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 spin_lock(&inode->lock);
51 i_size = new_i_size ?: i_size_read(&inode->vfs_inode);
52 if (btrfs_fs_incompat(fs_info, NO_HOLES)) {
53 inode->disk_i_size = i_size;
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;
65 spin_unlock(&inode->lock);
69 * Mark range within a file as having a new extent inserted
71 * @inode: inode being modified
72 * @start: start file offset of the file extent we've inserted
73 * @len: logical length of the file extent item
75 * Call when we are inserting a new file extent where there was none before.
76 * Does not need to call this in the case where we're replacing an existing file
77 * extent, however if not sure it's fine to call this multiple times.
79 * The start and len must match the file extent item, so thus must be sectorsize
82 int btrfs_inode_set_file_extent_range(struct btrfs_inode *inode, u64 start,
88 ASSERT(IS_ALIGNED(start + len, inode->root->fs_info->sectorsize));
90 if (btrfs_fs_incompat(inode->root->fs_info, NO_HOLES))
92 return set_extent_bits(&inode->file_extent_tree, start, start + len - 1,
97 * Marks an inode range as not having a backing extent
99 * @inode: inode being modified
100 * @start: start file offset of the file extent we've inserted
101 * @len: logical length of the file extent item
103 * Called when we drop a file extent, for example when we truncate. Doesn't
104 * need to be called for cases where we're replacing a file extent, like when
105 * we've COWed a file extent.
107 * The start and len must match the file extent item, so thus must be sectorsize
110 int btrfs_inode_clear_file_extent_range(struct btrfs_inode *inode, u64 start,
116 ASSERT(IS_ALIGNED(start + len, inode->root->fs_info->sectorsize) ||
119 if (btrfs_fs_incompat(inode->root->fs_info, NO_HOLES))
121 return clear_extent_bit(&inode->file_extent_tree, start,
122 start + len - 1, EXTENT_DIRTY, NULL);
125 static inline u32 max_ordered_sum_bytes(struct btrfs_fs_info *fs_info,
128 u32 ncsums = (PAGE_SIZE - sizeof(struct btrfs_ordered_sum)) / csum_size;
130 return ncsums * fs_info->sectorsize;
134 * Calculate the total size needed to allocate for an ordered sum structure
135 * spanning @bytes in the file.
137 static int btrfs_ordered_sum_size(struct btrfs_fs_info *fs_info, unsigned long bytes)
139 int num_sectors = (int)DIV_ROUND_UP(bytes, fs_info->sectorsize);
141 return sizeof(struct btrfs_ordered_sum) + num_sectors * fs_info->csum_size;
144 int btrfs_insert_hole_extent(struct btrfs_trans_handle *trans,
145 struct btrfs_root *root,
146 u64 objectid, u64 pos, u64 num_bytes)
149 struct btrfs_file_extent_item *item;
150 struct btrfs_key file_key;
151 struct btrfs_path *path;
152 struct extent_buffer *leaf;
154 path = btrfs_alloc_path();
157 file_key.objectid = objectid;
158 file_key.offset = pos;
159 file_key.type = BTRFS_EXTENT_DATA_KEY;
161 ret = btrfs_insert_empty_item(trans, root, path, &file_key,
165 BUG_ON(ret); /* Can't happen */
166 leaf = path->nodes[0];
167 item = btrfs_item_ptr(leaf, path->slots[0],
168 struct btrfs_file_extent_item);
169 btrfs_set_file_extent_disk_bytenr(leaf, item, 0);
170 btrfs_set_file_extent_disk_num_bytes(leaf, item, 0);
171 btrfs_set_file_extent_offset(leaf, item, 0);
172 btrfs_set_file_extent_num_bytes(leaf, item, num_bytes);
173 btrfs_set_file_extent_ram_bytes(leaf, item, num_bytes);
174 btrfs_set_file_extent_generation(leaf, item, trans->transid);
175 btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
176 btrfs_set_file_extent_compression(leaf, item, 0);
177 btrfs_set_file_extent_encryption(leaf, item, 0);
178 btrfs_set_file_extent_other_encoding(leaf, item, 0);
180 btrfs_mark_buffer_dirty(leaf);
182 btrfs_free_path(path);
186 static struct btrfs_csum_item *
187 btrfs_lookup_csum(struct btrfs_trans_handle *trans,
188 struct btrfs_root *root,
189 struct btrfs_path *path,
192 struct btrfs_fs_info *fs_info = root->fs_info;
194 struct btrfs_key file_key;
195 struct btrfs_key found_key;
196 struct btrfs_csum_item *item;
197 struct extent_buffer *leaf;
199 const u32 csum_size = fs_info->csum_size;
202 file_key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
203 file_key.offset = bytenr;
204 file_key.type = BTRFS_EXTENT_CSUM_KEY;
205 ret = btrfs_search_slot(trans, root, &file_key, path, 0, cow);
208 leaf = path->nodes[0];
211 if (path->slots[0] == 0)
214 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
215 if (found_key.type != BTRFS_EXTENT_CSUM_KEY)
218 csum_offset = (bytenr - found_key.offset) >>
219 fs_info->sectorsize_bits;
220 csums_in_item = btrfs_item_size(leaf, path->slots[0]);
221 csums_in_item /= csum_size;
223 if (csum_offset == csums_in_item) {
226 } else if (csum_offset > csums_in_item) {
230 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item);
231 item = (struct btrfs_csum_item *)((unsigned char *)item +
232 csum_offset * csum_size);
240 int btrfs_lookup_file_extent(struct btrfs_trans_handle *trans,
241 struct btrfs_root *root,
242 struct btrfs_path *path, u64 objectid,
245 struct btrfs_key file_key;
246 int ins_len = mod < 0 ? -1 : 0;
249 file_key.objectid = objectid;
250 file_key.offset = offset;
251 file_key.type = BTRFS_EXTENT_DATA_KEY;
253 return btrfs_search_slot(trans, root, &file_key, path, ins_len, cow);
257 * Find checksums for logical bytenr range [disk_bytenr, disk_bytenr + len) and
258 * estore the result to @dst.
260 * Return >0 for the number of sectors we found.
261 * Return 0 for the range [disk_bytenr, disk_bytenr + sectorsize) has no csum
262 * for it. Caller may want to try next sector until one range is hit.
263 * Return <0 for fatal error.
265 static int search_csum_tree(struct btrfs_fs_info *fs_info,
266 struct btrfs_path *path, u64 disk_bytenr,
269 struct btrfs_root *csum_root;
270 struct btrfs_csum_item *item = NULL;
271 struct btrfs_key key;
272 const u32 sectorsize = fs_info->sectorsize;
273 const u32 csum_size = fs_info->csum_size;
279 ASSERT(IS_ALIGNED(disk_bytenr, sectorsize) &&
280 IS_ALIGNED(len, sectorsize));
282 /* Check if the current csum item covers disk_bytenr */
283 if (path->nodes[0]) {
284 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
285 struct btrfs_csum_item);
286 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
287 itemsize = btrfs_item_size(path->nodes[0], path->slots[0]);
289 csum_start = key.offset;
290 csum_len = (itemsize / csum_size) * sectorsize;
292 if (in_range(disk_bytenr, csum_start, csum_len))
296 /* Current item doesn't contain the desired range, search again */
297 btrfs_release_path(path);
298 csum_root = btrfs_csum_root(fs_info, disk_bytenr);
299 item = btrfs_lookup_csum(NULL, csum_root, path, disk_bytenr, 0);
304 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
305 itemsize = btrfs_item_size(path->nodes[0], path->slots[0]);
307 csum_start = key.offset;
308 csum_len = (itemsize / csum_size) * sectorsize;
309 ASSERT(in_range(disk_bytenr, csum_start, csum_len));
312 ret = (min(csum_start + csum_len, disk_bytenr + len) -
313 disk_bytenr) >> fs_info->sectorsize_bits;
314 read_extent_buffer(path->nodes[0], dst, (unsigned long)item,
317 if (ret == -ENOENT || ret == -EFBIG)
323 * Locate the file_offset of @cur_disk_bytenr of a @bio.
325 * Bio of btrfs represents read range of
326 * [bi_sector << 9, bi_sector << 9 + bi_size).
327 * Knowing this, we can iterate through each bvec to locate the page belong to
328 * @cur_disk_bytenr and get the file offset.
330 * @inode is used to determine if the bvec page really belongs to @inode.
332 * Return 0 if we can't find the file offset
333 * Return >0 if we find the file offset and restore it to @file_offset_ret
335 static int search_file_offset_in_bio(struct bio *bio, struct inode *inode,
336 u64 disk_bytenr, u64 *file_offset_ret)
338 struct bvec_iter iter;
340 u64 cur = bio->bi_iter.bi_sector << SECTOR_SHIFT;
343 bio_for_each_segment(bvec, bio, iter) {
344 struct page *page = bvec.bv_page;
346 if (cur > disk_bytenr)
348 if (cur + bvec.bv_len <= disk_bytenr) {
352 ASSERT(in_range(disk_bytenr, cur, bvec.bv_len));
353 if (page->mapping && page->mapping->host &&
354 page->mapping->host == inode) {
356 *file_offset_ret = page_offset(page) + bvec.bv_offset +
365 * Lookup the checksum for the read bio in csum tree.
367 * @inode: inode that the bio is for.
368 * @bio: bio to look up.
369 * @dst: Buffer of size nblocks * btrfs_super_csum_size() used to return
370 * checksum (nblocks = bio->bi_iter.bi_size / fs_info->sectorsize). If
371 * NULL, the checksum buffer is allocated and returned in
372 * btrfs_bio(bio)->csum instead.
374 * Return: BLK_STS_RESOURCE if allocating memory fails, BLK_STS_OK otherwise.
376 blk_status_t btrfs_lookup_bio_sums(struct inode *inode, struct bio *bio, u8 *dst)
378 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
379 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
380 struct btrfs_bio *bbio = NULL;
381 struct btrfs_path *path;
382 const u32 sectorsize = fs_info->sectorsize;
383 const u32 csum_size = fs_info->csum_size;
384 u32 orig_len = bio->bi_iter.bi_size;
385 u64 orig_disk_bytenr = bio->bi_iter.bi_sector << SECTOR_SHIFT;
388 const unsigned int nblocks = orig_len >> fs_info->sectorsize_bits;
390 blk_status_t ret = BLK_STS_OK;
392 if ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM) ||
393 test_bit(BTRFS_FS_STATE_NO_CSUMS, &fs_info->fs_state))
397 * This function is only called for read bio.
399 * This means two things:
400 * - All our csums should only be in csum tree
401 * No ordered extents csums, as ordered extents are only for write
403 * - No need to bother any other info from bvec
404 * Since we're looking up csums, the only important info is the
405 * disk_bytenr and the length, which can be extracted from bi_iter
408 ASSERT(bio_op(bio) == REQ_OP_READ);
409 path = btrfs_alloc_path();
411 return BLK_STS_RESOURCE;
414 bbio = btrfs_bio(bio);
416 if (nblocks * csum_size > BTRFS_BIO_INLINE_CSUM_SIZE) {
417 bbio->csum = kmalloc_array(nblocks, csum_size, GFP_NOFS);
419 btrfs_free_path(path);
420 return BLK_STS_RESOURCE;
423 bbio->csum = bbio->csum_inline;
431 * If requested number of sectors is larger than one leaf can contain,
432 * kick the readahead for csum tree.
434 if (nblocks > fs_info->csums_per_leaf)
435 path->reada = READA_FORWARD;
438 * the free space stuff is only read when it hasn't been
439 * updated in the current transaction. So, we can safely
440 * read from the commit root and sidestep a nasty deadlock
441 * between reading the free space cache and updating the csum tree.
443 if (btrfs_is_free_space_inode(BTRFS_I(inode))) {
444 path->search_commit_root = 1;
445 path->skip_locking = 1;
448 for (cur_disk_bytenr = orig_disk_bytenr;
449 cur_disk_bytenr < orig_disk_bytenr + orig_len;
450 cur_disk_bytenr += (count * sectorsize)) {
451 u64 search_len = orig_disk_bytenr + orig_len - cur_disk_bytenr;
452 unsigned int sector_offset;
456 * Although both cur_disk_bytenr and orig_disk_bytenr is u64,
457 * we're calculating the offset to the bio start.
459 * Bio size is limited to UINT_MAX, thus unsigned int is large
460 * enough to contain the raw result, not to mention the right
463 ASSERT(cur_disk_bytenr - orig_disk_bytenr < UINT_MAX);
464 sector_offset = (cur_disk_bytenr - orig_disk_bytenr) >>
465 fs_info->sectorsize_bits;
466 csum_dst = csum + sector_offset * csum_size;
468 count = search_csum_tree(fs_info, path, cur_disk_bytenr,
469 search_len, csum_dst);
471 ret = errno_to_blk_status(count);
473 btrfs_bio_free_csum(bbio);
478 * We didn't find a csum for this range. We need to make sure
479 * we complain loudly about this, because we are not NODATASUM.
481 * However for the DATA_RELOC inode we could potentially be
482 * relocating data extents for a NODATASUM inode, so the inode
483 * itself won't be marked with NODATASUM, but the extent we're
484 * copying is in fact NODATASUM. If we don't find a csum we
485 * assume this is the case.
488 memset(csum_dst, 0, csum_size);
491 if (BTRFS_I(inode)->root->root_key.objectid ==
492 BTRFS_DATA_RELOC_TREE_OBJECTID) {
496 ret = search_file_offset_in_bio(bio, inode,
497 cur_disk_bytenr, &file_offset);
499 set_extent_bits(io_tree, file_offset,
500 file_offset + sectorsize - 1,
503 btrfs_warn_rl(fs_info,
504 "csum hole found for disk bytenr range [%llu, %llu)",
505 cur_disk_bytenr, cur_disk_bytenr + sectorsize);
510 btrfs_free_path(path);
514 int btrfs_lookup_csums_range(struct btrfs_root *root, u64 start, u64 end,
515 struct list_head *list, int search_commit,
518 struct btrfs_fs_info *fs_info = root->fs_info;
519 struct btrfs_key key;
520 struct btrfs_path *path;
521 struct extent_buffer *leaf;
522 struct btrfs_ordered_sum *sums;
523 struct btrfs_csum_item *item;
525 unsigned long offset;
529 const u32 csum_size = fs_info->csum_size;
531 ASSERT(IS_ALIGNED(start, fs_info->sectorsize) &&
532 IS_ALIGNED(end + 1, fs_info->sectorsize));
534 path = btrfs_alloc_path();
538 path->nowait = nowait;
540 path->skip_locking = 1;
541 path->reada = READA_FORWARD;
542 path->search_commit_root = 1;
545 key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
547 key.type = BTRFS_EXTENT_CSUM_KEY;
549 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
552 if (ret > 0 && path->slots[0] > 0) {
553 leaf = path->nodes[0];
554 btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
555 if (key.objectid == BTRFS_EXTENT_CSUM_OBJECTID &&
556 key.type == BTRFS_EXTENT_CSUM_KEY) {
557 offset = (start - key.offset) >> fs_info->sectorsize_bits;
558 if (offset * csum_size <
559 btrfs_item_size(leaf, path->slots[0] - 1))
564 while (start <= end) {
565 leaf = path->nodes[0];
566 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
567 ret = btrfs_next_leaf(root, path);
572 leaf = path->nodes[0];
575 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
576 if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
577 key.type != BTRFS_EXTENT_CSUM_KEY ||
581 if (key.offset > start)
584 size = btrfs_item_size(leaf, path->slots[0]);
585 csum_end = key.offset + (size / csum_size) * fs_info->sectorsize;
586 if (csum_end <= start) {
591 csum_end = min(csum_end, end + 1);
592 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
593 struct btrfs_csum_item);
594 while (start < csum_end) {
595 size = min_t(size_t, csum_end - start,
596 max_ordered_sum_bytes(fs_info, csum_size));
597 sums = kzalloc(btrfs_ordered_sum_size(fs_info, size),
604 sums->bytenr = start;
607 offset = (start - key.offset) >> fs_info->sectorsize_bits;
609 size >>= fs_info->sectorsize_bits;
611 read_extent_buffer(path->nodes[0],
613 ((unsigned long)item) + offset,
616 start += fs_info->sectorsize * size;
617 list_add_tail(&sums->list, &tmplist);
623 while (ret < 0 && !list_empty(&tmplist)) {
624 sums = list_entry(tmplist.next, struct btrfs_ordered_sum, list);
625 list_del(&sums->list);
628 list_splice_tail(&tmplist, list);
630 btrfs_free_path(path);
635 * Calculate checksums of the data contained inside a bio
637 * @inode: Owner of the data inside the bio
638 * @bio: Contains the data to be checksummed
639 * @offset: If (u64)-1, @bio may contain discontiguous bio vecs, so the
640 * file offsets are determined from the page offsets in the bio.
641 * Otherwise, this is the starting file offset of the bio vecs in
642 * @bio, which must be contiguous.
643 * @one_ordered: If true, @bio only refers to one ordered extent.
645 blk_status_t btrfs_csum_one_bio(struct btrfs_inode *inode, struct bio *bio,
646 u64 offset, bool one_ordered)
648 struct btrfs_fs_info *fs_info = inode->root->fs_info;
649 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
650 struct btrfs_ordered_sum *sums;
651 struct btrfs_ordered_extent *ordered = NULL;
652 const bool use_page_offsets = (offset == (u64)-1);
654 struct bvec_iter iter;
657 unsigned int blockcount;
658 unsigned long total_bytes = 0;
659 unsigned long this_sum_bytes = 0;
663 nofs_flag = memalloc_nofs_save();
664 sums = kvzalloc(btrfs_ordered_sum_size(fs_info, bio->bi_iter.bi_size),
666 memalloc_nofs_restore(nofs_flag);
669 return BLK_STS_RESOURCE;
671 sums->len = bio->bi_iter.bi_size;
672 INIT_LIST_HEAD(&sums->list);
674 sums->bytenr = bio->bi_iter.bi_sector << 9;
677 shash->tfm = fs_info->csum_shash;
679 bio_for_each_segment(bvec, bio, iter) {
680 if (use_page_offsets)
681 offset = page_offset(bvec.bv_page) + bvec.bv_offset;
684 ordered = btrfs_lookup_ordered_extent(inode, offset);
686 * The bio range is not covered by any ordered extent,
687 * must be a code logic error.
689 if (unlikely(!ordered)) {
691 "no ordered extent for root %llu ino %llu offset %llu\n",
692 inode->root->root_key.objectid,
693 btrfs_ino(inode), offset);
695 return BLK_STS_IOERR;
699 blockcount = BTRFS_BYTES_TO_BLKS(fs_info,
700 bvec.bv_len + fs_info->sectorsize
703 for (i = 0; i < blockcount; i++) {
705 !in_range(offset, ordered->file_offset,
706 ordered->num_bytes)) {
707 unsigned long bytes_left;
709 sums->len = this_sum_bytes;
711 btrfs_add_ordered_sum(ordered, sums);
712 btrfs_put_ordered_extent(ordered);
714 bytes_left = bio->bi_iter.bi_size - total_bytes;
716 nofs_flag = memalloc_nofs_save();
717 sums = kvzalloc(btrfs_ordered_sum_size(fs_info,
718 bytes_left), GFP_KERNEL);
719 memalloc_nofs_restore(nofs_flag);
721 return BLK_STS_RESOURCE;
723 sums->len = bytes_left;
724 ordered = btrfs_lookup_ordered_extent(inode,
726 ASSERT(ordered); /* Logic error */
727 sums->bytenr = (bio->bi_iter.bi_sector << 9)
732 data = bvec_kmap_local(&bvec);
733 crypto_shash_digest(shash,
734 data + (i * fs_info->sectorsize),
738 index += fs_info->csum_size;
739 offset += fs_info->sectorsize;
740 this_sum_bytes += fs_info->sectorsize;
741 total_bytes += fs_info->sectorsize;
746 btrfs_add_ordered_sum(ordered, sums);
747 btrfs_put_ordered_extent(ordered);
752 * helper function for csum removal, this expects the
753 * key to describe the csum pointed to by the path, and it expects
754 * the csum to overlap the range [bytenr, len]
756 * The csum should not be entirely contained in the range and the
757 * range should not be entirely contained in the csum.
759 * This calls btrfs_truncate_item with the correct args based on the
760 * overlap, and fixes up the key as required.
762 static noinline void truncate_one_csum(struct btrfs_fs_info *fs_info,
763 struct btrfs_path *path,
764 struct btrfs_key *key,
767 struct extent_buffer *leaf;
768 const u32 csum_size = fs_info->csum_size;
770 u64 end_byte = bytenr + len;
771 u32 blocksize_bits = fs_info->sectorsize_bits;
773 leaf = path->nodes[0];
774 csum_end = btrfs_item_size(leaf, path->slots[0]) / csum_size;
775 csum_end <<= blocksize_bits;
776 csum_end += key->offset;
778 if (key->offset < bytenr && csum_end <= end_byte) {
783 * A simple truncate off the end of the item
785 u32 new_size = (bytenr - key->offset) >> blocksize_bits;
786 new_size *= csum_size;
787 btrfs_truncate_item(path, new_size, 1);
788 } else if (key->offset >= bytenr && csum_end > end_byte &&
789 end_byte > key->offset) {
794 * we need to truncate from the beginning of the csum
796 u32 new_size = (csum_end - end_byte) >> blocksize_bits;
797 new_size *= csum_size;
799 btrfs_truncate_item(path, new_size, 0);
801 key->offset = end_byte;
802 btrfs_set_item_key_safe(fs_info, path, key);
809 * deletes the csum items from the csum tree for a given
812 int btrfs_del_csums(struct btrfs_trans_handle *trans,
813 struct btrfs_root *root, u64 bytenr, u64 len)
815 struct btrfs_fs_info *fs_info = trans->fs_info;
816 struct btrfs_path *path;
817 struct btrfs_key key;
818 u64 end_byte = bytenr + len;
820 struct extent_buffer *leaf;
822 const u32 csum_size = fs_info->csum_size;
823 u32 blocksize_bits = fs_info->sectorsize_bits;
825 ASSERT(root->root_key.objectid == BTRFS_CSUM_TREE_OBJECTID ||
826 root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID);
828 path = btrfs_alloc_path();
833 key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
834 key.offset = end_byte - 1;
835 key.type = BTRFS_EXTENT_CSUM_KEY;
837 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
840 if (path->slots[0] == 0)
843 } else if (ret < 0) {
847 leaf = path->nodes[0];
848 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
850 if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
851 key.type != BTRFS_EXTENT_CSUM_KEY) {
855 if (key.offset >= end_byte)
858 csum_end = btrfs_item_size(leaf, path->slots[0]) / csum_size;
859 csum_end <<= blocksize_bits;
860 csum_end += key.offset;
862 /* this csum ends before we start, we're done */
863 if (csum_end <= bytenr)
866 /* delete the entire item, it is inside our range */
867 if (key.offset >= bytenr && csum_end <= end_byte) {
871 * Check how many csum items preceding this one in this
872 * leaf correspond to our range and then delete them all
875 if (key.offset > bytenr && path->slots[0] > 0) {
876 int slot = path->slots[0] - 1;
881 btrfs_item_key_to_cpu(leaf, &pk, slot);
882 if (pk.offset < bytenr ||
883 pk.type != BTRFS_EXTENT_CSUM_KEY ||
885 BTRFS_EXTENT_CSUM_OBJECTID)
887 path->slots[0] = slot;
889 key.offset = pk.offset;
893 ret = btrfs_del_items(trans, root, path,
894 path->slots[0], del_nr);
897 if (key.offset == bytenr)
899 } else if (key.offset < bytenr && csum_end > end_byte) {
900 unsigned long offset;
901 unsigned long shift_len;
902 unsigned long item_offset;
907 * Our bytes are in the middle of the csum,
908 * we need to split this item and insert a new one.
910 * But we can't drop the path because the
911 * csum could change, get removed, extended etc.
913 * The trick here is the max size of a csum item leaves
914 * enough room in the tree block for a single
915 * item header. So, we split the item in place,
916 * adding a new header pointing to the existing
917 * bytes. Then we loop around again and we have
918 * a nicely formed csum item that we can neatly
921 offset = (bytenr - key.offset) >> blocksize_bits;
924 shift_len = (len >> blocksize_bits) * csum_size;
926 item_offset = btrfs_item_ptr_offset(leaf,
929 memzero_extent_buffer(leaf, item_offset + offset,
934 * btrfs_split_item returns -EAGAIN when the
935 * item changed size or key
937 ret = btrfs_split_item(trans, root, path, &key, offset);
938 if (ret && ret != -EAGAIN) {
939 btrfs_abort_transaction(trans, ret);
944 key.offset = end_byte - 1;
946 truncate_one_csum(fs_info, path, &key, bytenr, len);
947 if (key.offset < bytenr)
950 btrfs_release_path(path);
952 btrfs_free_path(path);
956 static int find_next_csum_offset(struct btrfs_root *root,
957 struct btrfs_path *path,
960 const u32 nritems = btrfs_header_nritems(path->nodes[0]);
961 struct btrfs_key found_key;
962 int slot = path->slots[0] + 1;
965 if (nritems == 0 || slot >= nritems) {
966 ret = btrfs_next_leaf(root, path);
969 } else if (ret > 0) {
970 *next_offset = (u64)-1;
973 slot = path->slots[0];
976 btrfs_item_key_to_cpu(path->nodes[0], &found_key, slot);
978 if (found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
979 found_key.type != BTRFS_EXTENT_CSUM_KEY)
980 *next_offset = (u64)-1;
982 *next_offset = found_key.offset;
987 int btrfs_csum_file_blocks(struct btrfs_trans_handle *trans,
988 struct btrfs_root *root,
989 struct btrfs_ordered_sum *sums)
991 struct btrfs_fs_info *fs_info = root->fs_info;
992 struct btrfs_key file_key;
993 struct btrfs_key found_key;
994 struct btrfs_path *path;
995 struct btrfs_csum_item *item;
996 struct btrfs_csum_item *item_end;
997 struct extent_buffer *leaf = NULL;
1006 const u32 csum_size = fs_info->csum_size;
1008 path = btrfs_alloc_path();
1012 next_offset = (u64)-1;
1014 bytenr = sums->bytenr + total_bytes;
1015 file_key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
1016 file_key.offset = bytenr;
1017 file_key.type = BTRFS_EXTENT_CSUM_KEY;
1019 item = btrfs_lookup_csum(trans, root, path, bytenr, 1);
1020 if (!IS_ERR(item)) {
1022 leaf = path->nodes[0];
1023 item_end = btrfs_item_ptr(leaf, path->slots[0],
1024 struct btrfs_csum_item);
1025 item_end = (struct btrfs_csum_item *)((char *)item_end +
1026 btrfs_item_size(leaf, path->slots[0]));
1029 ret = PTR_ERR(item);
1030 if (ret != -EFBIG && ret != -ENOENT)
1033 if (ret == -EFBIG) {
1035 /* we found one, but it isn't big enough yet */
1036 leaf = path->nodes[0];
1037 item_size = btrfs_item_size(leaf, path->slots[0]);
1038 if ((item_size / csum_size) >=
1039 MAX_CSUM_ITEMS(fs_info, csum_size)) {
1040 /* already at max size, make a new one */
1044 /* We didn't find a csum item, insert one. */
1045 ret = find_next_csum_offset(root, path, &next_offset);
1053 * At this point, we know the tree has a checksum item that ends at an
1054 * offset matching the start of the checksum range we want to insert.
1055 * We try to extend that item as much as possible and then add as many
1056 * checksums to it as they fit.
1058 * First check if the leaf has enough free space for at least one
1059 * checksum. If it has go directly to the item extension code, otherwise
1060 * release the path and do a search for insertion before the extension.
1062 if (btrfs_leaf_free_space(leaf) >= csum_size) {
1063 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1064 csum_offset = (bytenr - found_key.offset) >>
1065 fs_info->sectorsize_bits;
1069 btrfs_release_path(path);
1070 path->search_for_extension = 1;
1071 ret = btrfs_search_slot(trans, root, &file_key, path,
1073 path->search_for_extension = 0;
1078 if (path->slots[0] == 0)
1083 leaf = path->nodes[0];
1084 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1085 csum_offset = (bytenr - found_key.offset) >> fs_info->sectorsize_bits;
1087 if (found_key.type != BTRFS_EXTENT_CSUM_KEY ||
1088 found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
1089 csum_offset >= MAX_CSUM_ITEMS(fs_info, csum_size)) {
1094 if (csum_offset == btrfs_item_size(leaf, path->slots[0]) /
1100 tmp = sums->len - total_bytes;
1101 tmp >>= fs_info->sectorsize_bits;
1103 extend_nr = max_t(int, 1, tmp);
1106 * A log tree can already have checksum items with a subset of
1107 * the checksums we are trying to log. This can happen after
1108 * doing a sequence of partial writes into prealloc extents and
1109 * fsyncs in between, with a full fsync logging a larger subrange
1110 * of an extent for which a previous fast fsync logged a smaller
1111 * subrange. And this happens in particular due to merging file
1112 * extent items when we complete an ordered extent for a range
1113 * covered by a prealloc extent - this is done at
1114 * btrfs_mark_extent_written().
1116 * So if we try to extend the previous checksum item, which has
1117 * a range that ends at the start of the range we want to insert,
1118 * make sure we don't extend beyond the start offset of the next
1119 * checksum item. If we are at the last item in the leaf, then
1120 * forget the optimization of extending and add a new checksum
1121 * item - it is not worth the complexity of releasing the path,
1122 * getting the first key for the next leaf, repeat the btree
1123 * search, etc, because log trees are temporary anyway and it
1124 * would only save a few bytes of leaf space.
1126 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
1127 if (path->slots[0] + 1 >=
1128 btrfs_header_nritems(path->nodes[0])) {
1129 ret = find_next_csum_offset(root, path, &next_offset);
1136 ret = find_next_csum_offset(root, path, &next_offset);
1140 tmp = (next_offset - bytenr) >> fs_info->sectorsize_bits;
1142 extend_nr = min_t(int, extend_nr, tmp);
1145 diff = (csum_offset + extend_nr) * csum_size;
1147 MAX_CSUM_ITEMS(fs_info, csum_size) * csum_size);
1149 diff = diff - btrfs_item_size(leaf, path->slots[0]);
1150 diff = min_t(u32, btrfs_leaf_free_space(leaf), diff);
1154 btrfs_extend_item(path, diff);
1160 btrfs_release_path(path);
1165 tmp = sums->len - total_bytes;
1166 tmp >>= fs_info->sectorsize_bits;
1167 tmp = min(tmp, (next_offset - file_key.offset) >>
1168 fs_info->sectorsize_bits);
1170 tmp = max_t(u64, 1, tmp);
1171 tmp = min_t(u64, tmp, MAX_CSUM_ITEMS(fs_info, csum_size));
1172 ins_size = csum_size * tmp;
1174 ins_size = csum_size;
1176 ret = btrfs_insert_empty_item(trans, root, path, &file_key,
1180 if (WARN_ON(ret != 0))
1182 leaf = path->nodes[0];
1184 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item);
1185 item_end = (struct btrfs_csum_item *)((unsigned char *)item +
1186 btrfs_item_size(leaf, path->slots[0]));
1187 item = (struct btrfs_csum_item *)((unsigned char *)item +
1188 csum_offset * csum_size);
1190 ins_size = (u32)(sums->len - total_bytes) >> fs_info->sectorsize_bits;
1191 ins_size *= csum_size;
1192 ins_size = min_t(u32, (unsigned long)item_end - (unsigned long)item,
1194 write_extent_buffer(leaf, sums->sums + index, (unsigned long)item,
1198 ins_size /= csum_size;
1199 total_bytes += ins_size * fs_info->sectorsize;
1201 btrfs_mark_buffer_dirty(path->nodes[0]);
1202 if (total_bytes < sums->len) {
1203 btrfs_release_path(path);
1208 btrfs_free_path(path);
1212 void btrfs_extent_item_to_extent_map(struct btrfs_inode *inode,
1213 const struct btrfs_path *path,
1214 struct btrfs_file_extent_item *fi,
1215 const bool new_inline,
1216 struct extent_map *em)
1218 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1219 struct btrfs_root *root = inode->root;
1220 struct extent_buffer *leaf = path->nodes[0];
1221 const int slot = path->slots[0];
1222 struct btrfs_key key;
1223 u64 extent_start, extent_end;
1225 u8 type = btrfs_file_extent_type(leaf, fi);
1226 int compress_type = btrfs_file_extent_compression(leaf, fi);
1228 btrfs_item_key_to_cpu(leaf, &key, slot);
1229 extent_start = key.offset;
1230 extent_end = btrfs_file_extent_end(path);
1231 em->ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1232 em->generation = btrfs_file_extent_generation(leaf, fi);
1233 if (type == BTRFS_FILE_EXTENT_REG ||
1234 type == BTRFS_FILE_EXTENT_PREALLOC) {
1235 em->start = extent_start;
1236 em->len = extent_end - extent_start;
1237 em->orig_start = extent_start -
1238 btrfs_file_extent_offset(leaf, fi);
1239 em->orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
1240 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1242 em->block_start = EXTENT_MAP_HOLE;
1245 if (compress_type != BTRFS_COMPRESS_NONE) {
1246 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
1247 em->compress_type = compress_type;
1248 em->block_start = bytenr;
1249 em->block_len = em->orig_block_len;
1251 bytenr += btrfs_file_extent_offset(leaf, fi);
1252 em->block_start = bytenr;
1253 em->block_len = em->len;
1254 if (type == BTRFS_FILE_EXTENT_PREALLOC)
1255 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
1257 } else if (type == BTRFS_FILE_EXTENT_INLINE) {
1258 em->block_start = EXTENT_MAP_INLINE;
1259 em->start = extent_start;
1260 em->len = extent_end - extent_start;
1262 * Initialize orig_start and block_len with the same values
1263 * as in inode.c:btrfs_get_extent().
1265 em->orig_start = EXTENT_MAP_HOLE;
1266 em->block_len = (u64)-1;
1267 if (!new_inline && compress_type != BTRFS_COMPRESS_NONE) {
1268 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
1269 em->compress_type = compress_type;
1273 "unknown file extent item type %d, inode %llu, offset %llu, "
1274 "root %llu", type, btrfs_ino(inode), extent_start,
1275 root->root_key.objectid);
1280 * Returns the end offset (non inclusive) of the file extent item the given path
1281 * points to. If it points to an inline extent, the returned offset is rounded
1282 * up to the sector size.
1284 u64 btrfs_file_extent_end(const struct btrfs_path *path)
1286 const struct extent_buffer *leaf = path->nodes[0];
1287 const int slot = path->slots[0];
1288 struct btrfs_file_extent_item *fi;
1289 struct btrfs_key key;
1292 btrfs_item_key_to_cpu(leaf, &key, slot);
1293 ASSERT(key.type == BTRFS_EXTENT_DATA_KEY);
1294 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
1296 if (btrfs_file_extent_type(leaf, fi) == BTRFS_FILE_EXTENT_INLINE) {
1297 end = btrfs_file_extent_ram_bytes(leaf, fi);
1298 end = ALIGN(key.offset + end, leaf->fs_info->sectorsize);
1300 end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);