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;
738 struct btrfs_key key;
739 struct btrfs_key found_key;
740 struct extent_buffer *l;
741 struct btrfs_extent_item *item;
742 struct btrfs_extent_ref *ref_item;
745 path = btrfs_alloc_path();
748 bytenr = first_extent;
750 bytenr = count_path->nodes[level]->start;
753 key.objectid = bytenr;
756 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
757 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
763 btrfs_item_key_to_cpu(l, &found_key, path->slots[0]);
765 if (found_key.objectid != bytenr ||
766 found_key.type != BTRFS_EXTENT_ITEM_KEY) {
770 item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
771 extent_refs = btrfs_extent_refs(l, item);
774 nritems = btrfs_header_nritems(l);
775 if (path->slots[0] >= nritems) {
776 ret = btrfs_next_leaf(extent_root, path);
781 btrfs_item_key_to_cpu(l, &found_key, path->slots[0]);
782 if (found_key.objectid != bytenr)
785 if (found_key.type != BTRFS_EXTENT_REF_KEY) {
791 ref_item = btrfs_item_ptr(l, path->slots[0],
792 struct btrfs_extent_ref);
793 found_objectid = btrfs_ref_root(l, ref_item);
795 if (found_objectid != root_objectid) {
800 found_owner = btrfs_ref_objectid(l, ref_item);
801 if (found_owner != expected_owner) {
806 * nasty. we don't count a reference held by
807 * the running transaction. This allows nodatacow
808 * to avoid cow most of the time
810 if (found_owner >= BTRFS_FIRST_FREE_OBJECTID &&
811 btrfs_ref_generation(l, ref_item) ==
812 root->fs_info->generation) {
820 * if there is more than one reference against a data extent,
821 * we have to assume the other ref is another snapshot
823 if (level == -1 && extent_refs > 1) {
827 if (cur_count == 0) {
831 if (level >= 0 && root->node == count_path->nodes[level])
834 btrfs_release_path(root, path);
838 btrfs_free_path(path);
841 int btrfs_inc_root_ref(struct btrfs_trans_handle *trans,
842 struct btrfs_root *root, u64 owner_objectid)
848 struct btrfs_disk_key disk_key;
850 level = btrfs_header_level(root->node);
851 generation = trans->transid;
852 nritems = btrfs_header_nritems(root->node);
855 btrfs_item_key(root->node, &disk_key, 0);
857 btrfs_node_key(root->node, &disk_key, 0);
858 key_objectid = btrfs_disk_key_objectid(&disk_key);
862 return btrfs_inc_extent_ref(trans, root, root->node->start,
863 root->node->len, owner_objectid,
864 generation, level, key_objectid);
867 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
868 struct extent_buffer *buf)
872 struct btrfs_key key;
873 struct btrfs_file_extent_item *fi;
882 level = btrfs_header_level(buf);
883 nritems = btrfs_header_nritems(buf);
884 for (i = 0; i < nritems; i++) {
887 btrfs_item_key_to_cpu(buf, &key, i);
888 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
890 fi = btrfs_item_ptr(buf, i,
891 struct btrfs_file_extent_item);
892 if (btrfs_file_extent_type(buf, fi) ==
893 BTRFS_FILE_EXTENT_INLINE)
895 disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
896 if (disk_bytenr == 0)
898 ret = btrfs_inc_extent_ref(trans, root, disk_bytenr,
899 btrfs_file_extent_disk_num_bytes(buf, fi),
900 root->root_key.objectid, trans->transid,
901 key.objectid, key.offset);
907 bytenr = btrfs_node_blockptr(buf, i);
908 btrfs_node_key_to_cpu(buf, &key, i);
909 ret = btrfs_inc_extent_ref(trans, root, bytenr,
910 btrfs_level_size(root, level - 1),
911 root->root_key.objectid,
913 level - 1, key.objectid);
924 for (i =0; i < faili; i++) {
927 btrfs_item_key_to_cpu(buf, &key, i);
928 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
930 fi = btrfs_item_ptr(buf, i,
931 struct btrfs_file_extent_item);
932 if (btrfs_file_extent_type(buf, fi) ==
933 BTRFS_FILE_EXTENT_INLINE)
935 disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
936 if (disk_bytenr == 0)
938 err = btrfs_free_extent(trans, root, disk_bytenr,
939 btrfs_file_extent_disk_num_bytes(buf,
943 bytenr = btrfs_node_blockptr(buf, i);
944 err = btrfs_free_extent(trans, root, bytenr,
945 btrfs_level_size(root, level - 1), 0);
953 static int write_one_cache_group(struct btrfs_trans_handle *trans,
954 struct btrfs_root *root,
955 struct btrfs_path *path,
956 struct btrfs_block_group_cache *cache)
960 struct btrfs_root *extent_root = root->fs_info->extent_root;
962 struct extent_buffer *leaf;
964 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
969 leaf = path->nodes[0];
970 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
971 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
972 btrfs_mark_buffer_dirty(leaf);
973 btrfs_release_path(extent_root, path);
975 finish_current_insert(trans, extent_root);
976 pending_ret = del_pending_extents(trans, extent_root);
985 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
986 struct btrfs_root *root)
988 struct extent_io_tree *block_group_cache;
989 struct btrfs_block_group_cache *cache;
993 struct btrfs_path *path;
999 block_group_cache = &root->fs_info->block_group_cache;
1000 path = btrfs_alloc_path();
1005 ret = find_first_extent_bit(block_group_cache, last,
1006 &start, &end, BLOCK_GROUP_DIRTY);
1011 ret = get_state_private(block_group_cache, start, &ptr);
1014 cache = (struct btrfs_block_group_cache *)(unsigned long)ptr;
1015 err = write_one_cache_group(trans, root,
1018 * if we fail to write the cache group, we want
1019 * to keep it marked dirty in hopes that a later
1026 clear_extent_bits(block_group_cache, start, end,
1027 BLOCK_GROUP_DIRTY, GFP_NOFS);
1029 btrfs_free_path(path);
1033 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
1036 struct list_head *head = &info->space_info;
1037 struct list_head *cur;
1038 struct btrfs_space_info *found;
1039 list_for_each(cur, head) {
1040 found = list_entry(cur, struct btrfs_space_info, list);
1041 if (found->flags == flags)
1048 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
1049 u64 total_bytes, u64 bytes_used,
1050 struct btrfs_space_info **space_info)
1052 struct btrfs_space_info *found;
1054 found = __find_space_info(info, flags);
1056 found->total_bytes += total_bytes;
1057 found->bytes_used += bytes_used;
1059 WARN_ON(found->total_bytes < found->bytes_used);
1060 *space_info = found;
1063 found = kmalloc(sizeof(*found), GFP_NOFS);
1067 list_add(&found->list, &info->space_info);
1068 found->flags = flags;
1069 found->total_bytes = total_bytes;
1070 found->bytes_used = bytes_used;
1071 found->bytes_pinned = 0;
1073 *space_info = found;
1077 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
1079 u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
1080 BTRFS_BLOCK_GROUP_RAID1 |
1081 BTRFS_BLOCK_GROUP_RAID10 |
1082 BTRFS_BLOCK_GROUP_DUP);
1084 if (flags & BTRFS_BLOCK_GROUP_DATA)
1085 fs_info->avail_data_alloc_bits |= extra_flags;
1086 if (flags & BTRFS_BLOCK_GROUP_METADATA)
1087 fs_info->avail_metadata_alloc_bits |= extra_flags;
1088 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
1089 fs_info->avail_system_alloc_bits |= extra_flags;
1093 static u64 reduce_alloc_profile(struct btrfs_root *root, u64 flags)
1095 u64 num_devices = root->fs_info->fs_devices->num_devices;
1097 if (num_devices == 1)
1098 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
1099 if (num_devices < 4)
1100 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
1102 if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
1103 (flags & (BTRFS_BLOCK_GROUP_RAID1 |
1104 BTRFS_BLOCK_GROUP_RAID10))) {
1105 flags &= ~BTRFS_BLOCK_GROUP_DUP;
1108 if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
1109 (flags & BTRFS_BLOCK_GROUP_RAID10)) {
1110 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
1113 if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
1114 ((flags & BTRFS_BLOCK_GROUP_RAID1) |
1115 (flags & BTRFS_BLOCK_GROUP_RAID10) |
1116 (flags & BTRFS_BLOCK_GROUP_DUP)))
1117 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
1121 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
1122 struct btrfs_root *extent_root, u64 alloc_bytes,
1125 struct btrfs_space_info *space_info;
1131 flags = reduce_alloc_profile(extent_root, flags);
1133 space_info = __find_space_info(extent_root->fs_info, flags);
1135 ret = update_space_info(extent_root->fs_info, flags,
1139 BUG_ON(!space_info);
1141 if (space_info->full)
1144 thresh = div_factor(space_info->total_bytes, 6);
1145 if ((space_info->bytes_used + space_info->bytes_pinned + alloc_bytes) <
1149 ret = btrfs_alloc_chunk(trans, extent_root, &start, &num_bytes, flags);
1150 if (ret == -ENOSPC) {
1151 printk("space info full %Lu\n", flags);
1152 space_info->full = 1;
1158 ret = btrfs_make_block_group(trans, extent_root, 0, flags,
1159 BTRFS_FIRST_CHUNK_TREE_OBJECTID, start, num_bytes);
1165 static int update_block_group(struct btrfs_trans_handle *trans,
1166 struct btrfs_root *root,
1167 u64 bytenr, u64 num_bytes, int alloc,
1170 struct btrfs_block_group_cache *cache;
1171 struct btrfs_fs_info *info = root->fs_info;
1172 u64 total = num_bytes;
1179 cache = btrfs_lookup_block_group(info, bytenr);
1183 byte_in_group = bytenr - cache->key.objectid;
1184 WARN_ON(byte_in_group > cache->key.offset);
1185 start = cache->key.objectid;
1186 end = start + cache->key.offset - 1;
1187 set_extent_bits(&info->block_group_cache, start, end,
1188 BLOCK_GROUP_DIRTY, GFP_NOFS);
1190 old_val = btrfs_block_group_used(&cache->item);
1191 num_bytes = min(total, cache->key.offset - byte_in_group);
1193 old_val += num_bytes;
1194 cache->space_info->bytes_used += num_bytes;
1196 old_val -= num_bytes;
1197 cache->space_info->bytes_used -= num_bytes;
1199 set_extent_dirty(&info->free_space_cache,
1200 bytenr, bytenr + num_bytes - 1,
1204 btrfs_set_block_group_used(&cache->item, old_val);
1206 bytenr += num_bytes;
1211 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
1216 ret = find_first_extent_bit(&root->fs_info->block_group_cache,
1217 search_start, &start, &end,
1218 BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA |
1219 BLOCK_GROUP_SYSTEM);
1226 static int update_pinned_extents(struct btrfs_root *root,
1227 u64 bytenr, u64 num, int pin)
1230 struct btrfs_block_group_cache *cache;
1231 struct btrfs_fs_info *fs_info = root->fs_info;
1234 set_extent_dirty(&fs_info->pinned_extents,
1235 bytenr, bytenr + num - 1, GFP_NOFS);
1237 clear_extent_dirty(&fs_info->pinned_extents,
1238 bytenr, bytenr + num - 1, GFP_NOFS);
1241 cache = btrfs_lookup_block_group(fs_info, bytenr);
1243 u64 first = first_logical_byte(root, bytenr);
1244 WARN_ON(first < bytenr);
1245 len = min(first - bytenr, num);
1247 len = min(num, cache->key.offset -
1248 (bytenr - cache->key.objectid));
1252 cache->pinned += len;
1253 cache->space_info->bytes_pinned += len;
1255 fs_info->total_pinned += len;
1258 cache->pinned -= len;
1259 cache->space_info->bytes_pinned -= len;
1261 fs_info->total_pinned -= len;
1269 int btrfs_copy_pinned(struct btrfs_root *root, struct extent_io_tree *copy)
1274 struct extent_io_tree *pinned_extents = &root->fs_info->pinned_extents;
1278 ret = find_first_extent_bit(pinned_extents, last,
1279 &start, &end, EXTENT_DIRTY);
1282 set_extent_dirty(copy, start, end, GFP_NOFS);
1288 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
1289 struct btrfs_root *root,
1290 struct extent_io_tree *unpin)
1295 struct extent_io_tree *free_space_cache;
1296 free_space_cache = &root->fs_info->free_space_cache;
1299 ret = find_first_extent_bit(unpin, 0, &start, &end,
1303 update_pinned_extents(root, start, end + 1 - start, 0);
1304 clear_extent_dirty(unpin, start, end, GFP_NOFS);
1305 set_extent_dirty(free_space_cache, start, end, GFP_NOFS);
1310 static int finish_current_insert(struct btrfs_trans_handle *trans,
1311 struct btrfs_root *extent_root)
1315 struct btrfs_fs_info *info = extent_root->fs_info;
1316 struct extent_buffer *eb;
1317 struct btrfs_path *path;
1318 struct btrfs_key ins;
1319 struct btrfs_disk_key first;
1320 struct btrfs_extent_item extent_item;
1325 btrfs_set_stack_extent_refs(&extent_item, 1);
1326 btrfs_set_key_type(&ins, BTRFS_EXTENT_ITEM_KEY);
1327 path = btrfs_alloc_path();
1330 ret = find_first_extent_bit(&info->extent_ins, 0, &start,
1331 &end, EXTENT_LOCKED);
1335 ins.objectid = start;
1336 ins.offset = end + 1 - start;
1337 err = btrfs_insert_item(trans, extent_root, &ins,
1338 &extent_item, sizeof(extent_item));
1339 clear_extent_bits(&info->extent_ins, start, end, EXTENT_LOCKED,
1341 eb = read_tree_block(extent_root, ins.objectid, ins.offset);
1342 level = btrfs_header_level(eb);
1344 btrfs_item_key(eb, &first, 0);
1346 btrfs_node_key(eb, &first, 0);
1348 err = btrfs_insert_extent_backref(trans, extent_root, path,
1349 start, extent_root->root_key.objectid,
1351 btrfs_disk_key_objectid(&first));
1353 free_extent_buffer(eb);
1355 btrfs_free_path(path);
1359 static int pin_down_bytes(struct btrfs_root *root, u64 bytenr, u32 num_bytes,
1363 struct extent_buffer *buf;
1366 buf = btrfs_find_tree_block(root, bytenr, num_bytes);
1368 if (btrfs_buffer_uptodate(buf)) {
1370 root->fs_info->running_transaction->transid;
1371 u64 header_transid =
1372 btrfs_header_generation(buf);
1373 if (header_transid == transid &&
1374 !btrfs_header_flag(buf,
1375 BTRFS_HEADER_FLAG_WRITTEN)) {
1376 clean_tree_block(NULL, root, buf);
1377 free_extent_buffer(buf);
1381 free_extent_buffer(buf);
1383 update_pinned_extents(root, bytenr, num_bytes, 1);
1385 set_extent_bits(&root->fs_info->pending_del,
1386 bytenr, bytenr + num_bytes - 1,
1387 EXTENT_LOCKED, GFP_NOFS);
1394 * remove an extent from the root, returns 0 on success
1396 static int __free_extent(struct btrfs_trans_handle *trans, struct btrfs_root
1397 *root, u64 bytenr, u64 num_bytes,
1398 u64 root_objectid, u64 ref_generation,
1399 u64 owner_objectid, u64 owner_offset, int pin,
1402 struct btrfs_path *path;
1403 struct btrfs_key key;
1404 struct btrfs_fs_info *info = root->fs_info;
1405 struct btrfs_root *extent_root = info->extent_root;
1406 struct extent_buffer *leaf;
1408 int extent_slot = 0;
1409 int found_extent = 0;
1411 struct btrfs_extent_item *ei;
1414 key.objectid = bytenr;
1415 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
1416 key.offset = num_bytes;
1417 path = btrfs_alloc_path();
1422 ret = lookup_extent_backref(trans, extent_root, path,
1423 bytenr, root_objectid,
1425 owner_objectid, owner_offset, 1);
1427 struct btrfs_key found_key;
1428 extent_slot = path->slots[0];
1429 while(extent_slot > 0) {
1431 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1433 if (found_key.objectid != bytenr)
1435 if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
1436 found_key.offset == num_bytes) {
1440 if (path->slots[0] - extent_slot > 5)
1444 ret = btrfs_del_item(trans, extent_root, path);
1446 btrfs_print_leaf(extent_root, path->nodes[0]);
1448 printk("Unable to find ref byte nr %Lu root %Lu "
1449 " gen %Lu owner %Lu offset %Lu\n", bytenr,
1450 root_objectid, ref_generation, owner_objectid,
1453 if (!found_extent) {
1454 btrfs_release_path(extent_root, path);
1455 ret = btrfs_search_slot(trans, extent_root, &key, path, -1, 1);
1459 extent_slot = path->slots[0];
1462 leaf = path->nodes[0];
1463 ei = btrfs_item_ptr(leaf, extent_slot,
1464 struct btrfs_extent_item);
1465 refs = btrfs_extent_refs(leaf, ei);
1468 btrfs_set_extent_refs(leaf, ei, refs);
1470 btrfs_mark_buffer_dirty(leaf);
1472 if (refs == 0 && found_extent && path->slots[0] == extent_slot + 1) {
1473 /* if the back ref and the extent are next to each other
1474 * they get deleted below in one shot
1476 path->slots[0] = extent_slot;
1478 } else if (found_extent) {
1479 /* otherwise delete the extent back ref */
1480 ret = btrfs_del_item(trans, extent_root, path);
1482 /* if refs are 0, we need to setup the path for deletion */
1484 btrfs_release_path(extent_root, path);
1485 ret = btrfs_search_slot(trans, extent_root, &key, path,
1498 ret = pin_down_bytes(root, bytenr, num_bytes, 0);
1504 /* block accounting for super block */
1505 super_used = btrfs_super_bytes_used(&info->super_copy);
1506 btrfs_set_super_bytes_used(&info->super_copy,
1507 super_used - num_bytes);
1509 /* block accounting for root item */
1510 root_used = btrfs_root_used(&root->root_item);
1511 btrfs_set_root_used(&root->root_item,
1512 root_used - num_bytes);
1513 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
1518 ret = update_block_group(trans, root, bytenr, num_bytes, 0,
1522 btrfs_free_path(path);
1523 finish_current_insert(trans, extent_root);
1528 * find all the blocks marked as pending in the radix tree and remove
1529 * them from the extent map
1531 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
1532 btrfs_root *extent_root)
1538 struct extent_io_tree *pending_del;
1539 struct extent_io_tree *pinned_extents;
1541 pending_del = &extent_root->fs_info->pending_del;
1542 pinned_extents = &extent_root->fs_info->pinned_extents;
1545 ret = find_first_extent_bit(pending_del, 0, &start, &end,
1549 update_pinned_extents(extent_root, start, end + 1 - start, 1);
1550 clear_extent_bits(pending_del, start, end, EXTENT_LOCKED,
1552 ret = __free_extent(trans, extent_root,
1553 start, end + 1 - start,
1554 extent_root->root_key.objectid,
1563 * remove an extent from the root, returns 0 on success
1565 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root
1566 *root, u64 bytenr, u64 num_bytes,
1567 u64 root_objectid, u64 ref_generation,
1568 u64 owner_objectid, u64 owner_offset, int pin)
1570 struct btrfs_root *extent_root = root->fs_info->extent_root;
1574 WARN_ON(num_bytes < root->sectorsize);
1575 if (!root->ref_cows)
1578 if (root == extent_root) {
1579 pin_down_bytes(root, bytenr, num_bytes, 1);
1582 ret = __free_extent(trans, root, bytenr, num_bytes, root_objectid,
1583 ref_generation, owner_objectid, owner_offset,
1585 pending_ret = del_pending_extents(trans, root->fs_info->extent_root);
1586 return ret ? ret : pending_ret;
1589 static u64 stripe_align(struct btrfs_root *root, u64 val)
1591 u64 mask = ((u64)root->stripesize - 1);
1592 u64 ret = (val + mask) & ~mask;
1597 * walks the btree of allocated extents and find a hole of a given size.
1598 * The key ins is changed to record the hole:
1599 * ins->objectid == block start
1600 * ins->flags = BTRFS_EXTENT_ITEM_KEY
1601 * ins->offset == number of blocks
1602 * Any available blocks before search_start are skipped.
1604 static int noinline find_free_extent(struct btrfs_trans_handle *trans,
1605 struct btrfs_root *orig_root,
1606 u64 num_bytes, u64 empty_size,
1607 u64 search_start, u64 search_end,
1608 u64 hint_byte, struct btrfs_key *ins,
1609 u64 exclude_start, u64 exclude_nr,
1613 u64 orig_search_start;
1614 struct btrfs_root * root = orig_root->fs_info->extent_root;
1615 struct btrfs_fs_info *info = root->fs_info;
1616 u64 total_needed = num_bytes;
1617 u64 *last_ptr = NULL;
1618 struct btrfs_block_group_cache *block_group;
1621 int empty_cluster = 2 * 1024 * 1024;
1623 WARN_ON(num_bytes < root->sectorsize);
1624 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
1626 if (data & BTRFS_BLOCK_GROUP_METADATA) {
1627 last_ptr = &root->fs_info->last_alloc;
1628 empty_cluster = 256 * 1024;
1631 if ((data & BTRFS_BLOCK_GROUP_DATA) && btrfs_test_opt(root, SSD)) {
1632 last_ptr = &root->fs_info->last_data_alloc;
1637 hint_byte = *last_ptr;
1639 empty_size += empty_cluster;
1643 search_start = max(search_start, first_logical_byte(root, 0));
1644 orig_search_start = search_start;
1646 if (search_end == (u64)-1)
1647 search_end = btrfs_super_total_bytes(&info->super_copy);
1650 block_group = btrfs_lookup_block_group(info, hint_byte);
1652 hint_byte = search_start;
1653 block_group = btrfs_find_block_group(root, block_group,
1654 hint_byte, data, 1);
1655 if (last_ptr && *last_ptr == 0 && block_group)
1656 hint_byte = block_group->key.objectid;
1658 block_group = btrfs_find_block_group(root,
1660 search_start, data, 1);
1662 search_start = max(search_start, hint_byte);
1664 total_needed += empty_size;
1668 block_group = btrfs_lookup_block_group(info, search_start);
1670 block_group = btrfs_lookup_block_group(info,
1673 ret = find_search_start(root, &block_group, &search_start,
1674 total_needed, data);
1675 if (ret == -ENOSPC && last_ptr && *last_ptr) {
1677 block_group = btrfs_lookup_block_group(info,
1679 search_start = orig_search_start;
1680 ret = find_search_start(root, &block_group, &search_start,
1681 total_needed, data);
1688 if (last_ptr && *last_ptr && search_start != *last_ptr) {
1691 empty_size += empty_cluster;
1692 total_needed += empty_size;
1694 block_group = btrfs_lookup_block_group(info,
1696 search_start = orig_search_start;
1697 ret = find_search_start(root, &block_group,
1698 &search_start, total_needed, data);
1705 search_start = stripe_align(root, search_start);
1706 ins->objectid = search_start;
1707 ins->offset = num_bytes;
1709 if (ins->objectid + num_bytes >= search_end)
1712 if (ins->objectid + num_bytes >
1713 block_group->key.objectid + block_group->key.offset) {
1714 search_start = block_group->key.objectid +
1715 block_group->key.offset;
1719 if (test_range_bit(&info->extent_ins, ins->objectid,
1720 ins->objectid + num_bytes -1, EXTENT_LOCKED, 0)) {
1721 search_start = ins->objectid + num_bytes;
1725 if (test_range_bit(&info->pinned_extents, ins->objectid,
1726 ins->objectid + num_bytes -1, EXTENT_DIRTY, 0)) {
1727 search_start = ins->objectid + num_bytes;
1731 if (exclude_nr > 0 && (ins->objectid + num_bytes > exclude_start &&
1732 ins->objectid < exclude_start + exclude_nr)) {
1733 search_start = exclude_start + exclude_nr;
1737 if (!(data & BTRFS_BLOCK_GROUP_DATA)) {
1738 block_group = btrfs_lookup_block_group(info, ins->objectid);
1740 trans->block_group = block_group;
1742 ins->offset = num_bytes;
1744 *last_ptr = ins->objectid + ins->offset;
1746 btrfs_super_total_bytes(&root->fs_info->super_copy)) {
1753 if (search_start + num_bytes >= search_end) {
1755 search_start = orig_search_start;
1762 total_needed -= empty_size;
1767 block_group = btrfs_lookup_block_group(info, search_start);
1769 block_group = btrfs_find_block_group(root, block_group,
1770 search_start, data, 0);
1778 * finds a free extent and does all the dirty work required for allocation
1779 * returns the key for the extent through ins, and a tree buffer for
1780 * the first block of the extent through buf.
1782 * returns 0 if everything worked, non-zero otherwise.
1784 int btrfs_alloc_extent(struct btrfs_trans_handle *trans,
1785 struct btrfs_root *root,
1786 u64 num_bytes, u64 min_alloc_size,
1787 u64 root_objectid, u64 ref_generation,
1788 u64 owner, u64 owner_offset,
1789 u64 empty_size, u64 hint_byte,
1790 u64 search_end, struct btrfs_key *ins, u64 data)
1796 u64 search_start = 0;
1799 struct btrfs_fs_info *info = root->fs_info;
1800 struct btrfs_root *extent_root = info->extent_root;
1801 struct btrfs_extent_item *extent_item;
1802 struct btrfs_extent_ref *ref;
1803 struct btrfs_path *path;
1804 struct btrfs_key keys[2];
1807 alloc_profile = info->avail_data_alloc_bits &
1808 info->data_alloc_profile;
1809 data = BTRFS_BLOCK_GROUP_DATA | alloc_profile;
1810 } else if (root == root->fs_info->chunk_root) {
1811 alloc_profile = info->avail_system_alloc_bits &
1812 info->system_alloc_profile;
1813 data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile;
1815 alloc_profile = info->avail_metadata_alloc_bits &
1816 info->metadata_alloc_profile;
1817 data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile;
1820 data = reduce_alloc_profile(root, data);
1821 if (root->ref_cows) {
1822 if (!(data & BTRFS_BLOCK_GROUP_METADATA)) {
1823 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
1825 BTRFS_BLOCK_GROUP_METADATA |
1826 (info->metadata_alloc_profile &
1827 info->avail_metadata_alloc_bits));
1830 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
1831 num_bytes + 2 * 1024 * 1024, data);
1835 WARN_ON(num_bytes < root->sectorsize);
1836 ret = find_free_extent(trans, root, num_bytes, empty_size,
1837 search_start, search_end, hint_byte, ins,
1838 trans->alloc_exclude_start,
1839 trans->alloc_exclude_nr, data);
1841 if (ret == -ENOSPC && num_bytes > min_alloc_size) {
1842 num_bytes = num_bytes >> 1;
1843 num_bytes = max(num_bytes, min_alloc_size);
1847 printk("allocation failed flags %Lu\n", data);
1853 /* block accounting for super block */
1854 super_used = btrfs_super_bytes_used(&info->super_copy);
1855 btrfs_set_super_bytes_used(&info->super_copy, super_used + num_bytes);
1857 /* block accounting for root item */
1858 root_used = btrfs_root_used(&root->root_item);
1859 btrfs_set_root_used(&root->root_item, root_used + num_bytes);
1861 clear_extent_dirty(&root->fs_info->free_space_cache,
1862 ins->objectid, ins->objectid + ins->offset - 1,
1865 if (root == extent_root) {
1866 set_extent_bits(&root->fs_info->extent_ins, ins->objectid,
1867 ins->objectid + ins->offset - 1,
1868 EXTENT_LOCKED, GFP_NOFS);
1872 WARN_ON(trans->alloc_exclude_nr);
1873 trans->alloc_exclude_start = ins->objectid;
1874 trans->alloc_exclude_nr = ins->offset;
1876 memcpy(&keys[0], ins, sizeof(*ins));
1877 keys[1].offset = hash_extent_ref(root_objectid, ref_generation,
1878 owner, owner_offset);
1879 keys[1].objectid = ins->objectid;
1880 keys[1].type = BTRFS_EXTENT_REF_KEY;
1881 sizes[0] = sizeof(*extent_item);
1882 sizes[1] = sizeof(*ref);
1884 path = btrfs_alloc_path();
1887 ret = btrfs_insert_empty_items(trans, extent_root, path, keys,
1891 extent_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1892 struct btrfs_extent_item);
1893 btrfs_set_extent_refs(path->nodes[0], extent_item, 1);
1894 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
1895 struct btrfs_extent_ref);
1897 btrfs_set_ref_root(path->nodes[0], ref, root_objectid);
1898 btrfs_set_ref_generation(path->nodes[0], ref, ref_generation);
1899 btrfs_set_ref_objectid(path->nodes[0], ref, owner);
1900 btrfs_set_ref_offset(path->nodes[0], ref, owner_offset);
1902 btrfs_mark_buffer_dirty(path->nodes[0]);
1904 trans->alloc_exclude_start = 0;
1905 trans->alloc_exclude_nr = 0;
1906 btrfs_free_path(path);
1907 finish_current_insert(trans, extent_root);
1908 pending_ret = del_pending_extents(trans, extent_root);
1918 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1, 0);
1920 printk("update block group failed for %Lu %Lu\n",
1921 ins->objectid, ins->offset);
1928 * helper function to allocate a block for a given tree
1929 * returns the tree buffer or NULL.
1931 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
1932 struct btrfs_root *root,
1934 u64 root_objectid, u64 hint,
1940 ref_generation = trans->transid;
1945 return __btrfs_alloc_free_block(trans, root, blocksize, root_objectid,
1946 ref_generation, 0, 0, hint, empty_size);
1950 * helper function to allocate a block for a given tree
1951 * returns the tree buffer or NULL.
1953 struct extent_buffer *__btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
1954 struct btrfs_root *root,
1963 struct btrfs_key ins;
1965 struct extent_buffer *buf;
1967 ret = btrfs_alloc_extent(trans, root, blocksize, blocksize,
1968 root_objectid, ref_generation,
1969 level, first_objectid, empty_size, hint,
1973 return ERR_PTR(ret);
1975 buf = btrfs_find_create_tree_block(root, ins.objectid, blocksize);
1977 btrfs_free_extent(trans, root, ins.objectid, blocksize,
1978 root->root_key.objectid, ref_generation,
1980 return ERR_PTR(-ENOMEM);
1982 btrfs_set_header_generation(buf, trans->transid);
1983 clean_tree_block(trans, root, buf);
1984 btrfs_set_buffer_uptodate(buf);
1986 if (PageDirty(buf->first_page)) {
1987 printk("page %lu dirty\n", buf->first_page->index);
1991 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
1992 buf->start + buf->len - 1, GFP_NOFS);
1993 if (!btrfs_test_opt(root, SSD))
1994 btrfs_set_buffer_defrag(buf);
1995 trans->blocks_used++;
1999 static int noinline drop_leaf_ref(struct btrfs_trans_handle *trans,
2000 struct btrfs_root *root,
2001 struct extent_buffer *leaf)
2004 u64 leaf_generation;
2005 struct btrfs_key key;
2006 struct btrfs_file_extent_item *fi;
2011 BUG_ON(!btrfs_is_leaf(leaf));
2012 nritems = btrfs_header_nritems(leaf);
2013 leaf_owner = btrfs_header_owner(leaf);
2014 leaf_generation = btrfs_header_generation(leaf);
2016 for (i = 0; i < nritems; i++) {
2019 btrfs_item_key_to_cpu(leaf, &key, i);
2020 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2022 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
2023 if (btrfs_file_extent_type(leaf, fi) ==
2024 BTRFS_FILE_EXTENT_INLINE)
2027 * FIXME make sure to insert a trans record that
2028 * repeats the snapshot del on crash
2030 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
2031 if (disk_bytenr == 0)
2033 ret = btrfs_free_extent(trans, root, disk_bytenr,
2034 btrfs_file_extent_disk_num_bytes(leaf, fi),
2035 leaf_owner, leaf_generation,
2036 key.objectid, key.offset, 0);
2042 static void noinline reada_walk_down(struct btrfs_root *root,
2043 struct extent_buffer *node,
2056 nritems = btrfs_header_nritems(node);
2057 level = btrfs_header_level(node);
2061 for (i = slot; i < nritems && skipped < 32; i++) {
2062 bytenr = btrfs_node_blockptr(node, i);
2063 if (last && ((bytenr > last && bytenr - last > 32 * 1024) ||
2064 (last > bytenr && last - bytenr > 32 * 1024))) {
2068 blocksize = btrfs_level_size(root, level - 1);
2070 ret = lookup_extent_ref(NULL, root, bytenr,
2078 mutex_unlock(&root->fs_info->fs_mutex);
2079 ret = readahead_tree_block(root, bytenr, blocksize);
2080 last = bytenr + blocksize;
2082 mutex_lock(&root->fs_info->fs_mutex);
2089 * helper function for drop_snapshot, this walks down the tree dropping ref
2090 * counts as it goes.
2092 static int noinline walk_down_tree(struct btrfs_trans_handle *trans,
2093 struct btrfs_root *root,
2094 struct btrfs_path *path, int *level)
2099 struct extent_buffer *next;
2100 struct extent_buffer *cur;
2101 struct extent_buffer *parent;
2106 WARN_ON(*level < 0);
2107 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2108 ret = lookup_extent_ref(trans, root,
2109 path->nodes[*level]->start,
2110 path->nodes[*level]->len, &refs);
2116 * walk down to the last node level and free all the leaves
2118 while(*level >= 0) {
2119 WARN_ON(*level < 0);
2120 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2121 cur = path->nodes[*level];
2123 if (btrfs_header_level(cur) != *level)
2126 if (path->slots[*level] >=
2127 btrfs_header_nritems(cur))
2130 ret = drop_leaf_ref(trans, root, cur);
2134 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
2135 blocksize = btrfs_level_size(root, *level - 1);
2136 ret = lookup_extent_ref(trans, root, bytenr, blocksize, &refs);
2139 parent = path->nodes[*level];
2140 root_owner = btrfs_header_owner(parent);
2141 root_gen = btrfs_header_generation(parent);
2142 path->slots[*level]++;
2143 ret = btrfs_free_extent(trans, root, bytenr,
2144 blocksize, root_owner,
2149 next = btrfs_find_tree_block(root, bytenr, blocksize);
2150 if (!next || !btrfs_buffer_uptodate(next)) {
2151 free_extent_buffer(next);
2152 reada_walk_down(root, cur, path->slots[*level]);
2154 mutex_unlock(&root->fs_info->fs_mutex);
2155 next = read_tree_block(root, bytenr, blocksize);
2156 mutex_lock(&root->fs_info->fs_mutex);
2158 /* we've dropped the lock, double check */
2159 ret = lookup_extent_ref(trans, root, bytenr,
2163 parent = path->nodes[*level];
2164 root_owner = btrfs_header_owner(parent);
2165 root_gen = btrfs_header_generation(parent);
2167 path->slots[*level]++;
2168 free_extent_buffer(next);
2169 ret = btrfs_free_extent(trans, root, bytenr,
2177 btrfs_verify_block_csum(root, next);
2179 WARN_ON(*level <= 0);
2180 if (path->nodes[*level-1])
2181 free_extent_buffer(path->nodes[*level-1]);
2182 path->nodes[*level-1] = next;
2183 *level = btrfs_header_level(next);
2184 path->slots[*level] = 0;
2187 WARN_ON(*level < 0);
2188 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2190 if (path->nodes[*level] == root->node) {
2191 root_owner = root->root_key.objectid;
2192 parent = path->nodes[*level];
2194 parent = path->nodes[*level + 1];
2195 root_owner = btrfs_header_owner(parent);
2198 root_gen = btrfs_header_generation(parent);
2199 ret = btrfs_free_extent(trans, root, path->nodes[*level]->start,
2200 path->nodes[*level]->len,
2201 root_owner, root_gen, 0, 0, 1);
2202 free_extent_buffer(path->nodes[*level]);
2203 path->nodes[*level] = NULL;
2210 * helper for dropping snapshots. This walks back up the tree in the path
2211 * to find the first node higher up where we haven't yet gone through
2214 static int noinline walk_up_tree(struct btrfs_trans_handle *trans,
2215 struct btrfs_root *root,
2216 struct btrfs_path *path, int *level)
2220 struct btrfs_root_item *root_item = &root->root_item;
2225 for(i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
2226 slot = path->slots[i];
2227 if (slot < btrfs_header_nritems(path->nodes[i]) - 1) {
2228 struct extent_buffer *node;
2229 struct btrfs_disk_key disk_key;
2230 node = path->nodes[i];
2233 WARN_ON(*level == 0);
2234 btrfs_node_key(node, &disk_key, path->slots[i]);
2235 memcpy(&root_item->drop_progress,
2236 &disk_key, sizeof(disk_key));
2237 root_item->drop_level = i;
2240 if (path->nodes[*level] == root->node) {
2241 root_owner = root->root_key.objectid;
2243 btrfs_header_generation(path->nodes[*level]);
2245 struct extent_buffer *node;
2246 node = path->nodes[*level + 1];
2247 root_owner = btrfs_header_owner(node);
2248 root_gen = btrfs_header_generation(node);
2250 ret = btrfs_free_extent(trans, root,
2251 path->nodes[*level]->start,
2252 path->nodes[*level]->len,
2253 root_owner, root_gen, 0, 0, 1);
2255 free_extent_buffer(path->nodes[*level]);
2256 path->nodes[*level] = NULL;
2264 * drop the reference count on the tree rooted at 'snap'. This traverses
2265 * the tree freeing any blocks that have a ref count of zero after being
2268 int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root
2274 struct btrfs_path *path;
2277 struct btrfs_root_item *root_item = &root->root_item;
2279 path = btrfs_alloc_path();
2282 level = btrfs_header_level(root->node);
2284 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
2285 path->nodes[level] = root->node;
2286 extent_buffer_get(root->node);
2287 path->slots[level] = 0;
2289 struct btrfs_key key;
2290 struct btrfs_disk_key found_key;
2291 struct extent_buffer *node;
2293 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
2294 level = root_item->drop_level;
2295 path->lowest_level = level;
2296 wret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2301 node = path->nodes[level];
2302 btrfs_node_key(node, &found_key, path->slots[level]);
2303 WARN_ON(memcmp(&found_key, &root_item->drop_progress,
2304 sizeof(found_key)));
2307 wret = walk_down_tree(trans, root, path, &level);
2313 wret = walk_up_tree(trans, root, path, &level);
2321 for (i = 0; i <= orig_level; i++) {
2322 if (path->nodes[i]) {
2323 free_extent_buffer(path->nodes[i]);
2324 path->nodes[i] = NULL;
2328 btrfs_free_path(path);
2332 int btrfs_free_block_groups(struct btrfs_fs_info *info)
2339 ret = find_first_extent_bit(&info->block_group_cache, 0,
2340 &start, &end, (unsigned int)-1);
2343 ret = get_state_private(&info->block_group_cache, start, &ptr);
2345 kfree((void *)(unsigned long)ptr);
2346 clear_extent_bits(&info->block_group_cache, start,
2347 end, (unsigned int)-1, GFP_NOFS);
2350 ret = find_first_extent_bit(&info->free_space_cache, 0,
2351 &start, &end, EXTENT_DIRTY);
2354 clear_extent_dirty(&info->free_space_cache, start,
2360 static unsigned long calc_ra(unsigned long start, unsigned long last,
2363 return min(last, start + nr - 1);
2366 static int noinline relocate_inode_pages(struct inode *inode, u64 start,
2371 unsigned long last_index;
2374 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2375 struct file_ra_state *ra;
2376 unsigned long total_read = 0;
2377 unsigned long ra_pages;
2378 struct btrfs_trans_handle *trans;
2380 ra = kzalloc(sizeof(*ra), GFP_NOFS);
2382 mutex_lock(&inode->i_mutex);
2383 i = start >> PAGE_CACHE_SHIFT;
2384 last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
2386 ra_pages = BTRFS_I(inode)->root->fs_info->bdi.ra_pages;
2388 file_ra_state_init(ra, inode->i_mapping);
2390 for (; i <= last_index; i++) {
2391 if (total_read % ra_pages == 0) {
2392 btrfs_force_ra(inode->i_mapping, ra, NULL, i,
2393 calc_ra(i, last_index, ra_pages));
2396 if (((u64)i << PAGE_CACHE_SHIFT) > inode->i_size)
2397 goto truncate_racing;
2399 page = grab_cache_page(inode->i_mapping, i);
2403 if (!PageUptodate(page)) {
2404 btrfs_readpage(NULL, page);
2406 if (!PageUptodate(page)) {
2408 page_cache_release(page);
2412 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
2413 ClearPageDirty(page);
2415 cancel_dirty_page(page, PAGE_CACHE_SIZE);
2417 wait_on_page_writeback(page);
2418 set_page_extent_mapped(page);
2419 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
2420 page_end = page_start + PAGE_CACHE_SIZE - 1;
2422 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
2424 set_extent_delalloc(io_tree, page_start,
2425 page_end, GFP_NOFS);
2426 set_page_dirty(page);
2428 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
2430 page_cache_release(page);
2432 balance_dirty_pages_ratelimited_nr(inode->i_mapping,
2437 trans = btrfs_start_transaction(BTRFS_I(inode)->root, 1);
2439 btrfs_add_ordered_inode(inode);
2440 btrfs_end_transaction(trans, BTRFS_I(inode)->root);
2441 mark_inode_dirty(inode);
2443 mutex_unlock(&inode->i_mutex);
2447 vmtruncate(inode, inode->i_size);
2448 balance_dirty_pages_ratelimited_nr(inode->i_mapping,
2454 * The back references tell us which tree holds a ref on a block,
2455 * but it is possible for the tree root field in the reference to
2456 * reflect the original root before a snapshot was made. In this
2457 * case we should search through all the children of a given root
2458 * to find potential holders of references on a block.
2460 * Instead, we do something a little less fancy and just search
2461 * all the roots for a given key/block combination.
2463 static int find_root_for_ref(struct btrfs_root *root,
2464 struct btrfs_path *path,
2465 struct btrfs_key *key0,
2468 struct btrfs_root **found_root,
2471 struct btrfs_key root_location;
2472 struct btrfs_root *cur_root = *found_root;
2473 struct btrfs_file_extent_item *file_extent;
2474 u64 root_search_start = BTRFS_FS_TREE_OBJECTID;
2479 root_location.offset = (u64)-1;
2480 root_location.type = BTRFS_ROOT_ITEM_KEY;
2481 path->lowest_level = level;
2484 ret = btrfs_search_slot(NULL, cur_root, key0, path, 0, 0);
2486 if (ret == 0 && file_key) {
2487 struct extent_buffer *leaf = path->nodes[0];
2488 file_extent = btrfs_item_ptr(leaf, path->slots[0],
2489 struct btrfs_file_extent_item);
2490 if (btrfs_file_extent_type(leaf, file_extent) ==
2491 BTRFS_FILE_EXTENT_REG) {
2493 btrfs_file_extent_disk_bytenr(leaf,
2496 } else if (ret == 0) {
2497 if (path->nodes[level])
2498 found_bytenr = path->nodes[level]->start;
2501 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
2502 if (!path->nodes[i])
2504 free_extent_buffer(path->nodes[i]);
2505 path->nodes[i] = NULL;
2507 btrfs_release_path(cur_root, path);
2509 if (found_bytenr == bytenr) {
2510 *found_root = cur_root;
2514 ret = btrfs_search_root(root->fs_info->tree_root,
2515 root_search_start, &root_search_start);
2519 root_location.objectid = root_search_start;
2520 cur_root = btrfs_read_fs_root_no_name(root->fs_info,
2528 path->lowest_level = 0;
2533 * note, this releases the path
2535 static int noinline relocate_one_reference(struct btrfs_root *extent_root,
2536 struct btrfs_path *path,
2537 struct btrfs_key *extent_key)
2539 struct inode *inode;
2540 struct btrfs_root *found_root;
2541 struct btrfs_key root_location;
2542 struct btrfs_key found_key;
2543 struct btrfs_extent_ref *ref;
2551 ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
2552 struct btrfs_extent_ref);
2553 ref_root = btrfs_ref_root(path->nodes[0], ref);
2554 ref_gen = btrfs_ref_generation(path->nodes[0], ref);
2555 ref_objectid = btrfs_ref_objectid(path->nodes[0], ref);
2556 ref_offset = btrfs_ref_offset(path->nodes[0], ref);
2557 btrfs_release_path(extent_root, path);
2559 root_location.objectid = ref_root;
2561 root_location.offset = 0;
2563 root_location.offset = (u64)-1;
2564 root_location.type = BTRFS_ROOT_ITEM_KEY;
2566 found_root = btrfs_read_fs_root_no_name(extent_root->fs_info,
2568 BUG_ON(!found_root);
2570 if (ref_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
2571 found_key.objectid = ref_objectid;
2572 found_key.type = BTRFS_EXTENT_DATA_KEY;
2573 found_key.offset = ref_offset;
2576 ret = find_root_for_ref(extent_root, path, &found_key,
2577 level, 1, &found_root,
2578 extent_key->objectid);
2583 mutex_unlock(&extent_root->fs_info->fs_mutex);
2584 inode = btrfs_iget_locked(extent_root->fs_info->sb,
2585 ref_objectid, found_root);
2586 if (inode->i_state & I_NEW) {
2587 /* the inode and parent dir are two different roots */
2588 BTRFS_I(inode)->root = found_root;
2589 BTRFS_I(inode)->location.objectid = ref_objectid;
2590 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
2591 BTRFS_I(inode)->location.offset = 0;
2592 btrfs_read_locked_inode(inode);
2593 unlock_new_inode(inode);
2596 /* this can happen if the reference is not against
2597 * the latest version of the tree root
2599 if (is_bad_inode(inode)) {
2600 mutex_lock(&extent_root->fs_info->fs_mutex);
2603 relocate_inode_pages(inode, ref_offset, extent_key->offset);
2605 mutex_lock(&extent_root->fs_info->fs_mutex);
2607 struct btrfs_trans_handle *trans;
2608 struct extent_buffer *eb;
2611 eb = read_tree_block(found_root, extent_key->objectid,
2612 extent_key->offset);
2613 level = btrfs_header_level(eb);
2616 btrfs_item_key_to_cpu(eb, &found_key, 0);
2618 btrfs_node_key_to_cpu(eb, &found_key, 0);
2620 free_extent_buffer(eb);
2622 ret = find_root_for_ref(extent_root, path, &found_key,
2623 level, 0, &found_root,
2624 extent_key->objectid);
2629 trans = btrfs_start_transaction(found_root, 1);
2631 path->lowest_level = level;
2633 ret = btrfs_search_slot(trans, found_root, &found_key, path,
2635 path->lowest_level = 0;
2636 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
2637 if (!path->nodes[i])
2639 free_extent_buffer(path->nodes[i]);
2640 path->nodes[i] = NULL;
2642 btrfs_release_path(found_root, path);
2643 btrfs_end_transaction(trans, found_root);
2650 static int noinline del_extent_zero(struct btrfs_root *extent_root,
2651 struct btrfs_path *path,
2652 struct btrfs_key *extent_key)
2655 struct btrfs_trans_handle *trans;
2657 trans = btrfs_start_transaction(extent_root, 1);
2658 ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1);
2665 ret = btrfs_del_item(trans, extent_root, path);
2667 btrfs_end_transaction(trans, extent_root);
2671 static int noinline relocate_one_extent(struct btrfs_root *extent_root,
2672 struct btrfs_path *path,
2673 struct btrfs_key *extent_key)
2675 struct btrfs_key key;
2676 struct btrfs_key found_key;
2677 struct extent_buffer *leaf;
2682 if (extent_key->objectid == 0) {
2683 ret = del_extent_zero(extent_root, path, extent_key);
2686 key.objectid = extent_key->objectid;
2687 key.type = BTRFS_EXTENT_REF_KEY;
2691 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2697 leaf = path->nodes[0];
2698 nritems = btrfs_header_nritems(leaf);
2699 if (path->slots[0] == nritems) {
2700 ret = btrfs_next_leaf(extent_root, path);
2707 leaf = path->nodes[0];
2710 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2711 if (found_key.objectid != extent_key->objectid) {
2715 if (found_key.type != BTRFS_EXTENT_REF_KEY) {
2719 key.offset = found_key.offset + 1;
2720 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2722 ret = relocate_one_reference(extent_root, path, extent_key);
2728 btrfs_release_path(extent_root, path);
2732 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
2735 u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
2736 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
2738 num_devices = root->fs_info->fs_devices->num_devices;
2739 if (num_devices == 1) {
2740 stripped |= BTRFS_BLOCK_GROUP_DUP;
2741 stripped = flags & ~stripped;
2743 /* turn raid0 into single device chunks */
2744 if (flags & BTRFS_BLOCK_GROUP_RAID0)
2747 /* turn mirroring into duplication */
2748 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
2749 BTRFS_BLOCK_GROUP_RAID10))
2750 return stripped | BTRFS_BLOCK_GROUP_DUP;
2753 /* they already had raid on here, just return */
2754 if (flags & stripped)
2757 stripped |= BTRFS_BLOCK_GROUP_DUP;
2758 stripped = flags & ~stripped;
2760 /* switch duplicated blocks with raid1 */
2761 if (flags & BTRFS_BLOCK_GROUP_DUP)
2762 return stripped | BTRFS_BLOCK_GROUP_RAID1;
2764 /* turn single device chunks into raid0 */
2765 return stripped | BTRFS_BLOCK_GROUP_RAID0;
2770 int btrfs_shrink_extent_tree(struct btrfs_root *root, u64 shrink_start)
2772 struct btrfs_trans_handle *trans;
2773 struct btrfs_root *tree_root = root->fs_info->tree_root;
2774 struct btrfs_path *path;
2777 u64 shrink_last_byte;
2778 u64 new_alloc_flags;
2779 struct btrfs_block_group_cache *shrink_block_group;
2780 struct btrfs_fs_info *info = root->fs_info;
2781 struct btrfs_key key;
2782 struct btrfs_key found_key;
2783 struct extent_buffer *leaf;
2788 shrink_block_group = btrfs_lookup_block_group(root->fs_info,
2790 BUG_ON(!shrink_block_group);
2792 shrink_last_byte = shrink_start + shrink_block_group->key.offset;
2794 shrink_block_group->space_info->total_bytes -=
2795 shrink_block_group->key.offset;
2796 path = btrfs_alloc_path();
2797 root = root->fs_info->extent_root;
2801 if (btrfs_block_group_used(&shrink_block_group->item) > 0) {
2802 trans = btrfs_start_transaction(root, 1);
2803 new_alloc_flags = update_block_group_flags(root,
2804 shrink_block_group->flags);
2805 do_chunk_alloc(trans, root->fs_info->extent_root,
2806 btrfs_block_group_used(&shrink_block_group->item) +
2807 2 * 1024 * 1024, new_alloc_flags);
2808 btrfs_end_transaction(trans, root);
2810 shrink_block_group->ro = 1;
2814 key.objectid = shrink_start;
2817 cur_byte = key.objectid;
2819 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2823 ret = btrfs_previous_item(root, path, 0, BTRFS_EXTENT_ITEM_KEY);
2828 leaf = path->nodes[0];
2829 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2830 if (found_key.objectid + found_key.offset > shrink_start &&
2831 found_key.objectid < shrink_last_byte) {
2832 cur_byte = found_key.objectid;
2833 key.objectid = cur_byte;
2836 btrfs_release_path(root, path);
2839 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2843 leaf = path->nodes[0];
2844 nritems = btrfs_header_nritems(leaf);
2846 if (path->slots[0] >= nritems) {
2847 ret = btrfs_next_leaf(root, path);
2854 leaf = path->nodes[0];
2855 nritems = btrfs_header_nritems(leaf);
2858 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2860 if (found_key.objectid >= shrink_last_byte)
2863 if (progress && need_resched()) {
2864 memcpy(&key, &found_key, sizeof(key));
2865 mutex_unlock(&root->fs_info->fs_mutex);
2867 mutex_lock(&root->fs_info->fs_mutex);
2868 btrfs_release_path(root, path);
2869 btrfs_search_slot(NULL, root, &key, path, 0, 0);
2875 if (btrfs_key_type(&found_key) != BTRFS_EXTENT_ITEM_KEY ||
2876 found_key.objectid + found_key.offset <= cur_byte) {
2882 cur_byte = found_key.objectid + found_key.offset;
2883 key.objectid = cur_byte;
2884 btrfs_release_path(root, path);
2885 ret = relocate_one_extent(root, path, &found_key);
2888 btrfs_release_path(root, path);
2890 if (total_found > 0) {
2891 trans = btrfs_start_transaction(tree_root, 1);
2892 btrfs_commit_transaction(trans, tree_root);
2894 mutex_unlock(&root->fs_info->fs_mutex);
2895 btrfs_clean_old_snapshots(tree_root);
2896 mutex_lock(&root->fs_info->fs_mutex);
2898 trans = btrfs_start_transaction(tree_root, 1);
2899 btrfs_commit_transaction(trans, tree_root);
2904 * we've freed all the extents, now remove the block
2905 * group item from the tree
2907 trans = btrfs_start_transaction(root, 1);
2908 memcpy(&key, &shrink_block_group->key, sizeof(key));
2910 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2916 leaf = path->nodes[0];
2917 nritems = btrfs_header_nritems(leaf);
2918 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2919 kfree(shrink_block_group);
2921 clear_extent_bits(&info->block_group_cache, found_key.objectid,
2922 found_key.objectid + found_key.offset - 1,
2923 (unsigned int)-1, GFP_NOFS);
2925 btrfs_del_item(trans, root, path);
2926 clear_extent_dirty(&info->free_space_cache,
2927 shrink_start, shrink_last_byte - 1,
2929 btrfs_commit_transaction(trans, root);
2931 btrfs_free_path(path);
2935 int find_first_block_group(struct btrfs_root *root, struct btrfs_path *path,
2936 struct btrfs_key *key)
2939 struct btrfs_key found_key;
2940 struct extent_buffer *leaf;
2943 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
2947 slot = path->slots[0];
2948 leaf = path->nodes[0];
2949 if (slot >= btrfs_header_nritems(leaf)) {
2950 ret = btrfs_next_leaf(root, path);
2957 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2959 if (found_key.objectid >= key->objectid &&
2960 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY)
2969 int btrfs_read_block_groups(struct btrfs_root *root)
2971 struct btrfs_path *path;
2974 struct btrfs_block_group_cache *cache;
2975 struct btrfs_fs_info *info = root->fs_info;
2976 struct btrfs_space_info *space_info;
2977 struct extent_io_tree *block_group_cache;
2978 struct btrfs_key key;
2979 struct btrfs_key found_key;
2980 struct extent_buffer *leaf;
2982 block_group_cache = &info->block_group_cache;
2983 root = info->extent_root;
2986 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
2987 path = btrfs_alloc_path();
2992 ret = find_first_block_group(root, path, &key);
3000 leaf = path->nodes[0];
3001 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3002 cache = kzalloc(sizeof(*cache), GFP_NOFS);
3008 read_extent_buffer(leaf, &cache->item,
3009 btrfs_item_ptr_offset(leaf, path->slots[0]),
3010 sizeof(cache->item));
3011 memcpy(&cache->key, &found_key, sizeof(found_key));
3013 key.objectid = found_key.objectid + found_key.offset;
3014 btrfs_release_path(root, path);
3015 cache->flags = btrfs_block_group_flags(&cache->item);
3017 if (cache->flags & BTRFS_BLOCK_GROUP_DATA) {
3018 bit = BLOCK_GROUP_DATA;
3019 } else if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) {
3020 bit = BLOCK_GROUP_SYSTEM;
3021 } else if (cache->flags & BTRFS_BLOCK_GROUP_METADATA) {
3022 bit = BLOCK_GROUP_METADATA;
3024 set_avail_alloc_bits(info, cache->flags);
3026 ret = update_space_info(info, cache->flags, found_key.offset,
3027 btrfs_block_group_used(&cache->item),
3030 cache->space_info = space_info;
3032 /* use EXTENT_LOCKED to prevent merging */
3033 set_extent_bits(block_group_cache, found_key.objectid,
3034 found_key.objectid + found_key.offset - 1,
3035 bit | EXTENT_LOCKED, GFP_NOFS);
3036 set_state_private(block_group_cache, found_key.objectid,
3037 (unsigned long)cache);
3040 btrfs_super_total_bytes(&info->super_copy))
3045 btrfs_free_path(path);
3049 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
3050 struct btrfs_root *root, u64 bytes_used,
3051 u64 type, u64 chunk_objectid, u64 chunk_offset,
3056 struct btrfs_root *extent_root;
3057 struct btrfs_block_group_cache *cache;
3058 struct extent_io_tree *block_group_cache;
3060 extent_root = root->fs_info->extent_root;
3061 block_group_cache = &root->fs_info->block_group_cache;
3063 cache = kzalloc(sizeof(*cache), GFP_NOFS);
3065 cache->key.objectid = chunk_offset;
3066 cache->key.offset = size;
3068 btrfs_set_key_type(&cache->key, BTRFS_BLOCK_GROUP_ITEM_KEY);
3069 memset(&cache->item, 0, sizeof(cache->item));
3070 btrfs_set_block_group_used(&cache->item, bytes_used);
3071 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
3072 cache->flags = type;
3073 btrfs_set_block_group_flags(&cache->item, type);
3075 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
3076 &cache->space_info);
3079 bit = block_group_state_bits(type);
3080 set_extent_bits(block_group_cache, chunk_offset,
3081 chunk_offset + size - 1,
3082 bit | EXTENT_LOCKED, GFP_NOFS);
3084 set_state_private(block_group_cache, chunk_offset,
3085 (unsigned long)cache);
3086 ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
3087 sizeof(cache->item));
3090 finish_current_insert(trans, extent_root);
3091 ret = del_pending_extents(trans, extent_root);
3093 set_avail_alloc_bits(extent_root->fs_info, type);