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);
181 * the free space stuff is only read when it hasn't been
182 * updated in the current transaction. So, we can safely
183 * read from the commit root and sidestep a nasty deadlock
184 * between reading the free space cache and updating the csum tree.
186 if (btrfs_is_free_space_inode(root, inode))
187 path->search_commit_root = 1;
189 disk_bytenr = (u64)bio->bi_sector << 9;
191 offset = logical_offset;
192 while (bio_index < bio->bi_vcnt) {
194 offset = page_offset(bvec->bv_page) + bvec->bv_offset;
195 ret = btrfs_find_ordered_sum(inode, offset, disk_bytenr, &sum);
199 if (!item || disk_bytenr < item_start_offset ||
200 disk_bytenr >= item_last_offset) {
201 struct btrfs_key found_key;
205 btrfs_release_path(path);
206 item = btrfs_lookup_csum(NULL, root->fs_info->csum_root,
207 path, disk_bytenr, 0);
210 if (ret == -ENOENT || ret == -EFBIG)
213 if (BTRFS_I(inode)->root->root_key.objectid ==
214 BTRFS_DATA_RELOC_TREE_OBJECTID) {
215 set_extent_bits(io_tree, offset,
216 offset + bvec->bv_len - 1,
217 EXTENT_NODATASUM, GFP_NOFS);
219 printk(KERN_INFO "btrfs no csum found "
220 "for inode %llu start %llu\n",
223 (unsigned long long)offset);
226 btrfs_release_path(path);
229 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
232 item_start_offset = found_key.offset;
233 item_size = btrfs_item_size_nr(path->nodes[0],
235 item_last_offset = item_start_offset +
236 (item_size / csum_size) *
238 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
239 struct btrfs_csum_item);
242 * this byte range must be able to fit inside
243 * a single leaf so it will also fit inside a u32
245 diff = disk_bytenr - item_start_offset;
246 diff = diff / root->sectorsize;
247 diff = diff * csum_size;
249 read_extent_buffer(path->nodes[0], &sum,
250 ((unsigned long)item) + diff,
256 set_state_private(io_tree, offset, sum);
257 disk_bytenr += bvec->bv_len;
258 offset += bvec->bv_len;
262 btrfs_free_path(path);
266 int btrfs_lookup_bio_sums(struct btrfs_root *root, struct inode *inode,
267 struct bio *bio, u32 *dst)
269 return __btrfs_lookup_bio_sums(root, inode, bio, 0, dst, 0);
272 int btrfs_lookup_bio_sums_dio(struct btrfs_root *root, struct inode *inode,
273 struct bio *bio, u64 offset, u32 *dst)
275 return __btrfs_lookup_bio_sums(root, inode, bio, offset, dst, 1);
278 int btrfs_lookup_csums_range(struct btrfs_root *root, u64 start, u64 end,
279 struct list_head *list, int search_commit)
281 struct btrfs_key key;
282 struct btrfs_path *path;
283 struct extent_buffer *leaf;
284 struct btrfs_ordered_sum *sums;
285 struct btrfs_sector_sum *sector_sum;
286 struct btrfs_csum_item *item;
287 unsigned long offset;
291 u16 csum_size = btrfs_super_csum_size(&root->fs_info->super_copy);
293 path = btrfs_alloc_path();
298 path->skip_locking = 1;
300 path->search_commit_root = 1;
303 key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
305 key.type = BTRFS_EXTENT_CSUM_KEY;
307 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
310 if (ret > 0 && path->slots[0] > 0) {
311 leaf = path->nodes[0];
312 btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
313 if (key.objectid == BTRFS_EXTENT_CSUM_OBJECTID &&
314 key.type == BTRFS_EXTENT_CSUM_KEY) {
315 offset = (start - key.offset) >>
316 root->fs_info->sb->s_blocksize_bits;
317 if (offset * csum_size <
318 btrfs_item_size_nr(leaf, path->slots[0] - 1))
323 while (start <= end) {
324 leaf = path->nodes[0];
325 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
326 ret = btrfs_next_leaf(root, path);
331 leaf = path->nodes[0];
334 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
335 if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
336 key.type != BTRFS_EXTENT_CSUM_KEY)
339 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
340 if (key.offset > end)
343 if (key.offset > start)
346 size = btrfs_item_size_nr(leaf, path->slots[0]);
347 csum_end = key.offset + (size / csum_size) * root->sectorsize;
348 if (csum_end <= start) {
353 csum_end = min(csum_end, end + 1);
354 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
355 struct btrfs_csum_item);
356 while (start < csum_end) {
357 size = min_t(size_t, csum_end - start,
358 MAX_ORDERED_SUM_BYTES(root));
359 sums = kzalloc(btrfs_ordered_sum_size(root, size),
363 sector_sum = sums->sums;
364 sums->bytenr = start;
367 offset = (start - key.offset) >>
368 root->fs_info->sb->s_blocksize_bits;
372 read_extent_buffer(path->nodes[0],
374 ((unsigned long)item) +
376 sector_sum->bytenr = start;
378 size -= root->sectorsize;
379 start += root->sectorsize;
383 list_add_tail(&sums->list, list);
389 btrfs_free_path(path);
393 int btrfs_csum_one_bio(struct btrfs_root *root, struct inode *inode,
394 struct bio *bio, u64 file_start, int contig)
396 struct btrfs_ordered_sum *sums;
397 struct btrfs_sector_sum *sector_sum;
398 struct btrfs_ordered_extent *ordered;
400 struct bio_vec *bvec = bio->bi_io_vec;
402 unsigned long total_bytes = 0;
403 unsigned long this_sum_bytes = 0;
407 WARN_ON(bio->bi_vcnt <= 0);
408 sums = kzalloc(btrfs_ordered_sum_size(root, bio->bi_size), GFP_NOFS);
412 sector_sum = sums->sums;
413 disk_bytenr = (u64)bio->bi_sector << 9;
414 sums->len = bio->bi_size;
415 INIT_LIST_HEAD(&sums->list);
420 offset = page_offset(bvec->bv_page) + bvec->bv_offset;
422 ordered = btrfs_lookup_ordered_extent(inode, offset);
424 sums->bytenr = ordered->start;
426 while (bio_index < bio->bi_vcnt) {
428 offset = page_offset(bvec->bv_page) + bvec->bv_offset;
430 if (!contig && (offset >= ordered->file_offset + ordered->len ||
431 offset < ordered->file_offset)) {
432 unsigned long bytes_left;
433 sums->len = this_sum_bytes;
435 btrfs_add_ordered_sum(inode, ordered, sums);
436 btrfs_put_ordered_extent(ordered);
438 bytes_left = bio->bi_size - total_bytes;
440 sums = kzalloc(btrfs_ordered_sum_size(root, bytes_left),
443 sector_sum = sums->sums;
444 sums->len = bytes_left;
445 ordered = btrfs_lookup_ordered_extent(inode, offset);
447 sums->bytenr = ordered->start;
450 data = kmap_atomic(bvec->bv_page, KM_USER0);
451 sector_sum->sum = ~(u32)0;
452 sector_sum->sum = btrfs_csum_data(root,
453 data + bvec->bv_offset,
456 kunmap_atomic(data, KM_USER0);
457 btrfs_csum_final(sector_sum->sum,
458 (char *)§or_sum->sum);
459 sector_sum->bytenr = disk_bytenr;
463 total_bytes += bvec->bv_len;
464 this_sum_bytes += bvec->bv_len;
465 disk_bytenr += bvec->bv_len;
466 offset += bvec->bv_len;
470 btrfs_add_ordered_sum(inode, ordered, sums);
471 btrfs_put_ordered_extent(ordered);
476 * helper function for csum removal, this expects the
477 * key to describe the csum pointed to by the path, and it expects
478 * the csum to overlap the range [bytenr, len]
480 * The csum should not be entirely contained in the range and the
481 * range should not be entirely contained in the csum.
483 * This calls btrfs_truncate_item with the correct args based on the
484 * overlap, and fixes up the key as required.
486 static noinline int truncate_one_csum(struct btrfs_trans_handle *trans,
487 struct btrfs_root *root,
488 struct btrfs_path *path,
489 struct btrfs_key *key,
492 struct extent_buffer *leaf;
494 btrfs_super_csum_size(&root->fs_info->super_copy);
496 u64 end_byte = bytenr + len;
497 u32 blocksize_bits = root->fs_info->sb->s_blocksize_bits;
500 leaf = path->nodes[0];
501 csum_end = btrfs_item_size_nr(leaf, path->slots[0]) / csum_size;
502 csum_end <<= root->fs_info->sb->s_blocksize_bits;
503 csum_end += key->offset;
505 if (key->offset < bytenr && csum_end <= end_byte) {
510 * A simple truncate off the end of the item
512 u32 new_size = (bytenr - key->offset) >> blocksize_bits;
513 new_size *= csum_size;
514 ret = btrfs_truncate_item(trans, root, path, new_size, 1);
515 } else if (key->offset >= bytenr && csum_end > end_byte &&
516 end_byte > key->offset) {
521 * we need to truncate from the beginning of the csum
523 u32 new_size = (csum_end - end_byte) >> blocksize_bits;
524 new_size *= csum_size;
526 ret = btrfs_truncate_item(trans, root, path, new_size, 0);
528 key->offset = end_byte;
529 ret = btrfs_set_item_key_safe(trans, root, path, key);
538 * deletes the csum items from the csum tree for a given
541 int btrfs_del_csums(struct btrfs_trans_handle *trans,
542 struct btrfs_root *root, u64 bytenr, u64 len)
544 struct btrfs_path *path;
545 struct btrfs_key key;
546 u64 end_byte = bytenr + len;
548 struct extent_buffer *leaf;
551 btrfs_super_csum_size(&root->fs_info->super_copy);
552 int blocksize_bits = root->fs_info->sb->s_blocksize_bits;
554 root = root->fs_info->csum_root;
556 path = btrfs_alloc_path();
561 key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
562 key.offset = end_byte - 1;
563 key.type = BTRFS_EXTENT_CSUM_KEY;
565 path->leave_spinning = 1;
566 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
568 if (path->slots[0] == 0)
571 } else if (ret < 0) {
575 leaf = path->nodes[0];
576 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
578 if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
579 key.type != BTRFS_EXTENT_CSUM_KEY) {
583 if (key.offset >= end_byte)
586 csum_end = btrfs_item_size_nr(leaf, path->slots[0]) / csum_size;
587 csum_end <<= blocksize_bits;
588 csum_end += key.offset;
590 /* this csum ends before we start, we're done */
591 if (csum_end <= bytenr)
594 /* delete the entire item, it is inside our range */
595 if (key.offset >= bytenr && csum_end <= end_byte) {
596 ret = btrfs_del_item(trans, root, path);
599 if (key.offset == bytenr)
601 } else if (key.offset < bytenr && csum_end > end_byte) {
602 unsigned long offset;
603 unsigned long shift_len;
604 unsigned long item_offset;
609 * Our bytes are in the middle of the csum,
610 * we need to split this item and insert a new one.
612 * But we can't drop the path because the
613 * csum could change, get removed, extended etc.
615 * The trick here is the max size of a csum item leaves
616 * enough room in the tree block for a single
617 * item header. So, we split the item in place,
618 * adding a new header pointing to the existing
619 * bytes. Then we loop around again and we have
620 * a nicely formed csum item that we can neatly
623 offset = (bytenr - key.offset) >> blocksize_bits;
626 shift_len = (len >> blocksize_bits) * csum_size;
628 item_offset = btrfs_item_ptr_offset(leaf,
631 memset_extent_buffer(leaf, 0, item_offset + offset,
636 * btrfs_split_item returns -EAGAIN when the
637 * item changed size or key
639 ret = btrfs_split_item(trans, root, path, &key, offset);
640 BUG_ON(ret && ret != -EAGAIN);
642 key.offset = end_byte - 1;
644 ret = truncate_one_csum(trans, root, path,
647 if (key.offset < bytenr)
650 btrfs_release_path(path);
654 btrfs_free_path(path);
658 int btrfs_csum_file_blocks(struct btrfs_trans_handle *trans,
659 struct btrfs_root *root,
660 struct btrfs_ordered_sum *sums)
664 struct btrfs_key file_key;
665 struct btrfs_key found_key;
669 struct btrfs_path *path;
670 struct btrfs_csum_item *item;
671 struct btrfs_csum_item *item_end;
672 struct extent_buffer *leaf = NULL;
674 struct btrfs_sector_sum *sector_sum;
678 btrfs_super_csum_size(&root->fs_info->super_copy);
680 path = btrfs_alloc_path();
684 sector_sum = sums->sums;
686 next_offset = (u64)-1;
688 file_key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
689 file_key.offset = sector_sum->bytenr;
690 bytenr = sector_sum->bytenr;
691 btrfs_set_key_type(&file_key, BTRFS_EXTENT_CSUM_KEY);
693 item = btrfs_lookup_csum(trans, root, path, sector_sum->bytenr, 1);
695 leaf = path->nodes[0];
700 if (ret != -EFBIG && ret != -ENOENT)
705 /* we found one, but it isn't big enough yet */
706 leaf = path->nodes[0];
707 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
708 if ((item_size / csum_size) >=
709 MAX_CSUM_ITEMS(root, csum_size)) {
710 /* already at max size, make a new one */
714 int slot = path->slots[0] + 1;
715 /* we didn't find a csum item, insert one */
716 nritems = btrfs_header_nritems(path->nodes[0]);
717 if (path->slots[0] >= nritems - 1) {
718 ret = btrfs_next_leaf(root, path);
725 btrfs_item_key_to_cpu(path->nodes[0], &found_key, slot);
726 if (found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
727 found_key.type != BTRFS_EXTENT_CSUM_KEY) {
731 next_offset = found_key.offset;
737 * at this point, we know the tree has an item, but it isn't big
738 * enough yet to put our csum in. Grow it
740 btrfs_release_path(path);
741 ret = btrfs_search_slot(trans, root, &file_key, path,
747 if (path->slots[0] == 0)
752 leaf = path->nodes[0];
753 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
754 csum_offset = (bytenr - found_key.offset) >>
755 root->fs_info->sb->s_blocksize_bits;
757 if (btrfs_key_type(&found_key) != BTRFS_EXTENT_CSUM_KEY ||
758 found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
759 csum_offset >= MAX_CSUM_ITEMS(root, csum_size)) {
763 if (csum_offset >= btrfs_item_size_nr(leaf, path->slots[0]) /
765 u32 diff = (csum_offset + 1) * csum_size;
768 * is the item big enough already? we dropped our lock
769 * before and need to recheck
771 if (diff < btrfs_item_size_nr(leaf, path->slots[0]))
774 diff = diff - btrfs_item_size_nr(leaf, path->slots[0]);
775 if (diff != csum_size)
778 ret = btrfs_extend_item(trans, root, path, diff);
783 btrfs_release_path(path);
786 u64 tmp = total_bytes + root->sectorsize;
787 u64 next_sector = sector_sum->bytenr;
788 struct btrfs_sector_sum *next = sector_sum + 1;
790 while (tmp < sums->len) {
791 if (next_sector + root->sectorsize != next->bytenr)
793 tmp += root->sectorsize;
794 next_sector = next->bytenr;
797 tmp = min(tmp, next_offset - file_key.offset);
798 tmp >>= root->fs_info->sb->s_blocksize_bits;
799 tmp = max((u64)1, tmp);
800 tmp = min(tmp, (u64)MAX_CSUM_ITEMS(root, csum_size));
801 ins_size = csum_size * tmp;
803 ins_size = csum_size;
805 path->leave_spinning = 1;
806 ret = btrfs_insert_empty_item(trans, root, path, &file_key,
808 path->leave_spinning = 0;
816 leaf = path->nodes[0];
817 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item);
819 item = (struct btrfs_csum_item *)((unsigned char *)item +
820 csum_offset * csum_size);
822 item_end = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item);
823 item_end = (struct btrfs_csum_item *)((unsigned char *)item_end +
824 btrfs_item_size_nr(leaf, path->slots[0]));
827 write_extent_buffer(leaf, §or_sum->sum, (unsigned long)item, csum_size);
829 total_bytes += root->sectorsize;
831 if (total_bytes < sums->len) {
832 item = (struct btrfs_csum_item *)((char *)item +
834 if (item < item_end && bytenr + PAGE_CACHE_SIZE ==
835 sector_sum->bytenr) {
836 bytenr = sector_sum->bytenr;
841 btrfs_mark_buffer_dirty(path->nodes[0]);
842 if (total_bytes < sums->len) {
843 btrfs_release_path(path);
848 btrfs_free_path(path);