blk-mq: don't count completed flush data request as inflight in case of quiesce
[linux-block.git] / fs / remap_range.c
CommitLineData
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>
6f7db389 17#include <linux/dax.h>
b7a9a503 18#include <linux/overflow.h>
02e83f46
DW
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 */
1b2c54d6
DW
31static 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)
02e83f46
DW
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_in);
52 size_out = i_size_read(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_out, pos_out, &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 */
5750676b
DC
76 if (pos_in + count == size_in &&
77 (!(remap_flags & REMAP_FILE_DEDUP) || pos_out + count == size_out)) {
02e83f46
DW
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}
1b2c54d6
DW
101
102static int remap_verify_area(struct file *file, loff_t pos, loff_t len,
103 bool write)
104{
b7a9a503
DS
105 loff_t tmp;
106
1b2c54d6
DW
107 if (unlikely(pos < 0 || len < 0))
108 return -EINVAL;
109
b7a9a503 110 if (unlikely(check_add_overflow(pos, len, &tmp)))
1b2c54d6
DW
111 return -EINVAL;
112
1b2c54d6
DW
113 return security_file_permission(file, write ? MAY_WRITE : MAY_READ);
114}
115
116/*
117 * Ensure that we don't remap a partial EOF block in the middle of something
118 * else. Assume that the offsets have already been checked for block
119 * alignment.
120 *
121 * For clone we only link a partial EOF block above or at the destination file's
122 * EOF. For deduplication we accept a partial EOF block only if it ends at the
123 * destination file's EOF (can not link it into the middle of a file).
124 *
125 * Shorten the request if possible.
126 */
127static int generic_remap_check_len(struct inode *inode_in,
128 struct inode *inode_out,
129 loff_t pos_out,
130 loff_t *len,
131 unsigned int remap_flags)
132{
133 u64 blkmask = i_blocksize(inode_in) - 1;
134 loff_t new_len = *len;
135
136 if ((*len & blkmask) == 0)
137 return 0;
138
139 if (pos_out + *len < i_size_read(inode_out))
140 new_len &= ~blkmask;
141
142 if (new_len == *len)
143 return 0;
144
145 if (remap_flags & REMAP_FILE_CAN_SHORTEN) {
146 *len = new_len;
147 return 0;
148 }
149
150 return (remap_flags & REMAP_FILE_DEDUP) ? -EBADE : -EINVAL;
151}
152
153/* Read a page's worth of file data into the page cache. */
4495a96c 154static struct folio *vfs_dedupe_get_folio(struct file *file, loff_t pos)
1b2c54d6 155{
da028b6b 156 return read_mapping_folio(file->f_mapping, pos >> PAGE_SHIFT, file);
1b2c54d6
DW
157}
158
159/*
338f379c
MWO
160 * Lock two folios, ensuring that we lock in offset order if the folios
161 * are from the same file.
1b2c54d6 162 */
338f379c 163static void vfs_lock_two_folios(struct folio *folio1, struct folio *folio2)
1b2c54d6
DW
164{
165 /* Always lock in order of increasing index. */
338f379c
MWO
166 if (folio1->index > folio2->index)
167 swap(folio1, folio2);
1b2c54d6 168
338f379c
MWO
169 folio_lock(folio1);
170 if (folio1 != folio2)
171 folio_lock(folio2);
1b2c54d6
DW
172}
173
338f379c
MWO
174/* Unlock two folios, being careful not to unlock the same folio twice. */
175static void vfs_unlock_two_folios(struct folio *folio1, struct folio *folio2)
1b2c54d6 176{
338f379c
MWO
177 folio_unlock(folio1);
178 if (folio1 != folio2)
179 folio_unlock(folio2);
1b2c54d6
DW
180}
181
182/*
183 * Compare extents of two files to see if they are the same.
184 * Caller must have locked both inodes to prevent write races.
185 */
4495a96c
MWO
186static int vfs_dedupe_file_range_compare(struct file *src, loff_t srcoff,
187 struct file *dest, loff_t dstoff,
1b2c54d6
DW
188 loff_t len, bool *is_same)
189{
338f379c
MWO
190 bool same = true;
191 int error = -EINVAL;
192
1b2c54d6 193 while (len) {
338f379c
MWO
194 struct folio *src_folio, *dst_folio;
195 void *src_addr, *dst_addr;
196 loff_t cmp_len = min(PAGE_SIZE - offset_in_page(srcoff),
197 PAGE_SIZE - offset_in_page(dstoff));
198
1b2c54d6
DW
199 cmp_len = min(cmp_len, len);
200 if (cmp_len <= 0)
201 goto out_error;
202
338f379c
MWO
203 src_folio = vfs_dedupe_get_folio(src, srcoff);
204 if (IS_ERR(src_folio)) {
205 error = PTR_ERR(src_folio);
1b2c54d6
DW
206 goto out_error;
207 }
338f379c
MWO
208 dst_folio = vfs_dedupe_get_folio(dest, dstoff);
209 if (IS_ERR(dst_folio)) {
210 error = PTR_ERR(dst_folio);
211 folio_put(src_folio);
1b2c54d6
DW
212 goto out_error;
213 }
214
338f379c 215 vfs_lock_two_folios(src_folio, dst_folio);
1b2c54d6
DW
216
217 /*
338f379c 218 * Now that we've locked both folios, make sure they're still
1b2c54d6
DW
219 * mapped to the file data we're interested in. If not,
220 * someone is invalidating pages on us and we lose.
221 */
338f379c 222 if (!folio_test_uptodate(src_folio) || !folio_test_uptodate(dst_folio) ||
4495a96c
MWO
223 src_folio->mapping != src->f_mapping ||
224 dst_folio->mapping != dest->f_mapping) {
1b2c54d6
DW
225 same = false;
226 goto unlock;
227 }
228
338f379c
MWO
229 src_addr = kmap_local_folio(src_folio,
230 offset_in_folio(src_folio, srcoff));
231 dst_addr = kmap_local_folio(dst_folio,
232 offset_in_folio(dst_folio, dstoff));
1b2c54d6 233
338f379c
MWO
234 flush_dcache_folio(src_folio);
235 flush_dcache_folio(dst_folio);
1b2c54d6 236
338f379c 237 if (memcmp(src_addr, dst_addr, cmp_len))
1b2c54d6
DW
238 same = false;
239
338f379c
MWO
240 kunmap_local(dst_addr);
241 kunmap_local(src_addr);
1b2c54d6 242unlock:
338f379c
MWO
243 vfs_unlock_two_folios(src_folio, dst_folio);
244 folio_put(dst_folio);
245 folio_put(src_folio);
1b2c54d6
DW
246
247 if (!same)
248 break;
249
250 srcoff += cmp_len;
338f379c 251 dstoff += cmp_len;
1b2c54d6
DW
252 len -= cmp_len;
253 }
254
255 *is_same = same;
256 return 0;
257
258out_error:
259 return error;
260}
261
262/*
263 * Check that the two inodes are eligible for cloning, the ranges make
264 * sense, and then flush all dirty data. Caller must ensure that the
265 * inodes have been locked against any other modifications.
266 *
267 * If there's an error, then the usual negative error code is returned.
268 * Otherwise returns 0 with *len set to the request length.
269 */
6f7db389
SR
270int
271__generic_remap_file_range_prep(struct file *file_in, loff_t pos_in,
272 struct file *file_out, loff_t pos_out,
273 loff_t *len, unsigned int remap_flags,
274 const struct iomap_ops *dax_read_ops)
1b2c54d6
DW
275{
276 struct inode *inode_in = file_inode(file_in);
277 struct inode *inode_out = file_inode(file_out);
278 bool same_inode = (inode_in == inode_out);
279 int ret;
280
281 /* Don't touch certain kinds of inodes */
282 if (IS_IMMUTABLE(inode_out))
283 return -EPERM;
284
285 if (IS_SWAPFILE(inode_in) || IS_SWAPFILE(inode_out))
286 return -ETXTBSY;
287
288 /* Don't reflink dirs, pipes, sockets... */
289 if (S_ISDIR(inode_in->i_mode) || S_ISDIR(inode_out->i_mode))
290 return -EISDIR;
291 if (!S_ISREG(inode_in->i_mode) || !S_ISREG(inode_out->i_mode))
292 return -EINVAL;
293
294 /* Zero length dedupe exits immediately; reflink goes to EOF. */
295 if (*len == 0) {
296 loff_t isize = i_size_read(inode_in);
297
298 if ((remap_flags & REMAP_FILE_DEDUP) || pos_in == isize)
299 return 0;
300 if (pos_in > isize)
301 return -EINVAL;
302 *len = isize - pos_in;
303 if (*len == 0)
304 return 0;
305 }
306
307 /* Check that we don't violate system file offset limits. */
308 ret = generic_remap_checks(file_in, pos_in, file_out, pos_out, len,
309 remap_flags);
a79168a0 310 if (ret || *len == 0)
1b2c54d6
DW
311 return ret;
312
313 /* Wait for the completion of any pending IOs on both files */
314 inode_dio_wait(inode_in);
315 if (!same_inode)
316 inode_dio_wait(inode_out);
317
318 ret = filemap_write_and_wait_range(inode_in->i_mapping,
319 pos_in, pos_in + *len - 1);
320 if (ret)
321 return ret;
322
323 ret = filemap_write_and_wait_range(inode_out->i_mapping,
324 pos_out, pos_out + *len - 1);
325 if (ret)
326 return ret;
327
328 /*
329 * Check that the extents are the same.
330 */
331 if (remap_flags & REMAP_FILE_DEDUP) {
332 bool is_same = false;
333
6f7db389
SR
334 if (!IS_DAX(inode_in))
335 ret = vfs_dedupe_file_range_compare(file_in, pos_in,
336 file_out, pos_out, *len, &is_same);
337 else if (dax_read_ops)
338 ret = dax_dedupe_file_range_compare(inode_in, pos_in,
339 inode_out, pos_out, *len, &is_same,
340 dax_read_ops);
341 else
342 return -EINVAL;
1b2c54d6
DW
343 if (ret)
344 return ret;
345 if (!is_same)
346 return -EBADE;
347 }
348
349 ret = generic_remap_check_len(inode_in, inode_out, pos_out, len,
350 remap_flags);
a79168a0 351 if (ret || *len == 0)
1b2c54d6
DW
352 return ret;
353
354 /* If can't alter the file contents, we're done. */
355 if (!(remap_flags & REMAP_FILE_DEDUP))
356 ret = file_modified(file_out);
357
358 return ret;
359}
6f7db389
SR
360
361int generic_remap_file_range_prep(struct file *file_in, loff_t pos_in,
362 struct file *file_out, loff_t pos_out,
363 loff_t *len, unsigned int remap_flags)
364{
365 return __generic_remap_file_range_prep(file_in, pos_in, file_out,
366 pos_out, len, remap_flags, NULL);
367}
1b2c54d6
DW
368EXPORT_SYMBOL(generic_remap_file_range_prep);
369
370loff_t do_clone_file_range(struct file *file_in, loff_t pos_in,
371 struct file *file_out, loff_t pos_out,
372 loff_t len, unsigned int remap_flags)
373{
374 loff_t ret;
375
376 WARN_ON_ONCE(remap_flags & REMAP_FILE_DEDUP);
377
1b2c54d6
DW
378 if (file_inode(file_in)->i_sb != file_inode(file_out)->i_sb)
379 return -EXDEV;
380
381 ret = generic_file_rw_checks(file_in, file_out);
382 if (ret < 0)
383 return ret;
384
385 if (!file_in->f_op->remap_file_range)
386 return -EOPNOTSUPP;
387
388 ret = remap_verify_area(file_in, pos_in, len, false);
389 if (ret)
390 return ret;
391
392 ret = remap_verify_area(file_out, pos_out, len, true);
393 if (ret)
394 return ret;
395
396 ret = file_in->f_op->remap_file_range(file_in, pos_in,
397 file_out, pos_out, len, remap_flags);
398 if (ret < 0)
399 return ret;
400
401 fsnotify_access(file_in);
402 fsnotify_modify(file_out);
403 return ret;
404}
405EXPORT_SYMBOL(do_clone_file_range);
406
407loff_t vfs_clone_file_range(struct file *file_in, loff_t pos_in,
408 struct file *file_out, loff_t pos_out,
409 loff_t len, unsigned int remap_flags)
410{
411 loff_t ret;
412
413 file_start_write(file_out);
414 ret = do_clone_file_range(file_in, pos_in, file_out, pos_out, len,
415 remap_flags);
416 file_end_write(file_out);
417
418 return ret;
419}
420EXPORT_SYMBOL(vfs_clone_file_range);
421
422/* Check whether we are allowed to dedupe the destination file */
423static bool allow_file_dedupe(struct file *file)
424{
4609e1f1 425 struct mnt_idmap *idmap = file_mnt_idmap(file);
0f5d220b
CB
426 struct inode *inode = file_inode(file);
427
1b2c54d6
DW
428 if (capable(CAP_SYS_ADMIN))
429 return true;
430 if (file->f_mode & FMODE_WRITE)
431 return true;
e67fe633 432 if (vfsuid_eq_kuid(i_uid_into_vfsuid(idmap, inode), current_fsuid()))
1b2c54d6 433 return true;
4609e1f1 434 if (!inode_permission(idmap, inode, MAY_WRITE))
1b2c54d6
DW
435 return true;
436 return false;
437}
438
439loff_t vfs_dedupe_file_range_one(struct file *src_file, loff_t src_pos,
440 struct file *dst_file, loff_t dst_pos,
441 loff_t len, unsigned int remap_flags)
442{
443 loff_t ret;
444
445 WARN_ON_ONCE(remap_flags & ~(REMAP_FILE_DEDUP |
446 REMAP_FILE_CAN_SHORTEN));
447
448 ret = mnt_want_write_file(dst_file);
449 if (ret)
450 return ret;
451
3078d85c
MS
452 /*
453 * This is redundant if called from vfs_dedupe_file_range(), but other
454 * callers need it and it's not performance sesitive...
455 */
456 ret = remap_verify_area(src_file, src_pos, len, false);
457 if (ret)
458 goto out_drop_write;
459
1b2c54d6 460 ret = remap_verify_area(dst_file, dst_pos, len, true);
3078d85c 461 if (ret)
1b2c54d6
DW
462 goto out_drop_write;
463
464 ret = -EPERM;
465 if (!allow_file_dedupe(dst_file))
466 goto out_drop_write;
467
468 ret = -EXDEV;
9f5710bb 469 if (file_inode(src_file)->i_sb != file_inode(dst_file)->i_sb)
1b2c54d6
DW
470 goto out_drop_write;
471
472 ret = -EISDIR;
473 if (S_ISDIR(file_inode(dst_file)->i_mode))
474 goto out_drop_write;
475
476 ret = -EINVAL;
477 if (!dst_file->f_op->remap_file_range)
478 goto out_drop_write;
479
480 if (len == 0) {
481 ret = 0;
482 goto out_drop_write;
483 }
484
485 ret = dst_file->f_op->remap_file_range(src_file, src_pos, dst_file,
486 dst_pos, len, remap_flags | REMAP_FILE_DEDUP);
487out_drop_write:
488 mnt_drop_write_file(dst_file);
489
490 return ret;
491}
492EXPORT_SYMBOL(vfs_dedupe_file_range_one);
493
494int vfs_dedupe_file_range(struct file *file, struct file_dedupe_range *same)
495{
496 struct file_dedupe_range_info *info;
497 struct inode *src = file_inode(file);
498 u64 off;
499 u64 len;
500 int i;
501 int ret;
502 u16 count = same->dest_count;
503 loff_t deduped;
504
505 if (!(file->f_mode & FMODE_READ))
506 return -EINVAL;
507
508 if (same->reserved1 || same->reserved2)
509 return -EINVAL;
510
511 off = same->src_offset;
512 len = same->src_length;
513
514 if (S_ISDIR(src->i_mode))
515 return -EISDIR;
516
517 if (!S_ISREG(src->i_mode))
518 return -EINVAL;
519
520 if (!file->f_op->remap_file_range)
521 return -EOPNOTSUPP;
522
523 ret = remap_verify_area(file, off, len, false);
524 if (ret < 0)
525 return ret;
526 ret = 0;
527
528 if (off + len > i_size_read(src))
529 return -EINVAL;
530
531 /* Arbitrary 1G limit on a single dedupe request, can be raised. */
532 len = min_t(u64, len, 1 << 30);
533
534 /* pre-format output fields to sane values */
535 for (i = 0; i < count; i++) {
536 same->info[i].bytes_deduped = 0ULL;
537 same->info[i].status = FILE_DEDUPE_RANGE_SAME;
538 }
539
540 for (i = 0, info = same->info; i < count; i++, info++) {
541 struct fd dst_fd = fdget(info->dest_fd);
542 struct file *dst_file = dst_fd.file;
543
544 if (!dst_file) {
545 info->status = -EBADF;
546 goto next_loop;
547 }
548
549 if (info->reserved) {
550 info->status = -EINVAL;
551 goto next_fdput;
552 }
553
554 deduped = vfs_dedupe_file_range_one(file, off, dst_file,
555 info->dest_offset, len,
556 REMAP_FILE_CAN_SHORTEN);
557 if (deduped == -EBADE)
558 info->status = FILE_DEDUPE_RANGE_DIFFERS;
559 else if (deduped < 0)
560 info->status = deduped;
561 else
b926f2ad 562 info->bytes_deduped = len;
1b2c54d6
DW
563
564next_fdput:
565 fdput(dst_fd);
566next_loop:
567 if (fatal_signal_pending(current))
568 break;
569 }
570 return ret;
571}
572EXPORT_SYMBOL(vfs_dedupe_file_range);