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();
52 file_key.objectid = objectid;
53 file_key.offset = pos;
54 btrfs_set_key_type(&file_key, BTRFS_EXTENT_DATA_KEY);
56 path->leave_spinning = 1;
57 ret = btrfs_insert_empty_item(trans, root, path, &file_key,
62 leaf = path->nodes[0];
63 item = btrfs_item_ptr(leaf, path->slots[0],
64 struct btrfs_file_extent_item);
65 btrfs_set_file_extent_disk_bytenr(leaf, item, disk_offset);
66 btrfs_set_file_extent_disk_num_bytes(leaf, item, disk_num_bytes);
67 btrfs_set_file_extent_offset(leaf, item, offset);
68 btrfs_set_file_extent_num_bytes(leaf, item, num_bytes);
69 btrfs_set_file_extent_ram_bytes(leaf, item, ram_bytes);
70 btrfs_set_file_extent_generation(leaf, item, trans->transid);
71 btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
72 btrfs_set_file_extent_compression(leaf, item, compression);
73 btrfs_set_file_extent_encryption(leaf, item, encryption);
74 btrfs_set_file_extent_other_encoding(leaf, item, other_encoding);
76 btrfs_mark_buffer_dirty(leaf);
78 btrfs_free_path(path);
82 struct btrfs_csum_item *btrfs_lookup_csum(struct btrfs_trans_handle *trans,
83 struct btrfs_root *root,
84 struct btrfs_path *path,
88 struct btrfs_key file_key;
89 struct btrfs_key found_key;
90 struct btrfs_csum_item *item;
91 struct extent_buffer *leaf;
94 btrfs_super_csum_size(&root->fs_info->super_copy);
97 file_key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
98 file_key.offset = bytenr;
99 btrfs_set_key_type(&file_key, BTRFS_EXTENT_CSUM_KEY);
100 ret = btrfs_search_slot(trans, root, &file_key, path, 0, cow);
103 leaf = path->nodes[0];
106 if (path->slots[0] == 0)
109 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
110 if (btrfs_key_type(&found_key) != BTRFS_EXTENT_CSUM_KEY)
113 csum_offset = (bytenr - found_key.offset) >>
114 root->fs_info->sb->s_blocksize_bits;
115 csums_in_item = btrfs_item_size_nr(leaf, path->slots[0]);
116 csums_in_item /= csum_size;
118 if (csum_offset >= csums_in_item) {
123 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item);
124 item = (struct btrfs_csum_item *)((unsigned char *)item +
125 csum_offset * csum_size);
134 int btrfs_lookup_file_extent(struct btrfs_trans_handle *trans,
135 struct btrfs_root *root,
136 struct btrfs_path *path, u64 objectid,
140 struct btrfs_key file_key;
141 int ins_len = mod < 0 ? -1 : 0;
144 file_key.objectid = objectid;
145 file_key.offset = offset;
146 btrfs_set_key_type(&file_key, BTRFS_EXTENT_DATA_KEY);
147 ret = btrfs_search_slot(trans, root, &file_key, path, ins_len, cow);
152 static int __btrfs_lookup_bio_sums(struct btrfs_root *root,
153 struct inode *inode, struct bio *bio,
154 u64 logical_offset, u32 *dst, int dio)
157 struct bio_vec *bvec = bio->bi_io_vec;
160 u64 item_start_offset = 0;
161 u64 item_last_offset = 0;
165 btrfs_super_csum_size(&root->fs_info->super_copy);
167 struct btrfs_path *path;
168 struct btrfs_csum_item *item = NULL;
169 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
171 path = btrfs_alloc_path();
172 if (bio->bi_size > PAGE_CACHE_SIZE * 8)
175 WARN_ON(bio->bi_vcnt <= 0);
177 disk_bytenr = (u64)bio->bi_sector << 9;
179 offset = logical_offset;
180 while (bio_index < bio->bi_vcnt) {
182 offset = page_offset(bvec->bv_page) + bvec->bv_offset;
183 ret = btrfs_find_ordered_sum(inode, offset, disk_bytenr, &sum);
187 if (!item || disk_bytenr < item_start_offset ||
188 disk_bytenr >= item_last_offset) {
189 struct btrfs_key found_key;
193 btrfs_release_path(root, path);
194 item = btrfs_lookup_csum(NULL, root->fs_info->csum_root,
195 path, disk_bytenr, 0);
198 if (ret == -ENOENT || ret == -EFBIG)
201 if (BTRFS_I(inode)->root->root_key.objectid ==
202 BTRFS_DATA_RELOC_TREE_OBJECTID) {
203 set_extent_bits(io_tree, offset,
204 offset + bvec->bv_len - 1,
205 EXTENT_NODATASUM, GFP_NOFS);
207 printk(KERN_INFO "btrfs no csum found "
208 "for inode %lu start %llu\n",
210 (unsigned long long)offset);
213 btrfs_release_path(root, path);
216 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
219 item_start_offset = found_key.offset;
220 item_size = btrfs_item_size_nr(path->nodes[0],
222 item_last_offset = item_start_offset +
223 (item_size / csum_size) *
225 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
226 struct btrfs_csum_item);
229 * this byte range must be able to fit inside
230 * a single leaf so it will also fit inside a u32
232 diff = disk_bytenr - item_start_offset;
233 diff = diff / root->sectorsize;
234 diff = diff * csum_size;
236 read_extent_buffer(path->nodes[0], &sum,
237 ((unsigned long)item) + diff,
243 set_state_private(io_tree, offset, sum);
244 disk_bytenr += bvec->bv_len;
245 offset += bvec->bv_len;
249 btrfs_free_path(path);
253 int btrfs_lookup_bio_sums(struct btrfs_root *root, struct inode *inode,
254 struct bio *bio, u32 *dst)
256 return __btrfs_lookup_bio_sums(root, inode, bio, 0, dst, 0);
259 int btrfs_lookup_bio_sums_dio(struct btrfs_root *root, struct inode *inode,
260 struct bio *bio, u64 offset, u32 *dst)
262 return __btrfs_lookup_bio_sums(root, inode, bio, offset, dst, 1);
265 int btrfs_lookup_csums_range(struct btrfs_root *root, u64 start, u64 end,
266 struct list_head *list)
268 struct btrfs_key key;
269 struct btrfs_path *path;
270 struct extent_buffer *leaf;
271 struct btrfs_ordered_sum *sums;
272 struct btrfs_sector_sum *sector_sum;
273 struct btrfs_csum_item *item;
274 unsigned long offset;
278 u16 csum_size = btrfs_super_csum_size(&root->fs_info->super_copy);
280 path = btrfs_alloc_path();
283 key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
285 key.type = BTRFS_EXTENT_CSUM_KEY;
287 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
290 if (ret > 0 && path->slots[0] > 0) {
291 leaf = path->nodes[0];
292 btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
293 if (key.objectid == BTRFS_EXTENT_CSUM_OBJECTID &&
294 key.type == BTRFS_EXTENT_CSUM_KEY) {
295 offset = (start - key.offset) >>
296 root->fs_info->sb->s_blocksize_bits;
297 if (offset * csum_size <
298 btrfs_item_size_nr(leaf, path->slots[0] - 1))
303 while (start <= end) {
304 leaf = path->nodes[0];
305 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
306 ret = btrfs_next_leaf(root, path);
311 leaf = path->nodes[0];
314 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
315 if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
316 key.type != BTRFS_EXTENT_CSUM_KEY)
319 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
320 if (key.offset > end)
323 if (key.offset > start)
326 size = btrfs_item_size_nr(leaf, path->slots[0]);
327 csum_end = key.offset + (size / csum_size) * root->sectorsize;
328 if (csum_end <= start) {
333 csum_end = min(csum_end, end + 1);
334 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
335 struct btrfs_csum_item);
336 while (start < csum_end) {
337 size = min_t(size_t, csum_end - start,
338 MAX_ORDERED_SUM_BYTES(root));
339 sums = kzalloc(btrfs_ordered_sum_size(root, size),
343 sector_sum = sums->sums;
344 sums->bytenr = start;
347 offset = (start - key.offset) >>
348 root->fs_info->sb->s_blocksize_bits;
352 read_extent_buffer(path->nodes[0],
354 ((unsigned long)item) +
356 sector_sum->bytenr = start;
358 size -= root->sectorsize;
359 start += root->sectorsize;
363 list_add_tail(&sums->list, list);
369 btrfs_free_path(path);
373 int btrfs_csum_one_bio(struct btrfs_root *root, struct inode *inode,
374 struct bio *bio, u64 file_start, int contig)
376 struct btrfs_ordered_sum *sums;
377 struct btrfs_sector_sum *sector_sum;
378 struct btrfs_ordered_extent *ordered;
380 struct bio_vec *bvec = bio->bi_io_vec;
382 unsigned long total_bytes = 0;
383 unsigned long this_sum_bytes = 0;
387 WARN_ON(bio->bi_vcnt <= 0);
388 sums = kzalloc(btrfs_ordered_sum_size(root, bio->bi_size), GFP_NOFS);
392 sector_sum = sums->sums;
393 disk_bytenr = (u64)bio->bi_sector << 9;
394 sums->len = bio->bi_size;
395 INIT_LIST_HEAD(&sums->list);
400 offset = page_offset(bvec->bv_page) + bvec->bv_offset;
402 ordered = btrfs_lookup_ordered_extent(inode, offset);
404 sums->bytenr = ordered->start;
406 while (bio_index < bio->bi_vcnt) {
408 offset = page_offset(bvec->bv_page) + bvec->bv_offset;
410 if (!contig && (offset >= ordered->file_offset + ordered->len ||
411 offset < ordered->file_offset)) {
412 unsigned long bytes_left;
413 sums->len = this_sum_bytes;
415 btrfs_add_ordered_sum(inode, ordered, sums);
416 btrfs_put_ordered_extent(ordered);
418 bytes_left = bio->bi_size - total_bytes;
420 sums = kzalloc(btrfs_ordered_sum_size(root, bytes_left),
423 sector_sum = sums->sums;
424 sums->len = bytes_left;
425 ordered = btrfs_lookup_ordered_extent(inode, offset);
427 sums->bytenr = ordered->start;
430 data = kmap_atomic(bvec->bv_page, KM_USER0);
431 sector_sum->sum = ~(u32)0;
432 sector_sum->sum = btrfs_csum_data(root,
433 data + bvec->bv_offset,
436 kunmap_atomic(data, KM_USER0);
437 btrfs_csum_final(sector_sum->sum,
438 (char *)§or_sum->sum);
439 sector_sum->bytenr = disk_bytenr;
443 total_bytes += bvec->bv_len;
444 this_sum_bytes += bvec->bv_len;
445 disk_bytenr += bvec->bv_len;
446 offset += bvec->bv_len;
450 btrfs_add_ordered_sum(inode, ordered, sums);
451 btrfs_put_ordered_extent(ordered);
456 * helper function for csum removal, this expects the
457 * key to describe the csum pointed to by the path, and it expects
458 * the csum to overlap the range [bytenr, len]
460 * The csum should not be entirely contained in the range and the
461 * range should not be entirely contained in the csum.
463 * This calls btrfs_truncate_item with the correct args based on the
464 * overlap, and fixes up the key as required.
466 static noinline int truncate_one_csum(struct btrfs_trans_handle *trans,
467 struct btrfs_root *root,
468 struct btrfs_path *path,
469 struct btrfs_key *key,
472 struct extent_buffer *leaf;
474 btrfs_super_csum_size(&root->fs_info->super_copy);
476 u64 end_byte = bytenr + len;
477 u32 blocksize_bits = root->fs_info->sb->s_blocksize_bits;
480 leaf = path->nodes[0];
481 csum_end = btrfs_item_size_nr(leaf, path->slots[0]) / csum_size;
482 csum_end <<= root->fs_info->sb->s_blocksize_bits;
483 csum_end += key->offset;
485 if (key->offset < bytenr && csum_end <= end_byte) {
490 * A simple truncate off the end of the item
492 u32 new_size = (bytenr - key->offset) >> blocksize_bits;
493 new_size *= csum_size;
494 ret = btrfs_truncate_item(trans, root, path, new_size, 1);
496 } else if (key->offset >= bytenr && csum_end > end_byte &&
497 end_byte > key->offset) {
502 * we need to truncate from the beginning of the csum
504 u32 new_size = (csum_end - end_byte) >> blocksize_bits;
505 new_size *= csum_size;
507 ret = btrfs_truncate_item(trans, root, path, new_size, 0);
510 key->offset = end_byte;
511 ret = btrfs_set_item_key_safe(trans, root, path, key);
520 * deletes the csum items from the csum tree for a given
523 int btrfs_del_csums(struct btrfs_trans_handle *trans,
524 struct btrfs_root *root, u64 bytenr, u64 len)
526 struct btrfs_path *path;
527 struct btrfs_key key;
528 u64 end_byte = bytenr + len;
530 struct extent_buffer *leaf;
533 btrfs_super_csum_size(&root->fs_info->super_copy);
534 int blocksize_bits = root->fs_info->sb->s_blocksize_bits;
536 root = root->fs_info->csum_root;
538 path = btrfs_alloc_path();
541 key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
542 key.offset = end_byte - 1;
543 key.type = BTRFS_EXTENT_CSUM_KEY;
545 path->leave_spinning = 1;
546 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
548 if (path->slots[0] == 0)
552 leaf = path->nodes[0];
553 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
555 if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
556 key.type != BTRFS_EXTENT_CSUM_KEY) {
560 if (key.offset >= end_byte)
563 csum_end = btrfs_item_size_nr(leaf, path->slots[0]) / csum_size;
564 csum_end <<= blocksize_bits;
565 csum_end += key.offset;
567 /* this csum ends before we start, we're done */
568 if (csum_end <= bytenr)
571 /* delete the entire item, it is inside our range */
572 if (key.offset >= bytenr && csum_end <= end_byte) {
573 ret = btrfs_del_item(trans, root, path);
575 if (key.offset == bytenr)
577 } else if (key.offset < bytenr && csum_end > end_byte) {
578 unsigned long offset;
579 unsigned long shift_len;
580 unsigned long item_offset;
585 * Our bytes are in the middle of the csum,
586 * we need to split this item and insert a new one.
588 * But we can't drop the path because the
589 * csum could change, get removed, extended etc.
591 * The trick here is the max size of a csum item leaves
592 * enough room in the tree block for a single
593 * item header. So, we split the item in place,
594 * adding a new header pointing to the existing
595 * bytes. Then we loop around again and we have
596 * a nicely formed csum item that we can neatly
599 offset = (bytenr - key.offset) >> blocksize_bits;
602 shift_len = (len >> blocksize_bits) * csum_size;
604 item_offset = btrfs_item_ptr_offset(leaf,
607 memset_extent_buffer(leaf, 0, item_offset + offset,
612 * btrfs_split_item returns -EAGAIN when the
613 * item changed size or key
615 ret = btrfs_split_item(trans, root, path, &key, offset);
616 BUG_ON(ret && ret != -EAGAIN);
618 key.offset = end_byte - 1;
620 ret = truncate_one_csum(trans, root, path,
623 if (key.offset < bytenr)
626 btrfs_release_path(root, path);
629 btrfs_free_path(path);
633 int btrfs_csum_file_blocks(struct btrfs_trans_handle *trans,
634 struct btrfs_root *root,
635 struct btrfs_ordered_sum *sums)
639 struct btrfs_key file_key;
640 struct btrfs_key found_key;
644 struct btrfs_path *path;
645 struct btrfs_csum_item *item;
646 struct btrfs_csum_item *item_end;
647 struct extent_buffer *leaf = NULL;
649 struct btrfs_sector_sum *sector_sum;
654 unsigned long map_len;
655 unsigned long map_start;
657 btrfs_super_csum_size(&root->fs_info->super_copy);
659 path = btrfs_alloc_path();
661 sector_sum = sums->sums;
663 next_offset = (u64)-1;
665 file_key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
666 file_key.offset = sector_sum->bytenr;
667 bytenr = sector_sum->bytenr;
668 btrfs_set_key_type(&file_key, BTRFS_EXTENT_CSUM_KEY);
670 item = btrfs_lookup_csum(trans, root, path, sector_sum->bytenr, 1);
672 leaf = path->nodes[0];
677 if (ret != -EFBIG && ret != -ENOENT)
682 /* we found one, but it isn't big enough yet */
683 leaf = path->nodes[0];
684 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
685 if ((item_size / csum_size) >=
686 MAX_CSUM_ITEMS(root, csum_size)) {
687 /* already at max size, make a new one */
691 int slot = path->slots[0] + 1;
692 /* we didn't find a csum item, insert one */
693 nritems = btrfs_header_nritems(path->nodes[0]);
694 if (path->slots[0] >= nritems - 1) {
695 ret = btrfs_next_leaf(root, path);
702 btrfs_item_key_to_cpu(path->nodes[0], &found_key, slot);
703 if (found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
704 found_key.type != BTRFS_EXTENT_CSUM_KEY) {
708 next_offset = found_key.offset;
714 * at this point, we know the tree has an item, but it isn't big
715 * enough yet to put our csum in. Grow it
717 btrfs_release_path(root, path);
718 ret = btrfs_search_slot(trans, root, &file_key, path,
724 if (path->slots[0] == 0)
729 leaf = path->nodes[0];
730 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
731 csum_offset = (bytenr - found_key.offset) >>
732 root->fs_info->sb->s_blocksize_bits;
734 if (btrfs_key_type(&found_key) != BTRFS_EXTENT_CSUM_KEY ||
735 found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
736 csum_offset >= MAX_CSUM_ITEMS(root, csum_size)) {
740 if (csum_offset >= btrfs_item_size_nr(leaf, path->slots[0]) /
742 u32 diff = (csum_offset + 1) * csum_size;
745 * is the item big enough already? we dropped our lock
746 * before and need to recheck
748 if (diff < btrfs_item_size_nr(leaf, path->slots[0]))
751 diff = diff - btrfs_item_size_nr(leaf, path->slots[0]);
752 if (diff != csum_size)
755 ret = btrfs_extend_item(trans, root, path, diff);
761 btrfs_release_path(root, path);
764 u64 tmp = total_bytes + root->sectorsize;
765 u64 next_sector = sector_sum->bytenr;
766 struct btrfs_sector_sum *next = sector_sum + 1;
768 while (tmp < sums->len) {
769 if (next_sector + root->sectorsize != next->bytenr)
771 tmp += root->sectorsize;
772 next_sector = next->bytenr;
775 tmp = min(tmp, next_offset - file_key.offset);
776 tmp >>= root->fs_info->sb->s_blocksize_bits;
777 tmp = max((u64)1, tmp);
778 tmp = min(tmp, (u64)MAX_CSUM_ITEMS(root, csum_size));
779 ins_size = csum_size * tmp;
781 ins_size = csum_size;
783 path->leave_spinning = 1;
784 ret = btrfs_insert_empty_item(trans, root, path, &file_key,
786 path->leave_spinning = 0;
794 leaf = path->nodes[0];
795 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item);
797 item = (struct btrfs_csum_item *)((unsigned char *)item +
798 csum_offset * csum_size);
800 item_end = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item);
801 item_end = (struct btrfs_csum_item *)((unsigned char *)item_end +
802 btrfs_item_size_nr(leaf, path->slots[0]));
807 (unsigned long)item + csum_size >= map_start + map_len) {
811 unmap_extent_buffer(leaf, eb_token, KM_USER1);
813 err = map_private_extent_buffer(leaf, (unsigned long)item,
816 &map_start, &map_len, KM_USER1);
821 memcpy(eb_token + ((unsigned long)item & (PAGE_CACHE_SIZE - 1)),
822 §or_sum->sum, csum_size);
824 write_extent_buffer(leaf, §or_sum->sum,
825 (unsigned long)item, csum_size);
828 total_bytes += root->sectorsize;
830 if (total_bytes < sums->len) {
831 item = (struct btrfs_csum_item *)((char *)item +
833 if (item < item_end && bytenr + PAGE_CACHE_SIZE ==
834 sector_sum->bytenr) {
835 bytenr = sector_sum->bytenr;
840 unmap_extent_buffer(leaf, eb_token, KM_USER1);
843 btrfs_mark_buffer_dirty(path->nodes[0]);
844 if (total_bytes < sums->len) {
845 btrfs_release_path(root, path);
850 btrfs_free_path(path);