1 // SPDX-License-Identifier: GPL-2.0
5 * Copyright (C) 1998-1999, Stephen Tweedie and Bill Hawes
7 * Manage the dynamic fd arrays in the process files_struct.
10 #include <linux/syscalls.h>
11 #include <linux/export.h>
13 #include <linux/kernel.h>
15 #include <linux/sched/signal.h>
16 #include <linux/slab.h>
17 #include <linux/file.h>
18 #include <linux/fdtable.h>
19 #include <linux/bitops.h>
20 #include <linux/spinlock.h>
21 #include <linux/rcupdate.h>
22 #include <linux/close_range.h>
27 unsigned int sysctl_nr_open __read_mostly
= 1024*1024;
28 unsigned int sysctl_nr_open_min
= BITS_PER_LONG
;
29 /* our min() is unusable in constant expressions ;-/ */
30 #define __const_min(x, y) ((x) < (y) ? (x) : (y))
31 unsigned int sysctl_nr_open_max
=
32 __const_min(INT_MAX
, ~(size_t)0/sizeof(void *)) & -BITS_PER_LONG
;
34 static void __free_fdtable(struct fdtable
*fdt
)
37 kvfree(fdt
->open_fds
);
41 static void free_fdtable_rcu(struct rcu_head
*rcu
)
43 __free_fdtable(container_of(rcu
, struct fdtable
, rcu
));
46 #define BITBIT_NR(nr) BITS_TO_LONGS(BITS_TO_LONGS(nr))
47 #define BITBIT_SIZE(nr) (BITBIT_NR(nr) * sizeof(long))
50 * Copy 'count' fd bits from the old table to the new table and clear the extra
51 * space if any. This does not copy the file pointers. Called with the files
52 * spinlock held for write.
54 static void copy_fd_bitmaps(struct fdtable
*nfdt
, struct fdtable
*ofdt
,
57 unsigned int cpy
, set
;
59 cpy
= count
/ BITS_PER_BYTE
;
60 set
= (nfdt
->max_fds
- count
) / BITS_PER_BYTE
;
61 memcpy(nfdt
->open_fds
, ofdt
->open_fds
, cpy
);
62 memset((char *)nfdt
->open_fds
+ cpy
, 0, set
);
63 memcpy(nfdt
->close_on_exec
, ofdt
->close_on_exec
, cpy
);
64 memset((char *)nfdt
->close_on_exec
+ cpy
, 0, set
);
66 cpy
= BITBIT_SIZE(count
);
67 set
= BITBIT_SIZE(nfdt
->max_fds
) - cpy
;
68 memcpy(nfdt
->full_fds_bits
, ofdt
->full_fds_bits
, cpy
);
69 memset((char *)nfdt
->full_fds_bits
+ cpy
, 0, set
);
73 * Copy all file descriptors from the old table to the new, expanded table and
74 * clear the extra space. Called with the files spinlock held for write.
76 static void copy_fdtable(struct fdtable
*nfdt
, struct fdtable
*ofdt
)
80 BUG_ON(nfdt
->max_fds
< ofdt
->max_fds
);
82 cpy
= ofdt
->max_fds
* sizeof(struct file
*);
83 set
= (nfdt
->max_fds
- ofdt
->max_fds
) * sizeof(struct file
*);
84 memcpy(nfdt
->fd
, ofdt
->fd
, cpy
);
85 memset((char *)nfdt
->fd
+ cpy
, 0, set
);
87 copy_fd_bitmaps(nfdt
, ofdt
, ofdt
->max_fds
);
90 static struct fdtable
* alloc_fdtable(unsigned int nr
)
96 * Figure out how many fds we actually want to support in this fdtable.
97 * Allocation steps are keyed to the size of the fdarray, since it
98 * grows far faster than any of the other dynamic data. We try to fit
99 * the fdarray into comfortable page-tuned chunks: starting at 1024B
100 * and growing in powers of two from there on.
102 nr
/= (1024 / sizeof(struct file
*));
103 nr
= roundup_pow_of_two(nr
+ 1);
104 nr
*= (1024 / sizeof(struct file
*));
106 * Note that this can drive nr *below* what we had passed if sysctl_nr_open
107 * had been set lower between the check in expand_files() and here. Deal
108 * with that in caller, it's cheaper that way.
110 * We make sure that nr remains a multiple of BITS_PER_LONG - otherwise
111 * bitmaps handling below becomes unpleasant, to put it mildly...
113 if (unlikely(nr
> sysctl_nr_open
))
114 nr
= ((sysctl_nr_open
- 1) | (BITS_PER_LONG
- 1)) + 1;
116 fdt
= kmalloc(sizeof(struct fdtable
), GFP_KERNEL_ACCOUNT
);
120 data
= kvmalloc_array(nr
, sizeof(struct file
*), GFP_KERNEL_ACCOUNT
);
125 data
= kvmalloc(max_t(size_t,
126 2 * nr
/ BITS_PER_BYTE
+ BITBIT_SIZE(nr
), L1_CACHE_BYTES
),
130 fdt
->open_fds
= data
;
131 data
+= nr
/ BITS_PER_BYTE
;
132 fdt
->close_on_exec
= data
;
133 data
+= nr
/ BITS_PER_BYTE
;
134 fdt
->full_fds_bits
= data
;
147 * Expand the file descriptor table.
148 * This function will allocate a new fdtable and both fd array and fdset, of
150 * Return <0 error code on error; 1 on successful completion.
151 * The files->file_lock should be held on entry, and will be held on exit.
153 static int expand_fdtable(struct files_struct
*files
, unsigned int nr
)
154 __releases(files
->file_lock
)
155 __acquires(files
->file_lock
)
157 struct fdtable
*new_fdt
, *cur_fdt
;
159 spin_unlock(&files
->file_lock
);
160 new_fdt
= alloc_fdtable(nr
);
162 /* make sure all fd_install() have seen resize_in_progress
163 * or have finished their rcu_read_lock_sched() section.
165 if (atomic_read(&files
->count
) > 1)
168 spin_lock(&files
->file_lock
);
172 * extremely unlikely race - sysctl_nr_open decreased between the check in
173 * caller and alloc_fdtable(). Cheaper to catch it here...
175 if (unlikely(new_fdt
->max_fds
<= nr
)) {
176 __free_fdtable(new_fdt
);
179 cur_fdt
= files_fdtable(files
);
180 BUG_ON(nr
< cur_fdt
->max_fds
);
181 copy_fdtable(new_fdt
, cur_fdt
);
182 rcu_assign_pointer(files
->fdt
, new_fdt
);
183 if (cur_fdt
!= &files
->fdtab
)
184 call_rcu(&cur_fdt
->rcu
, free_fdtable_rcu
);
185 /* coupled with smp_rmb() in fd_install() */
192 * This function will expand the file structures, if the requested size exceeds
193 * the current capacity and there is room for expansion.
194 * Return <0 error code on error; 0 when nothing done; 1 when files were
195 * expanded and execution may have blocked.
196 * The files->file_lock should be held on entry, and will be held on exit.
198 static int expand_files(struct files_struct
*files
, unsigned int nr
)
199 __releases(files
->file_lock
)
200 __acquires(files
->file_lock
)
206 fdt
= files_fdtable(files
);
208 /* Do we need to expand? */
209 if (nr
< fdt
->max_fds
)
213 if (nr
>= sysctl_nr_open
)
216 if (unlikely(files
->resize_in_progress
)) {
217 spin_unlock(&files
->file_lock
);
219 wait_event(files
->resize_wait
, !files
->resize_in_progress
);
220 spin_lock(&files
->file_lock
);
224 /* All good, so we try */
225 files
->resize_in_progress
= true;
226 expanded
= expand_fdtable(files
, nr
);
227 files
->resize_in_progress
= false;
229 wake_up_all(&files
->resize_wait
);
233 static inline void __set_close_on_exec(unsigned int fd
, struct fdtable
*fdt
)
235 __set_bit(fd
, fdt
->close_on_exec
);
238 static inline void __clear_close_on_exec(unsigned int fd
, struct fdtable
*fdt
)
240 if (test_bit(fd
, fdt
->close_on_exec
))
241 __clear_bit(fd
, fdt
->close_on_exec
);
244 static inline void __set_open_fd(unsigned int fd
, struct fdtable
*fdt
)
246 __set_bit(fd
, fdt
->open_fds
);
248 if (!~fdt
->open_fds
[fd
])
249 __set_bit(fd
, fdt
->full_fds_bits
);
252 static inline void __clear_open_fd(unsigned int fd
, struct fdtable
*fdt
)
254 __clear_bit(fd
, fdt
->open_fds
);
255 __clear_bit(fd
/ BITS_PER_LONG
, fdt
->full_fds_bits
);
258 static unsigned int count_open_files(struct fdtable
*fdt
)
260 unsigned int size
= fdt
->max_fds
;
263 /* Find the last open fd */
264 for (i
= size
/ BITS_PER_LONG
; i
> 0; ) {
265 if (fdt
->open_fds
[--i
])
268 i
= (i
+ 1) * BITS_PER_LONG
;
272 static unsigned int sane_fdtable_size(struct fdtable
*fdt
, unsigned int max_fds
)
276 count
= count_open_files(fdt
);
277 if (max_fds
< NR_OPEN_DEFAULT
)
278 max_fds
= NR_OPEN_DEFAULT
;
279 return min(count
, max_fds
);
283 * Allocate a new files structure and copy contents from the
284 * passed in files structure.
285 * errorp will be valid only when the returned files_struct is NULL.
287 struct files_struct
*dup_fd(struct files_struct
*oldf
, unsigned int max_fds
, int *errorp
)
289 struct files_struct
*newf
;
290 struct file
**old_fds
, **new_fds
;
291 unsigned int open_files
, i
;
292 struct fdtable
*old_fdt
, *new_fdt
;
295 newf
= kmem_cache_alloc(files_cachep
, GFP_KERNEL
);
299 atomic_set(&newf
->count
, 1);
301 spin_lock_init(&newf
->file_lock
);
302 newf
->resize_in_progress
= false;
303 init_waitqueue_head(&newf
->resize_wait
);
305 new_fdt
= &newf
->fdtab
;
306 new_fdt
->max_fds
= NR_OPEN_DEFAULT
;
307 new_fdt
->close_on_exec
= newf
->close_on_exec_init
;
308 new_fdt
->open_fds
= newf
->open_fds_init
;
309 new_fdt
->full_fds_bits
= newf
->full_fds_bits_init
;
310 new_fdt
->fd
= &newf
->fd_array
[0];
312 spin_lock(&oldf
->file_lock
);
313 old_fdt
= files_fdtable(oldf
);
314 open_files
= sane_fdtable_size(old_fdt
, max_fds
);
317 * Check whether we need to allocate a larger fd array and fd set.
319 while (unlikely(open_files
> new_fdt
->max_fds
)) {
320 spin_unlock(&oldf
->file_lock
);
322 if (new_fdt
!= &newf
->fdtab
)
323 __free_fdtable(new_fdt
);
325 new_fdt
= alloc_fdtable(open_files
- 1);
331 /* beyond sysctl_nr_open; nothing to do */
332 if (unlikely(new_fdt
->max_fds
< open_files
)) {
333 __free_fdtable(new_fdt
);
339 * Reacquire the oldf lock and a pointer to its fd table
340 * who knows it may have a new bigger fd table. We need
341 * the latest pointer.
343 spin_lock(&oldf
->file_lock
);
344 old_fdt
= files_fdtable(oldf
);
345 open_files
= sane_fdtable_size(old_fdt
, max_fds
);
348 copy_fd_bitmaps(new_fdt
, old_fdt
, open_files
);
350 old_fds
= old_fdt
->fd
;
351 new_fds
= new_fdt
->fd
;
353 for (i
= open_files
; i
!= 0; i
--) {
354 struct file
*f
= *old_fds
++;
359 * The fd may be claimed in the fd bitmap but not yet
360 * instantiated in the files array if a sibling thread
361 * is partway through open(). So make sure that this
362 * fd is available to the new process.
364 __clear_open_fd(open_files
- i
, new_fdt
);
366 rcu_assign_pointer(*new_fds
++, f
);
368 spin_unlock(&oldf
->file_lock
);
370 /* clear the remainder */
371 memset(new_fds
, 0, (new_fdt
->max_fds
- open_files
) * sizeof(struct file
*));
373 rcu_assign_pointer(newf
->fdt
, new_fdt
);
378 kmem_cache_free(files_cachep
, newf
);
383 static struct fdtable
*close_files(struct files_struct
* files
)
386 * It is safe to dereference the fd table without RCU or
387 * ->file_lock because this is the last reference to the
390 struct fdtable
*fdt
= rcu_dereference_raw(files
->fdt
);
391 unsigned int i
, j
= 0;
395 i
= j
* BITS_PER_LONG
;
396 if (i
>= fdt
->max_fds
)
398 set
= fdt
->open_fds
[j
++];
401 struct file
* file
= xchg(&fdt
->fd
[i
], NULL
);
403 filp_close(file
, files
);
415 void put_files_struct(struct files_struct
*files
)
417 if (atomic_dec_and_test(&files
->count
)) {
418 struct fdtable
*fdt
= close_files(files
);
420 /* free the arrays if they are not embedded */
421 if (fdt
!= &files
->fdtab
)
423 kmem_cache_free(files_cachep
, files
);
427 void exit_files(struct task_struct
*tsk
)
429 struct files_struct
* files
= tsk
->files
;
435 put_files_struct(files
);
439 struct files_struct init_files
= {
440 .count
= ATOMIC_INIT(1),
441 .fdt
= &init_files
.fdtab
,
443 .max_fds
= NR_OPEN_DEFAULT
,
444 .fd
= &init_files
.fd_array
[0],
445 .close_on_exec
= init_files
.close_on_exec_init
,
446 .open_fds
= init_files
.open_fds_init
,
447 .full_fds_bits
= init_files
.full_fds_bits_init
,
449 .file_lock
= __SPIN_LOCK_UNLOCKED(init_files
.file_lock
),
450 .resize_wait
= __WAIT_QUEUE_HEAD_INITIALIZER(init_files
.resize_wait
),
453 static unsigned int find_next_fd(struct fdtable
*fdt
, unsigned int start
)
455 unsigned int maxfd
= fdt
->max_fds
;
456 unsigned int maxbit
= maxfd
/ BITS_PER_LONG
;
457 unsigned int bitbit
= start
/ BITS_PER_LONG
;
459 bitbit
= find_next_zero_bit(fdt
->full_fds_bits
, maxbit
, bitbit
) * BITS_PER_LONG
;
464 return find_next_zero_bit(fdt
->open_fds
, maxfd
, start
);
468 * allocate a file descriptor, mark it busy.
470 static int alloc_fd(unsigned start
, unsigned end
, unsigned flags
)
472 struct files_struct
*files
= current
->files
;
477 spin_lock(&files
->file_lock
);
479 fdt
= files_fdtable(files
);
481 if (fd
< files
->next_fd
)
484 if (fd
< fdt
->max_fds
)
485 fd
= find_next_fd(fdt
, fd
);
488 * N.B. For clone tasks sharing a files structure, this test
489 * will limit the total number of files that can be opened.
495 error
= expand_files(files
, fd
);
500 * If we needed to expand the fs array we
501 * might have blocked - try again.
506 if (start
<= files
->next_fd
)
507 files
->next_fd
= fd
+ 1;
509 __set_open_fd(fd
, fdt
);
510 if (flags
& O_CLOEXEC
)
511 __set_close_on_exec(fd
, fdt
);
513 __clear_close_on_exec(fd
, fdt
);
517 if (rcu_access_pointer(fdt
->fd
[fd
]) != NULL
) {
518 printk(KERN_WARNING
"alloc_fd: slot %d not NULL!\n", fd
);
519 rcu_assign_pointer(fdt
->fd
[fd
], NULL
);
524 spin_unlock(&files
->file_lock
);
528 int __get_unused_fd_flags(unsigned flags
, unsigned long nofile
)
530 return alloc_fd(0, nofile
, flags
);
533 int get_unused_fd_flags(unsigned flags
)
535 return __get_unused_fd_flags(flags
, rlimit(RLIMIT_NOFILE
));
537 EXPORT_SYMBOL(get_unused_fd_flags
);
539 static void __put_unused_fd(struct files_struct
*files
, unsigned int fd
)
541 struct fdtable
*fdt
= files_fdtable(files
);
542 __clear_open_fd(fd
, fdt
);
543 if (fd
< files
->next_fd
)
547 void put_unused_fd(unsigned int fd
)
549 struct files_struct
*files
= current
->files
;
550 spin_lock(&files
->file_lock
);
551 __put_unused_fd(files
, fd
);
552 spin_unlock(&files
->file_lock
);
555 EXPORT_SYMBOL(put_unused_fd
);
558 * Install a file pointer in the fd array.
560 * The VFS is full of places where we drop the files lock between
561 * setting the open_fds bitmap and installing the file in the file
562 * array. At any such point, we are vulnerable to a dup2() race
563 * installing a file in the array before us. We need to detect this and
564 * fput() the struct file we are about to overwrite in this case.
566 * It should never happen - if we allow dup2() do it, _really_ bad things
569 * This consumes the "file" refcount, so callers should treat it
570 * as if they had called fput(file).
573 void fd_install(unsigned int fd
, struct file
*file
)
575 struct files_struct
*files
= current
->files
;
578 rcu_read_lock_sched();
580 if (unlikely(files
->resize_in_progress
)) {
581 rcu_read_unlock_sched();
582 spin_lock(&files
->file_lock
);
583 fdt
= files_fdtable(files
);
584 BUG_ON(fdt
->fd
[fd
] != NULL
);
585 rcu_assign_pointer(fdt
->fd
[fd
], file
);
586 spin_unlock(&files
->file_lock
);
589 /* coupled with smp_wmb() in expand_fdtable() */
591 fdt
= rcu_dereference_sched(files
->fdt
);
592 BUG_ON(fdt
->fd
[fd
] != NULL
);
593 rcu_assign_pointer(fdt
->fd
[fd
], file
);
594 rcu_read_unlock_sched();
597 EXPORT_SYMBOL(fd_install
);
599 static struct file
*pick_file(struct files_struct
*files
, unsigned fd
)
601 struct file
*file
= NULL
;
604 spin_lock(&files
->file_lock
);
605 fdt
= files_fdtable(files
);
606 if (fd
>= fdt
->max_fds
)
611 rcu_assign_pointer(fdt
->fd
[fd
], NULL
);
612 __put_unused_fd(files
, fd
);
615 spin_unlock(&files
->file_lock
);
619 int close_fd(unsigned fd
)
621 struct files_struct
*files
= current
->files
;
624 file
= pick_file(files
, fd
);
628 return filp_close(file
, files
);
630 EXPORT_SYMBOL(close_fd
); /* for ksys_close() */
633 * last_fd - return last valid index into fd table
634 * @cur_fds: files struct
636 * Context: Either rcu read lock or files_lock must be held.
638 * Returns: Last valid index into fdtable.
640 static inline unsigned last_fd(struct fdtable
*fdt
)
642 return fdt
->max_fds
- 1;
645 static inline void __range_cloexec(struct files_struct
*cur_fds
,
646 unsigned int fd
, unsigned int max_fd
)
650 /* make sure we're using the correct maximum value */
651 spin_lock(&cur_fds
->file_lock
);
652 fdt
= files_fdtable(cur_fds
);
653 max_fd
= min(last_fd(fdt
), max_fd
);
655 bitmap_set(fdt
->close_on_exec
, fd
, max_fd
- fd
+ 1);
656 spin_unlock(&cur_fds
->file_lock
);
659 static inline void __range_close(struct files_struct
*cur_fds
, unsigned int fd
,
662 while (fd
<= max_fd
) {
665 file
= pick_file(cur_fds
, fd
++);
669 filp_close(file
, cur_fds
);
675 * __close_range() - Close all file descriptors in a given range.
677 * @fd: starting file descriptor to close
678 * @max_fd: last file descriptor to close
680 * This closes a range of file descriptors. All file descriptors
681 * from @fd up to and including @max_fd are closed.
683 int __close_range(unsigned fd
, unsigned max_fd
, unsigned int flags
)
685 unsigned int cur_max
;
686 struct task_struct
*me
= current
;
687 struct files_struct
*cur_fds
= me
->files
, *fds
= NULL
;
689 if (flags
& ~(CLOSE_RANGE_UNSHARE
| CLOSE_RANGE_CLOEXEC
))
696 cur_max
= files_fdtable(cur_fds
)->max_fds
;
699 /* cap to last valid index into fdtable */
702 if (flags
& CLOSE_RANGE_UNSHARE
) {
704 unsigned int max_unshare_fds
= NR_OPEN_MAX
;
707 * If the requested range is greater than the current maximum,
708 * we're closing everything so only copy all file descriptors
709 * beneath the lowest file descriptor.
710 * If the caller requested all fds to be made cloexec copy all
711 * of the file descriptors since they still want to use them.
713 if (!(flags
& CLOSE_RANGE_CLOEXEC
) && (max_fd
>= cur_max
))
714 max_unshare_fds
= fd
;
716 ret
= unshare_fd(CLONE_FILES
, max_unshare_fds
, &fds
);
721 * We used to share our file descriptor table, and have now
722 * created a private one, make sure we're using it below.
728 max_fd
= min(max_fd
, cur_max
);
730 if (flags
& CLOSE_RANGE_CLOEXEC
)
731 __range_cloexec(cur_fds
, fd
, max_fd
);
733 __range_close(cur_fds
, fd
, max_fd
);
737 * We're done closing the files we were supposed to. Time to install
738 * the new file descriptor table and drop the old one.
743 put_files_struct(fds
);
750 * See close_fd_get_file() below, this variant assumes current->files->file_lock
753 int __close_fd_get_file(unsigned int fd
, struct file
**res
)
755 struct files_struct
*files
= current
->files
;
759 fdt
= files_fdtable(files
);
760 if (fd
>= fdt
->max_fds
)
765 rcu_assign_pointer(fdt
->fd
[fd
], NULL
);
766 __put_unused_fd(files
, fd
);
776 * variant of close_fd that gets a ref on the file for later fput.
777 * The caller must ensure that filp_close() called on the file, and then
780 int close_fd_get_file(unsigned int fd
, struct file
**res
)
782 struct files_struct
*files
= current
->files
;
785 spin_lock(&files
->file_lock
);
786 ret
= __close_fd_get_file(fd
, res
);
787 spin_unlock(&files
->file_lock
);
792 void do_close_on_exec(struct files_struct
*files
)
797 /* exec unshares first */
798 spin_lock(&files
->file_lock
);
801 unsigned fd
= i
* BITS_PER_LONG
;
802 fdt
= files_fdtable(files
);
803 if (fd
>= fdt
->max_fds
)
805 set
= fdt
->close_on_exec
[i
];
808 fdt
->close_on_exec
[i
] = 0;
809 for ( ; set
; fd
++, set
>>= 1) {
816 rcu_assign_pointer(fdt
->fd
[fd
], NULL
);
817 __put_unused_fd(files
, fd
);
818 spin_unlock(&files
->file_lock
);
819 filp_close(file
, files
);
821 spin_lock(&files
->file_lock
);
825 spin_unlock(&files
->file_lock
);
828 static struct file
*__fget_files(struct files_struct
*files
, unsigned int fd
,
829 fmode_t mask
, unsigned int refs
)
835 file
= files_lookup_fd_rcu(files
, fd
);
837 /* File object ref couldn't be taken.
838 * dup2() atomicity guarantee is the reason
839 * we loop to catch the new file (or NULL pointer)
841 if (file
->f_mode
& mask
)
843 else if (!get_file_rcu_many(file
, refs
))
851 static inline struct file
*__fget(unsigned int fd
, fmode_t mask
,
854 return __fget_files(current
->files
, fd
, mask
, refs
);
857 struct file
*fget_many(unsigned int fd
, unsigned int refs
)
859 return __fget(fd
, FMODE_PATH
, refs
);
862 struct file
*fget(unsigned int fd
)
864 return __fget(fd
, FMODE_PATH
, 1);
868 struct file
*fget_raw(unsigned int fd
)
870 return __fget(fd
, 0, 1);
872 EXPORT_SYMBOL(fget_raw
);
874 struct file
*fget_task(struct task_struct
*task
, unsigned int fd
)
876 struct file
*file
= NULL
;
880 file
= __fget_files(task
->files
, fd
, 0, 1);
886 struct file
*task_lookup_fd_rcu(struct task_struct
*task
, unsigned int fd
)
888 /* Must be called with rcu_read_lock held */
889 struct files_struct
*files
;
890 struct file
*file
= NULL
;
895 file
= files_lookup_fd_rcu(files
, fd
);
901 struct file
*task_lookup_next_fd_rcu(struct task_struct
*task
, unsigned int *ret_fd
)
903 /* Must be called with rcu_read_lock held */
904 struct files_struct
*files
;
905 unsigned int fd
= *ret_fd
;
906 struct file
*file
= NULL
;
911 for (; fd
< files_fdtable(files
)->max_fds
; fd
++) {
912 file
= files_lookup_fd_rcu(files
, fd
);
923 * Lightweight file lookup - no refcnt increment if fd table isn't shared.
925 * You can use this instead of fget if you satisfy all of the following
927 * 1) You must call fput_light before exiting the syscall and returning control
928 * to userspace (i.e. you cannot remember the returned struct file * after
929 * returning to userspace).
930 * 2) You must not call filp_close on the returned struct file * in between
931 * calls to fget_light and fput_light.
932 * 3) You must not clone the current task in between the calls to fget_light
935 * The fput_needed flag returned by fget_light should be passed to the
936 * corresponding fput_light.
938 static unsigned long __fget_light(unsigned int fd
, fmode_t mask
)
940 struct files_struct
*files
= current
->files
;
943 if (atomic_read(&files
->count
) == 1) {
944 file
= files_lookup_fd_raw(files
, fd
);
945 if (!file
|| unlikely(file
->f_mode
& mask
))
947 return (unsigned long)file
;
949 file
= __fget(fd
, mask
, 1);
952 return FDPUT_FPUT
| (unsigned long)file
;
955 unsigned long __fdget(unsigned int fd
)
957 return __fget_light(fd
, FMODE_PATH
);
959 EXPORT_SYMBOL(__fdget
);
961 unsigned long __fdget_raw(unsigned int fd
)
963 return __fget_light(fd
, 0);
966 unsigned long __fdget_pos(unsigned int fd
)
968 unsigned long v
= __fdget(fd
);
969 struct file
*file
= (struct file
*)(v
& ~3);
971 if (file
&& (file
->f_mode
& FMODE_ATOMIC_POS
)) {
972 if (file_count(file
) > 1) {
973 v
|= FDPUT_POS_UNLOCK
;
974 mutex_lock(&file
->f_pos_lock
);
980 void __f_unlock_pos(struct file
*f
)
982 mutex_unlock(&f
->f_pos_lock
);
986 * We only lock f_pos if we have threads or if the file might be
987 * shared with another process. In both cases we'll have an elevated
988 * file count (done either by fdget() or by fork()).
991 void set_close_on_exec(unsigned int fd
, int flag
)
993 struct files_struct
*files
= current
->files
;
995 spin_lock(&files
->file_lock
);
996 fdt
= files_fdtable(files
);
998 __set_close_on_exec(fd
, fdt
);
1000 __clear_close_on_exec(fd
, fdt
);
1001 spin_unlock(&files
->file_lock
);
1004 bool get_close_on_exec(unsigned int fd
)
1006 struct files_struct
*files
= current
->files
;
1007 struct fdtable
*fdt
;
1010 fdt
= files_fdtable(files
);
1011 res
= close_on_exec(fd
, fdt
);
1016 static int do_dup2(struct files_struct
*files
,
1017 struct file
*file
, unsigned fd
, unsigned flags
)
1018 __releases(&files
->file_lock
)
1020 struct file
*tofree
;
1021 struct fdtable
*fdt
;
1024 * We need to detect attempts to do dup2() over allocated but still
1025 * not finished descriptor. NB: OpenBSD avoids that at the price of
1026 * extra work in their equivalent of fget() - they insert struct
1027 * file immediately after grabbing descriptor, mark it larval if
1028 * more work (e.g. actual opening) is needed and make sure that
1029 * fget() treats larval files as absent. Potentially interesting,
1030 * but while extra work in fget() is trivial, locking implications
1031 * and amount of surgery on open()-related paths in VFS are not.
1032 * FreeBSD fails with -EBADF in the same situation, NetBSD "solution"
1033 * deadlocks in rather amusing ways, AFAICS. All of that is out of
1034 * scope of POSIX or SUS, since neither considers shared descriptor
1035 * tables and this condition does not arise without those.
1037 fdt
= files_fdtable(files
);
1038 tofree
= fdt
->fd
[fd
];
1039 if (!tofree
&& fd_is_open(fd
, fdt
))
1042 rcu_assign_pointer(fdt
->fd
[fd
], file
);
1043 __set_open_fd(fd
, fdt
);
1044 if (flags
& O_CLOEXEC
)
1045 __set_close_on_exec(fd
, fdt
);
1047 __clear_close_on_exec(fd
, fdt
);
1048 spin_unlock(&files
->file_lock
);
1051 filp_close(tofree
, files
);
1056 spin_unlock(&files
->file_lock
);
1060 int replace_fd(unsigned fd
, struct file
*file
, unsigned flags
)
1063 struct files_struct
*files
= current
->files
;
1066 return close_fd(fd
);
1068 if (fd
>= rlimit(RLIMIT_NOFILE
))
1071 spin_lock(&files
->file_lock
);
1072 err
= expand_files(files
, fd
);
1073 if (unlikely(err
< 0))
1075 return do_dup2(files
, file
, fd
, flags
);
1078 spin_unlock(&files
->file_lock
);
1083 * __receive_fd() - Install received file into file descriptor table
1084 * @file: struct file that was received from another process
1085 * @ufd: __user pointer to write new fd number to
1086 * @o_flags: the O_* flags to apply to the new fd entry
1088 * Installs a received file into the file descriptor table, with appropriate
1089 * checks and count updates. Optionally writes the fd number to userspace, if
1092 * This helper handles its own reference counting of the incoming
1095 * Returns newly install fd or -ve on error.
1097 int __receive_fd(struct file
*file
, int __user
*ufd
, unsigned int o_flags
)
1102 error
= security_file_receive(file
);
1106 new_fd
= get_unused_fd_flags(o_flags
);
1111 error
= put_user(new_fd
, ufd
);
1113 put_unused_fd(new_fd
);
1118 fd_install(new_fd
, get_file(file
));
1119 __receive_sock(file
);
1123 int receive_fd_replace(int new_fd
, struct file
*file
, unsigned int o_flags
)
1127 error
= security_file_receive(file
);
1130 error
= replace_fd(new_fd
, file
, o_flags
);
1133 __receive_sock(file
);
1137 static int ksys_dup3(unsigned int oldfd
, unsigned int newfd
, int flags
)
1141 struct files_struct
*files
= current
->files
;
1143 if ((flags
& ~O_CLOEXEC
) != 0)
1146 if (unlikely(oldfd
== newfd
))
1149 if (newfd
>= rlimit(RLIMIT_NOFILE
))
1152 spin_lock(&files
->file_lock
);
1153 err
= expand_files(files
, newfd
);
1154 file
= files_lookup_fd_locked(files
, oldfd
);
1155 if (unlikely(!file
))
1157 if (unlikely(err
< 0)) {
1162 return do_dup2(files
, file
, newfd
, flags
);
1167 spin_unlock(&files
->file_lock
);
1171 SYSCALL_DEFINE3(dup3
, unsigned int, oldfd
, unsigned int, newfd
, int, flags
)
1173 return ksys_dup3(oldfd
, newfd
, flags
);
1176 SYSCALL_DEFINE2(dup2
, unsigned int, oldfd
, unsigned int, newfd
)
1178 if (unlikely(newfd
== oldfd
)) { /* corner case */
1179 struct files_struct
*files
= current
->files
;
1183 if (!files_lookup_fd_rcu(files
, oldfd
))
1188 return ksys_dup3(oldfd
, newfd
, 0);
1191 SYSCALL_DEFINE1(dup
, unsigned int, fildes
)
1194 struct file
*file
= fget_raw(fildes
);
1197 ret
= get_unused_fd_flags(0);
1199 fd_install(ret
, file
);
1206 int f_dupfd(unsigned int from
, struct file
*file
, unsigned flags
)
1208 unsigned long nofile
= rlimit(RLIMIT_NOFILE
);
1212 err
= alloc_fd(from
, nofile
, flags
);
1215 fd_install(err
, file
);
1220 int iterate_fd(struct files_struct
*files
, unsigned n
,
1221 int (*f
)(const void *, struct file
*, unsigned),
1224 struct fdtable
*fdt
;
1228 spin_lock(&files
->file_lock
);
1229 for (fdt
= files_fdtable(files
); n
< fdt
->max_fds
; n
++) {
1231 file
= rcu_dereference_check_fdtable(files
, fdt
->fd
[n
]);
1234 res
= f(p
, file
, n
);
1238 spin_unlock(&files
->file_lock
);
1241 EXPORT_SYMBOL(iterate_fd
);