Loading...
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Copyright (C) 2007 Oracle. All rights reserved.
4 */
5
6#include <linux/bio.h>
7#include <linux/slab.h>
8#include <linux/pagemap.h>
9#include <linux/highmem.h>
10#include <linux/sched/mm.h>
11#include <crypto/hash.h>
12#include "messages.h"
13#include "ctree.h"
14#include "disk-io.h"
15#include "transaction.h"
16#include "bio.h"
17#include "compression.h"
18#include "fs.h"
19#include "accessors.h"
20#include "file-item.h"
21
22#define __MAX_CSUM_ITEMS(r, size) ((unsigned long)(((BTRFS_LEAF_DATA_SIZE(r) - \
23 sizeof(struct btrfs_item) * 2) / \
24 size) - 1))
25
26#define MAX_CSUM_ITEMS(r, size) (min_t(u32, __MAX_CSUM_ITEMS(r, size), \
27 PAGE_SIZE))
28
29/*
30 * Set inode's size according to filesystem options.
31 *
32 * @inode: inode we want to update the disk_i_size for
33 * @new_i_size: i_size we want to set to, 0 if we use i_size
34 *
35 * With NO_HOLES set this simply sets the disk_is_size to whatever i_size_read()
36 * returns as it is perfectly fine with a file that has holes without hole file
37 * extent items.
38 *
39 * However without NO_HOLES we need to only return the area that is contiguous
40 * from the 0 offset of the file. Otherwise we could end up adjust i_size up
41 * to an extent that has a gap in between.
42 *
43 * Finally new_i_size should only be set in the case of truncate where we're not
44 * ready to use i_size_read() as the limiter yet.
45 */
46void btrfs_inode_safe_disk_i_size_write(struct btrfs_inode *inode, u64 new_i_size)
47{
48 u64 start, end, i_size;
49 int ret;
50
51 spin_lock(&inode->lock);
52 i_size = new_i_size ?: i_size_read(&inode->vfs_inode);
53 if (!inode->file_extent_tree) {
54 inode->disk_i_size = i_size;
55 goto out_unlock;
56 }
57
58 ret = find_contiguous_extent_bit(inode->file_extent_tree, 0, &start,
59 &end, EXTENT_DIRTY);
60 if (!ret && start == 0)
61 i_size = min(i_size, end + 1);
62 else
63 i_size = 0;
64 inode->disk_i_size = i_size;
65out_unlock:
66 spin_unlock(&inode->lock);
67}
68
69/*
70 * Mark range within a file as having a new extent inserted.
71 *
72 * @inode: inode being modified
73 * @start: start file offset of the file extent we've inserted
74 * @len: logical length of the file extent item
75 *
76 * Call when we are inserting a new file extent where there was none before.
77 * Does not need to call this in the case where we're replacing an existing file
78 * extent, however if not sure it's fine to call this multiple times.
79 *
80 * The start and len must match the file extent item, so thus must be sectorsize
81 * aligned.
82 */
83int btrfs_inode_set_file_extent_range(struct btrfs_inode *inode, u64 start,
84 u64 len)
85{
86 if (!inode->file_extent_tree)
87 return 0;
88
89 if (len == 0)
90 return 0;
91
92 ASSERT(IS_ALIGNED(start + len, inode->root->fs_info->sectorsize));
93
94 return set_extent_bit(inode->file_extent_tree, start, start + len - 1,
95 EXTENT_DIRTY, NULL);
96}
97
98/*
99 * Mark an inode range as not having a backing extent.
100 *
101 * @inode: inode being modified
102 * @start: start file offset of the file extent we've inserted
103 * @len: logical length of the file extent item
104 *
105 * Called when we drop a file extent, for example when we truncate. Doesn't
106 * need to be called for cases where we're replacing a file extent, like when
107 * we've COWed a file extent.
108 *
109 * The start and len must match the file extent item, so thus must be sectorsize
110 * aligned.
111 */
112int btrfs_inode_clear_file_extent_range(struct btrfs_inode *inode, u64 start,
113 u64 len)
114{
115 if (!inode->file_extent_tree)
116 return 0;
117
118 if (len == 0)
119 return 0;
120
121 ASSERT(IS_ALIGNED(start + len, inode->root->fs_info->sectorsize) ||
122 len == (u64)-1);
123
124 return clear_extent_bit(inode->file_extent_tree, start,
125 start + len - 1, EXTENT_DIRTY, NULL);
126}
127
128static size_t bytes_to_csum_size(const struct btrfs_fs_info *fs_info, u32 bytes)
129{
130 ASSERT(IS_ALIGNED(bytes, fs_info->sectorsize));
131
132 return (bytes >> fs_info->sectorsize_bits) * fs_info->csum_size;
133}
134
135static size_t csum_size_to_bytes(const struct btrfs_fs_info *fs_info, u32 csum_size)
136{
137 ASSERT(IS_ALIGNED(csum_size, fs_info->csum_size));
138
139 return (csum_size / fs_info->csum_size) << fs_info->sectorsize_bits;
140}
141
142static inline u32 max_ordered_sum_bytes(const struct btrfs_fs_info *fs_info)
143{
144 u32 max_csum_size = round_down(PAGE_SIZE - sizeof(struct btrfs_ordered_sum),
145 fs_info->csum_size);
146
147 return csum_size_to_bytes(fs_info, max_csum_size);
148}
149
150/*
151 * Calculate the total size needed to allocate for an ordered sum structure
152 * spanning @bytes in the file.
153 */
154static int btrfs_ordered_sum_size(const struct btrfs_fs_info *fs_info, unsigned long bytes)
155{
156 return sizeof(struct btrfs_ordered_sum) + bytes_to_csum_size(fs_info, bytes);
157}
158
159int btrfs_insert_hole_extent(struct btrfs_trans_handle *trans,
160 struct btrfs_root *root,
161 u64 objectid, u64 pos, u64 num_bytes)
162{
163 int ret = 0;
164 struct btrfs_file_extent_item *item;
165 struct btrfs_key file_key;
166 struct btrfs_path *path;
167 struct extent_buffer *leaf;
168
169 path = btrfs_alloc_path();
170 if (!path)
171 return -ENOMEM;
172 file_key.objectid = objectid;
173 file_key.offset = pos;
174 file_key.type = BTRFS_EXTENT_DATA_KEY;
175
176 ret = btrfs_insert_empty_item(trans, root, path, &file_key,
177 sizeof(*item));
178 if (ret < 0)
179 goto out;
180 leaf = path->nodes[0];
181 item = btrfs_item_ptr(leaf, path->slots[0],
182 struct btrfs_file_extent_item);
183 btrfs_set_file_extent_disk_bytenr(leaf, item, 0);
184 btrfs_set_file_extent_disk_num_bytes(leaf, item, 0);
185 btrfs_set_file_extent_offset(leaf, item, 0);
186 btrfs_set_file_extent_num_bytes(leaf, item, num_bytes);
187 btrfs_set_file_extent_ram_bytes(leaf, item, num_bytes);
188 btrfs_set_file_extent_generation(leaf, item, trans->transid);
189 btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
190 btrfs_set_file_extent_compression(leaf, item, 0);
191 btrfs_set_file_extent_encryption(leaf, item, 0);
192 btrfs_set_file_extent_other_encoding(leaf, item, 0);
193
194 btrfs_mark_buffer_dirty(trans, leaf);
195out:
196 btrfs_free_path(path);
197 return ret;
198}
199
200static struct btrfs_csum_item *
201btrfs_lookup_csum(struct btrfs_trans_handle *trans,
202 struct btrfs_root *root,
203 struct btrfs_path *path,
204 u64 bytenr, int cow)
205{
206 struct btrfs_fs_info *fs_info = root->fs_info;
207 int ret;
208 struct btrfs_key file_key;
209 struct btrfs_key found_key;
210 struct btrfs_csum_item *item;
211 struct extent_buffer *leaf;
212 u64 csum_offset = 0;
213 const u32 csum_size = fs_info->csum_size;
214 int csums_in_item;
215
216 file_key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
217 file_key.offset = bytenr;
218 file_key.type = BTRFS_EXTENT_CSUM_KEY;
219 ret = btrfs_search_slot(trans, root, &file_key, path, 0, cow);
220 if (ret < 0)
221 goto fail;
222 leaf = path->nodes[0];
223 if (ret > 0) {
224 ret = 1;
225 if (path->slots[0] == 0)
226 goto fail;
227 path->slots[0]--;
228 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
229 if (found_key.type != BTRFS_EXTENT_CSUM_KEY)
230 goto fail;
231
232 csum_offset = (bytenr - found_key.offset) >>
233 fs_info->sectorsize_bits;
234 csums_in_item = btrfs_item_size(leaf, path->slots[0]);
235 csums_in_item /= csum_size;
236
237 if (csum_offset == csums_in_item) {
238 ret = -EFBIG;
239 goto fail;
240 } else if (csum_offset > csums_in_item) {
241 goto fail;
242 }
243 }
244 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item);
245 item = (struct btrfs_csum_item *)((unsigned char *)item +
246 csum_offset * csum_size);
247 return item;
248fail:
249 if (ret > 0)
250 ret = -ENOENT;
251 return ERR_PTR(ret);
252}
253
254int btrfs_lookup_file_extent(struct btrfs_trans_handle *trans,
255 struct btrfs_root *root,
256 struct btrfs_path *path, u64 objectid,
257 u64 offset, int mod)
258{
259 struct btrfs_key file_key;
260 int ins_len = mod < 0 ? -1 : 0;
261 int cow = mod != 0;
262
263 file_key.objectid = objectid;
264 file_key.offset = offset;
265 file_key.type = BTRFS_EXTENT_DATA_KEY;
266
267 return btrfs_search_slot(trans, root, &file_key, path, ins_len, cow);
268}
269
270/*
271 * Find checksums for logical bytenr range [disk_bytenr, disk_bytenr + len) and
272 * store the result to @dst.
273 *
274 * Return >0 for the number of sectors we found.
275 * Return 0 for the range [disk_bytenr, disk_bytenr + sectorsize) has no csum
276 * for it. Caller may want to try next sector until one range is hit.
277 * Return <0 for fatal error.
278 */
279static int search_csum_tree(struct btrfs_fs_info *fs_info,
280 struct btrfs_path *path, u64 disk_bytenr,
281 u64 len, u8 *dst)
282{
283 struct btrfs_root *csum_root;
284 struct btrfs_csum_item *item = NULL;
285 struct btrfs_key key;
286 const u32 sectorsize = fs_info->sectorsize;
287 const u32 csum_size = fs_info->csum_size;
288 u32 itemsize;
289 int ret;
290 u64 csum_start;
291 u64 csum_len;
292
293 ASSERT(IS_ALIGNED(disk_bytenr, sectorsize) &&
294 IS_ALIGNED(len, sectorsize));
295
296 /* Check if the current csum item covers disk_bytenr */
297 if (path->nodes[0]) {
298 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
299 struct btrfs_csum_item);
300 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
301 itemsize = btrfs_item_size(path->nodes[0], path->slots[0]);
302
303 csum_start = key.offset;
304 csum_len = (itemsize / csum_size) * sectorsize;
305
306 if (in_range(disk_bytenr, csum_start, csum_len))
307 goto found;
308 }
309
310 /* Current item doesn't contain the desired range, search again */
311 btrfs_release_path(path);
312 csum_root = btrfs_csum_root(fs_info, disk_bytenr);
313 item = btrfs_lookup_csum(NULL, csum_root, path, disk_bytenr, 0);
314 if (IS_ERR(item)) {
315 ret = PTR_ERR(item);
316 goto out;
317 }
318 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
319 itemsize = btrfs_item_size(path->nodes[0], path->slots[0]);
320
321 csum_start = key.offset;
322 csum_len = (itemsize / csum_size) * sectorsize;
323 ASSERT(in_range(disk_bytenr, csum_start, csum_len));
324
325found:
326 ret = (min(csum_start + csum_len, disk_bytenr + len) -
327 disk_bytenr) >> fs_info->sectorsize_bits;
328 read_extent_buffer(path->nodes[0], dst, (unsigned long)item,
329 ret * csum_size);
330out:
331 if (ret == -ENOENT || ret == -EFBIG)
332 ret = 0;
333 return ret;
334}
335
336/*
337 * Lookup the checksum for the read bio in csum tree.
338 *
339 * Return: BLK_STS_RESOURCE if allocating memory fails, BLK_STS_OK otherwise.
340 */
341blk_status_t btrfs_lookup_bio_sums(struct btrfs_bio *bbio)
342{
343 struct btrfs_inode *inode = bbio->inode;
344 struct btrfs_fs_info *fs_info = inode->root->fs_info;
345 struct bio *bio = &bbio->bio;
346 struct btrfs_path *path;
347 const u32 sectorsize = fs_info->sectorsize;
348 const u32 csum_size = fs_info->csum_size;
349 u32 orig_len = bio->bi_iter.bi_size;
350 u64 orig_disk_bytenr = bio->bi_iter.bi_sector << SECTOR_SHIFT;
351 const unsigned int nblocks = orig_len >> fs_info->sectorsize_bits;
352 blk_status_t ret = BLK_STS_OK;
353 u32 bio_offset = 0;
354
355 if ((inode->flags & BTRFS_INODE_NODATASUM) ||
356 test_bit(BTRFS_FS_STATE_NO_DATA_CSUMS, &fs_info->fs_state))
357 return BLK_STS_OK;
358
359 /*
360 * This function is only called for read bio.
361 *
362 * This means two things:
363 * - All our csums should only be in csum tree
364 * No ordered extents csums, as ordered extents are only for write
365 * path.
366 * - No need to bother any other info from bvec
367 * Since we're looking up csums, the only important info is the
368 * disk_bytenr and the length, which can be extracted from bi_iter
369 * directly.
370 */
371 ASSERT(bio_op(bio) == REQ_OP_READ);
372 path = btrfs_alloc_path();
373 if (!path)
374 return BLK_STS_RESOURCE;
375
376 if (nblocks * csum_size > BTRFS_BIO_INLINE_CSUM_SIZE) {
377 bbio->csum = kmalloc_array(nblocks, csum_size, GFP_NOFS);
378 if (!bbio->csum) {
379 btrfs_free_path(path);
380 return BLK_STS_RESOURCE;
381 }
382 } else {
383 bbio->csum = bbio->csum_inline;
384 }
385
386 /*
387 * If requested number of sectors is larger than one leaf can contain,
388 * kick the readahead for csum tree.
389 */
390 if (nblocks > fs_info->csums_per_leaf)
391 path->reada = READA_FORWARD;
392
393 /*
394 * the free space stuff is only read when it hasn't been
395 * updated in the current transaction. So, we can safely
396 * read from the commit root and sidestep a nasty deadlock
397 * between reading the free space cache and updating the csum tree.
398 */
399 if (btrfs_is_free_space_inode(inode)) {
400 path->search_commit_root = 1;
401 path->skip_locking = 1;
402 }
403
404 while (bio_offset < orig_len) {
405 int count;
406 u64 cur_disk_bytenr = orig_disk_bytenr + bio_offset;
407 u8 *csum_dst = bbio->csum +
408 (bio_offset >> fs_info->sectorsize_bits) * csum_size;
409
410 count = search_csum_tree(fs_info, path, cur_disk_bytenr,
411 orig_len - bio_offset, csum_dst);
412 if (count < 0) {
413 ret = errno_to_blk_status(count);
414 if (bbio->csum != bbio->csum_inline)
415 kfree(bbio->csum);
416 bbio->csum = NULL;
417 break;
418 }
419
420 /*
421 * We didn't find a csum for this range. We need to make sure
422 * we complain loudly about this, because we are not NODATASUM.
423 *
424 * However for the DATA_RELOC inode we could potentially be
425 * relocating data extents for a NODATASUM inode, so the inode
426 * itself won't be marked with NODATASUM, but the extent we're
427 * copying is in fact NODATASUM. If we don't find a csum we
428 * assume this is the case.
429 */
430 if (count == 0) {
431 memset(csum_dst, 0, csum_size);
432 count = 1;
433
434 if (btrfs_root_id(inode->root) == BTRFS_DATA_RELOC_TREE_OBJECTID) {
435 u64 file_offset = bbio->file_offset + bio_offset;
436
437 set_extent_bit(&inode->io_tree, file_offset,
438 file_offset + sectorsize - 1,
439 EXTENT_NODATASUM, NULL);
440 } else {
441 btrfs_warn_rl(fs_info,
442 "csum hole found for disk bytenr range [%llu, %llu)",
443 cur_disk_bytenr, cur_disk_bytenr + sectorsize);
444 }
445 }
446 bio_offset += count * sectorsize;
447 }
448
449 btrfs_free_path(path);
450 return ret;
451}
452
453/*
454 * Search for checksums for a given logical range.
455 *
456 * @root: The root where to look for checksums.
457 * @start: Logical address of target checksum range.
458 * @end: End offset (inclusive) of the target checksum range.
459 * @list: List for adding each checksum that was found.
460 * Can be NULL in case the caller only wants to check if
461 * there any checksums for the range.
462 * @nowait: Indicate if the search must be non-blocking or not.
463 *
464 * Return < 0 on error, 0 if no checksums were found, or 1 if checksums were
465 * found.
466 */
467int btrfs_lookup_csums_list(struct btrfs_root *root, u64 start, u64 end,
468 struct list_head *list, bool nowait)
469{
470 struct btrfs_fs_info *fs_info = root->fs_info;
471 struct btrfs_key key;
472 struct btrfs_path *path;
473 struct extent_buffer *leaf;
474 struct btrfs_ordered_sum *sums;
475 struct btrfs_csum_item *item;
476 int ret;
477 bool found_csums = false;
478
479 ASSERT(IS_ALIGNED(start, fs_info->sectorsize) &&
480 IS_ALIGNED(end + 1, fs_info->sectorsize));
481
482 path = btrfs_alloc_path();
483 if (!path)
484 return -ENOMEM;
485
486 path->nowait = nowait;
487
488 key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
489 key.offset = start;
490 key.type = BTRFS_EXTENT_CSUM_KEY;
491
492 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
493 if (ret < 0)
494 goto out;
495 if (ret > 0 && path->slots[0] > 0) {
496 leaf = path->nodes[0];
497 btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
498
499 /*
500 * There are two cases we can hit here for the previous csum
501 * item:
502 *
503 * |<- search range ->|
504 * |<- csum item ->|
505 *
506 * Or
507 * |<- search range ->|
508 * |<- csum item ->|
509 *
510 * Check if the previous csum item covers the leading part of
511 * the search range. If so we have to start from previous csum
512 * item.
513 */
514 if (key.objectid == BTRFS_EXTENT_CSUM_OBJECTID &&
515 key.type == BTRFS_EXTENT_CSUM_KEY) {
516 if (bytes_to_csum_size(fs_info, start - key.offset) <
517 btrfs_item_size(leaf, path->slots[0] - 1))
518 path->slots[0]--;
519 }
520 }
521
522 while (start <= end) {
523 u64 csum_end;
524
525 leaf = path->nodes[0];
526 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
527 ret = btrfs_next_leaf(root, path);
528 if (ret < 0)
529 goto out;
530 if (ret > 0)
531 break;
532 leaf = path->nodes[0];
533 }
534
535 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
536 if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
537 key.type != BTRFS_EXTENT_CSUM_KEY ||
538 key.offset > end)
539 break;
540
541 if (key.offset > start)
542 start = key.offset;
543
544 csum_end = key.offset + csum_size_to_bytes(fs_info,
545 btrfs_item_size(leaf, path->slots[0]));
546 if (csum_end <= start) {
547 path->slots[0]++;
548 continue;
549 }
550
551 found_csums = true;
552 if (!list)
553 goto out;
554
555 csum_end = min(csum_end, end + 1);
556 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
557 struct btrfs_csum_item);
558 while (start < csum_end) {
559 unsigned long offset;
560 size_t size;
561
562 size = min_t(size_t, csum_end - start,
563 max_ordered_sum_bytes(fs_info));
564 sums = kzalloc(btrfs_ordered_sum_size(fs_info, size),
565 GFP_NOFS);
566 if (!sums) {
567 ret = -ENOMEM;
568 goto out;
569 }
570
571 sums->logical = start;
572 sums->len = size;
573
574 offset = bytes_to_csum_size(fs_info, start - key.offset);
575
576 read_extent_buffer(path->nodes[0],
577 sums->sums,
578 ((unsigned long)item) + offset,
579 bytes_to_csum_size(fs_info, size));
580
581 start += size;
582 list_add_tail(&sums->list, list);
583 }
584 path->slots[0]++;
585 }
586out:
587 btrfs_free_path(path);
588 if (ret < 0) {
589 if (list) {
590 struct btrfs_ordered_sum *tmp_sums;
591
592 list_for_each_entry_safe(sums, tmp_sums, list, list)
593 kfree(sums);
594 }
595
596 return ret;
597 }
598
599 return found_csums ? 1 : 0;
600}
601
602/*
603 * Do the same work as btrfs_lookup_csums_list(), the difference is in how
604 * we return the result.
605 *
606 * This version will set the corresponding bits in @csum_bitmap to represent
607 * that there is a csum found.
608 * Each bit represents a sector. Thus caller should ensure @csum_buf passed
609 * in is large enough to contain all csums.
610 */
611int btrfs_lookup_csums_bitmap(struct btrfs_root *root, struct btrfs_path *path,
612 u64 start, u64 end, u8 *csum_buf,
613 unsigned long *csum_bitmap)
614{
615 struct btrfs_fs_info *fs_info = root->fs_info;
616 struct btrfs_key key;
617 struct extent_buffer *leaf;
618 struct btrfs_csum_item *item;
619 const u64 orig_start = start;
620 bool free_path = false;
621 int ret;
622
623 ASSERT(IS_ALIGNED(start, fs_info->sectorsize) &&
624 IS_ALIGNED(end + 1, fs_info->sectorsize));
625
626 if (!path) {
627 path = btrfs_alloc_path();
628 if (!path)
629 return -ENOMEM;
630 free_path = true;
631 }
632
633 /* Check if we can reuse the previous path. */
634 if (path->nodes[0]) {
635 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
636
637 if (key.objectid == BTRFS_EXTENT_CSUM_OBJECTID &&
638 key.type == BTRFS_EXTENT_CSUM_KEY &&
639 key.offset <= start)
640 goto search_forward;
641 btrfs_release_path(path);
642 }
643
644 key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
645 key.type = BTRFS_EXTENT_CSUM_KEY;
646 key.offset = start;
647
648 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
649 if (ret < 0)
650 goto fail;
651 if (ret > 0 && path->slots[0] > 0) {
652 leaf = path->nodes[0];
653 btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
654
655 /*
656 * There are two cases we can hit here for the previous csum
657 * item:
658 *
659 * |<- search range ->|
660 * |<- csum item ->|
661 *
662 * Or
663 * |<- search range ->|
664 * |<- csum item ->|
665 *
666 * Check if the previous csum item covers the leading part of
667 * the search range. If so we have to start from previous csum
668 * item.
669 */
670 if (key.objectid == BTRFS_EXTENT_CSUM_OBJECTID &&
671 key.type == BTRFS_EXTENT_CSUM_KEY) {
672 if (bytes_to_csum_size(fs_info, start - key.offset) <
673 btrfs_item_size(leaf, path->slots[0] - 1))
674 path->slots[0]--;
675 }
676 }
677
678search_forward:
679 while (start <= end) {
680 u64 csum_end;
681
682 leaf = path->nodes[0];
683 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
684 ret = btrfs_next_leaf(root, path);
685 if (ret < 0)
686 goto fail;
687 if (ret > 0)
688 break;
689 leaf = path->nodes[0];
690 }
691
692 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
693 if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
694 key.type != BTRFS_EXTENT_CSUM_KEY ||
695 key.offset > end)
696 break;
697
698 if (key.offset > start)
699 start = key.offset;
700
701 csum_end = key.offset + csum_size_to_bytes(fs_info,
702 btrfs_item_size(leaf, path->slots[0]));
703 if (csum_end <= start) {
704 path->slots[0]++;
705 continue;
706 }
707
708 csum_end = min(csum_end, end + 1);
709 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
710 struct btrfs_csum_item);
711 while (start < csum_end) {
712 unsigned long offset;
713 size_t size;
714 u8 *csum_dest = csum_buf + bytes_to_csum_size(fs_info,
715 start - orig_start);
716
717 size = min_t(size_t, csum_end - start, end + 1 - start);
718
719 offset = bytes_to_csum_size(fs_info, start - key.offset);
720
721 read_extent_buffer(path->nodes[0], csum_dest,
722 ((unsigned long)item) + offset,
723 bytes_to_csum_size(fs_info, size));
724
725 bitmap_set(csum_bitmap,
726 (start - orig_start) >> fs_info->sectorsize_bits,
727 size >> fs_info->sectorsize_bits);
728
729 start += size;
730 }
731 path->slots[0]++;
732 }
733 ret = 0;
734fail:
735 if (free_path)
736 btrfs_free_path(path);
737 return ret;
738}
739
740/*
741 * Calculate checksums of the data contained inside a bio.
742 */
743blk_status_t btrfs_csum_one_bio(struct btrfs_bio *bbio)
744{
745 struct btrfs_ordered_extent *ordered = bbio->ordered;
746 struct btrfs_inode *inode = bbio->inode;
747 struct btrfs_fs_info *fs_info = inode->root->fs_info;
748 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
749 struct bio *bio = &bbio->bio;
750 struct btrfs_ordered_sum *sums;
751 char *data;
752 struct bvec_iter iter;
753 struct bio_vec bvec;
754 int index;
755 unsigned int blockcount;
756 int i;
757 unsigned nofs_flag;
758
759 nofs_flag = memalloc_nofs_save();
760 sums = kvzalloc(btrfs_ordered_sum_size(fs_info, bio->bi_iter.bi_size),
761 GFP_KERNEL);
762 memalloc_nofs_restore(nofs_flag);
763
764 if (!sums)
765 return BLK_STS_RESOURCE;
766
767 sums->len = bio->bi_iter.bi_size;
768 INIT_LIST_HEAD(&sums->list);
769
770 sums->logical = bio->bi_iter.bi_sector << SECTOR_SHIFT;
771 index = 0;
772
773 shash->tfm = fs_info->csum_shash;
774
775 bio_for_each_segment(bvec, bio, iter) {
776 blockcount = BTRFS_BYTES_TO_BLKS(fs_info,
777 bvec.bv_len + fs_info->sectorsize
778 - 1);
779
780 for (i = 0; i < blockcount; i++) {
781 data = bvec_kmap_local(&bvec);
782 crypto_shash_digest(shash,
783 data + (i * fs_info->sectorsize),
784 fs_info->sectorsize,
785 sums->sums + index);
786 kunmap_local(data);
787 index += fs_info->csum_size;
788 }
789
790 }
791
792 bbio->sums = sums;
793 btrfs_add_ordered_sum(ordered, sums);
794 return 0;
795}
796
797/*
798 * Nodatasum I/O on zoned file systems still requires an btrfs_ordered_sum to
799 * record the updated logical address on Zone Append completion.
800 * Allocate just the structure with an empty sums array here for that case.
801 */
802blk_status_t btrfs_alloc_dummy_sum(struct btrfs_bio *bbio)
803{
804 bbio->sums = kmalloc(sizeof(*bbio->sums), GFP_NOFS);
805 if (!bbio->sums)
806 return BLK_STS_RESOURCE;
807 bbio->sums->len = bbio->bio.bi_iter.bi_size;
808 bbio->sums->logical = bbio->bio.bi_iter.bi_sector << SECTOR_SHIFT;
809 btrfs_add_ordered_sum(bbio->ordered, bbio->sums);
810 return 0;
811}
812
813/*
814 * Remove one checksum overlapping a range.
815 *
816 * This expects the key to describe the csum pointed to by the path, and it
817 * expects the csum to overlap the range [bytenr, len]
818 *
819 * The csum should not be entirely contained in the range and the range should
820 * not be entirely contained in the csum.
821 *
822 * This calls btrfs_truncate_item with the correct args based on the overlap,
823 * and fixes up the key as required.
824 */
825static noinline void truncate_one_csum(struct btrfs_trans_handle *trans,
826 struct btrfs_path *path,
827 struct btrfs_key *key,
828 u64 bytenr, u64 len)
829{
830 struct btrfs_fs_info *fs_info = trans->fs_info;
831 struct extent_buffer *leaf;
832 const u32 csum_size = fs_info->csum_size;
833 u64 csum_end;
834 u64 end_byte = bytenr + len;
835 u32 blocksize_bits = fs_info->sectorsize_bits;
836
837 leaf = path->nodes[0];
838 csum_end = btrfs_item_size(leaf, path->slots[0]) / csum_size;
839 csum_end <<= blocksize_bits;
840 csum_end += key->offset;
841
842 if (key->offset < bytenr && csum_end <= end_byte) {
843 /*
844 * [ bytenr - len ]
845 * [ ]
846 * [csum ]
847 * A simple truncate off the end of the item
848 */
849 u32 new_size = (bytenr - key->offset) >> blocksize_bits;
850 new_size *= csum_size;
851 btrfs_truncate_item(trans, path, new_size, 1);
852 } else if (key->offset >= bytenr && csum_end > end_byte &&
853 end_byte > key->offset) {
854 /*
855 * [ bytenr - len ]
856 * [ ]
857 * [csum ]
858 * we need to truncate from the beginning of the csum
859 */
860 u32 new_size = (csum_end - end_byte) >> blocksize_bits;
861 new_size *= csum_size;
862
863 btrfs_truncate_item(trans, path, new_size, 0);
864
865 key->offset = end_byte;
866 btrfs_set_item_key_safe(trans, path, key);
867 } else {
868 BUG();
869 }
870}
871
872/*
873 * Delete the csum items from the csum tree for a given range of bytes.
874 */
875int btrfs_del_csums(struct btrfs_trans_handle *trans,
876 struct btrfs_root *root, u64 bytenr, u64 len)
877{
878 struct btrfs_fs_info *fs_info = trans->fs_info;
879 struct btrfs_path *path;
880 struct btrfs_key key;
881 u64 end_byte = bytenr + len;
882 u64 csum_end;
883 struct extent_buffer *leaf;
884 int ret = 0;
885 const u32 csum_size = fs_info->csum_size;
886 u32 blocksize_bits = fs_info->sectorsize_bits;
887
888 ASSERT(btrfs_root_id(root) == BTRFS_CSUM_TREE_OBJECTID ||
889 btrfs_root_id(root) == BTRFS_TREE_LOG_OBJECTID);
890
891 path = btrfs_alloc_path();
892 if (!path)
893 return -ENOMEM;
894
895 while (1) {
896 key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
897 key.offset = end_byte - 1;
898 key.type = BTRFS_EXTENT_CSUM_KEY;
899
900 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
901 if (ret > 0) {
902 ret = 0;
903 if (path->slots[0] == 0)
904 break;
905 path->slots[0]--;
906 } else if (ret < 0) {
907 break;
908 }
909
910 leaf = path->nodes[0];
911 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
912
913 if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
914 key.type != BTRFS_EXTENT_CSUM_KEY) {
915 break;
916 }
917
918 if (key.offset >= end_byte)
919 break;
920
921 csum_end = btrfs_item_size(leaf, path->slots[0]) / csum_size;
922 csum_end <<= blocksize_bits;
923 csum_end += key.offset;
924
925 /* this csum ends before we start, we're done */
926 if (csum_end <= bytenr)
927 break;
928
929 /* delete the entire item, it is inside our range */
930 if (key.offset >= bytenr && csum_end <= end_byte) {
931 int del_nr = 1;
932
933 /*
934 * Check how many csum items preceding this one in this
935 * leaf correspond to our range and then delete them all
936 * at once.
937 */
938 if (key.offset > bytenr && path->slots[0] > 0) {
939 int slot = path->slots[0] - 1;
940
941 while (slot >= 0) {
942 struct btrfs_key pk;
943
944 btrfs_item_key_to_cpu(leaf, &pk, slot);
945 if (pk.offset < bytenr ||
946 pk.type != BTRFS_EXTENT_CSUM_KEY ||
947 pk.objectid !=
948 BTRFS_EXTENT_CSUM_OBJECTID)
949 break;
950 path->slots[0] = slot;
951 del_nr++;
952 key.offset = pk.offset;
953 slot--;
954 }
955 }
956 ret = btrfs_del_items(trans, root, path,
957 path->slots[0], del_nr);
958 if (ret)
959 break;
960 if (key.offset == bytenr)
961 break;
962 } else if (key.offset < bytenr && csum_end > end_byte) {
963 unsigned long offset;
964 unsigned long shift_len;
965 unsigned long item_offset;
966 /*
967 * [ bytenr - len ]
968 * [csum ]
969 *
970 * Our bytes are in the middle of the csum,
971 * we need to split this item and insert a new one.
972 *
973 * But we can't drop the path because the
974 * csum could change, get removed, extended etc.
975 *
976 * The trick here is the max size of a csum item leaves
977 * enough room in the tree block for a single
978 * item header. So, we split the item in place,
979 * adding a new header pointing to the existing
980 * bytes. Then we loop around again and we have
981 * a nicely formed csum item that we can neatly
982 * truncate.
983 */
984 offset = (bytenr - key.offset) >> blocksize_bits;
985 offset *= csum_size;
986
987 shift_len = (len >> blocksize_bits) * csum_size;
988
989 item_offset = btrfs_item_ptr_offset(leaf,
990 path->slots[0]);
991
992 memzero_extent_buffer(leaf, item_offset + offset,
993 shift_len);
994 key.offset = bytenr;
995
996 /*
997 * btrfs_split_item returns -EAGAIN when the
998 * item changed size or key
999 */
1000 ret = btrfs_split_item(trans, root, path, &key, offset);
1001 if (ret && ret != -EAGAIN) {
1002 btrfs_abort_transaction(trans, ret);
1003 break;
1004 }
1005 ret = 0;
1006
1007 key.offset = end_byte - 1;
1008 } else {
1009 truncate_one_csum(trans, path, &key, bytenr, len);
1010 if (key.offset < bytenr)
1011 break;
1012 }
1013 btrfs_release_path(path);
1014 }
1015 btrfs_free_path(path);
1016 return ret;
1017}
1018
1019static int find_next_csum_offset(struct btrfs_root *root,
1020 struct btrfs_path *path,
1021 u64 *next_offset)
1022{
1023 const u32 nritems = btrfs_header_nritems(path->nodes[0]);
1024 struct btrfs_key found_key;
1025 int slot = path->slots[0] + 1;
1026 int ret;
1027
1028 if (nritems == 0 || slot >= nritems) {
1029 ret = btrfs_next_leaf(root, path);
1030 if (ret < 0) {
1031 return ret;
1032 } else if (ret > 0) {
1033 *next_offset = (u64)-1;
1034 return 0;
1035 }
1036 slot = path->slots[0];
1037 }
1038
1039 btrfs_item_key_to_cpu(path->nodes[0], &found_key, slot);
1040
1041 if (found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
1042 found_key.type != BTRFS_EXTENT_CSUM_KEY)
1043 *next_offset = (u64)-1;
1044 else
1045 *next_offset = found_key.offset;
1046
1047 return 0;
1048}
1049
1050int btrfs_csum_file_blocks(struct btrfs_trans_handle *trans,
1051 struct btrfs_root *root,
1052 struct btrfs_ordered_sum *sums)
1053{
1054 struct btrfs_fs_info *fs_info = root->fs_info;
1055 struct btrfs_key file_key;
1056 struct btrfs_key found_key;
1057 struct btrfs_path *path;
1058 struct btrfs_csum_item *item;
1059 struct btrfs_csum_item *item_end;
1060 struct extent_buffer *leaf = NULL;
1061 u64 next_offset;
1062 u64 total_bytes = 0;
1063 u64 csum_offset;
1064 u64 bytenr;
1065 u32 ins_size;
1066 int index = 0;
1067 int found_next;
1068 int ret;
1069 const u32 csum_size = fs_info->csum_size;
1070
1071 path = btrfs_alloc_path();
1072 if (!path)
1073 return -ENOMEM;
1074again:
1075 next_offset = (u64)-1;
1076 found_next = 0;
1077 bytenr = sums->logical + total_bytes;
1078 file_key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
1079 file_key.offset = bytenr;
1080 file_key.type = BTRFS_EXTENT_CSUM_KEY;
1081
1082 item = btrfs_lookup_csum(trans, root, path, bytenr, 1);
1083 if (!IS_ERR(item)) {
1084 ret = 0;
1085 leaf = path->nodes[0];
1086 item_end = btrfs_item_ptr(leaf, path->slots[0],
1087 struct btrfs_csum_item);
1088 item_end = (struct btrfs_csum_item *)((char *)item_end +
1089 btrfs_item_size(leaf, path->slots[0]));
1090 goto found;
1091 }
1092 ret = PTR_ERR(item);
1093 if (ret != -EFBIG && ret != -ENOENT)
1094 goto out;
1095
1096 if (ret == -EFBIG) {
1097 u32 item_size;
1098 /* we found one, but it isn't big enough yet */
1099 leaf = path->nodes[0];
1100 item_size = btrfs_item_size(leaf, path->slots[0]);
1101 if ((item_size / csum_size) >=
1102 MAX_CSUM_ITEMS(fs_info, csum_size)) {
1103 /* already at max size, make a new one */
1104 goto insert;
1105 }
1106 } else {
1107 /* We didn't find a csum item, insert one. */
1108 ret = find_next_csum_offset(root, path, &next_offset);
1109 if (ret < 0)
1110 goto out;
1111 found_next = 1;
1112 goto insert;
1113 }
1114
1115 /*
1116 * At this point, we know the tree has a checksum item that ends at an
1117 * offset matching the start of the checksum range we want to insert.
1118 * We try to extend that item as much as possible and then add as many
1119 * checksums to it as they fit.
1120 *
1121 * First check if the leaf has enough free space for at least one
1122 * checksum. If it has go directly to the item extension code, otherwise
1123 * release the path and do a search for insertion before the extension.
1124 */
1125 if (btrfs_leaf_free_space(leaf) >= csum_size) {
1126 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1127 csum_offset = (bytenr - found_key.offset) >>
1128 fs_info->sectorsize_bits;
1129 goto extend_csum;
1130 }
1131
1132 btrfs_release_path(path);
1133 path->search_for_extension = 1;
1134 ret = btrfs_search_slot(trans, root, &file_key, path,
1135 csum_size, 1);
1136 path->search_for_extension = 0;
1137 if (ret < 0)
1138 goto out;
1139
1140 if (ret > 0) {
1141 if (path->slots[0] == 0)
1142 goto insert;
1143 path->slots[0]--;
1144 }
1145
1146 leaf = path->nodes[0];
1147 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1148 csum_offset = (bytenr - found_key.offset) >> fs_info->sectorsize_bits;
1149
1150 if (found_key.type != BTRFS_EXTENT_CSUM_KEY ||
1151 found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
1152 csum_offset >= MAX_CSUM_ITEMS(fs_info, csum_size)) {
1153 goto insert;
1154 }
1155
1156extend_csum:
1157 if (csum_offset == btrfs_item_size(leaf, path->slots[0]) /
1158 csum_size) {
1159 int extend_nr;
1160 u64 tmp;
1161 u32 diff;
1162
1163 tmp = sums->len - total_bytes;
1164 tmp >>= fs_info->sectorsize_bits;
1165 WARN_ON(tmp < 1);
1166 extend_nr = max_t(int, 1, tmp);
1167
1168 /*
1169 * A log tree can already have checksum items with a subset of
1170 * the checksums we are trying to log. This can happen after
1171 * doing a sequence of partial writes into prealloc extents and
1172 * fsyncs in between, with a full fsync logging a larger subrange
1173 * of an extent for which a previous fast fsync logged a smaller
1174 * subrange. And this happens in particular due to merging file
1175 * extent items when we complete an ordered extent for a range
1176 * covered by a prealloc extent - this is done at
1177 * btrfs_mark_extent_written().
1178 *
1179 * So if we try to extend the previous checksum item, which has
1180 * a range that ends at the start of the range we want to insert,
1181 * make sure we don't extend beyond the start offset of the next
1182 * checksum item. If we are at the last item in the leaf, then
1183 * forget the optimization of extending and add a new checksum
1184 * item - it is not worth the complexity of releasing the path,
1185 * getting the first key for the next leaf, repeat the btree
1186 * search, etc, because log trees are temporary anyway and it
1187 * would only save a few bytes of leaf space.
1188 */
1189 if (btrfs_root_id(root) == BTRFS_TREE_LOG_OBJECTID) {
1190 if (path->slots[0] + 1 >=
1191 btrfs_header_nritems(path->nodes[0])) {
1192 ret = find_next_csum_offset(root, path, &next_offset);
1193 if (ret < 0)
1194 goto out;
1195 found_next = 1;
1196 goto insert;
1197 }
1198
1199 ret = find_next_csum_offset(root, path, &next_offset);
1200 if (ret < 0)
1201 goto out;
1202
1203 tmp = (next_offset - bytenr) >> fs_info->sectorsize_bits;
1204 if (tmp <= INT_MAX)
1205 extend_nr = min_t(int, extend_nr, tmp);
1206 }
1207
1208 diff = (csum_offset + extend_nr) * csum_size;
1209 diff = min(diff,
1210 MAX_CSUM_ITEMS(fs_info, csum_size) * csum_size);
1211
1212 diff = diff - btrfs_item_size(leaf, path->slots[0]);
1213 diff = min_t(u32, btrfs_leaf_free_space(leaf), diff);
1214 diff /= csum_size;
1215 diff *= csum_size;
1216
1217 btrfs_extend_item(trans, path, diff);
1218 ret = 0;
1219 goto csum;
1220 }
1221
1222insert:
1223 btrfs_release_path(path);
1224 csum_offset = 0;
1225 if (found_next) {
1226 u64 tmp;
1227
1228 tmp = sums->len - total_bytes;
1229 tmp >>= fs_info->sectorsize_bits;
1230 tmp = min(tmp, (next_offset - file_key.offset) >>
1231 fs_info->sectorsize_bits);
1232
1233 tmp = max_t(u64, 1, tmp);
1234 tmp = min_t(u64, tmp, MAX_CSUM_ITEMS(fs_info, csum_size));
1235 ins_size = csum_size * tmp;
1236 } else {
1237 ins_size = csum_size;
1238 }
1239 ret = btrfs_insert_empty_item(trans, root, path, &file_key,
1240 ins_size);
1241 if (ret < 0)
1242 goto out;
1243 leaf = path->nodes[0];
1244csum:
1245 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item);
1246 item_end = (struct btrfs_csum_item *)((unsigned char *)item +
1247 btrfs_item_size(leaf, path->slots[0]));
1248 item = (struct btrfs_csum_item *)((unsigned char *)item +
1249 csum_offset * csum_size);
1250found:
1251 ins_size = (u32)(sums->len - total_bytes) >> fs_info->sectorsize_bits;
1252 ins_size *= csum_size;
1253 ins_size = min_t(u32, (unsigned long)item_end - (unsigned long)item,
1254 ins_size);
1255 write_extent_buffer(leaf, sums->sums + index, (unsigned long)item,
1256 ins_size);
1257
1258 index += ins_size;
1259 ins_size /= csum_size;
1260 total_bytes += ins_size * fs_info->sectorsize;
1261
1262 btrfs_mark_buffer_dirty(trans, path->nodes[0]);
1263 if (total_bytes < sums->len) {
1264 btrfs_release_path(path);
1265 cond_resched();
1266 goto again;
1267 }
1268out:
1269 btrfs_free_path(path);
1270 return ret;
1271}
1272
1273void btrfs_extent_item_to_extent_map(struct btrfs_inode *inode,
1274 const struct btrfs_path *path,
1275 const struct btrfs_file_extent_item *fi,
1276 struct extent_map *em)
1277{
1278 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1279 struct btrfs_root *root = inode->root;
1280 struct extent_buffer *leaf = path->nodes[0];
1281 const int slot = path->slots[0];
1282 struct btrfs_key key;
1283 u64 extent_start;
1284 u8 type = btrfs_file_extent_type(leaf, fi);
1285 int compress_type = btrfs_file_extent_compression(leaf, fi);
1286
1287 btrfs_item_key_to_cpu(leaf, &key, slot);
1288 extent_start = key.offset;
1289 em->ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1290 em->generation = btrfs_file_extent_generation(leaf, fi);
1291 if (type == BTRFS_FILE_EXTENT_REG ||
1292 type == BTRFS_FILE_EXTENT_PREALLOC) {
1293 const u64 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1294
1295 em->start = extent_start;
1296 em->len = btrfs_file_extent_end(path) - extent_start;
1297 if (disk_bytenr == 0) {
1298 em->disk_bytenr = EXTENT_MAP_HOLE;
1299 em->disk_num_bytes = 0;
1300 em->offset = 0;
1301 return;
1302 }
1303 em->disk_bytenr = disk_bytenr;
1304 em->disk_num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
1305 em->offset = btrfs_file_extent_offset(leaf, fi);
1306 if (compress_type != BTRFS_COMPRESS_NONE) {
1307 extent_map_set_compression(em, compress_type);
1308 } else {
1309 /*
1310 * Older kernels can create regular non-hole data
1311 * extents with ram_bytes smaller than disk_num_bytes.
1312 * Not a big deal, just always use disk_num_bytes
1313 * for ram_bytes.
1314 */
1315 em->ram_bytes = em->disk_num_bytes;
1316 if (type == BTRFS_FILE_EXTENT_PREALLOC)
1317 em->flags |= EXTENT_FLAG_PREALLOC;
1318 }
1319 } else if (type == BTRFS_FILE_EXTENT_INLINE) {
1320 /* Tree-checker has ensured this. */
1321 ASSERT(extent_start == 0);
1322
1323 em->disk_bytenr = EXTENT_MAP_INLINE;
1324 em->start = 0;
1325 em->len = fs_info->sectorsize;
1326 em->offset = 0;
1327 extent_map_set_compression(em, compress_type);
1328 } else {
1329 btrfs_err(fs_info,
1330 "unknown file extent item type %d, inode %llu, offset %llu, "
1331 "root %llu", type, btrfs_ino(inode), extent_start,
1332 btrfs_root_id(root));
1333 }
1334}
1335
1336/*
1337 * Returns the end offset (non inclusive) of the file extent item the given path
1338 * points to. If it points to an inline extent, the returned offset is rounded
1339 * up to the sector size.
1340 */
1341u64 btrfs_file_extent_end(const struct btrfs_path *path)
1342{
1343 const struct extent_buffer *leaf = path->nodes[0];
1344 const int slot = path->slots[0];
1345 struct btrfs_file_extent_item *fi;
1346 struct btrfs_key key;
1347 u64 end;
1348
1349 btrfs_item_key_to_cpu(leaf, &key, slot);
1350 ASSERT(key.type == BTRFS_EXTENT_DATA_KEY);
1351 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
1352
1353 if (btrfs_file_extent_type(leaf, fi) == BTRFS_FILE_EXTENT_INLINE)
1354 end = leaf->fs_info->sectorsize;
1355 else
1356 end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
1357
1358 return end;
1359}
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Copyright (C) 2007 Oracle. All rights reserved.
4 */
5
6#include <linux/bio.h>
7#include <linux/slab.h>
8#include <linux/pagemap.h>
9#include <linux/highmem.h>
10#include <linux/sched/mm.h>
11#include <crypto/hash.h>
12#include "misc.h"
13#include "ctree.h"
14#include "disk-io.h"
15#include "transaction.h"
16#include "volumes.h"
17#include "print-tree.h"
18#include "compression.h"
19
20#define __MAX_CSUM_ITEMS(r, size) ((unsigned long)(((BTRFS_LEAF_DATA_SIZE(r) - \
21 sizeof(struct btrfs_item) * 2) / \
22 size) - 1))
23
24#define MAX_CSUM_ITEMS(r, size) (min_t(u32, __MAX_CSUM_ITEMS(r, size), \
25 PAGE_SIZE))
26
27/**
28 * Set inode's size according to filesystem options
29 *
30 * @inode: inode we want to update the disk_i_size for
31 * @new_i_size: i_size we want to set to, 0 if we use i_size
32 *
33 * With NO_HOLES set this simply sets the disk_is_size to whatever i_size_read()
34 * returns as it is perfectly fine with a file that has holes without hole file
35 * extent items.
36 *
37 * However without NO_HOLES we need to only return the area that is contiguous
38 * from the 0 offset of the file. Otherwise we could end up adjust i_size up
39 * to an extent that has a gap in between.
40 *
41 * Finally new_i_size should only be set in the case of truncate where we're not
42 * ready to use i_size_read() as the limiter yet.
43 */
44void btrfs_inode_safe_disk_i_size_write(struct btrfs_inode *inode, u64 new_i_size)
45{
46 struct btrfs_fs_info *fs_info = inode->root->fs_info;
47 u64 start, end, i_size;
48 int ret;
49
50 i_size = new_i_size ?: i_size_read(&inode->vfs_inode);
51 if (btrfs_fs_incompat(fs_info, NO_HOLES)) {
52 inode->disk_i_size = i_size;
53 return;
54 }
55
56 spin_lock(&inode->lock);
57 ret = find_contiguous_extent_bit(&inode->file_extent_tree, 0, &start,
58 &end, EXTENT_DIRTY);
59 if (!ret && start == 0)
60 i_size = min(i_size, end + 1);
61 else
62 i_size = 0;
63 inode->disk_i_size = i_size;
64 spin_unlock(&inode->lock);
65}
66
67/**
68 * Mark range within a file as having a new extent inserted
69 *
70 * @inode: inode being modified
71 * @start: start file offset of the file extent we've inserted
72 * @len: logical length of the file extent item
73 *
74 * Call when we are inserting a new file extent where there was none before.
75 * Does not need to call this in the case where we're replacing an existing file
76 * extent, however if not sure it's fine to call this multiple times.
77 *
78 * The start and len must match the file extent item, so thus must be sectorsize
79 * aligned.
80 */
81int btrfs_inode_set_file_extent_range(struct btrfs_inode *inode, u64 start,
82 u64 len)
83{
84 if (len == 0)
85 return 0;
86
87 ASSERT(IS_ALIGNED(start + len, inode->root->fs_info->sectorsize));
88
89 if (btrfs_fs_incompat(inode->root->fs_info, NO_HOLES))
90 return 0;
91 return set_extent_bits(&inode->file_extent_tree, start, start + len - 1,
92 EXTENT_DIRTY);
93}
94
95/**
96 * Marks an inode range as not having a backing extent
97 *
98 * @inode: inode being modified
99 * @start: start file offset of the file extent we've inserted
100 * @len: logical length of the file extent item
101 *
102 * Called when we drop a file extent, for example when we truncate. Doesn't
103 * need to be called for cases where we're replacing a file extent, like when
104 * we've COWed a file extent.
105 *
106 * The start and len must match the file extent item, so thus must be sectorsize
107 * aligned.
108 */
109int btrfs_inode_clear_file_extent_range(struct btrfs_inode *inode, u64 start,
110 u64 len)
111{
112 if (len == 0)
113 return 0;
114
115 ASSERT(IS_ALIGNED(start + len, inode->root->fs_info->sectorsize) ||
116 len == (u64)-1);
117
118 if (btrfs_fs_incompat(inode->root->fs_info, NO_HOLES))
119 return 0;
120 return clear_extent_bit(&inode->file_extent_tree, start,
121 start + len - 1, EXTENT_DIRTY, 0, 0, NULL);
122}
123
124static inline u32 max_ordered_sum_bytes(struct btrfs_fs_info *fs_info,
125 u16 csum_size)
126{
127 u32 ncsums = (PAGE_SIZE - sizeof(struct btrfs_ordered_sum)) / csum_size;
128
129 return ncsums * fs_info->sectorsize;
130}
131
132int btrfs_insert_file_extent(struct btrfs_trans_handle *trans,
133 struct btrfs_root *root,
134 u64 objectid, u64 pos,
135 u64 disk_offset, u64 disk_num_bytes,
136 u64 num_bytes, u64 offset, u64 ram_bytes,
137 u8 compression, u8 encryption, u16 other_encoding)
138{
139 int ret = 0;
140 struct btrfs_file_extent_item *item;
141 struct btrfs_key file_key;
142 struct btrfs_path *path;
143 struct extent_buffer *leaf;
144
145 path = btrfs_alloc_path();
146 if (!path)
147 return -ENOMEM;
148 file_key.objectid = objectid;
149 file_key.offset = pos;
150 file_key.type = BTRFS_EXTENT_DATA_KEY;
151
152 ret = btrfs_insert_empty_item(trans, root, path, &file_key,
153 sizeof(*item));
154 if (ret < 0)
155 goto out;
156 BUG_ON(ret); /* Can't happen */
157 leaf = path->nodes[0];
158 item = btrfs_item_ptr(leaf, path->slots[0],
159 struct btrfs_file_extent_item);
160 btrfs_set_file_extent_disk_bytenr(leaf, item, disk_offset);
161 btrfs_set_file_extent_disk_num_bytes(leaf, item, disk_num_bytes);
162 btrfs_set_file_extent_offset(leaf, item, offset);
163 btrfs_set_file_extent_num_bytes(leaf, item, num_bytes);
164 btrfs_set_file_extent_ram_bytes(leaf, item, ram_bytes);
165 btrfs_set_file_extent_generation(leaf, item, trans->transid);
166 btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
167 btrfs_set_file_extent_compression(leaf, item, compression);
168 btrfs_set_file_extent_encryption(leaf, item, encryption);
169 btrfs_set_file_extent_other_encoding(leaf, item, other_encoding);
170
171 btrfs_mark_buffer_dirty(leaf);
172out:
173 btrfs_free_path(path);
174 return ret;
175}
176
177static struct btrfs_csum_item *
178btrfs_lookup_csum(struct btrfs_trans_handle *trans,
179 struct btrfs_root *root,
180 struct btrfs_path *path,
181 u64 bytenr, int cow)
182{
183 struct btrfs_fs_info *fs_info = root->fs_info;
184 int ret;
185 struct btrfs_key file_key;
186 struct btrfs_key found_key;
187 struct btrfs_csum_item *item;
188 struct extent_buffer *leaf;
189 u64 csum_offset = 0;
190 const u32 csum_size = fs_info->csum_size;
191 int csums_in_item;
192
193 file_key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
194 file_key.offset = bytenr;
195 file_key.type = BTRFS_EXTENT_CSUM_KEY;
196 ret = btrfs_search_slot(trans, root, &file_key, path, 0, cow);
197 if (ret < 0)
198 goto fail;
199 leaf = path->nodes[0];
200 if (ret > 0) {
201 ret = 1;
202 if (path->slots[0] == 0)
203 goto fail;
204 path->slots[0]--;
205 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
206 if (found_key.type != BTRFS_EXTENT_CSUM_KEY)
207 goto fail;
208
209 csum_offset = (bytenr - found_key.offset) >>
210 fs_info->sectorsize_bits;
211 csums_in_item = btrfs_item_size_nr(leaf, path->slots[0]);
212 csums_in_item /= csum_size;
213
214 if (csum_offset == csums_in_item) {
215 ret = -EFBIG;
216 goto fail;
217 } else if (csum_offset > csums_in_item) {
218 goto fail;
219 }
220 }
221 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item);
222 item = (struct btrfs_csum_item *)((unsigned char *)item +
223 csum_offset * csum_size);
224 return item;
225fail:
226 if (ret > 0)
227 ret = -ENOENT;
228 return ERR_PTR(ret);
229}
230
231int btrfs_lookup_file_extent(struct btrfs_trans_handle *trans,
232 struct btrfs_root *root,
233 struct btrfs_path *path, u64 objectid,
234 u64 offset, int mod)
235{
236 int ret;
237 struct btrfs_key file_key;
238 int ins_len = mod < 0 ? -1 : 0;
239 int cow = mod != 0;
240
241 file_key.objectid = objectid;
242 file_key.offset = offset;
243 file_key.type = BTRFS_EXTENT_DATA_KEY;
244 ret = btrfs_search_slot(trans, root, &file_key, path, ins_len, cow);
245 return ret;
246}
247
248/*
249 * Find checksums for logical bytenr range [disk_bytenr, disk_bytenr + len) and
250 * estore the result to @dst.
251 *
252 * Return >0 for the number of sectors we found.
253 * Return 0 for the range [disk_bytenr, disk_bytenr + sectorsize) has no csum
254 * for it. Caller may want to try next sector until one range is hit.
255 * Return <0 for fatal error.
256 */
257static int search_csum_tree(struct btrfs_fs_info *fs_info,
258 struct btrfs_path *path, u64 disk_bytenr,
259 u64 len, u8 *dst)
260{
261 struct btrfs_csum_item *item = NULL;
262 struct btrfs_key key;
263 const u32 sectorsize = fs_info->sectorsize;
264 const u32 csum_size = fs_info->csum_size;
265 u32 itemsize;
266 int ret;
267 u64 csum_start;
268 u64 csum_len;
269
270 ASSERT(IS_ALIGNED(disk_bytenr, sectorsize) &&
271 IS_ALIGNED(len, sectorsize));
272
273 /* Check if the current csum item covers disk_bytenr */
274 if (path->nodes[0]) {
275 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
276 struct btrfs_csum_item);
277 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
278 itemsize = btrfs_item_size_nr(path->nodes[0], path->slots[0]);
279
280 csum_start = key.offset;
281 csum_len = (itemsize / csum_size) * sectorsize;
282
283 if (in_range(disk_bytenr, csum_start, csum_len))
284 goto found;
285 }
286
287 /* Current item doesn't contain the desired range, search again */
288 btrfs_release_path(path);
289 item = btrfs_lookup_csum(NULL, fs_info->csum_root, path, disk_bytenr, 0);
290 if (IS_ERR(item)) {
291 ret = PTR_ERR(item);
292 goto out;
293 }
294 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
295 itemsize = btrfs_item_size_nr(path->nodes[0], path->slots[0]);
296
297 csum_start = key.offset;
298 csum_len = (itemsize / csum_size) * sectorsize;
299 ASSERT(in_range(disk_bytenr, csum_start, csum_len));
300
301found:
302 ret = (min(csum_start + csum_len, disk_bytenr + len) -
303 disk_bytenr) >> fs_info->sectorsize_bits;
304 read_extent_buffer(path->nodes[0], dst, (unsigned long)item,
305 ret * csum_size);
306out:
307 if (ret == -ENOENT)
308 ret = 0;
309 return ret;
310}
311
312/*
313 * Locate the file_offset of @cur_disk_bytenr of a @bio.
314 *
315 * Bio of btrfs represents read range of
316 * [bi_sector << 9, bi_sector << 9 + bi_size).
317 * Knowing this, we can iterate through each bvec to locate the page belong to
318 * @cur_disk_bytenr and get the file offset.
319 *
320 * @inode is used to determine if the bvec page really belongs to @inode.
321 *
322 * Return 0 if we can't find the file offset
323 * Return >0 if we find the file offset and restore it to @file_offset_ret
324 */
325static int search_file_offset_in_bio(struct bio *bio, struct inode *inode,
326 u64 disk_bytenr, u64 *file_offset_ret)
327{
328 struct bvec_iter iter;
329 struct bio_vec bvec;
330 u64 cur = bio->bi_iter.bi_sector << SECTOR_SHIFT;
331 int ret = 0;
332
333 bio_for_each_segment(bvec, bio, iter) {
334 struct page *page = bvec.bv_page;
335
336 if (cur > disk_bytenr)
337 break;
338 if (cur + bvec.bv_len <= disk_bytenr) {
339 cur += bvec.bv_len;
340 continue;
341 }
342 ASSERT(in_range(disk_bytenr, cur, bvec.bv_len));
343 if (page->mapping && page->mapping->host &&
344 page->mapping->host == inode) {
345 ret = 1;
346 *file_offset_ret = page_offset(page) + bvec.bv_offset +
347 disk_bytenr - cur;
348 break;
349 }
350 }
351 return ret;
352}
353
354/**
355 * Lookup the checksum for the read bio in csum tree.
356 *
357 * @inode: inode that the bio is for.
358 * @bio: bio to look up.
359 * @dst: Buffer of size nblocks * btrfs_super_csum_size() used to return
360 * checksum (nblocks = bio->bi_iter.bi_size / fs_info->sectorsize). If
361 * NULL, the checksum buffer is allocated and returned in
362 * btrfs_io_bio(bio)->csum instead.
363 *
364 * Return: BLK_STS_RESOURCE if allocating memory fails, BLK_STS_OK otherwise.
365 */
366blk_status_t btrfs_lookup_bio_sums(struct inode *inode, struct bio *bio, u8 *dst)
367{
368 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
369 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
370 struct btrfs_path *path;
371 const u32 sectorsize = fs_info->sectorsize;
372 const u32 csum_size = fs_info->csum_size;
373 u32 orig_len = bio->bi_iter.bi_size;
374 u64 orig_disk_bytenr = bio->bi_iter.bi_sector << SECTOR_SHIFT;
375 u64 cur_disk_bytenr;
376 u8 *csum;
377 const unsigned int nblocks = orig_len >> fs_info->sectorsize_bits;
378 int count = 0;
379
380 if (!fs_info->csum_root || (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM))
381 return BLK_STS_OK;
382
383 /*
384 * This function is only called for read bio.
385 *
386 * This means two things:
387 * - All our csums should only be in csum tree
388 * No ordered extents csums, as ordered extents are only for write
389 * path.
390 * - No need to bother any other info from bvec
391 * Since we're looking up csums, the only important info is the
392 * disk_bytenr and the length, which can be extracted from bi_iter
393 * directly.
394 */
395 ASSERT(bio_op(bio) == REQ_OP_READ);
396 path = btrfs_alloc_path();
397 if (!path)
398 return BLK_STS_RESOURCE;
399
400 if (!dst) {
401 struct btrfs_io_bio *btrfs_bio = btrfs_io_bio(bio);
402
403 if (nblocks * csum_size > BTRFS_BIO_INLINE_CSUM_SIZE) {
404 btrfs_bio->csum = kmalloc_array(nblocks, csum_size,
405 GFP_NOFS);
406 if (!btrfs_bio->csum) {
407 btrfs_free_path(path);
408 return BLK_STS_RESOURCE;
409 }
410 } else {
411 btrfs_bio->csum = btrfs_bio->csum_inline;
412 }
413 csum = btrfs_bio->csum;
414 } else {
415 csum = dst;
416 }
417
418 /*
419 * If requested number of sectors is larger than one leaf can contain,
420 * kick the readahead for csum tree.
421 */
422 if (nblocks > fs_info->csums_per_leaf)
423 path->reada = READA_FORWARD;
424
425 /*
426 * the free space stuff is only read when it hasn't been
427 * updated in the current transaction. So, we can safely
428 * read from the commit root and sidestep a nasty deadlock
429 * between reading the free space cache and updating the csum tree.
430 */
431 if (btrfs_is_free_space_inode(BTRFS_I(inode))) {
432 path->search_commit_root = 1;
433 path->skip_locking = 1;
434 }
435
436 for (cur_disk_bytenr = orig_disk_bytenr;
437 cur_disk_bytenr < orig_disk_bytenr + orig_len;
438 cur_disk_bytenr += (count * sectorsize)) {
439 u64 search_len = orig_disk_bytenr + orig_len - cur_disk_bytenr;
440 unsigned int sector_offset;
441 u8 *csum_dst;
442
443 /*
444 * Although both cur_disk_bytenr and orig_disk_bytenr is u64,
445 * we're calculating the offset to the bio start.
446 *
447 * Bio size is limited to UINT_MAX, thus unsigned int is large
448 * enough to contain the raw result, not to mention the right
449 * shifted result.
450 */
451 ASSERT(cur_disk_bytenr - orig_disk_bytenr < UINT_MAX);
452 sector_offset = (cur_disk_bytenr - orig_disk_bytenr) >>
453 fs_info->sectorsize_bits;
454 csum_dst = csum + sector_offset * csum_size;
455
456 count = search_csum_tree(fs_info, path, cur_disk_bytenr,
457 search_len, csum_dst);
458 if (count <= 0) {
459 /*
460 * Either we hit a critical error or we didn't find
461 * the csum.
462 * Either way, we put zero into the csums dst, and skip
463 * to the next sector.
464 */
465 memset(csum_dst, 0, csum_size);
466 count = 1;
467
468 /*
469 * For data reloc inode, we need to mark the range
470 * NODATASUM so that balance won't report false csum
471 * error.
472 */
473 if (BTRFS_I(inode)->root->root_key.objectid ==
474 BTRFS_DATA_RELOC_TREE_OBJECTID) {
475 u64 file_offset;
476 int ret;
477
478 ret = search_file_offset_in_bio(bio, inode,
479 cur_disk_bytenr, &file_offset);
480 if (ret)
481 set_extent_bits(io_tree, file_offset,
482 file_offset + sectorsize - 1,
483 EXTENT_NODATASUM);
484 } else {
485 btrfs_warn_rl(fs_info,
486 "csum hole found for disk bytenr range [%llu, %llu)",
487 cur_disk_bytenr, cur_disk_bytenr + sectorsize);
488 }
489 }
490 }
491
492 btrfs_free_path(path);
493 return BLK_STS_OK;
494}
495
496int btrfs_lookup_csums_range(struct btrfs_root *root, u64 start, u64 end,
497 struct list_head *list, int search_commit)
498{
499 struct btrfs_fs_info *fs_info = root->fs_info;
500 struct btrfs_key key;
501 struct btrfs_path *path;
502 struct extent_buffer *leaf;
503 struct btrfs_ordered_sum *sums;
504 struct btrfs_csum_item *item;
505 LIST_HEAD(tmplist);
506 unsigned long offset;
507 int ret;
508 size_t size;
509 u64 csum_end;
510 const u32 csum_size = fs_info->csum_size;
511
512 ASSERT(IS_ALIGNED(start, fs_info->sectorsize) &&
513 IS_ALIGNED(end + 1, fs_info->sectorsize));
514
515 path = btrfs_alloc_path();
516 if (!path)
517 return -ENOMEM;
518
519 if (search_commit) {
520 path->skip_locking = 1;
521 path->reada = READA_FORWARD;
522 path->search_commit_root = 1;
523 }
524
525 key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
526 key.offset = start;
527 key.type = BTRFS_EXTENT_CSUM_KEY;
528
529 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
530 if (ret < 0)
531 goto fail;
532 if (ret > 0 && path->slots[0] > 0) {
533 leaf = path->nodes[0];
534 btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
535 if (key.objectid == BTRFS_EXTENT_CSUM_OBJECTID &&
536 key.type == BTRFS_EXTENT_CSUM_KEY) {
537 offset = (start - key.offset) >> fs_info->sectorsize_bits;
538 if (offset * csum_size <
539 btrfs_item_size_nr(leaf, path->slots[0] - 1))
540 path->slots[0]--;
541 }
542 }
543
544 while (start <= end) {
545 leaf = path->nodes[0];
546 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
547 ret = btrfs_next_leaf(root, path);
548 if (ret < 0)
549 goto fail;
550 if (ret > 0)
551 break;
552 leaf = path->nodes[0];
553 }
554
555 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
556 if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
557 key.type != BTRFS_EXTENT_CSUM_KEY ||
558 key.offset > end)
559 break;
560
561 if (key.offset > start)
562 start = key.offset;
563
564 size = btrfs_item_size_nr(leaf, path->slots[0]);
565 csum_end = key.offset + (size / csum_size) * fs_info->sectorsize;
566 if (csum_end <= start) {
567 path->slots[0]++;
568 continue;
569 }
570
571 csum_end = min(csum_end, end + 1);
572 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
573 struct btrfs_csum_item);
574 while (start < csum_end) {
575 size = min_t(size_t, csum_end - start,
576 max_ordered_sum_bytes(fs_info, csum_size));
577 sums = kzalloc(btrfs_ordered_sum_size(fs_info, size),
578 GFP_NOFS);
579 if (!sums) {
580 ret = -ENOMEM;
581 goto fail;
582 }
583
584 sums->bytenr = start;
585 sums->len = (int)size;
586
587 offset = (start - key.offset) >> fs_info->sectorsize_bits;
588 offset *= csum_size;
589 size >>= fs_info->sectorsize_bits;
590
591 read_extent_buffer(path->nodes[0],
592 sums->sums,
593 ((unsigned long)item) + offset,
594 csum_size * size);
595
596 start += fs_info->sectorsize * size;
597 list_add_tail(&sums->list, &tmplist);
598 }
599 path->slots[0]++;
600 }
601 ret = 0;
602fail:
603 while (ret < 0 && !list_empty(&tmplist)) {
604 sums = list_entry(tmplist.next, struct btrfs_ordered_sum, list);
605 list_del(&sums->list);
606 kfree(sums);
607 }
608 list_splice_tail(&tmplist, list);
609
610 btrfs_free_path(path);
611 return ret;
612}
613
614/*
615 * btrfs_csum_one_bio - Calculates checksums of the data contained inside a bio
616 * @inode: Owner of the data inside the bio
617 * @bio: Contains the data to be checksummed
618 * @file_start: offset in file this bio begins to describe
619 * @contig: Boolean. If true/1 means all bio vecs in this bio are
620 * contiguous and they begin at @file_start in the file. False/0
621 * means this bio can contain potentially discontiguous bio vecs
622 * so the logical offset of each should be calculated separately.
623 */
624blk_status_t btrfs_csum_one_bio(struct btrfs_inode *inode, struct bio *bio,
625 u64 file_start, int contig)
626{
627 struct btrfs_fs_info *fs_info = inode->root->fs_info;
628 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
629 struct btrfs_ordered_sum *sums;
630 struct btrfs_ordered_extent *ordered = NULL;
631 char *data;
632 struct bvec_iter iter;
633 struct bio_vec bvec;
634 int index;
635 int nr_sectors;
636 unsigned long total_bytes = 0;
637 unsigned long this_sum_bytes = 0;
638 int i;
639 u64 offset;
640 unsigned nofs_flag;
641
642 nofs_flag = memalloc_nofs_save();
643 sums = kvzalloc(btrfs_ordered_sum_size(fs_info, bio->bi_iter.bi_size),
644 GFP_KERNEL);
645 memalloc_nofs_restore(nofs_flag);
646
647 if (!sums)
648 return BLK_STS_RESOURCE;
649
650 sums->len = bio->bi_iter.bi_size;
651 INIT_LIST_HEAD(&sums->list);
652
653 if (contig)
654 offset = file_start;
655 else
656 offset = 0; /* shut up gcc */
657
658 sums->bytenr = bio->bi_iter.bi_sector << 9;
659 index = 0;
660
661 shash->tfm = fs_info->csum_shash;
662
663 bio_for_each_segment(bvec, bio, iter) {
664 if (!contig)
665 offset = page_offset(bvec.bv_page) + bvec.bv_offset;
666
667 if (!ordered) {
668 ordered = btrfs_lookup_ordered_extent(inode, offset);
669 /*
670 * The bio range is not covered by any ordered extent,
671 * must be a code logic error.
672 */
673 if (unlikely(!ordered)) {
674 WARN(1, KERN_WARNING
675 "no ordered extent for root %llu ino %llu offset %llu\n",
676 inode->root->root_key.objectid,
677 btrfs_ino(inode), offset);
678 kvfree(sums);
679 return BLK_STS_IOERR;
680 }
681 }
682
683 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info,
684 bvec.bv_len + fs_info->sectorsize
685 - 1);
686
687 for (i = 0; i < nr_sectors; i++) {
688 if (offset >= ordered->file_offset + ordered->num_bytes ||
689 offset < ordered->file_offset) {
690 unsigned long bytes_left;
691
692 sums->len = this_sum_bytes;
693 this_sum_bytes = 0;
694 btrfs_add_ordered_sum(ordered, sums);
695 btrfs_put_ordered_extent(ordered);
696
697 bytes_left = bio->bi_iter.bi_size - total_bytes;
698
699 nofs_flag = memalloc_nofs_save();
700 sums = kvzalloc(btrfs_ordered_sum_size(fs_info,
701 bytes_left), GFP_KERNEL);
702 memalloc_nofs_restore(nofs_flag);
703 BUG_ON(!sums); /* -ENOMEM */
704 sums->len = bytes_left;
705 ordered = btrfs_lookup_ordered_extent(inode,
706 offset);
707 ASSERT(ordered); /* Logic error */
708 sums->bytenr = (bio->bi_iter.bi_sector << 9)
709 + total_bytes;
710 index = 0;
711 }
712
713 data = kmap_atomic(bvec.bv_page);
714 crypto_shash_digest(shash, data + bvec.bv_offset
715 + (i * fs_info->sectorsize),
716 fs_info->sectorsize,
717 sums->sums + index);
718 kunmap_atomic(data);
719 index += fs_info->csum_size;
720 offset += fs_info->sectorsize;
721 this_sum_bytes += fs_info->sectorsize;
722 total_bytes += fs_info->sectorsize;
723 }
724
725 }
726 this_sum_bytes = 0;
727 btrfs_add_ordered_sum(ordered, sums);
728 btrfs_put_ordered_extent(ordered);
729 return 0;
730}
731
732/*
733 * helper function for csum removal, this expects the
734 * key to describe the csum pointed to by the path, and it expects
735 * the csum to overlap the range [bytenr, len]
736 *
737 * The csum should not be entirely contained in the range and the
738 * range should not be entirely contained in the csum.
739 *
740 * This calls btrfs_truncate_item with the correct args based on the
741 * overlap, and fixes up the key as required.
742 */
743static noinline void truncate_one_csum(struct btrfs_fs_info *fs_info,
744 struct btrfs_path *path,
745 struct btrfs_key *key,
746 u64 bytenr, u64 len)
747{
748 struct extent_buffer *leaf;
749 const u32 csum_size = fs_info->csum_size;
750 u64 csum_end;
751 u64 end_byte = bytenr + len;
752 u32 blocksize_bits = fs_info->sectorsize_bits;
753
754 leaf = path->nodes[0];
755 csum_end = btrfs_item_size_nr(leaf, path->slots[0]) / csum_size;
756 csum_end <<= blocksize_bits;
757 csum_end += key->offset;
758
759 if (key->offset < bytenr && csum_end <= end_byte) {
760 /*
761 * [ bytenr - len ]
762 * [ ]
763 * [csum ]
764 * A simple truncate off the end of the item
765 */
766 u32 new_size = (bytenr - key->offset) >> blocksize_bits;
767 new_size *= csum_size;
768 btrfs_truncate_item(path, new_size, 1);
769 } else if (key->offset >= bytenr && csum_end > end_byte &&
770 end_byte > key->offset) {
771 /*
772 * [ bytenr - len ]
773 * [ ]
774 * [csum ]
775 * we need to truncate from the beginning of the csum
776 */
777 u32 new_size = (csum_end - end_byte) >> blocksize_bits;
778 new_size *= csum_size;
779
780 btrfs_truncate_item(path, new_size, 0);
781
782 key->offset = end_byte;
783 btrfs_set_item_key_safe(fs_info, path, key);
784 } else {
785 BUG();
786 }
787}
788
789/*
790 * deletes the csum items from the csum tree for a given
791 * range of bytes.
792 */
793int btrfs_del_csums(struct btrfs_trans_handle *trans,
794 struct btrfs_root *root, u64 bytenr, u64 len)
795{
796 struct btrfs_fs_info *fs_info = trans->fs_info;
797 struct btrfs_path *path;
798 struct btrfs_key key;
799 u64 end_byte = bytenr + len;
800 u64 csum_end;
801 struct extent_buffer *leaf;
802 int ret = 0;
803 const u32 csum_size = fs_info->csum_size;
804 u32 blocksize_bits = fs_info->sectorsize_bits;
805
806 ASSERT(root == fs_info->csum_root ||
807 root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID);
808
809 path = btrfs_alloc_path();
810 if (!path)
811 return -ENOMEM;
812
813 while (1) {
814 key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
815 key.offset = end_byte - 1;
816 key.type = BTRFS_EXTENT_CSUM_KEY;
817
818 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
819 if (ret > 0) {
820 ret = 0;
821 if (path->slots[0] == 0)
822 break;
823 path->slots[0]--;
824 } else if (ret < 0) {
825 break;
826 }
827
828 leaf = path->nodes[0];
829 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
830
831 if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
832 key.type != BTRFS_EXTENT_CSUM_KEY) {
833 break;
834 }
835
836 if (key.offset >= end_byte)
837 break;
838
839 csum_end = btrfs_item_size_nr(leaf, path->slots[0]) / csum_size;
840 csum_end <<= blocksize_bits;
841 csum_end += key.offset;
842
843 /* this csum ends before we start, we're done */
844 if (csum_end <= bytenr)
845 break;
846
847 /* delete the entire item, it is inside our range */
848 if (key.offset >= bytenr && csum_end <= end_byte) {
849 int del_nr = 1;
850
851 /*
852 * Check how many csum items preceding this one in this
853 * leaf correspond to our range and then delete them all
854 * at once.
855 */
856 if (key.offset > bytenr && path->slots[0] > 0) {
857 int slot = path->slots[0] - 1;
858
859 while (slot >= 0) {
860 struct btrfs_key pk;
861
862 btrfs_item_key_to_cpu(leaf, &pk, slot);
863 if (pk.offset < bytenr ||
864 pk.type != BTRFS_EXTENT_CSUM_KEY ||
865 pk.objectid !=
866 BTRFS_EXTENT_CSUM_OBJECTID)
867 break;
868 path->slots[0] = slot;
869 del_nr++;
870 key.offset = pk.offset;
871 slot--;
872 }
873 }
874 ret = btrfs_del_items(trans, root, path,
875 path->slots[0], del_nr);
876 if (ret)
877 break;
878 if (key.offset == bytenr)
879 break;
880 } else if (key.offset < bytenr && csum_end > end_byte) {
881 unsigned long offset;
882 unsigned long shift_len;
883 unsigned long item_offset;
884 /*
885 * [ bytenr - len ]
886 * [csum ]
887 *
888 * Our bytes are in the middle of the csum,
889 * we need to split this item and insert a new one.
890 *
891 * But we can't drop the path because the
892 * csum could change, get removed, extended etc.
893 *
894 * The trick here is the max size of a csum item leaves
895 * enough room in the tree block for a single
896 * item header. So, we split the item in place,
897 * adding a new header pointing to the existing
898 * bytes. Then we loop around again and we have
899 * a nicely formed csum item that we can neatly
900 * truncate.
901 */
902 offset = (bytenr - key.offset) >> blocksize_bits;
903 offset *= csum_size;
904
905 shift_len = (len >> blocksize_bits) * csum_size;
906
907 item_offset = btrfs_item_ptr_offset(leaf,
908 path->slots[0]);
909
910 memzero_extent_buffer(leaf, item_offset + offset,
911 shift_len);
912 key.offset = bytenr;
913
914 /*
915 * btrfs_split_item returns -EAGAIN when the
916 * item changed size or key
917 */
918 ret = btrfs_split_item(trans, root, path, &key, offset);
919 if (ret && ret != -EAGAIN) {
920 btrfs_abort_transaction(trans, ret);
921 break;
922 }
923 ret = 0;
924
925 key.offset = end_byte - 1;
926 } else {
927 truncate_one_csum(fs_info, path, &key, bytenr, len);
928 if (key.offset < bytenr)
929 break;
930 }
931 btrfs_release_path(path);
932 }
933 btrfs_free_path(path);
934 return ret;
935}
936
937static int find_next_csum_offset(struct btrfs_root *root,
938 struct btrfs_path *path,
939 u64 *next_offset)
940{
941 const u32 nritems = btrfs_header_nritems(path->nodes[0]);
942 struct btrfs_key found_key;
943 int slot = path->slots[0] + 1;
944 int ret;
945
946 if (nritems == 0 || slot >= nritems) {
947 ret = btrfs_next_leaf(root, path);
948 if (ret < 0) {
949 return ret;
950 } else if (ret > 0) {
951 *next_offset = (u64)-1;
952 return 0;
953 }
954 slot = path->slots[0];
955 }
956
957 btrfs_item_key_to_cpu(path->nodes[0], &found_key, slot);
958
959 if (found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
960 found_key.type != BTRFS_EXTENT_CSUM_KEY)
961 *next_offset = (u64)-1;
962 else
963 *next_offset = found_key.offset;
964
965 return 0;
966}
967
968int btrfs_csum_file_blocks(struct btrfs_trans_handle *trans,
969 struct btrfs_root *root,
970 struct btrfs_ordered_sum *sums)
971{
972 struct btrfs_fs_info *fs_info = root->fs_info;
973 struct btrfs_key file_key;
974 struct btrfs_key found_key;
975 struct btrfs_path *path;
976 struct btrfs_csum_item *item;
977 struct btrfs_csum_item *item_end;
978 struct extent_buffer *leaf = NULL;
979 u64 next_offset;
980 u64 total_bytes = 0;
981 u64 csum_offset;
982 u64 bytenr;
983 u32 ins_size;
984 int index = 0;
985 int found_next;
986 int ret;
987 const u32 csum_size = fs_info->csum_size;
988
989 path = btrfs_alloc_path();
990 if (!path)
991 return -ENOMEM;
992again:
993 next_offset = (u64)-1;
994 found_next = 0;
995 bytenr = sums->bytenr + total_bytes;
996 file_key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
997 file_key.offset = bytenr;
998 file_key.type = BTRFS_EXTENT_CSUM_KEY;
999
1000 item = btrfs_lookup_csum(trans, root, path, bytenr, 1);
1001 if (!IS_ERR(item)) {
1002 ret = 0;
1003 leaf = path->nodes[0];
1004 item_end = btrfs_item_ptr(leaf, path->slots[0],
1005 struct btrfs_csum_item);
1006 item_end = (struct btrfs_csum_item *)((char *)item_end +
1007 btrfs_item_size_nr(leaf, path->slots[0]));
1008 goto found;
1009 }
1010 ret = PTR_ERR(item);
1011 if (ret != -EFBIG && ret != -ENOENT)
1012 goto out;
1013
1014 if (ret == -EFBIG) {
1015 u32 item_size;
1016 /* we found one, but it isn't big enough yet */
1017 leaf = path->nodes[0];
1018 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1019 if ((item_size / csum_size) >=
1020 MAX_CSUM_ITEMS(fs_info, csum_size)) {
1021 /* already at max size, make a new one */
1022 goto insert;
1023 }
1024 } else {
1025 /* We didn't find a csum item, insert one. */
1026 ret = find_next_csum_offset(root, path, &next_offset);
1027 if (ret < 0)
1028 goto out;
1029 found_next = 1;
1030 goto insert;
1031 }
1032
1033 /*
1034 * At this point, we know the tree has a checksum item that ends at an
1035 * offset matching the start of the checksum range we want to insert.
1036 * We try to extend that item as much as possible and then add as many
1037 * checksums to it as they fit.
1038 *
1039 * First check if the leaf has enough free space for at least one
1040 * checksum. If it has go directly to the item extension code, otherwise
1041 * release the path and do a search for insertion before the extension.
1042 */
1043 if (btrfs_leaf_free_space(leaf) >= csum_size) {
1044 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1045 csum_offset = (bytenr - found_key.offset) >>
1046 fs_info->sectorsize_bits;
1047 goto extend_csum;
1048 }
1049
1050 btrfs_release_path(path);
1051 path->search_for_extension = 1;
1052 ret = btrfs_search_slot(trans, root, &file_key, path,
1053 csum_size, 1);
1054 path->search_for_extension = 0;
1055 if (ret < 0)
1056 goto out;
1057
1058 if (ret > 0) {
1059 if (path->slots[0] == 0)
1060 goto insert;
1061 path->slots[0]--;
1062 }
1063
1064 leaf = path->nodes[0];
1065 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1066 csum_offset = (bytenr - found_key.offset) >> fs_info->sectorsize_bits;
1067
1068 if (found_key.type != BTRFS_EXTENT_CSUM_KEY ||
1069 found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
1070 csum_offset >= MAX_CSUM_ITEMS(fs_info, csum_size)) {
1071 goto insert;
1072 }
1073
1074extend_csum:
1075 if (csum_offset == btrfs_item_size_nr(leaf, path->slots[0]) /
1076 csum_size) {
1077 int extend_nr;
1078 u64 tmp;
1079 u32 diff;
1080
1081 tmp = sums->len - total_bytes;
1082 tmp >>= fs_info->sectorsize_bits;
1083 WARN_ON(tmp < 1);
1084 extend_nr = max_t(int, 1, tmp);
1085
1086 /*
1087 * A log tree can already have checksum items with a subset of
1088 * the checksums we are trying to log. This can happen after
1089 * doing a sequence of partial writes into prealloc extents and
1090 * fsyncs in between, with a full fsync logging a larger subrange
1091 * of an extent for which a previous fast fsync logged a smaller
1092 * subrange. And this happens in particular due to merging file
1093 * extent items when we complete an ordered extent for a range
1094 * covered by a prealloc extent - this is done at
1095 * btrfs_mark_extent_written().
1096 *
1097 * So if we try to extend the previous checksum item, which has
1098 * a range that ends at the start of the range we want to insert,
1099 * make sure we don't extend beyond the start offset of the next
1100 * checksum item. If we are at the last item in the leaf, then
1101 * forget the optimization of extending and add a new checksum
1102 * item - it is not worth the complexity of releasing the path,
1103 * getting the first key for the next leaf, repeat the btree
1104 * search, etc, because log trees are temporary anyway and it
1105 * would only save a few bytes of leaf space.
1106 */
1107 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
1108 if (path->slots[0] + 1 >=
1109 btrfs_header_nritems(path->nodes[0])) {
1110 ret = find_next_csum_offset(root, path, &next_offset);
1111 if (ret < 0)
1112 goto out;
1113 found_next = 1;
1114 goto insert;
1115 }
1116
1117 ret = find_next_csum_offset(root, path, &next_offset);
1118 if (ret < 0)
1119 goto out;
1120
1121 tmp = (next_offset - bytenr) >> fs_info->sectorsize_bits;
1122 if (tmp <= INT_MAX)
1123 extend_nr = min_t(int, extend_nr, tmp);
1124 }
1125
1126 diff = (csum_offset + extend_nr) * csum_size;
1127 diff = min(diff,
1128 MAX_CSUM_ITEMS(fs_info, csum_size) * csum_size);
1129
1130 diff = diff - btrfs_item_size_nr(leaf, path->slots[0]);
1131 diff = min_t(u32, btrfs_leaf_free_space(leaf), diff);
1132 diff /= csum_size;
1133 diff *= csum_size;
1134
1135 btrfs_extend_item(path, diff);
1136 ret = 0;
1137 goto csum;
1138 }
1139
1140insert:
1141 btrfs_release_path(path);
1142 csum_offset = 0;
1143 if (found_next) {
1144 u64 tmp;
1145
1146 tmp = sums->len - total_bytes;
1147 tmp >>= fs_info->sectorsize_bits;
1148 tmp = min(tmp, (next_offset - file_key.offset) >>
1149 fs_info->sectorsize_bits);
1150
1151 tmp = max_t(u64, 1, tmp);
1152 tmp = min_t(u64, tmp, MAX_CSUM_ITEMS(fs_info, csum_size));
1153 ins_size = csum_size * tmp;
1154 } else {
1155 ins_size = csum_size;
1156 }
1157 ret = btrfs_insert_empty_item(trans, root, path, &file_key,
1158 ins_size);
1159 if (ret < 0)
1160 goto out;
1161 if (WARN_ON(ret != 0))
1162 goto out;
1163 leaf = path->nodes[0];
1164csum:
1165 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item);
1166 item_end = (struct btrfs_csum_item *)((unsigned char *)item +
1167 btrfs_item_size_nr(leaf, path->slots[0]));
1168 item = (struct btrfs_csum_item *)((unsigned char *)item +
1169 csum_offset * csum_size);
1170found:
1171 ins_size = (u32)(sums->len - total_bytes) >> fs_info->sectorsize_bits;
1172 ins_size *= csum_size;
1173 ins_size = min_t(u32, (unsigned long)item_end - (unsigned long)item,
1174 ins_size);
1175 write_extent_buffer(leaf, sums->sums + index, (unsigned long)item,
1176 ins_size);
1177
1178 index += ins_size;
1179 ins_size /= csum_size;
1180 total_bytes += ins_size * fs_info->sectorsize;
1181
1182 btrfs_mark_buffer_dirty(path->nodes[0]);
1183 if (total_bytes < sums->len) {
1184 btrfs_release_path(path);
1185 cond_resched();
1186 goto again;
1187 }
1188out:
1189 btrfs_free_path(path);
1190 return ret;
1191}
1192
1193void btrfs_extent_item_to_extent_map(struct btrfs_inode *inode,
1194 const struct btrfs_path *path,
1195 struct btrfs_file_extent_item *fi,
1196 const bool new_inline,
1197 struct extent_map *em)
1198{
1199 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1200 struct btrfs_root *root = inode->root;
1201 struct extent_buffer *leaf = path->nodes[0];
1202 const int slot = path->slots[0];
1203 struct btrfs_key key;
1204 u64 extent_start, extent_end;
1205 u64 bytenr;
1206 u8 type = btrfs_file_extent_type(leaf, fi);
1207 int compress_type = btrfs_file_extent_compression(leaf, fi);
1208
1209 btrfs_item_key_to_cpu(leaf, &key, slot);
1210 extent_start = key.offset;
1211 extent_end = btrfs_file_extent_end(path);
1212 em->ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1213 if (type == BTRFS_FILE_EXTENT_REG ||
1214 type == BTRFS_FILE_EXTENT_PREALLOC) {
1215 em->start = extent_start;
1216 em->len = extent_end - extent_start;
1217 em->orig_start = extent_start -
1218 btrfs_file_extent_offset(leaf, fi);
1219 em->orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
1220 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1221 if (bytenr == 0) {
1222 em->block_start = EXTENT_MAP_HOLE;
1223 return;
1224 }
1225 if (compress_type != BTRFS_COMPRESS_NONE) {
1226 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
1227 em->compress_type = compress_type;
1228 em->block_start = bytenr;
1229 em->block_len = em->orig_block_len;
1230 } else {
1231 bytenr += btrfs_file_extent_offset(leaf, fi);
1232 em->block_start = bytenr;
1233 em->block_len = em->len;
1234 if (type == BTRFS_FILE_EXTENT_PREALLOC)
1235 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
1236 }
1237 } else if (type == BTRFS_FILE_EXTENT_INLINE) {
1238 em->block_start = EXTENT_MAP_INLINE;
1239 em->start = extent_start;
1240 em->len = extent_end - extent_start;
1241 /*
1242 * Initialize orig_start and block_len with the same values
1243 * as in inode.c:btrfs_get_extent().
1244 */
1245 em->orig_start = EXTENT_MAP_HOLE;
1246 em->block_len = (u64)-1;
1247 if (!new_inline && compress_type != BTRFS_COMPRESS_NONE) {
1248 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
1249 em->compress_type = compress_type;
1250 }
1251 } else {
1252 btrfs_err(fs_info,
1253 "unknown file extent item type %d, inode %llu, offset %llu, "
1254 "root %llu", type, btrfs_ino(inode), extent_start,
1255 root->root_key.objectid);
1256 }
1257}
1258
1259/*
1260 * Returns the end offset (non inclusive) of the file extent item the given path
1261 * points to. If it points to an inline extent, the returned offset is rounded
1262 * up to the sector size.
1263 */
1264u64 btrfs_file_extent_end(const struct btrfs_path *path)
1265{
1266 const struct extent_buffer *leaf = path->nodes[0];
1267 const int slot = path->slots[0];
1268 struct btrfs_file_extent_item *fi;
1269 struct btrfs_key key;
1270 u64 end;
1271
1272 btrfs_item_key_to_cpu(leaf, &key, slot);
1273 ASSERT(key.type == BTRFS_EXTENT_DATA_KEY);
1274 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
1275
1276 if (btrfs_file_extent_type(leaf, fi) == BTRFS_FILE_EXTENT_INLINE) {
1277 end = btrfs_file_extent_ram_bytes(leaf, fi);
1278 end = ALIGN(key.offset + end, leaf->fs_info->sectorsize);
1279 } else {
1280 end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
1281 }
1282
1283 return end;
1284}