xref: /freebsd/sys/kern/kern_descrip.c (revision a9e8641da961bcf3d24afc85fd657f2083a872a2)
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.  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 	/* copy the old data over and point at the new tables */
1533 	memcpy(ntable, otable, onfiles * sizeof(*otable));
1534 	fdp->fd_ofiles = ntable;
1535 
1536 	/* Allocate a new map only if the old is not large enough.  It will
1537 	 * grow at a slower rate than the table as it can map more
1538 	 * entries than the table can hold. */
1539 	if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) {
1540 		nmap = malloc(NDSLOTS(nnfiles) * NDSLOTSIZE, M_FILEDESC,
1541 		    M_ZERO | M_WAITOK);
1542 		/* copy over the old data and update the pointer */
1543 		memcpy(nmap, omap, NDSLOTS(onfiles) * sizeof(*omap));
1544 		fdp->fd_map = nmap;
1545 	}
1546 
1547 	/*
1548 	 * In order to have a valid pattern for fget_unlocked()
1549 	 * fdp->fd_nfiles must be the last member to be updated, otherwise
1550 	 * fget_unlocked() consumers may reference a new, higher value for
1551 	 * fdp->fd_nfiles before to access the fdp->fd_ofiles array,
1552 	 * resulting in OOB accesses.
1553 	 */
1554 	atomic_store_rel_int(&fdp->fd_nfiles, nnfiles);
1555 
1556 	/*
1557 	 * Do not free the old file table, as some threads may still
1558 	 * reference entries within it.  Instead, place it on a freelist
1559 	 * which will be processed when the struct filedesc is released.
1560 	 *
1561 	 * Note that if onfiles == NDFILE, we're dealing with the original
1562 	 * static allocation contained within (struct filedesc0 *)fdp,
1563 	 * which must not be freed.
1564 	 */
1565 	if (onfiles > NDFILE) {
1566 		ft = (struct freetable *)&otable[onfiles];
1567 		fdp0 = (struct filedesc0 *)fdp;
1568 		ft->ft_table = otable;
1569 		SLIST_INSERT_HEAD(&fdp0->fd_free, ft, ft_next);
1570 	}
1571 	/* The map does not have the same possibility of threads still
1572 	 * holding references to it.  So always free it as long as it
1573 	 * does not reference the original static allocation. */
1574 	if (NDSLOTS(onfiles) > NDSLOTS(NDFILE))
1575 		free(omap, M_FILEDESC);
1576 }
1577 
1578 /*
1579  * Allocate a file descriptor for the process.
1580  */
1581 int
1582 fdalloc(struct thread *td, int minfd, int *result)
1583 {
1584 	struct proc *p = td->td_proc;
1585 	struct filedesc *fdp = p->p_fd;
1586 	int fd = -1, maxfd, allocfd;
1587 #ifdef RACCT
1588 	int error;
1589 #endif
1590 
1591 	FILEDESC_XLOCK_ASSERT(fdp);
1592 
1593 	if (fdp->fd_freefile > minfd)
1594 		minfd = fdp->fd_freefile;
1595 
1596 	maxfd = getmaxfd(p);
1597 
1598 	/*
1599 	 * Search the bitmap for a free descriptor starting at minfd.
1600 	 * If none is found, grow the file table.
1601 	 */
1602 	fd = fd_first_free(fdp, minfd, fdp->fd_nfiles);
1603 	if (fd >= maxfd)
1604 		return (EMFILE);
1605 	if (fd >= fdp->fd_nfiles) {
1606 		allocfd = min(fd * 2, maxfd);
1607 #ifdef RACCT
1608 		PROC_LOCK(p);
1609 		error = racct_set(p, RACCT_NOFILE, allocfd);
1610 		PROC_UNLOCK(p);
1611 		if (error != 0)
1612 			return (EMFILE);
1613 #endif
1614 		/*
1615 		 * fd is already equal to first free descriptor >= minfd, so
1616 		 * we only need to grow the table and we are done.
1617 		 */
1618 		fdgrowtable_exp(fdp, allocfd);
1619 	}
1620 
1621 	/*
1622 	 * Perform some sanity checks, then mark the file descriptor as
1623 	 * used and return it to the caller.
1624 	 */
1625 	KASSERT(fd >= 0 && fd < min(maxfd, fdp->fd_nfiles),
1626 	    ("invalid descriptor %d", fd));
1627 	KASSERT(!fdisused(fdp, fd),
1628 	    ("fd_first_free() returned non-free descriptor"));
1629 	KASSERT(fdp->fd_ofiles[fd].fde_file == NULL,
1630 	    ("file descriptor isn't free"));
1631 	KASSERT(fdp->fd_ofiles[fd].fde_flags == 0, ("file flags are set"));
1632 	fdused(fdp, fd);
1633 	*result = fd;
1634 	return (0);
1635 }
1636 
1637 /*
1638  * Allocate n file descriptors for the process.
1639  */
1640 int
1641 fdallocn(struct thread *td, int minfd, int *fds, int n)
1642 {
1643 	struct proc *p = td->td_proc;
1644 	struct filedesc *fdp = p->p_fd;
1645 	int i;
1646 
1647 	FILEDESC_XLOCK_ASSERT(fdp);
1648 
1649 	if (!fdavail(td, n))
1650 		return (EMFILE);
1651 
1652 	for (i = 0; i < n; i++)
1653 		if (fdalloc(td, 0, &fds[i]) != 0)
1654 			break;
1655 
1656 	if (i < n) {
1657 		for (i--; i >= 0; i--)
1658 			fdunused(fdp, fds[i]);
1659 		return (EMFILE);
1660 	}
1661 
1662 	return (0);
1663 }
1664 
1665 /*
1666  * Check to see whether n user file descriptors are available to the process
1667  * p.
1668  */
1669 int
1670 fdavail(struct thread *td, int n)
1671 {
1672 	struct proc *p = td->td_proc;
1673 	struct filedesc *fdp = td->td_proc->p_fd;
1674 	int i, lim, last;
1675 
1676 	FILEDESC_LOCK_ASSERT(fdp);
1677 
1678 	/*
1679 	 * XXX: This is only called from uipc_usrreq.c:unp_externalize();
1680 	 *      call racct_add() from there instead of dealing with containers
1681 	 *      here.
1682 	 */
1683 	lim = getmaxfd(p);
1684 	if ((i = lim - fdp->fd_nfiles) > 0 && (n -= i) <= 0)
1685 		return (1);
1686 	last = min(fdp->fd_nfiles, lim);
1687 	for (i = fdp->fd_freefile; i < last; i++) {
1688 		if (fdp->fd_ofiles[i].fde_file == NULL && --n <= 0)
1689 			return (1);
1690 	}
1691 	return (0);
1692 }
1693 
1694 /*
1695  * Create a new open file structure and allocate a file decriptor for the
1696  * process that refers to it.  We add one reference to the file for the
1697  * descriptor table and one reference for resultfp. This is to prevent us
1698  * being preempted and the entry in the descriptor table closed after we
1699  * release the FILEDESC lock.
1700  */
1701 int
1702 falloc(struct thread *td, struct file **resultfp, int *resultfd, int flags)
1703 {
1704 	struct file *fp;
1705 	int error, fd;
1706 
1707 	error = falloc_noinstall(td, &fp);
1708 	if (error)
1709 		return (error);		/* no reference held on error */
1710 
1711 	error = finstall(td, fp, &fd, flags, NULL);
1712 	if (error) {
1713 		fdrop(fp, td);		/* one reference (fp only) */
1714 		return (error);
1715 	}
1716 
1717 	if (resultfp != NULL)
1718 		*resultfp = fp;		/* copy out result */
1719 	else
1720 		fdrop(fp, td);		/* release local reference */
1721 
1722 	if (resultfd != NULL)
1723 		*resultfd = fd;
1724 
1725 	return (0);
1726 }
1727 
1728 /*
1729  * Create a new open file structure without allocating a file descriptor.
1730  */
1731 int
1732 falloc_noinstall(struct thread *td, struct file **resultfp)
1733 {
1734 	struct file *fp;
1735 	int maxuserfiles = maxfiles - (maxfiles / 20);
1736 	static struct timeval lastfail;
1737 	static int curfail;
1738 
1739 	KASSERT(resultfp != NULL, ("%s: resultfp == NULL", __func__));
1740 
1741 	if ((openfiles >= maxuserfiles &&
1742 	    priv_check(td, PRIV_MAXFILES) != 0) ||
1743 	    openfiles >= maxfiles) {
1744 		if (ppsratecheck(&lastfail, &curfail, 1)) {
1745 			printf("kern.maxfiles limit exceeded by uid %i, "
1746 			    "please see tuning(7).\n", td->td_ucred->cr_ruid);
1747 		}
1748 		return (ENFILE);
1749 	}
1750 	atomic_add_int(&openfiles, 1);
1751 	fp = uma_zalloc(file_zone, M_WAITOK | M_ZERO);
1752 	refcount_init(&fp->f_count, 1);
1753 	fp->f_cred = crhold(td->td_ucred);
1754 	fp->f_ops = &badfileops;
1755 	fp->f_data = NULL;
1756 	fp->f_vnode = NULL;
1757 	*resultfp = fp;
1758 	return (0);
1759 }
1760 
1761 /*
1762  * Install a file in a file descriptor table.
1763  */
1764 int
1765 finstall(struct thread *td, struct file *fp, int *fd, int flags,
1766     struct filecaps *fcaps)
1767 {
1768 	struct filedesc *fdp = td->td_proc->p_fd;
1769 	struct filedescent *fde;
1770 	int error;
1771 
1772 	KASSERT(fd != NULL, ("%s: fd == NULL", __func__));
1773 	KASSERT(fp != NULL, ("%s: fp == NULL", __func__));
1774 	if (fcaps != NULL)
1775 		filecaps_validate(fcaps, __func__);
1776 
1777 	FILEDESC_XLOCK(fdp);
1778 	if ((error = fdalloc(td, 0, fd))) {
1779 		FILEDESC_XUNLOCK(fdp);
1780 		return (error);
1781 	}
1782 	fhold(fp);
1783 	fde = &fdp->fd_ofiles[*fd];
1784 	fde->fde_file = fp;
1785 	if ((flags & O_CLOEXEC) != 0)
1786 		fde->fde_flags |= UF_EXCLOSE;
1787 	if (fcaps != NULL)
1788 		filecaps_move(fcaps, &fde->fde_caps);
1789 	else
1790 		filecaps_fill(&fde->fde_caps);
1791 	FILEDESC_XUNLOCK(fdp);
1792 	return (0);
1793 }
1794 
1795 /*
1796  * Build a new filedesc structure from another.
1797  * Copy the current, root, and jail root vnode references.
1798  */
1799 struct filedesc *
1800 fdinit(struct filedesc *fdp)
1801 {
1802 	struct filedesc0 *newfdp;
1803 
1804 	newfdp = malloc(sizeof *newfdp, M_FILEDESC, M_WAITOK | M_ZERO);
1805 	FILEDESC_LOCK_INIT(&newfdp->fd_fd);
1806 	if (fdp != NULL) {
1807 		FILEDESC_XLOCK(fdp);
1808 		newfdp->fd_fd.fd_cdir = fdp->fd_cdir;
1809 		if (newfdp->fd_fd.fd_cdir)
1810 			VREF(newfdp->fd_fd.fd_cdir);
1811 		newfdp->fd_fd.fd_rdir = fdp->fd_rdir;
1812 		if (newfdp->fd_fd.fd_rdir)
1813 			VREF(newfdp->fd_fd.fd_rdir);
1814 		newfdp->fd_fd.fd_jdir = fdp->fd_jdir;
1815 		if (newfdp->fd_fd.fd_jdir)
1816 			VREF(newfdp->fd_fd.fd_jdir);
1817 		FILEDESC_XUNLOCK(fdp);
1818 	}
1819 
1820 	/* Create the file descriptor table. */
1821 	newfdp->fd_fd.fd_refcnt = 1;
1822 	newfdp->fd_fd.fd_holdcnt = 1;
1823 	newfdp->fd_fd.fd_cmask = CMASK;
1824 	newfdp->fd_fd.fd_ofiles = newfdp->fd_dfiles;
1825 	newfdp->fd_fd.fd_nfiles = NDFILE;
1826 	newfdp->fd_fd.fd_map = newfdp->fd_dmap;
1827 	newfdp->fd_fd.fd_lastfile = -1;
1828 	return (&newfdp->fd_fd);
1829 }
1830 
1831 static struct filedesc *
1832 fdhold(struct proc *p)
1833 {
1834 	struct filedesc *fdp;
1835 
1836 	mtx_lock(&fdesc_mtx);
1837 	fdp = p->p_fd;
1838 	if (fdp != NULL)
1839 		fdp->fd_holdcnt++;
1840 	mtx_unlock(&fdesc_mtx);
1841 	return (fdp);
1842 }
1843 
1844 static void
1845 fddrop(struct filedesc *fdp)
1846 {
1847 	struct filedesc0 *fdp0;
1848 	struct freetable *ft;
1849 	int i;
1850 
1851 	mtx_lock(&fdesc_mtx);
1852 	i = --fdp->fd_holdcnt;
1853 	mtx_unlock(&fdesc_mtx);
1854 	if (i > 0)
1855 		return;
1856 
1857 	FILEDESC_LOCK_DESTROY(fdp);
1858 	fdp0 = (struct filedesc0 *)fdp;
1859 	while ((ft = SLIST_FIRST(&fdp0->fd_free)) != NULL) {
1860 		SLIST_REMOVE_HEAD(&fdp0->fd_free, ft_next);
1861 		free(ft->ft_table, M_FILEDESC);
1862 	}
1863 	free(fdp, M_FILEDESC);
1864 }
1865 
1866 /*
1867  * Share a filedesc structure.
1868  */
1869 struct filedesc *
1870 fdshare(struct filedesc *fdp)
1871 {
1872 
1873 	FILEDESC_XLOCK(fdp);
1874 	fdp->fd_refcnt++;
1875 	FILEDESC_XUNLOCK(fdp);
1876 	return (fdp);
1877 }
1878 
1879 /*
1880  * Unshare a filedesc structure, if necessary by making a copy
1881  */
1882 void
1883 fdunshare(struct proc *p, struct thread *td)
1884 {
1885 
1886 	FILEDESC_XLOCK(p->p_fd);
1887 	if (p->p_fd->fd_refcnt > 1) {
1888 		struct filedesc *tmp;
1889 
1890 		FILEDESC_XUNLOCK(p->p_fd);
1891 		tmp = fdcopy(p->p_fd);
1892 		fdescfree(td);
1893 		p->p_fd = tmp;
1894 	} else
1895 		FILEDESC_XUNLOCK(p->p_fd);
1896 }
1897 
1898 /*
1899  * Copy a filedesc structure.  A NULL pointer in returns a NULL reference,
1900  * this is to ease callers, not catch errors.
1901  */
1902 struct filedesc *
1903 fdcopy(struct filedesc *fdp)
1904 {
1905 	struct filedesc *newfdp;
1906 	struct filedescent *nfde, *ofde;
1907 	int i;
1908 
1909 	/* Certain daemons might not have file descriptors. */
1910 	if (fdp == NULL)
1911 		return (NULL);
1912 
1913 	newfdp = fdinit(fdp);
1914 	FILEDESC_SLOCK(fdp);
1915 	while (fdp->fd_lastfile >= newfdp->fd_nfiles) {
1916 		FILEDESC_SUNLOCK(fdp);
1917 		FILEDESC_XLOCK(newfdp);
1918 		fdgrowtable(newfdp, fdp->fd_lastfile + 1);
1919 		FILEDESC_XUNLOCK(newfdp);
1920 		FILEDESC_SLOCK(fdp);
1921 	}
1922 	/* copy all passable descriptors (i.e. not kqueue) */
1923 	newfdp->fd_freefile = -1;
1924 	for (i = 0; i <= fdp->fd_lastfile; ++i) {
1925 		ofde = &fdp->fd_ofiles[i];
1926 		if (fdisused(fdp, i) &&
1927 		    (ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) &&
1928 		    ofde->fde_file->f_ops != &badfileops) {
1929 			nfde = &newfdp->fd_ofiles[i];
1930 			*nfde = *ofde;
1931 			filecaps_copy(&ofde->fde_caps, &nfde->fde_caps);
1932 			fhold(nfde->fde_file);
1933 			newfdp->fd_lastfile = i;
1934 		} else {
1935 			if (newfdp->fd_freefile == -1)
1936 				newfdp->fd_freefile = i;
1937 		}
1938 	}
1939 	newfdp->fd_cmask = fdp->fd_cmask;
1940 	FILEDESC_SUNLOCK(fdp);
1941 	FILEDESC_XLOCK(newfdp);
1942 	for (i = 0; i <= newfdp->fd_lastfile; ++i) {
1943 		if (newfdp->fd_ofiles[i].fde_file != NULL)
1944 			fdused(newfdp, i);
1945 	}
1946 	if (newfdp->fd_freefile == -1)
1947 		newfdp->fd_freefile = i;
1948 	FILEDESC_XUNLOCK(newfdp);
1949 	return (newfdp);
1950 }
1951 
1952 /*
1953  * Release a filedesc structure.
1954  */
1955 void
1956 fdescfree(struct thread *td)
1957 {
1958 	struct filedesc *fdp;
1959 	int i;
1960 	struct filedesc_to_leader *fdtol;
1961 	struct file *fp;
1962 	struct vnode *cdir, *jdir, *rdir, *vp;
1963 	struct flock lf;
1964 
1965 	/* Certain daemons might not have file descriptors. */
1966 	fdp = td->td_proc->p_fd;
1967 	if (fdp == NULL)
1968 		return;
1969 
1970 #ifdef RACCT
1971 	PROC_LOCK(td->td_proc);
1972 	racct_set(td->td_proc, RACCT_NOFILE, 0);
1973 	PROC_UNLOCK(td->td_proc);
1974 #endif
1975 
1976 	/* Check for special need to clear POSIX style locks */
1977 	fdtol = td->td_proc->p_fdtol;
1978 	if (fdtol != NULL) {
1979 		FILEDESC_XLOCK(fdp);
1980 		KASSERT(fdtol->fdl_refcount > 0,
1981 		    ("filedesc_to_refcount botch: fdl_refcount=%d",
1982 		    fdtol->fdl_refcount));
1983 		if (fdtol->fdl_refcount == 1 &&
1984 		    (td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
1985 			for (i = 0; i <= fdp->fd_lastfile; i++) {
1986 				fp = fdp->fd_ofiles[i].fde_file;
1987 				if (fp == NULL || fp->f_type != DTYPE_VNODE)
1988 					continue;
1989 				fhold(fp);
1990 				FILEDESC_XUNLOCK(fdp);
1991 				lf.l_whence = SEEK_SET;
1992 				lf.l_start = 0;
1993 				lf.l_len = 0;
1994 				lf.l_type = F_UNLCK;
1995 				vp = fp->f_vnode;
1996 				(void) VOP_ADVLOCK(vp,
1997 				    (caddr_t)td->td_proc->p_leader, F_UNLCK,
1998 				    &lf, F_POSIX);
1999 				FILEDESC_XLOCK(fdp);
2000 				fdrop(fp, td);
2001 			}
2002 		}
2003 	retry:
2004 		if (fdtol->fdl_refcount == 1) {
2005 			if (fdp->fd_holdleaderscount > 0 &&
2006 			    (td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
2007 				/*
2008 				 * close() or do_dup() has cleared a reference
2009 				 * in a shared file descriptor table.
2010 				 */
2011 				fdp->fd_holdleaderswakeup = 1;
2012 				sx_sleep(&fdp->fd_holdleaderscount,
2013 				    FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0);
2014 				goto retry;
2015 			}
2016 			if (fdtol->fdl_holdcount > 0) {
2017 				/*
2018 				 * Ensure that fdtol->fdl_leader remains
2019 				 * valid in closef().
2020 				 */
2021 				fdtol->fdl_wakeup = 1;
2022 				sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK,
2023 				    "fdlhold", 0);
2024 				goto retry;
2025 			}
2026 		}
2027 		fdtol->fdl_refcount--;
2028 		if (fdtol->fdl_refcount == 0 &&
2029 		    fdtol->fdl_holdcount == 0) {
2030 			fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
2031 			fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
2032 		} else
2033 			fdtol = NULL;
2034 		td->td_proc->p_fdtol = NULL;
2035 		FILEDESC_XUNLOCK(fdp);
2036 		if (fdtol != NULL)
2037 			free(fdtol, M_FILEDESC_TO_LEADER);
2038 	}
2039 	FILEDESC_XLOCK(fdp);
2040 	i = --fdp->fd_refcnt;
2041 	FILEDESC_XUNLOCK(fdp);
2042 	if (i > 0)
2043 		return;
2044 
2045 	for (i = 0; i <= fdp->fd_lastfile; i++) {
2046 		fp = fdp->fd_ofiles[i].fde_file;
2047 		if (fp != NULL) {
2048 			FILEDESC_XLOCK(fdp);
2049 			fdfree(fdp, i);
2050 			FILEDESC_XUNLOCK(fdp);
2051 			(void) closef(fp, td);
2052 		}
2053 	}
2054 	FILEDESC_XLOCK(fdp);
2055 
2056 	/* XXX This should happen earlier. */
2057 	mtx_lock(&fdesc_mtx);
2058 	td->td_proc->p_fd = NULL;
2059 	mtx_unlock(&fdesc_mtx);
2060 
2061 	if (fdp->fd_nfiles > NDFILE)
2062 		free(fdp->fd_ofiles, M_FILEDESC);
2063 	if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE))
2064 		free(fdp->fd_map, M_FILEDESC);
2065 
2066 	fdp->fd_nfiles = 0;
2067 
2068 	cdir = fdp->fd_cdir;
2069 	fdp->fd_cdir = NULL;
2070 	rdir = fdp->fd_rdir;
2071 	fdp->fd_rdir = NULL;
2072 	jdir = fdp->fd_jdir;
2073 	fdp->fd_jdir = NULL;
2074 	FILEDESC_XUNLOCK(fdp);
2075 
2076 	if (cdir != NULL)
2077 		vrele(cdir);
2078 	if (rdir != NULL)
2079 		vrele(rdir);
2080 	if (jdir != NULL)
2081 		vrele(jdir);
2082 
2083 	fddrop(fdp);
2084 }
2085 
2086 /*
2087  * For setugid programs, we don't want to people to use that setugidness
2088  * to generate error messages which write to a file which otherwise would
2089  * otherwise be off-limits to the process.  We check for filesystems where
2090  * the vnode can change out from under us after execve (like [lin]procfs).
2091  *
2092  * Since setugidsafety calls this only for fd 0, 1 and 2, this check is
2093  * sufficient.  We also don't check for setugidness since we know we are.
2094  */
2095 static int
2096 is_unsafe(struct file *fp)
2097 {
2098 	if (fp->f_type == DTYPE_VNODE) {
2099 		struct vnode *vp = fp->f_vnode;
2100 
2101 		if ((vp->v_vflag & VV_PROCDEP) != 0)
2102 			return (1);
2103 	}
2104 	return (0);
2105 }
2106 
2107 /*
2108  * Make this setguid thing safe, if at all possible.
2109  */
2110 void
2111 setugidsafety(struct thread *td)
2112 {
2113 	struct filedesc *fdp;
2114 	struct file *fp;
2115 	int i;
2116 
2117 	/* Certain daemons might not have file descriptors. */
2118 	fdp = td->td_proc->p_fd;
2119 	if (fdp == NULL)
2120 		return;
2121 
2122 	/*
2123 	 * Note: fdp->fd_ofiles may be reallocated out from under us while
2124 	 * we are blocked in a close.  Be careful!
2125 	 */
2126 	FILEDESC_XLOCK(fdp);
2127 	for (i = 0; i <= fdp->fd_lastfile; i++) {
2128 		if (i > 2)
2129 			break;
2130 		fp = fdp->fd_ofiles[i].fde_file;
2131 		if (fp != NULL && is_unsafe(fp)) {
2132 			knote_fdclose(td, i);
2133 			/*
2134 			 * NULL-out descriptor prior to close to avoid
2135 			 * a race while close blocks.
2136 			 */
2137 			fdfree(fdp, i);
2138 			FILEDESC_XUNLOCK(fdp);
2139 			(void) closef(fp, td);
2140 			FILEDESC_XLOCK(fdp);
2141 		}
2142 	}
2143 	FILEDESC_XUNLOCK(fdp);
2144 }
2145 
2146 /*
2147  * If a specific file object occupies a specific file descriptor, close the
2148  * file descriptor entry and drop a reference on the file object.  This is a
2149  * convenience function to handle a subsequent error in a function that calls
2150  * falloc() that handles the race that another thread might have closed the
2151  * file descriptor out from under the thread creating the file object.
2152  */
2153 void
2154 fdclose(struct filedesc *fdp, struct file *fp, int idx, struct thread *td)
2155 {
2156 
2157 	FILEDESC_XLOCK(fdp);
2158 	if (fdp->fd_ofiles[idx].fde_file == fp) {
2159 		fdfree(fdp, idx);
2160 		FILEDESC_XUNLOCK(fdp);
2161 		fdrop(fp, td);
2162 	} else
2163 		FILEDESC_XUNLOCK(fdp);
2164 }
2165 
2166 /*
2167  * Close any files on exec?
2168  */
2169 void
2170 fdcloseexec(struct thread *td)
2171 {
2172 	struct filedesc *fdp;
2173 	struct filedescent *fde;
2174 	struct file *fp;
2175 	int i;
2176 
2177 	/* Certain daemons might not have file descriptors. */
2178 	fdp = td->td_proc->p_fd;
2179 	if (fdp == NULL)
2180 		return;
2181 
2182 	/*
2183 	 * We cannot cache fd_ofiles since operations
2184 	 * may block and rip them out from under us.
2185 	 */
2186 	FILEDESC_XLOCK(fdp);
2187 	for (i = 0; i <= fdp->fd_lastfile; i++) {
2188 		fde = &fdp->fd_ofiles[i];
2189 		fp = fde->fde_file;
2190 		if (fp != NULL && (fp->f_type == DTYPE_MQUEUE ||
2191 		    (fde->fde_flags & UF_EXCLOSE))) {
2192 			fdfree(fdp, i);
2193 			(void) closefp(fdp, i, fp, td, 0);
2194 			/* closefp() drops the FILEDESC lock. */
2195 			FILEDESC_XLOCK(fdp);
2196 		}
2197 	}
2198 	FILEDESC_XUNLOCK(fdp);
2199 }
2200 
2201 /*
2202  * It is unsafe for set[ug]id processes to be started with file
2203  * descriptors 0..2 closed, as these descriptors are given implicit
2204  * significance in the Standard C library.  fdcheckstd() will create a
2205  * descriptor referencing /dev/null for each of stdin, stdout, and
2206  * stderr that is not already open.
2207  */
2208 int
2209 fdcheckstd(struct thread *td)
2210 {
2211 	struct filedesc *fdp;
2212 	register_t retval, save;
2213 	int i, error, devnull;
2214 
2215 	fdp = td->td_proc->p_fd;
2216 	if (fdp == NULL)
2217 		return (0);
2218 	KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
2219 	devnull = -1;
2220 	error = 0;
2221 	for (i = 0; i < 3; i++) {
2222 		if (fdp->fd_ofiles[i].fde_file != NULL)
2223 			continue;
2224 		if (devnull < 0) {
2225 			save = td->td_retval[0];
2226 			error = kern_open(td, "/dev/null", UIO_SYSSPACE,
2227 			    O_RDWR, 0);
2228 			devnull = td->td_retval[0];
2229 			td->td_retval[0] = save;
2230 			if (error)
2231 				break;
2232 			KASSERT(devnull == i, ("oof, we didn't get our fd"));
2233 		} else {
2234 			error = do_dup(td, DUP_FIXED, devnull, i, &retval);
2235 			if (error != 0)
2236 				break;
2237 		}
2238 	}
2239 	return (error);
2240 }
2241 
2242 /*
2243  * Internal form of close.  Decrement reference count on file structure.
2244  * Note: td may be NULL when closing a file that was being passed in a
2245  * message.
2246  *
2247  * XXXRW: Giant is not required for the caller, but often will be held; this
2248  * makes it moderately likely the Giant will be recursed in the VFS case.
2249  */
2250 int
2251 closef(struct file *fp, struct thread *td)
2252 {
2253 	struct vnode *vp;
2254 	struct flock lf;
2255 	struct filedesc_to_leader *fdtol;
2256 	struct filedesc *fdp;
2257 
2258 	/*
2259 	 * POSIX record locking dictates that any close releases ALL
2260 	 * locks owned by this process.  This is handled by setting
2261 	 * a flag in the unlock to free ONLY locks obeying POSIX
2262 	 * semantics, and not to free BSD-style file locks.
2263 	 * If the descriptor was in a message, POSIX-style locks
2264 	 * aren't passed with the descriptor, and the thread pointer
2265 	 * will be NULL.  Callers should be careful only to pass a
2266 	 * NULL thread pointer when there really is no owning
2267 	 * context that might have locks, or the locks will be
2268 	 * leaked.
2269 	 */
2270 	if (fp->f_type == DTYPE_VNODE && td != NULL) {
2271 		vp = fp->f_vnode;
2272 		if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
2273 			lf.l_whence = SEEK_SET;
2274 			lf.l_start = 0;
2275 			lf.l_len = 0;
2276 			lf.l_type = F_UNLCK;
2277 			(void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader,
2278 			    F_UNLCK, &lf, F_POSIX);
2279 		}
2280 		fdtol = td->td_proc->p_fdtol;
2281 		if (fdtol != NULL) {
2282 			/*
2283 			 * Handle special case where file descriptor table is
2284 			 * shared between multiple process leaders.
2285 			 */
2286 			fdp = td->td_proc->p_fd;
2287 			FILEDESC_XLOCK(fdp);
2288 			for (fdtol = fdtol->fdl_next;
2289 			     fdtol != td->td_proc->p_fdtol;
2290 			     fdtol = fdtol->fdl_next) {
2291 				if ((fdtol->fdl_leader->p_flag &
2292 				     P_ADVLOCK) == 0)
2293 					continue;
2294 				fdtol->fdl_holdcount++;
2295 				FILEDESC_XUNLOCK(fdp);
2296 				lf.l_whence = SEEK_SET;
2297 				lf.l_start = 0;
2298 				lf.l_len = 0;
2299 				lf.l_type = F_UNLCK;
2300 				vp = fp->f_vnode;
2301 				(void) VOP_ADVLOCK(vp,
2302 				    (caddr_t)fdtol->fdl_leader, F_UNLCK, &lf,
2303 				    F_POSIX);
2304 				FILEDESC_XLOCK(fdp);
2305 				fdtol->fdl_holdcount--;
2306 				if (fdtol->fdl_holdcount == 0 &&
2307 				    fdtol->fdl_wakeup != 0) {
2308 					fdtol->fdl_wakeup = 0;
2309 					wakeup(fdtol);
2310 				}
2311 			}
2312 			FILEDESC_XUNLOCK(fdp);
2313 		}
2314 	}
2315 	return (fdrop(fp, td));
2316 }
2317 
2318 /*
2319  * Initialize the file pointer with the specified properties.
2320  *
2321  * The ops are set with release semantics to be certain that the flags, type,
2322  * and data are visible when ops is.  This is to prevent ops methods from being
2323  * called with bad data.
2324  */
2325 void
2326 finit(struct file *fp, u_int flag, short type, void *data, struct fileops *ops)
2327 {
2328 	fp->f_data = data;
2329 	fp->f_flag = flag;
2330 	fp->f_type = type;
2331 	atomic_store_rel_ptr((volatile uintptr_t *)&fp->f_ops, (uintptr_t)ops);
2332 }
2333 
2334 int
2335 fget_unlocked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp,
2336     int needfcntl, struct file **fpp, cap_rights_t *haverightsp)
2337 {
2338 	struct file *fp;
2339 	u_int count;
2340 #ifdef CAPABILITIES
2341 	cap_rights_t haverights;
2342 	int error;
2343 #endif
2344 
2345 	/*
2346 	 * Avoid reads reordering and then a first access to the
2347 	 * fdp->fd_ofiles table which could result in OOB operation.
2348 	 */
2349 	if (fd < 0 || fd >= atomic_load_acq_int(&fdp->fd_nfiles))
2350 		return (EBADF);
2351 	/*
2352 	 * Fetch the descriptor locklessly.  We avoid fdrop() races by
2353 	 * never raising a refcount above 0.  To accomplish this we have
2354 	 * to use a cmpset loop rather than an atomic_add.  The descriptor
2355 	 * must be re-verified once we acquire a reference to be certain
2356 	 * that the identity is still correct and we did not lose a race
2357 	 * due to preemption.
2358 	 */
2359 	for (;;) {
2360 		fp = fdp->fd_ofiles[fd].fde_file;
2361 		if (fp == NULL)
2362 			return (EBADF);
2363 #ifdef CAPABILITIES
2364 		haverights = *cap_rights(fdp, fd);
2365 		if (needrightsp != NULL) {
2366 			error = cap_check(&haverights, needrightsp);
2367 			if (error != 0)
2368 				return (error);
2369 			if (cap_rights_is_set(needrightsp, CAP_FCNTL)) {
2370 				error = cap_fcntl_check(fdp, fd, needfcntl);
2371 				if (error != 0)
2372 					return (error);
2373 			}
2374 		}
2375 #endif
2376 		count = fp->f_count;
2377 		if (count == 0)
2378 			continue;
2379 		/*
2380 		 * Use an acquire barrier to prevent caching of fd_ofiles
2381 		 * so it is refreshed for verification.
2382 		 */
2383 		if (atomic_cmpset_acq_int(&fp->f_count, count, count + 1) != 1)
2384 			continue;
2385 		if (fp == fdp->fd_ofiles[fd].fde_file)
2386 			break;
2387 		fdrop(fp, curthread);
2388 	}
2389 	*fpp = fp;
2390 	if (haverightsp != NULL) {
2391 #ifdef CAPABILITIES
2392 		*haverightsp = haverights;
2393 #else
2394 		CAP_ALL(haverightsp);
2395 #endif
2396 	}
2397 	return (0);
2398 }
2399 
2400 /*
2401  * Extract the file pointer associated with the specified descriptor for the
2402  * current user process.
2403  *
2404  * If the descriptor doesn't exist or doesn't match 'flags', EBADF is
2405  * returned.
2406  *
2407  * File's rights will be checked against the capability rights mask.
2408  *
2409  * If an error occured the non-zero error is returned and *fpp is set to
2410  * NULL.  Otherwise *fpp is held and set and zero is returned.  Caller is
2411  * responsible for fdrop().
2412  */
2413 static __inline int
2414 _fget(struct thread *td, int fd, struct file **fpp, int flags,
2415     cap_rights_t *needrightsp, u_char *maxprotp)
2416 {
2417 	struct filedesc *fdp;
2418 	struct file *fp;
2419 	cap_rights_t haverights, needrights;
2420 	int error;
2421 
2422 	*fpp = NULL;
2423 	if (td == NULL || (fdp = td->td_proc->p_fd) == NULL)
2424 		return (EBADF);
2425 	if (needrightsp != NULL)
2426 		needrights = *needrightsp;
2427 	else
2428 		cap_rights_init(&needrights);
2429 	if (maxprotp != NULL)
2430 		cap_rights_set(&needrights, CAP_MMAP);
2431 	error = fget_unlocked(fdp, fd, &needrights, 0, &fp, &haverights);
2432 	if (error != 0)
2433 		return (error);
2434 	if (fp->f_ops == &badfileops) {
2435 		fdrop(fp, td);
2436 		return (EBADF);
2437 	}
2438 
2439 #ifdef CAPABILITIES
2440 	/*
2441 	 * If requested, convert capability rights to access flags.
2442 	 */
2443 	if (maxprotp != NULL)
2444 		*maxprotp = cap_rights_to_vmprot(&haverights);
2445 #else /* !CAPABILITIES */
2446 	if (maxprotp != NULL)
2447 		*maxprotp = VM_PROT_ALL;
2448 #endif /* CAPABILITIES */
2449 
2450 	/*
2451 	 * FREAD and FWRITE failure return EBADF as per POSIX.
2452 	 */
2453 	error = 0;
2454 	switch (flags) {
2455 	case FREAD:
2456 	case FWRITE:
2457 		if ((fp->f_flag & flags) == 0)
2458 			error = EBADF;
2459 		break;
2460 	case FEXEC:
2461 	    	if ((fp->f_flag & (FREAD | FEXEC)) == 0 ||
2462 		    ((fp->f_flag & FWRITE) != 0))
2463 			error = EBADF;
2464 		break;
2465 	case 0:
2466 		break;
2467 	default:
2468 		KASSERT(0, ("wrong flags"));
2469 	}
2470 
2471 	if (error != 0) {
2472 		fdrop(fp, td);
2473 		return (error);
2474 	}
2475 
2476 	*fpp = fp;
2477 	return (0);
2478 }
2479 
2480 int
2481 fget(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
2482 {
2483 
2484 	return(_fget(td, fd, fpp, 0, rightsp, NULL));
2485 }
2486 
2487 int
2488 fget_mmap(struct thread *td, int fd, cap_rights_t *rightsp, u_char *maxprotp,
2489     struct file **fpp)
2490 {
2491 
2492 	return (_fget(td, fd, fpp, 0, rightsp, maxprotp));
2493 }
2494 
2495 int
2496 fget_read(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
2497 {
2498 
2499 	return(_fget(td, fd, fpp, FREAD, rightsp, NULL));
2500 }
2501 
2502 int
2503 fget_write(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
2504 {
2505 
2506 	return (_fget(td, fd, fpp, FWRITE, rightsp, NULL));
2507 }
2508 
2509 /*
2510  * Like fget() but loads the underlying vnode, or returns an error if the
2511  * descriptor does not represent a vnode.  Note that pipes use vnodes but
2512  * never have VM objects.  The returned vnode will be vref()'d.
2513  *
2514  * XXX: what about the unused flags ?
2515  */
2516 static __inline int
2517 _fgetvp(struct thread *td, int fd, int flags, cap_rights_t *needrightsp,
2518     struct vnode **vpp)
2519 {
2520 	struct file *fp;
2521 	int error;
2522 
2523 	*vpp = NULL;
2524 	error = _fget(td, fd, &fp, flags, needrightsp, NULL);
2525 	if (error != 0)
2526 		return (error);
2527 	if (fp->f_vnode == NULL) {
2528 		error = EINVAL;
2529 	} else {
2530 		*vpp = fp->f_vnode;
2531 		vref(*vpp);
2532 	}
2533 	fdrop(fp, td);
2534 
2535 	return (error);
2536 }
2537 
2538 int
2539 fgetvp(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
2540 {
2541 
2542 	return (_fgetvp(td, fd, 0, rightsp, vpp));
2543 }
2544 
2545 int
2546 fgetvp_rights(struct thread *td, int fd, cap_rights_t *needrightsp,
2547     struct filecaps *havecaps, struct vnode **vpp)
2548 {
2549 	struct filedesc *fdp;
2550 	struct file *fp;
2551 #ifdef CAPABILITIES
2552 	int error;
2553 #endif
2554 
2555 	if (td == NULL || (fdp = td->td_proc->p_fd) == NULL)
2556 		return (EBADF);
2557 
2558 	fp = fget_locked(fdp, fd);
2559 	if (fp == NULL || fp->f_ops == &badfileops)
2560 		return (EBADF);
2561 
2562 #ifdef CAPABILITIES
2563 	if (needrightsp != NULL) {
2564 		error = cap_check(cap_rights(fdp, fd), needrightsp);
2565 		if (error != 0)
2566 			return (error);
2567 	}
2568 #endif
2569 
2570 	if (fp->f_vnode == NULL)
2571 		return (EINVAL);
2572 
2573 	*vpp = fp->f_vnode;
2574 	vref(*vpp);
2575 	filecaps_copy(&fdp->fd_ofiles[fd].fde_caps, havecaps);
2576 
2577 	return (0);
2578 }
2579 
2580 int
2581 fgetvp_read(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
2582 {
2583 
2584 	return (_fgetvp(td, fd, FREAD, rightsp, vpp));
2585 }
2586 
2587 int
2588 fgetvp_exec(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp)
2589 {
2590 
2591 	return (_fgetvp(td, fd, FEXEC, rightsp, vpp));
2592 }
2593 
2594 #ifdef notyet
2595 int
2596 fgetvp_write(struct thread *td, int fd, cap_rights_t *rightsp,
2597     struct vnode **vpp)
2598 {
2599 
2600 	return (_fgetvp(td, fd, FWRITE, rightsp, vpp));
2601 }
2602 #endif
2603 
2604 /*
2605  * Like fget() but loads the underlying socket, or returns an error if the
2606  * descriptor does not represent a socket.
2607  *
2608  * We bump the ref count on the returned socket.  XXX Also obtain the SX lock
2609  * in the future.
2610  *
2611  * Note: fgetsock() and fputsock() are deprecated, as consumers should rely
2612  * on their file descriptor reference to prevent the socket from being free'd
2613  * during use.
2614  */
2615 int
2616 fgetsock(struct thread *td, int fd, cap_rights_t *rightsp, struct socket **spp,
2617     u_int *fflagp)
2618 {
2619 	struct file *fp;
2620 	int error;
2621 
2622 	*spp = NULL;
2623 	if (fflagp != NULL)
2624 		*fflagp = 0;
2625 	if ((error = _fget(td, fd, &fp, 0, rightsp, NULL)) != 0)
2626 		return (error);
2627 	if (fp->f_type != DTYPE_SOCKET) {
2628 		error = ENOTSOCK;
2629 	} else {
2630 		*spp = fp->f_data;
2631 		if (fflagp)
2632 			*fflagp = fp->f_flag;
2633 		SOCK_LOCK(*spp);
2634 		soref(*spp);
2635 		SOCK_UNLOCK(*spp);
2636 	}
2637 	fdrop(fp, td);
2638 
2639 	return (error);
2640 }
2641 
2642 /*
2643  * Drop the reference count on the socket and XXX release the SX lock in the
2644  * future.  The last reference closes the socket.
2645  *
2646  * Note: fputsock() is deprecated, see comment for fgetsock().
2647  */
2648 void
2649 fputsock(struct socket *so)
2650 {
2651 
2652 	ACCEPT_LOCK();
2653 	SOCK_LOCK(so);
2654 	CURVNET_SET(so->so_vnet);
2655 	sorele(so);
2656 	CURVNET_RESTORE();
2657 }
2658 
2659 /*
2660  * Handle the last reference to a file being closed.
2661  */
2662 int
2663 _fdrop(struct file *fp, struct thread *td)
2664 {
2665 	int error;
2666 
2667 	error = 0;
2668 	if (fp->f_count != 0)
2669 		panic("fdrop: count %d", fp->f_count);
2670 	if (fp->f_ops != &badfileops)
2671 		error = fo_close(fp, td);
2672 	atomic_subtract_int(&openfiles, 1);
2673 	crfree(fp->f_cred);
2674 	free(fp->f_advice, M_FADVISE);
2675 	uma_zfree(file_zone, fp);
2676 
2677 	return (error);
2678 }
2679 
2680 /*
2681  * Apply an advisory lock on a file descriptor.
2682  *
2683  * Just attempt to get a record lock of the requested type on the entire file
2684  * (l_whence = SEEK_SET, l_start = 0, l_len = 0).
2685  */
2686 #ifndef _SYS_SYSPROTO_H_
2687 struct flock_args {
2688 	int	fd;
2689 	int	how;
2690 };
2691 #endif
2692 /* ARGSUSED */
2693 int
2694 sys_flock(struct thread *td, struct flock_args *uap)
2695 {
2696 	struct file *fp;
2697 	struct vnode *vp;
2698 	struct flock lf;
2699 	cap_rights_t rights;
2700 	int error;
2701 
2702 	error = fget(td, uap->fd, cap_rights_init(&rights, CAP_FLOCK), &fp);
2703 	if (error != 0)
2704 		return (error);
2705 	if (fp->f_type != DTYPE_VNODE) {
2706 		fdrop(fp, td);
2707 		return (EOPNOTSUPP);
2708 	}
2709 
2710 	vp = fp->f_vnode;
2711 	lf.l_whence = SEEK_SET;
2712 	lf.l_start = 0;
2713 	lf.l_len = 0;
2714 	if (uap->how & LOCK_UN) {
2715 		lf.l_type = F_UNLCK;
2716 		atomic_clear_int(&fp->f_flag, FHASLOCK);
2717 		error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK);
2718 		goto done2;
2719 	}
2720 	if (uap->how & LOCK_EX)
2721 		lf.l_type = F_WRLCK;
2722 	else if (uap->how & LOCK_SH)
2723 		lf.l_type = F_RDLCK;
2724 	else {
2725 		error = EBADF;
2726 		goto done2;
2727 	}
2728 	atomic_set_int(&fp->f_flag, FHASLOCK);
2729 	error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf,
2730 	    (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT);
2731 done2:
2732 	fdrop(fp, td);
2733 	return (error);
2734 }
2735 /*
2736  * Duplicate the specified descriptor to a free descriptor.
2737  */
2738 int
2739 dupfdopen(struct thread *td, struct filedesc *fdp, int dfd, int mode,
2740     int openerror, int *indxp)
2741 {
2742 	struct file *fp;
2743 	int error, indx;
2744 
2745 	KASSERT(openerror == ENODEV || openerror == ENXIO,
2746 	    ("unexpected error %d in %s", openerror, __func__));
2747 
2748 	/*
2749 	 * If the to-be-dup'd fd number is greater than the allowed number
2750 	 * of file descriptors, or the fd to be dup'd has already been
2751 	 * closed, then reject.
2752 	 */
2753 	FILEDESC_XLOCK(fdp);
2754 	if ((fp = fget_locked(fdp, dfd)) == NULL) {
2755 		FILEDESC_XUNLOCK(fdp);
2756 		return (EBADF);
2757 	}
2758 
2759 	error = fdalloc(td, 0, &indx);
2760 	if (error != 0) {
2761 		FILEDESC_XUNLOCK(fdp);
2762 		return (error);
2763 	}
2764 
2765 	/*
2766 	 * There are two cases of interest here.
2767 	 *
2768 	 * For ENODEV simply dup (dfd) to file descriptor (indx) and return.
2769 	 *
2770 	 * For ENXIO steal away the file structure from (dfd) and store it in
2771 	 * (indx).  (dfd) is effectively closed by this operation.
2772 	 */
2773 	switch (openerror) {
2774 	case ENODEV:
2775 		/*
2776 		 * Check that the mode the file is being opened for is a
2777 		 * subset of the mode of the existing descriptor.
2778 		 */
2779 		if (((mode & (FREAD|FWRITE)) | fp->f_flag) != fp->f_flag) {
2780 			fdunused(fdp, indx);
2781 			FILEDESC_XUNLOCK(fdp);
2782 			return (EACCES);
2783 		}
2784 		fhold(fp);
2785 		fdp->fd_ofiles[indx] = fdp->fd_ofiles[dfd];
2786 		filecaps_copy(&fdp->fd_ofiles[dfd].fde_caps,
2787 		    &fdp->fd_ofiles[indx].fde_caps);
2788 		break;
2789 	case ENXIO:
2790 		/*
2791 		 * Steal away the file pointer from dfd and stuff it into indx.
2792 		 */
2793 		fdp->fd_ofiles[indx] = fdp->fd_ofiles[dfd];
2794 		bzero(&fdp->fd_ofiles[dfd], sizeof(fdp->fd_ofiles[dfd]));
2795 		fdunused(fdp, dfd);
2796 		break;
2797 	}
2798 	FILEDESC_XUNLOCK(fdp);
2799 	*indxp = indx;
2800 	return (0);
2801 }
2802 
2803 /*
2804  * Scan all active processes and prisons to see if any of them have a current
2805  * or root directory of `olddp'. If so, replace them with the new mount point.
2806  */
2807 void
2808 mountcheckdirs(struct vnode *olddp, struct vnode *newdp)
2809 {
2810 	struct filedesc *fdp;
2811 	struct prison *pr;
2812 	struct proc *p;
2813 	int nrele;
2814 
2815 	if (vrefcnt(olddp) == 1)
2816 		return;
2817 	nrele = 0;
2818 	sx_slock(&allproc_lock);
2819 	FOREACH_PROC_IN_SYSTEM(p) {
2820 		fdp = fdhold(p);
2821 		if (fdp == NULL)
2822 			continue;
2823 		FILEDESC_XLOCK(fdp);
2824 		if (fdp->fd_cdir == olddp) {
2825 			vref(newdp);
2826 			fdp->fd_cdir = newdp;
2827 			nrele++;
2828 		}
2829 		if (fdp->fd_rdir == olddp) {
2830 			vref(newdp);
2831 			fdp->fd_rdir = newdp;
2832 			nrele++;
2833 		}
2834 		if (fdp->fd_jdir == olddp) {
2835 			vref(newdp);
2836 			fdp->fd_jdir = newdp;
2837 			nrele++;
2838 		}
2839 		FILEDESC_XUNLOCK(fdp);
2840 		fddrop(fdp);
2841 	}
2842 	sx_sunlock(&allproc_lock);
2843 	if (rootvnode == olddp) {
2844 		vref(newdp);
2845 		rootvnode = newdp;
2846 		nrele++;
2847 	}
2848 	mtx_lock(&prison0.pr_mtx);
2849 	if (prison0.pr_root == olddp) {
2850 		vref(newdp);
2851 		prison0.pr_root = newdp;
2852 		nrele++;
2853 	}
2854 	mtx_unlock(&prison0.pr_mtx);
2855 	sx_slock(&allprison_lock);
2856 	TAILQ_FOREACH(pr, &allprison, pr_list) {
2857 		mtx_lock(&pr->pr_mtx);
2858 		if (pr->pr_root == olddp) {
2859 			vref(newdp);
2860 			pr->pr_root = newdp;
2861 			nrele++;
2862 		}
2863 		mtx_unlock(&pr->pr_mtx);
2864 	}
2865 	sx_sunlock(&allprison_lock);
2866 	while (nrele--)
2867 		vrele(olddp);
2868 }
2869 
2870 struct filedesc_to_leader *
2871 filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader)
2872 {
2873 	struct filedesc_to_leader *fdtol;
2874 
2875 	fdtol = malloc(sizeof(struct filedesc_to_leader),
2876 	       M_FILEDESC_TO_LEADER,
2877 	       M_WAITOK);
2878 	fdtol->fdl_refcount = 1;
2879 	fdtol->fdl_holdcount = 0;
2880 	fdtol->fdl_wakeup = 0;
2881 	fdtol->fdl_leader = leader;
2882 	if (old != NULL) {
2883 		FILEDESC_XLOCK(fdp);
2884 		fdtol->fdl_next = old->fdl_next;
2885 		fdtol->fdl_prev = old;
2886 		old->fdl_next = fdtol;
2887 		fdtol->fdl_next->fdl_prev = fdtol;
2888 		FILEDESC_XUNLOCK(fdp);
2889 	} else {
2890 		fdtol->fdl_next = fdtol;
2891 		fdtol->fdl_prev = fdtol;
2892 	}
2893 	return (fdtol);
2894 }
2895 
2896 /*
2897  * Get file structures globally.
2898  */
2899 static int
2900 sysctl_kern_file(SYSCTL_HANDLER_ARGS)
2901 {
2902 	struct xfile xf;
2903 	struct filedesc *fdp;
2904 	struct file *fp;
2905 	struct proc *p;
2906 	int error, n;
2907 
2908 	error = sysctl_wire_old_buffer(req, 0);
2909 	if (error != 0)
2910 		return (error);
2911 	if (req->oldptr == NULL) {
2912 		n = 0;
2913 		sx_slock(&allproc_lock);
2914 		FOREACH_PROC_IN_SYSTEM(p) {
2915 			if (p->p_state == PRS_NEW)
2916 				continue;
2917 			fdp = fdhold(p);
2918 			if (fdp == NULL)
2919 				continue;
2920 			/* overestimates sparse tables. */
2921 			if (fdp->fd_lastfile > 0)
2922 				n += fdp->fd_lastfile;
2923 			fddrop(fdp);
2924 		}
2925 		sx_sunlock(&allproc_lock);
2926 		return (SYSCTL_OUT(req, 0, n * sizeof(xf)));
2927 	}
2928 	error = 0;
2929 	bzero(&xf, sizeof(xf));
2930 	xf.xf_size = sizeof(xf);
2931 	sx_slock(&allproc_lock);
2932 	FOREACH_PROC_IN_SYSTEM(p) {
2933 		PROC_LOCK(p);
2934 		if (p->p_state == PRS_NEW) {
2935 			PROC_UNLOCK(p);
2936 			continue;
2937 		}
2938 		if (p_cansee(req->td, p) != 0) {
2939 			PROC_UNLOCK(p);
2940 			continue;
2941 		}
2942 		xf.xf_pid = p->p_pid;
2943 		xf.xf_uid = p->p_ucred->cr_uid;
2944 		PROC_UNLOCK(p);
2945 		fdp = fdhold(p);
2946 		if (fdp == NULL)
2947 			continue;
2948 		FILEDESC_SLOCK(fdp);
2949 		for (n = 0; fdp->fd_refcnt > 0 && n < fdp->fd_nfiles; ++n) {
2950 			if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
2951 				continue;
2952 			xf.xf_fd = n;
2953 			xf.xf_file = fp;
2954 			xf.xf_data = fp->f_data;
2955 			xf.xf_vnode = fp->f_vnode;
2956 			xf.xf_type = fp->f_type;
2957 			xf.xf_count = fp->f_count;
2958 			xf.xf_msgcount = 0;
2959 			xf.xf_offset = foffset_get(fp);
2960 			xf.xf_flag = fp->f_flag;
2961 			error = SYSCTL_OUT(req, &xf, sizeof(xf));
2962 			if (error)
2963 				break;
2964 		}
2965 		FILEDESC_SUNLOCK(fdp);
2966 		fddrop(fdp);
2967 		if (error)
2968 			break;
2969 	}
2970 	sx_sunlock(&allproc_lock);
2971 	return (error);
2972 }
2973 
2974 SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD,
2975     0, 0, sysctl_kern_file, "S,xfile", "Entire file table");
2976 
2977 #ifdef KINFO_OFILE_SIZE
2978 CTASSERT(sizeof(struct kinfo_ofile) == KINFO_OFILE_SIZE);
2979 #endif
2980 
2981 #ifdef COMPAT_FREEBSD7
2982 static int
2983 export_vnode_for_osysctl(struct vnode *vp, int type,
2984     struct kinfo_ofile *kif, struct filedesc *fdp, struct sysctl_req *req)
2985 {
2986 	int error;
2987 	char *fullpath, *freepath;
2988 
2989 	bzero(kif, sizeof(*kif));
2990 	kif->kf_structsize = sizeof(*kif);
2991 
2992 	vref(vp);
2993 	kif->kf_fd = type;
2994 	kif->kf_type = KF_TYPE_VNODE;
2995 	/* This function only handles directories. */
2996 	if (vp->v_type != VDIR) {
2997 		vrele(vp);
2998 		return (ENOTDIR);
2999 	}
3000 	kif->kf_vnode_type = KF_VTYPE_VDIR;
3001 
3002 	/*
3003 	 * This is not a true file descriptor, so we set a bogus refcount
3004 	 * and offset to indicate these fields should be ignored.
3005 	 */
3006 	kif->kf_ref_count = -1;
3007 	kif->kf_offset = -1;
3008 
3009 	freepath = NULL;
3010 	fullpath = "-";
3011 	FILEDESC_SUNLOCK(fdp);
3012 	vn_fullpath(curthread, vp, &fullpath, &freepath);
3013 	vrele(vp);
3014 	strlcpy(kif->kf_path, fullpath, sizeof(kif->kf_path));
3015 	if (freepath != NULL)
3016 		free(freepath, M_TEMP);
3017 	error = SYSCTL_OUT(req, kif, sizeof(*kif));
3018 	FILEDESC_SLOCK(fdp);
3019 	return (error);
3020 }
3021 
3022 /*
3023  * Get per-process file descriptors for use by procstat(1), et al.
3024  */
3025 static int
3026 sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS)
3027 {
3028 	char *fullpath, *freepath;
3029 	struct kinfo_ofile *kif;
3030 	struct filedesc *fdp;
3031 	int error, i, *name;
3032 	struct shmfd *shmfd;
3033 	struct socket *so;
3034 	struct vnode *vp;
3035 	struct ksem *ks;
3036 	struct file *fp;
3037 	struct proc *p;
3038 	struct tty *tp;
3039 
3040 	name = (int *)arg1;
3041 	error = pget((pid_t)name[0], PGET_CANDEBUG, &p);
3042 	if (error != 0)
3043 		return (error);
3044 	fdp = fdhold(p);
3045 	PROC_UNLOCK(p);
3046 	if (fdp == NULL)
3047 		return (ENOENT);
3048 	kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK);
3049 	FILEDESC_SLOCK(fdp);
3050 	if (fdp->fd_cdir != NULL)
3051 		export_vnode_for_osysctl(fdp->fd_cdir, KF_FD_TYPE_CWD, kif,
3052 				fdp, req);
3053 	if (fdp->fd_rdir != NULL)
3054 		export_vnode_for_osysctl(fdp->fd_rdir, KF_FD_TYPE_ROOT, kif,
3055 				fdp, req);
3056 	if (fdp->fd_jdir != NULL)
3057 		export_vnode_for_osysctl(fdp->fd_jdir, KF_FD_TYPE_JAIL, kif,
3058 				fdp, req);
3059 	for (i = 0; i < fdp->fd_nfiles; i++) {
3060 		if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
3061 			continue;
3062 		bzero(kif, sizeof(*kif));
3063 		kif->kf_structsize = sizeof(*kif);
3064 		ks = NULL;
3065 		vp = NULL;
3066 		so = NULL;
3067 		tp = NULL;
3068 		shmfd = NULL;
3069 		kif->kf_fd = i;
3070 
3071 		switch (fp->f_type) {
3072 		case DTYPE_VNODE:
3073 			kif->kf_type = KF_TYPE_VNODE;
3074 			vp = fp->f_vnode;
3075 			break;
3076 
3077 		case DTYPE_SOCKET:
3078 			kif->kf_type = KF_TYPE_SOCKET;
3079 			so = fp->f_data;
3080 			break;
3081 
3082 		case DTYPE_PIPE:
3083 			kif->kf_type = KF_TYPE_PIPE;
3084 			break;
3085 
3086 		case DTYPE_FIFO:
3087 			kif->kf_type = KF_TYPE_FIFO;
3088 			vp = fp->f_vnode;
3089 			break;
3090 
3091 		case DTYPE_KQUEUE:
3092 			kif->kf_type = KF_TYPE_KQUEUE;
3093 			break;
3094 
3095 		case DTYPE_CRYPTO:
3096 			kif->kf_type = KF_TYPE_CRYPTO;
3097 			break;
3098 
3099 		case DTYPE_MQUEUE:
3100 			kif->kf_type = KF_TYPE_MQUEUE;
3101 			break;
3102 
3103 		case DTYPE_SHM:
3104 			kif->kf_type = KF_TYPE_SHM;
3105 			shmfd = fp->f_data;
3106 			break;
3107 
3108 		case DTYPE_SEM:
3109 			kif->kf_type = KF_TYPE_SEM;
3110 			ks = fp->f_data;
3111 			break;
3112 
3113 		case DTYPE_PTS:
3114 			kif->kf_type = KF_TYPE_PTS;
3115 			tp = fp->f_data;
3116 			break;
3117 
3118 		case DTYPE_PROCDESC:
3119 			kif->kf_type = KF_TYPE_PROCDESC;
3120 			break;
3121 
3122 		default:
3123 			kif->kf_type = KF_TYPE_UNKNOWN;
3124 			break;
3125 		}
3126 		kif->kf_ref_count = fp->f_count;
3127 		if (fp->f_flag & FREAD)
3128 			kif->kf_flags |= KF_FLAG_READ;
3129 		if (fp->f_flag & FWRITE)
3130 			kif->kf_flags |= KF_FLAG_WRITE;
3131 		if (fp->f_flag & FAPPEND)
3132 			kif->kf_flags |= KF_FLAG_APPEND;
3133 		if (fp->f_flag & FASYNC)
3134 			kif->kf_flags |= KF_FLAG_ASYNC;
3135 		if (fp->f_flag & FFSYNC)
3136 			kif->kf_flags |= KF_FLAG_FSYNC;
3137 		if (fp->f_flag & FNONBLOCK)
3138 			kif->kf_flags |= KF_FLAG_NONBLOCK;
3139 		if (fp->f_flag & O_DIRECT)
3140 			kif->kf_flags |= KF_FLAG_DIRECT;
3141 		if (fp->f_flag & FHASLOCK)
3142 			kif->kf_flags |= KF_FLAG_HASLOCK;
3143 		kif->kf_offset = foffset_get(fp);
3144 		if (vp != NULL) {
3145 			vref(vp);
3146 			switch (vp->v_type) {
3147 			case VNON:
3148 				kif->kf_vnode_type = KF_VTYPE_VNON;
3149 				break;
3150 			case VREG:
3151 				kif->kf_vnode_type = KF_VTYPE_VREG;
3152 				break;
3153 			case VDIR:
3154 				kif->kf_vnode_type = KF_VTYPE_VDIR;
3155 				break;
3156 			case VBLK:
3157 				kif->kf_vnode_type = KF_VTYPE_VBLK;
3158 				break;
3159 			case VCHR:
3160 				kif->kf_vnode_type = KF_VTYPE_VCHR;
3161 				break;
3162 			case VLNK:
3163 				kif->kf_vnode_type = KF_VTYPE_VLNK;
3164 				break;
3165 			case VSOCK:
3166 				kif->kf_vnode_type = KF_VTYPE_VSOCK;
3167 				break;
3168 			case VFIFO:
3169 				kif->kf_vnode_type = KF_VTYPE_VFIFO;
3170 				break;
3171 			case VBAD:
3172 				kif->kf_vnode_type = KF_VTYPE_VBAD;
3173 				break;
3174 			default:
3175 				kif->kf_vnode_type = KF_VTYPE_UNKNOWN;
3176 				break;
3177 			}
3178 			/*
3179 			 * It is OK to drop the filedesc lock here as we will
3180 			 * re-validate and re-evaluate its properties when
3181 			 * the loop continues.
3182 			 */
3183 			freepath = NULL;
3184 			fullpath = "-";
3185 			FILEDESC_SUNLOCK(fdp);
3186 			vn_fullpath(curthread, vp, &fullpath, &freepath);
3187 			vrele(vp);
3188 			strlcpy(kif->kf_path, fullpath,
3189 			    sizeof(kif->kf_path));
3190 			if (freepath != NULL)
3191 				free(freepath, M_TEMP);
3192 			FILEDESC_SLOCK(fdp);
3193 		}
3194 		if (so != NULL) {
3195 			struct sockaddr *sa;
3196 
3197 			if (so->so_proto->pr_usrreqs->pru_sockaddr(so, &sa)
3198 			    == 0 && sa->sa_len <= sizeof(kif->kf_sa_local)) {
3199 				bcopy(sa, &kif->kf_sa_local, sa->sa_len);
3200 				free(sa, M_SONAME);
3201 			}
3202 			if (so->so_proto->pr_usrreqs->pru_peeraddr(so, &sa)
3203 			    == 0 && sa->sa_len <= sizeof(kif->kf_sa_peer)) {
3204 				bcopy(sa, &kif->kf_sa_peer, sa->sa_len);
3205 				free(sa, M_SONAME);
3206 			}
3207 			kif->kf_sock_domain =
3208 			    so->so_proto->pr_domain->dom_family;
3209 			kif->kf_sock_type = so->so_type;
3210 			kif->kf_sock_protocol = so->so_proto->pr_protocol;
3211 		}
3212 		if (tp != NULL) {
3213 			strlcpy(kif->kf_path, tty_devname(tp),
3214 			    sizeof(kif->kf_path));
3215 		}
3216 		if (shmfd != NULL)
3217 			shm_path(shmfd, kif->kf_path, sizeof(kif->kf_path));
3218 		if (ks != NULL && ksem_info != NULL)
3219 			ksem_info(ks, kif->kf_path, sizeof(kif->kf_path), NULL);
3220 		error = SYSCTL_OUT(req, kif, sizeof(*kif));
3221 		if (error)
3222 			break;
3223 	}
3224 	FILEDESC_SUNLOCK(fdp);
3225 	fddrop(fdp);
3226 	free(kif, M_TEMP);
3227 	return (0);
3228 }
3229 
3230 static SYSCTL_NODE(_kern_proc, KERN_PROC_OFILEDESC, ofiledesc, CTLFLAG_RD,
3231     sysctl_kern_proc_ofiledesc, "Process ofiledesc entries");
3232 #endif	/* COMPAT_FREEBSD7 */
3233 
3234 #ifdef KINFO_FILE_SIZE
3235 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
3236 #endif
3237 
3238 struct export_fd_buf {
3239 	struct filedesc		*fdp;
3240 	struct sbuf 		*sb;
3241 	ssize_t			remainder;
3242 	struct kinfo_file	kif;
3243 };
3244 
3245 static int
3246 export_fd_to_sb(void *data, int type, int fd, int fflags, int refcnt,
3247     int64_t offset, cap_rights_t *rightsp, struct export_fd_buf *efbuf)
3248 {
3249 	struct {
3250 		int	fflag;
3251 		int	kf_fflag;
3252 	} fflags_table[] = {
3253 		{ FAPPEND, KF_FLAG_APPEND },
3254 		{ FASYNC, KF_FLAG_ASYNC },
3255 		{ FFSYNC, KF_FLAG_FSYNC },
3256 		{ FHASLOCK, KF_FLAG_HASLOCK },
3257 		{ FNONBLOCK, KF_FLAG_NONBLOCK },
3258 		{ FREAD, KF_FLAG_READ },
3259 		{ FWRITE, KF_FLAG_WRITE },
3260 		{ O_CREAT, KF_FLAG_CREAT },
3261 		{ O_DIRECT, KF_FLAG_DIRECT },
3262 		{ O_EXCL, KF_FLAG_EXCL },
3263 		{ O_EXEC, KF_FLAG_EXEC },
3264 		{ O_EXLOCK, KF_FLAG_EXLOCK },
3265 		{ O_NOFOLLOW, KF_FLAG_NOFOLLOW },
3266 		{ O_SHLOCK, KF_FLAG_SHLOCK },
3267 		{ O_TRUNC, KF_FLAG_TRUNC }
3268 	};
3269 #define	NFFLAGS	(sizeof(fflags_table) / sizeof(*fflags_table))
3270 	struct kinfo_file *kif;
3271 	struct vnode *vp;
3272 	int error, locked;
3273 	unsigned int i;
3274 
3275 	if (efbuf->remainder == 0)
3276 		return (0);
3277 	kif = &efbuf->kif;
3278 	bzero(kif, sizeof(*kif));
3279 	locked = efbuf->fdp != NULL;
3280 	switch (type) {
3281 	case KF_TYPE_FIFO:
3282 	case KF_TYPE_VNODE:
3283 		if (locked) {
3284 			FILEDESC_SUNLOCK(efbuf->fdp);
3285 			locked = 0;
3286 		}
3287 		vp = (struct vnode *)data;
3288 		error = fill_vnode_info(vp, kif);
3289 		vrele(vp);
3290 		break;
3291 	case KF_TYPE_SOCKET:
3292 		error = fill_socket_info((struct socket *)data, kif);
3293 		break;
3294 	case KF_TYPE_PIPE:
3295 		error = fill_pipe_info((struct pipe *)data, kif);
3296 		break;
3297 	case KF_TYPE_PTS:
3298 		error = fill_pts_info((struct tty *)data, kif);
3299 		break;
3300 	case KF_TYPE_PROCDESC:
3301 		error = fill_procdesc_info((struct procdesc *)data, kif);
3302 		break;
3303 	case KF_TYPE_SEM:
3304 		error = fill_sem_info((struct file *)data, kif);
3305 		break;
3306 	case KF_TYPE_SHM:
3307 		error = fill_shm_info((struct file *)data, kif);
3308 		break;
3309 	default:
3310 		error = 0;
3311 	}
3312 	if (error == 0)
3313 		kif->kf_status |= KF_ATTR_VALID;
3314 
3315 	/*
3316 	 * Translate file access flags.
3317 	 */
3318 	for (i = 0; i < NFFLAGS; i++)
3319 		if (fflags & fflags_table[i].fflag)
3320 			kif->kf_flags |=  fflags_table[i].kf_fflag;
3321 	if (rightsp != NULL)
3322 		kif->kf_cap_rights = *rightsp;
3323 	else
3324 		cap_rights_init(&kif->kf_cap_rights);
3325 	kif->kf_fd = fd;
3326 	kif->kf_type = type;
3327 	kif->kf_ref_count = refcnt;
3328 	kif->kf_offset = offset;
3329 	/* Pack record size down */
3330 	kif->kf_structsize = offsetof(struct kinfo_file, kf_path) +
3331 	    strlen(kif->kf_path) + 1;
3332 	kif->kf_structsize = roundup(kif->kf_structsize, sizeof(uint64_t));
3333 	if (efbuf->remainder != -1) {
3334 		if (efbuf->remainder < kif->kf_structsize) {
3335 			/* Terminate export. */
3336 			efbuf->remainder = 0;
3337 			if (efbuf->fdp != NULL && !locked)
3338 				FILEDESC_SLOCK(efbuf->fdp);
3339 			return (0);
3340 		}
3341 		efbuf->remainder -= kif->kf_structsize;
3342 	}
3343 	if (locked)
3344 		FILEDESC_SUNLOCK(efbuf->fdp);
3345 	error = sbuf_bcat(efbuf->sb, kif, kif->kf_structsize);
3346 	if (efbuf->fdp != NULL)
3347 		FILEDESC_SLOCK(efbuf->fdp);
3348 	return (error);
3349 }
3350 
3351 /*
3352  * Store a process file descriptor information to sbuf.
3353  *
3354  * Takes a locked proc as argument, and returns with the proc unlocked.
3355  */
3356 int
3357 kern_proc_filedesc_out(struct proc *p,  struct sbuf *sb, ssize_t maxlen)
3358 {
3359 	struct file *fp;
3360 	struct filedesc *fdp;
3361 	struct export_fd_buf *efbuf;
3362 	struct vnode *cttyvp, *textvp, *tracevp;
3363 	int64_t offset;
3364 	void *data;
3365 	int error, i;
3366 	int type, refcnt, fflags;
3367 	cap_rights_t rights;
3368 
3369 	PROC_LOCK_ASSERT(p, MA_OWNED);
3370 
3371 	/* ktrace vnode */
3372 	tracevp = p->p_tracevp;
3373 	if (tracevp != NULL)
3374 		vref(tracevp);
3375 	/* text vnode */
3376 	textvp = p->p_textvp;
3377 	if (textvp != NULL)
3378 		vref(textvp);
3379 	/* Controlling tty. */
3380 	cttyvp = NULL;
3381 	if (p->p_pgrp != NULL && p->p_pgrp->pg_session != NULL) {
3382 		cttyvp = p->p_pgrp->pg_session->s_ttyvp;
3383 		if (cttyvp != NULL)
3384 			vref(cttyvp);
3385 	}
3386 	fdp = fdhold(p);
3387 	PROC_UNLOCK(p);
3388 	efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK);
3389 	efbuf->fdp = NULL;
3390 	efbuf->sb = sb;
3391 	efbuf->remainder = maxlen;
3392 	if (tracevp != NULL)
3393 		export_fd_to_sb(tracevp, KF_TYPE_VNODE, KF_FD_TYPE_TRACE,
3394 		    FREAD | FWRITE, -1, -1, NULL, efbuf);
3395 	if (textvp != NULL)
3396 		export_fd_to_sb(textvp, KF_TYPE_VNODE, KF_FD_TYPE_TEXT,
3397 		    FREAD, -1, -1, NULL, efbuf);
3398 	if (cttyvp != NULL)
3399 		export_fd_to_sb(cttyvp, KF_TYPE_VNODE, KF_FD_TYPE_CTTY,
3400 		    FREAD | FWRITE, -1, -1, NULL, efbuf);
3401 	error = 0;
3402 	if (fdp == NULL)
3403 		goto fail;
3404 	efbuf->fdp = fdp;
3405 	FILEDESC_SLOCK(fdp);
3406 	/* working directory */
3407 	if (fdp->fd_cdir != NULL) {
3408 		vref(fdp->fd_cdir);
3409 		data = fdp->fd_cdir;
3410 		export_fd_to_sb(data, KF_TYPE_VNODE, KF_FD_TYPE_CWD,
3411 		    FREAD, -1, -1, NULL, efbuf);
3412 	}
3413 	/* root directory */
3414 	if (fdp->fd_rdir != NULL) {
3415 		vref(fdp->fd_rdir);
3416 		data = fdp->fd_rdir;
3417 		export_fd_to_sb(data, KF_TYPE_VNODE, KF_FD_TYPE_ROOT,
3418 		    FREAD, -1, -1, NULL, efbuf);
3419 	}
3420 	/* jail directory */
3421 	if (fdp->fd_jdir != NULL) {
3422 		vref(fdp->fd_jdir);
3423 		data = fdp->fd_jdir;
3424 		export_fd_to_sb(data, KF_TYPE_VNODE, KF_FD_TYPE_JAIL,
3425 		    FREAD, -1, -1, NULL, efbuf);
3426 	}
3427 	for (i = 0; i < fdp->fd_nfiles; i++) {
3428 		if ((fp = fdp->fd_ofiles[i].fde_file) == NULL)
3429 			continue;
3430 		data = NULL;
3431 #ifdef CAPABILITIES
3432 		rights = *cap_rights(fdp, i);
3433 #else /* !CAPABILITIES */
3434 		cap_rights_init(&rights);
3435 #endif
3436 		switch (fp->f_type) {
3437 		case DTYPE_VNODE:
3438 			type = KF_TYPE_VNODE;
3439 			vref(fp->f_vnode);
3440 			data = fp->f_vnode;
3441 			break;
3442 
3443 		case DTYPE_SOCKET:
3444 			type = KF_TYPE_SOCKET;
3445 			data = fp->f_data;
3446 			break;
3447 
3448 		case DTYPE_PIPE:
3449 			type = KF_TYPE_PIPE;
3450 			data = fp->f_data;
3451 			break;
3452 
3453 		case DTYPE_FIFO:
3454 			type = KF_TYPE_FIFO;
3455 			vref(fp->f_vnode);
3456 			data = fp->f_vnode;
3457 			break;
3458 
3459 		case DTYPE_KQUEUE:
3460 			type = KF_TYPE_KQUEUE;
3461 			break;
3462 
3463 		case DTYPE_CRYPTO:
3464 			type = KF_TYPE_CRYPTO;
3465 			break;
3466 
3467 		case DTYPE_MQUEUE:
3468 			type = KF_TYPE_MQUEUE;
3469 			break;
3470 
3471 		case DTYPE_SHM:
3472 			type = KF_TYPE_SHM;
3473 			data = fp;
3474 			break;
3475 
3476 		case DTYPE_SEM:
3477 			type = KF_TYPE_SEM;
3478 			data = fp;
3479 			break;
3480 
3481 		case DTYPE_PTS:
3482 			type = KF_TYPE_PTS;
3483 			data = fp->f_data;
3484 			break;
3485 
3486 		case DTYPE_PROCDESC:
3487 			type = KF_TYPE_PROCDESC;
3488 			data = fp->f_data;
3489 			break;
3490 
3491 		default:
3492 			type = KF_TYPE_UNKNOWN;
3493 			break;
3494 		}
3495 		refcnt = fp->f_count;
3496 		fflags = fp->f_flag;
3497 		offset = foffset_get(fp);
3498 
3499 		/*
3500 		 * Create sysctl entry.
3501 		 * It is OK to drop the filedesc lock here as we will
3502 		 * re-validate and re-evaluate its properties when
3503 		 * the loop continues.
3504 		 */
3505 		error = export_fd_to_sb(data, type, i, fflags, refcnt,
3506 		    offset, &rights, efbuf);
3507 		if (error != 0)
3508 			break;
3509 	}
3510 	FILEDESC_SUNLOCK(fdp);
3511 	fddrop(fdp);
3512 fail:
3513 	free(efbuf, M_TEMP);
3514 	return (error);
3515 }
3516 
3517 #define FILEDESC_SBUF_SIZE	(sizeof(struct kinfo_file) * 5)
3518 
3519 /*
3520  * Get per-process file descriptors for use by procstat(1), et al.
3521  */
3522 static int
3523 sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS)
3524 {
3525 	struct sbuf sb;
3526 	struct proc *p;
3527 	ssize_t maxlen;
3528 	int error, error2, *name;
3529 
3530 	name = (int *)arg1;
3531 
3532 	sbuf_new_for_sysctl(&sb, NULL, FILEDESC_SBUF_SIZE, req);
3533 	error = pget((pid_t)name[0], PGET_CANDEBUG, &p);
3534 	if (error != 0) {
3535 		sbuf_delete(&sb);
3536 		return (error);
3537 	}
3538 	maxlen = req->oldptr != NULL ? req->oldlen : -1;
3539 	error = kern_proc_filedesc_out(p, &sb, maxlen);
3540 	error2 = sbuf_finish(&sb);
3541 	sbuf_delete(&sb);
3542 	return (error != 0 ? error : error2);
3543 }
3544 
3545 int
3546 vntype_to_kinfo(int vtype)
3547 {
3548 	struct {
3549 		int	vtype;
3550 		int	kf_vtype;
3551 	} vtypes_table[] = {
3552 		{ VBAD, KF_VTYPE_VBAD },
3553 		{ VBLK, KF_VTYPE_VBLK },
3554 		{ VCHR, KF_VTYPE_VCHR },
3555 		{ VDIR, KF_VTYPE_VDIR },
3556 		{ VFIFO, KF_VTYPE_VFIFO },
3557 		{ VLNK, KF_VTYPE_VLNK },
3558 		{ VNON, KF_VTYPE_VNON },
3559 		{ VREG, KF_VTYPE_VREG },
3560 		{ VSOCK, KF_VTYPE_VSOCK }
3561 	};
3562 #define	NVTYPES	(sizeof(vtypes_table) / sizeof(*vtypes_table))
3563 	unsigned int i;
3564 
3565 	/*
3566 	 * Perform vtype translation.
3567 	 */
3568 	for (i = 0; i < NVTYPES; i++)
3569 		if (vtypes_table[i].vtype == vtype)
3570 			break;
3571 	if (i < NVTYPES)
3572 		return (vtypes_table[i].kf_vtype);
3573 
3574 	return (KF_VTYPE_UNKNOWN);
3575 }
3576 
3577 static int
3578 fill_vnode_info(struct vnode *vp, struct kinfo_file *kif)
3579 {
3580 	struct vattr va;
3581 	char *fullpath, *freepath;
3582 	int error;
3583 
3584 	if (vp == NULL)
3585 		return (1);
3586 	kif->kf_vnode_type = vntype_to_kinfo(vp->v_type);
3587 	freepath = NULL;
3588 	fullpath = "-";
3589 	error = vn_fullpath(curthread, vp, &fullpath, &freepath);
3590 	if (error == 0) {
3591 		strlcpy(kif->kf_path, fullpath, sizeof(kif->kf_path));
3592 	}
3593 	if (freepath != NULL)
3594 		free(freepath, M_TEMP);
3595 
3596 	/*
3597 	 * Retrieve vnode attributes.
3598 	 */
3599 	va.va_fsid = VNOVAL;
3600 	va.va_rdev = NODEV;
3601 	vn_lock(vp, LK_SHARED | LK_RETRY);
3602 	error = VOP_GETATTR(vp, &va, curthread->td_ucred);
3603 	VOP_UNLOCK(vp, 0);
3604 	if (error != 0)
3605 		return (error);
3606 	if (va.va_fsid != VNOVAL)
3607 		kif->kf_un.kf_file.kf_file_fsid = va.va_fsid;
3608 	else
3609 		kif->kf_un.kf_file.kf_file_fsid =
3610 		    vp->v_mount->mnt_stat.f_fsid.val[0];
3611 	kif->kf_un.kf_file.kf_file_fileid = va.va_fileid;
3612 	kif->kf_un.kf_file.kf_file_mode = MAKEIMODE(va.va_type, va.va_mode);
3613 	kif->kf_un.kf_file.kf_file_size = va.va_size;
3614 	kif->kf_un.kf_file.kf_file_rdev = va.va_rdev;
3615 	return (0);
3616 }
3617 
3618 static int
3619 fill_socket_info(struct socket *so, struct kinfo_file *kif)
3620 {
3621 	struct sockaddr *sa;
3622 	struct inpcb *inpcb;
3623 	struct unpcb *unpcb;
3624 	int error;
3625 
3626 	if (so == NULL)
3627 		return (1);
3628 	kif->kf_sock_domain = so->so_proto->pr_domain->dom_family;
3629 	kif->kf_sock_type = so->so_type;
3630 	kif->kf_sock_protocol = so->so_proto->pr_protocol;
3631 	kif->kf_un.kf_sock.kf_sock_pcb = (uintptr_t)so->so_pcb;
3632 	switch(kif->kf_sock_domain) {
3633 	case AF_INET:
3634 	case AF_INET6:
3635 		if (kif->kf_sock_protocol == IPPROTO_TCP) {
3636 			if (so->so_pcb != NULL) {
3637 				inpcb = (struct inpcb *)(so->so_pcb);
3638 				kif->kf_un.kf_sock.kf_sock_inpcb =
3639 				    (uintptr_t)inpcb->inp_ppcb;
3640 			}
3641 		}
3642 		break;
3643 	case AF_UNIX:
3644 		if (so->so_pcb != NULL) {
3645 			unpcb = (struct unpcb *)(so->so_pcb);
3646 			if (unpcb->unp_conn) {
3647 				kif->kf_un.kf_sock.kf_sock_unpconn =
3648 				    (uintptr_t)unpcb->unp_conn;
3649 				kif->kf_un.kf_sock.kf_sock_rcv_sb_state =
3650 				    so->so_rcv.sb_state;
3651 				kif->kf_un.kf_sock.kf_sock_snd_sb_state =
3652 				    so->so_snd.sb_state;
3653 			}
3654 		}
3655 		break;
3656 	}
3657 	error = so->so_proto->pr_usrreqs->pru_sockaddr(so, &sa);
3658 	if (error == 0 && sa->sa_len <= sizeof(kif->kf_sa_local)) {
3659 		bcopy(sa, &kif->kf_sa_local, sa->sa_len);
3660 		free(sa, M_SONAME);
3661 	}
3662 	error = so->so_proto->pr_usrreqs->pru_peeraddr(so, &sa);
3663 	if (error == 0 && sa->sa_len <= sizeof(kif->kf_sa_peer)) {
3664 		bcopy(sa, &kif->kf_sa_peer, sa->sa_len);
3665 		free(sa, M_SONAME);
3666 	}
3667 	strncpy(kif->kf_path, so->so_proto->pr_domain->dom_name,
3668 	    sizeof(kif->kf_path));
3669 	return (0);
3670 }
3671 
3672 static int
3673 fill_pts_info(struct tty *tp, struct kinfo_file *kif)
3674 {
3675 
3676 	if (tp == NULL)
3677 		return (1);
3678 	kif->kf_un.kf_pts.kf_pts_dev = tty_udev(tp);
3679 	strlcpy(kif->kf_path, tty_devname(tp), sizeof(kif->kf_path));
3680 	return (0);
3681 }
3682 
3683 static int
3684 fill_pipe_info(struct pipe *pi, struct kinfo_file *kif)
3685 {
3686 
3687 	if (pi == NULL)
3688 		return (1);
3689 	kif->kf_un.kf_pipe.kf_pipe_addr = (uintptr_t)pi;
3690 	kif->kf_un.kf_pipe.kf_pipe_peer = (uintptr_t)pi->pipe_peer;
3691 	kif->kf_un.kf_pipe.kf_pipe_buffer_cnt = pi->pipe_buffer.cnt;
3692 	return (0);
3693 }
3694 
3695 static int
3696 fill_procdesc_info(struct procdesc *pdp, struct kinfo_file *kif)
3697 {
3698 
3699 	if (pdp == NULL)
3700 		return (1);
3701 	kif->kf_un.kf_proc.kf_pid = pdp->pd_pid;
3702 	return (0);
3703 }
3704 
3705 static int
3706 fill_sem_info(struct file *fp, struct kinfo_file *kif)
3707 {
3708 	struct thread *td;
3709 	struct stat sb;
3710 
3711 	td = curthread;
3712 	if (fp->f_data == NULL)
3713 		return (1);
3714 	if (fo_stat(fp, &sb, td->td_ucred, td) != 0)
3715 		return (1);
3716 	if (ksem_info == NULL)
3717 		return (1);
3718 	ksem_info(fp->f_data, kif->kf_path, sizeof(kif->kf_path),
3719 	    &kif->kf_un.kf_sem.kf_sem_value);
3720 	kif->kf_un.kf_sem.kf_sem_mode = sb.st_mode;
3721 	return (0);
3722 }
3723 
3724 static int
3725 fill_shm_info(struct file *fp, struct kinfo_file *kif)
3726 {
3727 	struct thread *td;
3728 	struct stat sb;
3729 
3730 	td = curthread;
3731 	if (fp->f_data == NULL)
3732 		return (1);
3733 	if (fo_stat(fp, &sb, td->td_ucred, td) != 0)
3734 		return (1);
3735 	shm_path(fp->f_data, kif->kf_path, sizeof(kif->kf_path));
3736 	kif->kf_un.kf_file.kf_file_mode = sb.st_mode;
3737 	kif->kf_un.kf_file.kf_file_size = sb.st_size;
3738 	return (0);
3739 }
3740 
3741 static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc, CTLFLAG_RD,
3742     sysctl_kern_proc_filedesc, "Process filedesc entries");
3743 
3744 #ifdef DDB
3745 /*
3746  * For the purposes of debugging, generate a human-readable string for the
3747  * file type.
3748  */
3749 static const char *
3750 file_type_to_name(short type)
3751 {
3752 
3753 	switch (type) {
3754 	case 0:
3755 		return ("zero");
3756 	case DTYPE_VNODE:
3757 		return ("vnod");
3758 	case DTYPE_SOCKET:
3759 		return ("sock");
3760 	case DTYPE_PIPE:
3761 		return ("pipe");
3762 	case DTYPE_FIFO:
3763 		return ("fifo");
3764 	case DTYPE_KQUEUE:
3765 		return ("kque");
3766 	case DTYPE_CRYPTO:
3767 		return ("crpt");
3768 	case DTYPE_MQUEUE:
3769 		return ("mque");
3770 	case DTYPE_SHM:
3771 		return ("shm");
3772 	case DTYPE_SEM:
3773 		return ("ksem");
3774 	default:
3775 		return ("unkn");
3776 	}
3777 }
3778 
3779 /*
3780  * For the purposes of debugging, identify a process (if any, perhaps one of
3781  * many) that references the passed file in its file descriptor array. Return
3782  * NULL if none.
3783  */
3784 static struct proc *
3785 file_to_first_proc(struct file *fp)
3786 {
3787 	struct filedesc *fdp;
3788 	struct proc *p;
3789 	int n;
3790 
3791 	FOREACH_PROC_IN_SYSTEM(p) {
3792 		if (p->p_state == PRS_NEW)
3793 			continue;
3794 		fdp = p->p_fd;
3795 		if (fdp == NULL)
3796 			continue;
3797 		for (n = 0; n < fdp->fd_nfiles; n++) {
3798 			if (fp == fdp->fd_ofiles[n].fde_file)
3799 				return (p);
3800 		}
3801 	}
3802 	return (NULL);
3803 }
3804 
3805 static void
3806 db_print_file(struct file *fp, int header)
3807 {
3808 	struct proc *p;
3809 
3810 	if (header)
3811 		db_printf("%8s %4s %8s %8s %4s %5s %6s %8s %5s %12s\n",
3812 		    "File", "Type", "Data", "Flag", "GCFl", "Count",
3813 		    "MCount", "Vnode", "FPID", "FCmd");
3814 	p = file_to_first_proc(fp);
3815 	db_printf("%8p %4s %8p %08x %04x %5d %6d %8p %5d %12s\n", fp,
3816 	    file_type_to_name(fp->f_type), fp->f_data, fp->f_flag,
3817 	    0, fp->f_count, 0, fp->f_vnode,
3818 	    p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-");
3819 }
3820 
3821 DB_SHOW_COMMAND(file, db_show_file)
3822 {
3823 	struct file *fp;
3824 
3825 	if (!have_addr) {
3826 		db_printf("usage: show file <addr>\n");
3827 		return;
3828 	}
3829 	fp = (struct file *)addr;
3830 	db_print_file(fp, 1);
3831 }
3832 
3833 DB_SHOW_COMMAND(files, db_show_files)
3834 {
3835 	struct filedesc *fdp;
3836 	struct file *fp;
3837 	struct proc *p;
3838 	int header;
3839 	int n;
3840 
3841 	header = 1;
3842 	FOREACH_PROC_IN_SYSTEM(p) {
3843 		if (p->p_state == PRS_NEW)
3844 			continue;
3845 		if ((fdp = p->p_fd) == NULL)
3846 			continue;
3847 		for (n = 0; n < fdp->fd_nfiles; ++n) {
3848 			if ((fp = fdp->fd_ofiles[n].fde_file) == NULL)
3849 				continue;
3850 			db_print_file(fp, header);
3851 			header = 0;
3852 		}
3853 	}
3854 }
3855 #endif
3856 
3857 SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
3858     &maxfilesperproc, 0, "Maximum files allowed open per process");
3859 
3860 SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
3861     &maxfiles, 0, "Maximum number of files");
3862 
3863 SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
3864     __DEVOLATILE(int *, &openfiles), 0, "System-wide number of open files");
3865 
3866 /* ARGSUSED*/
3867 static void
3868 filelistinit(void *dummy)
3869 {
3870 
3871 	file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL,
3872 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
3873 	mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF);
3874 	mtx_init(&fdesc_mtx, "fdesc", NULL, MTX_DEF);
3875 }
3876 SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL);
3877 
3878 /*-------------------------------------------------------------------*/
3879 
3880 static int
3881 badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred,
3882     int flags, struct thread *td)
3883 {
3884 
3885 	return (EBADF);
3886 }
3887 
3888 static int
3889 badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred,
3890     struct thread *td)
3891 {
3892 
3893 	return (EINVAL);
3894 }
3895 
3896 static int
3897 badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred,
3898     struct thread *td)
3899 {
3900 
3901 	return (EBADF);
3902 }
3903 
3904 static int
3905 badfo_poll(struct file *fp, int events, struct ucred *active_cred,
3906     struct thread *td)
3907 {
3908 
3909 	return (0);
3910 }
3911 
3912 static int
3913 badfo_kqfilter(struct file *fp, struct knote *kn)
3914 {
3915 
3916 	return (EBADF);
3917 }
3918 
3919 static int
3920 badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred,
3921     struct thread *td)
3922 {
3923 
3924 	return (EBADF);
3925 }
3926 
3927 static int
3928 badfo_close(struct file *fp, struct thread *td)
3929 {
3930 
3931 	return (EBADF);
3932 }
3933 
3934 static int
3935 badfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
3936     struct thread *td)
3937 {
3938 
3939 	return (EBADF);
3940 }
3941 
3942 static int
3943 badfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
3944     struct thread *td)
3945 {
3946 
3947 	return (EBADF);
3948 }
3949 
3950 static int
3951 badfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
3952     struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
3953     int kflags, struct sendfile_sync *sfs, struct thread *td)
3954 {
3955 
3956 	return (EBADF);
3957 }
3958 
3959 struct fileops badfileops = {
3960 	.fo_read = badfo_readwrite,
3961 	.fo_write = badfo_readwrite,
3962 	.fo_truncate = badfo_truncate,
3963 	.fo_ioctl = badfo_ioctl,
3964 	.fo_poll = badfo_poll,
3965 	.fo_kqfilter = badfo_kqfilter,
3966 	.fo_stat = badfo_stat,
3967 	.fo_close = badfo_close,
3968 	.fo_chmod = badfo_chmod,
3969 	.fo_chown = badfo_chown,
3970 	.fo_sendfile = badfo_sendfile,
3971 };
3972 
3973 int
3974 invfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
3975     struct thread *td)
3976 {
3977 
3978 	return (EINVAL);
3979 }
3980 
3981 int
3982 invfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
3983     struct thread *td)
3984 {
3985 
3986 	return (EINVAL);
3987 }
3988 
3989 int
3990 invfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio,
3991     struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags,
3992     int kflags, struct sendfile_sync *sfs, struct thread *td)
3993 {
3994 
3995 	return (EINVAL);
3996 }
3997 
3998 /*-------------------------------------------------------------------*/
3999 
4000 /*
4001  * File Descriptor pseudo-device driver (/dev/fd/).
4002  *
4003  * Opening minor device N dup()s the file (if any) connected to file
4004  * descriptor N belonging to the calling process.  Note that this driver
4005  * consists of only the ``open()'' routine, because all subsequent
4006  * references to this file will be direct to the other driver.
4007  *
4008  * XXX: we could give this one a cloning event handler if necessary.
4009  */
4010 
4011 /* ARGSUSED */
4012 static int
4013 fdopen(struct cdev *dev, int mode, int type, struct thread *td)
4014 {
4015 
4016 	/*
4017 	 * XXX Kludge: set curthread->td_dupfd to contain the value of the
4018 	 * the file descriptor being sought for duplication. The error
4019 	 * return ensures that the vnode for this device will be released
4020 	 * by vn_open. Open will detect this special error and take the
4021 	 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
4022 	 * will simply report the error.
4023 	 */
4024 	td->td_dupfd = dev2unit(dev);
4025 	return (ENODEV);
4026 }
4027 
4028 static struct cdevsw fildesc_cdevsw = {
4029 	.d_version =	D_VERSION,
4030 	.d_open =	fdopen,
4031 	.d_name =	"FD",
4032 };
4033 
4034 static void
4035 fildesc_drvinit(void *unused)
4036 {
4037 	struct cdev *dev;
4038 
4039 	dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 0, NULL,
4040 	    UID_ROOT, GID_WHEEL, 0666, "fd/0");
4041 	make_dev_alias(dev, "stdin");
4042 	dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 1, NULL,
4043 	    UID_ROOT, GID_WHEEL, 0666, "fd/1");
4044 	make_dev_alias(dev, "stdout");
4045 	dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 2, NULL,
4046 	    UID_ROOT, GID_WHEEL, 0666, "fd/2");
4047 	make_dev_alias(dev, "stderr");
4048 }
4049 
4050 SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL);
4051