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.
18 #include <linux/sched.h>
19 #include <linux/pagemap.h>
20 #include <linux/writeback.h>
25 #include "print-tree.h"
26 #include "transaction.h"
29 #define BLOCK_GROUP_DATA EXTENT_WRITEBACK
30 #define BLOCK_GROUP_METADATA EXTENT_UPTODATE
31 #define BLOCK_GROUP_SYSTEM EXTENT_NEW
33 #define BLOCK_GROUP_DIRTY EXTENT_DIRTY
35 static int finish_current_insert(struct btrfs_trans_handle *trans, struct
36 btrfs_root *extent_root);
37 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
38 btrfs_root *extent_root);
41 static int cache_block_group(struct btrfs_root *root,
42 struct btrfs_block_group_cache *block_group)
44 struct btrfs_path *path;
47 struct extent_buffer *leaf;
48 struct extent_io_tree *free_space_cache;
58 root = root->fs_info->extent_root;
59 free_space_cache = &root->fs_info->free_space_cache;
61 if (block_group->cached)
64 path = btrfs_alloc_path();
69 first_free = block_group->key.objectid;
70 key.objectid = block_group->key.objectid;
72 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
73 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
76 ret = btrfs_previous_item(root, path, 0, BTRFS_EXTENT_ITEM_KEY);
80 leaf = path->nodes[0];
81 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
82 if (key.objectid + key.offset > first_free)
83 first_free = key.objectid + key.offset;
86 leaf = path->nodes[0];
87 slot = path->slots[0];
88 if (slot >= btrfs_header_nritems(leaf)) {
89 ret = btrfs_next_leaf(root, path);
98 btrfs_item_key_to_cpu(leaf, &key, slot);
99 if (key.objectid < block_group->key.objectid) {
102 if (key.objectid >= block_group->key.objectid +
103 block_group->key.offset) {
107 if (btrfs_key_type(&key) == BTRFS_EXTENT_ITEM_KEY) {
112 if (key.objectid > last) {
113 hole_size = key.objectid - last;
114 set_extent_dirty(free_space_cache, last,
115 last + hole_size - 1,
118 last = key.objectid + key.offset;
126 if (block_group->key.objectid +
127 block_group->key.offset > last) {
128 hole_size = block_group->key.objectid +
129 block_group->key.offset - last;
130 set_extent_dirty(free_space_cache, last,
131 last + hole_size - 1, GFP_NOFS);
133 block_group->cached = 1;
135 btrfs_free_path(path);
139 struct btrfs_block_group_cache *btrfs_lookup_block_group(struct
143 struct extent_io_tree *block_group_cache;
144 struct btrfs_block_group_cache *block_group = NULL;
150 bytenr = max_t(u64, bytenr,
151 BTRFS_SUPER_INFO_OFFSET + BTRFS_SUPER_INFO_SIZE);
152 block_group_cache = &info->block_group_cache;
153 ret = find_first_extent_bit(block_group_cache,
154 bytenr, &start, &end,
155 BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA |
160 ret = get_state_private(block_group_cache, start, &ptr);
164 block_group = (struct btrfs_block_group_cache *)(unsigned long)ptr;
165 if (block_group->key.objectid <= bytenr && bytenr <
166 block_group->key.objectid + block_group->key.offset)
171 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
173 return (cache->flags & bits) == bits;
176 static int noinline find_search_start(struct btrfs_root *root,
177 struct btrfs_block_group_cache **cache_ret,
178 u64 *start_ret, int num, int data)
181 struct btrfs_block_group_cache *cache = *cache_ret;
182 struct extent_io_tree *free_space_cache;
183 struct extent_state *state;
188 u64 search_start = *start_ret;
194 total_fs_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
195 free_space_cache = &root->fs_info->free_space_cache;
198 ret = cache_block_group(root, cache);
202 last = max(search_start, cache->key.objectid);
203 if (!block_group_bits(cache, data) || cache->ro) {
207 spin_lock_irq(&free_space_cache->lock);
208 state = find_first_extent_bit_state(free_space_cache, last, EXTENT_DIRTY);
213 spin_unlock_irq(&free_space_cache->lock);
217 start = max(last, state->start);
218 last = state->end + 1;
219 if (last - start < num) {
220 if (last == cache->key.objectid + cache->key.offset)
223 state = extent_state_next(state);
224 } while(state && !(state->state & EXTENT_DIRTY));
227 spin_unlock_irq(&free_space_cache->lock);
230 if (start + num > cache->key.objectid + cache->key.offset)
232 if (start + num > total_fs_bytes)
234 if (!block_group_bits(cache, data)) {
235 printk("block group bits don't match %Lu %d\n", cache->flags, data);
241 cache = btrfs_lookup_block_group(root->fs_info, search_start);
243 printk("Unable to find block group for %Lu\n", search_start);
249 last = cache->key.objectid + cache->key.offset;
251 cache = btrfs_lookup_block_group(root->fs_info, last);
252 if (!cache || cache->key.objectid >= total_fs_bytes) {
261 if (cache_miss && !cache->cached) {
262 cache_block_group(root, cache);
264 cache = btrfs_lookup_block_group(root->fs_info, last);
266 cache = btrfs_find_block_group(root, cache, last, data, 0);
274 static u64 div_factor(u64 num, int factor)
283 static int block_group_state_bits(u64 flags)
286 if (flags & BTRFS_BLOCK_GROUP_DATA)
287 bits |= BLOCK_GROUP_DATA;
288 if (flags & BTRFS_BLOCK_GROUP_METADATA)
289 bits |= BLOCK_GROUP_METADATA;
290 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
291 bits |= BLOCK_GROUP_SYSTEM;
295 struct btrfs_block_group_cache *btrfs_find_block_group(struct btrfs_root *root,
296 struct btrfs_block_group_cache
297 *hint, u64 search_start,
300 struct btrfs_block_group_cache *cache;
301 struct extent_io_tree *block_group_cache;
302 struct btrfs_block_group_cache *found_group = NULL;
303 struct btrfs_fs_info *info = root->fs_info;
317 block_group_cache = &info->block_group_cache;
318 total_fs_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
320 if (data & BTRFS_BLOCK_GROUP_METADATA)
323 bit = block_group_state_bits(data);
325 if (search_start && search_start < total_fs_bytes) {
326 struct btrfs_block_group_cache *shint;
327 shint = btrfs_lookup_block_group(info, search_start);
328 if (shint && block_group_bits(shint, data) && !shint->ro) {
329 used = btrfs_block_group_used(&shint->item);
330 if (used + shint->pinned <
331 div_factor(shint->key.offset, factor)) {
336 if (hint && !hint->ro && block_group_bits(hint, data) &&
337 hint->key.objectid < total_fs_bytes) {
338 used = btrfs_block_group_used(&hint->item);
339 if (used + hint->pinned <
340 div_factor(hint->key.offset, factor)) {
343 last = hint->key.objectid + hint->key.offset;
347 hint_last = max(hint->key.objectid, search_start);
349 hint_last = search_start;
351 if (hint_last >= total_fs_bytes)
352 hint_last = search_start;
357 ret = find_first_extent_bit(block_group_cache, last,
362 ret = get_state_private(block_group_cache, start, &ptr);
366 cache = (struct btrfs_block_group_cache *)(unsigned long)ptr;
367 last = cache->key.objectid + cache->key.offset;
368 used = btrfs_block_group_used(&cache->item);
370 if (cache->key.objectid > total_fs_bytes)
373 if (!cache->ro && block_group_bits(cache, data)) {
375 free_check = cache->key.offset;
377 free_check = div_factor(cache->key.offset,
380 if (used + cache->pinned < free_check) {
396 static u64 hash_extent_ref(u64 root_objectid, u64 ref_generation,
397 u64 owner, u64 owner_offset)
399 u32 high_crc = ~(u32)0;
400 u32 low_crc = ~(u32)0;
402 lenum = cpu_to_le64(root_objectid);
403 high_crc = btrfs_crc32c(high_crc, &lenum, sizeof(lenum));
404 lenum = cpu_to_le64(ref_generation);
405 low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
406 if (owner >= BTRFS_FIRST_FREE_OBJECTID) {
407 lenum = cpu_to_le64(owner);
408 low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
409 lenum = cpu_to_le64(owner_offset);
410 low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
412 return ((u64)high_crc << 32) | (u64)low_crc;
415 static int match_extent_ref(struct extent_buffer *leaf,
416 struct btrfs_extent_ref *disk_ref,
417 struct btrfs_extent_ref *cpu_ref)
422 if (cpu_ref->objectid)
423 len = sizeof(*cpu_ref);
425 len = 2 * sizeof(u64);
426 ret = memcmp_extent_buffer(leaf, cpu_ref, (unsigned long)disk_ref,
431 static int noinline lookup_extent_backref(struct btrfs_trans_handle *trans,
432 struct btrfs_root *root,
433 struct btrfs_path *path, u64 bytenr,
435 u64 ref_generation, u64 owner,
436 u64 owner_offset, int del)
439 struct btrfs_key key;
440 struct btrfs_key found_key;
441 struct btrfs_extent_ref ref;
442 struct extent_buffer *leaf;
443 struct btrfs_extent_ref *disk_ref;
447 btrfs_set_stack_ref_root(&ref, root_objectid);
448 btrfs_set_stack_ref_generation(&ref, ref_generation);
449 btrfs_set_stack_ref_objectid(&ref, owner);
450 btrfs_set_stack_ref_offset(&ref, owner_offset);
452 hash = hash_extent_ref(root_objectid, ref_generation, owner,
455 key.objectid = bytenr;
456 key.type = BTRFS_EXTENT_REF_KEY;
459 ret = btrfs_search_slot(trans, root, &key, path,
463 leaf = path->nodes[0];
465 u32 nritems = btrfs_header_nritems(leaf);
466 if (path->slots[0] >= nritems) {
467 ret2 = btrfs_next_leaf(root, path);
470 leaf = path->nodes[0];
472 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
473 if (found_key.objectid != bytenr ||
474 found_key.type != BTRFS_EXTENT_REF_KEY)
476 key.offset = found_key.offset;
478 btrfs_release_path(root, path);
482 disk_ref = btrfs_item_ptr(path->nodes[0],
484 struct btrfs_extent_ref);
485 if (match_extent_ref(path->nodes[0], disk_ref, &ref)) {
489 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
490 key.offset = found_key.offset + 1;
491 btrfs_release_path(root, path);
498 * Back reference rules. Back refs have three main goals:
500 * 1) differentiate between all holders of references to an extent so that
501 * when a reference is dropped we can make sure it was a valid reference
502 * before freeing the extent.
504 * 2) Provide enough information to quickly find the holders of an extent
505 * if we notice a given block is corrupted or bad.
507 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
508 * maintenance. This is actually the same as #2, but with a slightly
509 * different use case.
511 * File extents can be referenced by:
513 * - multiple snapshots, subvolumes, or different generations in one subvol
514 * - different files inside a single subvolume (in theory, not implemented yet)
515 * - different offsets inside a file (bookend extents in file.c)
517 * The extent ref structure has fields for:
519 * - Objectid of the subvolume root
520 * - Generation number of the tree holding the reference
521 * - objectid of the file holding the reference
522 * - offset in the file corresponding to the key holding the reference
524 * When a file extent is allocated the fields are filled in:
525 * (root_key.objectid, trans->transid, inode objectid, offset in file)
527 * When a leaf is cow'd new references are added for every file extent found
528 * in the leaf. It looks the same as the create case, but trans->transid
529 * will be different when the block is cow'd.
531 * (root_key.objectid, trans->transid, inode objectid, offset in file)
533 * When a file extent is removed either during snapshot deletion or file
534 * truncation, the corresponding back reference is found
537 * (btrfs_header_owner(leaf), btrfs_header_generation(leaf),
538 * inode objectid, offset in file)
540 * Btree extents can be referenced by:
542 * - Different subvolumes
543 * - Different generations of the same subvolume
545 * Storing sufficient information for a full reverse mapping of a btree
546 * block would require storing the lowest key of the block in the backref,
547 * and it would require updating that lowest key either before write out or
548 * every time it changed. Instead, the objectid of the lowest key is stored
549 * along with the level of the tree block. This provides a hint
550 * about where in the btree the block can be found. Searches through the
551 * btree only need to look for a pointer to that block, so they stop one
552 * level higher than the level recorded in the backref.
554 * Some btrees do not do reference counting on their extents. These
555 * include the extent tree and the tree of tree roots. Backrefs for these
556 * trees always have a generation of zero.
558 * When a tree block is created, back references are inserted:
560 * (root->root_key.objectid, trans->transid or zero, level, lowest_key_objectid)
562 * When a tree block is cow'd in a reference counted root,
563 * new back references are added for all the blocks it points to.
564 * These are of the form (trans->transid will have increased since creation):
566 * (root->root_key.objectid, trans->transid, level, lowest_key_objectid)
568 * Because the lowest_key_objectid and the level are just hints
569 * they are not used when backrefs are deleted. When a backref is deleted:
571 * if backref was for a tree root:
572 * root_objectid = root->root_key.objectid
574 * root_objectid = btrfs_header_owner(parent)
576 * (root_objectid, btrfs_header_generation(parent) or zero, 0, 0)
578 * Back Reference Key hashing:
580 * Back references have four fields, each 64 bits long. Unfortunately,
581 * This is hashed into a single 64 bit number and placed into the key offset.
582 * The key objectid corresponds to the first byte in the extent, and the
583 * key type is set to BTRFS_EXTENT_REF_KEY
585 int btrfs_insert_extent_backref(struct btrfs_trans_handle *trans,
586 struct btrfs_root *root,
587 struct btrfs_path *path, u64 bytenr,
588 u64 root_objectid, u64 ref_generation,
589 u64 owner, u64 owner_offset)
592 struct btrfs_key key;
593 struct btrfs_extent_ref ref;
594 struct btrfs_extent_ref *disk_ref;
597 btrfs_set_stack_ref_root(&ref, root_objectid);
598 btrfs_set_stack_ref_generation(&ref, ref_generation);
599 btrfs_set_stack_ref_objectid(&ref, owner);
600 btrfs_set_stack_ref_offset(&ref, owner_offset);
602 hash = hash_extent_ref(root_objectid, ref_generation, owner,
605 key.objectid = bytenr;
606 key.type = BTRFS_EXTENT_REF_KEY;
608 ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(ref));
609 while (ret == -EEXIST) {
610 disk_ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
611 struct btrfs_extent_ref);
612 if (match_extent_ref(path->nodes[0], disk_ref, &ref))
615 btrfs_release_path(root, path);
616 ret = btrfs_insert_empty_item(trans, root, path, &key,
621 disk_ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
622 struct btrfs_extent_ref);
623 write_extent_buffer(path->nodes[0], &ref, (unsigned long)disk_ref,
625 btrfs_mark_buffer_dirty(path->nodes[0]);
627 btrfs_release_path(root, path);
631 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
632 struct btrfs_root *root,
633 u64 bytenr, u64 num_bytes,
634 u64 root_objectid, u64 ref_generation,
635 u64 owner, u64 owner_offset)
637 struct btrfs_path *path;
639 struct btrfs_key key;
640 struct extent_buffer *l;
641 struct btrfs_extent_item *item;
644 WARN_ON(num_bytes < root->sectorsize);
645 path = btrfs_alloc_path();
650 key.objectid = bytenr;
651 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
652 key.offset = num_bytes;
653 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
662 item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
663 refs = btrfs_extent_refs(l, item);
664 btrfs_set_extent_refs(l, item, refs + 1);
665 btrfs_mark_buffer_dirty(path->nodes[0]);
667 btrfs_release_path(root->fs_info->extent_root, path);
670 ret = btrfs_insert_extent_backref(trans, root->fs_info->extent_root,
671 path, bytenr, root_objectid,
672 ref_generation, owner, owner_offset);
674 finish_current_insert(trans, root->fs_info->extent_root);
675 del_pending_extents(trans, root->fs_info->extent_root);
677 btrfs_free_path(path);
681 int btrfs_extent_post_op(struct btrfs_trans_handle *trans,
682 struct btrfs_root *root)
684 finish_current_insert(trans, root->fs_info->extent_root);
685 del_pending_extents(trans, root->fs_info->extent_root);
689 static int lookup_extent_ref(struct btrfs_trans_handle *trans,
690 struct btrfs_root *root, u64 bytenr,
691 u64 num_bytes, u32 *refs)
693 struct btrfs_path *path;
695 struct btrfs_key key;
696 struct extent_buffer *l;
697 struct btrfs_extent_item *item;
699 WARN_ON(num_bytes < root->sectorsize);
700 path = btrfs_alloc_path();
702 key.objectid = bytenr;
703 key.offset = num_bytes;
704 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
705 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
710 btrfs_print_leaf(root, path->nodes[0]);
711 printk("failed to find block number %Lu\n", bytenr);
715 item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
716 *refs = btrfs_extent_refs(l, item);
718 btrfs_free_path(path);
722 u32 btrfs_count_snapshots_in_path(struct btrfs_root *root,
723 struct btrfs_path *count_path,
726 struct btrfs_root *extent_root = root->fs_info->extent_root;
727 struct btrfs_path *path;
730 u64 root_objectid = root->root_key.objectid;
735 struct btrfs_key key;
736 struct btrfs_key found_key;
737 struct extent_buffer *l;
738 struct btrfs_extent_item *item;
739 struct btrfs_extent_ref *ref_item;
742 path = btrfs_alloc_path();
745 bytenr = first_extent;
747 bytenr = count_path->nodes[level]->start;
750 key.objectid = bytenr;
753 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
754 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
760 btrfs_item_key_to_cpu(l, &found_key, path->slots[0]);
762 if (found_key.objectid != bytenr ||
763 found_key.type != BTRFS_EXTENT_ITEM_KEY) {
767 item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
770 nritems = btrfs_header_nritems(l);
771 if (path->slots[0] >= nritems) {
772 ret = btrfs_next_leaf(extent_root, path);
777 btrfs_item_key_to_cpu(l, &found_key, path->slots[0]);
778 if (found_key.objectid != bytenr)
781 if (found_key.type != BTRFS_EXTENT_REF_KEY) {
787 ref_item = btrfs_item_ptr(l, path->slots[0],
788 struct btrfs_extent_ref);
789 found_objectid = btrfs_ref_root(l, ref_item);
791 if (found_objectid != root_objectid) {
798 if (cur_count == 0) {
802 if (level >= 0 && root->node == count_path->nodes[level])
805 btrfs_release_path(root, path);
809 btrfs_free_path(path);
812 int btrfs_inc_root_ref(struct btrfs_trans_handle *trans,
813 struct btrfs_root *root, u64 owner_objectid)
819 struct btrfs_disk_key disk_key;
821 level = btrfs_header_level(root->node);
822 generation = trans->transid;
823 nritems = btrfs_header_nritems(root->node);
826 btrfs_item_key(root->node, &disk_key, 0);
828 btrfs_node_key(root->node, &disk_key, 0);
829 key_objectid = btrfs_disk_key_objectid(&disk_key);
833 return btrfs_inc_extent_ref(trans, root, root->node->start,
834 root->node->len, owner_objectid,
835 generation, level, key_objectid);
838 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
839 struct extent_buffer *buf)
843 struct btrfs_key key;
844 struct btrfs_file_extent_item *fi;
853 level = btrfs_header_level(buf);
854 nritems = btrfs_header_nritems(buf);
855 for (i = 0; i < nritems; i++) {
858 btrfs_item_key_to_cpu(buf, &key, i);
859 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
861 fi = btrfs_item_ptr(buf, i,
862 struct btrfs_file_extent_item);
863 if (btrfs_file_extent_type(buf, fi) ==
864 BTRFS_FILE_EXTENT_INLINE)
866 disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
867 if (disk_bytenr == 0)
869 ret = btrfs_inc_extent_ref(trans, root, disk_bytenr,
870 btrfs_file_extent_disk_num_bytes(buf, fi),
871 root->root_key.objectid, trans->transid,
872 key.objectid, key.offset);
878 bytenr = btrfs_node_blockptr(buf, i);
879 btrfs_node_key_to_cpu(buf, &key, i);
880 ret = btrfs_inc_extent_ref(trans, root, bytenr,
881 btrfs_level_size(root, level - 1),
882 root->root_key.objectid,
884 level - 1, key.objectid);
895 for (i =0; i < faili; i++) {
898 btrfs_item_key_to_cpu(buf, &key, i);
899 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
901 fi = btrfs_item_ptr(buf, i,
902 struct btrfs_file_extent_item);
903 if (btrfs_file_extent_type(buf, fi) ==
904 BTRFS_FILE_EXTENT_INLINE)
906 disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
907 if (disk_bytenr == 0)
909 err = btrfs_free_extent(trans, root, disk_bytenr,
910 btrfs_file_extent_disk_num_bytes(buf,
914 bytenr = btrfs_node_blockptr(buf, i);
915 err = btrfs_free_extent(trans, root, bytenr,
916 btrfs_level_size(root, level - 1), 0);
924 static int write_one_cache_group(struct btrfs_trans_handle *trans,
925 struct btrfs_root *root,
926 struct btrfs_path *path,
927 struct btrfs_block_group_cache *cache)
931 struct btrfs_root *extent_root = root->fs_info->extent_root;
933 struct extent_buffer *leaf;
935 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
940 leaf = path->nodes[0];
941 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
942 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
943 btrfs_mark_buffer_dirty(leaf);
944 btrfs_release_path(extent_root, path);
946 finish_current_insert(trans, extent_root);
947 pending_ret = del_pending_extents(trans, extent_root);
956 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
957 struct btrfs_root *root)
959 struct extent_io_tree *block_group_cache;
960 struct btrfs_block_group_cache *cache;
964 struct btrfs_path *path;
970 block_group_cache = &root->fs_info->block_group_cache;
971 path = btrfs_alloc_path();
976 ret = find_first_extent_bit(block_group_cache, last,
977 &start, &end, BLOCK_GROUP_DIRTY);
982 ret = get_state_private(block_group_cache, start, &ptr);
985 cache = (struct btrfs_block_group_cache *)(unsigned long)ptr;
986 err = write_one_cache_group(trans, root,
989 * if we fail to write the cache group, we want
990 * to keep it marked dirty in hopes that a later
997 clear_extent_bits(block_group_cache, start, end,
998 BLOCK_GROUP_DIRTY, GFP_NOFS);
1000 btrfs_free_path(path);
1004 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
1007 struct list_head *head = &info->space_info;
1008 struct list_head *cur;
1009 struct btrfs_space_info *found;
1010 list_for_each(cur, head) {
1011 found = list_entry(cur, struct btrfs_space_info, list);
1012 if (found->flags == flags)
1019 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
1020 u64 total_bytes, u64 bytes_used,
1021 struct btrfs_space_info **space_info)
1023 struct btrfs_space_info *found;
1025 found = __find_space_info(info, flags);
1027 found->total_bytes += total_bytes;
1028 found->bytes_used += bytes_used;
1030 WARN_ON(found->total_bytes < found->bytes_used);
1031 *space_info = found;
1034 found = kmalloc(sizeof(*found), GFP_NOFS);
1038 list_add(&found->list, &info->space_info);
1039 found->flags = flags;
1040 found->total_bytes = total_bytes;
1041 found->bytes_used = bytes_used;
1042 found->bytes_pinned = 0;
1044 *space_info = found;
1048 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
1050 u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
1051 BTRFS_BLOCK_GROUP_RAID1 |
1052 BTRFS_BLOCK_GROUP_RAID10 |
1053 BTRFS_BLOCK_GROUP_DUP);
1055 if (flags & BTRFS_BLOCK_GROUP_DATA)
1056 fs_info->avail_data_alloc_bits |= extra_flags;
1057 if (flags & BTRFS_BLOCK_GROUP_METADATA)
1058 fs_info->avail_metadata_alloc_bits |= extra_flags;
1059 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
1060 fs_info->avail_system_alloc_bits |= extra_flags;
1064 static u64 reduce_alloc_profile(struct btrfs_root *root, u64 flags)
1066 u64 num_devices = root->fs_info->fs_devices->num_devices;
1068 if (num_devices == 1)
1069 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
1070 if (num_devices < 4)
1071 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
1073 if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
1074 (flags & (BTRFS_BLOCK_GROUP_RAID1 |
1075 BTRFS_BLOCK_GROUP_RAID10))) {
1076 flags &= ~BTRFS_BLOCK_GROUP_DUP;
1079 if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
1080 (flags & BTRFS_BLOCK_GROUP_RAID10)) {
1081 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
1084 if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
1085 ((flags & BTRFS_BLOCK_GROUP_RAID1) |
1086 (flags & BTRFS_BLOCK_GROUP_RAID10) |
1087 (flags & BTRFS_BLOCK_GROUP_DUP)))
1088 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
1092 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
1093 struct btrfs_root *extent_root, u64 alloc_bytes,
1096 struct btrfs_space_info *space_info;
1102 flags = reduce_alloc_profile(extent_root, flags);
1104 space_info = __find_space_info(extent_root->fs_info, flags);
1106 ret = update_space_info(extent_root->fs_info, flags,
1110 BUG_ON(!space_info);
1112 if (space_info->full)
1115 thresh = div_factor(space_info->total_bytes, 6);
1116 if ((space_info->bytes_used + space_info->bytes_pinned + alloc_bytes) <
1120 ret = btrfs_alloc_chunk(trans, extent_root, &start, &num_bytes, flags);
1121 if (ret == -ENOSPC) {
1122 printk("space info full %Lu\n", flags);
1123 space_info->full = 1;
1129 ret = btrfs_make_block_group(trans, extent_root, 0, flags,
1130 BTRFS_FIRST_CHUNK_TREE_OBJECTID, start, num_bytes);
1136 static int update_block_group(struct btrfs_trans_handle *trans,
1137 struct btrfs_root *root,
1138 u64 bytenr, u64 num_bytes, int alloc,
1141 struct btrfs_block_group_cache *cache;
1142 struct btrfs_fs_info *info = root->fs_info;
1143 u64 total = num_bytes;
1150 cache = btrfs_lookup_block_group(info, bytenr);
1154 byte_in_group = bytenr - cache->key.objectid;
1155 WARN_ON(byte_in_group > cache->key.offset);
1156 start = cache->key.objectid;
1157 end = start + cache->key.offset - 1;
1158 set_extent_bits(&info->block_group_cache, start, end,
1159 BLOCK_GROUP_DIRTY, GFP_NOFS);
1161 old_val = btrfs_block_group_used(&cache->item);
1162 num_bytes = min(total, cache->key.offset - byte_in_group);
1164 old_val += num_bytes;
1165 cache->space_info->bytes_used += num_bytes;
1167 old_val -= num_bytes;
1168 cache->space_info->bytes_used -= num_bytes;
1170 set_extent_dirty(&info->free_space_cache,
1171 bytenr, bytenr + num_bytes - 1,
1175 btrfs_set_block_group_used(&cache->item, old_val);
1177 bytenr += num_bytes;
1182 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
1187 ret = find_first_extent_bit(&root->fs_info->block_group_cache,
1188 search_start, &start, &end,
1189 BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA |
1190 BLOCK_GROUP_SYSTEM);
1197 static int update_pinned_extents(struct btrfs_root *root,
1198 u64 bytenr, u64 num, int pin)
1201 struct btrfs_block_group_cache *cache;
1202 struct btrfs_fs_info *fs_info = root->fs_info;
1205 set_extent_dirty(&fs_info->pinned_extents,
1206 bytenr, bytenr + num - 1, GFP_NOFS);
1208 clear_extent_dirty(&fs_info->pinned_extents,
1209 bytenr, bytenr + num - 1, GFP_NOFS);
1212 cache = btrfs_lookup_block_group(fs_info, bytenr);
1214 u64 first = first_logical_byte(root, bytenr);
1215 WARN_ON(first < bytenr);
1216 len = min(first - bytenr, num);
1218 len = min(num, cache->key.offset -
1219 (bytenr - cache->key.objectid));
1223 cache->pinned += len;
1224 cache->space_info->bytes_pinned += len;
1226 fs_info->total_pinned += len;
1229 cache->pinned -= len;
1230 cache->space_info->bytes_pinned -= len;
1232 fs_info->total_pinned -= len;
1240 int btrfs_copy_pinned(struct btrfs_root *root, struct extent_io_tree *copy)
1245 struct extent_io_tree *pinned_extents = &root->fs_info->pinned_extents;
1249 ret = find_first_extent_bit(pinned_extents, last,
1250 &start, &end, EXTENT_DIRTY);
1253 set_extent_dirty(copy, start, end, GFP_NOFS);
1259 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
1260 struct btrfs_root *root,
1261 struct extent_io_tree *unpin)
1266 struct extent_io_tree *free_space_cache;
1267 free_space_cache = &root->fs_info->free_space_cache;
1270 ret = find_first_extent_bit(unpin, 0, &start, &end,
1274 update_pinned_extents(root, start, end + 1 - start, 0);
1275 clear_extent_dirty(unpin, start, end, GFP_NOFS);
1276 set_extent_dirty(free_space_cache, start, end, GFP_NOFS);
1281 static int finish_current_insert(struct btrfs_trans_handle *trans,
1282 struct btrfs_root *extent_root)
1286 struct btrfs_fs_info *info = extent_root->fs_info;
1287 struct extent_buffer *eb;
1288 struct btrfs_path *path;
1289 struct btrfs_key ins;
1290 struct btrfs_disk_key first;
1291 struct btrfs_extent_item extent_item;
1296 btrfs_set_stack_extent_refs(&extent_item, 1);
1297 btrfs_set_key_type(&ins, BTRFS_EXTENT_ITEM_KEY);
1298 path = btrfs_alloc_path();
1301 ret = find_first_extent_bit(&info->extent_ins, 0, &start,
1302 &end, EXTENT_LOCKED);
1306 ins.objectid = start;
1307 ins.offset = end + 1 - start;
1308 err = btrfs_insert_item(trans, extent_root, &ins,
1309 &extent_item, sizeof(extent_item));
1310 clear_extent_bits(&info->extent_ins, start, end, EXTENT_LOCKED,
1312 eb = read_tree_block(extent_root, ins.objectid, ins.offset);
1313 level = btrfs_header_level(eb);
1315 btrfs_item_key(eb, &first, 0);
1317 btrfs_node_key(eb, &first, 0);
1319 err = btrfs_insert_extent_backref(trans, extent_root, path,
1320 start, extent_root->root_key.objectid,
1322 btrfs_disk_key_objectid(&first));
1324 free_extent_buffer(eb);
1326 btrfs_free_path(path);
1330 static int pin_down_bytes(struct btrfs_root *root, u64 bytenr, u32 num_bytes,
1334 struct extent_buffer *buf;
1337 buf = btrfs_find_tree_block(root, bytenr, num_bytes);
1339 if (btrfs_buffer_uptodate(buf)) {
1341 root->fs_info->running_transaction->transid;
1342 u64 header_transid =
1343 btrfs_header_generation(buf);
1344 if (header_transid == transid &&
1345 !btrfs_header_flag(buf,
1346 BTRFS_HEADER_FLAG_WRITTEN)) {
1347 clean_tree_block(NULL, root, buf);
1348 free_extent_buffer(buf);
1352 free_extent_buffer(buf);
1354 update_pinned_extents(root, bytenr, num_bytes, 1);
1356 set_extent_bits(&root->fs_info->pending_del,
1357 bytenr, bytenr + num_bytes - 1,
1358 EXTENT_LOCKED, GFP_NOFS);
1365 * remove an extent from the root, returns 0 on success
1367 static int __free_extent(struct btrfs_trans_handle *trans, struct btrfs_root
1368 *root, u64 bytenr, u64 num_bytes,
1369 u64 root_objectid, u64 ref_generation,
1370 u64 owner_objectid, u64 owner_offset, int pin,
1373 struct btrfs_path *path;
1374 struct btrfs_key key;
1375 struct btrfs_fs_info *info = root->fs_info;
1376 struct btrfs_root *extent_root = info->extent_root;
1377 struct extent_buffer *leaf;
1379 int extent_slot = 0;
1380 int found_extent = 0;
1382 struct btrfs_extent_item *ei;
1385 key.objectid = bytenr;
1386 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
1387 key.offset = num_bytes;
1388 path = btrfs_alloc_path();
1393 ret = lookup_extent_backref(trans, extent_root, path,
1394 bytenr, root_objectid,
1396 owner_objectid, owner_offset, 1);
1398 struct btrfs_key found_key;
1399 extent_slot = path->slots[0];
1400 while(extent_slot > 0) {
1402 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1404 if (found_key.objectid != bytenr)
1406 if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
1407 found_key.offset == num_bytes) {
1411 if (path->slots[0] - extent_slot > 5)
1415 ret = btrfs_del_item(trans, extent_root, path);
1417 btrfs_print_leaf(extent_root, path->nodes[0]);
1419 printk("Unable to find ref byte nr %Lu root %Lu "
1420 " gen %Lu owner %Lu offset %Lu\n", bytenr,
1421 root_objectid, ref_generation, owner_objectid,
1424 if (!found_extent) {
1425 btrfs_release_path(extent_root, path);
1426 ret = btrfs_search_slot(trans, extent_root, &key, path, -1, 1);
1430 extent_slot = path->slots[0];
1433 leaf = path->nodes[0];
1434 ei = btrfs_item_ptr(leaf, extent_slot,
1435 struct btrfs_extent_item);
1436 refs = btrfs_extent_refs(leaf, ei);
1439 btrfs_set_extent_refs(leaf, ei, refs);
1441 btrfs_mark_buffer_dirty(leaf);
1443 if (refs == 0 && found_extent && path->slots[0] == extent_slot + 1) {
1444 /* if the back ref and the extent are next to each other
1445 * they get deleted below in one shot
1447 path->slots[0] = extent_slot;
1449 } else if (found_extent) {
1450 /* otherwise delete the extent back ref */
1451 ret = btrfs_del_item(trans, extent_root, path);
1453 /* if refs are 0, we need to setup the path for deletion */
1455 btrfs_release_path(extent_root, path);
1456 ret = btrfs_search_slot(trans, extent_root, &key, path,
1469 ret = pin_down_bytes(root, bytenr, num_bytes, 0);
1475 /* block accounting for super block */
1476 super_used = btrfs_super_bytes_used(&info->super_copy);
1477 btrfs_set_super_bytes_used(&info->super_copy,
1478 super_used - num_bytes);
1480 /* block accounting for root item */
1481 root_used = btrfs_root_used(&root->root_item);
1482 btrfs_set_root_used(&root->root_item,
1483 root_used - num_bytes);
1484 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
1489 ret = update_block_group(trans, root, bytenr, num_bytes, 0,
1493 btrfs_free_path(path);
1494 finish_current_insert(trans, extent_root);
1499 * find all the blocks marked as pending in the radix tree and remove
1500 * them from the extent map
1502 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
1503 btrfs_root *extent_root)
1509 struct extent_io_tree *pending_del;
1510 struct extent_io_tree *pinned_extents;
1512 pending_del = &extent_root->fs_info->pending_del;
1513 pinned_extents = &extent_root->fs_info->pinned_extents;
1516 ret = find_first_extent_bit(pending_del, 0, &start, &end,
1520 update_pinned_extents(extent_root, start, end + 1 - start, 1);
1521 clear_extent_bits(pending_del, start, end, EXTENT_LOCKED,
1523 ret = __free_extent(trans, extent_root,
1524 start, end + 1 - start,
1525 extent_root->root_key.objectid,
1534 * remove an extent from the root, returns 0 on success
1536 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root
1537 *root, u64 bytenr, u64 num_bytes,
1538 u64 root_objectid, u64 ref_generation,
1539 u64 owner_objectid, u64 owner_offset, int pin)
1541 struct btrfs_root *extent_root = root->fs_info->extent_root;
1545 WARN_ON(num_bytes < root->sectorsize);
1546 if (!root->ref_cows)
1549 if (root == extent_root) {
1550 pin_down_bytes(root, bytenr, num_bytes, 1);
1553 ret = __free_extent(trans, root, bytenr, num_bytes, root_objectid,
1554 ref_generation, owner_objectid, owner_offset,
1556 pending_ret = del_pending_extents(trans, root->fs_info->extent_root);
1557 return ret ? ret : pending_ret;
1560 static u64 stripe_align(struct btrfs_root *root, u64 val)
1562 u64 mask = ((u64)root->stripesize - 1);
1563 u64 ret = (val + mask) & ~mask;
1568 * walks the btree of allocated extents and find a hole of a given size.
1569 * The key ins is changed to record the hole:
1570 * ins->objectid == block start
1571 * ins->flags = BTRFS_EXTENT_ITEM_KEY
1572 * ins->offset == number of blocks
1573 * Any available blocks before search_start are skipped.
1575 static int noinline find_free_extent(struct btrfs_trans_handle *trans,
1576 struct btrfs_root *orig_root,
1577 u64 num_bytes, u64 empty_size,
1578 u64 search_start, u64 search_end,
1579 u64 hint_byte, struct btrfs_key *ins,
1580 u64 exclude_start, u64 exclude_nr,
1584 u64 orig_search_start;
1585 struct btrfs_root * root = orig_root->fs_info->extent_root;
1586 struct btrfs_fs_info *info = root->fs_info;
1587 u64 total_needed = num_bytes;
1588 u64 *last_ptr = NULL;
1589 struct btrfs_block_group_cache *block_group;
1592 int empty_cluster = 2 * 1024 * 1024;
1594 WARN_ON(num_bytes < root->sectorsize);
1595 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
1597 if (data & BTRFS_BLOCK_GROUP_METADATA) {
1598 last_ptr = &root->fs_info->last_alloc;
1599 empty_cluster = 256 * 1024;
1602 if ((data & BTRFS_BLOCK_GROUP_DATA) && btrfs_test_opt(root, SSD)) {
1603 last_ptr = &root->fs_info->last_data_alloc;
1608 hint_byte = *last_ptr;
1610 empty_size += empty_cluster;
1614 search_start = max(search_start, first_logical_byte(root, 0));
1615 orig_search_start = search_start;
1617 if (search_end == (u64)-1)
1618 search_end = btrfs_super_total_bytes(&info->super_copy);
1621 block_group = btrfs_lookup_block_group(info, hint_byte);
1623 hint_byte = search_start;
1624 block_group = btrfs_find_block_group(root, block_group,
1625 hint_byte, data, 1);
1626 if (last_ptr && *last_ptr == 0 && block_group)
1627 hint_byte = block_group->key.objectid;
1629 block_group = btrfs_find_block_group(root,
1631 search_start, data, 1);
1633 search_start = max(search_start, hint_byte);
1635 total_needed += empty_size;
1639 block_group = btrfs_lookup_block_group(info, search_start);
1641 block_group = btrfs_lookup_block_group(info,
1644 ret = find_search_start(root, &block_group, &search_start,
1645 total_needed, data);
1646 if (ret == -ENOSPC && last_ptr && *last_ptr) {
1648 block_group = btrfs_lookup_block_group(info,
1650 search_start = orig_search_start;
1651 ret = find_search_start(root, &block_group, &search_start,
1652 total_needed, data);
1659 if (last_ptr && *last_ptr && search_start != *last_ptr) {
1662 empty_size += empty_cluster;
1663 total_needed += empty_size;
1665 block_group = btrfs_lookup_block_group(info,
1667 search_start = orig_search_start;
1668 ret = find_search_start(root, &block_group,
1669 &search_start, total_needed, data);
1676 search_start = stripe_align(root, search_start);
1677 ins->objectid = search_start;
1678 ins->offset = num_bytes;
1680 if (ins->objectid + num_bytes >= search_end)
1683 if (ins->objectid + num_bytes >
1684 block_group->key.objectid + block_group->key.offset) {
1685 search_start = block_group->key.objectid +
1686 block_group->key.offset;
1690 if (test_range_bit(&info->extent_ins, ins->objectid,
1691 ins->objectid + num_bytes -1, EXTENT_LOCKED, 0)) {
1692 search_start = ins->objectid + num_bytes;
1696 if (test_range_bit(&info->pinned_extents, ins->objectid,
1697 ins->objectid + num_bytes -1, EXTENT_DIRTY, 0)) {
1698 search_start = ins->objectid + num_bytes;
1702 if (exclude_nr > 0 && (ins->objectid + num_bytes > exclude_start &&
1703 ins->objectid < exclude_start + exclude_nr)) {
1704 search_start = exclude_start + exclude_nr;
1708 if (!(data & BTRFS_BLOCK_GROUP_DATA)) {
1709 block_group = btrfs_lookup_block_group(info, ins->objectid);
1711 trans->block_group = block_group;
1713 ins->offset = num_bytes;
1715 *last_ptr = ins->objectid + ins->offset;
1717 btrfs_super_total_bytes(&root->fs_info->super_copy)) {
1724 if (search_start + num_bytes >= search_end) {
1726 search_start = orig_search_start;
1733 total_needed -= empty_size;
1738 block_group = btrfs_lookup_block_group(info, search_start);
1740 block_group = btrfs_find_block_group(root, block_group,
1741 search_start, data, 0);
1749 * finds a free extent and does all the dirty work required for allocation
1750 * returns the key for the extent through ins, and a tree buffer for
1751 * the first block of the extent through buf.
1753 * returns 0 if everything worked, non-zero otherwise.
1755 int btrfs_alloc_extent(struct btrfs_trans_handle *trans,
1756 struct btrfs_root *root,
1757 u64 num_bytes, u64 min_alloc_size,
1758 u64 root_objectid, u64 ref_generation,
1759 u64 owner, u64 owner_offset,
1760 u64 empty_size, u64 hint_byte,
1761 u64 search_end, struct btrfs_key *ins, u64 data)
1767 u64 search_start = 0;
1770 struct btrfs_fs_info *info = root->fs_info;
1771 struct btrfs_root *extent_root = info->extent_root;
1772 struct btrfs_extent_item *extent_item;
1773 struct btrfs_extent_ref *ref;
1774 struct btrfs_path *path;
1775 struct btrfs_key keys[2];
1778 alloc_profile = info->avail_data_alloc_bits &
1779 info->data_alloc_profile;
1780 data = BTRFS_BLOCK_GROUP_DATA | alloc_profile;
1781 } else if (root == root->fs_info->chunk_root) {
1782 alloc_profile = info->avail_system_alloc_bits &
1783 info->system_alloc_profile;
1784 data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile;
1786 alloc_profile = info->avail_metadata_alloc_bits &
1787 info->metadata_alloc_profile;
1788 data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile;
1791 data = reduce_alloc_profile(root, data);
1792 if (root->ref_cows) {
1793 if (!(data & BTRFS_BLOCK_GROUP_METADATA)) {
1794 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
1796 BTRFS_BLOCK_GROUP_METADATA |
1797 (info->metadata_alloc_profile &
1798 info->avail_metadata_alloc_bits));
1801 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
1802 num_bytes + 2 * 1024 * 1024, data);
1806 WARN_ON(num_bytes < root->sectorsize);
1807 ret = find_free_extent(trans, root, num_bytes, empty_size,
1808 search_start, search_end, hint_byte, ins,
1809 trans->alloc_exclude_start,
1810 trans->alloc_exclude_nr, data);
1812 if (ret == -ENOSPC && num_bytes > min_alloc_size) {
1813 num_bytes = num_bytes >> 1;
1814 num_bytes = max(num_bytes, min_alloc_size);
1818 printk("allocation failed flags %Lu\n", data);
1824 /* block accounting for super block */
1825 super_used = btrfs_super_bytes_used(&info->super_copy);
1826 btrfs_set_super_bytes_used(&info->super_copy, super_used + num_bytes);
1828 /* block accounting for root item */
1829 root_used = btrfs_root_used(&root->root_item);
1830 btrfs_set_root_used(&root->root_item, root_used + num_bytes);
1832 clear_extent_dirty(&root->fs_info->free_space_cache,
1833 ins->objectid, ins->objectid + ins->offset - 1,
1836 if (root == extent_root) {
1837 set_extent_bits(&root->fs_info->extent_ins, ins->objectid,
1838 ins->objectid + ins->offset - 1,
1839 EXTENT_LOCKED, GFP_NOFS);
1843 WARN_ON(trans->alloc_exclude_nr);
1844 trans->alloc_exclude_start = ins->objectid;
1845 trans->alloc_exclude_nr = ins->offset;
1847 memcpy(&keys[0], ins, sizeof(*ins));
1848 keys[1].offset = hash_extent_ref(root_objectid, ref_generation,
1849 owner, owner_offset);
1850 keys[1].objectid = ins->objectid;
1851 keys[1].type = BTRFS_EXTENT_REF_KEY;
1852 sizes[0] = sizeof(*extent_item);
1853 sizes[1] = sizeof(*ref);
1855 path = btrfs_alloc_path();
1858 ret = btrfs_insert_empty_items(trans, extent_root, path, keys,
1862 extent_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1863 struct btrfs_extent_item);
1864 btrfs_set_extent_refs(path->nodes[0], extent_item, 1);
1865 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
1866 struct btrfs_extent_ref);
1868 btrfs_set_ref_root(path->nodes[0], ref, root_objectid);
1869 btrfs_set_ref_generation(path->nodes[0], ref, ref_generation);
1870 btrfs_set_ref_objectid(path->nodes[0], ref, owner);
1871 btrfs_set_ref_offset(path->nodes[0], ref, owner_offset);
1873 btrfs_mark_buffer_dirty(path->nodes[0]);
1875 trans->alloc_exclude_start = 0;
1876 trans->alloc_exclude_nr = 0;
1877 btrfs_free_path(path);
1878 finish_current_insert(trans, extent_root);
1879 pending_ret = del_pending_extents(trans, extent_root);
1889 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1, 0);
1891 printk("update block group failed for %Lu %Lu\n",
1892 ins->objectid, ins->offset);
1899 * helper function to allocate a block for a given tree
1900 * returns the tree buffer or NULL.
1902 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
1903 struct btrfs_root *root,
1905 u64 root_objectid, u64 hint,
1911 ref_generation = trans->transid;
1916 return __btrfs_alloc_free_block(trans, root, blocksize, root_objectid,
1917 ref_generation, 0, 0, hint, empty_size);
1921 * helper function to allocate a block for a given tree
1922 * returns the tree buffer or NULL.
1924 struct extent_buffer *__btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
1925 struct btrfs_root *root,
1934 struct btrfs_key ins;
1936 struct extent_buffer *buf;
1938 ret = btrfs_alloc_extent(trans, root, blocksize, blocksize,
1939 root_objectid, ref_generation,
1940 level, first_objectid, empty_size, hint,
1944 return ERR_PTR(ret);
1946 buf = btrfs_find_create_tree_block(root, ins.objectid, blocksize);
1948 btrfs_free_extent(trans, root, ins.objectid, blocksize,
1949 root->root_key.objectid, ref_generation,
1951 return ERR_PTR(-ENOMEM);
1953 btrfs_set_header_generation(buf, trans->transid);
1954 clean_tree_block(trans, root, buf);
1955 btrfs_set_buffer_uptodate(buf);
1957 if (PageDirty(buf->first_page)) {
1958 printk("page %lu dirty\n", buf->first_page->index);
1962 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
1963 buf->start + buf->len - 1, GFP_NOFS);
1964 if (!btrfs_test_opt(root, SSD))
1965 btrfs_set_buffer_defrag(buf);
1966 trans->blocks_used++;
1970 static int noinline drop_leaf_ref(struct btrfs_trans_handle *trans,
1971 struct btrfs_root *root,
1972 struct extent_buffer *leaf)
1975 u64 leaf_generation;
1976 struct btrfs_key key;
1977 struct btrfs_file_extent_item *fi;
1982 BUG_ON(!btrfs_is_leaf(leaf));
1983 nritems = btrfs_header_nritems(leaf);
1984 leaf_owner = btrfs_header_owner(leaf);
1985 leaf_generation = btrfs_header_generation(leaf);
1987 for (i = 0; i < nritems; i++) {
1990 btrfs_item_key_to_cpu(leaf, &key, i);
1991 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1993 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
1994 if (btrfs_file_extent_type(leaf, fi) ==
1995 BTRFS_FILE_EXTENT_INLINE)
1998 * FIXME make sure to insert a trans record that
1999 * repeats the snapshot del on crash
2001 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
2002 if (disk_bytenr == 0)
2004 ret = btrfs_free_extent(trans, root, disk_bytenr,
2005 btrfs_file_extent_disk_num_bytes(leaf, fi),
2006 leaf_owner, leaf_generation,
2007 key.objectid, key.offset, 0);
2013 static void noinline reada_walk_down(struct btrfs_root *root,
2014 struct extent_buffer *node,
2027 nritems = btrfs_header_nritems(node);
2028 level = btrfs_header_level(node);
2032 for (i = slot; i < nritems && skipped < 32; i++) {
2033 bytenr = btrfs_node_blockptr(node, i);
2034 if (last && ((bytenr > last && bytenr - last > 32 * 1024) ||
2035 (last > bytenr && last - bytenr > 32 * 1024))) {
2039 blocksize = btrfs_level_size(root, level - 1);
2041 ret = lookup_extent_ref(NULL, root, bytenr,
2049 mutex_unlock(&root->fs_info->fs_mutex);
2050 ret = readahead_tree_block(root, bytenr, blocksize);
2051 last = bytenr + blocksize;
2053 mutex_lock(&root->fs_info->fs_mutex);
2060 * helper function for drop_snapshot, this walks down the tree dropping ref
2061 * counts as it goes.
2063 static int noinline walk_down_tree(struct btrfs_trans_handle *trans,
2064 struct btrfs_root *root,
2065 struct btrfs_path *path, int *level)
2070 struct extent_buffer *next;
2071 struct extent_buffer *cur;
2072 struct extent_buffer *parent;
2077 WARN_ON(*level < 0);
2078 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2079 ret = lookup_extent_ref(trans, root,
2080 path->nodes[*level]->start,
2081 path->nodes[*level]->len, &refs);
2087 * walk down to the last node level and free all the leaves
2089 while(*level >= 0) {
2090 WARN_ON(*level < 0);
2091 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2092 cur = path->nodes[*level];
2094 if (btrfs_header_level(cur) != *level)
2097 if (path->slots[*level] >=
2098 btrfs_header_nritems(cur))
2101 ret = drop_leaf_ref(trans, root, cur);
2105 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
2106 blocksize = btrfs_level_size(root, *level - 1);
2107 ret = lookup_extent_ref(trans, root, bytenr, blocksize, &refs);
2110 parent = path->nodes[*level];
2111 root_owner = btrfs_header_owner(parent);
2112 root_gen = btrfs_header_generation(parent);
2113 path->slots[*level]++;
2114 ret = btrfs_free_extent(trans, root, bytenr,
2115 blocksize, root_owner,
2120 next = btrfs_find_tree_block(root, bytenr, blocksize);
2121 if (!next || !btrfs_buffer_uptodate(next)) {
2122 free_extent_buffer(next);
2123 reada_walk_down(root, cur, path->slots[*level]);
2125 mutex_unlock(&root->fs_info->fs_mutex);
2126 next = read_tree_block(root, bytenr, blocksize);
2127 mutex_lock(&root->fs_info->fs_mutex);
2129 /* we've dropped the lock, double check */
2130 ret = lookup_extent_ref(trans, root, bytenr,
2134 parent = path->nodes[*level];
2135 root_owner = btrfs_header_owner(parent);
2136 root_gen = btrfs_header_generation(parent);
2138 path->slots[*level]++;
2139 free_extent_buffer(next);
2140 ret = btrfs_free_extent(trans, root, bytenr,
2148 btrfs_verify_block_csum(root, next);
2150 WARN_ON(*level <= 0);
2151 if (path->nodes[*level-1])
2152 free_extent_buffer(path->nodes[*level-1]);
2153 path->nodes[*level-1] = next;
2154 *level = btrfs_header_level(next);
2155 path->slots[*level] = 0;
2158 WARN_ON(*level < 0);
2159 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2161 if (path->nodes[*level] == root->node) {
2162 root_owner = root->root_key.objectid;
2163 parent = path->nodes[*level];
2165 parent = path->nodes[*level + 1];
2166 root_owner = btrfs_header_owner(parent);
2169 root_gen = btrfs_header_generation(parent);
2170 ret = btrfs_free_extent(trans, root, path->nodes[*level]->start,
2171 path->nodes[*level]->len,
2172 root_owner, root_gen, 0, 0, 1);
2173 free_extent_buffer(path->nodes[*level]);
2174 path->nodes[*level] = NULL;
2181 * helper for dropping snapshots. This walks back up the tree in the path
2182 * to find the first node higher up where we haven't yet gone through
2185 static int noinline walk_up_tree(struct btrfs_trans_handle *trans,
2186 struct btrfs_root *root,
2187 struct btrfs_path *path, int *level)
2191 struct btrfs_root_item *root_item = &root->root_item;
2196 for(i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
2197 slot = path->slots[i];
2198 if (slot < btrfs_header_nritems(path->nodes[i]) - 1) {
2199 struct extent_buffer *node;
2200 struct btrfs_disk_key disk_key;
2201 node = path->nodes[i];
2204 WARN_ON(*level == 0);
2205 btrfs_node_key(node, &disk_key, path->slots[i]);
2206 memcpy(&root_item->drop_progress,
2207 &disk_key, sizeof(disk_key));
2208 root_item->drop_level = i;
2211 if (path->nodes[*level] == root->node) {
2212 root_owner = root->root_key.objectid;
2214 btrfs_header_generation(path->nodes[*level]);
2216 struct extent_buffer *node;
2217 node = path->nodes[*level + 1];
2218 root_owner = btrfs_header_owner(node);
2219 root_gen = btrfs_header_generation(node);
2221 ret = btrfs_free_extent(trans, root,
2222 path->nodes[*level]->start,
2223 path->nodes[*level]->len,
2224 root_owner, root_gen, 0, 0, 1);
2226 free_extent_buffer(path->nodes[*level]);
2227 path->nodes[*level] = NULL;
2235 * drop the reference count on the tree rooted at 'snap'. This traverses
2236 * the tree freeing any blocks that have a ref count of zero after being
2239 int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root
2245 struct btrfs_path *path;
2248 struct btrfs_root_item *root_item = &root->root_item;
2250 path = btrfs_alloc_path();
2253 level = btrfs_header_level(root->node);
2255 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
2256 path->nodes[level] = root->node;
2257 extent_buffer_get(root->node);
2258 path->slots[level] = 0;
2260 struct btrfs_key key;
2261 struct btrfs_disk_key found_key;
2262 struct extent_buffer *node;
2264 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
2265 level = root_item->drop_level;
2266 path->lowest_level = level;
2267 wret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2272 node = path->nodes[level];
2273 btrfs_node_key(node, &found_key, path->slots[level]);
2274 WARN_ON(memcmp(&found_key, &root_item->drop_progress,
2275 sizeof(found_key)));
2278 wret = walk_down_tree(trans, root, path, &level);
2284 wret = walk_up_tree(trans, root, path, &level);
2292 for (i = 0; i <= orig_level; i++) {
2293 if (path->nodes[i]) {
2294 free_extent_buffer(path->nodes[i]);
2295 path->nodes[i] = NULL;
2299 btrfs_free_path(path);
2303 int btrfs_free_block_groups(struct btrfs_fs_info *info)
2310 ret = find_first_extent_bit(&info->block_group_cache, 0,
2311 &start, &end, (unsigned int)-1);
2314 ret = get_state_private(&info->block_group_cache, start, &ptr);
2316 kfree((void *)(unsigned long)ptr);
2317 clear_extent_bits(&info->block_group_cache, start,
2318 end, (unsigned int)-1, GFP_NOFS);
2321 ret = find_first_extent_bit(&info->free_space_cache, 0,
2322 &start, &end, EXTENT_DIRTY);
2325 clear_extent_dirty(&info->free_space_cache, start,
2331 static unsigned long calc_ra(unsigned long start, unsigned long last,
2334 return min(last, start + nr - 1);
2337 static int noinline relocate_inode_pages(struct inode *inode, u64 start,
2342 unsigned long last_index;
2345 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2346 struct file_ra_state *ra;
2347 unsigned long total_read = 0;
2348 unsigned long ra_pages;
2349 struct btrfs_trans_handle *trans;
2351 ra = kzalloc(sizeof(*ra), GFP_NOFS);
2353 mutex_lock(&inode->i_mutex);
2354 i = start >> PAGE_CACHE_SHIFT;
2355 last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
2357 ra_pages = BTRFS_I(inode)->root->fs_info->bdi.ra_pages;
2359 file_ra_state_init(ra, inode->i_mapping);
2361 for (; i <= last_index; i++) {
2362 if (total_read % ra_pages == 0) {
2363 btrfs_force_ra(inode->i_mapping, ra, NULL, i,
2364 calc_ra(i, last_index, ra_pages));
2367 if (((u64)i << PAGE_CACHE_SHIFT) > inode->i_size)
2368 goto truncate_racing;
2370 page = grab_cache_page(inode->i_mapping, i);
2374 if (!PageUptodate(page)) {
2375 btrfs_readpage(NULL, page);
2377 if (!PageUptodate(page)) {
2379 page_cache_release(page);
2383 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
2384 ClearPageDirty(page);
2386 cancel_dirty_page(page, PAGE_CACHE_SIZE);
2388 wait_on_page_writeback(page);
2389 set_page_extent_mapped(page);
2390 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
2391 page_end = page_start + PAGE_CACHE_SIZE - 1;
2393 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
2395 set_extent_delalloc(io_tree, page_start,
2396 page_end, GFP_NOFS);
2397 set_page_dirty(page);
2399 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
2401 page_cache_release(page);
2403 balance_dirty_pages_ratelimited_nr(inode->i_mapping,
2408 trans = btrfs_start_transaction(BTRFS_I(inode)->root, 1);
2410 btrfs_add_ordered_inode(inode);
2411 btrfs_end_transaction(trans, BTRFS_I(inode)->root);
2412 mark_inode_dirty(inode);
2414 mutex_unlock(&inode->i_mutex);
2418 vmtruncate(inode, inode->i_size);
2419 balance_dirty_pages_ratelimited_nr(inode->i_mapping,
2425 * The back references tell us which tree holds a ref on a block,
2426 * but it is possible for the tree root field in the reference to
2427 * reflect the original root before a snapshot was made. In this
2428 * case we should search through all the children of a given root
2429 * to find potential holders of references on a block.
2431 * Instead, we do something a little less fancy and just search
2432 * all the roots for a given key/block combination.
2434 static int find_root_for_ref(struct btrfs_root *root,
2435 struct btrfs_path *path,
2436 struct btrfs_key *key0,
2439 struct btrfs_root **found_root,
2442 struct btrfs_key root_location;
2443 struct btrfs_root *cur_root = *found_root;
2444 struct btrfs_file_extent_item *file_extent;
2445 u64 root_search_start = BTRFS_FS_TREE_OBJECTID;
2450 root_location.offset = (u64)-1;
2451 root_location.type = BTRFS_ROOT_ITEM_KEY;
2452 path->lowest_level = level;
2455 ret = btrfs_search_slot(NULL, cur_root, key0, path, 0, 0);
2457 if (ret == 0 && file_key) {
2458 struct extent_buffer *leaf = path->nodes[0];
2459 file_extent = btrfs_item_ptr(leaf, path->slots[0],
2460 struct btrfs_file_extent_item);
2461 if (btrfs_file_extent_type(leaf, file_extent) ==
2462 BTRFS_FILE_EXTENT_REG) {
2464 btrfs_file_extent_disk_bytenr(leaf,
2467 } else if (ret == 0) {
2468 if (path->nodes[level])
2469 found_bytenr = path->nodes[level]->start;
2472 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
2473 if (!path->nodes[i])
2475 free_extent_buffer(path->nodes[i]);
2476 path->nodes[i] = NULL;
2478 btrfs_release_path(cur_root, path);
2480 if (found_bytenr == bytenr) {
2481 *found_root = cur_root;
2485 ret = btrfs_search_root(root->fs_info->tree_root,
2486 root_search_start, &root_search_start);
2490 root_location.objectid = root_search_start;
2491 cur_root = btrfs_read_fs_root_no_name(root->fs_info,
2499 path->lowest_level = 0;
2504 * note, this releases the path
2506 static int noinline relocate_one_reference(struct btrfs_root *extent_root,
2507 struct btrfs_path *path,
2508 struct btrfs_key *extent_key)
2510 struct inode *inode;
2511 struct btrfs_root *found_root;
2512 struct btrfs_key root_location;
2513 struct btrfs_key found_key;
2514 struct btrfs_extent_ref *ref;
2522 ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
2523 struct btrfs_extent_ref);
2524 ref_root = btrfs_ref_root(path->nodes[0], ref);
2525 ref_gen = btrfs_ref_generation(path->nodes[0], ref);
2526 ref_objectid = btrfs_ref_objectid(path->nodes[0], ref);
2527 ref_offset = btrfs_ref_offset(path->nodes[0], ref);
2528 btrfs_release_path(extent_root, path);
2530 root_location.objectid = ref_root;
2532 root_location.offset = 0;
2534 root_location.offset = (u64)-1;
2535 root_location.type = BTRFS_ROOT_ITEM_KEY;
2537 found_root = btrfs_read_fs_root_no_name(extent_root->fs_info,
2539 BUG_ON(!found_root);
2541 if (ref_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
2542 found_key.objectid = ref_objectid;
2543 found_key.type = BTRFS_EXTENT_DATA_KEY;
2544 found_key.offset = ref_offset;
2547 ret = find_root_for_ref(extent_root, path, &found_key,
2548 level, 1, &found_root,
2549 extent_key->objectid);
2554 mutex_unlock(&extent_root->fs_info->fs_mutex);
2555 inode = btrfs_iget_locked(extent_root->fs_info->sb,
2556 ref_objectid, found_root);
2557 if (inode->i_state & I_NEW) {
2558 /* the inode and parent dir are two different roots */
2559 BTRFS_I(inode)->root = found_root;
2560 BTRFS_I(inode)->location.objectid = ref_objectid;
2561 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
2562 BTRFS_I(inode)->location.offset = 0;
2563 btrfs_read_locked_inode(inode);
2564 unlock_new_inode(inode);
2567 /* this can happen if the reference is not against
2568 * the latest version of the tree root
2570 if (is_bad_inode(inode)) {
2571 mutex_lock(&extent_root->fs_info->fs_mutex);
2574 relocate_inode_pages(inode, ref_offset, extent_key->offset);
2576 mutex_lock(&extent_root->fs_info->fs_mutex);
2578 struct btrfs_trans_handle *trans;
2579 struct extent_buffer *eb;
2582 eb = read_tree_block(found_root, extent_key->objectid,
2583 extent_key->offset);
2584 level = btrfs_header_level(eb);
2587 btrfs_item_key_to_cpu(eb, &found_key, 0);
2589 btrfs_node_key_to_cpu(eb, &found_key, 0);
2591 free_extent_buffer(eb);
2593 ret = find_root_for_ref(extent_root, path, &found_key,
2594 level, 0, &found_root,
2595 extent_key->objectid);
2600 trans = btrfs_start_transaction(found_root, 1);
2602 path->lowest_level = level;
2604 ret = btrfs_search_slot(trans, found_root, &found_key, path,
2606 path->lowest_level = 0;
2607 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
2608 if (!path->nodes[i])
2610 free_extent_buffer(path->nodes[i]);
2611 path->nodes[i] = NULL;
2613 btrfs_release_path(found_root, path);
2614 btrfs_end_transaction(trans, found_root);
2621 static int noinline del_extent_zero(struct btrfs_root *extent_root,
2622 struct btrfs_path *path,
2623 struct btrfs_key *extent_key)
2626 struct btrfs_trans_handle *trans;
2628 trans = btrfs_start_transaction(extent_root, 1);
2629 ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1);
2636 ret = btrfs_del_item(trans, extent_root, path);
2638 btrfs_end_transaction(trans, extent_root);
2642 static int noinline relocate_one_extent(struct btrfs_root *extent_root,
2643 struct btrfs_path *path,
2644 struct btrfs_key *extent_key)
2646 struct btrfs_key key;
2647 struct btrfs_key found_key;
2648 struct extent_buffer *leaf;
2653 if (extent_key->objectid == 0) {
2654 ret = del_extent_zero(extent_root, path, extent_key);
2657 key.objectid = extent_key->objectid;
2658 key.type = BTRFS_EXTENT_REF_KEY;
2662 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2668 leaf = path->nodes[0];
2669 nritems = btrfs_header_nritems(leaf);
2670 if (path->slots[0] == nritems) {
2671 ret = btrfs_next_leaf(extent_root, path);
2678 leaf = path->nodes[0];
2681 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2682 if (found_key.objectid != extent_key->objectid) {
2686 if (found_key.type != BTRFS_EXTENT_REF_KEY) {
2690 key.offset = found_key.offset + 1;
2691 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2693 ret = relocate_one_reference(extent_root, path, extent_key);
2699 btrfs_release_path(extent_root, path);
2703 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
2706 u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
2707 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
2709 num_devices = root->fs_info->fs_devices->num_devices;
2710 if (num_devices == 1) {
2711 stripped |= BTRFS_BLOCK_GROUP_DUP;
2712 stripped = flags & ~stripped;
2714 /* turn raid0 into single device chunks */
2715 if (flags & BTRFS_BLOCK_GROUP_RAID0)
2718 /* turn mirroring into duplication */
2719 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
2720 BTRFS_BLOCK_GROUP_RAID10))
2721 return stripped | BTRFS_BLOCK_GROUP_DUP;
2724 /* they already had raid on here, just return */
2725 if (flags & stripped)
2728 stripped |= BTRFS_BLOCK_GROUP_DUP;
2729 stripped = flags & ~stripped;
2731 /* switch duplicated blocks with raid1 */
2732 if (flags & BTRFS_BLOCK_GROUP_DUP)
2733 return stripped | BTRFS_BLOCK_GROUP_RAID1;
2735 /* turn single device chunks into raid0 */
2736 return stripped | BTRFS_BLOCK_GROUP_RAID0;
2741 int btrfs_shrink_extent_tree(struct btrfs_root *root, u64 shrink_start)
2743 struct btrfs_trans_handle *trans;
2744 struct btrfs_root *tree_root = root->fs_info->tree_root;
2745 struct btrfs_path *path;
2748 u64 shrink_last_byte;
2749 u64 new_alloc_flags;
2750 struct btrfs_block_group_cache *shrink_block_group;
2751 struct btrfs_fs_info *info = root->fs_info;
2752 struct btrfs_key key;
2753 struct btrfs_key found_key;
2754 struct extent_buffer *leaf;
2759 shrink_block_group = btrfs_lookup_block_group(root->fs_info,
2761 BUG_ON(!shrink_block_group);
2763 shrink_last_byte = shrink_start + shrink_block_group->key.offset;
2765 shrink_block_group->space_info->total_bytes -=
2766 shrink_block_group->key.offset;
2767 path = btrfs_alloc_path();
2768 root = root->fs_info->extent_root;
2772 if (btrfs_block_group_used(&shrink_block_group->item) > 0) {
2773 trans = btrfs_start_transaction(root, 1);
2774 new_alloc_flags = update_block_group_flags(root,
2775 shrink_block_group->flags);
2776 do_chunk_alloc(trans, root->fs_info->extent_root,
2777 btrfs_block_group_used(&shrink_block_group->item) +
2778 2 * 1024 * 1024, new_alloc_flags);
2779 btrfs_end_transaction(trans, root);
2781 shrink_block_group->ro = 1;
2785 key.objectid = shrink_start;
2788 cur_byte = key.objectid;
2790 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2794 ret = btrfs_previous_item(root, path, 0, BTRFS_EXTENT_ITEM_KEY);
2799 leaf = path->nodes[0];
2800 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2801 if (found_key.objectid + found_key.offset > shrink_start &&
2802 found_key.objectid < shrink_last_byte) {
2803 cur_byte = found_key.objectid;
2804 key.objectid = cur_byte;
2807 btrfs_release_path(root, path);
2810 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2814 leaf = path->nodes[0];
2815 nritems = btrfs_header_nritems(leaf);
2817 if (path->slots[0] >= nritems) {
2818 ret = btrfs_next_leaf(root, path);
2825 leaf = path->nodes[0];
2826 nritems = btrfs_header_nritems(leaf);
2829 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2831 if (found_key.objectid >= shrink_last_byte)
2834 if (progress && need_resched()) {
2835 memcpy(&key, &found_key, sizeof(key));
2836 mutex_unlock(&root->fs_info->fs_mutex);
2838 mutex_lock(&root->fs_info->fs_mutex);
2839 btrfs_release_path(root, path);
2840 btrfs_search_slot(NULL, root, &key, path, 0, 0);
2846 if (btrfs_key_type(&found_key) != BTRFS_EXTENT_ITEM_KEY ||
2847 found_key.objectid + found_key.offset <= cur_byte) {
2853 cur_byte = found_key.objectid + found_key.offset;
2854 key.objectid = cur_byte;
2855 btrfs_release_path(root, path);
2856 ret = relocate_one_extent(root, path, &found_key);
2859 btrfs_release_path(root, path);
2861 if (total_found > 0) {
2862 trans = btrfs_start_transaction(tree_root, 1);
2863 btrfs_commit_transaction(trans, tree_root);
2865 mutex_unlock(&root->fs_info->fs_mutex);
2866 btrfs_clean_old_snapshots(tree_root);
2867 mutex_lock(&root->fs_info->fs_mutex);
2869 trans = btrfs_start_transaction(tree_root, 1);
2870 btrfs_commit_transaction(trans, tree_root);
2875 * we've freed all the extents, now remove the block
2876 * group item from the tree
2878 trans = btrfs_start_transaction(root, 1);
2879 memcpy(&key, &shrink_block_group->key, sizeof(key));
2881 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2887 leaf = path->nodes[0];
2888 nritems = btrfs_header_nritems(leaf);
2889 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2890 kfree(shrink_block_group);
2892 clear_extent_bits(&info->block_group_cache, found_key.objectid,
2893 found_key.objectid + found_key.offset - 1,
2894 (unsigned int)-1, GFP_NOFS);
2896 btrfs_del_item(trans, root, path);
2897 clear_extent_dirty(&info->free_space_cache,
2898 shrink_start, shrink_last_byte - 1,
2900 btrfs_commit_transaction(trans, root);
2902 btrfs_free_path(path);
2906 int find_first_block_group(struct btrfs_root *root, struct btrfs_path *path,
2907 struct btrfs_key *key)
2910 struct btrfs_key found_key;
2911 struct extent_buffer *leaf;
2914 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
2918 slot = path->slots[0];
2919 leaf = path->nodes[0];
2920 if (slot >= btrfs_header_nritems(leaf)) {
2921 ret = btrfs_next_leaf(root, path);
2928 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2930 if (found_key.objectid >= key->objectid &&
2931 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY)
2940 int btrfs_read_block_groups(struct btrfs_root *root)
2942 struct btrfs_path *path;
2945 struct btrfs_block_group_cache *cache;
2946 struct btrfs_fs_info *info = root->fs_info;
2947 struct btrfs_space_info *space_info;
2948 struct extent_io_tree *block_group_cache;
2949 struct btrfs_key key;
2950 struct btrfs_key found_key;
2951 struct extent_buffer *leaf;
2953 block_group_cache = &info->block_group_cache;
2954 root = info->extent_root;
2957 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
2958 path = btrfs_alloc_path();
2963 ret = find_first_block_group(root, path, &key);
2971 leaf = path->nodes[0];
2972 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2973 cache = kzalloc(sizeof(*cache), GFP_NOFS);
2979 read_extent_buffer(leaf, &cache->item,
2980 btrfs_item_ptr_offset(leaf, path->slots[0]),
2981 sizeof(cache->item));
2982 memcpy(&cache->key, &found_key, sizeof(found_key));
2984 key.objectid = found_key.objectid + found_key.offset;
2985 btrfs_release_path(root, path);
2986 cache->flags = btrfs_block_group_flags(&cache->item);
2988 if (cache->flags & BTRFS_BLOCK_GROUP_DATA) {
2989 bit = BLOCK_GROUP_DATA;
2990 } else if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) {
2991 bit = BLOCK_GROUP_SYSTEM;
2992 } else if (cache->flags & BTRFS_BLOCK_GROUP_METADATA) {
2993 bit = BLOCK_GROUP_METADATA;
2995 set_avail_alloc_bits(info, cache->flags);
2997 ret = update_space_info(info, cache->flags, found_key.offset,
2998 btrfs_block_group_used(&cache->item),
3001 cache->space_info = space_info;
3003 /* use EXTENT_LOCKED to prevent merging */
3004 set_extent_bits(block_group_cache, found_key.objectid,
3005 found_key.objectid + found_key.offset - 1,
3006 bit | EXTENT_LOCKED, GFP_NOFS);
3007 set_state_private(block_group_cache, found_key.objectid,
3008 (unsigned long)cache);
3011 btrfs_super_total_bytes(&info->super_copy))
3016 btrfs_free_path(path);
3020 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
3021 struct btrfs_root *root, u64 bytes_used,
3022 u64 type, u64 chunk_objectid, u64 chunk_offset,
3027 struct btrfs_root *extent_root;
3028 struct btrfs_block_group_cache *cache;
3029 struct extent_io_tree *block_group_cache;
3031 extent_root = root->fs_info->extent_root;
3032 block_group_cache = &root->fs_info->block_group_cache;
3034 cache = kzalloc(sizeof(*cache), GFP_NOFS);
3036 cache->key.objectid = chunk_offset;
3037 cache->key.offset = size;
3039 btrfs_set_key_type(&cache->key, BTRFS_BLOCK_GROUP_ITEM_KEY);
3040 memset(&cache->item, 0, sizeof(cache->item));
3041 btrfs_set_block_group_used(&cache->item, bytes_used);
3042 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
3043 cache->flags = type;
3044 btrfs_set_block_group_flags(&cache->item, type);
3046 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
3047 &cache->space_info);
3050 bit = block_group_state_bits(type);
3051 set_extent_bits(block_group_cache, chunk_offset,
3052 chunk_offset + size - 1,
3053 bit | EXTENT_LOCKED, GFP_NOFS);
3055 set_state_private(block_group_cache, chunk_offset,
3056 (unsigned long)cache);
3057 ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
3058 sizeof(cache->item));
3061 finish_current_insert(trans, extent_root);
3062 ret = del_pending_extents(trans, extent_root);
3064 set_avail_alloc_bits(extent_root->fs_info, type);