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 (inode->root->root_key.objectid ==
434 BTRFS_DATA_RELOC_TREE_OBJECTID) {
435 u64 file_offset = bbio->file_offset + bio_offset;
437 set_extent_bit(&inode->io_tree, file_offset,
438 file_offset + sectorsize - 1,
439 EXTENT_NODATASUM, NULL);
441 btrfs_warn_rl(fs_info,
442 "csum hole found for disk bytenr range [%llu, %llu)",
443 cur_disk_bytenr, cur_disk_bytenr + sectorsize);
446 bio_offset += count * sectorsize;
449 btrfs_free_path(path);
453 int btrfs_lookup_csums_list(struct btrfs_root *root, u64 start, u64 end,
454 struct list_head *list, int search_commit,
457 struct btrfs_fs_info *fs_info = root->fs_info;
458 struct btrfs_key key;
459 struct btrfs_path *path;
460 struct extent_buffer *leaf;
461 struct btrfs_ordered_sum *sums;
462 struct btrfs_csum_item *item;
466 ASSERT(IS_ALIGNED(start, fs_info->sectorsize) &&
467 IS_ALIGNED(end + 1, fs_info->sectorsize));
469 path = btrfs_alloc_path();
473 path->nowait = nowait;
475 path->skip_locking = 1;
476 path->reada = READA_FORWARD;
477 path->search_commit_root = 1;
480 key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
482 key.type = BTRFS_EXTENT_CSUM_KEY;
484 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
487 if (ret > 0 && path->slots[0] > 0) {
488 leaf = path->nodes[0];
489 btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
492 * There are two cases we can hit here for the previous csum
495 * |<- search range ->|
499 * |<- search range ->|
502 * Check if the previous csum item covers the leading part of
503 * the search range. If so we have to start from previous csum
506 if (key.objectid == BTRFS_EXTENT_CSUM_OBJECTID &&
507 key.type == BTRFS_EXTENT_CSUM_KEY) {
508 if (bytes_to_csum_size(fs_info, start - key.offset) <
509 btrfs_item_size(leaf, path->slots[0] - 1))
514 while (start <= end) {
517 leaf = path->nodes[0];
518 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
519 ret = btrfs_next_leaf(root, path);
524 leaf = path->nodes[0];
527 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
528 if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
529 key.type != BTRFS_EXTENT_CSUM_KEY ||
533 if (key.offset > start)
536 csum_end = key.offset + csum_size_to_bytes(fs_info,
537 btrfs_item_size(leaf, path->slots[0]));
538 if (csum_end <= start) {
543 csum_end = min(csum_end, end + 1);
544 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
545 struct btrfs_csum_item);
546 while (start < csum_end) {
547 unsigned long offset;
550 size = min_t(size_t, csum_end - start,
551 max_ordered_sum_bytes(fs_info));
552 sums = kzalloc(btrfs_ordered_sum_size(fs_info, size),
559 sums->logical = start;
562 offset = bytes_to_csum_size(fs_info, start - key.offset);
564 read_extent_buffer(path->nodes[0],
566 ((unsigned long)item) + offset,
567 bytes_to_csum_size(fs_info, size));
570 list_add_tail(&sums->list, &tmplist);
576 while (ret < 0 && !list_empty(&tmplist)) {
577 sums = list_entry(tmplist.next, struct btrfs_ordered_sum, list);
578 list_del(&sums->list);
581 list_splice_tail(&tmplist, list);
583 btrfs_free_path(path);
588 * Do the same work as btrfs_lookup_csums_list(), the difference is in how
589 * we return the result.
591 * This version will set the corresponding bits in @csum_bitmap to represent
592 * that there is a csum found.
593 * Each bit represents a sector. Thus caller should ensure @csum_buf passed
594 * in is large enough to contain all csums.
596 int btrfs_lookup_csums_bitmap(struct btrfs_root *root, struct btrfs_path *path,
597 u64 start, u64 end, u8 *csum_buf,
598 unsigned long *csum_bitmap)
600 struct btrfs_fs_info *fs_info = root->fs_info;
601 struct btrfs_key key;
602 struct extent_buffer *leaf;
603 struct btrfs_csum_item *item;
604 const u64 orig_start = start;
605 bool free_path = false;
608 ASSERT(IS_ALIGNED(start, fs_info->sectorsize) &&
609 IS_ALIGNED(end + 1, fs_info->sectorsize));
612 path = btrfs_alloc_path();
618 /* Check if we can reuse the previous path. */
619 if (path->nodes[0]) {
620 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
622 if (key.objectid == BTRFS_EXTENT_CSUM_OBJECTID &&
623 key.type == BTRFS_EXTENT_CSUM_KEY &&
626 btrfs_release_path(path);
629 key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
630 key.type = BTRFS_EXTENT_CSUM_KEY;
633 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
636 if (ret > 0 && path->slots[0] > 0) {
637 leaf = path->nodes[0];
638 btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
641 * There are two cases we can hit here for the previous csum
644 * |<- search range ->|
648 * |<- search range ->|
651 * Check if the previous csum item covers the leading part of
652 * the search range. If so we have to start from previous csum
655 if (key.objectid == BTRFS_EXTENT_CSUM_OBJECTID &&
656 key.type == BTRFS_EXTENT_CSUM_KEY) {
657 if (bytes_to_csum_size(fs_info, start - key.offset) <
658 btrfs_item_size(leaf, path->slots[0] - 1))
664 while (start <= end) {
667 leaf = path->nodes[0];
668 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
669 ret = btrfs_next_leaf(root, path);
674 leaf = path->nodes[0];
677 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
678 if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
679 key.type != BTRFS_EXTENT_CSUM_KEY ||
683 if (key.offset > start)
686 csum_end = key.offset + csum_size_to_bytes(fs_info,
687 btrfs_item_size(leaf, path->slots[0]));
688 if (csum_end <= start) {
693 csum_end = min(csum_end, end + 1);
694 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
695 struct btrfs_csum_item);
696 while (start < csum_end) {
697 unsigned long offset;
699 u8 *csum_dest = csum_buf + bytes_to_csum_size(fs_info,
702 size = min_t(size_t, csum_end - start, end + 1 - start);
704 offset = bytes_to_csum_size(fs_info, start - key.offset);
706 read_extent_buffer(path->nodes[0], csum_dest,
707 ((unsigned long)item) + offset,
708 bytes_to_csum_size(fs_info, size));
710 bitmap_set(csum_bitmap,
711 (start - orig_start) >> fs_info->sectorsize_bits,
712 size >> fs_info->sectorsize_bits);
721 btrfs_free_path(path);
726 * Calculate checksums of the data contained inside a bio.
728 blk_status_t btrfs_csum_one_bio(struct btrfs_bio *bbio)
730 struct btrfs_ordered_extent *ordered = bbio->ordered;
731 struct btrfs_inode *inode = bbio->inode;
732 struct btrfs_fs_info *fs_info = inode->root->fs_info;
733 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
734 struct bio *bio = &bbio->bio;
735 struct btrfs_ordered_sum *sums;
737 struct bvec_iter iter;
740 unsigned int blockcount;
744 nofs_flag = memalloc_nofs_save();
745 sums = kvzalloc(btrfs_ordered_sum_size(fs_info, bio->bi_iter.bi_size),
747 memalloc_nofs_restore(nofs_flag);
750 return BLK_STS_RESOURCE;
752 sums->len = bio->bi_iter.bi_size;
753 INIT_LIST_HEAD(&sums->list);
755 sums->logical = bio->bi_iter.bi_sector << SECTOR_SHIFT;
758 shash->tfm = fs_info->csum_shash;
760 bio_for_each_segment(bvec, bio, iter) {
761 blockcount = BTRFS_BYTES_TO_BLKS(fs_info,
762 bvec.bv_len + fs_info->sectorsize
765 for (i = 0; i < blockcount; i++) {
766 data = bvec_kmap_local(&bvec);
767 crypto_shash_digest(shash,
768 data + (i * fs_info->sectorsize),
772 index += fs_info->csum_size;
778 btrfs_add_ordered_sum(ordered, sums);
783 * Nodatasum I/O on zoned file systems still requires an btrfs_ordered_sum to
784 * record the updated logical address on Zone Append completion.
785 * Allocate just the structure with an empty sums array here for that case.
787 blk_status_t btrfs_alloc_dummy_sum(struct btrfs_bio *bbio)
789 bbio->sums = kmalloc(sizeof(*bbio->sums), GFP_NOFS);
791 return BLK_STS_RESOURCE;
792 bbio->sums->len = bbio->bio.bi_iter.bi_size;
793 bbio->sums->logical = bbio->bio.bi_iter.bi_sector << SECTOR_SHIFT;
794 btrfs_add_ordered_sum(bbio->ordered, bbio->sums);
799 * Remove one checksum overlapping a range.
801 * This expects the key to describe the csum pointed to by the path, and it
802 * expects the csum to overlap the range [bytenr, len]
804 * The csum should not be entirely contained in the range and the range should
805 * not be entirely contained in the csum.
807 * This calls btrfs_truncate_item with the correct args based on the overlap,
808 * and fixes up the key as required.
810 static noinline void truncate_one_csum(struct btrfs_trans_handle *trans,
811 struct btrfs_path *path,
812 struct btrfs_key *key,
815 struct btrfs_fs_info *fs_info = trans->fs_info;
816 struct extent_buffer *leaf;
817 const u32 csum_size = fs_info->csum_size;
819 u64 end_byte = bytenr + len;
820 u32 blocksize_bits = fs_info->sectorsize_bits;
822 leaf = path->nodes[0];
823 csum_end = btrfs_item_size(leaf, path->slots[0]) / csum_size;
824 csum_end <<= blocksize_bits;
825 csum_end += key->offset;
827 if (key->offset < bytenr && csum_end <= end_byte) {
832 * A simple truncate off the end of the item
834 u32 new_size = (bytenr - key->offset) >> blocksize_bits;
835 new_size *= csum_size;
836 btrfs_truncate_item(trans, path, new_size, 1);
837 } else if (key->offset >= bytenr && csum_end > end_byte &&
838 end_byte > key->offset) {
843 * we need to truncate from the beginning of the csum
845 u32 new_size = (csum_end - end_byte) >> blocksize_bits;
846 new_size *= csum_size;
848 btrfs_truncate_item(trans, path, new_size, 0);
850 key->offset = end_byte;
851 btrfs_set_item_key_safe(trans, path, key);
858 * Delete the csum items from the csum tree for a given range of bytes.
860 int btrfs_del_csums(struct btrfs_trans_handle *trans,
861 struct btrfs_root *root, u64 bytenr, u64 len)
863 struct btrfs_fs_info *fs_info = trans->fs_info;
864 struct btrfs_path *path;
865 struct btrfs_key key;
866 u64 end_byte = bytenr + len;
868 struct extent_buffer *leaf;
870 const u32 csum_size = fs_info->csum_size;
871 u32 blocksize_bits = fs_info->sectorsize_bits;
873 ASSERT(root->root_key.objectid == BTRFS_CSUM_TREE_OBJECTID ||
874 root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID);
876 path = btrfs_alloc_path();
881 key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
882 key.offset = end_byte - 1;
883 key.type = BTRFS_EXTENT_CSUM_KEY;
885 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
888 if (path->slots[0] == 0)
891 } else if (ret < 0) {
895 leaf = path->nodes[0];
896 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
898 if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
899 key.type != BTRFS_EXTENT_CSUM_KEY) {
903 if (key.offset >= end_byte)
906 csum_end = btrfs_item_size(leaf, path->slots[0]) / csum_size;
907 csum_end <<= blocksize_bits;
908 csum_end += key.offset;
910 /* this csum ends before we start, we're done */
911 if (csum_end <= bytenr)
914 /* delete the entire item, it is inside our range */
915 if (key.offset >= bytenr && csum_end <= end_byte) {
919 * Check how many csum items preceding this one in this
920 * leaf correspond to our range and then delete them all
923 if (key.offset > bytenr && path->slots[0] > 0) {
924 int slot = path->slots[0] - 1;
929 btrfs_item_key_to_cpu(leaf, &pk, slot);
930 if (pk.offset < bytenr ||
931 pk.type != BTRFS_EXTENT_CSUM_KEY ||
933 BTRFS_EXTENT_CSUM_OBJECTID)
935 path->slots[0] = slot;
937 key.offset = pk.offset;
941 ret = btrfs_del_items(trans, root, path,
942 path->slots[0], del_nr);
945 if (key.offset == bytenr)
947 } else if (key.offset < bytenr && csum_end > end_byte) {
948 unsigned long offset;
949 unsigned long shift_len;
950 unsigned long item_offset;
955 * Our bytes are in the middle of the csum,
956 * we need to split this item and insert a new one.
958 * But we can't drop the path because the
959 * csum could change, get removed, extended etc.
961 * The trick here is the max size of a csum item leaves
962 * enough room in the tree block for a single
963 * item header. So, we split the item in place,
964 * adding a new header pointing to the existing
965 * bytes. Then we loop around again and we have
966 * a nicely formed csum item that we can neatly
969 offset = (bytenr - key.offset) >> blocksize_bits;
972 shift_len = (len >> blocksize_bits) * csum_size;
974 item_offset = btrfs_item_ptr_offset(leaf,
977 memzero_extent_buffer(leaf, item_offset + offset,
982 * btrfs_split_item returns -EAGAIN when the
983 * item changed size or key
985 ret = btrfs_split_item(trans, root, path, &key, offset);
986 if (ret && ret != -EAGAIN) {
987 btrfs_abort_transaction(trans, ret);
992 key.offset = end_byte - 1;
994 truncate_one_csum(trans, path, &key, bytenr, len);
995 if (key.offset < bytenr)
998 btrfs_release_path(path);
1000 btrfs_free_path(path);
1004 static int find_next_csum_offset(struct btrfs_root *root,
1005 struct btrfs_path *path,
1008 const u32 nritems = btrfs_header_nritems(path->nodes[0]);
1009 struct btrfs_key found_key;
1010 int slot = path->slots[0] + 1;
1013 if (nritems == 0 || slot >= nritems) {
1014 ret = btrfs_next_leaf(root, path);
1017 } else if (ret > 0) {
1018 *next_offset = (u64)-1;
1021 slot = path->slots[0];
1024 btrfs_item_key_to_cpu(path->nodes[0], &found_key, slot);
1026 if (found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
1027 found_key.type != BTRFS_EXTENT_CSUM_KEY)
1028 *next_offset = (u64)-1;
1030 *next_offset = found_key.offset;
1035 int btrfs_csum_file_blocks(struct btrfs_trans_handle *trans,
1036 struct btrfs_root *root,
1037 struct btrfs_ordered_sum *sums)
1039 struct btrfs_fs_info *fs_info = root->fs_info;
1040 struct btrfs_key file_key;
1041 struct btrfs_key found_key;
1042 struct btrfs_path *path;
1043 struct btrfs_csum_item *item;
1044 struct btrfs_csum_item *item_end;
1045 struct extent_buffer *leaf = NULL;
1047 u64 total_bytes = 0;
1054 const u32 csum_size = fs_info->csum_size;
1056 path = btrfs_alloc_path();
1060 next_offset = (u64)-1;
1062 bytenr = sums->logical + total_bytes;
1063 file_key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
1064 file_key.offset = bytenr;
1065 file_key.type = BTRFS_EXTENT_CSUM_KEY;
1067 item = btrfs_lookup_csum(trans, root, path, bytenr, 1);
1068 if (!IS_ERR(item)) {
1070 leaf = path->nodes[0];
1071 item_end = btrfs_item_ptr(leaf, path->slots[0],
1072 struct btrfs_csum_item);
1073 item_end = (struct btrfs_csum_item *)((char *)item_end +
1074 btrfs_item_size(leaf, path->slots[0]));
1077 ret = PTR_ERR(item);
1078 if (ret != -EFBIG && ret != -ENOENT)
1081 if (ret == -EFBIG) {
1083 /* we found one, but it isn't big enough yet */
1084 leaf = path->nodes[0];
1085 item_size = btrfs_item_size(leaf, path->slots[0]);
1086 if ((item_size / csum_size) >=
1087 MAX_CSUM_ITEMS(fs_info, csum_size)) {
1088 /* already at max size, make a new one */
1092 /* We didn't find a csum item, insert one. */
1093 ret = find_next_csum_offset(root, path, &next_offset);
1101 * At this point, we know the tree has a checksum item that ends at an
1102 * offset matching the start of the checksum range we want to insert.
1103 * We try to extend that item as much as possible and then add as many
1104 * checksums to it as they fit.
1106 * First check if the leaf has enough free space for at least one
1107 * checksum. If it has go directly to the item extension code, otherwise
1108 * release the path and do a search for insertion before the extension.
1110 if (btrfs_leaf_free_space(leaf) >= csum_size) {
1111 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1112 csum_offset = (bytenr - found_key.offset) >>
1113 fs_info->sectorsize_bits;
1117 btrfs_release_path(path);
1118 path->search_for_extension = 1;
1119 ret = btrfs_search_slot(trans, root, &file_key, path,
1121 path->search_for_extension = 0;
1126 if (path->slots[0] == 0)
1131 leaf = path->nodes[0];
1132 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1133 csum_offset = (bytenr - found_key.offset) >> fs_info->sectorsize_bits;
1135 if (found_key.type != BTRFS_EXTENT_CSUM_KEY ||
1136 found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
1137 csum_offset >= MAX_CSUM_ITEMS(fs_info, csum_size)) {
1142 if (csum_offset == btrfs_item_size(leaf, path->slots[0]) /
1148 tmp = sums->len - total_bytes;
1149 tmp >>= fs_info->sectorsize_bits;
1151 extend_nr = max_t(int, 1, tmp);
1154 * A log tree can already have checksum items with a subset of
1155 * the checksums we are trying to log. This can happen after
1156 * doing a sequence of partial writes into prealloc extents and
1157 * fsyncs in between, with a full fsync logging a larger subrange
1158 * of an extent for which a previous fast fsync logged a smaller
1159 * subrange. And this happens in particular due to merging file
1160 * extent items when we complete an ordered extent for a range
1161 * covered by a prealloc extent - this is done at
1162 * btrfs_mark_extent_written().
1164 * So if we try to extend the previous checksum item, which has
1165 * a range that ends at the start of the range we want to insert,
1166 * make sure we don't extend beyond the start offset of the next
1167 * checksum item. If we are at the last item in the leaf, then
1168 * forget the optimization of extending and add a new checksum
1169 * item - it is not worth the complexity of releasing the path,
1170 * getting the first key for the next leaf, repeat the btree
1171 * search, etc, because log trees are temporary anyway and it
1172 * would only save a few bytes of leaf space.
1174 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
1175 if (path->slots[0] + 1 >=
1176 btrfs_header_nritems(path->nodes[0])) {
1177 ret = find_next_csum_offset(root, path, &next_offset);
1184 ret = find_next_csum_offset(root, path, &next_offset);
1188 tmp = (next_offset - bytenr) >> fs_info->sectorsize_bits;
1190 extend_nr = min_t(int, extend_nr, tmp);
1193 diff = (csum_offset + extend_nr) * csum_size;
1195 MAX_CSUM_ITEMS(fs_info, csum_size) * csum_size);
1197 diff = diff - btrfs_item_size(leaf, path->slots[0]);
1198 diff = min_t(u32, btrfs_leaf_free_space(leaf), diff);
1202 btrfs_extend_item(trans, path, diff);
1208 btrfs_release_path(path);
1213 tmp = sums->len - total_bytes;
1214 tmp >>= fs_info->sectorsize_bits;
1215 tmp = min(tmp, (next_offset - file_key.offset) >>
1216 fs_info->sectorsize_bits);
1218 tmp = max_t(u64, 1, tmp);
1219 tmp = min_t(u64, tmp, MAX_CSUM_ITEMS(fs_info, csum_size));
1220 ins_size = csum_size * tmp;
1222 ins_size = csum_size;
1224 ret = btrfs_insert_empty_item(trans, root, path, &file_key,
1228 leaf = path->nodes[0];
1230 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item);
1231 item_end = (struct btrfs_csum_item *)((unsigned char *)item +
1232 btrfs_item_size(leaf, path->slots[0]));
1233 item = (struct btrfs_csum_item *)((unsigned char *)item +
1234 csum_offset * csum_size);
1236 ins_size = (u32)(sums->len - total_bytes) >> fs_info->sectorsize_bits;
1237 ins_size *= csum_size;
1238 ins_size = min_t(u32, (unsigned long)item_end - (unsigned long)item,
1240 write_extent_buffer(leaf, sums->sums + index, (unsigned long)item,
1244 ins_size /= csum_size;
1245 total_bytes += ins_size * fs_info->sectorsize;
1247 btrfs_mark_buffer_dirty(trans, path->nodes[0]);
1248 if (total_bytes < sums->len) {
1249 btrfs_release_path(path);
1254 btrfs_free_path(path);
1258 void btrfs_extent_item_to_extent_map(struct btrfs_inode *inode,
1259 const struct btrfs_path *path,
1260 struct btrfs_file_extent_item *fi,
1261 struct extent_map *em)
1263 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1264 struct btrfs_root *root = inode->root;
1265 struct extent_buffer *leaf = path->nodes[0];
1266 const int slot = path->slots[0];
1267 struct btrfs_key key;
1268 u64 extent_start, extent_end;
1270 u8 type = btrfs_file_extent_type(leaf, fi);
1271 int compress_type = btrfs_file_extent_compression(leaf, fi);
1273 btrfs_item_key_to_cpu(leaf, &key, slot);
1274 extent_start = key.offset;
1275 extent_end = btrfs_file_extent_end(path);
1276 em->ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1277 em->generation = btrfs_file_extent_generation(leaf, fi);
1278 if (type == BTRFS_FILE_EXTENT_REG ||
1279 type == BTRFS_FILE_EXTENT_PREALLOC) {
1280 em->start = extent_start;
1281 em->len = extent_end - extent_start;
1282 em->orig_start = extent_start -
1283 btrfs_file_extent_offset(leaf, fi);
1284 em->orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
1285 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1287 em->block_start = EXTENT_MAP_HOLE;
1290 if (compress_type != BTRFS_COMPRESS_NONE) {
1291 extent_map_set_compression(em, compress_type);
1292 em->block_start = bytenr;
1293 em->block_len = em->orig_block_len;
1295 bytenr += btrfs_file_extent_offset(leaf, fi);
1296 em->block_start = bytenr;
1297 em->block_len = em->len;
1298 if (type == BTRFS_FILE_EXTENT_PREALLOC)
1299 em->flags |= EXTENT_FLAG_PREALLOC;
1301 } else if (type == BTRFS_FILE_EXTENT_INLINE) {
1302 em->block_start = EXTENT_MAP_INLINE;
1303 em->start = extent_start;
1304 em->len = extent_end - extent_start;
1306 * Initialize orig_start and block_len with the same values
1307 * as in inode.c:btrfs_get_extent().
1309 em->orig_start = EXTENT_MAP_HOLE;
1310 em->block_len = (u64)-1;
1311 extent_map_set_compression(em, compress_type);
1314 "unknown file extent item type %d, inode %llu, offset %llu, "
1315 "root %llu", type, btrfs_ino(inode), extent_start,
1316 root->root_key.objectid);
1321 * Returns the end offset (non inclusive) of the file extent item the given path
1322 * points to. If it points to an inline extent, the returned offset is rounded
1323 * up to the sector size.
1325 u64 btrfs_file_extent_end(const struct btrfs_path *path)
1327 const struct extent_buffer *leaf = path->nodes[0];
1328 const int slot = path->slots[0];
1329 struct btrfs_file_extent_item *fi;
1330 struct btrfs_key key;
1333 btrfs_item_key_to_cpu(leaf, &key, slot);
1334 ASSERT(key.type == BTRFS_EXTENT_DATA_KEY);
1335 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
1337 if (btrfs_file_extent_type(leaf, fi) == BTRFS_FILE_EXTENT_INLINE) {
1338 end = btrfs_file_extent_ram_bytes(leaf, fi);
1339 end = ALIGN(key.offset + end, leaf->fs_info->sectorsize);
1341 end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);