1 // SPDX-License-Identifier: GPL-2.0-only
2 #include <linux/slab.h>
3 #include <linux/stat.h>
4 #include <linux/sched/xacct.h>
5 #include <linux/fcntl.h>
6 #include <linux/file.h>
8 #include <linux/fsnotify.h>
9 #include <linux/security.h>
10 #include <linux/export.h>
11 #include <linux/syscalls.h>
12 #include <linux/pagemap.h>
13 #include <linux/splice.h>
14 #include <linux/compat.h>
15 #include <linux/mount.h>
19 #include <linux/uaccess.h>
20 #include <asm/unistd.h>
23 * Performs necessary checks before doing a clone.
25 * Can adjust amount of bytes to clone via @req_count argument.
26 * Returns appropriate error code that caller should return or
27 * zero in case the clone should be allowed.
29 static int generic_remap_checks(struct file *file_in, loff_t pos_in,
30 struct file *file_out, loff_t pos_out,
31 loff_t *req_count, unsigned int remap_flags)
33 struct inode *inode_in = file_in->f_mapping->host;
34 struct inode *inode_out = file_out->f_mapping->host;
35 uint64_t count = *req_count;
37 loff_t size_in, size_out;
38 loff_t bs = inode_out->i_sb->s_blocksize;
41 /* The start of both ranges must be aligned to an fs block. */
42 if (!IS_ALIGNED(pos_in, bs) || !IS_ALIGNED(pos_out, bs))
45 /* Ensure offsets don't wrap. */
46 if (pos_in + count < pos_in || pos_out + count < pos_out)
49 size_in = i_size_read(inode_in);
50 size_out = i_size_read(inode_out);
52 /* Dedupe requires both ranges to be within EOF. */
53 if ((remap_flags & REMAP_FILE_DEDUP) &&
54 (pos_in >= size_in || pos_in + count > size_in ||
55 pos_out >= size_out || pos_out + count > size_out))
58 /* Ensure the infile range is within the infile. */
59 if (pos_in >= size_in)
61 count = min(count, size_in - (uint64_t)pos_in);
63 ret = generic_write_check_limits(file_out, pos_out, &count);
68 * If the user wanted us to link to the infile's EOF, round up to the
69 * next block boundary for this check.
71 * Otherwise, make sure the count is also block-aligned, having
72 * already confirmed the starting offsets' block alignment.
74 if (pos_in + count == size_in &&
75 (!(remap_flags & REMAP_FILE_DEDUP) || pos_out + count == size_out)) {
76 bcount = ALIGN(size_in, bs) - pos_in;
78 if (!IS_ALIGNED(count, bs))
79 count = ALIGN_DOWN(count, bs);
83 /* Don't allow overlapped cloning within the same file. */
84 if (inode_in == inode_out &&
85 pos_out + bcount > pos_in &&
86 pos_out < pos_in + bcount)
90 * We shortened the request but the caller can't deal with that, so
91 * bounce the request back to userspace.
93 if (*req_count != count && !(remap_flags & REMAP_FILE_CAN_SHORTEN))
100 static int remap_verify_area(struct file *file, loff_t pos, loff_t len,
103 struct inode *inode = file_inode(file);
105 if (unlikely(pos < 0 || len < 0))
108 if (unlikely((loff_t) (pos + len) < 0))
111 if (unlikely(inode->i_flctx && mandatory_lock(inode))) {
112 loff_t end = len ? pos + len - 1 : OFFSET_MAX;
115 retval = locks_mandatory_area(inode, file, pos, end,
116 write ? F_WRLCK : F_RDLCK);
121 return security_file_permission(file, write ? MAY_WRITE : MAY_READ);
125 * Ensure that we don't remap a partial EOF block in the middle of something
126 * else. Assume that the offsets have already been checked for block
129 * For clone we only link a partial EOF block above or at the destination file's
130 * EOF. For deduplication we accept a partial EOF block only if it ends at the
131 * destination file's EOF (can not link it into the middle of a file).
133 * Shorten the request if possible.
135 static int generic_remap_check_len(struct inode *inode_in,
136 struct inode *inode_out,
139 unsigned int remap_flags)
141 u64 blkmask = i_blocksize(inode_in) - 1;
142 loff_t new_len = *len;
144 if ((*len & blkmask) == 0)
147 if (pos_out + *len < i_size_read(inode_out))
153 if (remap_flags & REMAP_FILE_CAN_SHORTEN) {
158 return (remap_flags & REMAP_FILE_DEDUP) ? -EBADE : -EINVAL;
161 /* Read a page's worth of file data into the page cache. */
162 static struct page *vfs_dedupe_get_page(struct inode *inode, loff_t offset)
166 page = read_mapping_page(inode->i_mapping, offset >> PAGE_SHIFT, NULL);
169 if (!PageUptodate(page)) {
171 return ERR_PTR(-EIO);
177 * Lock two pages, ensuring that we lock in offset order if the pages are from
180 static void vfs_lock_two_pages(struct page *page1, struct page *page2)
182 /* Always lock in order of increasing index. */
183 if (page1->index > page2->index)
191 /* Unlock two pages, being careful not to unlock the same page twice. */
192 static void vfs_unlock_two_pages(struct page *page1, struct page *page2)
200 * Compare extents of two files to see if they are the same.
201 * Caller must have locked both inodes to prevent write races.
203 static int vfs_dedupe_file_range_compare(struct inode *src, loff_t srcoff,
204 struct inode *dest, loff_t destoff,
205 loff_t len, bool *is_same)
211 struct page *src_page;
212 struct page *dest_page;
220 src_poff = srcoff & (PAGE_SIZE - 1);
221 dest_poff = destoff & (PAGE_SIZE - 1);
222 cmp_len = min(PAGE_SIZE - src_poff,
223 PAGE_SIZE - dest_poff);
224 cmp_len = min(cmp_len, len);
228 src_page = vfs_dedupe_get_page(src, srcoff);
229 if (IS_ERR(src_page)) {
230 error = PTR_ERR(src_page);
233 dest_page = vfs_dedupe_get_page(dest, destoff);
234 if (IS_ERR(dest_page)) {
235 error = PTR_ERR(dest_page);
240 vfs_lock_two_pages(src_page, dest_page);
243 * Now that we've locked both pages, make sure they're still
244 * mapped to the file data we're interested in. If not,
245 * someone is invalidating pages on us and we lose.
247 if (!PageUptodate(src_page) || !PageUptodate(dest_page) ||
248 src_page->mapping != src->i_mapping ||
249 dest_page->mapping != dest->i_mapping) {
254 src_addr = kmap_atomic(src_page);
255 dest_addr = kmap_atomic(dest_page);
257 flush_dcache_page(src_page);
258 flush_dcache_page(dest_page);
260 if (memcmp(src_addr + src_poff, dest_addr + dest_poff, cmp_len))
263 kunmap_atomic(dest_addr);
264 kunmap_atomic(src_addr);
266 vfs_unlock_two_pages(src_page, dest_page);
286 * Check that the two inodes are eligible for cloning, the ranges make
287 * sense, and then flush all dirty data. Caller must ensure that the
288 * inodes have been locked against any other modifications.
290 * If there's an error, then the usual negative error code is returned.
291 * Otherwise returns 0 with *len set to the request length.
293 int generic_remap_file_range_prep(struct file *file_in, loff_t pos_in,
294 struct file *file_out, loff_t pos_out,
295 loff_t *len, unsigned int remap_flags)
297 struct inode *inode_in = file_inode(file_in);
298 struct inode *inode_out = file_inode(file_out);
299 bool same_inode = (inode_in == inode_out);
302 /* Don't touch certain kinds of inodes */
303 if (IS_IMMUTABLE(inode_out))
306 if (IS_SWAPFILE(inode_in) || IS_SWAPFILE(inode_out))
309 /* Don't reflink dirs, pipes, sockets... */
310 if (S_ISDIR(inode_in->i_mode) || S_ISDIR(inode_out->i_mode))
312 if (!S_ISREG(inode_in->i_mode) || !S_ISREG(inode_out->i_mode))
315 /* Zero length dedupe exits immediately; reflink goes to EOF. */
317 loff_t isize = i_size_read(inode_in);
319 if ((remap_flags & REMAP_FILE_DEDUP) || pos_in == isize)
323 *len = isize - pos_in;
328 /* Check that we don't violate system file offset limits. */
329 ret = generic_remap_checks(file_in, pos_in, file_out, pos_out, len,
334 /* Wait for the completion of any pending IOs on both files */
335 inode_dio_wait(inode_in);
337 inode_dio_wait(inode_out);
339 ret = filemap_write_and_wait_range(inode_in->i_mapping,
340 pos_in, pos_in + *len - 1);
344 ret = filemap_write_and_wait_range(inode_out->i_mapping,
345 pos_out, pos_out + *len - 1);
350 * Check that the extents are the same.
352 if (remap_flags & REMAP_FILE_DEDUP) {
353 bool is_same = false;
355 ret = vfs_dedupe_file_range_compare(inode_in, pos_in,
356 inode_out, pos_out, *len, &is_same);
363 ret = generic_remap_check_len(inode_in, inode_out, pos_out, len,
368 /* If can't alter the file contents, we're done. */
369 if (!(remap_flags & REMAP_FILE_DEDUP))
370 ret = file_modified(file_out);
374 EXPORT_SYMBOL(generic_remap_file_range_prep);
376 loff_t do_clone_file_range(struct file *file_in, loff_t pos_in,
377 struct file *file_out, loff_t pos_out,
378 loff_t len, unsigned int remap_flags)
382 WARN_ON_ONCE(remap_flags & REMAP_FILE_DEDUP);
385 * FICLONE/FICLONERANGE ioctls enforce that src and dest files are on
386 * the same mount. Practically, they only need to be on the same file
389 if (file_inode(file_in)->i_sb != file_inode(file_out)->i_sb)
392 ret = generic_file_rw_checks(file_in, file_out);
396 if (!file_in->f_op->remap_file_range)
399 ret = remap_verify_area(file_in, pos_in, len, false);
403 ret = remap_verify_area(file_out, pos_out, len, true);
407 ret = file_in->f_op->remap_file_range(file_in, pos_in,
408 file_out, pos_out, len, remap_flags);
412 fsnotify_access(file_in);
413 fsnotify_modify(file_out);
416 EXPORT_SYMBOL(do_clone_file_range);
418 loff_t vfs_clone_file_range(struct file *file_in, loff_t pos_in,
419 struct file *file_out, loff_t pos_out,
420 loff_t len, unsigned int remap_flags)
424 file_start_write(file_out);
425 ret = do_clone_file_range(file_in, pos_in, file_out, pos_out, len,
427 file_end_write(file_out);
431 EXPORT_SYMBOL(vfs_clone_file_range);
433 /* Check whether we are allowed to dedupe the destination file */
434 static bool allow_file_dedupe(struct file *file)
436 if (capable(CAP_SYS_ADMIN))
438 if (file->f_mode & FMODE_WRITE)
440 if (uid_eq(current_fsuid(), file_inode(file)->i_uid))
442 if (!inode_permission(file_inode(file), MAY_WRITE))
447 loff_t vfs_dedupe_file_range_one(struct file *src_file, loff_t src_pos,
448 struct file *dst_file, loff_t dst_pos,
449 loff_t len, unsigned int remap_flags)
453 WARN_ON_ONCE(remap_flags & ~(REMAP_FILE_DEDUP |
454 REMAP_FILE_CAN_SHORTEN));
456 ret = mnt_want_write_file(dst_file);
460 ret = remap_verify_area(dst_file, dst_pos, len, true);
465 if (!allow_file_dedupe(dst_file))
469 if (src_file->f_path.mnt != dst_file->f_path.mnt)
473 if (S_ISDIR(file_inode(dst_file)->i_mode))
477 if (!dst_file->f_op->remap_file_range)
485 ret = dst_file->f_op->remap_file_range(src_file, src_pos, dst_file,
486 dst_pos, len, remap_flags | REMAP_FILE_DEDUP);
488 mnt_drop_write_file(dst_file);
492 EXPORT_SYMBOL(vfs_dedupe_file_range_one);
494 int vfs_dedupe_file_range(struct file *file, struct file_dedupe_range *same)
496 struct file_dedupe_range_info *info;
497 struct inode *src = file_inode(file);
502 u16 count = same->dest_count;
505 if (!(file->f_mode & FMODE_READ))
508 if (same->reserved1 || same->reserved2)
511 off = same->src_offset;
512 len = same->src_length;
514 if (S_ISDIR(src->i_mode))
517 if (!S_ISREG(src->i_mode))
520 if (!file->f_op->remap_file_range)
523 ret = remap_verify_area(file, off, len, false);
528 if (off + len > i_size_read(src))
531 /* Arbitrary 1G limit on a single dedupe request, can be raised. */
532 len = min_t(u64, len, 1 << 30);
534 /* pre-format output fields to sane values */
535 for (i = 0; i < count; i++) {
536 same->info[i].bytes_deduped = 0ULL;
537 same->info[i].status = FILE_DEDUPE_RANGE_SAME;
540 for (i = 0, info = same->info; i < count; i++, info++) {
541 struct fd dst_fd = fdget(info->dest_fd);
542 struct file *dst_file = dst_fd.file;
545 info->status = -EBADF;
549 if (info->reserved) {
550 info->status = -EINVAL;
554 deduped = vfs_dedupe_file_range_one(file, off, dst_file,
555 info->dest_offset, len,
556 REMAP_FILE_CAN_SHORTEN);
557 if (deduped == -EBADE)
558 info->status = FILE_DEDUPE_RANGE_DIFFERS;
559 else if (deduped < 0)
560 info->status = deduped;
562 info->bytes_deduped = len;
567 if (fatal_signal_pending(current))
572 EXPORT_SYMBOL(vfs_dedupe_file_range);