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 defined(__ARCH_WANT_SYS_LLSEEK)
336 SYSCALL_DEFINE5(llseek, unsigned int, fd, unsigned long, offset_high,
337 unsigned long, offset_low, loff_t __user *, result,
338 unsigned int, whence)
341 struct fd f = fdget_pos(fd);
348 if (whence > SEEK_MAX)
351 offset = vfs_llseek(f.file, ((loff_t) offset_high << 32) | offset_low,
354 retval = (int)offset;
357 if (!copy_to_user(result, &offset, sizeof(offset)))
366 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))
376 * ranged mandatory locking does not apply to streams - it makes sense
377 * only for files where position has a meaning.
382 if (unlikely(pos < 0)) {
383 if (!unsigned_offsets(file))
385 if (count >= -pos) /* both values are in 0..LLONG_MAX */
387 } else if (unlikely((loff_t) (pos + count) < 0)) {
388 if (!unsigned_offsets(file))
392 if (unlikely(inode->i_flctx && mandatory_lock(inode))) {
393 retval = locks_mandatory_area(inode, file, pos, pos + count - 1,
394 read_write == READ ? F_RDLCK : F_WRLCK);
400 return security_file_permission(file,
401 read_write == READ ? MAY_READ : MAY_WRITE);
404 static ssize_t new_sync_read(struct file *filp, char __user *buf, size_t len, loff_t *ppos)
406 struct iovec iov = { .iov_base = buf, .iov_len = len };
408 struct iov_iter iter;
411 init_sync_kiocb(&kiocb, filp);
412 kiocb.ki_pos = (ppos ? *ppos : 0);
413 iov_iter_init(&iter, READ, &iov, 1, len);
415 ret = call_read_iter(filp, &kiocb, &iter);
416 BUG_ON(ret == -EIOCBQUEUED);
418 *ppos = kiocb.ki_pos;
422 ssize_t __vfs_read(struct file *file, char __user *buf, size_t count,
425 if (file->f_op->read)
426 return file->f_op->read(file, buf, count, pos);
427 else if (file->f_op->read_iter)
428 return new_sync_read(file, buf, count, pos);
433 ssize_t kernel_read(struct file *file, void *buf, size_t count, loff_t *pos)
440 /* The cast to a user pointer is valid due to the set_fs() */
441 result = vfs_read(file, (void __user *)buf, count, pos);
445 EXPORT_SYMBOL(kernel_read);
447 ssize_t vfs_read(struct file *file, char __user *buf, size_t count, loff_t *pos)
451 if (!(file->f_mode & FMODE_READ))
453 if (!(file->f_mode & FMODE_CAN_READ))
455 if (unlikely(!access_ok(buf, count)))
458 ret = rw_verify_area(READ, file, pos, count);
460 if (count > MAX_RW_COUNT)
461 count = MAX_RW_COUNT;
462 ret = __vfs_read(file, buf, count, pos);
464 fsnotify_access(file);
465 add_rchar(current, ret);
473 static ssize_t new_sync_write(struct file *filp, const char __user *buf, size_t len, loff_t *ppos)
475 struct iovec iov = { .iov_base = (void __user *)buf, .iov_len = len };
477 struct iov_iter iter;
480 init_sync_kiocb(&kiocb, filp);
481 kiocb.ki_pos = (ppos ? *ppos : 0);
482 iov_iter_init(&iter, WRITE, &iov, 1, len);
484 ret = call_write_iter(filp, &kiocb, &iter);
485 BUG_ON(ret == -EIOCBQUEUED);
487 *ppos = kiocb.ki_pos;
491 static ssize_t __vfs_write(struct file *file, const char __user *p,
492 size_t count, loff_t *pos)
494 if (file->f_op->write)
495 return file->f_op->write(file, p, count, pos);
496 else if (file->f_op->write_iter)
497 return new_sync_write(file, p, count, pos);
502 ssize_t __kernel_write(struct file *file, const void *buf, size_t count, loff_t *pos)
505 const char __user *p;
508 if (!(file->f_mode & FMODE_CAN_WRITE))
513 p = (__force const char __user *)buf;
514 if (count > MAX_RW_COUNT)
515 count = MAX_RW_COUNT;
516 ret = __vfs_write(file, p, count, pos);
519 fsnotify_modify(file);
520 add_wchar(current, ret);
525 EXPORT_SYMBOL(__kernel_write);
527 ssize_t kernel_write(struct file *file, const void *buf, size_t count,
535 /* The cast to a user pointer is valid due to the set_fs() */
536 res = vfs_write(file, (__force const char __user *)buf, count, pos);
541 EXPORT_SYMBOL(kernel_write);
543 ssize_t vfs_write(struct file *file, const char __user *buf, size_t count, loff_t *pos)
547 if (!(file->f_mode & FMODE_WRITE))
549 if (!(file->f_mode & FMODE_CAN_WRITE))
551 if (unlikely(!access_ok(buf, count)))
554 ret = rw_verify_area(WRITE, file, pos, count);
556 if (count > MAX_RW_COUNT)
557 count = MAX_RW_COUNT;
558 file_start_write(file);
559 ret = __vfs_write(file, buf, count, pos);
561 fsnotify_modify(file);
562 add_wchar(current, ret);
565 file_end_write(file);
571 /* file_ppos returns &file->f_pos or NULL if file is stream */
572 static inline loff_t *file_ppos(struct file *file)
574 return file->f_mode & FMODE_STREAM ? NULL : &file->f_pos;
577 ssize_t ksys_read(unsigned int fd, char __user *buf, size_t count)
579 struct fd f = fdget_pos(fd);
580 ssize_t ret = -EBADF;
583 loff_t pos, *ppos = file_ppos(f.file);
588 ret = vfs_read(f.file, buf, count, ppos);
589 if (ret >= 0 && ppos)
596 SYSCALL_DEFINE3(read, unsigned int, fd, char __user *, buf, size_t, count)
598 return ksys_read(fd, buf, count);
601 ssize_t ksys_write(unsigned int fd, const char __user *buf, size_t count)
603 struct fd f = fdget_pos(fd);
604 ssize_t ret = -EBADF;
607 loff_t pos, *ppos = file_ppos(f.file);
612 ret = vfs_write(f.file, buf, count, ppos);
613 if (ret >= 0 && ppos)
621 SYSCALL_DEFINE3(write, unsigned int, fd, const char __user *, buf,
624 return ksys_write(fd, buf, count);
627 ssize_t ksys_pread64(unsigned int fd, char __user *buf, size_t count,
631 ssize_t ret = -EBADF;
639 if (f.file->f_mode & FMODE_PREAD)
640 ret = vfs_read(f.file, buf, count, &pos);
647 SYSCALL_DEFINE4(pread64, unsigned int, fd, char __user *, buf,
648 size_t, count, loff_t, pos)
650 return ksys_pread64(fd, buf, count, pos);
653 ssize_t ksys_pwrite64(unsigned int fd, const char __user *buf,
654 size_t count, loff_t pos)
657 ssize_t ret = -EBADF;
665 if (f.file->f_mode & FMODE_PWRITE)
666 ret = vfs_write(f.file, buf, count, &pos);
673 SYSCALL_DEFINE4(pwrite64, unsigned int, fd, const char __user *, buf,
674 size_t, count, loff_t, pos)
676 return ksys_pwrite64(fd, buf, count, pos);
679 static ssize_t do_iter_readv_writev(struct file *filp, struct iov_iter *iter,
680 loff_t *ppos, int type, rwf_t flags)
685 init_sync_kiocb(&kiocb, filp);
686 ret = kiocb_set_rw_flags(&kiocb, flags);
689 kiocb.ki_pos = (ppos ? *ppos : 0);
692 ret = call_read_iter(filp, &kiocb, iter);
694 ret = call_write_iter(filp, &kiocb, iter);
695 BUG_ON(ret == -EIOCBQUEUED);
697 *ppos = kiocb.ki_pos;
701 /* Do it by hand, with file-ops */
702 static ssize_t do_loop_readv_writev(struct file *filp, struct iov_iter *iter,
703 loff_t *ppos, int type, rwf_t flags)
707 if (flags & ~RWF_HIPRI)
710 while (iov_iter_count(iter)) {
711 struct iovec iovec = iov_iter_iovec(iter);
715 nr = filp->f_op->read(filp, iovec.iov_base,
716 iovec.iov_len, ppos);
718 nr = filp->f_op->write(filp, iovec.iov_base,
719 iovec.iov_len, ppos);
728 if (nr != iovec.iov_len)
730 iov_iter_advance(iter, nr);
737 * rw_copy_check_uvector() - Copy an array of &struct iovec from userspace
738 * into the kernel and check that it is valid.
740 * @type: One of %CHECK_IOVEC_ONLY, %READ, or %WRITE.
741 * @uvector: Pointer to the userspace array.
742 * @nr_segs: Number of elements in userspace array.
743 * @fast_segs: Number of elements in @fast_pointer.
744 * @fast_pointer: Pointer to (usually small on-stack) kernel array.
745 * @ret_pointer: (output parameter) Pointer to a variable that will point to
746 * either @fast_pointer, a newly allocated kernel array, or NULL,
747 * depending on which array was used.
749 * This function copies an array of &struct iovec of @nr_segs from
750 * userspace into the kernel and checks that each element is valid (e.g.
751 * it does not point to a kernel address or cause overflow by being too
754 * As an optimization, the caller may provide a pointer to a small
755 * on-stack array in @fast_pointer, typically %UIO_FASTIOV elements long
756 * (the size of this array, or 0 if unused, should be given in @fast_segs).
758 * @ret_pointer will always point to the array that was used, so the
759 * caller must take care not to call kfree() on it e.g. in case the
760 * @fast_pointer array was used and it was allocated on the stack.
762 * Return: The total number of bytes covered by the iovec array on success
763 * or a negative error code on error.
765 ssize_t rw_copy_check_uvector(int type, const struct iovec __user * uvector,
766 unsigned long nr_segs, unsigned long fast_segs,
767 struct iovec *fast_pointer,
768 struct iovec **ret_pointer)
772 struct iovec *iov = fast_pointer;
775 * SuS says "The readv() function *may* fail if the iovcnt argument
776 * was less than or equal to 0, or greater than {IOV_MAX}. Linux has
777 * traditionally returned zero for zero segments, so...
785 * First get the "struct iovec" from user memory and
786 * verify all the pointers
788 if (nr_segs > UIO_MAXIOV) {
792 if (nr_segs > fast_segs) {
793 iov = kmalloc_array(nr_segs, sizeof(struct iovec), GFP_KERNEL);
799 if (copy_from_user(iov, uvector, nr_segs*sizeof(*uvector))) {
805 * According to the Single Unix Specification we should return EINVAL
806 * if an element length is < 0 when cast to ssize_t or if the
807 * total length would overflow the ssize_t return value of the
810 * Linux caps all read/write calls to MAX_RW_COUNT, and avoids the
814 for (seg = 0; seg < nr_segs; seg++) {
815 void __user *buf = iov[seg].iov_base;
816 ssize_t len = (ssize_t)iov[seg].iov_len;
818 /* see if we we're about to use an invalid len or if
819 * it's about to overflow ssize_t */
825 && unlikely(!access_ok(buf, len))) {
829 if (len > MAX_RW_COUNT - ret) {
830 len = MAX_RW_COUNT - ret;
831 iov[seg].iov_len = len;
841 ssize_t compat_rw_copy_check_uvector(int type,
842 const struct compat_iovec __user *uvector, unsigned long nr_segs,
843 unsigned long fast_segs, struct iovec *fast_pointer,
844 struct iovec **ret_pointer)
846 compat_ssize_t tot_len;
847 struct iovec *iov = *ret_pointer = fast_pointer;
852 * SuS says "The readv() function *may* fail if the iovcnt argument
853 * was less than or equal to 0, or greater than {IOV_MAX}. Linux has
854 * traditionally returned zero for zero segments, so...
860 if (nr_segs > UIO_MAXIOV)
862 if (nr_segs > fast_segs) {
864 iov = kmalloc_array(nr_segs, sizeof(struct iovec), GFP_KERNEL);
871 if (!access_ok(uvector, nr_segs*sizeof(*uvector)))
875 * Single unix specification:
876 * We should -EINVAL if an element length is not >= 0 and fitting an
879 * In Linux, the total length is limited to MAX_RW_COUNT, there is
880 * no overflow possibility.
884 for (seg = 0; seg < nr_segs; seg++) {
888 if (__get_user(len, &uvector->iov_len) ||
889 __get_user(buf, &uvector->iov_base)) {
893 if (len < 0) /* size_t not fitting in compat_ssize_t .. */
896 !access_ok(compat_ptr(buf), len)) {
900 if (len > MAX_RW_COUNT - tot_len)
901 len = MAX_RW_COUNT - tot_len;
903 iov->iov_base = compat_ptr(buf);
904 iov->iov_len = (compat_size_t) len;
915 static ssize_t do_iter_read(struct file *file, struct iov_iter *iter,
916 loff_t *pos, rwf_t flags)
921 if (!(file->f_mode & FMODE_READ))
923 if (!(file->f_mode & FMODE_CAN_READ))
926 tot_len = iov_iter_count(iter);
929 ret = rw_verify_area(READ, file, pos, tot_len);
933 if (file->f_op->read_iter)
934 ret = do_iter_readv_writev(file, iter, pos, READ, flags);
936 ret = do_loop_readv_writev(file, iter, pos, READ, flags);
939 fsnotify_access(file);
943 ssize_t vfs_iocb_iter_read(struct file *file, struct kiocb *iocb,
944 struct iov_iter *iter)
949 if (!file->f_op->read_iter)
951 if (!(file->f_mode & FMODE_READ))
953 if (!(file->f_mode & FMODE_CAN_READ))
956 tot_len = iov_iter_count(iter);
959 ret = rw_verify_area(READ, file, &iocb->ki_pos, tot_len);
963 ret = call_read_iter(file, iocb, iter);
966 fsnotify_access(file);
969 EXPORT_SYMBOL(vfs_iocb_iter_read);
971 ssize_t vfs_iter_read(struct file *file, struct iov_iter *iter, loff_t *ppos,
974 if (!file->f_op->read_iter)
976 return do_iter_read(file, iter, ppos, flags);
978 EXPORT_SYMBOL(vfs_iter_read);
980 static ssize_t do_iter_write(struct file *file, struct iov_iter *iter,
981 loff_t *pos, rwf_t flags)
986 if (!(file->f_mode & FMODE_WRITE))
988 if (!(file->f_mode & FMODE_CAN_WRITE))
991 tot_len = iov_iter_count(iter);
994 ret = rw_verify_area(WRITE, file, pos, tot_len);
998 if (file->f_op->write_iter)
999 ret = do_iter_readv_writev(file, iter, pos, WRITE, flags);
1001 ret = do_loop_readv_writev(file, iter, pos, WRITE, flags);
1003 fsnotify_modify(file);
1007 ssize_t vfs_iocb_iter_write(struct file *file, struct kiocb *iocb,
1008 struct iov_iter *iter)
1013 if (!file->f_op->write_iter)
1015 if (!(file->f_mode & FMODE_WRITE))
1017 if (!(file->f_mode & FMODE_CAN_WRITE))
1020 tot_len = iov_iter_count(iter);
1023 ret = rw_verify_area(WRITE, file, &iocb->ki_pos, tot_len);
1027 ret = call_write_iter(file, iocb, iter);
1029 fsnotify_modify(file);
1033 EXPORT_SYMBOL(vfs_iocb_iter_write);
1035 ssize_t vfs_iter_write(struct file *file, struct iov_iter *iter, loff_t *ppos,
1038 if (!file->f_op->write_iter)
1040 return do_iter_write(file, iter, ppos, flags);
1042 EXPORT_SYMBOL(vfs_iter_write);
1044 ssize_t vfs_readv(struct file *file, const struct iovec __user *vec,
1045 unsigned long vlen, loff_t *pos, rwf_t flags)
1047 struct iovec iovstack[UIO_FASTIOV];
1048 struct iovec *iov = iovstack;
1049 struct iov_iter iter;
1052 ret = import_iovec(READ, vec, vlen, ARRAY_SIZE(iovstack), &iov, &iter);
1054 ret = do_iter_read(file, &iter, pos, flags);
1061 static ssize_t vfs_writev(struct file *file, const struct iovec __user *vec,
1062 unsigned long vlen, loff_t *pos, rwf_t flags)
1064 struct iovec iovstack[UIO_FASTIOV];
1065 struct iovec *iov = iovstack;
1066 struct iov_iter iter;
1069 ret = import_iovec(WRITE, vec, vlen, ARRAY_SIZE(iovstack), &iov, &iter);
1071 file_start_write(file);
1072 ret = do_iter_write(file, &iter, pos, flags);
1073 file_end_write(file);
1079 static ssize_t do_readv(unsigned long fd, const struct iovec __user *vec,
1080 unsigned long vlen, rwf_t flags)
1082 struct fd f = fdget_pos(fd);
1083 ssize_t ret = -EBADF;
1086 loff_t pos, *ppos = file_ppos(f.file);
1091 ret = vfs_readv(f.file, vec, vlen, ppos, flags);
1092 if (ret >= 0 && ppos)
1093 f.file->f_pos = pos;
1098 add_rchar(current, ret);
1103 static ssize_t do_writev(unsigned long fd, const struct iovec __user *vec,
1104 unsigned long vlen, rwf_t flags)
1106 struct fd f = fdget_pos(fd);
1107 ssize_t ret = -EBADF;
1110 loff_t pos, *ppos = file_ppos(f.file);
1115 ret = vfs_writev(f.file, vec, vlen, ppos, flags);
1116 if (ret >= 0 && ppos)
1117 f.file->f_pos = pos;
1122 add_wchar(current, ret);
1127 static inline loff_t pos_from_hilo(unsigned long high, unsigned long low)
1129 #define HALF_LONG_BITS (BITS_PER_LONG / 2)
1130 return (((loff_t)high << HALF_LONG_BITS) << HALF_LONG_BITS) | low;
1133 static ssize_t do_preadv(unsigned long fd, const struct iovec __user *vec,
1134 unsigned long vlen, loff_t pos, rwf_t flags)
1137 ssize_t ret = -EBADF;
1145 if (f.file->f_mode & FMODE_PREAD)
1146 ret = vfs_readv(f.file, vec, vlen, &pos, flags);
1151 add_rchar(current, ret);
1156 static ssize_t do_pwritev(unsigned long fd, const struct iovec __user *vec,
1157 unsigned long vlen, loff_t pos, rwf_t flags)
1160 ssize_t ret = -EBADF;
1168 if (f.file->f_mode & FMODE_PWRITE)
1169 ret = vfs_writev(f.file, vec, vlen, &pos, flags);
1174 add_wchar(current, ret);
1179 SYSCALL_DEFINE3(readv, unsigned long, fd, const struct iovec __user *, vec,
1180 unsigned long, vlen)
1182 return do_readv(fd, vec, vlen, 0);
1185 SYSCALL_DEFINE3(writev, unsigned long, fd, const struct iovec __user *, vec,
1186 unsigned long, vlen)
1188 return do_writev(fd, vec, vlen, 0);
1191 SYSCALL_DEFINE5(preadv, unsigned long, fd, const struct iovec __user *, vec,
1192 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
1194 loff_t pos = pos_from_hilo(pos_h, pos_l);
1196 return do_preadv(fd, vec, vlen, pos, 0);
1199 SYSCALL_DEFINE6(preadv2, unsigned long, fd, const struct iovec __user *, vec,
1200 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h,
1203 loff_t pos = pos_from_hilo(pos_h, pos_l);
1206 return do_readv(fd, vec, vlen, flags);
1208 return do_preadv(fd, vec, vlen, pos, flags);
1211 SYSCALL_DEFINE5(pwritev, unsigned long, fd, const struct iovec __user *, vec,
1212 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
1214 loff_t pos = pos_from_hilo(pos_h, pos_l);
1216 return do_pwritev(fd, vec, vlen, pos, 0);
1219 SYSCALL_DEFINE6(pwritev2, unsigned long, fd, const struct iovec __user *, vec,
1220 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h,
1223 loff_t pos = pos_from_hilo(pos_h, pos_l);
1226 return do_writev(fd, vec, vlen, flags);
1228 return do_pwritev(fd, vec, vlen, pos, flags);
1231 #ifdef CONFIG_COMPAT
1232 static size_t compat_readv(struct file *file,
1233 const struct compat_iovec __user *vec,
1234 unsigned long vlen, loff_t *pos, rwf_t flags)
1236 struct iovec iovstack[UIO_FASTIOV];
1237 struct iovec *iov = iovstack;
1238 struct iov_iter iter;
1241 ret = compat_import_iovec(READ, vec, vlen, UIO_FASTIOV, &iov, &iter);
1243 ret = do_iter_read(file, &iter, pos, flags);
1247 add_rchar(current, ret);
1252 static size_t do_compat_readv(compat_ulong_t fd,
1253 const struct compat_iovec __user *vec,
1254 compat_ulong_t vlen, rwf_t flags)
1256 struct fd f = fdget_pos(fd);
1262 pos = f.file->f_pos;
1263 ret = compat_readv(f.file, vec, vlen, &pos, flags);
1265 f.file->f_pos = pos;
1271 COMPAT_SYSCALL_DEFINE3(readv, compat_ulong_t, fd,
1272 const struct compat_iovec __user *,vec,
1273 compat_ulong_t, vlen)
1275 return do_compat_readv(fd, vec, vlen, 0);
1278 static long do_compat_preadv64(unsigned long fd,
1279 const struct compat_iovec __user *vec,
1280 unsigned long vlen, loff_t pos, rwf_t flags)
1291 if (f.file->f_mode & FMODE_PREAD)
1292 ret = compat_readv(f.file, vec, vlen, &pos, flags);
1297 #ifdef __ARCH_WANT_COMPAT_SYS_PREADV64
1298 COMPAT_SYSCALL_DEFINE4(preadv64, unsigned long, fd,
1299 const struct compat_iovec __user *,vec,
1300 unsigned long, vlen, loff_t, pos)
1302 return do_compat_preadv64(fd, vec, vlen, pos, 0);
1306 COMPAT_SYSCALL_DEFINE5(preadv, compat_ulong_t, fd,
1307 const struct compat_iovec __user *,vec,
1308 compat_ulong_t, vlen, u32, pos_low, u32, pos_high)
1310 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1312 return do_compat_preadv64(fd, vec, vlen, pos, 0);
1315 #ifdef __ARCH_WANT_COMPAT_SYS_PREADV64V2
1316 COMPAT_SYSCALL_DEFINE5(preadv64v2, unsigned long, fd,
1317 const struct compat_iovec __user *,vec,
1318 unsigned long, vlen, loff_t, pos, rwf_t, flags)
1321 return do_compat_readv(fd, vec, vlen, flags);
1323 return do_compat_preadv64(fd, vec, vlen, pos, flags);
1327 COMPAT_SYSCALL_DEFINE6(preadv2, compat_ulong_t, fd,
1328 const struct compat_iovec __user *,vec,
1329 compat_ulong_t, vlen, u32, pos_low, u32, pos_high,
1332 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1335 return do_compat_readv(fd, vec, vlen, flags);
1337 return do_compat_preadv64(fd, vec, vlen, pos, flags);
1340 static size_t compat_writev(struct file *file,
1341 const struct compat_iovec __user *vec,
1342 unsigned long vlen, loff_t *pos, rwf_t flags)
1344 struct iovec iovstack[UIO_FASTIOV];
1345 struct iovec *iov = iovstack;
1346 struct iov_iter iter;
1349 ret = compat_import_iovec(WRITE, vec, vlen, UIO_FASTIOV, &iov, &iter);
1351 file_start_write(file);
1352 ret = do_iter_write(file, &iter, pos, flags);
1353 file_end_write(file);
1357 add_wchar(current, ret);
1362 static size_t do_compat_writev(compat_ulong_t fd,
1363 const struct compat_iovec __user* vec,
1364 compat_ulong_t vlen, rwf_t flags)
1366 struct fd f = fdget_pos(fd);
1372 pos = f.file->f_pos;
1373 ret = compat_writev(f.file, vec, vlen, &pos, flags);
1375 f.file->f_pos = pos;
1380 COMPAT_SYSCALL_DEFINE3(writev, compat_ulong_t, fd,
1381 const struct compat_iovec __user *, vec,
1382 compat_ulong_t, vlen)
1384 return do_compat_writev(fd, vec, vlen, 0);
1387 static long do_compat_pwritev64(unsigned long fd,
1388 const struct compat_iovec __user *vec,
1389 unsigned long vlen, loff_t pos, rwf_t flags)
1400 if (f.file->f_mode & FMODE_PWRITE)
1401 ret = compat_writev(f.file, vec, vlen, &pos, flags);
1406 #ifdef __ARCH_WANT_COMPAT_SYS_PWRITEV64
1407 COMPAT_SYSCALL_DEFINE4(pwritev64, unsigned long, fd,
1408 const struct compat_iovec __user *,vec,
1409 unsigned long, vlen, loff_t, pos)
1411 return do_compat_pwritev64(fd, vec, vlen, pos, 0);
1415 COMPAT_SYSCALL_DEFINE5(pwritev, compat_ulong_t, fd,
1416 const struct compat_iovec __user *,vec,
1417 compat_ulong_t, vlen, u32, pos_low, u32, pos_high)
1419 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1421 return do_compat_pwritev64(fd, vec, vlen, pos, 0);
1424 #ifdef __ARCH_WANT_COMPAT_SYS_PWRITEV64V2
1425 COMPAT_SYSCALL_DEFINE5(pwritev64v2, unsigned long, fd,
1426 const struct compat_iovec __user *,vec,
1427 unsigned long, vlen, loff_t, pos, rwf_t, flags)
1430 return do_compat_writev(fd, vec, vlen, flags);
1432 return do_compat_pwritev64(fd, vec, vlen, pos, flags);
1436 COMPAT_SYSCALL_DEFINE6(pwritev2, compat_ulong_t, fd,
1437 const struct compat_iovec __user *,vec,
1438 compat_ulong_t, vlen, u32, pos_low, u32, pos_high, rwf_t, flags)
1440 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1443 return do_compat_writev(fd, vec, vlen, flags);
1445 return do_compat_pwritev64(fd, vec, vlen, pos, flags);
1450 static ssize_t do_sendfile(int out_fd, int in_fd, loff_t *ppos,
1451 size_t count, loff_t max)
1454 struct inode *in_inode, *out_inode;
1461 * Get input file, and verify that it is ok..
1467 if (!(in.file->f_mode & FMODE_READ))
1471 pos = in.file->f_pos;
1474 if (!(in.file->f_mode & FMODE_PREAD))
1477 retval = rw_verify_area(READ, in.file, &pos, count);
1480 if (count > MAX_RW_COUNT)
1481 count = MAX_RW_COUNT;
1484 * Get output file, and verify that it is ok..
1487 out = fdget(out_fd);
1490 if (!(out.file->f_mode & FMODE_WRITE))
1492 in_inode = file_inode(in.file);
1493 out_inode = file_inode(out.file);
1494 out_pos = out.file->f_pos;
1495 retval = rw_verify_area(WRITE, out.file, &out_pos, count);
1500 max = min(in_inode->i_sb->s_maxbytes, out_inode->i_sb->s_maxbytes);
1502 if (unlikely(pos + count > max)) {
1503 retval = -EOVERFLOW;
1512 * We need to debate whether we can enable this or not. The
1513 * man page documents EAGAIN return for the output at least,
1514 * and the application is arguably buggy if it doesn't expect
1515 * EAGAIN on a non-blocking file descriptor.
1517 if (in.file->f_flags & O_NONBLOCK)
1518 fl = SPLICE_F_NONBLOCK;
1520 file_start_write(out.file);
1521 retval = do_splice_direct(in.file, &pos, out.file, &out_pos, count, fl);
1522 file_end_write(out.file);
1525 add_rchar(current, retval);
1526 add_wchar(current, retval);
1527 fsnotify_access(in.file);
1528 fsnotify_modify(out.file);
1529 out.file->f_pos = out_pos;
1533 in.file->f_pos = pos;
1539 retval = -EOVERFLOW;
1549 SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd, off_t __user *, offset, size_t, count)
1556 if (unlikely(get_user(off, offset)))
1559 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
1560 if (unlikely(put_user(pos, offset)))
1565 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1568 SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd, loff_t __user *, offset, size_t, count)
1574 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
1576 ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
1577 if (unlikely(put_user(pos, offset)))
1582 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1585 #ifdef CONFIG_COMPAT
1586 COMPAT_SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd,
1587 compat_off_t __user *, offset, compat_size_t, count)
1594 if (unlikely(get_user(off, offset)))
1597 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
1598 if (unlikely(put_user(pos, offset)))
1603 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1606 COMPAT_SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd,
1607 compat_loff_t __user *, offset, compat_size_t, count)
1613 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
1615 ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
1616 if (unlikely(put_user(pos, offset)))
1621 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1626 * generic_copy_file_range - copy data between two files
1627 * @file_in: file structure to read from
1628 * @pos_in: file offset to read from
1629 * @file_out: file structure to write data to
1630 * @pos_out: file offset to write data to
1631 * @len: amount of data to copy
1632 * @flags: copy flags
1634 * This is a generic filesystem helper to copy data from one file to another.
1635 * It has no constraints on the source or destination file owners - the files
1636 * can belong to different superblocks and different filesystem types. Short
1637 * copies are allowed.
1639 * This should be called from the @file_out filesystem, as per the
1640 * ->copy_file_range() method.
1642 * Returns the number of bytes copied or a negative error indicating the
1646 ssize_t generic_copy_file_range(struct file *file_in, loff_t pos_in,
1647 struct file *file_out, loff_t pos_out,
1648 size_t len, unsigned int flags)
1650 return do_splice_direct(file_in, &pos_in, file_out, &pos_out,
1651 len > MAX_RW_COUNT ? MAX_RW_COUNT : len, 0);
1653 EXPORT_SYMBOL(generic_copy_file_range);
1655 static ssize_t do_copy_file_range(struct file *file_in, loff_t pos_in,
1656 struct file *file_out, loff_t pos_out,
1657 size_t len, unsigned int flags)
1660 * Although we now allow filesystems to handle cross sb copy, passing
1661 * a file of the wrong filesystem type to filesystem driver can result
1662 * in an attempt to dereference the wrong type of ->private_data, so
1663 * avoid doing that until we really have a good reason. NFS defines
1664 * several different file_system_type structures, but they all end up
1665 * using the same ->copy_file_range() function pointer.
1667 if (file_out->f_op->copy_file_range &&
1668 file_out->f_op->copy_file_range == file_in->f_op->copy_file_range)
1669 return file_out->f_op->copy_file_range(file_in, pos_in,
1673 return generic_copy_file_range(file_in, pos_in, file_out, pos_out, len,
1678 * copy_file_range() differs from regular file read and write in that it
1679 * specifically allows return partial success. When it does so is up to
1680 * the copy_file_range method.
1682 ssize_t vfs_copy_file_range(struct file *file_in, loff_t pos_in,
1683 struct file *file_out, loff_t pos_out,
1684 size_t len, unsigned int flags)
1691 ret = generic_copy_file_checks(file_in, pos_in, file_out, pos_out, &len,
1696 ret = rw_verify_area(READ, file_in, &pos_in, len);
1700 ret = rw_verify_area(WRITE, file_out, &pos_out, len);
1707 file_start_write(file_out);
1710 * Try cloning first, this is supported by more file systems, and
1711 * more efficient if both clone and copy are supported (e.g. NFS).
1713 if (file_in->f_op->remap_file_range &&
1714 file_inode(file_in)->i_sb == file_inode(file_out)->i_sb) {
1717 cloned = file_in->f_op->remap_file_range(file_in, pos_in,
1719 min_t(loff_t, MAX_RW_COUNT, len),
1720 REMAP_FILE_CAN_SHORTEN);
1727 ret = do_copy_file_range(file_in, pos_in, file_out, pos_out, len,
1729 WARN_ON_ONCE(ret == -EOPNOTSUPP);
1732 fsnotify_access(file_in);
1733 add_rchar(current, ret);
1734 fsnotify_modify(file_out);
1735 add_wchar(current, ret);
1741 file_end_write(file_out);
1745 EXPORT_SYMBOL(vfs_copy_file_range);
1747 SYSCALL_DEFINE6(copy_file_range, int, fd_in, loff_t __user *, off_in,
1748 int, fd_out, loff_t __user *, off_out,
1749 size_t, len, unsigned int, flags)
1755 ssize_t ret = -EBADF;
1757 f_in = fdget(fd_in);
1761 f_out = fdget(fd_out);
1767 if (copy_from_user(&pos_in, off_in, sizeof(loff_t)))
1770 pos_in = f_in.file->f_pos;
1774 if (copy_from_user(&pos_out, off_out, sizeof(loff_t)))
1777 pos_out = f_out.file->f_pos;
1780 ret = vfs_copy_file_range(f_in.file, pos_in, f_out.file, pos_out, len,
1787 if (copy_to_user(off_in, &pos_in, sizeof(loff_t)))
1790 f_in.file->f_pos = pos_in;
1794 if (copy_to_user(off_out, &pos_out, sizeof(loff_t)))
1797 f_out.file->f_pos = pos_out;
1809 static int remap_verify_area(struct file *file, loff_t pos, loff_t len,
1812 struct inode *inode = file_inode(file);
1814 if (unlikely(pos < 0 || len < 0))
1817 if (unlikely((loff_t) (pos + len) < 0))
1820 if (unlikely(inode->i_flctx && mandatory_lock(inode))) {
1821 loff_t end = len ? pos + len - 1 : OFFSET_MAX;
1824 retval = locks_mandatory_area(inode, file, pos, end,
1825 write ? F_WRLCK : F_RDLCK);
1830 return security_file_permission(file, write ? MAY_WRITE : MAY_READ);
1833 * Ensure that we don't remap a partial EOF block in the middle of something
1834 * else. Assume that the offsets have already been checked for block
1837 * For clone we only link a partial EOF block above or at the destination file's
1838 * EOF. For deduplication we accept a partial EOF block only if it ends at the
1839 * destination file's EOF (can not link it into the middle of a file).
1841 * Shorten the request if possible.
1843 static int generic_remap_check_len(struct inode *inode_in,
1844 struct inode *inode_out,
1847 unsigned int remap_flags)
1849 u64 blkmask = i_blocksize(inode_in) - 1;
1850 loff_t new_len = *len;
1852 if ((*len & blkmask) == 0)
1855 if (pos_out + *len < i_size_read(inode_out))
1856 new_len &= ~blkmask;
1858 if (new_len == *len)
1861 if (remap_flags & REMAP_FILE_CAN_SHORTEN) {
1866 return (remap_flags & REMAP_FILE_DEDUP) ? -EBADE : -EINVAL;
1869 /* Read a page's worth of file data into the page cache. */
1870 static struct page *vfs_dedupe_get_page(struct inode *inode, loff_t offset)
1874 page = read_mapping_page(inode->i_mapping, offset >> PAGE_SHIFT, NULL);
1877 if (!PageUptodate(page)) {
1879 return ERR_PTR(-EIO);
1885 * Lock two pages, ensuring that we lock in offset order if the pages are from
1888 static void vfs_lock_two_pages(struct page *page1, struct page *page2)
1890 /* Always lock in order of increasing index. */
1891 if (page1->index > page2->index)
1899 /* Unlock two pages, being careful not to unlock the same page twice. */
1900 static void vfs_unlock_two_pages(struct page *page1, struct page *page2)
1908 * Compare extents of two files to see if they are the same.
1909 * Caller must have locked both inodes to prevent write races.
1911 static int vfs_dedupe_file_range_compare(struct inode *src, loff_t srcoff,
1912 struct inode *dest, loff_t destoff,
1913 loff_t len, bool *is_same)
1919 struct page *src_page;
1920 struct page *dest_page;
1928 src_poff = srcoff & (PAGE_SIZE - 1);
1929 dest_poff = destoff & (PAGE_SIZE - 1);
1930 cmp_len = min(PAGE_SIZE - src_poff,
1931 PAGE_SIZE - dest_poff);
1932 cmp_len = min(cmp_len, len);
1936 src_page = vfs_dedupe_get_page(src, srcoff);
1937 if (IS_ERR(src_page)) {
1938 error = PTR_ERR(src_page);
1941 dest_page = vfs_dedupe_get_page(dest, destoff);
1942 if (IS_ERR(dest_page)) {
1943 error = PTR_ERR(dest_page);
1948 vfs_lock_two_pages(src_page, dest_page);
1951 * Now that we've locked both pages, make sure they're still
1952 * mapped to the file data we're interested in. If not,
1953 * someone is invalidating pages on us and we lose.
1955 if (!PageUptodate(src_page) || !PageUptodate(dest_page) ||
1956 src_page->mapping != src->i_mapping ||
1957 dest_page->mapping != dest->i_mapping) {
1962 src_addr = kmap_atomic(src_page);
1963 dest_addr = kmap_atomic(dest_page);
1965 flush_dcache_page(src_page);
1966 flush_dcache_page(dest_page);
1968 if (memcmp(src_addr + src_poff, dest_addr + dest_poff, cmp_len))
1971 kunmap_atomic(dest_addr);
1972 kunmap_atomic(src_addr);
1974 vfs_unlock_two_pages(src_page, dest_page);
1975 put_page(dest_page);
1994 * Check that the two inodes are eligible for cloning, the ranges make
1995 * sense, and then flush all dirty data. Caller must ensure that the
1996 * inodes have been locked against any other modifications.
1998 * If there's an error, then the usual negative error code is returned.
1999 * Otherwise returns 0 with *len set to the request length.
2001 int generic_remap_file_range_prep(struct file *file_in, loff_t pos_in,
2002 struct file *file_out, loff_t pos_out,
2003 loff_t *len, unsigned int remap_flags)
2005 struct inode *inode_in = file_inode(file_in);
2006 struct inode *inode_out = file_inode(file_out);
2007 bool same_inode = (inode_in == inode_out);
2010 /* Don't touch certain kinds of inodes */
2011 if (IS_IMMUTABLE(inode_out))
2014 if (IS_SWAPFILE(inode_in) || IS_SWAPFILE(inode_out))
2017 /* Don't reflink dirs, pipes, sockets... */
2018 if (S_ISDIR(inode_in->i_mode) || S_ISDIR(inode_out->i_mode))
2020 if (!S_ISREG(inode_in->i_mode) || !S_ISREG(inode_out->i_mode))
2023 /* Zero length dedupe exits immediately; reflink goes to EOF. */
2025 loff_t isize = i_size_read(inode_in);
2027 if ((remap_flags & REMAP_FILE_DEDUP) || pos_in == isize)
2031 *len = isize - pos_in;
2036 /* Check that we don't violate system file offset limits. */
2037 ret = generic_remap_checks(file_in, pos_in, file_out, pos_out, len,
2042 /* Wait for the completion of any pending IOs on both files */
2043 inode_dio_wait(inode_in);
2045 inode_dio_wait(inode_out);
2047 ret = filemap_write_and_wait_range(inode_in->i_mapping,
2048 pos_in, pos_in + *len - 1);
2052 ret = filemap_write_and_wait_range(inode_out->i_mapping,
2053 pos_out, pos_out + *len - 1);
2058 * Check that the extents are the same.
2060 if (remap_flags & REMAP_FILE_DEDUP) {
2061 bool is_same = false;
2063 ret = vfs_dedupe_file_range_compare(inode_in, pos_in,
2064 inode_out, pos_out, *len, &is_same);
2071 ret = generic_remap_check_len(inode_in, inode_out, pos_out, len,
2076 /* If can't alter the file contents, we're done. */
2077 if (!(remap_flags & REMAP_FILE_DEDUP))
2078 ret = file_modified(file_out);
2082 EXPORT_SYMBOL(generic_remap_file_range_prep);
2084 loff_t do_clone_file_range(struct file *file_in, loff_t pos_in,
2085 struct file *file_out, loff_t pos_out,
2086 loff_t len, unsigned int remap_flags)
2090 WARN_ON_ONCE(remap_flags & REMAP_FILE_DEDUP);
2093 * FICLONE/FICLONERANGE ioctls enforce that src and dest files are on
2094 * the same mount. Practically, they only need to be on the same file
2097 if (file_inode(file_in)->i_sb != file_inode(file_out)->i_sb)
2100 ret = generic_file_rw_checks(file_in, file_out);
2104 if (!file_in->f_op->remap_file_range)
2107 ret = remap_verify_area(file_in, pos_in, len, false);
2111 ret = remap_verify_area(file_out, pos_out, len, true);
2115 ret = file_in->f_op->remap_file_range(file_in, pos_in,
2116 file_out, pos_out, len, remap_flags);
2120 fsnotify_access(file_in);
2121 fsnotify_modify(file_out);
2124 EXPORT_SYMBOL(do_clone_file_range);
2126 loff_t vfs_clone_file_range(struct file *file_in, loff_t pos_in,
2127 struct file *file_out, loff_t pos_out,
2128 loff_t len, unsigned int remap_flags)
2132 file_start_write(file_out);
2133 ret = do_clone_file_range(file_in, pos_in, file_out, pos_out, len,
2135 file_end_write(file_out);
2139 EXPORT_SYMBOL(vfs_clone_file_range);
2141 /* Check whether we are allowed to dedupe the destination file */
2142 static bool allow_file_dedupe(struct file *file)
2144 if (capable(CAP_SYS_ADMIN))
2146 if (file->f_mode & FMODE_WRITE)
2148 if (uid_eq(current_fsuid(), file_inode(file)->i_uid))
2150 if (!inode_permission(file_inode(file), MAY_WRITE))
2155 loff_t vfs_dedupe_file_range_one(struct file *src_file, loff_t src_pos,
2156 struct file *dst_file, loff_t dst_pos,
2157 loff_t len, unsigned int remap_flags)
2161 WARN_ON_ONCE(remap_flags & ~(REMAP_FILE_DEDUP |
2162 REMAP_FILE_CAN_SHORTEN));
2164 ret = mnt_want_write_file(dst_file);
2168 ret = remap_verify_area(dst_file, dst_pos, len, true);
2170 goto out_drop_write;
2173 if (!allow_file_dedupe(dst_file))
2174 goto out_drop_write;
2177 if (src_file->f_path.mnt != dst_file->f_path.mnt)
2178 goto out_drop_write;
2181 if (S_ISDIR(file_inode(dst_file)->i_mode))
2182 goto out_drop_write;
2185 if (!dst_file->f_op->remap_file_range)
2186 goto out_drop_write;
2190 goto out_drop_write;
2193 ret = dst_file->f_op->remap_file_range(src_file, src_pos, dst_file,
2194 dst_pos, len, remap_flags | REMAP_FILE_DEDUP);
2196 mnt_drop_write_file(dst_file);
2200 EXPORT_SYMBOL(vfs_dedupe_file_range_one);
2202 int vfs_dedupe_file_range(struct file *file, struct file_dedupe_range *same)
2204 struct file_dedupe_range_info *info;
2205 struct inode *src = file_inode(file);
2210 u16 count = same->dest_count;
2213 if (!(file->f_mode & FMODE_READ))
2216 if (same->reserved1 || same->reserved2)
2219 off = same->src_offset;
2220 len = same->src_length;
2222 if (S_ISDIR(src->i_mode))
2225 if (!S_ISREG(src->i_mode))
2228 if (!file->f_op->remap_file_range)
2231 ret = remap_verify_area(file, off, len, false);
2236 if (off + len > i_size_read(src))
2239 /* Arbitrary 1G limit on a single dedupe request, can be raised. */
2240 len = min_t(u64, len, 1 << 30);
2242 /* pre-format output fields to sane values */
2243 for (i = 0; i < count; i++) {
2244 same->info[i].bytes_deduped = 0ULL;
2245 same->info[i].status = FILE_DEDUPE_RANGE_SAME;
2248 for (i = 0, info = same->info; i < count; i++, info++) {
2249 struct fd dst_fd = fdget(info->dest_fd);
2250 struct file *dst_file = dst_fd.file;
2253 info->status = -EBADF;
2257 if (info->reserved) {
2258 info->status = -EINVAL;
2262 deduped = vfs_dedupe_file_range_one(file, off, dst_file,
2263 info->dest_offset, len,
2264 REMAP_FILE_CAN_SHORTEN);
2265 if (deduped == -EBADE)
2266 info->status = FILE_DEDUPE_RANGE_DIFFERS;
2267 else if (deduped < 0)
2268 info->status = deduped;
2270 info->bytes_deduped = len;
2275 if (fatal_signal_pending(current))
2280 EXPORT_SYMBOL(vfs_dedupe_file_range);