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;
188 path->skip_locking = 1;
191 disk_bytenr = (u64)bio->bi_sector << 9;
193 offset = logical_offset;
194 while (bio_index < bio->bi_vcnt) {
196 offset = page_offset(bvec->bv_page) + bvec->bv_offset;
197 ret = btrfs_find_ordered_sum(inode, offset, disk_bytenr, &sum);
201 if (!item || disk_bytenr < item_start_offset ||
202 disk_bytenr >= item_last_offset) {
203 struct btrfs_key found_key;
207 btrfs_release_path(path);
208 item = btrfs_lookup_csum(NULL, root->fs_info->csum_root,
209 path, disk_bytenr, 0);
212 if (ret == -ENOENT || ret == -EFBIG)
215 if (BTRFS_I(inode)->root->root_key.objectid ==
216 BTRFS_DATA_RELOC_TREE_OBJECTID) {
217 set_extent_bits(io_tree, offset,
218 offset + bvec->bv_len - 1,
219 EXTENT_NODATASUM, GFP_NOFS);
221 printk(KERN_INFO "btrfs no csum found "
222 "for inode %llu start %llu\n",
225 (unsigned long long)offset);
228 btrfs_release_path(path);
231 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
234 item_start_offset = found_key.offset;
235 item_size = btrfs_item_size_nr(path->nodes[0],
237 item_last_offset = item_start_offset +
238 (item_size / csum_size) *
240 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
241 struct btrfs_csum_item);
244 * this byte range must be able to fit inside
245 * a single leaf so it will also fit inside a u32
247 diff = disk_bytenr - item_start_offset;
248 diff = diff / root->sectorsize;
249 diff = diff * csum_size;
251 read_extent_buffer(path->nodes[0], &sum,
252 ((unsigned long)item) + diff,
258 set_state_private(io_tree, offset, sum);
259 disk_bytenr += bvec->bv_len;
260 offset += bvec->bv_len;
264 btrfs_free_path(path);
268 int btrfs_lookup_bio_sums(struct btrfs_root *root, struct inode *inode,
269 struct bio *bio, u32 *dst)
271 return __btrfs_lookup_bio_sums(root, inode, bio, 0, dst, 0);
274 int btrfs_lookup_bio_sums_dio(struct btrfs_root *root, struct inode *inode,
275 struct bio *bio, u64 offset, u32 *dst)
277 return __btrfs_lookup_bio_sums(root, inode, bio, offset, dst, 1);
280 int btrfs_lookup_csums_range(struct btrfs_root *root, u64 start, u64 end,
281 struct list_head *list, int search_commit)
283 struct btrfs_key key;
284 struct btrfs_path *path;
285 struct extent_buffer *leaf;
286 struct btrfs_ordered_sum *sums;
287 struct btrfs_sector_sum *sector_sum;
288 struct btrfs_csum_item *item;
289 unsigned long offset;
293 u16 csum_size = btrfs_super_csum_size(&root->fs_info->super_copy);
295 path = btrfs_alloc_path();
300 path->skip_locking = 1;
302 path->search_commit_root = 1;
305 key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
307 key.type = BTRFS_EXTENT_CSUM_KEY;
309 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
312 if (ret > 0 && path->slots[0] > 0) {
313 leaf = path->nodes[0];
314 btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
315 if (key.objectid == BTRFS_EXTENT_CSUM_OBJECTID &&
316 key.type == BTRFS_EXTENT_CSUM_KEY) {
317 offset = (start - key.offset) >>
318 root->fs_info->sb->s_blocksize_bits;
319 if (offset * csum_size <
320 btrfs_item_size_nr(leaf, path->slots[0] - 1))
325 while (start <= end) {
326 leaf = path->nodes[0];
327 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
328 ret = btrfs_next_leaf(root, path);
333 leaf = path->nodes[0];
336 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
337 if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
338 key.type != BTRFS_EXTENT_CSUM_KEY)
341 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
342 if (key.offset > end)
345 if (key.offset > start)
348 size = btrfs_item_size_nr(leaf, path->slots[0]);
349 csum_end = key.offset + (size / csum_size) * root->sectorsize;
350 if (csum_end <= start) {
355 csum_end = min(csum_end, end + 1);
356 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
357 struct btrfs_csum_item);
358 while (start < csum_end) {
359 size = min_t(size_t, csum_end - start,
360 MAX_ORDERED_SUM_BYTES(root));
361 sums = kzalloc(btrfs_ordered_sum_size(root, size),
365 sector_sum = sums->sums;
366 sums->bytenr = start;
369 offset = (start - key.offset) >>
370 root->fs_info->sb->s_blocksize_bits;
374 read_extent_buffer(path->nodes[0],
376 ((unsigned long)item) +
378 sector_sum->bytenr = start;
380 size -= root->sectorsize;
381 start += root->sectorsize;
385 list_add_tail(&sums->list, list);
391 btrfs_free_path(path);
395 int btrfs_csum_one_bio(struct btrfs_root *root, struct inode *inode,
396 struct bio *bio, u64 file_start, int contig)
398 struct btrfs_ordered_sum *sums;
399 struct btrfs_sector_sum *sector_sum;
400 struct btrfs_ordered_extent *ordered;
402 struct bio_vec *bvec = bio->bi_io_vec;
404 unsigned long total_bytes = 0;
405 unsigned long this_sum_bytes = 0;
409 WARN_ON(bio->bi_vcnt <= 0);
410 sums = kzalloc(btrfs_ordered_sum_size(root, bio->bi_size), GFP_NOFS);
414 sector_sum = sums->sums;
415 disk_bytenr = (u64)bio->bi_sector << 9;
416 sums->len = bio->bi_size;
417 INIT_LIST_HEAD(&sums->list);
422 offset = page_offset(bvec->bv_page) + bvec->bv_offset;
424 ordered = btrfs_lookup_ordered_extent(inode, offset);
426 sums->bytenr = ordered->start;
428 while (bio_index < bio->bi_vcnt) {
430 offset = page_offset(bvec->bv_page) + bvec->bv_offset;
432 if (!contig && (offset >= ordered->file_offset + ordered->len ||
433 offset < ordered->file_offset)) {
434 unsigned long bytes_left;
435 sums->len = this_sum_bytes;
437 btrfs_add_ordered_sum(inode, ordered, sums);
438 btrfs_put_ordered_extent(ordered);
440 bytes_left = bio->bi_size - total_bytes;
442 sums = kzalloc(btrfs_ordered_sum_size(root, bytes_left),
445 sector_sum = sums->sums;
446 sums->len = bytes_left;
447 ordered = btrfs_lookup_ordered_extent(inode, offset);
449 sums->bytenr = ordered->start;
452 data = kmap_atomic(bvec->bv_page, KM_USER0);
453 sector_sum->sum = ~(u32)0;
454 sector_sum->sum = btrfs_csum_data(root,
455 data + bvec->bv_offset,
458 kunmap_atomic(data, KM_USER0);
459 btrfs_csum_final(sector_sum->sum,
460 (char *)§or_sum->sum);
461 sector_sum->bytenr = disk_bytenr;
465 total_bytes += bvec->bv_len;
466 this_sum_bytes += bvec->bv_len;
467 disk_bytenr += bvec->bv_len;
468 offset += bvec->bv_len;
472 btrfs_add_ordered_sum(inode, ordered, sums);
473 btrfs_put_ordered_extent(ordered);
478 * helper function for csum removal, this expects the
479 * key to describe the csum pointed to by the path, and it expects
480 * the csum to overlap the range [bytenr, len]
482 * The csum should not be entirely contained in the range and the
483 * range should not be entirely contained in the csum.
485 * This calls btrfs_truncate_item with the correct args based on the
486 * overlap, and fixes up the key as required.
488 static noinline int truncate_one_csum(struct btrfs_trans_handle *trans,
489 struct btrfs_root *root,
490 struct btrfs_path *path,
491 struct btrfs_key *key,
494 struct extent_buffer *leaf;
496 btrfs_super_csum_size(&root->fs_info->super_copy);
498 u64 end_byte = bytenr + len;
499 u32 blocksize_bits = root->fs_info->sb->s_blocksize_bits;
502 leaf = path->nodes[0];
503 csum_end = btrfs_item_size_nr(leaf, path->slots[0]) / csum_size;
504 csum_end <<= root->fs_info->sb->s_blocksize_bits;
505 csum_end += key->offset;
507 if (key->offset < bytenr && csum_end <= end_byte) {
512 * A simple truncate off the end of the item
514 u32 new_size = (bytenr - key->offset) >> blocksize_bits;
515 new_size *= csum_size;
516 ret = btrfs_truncate_item(trans, root, path, new_size, 1);
517 } else if (key->offset >= bytenr && csum_end > end_byte &&
518 end_byte > key->offset) {
523 * we need to truncate from the beginning of the csum
525 u32 new_size = (csum_end - end_byte) >> blocksize_bits;
526 new_size *= csum_size;
528 ret = btrfs_truncate_item(trans, root, path, new_size, 0);
530 key->offset = end_byte;
531 ret = btrfs_set_item_key_safe(trans, root, path, key);
540 * deletes the csum items from the csum tree for a given
543 int btrfs_del_csums(struct btrfs_trans_handle *trans,
544 struct btrfs_root *root, u64 bytenr, u64 len)
546 struct btrfs_path *path;
547 struct btrfs_key key;
548 u64 end_byte = bytenr + len;
550 struct extent_buffer *leaf;
553 btrfs_super_csum_size(&root->fs_info->super_copy);
554 int blocksize_bits = root->fs_info->sb->s_blocksize_bits;
556 root = root->fs_info->csum_root;
558 path = btrfs_alloc_path();
563 key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
564 key.offset = end_byte - 1;
565 key.type = BTRFS_EXTENT_CSUM_KEY;
567 path->leave_spinning = 1;
568 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
570 if (path->slots[0] == 0)
573 } else if (ret < 0) {
577 leaf = path->nodes[0];
578 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
580 if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
581 key.type != BTRFS_EXTENT_CSUM_KEY) {
585 if (key.offset >= end_byte)
588 csum_end = btrfs_item_size_nr(leaf, path->slots[0]) / csum_size;
589 csum_end <<= blocksize_bits;
590 csum_end += key.offset;
592 /* this csum ends before we start, we're done */
593 if (csum_end <= bytenr)
596 /* delete the entire item, it is inside our range */
597 if (key.offset >= bytenr && csum_end <= end_byte) {
598 ret = btrfs_del_item(trans, root, path);
601 if (key.offset == bytenr)
603 } else if (key.offset < bytenr && csum_end > end_byte) {
604 unsigned long offset;
605 unsigned long shift_len;
606 unsigned long item_offset;
611 * Our bytes are in the middle of the csum,
612 * we need to split this item and insert a new one.
614 * But we can't drop the path because the
615 * csum could change, get removed, extended etc.
617 * The trick here is the max size of a csum item leaves
618 * enough room in the tree block for a single
619 * item header. So, we split the item in place,
620 * adding a new header pointing to the existing
621 * bytes. Then we loop around again and we have
622 * a nicely formed csum item that we can neatly
625 offset = (bytenr - key.offset) >> blocksize_bits;
628 shift_len = (len >> blocksize_bits) * csum_size;
630 item_offset = btrfs_item_ptr_offset(leaf,
633 memset_extent_buffer(leaf, 0, item_offset + offset,
638 * btrfs_split_item returns -EAGAIN when the
639 * item changed size or key
641 ret = btrfs_split_item(trans, root, path, &key, offset);
642 BUG_ON(ret && ret != -EAGAIN);
644 key.offset = end_byte - 1;
646 ret = truncate_one_csum(trans, root, path,
649 if (key.offset < bytenr)
652 btrfs_release_path(path);
656 btrfs_free_path(path);
660 int btrfs_csum_file_blocks(struct btrfs_trans_handle *trans,
661 struct btrfs_root *root,
662 struct btrfs_ordered_sum *sums)
666 struct btrfs_key file_key;
667 struct btrfs_key found_key;
671 struct btrfs_path *path;
672 struct btrfs_csum_item *item;
673 struct btrfs_csum_item *item_end;
674 struct extent_buffer *leaf = NULL;
676 struct btrfs_sector_sum *sector_sum;
680 btrfs_super_csum_size(&root->fs_info->super_copy);
682 path = btrfs_alloc_path();
686 sector_sum = sums->sums;
688 next_offset = (u64)-1;
690 file_key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
691 file_key.offset = sector_sum->bytenr;
692 bytenr = sector_sum->bytenr;
693 btrfs_set_key_type(&file_key, BTRFS_EXTENT_CSUM_KEY);
695 item = btrfs_lookup_csum(trans, root, path, sector_sum->bytenr, 1);
697 leaf = path->nodes[0];
702 if (ret != -EFBIG && ret != -ENOENT)
707 /* we found one, but it isn't big enough yet */
708 leaf = path->nodes[0];
709 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
710 if ((item_size / csum_size) >=
711 MAX_CSUM_ITEMS(root, csum_size)) {
712 /* already at max size, make a new one */
716 int slot = path->slots[0] + 1;
717 /* we didn't find a csum item, insert one */
718 nritems = btrfs_header_nritems(path->nodes[0]);
719 if (path->slots[0] >= nritems - 1) {
720 ret = btrfs_next_leaf(root, path);
727 btrfs_item_key_to_cpu(path->nodes[0], &found_key, slot);
728 if (found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
729 found_key.type != BTRFS_EXTENT_CSUM_KEY) {
733 next_offset = found_key.offset;
739 * at this point, we know the tree has an item, but it isn't big
740 * enough yet to put our csum in. Grow it
742 btrfs_release_path(path);
743 ret = btrfs_search_slot(trans, root, &file_key, path,
749 if (path->slots[0] == 0)
754 leaf = path->nodes[0];
755 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
756 csum_offset = (bytenr - found_key.offset) >>
757 root->fs_info->sb->s_blocksize_bits;
759 if (btrfs_key_type(&found_key) != BTRFS_EXTENT_CSUM_KEY ||
760 found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
761 csum_offset >= MAX_CSUM_ITEMS(root, csum_size)) {
765 if (csum_offset >= btrfs_item_size_nr(leaf, path->slots[0]) /
767 u32 diff = (csum_offset + 1) * csum_size;
770 * is the item big enough already? we dropped our lock
771 * before and need to recheck
773 if (diff < btrfs_item_size_nr(leaf, path->slots[0]))
776 diff = diff - btrfs_item_size_nr(leaf, path->slots[0]);
777 if (diff != csum_size)
780 ret = btrfs_extend_item(trans, root, path, diff);
785 btrfs_release_path(path);
788 u64 tmp = total_bytes + root->sectorsize;
789 u64 next_sector = sector_sum->bytenr;
790 struct btrfs_sector_sum *next = sector_sum + 1;
792 while (tmp < sums->len) {
793 if (next_sector + root->sectorsize != next->bytenr)
795 tmp += root->sectorsize;
796 next_sector = next->bytenr;
799 tmp = min(tmp, next_offset - file_key.offset);
800 tmp >>= root->fs_info->sb->s_blocksize_bits;
801 tmp = max((u64)1, tmp);
802 tmp = min(tmp, (u64)MAX_CSUM_ITEMS(root, csum_size));
803 ins_size = csum_size * tmp;
805 ins_size = csum_size;
807 path->leave_spinning = 1;
808 ret = btrfs_insert_empty_item(trans, root, path, &file_key,
810 path->leave_spinning = 0;
818 leaf = path->nodes[0];
819 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item);
821 item = (struct btrfs_csum_item *)((unsigned char *)item +
822 csum_offset * csum_size);
824 item_end = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item);
825 item_end = (struct btrfs_csum_item *)((unsigned char *)item_end +
826 btrfs_item_size_nr(leaf, path->slots[0]));
829 write_extent_buffer(leaf, §or_sum->sum, (unsigned long)item, csum_size);
831 total_bytes += root->sectorsize;
833 if (total_bytes < sums->len) {
834 item = (struct btrfs_csum_item *)((char *)item +
836 if (item < item_end && bytenr + PAGE_CACHE_SIZE ==
837 sector_sum->bytenr) {
838 bytenr = sector_sum->bytenr;
843 btrfs_mark_buffer_dirty(path->nodes[0]);
844 if (total_bytes < sums->len) {
845 btrfs_release_path(path);
850 btrfs_free_path(path);