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/pagemap.h>
21 #include <linux/highmem.h>
24 #include "transaction.h"
25 #include "print-tree.h"
27 #define MAX_CSUM_ITEMS(r, size) ((((BTRFS_LEAF_DATA_SIZE(r) - \
28 sizeof(struct btrfs_item) * 2) / \
30 int btrfs_insert_file_extent(struct btrfs_trans_handle *trans,
31 struct btrfs_root *root,
32 u64 objectid, u64 pos,
33 u64 disk_offset, u64 disk_num_bytes,
34 u64 num_bytes, u64 offset, u64 ram_bytes,
35 u8 compression, u8 encryption, u16 other_encoding)
38 struct btrfs_file_extent_item *item;
39 struct btrfs_key file_key;
40 struct btrfs_path *path;
41 struct extent_buffer *leaf;
43 path = btrfs_alloc_path();
45 file_key.objectid = objectid;
46 file_key.offset = pos;
47 btrfs_set_key_type(&file_key, BTRFS_EXTENT_DATA_KEY);
49 ret = btrfs_insert_empty_item(trans, root, path, &file_key,
54 leaf = path->nodes[0];
55 item = btrfs_item_ptr(leaf, path->slots[0],
56 struct btrfs_file_extent_item);
57 btrfs_set_file_extent_disk_bytenr(leaf, item, disk_offset);
58 btrfs_set_file_extent_disk_num_bytes(leaf, item, disk_num_bytes);
59 btrfs_set_file_extent_offset(leaf, item, offset);
60 btrfs_set_file_extent_num_bytes(leaf, item, num_bytes);
61 btrfs_set_file_extent_ram_bytes(leaf, item, ram_bytes);
62 btrfs_set_file_extent_generation(leaf, item, trans->transid);
63 btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
64 btrfs_set_file_extent_compression(leaf, item, compression);
65 btrfs_set_file_extent_encryption(leaf, item, encryption);
66 btrfs_set_file_extent_other_encoding(leaf, item, other_encoding);
68 btrfs_mark_buffer_dirty(leaf);
70 btrfs_free_path(path);
74 struct btrfs_csum_item *btrfs_lookup_csum(struct btrfs_trans_handle *trans,
75 struct btrfs_root *root,
76 struct btrfs_path *path,
80 struct btrfs_key file_key;
81 struct btrfs_key found_key;
82 struct btrfs_csum_item *item;
83 struct extent_buffer *leaf;
86 btrfs_super_csum_size(&root->fs_info->super_copy);
89 file_key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
90 file_key.offset = bytenr;
91 btrfs_set_key_type(&file_key, BTRFS_EXTENT_CSUM_KEY);
92 ret = btrfs_search_slot(trans, root, &file_key, path, 0, cow);
95 leaf = path->nodes[0];
98 if (path->slots[0] == 0)
101 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
102 if (btrfs_key_type(&found_key) != BTRFS_EXTENT_CSUM_KEY)
105 csum_offset = (bytenr - found_key.offset) >>
106 root->fs_info->sb->s_blocksize_bits;
107 csums_in_item = btrfs_item_size_nr(leaf, path->slots[0]);
108 csums_in_item /= csum_size;
110 if (csum_offset >= csums_in_item) {
115 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item);
116 item = (struct btrfs_csum_item *)((unsigned char *)item +
117 csum_offset * csum_size);
126 int btrfs_lookup_file_extent(struct btrfs_trans_handle *trans,
127 struct btrfs_root *root,
128 struct btrfs_path *path, u64 objectid,
132 struct btrfs_key file_key;
133 int ins_len = mod < 0 ? -1 : 0;
136 file_key.objectid = objectid;
137 file_key.offset = offset;
138 btrfs_set_key_type(&file_key, BTRFS_EXTENT_DATA_KEY);
139 ret = btrfs_search_slot(trans, root, &file_key, path, ins_len, cow);
144 int btrfs_lookup_bio_sums(struct btrfs_root *root, struct inode *inode,
145 struct bio *bio, u32 *dst)
148 struct bio_vec *bvec = bio->bi_io_vec;
151 u64 item_start_offset = 0;
152 u64 item_last_offset = 0;
156 btrfs_super_csum_size(&root->fs_info->super_copy);
158 struct btrfs_path *path;
159 struct btrfs_csum_item *item = NULL;
160 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
162 path = btrfs_alloc_path();
163 if (bio->bi_size > PAGE_CACHE_SIZE * 8)
166 WARN_ON(bio->bi_vcnt <= 0);
168 disk_bytenr = (u64)bio->bi_sector << 9;
169 while (bio_index < bio->bi_vcnt) {
170 offset = page_offset(bvec->bv_page) + bvec->bv_offset;
171 ret = btrfs_find_ordered_sum(inode, offset, disk_bytenr, &sum);
175 if (!item || disk_bytenr < item_start_offset ||
176 disk_bytenr >= item_last_offset) {
177 struct btrfs_key found_key;
181 btrfs_release_path(root, path);
182 item = btrfs_lookup_csum(NULL, root->fs_info->csum_root,
183 path, disk_bytenr, 0);
186 if (ret == -ENOENT || ret == -EFBIG)
189 if (BTRFS_I(inode)->root->root_key.objectid ==
190 BTRFS_DATA_RELOC_TREE_OBJECTID) {
191 set_extent_bits(io_tree, offset,
192 offset + bvec->bv_len - 1,
193 EXTENT_NODATASUM, GFP_NOFS);
195 printk(KERN_INFO "btrfs no csum found "
196 "for inode %lu start %llu\n",
198 (unsigned long long)offset);
201 btrfs_release_path(root, path);
204 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
207 item_start_offset = found_key.offset;
208 item_size = btrfs_item_size_nr(path->nodes[0],
210 item_last_offset = item_start_offset +
211 (item_size / csum_size) *
213 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
214 struct btrfs_csum_item);
217 * this byte range must be able to fit inside
218 * a single leaf so it will also fit inside a u32
220 diff = disk_bytenr - item_start_offset;
221 diff = diff / root->sectorsize;
222 diff = diff * csum_size;
224 read_extent_buffer(path->nodes[0], &sum,
225 ((unsigned long)item) + diff,
231 set_state_private(io_tree, offset, sum);
232 disk_bytenr += bvec->bv_len;
236 btrfs_free_path(path);
240 int btrfs_lookup_csums_range(struct btrfs_root *root, u64 start, u64 end,
241 struct list_head *list)
243 struct btrfs_key key;
244 struct btrfs_path *path;
245 struct extent_buffer *leaf;
246 struct btrfs_ordered_sum *sums;
247 struct btrfs_sector_sum *sector_sum;
248 struct btrfs_csum_item *item;
249 unsigned long offset;
253 u16 csum_size = btrfs_super_csum_size(&root->fs_info->super_copy);
255 path = btrfs_alloc_path();
258 key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
260 key.type = BTRFS_EXTENT_CSUM_KEY;
262 ret = btrfs_search_slot(NULL, root->fs_info->csum_root,
266 if (ret > 0 && path->slots[0] > 0) {
267 leaf = path->nodes[0];
268 btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
269 if (key.objectid == BTRFS_EXTENT_CSUM_OBJECTID &&
270 key.type == BTRFS_EXTENT_CSUM_KEY) {
271 offset = (start - key.offset) >>
272 root->fs_info->sb->s_blocksize_bits;
273 if (offset * csum_size <
274 btrfs_item_size_nr(leaf, path->slots[0] - 1))
279 while (start <= end) {
280 leaf = path->nodes[0];
281 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
282 ret = btrfs_next_leaf(root->fs_info->csum_root, path);
287 leaf = path->nodes[0];
290 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
291 if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
292 key.type != BTRFS_EXTENT_CSUM_KEY)
295 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
296 if (key.offset > end)
299 if (key.offset > start)
302 size = btrfs_item_size_nr(leaf, path->slots[0]);
303 csum_end = key.offset + (size / csum_size) * root->sectorsize;
304 if (csum_end <= start) {
309 size = min(csum_end, end + 1) - start;
310 sums = kzalloc(btrfs_ordered_sum_size(root, size), GFP_NOFS);
313 sector_sum = sums->sums;
314 sums->bytenr = start;
317 offset = (start - key.offset) >>
318 root->fs_info->sb->s_blocksize_bits;
321 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
322 struct btrfs_csum_item);
324 read_extent_buffer(path->nodes[0], §or_sum->sum,
325 ((unsigned long)item) + offset,
327 sector_sum->bytenr = start;
329 size -= root->sectorsize;
330 start += root->sectorsize;
334 list_add_tail(&sums->list, list);
340 btrfs_free_path(path);
344 int btrfs_csum_one_bio(struct btrfs_root *root, struct inode *inode,
345 struct bio *bio, u64 file_start, int contig)
347 struct btrfs_ordered_sum *sums;
348 struct btrfs_sector_sum *sector_sum;
349 struct btrfs_ordered_extent *ordered;
351 struct bio_vec *bvec = bio->bi_io_vec;
353 unsigned long total_bytes = 0;
354 unsigned long this_sum_bytes = 0;
358 WARN_ON(bio->bi_vcnt <= 0);
359 sums = kzalloc(btrfs_ordered_sum_size(root, bio->bi_size), GFP_NOFS);
363 sector_sum = sums->sums;
364 disk_bytenr = (u64)bio->bi_sector << 9;
365 sums->len = bio->bi_size;
366 INIT_LIST_HEAD(&sums->list);
371 offset = page_offset(bvec->bv_page) + bvec->bv_offset;
373 ordered = btrfs_lookup_ordered_extent(inode, offset);
375 sums->bytenr = ordered->start;
377 while (bio_index < bio->bi_vcnt) {
379 offset = page_offset(bvec->bv_page) + bvec->bv_offset;
381 if (!contig && (offset >= ordered->file_offset + ordered->len ||
382 offset < ordered->file_offset)) {
383 unsigned long bytes_left;
384 sums->len = this_sum_bytes;
386 btrfs_add_ordered_sum(inode, ordered, sums);
387 btrfs_put_ordered_extent(ordered);
389 bytes_left = bio->bi_size - total_bytes;
391 sums = kzalloc(btrfs_ordered_sum_size(root, bytes_left),
394 sector_sum = sums->sums;
395 sums->len = bytes_left;
396 ordered = btrfs_lookup_ordered_extent(inode, offset);
398 sums->bytenr = ordered->start;
401 data = kmap_atomic(bvec->bv_page, KM_USER0);
402 sector_sum->sum = ~(u32)0;
403 sector_sum->sum = btrfs_csum_data(root,
404 data + bvec->bv_offset,
407 kunmap_atomic(data, KM_USER0);
408 btrfs_csum_final(sector_sum->sum,
409 (char *)§or_sum->sum);
410 sector_sum->bytenr = disk_bytenr;
414 total_bytes += bvec->bv_len;
415 this_sum_bytes += bvec->bv_len;
416 disk_bytenr += bvec->bv_len;
417 offset += bvec->bv_len;
421 btrfs_add_ordered_sum(inode, ordered, sums);
422 btrfs_put_ordered_extent(ordered);
427 * helper function for csum removal, this expects the
428 * key to describe the csum pointed to by the path, and it expects
429 * the csum to overlap the range [bytenr, len]
431 * The csum should not be entirely contained in the range and the
432 * range should not be entirely contained in the csum.
434 * This calls btrfs_truncate_item with the correct args based on the
435 * overlap, and fixes up the key as required.
437 static noinline int truncate_one_csum(struct btrfs_trans_handle *trans,
438 struct btrfs_root *root,
439 struct btrfs_path *path,
440 struct btrfs_key *key,
443 struct extent_buffer *leaf;
445 btrfs_super_csum_size(&root->fs_info->super_copy);
447 u64 end_byte = bytenr + len;
448 u32 blocksize_bits = root->fs_info->sb->s_blocksize_bits;
451 leaf = path->nodes[0];
452 csum_end = btrfs_item_size_nr(leaf, path->slots[0]) / csum_size;
453 csum_end <<= root->fs_info->sb->s_blocksize_bits;
454 csum_end += key->offset;
456 if (key->offset < bytenr && csum_end <= end_byte) {
461 * A simple truncate off the end of the item
463 u32 new_size = (bytenr - key->offset) >> blocksize_bits;
464 new_size *= csum_size;
465 ret = btrfs_truncate_item(trans, root, path, new_size, 1);
467 } else if (key->offset >= bytenr && csum_end > end_byte &&
468 end_byte > key->offset) {
473 * we need to truncate from the beginning of the csum
475 u32 new_size = (csum_end - end_byte) >> blocksize_bits;
476 new_size *= csum_size;
478 ret = btrfs_truncate_item(trans, root, path, new_size, 0);
481 key->offset = end_byte;
482 ret = btrfs_set_item_key_safe(trans, root, path, key);
491 * deletes the csum items from the csum tree for a given
494 int btrfs_del_csums(struct btrfs_trans_handle *trans,
495 struct btrfs_root *root, u64 bytenr, u64 len)
497 struct btrfs_path *path;
498 struct btrfs_key key;
499 u64 end_byte = bytenr + len;
501 struct extent_buffer *leaf;
504 btrfs_super_csum_size(&root->fs_info->super_copy);
505 int blocksize_bits = root->fs_info->sb->s_blocksize_bits;
507 root = root->fs_info->csum_root;
509 path = btrfs_alloc_path();
512 key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
513 key.offset = end_byte - 1;
514 key.type = BTRFS_EXTENT_CSUM_KEY;
516 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
518 if (path->slots[0] == 0)
522 leaf = path->nodes[0];
523 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
525 if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
526 key.type != BTRFS_EXTENT_CSUM_KEY) {
530 if (key.offset >= end_byte)
533 csum_end = btrfs_item_size_nr(leaf, path->slots[0]) / csum_size;
534 csum_end <<= blocksize_bits;
535 csum_end += key.offset;
537 /* this csum ends before we start, we're done */
538 if (csum_end <= bytenr)
541 /* delete the entire item, it is inside our range */
542 if (key.offset >= bytenr && csum_end <= end_byte) {
543 ret = btrfs_del_item(trans, root, path);
545 if (key.offset == bytenr)
547 } else if (key.offset < bytenr && csum_end > end_byte) {
548 unsigned long offset;
549 unsigned long shift_len;
550 unsigned long item_offset;
555 * Our bytes are in the middle of the csum,
556 * we need to split this item and insert a new one.
558 * But we can't drop the path because the
559 * csum could change, get removed, extended etc.
561 * The trick here is the max size of a csum item leaves
562 * enough room in the tree block for a single
563 * item header. So, we split the item in place,
564 * adding a new header pointing to the existing
565 * bytes. Then we loop around again and we have
566 * a nicely formed csum item that we can neatly
569 offset = (bytenr - key.offset) >> blocksize_bits;
572 shift_len = (len >> blocksize_bits) * csum_size;
574 item_offset = btrfs_item_ptr_offset(leaf,
577 memset_extent_buffer(leaf, 0, item_offset + offset,
582 * btrfs_split_item returns -EAGAIN when the
583 * item changed size or key
585 ret = btrfs_split_item(trans, root, path, &key, offset);
586 BUG_ON(ret && ret != -EAGAIN);
588 key.offset = end_byte - 1;
590 ret = truncate_one_csum(trans, root, path,
593 if (key.offset < bytenr)
596 btrfs_release_path(root, path);
599 btrfs_free_path(path);
603 int btrfs_csum_file_blocks(struct btrfs_trans_handle *trans,
604 struct btrfs_root *root,
605 struct btrfs_ordered_sum *sums)
609 struct btrfs_key file_key;
610 struct btrfs_key found_key;
614 struct btrfs_path *path;
615 struct btrfs_csum_item *item;
616 struct btrfs_csum_item *item_end;
617 struct extent_buffer *leaf = NULL;
619 struct btrfs_sector_sum *sector_sum;
624 unsigned long map_len;
625 unsigned long map_start;
627 btrfs_super_csum_size(&root->fs_info->super_copy);
629 path = btrfs_alloc_path();
631 sector_sum = sums->sums;
633 next_offset = (u64)-1;
635 file_key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
636 file_key.offset = sector_sum->bytenr;
637 bytenr = sector_sum->bytenr;
638 btrfs_set_key_type(&file_key, BTRFS_EXTENT_CSUM_KEY);
640 item = btrfs_lookup_csum(trans, root, path, sector_sum->bytenr, 1);
642 leaf = path->nodes[0];
649 /* we found one, but it isn't big enough yet */
650 leaf = path->nodes[0];
651 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
652 if ((item_size / csum_size) >=
653 MAX_CSUM_ITEMS(root, csum_size)) {
654 /* already at max size, make a new one */
658 int slot = path->slots[0] + 1;
659 /* we didn't find a csum item, insert one */
660 nritems = btrfs_header_nritems(path->nodes[0]);
661 if (path->slots[0] >= nritems - 1) {
662 ret = btrfs_next_leaf(root, path);
669 btrfs_item_key_to_cpu(path->nodes[0], &found_key, slot);
670 if (found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
671 found_key.type != BTRFS_EXTENT_CSUM_KEY) {
675 next_offset = found_key.offset;
681 * at this point, we know the tree has an item, but it isn't big
682 * enough yet to put our csum in. Grow it
684 btrfs_release_path(root, path);
685 ret = btrfs_search_slot(trans, root, &file_key, path,
691 if (path->slots[0] == 0)
696 leaf = path->nodes[0];
697 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
698 csum_offset = (bytenr - found_key.offset) >>
699 root->fs_info->sb->s_blocksize_bits;
701 if (btrfs_key_type(&found_key) != BTRFS_EXTENT_CSUM_KEY ||
702 found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
703 csum_offset >= MAX_CSUM_ITEMS(root, csum_size)) {
707 if (csum_offset >= btrfs_item_size_nr(leaf, path->slots[0]) /
709 u32 diff = (csum_offset + 1) * csum_size;
712 * is the item big enough already? we dropped our lock
713 * before and need to recheck
715 if (diff < btrfs_item_size_nr(leaf, path->slots[0]))
718 diff = diff - btrfs_item_size_nr(leaf, path->slots[0]);
719 if (diff != csum_size)
722 ret = btrfs_extend_item(trans, root, path, diff);
728 btrfs_release_path(root, path);
731 u64 tmp = total_bytes + root->sectorsize;
732 u64 next_sector = sector_sum->bytenr;
733 struct btrfs_sector_sum *next = sector_sum + 1;
735 while (tmp < sums->len) {
736 if (next_sector + root->sectorsize != next->bytenr)
738 tmp += root->sectorsize;
739 next_sector = next->bytenr;
742 tmp = min(tmp, next_offset - file_key.offset);
743 tmp >>= root->fs_info->sb->s_blocksize_bits;
744 tmp = max((u64)1, tmp);
745 tmp = min(tmp, (u64)MAX_CSUM_ITEMS(root, csum_size));
746 ins_size = csum_size * tmp;
748 ins_size = csum_size;
750 ret = btrfs_insert_empty_item(trans, root, path, &file_key,
759 leaf = path->nodes[0];
760 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item);
762 item = (struct btrfs_csum_item *)((unsigned char *)item +
763 csum_offset * csum_size);
765 item_end = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item);
766 item_end = (struct btrfs_csum_item *)((unsigned char *)item_end +
767 btrfs_item_size_nr(leaf, path->slots[0]));
773 (unsigned long)item + csum_size >= map_start + map_len) {
777 unmap_extent_buffer(leaf, eb_token, KM_USER1);
779 err = map_private_extent_buffer(leaf, (unsigned long)item,
782 &map_start, &map_len, KM_USER1);
787 memcpy(eb_token + ((unsigned long)item & (PAGE_CACHE_SIZE - 1)),
788 §or_sum->sum, csum_size);
790 write_extent_buffer(leaf, §or_sum->sum,
791 (unsigned long)item, csum_size);
794 total_bytes += root->sectorsize;
796 if (total_bytes < sums->len) {
797 item = (struct btrfs_csum_item *)((char *)item +
799 if (item < item_end && bytenr + PAGE_CACHE_SIZE ==
800 sector_sum->bytenr) {
801 bytenr = sector_sum->bytenr;
806 unmap_extent_buffer(leaf, eb_token, KM_USER1);
809 btrfs_mark_buffer_dirty(path->nodes[0]);
811 if (total_bytes < sums->len) {
812 btrfs_release_path(root, path);
816 btrfs_free_path(path);