++;
2842 	}
2843 	mtx_lock(&prison0.pr_mtx);
2844 	if (prison0.pr_root == olddp) {
2845 		vref(newdp);
2846 		prison0.pr_root = newdp;
2847 		nrele++;
2848 	}
2849 	mtx_unlock(&prison0.pr_mtx);
2850 	sx_slock(&allprison_lock);
2851 	TAILQ_FOREACH(pr, &allprison, pr_list) {
2852 		mtx_lock(&pr->pr_mtx);
2853 		if (pr->pr_root == olddp) {
2854 			vref(newdp);
2855 			pr->pr_root = newdp;
2856 			nrele++;
2857 		}
2858 		mtx_unlock(&pr->pr_mtx);
2859 	}
2860 	sx_sunlock(&allprison_lock);
2861 	while (nrele--)
2862 		vrele(olddp);
2863 }
2864 
2865 struct filedesc_to_leader *
2866 filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader)
2867 {
2868 	struct filedesc_to_leader *fdtol;
2869 
2870 	fdtol = malloc(sizeof(struct filedesc_to_leader),
2871 	       M_FILEDESC_TO_LEADER,
2872 	       M_WAITOK);
2873 	fdtol->fdl_refcount = 1;
2874 	fdtol->fdl_holdcount = 0;
2875 	fdtol->fdl_wakeup = 0;
2876 	fdtol->fdl_leader = leader;
2877 	if (old != NULL) {
2878 		FILEDESC_XLOCK(fdp);
2879 		fdtol->fdl_next = old->fdl_next;
2880 		fdtol->fdl_prev = old;
2881 		old->fdl_next = fdtol;
2882 		fdtol->fdl_next->fdl_prev = fdtol;
2883 		FILEDESC_XUNLOCK(fdp);
2884 	} else {
2885 		fdtol->fdl_next = fdtol;
2886 		fdtol->fdl_prev = fdtol;
2887 	}
2888 	return (fdtol);
2889 }
2890 
2891 /*
2892  * Get file structures globally.
2893  */
2894 static int
2895 sysctl_kern_file(SYSCTL_HANDLER_ARGS)
2896 {
2897 	struct xfile xf;
2898 	struct filedesc *fdp;
2899 	struct file *fp;
2900 	struct proc *p;
2901 	int error, n;
2902 
2903 	error = sysctl_wire_old_buffer(req, 0);
2904 	if (error != 0)
2905 		return (error);
2906 	if (req->oldptr == NULL) {
2907 		n = 0;
2908 		sx_slock(&allproc_lock);
2909 		FOREACH_PROC_IN_SYSTEM(p) {
2910 			if (p->p_state == PRS_NEW)
2911 				continue;
2912 			fdp = fdhold(p);
2913 			if (fdp == NULL)
2914 				continue;
2915 			/* overestimates sparse tables. */
2916 			if (fdp->fd_lastfile > 0)
2917 				n += fdp->fd_lastfile;
2918 			fddrop(fdp);
2919 		}
2920 		sx_sunlock(&allproc_lock);
2921 		return (SYSCTL_OUT(req, 0, n * sizeof(xf)));
2922 	}
2923 	error = 0;
2924 	bzero(&xf, sizeof(xf));
2925 	xf.xf_size = sizeof(xf);
2926 	sx_slock(&allproc_lock);
2927 	FOREACH_PROC_IN_SYSTEM(p) {
2928 		PROC_LOCK(p);
2929 		if (p->p_state == PRS_NEW) {
2930 			PROC_UNLOCK(p);
2931 			continue;
2932 		}
2933 		if (p_cansee(req->td, p) != 0) {
2934 			PROC_UNLOCK(p);
2935 			continue;
2936 		}
2937 		xf.xf_pid = p->p_pid;
2938 		xf.xf_uid = p->p_ucred->cr_uid;
2939 		PROC_UNLOCK(p);
2940 		fdp = fdhold(p);
2941 		if (fdp == NULL)
2942 			continue;
2943 		FILEDESC_SLOCK(fdp);
2944 		for (n = 0; fdp->fd_refcnt > 0 && n < fdp->fd_nfiles; ++n) {
2945 			if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
2946 				continue;
2947 			xf.xf_fd = n;
2948 			xf.xf_file = fp;
2949 			xf.xf_data = fp->f_data;
2950 			xf.xf_vnode = fp->f_vnode;
2951 			xf.xf_type = fp->f_type;
2952 			xf.xf_count = fp->f_count;
2953 			xf.xf_msgcount = 0;
2954 			xf.xf_offset = foffset_get(fp);
2955 			xf.xf_flag = fp->f_flag;
2956 			error = SYSCTL_OUT(req, &xf, sizeof(xf));
2957 			if (error)
2958 				break;
2959 		}
2960 		FILEDESC_SUNLOCK(fdp);
2961 		fddrop(fdp);
2962 		if (error)
2963 			break;
2964 	}
2965 	sx_sunlock(&allproc_lock);
2966 	return (error);
2967 }
2968 
2969 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD,
2970     0, 0, sysctl_kern_file, "S,xfile", "Entire file table");
2971 
2972 #ifdef KINFO_OFILE_SIZE
2973 CTASSERT(sizeof(struct kinfo_ofile) == KINFO_OFILE_SIZE);
2974 #endif
2975 
2976 #ifdef COMPAT_FREEBSD7
2977 static int
2978 export_vnode_for_osysctl(struct vnode *vp, int type,
2979     struct kinfo_ofile *kif, struct filedesc *fdp, struct sysctl_req *req)
2980 {
2981 	int error;
2982 	char *fullpath, *freepath;
2983 
2984 	bzero(kif, sizeof(*kif));
2985 	kif->kf_structsize = sizeof(*kif);
2986 
2987 	vref(vp);
2988 	kif->kf_fd = type;
2989 	kif->kf_type = KF_TYPE_VNODE;
2990 	/* This function only handles directories. */
2991 	if (vp->v_type != VDIR) {
2992 		vrele(vp);
2993 		return (ENOTDIR);
2994 	}
2995 	kif->kf_vnode_type = KF_VTYPE_VDIR;
2996 
2997 	/*
2998 	 * This is not a true file descriptor, so we set a bogus refcount
2999 	 * and offset to indicate these fields should be ignored.
3000 	 */
3001 	kif->kf_ref_count = -1;
3002 	kif->kf_offset = -1;
3003 
3004 	freepath = NULL;
3005 	fullpath = "-";
3006 	FILEDESC_SUNLOCK(fdp);
3007 	vn_fullpath(curthread, vp, &fullpath, &freepath);
3008 	vrele(vp);
3009 	strlcpy(kif->kf_path, fullpath, sizeof(kif->kf_path));
3010 	if (freepath != NULL)
3011 		free(freepath, M_TEMP);
3012 	error = SYSCTL_OUT(req, kif, sizeof(*kif));
3013 	FILEDESC_SLOCK(fdp);
3014 	return (error);
3015 }
3016 
3017 /*
3018  * Get per-process file descriptors for use by procstat(1), et al.
3019  */
3020 static int
3021 sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS)
3022 {
3023 	char *fullpath, *freepath;
3024 	struct kinfo_ofile *kif;
3025 	struct filedesc *fdp;
3026 	int error, i, *name;
3027 	struct shmfd *shmfd;
3028 	struct socket *so;
3029 	struct vnode *vp;
3030 	struct ksem *ks;
3031 	struct file *fp;
3032 	struct proc *p;
3033 	struct tty *tp;
3034 
3035 	name = (int *)arg1;
3036 	error = pget((pid_t)name[0], PGET_CANDEBUG, &p);
3037 	if (error != 0)
3038 		return (error);
3039 	fdp = fdhold(p);
3040 	PROC_UNLOCK(p);
3041 	if (fdp == NULL)
3042 		return (ENOENT);
3043 	kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK);
3044 	FILEDESC_SLOCK(fdp);
3045 	if (fdp->fd_cdir != NULL)
3046 		export_vnode_for_osysctl(fdp->fd_cdir, KF_FD_TYPE_CWD, kif,
3047 				fdp, req);
3048 	if (fdp->fd_rdir != NULL)
3049 		export_vnode_for_osysctl(fdp->fd_rdir, KF_FD_TYPE_ROOT, kif,
3050 				fdp, req);
3051 	if (fdp->fd_jdir != NULL)
3052 		export_vnode_for_osysctl(fdp->fd_jdir, KF_FD_TYPE_JAIL, kif,
3053 				fdp, req);
3054 	for (i = 0; i < fdp->fd_nfiles; i++) {
3055 		if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
3056 			continue;
3057 		bzero(kif, sizeof(*kif));
3058 		kif->kf_structsize = sizeof(*kif);
3059 		ks = NULL;
3060 		vp = NULL;
3061 		so = NULL;
3062 		tp = NULL;
3063 		shmfd = NULL;
3064 		kif->kf_fd = i;
3065 
3066 		switch (fp->f_type) {
3067 		case DTYPE_VNODE:
3068 			kif->kf_type = KF_TYPE_VNODE;
3069 			vp = fp->f_vnode;
3070 			break;
3071 
3072 		case DTYPE_SOCKET:
3073 			kif->kf_type = KF_TYPE_SOCKET;
3074 			so = fp->f_data;
3075 			break;
3076 
3077 		case DTYPE_PIPE:
3078 			kif->kf_type = KF_TYPE_PIPE;
3079 			break;
3080 
3081 		case DTYPE_FIFO:
3082 			kif->kf_type = KF_TYPE_FIFO;
3083 			vp = fp->f_vnode;
3084 			break;
3085 
3086 		case DTYPE_KQUEUE:
3087 			kif->kf_type = KF_TYPE_KQUEUE;
3088 			break;
3089 
3090 		case DTYPE_CRYPTO:
3091 			kif->kf_type = KF_TYPE_CRYPTO;
3092 			break;
3093 
3094 		case DTYPE_MQUEUE:
3095 			kif->kf_type = KF_TYPE_MQUEUE;
3096 			break;
3097 
3098 		case DTYPE_SHM:
3099 			kif->kf_type = KF_TYPE_SHM;
3100 			shmfd = fp->f_data;
3101 			break;
3102 
3103 		case DTYPE_SEM:
3104 			kif->kf_type = KF_TYPE_SEM;
3105 			ks = fp->f_data;
3106 			break;
3107 
3108 		case DTYPE_PTS:
3109 			kif->kf_type = KF_TYPE_PTS;
3110 			tp = fp->f_data;
3111 			break;
3112 
3113 		case DTYPE_PROCDESC:
3114 			kif->kf_type = KF_TYPE_PROCDESC;
3115 			break;
3116 
3117 		default:
3118 			kif->kf_type = KF_TYPE_UNKNOWN;
3119 			break;
3120 		}
3121 		kif->kf_ref_count = fp->f_count;
3122 		if (fp->f_flag & FREAD)
3123 			kif->kf_flags |= KF_FLAG_READ;
3124 		if (fp->f_flag & FWRITE)
3125 			kif->kf_flags |= KF_FLAG_WRITE;
3126 		if (fp->f_flag & FAPPEND)
3127 			kif->kf_flags |= KF_FLAG_APPEND;
3128 		if (fp->f_flag & FASYNC)
3129 			kif->kf_flags |= KF_FLAG_ASYNC;
3130 		if (fp->f_flag & FFSYNC)
3131 			kif->kf_flags |= KF_FLAG_FSYNC;
3132 		if (fp->f_flag & FNONBLOCK)
3133 			kif->kf_flags |= KF_FLAG_NONBLOCK;
3134 		if (fp->f_flag & O_DIRECT)
3135 			kif->kf_flags |= KF_FLAG_DIRECT;
3136 		if (fp->f_flag & FHASLOCK)
3137 			kif->kf_flags |= KF_FLAG_HASLOCK;
3138 		kif->kf_offset = foffset_get(fp);
3139 		if (vp != NULL) {
3140 			vref(vp);
3141 			switch (vp->v_type) {
3142 			case VNON:
3143 				kif->kf_vnode_type = KF_VTYPE_VNON;
3144 				break;
3145 			case VREG:
3146 				kif->kf_vnode_type = KF_VTYPE_VREG;
3147 				break;
3148 			case VDIR:
3149 				kif->kf_vnode_type = KF_VTYPE_VDIR;
3150 				break;
3151 			case VBLK:
3152 				kif->kf_vnode_type = KF_VTYPE_VBLK;
3153 				break;
3154 			case VCHR:
3155 				kif->kf_vnode_type = KF_VTYPE_VCHR;
3156 				break;
3157 			case VLNK:
3158 				kif->kf_vnode_type = KF_VTYPE_VLNK;
3159 				break;
3160 			case VSOCK:
3161 				kif->kf_vnode_type = KF_VTYPE_VSOCK;
3162 				break;
3163 			case VFIFO:
3164 				kif->kf_vnode_type = KF_VTYPE_VFIFO;
3165 				break;
3166 			case VBAD:
3167 				kif->kf_vnode_type = KF_VTYPE_VBAD;
3168 				break;
3169 			default:
3170 				kif->kf_vnode_type = KF_VTYPE_UNKNOWN;
3171 				break;
3172 			}
3173 			/*
3174 			 * It is OK to drop the filedesc lock here as we will
3175 			 * re-validate and re-evaluate its properties when
3176 			 * the loop continues.
3177 			 */
3178 			freepath = NULL;
3179 			fullpath = "-";
3180 			FILEDESC_SUNLOCK(fdp);
3181 			vn_fullpath(curthread, vp, &fullpath, &freepath);
3182 			vrele(vp);
3183 			strlcpy(kif->kf_path, fullpath,
3184 			    sizeof(kif->kf_path));
3185 			if (freepath != NULL)
3186 				free(freepath, M_TEMP);
3187 			FILEDESC_SLOCK(fdp);
3188 		}
3189 		if (so != NULL) {
3190 			struct sockaddr *sa;
3191 
3192 			if (so->so_proto->pr_usrreqs->pru_sockaddr(so, &sa)
3193 			    == 0 && sa->sa_len <= sizeof(kif->kf_sa_local)) {
3194 				bcopy(sa, &kif->kf_sa_local, sa->sa_len);
3195 				free(sa, M_SONAME);
3196 			}
3197 			if (so->so_proto->pr_usrreqs->pru_peeraddr(so, &sa)
3198 			    == 0 && sa->sa_len <= sizeof(kif->kf_sa_peer)) {
3199 				bcopy(sa, &kif->kf_sa_peer, sa->sa_len);
3200 				free(sa, M_SONAME);
3201 			}
3202 			kif->kf_sock_domain =
3203 			    so->so_proto->pr_domain->dom_family;
3204 			kif->kf_sock_type = so->so_type;
3205 			kif->kf_sock_protocol = so->so_proto->pr_protocol;
3206 		}
3207 		if (tp != NULL) {
3208 			strlcpy(kif->kf_path, tty_devname(tp),
3209 			    sizeof(kif->kf_path));
3210 		}
3211 		if (shmfd != NULL)
3212 			shm_path(shmfd, kif->kf_path, sizeof(kif->kf_path));
3213 		if (ks != NULL && ksem_info != NULL)
3214 			ksem_info(ks, kif->kf_path, sizeof(kif->kf_path), NULL);
3215 		error = SYSCTL_OUT(req, kif, sizeof(*kif));
3216 		if (error)
3217 			break;
3218 	}
3219 	FILEDESC_SUNLOCK(fdp);
3220 	fddrop(fdp);
3221 	free(kif, M_TEMP);
3222 	return (0);
3223 }
3224 
3225 static SYSCTL_NODE(_kern_proc, KERN_PROC_OFILEDESC, ofiledesc, CTLFLAG_RD,
3226     sysctl_kern_proc_ofiledesc, "Process ofiledesc entries");
3227 #endif	/* COMPAT_FREEBSD7 */
3228 
3229 #ifdef KINFO_FILE_SIZE
3230 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
3231 #endif
3232 
3233 struct export_fd_buf {
3234 	struct filedesc		*fdp;
3235 	struct sbuf 		*sb;
3236 	ssize_t			remainder;
3237 	struct kinfo_file	kif;
3238 };
3239 
3240 static int
3241 export_fd_to_sb(void *data, int type, int fd, int fflags, int refcnt,
3242     int64_t offset, cap_rights_t *rightsp, struct export_fd_buf *efbuf)
3243 {
3244 	struct {
3245 		int	fflag;
3246 		int	kf_fflag;
3247 	} fflags_table[] = {
3248 		{ FAPPEND, KF_FLAG_APPEND },
3249 		{ FASYNC, KF_FLAG_ASYNC },
3250 		{ FFSYNC, KF_FLAG_FSYNC },
3251 		{ FHASLOCK, KF_FLAG_HASLOCK },
3252 		{ FNONBLOCK, KF_FLAG_NONBLOCK },
3253 		{ FREAD, KF_FLAG_READ },
3254 		{ FWRITE, KF_FLAG_WRITE },
3255 		{ O_CREAT, KF_FLAG_CREAT },
3256 		{ O_DIRECT, KF_FLAG_DIRECT },
3257 		{ O_EXCL, KF_FLAG_EXCL },
3258 		{ O_EXEC, KF_FLAG_EXEC },
3259 		{ O_EXLOCK, KF_FLAG_EXLOCK },
3260 		{ O_NOFOLLOW, KF_FLAG_NOFOLLOW },
3261 		{ O_SHLOCK, KF_FLAG_SHLOCK },
3262 		{ O_TRUNC, KF_FLAG_TRUNC }
3263 	};
3264 #define	NFFLAGS	(sizeof(fflags_table) / sizeof(*fflags_table))
3265 	struct kinfo_file *kif;
3266 	struct vnode *vp;
3267 	int error, locked;
3268 	unsigned int i;
3269 
3270 	if (efbuf->remainder == 0)
3271 		return (0);
3272 	kif = &efbuf->kif;
3273 	bzero(kif, sizeof(*kif));
3274 	locked = efbuf->fdp != NULL;
3275 	switch (type) {
3276 	case KF_TYPE_FIFO:
3277 	case KF_TYPE_VNODE:
3278 		if (locked) {
3279 			FILEDESC_SUNLOCK(efbuf->fdp);
3280 			locked = 0;
3281 		}
3282 		vp = (struct vnode *)data;
3283 		error = fill_vnode_info(vp, kif);
3284 		vrele(vp);
3285 		break;
3286 	case KF_TYPE_SOCKET:
3287 		error = fill_socket_info((struct socket *)data, kif);
3288 		break;
3289 	case KF_TYPE_PIPE:
3290 		error = fill_pipe_info((struct pipe *)data, kif);
3291 		break;
3292 	case KF_TYPE_PTS:
3293 		error = fill_pts_info((struct tty *)data, kif);
3294 		break;
3295 	case KF_TYPE_PROCDESC:
3296 		error = fill_procdesc_info((struct procdesc *)data, kif);
3297 		break;
3298 	case KF_TYPE_SEM:
3299 		error = fill_sem_info((struct file *)data, kif);
3300 		break;
3301 	case KF_TYPE_SHM:
3302 		error = fill_shm_info((struct file *)data, kif);
3303 		break;
3304 	default:
3305 		error = 0;
3306 	}
3307 	if (error == 0)
3308 		kif->kf_status |= KF_ATTR_VALID;
3309 
3310 	/*
3311 	 * Translate file access flags.
3312 	 */
3313 	for (i = 0; i < NFFLAGS; i++)
3314 		if (fflags & fflags_table[i].fflag)
3315 			kif->kf_flags |=  fflags_table[i].kf_fflag;
3316 	if (rightsp != NULL)
3317 		kif->kf_cap_rights = *rightsp;
3318 	else
3319 		cap_rights_init(&kif->kf_cap_rights);
3320 	kif->kf_fd = fd;
3321 	kif->kf_type = type;
3322 	kif->kf_ref_count = refcnt;
3323 	kif->kf_offset = offset;
3324 	/* Pack record size down */
3325 	kif->kf_structsize = offsetof(struct kinfo_file, kf_path) +
3326 	    strlen(kif->kf_path) + 1;
3327 	kif->kf_structsize = roundup(kif->kf_structsize, sizeof(uint64_t));
3328 	if (efbuf->remainder != -1) {
3329 		if (efbuf->remainder < kif->kf_structsize) {
3330 			/* Terminate export. */
3331 			efbuf->remainder = 0;
3332 			if (efbuf->fdp != NULL && !locked)
3333 				FILEDESC_SLOCK(efbuf->fdp);
3334 			return (0);
3335 		}
3336 		efbuf->remainder -= kif->kf_structsize;
3337 	}
3338 	if (locked)
3339 		FILEDESC_SUNLOCK(efbuf->fdp);
3340 	error = sbuf_bcat(efbuf->sb, kif, kif->kf_structsize);
3341 	if (efbuf->fdp != NULL)
3342 		FILEDESC_SLOCK(efbuf->fdp);
3343 	return (error);
3344 }
3345 
3346 /*
3347  * Store a process file descriptor information to sbuf.
3348  *
3349  * Takes a locked proc as argument, and returns with the proc unlocked.
3350  */
3351 int
3352 kern_proc_filedesc_out(struct proc *p,  struct sbuf *sb, ssize_t maxlen)
3353 {
3354 	struct file *fp;
3355 	struct filedesc *fdp;
3356 	struct export_fd_buf *efbuf;
3357 	struct vnode *cttyvp, *textvp, *tracevp;
3358 	int64_t offset;
3359 	void *data;
3360 	int error, i;
3361 	int type, refcnt, fflags;
3362 	cap_rights_t rights;
3363 
3364 	PROC_LOCK_ASSERT(p, MA_OWNED);
3365 
3366 	/* ktrace vnode */
3367 	tracevp = p->p_tracevp;
3368 	if (tracevp != NULL)
3369 		vref(tracevp);
3370 	/* text vnode */
3371 	textvp = p->p_textvp;
3372 	if (textvp != NULL)
3373 		vref(textvp);
3374 	/* Controlling tty. */
3375 	cttyvp = NULL;
3376 	if (p->p_pgrp != NULL && p->p_pgrp->pg_session != NULL) {
3377 		cttyvp = p->p_pgrp->pg_session->s_ttyvp;
3378 		if (cttyvp != NULL)
3379 			vref(cttyvp);
3380 	}
3381 	fdp = fdhold(p);
3382 	PROC_UNLOCK(p);
3383 	efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
3384 	efbuf->fdp = NULL;
3385 	efbuf->sb = sb;
3386 	efbuf->remainder = maxlen;
3387 	if (tracevp != NULL)
3388 		export_fd_to_sb(tracevp, KF_TYPE_VNODE, KF_FD_TYPE_TRACE,
3389 		    FREAD | FWRITE, -1, -1, NULL, efbuf);
3390 	if (textvp != NULL)
3391 		export_fd_to_sb(textvp, KF_TYPE_VNODE, KF_FD_TYPE_TEXT,
3392 		    FREAD, -1, -1, NULL, efbuf);
3393 	if (cttyvp != NULL)
3394 		export_fd_to_sb(cttyvp, KF_TYPE_VNODE, KF_FD_TYPE_CTTY,
3395 		    FREAD | FWRITE, -1, -1, NULL, efbuf);
3396 	error = 0;
3397 	if (fdp == NULL)
3398 		goto fail;
3399 	efbuf->fdp = fdp;
3400 	FILEDESC_SLOCK(fdp);
3401 	/* working directory */
3402 	if (fdp->fd_cdir != NULL) {
3403 		vref(fdp->fd_cdir);
3404 		data = fdp->fd_cdir;
3405 		export_fd_to_sb(data, KF_TYPE_VNODE, KF_FD_TYPE_CWD,
3406 		    FREAD, -1, -1, NULL, efbuf);
3407 	}
3408 	/* root directory */
3409 	if (fdp->fd_rdir != NULL) {
3410 		vref(fdp->fd_rdir);
3411 		data = fdp->fd_rdir;
3412 		export_fd_to_sb(data, KF_TYPE_VNODE, KF_FD_TYPE_ROOT,
3413 		    FREAD, -1, -1, NULL, efbuf);
3414 	}
3415 	/* jail directory */
3416 	if (fdp->fd_jdir != NULL) {
3417 		vref(fdp->fd_jdir);
3418 		data = fdp->fd_jdir;
3419 		export_fd_to_sb(data, KF_TYPE_VNODE, KF_FD_TYPE_JAIL,
3420 		    FREAD, -1, -1, NULL, efbuf);
3421 	}
3422 	for (i = 0; i < fdp->fd_nfiles; i++) {
3423 		if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
3424 			continue;
3425 		data = NULL;
3426 #ifdef CAPABILITIES
3427 		rights = *cap_rights(fdp, i);
3428 #else /* !CAPABILITIES */
3429 		cap_rights_init(&rights);
3430 #endif
3431 		switch (fp->f_type) {
3432 		case DTYPE_VNODE:
3433 			type = KF_TYPE_VNODE;
3434 			vref(fp->f_vnode);
3435 			data = fp->f_vnode;
3436 			break;
3437 
3438 		case DTYPE_SOCKET:
3439 			type = KF_TYPE_SOCKET;
3440 			data = fp->f_data;
3441 			break;
3442 
3443 		case DTYPE_PIPE:
3444 			type = KF_TYPE_PIPE;
3445 			data = fp->f_data;
3446 			break;
3447 
3448 		case DTYPE_FIFO:
3449 			type = KF_TYPE_FIFO;
3450 			vref(fp->f_vnode);
3451 			data = fp->f_vnode;
3452 			break;
3453 
3454 		case DTYPE_KQUEUE:
3455 			type = KF_TYPE_KQUEUE;
3456 			break;
3457 
3458 		case DTYPE_CRYPTO:
3459 			type = KF_TYPE_CRYPTO;
3460 			break;
3461 
3462 		case DTYPE_MQUEUE:
3463 			type = KF_TYPE_MQUEUE;
3464 			break;
3465 
3466 		case DTYPE_SHM:
3467 			type = KF_TYPE_SHM;
3468 			data = fp;
3469 			break;
3470 
3471 		case DTYPE_SEM:
3472 			type = KF_TYPE_SEM;
3473 			data = fp;
3474 			break;
3475 
3476 		case DTYPE_PTS:
3477 			type = KF_TYPE_PTS;
3478 			data = fp->f_data;
3479 			break;
3480 
3481 		case DTYPE_PROCDESC:
3482 			type = KF_TYPE_PROCDESC;
3483 			data = fp->f_data;
3484 			break;
3485 
3486 		default:
3487 			type = KF_TYPE_UNKNOWN;
3488 			break;
3489 		}
3490 		refcnt = fp->f_count;
3491 		fflags = fp->f_flag;
3492 		offset = foffset_get(fp);
3493 
3494 		/*
3495 		 * Create sysctl entry.
3496 		 * It is OK to drop the filedesc lock here as we will
3497 		 * re-validate and re-evaluate its properties when
3498 		 * the loop continues.
3499 		 */
3500 		error = export_fd_to_sb(data, type, i, fflags, refcnt,
3501 		    offset, &rights, efbuf);
3502 		if (error != 0)
3503 			break;
3504 	}
3505 	FILEDESC_SUNLOCK(fdp);
3506 	fddrop(fdp);
3507 fail:
3508 	free(efbuf, M_TEMP);
3509 	return (error);
3510 }
3511 
3512 #define FILEDESC_SBUF_SIZE	(sizeof(struct kinfo_file) * 5)
3513 
3514 /*
3515  * Get per-process file descriptors for use by procstat(1), et al.
3516  */
3517 static int
3518 sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS)
3519 {
3520 	struct sbuf sb;
3521 	struct proc *p;
3522 	ssize_t maxlen;
3523 	int error, error2, *name;
3524 
3525 	name = (int *)arg1;
3526 
3527 	sbuf_new_for_sysctl(&sb, NULL, FILEDESC_SBUF_SIZE, req);
3528 	error = pget((pid_t)name[0], PGET_CANDEBUG, &p);
3529 	if (error != 0) {
3530 		sbuf_delete(&sb);
3531 		return (error);
3532 	}
3533 	maxlen = req->oldptr != NULL ? req->oldlen : -1;
3534 	error = kern_proc_filedesc_out(p, &sb, maxlen);
3535 	error2 = sbuf_finish(&sb);
3536 	sbuf_delete(&sb);
3537 	return (error != 0 ? error : error2);
3538 }
3539 
3540 int
3541 vntype_to_kinfo(int vtype)
3542 {
3543 	struct {
3544 		int	vtype;
3545 		int	kf_vtype;
3546 	} vtypes_table[] = {
3547 		{ VBAD, KF_VTYPE_VBAD },
3548 		{ VBLK, KF_VTYPE_VBLK },
3549 		{ VCHR, KF_VTYPE_VCHR },
3550 		{ VDIR, KF_VTYPE_VDIR },
3551 		{ VFIFO, KF_VTYPE_VFIFO },
3552 		{ VLNK, KF_VTYPE_VLNK },
3553 		{ VNON, KF_VTYPE_VNON },
3554 		{ VREG, KF_VTYPE_VREG },
3555 		{ VSOCK, KF_VTYPE_VSOCK }
3556 	};
3557 #define	NVTYPES	(sizeof(vtypes_table) / sizeof(*vtypes_table))
3558 	unsigned int i;
3559 
3560 	/*
3561 	 * Perform vtype translation.
3562 	 */
3563 	for (i = 0; i < NVTYPES; i++)
3564 		if (vtypes_table[i].vtype == vtype)
3565 			break;
3566 	if (i < NVTYPES)
3567 		return (vtypes_table[i].kf_vtype);
3568 
3569 	return (KF_VTYPE_UNKNOWN);
3570 }
3571 
3572 static int
3573 fill_vnode_info(struct vnode *vp, struct kinfo_file *kif)
3574 {
3575 	struct vattr va;
3576 	char *fullpath, *freepath;
3577 	int error;
3578 
3579 	if (vp == NULL)
3580 		return (1);
3581 	kif->kf_vnode_type = vntype_to_kinfo(vp->v_type);
3582 	freepath = NULL;
3583 	fullpath = "-";
3584 	error = vn_fullpath(curthread, vp, &fullpath, &freepath);
3585 	if (error == 0) {
3586 		strlcpy(kif->kf_path, fullpath, sizeof(kif->kf_path));
3587 	}
3588 	if (freepath != NULL)
3589 		free(freepath, M_TEMP);
3590 
3591 	/*
3592 	 * Retrieve vnode attributes.
3593 	 */
3594 	va.va_fsid = VNOVAL;
3595 	va.va_rdev = NODEV;
3596 	vn_lock(vp, LK_SHARED | LK_RETRY);
3597 	error = VOP_GETATTR(vp, &va, curthread->td_ucred);
3598 	VOP_UNLOCK(vp, 0);
3599 	if (error != 0)
3600 		return (error);
3601 	if (va.va_fsid != VNOVAL)
3602 		kif->kf_un.kf_file.kf_file_fsid = va.va_fsid;
3603 	else
3604 		kif->kf_un.kf_file.kf_file_fsid =
3605 		    vp->v_mount->mnt_stat.f_fsid.val[0];
3606 	kif->kf_un.kf_file.kf_file_fileid = va.va_fileid;
3607 	kif->kf_un.kf_file.kf_file_mode = MAKEIMODE(va.va_type, va.va_mode);
3608 	kif->kf_un.kf_file.kf_file_size = va.va_size;
3609 	kif->kf_un.kf_file.kf_file_rdev = va.va_rdev;
3610 	return (0);
3611 }
3612 
3613 static int
3614 fill_socket_info(struct socket *so, struct kinfo_file *kif)
3615 {
3616 	struct sockaddr *sa;
3617 	struct inpcb *inpcb;
3618 	struct unpcb *unpcb;
3619 	int error;
3620 
3621 	if (so == NULL)
3622 		return (1);
3623 	kif->kf_sock_domain = so->so_proto->pr_domain->dom_family;
3624 	kif->kf_sock_type = so->so_type;
3625 	kif->kf_sock_protocol = so->so_proto->pr_protocol;
3626 	kif->kf_un.kf_sock.kf_sock_pcb = (uintptr_t)so->so_pcb;
3627 	switch(kif->kf_sock_domain) {
3628 	case AF_INET:
3629 	case AF_INET6:
3630 		if (kif->kf_sock_protocol == IPPROTO_TCP) {
3631 			if (so->so_pcb != NULL) {
3632 				inpcb = (struct inpcb *)(so->so_pcb);
3633 				kif->kf_un.kf_sock.kf_sock_inpcb =
3634 				    (uintptr_t)inpcb->inp_ppcb;
3635 			}
3636 		}
3637 		break;
3638 	case AF_UNIX:
3639 		if (so->so_pcb != NULL) {
3640 			unpcb = (struct unpcb *)(so->so_pcb);
3641 			if (unpcb->unp_conn) {
3642 				kif->kf_un.kf_sock.kf_sock_unpconn =
3643 				    (uintptr_t)unpcb->unp_conn;
3644 				kif->kf_un.kf_sock.kf_sock_rcv_sb_state =
3645 				    so->so_rcv.sb_state;
3646 				kif->kf_un.kf_sock.kf_sock_snd_sb_state =
3647 				    so->so_snd.sb_state;
3648 			}
3649 		}
3650 		break;
3651 	}
3652 	error = so->so_proto->pr_usrreqs->pru_sockaddr(so, &sa);
3653 	if (error == 0 && sa->sa_len <= sizeof(kif->kf_sa_local)) {
3654 		bcopy(sa, &kif->kf_sa_local, sa->sa_len);
3655 		free(sa, M_SONAME);
3656 	}
3657 	error = so->so_proto->pr_usrreqs->pru_peeraddr(so, &sa);
3658 	if (error == 0 && sa->sa_len <= sizeof(kif->kf_sa_peer)) {
3659 		bcopy(sa, &kif->kf_sa_peer, sa->sa_len);
3660 		free(sa, M_SONAME);
3661 	}
3662 	strncpy(kif->kf_path, so->so_proto->pr_domain->dom_name,
3663 	    sizeof(kif->kf_path));
3664 	return (0);
3665 }
3666 
3667 static int
3668 fill_pts_info(struct tty *tp, struct kinfo_file *kif)
3669 {
3670 
3671 	if (tp == NULL)
3672 		return (1);
3673 	kif->kf_un.kf_pts.kf_pts_dev = tty_udev(tp);
3674 	strlcpy(kif->kf_path, tty_devname(tp), sizeof(kif->kf_path));
3675 	return (0);
3676 }
3677 
3678 static int
3679 fill_pipe_info(struct pipe *pi, struct kinfo_file *kif)
3680 {
3681 
3682 	if (pi == NULL)
3683 		return (1);
3684 	kif->kf_un.kf_pipe.kf_pipe_addr = (uintptr_t)pi;
3685 	kif->kf_un.kf_pipe.kf_pipe_peer = (uintptr_t)pi->pipe_peer;
3686 	kif->kf_un.kf_pipe.kf_pipe_buffer_cnt = pi->pipe_buffer.cnt;
3687 	return (0);
3688 }
3689 
3690 static int
3691 fill_procdesc_info(struct procdesc *pdp, struct kinfo_file *kif)
3692 {
3693 
3694 	if (pdp == NULL)
3695 		return (1);
3696 	kif->kf_un.kf_proc.kf_pid = pdp->pd_pid;
3697 	return (0);
3698 }
3699 
3700 static int
3701 fill_sem_info(struct file *fp, struct kinfo_file *kif)
3702 {
3703 	struct thread *td;
3704 	struct stat sb;
3705 
3706 	td = curthread;
3707 	if (fp->f_data == NULL)
3708 		return (1);
3709 	if (fo_stat(fp, &sb, td->td_ucred, td) != 0)
3710 		return (1);
3711 	if (ksem_info == NULL)
3712 		return (1);
3713 	ksem_info(fp->f_data, kif->kf_path, sizeof(kif->kf_path),
3714 	    &kif->kf_un.kf_sem.kf_sem_value);
3715 	kif->kf_un.kf_sem.kf_sem_mode = sb.st_mode;
3716 	return (0);
3717 }
3718 
3719 static int
3720 fill_shm_info(struct file *fp, struct kinfo_file *kif)
3721 {
3722 	struct thread *td;
3723 	struct stat sb;
3724 
3725 	td = curthread;
3726 	if (fp->f_data == NULL)
3727 		return (1);
3728 	if (fo_stat(fp, &sb, td->td_ucred, td) != 0)
3729 		return (1);
3730 	shm_path(fp->f_data, kif->kf_path, sizeof(kif->kf_path));
3731 	kif->kf_un.kf_file.kf_file_mode = sb.st_mode;
3732 	kif->kf_un.kf_file.kf_file_size = sb.st_size;
3733 	return (0);
3734 }
3735 
3736 static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc, CTLFLAG_RD,
3737     sysctl_kern_proc_filedesc, "Process filedesc entries");
3738 
3739 #ifdef DDB
3740 /*
3741  * For the purposes of debugging, generate a human-readable string for the
3742  * file type.
3743  */
3744 static const char *
3745 file_type_to_name(short type)
3746 {
3747 
3748 	switch (type) {
3749 	case 0:
3750 		return ("zero");
3751 	case DTYPE_VNODE:
3752 		return ("vnod");
3753 	case DTYPE_SOCKET:
3754 		return ("sock");
3755 	case DTYPE_PIPE:
3756 		return ("pipe");
3757 	case DTYPE_FIFO:
3758 		return ("fifo");
3759 	case DTYPE_KQUEUE:
3760 		return ("kque");
3761 	case DTYPE_CRYPTO:
3762 		return ("crpt");
3763 	case DTYPE_MQUEUE:
3764 		return ("mque");
3765 	case DTYPE_SHM:
3766 		return ("shm");
3767 	case DTYPE_SEM:
3768 		return ("ksem");
3769 	default:
3770 		return ("unkn");
3771 	}
3772 }
3773 
3774 /*
3775  * For the purposes of debugging, identify a process (if any, perhaps one of
3776  * many) that references the passed file in its file descriptor array. Return
3777  * NULL if none.
3778  */
3779 static struct proc *
3780 file_to_first_proc(struct file *fp)
3781 {
3782 	struct filedesc *fdp;
3783 	struct proc *p;
3784 	int n;
3785 
3786 	FOREACH_PROC_IN_SYSTEM(p) {
3787 		if (p->p_state == PRS_NEW)
3788 			continue;
3789 		fdp = p->p_fd;
3790 		if (fdp == NULL)
3791 			continue;
3792 		for (n = 0; n < fdp->fd_nfiles; n++) {
3793 			if (fp == fdp->fd_ofiles[n].fde_file)
3794 				return (p);
3795 		}
3796 	}
3797 	return (NULL);
3798 }
3799 
3800 static void
3801 db_print_file(struct file *fp, int header)
3802 {
3803 	struct proc *p;
3804 
3805 	if (header)
3806 		db_printf("%8s %4s %8s %8s %4s %5s %6s %8s %5s %12s\n",
3807 		    "File", "Type", "Data", "Flag", "GCFl", "Count",
3808 		    "MCount", "Vnode", "FPID", "FCmd");
3809 	p = file_to_first_proc(fp);
3810 	db_printf("%8p %4s %8p %08x %04x %5d %6d %8p %5d %12s\n", fp,
3811 	    file_type_to_name(fp->f_type), fp->f_data, fp->f_flag,
3812 	    0, fp->f_count, 0, fp->f_vnode,
3813 	    p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-");
3814 }
3815 
3816 DB_SHOW_COMMAND(file, db_show_file)
3817 {
3818 	struct file *fp;
3819 
3820 	if (!have_addr) {
3821 		db_printf("usage: show file <addr>\n");
3822 		return;
3823 	}
3824 	fp = (struct file *)addr;
3825 	db_print_file(fp, 1);
3826 }
3827 
3828 DB_SHOW_COMMAND(files, db_show_files)
3829 {
3830 	struct filedesc *fdp;
3831 	struct file *fp;
3832 	struct proc *p;
3833 	int header;
3834 	int n;
3835 
3836 	header = 1;
3837 	FOREACH_PROC_IN_SYSTEM(p) {
3838 		if (p->p_state == PRS_NEW)
3839 			continue;
3840 		if ((fdp = p->p_fd) == NULL)
3841 			continue;
3842 		for (n = 0; n < fdp->fd_nfiles; ++n) {
3843 			if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
3844 				continue;
3845 			db_print_file(fp, header);
3846 			header = 0;
3847 		}
3848 	}
3849 }
3850 #endif
3851 
3852 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
3853     &maxfilesperproc, 0, "Maximum files allowed open per process");
3854 
3855 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
3856     &maxfiles, 0, "Maximum number of files");
3857 
3858 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
3859     __DEVOLATILE(int *, &openfiles), 0, "System-wide number of open files");
3860 
3861 /* ARGSUSED*/
3862 static void
3863 filelistinit(void *dummy)
3864 {
3865 
3866 	file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL,
3867 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
3868 	mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF);
3869 	mtx_init(&fdesc_mtx, "fdesc", NULL, MTX_DEF);
3870 }
3871 SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL);
3872 
3873 /*-------------------------------------------------------------------*/
3874 
3875 static int
3876 badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred,
3877     int flags, struct thread *td)
3878 {
3879 
3880 	return (EBADF);
3881 }
3882 
3883 static int
3884 badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
3885     struct thread *td)
3886 {
3887 
3888 	return (EINVAL);
3889 }
3890 
3891 static int
3892 badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred,
3893     struct thread *td)
3894 {
3895 
3896 	return (EBADF);
3897 }
3898 
3899 static int
3900 badfo_poll(struct file *fp, int events, struct ucred *active_cred,
3901     struct thread *td)
3902 {
3903 
3904 	return (0);
3905 }
3906 
3907 static int
3908 badfo_kqfilter(struct file *fp, struct knote *kn)
3909 {
3910 
3911 	return (EBADF);
3912 }
3913 
3914 static int
3915 badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred,
3916     struct thread *td)
3917 {
3918 
3919 	return (EBADF);
3920 }
3921 
3922 static int
3923 badfo_close(struct file *fp, struct thread *td)
3924 {
3925 
3926 	return (EBADF);
3927 }
3928 
3929 static int
3930 badfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
3931     struct thread *td)
3932 {
3933 
3934 	return (EBADF);
3935 }
3936 
3937 static int
3938 badfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
3939     struct thread *td)
3940 {
3941 
3942 	return (EBADF);
3943 }
3944 
3945 static int
3946 badfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
3947     struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
3948     int kflags, struct sendfile_sync *sfs, struct thread *td)
3949 {
3950 
3951 	return (EBADF);
3952 }
3953 
3954 struct fileops badfileops = {
3955 	.fo_read = badfo_readwrite,
3956 	.fo_write = badfo_readwrite,
3957 	.fo_truncate = badfo_truncate,
3958 	.fo_ioctl = badfo_ioctl,
3959 	.fo_poll = badfo_poll,
3960 	.fo_kqfilter = badfo_kqfilter,
3961 	.fo_stat = badfo_stat,
3962 	.fo_close = badfo_close,
3963 	.fo_chmod = badfo_chmod,
3964 	.fo_chown = badfo_chown,
3965 	.fo_sendfile = badfo_sendfile,
3966 };
3967 
3968 int
3969 invfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
3970     struct thread *td)
3971 {
3972 
3973 	return (EINVAL);
3974 }
3975 
3976 int
3977 invfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
3978     struct thread *td)
3979 {
3980 
3981 	return (EINVAL);
3982 }
3983 
3984 int
3985 invfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
3986     struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
3987     int kflags, struct sendfile_sync *sfs, struct thread *td)
3988 {
3989 
3990 	return (EINVAL);
3991 }
3992 
3993 /*-------------------------------------------------------------------*/
3994 
3995 /*
3996  * File Descriptor pseudo-device driver (/dev/fd/).
3997  *
3998  * Opening minor device N dup()s the file (if any) connected to file
3999  * descriptor N belonging to the calling process.  Note that this driver
4000  * consists of only the ``open()'' routine, because all subsequent
4001  * references to this file will be direct to the other driver.
4002  *
4003  * XXX: we could give this one a cloning event handler if necessary.
4004  */
4005 
4006 /* ARGSUSED */
4007 static int
4008 fdopen(struct cdev *dev, int mode, int type, struct thread *td)
4009 {
4010 
4011 	/*
4012 	 * XXX Kludge: set curthread->td_dupfd to contain the value of the
4013 	 * the file descriptor being sought for duplication. The error
4014 	 * return ensures that the vnode for this device will be released
4015 	 * by vn_open. Open will detect this special error and take the
4016 	 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
4017 	 * will simply report the error.
4018 	 */
4019 	td->td_dupfd = dev2unit(dev);
4020 	return (ENODEV);
4021 }
4022 
4023 static struct cdevsw fildesc_cdevsw = {
4024 	.d_version =	D_VERSION,
4025 	.d_open =	fdopen,
4026 	.d_name =	"FD",
4027 };
4028 
4029 static void
4030 fildesc_drvinit(void *unused)
4031 {
4032 	struct cdev *dev;
4033 
4034 	dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 0, NULL,
4035 	    UID_ROOT, GID_WHEEL, 0666, "fd/0");
4036 	make_dev_alias(dev, "stdin");
4037 	dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 1, NULL,
4038 	    UID_ROOT, GID_WHEEL, 0666, "fd/1");
4039 	make_dev_alias(dev, "stdout");
4040 	dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 2, NULL,
4041 	    UID_ROOT, GID_WHEEL, 0666, "fd/2");
4042 	make_dev_alias(dev, "stderr");
4043 }
4044 
4045 SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL);
4046