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 size_t bytes_to_csum_size(const struct btrfs_fs_info *fs_info, u32 bytes)
131 ASSERT(IS_ALIGNED(bytes, fs_info->sectorsize));
133 return (bytes >> fs_info->sectorsize_bits) * fs_info->csum_size;
136 static size_t csum_size_to_bytes(const struct btrfs_fs_info *fs_info, u32 csum_size)
138 ASSERT(IS_ALIGNED(csum_size, fs_info->csum_size));
140 return (csum_size / fs_info->csum_size) << fs_info->sectorsize_bits;
143 static inline u32 max_ordered_sum_bytes(const struct btrfs_fs_info *fs_info)
145 u32 max_csum_size = round_down(PAGE_SIZE - sizeof(struct btrfs_ordered_sum),
148 return csum_size_to_bytes(fs_info, max_csum_size);
152 * Calculate the total size needed to allocate for an ordered sum structure
153 * spanning @bytes in the file.
155 static int btrfs_ordered_sum_size(struct btrfs_fs_info *fs_info, unsigned long bytes)
157 return sizeof(struct btrfs_ordered_sum) + bytes_to_csum_size(fs_info, bytes);
160 int btrfs_insert_hole_extent(struct btrfs_trans_handle *trans,
161 struct btrfs_root *root,
162 u64 objectid, u64 pos, u64 num_bytes)
165 struct btrfs_file_extent_item *item;
166 struct btrfs_key file_key;
167 struct btrfs_path *path;
168 struct extent_buffer *leaf;
170 path = btrfs_alloc_path();
173 file_key.objectid = objectid;
174 file_key.offset = pos;
175 file_key.type = BTRFS_EXTENT_DATA_KEY;
177 ret = btrfs_insert_empty_item(trans, root, path, &file_key,
181 BUG_ON(ret); /* Can't happen */
182 leaf = path->nodes[0];
183 item = btrfs_item_ptr(leaf, path->slots[0],
184 struct btrfs_file_extent_item);
185 btrfs_set_file_extent_disk_bytenr(leaf, item, 0);
186 btrfs_set_file_extent_disk_num_bytes(leaf, item, 0);
187 btrfs_set_file_extent_offset(leaf, item, 0);
188 btrfs_set_file_extent_num_bytes(leaf, item, num_bytes);
189 btrfs_set_file_extent_ram_bytes(leaf, item, num_bytes);
190 btrfs_set_file_extent_generation(leaf, item, trans->transid);
191 btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
192 btrfs_set_file_extent_compression(leaf, item, 0);
193 btrfs_set_file_extent_encryption(leaf, item, 0);
194 btrfs_set_file_extent_other_encoding(leaf, item, 0);
196 btrfs_mark_buffer_dirty(leaf);
198 btrfs_free_path(path);
202 static struct btrfs_csum_item *
203 btrfs_lookup_csum(struct btrfs_trans_handle *trans,
204 struct btrfs_root *root,
205 struct btrfs_path *path,
208 struct btrfs_fs_info *fs_info = root->fs_info;
210 struct btrfs_key file_key;
211 struct btrfs_key found_key;
212 struct btrfs_csum_item *item;
213 struct extent_buffer *leaf;
215 const u32 csum_size = fs_info->csum_size;
218 file_key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
219 file_key.offset = bytenr;
220 file_key.type = BTRFS_EXTENT_CSUM_KEY;
221 ret = btrfs_search_slot(trans, root, &file_key, path, 0, cow);
224 leaf = path->nodes[0];
227 if (path->slots[0] == 0)
230 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
231 if (found_key.type != BTRFS_EXTENT_CSUM_KEY)
234 csum_offset = (bytenr - found_key.offset) >>
235 fs_info->sectorsize_bits;
236 csums_in_item = btrfs_item_size(leaf, path->slots[0]);
237 csums_in_item /= csum_size;
239 if (csum_offset == csums_in_item) {
242 } else if (csum_offset > csums_in_item) {
246 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item);
247 item = (struct btrfs_csum_item *)((unsigned char *)item +
248 csum_offset * csum_size);
256 int btrfs_lookup_file_extent(struct btrfs_trans_handle *trans,
257 struct btrfs_root *root,
258 struct btrfs_path *path, u64 objectid,
261 struct btrfs_key file_key;
262 int ins_len = mod < 0 ? -1 : 0;
265 file_key.objectid = objectid;
266 file_key.offset = offset;
267 file_key.type = BTRFS_EXTENT_DATA_KEY;
269 return btrfs_search_slot(trans, root, &file_key, path, ins_len, cow);
273 * Find checksums for logical bytenr range [disk_bytenr, disk_bytenr + len) and
274 * store the result to @dst.
276 * Return >0 for the number of sectors we found.
277 * Return 0 for the range [disk_bytenr, disk_bytenr + sectorsize) has no csum
278 * for it. Caller may want to try next sector until one range is hit.
279 * Return <0 for fatal error.
281 static int search_csum_tree(struct btrfs_fs_info *fs_info,
282 struct btrfs_path *path, u64 disk_bytenr,
285 struct btrfs_root *csum_root;
286 struct btrfs_csum_item *item = NULL;
287 struct btrfs_key key;
288 const u32 sectorsize = fs_info->sectorsize;
289 const u32 csum_size = fs_info->csum_size;
295 ASSERT(IS_ALIGNED(disk_bytenr, sectorsize) &&
296 IS_ALIGNED(len, sectorsize));
298 /* Check if the current csum item covers disk_bytenr */
299 if (path->nodes[0]) {
300 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
301 struct btrfs_csum_item);
302 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
303 itemsize = btrfs_item_size(path->nodes[0], path->slots[0]);
305 csum_start = key.offset;
306 csum_len = (itemsize / csum_size) * sectorsize;
308 if (in_range(disk_bytenr, csum_start, csum_len))
312 /* Current item doesn't contain the desired range, search again */
313 btrfs_release_path(path);
314 csum_root = btrfs_csum_root(fs_info, disk_bytenr);
315 item = btrfs_lookup_csum(NULL, csum_root, path, disk_bytenr, 0);
320 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
321 itemsize = btrfs_item_size(path->nodes[0], path->slots[0]);
323 csum_start = key.offset;
324 csum_len = (itemsize / csum_size) * sectorsize;
325 ASSERT(in_range(disk_bytenr, csum_start, csum_len));
328 ret = (min(csum_start + csum_len, disk_bytenr + len) -
329 disk_bytenr) >> fs_info->sectorsize_bits;
330 read_extent_buffer(path->nodes[0], dst, (unsigned long)item,
333 if (ret == -ENOENT || ret == -EFBIG)
339 * Locate the file_offset of @cur_disk_bytenr of a @bio.
341 * Bio of btrfs represents read range of
342 * [bi_sector << 9, bi_sector << 9 + bi_size).
343 * Knowing this, we can iterate through each bvec to locate the page belong to
344 * @cur_disk_bytenr and get the file offset.
346 * @inode is used to determine if the bvec page really belongs to @inode.
348 * Return 0 if we can't find the file offset
349 * Return >0 if we find the file offset and restore it to @file_offset_ret
351 static int search_file_offset_in_bio(struct bio *bio, struct inode *inode,
352 u64 disk_bytenr, u64 *file_offset_ret)
354 struct bvec_iter iter;
356 u64 cur = bio->bi_iter.bi_sector << SECTOR_SHIFT;
359 bio_for_each_segment(bvec, bio, iter) {
360 struct page *page = bvec.bv_page;
362 if (cur > disk_bytenr)
364 if (cur + bvec.bv_len <= disk_bytenr) {
368 ASSERT(in_range(disk_bytenr, cur, bvec.bv_len));
369 if (page->mapping && page->mapping->host &&
370 page->mapping->host == inode) {
372 *file_offset_ret = page_offset(page) + bvec.bv_offset +
381 * Lookup the checksum for the read bio in csum tree.
383 * @inode: inode that the bio is for.
384 * @bio: bio to look up.
385 * @dst: Buffer of size nblocks * btrfs_super_csum_size() used to return
386 * checksum (nblocks = bio->bi_iter.bi_size / fs_info->sectorsize). If
387 * NULL, the checksum buffer is allocated and returned in
388 * btrfs_bio(bio)->csum instead.
390 * Return: BLK_STS_RESOURCE if allocating memory fails, BLK_STS_OK otherwise.
392 blk_status_t btrfs_lookup_bio_sums(struct inode *inode, struct bio *bio, u8 *dst)
394 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
395 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
396 struct btrfs_bio *bbio = NULL;
397 struct btrfs_path *path;
398 const u32 sectorsize = fs_info->sectorsize;
399 const u32 csum_size = fs_info->csum_size;
400 u32 orig_len = bio->bi_iter.bi_size;
401 u64 orig_disk_bytenr = bio->bi_iter.bi_sector << SECTOR_SHIFT;
404 const unsigned int nblocks = orig_len >> fs_info->sectorsize_bits;
406 blk_status_t ret = BLK_STS_OK;
408 if ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM) ||
409 test_bit(BTRFS_FS_STATE_NO_CSUMS, &fs_info->fs_state))
413 * This function is only called for read bio.
415 * This means two things:
416 * - All our csums should only be in csum tree
417 * No ordered extents csums, as ordered extents are only for write
419 * - No need to bother any other info from bvec
420 * Since we're looking up csums, the only important info is the
421 * disk_bytenr and the length, which can be extracted from bi_iter
424 ASSERT(bio_op(bio) == REQ_OP_READ);
425 path = btrfs_alloc_path();
427 return BLK_STS_RESOURCE;
430 bbio = btrfs_bio(bio);
432 if (nblocks * csum_size > BTRFS_BIO_INLINE_CSUM_SIZE) {
433 bbio->csum = kmalloc_array(nblocks, csum_size, GFP_NOFS);
435 btrfs_free_path(path);
436 return BLK_STS_RESOURCE;
439 bbio->csum = bbio->csum_inline;
447 * If requested number of sectors is larger than one leaf can contain,
448 * kick the readahead for csum tree.
450 if (nblocks > fs_info->csums_per_leaf)
451 path->reada = READA_FORWARD;
454 * the free space stuff is only read when it hasn't been
455 * updated in the current transaction. So, we can safely
456 * read from the commit root and sidestep a nasty deadlock
457 * between reading the free space cache and updating the csum tree.
459 if (btrfs_is_free_space_inode(BTRFS_I(inode))) {
460 path->search_commit_root = 1;
461 path->skip_locking = 1;
464 for (cur_disk_bytenr = orig_disk_bytenr;
465 cur_disk_bytenr < orig_disk_bytenr + orig_len;
466 cur_disk_bytenr += (count * sectorsize)) {
467 u64 search_len = orig_disk_bytenr + orig_len - cur_disk_bytenr;
468 unsigned int sector_offset;
472 * Although both cur_disk_bytenr and orig_disk_bytenr is u64,
473 * we're calculating the offset to the bio start.
475 * Bio size is limited to UINT_MAX, thus unsigned int is large
476 * enough to contain the raw result, not to mention the right
479 ASSERT(cur_disk_bytenr - orig_disk_bytenr < UINT_MAX);
480 sector_offset = (cur_disk_bytenr - orig_disk_bytenr) >>
481 fs_info->sectorsize_bits;
482 csum_dst = csum + sector_offset * csum_size;
484 count = search_csum_tree(fs_info, path, cur_disk_bytenr,
485 search_len, csum_dst);
487 ret = errno_to_blk_status(count);
489 btrfs_bio_free_csum(bbio);
494 * We didn't find a csum for this range. We need to make sure
495 * we complain loudly about this, because we are not NODATASUM.
497 * However for the DATA_RELOC inode we could potentially be
498 * relocating data extents for a NODATASUM inode, so the inode
499 * itself won't be marked with NODATASUM, but the extent we're
500 * copying is in fact NODATASUM. If we don't find a csum we
501 * assume this is the case.
504 memset(csum_dst, 0, csum_size);
507 if (BTRFS_I(inode)->root->root_key.objectid ==
508 BTRFS_DATA_RELOC_TREE_OBJECTID) {
512 ret = search_file_offset_in_bio(bio, inode,
513 cur_disk_bytenr, &file_offset);
515 set_extent_bits(io_tree, file_offset,
516 file_offset + sectorsize - 1,
519 btrfs_warn_rl(fs_info,
520 "csum hole found for disk bytenr range [%llu, %llu)",
521 cur_disk_bytenr, cur_disk_bytenr + sectorsize);
526 btrfs_free_path(path);
530 int btrfs_lookup_csums_list(struct btrfs_root *root, u64 start, u64 end,
531 struct list_head *list, int search_commit,
534 struct btrfs_fs_info *fs_info = root->fs_info;
535 struct btrfs_key key;
536 struct btrfs_path *path;
537 struct extent_buffer *leaf;
538 struct btrfs_ordered_sum *sums;
539 struct btrfs_csum_item *item;
543 ASSERT(IS_ALIGNED(start, fs_info->sectorsize) &&
544 IS_ALIGNED(end + 1, fs_info->sectorsize));
546 path = btrfs_alloc_path();
550 path->nowait = nowait;
552 path->skip_locking = 1;
553 path->reada = READA_FORWARD;
554 path->search_commit_root = 1;
557 key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
559 key.type = BTRFS_EXTENT_CSUM_KEY;
561 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
564 if (ret > 0 && path->slots[0] > 0) {
565 leaf = path->nodes[0];
566 btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
569 * There are two cases we can hit here for the previous csum
572 * |<- search range ->|
576 * |<- search range ->|
579 * Check if the previous csum item covers the leading part of
580 * the search range. If so we have to start from previous csum
583 if (key.objectid == BTRFS_EXTENT_CSUM_OBJECTID &&
584 key.type == BTRFS_EXTENT_CSUM_KEY) {
585 if (bytes_to_csum_size(fs_info, start - key.offset) <
586 btrfs_item_size(leaf, path->slots[0] - 1))
591 while (start <= end) {
594 leaf = path->nodes[0];
595 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
596 ret = btrfs_next_leaf(root, path);
601 leaf = path->nodes[0];
604 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
605 if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
606 key.type != BTRFS_EXTENT_CSUM_KEY ||
610 if (key.offset > start)
613 csum_end = key.offset + csum_size_to_bytes(fs_info,
614 btrfs_item_size(leaf, path->slots[0]));
615 if (csum_end <= start) {
620 csum_end = min(csum_end, end + 1);
621 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
622 struct btrfs_csum_item);
623 while (start < csum_end) {
624 unsigned long offset;
627 size = min_t(size_t, csum_end - start,
628 max_ordered_sum_bytes(fs_info));
629 sums = kzalloc(btrfs_ordered_sum_size(fs_info, size),
636 sums->bytenr = start;
637 sums->len = (int)size;
639 offset = bytes_to_csum_size(fs_info, start - key.offset);
641 read_extent_buffer(path->nodes[0],
643 ((unsigned long)item) + offset,
644 bytes_to_csum_size(fs_info, size));
647 list_add_tail(&sums->list, &tmplist);
653 while (ret < 0 && !list_empty(&tmplist)) {
654 sums = list_entry(tmplist.next, struct btrfs_ordered_sum, list);
655 list_del(&sums->list);
658 list_splice_tail(&tmplist, list);
660 btrfs_free_path(path);
665 * Do the same work as btrfs_lookup_csums_list(), the difference is in how
666 * we return the result.
668 * This version will set the corresponding bits in @csum_bitmap to represent
669 * that there is a csum found.
670 * Each bit represents a sector. Thus caller should ensure @csum_buf passed
671 * in is large enough to contain all csums.
673 int btrfs_lookup_csums_bitmap(struct btrfs_root *root, u64 start, u64 end,
674 u8 *csum_buf, unsigned long *csum_bitmap)
676 struct btrfs_fs_info *fs_info = root->fs_info;
677 struct btrfs_key key;
678 struct btrfs_path *path;
679 struct extent_buffer *leaf;
680 struct btrfs_csum_item *item;
681 const u64 orig_start = start;
684 ASSERT(IS_ALIGNED(start, fs_info->sectorsize) &&
685 IS_ALIGNED(end + 1, fs_info->sectorsize));
687 path = btrfs_alloc_path();
691 key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
692 key.type = BTRFS_EXTENT_CSUM_KEY;
695 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
698 if (ret > 0 && path->slots[0] > 0) {
699 leaf = path->nodes[0];
700 btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
703 * There are two cases we can hit here for the previous csum
706 * |<- search range ->|
710 * |<- search range ->|
713 * Check if the previous csum item covers the leading part of
714 * the search range. If so we have to start from previous csum
717 if (key.objectid == BTRFS_EXTENT_CSUM_OBJECTID &&
718 key.type == BTRFS_EXTENT_CSUM_KEY) {
719 if (bytes_to_csum_size(fs_info, start - key.offset) <
720 btrfs_item_size(leaf, path->slots[0] - 1))
725 while (start <= end) {
728 leaf = path->nodes[0];
729 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
730 ret = btrfs_next_leaf(root, path);
735 leaf = path->nodes[0];
738 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
739 if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
740 key.type != BTRFS_EXTENT_CSUM_KEY ||
744 if (key.offset > start)
747 csum_end = key.offset + csum_size_to_bytes(fs_info,
748 btrfs_item_size(leaf, path->slots[0]));
749 if (csum_end <= start) {
754 csum_end = min(csum_end, end + 1);
755 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
756 struct btrfs_csum_item);
757 while (start < csum_end) {
758 unsigned long offset;
760 u8 *csum_dest = csum_buf + bytes_to_csum_size(fs_info,
763 size = min_t(size_t, csum_end - start, end + 1 - start);
765 offset = bytes_to_csum_size(fs_info, start - key.offset);
767 read_extent_buffer(path->nodes[0], csum_dest,
768 ((unsigned long)item) + offset,
769 bytes_to_csum_size(fs_info, size));
771 bitmap_set(csum_bitmap,
772 (start - orig_start) >> fs_info->sectorsize_bits,
773 size >> fs_info->sectorsize_bits);
781 btrfs_free_path(path);
786 * Calculate checksums of the data contained inside a bio.
788 * @inode: Owner of the data inside the bio
789 * @bio: Contains the data to be checksummed
790 * @offset: If (u64)-1, @bio may contain discontiguous bio vecs, so the
791 * file offsets are determined from the page offsets in the bio.
792 * Otherwise, this is the starting file offset of the bio vecs in
793 * @bio, which must be contiguous.
794 * @one_ordered: If true, @bio only refers to one ordered extent.
796 blk_status_t btrfs_csum_one_bio(struct btrfs_inode *inode, struct bio *bio,
797 u64 offset, bool one_ordered)
799 struct btrfs_fs_info *fs_info = inode->root->fs_info;
800 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
801 struct btrfs_ordered_sum *sums;
802 struct btrfs_ordered_extent *ordered = NULL;
803 const bool use_page_offsets = (offset == (u64)-1);
805 struct bvec_iter iter;
808 unsigned int blockcount;
809 unsigned long total_bytes = 0;
810 unsigned long this_sum_bytes = 0;
814 nofs_flag = memalloc_nofs_save();
815 sums = kvzalloc(btrfs_ordered_sum_size(fs_info, bio->bi_iter.bi_size),
817 memalloc_nofs_restore(nofs_flag);
820 return BLK_STS_RESOURCE;
822 sums->len = bio->bi_iter.bi_size;
823 INIT_LIST_HEAD(&sums->list);
825 sums->bytenr = bio->bi_iter.bi_sector << 9;
828 shash->tfm = fs_info->csum_shash;
830 bio_for_each_segment(bvec, bio, iter) {
831 if (use_page_offsets)
832 offset = page_offset(bvec.bv_page) + bvec.bv_offset;
835 ordered = btrfs_lookup_ordered_extent(inode, offset);
837 * The bio range is not covered by any ordered extent,
838 * must be a code logic error.
840 if (unlikely(!ordered)) {
842 "no ordered extent for root %llu ino %llu offset %llu\n",
843 inode->root->root_key.objectid,
844 btrfs_ino(inode), offset);
846 return BLK_STS_IOERR;
850 blockcount = BTRFS_BYTES_TO_BLKS(fs_info,
851 bvec.bv_len + fs_info->sectorsize
854 for (i = 0; i < blockcount; i++) {
856 !in_range(offset, ordered->file_offset,
857 ordered->num_bytes)) {
858 unsigned long bytes_left;
860 sums->len = this_sum_bytes;
862 btrfs_add_ordered_sum(ordered, sums);
863 btrfs_put_ordered_extent(ordered);
865 bytes_left = bio->bi_iter.bi_size - total_bytes;
867 nofs_flag = memalloc_nofs_save();
868 sums = kvzalloc(btrfs_ordered_sum_size(fs_info,
869 bytes_left), GFP_KERNEL);
870 memalloc_nofs_restore(nofs_flag);
871 BUG_ON(!sums); /* -ENOMEM */
872 sums->len = bytes_left;
873 ordered = btrfs_lookup_ordered_extent(inode,
875 ASSERT(ordered); /* Logic error */
876 sums->bytenr = (bio->bi_iter.bi_sector << 9)
881 data = bvec_kmap_local(&bvec);
882 crypto_shash_digest(shash,
883 data + (i * fs_info->sectorsize),
887 index += fs_info->csum_size;
888 offset += fs_info->sectorsize;
889 this_sum_bytes += fs_info->sectorsize;
890 total_bytes += fs_info->sectorsize;
895 btrfs_add_ordered_sum(ordered, sums);
896 btrfs_put_ordered_extent(ordered);
901 * Remove one checksum overlapping a range.
903 * This expects the key to describe the csum pointed to by the path, and it
904 * expects the csum to overlap the range [bytenr, len]
906 * The csum should not be entirely contained in the range and the range should
907 * not be entirely contained in the csum.
909 * This calls btrfs_truncate_item with the correct args based on the overlap,
910 * and fixes up the key as required.
912 static noinline void truncate_one_csum(struct btrfs_fs_info *fs_info,
913 struct btrfs_path *path,
914 struct btrfs_key *key,
917 struct extent_buffer *leaf;
918 const u32 csum_size = fs_info->csum_size;
920 u64 end_byte = bytenr + len;
921 u32 blocksize_bits = fs_info->sectorsize_bits;
923 leaf = path->nodes[0];
924 csum_end = btrfs_item_size(leaf, path->slots[0]) / csum_size;
925 csum_end <<= blocksize_bits;
926 csum_end += key->offset;
928 if (key->offset < bytenr && csum_end <= end_byte) {
933 * A simple truncate off the end of the item
935 u32 new_size = (bytenr - key->offset) >> blocksize_bits;
936 new_size *= csum_size;
937 btrfs_truncate_item(path, new_size, 1);
938 } else if (key->offset >= bytenr && csum_end > end_byte &&
939 end_byte > key->offset) {
944 * we need to truncate from the beginning of the csum
946 u32 new_size = (csum_end - end_byte) >> blocksize_bits;
947 new_size *= csum_size;
949 btrfs_truncate_item(path, new_size, 0);
951 key->offset = end_byte;
952 btrfs_set_item_key_safe(fs_info, path, key);
959 * Delete the csum items from the csum tree for a given range of bytes.
961 int btrfs_del_csums(struct btrfs_trans_handle *trans,
962 struct btrfs_root *root, u64 bytenr, u64 len)
964 struct btrfs_fs_info *fs_info = trans->fs_info;
965 struct btrfs_path *path;
966 struct btrfs_key key;
967 u64 end_byte = bytenr + len;
969 struct extent_buffer *leaf;
971 const u32 csum_size = fs_info->csum_size;
972 u32 blocksize_bits = fs_info->sectorsize_bits;
974 ASSERT(root->root_key.objectid == BTRFS_CSUM_TREE_OBJECTID ||
975 root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID);
977 path = btrfs_alloc_path();
982 key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
983 key.offset = end_byte - 1;
984 key.type = BTRFS_EXTENT_CSUM_KEY;
986 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
989 if (path->slots[0] == 0)
992 } else if (ret < 0) {
996 leaf = path->nodes[0];
997 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
999 if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
1000 key.type != BTRFS_EXTENT_CSUM_KEY) {
1004 if (key.offset >= end_byte)
1007 csum_end = btrfs_item_size(leaf, path->slots[0]) / csum_size;
1008 csum_end <<= blocksize_bits;
1009 csum_end += key.offset;
1011 /* this csum ends before we start, we're done */
1012 if (csum_end <= bytenr)
1015 /* delete the entire item, it is inside our range */
1016 if (key.offset >= bytenr && csum_end <= end_byte) {
1020 * Check how many csum items preceding this one in this
1021 * leaf correspond to our range and then delete them all
1024 if (key.offset > bytenr && path->slots[0] > 0) {
1025 int slot = path->slots[0] - 1;
1028 struct btrfs_key pk;
1030 btrfs_item_key_to_cpu(leaf, &pk, slot);
1031 if (pk.offset < bytenr ||
1032 pk.type != BTRFS_EXTENT_CSUM_KEY ||
1034 BTRFS_EXTENT_CSUM_OBJECTID)
1036 path->slots[0] = slot;
1038 key.offset = pk.offset;
1042 ret = btrfs_del_items(trans, root, path,
1043 path->slots[0], del_nr);
1046 if (key.offset == bytenr)
1048 } else if (key.offset < bytenr && csum_end > end_byte) {
1049 unsigned long offset;
1050 unsigned long shift_len;
1051 unsigned long item_offset;
1056 * Our bytes are in the middle of the csum,
1057 * we need to split this item and insert a new one.
1059 * But we can't drop the path because the
1060 * csum could change, get removed, extended etc.
1062 * The trick here is the max size of a csum item leaves
1063 * enough room in the tree block for a single
1064 * item header. So, we split the item in place,
1065 * adding a new header pointing to the existing
1066 * bytes. Then we loop around again and we have
1067 * a nicely formed csum item that we can neatly
1070 offset = (bytenr - key.offset) >> blocksize_bits;
1071 offset *= csum_size;
1073 shift_len = (len >> blocksize_bits) * csum_size;
1075 item_offset = btrfs_item_ptr_offset(leaf,
1078 memzero_extent_buffer(leaf, item_offset + offset,
1080 key.offset = bytenr;
1083 * btrfs_split_item returns -EAGAIN when the
1084 * item changed size or key
1086 ret = btrfs_split_item(trans, root, path, &key, offset);
1087 if (ret && ret != -EAGAIN) {
1088 btrfs_abort_transaction(trans, ret);
1093 key.offset = end_byte - 1;
1095 truncate_one_csum(fs_info, path, &key, bytenr, len);
1096 if (key.offset < bytenr)
1099 btrfs_release_path(path);
1101 btrfs_free_path(path);
1105 static int find_next_csum_offset(struct btrfs_root *root,
1106 struct btrfs_path *path,
1109 const u32 nritems = btrfs_header_nritems(path->nodes[0]);
1110 struct btrfs_key found_key;
1111 int slot = path->slots[0] + 1;
1114 if (nritems == 0 || slot >= nritems) {
1115 ret = btrfs_next_leaf(root, path);
1118 } else if (ret > 0) {
1119 *next_offset = (u64)-1;
1122 slot = path->slots[0];
1125 btrfs_item_key_to_cpu(path->nodes[0], &found_key, slot);
1127 if (found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
1128 found_key.type != BTRFS_EXTENT_CSUM_KEY)
1129 *next_offset = (u64)-1;
1131 *next_offset = found_key.offset;
1136 int btrfs_csum_file_blocks(struct btrfs_trans_handle *trans,
1137 struct btrfs_root *root,
1138 struct btrfs_ordered_sum *sums)
1140 struct btrfs_fs_info *fs_info = root->fs_info;
1141 struct btrfs_key file_key;
1142 struct btrfs_key found_key;
1143 struct btrfs_path *path;
1144 struct btrfs_csum_item *item;
1145 struct btrfs_csum_item *item_end;
1146 struct extent_buffer *leaf = NULL;
1148 u64 total_bytes = 0;
1155 const u32 csum_size = fs_info->csum_size;
1157 path = btrfs_alloc_path();
1161 next_offset = (u64)-1;
1163 bytenr = sums->bytenr + total_bytes;
1164 file_key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
1165 file_key.offset = bytenr;
1166 file_key.type = BTRFS_EXTENT_CSUM_KEY;
1168 item = btrfs_lookup_csum(trans, root, path, bytenr, 1);
1169 if (!IS_ERR(item)) {
1171 leaf = path->nodes[0];
1172 item_end = btrfs_item_ptr(leaf, path->slots[0],
1173 struct btrfs_csum_item);
1174 item_end = (struct btrfs_csum_item *)((char *)item_end +
1175 btrfs_item_size(leaf, path->slots[0]));
1178 ret = PTR_ERR(item);
1179 if (ret != -EFBIG && ret != -ENOENT)
1182 if (ret == -EFBIG) {
1184 /* we found one, but it isn't big enough yet */
1185 leaf = path->nodes[0];
1186 item_size = btrfs_item_size(leaf, path->slots[0]);
1187 if ((item_size / csum_size) >=
1188 MAX_CSUM_ITEMS(fs_info, csum_size)) {
1189 /* already at max size, make a new one */
1193 /* We didn't find a csum item, insert one. */
1194 ret = find_next_csum_offset(root, path, &next_offset);
1202 * At this point, we know the tree has a checksum item that ends at an
1203 * offset matching the start of the checksum range we want to insert.
1204 * We try to extend that item as much as possible and then add as many
1205 * checksums to it as they fit.
1207 * First check if the leaf has enough free space for at least one
1208 * checksum. If it has go directly to the item extension code, otherwise
1209 * release the path and do a search for insertion before the extension.
1211 if (btrfs_leaf_free_space(leaf) >= csum_size) {
1212 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1213 csum_offset = (bytenr - found_key.offset) >>
1214 fs_info->sectorsize_bits;
1218 btrfs_release_path(path);
1219 path->search_for_extension = 1;
1220 ret = btrfs_search_slot(trans, root, &file_key, path,
1222 path->search_for_extension = 0;
1227 if (path->slots[0] == 0)
1232 leaf = path->nodes[0];
1233 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1234 csum_offset = (bytenr - found_key.offset) >> fs_info->sectorsize_bits;
1236 if (found_key.type != BTRFS_EXTENT_CSUM_KEY ||
1237 found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
1238 csum_offset >= MAX_CSUM_ITEMS(fs_info, csum_size)) {
1243 if (csum_offset == btrfs_item_size(leaf, path->slots[0]) /
1249 tmp = sums->len - total_bytes;
1250 tmp >>= fs_info->sectorsize_bits;
1252 extend_nr = max_t(int, 1, tmp);
1255 * A log tree can already have checksum items with a subset of
1256 * the checksums we are trying to log. This can happen after
1257 * doing a sequence of partial writes into prealloc extents and
1258 * fsyncs in between, with a full fsync logging a larger subrange
1259 * of an extent for which a previous fast fsync logged a smaller
1260 * subrange. And this happens in particular due to merging file
1261 * extent items when we complete an ordered extent for a range
1262 * covered by a prealloc extent - this is done at
1263 * btrfs_mark_extent_written().
1265 * So if we try to extend the previous checksum item, which has
1266 * a range that ends at the start of the range we want to insert,
1267 * make sure we don't extend beyond the start offset of the next
1268 * checksum item. If we are at the last item in the leaf, then
1269 * forget the optimization of extending and add a new checksum
1270 * item - it is not worth the complexity of releasing the path,
1271 * getting the first key for the next leaf, repeat the btree
1272 * search, etc, because log trees are temporary anyway and it
1273 * would only save a few bytes of leaf space.
1275 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
1276 if (path->slots[0] + 1 >=
1277 btrfs_header_nritems(path->nodes[0])) {
1278 ret = find_next_csum_offset(root, path, &next_offset);
1285 ret = find_next_csum_offset(root, path, &next_offset);
1289 tmp = (next_offset - bytenr) >> fs_info->sectorsize_bits;
1291 extend_nr = min_t(int, extend_nr, tmp);
1294 diff = (csum_offset + extend_nr) * csum_size;
1296 MAX_CSUM_ITEMS(fs_info, csum_size) * csum_size);
1298 diff = diff - btrfs_item_size(leaf, path->slots[0]);
1299 diff = min_t(u32, btrfs_leaf_free_space(leaf), diff);
1303 btrfs_extend_item(path, diff);
1309 btrfs_release_path(path);
1314 tmp = sums->len - total_bytes;
1315 tmp >>= fs_info->sectorsize_bits;
1316 tmp = min(tmp, (next_offset - file_key.offset) >>
1317 fs_info->sectorsize_bits);
1319 tmp = max_t(u64, 1, tmp);
1320 tmp = min_t(u64, tmp, MAX_CSUM_ITEMS(fs_info, csum_size));
1321 ins_size = csum_size * tmp;
1323 ins_size = csum_size;
1325 ret = btrfs_insert_empty_item(trans, root, path, &file_key,
1329 if (WARN_ON(ret != 0))
1331 leaf = path->nodes[0];
1333 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item);
1334 item_end = (struct btrfs_csum_item *)((unsigned char *)item +
1335 btrfs_item_size(leaf, path->slots[0]));
1336 item = (struct btrfs_csum_item *)((unsigned char *)item +
1337 csum_offset * csum_size);
1339 ins_size = (u32)(sums->len - total_bytes) >> fs_info->sectorsize_bits;
1340 ins_size *= csum_size;
1341 ins_size = min_t(u32, (unsigned long)item_end - (unsigned long)item,
1343 write_extent_buffer(leaf, sums->sums + index, (unsigned long)item,
1347 ins_size /= csum_size;
1348 total_bytes += ins_size * fs_info->sectorsize;
1350 btrfs_mark_buffer_dirty(path->nodes[0]);
1351 if (total_bytes < sums->len) {
1352 btrfs_release_path(path);
1357 btrfs_free_path(path);
1361 void btrfs_extent_item_to_extent_map(struct btrfs_inode *inode,
1362 const struct btrfs_path *path,
1363 struct btrfs_file_extent_item *fi,
1364 struct extent_map *em)
1366 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1367 struct btrfs_root *root = inode->root;
1368 struct extent_buffer *leaf = path->nodes[0];
1369 const int slot = path->slots[0];
1370 struct btrfs_key key;
1371 u64 extent_start, extent_end;
1373 u8 type = btrfs_file_extent_type(leaf, fi);
1374 int compress_type = btrfs_file_extent_compression(leaf, fi);
1376 btrfs_item_key_to_cpu(leaf, &key, slot);
1377 extent_start = key.offset;
1378 extent_end = btrfs_file_extent_end(path);
1379 em->ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1380 em->generation = btrfs_file_extent_generation(leaf, fi);
1381 if (type == BTRFS_FILE_EXTENT_REG ||
1382 type == BTRFS_FILE_EXTENT_PREALLOC) {
1383 em->start = extent_start;
1384 em->len = extent_end - extent_start;
1385 em->orig_start = extent_start -
1386 btrfs_file_extent_offset(leaf, fi);
1387 em->orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
1388 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1390 em->block_start = EXTENT_MAP_HOLE;
1393 if (compress_type != BTRFS_COMPRESS_NONE) {
1394 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
1395 em->compress_type = compress_type;
1396 em->block_start = bytenr;
1397 em->block_len = em->orig_block_len;
1399 bytenr += btrfs_file_extent_offset(leaf, fi);
1400 em->block_start = bytenr;
1401 em->block_len = em->len;
1402 if (type == BTRFS_FILE_EXTENT_PREALLOC)
1403 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
1405 } else if (type == BTRFS_FILE_EXTENT_INLINE) {
1406 em->block_start = EXTENT_MAP_INLINE;
1407 em->start = extent_start;
1408 em->len = extent_end - extent_start;
1410 * Initialize orig_start and block_len with the same values
1411 * as in inode.c:btrfs_get_extent().
1413 em->orig_start = EXTENT_MAP_HOLE;
1414 em->block_len = (u64)-1;
1415 em->compress_type = compress_type;
1416 if (compress_type != BTRFS_COMPRESS_NONE)
1417 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
1420 "unknown file extent item type %d, inode %llu, offset %llu, "
1421 "root %llu", type, btrfs_ino(inode), extent_start,
1422 root->root_key.objectid);
1427 * Returns the end offset (non inclusive) of the file extent item the given path
1428 * points to. If it points to an inline extent, the returned offset is rounded
1429 * up to the sector size.
1431 u64 btrfs_file_extent_end(const struct btrfs_path *path)
1433 const struct extent_buffer *leaf = path->nodes[0];
1434 const int slot = path->slots[0];
1435 struct btrfs_file_extent_item *fi;
1436 struct btrfs_key key;
1439 btrfs_item_key_to_cpu(leaf, &key, slot);
1440 ASSERT(key.type == BTRFS_EXTENT_DATA_KEY);
1441 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
1443 if (btrfs_file_extent_type(leaf, fi) == BTRFS_FILE_EXTENT_INLINE) {
1444 end = btrfs_file_extent_ram_bytes(leaf, fi);
1445 end = ALIGN(key.offset + end, leaf->fs_info->sectorsize);
1447 end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);