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>
20 #include <linux/uaccess.h>
21 #include <asm/unistd.h>
24 * Performs necessary checks before doing a clone.
26 * Can adjust amount of bytes to clone via @req_count argument.
27 * Returns appropriate error code that caller should return or
28 * zero in case the clone should be allowed.
30 static int generic_remap_checks(struct file *file_in, loff_t pos_in,
31 struct file *file_out, loff_t pos_out,
32 loff_t *req_count, unsigned int remap_flags)
34 struct inode *inode_in = file_in->f_mapping->host;
35 struct inode *inode_out = file_out->f_mapping->host;
36 uint64_t count = *req_count;
38 loff_t size_in, size_out;
39 loff_t bs = inode_out->i_sb->s_blocksize;
42 /* The start of both ranges must be aligned to an fs block. */
43 if (!IS_ALIGNED(pos_in, bs) || !IS_ALIGNED(pos_out, bs))
46 /* Ensure offsets don't wrap. */
47 if (pos_in + count < pos_in || pos_out + count < pos_out)
50 size_in = i_size_read(inode_in);
51 size_out = i_size_read(inode_out);
53 /* Dedupe requires both ranges to be within EOF. */
54 if ((remap_flags & REMAP_FILE_DEDUP) &&
55 (pos_in >= size_in || pos_in + count > size_in ||
56 pos_out >= size_out || pos_out + count > size_out))
59 /* Ensure the infile range is within the infile. */
60 if (pos_in >= size_in)
62 count = min(count, size_in - (uint64_t)pos_in);
64 ret = generic_write_check_limits(file_out, pos_out, &count);
69 * If the user wanted us to link to the infile's EOF, round up to the
70 * next block boundary for this check.
72 * Otherwise, make sure the count is also block-aligned, having
73 * already confirmed the starting offsets' block alignment.
75 if (pos_in + count == size_in &&
76 (!(remap_flags & REMAP_FILE_DEDUP) || pos_out + count == size_out)) {
77 bcount = ALIGN(size_in, bs) - pos_in;
79 if (!IS_ALIGNED(count, bs))
80 count = ALIGN_DOWN(count, bs);
84 /* Don't allow overlapped cloning within the same file. */
85 if (inode_in == inode_out &&
86 pos_out + bcount > pos_in &&
87 pos_out < pos_in + bcount)
91 * We shortened the request but the caller can't deal with that, so
92 * bounce the request back to userspace.
94 if (*req_count != count && !(remap_flags & REMAP_FILE_CAN_SHORTEN))
101 static int remap_verify_area(struct file *file, loff_t pos, loff_t len,
104 if (unlikely(pos < 0 || len < 0))
107 if (unlikely((loff_t) (pos + len) < 0))
110 return security_file_permission(file, write ? MAY_WRITE : MAY_READ);
114 * Ensure that we don't remap a partial EOF block in the middle of something
115 * else. Assume that the offsets have already been checked for block
118 * For clone we only link a partial EOF block above or at the destination file's
119 * EOF. For deduplication we accept a partial EOF block only if it ends at the
120 * destination file's EOF (can not link it into the middle of a file).
122 * Shorten the request if possible.
124 static int generic_remap_check_len(struct inode *inode_in,
125 struct inode *inode_out,
128 unsigned int remap_flags)
130 u64 blkmask = i_blocksize(inode_in) - 1;
131 loff_t new_len = *len;
133 if ((*len & blkmask) == 0)
136 if (pos_out + *len < i_size_read(inode_out))
142 if (remap_flags & REMAP_FILE_CAN_SHORTEN) {
147 return (remap_flags & REMAP_FILE_DEDUP) ? -EBADE : -EINVAL;
150 /* Read a page's worth of file data into the page cache. */
151 static struct folio *vfs_dedupe_get_folio(struct file *file, loff_t pos)
153 return read_mapping_folio(file->f_mapping, pos >> PAGE_SHIFT, file);
157 * Lock two folios, ensuring that we lock in offset order if the folios
158 * are from the same file.
160 static void vfs_lock_two_folios(struct folio *folio1, struct folio *folio2)
162 /* Always lock in order of increasing index. */
163 if (folio1->index > folio2->index)
164 swap(folio1, folio2);
167 if (folio1 != folio2)
171 /* Unlock two folios, being careful not to unlock the same folio twice. */
172 static void vfs_unlock_two_folios(struct folio *folio1, struct folio *folio2)
174 folio_unlock(folio1);
175 if (folio1 != folio2)
176 folio_unlock(folio2);
180 * Compare extents of two files to see if they are the same.
181 * Caller must have locked both inodes to prevent write races.
183 static int vfs_dedupe_file_range_compare(struct file *src, loff_t srcoff,
184 struct file *dest, loff_t dstoff,
185 loff_t len, bool *is_same)
191 struct folio *src_folio, *dst_folio;
192 void *src_addr, *dst_addr;
193 loff_t cmp_len = min(PAGE_SIZE - offset_in_page(srcoff),
194 PAGE_SIZE - offset_in_page(dstoff));
196 cmp_len = min(cmp_len, len);
200 src_folio = vfs_dedupe_get_folio(src, srcoff);
201 if (IS_ERR(src_folio)) {
202 error = PTR_ERR(src_folio);
205 dst_folio = vfs_dedupe_get_folio(dest, dstoff);
206 if (IS_ERR(dst_folio)) {
207 error = PTR_ERR(dst_folio);
208 folio_put(src_folio);
212 vfs_lock_two_folios(src_folio, dst_folio);
215 * Now that we've locked both folios, make sure they're still
216 * mapped to the file data we're interested in. If not,
217 * someone is invalidating pages on us and we lose.
219 if (!folio_test_uptodate(src_folio) || !folio_test_uptodate(dst_folio) ||
220 src_folio->mapping != src->f_mapping ||
221 dst_folio->mapping != dest->f_mapping) {
226 src_addr = kmap_local_folio(src_folio,
227 offset_in_folio(src_folio, srcoff));
228 dst_addr = kmap_local_folio(dst_folio,
229 offset_in_folio(dst_folio, dstoff));
231 flush_dcache_folio(src_folio);
232 flush_dcache_folio(dst_folio);
234 if (memcmp(src_addr, dst_addr, cmp_len))
237 kunmap_local(dst_addr);
238 kunmap_local(src_addr);
240 vfs_unlock_two_folios(src_folio, dst_folio);
241 folio_put(dst_folio);
242 folio_put(src_folio);
260 * Check that the two inodes are eligible for cloning, the ranges make
261 * sense, and then flush all dirty data. Caller must ensure that the
262 * inodes have been locked against any other modifications.
264 * If there's an error, then the usual negative error code is returned.
265 * Otherwise returns 0 with *len set to the request length.
268 __generic_remap_file_range_prep(struct file *file_in, loff_t pos_in,
269 struct file *file_out, loff_t pos_out,
270 loff_t *len, unsigned int remap_flags,
271 const struct iomap_ops *dax_read_ops)
273 struct inode *inode_in = file_inode(file_in);
274 struct inode *inode_out = file_inode(file_out);
275 bool same_inode = (inode_in == inode_out);
278 /* Don't touch certain kinds of inodes */
279 if (IS_IMMUTABLE(inode_out))
282 if (IS_SWAPFILE(inode_in) || IS_SWAPFILE(inode_out))
285 /* Don't reflink dirs, pipes, sockets... */
286 if (S_ISDIR(inode_in->i_mode) || S_ISDIR(inode_out->i_mode))
288 if (!S_ISREG(inode_in->i_mode) || !S_ISREG(inode_out->i_mode))
291 /* Zero length dedupe exits immediately; reflink goes to EOF. */
293 loff_t isize = i_size_read(inode_in);
295 if ((remap_flags & REMAP_FILE_DEDUP) || pos_in == isize)
299 *len = isize - pos_in;
304 /* Check that we don't violate system file offset limits. */
305 ret = generic_remap_checks(file_in, pos_in, file_out, pos_out, len,
307 if (ret || *len == 0)
310 /* Wait for the completion of any pending IOs on both files */
311 inode_dio_wait(inode_in);
313 inode_dio_wait(inode_out);
315 ret = filemap_write_and_wait_range(inode_in->i_mapping,
316 pos_in, pos_in + *len - 1);
320 ret = filemap_write_and_wait_range(inode_out->i_mapping,
321 pos_out, pos_out + *len - 1);
326 * Check that the extents are the same.
328 if (remap_flags & REMAP_FILE_DEDUP) {
329 bool is_same = false;
331 if (!IS_DAX(inode_in))
332 ret = vfs_dedupe_file_range_compare(file_in, pos_in,
333 file_out, pos_out, *len, &is_same);
334 else if (dax_read_ops)
335 ret = dax_dedupe_file_range_compare(inode_in, pos_in,
336 inode_out, pos_out, *len, &is_same,
346 ret = generic_remap_check_len(inode_in, inode_out, pos_out, len,
348 if (ret || *len == 0)
351 /* If can't alter the file contents, we're done. */
352 if (!(remap_flags & REMAP_FILE_DEDUP))
353 ret = file_modified(file_out);
358 int generic_remap_file_range_prep(struct file *file_in, loff_t pos_in,
359 struct file *file_out, loff_t pos_out,
360 loff_t *len, unsigned int remap_flags)
362 return __generic_remap_file_range_prep(file_in, pos_in, file_out,
363 pos_out, len, remap_flags, NULL);
365 EXPORT_SYMBOL(generic_remap_file_range_prep);
367 loff_t do_clone_file_range(struct file *file_in, loff_t pos_in,
368 struct file *file_out, loff_t pos_out,
369 loff_t len, unsigned int remap_flags)
373 WARN_ON_ONCE(remap_flags & REMAP_FILE_DEDUP);
375 if (file_inode(file_in)->i_sb != file_inode(file_out)->i_sb)
378 ret = generic_file_rw_checks(file_in, file_out);
382 if (!file_in->f_op->remap_file_range)
385 ret = remap_verify_area(file_in, pos_in, len, false);
389 ret = remap_verify_area(file_out, pos_out, len, true);
393 ret = file_in->f_op->remap_file_range(file_in, pos_in,
394 file_out, pos_out, len, remap_flags);
398 fsnotify_access(file_in);
399 fsnotify_modify(file_out);
402 EXPORT_SYMBOL(do_clone_file_range);
404 loff_t vfs_clone_file_range(struct file *file_in, loff_t pos_in,
405 struct file *file_out, loff_t pos_out,
406 loff_t len, unsigned int remap_flags)
410 file_start_write(file_out);
411 ret = do_clone_file_range(file_in, pos_in, file_out, pos_out, len,
413 file_end_write(file_out);
417 EXPORT_SYMBOL(vfs_clone_file_range);
419 /* Check whether we are allowed to dedupe the destination file */
420 static bool allow_file_dedupe(struct file *file)
422 struct user_namespace *mnt_userns = file_mnt_user_ns(file);
423 struct inode *inode = file_inode(file);
425 if (capable(CAP_SYS_ADMIN))
427 if (file->f_mode & FMODE_WRITE)
429 if (vfsuid_eq_kuid(i_uid_into_vfsuid(mnt_userns, inode), current_fsuid()))
431 if (!inode_permission(mnt_userns, inode, MAY_WRITE))
436 loff_t vfs_dedupe_file_range_one(struct file *src_file, loff_t src_pos,
437 struct file *dst_file, loff_t dst_pos,
438 loff_t len, unsigned int remap_flags)
442 WARN_ON_ONCE(remap_flags & ~(REMAP_FILE_DEDUP |
443 REMAP_FILE_CAN_SHORTEN));
445 ret = mnt_want_write_file(dst_file);
450 * This is redundant if called from vfs_dedupe_file_range(), but other
451 * callers need it and it's not performance sesitive...
453 ret = remap_verify_area(src_file, src_pos, len, false);
457 ret = remap_verify_area(dst_file, dst_pos, len, true);
462 if (!allow_file_dedupe(dst_file))
466 if (file_inode(src_file)->i_sb != file_inode(dst_file)->i_sb)
470 if (S_ISDIR(file_inode(dst_file)->i_mode))
474 if (!dst_file->f_op->remap_file_range)
482 ret = dst_file->f_op->remap_file_range(src_file, src_pos, dst_file,
483 dst_pos, len, remap_flags | REMAP_FILE_DEDUP);
485 mnt_drop_write_file(dst_file);
489 EXPORT_SYMBOL(vfs_dedupe_file_range_one);
491 int vfs_dedupe_file_range(struct file *file, struct file_dedupe_range *same)
493 struct file_dedupe_range_info *info;
494 struct inode *src = file_inode(file);
499 u16 count = same->dest_count;
502 if (!(file->f_mode & FMODE_READ))
505 if (same->reserved1 || same->reserved2)
508 off = same->src_offset;
509 len = same->src_length;
511 if (S_ISDIR(src->i_mode))
514 if (!S_ISREG(src->i_mode))
517 if (!file->f_op->remap_file_range)
520 ret = remap_verify_area(file, off, len, false);
525 if (off + len > i_size_read(src))
528 /* Arbitrary 1G limit on a single dedupe request, can be raised. */
529 len = min_t(u64, len, 1 << 30);
531 /* pre-format output fields to sane values */
532 for (i = 0; i < count; i++) {
533 same->info[i].bytes_deduped = 0ULL;
534 same->info[i].status = FILE_DEDUPE_RANGE_SAME;
537 for (i = 0, info = same->info; i < count; i++, info++) {
538 struct fd dst_fd = fdget(info->dest_fd);
539 struct file *dst_file = dst_fd.file;
542 info->status = -EBADF;
546 if (info->reserved) {
547 info->status = -EINVAL;
551 deduped = vfs_dedupe_file_range_one(file, off, dst_file,
552 info->dest_offset, len,
553 REMAP_FILE_CAN_SHORTEN);
554 if (deduped == -EBADE)
555 info->status = FILE_DEDUPE_RANGE_DIFFERS;
556 else if (deduped < 0)
557 info->status = deduped;
559 info->bytes_deduped = len;
564 if (fatal_signal_pending(current))
569 EXPORT_SYMBOL(vfs_dedupe_file_range);