Line data Source code
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 : */
29 0 : 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)
32 : {
33 0 : struct inode *inode_in = file_in->f_mapping->host;
34 0 : struct inode *inode_out = file_out->f_mapping->host;
35 0 : uint64_t count = *req_count;
36 : uint64_t bcount;
37 : loff_t size_in, size_out;
38 0 : 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 0 : if (!IS_ALIGNED(pos_in, bs) || !IS_ALIGNED(pos_out, bs))
43 : return -EINVAL;
44 :
45 : /* Ensure offsets don't wrap. */
46 0 : if (pos_in + count < pos_in || pos_out + count < pos_out)
47 : return -EINVAL;
48 :
49 0 : size_in = i_size_read(inode_in);
50 0 : size_out = i_size_read(inode_out);
51 :
52 : /* Dedupe requires both ranges to be within EOF. */
53 0 : if ((remap_flags & REMAP_FILE_DEDUP) &&
54 0 : (pos_in >= size_in || pos_in + count > size_in ||
55 0 : pos_out >= size_out || pos_out + count > size_out))
56 : return -EINVAL;
57 :
58 : /* Ensure the infile range is within the infile. */
59 0 : if (pos_in >= size_in)
60 : return -EINVAL;
61 0 : count = min(count, size_in - (uint64_t)pos_in);
62 :
63 0 : ret = generic_write_check_limits(file_out, pos_out, &count);
64 0 : 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 : */
74 0 : if (pos_in + count == size_in) {
75 0 : bcount = ALIGN(size_in, bs) - pos_in;
76 : } else {
77 0 : if (!IS_ALIGNED(count, bs))
78 0 : count = ALIGN_DOWN(count, bs);
79 0 : bcount = count;
80 : }
81 :
82 : /* Don't allow overlapped cloning within the same file. */
83 0 : if (inode_in == inode_out &&
84 0 : pos_out + bcount > pos_in &&
85 0 : pos_out < pos_in + bcount)
86 : return -EINVAL;
87 :
88 : /*
89 : * We shortened the request but the caller can't deal with that, so
90 : * bounce the request back to userspace.
91 : */
92 0 : if (*req_count != count && !(remap_flags & REMAP_FILE_CAN_SHORTEN))
93 : return -EINVAL;
94 :
95 0 : *req_count = count;
96 : return 0;
97 : }
98 :
99 : static int remap_verify_area(struct file *file, loff_t pos, loff_t len,
100 : bool write)
101 : {
102 0 : if (unlikely(pos < 0 || len < 0))
103 : return -EINVAL;
104 :
105 0 : if (unlikely((loff_t) (pos + len) < 0))
106 : return -EINVAL;
107 :
108 : return security_file_permission(file, write ? MAY_WRITE : MAY_READ);
109 : }
110 :
111 : /*
112 : * Ensure that we don't remap a partial EOF block in the middle of something
113 : * else. Assume that the offsets have already been checked for block
114 : * alignment.
115 : *
116 : * For clone we only link a partial EOF block above or at the destination file's
117 : * EOF. For deduplication we accept a partial EOF block only if it ends at the
118 : * destination file's EOF (can not link it into the middle of a file).
119 : *
120 : * Shorten the request if possible.
121 : */
122 : static int generic_remap_check_len(struct inode *inode_in,
123 : struct inode *inode_out,
124 : loff_t pos_out,
125 : loff_t *len,
126 : unsigned int remap_flags)
127 : {
128 0 : u64 blkmask = i_blocksize(inode_in) - 1;
129 0 : loff_t new_len = *len;
130 :
131 0 : if ((*len & blkmask) == 0)
132 : return 0;
133 :
134 0 : if (pos_out + *len < i_size_read(inode_out))
135 0 : new_len &= ~blkmask;
136 :
137 0 : if (new_len == *len)
138 : return 0;
139 :
140 0 : if (remap_flags & REMAP_FILE_CAN_SHORTEN) {
141 0 : *len = new_len;
142 : return 0;
143 : }
144 :
145 0 : return (remap_flags & REMAP_FILE_DEDUP) ? -EBADE : -EINVAL;
146 : }
147 :
148 : /* Read a page's worth of file data into the page cache. */
149 0 : static struct folio *vfs_dedupe_get_folio(struct file *file, loff_t pos)
150 : {
151 : struct folio *folio;
152 :
153 0 : folio = read_mapping_folio(file->f_mapping, pos >> PAGE_SHIFT, file);
154 0 : if (IS_ERR(folio))
155 : return folio;
156 0 : if (!folio_test_uptodate(folio)) {
157 : folio_put(folio);
158 : return ERR_PTR(-EIO);
159 : }
160 : return folio;
161 : }
162 :
163 : /*
164 : * Lock two folios, ensuring that we lock in offset order if the folios
165 : * are from the same file.
166 : */
167 0 : static void vfs_lock_two_folios(struct folio *folio1, struct folio *folio2)
168 : {
169 : /* Always lock in order of increasing index. */
170 0 : if (folio1->index > folio2->index)
171 0 : swap(folio1, folio2);
172 :
173 0 : folio_lock(folio1);
174 0 : if (folio1 != folio2)
175 : folio_lock(folio2);
176 0 : }
177 :
178 : /* Unlock two folios, being careful not to unlock the same folio twice. */
179 : static void vfs_unlock_two_folios(struct folio *folio1, struct folio *folio2)
180 : {
181 0 : folio_unlock(folio1);
182 0 : if (folio1 != folio2)
183 0 : folio_unlock(folio2);
184 : }
185 :
186 : /*
187 : * Compare extents of two files to see if they are the same.
188 : * Caller must have locked both inodes to prevent write races.
189 : */
190 0 : static int vfs_dedupe_file_range_compare(struct file *src, loff_t srcoff,
191 : struct file *dest, loff_t dstoff,
192 : loff_t len, bool *is_same)
193 : {
194 0 : bool same = true;
195 0 : int error = -EINVAL;
196 :
197 0 : while (len) {
198 : struct folio *src_folio, *dst_folio;
199 : void *src_addr, *dst_addr;
200 0 : loff_t cmp_len = min(PAGE_SIZE - offset_in_page(srcoff),
201 : PAGE_SIZE - offset_in_page(dstoff));
202 :
203 0 : cmp_len = min(cmp_len, len);
204 0 : if (cmp_len <= 0)
205 : goto out_error;
206 :
207 0 : src_folio = vfs_dedupe_get_folio(src, srcoff);
208 0 : if (IS_ERR(src_folio)) {
209 0 : error = PTR_ERR(src_folio);
210 0 : goto out_error;
211 : }
212 0 : dst_folio = vfs_dedupe_get_folio(dest, dstoff);
213 0 : if (IS_ERR(dst_folio)) {
214 0 : error = PTR_ERR(dst_folio);
215 : folio_put(src_folio);
216 : goto out_error;
217 : }
218 :
219 0 : vfs_lock_two_folios(src_folio, dst_folio);
220 :
221 : /*
222 : * Now that we've locked both folios, make sure they're still
223 : * mapped to the file data we're interested in. If not,
224 : * someone is invalidating pages on us and we lose.
225 : */
226 0 : if (!folio_test_uptodate(src_folio) || !folio_test_uptodate(dst_folio) ||
227 0 : src_folio->mapping != src->f_mapping ||
228 0 : dst_folio->mapping != dest->f_mapping) {
229 : same = false;
230 : goto unlock;
231 : }
232 :
233 0 : src_addr = kmap_local_folio(src_folio,
234 0 : offset_in_folio(src_folio, srcoff));
235 0 : dst_addr = kmap_local_folio(dst_folio,
236 0 : offset_in_folio(dst_folio, dstoff));
237 :
238 0 : flush_dcache_folio(src_folio);
239 0 : flush_dcache_folio(dst_folio);
240 :
241 0 : if (memcmp(src_addr, dst_addr, cmp_len))
242 0 : same = false;
243 :
244 : kunmap_local(dst_addr);
245 : kunmap_local(src_addr);
246 : unlock:
247 0 : vfs_unlock_two_folios(src_folio, dst_folio);
248 0 : folio_put(dst_folio);
249 0 : folio_put(src_folio);
250 :
251 0 : if (!same)
252 : break;
253 :
254 0 : srcoff += cmp_len;
255 0 : dstoff += cmp_len;
256 0 : len -= cmp_len;
257 : }
258 :
259 0 : *is_same = same;
260 0 : return 0;
261 :
262 : out_error:
263 : return error;
264 : }
265 :
266 : /*
267 : * Check that the two inodes are eligible for cloning, the ranges make
268 : * sense, and then flush all dirty data. Caller must ensure that the
269 : * inodes have been locked against any other modifications.
270 : *
271 : * If there's an error, then the usual negative error code is returned.
272 : * Otherwise returns 0 with *len set to the request length.
273 : */
274 0 : int generic_remap_file_range_prep(struct file *file_in, loff_t pos_in,
275 : struct file *file_out, loff_t pos_out,
276 : loff_t *len, unsigned int remap_flags)
277 : {
278 0 : struct inode *inode_in = file_inode(file_in);
279 0 : struct inode *inode_out = file_inode(file_out);
280 0 : bool same_inode = (inode_in == inode_out);
281 : int ret;
282 :
283 : /* Don't touch certain kinds of inodes */
284 0 : if (IS_IMMUTABLE(inode_out))
285 : return -EPERM;
286 :
287 0 : if (IS_SWAPFILE(inode_in) || IS_SWAPFILE(inode_out))
288 : return -ETXTBSY;
289 :
290 : /* Don't reflink dirs, pipes, sockets... */
291 0 : if (S_ISDIR(inode_in->i_mode) || S_ISDIR(inode_out->i_mode))
292 : return -EISDIR;
293 0 : if (!S_ISREG(inode_in->i_mode) || !S_ISREG(inode_out->i_mode))
294 : return -EINVAL;
295 :
296 : /* Zero length dedupe exits immediately; reflink goes to EOF. */
297 0 : if (*len == 0) {
298 0 : loff_t isize = i_size_read(inode_in);
299 :
300 0 : if ((remap_flags & REMAP_FILE_DEDUP) || pos_in == isize)
301 : return 0;
302 0 : if (pos_in > isize)
303 : return -EINVAL;
304 0 : *len = isize - pos_in;
305 0 : if (*len == 0)
306 : return 0;
307 : }
308 :
309 : /* Check that we don't violate system file offset limits. */
310 0 : ret = generic_remap_checks(file_in, pos_in, file_out, pos_out, len,
311 : remap_flags);
312 0 : if (ret)
313 : return ret;
314 :
315 : /* Wait for the completion of any pending IOs on both files */
316 0 : inode_dio_wait(inode_in);
317 0 : if (!same_inode)
318 0 : inode_dio_wait(inode_out);
319 :
320 0 : ret = filemap_write_and_wait_range(inode_in->i_mapping,
321 0 : pos_in, pos_in + *len - 1);
322 0 : if (ret)
323 : return ret;
324 :
325 0 : ret = filemap_write_and_wait_range(inode_out->i_mapping,
326 0 : pos_out, pos_out + *len - 1);
327 0 : if (ret)
328 : return ret;
329 :
330 : /*
331 : * Check that the extents are the same.
332 : */
333 0 : if (remap_flags & REMAP_FILE_DEDUP) {
334 0 : bool is_same = false;
335 :
336 0 : ret = vfs_dedupe_file_range_compare(file_in, pos_in,
337 : file_out, pos_out, *len, &is_same);
338 0 : if (ret)
339 0 : return ret;
340 0 : if (!is_same)
341 : return -EBADE;
342 : }
343 :
344 0 : ret = generic_remap_check_len(inode_in, inode_out, pos_out, len,
345 : remap_flags);
346 0 : if (ret)
347 : return ret;
348 :
349 : /* If can't alter the file contents, we're done. */
350 0 : if (!(remap_flags & REMAP_FILE_DEDUP))
351 0 : ret = file_modified(file_out);
352 :
353 : return ret;
354 : }
355 : EXPORT_SYMBOL(generic_remap_file_range_prep);
356 :
357 0 : loff_t do_clone_file_range(struct file *file_in, loff_t pos_in,
358 : struct file *file_out, loff_t pos_out,
359 : loff_t len, unsigned int remap_flags)
360 : {
361 : loff_t ret;
362 :
363 0 : WARN_ON_ONCE(remap_flags & REMAP_FILE_DEDUP);
364 :
365 0 : if (file_inode(file_in)->i_sb != file_inode(file_out)->i_sb)
366 : return -EXDEV;
367 :
368 0 : ret = generic_file_rw_checks(file_in, file_out);
369 0 : if (ret < 0)
370 : return ret;
371 :
372 0 : if (!file_in->f_op->remap_file_range)
373 : return -EOPNOTSUPP;
374 :
375 0 : ret = remap_verify_area(file_in, pos_in, len, false);
376 0 : if (ret)
377 : return ret;
378 :
379 0 : ret = remap_verify_area(file_out, pos_out, len, true);
380 0 : if (ret)
381 : return ret;
382 :
383 0 : ret = file_in->f_op->remap_file_range(file_in, pos_in,
384 : file_out, pos_out, len, remap_flags);
385 0 : if (ret < 0)
386 : return ret;
387 :
388 0 : fsnotify_access(file_in);
389 : fsnotify_modify(file_out);
390 : return ret;
391 : }
392 : EXPORT_SYMBOL(do_clone_file_range);
393 :
394 0 : loff_t vfs_clone_file_range(struct file *file_in, loff_t pos_in,
395 : struct file *file_out, loff_t pos_out,
396 : loff_t len, unsigned int remap_flags)
397 : {
398 : loff_t ret;
399 :
400 0 : file_start_write(file_out);
401 0 : ret = do_clone_file_range(file_in, pos_in, file_out, pos_out, len,
402 : remap_flags);
403 0 : file_end_write(file_out);
404 :
405 0 : return ret;
406 : }
407 : EXPORT_SYMBOL(vfs_clone_file_range);
408 :
409 : /* Check whether we are allowed to dedupe the destination file */
410 0 : static bool allow_file_dedupe(struct file *file)
411 : {
412 0 : struct user_namespace *mnt_userns = file_mnt_user_ns(file);
413 0 : struct inode *inode = file_inode(file);
414 :
415 0 : if (capable(CAP_SYS_ADMIN))
416 : return true;
417 0 : if (file->f_mode & FMODE_WRITE)
418 : return true;
419 0 : if (uid_eq(current_fsuid(), i_uid_into_mnt(mnt_userns, inode)))
420 : return true;
421 0 : if (!inode_permission(mnt_userns, inode, MAY_WRITE))
422 : return true;
423 0 : return false;
424 : }
425 :
426 0 : loff_t vfs_dedupe_file_range_one(struct file *src_file, loff_t src_pos,
427 : struct file *dst_file, loff_t dst_pos,
428 : loff_t len, unsigned int remap_flags)
429 : {
430 : loff_t ret;
431 :
432 0 : WARN_ON_ONCE(remap_flags & ~(REMAP_FILE_DEDUP |
433 : REMAP_FILE_CAN_SHORTEN));
434 :
435 0 : ret = mnt_want_write_file(dst_file);
436 0 : if (ret)
437 : return ret;
438 :
439 : /*
440 : * This is redundant if called from vfs_dedupe_file_range(), but other
441 : * callers need it and it's not performance sesitive...
442 : */
443 0 : ret = remap_verify_area(src_file, src_pos, len, false);
444 0 : if (ret)
445 : goto out_drop_write;
446 :
447 0 : ret = remap_verify_area(dst_file, dst_pos, len, true);
448 0 : if (ret)
449 : goto out_drop_write;
450 :
451 0 : ret = -EPERM;
452 0 : if (!allow_file_dedupe(dst_file))
453 : goto out_drop_write;
454 :
455 0 : ret = -EXDEV;
456 0 : if (file_inode(src_file)->i_sb != file_inode(dst_file)->i_sb)
457 : goto out_drop_write;
458 :
459 0 : ret = -EISDIR;
460 0 : if (S_ISDIR(file_inode(dst_file)->i_mode))
461 : goto out_drop_write;
462 :
463 0 : ret = -EINVAL;
464 0 : if (!dst_file->f_op->remap_file_range)
465 : goto out_drop_write;
466 :
467 0 : if (len == 0) {
468 : ret = 0;
469 : goto out_drop_write;
470 : }
471 :
472 0 : ret = dst_file->f_op->remap_file_range(src_file, src_pos, dst_file,
473 : dst_pos, len, remap_flags | REMAP_FILE_DEDUP);
474 : out_drop_write:
475 0 : mnt_drop_write_file(dst_file);
476 :
477 0 : return ret;
478 : }
479 : EXPORT_SYMBOL(vfs_dedupe_file_range_one);
480 :
481 0 : int vfs_dedupe_file_range(struct file *file, struct file_dedupe_range *same)
482 : {
483 : struct file_dedupe_range_info *info;
484 0 : struct inode *src = file_inode(file);
485 : u64 off;
486 : u64 len;
487 : int i;
488 : int ret;
489 0 : u16 count = same->dest_count;
490 : loff_t deduped;
491 :
492 0 : if (!(file->f_mode & FMODE_READ))
493 : return -EINVAL;
494 :
495 0 : if (same->reserved1 || same->reserved2)
496 : return -EINVAL;
497 :
498 0 : off = same->src_offset;
499 0 : len = same->src_length;
500 :
501 0 : if (S_ISDIR(src->i_mode))
502 : return -EISDIR;
503 :
504 0 : if (!S_ISREG(src->i_mode))
505 : return -EINVAL;
506 :
507 0 : if (!file->f_op->remap_file_range)
508 : return -EOPNOTSUPP;
509 :
510 0 : ret = remap_verify_area(file, off, len, false);
511 0 : if (ret < 0)
512 : return ret;
513 0 : ret = 0;
514 :
515 0 : if (off + len > i_size_read(src))
516 : return -EINVAL;
517 :
518 : /* Arbitrary 1G limit on a single dedupe request, can be raised. */
519 0 : len = min_t(u64, len, 1 << 30);
520 :
521 : /* pre-format output fields to sane values */
522 0 : for (i = 0; i < count; i++) {
523 0 : same->info[i].bytes_deduped = 0ULL;
524 0 : same->info[i].status = FILE_DEDUPE_RANGE_SAME;
525 : }
526 :
527 0 : for (i = 0, info = same->info; i < count; i++, info++) {
528 0 : struct fd dst_fd = fdget(info->dest_fd);
529 0 : struct file *dst_file = dst_fd.file;
530 :
531 0 : if (!dst_file) {
532 0 : info->status = -EBADF;
533 0 : goto next_loop;
534 : }
535 :
536 0 : if (info->reserved) {
537 0 : info->status = -EINVAL;
538 0 : goto next_fdput;
539 : }
540 :
541 0 : deduped = vfs_dedupe_file_range_one(file, off, dst_file,
542 0 : info->dest_offset, len,
543 : REMAP_FILE_CAN_SHORTEN);
544 0 : if (deduped == -EBADE)
545 0 : info->status = FILE_DEDUPE_RANGE_DIFFERS;
546 0 : else if (deduped < 0)
547 0 : info->status = deduped;
548 : else
549 0 : info->bytes_deduped = len;
550 :
551 : next_fdput:
552 0 : fdput(dst_fd);
553 : next_loop:
554 0 : if (fatal_signal_pending(current))
555 : break;
556 : }
557 : return ret;
558 : }
559 : EXPORT_SYMBOL(vfs_dedupe_file_range);
|
