Commit | Line | Data |
---|---|---|
02e83f46 DW |
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 "internal.h" | |
18 | ||
19 | #include <linux/uaccess.h> | |
20 | #include <asm/unistd.h> | |
21 | ||
22 | /* | |
23 | * Performs necessary checks before doing a clone. | |
24 | * | |
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. | |
28 | */ | |
1b2c54d6 DW |
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) | |
02e83f46 DW |
32 | { |
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; | |
36 | uint64_t bcount; | |
37 | loff_t size_in, size_out; | |
38 | loff_t bs = inode_out->i_sb->s_blocksize; | |
39 | int ret; | |
40 | ||
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)) | |
43 | return -EINVAL; | |
44 | ||
45 | /* Ensure offsets don't wrap. */ | |
46 | if (pos_in + count < pos_in || pos_out + count < pos_out) | |
47 | return -EINVAL; | |
48 | ||
49 | size_in = i_size_read(inode_in); | |
50 | size_out = i_size_read(inode_out); | |
51 | ||
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)) | |
56 | return -EINVAL; | |
57 | ||
58 | /* Ensure the infile range is within the infile. */ | |
59 | if (pos_in >= size_in) | |
60 | return -EINVAL; | |
61 | count = min(count, size_in - (uint64_t)pos_in); | |
62 | ||
63 | ret = generic_write_check_limits(file_out, pos_out, &count); | |
64 | if (ret) | |
65 | return ret; | |
66 | ||
67 | /* | |
68 | * If the user wanted us to link to the infile's EOF, round up to the | |
69 | * next block boundary for this check. | |
70 | * | |
71 | * Otherwise, make sure the count is also block-aligned, having | |
72 | * already confirmed the starting offsets' block alignment. | |
73 | */ | |
5750676b DC |
74 | if (pos_in + count == size_in && |
75 | (!(remap_flags & REMAP_FILE_DEDUP) || pos_out + count == size_out)) { | |
02e83f46 DW |
76 | bcount = ALIGN(size_in, bs) - pos_in; |
77 | } else { | |
78 | if (!IS_ALIGNED(count, bs)) | |
79 | count = ALIGN_DOWN(count, bs); | |
80 | bcount = count; | |
81 | } | |
82 | ||
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) | |
87 | return -EINVAL; | |
88 | ||
89 | /* | |
90 | * We shortened the request but the caller can't deal with that, so | |
91 | * bounce the request back to userspace. | |
92 | */ | |
93 | if (*req_count != count && !(remap_flags & REMAP_FILE_CAN_SHORTEN)) | |
94 | return -EINVAL; | |
95 | ||
96 | *req_count = count; | |
97 | return 0; | |
98 | } | |
1b2c54d6 DW |
99 | |
100 | static int remap_verify_area(struct file *file, loff_t pos, loff_t len, | |
101 | bool write) | |
102 | { | |
1b2c54d6 DW |
103 | if (unlikely(pos < 0 || len < 0)) |
104 | return -EINVAL; | |
105 | ||
106 | if (unlikely((loff_t) (pos + len) < 0)) | |
107 | return -EINVAL; | |
108 | ||
1b2c54d6 DW |
109 | return security_file_permission(file, write ? MAY_WRITE : MAY_READ); |
110 | } | |
111 | ||
112 | /* | |
113 | * Ensure that we don't remap a partial EOF block in the middle of something | |
114 | * else. Assume that the offsets have already been checked for block | |
115 | * alignment. | |
116 | * | |
117 | * For clone we only link a partial EOF block above or at the destination file's | |
118 | * EOF. For deduplication we accept a partial EOF block only if it ends at the | |
119 | * destination file's EOF (can not link it into the middle of a file). | |
120 | * | |
121 | * Shorten the request if possible. | |
122 | */ | |
123 | static int generic_remap_check_len(struct inode *inode_in, | |
124 | struct inode *inode_out, | |
125 | loff_t pos_out, | |
126 | loff_t *len, | |
127 | unsigned int remap_flags) | |
128 | { | |
129 | u64 blkmask = i_blocksize(inode_in) - 1; | |
130 | loff_t new_len = *len; | |
131 | ||
132 | if ((*len & blkmask) == 0) | |
133 | return 0; | |
134 | ||
135 | if (pos_out + *len < i_size_read(inode_out)) | |
136 | new_len &= ~blkmask; | |
137 | ||
138 | if (new_len == *len) | |
139 | return 0; | |
140 | ||
141 | if (remap_flags & REMAP_FILE_CAN_SHORTEN) { | |
142 | *len = new_len; | |
143 | return 0; | |
144 | } | |
145 | ||
146 | return (remap_flags & REMAP_FILE_DEDUP) ? -EBADE : -EINVAL; | |
147 | } | |
148 | ||
149 | /* Read a page's worth of file data into the page cache. */ | |
4495a96c | 150 | static struct folio *vfs_dedupe_get_folio(struct file *file, loff_t pos) |
1b2c54d6 | 151 | { |
da028b6b | 152 | return read_mapping_folio(file->f_mapping, pos >> PAGE_SHIFT, file); |
1b2c54d6 DW |
153 | } |
154 | ||
155 | /* | |
338f379c MWO |
156 | * Lock two folios, ensuring that we lock in offset order if the folios |
157 | * are from the same file. | |
1b2c54d6 | 158 | */ |
338f379c | 159 | static void vfs_lock_two_folios(struct folio *folio1, struct folio *folio2) |
1b2c54d6 DW |
160 | { |
161 | /* Always lock in order of increasing index. */ | |
338f379c MWO |
162 | if (folio1->index > folio2->index) |
163 | swap(folio1, folio2); | |
1b2c54d6 | 164 | |
338f379c MWO |
165 | folio_lock(folio1); |
166 | if (folio1 != folio2) | |
167 | folio_lock(folio2); | |
1b2c54d6 DW |
168 | } |
169 | ||
338f379c MWO |
170 | /* Unlock two folios, being careful not to unlock the same folio twice. */ |
171 | static void vfs_unlock_two_folios(struct folio *folio1, struct folio *folio2) | |
1b2c54d6 | 172 | { |
338f379c MWO |
173 | folio_unlock(folio1); |
174 | if (folio1 != folio2) | |
175 | folio_unlock(folio2); | |
1b2c54d6 DW |
176 | } |
177 | ||
178 | /* | |
179 | * Compare extents of two files to see if they are the same. | |
180 | * Caller must have locked both inodes to prevent write races. | |
181 | */ | |
4495a96c MWO |
182 | static int vfs_dedupe_file_range_compare(struct file *src, loff_t srcoff, |
183 | struct file *dest, loff_t dstoff, | |
1b2c54d6 DW |
184 | loff_t len, bool *is_same) |
185 | { | |
338f379c MWO |
186 | bool same = true; |
187 | int error = -EINVAL; | |
188 | ||
1b2c54d6 | 189 | while (len) { |
338f379c MWO |
190 | struct folio *src_folio, *dst_folio; |
191 | void *src_addr, *dst_addr; | |
192 | loff_t cmp_len = min(PAGE_SIZE - offset_in_page(srcoff), | |
193 | PAGE_SIZE - offset_in_page(dstoff)); | |
194 | ||
1b2c54d6 DW |
195 | cmp_len = min(cmp_len, len); |
196 | if (cmp_len <= 0) | |
197 | goto out_error; | |
198 | ||
338f379c MWO |
199 | src_folio = vfs_dedupe_get_folio(src, srcoff); |
200 | if (IS_ERR(src_folio)) { | |
201 | error = PTR_ERR(src_folio); | |
1b2c54d6 DW |
202 | goto out_error; |
203 | } | |
338f379c MWO |
204 | dst_folio = vfs_dedupe_get_folio(dest, dstoff); |
205 | if (IS_ERR(dst_folio)) { | |
206 | error = PTR_ERR(dst_folio); | |
207 | folio_put(src_folio); | |
1b2c54d6 DW |
208 | goto out_error; |
209 | } | |
210 | ||
338f379c | 211 | vfs_lock_two_folios(src_folio, dst_folio); |
1b2c54d6 DW |
212 | |
213 | /* | |
338f379c | 214 | * Now that we've locked both folios, make sure they're still |
1b2c54d6 DW |
215 | * mapped to the file data we're interested in. If not, |
216 | * someone is invalidating pages on us and we lose. | |
217 | */ | |
338f379c | 218 | if (!folio_test_uptodate(src_folio) || !folio_test_uptodate(dst_folio) || |
4495a96c MWO |
219 | src_folio->mapping != src->f_mapping || |
220 | dst_folio->mapping != dest->f_mapping) { | |
1b2c54d6 DW |
221 | same = false; |
222 | goto unlock; | |
223 | } | |
224 | ||
338f379c MWO |
225 | src_addr = kmap_local_folio(src_folio, |
226 | offset_in_folio(src_folio, srcoff)); | |
227 | dst_addr = kmap_local_folio(dst_folio, | |
228 | offset_in_folio(dst_folio, dstoff)); | |
1b2c54d6 | 229 | |
338f379c MWO |
230 | flush_dcache_folio(src_folio); |
231 | flush_dcache_folio(dst_folio); | |
1b2c54d6 | 232 | |
338f379c | 233 | if (memcmp(src_addr, dst_addr, cmp_len)) |
1b2c54d6 DW |
234 | same = false; |
235 | ||
338f379c MWO |
236 | kunmap_local(dst_addr); |
237 | kunmap_local(src_addr); | |
1b2c54d6 | 238 | unlock: |
338f379c MWO |
239 | vfs_unlock_two_folios(src_folio, dst_folio); |
240 | folio_put(dst_folio); | |
241 | folio_put(src_folio); | |
1b2c54d6 DW |
242 | |
243 | if (!same) | |
244 | break; | |
245 | ||
246 | srcoff += cmp_len; | |
338f379c | 247 | dstoff += cmp_len; |
1b2c54d6 DW |
248 | len -= cmp_len; |
249 | } | |
250 | ||
251 | *is_same = same; | |
252 | return 0; | |
253 | ||
254 | out_error: | |
255 | return error; | |
256 | } | |
257 | ||
258 | /* | |
259 | * Check that the two inodes are eligible for cloning, the ranges make | |
260 | * sense, and then flush all dirty data. Caller must ensure that the | |
261 | * inodes have been locked against any other modifications. | |
262 | * | |
263 | * If there's an error, then the usual negative error code is returned. | |
264 | * Otherwise returns 0 with *len set to the request length. | |
265 | */ | |
266 | int generic_remap_file_range_prep(struct file *file_in, loff_t pos_in, | |
267 | struct file *file_out, loff_t pos_out, | |
268 | loff_t *len, unsigned int remap_flags) | |
269 | { | |
270 | struct inode *inode_in = file_inode(file_in); | |
271 | struct inode *inode_out = file_inode(file_out); | |
272 | bool same_inode = (inode_in == inode_out); | |
273 | int ret; | |
274 | ||
275 | /* Don't touch certain kinds of inodes */ | |
276 | if (IS_IMMUTABLE(inode_out)) | |
277 | return -EPERM; | |
278 | ||
279 | if (IS_SWAPFILE(inode_in) || IS_SWAPFILE(inode_out)) | |
280 | return -ETXTBSY; | |
281 | ||
282 | /* Don't reflink dirs, pipes, sockets... */ | |
283 | if (S_ISDIR(inode_in->i_mode) || S_ISDIR(inode_out->i_mode)) | |
284 | return -EISDIR; | |
285 | if (!S_ISREG(inode_in->i_mode) || !S_ISREG(inode_out->i_mode)) | |
286 | return -EINVAL; | |
287 | ||
288 | /* Zero length dedupe exits immediately; reflink goes to EOF. */ | |
289 | if (*len == 0) { | |
290 | loff_t isize = i_size_read(inode_in); | |
291 | ||
292 | if ((remap_flags & REMAP_FILE_DEDUP) || pos_in == isize) | |
293 | return 0; | |
294 | if (pos_in > isize) | |
295 | return -EINVAL; | |
296 | *len = isize - pos_in; | |
297 | if (*len == 0) | |
298 | return 0; | |
299 | } | |
300 | ||
301 | /* Check that we don't violate system file offset limits. */ | |
302 | ret = generic_remap_checks(file_in, pos_in, file_out, pos_out, len, | |
303 | remap_flags); | |
304 | if (ret) | |
305 | return ret; | |
306 | ||
307 | /* Wait for the completion of any pending IOs on both files */ | |
308 | inode_dio_wait(inode_in); | |
309 | if (!same_inode) | |
310 | inode_dio_wait(inode_out); | |
311 | ||
312 | ret = filemap_write_and_wait_range(inode_in->i_mapping, | |
313 | pos_in, pos_in + *len - 1); | |
314 | if (ret) | |
315 | return ret; | |
316 | ||
317 | ret = filemap_write_and_wait_range(inode_out->i_mapping, | |
318 | pos_out, pos_out + *len - 1); | |
319 | if (ret) | |
320 | return ret; | |
321 | ||
322 | /* | |
323 | * Check that the extents are the same. | |
324 | */ | |
325 | if (remap_flags & REMAP_FILE_DEDUP) { | |
326 | bool is_same = false; | |
327 | ||
4495a96c MWO |
328 | ret = vfs_dedupe_file_range_compare(file_in, pos_in, |
329 | file_out, pos_out, *len, &is_same); | |
1b2c54d6 DW |
330 | if (ret) |
331 | return ret; | |
332 | if (!is_same) | |
333 | return -EBADE; | |
334 | } | |
335 | ||
336 | ret = generic_remap_check_len(inode_in, inode_out, pos_out, len, | |
337 | remap_flags); | |
338 | if (ret) | |
339 | return ret; | |
340 | ||
341 | /* If can't alter the file contents, we're done. */ | |
342 | if (!(remap_flags & REMAP_FILE_DEDUP)) | |
343 | ret = file_modified(file_out); | |
344 | ||
345 | return ret; | |
346 | } | |
347 | EXPORT_SYMBOL(generic_remap_file_range_prep); | |
348 | ||
349 | loff_t do_clone_file_range(struct file *file_in, loff_t pos_in, | |
350 | struct file *file_out, loff_t pos_out, | |
351 | loff_t len, unsigned int remap_flags) | |
352 | { | |
353 | loff_t ret; | |
354 | ||
355 | WARN_ON_ONCE(remap_flags & REMAP_FILE_DEDUP); | |
356 | ||
1b2c54d6 DW |
357 | if (file_inode(file_in)->i_sb != file_inode(file_out)->i_sb) |
358 | return -EXDEV; | |
359 | ||
360 | ret = generic_file_rw_checks(file_in, file_out); | |
361 | if (ret < 0) | |
362 | return ret; | |
363 | ||
364 | if (!file_in->f_op->remap_file_range) | |
365 | return -EOPNOTSUPP; | |
366 | ||
367 | ret = remap_verify_area(file_in, pos_in, len, false); | |
368 | if (ret) | |
369 | return ret; | |
370 | ||
371 | ret = remap_verify_area(file_out, pos_out, len, true); | |
372 | if (ret) | |
373 | return ret; | |
374 | ||
375 | ret = file_in->f_op->remap_file_range(file_in, pos_in, | |
376 | file_out, pos_out, len, remap_flags); | |
377 | if (ret < 0) | |
378 | return ret; | |
379 | ||
380 | fsnotify_access(file_in); | |
381 | fsnotify_modify(file_out); | |
382 | return ret; | |
383 | } | |
384 | EXPORT_SYMBOL(do_clone_file_range); | |
385 | ||
386 | loff_t vfs_clone_file_range(struct file *file_in, loff_t pos_in, | |
387 | struct file *file_out, loff_t pos_out, | |
388 | loff_t len, unsigned int remap_flags) | |
389 | { | |
390 | loff_t ret; | |
391 | ||
392 | file_start_write(file_out); | |
393 | ret = do_clone_file_range(file_in, pos_in, file_out, pos_out, len, | |
394 | remap_flags); | |
395 | file_end_write(file_out); | |
396 | ||
397 | return ret; | |
398 | } | |
399 | EXPORT_SYMBOL(vfs_clone_file_range); | |
400 | ||
401 | /* Check whether we are allowed to dedupe the destination file */ | |
402 | static bool allow_file_dedupe(struct file *file) | |
403 | { | |
0f5d220b CB |
404 | struct user_namespace *mnt_userns = file_mnt_user_ns(file); |
405 | struct inode *inode = file_inode(file); | |
406 | ||
1b2c54d6 DW |
407 | if (capable(CAP_SYS_ADMIN)) |
408 | return true; | |
409 | if (file->f_mode & FMODE_WRITE) | |
410 | return true; | |
0f5d220b | 411 | if (uid_eq(current_fsuid(), i_uid_into_mnt(mnt_userns, inode))) |
1b2c54d6 | 412 | return true; |
0f5d220b | 413 | if (!inode_permission(mnt_userns, inode, MAY_WRITE)) |
1b2c54d6 DW |
414 | return true; |
415 | return false; | |
416 | } | |
417 | ||
418 | loff_t vfs_dedupe_file_range_one(struct file *src_file, loff_t src_pos, | |
419 | struct file *dst_file, loff_t dst_pos, | |
420 | loff_t len, unsigned int remap_flags) | |
421 | { | |
422 | loff_t ret; | |
423 | ||
424 | WARN_ON_ONCE(remap_flags & ~(REMAP_FILE_DEDUP | | |
425 | REMAP_FILE_CAN_SHORTEN)); | |
426 | ||
427 | ret = mnt_want_write_file(dst_file); | |
428 | if (ret) | |
429 | return ret; | |
430 | ||
3078d85c MS |
431 | /* |
432 | * This is redundant if called from vfs_dedupe_file_range(), but other | |
433 | * callers need it and it's not performance sesitive... | |
434 | */ | |
435 | ret = remap_verify_area(src_file, src_pos, len, false); | |
436 | if (ret) | |
437 | goto out_drop_write; | |
438 | ||
1b2c54d6 | 439 | ret = remap_verify_area(dst_file, dst_pos, len, true); |
3078d85c | 440 | if (ret) |
1b2c54d6 DW |
441 | goto out_drop_write; |
442 | ||
443 | ret = -EPERM; | |
444 | if (!allow_file_dedupe(dst_file)) | |
445 | goto out_drop_write; | |
446 | ||
447 | ret = -EXDEV; | |
9f5710bb | 448 | if (file_inode(src_file)->i_sb != file_inode(dst_file)->i_sb) |
1b2c54d6 DW |
449 | goto out_drop_write; |
450 | ||
451 | ret = -EISDIR; | |
452 | if (S_ISDIR(file_inode(dst_file)->i_mode)) | |
453 | goto out_drop_write; | |
454 | ||
455 | ret = -EINVAL; | |
456 | if (!dst_file->f_op->remap_file_range) | |
457 | goto out_drop_write; | |
458 | ||
459 | if (len == 0) { | |
460 | ret = 0; | |
461 | goto out_drop_write; | |
462 | } | |
463 | ||
464 | ret = dst_file->f_op->remap_file_range(src_file, src_pos, dst_file, | |
465 | dst_pos, len, remap_flags | REMAP_FILE_DEDUP); | |
466 | out_drop_write: | |
467 | mnt_drop_write_file(dst_file); | |
468 | ||
469 | return ret; | |
470 | } | |
471 | EXPORT_SYMBOL(vfs_dedupe_file_range_one); | |
472 | ||
473 | int vfs_dedupe_file_range(struct file *file, struct file_dedupe_range *same) | |
474 | { | |
475 | struct file_dedupe_range_info *info; | |
476 | struct inode *src = file_inode(file); | |
477 | u64 off; | |
478 | u64 len; | |
479 | int i; | |
480 | int ret; | |
481 | u16 count = same->dest_count; | |
482 | loff_t deduped; | |
483 | ||
484 | if (!(file->f_mode & FMODE_READ)) | |
485 | return -EINVAL; | |
486 | ||
487 | if (same->reserved1 || same->reserved2) | |
488 | return -EINVAL; | |
489 | ||
490 | off = same->src_offset; | |
491 | len = same->src_length; | |
492 | ||
493 | if (S_ISDIR(src->i_mode)) | |
494 | return -EISDIR; | |
495 | ||
496 | if (!S_ISREG(src->i_mode)) | |
497 | return -EINVAL; | |
498 | ||
499 | if (!file->f_op->remap_file_range) | |
500 | return -EOPNOTSUPP; | |
501 | ||
502 | ret = remap_verify_area(file, off, len, false); | |
503 | if (ret < 0) | |
504 | return ret; | |
505 | ret = 0; | |
506 | ||
507 | if (off + len > i_size_read(src)) | |
508 | return -EINVAL; | |
509 | ||
510 | /* Arbitrary 1G limit on a single dedupe request, can be raised. */ | |
511 | len = min_t(u64, len, 1 << 30); | |
512 | ||
513 | /* pre-format output fields to sane values */ | |
514 | for (i = 0; i < count; i++) { | |
515 | same->info[i].bytes_deduped = 0ULL; | |
516 | same->info[i].status = FILE_DEDUPE_RANGE_SAME; | |
517 | } | |
518 | ||
519 | for (i = 0, info = same->info; i < count; i++, info++) { | |
520 | struct fd dst_fd = fdget(info->dest_fd); | |
521 | struct file *dst_file = dst_fd.file; | |
522 | ||
523 | if (!dst_file) { | |
524 | info->status = -EBADF; | |
525 | goto next_loop; | |
526 | } | |
527 | ||
528 | if (info->reserved) { | |
529 | info->status = -EINVAL; | |
530 | goto next_fdput; | |
531 | } | |
532 | ||
533 | deduped = vfs_dedupe_file_range_one(file, off, dst_file, | |
534 | info->dest_offset, len, | |
535 | REMAP_FILE_CAN_SHORTEN); | |
536 | if (deduped == -EBADE) | |
537 | info->status = FILE_DEDUPE_RANGE_DIFFERS; | |
538 | else if (deduped < 0) | |
539 | info->status = deduped; | |
540 | else | |
b926f2ad | 541 | info->bytes_deduped = len; |
1b2c54d6 DW |
542 | |
543 | next_fdput: | |
544 | fdput(dst_fd); | |
545 | next_loop: | |
546 | if (fatal_signal_pending(current)) | |
547 | break; | |
548 | } | |
549 | return ret; | |
550 | } | |
551 | EXPORT_SYMBOL(vfs_dedupe_file_range); |