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>
14 #include <linux/sched/signal.h>
15 #include <linux/slab.h>
16 #include <linux/file.h>
17 #include <linux/fdtable.h>
18 #include <linux/bitops.h>
19 #include <linux/spinlock.h>
20 #include <linux/rcupdate.h>
23 unsigned int sysctl_nr_open __read_mostly = 1024*1024;
24 unsigned int sysctl_nr_open_min = BITS_PER_LONG;
25 /* our min() is unusable in constant expressions ;-/ */
26 #define __const_min(x, y) ((x) < (y) ? (x) : (y))
27 unsigned int sysctl_nr_open_max =
28 __const_min(INT_MAX, ~(size_t)0/sizeof(void *)) & -BITS_PER_LONG;
30 static void __free_fdtable(struct fdtable *fdt)
33 kvfree(fdt->open_fds);
37 static void free_fdtable_rcu(struct rcu_head *rcu)
39 __free_fdtable(container_of(rcu, struct fdtable, rcu));
42 #define BITBIT_NR(nr) BITS_TO_LONGS(BITS_TO_LONGS(nr))
43 #define BITBIT_SIZE(nr) (BITBIT_NR(nr) * sizeof(long))
46 * Copy 'count' fd bits from the old table to the new table and clear the extra
47 * space if any. This does not copy the file pointers. Called with the files
48 * spinlock held for write.
50 static void copy_fd_bitmaps(struct fdtable *nfdt, struct fdtable *ofdt,
53 unsigned int cpy, set;
55 cpy = count / BITS_PER_BYTE;
56 set = (nfdt->max_fds - count) / BITS_PER_BYTE;
57 memcpy(nfdt->open_fds, ofdt->open_fds, cpy);
58 memset((char *)nfdt->open_fds + cpy, 0, set);
59 memcpy(nfdt->close_on_exec, ofdt->close_on_exec, cpy);
60 memset((char *)nfdt->close_on_exec + cpy, 0, set);
62 cpy = BITBIT_SIZE(count);
63 set = BITBIT_SIZE(nfdt->max_fds) - cpy;
64 memcpy(nfdt->full_fds_bits, ofdt->full_fds_bits, cpy);
65 memset((char *)nfdt->full_fds_bits + cpy, 0, set);
69 * Copy all file descriptors from the old table to the new, expanded table and
70 * clear the extra space. Called with the files spinlock held for write.
72 static void copy_fdtable(struct fdtable *nfdt, struct fdtable *ofdt)
76 BUG_ON(nfdt->max_fds < ofdt->max_fds);
78 cpy = ofdt->max_fds * sizeof(struct file *);
79 set = (nfdt->max_fds - ofdt->max_fds) * sizeof(struct file *);
80 memcpy(nfdt->fd, ofdt->fd, cpy);
81 memset((char *)nfdt->fd + cpy, 0, set);
83 copy_fd_bitmaps(nfdt, ofdt, ofdt->max_fds);
86 static struct fdtable * alloc_fdtable(unsigned int nr)
92 * Figure out how many fds we actually want to support in this fdtable.
93 * Allocation steps are keyed to the size of the fdarray, since it
94 * grows far faster than any of the other dynamic data. We try to fit
95 * the fdarray into comfortable page-tuned chunks: starting at 1024B
96 * and growing in powers of two from there on.
98 nr /= (1024 / sizeof(struct file *));
99 nr = roundup_pow_of_two(nr + 1);
100 nr *= (1024 / sizeof(struct file *));
102 * Note that this can drive nr *below* what we had passed if sysctl_nr_open
103 * had been set lower between the check in expand_files() and here. Deal
104 * with that in caller, it's cheaper that way.
106 * We make sure that nr remains a multiple of BITS_PER_LONG - otherwise
107 * bitmaps handling below becomes unpleasant, to put it mildly...
109 if (unlikely(nr > sysctl_nr_open))
110 nr = ((sysctl_nr_open - 1) | (BITS_PER_LONG - 1)) + 1;
112 fdt = kmalloc(sizeof(struct fdtable), GFP_KERNEL_ACCOUNT);
116 data = kvmalloc_array(nr, sizeof(struct file *), GFP_KERNEL_ACCOUNT);
121 data = kvmalloc(max_t(size_t,
122 2 * nr / BITS_PER_BYTE + BITBIT_SIZE(nr), L1_CACHE_BYTES),
126 fdt->open_fds = data;
127 data += nr / BITS_PER_BYTE;
128 fdt->close_on_exec = data;
129 data += nr / BITS_PER_BYTE;
130 fdt->full_fds_bits = data;
143 * Expand the file descriptor table.
144 * This function will allocate a new fdtable and both fd array and fdset, of
146 * Return <0 error code on error; 1 on successful completion.
147 * The files->file_lock should be held on entry, and will be held on exit.
149 static int expand_fdtable(struct files_struct *files, unsigned int nr)
150 __releases(files->file_lock)
151 __acquires(files->file_lock)
153 struct fdtable *new_fdt, *cur_fdt;
155 spin_unlock(&files->file_lock);
156 new_fdt = alloc_fdtable(nr);
158 /* make sure all __fd_install() have seen resize_in_progress
159 * or have finished their rcu_read_lock_sched() section.
161 if (atomic_read(&files->count) > 1)
164 spin_lock(&files->file_lock);
168 * extremely unlikely race - sysctl_nr_open decreased between the check in
169 * caller and alloc_fdtable(). Cheaper to catch it here...
171 if (unlikely(new_fdt->max_fds <= nr)) {
172 __free_fdtable(new_fdt);
175 cur_fdt = files_fdtable(files);
176 BUG_ON(nr < cur_fdt->max_fds);
177 copy_fdtable(new_fdt, cur_fdt);
178 rcu_assign_pointer(files->fdt, new_fdt);
179 if (cur_fdt != &files->fdtab)
180 call_rcu(&cur_fdt->rcu, free_fdtable_rcu);
181 /* coupled with smp_rmb() in __fd_install() */
188 * This function will expand the file structures, if the requested size exceeds
189 * the current capacity and there is room for expansion.
190 * Return <0 error code on error; 0 when nothing done; 1 when files were
191 * expanded and execution may have blocked.
192 * The files->file_lock should be held on entry, and will be held on exit.
194 static int expand_files(struct files_struct *files, unsigned int nr)
195 __releases(files->file_lock)
196 __acquires(files->file_lock)
202 fdt = files_fdtable(files);
204 /* Do we need to expand? */
205 if (nr < fdt->max_fds)
209 if (nr >= sysctl_nr_open)
212 if (unlikely(files->resize_in_progress)) {
213 spin_unlock(&files->file_lock);
215 wait_event(files->resize_wait, !files->resize_in_progress);
216 spin_lock(&files->file_lock);
220 /* All good, so we try */
221 files->resize_in_progress = true;
222 expanded = expand_fdtable(files, nr);
223 files->resize_in_progress = false;
225 wake_up_all(&files->resize_wait);
229 static inline void __set_close_on_exec(unsigned int fd, struct fdtable *fdt)
231 __set_bit(fd, fdt->close_on_exec);
234 static inline void __clear_close_on_exec(unsigned int fd, struct fdtable *fdt)
236 if (test_bit(fd, fdt->close_on_exec))
237 __clear_bit(fd, fdt->close_on_exec);
240 static inline void __set_open_fd(unsigned int fd, struct fdtable *fdt)
242 __set_bit(fd, fdt->open_fds);
244 if (!~fdt->open_fds[fd])
245 __set_bit(fd, fdt->full_fds_bits);
248 static inline void __clear_open_fd(unsigned int fd, struct fdtable *fdt)
250 __clear_bit(fd, fdt->open_fds);
251 __clear_bit(fd / BITS_PER_LONG, fdt->full_fds_bits);
254 static unsigned int count_open_files(struct fdtable *fdt)
256 unsigned int size = fdt->max_fds;
259 /* Find the last open fd */
260 for (i = size / BITS_PER_LONG; i > 0; ) {
261 if (fdt->open_fds[--i])
264 i = (i + 1) * BITS_PER_LONG;
269 * Allocate a new files structure and copy contents from the
270 * passed in files structure.
271 * errorp will be valid only when the returned files_struct is NULL.
273 struct files_struct *dup_fd(struct files_struct *oldf, int *errorp)
275 struct files_struct *newf;
276 struct file **old_fds, **new_fds;
277 unsigned int open_files, i;
278 struct fdtable *old_fdt, *new_fdt;
281 newf = kmem_cache_alloc(files_cachep, GFP_KERNEL);
285 atomic_set(&newf->count, 1);
287 spin_lock_init(&newf->file_lock);
288 newf->resize_in_progress = false;
289 init_waitqueue_head(&newf->resize_wait);
291 new_fdt = &newf->fdtab;
292 new_fdt->max_fds = NR_OPEN_DEFAULT;
293 new_fdt->close_on_exec = newf->close_on_exec_init;
294 new_fdt->open_fds = newf->open_fds_init;
295 new_fdt->full_fds_bits = newf->full_fds_bits_init;
296 new_fdt->fd = &newf->fd_array[0];
298 spin_lock(&oldf->file_lock);
299 old_fdt = files_fdtable(oldf);
300 open_files = count_open_files(old_fdt);
303 * Check whether we need to allocate a larger fd array and fd set.
305 while (unlikely(open_files > new_fdt->max_fds)) {
306 spin_unlock(&oldf->file_lock);
308 if (new_fdt != &newf->fdtab)
309 __free_fdtable(new_fdt);
311 new_fdt = alloc_fdtable(open_files - 1);
317 /* beyond sysctl_nr_open; nothing to do */
318 if (unlikely(new_fdt->max_fds < open_files)) {
319 __free_fdtable(new_fdt);
325 * Reacquire the oldf lock and a pointer to its fd table
326 * who knows it may have a new bigger fd table. We need
327 * the latest pointer.
329 spin_lock(&oldf->file_lock);
330 old_fdt = files_fdtable(oldf);
331 open_files = count_open_files(old_fdt);
334 copy_fd_bitmaps(new_fdt, old_fdt, open_files);
336 old_fds = old_fdt->fd;
337 new_fds = new_fdt->fd;
339 for (i = open_files; i != 0; i--) {
340 struct file *f = *old_fds++;
345 * The fd may be claimed in the fd bitmap but not yet
346 * instantiated in the files array if a sibling thread
347 * is partway through open(). So make sure that this
348 * fd is available to the new process.
350 __clear_open_fd(open_files - i, new_fdt);
352 rcu_assign_pointer(*new_fds++, f);
354 spin_unlock(&oldf->file_lock);
356 /* clear the remainder */
357 memset(new_fds, 0, (new_fdt->max_fds - open_files) * sizeof(struct file *));
359 rcu_assign_pointer(newf->fdt, new_fdt);
364 kmem_cache_free(files_cachep, newf);
369 static struct fdtable *close_files(struct files_struct * files)
372 * It is safe to dereference the fd table without RCU or
373 * ->file_lock because this is the last reference to the
376 struct fdtable *fdt = rcu_dereference_raw(files->fdt);
377 unsigned int i, j = 0;
381 i = j * BITS_PER_LONG;
382 if (i >= fdt->max_fds)
384 set = fdt->open_fds[j++];
387 struct file * file = xchg(&fdt->fd[i], NULL);
389 filp_close(file, files);
401 struct files_struct *get_files_struct(struct task_struct *task)
403 struct files_struct *files;
408 atomic_inc(&files->count);
414 void put_files_struct(struct files_struct *files)
416 if (atomic_dec_and_test(&files->count)) {
417 struct fdtable *fdt = close_files(files);
419 /* free the arrays if they are not embedded */
420 if (fdt != &files->fdtab)
422 kmem_cache_free(files_cachep, files);
426 void reset_files_struct(struct files_struct *files)
428 struct task_struct *tsk = current;
429 struct files_struct *old;
435 put_files_struct(old);
438 void exit_files(struct task_struct *tsk)
440 struct files_struct * files = tsk->files;
446 put_files_struct(files);
450 struct files_struct init_files = {
451 .count = ATOMIC_INIT(1),
452 .fdt = &init_files.fdtab,
454 .max_fds = NR_OPEN_DEFAULT,
455 .fd = &init_files.fd_array[0],
456 .close_on_exec = init_files.close_on_exec_init,
457 .open_fds = init_files.open_fds_init,
458 .full_fds_bits = init_files.full_fds_bits_init,
460 .file_lock = __SPIN_LOCK_UNLOCKED(init_files.file_lock),
461 .resize_wait = __WAIT_QUEUE_HEAD_INITIALIZER(init_files.resize_wait),
464 static unsigned int find_next_fd(struct fdtable *fdt, unsigned int start)
466 unsigned int maxfd = fdt->max_fds;
467 unsigned int maxbit = maxfd / BITS_PER_LONG;
468 unsigned int bitbit = start / BITS_PER_LONG;
470 bitbit = find_next_zero_bit(fdt->full_fds_bits, maxbit, bitbit) * BITS_PER_LONG;
475 return find_next_zero_bit(fdt->open_fds, maxfd, start);
479 * allocate a file descriptor, mark it busy.
481 int __alloc_fd(struct files_struct *files,
482 unsigned start, unsigned end, unsigned flags)
488 spin_lock(&files->file_lock);
490 fdt = files_fdtable(files);
492 if (fd < files->next_fd)
495 if (fd < fdt->max_fds)
496 fd = find_next_fd(fdt, fd);
499 * N.B. For clone tasks sharing a files structure, this test
500 * will limit the total number of files that can be opened.
506 error = expand_files(files, fd);
511 * If we needed to expand the fs array we
512 * might have blocked - try again.
517 if (start <= files->next_fd)
518 files->next_fd = fd + 1;
520 __set_open_fd(fd, fdt);
521 if (flags & O_CLOEXEC)
522 __set_close_on_exec(fd, fdt);
524 __clear_close_on_exec(fd, fdt);
528 if (rcu_access_pointer(fdt->fd[fd]) != NULL) {
529 printk(KERN_WARNING "alloc_fd: slot %d not NULL!\n", fd);
530 rcu_assign_pointer(fdt->fd[fd], NULL);
535 spin_unlock(&files->file_lock);
539 static int alloc_fd(unsigned start, unsigned flags)
541 return __alloc_fd(current->files, start, rlimit(RLIMIT_NOFILE), flags);
544 int __get_unused_fd_flags(unsigned flags, unsigned long nofile)
546 return __alloc_fd(current->files, 0, nofile, flags);
549 int get_unused_fd_flags(unsigned flags)
551 return __get_unused_fd_flags(flags, rlimit(RLIMIT_NOFILE));
553 EXPORT_SYMBOL(get_unused_fd_flags);
555 static void __put_unused_fd(struct files_struct *files, unsigned int fd)
557 struct fdtable *fdt = files_fdtable(files);
558 __clear_open_fd(fd, fdt);
559 if (fd < files->next_fd)
563 void put_unused_fd(unsigned int fd)
565 struct files_struct *files = current->files;
566 spin_lock(&files->file_lock);
567 __put_unused_fd(files, fd);
568 spin_unlock(&files->file_lock);
571 EXPORT_SYMBOL(put_unused_fd);
574 * Install a file pointer in the fd array.
576 * The VFS is full of places where we drop the files lock between
577 * setting the open_fds bitmap and installing the file in the file
578 * array. At any such point, we are vulnerable to a dup2() race
579 * installing a file in the array before us. We need to detect this and
580 * fput() the struct file we are about to overwrite in this case.
582 * It should never happen - if we allow dup2() do it, _really_ bad things
585 * NOTE: __fd_install() variant is really, really low-level; don't
586 * use it unless you are forced to by truly lousy API shoved down
587 * your throat. 'files' *MUST* be either current->files or obtained
588 * by get_files_struct(current) done by whoever had given it to you,
589 * or really bad things will happen. Normally you want to use
590 * fd_install() instead.
593 void __fd_install(struct files_struct *files, unsigned int fd,
598 rcu_read_lock_sched();
600 if (unlikely(files->resize_in_progress)) {
601 rcu_read_unlock_sched();
602 spin_lock(&files->file_lock);
603 fdt = files_fdtable(files);
604 BUG_ON(fdt->fd[fd] != NULL);
605 rcu_assign_pointer(fdt->fd[fd], file);
606 spin_unlock(&files->file_lock);
609 /* coupled with smp_wmb() in expand_fdtable() */
611 fdt = rcu_dereference_sched(files->fdt);
612 BUG_ON(fdt->fd[fd] != NULL);
613 rcu_assign_pointer(fdt->fd[fd], file);
614 rcu_read_unlock_sched();
618 * This consumes the "file" refcount, so callers should treat it
619 * as if they had called fput(file).
621 void fd_install(unsigned int fd, struct file *file)
623 __fd_install(current->files, fd, file);
626 EXPORT_SYMBOL(fd_install);
629 * The same warnings as for __alloc_fd()/__fd_install() apply here...
631 int __close_fd(struct files_struct *files, unsigned fd)
636 spin_lock(&files->file_lock);
637 fdt = files_fdtable(files);
638 if (fd >= fdt->max_fds)
643 rcu_assign_pointer(fdt->fd[fd], NULL);
644 __put_unused_fd(files, fd);
645 spin_unlock(&files->file_lock);
646 return filp_close(file, files);
649 spin_unlock(&files->file_lock);
652 EXPORT_SYMBOL(__close_fd); /* for ksys_close() */
655 * variant of __close_fd that gets a ref on the file for later fput.
656 * The caller must ensure that filp_close() called on the file, and then
659 int __close_fd_get_file(unsigned int fd, struct file **res)
661 struct files_struct *files = current->files;
665 spin_lock(&files->file_lock);
666 fdt = files_fdtable(files);
667 if (fd >= fdt->max_fds)
672 rcu_assign_pointer(fdt->fd[fd], NULL);
673 __put_unused_fd(files, fd);
674 spin_unlock(&files->file_lock);
680 spin_unlock(&files->file_lock);
685 void do_close_on_exec(struct files_struct *files)
690 /* exec unshares first */
691 spin_lock(&files->file_lock);
694 unsigned fd = i * BITS_PER_LONG;
695 fdt = files_fdtable(files);
696 if (fd >= fdt->max_fds)
698 set = fdt->close_on_exec[i];
701 fdt->close_on_exec[i] = 0;
702 for ( ; set ; fd++, set >>= 1) {
709 rcu_assign_pointer(fdt->fd[fd], NULL);
710 __put_unused_fd(files, fd);
711 spin_unlock(&files->file_lock);
712 filp_close(file, files);
714 spin_lock(&files->file_lock);
718 spin_unlock(&files->file_lock);
721 static struct file *__fget_files(struct files_struct *files, unsigned int fd,
722 fmode_t mask, unsigned int refs)
728 file = fcheck_files(files, fd);
730 /* File object ref couldn't be taken.
731 * dup2() atomicity guarantee is the reason
732 * we loop to catch the new file (or NULL pointer)
734 if (file->f_mode & mask)
736 else if (!get_file_rcu_many(file, refs))
744 static inline struct file *__fget(unsigned int fd, fmode_t mask,
747 return __fget_files(current->files, fd, mask, refs);
750 struct file *fget_many(unsigned int fd, unsigned int refs)
752 return __fget(fd, FMODE_PATH, refs);
755 struct file *fget(unsigned int fd)
757 return __fget(fd, FMODE_PATH, 1);
761 struct file *fget_raw(unsigned int fd)
763 return __fget(fd, 0, 1);
765 EXPORT_SYMBOL(fget_raw);
767 struct file *fget_task(struct task_struct *task, unsigned int fd)
769 struct file *file = NULL;
773 file = __fget_files(task->files, fd, 0, 1);
780 * Lightweight file lookup - no refcnt increment if fd table isn't shared.
782 * You can use this instead of fget if you satisfy all of the following
784 * 1) You must call fput_light before exiting the syscall and returning control
785 * to userspace (i.e. you cannot remember the returned struct file * after
786 * returning to userspace).
787 * 2) You must not call filp_close on the returned struct file * in between
788 * calls to fget_light and fput_light.
789 * 3) You must not clone the current task in between the calls to fget_light
792 * The fput_needed flag returned by fget_light should be passed to the
793 * corresponding fput_light.
795 static unsigned long __fget_light(unsigned int fd, fmode_t mask)
797 struct files_struct *files = current->files;
800 if (atomic_read(&files->count) == 1) {
801 file = __fcheck_files(files, fd);
802 if (!file || unlikely(file->f_mode & mask))
804 return (unsigned long)file;
806 file = __fget(fd, mask, 1);
809 return FDPUT_FPUT | (unsigned long)file;
812 unsigned long __fdget(unsigned int fd)
814 return __fget_light(fd, FMODE_PATH);
816 EXPORT_SYMBOL(__fdget);
818 unsigned long __fdget_raw(unsigned int fd)
820 return __fget_light(fd, 0);
823 unsigned long __fdget_pos(unsigned int fd)
825 unsigned long v = __fdget(fd);
826 struct file *file = (struct file *)(v & ~3);
828 if (file && (file->f_mode & FMODE_ATOMIC_POS)) {
829 if (file_count(file) > 1) {
830 v |= FDPUT_POS_UNLOCK;
831 mutex_lock(&file->f_pos_lock);
837 void __f_unlock_pos(struct file *f)
839 mutex_unlock(&f->f_pos_lock);
843 * We only lock f_pos if we have threads or if the file might be
844 * shared with another process. In both cases we'll have an elevated
845 * file count (done either by fdget() or by fork()).
848 void set_close_on_exec(unsigned int fd, int flag)
850 struct files_struct *files = current->files;
852 spin_lock(&files->file_lock);
853 fdt = files_fdtable(files);
855 __set_close_on_exec(fd, fdt);
857 __clear_close_on_exec(fd, fdt);
858 spin_unlock(&files->file_lock);
861 bool get_close_on_exec(unsigned int fd)
863 struct files_struct *files = current->files;
867 fdt = files_fdtable(files);
868 res = close_on_exec(fd, fdt);
873 static int do_dup2(struct files_struct *files,
874 struct file *file, unsigned fd, unsigned flags)
875 __releases(&files->file_lock)
881 * We need to detect attempts to do dup2() over allocated but still
882 * not finished descriptor. NB: OpenBSD avoids that at the price of
883 * extra work in their equivalent of fget() - they insert struct
884 * file immediately after grabbing descriptor, mark it larval if
885 * more work (e.g. actual opening) is needed and make sure that
886 * fget() treats larval files as absent. Potentially interesting,
887 * but while extra work in fget() is trivial, locking implications
888 * and amount of surgery on open()-related paths in VFS are not.
889 * FreeBSD fails with -EBADF in the same situation, NetBSD "solution"
890 * deadlocks in rather amusing ways, AFAICS. All of that is out of
891 * scope of POSIX or SUS, since neither considers shared descriptor
892 * tables and this condition does not arise without those.
894 fdt = files_fdtable(files);
895 tofree = fdt->fd[fd];
896 if (!tofree && fd_is_open(fd, fdt))
899 rcu_assign_pointer(fdt->fd[fd], file);
900 __set_open_fd(fd, fdt);
901 if (flags & O_CLOEXEC)
902 __set_close_on_exec(fd, fdt);
904 __clear_close_on_exec(fd, fdt);
905 spin_unlock(&files->file_lock);
908 filp_close(tofree, files);
913 spin_unlock(&files->file_lock);
917 int replace_fd(unsigned fd, struct file *file, unsigned flags)
920 struct files_struct *files = current->files;
923 return __close_fd(files, fd);
925 if (fd >= rlimit(RLIMIT_NOFILE))
928 spin_lock(&files->file_lock);
929 err = expand_files(files, fd);
930 if (unlikely(err < 0))
932 return do_dup2(files, file, fd, flags);
935 spin_unlock(&files->file_lock);
940 * __receive_fd() - Install received file into file descriptor table
942 * @fd: fd to install into (if negative, a new fd will be allocated)
943 * @file: struct file that was received from another process
944 * @ufd: __user pointer to write new fd number to
945 * @o_flags: the O_* flags to apply to the new fd entry
947 * Installs a received file into the file descriptor table, with appropriate
948 * checks and count updates. Optionally writes the fd number to userspace, if
951 * This helper handles its own reference counting of the incoming
954 * Returns newly install fd or -ve on error.
956 int __receive_fd(int fd, struct file *file, int __user *ufd, unsigned int o_flags)
961 error = security_file_receive(file);
966 new_fd = get_unused_fd_flags(o_flags);
974 error = put_user(new_fd, ufd);
977 put_unused_fd(new_fd);
983 fd_install(new_fd, get_file(file));
985 error = replace_fd(new_fd, file, o_flags);
990 /* Bump the sock usage counts, if any. */
991 __receive_sock(file);
995 static int ksys_dup3(unsigned int oldfd, unsigned int newfd, int flags)
999 struct files_struct *files = current->files;
1001 if ((flags & ~O_CLOEXEC) != 0)
1004 if (unlikely(oldfd == newfd))
1007 if (newfd >= rlimit(RLIMIT_NOFILE))
1010 spin_lock(&files->file_lock);
1011 err = expand_files(files, newfd);
1012 file = fcheck(oldfd);
1013 if (unlikely(!file))
1015 if (unlikely(err < 0)) {
1020 return do_dup2(files, file, newfd, flags);
1025 spin_unlock(&files->file_lock);
1029 SYSCALL_DEFINE3(dup3, unsigned int, oldfd, unsigned int, newfd, int, flags)
1031 return ksys_dup3(oldfd, newfd, flags);
1034 SYSCALL_DEFINE2(dup2, unsigned int, oldfd, unsigned int, newfd)
1036 if (unlikely(newfd == oldfd)) { /* corner case */
1037 struct files_struct *files = current->files;
1041 if (!fcheck_files(files, oldfd))
1046 return ksys_dup3(oldfd, newfd, 0);
1049 int ksys_dup(unsigned int fildes)
1052 struct file *file = fget_raw(fildes);
1055 ret = get_unused_fd_flags(0);
1057 fd_install(ret, file);
1064 SYSCALL_DEFINE1(dup, unsigned int, fildes)
1066 return ksys_dup(fildes);
1069 int f_dupfd(unsigned int from, struct file *file, unsigned flags)
1072 if (from >= rlimit(RLIMIT_NOFILE))
1074 err = alloc_fd(from, flags);
1077 fd_install(err, file);
1082 int iterate_fd(struct files_struct *files, unsigned n,
1083 int (*f)(const void *, struct file *, unsigned),
1086 struct fdtable *fdt;
1090 spin_lock(&files->file_lock);
1091 for (fdt = files_fdtable(files); n < fdt->max_fds; n++) {
1093 file = rcu_dereference_check_fdtable(files, fdt->fd[n]);
1096 res = f(p, file, n);
1100 spin_unlock(&files->file_lock);
1103 EXPORT_SYMBOL(iterate_fd);