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>
25 unsigned int sysctl_nr_open __read_mostly
= 1024*1024;
26 unsigned int sysctl_nr_open_min
= BITS_PER_LONG
;
27 /* our min() is unusable in constant expressions ;-/ */
28 #define __const_min(x, y) ((x) < (y) ? (x) : (y))
29 unsigned int sysctl_nr_open_max
=
30 __const_min(INT_MAX
, ~(size_t)0/sizeof(void *)) & -BITS_PER_LONG
;
32 static void __free_fdtable(struct fdtable
*fdt
)
35 kvfree(fdt
->open_fds
);
39 static void free_fdtable_rcu(struct rcu_head
*rcu
)
41 __free_fdtable(container_of(rcu
, struct fdtable
, rcu
));
44 #define BITBIT_NR(nr) BITS_TO_LONGS(BITS_TO_LONGS(nr))
45 #define BITBIT_SIZE(nr) (BITBIT_NR(nr) * sizeof(long))
48 * Copy 'count' fd bits from the old table to the new table and clear the extra
49 * space if any. This does not copy the file pointers. Called with the files
50 * spinlock held for write.
52 static void copy_fd_bitmaps(struct fdtable
*nfdt
, struct fdtable
*ofdt
,
55 unsigned int cpy
, set
;
57 cpy
= count
/ BITS_PER_BYTE
;
58 set
= (nfdt
->max_fds
- count
) / BITS_PER_BYTE
;
59 memcpy(nfdt
->open_fds
, ofdt
->open_fds
, cpy
);
60 memset((char *)nfdt
->open_fds
+ cpy
, 0, set
);
61 memcpy(nfdt
->close_on_exec
, ofdt
->close_on_exec
, cpy
);
62 memset((char *)nfdt
->close_on_exec
+ cpy
, 0, set
);
64 cpy
= BITBIT_SIZE(count
);
65 set
= BITBIT_SIZE(nfdt
->max_fds
) - cpy
;
66 memcpy(nfdt
->full_fds_bits
, ofdt
->full_fds_bits
, cpy
);
67 memset((char *)nfdt
->full_fds_bits
+ cpy
, 0, set
);
71 * Copy all file descriptors from the old table to the new, expanded table and
72 * clear the extra space. Called with the files spinlock held for write.
74 static void copy_fdtable(struct fdtable
*nfdt
, struct fdtable
*ofdt
)
78 BUG_ON(nfdt
->max_fds
< ofdt
->max_fds
);
80 cpy
= ofdt
->max_fds
* sizeof(struct file
*);
81 set
= (nfdt
->max_fds
- ofdt
->max_fds
) * sizeof(struct file
*);
82 memcpy(nfdt
->fd
, ofdt
->fd
, cpy
);
83 memset((char *)nfdt
->fd
+ cpy
, 0, set
);
85 copy_fd_bitmaps(nfdt
, ofdt
, ofdt
->max_fds
);
88 static struct fdtable
* alloc_fdtable(unsigned int nr
)
94 * Figure out how many fds we actually want to support in this fdtable.
95 * Allocation steps are keyed to the size of the fdarray, since it
96 * grows far faster than any of the other dynamic data. We try to fit
97 * the fdarray into comfortable page-tuned chunks: starting at 1024B
98 * and growing in powers of two from there on.
100 nr
/= (1024 / sizeof(struct file
*));
101 nr
= roundup_pow_of_two(nr
+ 1);
102 nr
*= (1024 / sizeof(struct file
*));
104 * Note that this can drive nr *below* what we had passed if sysctl_nr_open
105 * had been set lower between the check in expand_files() and here. Deal
106 * with that in caller, it's cheaper that way.
108 * We make sure that nr remains a multiple of BITS_PER_LONG - otherwise
109 * bitmaps handling below becomes unpleasant, to put it mildly...
111 if (unlikely(nr
> sysctl_nr_open
))
112 nr
= ((sysctl_nr_open
- 1) | (BITS_PER_LONG
- 1)) + 1;
114 fdt
= kmalloc(sizeof(struct fdtable
), GFP_KERNEL_ACCOUNT
);
118 data
= kvmalloc_array(nr
, sizeof(struct file
*), GFP_KERNEL_ACCOUNT
);
123 data
= kvmalloc(max_t(size_t,
124 2 * nr
/ BITS_PER_BYTE
+ BITBIT_SIZE(nr
), L1_CACHE_BYTES
),
128 fdt
->open_fds
= data
;
129 data
+= nr
/ BITS_PER_BYTE
;
130 fdt
->close_on_exec
= data
;
131 data
+= nr
/ BITS_PER_BYTE
;
132 fdt
->full_fds_bits
= data
;
145 * Expand the file descriptor table.
146 * This function will allocate a new fdtable and both fd array and fdset, of
148 * Return <0 error code on error; 1 on successful completion.
149 * The files->file_lock should be held on entry, and will be held on exit.
151 static int expand_fdtable(struct files_struct
*files
, unsigned int nr
)
152 __releases(files
->file_lock
)
153 __acquires(files
->file_lock
)
155 struct fdtable
*new_fdt
, *cur_fdt
;
157 spin_unlock(&files
->file_lock
);
158 new_fdt
= alloc_fdtable(nr
);
160 /* make sure all __fd_install() have seen resize_in_progress
161 * or have finished their rcu_read_lock_sched() section.
163 if (atomic_read(&files
->count
) > 1)
166 spin_lock(&files
->file_lock
);
170 * extremely unlikely race - sysctl_nr_open decreased between the check in
171 * caller and alloc_fdtable(). Cheaper to catch it here...
173 if (unlikely(new_fdt
->max_fds
<= nr
)) {
174 __free_fdtable(new_fdt
);
177 cur_fdt
= files_fdtable(files
);
178 BUG_ON(nr
< cur_fdt
->max_fds
);
179 copy_fdtable(new_fdt
, cur_fdt
);
180 rcu_assign_pointer(files
->fdt
, new_fdt
);
181 if (cur_fdt
!= &files
->fdtab
)
182 call_rcu(&cur_fdt
->rcu
, free_fdtable_rcu
);
183 /* coupled with smp_rmb() in __fd_install() */
190 * This function will expand the file structures, if the requested size exceeds
191 * the current capacity and there is room for expansion.
192 * Return <0 error code on error; 0 when nothing done; 1 when files were
193 * expanded and execution may have blocked.
194 * The files->file_lock should be held on entry, and will be held on exit.
196 static int expand_files(struct files_struct
*files
, unsigned int nr
)
197 __releases(files
->file_lock
)
198 __acquires(files
->file_lock
)
204 fdt
= files_fdtable(files
);
206 /* Do we need to expand? */
207 if (nr
< fdt
->max_fds
)
211 if (nr
>= sysctl_nr_open
)
214 if (unlikely(files
->resize_in_progress
)) {
215 spin_unlock(&files
->file_lock
);
217 wait_event(files
->resize_wait
, !files
->resize_in_progress
);
218 spin_lock(&files
->file_lock
);
222 /* All good, so we try */
223 files
->resize_in_progress
= true;
224 expanded
= expand_fdtable(files
, nr
);
225 files
->resize_in_progress
= false;
227 wake_up_all(&files
->resize_wait
);
231 static inline void __set_close_on_exec(unsigned int fd
, struct fdtable
*fdt
)
233 __set_bit(fd
, fdt
->close_on_exec
);
236 static inline void __clear_close_on_exec(unsigned int fd
, struct fdtable
*fdt
)
238 if (test_bit(fd
, fdt
->close_on_exec
))
239 __clear_bit(fd
, fdt
->close_on_exec
);
242 static inline void __set_open_fd(unsigned int fd
, struct fdtable
*fdt
)
244 __set_bit(fd
, fdt
->open_fds
);
246 if (!~fdt
->open_fds
[fd
])
247 __set_bit(fd
, fdt
->full_fds_bits
);
250 static inline void __clear_open_fd(unsigned int fd
, struct fdtable
*fdt
)
252 __clear_bit(fd
, fdt
->open_fds
);
253 __clear_bit(fd
/ BITS_PER_LONG
, fdt
->full_fds_bits
);
256 static unsigned int count_open_files(struct fdtable
*fdt
)
258 unsigned int size
= fdt
->max_fds
;
261 /* Find the last open fd */
262 for (i
= size
/ BITS_PER_LONG
; i
> 0; ) {
263 if (fdt
->open_fds
[--i
])
266 i
= (i
+ 1) * BITS_PER_LONG
;
270 static unsigned int sane_fdtable_size(struct fdtable
*fdt
, unsigned int max_fds
)
274 count
= count_open_files(fdt
);
275 if (max_fds
< NR_OPEN_DEFAULT
)
276 max_fds
= NR_OPEN_DEFAULT
;
277 return min(count
, max_fds
);
281 * Allocate a new files structure and copy contents from the
282 * passed in files structure.
283 * errorp will be valid only when the returned files_struct is NULL.
285 struct files_struct
*dup_fd(struct files_struct
*oldf
, unsigned int max_fds
, int *errorp
)
287 struct files_struct
*newf
;
288 struct file
**old_fds
, **new_fds
;
289 unsigned int open_files
, i
;
290 struct fdtable
*old_fdt
, *new_fdt
;
293 newf
= kmem_cache_alloc(files_cachep
, GFP_KERNEL
);
297 atomic_set(&newf
->count
, 1);
299 spin_lock_init(&newf
->file_lock
);
300 newf
->resize_in_progress
= false;
301 init_waitqueue_head(&newf
->resize_wait
);
303 new_fdt
= &newf
->fdtab
;
304 new_fdt
->max_fds
= NR_OPEN_DEFAULT
;
305 new_fdt
->close_on_exec
= newf
->close_on_exec_init
;
306 new_fdt
->open_fds
= newf
->open_fds_init
;
307 new_fdt
->full_fds_bits
= newf
->full_fds_bits_init
;
308 new_fdt
->fd
= &newf
->fd_array
[0];
310 spin_lock(&oldf
->file_lock
);
311 old_fdt
= files_fdtable(oldf
);
312 open_files
= sane_fdtable_size(old_fdt
, max_fds
);
315 * Check whether we need to allocate a larger fd array and fd set.
317 while (unlikely(open_files
> new_fdt
->max_fds
)) {
318 spin_unlock(&oldf
->file_lock
);
320 if (new_fdt
!= &newf
->fdtab
)
321 __free_fdtable(new_fdt
);
323 new_fdt
= alloc_fdtable(open_files
- 1);
329 /* beyond sysctl_nr_open; nothing to do */
330 if (unlikely(new_fdt
->max_fds
< open_files
)) {
331 __free_fdtable(new_fdt
);
337 * Reacquire the oldf lock and a pointer to its fd table
338 * who knows it may have a new bigger fd table. We need
339 * the latest pointer.
341 spin_lock(&oldf
->file_lock
);
342 old_fdt
= files_fdtable(oldf
);
343 open_files
= sane_fdtable_size(old_fdt
, max_fds
);
346 copy_fd_bitmaps(new_fdt
, old_fdt
, open_files
);
348 old_fds
= old_fdt
->fd
;
349 new_fds
= new_fdt
->fd
;
351 for (i
= open_files
; i
!= 0; i
--) {
352 struct file
*f
= *old_fds
++;
357 * The fd may be claimed in the fd bitmap but not yet
358 * instantiated in the files array if a sibling thread
359 * is partway through open(). So make sure that this
360 * fd is available to the new process.
362 __clear_open_fd(open_files
- i
, new_fdt
);
364 rcu_assign_pointer(*new_fds
++, f
);
366 spin_unlock(&oldf
->file_lock
);
368 /* clear the remainder */
369 memset(new_fds
, 0, (new_fdt
->max_fds
- open_files
) * sizeof(struct file
*));
371 rcu_assign_pointer(newf
->fdt
, new_fdt
);
376 kmem_cache_free(files_cachep
, newf
);
381 static struct fdtable
*close_files(struct files_struct
* files
)
384 * It is safe to dereference the fd table without RCU or
385 * ->file_lock because this is the last reference to the
388 struct fdtable
*fdt
= rcu_dereference_raw(files
->fdt
);
389 unsigned int i
, j
= 0;
393 i
= j
* BITS_PER_LONG
;
394 if (i
>= fdt
->max_fds
)
396 set
= fdt
->open_fds
[j
++];
399 struct file
* file
= xchg(&fdt
->fd
[i
], NULL
);
401 filp_close(file
, files
);
413 struct files_struct
*get_files_struct(struct task_struct
*task
)
415 struct files_struct
*files
;
420 atomic_inc(&files
->count
);
426 void put_files_struct(struct files_struct
*files
)
428 if (atomic_dec_and_test(&files
->count
)) {
429 struct fdtable
*fdt
= close_files(files
);
431 /* free the arrays if they are not embedded */
432 if (fdt
!= &files
->fdtab
)
434 kmem_cache_free(files_cachep
, files
);
438 void reset_files_struct(struct files_struct
*files
)
440 struct task_struct
*tsk
= current
;
441 struct files_struct
*old
;
447 put_files_struct(old
);
450 void exit_files(struct task_struct
*tsk
)
452 struct files_struct
* files
= tsk
->files
;
458 put_files_struct(files
);
462 struct files_struct init_files
= {
463 .count
= ATOMIC_INIT(1),
464 .fdt
= &init_files
.fdtab
,
466 .max_fds
= NR_OPEN_DEFAULT
,
467 .fd
= &init_files
.fd_array
[0],
468 .close_on_exec
= init_files
.close_on_exec_init
,
469 .open_fds
= init_files
.open_fds_init
,
470 .full_fds_bits
= init_files
.full_fds_bits_init
,
472 .file_lock
= __SPIN_LOCK_UNLOCKED(init_files
.file_lock
),
473 .resize_wait
= __WAIT_QUEUE_HEAD_INITIALIZER(init_files
.resize_wait
),
476 static unsigned int find_next_fd(struct fdtable
*fdt
, unsigned int start
)
478 unsigned int maxfd
= fdt
->max_fds
;
479 unsigned int maxbit
= maxfd
/ BITS_PER_LONG
;
480 unsigned int bitbit
= start
/ BITS_PER_LONG
;
482 bitbit
= find_next_zero_bit(fdt
->full_fds_bits
, maxbit
, bitbit
) * BITS_PER_LONG
;
487 return find_next_zero_bit(fdt
->open_fds
, maxfd
, start
);
491 * allocate a file descriptor, mark it busy.
493 int __alloc_fd(struct files_struct
*files
,
494 unsigned start
, unsigned end
, unsigned flags
)
500 spin_lock(&files
->file_lock
);
502 fdt
= files_fdtable(files
);
504 if (fd
< files
->next_fd
)
507 if (fd
< fdt
->max_fds
)
508 fd
= find_next_fd(fdt
, fd
);
511 * N.B. For clone tasks sharing a files structure, this test
512 * will limit the total number of files that can be opened.
518 error
= expand_files(files
, fd
);
523 * If we needed to expand the fs array we
524 * might have blocked - try again.
529 if (start
<= files
->next_fd
)
530 files
->next_fd
= fd
+ 1;
532 __set_open_fd(fd
, fdt
);
533 if (flags
& O_CLOEXEC
)
534 __set_close_on_exec(fd
, fdt
);
536 __clear_close_on_exec(fd
, fdt
);
540 if (rcu_access_pointer(fdt
->fd
[fd
]) != NULL
) {
541 printk(KERN_WARNING
"alloc_fd: slot %d not NULL!\n", fd
);
542 rcu_assign_pointer(fdt
->fd
[fd
], NULL
);
547 spin_unlock(&files
->file_lock
);
551 static int alloc_fd(unsigned start
, unsigned flags
)
553 return __alloc_fd(current
->files
, start
, rlimit(RLIMIT_NOFILE
), flags
);
556 int __get_unused_fd_flags(unsigned flags
, unsigned long nofile
)
558 return __alloc_fd(current
->files
, 0, nofile
, flags
);
561 int get_unused_fd_flags(unsigned flags
)
563 return __get_unused_fd_flags(flags
, rlimit(RLIMIT_NOFILE
));
565 EXPORT_SYMBOL(get_unused_fd_flags
);
567 static void __put_unused_fd(struct files_struct
*files
, unsigned int fd
)
569 struct fdtable
*fdt
= files_fdtable(files
);
570 __clear_open_fd(fd
, fdt
);
571 if (fd
< files
->next_fd
)
575 void put_unused_fd(unsigned int fd
)
577 struct files_struct
*files
= current
->files
;
578 spin_lock(&files
->file_lock
);
579 __put_unused_fd(files
, fd
);
580 spin_unlock(&files
->file_lock
);
583 EXPORT_SYMBOL(put_unused_fd
);
586 * Install a file pointer in the fd array.
588 * The VFS is full of places where we drop the files lock between
589 * setting the open_fds bitmap and installing the file in the file
590 * array. At any such point, we are vulnerable to a dup2() race
591 * installing a file in the array before us. We need to detect this and
592 * fput() the struct file we are about to overwrite in this case.
594 * It should never happen - if we allow dup2() do it, _really_ bad things
597 * NOTE: __fd_install() variant is really, really low-level; don't
598 * use it unless you are forced to by truly lousy API shoved down
599 * your throat. 'files' *MUST* be either current->files or obtained
600 * by get_files_struct(current) done by whoever had given it to you,
601 * or really bad things will happen. Normally you want to use
602 * fd_install() instead.
605 void __fd_install(struct files_struct
*files
, unsigned int fd
,
610 rcu_read_lock_sched();
612 if (unlikely(files
->resize_in_progress
)) {
613 rcu_read_unlock_sched();
614 spin_lock(&files
->file_lock
);
615 fdt
= files_fdtable(files
);
616 BUG_ON(fdt
->fd
[fd
] != NULL
);
617 rcu_assign_pointer(fdt
->fd
[fd
], file
);
618 spin_unlock(&files
->file_lock
);
621 /* coupled with smp_wmb() in expand_fdtable() */
623 fdt
= rcu_dereference_sched(files
->fdt
);
624 BUG_ON(fdt
->fd
[fd
] != NULL
);
625 rcu_assign_pointer(fdt
->fd
[fd
], file
);
626 rcu_read_unlock_sched();
630 * This consumes the "file" refcount, so callers should treat it
631 * as if they had called fput(file).
633 void fd_install(unsigned int fd
, struct file
*file
)
635 __fd_install(current
->files
, fd
, file
);
638 EXPORT_SYMBOL(fd_install
);
640 static struct file
*pick_file(struct files_struct
*files
, unsigned fd
)
642 struct file
*file
= NULL
;
645 spin_lock(&files
->file_lock
);
646 fdt
= files_fdtable(files
);
647 if (fd
>= fdt
->max_fds
)
652 rcu_assign_pointer(fdt
->fd
[fd
], NULL
);
653 __put_unused_fd(files
, fd
);
656 spin_unlock(&files
->file_lock
);
661 * The same warnings as for __alloc_fd()/__fd_install() apply here...
663 int __close_fd(struct files_struct
*files
, unsigned fd
)
667 file
= pick_file(files
, fd
);
671 return filp_close(file
, files
);
673 EXPORT_SYMBOL(__close_fd
); /* for ksys_close() */
676 * __close_range() - Close all file descriptors in a given range.
678 * @fd: starting file descriptor to close
679 * @max_fd: last file descriptor to close
681 * This closes a range of file descriptors. All file descriptors
682 * from @fd up to and including @max_fd are closed.
684 int __close_range(unsigned fd
, unsigned max_fd
, unsigned int flags
)
686 unsigned int cur_max
;
687 struct task_struct
*me
= current
;
688 struct files_struct
*cur_fds
= me
->files
, *fds
= NULL
;
690 if (flags
& ~CLOSE_RANGE_UNSHARE
)
697 cur_max
= files_fdtable(cur_fds
)->max_fds
;
700 /* cap to last valid index into fdtable */
703 if (flags
& CLOSE_RANGE_UNSHARE
) {
705 unsigned int max_unshare_fds
= NR_OPEN_MAX
;
708 * If the requested range is greater than the current maximum,
709 * we're closing everything so only copy all file descriptors
710 * beneath the lowest file descriptor.
712 if (max_fd
>= cur_max
)
713 max_unshare_fds
= fd
;
715 ret
= unshare_fd(CLONE_FILES
, max_unshare_fds
, &fds
);
720 * We used to share our file descriptor table, and have now
721 * created a private one, make sure we're using it below.
727 max_fd
= min(max_fd
, cur_max
);
728 while (fd
<= max_fd
) {
731 file
= pick_file(cur_fds
, fd
++);
735 filp_close(file
, cur_fds
);
741 * We're done closing the files we were supposed to. Time to install
742 * the new file descriptor table and drop the old one.
747 put_files_struct(fds
);
754 * variant of __close_fd that gets a ref on the file for later fput.
755 * The caller must ensure that filp_close() called on the file, and then
758 int __close_fd_get_file(unsigned int fd
, struct file
**res
)
760 struct files_struct
*files
= current
->files
;
764 spin_lock(&files
->file_lock
);
765 fdt
= files_fdtable(files
);
766 if (fd
>= fdt
->max_fds
)
771 rcu_assign_pointer(fdt
->fd
[fd
], NULL
);
772 __put_unused_fd(files
, fd
);
773 spin_unlock(&files
->file_lock
);
779 spin_unlock(&files
->file_lock
);
784 void do_close_on_exec(struct files_struct
*files
)
789 /* exec unshares first */
790 spin_lock(&files
->file_lock
);
793 unsigned fd
= i
* BITS_PER_LONG
;
794 fdt
= files_fdtable(files
);
795 if (fd
>= fdt
->max_fds
)
797 set
= fdt
->close_on_exec
[i
];
800 fdt
->close_on_exec
[i
] = 0;
801 for ( ; set
; fd
++, set
>>= 1) {
808 rcu_assign_pointer(fdt
->fd
[fd
], NULL
);
809 __put_unused_fd(files
, fd
);
810 spin_unlock(&files
->file_lock
);
811 filp_close(file
, files
);
813 spin_lock(&files
->file_lock
);
817 spin_unlock(&files
->file_lock
);
820 static struct file
*__fget_files(struct files_struct
*files
, unsigned int fd
,
821 fmode_t mask
, unsigned int refs
)
827 file
= fcheck_files(files
, fd
);
829 /* File object ref couldn't be taken.
830 * dup2() atomicity guarantee is the reason
831 * we loop to catch the new file (or NULL pointer)
833 if (file
->f_mode
& mask
)
835 else if (!get_file_rcu_many(file
, refs
))
843 static inline struct file
*__fget(unsigned int fd
, fmode_t mask
,
846 return __fget_files(current
->files
, fd
, mask
, refs
);
849 struct file
*fget_many(unsigned int fd
, unsigned int refs
)
851 return __fget(fd
, FMODE_PATH
, refs
);
854 struct file
*fget(unsigned int fd
)
856 return __fget(fd
, FMODE_PATH
, 1);
860 struct file
*fget_raw(unsigned int fd
)
862 return __fget(fd
, 0, 1);
864 EXPORT_SYMBOL(fget_raw
);
866 struct file
*fget_task(struct task_struct
*task
, unsigned int fd
)
868 struct file
*file
= NULL
;
872 file
= __fget_files(task
->files
, fd
, 0, 1);
879 * Lightweight file lookup - no refcnt increment if fd table isn't shared.
881 * You can use this instead of fget if you satisfy all of the following
883 * 1) You must call fput_light before exiting the syscall and returning control
884 * to userspace (i.e. you cannot remember the returned struct file * after
885 * returning to userspace).
886 * 2) You must not call filp_close on the returned struct file * in between
887 * calls to fget_light and fput_light.
888 * 3) You must not clone the current task in between the calls to fget_light
891 * The fput_needed flag returned by fget_light should be passed to the
892 * corresponding fput_light.
894 static unsigned long __fget_light(unsigned int fd
, fmode_t mask
)
896 struct files_struct
*files
= current
->files
;
899 if (atomic_read(&files
->count
) == 1) {
900 file
= __fcheck_files(files
, fd
);
901 if (!file
|| unlikely(file
->f_mode
& mask
))
903 return (unsigned long)file
;
905 file
= __fget(fd
, mask
, 1);
908 return FDPUT_FPUT
| (unsigned long)file
;
911 unsigned long __fdget(unsigned int fd
)
913 return __fget_light(fd
, FMODE_PATH
);
915 EXPORT_SYMBOL(__fdget
);
917 unsigned long __fdget_raw(unsigned int fd
)
919 return __fget_light(fd
, 0);
922 unsigned long __fdget_pos(unsigned int fd
)
924 unsigned long v
= __fdget(fd
);
925 struct file
*file
= (struct file
*)(v
& ~3);
927 if (file
&& (file
->f_mode
& FMODE_ATOMIC_POS
)) {
928 if (file_count(file
) > 1) {
929 v
|= FDPUT_POS_UNLOCK
;
930 mutex_lock(&file
->f_pos_lock
);
936 void __f_unlock_pos(struct file
*f
)
938 mutex_unlock(&f
->f_pos_lock
);
942 * We only lock f_pos if we have threads or if the file might be
943 * shared with another process. In both cases we'll have an elevated
944 * file count (done either by fdget() or by fork()).
947 void set_close_on_exec(unsigned int fd
, int flag
)
949 struct files_struct
*files
= current
->files
;
951 spin_lock(&files
->file_lock
);
952 fdt
= files_fdtable(files
);
954 __set_close_on_exec(fd
, fdt
);
956 __clear_close_on_exec(fd
, fdt
);
957 spin_unlock(&files
->file_lock
);
960 bool get_close_on_exec(unsigned int fd
)
962 struct files_struct
*files
= current
->files
;
966 fdt
= files_fdtable(files
);
967 res
= close_on_exec(fd
, fdt
);
972 static int do_dup2(struct files_struct
*files
,
973 struct file
*file
, unsigned fd
, unsigned flags
)
974 __releases(&files
->file_lock
)
980 * We need to detect attempts to do dup2() over allocated but still
981 * not finished descriptor. NB: OpenBSD avoids that at the price of
982 * extra work in their equivalent of fget() - they insert struct
983 * file immediately after grabbing descriptor, mark it larval if
984 * more work (e.g. actual opening) is needed and make sure that
985 * fget() treats larval files as absent. Potentially interesting,
986 * but while extra work in fget() is trivial, locking implications
987 * and amount of surgery on open()-related paths in VFS are not.
988 * FreeBSD fails with -EBADF in the same situation, NetBSD "solution"
989 * deadlocks in rather amusing ways, AFAICS. All of that is out of
990 * scope of POSIX or SUS, since neither considers shared descriptor
991 * tables and this condition does not arise without those.
993 fdt
= files_fdtable(files
);
994 tofree
= fdt
->fd
[fd
];
995 if (!tofree
&& fd_is_open(fd
, fdt
))
998 rcu_assign_pointer(fdt
->fd
[fd
], file
);
999 __set_open_fd(fd
, fdt
);
1000 if (flags
& O_CLOEXEC
)
1001 __set_close_on_exec(fd
, fdt
);
1003 __clear_close_on_exec(fd
, fdt
);
1004 spin_unlock(&files
->file_lock
);
1007 filp_close(tofree
, files
);
1012 spin_unlock(&files
->file_lock
);
1016 int replace_fd(unsigned fd
, struct file
*file
, unsigned flags
)
1019 struct files_struct
*files
= current
->files
;
1022 return __close_fd(files
, fd
);
1024 if (fd
>= rlimit(RLIMIT_NOFILE
))
1027 spin_lock(&files
->file_lock
);
1028 err
= expand_files(files
, fd
);
1029 if (unlikely(err
< 0))
1031 return do_dup2(files
, file
, fd
, flags
);
1034 spin_unlock(&files
->file_lock
);
1039 * __receive_fd() - Install received file into file descriptor table
1041 * @fd: fd to install into (if negative, a new fd will be allocated)
1042 * @file: struct file that was received from another process
1043 * @ufd: __user pointer to write new fd number to
1044 * @o_flags: the O_* flags to apply to the new fd entry
1046 * Installs a received file into the file descriptor table, with appropriate
1047 * checks and count updates. Optionally writes the fd number to userspace, if
1050 * This helper handles its own reference counting of the incoming
1053 * Returns newly install fd or -ve on error.
1055 int __receive_fd(int fd
, struct file
*file
, int __user
*ufd
, unsigned int o_flags
)
1060 error
= security_file_receive(file
);
1065 new_fd
= get_unused_fd_flags(o_flags
);
1073 error
= put_user(new_fd
, ufd
);
1076 put_unused_fd(new_fd
);
1082 fd_install(new_fd
, get_file(file
));
1084 error
= replace_fd(new_fd
, file
, o_flags
);
1089 /* Bump the sock usage counts, if any. */
1090 __receive_sock(file
);
1094 static int ksys_dup3(unsigned int oldfd
, unsigned int newfd
, int flags
)
1098 struct files_struct
*files
= current
->files
;
1100 if ((flags
& ~O_CLOEXEC
) != 0)
1103 if (unlikely(oldfd
== newfd
))
1106 if (newfd
>= rlimit(RLIMIT_NOFILE
))
1109 spin_lock(&files
->file_lock
);
1110 err
= expand_files(files
, newfd
);
1111 file
= fcheck(oldfd
);
1112 if (unlikely(!file
))
1114 if (unlikely(err
< 0)) {
1119 return do_dup2(files
, file
, newfd
, flags
);
1124 spin_unlock(&files
->file_lock
);
1128 SYSCALL_DEFINE3(dup3
, unsigned int, oldfd
, unsigned int, newfd
, int, flags
)
1130 return ksys_dup3(oldfd
, newfd
, flags
);
1133 SYSCALL_DEFINE2(dup2
, unsigned int, oldfd
, unsigned int, newfd
)
1135 if (unlikely(newfd
== oldfd
)) { /* corner case */
1136 struct files_struct
*files
= current
->files
;
1140 if (!fcheck_files(files
, oldfd
))
1145 return ksys_dup3(oldfd
, newfd
, 0);
1148 SYSCALL_DEFINE1(dup
, unsigned int, fildes
)
1151 struct file
*file
= fget_raw(fildes
);
1154 ret
= get_unused_fd_flags(0);
1156 fd_install(ret
, file
);
1163 int f_dupfd(unsigned int from
, struct file
*file
, unsigned flags
)
1166 if (from
>= rlimit(RLIMIT_NOFILE
))
1168 err
= alloc_fd(from
, flags
);
1171 fd_install(err
, file
);
1176 int iterate_fd(struct files_struct
*files
, unsigned n
,
1177 int (*f
)(const void *, struct file
*, unsigned),
1180 struct fdtable
*fdt
;
1184 spin_lock(&files
->file_lock
);
1185 for (fdt
= files_fdtable(files
); n
< fdt
->max_fds
; n
++) {
1187 file
= rcu_dereference_check_fdtable(files
, fdt
->fd
[n
]);
1190 res
= f(p
, file
, n
);
1194 spin_unlock(&files
->file_lock
);
1197 EXPORT_SYMBOL(iterate_fd
);