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 "compression.h"
19 #include "accessors.h"
20 #include "file-item.h"
22 #define __MAX_CSUM_ITEMS(r, size) ((unsigned long)(((BTRFS_LEAF_DATA_SIZE(r) - \
23 sizeof(struct btrfs_item) * 2) / \
26 #define MAX_CSUM_ITEMS(r, size) (min_t(u32, __MAX_CSUM_ITEMS(r, size), \
30 * Set inode's size according to filesystem options.
32 * @inode: inode we want to update the disk_i_size for
33 * @new_i_size: i_size we want to set to, 0 if we use i_size
35 * With NO_HOLES set this simply sets the disk_is_size to whatever i_size_read()
36 * returns as it is perfectly fine with a file that has holes without hole file
39 * However without NO_HOLES we need to only return the area that is contiguous
40 * from the 0 offset of the file. Otherwise we could end up adjust i_size up
41 * to an extent that has a gap in between.
43 * Finally new_i_size should only be set in the case of truncate where we're not
44 * ready to use i_size_read() as the limiter yet.
46 void btrfs_inode_safe_disk_i_size_write(struct btrfs_inode *inode, u64 new_i_size)
48 struct btrfs_fs_info *fs_info = inode->root->fs_info;
49 u64 start, end, i_size;
52 spin_lock(&inode->lock);
53 i_size = new_i_size ?: i_size_read(&inode->vfs_inode);
54 if (btrfs_fs_incompat(fs_info, NO_HOLES)) {
55 inode->disk_i_size = i_size;
59 ret = find_contiguous_extent_bit(inode->file_extent_tree, 0, &start,
61 if (!ret && start == 0)
62 i_size = min(i_size, end + 1);
65 inode->disk_i_size = i_size;
67 spin_unlock(&inode->lock);
71 * Mark range within a file as having a new extent inserted.
73 * @inode: inode being modified
74 * @start: start file offset of the file extent we've inserted
75 * @len: logical length of the file extent item
77 * Call when we are inserting a new file extent where there was none before.
78 * Does not need to call this in the case where we're replacing an existing file
79 * extent, however if not sure it's fine to call this multiple times.
81 * The start and len must match the file extent item, so thus must be sectorsize
84 int btrfs_inode_set_file_extent_range(struct btrfs_inode *inode, u64 start,
90 ASSERT(IS_ALIGNED(start + len, inode->root->fs_info->sectorsize));
92 if (btrfs_fs_incompat(inode->root->fs_info, NO_HOLES))
94 return set_extent_bit(inode->file_extent_tree, start, start + len - 1,
99 * Mark an inode range as not having a backing extent.
101 * @inode: inode being modified
102 * @start: start file offset of the file extent we've inserted
103 * @len: logical length of the file extent item
105 * Called when we drop a file extent, for example when we truncate. Doesn't
106 * need to be called for cases where we're replacing a file extent, like when
107 * we've COWed a file extent.
109 * The start and len must match the file extent item, so thus must be sectorsize
112 int btrfs_inode_clear_file_extent_range(struct btrfs_inode *inode, u64 start,
118 ASSERT(IS_ALIGNED(start + len, inode->root->fs_info->sectorsize) ||
121 if (btrfs_fs_incompat(inode->root->fs_info, NO_HOLES))
123 return clear_extent_bit(inode->file_extent_tree, start,
124 start + len - 1, EXTENT_DIRTY, NULL);
127 static size_t bytes_to_csum_size(const struct btrfs_fs_info *fs_info, u32 bytes)
129 ASSERT(IS_ALIGNED(bytes, fs_info->sectorsize));
131 return (bytes >> fs_info->sectorsize_bits) * fs_info->csum_size;
134 static size_t csum_size_to_bytes(const struct btrfs_fs_info *fs_info, u32 csum_size)
136 ASSERT(IS_ALIGNED(csum_size, fs_info->csum_size));
138 return (csum_size / fs_info->csum_size) << fs_info->sectorsize_bits;
141 static inline u32 max_ordered_sum_bytes(const struct btrfs_fs_info *fs_info)
143 u32 max_csum_size = round_down(PAGE_SIZE - sizeof(struct btrfs_ordered_sum),
146 return csum_size_to_bytes(fs_info, max_csum_size);
150 * Calculate the total size needed to allocate for an ordered sum structure
151 * spanning @bytes in the file.
153 static int btrfs_ordered_sum_size(struct btrfs_fs_info *fs_info, unsigned long bytes)
155 return sizeof(struct btrfs_ordered_sum) + bytes_to_csum_size(fs_info, bytes);
158 int btrfs_insert_hole_extent(struct btrfs_trans_handle *trans,
159 struct btrfs_root *root,
160 u64 objectid, u64 pos, u64 num_bytes)
163 struct btrfs_file_extent_item *item;
164 struct btrfs_key file_key;
165 struct btrfs_path *path;
166 struct extent_buffer *leaf;
168 path = btrfs_alloc_path();
171 file_key.objectid = objectid;
172 file_key.offset = pos;
173 file_key.type = BTRFS_EXTENT_DATA_KEY;
175 ret = btrfs_insert_empty_item(trans, root, path, &file_key,
179 leaf = path->nodes[0];
180 item = btrfs_item_ptr(leaf, path->slots[0],
181 struct btrfs_file_extent_item);
182 btrfs_set_file_extent_disk_bytenr(leaf, item, 0);
183 btrfs_set_file_extent_disk_num_bytes(leaf, item, 0);
184 btrfs_set_file_extent_offset(leaf, item, 0);
185 btrfs_set_file_extent_num_bytes(leaf, item, num_bytes);
186 btrfs_set_file_extent_ram_bytes(leaf, item, num_bytes);
187 btrfs_set_file_extent_generation(leaf, item, trans->transid);
188 btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
189 btrfs_set_file_extent_compression(leaf, item, 0);
190 btrfs_set_file_extent_encryption(leaf, item, 0);
191 btrfs_set_file_extent_other_encoding(leaf, item, 0);
193 btrfs_mark_buffer_dirty(trans, leaf);
195 btrfs_free_path(path);
199 static struct btrfs_csum_item *
200 btrfs_lookup_csum(struct btrfs_trans_handle *trans,
201 struct btrfs_root *root,
202 struct btrfs_path *path,
205 struct btrfs_fs_info *fs_info = root->fs_info;
207 struct btrfs_key file_key;
208 struct btrfs_key found_key;
209 struct btrfs_csum_item *item;
210 struct extent_buffer *leaf;
212 const u32 csum_size = fs_info->csum_size;
215 file_key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
216 file_key.offset = bytenr;
217 file_key.type = BTRFS_EXTENT_CSUM_KEY;
218 ret = btrfs_search_slot(trans, root, &file_key, path, 0, cow);
221 leaf = path->nodes[0];
224 if (path->slots[0] == 0)
227 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
228 if (found_key.type != BTRFS_EXTENT_CSUM_KEY)
231 csum_offset = (bytenr - found_key.offset) >>
232 fs_info->sectorsize_bits;
233 csums_in_item = btrfs_item_size(leaf, path->slots[0]);
234 csums_in_item /= csum_size;
236 if (csum_offset == csums_in_item) {
239 } else if (csum_offset > csums_in_item) {
243 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item);
244 item = (struct btrfs_csum_item *)((unsigned char *)item +
245 csum_offset * csum_size);
253 int btrfs_lookup_file_extent(struct btrfs_trans_handle *trans,
254 struct btrfs_root *root,
255 struct btrfs_path *path, u64 objectid,
258 struct btrfs_key file_key;
259 int ins_len = mod < 0 ? -1 : 0;
262 file_key.objectid = objectid;
263 file_key.offset = offset;
264 file_key.type = BTRFS_EXTENT_DATA_KEY;
266 return btrfs_search_slot(trans, root, &file_key, path, ins_len, cow);
270 * Find checksums for logical bytenr range [disk_bytenr, disk_bytenr + len) and
271 * store the result to @dst.
273 * Return >0 for the number of sectors we found.
274 * Return 0 for the range [disk_bytenr, disk_bytenr + sectorsize) has no csum
275 * for it. Caller may want to try next sector until one range is hit.
276 * Return <0 for fatal error.
278 static int search_csum_tree(struct btrfs_fs_info *fs_info,
279 struct btrfs_path *path, u64 disk_bytenr,
282 struct btrfs_root *csum_root;
283 struct btrfs_csum_item *item = NULL;
284 struct btrfs_key key;
285 const u32 sectorsize = fs_info->sectorsize;
286 const u32 csum_size = fs_info->csum_size;
292 ASSERT(IS_ALIGNED(disk_bytenr, sectorsize) &&
293 IS_ALIGNED(len, sectorsize));
295 /* Check if the current csum item covers disk_bytenr */
296 if (path->nodes[0]) {
297 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
298 struct btrfs_csum_item);
299 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
300 itemsize = btrfs_item_size(path->nodes[0], path->slots[0]);
302 csum_start = key.offset;
303 csum_len = (itemsize / csum_size) * sectorsize;
305 if (in_range(disk_bytenr, csum_start, csum_len))
309 /* Current item doesn't contain the desired range, search again */
310 btrfs_release_path(path);
311 csum_root = btrfs_csum_root(fs_info, disk_bytenr);
312 item = btrfs_lookup_csum(NULL, csum_root, path, disk_bytenr, 0);
317 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
318 itemsize = btrfs_item_size(path->nodes[0], path->slots[0]);
320 csum_start = key.offset;
321 csum_len = (itemsize / csum_size) * sectorsize;
322 ASSERT(in_range(disk_bytenr, csum_start, csum_len));
325 ret = (min(csum_start + csum_len, disk_bytenr + len) -
326 disk_bytenr) >> fs_info->sectorsize_bits;
327 read_extent_buffer(path->nodes[0], dst, (unsigned long)item,
330 if (ret == -ENOENT || ret == -EFBIG)
336 * Lookup the checksum for the read bio in csum tree.
338 * Return: BLK_STS_RESOURCE if allocating memory fails, BLK_STS_OK otherwise.
340 blk_status_t btrfs_lookup_bio_sums(struct btrfs_bio *bbio)
342 struct btrfs_inode *inode = bbio->inode;
343 struct btrfs_fs_info *fs_info = inode->root->fs_info;
344 struct bio *bio = &bbio->bio;
345 struct btrfs_path *path;
346 const u32 sectorsize = fs_info->sectorsize;
347 const u32 csum_size = fs_info->csum_size;
348 u32 orig_len = bio->bi_iter.bi_size;
349 u64 orig_disk_bytenr = bio->bi_iter.bi_sector << SECTOR_SHIFT;
350 const unsigned int nblocks = orig_len >> fs_info->sectorsize_bits;
351 blk_status_t ret = BLK_STS_OK;
354 if ((inode->flags & BTRFS_INODE_NODATASUM) ||
355 test_bit(BTRFS_FS_STATE_NO_CSUMS, &fs_info->fs_state))
359 * This function is only called for read bio.
361 * This means two things:
362 * - All our csums should only be in csum tree
363 * No ordered extents csums, as ordered extents are only for write
365 * - No need to bother any other info from bvec
366 * Since we're looking up csums, the only important info is the
367 * disk_bytenr and the length, which can be extracted from bi_iter
370 ASSERT(bio_op(bio) == REQ_OP_READ);
371 path = btrfs_alloc_path();
373 return BLK_STS_RESOURCE;
375 if (nblocks * csum_size > BTRFS_BIO_INLINE_CSUM_SIZE) {
376 bbio->csum = kmalloc_array(nblocks, csum_size, GFP_NOFS);
378 btrfs_free_path(path);
379 return BLK_STS_RESOURCE;
382 bbio->csum = bbio->csum_inline;
386 * If requested number of sectors is larger than one leaf can contain,
387 * kick the readahead for csum tree.
389 if (nblocks > fs_info->csums_per_leaf)
390 path->reada = READA_FORWARD;
393 * the free space stuff is only read when it hasn't been
394 * updated in the current transaction. So, we can safely
395 * read from the commit root and sidestep a nasty deadlock
396 * between reading the free space cache and updating the csum tree.
398 if (btrfs_is_free_space_inode(inode)) {
399 path->search_commit_root = 1;
400 path->skip_locking = 1;
403 while (bio_offset < orig_len) {
405 u64 cur_disk_bytenr = orig_disk_bytenr + bio_offset;
406 u8 *csum_dst = bbio->csum +
407 (bio_offset >> fs_info->sectorsize_bits) * csum_size;
409 count = search_csum_tree(fs_info, path, cur_disk_bytenr,
410 orig_len - bio_offset, csum_dst);
412 ret = errno_to_blk_status(count);
413 if (bbio->csum != bbio->csum_inline)
420 * We didn't find a csum for this range. We need to make sure
421 * we complain loudly about this, because we are not NODATASUM.
423 * However for the DATA_RELOC inode we could potentially be
424 * relocating data extents for a NODATASUM inode, so the inode
425 * itself won't be marked with NODATASUM, but the extent we're
426 * copying is in fact NODATASUM. If we don't find a csum we
427 * assume this is the case.
430 memset(csum_dst, 0, csum_size);
433 if (btrfs_root_id(inode->root) == BTRFS_DATA_RELOC_TREE_OBJECTID) {
434 u64 file_offset = bbio->file_offset + bio_offset;
436 set_extent_bit(&inode->io_tree, file_offset,
437 file_offset + sectorsize - 1,
438 EXTENT_NODATASUM, NULL);
440 btrfs_warn_rl(fs_info,
441 "csum hole found for disk bytenr range [%llu, %llu)",
442 cur_disk_bytenr, cur_disk_bytenr + sectorsize);
445 bio_offset += count * sectorsize;
448 btrfs_free_path(path);
453 * Search for checksums for a given logical range.
455 * @root: The root where to look for checksums.
456 * @start: Logical address of target checksum range.
457 * @end: End offset (inclusive) of the target checksum range.
458 * @list: List for adding each checksum that was found.
459 * Can be NULL in case the caller only wants to check if
460 * there any checksums for the range.
461 * @nowait: Indicate if the search must be non-blocking or not.
463 * Return < 0 on error, 0 if no checksums were found, or 1 if checksums were
466 int btrfs_lookup_csums_list(struct btrfs_root *root, u64 start, u64 end,
467 struct list_head *list, bool nowait)
469 struct btrfs_fs_info *fs_info = root->fs_info;
470 struct btrfs_key key;
471 struct btrfs_path *path;
472 struct extent_buffer *leaf;
473 struct btrfs_ordered_sum *sums;
474 struct btrfs_csum_item *item;
476 bool found_csums = false;
478 ASSERT(IS_ALIGNED(start, fs_info->sectorsize) &&
479 IS_ALIGNED(end + 1, fs_info->sectorsize));
481 path = btrfs_alloc_path();
485 path->nowait = nowait;
487 key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
489 key.type = BTRFS_EXTENT_CSUM_KEY;
491 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
494 if (ret > 0 && path->slots[0] > 0) {
495 leaf = path->nodes[0];
496 btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
499 * There are two cases we can hit here for the previous csum
502 * |<- search range ->|
506 * |<- search range ->|
509 * Check if the previous csum item covers the leading part of
510 * the search range. If so we have to start from previous csum
513 if (key.objectid == BTRFS_EXTENT_CSUM_OBJECTID &&
514 key.type == BTRFS_EXTENT_CSUM_KEY) {
515 if (bytes_to_csum_size(fs_info, start - key.offset) <
516 btrfs_item_size(leaf, path->slots[0] - 1))
521 while (start <= end) {
524 leaf = path->nodes[0];
525 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
526 ret = btrfs_next_leaf(root, path);
531 leaf = path->nodes[0];
534 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
535 if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
536 key.type != BTRFS_EXTENT_CSUM_KEY ||
540 if (key.offset > start)
543 csum_end = key.offset + csum_size_to_bytes(fs_info,
544 btrfs_item_size(leaf, path->slots[0]));
545 if (csum_end <= start) {
554 csum_end = min(csum_end, end + 1);
555 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
556 struct btrfs_csum_item);
557 while (start < csum_end) {
558 unsigned long offset;
561 size = min_t(size_t, csum_end - start,
562 max_ordered_sum_bytes(fs_info));
563 sums = kzalloc(btrfs_ordered_sum_size(fs_info, size),
570 sums->logical = start;
573 offset = bytes_to_csum_size(fs_info, start - key.offset);
575 read_extent_buffer(path->nodes[0],
577 ((unsigned long)item) + offset,
578 bytes_to_csum_size(fs_info, size));
581 list_add_tail(&sums->list, list);
586 btrfs_free_path(path);
589 struct btrfs_ordered_sum *tmp_sums;
591 list_for_each_entry_safe(sums, tmp_sums, list, list)
598 return found_csums ? 1 : 0;
602 * Do the same work as btrfs_lookup_csums_list(), the difference is in how
603 * we return the result.
605 * This version will set the corresponding bits in @csum_bitmap to represent
606 * that there is a csum found.
607 * Each bit represents a sector. Thus caller should ensure @csum_buf passed
608 * in is large enough to contain all csums.
610 int btrfs_lookup_csums_bitmap(struct btrfs_root *root, struct btrfs_path *path,
611 u64 start, u64 end, u8 *csum_buf,
612 unsigned long *csum_bitmap)
614 struct btrfs_fs_info *fs_info = root->fs_info;
615 struct btrfs_key key;
616 struct extent_buffer *leaf;
617 struct btrfs_csum_item *item;
618 const u64 orig_start = start;
619 bool free_path = false;
622 ASSERT(IS_ALIGNED(start, fs_info->sectorsize) &&
623 IS_ALIGNED(end + 1, fs_info->sectorsize));
626 path = btrfs_alloc_path();
632 /* Check if we can reuse the previous path. */
633 if (path->nodes[0]) {
634 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
636 if (key.objectid == BTRFS_EXTENT_CSUM_OBJECTID &&
637 key.type == BTRFS_EXTENT_CSUM_KEY &&
640 btrfs_release_path(path);
643 key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
644 key.type = BTRFS_EXTENT_CSUM_KEY;
647 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
650 if (ret > 0 && path->slots[0] > 0) {
651 leaf = path->nodes[0];
652 btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
655 * There are two cases we can hit here for the previous csum
658 * |<- search range ->|
662 * |<- search range ->|
665 * Check if the previous csum item covers the leading part of
666 * the search range. If so we have to start from previous csum
669 if (key.objectid == BTRFS_EXTENT_CSUM_OBJECTID &&
670 key.type == BTRFS_EXTENT_CSUM_KEY) {
671 if (bytes_to_csum_size(fs_info, start - key.offset) <
672 btrfs_item_size(leaf, path->slots[0] - 1))
678 while (start <= end) {
681 leaf = path->nodes[0];
682 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
683 ret = btrfs_next_leaf(root, path);
688 leaf = path->nodes[0];
691 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
692 if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
693 key.type != BTRFS_EXTENT_CSUM_KEY ||
697 if (key.offset > start)
700 csum_end = key.offset + csum_size_to_bytes(fs_info,
701 btrfs_item_size(leaf, path->slots[0]));
702 if (csum_end <= start) {
707 csum_end = min(csum_end, end + 1);
708 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
709 struct btrfs_csum_item);
710 while (start < csum_end) {
711 unsigned long offset;
713 u8 *csum_dest = csum_buf + bytes_to_csum_size(fs_info,
716 size = min_t(size_t, csum_end - start, end + 1 - start);
718 offset = bytes_to_csum_size(fs_info, start - key.offset);
720 read_extent_buffer(path->nodes[0], csum_dest,
721 ((unsigned long)item) + offset,
722 bytes_to_csum_size(fs_info, size));
724 bitmap_set(csum_bitmap,
725 (start - orig_start) >> fs_info->sectorsize_bits,
726 size >> fs_info->sectorsize_bits);
735 btrfs_free_path(path);
740 * Calculate checksums of the data contained inside a bio.
742 blk_status_t btrfs_csum_one_bio(struct btrfs_bio *bbio)
744 struct btrfs_ordered_extent *ordered = bbio->ordered;
745 struct btrfs_inode *inode = bbio->inode;
746 struct btrfs_fs_info *fs_info = inode->root->fs_info;
747 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
748 struct bio *bio = &bbio->bio;
749 struct btrfs_ordered_sum *sums;
751 struct bvec_iter iter;
754 unsigned int blockcount;
758 nofs_flag = memalloc_nofs_save();
759 sums = kvzalloc(btrfs_ordered_sum_size(fs_info, bio->bi_iter.bi_size),
761 memalloc_nofs_restore(nofs_flag);
764 return BLK_STS_RESOURCE;
766 sums->len = bio->bi_iter.bi_size;
767 INIT_LIST_HEAD(&sums->list);
769 sums->logical = bio->bi_iter.bi_sector << SECTOR_SHIFT;
772 shash->tfm = fs_info->csum_shash;
774 bio_for_each_segment(bvec, bio, iter) {
775 blockcount = BTRFS_BYTES_TO_BLKS(fs_info,
776 bvec.bv_len + fs_info->sectorsize
779 for (i = 0; i < blockcount; i++) {
780 data = bvec_kmap_local(&bvec);
781 crypto_shash_digest(shash,
782 data + (i * fs_info->sectorsize),
786 index += fs_info->csum_size;
792 btrfs_add_ordered_sum(ordered, sums);
797 * Nodatasum I/O on zoned file systems still requires an btrfs_ordered_sum to
798 * record the updated logical address on Zone Append completion.
799 * Allocate just the structure with an empty sums array here for that case.
801 blk_status_t btrfs_alloc_dummy_sum(struct btrfs_bio *bbio)
803 bbio->sums = kmalloc(sizeof(*bbio->sums), GFP_NOFS);
805 return BLK_STS_RESOURCE;
806 bbio->sums->len = bbio->bio.bi_iter.bi_size;
807 bbio->sums->logical = bbio->bio.bi_iter.bi_sector << SECTOR_SHIFT;
808 btrfs_add_ordered_sum(bbio->ordered, bbio->sums);
813 * Remove one checksum overlapping a range.
815 * This expects the key to describe the csum pointed to by the path, and it
816 * expects the csum to overlap the range [bytenr, len]
818 * The csum should not be entirely contained in the range and the range should
819 * not be entirely contained in the csum.
821 * This calls btrfs_truncate_item with the correct args based on the overlap,
822 * and fixes up the key as required.
824 static noinline void truncate_one_csum(struct btrfs_trans_handle *trans,
825 struct btrfs_path *path,
826 struct btrfs_key *key,
829 struct btrfs_fs_info *fs_info = trans->fs_info;
830 struct extent_buffer *leaf;
831 const u32 csum_size = fs_info->csum_size;
833 u64 end_byte = bytenr + len;
834 u32 blocksize_bits = fs_info->sectorsize_bits;
836 leaf = path->nodes[0];
837 csum_end = btrfs_item_size(leaf, path->slots[0]) / csum_size;
838 csum_end <<= blocksize_bits;
839 csum_end += key->offset;
841 if (key->offset < bytenr && csum_end <= end_byte) {
846 * A simple truncate off the end of the item
848 u32 new_size = (bytenr - key->offset) >> blocksize_bits;
849 new_size *= csum_size;
850 btrfs_truncate_item(trans, path, new_size, 1);
851 } else if (key->offset >= bytenr && csum_end > end_byte &&
852 end_byte > key->offset) {
857 * we need to truncate from the beginning of the csum
859 u32 new_size = (csum_end - end_byte) >> blocksize_bits;
860 new_size *= csum_size;
862 btrfs_truncate_item(trans, path, new_size, 0);
864 key->offset = end_byte;
865 btrfs_set_item_key_safe(trans, path, key);
872 * Delete the csum items from the csum tree for a given range of bytes.
874 int btrfs_del_csums(struct btrfs_trans_handle *trans,
875 struct btrfs_root *root, u64 bytenr, u64 len)
877 struct btrfs_fs_info *fs_info = trans->fs_info;
878 struct btrfs_path *path;
879 struct btrfs_key key;
880 u64 end_byte = bytenr + len;
882 struct extent_buffer *leaf;
884 const u32 csum_size = fs_info->csum_size;
885 u32 blocksize_bits = fs_info->sectorsize_bits;
887 ASSERT(btrfs_root_id(root) == BTRFS_CSUM_TREE_OBJECTID ||
888 btrfs_root_id(root) == BTRFS_TREE_LOG_OBJECTID);
890 path = btrfs_alloc_path();
895 key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
896 key.offset = end_byte - 1;
897 key.type = BTRFS_EXTENT_CSUM_KEY;
899 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
902 if (path->slots[0] == 0)
905 } else if (ret < 0) {
909 leaf = path->nodes[0];
910 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
912 if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
913 key.type != BTRFS_EXTENT_CSUM_KEY) {
917 if (key.offset >= end_byte)
920 csum_end = btrfs_item_size(leaf, path->slots[0]) / csum_size;
921 csum_end <<= blocksize_bits;
922 csum_end += key.offset;
924 /* this csum ends before we start, we're done */
925 if (csum_end <= bytenr)
928 /* delete the entire item, it is inside our range */
929 if (key.offset >= bytenr && csum_end <= end_byte) {
933 * Check how many csum items preceding this one in this
934 * leaf correspond to our range and then delete them all
937 if (key.offset > bytenr && path->slots[0] > 0) {
938 int slot = path->slots[0] - 1;
943 btrfs_item_key_to_cpu(leaf, &pk, slot);
944 if (pk.offset < bytenr ||
945 pk.type != BTRFS_EXTENT_CSUM_KEY ||
947 BTRFS_EXTENT_CSUM_OBJECTID)
949 path->slots[0] = slot;
951 key.offset = pk.offset;
955 ret = btrfs_del_items(trans, root, path,
956 path->slots[0], del_nr);
959 if (key.offset == bytenr)
961 } else if (key.offset < bytenr && csum_end > end_byte) {
962 unsigned long offset;
963 unsigned long shift_len;
964 unsigned long item_offset;
969 * Our bytes are in the middle of the csum,
970 * we need to split this item and insert a new one.
972 * But we can't drop the path because the
973 * csum could change, get removed, extended etc.
975 * The trick here is the max size of a csum item leaves
976 * enough room in the tree block for a single
977 * item header. So, we split the item in place,
978 * adding a new header pointing to the existing
979 * bytes. Then we loop around again and we have
980 * a nicely formed csum item that we can neatly
983 offset = (bytenr - key.offset) >> blocksize_bits;
986 shift_len = (len >> blocksize_bits) * csum_size;
988 item_offset = btrfs_item_ptr_offset(leaf,
991 memzero_extent_buffer(leaf, item_offset + offset,
996 * btrfs_split_item returns -EAGAIN when the
997 * item changed size or key
999 ret = btrfs_split_item(trans, root, path, &key, offset);
1000 if (ret && ret != -EAGAIN) {
1001 btrfs_abort_transaction(trans, ret);
1006 key.offset = end_byte - 1;
1008 truncate_one_csum(trans, path, &key, bytenr, len);
1009 if (key.offset < bytenr)
1012 btrfs_release_path(path);
1014 btrfs_free_path(path);
1018 static int find_next_csum_offset(struct btrfs_root *root,
1019 struct btrfs_path *path,
1022 const u32 nritems = btrfs_header_nritems(path->nodes[0]);
1023 struct btrfs_key found_key;
1024 int slot = path->slots[0] + 1;
1027 if (nritems == 0 || slot >= nritems) {
1028 ret = btrfs_next_leaf(root, path);
1031 } else if (ret > 0) {
1032 *next_offset = (u64)-1;
1035 slot = path->slots[0];
1038 btrfs_item_key_to_cpu(path->nodes[0], &found_key, slot);
1040 if (found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
1041 found_key.type != BTRFS_EXTENT_CSUM_KEY)
1042 *next_offset = (u64)-1;
1044 *next_offset = found_key.offset;
1049 int btrfs_csum_file_blocks(struct btrfs_trans_handle *trans,
1050 struct btrfs_root *root,
1051 struct btrfs_ordered_sum *sums)
1053 struct btrfs_fs_info *fs_info = root->fs_info;
1054 struct btrfs_key file_key;
1055 struct btrfs_key found_key;
1056 struct btrfs_path *path;
1057 struct btrfs_csum_item *item;
1058 struct btrfs_csum_item *item_end;
1059 struct extent_buffer *leaf = NULL;
1061 u64 total_bytes = 0;
1068 const u32 csum_size = fs_info->csum_size;
1070 path = btrfs_alloc_path();
1074 next_offset = (u64)-1;
1076 bytenr = sums->logical + total_bytes;
1077 file_key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
1078 file_key.offset = bytenr;
1079 file_key.type = BTRFS_EXTENT_CSUM_KEY;
1081 item = btrfs_lookup_csum(trans, root, path, bytenr, 1);
1082 if (!IS_ERR(item)) {
1084 leaf = path->nodes[0];
1085 item_end = btrfs_item_ptr(leaf, path->slots[0],
1086 struct btrfs_csum_item);
1087 item_end = (struct btrfs_csum_item *)((char *)item_end +
1088 btrfs_item_size(leaf, path->slots[0]));
1091 ret = PTR_ERR(item);
1092 if (ret != -EFBIG && ret != -ENOENT)
1095 if (ret == -EFBIG) {
1097 /* we found one, but it isn't big enough yet */
1098 leaf = path->nodes[0];
1099 item_size = btrfs_item_size(leaf, path->slots[0]);
1100 if ((item_size / csum_size) >=
1101 MAX_CSUM_ITEMS(fs_info, csum_size)) {
1102 /* already at max size, make a new one */
1106 /* We didn't find a csum item, insert one. */
1107 ret = find_next_csum_offset(root, path, &next_offset);
1115 * At this point, we know the tree has a checksum item that ends at an
1116 * offset matching the start of the checksum range we want to insert.
1117 * We try to extend that item as much as possible and then add as many
1118 * checksums to it as they fit.
1120 * First check if the leaf has enough free space for at least one
1121 * checksum. If it has go directly to the item extension code, otherwise
1122 * release the path and do a search for insertion before the extension.
1124 if (btrfs_leaf_free_space(leaf) >= csum_size) {
1125 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1126 csum_offset = (bytenr - found_key.offset) >>
1127 fs_info->sectorsize_bits;
1131 btrfs_release_path(path);
1132 path->search_for_extension = 1;
1133 ret = btrfs_search_slot(trans, root, &file_key, path,
1135 path->search_for_extension = 0;
1140 if (path->slots[0] == 0)
1145 leaf = path->nodes[0];
1146 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1147 csum_offset = (bytenr - found_key.offset) >> fs_info->sectorsize_bits;
1149 if (found_key.type != BTRFS_EXTENT_CSUM_KEY ||
1150 found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
1151 csum_offset >= MAX_CSUM_ITEMS(fs_info, csum_size)) {
1156 if (csum_offset == btrfs_item_size(leaf, path->slots[0]) /
1162 tmp = sums->len - total_bytes;
1163 tmp >>= fs_info->sectorsize_bits;
1165 extend_nr = max_t(int, 1, tmp);
1168 * A log tree can already have checksum items with a subset of
1169 * the checksums we are trying to log. This can happen after
1170 * doing a sequence of partial writes into prealloc extents and
1171 * fsyncs in between, with a full fsync logging a larger subrange
1172 * of an extent for which a previous fast fsync logged a smaller
1173 * subrange. And this happens in particular due to merging file
1174 * extent items when we complete an ordered extent for a range
1175 * covered by a prealloc extent - this is done at
1176 * btrfs_mark_extent_written().
1178 * So if we try to extend the previous checksum item, which has
1179 * a range that ends at the start of the range we want to insert,
1180 * make sure we don't extend beyond the start offset of the next
1181 * checksum item. If we are at the last item in the leaf, then
1182 * forget the optimization of extending and add a new checksum
1183 * item - it is not worth the complexity of releasing the path,
1184 * getting the first key for the next leaf, repeat the btree
1185 * search, etc, because log trees are temporary anyway and it
1186 * would only save a few bytes of leaf space.
1188 if (btrfs_root_id(root) == BTRFS_TREE_LOG_OBJECTID) {
1189 if (path->slots[0] + 1 >=
1190 btrfs_header_nritems(path->nodes[0])) {
1191 ret = find_next_csum_offset(root, path, &next_offset);
1198 ret = find_next_csum_offset(root, path, &next_offset);
1202 tmp = (next_offset - bytenr) >> fs_info->sectorsize_bits;
1204 extend_nr = min_t(int, extend_nr, tmp);
1207 diff = (csum_offset + extend_nr) * csum_size;
1209 MAX_CSUM_ITEMS(fs_info, csum_size) * csum_size);
1211 diff = diff - btrfs_item_size(leaf, path->slots[0]);
1212 diff = min_t(u32, btrfs_leaf_free_space(leaf), diff);
1216 btrfs_extend_item(trans, path, diff);
1222 btrfs_release_path(path);
1227 tmp = sums->len - total_bytes;
1228 tmp >>= fs_info->sectorsize_bits;
1229 tmp = min(tmp, (next_offset - file_key.offset) >>
1230 fs_info->sectorsize_bits);
1232 tmp = max_t(u64, 1, tmp);
1233 tmp = min_t(u64, tmp, MAX_CSUM_ITEMS(fs_info, csum_size));
1234 ins_size = csum_size * tmp;
1236 ins_size = csum_size;
1238 ret = btrfs_insert_empty_item(trans, root, path, &file_key,
1242 leaf = path->nodes[0];
1244 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item);
1245 item_end = (struct btrfs_csum_item *)((unsigned char *)item +
1246 btrfs_item_size(leaf, path->slots[0]));
1247 item = (struct btrfs_csum_item *)((unsigned char *)item +
1248 csum_offset * csum_size);
1250 ins_size = (u32)(sums->len - total_bytes) >> fs_info->sectorsize_bits;
1251 ins_size *= csum_size;
1252 ins_size = min_t(u32, (unsigned long)item_end - (unsigned long)item,
1254 write_extent_buffer(leaf, sums->sums + index, (unsigned long)item,
1258 ins_size /= csum_size;
1259 total_bytes += ins_size * fs_info->sectorsize;
1261 btrfs_mark_buffer_dirty(trans, path->nodes[0]);
1262 if (total_bytes < sums->len) {
1263 btrfs_release_path(path);
1268 btrfs_free_path(path);
1272 void btrfs_extent_item_to_extent_map(struct btrfs_inode *inode,
1273 const struct btrfs_path *path,
1274 struct btrfs_file_extent_item *fi,
1275 struct extent_map *em)
1277 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1278 struct btrfs_root *root = inode->root;
1279 struct extent_buffer *leaf = path->nodes[0];
1280 const int slot = path->slots[0];
1281 struct btrfs_key key;
1284 u8 type = btrfs_file_extent_type(leaf, fi);
1285 int compress_type = btrfs_file_extent_compression(leaf, fi);
1287 btrfs_item_key_to_cpu(leaf, &key, slot);
1288 extent_start = key.offset;
1289 em->ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1290 em->generation = btrfs_file_extent_generation(leaf, fi);
1291 if (type == BTRFS_FILE_EXTENT_REG ||
1292 type == BTRFS_FILE_EXTENT_PREALLOC) {
1293 em->start = extent_start;
1294 em->len = btrfs_file_extent_end(path) - extent_start;
1295 em->orig_start = extent_start -
1296 btrfs_file_extent_offset(leaf, fi);
1297 em->orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
1298 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1300 em->block_start = EXTENT_MAP_HOLE;
1303 if (compress_type != BTRFS_COMPRESS_NONE) {
1304 extent_map_set_compression(em, compress_type);
1305 em->block_start = bytenr;
1306 em->block_len = em->orig_block_len;
1308 bytenr += btrfs_file_extent_offset(leaf, fi);
1309 em->block_start = bytenr;
1310 em->block_len = em->len;
1311 if (type == BTRFS_FILE_EXTENT_PREALLOC)
1312 em->flags |= EXTENT_FLAG_PREALLOC;
1314 } else if (type == BTRFS_FILE_EXTENT_INLINE) {
1315 /* Tree-checker has ensured this. */
1316 ASSERT(extent_start == 0);
1318 em->block_start = EXTENT_MAP_INLINE;
1320 em->len = fs_info->sectorsize;
1322 * Initialize orig_start and block_len with the same values
1323 * as in inode.c:btrfs_get_extent().
1325 em->orig_start = EXTENT_MAP_HOLE;
1326 em->block_len = (u64)-1;
1327 extent_map_set_compression(em, compress_type);
1330 "unknown file extent item type %d, inode %llu, offset %llu, "
1331 "root %llu", type, btrfs_ino(inode), extent_start,
1332 btrfs_root_id(root));
1337 * Returns the end offset (non inclusive) of the file extent item the given path
1338 * points to. If it points to an inline extent, the returned offset is rounded
1339 * up to the sector size.
1341 u64 btrfs_file_extent_end(const struct btrfs_path *path)
1343 const struct extent_buffer *leaf = path->nodes[0];
1344 const int slot = path->slots[0];
1345 struct btrfs_file_extent_item *fi;
1346 struct btrfs_key key;
1349 btrfs_item_key_to_cpu(leaf, &key, slot);
1350 ASSERT(key.type == BTRFS_EXTENT_DATA_KEY);
1351 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
1353 if (btrfs_file_extent_type(leaf, fi) == BTRFS_FILE_EXTENT_INLINE)
1354 end = leaf->fs_info->sectorsize;
1356 end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);