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);
600 * pick_file - return file associatd with fd
601 * @files: file struct to retrieve file from
602 * @fd: file descriptor to retrieve file for
604 * If this functions returns an EINVAL error pointer the fd was beyond the
605 * current maximum number of file descriptors for that fdtable.
607 * Returns: The file associated with @fd, on error returns an error pointer.
609 static struct file *pick_file(struct files_struct *files, unsigned fd)
614 spin_lock(&files->file_lock);
615 fdt = files_fdtable(files);
616 if (fd >= fdt->max_fds) {
617 file = ERR_PTR(-EINVAL);
622 file = ERR_PTR(-EBADF);
625 rcu_assign_pointer(fdt->fd[fd], NULL);
626 __put_unused_fd(files, fd);
629 spin_unlock(&files->file_lock);
633 int close_fd(unsigned fd)
635 struct files_struct *files = current->files;
638 file = pick_file(files, fd);
642 return filp_close(file, files);
644 EXPORT_SYMBOL(close_fd); /* for ksys_close() */
647 * last_fd - return last valid index into fd table
648 * @cur_fds: files struct
650 * Context: Either rcu read lock or files_lock must be held.
652 * Returns: Last valid index into fdtable.
654 static inline unsigned last_fd(struct fdtable *fdt)
656 return fdt->max_fds - 1;
659 static inline void __range_cloexec(struct files_struct *cur_fds,
660 unsigned int fd, unsigned int max_fd)
664 /* make sure we're using the correct maximum value */
665 spin_lock(&cur_fds->file_lock);
666 fdt = files_fdtable(cur_fds);
667 max_fd = min(last_fd(fdt), max_fd);
669 bitmap_set(fdt->close_on_exec, fd, max_fd - fd + 1);
670 spin_unlock(&cur_fds->file_lock);
673 static inline void __range_close(struct files_struct *cur_fds, unsigned int fd,
676 while (fd <= max_fd) {
679 file = pick_file(cur_fds, fd++);
681 /* found a valid file to close */
682 filp_close(file, cur_fds);
687 /* beyond the last fd in that table */
688 if (PTR_ERR(file) == -EINVAL)
694 * __close_range() - Close all file descriptors in a given range.
696 * @fd: starting file descriptor to close
697 * @max_fd: last file descriptor to close
699 * This closes a range of file descriptors. All file descriptors
700 * from @fd up to and including @max_fd are closed.
702 int __close_range(unsigned fd, unsigned max_fd, unsigned int flags)
704 struct task_struct *me = current;
705 struct files_struct *cur_fds = me->files, *fds = NULL;
707 if (flags & ~(CLOSE_RANGE_UNSHARE | CLOSE_RANGE_CLOEXEC))
713 if (flags & CLOSE_RANGE_UNSHARE) {
715 unsigned int max_unshare_fds = NR_OPEN_MAX;
718 * If the caller requested all fds to be made cloexec we always
719 * copy all of the file descriptors since they still want to
722 if (!(flags & CLOSE_RANGE_CLOEXEC)) {
724 * If the requested range is greater than the current
725 * maximum, we're closing everything so only copy all
726 * file descriptors beneath the lowest file descriptor.
729 if (max_fd >= last_fd(files_fdtable(cur_fds)))
730 max_unshare_fds = fd;
734 ret = unshare_fd(CLONE_FILES, max_unshare_fds, &fds);
739 * We used to share our file descriptor table, and have now
740 * created a private one, make sure we're using it below.
746 if (flags & CLOSE_RANGE_CLOEXEC)
747 __range_cloexec(cur_fds, fd, max_fd);
749 __range_close(cur_fds, fd, max_fd);
753 * We're done closing the files we were supposed to. Time to install
754 * the new file descriptor table and drop the old one.
759 put_files_struct(fds);
766 * See close_fd_get_file() below, this variant assumes current->files->file_lock
769 int __close_fd_get_file(unsigned int fd, struct file **res)
771 struct files_struct *files = current->files;
775 fdt = files_fdtable(files);
776 if (fd >= fdt->max_fds)
781 rcu_assign_pointer(fdt->fd[fd], NULL);
782 __put_unused_fd(files, fd);
792 * variant of close_fd that gets a ref on the file for later fput.
793 * The caller must ensure that filp_close() called on the file, and then
796 int close_fd_get_file(unsigned int fd, struct file **res)
798 struct files_struct *files = current->files;
801 spin_lock(&files->file_lock);
802 ret = __close_fd_get_file(fd, res);
803 spin_unlock(&files->file_lock);
808 void do_close_on_exec(struct files_struct *files)
813 /* exec unshares first */
814 spin_lock(&files->file_lock);
817 unsigned fd = i * BITS_PER_LONG;
818 fdt = files_fdtable(files);
819 if (fd >= fdt->max_fds)
821 set = fdt->close_on_exec[i];
824 fdt->close_on_exec[i] = 0;
825 for ( ; set ; fd++, set >>= 1) {
832 rcu_assign_pointer(fdt->fd[fd], NULL);
833 __put_unused_fd(files, fd);
834 spin_unlock(&files->file_lock);
835 filp_close(file, files);
837 spin_lock(&files->file_lock);
841 spin_unlock(&files->file_lock);
844 static inline struct file *__fget_files_rcu(struct files_struct *files,
845 unsigned int fd, fmode_t mask, unsigned int refs)
849 struct fdtable *fdt = rcu_dereference_raw(files->fdt);
850 struct file __rcu **fdentry;
852 if (unlikely(fd >= fdt->max_fds))
855 fdentry = fdt->fd + array_index_nospec(fd, fdt->max_fds);
856 file = rcu_dereference_raw(*fdentry);
860 if (unlikely(file->f_mode & mask))
864 * Ok, we have a file pointer. However, because we do
865 * this all locklessly under RCU, we may be racing with
866 * that file being closed.
868 * Such a race can take two forms:
870 * (a) the file ref already went down to zero,
871 * and get_file_rcu_many() fails. Just try
874 if (unlikely(!get_file_rcu_many(file, refs)))
878 * (b) the file table entry has changed under us.
879 * Note that we don't need to re-check the 'fdt->fd'
880 * pointer having changed, because it always goes
881 * hand-in-hand with 'fdt'.
883 * If so, we need to put our refs and try again.
885 if (unlikely(rcu_dereference_raw(files->fdt) != fdt) ||
886 unlikely(rcu_dereference_raw(*fdentry) != file)) {
887 fput_many(file, refs);
892 * Ok, we have a ref to the file, and checked that it
899 static struct file *__fget_files(struct files_struct *files, unsigned int fd,
900 fmode_t mask, unsigned int refs)
905 file = __fget_files_rcu(files, fd, mask, refs);
911 static inline struct file *__fget(unsigned int fd, fmode_t mask,
914 return __fget_files(current->files, fd, mask, refs);
917 struct file *fget_many(unsigned int fd, unsigned int refs)
919 return __fget(fd, FMODE_PATH, refs);
922 struct file *fget(unsigned int fd)
924 return __fget(fd, FMODE_PATH, 1);
928 struct file *fget_raw(unsigned int fd)
930 return __fget(fd, 0, 1);
932 EXPORT_SYMBOL(fget_raw);
934 struct file *fget_task(struct task_struct *task, unsigned int fd)
936 struct file *file = NULL;
940 file = __fget_files(task->files, fd, 0, 1);
946 struct file *task_lookup_fd_rcu(struct task_struct *task, unsigned int fd)
948 /* Must be called with rcu_read_lock held */
949 struct files_struct *files;
950 struct file *file = NULL;
955 file = files_lookup_fd_rcu(files, fd);
961 struct file *task_lookup_next_fd_rcu(struct task_struct *task, unsigned int *ret_fd)
963 /* Must be called with rcu_read_lock held */
964 struct files_struct *files;
965 unsigned int fd = *ret_fd;
966 struct file *file = NULL;
971 for (; fd < files_fdtable(files)->max_fds; fd++) {
972 file = files_lookup_fd_rcu(files, fd);
983 * Lightweight file lookup - no refcnt increment if fd table isn't shared.
985 * You can use this instead of fget if you satisfy all of the following
987 * 1) You must call fput_light before exiting the syscall and returning control
988 * to userspace (i.e. you cannot remember the returned struct file * after
989 * returning to userspace).
990 * 2) You must not call filp_close on the returned struct file * in between
991 * calls to fget_light and fput_light.
992 * 3) You must not clone the current task in between the calls to fget_light
995 * The fput_needed flag returned by fget_light should be passed to the
996 * corresponding fput_light.
998 static unsigned long __fget_light(unsigned int fd, fmode_t mask)
1000 struct files_struct *files = current->files;
1003 if (atomic_read(&files->count) == 1) {
1004 file = files_lookup_fd_raw(files, fd);
1005 if (!file || unlikely(file->f_mode & mask))
1007 return (unsigned long)file;
1009 file = __fget(fd, mask, 1);
1012 return FDPUT_FPUT | (unsigned long)file;
1015 unsigned long __fdget(unsigned int fd)
1017 return __fget_light(fd, FMODE_PATH);
1019 EXPORT_SYMBOL(__fdget);
1021 unsigned long __fdget_raw(unsigned int fd)
1023 return __fget_light(fd, 0);
1026 unsigned long __fdget_pos(unsigned int fd)
1028 unsigned long v = __fdget(fd);
1029 struct file *file = (struct file *)(v & ~3);
1031 if (file && (file->f_mode & FMODE_ATOMIC_POS)) {
1032 if (file_count(file) > 1) {
1033 v |= FDPUT_POS_UNLOCK;
1034 mutex_lock(&file->f_pos_lock);
1040 void __f_unlock_pos(struct file *f)
1042 mutex_unlock(&f->f_pos_lock);
1046 * We only lock f_pos if we have threads or if the file might be
1047 * shared with another process. In both cases we'll have an elevated
1048 * file count (done either by fdget() or by fork()).
1051 void set_close_on_exec(unsigned int fd, int flag)
1053 struct files_struct *files = current->files;
1054 struct fdtable *fdt;
1055 spin_lock(&files->file_lock);
1056 fdt = files_fdtable(files);
1058 __set_close_on_exec(fd, fdt);
1060 __clear_close_on_exec(fd, fdt);
1061 spin_unlock(&files->file_lock);
1064 bool get_close_on_exec(unsigned int fd)
1066 struct files_struct *files = current->files;
1067 struct fdtable *fdt;
1070 fdt = files_fdtable(files);
1071 res = close_on_exec(fd, fdt);
1076 static int do_dup2(struct files_struct *files,
1077 struct file *file, unsigned fd, unsigned flags)
1078 __releases(&files->file_lock)
1080 struct file *tofree;
1081 struct fdtable *fdt;
1084 * We need to detect attempts to do dup2() over allocated but still
1085 * not finished descriptor. NB: OpenBSD avoids that at the price of
1086 * extra work in their equivalent of fget() - they insert struct
1087 * file immediately after grabbing descriptor, mark it larval if
1088 * more work (e.g. actual opening) is needed and make sure that
1089 * fget() treats larval files as absent. Potentially interesting,
1090 * but while extra work in fget() is trivial, locking implications
1091 * and amount of surgery on open()-related paths in VFS are not.
1092 * FreeBSD fails with -EBADF in the same situation, NetBSD "solution"
1093 * deadlocks in rather amusing ways, AFAICS. All of that is out of
1094 * scope of POSIX or SUS, since neither considers shared descriptor
1095 * tables and this condition does not arise without those.
1097 fdt = files_fdtable(files);
1098 tofree = fdt->fd[fd];
1099 if (!tofree && fd_is_open(fd, fdt))
1102 rcu_assign_pointer(fdt->fd[fd], file);
1103 __set_open_fd(fd, fdt);
1104 if (flags & O_CLOEXEC)
1105 __set_close_on_exec(fd, fdt);
1107 __clear_close_on_exec(fd, fdt);
1108 spin_unlock(&files->file_lock);
1111 filp_close(tofree, files);
1116 spin_unlock(&files->file_lock);
1120 int replace_fd(unsigned fd, struct file *file, unsigned flags)
1123 struct files_struct *files = current->files;
1126 return close_fd(fd);
1128 if (fd >= rlimit(RLIMIT_NOFILE))
1131 spin_lock(&files->file_lock);
1132 err = expand_files(files, fd);
1133 if (unlikely(err < 0))
1135 return do_dup2(files, file, fd, flags);
1138 spin_unlock(&files->file_lock);
1143 * __receive_fd() - Install received file into file descriptor table
1144 * @file: struct file that was received from another process
1145 * @ufd: __user pointer to write new fd number to
1146 * @o_flags: the O_* flags to apply to the new fd entry
1148 * Installs a received file into the file descriptor table, with appropriate
1149 * checks and count updates. Optionally writes the fd number to userspace, if
1152 * This helper handles its own reference counting of the incoming
1155 * Returns newly install fd or -ve on error.
1157 int __receive_fd(struct file *file, int __user *ufd, unsigned int o_flags)
1162 error = security_file_receive(file);
1166 new_fd = get_unused_fd_flags(o_flags);
1171 error = put_user(new_fd, ufd);
1173 put_unused_fd(new_fd);
1178 fd_install(new_fd, get_file(file));
1179 __receive_sock(file);
1183 int receive_fd_replace(int new_fd, struct file *file, unsigned int o_flags)
1187 error = security_file_receive(file);
1190 error = replace_fd(new_fd, file, o_flags);
1193 __receive_sock(file);
1197 int receive_fd(struct file *file, unsigned int o_flags)
1199 return __receive_fd(file, NULL, o_flags);
1201 EXPORT_SYMBOL_GPL(receive_fd);
1203 static int ksys_dup3(unsigned int oldfd, unsigned int newfd, int flags)
1207 struct files_struct *files = current->files;
1209 if ((flags & ~O_CLOEXEC) != 0)
1212 if (unlikely(oldfd == newfd))
1215 if (newfd >= rlimit(RLIMIT_NOFILE))
1218 spin_lock(&files->file_lock);
1219 err = expand_files(files, newfd);
1220 file = files_lookup_fd_locked(files, oldfd);
1221 if (unlikely(!file))
1223 if (unlikely(err < 0)) {
1228 return do_dup2(files, file, newfd, flags);
1233 spin_unlock(&files->file_lock);
1237 SYSCALL_DEFINE3(dup3, unsigned int, oldfd, unsigned int, newfd, int, flags)
1239 return ksys_dup3(oldfd, newfd, flags);
1242 SYSCALL_DEFINE2(dup2, unsigned int, oldfd, unsigned int, newfd)
1244 if (unlikely(newfd == oldfd)) { /* corner case */
1245 struct files_struct *files = current->files;
1249 if (!files_lookup_fd_rcu(files, oldfd))
1254 return ksys_dup3(oldfd, newfd, 0);
1257 SYSCALL_DEFINE1(dup, unsigned int, fildes)
1260 struct file *file = fget_raw(fildes);
1263 ret = get_unused_fd_flags(0);
1265 fd_install(ret, file);
1272 int f_dupfd(unsigned int from, struct file *file, unsigned flags)
1274 unsigned long nofile = rlimit(RLIMIT_NOFILE);
1278 err = alloc_fd(from, nofile, flags);
1281 fd_install(err, file);
1286 int iterate_fd(struct files_struct *files, unsigned n,
1287 int (*f)(const void *, struct file *, unsigned),
1290 struct fdtable *fdt;
1294 spin_lock(&files->file_lock);
1295 for (fdt = files_fdtable(files); n < fdt->max_fds; n++) {
1297 file = rcu_dereference_check_fdtable(files, fdt->fd[n]);
1300 res = f(p, file, n);
1304 spin_unlock(&files->file_lock);
1307 EXPORT_SYMBOL(iterate_fd);