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: head/sys/kern/kern_descrip.c 176957 2008-03-08 22:02:21Z antoine $");
39
40#include "opt_compat.h"
41#include "opt_ddb.h"
42#include "opt_ktrace.h"
43
44#include <sys/param.h>
45#include <sys/systm.h>
46
47#include <sys/conf.h>
48#include <sys/domain.h>
49#include <sys/fcntl.h>
50#include <sys/file.h>
51#include <sys/filedesc.h>
52#include <sys/filio.h>
53#include <sys/jail.h>
54#include <sys/kernel.h>
55#include <sys/limits.h>
56#include <sys/lock.h>
57#include <sys/malloc.h>
58#include <sys/mount.h>
59#include <sys/mqueue.h>
60#include <sys/mutex.h>
61#include <sys/namei.h>
62#include <sys/priv.h>
63#include <sys/proc.h>
64#include <sys/protosw.h>
65#include <sys/resourcevar.h>
66#include <sys/signalvar.h>
67#include <sys/socketvar.h>
68#include <sys/stat.h>
69#include <sys/sx.h>
70#include <sys/syscallsubr.h>
71#include <sys/sysctl.h>
72#include <sys/sysproto.h>
73#include <sys/unistd.h>
74#include <sys/user.h>
75#include <sys/vnode.h>
76#ifdef KTRACE
77#include <sys/ktrace.h>
78#endif
79
80#include <security/audit/audit.h>
81
82#include <vm/uma.h>
83
84#include <ddb/ddb.h>
85
86static MALLOC_DEFINE(M_FILEDESC, "filedesc", "Open file descriptor table");
87static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "filedesc_to_leader",
88 "file desc to leader structures");
89static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures");
90
91static uma_zone_t file_zone;
92
93
94/* How to treat 'new' parameter when allocating a fd for do_dup(). */
95enum dup_type { DUP_VARIABLE, DUP_FIXED };
96
97static int do_dup(struct thread *td, enum dup_type type, int old, int new,
98 register_t *retval);
99static int fd_first_free(struct filedesc *, int, int);
100static int fd_last_used(struct filedesc *, int, int);
101static void fdgrowtable(struct filedesc *, int);
102static void fdunused(struct filedesc *fdp, int fd);
103static void fdused(struct filedesc *fdp, int fd);
104
105/*
106 * A process is initially started out with NDFILE descriptors stored within
107 * this structure, selected to be enough for typical applications based on
108 * the historical limit of 20 open files (and the usage of descriptors by
109 * shells). If these descriptors are exhausted, a larger descriptor table
110 * may be allocated, up to a process' resource limit; the internal arrays
111 * are then unused.
112 */
113#define NDFILE 20
114#define NDSLOTSIZE sizeof(NDSLOTTYPE)
115#define NDENTRIES (NDSLOTSIZE * __CHAR_BIT)
116#define NDSLOT(x) ((x) / NDENTRIES)
117#define NDBIT(x) ((NDSLOTTYPE)1 << ((x) % NDENTRIES))
118#define NDSLOTS(x) (((x) + NDENTRIES - 1) / NDENTRIES)
119
120/*
121 * Storage required per open file descriptor.
122 */
123#define OFILESIZE (sizeof(struct file *) + sizeof(char))
124
125/*
126 * Basic allocation of descriptors:
127 * one of the above, plus arrays for NDFILE descriptors.
128 */
129struct filedesc0 {
130 struct filedesc fd_fd;
131 /*
132 * These arrays are used when the number of open files is
133 * <= NDFILE, and are then pointed to by the pointers above.
134 */
135 struct file *fd_dfiles[NDFILE];
136 char fd_dfileflags[NDFILE];
137 NDSLOTTYPE fd_dmap[NDSLOTS(NDFILE)];
138};
139
140/*
141 * Descriptor management.
142 */
143volatile int openfiles; /* actual number of open files */
144struct mtx sigio_lock; /* mtx to protect pointers to sigio */
145void (*mq_fdclose)(struct thread *td, int fd, struct file *fp);
146
147/* A mutex to protect the association between a proc and filedesc. */
148static struct mtx fdesc_mtx;
149
150/*
151 * Find the first zero bit in the given bitmap, starting at low and not
152 * exceeding size - 1.
153 */
154static int
155fd_first_free(struct filedesc *fdp, int low, int size)
156{
157 NDSLOTTYPE *map = fdp->fd_map;
158 NDSLOTTYPE mask;
159 int off, maxoff;
160
161 if (low >= size)
162 return (low);
163
164 off = NDSLOT(low);
165 if (low % NDENTRIES) {
166 mask = ~(~(NDSLOTTYPE)0 >> (NDENTRIES - (low % NDENTRIES)));
167 if ((mask &= ~map[off]) != 0UL)
168 return (off * NDENTRIES + ffsl(mask) - 1);
169 ++off;
170 }
171 for (maxoff = NDSLOTS(size); off < maxoff; ++off)
172 if (map[off] != ~0UL)
173 return (off * NDENTRIES + ffsl(~map[off]) - 1);
174 return (size);
175}
176
177/*
178 * Find the highest non-zero bit in the given bitmap, starting at low and
179 * not exceeding size - 1.
180 */
181static int
182fd_last_used(struct filedesc *fdp, int low, int size)
183{
184 NDSLOTTYPE *map = fdp->fd_map;
185 NDSLOTTYPE mask;
186 int off, minoff;
187
188 if (low >= size)
189 return (-1);
190
191 off = NDSLOT(size);
192 if (size % NDENTRIES) {
193 mask = ~(~(NDSLOTTYPE)0 << (size % NDENTRIES));
194 if ((mask &= map[off]) != 0)
195 return (off * NDENTRIES + flsl(mask) - 1);
196 --off;
197 }
198 for (minoff = NDSLOT(low); off >= minoff; --off)
199 if (map[off] != 0)
200 return (off * NDENTRIES + flsl(map[off]) - 1);
201 return (low - 1);
202}
203
204static int
205fdisused(struct filedesc *fdp, int fd)
206{
207 KASSERT(fd >= 0 && fd < fdp->fd_nfiles,
208 ("file descriptor %d out of range (0, %d)", fd, fdp->fd_nfiles));
209 return ((fdp->fd_map[NDSLOT(fd)] & NDBIT(fd)) != 0);
210}
211
212/*
213 * Mark a file descriptor as used.
214 */
215static void
216fdused(struct filedesc *fdp, int fd)
217{
218
219 FILEDESC_XLOCK_ASSERT(fdp);
220 KASSERT(!fdisused(fdp, fd),
221 ("fd already used"));
222
223 fdp->fd_map[NDSLOT(fd)] |= NDBIT(fd);
224 if (fd > fdp->fd_lastfile)
225 fdp->fd_lastfile = fd;
226 if (fd == fdp->fd_freefile)
227 fdp->fd_freefile = fd_first_free(fdp, fd, fdp->fd_nfiles);
228}
229
230/*
231 * Mark a file descriptor as unused.
232 */
233static void
234fdunused(struct filedesc *fdp, int fd)
235{
236
237 FILEDESC_XLOCK_ASSERT(fdp);
238 KASSERT(fdisused(fdp, fd),
239 ("fd is already unused"));
240 KASSERT(fdp->fd_ofiles[fd] == NULL,
241 ("fd is still in use"));
242
243 fdp->fd_map[NDSLOT(fd)] &= ~NDBIT(fd);
244 if (fd < fdp->fd_freefile)
245 fdp->fd_freefile = fd;
246 if (fd == fdp->fd_lastfile)
247 fdp->fd_lastfile = fd_last_used(fdp, 0, fd);
248}
249
250/*
251 * System calls on descriptors.
252 */
253#ifndef _SYS_SYSPROTO_H_
254struct getdtablesize_args {
255 int dummy;
256};
257#endif
258/* ARGSUSED */
259int
260getdtablesize(struct thread *td, struct getdtablesize_args *uap)
261{
262 struct proc *p = td->td_proc;
263
264 PROC_LOCK(p);
265 td->td_retval[0] =
266 min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc);
267 PROC_UNLOCK(p);
268 return (0);
269}
270
271/*
272 * Duplicate a file descriptor to a particular value.
273 *
274 * Note: keep in mind that a potential race condition exists when closing
275 * descriptors from a shared descriptor table (via rfork).
276 */
277#ifndef _SYS_SYSPROTO_H_
278struct dup2_args {
279 u_int from;
280 u_int to;
281};
282#endif
283/* ARGSUSED */
284int
285dup2(struct thread *td, struct dup2_args *uap)
286{
287
288 return (do_dup(td, DUP_FIXED, (int)uap->from, (int)uap->to,
289 td->td_retval));
290}
291
292/*
293 * Duplicate a file descriptor.
294 */
295#ifndef _SYS_SYSPROTO_H_
296struct dup_args {
297 u_int fd;
298};
299#endif
300/* ARGSUSED */
301int
302dup(struct thread *td, struct dup_args *uap)
303{
304
305 return (do_dup(td, DUP_VARIABLE, (int)uap->fd, 0, td->td_retval));
306}
307
308/*
309 * The file control system call.
310 */
311#ifndef _SYS_SYSPROTO_H_
312struct fcntl_args {
313 int fd;
314 int cmd;
315 long arg;
316};
317#endif
318/* ARGSUSED */
319int
320fcntl(struct thread *td, struct fcntl_args *uap)
321{
322 struct flock fl;
323 intptr_t arg;
324 int error;
325
326 error = 0;
327 switch (uap->cmd) {
328 case F_GETLK:
329 case F_SETLK:
330 case F_SETLKW:
331 error = copyin((void *)(intptr_t)uap->arg, &fl, sizeof(fl));
332 arg = (intptr_t)&fl;
333 break;
334 default:
335 arg = uap->arg;
336 break;
337 }
338 if (error)
339 return (error);
340 error = kern_fcntl(td, uap->fd, uap->cmd, arg);
341 if (error)
342 return (error);
343 if (uap->cmd == F_GETLK)
344 error = copyout(&fl, (void *)(intptr_t)uap->arg, sizeof(fl));
345 return (error);
346}
347
348static inline struct file *
349fdtofp(int fd, struct filedesc *fdp)
350{
351 struct file *fp;
352
353 FILEDESC_LOCK_ASSERT(fdp);
354 if ((unsigned)fd >= fdp->fd_nfiles ||
355 (fp = fdp->fd_ofiles[fd]) == NULL)
356 return (NULL);
357 return (fp);
358}
359
360int
361kern_fcntl(struct thread *td, int fd, int cmd, intptr_t arg)
362{
363 struct filedesc *fdp;
364 struct flock *flp;
365 struct file *fp;
366 struct proc *p;
367 char *pop;
368 struct vnode *vp;
369 u_int newmin;
370 int error, flg, tmp;
371 int vfslocked;
372
373 vfslocked = 0;
374 error = 0;
375 flg = F_POSIX;
376 p = td->td_proc;
377 fdp = p->p_fd;
378
379 switch (cmd) {
380 case F_DUPFD:
381 FILEDESC_SLOCK(fdp);
382 if ((fp = fdtofp(fd, fdp)) == NULL) {
383 FILEDESC_SUNLOCK(fdp);
384 error = EBADF;
385 break;
386 }
387 FILEDESC_SUNLOCK(fdp);
388 newmin = arg;
389 PROC_LOCK(p);
390 if (newmin >= lim_cur(p, RLIMIT_NOFILE) ||
391 newmin >= maxfilesperproc) {
392 PROC_UNLOCK(p);
393 error = EINVAL;
394 break;
395 }
396 PROC_UNLOCK(p);
397 error = do_dup(td, DUP_VARIABLE, fd, newmin, td->td_retval);
398 break;
399
400 case F_DUP2FD:
401 tmp = arg;
402 error = do_dup(td, DUP_FIXED, fd, tmp, td->td_retval);
403 break;
404
405 case F_GETFD:
406 FILEDESC_SLOCK(fdp);
407 if ((fp = fdtofp(fd, fdp)) == NULL) {
408 FILEDESC_SUNLOCK(fdp);
409 error = EBADF;
410 break;
411 }
412 pop = &fdp->fd_ofileflags[fd];
413 td->td_retval[0] = (*pop & UF_EXCLOSE) ? FD_CLOEXEC : 0;
414 FILEDESC_SUNLOCK(fdp);
415 break;
416
417 case F_SETFD:
418 FILEDESC_XLOCK(fdp);
419 if ((fp = fdtofp(fd, fdp)) == NULL) {
420 FILEDESC_XUNLOCK(fdp);
421 error = EBADF;
422 break;
423 }
424 pop = &fdp->fd_ofileflags[fd];
425 *pop = (*pop &~ UF_EXCLOSE) |
426 (arg & FD_CLOEXEC ? UF_EXCLOSE : 0);
427 FILEDESC_XUNLOCK(fdp);
428 break;
429
430 case F_GETFL:
431 FILEDESC_SLOCK(fdp);
432 if ((fp = fdtofp(fd, fdp)) == NULL) {
433 FILEDESC_SUNLOCK(fdp);
434 error = EBADF;
435 break;
436 }
437 td->td_retval[0] = OFLAGS(fp->f_flag);
438 FILEDESC_SUNLOCK(fdp);
439 break;
440
441 case F_SETFL:
442 FILEDESC_SLOCK(fdp);
443 if ((fp = fdtofp(fd, fdp)) == NULL) {
444 FILEDESC_SUNLOCK(fdp);
445 error = EBADF;
446 break;
447 }
448 fhold(fp);
449 FILEDESC_SUNLOCK(fdp);
450 do {
451 tmp = flg = fp->f_flag;
452 tmp &= ~FCNTLFLAGS;
453 tmp |= FFLAGS(arg & ~O_ACCMODE) & FCNTLFLAGS;
454 } while(atomic_cmpset_int(&fp->f_flag, flg, tmp) == 0);
455 tmp = fp->f_flag & FNONBLOCK;
456 error = fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
457 if (error) {
458 fdrop(fp, td);
459 break;
460 }
461 tmp = fp->f_flag & FASYNC;
462 error = fo_ioctl(fp, FIOASYNC, &tmp, td->td_ucred, td);
463 if (error == 0) {
464 fdrop(fp, td);
465 break;
466 }
467 atomic_clear_int(&fp->f_flag, FNONBLOCK);
468 tmp = 0;
469 (void)fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td);
470 fdrop(fp, td);
471 break;
472
473 case F_GETOWN:
474 FILEDESC_SLOCK(fdp);
475 if ((fp = fdtofp(fd, fdp)) == NULL) {
476 FILEDESC_SUNLOCK(fdp);
477 error = EBADF;
478 break;
479 }
480 fhold(fp);
481 FILEDESC_SUNLOCK(fdp);
482 error = fo_ioctl(fp, FIOGETOWN, &tmp, td->td_ucred, td);
483 if (error == 0)
484 td->td_retval[0] = tmp;
485 fdrop(fp, td);
486 break;
487
488 case F_SETOWN:
489 FILEDESC_SLOCK(fdp);
490 if ((fp = fdtofp(fd, fdp)) == NULL) {
491 FILEDESC_SUNLOCK(fdp);
492 error = EBADF;
493 break;
494 }
495 fhold(fp);
496 FILEDESC_SUNLOCK(fdp);
497 tmp = arg;
498 error = fo_ioctl(fp, FIOSETOWN, &tmp, td->td_ucred, td);
499 fdrop(fp, td);
500 break;
501
502 case F_SETLKW:
503 flg |= F_WAIT;
504 /* FALLTHROUGH F_SETLK */
505
506 case F_SETLK:
507 FILEDESC_SLOCK(fdp);
508 if ((fp = fdtofp(fd, fdp)) == NULL) {
509 FILEDESC_SUNLOCK(fdp);
510 error = EBADF;
511 break;
512 }
513 if (fp->f_type != DTYPE_VNODE) {
514 FILEDESC_SUNLOCK(fdp);
515 error = EBADF;
516 break;
517 }
518 flp = (struct flock *)arg;
519 if (flp->l_whence == SEEK_CUR) {
520 if (fp->f_offset < 0 ||
521 (flp->l_start > 0 &&
522 fp->f_offset > OFF_MAX - flp->l_start)) {
523 FILEDESC_SUNLOCK(fdp);
524 error = EOVERFLOW;
525 break;
526 }
527 flp->l_start += fp->f_offset;
528 }
529
530 /*
531 * VOP_ADVLOCK() may block.
532 */
533 fhold(fp);
534 FILEDESC_SUNLOCK(fdp);
535 vp = fp->f_vnode;
536 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
537 switch (flp->l_type) {
538 case F_RDLCK:
539 if ((fp->f_flag & FREAD) == 0) {
540 error = EBADF;
541 break;
542 }
543 PROC_LOCK(p->p_leader);
544 p->p_leader->p_flag |= P_ADVLOCK;
545 PROC_UNLOCK(p->p_leader);
546 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
547 flp, flg);
548 break;
549 case F_WRLCK:
550 if ((fp->f_flag & FWRITE) == 0) {
551 error = EBADF;
552 break;
553 }
554 PROC_LOCK(p->p_leader);
555 p->p_leader->p_flag |= P_ADVLOCK;
556 PROC_UNLOCK(p->p_leader);
557 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK,
558 flp, flg);
559 break;
560 case F_UNLCK:
561 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK,
562 flp, F_POSIX);
563 break;
564 default:
565 error = EINVAL;
566 break;
567 }
568 VFS_UNLOCK_GIANT(vfslocked);
569 vfslocked = 0;
570 /* Check for race with close */
571 FILEDESC_SLOCK(fdp);
572 if ((unsigned) fd >= fdp->fd_nfiles ||
573 fp != fdp->fd_ofiles[fd]) {
574 FILEDESC_SUNLOCK(fdp);
575 flp->l_whence = SEEK_SET;
576 flp->l_start = 0;
577 flp->l_len = 0;
578 flp->l_type = F_UNLCK;
579 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
580 (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader,
581 F_UNLCK, flp, F_POSIX);
582 VFS_UNLOCK_GIANT(vfslocked);
583 vfslocked = 0;
584 } else
585 FILEDESC_SUNLOCK(fdp);
586 fdrop(fp, td);
587 break;
588
589 case F_GETLK:
590 FILEDESC_SLOCK(fdp);
591 if ((fp = fdtofp(fd, fdp)) == NULL) {
592 FILEDESC_SUNLOCK(fdp);
593 error = EBADF;
594 break;
595 }
596 if (fp->f_type != DTYPE_VNODE) {
597 FILEDESC_SUNLOCK(fdp);
598 error = EBADF;
599 break;
600 }
601 flp = (struct flock *)arg;
602 if (flp->l_type != F_RDLCK && flp->l_type != F_WRLCK &&
603 flp->l_type != F_UNLCK) {
604 FILEDESC_SUNLOCK(fdp);
605 error = EINVAL;
606 break;
607 }
608 if (flp->l_whence == SEEK_CUR) {
609 if ((flp->l_start > 0 &&
610 fp->f_offset > OFF_MAX - flp->l_start) ||
611 (flp->l_start < 0 &&
612 fp->f_offset < OFF_MIN - flp->l_start)) {
613 FILEDESC_SUNLOCK(fdp);
614 error = EOVERFLOW;
615 break;
616 }
617 flp->l_start += fp->f_offset;
618 }
619 /*
620 * VOP_ADVLOCK() may block.
621 */
622 fhold(fp);
623 FILEDESC_SUNLOCK(fdp);
624 vp = fp->f_vnode;
625 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
626 error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, flp,
627 F_POSIX);
628 VFS_UNLOCK_GIANT(vfslocked);
629 vfslocked = 0;
630 fdrop(fp, td);
631 break;
632 default:
633 error = EINVAL;
634 break;
635 }
636 VFS_UNLOCK_GIANT(vfslocked);
637 return (error);
638}
639
640/*
641 * Common code for dup, dup2, fcntl(F_DUPFD) and fcntl(F_DUP2FD).
642 */
643static int
644do_dup(struct thread *td, enum dup_type type, int old, int new,
645 register_t *retval)
646{
647 struct filedesc *fdp;
648 struct proc *p;
649 struct file *fp;
650 struct file *delfp;
651 int error, holdleaders, maxfd;
652
653 KASSERT((type == DUP_VARIABLE || type == DUP_FIXED),
654 ("invalid dup type %d", type));
655
656 p = td->td_proc;
657 fdp = p->p_fd;
658
659 /*
660 * Verify we have a valid descriptor to dup from and possibly to
661 * dup to.
662 */
663 if (old < 0 || new < 0)
664 return (EBADF);
665 PROC_LOCK(p);
666 maxfd = min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc);
667 PROC_UNLOCK(p);
668 if (new >= maxfd)
669 return (EMFILE);
670
671 FILEDESC_XLOCK(fdp);
672 if (old >= fdp->fd_nfiles || fdp->fd_ofiles[old] == NULL) {
673 FILEDESC_XUNLOCK(fdp);
674 return (EBADF);
675 }
676 if (type == DUP_FIXED && old == new) {
677 *retval = new;
678 FILEDESC_XUNLOCK(fdp);
679 return (0);
680 }
681 fp = fdp->fd_ofiles[old];
682 fhold(fp);
683
684 /*
685 * If the caller specified a file descriptor, make sure the file
686 * table is large enough to hold it, and grab it. Otherwise, just
687 * allocate a new descriptor the usual way. Since the filedesc
688 * lock may be temporarily dropped in the process, we have to look
689 * out for a race.
690 */
691 if (type == DUP_FIXED) {
692 if (new >= fdp->fd_nfiles)
693 fdgrowtable(fdp, new + 1);
694 if (fdp->fd_ofiles[new] == NULL)
695 fdused(fdp, new);
696 } else {
697 if ((error = fdalloc(td, new, &new)) != 0) {
698 FILEDESC_XUNLOCK(fdp);
699 fdrop(fp, td);
700 return (error);
701 }
702 }
703
704 /*
705 * If the old file changed out from under us then treat it as a
706 * bad file descriptor. Userland should do its own locking to
707 * avoid this case.
708 */
709 if (fdp->fd_ofiles[old] != fp) {
710 /* we've allocated a descriptor which we won't use */
711 if (fdp->fd_ofiles[new] == NULL)
712 fdunused(fdp, new);
713 FILEDESC_XUNLOCK(fdp);
714 fdrop(fp, td);
715 return (EBADF);
716 }
717 KASSERT(old != new,
718 ("new fd is same as old"));
719
720 /*
721 * Save info on the descriptor being overwritten. We cannot close
722 * it without introducing an ownership race for the slot, since we
723 * need to drop the filedesc lock to call closef().
724 *
725 * XXX this duplicates parts of close().
726 */
727 delfp = fdp->fd_ofiles[new];
728 holdleaders = 0;
729 if (delfp != NULL) {
730 if (td->td_proc->p_fdtol != NULL) {
731 /*
732 * Ask fdfree() to sleep to ensure that all relevant
733 * process leaders can be traversed in closef().
734 */
735 fdp->fd_holdleaderscount++;
736 holdleaders = 1;
737 }
738 }
739
740 /*
741 * Duplicate the source descriptor
742 */
743 fdp->fd_ofiles[new] = fp;
744 fdp->fd_ofileflags[new] = fdp->fd_ofileflags[old] &~ UF_EXCLOSE;
745 if (new > fdp->fd_lastfile)
746 fdp->fd_lastfile = new;
747 *retval = new;
748
749 /*
750 * If we dup'd over a valid file, we now own the reference to it
751 * and must dispose of it using closef() semantics (as if a
752 * close() were performed on it).
753 *
754 * XXX this duplicates parts of close().
755 */
756 if (delfp != NULL) {
757 knote_fdclose(td, new);
758 if (delfp->f_type == DTYPE_MQUEUE)
759 mq_fdclose(td, new, delfp);
760 FILEDESC_XUNLOCK(fdp);
761 (void) closef(delfp, td);
762 if (holdleaders) {
763 FILEDESC_XLOCK(fdp);
764 fdp->fd_holdleaderscount--;
765 if (fdp->fd_holdleaderscount == 0 &&
766 fdp->fd_holdleaderswakeup != 0) {
767 fdp->fd_holdleaderswakeup = 0;
768 wakeup(&fdp->fd_holdleaderscount);
769 }
770 FILEDESC_XUNLOCK(fdp);
771 }
772 } else {
773 FILEDESC_XUNLOCK(fdp);
774 }
775 return (0);
776}
777
778/*
779 * If sigio is on the list associated with a process or process group,
780 * disable signalling from the device, remove sigio from the list and
781 * free sigio.
782 */
783void
784funsetown(struct sigio **sigiop)
785{
786 struct sigio *sigio;
787
788 SIGIO_LOCK();
789 sigio = *sigiop;
790 if (sigio == NULL) {
791 SIGIO_UNLOCK();
792 return;
793 }
794 *(sigio->sio_myref) = NULL;
795 if ((sigio)->sio_pgid < 0) {
796 struct pgrp *pg = (sigio)->sio_pgrp;
797 PGRP_LOCK(pg);
798 SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio,
799 sigio, sio_pgsigio);
800 PGRP_UNLOCK(pg);
801 } else {
802 struct proc *p = (sigio)->sio_proc;
803 PROC_LOCK(p);
804 SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio,
805 sigio, sio_pgsigio);
806 PROC_UNLOCK(p);
807 }
808 SIGIO_UNLOCK();
809 crfree(sigio->sio_ucred);
810 FREE(sigio, M_SIGIO);
811}
812
813/*
814 * Free a list of sigio structures.
815 * We only need to lock the SIGIO_LOCK because we have made ourselves
816 * inaccessible to callers of fsetown and therefore do not need to lock
817 * the proc or pgrp struct for the list manipulation.
818 */
819void
820funsetownlst(struct sigiolst *sigiolst)
821{
822 struct proc *p;
823 struct pgrp *pg;
824 struct sigio *sigio;
825
826 sigio = SLIST_FIRST(sigiolst);
827 if (sigio == NULL)
828 return;
829 p = NULL;
830 pg = NULL;
831
832 /*
833 * Every entry of the list should belong
834 * to a single proc or pgrp.
835 */
836 if (sigio->sio_pgid < 0) {
837 pg = sigio->sio_pgrp;
838 PGRP_LOCK_ASSERT(pg, MA_NOTOWNED);
839 } else /* if (sigio->sio_pgid > 0) */ {
840 p = sigio->sio_proc;
841 PROC_LOCK_ASSERT(p, MA_NOTOWNED);
842 }
843
844 SIGIO_LOCK();
845 while ((sigio = SLIST_FIRST(sigiolst)) != NULL) {
846 *(sigio->sio_myref) = NULL;
847 if (pg != NULL) {
848 KASSERT(sigio->sio_pgid < 0,
849 ("Proc sigio in pgrp sigio list"));
850 KASSERT(sigio->sio_pgrp == pg,
851 ("Bogus pgrp in sigio list"));
852 PGRP_LOCK(pg);
853 SLIST_REMOVE(&pg->pg_sigiolst, sigio, sigio,
854 sio_pgsigio);
855 PGRP_UNLOCK(pg);
856 } else /* if (p != NULL) */ {
857 KASSERT(sigio->sio_pgid > 0,
858 ("Pgrp sigio in proc sigio list"));
859 KASSERT(sigio->sio_proc == p,
860 ("Bogus proc in sigio list"));
861 PROC_LOCK(p);
862 SLIST_REMOVE(&p->p_sigiolst, sigio, sigio,
863 sio_pgsigio);
864 PROC_UNLOCK(p);
865 }
866 SIGIO_UNLOCK();
867 crfree(sigio->sio_ucred);
868 FREE(sigio, M_SIGIO);
869 SIGIO_LOCK();
870 }
871 SIGIO_UNLOCK();
872}
873
874/*
875 * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg).
876 *
877 * After permission checking, add a sigio structure to the sigio list for
878 * the process or process group.
879 */
880int
881fsetown(pid_t pgid, struct sigio **sigiop)
882{
883 struct proc *proc;
884 struct pgrp *pgrp;
885 struct sigio *sigio;
886 int ret;
887
888 if (pgid == 0) {
889 funsetown(sigiop);
890 return (0);
891 }
892
893 ret = 0;
894
895 /* Allocate and fill in the new sigio out of locks. */
896 MALLOC(sigio, struct sigio *, sizeof(struct sigio), M_SIGIO, M_WAITOK);
897 sigio->sio_pgid = pgid;
898 sigio->sio_ucred = crhold(curthread->td_ucred);
899 sigio->sio_myref = sigiop;
900
901 sx_slock(&proctree_lock);
902 if (pgid > 0) {
903 proc = pfind(pgid);
904 if (proc == NULL) {
905 ret = ESRCH;
906 goto fail;
907 }
908
909 /*
910 * Policy - Don't allow a process to FSETOWN a process
911 * in another session.
912 *
913 * Remove this test to allow maximum flexibility or
914 * restrict FSETOWN to the current process or process
915 * group for maximum safety.
916 */
917 PROC_UNLOCK(proc);
918 if (proc->p_session != curthread->td_proc->p_session) {
919 ret = EPERM;
920 goto fail;
921 }
922
923 pgrp = NULL;
924 } else /* if (pgid < 0) */ {
925 pgrp = pgfind(-pgid);
926 if (pgrp == NULL) {
927 ret = ESRCH;
928 goto fail;
929 }
930 PGRP_UNLOCK(pgrp);
931
932 /*
933 * Policy - Don't allow a process to FSETOWN a process
934 * in another session.
935 *
936 * Remove this test to allow maximum flexibility or
937 * restrict FSETOWN to the current process or process
938 * group for maximum safety.
939 */
940 if (pgrp->pg_session != curthread->td_proc->p_session) {
941 ret = EPERM;
942 goto fail;
943 }
944
945 proc = NULL;
946 }
947 funsetown(sigiop);
948 if (pgid > 0) {
949 PROC_LOCK(proc);
950 /*
951 * Since funsetownlst() is called without the proctree
952 * locked, we need to check for P_WEXIT.
953 * XXX: is ESRCH correct?
954 */
955 if ((proc->p_flag & P_WEXIT) != 0) {
956 PROC_UNLOCK(proc);
957 ret = ESRCH;
958 goto fail;
959 }
960 SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio);
961 sigio->sio_proc = proc;
962 PROC_UNLOCK(proc);
963 } else {
964 PGRP_LOCK(pgrp);
965 SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio);
966 sigio->sio_pgrp = pgrp;
967 PGRP_UNLOCK(pgrp);
968 }
969 sx_sunlock(&proctree_lock);
970 SIGIO_LOCK();
971 *sigiop = sigio;
972 SIGIO_UNLOCK();
973 return (0);
974
975fail:
976 sx_sunlock(&proctree_lock);
977 crfree(sigio->sio_ucred);
978 FREE(sigio, M_SIGIO);
979 return (ret);
980}
981
982/*
983 * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg).
984 */
985pid_t
986fgetown(sigiop)
987 struct sigio **sigiop;
988{
989 pid_t pgid;
990
991 SIGIO_LOCK();
992 pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0;
993 SIGIO_UNLOCK();
994 return (pgid);
995}
996
997/*
998 * Close a file descriptor.
999 */
1000#ifndef _SYS_SYSPROTO_H_
1001struct close_args {
1002 int fd;
1003};
1004#endif
1005/* ARGSUSED */
1006int
1007close(td, uap)
1008 struct thread *td;
1009 struct close_args *uap;
1010{
1011
1012 return (kern_close(td, uap->fd));
1013}
1014
1015int
1016kern_close(td, fd)
1017 struct thread *td;
1018 int fd;
1019{
1020 struct filedesc *fdp;
1021 struct file *fp;
1022 int error;
1023 int holdleaders;
1024
1025 error = 0;
1026 holdleaders = 0;
1027 fdp = td->td_proc->p_fd;
1028
1029 AUDIT_SYSCLOSE(td, fd);
1030
1031 FILEDESC_XLOCK(fdp);
1032 if ((unsigned)fd >= fdp->fd_nfiles ||
1033 (fp = fdp->fd_ofiles[fd]) == NULL) {
1034 FILEDESC_XUNLOCK(fdp);
1035 return (EBADF);
1036 }
1037 fdp->fd_ofiles[fd] = NULL;
1038 fdp->fd_ofileflags[fd] = 0;
1039 fdunused(fdp, fd);
1040 if (td->td_proc->p_fdtol != NULL) {
1041 /*
1042 * Ask fdfree() to sleep to ensure that all relevant
1043 * process leaders can be traversed in closef().
1044 */
1045 fdp->fd_holdleaderscount++;
1046 holdleaders = 1;
1047 }
1048
1049 /*
1050 * We now hold the fp reference that used to be owned by the
1051 * descriptor array. We have to unlock the FILEDESC *AFTER*
1052 * knote_fdclose to prevent a race of the fd getting opened, a knote
1053 * added, and deleteing a knote for the new fd.
1054 */
1055 knote_fdclose(td, fd);
1056 if (fp->f_type == DTYPE_MQUEUE)
1057 mq_fdclose(td, fd, fp);
1058 FILEDESC_XUNLOCK(fdp);
1059
1060 error = closef(fp, td);
1061 if (holdleaders) {
1062 FILEDESC_XLOCK(fdp);
1063 fdp->fd_holdleaderscount--;
1064 if (fdp->fd_holdleaderscount == 0 &&
1065 fdp->fd_holdleaderswakeup != 0) {
1066 fdp->fd_holdleaderswakeup = 0;
1067 wakeup(&fdp->fd_holdleaderscount);
1068 }
1069 FILEDESC_XUNLOCK(fdp);
1070 }
1071 return (error);
1072}
1073
1074#if defined(COMPAT_43)
1075/*
1076 * Return status information about a file descriptor.
1077 */
1078#ifndef _SYS_SYSPROTO_H_
1079struct ofstat_args {
1080 int fd;
1081 struct ostat *sb;
1082};
1083#endif
1084/* ARGSUSED */
1085int
1086ofstat(struct thread *td, struct ofstat_args *uap)
1087{
1088 struct ostat oub;
1089 struct stat ub;
1090 int error;
1091
1092 error = kern_fstat(td, uap->fd, &ub);
1093 if (error == 0) {
1094 cvtstat(&ub, &oub);
1095 error = copyout(&oub, uap->sb, sizeof(oub));
1096 }
1097 return (error);
1098}
1099#endif /* COMPAT_43 */
1100
1101/*
1102 * Return status information about a file descriptor.
1103 */
1104#ifndef _SYS_SYSPROTO_H_
1105struct fstat_args {
1106 int fd;
1107 struct stat *sb;
1108};
1109#endif
1110/* ARGSUSED */
1111int
1112fstat(struct thread *td, struct fstat_args *uap)
1113{
1114 struct stat ub;
1115 int error;
1116
1117 error = kern_fstat(td, uap->fd, &ub);
1118 if (error == 0)
1119 error = copyout(&ub, uap->sb, sizeof(ub));
1120 return (error);
1121}
1122
1123int
1124kern_fstat(struct thread *td, int fd, struct stat *sbp)
1125{
1126 struct file *fp;
1127 int error;
1128
1129 AUDIT_ARG(fd, fd);
1130
1131 if ((error = fget(td, fd, &fp)) != 0)
1132 return (error);
1133
1134 AUDIT_ARG(file, td->td_proc, fp);
1135
1136 error = fo_stat(fp, sbp, td->td_ucred, td);
1137 fdrop(fp, td);
1138#ifdef KTRACE
1139 if (error == 0 && KTRPOINT(td, KTR_STRUCT))
1140 ktrstat(sbp);
1141#endif
1142 return (error);
1143}
1144
1145/*
1146 * Return status information about a file descriptor.
1147 */
1148#ifndef _SYS_SYSPROTO_H_
1149struct nfstat_args {
1150 int fd;
1151 struct nstat *sb;
1152};
1153#endif
1154/* ARGSUSED */
1155int
1156nfstat(struct thread *td, struct nfstat_args *uap)
1157{
1158 struct nstat nub;
1159 struct stat ub;
1160 int error;
1161
1162 error = kern_fstat(td, uap->fd, &ub);
1163 if (error == 0) {
1164 cvtnstat(&ub, &nub);
1165 error = copyout(&nub, uap->sb, sizeof(nub));
1166 }
1167 return (error);
1168}
1169
1170/*
1171 * Return pathconf information about a file descriptor.
1172 */
1173#ifndef _SYS_SYSPROTO_H_
1174struct fpathconf_args {
1175 int fd;
1176 int name;
1177};
1178#endif
1179/* ARGSUSED */
1180int
1181fpathconf(struct thread *td, struct fpathconf_args *uap)
1182{
1183 struct file *fp;
1184 struct vnode *vp;
1185 int error;
1186
1187 if ((error = fget(td, uap->fd, &fp)) != 0)
1188 return (error);
1189
1190 /* If asynchronous I/O is available, it works for all descriptors. */
1191 if (uap->name == _PC_ASYNC_IO) {
1192 td->td_retval[0] = async_io_version;
1193 goto out;
1194 }
1195 vp = fp->f_vnode;
1196 if (vp != NULL) {
1197 int vfslocked;
1198 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
1199 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1200 error = VOP_PATHCONF(vp, uap->name, td->td_retval);
1201 VOP_UNLOCK(vp, 0);
1202 VFS_UNLOCK_GIANT(vfslocked);
1203 } else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) {
1204 if (uap->name != _PC_PIPE_BUF) {
1205 error = EINVAL;
1206 } else {
1207 td->td_retval[0] = PIPE_BUF;
1208 error = 0;
1209 }
1210 } else {
1211 error = EOPNOTSUPP;
1212 }
1213out:
1214 fdrop(fp, td);
1215 return (error);
1216}
1217
1218/*
1219 * Grow the file table to accomodate (at least) nfd descriptors. This may
1220 * block and drop the filedesc lock, but it will reacquire it before
1221 * returning.
1222 */
1223static void
1224fdgrowtable(struct filedesc *fdp, int nfd)
1225{
1226 struct file **ntable;
1227 char *nfileflags;
1228 int nnfiles, onfiles;
1229 NDSLOTTYPE *nmap;
1230
1231 FILEDESC_XLOCK_ASSERT(fdp);
1232
1233 KASSERT(fdp->fd_nfiles > 0,
1234 ("zero-length file table"));
1235
1236 /* compute the size of the new table */
1237 onfiles = fdp->fd_nfiles;
1238 nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */
1239 if (nnfiles <= onfiles)
1240 /* the table is already large enough */
1241 return;
1242
1243 /* allocate a new table and (if required) new bitmaps */
1244 FILEDESC_XUNLOCK(fdp);
1245 MALLOC(ntable, struct file **, nnfiles * OFILESIZE,
1246 M_FILEDESC, M_ZERO | M_WAITOK);
1247 nfileflags = (char *)&ntable[nnfiles];
1248 if (NDSLOTS(nnfiles) > NDSLOTS(onfiles))
1249 MALLOC(nmap, NDSLOTTYPE *, NDSLOTS(nnfiles) * NDSLOTSIZE,
1250 M_FILEDESC, M_ZERO | M_WAITOK);
1251 else
1252 nmap = NULL;
1253 FILEDESC_XLOCK(fdp);
1254
1255 /*
1256 * We now have new tables ready to go. Since we dropped the
1257 * filedesc lock to call malloc(), watch out for a race.
1258 */
1259 onfiles = fdp->fd_nfiles;
1260 if (onfiles >= nnfiles) {
1261 /* we lost the race, but that's OK */
1262 free(ntable, M_FILEDESC);
1263 if (nmap != NULL)
1264 free(nmap, M_FILEDESC);
1265 return;
1266 }
1267 bcopy(fdp->fd_ofiles, ntable, onfiles * sizeof(*ntable));
1268 bcopy(fdp->fd_ofileflags, nfileflags, onfiles);
1269 if (onfiles > NDFILE)
1270 free(fdp->fd_ofiles, M_FILEDESC);
1271 fdp->fd_ofiles = ntable;
1272 fdp->fd_ofileflags = nfileflags;
1273 if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) {
1274 bcopy(fdp->fd_map, nmap, NDSLOTS(onfiles) * sizeof(*nmap));
1275 if (NDSLOTS(onfiles) > NDSLOTS(NDFILE))
1276 free(fdp->fd_map, M_FILEDESC);
1277 fdp->fd_map = nmap;
1278 }
1279 fdp->fd_nfiles = nnfiles;
1280}
1281
1282/*
1283 * Allocate a file descriptor for the process.
1284 */
1285int
1286fdalloc(struct thread *td, int minfd, int *result)
1287{
1288 struct proc *p = td->td_proc;
1289 struct filedesc *fdp = p->p_fd;
1290 int fd = -1, maxfd;
1291
1292 FILEDESC_XLOCK_ASSERT(fdp);
1293
1294 if (fdp->fd_freefile > minfd)
1295 minfd = fdp->fd_freefile;
1296
1297 PROC_LOCK(p);
1298 maxfd = min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc);
1299 PROC_UNLOCK(p);
1300
1301 /*
1302 * Search the bitmap for a free descriptor. If none is found, try
1303 * to grow the file table. Keep at it until we either get a file
1304 * descriptor or run into process or system limits; fdgrowtable()
1305 * may drop the filedesc lock, so we're in a race.
1306 */
1307 for (;;) {
1308 fd = fd_first_free(fdp, minfd, fdp->fd_nfiles);
1309 if (fd >= maxfd)
1310 return (EMFILE);
1311 if (fd < fdp->fd_nfiles)
1312 break;
1313 fdgrowtable(fdp, min(fdp->fd_nfiles * 2, maxfd));
1314 }
1315
1316 /*
1317 * Perform some sanity checks, then mark the file descriptor as
1318 * used and return it to the caller.
1319 */
1320 KASSERT(!fdisused(fdp, fd),
1321 ("fd_first_free() returned non-free descriptor"));
1322 KASSERT(fdp->fd_ofiles[fd] == NULL,
1323 ("free descriptor isn't"));
1324 fdp->fd_ofileflags[fd] = 0; /* XXX needed? */
1325 fdused(fdp, fd);
1326 *result = fd;
1327 return (0);
1328}
1329
1330/*
1331 * Check to see whether n user file descriptors are available to the process
1332 * p.
1333 */
1334int
1335fdavail(struct thread *td, int n)
1336{
1337 struct proc *p = td->td_proc;
1338 struct filedesc *fdp = td->td_proc->p_fd;
1339 struct file **fpp;
1340 int i, lim, last;
1341
1342 FILEDESC_LOCK_ASSERT(fdp);
1343
1344 PROC_LOCK(p);
1345 lim = min((int)lim_cur(p, RLIMIT_NOFILE), maxfilesperproc);
1346 PROC_UNLOCK(p);
1347 if ((i = lim - fdp->fd_nfiles) > 0 && (n -= i) <= 0)
1348 return (1);
1349 last = min(fdp->fd_nfiles, lim);
1350 fpp = &fdp->fd_ofiles[fdp->fd_freefile];
1351 for (i = last - fdp->fd_freefile; --i >= 0; fpp++) {
1352 if (*fpp == NULL && --n <= 0)
1353 return (1);
1354 }
1355 return (0);
1356}
1357
1358/*
1359 * Create a new open file structure and allocate a file decriptor for the
1360 * process that refers to it. We add one reference to the file for the
1361 * descriptor table and one reference for resultfp. This is to prevent us
1362 * being preempted and the entry in the descriptor table closed after we
1363 * release the FILEDESC lock.
1364 */
1365int
1366falloc(struct thread *td, struct file **resultfp, int *resultfd)
1367{
1368 struct proc *p = td->td_proc;
1369 struct file *fp;
1370 int error, i;
1371 int maxuserfiles = maxfiles - (maxfiles / 20);
1372 static struct timeval lastfail;
1373 static int curfail;
1374
1375 fp = uma_zalloc(file_zone, M_WAITOK | M_ZERO);
1376 if ((openfiles >= maxuserfiles &&
1377 priv_check(td, PRIV_MAXFILES) != 0) ||
1378 openfiles >= maxfiles) {
1379 if (ppsratecheck(&lastfail, &curfail, 1)) {
1380 printf("kern.maxfiles limit exceeded by uid %i, please see tuning(7).\n",
1381 td->td_ucred->cr_ruid);
1382 }
1383 uma_zfree(file_zone, fp);
1384 return (ENFILE);
1385 }
1386 atomic_add_int(&openfiles, 1);
1387
1388 /*
1389 * If the process has file descriptor zero open, add the new file
1390 * descriptor to the list of open files at that point, otherwise
1391 * put it at the front of the list of open files.
1392 */
1393 fp->f_count = 1;
1394 if (resultfp)
1395 fp->f_count++;
1396 fp->f_cred = crhold(td->td_ucred);
1397 fp->f_ops = &badfileops;
1398 fp->f_data = NULL;
1399 fp->f_vnode = NULL;
1400 FILEDESC_XLOCK(p->p_fd);
1401 if ((error = fdalloc(td, 0, &i))) {
1402 FILEDESC_XUNLOCK(p->p_fd);
1403 fdrop(fp, td);
1404 if (resultfp)
1405 fdrop(fp, td);
1406 return (error);
1407 }
1408 p->p_fd->fd_ofiles[i] = fp;
1409 FILEDESC_XUNLOCK(p->p_fd);
1410 if (resultfp)
1411 *resultfp = fp;
1412 if (resultfd)
1413 *resultfd = i;
1414 return (0);
1415}
1416
1417/*
1418 * Build a new filedesc structure from another.
1419 * Copy the current, root, and jail root vnode references.
1420 */
1421struct filedesc *
1422fdinit(struct filedesc *fdp)
1423{
1424 struct filedesc0 *newfdp;
1425
1426 newfdp = malloc(sizeof *newfdp, M_FILEDESC, M_WAITOK | M_ZERO);
1427 FILEDESC_LOCK_INIT(&newfdp->fd_fd);
1428 if (fdp != NULL) {
1429 FILEDESC_XLOCK(fdp);
1430 newfdp->fd_fd.fd_cdir = fdp->fd_cdir;
1431 if (newfdp->fd_fd.fd_cdir)
1432 VREF(newfdp->fd_fd.fd_cdir);
1433 newfdp->fd_fd.fd_rdir = fdp->fd_rdir;
1434 if (newfdp->fd_fd.fd_rdir)
1435 VREF(newfdp->fd_fd.fd_rdir);
1436 newfdp->fd_fd.fd_jdir = fdp->fd_jdir;
1437 if (newfdp->fd_fd.fd_jdir)
1438 VREF(newfdp->fd_fd.fd_jdir);
1439 FILEDESC_XUNLOCK(fdp);
1440 }
1441
1442 /* Create the file descriptor table. */
1443 newfdp->fd_fd.fd_refcnt = 1;
1444 newfdp->fd_fd.fd_holdcnt = 1;
1445 newfdp->fd_fd.fd_cmask = CMASK;
1446 newfdp->fd_fd.fd_ofiles = newfdp->fd_dfiles;
1447 newfdp->fd_fd.fd_ofileflags = newfdp->fd_dfileflags;
1448 newfdp->fd_fd.fd_nfiles = NDFILE;
1449 newfdp->fd_fd.fd_map = newfdp->fd_dmap;
1450 newfdp->fd_fd.fd_lastfile = -1;
1451 return (&newfdp->fd_fd);
1452}
1453
1454static struct filedesc *
1455fdhold(struct proc *p)
1456{
1457 struct filedesc *fdp;
1458
1459 mtx_lock(&fdesc_mtx);
1460 fdp = p->p_fd;
1461 if (fdp != NULL)
1462 fdp->fd_holdcnt++;
1463 mtx_unlock(&fdesc_mtx);
1464 return (fdp);
1465}
1466
1467static void
1468fddrop(struct filedesc *fdp)
1469{
1470 int i;
1471
1472 mtx_lock(&fdesc_mtx);
1473 i = --fdp->fd_holdcnt;
1474 mtx_unlock(&fdesc_mtx);
1475 if (i > 0)
1476 return;
1477
1478 FILEDESC_LOCK_DESTROY(fdp);
1479 FREE(fdp, M_FILEDESC);
1480}
1481
1482/*
1483 * Share a filedesc structure.
1484 */
1485struct filedesc *
1486fdshare(struct filedesc *fdp)
1487{
1488
1489 FILEDESC_XLOCK(fdp);
1490 fdp->fd_refcnt++;
1491 FILEDESC_XUNLOCK(fdp);
1492 return (fdp);
1493}
1494
1495/*
1496 * Unshare a filedesc structure, if necessary by making a copy
1497 */
1498void
1499fdunshare(struct proc *p, struct thread *td)
1500{
1501
1502 FILEDESC_XLOCK(p->p_fd);
1503 if (p->p_fd->fd_refcnt > 1) {
1504 struct filedesc *tmp;
1505
1506 FILEDESC_XUNLOCK(p->p_fd);
1507 tmp = fdcopy(p->p_fd);
1508 fdfree(td);
1509 p->p_fd = tmp;
1510 } else
1511 FILEDESC_XUNLOCK(p->p_fd);
1512}
1513
1514/*
1515 * Copy a filedesc structure. A NULL pointer in returns a NULL reference,
1516 * this is to ease callers, not catch errors.
1517 */
1518struct filedesc *
1519fdcopy(struct filedesc *fdp)
1520{
1521 struct filedesc *newfdp;
1522 int i;
1523
1524 /* Certain daemons might not have file descriptors. */
1525 if (fdp == NULL)
1526 return (NULL);
1527
1528 newfdp = fdinit(fdp);
1529 FILEDESC_SLOCK(fdp);
1530 while (fdp->fd_lastfile >= newfdp->fd_nfiles) {
1531 FILEDESC_SUNLOCK(fdp);
1532 FILEDESC_XLOCK(newfdp);
1533 fdgrowtable(newfdp, fdp->fd_lastfile + 1);
1534 FILEDESC_XUNLOCK(newfdp);
1535 FILEDESC_SLOCK(fdp);
1536 }
1537 /* copy everything except kqueue descriptors */
1538 newfdp->fd_freefile = -1;
1539 for (i = 0; i <= fdp->fd_lastfile; ++i) {
1540 if (fdisused(fdp, i) &&
1541 fdp->fd_ofiles[i]->f_type != DTYPE_KQUEUE) {
1542 newfdp->fd_ofiles[i] = fdp->fd_ofiles[i];
1543 newfdp->fd_ofileflags[i] = fdp->fd_ofileflags[i];
1544 fhold(newfdp->fd_ofiles[i]);
1545 newfdp->fd_lastfile = i;
1546 } else {
1547 if (newfdp->fd_freefile == -1)
1548 newfdp->fd_freefile = i;
1549 }
1550 }
1551 FILEDESC_SUNLOCK(fdp);
1552 FILEDESC_XLOCK(newfdp);
1553 for (i = 0; i <= newfdp->fd_lastfile; ++i)
1554 if (newfdp->fd_ofiles[i] != NULL)
1555 fdused(newfdp, i);
1556 FILEDESC_XUNLOCK(newfdp);
1557 FILEDESC_SLOCK(fdp);
1558 if (newfdp->fd_freefile == -1)
1559 newfdp->fd_freefile = i;
1560 newfdp->fd_cmask = fdp->fd_cmask;
1561 FILEDESC_SUNLOCK(fdp);
1562 return (newfdp);
1563}
1564
1565/*
1566 * Release a filedesc structure.
1567 */
1568void
1569fdfree(struct thread *td)
1570{
1571 struct filedesc *fdp;
1572 struct file **fpp;
1573 int i, locked;
1574 struct filedesc_to_leader *fdtol;
1575 struct file *fp;
1576 struct vnode *cdir, *jdir, *rdir, *vp;
1577 struct flock lf;
1578
1579 /* Certain daemons might not have file descriptors. */
1580 fdp = td->td_proc->p_fd;
1581 if (fdp == NULL)
1582 return;
1583
1584 /* Check for special need to clear POSIX style locks */
1585 fdtol = td->td_proc->p_fdtol;
1586 if (fdtol != NULL) {
1587 FILEDESC_XLOCK(fdp);
1588 KASSERT(fdtol->fdl_refcount > 0,
1589 ("filedesc_to_refcount botch: fdl_refcount=%d",
1590 fdtol->fdl_refcount));
1591 if (fdtol->fdl_refcount == 1 &&
1592 (td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
1593 for (i = 0, fpp = fdp->fd_ofiles;
1594 i <= fdp->fd_lastfile;
1595 i++, fpp++) {
1596 if (*fpp == NULL ||
1597 (*fpp)->f_type != DTYPE_VNODE)
1598 continue;
1599 fp = *fpp;
1600 fhold(fp);
1601 FILEDESC_XUNLOCK(fdp);
1602 lf.l_whence = SEEK_SET;
1603 lf.l_start = 0;
1604 lf.l_len = 0;
1605 lf.l_type = F_UNLCK;
1606 vp = fp->f_vnode;
1607 locked = VFS_LOCK_GIANT(vp->v_mount);
1608 (void) VOP_ADVLOCK(vp,
1609 (caddr_t)td->td_proc->
1610 p_leader,
1611 F_UNLCK,
1612 &lf,
1613 F_POSIX);
1614 VFS_UNLOCK_GIANT(locked);
1615 FILEDESC_XLOCK(fdp);
1616 fdrop(fp, td);
1617 fpp = fdp->fd_ofiles + i;
1618 }
1619 }
1620 retry:
1621 if (fdtol->fdl_refcount == 1) {
1622 if (fdp->fd_holdleaderscount > 0 &&
1623 (td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
1624 /*
1625 * close() or do_dup() has cleared a reference
1626 * in a shared file descriptor table.
1627 */
1628 fdp->fd_holdleaderswakeup = 1;
1629 sx_sleep(&fdp->fd_holdleaderscount,
1630 FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0);
1631 goto retry;
1632 }
1633 if (fdtol->fdl_holdcount > 0) {
1634 /*
1635 * Ensure that fdtol->fdl_leader remains
1636 * valid in closef().
1637 */
1638 fdtol->fdl_wakeup = 1;
1639 sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK,
1640 "fdlhold", 0);
1641 goto retry;
1642 }
1643 }
1644 fdtol->fdl_refcount--;
1645 if (fdtol->fdl_refcount == 0 &&
1646 fdtol->fdl_holdcount == 0) {
1647 fdtol->fdl_next->fdl_prev = fdtol->fdl_prev;
1648 fdtol->fdl_prev->fdl_next = fdtol->fdl_next;
1649 } else
1650 fdtol = NULL;
1651 td->td_proc->p_fdtol = NULL;
1652 FILEDESC_XUNLOCK(fdp);
1653 if (fdtol != NULL)
1654 FREE(fdtol, M_FILEDESC_TO_LEADER);
1655 }
1656 FILEDESC_XLOCK(fdp);
1657 i = --fdp->fd_refcnt;
1658 FILEDESC_XUNLOCK(fdp);
1659 if (i > 0)
1660 return;
1661 /*
1662 * We are the last reference to the structure, so we can
1663 * safely assume it will not change out from under us.
1664 */
1665 fpp = fdp->fd_ofiles;
1666 for (i = fdp->fd_lastfile; i-- >= 0; fpp++) {
1667 if (*fpp)
1668 (void) closef(*fpp, td);
1669 }
1670 FILEDESC_XLOCK(fdp);
1671
1672 /* XXX This should happen earlier. */
1673 mtx_lock(&fdesc_mtx);
1674 td->td_proc->p_fd = NULL;
1675 mtx_unlock(&fdesc_mtx);
1676
1677 if (fdp->fd_nfiles > NDFILE)
1678 FREE(fdp->fd_ofiles, M_FILEDESC);
1679 if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE))
1680 FREE(fdp->fd_map, M_FILEDESC);
1681
1682 fdp->fd_nfiles = 0;
1683
1684 cdir = fdp->fd_cdir;
1685 fdp->fd_cdir = NULL;
1686 rdir = fdp->fd_rdir;
1687 fdp->fd_rdir = NULL;
1688 jdir = fdp->fd_jdir;
1689 fdp->fd_jdir = NULL;
1690 FILEDESC_XUNLOCK(fdp);
1691
1692 if (cdir) {
1693 locked = VFS_LOCK_GIANT(cdir->v_mount);
1694 vrele(cdir);
1695 VFS_UNLOCK_GIANT(locked);
1696 }
1697 if (rdir) {
1698 locked = VFS_LOCK_GIANT(rdir->v_mount);
1699 vrele(rdir);
1700 VFS_UNLOCK_GIANT(locked);
1701 }
1702 if (jdir) {
1703 locked = VFS_LOCK_GIANT(jdir->v_mount);
1704 vrele(jdir);
1705 VFS_UNLOCK_GIANT(locked);
1706 }
1707
1708 fddrop(fdp);
1709}
1710
1711/*
1712 * For setugid programs, we don't want to people to use that setugidness
1713 * to generate error messages which write to a file which otherwise would
1714 * otherwise be off-limits to the process. We check for filesystems where
1715 * the vnode can change out from under us after execve (like [lin]procfs).
1716 *
1717 * Since setugidsafety calls this only for fd 0, 1 and 2, this check is
1718 * sufficient. We also don't check for setugidness since we know we are.
1719 */
1720static int
1721is_unsafe(struct file *fp)
1722{
1723 if (fp->f_type == DTYPE_VNODE) {
1724 struct vnode *vp = fp->f_vnode;
1725
1726 if ((vp->v_vflag & VV_PROCDEP) != 0)
1727 return (1);
1728 }
1729 return (0);
1730}
1731
1732/*
1733 * Make this setguid thing safe, if at all possible.
1734 */
1735void
1736setugidsafety(struct thread *td)
1737{
1738 struct filedesc *fdp;
1739 int i;
1740
1741 /* Certain daemons might not have file descriptors. */
1742 fdp = td->td_proc->p_fd;
1743 if (fdp == NULL)
1744 return;
1745
1746 /*
1747 * Note: fdp->fd_ofiles may be reallocated out from under us while
1748 * we are blocked in a close. Be careful!
1749 */
1750 FILEDESC_XLOCK(fdp);
1751 for (i = 0; i <= fdp->fd_lastfile; i++) {
1752 if (i > 2)
1753 break;
1754 if (fdp->fd_ofiles[i] && is_unsafe(fdp->fd_ofiles[i])) {
1755 struct file *fp;
1756
1757 knote_fdclose(td, i);
1758 /*
1759 * NULL-out descriptor prior to close to avoid
1760 * a race while close blocks.
1761 */
1762 fp = fdp->fd_ofiles[i];
1763 fdp->fd_ofiles[i] = NULL;
1764 fdp->fd_ofileflags[i] = 0;
1765 fdunused(fdp, i);
1766 FILEDESC_XUNLOCK(fdp);
1767 (void) closef(fp, td);
1768 FILEDESC_XLOCK(fdp);
1769 }
1770 }
1771 FILEDESC_XUNLOCK(fdp);
1772}
1773
1774/*
1775 * If a specific file object occupies a specific file descriptor, close the
1776 * file descriptor entry and drop a reference on the file object. This is a
1777 * convenience function to handle a subsequent error in a function that calls
1778 * falloc() that handles the race that another thread might have closed the
1779 * file descriptor out from under the thread creating the file object.
1780 */
1781void
1782fdclose(struct filedesc *fdp, struct file *fp, int idx, struct thread *td)
1783{
1784
1785 FILEDESC_XLOCK(fdp);
1786 if (fdp->fd_ofiles[idx] == fp) {
1787 fdp->fd_ofiles[idx] = NULL;
1788 fdunused(fdp, idx);
1789 FILEDESC_XUNLOCK(fdp);
1790 fdrop(fp, td);
1791 } else
1792 FILEDESC_XUNLOCK(fdp);
1793}
1794
1795/*
1796 * Close any files on exec?
1797 */
1798void
1799fdcloseexec(struct thread *td)
1800{
1801 struct filedesc *fdp;
1802 int i;
1803
1804 /* Certain daemons might not have file descriptors. */
1805 fdp = td->td_proc->p_fd;
1806 if (fdp == NULL)
1807 return;
1808
1809 FILEDESC_XLOCK(fdp);
1810
1811 /*
1812 * We cannot cache fd_ofiles or fd_ofileflags since operations
1813 * may block and rip them out from under us.
1814 */
1815 for (i = 0; i <= fdp->fd_lastfile; i++) {
1816 if (fdp->fd_ofiles[i] != NULL &&
1817 (fdp->fd_ofiles[i]->f_type == DTYPE_MQUEUE ||
1818 (fdp->fd_ofileflags[i] & UF_EXCLOSE))) {
1819 struct file *fp;
1820
1821 knote_fdclose(td, i);
1822 /*
1823 * NULL-out descriptor prior to close to avoid
1824 * a race while close blocks.
1825 */
1826 fp = fdp->fd_ofiles[i];
1827 fdp->fd_ofiles[i] = NULL;
1828 fdp->fd_ofileflags[i] = 0;
1829 fdunused(fdp, i);
1830 if (fp->f_type == DTYPE_MQUEUE)
1831 mq_fdclose(td, i, fp);
1832 FILEDESC_XUNLOCK(fdp);
1833 (void) closef(fp, td);
1834 FILEDESC_XLOCK(fdp);
1835 }
1836 }
1837 FILEDESC_XUNLOCK(fdp);
1838}
1839
1840/*
1841 * It is unsafe for set[ug]id processes to be started with file
1842 * descriptors 0..2 closed, as these descriptors are given implicit
1843 * significance in the Standard C library. fdcheckstd() will create a
1844 * descriptor referencing /dev/null for each of stdin, stdout, and
1845 * stderr that is not already open.
1846 */
1847int
1848fdcheckstd(struct thread *td)
1849{
1850 struct filedesc *fdp;
1851 register_t retval, save;
1852 int i, error, devnull;
1853
1854 fdp = td->td_proc->p_fd;
1855 if (fdp == NULL)
1856 return (0);
1857 KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared"));
1858 devnull = -1;
1859 error = 0;
1860 for (i = 0; i < 3; i++) {
1861 if (fdp->fd_ofiles[i] != NULL)
1862 continue;
1863 if (devnull < 0) {
1864 save = td->td_retval[0];
1865 error = kern_open(td, "/dev/null", UIO_SYSSPACE,
1866 O_RDWR, 0);
1867 devnull = td->td_retval[0];
1868 KASSERT(devnull == i, ("oof, we didn't get our fd"));
1869 td->td_retval[0] = save;
1870 if (error)
1871 break;
1872 } else {
1873 error = do_dup(td, DUP_FIXED, devnull, i, &retval);
1874 if (error != 0)
1875 break;
1876 }
1877 }
1878 return (error);
1879}
1880
1881/*
1882 * Internal form of close. Decrement reference count on file structure.
1883 * Note: td may be NULL when closing a file that was being passed in a
1884 * message.
1885 *
1886 * XXXRW: Giant is not required for the caller, but often will be held; this
1887 * makes it moderately likely the Giant will be recursed in the VFS case.
1888 */
1889int
1890closef(struct file *fp, struct thread *td)
1891{
1892 struct vnode *vp;
1893 struct flock lf;
1894 struct filedesc_to_leader *fdtol;
1895 struct filedesc *fdp;
1896
1897 /*
1898 * POSIX record locking dictates that any close releases ALL
1899 * locks owned by this process. This is handled by setting
1900 * a flag in the unlock to free ONLY locks obeying POSIX
1901 * semantics, and not to free BSD-style file locks.
1902 * If the descriptor was in a message, POSIX-style locks
1903 * aren't passed with the descriptor, and the thread pointer
1904 * will be NULL. Callers should be careful only to pass a
1905 * NULL thread pointer when there really is no owning
1906 * context that might have locks, or the locks will be
1907 * leaked.
1908 */
1909 if (fp->f_type == DTYPE_VNODE && td != NULL) {
1910 int vfslocked;
1911
1912 vp = fp->f_vnode;
1913 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
1914 if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) {
1915 lf.l_whence = SEEK_SET;
1916 lf.l_start = 0;
1917 lf.l_len = 0;
1918 lf.l_type = F_UNLCK;
1919 (void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader,
1920 F_UNLCK, &lf, F_POSIX);
1921 }
1922 fdtol = td->td_proc->p_fdtol;
1923 if (fdtol != NULL) {
1924 /*
1925 * Handle special case where file descriptor table is
1926 * shared between multiple process leaders.
1927 */
1928 fdp = td->td_proc->p_fd;
1929 FILEDESC_XLOCK(fdp);
1930 for (fdtol = fdtol->fdl_next;
1931 fdtol != td->td_proc->p_fdtol;
1932 fdtol = fdtol->fdl_next) {
1933 if ((fdtol->fdl_leader->p_flag &
1934 P_ADVLOCK) == 0)
1935 continue;
1936 fdtol->fdl_holdcount++;
1937 FILEDESC_XUNLOCK(fdp);
1938 lf.l_whence = SEEK_SET;
1939 lf.l_start = 0;
1940 lf.l_len = 0;
1941 lf.l_type = F_UNLCK;
1942 vp = fp->f_vnode;
1943 (void) VOP_ADVLOCK(vp,
1944 (caddr_t)fdtol->fdl_leader,
1945 F_UNLCK, &lf, F_POSIX);
1946 FILEDESC_XLOCK(fdp);
1947 fdtol->fdl_holdcount--;
1948 if (fdtol->fdl_holdcount == 0 &&
1949 fdtol->fdl_wakeup != 0) {
1950 fdtol->fdl_wakeup = 0;
1951 wakeup(fdtol);
1952 }
1953 }
1954 FILEDESC_XUNLOCK(fdp);
1955 }
1956 VFS_UNLOCK_GIANT(vfslocked);
1957 }
1958 return (fdrop(fp, td));
1959}
1960
1961/*
1962 * Initialize the file pointer with the specified properties.
1963 *
1964 * The ops are set with release semantics to be certain that the flags, type,
1965 * and data are visible when ops is. This is to prevent ops methods from being
1966 * called with bad data.
1967 */
1968void
1969finit(struct file *fp, u_int flag, short type, void *data, struct fileops *ops)
1970{
1971 fp->f_data = data;
1972 fp->f_flag = flag;
1973 fp->f_type = type;
1974 atomic_store_rel_ptr((volatile uintptr_t *)&fp->f_ops, (uintptr_t)ops);
1975}
1976
1977
1978/*
1979 * Extract the file pointer associated with the specified descriptor for the
1980 * current user process.
1981 *
1982 * If the descriptor doesn't exist, EBADF is returned.
1983 *
1984 * If the descriptor exists but doesn't match 'flags' then return EBADF for
1985 * read attempts and EINVAL for write attempts.
1986 *
1987 * If 'hold' is set (non-zero) the file's refcount will be bumped on return.
1988 * It should be dropped with fdrop(). If it is not set, then the refcount
1989 * will not be bumped however the thread's filedesc struct will be returned
1990 * locked (for fgetsock).
1991 *
1992 * If an error occured the non-zero error is returned and *fpp is set to
1993 * NULL. Otherwise *fpp is set and zero is returned.
1994 */
1995static __inline int
1996_fget(struct thread *td, int fd, struct file **fpp, int flags, int hold)
1997{
1998 struct filedesc *fdp;
1999 struct file *fp;
2000
2001 *fpp = NULL;
2002 if (td == NULL || (fdp = td->td_proc->p_fd) == NULL)
2003 return (EBADF);
2004 FILEDESC_SLOCK(fdp);
2005 if ((fp = fget_locked(fdp, fd)) == NULL || fp->f_ops == &badfileops) {
2006 FILEDESC_SUNLOCK(fdp);
2007 return (EBADF);
2008 }
2009
2010 /*
2011 * FREAD and FWRITE failure return EBADF as per POSIX.
2012 *
2013 * Only one flag, or 0, may be specified.
2014 */
2015 if (flags == FREAD && (fp->f_flag & FREAD) == 0) {
2016 FILEDESC_SUNLOCK(fdp);
2017 return (EBADF);
2018 }
2019 if (flags == FWRITE && (fp->f_flag & FWRITE) == 0) {
2020 FILEDESC_SUNLOCK(fdp);
2021 return (EBADF);
2022 }
2023 if (hold) {
2024 fhold(fp);
2025 FILEDESC_SUNLOCK(fdp);
2026 }
2027 *fpp = fp;
2028 return (0);
2029}
2030
2031int
2032fget(struct thread *td, int fd, struct file **fpp)
2033{
2034
2035 return(_fget(td, fd, fpp, 0, 1));
2036}
2037
2038int
2039fget_read(struct thread *td, int fd, struct file **fpp)
2040{
2041
2042 return(_fget(td, fd, fpp, FREAD, 1));
2043}
2044
2045int
2046fget_write(struct thread *td, int fd, struct file **fpp)
2047{
2048
2049 return(_fget(td, fd, fpp, FWRITE, 1));
2050}
2051
2052/*
2053 * Like fget() but loads the underlying vnode, or returns an error if the
2054 * descriptor does not represent a vnode. Note that pipes use vnodes but
2055 * never have VM objects. The returned vnode will be vref()'d.
2056 *
2057 * XXX: what about the unused flags ?
2058 */
2059static __inline int
2060_fgetvp(struct thread *td, int fd, struct vnode **vpp, int flags)
2061{
2062 struct file *fp;
2063 int error;
2064
2065 *vpp = NULL;
2066 if ((error = _fget(td, fd, &fp, flags, 0)) != 0)
2067 return (error);
2068 if (fp->f_vnode == NULL) {
2069 error = EINVAL;
2070 } else {
2071 *vpp = fp->f_vnode;
2072 vref(*vpp);
2073 }
2074 FILEDESC_SUNLOCK(td->td_proc->p_fd);
2075 return (error);
2076}
2077
2078int
2079fgetvp(struct thread *td, int fd, struct vnode **vpp)
2080{
2081
2082 return (_fgetvp(td, fd, vpp, 0));
2083}
2084
2085int
2086fgetvp_read(struct thread *td, int fd, struct vnode **vpp)
2087{
2088
2089 return (_fgetvp(td, fd, vpp, FREAD));
2090}
2091
2092#ifdef notyet
2093int
2094fgetvp_write(struct thread *td, int fd, struct vnode **vpp)
2095{
2096
2097 return (_fgetvp(td, fd, vpp, FWRITE));
2098}
2099#endif
2100
2101/*
2102 * Like fget() but loads the underlying socket, or returns an error if the
2103 * descriptor does not represent a socket.
2104 *
2105 * We bump the ref count on the returned socket. XXX Also obtain the SX lock
2106 * in the future.
2107 *
2108 * XXXRW: fgetsock() and fputsock() are deprecated, as consumers should rely
2109 * on their file descriptor reference to prevent the socket from being free'd
2110 * during use.
2111 */
2112int
2113fgetsock(struct thread *td, int fd, struct socket **spp, u_int *fflagp)
2114{
2115 struct file *fp;
2116 int error;
2117
2118 *spp = NULL;
2119 if (fflagp != NULL)
2120 *fflagp = 0;
2121 if ((error = _fget(td, fd, &fp, 0, 0)) != 0)
2122 return (error);
2123 if (fp->f_type != DTYPE_SOCKET) {
2124 error = ENOTSOCK;
2125 } else {
2126 *spp = fp->f_data;
2127 if (fflagp)
2128 *fflagp = fp->f_flag;
2129 SOCK_LOCK(*spp);
2130 soref(*spp);
2131 SOCK_UNLOCK(*spp);
2132 }
2133 FILEDESC_SUNLOCK(td->td_proc->p_fd);
2134 return (error);
2135}
2136
2137/*
2138 * Drop the reference count on the socket and XXX release the SX lock in the
2139 * future. The last reference closes the socket.
2140 *
2141 * XXXRW: fputsock() is deprecated, see comment for fgetsock().
2142 */
2143void
2144fputsock(struct socket *so)
2145{
2146
2147 ACCEPT_LOCK();
2148 SOCK_LOCK(so);
2149 sorele(so);
2150}
2151
2152/*
2153 * Handle the last reference to a file being closed.
2154 */
2155int
2156_fdrop(struct file *fp, struct thread *td)
2157{
2158 int error;
2159
2160 error = 0;
2161 if (fp->f_count != 0)
2162 panic("fdrop: count %d", fp->f_count);
2163 if (fp->f_ops != &badfileops)
2164 error = fo_close(fp, td);
2165 atomic_subtract_int(&openfiles, 1);
2166 crfree(fp->f_cred);
2167 uma_zfree(file_zone, fp);
2168
2169 return (error);
2170}
2171
2172/*
2173 * Apply an advisory lock on a file descriptor.
2174 *
2175 * Just attempt to get a record lock of the requested type on the entire file
2176 * (l_whence = SEEK_SET, l_start = 0, l_len = 0).
2177 */
2178#ifndef _SYS_SYSPROTO_H_
2179struct flock_args {
2180 int fd;
2181 int how;
2182};
2183#endif
2184/* ARGSUSED */
2185int
2186flock(struct thread *td, struct flock_args *uap)
2187{
2188 struct file *fp;
2189 struct vnode *vp;
2190 struct flock lf;
2191 int vfslocked;
2192 int error;
2193
2194 if ((error = fget(td, uap->fd, &fp)) != 0)
2195 return (error);
2196 if (fp->f_type != DTYPE_VNODE) {
2197 fdrop(fp, td);
2198 return (EOPNOTSUPP);
2199 }
2200
2201 vp = fp->f_vnode;
2202 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
2203 lf.l_whence = SEEK_SET;
2204 lf.l_start = 0;
2205 lf.l_len = 0;
2206 if (uap->how & LOCK_UN) {
2207 lf.l_type = F_UNLCK;
2208 atomic_clear_int(&fp->f_flag, FHASLOCK);
2209 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK);
2210 goto done2;
2211 }
2212 if (uap->how & LOCK_EX)
2213 lf.l_type = F_WRLCK;
2214 else if (uap->how & LOCK_SH)
2215 lf.l_type = F_RDLCK;
2216 else {
2217 error = EBADF;
2218 goto done2;
2219 }
2220 atomic_set_int(&fp->f_flag, FHASLOCK);
2221 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf,
2222 (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT);
2223done2:
2224 fdrop(fp, td);
2225 VFS_UNLOCK_GIANT(vfslocked);
2226 return (error);
2227}
2228/*
2229 * Duplicate the specified descriptor to a free descriptor.
2230 */
2231int
2232dupfdopen(struct thread *td, struct filedesc *fdp, int indx, int dfd, int mode, int error)
2233{
2234 struct file *wfp;
2235 struct file *fp;
2236
2237 /*
2238 * If the to-be-dup'd fd number is greater than the allowed number
2239 * of file descriptors, or the fd to be dup'd has already been
2240 * closed, then reject.
2241 */
2242 FILEDESC_XLOCK(fdp);
2243 if (dfd < 0 || dfd >= fdp->fd_nfiles ||
2244 (wfp = fdp->fd_ofiles[dfd]) == NULL) {
2245 FILEDESC_XUNLOCK(fdp);
2246 return (EBADF);
2247 }
2248
2249 /*
2250 * There are two cases of interest here.
2251 *
2252 * For ENODEV simply dup (dfd) to file descriptor (indx) and return.
2253 *
2254 * For ENXIO steal away the file structure from (dfd) and store it in
2255 * (indx). (dfd) is effectively closed by this operation.
2256 *
2257 * Any other error code is just returned.
2258 */
2259 switch (error) {
2260 case ENODEV:
2261 /*
2262 * Check that the mode the file is being opened for is a
2263 * subset of the mode of the existing descriptor.
2264 */
2265 if (((mode & (FREAD|FWRITE)) | wfp->f_flag) != wfp->f_flag) {
2266 FILEDESC_XUNLOCK(fdp);
2267 return (EACCES);
2268 }
2269 fp = fdp->fd_ofiles[indx];
2270 fdp->fd_ofiles[indx] = wfp;
2271 fdp->fd_ofileflags[indx] = fdp->fd_ofileflags[dfd];
2272 if (fp == NULL)
2273 fdused(fdp, indx);
2274 fhold(wfp);
2275 FILEDESC_XUNLOCK(fdp);
2276 if (fp != NULL)
2277 /*
2278 * We now own the reference to fp that the ofiles[]
2279 * array used to own. Release it.
2280 */
2281 fdrop(fp, td);
2282 return (0);
2283
2284 case ENXIO:
2285 /*
2286 * Steal away the file pointer from dfd and stuff it into indx.
2287 */
2288 fp = fdp->fd_ofiles[indx];
2289 fdp->fd_ofiles[indx] = fdp->fd_ofiles[dfd];
2290 fdp->fd_ofiles[dfd] = NULL;
2291 fdp->fd_ofileflags[indx] = fdp->fd_ofileflags[dfd];
2292 fdp->fd_ofileflags[dfd] = 0;
2293 fdunused(fdp, dfd);
2294 if (fp == NULL)
2295 fdused(fdp, indx);
2296 FILEDESC_XUNLOCK(fdp);
2297
2298 /*
2299 * We now own the reference to fp that the ofiles[] array
2300 * used to own. Release it.
2301 */
2302 if (fp != NULL)
2303 fdrop(fp, td);
2304 return (0);
2305
2306 default:
2307 FILEDESC_XUNLOCK(fdp);
2308 return (error);
2309 }
2310 /* NOTREACHED */
2311}
2312
2313/*
2314 * Scan all active processes to see if any of them have a current or root
2315 * directory of `olddp'. If so, replace them with the new mount point.
2316 */
2317void
2318mountcheckdirs(struct vnode *olddp, struct vnode *newdp)
2319{
2320 struct filedesc *fdp;
2321 struct proc *p;
2322 int nrele;
2323
2324 if (vrefcnt(olddp) == 1)
2325 return;
2326 sx_slock(&allproc_lock);
2327 FOREACH_PROC_IN_SYSTEM(p) {
2328 fdp = fdhold(p);
2329 if (fdp == NULL)
2330 continue;
2331 nrele = 0;
2332 FILEDESC_XLOCK(fdp);
2333 if (fdp->fd_cdir == olddp) {
2334 vref(newdp);
2335 fdp->fd_cdir = newdp;
2336 nrele++;
2337 }
2338 if (fdp->fd_rdir == olddp) {
2339 vref(newdp);
2340 fdp->fd_rdir = newdp;
2341 nrele++;
2342 }
2343 FILEDESC_XUNLOCK(fdp);
2344 fddrop(fdp);
2345 while (nrele--)
2346 vrele(olddp);
2347 }
2348 sx_sunlock(&allproc_lock);
2349 if (rootvnode == olddp) {
2350 vrele(rootvnode);
2351 vref(newdp);
2352 rootvnode = newdp;
2353 }
2354}
2355
2356struct filedesc_to_leader *
2357filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader)
2358{
2359 struct filedesc_to_leader *fdtol;
2360
2361 MALLOC(fdtol, struct filedesc_to_leader *,
2362 sizeof(struct filedesc_to_leader),
2363 M_FILEDESC_TO_LEADER,
2364 M_WAITOK);
2365 fdtol->fdl_refcount = 1;
2366 fdtol->fdl_holdcount = 0;
2367 fdtol->fdl_wakeup = 0;
2368 fdtol->fdl_leader = leader;
2369 if (old != NULL) {
2370 FILEDESC_XLOCK(fdp);
2371 fdtol->fdl_next = old->fdl_next;
2372 fdtol->fdl_prev = old;
2373 old->fdl_next = fdtol;
2374 fdtol->fdl_next->fdl_prev = fdtol;
2375 FILEDESC_XUNLOCK(fdp);
2376 } else {
2377 fdtol->fdl_next = fdtol;
2378 fdtol->fdl_prev = fdtol;
2379 }
2380 return (fdtol);
2381}
2382
2383/*
2384 * Get file structures globally.
2385 */
2386static int
2387sysctl_kern_file(SYSCTL_HANDLER_ARGS)
2388{
2389 struct xfile xf;
2390 struct filedesc *fdp;
2391 struct file *fp;
2392 struct proc *p;
2393 int error, n;
2394
2395 error = sysctl_wire_old_buffer(req, 0);
2396 if (error != 0)
2397 return (error);
2398 if (req->oldptr == NULL) {
2399 n = 0;
2400 sx_slock(&allproc_lock);
2401 FOREACH_PROC_IN_SYSTEM(p) {
2402 if (p->p_state == PRS_NEW)
2403 continue;
2404 fdp = fdhold(p);
2405 if (fdp == NULL)
2406 continue;
2407 /* overestimates sparse tables. */
2408 if (fdp->fd_lastfile > 0)
2409 n += fdp->fd_lastfile;
2410 fddrop(fdp);
2411 }
2412 sx_sunlock(&allproc_lock);
2413 return (SYSCTL_OUT(req, 0, n * sizeof(xf)));
2414 }
2415 error = 0;
2416 bzero(&xf, sizeof(xf));
2417 xf.xf_size = sizeof(xf);
2418 sx_slock(&allproc_lock);
2419 FOREACH_PROC_IN_SYSTEM(p) {
2420 if (p->p_state == PRS_NEW)
2421 continue;
2422 PROC_LOCK(p);
2423 if (p_cansee(req->td, p) != 0) {
2424 PROC_UNLOCK(p);
2425 continue;
2426 }
2427 xf.xf_pid = p->p_pid;
2428 xf.xf_uid = p->p_ucred->cr_uid;
2429 PROC_UNLOCK(p);
2430 fdp = fdhold(p);
2431 if (fdp == NULL)
2432 continue;
2433 FILEDESC_SLOCK(fdp);
2434 for (n = 0; fdp->fd_refcnt > 0 && n < fdp->fd_nfiles; ++n) {
2435 if ((fp = fdp->fd_ofiles[n]) == NULL)
2436 continue;
2437 xf.xf_fd = n;
2438 xf.xf_file = fp;
2439 xf.xf_data = fp->f_data;
2440 xf.xf_vnode = fp->f_vnode;
2441 xf.xf_type = fp->f_type;
2442 xf.xf_count = fp->f_count;
2443 xf.xf_msgcount = 0;
2444 xf.xf_offset = fp->f_offset;
2445 xf.xf_flag = fp->f_flag;
2446 error = SYSCTL_OUT(req, &xf, sizeof(xf));
2447 if (error)
2448 break;
2449 }
2450 FILEDESC_SUNLOCK(fdp);
2451 fddrop(fdp);
2452 if (error)
2453 break;
2454 }
2455 sx_sunlock(&allproc_lock);
2456 return (error);
2457}
2458
2459SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD,
2460 0, 0, sysctl_kern_file, "S,xfile", "Entire file table");
2461
2462static int
2463export_vnode_for_sysctl(struct vnode *vp, int type,
2464 struct kinfo_file *kif, struct filedesc *fdp, struct sysctl_req *req)
2465{
2466 int error;
2467 char *fullpath, *freepath;
2468 int vfslocked;
2469
2470 bzero(kif, sizeof(*kif));
2471 kif->kf_structsize = sizeof(*kif);
2472
2473 vref(vp);
2474 kif->kf_fd = type;
2475 kif->kf_type = KF_TYPE_VNODE;
2476 /* This function only handles directories. */
2477 KASSERT(vp->v_type == VDIR, ("export_vnode_for_sysctl: vnode not directory"));
2478 kif->kf_vnode_type = KF_VTYPE_VDIR;
2479
2480 /*
2481 * This is not a true file descriptor, so we set a bogus refcount
2482 * and offset to indicate these fields should be ignored.
2483 */
2484 kif->kf_ref_count = -1;
2485 kif->kf_offset = -1;
2486
2487 freepath = NULL;
2488 fullpath = "-";
2489 FILEDESC_SUNLOCK(fdp);
2490 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
2491 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2492 vn_fullpath(curthread, vp, &fullpath, &freepath);
2493 vput(vp);
2494 VFS_UNLOCK_GIANT(vfslocked);
2495 strlcpy(kif->kf_path, fullpath, sizeof(kif->kf_path));
2496 if (freepath != NULL)
2497 free(freepath, M_TEMP);
2498 error = SYSCTL_OUT(req, kif, sizeof(*kif));
2499 FILEDESC_SLOCK(fdp);
2500 return (error);
2501}
2502
2503/*
2504 * Get per-process file descriptors for use by procstat(1), et al.
2505 */
2506static int
2507sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS)
2508{
2509 char *fullpath, *freepath;
2510 struct kinfo_file *kif;
2511 struct filedesc *fdp;
2512 int error, i, *name;
2513 struct socket *so;
2514 struct vnode *vp;
2515 struct file *fp;
2516 struct proc *p;
2517 int vfslocked;
2518
2519 name = (int *)arg1;
2520 if ((p = pfind((pid_t)name[0])) == NULL)
2521 return (ESRCH);
2522 if ((error = p_candebug(curthread, p))) {
2523 PROC_UNLOCK(p);
2524 return (error);
2525 }
2526 fdp = fdhold(p);
2527 PROC_UNLOCK(p);
2528 kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK);
2529 FILEDESC_SLOCK(fdp);
2530 if (fdp->fd_cdir != NULL)
2531 export_vnode_for_sysctl(fdp->fd_cdir, KF_FD_TYPE_CWD, kif,
2532 fdp, req);
2533 if (fdp->fd_rdir != NULL)
2534 export_vnode_for_sysctl(fdp->fd_rdir, KF_FD_TYPE_ROOT, kif,
2535 fdp, req);
2536 if (fdp->fd_jdir != NULL)
2537 export_vnode_for_sysctl(fdp->fd_jdir, KF_FD_TYPE_JAIL, kif,
2538 fdp, req);
2539 for (i = 0; i < fdp->fd_nfiles; i++) {
2540 if ((fp = fdp->fd_ofiles[i]) == NULL)
2541 continue;
2542 bzero(kif, sizeof(*kif));
2543 kif->kf_structsize = sizeof(*kif);
2544 vp = NULL;
2545 so = NULL;
2546 kif->kf_fd = i;
2547 switch (fp->f_type) {
2548 case DTYPE_VNODE:
2549 kif->kf_type = KF_TYPE_VNODE;
2550 vp = fp->f_vnode;
2551 break;
2552
2553 case DTYPE_SOCKET:
2554 kif->kf_type = KF_TYPE_SOCKET;
2555 so = fp->f_data;
2556 break;
2557
2558 case DTYPE_PIPE:
2559 kif->kf_type = KF_TYPE_PIPE;
2560 break;
2561
2562 case DTYPE_FIFO:
2563 kif->kf_type = KF_TYPE_FIFO;
2564 vp = fp->f_vnode;
2565 vref(vp);
2566 break;
2567
2568 case DTYPE_KQUEUE:
2569 kif->kf_type = KF_TYPE_KQUEUE;
2570 break;
2571
2572 case DTYPE_CRYPTO:
2573 kif->kf_type = KF_TYPE_CRYPTO;
2574 break;
2575
2576 case DTYPE_MQUEUE:
2577 kif->kf_type = KF_TYPE_MQUEUE;
2578 break;
2579
2580 case DTYPE_SHM:
2581 kif->kf_type = KF_TYPE_SHM;
2582 break;
2583
2584 default:
2585 kif->kf_type = KF_TYPE_UNKNOWN;
2586 break;
2587 }
2588 kif->kf_ref_count = fp->f_count;
2589 if (fp->f_flag & FREAD)
2590 kif->kf_flags |= KF_FLAG_READ;
2591 if (fp->f_flag & FWRITE)
2592 kif->kf_flags |= KF_FLAG_WRITE;
2593 if (fp->f_flag & FAPPEND)
2594 kif->kf_flags |= KF_FLAG_APPEND;
2595 if (fp->f_flag & FASYNC)
2596 kif->kf_flags |= KF_FLAG_ASYNC;
2597 if (fp->f_flag & FFSYNC)
2598 kif->kf_flags |= KF_FLAG_FSYNC;
2599 if (fp->f_flag & FNONBLOCK)
2600 kif->kf_flags |= KF_FLAG_NONBLOCK;
2601 if (fp->f_flag & O_DIRECT)
2602 kif->kf_flags |= KF_FLAG_DIRECT;
2603 if (fp->f_flag & FHASLOCK)
2604 kif->kf_flags |= KF_FLAG_HASLOCK;
2605 kif->kf_offset = fp->f_offset;
2606 if (vp != NULL) {
2607 vref(vp);
2608 switch (vp->v_type) {
2609 case VNON:
2610 kif->kf_vnode_type = KF_VTYPE_VNON;
2611 break;
2612 case VREG:
2613 kif->kf_vnode_type = KF_VTYPE_VREG;
2614 break;
2615 case VDIR:
2616 kif->kf_vnode_type = KF_VTYPE_VDIR;
2617 break;
2618 case VBLK:
2619 kif->kf_vnode_type = KF_VTYPE_VBLK;
2620 break;
2621 case VCHR:
2622 kif->kf_vnode_type = KF_VTYPE_VCHR;
2623 break;
2624 case VLNK:
2625 kif->kf_vnode_type = KF_VTYPE_VLNK;
2626 break;
2627 case VSOCK:
2628 kif->kf_vnode_type = KF_VTYPE_VSOCK;
2629 break;
2630 case VFIFO:
2631 kif->kf_vnode_type = KF_VTYPE_VFIFO;
2632 break;
2633 case VBAD:
2634 kif->kf_vnode_type = KF_VTYPE_VBAD;
2635 break;
2636 default:
2637 kif->kf_vnode_type = KF_VTYPE_UNKNOWN;
2638 break;
2639 }
2640 /*
2641 * It is OK to drop the filedesc lock here as we will
2642 * re-validate and re-evaluate its properties when
2643 * the loop continues.
2644 */
2645 freepath = NULL;
2646 fullpath = "-";
2647 FILEDESC_SUNLOCK(fdp);
2648 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
2649 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2650 vn_fullpath(curthread, vp, &fullpath, &freepath);
2651 vput(vp);
2652 VFS_UNLOCK_GIANT(vfslocked);
2653 strlcpy(kif->kf_path, fullpath,
2654 sizeof(kif->kf_path));
2655 if (freepath != NULL)
2656 free(freepath, M_TEMP);
2657 FILEDESC_SLOCK(fdp);
2658 }
2659 if (so != NULL) {
2660 struct sockaddr *sa;
2661
2662 if (so->so_proto->pr_usrreqs->pru_sockaddr(so, &sa)
2663 == 0 && sa->sa_len <= sizeof(kif->kf_sa_local)) {
2664 bcopy(sa, &kif->kf_sa_local, sa->sa_len);
2665 free(sa, M_SONAME);
2666 }
2667 if (so->so_proto->pr_usrreqs->pru_peeraddr(so, &sa)
2668 == 00 && sa->sa_len <= sizeof(kif->kf_sa_peer)) {
2669 bcopy(sa, &kif->kf_sa_peer, sa->sa_len);
2670 free(sa, M_SONAME);
2671 }
2672 kif->kf_sock_domain =
2673 so->so_proto->pr_domain->dom_family;
2674 kif->kf_sock_type = so->so_type;
2675 kif->kf_sock_protocol = so->so_proto->pr_protocol;
2676 }
2677 error = SYSCTL_OUT(req, kif, sizeof(*kif));
2678 if (error)
2679 break;
2680 }
2681 FILEDESC_SUNLOCK(fdp);
2682 fddrop(fdp);
2683 free(kif, M_TEMP);
2684 return (0);
2685}
2686
2687static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc, CTLFLAG_RD,
2688 sysctl_kern_proc_filedesc, "Process filedesc entries");
2689
2690#ifdef DDB
2691/*
2692 * For the purposes of debugging, generate a human-readable string for the
2693 * file type.
2694 */
2695static const char *
2696file_type_to_name(short type)
2697{
2698
2699 switch (type) {
2700 case 0:
2701 return ("zero");
2702 case DTYPE_VNODE:
2703 return ("vnod");
2704 case DTYPE_SOCKET:
2705 return ("sock");
2706 case DTYPE_PIPE:
2707 return ("pipe");
2708 case DTYPE_FIFO:
2709 return ("fifo");
2710 case DTYPE_KQUEUE:
2711 return ("kque");
2712 case DTYPE_CRYPTO:
2713 return ("crpt");
2714 case DTYPE_MQUEUE:
2715 return ("mque");
2716 case DTYPE_SHM:
2717 return ("shm");
2718 default:
2719 return ("unkn");
2720 }
2721}
2722
2723/*
2724 * For the purposes of debugging, identify a process (if any, perhaps one of
2725 * many) that references the passed file in its file descriptor array. Return
2726 * NULL if none.
2727 */
2728static struct proc *
2729file_to_first_proc(struct file *fp)
2730{
2731 struct filedesc *fdp;
2732 struct proc *p;
2733 int n;
2734
2735 FOREACH_PROC_IN_SYSTEM(p) {
2736 if (p->p_state == PRS_NEW)
2737 continue;
2738 fdp = p->p_fd;
2739 if (fdp == NULL)
2740 continue;
2741 for (n = 0; n < fdp->fd_nfiles; n++) {
2742 if (fp == fdp->fd_ofiles[n])
2743 return (p);
2744 }
2745 }
2746 return (NULL);
2747}
2748
2749static void
2750db_print_file(struct file *fp, int header)
2751{
2752 struct proc *p;
2753
2754 if (header)
2755 db_printf("%8s %4s %8s %8s %4s %5s %6s %8s %5s %12s\n",
2756 "File", "Type", "Data", "Flag", "GCFl", "Count",
2757 "MCount", "Vnode", "FPID", "FCmd");
2758 p = file_to_first_proc(fp);
2759 db_printf("%8p %4s %8p %08x %04x %5d %6d %8p %5d %12s\n", fp,
2760 file_type_to_name(fp->f_type), fp->f_data, fp->f_flag,
2761 0, fp->f_count, 0, fp->f_vnode,
2762 p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-");
2763}
2764
2765DB_SHOW_COMMAND(file, db_show_file)
2766{
2767 struct file *fp;
2768
2769 if (!have_addr) {
2770 db_printf("usage: show file <addr>\n");
2771 return;
2772 }
2773 fp = (struct file *)addr;
2774 db_print_file(fp, 1);
2775}
2776
2777DB_SHOW_COMMAND(files, db_show_files)
2778{
2779 struct filedesc *fdp;
2780 struct file *fp;
2781 struct proc *p;
2782 int header;
2783 int n;
2784
2785 header = 1;
2786 FOREACH_PROC_IN_SYSTEM(p) {
2787 if (p->p_state == PRS_NEW)
2788 continue;
2789 if ((fdp = p->p_fd) == NULL)
2790 continue;
2791 for (n = 0; n < fdp->fd_nfiles; ++n) {
2792 if ((fp = fdp->fd_ofiles[n]) == NULL)
2793 continue;
2794 db_print_file(fp, header);
2795 header = 0;
2796 }
2797 }
2798}
2799#endif
2800
2801SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW,
2802 &maxfilesperproc, 0, "Maximum files allowed open per process");
2803
2804SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW,
2805 &maxfiles, 0, "Maximum number of files");
2806
2807SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD,
2808 __DEVOLATILE(int *, &openfiles), 0, "System-wide number of open files");
2809
2810/* ARGSUSED*/
2811static void
2812filelistinit(void *dummy)
2813{
2814
2815 file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL,
2816 NULL, NULL, UMA_ALIGN_PTR, 0);
2817 mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF);
2818 mtx_init(&fdesc_mtx, "fdesc", NULL, MTX_DEF);
2819}
2820SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL)
2821
2822/*-------------------------------------------------------------------*/
2823
2824static int
2825badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred, int flags, struct thread *td)
2826{
2827
2828 return (EBADF);
2829}
2830
2831static int
2832badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred, struct thread *td)
2833{
2834
2835 return (EINVAL);
2836}
2837
2838static int
2839badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred, struct thread *td)
2840{
2841
2842 return (EBADF);
2843}
2844
2845static int
2846badfo_poll(struct file *fp, int events, struct ucred *active_cred, struct thread *td)
2847{
2848
2849 return (0);
2850}
2851
2852static int
2853badfo_kqfilter(struct file *fp, struct knote *kn)
2854{
2855
2856 return (EBADF);
2857}
2858
2859static int
2860badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred, struct thread *td)
2861{
2862
2863 return (EBADF);
2864}
2865
2866static int
2867badfo_close(struct file *fp, struct thread *td)
2868{
2869
2870 return (EBADF);
2871}
2872
2873struct fileops badfileops = {
2874 .fo_read = badfo_readwrite,
2875 .fo_write = badfo_readwrite,
2876 .fo_truncate = badfo_truncate,
2877 .fo_ioctl = badfo_ioctl,
2878 .fo_poll = badfo_poll,
2879 .fo_kqfilter = badfo_kqfilter,
2880 .fo_stat = badfo_stat,
2881 .fo_close = badfo_close,
2882};
2883
2884
2885/*-------------------------------------------------------------------*/
2886
2887/*
2888 * File Descriptor pseudo-device driver (/dev/fd/).
2889 *
2890 * Opening minor device N dup()s the file (if any) connected to file
2891 * descriptor N belonging to the calling process. Note that this driver
2892 * consists of only the ``open()'' routine, because all subsequent
2893 * references to this file will be direct to the other driver.
2894 *
2895 * XXX: we could give this one a cloning event handler if necessary.
2896 */
2897
2898/* ARGSUSED */
2899static int
2900fdopen(struct cdev *dev, int mode, int type, struct thread *td)
2901{
2902
2903 /*
2904 * XXX Kludge: set curthread->td_dupfd to contain the value of the
2905 * the file descriptor being sought for duplication. The error
2906 * return ensures that the vnode for this device will be released
2907 * by vn_open. Open will detect this special error and take the
2908 * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN
2909 * will simply report the error.
2910 */
2911 td->td_dupfd = dev2unit(dev);
2912 return (ENODEV);
2913}
2914
2915static struct cdevsw fildesc_cdevsw = {
2916 .d_version = D_VERSION,
2917 .d_flags = D_NEEDGIANT,
2918 .d_open = fdopen,
2919 .d_name = "FD",
2920};
2921
2922static void
2923fildesc_drvinit(void *unused)
2924{
2925 struct cdev *dev;
2926
2927 dev = make_dev(&fildesc_cdevsw, 0, UID_ROOT, GID_WHEEL, 0666, "fd/0");
2928 make_dev_alias(dev, "stdin");
2929 dev = make_dev(&fildesc_cdevsw, 1, UID_ROOT, GID_WHEEL, 0666, "fd/1");
2930 make_dev_alias(dev, "stdout");
2931 dev = make_dev(&fildesc_cdevsw, 2, UID_ROOT, GID_WHEEL, 0666, "fd/2");
2932 make_dev_alias(dev, "stderr");
2933}
2934
2935SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL)