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);
91 * Note how the fdtable bitmap allocations very much have to be a multiple of
92 * BITS_PER_LONG. This is not only because we walk those things in chunks of
93 * 'unsigned long' in some places, but simply because that is how the Linux
94 * kernel bitmaps are defined to work: they are not "bits in an array of bytes",
95 * they are very much "bits in an array of unsigned long".
97 * The ALIGN(nr, BITS_PER_LONG) here is for clarity: since we just multiplied
98 * by that "1024/sizeof(ptr)" before, we already know there are sufficient
99 * clear low bits. Clang seems to realize that, gcc ends up being confused.
101 * On a 128-bit machine, the ALIGN() would actually matter. In the meantime,
102 * let's consider it documentation (and maybe a test-case for gcc to improve
103 * its code generation ;)
105 static struct fdtable * alloc_fdtable(unsigned int nr)
111 * Figure out how many fds we actually want to support in this fdtable.
112 * Allocation steps are keyed to the size of the fdarray, since it
113 * grows far faster than any of the other dynamic data. We try to fit
114 * the fdarray into comfortable page-tuned chunks: starting at 1024B
115 * and growing in powers of two from there on.
117 nr /= (1024 / sizeof(struct file *));
118 nr = roundup_pow_of_two(nr + 1);
119 nr *= (1024 / sizeof(struct file *));
120 nr = ALIGN(nr, BITS_PER_LONG);
122 * Note that this can drive nr *below* what we had passed if sysctl_nr_open
123 * had been set lower between the check in expand_files() and here. Deal
124 * with that in caller, it's cheaper that way.
126 * We make sure that nr remains a multiple of BITS_PER_LONG - otherwise
127 * bitmaps handling below becomes unpleasant, to put it mildly...
129 if (unlikely(nr > sysctl_nr_open))
130 nr = ((sysctl_nr_open - 1) | (BITS_PER_LONG - 1)) + 1;
132 fdt = kmalloc(sizeof(struct fdtable), GFP_KERNEL_ACCOUNT);
136 data = kvmalloc_array(nr, sizeof(struct file *), GFP_KERNEL_ACCOUNT);
141 data = kvmalloc(max_t(size_t,
142 2 * nr / BITS_PER_BYTE + BITBIT_SIZE(nr), L1_CACHE_BYTES),
146 fdt->open_fds = data;
147 data += nr / BITS_PER_BYTE;
148 fdt->close_on_exec = data;
149 data += nr / BITS_PER_BYTE;
150 fdt->full_fds_bits = data;
163 * Expand the file descriptor table.
164 * This function will allocate a new fdtable and both fd array and fdset, of
166 * Return <0 error code on error; 1 on successful completion.
167 * The files->file_lock should be held on entry, and will be held on exit.
169 static int expand_fdtable(struct files_struct *files, unsigned int nr)
170 __releases(files->file_lock)
171 __acquires(files->file_lock)
173 struct fdtable *new_fdt, *cur_fdt;
175 spin_unlock(&files->file_lock);
176 new_fdt = alloc_fdtable(nr);
178 /* make sure all __fd_install() have seen resize_in_progress
179 * or have finished their rcu_read_lock_sched() section.
181 if (atomic_read(&files->count) > 1)
184 spin_lock(&files->file_lock);
188 * extremely unlikely race - sysctl_nr_open decreased between the check in
189 * caller and alloc_fdtable(). Cheaper to catch it here...
191 if (unlikely(new_fdt->max_fds <= nr)) {
192 __free_fdtable(new_fdt);
195 cur_fdt = files_fdtable(files);
196 BUG_ON(nr < cur_fdt->max_fds);
197 copy_fdtable(new_fdt, cur_fdt);
198 rcu_assign_pointer(files->fdt, new_fdt);
199 if (cur_fdt != &files->fdtab)
200 call_rcu(&cur_fdt->rcu, free_fdtable_rcu);
201 /* coupled with smp_rmb() in __fd_install() */
208 * This function will expand the file structures, if the requested size exceeds
209 * the current capacity and there is room for expansion.
210 * Return <0 error code on error; 0 when nothing done; 1 when files were
211 * expanded and execution may have blocked.
212 * The files->file_lock should be held on entry, and will be held on exit.
214 static int expand_files(struct files_struct *files, unsigned int nr)
215 __releases(files->file_lock)
216 __acquires(files->file_lock)
222 fdt = files_fdtable(files);
224 /* Do we need to expand? */
225 if (nr < fdt->max_fds)
229 if (nr >= sysctl_nr_open)
232 if (unlikely(files->resize_in_progress)) {
233 spin_unlock(&files->file_lock);
235 wait_event(files->resize_wait, !files->resize_in_progress);
236 spin_lock(&files->file_lock);
240 /* All good, so we try */
241 files->resize_in_progress = true;
242 expanded = expand_fdtable(files, nr);
243 files->resize_in_progress = false;
245 wake_up_all(&files->resize_wait);
249 static inline void __set_close_on_exec(unsigned int fd, struct fdtable *fdt)
251 __set_bit(fd, fdt->close_on_exec);
254 static inline void __clear_close_on_exec(unsigned int fd, struct fdtable *fdt)
256 if (test_bit(fd, fdt->close_on_exec))
257 __clear_bit(fd, fdt->close_on_exec);
260 static inline void __set_open_fd(unsigned int fd, struct fdtable *fdt)
262 __set_bit(fd, fdt->open_fds);
264 if (!~fdt->open_fds[fd])
265 __set_bit(fd, fdt->full_fds_bits);
268 static inline void __clear_open_fd(unsigned int fd, struct fdtable *fdt)
270 __clear_bit(fd, fdt->open_fds);
271 __clear_bit(fd / BITS_PER_LONG, fdt->full_fds_bits);
274 static unsigned int count_open_files(struct fdtable *fdt)
276 unsigned int size = fdt->max_fds;
279 /* Find the last open fd */
280 for (i = size / BITS_PER_LONG; i > 0; ) {
281 if (fdt->open_fds[--i])
284 i = (i + 1) * BITS_PER_LONG;
289 * Note that a sane fdtable size always has to be a multiple of
290 * BITS_PER_LONG, since we have bitmaps that are sized by this.
292 * 'max_fds' will normally already be properly aligned, but it
293 * turns out that in the close_range() -> __close_range() ->
294 * unshare_fd() -> dup_fd() -> sane_fdtable_size() we can end
295 * up having a 'max_fds' value that isn't already aligned.
297 * Rather than make close_range() have to worry about this,
298 * just make that BITS_PER_LONG alignment be part of a sane
299 * fdtable size. Becuase that's really what it is.
301 static unsigned int sane_fdtable_size(struct fdtable *fdt, unsigned int max_fds)
305 count = count_open_files(fdt);
306 if (max_fds < NR_OPEN_DEFAULT)
307 max_fds = NR_OPEN_DEFAULT;
308 return ALIGN(min(count, max_fds), BITS_PER_LONG);
312 * Allocate a new files structure and copy contents from the
313 * passed in files structure.
314 * errorp will be valid only when the returned files_struct is NULL.
316 struct files_struct *dup_fd(struct files_struct *oldf, unsigned int max_fds, int *errorp)
318 struct files_struct *newf;
319 struct file **old_fds, **new_fds;
320 unsigned int open_files, i;
321 struct fdtable *old_fdt, *new_fdt;
324 newf = kmem_cache_alloc(files_cachep, GFP_KERNEL);
328 atomic_set(&newf->count, 1);
330 spin_lock_init(&newf->file_lock);
331 newf->resize_in_progress = false;
332 init_waitqueue_head(&newf->resize_wait);
334 new_fdt = &newf->fdtab;
335 new_fdt->max_fds = NR_OPEN_DEFAULT;
336 new_fdt->close_on_exec = newf->close_on_exec_init;
337 new_fdt->open_fds = newf->open_fds_init;
338 new_fdt->full_fds_bits = newf->full_fds_bits_init;
339 new_fdt->fd = &newf->fd_array[0];
341 spin_lock(&oldf->file_lock);
342 old_fdt = files_fdtable(oldf);
343 open_files = sane_fdtable_size(old_fdt, max_fds);
346 * Check whether we need to allocate a larger fd array and fd set.
348 while (unlikely(open_files > new_fdt->max_fds)) {
349 spin_unlock(&oldf->file_lock);
351 if (new_fdt != &newf->fdtab)
352 __free_fdtable(new_fdt);
354 new_fdt = alloc_fdtable(open_files - 1);
360 /* beyond sysctl_nr_open; nothing to do */
361 if (unlikely(new_fdt->max_fds < open_files)) {
362 __free_fdtable(new_fdt);
368 * Reacquire the oldf lock and a pointer to its fd table
369 * who knows it may have a new bigger fd table. We need
370 * the latest pointer.
372 spin_lock(&oldf->file_lock);
373 old_fdt = files_fdtable(oldf);
374 open_files = sane_fdtable_size(old_fdt, max_fds);
377 copy_fd_bitmaps(new_fdt, old_fdt, open_files);
379 old_fds = old_fdt->fd;
380 new_fds = new_fdt->fd;
382 for (i = open_files; i != 0; i--) {
383 struct file *f = *old_fds++;
388 * The fd may be claimed in the fd bitmap but not yet
389 * instantiated in the files array if a sibling thread
390 * is partway through open(). So make sure that this
391 * fd is available to the new process.
393 __clear_open_fd(open_files - i, new_fdt);
395 rcu_assign_pointer(*new_fds++, f);
397 spin_unlock(&oldf->file_lock);
399 /* clear the remainder */
400 memset(new_fds, 0, (new_fdt->max_fds - open_files) * sizeof(struct file *));
402 rcu_assign_pointer(newf->fdt, new_fdt);
407 kmem_cache_free(files_cachep, newf);
412 static struct fdtable *close_files(struct files_struct * files)
415 * It is safe to dereference the fd table without RCU or
416 * ->file_lock because this is the last reference to the
419 struct fdtable *fdt = rcu_dereference_raw(files->fdt);
420 unsigned int i, j = 0;
424 i = j * BITS_PER_LONG;
425 if (i >= fdt->max_fds)
427 set = fdt->open_fds[j++];
430 struct file * file = xchg(&fdt->fd[i], NULL);
432 filp_close(file, files);
444 struct files_struct *get_files_struct(struct task_struct *task)
446 struct files_struct *files;
451 atomic_inc(&files->count);
457 void put_files_struct(struct files_struct *files)
459 if (atomic_dec_and_test(&files->count)) {
460 struct fdtable *fdt = close_files(files);
462 /* free the arrays if they are not embedded */
463 if (fdt != &files->fdtab)
465 kmem_cache_free(files_cachep, files);
469 void reset_files_struct(struct files_struct *files)
471 struct task_struct *tsk = current;
472 struct files_struct *old;
478 put_files_struct(old);
481 void exit_files(struct task_struct *tsk)
483 struct files_struct * files = tsk->files;
489 put_files_struct(files);
493 struct files_struct init_files = {
494 .count = ATOMIC_INIT(1),
495 .fdt = &init_files.fdtab,
497 .max_fds = NR_OPEN_DEFAULT,
498 .fd = &init_files.fd_array[0],
499 .close_on_exec = init_files.close_on_exec_init,
500 .open_fds = init_files.open_fds_init,
501 .full_fds_bits = init_files.full_fds_bits_init,
503 .file_lock = __SPIN_LOCK_UNLOCKED(init_files.file_lock),
504 .resize_wait = __WAIT_QUEUE_HEAD_INITIALIZER(init_files.resize_wait),
507 static unsigned int find_next_fd(struct fdtable *fdt, unsigned int start)
509 unsigned int maxfd = fdt->max_fds;
510 unsigned int maxbit = maxfd / BITS_PER_LONG;
511 unsigned int bitbit = start / BITS_PER_LONG;
513 bitbit = find_next_zero_bit(fdt->full_fds_bits, maxbit, bitbit) * BITS_PER_LONG;
518 return find_next_zero_bit(fdt->open_fds, maxfd, start);
522 * allocate a file descriptor, mark it busy.
524 int __alloc_fd(struct files_struct *files,
525 unsigned start, unsigned end, unsigned flags)
531 spin_lock(&files->file_lock);
533 fdt = files_fdtable(files);
535 if (fd < files->next_fd)
538 if (fd < fdt->max_fds)
539 fd = find_next_fd(fdt, fd);
542 * N.B. For clone tasks sharing a files structure, this test
543 * will limit the total number of files that can be opened.
549 error = expand_files(files, fd);
554 * If we needed to expand the fs array we
555 * might have blocked - try again.
560 if (start <= files->next_fd)
561 files->next_fd = fd + 1;
563 __set_open_fd(fd, fdt);
564 if (flags & O_CLOEXEC)
565 __set_close_on_exec(fd, fdt);
567 __clear_close_on_exec(fd, fdt);
571 if (rcu_access_pointer(fdt->fd[fd]) != NULL) {
572 printk(KERN_WARNING "alloc_fd: slot %d not NULL!\n", fd);
573 rcu_assign_pointer(fdt->fd[fd], NULL);
578 spin_unlock(&files->file_lock);
582 static int alloc_fd(unsigned start, unsigned flags)
584 return __alloc_fd(current->files, start, rlimit(RLIMIT_NOFILE), flags);
587 int __get_unused_fd_flags(unsigned flags, unsigned long nofile)
589 return __alloc_fd(current->files, 0, nofile, flags);
592 int get_unused_fd_flags(unsigned flags)
594 return __get_unused_fd_flags(flags, rlimit(RLIMIT_NOFILE));
596 EXPORT_SYMBOL(get_unused_fd_flags);
598 static void __put_unused_fd(struct files_struct *files, unsigned int fd)
600 struct fdtable *fdt = files_fdtable(files);
601 __clear_open_fd(fd, fdt);
602 if (fd < files->next_fd)
606 void put_unused_fd(unsigned int fd)
608 struct files_struct *files = current->files;
609 spin_lock(&files->file_lock);
610 __put_unused_fd(files, fd);
611 spin_unlock(&files->file_lock);
614 EXPORT_SYMBOL(put_unused_fd);
617 * Install a file pointer in the fd array.
619 * The VFS is full of places where we drop the files lock between
620 * setting the open_fds bitmap and installing the file in the file
621 * array. At any such point, we are vulnerable to a dup2() race
622 * installing a file in the array before us. We need to detect this and
623 * fput() the struct file we are about to overwrite in this case.
625 * It should never happen - if we allow dup2() do it, _really_ bad things
628 * NOTE: __fd_install() variant is really, really low-level; don't
629 * use it unless you are forced to by truly lousy API shoved down
630 * your throat. 'files' *MUST* be either current->files or obtained
631 * by get_files_struct(current) done by whoever had given it to you,
632 * or really bad things will happen. Normally you want to use
633 * fd_install() instead.
636 void __fd_install(struct files_struct *files, unsigned int fd,
641 rcu_read_lock_sched();
643 if (unlikely(files->resize_in_progress)) {
644 rcu_read_unlock_sched();
645 spin_lock(&files->file_lock);
646 fdt = files_fdtable(files);
647 BUG_ON(fdt->fd[fd] != NULL);
648 rcu_assign_pointer(fdt->fd[fd], file);
649 spin_unlock(&files->file_lock);
652 /* coupled with smp_wmb() in expand_fdtable() */
654 fdt = rcu_dereference_sched(files->fdt);
655 BUG_ON(fdt->fd[fd] != NULL);
656 rcu_assign_pointer(fdt->fd[fd], file);
657 rcu_read_unlock_sched();
661 * This consumes the "file" refcount, so callers should treat it
662 * as if they had called fput(file).
664 void fd_install(unsigned int fd, struct file *file)
666 __fd_install(current->files, fd, file);
669 EXPORT_SYMBOL(fd_install);
671 static struct file *pick_file(struct files_struct *files, unsigned fd)
673 struct file *file = NULL;
676 spin_lock(&files->file_lock);
677 fdt = files_fdtable(files);
678 if (fd >= fdt->max_fds)
680 fd = array_index_nospec(fd, fdt->max_fds);
684 rcu_assign_pointer(fdt->fd[fd], NULL);
685 __put_unused_fd(files, fd);
688 spin_unlock(&files->file_lock);
693 * The same warnings as for __alloc_fd()/__fd_install() apply here...
695 int __close_fd(struct files_struct *files, unsigned fd)
699 file = pick_file(files, fd);
703 return filp_close(file, files);
705 EXPORT_SYMBOL(__close_fd); /* for ksys_close() */
708 * __close_range() - Close all file descriptors in a given range.
710 * @fd: starting file descriptor to close
711 * @max_fd: last file descriptor to close
713 * This closes a range of file descriptors. All file descriptors
714 * from @fd up to and including @max_fd are closed.
716 int __close_range(unsigned fd, unsigned max_fd, unsigned int flags)
718 unsigned int cur_max;
719 struct task_struct *me = current;
720 struct files_struct *cur_fds = me->files, *fds = NULL;
722 if (flags & ~CLOSE_RANGE_UNSHARE)
729 cur_max = files_fdtable(cur_fds)->max_fds;
732 /* cap to last valid index into fdtable */
735 if (flags & CLOSE_RANGE_UNSHARE) {
737 unsigned int max_unshare_fds = NR_OPEN_MAX;
740 * If the requested range is greater than the current maximum,
741 * we're closing everything so only copy all file descriptors
742 * beneath the lowest file descriptor.
744 if (max_fd >= cur_max)
745 max_unshare_fds = fd;
747 ret = unshare_fd(CLONE_FILES, max_unshare_fds, &fds);
752 * We used to share our file descriptor table, and have now
753 * created a private one, make sure we're using it below.
759 max_fd = min(max_fd, cur_max);
760 while (fd <= max_fd) {
763 file = pick_file(cur_fds, fd++);
767 filp_close(file, cur_fds);
773 * We're done closing the files we were supposed to. Time to install
774 * the new file descriptor table and drop the old one.
779 put_files_struct(fds);
786 * See close_fd_get_file() below, this variant assumes current->files->file_lock
789 int __close_fd_get_file(unsigned int fd, struct file **res)
791 struct files_struct *files = current->files;
795 fdt = files_fdtable(files);
796 if (fd >= fdt->max_fds)
801 rcu_assign_pointer(fdt->fd[fd], NULL);
802 __put_unused_fd(files, fd);
812 * variant of close_fd that gets a ref on the file for later fput.
813 * The caller must ensure that filp_close() called on the file, and then
816 int close_fd_get_file(unsigned int fd, struct file **res)
818 struct files_struct *files = current->files;
821 spin_lock(&files->file_lock);
822 ret = __close_fd_get_file(fd, res);
823 spin_unlock(&files->file_lock);
828 void do_close_on_exec(struct files_struct *files)
833 /* exec unshares first */
834 spin_lock(&files->file_lock);
837 unsigned fd = i * BITS_PER_LONG;
838 fdt = files_fdtable(files);
839 if (fd >= fdt->max_fds)
841 set = fdt->close_on_exec[i];
844 fdt->close_on_exec[i] = 0;
845 for ( ; set ; fd++, set >>= 1) {
852 rcu_assign_pointer(fdt->fd[fd], NULL);
853 __put_unused_fd(files, fd);
854 spin_unlock(&files->file_lock);
855 filp_close(file, files);
857 spin_lock(&files->file_lock);
861 spin_unlock(&files->file_lock);
864 static inline struct file *__fget_files_rcu(struct files_struct *files,
865 unsigned int fd, fmode_t mask, unsigned int refs)
869 struct fdtable *fdt = rcu_dereference_raw(files->fdt);
870 struct file __rcu **fdentry;
872 if (unlikely(fd >= fdt->max_fds))
875 fdentry = fdt->fd + array_index_nospec(fd, fdt->max_fds);
876 file = rcu_dereference_raw(*fdentry);
880 if (unlikely(file->f_mode & mask))
884 * Ok, we have a file pointer. However, because we do
885 * this all locklessly under RCU, we may be racing with
886 * that file being closed.
888 * Such a race can take two forms:
890 * (a) the file ref already went down to zero,
891 * and get_file_rcu_many() fails. Just try
894 if (unlikely(!get_file_rcu_many(file, refs)))
898 * (b) the file table entry has changed under us.
899 * Note that we don't need to re-check the 'fdt->fd'
900 * pointer having changed, because it always goes
901 * hand-in-hand with 'fdt'.
903 * If so, we need to put our refs and try again.
905 if (unlikely(rcu_dereference_raw(files->fdt) != fdt) ||
906 unlikely(rcu_dereference_raw(*fdentry) != file)) {
907 fput_many(file, refs);
912 * Ok, we have a ref to the file, and checked that it
919 static struct file *__fget_files(struct files_struct *files, unsigned int fd,
920 fmode_t mask, unsigned int refs)
925 file = __fget_files_rcu(files, fd, mask, refs);
931 static inline struct file *__fget(unsigned int fd, fmode_t mask,
934 return __fget_files(current->files, fd, mask, refs);
937 struct file *fget_many(unsigned int fd, unsigned int refs)
939 return __fget(fd, FMODE_PATH, refs);
942 struct file *fget(unsigned int fd)
944 return __fget(fd, FMODE_PATH, 1);
948 struct file *fget_raw(unsigned int fd)
950 return __fget(fd, 0, 1);
952 EXPORT_SYMBOL(fget_raw);
954 struct file *fget_task(struct task_struct *task, unsigned int fd)
956 struct file *file = NULL;
960 file = __fget_files(task->files, fd, 0, 1);
967 * Lightweight file lookup - no refcnt increment if fd table isn't shared.
969 * You can use this instead of fget if you satisfy all of the following
971 * 1) You must call fput_light before exiting the syscall and returning control
972 * to userspace (i.e. you cannot remember the returned struct file * after
973 * returning to userspace).
974 * 2) You must not call filp_close on the returned struct file * in between
975 * calls to fget_light and fput_light.
976 * 3) You must not clone the current task in between the calls to fget_light
979 * The fput_needed flag returned by fget_light should be passed to the
980 * corresponding fput_light.
982 static unsigned long __fget_light(unsigned int fd, fmode_t mask)
984 struct files_struct *files = current->files;
987 if (atomic_read(&files->count) == 1) {
988 file = __fcheck_files(files, fd);
989 if (!file || unlikely(file->f_mode & mask))
991 return (unsigned long)file;
993 file = __fget(fd, mask, 1);
996 return FDPUT_FPUT | (unsigned long)file;
999 unsigned long __fdget(unsigned int fd)
1001 return __fget_light(fd, FMODE_PATH);
1003 EXPORT_SYMBOL(__fdget);
1005 unsigned long __fdget_raw(unsigned int fd)
1007 return __fget_light(fd, 0);
1011 * Try to avoid f_pos locking. We only need it if the
1012 * file is marked for FMODE_ATOMIC_POS, and it can be
1013 * accessed multiple ways.
1015 * Always do it for directories, because pidfd_getfd()
1016 * can make a file accessible even if it otherwise would
1017 * not be, and for directories this is a correctness
1018 * issue, not a "POSIX requirement".
1020 static inline bool file_needs_f_pos_lock(struct file *file)
1022 return (file->f_mode & FMODE_ATOMIC_POS) &&
1023 (file_count(file) > 1 || S_ISDIR(file_inode(file)->i_mode));
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_needs_f_pos_lock(file)) {
1032 v |= FDPUT_POS_UNLOCK;
1033 mutex_lock(&file->f_pos_lock);
1038 void __f_unlock_pos(struct file *f)
1040 mutex_unlock(&f->f_pos_lock);
1044 * We only lock f_pos if we have threads or if the file might be
1045 * shared with another process. In both cases we'll have an elevated
1046 * file count (done either by fdget() or by fork()).
1049 void set_close_on_exec(unsigned int fd, int flag)
1051 struct files_struct *files = current->files;
1052 struct fdtable *fdt;
1053 spin_lock(&files->file_lock);
1054 fdt = files_fdtable(files);
1056 __set_close_on_exec(fd, fdt);
1058 __clear_close_on_exec(fd, fdt);
1059 spin_unlock(&files->file_lock);
1062 bool get_close_on_exec(unsigned int fd)
1064 struct files_struct *files = current->files;
1065 struct fdtable *fdt;
1068 fdt = files_fdtable(files);
1069 res = close_on_exec(fd, fdt);
1074 static int do_dup2(struct files_struct *files,
1075 struct file *file, unsigned fd, unsigned flags)
1076 __releases(&files->file_lock)
1078 struct file *tofree;
1079 struct fdtable *fdt;
1082 * We need to detect attempts to do dup2() over allocated but still
1083 * not finished descriptor. NB: OpenBSD avoids that at the price of
1084 * extra work in their equivalent of fget() - they insert struct
1085 * file immediately after grabbing descriptor, mark it larval if
1086 * more work (e.g. actual opening) is needed and make sure that
1087 * fget() treats larval files as absent. Potentially interesting,
1088 * but while extra work in fget() is trivial, locking implications
1089 * and amount of surgery on open()-related paths in VFS are not.
1090 * FreeBSD fails with -EBADF in the same situation, NetBSD "solution"
1091 * deadlocks in rather amusing ways, AFAICS. All of that is out of
1092 * scope of POSIX or SUS, since neither considers shared descriptor
1093 * tables and this condition does not arise without those.
1095 fdt = files_fdtable(files);
1096 tofree = fdt->fd[fd];
1097 if (!tofree && fd_is_open(fd, fdt))
1100 rcu_assign_pointer(fdt->fd[fd], file);
1101 __set_open_fd(fd, fdt);
1102 if (flags & O_CLOEXEC)
1103 __set_close_on_exec(fd, fdt);
1105 __clear_close_on_exec(fd, fdt);
1106 spin_unlock(&files->file_lock);
1109 filp_close(tofree, files);
1114 spin_unlock(&files->file_lock);
1118 int replace_fd(unsigned fd, struct file *file, unsigned flags)
1121 struct files_struct *files = current->files;
1124 return __close_fd(files, fd);
1126 if (fd >= rlimit(RLIMIT_NOFILE))
1129 spin_lock(&files->file_lock);
1130 err = expand_files(files, fd);
1131 if (unlikely(err < 0))
1133 return do_dup2(files, file, fd, flags);
1136 spin_unlock(&files->file_lock);
1141 * __receive_fd() - Install received file into file descriptor table
1143 * @fd: fd to install into (if negative, a new fd will be allocated)
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(int fd, struct file *file, int __user *ufd, unsigned int o_flags)
1162 error = security_file_receive(file);
1167 new_fd = get_unused_fd_flags(o_flags);
1175 error = put_user(new_fd, ufd);
1178 put_unused_fd(new_fd);
1184 fd_install(new_fd, get_file(file));
1186 error = replace_fd(new_fd, file, o_flags);
1191 /* Bump the sock usage counts, if any. */
1192 __receive_sock(file);
1196 static int ksys_dup3(unsigned int oldfd, unsigned int newfd, int flags)
1200 struct files_struct *files = current->files;
1202 if ((flags & ~O_CLOEXEC) != 0)
1205 if (unlikely(oldfd == newfd))
1208 if (newfd >= rlimit(RLIMIT_NOFILE))
1211 spin_lock(&files->file_lock);
1212 err = expand_files(files, newfd);
1213 file = fcheck(oldfd);
1214 if (unlikely(!file))
1216 if (unlikely(err < 0)) {
1221 return do_dup2(files, file, newfd, flags);
1226 spin_unlock(&files->file_lock);
1230 SYSCALL_DEFINE3(dup3, unsigned int, oldfd, unsigned int, newfd, int, flags)
1232 return ksys_dup3(oldfd, newfd, flags);
1235 SYSCALL_DEFINE2(dup2, unsigned int, oldfd, unsigned int, newfd)
1237 if (unlikely(newfd == oldfd)) { /* corner case */
1238 struct files_struct *files = current->files;
1242 if (!fcheck_files(files, oldfd))
1247 return ksys_dup3(oldfd, newfd, 0);
1250 SYSCALL_DEFINE1(dup, unsigned int, fildes)
1253 struct file *file = fget_raw(fildes);
1256 ret = get_unused_fd_flags(0);
1258 fd_install(ret, file);
1265 int f_dupfd(unsigned int from, struct file *file, unsigned flags)
1268 if (from >= rlimit(RLIMIT_NOFILE))
1270 err = alloc_fd(from, flags);
1273 fd_install(err, file);
1278 int iterate_fd(struct files_struct *files, unsigned n,
1279 int (*f)(const void *, struct file *, unsigned),
1282 struct fdtable *fdt;
1286 spin_lock(&files->file_lock);
1287 for (fdt = files_fdtable(files); n < fdt->max_fds; n++) {
1289 file = rcu_dereference_check_fdtable(files, fdt->fd[n]);
1292 res = f(p, file, n);
1296 spin_unlock(&files->file_lock);
1299 EXPORT_SYMBOL(iterate_fd);