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> |
7 | #include <linux/uio.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> |
16 | #include <linux/fs.h> |
17 | #include <linux/dax.h> |
18 | #include <linux/overflow.h> |
19 | #include "internal.h" |
20 | |
21 | #include <linux/uaccess.h> |
22 | #include <asm/unistd.h> |
23 | |
24 | /* |
25 | * Performs necessary checks before doing a clone. |
26 | * |
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. |
30 | */ |
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) |
34 | { |
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; |
38 | uint64_t bcount; |
39 | loff_t size_in, size_out; |
40 | loff_t bs = inode_out->i_sb->s_blocksize; |
41 | int ret; |
42 | |
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)) |
45 | return -EINVAL; |
46 | |
47 | /* Ensure offsets don't wrap. */ |
48 | if (pos_in + count < pos_in || pos_out + count < pos_out) |
49 | return -EINVAL; |
50 | |
51 | size_in = i_size_read(inode: inode_in); |
52 | size_out = i_size_read(inode: inode_out); |
53 | |
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)) |
58 | return -EINVAL; |
59 | |
60 | /* Ensure the infile range is within the infile. */ |
61 | if (pos_in >= size_in) |
62 | return -EINVAL; |
63 | count = min(count, size_in - (uint64_t)pos_in); |
64 | |
65 | ret = generic_write_check_limits(file: file_out, pos: pos_out, count: &count); |
66 | if (ret) |
67 | return ret; |
68 | |
69 | /* |
70 | * If the user wanted us to link to the infile's EOF, round up to the |
71 | * next block boundary for this check. |
72 | * |
73 | * Otherwise, make sure the count is also block-aligned, having |
74 | * already confirmed the starting offsets' block alignment. |
75 | */ |
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; |
79 | } else { |
80 | if (!IS_ALIGNED(count, bs)) |
81 | count = ALIGN_DOWN(count, bs); |
82 | bcount = count; |
83 | } |
84 | |
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) |
89 | return -EINVAL; |
90 | |
91 | /* |
92 | * We shortened the request but the caller can't deal with that, so |
93 | * bounce the request back to userspace. |
94 | */ |
95 | if (*req_count != count && !(remap_flags & REMAP_FILE_CAN_SHORTEN)) |
96 | return -EINVAL; |
97 | |
98 | *req_count = count; |
99 | return 0; |
100 | } |
101 | |
102 | static int remap_verify_area(struct file *file, loff_t pos, loff_t len, |
103 | bool write) |
104 | { |
105 | int mask = write ? MAY_WRITE : MAY_READ; |
106 | loff_t tmp; |
107 | int ret; |
108 | |
109 | if (unlikely(pos < 0 || len < 0)) |
110 | return -EINVAL; |
111 | |
112 | if (unlikely(check_add_overflow(pos, len, &tmp))) |
113 | return -EINVAL; |
114 | |
115 | ret = security_file_permission(file, mask); |
116 | if (ret) |
117 | return ret; |
118 | |
119 | return fsnotify_file_area_perm(file, perm_mask: mask, ppos: &pos, count: len); |
120 | } |
121 | |
122 | /* |
123 | * Ensure that we don't remap a partial EOF block in the middle of something |
124 | * else. Assume that the offsets have already been checked for block |
125 | * alignment. |
126 | * |
127 | * For clone we only link a partial EOF block above or at the destination file's |
128 | * EOF. For deduplication we accept a partial EOF block only if it ends at the |
129 | * destination file's EOF (can not link it into the middle of a file). |
130 | * |
131 | * Shorten the request if possible. |
132 | */ |
133 | static int generic_remap_check_len(struct inode *inode_in, |
134 | struct inode *inode_out, |
135 | loff_t pos_out, |
136 | loff_t *len, |
137 | unsigned int remap_flags) |
138 | { |
139 | u64 blkmask = i_blocksize(node: inode_in) - 1; |
140 | loff_t new_len = *len; |
141 | |
142 | if ((*len & blkmask) == 0) |
143 | return 0; |
144 | |
145 | if (pos_out + *len < i_size_read(inode: inode_out)) |
146 | new_len &= ~blkmask; |
147 | |
148 | if (new_len == *len) |
149 | return 0; |
150 | |
151 | if (remap_flags & REMAP_FILE_CAN_SHORTEN) { |
152 | *len = new_len; |
153 | return 0; |
154 | } |
155 | |
156 | return (remap_flags & REMAP_FILE_DEDUP) ? -EBADE : -EINVAL; |
157 | } |
158 | |
159 | /* Read a page's worth of file data into the page cache. */ |
160 | static struct folio *vfs_dedupe_get_folio(struct file *file, loff_t pos) |
161 | { |
162 | return read_mapping_folio(mapping: file->f_mapping, index: pos >> PAGE_SHIFT, file); |
163 | } |
164 | |
165 | /* |
166 | * Lock two folios, ensuring that we lock in offset order if the folios |
167 | * are from the same file. |
168 | */ |
169 | static void vfs_lock_two_folios(struct folio *folio1, struct folio *folio2) |
170 | { |
171 | /* Always lock in order of increasing index. */ |
172 | if (folio1->index > folio2->index) |
173 | swap(folio1, folio2); |
174 | |
175 | folio_lock(folio: folio1); |
176 | if (folio1 != folio2) |
177 | folio_lock(folio: folio2); |
178 | } |
179 | |
180 | /* Unlock two folios, being careful not to unlock the same folio twice. */ |
181 | static void vfs_unlock_two_folios(struct folio *folio1, struct folio *folio2) |
182 | { |
183 | folio_unlock(folio: folio1); |
184 | if (folio1 != folio2) |
185 | folio_unlock(folio: folio2); |
186 | } |
187 | |
188 | /* |
189 | * Compare extents of two files to see if they are the same. |
190 | * Caller must have locked both inodes to prevent write races. |
191 | */ |
192 | static int vfs_dedupe_file_range_compare(struct file *src, loff_t srcoff, |
193 | struct file *dest, loff_t dstoff, |
194 | loff_t len, bool *is_same) |
195 | { |
196 | bool same = true; |
197 | int error = -EINVAL; |
198 | |
199 | while (len) { |
200 | struct folio *src_folio, *dst_folio; |
201 | void *src_addr, *dst_addr; |
202 | loff_t cmp_len = min(PAGE_SIZE - offset_in_page(srcoff), |
203 | PAGE_SIZE - offset_in_page(dstoff)); |
204 | |
205 | cmp_len = min(cmp_len, len); |
206 | if (cmp_len <= 0) |
207 | goto out_error; |
208 | |
209 | src_folio = vfs_dedupe_get_folio(file: src, pos: srcoff); |
210 | if (IS_ERR(ptr: src_folio)) { |
211 | error = PTR_ERR(ptr: src_folio); |
212 | goto out_error; |
213 | } |
214 | dst_folio = vfs_dedupe_get_folio(file: dest, pos: dstoff); |
215 | if (IS_ERR(ptr: dst_folio)) { |
216 | error = PTR_ERR(ptr: dst_folio); |
217 | folio_put(folio: src_folio); |
218 | goto out_error; |
219 | } |
220 | |
221 | vfs_lock_two_folios(folio1: src_folio, folio2: dst_folio); |
222 | |
223 | /* |
224 | * Now that we've locked both folios, make sure they're still |
225 | * mapped to the file data we're interested in. If not, |
226 | * someone is invalidating pages on us and we lose. |
227 | */ |
228 | if (!folio_test_uptodate(folio: src_folio) || !folio_test_uptodate(folio: dst_folio) || |
229 | src_folio->mapping != src->f_mapping || |
230 | dst_folio->mapping != dest->f_mapping) { |
231 | same = false; |
232 | goto unlock; |
233 | } |
234 | |
235 | src_addr = kmap_local_folio(folio: src_folio, |
236 | offset_in_folio(src_folio, srcoff)); |
237 | dst_addr = kmap_local_folio(folio: dst_folio, |
238 | offset_in_folio(dst_folio, dstoff)); |
239 | |
240 | flush_dcache_folio(folio: src_folio); |
241 | flush_dcache_folio(folio: dst_folio); |
242 | |
243 | if (memcmp(p: src_addr, q: dst_addr, size: cmp_len)) |
244 | same = false; |
245 | |
246 | kunmap_local(dst_addr); |
247 | kunmap_local(src_addr); |
248 | unlock: |
249 | vfs_unlock_two_folios(folio1: src_folio, folio2: dst_folio); |
250 | folio_put(folio: dst_folio); |
251 | folio_put(folio: src_folio); |
252 | |
253 | if (!same) |
254 | break; |
255 | |
256 | srcoff += cmp_len; |
257 | dstoff += cmp_len; |
258 | len -= cmp_len; |
259 | } |
260 | |
261 | *is_same = same; |
262 | return 0; |
263 | |
264 | out_error: |
265 | return error; |
266 | } |
267 | |
268 | /* |
269 | * Check that the two inodes are eligible for cloning, the ranges make |
270 | * sense, and then flush all dirty data. Caller must ensure that the |
271 | * inodes have been locked against any other modifications. |
272 | * |
273 | * If there's an error, then the usual negative error code is returned. |
274 | * Otherwise returns 0 with *len set to the request length. |
275 | */ |
276 | int |
277 | __generic_remap_file_range_prep(struct file *file_in, loff_t pos_in, |
278 | struct file *file_out, loff_t pos_out, |
279 | loff_t *len, unsigned int remap_flags, |
280 | const struct iomap_ops *dax_read_ops) |
281 | { |
282 | struct inode *inode_in = file_inode(f: file_in); |
283 | struct inode *inode_out = file_inode(f: file_out); |
284 | bool same_inode = (inode_in == inode_out); |
285 | int ret; |
286 | |
287 | /* Don't touch certain kinds of inodes */ |
288 | if (IS_IMMUTABLE(inode_out)) |
289 | return -EPERM; |
290 | |
291 | if (IS_SWAPFILE(inode_in) || IS_SWAPFILE(inode_out)) |
292 | return -ETXTBSY; |
293 | |
294 | /* Don't reflink dirs, pipes, sockets... */ |
295 | if (S_ISDIR(inode_in->i_mode) || S_ISDIR(inode_out->i_mode)) |
296 | return -EISDIR; |
297 | if (!S_ISREG(inode_in->i_mode) || !S_ISREG(inode_out->i_mode)) |
298 | return -EINVAL; |
299 | |
300 | /* Zero length dedupe exits immediately; reflink goes to EOF. */ |
301 | if (*len == 0) { |
302 | loff_t isize = i_size_read(inode: inode_in); |
303 | |
304 | if ((remap_flags & REMAP_FILE_DEDUP) || pos_in == isize) |
305 | return 0; |
306 | if (pos_in > isize) |
307 | return -EINVAL; |
308 | *len = isize - pos_in; |
309 | if (*len == 0) |
310 | return 0; |
311 | } |
312 | |
313 | /* Check that we don't violate system file offset limits. */ |
314 | ret = generic_remap_checks(file_in, pos_in, file_out, pos_out, req_count: len, |
315 | remap_flags); |
316 | if (ret || *len == 0) |
317 | return ret; |
318 | |
319 | /* Wait for the completion of any pending IOs on both files */ |
320 | inode_dio_wait(inode: inode_in); |
321 | if (!same_inode) |
322 | inode_dio_wait(inode: inode_out); |
323 | |
324 | ret = filemap_write_and_wait_range(mapping: inode_in->i_mapping, |
325 | lstart: pos_in, lend: pos_in + *len - 1); |
326 | if (ret) |
327 | return ret; |
328 | |
329 | ret = filemap_write_and_wait_range(mapping: inode_out->i_mapping, |
330 | lstart: pos_out, lend: pos_out + *len - 1); |
331 | if (ret) |
332 | return ret; |
333 | |
334 | /* |
335 | * Check that the extents are the same. |
336 | */ |
337 | if (remap_flags & REMAP_FILE_DEDUP) { |
338 | bool is_same = false; |
339 | |
340 | if (!IS_DAX(inode_in)) |
341 | ret = vfs_dedupe_file_range_compare(src: file_in, srcoff: pos_in, |
342 | dest: file_out, dstoff: pos_out, len: *len, is_same: &is_same); |
343 | else if (dax_read_ops) |
344 | ret = dax_dedupe_file_range_compare(src: inode_in, srcoff: pos_in, |
345 | dest: inode_out, destoff: pos_out, len: *len, is_same: &is_same, |
346 | ops: dax_read_ops); |
347 | else |
348 | return -EINVAL; |
349 | if (ret) |
350 | return ret; |
351 | if (!is_same) |
352 | return -EBADE; |
353 | } |
354 | |
355 | ret = generic_remap_check_len(inode_in, inode_out, pos_out, len, |
356 | remap_flags); |
357 | if (ret || *len == 0) |
358 | return ret; |
359 | |
360 | /* If can't alter the file contents, we're done. */ |
361 | if (!(remap_flags & REMAP_FILE_DEDUP)) |
362 | ret = file_modified(file: file_out); |
363 | |
364 | return ret; |
365 | } |
366 | |
367 | int generic_remap_file_range_prep(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) |
370 | { |
371 | return __generic_remap_file_range_prep(file_in, pos_in, file_out, |
372 | pos_out, len, remap_flags, NULL); |
373 | } |
374 | EXPORT_SYMBOL(generic_remap_file_range_prep); |
375 | |
376 | loff_t vfs_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) |
379 | { |
380 | loff_t ret; |
381 | |
382 | WARN_ON_ONCE(remap_flags & REMAP_FILE_DEDUP); |
383 | |
384 | if (file_inode(f: file_in)->i_sb != file_inode(f: file_out)->i_sb) |
385 | return -EXDEV; |
386 | |
387 | ret = generic_file_rw_checks(file_in, file_out); |
388 | if (ret < 0) |
389 | return ret; |
390 | |
391 | if (!file_in->f_op->remap_file_range) |
392 | return -EOPNOTSUPP; |
393 | |
394 | ret = remap_verify_area(file: file_in, pos: pos_in, len, write: false); |
395 | if (ret) |
396 | return ret; |
397 | |
398 | ret = remap_verify_area(file: file_out, pos: pos_out, len, write: true); |
399 | if (ret) |
400 | return ret; |
401 | |
402 | file_start_write(file: file_out); |
403 | ret = file_in->f_op->remap_file_range(file_in, pos_in, |
404 | file_out, pos_out, len, remap_flags); |
405 | file_end_write(file: file_out); |
406 | if (ret < 0) |
407 | return ret; |
408 | |
409 | fsnotify_access(file: file_in); |
410 | fsnotify_modify(file: file_out); |
411 | return ret; |
412 | } |
413 | EXPORT_SYMBOL(vfs_clone_file_range); |
414 | |
415 | /* Check whether we are allowed to dedupe the destination file */ |
416 | static bool may_dedupe_file(struct file *file) |
417 | { |
418 | struct mnt_idmap *idmap = file_mnt_idmap(file); |
419 | struct inode *inode = file_inode(f: file); |
420 | |
421 | if (capable(CAP_SYS_ADMIN)) |
422 | return true; |
423 | if (file->f_mode & FMODE_WRITE) |
424 | return true; |
425 | if (vfsuid_eq_kuid(vfsuid: i_uid_into_vfsuid(idmap, inode), current_fsuid())) |
426 | return true; |
427 | if (!inode_permission(idmap, inode, MAY_WRITE)) |
428 | return true; |
429 | return false; |
430 | } |
431 | |
432 | loff_t vfs_dedupe_file_range_one(struct file *src_file, loff_t src_pos, |
433 | struct file *dst_file, loff_t dst_pos, |
434 | loff_t len, unsigned int remap_flags) |
435 | { |
436 | loff_t ret; |
437 | |
438 | WARN_ON_ONCE(remap_flags & ~(REMAP_FILE_DEDUP | |
439 | REMAP_FILE_CAN_SHORTEN)); |
440 | |
441 | /* |
442 | * This is redundant if called from vfs_dedupe_file_range(), but other |
443 | * callers need it and it's not performance sesitive... |
444 | */ |
445 | ret = remap_verify_area(file: src_file, pos: src_pos, len, write: false); |
446 | if (ret) |
447 | return ret; |
448 | |
449 | ret = remap_verify_area(file: dst_file, pos: dst_pos, len, write: true); |
450 | if (ret) |
451 | return ret; |
452 | |
453 | /* |
454 | * This needs to be called after remap_verify_area() because of |
455 | * sb_start_write() and before may_dedupe_file() because the mount's |
456 | * MAY_WRITE need to be checked with mnt_get_write_access_file() held. |
457 | */ |
458 | ret = mnt_want_write_file(file: dst_file); |
459 | if (ret) |
460 | return ret; |
461 | |
462 | ret = -EPERM; |
463 | if (!may_dedupe_file(file: dst_file)) |
464 | goto out_drop_write; |
465 | |
466 | ret = -EXDEV; |
467 | if (file_inode(f: src_file)->i_sb != file_inode(f: dst_file)->i_sb) |
468 | goto out_drop_write; |
469 | |
470 | ret = -EISDIR; |
471 | if (S_ISDIR(file_inode(dst_file)->i_mode)) |
472 | goto out_drop_write; |
473 | |
474 | ret = -EINVAL; |
475 | if (!dst_file->f_op->remap_file_range) |
476 | goto out_drop_write; |
477 | |
478 | if (len == 0) { |
479 | ret = 0; |
480 | goto out_drop_write; |
481 | } |
482 | |
483 | ret = dst_file->f_op->remap_file_range(src_file, src_pos, dst_file, |
484 | dst_pos, len, remap_flags | REMAP_FILE_DEDUP); |
485 | out_drop_write: |
486 | mnt_drop_write_file(file: dst_file); |
487 | |
488 | return ret; |
489 | } |
490 | EXPORT_SYMBOL(vfs_dedupe_file_range_one); |
491 | |
492 | int vfs_dedupe_file_range(struct file *file, struct file_dedupe_range *same) |
493 | { |
494 | struct file_dedupe_range_info *info; |
495 | struct inode *src = file_inode(f: file); |
496 | u64 off; |
497 | u64 len; |
498 | int i; |
499 | int ret; |
500 | u16 count = same->dest_count; |
501 | loff_t deduped; |
502 | |
503 | if (!(file->f_mode & FMODE_READ)) |
504 | return -EINVAL; |
505 | |
506 | if (same->reserved1 || same->reserved2) |
507 | return -EINVAL; |
508 | |
509 | off = same->src_offset; |
510 | len = same->src_length; |
511 | |
512 | if (S_ISDIR(src->i_mode)) |
513 | return -EISDIR; |
514 | |
515 | if (!S_ISREG(src->i_mode)) |
516 | return -EINVAL; |
517 | |
518 | if (!file->f_op->remap_file_range) |
519 | return -EOPNOTSUPP; |
520 | |
521 | ret = remap_verify_area(file, pos: off, len, write: false); |
522 | if (ret < 0) |
523 | return ret; |
524 | ret = 0; |
525 | |
526 | if (off + len > i_size_read(inode: src)) |
527 | return -EINVAL; |
528 | |
529 | /* Arbitrary 1G limit on a single dedupe request, can be raised. */ |
530 | len = min_t(u64, len, 1 << 30); |
531 | |
532 | /* pre-format output fields to sane values */ |
533 | for (i = 0; i < count; i++) { |
534 | same->info[i].bytes_deduped = 0ULL; |
535 | same->info[i].status = FILE_DEDUPE_RANGE_SAME; |
536 | } |
537 | |
538 | for (i = 0, info = same->info; i < count; i++, info++) { |
539 | struct fd dst_fd = fdget(fd: info->dest_fd); |
540 | struct file *dst_file = dst_fd.file; |
541 | |
542 | if (!dst_file) { |
543 | info->status = -EBADF; |
544 | goto next_loop; |
545 | } |
546 | |
547 | if (info->reserved) { |
548 | info->status = -EINVAL; |
549 | goto next_fdput; |
550 | } |
551 | |
552 | deduped = vfs_dedupe_file_range_one(file, off, dst_file, |
553 | info->dest_offset, len, |
554 | REMAP_FILE_CAN_SHORTEN); |
555 | if (deduped == -EBADE) |
556 | info->status = FILE_DEDUPE_RANGE_DIFFERS; |
557 | else if (deduped < 0) |
558 | info->status = deduped; |
559 | else |
560 | info->bytes_deduped = len; |
561 | |
562 | next_fdput: |
563 | fdput(fd: dst_fd); |
564 | next_loop: |
565 | if (fatal_signal_pending(current)) |
566 | break; |
567 | } |
568 | return ret; |
569 | } |
570 | EXPORT_SYMBOL(vfs_dedupe_file_range); |
571 | |