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,
236 struct btrfs_key file_key;
237 int ins_len = mod < 0 ? -1 : 0;
240 file_key.objectid = objectid;
241 file_key.offset = offset;
242 file_key.type = BTRFS_EXTENT_DATA_KEY;
244 return btrfs_search_slot(trans, root, &file_key, path, ins_len, cow);
248 * Find checksums for logical bytenr range [disk_bytenr, disk_bytenr + len) and
249 * estore the result to @dst.
251 * Return >0 for the number of sectors we found.
252 * Return 0 for the range [disk_bytenr, disk_bytenr + sectorsize) has no csum
253 * for it. Caller may want to try next sector until one range is hit.
254 * Return <0 for fatal error.
256 static int search_csum_tree(struct btrfs_fs_info *fs_info,
257 struct btrfs_path *path, u64 disk_bytenr,
260 struct btrfs_csum_item *item = NULL;
261 struct btrfs_key key;
262 const u32 sectorsize = fs_info->sectorsize;
263 const u32 csum_size = fs_info->csum_size;
269 ASSERT(IS_ALIGNED(disk_bytenr, sectorsize) &&
270 IS_ALIGNED(len, sectorsize));
272 /* Check if the current csum item covers disk_bytenr */
273 if (path->nodes[0]) {
274 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
275 struct btrfs_csum_item);
276 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
277 itemsize = btrfs_item_size_nr(path->nodes[0], path->slots[0]);
279 csum_start = key.offset;
280 csum_len = (itemsize / csum_size) * sectorsize;
282 if (in_range(disk_bytenr, csum_start, csum_len))
286 /* Current item doesn't contain the desired range, search again */
287 btrfs_release_path(path);
288 item = btrfs_lookup_csum(NULL, fs_info->csum_root, path, disk_bytenr, 0);
293 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
294 itemsize = btrfs_item_size_nr(path->nodes[0], path->slots[0]);
296 csum_start = key.offset;
297 csum_len = (itemsize / csum_size) * sectorsize;
298 ASSERT(in_range(disk_bytenr, csum_start, csum_len));
301 ret = (min(csum_start + csum_len, disk_bytenr + len) -
302 disk_bytenr) >> fs_info->sectorsize_bits;
303 read_extent_buffer(path->nodes[0], dst, (unsigned long)item,
312 * Locate the file_offset of @cur_disk_bytenr of a @bio.
314 * Bio of btrfs represents read range of
315 * [bi_sector << 9, bi_sector << 9 + bi_size).
316 * Knowing this, we can iterate through each bvec to locate the page belong to
317 * @cur_disk_bytenr and get the file offset.
319 * @inode is used to determine if the bvec page really belongs to @inode.
321 * Return 0 if we can't find the file offset
322 * Return >0 if we find the file offset and restore it to @file_offset_ret
324 static int search_file_offset_in_bio(struct bio *bio, struct inode *inode,
325 u64 disk_bytenr, u64 *file_offset_ret)
327 struct bvec_iter iter;
329 u64 cur = bio->bi_iter.bi_sector << SECTOR_SHIFT;
332 bio_for_each_segment(bvec, bio, iter) {
333 struct page *page = bvec.bv_page;
335 if (cur > disk_bytenr)
337 if (cur + bvec.bv_len <= disk_bytenr) {
341 ASSERT(in_range(disk_bytenr, cur, bvec.bv_len));
342 if (page->mapping && page->mapping->host &&
343 page->mapping->host == inode) {
345 *file_offset_ret = page_offset(page) + bvec.bv_offset +
354 * Lookup the checksum for the read bio in csum tree.
356 * @inode: inode that the bio is for.
357 * @bio: bio to look up.
358 * @dst: Buffer of size nblocks * btrfs_super_csum_size() used to return
359 * checksum (nblocks = bio->bi_iter.bi_size / fs_info->sectorsize). If
360 * NULL, the checksum buffer is allocated and returned in
361 * btrfs_io_bio(bio)->csum instead.
363 * Return: BLK_STS_RESOURCE if allocating memory fails, BLK_STS_OK otherwise.
365 blk_status_t btrfs_lookup_bio_sums(struct inode *inode, struct bio *bio, u8 *dst)
367 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
368 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
369 struct btrfs_path *path;
370 const u32 sectorsize = fs_info->sectorsize;
371 const u32 csum_size = fs_info->csum_size;
372 u32 orig_len = bio->bi_iter.bi_size;
373 u64 orig_disk_bytenr = bio->bi_iter.bi_sector << SECTOR_SHIFT;
376 const unsigned int nblocks = orig_len >> fs_info->sectorsize_bits;
379 if (!fs_info->csum_root || (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM))
383 * This function is only called for read bio.
385 * This means two things:
386 * - All our csums should only be in csum tree
387 * No ordered extents csums, as ordered extents are only for write
389 * - No need to bother any other info from bvec
390 * Since we're looking up csums, the only important info is the
391 * disk_bytenr and the length, which can be extracted from bi_iter
394 ASSERT(bio_op(bio) == REQ_OP_READ);
395 path = btrfs_alloc_path();
397 return BLK_STS_RESOURCE;
400 struct btrfs_io_bio *btrfs_bio = btrfs_io_bio(bio);
402 if (nblocks * csum_size > BTRFS_BIO_INLINE_CSUM_SIZE) {
403 btrfs_bio->csum = kmalloc_array(nblocks, csum_size,
405 if (!btrfs_bio->csum) {
406 btrfs_free_path(path);
407 return BLK_STS_RESOURCE;
410 btrfs_bio->csum = btrfs_bio->csum_inline;
412 csum = btrfs_bio->csum;
418 * If requested number of sectors is larger than one leaf can contain,
419 * kick the readahead for csum tree.
421 if (nblocks > fs_info->csums_per_leaf)
422 path->reada = READA_FORWARD;
425 * the free space stuff is only read when it hasn't been
426 * updated in the current transaction. So, we can safely
427 * read from the commit root and sidestep a nasty deadlock
428 * between reading the free space cache and updating the csum tree.
430 if (btrfs_is_free_space_inode(BTRFS_I(inode))) {
431 path->search_commit_root = 1;
432 path->skip_locking = 1;
435 for (cur_disk_bytenr = orig_disk_bytenr;
436 cur_disk_bytenr < orig_disk_bytenr + orig_len;
437 cur_disk_bytenr += (count * sectorsize)) {
438 u64 search_len = orig_disk_bytenr + orig_len - cur_disk_bytenr;
439 unsigned int sector_offset;
443 * Although both cur_disk_bytenr and orig_disk_bytenr is u64,
444 * we're calculating the offset to the bio start.
446 * Bio size is limited to UINT_MAX, thus unsigned int is large
447 * enough to contain the raw result, not to mention the right
450 ASSERT(cur_disk_bytenr - orig_disk_bytenr < UINT_MAX);
451 sector_offset = (cur_disk_bytenr - orig_disk_bytenr) >>
452 fs_info->sectorsize_bits;
453 csum_dst = csum + sector_offset * csum_size;
455 count = search_csum_tree(fs_info, path, cur_disk_bytenr,
456 search_len, csum_dst);
459 * Either we hit a critical error or we didn't find
461 * Either way, we put zero into the csums dst, and skip
462 * to the next sector.
464 memset(csum_dst, 0, csum_size);
468 * For data reloc inode, we need to mark the range
469 * NODATASUM so that balance won't report false csum
472 if (BTRFS_I(inode)->root->root_key.objectid ==
473 BTRFS_DATA_RELOC_TREE_OBJECTID) {
477 ret = search_file_offset_in_bio(bio, inode,
478 cur_disk_bytenr, &file_offset);
480 set_extent_bits(io_tree, file_offset,
481 file_offset + sectorsize - 1,
484 btrfs_warn_rl(fs_info,
485 "csum hole found for disk bytenr range [%llu, %llu)",
486 cur_disk_bytenr, cur_disk_bytenr + sectorsize);
491 btrfs_free_path(path);
495 int btrfs_lookup_csums_range(struct btrfs_root *root, u64 start, u64 end,
496 struct list_head *list, int search_commit)
498 struct btrfs_fs_info *fs_info = root->fs_info;
499 struct btrfs_key key;
500 struct btrfs_path *path;
501 struct extent_buffer *leaf;
502 struct btrfs_ordered_sum *sums;
503 struct btrfs_csum_item *item;
505 unsigned long offset;
509 const u32 csum_size = fs_info->csum_size;
511 ASSERT(IS_ALIGNED(start, fs_info->sectorsize) &&
512 IS_ALIGNED(end + 1, fs_info->sectorsize));
514 path = btrfs_alloc_path();
519 path->skip_locking = 1;
520 path->reada = READA_FORWARD;
521 path->search_commit_root = 1;
524 key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
526 key.type = BTRFS_EXTENT_CSUM_KEY;
528 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
531 if (ret > 0 && path->slots[0] > 0) {
532 leaf = path->nodes[0];
533 btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
534 if (key.objectid == BTRFS_EXTENT_CSUM_OBJECTID &&
535 key.type == BTRFS_EXTENT_CSUM_KEY) {
536 offset = (start - key.offset) >> fs_info->sectorsize_bits;
537 if (offset * csum_size <
538 btrfs_item_size_nr(leaf, path->slots[0] - 1))
543 while (start <= end) {
544 leaf = path->nodes[0];
545 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
546 ret = btrfs_next_leaf(root, path);
551 leaf = path->nodes[0];
554 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
555 if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
556 key.type != BTRFS_EXTENT_CSUM_KEY ||
560 if (key.offset > start)
563 size = btrfs_item_size_nr(leaf, path->slots[0]);
564 csum_end = key.offset + (size / csum_size) * fs_info->sectorsize;
565 if (csum_end <= start) {
570 csum_end = min(csum_end, end + 1);
571 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
572 struct btrfs_csum_item);
573 while (start < csum_end) {
574 size = min_t(size_t, csum_end - start,
575 max_ordered_sum_bytes(fs_info, csum_size));
576 sums = kzalloc(btrfs_ordered_sum_size(fs_info, size),
583 sums->bytenr = start;
584 sums->len = (int)size;
586 offset = (start - key.offset) >> fs_info->sectorsize_bits;
588 size >>= fs_info->sectorsize_bits;
590 read_extent_buffer(path->nodes[0],
592 ((unsigned long)item) + offset,
595 start += fs_info->sectorsize * size;
596 list_add_tail(&sums->list, &tmplist);
602 while (ret < 0 && !list_empty(&tmplist)) {
603 sums = list_entry(tmplist.next, struct btrfs_ordered_sum, list);
604 list_del(&sums->list);
607 list_splice_tail(&tmplist, list);
609 btrfs_free_path(path);
614 * btrfs_csum_one_bio - Calculates checksums of the data contained inside a bio
615 * @inode: Owner of the data inside the bio
616 * @bio: Contains the data to be checksummed
617 * @file_start: offset in file this bio begins to describe
618 * @contig: Boolean. If true/1 means all bio vecs in this bio are
619 * contiguous and they begin at @file_start in the file. False/0
620 * means this bio can contain potentially discontiguous bio vecs
621 * so the logical offset of each should be calculated separately.
623 blk_status_t btrfs_csum_one_bio(struct btrfs_inode *inode, struct bio *bio,
624 u64 file_start, int contig)
626 struct btrfs_fs_info *fs_info = inode->root->fs_info;
627 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
628 struct btrfs_ordered_sum *sums;
629 struct btrfs_ordered_extent *ordered = NULL;
631 struct bvec_iter iter;
635 unsigned long total_bytes = 0;
636 unsigned long this_sum_bytes = 0;
641 nofs_flag = memalloc_nofs_save();
642 sums = kvzalloc(btrfs_ordered_sum_size(fs_info, bio->bi_iter.bi_size),
644 memalloc_nofs_restore(nofs_flag);
647 return BLK_STS_RESOURCE;
649 sums->len = bio->bi_iter.bi_size;
650 INIT_LIST_HEAD(&sums->list);
655 offset = 0; /* shut up gcc */
657 sums->bytenr = bio->bi_iter.bi_sector << 9;
660 shash->tfm = fs_info->csum_shash;
662 bio_for_each_segment(bvec, bio, iter) {
664 offset = page_offset(bvec.bv_page) + bvec.bv_offset;
667 ordered = btrfs_lookup_ordered_extent(inode, offset);
668 BUG_ON(!ordered); /* Logic error */
671 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info,
672 bvec.bv_len + fs_info->sectorsize
675 for (i = 0; i < nr_sectors; i++) {
676 if (offset >= ordered->file_offset + ordered->num_bytes ||
677 offset < ordered->file_offset) {
678 unsigned long bytes_left;
680 sums->len = this_sum_bytes;
682 btrfs_add_ordered_sum(ordered, sums);
683 btrfs_put_ordered_extent(ordered);
685 bytes_left = bio->bi_iter.bi_size - total_bytes;
687 nofs_flag = memalloc_nofs_save();
688 sums = kvzalloc(btrfs_ordered_sum_size(fs_info,
689 bytes_left), GFP_KERNEL);
690 memalloc_nofs_restore(nofs_flag);
691 BUG_ON(!sums); /* -ENOMEM */
692 sums->len = bytes_left;
693 ordered = btrfs_lookup_ordered_extent(inode,
695 ASSERT(ordered); /* Logic error */
696 sums->bytenr = (bio->bi_iter.bi_sector << 9)
701 data = kmap_atomic(bvec.bv_page);
702 crypto_shash_digest(shash, data + bvec.bv_offset
703 + (i * fs_info->sectorsize),
707 index += fs_info->csum_size;
708 offset += fs_info->sectorsize;
709 this_sum_bytes += fs_info->sectorsize;
710 total_bytes += fs_info->sectorsize;
715 btrfs_add_ordered_sum(ordered, sums);
716 btrfs_put_ordered_extent(ordered);
721 * helper function for csum removal, this expects the
722 * key to describe the csum pointed to by the path, and it expects
723 * the csum to overlap the range [bytenr, len]
725 * The csum should not be entirely contained in the range and the
726 * range should not be entirely contained in the csum.
728 * This calls btrfs_truncate_item with the correct args based on the
729 * overlap, and fixes up the key as required.
731 static noinline void truncate_one_csum(struct btrfs_fs_info *fs_info,
732 struct btrfs_path *path,
733 struct btrfs_key *key,
736 struct extent_buffer *leaf;
737 const u32 csum_size = fs_info->csum_size;
739 u64 end_byte = bytenr + len;
740 u32 blocksize_bits = fs_info->sectorsize_bits;
742 leaf = path->nodes[0];
743 csum_end = btrfs_item_size_nr(leaf, path->slots[0]) / csum_size;
744 csum_end <<= blocksize_bits;
745 csum_end += key->offset;
747 if (key->offset < bytenr && csum_end <= end_byte) {
752 * A simple truncate off the end of the item
754 u32 new_size = (bytenr - key->offset) >> blocksize_bits;
755 new_size *= csum_size;
756 btrfs_truncate_item(path, new_size, 1);
757 } else if (key->offset >= bytenr && csum_end > end_byte &&
758 end_byte > key->offset) {
763 * we need to truncate from the beginning of the csum
765 u32 new_size = (csum_end - end_byte) >> blocksize_bits;
766 new_size *= csum_size;
768 btrfs_truncate_item(path, new_size, 0);
770 key->offset = end_byte;
771 btrfs_set_item_key_safe(fs_info, path, key);
778 * deletes the csum items from the csum tree for a given
781 int btrfs_del_csums(struct btrfs_trans_handle *trans,
782 struct btrfs_root *root, u64 bytenr, u64 len)
784 struct btrfs_fs_info *fs_info = trans->fs_info;
785 struct btrfs_path *path;
786 struct btrfs_key key;
787 u64 end_byte = bytenr + len;
789 struct extent_buffer *leaf;
791 const u32 csum_size = fs_info->csum_size;
792 u32 blocksize_bits = fs_info->sectorsize_bits;
794 ASSERT(root == fs_info->csum_root ||
795 root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID);
797 path = btrfs_alloc_path();
802 key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
803 key.offset = end_byte - 1;
804 key.type = BTRFS_EXTENT_CSUM_KEY;
806 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);
913 key.offset = end_byte - 1;
915 truncate_one_csum(fs_info, path, &key, bytenr, len);
916 if (key.offset < bytenr)
919 btrfs_release_path(path);
921 btrfs_free_path(path);
925 static int find_next_csum_offset(struct btrfs_root *root,
926 struct btrfs_path *path,
929 const u32 nritems = btrfs_header_nritems(path->nodes[0]);
930 struct btrfs_key found_key;
931 int slot = path->slots[0] + 1;
934 if (nritems == 0 || slot >= nritems) {
935 ret = btrfs_next_leaf(root, path);
938 } else if (ret > 0) {
939 *next_offset = (u64)-1;
942 slot = path->slots[0];
945 btrfs_item_key_to_cpu(path->nodes[0], &found_key, slot);
947 if (found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
948 found_key.type != BTRFS_EXTENT_CSUM_KEY)
949 *next_offset = (u64)-1;
951 *next_offset = found_key.offset;
956 int btrfs_csum_file_blocks(struct btrfs_trans_handle *trans,
957 struct btrfs_root *root,
958 struct btrfs_ordered_sum *sums)
960 struct btrfs_fs_info *fs_info = root->fs_info;
961 struct btrfs_key file_key;
962 struct btrfs_key found_key;
963 struct btrfs_path *path;
964 struct btrfs_csum_item *item;
965 struct btrfs_csum_item *item_end;
966 struct extent_buffer *leaf = NULL;
975 const u32 csum_size = fs_info->csum_size;
977 path = btrfs_alloc_path();
981 next_offset = (u64)-1;
983 bytenr = sums->bytenr + total_bytes;
984 file_key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
985 file_key.offset = bytenr;
986 file_key.type = BTRFS_EXTENT_CSUM_KEY;
988 item = btrfs_lookup_csum(trans, root, path, bytenr, 1);
991 leaf = path->nodes[0];
992 item_end = btrfs_item_ptr(leaf, path->slots[0],
993 struct btrfs_csum_item);
994 item_end = (struct btrfs_csum_item *)((char *)item_end +
995 btrfs_item_size_nr(leaf, path->slots[0]));
999 if (ret != -EFBIG && ret != -ENOENT)
1002 if (ret == -EFBIG) {
1004 /* we found one, but it isn't big enough yet */
1005 leaf = path->nodes[0];
1006 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1007 if ((item_size / csum_size) >=
1008 MAX_CSUM_ITEMS(fs_info, csum_size)) {
1009 /* already at max size, make a new one */
1013 /* We didn't find a csum item, insert one. */
1014 ret = find_next_csum_offset(root, path, &next_offset);
1022 * At this point, we know the tree has a checksum item that ends at an
1023 * offset matching the start of the checksum range we want to insert.
1024 * We try to extend that item as much as possible and then add as many
1025 * checksums to it as they fit.
1027 * First check if the leaf has enough free space for at least one
1028 * checksum. If it has go directly to the item extension code, otherwise
1029 * release the path and do a search for insertion before the extension.
1031 if (btrfs_leaf_free_space(leaf) >= csum_size) {
1032 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1033 csum_offset = (bytenr - found_key.offset) >>
1034 fs_info->sectorsize_bits;
1038 btrfs_release_path(path);
1039 path->search_for_extension = 1;
1040 ret = btrfs_search_slot(trans, root, &file_key, path,
1042 path->search_for_extension = 0;
1047 if (path->slots[0] == 0)
1052 leaf = path->nodes[0];
1053 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1054 csum_offset = (bytenr - found_key.offset) >> fs_info->sectorsize_bits;
1056 if (found_key.type != BTRFS_EXTENT_CSUM_KEY ||
1057 found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
1058 csum_offset >= MAX_CSUM_ITEMS(fs_info, csum_size)) {
1063 if (csum_offset == btrfs_item_size_nr(leaf, path->slots[0]) /
1069 tmp = sums->len - total_bytes;
1070 tmp >>= fs_info->sectorsize_bits;
1072 extend_nr = max_t(int, 1, tmp);
1075 * A log tree can already have checksum items with a subset of
1076 * the checksums we are trying to log. This can happen after
1077 * doing a sequence of partial writes into prealloc extents and
1078 * fsyncs in between, with a full fsync logging a larger subrange
1079 * of an extent for which a previous fast fsync logged a smaller
1080 * subrange. And this happens in particular due to merging file
1081 * extent items when we complete an ordered extent for a range
1082 * covered by a prealloc extent - this is done at
1083 * btrfs_mark_extent_written().
1085 * So if we try to extend the previous checksum item, which has
1086 * a range that ends at the start of the range we want to insert,
1087 * make sure we don't extend beyond the start offset of the next
1088 * checksum item. If we are at the last item in the leaf, then
1089 * forget the optimization of extending and add a new checksum
1090 * item - it is not worth the complexity of releasing the path,
1091 * getting the first key for the next leaf, repeat the btree
1092 * search, etc, because log trees are temporary anyway and it
1093 * would only save a few bytes of leaf space.
1095 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
1096 if (path->slots[0] + 1 >=
1097 btrfs_header_nritems(path->nodes[0])) {
1098 ret = find_next_csum_offset(root, path, &next_offset);
1105 ret = find_next_csum_offset(root, path, &next_offset);
1109 tmp = (next_offset - bytenr) >> fs_info->sectorsize_bits;
1111 extend_nr = min_t(int, extend_nr, tmp);
1114 diff = (csum_offset + extend_nr) * csum_size;
1116 MAX_CSUM_ITEMS(fs_info, csum_size) * csum_size);
1118 diff = diff - btrfs_item_size_nr(leaf, path->slots[0]);
1119 diff = min_t(u32, btrfs_leaf_free_space(leaf), diff);
1123 btrfs_extend_item(path, diff);
1129 btrfs_release_path(path);
1134 tmp = sums->len - total_bytes;
1135 tmp >>= fs_info->sectorsize_bits;
1136 tmp = min(tmp, (next_offset - file_key.offset) >>
1137 fs_info->sectorsize_bits);
1139 tmp = max_t(u64, 1, tmp);
1140 tmp = min_t(u64, tmp, MAX_CSUM_ITEMS(fs_info, csum_size));
1141 ins_size = csum_size * tmp;
1143 ins_size = csum_size;
1145 ret = btrfs_insert_empty_item(trans, root, path, &file_key,
1149 if (WARN_ON(ret != 0))
1151 leaf = path->nodes[0];
1153 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item);
1154 item_end = (struct btrfs_csum_item *)((unsigned char *)item +
1155 btrfs_item_size_nr(leaf, path->slots[0]));
1156 item = (struct btrfs_csum_item *)((unsigned char *)item +
1157 csum_offset * csum_size);
1159 ins_size = (u32)(sums->len - total_bytes) >> fs_info->sectorsize_bits;
1160 ins_size *= csum_size;
1161 ins_size = min_t(u32, (unsigned long)item_end - (unsigned long)item,
1163 write_extent_buffer(leaf, sums->sums + index, (unsigned long)item,
1167 ins_size /= csum_size;
1168 total_bytes += ins_size * fs_info->sectorsize;
1170 btrfs_mark_buffer_dirty(path->nodes[0]);
1171 if (total_bytes < sums->len) {
1172 btrfs_release_path(path);
1177 btrfs_free_path(path);
1181 void btrfs_extent_item_to_extent_map(struct btrfs_inode *inode,
1182 const struct btrfs_path *path,
1183 struct btrfs_file_extent_item *fi,
1184 const bool new_inline,
1185 struct extent_map *em)
1187 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1188 struct btrfs_root *root = inode->root;
1189 struct extent_buffer *leaf = path->nodes[0];
1190 const int slot = path->slots[0];
1191 struct btrfs_key key;
1192 u64 extent_start, extent_end;
1194 u8 type = btrfs_file_extent_type(leaf, fi);
1195 int compress_type = btrfs_file_extent_compression(leaf, fi);
1197 btrfs_item_key_to_cpu(leaf, &key, slot);
1198 extent_start = key.offset;
1199 extent_end = btrfs_file_extent_end(path);
1200 em->ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1201 if (type == BTRFS_FILE_EXTENT_REG ||
1202 type == BTRFS_FILE_EXTENT_PREALLOC) {
1203 em->start = extent_start;
1204 em->len = extent_end - extent_start;
1205 em->orig_start = extent_start -
1206 btrfs_file_extent_offset(leaf, fi);
1207 em->orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
1208 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1210 em->block_start = EXTENT_MAP_HOLE;
1213 if (compress_type != BTRFS_COMPRESS_NONE) {
1214 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
1215 em->compress_type = compress_type;
1216 em->block_start = bytenr;
1217 em->block_len = em->orig_block_len;
1219 bytenr += btrfs_file_extent_offset(leaf, fi);
1220 em->block_start = bytenr;
1221 em->block_len = em->len;
1222 if (type == BTRFS_FILE_EXTENT_PREALLOC)
1223 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
1225 } else if (type == BTRFS_FILE_EXTENT_INLINE) {
1226 em->block_start = EXTENT_MAP_INLINE;
1227 em->start = extent_start;
1228 em->len = extent_end - extent_start;
1230 * Initialize orig_start and block_len with the same values
1231 * as in inode.c:btrfs_get_extent().
1233 em->orig_start = EXTENT_MAP_HOLE;
1234 em->block_len = (u64)-1;
1235 if (!new_inline && compress_type != BTRFS_COMPRESS_NONE) {
1236 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
1237 em->compress_type = compress_type;
1241 "unknown file extent item type %d, inode %llu, offset %llu, "
1242 "root %llu", type, btrfs_ino(inode), extent_start,
1243 root->root_key.objectid);
1248 * Returns the end offset (non inclusive) of the file extent item the given path
1249 * points to. If it points to an inline extent, the returned offset is rounded
1250 * up to the sector size.
1252 u64 btrfs_file_extent_end(const struct btrfs_path *path)
1254 const struct extent_buffer *leaf = path->nodes[0];
1255 const int slot = path->slots[0];
1256 struct btrfs_file_extent_item *fi;
1257 struct btrfs_key key;
1260 btrfs_item_key_to_cpu(leaf, &key, slot);
1261 ASSERT(key.type == BTRFS_EXTENT_DATA_KEY);
1262 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
1264 if (btrfs_file_extent_type(leaf, fi) == BTRFS_FILE_EXTENT_INLINE) {
1265 end = btrfs_file_extent_ram_bytes(leaf, fi);
1266 end = ALIGN(key.offset + end, leaf->fs_info->sectorsize);
1268 end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);