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,
727 struct btrfs_root *extent_root = root->fs_info->extent_root;
728 struct btrfs_path *path;
732 u64 root_objectid = root->root_key.objectid;
737 struct btrfs_key key;
738 struct btrfs_key found_key;
739 struct extent_buffer *l;
740 struct btrfs_extent_item *item;
741 struct btrfs_extent_ref *ref_item;
744 path = btrfs_alloc_path();
747 bytenr = first_extent;
749 bytenr = count_path->nodes[level]->start;
752 key.objectid = bytenr;
755 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
756 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
762 btrfs_item_key_to_cpu(l, &found_key, path->slots[0]);
764 if (found_key.objectid != bytenr ||
765 found_key.type != BTRFS_EXTENT_ITEM_KEY) {
769 item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
772 nritems = btrfs_header_nritems(l);
773 if (path->slots[0] >= nritems) {
774 ret = btrfs_next_leaf(extent_root, path);
779 btrfs_item_key_to_cpu(l, &found_key, path->slots[0]);
780 if (found_key.objectid != bytenr)
783 if (found_key.type != BTRFS_EXTENT_REF_KEY) {
789 ref_item = btrfs_item_ptr(l, path->slots[0],
790 struct btrfs_extent_ref);
791 found_objectid = btrfs_ref_root(l, ref_item);
793 if (found_objectid != root_objectid) {
798 found_owner = btrfs_ref_objectid(l, ref_item);
799 if (found_owner != expected_owner) {
807 if (cur_count == 0) {
811 if (level >= 0 && root->node == count_path->nodes[level])
814 btrfs_release_path(root, path);
818 btrfs_free_path(path);
821 int btrfs_inc_root_ref(struct btrfs_trans_handle *trans,
822 struct btrfs_root *root, u64 owner_objectid)
828 struct btrfs_disk_key disk_key;
830 level = btrfs_header_level(root->node);
831 generation = trans->transid;
832 nritems = btrfs_header_nritems(root->node);
835 btrfs_item_key(root->node, &disk_key, 0);
837 btrfs_node_key(root->node, &disk_key, 0);
838 key_objectid = btrfs_disk_key_objectid(&disk_key);
842 return btrfs_inc_extent_ref(trans, root, root->node->start,
843 root->node->len, owner_objectid,
844 generation, level, key_objectid);
847 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
848 struct extent_buffer *buf)
852 struct btrfs_key key;
853 struct btrfs_file_extent_item *fi;
862 level = btrfs_header_level(buf);
863 nritems = btrfs_header_nritems(buf);
864 for (i = 0; i < nritems; i++) {
867 btrfs_item_key_to_cpu(buf, &key, i);
868 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
870 fi = btrfs_item_ptr(buf, i,
871 struct btrfs_file_extent_item);
872 if (btrfs_file_extent_type(buf, fi) ==
873 BTRFS_FILE_EXTENT_INLINE)
875 disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
876 if (disk_bytenr == 0)
878 ret = btrfs_inc_extent_ref(trans, root, disk_bytenr,
879 btrfs_file_extent_disk_num_bytes(buf, fi),
880 root->root_key.objectid, trans->transid,
881 key.objectid, key.offset);
887 bytenr = btrfs_node_blockptr(buf, i);
888 btrfs_node_key_to_cpu(buf, &key, i);
889 ret = btrfs_inc_extent_ref(trans, root, bytenr,
890 btrfs_level_size(root, level - 1),
891 root->root_key.objectid,
893 level - 1, key.objectid);
904 for (i =0; i < faili; i++) {
907 btrfs_item_key_to_cpu(buf, &key, i);
908 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
910 fi = btrfs_item_ptr(buf, i,
911 struct btrfs_file_extent_item);
912 if (btrfs_file_extent_type(buf, fi) ==
913 BTRFS_FILE_EXTENT_INLINE)
915 disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
916 if (disk_bytenr == 0)
918 err = btrfs_free_extent(trans, root, disk_bytenr,
919 btrfs_file_extent_disk_num_bytes(buf,
923 bytenr = btrfs_node_blockptr(buf, i);
924 err = btrfs_free_extent(trans, root, bytenr,
925 btrfs_level_size(root, level - 1), 0);
933 static int write_one_cache_group(struct btrfs_trans_handle *trans,
934 struct btrfs_root *root,
935 struct btrfs_path *path,
936 struct btrfs_block_group_cache *cache)
940 struct btrfs_root *extent_root = root->fs_info->extent_root;
942 struct extent_buffer *leaf;
944 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
949 leaf = path->nodes[0];
950 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
951 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
952 btrfs_mark_buffer_dirty(leaf);
953 btrfs_release_path(extent_root, path);
955 finish_current_insert(trans, extent_root);
956 pending_ret = del_pending_extents(trans, extent_root);
965 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
966 struct btrfs_root *root)
968 struct extent_io_tree *block_group_cache;
969 struct btrfs_block_group_cache *cache;
973 struct btrfs_path *path;
979 block_group_cache = &root->fs_info->block_group_cache;
980 path = btrfs_alloc_path();
985 ret = find_first_extent_bit(block_group_cache, last,
986 &start, &end, BLOCK_GROUP_DIRTY);
991 ret = get_state_private(block_group_cache, start, &ptr);
994 cache = (struct btrfs_block_group_cache *)(unsigned long)ptr;
995 err = write_one_cache_group(trans, root,
998 * if we fail to write the cache group, we want
999 * to keep it marked dirty in hopes that a later
1006 clear_extent_bits(block_group_cache, start, end,
1007 BLOCK_GROUP_DIRTY, GFP_NOFS);
1009 btrfs_free_path(path);
1013 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
1016 struct list_head *head = &info->space_info;
1017 struct list_head *cur;
1018 struct btrfs_space_info *found;
1019 list_for_each(cur, head) {
1020 found = list_entry(cur, struct btrfs_space_info, list);
1021 if (found->flags == flags)
1028 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
1029 u64 total_bytes, u64 bytes_used,
1030 struct btrfs_space_info **space_info)
1032 struct btrfs_space_info *found;
1034 found = __find_space_info(info, flags);
1036 found->total_bytes += total_bytes;
1037 found->bytes_used += bytes_used;
1039 WARN_ON(found->total_bytes < found->bytes_used);
1040 *space_info = found;
1043 found = kmalloc(sizeof(*found), GFP_NOFS);
1047 list_add(&found->list, &info->space_info);
1048 found->flags = flags;
1049 found->total_bytes = total_bytes;
1050 found->bytes_used = bytes_used;
1051 found->bytes_pinned = 0;
1053 *space_info = found;
1057 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
1059 u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
1060 BTRFS_BLOCK_GROUP_RAID1 |
1061 BTRFS_BLOCK_GROUP_RAID10 |
1062 BTRFS_BLOCK_GROUP_DUP);
1064 if (flags & BTRFS_BLOCK_GROUP_DATA)
1065 fs_info->avail_data_alloc_bits |= extra_flags;
1066 if (flags & BTRFS_BLOCK_GROUP_METADATA)
1067 fs_info->avail_metadata_alloc_bits |= extra_flags;
1068 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
1069 fs_info->avail_system_alloc_bits |= extra_flags;
1073 static u64 reduce_alloc_profile(struct btrfs_root *root, u64 flags)
1075 u64 num_devices = root->fs_info->fs_devices->num_devices;
1077 if (num_devices == 1)
1078 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
1079 if (num_devices < 4)
1080 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
1082 if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
1083 (flags & (BTRFS_BLOCK_GROUP_RAID1 |
1084 BTRFS_BLOCK_GROUP_RAID10))) {
1085 flags &= ~BTRFS_BLOCK_GROUP_DUP;
1088 if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
1089 (flags & BTRFS_BLOCK_GROUP_RAID10)) {
1090 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
1093 if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
1094 ((flags & BTRFS_BLOCK_GROUP_RAID1) |
1095 (flags & BTRFS_BLOCK_GROUP_RAID10) |
1096 (flags & BTRFS_BLOCK_GROUP_DUP)))
1097 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
1101 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
1102 struct btrfs_root *extent_root, u64 alloc_bytes,
1105 struct btrfs_space_info *space_info;
1111 flags = reduce_alloc_profile(extent_root, flags);
1113 space_info = __find_space_info(extent_root->fs_info, flags);
1115 ret = update_space_info(extent_root->fs_info, flags,
1119 BUG_ON(!space_info);
1121 if (space_info->full)
1124 thresh = div_factor(space_info->total_bytes, 6);
1125 if ((space_info->bytes_used + space_info->bytes_pinned + alloc_bytes) <
1129 ret = btrfs_alloc_chunk(trans, extent_root, &start, &num_bytes, flags);
1130 if (ret == -ENOSPC) {
1131 printk("space info full %Lu\n", flags);
1132 space_info->full = 1;
1138 ret = btrfs_make_block_group(trans, extent_root, 0, flags,
1139 BTRFS_FIRST_CHUNK_TREE_OBJECTID, start, num_bytes);
1145 static int update_block_group(struct btrfs_trans_handle *trans,
1146 struct btrfs_root *root,
1147 u64 bytenr, u64 num_bytes, int alloc,
1150 struct btrfs_block_group_cache *cache;
1151 struct btrfs_fs_info *info = root->fs_info;
1152 u64 total = num_bytes;
1159 cache = btrfs_lookup_block_group(info, bytenr);
1163 byte_in_group = bytenr - cache->key.objectid;
1164 WARN_ON(byte_in_group > cache->key.offset);
1165 start = cache->key.objectid;
1166 end = start + cache->key.offset - 1;
1167 set_extent_bits(&info->block_group_cache, start, end,
1168 BLOCK_GROUP_DIRTY, GFP_NOFS);
1170 old_val = btrfs_block_group_used(&cache->item);
1171 num_bytes = min(total, cache->key.offset - byte_in_group);
1173 old_val += num_bytes;
1174 cache->space_info->bytes_used += num_bytes;
1176 old_val -= num_bytes;
1177 cache->space_info->bytes_used -= num_bytes;
1179 set_extent_dirty(&info->free_space_cache,
1180 bytenr, bytenr + num_bytes - 1,
1184 btrfs_set_block_group_used(&cache->item, old_val);
1186 bytenr += num_bytes;
1191 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
1196 ret = find_first_extent_bit(&root->fs_info->block_group_cache,
1197 search_start, &start, &end,
1198 BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA |
1199 BLOCK_GROUP_SYSTEM);
1206 static int update_pinned_extents(struct btrfs_root *root,
1207 u64 bytenr, u64 num, int pin)
1210 struct btrfs_block_group_cache *cache;
1211 struct btrfs_fs_info *fs_info = root->fs_info;
1214 set_extent_dirty(&fs_info->pinned_extents,
1215 bytenr, bytenr + num - 1, GFP_NOFS);
1217 clear_extent_dirty(&fs_info->pinned_extents,
1218 bytenr, bytenr + num - 1, GFP_NOFS);
1221 cache = btrfs_lookup_block_group(fs_info, bytenr);
1223 u64 first = first_logical_byte(root, bytenr);
1224 WARN_ON(first < bytenr);
1225 len = min(first - bytenr, num);
1227 len = min(num, cache->key.offset -
1228 (bytenr - cache->key.objectid));
1232 cache->pinned += len;
1233 cache->space_info->bytes_pinned += len;
1235 fs_info->total_pinned += len;
1238 cache->pinned -= len;
1239 cache->space_info->bytes_pinned -= len;
1241 fs_info->total_pinned -= len;
1249 int btrfs_copy_pinned(struct btrfs_root *root, struct extent_io_tree *copy)
1254 struct extent_io_tree *pinned_extents = &root->fs_info->pinned_extents;
1258 ret = find_first_extent_bit(pinned_extents, last,
1259 &start, &end, EXTENT_DIRTY);
1262 set_extent_dirty(copy, start, end, GFP_NOFS);
1268 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
1269 struct btrfs_root *root,
1270 struct extent_io_tree *unpin)
1275 struct extent_io_tree *free_space_cache;
1276 free_space_cache = &root->fs_info->free_space_cache;
1279 ret = find_first_extent_bit(unpin, 0, &start, &end,
1283 update_pinned_extents(root, start, end + 1 - start, 0);
1284 clear_extent_dirty(unpin, start, end, GFP_NOFS);
1285 set_extent_dirty(free_space_cache, start, end, GFP_NOFS);
1290 static int finish_current_insert(struct btrfs_trans_handle *trans,
1291 struct btrfs_root *extent_root)
1295 struct btrfs_fs_info *info = extent_root->fs_info;
1296 struct extent_buffer *eb;
1297 struct btrfs_path *path;
1298 struct btrfs_key ins;
1299 struct btrfs_disk_key first;
1300 struct btrfs_extent_item extent_item;
1305 btrfs_set_stack_extent_refs(&extent_item, 1);
1306 btrfs_set_key_type(&ins, BTRFS_EXTENT_ITEM_KEY);
1307 path = btrfs_alloc_path();
1310 ret = find_first_extent_bit(&info->extent_ins, 0, &start,
1311 &end, EXTENT_LOCKED);
1315 ins.objectid = start;
1316 ins.offset = end + 1 - start;
1317 err = btrfs_insert_item(trans, extent_root, &ins,
1318 &extent_item, sizeof(extent_item));
1319 clear_extent_bits(&info->extent_ins, start, end, EXTENT_LOCKED,
1321 eb = read_tree_block(extent_root, ins.objectid, ins.offset);
1322 level = btrfs_header_level(eb);
1324 btrfs_item_key(eb, &first, 0);
1326 btrfs_node_key(eb, &first, 0);
1328 err = btrfs_insert_extent_backref(trans, extent_root, path,
1329 start, extent_root->root_key.objectid,
1331 btrfs_disk_key_objectid(&first));
1333 free_extent_buffer(eb);
1335 btrfs_free_path(path);
1339 static int pin_down_bytes(struct btrfs_root *root, u64 bytenr, u32 num_bytes,
1343 struct extent_buffer *buf;
1346 buf = btrfs_find_tree_block(root, bytenr, num_bytes);
1348 if (btrfs_buffer_uptodate(buf)) {
1350 root->fs_info->running_transaction->transid;
1351 u64 header_transid =
1352 btrfs_header_generation(buf);
1353 if (header_transid == transid &&
1354 !btrfs_header_flag(buf,
1355 BTRFS_HEADER_FLAG_WRITTEN)) {
1356 clean_tree_block(NULL, root, buf);
1357 free_extent_buffer(buf);
1361 free_extent_buffer(buf);
1363 update_pinned_extents(root, bytenr, num_bytes, 1);
1365 set_extent_bits(&root->fs_info->pending_del,
1366 bytenr, bytenr + num_bytes - 1,
1367 EXTENT_LOCKED, GFP_NOFS);
1374 * remove an extent from the root, returns 0 on success
1376 static int __free_extent(struct btrfs_trans_handle *trans, struct btrfs_root
1377 *root, u64 bytenr, u64 num_bytes,
1378 u64 root_objectid, u64 ref_generation,
1379 u64 owner_objectid, u64 owner_offset, int pin,
1382 struct btrfs_path *path;
1383 struct btrfs_key key;
1384 struct btrfs_fs_info *info = root->fs_info;
1385 struct btrfs_root *extent_root = info->extent_root;
1386 struct extent_buffer *leaf;
1388 int extent_slot = 0;
1389 int found_extent = 0;
1391 struct btrfs_extent_item *ei;
1394 key.objectid = bytenr;
1395 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
1396 key.offset = num_bytes;
1397 path = btrfs_alloc_path();
1402 ret = lookup_extent_backref(trans, extent_root, path,
1403 bytenr, root_objectid,
1405 owner_objectid, owner_offset, 1);
1407 struct btrfs_key found_key;
1408 extent_slot = path->slots[0];
1409 while(extent_slot > 0) {
1411 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1413 if (found_key.objectid != bytenr)
1415 if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
1416 found_key.offset == num_bytes) {
1420 if (path->slots[0] - extent_slot > 5)
1424 ret = btrfs_del_item(trans, extent_root, path);
1426 btrfs_print_leaf(extent_root, path->nodes[0]);
1428 printk("Unable to find ref byte nr %Lu root %Lu "
1429 " gen %Lu owner %Lu offset %Lu\n", bytenr,
1430 root_objectid, ref_generation, owner_objectid,
1433 if (!found_extent) {
1434 btrfs_release_path(extent_root, path);
1435 ret = btrfs_search_slot(trans, extent_root, &key, path, -1, 1);
1439 extent_slot = path->slots[0];
1442 leaf = path->nodes[0];
1443 ei = btrfs_item_ptr(leaf, extent_slot,
1444 struct btrfs_extent_item);
1445 refs = btrfs_extent_refs(leaf, ei);
1448 btrfs_set_extent_refs(leaf, ei, refs);
1450 btrfs_mark_buffer_dirty(leaf);
1452 if (refs == 0 && found_extent && path->slots[0] == extent_slot + 1) {
1453 /* if the back ref and the extent are next to each other
1454 * they get deleted below in one shot
1456 path->slots[0] = extent_slot;
1458 } else if (found_extent) {
1459 /* otherwise delete the extent back ref */
1460 ret = btrfs_del_item(trans, extent_root, path);
1462 /* if refs are 0, we need to setup the path for deletion */
1464 btrfs_release_path(extent_root, path);
1465 ret = btrfs_search_slot(trans, extent_root, &key, path,
1478 ret = pin_down_bytes(root, bytenr, num_bytes, 0);
1484 /* block accounting for super block */
1485 super_used = btrfs_super_bytes_used(&info->super_copy);
1486 btrfs_set_super_bytes_used(&info->super_copy,
1487 super_used - num_bytes);
1489 /* block accounting for root item */
1490 root_used = btrfs_root_used(&root->root_item);
1491 btrfs_set_root_used(&root->root_item,
1492 root_used - num_bytes);
1493 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
1498 ret = update_block_group(trans, root, bytenr, num_bytes, 0,
1502 btrfs_free_path(path);
1503 finish_current_insert(trans, extent_root);
1508 * find all the blocks marked as pending in the radix tree and remove
1509 * them from the extent map
1511 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
1512 btrfs_root *extent_root)
1518 struct extent_io_tree *pending_del;
1519 struct extent_io_tree *pinned_extents;
1521 pending_del = &extent_root->fs_info->pending_del;
1522 pinned_extents = &extent_root->fs_info->pinned_extents;
1525 ret = find_first_extent_bit(pending_del, 0, &start, &end,
1529 update_pinned_extents(extent_root, start, end + 1 - start, 1);
1530 clear_extent_bits(pending_del, start, end, EXTENT_LOCKED,
1532 ret = __free_extent(trans, extent_root,
1533 start, end + 1 - start,
1534 extent_root->root_key.objectid,
1543 * remove an extent from the root, returns 0 on success
1545 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root
1546 *root, u64 bytenr, u64 num_bytes,
1547 u64 root_objectid, u64 ref_generation,
1548 u64 owner_objectid, u64 owner_offset, int pin)
1550 struct btrfs_root *extent_root = root->fs_info->extent_root;
1554 WARN_ON(num_bytes < root->sectorsize);
1555 if (!root->ref_cows)
1558 if (root == extent_root) {
1559 pin_down_bytes(root, bytenr, num_bytes, 1);
1562 ret = __free_extent(trans, root, bytenr, num_bytes, root_objectid,
1563 ref_generation, owner_objectid, owner_offset,
1565 pending_ret = del_pending_extents(trans, root->fs_info->extent_root);
1566 return ret ? ret : pending_ret;
1569 static u64 stripe_align(struct btrfs_root *root, u64 val)
1571 u64 mask = ((u64)root->stripesize - 1);
1572 u64 ret = (val + mask) & ~mask;
1577 * walks the btree of allocated extents and find a hole of a given size.
1578 * The key ins is changed to record the hole:
1579 * ins->objectid == block start
1580 * ins->flags = BTRFS_EXTENT_ITEM_KEY
1581 * ins->offset == number of blocks
1582 * Any available blocks before search_start are skipped.
1584 static int noinline find_free_extent(struct btrfs_trans_handle *trans,
1585 struct btrfs_root *orig_root,
1586 u64 num_bytes, u64 empty_size,
1587 u64 search_start, u64 search_end,
1588 u64 hint_byte, struct btrfs_key *ins,
1589 u64 exclude_start, u64 exclude_nr,
1593 u64 orig_search_start;
1594 struct btrfs_root * root = orig_root->fs_info->extent_root;
1595 struct btrfs_fs_info *info = root->fs_info;
1596 u64 total_needed = num_bytes;
1597 u64 *last_ptr = NULL;
1598 struct btrfs_block_group_cache *block_group;
1601 int empty_cluster = 2 * 1024 * 1024;
1603 WARN_ON(num_bytes < root->sectorsize);
1604 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
1606 if (data & BTRFS_BLOCK_GROUP_METADATA) {
1607 last_ptr = &root->fs_info->last_alloc;
1608 empty_cluster = 256 * 1024;
1611 if ((data & BTRFS_BLOCK_GROUP_DATA) && btrfs_test_opt(root, SSD)) {
1612 last_ptr = &root->fs_info->last_data_alloc;
1617 hint_byte = *last_ptr;
1619 empty_size += empty_cluster;
1623 search_start = max(search_start, first_logical_byte(root, 0));
1624 orig_search_start = search_start;
1626 if (search_end == (u64)-1)
1627 search_end = btrfs_super_total_bytes(&info->super_copy);
1630 block_group = btrfs_lookup_block_group(info, hint_byte);
1632 hint_byte = search_start;
1633 block_group = btrfs_find_block_group(root, block_group,
1634 hint_byte, data, 1);
1635 if (last_ptr && *last_ptr == 0 && block_group)
1636 hint_byte = block_group->key.objectid;
1638 block_group = btrfs_find_block_group(root,
1640 search_start, data, 1);
1642 search_start = max(search_start, hint_byte);
1644 total_needed += empty_size;
1648 block_group = btrfs_lookup_block_group(info, search_start);
1650 block_group = btrfs_lookup_block_group(info,
1653 ret = find_search_start(root, &block_group, &search_start,
1654 total_needed, data);
1655 if (ret == -ENOSPC && last_ptr && *last_ptr) {
1657 block_group = btrfs_lookup_block_group(info,
1659 search_start = orig_search_start;
1660 ret = find_search_start(root, &block_group, &search_start,
1661 total_needed, data);
1668 if (last_ptr && *last_ptr && search_start != *last_ptr) {
1671 empty_size += empty_cluster;
1672 total_needed += empty_size;
1674 block_group = btrfs_lookup_block_group(info,
1676 search_start = orig_search_start;
1677 ret = find_search_start(root, &block_group,
1678 &search_start, total_needed, data);
1685 search_start = stripe_align(root, search_start);
1686 ins->objectid = search_start;
1687 ins->offset = num_bytes;
1689 if (ins->objectid + num_bytes >= search_end)
1692 if (ins->objectid + num_bytes >
1693 block_group->key.objectid + block_group->key.offset) {
1694 search_start = block_group->key.objectid +
1695 block_group->key.offset;
1699 if (test_range_bit(&info->extent_ins, ins->objectid,
1700 ins->objectid + num_bytes -1, EXTENT_LOCKED, 0)) {
1701 search_start = ins->objectid + num_bytes;
1705 if (test_range_bit(&info->pinned_extents, ins->objectid,
1706 ins->objectid + num_bytes -1, EXTENT_DIRTY, 0)) {
1707 search_start = ins->objectid + num_bytes;
1711 if (exclude_nr > 0 && (ins->objectid + num_bytes > exclude_start &&
1712 ins->objectid < exclude_start + exclude_nr)) {
1713 search_start = exclude_start + exclude_nr;
1717 if (!(data & BTRFS_BLOCK_GROUP_DATA)) {
1718 block_group = btrfs_lookup_block_group(info, ins->objectid);
1720 trans->block_group = block_group;
1722 ins->offset = num_bytes;
1724 *last_ptr = ins->objectid + ins->offset;
1726 btrfs_super_total_bytes(&root->fs_info->super_copy)) {
1733 if (search_start + num_bytes >= search_end) {
1735 search_start = orig_search_start;
1742 total_needed -= empty_size;
1747 block_group = btrfs_lookup_block_group(info, search_start);
1749 block_group = btrfs_find_block_group(root, block_group,
1750 search_start, data, 0);
1758 * finds a free extent and does all the dirty work required for allocation
1759 * returns the key for the extent through ins, and a tree buffer for
1760 * the first block of the extent through buf.
1762 * returns 0 if everything worked, non-zero otherwise.
1764 int btrfs_alloc_extent(struct btrfs_trans_handle *trans,
1765 struct btrfs_root *root,
1766 u64 num_bytes, u64 min_alloc_size,
1767 u64 root_objectid, u64 ref_generation,
1768 u64 owner, u64 owner_offset,
1769 u64 empty_size, u64 hint_byte,
1770 u64 search_end, struct btrfs_key *ins, u64 data)
1776 u64 search_start = 0;
1779 struct btrfs_fs_info *info = root->fs_info;
1780 struct btrfs_root *extent_root = info->extent_root;
1781 struct btrfs_extent_item *extent_item;
1782 struct btrfs_extent_ref *ref;
1783 struct btrfs_path *path;
1784 struct btrfs_key keys[2];
1787 alloc_profile = info->avail_data_alloc_bits &
1788 info->data_alloc_profile;
1789 data = BTRFS_BLOCK_GROUP_DATA | alloc_profile;
1790 } else if (root == root->fs_info->chunk_root) {
1791 alloc_profile = info->avail_system_alloc_bits &
1792 info->system_alloc_profile;
1793 data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile;
1795 alloc_profile = info->avail_metadata_alloc_bits &
1796 info->metadata_alloc_profile;
1797 data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile;
1800 data = reduce_alloc_profile(root, data);
1801 if (root->ref_cows) {
1802 if (!(data & BTRFS_BLOCK_GROUP_METADATA)) {
1803 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
1805 BTRFS_BLOCK_GROUP_METADATA |
1806 (info->metadata_alloc_profile &
1807 info->avail_metadata_alloc_bits));
1810 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
1811 num_bytes + 2 * 1024 * 1024, data);
1815 WARN_ON(num_bytes < root->sectorsize);
1816 ret = find_free_extent(trans, root, num_bytes, empty_size,
1817 search_start, search_end, hint_byte, ins,
1818 trans->alloc_exclude_start,
1819 trans->alloc_exclude_nr, data);
1821 if (ret == -ENOSPC && num_bytes > min_alloc_size) {
1822 num_bytes = num_bytes >> 1;
1823 num_bytes = max(num_bytes, min_alloc_size);
1827 printk("allocation failed flags %Lu\n", data);
1833 /* block accounting for super block */
1834 super_used = btrfs_super_bytes_used(&info->super_copy);
1835 btrfs_set_super_bytes_used(&info->super_copy, super_used + num_bytes);
1837 /* block accounting for root item */
1838 root_used = btrfs_root_used(&root->root_item);
1839 btrfs_set_root_used(&root->root_item, root_used + num_bytes);
1841 clear_extent_dirty(&root->fs_info->free_space_cache,
1842 ins->objectid, ins->objectid + ins->offset - 1,
1845 if (root == extent_root) {
1846 set_extent_bits(&root->fs_info->extent_ins, ins->objectid,
1847 ins->objectid + ins->offset - 1,
1848 EXTENT_LOCKED, GFP_NOFS);
1852 WARN_ON(trans->alloc_exclude_nr);
1853 trans->alloc_exclude_start = ins->objectid;
1854 trans->alloc_exclude_nr = ins->offset;
1856 memcpy(&keys[0], ins, sizeof(*ins));
1857 keys[1].offset = hash_extent_ref(root_objectid, ref_generation,
1858 owner, owner_offset);
1859 keys[1].objectid = ins->objectid;
1860 keys[1].type = BTRFS_EXTENT_REF_KEY;
1861 sizes[0] = sizeof(*extent_item);
1862 sizes[1] = sizeof(*ref);
1864 path = btrfs_alloc_path();
1867 ret = btrfs_insert_empty_items(trans, extent_root, path, keys,
1871 extent_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1872 struct btrfs_extent_item);
1873 btrfs_set_extent_refs(path->nodes[0], extent_item, 1);
1874 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
1875 struct btrfs_extent_ref);
1877 btrfs_set_ref_root(path->nodes[0], ref, root_objectid);
1878 btrfs_set_ref_generation(path->nodes[0], ref, ref_generation);
1879 btrfs_set_ref_objectid(path->nodes[0], ref, owner);
1880 btrfs_set_ref_offset(path->nodes[0], ref, owner_offset);
1882 btrfs_mark_buffer_dirty(path->nodes[0]);
1884 trans->alloc_exclude_start = 0;
1885 trans->alloc_exclude_nr = 0;
1886 btrfs_free_path(path);
1887 finish_current_insert(trans, extent_root);
1888 pending_ret = del_pending_extents(trans, extent_root);
1898 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1, 0);
1900 printk("update block group failed for %Lu %Lu\n",
1901 ins->objectid, ins->offset);
1908 * helper function to allocate a block for a given tree
1909 * returns the tree buffer or NULL.
1911 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
1912 struct btrfs_root *root,
1914 u64 root_objectid, u64 hint,
1920 ref_generation = trans->transid;
1925 return __btrfs_alloc_free_block(trans, root, blocksize, root_objectid,
1926 ref_generation, 0, 0, hint, empty_size);
1930 * helper function to allocate a block for a given tree
1931 * returns the tree buffer or NULL.
1933 struct extent_buffer *__btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
1934 struct btrfs_root *root,
1943 struct btrfs_key ins;
1945 struct extent_buffer *buf;
1947 ret = btrfs_alloc_extent(trans, root, blocksize, blocksize,
1948 root_objectid, ref_generation,
1949 level, first_objectid, empty_size, hint,
1953 return ERR_PTR(ret);
1955 buf = btrfs_find_create_tree_block(root, ins.objectid, blocksize);
1957 btrfs_free_extent(trans, root, ins.objectid, blocksize,
1958 root->root_key.objectid, ref_generation,
1960 return ERR_PTR(-ENOMEM);
1962 btrfs_set_header_generation(buf, trans->transid);
1963 clean_tree_block(trans, root, buf);
1964 btrfs_set_buffer_uptodate(buf);
1966 if (PageDirty(buf->first_page)) {
1967 printk("page %lu dirty\n", buf->first_page->index);
1971 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
1972 buf->start + buf->len - 1, GFP_NOFS);
1973 if (!btrfs_test_opt(root, SSD))
1974 btrfs_set_buffer_defrag(buf);
1975 trans->blocks_used++;
1979 static int noinline drop_leaf_ref(struct btrfs_trans_handle *trans,
1980 struct btrfs_root *root,
1981 struct extent_buffer *leaf)
1984 u64 leaf_generation;
1985 struct btrfs_key key;
1986 struct btrfs_file_extent_item *fi;
1991 BUG_ON(!btrfs_is_leaf(leaf));
1992 nritems = btrfs_header_nritems(leaf);
1993 leaf_owner = btrfs_header_owner(leaf);
1994 leaf_generation = btrfs_header_generation(leaf);
1996 for (i = 0; i < nritems; i++) {
1999 btrfs_item_key_to_cpu(leaf, &key, i);
2000 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2002 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
2003 if (btrfs_file_extent_type(leaf, fi) ==
2004 BTRFS_FILE_EXTENT_INLINE)
2007 * FIXME make sure to insert a trans record that
2008 * repeats the snapshot del on crash
2010 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
2011 if (disk_bytenr == 0)
2013 ret = btrfs_free_extent(trans, root, disk_bytenr,
2014 btrfs_file_extent_disk_num_bytes(leaf, fi),
2015 leaf_owner, leaf_generation,
2016 key.objectid, key.offset, 0);
2022 static void noinline reada_walk_down(struct btrfs_root *root,
2023 struct extent_buffer *node,
2036 nritems = btrfs_header_nritems(node);
2037 level = btrfs_header_level(node);
2041 for (i = slot; i < nritems && skipped < 32; i++) {
2042 bytenr = btrfs_node_blockptr(node, i);
2043 if (last && ((bytenr > last && bytenr - last > 32 * 1024) ||
2044 (last > bytenr && last - bytenr > 32 * 1024))) {
2048 blocksize = btrfs_level_size(root, level - 1);
2050 ret = lookup_extent_ref(NULL, root, bytenr,
2058 mutex_unlock(&root->fs_info->fs_mutex);
2059 ret = readahead_tree_block(root, bytenr, blocksize);
2060 last = bytenr + blocksize;
2062 mutex_lock(&root->fs_info->fs_mutex);
2069 * helper function for drop_snapshot, this walks down the tree dropping ref
2070 * counts as it goes.
2072 static int noinline walk_down_tree(struct btrfs_trans_handle *trans,
2073 struct btrfs_root *root,
2074 struct btrfs_path *path, int *level)
2079 struct extent_buffer *next;
2080 struct extent_buffer *cur;
2081 struct extent_buffer *parent;
2086 WARN_ON(*level < 0);
2087 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2088 ret = lookup_extent_ref(trans, root,
2089 path->nodes[*level]->start,
2090 path->nodes[*level]->len, &refs);
2096 * walk down to the last node level and free all the leaves
2098 while(*level >= 0) {
2099 WARN_ON(*level < 0);
2100 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2101 cur = path->nodes[*level];
2103 if (btrfs_header_level(cur) != *level)
2106 if (path->slots[*level] >=
2107 btrfs_header_nritems(cur))
2110 ret = drop_leaf_ref(trans, root, cur);
2114 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
2115 blocksize = btrfs_level_size(root, *level - 1);
2116 ret = lookup_extent_ref(trans, root, bytenr, blocksize, &refs);
2119 parent = path->nodes[*level];
2120 root_owner = btrfs_header_owner(parent);
2121 root_gen = btrfs_header_generation(parent);
2122 path->slots[*level]++;
2123 ret = btrfs_free_extent(trans, root, bytenr,
2124 blocksize, root_owner,
2129 next = btrfs_find_tree_block(root, bytenr, blocksize);
2130 if (!next || !btrfs_buffer_uptodate(next)) {
2131 free_extent_buffer(next);
2132 reada_walk_down(root, cur, path->slots[*level]);
2134 mutex_unlock(&root->fs_info->fs_mutex);
2135 next = read_tree_block(root, bytenr, blocksize);
2136 mutex_lock(&root->fs_info->fs_mutex);
2138 /* we've dropped the lock, double check */
2139 ret = lookup_extent_ref(trans, root, bytenr,
2143 parent = path->nodes[*level];
2144 root_owner = btrfs_header_owner(parent);
2145 root_gen = btrfs_header_generation(parent);
2147 path->slots[*level]++;
2148 free_extent_buffer(next);
2149 ret = btrfs_free_extent(trans, root, bytenr,
2157 btrfs_verify_block_csum(root, next);
2159 WARN_ON(*level <= 0);
2160 if (path->nodes[*level-1])
2161 free_extent_buffer(path->nodes[*level-1]);
2162 path->nodes[*level-1] = next;
2163 *level = btrfs_header_level(next);
2164 path->slots[*level] = 0;
2167 WARN_ON(*level < 0);
2168 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2170 if (path->nodes[*level] == root->node) {
2171 root_owner = root->root_key.objectid;
2172 parent = path->nodes[*level];
2174 parent = path->nodes[*level + 1];
2175 root_owner = btrfs_header_owner(parent);
2178 root_gen = btrfs_header_generation(parent);
2179 ret = btrfs_free_extent(trans, root, path->nodes[*level]->start,
2180 path->nodes[*level]->len,
2181 root_owner, root_gen, 0, 0, 1);
2182 free_extent_buffer(path->nodes[*level]);
2183 path->nodes[*level] = NULL;
2190 * helper for dropping snapshots. This walks back up the tree in the path
2191 * to find the first node higher up where we haven't yet gone through
2194 static int noinline walk_up_tree(struct btrfs_trans_handle *trans,
2195 struct btrfs_root *root,
2196 struct btrfs_path *path, int *level)
2200 struct btrfs_root_item *root_item = &root->root_item;
2205 for(i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
2206 slot = path->slots[i];
2207 if (slot < btrfs_header_nritems(path->nodes[i]) - 1) {
2208 struct extent_buffer *node;
2209 struct btrfs_disk_key disk_key;
2210 node = path->nodes[i];
2213 WARN_ON(*level == 0);
2214 btrfs_node_key(node, &disk_key, path->slots[i]);
2215 memcpy(&root_item->drop_progress,
2216 &disk_key, sizeof(disk_key));
2217 root_item->drop_level = i;
2220 if (path->nodes[*level] == root->node) {
2221 root_owner = root->root_key.objectid;
2223 btrfs_header_generation(path->nodes[*level]);
2225 struct extent_buffer *node;
2226 node = path->nodes[*level + 1];
2227 root_owner = btrfs_header_owner(node);
2228 root_gen = btrfs_header_generation(node);
2230 ret = btrfs_free_extent(trans, root,
2231 path->nodes[*level]->start,
2232 path->nodes[*level]->len,
2233 root_owner, root_gen, 0, 0, 1);
2235 free_extent_buffer(path->nodes[*level]);
2236 path->nodes[*level] = NULL;
2244 * drop the reference count on the tree rooted at 'snap'. This traverses
2245 * the tree freeing any blocks that have a ref count of zero after being
2248 int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root
2254 struct btrfs_path *path;
2257 struct btrfs_root_item *root_item = &root->root_item;
2259 path = btrfs_alloc_path();
2262 level = btrfs_header_level(root->node);
2264 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
2265 path->nodes[level] = root->node;
2266 extent_buffer_get(root->node);
2267 path->slots[level] = 0;
2269 struct btrfs_key key;
2270 struct btrfs_disk_key found_key;
2271 struct extent_buffer *node;
2273 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
2274 level = root_item->drop_level;
2275 path->lowest_level = level;
2276 wret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2281 node = path->nodes[level];
2282 btrfs_node_key(node, &found_key, path->slots[level]);
2283 WARN_ON(memcmp(&found_key, &root_item->drop_progress,
2284 sizeof(found_key)));
2287 wret = walk_down_tree(trans, root, path, &level);
2293 wret = walk_up_tree(trans, root, path, &level);
2301 for (i = 0; i <= orig_level; i++) {
2302 if (path->nodes[i]) {
2303 free_extent_buffer(path->nodes[i]);
2304 path->nodes[i] = NULL;
2308 btrfs_free_path(path);
2312 int btrfs_free_block_groups(struct btrfs_fs_info *info)
2319 ret = find_first_extent_bit(&info->block_group_cache, 0,
2320 &start, &end, (unsigned int)-1);
2323 ret = get_state_private(&info->block_group_cache, start, &ptr);
2325 kfree((void *)(unsigned long)ptr);
2326 clear_extent_bits(&info->block_group_cache, start,
2327 end, (unsigned int)-1, GFP_NOFS);
2330 ret = find_first_extent_bit(&info->free_space_cache, 0,
2331 &start, &end, EXTENT_DIRTY);
2334 clear_extent_dirty(&info->free_space_cache, start,
2340 static unsigned long calc_ra(unsigned long start, unsigned long last,
2343 return min(last, start + nr - 1);
2346 static int noinline relocate_inode_pages(struct inode *inode, u64 start,
2351 unsigned long last_index;
2354 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2355 struct file_ra_state *ra;
2356 unsigned long total_read = 0;
2357 unsigned long ra_pages;
2358 struct btrfs_trans_handle *trans;
2360 ra = kzalloc(sizeof(*ra), GFP_NOFS);
2362 mutex_lock(&inode->i_mutex);
2363 i = start >> PAGE_CACHE_SHIFT;
2364 last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
2366 ra_pages = BTRFS_I(inode)->root->fs_info->bdi.ra_pages;
2368 file_ra_state_init(ra, inode->i_mapping);
2370 for (; i <= last_index; i++) {
2371 if (total_read % ra_pages == 0) {
2372 btrfs_force_ra(inode->i_mapping, ra, NULL, i,
2373 calc_ra(i, last_index, ra_pages));
2376 if (((u64)i << PAGE_CACHE_SHIFT) > inode->i_size)
2377 goto truncate_racing;
2379 page = grab_cache_page(inode->i_mapping, i);
2383 if (!PageUptodate(page)) {
2384 btrfs_readpage(NULL, page);
2386 if (!PageUptodate(page)) {
2388 page_cache_release(page);
2392 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
2393 ClearPageDirty(page);
2395 cancel_dirty_page(page, PAGE_CACHE_SIZE);
2397 wait_on_page_writeback(page);
2398 set_page_extent_mapped(page);
2399 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
2400 page_end = page_start + PAGE_CACHE_SIZE - 1;
2402 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
2404 set_extent_delalloc(io_tree, page_start,
2405 page_end, GFP_NOFS);
2406 set_page_dirty(page);
2408 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
2410 page_cache_release(page);
2412 balance_dirty_pages_ratelimited_nr(inode->i_mapping,
2417 trans = btrfs_start_transaction(BTRFS_I(inode)->root, 1);
2419 btrfs_add_ordered_inode(inode);
2420 btrfs_end_transaction(trans, BTRFS_I(inode)->root);
2421 mark_inode_dirty(inode);
2423 mutex_unlock(&inode->i_mutex);
2427 vmtruncate(inode, inode->i_size);
2428 balance_dirty_pages_ratelimited_nr(inode->i_mapping,
2434 * The back references tell us which tree holds a ref on a block,
2435 * but it is possible for the tree root field in the reference to
2436 * reflect the original root before a snapshot was made. In this
2437 * case we should search through all the children of a given root
2438 * to find potential holders of references on a block.
2440 * Instead, we do something a little less fancy and just search
2441 * all the roots for a given key/block combination.
2443 static int find_root_for_ref(struct btrfs_root *root,
2444 struct btrfs_path *path,
2445 struct btrfs_key *key0,
2448 struct btrfs_root **found_root,
2451 struct btrfs_key root_location;
2452 struct btrfs_root *cur_root = *found_root;
2453 struct btrfs_file_extent_item *file_extent;
2454 u64 root_search_start = BTRFS_FS_TREE_OBJECTID;
2459 root_location.offset = (u64)-1;
2460 root_location.type = BTRFS_ROOT_ITEM_KEY;
2461 path->lowest_level = level;
2464 ret = btrfs_search_slot(NULL, cur_root, key0, path, 0, 0);
2466 if (ret == 0 && file_key) {
2467 struct extent_buffer *leaf = path->nodes[0];
2468 file_extent = btrfs_item_ptr(leaf, path->slots[0],
2469 struct btrfs_file_extent_item);
2470 if (btrfs_file_extent_type(leaf, file_extent) ==
2471 BTRFS_FILE_EXTENT_REG) {
2473 btrfs_file_extent_disk_bytenr(leaf,
2476 } else if (ret == 0) {
2477 if (path->nodes[level])
2478 found_bytenr = path->nodes[level]->start;
2481 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
2482 if (!path->nodes[i])
2484 free_extent_buffer(path->nodes[i]);
2485 path->nodes[i] = NULL;
2487 btrfs_release_path(cur_root, path);
2489 if (found_bytenr == bytenr) {
2490 *found_root = cur_root;
2494 ret = btrfs_search_root(root->fs_info->tree_root,
2495 root_search_start, &root_search_start);
2499 root_location.objectid = root_search_start;
2500 cur_root = btrfs_read_fs_root_no_name(root->fs_info,
2508 path->lowest_level = 0;
2513 * note, this releases the path
2515 static int noinline relocate_one_reference(struct btrfs_root *extent_root,
2516 struct btrfs_path *path,
2517 struct btrfs_key *extent_key)
2519 struct inode *inode;
2520 struct btrfs_root *found_root;
2521 struct btrfs_key root_location;
2522 struct btrfs_key found_key;
2523 struct btrfs_extent_ref *ref;
2531 ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
2532 struct btrfs_extent_ref);
2533 ref_root = btrfs_ref_root(path->nodes[0], ref);
2534 ref_gen = btrfs_ref_generation(path->nodes[0], ref);
2535 ref_objectid = btrfs_ref_objectid(path->nodes[0], ref);
2536 ref_offset = btrfs_ref_offset(path->nodes[0], ref);
2537 btrfs_release_path(extent_root, path);
2539 root_location.objectid = ref_root;
2541 root_location.offset = 0;
2543 root_location.offset = (u64)-1;
2544 root_location.type = BTRFS_ROOT_ITEM_KEY;
2546 found_root = btrfs_read_fs_root_no_name(extent_root->fs_info,
2548 BUG_ON(!found_root);
2550 if (ref_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
2551 found_key.objectid = ref_objectid;
2552 found_key.type = BTRFS_EXTENT_DATA_KEY;
2553 found_key.offset = ref_offset;
2556 ret = find_root_for_ref(extent_root, path, &found_key,
2557 level, 1, &found_root,
2558 extent_key->objectid);
2563 mutex_unlock(&extent_root->fs_info->fs_mutex);
2564 inode = btrfs_iget_locked(extent_root->fs_info->sb,
2565 ref_objectid, found_root);
2566 if (inode->i_state & I_NEW) {
2567 /* the inode and parent dir are two different roots */
2568 BTRFS_I(inode)->root = found_root;
2569 BTRFS_I(inode)->location.objectid = ref_objectid;
2570 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
2571 BTRFS_I(inode)->location.offset = 0;
2572 btrfs_read_locked_inode(inode);
2573 unlock_new_inode(inode);
2576 /* this can happen if the reference is not against
2577 * the latest version of the tree root
2579 if (is_bad_inode(inode)) {
2580 mutex_lock(&extent_root->fs_info->fs_mutex);
2583 relocate_inode_pages(inode, ref_offset, extent_key->offset);
2585 mutex_lock(&extent_root->fs_info->fs_mutex);
2587 struct btrfs_trans_handle *trans;
2588 struct extent_buffer *eb;
2591 eb = read_tree_block(found_root, extent_key->objectid,
2592 extent_key->offset);
2593 level = btrfs_header_level(eb);
2596 btrfs_item_key_to_cpu(eb, &found_key, 0);
2598 btrfs_node_key_to_cpu(eb, &found_key, 0);
2600 free_extent_buffer(eb);
2602 ret = find_root_for_ref(extent_root, path, &found_key,
2603 level, 0, &found_root,
2604 extent_key->objectid);
2609 trans = btrfs_start_transaction(found_root, 1);
2611 path->lowest_level = level;
2613 ret = btrfs_search_slot(trans, found_root, &found_key, path,
2615 path->lowest_level = 0;
2616 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
2617 if (!path->nodes[i])
2619 free_extent_buffer(path->nodes[i]);
2620 path->nodes[i] = NULL;
2622 btrfs_release_path(found_root, path);
2623 btrfs_end_transaction(trans, found_root);
2630 static int noinline del_extent_zero(struct btrfs_root *extent_root,
2631 struct btrfs_path *path,
2632 struct btrfs_key *extent_key)
2635 struct btrfs_trans_handle *trans;
2637 trans = btrfs_start_transaction(extent_root, 1);
2638 ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1);
2645 ret = btrfs_del_item(trans, extent_root, path);
2647 btrfs_end_transaction(trans, extent_root);
2651 static int noinline relocate_one_extent(struct btrfs_root *extent_root,
2652 struct btrfs_path *path,
2653 struct btrfs_key *extent_key)
2655 struct btrfs_key key;
2656 struct btrfs_key found_key;
2657 struct extent_buffer *leaf;
2662 if (extent_key->objectid == 0) {
2663 ret = del_extent_zero(extent_root, path, extent_key);
2666 key.objectid = extent_key->objectid;
2667 key.type = BTRFS_EXTENT_REF_KEY;
2671 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2677 leaf = path->nodes[0];
2678 nritems = btrfs_header_nritems(leaf);
2679 if (path->slots[0] == nritems) {
2680 ret = btrfs_next_leaf(extent_root, path);
2687 leaf = path->nodes[0];
2690 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2691 if (found_key.objectid != extent_key->objectid) {
2695 if (found_key.type != BTRFS_EXTENT_REF_KEY) {
2699 key.offset = found_key.offset + 1;
2700 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2702 ret = relocate_one_reference(extent_root, path, extent_key);
2708 btrfs_release_path(extent_root, path);
2712 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
2715 u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
2716 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
2718 num_devices = root->fs_info->fs_devices->num_devices;
2719 if (num_devices == 1) {
2720 stripped |= BTRFS_BLOCK_GROUP_DUP;
2721 stripped = flags & ~stripped;
2723 /* turn raid0 into single device chunks */
2724 if (flags & BTRFS_BLOCK_GROUP_RAID0)
2727 /* turn mirroring into duplication */
2728 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
2729 BTRFS_BLOCK_GROUP_RAID10))
2730 return stripped | BTRFS_BLOCK_GROUP_DUP;
2733 /* they already had raid on here, just return */
2734 if (flags & stripped)
2737 stripped |= BTRFS_BLOCK_GROUP_DUP;
2738 stripped = flags & ~stripped;
2740 /* switch duplicated blocks with raid1 */
2741 if (flags & BTRFS_BLOCK_GROUP_DUP)
2742 return stripped | BTRFS_BLOCK_GROUP_RAID1;
2744 /* turn single device chunks into raid0 */
2745 return stripped | BTRFS_BLOCK_GROUP_RAID0;
2750 int btrfs_shrink_extent_tree(struct btrfs_root *root, u64 shrink_start)
2752 struct btrfs_trans_handle *trans;
2753 struct btrfs_root *tree_root = root->fs_info->tree_root;
2754 struct btrfs_path *path;
2757 u64 shrink_last_byte;
2758 u64 new_alloc_flags;
2759 struct btrfs_block_group_cache *shrink_block_group;
2760 struct btrfs_fs_info *info = root->fs_info;
2761 struct btrfs_key key;
2762 struct btrfs_key found_key;
2763 struct extent_buffer *leaf;
2768 shrink_block_group = btrfs_lookup_block_group(root->fs_info,
2770 BUG_ON(!shrink_block_group);
2772 shrink_last_byte = shrink_start + shrink_block_group->key.offset;
2774 shrink_block_group->space_info->total_bytes -=
2775 shrink_block_group->key.offset;
2776 path = btrfs_alloc_path();
2777 root = root->fs_info->extent_root;
2781 if (btrfs_block_group_used(&shrink_block_group->item) > 0) {
2782 trans = btrfs_start_transaction(root, 1);
2783 new_alloc_flags = update_block_group_flags(root,
2784 shrink_block_group->flags);
2785 do_chunk_alloc(trans, root->fs_info->extent_root,
2786 btrfs_block_group_used(&shrink_block_group->item) +
2787 2 * 1024 * 1024, new_alloc_flags);
2788 btrfs_end_transaction(trans, root);
2790 shrink_block_group->ro = 1;
2794 key.objectid = shrink_start;
2797 cur_byte = key.objectid;
2799 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2803 ret = btrfs_previous_item(root, path, 0, BTRFS_EXTENT_ITEM_KEY);
2808 leaf = path->nodes[0];
2809 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2810 if (found_key.objectid + found_key.offset > shrink_start &&
2811 found_key.objectid < shrink_last_byte) {
2812 cur_byte = found_key.objectid;
2813 key.objectid = cur_byte;
2816 btrfs_release_path(root, path);
2819 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2823 leaf = path->nodes[0];
2824 nritems = btrfs_header_nritems(leaf);
2826 if (path->slots[0] >= nritems) {
2827 ret = btrfs_next_leaf(root, path);
2834 leaf = path->nodes[0];
2835 nritems = btrfs_header_nritems(leaf);
2838 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2840 if (found_key.objectid >= shrink_last_byte)
2843 if (progress && need_resched()) {
2844 memcpy(&key, &found_key, sizeof(key));
2845 mutex_unlock(&root->fs_info->fs_mutex);
2847 mutex_lock(&root->fs_info->fs_mutex);
2848 btrfs_release_path(root, path);
2849 btrfs_search_slot(NULL, root, &key, path, 0, 0);
2855 if (btrfs_key_type(&found_key) != BTRFS_EXTENT_ITEM_KEY ||
2856 found_key.objectid + found_key.offset <= cur_byte) {
2862 cur_byte = found_key.objectid + found_key.offset;
2863 key.objectid = cur_byte;
2864 btrfs_release_path(root, path);
2865 ret = relocate_one_extent(root, path, &found_key);
2868 btrfs_release_path(root, path);
2870 if (total_found > 0) {
2871 trans = btrfs_start_transaction(tree_root, 1);
2872 btrfs_commit_transaction(trans, tree_root);
2874 mutex_unlock(&root->fs_info->fs_mutex);
2875 btrfs_clean_old_snapshots(tree_root);
2876 mutex_lock(&root->fs_info->fs_mutex);
2878 trans = btrfs_start_transaction(tree_root, 1);
2879 btrfs_commit_transaction(trans, tree_root);
2884 * we've freed all the extents, now remove the block
2885 * group item from the tree
2887 trans = btrfs_start_transaction(root, 1);
2888 memcpy(&key, &shrink_block_group->key, sizeof(key));
2890 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2896 leaf = path->nodes[0];
2897 nritems = btrfs_header_nritems(leaf);
2898 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2899 kfree(shrink_block_group);
2901 clear_extent_bits(&info->block_group_cache, found_key.objectid,
2902 found_key.objectid + found_key.offset - 1,
2903 (unsigned int)-1, GFP_NOFS);
2905 btrfs_del_item(trans, root, path);
2906 clear_extent_dirty(&info->free_space_cache,
2907 shrink_start, shrink_last_byte - 1,
2909 btrfs_commit_transaction(trans, root);
2911 btrfs_free_path(path);
2915 int find_first_block_group(struct btrfs_root *root, struct btrfs_path *path,
2916 struct btrfs_key *key)
2919 struct btrfs_key found_key;
2920 struct extent_buffer *leaf;
2923 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
2927 slot = path->slots[0];
2928 leaf = path->nodes[0];
2929 if (slot >= btrfs_header_nritems(leaf)) {
2930 ret = btrfs_next_leaf(root, path);
2937 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2939 if (found_key.objectid >= key->objectid &&
2940 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY)
2949 int btrfs_read_block_groups(struct btrfs_root *root)
2951 struct btrfs_path *path;
2954 struct btrfs_block_group_cache *cache;
2955 struct btrfs_fs_info *info = root->fs_info;
2956 struct btrfs_space_info *space_info;
2957 struct extent_io_tree *block_group_cache;
2958 struct btrfs_key key;
2959 struct btrfs_key found_key;
2960 struct extent_buffer *leaf;
2962 block_group_cache = &info->block_group_cache;
2963 root = info->extent_root;
2966 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
2967 path = btrfs_alloc_path();
2972 ret = find_first_block_group(root, path, &key);
2980 leaf = path->nodes[0];
2981 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2982 cache = kzalloc(sizeof(*cache), GFP_NOFS);
2988 read_extent_buffer(leaf, &cache->item,
2989 btrfs_item_ptr_offset(leaf, path->slots[0]),
2990 sizeof(cache->item));
2991 memcpy(&cache->key, &found_key, sizeof(found_key));
2993 key.objectid = found_key.objectid + found_key.offset;
2994 btrfs_release_path(root, path);
2995 cache->flags = btrfs_block_group_flags(&cache->item);
2997 if (cache->flags & BTRFS_BLOCK_GROUP_DATA) {
2998 bit = BLOCK_GROUP_DATA;
2999 } else if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) {
3000 bit = BLOCK_GROUP_SYSTEM;
3001 } else if (cache->flags & BTRFS_BLOCK_GROUP_METADATA) {
3002 bit = BLOCK_GROUP_METADATA;
3004 set_avail_alloc_bits(info, cache->flags);
3006 ret = update_space_info(info, cache->flags, found_key.offset,
3007 btrfs_block_group_used(&cache->item),
3010 cache->space_info = space_info;
3012 /* use EXTENT_LOCKED to prevent merging */
3013 set_extent_bits(block_group_cache, found_key.objectid,
3014 found_key.objectid + found_key.offset - 1,
3015 bit | EXTENT_LOCKED, GFP_NOFS);
3016 set_state_private(block_group_cache, found_key.objectid,
3017 (unsigned long)cache);
3020 btrfs_super_total_bytes(&info->super_copy))
3025 btrfs_free_path(path);
3029 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
3030 struct btrfs_root *root, u64 bytes_used,
3031 u64 type, u64 chunk_objectid, u64 chunk_offset,
3036 struct btrfs_root *extent_root;
3037 struct btrfs_block_group_cache *cache;
3038 struct extent_io_tree *block_group_cache;
3040 extent_root = root->fs_info->extent_root;
3041 block_group_cache = &root->fs_info->block_group_cache;
3043 cache = kzalloc(sizeof(*cache), GFP_NOFS);
3045 cache->key.objectid = chunk_offset;
3046 cache->key.offset = size;
3048 btrfs_set_key_type(&cache->key, BTRFS_BLOCK_GROUP_ITEM_KEY);
3049 memset(&cache->item, 0, sizeof(cache->item));
3050 btrfs_set_block_group_used(&cache->item, bytes_used);
3051 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
3052 cache->flags = type;
3053 btrfs_set_block_group_flags(&cache->item, type);
3055 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
3056 &cache->space_info);
3059 bit = block_group_state_bits(type);
3060 set_extent_bits(block_group_cache, chunk_offset,
3061 chunk_offset + size - 1,
3062 bit | EXTENT_LOCKED, GFP_NOFS);
3064 set_state_private(block_group_cache, chunk_offset,
3065 (unsigned long)cache);
3066 ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
3067 sizeof(cache->item));
3070 finish_current_insert(trans, extent_root);
3071 ret = del_pending_extents(trans, extent_root);
3073 set_avail_alloc_bits(extent_root->fs_info, type);