1 // SPDX-License-Identifier: GPL-2.0
3 * linux/fs/read_write.c
5 * Copyright (C) 1991, 1992 Linus Torvalds
8 #include <linux/slab.h>
9 #include <linux/stat.h>
10 #include <linux/sched/xacct.h>
11 #include <linux/fcntl.h>
12 #include <linux/file.h>
13 #include <linux/uio.h>
14 #include <linux/fsnotify.h>
15 #include <linux/security.h>
16 #include <linux/export.h>
17 #include <linux/syscalls.h>
18 #include <linux/pagemap.h>
19 #include <linux/splice.h>
20 #include <linux/compat.h>
21 #include <linux/mount.h>
25 #include <linux/uaccess.h>
26 #include <asm/unistd.h>
28 const struct file_operations generic_ro_fops = {
29 .llseek = generic_file_llseek,
30 .read_iter = generic_file_read_iter,
31 .mmap = generic_file_readonly_mmap,
32 .splice_read = generic_file_splice_read,
35 EXPORT_SYMBOL(generic_ro_fops);
37 static inline bool unsigned_offsets(struct file *file)
39 return file->f_mode & FMODE_UNSIGNED_OFFSET;
43 * vfs_setpos - update the file offset for lseek
44 * @file: file structure in question
45 * @offset: file offset to seek to
46 * @maxsize: maximum file size
48 * This is a low-level filesystem helper for updating the file offset to
49 * the value specified by @offset if the given offset is valid and it is
50 * not equal to the current file offset.
52 * Return the specified offset on success and -EINVAL on invalid offset.
54 loff_t vfs_setpos(struct file *file, loff_t offset, loff_t maxsize)
56 if (offset < 0 && !unsigned_offsets(file))
61 if (offset != file->f_pos) {
67 EXPORT_SYMBOL(vfs_setpos);
70 * generic_file_llseek_size - generic llseek implementation for regular files
71 * @file: file structure to seek on
72 * @offset: file offset to seek to
73 * @whence: type of seek
74 * @size: max size of this file in file system
75 * @eof: offset used for SEEK_END position
77 * This is a variant of generic_file_llseek that allows passing in a custom
78 * maximum file size and a custom EOF position, for e.g. hashed directories
81 * SEEK_SET and SEEK_END are unsynchronized (but atomic on 64bit platforms)
82 * SEEK_CUR is synchronized against other SEEK_CURs, but not read/writes.
83 * read/writes behave like SEEK_SET against seeks.
86 generic_file_llseek_size(struct file *file, loff_t offset, int whence,
87 loff_t maxsize, loff_t eof)
95 * Here we special-case the lseek(fd, 0, SEEK_CUR)
96 * position-querying operation. Avoid rewriting the "same"
97 * f_pos value back to the file because a concurrent read(),
98 * write() or lseek() might have altered it
103 * f_lock protects against read/modify/write race with other
104 * SEEK_CURs. Note that parallel writes and reads behave
107 spin_lock(&file->f_lock);
108 offset = vfs_setpos(file, file->f_pos + offset, maxsize);
109 spin_unlock(&file->f_lock);
113 * In the generic case the entire file is data, so as long as
114 * offset isn't at the end of the file then the offset is data.
116 if ((unsigned long long)offset >= eof)
121 * There is a virtual hole at the end of the file, so as long as
122 * offset isn't i_size or larger, return i_size.
124 if ((unsigned long long)offset >= eof)
130 return vfs_setpos(file, offset, maxsize);
132 EXPORT_SYMBOL(generic_file_llseek_size);
135 * generic_file_llseek - generic llseek implementation for regular files
136 * @file: file structure to seek on
137 * @offset: file offset to seek to
138 * @whence: type of seek
140 * This is a generic implemenation of ->llseek useable for all normal local
141 * filesystems. It just updates the file offset to the value specified by
142 * @offset and @whence.
144 loff_t generic_file_llseek(struct file *file, loff_t offset, int whence)
146 struct inode *inode = file->f_mapping->host;
148 return generic_file_llseek_size(file, offset, whence,
149 inode->i_sb->s_maxbytes,
152 EXPORT_SYMBOL(generic_file_llseek);
155 * fixed_size_llseek - llseek implementation for fixed-sized devices
156 * @file: file structure to seek on
157 * @offset: file offset to seek to
158 * @whence: type of seek
159 * @size: size of the file
162 loff_t fixed_size_llseek(struct file *file, loff_t offset, int whence, loff_t size)
165 case SEEK_SET: case SEEK_CUR: case SEEK_END:
166 return generic_file_llseek_size(file, offset, whence,
172 EXPORT_SYMBOL(fixed_size_llseek);
175 * no_seek_end_llseek - llseek implementation for fixed-sized devices
176 * @file: file structure to seek on
177 * @offset: file offset to seek to
178 * @whence: type of seek
181 loff_t no_seek_end_llseek(struct file *file, loff_t offset, int whence)
184 case SEEK_SET: case SEEK_CUR:
185 return generic_file_llseek_size(file, offset, whence,
191 EXPORT_SYMBOL(no_seek_end_llseek);
194 * no_seek_end_llseek_size - llseek implementation for fixed-sized devices
195 * @file: file structure to seek on
196 * @offset: file offset to seek to
197 * @whence: type of seek
198 * @size: maximal offset allowed
201 loff_t no_seek_end_llseek_size(struct file *file, loff_t offset, int whence, loff_t size)
204 case SEEK_SET: case SEEK_CUR:
205 return generic_file_llseek_size(file, offset, whence,
211 EXPORT_SYMBOL(no_seek_end_llseek_size);
214 * noop_llseek - No Operation Performed llseek implementation
215 * @file: file structure to seek on
216 * @offset: file offset to seek to
217 * @whence: type of seek
219 * This is an implementation of ->llseek useable for the rare special case when
220 * userspace expects the seek to succeed but the (device) file is actually not
221 * able to perform the seek. In this case you use noop_llseek() instead of
222 * falling back to the default implementation of ->llseek.
224 loff_t noop_llseek(struct file *file, loff_t offset, int whence)
228 EXPORT_SYMBOL(noop_llseek);
230 loff_t no_llseek(struct file *file, loff_t offset, int whence)
234 EXPORT_SYMBOL(no_llseek);
236 loff_t default_llseek(struct file *file, loff_t offset, int whence)
238 struct inode *inode = file_inode(file);
244 offset += i_size_read(inode);
248 retval = file->f_pos;
251 offset += file->f_pos;
255 * In the generic case the entire file is data, so as
256 * long as offset isn't at the end of the file then the
259 if (offset >= inode->i_size) {
266 * There is a virtual hole at the end of the file, so
267 * as long as offset isn't i_size or larger, return
270 if (offset >= inode->i_size) {
274 offset = inode->i_size;
278 if (offset >= 0 || unsigned_offsets(file)) {
279 if (offset != file->f_pos) {
280 file->f_pos = offset;
289 EXPORT_SYMBOL(default_llseek);
291 loff_t vfs_llseek(struct file *file, loff_t offset, int whence)
293 loff_t (*fn)(struct file *, loff_t, int);
296 if (file->f_mode & FMODE_LSEEK) {
297 if (file->f_op->llseek)
298 fn = file->f_op->llseek;
300 return fn(file, offset, whence);
302 EXPORT_SYMBOL(vfs_llseek);
304 off_t ksys_lseek(unsigned int fd, off_t offset, unsigned int whence)
307 struct fd f = fdget_pos(fd);
312 if (whence <= SEEK_MAX) {
313 loff_t res = vfs_llseek(f.file, offset, whence);
315 if (res != (loff_t)retval)
316 retval = -EOVERFLOW; /* LFS: should only happen on 32 bit platforms */
322 SYSCALL_DEFINE3(lseek, unsigned int, fd, off_t, offset, unsigned int, whence)
324 return ksys_lseek(fd, offset, whence);
328 COMPAT_SYSCALL_DEFINE3(lseek, unsigned int, fd, compat_off_t, offset, unsigned int, whence)
330 return ksys_lseek(fd, offset, whence);
334 #ifdef __ARCH_WANT_SYS_LLSEEK
335 SYSCALL_DEFINE5(llseek, unsigned int, fd, unsigned long, offset_high,
336 unsigned long, offset_low, loff_t __user *, result,
337 unsigned int, whence)
340 struct fd f = fdget_pos(fd);
347 if (whence > SEEK_MAX)
350 offset = vfs_llseek(f.file, ((loff_t) offset_high << 32) | offset_low,
353 retval = (int)offset;
356 if (!copy_to_user(result, &offset, sizeof(offset)))
365 int rw_verify_area(int read_write, struct file *file, const loff_t *ppos, size_t count)
369 int retval = -EINVAL;
371 inode = file_inode(file);
372 if (unlikely((ssize_t) count < 0))
375 if (unlikely(pos < 0)) {
376 if (!unsigned_offsets(file))
378 if (count >= -pos) /* both values are in 0..LLONG_MAX */
380 } else if (unlikely((loff_t) (pos + count) < 0)) {
381 if (!unsigned_offsets(file))
385 if (unlikely(inode->i_flctx && mandatory_lock(inode))) {
386 retval = locks_mandatory_area(inode, file, pos, pos + count - 1,
387 read_write == READ ? F_RDLCK : F_WRLCK);
391 return security_file_permission(file,
392 read_write == READ ? MAY_READ : MAY_WRITE);
395 static ssize_t new_sync_read(struct file *filp, char __user *buf, size_t len, loff_t *ppos)
397 struct iovec iov = { .iov_base = buf, .iov_len = len };
399 struct iov_iter iter;
402 init_sync_kiocb(&kiocb, filp);
403 kiocb.ki_pos = *ppos;
404 iov_iter_init(&iter, READ, &iov, 1, len);
406 ret = call_read_iter(filp, &kiocb, &iter);
407 BUG_ON(ret == -EIOCBQUEUED);
408 *ppos = kiocb.ki_pos;
412 ssize_t __vfs_read(struct file *file, char __user *buf, size_t count,
415 if (file->f_op->read)
416 return file->f_op->read(file, buf, count, pos);
417 else if (file->f_op->read_iter)
418 return new_sync_read(file, buf, count, pos);
423 ssize_t kernel_read(struct file *file, void *buf, size_t count, loff_t *pos)
430 /* The cast to a user pointer is valid due to the set_fs() */
431 result = vfs_read(file, (void __user *)buf, count, pos);
435 EXPORT_SYMBOL(kernel_read);
437 ssize_t vfs_read(struct file *file, char __user *buf, size_t count, loff_t *pos)
441 if (!(file->f_mode & FMODE_READ))
443 if (!(file->f_mode & FMODE_CAN_READ))
445 if (unlikely(!access_ok(VERIFY_WRITE, buf, count)))
448 ret = rw_verify_area(READ, file, pos, count);
450 if (count > MAX_RW_COUNT)
451 count = MAX_RW_COUNT;
452 ret = __vfs_read(file, buf, count, pos);
454 fsnotify_access(file);
455 add_rchar(current, ret);
463 static ssize_t new_sync_write(struct file *filp, const char __user *buf, size_t len, loff_t *ppos)
465 struct iovec iov = { .iov_base = (void __user *)buf, .iov_len = len };
467 struct iov_iter iter;
470 init_sync_kiocb(&kiocb, filp);
471 kiocb.ki_pos = *ppos;
472 iov_iter_init(&iter, WRITE, &iov, 1, len);
474 ret = call_write_iter(filp, &kiocb, &iter);
475 BUG_ON(ret == -EIOCBQUEUED);
477 *ppos = kiocb.ki_pos;
481 ssize_t __vfs_write(struct file *file, const char __user *p, size_t count,
484 if (file->f_op->write)
485 return file->f_op->write(file, p, count, pos);
486 else if (file->f_op->write_iter)
487 return new_sync_write(file, p, count, pos);
492 ssize_t __kernel_write(struct file *file, const void *buf, size_t count, loff_t *pos)
495 const char __user *p;
498 if (!(file->f_mode & FMODE_CAN_WRITE))
503 p = (__force const char __user *)buf;
504 if (count > MAX_RW_COUNT)
505 count = MAX_RW_COUNT;
506 ret = __vfs_write(file, p, count, pos);
509 fsnotify_modify(file);
510 add_wchar(current, ret);
515 EXPORT_SYMBOL(__kernel_write);
517 ssize_t kernel_write(struct file *file, const void *buf, size_t count,
525 /* The cast to a user pointer is valid due to the set_fs() */
526 res = vfs_write(file, (__force const char __user *)buf, count, pos);
531 EXPORT_SYMBOL(kernel_write);
533 ssize_t vfs_write(struct file *file, const char __user *buf, size_t count, loff_t *pos)
537 if (!(file->f_mode & FMODE_WRITE))
539 if (!(file->f_mode & FMODE_CAN_WRITE))
541 if (unlikely(!access_ok(VERIFY_READ, buf, count)))
544 ret = rw_verify_area(WRITE, file, pos, count);
546 if (count > MAX_RW_COUNT)
547 count = MAX_RW_COUNT;
548 file_start_write(file);
549 ret = __vfs_write(file, buf, count, pos);
551 fsnotify_modify(file);
552 add_wchar(current, ret);
555 file_end_write(file);
561 static inline loff_t file_pos_read(struct file *file)
563 return file->f_mode & FMODE_STREAM ? 0 : file->f_pos;
566 static inline void file_pos_write(struct file *file, loff_t pos)
568 if ((file->f_mode & FMODE_STREAM) == 0)
572 ssize_t ksys_read(unsigned int fd, char __user *buf, size_t count)
574 struct fd f = fdget_pos(fd);
575 ssize_t ret = -EBADF;
578 loff_t pos = file_pos_read(f.file);
579 ret = vfs_read(f.file, buf, count, &pos);
581 file_pos_write(f.file, pos);
587 SYSCALL_DEFINE3(read, unsigned int, fd, char __user *, buf, size_t, count)
589 return ksys_read(fd, buf, count);
592 ssize_t ksys_write(unsigned int fd, const char __user *buf, size_t count)
594 struct fd f = fdget_pos(fd);
595 ssize_t ret = -EBADF;
598 loff_t pos = file_pos_read(f.file);
599 ret = vfs_write(f.file, buf, count, &pos);
601 file_pos_write(f.file, pos);
608 SYSCALL_DEFINE3(write, unsigned int, fd, const char __user *, buf,
611 return ksys_write(fd, buf, count);
614 ssize_t ksys_pread64(unsigned int fd, char __user *buf, size_t count,
618 ssize_t ret = -EBADF;
626 if (f.file->f_mode & FMODE_PREAD)
627 ret = vfs_read(f.file, buf, count, &pos);
634 SYSCALL_DEFINE4(pread64, unsigned int, fd, char __user *, buf,
635 size_t, count, loff_t, pos)
637 return ksys_pread64(fd, buf, count, pos);
640 ssize_t ksys_pwrite64(unsigned int fd, const char __user *buf,
641 size_t count, loff_t pos)
644 ssize_t ret = -EBADF;
652 if (f.file->f_mode & FMODE_PWRITE)
653 ret = vfs_write(f.file, buf, count, &pos);
660 SYSCALL_DEFINE4(pwrite64, unsigned int, fd, const char __user *, buf,
661 size_t, count, loff_t, pos)
663 return ksys_pwrite64(fd, buf, count, pos);
666 static ssize_t do_iter_readv_writev(struct file *filp, struct iov_iter *iter,
667 loff_t *ppos, int type, rwf_t flags)
672 init_sync_kiocb(&kiocb, filp);
673 ret = kiocb_set_rw_flags(&kiocb, flags);
676 kiocb.ki_pos = *ppos;
679 ret = call_read_iter(filp, &kiocb, iter);
681 ret = call_write_iter(filp, &kiocb, iter);
682 BUG_ON(ret == -EIOCBQUEUED);
683 *ppos = kiocb.ki_pos;
687 /* Do it by hand, with file-ops */
688 static ssize_t do_loop_readv_writev(struct file *filp, struct iov_iter *iter,
689 loff_t *ppos, int type, rwf_t flags)
693 if (flags & ~RWF_HIPRI)
696 while (iov_iter_count(iter)) {
697 struct iovec iovec = iov_iter_iovec(iter);
701 nr = filp->f_op->read(filp, iovec.iov_base,
702 iovec.iov_len, ppos);
704 nr = filp->f_op->write(filp, iovec.iov_base,
705 iovec.iov_len, ppos);
714 if (nr != iovec.iov_len)
716 iov_iter_advance(iter, nr);
722 /* A write operation does a read from user space and vice versa */
723 #define vrfy_dir(type) ((type) == READ ? VERIFY_WRITE : VERIFY_READ)
726 * rw_copy_check_uvector() - Copy an array of &struct iovec from userspace
727 * into the kernel and check that it is valid.
729 * @type: One of %CHECK_IOVEC_ONLY, %READ, or %WRITE.
730 * @uvector: Pointer to the userspace array.
731 * @nr_segs: Number of elements in userspace array.
732 * @fast_segs: Number of elements in @fast_pointer.
733 * @fast_pointer: Pointer to (usually small on-stack) kernel array.
734 * @ret_pointer: (output parameter) Pointer to a variable that will point to
735 * either @fast_pointer, a newly allocated kernel array, or NULL,
736 * depending on which array was used.
738 * This function copies an array of &struct iovec of @nr_segs from
739 * userspace into the kernel and checks that each element is valid (e.g.
740 * it does not point to a kernel address or cause overflow by being too
743 * As an optimization, the caller may provide a pointer to a small
744 * on-stack array in @fast_pointer, typically %UIO_FASTIOV elements long
745 * (the size of this array, or 0 if unused, should be given in @fast_segs).
747 * @ret_pointer will always point to the array that was used, so the
748 * caller must take care not to call kfree() on it e.g. in case the
749 * @fast_pointer array was used and it was allocated on the stack.
751 * Return: The total number of bytes covered by the iovec array on success
752 * or a negative error code on error.
754 ssize_t rw_copy_check_uvector(int type, const struct iovec __user * uvector,
755 unsigned long nr_segs, unsigned long fast_segs,
756 struct iovec *fast_pointer,
757 struct iovec **ret_pointer)
761 struct iovec *iov = fast_pointer;
764 * SuS says "The readv() function *may* fail if the iovcnt argument
765 * was less than or equal to 0, or greater than {IOV_MAX}. Linux has
766 * traditionally returned zero for zero segments, so...
774 * First get the "struct iovec" from user memory and
775 * verify all the pointers
777 if (nr_segs > UIO_MAXIOV) {
781 if (nr_segs > fast_segs) {
782 iov = kmalloc_array(nr_segs, sizeof(struct iovec), GFP_KERNEL);
788 if (copy_from_user(iov, uvector, nr_segs*sizeof(*uvector))) {
794 * According to the Single Unix Specification we should return EINVAL
795 * if an element length is < 0 when cast to ssize_t or if the
796 * total length would overflow the ssize_t return value of the
799 * Linux caps all read/write calls to MAX_RW_COUNT, and avoids the
803 for (seg = 0; seg < nr_segs; seg++) {
804 void __user *buf = iov[seg].iov_base;
805 ssize_t len = (ssize_t)iov[seg].iov_len;
807 /* see if we we're about to use an invalid len or if
808 * it's about to overflow ssize_t */
814 && unlikely(!access_ok(vrfy_dir(type), buf, len))) {
818 if (len > MAX_RW_COUNT - ret) {
819 len = MAX_RW_COUNT - ret;
820 iov[seg].iov_len = len;
830 ssize_t compat_rw_copy_check_uvector(int type,
831 const struct compat_iovec __user *uvector, unsigned long nr_segs,
832 unsigned long fast_segs, struct iovec *fast_pointer,
833 struct iovec **ret_pointer)
835 compat_ssize_t tot_len;
836 struct iovec *iov = *ret_pointer = fast_pointer;
841 * SuS says "The readv() function *may* fail if the iovcnt argument
842 * was less than or equal to 0, or greater than {IOV_MAX}. Linux has
843 * traditionally returned zero for zero segments, so...
849 if (nr_segs > UIO_MAXIOV)
851 if (nr_segs > fast_segs) {
853 iov = kmalloc_array(nr_segs, sizeof(struct iovec), GFP_KERNEL);
860 if (!access_ok(VERIFY_READ, uvector, nr_segs*sizeof(*uvector)))
864 * Single unix specification:
865 * We should -EINVAL if an element length is not >= 0 and fitting an
868 * In Linux, the total length is limited to MAX_RW_COUNT, there is
869 * no overflow possibility.
873 for (seg = 0; seg < nr_segs; seg++) {
877 if (__get_user(len, &uvector->iov_len) ||
878 __get_user(buf, &uvector->iov_base)) {
882 if (len < 0) /* size_t not fitting in compat_ssize_t .. */
885 !access_ok(vrfy_dir(type), compat_ptr(buf), len)) {
889 if (len > MAX_RW_COUNT - tot_len)
890 len = MAX_RW_COUNT - tot_len;
892 iov->iov_base = compat_ptr(buf);
893 iov->iov_len = (compat_size_t) len;
904 static ssize_t do_iter_read(struct file *file, struct iov_iter *iter,
905 loff_t *pos, rwf_t flags)
910 if (!(file->f_mode & FMODE_READ))
912 if (!(file->f_mode & FMODE_CAN_READ))
915 tot_len = iov_iter_count(iter);
918 ret = rw_verify_area(READ, file, pos, tot_len);
922 if (file->f_op->read_iter)
923 ret = do_iter_readv_writev(file, iter, pos, READ, flags);
925 ret = do_loop_readv_writev(file, iter, pos, READ, flags);
928 fsnotify_access(file);
932 ssize_t vfs_iter_read(struct file *file, struct iov_iter *iter, loff_t *ppos,
935 if (!file->f_op->read_iter)
937 return do_iter_read(file, iter, ppos, flags);
939 EXPORT_SYMBOL(vfs_iter_read);
941 static ssize_t do_iter_write(struct file *file, struct iov_iter *iter,
942 loff_t *pos, rwf_t flags)
947 if (!(file->f_mode & FMODE_WRITE))
949 if (!(file->f_mode & FMODE_CAN_WRITE))
952 tot_len = iov_iter_count(iter);
955 ret = rw_verify_area(WRITE, file, pos, tot_len);
959 if (file->f_op->write_iter)
960 ret = do_iter_readv_writev(file, iter, pos, WRITE, flags);
962 ret = do_loop_readv_writev(file, iter, pos, WRITE, flags);
964 fsnotify_modify(file);
968 ssize_t vfs_iter_write(struct file *file, struct iov_iter *iter, loff_t *ppos,
971 if (!file->f_op->write_iter)
973 return do_iter_write(file, iter, ppos, flags);
975 EXPORT_SYMBOL(vfs_iter_write);
977 ssize_t vfs_readv(struct file *file, const struct iovec __user *vec,
978 unsigned long vlen, loff_t *pos, rwf_t flags)
980 struct iovec iovstack[UIO_FASTIOV];
981 struct iovec *iov = iovstack;
982 struct iov_iter iter;
985 ret = import_iovec(READ, vec, vlen, ARRAY_SIZE(iovstack), &iov, &iter);
987 ret = do_iter_read(file, &iter, pos, flags);
994 static ssize_t vfs_writev(struct file *file, const struct iovec __user *vec,
995 unsigned long vlen, loff_t *pos, rwf_t flags)
997 struct iovec iovstack[UIO_FASTIOV];
998 struct iovec *iov = iovstack;
999 struct iov_iter iter;
1002 ret = import_iovec(WRITE, vec, vlen, ARRAY_SIZE(iovstack), &iov, &iter);
1004 file_start_write(file);
1005 ret = do_iter_write(file, &iter, pos, flags);
1006 file_end_write(file);
1012 static ssize_t do_readv(unsigned long fd, const struct iovec __user *vec,
1013 unsigned long vlen, rwf_t flags)
1015 struct fd f = fdget_pos(fd);
1016 ssize_t ret = -EBADF;
1019 loff_t pos = file_pos_read(f.file);
1020 ret = vfs_readv(f.file, vec, vlen, &pos, flags);
1022 file_pos_write(f.file, pos);
1027 add_rchar(current, ret);
1032 static ssize_t do_writev(unsigned long fd, const struct iovec __user *vec,
1033 unsigned long vlen, rwf_t flags)
1035 struct fd f = fdget_pos(fd);
1036 ssize_t ret = -EBADF;
1039 loff_t pos = file_pos_read(f.file);
1040 ret = vfs_writev(f.file, vec, vlen, &pos, flags);
1042 file_pos_write(f.file, pos);
1047 add_wchar(current, ret);
1052 static inline loff_t pos_from_hilo(unsigned long high, unsigned long low)
1054 #define HALF_LONG_BITS (BITS_PER_LONG / 2)
1055 return (((loff_t)high << HALF_LONG_BITS) << HALF_LONG_BITS) | low;
1058 static ssize_t do_preadv(unsigned long fd, const struct iovec __user *vec,
1059 unsigned long vlen, loff_t pos, rwf_t flags)
1062 ssize_t ret = -EBADF;
1070 if (f.file->f_mode & FMODE_PREAD)
1071 ret = vfs_readv(f.file, vec, vlen, &pos, flags);
1076 add_rchar(current, ret);
1081 static ssize_t do_pwritev(unsigned long fd, const struct iovec __user *vec,
1082 unsigned long vlen, loff_t pos, rwf_t flags)
1085 ssize_t ret = -EBADF;
1093 if (f.file->f_mode & FMODE_PWRITE)
1094 ret = vfs_writev(f.file, vec, vlen, &pos, flags);
1099 add_wchar(current, ret);
1104 SYSCALL_DEFINE3(readv, unsigned long, fd, const struct iovec __user *, vec,
1105 unsigned long, vlen)
1107 return do_readv(fd, vec, vlen, 0);
1110 SYSCALL_DEFINE3(writev, unsigned long, fd, const struct iovec __user *, vec,
1111 unsigned long, vlen)
1113 return do_writev(fd, vec, vlen, 0);
1116 SYSCALL_DEFINE5(preadv, unsigned long, fd, const struct iovec __user *, vec,
1117 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
1119 loff_t pos = pos_from_hilo(pos_h, pos_l);
1121 return do_preadv(fd, vec, vlen, pos, 0);
1124 SYSCALL_DEFINE6(preadv2, unsigned long, fd, const struct iovec __user *, vec,
1125 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h,
1128 loff_t pos = pos_from_hilo(pos_h, pos_l);
1131 return do_readv(fd, vec, vlen, flags);
1133 return do_preadv(fd, vec, vlen, pos, flags);
1136 SYSCALL_DEFINE5(pwritev, unsigned long, fd, const struct iovec __user *, vec,
1137 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
1139 loff_t pos = pos_from_hilo(pos_h, pos_l);
1141 return do_pwritev(fd, vec, vlen, pos, 0);
1144 SYSCALL_DEFINE6(pwritev2, unsigned long, fd, const struct iovec __user *, vec,
1145 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h,
1148 loff_t pos = pos_from_hilo(pos_h, pos_l);
1151 return do_writev(fd, vec, vlen, flags);
1153 return do_pwritev(fd, vec, vlen, pos, flags);
1156 #ifdef CONFIG_COMPAT
1157 static size_t compat_readv(struct file *file,
1158 const struct compat_iovec __user *vec,
1159 unsigned long vlen, loff_t *pos, rwf_t flags)
1161 struct iovec iovstack[UIO_FASTIOV];
1162 struct iovec *iov = iovstack;
1163 struct iov_iter iter;
1166 ret = compat_import_iovec(READ, vec, vlen, UIO_FASTIOV, &iov, &iter);
1168 ret = do_iter_read(file, &iter, pos, flags);
1172 add_rchar(current, ret);
1177 static size_t do_compat_readv(compat_ulong_t fd,
1178 const struct compat_iovec __user *vec,
1179 compat_ulong_t vlen, rwf_t flags)
1181 struct fd f = fdget_pos(fd);
1187 pos = f.file->f_pos;
1188 ret = compat_readv(f.file, vec, vlen, &pos, flags);
1190 f.file->f_pos = pos;
1196 COMPAT_SYSCALL_DEFINE3(readv, compat_ulong_t, fd,
1197 const struct compat_iovec __user *,vec,
1198 compat_ulong_t, vlen)
1200 return do_compat_readv(fd, vec, vlen, 0);
1203 static long do_compat_preadv64(unsigned long fd,
1204 const struct compat_iovec __user *vec,
1205 unsigned long vlen, loff_t pos, rwf_t flags)
1216 if (f.file->f_mode & FMODE_PREAD)
1217 ret = compat_readv(f.file, vec, vlen, &pos, flags);
1222 #ifdef __ARCH_WANT_COMPAT_SYS_PREADV64
1223 COMPAT_SYSCALL_DEFINE4(preadv64, unsigned long, fd,
1224 const struct compat_iovec __user *,vec,
1225 unsigned long, vlen, loff_t, pos)
1227 return do_compat_preadv64(fd, vec, vlen, pos, 0);
1231 COMPAT_SYSCALL_DEFINE5(preadv, compat_ulong_t, fd,
1232 const struct compat_iovec __user *,vec,
1233 compat_ulong_t, vlen, u32, pos_low, u32, pos_high)
1235 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1237 return do_compat_preadv64(fd, vec, vlen, pos, 0);
1240 #ifdef __ARCH_WANT_COMPAT_SYS_PREADV64V2
1241 COMPAT_SYSCALL_DEFINE5(preadv64v2, unsigned long, fd,
1242 const struct compat_iovec __user *,vec,
1243 unsigned long, vlen, loff_t, pos, rwf_t, flags)
1246 return do_compat_readv(fd, vec, vlen, flags);
1248 return do_compat_preadv64(fd, vec, vlen, pos, flags);
1252 COMPAT_SYSCALL_DEFINE6(preadv2, compat_ulong_t, fd,
1253 const struct compat_iovec __user *,vec,
1254 compat_ulong_t, vlen, u32, pos_low, u32, pos_high,
1257 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1260 return do_compat_readv(fd, vec, vlen, flags);
1262 return do_compat_preadv64(fd, vec, vlen, pos, flags);
1265 static size_t compat_writev(struct file *file,
1266 const struct compat_iovec __user *vec,
1267 unsigned long vlen, loff_t *pos, rwf_t flags)
1269 struct iovec iovstack[UIO_FASTIOV];
1270 struct iovec *iov = iovstack;
1271 struct iov_iter iter;
1274 ret = compat_import_iovec(WRITE, vec, vlen, UIO_FASTIOV, &iov, &iter);
1276 file_start_write(file);
1277 ret = do_iter_write(file, &iter, pos, flags);
1278 file_end_write(file);
1282 add_wchar(current, ret);
1287 static size_t do_compat_writev(compat_ulong_t fd,
1288 const struct compat_iovec __user* vec,
1289 compat_ulong_t vlen, rwf_t flags)
1291 struct fd f = fdget_pos(fd);
1297 pos = f.file->f_pos;
1298 ret = compat_writev(f.file, vec, vlen, &pos, flags);
1300 f.file->f_pos = pos;
1305 COMPAT_SYSCALL_DEFINE3(writev, compat_ulong_t, fd,
1306 const struct compat_iovec __user *, vec,
1307 compat_ulong_t, vlen)
1309 return do_compat_writev(fd, vec, vlen, 0);
1312 static long do_compat_pwritev64(unsigned long fd,
1313 const struct compat_iovec __user *vec,
1314 unsigned long vlen, loff_t pos, rwf_t flags)
1325 if (f.file->f_mode & FMODE_PWRITE)
1326 ret = compat_writev(f.file, vec, vlen, &pos, flags);
1331 #ifdef __ARCH_WANT_COMPAT_SYS_PWRITEV64
1332 COMPAT_SYSCALL_DEFINE4(pwritev64, unsigned long, fd,
1333 const struct compat_iovec __user *,vec,
1334 unsigned long, vlen, loff_t, pos)
1336 return do_compat_pwritev64(fd, vec, vlen, pos, 0);
1340 COMPAT_SYSCALL_DEFINE5(pwritev, compat_ulong_t, fd,
1341 const struct compat_iovec __user *,vec,
1342 compat_ulong_t, vlen, u32, pos_low, u32, pos_high)
1344 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1346 return do_compat_pwritev64(fd, vec, vlen, pos, 0);
1349 #ifdef __ARCH_WANT_COMPAT_SYS_PWRITEV64V2
1350 COMPAT_SYSCALL_DEFINE5(pwritev64v2, unsigned long, fd,
1351 const struct compat_iovec __user *,vec,
1352 unsigned long, vlen, loff_t, pos, rwf_t, flags)
1355 return do_compat_writev(fd, vec, vlen, flags);
1357 return do_compat_pwritev64(fd, vec, vlen, pos, flags);
1361 COMPAT_SYSCALL_DEFINE6(pwritev2, compat_ulong_t, fd,
1362 const struct compat_iovec __user *,vec,
1363 compat_ulong_t, vlen, u32, pos_low, u32, pos_high, rwf_t, flags)
1365 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1368 return do_compat_writev(fd, vec, vlen, flags);
1370 return do_compat_pwritev64(fd, vec, vlen, pos, flags);
1375 static ssize_t do_sendfile(int out_fd, int in_fd, loff_t *ppos,
1376 size_t count, loff_t max)
1379 struct inode *in_inode, *out_inode;
1386 * Get input file, and verify that it is ok..
1392 if (!(in.file->f_mode & FMODE_READ))
1396 pos = in.file->f_pos;
1399 if (!(in.file->f_mode & FMODE_PREAD))
1402 retval = rw_verify_area(READ, in.file, &pos, count);
1405 if (count > MAX_RW_COUNT)
1406 count = MAX_RW_COUNT;
1409 * Get output file, and verify that it is ok..
1412 out = fdget(out_fd);
1415 if (!(out.file->f_mode & FMODE_WRITE))
1418 in_inode = file_inode(in.file);
1419 out_inode = file_inode(out.file);
1420 out_pos = out.file->f_pos;
1421 retval = rw_verify_area(WRITE, out.file, &out_pos, count);
1426 max = min(in_inode->i_sb->s_maxbytes, out_inode->i_sb->s_maxbytes);
1428 if (unlikely(pos + count > max)) {
1429 retval = -EOVERFLOW;
1438 * We need to debate whether we can enable this or not. The
1439 * man page documents EAGAIN return for the output at least,
1440 * and the application is arguably buggy if it doesn't expect
1441 * EAGAIN on a non-blocking file descriptor.
1443 if (in.file->f_flags & O_NONBLOCK)
1444 fl = SPLICE_F_NONBLOCK;
1446 file_start_write(out.file);
1447 retval = do_splice_direct(in.file, &pos, out.file, &out_pos, count, fl);
1448 file_end_write(out.file);
1451 add_rchar(current, retval);
1452 add_wchar(current, retval);
1453 fsnotify_access(in.file);
1454 fsnotify_modify(out.file);
1455 out.file->f_pos = out_pos;
1459 in.file->f_pos = pos;
1465 retval = -EOVERFLOW;
1475 SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd, off_t __user *, offset, size_t, count)
1482 if (unlikely(get_user(off, offset)))
1485 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
1486 if (unlikely(put_user(pos, offset)))
1491 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1494 SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd, loff_t __user *, offset, size_t, count)
1500 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
1502 ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
1503 if (unlikely(put_user(pos, offset)))
1508 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1511 #ifdef CONFIG_COMPAT
1512 COMPAT_SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd,
1513 compat_off_t __user *, offset, compat_size_t, count)
1520 if (unlikely(get_user(off, offset)))
1523 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
1524 if (unlikely(put_user(pos, offset)))
1529 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1532 COMPAT_SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd,
1533 compat_loff_t __user *, offset, compat_size_t, count)
1539 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
1541 ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
1542 if (unlikely(put_user(pos, offset)))
1547 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1552 * copy_file_range() differs from regular file read and write in that it
1553 * specifically allows return partial success. When it does so is up to
1554 * the copy_file_range method.
1556 ssize_t vfs_copy_file_range(struct file *file_in, loff_t pos_in,
1557 struct file *file_out, loff_t pos_out,
1558 size_t len, unsigned int flags)
1560 struct inode *inode_in = file_inode(file_in);
1561 struct inode *inode_out = file_inode(file_out);
1567 if (S_ISDIR(inode_in->i_mode) || S_ISDIR(inode_out->i_mode))
1569 if (!S_ISREG(inode_in->i_mode) || !S_ISREG(inode_out->i_mode))
1572 ret = rw_verify_area(READ, file_in, &pos_in, len);
1576 ret = rw_verify_area(WRITE, file_out, &pos_out, len);
1580 if (!(file_in->f_mode & FMODE_READ) ||
1581 !(file_out->f_mode & FMODE_WRITE) ||
1582 (file_out->f_flags & O_APPEND))
1585 /* this could be relaxed once a method supports cross-fs copies */
1586 if (inode_in->i_sb != inode_out->i_sb)
1592 file_start_write(file_out);
1595 * Try cloning first, this is supported by more file systems, and
1596 * more efficient if both clone and copy are supported (e.g. NFS).
1598 if (file_in->f_op->clone_file_range) {
1599 ret = file_in->f_op->clone_file_range(file_in, pos_in,
1600 file_out, pos_out, len);
1607 if (file_out->f_op->copy_file_range) {
1608 ret = file_out->f_op->copy_file_range(file_in, pos_in, file_out,
1609 pos_out, len, flags);
1610 if (ret != -EOPNOTSUPP)
1614 ret = do_splice_direct(file_in, &pos_in, file_out, &pos_out,
1615 len > MAX_RW_COUNT ? MAX_RW_COUNT : len, 0);
1619 fsnotify_access(file_in);
1620 add_rchar(current, ret);
1621 fsnotify_modify(file_out);
1622 add_wchar(current, ret);
1628 file_end_write(file_out);
1632 EXPORT_SYMBOL(vfs_copy_file_range);
1634 SYSCALL_DEFINE6(copy_file_range, int, fd_in, loff_t __user *, off_in,
1635 int, fd_out, loff_t __user *, off_out,
1636 size_t, len, unsigned int, flags)
1642 ssize_t ret = -EBADF;
1644 f_in = fdget(fd_in);
1648 f_out = fdget(fd_out);
1654 if (copy_from_user(&pos_in, off_in, sizeof(loff_t)))
1657 pos_in = f_in.file->f_pos;
1661 if (copy_from_user(&pos_out, off_out, sizeof(loff_t)))
1664 pos_out = f_out.file->f_pos;
1667 ret = vfs_copy_file_range(f_in.file, pos_in, f_out.file, pos_out, len,
1674 if (copy_to_user(off_in, &pos_in, sizeof(loff_t)))
1677 f_in.file->f_pos = pos_in;
1681 if (copy_to_user(off_out, &pos_out, sizeof(loff_t)))
1684 f_out.file->f_pos = pos_out;
1696 static int clone_verify_area(struct file *file, loff_t pos, u64 len, bool write)
1698 struct inode *inode = file_inode(file);
1700 if (unlikely(pos < 0))
1703 if (unlikely((loff_t) (pos + len) < 0))
1706 if (unlikely(inode->i_flctx && mandatory_lock(inode))) {
1707 loff_t end = len ? pos + len - 1 : OFFSET_MAX;
1710 retval = locks_mandatory_area(inode, file, pos, end,
1711 write ? F_WRLCK : F_RDLCK);
1716 return security_file_permission(file, write ? MAY_WRITE : MAY_READ);
1719 * Ensure that we don't remap a partial EOF block in the middle of something
1720 * else. Assume that the offsets have already been checked for block
1723 * For deduplication we always scale down to the previous block because we
1724 * can't meaningfully compare post-EOF contents.
1726 * For clone we only link a partial EOF block above the destination file's EOF.
1728 static int generic_remap_check_len(struct inode *inode_in,
1729 struct inode *inode_out,
1734 u64 blkmask = i_blocksize(inode_in) - 1;
1736 if ((*len & blkmask) == 0)
1741 else if (pos_out + *len < i_size_read(inode_out))
1748 * Check that the two inodes are eligible for cloning, the ranges make
1749 * sense, and then flush all dirty data. Caller must ensure that the
1750 * inodes have been locked against any other modifications.
1752 * Returns: 0 for "nothing to clone", 1 for "something to clone", or
1753 * the usual negative error code.
1755 int vfs_clone_file_prep_inodes(struct inode *inode_in, loff_t pos_in,
1756 struct inode *inode_out, loff_t pos_out,
1757 u64 *len, bool is_dedupe)
1759 loff_t bs = inode_out->i_sb->s_blocksize;
1762 bool same_inode = (inode_in == inode_out);
1765 /* Don't touch certain kinds of inodes */
1766 if (IS_IMMUTABLE(inode_out))
1769 if (IS_SWAPFILE(inode_in) || IS_SWAPFILE(inode_out))
1772 /* Don't reflink dirs, pipes, sockets... */
1773 if (S_ISDIR(inode_in->i_mode) || S_ISDIR(inode_out->i_mode))
1775 if (!S_ISREG(inode_in->i_mode) || !S_ISREG(inode_out->i_mode))
1778 /* Are we going all the way to the end? */
1779 isize = i_size_read(inode_in);
1783 /* Zero length dedupe exits immediately; reflink goes to EOF. */
1785 if (is_dedupe || pos_in == isize)
1789 *len = isize - pos_in;
1792 /* Ensure offsets don't wrap and the input is inside i_size */
1793 if (pos_in + *len < pos_in || pos_out + *len < pos_out ||
1794 pos_in + *len > isize)
1797 /* Don't allow dedupe past EOF in the dest file */
1801 disize = i_size_read(inode_out);
1802 if (pos_out >= disize || pos_out + *len > disize)
1806 /* If we're linking to EOF, continue to the block boundary. */
1807 if (pos_in + *len == isize)
1808 blen = ALIGN(isize, bs) - pos_in;
1812 /* Only reflink if we're aligned to block boundaries */
1813 if (!IS_ALIGNED(pos_in, bs) || !IS_ALIGNED(pos_in + blen, bs) ||
1814 !IS_ALIGNED(pos_out, bs) || !IS_ALIGNED(pos_out + blen, bs))
1817 /* Don't allow overlapped reflink within the same file */
1819 if (pos_out + blen > pos_in && pos_out < pos_in + blen)
1823 /* Wait for the completion of any pending IOs on both files */
1824 inode_dio_wait(inode_in);
1826 inode_dio_wait(inode_out);
1828 ret = filemap_write_and_wait_range(inode_in->i_mapping,
1829 pos_in, pos_in + *len - 1);
1833 ret = filemap_write_and_wait_range(inode_out->i_mapping,
1834 pos_out, pos_out + *len - 1);
1839 * Check that the extents are the same.
1842 bool is_same = false;
1844 ret = vfs_dedupe_file_range_compare(inode_in, pos_in,
1845 inode_out, pos_out, *len, &is_same);
1852 ret = generic_remap_check_len(inode_in, inode_out, pos_out, len,
1859 EXPORT_SYMBOL(vfs_clone_file_prep_inodes);
1861 int do_clone_file_range(struct file *file_in, loff_t pos_in,
1862 struct file *file_out, loff_t pos_out, u64 len)
1864 struct inode *inode_in = file_inode(file_in);
1865 struct inode *inode_out = file_inode(file_out);
1868 if (S_ISDIR(inode_in->i_mode) || S_ISDIR(inode_out->i_mode))
1870 if (!S_ISREG(inode_in->i_mode) || !S_ISREG(inode_out->i_mode))
1874 * FICLONE/FICLONERANGE ioctls enforce that src and dest files are on
1875 * the same mount. Practically, they only need to be on the same file
1878 if (inode_in->i_sb != inode_out->i_sb)
1881 if (!(file_in->f_mode & FMODE_READ) ||
1882 !(file_out->f_mode & FMODE_WRITE) ||
1883 (file_out->f_flags & O_APPEND))
1886 if (!file_in->f_op->clone_file_range)
1889 ret = clone_verify_area(file_in, pos_in, len, false);
1893 ret = clone_verify_area(file_out, pos_out, len, true);
1897 if (pos_in + len > i_size_read(inode_in))
1900 ret = file_in->f_op->clone_file_range(file_in, pos_in,
1901 file_out, pos_out, len);
1903 fsnotify_access(file_in);
1904 fsnotify_modify(file_out);
1909 EXPORT_SYMBOL(do_clone_file_range);
1911 int vfs_clone_file_range(struct file *file_in, loff_t pos_in,
1912 struct file *file_out, loff_t pos_out, u64 len)
1916 file_start_write(file_out);
1917 ret = do_clone_file_range(file_in, pos_in, file_out, pos_out, len);
1918 file_end_write(file_out);
1922 EXPORT_SYMBOL(vfs_clone_file_range);
1924 /* Read a page's worth of file data into the page cache. */
1925 static struct page *vfs_dedupe_get_page(struct inode *inode, loff_t offset)
1927 struct address_space *mapping;
1931 n = offset >> PAGE_SHIFT;
1932 mapping = inode->i_mapping;
1933 page = read_mapping_page(mapping, n, NULL);
1936 if (!PageUptodate(page)) {
1938 return ERR_PTR(-EIO);
1944 * Lock two pages, ensuring that we lock in offset order if the pages are from
1947 static void vfs_lock_two_pages(struct page *page1, struct page *page2)
1949 /* Always lock in order of increasing index. */
1950 if (page1->index > page2->index)
1958 /* Unlock two pages, being careful not to unlock the same page twice. */
1959 static void vfs_unlock_two_pages(struct page *page1, struct page *page2)
1967 * Compare extents of two files to see if they are the same.
1968 * Caller must have locked both inodes to prevent write races.
1970 int vfs_dedupe_file_range_compare(struct inode *src, loff_t srcoff,
1971 struct inode *dest, loff_t destoff,
1972 loff_t len, bool *is_same)
1978 struct page *src_page;
1979 struct page *dest_page;
1987 src_poff = srcoff & (PAGE_SIZE - 1);
1988 dest_poff = destoff & (PAGE_SIZE - 1);
1989 cmp_len = min(PAGE_SIZE - src_poff,
1990 PAGE_SIZE - dest_poff);
1991 cmp_len = min(cmp_len, len);
1995 src_page = vfs_dedupe_get_page(src, srcoff);
1996 if (IS_ERR(src_page)) {
1997 error = PTR_ERR(src_page);
2000 dest_page = vfs_dedupe_get_page(dest, destoff);
2001 if (IS_ERR(dest_page)) {
2002 error = PTR_ERR(dest_page);
2007 vfs_lock_two_pages(src_page, dest_page);
2010 * Now that we've locked both pages, make sure they're still
2011 * mapped to the file data we're interested in. If not,
2012 * someone is invalidating pages on us and we lose.
2014 if (!PageUptodate(src_page) || !PageUptodate(dest_page) ||
2015 src_page->mapping != src->i_mapping ||
2016 dest_page->mapping != dest->i_mapping) {
2021 src_addr = kmap_atomic(src_page);
2022 dest_addr = kmap_atomic(dest_page);
2024 flush_dcache_page(src_page);
2025 flush_dcache_page(dest_page);
2027 if (memcmp(src_addr + src_poff, dest_addr + dest_poff, cmp_len))
2030 kunmap_atomic(dest_addr);
2031 kunmap_atomic(src_addr);
2033 vfs_unlock_two_pages(src_page, dest_page);
2034 put_page(dest_page);
2051 EXPORT_SYMBOL(vfs_dedupe_file_range_compare);
2053 int vfs_dedupe_file_range_one(struct file *src_file, loff_t src_pos,
2054 struct file *dst_file, loff_t dst_pos, u64 len)
2058 ret = mnt_want_write_file(dst_file);
2062 ret = clone_verify_area(dst_file, dst_pos, len, true);
2064 goto out_drop_write;
2067 if (!(capable(CAP_SYS_ADMIN) || (dst_file->f_mode & FMODE_WRITE)))
2068 goto out_drop_write;
2071 if (src_file->f_path.mnt != dst_file->f_path.mnt)
2072 goto out_drop_write;
2075 if (S_ISDIR(file_inode(dst_file)->i_mode))
2076 goto out_drop_write;
2079 if (!dst_file->f_op->dedupe_file_range)
2080 goto out_drop_write;
2082 ret = dst_file->f_op->dedupe_file_range(src_file, src_pos,
2083 dst_file, dst_pos, len);
2085 mnt_drop_write_file(dst_file);
2089 EXPORT_SYMBOL(vfs_dedupe_file_range_one);
2091 int vfs_dedupe_file_range(struct file *file, struct file_dedupe_range *same)
2093 struct file_dedupe_range_info *info;
2094 struct inode *src = file_inode(file);
2099 u16 count = same->dest_count;
2102 if (!(file->f_mode & FMODE_READ))
2105 if (same->reserved1 || same->reserved2)
2108 off = same->src_offset;
2109 len = same->src_length;
2112 if (S_ISDIR(src->i_mode))
2116 if (!S_ISREG(src->i_mode))
2119 ret = clone_verify_area(file, off, len, false);
2124 if (off + len > i_size_read(src))
2127 /* Arbitrary 1G limit on a single dedupe request, can be raised. */
2128 len = min_t(u64, len, 1 << 30);
2130 /* pre-format output fields to sane values */
2131 for (i = 0; i < count; i++) {
2132 same->info[i].bytes_deduped = 0ULL;
2133 same->info[i].status = FILE_DEDUPE_RANGE_SAME;
2136 for (i = 0, info = same->info; i < count; i++, info++) {
2137 struct fd dst_fd = fdget(info->dest_fd);
2138 struct file *dst_file = dst_fd.file;
2141 info->status = -EBADF;
2145 if (info->reserved) {
2146 info->status = -EINVAL;
2150 deduped = vfs_dedupe_file_range_one(file, off, dst_file,
2151 info->dest_offset, len);
2152 if (deduped == -EBADE)
2153 info->status = FILE_DEDUPE_RANGE_DIFFERS;
2154 else if (deduped < 0)
2155 info->status = deduped;
2157 info->bytes_deduped = len;
2162 if (fatal_signal_pending(current))
2169 EXPORT_SYMBOL(vfs_dedupe_file_range);