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_CSUM_ITEMS(r, size) (min(__MAX_CSUM_ITEMS(r, size), PAGE_CACHE_SIZE))
34 #define MAX_ORDERED_SUM_BYTES(r) ((PAGE_SIZE - \
35 sizeof(struct btrfs_ordered_sum)) / \
36 sizeof(struct btrfs_sector_sum) * \
37 (r)->sectorsize - (r)->sectorsize)
39 int btrfs_insert_file_extent(struct btrfs_trans_handle *trans,
40 struct btrfs_root *root,
41 u64 objectid, u64 pos,
42 u64 disk_offset, u64 disk_num_bytes,
43 u64 num_bytes, u64 offset, u64 ram_bytes,
44 u8 compression, u8 encryption, u16 other_encoding)
47 struct btrfs_file_extent_item *item;
48 struct btrfs_key file_key;
49 struct btrfs_path *path;
50 struct extent_buffer *leaf;
52 path = btrfs_alloc_path();
55 file_key.objectid = objectid;
56 file_key.offset = pos;
57 btrfs_set_key_type(&file_key, BTRFS_EXTENT_DATA_KEY);
59 path->leave_spinning = 1;
60 ret = btrfs_insert_empty_item(trans, root, path, &file_key,
64 BUG_ON(ret); /* Can't happen */
65 leaf = path->nodes[0];
66 item = btrfs_item_ptr(leaf, path->slots[0],
67 struct btrfs_file_extent_item);
68 btrfs_set_file_extent_disk_bytenr(leaf, item, disk_offset);
69 btrfs_set_file_extent_disk_num_bytes(leaf, item, disk_num_bytes);
70 btrfs_set_file_extent_offset(leaf, item, offset);
71 btrfs_set_file_extent_num_bytes(leaf, item, num_bytes);
72 btrfs_set_file_extent_ram_bytes(leaf, item, ram_bytes);
73 btrfs_set_file_extent_generation(leaf, item, trans->transid);
74 btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
75 btrfs_set_file_extent_compression(leaf, item, compression);
76 btrfs_set_file_extent_encryption(leaf, item, encryption);
77 btrfs_set_file_extent_other_encoding(leaf, item, other_encoding);
79 btrfs_mark_buffer_dirty(leaf);
81 btrfs_free_path(path);
85 struct btrfs_csum_item *btrfs_lookup_csum(struct btrfs_trans_handle *trans,
86 struct btrfs_root *root,
87 struct btrfs_path *path,
91 struct btrfs_key file_key;
92 struct btrfs_key found_key;
93 struct btrfs_csum_item *item;
94 struct extent_buffer *leaf;
96 u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
99 file_key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
100 file_key.offset = bytenr;
101 btrfs_set_key_type(&file_key, BTRFS_EXTENT_CSUM_KEY);
102 ret = btrfs_search_slot(trans, root, &file_key, path, 0, cow);
105 leaf = path->nodes[0];
108 if (path->slots[0] == 0)
111 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
112 if (btrfs_key_type(&found_key) != BTRFS_EXTENT_CSUM_KEY)
115 csum_offset = (bytenr - found_key.offset) >>
116 root->fs_info->sb->s_blocksize_bits;
117 csums_in_item = btrfs_item_size_nr(leaf, path->slots[0]);
118 csums_in_item /= csum_size;
120 if (csum_offset >= csums_in_item) {
125 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item);
126 item = (struct btrfs_csum_item *)((unsigned char *)item +
127 csum_offset * csum_size);
136 int btrfs_lookup_file_extent(struct btrfs_trans_handle *trans,
137 struct btrfs_root *root,
138 struct btrfs_path *path, u64 objectid,
142 struct btrfs_key file_key;
143 int ins_len = mod < 0 ? -1 : 0;
146 file_key.objectid = objectid;
147 file_key.offset = offset;
148 btrfs_set_key_type(&file_key, BTRFS_EXTENT_DATA_KEY);
149 ret = btrfs_search_slot(trans, root, &file_key, path, ins_len, cow);
154 static int __btrfs_lookup_bio_sums(struct btrfs_root *root,
155 struct inode *inode, struct bio *bio,
156 u64 logical_offset, u32 *dst, int dio)
159 struct bio_vec *bvec = bio->bi_io_vec;
162 u64 item_start_offset = 0;
163 u64 item_last_offset = 0;
166 u16 csum_size = 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(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)
277 return __btrfs_lookup_bio_sums(root, inode, bio, offset, NULL, 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;
290 unsigned long offset;
294 u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
296 path = btrfs_alloc_path();
301 path->skip_locking = 1;
303 path->search_commit_root = 1;
306 key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
308 key.type = BTRFS_EXTENT_CSUM_KEY;
310 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
313 if (ret > 0 && path->slots[0] > 0) {
314 leaf = path->nodes[0];
315 btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
316 if (key.objectid == BTRFS_EXTENT_CSUM_OBJECTID &&
317 key.type == BTRFS_EXTENT_CSUM_KEY) {
318 offset = (start - key.offset) >>
319 root->fs_info->sb->s_blocksize_bits;
320 if (offset * csum_size <
321 btrfs_item_size_nr(leaf, path->slots[0] - 1))
326 while (start <= end) {
327 leaf = path->nodes[0];
328 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
329 ret = btrfs_next_leaf(root, path);
334 leaf = path->nodes[0];
337 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
338 if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
339 key.type != BTRFS_EXTENT_CSUM_KEY)
342 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
343 if (key.offset > end)
346 if (key.offset > start)
349 size = btrfs_item_size_nr(leaf, path->slots[0]);
350 csum_end = key.offset + (size / csum_size) * root->sectorsize;
351 if (csum_end <= start) {
356 csum_end = min(csum_end, end + 1);
357 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
358 struct btrfs_csum_item);
359 while (start < csum_end) {
360 size = min_t(size_t, csum_end - start,
361 MAX_ORDERED_SUM_BYTES(root));
362 sums = kzalloc(btrfs_ordered_sum_size(root, size),
369 sector_sum = sums->sums;
370 sums->bytenr = start;
373 offset = (start - key.offset) >>
374 root->fs_info->sb->s_blocksize_bits;
378 read_extent_buffer(path->nodes[0],
380 ((unsigned long)item) +
382 sector_sum->bytenr = start;
384 size -= root->sectorsize;
385 start += root->sectorsize;
389 list_add_tail(&sums->list, &tmplist);
395 while (ret < 0 && !list_empty(&tmplist)) {
396 sums = list_entry(&tmplist, struct btrfs_ordered_sum, list);
397 list_del(&sums->list);
400 list_splice_tail(&tmplist, list);
402 btrfs_free_path(path);
406 int btrfs_csum_one_bio(struct btrfs_root *root, struct inode *inode,
407 struct bio *bio, u64 file_start, int contig)
409 struct btrfs_ordered_sum *sums;
410 struct btrfs_sector_sum *sector_sum;
411 struct btrfs_ordered_extent *ordered;
413 struct bio_vec *bvec = bio->bi_io_vec;
415 unsigned long total_bytes = 0;
416 unsigned long this_sum_bytes = 0;
420 WARN_ON(bio->bi_vcnt <= 0);
421 sums = kzalloc(btrfs_ordered_sum_size(root, bio->bi_size), GFP_NOFS);
425 sector_sum = sums->sums;
426 disk_bytenr = (u64)bio->bi_sector << 9;
427 sums->len = bio->bi_size;
428 INIT_LIST_HEAD(&sums->list);
433 offset = page_offset(bvec->bv_page) + bvec->bv_offset;
435 ordered = btrfs_lookup_ordered_extent(inode, offset);
436 BUG_ON(!ordered); /* Logic error */
437 sums->bytenr = ordered->start;
439 while (bio_index < bio->bi_vcnt) {
441 offset = page_offset(bvec->bv_page) + bvec->bv_offset;
443 if (!contig && (offset >= ordered->file_offset + ordered->len ||
444 offset < ordered->file_offset)) {
445 unsigned long bytes_left;
446 sums->len = this_sum_bytes;
448 btrfs_add_ordered_sum(inode, ordered, sums);
449 btrfs_put_ordered_extent(ordered);
451 bytes_left = bio->bi_size - total_bytes;
453 sums = kzalloc(btrfs_ordered_sum_size(root, bytes_left),
455 BUG_ON(!sums); /* -ENOMEM */
456 sector_sum = sums->sums;
457 sums->len = bytes_left;
458 ordered = btrfs_lookup_ordered_extent(inode, offset);
459 BUG_ON(!ordered); /* Logic error */
460 sums->bytenr = ordered->start;
463 data = kmap_atomic(bvec->bv_page);
464 sector_sum->sum = ~(u32)0;
465 sector_sum->sum = btrfs_csum_data(root,
466 data + bvec->bv_offset,
470 btrfs_csum_final(sector_sum->sum,
471 (char *)§or_sum->sum);
472 sector_sum->bytenr = disk_bytenr;
476 total_bytes += bvec->bv_len;
477 this_sum_bytes += bvec->bv_len;
478 disk_bytenr += bvec->bv_len;
479 offset += bvec->bv_len;
483 btrfs_add_ordered_sum(inode, ordered, sums);
484 btrfs_put_ordered_extent(ordered);
489 * helper function for csum removal, this expects the
490 * key to describe the csum pointed to by the path, and it expects
491 * the csum to overlap the range [bytenr, len]
493 * The csum should not be entirely contained in the range and the
494 * range should not be entirely contained in the csum.
496 * This calls btrfs_truncate_item with the correct args based on the
497 * overlap, and fixes up the key as required.
499 static noinline void truncate_one_csum(struct btrfs_trans_handle *trans,
500 struct btrfs_root *root,
501 struct btrfs_path *path,
502 struct btrfs_key *key,
505 struct extent_buffer *leaf;
506 u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
508 u64 end_byte = bytenr + len;
509 u32 blocksize_bits = root->fs_info->sb->s_blocksize_bits;
511 leaf = path->nodes[0];
512 csum_end = btrfs_item_size_nr(leaf, path->slots[0]) / csum_size;
513 csum_end <<= root->fs_info->sb->s_blocksize_bits;
514 csum_end += key->offset;
516 if (key->offset < bytenr && csum_end <= end_byte) {
521 * A simple truncate off the end of the item
523 u32 new_size = (bytenr - key->offset) >> blocksize_bits;
524 new_size *= csum_size;
525 btrfs_truncate_item(trans, root, path, new_size, 1);
526 } else if (key->offset >= bytenr && csum_end > end_byte &&
527 end_byte > key->offset) {
532 * we need to truncate from the beginning of the csum
534 u32 new_size = (csum_end - end_byte) >> blocksize_bits;
535 new_size *= csum_size;
537 btrfs_truncate_item(trans, root, path, new_size, 0);
539 key->offset = end_byte;
540 btrfs_set_item_key_safe(trans, root, path, key);
547 * deletes the csum items from the csum tree for a given
550 int btrfs_del_csums(struct btrfs_trans_handle *trans,
551 struct btrfs_root *root, u64 bytenr, u64 len)
553 struct btrfs_path *path;
554 struct btrfs_key key;
555 u64 end_byte = bytenr + len;
557 struct extent_buffer *leaf;
559 u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
560 int blocksize_bits = root->fs_info->sb->s_blocksize_bits;
562 root = root->fs_info->csum_root;
564 path = btrfs_alloc_path();
569 key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
570 key.offset = end_byte - 1;
571 key.type = BTRFS_EXTENT_CSUM_KEY;
573 path->leave_spinning = 1;
574 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
576 if (path->slots[0] == 0)
579 } else if (ret < 0) {
583 leaf = path->nodes[0];
584 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
586 if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
587 key.type != BTRFS_EXTENT_CSUM_KEY) {
591 if (key.offset >= end_byte)
594 csum_end = btrfs_item_size_nr(leaf, path->slots[0]) / csum_size;
595 csum_end <<= blocksize_bits;
596 csum_end += key.offset;
598 /* this csum ends before we start, we're done */
599 if (csum_end <= bytenr)
602 /* delete the entire item, it is inside our range */
603 if (key.offset >= bytenr && csum_end <= end_byte) {
604 ret = btrfs_del_item(trans, root, path);
607 if (key.offset == bytenr)
609 } else if (key.offset < bytenr && csum_end > end_byte) {
610 unsigned long offset;
611 unsigned long shift_len;
612 unsigned long item_offset;
617 * Our bytes are in the middle of the csum,
618 * we need to split this item and insert a new one.
620 * But we can't drop the path because the
621 * csum could change, get removed, extended etc.
623 * The trick here is the max size of a csum item leaves
624 * enough room in the tree block for a single
625 * item header. So, we split the item in place,
626 * adding a new header pointing to the existing
627 * bytes. Then we loop around again and we have
628 * a nicely formed csum item that we can neatly
631 offset = (bytenr - key.offset) >> blocksize_bits;
634 shift_len = (len >> blocksize_bits) * csum_size;
636 item_offset = btrfs_item_ptr_offset(leaf,
639 memset_extent_buffer(leaf, 0, item_offset + offset,
644 * btrfs_split_item returns -EAGAIN when the
645 * item changed size or key
647 ret = btrfs_split_item(trans, root, path, &key, offset);
648 if (ret && ret != -EAGAIN) {
649 btrfs_abort_transaction(trans, root, ret);
653 key.offset = end_byte - 1;
655 truncate_one_csum(trans, root, path, &key, bytenr, len);
656 if (key.offset < bytenr)
659 btrfs_release_path(path);
663 btrfs_free_path(path);
667 int btrfs_csum_file_blocks(struct btrfs_trans_handle *trans,
668 struct btrfs_root *root,
669 struct btrfs_ordered_sum *sums)
673 struct btrfs_key file_key;
674 struct btrfs_key found_key;
678 struct btrfs_path *path;
679 struct btrfs_csum_item *item;
680 struct btrfs_csum_item *item_end;
681 struct extent_buffer *leaf = NULL;
683 struct btrfs_sector_sum *sector_sum;
686 u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
688 path = btrfs_alloc_path();
692 sector_sum = sums->sums;
693 trans->adding_csums = 1;
695 next_offset = (u64)-1;
697 file_key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
698 file_key.offset = sector_sum->bytenr;
699 bytenr = sector_sum->bytenr;
700 btrfs_set_key_type(&file_key, BTRFS_EXTENT_CSUM_KEY);
702 item = btrfs_lookup_csum(trans, root, path, sector_sum->bytenr, 1);
704 leaf = path->nodes[0];
709 if (ret != -EFBIG && ret != -ENOENT)
714 /* we found one, but it isn't big enough yet */
715 leaf = path->nodes[0];
716 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
717 if ((item_size / csum_size) >=
718 MAX_CSUM_ITEMS(root, csum_size)) {
719 /* already at max size, make a new one */
723 int slot = path->slots[0] + 1;
724 /* we didn't find a csum item, insert one */
725 nritems = btrfs_header_nritems(path->nodes[0]);
726 if (path->slots[0] >= nritems - 1) {
727 ret = btrfs_next_leaf(root, path);
734 btrfs_item_key_to_cpu(path->nodes[0], &found_key, slot);
735 if (found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
736 found_key.type != BTRFS_EXTENT_CSUM_KEY) {
740 next_offset = found_key.offset;
746 * at this point, we know the tree has an item, but it isn't big
747 * enough yet to put our csum in. Grow it
749 btrfs_release_path(path);
750 ret = btrfs_search_slot(trans, root, &file_key, path,
756 if (path->slots[0] == 0)
761 leaf = path->nodes[0];
762 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
763 csum_offset = (bytenr - found_key.offset) >>
764 root->fs_info->sb->s_blocksize_bits;
766 if (btrfs_key_type(&found_key) != BTRFS_EXTENT_CSUM_KEY ||
767 found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
768 csum_offset >= MAX_CSUM_ITEMS(root, csum_size)) {
772 if (csum_offset >= btrfs_item_size_nr(leaf, path->slots[0]) /
774 u32 diff = (csum_offset + 1) * csum_size;
777 * is the item big enough already? we dropped our lock
778 * before and need to recheck
780 if (diff < btrfs_item_size_nr(leaf, path->slots[0]))
783 diff = diff - btrfs_item_size_nr(leaf, path->slots[0]);
784 if (diff != csum_size)
787 btrfs_extend_item(trans, root, path, diff);
792 btrfs_release_path(path);
795 u64 tmp = total_bytes + root->sectorsize;
796 u64 next_sector = sector_sum->bytenr;
797 struct btrfs_sector_sum *next = sector_sum + 1;
799 while (tmp < sums->len) {
800 if (next_sector + root->sectorsize != next->bytenr)
802 tmp += root->sectorsize;
803 next_sector = next->bytenr;
806 tmp = min(tmp, next_offset - file_key.offset);
807 tmp >>= root->fs_info->sb->s_blocksize_bits;
808 tmp = max((u64)1, tmp);
809 tmp = min(tmp, (u64)MAX_CSUM_ITEMS(root, csum_size));
810 ins_size = csum_size * tmp;
812 ins_size = csum_size;
814 path->leave_spinning = 1;
815 ret = btrfs_insert_empty_item(trans, root, path, &file_key,
817 path->leave_spinning = 0;
825 leaf = path->nodes[0];
826 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item);
828 item = (struct btrfs_csum_item *)((unsigned char *)item +
829 csum_offset * csum_size);
831 item_end = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item);
832 item_end = (struct btrfs_csum_item *)((unsigned char *)item_end +
833 btrfs_item_size_nr(leaf, path->slots[0]));
836 write_extent_buffer(leaf, §or_sum->sum, (unsigned long)item, csum_size);
838 total_bytes += root->sectorsize;
840 if (total_bytes < sums->len) {
841 item = (struct btrfs_csum_item *)((char *)item +
843 if (item < item_end && bytenr + PAGE_CACHE_SIZE ==
844 sector_sum->bytenr) {
845 bytenr = sector_sum->bytenr;
850 btrfs_mark_buffer_dirty(path->nodes[0]);
851 if (total_bytes < sums->len) {
852 btrfs_release_path(path);
857 trans->adding_csums = 0;
858 btrfs_free_path(path);