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);
89 * Note how the fdtable bitmap allocations very much have to be a multiple of
90 * BITS_PER_LONG. This is not only because we walk those things in chunks of
91 * 'unsigned long' in some places, but simply because that is how the Linux
92 * kernel bitmaps are defined to work: they are not "bits in an array of bytes",
93 * they are very much "bits in an array of unsigned long".
95 * The ALIGN(nr, BITS_PER_LONG) here is for clarity: since we just multiplied
96 * by that "1024/sizeof(ptr)" before, we already know there are sufficient
97 * clear low bits. Clang seems to realize that, gcc ends up being confused.
99 * On a 128-bit machine, the ALIGN() would actually matter. In the meantime,
100 * let's consider it documentation (and maybe a test-case for gcc to improve
101 * its code generation ;)
103 static struct fdtable * alloc_fdtable(unsigned int nr)
109 * Figure out how many fds we actually want to support in this fdtable.
110 * Allocation steps are keyed to the size of the fdarray, since it
111 * grows far faster than any of the other dynamic data. We try to fit
112 * the fdarray into comfortable page-tuned chunks: starting at 1024B
113 * and growing in powers of two from there on.
115 nr /= (1024 / sizeof(struct file *));
116 nr = roundup_pow_of_two(nr + 1);
117 nr *= (1024 / sizeof(struct file *));
118 nr = ALIGN(nr, BITS_PER_LONG);
120 * Note that this can drive nr *below* what we had passed if sysctl_nr_open
121 * had been set lower between the check in expand_files() and here. Deal
122 * with that in caller, it's cheaper that way.
124 * We make sure that nr remains a multiple of BITS_PER_LONG - otherwise
125 * bitmaps handling below becomes unpleasant, to put it mildly...
127 if (unlikely(nr > sysctl_nr_open))
128 nr = ((sysctl_nr_open - 1) | (BITS_PER_LONG - 1)) + 1;
130 fdt = kmalloc(sizeof(struct fdtable), GFP_KERNEL_ACCOUNT);
134 data = kvmalloc_array(nr, sizeof(struct file *), GFP_KERNEL_ACCOUNT);
139 data = kvmalloc(max_t(size_t,
140 2 * nr / BITS_PER_BYTE + BITBIT_SIZE(nr), L1_CACHE_BYTES),
144 fdt->open_fds = data;
145 data += nr / BITS_PER_BYTE;
146 fdt->close_on_exec = data;
147 data += nr / BITS_PER_BYTE;
148 fdt->full_fds_bits = data;
161 * Expand the file descriptor table.
162 * This function will allocate a new fdtable and both fd array and fdset, of
164 * Return <0 error code on error; 1 on successful completion.
165 * The files->file_lock should be held on entry, and will be held on exit.
167 static int expand_fdtable(struct files_struct *files, unsigned int nr)
168 __releases(files->file_lock)
169 __acquires(files->file_lock)
171 struct fdtable *new_fdt, *cur_fdt;
173 spin_unlock(&files->file_lock);
174 new_fdt = alloc_fdtable(nr);
176 /* make sure all __fd_install() have seen resize_in_progress
177 * or have finished their rcu_read_lock_sched() section.
179 if (atomic_read(&files->count) > 1)
182 spin_lock(&files->file_lock);
186 * extremely unlikely race - sysctl_nr_open decreased between the check in
187 * caller and alloc_fdtable(). Cheaper to catch it here...
189 if (unlikely(new_fdt->max_fds <= nr)) {
190 __free_fdtable(new_fdt);
193 cur_fdt = files_fdtable(files);
194 BUG_ON(nr < cur_fdt->max_fds);
195 copy_fdtable(new_fdt, cur_fdt);
196 rcu_assign_pointer(files->fdt, new_fdt);
197 if (cur_fdt != &files->fdtab)
198 call_rcu(&cur_fdt->rcu, free_fdtable_rcu);
199 /* coupled with smp_rmb() in __fd_install() */
206 * This function will expand the file structures, if the requested size exceeds
207 * the current capacity and there is room for expansion.
208 * Return <0 error code on error; 0 when nothing done; 1 when files were
209 * expanded and execution may have blocked.
210 * The files->file_lock should be held on entry, and will be held on exit.
212 static int expand_files(struct files_struct *files, unsigned int nr)
213 __releases(files->file_lock)
214 __acquires(files->file_lock)
220 fdt = files_fdtable(files);
222 /* Do we need to expand? */
223 if (nr < fdt->max_fds)
227 if (nr >= sysctl_nr_open)
230 if (unlikely(files->resize_in_progress)) {
231 spin_unlock(&files->file_lock);
233 wait_event(files->resize_wait, !files->resize_in_progress);
234 spin_lock(&files->file_lock);
238 /* All good, so we try */
239 files->resize_in_progress = true;
240 expanded = expand_fdtable(files, nr);
241 files->resize_in_progress = false;
243 wake_up_all(&files->resize_wait);
247 static inline void __set_close_on_exec(unsigned int fd, struct fdtable *fdt)
249 __set_bit(fd, fdt->close_on_exec);
252 static inline void __clear_close_on_exec(unsigned int fd, struct fdtable *fdt)
254 if (test_bit(fd, fdt->close_on_exec))
255 __clear_bit(fd, fdt->close_on_exec);
258 static inline void __set_open_fd(unsigned int fd, struct fdtable *fdt)
260 __set_bit(fd, fdt->open_fds);
262 if (!~fdt->open_fds[fd])
263 __set_bit(fd, fdt->full_fds_bits);
266 static inline void __clear_open_fd(unsigned int fd, struct fdtable *fdt)
268 __clear_bit(fd, fdt->open_fds);
269 __clear_bit(fd / BITS_PER_LONG, fdt->full_fds_bits);
272 static unsigned int count_open_files(struct fdtable *fdt)
274 unsigned int size = fdt->max_fds;
277 /* Find the last open fd */
278 for (i = size / BITS_PER_LONG; i > 0; ) {
279 if (fdt->open_fds[--i])
282 i = (i + 1) * BITS_PER_LONG;
287 * Note that a sane fdtable size always has to be a multiple of
288 * BITS_PER_LONG, since we have bitmaps that are sized by this.
290 * 'max_fds' will normally already be properly aligned, but it
291 * turns out that in the close_range() -> __close_range() ->
292 * unshare_fd() -> dup_fd() -> sane_fdtable_size() we can end
293 * up having a 'max_fds' value that isn't already aligned.
295 * Rather than make close_range() have to worry about this,
296 * just make that BITS_PER_LONG alignment be part of a sane
297 * fdtable size. Becuase that's really what it is.
299 static unsigned int sane_fdtable_size(struct fdtable *fdt, unsigned int max_fds)
303 count = count_open_files(fdt);
304 if (max_fds < NR_OPEN_DEFAULT)
305 max_fds = NR_OPEN_DEFAULT;
306 return ALIGN(min(count, max_fds), BITS_PER_LONG);
310 * Allocate a new files structure and copy contents from the
311 * passed in files structure.
312 * errorp will be valid only when the returned files_struct is NULL.
314 struct files_struct *dup_fd(struct files_struct *oldf, unsigned int max_fds, int *errorp)
316 struct files_struct *newf;
317 struct file **old_fds, **new_fds;
318 unsigned int open_files, i;
319 struct fdtable *old_fdt, *new_fdt;
322 newf = kmem_cache_alloc(files_cachep, GFP_KERNEL);
326 atomic_set(&newf->count, 1);
328 spin_lock_init(&newf->file_lock);
329 newf->resize_in_progress = false;
330 init_waitqueue_head(&newf->resize_wait);
332 new_fdt = &newf->fdtab;
333 new_fdt->max_fds = NR_OPEN_DEFAULT;
334 new_fdt->close_on_exec = newf->close_on_exec_init;
335 new_fdt->open_fds = newf->open_fds_init;
336 new_fdt->full_fds_bits = newf->full_fds_bits_init;
337 new_fdt->fd = &newf->fd_array[0];
339 spin_lock(&oldf->file_lock);
340 old_fdt = files_fdtable(oldf);
341 open_files = sane_fdtable_size(old_fdt, max_fds);
344 * Check whether we need to allocate a larger fd array and fd set.
346 while (unlikely(open_files > new_fdt->max_fds)) {
347 spin_unlock(&oldf->file_lock);
349 if (new_fdt != &newf->fdtab)
350 __free_fdtable(new_fdt);
352 new_fdt = alloc_fdtable(open_files - 1);
358 /* beyond sysctl_nr_open; nothing to do */
359 if (unlikely(new_fdt->max_fds < open_files)) {
360 __free_fdtable(new_fdt);
366 * Reacquire the oldf lock and a pointer to its fd table
367 * who knows it may have a new bigger fd table. We need
368 * the latest pointer.
370 spin_lock(&oldf->file_lock);
371 old_fdt = files_fdtable(oldf);
372 open_files = sane_fdtable_size(old_fdt, max_fds);
375 copy_fd_bitmaps(new_fdt, old_fdt, open_files);
377 old_fds = old_fdt->fd;
378 new_fds = new_fdt->fd;
380 for (i = open_files; i != 0; i--) {
381 struct file *f = *old_fds++;
386 * The fd may be claimed in the fd bitmap but not yet
387 * instantiated in the files array if a sibling thread
388 * is partway through open(). So make sure that this
389 * fd is available to the new process.
391 __clear_open_fd(open_files - i, new_fdt);
393 rcu_assign_pointer(*new_fds++, f);
395 spin_unlock(&oldf->file_lock);
397 /* clear the remainder */
398 memset(new_fds, 0, (new_fdt->max_fds - open_files) * sizeof(struct file *));
400 rcu_assign_pointer(newf->fdt, new_fdt);
405 kmem_cache_free(files_cachep, newf);
410 static struct fdtable *close_files(struct files_struct * files)
413 * It is safe to dereference the fd table without RCU or
414 * ->file_lock because this is the last reference to the
417 struct fdtable *fdt = rcu_dereference_raw(files->fdt);
418 unsigned int i, j = 0;
422 i = j * BITS_PER_LONG;
423 if (i >= fdt->max_fds)
425 set = fdt->open_fds[j++];
428 struct file * file = xchg(&fdt->fd[i], NULL);
430 filp_close(file, files);
442 struct files_struct *get_files_struct(struct task_struct *task)
444 struct files_struct *files;
449 atomic_inc(&files->count);
455 void put_files_struct(struct files_struct *files)
457 if (atomic_dec_and_test(&files->count)) {
458 struct fdtable *fdt = close_files(files);
460 /* free the arrays if they are not embedded */
461 if (fdt != &files->fdtab)
463 kmem_cache_free(files_cachep, files);
467 void reset_files_struct(struct files_struct *files)
469 struct task_struct *tsk = current;
470 struct files_struct *old;
476 put_files_struct(old);
479 void exit_files(struct task_struct *tsk)
481 struct files_struct * files = tsk->files;
487 put_files_struct(files);
491 struct files_struct init_files = {
492 .count = ATOMIC_INIT(1),
493 .fdt = &init_files.fdtab,
495 .max_fds = NR_OPEN_DEFAULT,
496 .fd = &init_files.fd_array[0],
497 .close_on_exec = init_files.close_on_exec_init,
498 .open_fds = init_files.open_fds_init,
499 .full_fds_bits = init_files.full_fds_bits_init,
501 .file_lock = __SPIN_LOCK_UNLOCKED(init_files.file_lock),
502 .resize_wait = __WAIT_QUEUE_HEAD_INITIALIZER(init_files.resize_wait),
505 static unsigned int find_next_fd(struct fdtable *fdt, unsigned int start)
507 unsigned int maxfd = fdt->max_fds;
508 unsigned int maxbit = maxfd / BITS_PER_LONG;
509 unsigned int bitbit = start / BITS_PER_LONG;
511 bitbit = find_next_zero_bit(fdt->full_fds_bits, maxbit, bitbit) * BITS_PER_LONG;
516 return find_next_zero_bit(fdt->open_fds, maxfd, start);
520 * allocate a file descriptor, mark it busy.
522 int __alloc_fd(struct files_struct *files,
523 unsigned start, unsigned end, unsigned flags)
529 spin_lock(&files->file_lock);
531 fdt = files_fdtable(files);
533 if (fd < files->next_fd)
536 if (fd < fdt->max_fds)
537 fd = find_next_fd(fdt, fd);
540 * N.B. For clone tasks sharing a files structure, this test
541 * will limit the total number of files that can be opened.
547 error = expand_files(files, fd);
552 * If we needed to expand the fs array we
553 * might have blocked - try again.
558 if (start <= files->next_fd)
559 files->next_fd = fd + 1;
561 __set_open_fd(fd, fdt);
562 if (flags & O_CLOEXEC)
563 __set_close_on_exec(fd, fdt);
565 __clear_close_on_exec(fd, fdt);
569 if (rcu_access_pointer(fdt->fd[fd]) != NULL) {
570 printk(KERN_WARNING "alloc_fd: slot %d not NULL!\n", fd);
571 rcu_assign_pointer(fdt->fd[fd], NULL);
576 spin_unlock(&files->file_lock);
580 static int alloc_fd(unsigned start, unsigned flags)
582 return __alloc_fd(current->files, start, rlimit(RLIMIT_NOFILE), flags);
585 int __get_unused_fd_flags(unsigned flags, unsigned long nofile)
587 return __alloc_fd(current->files, 0, nofile, flags);
590 int get_unused_fd_flags(unsigned flags)
592 return __get_unused_fd_flags(flags, rlimit(RLIMIT_NOFILE));
594 EXPORT_SYMBOL(get_unused_fd_flags);
596 static void __put_unused_fd(struct files_struct *files, unsigned int fd)
598 struct fdtable *fdt = files_fdtable(files);
599 __clear_open_fd(fd, fdt);
600 if (fd < files->next_fd)
604 void put_unused_fd(unsigned int fd)
606 struct files_struct *files = current->files;
607 spin_lock(&files->file_lock);
608 __put_unused_fd(files, fd);
609 spin_unlock(&files->file_lock);
612 EXPORT_SYMBOL(put_unused_fd);
615 * Install a file pointer in the fd array.
617 * The VFS is full of places where we drop the files lock between
618 * setting the open_fds bitmap and installing the file in the file
619 * array. At any such point, we are vulnerable to a dup2() race
620 * installing a file in the array before us. We need to detect this and
621 * fput() the struct file we are about to overwrite in this case.
623 * It should never happen - if we allow dup2() do it, _really_ bad things
626 * NOTE: __fd_install() variant is really, really low-level; don't
627 * use it unless you are forced to by truly lousy API shoved down
628 * your throat. 'files' *MUST* be either current->files or obtained
629 * by get_files_struct(current) done by whoever had given it to you,
630 * or really bad things will happen. Normally you want to use
631 * fd_install() instead.
634 void __fd_install(struct files_struct *files, unsigned int fd,
639 rcu_read_lock_sched();
641 if (unlikely(files->resize_in_progress)) {
642 rcu_read_unlock_sched();
643 spin_lock(&files->file_lock);
644 fdt = files_fdtable(files);
645 BUG_ON(fdt->fd[fd] != NULL);
646 rcu_assign_pointer(fdt->fd[fd], file);
647 spin_unlock(&files->file_lock);
650 /* coupled with smp_wmb() in expand_fdtable() */
652 fdt = rcu_dereference_sched(files->fdt);
653 BUG_ON(fdt->fd[fd] != NULL);
654 rcu_assign_pointer(fdt->fd[fd], file);
655 rcu_read_unlock_sched();
659 * This consumes the "file" refcount, so callers should treat it
660 * as if they had called fput(file).
662 void fd_install(unsigned int fd, struct file *file)
664 __fd_install(current->files, fd, file);
667 EXPORT_SYMBOL(fd_install);
669 static struct file *pick_file(struct files_struct *files, unsigned fd)
671 struct file *file = NULL;
674 spin_lock(&files->file_lock);
675 fdt = files_fdtable(files);
676 if (fd >= fdt->max_fds)
681 rcu_assign_pointer(fdt->fd[fd], NULL);
682 __put_unused_fd(files, fd);
685 spin_unlock(&files->file_lock);
690 * The same warnings as for __alloc_fd()/__fd_install() apply here...
692 int __close_fd(struct files_struct *files, unsigned fd)
696 file = pick_file(files, fd);
700 return filp_close(file, files);
702 EXPORT_SYMBOL(__close_fd); /* for ksys_close() */
705 * __close_range() - Close all file descriptors in a given range.
707 * @fd: starting file descriptor to close
708 * @max_fd: last file descriptor to close
710 * This closes a range of file descriptors. All file descriptors
711 * from @fd up to and including @max_fd are closed.
713 int __close_range(unsigned fd, unsigned max_fd, unsigned int flags)
715 unsigned int cur_max;
716 struct task_struct *me = current;
717 struct files_struct *cur_fds = me->files, *fds = NULL;
719 if (flags & ~CLOSE_RANGE_UNSHARE)
726 cur_max = files_fdtable(cur_fds)->max_fds;
729 /* cap to last valid index into fdtable */
732 if (flags & CLOSE_RANGE_UNSHARE) {
734 unsigned int max_unshare_fds = NR_OPEN_MAX;
737 * If the requested range is greater than the current maximum,
738 * we're closing everything so only copy all file descriptors
739 * beneath the lowest file descriptor.
741 if (max_fd >= cur_max)
742 max_unshare_fds = fd;
744 ret = unshare_fd(CLONE_FILES, max_unshare_fds, &fds);
749 * We used to share our file descriptor table, and have now
750 * created a private one, make sure we're using it below.
756 max_fd = min(max_fd, cur_max);
757 while (fd <= max_fd) {
760 file = pick_file(cur_fds, fd++);
764 filp_close(file, cur_fds);
770 * We're done closing the files we were supposed to. Time to install
771 * the new file descriptor table and drop the old one.
776 put_files_struct(fds);
783 * variant of __close_fd that gets a ref on the file for later fput.
784 * The caller must ensure that filp_close() called on the file, and then
787 int __close_fd_get_file(unsigned int fd, struct file **res)
789 struct files_struct *files = current->files;
793 spin_lock(&files->file_lock);
794 fdt = files_fdtable(files);
795 if (fd >= fdt->max_fds)
800 rcu_assign_pointer(fdt->fd[fd], NULL);
801 __put_unused_fd(files, fd);
802 spin_unlock(&files->file_lock);
808 spin_unlock(&files->file_lock);
813 void do_close_on_exec(struct files_struct *files)
818 /* exec unshares first */
819 spin_lock(&files->file_lock);
822 unsigned fd = i * BITS_PER_LONG;
823 fdt = files_fdtable(files);
824 if (fd >= fdt->max_fds)
826 set = fdt->close_on_exec[i];
829 fdt->close_on_exec[i] = 0;
830 for ( ; set ; fd++, set >>= 1) {
837 rcu_assign_pointer(fdt->fd[fd], NULL);
838 __put_unused_fd(files, fd);
839 spin_unlock(&files->file_lock);
840 filp_close(file, files);
842 spin_lock(&files->file_lock);
846 spin_unlock(&files->file_lock);
849 static inline struct file *__fget_files_rcu(struct files_struct *files,
850 unsigned int fd, fmode_t mask, unsigned int refs)
854 struct fdtable *fdt = rcu_dereference_raw(files->fdt);
855 struct file __rcu **fdentry;
857 if (unlikely(fd >= fdt->max_fds))
860 fdentry = fdt->fd + array_index_nospec(fd, fdt->max_fds);
861 file = rcu_dereference_raw(*fdentry);
865 if (unlikely(file->f_mode & mask))
869 * Ok, we have a file pointer. However, because we do
870 * this all locklessly under RCU, we may be racing with
871 * that file being closed.
873 * Such a race can take two forms:
875 * (a) the file ref already went down to zero,
876 * and get_file_rcu_many() fails. Just try
879 if (unlikely(!get_file_rcu_many(file, refs)))
883 * (b) the file table entry has changed under us.
884 * Note that we don't need to re-check the 'fdt->fd'
885 * pointer having changed, because it always goes
886 * hand-in-hand with 'fdt'.
888 * If so, we need to put our refs and try again.
890 if (unlikely(rcu_dereference_raw(files->fdt) != fdt) ||
891 unlikely(rcu_dereference_raw(*fdentry) != file)) {
892 fput_many(file, refs);
897 * Ok, we have a ref to the file, and checked that it
904 static struct file *__fget_files(struct files_struct *files, unsigned int fd,
905 fmode_t mask, unsigned int refs)
910 file = __fget_files_rcu(files, fd, mask, refs);
916 static inline struct file *__fget(unsigned int fd, fmode_t mask,
919 return __fget_files(current->files, fd, mask, refs);
922 struct file *fget_many(unsigned int fd, unsigned int refs)
924 return __fget(fd, FMODE_PATH, refs);
927 struct file *fget(unsigned int fd)
929 return __fget(fd, FMODE_PATH, 1);
933 struct file *fget_raw(unsigned int fd)
935 return __fget(fd, 0, 1);
937 EXPORT_SYMBOL(fget_raw);
939 struct file *fget_task(struct task_struct *task, unsigned int fd)
941 struct file *file = NULL;
945 file = __fget_files(task->files, fd, 0, 1);
952 * Lightweight file lookup - no refcnt increment if fd table isn't shared.
954 * You can use this instead of fget if you satisfy all of the following
956 * 1) You must call fput_light before exiting the syscall and returning control
957 * to userspace (i.e. you cannot remember the returned struct file * after
958 * returning to userspace).
959 * 2) You must not call filp_close on the returned struct file * in between
960 * calls to fget_light and fput_light.
961 * 3) You must not clone the current task in between the calls to fget_light
964 * The fput_needed flag returned by fget_light should be passed to the
965 * corresponding fput_light.
967 static unsigned long __fget_light(unsigned int fd, fmode_t mask)
969 struct files_struct *files = current->files;
972 if (atomic_read(&files->count) == 1) {
973 file = __fcheck_files(files, fd);
974 if (!file || unlikely(file->f_mode & mask))
976 return (unsigned long)file;
978 file = __fget(fd, mask, 1);
981 return FDPUT_FPUT | (unsigned long)file;
984 unsigned long __fdget(unsigned int fd)
986 return __fget_light(fd, FMODE_PATH);
988 EXPORT_SYMBOL(__fdget);
990 unsigned long __fdget_raw(unsigned int fd)
992 return __fget_light(fd, 0);
995 unsigned long __fdget_pos(unsigned int fd)
997 unsigned long v = __fdget(fd);
998 struct file *file = (struct file *)(v & ~3);
1000 if (file && (file->f_mode & FMODE_ATOMIC_POS)) {
1001 if (file_count(file) > 1) {
1002 v |= FDPUT_POS_UNLOCK;
1003 mutex_lock(&file->f_pos_lock);
1009 void __f_unlock_pos(struct file *f)
1011 mutex_unlock(&f->f_pos_lock);
1015 * We only lock f_pos if we have threads or if the file might be
1016 * shared with another process. In both cases we'll have an elevated
1017 * file count (done either by fdget() or by fork()).
1020 void set_close_on_exec(unsigned int fd, int flag)
1022 struct files_struct *files = current->files;
1023 struct fdtable *fdt;
1024 spin_lock(&files->file_lock);
1025 fdt = files_fdtable(files);
1027 __set_close_on_exec(fd, fdt);
1029 __clear_close_on_exec(fd, fdt);
1030 spin_unlock(&files->file_lock);
1033 bool get_close_on_exec(unsigned int fd)
1035 struct files_struct *files = current->files;
1036 struct fdtable *fdt;
1039 fdt = files_fdtable(files);
1040 res = close_on_exec(fd, fdt);
1045 static int do_dup2(struct files_struct *files,
1046 struct file *file, unsigned fd, unsigned flags)
1047 __releases(&files->file_lock)
1049 struct file *tofree;
1050 struct fdtable *fdt;
1053 * We need to detect attempts to do dup2() over allocated but still
1054 * not finished descriptor. NB: OpenBSD avoids that at the price of
1055 * extra work in their equivalent of fget() - they insert struct
1056 * file immediately after grabbing descriptor, mark it larval if
1057 * more work (e.g. actual opening) is needed and make sure that
1058 * fget() treats larval files as absent. Potentially interesting,
1059 * but while extra work in fget() is trivial, locking implications
1060 * and amount of surgery on open()-related paths in VFS are not.
1061 * FreeBSD fails with -EBADF in the same situation, NetBSD "solution"
1062 * deadlocks in rather amusing ways, AFAICS. All of that is out of
1063 * scope of POSIX or SUS, since neither considers shared descriptor
1064 * tables and this condition does not arise without those.
1066 fdt = files_fdtable(files);
1067 tofree = fdt->fd[fd];
1068 if (!tofree && fd_is_open(fd, fdt))
1071 rcu_assign_pointer(fdt->fd[fd], file);
1072 __set_open_fd(fd, fdt);
1073 if (flags & O_CLOEXEC)
1074 __set_close_on_exec(fd, fdt);
1076 __clear_close_on_exec(fd, fdt);
1077 spin_unlock(&files->file_lock);
1080 filp_close(tofree, files);
1085 spin_unlock(&files->file_lock);
1089 int replace_fd(unsigned fd, struct file *file, unsigned flags)
1092 struct files_struct *files = current->files;
1095 return __close_fd(files, fd);
1097 if (fd >= rlimit(RLIMIT_NOFILE))
1100 spin_lock(&files->file_lock);
1101 err = expand_files(files, fd);
1102 if (unlikely(err < 0))
1104 return do_dup2(files, file, fd, flags);
1107 spin_unlock(&files->file_lock);
1112 * __receive_fd() - Install received file into file descriptor table
1114 * @fd: fd to install into (if negative, a new fd will be allocated)
1115 * @file: struct file that was received from another process
1116 * @ufd: __user pointer to write new fd number to
1117 * @o_flags: the O_* flags to apply to the new fd entry
1119 * Installs a received file into the file descriptor table, with appropriate
1120 * checks and count updates. Optionally writes the fd number to userspace, if
1123 * This helper handles its own reference counting of the incoming
1126 * Returns newly install fd or -ve on error.
1128 int __receive_fd(int fd, struct file *file, int __user *ufd, unsigned int o_flags)
1133 error = security_file_receive(file);
1138 new_fd = get_unused_fd_flags(o_flags);
1146 error = put_user(new_fd, ufd);
1149 put_unused_fd(new_fd);
1155 fd_install(new_fd, get_file(file));
1157 error = replace_fd(new_fd, file, o_flags);
1162 /* Bump the sock usage counts, if any. */
1163 __receive_sock(file);
1167 static int ksys_dup3(unsigned int oldfd, unsigned int newfd, int flags)
1171 struct files_struct *files = current->files;
1173 if ((flags & ~O_CLOEXEC) != 0)
1176 if (unlikely(oldfd == newfd))
1179 if (newfd >= rlimit(RLIMIT_NOFILE))
1182 spin_lock(&files->file_lock);
1183 err = expand_files(files, newfd);
1184 file = fcheck(oldfd);
1185 if (unlikely(!file))
1187 if (unlikely(err < 0)) {
1192 return do_dup2(files, file, newfd, flags);
1197 spin_unlock(&files->file_lock);
1201 SYSCALL_DEFINE3(dup3, unsigned int, oldfd, unsigned int, newfd, int, flags)
1203 return ksys_dup3(oldfd, newfd, flags);
1206 SYSCALL_DEFINE2(dup2, unsigned int, oldfd, unsigned int, newfd)
1208 if (unlikely(newfd == oldfd)) { /* corner case */
1209 struct files_struct *files = current->files;
1213 if (!fcheck_files(files, oldfd))
1218 return ksys_dup3(oldfd, newfd, 0);
1221 SYSCALL_DEFINE1(dup, unsigned int, fildes)
1224 struct file *file = fget_raw(fildes);
1227 ret = get_unused_fd_flags(0);
1229 fd_install(ret, file);
1236 int f_dupfd(unsigned int from, struct file *file, unsigned flags)
1239 if (from >= rlimit(RLIMIT_NOFILE))
1241 err = alloc_fd(from, flags);
1244 fd_install(err, file);
1249 int iterate_fd(struct files_struct *files, unsigned n,
1250 int (*f)(const void *, struct file *, unsigned),
1253 struct fdtable *fdt;
1257 spin_lock(&files->file_lock);
1258 for (fdt = files_fdtable(files); n < fdt->max_fds; n++) {
1260 file = rcu_dereference_check_fdtable(files, fdt->fd[n]);
1263 res = f(p, file, n);
1267 spin_unlock(&files->file_lock);
1270 EXPORT_SYMBOL(iterate_fd);