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>
17 #include <linux/dax.h>
18 #include <linux/overflow.h>
21 #include <linux/uaccess.h>
22 #include <asm/unistd.h>
25 * Performs necessary checks before doing a clone.
27 * Can adjust amount of bytes to clone via @req_count argument.
28 * Returns appropriate error code that caller should return or
29 * zero in case the clone should be allowed.
31 static int generic_remap_checks(struct file *file_in, loff_t pos_in,
32 struct file *file_out, loff_t pos_out,
33 loff_t *req_count, unsigned int remap_flags)
35 struct inode *inode_in = file_in->f_mapping->host;
36 struct inode *inode_out = file_out->f_mapping->host;
37 uint64_t count = *req_count;
39 loff_t size_in, size_out;
40 loff_t bs = inode_out->i_sb->s_blocksize;
43 /* The start of both ranges must be aligned to an fs block. */
44 if (!IS_ALIGNED(pos_in, bs) || !IS_ALIGNED(pos_out, bs))
47 /* Ensure offsets don't wrap. */
48 if (pos_in + count < pos_in || pos_out + count < pos_out)
51 size_in = i_size_read(inode_in);
52 size_out = i_size_read(inode_out);
54 /* Dedupe requires both ranges to be within EOF. */
55 if ((remap_flags & REMAP_FILE_DEDUP) &&
56 (pos_in >= size_in || pos_in + count > size_in ||
57 pos_out >= size_out || pos_out + count > size_out))
60 /* Ensure the infile range is within the infile. */
61 if (pos_in >= size_in)
63 count = min(count, size_in - (uint64_t)pos_in);
65 ret = generic_write_check_limits(file_out, pos_out, &count);
70 * If the user wanted us to link to the infile's EOF, round up to the
71 * next block boundary for this check.
73 * Otherwise, make sure the count is also block-aligned, having
74 * already confirmed the starting offsets' block alignment.
76 if (pos_in + count == size_in &&
77 (!(remap_flags & REMAP_FILE_DEDUP) || pos_out + count == size_out)) {
78 bcount = ALIGN(size_in, bs) - pos_in;
80 if (!IS_ALIGNED(count, bs))
81 count = ALIGN_DOWN(count, bs);
85 /* Don't allow overlapped cloning within the same file. */
86 if (inode_in == inode_out &&
87 pos_out + bcount > pos_in &&
88 pos_out < pos_in + bcount)
92 * We shortened the request but the caller can't deal with that, so
93 * bounce the request back to userspace.
95 if (*req_count != count && !(remap_flags & REMAP_FILE_CAN_SHORTEN))
102 static int remap_verify_area(struct file *file, loff_t pos, loff_t len,
107 if (unlikely(pos < 0 || len < 0))
110 if (unlikely(check_add_overflow(pos, len, &tmp)))
113 return security_file_permission(file, write ? MAY_WRITE : MAY_READ);
117 * Ensure that we don't remap a partial EOF block in the middle of something
118 * else. Assume that the offsets have already been checked for block
121 * For clone we only link a partial EOF block above or at the destination file's
122 * EOF. For deduplication we accept a partial EOF block only if it ends at the
123 * destination file's EOF (can not link it into the middle of a file).
125 * Shorten the request if possible.
127 static int generic_remap_check_len(struct inode *inode_in,
128 struct inode *inode_out,
131 unsigned int remap_flags)
133 u64 blkmask = i_blocksize(inode_in) - 1;
134 loff_t new_len = *len;
136 if ((*len & blkmask) == 0)
139 if (pos_out + *len < i_size_read(inode_out))
145 if (remap_flags & REMAP_FILE_CAN_SHORTEN) {
150 return (remap_flags & REMAP_FILE_DEDUP) ? -EBADE : -EINVAL;
153 /* Read a page's worth of file data into the page cache. */
154 static struct folio *vfs_dedupe_get_folio(struct file *file, loff_t pos)
156 return read_mapping_folio(file->f_mapping, pos >> PAGE_SHIFT, file);
160 * Lock two folios, ensuring that we lock in offset order if the folios
161 * are from the same file.
163 static void vfs_lock_two_folios(struct folio *folio1, struct folio *folio2)
165 /* Always lock in order of increasing index. */
166 if (folio1->index > folio2->index)
167 swap(folio1, folio2);
170 if (folio1 != folio2)
174 /* Unlock two folios, being careful not to unlock the same folio twice. */
175 static void vfs_unlock_two_folios(struct folio *folio1, struct folio *folio2)
177 folio_unlock(folio1);
178 if (folio1 != folio2)
179 folio_unlock(folio2);
183 * Compare extents of two files to see if they are the same.
184 * Caller must have locked both inodes to prevent write races.
186 static int vfs_dedupe_file_range_compare(struct file *src, loff_t srcoff,
187 struct file *dest, loff_t dstoff,
188 loff_t len, bool *is_same)
194 struct folio *src_folio, *dst_folio;
195 void *src_addr, *dst_addr;
196 loff_t cmp_len = min(PAGE_SIZE - offset_in_page(srcoff),
197 PAGE_SIZE - offset_in_page(dstoff));
199 cmp_len = min(cmp_len, len);
203 src_folio = vfs_dedupe_get_folio(src, srcoff);
204 if (IS_ERR(src_folio)) {
205 error = PTR_ERR(src_folio);
208 dst_folio = vfs_dedupe_get_folio(dest, dstoff);
209 if (IS_ERR(dst_folio)) {
210 error = PTR_ERR(dst_folio);
211 folio_put(src_folio);
215 vfs_lock_two_folios(src_folio, dst_folio);
218 * Now that we've locked both folios, make sure they're still
219 * mapped to the file data we're interested in. If not,
220 * someone is invalidating pages on us and we lose.
222 if (!folio_test_uptodate(src_folio) || !folio_test_uptodate(dst_folio) ||
223 src_folio->mapping != src->f_mapping ||
224 dst_folio->mapping != dest->f_mapping) {
229 src_addr = kmap_local_folio(src_folio,
230 offset_in_folio(src_folio, srcoff));
231 dst_addr = kmap_local_folio(dst_folio,
232 offset_in_folio(dst_folio, dstoff));
234 flush_dcache_folio(src_folio);
235 flush_dcache_folio(dst_folio);
237 if (memcmp(src_addr, dst_addr, cmp_len))
240 kunmap_local(dst_addr);
241 kunmap_local(src_addr);
243 vfs_unlock_two_folios(src_folio, dst_folio);
244 folio_put(dst_folio);
245 folio_put(src_folio);
263 * Check that the two inodes are eligible for cloning, the ranges make
264 * sense, and then flush all dirty data. Caller must ensure that the
265 * inodes have been locked against any other modifications.
267 * If there's an error, then the usual negative error code is returned.
268 * Otherwise returns 0 with *len set to the request length.
271 __generic_remap_file_range_prep(struct file *file_in, loff_t pos_in,
272 struct file *file_out, loff_t pos_out,
273 loff_t *len, unsigned int remap_flags,
274 const struct iomap_ops *dax_read_ops)
276 struct inode *inode_in = file_inode(file_in);
277 struct inode *inode_out = file_inode(file_out);
278 bool same_inode = (inode_in == inode_out);
281 /* Don't touch certain kinds of inodes */
282 if (IS_IMMUTABLE(inode_out))
285 if (IS_SWAPFILE(inode_in) || IS_SWAPFILE(inode_out))
288 /* Don't reflink dirs, pipes, sockets... */
289 if (S_ISDIR(inode_in->i_mode) || S_ISDIR(inode_out->i_mode))
291 if (!S_ISREG(inode_in->i_mode) || !S_ISREG(inode_out->i_mode))
294 /* Zero length dedupe exits immediately; reflink goes to EOF. */
296 loff_t isize = i_size_read(inode_in);
298 if ((remap_flags & REMAP_FILE_DEDUP) || pos_in == isize)
302 *len = isize - pos_in;
307 /* Check that we don't violate system file offset limits. */
308 ret = generic_remap_checks(file_in, pos_in, file_out, pos_out, len,
310 if (ret || *len == 0)
313 /* Wait for the completion of any pending IOs on both files */
314 inode_dio_wait(inode_in);
316 inode_dio_wait(inode_out);
318 ret = filemap_write_and_wait_range(inode_in->i_mapping,
319 pos_in, pos_in + *len - 1);
323 ret = filemap_write_and_wait_range(inode_out->i_mapping,
324 pos_out, pos_out + *len - 1);
329 * Check that the extents are the same.
331 if (remap_flags & REMAP_FILE_DEDUP) {
332 bool is_same = false;
334 if (!IS_DAX(inode_in))
335 ret = vfs_dedupe_file_range_compare(file_in, pos_in,
336 file_out, pos_out, *len, &is_same);
337 else if (dax_read_ops)
338 ret = dax_dedupe_file_range_compare(inode_in, pos_in,
339 inode_out, pos_out, *len, &is_same,
349 ret = generic_remap_check_len(inode_in, inode_out, pos_out, len,
351 if (ret || *len == 0)
354 /* If can't alter the file contents, we're done. */
355 if (!(remap_flags & REMAP_FILE_DEDUP))
356 ret = file_modified(file_out);
361 int generic_remap_file_range_prep(struct file *file_in, loff_t pos_in,
362 struct file *file_out, loff_t pos_out,
363 loff_t *len, unsigned int remap_flags)
365 return __generic_remap_file_range_prep(file_in, pos_in, file_out,
366 pos_out, len, remap_flags, NULL);
368 EXPORT_SYMBOL(generic_remap_file_range_prep);
370 loff_t do_clone_file_range(struct file *file_in, loff_t pos_in,
371 struct file *file_out, loff_t pos_out,
372 loff_t len, unsigned int remap_flags)
376 WARN_ON_ONCE(remap_flags & REMAP_FILE_DEDUP);
378 if (file_inode(file_in)->i_sb != file_inode(file_out)->i_sb)
381 ret = generic_file_rw_checks(file_in, file_out);
385 if (!file_in->f_op->remap_file_range)
388 ret = remap_verify_area(file_in, pos_in, len, false);
392 ret = remap_verify_area(file_out, pos_out, len, true);
396 ret = file_in->f_op->remap_file_range(file_in, pos_in,
397 file_out, pos_out, len, remap_flags);
401 fsnotify_access(file_in);
402 fsnotify_modify(file_out);
405 EXPORT_SYMBOL(do_clone_file_range);
407 loff_t vfs_clone_file_range(struct file *file_in, loff_t pos_in,
408 struct file *file_out, loff_t pos_out,
409 loff_t len, unsigned int remap_flags)
413 file_start_write(file_out);
414 ret = do_clone_file_range(file_in, pos_in, file_out, pos_out, len,
416 file_end_write(file_out);
420 EXPORT_SYMBOL(vfs_clone_file_range);
422 /* Check whether we are allowed to dedupe the destination file */
423 static bool allow_file_dedupe(struct file *file)
425 struct mnt_idmap *idmap = file_mnt_idmap(file);
426 struct inode *inode = file_inode(file);
428 if (capable(CAP_SYS_ADMIN))
430 if (file->f_mode & FMODE_WRITE)
432 if (vfsuid_eq_kuid(i_uid_into_vfsuid(idmap, inode), current_fsuid()))
434 if (!inode_permission(idmap, inode, MAY_WRITE))
439 loff_t vfs_dedupe_file_range_one(struct file *src_file, loff_t src_pos,
440 struct file *dst_file, loff_t dst_pos,
441 loff_t len, unsigned int remap_flags)
445 WARN_ON_ONCE(remap_flags & ~(REMAP_FILE_DEDUP |
446 REMAP_FILE_CAN_SHORTEN));
448 ret = mnt_want_write_file(dst_file);
453 * This is redundant if called from vfs_dedupe_file_range(), but other
454 * callers need it and it's not performance sesitive...
456 ret = remap_verify_area(src_file, src_pos, len, false);
460 ret = remap_verify_area(dst_file, dst_pos, len, true);
465 if (!allow_file_dedupe(dst_file))
469 if (file_inode(src_file)->i_sb != file_inode(dst_file)->i_sb)
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);