1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2007 Oracle. All rights reserved.
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>
15 #include "transaction.h"
17 #include "print-tree.h"
18 #include "compression.h"
20 #define __MAX_CSUM_ITEMS(r, size) ((unsigned long)(((BTRFS_LEAF_DATA_SIZE(r) - \
21 sizeof(struct btrfs_item) * 2) / \
24 #define MAX_CSUM_ITEMS(r, size) (min_t(u32, __MAX_CSUM_ITEMS(r, size), \
28 * Set inode's size according to filesystem options
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
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
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.
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.
44 void btrfs_inode_safe_disk_i_size_write(struct btrfs_inode *inode, u64 new_i_size)
46 struct btrfs_fs_info *fs_info = inode->root->fs_info;
47 u64 start, end, i_size;
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;
56 spin_lock(&inode->lock);
57 ret = find_contiguous_extent_bit(&inode->file_extent_tree, 0, &start,
59 if (!ret && start == 0)
60 i_size = min(i_size, end + 1);
63 inode->disk_i_size = i_size;
64 spin_unlock(&inode->lock);
68 * Mark range within a file as having a new extent inserted
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
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.
78 * The start and len must match the file extent item, so thus must be sectorsize
81 int btrfs_inode_set_file_extent_range(struct btrfs_inode *inode, u64 start,
87 ASSERT(IS_ALIGNED(start + len, inode->root->fs_info->sectorsize));
89 if (btrfs_fs_incompat(inode->root->fs_info, NO_HOLES))
91 return set_extent_bits(&inode->file_extent_tree, start, start + len - 1,
96 * Marks an inode range as not having a backing extent
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
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.
106 * The start and len must match the file extent item, so thus must be sectorsize
109 int btrfs_inode_clear_file_extent_range(struct btrfs_inode *inode, u64 start,
115 ASSERT(IS_ALIGNED(start + len, inode->root->fs_info->sectorsize) ||
118 if (btrfs_fs_incompat(inode->root->fs_info, NO_HOLES))
120 return clear_extent_bit(&inode->file_extent_tree, start,
121 start + len - 1, EXTENT_DIRTY, 0, 0, NULL);
124 static inline u32 max_ordered_sum_bytes(struct btrfs_fs_info *fs_info,
127 u32 ncsums = (PAGE_SIZE - sizeof(struct btrfs_ordered_sum)) / csum_size;
129 return ncsums * fs_info->sectorsize;
132 int 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)
140 struct btrfs_file_extent_item *item;
141 struct btrfs_key file_key;
142 struct btrfs_path *path;
143 struct extent_buffer *leaf;
145 path = btrfs_alloc_path();
148 file_key.objectid = objectid;
149 file_key.offset = pos;
150 file_key.type = BTRFS_EXTENT_DATA_KEY;
152 ret = btrfs_insert_empty_item(trans, root, path, &file_key,
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);
171 btrfs_mark_buffer_dirty(leaf);
173 btrfs_free_path(path);
177 static struct btrfs_csum_item *
178 btrfs_lookup_csum(struct btrfs_trans_handle *trans,
179 struct btrfs_root *root,
180 struct btrfs_path *path,
183 struct btrfs_fs_info *fs_info = root->fs_info;
185 struct btrfs_key file_key;
186 struct btrfs_key found_key;
187 struct btrfs_csum_item *item;
188 struct extent_buffer *leaf;
190 const u32 csum_size = fs_info->csum_size;
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);
199 leaf = path->nodes[0];
202 if (path->slots[0] == 0)
205 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
206 if (found_key.type != BTRFS_EXTENT_CSUM_KEY)
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;
214 if (csum_offset == csums_in_item) {
217 } else if (csum_offset > csums_in_item) {
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);
231 int btrfs_lookup_file_extent(struct btrfs_trans_handle *trans,
232 struct btrfs_root *root,
233 struct btrfs_path *path, u64 objectid,
237 struct btrfs_key file_key;
238 int ins_len = mod < 0 ? -1 : 0;
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);
249 * Find checksums for logical bytenr range [disk_bytenr, disk_bytenr + len) and
250 * estore the result to @dst.
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.
257 static int search_csum_tree(struct btrfs_fs_info *fs_info,
258 struct btrfs_path *path, u64 disk_bytenr,
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;
270 ASSERT(IS_ALIGNED(disk_bytenr, sectorsize) &&
271 IS_ALIGNED(len, sectorsize));
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]);
280 csum_start = key.offset;
281 csum_len = (itemsize / csum_size) * sectorsize;
283 if (in_range(disk_bytenr, csum_start, csum_len))
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);
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]);
297 csum_start = key.offset;
298 csum_len = (itemsize / csum_size) * sectorsize;
299 ASSERT(in_range(disk_bytenr, csum_start, csum_len));
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,
313 * Locate the file_offset of @cur_disk_bytenr of a @bio.
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.
320 * @inode is used to determine if the bvec page really belongs to @inode.
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
325 static int search_file_offset_in_bio(struct bio *bio, struct inode *inode,
326 u64 disk_bytenr, u64 *file_offset_ret)
328 struct bvec_iter iter;
330 u64 cur = bio->bi_iter.bi_sector << SECTOR_SHIFT;
333 bio_for_each_segment(bvec, bio, iter) {
334 struct page *page = bvec.bv_page;
336 if (cur > disk_bytenr)
338 if (cur + bvec.bv_len <= disk_bytenr) {
342 ASSERT(in_range(disk_bytenr, cur, bvec.bv_len));
343 if (page->mapping && page->mapping->host &&
344 page->mapping->host == inode) {
346 *file_offset_ret = page_offset(page) + bvec.bv_offset +
355 * Lookup the checksum for the read bio in csum tree.
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.
364 * Return: BLK_STS_RESOURCE if allocating memory fails, BLK_STS_OK otherwise.
366 blk_status_t btrfs_lookup_bio_sums(struct inode *inode, struct bio *bio, u8 *dst)
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;
377 const unsigned int nblocks = orig_len >> fs_info->sectorsize_bits;
380 if (!fs_info->csum_root || (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM))
384 * This function is only called for read bio.
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
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
395 ASSERT(bio_op(bio) == REQ_OP_READ);
396 path = btrfs_alloc_path();
398 return BLK_STS_RESOURCE;
401 struct btrfs_io_bio *btrfs_bio = btrfs_io_bio(bio);
403 if (nblocks * csum_size > BTRFS_BIO_INLINE_CSUM_SIZE) {
404 btrfs_bio->csum = kmalloc_array(nblocks, csum_size,
406 if (!btrfs_bio->csum) {
407 btrfs_free_path(path);
408 return BLK_STS_RESOURCE;
411 btrfs_bio->csum = btrfs_bio->csum_inline;
413 csum = btrfs_bio->csum;
419 * If requested number of sectors is larger than one leaf can contain,
420 * kick the readahead for csum tree.
422 if (nblocks > fs_info->csums_per_leaf)
423 path->reada = READA_FORWARD;
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.
431 if (btrfs_is_free_space_inode(BTRFS_I(inode))) {
432 path->search_commit_root = 1;
433 path->skip_locking = 1;
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;
444 * Although both cur_disk_bytenr and orig_disk_bytenr is u64,
445 * we're calculating the offset to the bio start.
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
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;
456 count = search_csum_tree(fs_info, path, cur_disk_bytenr,
457 search_len, csum_dst);
460 * Either we hit a critical error or we didn't find
462 * Either way, we put zero into the csums dst, and skip
463 * to the next sector.
465 memset(csum_dst, 0, csum_size);
469 * For data reloc inode, we need to mark the range
470 * NODATASUM so that balance won't report false csum
473 if (BTRFS_I(inode)->root->root_key.objectid ==
474 BTRFS_DATA_RELOC_TREE_OBJECTID) {
478 ret = search_file_offset_in_bio(bio, inode,
479 cur_disk_bytenr, &file_offset);
481 set_extent_bits(io_tree, file_offset,
482 file_offset + sectorsize - 1,
485 btrfs_warn_rl(fs_info,
486 "csum hole found for disk bytenr range [%llu, %llu)",
487 cur_disk_bytenr, cur_disk_bytenr + sectorsize);
492 btrfs_free_path(path);
496 int btrfs_lookup_csums_range(struct btrfs_root *root, u64 start, u64 end,
497 struct list_head *list, int search_commit)
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;
506 unsigned long offset;
510 const u32 csum_size = fs_info->csum_size;
512 ASSERT(IS_ALIGNED(start, fs_info->sectorsize) &&
513 IS_ALIGNED(end + 1, fs_info->sectorsize));
515 path = btrfs_alloc_path();
520 path->skip_locking = 1;
521 path->reada = READA_FORWARD;
522 path->search_commit_root = 1;
525 key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
527 key.type = BTRFS_EXTENT_CSUM_KEY;
529 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
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))
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);
552 leaf = path->nodes[0];
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 ||
561 if (key.offset > start)
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) {
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),
584 sums->bytenr = start;
585 sums->len = (int)size;
587 offset = (start - key.offset) >> fs_info->sectorsize_bits;
589 size >>= fs_info->sectorsize_bits;
591 read_extent_buffer(path->nodes[0],
593 ((unsigned long)item) + offset,
596 start += fs_info->sectorsize * size;
597 list_add_tail(&sums->list, &tmplist);
603 while (ret < 0 && !list_empty(&tmplist)) {
604 sums = list_entry(tmplist.next, struct btrfs_ordered_sum, list);
605 list_del(&sums->list);
608 list_splice_tail(&tmplist, list);
610 btrfs_free_path(path);
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 contains potentially discontigous bio vecs
622 * so the logical offset of each should be calculated separately.
624 blk_status_t btrfs_csum_one_bio(struct btrfs_inode *inode, struct bio *bio,
625 u64 file_start, int contig)
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;
632 struct bvec_iter iter;
636 unsigned long total_bytes = 0;
637 unsigned long this_sum_bytes = 0;
642 nofs_flag = memalloc_nofs_save();
643 sums = kvzalloc(btrfs_ordered_sum_size(fs_info, bio->bi_iter.bi_size),
645 memalloc_nofs_restore(nofs_flag);
648 return BLK_STS_RESOURCE;
650 sums->len = bio->bi_iter.bi_size;
651 INIT_LIST_HEAD(&sums->list);
656 offset = 0; /* shut up gcc */
658 sums->bytenr = bio->bi_iter.bi_sector << 9;
661 shash->tfm = fs_info->csum_shash;
663 bio_for_each_segment(bvec, bio, iter) {
665 offset = page_offset(bvec.bv_page) + bvec.bv_offset;
668 ordered = btrfs_lookup_ordered_extent(inode, offset);
669 BUG_ON(!ordered); /* Logic error */
672 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info,
673 bvec.bv_len + fs_info->sectorsize
676 for (i = 0; i < nr_sectors; i++) {
677 if (offset >= ordered->file_offset + ordered->num_bytes ||
678 offset < ordered->file_offset) {
679 unsigned long bytes_left;
681 sums->len = this_sum_bytes;
683 btrfs_add_ordered_sum(ordered, sums);
684 btrfs_put_ordered_extent(ordered);
686 bytes_left = bio->bi_iter.bi_size - total_bytes;
688 nofs_flag = memalloc_nofs_save();
689 sums = kvzalloc(btrfs_ordered_sum_size(fs_info,
690 bytes_left), GFP_KERNEL);
691 memalloc_nofs_restore(nofs_flag);
692 BUG_ON(!sums); /* -ENOMEM */
693 sums->len = bytes_left;
694 ordered = btrfs_lookup_ordered_extent(inode,
696 ASSERT(ordered); /* Logic error */
697 sums->bytenr = (bio->bi_iter.bi_sector << 9)
702 data = kmap_atomic(bvec.bv_page);
703 crypto_shash_digest(shash, data + bvec.bv_offset
704 + (i * fs_info->sectorsize),
708 index += fs_info->csum_size;
709 offset += fs_info->sectorsize;
710 this_sum_bytes += fs_info->sectorsize;
711 total_bytes += fs_info->sectorsize;
716 btrfs_add_ordered_sum(ordered, sums);
717 btrfs_put_ordered_extent(ordered);
722 * helper function for csum removal, this expects the
723 * key to describe the csum pointed to by the path, and it expects
724 * the csum to overlap the range [bytenr, len]
726 * The csum should not be entirely contained in the range and the
727 * range should not be entirely contained in the csum.
729 * This calls btrfs_truncate_item with the correct args based on the
730 * overlap, and fixes up the key as required.
732 static noinline void truncate_one_csum(struct btrfs_fs_info *fs_info,
733 struct btrfs_path *path,
734 struct btrfs_key *key,
737 struct extent_buffer *leaf;
738 const u32 csum_size = fs_info->csum_size;
740 u64 end_byte = bytenr + len;
741 u32 blocksize_bits = fs_info->sectorsize_bits;
743 leaf = path->nodes[0];
744 csum_end = btrfs_item_size_nr(leaf, path->slots[0]) / csum_size;
745 csum_end <<= blocksize_bits;
746 csum_end += key->offset;
748 if (key->offset < bytenr && csum_end <= end_byte) {
753 * A simple truncate off the end of the item
755 u32 new_size = (bytenr - key->offset) >> blocksize_bits;
756 new_size *= csum_size;
757 btrfs_truncate_item(path, new_size, 1);
758 } else if (key->offset >= bytenr && csum_end > end_byte &&
759 end_byte > key->offset) {
764 * we need to truncate from the beginning of the csum
766 u32 new_size = (csum_end - end_byte) >> blocksize_bits;
767 new_size *= csum_size;
769 btrfs_truncate_item(path, new_size, 0);
771 key->offset = end_byte;
772 btrfs_set_item_key_safe(fs_info, path, key);
779 * deletes the csum items from the csum tree for a given
782 int btrfs_del_csums(struct btrfs_trans_handle *trans,
783 struct btrfs_root *root, u64 bytenr, u64 len)
785 struct btrfs_fs_info *fs_info = trans->fs_info;
786 struct btrfs_path *path;
787 struct btrfs_key key;
788 u64 end_byte = bytenr + len;
790 struct extent_buffer *leaf;
792 const u32 csum_size = fs_info->csum_size;
793 u32 blocksize_bits = fs_info->sectorsize_bits;
795 ASSERT(root == fs_info->csum_root ||
796 root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID);
798 path = btrfs_alloc_path();
803 key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
804 key.offset = end_byte - 1;
805 key.type = BTRFS_EXTENT_CSUM_KEY;
807 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
809 if (path->slots[0] == 0)
812 } else if (ret < 0) {
816 leaf = path->nodes[0];
817 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
819 if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
820 key.type != BTRFS_EXTENT_CSUM_KEY) {
824 if (key.offset >= end_byte)
827 csum_end = btrfs_item_size_nr(leaf, path->slots[0]) / csum_size;
828 csum_end <<= blocksize_bits;
829 csum_end += key.offset;
831 /* this csum ends before we start, we're done */
832 if (csum_end <= bytenr)
835 /* delete the entire item, it is inside our range */
836 if (key.offset >= bytenr && csum_end <= end_byte) {
840 * Check how many csum items preceding this one in this
841 * leaf correspond to our range and then delete them all
844 if (key.offset > bytenr && path->slots[0] > 0) {
845 int slot = path->slots[0] - 1;
850 btrfs_item_key_to_cpu(leaf, &pk, slot);
851 if (pk.offset < bytenr ||
852 pk.type != BTRFS_EXTENT_CSUM_KEY ||
854 BTRFS_EXTENT_CSUM_OBJECTID)
856 path->slots[0] = slot;
858 key.offset = pk.offset;
862 ret = btrfs_del_items(trans, root, path,
863 path->slots[0], del_nr);
866 if (key.offset == bytenr)
868 } else if (key.offset < bytenr && csum_end > end_byte) {
869 unsigned long offset;
870 unsigned long shift_len;
871 unsigned long item_offset;
876 * Our bytes are in the middle of the csum,
877 * we need to split this item and insert a new one.
879 * But we can't drop the path because the
880 * csum could change, get removed, extended etc.
882 * The trick here is the max size of a csum item leaves
883 * enough room in the tree block for a single
884 * item header. So, we split the item in place,
885 * adding a new header pointing to the existing
886 * bytes. Then we loop around again and we have
887 * a nicely formed csum item that we can neatly
890 offset = (bytenr - key.offset) >> blocksize_bits;
893 shift_len = (len >> blocksize_bits) * csum_size;
895 item_offset = btrfs_item_ptr_offset(leaf,
898 memzero_extent_buffer(leaf, item_offset + offset,
903 * btrfs_split_item returns -EAGAIN when the
904 * item changed size or key
906 ret = btrfs_split_item(trans, root, path, &key, offset);
907 if (ret && ret != -EAGAIN) {
908 btrfs_abort_transaction(trans, ret);
912 key.offset = end_byte - 1;
914 truncate_one_csum(fs_info, path, &key, bytenr, len);
915 if (key.offset < bytenr)
918 btrfs_release_path(path);
922 btrfs_free_path(path);
926 int btrfs_csum_file_blocks(struct btrfs_trans_handle *trans,
927 struct btrfs_root *root,
928 struct btrfs_ordered_sum *sums)
930 struct btrfs_fs_info *fs_info = root->fs_info;
931 struct btrfs_key file_key;
932 struct btrfs_key found_key;
933 struct btrfs_path *path;
934 struct btrfs_csum_item *item;
935 struct btrfs_csum_item *item_end;
936 struct extent_buffer *leaf = NULL;
946 const u32 csum_size = fs_info->csum_size;
948 path = btrfs_alloc_path();
952 next_offset = (u64)-1;
954 bytenr = sums->bytenr + total_bytes;
955 file_key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
956 file_key.offset = bytenr;
957 file_key.type = BTRFS_EXTENT_CSUM_KEY;
959 item = btrfs_lookup_csum(trans, root, path, bytenr, 1);
962 leaf = path->nodes[0];
963 item_end = btrfs_item_ptr(leaf, path->slots[0],
964 struct btrfs_csum_item);
965 item_end = (struct btrfs_csum_item *)((char *)item_end +
966 btrfs_item_size_nr(leaf, path->slots[0]));
970 if (ret != -EFBIG && ret != -ENOENT)
975 /* we found one, but it isn't big enough yet */
976 leaf = path->nodes[0];
977 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
978 if ((item_size / csum_size) >=
979 MAX_CSUM_ITEMS(fs_info, csum_size)) {
980 /* already at max size, make a new one */
984 int slot = path->slots[0] + 1;
985 /* we didn't find a csum item, insert one */
986 nritems = btrfs_header_nritems(path->nodes[0]);
987 if (!nritems || (path->slots[0] >= nritems - 1)) {
988 ret = btrfs_next_leaf(root, path);
991 } else if (ret > 0) {
995 slot = path->slots[0];
997 btrfs_item_key_to_cpu(path->nodes[0], &found_key, slot);
998 if (found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
999 found_key.type != BTRFS_EXTENT_CSUM_KEY) {
1003 next_offset = found_key.offset;
1009 * At this point, we know the tree has a checksum item that ends at an
1010 * offset matching the start of the checksum range we want to insert.
1011 * We try to extend that item as much as possible and then add as many
1012 * checksums to it as they fit.
1014 * First check if the leaf has enough free space for at least one
1015 * checksum. If it has go directly to the item extension code, otherwise
1016 * release the path and do a search for insertion before the extension.
1018 if (btrfs_leaf_free_space(leaf) >= csum_size) {
1019 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1020 csum_offset = (bytenr - found_key.offset) >>
1021 fs_info->sectorsize_bits;
1025 btrfs_release_path(path);
1026 path->search_for_extension = 1;
1027 ret = btrfs_search_slot(trans, root, &file_key, path,
1029 path->search_for_extension = 0;
1034 if (path->slots[0] == 0)
1039 leaf = path->nodes[0];
1040 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1041 csum_offset = (bytenr - found_key.offset) >> fs_info->sectorsize_bits;
1043 if (found_key.type != BTRFS_EXTENT_CSUM_KEY ||
1044 found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
1045 csum_offset >= MAX_CSUM_ITEMS(fs_info, csum_size)) {
1050 if (csum_offset == btrfs_item_size_nr(leaf, path->slots[0]) /
1056 tmp = sums->len - total_bytes;
1057 tmp >>= fs_info->sectorsize_bits;
1060 extend_nr = max_t(int, 1, (int)tmp);
1061 diff = (csum_offset + extend_nr) * csum_size;
1063 MAX_CSUM_ITEMS(fs_info, csum_size) * csum_size);
1065 diff = diff - btrfs_item_size_nr(leaf, path->slots[0]);
1066 diff = min_t(u32, btrfs_leaf_free_space(leaf), diff);
1070 btrfs_extend_item(path, diff);
1076 btrfs_release_path(path);
1081 tmp = sums->len - total_bytes;
1082 tmp >>= fs_info->sectorsize_bits;
1083 tmp = min(tmp, (next_offset - file_key.offset) >>
1084 fs_info->sectorsize_bits);
1086 tmp = max_t(u64, 1, tmp);
1087 tmp = min_t(u64, tmp, MAX_CSUM_ITEMS(fs_info, csum_size));
1088 ins_size = csum_size * tmp;
1090 ins_size = csum_size;
1092 ret = btrfs_insert_empty_item(trans, root, path, &file_key,
1096 if (WARN_ON(ret != 0))
1098 leaf = path->nodes[0];
1100 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item);
1101 item_end = (struct btrfs_csum_item *)((unsigned char *)item +
1102 btrfs_item_size_nr(leaf, path->slots[0]));
1103 item = (struct btrfs_csum_item *)((unsigned char *)item +
1104 csum_offset * csum_size);
1106 ins_size = (u32)(sums->len - total_bytes) >> fs_info->sectorsize_bits;
1107 ins_size *= csum_size;
1108 ins_size = min_t(u32, (unsigned long)item_end - (unsigned long)item,
1110 write_extent_buffer(leaf, sums->sums + index, (unsigned long)item,
1114 ins_size /= csum_size;
1115 total_bytes += ins_size * fs_info->sectorsize;
1117 btrfs_mark_buffer_dirty(path->nodes[0]);
1118 if (total_bytes < sums->len) {
1119 btrfs_release_path(path);
1124 btrfs_free_path(path);
1128 void btrfs_extent_item_to_extent_map(struct btrfs_inode *inode,
1129 const struct btrfs_path *path,
1130 struct btrfs_file_extent_item *fi,
1131 const bool new_inline,
1132 struct extent_map *em)
1134 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1135 struct btrfs_root *root = inode->root;
1136 struct extent_buffer *leaf = path->nodes[0];
1137 const int slot = path->slots[0];
1138 struct btrfs_key key;
1139 u64 extent_start, extent_end;
1141 u8 type = btrfs_file_extent_type(leaf, fi);
1142 int compress_type = btrfs_file_extent_compression(leaf, fi);
1144 btrfs_item_key_to_cpu(leaf, &key, slot);
1145 extent_start = key.offset;
1146 extent_end = btrfs_file_extent_end(path);
1147 em->ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1148 if (type == BTRFS_FILE_EXTENT_REG ||
1149 type == BTRFS_FILE_EXTENT_PREALLOC) {
1150 em->start = extent_start;
1151 em->len = extent_end - extent_start;
1152 em->orig_start = extent_start -
1153 btrfs_file_extent_offset(leaf, fi);
1154 em->orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
1155 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1157 em->block_start = EXTENT_MAP_HOLE;
1160 if (compress_type != BTRFS_COMPRESS_NONE) {
1161 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
1162 em->compress_type = compress_type;
1163 em->block_start = bytenr;
1164 em->block_len = em->orig_block_len;
1166 bytenr += btrfs_file_extent_offset(leaf, fi);
1167 em->block_start = bytenr;
1168 em->block_len = em->len;
1169 if (type == BTRFS_FILE_EXTENT_PREALLOC)
1170 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
1172 } else if (type == BTRFS_FILE_EXTENT_INLINE) {
1173 em->block_start = EXTENT_MAP_INLINE;
1174 em->start = extent_start;
1175 em->len = extent_end - extent_start;
1177 * Initialize orig_start and block_len with the same values
1178 * as in inode.c:btrfs_get_extent().
1180 em->orig_start = EXTENT_MAP_HOLE;
1181 em->block_len = (u64)-1;
1182 if (!new_inline && compress_type != BTRFS_COMPRESS_NONE) {
1183 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
1184 em->compress_type = compress_type;
1188 "unknown file extent item type %d, inode %llu, offset %llu, "
1189 "root %llu", type, btrfs_ino(inode), extent_start,
1190 root->root_key.objectid);
1195 * Returns the end offset (non inclusive) of the file extent item the given path
1196 * points to. If it points to an inline extent, the returned offset is rounded
1197 * up to the sector size.
1199 u64 btrfs_file_extent_end(const struct btrfs_path *path)
1201 const struct extent_buffer *leaf = path->nodes[0];
1202 const int slot = path->slots[0];
1203 struct btrfs_file_extent_item *fi;
1204 struct btrfs_key key;
1207 btrfs_item_key_to_cpu(leaf, &key, slot);
1208 ASSERT(key.type == BTRFS_EXTENT_DATA_KEY);
1209 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
1211 if (btrfs_file_extent_type(leaf, fi) == BTRFS_FILE_EXTENT_INLINE) {
1212 end = btrfs_file_extent_ram_bytes(leaf, fi);
1213 end = ALIGN(key.offset + end, leaf->fs_info->sectorsize);
1215 end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);