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>
16 #include "transaction.h"
18 #include "print-tree.h"
19 #include "compression.h"
21 #include "accessors.h"
22 #include "file-item.h"
25 #define __MAX_CSUM_ITEMS(r, size) ((unsigned long)(((BTRFS_LEAF_DATA_SIZE(r) - \
26 sizeof(struct btrfs_item) * 2) / \
29 #define MAX_CSUM_ITEMS(r, size) (min_t(u32, __MAX_CSUM_ITEMS(r, size), \
33 * Set inode's size according to filesystem options.
35 * @inode: inode we want to update the disk_i_size for
36 * @new_i_size: i_size we want to set to, 0 if we use i_size
38 * With NO_HOLES set this simply sets the disk_is_size to whatever i_size_read()
39 * returns as it is perfectly fine with a file that has holes without hole file
42 * However without NO_HOLES we need to only return the area that is contiguous
43 * from the 0 offset of the file. Otherwise we could end up adjust i_size up
44 * to an extent that has a gap in between.
46 * Finally new_i_size should only be set in the case of truncate where we're not
47 * ready to use i_size_read() as the limiter yet.
49 void btrfs_inode_safe_disk_i_size_write(struct btrfs_inode *inode, u64 new_i_size)
51 struct btrfs_fs_info *fs_info = inode->root->fs_info;
52 u64 start, end, i_size;
55 spin_lock(&inode->lock);
56 i_size = new_i_size ?: i_size_read(&inode->vfs_inode);
57 if (btrfs_fs_incompat(fs_info, NO_HOLES)) {
58 inode->disk_i_size = i_size;
62 ret = find_contiguous_extent_bit(&inode->file_extent_tree, 0, &start,
64 if (!ret && start == 0)
65 i_size = min(i_size, end + 1);
68 inode->disk_i_size = i_size;
70 spin_unlock(&inode->lock);
74 * Mark range within a file as having a new extent inserted.
76 * @inode: inode being modified
77 * @start: start file offset of the file extent we've inserted
78 * @len: logical length of the file extent item
80 * Call when we are inserting a new file extent where there was none before.
81 * Does not need to call this in the case where we're replacing an existing file
82 * extent, however if not sure it's fine to call this multiple times.
84 * The start and len must match the file extent item, so thus must be sectorsize
87 int btrfs_inode_set_file_extent_range(struct btrfs_inode *inode, u64 start,
93 ASSERT(IS_ALIGNED(start + len, inode->root->fs_info->sectorsize));
95 if (btrfs_fs_incompat(inode->root->fs_info, NO_HOLES))
97 return set_extent_bits(&inode->file_extent_tree, start, start + len - 1,
102 * Mark an inode range as not having a backing extent.
104 * @inode: inode being modified
105 * @start: start file offset of the file extent we've inserted
106 * @len: logical length of the file extent item
108 * Called when we drop a file extent, for example when we truncate. Doesn't
109 * need to be called for cases where we're replacing a file extent, like when
110 * we've COWed a file extent.
112 * The start and len must match the file extent item, so thus must be sectorsize
115 int btrfs_inode_clear_file_extent_range(struct btrfs_inode *inode, u64 start,
121 ASSERT(IS_ALIGNED(start + len, inode->root->fs_info->sectorsize) ||
124 if (btrfs_fs_incompat(inode->root->fs_info, NO_HOLES))
126 return clear_extent_bit(&inode->file_extent_tree, start,
127 start + len - 1, EXTENT_DIRTY, NULL);
130 static size_t bytes_to_csum_size(const struct btrfs_fs_info *fs_info, u32 bytes)
132 ASSERT(IS_ALIGNED(bytes, fs_info->sectorsize));
134 return (bytes >> fs_info->sectorsize_bits) * fs_info->csum_size;
137 static size_t csum_size_to_bytes(const struct btrfs_fs_info *fs_info, u32 csum_size)
139 ASSERT(IS_ALIGNED(csum_size, fs_info->csum_size));
141 return (csum_size / fs_info->csum_size) << fs_info->sectorsize_bits;
144 static inline u32 max_ordered_sum_bytes(const struct btrfs_fs_info *fs_info)
146 u32 max_csum_size = round_down(PAGE_SIZE - sizeof(struct btrfs_ordered_sum),
149 return csum_size_to_bytes(fs_info, max_csum_size);
153 * Calculate the total size needed to allocate for an ordered sum structure
154 * spanning @bytes in the file.
156 static int btrfs_ordered_sum_size(struct btrfs_fs_info *fs_info, unsigned long bytes)
158 return sizeof(struct btrfs_ordered_sum) + bytes_to_csum_size(fs_info, bytes);
161 int btrfs_insert_hole_extent(struct btrfs_trans_handle *trans,
162 struct btrfs_root *root,
163 u64 objectid, u64 pos, u64 num_bytes)
166 struct btrfs_file_extent_item *item;
167 struct btrfs_key file_key;
168 struct btrfs_path *path;
169 struct extent_buffer *leaf;
171 path = btrfs_alloc_path();
174 file_key.objectid = objectid;
175 file_key.offset = pos;
176 file_key.type = BTRFS_EXTENT_DATA_KEY;
178 ret = btrfs_insert_empty_item(trans, root, path, &file_key,
182 BUG_ON(ret); /* Can't happen */
183 leaf = path->nodes[0];
184 item = btrfs_item_ptr(leaf, path->slots[0],
185 struct btrfs_file_extent_item);
186 btrfs_set_file_extent_disk_bytenr(leaf, item, 0);
187 btrfs_set_file_extent_disk_num_bytes(leaf, item, 0);
188 btrfs_set_file_extent_offset(leaf, item, 0);
189 btrfs_set_file_extent_num_bytes(leaf, item, num_bytes);
190 btrfs_set_file_extent_ram_bytes(leaf, item, num_bytes);
191 btrfs_set_file_extent_generation(leaf, item, trans->transid);
192 btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
193 btrfs_set_file_extent_compression(leaf, item, 0);
194 btrfs_set_file_extent_encryption(leaf, item, 0);
195 btrfs_set_file_extent_other_encoding(leaf, item, 0);
197 btrfs_mark_buffer_dirty(leaf);
199 btrfs_free_path(path);
203 static struct btrfs_csum_item *
204 btrfs_lookup_csum(struct btrfs_trans_handle *trans,
205 struct btrfs_root *root,
206 struct btrfs_path *path,
209 struct btrfs_fs_info *fs_info = root->fs_info;
211 struct btrfs_key file_key;
212 struct btrfs_key found_key;
213 struct btrfs_csum_item *item;
214 struct extent_buffer *leaf;
216 const u32 csum_size = fs_info->csum_size;
219 file_key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
220 file_key.offset = bytenr;
221 file_key.type = BTRFS_EXTENT_CSUM_KEY;
222 ret = btrfs_search_slot(trans, root, &file_key, path, 0, cow);
225 leaf = path->nodes[0];
228 if (path->slots[0] == 0)
231 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
232 if (found_key.type != BTRFS_EXTENT_CSUM_KEY)
235 csum_offset = (bytenr - found_key.offset) >>
236 fs_info->sectorsize_bits;
237 csums_in_item = btrfs_item_size(leaf, path->slots[0]);
238 csums_in_item /= csum_size;
240 if (csum_offset == csums_in_item) {
243 } else if (csum_offset > csums_in_item) {
247 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item);
248 item = (struct btrfs_csum_item *)((unsigned char *)item +
249 csum_offset * csum_size);
257 int btrfs_lookup_file_extent(struct btrfs_trans_handle *trans,
258 struct btrfs_root *root,
259 struct btrfs_path *path, u64 objectid,
262 struct btrfs_key file_key;
263 int ins_len = mod < 0 ? -1 : 0;
266 file_key.objectid = objectid;
267 file_key.offset = offset;
268 file_key.type = BTRFS_EXTENT_DATA_KEY;
270 return btrfs_search_slot(trans, root, &file_key, path, ins_len, cow);
274 * Find checksums for logical bytenr range [disk_bytenr, disk_bytenr + len) and
275 * store the result to @dst.
277 * Return >0 for the number of sectors we found.
278 * Return 0 for the range [disk_bytenr, disk_bytenr + sectorsize) has no csum
279 * for it. Caller may want to try next sector until one range is hit.
280 * Return <0 for fatal error.
282 static int search_csum_tree(struct btrfs_fs_info *fs_info,
283 struct btrfs_path *path, u64 disk_bytenr,
286 struct btrfs_root *csum_root;
287 struct btrfs_csum_item *item = NULL;
288 struct btrfs_key key;
289 const u32 sectorsize = fs_info->sectorsize;
290 const u32 csum_size = fs_info->csum_size;
296 ASSERT(IS_ALIGNED(disk_bytenr, sectorsize) &&
297 IS_ALIGNED(len, sectorsize));
299 /* Check if the current csum item covers disk_bytenr */
300 if (path->nodes[0]) {
301 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
302 struct btrfs_csum_item);
303 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
304 itemsize = btrfs_item_size(path->nodes[0], path->slots[0]);
306 csum_start = key.offset;
307 csum_len = (itemsize / csum_size) * sectorsize;
309 if (in_range(disk_bytenr, csum_start, csum_len))
313 /* Current item doesn't contain the desired range, search again */
314 btrfs_release_path(path);
315 csum_root = btrfs_csum_root(fs_info, disk_bytenr);
316 item = btrfs_lookup_csum(NULL, csum_root, path, disk_bytenr, 0);
321 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
322 itemsize = btrfs_item_size(path->nodes[0], path->slots[0]);
324 csum_start = key.offset;
325 csum_len = (itemsize / csum_size) * sectorsize;
326 ASSERT(in_range(disk_bytenr, csum_start, csum_len));
329 ret = (min(csum_start + csum_len, disk_bytenr + len) -
330 disk_bytenr) >> fs_info->sectorsize_bits;
331 read_extent_buffer(path->nodes[0], dst, (unsigned long)item,
334 if (ret == -ENOENT || ret == -EFBIG)
340 * Lookup the checksum for the read bio in csum tree.
342 * Return: BLK_STS_RESOURCE if allocating memory fails, BLK_STS_OK otherwise.
344 blk_status_t btrfs_lookup_bio_sums(struct btrfs_bio *bbio)
346 struct btrfs_inode *inode = bbio->inode;
347 struct btrfs_fs_info *fs_info = inode->root->fs_info;
348 struct bio *bio = &bbio->bio;
349 struct btrfs_path *path;
350 const u32 sectorsize = fs_info->sectorsize;
351 const u32 csum_size = fs_info->csum_size;
352 u32 orig_len = bio->bi_iter.bi_size;
353 u64 orig_disk_bytenr = bio->bi_iter.bi_sector << SECTOR_SHIFT;
354 const unsigned int nblocks = orig_len >> fs_info->sectorsize_bits;
355 blk_status_t ret = BLK_STS_OK;
358 if ((inode->flags & BTRFS_INODE_NODATASUM) ||
359 test_bit(BTRFS_FS_STATE_NO_CSUMS, &fs_info->fs_state))
363 * This function is only called for read bio.
365 * This means two things:
366 * - All our csums should only be in csum tree
367 * No ordered extents csums, as ordered extents are only for write
369 * - No need to bother any other info from bvec
370 * Since we're looking up csums, the only important info is the
371 * disk_bytenr and the length, which can be extracted from bi_iter
374 ASSERT(bio_op(bio) == REQ_OP_READ);
375 path = btrfs_alloc_path();
377 return BLK_STS_RESOURCE;
379 if (nblocks * csum_size > BTRFS_BIO_INLINE_CSUM_SIZE) {
380 bbio->csum = kmalloc_array(nblocks, csum_size, GFP_NOFS);
382 btrfs_free_path(path);
383 return BLK_STS_RESOURCE;
386 bbio->csum = bbio->csum_inline;
390 * If requested number of sectors is larger than one leaf can contain,
391 * kick the readahead for csum tree.
393 if (nblocks > fs_info->csums_per_leaf)
394 path->reada = READA_FORWARD;
397 * the free space stuff is only read when it hasn't been
398 * updated in the current transaction. So, we can safely
399 * read from the commit root and sidestep a nasty deadlock
400 * between reading the free space cache and updating the csum tree.
402 if (btrfs_is_free_space_inode(inode)) {
403 path->search_commit_root = 1;
404 path->skip_locking = 1;
407 while (bio_offset < orig_len) {
409 u64 cur_disk_bytenr = orig_disk_bytenr + bio_offset;
410 u8 *csum_dst = bbio->csum +
411 (bio_offset >> fs_info->sectorsize_bits) * csum_size;
413 count = search_csum_tree(fs_info, path, cur_disk_bytenr,
414 orig_len - bio_offset, csum_dst);
416 ret = errno_to_blk_status(count);
417 if (bbio->csum != bbio->csum_inline)
424 * We didn't find a csum for this range. We need to make sure
425 * we complain loudly about this, because we are not NODATASUM.
427 * However for the DATA_RELOC inode we could potentially be
428 * relocating data extents for a NODATASUM inode, so the inode
429 * itself won't be marked with NODATASUM, but the extent we're
430 * copying is in fact NODATASUM. If we don't find a csum we
431 * assume this is the case.
434 memset(csum_dst, 0, csum_size);
437 if (inode->root->root_key.objectid ==
438 BTRFS_DATA_RELOC_TREE_OBJECTID) {
439 u64 file_offset = bbio->file_offset + bio_offset;
441 set_extent_bits(&inode->io_tree, file_offset,
442 file_offset + sectorsize - 1,
445 btrfs_warn_rl(fs_info,
446 "csum hole found for disk bytenr range [%llu, %llu)",
447 cur_disk_bytenr, cur_disk_bytenr + sectorsize);
450 bio_offset += count * sectorsize;
453 btrfs_free_path(path);
457 int btrfs_lookup_csums_list(struct btrfs_root *root, u64 start, u64 end,
458 struct list_head *list, int search_commit,
461 struct btrfs_fs_info *fs_info = root->fs_info;
462 struct btrfs_key key;
463 struct btrfs_path *path;
464 struct extent_buffer *leaf;
465 struct btrfs_ordered_sum *sums;
466 struct btrfs_csum_item *item;
470 ASSERT(IS_ALIGNED(start, fs_info->sectorsize) &&
471 IS_ALIGNED(end + 1, fs_info->sectorsize));
473 path = btrfs_alloc_path();
477 path->nowait = nowait;
479 path->skip_locking = 1;
480 path->reada = READA_FORWARD;
481 path->search_commit_root = 1;
484 key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
486 key.type = BTRFS_EXTENT_CSUM_KEY;
488 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
491 if (ret > 0 && path->slots[0] > 0) {
492 leaf = path->nodes[0];
493 btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
496 * There are two cases we can hit here for the previous csum
499 * |<- search range ->|
503 * |<- search range ->|
506 * Check if the previous csum item covers the leading part of
507 * the search range. If so we have to start from previous csum
510 if (key.objectid == BTRFS_EXTENT_CSUM_OBJECTID &&
511 key.type == BTRFS_EXTENT_CSUM_KEY) {
512 if (bytes_to_csum_size(fs_info, start - key.offset) <
513 btrfs_item_size(leaf, path->slots[0] - 1))
518 while (start <= end) {
521 leaf = path->nodes[0];
522 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
523 ret = btrfs_next_leaf(root, path);
528 leaf = path->nodes[0];
531 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
532 if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
533 key.type != BTRFS_EXTENT_CSUM_KEY ||
537 if (key.offset > start)
540 csum_end = key.offset + csum_size_to_bytes(fs_info,
541 btrfs_item_size(leaf, path->slots[0]));
542 if (csum_end <= start) {
547 csum_end = min(csum_end, end + 1);
548 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
549 struct btrfs_csum_item);
550 while (start < csum_end) {
551 unsigned long offset;
554 size = min_t(size_t, csum_end - start,
555 max_ordered_sum_bytes(fs_info));
556 sums = kzalloc(btrfs_ordered_sum_size(fs_info, size),
563 sums->bytenr = start;
564 sums->len = (int)size;
566 offset = bytes_to_csum_size(fs_info, start - key.offset);
568 read_extent_buffer(path->nodes[0],
570 ((unsigned long)item) + offset,
571 bytes_to_csum_size(fs_info, size));
574 list_add_tail(&sums->list, &tmplist);
580 while (ret < 0 && !list_empty(&tmplist)) {
581 sums = list_entry(tmplist.next, struct btrfs_ordered_sum, list);
582 list_del(&sums->list);
585 list_splice_tail(&tmplist, list);
587 btrfs_free_path(path);
592 * Do the same work as btrfs_lookup_csums_list(), the difference is in how
593 * we return the result.
595 * This version will set the corresponding bits in @csum_bitmap to represent
596 * that there is a csum found.
597 * Each bit represents a sector. Thus caller should ensure @csum_buf passed
598 * in is large enough to contain all csums.
600 int btrfs_lookup_csums_bitmap(struct btrfs_root *root, u64 start, u64 end,
601 u8 *csum_buf, unsigned long *csum_bitmap,
604 struct btrfs_fs_info *fs_info = root->fs_info;
605 struct btrfs_key key;
606 struct btrfs_path *path;
607 struct extent_buffer *leaf;
608 struct btrfs_csum_item *item;
609 const u64 orig_start = start;
612 ASSERT(IS_ALIGNED(start, fs_info->sectorsize) &&
613 IS_ALIGNED(end + 1, fs_info->sectorsize));
615 path = btrfs_alloc_path();
620 path->skip_locking = 1;
621 path->reada = READA_FORWARD;
622 path->search_commit_root = 1;
625 key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
626 key.type = BTRFS_EXTENT_CSUM_KEY;
629 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
632 if (ret > 0 && path->slots[0] > 0) {
633 leaf = path->nodes[0];
634 btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
637 * There are two cases we can hit here for the previous csum
640 * |<- search range ->|
644 * |<- search range ->|
647 * Check if the previous csum item covers the leading part of
648 * the search range. If so we have to start from previous csum
651 if (key.objectid == BTRFS_EXTENT_CSUM_OBJECTID &&
652 key.type == BTRFS_EXTENT_CSUM_KEY) {
653 if (bytes_to_csum_size(fs_info, start - key.offset) <
654 btrfs_item_size(leaf, path->slots[0] - 1))
659 while (start <= end) {
662 leaf = path->nodes[0];
663 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
664 ret = btrfs_next_leaf(root, path);
669 leaf = path->nodes[0];
672 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
673 if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
674 key.type != BTRFS_EXTENT_CSUM_KEY ||
678 if (key.offset > start)
681 csum_end = key.offset + csum_size_to_bytes(fs_info,
682 btrfs_item_size(leaf, path->slots[0]));
683 if (csum_end <= start) {
688 csum_end = min(csum_end, end + 1);
689 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
690 struct btrfs_csum_item);
691 while (start < csum_end) {
692 unsigned long offset;
694 u8 *csum_dest = csum_buf + bytes_to_csum_size(fs_info,
697 size = min_t(size_t, csum_end - start, end + 1 - start);
699 offset = bytes_to_csum_size(fs_info, start - key.offset);
701 read_extent_buffer(path->nodes[0], csum_dest,
702 ((unsigned long)item) + offset,
703 bytes_to_csum_size(fs_info, size));
705 bitmap_set(csum_bitmap,
706 (start - orig_start) >> fs_info->sectorsize_bits,
707 size >> fs_info->sectorsize_bits);
715 btrfs_free_path(path);
720 * Calculate checksums of the data contained inside a bio.
722 blk_status_t btrfs_csum_one_bio(struct btrfs_bio *bbio)
724 struct btrfs_inode *inode = bbio->inode;
725 struct btrfs_fs_info *fs_info = inode->root->fs_info;
726 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
727 struct bio *bio = &bbio->bio;
728 u64 offset = bbio->file_offset;
729 struct btrfs_ordered_sum *sums;
730 struct btrfs_ordered_extent *ordered = NULL;
732 struct bvec_iter iter;
735 unsigned int blockcount;
736 unsigned long total_bytes = 0;
737 unsigned long this_sum_bytes = 0;
741 nofs_flag = memalloc_nofs_save();
742 sums = kvzalloc(btrfs_ordered_sum_size(fs_info, bio->bi_iter.bi_size),
744 memalloc_nofs_restore(nofs_flag);
747 return BLK_STS_RESOURCE;
749 sums->len = bio->bi_iter.bi_size;
750 INIT_LIST_HEAD(&sums->list);
752 sums->bytenr = bio->bi_iter.bi_sector << 9;
755 shash->tfm = fs_info->csum_shash;
757 bio_for_each_segment(bvec, bio, iter) {
759 ordered = btrfs_lookup_ordered_extent(inode, offset);
761 * The bio range is not covered by any ordered extent,
762 * must be a code logic error.
764 if (unlikely(!ordered)) {
766 "no ordered extent for root %llu ino %llu offset %llu\n",
767 inode->root->root_key.objectid,
768 btrfs_ino(inode), offset);
770 return BLK_STS_IOERR;
774 blockcount = BTRFS_BYTES_TO_BLKS(fs_info,
775 bvec.bv_len + fs_info->sectorsize
778 for (i = 0; i < blockcount; i++) {
779 if (!(bio->bi_opf & REQ_BTRFS_ONE_ORDERED) &&
780 !in_range(offset, ordered->file_offset,
781 ordered->num_bytes)) {
782 unsigned long bytes_left;
784 sums->len = this_sum_bytes;
786 btrfs_add_ordered_sum(ordered, sums);
787 btrfs_put_ordered_extent(ordered);
789 bytes_left = bio->bi_iter.bi_size - total_bytes;
791 nofs_flag = memalloc_nofs_save();
792 sums = kvzalloc(btrfs_ordered_sum_size(fs_info,
793 bytes_left), GFP_KERNEL);
794 memalloc_nofs_restore(nofs_flag);
795 BUG_ON(!sums); /* -ENOMEM */
796 sums->len = bytes_left;
797 ordered = btrfs_lookup_ordered_extent(inode,
799 ASSERT(ordered); /* Logic error */
800 sums->bytenr = (bio->bi_iter.bi_sector << 9)
805 data = bvec_kmap_local(&bvec);
806 crypto_shash_digest(shash,
807 data + (i * fs_info->sectorsize),
811 index += fs_info->csum_size;
812 offset += fs_info->sectorsize;
813 this_sum_bytes += fs_info->sectorsize;
814 total_bytes += fs_info->sectorsize;
819 btrfs_add_ordered_sum(ordered, sums);
820 btrfs_put_ordered_extent(ordered);
825 * Remove one checksum overlapping a range.
827 * This expects the key to describe the csum pointed to by the path, and it
828 * expects the csum to overlap the range [bytenr, len]
830 * The csum should not be entirely contained in the range and the range should
831 * not be entirely contained in the csum.
833 * This calls btrfs_truncate_item with the correct args based on the overlap,
834 * and fixes up the key as required.
836 static noinline void truncate_one_csum(struct btrfs_fs_info *fs_info,
837 struct btrfs_path *path,
838 struct btrfs_key *key,
841 struct extent_buffer *leaf;
842 const u32 csum_size = fs_info->csum_size;
844 u64 end_byte = bytenr + len;
845 u32 blocksize_bits = fs_info->sectorsize_bits;
847 leaf = path->nodes[0];
848 csum_end = btrfs_item_size(leaf, path->slots[0]) / csum_size;
849 csum_end <<= blocksize_bits;
850 csum_end += key->offset;
852 if (key->offset < bytenr && csum_end <= end_byte) {
857 * A simple truncate off the end of the item
859 u32 new_size = (bytenr - key->offset) >> blocksize_bits;
860 new_size *= csum_size;
861 btrfs_truncate_item(path, new_size, 1);
862 } else if (key->offset >= bytenr && csum_end > end_byte &&
863 end_byte > key->offset) {
868 * we need to truncate from the beginning of the csum
870 u32 new_size = (csum_end - end_byte) >> blocksize_bits;
871 new_size *= csum_size;
873 btrfs_truncate_item(path, new_size, 0);
875 key->offset = end_byte;
876 btrfs_set_item_key_safe(fs_info, path, key);
883 * Delete the csum items from the csum tree for a given range of bytes.
885 int btrfs_del_csums(struct btrfs_trans_handle *trans,
886 struct btrfs_root *root, u64 bytenr, u64 len)
888 struct btrfs_fs_info *fs_info = trans->fs_info;
889 struct btrfs_path *path;
890 struct btrfs_key key;
891 u64 end_byte = bytenr + len;
893 struct extent_buffer *leaf;
895 const u32 csum_size = fs_info->csum_size;
896 u32 blocksize_bits = fs_info->sectorsize_bits;
898 ASSERT(root->root_key.objectid == BTRFS_CSUM_TREE_OBJECTID ||
899 root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID);
901 path = btrfs_alloc_path();
906 key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
907 key.offset = end_byte - 1;
908 key.type = BTRFS_EXTENT_CSUM_KEY;
910 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
913 if (path->slots[0] == 0)
916 } else if (ret < 0) {
920 leaf = path->nodes[0];
921 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
923 if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
924 key.type != BTRFS_EXTENT_CSUM_KEY) {
928 if (key.offset >= end_byte)
931 csum_end = btrfs_item_size(leaf, path->slots[0]) / csum_size;
932 csum_end <<= blocksize_bits;
933 csum_end += key.offset;
935 /* this csum ends before we start, we're done */
936 if (csum_end <= bytenr)
939 /* delete the entire item, it is inside our range */
940 if (key.offset >= bytenr && csum_end <= end_byte) {
944 * Check how many csum items preceding this one in this
945 * leaf correspond to our range and then delete them all
948 if (key.offset > bytenr && path->slots[0] > 0) {
949 int slot = path->slots[0] - 1;
954 btrfs_item_key_to_cpu(leaf, &pk, slot);
955 if (pk.offset < bytenr ||
956 pk.type != BTRFS_EXTENT_CSUM_KEY ||
958 BTRFS_EXTENT_CSUM_OBJECTID)
960 path->slots[0] = slot;
962 key.offset = pk.offset;
966 ret = btrfs_del_items(trans, root, path,
967 path->slots[0], del_nr);
970 if (key.offset == bytenr)
972 } else if (key.offset < bytenr && csum_end > end_byte) {
973 unsigned long offset;
974 unsigned long shift_len;
975 unsigned long item_offset;
980 * Our bytes are in the middle of the csum,
981 * we need to split this item and insert a new one.
983 * But we can't drop the path because the
984 * csum could change, get removed, extended etc.
986 * The trick here is the max size of a csum item leaves
987 * enough room in the tree block for a single
988 * item header. So, we split the item in place,
989 * adding a new header pointing to the existing
990 * bytes. Then we loop around again and we have
991 * a nicely formed csum item that we can neatly
994 offset = (bytenr - key.offset) >> blocksize_bits;
997 shift_len = (len >> blocksize_bits) * csum_size;
999 item_offset = btrfs_item_ptr_offset(leaf,
1002 memzero_extent_buffer(leaf, item_offset + offset,
1004 key.offset = bytenr;
1007 * btrfs_split_item returns -EAGAIN when the
1008 * item changed size or key
1010 ret = btrfs_split_item(trans, root, path, &key, offset);
1011 if (ret && ret != -EAGAIN) {
1012 btrfs_abort_transaction(trans, ret);
1017 key.offset = end_byte - 1;
1019 truncate_one_csum(fs_info, path, &key, bytenr, len);
1020 if (key.offset < bytenr)
1023 btrfs_release_path(path);
1025 btrfs_free_path(path);
1029 static int find_next_csum_offset(struct btrfs_root *root,
1030 struct btrfs_path *path,
1033 const u32 nritems = btrfs_header_nritems(path->nodes[0]);
1034 struct btrfs_key found_key;
1035 int slot = path->slots[0] + 1;
1038 if (nritems == 0 || slot >= nritems) {
1039 ret = btrfs_next_leaf(root, path);
1042 } else if (ret > 0) {
1043 *next_offset = (u64)-1;
1046 slot = path->slots[0];
1049 btrfs_item_key_to_cpu(path->nodes[0], &found_key, slot);
1051 if (found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
1052 found_key.type != BTRFS_EXTENT_CSUM_KEY)
1053 *next_offset = (u64)-1;
1055 *next_offset = found_key.offset;
1060 int btrfs_csum_file_blocks(struct btrfs_trans_handle *trans,
1061 struct btrfs_root *root,
1062 struct btrfs_ordered_sum *sums)
1064 struct btrfs_fs_info *fs_info = root->fs_info;
1065 struct btrfs_key file_key;
1066 struct btrfs_key found_key;
1067 struct btrfs_path *path;
1068 struct btrfs_csum_item *item;
1069 struct btrfs_csum_item *item_end;
1070 struct extent_buffer *leaf = NULL;
1072 u64 total_bytes = 0;
1079 const u32 csum_size = fs_info->csum_size;
1081 path = btrfs_alloc_path();
1085 next_offset = (u64)-1;
1087 bytenr = sums->bytenr + total_bytes;
1088 file_key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
1089 file_key.offset = bytenr;
1090 file_key.type = BTRFS_EXTENT_CSUM_KEY;
1092 item = btrfs_lookup_csum(trans, root, path, bytenr, 1);
1093 if (!IS_ERR(item)) {
1095 leaf = path->nodes[0];
1096 item_end = btrfs_item_ptr(leaf, path->slots[0],
1097 struct btrfs_csum_item);
1098 item_end = (struct btrfs_csum_item *)((char *)item_end +
1099 btrfs_item_size(leaf, path->slots[0]));
1102 ret = PTR_ERR(item);
1103 if (ret != -EFBIG && ret != -ENOENT)
1106 if (ret == -EFBIG) {
1108 /* we found one, but it isn't big enough yet */
1109 leaf = path->nodes[0];
1110 item_size = btrfs_item_size(leaf, path->slots[0]);
1111 if ((item_size / csum_size) >=
1112 MAX_CSUM_ITEMS(fs_info, csum_size)) {
1113 /* already at max size, make a new one */
1117 /* We didn't find a csum item, insert one. */
1118 ret = find_next_csum_offset(root, path, &next_offset);
1126 * At this point, we know the tree has a checksum item that ends at an
1127 * offset matching the start of the checksum range we want to insert.
1128 * We try to extend that item as much as possible and then add as many
1129 * checksums to it as they fit.
1131 * First check if the leaf has enough free space for at least one
1132 * checksum. If it has go directly to the item extension code, otherwise
1133 * release the path and do a search for insertion before the extension.
1135 if (btrfs_leaf_free_space(leaf) >= csum_size) {
1136 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1137 csum_offset = (bytenr - found_key.offset) >>
1138 fs_info->sectorsize_bits;
1142 btrfs_release_path(path);
1143 path->search_for_extension = 1;
1144 ret = btrfs_search_slot(trans, root, &file_key, path,
1146 path->search_for_extension = 0;
1151 if (path->slots[0] == 0)
1156 leaf = path->nodes[0];
1157 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1158 csum_offset = (bytenr - found_key.offset) >> fs_info->sectorsize_bits;
1160 if (found_key.type != BTRFS_EXTENT_CSUM_KEY ||
1161 found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
1162 csum_offset >= MAX_CSUM_ITEMS(fs_info, csum_size)) {
1167 if (csum_offset == btrfs_item_size(leaf, path->slots[0]) /
1173 tmp = sums->len - total_bytes;
1174 tmp >>= fs_info->sectorsize_bits;
1176 extend_nr = max_t(int, 1, tmp);
1179 * A log tree can already have checksum items with a subset of
1180 * the checksums we are trying to log. This can happen after
1181 * doing a sequence of partial writes into prealloc extents and
1182 * fsyncs in between, with a full fsync logging a larger subrange
1183 * of an extent for which a previous fast fsync logged a smaller
1184 * subrange. And this happens in particular due to merging file
1185 * extent items when we complete an ordered extent for a range
1186 * covered by a prealloc extent - this is done at
1187 * btrfs_mark_extent_written().
1189 * So if we try to extend the previous checksum item, which has
1190 * a range that ends at the start of the range we want to insert,
1191 * make sure we don't extend beyond the start offset of the next
1192 * checksum item. If we are at the last item in the leaf, then
1193 * forget the optimization of extending and add a new checksum
1194 * item - it is not worth the complexity of releasing the path,
1195 * getting the first key for the next leaf, repeat the btree
1196 * search, etc, because log trees are temporary anyway and it
1197 * would only save a few bytes of leaf space.
1199 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
1200 if (path->slots[0] + 1 >=
1201 btrfs_header_nritems(path->nodes[0])) {
1202 ret = find_next_csum_offset(root, path, &next_offset);
1209 ret = find_next_csum_offset(root, path, &next_offset);
1213 tmp = (next_offset - bytenr) >> fs_info->sectorsize_bits;
1215 extend_nr = min_t(int, extend_nr, tmp);
1218 diff = (csum_offset + extend_nr) * csum_size;
1220 MAX_CSUM_ITEMS(fs_info, csum_size) * csum_size);
1222 diff = diff - btrfs_item_size(leaf, path->slots[0]);
1223 diff = min_t(u32, btrfs_leaf_free_space(leaf), diff);
1227 btrfs_extend_item(path, diff);
1233 btrfs_release_path(path);
1238 tmp = sums->len - total_bytes;
1239 tmp >>= fs_info->sectorsize_bits;
1240 tmp = min(tmp, (next_offset - file_key.offset) >>
1241 fs_info->sectorsize_bits);
1243 tmp = max_t(u64, 1, tmp);
1244 tmp = min_t(u64, tmp, MAX_CSUM_ITEMS(fs_info, csum_size));
1245 ins_size = csum_size * tmp;
1247 ins_size = csum_size;
1249 ret = btrfs_insert_empty_item(trans, root, path, &file_key,
1253 if (WARN_ON(ret != 0))
1255 leaf = path->nodes[0];
1257 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item);
1258 item_end = (struct btrfs_csum_item *)((unsigned char *)item +
1259 btrfs_item_size(leaf, path->slots[0]));
1260 item = (struct btrfs_csum_item *)((unsigned char *)item +
1261 csum_offset * csum_size);
1263 ins_size = (u32)(sums->len - total_bytes) >> fs_info->sectorsize_bits;
1264 ins_size *= csum_size;
1265 ins_size = min_t(u32, (unsigned long)item_end - (unsigned long)item,
1267 write_extent_buffer(leaf, sums->sums + index, (unsigned long)item,
1271 ins_size /= csum_size;
1272 total_bytes += ins_size * fs_info->sectorsize;
1274 btrfs_mark_buffer_dirty(path->nodes[0]);
1275 if (total_bytes < sums->len) {
1276 btrfs_release_path(path);
1281 btrfs_free_path(path);
1285 void btrfs_extent_item_to_extent_map(struct btrfs_inode *inode,
1286 const struct btrfs_path *path,
1287 struct btrfs_file_extent_item *fi,
1288 struct extent_map *em)
1290 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1291 struct btrfs_root *root = inode->root;
1292 struct extent_buffer *leaf = path->nodes[0];
1293 const int slot = path->slots[0];
1294 struct btrfs_key key;
1295 u64 extent_start, extent_end;
1297 u8 type = btrfs_file_extent_type(leaf, fi);
1298 int compress_type = btrfs_file_extent_compression(leaf, fi);
1300 btrfs_item_key_to_cpu(leaf, &key, slot);
1301 extent_start = key.offset;
1302 extent_end = btrfs_file_extent_end(path);
1303 em->ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1304 em->generation = btrfs_file_extent_generation(leaf, fi);
1305 if (type == BTRFS_FILE_EXTENT_REG ||
1306 type == BTRFS_FILE_EXTENT_PREALLOC) {
1307 em->start = extent_start;
1308 em->len = extent_end - extent_start;
1309 em->orig_start = extent_start -
1310 btrfs_file_extent_offset(leaf, fi);
1311 em->orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
1312 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1314 em->block_start = EXTENT_MAP_HOLE;
1317 if (compress_type != BTRFS_COMPRESS_NONE) {
1318 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
1319 em->compress_type = compress_type;
1320 em->block_start = bytenr;
1321 em->block_len = em->orig_block_len;
1323 bytenr += btrfs_file_extent_offset(leaf, fi);
1324 em->block_start = bytenr;
1325 em->block_len = em->len;
1326 if (type == BTRFS_FILE_EXTENT_PREALLOC)
1327 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
1329 } else if (type == BTRFS_FILE_EXTENT_INLINE) {
1330 em->block_start = EXTENT_MAP_INLINE;
1331 em->start = extent_start;
1332 em->len = extent_end - extent_start;
1334 * Initialize orig_start and block_len with the same values
1335 * as in inode.c:btrfs_get_extent().
1337 em->orig_start = EXTENT_MAP_HOLE;
1338 em->block_len = (u64)-1;
1339 em->compress_type = compress_type;
1340 if (compress_type != BTRFS_COMPRESS_NONE)
1341 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
1344 "unknown file extent item type %d, inode %llu, offset %llu, "
1345 "root %llu", type, btrfs_ino(inode), extent_start,
1346 root->root_key.objectid);
1351 * Returns the end offset (non inclusive) of the file extent item the given path
1352 * points to. If it points to an inline extent, the returned offset is rounded
1353 * up to the sector size.
1355 u64 btrfs_file_extent_end(const struct btrfs_path *path)
1357 const struct extent_buffer *leaf = path->nodes[0];
1358 const int slot = path->slots[0];
1359 struct btrfs_file_extent_item *fi;
1360 struct btrfs_key key;
1363 btrfs_item_key_to_cpu(leaf, &key, slot);
1364 ASSERT(key.type == BTRFS_EXTENT_DATA_KEY);
1365 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
1367 if (btrfs_file_extent_type(leaf, fi) == BTRFS_FILE_EXTENT_INLINE) {
1368 end = btrfs_file_extent_ram_bytes(leaf, fi);
1369 end = ALIGN(key.offset + end, leaf->fs_info->sectorsize);
1371 end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);