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>
24 #include <linux/io_uring.h>
26 unsigned int sysctl_nr_open __read_mostly
= 1024*1024;
27 unsigned int sysctl_nr_open_min
= BITS_PER_LONG
;
28 /* our min() is unusable in constant expressions ;-/ */
29 #define __const_min(x, y) ((x) < (y) ? (x) : (y))
30 unsigned int sysctl_nr_open_max
=
31 __const_min(INT_MAX
, ~(size_t)0/sizeof(void *)) & -BITS_PER_LONG
;
33 static void __free_fdtable(struct fdtable
*fdt
)
36 kvfree(fdt
->open_fds
);
40 static void free_fdtable_rcu(struct rcu_head
*rcu
)
42 __free_fdtable(container_of(rcu
, struct fdtable
, rcu
));
45 #define BITBIT_NR(nr) BITS_TO_LONGS(BITS_TO_LONGS(nr))
46 #define BITBIT_SIZE(nr) (BITBIT_NR(nr) * sizeof(long))
49 * Copy 'count' fd bits from the old table to the new table and clear the extra
50 * space if any. This does not copy the file pointers. Called with the files
51 * spinlock held for write.
53 static void copy_fd_bitmaps(struct fdtable
*nfdt
, struct fdtable
*ofdt
,
56 unsigned int cpy
, set
;
58 cpy
= count
/ BITS_PER_BYTE
;
59 set
= (nfdt
->max_fds
- count
) / BITS_PER_BYTE
;
60 memcpy(nfdt
->open_fds
, ofdt
->open_fds
, cpy
);
61 memset((char *)nfdt
->open_fds
+ cpy
, 0, set
);
62 memcpy(nfdt
->close_on_exec
, ofdt
->close_on_exec
, cpy
);
63 memset((char *)nfdt
->close_on_exec
+ cpy
, 0, set
);
65 cpy
= BITBIT_SIZE(count
);
66 set
= BITBIT_SIZE(nfdt
->max_fds
) - cpy
;
67 memcpy(nfdt
->full_fds_bits
, ofdt
->full_fds_bits
, cpy
);
68 memset((char *)nfdt
->full_fds_bits
+ cpy
, 0, set
);
72 * Copy all file descriptors from the old table to the new, expanded table and
73 * clear the extra space. Called with the files spinlock held for write.
75 static void copy_fdtable(struct fdtable
*nfdt
, struct fdtable
*ofdt
)
79 BUG_ON(nfdt
->max_fds
< ofdt
->max_fds
);
81 cpy
= ofdt
->max_fds
* sizeof(struct file
*);
82 set
= (nfdt
->max_fds
- ofdt
->max_fds
) * sizeof(struct file
*);
83 memcpy(nfdt
->fd
, ofdt
->fd
, cpy
);
84 memset((char *)nfdt
->fd
+ cpy
, 0, set
);
86 copy_fd_bitmaps(nfdt
, ofdt
, ofdt
->max_fds
);
89 static struct fdtable
* alloc_fdtable(unsigned int nr
)
95 * Figure out how many fds we actually want to support in this fdtable.
96 * Allocation steps are keyed to the size of the fdarray, since it
97 * grows far faster than any of the other dynamic data. We try to fit
98 * the fdarray into comfortable page-tuned chunks: starting at 1024B
99 * and growing in powers of two from there on.
101 nr
/= (1024 / sizeof(struct file
*));
102 nr
= roundup_pow_of_two(nr
+ 1);
103 nr
*= (1024 / sizeof(struct file
*));
105 * Note that this can drive nr *below* what we had passed if sysctl_nr_open
106 * had been set lower between the check in expand_files() and here. Deal
107 * with that in caller, it's cheaper that way.
109 * We make sure that nr remains a multiple of BITS_PER_LONG - otherwise
110 * bitmaps handling below becomes unpleasant, to put it mildly...
112 if (unlikely(nr
> sysctl_nr_open
))
113 nr
= ((sysctl_nr_open
- 1) | (BITS_PER_LONG
- 1)) + 1;
115 fdt
= kmalloc(sizeof(struct fdtable
), GFP_KERNEL_ACCOUNT
);
119 data
= kvmalloc_array(nr
, sizeof(struct file
*), GFP_KERNEL_ACCOUNT
);
124 data
= kvmalloc(max_t(size_t,
125 2 * nr
/ BITS_PER_BYTE
+ BITBIT_SIZE(nr
), L1_CACHE_BYTES
),
129 fdt
->open_fds
= data
;
130 data
+= nr
/ BITS_PER_BYTE
;
131 fdt
->close_on_exec
= data
;
132 data
+= nr
/ BITS_PER_BYTE
;
133 fdt
->full_fds_bits
= data
;
146 * Expand the file descriptor table.
147 * This function will allocate a new fdtable and both fd array and fdset, of
149 * Return <0 error code on error; 1 on successful completion.
150 * The files->file_lock should be held on entry, and will be held on exit.
152 static int expand_fdtable(struct files_struct
*files
, unsigned int nr
)
153 __releases(files
->file_lock
)
154 __acquires(files
->file_lock
)
156 struct fdtable
*new_fdt
, *cur_fdt
;
158 spin_unlock(&files
->file_lock
);
159 new_fdt
= alloc_fdtable(nr
);
161 /* make sure all __fd_install() have seen resize_in_progress
162 * or have finished their rcu_read_lock_sched() section.
164 if (atomic_read(&files
->count
) > 1)
167 spin_lock(&files
->file_lock
);
171 * extremely unlikely race - sysctl_nr_open decreased between the check in
172 * caller and alloc_fdtable(). Cheaper to catch it here...
174 if (unlikely(new_fdt
->max_fds
<= nr
)) {
175 __free_fdtable(new_fdt
);
178 cur_fdt
= files_fdtable(files
);
179 BUG_ON(nr
< cur_fdt
->max_fds
);
180 copy_fdtable(new_fdt
, cur_fdt
);
181 rcu_assign_pointer(files
->fdt
, new_fdt
);
182 if (cur_fdt
!= &files
->fdtab
)
183 call_rcu(&cur_fdt
->rcu
, free_fdtable_rcu
);
184 /* coupled with smp_rmb() in __fd_install() */
191 * This function will expand the file structures, if the requested size exceeds
192 * the current capacity and there is room for expansion.
193 * Return <0 error code on error; 0 when nothing done; 1 when files were
194 * expanded and execution may have blocked.
195 * The files->file_lock should be held on entry, and will be held on exit.
197 static int expand_files(struct files_struct
*files
, unsigned int nr
)
198 __releases(files
->file_lock
)
199 __acquires(files
->file_lock
)
205 fdt
= files_fdtable(files
);
207 /* Do we need to expand? */
208 if (nr
< fdt
->max_fds
)
212 if (nr
>= sysctl_nr_open
)
215 if (unlikely(files
->resize_in_progress
)) {
216 spin_unlock(&files
->file_lock
);
218 wait_event(files
->resize_wait
, !files
->resize_in_progress
);
219 spin_lock(&files
->file_lock
);
223 /* All good, so we try */
224 files
->resize_in_progress
= true;
225 expanded
= expand_fdtable(files
, nr
);
226 files
->resize_in_progress
= false;
228 wake_up_all(&files
->resize_wait
);
232 static inline void __set_close_on_exec(unsigned int fd
, struct fdtable
*fdt
)
234 __set_bit(fd
, fdt
->close_on_exec
);
237 static inline void __clear_close_on_exec(unsigned int fd
, struct fdtable
*fdt
)
239 if (test_bit(fd
, fdt
->close_on_exec
))
240 __clear_bit(fd
, fdt
->close_on_exec
);
243 static inline void __set_open_fd(unsigned int fd
, struct fdtable
*fdt
)
245 __set_bit(fd
, fdt
->open_fds
);
247 if (!~fdt
->open_fds
[fd
])
248 __set_bit(fd
, fdt
->full_fds_bits
);
251 static inline void __clear_open_fd(unsigned int fd
, struct fdtable
*fdt
)
253 __clear_bit(fd
, fdt
->open_fds
);
254 __clear_bit(fd
/ BITS_PER_LONG
, fdt
->full_fds_bits
);
257 static unsigned int count_open_files(struct fdtable
*fdt
)
259 unsigned int size
= fdt
->max_fds
;
262 /* Find the last open fd */
263 for (i
= size
/ BITS_PER_LONG
; i
> 0; ) {
264 if (fdt
->open_fds
[--i
])
267 i
= (i
+ 1) * BITS_PER_LONG
;
271 static unsigned int sane_fdtable_size(struct fdtable
*fdt
, unsigned int max_fds
)
275 count
= count_open_files(fdt
);
276 if (max_fds
< NR_OPEN_DEFAULT
)
277 max_fds
= NR_OPEN_DEFAULT
;
278 return min(count
, max_fds
);
282 * Allocate a new files structure and copy contents from the
283 * passed in files structure.
284 * errorp will be valid only when the returned files_struct is NULL.
286 struct files_struct
*dup_fd(struct files_struct
*oldf
, unsigned int max_fds
, int *errorp
)
288 struct files_struct
*newf
;
289 struct file
**old_fds
, **new_fds
;
290 unsigned int open_files
, i
;
291 struct fdtable
*old_fdt
, *new_fdt
;
294 newf
= kmem_cache_alloc(files_cachep
, GFP_KERNEL
);
298 atomic_set(&newf
->count
, 1);
300 spin_lock_init(&newf
->file_lock
);
301 newf
->resize_in_progress
= false;
302 init_waitqueue_head(&newf
->resize_wait
);
304 new_fdt
= &newf
->fdtab
;
305 new_fdt
->max_fds
= NR_OPEN_DEFAULT
;
306 new_fdt
->close_on_exec
= newf
->close_on_exec_init
;
307 new_fdt
->open_fds
= newf
->open_fds_init
;
308 new_fdt
->full_fds_bits
= newf
->full_fds_bits_init
;
309 new_fdt
->fd
= &newf
->fd_array
[0];
311 spin_lock(&oldf
->file_lock
);
312 old_fdt
= files_fdtable(oldf
);
313 open_files
= sane_fdtable_size(old_fdt
, max_fds
);
316 * Check whether we need to allocate a larger fd array and fd set.
318 while (unlikely(open_files
> new_fdt
->max_fds
)) {
319 spin_unlock(&oldf
->file_lock
);
321 if (new_fdt
!= &newf
->fdtab
)
322 __free_fdtable(new_fdt
);
324 new_fdt
= alloc_fdtable(open_files
- 1);
330 /* beyond sysctl_nr_open; nothing to do */
331 if (unlikely(new_fdt
->max_fds
< open_files
)) {
332 __free_fdtable(new_fdt
);
338 * Reacquire the oldf lock and a pointer to its fd table
339 * who knows it may have a new bigger fd table. We need
340 * the latest pointer.
342 spin_lock(&oldf
->file_lock
);
343 old_fdt
= files_fdtable(oldf
);
344 open_files
= sane_fdtable_size(old_fdt
, max_fds
);
347 copy_fd_bitmaps(new_fdt
, old_fdt
, open_files
);
349 old_fds
= old_fdt
->fd
;
350 new_fds
= new_fdt
->fd
;
352 for (i
= open_files
; i
!= 0; i
--) {
353 struct file
*f
= *old_fds
++;
358 * The fd may be claimed in the fd bitmap but not yet
359 * instantiated in the files array if a sibling thread
360 * is partway through open(). So make sure that this
361 * fd is available to the new process.
363 __clear_open_fd(open_files
- i
, new_fdt
);
365 rcu_assign_pointer(*new_fds
++, f
);
367 spin_unlock(&oldf
->file_lock
);
369 /* clear the remainder */
370 memset(new_fds
, 0, (new_fdt
->max_fds
- open_files
) * sizeof(struct file
*));
372 rcu_assign_pointer(newf
->fdt
, new_fdt
);
377 kmem_cache_free(files_cachep
, newf
);
382 static struct fdtable
*close_files(struct files_struct
* files
)
385 * It is safe to dereference the fd table without RCU or
386 * ->file_lock because this is the last reference to the
389 struct fdtable
*fdt
= rcu_dereference_raw(files
->fdt
);
390 unsigned int i
, j
= 0;
394 i
= j
* BITS_PER_LONG
;
395 if (i
>= fdt
->max_fds
)
397 set
= fdt
->open_fds
[j
++];
400 struct file
* file
= xchg(&fdt
->fd
[i
], NULL
);
402 filp_close(file
, files
);
414 struct files_struct
*get_files_struct(struct task_struct
*task
)
416 struct files_struct
*files
;
421 atomic_inc(&files
->count
);
427 void put_files_struct(struct files_struct
*files
)
429 if (atomic_dec_and_test(&files
->count
)) {
430 struct fdtable
*fdt
= close_files(files
);
432 /* free the arrays if they are not embedded */
433 if (fdt
!= &files
->fdtab
)
435 kmem_cache_free(files_cachep
, files
);
439 void reset_files_struct(struct files_struct
*files
)
441 struct task_struct
*tsk
= current
;
442 struct files_struct
*old
;
448 put_files_struct(old
);
451 void exit_files(struct task_struct
*tsk
)
453 struct files_struct
* files
= tsk
->files
;
456 io_uring_files_cancel(files
);
460 put_files_struct(files
);
464 struct files_struct init_files
= {
465 .count
= ATOMIC_INIT(1),
466 .fdt
= &init_files
.fdtab
,
468 .max_fds
= NR_OPEN_DEFAULT
,
469 .fd
= &init_files
.fd_array
[0],
470 .close_on_exec
= init_files
.close_on_exec_init
,
471 .open_fds
= init_files
.open_fds_init
,
472 .full_fds_bits
= init_files
.full_fds_bits_init
,
474 .file_lock
= __SPIN_LOCK_UNLOCKED(init_files
.file_lock
),
475 .resize_wait
= __WAIT_QUEUE_HEAD_INITIALIZER(init_files
.resize_wait
),
478 static unsigned int find_next_fd(struct fdtable
*fdt
, unsigned int start
)
480 unsigned int maxfd
= fdt
->max_fds
;
481 unsigned int maxbit
= maxfd
/ BITS_PER_LONG
;
482 unsigned int bitbit
= start
/ BITS_PER_LONG
;
484 bitbit
= find_next_zero_bit(fdt
->full_fds_bits
, maxbit
, bitbit
) * BITS_PER_LONG
;
489 return find_next_zero_bit(fdt
->open_fds
, maxfd
, start
);
493 * allocate a file descriptor, mark it busy.
495 int __alloc_fd(struct files_struct
*files
,
496 unsigned start
, unsigned end
, unsigned flags
)
502 spin_lock(&files
->file_lock
);
504 fdt
= files_fdtable(files
);
506 if (fd
< files
->next_fd
)
509 if (fd
< fdt
->max_fds
)
510 fd
= find_next_fd(fdt
, fd
);
513 * N.B. For clone tasks sharing a files structure, this test
514 * will limit the total number of files that can be opened.
520 error
= expand_files(files
, fd
);
525 * If we needed to expand the fs array we
526 * might have blocked - try again.
531 if (start
<= files
->next_fd
)
532 files
->next_fd
= fd
+ 1;
534 __set_open_fd(fd
, fdt
);
535 if (flags
& O_CLOEXEC
)
536 __set_close_on_exec(fd
, fdt
);
538 __clear_close_on_exec(fd
, fdt
);
542 if (rcu_access_pointer(fdt
->fd
[fd
]) != NULL
) {
543 printk(KERN_WARNING
"alloc_fd: slot %d not NULL!\n", fd
);
544 rcu_assign_pointer(fdt
->fd
[fd
], NULL
);
549 spin_unlock(&files
->file_lock
);
553 static int alloc_fd(unsigned start
, unsigned flags
)
555 return __alloc_fd(current
->files
, start
, rlimit(RLIMIT_NOFILE
), flags
);
558 int __get_unused_fd_flags(unsigned flags
, unsigned long nofile
)
560 return __alloc_fd(current
->files
, 0, nofile
, flags
);
563 int get_unused_fd_flags(unsigned flags
)
565 return __get_unused_fd_flags(flags
, rlimit(RLIMIT_NOFILE
));
567 EXPORT_SYMBOL(get_unused_fd_flags
);
569 static void __put_unused_fd(struct files_struct
*files
, unsigned int fd
)
571 struct fdtable
*fdt
= files_fdtable(files
);
572 __clear_open_fd(fd
, fdt
);
573 if (fd
< files
->next_fd
)
577 void put_unused_fd(unsigned int fd
)
579 struct files_struct
*files
= current
->files
;
580 spin_lock(&files
->file_lock
);
581 __put_unused_fd(files
, fd
);
582 spin_unlock(&files
->file_lock
);
585 EXPORT_SYMBOL(put_unused_fd
);
588 * Install a file pointer in the fd array.
590 * The VFS is full of places where we drop the files lock between
591 * setting the open_fds bitmap and installing the file in the file
592 * array. At any such point, we are vulnerable to a dup2() race
593 * installing a file in the array before us. We need to detect this and
594 * fput() the struct file we are about to overwrite in this case.
596 * It should never happen - if we allow dup2() do it, _really_ bad things
599 * NOTE: __fd_install() variant is really, really low-level; don't
600 * use it unless you are forced to by truly lousy API shoved down
601 * your throat. 'files' *MUST* be either current->files or obtained
602 * by get_files_struct(current) done by whoever had given it to you,
603 * or really bad things will happen. Normally you want to use
604 * fd_install() instead.
607 void __fd_install(struct files_struct
*files
, unsigned int fd
,
612 rcu_read_lock_sched();
614 if (unlikely(files
->resize_in_progress
)) {
615 rcu_read_unlock_sched();
616 spin_lock(&files
->file_lock
);
617 fdt
= files_fdtable(files
);
618 BUG_ON(fdt
->fd
[fd
] != NULL
);
619 rcu_assign_pointer(fdt
->fd
[fd
], file
);
620 spin_unlock(&files
->file_lock
);
623 /* coupled with smp_wmb() in expand_fdtable() */
625 fdt
= rcu_dereference_sched(files
->fdt
);
626 BUG_ON(fdt
->fd
[fd
] != NULL
);
627 rcu_assign_pointer(fdt
->fd
[fd
], file
);
628 rcu_read_unlock_sched();
632 * This consumes the "file" refcount, so callers should treat it
633 * as if they had called fput(file).
635 void fd_install(unsigned int fd
, struct file
*file
)
637 __fd_install(current
->files
, fd
, file
);
640 EXPORT_SYMBOL(fd_install
);
642 static struct file
*pick_file(struct files_struct
*files
, unsigned fd
)
644 struct file
*file
= NULL
;
647 spin_lock(&files
->file_lock
);
648 fdt
= files_fdtable(files
);
649 if (fd
>= fdt
->max_fds
)
654 rcu_assign_pointer(fdt
->fd
[fd
], NULL
);
655 __put_unused_fd(files
, fd
);
658 spin_unlock(&files
->file_lock
);
663 * The same warnings as for __alloc_fd()/__fd_install() apply here...
665 int __close_fd(struct files_struct
*files
, unsigned fd
)
669 file
= pick_file(files
, fd
);
673 return filp_close(file
, files
);
675 EXPORT_SYMBOL(__close_fd
); /* for ksys_close() */
678 * __close_range() - Close all file descriptors in a given range.
680 * @fd: starting file descriptor to close
681 * @max_fd: last file descriptor to close
683 * This closes a range of file descriptors. All file descriptors
684 * from @fd up to and including @max_fd are closed.
686 int __close_range(unsigned fd
, unsigned max_fd
, unsigned int flags
)
688 unsigned int cur_max
;
689 struct task_struct
*me
= current
;
690 struct files_struct
*cur_fds
= me
->files
, *fds
= NULL
;
692 if (flags
& ~CLOSE_RANGE_UNSHARE
)
699 cur_max
= files_fdtable(cur_fds
)->max_fds
;
702 /* cap to last valid index into fdtable */
705 if (flags
& CLOSE_RANGE_UNSHARE
) {
707 unsigned int max_unshare_fds
= NR_OPEN_MAX
;
710 * If the requested range is greater than the current maximum,
711 * we're closing everything so only copy all file descriptors
712 * beneath the lowest file descriptor.
714 if (max_fd
>= cur_max
)
715 max_unshare_fds
= fd
;
717 ret
= unshare_fd(CLONE_FILES
, max_unshare_fds
, &fds
);
722 * We used to share our file descriptor table, and have now
723 * created a private one, make sure we're using it below.
729 max_fd
= min(max_fd
, cur_max
);
730 while (fd
<= max_fd
) {
733 file
= pick_file(cur_fds
, fd
++);
737 filp_close(file
, cur_fds
);
743 * We're done closing the files we were supposed to. Time to install
744 * the new file descriptor table and drop the old one.
749 put_files_struct(fds
);
756 * variant of __close_fd that gets a ref on the file for later fput.
757 * The caller must ensure that filp_close() called on the file, and then
760 int __close_fd_get_file(unsigned int fd
, struct file
**res
)
762 struct files_struct
*files
= current
->files
;
766 spin_lock(&files
->file_lock
);
767 fdt
= files_fdtable(files
);
768 if (fd
>= fdt
->max_fds
)
773 rcu_assign_pointer(fdt
->fd
[fd
], NULL
);
774 __put_unused_fd(files
, fd
);
775 spin_unlock(&files
->file_lock
);
781 spin_unlock(&files
->file_lock
);
786 void do_close_on_exec(struct files_struct
*files
)
791 /* exec unshares first */
792 spin_lock(&files
->file_lock
);
795 unsigned fd
= i
* BITS_PER_LONG
;
796 fdt
= files_fdtable(files
);
797 if (fd
>= fdt
->max_fds
)
799 set
= fdt
->close_on_exec
[i
];
802 fdt
->close_on_exec
[i
] = 0;
803 for ( ; set
; fd
++, set
>>= 1) {
810 rcu_assign_pointer(fdt
->fd
[fd
], NULL
);
811 __put_unused_fd(files
, fd
);
812 spin_unlock(&files
->file_lock
);
813 filp_close(file
, files
);
815 spin_lock(&files
->file_lock
);
819 spin_unlock(&files
->file_lock
);
822 static struct file
*__fget_files(struct files_struct
*files
, unsigned int fd
,
823 fmode_t mask
, unsigned int refs
)
829 file
= fcheck_files(files
, fd
);
831 /* File object ref couldn't be taken.
832 * dup2() atomicity guarantee is the reason
833 * we loop to catch the new file (or NULL pointer)
835 if (file
->f_mode
& mask
)
837 else if (!get_file_rcu_many(file
, refs
))
845 static inline struct file
*__fget(unsigned int fd
, fmode_t mask
,
848 return __fget_files(current
->files
, fd
, mask
, refs
);
851 struct file
*fget_many(unsigned int fd
, unsigned int refs
)
853 return __fget(fd
, FMODE_PATH
, refs
);
856 struct file
*fget(unsigned int fd
)
858 return __fget(fd
, FMODE_PATH
, 1);
862 struct file
*fget_raw(unsigned int fd
)
864 return __fget(fd
, 0, 1);
866 EXPORT_SYMBOL(fget_raw
);
868 struct file
*fget_task(struct task_struct
*task
, unsigned int fd
)
870 struct file
*file
= NULL
;
874 file
= __fget_files(task
->files
, fd
, 0, 1);
881 * Lightweight file lookup - no refcnt increment if fd table isn't shared.
883 * You can use this instead of fget if you satisfy all of the following
885 * 1) You must call fput_light before exiting the syscall and returning control
886 * to userspace (i.e. you cannot remember the returned struct file * after
887 * returning to userspace).
888 * 2) You must not call filp_close on the returned struct file * in between
889 * calls to fget_light and fput_light.
890 * 3) You must not clone the current task in between the calls to fget_light
893 * The fput_needed flag returned by fget_light should be passed to the
894 * corresponding fput_light.
896 static unsigned long __fget_light(unsigned int fd
, fmode_t mask
)
898 struct files_struct
*files
= current
->files
;
901 if (atomic_read(&files
->count
) == 1) {
902 file
= __fcheck_files(files
, fd
);
903 if (!file
|| unlikely(file
->f_mode
& mask
))
905 return (unsigned long)file
;
907 file
= __fget(fd
, mask
, 1);
910 return FDPUT_FPUT
| (unsigned long)file
;
913 unsigned long __fdget(unsigned int fd
)
915 return __fget_light(fd
, FMODE_PATH
);
917 EXPORT_SYMBOL(__fdget
);
919 unsigned long __fdget_raw(unsigned int fd
)
921 return __fget_light(fd
, 0);
924 unsigned long __fdget_pos(unsigned int fd
)
926 unsigned long v
= __fdget(fd
);
927 struct file
*file
= (struct file
*)(v
& ~3);
929 if (file
&& (file
->f_mode
& FMODE_ATOMIC_POS
)) {
930 if (file_count(file
) > 1) {
931 v
|= FDPUT_POS_UNLOCK
;
932 mutex_lock(&file
->f_pos_lock
);
938 void __f_unlock_pos(struct file
*f
)
940 mutex_unlock(&f
->f_pos_lock
);
944 * We only lock f_pos if we have threads or if the file might be
945 * shared with another process. In both cases we'll have an elevated
946 * file count (done either by fdget() or by fork()).
949 void set_close_on_exec(unsigned int fd
, int flag
)
951 struct files_struct
*files
= current
->files
;
953 spin_lock(&files
->file_lock
);
954 fdt
= files_fdtable(files
);
956 __set_close_on_exec(fd
, fdt
);
958 __clear_close_on_exec(fd
, fdt
);
959 spin_unlock(&files
->file_lock
);
962 bool get_close_on_exec(unsigned int fd
)
964 struct files_struct
*files
= current
->files
;
968 fdt
= files_fdtable(files
);
969 res
= close_on_exec(fd
, fdt
);
974 static int do_dup2(struct files_struct
*files
,
975 struct file
*file
, unsigned fd
, unsigned flags
)
976 __releases(&files
->file_lock
)
982 * We need to detect attempts to do dup2() over allocated but still
983 * not finished descriptor. NB: OpenBSD avoids that at the price of
984 * extra work in their equivalent of fget() - they insert struct
985 * file immediately after grabbing descriptor, mark it larval if
986 * more work (e.g. actual opening) is needed and make sure that
987 * fget() treats larval files as absent. Potentially interesting,
988 * but while extra work in fget() is trivial, locking implications
989 * and amount of surgery on open()-related paths in VFS are not.
990 * FreeBSD fails with -EBADF in the same situation, NetBSD "solution"
991 * deadlocks in rather amusing ways, AFAICS. All of that is out of
992 * scope of POSIX or SUS, since neither considers shared descriptor
993 * tables and this condition does not arise without those.
995 fdt
= files_fdtable(files
);
996 tofree
= fdt
->fd
[fd
];
997 if (!tofree
&& fd_is_open(fd
, fdt
))
1000 rcu_assign_pointer(fdt
->fd
[fd
], file
);
1001 __set_open_fd(fd
, fdt
);
1002 if (flags
& O_CLOEXEC
)
1003 __set_close_on_exec(fd
, fdt
);
1005 __clear_close_on_exec(fd
, fdt
);
1006 spin_unlock(&files
->file_lock
);
1009 filp_close(tofree
, files
);
1014 spin_unlock(&files
->file_lock
);
1018 int replace_fd(unsigned fd
, struct file
*file
, unsigned flags
)
1021 struct files_struct
*files
= current
->files
;
1024 return __close_fd(files
, fd
);
1026 if (fd
>= rlimit(RLIMIT_NOFILE
))
1029 spin_lock(&files
->file_lock
);
1030 err
= expand_files(files
, fd
);
1031 if (unlikely(err
< 0))
1033 return do_dup2(files
, file
, fd
, flags
);
1036 spin_unlock(&files
->file_lock
);
1041 * __receive_fd() - Install received file into file descriptor table
1043 * @fd: fd to install into (if negative, a new fd will be allocated)
1044 * @file: struct file that was received from another process
1045 * @ufd: __user pointer to write new fd number to
1046 * @o_flags: the O_* flags to apply to the new fd entry
1048 * Installs a received file into the file descriptor table, with appropriate
1049 * checks and count updates. Optionally writes the fd number to userspace, if
1052 * This helper handles its own reference counting of the incoming
1055 * Returns newly install fd or -ve on error.
1057 int __receive_fd(int fd
, struct file
*file
, int __user
*ufd
, unsigned int o_flags
)
1062 error
= security_file_receive(file
);
1067 new_fd
= get_unused_fd_flags(o_flags
);
1075 error
= put_user(new_fd
, ufd
);
1078 put_unused_fd(new_fd
);
1084 fd_install(new_fd
, get_file(file
));
1086 error
= replace_fd(new_fd
, file
, o_flags
);
1091 /* Bump the sock usage counts, if any. */
1092 __receive_sock(file
);
1096 static int ksys_dup3(unsigned int oldfd
, unsigned int newfd
, int flags
)
1100 struct files_struct
*files
= current
->files
;
1102 if ((flags
& ~O_CLOEXEC
) != 0)
1105 if (unlikely(oldfd
== newfd
))
1108 if (newfd
>= rlimit(RLIMIT_NOFILE
))
1111 spin_lock(&files
->file_lock
);
1112 err
= expand_files(files
, newfd
);
1113 file
= fcheck(oldfd
);
1114 if (unlikely(!file
))
1116 if (unlikely(err
< 0)) {
1121 return do_dup2(files
, file
, newfd
, flags
);
1126 spin_unlock(&files
->file_lock
);
1130 SYSCALL_DEFINE3(dup3
, unsigned int, oldfd
, unsigned int, newfd
, int, flags
)
1132 return ksys_dup3(oldfd
, newfd
, flags
);
1135 SYSCALL_DEFINE2(dup2
, unsigned int, oldfd
, unsigned int, newfd
)
1137 if (unlikely(newfd
== oldfd
)) { /* corner case */
1138 struct files_struct
*files
= current
->files
;
1142 if (!fcheck_files(files
, oldfd
))
1147 return ksys_dup3(oldfd
, newfd
, 0);
1150 SYSCALL_DEFINE1(dup
, unsigned int, fildes
)
1153 struct file
*file
= fget_raw(fildes
);
1156 ret
= get_unused_fd_flags(0);
1158 fd_install(ret
, file
);
1165 int f_dupfd(unsigned int from
, struct file
*file
, unsigned flags
)
1168 if (from
>= rlimit(RLIMIT_NOFILE
))
1170 err
= alloc_fd(from
, flags
);
1173 fd_install(err
, file
);
1178 int iterate_fd(struct files_struct
*files
, unsigned n
,
1179 int (*f
)(const void *, struct file
*, unsigned),
1182 struct fdtable
*fdt
;
1186 spin_lock(&files
->file_lock
);
1187 for (fdt
= files_fdtable(files
); n
< fdt
->max_fds
; n
++) {
1189 file
= rcu_dereference_check_fdtable(files
, fdt
->fd
[n
]);
1192 res
= f(p
, file
, n
);
1196 spin_unlock(&files
->file_lock
);
1199 EXPORT_SYMBOL(iterate_fd
);