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 <linux/dax.h>
18 : #include "internal.h"
19 :
20 : #include <linux/uaccess.h>
21 : #include <asm/unistd.h>
22 :
23 : /*
24 : * Performs necessary checks before doing a clone.
25 : *
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.
29 : */
30 0 : 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)
33 : {
34 0 : struct inode *inode_in = file_in->f_mapping->host;
35 0 : struct inode *inode_out = file_out->f_mapping->host;
36 0 : uint64_t count = *req_count;
37 : uint64_t bcount;
38 : loff_t size_in, size_out;
39 0 : loff_t bs = inode_out->i_sb->s_blocksize;
40 : int ret;
41 :
42 : /* The start of both ranges must be aligned to an fs block. */
43 0 : if (!IS_ALIGNED(pos_in, bs) || !IS_ALIGNED(pos_out, bs))
44 : return -EINVAL;
45 :
46 : /* Ensure offsets don't wrap. */
47 0 : if (pos_in + count < pos_in || pos_out + count < pos_out)
48 : return -EINVAL;
49 :
50 0 : size_in = i_size_read(inode_in);
51 0 : size_out = i_size_read(inode_out);
52 :
53 : /* Dedupe requires both ranges to be within EOF. */
54 0 : if ((remap_flags & REMAP_FILE_DEDUP) &&
55 0 : (pos_in >= size_in || pos_in + count > size_in ||
56 0 : pos_out >= size_out || pos_out + count > size_out))
57 : return -EINVAL;
58 :
59 : /* Ensure the infile range is within the infile. */
60 0 : if (pos_in >= size_in)
61 : return -EINVAL;
62 0 : count = min(count, size_in - (uint64_t)pos_in);
63 :
64 0 : ret = generic_write_check_limits(file_out, pos_out, &count);
65 0 : if (ret)
66 : return ret;
67 :
68 : /*
69 : * If the user wanted us to link to the infile's EOF, round up to the
70 : * next block boundary for this check.
71 : *
72 : * Otherwise, make sure the count is also block-aligned, having
73 : * already confirmed the starting offsets' block alignment.
74 : */
75 0 : if (pos_in + count == size_in &&
76 0 : (!(remap_flags & REMAP_FILE_DEDUP) || pos_out + count == size_out)) {
77 0 : bcount = ALIGN(size_in, bs) - pos_in;
78 : } else {
79 0 : if (!IS_ALIGNED(count, bs))
80 0 : count = ALIGN_DOWN(count, bs);
81 0 : bcount = count;
82 : }
83 :
84 : /* Don't allow overlapped cloning within the same file. */
85 0 : if (inode_in == inode_out &&
86 0 : pos_out + bcount > pos_in &&
87 0 : pos_out < pos_in + bcount)
88 : return -EINVAL;
89 :
90 : /*
91 : * We shortened the request but the caller can't deal with that, so
92 : * bounce the request back to userspace.
93 : */
94 0 : if (*req_count != count && !(remap_flags & REMAP_FILE_CAN_SHORTEN))
95 : return -EINVAL;
96 :
97 0 : *req_count = count;
98 : return 0;
99 : }
100 :
101 : static int remap_verify_area(struct file *file, loff_t pos, loff_t len,
102 : bool write)
103 : {
104 0 : if (unlikely(pos < 0 || len < 0))
105 : return -EINVAL;
106 :
107 0 : if (unlikely((loff_t) (pos + len) < 0))
108 : return -EINVAL;
109 :
110 : return security_file_permission(file, write ? MAY_WRITE : MAY_READ);
111 : }
112 :
113 : /*
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
116 : * alignment.
117 : *
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).
121 : *
122 : * Shorten the request if possible.
123 : */
124 : static int generic_remap_check_len(struct inode *inode_in,
125 : struct inode *inode_out,
126 : loff_t pos_out,
127 : loff_t *len,
128 : unsigned int remap_flags)
129 : {
130 0 : u64 blkmask = i_blocksize(inode_in) - 1;
131 0 : loff_t new_len = *len;
132 :
133 0 : if ((*len & blkmask) == 0)
134 : return 0;
135 :
136 0 : if (pos_out + *len < i_size_read(inode_out))
137 0 : new_len &= ~blkmask;
138 :
139 0 : if (new_len == *len)
140 : return 0;
141 :
142 0 : if (remap_flags & REMAP_FILE_CAN_SHORTEN) {
143 0 : *len = new_len;
144 : return 0;
145 : }
146 :
147 0 : return (remap_flags & REMAP_FILE_DEDUP) ? -EBADE : -EINVAL;
148 : }
149 :
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)
152 : {
153 0 : return read_mapping_folio(file->f_mapping, pos >> PAGE_SHIFT, file);
154 : }
155 :
156 : /*
157 : * Lock two folios, ensuring that we lock in offset order if the folios
158 : * are from the same file.
159 : */
160 0 : static void vfs_lock_two_folios(struct folio *folio1, struct folio *folio2)
161 : {
162 : /* Always lock in order of increasing index. */
163 0 : if (folio1->index > folio2->index)
164 0 : swap(folio1, folio2);
165 :
166 0 : folio_lock(folio1);
167 0 : if (folio1 != folio2)
168 : folio_lock(folio2);
169 0 : }
170 :
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)
173 : {
174 0 : folio_unlock(folio1);
175 0 : if (folio1 != folio2)
176 0 : folio_unlock(folio2);
177 : }
178 :
179 : /*
180 : * Compare extents of two files to see if they are the same.
181 : * Caller must have locked both inodes to prevent write races.
182 : */
183 0 : 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)
186 : {
187 0 : bool same = true;
188 0 : int error = -EINVAL;
189 :
190 0 : while (len) {
191 : struct folio *src_folio, *dst_folio;
192 : void *src_addr, *dst_addr;
193 0 : loff_t cmp_len = min(PAGE_SIZE - offset_in_page(srcoff),
194 : PAGE_SIZE - offset_in_page(dstoff));
195 :
196 0 : cmp_len = min(cmp_len, len);
197 0 : if (cmp_len <= 0)
198 : goto out_error;
199 :
200 0 : src_folio = vfs_dedupe_get_folio(src, srcoff);
201 0 : if (IS_ERR(src_folio)) {
202 0 : error = PTR_ERR(src_folio);
203 0 : goto out_error;
204 : }
205 0 : dst_folio = vfs_dedupe_get_folio(dest, dstoff);
206 0 : if (IS_ERR(dst_folio)) {
207 0 : error = PTR_ERR(dst_folio);
208 : folio_put(src_folio);
209 : goto out_error;
210 : }
211 :
212 0 : vfs_lock_two_folios(src_folio, dst_folio);
213 :
214 : /*
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.
218 : */
219 0 : if (!folio_test_uptodate(src_folio) || !folio_test_uptodate(dst_folio) ||
220 0 : src_folio->mapping != src->f_mapping ||
221 0 : dst_folio->mapping != dest->f_mapping) {
222 : same = false;
223 : goto unlock;
224 : }
225 :
226 0 : src_addr = kmap_local_folio(src_folio,
227 0 : offset_in_folio(src_folio, srcoff));
228 0 : dst_addr = kmap_local_folio(dst_folio,
229 0 : offset_in_folio(dst_folio, dstoff));
230 :
231 : flush_dcache_folio(src_folio);
232 : flush_dcache_folio(dst_folio);
233 :
234 0 : if (memcmp(src_addr, dst_addr, cmp_len))
235 0 : same = false;
236 :
237 : kunmap_local(dst_addr);
238 : kunmap_local(src_addr);
239 : unlock:
240 0 : vfs_unlock_two_folios(src_folio, dst_folio);
241 0 : folio_put(dst_folio);
242 0 : folio_put(src_folio);
243 :
244 0 : if (!same)
245 : break;
246 :
247 0 : srcoff += cmp_len;
248 0 : dstoff += cmp_len;
249 0 : len -= cmp_len;
250 : }
251 :
252 0 : *is_same = same;
253 0 : return 0;
254 :
255 : out_error:
256 : return error;
257 : }
258 :
259 : /*
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.
263 : *
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.
266 : */
267 : int
268 0 : __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)
272 : {
273 0 : struct inode *inode_in = file_inode(file_in);
274 0 : struct inode *inode_out = file_inode(file_out);
275 0 : bool same_inode = (inode_in == inode_out);
276 : int ret;
277 :
278 : /* Don't touch certain kinds of inodes */
279 0 : if (IS_IMMUTABLE(inode_out))
280 : return -EPERM;
281 :
282 0 : if (IS_SWAPFILE(inode_in) || IS_SWAPFILE(inode_out))
283 : return -ETXTBSY;
284 :
285 : /* Don't reflink dirs, pipes, sockets... */
286 0 : if (S_ISDIR(inode_in->i_mode) || S_ISDIR(inode_out->i_mode))
287 : return -EISDIR;
288 0 : if (!S_ISREG(inode_in->i_mode) || !S_ISREG(inode_out->i_mode))
289 : return -EINVAL;
290 :
291 : /* Zero length dedupe exits immediately; reflink goes to EOF. */
292 0 : if (*len == 0) {
293 0 : loff_t isize = i_size_read(inode_in);
294 :
295 0 : if ((remap_flags & REMAP_FILE_DEDUP) || pos_in == isize)
296 : return 0;
297 0 : if (pos_in > isize)
298 : return -EINVAL;
299 0 : *len = isize - pos_in;
300 0 : if (*len == 0)
301 : return 0;
302 : }
303 :
304 : /* Check that we don't violate system file offset limits. */
305 0 : ret = generic_remap_checks(file_in, pos_in, file_out, pos_out, len,
306 : remap_flags);
307 0 : if (ret || *len == 0)
308 : return ret;
309 :
310 : /* Wait for the completion of any pending IOs on both files */
311 0 : inode_dio_wait(inode_in);
312 0 : if (!same_inode)
313 0 : inode_dio_wait(inode_out);
314 :
315 0 : ret = filemap_write_and_wait_range(inode_in->i_mapping,
316 0 : pos_in, pos_in + *len - 1);
317 0 : if (ret)
318 : return ret;
319 :
320 0 : ret = filemap_write_and_wait_range(inode_out->i_mapping,
321 0 : pos_out, pos_out + *len - 1);
322 0 : if (ret)
323 : return ret;
324 :
325 : /*
326 : * Check that the extents are the same.
327 : */
328 0 : if (remap_flags & REMAP_FILE_DEDUP) {
329 0 : bool is_same = false;
330 :
331 : if (!IS_DAX(inode_in))
332 0 : 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,
337 : dax_read_ops);
338 : else
339 0 : return -EINVAL;
340 0 : if (ret)
341 : return ret;
342 0 : if (!is_same)
343 : return -EBADE;
344 : }
345 :
346 0 : ret = generic_remap_check_len(inode_in, inode_out, pos_out, len,
347 : remap_flags);
348 0 : if (ret || *len == 0)
349 : return ret;
350 :
351 : /* If can't alter the file contents, we're done. */
352 0 : if (!(remap_flags & REMAP_FILE_DEDUP))
353 0 : ret = file_modified(file_out);
354 :
355 : return ret;
356 : }
357 :
358 0 : 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)
361 : {
362 0 : return __generic_remap_file_range_prep(file_in, pos_in, file_out,
363 : pos_out, len, remap_flags, NULL);
364 : }
365 : EXPORT_SYMBOL(generic_remap_file_range_prep);
366 :
367 0 : 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)
370 : {
371 : loff_t ret;
372 :
373 0 : WARN_ON_ONCE(remap_flags & REMAP_FILE_DEDUP);
374 :
375 0 : if (file_inode(file_in)->i_sb != file_inode(file_out)->i_sb)
376 : return -EXDEV;
377 :
378 0 : ret = generic_file_rw_checks(file_in, file_out);
379 0 : if (ret < 0)
380 : return ret;
381 :
382 0 : if (!file_in->f_op->remap_file_range)
383 : return -EOPNOTSUPP;
384 :
385 0 : ret = remap_verify_area(file_in, pos_in, len, false);
386 0 : if (ret)
387 : return ret;
388 :
389 0 : ret = remap_verify_area(file_out, pos_out, len, true);
390 0 : if (ret)
391 : return ret;
392 :
393 0 : ret = file_in->f_op->remap_file_range(file_in, pos_in,
394 : file_out, pos_out, len, remap_flags);
395 0 : if (ret < 0)
396 : return ret;
397 :
398 0 : fsnotify_access(file_in);
399 : fsnotify_modify(file_out);
400 : return ret;
401 : }
402 : EXPORT_SYMBOL(do_clone_file_range);
403 :
404 0 : 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)
407 : {
408 : loff_t ret;
409 :
410 0 : file_start_write(file_out);
411 0 : ret = do_clone_file_range(file_in, pos_in, file_out, pos_out, len,
412 : remap_flags);
413 0 : file_end_write(file_out);
414 :
415 0 : return ret;
416 : }
417 : EXPORT_SYMBOL(vfs_clone_file_range);
418 :
419 : /* Check whether we are allowed to dedupe the destination file */
420 0 : static bool allow_file_dedupe(struct file *file)
421 : {
422 0 : struct mnt_idmap *idmap = file_mnt_idmap(file);
423 0 : struct inode *inode = file_inode(file);
424 :
425 0 : if (capable(CAP_SYS_ADMIN))
426 : return true;
427 0 : if (file->f_mode & FMODE_WRITE)
428 : return true;
429 0 : if (vfsuid_eq_kuid(i_uid_into_vfsuid(idmap, inode), current_fsuid()))
430 : return true;
431 0 : if (!inode_permission(idmap, inode, MAY_WRITE))
432 : return true;
433 0 : return false;
434 : }
435 :
436 0 : 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)
439 : {
440 : loff_t ret;
441 :
442 0 : WARN_ON_ONCE(remap_flags & ~(REMAP_FILE_DEDUP |
443 : REMAP_FILE_CAN_SHORTEN));
444 :
445 0 : ret = mnt_want_write_file(dst_file);
446 0 : if (ret)
447 : return ret;
448 :
449 : /*
450 : * This is redundant if called from vfs_dedupe_file_range(), but other
451 : * callers need it and it's not performance sesitive...
452 : */
453 0 : ret = remap_verify_area(src_file, src_pos, len, false);
454 0 : if (ret)
455 : goto out_drop_write;
456 :
457 0 : ret = remap_verify_area(dst_file, dst_pos, len, true);
458 0 : if (ret)
459 : goto out_drop_write;
460 :
461 0 : ret = -EPERM;
462 0 : if (!allow_file_dedupe(dst_file))
463 : goto out_drop_write;
464 :
465 0 : ret = -EXDEV;
466 0 : if (file_inode(src_file)->i_sb != file_inode(dst_file)->i_sb)
467 : goto out_drop_write;
468 :
469 0 : ret = -EISDIR;
470 0 : if (S_ISDIR(file_inode(dst_file)->i_mode))
471 : goto out_drop_write;
472 :
473 0 : ret = -EINVAL;
474 0 : if (!dst_file->f_op->remap_file_range)
475 : goto out_drop_write;
476 :
477 0 : if (len == 0) {
478 : ret = 0;
479 : goto out_drop_write;
480 : }
481 :
482 0 : ret = dst_file->f_op->remap_file_range(src_file, src_pos, dst_file,
483 : dst_pos, len, remap_flags | REMAP_FILE_DEDUP);
484 : out_drop_write:
485 0 : mnt_drop_write_file(dst_file);
486 :
487 0 : return ret;
488 : }
489 : EXPORT_SYMBOL(vfs_dedupe_file_range_one);
490 :
491 0 : int vfs_dedupe_file_range(struct file *file, struct file_dedupe_range *same)
492 : {
493 : struct file_dedupe_range_info *info;
494 0 : struct inode *src = file_inode(file);
495 : u64 off;
496 : u64 len;
497 : int i;
498 : int ret;
499 0 : u16 count = same->dest_count;
500 : loff_t deduped;
501 :
502 0 : if (!(file->f_mode & FMODE_READ))
503 : return -EINVAL;
504 :
505 0 : if (same->reserved1 || same->reserved2)
506 : return -EINVAL;
507 :
508 0 : off = same->src_offset;
509 0 : len = same->src_length;
510 :
511 0 : if (S_ISDIR(src->i_mode))
512 : return -EISDIR;
513 :
514 0 : if (!S_ISREG(src->i_mode))
515 : return -EINVAL;
516 :
517 0 : if (!file->f_op->remap_file_range)
518 : return -EOPNOTSUPP;
519 :
520 0 : ret = remap_verify_area(file, off, len, false);
521 0 : if (ret < 0)
522 : return ret;
523 0 : ret = 0;
524 :
525 0 : if (off + len > i_size_read(src))
526 : return -EINVAL;
527 :
528 : /* Arbitrary 1G limit on a single dedupe request, can be raised. */
529 0 : len = min_t(u64, len, 1 << 30);
530 :
531 : /* pre-format output fields to sane values */
532 0 : for (i = 0; i < count; i++) {
533 0 : same->info[i].bytes_deduped = 0ULL;
534 0 : same->info[i].status = FILE_DEDUPE_RANGE_SAME;
535 : }
536 :
537 0 : for (i = 0, info = same->info; i < count; i++, info++) {
538 0 : struct fd dst_fd = fdget(info->dest_fd);
539 0 : struct file *dst_file = dst_fd.file;
540 :
541 0 : if (!dst_file) {
542 0 : info->status = -EBADF;
543 0 : goto next_loop;
544 : }
545 :
546 0 : if (info->reserved) {
547 0 : info->status = -EINVAL;
548 0 : goto next_fdput;
549 : }
550 :
551 0 : deduped = vfs_dedupe_file_range_one(file, off, dst_file,
552 0 : info->dest_offset, len,
553 : REMAP_FILE_CAN_SHORTEN);
554 0 : if (deduped == -EBADE)
555 0 : info->status = FILE_DEDUPE_RANGE_DIFFERS;
556 0 : else if (deduped < 0)
557 0 : info->status = deduped;
558 : else
559 0 : info->bytes_deduped = len;
560 :
561 : next_fdput:
562 0 : fdput(dst_fd);
563 : next_loop:
564 0 : if (fatal_signal_pending(current))
565 : break;
566 : }
567 : return ret;
568 : }
569 : EXPORT_SYMBOL(vfs_dedupe_file_range);
|
