2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
19 #include <linux/bio.h>
20 #include <linux/slab.h>
21 #include <linux/pagemap.h>
22 #include <linux/highmem.h>
25 #include "transaction.h"
26 #include "print-tree.h"
28 #define MAX_CSUM_ITEMS(r, size) ((((BTRFS_LEAF_DATA_SIZE(r) - \
29 sizeof(struct btrfs_item) * 2) / \
32 #define MAX_ORDERED_SUM_BYTES(r) ((PAGE_SIZE - \
33 sizeof(struct btrfs_ordered_sum)) / \
34 sizeof(struct btrfs_sector_sum) * \
35 (r)->sectorsize - (r)->sectorsize)
37 int btrfs_insert_file_extent(struct btrfs_trans_handle *trans,
38 struct btrfs_root *root,
39 u64 objectid, u64 pos,
40 u64 disk_offset, u64 disk_num_bytes,
41 u64 num_bytes, u64 offset, u64 ram_bytes,
42 u8 compression, u8 encryption, u16 other_encoding)
45 struct btrfs_file_extent_item *item;
46 struct btrfs_key file_key;
47 struct btrfs_path *path;
48 struct extent_buffer *leaf;
50 path = btrfs_alloc_path();
53 file_key.objectid = objectid;
54 file_key.offset = pos;
55 btrfs_set_key_type(&file_key, BTRFS_EXTENT_DATA_KEY);
57 path->leave_spinning = 1;
58 ret = btrfs_insert_empty_item(trans, root, path, &file_key,
63 leaf = path->nodes[0];
64 item = btrfs_item_ptr(leaf, path->slots[0],
65 struct btrfs_file_extent_item);
66 btrfs_set_file_extent_disk_bytenr(leaf, item, disk_offset);
67 btrfs_set_file_extent_disk_num_bytes(leaf, item, disk_num_bytes);
68 btrfs_set_file_extent_offset(leaf, item, offset);
69 btrfs_set_file_extent_num_bytes(leaf, item, num_bytes);
70 btrfs_set_file_extent_ram_bytes(leaf, item, ram_bytes);
71 btrfs_set_file_extent_generation(leaf, item, trans->transid);
72 btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
73 btrfs_set_file_extent_compression(leaf, item, compression);
74 btrfs_set_file_extent_encryption(leaf, item, encryption);
75 btrfs_set_file_extent_other_encoding(leaf, item, other_encoding);
77 btrfs_mark_buffer_dirty(leaf);
79 btrfs_free_path(path);
83 struct btrfs_csum_item *btrfs_lookup_csum(struct btrfs_trans_handle *trans,
84 struct btrfs_root *root,
85 struct btrfs_path *path,
89 struct btrfs_key file_key;
90 struct btrfs_key found_key;
91 struct btrfs_csum_item *item;
92 struct extent_buffer *leaf;
95 btrfs_super_csum_size(&root->fs_info->super_copy);
98 file_key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
99 file_key.offset = bytenr;
100 btrfs_set_key_type(&file_key, BTRFS_EXTENT_CSUM_KEY);
101 ret = btrfs_search_slot(trans, root, &file_key, path, 0, cow);
104 leaf = path->nodes[0];
107 if (path->slots[0] == 0)
110 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
111 if (btrfs_key_type(&found_key) != BTRFS_EXTENT_CSUM_KEY)
114 csum_offset = (bytenr - found_key.offset) >>
115 root->fs_info->sb->s_blocksize_bits;
116 csums_in_item = btrfs_item_size_nr(leaf, path->slots[0]);
117 csums_in_item /= csum_size;
119 if (csum_offset >= csums_in_item) {
124 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item);
125 item = (struct btrfs_csum_item *)((unsigned char *)item +
126 csum_offset * csum_size);
135 int btrfs_lookup_file_extent(struct btrfs_trans_handle *trans,
136 struct btrfs_root *root,
137 struct btrfs_path *path, u64 objectid,
141 struct btrfs_key file_key;
142 int ins_len = mod < 0 ? -1 : 0;
145 file_key.objectid = objectid;
146 file_key.offset = offset;
147 btrfs_set_key_type(&file_key, BTRFS_EXTENT_DATA_KEY);
148 ret = btrfs_search_slot(trans, root, &file_key, path, ins_len, cow);
153 static int __btrfs_lookup_bio_sums(struct btrfs_root *root,
154 struct inode *inode, struct bio *bio,
155 u64 logical_offset, u32 *dst, int dio)
158 struct bio_vec *bvec = bio->bi_io_vec;
161 u64 item_start_offset = 0;
162 u64 item_last_offset = 0;
166 btrfs_super_csum_size(&root->fs_info->super_copy);
168 struct btrfs_path *path;
169 struct btrfs_csum_item *item = NULL;
170 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
172 path = btrfs_alloc_path();
175 if (bio->bi_size > PAGE_CACHE_SIZE * 8)
178 WARN_ON(bio->bi_vcnt <= 0);
180 disk_bytenr = (u64)bio->bi_sector << 9;
182 offset = logical_offset;
183 while (bio_index < bio->bi_vcnt) {
185 offset = page_offset(bvec->bv_page) + bvec->bv_offset;
186 ret = btrfs_find_ordered_sum(inode, offset, disk_bytenr, &sum);
190 if (!item || disk_bytenr < item_start_offset ||
191 disk_bytenr >= item_last_offset) {
192 struct btrfs_key found_key;
196 btrfs_release_path(path);
197 item = btrfs_lookup_csum(NULL, root->fs_info->csum_root,
198 path, disk_bytenr, 0);
201 if (ret == -ENOENT || ret == -EFBIG)
204 if (BTRFS_I(inode)->root->root_key.objectid ==
205 BTRFS_DATA_RELOC_TREE_OBJECTID) {
206 set_extent_bits(io_tree, offset,
207 offset + bvec->bv_len - 1,
208 EXTENT_NODATASUM, GFP_NOFS);
210 printk(KERN_INFO "btrfs no csum found "
211 "for inode %llu start %llu\n",
214 (unsigned long long)offset);
217 btrfs_release_path(path);
220 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
223 item_start_offset = found_key.offset;
224 item_size = btrfs_item_size_nr(path->nodes[0],
226 item_last_offset = item_start_offset +
227 (item_size / csum_size) *
229 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
230 struct btrfs_csum_item);
233 * this byte range must be able to fit inside
234 * a single leaf so it will also fit inside a u32
236 diff = disk_bytenr - item_start_offset;
237 diff = diff / root->sectorsize;
238 diff = diff * csum_size;
240 read_extent_buffer(path->nodes[0], &sum,
241 ((unsigned long)item) + diff,
247 set_state_private(io_tree, offset, sum);
248 disk_bytenr += bvec->bv_len;
249 offset += bvec->bv_len;
253 btrfs_free_path(path);
257 int btrfs_lookup_bio_sums(struct btrfs_root *root, struct inode *inode,
258 struct bio *bio, u32 *dst)
260 return __btrfs_lookup_bio_sums(root, inode, bio, 0, dst, 0);
263 int btrfs_lookup_bio_sums_dio(struct btrfs_root *root, struct inode *inode,
264 struct bio *bio, u64 offset, u32 *dst)
266 return __btrfs_lookup_bio_sums(root, inode, bio, offset, dst, 1);
269 int btrfs_lookup_csums_range(struct btrfs_root *root, u64 start, u64 end,
270 struct list_head *list, int search_commit)
272 struct btrfs_key key;
273 struct btrfs_path *path;
274 struct extent_buffer *leaf;
275 struct btrfs_ordered_sum *sums;
276 struct btrfs_sector_sum *sector_sum;
277 struct btrfs_csum_item *item;
278 unsigned long offset;
282 u16 csum_size = btrfs_super_csum_size(&root->fs_info->super_copy);
284 path = btrfs_alloc_path();
289 path->skip_locking = 1;
291 path->search_commit_root = 1;
294 key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
296 key.type = BTRFS_EXTENT_CSUM_KEY;
298 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
301 if (ret > 0 && path->slots[0] > 0) {
302 leaf = path->nodes[0];
303 btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
304 if (key.objectid == BTRFS_EXTENT_CSUM_OBJECTID &&
305 key.type == BTRFS_EXTENT_CSUM_KEY) {
306 offset = (start - key.offset) >>
307 root->fs_info->sb->s_blocksize_bits;
308 if (offset * csum_size <
309 btrfs_item_size_nr(leaf, path->slots[0] - 1))
314 while (start <= end) {
315 leaf = path->nodes[0];
316 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
317 ret = btrfs_next_leaf(root, path);
322 leaf = path->nodes[0];
325 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
326 if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
327 key.type != BTRFS_EXTENT_CSUM_KEY)
330 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
331 if (key.offset > end)
334 if (key.offset > start)
337 size = btrfs_item_size_nr(leaf, path->slots[0]);
338 csum_end = key.offset + (size / csum_size) * root->sectorsize;
339 if (csum_end <= start) {
344 csum_end = min(csum_end, end + 1);
345 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
346 struct btrfs_csum_item);
347 while (start < csum_end) {
348 size = min_t(size_t, csum_end - start,
349 MAX_ORDERED_SUM_BYTES(root));
350 sums = kzalloc(btrfs_ordered_sum_size(root, size),
354 sector_sum = sums->sums;
355 sums->bytenr = start;
358 offset = (start - key.offset) >>
359 root->fs_info->sb->s_blocksize_bits;
363 read_extent_buffer(path->nodes[0],
365 ((unsigned long)item) +
367 sector_sum->bytenr = start;
369 size -= root->sectorsize;
370 start += root->sectorsize;
374 list_add_tail(&sums->list, list);
380 btrfs_free_path(path);
384 int btrfs_csum_one_bio(struct btrfs_root *root, struct inode *inode,
385 struct bio *bio, u64 file_start, int contig)
387 struct btrfs_ordered_sum *sums;
388 struct btrfs_sector_sum *sector_sum;
389 struct btrfs_ordered_extent *ordered;
391 struct bio_vec *bvec = bio->bi_io_vec;
393 unsigned long total_bytes = 0;
394 unsigned long this_sum_bytes = 0;
398 WARN_ON(bio->bi_vcnt <= 0);
399 sums = kzalloc(btrfs_ordered_sum_size(root, bio->bi_size), GFP_NOFS);
403 sector_sum = sums->sums;
404 disk_bytenr = (u64)bio->bi_sector << 9;
405 sums->len = bio->bi_size;
406 INIT_LIST_HEAD(&sums->list);
411 offset = page_offset(bvec->bv_page) + bvec->bv_offset;
413 ordered = btrfs_lookup_ordered_extent(inode, offset);
415 sums->bytenr = ordered->start;
417 while (bio_index < bio->bi_vcnt) {
419 offset = page_offset(bvec->bv_page) + bvec->bv_offset;
421 if (!contig && (offset >= ordered->file_offset + ordered->len ||
422 offset < ordered->file_offset)) {
423 unsigned long bytes_left;
424 sums->len = this_sum_bytes;
426 btrfs_add_ordered_sum(inode, ordered, sums);
427 btrfs_put_ordered_extent(ordered);
429 bytes_left = bio->bi_size - total_bytes;
431 sums = kzalloc(btrfs_ordered_sum_size(root, bytes_left),
434 sector_sum = sums->sums;
435 sums->len = bytes_left;
436 ordered = btrfs_lookup_ordered_extent(inode, offset);
438 sums->bytenr = ordered->start;
441 data = kmap_atomic(bvec->bv_page, KM_USER0);
442 sector_sum->sum = ~(u32)0;
443 sector_sum->sum = btrfs_csum_data(root,
444 data + bvec->bv_offset,
447 kunmap_atomic(data, KM_USER0);
448 btrfs_csum_final(sector_sum->sum,
449 (char *)§or_sum->sum);
450 sector_sum->bytenr = disk_bytenr;
454 total_bytes += bvec->bv_len;
455 this_sum_bytes += bvec->bv_len;
456 disk_bytenr += bvec->bv_len;
457 offset += bvec->bv_len;
461 btrfs_add_ordered_sum(inode, ordered, sums);
462 btrfs_put_ordered_extent(ordered);
467 * helper function for csum removal, this expects the
468 * key to describe the csum pointed to by the path, and it expects
469 * the csum to overlap the range [bytenr, len]
471 * The csum should not be entirely contained in the range and the
472 * range should not be entirely contained in the csum.
474 * This calls btrfs_truncate_item with the correct args based on the
475 * overlap, and fixes up the key as required.
477 static noinline int truncate_one_csum(struct btrfs_trans_handle *trans,
478 struct btrfs_root *root,
479 struct btrfs_path *path,
480 struct btrfs_key *key,
483 struct extent_buffer *leaf;
485 btrfs_super_csum_size(&root->fs_info->super_copy);
487 u64 end_byte = bytenr + len;
488 u32 blocksize_bits = root->fs_info->sb->s_blocksize_bits;
491 leaf = path->nodes[0];
492 csum_end = btrfs_item_size_nr(leaf, path->slots[0]) / csum_size;
493 csum_end <<= root->fs_info->sb->s_blocksize_bits;
494 csum_end += key->offset;
496 if (key->offset < bytenr && csum_end <= end_byte) {
501 * A simple truncate off the end of the item
503 u32 new_size = (bytenr - key->offset) >> blocksize_bits;
504 new_size *= csum_size;
505 ret = btrfs_truncate_item(trans, root, path, new_size, 1);
506 } else if (key->offset >= bytenr && csum_end > end_byte &&
507 end_byte > key->offset) {
512 * we need to truncate from the beginning of the csum
514 u32 new_size = (csum_end - end_byte) >> blocksize_bits;
515 new_size *= csum_size;
517 ret = btrfs_truncate_item(trans, root, path, new_size, 0);
519 key->offset = end_byte;
520 ret = btrfs_set_item_key_safe(trans, root, path, key);
529 * deletes the csum items from the csum tree for a given
532 int btrfs_del_csums(struct btrfs_trans_handle *trans,
533 struct btrfs_root *root, u64 bytenr, u64 len)
535 struct btrfs_path *path;
536 struct btrfs_key key;
537 u64 end_byte = bytenr + len;
539 struct extent_buffer *leaf;
542 btrfs_super_csum_size(&root->fs_info->super_copy);
543 int blocksize_bits = root->fs_info->sb->s_blocksize_bits;
545 root = root->fs_info->csum_root;
547 path = btrfs_alloc_path();
552 key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
553 key.offset = end_byte - 1;
554 key.type = BTRFS_EXTENT_CSUM_KEY;
556 path->leave_spinning = 1;
557 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
559 if (path->slots[0] == 0)
562 } else if (ret < 0) {
566 leaf = path->nodes[0];
567 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
569 if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
570 key.type != BTRFS_EXTENT_CSUM_KEY) {
574 if (key.offset >= end_byte)
577 csum_end = btrfs_item_size_nr(leaf, path->slots[0]) / csum_size;
578 csum_end <<= blocksize_bits;
579 csum_end += key.offset;
581 /* this csum ends before we start, we're done */
582 if (csum_end <= bytenr)
585 /* delete the entire item, it is inside our range */
586 if (key.offset >= bytenr && csum_end <= end_byte) {
587 ret = btrfs_del_item(trans, root, path);
590 if (key.offset == bytenr)
592 } else if (key.offset < bytenr && csum_end > end_byte) {
593 unsigned long offset;
594 unsigned long shift_len;
595 unsigned long item_offset;
600 * Our bytes are in the middle of the csum,
601 * we need to split this item and insert a new one.
603 * But we can't drop the path because the
604 * csum could change, get removed, extended etc.
606 * The trick here is the max size of a csum item leaves
607 * enough room in the tree block for a single
608 * item header. So, we split the item in place,
609 * adding a new header pointing to the existing
610 * bytes. Then we loop around again and we have
611 * a nicely formed csum item that we can neatly
614 offset = (bytenr - key.offset) >> blocksize_bits;
617 shift_len = (len >> blocksize_bits) * csum_size;
619 item_offset = btrfs_item_ptr_offset(leaf,
622 memset_extent_buffer(leaf, 0, item_offset + offset,
627 * btrfs_split_item returns -EAGAIN when the
628 * item changed size or key
630 ret = btrfs_split_item(trans, root, path, &key, offset);
631 BUG_ON(ret && ret != -EAGAIN);
633 key.offset = end_byte - 1;
635 ret = truncate_one_csum(trans, root, path,
638 if (key.offset < bytenr)
641 btrfs_release_path(path);
645 btrfs_free_path(path);
649 int btrfs_csum_file_blocks(struct btrfs_trans_handle *trans,
650 struct btrfs_root *root,
651 struct btrfs_ordered_sum *sums)
655 struct btrfs_key file_key;
656 struct btrfs_key found_key;
660 struct btrfs_path *path;
661 struct btrfs_csum_item *item;
662 struct btrfs_csum_item *item_end;
663 struct extent_buffer *leaf = NULL;
665 struct btrfs_sector_sum *sector_sum;
670 unsigned long map_len;
671 unsigned long map_start;
673 btrfs_super_csum_size(&root->fs_info->super_copy);
675 path = btrfs_alloc_path();
679 sector_sum = sums->sums;
681 next_offset = (u64)-1;
683 file_key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
684 file_key.offset = sector_sum->bytenr;
685 bytenr = sector_sum->bytenr;
686 btrfs_set_key_type(&file_key, BTRFS_EXTENT_CSUM_KEY);
688 item = btrfs_lookup_csum(trans, root, path, sector_sum->bytenr, 1);
690 leaf = path->nodes[0];
695 if (ret != -EFBIG && ret != -ENOENT)
700 /* we found one, but it isn't big enough yet */
701 leaf = path->nodes[0];
702 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
703 if ((item_size / csum_size) >=
704 MAX_CSUM_ITEMS(root, csum_size)) {
705 /* already at max size, make a new one */
709 int slot = path->slots[0] + 1;
710 /* we didn't find a csum item, insert one */
711 nritems = btrfs_header_nritems(path->nodes[0]);
712 if (path->slots[0] >= nritems - 1) {
713 ret = btrfs_next_leaf(root, path);
720 btrfs_item_key_to_cpu(path->nodes[0], &found_key, slot);
721 if (found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
722 found_key.type != BTRFS_EXTENT_CSUM_KEY) {
726 next_offset = found_key.offset;
732 * at this point, we know the tree has an item, but it isn't big
733 * enough yet to put our csum in. Grow it
735 btrfs_release_path(path);
736 ret = btrfs_search_slot(trans, root, &file_key, path,
742 if (path->slots[0] == 0)
747 leaf = path->nodes[0];
748 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
749 csum_offset = (bytenr - found_key.offset) >>
750 root->fs_info->sb->s_blocksize_bits;
752 if (btrfs_key_type(&found_key) != BTRFS_EXTENT_CSUM_KEY ||
753 found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
754 csum_offset >= MAX_CSUM_ITEMS(root, csum_size)) {
758 if (csum_offset >= btrfs_item_size_nr(leaf, path->slots[0]) /
760 u32 diff = (csum_offset + 1) * csum_size;
763 * is the item big enough already? we dropped our lock
764 * before and need to recheck
766 if (diff < btrfs_item_size_nr(leaf, path->slots[0]))
769 diff = diff - btrfs_item_size_nr(leaf, path->slots[0]);
770 if (diff != csum_size)
773 ret = btrfs_extend_item(trans, root, path, diff);
778 btrfs_release_path(path);
781 u64 tmp = total_bytes + root->sectorsize;
782 u64 next_sector = sector_sum->bytenr;
783 struct btrfs_sector_sum *next = sector_sum + 1;
785 while (tmp < sums->len) {
786 if (next_sector + root->sectorsize != next->bytenr)
788 tmp += root->sectorsize;
789 next_sector = next->bytenr;
792 tmp = min(tmp, next_offset - file_key.offset);
793 tmp >>= root->fs_info->sb->s_blocksize_bits;
794 tmp = max((u64)1, tmp);
795 tmp = min(tmp, (u64)MAX_CSUM_ITEMS(root, csum_size));
796 ins_size = csum_size * tmp;
798 ins_size = csum_size;
800 path->leave_spinning = 1;
801 ret = btrfs_insert_empty_item(trans, root, path, &file_key,
803 path->leave_spinning = 0;
811 leaf = path->nodes[0];
812 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item);
814 item = (struct btrfs_csum_item *)((unsigned char *)item +
815 csum_offset * csum_size);
817 item_end = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item);
818 item_end = (struct btrfs_csum_item *)((unsigned char *)item_end +
819 btrfs_item_size_nr(leaf, path->slots[0]));
824 (unsigned long)item + csum_size >= map_start + map_len) {
828 unmap_extent_buffer(leaf, eb_token, KM_USER1);
830 err = map_private_extent_buffer(leaf, (unsigned long)item,
833 &map_start, &map_len, KM_USER1);
838 memcpy(eb_token + ((unsigned long)item & (PAGE_CACHE_SIZE - 1)),
839 §or_sum->sum, csum_size);
841 write_extent_buffer(leaf, §or_sum->sum,
842 (unsigned long)item, csum_size);
845 total_bytes += root->sectorsize;
847 if (total_bytes < sums->len) {
848 item = (struct btrfs_csum_item *)((char *)item +
850 if (item < item_end && bytenr + PAGE_CACHE_SIZE ==
851 sector_sum->bytenr) {
852 bytenr = sector_sum->bytenr;
857 unmap_extent_buffer(leaf, eb_token, KM_USER1);
860 btrfs_mark_buffer_dirty(path->nodes[0]);
861 if (total_bytes < sums->len) {
862 btrfs_release_path(path);
867 btrfs_free_path(path);