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,
62 BUG_ON(ret); /* Can't happen */
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;
94 u16 csum_size = 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;
164 u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
166 struct btrfs_path *path;
167 struct btrfs_csum_item *item = NULL;
168 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
170 path = btrfs_alloc_path();
173 if (bio->bi_size > PAGE_CACHE_SIZE * 8)
176 WARN_ON(bio->bi_vcnt <= 0);
179 * the free space stuff is only read when it hasn't been
180 * updated in the current transaction. So, we can safely
181 * read from the commit root and sidestep a nasty deadlock
182 * between reading the free space cache and updating the csum tree.
184 if (btrfs_is_free_space_inode(root, inode)) {
185 path->search_commit_root = 1;
186 path->skip_locking = 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;
288 unsigned long offset;
292 u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
294 path = btrfs_alloc_path();
299 path->skip_locking = 1;
301 path->search_commit_root = 1;
304 key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
306 key.type = BTRFS_EXTENT_CSUM_KEY;
308 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
311 if (ret > 0 && path->slots[0] > 0) {
312 leaf = path->nodes[0];
313 btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
314 if (key.objectid == BTRFS_EXTENT_CSUM_OBJECTID &&
315 key.type == BTRFS_EXTENT_CSUM_KEY) {
316 offset = (start - key.offset) >>
317 root->fs_info->sb->s_blocksize_bits;
318 if (offset * csum_size <
319 btrfs_item_size_nr(leaf, path->slots[0] - 1))
324 while (start <= end) {
325 leaf = path->nodes[0];
326 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
327 ret = btrfs_next_leaf(root, path);
332 leaf = path->nodes[0];
335 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
336 if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
337 key.type != BTRFS_EXTENT_CSUM_KEY)
340 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
341 if (key.offset > end)
344 if (key.offset > start)
347 size = btrfs_item_size_nr(leaf, path->slots[0]);
348 csum_end = key.offset + (size / csum_size) * root->sectorsize;
349 if (csum_end <= start) {
354 csum_end = min(csum_end, end + 1);
355 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
356 struct btrfs_csum_item);
357 while (start < csum_end) {
358 size = min_t(size_t, csum_end - start,
359 MAX_ORDERED_SUM_BYTES(root));
360 sums = kzalloc(btrfs_ordered_sum_size(root, size),
367 sector_sum = sums->sums;
368 sums->bytenr = start;
371 offset = (start - key.offset) >>
372 root->fs_info->sb->s_blocksize_bits;
376 read_extent_buffer(path->nodes[0],
378 ((unsigned long)item) +
380 sector_sum->bytenr = start;
382 size -= root->sectorsize;
383 start += root->sectorsize;
387 list_add_tail(&sums->list, &tmplist);
393 while (ret < 0 && !list_empty(&tmplist)) {
394 sums = list_entry(&tmplist, struct btrfs_ordered_sum, list);
395 list_del(&sums->list);
398 list_splice_tail(&tmplist, list);
400 btrfs_free_path(path);
404 int btrfs_csum_one_bio(struct btrfs_root *root, struct inode *inode,
405 struct bio *bio, u64 file_start, int contig)
407 struct btrfs_ordered_sum *sums;
408 struct btrfs_sector_sum *sector_sum;
409 struct btrfs_ordered_extent *ordered;
411 struct bio_vec *bvec = bio->bi_io_vec;
413 unsigned long total_bytes = 0;
414 unsigned long this_sum_bytes = 0;
418 WARN_ON(bio->bi_vcnt <= 0);
419 sums = kzalloc(btrfs_ordered_sum_size(root, bio->bi_size), GFP_NOFS);
423 sector_sum = sums->sums;
424 disk_bytenr = (u64)bio->bi_sector << 9;
425 sums->len = bio->bi_size;
426 INIT_LIST_HEAD(&sums->list);
431 offset = page_offset(bvec->bv_page) + bvec->bv_offset;
433 ordered = btrfs_lookup_ordered_extent(inode, offset);
434 BUG_ON(!ordered); /* Logic error */
435 sums->bytenr = ordered->start;
437 while (bio_index < bio->bi_vcnt) {
439 offset = page_offset(bvec->bv_page) + bvec->bv_offset;
441 if (!contig && (offset >= ordered->file_offset + ordered->len ||
442 offset < ordered->file_offset)) {
443 unsigned long bytes_left;
444 sums->len = this_sum_bytes;
446 btrfs_add_ordered_sum(inode, ordered, sums);
447 btrfs_put_ordered_extent(ordered);
449 bytes_left = bio->bi_size - total_bytes;
451 sums = kzalloc(btrfs_ordered_sum_size(root, bytes_left),
453 BUG_ON(!sums); /* -ENOMEM */
454 sector_sum = sums->sums;
455 sums->len = bytes_left;
456 ordered = btrfs_lookup_ordered_extent(inode, offset);
457 BUG_ON(!ordered); /* Logic error */
458 sums->bytenr = ordered->start;
461 data = kmap_atomic(bvec->bv_page, KM_USER0);
462 sector_sum->sum = ~(u32)0;
463 sector_sum->sum = btrfs_csum_data(root,
464 data + bvec->bv_offset,
467 kunmap_atomic(data, KM_USER0);
468 btrfs_csum_final(sector_sum->sum,
469 (char *)§or_sum->sum);
470 sector_sum->bytenr = disk_bytenr;
474 total_bytes += bvec->bv_len;
475 this_sum_bytes += bvec->bv_len;
476 disk_bytenr += bvec->bv_len;
477 offset += bvec->bv_len;
481 btrfs_add_ordered_sum(inode, ordered, sums);
482 btrfs_put_ordered_extent(ordered);
487 * helper function for csum removal, this expects the
488 * key to describe the csum pointed to by the path, and it expects
489 * the csum to overlap the range [bytenr, len]
491 * The csum should not be entirely contained in the range and the
492 * range should not be entirely contained in the csum.
494 * This calls btrfs_truncate_item with the correct args based on the
495 * overlap, and fixes up the key as required.
497 static noinline void truncate_one_csum(struct btrfs_trans_handle *trans,
498 struct btrfs_root *root,
499 struct btrfs_path *path,
500 struct btrfs_key *key,
503 struct extent_buffer *leaf;
504 u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
506 u64 end_byte = bytenr + len;
507 u32 blocksize_bits = root->fs_info->sb->s_blocksize_bits;
509 leaf = path->nodes[0];
510 csum_end = btrfs_item_size_nr(leaf, path->slots[0]) / csum_size;
511 csum_end <<= root->fs_info->sb->s_blocksize_bits;
512 csum_end += key->offset;
514 if (key->offset < bytenr && csum_end <= end_byte) {
519 * A simple truncate off the end of the item
521 u32 new_size = (bytenr - key->offset) >> blocksize_bits;
522 new_size *= csum_size;
523 btrfs_truncate_item(trans, root, path, new_size, 1);
524 } else if (key->offset >= bytenr && csum_end > end_byte &&
525 end_byte > key->offset) {
530 * we need to truncate from the beginning of the csum
532 u32 new_size = (csum_end - end_byte) >> blocksize_bits;
533 new_size *= csum_size;
535 btrfs_truncate_item(trans, root, path, new_size, 0);
537 key->offset = end_byte;
538 btrfs_set_item_key_safe(trans, root, path, key);
545 * deletes the csum items from the csum tree for a given
548 int btrfs_del_csums(struct btrfs_trans_handle *trans,
549 struct btrfs_root *root, u64 bytenr, u64 len)
551 struct btrfs_path *path;
552 struct btrfs_key key;
553 u64 end_byte = bytenr + len;
555 struct extent_buffer *leaf;
557 u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
558 int blocksize_bits = root->fs_info->sb->s_blocksize_bits;
560 root = root->fs_info->csum_root;
562 path = btrfs_alloc_path();
567 key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
568 key.offset = end_byte - 1;
569 key.type = BTRFS_EXTENT_CSUM_KEY;
571 path->leave_spinning = 1;
572 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
574 if (path->slots[0] == 0)
577 } else if (ret < 0) {
581 leaf = path->nodes[0];
582 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
584 if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
585 key.type != BTRFS_EXTENT_CSUM_KEY) {
589 if (key.offset >= end_byte)
592 csum_end = btrfs_item_size_nr(leaf, path->slots[0]) / csum_size;
593 csum_end <<= blocksize_bits;
594 csum_end += key.offset;
596 /* this csum ends before we start, we're done */
597 if (csum_end <= bytenr)
600 /* delete the entire item, it is inside our range */
601 if (key.offset >= bytenr && csum_end <= end_byte) {
602 ret = btrfs_del_item(trans, root, path);
605 if (key.offset == bytenr)
607 } else if (key.offset < bytenr && csum_end > end_byte) {
608 unsigned long offset;
609 unsigned long shift_len;
610 unsigned long item_offset;
615 * Our bytes are in the middle of the csum,
616 * we need to split this item and insert a new one.
618 * But we can't drop the path because the
619 * csum could change, get removed, extended etc.
621 * The trick here is the max size of a csum item leaves
622 * enough room in the tree block for a single
623 * item header. So, we split the item in place,
624 * adding a new header pointing to the existing
625 * bytes. Then we loop around again and we have
626 * a nicely formed csum item that we can neatly
629 offset = (bytenr - key.offset) >> blocksize_bits;
632 shift_len = (len >> blocksize_bits) * csum_size;
634 item_offset = btrfs_item_ptr_offset(leaf,
637 memset_extent_buffer(leaf, 0, item_offset + offset,
642 * btrfs_split_item returns -EAGAIN when the
643 * item changed size or key
645 ret = btrfs_split_item(trans, root, path, &key, offset);
646 if (ret && ret != -EAGAIN) {
647 btrfs_abort_transaction(trans, root, ret);
651 key.offset = end_byte - 1;
653 truncate_one_csum(trans, root, path, &key, bytenr, len);
654 if (key.offset < bytenr)
657 btrfs_release_path(path);
661 btrfs_free_path(path);
665 int btrfs_csum_file_blocks(struct btrfs_trans_handle *trans,
666 struct btrfs_root *root,
667 struct btrfs_ordered_sum *sums)
671 struct btrfs_key file_key;
672 struct btrfs_key found_key;
676 struct btrfs_path *path;
677 struct btrfs_csum_item *item;
678 struct btrfs_csum_item *item_end;
679 struct extent_buffer *leaf = NULL;
681 struct btrfs_sector_sum *sector_sum;
684 u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
686 path = btrfs_alloc_path();
690 sector_sum = sums->sums;
692 next_offset = (u64)-1;
694 file_key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
695 file_key.offset = sector_sum->bytenr;
696 bytenr = sector_sum->bytenr;
697 btrfs_set_key_type(&file_key, BTRFS_EXTENT_CSUM_KEY);
699 item = btrfs_lookup_csum(trans, root, path, sector_sum->bytenr, 1);
701 leaf = path->nodes[0];
706 if (ret != -EFBIG && ret != -ENOENT)
711 /* we found one, but it isn't big enough yet */
712 leaf = path->nodes[0];
713 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
714 if ((item_size / csum_size) >=
715 MAX_CSUM_ITEMS(root, csum_size)) {
716 /* already at max size, make a new one */
720 int slot = path->slots[0] + 1;
721 /* we didn't find a csum item, insert one */
722 nritems = btrfs_header_nritems(path->nodes[0]);
723 if (path->slots[0] >= nritems - 1) {
724 ret = btrfs_next_leaf(root, path);
731 btrfs_item_key_to_cpu(path->nodes[0], &found_key, slot);
732 if (found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
733 found_key.type != BTRFS_EXTENT_CSUM_KEY) {
737 next_offset = found_key.offset;
743 * at this point, we know the tree has an item, but it isn't big
744 * enough yet to put our csum in. Grow it
746 btrfs_release_path(path);
747 ret = btrfs_search_slot(trans, root, &file_key, path,
753 if (path->slots[0] == 0)
758 leaf = path->nodes[0];
759 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
760 csum_offset = (bytenr - found_key.offset) >>
761 root->fs_info->sb->s_blocksize_bits;
763 if (btrfs_key_type(&found_key) != BTRFS_EXTENT_CSUM_KEY ||
764 found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
765 csum_offset >= MAX_CSUM_ITEMS(root, csum_size)) {
769 if (csum_offset >= btrfs_item_size_nr(leaf, path->slots[0]) /
771 u32 diff = (csum_offset + 1) * csum_size;
774 * is the item big enough already? we dropped our lock
775 * before and need to recheck
777 if (diff < btrfs_item_size_nr(leaf, path->slots[0]))
780 diff = diff - btrfs_item_size_nr(leaf, path->slots[0]);
781 if (diff != csum_size)
784 btrfs_extend_item(trans, root, path, diff);
789 btrfs_release_path(path);
792 u64 tmp = total_bytes + root->sectorsize;
793 u64 next_sector = sector_sum->bytenr;
794 struct btrfs_sector_sum *next = sector_sum + 1;
796 while (tmp < sums->len) {
797 if (next_sector + root->sectorsize != next->bytenr)
799 tmp += root->sectorsize;
800 next_sector = next->bytenr;
803 tmp = min(tmp, next_offset - file_key.offset);
804 tmp >>= root->fs_info->sb->s_blocksize_bits;
805 tmp = max((u64)1, tmp);
806 tmp = min(tmp, (u64)MAX_CSUM_ITEMS(root, csum_size));
807 ins_size = csum_size * tmp;
809 ins_size = csum_size;
811 path->leave_spinning = 1;
812 ret = btrfs_insert_empty_item(trans, root, path, &file_key,
814 path->leave_spinning = 0;
822 leaf = path->nodes[0];
823 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item);
825 item = (struct btrfs_csum_item *)((unsigned char *)item +
826 csum_offset * csum_size);
828 item_end = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item);
829 item_end = (struct btrfs_csum_item *)((unsigned char *)item_end +
830 btrfs_item_size_nr(leaf, path->slots[0]));
833 write_extent_buffer(leaf, §or_sum->sum, (unsigned long)item, csum_size);
835 total_bytes += root->sectorsize;
837 if (total_bytes < sums->len) {
838 item = (struct btrfs_csum_item *)((char *)item +
840 if (item < item_end && bytenr + PAGE_CACHE_SIZE ==
841 sector_sum->bytenr) {
842 bytenr = sector_sum->bytenr;
847 btrfs_mark_buffer_dirty(path->nodes[0]);
848 if (total_bytes < sums->len) {
849 btrfs_release_path(path);
854 btrfs_free_path(path);