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(root, 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 %lu start %llu\n",
213 (unsigned long long)offset);
216 btrfs_release_path(root, path);
219 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
222 item_start_offset = found_key.offset;
223 item_size = btrfs_item_size_nr(path->nodes[0],
225 item_last_offset = item_start_offset +
226 (item_size / csum_size) *
228 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
229 struct btrfs_csum_item);
232 * this byte range must be able to fit inside
233 * a single leaf so it will also fit inside a u32
235 diff = disk_bytenr - item_start_offset;
236 diff = diff / root->sectorsize;
237 diff = diff * csum_size;
239 read_extent_buffer(path->nodes[0], &sum,
240 ((unsigned long)item) + diff,
246 set_state_private(io_tree, offset, sum);
247 disk_bytenr += bvec->bv_len;
248 offset += bvec->bv_len;
252 btrfs_free_path(path);
256 int btrfs_lookup_bio_sums(struct btrfs_root *root, struct inode *inode,
257 struct bio *bio, u32 *dst)
259 return __btrfs_lookup_bio_sums(root, inode, bio, 0, dst, 0);
262 int btrfs_lookup_bio_sums_dio(struct btrfs_root *root, struct inode *inode,
263 struct bio *bio, u64 offset, u32 *dst)
265 return __btrfs_lookup_bio_sums(root, inode, bio, offset, dst, 1);
268 int btrfs_lookup_csums_range(struct btrfs_root *root, u64 start, u64 end,
269 struct list_head *list)
271 struct btrfs_key key;
272 struct btrfs_path *path;
273 struct extent_buffer *leaf;
274 struct btrfs_ordered_sum *sums;
275 struct btrfs_sector_sum *sector_sum;
276 struct btrfs_csum_item *item;
277 unsigned long offset;
281 u16 csum_size = btrfs_super_csum_size(&root->fs_info->super_copy);
283 path = btrfs_alloc_path();
286 key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
288 key.type = BTRFS_EXTENT_CSUM_KEY;
290 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
293 if (ret > 0 && path->slots[0] > 0) {
294 leaf = path->nodes[0];
295 btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
296 if (key.objectid == BTRFS_EXTENT_CSUM_OBJECTID &&
297 key.type == BTRFS_EXTENT_CSUM_KEY) {
298 offset = (start - key.offset) >>
299 root->fs_info->sb->s_blocksize_bits;
300 if (offset * csum_size <
301 btrfs_item_size_nr(leaf, path->slots[0] - 1))
306 while (start <= end) {
307 leaf = path->nodes[0];
308 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
309 ret = btrfs_next_leaf(root, path);
314 leaf = path->nodes[0];
317 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
318 if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
319 key.type != BTRFS_EXTENT_CSUM_KEY)
322 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
323 if (key.offset > end)
326 if (key.offset > start)
329 size = btrfs_item_size_nr(leaf, path->slots[0]);
330 csum_end = key.offset + (size / csum_size) * root->sectorsize;
331 if (csum_end <= start) {
336 csum_end = min(csum_end, end + 1);
337 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
338 struct btrfs_csum_item);
339 while (start < csum_end) {
340 size = min_t(size_t, csum_end - start,
341 MAX_ORDERED_SUM_BYTES(root));
342 sums = kzalloc(btrfs_ordered_sum_size(root, size),
346 sector_sum = sums->sums;
347 sums->bytenr = start;
350 offset = (start - key.offset) >>
351 root->fs_info->sb->s_blocksize_bits;
355 read_extent_buffer(path->nodes[0],
357 ((unsigned long)item) +
359 sector_sum->bytenr = start;
361 size -= root->sectorsize;
362 start += root->sectorsize;
366 list_add_tail(&sums->list, list);
372 btrfs_free_path(path);
376 int btrfs_csum_one_bio(struct btrfs_root *root, struct inode *inode,
377 struct bio *bio, u64 file_start, int contig)
379 struct btrfs_ordered_sum *sums;
380 struct btrfs_sector_sum *sector_sum;
381 struct btrfs_ordered_extent *ordered;
383 struct bio_vec *bvec = bio->bi_io_vec;
385 unsigned long total_bytes = 0;
386 unsigned long this_sum_bytes = 0;
390 WARN_ON(bio->bi_vcnt <= 0);
391 sums = kzalloc(btrfs_ordered_sum_size(root, bio->bi_size), GFP_NOFS);
395 sector_sum = sums->sums;
396 disk_bytenr = (u64)bio->bi_sector << 9;
397 sums->len = bio->bi_size;
398 INIT_LIST_HEAD(&sums->list);
403 offset = page_offset(bvec->bv_page) + bvec->bv_offset;
405 ordered = btrfs_lookup_ordered_extent(inode, offset);
407 sums->bytenr = ordered->start;
409 while (bio_index < bio->bi_vcnt) {
411 offset = page_offset(bvec->bv_page) + bvec->bv_offset;
413 if (!contig && (offset >= ordered->file_offset + ordered->len ||
414 offset < ordered->file_offset)) {
415 unsigned long bytes_left;
416 sums->len = this_sum_bytes;
418 btrfs_add_ordered_sum(inode, ordered, sums);
419 btrfs_put_ordered_extent(ordered);
421 bytes_left = bio->bi_size - total_bytes;
423 sums = kzalloc(btrfs_ordered_sum_size(root, bytes_left),
426 sector_sum = sums->sums;
427 sums->len = bytes_left;
428 ordered = btrfs_lookup_ordered_extent(inode, offset);
430 sums->bytenr = ordered->start;
433 data = kmap_atomic(bvec->bv_page, KM_USER0);
434 sector_sum->sum = ~(u32)0;
435 sector_sum->sum = btrfs_csum_data(root,
436 data + bvec->bv_offset,
439 kunmap_atomic(data, KM_USER0);
440 btrfs_csum_final(sector_sum->sum,
441 (char *)§or_sum->sum);
442 sector_sum->bytenr = disk_bytenr;
446 total_bytes += bvec->bv_len;
447 this_sum_bytes += bvec->bv_len;
448 disk_bytenr += bvec->bv_len;
449 offset += bvec->bv_len;
453 btrfs_add_ordered_sum(inode, ordered, sums);
454 btrfs_put_ordered_extent(ordered);
459 * helper function for csum removal, this expects the
460 * key to describe the csum pointed to by the path, and it expects
461 * the csum to overlap the range [bytenr, len]
463 * The csum should not be entirely contained in the range and the
464 * range should not be entirely contained in the csum.
466 * This calls btrfs_truncate_item with the correct args based on the
467 * overlap, and fixes up the key as required.
469 static noinline int truncate_one_csum(struct btrfs_trans_handle *trans,
470 struct btrfs_root *root,
471 struct btrfs_path *path,
472 struct btrfs_key *key,
475 struct extent_buffer *leaf;
477 btrfs_super_csum_size(&root->fs_info->super_copy);
479 u64 end_byte = bytenr + len;
480 u32 blocksize_bits = root->fs_info->sb->s_blocksize_bits;
483 leaf = path->nodes[0];
484 csum_end = btrfs_item_size_nr(leaf, path->slots[0]) / csum_size;
485 csum_end <<= root->fs_info->sb->s_blocksize_bits;
486 csum_end += key->offset;
488 if (key->offset < bytenr && csum_end <= end_byte) {
493 * A simple truncate off the end of the item
495 u32 new_size = (bytenr - key->offset) >> blocksize_bits;
496 new_size *= csum_size;
497 ret = btrfs_truncate_item(trans, root, path, new_size, 1);
499 } else if (key->offset >= bytenr && csum_end > end_byte &&
500 end_byte > key->offset) {
505 * we need to truncate from the beginning of the csum
507 u32 new_size = (csum_end - end_byte) >> blocksize_bits;
508 new_size *= csum_size;
510 ret = btrfs_truncate_item(trans, root, path, new_size, 0);
513 key->offset = end_byte;
514 ret = btrfs_set_item_key_safe(trans, root, path, key);
523 * deletes the csum items from the csum tree for a given
526 int btrfs_del_csums(struct btrfs_trans_handle *trans,
527 struct btrfs_root *root, u64 bytenr, u64 len)
529 struct btrfs_path *path;
530 struct btrfs_key key;
531 u64 end_byte = bytenr + len;
533 struct extent_buffer *leaf;
536 btrfs_super_csum_size(&root->fs_info->super_copy);
537 int blocksize_bits = root->fs_info->sb->s_blocksize_bits;
539 root = root->fs_info->csum_root;
541 path = btrfs_alloc_path();
546 key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
547 key.offset = end_byte - 1;
548 key.type = BTRFS_EXTENT_CSUM_KEY;
550 path->leave_spinning = 1;
551 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
553 if (path->slots[0] == 0)
556 } else if (ret < 0) {
560 leaf = path->nodes[0];
561 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
563 if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
564 key.type != BTRFS_EXTENT_CSUM_KEY) {
568 if (key.offset >= end_byte)
571 csum_end = btrfs_item_size_nr(leaf, path->slots[0]) / csum_size;
572 csum_end <<= blocksize_bits;
573 csum_end += key.offset;
575 /* this csum ends before we start, we're done */
576 if (csum_end <= bytenr)
579 /* delete the entire item, it is inside our range */
580 if (key.offset >= bytenr && csum_end <= end_byte) {
581 ret = btrfs_del_item(trans, root, path);
583 if (key.offset == bytenr)
585 } else if (key.offset < bytenr && csum_end > end_byte) {
586 unsigned long offset;
587 unsigned long shift_len;
588 unsigned long item_offset;
593 * Our bytes are in the middle of the csum,
594 * we need to split this item and insert a new one.
596 * But we can't drop the path because the
597 * csum could change, get removed, extended etc.
599 * The trick here is the max size of a csum item leaves
600 * enough room in the tree block for a single
601 * item header. So, we split the item in place,
602 * adding a new header pointing to the existing
603 * bytes. Then we loop around again and we have
604 * a nicely formed csum item that we can neatly
607 offset = (bytenr - key.offset) >> blocksize_bits;
610 shift_len = (len >> blocksize_bits) * csum_size;
612 item_offset = btrfs_item_ptr_offset(leaf,
615 memset_extent_buffer(leaf, 0, item_offset + offset,
620 * btrfs_split_item returns -EAGAIN when the
621 * item changed size or key
623 ret = btrfs_split_item(trans, root, path, &key, offset);
624 BUG_ON(ret && ret != -EAGAIN);
626 key.offset = end_byte - 1;
628 ret = truncate_one_csum(trans, root, path,
631 if (key.offset < bytenr)
634 btrfs_release_path(root, path);
637 btrfs_free_path(path);
641 int btrfs_csum_file_blocks(struct btrfs_trans_handle *trans,
642 struct btrfs_root *root,
643 struct btrfs_ordered_sum *sums)
647 struct btrfs_key file_key;
648 struct btrfs_key found_key;
652 struct btrfs_path *path;
653 struct btrfs_csum_item *item;
654 struct btrfs_csum_item *item_end;
655 struct extent_buffer *leaf = NULL;
657 struct btrfs_sector_sum *sector_sum;
662 unsigned long map_len;
663 unsigned long map_start;
665 btrfs_super_csum_size(&root->fs_info->super_copy);
667 path = btrfs_alloc_path();
669 sector_sum = sums->sums;
671 next_offset = (u64)-1;
673 file_key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
674 file_key.offset = sector_sum->bytenr;
675 bytenr = sector_sum->bytenr;
676 btrfs_set_key_type(&file_key, BTRFS_EXTENT_CSUM_KEY);
678 item = btrfs_lookup_csum(trans, root, path, sector_sum->bytenr, 1);
680 leaf = path->nodes[0];
685 if (ret != -EFBIG && ret != -ENOENT)
690 /* we found one, but it isn't big enough yet */
691 leaf = path->nodes[0];
692 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
693 if ((item_size / csum_size) >=
694 MAX_CSUM_ITEMS(root, csum_size)) {
695 /* already at max size, make a new one */
699 int slot = path->slots[0] + 1;
700 /* we didn't find a csum item, insert one */
701 nritems = btrfs_header_nritems(path->nodes[0]);
702 if (path->slots[0] >= nritems - 1) {
703 ret = btrfs_next_leaf(root, path);
710 btrfs_item_key_to_cpu(path->nodes[0], &found_key, slot);
711 if (found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
712 found_key.type != BTRFS_EXTENT_CSUM_KEY) {
716 next_offset = found_key.offset;
722 * at this point, we know the tree has an item, but it isn't big
723 * enough yet to put our csum in. Grow it
725 btrfs_release_path(root, path);
726 ret = btrfs_search_slot(trans, root, &file_key, path,
732 if (path->slots[0] == 0)
737 leaf = path->nodes[0];
738 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
739 csum_offset = (bytenr - found_key.offset) >>
740 root->fs_info->sb->s_blocksize_bits;
742 if (btrfs_key_type(&found_key) != BTRFS_EXTENT_CSUM_KEY ||
743 found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
744 csum_offset >= MAX_CSUM_ITEMS(root, csum_size)) {
748 if (csum_offset >= btrfs_item_size_nr(leaf, path->slots[0]) /
750 u32 diff = (csum_offset + 1) * csum_size;
753 * is the item big enough already? we dropped our lock
754 * before and need to recheck
756 if (diff < btrfs_item_size_nr(leaf, path->slots[0]))
759 diff = diff - btrfs_item_size_nr(leaf, path->slots[0]);
760 if (diff != csum_size)
763 ret = btrfs_extend_item(trans, root, path, diff);
769 btrfs_release_path(root, path);
772 u64 tmp = total_bytes + root->sectorsize;
773 u64 next_sector = sector_sum->bytenr;
774 struct btrfs_sector_sum *next = sector_sum + 1;
776 while (tmp < sums->len) {
777 if (next_sector + root->sectorsize != next->bytenr)
779 tmp += root->sectorsize;
780 next_sector = next->bytenr;
783 tmp = min(tmp, next_offset - file_key.offset);
784 tmp >>= root->fs_info->sb->s_blocksize_bits;
785 tmp = max((u64)1, tmp);
786 tmp = min(tmp, (u64)MAX_CSUM_ITEMS(root, csum_size));
787 ins_size = csum_size * tmp;
789 ins_size = csum_size;
791 path->leave_spinning = 1;
792 ret = btrfs_insert_empty_item(trans, root, path, &file_key,
794 path->leave_spinning = 0;
802 leaf = path->nodes[0];
803 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item);
805 item = (struct btrfs_csum_item *)((unsigned char *)item +
806 csum_offset * csum_size);
808 item_end = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item);
809 item_end = (struct btrfs_csum_item *)((unsigned char *)item_end +
810 btrfs_item_size_nr(leaf, path->slots[0]));
815 (unsigned long)item + csum_size >= map_start + map_len) {
819 unmap_extent_buffer(leaf, eb_token, KM_USER1);
821 err = map_private_extent_buffer(leaf, (unsigned long)item,
824 &map_start, &map_len, KM_USER1);
829 memcpy(eb_token + ((unsigned long)item & (PAGE_CACHE_SIZE - 1)),
830 §or_sum->sum, csum_size);
832 write_extent_buffer(leaf, §or_sum->sum,
833 (unsigned long)item, csum_size);
836 total_bytes += root->sectorsize;
838 if (total_bytes < sums->len) {
839 item = (struct btrfs_csum_item *)((char *)item +
841 if (item < item_end && bytenr + PAGE_CACHE_SIZE ==
842 sector_sum->bytenr) {
843 bytenr = sector_sum->bytenr;
848 unmap_extent_buffer(leaf, eb_token, KM_USER1);
851 btrfs_mark_buffer_dirty(path->nodes[0]);
852 if (total_bytes < sums->len) {
853 btrfs_release_path(root, path);
858 btrfs_free_path(path);