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 #if !defined(CONFIG_64BIT) || defined(CONFIG_COMPAT)
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)
368 int retval = -EINVAL;
370 inode = file_inode(file);
371 if (unlikely((ssize_t) count < 0))
375 * ranged mandatory locking does not apply to streams - it makes sense
376 * only for files where position has a meaning.
381 if (unlikely(pos < 0)) {
382 if (!unsigned_offsets(file))
384 if (count >= -pos) /* both values are in 0..LLONG_MAX */
386 } else if (unlikely((loff_t) (pos + count) < 0)) {
387 if (!unsigned_offsets(file))
391 if (unlikely(inode->i_flctx && mandatory_lock(inode))) {
392 retval = locks_mandatory_area(inode, file, pos, pos + count - 1,
393 read_write == READ ? F_RDLCK : F_WRLCK);
399 return security_file_permission(file,
400 read_write == READ ? MAY_READ : MAY_WRITE);
403 static ssize_t new_sync_read(struct file *filp, char __user *buf, size_t len, loff_t *ppos)
405 struct iovec iov = { .iov_base = buf, .iov_len = len };
407 struct iov_iter iter;
410 init_sync_kiocb(&kiocb, filp);
411 kiocb.ki_pos = (ppos ? *ppos : 0);
412 iov_iter_init(&iter, READ, &iov, 1, len);
414 ret = call_read_iter(filp, &kiocb, &iter);
415 BUG_ON(ret == -EIOCBQUEUED);
417 *ppos = kiocb.ki_pos;
421 ssize_t __vfs_read(struct file *file, char __user *buf, size_t count,
424 if (file->f_op->read)
425 return file->f_op->read(file, buf, count, pos);
426 else if (file->f_op->read_iter)
427 return new_sync_read(file, buf, count, pos);
432 ssize_t kernel_read(struct file *file, void *buf, size_t count, loff_t *pos)
439 /* The cast to a user pointer is valid due to the set_fs() */
440 result = vfs_read(file, (void __user *)buf, count, pos);
444 EXPORT_SYMBOL(kernel_read);
446 ssize_t vfs_read(struct file *file, char __user *buf, size_t count, loff_t *pos)
450 if (!(file->f_mode & FMODE_READ))
452 if (!(file->f_mode & FMODE_CAN_READ))
454 if (unlikely(!access_ok(buf, count)))
457 ret = rw_verify_area(READ, file, pos, count);
459 if (count > MAX_RW_COUNT)
460 count = MAX_RW_COUNT;
461 ret = __vfs_read(file, buf, count, pos);
463 fsnotify_access(file);
464 add_rchar(current, ret);
472 static ssize_t new_sync_write(struct file *filp, const char __user *buf, size_t len, loff_t *ppos)
474 struct iovec iov = { .iov_base = (void __user *)buf, .iov_len = len };
476 struct iov_iter iter;
479 init_sync_kiocb(&kiocb, filp);
480 kiocb.ki_pos = (ppos ? *ppos : 0);
481 iov_iter_init(&iter, WRITE, &iov, 1, len);
483 ret = call_write_iter(filp, &kiocb, &iter);
484 BUG_ON(ret == -EIOCBQUEUED);
486 *ppos = kiocb.ki_pos;
490 static ssize_t __vfs_write(struct file *file, const char __user *p,
491 size_t count, loff_t *pos)
493 if (file->f_op->write)
494 return file->f_op->write(file, p, count, pos);
495 else if (file->f_op->write_iter)
496 return new_sync_write(file, p, count, pos);
501 ssize_t __kernel_write(struct file *file, const void *buf, size_t count, loff_t *pos)
504 const char __user *p;
507 if (!(file->f_mode & FMODE_CAN_WRITE))
512 p = (__force const char __user *)buf;
513 if (count > MAX_RW_COUNT)
514 count = MAX_RW_COUNT;
515 ret = __vfs_write(file, p, count, pos);
518 fsnotify_modify(file);
519 add_wchar(current, ret);
524 EXPORT_SYMBOL(__kernel_write);
526 ssize_t kernel_write(struct file *file, const void *buf, size_t count,
534 /* The cast to a user pointer is valid due to the set_fs() */
535 res = vfs_write(file, (__force const char __user *)buf, count, pos);
540 EXPORT_SYMBOL(kernel_write);
542 ssize_t vfs_write(struct file *file, const char __user *buf, size_t count, loff_t *pos)
546 if (!(file->f_mode & FMODE_WRITE))
548 if (!(file->f_mode & FMODE_CAN_WRITE))
550 if (unlikely(!access_ok(buf, count)))
553 ret = rw_verify_area(WRITE, file, pos, count);
555 if (count > MAX_RW_COUNT)
556 count = MAX_RW_COUNT;
557 file_start_write(file);
558 ret = __vfs_write(file, buf, count, pos);
560 fsnotify_modify(file);
561 add_wchar(current, ret);
564 file_end_write(file);
570 /* file_ppos returns &file->f_pos or NULL if file is stream */
571 static inline loff_t *file_ppos(struct file *file)
573 return file->f_mode & FMODE_STREAM ? NULL : &file->f_pos;
576 ssize_t ksys_read(unsigned int fd, char __user *buf, size_t count)
578 struct fd f = fdget_pos(fd);
579 ssize_t ret = -EBADF;
582 loff_t pos, *ppos = file_ppos(f.file);
587 ret = vfs_read(f.file, buf, count, ppos);
588 if (ret >= 0 && ppos)
595 SYSCALL_DEFINE3(read, unsigned int, fd, char __user *, buf, size_t, count)
597 return ksys_read(fd, buf, count);
600 ssize_t ksys_write(unsigned int fd, const char __user *buf, size_t count)
602 struct fd f = fdget_pos(fd);
603 ssize_t ret = -EBADF;
606 loff_t pos, *ppos = file_ppos(f.file);
611 ret = vfs_write(f.file, buf, count, ppos);
612 if (ret >= 0 && ppos)
620 SYSCALL_DEFINE3(write, unsigned int, fd, const char __user *, buf,
623 return ksys_write(fd, buf, count);
626 ssize_t ksys_pread64(unsigned int fd, char __user *buf, size_t count,
630 ssize_t ret = -EBADF;
638 if (f.file->f_mode & FMODE_PREAD)
639 ret = vfs_read(f.file, buf, count, &pos);
646 SYSCALL_DEFINE4(pread64, unsigned int, fd, char __user *, buf,
647 size_t, count, loff_t, pos)
649 return ksys_pread64(fd, buf, count, pos);
652 ssize_t ksys_pwrite64(unsigned int fd, const char __user *buf,
653 size_t count, loff_t pos)
656 ssize_t ret = -EBADF;
664 if (f.file->f_mode & FMODE_PWRITE)
665 ret = vfs_write(f.file, buf, count, &pos);
672 SYSCALL_DEFINE4(pwrite64, unsigned int, fd, const char __user *, buf,
673 size_t, count, loff_t, pos)
675 return ksys_pwrite64(fd, buf, count, pos);
678 static ssize_t do_iter_readv_writev(struct file *filp, struct iov_iter *iter,
679 loff_t *ppos, int type, rwf_t flags)
684 init_sync_kiocb(&kiocb, filp);
685 ret = kiocb_set_rw_flags(&kiocb, flags);
688 kiocb.ki_pos = (ppos ? *ppos : 0);
691 ret = call_read_iter(filp, &kiocb, iter);
693 ret = call_write_iter(filp, &kiocb, iter);
694 BUG_ON(ret == -EIOCBQUEUED);
696 *ppos = kiocb.ki_pos;
700 /* Do it by hand, with file-ops */
701 static ssize_t do_loop_readv_writev(struct file *filp, struct iov_iter *iter,
702 loff_t *ppos, int type, rwf_t flags)
706 if (flags & ~RWF_HIPRI)
709 while (iov_iter_count(iter)) {
710 struct iovec iovec = iov_iter_iovec(iter);
714 nr = filp->f_op->read(filp, iovec.iov_base,
715 iovec.iov_len, ppos);
717 nr = filp->f_op->write(filp, iovec.iov_base,
718 iovec.iov_len, ppos);
727 if (nr != iovec.iov_len)
729 iov_iter_advance(iter, nr);
736 * rw_copy_check_uvector() - Copy an array of &struct iovec from userspace
737 * into the kernel and check that it is valid.
739 * @type: One of %CHECK_IOVEC_ONLY, %READ, or %WRITE.
740 * @uvector: Pointer to the userspace array.
741 * @nr_segs: Number of elements in userspace array.
742 * @fast_segs: Number of elements in @fast_pointer.
743 * @fast_pointer: Pointer to (usually small on-stack) kernel array.
744 * @ret_pointer: (output parameter) Pointer to a variable that will point to
745 * either @fast_pointer, a newly allocated kernel array, or NULL,
746 * depending on which array was used.
748 * This function copies an array of &struct iovec of @nr_segs from
749 * userspace into the kernel and checks that each element is valid (e.g.
750 * it does not point to a kernel address or cause overflow by being too
753 * As an optimization, the caller may provide a pointer to a small
754 * on-stack array in @fast_pointer, typically %UIO_FASTIOV elements long
755 * (the size of this array, or 0 if unused, should be given in @fast_segs).
757 * @ret_pointer will always point to the array that was used, so the
758 * caller must take care not to call kfree() on it e.g. in case the
759 * @fast_pointer array was used and it was allocated on the stack.
761 * Return: The total number of bytes covered by the iovec array on success
762 * or a negative error code on error.
764 ssize_t rw_copy_check_uvector(int type, const struct iovec __user * uvector,
765 unsigned long nr_segs, unsigned long fast_segs,
766 struct iovec *fast_pointer,
767 struct iovec **ret_pointer)
771 struct iovec *iov = fast_pointer;
774 * SuS says "The readv() function *may* fail if the iovcnt argument
775 * was less than or equal to 0, or greater than {IOV_MAX}. Linux has
776 * traditionally returned zero for zero segments, so...
784 * First get the "struct iovec" from user memory and
785 * verify all the pointers
787 if (nr_segs > UIO_MAXIOV) {
791 if (nr_segs > fast_segs) {
792 iov = kmalloc_array(nr_segs, sizeof(struct iovec), GFP_KERNEL);
798 if (copy_from_user(iov, uvector, nr_segs*sizeof(*uvector))) {
804 * According to the Single Unix Specification we should return EINVAL
805 * if an element length is < 0 when cast to ssize_t or if the
806 * total length would overflow the ssize_t return value of the
809 * Linux caps all read/write calls to MAX_RW_COUNT, and avoids the
813 for (seg = 0; seg < nr_segs; seg++) {
814 void __user *buf = iov[seg].iov_base;
815 ssize_t len = (ssize_t)iov[seg].iov_len;
817 /* see if we we're about to use an invalid len or if
818 * it's about to overflow ssize_t */
824 && unlikely(!access_ok(buf, len))) {
828 if (len > MAX_RW_COUNT - ret) {
829 len = MAX_RW_COUNT - ret;
830 iov[seg].iov_len = len;
840 ssize_t compat_rw_copy_check_uvector(int type,
841 const struct compat_iovec __user *uvector, unsigned long nr_segs,
842 unsigned long fast_segs, struct iovec *fast_pointer,
843 struct iovec **ret_pointer)
845 compat_ssize_t tot_len;
846 struct iovec *iov = *ret_pointer = fast_pointer;
851 * SuS says "The readv() function *may* fail if the iovcnt argument
852 * was less than or equal to 0, or greater than {IOV_MAX}. Linux has
853 * traditionally returned zero for zero segments, so...
859 if (nr_segs > UIO_MAXIOV)
861 if (nr_segs > fast_segs) {
863 iov = kmalloc_array(nr_segs, sizeof(struct iovec), GFP_KERNEL);
870 if (!access_ok(uvector, nr_segs*sizeof(*uvector)))
874 * Single unix specification:
875 * We should -EINVAL if an element length is not >= 0 and fitting an
878 * In Linux, the total length is limited to MAX_RW_COUNT, there is
879 * no overflow possibility.
883 for (seg = 0; seg < nr_segs; seg++) {
887 if (__get_user(len, &uvector->iov_len) ||
888 __get_user(buf, &uvector->iov_base)) {
892 if (len < 0) /* size_t not fitting in compat_ssize_t .. */
895 !access_ok(compat_ptr(buf), len)) {
899 if (len > MAX_RW_COUNT - tot_len)
900 len = MAX_RW_COUNT - tot_len;
902 iov->iov_base = compat_ptr(buf);
903 iov->iov_len = (compat_size_t) len;
914 static ssize_t do_iter_read(struct file *file, struct iov_iter *iter,
915 loff_t *pos, rwf_t flags)
920 if (!(file->f_mode & FMODE_READ))
922 if (!(file->f_mode & FMODE_CAN_READ))
925 tot_len = iov_iter_count(iter);
928 ret = rw_verify_area(READ, file, pos, tot_len);
932 if (file->f_op->read_iter)
933 ret = do_iter_readv_writev(file, iter, pos, READ, flags);
935 ret = do_loop_readv_writev(file, iter, pos, READ, flags);
938 fsnotify_access(file);
942 ssize_t vfs_iter_read(struct file *file, struct iov_iter *iter, loff_t *ppos,
945 if (!file->f_op->read_iter)
947 return do_iter_read(file, iter, ppos, flags);
949 EXPORT_SYMBOL(vfs_iter_read);
951 static ssize_t do_iter_write(struct file *file, struct iov_iter *iter,
952 loff_t *pos, rwf_t flags)
957 if (!(file->f_mode & FMODE_WRITE))
959 if (!(file->f_mode & FMODE_CAN_WRITE))
962 tot_len = iov_iter_count(iter);
965 ret = rw_verify_area(WRITE, file, pos, tot_len);
969 if (file->f_op->write_iter)
970 ret = do_iter_readv_writev(file, iter, pos, WRITE, flags);
972 ret = do_loop_readv_writev(file, iter, pos, WRITE, flags);
974 fsnotify_modify(file);
978 ssize_t vfs_iter_write(struct file *file, struct iov_iter *iter, loff_t *ppos,
981 if (!file->f_op->write_iter)
983 return do_iter_write(file, iter, ppos, flags);
985 EXPORT_SYMBOL(vfs_iter_write);
987 ssize_t vfs_readv(struct file *file, const struct iovec __user *vec,
988 unsigned long vlen, loff_t *pos, rwf_t flags)
990 struct iovec iovstack[UIO_FASTIOV];
991 struct iovec *iov = iovstack;
992 struct iov_iter iter;
995 ret = import_iovec(READ, vec, vlen, ARRAY_SIZE(iovstack), &iov, &iter);
997 ret = do_iter_read(file, &iter, pos, flags);
1004 static ssize_t vfs_writev(struct file *file, const struct iovec __user *vec,
1005 unsigned long vlen, loff_t *pos, rwf_t flags)
1007 struct iovec iovstack[UIO_FASTIOV];
1008 struct iovec *iov = iovstack;
1009 struct iov_iter iter;
1012 ret = import_iovec(WRITE, vec, vlen, ARRAY_SIZE(iovstack), &iov, &iter);
1014 file_start_write(file);
1015 ret = do_iter_write(file, &iter, pos, flags);
1016 file_end_write(file);
1022 static ssize_t do_readv(unsigned long fd, const struct iovec __user *vec,
1023 unsigned long vlen, rwf_t flags)
1025 struct fd f = fdget_pos(fd);
1026 ssize_t ret = -EBADF;
1029 loff_t pos, *ppos = file_ppos(f.file);
1034 ret = vfs_readv(f.file, vec, vlen, ppos, flags);
1035 if (ret >= 0 && ppos)
1036 f.file->f_pos = pos;
1041 add_rchar(current, ret);
1046 static ssize_t do_writev(unsigned long fd, const struct iovec __user *vec,
1047 unsigned long vlen, rwf_t flags)
1049 struct fd f = fdget_pos(fd);
1050 ssize_t ret = -EBADF;
1053 loff_t pos, *ppos = file_ppos(f.file);
1058 ret = vfs_writev(f.file, vec, vlen, ppos, flags);
1059 if (ret >= 0 && ppos)
1060 f.file->f_pos = pos;
1065 add_wchar(current, ret);
1070 static inline loff_t pos_from_hilo(unsigned long high, unsigned long low)
1072 #define HALF_LONG_BITS (BITS_PER_LONG / 2)
1073 return (((loff_t)high << HALF_LONG_BITS) << HALF_LONG_BITS) | low;
1076 static ssize_t do_preadv(unsigned long fd, const struct iovec __user *vec,
1077 unsigned long vlen, loff_t pos, rwf_t flags)
1080 ssize_t ret = -EBADF;
1088 if (f.file->f_mode & FMODE_PREAD)
1089 ret = vfs_readv(f.file, vec, vlen, &pos, flags);
1094 add_rchar(current, ret);
1099 static ssize_t do_pwritev(unsigned long fd, const struct iovec __user *vec,
1100 unsigned long vlen, loff_t pos, rwf_t flags)
1103 ssize_t ret = -EBADF;
1111 if (f.file->f_mode & FMODE_PWRITE)
1112 ret = vfs_writev(f.file, vec, vlen, &pos, flags);
1117 add_wchar(current, ret);
1122 SYSCALL_DEFINE3(readv, unsigned long, fd, const struct iovec __user *, vec,
1123 unsigned long, vlen)
1125 return do_readv(fd, vec, vlen, 0);
1128 SYSCALL_DEFINE3(writev, unsigned long, fd, const struct iovec __user *, vec,
1129 unsigned long, vlen)
1131 return do_writev(fd, vec, vlen, 0);
1134 SYSCALL_DEFINE5(preadv, unsigned long, fd, const struct iovec __user *, vec,
1135 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
1137 loff_t pos = pos_from_hilo(pos_h, pos_l);
1139 return do_preadv(fd, vec, vlen, pos, 0);
1142 SYSCALL_DEFINE6(preadv2, unsigned long, fd, const struct iovec __user *, vec,
1143 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h,
1146 loff_t pos = pos_from_hilo(pos_h, pos_l);
1149 return do_readv(fd, vec, vlen, flags);
1151 return do_preadv(fd, vec, vlen, pos, flags);
1154 SYSCALL_DEFINE5(pwritev, unsigned long, fd, const struct iovec __user *, vec,
1155 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
1157 loff_t pos = pos_from_hilo(pos_h, pos_l);
1159 return do_pwritev(fd, vec, vlen, pos, 0);
1162 SYSCALL_DEFINE6(pwritev2, unsigned long, fd, const struct iovec __user *, vec,
1163 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h,
1166 loff_t pos = pos_from_hilo(pos_h, pos_l);
1169 return do_writev(fd, vec, vlen, flags);
1171 return do_pwritev(fd, vec, vlen, pos, flags);
1174 #ifdef CONFIG_COMPAT
1175 static size_t compat_readv(struct file *file,
1176 const struct compat_iovec __user *vec,
1177 unsigned long vlen, loff_t *pos, rwf_t flags)
1179 struct iovec iovstack[UIO_FASTIOV];
1180 struct iovec *iov = iovstack;
1181 struct iov_iter iter;
1184 ret = compat_import_iovec(READ, vec, vlen, UIO_FASTIOV, &iov, &iter);
1186 ret = do_iter_read(file, &iter, pos, flags);
1190 add_rchar(current, ret);
1195 static size_t do_compat_readv(compat_ulong_t fd,
1196 const struct compat_iovec __user *vec,
1197 compat_ulong_t vlen, rwf_t flags)
1199 struct fd f = fdget_pos(fd);
1205 pos = f.file->f_pos;
1206 ret = compat_readv(f.file, vec, vlen, &pos, flags);
1208 f.file->f_pos = pos;
1214 COMPAT_SYSCALL_DEFINE3(readv, compat_ulong_t, fd,
1215 const struct compat_iovec __user *,vec,
1216 compat_ulong_t, vlen)
1218 return do_compat_readv(fd, vec, vlen, 0);
1221 static long do_compat_preadv64(unsigned long fd,
1222 const struct compat_iovec __user *vec,
1223 unsigned long vlen, loff_t pos, rwf_t flags)
1234 if (f.file->f_mode & FMODE_PREAD)
1235 ret = compat_readv(f.file, vec, vlen, &pos, flags);
1240 #ifdef __ARCH_WANT_COMPAT_SYS_PREADV64
1241 COMPAT_SYSCALL_DEFINE4(preadv64, unsigned long, fd,
1242 const struct compat_iovec __user *,vec,
1243 unsigned long, vlen, loff_t, pos)
1245 return do_compat_preadv64(fd, vec, vlen, pos, 0);
1249 COMPAT_SYSCALL_DEFINE5(preadv, compat_ulong_t, fd,
1250 const struct compat_iovec __user *,vec,
1251 compat_ulong_t, vlen, u32, pos_low, u32, pos_high)
1253 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1255 return do_compat_preadv64(fd, vec, vlen, pos, 0);
1258 #ifdef __ARCH_WANT_COMPAT_SYS_PREADV64V2
1259 COMPAT_SYSCALL_DEFINE5(preadv64v2, unsigned long, fd,
1260 const struct compat_iovec __user *,vec,
1261 unsigned long, vlen, loff_t, pos, rwf_t, flags)
1264 return do_compat_readv(fd, vec, vlen, flags);
1266 return do_compat_preadv64(fd, vec, vlen, pos, flags);
1270 COMPAT_SYSCALL_DEFINE6(preadv2, compat_ulong_t, fd,
1271 const struct compat_iovec __user *,vec,
1272 compat_ulong_t, vlen, u32, pos_low, u32, pos_high,
1275 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1278 return do_compat_readv(fd, vec, vlen, flags);
1280 return do_compat_preadv64(fd, vec, vlen, pos, flags);
1283 static size_t compat_writev(struct file *file,
1284 const struct compat_iovec __user *vec,
1285 unsigned long vlen, loff_t *pos, rwf_t flags)
1287 struct iovec iovstack[UIO_FASTIOV];
1288 struct iovec *iov = iovstack;
1289 struct iov_iter iter;
1292 ret = compat_import_iovec(WRITE, vec, vlen, UIO_FASTIOV, &iov, &iter);
1294 file_start_write(file);
1295 ret = do_iter_write(file, &iter, pos, flags);
1296 file_end_write(file);
1300 add_wchar(current, ret);
1305 static size_t do_compat_writev(compat_ulong_t fd,
1306 const struct compat_iovec __user* vec,
1307 compat_ulong_t vlen, rwf_t flags)
1309 struct fd f = fdget_pos(fd);
1315 pos = f.file->f_pos;
1316 ret = compat_writev(f.file, vec, vlen, &pos, flags);
1318 f.file->f_pos = pos;
1323 COMPAT_SYSCALL_DEFINE3(writev, compat_ulong_t, fd,
1324 const struct compat_iovec __user *, vec,
1325 compat_ulong_t, vlen)
1327 return do_compat_writev(fd, vec, vlen, 0);
1330 static long do_compat_pwritev64(unsigned long fd,
1331 const struct compat_iovec __user *vec,
1332 unsigned long vlen, loff_t pos, rwf_t flags)
1343 if (f.file->f_mode & FMODE_PWRITE)
1344 ret = compat_writev(f.file, vec, vlen, &pos, flags);
1349 #ifdef __ARCH_WANT_COMPAT_SYS_PWRITEV64
1350 COMPAT_SYSCALL_DEFINE4(pwritev64, unsigned long, fd,
1351 const struct compat_iovec __user *,vec,
1352 unsigned long, vlen, loff_t, pos)
1354 return do_compat_pwritev64(fd, vec, vlen, pos, 0);
1358 COMPAT_SYSCALL_DEFINE5(pwritev, compat_ulong_t, fd,
1359 const struct compat_iovec __user *,vec,
1360 compat_ulong_t, vlen, u32, pos_low, u32, pos_high)
1362 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1364 return do_compat_pwritev64(fd, vec, vlen, pos, 0);
1367 #ifdef __ARCH_WANT_COMPAT_SYS_PWRITEV64V2
1368 COMPAT_SYSCALL_DEFINE5(pwritev64v2, unsigned long, fd,
1369 const struct compat_iovec __user *,vec,
1370 unsigned long, vlen, loff_t, pos, rwf_t, flags)
1373 return do_compat_writev(fd, vec, vlen, flags);
1375 return do_compat_pwritev64(fd, vec, vlen, pos, flags);
1379 COMPAT_SYSCALL_DEFINE6(pwritev2, compat_ulong_t, fd,
1380 const struct compat_iovec __user *,vec,
1381 compat_ulong_t, vlen, u32, pos_low, u32, pos_high, rwf_t, flags)
1383 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1386 return do_compat_writev(fd, vec, vlen, flags);
1388 return do_compat_pwritev64(fd, vec, vlen, pos, flags);
1393 static ssize_t do_sendfile(int out_fd, int in_fd, loff_t *ppos,
1394 size_t count, loff_t max)
1397 struct inode *in_inode, *out_inode;
1404 * Get input file, and verify that it is ok..
1410 if (!(in.file->f_mode & FMODE_READ))
1414 pos = in.file->f_pos;
1417 if (!(in.file->f_mode & FMODE_PREAD))
1420 retval = rw_verify_area(READ, in.file, &pos, count);
1423 if (count > MAX_RW_COUNT)
1424 count = MAX_RW_COUNT;
1427 * Get output file, and verify that it is ok..
1430 out = fdget(out_fd);
1433 if (!(out.file->f_mode & FMODE_WRITE))
1435 in_inode = file_inode(in.file);
1436 out_inode = file_inode(out.file);
1437 out_pos = out.file->f_pos;
1438 retval = rw_verify_area(WRITE, out.file, &out_pos, count);
1443 max = min(in_inode->i_sb->s_maxbytes, out_inode->i_sb->s_maxbytes);
1445 if (unlikely(pos + count > max)) {
1446 retval = -EOVERFLOW;
1455 * We need to debate whether we can enable this or not. The
1456 * man page documents EAGAIN return for the output at least,
1457 * and the application is arguably buggy if it doesn't expect
1458 * EAGAIN on a non-blocking file descriptor.
1460 if (in.file->f_flags & O_NONBLOCK)
1461 fl = SPLICE_F_NONBLOCK;
1463 file_start_write(out.file);
1464 retval = do_splice_direct(in.file, &pos, out.file, &out_pos, count, fl);
1465 file_end_write(out.file);
1468 add_rchar(current, retval);
1469 add_wchar(current, retval);
1470 fsnotify_access(in.file);
1471 fsnotify_modify(out.file);
1472 out.file->f_pos = out_pos;
1476 in.file->f_pos = pos;
1482 retval = -EOVERFLOW;
1492 SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd, off_t __user *, offset, size_t, count)
1499 if (unlikely(get_user(off, offset)))
1502 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
1503 if (unlikely(put_user(pos, offset)))
1508 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1511 SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd, loff_t __user *, offset, size_t, count)
1517 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
1519 ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
1520 if (unlikely(put_user(pos, offset)))
1525 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1528 #ifdef CONFIG_COMPAT
1529 COMPAT_SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd,
1530 compat_off_t __user *, offset, compat_size_t, count)
1537 if (unlikely(get_user(off, offset)))
1540 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
1541 if (unlikely(put_user(pos, offset)))
1546 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1549 COMPAT_SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd,
1550 compat_loff_t __user *, offset, compat_size_t, count)
1556 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
1558 ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
1559 if (unlikely(put_user(pos, offset)))
1564 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1569 * generic_copy_file_range - copy data between two files
1570 * @file_in: file structure to read from
1571 * @pos_in: file offset to read from
1572 * @file_out: file structure to write data to
1573 * @pos_out: file offset to write data to
1574 * @len: amount of data to copy
1575 * @flags: copy flags
1577 * This is a generic filesystem helper to copy data from one file to another.
1578 * It has no constraints on the source or destination file owners - the files
1579 * can belong to different superblocks and different filesystem types. Short
1580 * copies are allowed.
1582 * This should be called from the @file_out filesystem, as per the
1583 * ->copy_file_range() method.
1585 * Returns the number of bytes copied or a negative error indicating the
1589 ssize_t generic_copy_file_range(struct file *file_in, loff_t pos_in,
1590 struct file *file_out, loff_t pos_out,
1591 size_t len, unsigned int flags)
1593 return do_splice_direct(file_in, &pos_in, file_out, &pos_out,
1594 len > MAX_RW_COUNT ? MAX_RW_COUNT : len, 0);
1596 EXPORT_SYMBOL(generic_copy_file_range);
1598 static ssize_t do_copy_file_range(struct file *file_in, loff_t pos_in,
1599 struct file *file_out, loff_t pos_out,
1600 size_t len, unsigned int flags)
1603 * Although we now allow filesystems to handle cross sb copy, passing
1604 * a file of the wrong filesystem type to filesystem driver can result
1605 * in an attempt to dereference the wrong type of ->private_data, so
1606 * avoid doing that until we really have a good reason. NFS defines
1607 * several different file_system_type structures, but they all end up
1608 * using the same ->copy_file_range() function pointer.
1610 if (file_out->f_op->copy_file_range &&
1611 file_out->f_op->copy_file_range == file_in->f_op->copy_file_range)
1612 return file_out->f_op->copy_file_range(file_in, pos_in,
1616 return generic_copy_file_range(file_in, pos_in, file_out, pos_out, len,
1621 * copy_file_range() differs from regular file read and write in that it
1622 * specifically allows return partial success. When it does so is up to
1623 * the copy_file_range method.
1625 ssize_t vfs_copy_file_range(struct file *file_in, loff_t pos_in,
1626 struct file *file_out, loff_t pos_out,
1627 size_t len, unsigned int flags)
1634 ret = generic_copy_file_checks(file_in, pos_in, file_out, pos_out, &len,
1639 ret = rw_verify_area(READ, file_in, &pos_in, len);
1643 ret = rw_verify_area(WRITE, file_out, &pos_out, len);
1650 file_start_write(file_out);
1653 * Try cloning first, this is supported by more file systems, and
1654 * more efficient if both clone and copy are supported (e.g. NFS).
1656 if (file_in->f_op->remap_file_range &&
1657 file_inode(file_in)->i_sb == file_inode(file_out)->i_sb) {
1660 cloned = file_in->f_op->remap_file_range(file_in, pos_in,
1662 min_t(loff_t, MAX_RW_COUNT, len),
1663 REMAP_FILE_CAN_SHORTEN);
1670 ret = do_copy_file_range(file_in, pos_in, file_out, pos_out, len,
1672 WARN_ON_ONCE(ret == -EOPNOTSUPP);
1675 fsnotify_access(file_in);
1676 add_rchar(current, ret);
1677 fsnotify_modify(file_out);
1678 add_wchar(current, ret);
1684 file_end_write(file_out);
1688 EXPORT_SYMBOL(vfs_copy_file_range);
1690 SYSCALL_DEFINE6(copy_file_range, int, fd_in, loff_t __user *, off_in,
1691 int, fd_out, loff_t __user *, off_out,
1692 size_t, len, unsigned int, flags)
1698 ssize_t ret = -EBADF;
1700 f_in = fdget(fd_in);
1704 f_out = fdget(fd_out);
1710 if (copy_from_user(&pos_in, off_in, sizeof(loff_t)))
1713 pos_in = f_in.file->f_pos;
1717 if (copy_from_user(&pos_out, off_out, sizeof(loff_t)))
1720 pos_out = f_out.file->f_pos;
1723 ret = vfs_copy_file_range(f_in.file, pos_in, f_out.file, pos_out, len,
1730 if (copy_to_user(off_in, &pos_in, sizeof(loff_t)))
1733 f_in.file->f_pos = pos_in;
1737 if (copy_to_user(off_out, &pos_out, sizeof(loff_t)))
1740 f_out.file->f_pos = pos_out;
1752 static int remap_verify_area(struct file *file, loff_t pos, loff_t len,
1755 struct inode *inode = file_inode(file);
1757 if (unlikely(pos < 0 || len < 0))
1760 if (unlikely((loff_t) (pos + len) < 0))
1763 if (unlikely(inode->i_flctx && mandatory_lock(inode))) {
1764 loff_t end = len ? pos + len - 1 : OFFSET_MAX;
1767 retval = locks_mandatory_area(inode, file, pos, end,
1768 write ? F_WRLCK : F_RDLCK);
1773 return security_file_permission(file, write ? MAY_WRITE : MAY_READ);
1776 * Ensure that we don't remap a partial EOF block in the middle of something
1777 * else. Assume that the offsets have already been checked for block
1780 * For deduplication we always scale down to the previous block because we
1781 * can't meaningfully compare post-EOF contents.
1783 * For clone we only link a partial EOF block above the destination file's EOF.
1785 * Shorten the request if possible.
1787 static int generic_remap_check_len(struct inode *inode_in,
1788 struct inode *inode_out,
1791 unsigned int remap_flags)
1793 u64 blkmask = i_blocksize(inode_in) - 1;
1794 loff_t new_len = *len;
1796 if ((*len & blkmask) == 0)
1799 if ((remap_flags & REMAP_FILE_DEDUP) ||
1800 pos_out + *len < i_size_read(inode_out))
1801 new_len &= ~blkmask;
1803 if (new_len == *len)
1806 if (remap_flags & REMAP_FILE_CAN_SHORTEN) {
1811 return (remap_flags & REMAP_FILE_DEDUP) ? -EBADE : -EINVAL;
1814 /* Read a page's worth of file data into the page cache. */
1815 static struct page *vfs_dedupe_get_page(struct inode *inode, loff_t offset)
1819 page = read_mapping_page(inode->i_mapping, offset >> PAGE_SHIFT, NULL);
1822 if (!PageUptodate(page)) {
1824 return ERR_PTR(-EIO);
1830 * Lock two pages, ensuring that we lock in offset order if the pages are from
1833 static void vfs_lock_two_pages(struct page *page1, struct page *page2)
1835 /* Always lock in order of increasing index. */
1836 if (page1->index > page2->index)
1844 /* Unlock two pages, being careful not to unlock the same page twice. */
1845 static void vfs_unlock_two_pages(struct page *page1, struct page *page2)
1853 * Compare extents of two files to see if they are the same.
1854 * Caller must have locked both inodes to prevent write races.
1856 static int vfs_dedupe_file_range_compare(struct inode *src, loff_t srcoff,
1857 struct inode *dest, loff_t destoff,
1858 loff_t len, bool *is_same)
1864 struct page *src_page;
1865 struct page *dest_page;
1873 src_poff = srcoff & (PAGE_SIZE - 1);
1874 dest_poff = destoff & (PAGE_SIZE - 1);
1875 cmp_len = min(PAGE_SIZE - src_poff,
1876 PAGE_SIZE - dest_poff);
1877 cmp_len = min(cmp_len, len);
1881 src_page = vfs_dedupe_get_page(src, srcoff);
1882 if (IS_ERR(src_page)) {
1883 error = PTR_ERR(src_page);
1886 dest_page = vfs_dedupe_get_page(dest, destoff);
1887 if (IS_ERR(dest_page)) {
1888 error = PTR_ERR(dest_page);
1893 vfs_lock_two_pages(src_page, dest_page);
1896 * Now that we've locked both pages, make sure they're still
1897 * mapped to the file data we're interested in. If not,
1898 * someone is invalidating pages on us and we lose.
1900 if (!PageUptodate(src_page) || !PageUptodate(dest_page) ||
1901 src_page->mapping != src->i_mapping ||
1902 dest_page->mapping != dest->i_mapping) {
1907 src_addr = kmap_atomic(src_page);
1908 dest_addr = kmap_atomic(dest_page);
1910 flush_dcache_page(src_page);
1911 flush_dcache_page(dest_page);
1913 if (memcmp(src_addr + src_poff, dest_addr + dest_poff, cmp_len))
1916 kunmap_atomic(dest_addr);
1917 kunmap_atomic(src_addr);
1919 vfs_unlock_two_pages(src_page, dest_page);
1920 put_page(dest_page);
1939 * Check that the two inodes are eligible for cloning, the ranges make
1940 * sense, and then flush all dirty data. Caller must ensure that the
1941 * inodes have been locked against any other modifications.
1943 * If there's an error, then the usual negative error code is returned.
1944 * Otherwise returns 0 with *len set to the request length.
1946 int generic_remap_file_range_prep(struct file *file_in, loff_t pos_in,
1947 struct file *file_out, loff_t pos_out,
1948 loff_t *len, unsigned int remap_flags)
1950 struct inode *inode_in = file_inode(file_in);
1951 struct inode *inode_out = file_inode(file_out);
1952 bool same_inode = (inode_in == inode_out);
1955 /* Don't touch certain kinds of inodes */
1956 if (IS_IMMUTABLE(inode_out))
1959 if (IS_SWAPFILE(inode_in) || IS_SWAPFILE(inode_out))
1962 /* Don't reflink dirs, pipes, sockets... */
1963 if (S_ISDIR(inode_in->i_mode) || S_ISDIR(inode_out->i_mode))
1965 if (!S_ISREG(inode_in->i_mode) || !S_ISREG(inode_out->i_mode))
1968 /* Zero length dedupe exits immediately; reflink goes to EOF. */
1970 loff_t isize = i_size_read(inode_in);
1972 if ((remap_flags & REMAP_FILE_DEDUP) || pos_in == isize)
1976 *len = isize - pos_in;
1981 /* Check that we don't violate system file offset limits. */
1982 ret = generic_remap_checks(file_in, pos_in, file_out, pos_out, len,
1987 /* Wait for the completion of any pending IOs on both files */
1988 inode_dio_wait(inode_in);
1990 inode_dio_wait(inode_out);
1992 ret = filemap_write_and_wait_range(inode_in->i_mapping,
1993 pos_in, pos_in + *len - 1);
1997 ret = filemap_write_and_wait_range(inode_out->i_mapping,
1998 pos_out, pos_out + *len - 1);
2003 * Check that the extents are the same.
2005 if (remap_flags & REMAP_FILE_DEDUP) {
2006 bool is_same = false;
2008 ret = vfs_dedupe_file_range_compare(inode_in, pos_in,
2009 inode_out, pos_out, *len, &is_same);
2016 ret = generic_remap_check_len(inode_in, inode_out, pos_out, len,
2021 /* If can't alter the file contents, we're done. */
2022 if (!(remap_flags & REMAP_FILE_DEDUP))
2023 ret = file_modified(file_out);
2027 EXPORT_SYMBOL(generic_remap_file_range_prep);
2029 loff_t do_clone_file_range(struct file *file_in, loff_t pos_in,
2030 struct file *file_out, loff_t pos_out,
2031 loff_t len, unsigned int remap_flags)
2035 WARN_ON_ONCE(remap_flags & REMAP_FILE_DEDUP);
2038 * FICLONE/FICLONERANGE ioctls enforce that src and dest files are on
2039 * the same mount. Practically, they only need to be on the same file
2042 if (file_inode(file_in)->i_sb != file_inode(file_out)->i_sb)
2045 ret = generic_file_rw_checks(file_in, file_out);
2049 if (!file_in->f_op->remap_file_range)
2052 ret = remap_verify_area(file_in, pos_in, len, false);
2056 ret = remap_verify_area(file_out, pos_out, len, true);
2060 ret = file_in->f_op->remap_file_range(file_in, pos_in,
2061 file_out, pos_out, len, remap_flags);
2065 fsnotify_access(file_in);
2066 fsnotify_modify(file_out);
2069 EXPORT_SYMBOL(do_clone_file_range);
2071 loff_t vfs_clone_file_range(struct file *file_in, loff_t pos_in,
2072 struct file *file_out, loff_t pos_out,
2073 loff_t len, unsigned int remap_flags)
2077 file_start_write(file_out);
2078 ret = do_clone_file_range(file_in, pos_in, file_out, pos_out, len,
2080 file_end_write(file_out);
2084 EXPORT_SYMBOL(vfs_clone_file_range);
2086 /* Check whether we are allowed to dedupe the destination file */
2087 static bool allow_file_dedupe(struct file *file)
2089 if (capable(CAP_SYS_ADMIN))
2091 if (file->f_mode & FMODE_WRITE)
2093 if (uid_eq(current_fsuid(), file_inode(file)->i_uid))
2095 if (!inode_permission(file_inode(file), MAY_WRITE))
2100 loff_t vfs_dedupe_file_range_one(struct file *src_file, loff_t src_pos,
2101 struct file *dst_file, loff_t dst_pos,
2102 loff_t len, unsigned int remap_flags)
2106 WARN_ON_ONCE(remap_flags & ~(REMAP_FILE_DEDUP |
2107 REMAP_FILE_CAN_SHORTEN));
2109 ret = mnt_want_write_file(dst_file);
2113 ret = remap_verify_area(dst_file, dst_pos, len, true);
2115 goto out_drop_write;
2118 if (!allow_file_dedupe(dst_file))
2119 goto out_drop_write;
2122 if (src_file->f_path.mnt != dst_file->f_path.mnt)
2123 goto out_drop_write;
2126 if (S_ISDIR(file_inode(dst_file)->i_mode))
2127 goto out_drop_write;
2130 if (!dst_file->f_op->remap_file_range)
2131 goto out_drop_write;
2135 goto out_drop_write;
2138 ret = dst_file->f_op->remap_file_range(src_file, src_pos, dst_file,
2139 dst_pos, len, remap_flags | REMAP_FILE_DEDUP);
2141 mnt_drop_write_file(dst_file);
2145 EXPORT_SYMBOL(vfs_dedupe_file_range_one);
2147 int vfs_dedupe_file_range(struct file *file, struct file_dedupe_range *same)
2149 struct file_dedupe_range_info *info;
2150 struct inode *src = file_inode(file);
2155 u16 count = same->dest_count;
2158 if (!(file->f_mode & FMODE_READ))
2161 if (same->reserved1 || same->reserved2)
2164 off = same->src_offset;
2165 len = same->src_length;
2167 if (S_ISDIR(src->i_mode))
2170 if (!S_ISREG(src->i_mode))
2173 if (!file->f_op->remap_file_range)
2176 ret = remap_verify_area(file, off, len, false);
2181 if (off + len > i_size_read(src))
2184 /* Arbitrary 1G limit on a single dedupe request, can be raised. */
2185 len = min_t(u64, len, 1 << 30);
2187 /* pre-format output fields to sane values */
2188 for (i = 0; i < count; i++) {
2189 same->info[i].bytes_deduped = 0ULL;
2190 same->info[i].status = FILE_DEDUPE_RANGE_SAME;
2193 for (i = 0, info = same->info; i < count; i++, info++) {
2194 struct fd dst_fd = fdget(info->dest_fd);
2195 struct file *dst_file = dst_fd.file;
2198 info->status = -EBADF;
2202 if (info->reserved) {
2203 info->status = -EINVAL;
2207 deduped = vfs_dedupe_file_range_one(file, off, dst_file,
2208 info->dest_offset, len,
2209 REMAP_FILE_CAN_SHORTEN);
2210 if (deduped == -EBADE)
2211 info->status = FILE_DEDUPE_RANGE_DIFFERS;
2212 else if (deduped < 0)
2213 info->status = deduped;
2215 info->bytes_deduped = len;
2220 if (fatal_signal_pending(current))
2225 EXPORT_SYMBOL(vfs_dedupe_file_range);