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 i_size = new_i_size ?: i_size_read(&inode->vfs_inode);
56 if (btrfs_fs_incompat(fs_info, NO_HOLES)) {
57 inode->disk_i_size = i_size;
61 spin_lock(&inode->lock);
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;
69 spin_unlock(&inode->lock);
73 * Mark range within a file as having a new extent inserted.
75 * @inode: inode being modified
76 * @start: start file offset of the file extent we've inserted
77 * @len: logical length of the file extent item
79 * Call when we are inserting a new file extent where there was none before.
80 * Does not need to call this in the case where we're replacing an existing file
81 * extent, however if not sure it's fine to call this multiple times.
83 * The start and len must match the file extent item, so thus must be sectorsize
86 int btrfs_inode_set_file_extent_range(struct btrfs_inode *inode, u64 start,
92 ASSERT(IS_ALIGNED(start + len, inode->root->fs_info->sectorsize));
94 if (btrfs_fs_incompat(inode->root->fs_info, NO_HOLES))
96 return set_extent_bits(&inode->file_extent_tree, start, start + len - 1,
101 * Mark an inode range as not having a backing extent.
103 * @inode: inode being modified
104 * @start: start file offset of the file extent we've inserted
105 * @len: logical length of the file extent item
107 * Called when we drop a file extent, for example when we truncate. Doesn't
108 * need to be called for cases where we're replacing a file extent, like when
109 * we've COWed a file extent.
111 * The start and len must match the file extent item, so thus must be sectorsize
114 int btrfs_inode_clear_file_extent_range(struct btrfs_inode *inode, u64 start,
120 ASSERT(IS_ALIGNED(start + len, inode->root->fs_info->sectorsize) ||
123 if (btrfs_fs_incompat(inode->root->fs_info, NO_HOLES))
125 return clear_extent_bit(&inode->file_extent_tree, start,
126 start + len - 1, EXTENT_DIRTY, NULL);
129 static inline u32 max_ordered_sum_bytes(struct btrfs_fs_info *fs_info,
132 u32 ncsums = (PAGE_SIZE - sizeof(struct btrfs_ordered_sum)) / csum_size;
134 return ncsums * fs_info->sectorsize;
138 * Calculate the total size needed to allocate for an ordered sum structure
139 * spanning @bytes in the file.
141 static int btrfs_ordered_sum_size(struct btrfs_fs_info *fs_info, unsigned long bytes)
143 int num_sectors = (int)DIV_ROUND_UP(bytes, fs_info->sectorsize);
145 return sizeof(struct btrfs_ordered_sum) + num_sectors * fs_info->csum_size;
148 int btrfs_insert_hole_extent(struct btrfs_trans_handle *trans,
149 struct btrfs_root *root,
150 u64 objectid, u64 pos, u64 num_bytes)
153 struct btrfs_file_extent_item *item;
154 struct btrfs_key file_key;
155 struct btrfs_path *path;
156 struct extent_buffer *leaf;
158 path = btrfs_alloc_path();
161 file_key.objectid = objectid;
162 file_key.offset = pos;
163 file_key.type = BTRFS_EXTENT_DATA_KEY;
165 ret = btrfs_insert_empty_item(trans, root, path, &file_key,
169 BUG_ON(ret); /* Can't happen */
170 leaf = path->nodes[0];
171 item = btrfs_item_ptr(leaf, path->slots[0],
172 struct btrfs_file_extent_item);
173 btrfs_set_file_extent_disk_bytenr(leaf, item, 0);
174 btrfs_set_file_extent_disk_num_bytes(leaf, item, 0);
175 btrfs_set_file_extent_offset(leaf, item, 0);
176 btrfs_set_file_extent_num_bytes(leaf, item, num_bytes);
177 btrfs_set_file_extent_ram_bytes(leaf, item, num_bytes);
178 btrfs_set_file_extent_generation(leaf, item, trans->transid);
179 btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
180 btrfs_set_file_extent_compression(leaf, item, 0);
181 btrfs_set_file_extent_encryption(leaf, item, 0);
182 btrfs_set_file_extent_other_encoding(leaf, item, 0);
184 btrfs_mark_buffer_dirty(leaf);
186 btrfs_free_path(path);
190 static struct btrfs_csum_item *
191 btrfs_lookup_csum(struct btrfs_trans_handle *trans,
192 struct btrfs_root *root,
193 struct btrfs_path *path,
196 struct btrfs_fs_info *fs_info = root->fs_info;
198 struct btrfs_key file_key;
199 struct btrfs_key found_key;
200 struct btrfs_csum_item *item;
201 struct extent_buffer *leaf;
203 const u32 csum_size = fs_info->csum_size;
206 file_key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
207 file_key.offset = bytenr;
208 file_key.type = BTRFS_EXTENT_CSUM_KEY;
209 ret = btrfs_search_slot(trans, root, &file_key, path, 0, cow);
212 leaf = path->nodes[0];
215 if (path->slots[0] == 0)
218 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
219 if (found_key.type != BTRFS_EXTENT_CSUM_KEY)
222 csum_offset = (bytenr - found_key.offset) >>
223 fs_info->sectorsize_bits;
224 csums_in_item = btrfs_item_size(leaf, path->slots[0]);
225 csums_in_item /= csum_size;
227 if (csum_offset == csums_in_item) {
230 } else if (csum_offset > csums_in_item) {
234 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item);
235 item = (struct btrfs_csum_item *)((unsigned char *)item +
236 csum_offset * csum_size);
244 int btrfs_lookup_file_extent(struct btrfs_trans_handle *trans,
245 struct btrfs_root *root,
246 struct btrfs_path *path, u64 objectid,
249 struct btrfs_key file_key;
250 int ins_len = mod < 0 ? -1 : 0;
253 file_key.objectid = objectid;
254 file_key.offset = offset;
255 file_key.type = BTRFS_EXTENT_DATA_KEY;
257 return btrfs_search_slot(trans, root, &file_key, path, ins_len, cow);
261 * Find checksums for logical bytenr range [disk_bytenr, disk_bytenr + len) and
262 * store the result to @dst.
264 * Return >0 for the number of sectors we found.
265 * Return 0 for the range [disk_bytenr, disk_bytenr + sectorsize) has no csum
266 * for it. Caller may want to try next sector until one range is hit.
267 * Return <0 for fatal error.
269 static int search_csum_tree(struct btrfs_fs_info *fs_info,
270 struct btrfs_path *path, u64 disk_bytenr,
273 struct btrfs_root *csum_root;
274 struct btrfs_csum_item *item = NULL;
275 struct btrfs_key key;
276 const u32 sectorsize = fs_info->sectorsize;
277 const u32 csum_size = fs_info->csum_size;
283 ASSERT(IS_ALIGNED(disk_bytenr, sectorsize) &&
284 IS_ALIGNED(len, sectorsize));
286 /* Check if the current csum item covers disk_bytenr */
287 if (path->nodes[0]) {
288 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
289 struct btrfs_csum_item);
290 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
291 itemsize = btrfs_item_size(path->nodes[0], path->slots[0]);
293 csum_start = key.offset;
294 csum_len = (itemsize / csum_size) * sectorsize;
296 if (in_range(disk_bytenr, csum_start, csum_len))
300 /* Current item doesn't contain the desired range, search again */
301 btrfs_release_path(path);
302 csum_root = btrfs_csum_root(fs_info, disk_bytenr);
303 item = btrfs_lookup_csum(NULL, csum_root, path, disk_bytenr, 0);
308 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
309 itemsize = btrfs_item_size(path->nodes[0], path->slots[0]);
311 csum_start = key.offset;
312 csum_len = (itemsize / csum_size) * sectorsize;
313 ASSERT(in_range(disk_bytenr, csum_start, csum_len));
316 ret = (min(csum_start + csum_len, disk_bytenr + len) -
317 disk_bytenr) >> fs_info->sectorsize_bits;
318 read_extent_buffer(path->nodes[0], dst, (unsigned long)item,
321 if (ret == -ENOENT || ret == -EFBIG)
327 * Locate the file_offset of @cur_disk_bytenr of a @bio.
329 * Bio of btrfs represents read range of
330 * [bi_sector << 9, bi_sector << 9 + bi_size).
331 * Knowing this, we can iterate through each bvec to locate the page belong to
332 * @cur_disk_bytenr and get the file offset.
334 * @inode is used to determine if the bvec page really belongs to @inode.
336 * Return 0 if we can't find the file offset
337 * Return >0 if we find the file offset and restore it to @file_offset_ret
339 static int search_file_offset_in_bio(struct bio *bio, struct inode *inode,
340 u64 disk_bytenr, u64 *file_offset_ret)
342 struct bvec_iter iter;
344 u64 cur = bio->bi_iter.bi_sector << SECTOR_SHIFT;
347 bio_for_each_segment(bvec, bio, iter) {
348 struct page *page = bvec.bv_page;
350 if (cur > disk_bytenr)
352 if (cur + bvec.bv_len <= disk_bytenr) {
356 ASSERT(in_range(disk_bytenr, cur, bvec.bv_len));
357 if (page->mapping && page->mapping->host &&
358 page->mapping->host == inode) {
360 *file_offset_ret = page_offset(page) + bvec.bv_offset +
369 * Lookup the checksum for the read bio in csum tree.
371 * @inode: inode that the bio is for.
372 * @bio: bio to look up.
373 * @dst: Buffer of size nblocks * btrfs_super_csum_size() used to return
374 * checksum (nblocks = bio->bi_iter.bi_size / fs_info->sectorsize). If
375 * NULL, the checksum buffer is allocated and returned in
376 * btrfs_bio(bio)->csum instead.
378 * Return: BLK_STS_RESOURCE if allocating memory fails, BLK_STS_OK otherwise.
380 blk_status_t btrfs_lookup_bio_sums(struct inode *inode, struct bio *bio, u8 *dst)
382 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
383 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
384 struct btrfs_bio *bbio = NULL;
385 struct btrfs_path *path;
386 const u32 sectorsize = fs_info->sectorsize;
387 const u32 csum_size = fs_info->csum_size;
388 u32 orig_len = bio->bi_iter.bi_size;
389 u64 orig_disk_bytenr = bio->bi_iter.bi_sector << SECTOR_SHIFT;
392 const unsigned int nblocks = orig_len >> fs_info->sectorsize_bits;
394 blk_status_t ret = BLK_STS_OK;
396 if ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM) ||
397 test_bit(BTRFS_FS_STATE_NO_CSUMS, &fs_info->fs_state))
401 * This function is only called for read bio.
403 * This means two things:
404 * - All our csums should only be in csum tree
405 * No ordered extents csums, as ordered extents are only for write
407 * - No need to bother any other info from bvec
408 * Since we're looking up csums, the only important info is the
409 * disk_bytenr and the length, which can be extracted from bi_iter
412 ASSERT(bio_op(bio) == REQ_OP_READ);
413 path = btrfs_alloc_path();
415 return BLK_STS_RESOURCE;
418 bbio = btrfs_bio(bio);
420 if (nblocks * csum_size > BTRFS_BIO_INLINE_CSUM_SIZE) {
421 bbio->csum = kmalloc_array(nblocks, csum_size, GFP_NOFS);
423 btrfs_free_path(path);
424 return BLK_STS_RESOURCE;
427 bbio->csum = bbio->csum_inline;
435 * If requested number of sectors is larger than one leaf can contain,
436 * kick the readahead for csum tree.
438 if (nblocks > fs_info->csums_per_leaf)
439 path->reada = READA_FORWARD;
442 * the free space stuff is only read when it hasn't been
443 * updated in the current transaction. So, we can safely
444 * read from the commit root and sidestep a nasty deadlock
445 * between reading the free space cache and updating the csum tree.
447 if (btrfs_is_free_space_inode(BTRFS_I(inode))) {
448 path->search_commit_root = 1;
449 path->skip_locking = 1;
452 for (cur_disk_bytenr = orig_disk_bytenr;
453 cur_disk_bytenr < orig_disk_bytenr + orig_len;
454 cur_disk_bytenr += (count * sectorsize)) {
455 u64 search_len = orig_disk_bytenr + orig_len - cur_disk_bytenr;
456 unsigned int sector_offset;
460 * Although both cur_disk_bytenr and orig_disk_bytenr is u64,
461 * we're calculating the offset to the bio start.
463 * Bio size is limited to UINT_MAX, thus unsigned int is large
464 * enough to contain the raw result, not to mention the right
467 ASSERT(cur_disk_bytenr - orig_disk_bytenr < UINT_MAX);
468 sector_offset = (cur_disk_bytenr - orig_disk_bytenr) >>
469 fs_info->sectorsize_bits;
470 csum_dst = csum + sector_offset * csum_size;
472 count = search_csum_tree(fs_info, path, cur_disk_bytenr,
473 search_len, csum_dst);
475 ret = errno_to_blk_status(count);
477 btrfs_bio_free_csum(bbio);
482 * We didn't find a csum for this range. We need to make sure
483 * we complain loudly about this, because we are not NODATASUM.
485 * However for the DATA_RELOC inode we could potentially be
486 * relocating data extents for a NODATASUM inode, so the inode
487 * itself won't be marked with NODATASUM, but the extent we're
488 * copying is in fact NODATASUM. If we don't find a csum we
489 * assume this is the case.
492 memset(csum_dst, 0, csum_size);
495 if (BTRFS_I(inode)->root->root_key.objectid ==
496 BTRFS_DATA_RELOC_TREE_OBJECTID) {
500 ret = search_file_offset_in_bio(bio, inode,
501 cur_disk_bytenr, &file_offset);
503 set_extent_bits(io_tree, file_offset,
504 file_offset + sectorsize - 1,
507 btrfs_warn_rl(fs_info,
508 "csum hole found for disk bytenr range [%llu, %llu)",
509 cur_disk_bytenr, cur_disk_bytenr + sectorsize);
514 btrfs_free_path(path);
518 int btrfs_lookup_csums_range(struct btrfs_root *root, u64 start, u64 end,
519 struct list_head *list, int search_commit,
522 struct btrfs_fs_info *fs_info = root->fs_info;
523 struct btrfs_key key;
524 struct btrfs_path *path;
525 struct extent_buffer *leaf;
526 struct btrfs_ordered_sum *sums;
527 struct btrfs_csum_item *item;
529 unsigned long offset;
533 const u32 csum_size = fs_info->csum_size;
535 ASSERT(IS_ALIGNED(start, fs_info->sectorsize) &&
536 IS_ALIGNED(end + 1, fs_info->sectorsize));
538 path = btrfs_alloc_path();
542 path->nowait = nowait;
544 path->skip_locking = 1;
545 path->reada = READA_FORWARD;
546 path->search_commit_root = 1;
549 key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
551 key.type = BTRFS_EXTENT_CSUM_KEY;
553 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
556 if (ret > 0 && path->slots[0] > 0) {
557 leaf = path->nodes[0];
558 btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
559 if (key.objectid == BTRFS_EXTENT_CSUM_OBJECTID &&
560 key.type == BTRFS_EXTENT_CSUM_KEY) {
561 offset = (start - key.offset) >> fs_info->sectorsize_bits;
562 if (offset * csum_size <
563 btrfs_item_size(leaf, path->slots[0] - 1))
568 while (start <= end) {
569 leaf = path->nodes[0];
570 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
571 ret = btrfs_next_leaf(root, path);
576 leaf = path->nodes[0];
579 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
580 if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
581 key.type != BTRFS_EXTENT_CSUM_KEY ||
585 if (key.offset > start)
588 size = btrfs_item_size(leaf, path->slots[0]);
589 csum_end = key.offset + (size / csum_size) * fs_info->sectorsize;
590 if (csum_end <= start) {
595 csum_end = min(csum_end, end + 1);
596 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
597 struct btrfs_csum_item);
598 while (start < csum_end) {
599 size = min_t(size_t, csum_end - start,
600 max_ordered_sum_bytes(fs_info, csum_size));
601 sums = kzalloc(btrfs_ordered_sum_size(fs_info, size),
608 sums->bytenr = start;
609 sums->len = (int)size;
611 offset = (start - key.offset) >> fs_info->sectorsize_bits;
613 size >>= fs_info->sectorsize_bits;
615 read_extent_buffer(path->nodes[0],
617 ((unsigned long)item) + offset,
620 start += fs_info->sectorsize * size;
621 list_add_tail(&sums->list, &tmplist);
627 while (ret < 0 && !list_empty(&tmplist)) {
628 sums = list_entry(tmplist.next, struct btrfs_ordered_sum, list);
629 list_del(&sums->list);
632 list_splice_tail(&tmplist, list);
634 btrfs_free_path(path);
639 * Calculate checksums of the data contained inside a bio.
641 * @inode: Owner of the data inside the bio
642 * @bio: Contains the data to be checksummed
643 * @offset: If (u64)-1, @bio may contain discontiguous bio vecs, so the
644 * file offsets are determined from the page offsets in the bio.
645 * Otherwise, this is the starting file offset of the bio vecs in
646 * @bio, which must be contiguous.
647 * @one_ordered: If true, @bio only refers to one ordered extent.
649 blk_status_t btrfs_csum_one_bio(struct btrfs_inode *inode, struct bio *bio,
650 u64 offset, bool one_ordered)
652 struct btrfs_fs_info *fs_info = inode->root->fs_info;
653 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
654 struct btrfs_ordered_sum *sums;
655 struct btrfs_ordered_extent *ordered = NULL;
656 const bool use_page_offsets = (offset == (u64)-1);
658 struct bvec_iter iter;
661 unsigned int blockcount;
662 unsigned long total_bytes = 0;
663 unsigned long this_sum_bytes = 0;
667 nofs_flag = memalloc_nofs_save();
668 sums = kvzalloc(btrfs_ordered_sum_size(fs_info, bio->bi_iter.bi_size),
670 memalloc_nofs_restore(nofs_flag);
673 return BLK_STS_RESOURCE;
675 sums->len = bio->bi_iter.bi_size;
676 INIT_LIST_HEAD(&sums->list);
678 sums->bytenr = bio->bi_iter.bi_sector << 9;
681 shash->tfm = fs_info->csum_shash;
683 bio_for_each_segment(bvec, bio, iter) {
684 if (use_page_offsets)
685 offset = page_offset(bvec.bv_page) + bvec.bv_offset;
688 ordered = btrfs_lookup_ordered_extent(inode, offset);
690 * The bio range is not covered by any ordered extent,
691 * must be a code logic error.
693 if (unlikely(!ordered)) {
695 "no ordered extent for root %llu ino %llu offset %llu\n",
696 inode->root->root_key.objectid,
697 btrfs_ino(inode), offset);
699 return BLK_STS_IOERR;
703 blockcount = BTRFS_BYTES_TO_BLKS(fs_info,
704 bvec.bv_len + fs_info->sectorsize
707 for (i = 0; i < blockcount; i++) {
709 !in_range(offset, ordered->file_offset,
710 ordered->num_bytes)) {
711 unsigned long bytes_left;
713 sums->len = this_sum_bytes;
715 btrfs_add_ordered_sum(ordered, sums);
716 btrfs_put_ordered_extent(ordered);
718 bytes_left = bio->bi_iter.bi_size - total_bytes;
720 nofs_flag = memalloc_nofs_save();
721 sums = kvzalloc(btrfs_ordered_sum_size(fs_info,
722 bytes_left), GFP_KERNEL);
723 memalloc_nofs_restore(nofs_flag);
724 BUG_ON(!sums); /* -ENOMEM */
725 sums->len = bytes_left;
726 ordered = btrfs_lookup_ordered_extent(inode,
728 ASSERT(ordered); /* Logic error */
729 sums->bytenr = (bio->bi_iter.bi_sector << 9)
734 data = bvec_kmap_local(&bvec);
735 crypto_shash_digest(shash,
736 data + (i * fs_info->sectorsize),
740 index += fs_info->csum_size;
741 offset += fs_info->sectorsize;
742 this_sum_bytes += fs_info->sectorsize;
743 total_bytes += fs_info->sectorsize;
748 btrfs_add_ordered_sum(ordered, sums);
749 btrfs_put_ordered_extent(ordered);
754 * Remove one checksum overlapping a range.
756 * This expects the key to describe the csum pointed to by the path, and it
757 * expects the csum to overlap the range [bytenr, len]
759 * The csum should not be entirely contained in the range and the range should
760 * not be entirely contained in the csum.
762 * This calls btrfs_truncate_item with the correct args based on the overlap,
763 * and fixes up the key as required.
765 static noinline void truncate_one_csum(struct btrfs_fs_info *fs_info,
766 struct btrfs_path *path,
767 struct btrfs_key *key,
770 struct extent_buffer *leaf;
771 const u32 csum_size = fs_info->csum_size;
773 u64 end_byte = bytenr + len;
774 u32 blocksize_bits = fs_info->sectorsize_bits;
776 leaf = path->nodes[0];
777 csum_end = btrfs_item_size(leaf, path->slots[0]) / csum_size;
778 csum_end <<= blocksize_bits;
779 csum_end += key->offset;
781 if (key->offset < bytenr && csum_end <= end_byte) {
786 * A simple truncate off the end of the item
788 u32 new_size = (bytenr - key->offset) >> blocksize_bits;
789 new_size *= csum_size;
790 btrfs_truncate_item(path, new_size, 1);
791 } else if (key->offset >= bytenr && csum_end > end_byte &&
792 end_byte > key->offset) {
797 * we need to truncate from the beginning of the csum
799 u32 new_size = (csum_end - end_byte) >> blocksize_bits;
800 new_size *= csum_size;
802 btrfs_truncate_item(path, new_size, 0);
804 key->offset = end_byte;
805 btrfs_set_item_key_safe(fs_info, path, key);
812 * Delete the csum items from the csum tree for a given range of bytes.
814 int btrfs_del_csums(struct btrfs_trans_handle *trans,
815 struct btrfs_root *root, u64 bytenr, u64 len)
817 struct btrfs_fs_info *fs_info = trans->fs_info;
818 struct btrfs_path *path;
819 struct btrfs_key key;
820 u64 end_byte = bytenr + len;
822 struct extent_buffer *leaf;
824 const u32 csum_size = fs_info->csum_size;
825 u32 blocksize_bits = fs_info->sectorsize_bits;
827 ASSERT(root->root_key.objectid == BTRFS_CSUM_TREE_OBJECTID ||
828 root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID);
830 path = btrfs_alloc_path();
835 key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
836 key.offset = end_byte - 1;
837 key.type = BTRFS_EXTENT_CSUM_KEY;
839 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
842 if (path->slots[0] == 0)
845 } else if (ret < 0) {
849 leaf = path->nodes[0];
850 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
852 if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
853 key.type != BTRFS_EXTENT_CSUM_KEY) {
857 if (key.offset >= end_byte)
860 csum_end = btrfs_item_size(leaf, path->slots[0]) / csum_size;
861 csum_end <<= blocksize_bits;
862 csum_end += key.offset;
864 /* this csum ends before we start, we're done */
865 if (csum_end <= bytenr)
868 /* delete the entire item, it is inside our range */
869 if (key.offset >= bytenr && csum_end <= end_byte) {
873 * Check how many csum items preceding this one in this
874 * leaf correspond to our range and then delete them all
877 if (key.offset > bytenr && path->slots[0] > 0) {
878 int slot = path->slots[0] - 1;
883 btrfs_item_key_to_cpu(leaf, &pk, slot);
884 if (pk.offset < bytenr ||
885 pk.type != BTRFS_EXTENT_CSUM_KEY ||
887 BTRFS_EXTENT_CSUM_OBJECTID)
889 path->slots[0] = slot;
891 key.offset = pk.offset;
895 ret = btrfs_del_items(trans, root, path,
896 path->slots[0], del_nr);
899 if (key.offset == bytenr)
901 } else if (key.offset < bytenr && csum_end > end_byte) {
902 unsigned long offset;
903 unsigned long shift_len;
904 unsigned long item_offset;
909 * Our bytes are in the middle of the csum,
910 * we need to split this item and insert a new one.
912 * But we can't drop the path because the
913 * csum could change, get removed, extended etc.
915 * The trick here is the max size of a csum item leaves
916 * enough room in the tree block for a single
917 * item header. So, we split the item in place,
918 * adding a new header pointing to the existing
919 * bytes. Then we loop around again and we have
920 * a nicely formed csum item that we can neatly
923 offset = (bytenr - key.offset) >> blocksize_bits;
926 shift_len = (len >> blocksize_bits) * csum_size;
928 item_offset = btrfs_item_ptr_offset(leaf,
931 memzero_extent_buffer(leaf, item_offset + offset,
936 * btrfs_split_item returns -EAGAIN when the
937 * item changed size or key
939 ret = btrfs_split_item(trans, root, path, &key, offset);
940 if (ret && ret != -EAGAIN) {
941 btrfs_abort_transaction(trans, ret);
946 key.offset = end_byte - 1;
948 truncate_one_csum(fs_info, path, &key, bytenr, len);
949 if (key.offset < bytenr)
952 btrfs_release_path(path);
954 btrfs_free_path(path);
958 static int find_next_csum_offset(struct btrfs_root *root,
959 struct btrfs_path *path,
962 const u32 nritems = btrfs_header_nritems(path->nodes[0]);
963 struct btrfs_key found_key;
964 int slot = path->slots[0] + 1;
967 if (nritems == 0 || slot >= nritems) {
968 ret = btrfs_next_leaf(root, path);
971 } else if (ret > 0) {
972 *next_offset = (u64)-1;
975 slot = path->slots[0];
978 btrfs_item_key_to_cpu(path->nodes[0], &found_key, slot);
980 if (found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
981 found_key.type != BTRFS_EXTENT_CSUM_KEY)
982 *next_offset = (u64)-1;
984 *next_offset = found_key.offset;
989 int btrfs_csum_file_blocks(struct btrfs_trans_handle *trans,
990 struct btrfs_root *root,
991 struct btrfs_ordered_sum *sums)
993 struct btrfs_fs_info *fs_info = root->fs_info;
994 struct btrfs_key file_key;
995 struct btrfs_key found_key;
996 struct btrfs_path *path;
997 struct btrfs_csum_item *item;
998 struct btrfs_csum_item *item_end;
999 struct extent_buffer *leaf = NULL;
1001 u64 total_bytes = 0;
1008 const u32 csum_size = fs_info->csum_size;
1010 path = btrfs_alloc_path();
1014 next_offset = (u64)-1;
1016 bytenr = sums->bytenr + total_bytes;
1017 file_key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
1018 file_key.offset = bytenr;
1019 file_key.type = BTRFS_EXTENT_CSUM_KEY;
1021 item = btrfs_lookup_csum(trans, root, path, bytenr, 1);
1022 if (!IS_ERR(item)) {
1024 leaf = path->nodes[0];
1025 item_end = btrfs_item_ptr(leaf, path->slots[0],
1026 struct btrfs_csum_item);
1027 item_end = (struct btrfs_csum_item *)((char *)item_end +
1028 btrfs_item_size(leaf, path->slots[0]));
1031 ret = PTR_ERR(item);
1032 if (ret != -EFBIG && ret != -ENOENT)
1035 if (ret == -EFBIG) {
1037 /* we found one, but it isn't big enough yet */
1038 leaf = path->nodes[0];
1039 item_size = btrfs_item_size(leaf, path->slots[0]);
1040 if ((item_size / csum_size) >=
1041 MAX_CSUM_ITEMS(fs_info, csum_size)) {
1042 /* already at max size, make a new one */
1046 /* We didn't find a csum item, insert one. */
1047 ret = find_next_csum_offset(root, path, &next_offset);
1055 * At this point, we know the tree has a checksum item that ends at an
1056 * offset matching the start of the checksum range we want to insert.
1057 * We try to extend that item as much as possible and then add as many
1058 * checksums to it as they fit.
1060 * First check if the leaf has enough free space for at least one
1061 * checksum. If it has go directly to the item extension code, otherwise
1062 * release the path and do a search for insertion before the extension.
1064 if (btrfs_leaf_free_space(leaf) >= csum_size) {
1065 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1066 csum_offset = (bytenr - found_key.offset) >>
1067 fs_info->sectorsize_bits;
1071 btrfs_release_path(path);
1072 path->search_for_extension = 1;
1073 ret = btrfs_search_slot(trans, root, &file_key, path,
1075 path->search_for_extension = 0;
1080 if (path->slots[0] == 0)
1085 leaf = path->nodes[0];
1086 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1087 csum_offset = (bytenr - found_key.offset) >> fs_info->sectorsize_bits;
1089 if (found_key.type != BTRFS_EXTENT_CSUM_KEY ||
1090 found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
1091 csum_offset >= MAX_CSUM_ITEMS(fs_info, csum_size)) {
1096 if (csum_offset == btrfs_item_size(leaf, path->slots[0]) /
1102 tmp = sums->len - total_bytes;
1103 tmp >>= fs_info->sectorsize_bits;
1105 extend_nr = max_t(int, 1, tmp);
1108 * A log tree can already have checksum items with a subset of
1109 * the checksums we are trying to log. This can happen after
1110 * doing a sequence of partial writes into prealloc extents and
1111 * fsyncs in between, with a full fsync logging a larger subrange
1112 * of an extent for which a previous fast fsync logged a smaller
1113 * subrange. And this happens in particular due to merging file
1114 * extent items when we complete an ordered extent for a range
1115 * covered by a prealloc extent - this is done at
1116 * btrfs_mark_extent_written().
1118 * So if we try to extend the previous checksum item, which has
1119 * a range that ends at the start of the range we want to insert,
1120 * make sure we don't extend beyond the start offset of the next
1121 * checksum item. If we are at the last item in the leaf, then
1122 * forget the optimization of extending and add a new checksum
1123 * item - it is not worth the complexity of releasing the path,
1124 * getting the first key for the next leaf, repeat the btree
1125 * search, etc, because log trees are temporary anyway and it
1126 * would only save a few bytes of leaf space.
1128 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
1129 if (path->slots[0] + 1 >=
1130 btrfs_header_nritems(path->nodes[0])) {
1131 ret = find_next_csum_offset(root, path, &next_offset);
1138 ret = find_next_csum_offset(root, path, &next_offset);
1142 tmp = (next_offset - bytenr) >> fs_info->sectorsize_bits;
1144 extend_nr = min_t(int, extend_nr, tmp);
1147 diff = (csum_offset + extend_nr) * csum_size;
1149 MAX_CSUM_ITEMS(fs_info, csum_size) * csum_size);
1151 diff = diff - btrfs_item_size(leaf, path->slots[0]);
1152 diff = min_t(u32, btrfs_leaf_free_space(leaf), diff);
1156 btrfs_extend_item(path, diff);
1162 btrfs_release_path(path);
1167 tmp = sums->len - total_bytes;
1168 tmp >>= fs_info->sectorsize_bits;
1169 tmp = min(tmp, (next_offset - file_key.offset) >>
1170 fs_info->sectorsize_bits);
1172 tmp = max_t(u64, 1, tmp);
1173 tmp = min_t(u64, tmp, MAX_CSUM_ITEMS(fs_info, csum_size));
1174 ins_size = csum_size * tmp;
1176 ins_size = csum_size;
1178 ret = btrfs_insert_empty_item(trans, root, path, &file_key,
1182 if (WARN_ON(ret != 0))
1184 leaf = path->nodes[0];
1186 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item);
1187 item_end = (struct btrfs_csum_item *)((unsigned char *)item +
1188 btrfs_item_size(leaf, path->slots[0]));
1189 item = (struct btrfs_csum_item *)((unsigned char *)item +
1190 csum_offset * csum_size);
1192 ins_size = (u32)(sums->len - total_bytes) >> fs_info->sectorsize_bits;
1193 ins_size *= csum_size;
1194 ins_size = min_t(u32, (unsigned long)item_end - (unsigned long)item,
1196 write_extent_buffer(leaf, sums->sums + index, (unsigned long)item,
1200 ins_size /= csum_size;
1201 total_bytes += ins_size * fs_info->sectorsize;
1203 btrfs_mark_buffer_dirty(path->nodes[0]);
1204 if (total_bytes < sums->len) {
1205 btrfs_release_path(path);
1210 btrfs_free_path(path);
1214 void btrfs_extent_item_to_extent_map(struct btrfs_inode *inode,
1215 const struct btrfs_path *path,
1216 struct btrfs_file_extent_item *fi,
1217 struct extent_map *em)
1219 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1220 struct btrfs_root *root = inode->root;
1221 struct extent_buffer *leaf = path->nodes[0];
1222 const int slot = path->slots[0];
1223 struct btrfs_key key;
1224 u64 extent_start, extent_end;
1226 u8 type = btrfs_file_extent_type(leaf, fi);
1227 int compress_type = btrfs_file_extent_compression(leaf, fi);
1229 btrfs_item_key_to_cpu(leaf, &key, slot);
1230 extent_start = key.offset;
1231 extent_end = btrfs_file_extent_end(path);
1232 em->ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1233 em->generation = btrfs_file_extent_generation(leaf, fi);
1234 if (type == BTRFS_FILE_EXTENT_REG ||
1235 type == BTRFS_FILE_EXTENT_PREALLOC) {
1236 em->start = extent_start;
1237 em->len = extent_end - extent_start;
1238 em->orig_start = extent_start -
1239 btrfs_file_extent_offset(leaf, fi);
1240 em->orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
1241 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1243 em->block_start = EXTENT_MAP_HOLE;
1246 if (compress_type != BTRFS_COMPRESS_NONE) {
1247 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
1248 em->compress_type = compress_type;
1249 em->block_start = bytenr;
1250 em->block_len = em->orig_block_len;
1252 bytenr += btrfs_file_extent_offset(leaf, fi);
1253 em->block_start = bytenr;
1254 em->block_len = em->len;
1255 if (type == BTRFS_FILE_EXTENT_PREALLOC)
1256 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
1258 } else if (type == BTRFS_FILE_EXTENT_INLINE) {
1259 em->block_start = EXTENT_MAP_INLINE;
1260 em->start = extent_start;
1261 em->len = extent_end - extent_start;
1263 * Initialize orig_start and block_len with the same values
1264 * as in inode.c:btrfs_get_extent().
1266 em->orig_start = EXTENT_MAP_HOLE;
1267 em->block_len = (u64)-1;
1268 em->compress_type = compress_type;
1269 if (compress_type != BTRFS_COMPRESS_NONE)
1270 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
1273 "unknown file extent item type %d, inode %llu, offset %llu, "
1274 "root %llu", type, btrfs_ino(inode), extent_start,
1275 root->root_key.objectid);
1280 * Returns the end offset (non inclusive) of the file extent item the given path
1281 * points to. If it points to an inline extent, the returned offset is rounded
1282 * up to the sector size.
1284 u64 btrfs_file_extent_end(const struct btrfs_path *path)
1286 const struct extent_buffer *leaf = path->nodes[0];
1287 const int slot = path->slots[0];
1288 struct btrfs_file_extent_item *fi;
1289 struct btrfs_key key;
1292 btrfs_item_key_to_cpu(leaf, &key, slot);
1293 ASSERT(key.type == BTRFS_EXTENT_DATA_KEY);
1294 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
1296 if (btrfs_file_extent_type(leaf, fi) == BTRFS_FILE_EXTENT_INLINE) {
1297 end = btrfs_file_extent_ram_bytes(leaf, fi);
1298 end = ALIGN(key.offset + end, leaf->fs_info->sectorsize);
1300 end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);