btrfs-progs: convert: Fix offset-by-one error in read_data_extent()
[platform/upstream/btrfs-progs.git] / extent-tree.c
index 544ab2f..e2ae74a 100644 (file)
@@ -18,6 +18,8 @@
 
 #include <stdio.h>
 #include <stdlib.h>
+#include <stdint.h>
+#include <math.h>
 #include "kerncompat.h"
 #include "radix-tree.h"
 #include "ctree.h"
 #include "transaction.h"
 #include "crc32c.h"
 #include "volumes.h"
-
-#define BLOCK_GROUP_DATA     EXTENT_WRITEBACK
-#define BLOCK_GROUP_METADATA EXTENT_UPTODATE
-#define BLOCK_GROUP_SYSTEM   EXTENT_NEW
-
-#define BLOCK_GROUP_DIRTY EXTENT_DIRTY
+#include "free-space-cache.h"
+#include "utils.h"
 
 #define PENDING_EXTENT_INSERT 0
 #define PENDING_EXTENT_DELETE 1
@@ -60,6 +58,9 @@ static int finish_current_insert(struct btrfs_trans_handle *trans, struct
                                 btrfs_root *extent_root);
 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
                               btrfs_root *extent_root);
+static struct btrfs_block_group_cache *
+btrfs_find_block_group(struct btrfs_root *root, struct btrfs_block_group_cache
+                      *hint, u64 search_start, int data, int owner);
 
 static int remove_sb_from_cache(struct btrfs_root *root,
                                struct btrfs_block_group_cache *cache)
@@ -68,18 +69,18 @@ static int remove_sb_from_cache(struct btrfs_root *root,
        u64 *logical;
        int stripe_len;
        int i, nr, ret;
+       struct btrfs_fs_info *fs_info = root->fs_info;
        struct extent_io_tree *free_space_cache;
 
-       free_space_cache = &root->fs_info->free_space_cache;
+       free_space_cache = &fs_info->free_space_cache;
        for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
                bytenr = btrfs_sb_offset(i);
-               ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
-                                      cache->key.objectid, bytenr, 0,
+               ret = btrfs_rmap_block(fs_info, cache->key.objectid, bytenr, 0,
                                       &logical, &nr, &stripe_len);
                BUG_ON(ret);
                while (nr--) {
                        clear_extent_dirty(free_space_cache, logical[nr],
-                               logical[nr] + stripe_len - 1, GFP_NOFS);
+                               logical[nr] + stripe_len - 1);
                }
                kfree(logical);
        }
@@ -115,7 +116,8 @@ static int cache_block_group(struct btrfs_root *root,
        last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
        key.objectid = last;
        key.offset = 0;
-       btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
+       key.type = 0;
+
        ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
        if (ret < 0)
                goto err;
@@ -142,14 +144,17 @@ static int cache_block_group(struct btrfs_root *root,
                        break;
                }
 
-               if (btrfs_key_type(&key) == BTRFS_EXTENT_ITEM_KEY) {
+               if (key.type == BTRFS_EXTENT_ITEM_KEY ||
+                   key.type == BTRFS_METADATA_ITEM_KEY) {
                        if (key.objectid > last) {
                                hole_size = key.objectid - last;
                                set_extent_dirty(free_space_cache, last,
-                                                last + hole_size - 1,
-                                                GFP_NOFS);
+                                                last + hole_size - 1);
                        }
-                       last = key.objectid + key.offset;
+                       if (key.type == BTRFS_METADATA_ITEM_KEY)
+                               last = key.objectid + root->fs_info->nodesize;
+                       else
+                               last = key.objectid + key.offset;
                }
 next:
                path->slots[0]++;
@@ -159,8 +164,7 @@ next:
            block_group->key.offset > last) {
                hole_size = block_group->key.objectid +
                        block_group->key.offset - last;
-               set_extent_dirty(free_space_cache, last,
-                                last + hole_size - 1, GFP_NOFS);
+               set_extent_dirty(free_space_cache, last, last + hole_size - 1);
        }
        remove_sb_from_cache(root, block_group);
        block_group->cached = 1;
@@ -239,24 +243,22 @@ static int noinline find_search_start(struct btrfs_root *root,
 {
        int ret;
        struct btrfs_block_group_cache *cache = *cache_ret;
-       u64 last;
+       u64 last = *start_ret;
        u64 start = 0;
        u64 end = 0;
        u64 search_start = *start_ret;
        int wrapped = 0;
 
-       if (!cache) {
+       if (!cache)
                goto out;
-       }
 again:
        ret = cache_block_group(root, cache);
        if (ret)
                goto out;
 
        last = max(search_start, cache->key.objectid);
-       if (cache->ro || !block_group_bits(cache, data)) {
+       if (cache->ro || !block_group_bits(cache, data))
                goto new_group;
-       }
 
        while(1) {
                ret = find_first_extent_bit(&root->fs_info->free_space_cache,
@@ -277,11 +279,12 @@ again:
                return 0;
        }
 out:
+       *start_ret = last;
        cache = btrfs_lookup_block_group(root->fs_info, search_start);
        if (!cache) {
                printk("Unable to find block group for %llu\n",
                        (unsigned long long)search_start);
-               WARN_ON(1);
+               return -ENOENT;
        }
        return -ENOSPC;
 
@@ -290,7 +293,6 @@ new_group:
 wrapped:
        cache = btrfs_lookup_first_block_group(root->fs_info, last);
        if (!cache) {
-no_cache:
                if (!wrapped) {
                        wrapped = 1;
                        last = search_start;
@@ -298,24 +300,10 @@ no_cache:
                }
                goto out;
        }
-       cache = btrfs_find_block_group(root, cache, last, data, 0);
-       cache = btrfs_find_block_group(root, cache, last, data, 0);
-       if (!cache)
-               goto no_cache;
-
        *cache_ret = cache;
        goto again;
 }
 
-static u64 div_factor(u64 num, int factor)
-{
-       if (factor == 10)
-               return num;
-       num *= factor;
-       num /= 10;
-       return num;
-}
-
 static int block_group_state_bits(u64 flags)
 {
        int bits = 0;
@@ -328,10 +316,9 @@ static int block_group_state_bits(u64 flags)
        return bits;
 }
 
-struct btrfs_block_group_cache *btrfs_find_block_group(struct btrfs_root *root,
-                                                struct btrfs_block_group_cache
-                                                *hint, u64 search_start,
-                                                int data, int owner)
+static struct btrfs_block_group_cache *
+btrfs_find_block_group(struct btrfs_root *root, struct btrfs_block_group_cache
+                      *hint, u64 search_start, int data, int owner)
 {
        struct btrfs_block_group_cache *cache;
        struct extent_io_tree *block_group_cache;
@@ -574,7 +561,7 @@ static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
                        break;
                }
        }
-       btrfs_release_path(root, path);
+       btrfs_release_path(path);
 
        if (owner < BTRFS_FIRST_FREE_OBJECTID)
                new_size += sizeof(*bi);
@@ -585,7 +572,7 @@ static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
                return ret;
        BUG_ON(ret);
 
-       ret = btrfs_extend_item(trans, root, path, new_size);
+       ret = btrfs_extend_item(root, path, new_size);
        BUG_ON(ret);
 
        leaf = path->nodes[0];
@@ -609,7 +596,7 @@ static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
 }
 #endif
 
-static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
+u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
 {
        u32 high_crc = ~(u32)0;
        u32 low_crc = ~(u32)0;
@@ -681,7 +668,7 @@ again:
                        return 0;
 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
                key.type = BTRFS_EXTENT_REF_V0_KEY;
-               btrfs_release_path(root, path);
+               btrfs_release_path(path);
                ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
                if (ret < 0) {
                        err = ret;
@@ -719,7 +706,7 @@ again:
                if (match_extent_data_ref(leaf, ref, root_objectid,
                                          owner, offset)) {
                        if (recow) {
-                               btrfs_release_path(root, path);
+                               btrfs_release_path(path);
                                goto again;
                        }
                        err = 0;
@@ -780,7 +767,7 @@ static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
                        if (match_extent_data_ref(leaf, ref, root_objectid,
                                                  owner, offset))
                                break;
-                       btrfs_release_path(root, path);
+                       btrfs_release_path(path);
 
                        key.offset++;
                        ret = btrfs_insert_empty_item(trans, root, path, &key,
@@ -807,7 +794,7 @@ static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
        btrfs_mark_buffer_dirty(leaf);
        ret = 0;
 fail:
-       btrfs_release_path(root, path);
+       btrfs_release_path(path);
        return ret;
 }
 
@@ -868,8 +855,7 @@ static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
        return ret;
 }
 
-static noinline u32 extent_data_ref_count(struct btrfs_root *root,
-                                         struct btrfs_path *path,
+static noinline u32 extent_data_ref_count(struct btrfs_path *path,
                                          struct btrfs_extent_inline_ref *iref)
 {
        struct btrfs_key key;
@@ -933,7 +919,7 @@ static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
                ret = -ENOENT;
 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
        if (ret == -ENOENT && parent) {
-               btrfs_release_path(root, path);
+               btrfs_release_path(path);
                key.type = BTRFS_EXTENT_REF_V0_KEY;
                ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
                if (ret > 0)
@@ -963,44 +949,25 @@ static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
 
        ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
 
-       btrfs_release_path(root, path);
+       btrfs_release_path(path);
        return ret;
 }
 
 static inline int extent_ref_type(u64 parent, u64 owner)
 {
+       int type;
        if (owner < BTRFS_FIRST_FREE_OBJECTID) {
                if (parent > 0)
-                       return BTRFS_SHARED_BLOCK_REF_KEY;
+                       type = BTRFS_SHARED_BLOCK_REF_KEY;
                else
-                       return BTRFS_TREE_BLOCK_REF_KEY;
+                       type = BTRFS_TREE_BLOCK_REF_KEY;
        } else {
                if (parent > 0)
-                       return BTRFS_SHARED_DATA_REF_KEY;
-               else
-                       return BTRFS_EXTENT_DATA_REF_KEY;
-       }
-}
-
-static int find_next_key(struct btrfs_path *path, struct btrfs_key *key)
-
-{
-       int level;
-       for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
-               if (!path->nodes[level])
-                       break;
-               if (path->slots[level] + 1 >=
-                   btrfs_header_nritems(path->nodes[level]))
-                       continue;
-               if (level == 0)
-                       btrfs_item_key_to_cpu(path->nodes[level], key,
-                                             path->slots[level] + 1);
+                       type = BTRFS_SHARED_DATA_REF_KEY;
                else
-                       btrfs_node_key_to_cpu(path->nodes[level], key,
-                                             path->slots[level] + 1);
-               return 0;
+                       type = BTRFS_EXTENT_DATA_REF_KEY;
        }
-       return 1;
+       return type;
 }
 
 static int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
@@ -1024,6 +991,8 @@ static int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
        int want;
        int ret;
        int err = 0;
+       int skinny_metadata =
+               btrfs_fs_incompat(root->fs_info, SKINNY_METADATA);
 
        key.objectid = bytenr;
        key.type = BTRFS_EXTENT_ITEM_KEY;
@@ -1034,11 +1003,44 @@ static int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
                extra_size = btrfs_extent_inline_ref_size(want);
        else
                extra_size = -1;
+
+       if (owner < BTRFS_FIRST_FREE_OBJECTID && skinny_metadata) {
+               key.type = BTRFS_METADATA_ITEM_KEY;
+               key.offset = owner;
+       } else if (skinny_metadata) {
+               skinny_metadata = 0;
+       }
+
+again:
        ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
        if (ret < 0) {
                err = ret;
                goto out;
        }
+
+       /*
+        * We may be a newly converted file system which still has the old fat
+        * extent entries for metadata, so try and see if we have one of those.
+        */
+       if (ret > 0 && skinny_metadata) {
+               skinny_metadata = 0;
+               if (path->slots[0]) {
+                       path->slots[0]--;
+                       btrfs_item_key_to_cpu(path->nodes[0], &key,
+                                             path->slots[0]);
+                       if (key.objectid == bytenr &&
+                           key.type == BTRFS_EXTENT_ITEM_KEY &&
+                           key.offset == num_bytes)
+                               ret = 0;
+               }
+               if (ret) {
+                       key.type = BTRFS_EXTENT_ITEM_KEY;
+                       key.offset = num_bytes;
+                       btrfs_release_path(path);
+                       goto again;
+               }
+       }
+
        if (ret) {
                printf("Failed to find [%llu, %u, %llu]\n", key.objectid, key.type, key.offset);
                return -ENOENT;
@@ -1079,10 +1081,10 @@ static int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
        ptr = (unsigned long)(ei + 1);
        end = (unsigned long)ei + item_size;
 
-       if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
+       if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) {
                ptr += sizeof(struct btrfs_tree_block_info);
                BUG_ON(ptr > end);
-       } else {
+       } else if (!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)) {
                if (!(flags & BTRFS_EXTENT_FLAG_DATA)) {
                        return -EIO;
                }
@@ -1158,8 +1160,7 @@ out:
        return err;
 }
 
-static int setup_inline_extent_backref(struct btrfs_trans_handle *trans,
-                               struct btrfs_root *root,
+static int setup_inline_extent_backref(struct btrfs_root *root,
                                struct btrfs_path *path,
                                struct btrfs_extent_inline_ref *iref,
                                u64 parent, u64 root_objectid,
@@ -1182,7 +1183,7 @@ static int setup_inline_extent_backref(struct btrfs_trans_handle *trans,
        type = extent_ref_type(parent, owner);
        size = btrfs_extent_inline_ref_size(type);
 
-       ret = btrfs_extend_item(trans, root, path, size);
+       ret = btrfs_extend_item(root, path, size);
        BUG_ON(ret);
 
        ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
@@ -1234,7 +1235,7 @@ static int lookup_extent_backref(struct btrfs_trans_handle *trans,
        if (ret != -ENOENT)
                return ret;
 
-       btrfs_release_path(root, path);
+       btrfs_release_path(path);
        *ref_ret = NULL;
 
        if (owner < BTRFS_FIRST_FREE_OBJECTID) {
@@ -1302,7 +1303,7 @@ static int update_inline_extent_backref(struct btrfs_trans_handle *trans,
                        memmove_extent_buffer(leaf, ptr, ptr + size,
                                              end - ptr - size);
                item_size -= size;
-               ret = btrfs_truncate_item(trans, root, path, item_size, 1);
+               ret = btrfs_truncate_item(root, path, item_size, 1);
                BUG_ON(ret);
        }
        btrfs_mark_buffer_dirty(leaf);
@@ -1327,7 +1328,7 @@ static int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
                ret = update_inline_extent_backref(trans, root, path, iref,
                                                   refs_to_add);
        } else if (ret == -ENOENT) {
-               ret = setup_inline_extent_backref(trans, root, path, iref,
+               ret = setup_inline_extent_backref(root, path, iref,
                                                  parent, root_objectid,
                                                  owner, offset, refs_to_add);
        }
@@ -1391,7 +1392,6 @@ int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
                return -ENOMEM;
 
        path->reada = 1;
-       path->leave_spinning = 1;
 
        ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
                                           path, bytenr, num_bytes, parent,
@@ -1410,10 +1410,9 @@ int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
        btrfs_set_extent_refs(leaf, item, refs + 1);
 
        btrfs_mark_buffer_dirty(leaf);
-       btrfs_release_path(root->fs_info->extent_root, path);
+       btrfs_release_path(path);
 
        path->reada = 1;
-       path->leave_spinning = 1;
 
        /* now insert the actual backref */
        ret = insert_extent_backref(trans, root->fs_info->extent_root,
@@ -1439,7 +1438,7 @@ int btrfs_extent_post_op(struct btrfs_trans_handle *trans,
 
 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
                             struct btrfs_root *root, u64 bytenr,
-                            u64 num_bytes, u64 *refs, u64 *flags)
+                            u64 offset, int metadata, u64 *refs, u64 *flags)
 {
        struct btrfs_path *path;
        int ret;
@@ -1450,16 +1449,56 @@ int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
        u64 num_refs;
        u64 extent_flags;
 
-       WARN_ON(num_bytes < root->sectorsize);
+       if (metadata &&
+           !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA)) {
+               offset = root->fs_info->nodesize;
+               metadata = 0;
+       }
+
        path = btrfs_alloc_path();
+       if (!path)
+               return -ENOMEM;
        path->reada = 1;
+
        key.objectid = bytenr;
-       key.offset = num_bytes;
-       btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
+       key.offset = offset;
+       if (metadata)
+               key.type = BTRFS_METADATA_ITEM_KEY;
+       else
+               key.type = BTRFS_EXTENT_ITEM_KEY;
+
+again:
        ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
                                0, 0);
        if (ret < 0)
                goto out;
+
+       /*
+        * Deal with the fact that we may have mixed SKINNY and normal refs.  If
+        * we didn't find what we wanted check and see if we have a normal ref
+        * right next to us, or re-search if we are on the edge of the leaf just
+        * to make sure.
+        */
+       if (ret > 0 && metadata) {
+               if (path->slots[0]) {
+                       path->slots[0]--;
+                       btrfs_item_key_to_cpu(path->nodes[0], &key,
+                                             path->slots[0]);
+                       if (key.objectid == bytenr &&
+                           key.type == BTRFS_EXTENT_ITEM_KEY &&
+                           key.offset == root->fs_info->nodesize)
+                               ret = 0;
+               }
+
+               if (ret) {
+                       btrfs_release_path(path);
+                       key.type = BTRFS_EXTENT_ITEM_KEY;
+                       key.offset = root->fs_info->nodesize;
+                       metadata = 0;
+                       goto again;
+               }
+       }
+
        if (ret != 0) {
                ret = -EIO;
                goto out;
@@ -1492,12 +1531,12 @@ int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
                *flags = extent_flags;
 out:
        btrfs_free_path(path);
-       return 0;
+       return ret;
 }
 
 int btrfs_set_block_flags(struct btrfs_trans_handle *trans,
                          struct btrfs_root *root,
-                         u64 bytenr, u64 num_bytes, u64 flags)
+                         u64 bytenr, int level, u64 flags)
 {
        struct btrfs_path *path;
        int ret;
@@ -1505,17 +1544,48 @@ int btrfs_set_block_flags(struct btrfs_trans_handle *trans,
        struct extent_buffer *l;
        struct btrfs_extent_item *item;
        u32 item_size;
+       int skinny_metadata =
+               btrfs_fs_incompat(root->fs_info, SKINNY_METADATA);
 
-       WARN_ON(num_bytes < root->sectorsize);
        path = btrfs_alloc_path();
+       if (!path)
+               return -ENOMEM;
        path->reada = 1;
+
        key.objectid = bytenr;
-       key.offset = num_bytes;
-       btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
+       if (skinny_metadata) {
+               key.offset = level;
+               key.type = BTRFS_METADATA_ITEM_KEY;
+       } else {
+               key.offset = root->fs_info->nodesize;
+               key.type = BTRFS_EXTENT_ITEM_KEY;
+       }
+
+again:
        ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
                                0, 0);
        if (ret < 0)
                goto out;
+
+       if (ret > 0 && skinny_metadata) {
+               skinny_metadata = 0;
+               if (path->slots[0]) {
+                       path->slots[0]--;
+                       btrfs_item_key_to_cpu(path->nodes[0], &key,
+                                             path->slots[0]);
+                       if (key.objectid == bytenr &&
+                           key.offset == root->fs_info->nodesize &&
+                           key.type == BTRFS_EXTENT_ITEM_KEY)
+                               ret = 0;
+               }
+               if (ret) {
+                       btrfs_release_path(path);
+                       key.offset = root->fs_info->nodesize;
+                       key.type = BTRFS_EXTENT_ITEM_KEY;
+                       goto again;
+               }
+       }
+
        if (ret != 0) {
                btrfs_print_leaf(root, path->nodes[0]);
                printk("failed to find block number %Lu\n",
@@ -1586,7 +1656,7 @@ static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
                cond_resched();
                if (level == 0) {
                        btrfs_item_key_to_cpu(buf, &key, i);
-                       if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
+                       if (key.type != BTRFS_EXTENT_DATA_KEY)
                                continue;
                        fi = btrfs_item_ptr(buf, i,
                                            struct btrfs_file_extent_item);
@@ -1608,7 +1678,7 @@ static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
                        }
                } else {
                        bytenr = btrfs_node_blockptr(buf, i);
-                       num_bytes = btrfs_level_size(root, level - 1);
+                       num_bytes = root->fs_info->nodesize;
                        ret = process_func(trans, root, bytenr, num_bytes,
                                           parent, ref_root, level - 1, 0);
                        if (ret) {
@@ -1655,7 +1725,7 @@ static int write_one_cache_group(struct btrfs_trans_handle *trans,
        bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
        write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
        btrfs_mark_buffer_dirty(leaf);
-       btrfs_release_path(extent_root, path);
+       btrfs_release_path(path);
 fail:
        finish_current_insert(trans, extent_root);
        pending_ret = del_pending_extents(trans, extent_root);
@@ -1699,11 +1769,10 @@ int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
                BUG_ON(ret);
 
                clear_extent_bits(block_group_cache, start, end,
-                                 BLOCK_GROUP_DIRTY, GFP_NOFS);
+                                 BLOCK_GROUP_DIRTY);
 
                cache = (struct btrfs_block_group_cache *)(unsigned long)ptr;
                ret = write_one_cache_group(trans, root, path, cache);
-               BUG_ON(ret);
        }
        btrfs_free_path(path);
        return 0;
@@ -1712,11 +1781,11 @@ int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
                                                  u64 flags)
 {
-       struct list_head *head = &info->space_info;
-       struct list_head *cur;
        struct btrfs_space_info *found;
-       list_for_each(cur, head) {
-               found = list_entry(cur, struct btrfs_space_info, list);
+
+       flags &= BTRFS_BLOCK_GROUP_TYPE_MASK;
+
+       list_for_each_entry(found, &info->space_info, list) {
                if (found->flags & flags)
                        return found;
        }
@@ -1724,6 +1793,31 @@ static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
 
 }
 
+static int free_space_info(struct btrfs_fs_info *fs_info, u64 flags,
+                          u64 total_bytes, u64 bytes_used,
+                          struct btrfs_space_info **space_info)
+{
+       struct btrfs_space_info *found;
+
+       /* only support free block group which is empty */
+       if (bytes_used)
+               return -ENOTEMPTY;
+
+       found = __find_space_info(fs_info, flags);
+       if (!found)
+               return -ENOENT;
+       if (found->total_bytes < total_bytes) {
+               fprintf(stderr,
+                       "WARNING: bad space info to free %llu only have %llu\n",
+                       total_bytes, found->total_bytes);
+               return -EINVAL;
+       }
+       found->total_bytes -= total_bytes;
+       if (space_info)
+               *space_info = found;
+       return 0;
+}
+
 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
                             u64 total_bytes, u64 bytes_used,
                             struct btrfs_space_info **space_info)
@@ -1748,7 +1842,7 @@ static int update_space_info(struct btrfs_fs_info *info, u64 flags,
                return -ENOMEM;
 
        list_add(&found->list, &info->space_info);
-       found->flags = flags;
+       found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
        found->total_bytes = total_bytes;
        found->bytes_used = bytes_used;
        found->bytes_pinned = 0;
@@ -1763,6 +1857,8 @@ static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
        u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
                                   BTRFS_BLOCK_GROUP_RAID1 |
                                   BTRFS_BLOCK_GROUP_RAID10 |
+                                  BTRFS_BLOCK_GROUP_RAID5 |
+                                  BTRFS_BLOCK_GROUP_RAID6 |
                                   BTRFS_BLOCK_GROUP_DUP);
        if (extra_flags) {
                if (flags & BTRFS_BLOCK_GROUP_DATA)
@@ -1775,7 +1871,7 @@ static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
 }
 
 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
-                         struct btrfs_root *extent_root, u64 alloc_bytes,
+                         struct btrfs_fs_info *fs_info, u64 alloc_bytes,
                          u64 flags)
 {
        struct btrfs_space_info *space_info;
@@ -1784,10 +1880,9 @@ static int do_chunk_alloc(struct btrfs_trans_handle *trans,
        u64 num_bytes;
        int ret;
 
-       space_info = __find_space_info(extent_root->fs_info, flags);
+       space_info = __find_space_info(fs_info, flags);
        if (!space_info) {
-               ret = update_space_info(extent_root->fs_info, flags,
-                                       0, 0, &space_info);
+               ret = update_space_info(fs_info, flags, 0, 0, &space_info);
                BUG_ON(ret);
        }
        BUG_ON(!space_info);
@@ -1800,7 +1895,17 @@ static int do_chunk_alloc(struct btrfs_trans_handle *trans,
            thresh)
                return 0;
 
-       ret = btrfs_alloc_chunk(trans, extent_root, &start, &num_bytes,
+       /*
+        * Avoid allocating given chunk type
+        */
+       if (fs_info->avoid_meta_chunk_alloc &&
+           (flags & BTRFS_BLOCK_GROUP_METADATA))
+               return 0;
+       if (fs_info->avoid_sys_chunk_alloc &&
+           (flags & BTRFS_BLOCK_GROUP_SYSTEM))
+               return 0;
+
+       ret = btrfs_alloc_chunk(trans, fs_info, &start, &num_bytes,
                                space_info->flags);
        if (ret == -ENOSPC) {
                space_info->full = 1;
@@ -1809,14 +1914,13 @@ static int do_chunk_alloc(struct btrfs_trans_handle *trans,
 
        BUG_ON(ret);
 
-       ret = btrfs_make_block_group(trans, extent_root, 0, space_info->flags,
-                    BTRFS_FIRST_CHUNK_TREE_OBJECTID, start, num_bytes);
+       ret = btrfs_make_block_group(trans, fs_info, 0, space_info->flags,
+                                    start, num_bytes);
        BUG_ON(ret);
        return 0;
 }
 
-static int update_block_group(struct btrfs_trans_handle *trans,
-                             struct btrfs_root *root,
+static int update_block_group(struct btrfs_root *root,
                              u64 bytenr, u64 num_bytes, int alloc,
                              int mark_free)
 {
@@ -1829,12 +1933,12 @@ static int update_block_group(struct btrfs_trans_handle *trans,
        u64 end;
 
        /* block accounting for super block */
-       old_val = btrfs_super_bytes_used(&info->super_copy);
+       old_val = btrfs_super_bytes_used(info->super_copy);
        if (alloc)
                old_val += num_bytes;
        else
                old_val -= num_bytes;
-       btrfs_set_super_bytes_used(&info->super_copy, old_val);
+       btrfs_set_super_bytes_used(info->super_copy, old_val);
 
        /* block accounting for root item */
        old_val = btrfs_root_used(&root->root_item);
@@ -1854,7 +1958,7 @@ static int update_block_group(struct btrfs_trans_handle *trans,
                start = cache->key.objectid;
                end = start + cache->key.offset - 1;
                set_extent_bits(&info->block_group_cache, start, end,
-                               BLOCK_GROUP_DIRTY, GFP_NOFS);
+                               BLOCK_GROUP_DIRTY);
 
                old_val = btrfs_block_group_used(&cache->item);
                num_bytes = min(total, cache->key.offset - byte_in_group);
@@ -1867,8 +1971,7 @@ static int update_block_group(struct btrfs_trans_handle *trans,
                        cache->space_info->bytes_used -= num_bytes;
                        if (mark_free) {
                                set_extent_dirty(&info->free_space_cache,
-                                                bytenr, bytenr + num_bytes - 1,
-                                                GFP_NOFS);
+                                               bytenr, bytenr + num_bytes - 1);
                        }
                }
                btrfs_set_block_group_used(&cache->item, old_val);
@@ -1887,13 +1990,17 @@ static int update_pinned_extents(struct btrfs_root *root,
 
        if (pin) {
                set_extent_dirty(&fs_info->pinned_extents,
-                               bytenr, bytenr + num - 1, GFP_NOFS);
+                               bytenr, bytenr + num - 1);
        } else {
                clear_extent_dirty(&fs_info->pinned_extents,
-                               bytenr, bytenr + num - 1, GFP_NOFS);
+                               bytenr, bytenr + num - 1);
        }
        while (num > 0) {
                cache = btrfs_lookup_block_group(fs_info, bytenr);
+               if (!cache) {
+                       len = min((u64)fs_info->sectorsize, num);
+                       goto next;
+               }
                WARN_ON(!cache);
                len = min(num, cache->key.offset -
                          (bytenr - cache->key.objectid));
@@ -1906,31 +2013,13 @@ static int update_pinned_extents(struct btrfs_root *root,
                        cache->space_info->bytes_pinned -= len;
                        fs_info->total_pinned -= len;
                }
+next:
                bytenr += len;
                num -= len;
        }
        return 0;
 }
 
-int btrfs_copy_pinned(struct btrfs_root *root, struct extent_io_tree *copy)
-{
-       u64 last = 0;
-       u64 start;
-       u64 end;
-       struct extent_io_tree *pinned_extents = &root->fs_info->pinned_extents;
-       int ret;
-
-       while(1) {
-               ret = find_first_extent_bit(pinned_extents, last,
-                                           &start, &end, EXTENT_DIRTY);
-               if (ret)
-                       break;
-               set_extent_dirty(copy, start, end, GFP_NOFS);
-               last = end + 1;
-       }
-       return 0;
-}
-
 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
                               struct btrfs_root *root,
                               struct extent_io_tree *unpin)
@@ -1947,12 +2036,27 @@ int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
                if (ret)
                        break;
                update_pinned_extents(root, start, end + 1 - start, 0);
-               clear_extent_dirty(unpin, start, end, GFP_NOFS);
-               set_extent_dirty(free_space_cache, start, end, GFP_NOFS);
+               clear_extent_dirty(unpin, start, end);
+               set_extent_dirty(free_space_cache, start, end);
        }
        return 0;
 }
 
+static int extent_root_pending_ops(struct btrfs_fs_info *info)
+{
+       u64 start;
+       u64 end;
+       int ret;
+
+       ret = find_first_extent_bit(&info->extent_ins, 0, &start,
+                                   &end, EXTENT_LOCKED);
+       if (!ret) {
+               ret = find_first_extent_bit(&info->pending_del, 0, &start, &end,
+                                           EXTENT_LOCKED);
+       }
+       return ret == 0;
+
+}
 static int finish_current_insert(struct btrfs_trans_handle *trans,
                                 struct btrfs_root *extent_root)
 {
@@ -1960,12 +2064,11 @@ static int finish_current_insert(struct btrfs_trans_handle *trans,
        u64 end;
        u64 priv;
        struct btrfs_fs_info *info = extent_root->fs_info;
-       struct btrfs_path *path;
        struct pending_extent_op *extent_op;
        struct btrfs_key key;
        int ret;
-
-       path = btrfs_alloc_path();
+       int skinny_metadata =
+               btrfs_fs_incompat(extent_root->fs_info, SKINNY_METADATA);
 
        while(1) {
                ret = find_first_extent_bit(&info->extent_ins, 0, &start,
@@ -1979,23 +2082,27 @@ static int finish_current_insert(struct btrfs_trans_handle *trans,
 
                if (extent_op->type == PENDING_EXTENT_INSERT) {
                        key.objectid = start;
-                       key.offset = end + 1 - start;
-                       key.type = BTRFS_EXTENT_ITEM_KEY;
+                       if (skinny_metadata) {
+                               key.offset = extent_op->level;
+                               key.type = BTRFS_METADATA_ITEM_KEY;
+                       } else {
+                               key.offset = extent_op->num_bytes;
+                               key.type = BTRFS_EXTENT_ITEM_KEY;
+                       }
                        ret = alloc_reserved_tree_block(trans, extent_root,
                                                extent_root->root_key.objectid,
                                                trans->transid,
                                                extent_op->flags,
                                                &extent_op->key,
-                                               extent_op->level, &key);
+                                               extent_op->level, &key);
+                       BUG_ON(ret);
                } else {
                        BUG_ON(1);
                }
 
-               clear_extent_bits(&info->extent_ins, start, end, EXTENT_LOCKED,
-                                 GFP_NOFS);
+               clear_extent_bits(&info->extent_ins, start, end, EXTENT_LOCKED);
                kfree(extent_op);
        }
-       btrfs_free_path(path);
        return 0;
 }
 
@@ -2009,7 +2116,7 @@ static int pin_down_bytes(struct btrfs_trans_handle *trans,
        if (is_data)
                goto pinit;
 
-       buf = btrfs_find_tree_block(root, bytenr, num_bytes);
+       buf = btrfs_find_tree_block(root->fs_info, bytenr, num_bytes);
        if (!buf)
                goto pinit;
 
@@ -2043,6 +2150,12 @@ void btrfs_pin_extent(struct btrfs_fs_info *fs_info,
        update_pinned_extents(fs_info->extent_root, bytenr, num_bytes, 1);
 }
 
+void btrfs_unpin_extent(struct btrfs_fs_info *fs_info,
+                       u64 bytenr, u64 num_bytes)
+{
+       update_pinned_extents(fs_info->extent_root, bytenr, num_bytes, 0);
+}
+
 /*
  * remove an extent from the root, returns 0 on success
  */
@@ -2055,7 +2168,6 @@ static int __free_extent(struct btrfs_trans_handle *trans,
 
        struct btrfs_key key;
        struct btrfs_path *path;
-       struct btrfs_extent_ops *ops = root->fs_info->extent_ops;
        struct btrfs_root *extent_root = root->fs_info->extent_root;
        struct extent_buffer *leaf;
        struct btrfs_extent_item *ei;
@@ -2067,15 +2179,24 @@ static int __free_extent(struct btrfs_trans_handle *trans,
        int num_to_del = 1;
        u32 item_size;
        u64 refs;
+       int skinny_metadata =
+               btrfs_fs_incompat(extent_root->fs_info, SKINNY_METADATA);
 
+       if (root->fs_info->free_extent_hook) {
+               root->fs_info->free_extent_hook(trans, root, bytenr, num_bytes,
+                                               parent, root_objectid, owner_objectid,
+                                               owner_offset, refs_to_drop);
+
+       }
        path = btrfs_alloc_path();
        if (!path)
                return -ENOMEM;
 
        path->reada = 1;
-       path->leave_spinning = 1;
 
        is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
+       if (is_data)
+               skinny_metadata = 0;
        BUG_ON(!is_data && refs_to_drop != 1);
 
        ret = lookup_extent_backref(trans, extent_root, path, &iref,
@@ -2094,6 +2215,11 @@ static int __free_extent(struct btrfs_trans_handle *trans,
                                found_extent = 1;
                                break;
                        }
+                       if (key.type == BTRFS_METADATA_ITEM_KEY &&
+                           key.offset == owner_objectid) {
+                               found_extent = 1;
+                               break;
+                       }
                        if (path->slots[0] - extent_slot > 5)
                                break;
                        extent_slot--;
@@ -2109,15 +2235,40 @@ static int __free_extent(struct btrfs_trans_handle *trans,
                                                    NULL, refs_to_drop,
                                                    is_data);
                        BUG_ON(ret);
-                       btrfs_release_path(extent_root, path);
-                       path->leave_spinning = 1;
+                       btrfs_release_path(path);
 
                        key.objectid = bytenr;
-                       key.type = BTRFS_EXTENT_ITEM_KEY;
-                       key.offset = num_bytes;
+
+                       if (skinny_metadata) {
+                               key.type = BTRFS_METADATA_ITEM_KEY;
+                               key.offset = owner_objectid;
+                       } else {
+                               key.type = BTRFS_EXTENT_ITEM_KEY;
+                               key.offset = num_bytes;
+                       }
 
                        ret = btrfs_search_slot(trans, extent_root,
                                                &key, path, -1, 1);
+                       if (ret > 0 && skinny_metadata && path->slots[0]) {
+                               path->slots[0]--;
+                               btrfs_item_key_to_cpu(path->nodes[0],
+                                                     &key,
+                                                     path->slots[0]);
+                               if (key.objectid == bytenr &&
+                                   key.type == BTRFS_EXTENT_ITEM_KEY &&
+                                   key.offset == num_bytes)
+                                       ret = 0;
+                       }
+
+                       if (ret > 0 && skinny_metadata) {
+                               skinny_metadata = 0;
+                               btrfs_release_path(path);
+                               key.type = BTRFS_EXTENT_ITEM_KEY;
+                               key.offset = num_bytes;
+                               ret = btrfs_search_slot(trans, extent_root,
+                                                       &key, path, -1, 1);
+                       }
+
                        if (ret) {
                                printk(KERN_ERR "umm, got %d back from search"
                                       ", was looking for %llu\n", ret,
@@ -2148,8 +2299,7 @@ static int __free_extent(struct btrfs_trans_handle *trans,
                                             owner_objectid, 0);
                BUG_ON(ret < 0);
 
-               btrfs_release_path(extent_root, path);
-               path->leave_spinning = 1;
+               btrfs_release_path(path);
 
                key.objectid = bytenr;
                key.type = BTRFS_EXTENT_ITEM_KEY;
@@ -2172,7 +2322,8 @@ static int __free_extent(struct btrfs_trans_handle *trans,
        BUG_ON(item_size < sizeof(*ei));
        ei = btrfs_item_ptr(leaf, extent_slot,
                            struct btrfs_extent_item);
-       if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
+       if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
+           key.type == BTRFS_EXTENT_ITEM_KEY) {
                struct btrfs_tree_block_info *bi;
                BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
                bi = (struct btrfs_tree_block_info *)(ei + 1);
@@ -2206,7 +2357,7 @@ static int __free_extent(struct btrfs_trans_handle *trans,
 
                if (found_extent) {
                        BUG_ON(is_data && refs_to_drop !=
-                              extent_data_ref_count(root, path, iref));
+                              extent_data_ref_count(path, iref));
                        if (iref) {
                                BUG_ON(path->slots[0] != extent_slot);
                        } else {
@@ -2216,14 +2367,6 @@ static int __free_extent(struct btrfs_trans_handle *trans,
                        }
                }
 
-               if (ops && ops->free_extent) {
-                       ret = ops->free_extent(root, bytenr, num_bytes);
-                       if (ret > 0) {
-                               pin = 0;
-                               mark_free = 0;
-                       }
-               }
-
                if (pin) {
                        ret = pin_down_bytes(trans, root, bytenr, num_bytes,
                                             is_data);
@@ -2235,16 +2378,14 @@ static int __free_extent(struct btrfs_trans_handle *trans,
                ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
                                      num_to_del);
                BUG_ON(ret);
-               btrfs_release_path(extent_root, path);
+               btrfs_release_path(path);
 
                if (is_data) {
                        ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
                        BUG_ON(ret);
                }
 
-               ret = update_block_group(trans, root, bytenr, num_bytes, 0,
-                                        mark_free);
-               BUG_ON(ret);
+               update_block_group(root, bytenr, num_bytes, 0, mark_free);
        }
 fail:
        btrfs_free_path(path);
@@ -2281,8 +2422,7 @@ static int del_pending_extents(struct btrfs_trans_handle *trans, struct
                BUG_ON(ret);
                extent_op = (struct pending_extent_op *)(unsigned long)priv;
 
-               clear_extent_bits(pending_del, start, end, EXTENT_LOCKED,
-                                 GFP_NOFS);
+               clear_extent_bits(pending_del, start, end, EXTENT_LOCKED);
 
                if (!test_range_bit(extent_ins, start, end,
                                    EXTENT_LOCKED, 0)) {
@@ -2299,7 +2439,7 @@ static int del_pending_extents(struct btrfs_trans_handle *trans, struct
                                                        (unsigned long)priv;
 
                        clear_extent_bits(extent_ins, start, end,
-                                         EXTENT_LOCKED, GFP_NOFS);
+                                         EXTENT_LOCKED);
 
                        if (extent_op->type == PENDING_BACKREF_UPDATE)
                                BUG_ON(1);
@@ -2312,6 +2452,17 @@ static int del_pending_extents(struct btrfs_trans_handle *trans, struct
        return err;
 }
 
+
+int btrfs_free_tree_block(struct btrfs_trans_handle *trans,
+                         struct btrfs_root *root,
+                         struct extent_buffer *buf,
+                         u64 parent, int last_ref)
+{
+       return btrfs_free_extent(trans, root, buf->start, buf->len, parent,
+                                root->root_key.objectid,
+                                btrfs_header_level(buf), 0);
+}
+
 /*
  * remove an extent from the root, returns 0 on success
  */
@@ -2325,7 +2476,7 @@ int btrfs_free_extent(struct btrfs_trans_handle *trans,
        int pending_ret;
        int ret;
 
-       WARN_ON(num_bytes < root->sectorsize);
+       WARN_ON(num_bytes < root->fs_info->sectorsize);
        if (root == extent_root) {
                struct pending_extent_op *extent_op;
 
@@ -2339,7 +2490,7 @@ int btrfs_free_extent(struct btrfs_trans_handle *trans,
 
                set_extent_bits(&root->fs_info->pending_del,
                                bytenr, bytenr + num_bytes - 1,
-                               EXTENT_LOCKED, GFP_NOFS);
+                               EXTENT_LOCKED);
                set_state_private(&root->fs_info->pending_del,
                                  bytenr, (unsigned long)extent_op);
                return 0;
@@ -2352,9 +2503,7 @@ int btrfs_free_extent(struct btrfs_trans_handle *trans,
 
 static u64 stripe_align(struct btrfs_root *root, u64 val)
 {
-       u64 mask = ((u64)root->stripesize - 1);
-       u64 ret = (val + mask) & ~mask;
-       return ret;
+       return round_up(val, (u64)root->fs_info->stripesize);
 }
 
 /*
@@ -2382,8 +2531,10 @@ static int noinline find_free_extent(struct btrfs_trans_handle *trans,
        int full_scan = 0;
        int wrapped = 0;
 
-       WARN_ON(num_bytes < root->sectorsize);
-       btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
+       WARN_ON(num_bytes < info->sectorsize);
+       ins->type = BTRFS_EXTENT_ITEM_KEY;
+
+       search_start = stripe_align(root, search_start);
 
        if (hint_byte) {
                block_group = btrfs_lookup_first_block_group(info, hint_byte);
@@ -2400,6 +2551,7 @@ static int noinline find_free_extent(struct btrfs_trans_handle *trans,
        total_needed += empty_size;
 
 check_failed:
+       search_start = stripe_align(root, search_start);
        if (!block_group) {
                block_group = btrfs_lookup_first_block_group(info,
                                                             search_start);
@@ -2410,9 +2562,8 @@ check_failed:
        ret = find_search_start(root, &block_group, &search_start,
                                total_needed, data);
        if (ret)
-               goto error;
+               goto new_group;
 
-       search_start = stripe_align(root, search_start);
        ins->objectid = search_start;
        ins->offset = num_bytes;
 
@@ -2435,6 +2586,13 @@ check_failed:
                goto new_group;
        }
 
+       if (info->excluded_extents &&
+           test_range_bit(info->excluded_extents, ins->objectid,
+                          ins->objectid + num_bytes -1, EXTENT_DIRTY, 0)) {
+               search_start = ins->objectid + num_bytes;
+               goto new_group;
+       }
+
        if (exclude_nr > 0 && (ins->objectid + num_bytes > exclude_start &&
            ins->objectid < exclude_start + exclude_nr)) {
                search_start = exclude_start + exclude_nr;
@@ -2442,6 +2600,21 @@ check_failed:
        }
 
        if (!(data & BTRFS_BLOCK_GROUP_DATA)) {
+               if (check_crossing_stripes(info, ins->objectid, num_bytes)) {
+                       struct btrfs_block_group_cache *bg_cache;
+                       u64 bg_offset;
+
+                       bg_cache = btrfs_lookup_block_group(info, ins->objectid);
+                       if (!bg_cache)
+                               goto no_bg_cache;
+                       bg_offset = ins->objectid - bg_cache->key.objectid;
+
+                       search_start = round_up(
+                               bg_offset + num_bytes, BTRFS_STRIPE_LEN) +
+                               bg_cache->key.objectid;
+                       goto new_group;
+               }
+no_bg_cache:
                block_group = btrfs_lookup_block_group(info, ins->objectid);
                if (block_group)
                        trans->block_group = block_group;
@@ -2473,61 +2646,53 @@ error:
        return ret;
 }
 
-static int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
-                               struct btrfs_root *root,
-                               u64 num_bytes, u64 empty_size,
-                               u64 hint_byte, u64 search_end,
-                               struct btrfs_key *ins, int data)
+int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
+                        struct btrfs_root *root,
+                        u64 num_bytes, u64 empty_size,
+                        u64 hint_byte, u64 search_end,
+                        struct btrfs_key *ins, bool is_data)
 {
        int ret;
        u64 search_start = 0;
        u64 alloc_profile;
+       u64 profile;
        struct btrfs_fs_info *info = root->fs_info;
 
-       if (info->extent_ops) {
-               struct btrfs_extent_ops *ops = info->extent_ops;
-               ret = ops->alloc_extent(root, num_bytes, hint_byte, ins);
-               BUG_ON(ret);
-               goto found;
-       }
-
-       if (data) {
+       if (is_data) {
                alloc_profile = info->avail_data_alloc_bits &
                                info->data_alloc_profile;
-               data = BTRFS_BLOCK_GROUP_DATA | alloc_profile;
-       } else if ((info->system_allocs > 0 || root == info->chunk_root) &&
-                  info->system_allocs >= 0) {
+               profile = BTRFS_BLOCK_GROUP_DATA | alloc_profile;
+       } else if (info->system_allocs == 1 || root == info->chunk_root) {
                alloc_profile = info->avail_system_alloc_bits &
                                info->system_alloc_profile;
-               data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile;
+               profile = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile;
        } else {
                alloc_profile = info->avail_metadata_alloc_bits &
                                info->metadata_alloc_profile;
-               data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile;
+               profile = BTRFS_BLOCK_GROUP_METADATA | alloc_profile;
        }
 
        if (root->ref_cows) {
-               if (!(data & BTRFS_BLOCK_GROUP_METADATA)) {
-                       ret = do_chunk_alloc(trans, root->fs_info->extent_root,
+               if (!(profile & BTRFS_BLOCK_GROUP_METADATA)) {
+                       ret = do_chunk_alloc(trans, info,
                                             num_bytes,
                                             BTRFS_BLOCK_GROUP_METADATA);
                        BUG_ON(ret);
                }
-               ret = do_chunk_alloc(trans, root->fs_info->extent_root,
-                                    num_bytes + 2 * 1024 * 1024, data);
+               ret = do_chunk_alloc(trans, info,
+                                    num_bytes + SZ_2M, profile);
                BUG_ON(ret);
        }
 
-       WARN_ON(num_bytes < root->sectorsize);
+       WARN_ON(num_bytes < info->sectorsize);
        ret = find_free_extent(trans, root, num_bytes, empty_size,
                               search_start, search_end, hint_byte, ins,
                               trans->alloc_exclude_start,
-                              trans->alloc_exclude_nr, data);
-       BUG_ON(ret);
-found:
-       clear_extent_dirty(&root->fs_info->free_space_cache,
-                          ins->objectid, ins->objectid + ins->offset - 1,
-                          GFP_NOFS);
+                              trans->alloc_exclude_nr, profile);
+       if (ret < 0)
+               return ret;
+       clear_extent_dirty(&info->free_space_cache,
+                          ins->objectid, ins->objectid + ins->offset - 1);
        return ret;
 }
 
@@ -2544,12 +2709,16 @@ static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
        struct btrfs_extent_inline_ref *iref;
        struct btrfs_path *path;
        struct extent_buffer *leaf;
-       u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref);
+       u32 size = sizeof(*extent_item) + sizeof(*iref);
+       int skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
+
+       if (!skinny_metadata)
+               size += sizeof(*block_info);
 
        path = btrfs_alloc_path();
-       BUG_ON(!path);
+       if (!path)
+               return -ENOMEM;
 
-       path->leave_spinning = 1;
        ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
                                      ins, size);
        BUG_ON(ret);
@@ -2561,26 +2730,24 @@ static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
        btrfs_set_extent_generation(leaf, extent_item, generation);
        btrfs_set_extent_flags(leaf, extent_item,
                               flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
-       block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
 
-       btrfs_set_tree_block_key(leaf, block_info, key);
-       btrfs_set_tree_block_level(leaf, block_info, level);
+       if (skinny_metadata) {
+               iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
+       } else {
+               block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
+               btrfs_set_tree_block_key(leaf, block_info, key);
+               btrfs_set_tree_block_level(leaf, block_info, level);
+               iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
+       }
 
-       iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
        btrfs_set_extent_inline_ref_type(leaf, iref, BTRFS_TREE_BLOCK_REF_KEY);
        btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
 
        btrfs_mark_buffer_dirty(leaf);
        btrfs_free_path(path);
 
-       ret = update_block_group(trans, root, ins->objectid, ins->offset,
+       ret = update_block_group(root, ins->objectid, fs_info->nodesize,
                                 1, 0);
-       if (ret) {
-               printk(KERN_ERR "btrfs update block group failed for %llu "
-                      "%llu\n", (unsigned long long)ins->objectid,
-                      (unsigned long long)ins->offset);
-               BUG();
-       }
        return ret;
 }
 
@@ -2611,10 +2778,14 @@ static int alloc_tree_block(struct btrfs_trans_handle *trans,
 
                set_extent_bits(&root->fs_info->extent_ins, ins->objectid,
                                ins->objectid + ins->offset - 1,
-                               EXTENT_LOCKED, GFP_NOFS);
+                               EXTENT_LOCKED);
                set_state_private(&root->fs_info->extent_ins,
                                  ins->objectid, (unsigned long)extent_op);
        } else {
+               if (btrfs_fs_incompat(root->fs_info, SKINNY_METADATA)) {
+                       ins->offset = level;
+                       ins->type = BTRFS_METADATA_ITEM_KEY;
+               }
                ret = alloc_reserved_tree_block(trans, root, root_objectid,
                                                generation, flags,
                                                key, level, ins);
@@ -2646,7 +2817,7 @@ struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
                return ERR_PTR(ret);
        }
 
-       buf = btrfs_find_create_tree_block(root, ins.objectid, blocksize);
+       buf = btrfs_find_create_tree_block(root->fs_info, ins.objectid);
        if (!buf) {
                btrfs_free_extent(trans, root, ins.objectid, ins.offset,
                                  0, root->root_key.objectid, level, 0);
@@ -2725,8 +2896,8 @@ static void noinline reada_walk_down(struct btrfs_root *root,
 
        for (i = slot; i < nritems && skipped < 32; i++) {
                bytenr = btrfs_node_blockptr(node, i);
-               if (last && ((bytenr > last && bytenr - last > 32 * 1024) ||
-                            (last > bytenr && last - bytenr > 32 * 1024))) {
+               if (last && ((bytenr > last && bytenr - last > SZ_32K) ||
+                            (last > bytenr && last - bytenr > SZ_32K))) {
                        skipped++;
                        continue;
                }
@@ -2823,6 +2994,13 @@ static int noinline walk_down_tree(struct btrfs_trans_handle *trans,
                        next = read_tree_block(root, bytenr, blocksize,
                                               ptr_gen);
                        mutex_lock(&root->fs_info->fs_mutex);
+                       if (!extent_buffer_uptodate(next)) {
+                               if (IS_ERR(next))
+                                       ret = PTR_ERR(next);
+                               else
+                                       ret = -EIO;
+                               break;
+                       }
                }
                WARN_ON(*level <= 0);
                if (path->nodes[*level-1])
@@ -2907,109 +3085,53 @@ static int noinline walk_up_tree(struct btrfs_trans_handle *trans,
        return 1;
 }
 
-/*
- * drop the reference count on the tree rooted at 'snap'.  This traverses
- * the tree freeing any blocks that have a ref count of zero after being
- * decremented.
- */
-int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root
-                       *root)
-{
-       int ret = 0;
-       int wret;
-       int level;
-       struct btrfs_path *path;
-       int i;
-       int orig_level;
-       struct btrfs_root_item *root_item = &root->root_item;
-
-       path = btrfs_alloc_path();
-       BUG_ON(!path);
-
-       level = btrfs_header_level(root->node);
-       orig_level = level;
-       if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
-               path->nodes[level] = root->node;
-               extent_buffer_get(root->node);
-               path->slots[level] = 0;
-       } else {
-               struct btrfs_key key;
-               struct btrfs_disk_key found_key;
-               struct extent_buffer *node;
-
-               btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
-               level = root_item->drop_level;
-               path->lowest_level = level;
-               wret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
-               if (wret < 0) {
-                       ret = wret;
-                       goto out;
-               }
-               node = path->nodes[level];
-               btrfs_node_key(node, &found_key, path->slots[level]);
-               WARN_ON(memcmp(&found_key, &root_item->drop_progress,
-                              sizeof(found_key)));
-       }
-       while(1) {
-               wret = walk_down_tree(trans, root, path, &level);
-               if (wret < 0)
-                       ret = wret;
-               if (wret != 0)
-                       break;
-
-               wret = walk_up_tree(trans, root, path, &level);
-               if (wret < 0)
-                       ret = wret;
-               if (wret != 0)
-                       break;
-               /*
-               ret = -EAGAIN;
-               break;
-               */
-       }
-       for (i = 0; i <= orig_level; i++) {
-               if (path->nodes[i]) {
-                       free_extent_buffer(path->nodes[i]);
-                       path->nodes[i] = NULL;
-               }
-       }
-out:
-       btrfs_free_path(path);
-       return ret;
-}
-
 #endif
 
 int btrfs_free_block_groups(struct btrfs_fs_info *info)
 {
+       struct btrfs_space_info *sinfo;
+       struct btrfs_block_group_cache *cache;
        u64 start;
        u64 end;
        u64 ptr;
        int ret;
+
        while(1) {
                ret = find_first_extent_bit(&info->block_group_cache, 0,
                                            &start, &end, (unsigned int)-1);
                if (ret)
                        break;
                ret = get_state_private(&info->block_group_cache, start, &ptr);
-               if (!ret)
-                       kfree((void *)(unsigned long)ptr);
+               if (!ret) {
+                       cache = u64_to_ptr(ptr);
+                       if (cache->free_space_ctl) {
+                               btrfs_remove_free_space_cache(cache);
+                               kfree(cache->free_space_ctl);
+                       }
+                       kfree(cache);
+               }
                clear_extent_bits(&info->block_group_cache, start,
-                                 end, (unsigned int)-1, GFP_NOFS);
+                                 end, (unsigned int)-1);
        }
        while(1) {
                ret = find_first_extent_bit(&info->free_space_cache, 0,
                                            &start, &end, EXTENT_DIRTY);
                if (ret)
                        break;
-               clear_extent_dirty(&info->free_space_cache, start,
-                                  end, GFP_NOFS);
+               clear_extent_dirty(&info->free_space_cache, start, end);
+       }
+
+       while (!list_empty(&info->space_info)) {
+               sinfo = list_entry(info->space_info.next,
+                                  struct btrfs_space_info, list);
+               list_del_init(&sinfo->list);
+               kfree(sinfo);
        }
        return 0;
 }
 
-int find_first_block_group(struct btrfs_root *root, struct btrfs_path *path,
-                          struct btrfs_key *key)
+static int find_first_block_group(struct btrfs_root *root,
+               struct btrfs_path *path, struct btrfs_key *key)
 {
        int ret;
        struct btrfs_key found_key;
@@ -3042,25 +3164,73 @@ error:
        return ret;
 }
 
-int btrfs_read_block_groups(struct btrfs_root *root)
+static void account_super_bytes(struct btrfs_fs_info *fs_info,
+                               struct btrfs_block_group_cache *cache)
 {
-       struct btrfs_path *path;
-       int ret;
-       int bit;
-       struct btrfs_block_group_cache *cache;
-       struct btrfs_fs_info *info = root->fs_info;
-       struct btrfs_space_info *space_info;
-       struct extent_io_tree *block_group_cache;
-       struct btrfs_key key;
-       struct btrfs_key found_key;
-       struct extent_buffer *leaf;
+       u64 bytenr;
+       u64 *logical;
+       int stripe_len;
+       int i, nr, ret;
 
-       block_group_cache = &info->block_group_cache;
+       if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
+               stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
+               cache->bytes_super += stripe_len;
+       }
+
+       for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
+               bytenr = btrfs_sb_offset(i);
+               ret = btrfs_rmap_block(fs_info,
+                                      cache->key.objectid, bytenr,
+                                      0, &logical, &nr, &stripe_len);
+               if (ret)
+                       return;
+
+               while (nr--) {
+                       u64 start, len;
+
+                       if (logical[nr] > cache->key.objectid +
+                           cache->key.offset)
+                               continue;
+
+                       if (logical[nr] + stripe_len <= cache->key.objectid)
+                               continue;
+
+                       start = logical[nr];
+                       if (start < cache->key.objectid) {
+                               start = cache->key.objectid;
+                               len = (logical[nr] + stripe_len) - start;
+                       } else {
+                               len = min_t(u64, stripe_len,
+                                           cache->key.objectid +
+                                           cache->key.offset - start);
+                       }
+
+                       cache->bytes_super += len;
+               }
+
+               kfree(logical);
+       }
+}
+
+int btrfs_read_block_groups(struct btrfs_root *root)
+{
+       struct btrfs_path *path;
+       int ret;
+       int bit;
+       struct btrfs_block_group_cache *cache;
+       struct btrfs_fs_info *info = root->fs_info;
+       struct btrfs_space_info *space_info;
+       struct extent_io_tree *block_group_cache;
+       struct btrfs_key key;
+       struct btrfs_key found_key;
+       struct extent_buffer *leaf;
+
+       block_group_cache = &info->block_group_cache;
 
        root = info->extent_root;
        key.objectid = 0;
        key.offset = 0;
-       btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
+       key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
        path = btrfs_alloc_path();
        if (!path)
                return -ENOMEM;
@@ -3076,10 +3246,11 @@ int btrfs_read_block_groups(struct btrfs_root *root)
                }
                leaf = path->nodes[0];
                btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
+
                cache = kzalloc(sizeof(*cache), GFP_NOFS);
                if (!cache) {
                        ret = -ENOMEM;
-                       break;
+                       goto error;
                }
 
                read_extent_buffer(leaf, &cache->item,
@@ -3089,7 +3260,20 @@ int btrfs_read_block_groups(struct btrfs_root *root)
                cache->cached = 0;
                cache->pinned = 0;
                key.objectid = found_key.objectid + found_key.offset;
-               btrfs_release_path(root, path);
+               if (found_key.offset == 0)
+                       key.objectid++;
+               btrfs_release_path(path);
+
+               /*
+                * Skip 0 sized block group, don't insert them into block
+                * group cache tree, as its length is 0, it won't get
+                * freed at close_ctree() time.
+                */
+               if (found_key.offset == 0) {
+                       free(cache);
+                       continue;
+               }
+
                cache->flags = btrfs_block_group_flags(&cache->item);
                bit = 0;
                if (cache->flags & BTRFS_BLOCK_GROUP_DATA) {
@@ -3100,9 +3284,11 @@ int btrfs_read_block_groups(struct btrfs_root *root)
                        bit = BLOCK_GROUP_METADATA;
                }
                set_avail_alloc_bits(info, cache->flags);
-               if (btrfs_chunk_readonly(root, cache->key.objectid))
+               if (btrfs_chunk_readonly(info, cache->key.objectid))
                        cache->ro = 1;
 
+               account_super_bytes(info, cache);
+
                ret = update_space_info(info, cache->flags, found_key.offset,
                                        btrfs_block_group_used(&cache->item),
                                        &space_info);
@@ -3112,7 +3298,7 @@ int btrfs_read_block_groups(struct btrfs_root *root)
                /* use EXTENT_LOCKED to prevent merging */
                set_extent_bits(block_group_cache, found_key.objectid,
                                found_key.objectid + found_key.offset - 1,
-                               bit | EXTENT_LOCKED, GFP_NOFS);
+                               bit | EXTENT_LOCKED);
                set_state_private(block_group_cache, found_key.objectid,
                                  (unsigned long)cache);
        }
@@ -3122,50 +3308,67 @@ error:
        return ret;
 }
 
-int btrfs_make_block_group(struct btrfs_trans_handle *trans,
-                          struct btrfs_root *root, u64 bytes_used,
-                          u64 type, u64 chunk_objectid, u64 chunk_offset,
-                          u64 size)
+struct btrfs_block_group_cache *
+btrfs_add_block_group(struct btrfs_fs_info *fs_info, u64 bytes_used, u64 type,
+                     u64 chunk_offset, u64 size)
 {
        int ret;
        int bit = 0;
-       struct btrfs_root *extent_root;
        struct btrfs_block_group_cache *cache;
        struct extent_io_tree *block_group_cache;
 
-       extent_root = root->fs_info->extent_root;
-       block_group_cache = &root->fs_info->block_group_cache;
+       block_group_cache = &fs_info->block_group_cache;
 
        cache = kzalloc(sizeof(*cache), GFP_NOFS);
        BUG_ON(!cache);
        cache->key.objectid = chunk_offset;
        cache->key.offset = size;
 
-       btrfs_set_key_type(&cache->key, BTRFS_BLOCK_GROUP_ITEM_KEY);
+       cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
        btrfs_set_block_group_used(&cache->item, bytes_used);
-       btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
+       btrfs_set_block_group_chunk_objectid(&cache->item,
+                                            BTRFS_FIRST_CHUNK_TREE_OBJECTID);
        cache->flags = type;
        btrfs_set_block_group_flags(&cache->item, type);
 
-       ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
+       account_super_bytes(fs_info, cache);
+       ret = update_space_info(fs_info, cache->flags, size, bytes_used,
                                &cache->space_info);
        BUG_ON(ret);
 
        bit = block_group_state_bits(type);
-       set_extent_bits(block_group_cache, chunk_offset,
-                       chunk_offset + size - 1,
-                       bit | EXTENT_LOCKED, GFP_NOFS);
+       ret = set_extent_bits(block_group_cache, chunk_offset,
+                             chunk_offset + size - 1,
+                             bit | EXTENT_LOCKED);
+       BUG_ON(ret);
+
+       ret = set_state_private(block_group_cache, chunk_offset,
+                               (unsigned long)cache);
+       BUG_ON(ret);
+       set_avail_alloc_bits(fs_info, type);
+
+       return cache;
+}
 
-       set_state_private(block_group_cache, chunk_offset,
-                         (unsigned long)cache);
+int btrfs_make_block_group(struct btrfs_trans_handle *trans,
+                          struct btrfs_fs_info *fs_info, u64 bytes_used,
+                          u64 type, u64 chunk_offset, u64 size)
+{
+       int ret;
+       struct btrfs_root *extent_root = fs_info->extent_root;
+       struct btrfs_block_group_cache *cache;
+
+       cache = btrfs_add_block_group(fs_info, bytes_used, type, chunk_offset,
+                                     size);
        ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
                                sizeof(cache->item));
        BUG_ON(ret);
 
-       finish_current_insert(trans, extent_root);
+       ret = finish_current_insert(trans, extent_root);
+       BUG_ON(ret);
        ret = del_pending_extents(trans, extent_root);
        BUG_ON(ret);
-       set_avail_alloc_bits(extent_root->fs_info, type);
+
        return 0;
 }
 
@@ -3177,7 +3380,7 @@ int btrfs_make_block_group(struct btrfs_trans_handle *trans,
  * before doing any block allocation.
  */
 int btrfs_make_block_groups(struct btrfs_trans_handle *trans,
-                           struct btrfs_root *root)
+                           struct btrfs_fs_info *fs_info)
 {
        u64 total_bytes;
        u64 cur_start;
@@ -3189,15 +3392,14 @@ int btrfs_make_block_groups(struct btrfs_trans_handle *trans,
        u64 chunk_objectid;
        int ret;
        int bit;
-       struct btrfs_root *extent_root;
+       struct btrfs_root *extent_root = fs_info->extent_root;
        struct btrfs_block_group_cache *cache;
        struct extent_io_tree *block_group_cache;
 
-       extent_root = root->fs_info->extent_root;
-       block_group_cache = &root->fs_info->block_group_cache;
+       block_group_cache = &fs_info->block_group_cache;
        chunk_objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
-       total_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
-       group_align = 64 * root->sectorsize;
+       total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
+       group_align = 64 * fs_info->sectorsize;
 
        cur_start = 0;
        while (cur_start < total_bytes) {
@@ -3208,19 +3410,18 @@ int btrfs_make_block_groups(struct btrfs_trans_handle *trans,
                        group_type = BTRFS_BLOCK_GROUP_SYSTEM;
                        group_size /= 4;
                        group_size &= ~(group_align - 1);
-                       group_size = max_t(u64, group_size, 8 * 1024 * 1024);
-                       group_size = min_t(u64, group_size, 32 * 1024 * 1024);
+                       group_size = max_t(u64, group_size, SZ_8M);
+                       group_size = min_t(u64, group_size, SZ_32M);
                } else {
                        group_size &= ~(group_align - 1);
                        if (total_data >= total_metadata * 2) {
                                group_type = BTRFS_BLOCK_GROUP_METADATA;
-                               group_size = min_t(u64, group_size,
-                                                  1ULL * 1024 * 1024 * 1024);
+                               group_size = min_t(u64, group_size, SZ_1G);
                                total_metadata += group_size;
                        } else {
                                group_type = BTRFS_BLOCK_GROUP_DATA;
                                group_size = min_t(u64, group_size,
-                                                  5ULL * 1024 * 1024 * 1024);
+                                                  5ULL * SZ_1G);
                                total_data += group_size;
                        }
                        if ((total_bytes - cur_start) * 4 < group_size * 5)
@@ -3232,7 +3433,7 @@ int btrfs_make_block_groups(struct btrfs_trans_handle *trans,
 
                cache->key.objectid = cur_start;
                cache->key.offset = group_size;
-               btrfs_set_key_type(&cache->key, BTRFS_BLOCK_GROUP_ITEM_KEY);
+               cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
 
                btrfs_set_block_group_used(&cache->item, 0);
                btrfs_set_block_group_chunk_objectid(&cache->item,
@@ -3241,14 +3442,14 @@ int btrfs_make_block_groups(struct btrfs_trans_handle *trans,
 
                cache->flags = group_type;
 
-               ret = update_space_info(root->fs_info, group_type, group_size,
+               ret = update_space_info(fs_info, group_type, group_size,
                                        0, &cache->space_info);
                BUG_ON(ret);
-               set_avail_alloc_bits(extent_root->fs_info, group_type);
+               set_avail_alloc_bits(fs_info, group_type);
 
                set_extent_bits(block_group_cache, cur_start,
                                cur_start + group_size - 1,
-                               bit | EXTENT_LOCKED, GFP_NOFS);
+                               bit | EXTENT_LOCKED);
                set_state_private(block_group_cache, cur_start,
                                  (unsigned long)cache);
                cur_start += group_size;
@@ -3256,7 +3457,7 @@ int btrfs_make_block_groups(struct btrfs_trans_handle *trans,
        /* then insert all the items */
        cur_start = 0;
        while(cur_start < total_bytes) {
-               cache = btrfs_lookup_block_group(root->fs_info, cur_start);
+               cache = btrfs_lookup_block_group(fs_info, cur_start);
                BUG_ON(!cache);
 
                ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
@@ -3272,93 +3473,363 @@ int btrfs_make_block_groups(struct btrfs_trans_handle *trans,
        return 0;
 }
 
-int btrfs_update_block_group(struct btrfs_trans_handle *trans,
-                            struct btrfs_root *root,
+int btrfs_update_block_group(struct btrfs_root *root,
                             u64 bytenr, u64 num_bytes, int alloc,
                             int mark_free)
 {
-       return update_block_group(trans, root, bytenr, num_bytes,
+       return update_block_group(root, bytenr, num_bytes,
                                  alloc, mark_free);
 }
 
-static int btrfs_count_extents_in_block_group(struct btrfs_root *root,
-                                             struct btrfs_path *path, u64 start,
-                                             u64 len,
-                                             u64 *total)
+/*
+ * Just remove a block group item in extent tree
+ * Caller should ensure the block group is empty and all space is pinned.
+ * Or new tree block/data may be allocated into it.
+ */
+static int free_block_group_item(struct btrfs_trans_handle *trans,
+                                struct btrfs_fs_info *fs_info,
+                                u64 bytenr, u64 len)
 {
+       struct btrfs_path *path;
+       struct btrfs_key key;
+       struct btrfs_root *root = fs_info->extent_root;
+       int ret = 0;
+
+       key.objectid = bytenr;
+       key.offset = len;
+       key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
+
+       path = btrfs_alloc_path();
+       if (!path)
+               return -ENOMEM;
+
+       ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
+       if (ret > 0) {
+               ret = -ENOENT;
+               goto out;
+       }
+       if (ret < 0)
+               goto out;
+
+       ret = btrfs_del_item(trans, root, path);
+out:
+       btrfs_free_path(path);
+       return ret;
+}
+
+static int free_dev_extent_item(struct btrfs_trans_handle *trans,
+                               struct btrfs_fs_info *fs_info,
+                               u64 devid, u64 dev_offset)
+{
+       struct btrfs_root *root = fs_info->dev_root;
+       struct btrfs_path *path;
        struct btrfs_key key;
-       struct extent_buffer *leaf;
-       u64 bytes_used = 0;
        int ret;
-       int slot;
 
-       key.offset = 0;
-       key.objectid = start;
-       btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
-       ret = btrfs_search_slot(NULL, root->fs_info->extent_root,
-                               &key, path, 0, 0);
+       path = btrfs_alloc_path();
+       if (!path)
+               return -ENOMEM;
+
+       key.objectid = devid;
+       key.type = BTRFS_DEV_EXTENT_KEY;
+       key.offset = dev_offset;
+
+       ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
        if (ret < 0)
-               return ret;
-       while(1) {
-               leaf = path->nodes[0];
-               slot = path->slots[0];
-               if (slot >= btrfs_header_nritems(leaf)) {
-                       ret = btrfs_next_leaf(root, path);
-                       if (ret < 0)
-                               return ret;
-                       if (ret > 0)
-                               break;
-                       leaf = path->nodes[0];
-                       slot = path->slots[0];
+               goto out;
+       if (ret > 0) {
+               ret = -ENOENT;
+               goto out;
+       }
+
+       ret = btrfs_del_item(trans, root, path);
+out:
+       btrfs_free_path(path);
+       return ret;
+}
+
+static int free_chunk_dev_extent_items(struct btrfs_trans_handle *trans,
+                                      struct btrfs_fs_info *fs_info,
+                                      u64 chunk_offset)
+{
+       struct btrfs_chunk *chunk = NULL;
+       struct btrfs_root *root= fs_info->chunk_root;
+       struct btrfs_path *path;
+       struct btrfs_key key;
+       u16 num_stripes;
+       int i;
+       int ret;
+
+       path = btrfs_alloc_path();
+       if (!path)
+               return -ENOMEM;
+
+       key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
+       key.type = BTRFS_CHUNK_ITEM_KEY;
+       key.offset = chunk_offset;
+
+       ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
+       if (ret < 0)
+               goto out;
+       if (ret > 0) {
+               ret = -ENOENT;
+               goto out;
+       }
+       chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
+                              struct btrfs_chunk);
+       num_stripes = btrfs_chunk_num_stripes(path->nodes[0], chunk);
+       for (i = 0; i < num_stripes; i++) {
+               ret = free_dev_extent_item(trans, fs_info,
+                       btrfs_stripe_devid_nr(path->nodes[0], chunk, i),
+                       btrfs_stripe_offset_nr(path->nodes[0], chunk, i));
+               if (ret < 0)
+                       goto out;
+       }
+out:
+       btrfs_free_path(path);
+       return ret;
+}
+
+static int free_system_chunk_item(struct btrfs_super_block *super,
+                                 struct btrfs_key *key)
+{
+       struct btrfs_disk_key *disk_key;
+       struct btrfs_key cpu_key;
+       u32 array_size = btrfs_super_sys_array_size(super);
+       char *ptr = (char *)super->sys_chunk_array;
+       int cur = 0;
+       int ret = -ENOENT;
+
+       while (cur < btrfs_super_sys_array_size(super)) {
+               struct btrfs_chunk *chunk;
+               u32 num_stripes;
+               u32 chunk_len;
+
+               disk_key = (struct btrfs_disk_key *)(ptr + cur);
+               btrfs_disk_key_to_cpu(&cpu_key, disk_key);
+               if (cpu_key.type != BTRFS_CHUNK_ITEM_KEY) {
+                       /* just in case */
+                       ret = -EIO;
+                       goto out;
                }
-               btrfs_item_key_to_cpu(leaf, &key, slot);
-               if (key.objectid > start + len)
-                       break;
-               if (key.type == BTRFS_EXTENT_ITEM_KEY)
-                       bytes_used += key.offset;
-               path->slots[0]++;
+
+               chunk = (struct btrfs_chunk *)(ptr + cur + sizeof(*disk_key));
+               num_stripes = btrfs_stack_chunk_num_stripes(chunk);
+               chunk_len = btrfs_chunk_item_size(num_stripes) +
+                           sizeof(*disk_key);
+
+               if (key->objectid == cpu_key.objectid &&
+                   key->offset == cpu_key.offset &&
+                   key->type == cpu_key.type) {
+                       memmove(ptr + cur, ptr + cur + chunk_len,
+                               array_size - cur - chunk_len);
+                       array_size -= chunk_len;
+                       btrfs_set_super_sys_array_size(super, array_size);
+                       ret = 0;
+                       goto out;
+               }
+
+               cur += chunk_len;
        }
-       *total = bytes_used;
-       btrfs_release_path(root, path);
-       return 0;
+out:
+       return ret;
 }
 
-int btrfs_check_block_accounting(struct btrfs_root *root)
+static int free_chunk_item(struct btrfs_trans_handle *trans,
+                          struct btrfs_fs_info *fs_info,
+                          u64 bytenr)
 {
+       struct btrfs_path *path;
+       struct btrfs_key key;
+       struct btrfs_root *root = fs_info->chunk_root;
+       struct btrfs_chunk *chunk;
+       u64 chunk_type;
        int ret;
-       u64 start = 0;
-       u64 bytes_used = 0;
-       struct btrfs_path path;
+
+       key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
+       key.offset = bytenr;
+       key.type = BTRFS_CHUNK_ITEM_KEY;
+
+       path = btrfs_alloc_path();
+       if (!path)
+               return -ENOMEM;
+
+       ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
+       if (ret > 0) {
+               ret = -ENOENT;
+               goto out;
+       }
+       if (ret < 0)
+               goto out;
+       chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
+                              struct btrfs_chunk);
+       chunk_type = btrfs_chunk_type(path->nodes[0], chunk);
+
+       ret = btrfs_del_item(trans, root, path);
+       if (ret < 0)
+               goto out;
+
+       if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
+               ret = free_system_chunk_item(fs_info->super_copy, &key);
+out:
+       btrfs_free_path(path);
+       return ret;
+}
+
+static u64 get_dev_extent_len(struct map_lookup *map)
+{
+       int div;
+
+       switch (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
+       case 0: /* Single */
+       case BTRFS_BLOCK_GROUP_DUP:
+       case BTRFS_BLOCK_GROUP_RAID1:
+               div = 1;
+               break;
+       case BTRFS_BLOCK_GROUP_RAID5:
+               div = (map->num_stripes - 1);
+               break;
+       case BTRFS_BLOCK_GROUP_RAID6:
+               div = (map->num_stripes - 2);
+               break;
+       case BTRFS_BLOCK_GROUP_RAID10:
+               div = (map->num_stripes / map->sub_stripes);
+               break;
+       default:
+               /* normally, read chunk security hook should handled it */
+               BUG_ON(1);
+       }
+       return map->ce.size / div;
+}
+
+/* free block group/chunk related caches */
+static int free_block_group_cache(struct btrfs_trans_handle *trans,
+                                 struct btrfs_fs_info *fs_info,
+                                 u64 bytenr, u64 len)
+{
        struct btrfs_block_group_cache *cache;
-       struct btrfs_fs_info *fs_info = root->fs_info;
+       struct cache_extent *ce;
+       struct map_lookup *map;
+       int ret;
+       int i;
+       u64 flags;
 
-       btrfs_init_path(&path);
+       /* Free block group cache first */
+       cache = btrfs_lookup_block_group(fs_info, bytenr);
+       if (!cache)
+               return -ENOENT;
+       flags = cache->flags;
+       if (cache->free_space_ctl) {
+               btrfs_remove_free_space_cache(cache);
+               kfree(cache->free_space_ctl);
+       }
+       clear_extent_bits(&fs_info->block_group_cache, bytenr, bytenr + len - 1,
+                         (unsigned int)-1);
+       ret = free_space_info(fs_info, flags, len, 0, NULL);
+       if (ret < 0)
+               goto out;
+       kfree(cache);
 
-       while(1) {
-               cache = btrfs_lookup_block_group(fs_info, start);
-               if (!cache)
-                       break;
+       /* Then free mapping info and dev usage info */
+       ce = search_cache_extent(&fs_info->mapping_tree.cache_tree, bytenr);
+       if (!ce || ce->start != bytenr) {
+               ret = -ENOENT;
+               goto out;
+       }
+       map = container_of(ce, struct map_lookup, ce);
+       for (i = 0; i < map->num_stripes; i++) {
+               struct btrfs_device *device;
 
-               ret = btrfs_count_extents_in_block_group(root, &path,
-                                                        cache->key.objectid,
-                                                        cache->key.offset,
-                                                        &bytes_used);
+               device = map->stripes[i].dev;
+               device->bytes_used -= get_dev_extent_len(map);
+               ret = btrfs_update_device(trans, device);
+               if (ret < 0)
+                       goto out;
+       }
+       remove_cache_extent(&fs_info->mapping_tree.cache_tree, ce);
+       free(map);
+out:
+       return ret;
+}
 
-               if (ret == 0) {
-                       u64 on_disk = btrfs_block_group_used(&cache->item);
-                       if (on_disk != bytes_used) {
-                               fprintf(stderr, "bad block group accounting found %llu "
-                                       "expected %llu block group %llu\n",
-                                       (unsigned long long)bytes_used,
-                                       (unsigned long long)on_disk,
-                                       (unsigned long long)cache->key.objectid);
-                       }
-               }
-               start = cache->key.objectid + cache->key.offset;
+int btrfs_free_block_group(struct btrfs_trans_handle *trans,
+                          struct btrfs_fs_info *fs_info, u64 bytenr, u64 len)
+{
+       struct btrfs_root *extent_root = fs_info->extent_root;
+       struct btrfs_path *path;
+       struct btrfs_block_group_item *bgi;
+       struct btrfs_key key;
+       int ret = 0;
 
-               cache->space_info->bytes_used = 0;
+       path = btrfs_alloc_path();
+       if (!path)
+               return -ENOMEM;
+
+       key.objectid = bytenr;
+       key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
+       key.offset = len;
+
+       /* Double check the block group to ensure it's empty */
+       ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
+       if (ret > 0) {
+               ret = -ENONET;
+               goto out;
        }
-       return 0;
+       if (ret < 0)
+               goto out;
+
+       bgi = btrfs_item_ptr(path->nodes[0], path->slots[0],
+                            struct btrfs_block_group_item);
+       if (btrfs_disk_block_group_used(path->nodes[0], bgi)) {
+               fprintf(stderr,
+                       "WARNING: block group [%llu,%llu) is not empty\n",
+                       bytenr, bytenr + len);
+               ret = -EINVAL;
+               goto out;
+       }
+       btrfs_release_path(path);
+
+       /*
+        * Now pin all space in the block group, to prevent further transaction
+        * allocate space from it.
+        * Every operation needs a transaction must be in the range.
+        */
+       btrfs_pin_extent(fs_info, bytenr, len);
+
+       /* delete block group item and chunk item */
+       ret = free_block_group_item(trans, fs_info, bytenr, len);
+       if (ret < 0) {
+               fprintf(stderr,
+                       "failed to free block group item for [%llu,%llu)\n",
+                       bytenr, bytenr + len);
+               btrfs_unpin_extent(fs_info, bytenr, len);
+               goto out;
+       }
+
+       ret = free_chunk_dev_extent_items(trans, fs_info, bytenr);
+       if (ret < 0) {
+               fprintf(stderr,
+                       "failed to dev extents belongs to [%llu,%llu)\n",
+                       bytenr, bytenr + len);
+               btrfs_unpin_extent(fs_info, bytenr, len);
+               goto out;
+       }
+       ret = free_chunk_item(trans, fs_info, bytenr);
+       if (ret < 0) {
+               fprintf(stderr,
+                       "failed to free chunk for [%llu,%llu)\n",
+                       bytenr, bytenr + len);
+               btrfs_unpin_extent(fs_info, bytenr, len);
+               goto out;
+       }
+
+       /* Now release the block_group_cache */
+       ret = free_block_group_cache(trans, fs_info, bytenr, len);
+       btrfs_unpin_extent(fs_info, bytenr, len);
+
+out:
+       btrfs_free_path(path);
+       return ret;
 }
 
 /*
@@ -3368,7 +3839,7 @@ int btrfs_check_block_accounting(struct btrfs_root *root)
 int btrfs_fix_block_accounting(struct btrfs_trans_handle *trans,
                               struct btrfs_root *root)
 {
-       int ret;
+       int ret = 0;
        int slot;
        u64 start = 0;
        u64 bytes_used = 0;
@@ -3378,19 +3849,34 @@ int btrfs_fix_block_accounting(struct btrfs_trans_handle *trans,
        struct btrfs_block_group_cache *cache;
        struct btrfs_fs_info *fs_info = root->fs_info;
 
+       root = root->fs_info->extent_root;
+
+       while(extent_root_pending_ops(fs_info)) {
+               ret = finish_current_insert(trans, root);
+               if (ret)
+                       return ret;
+               ret = del_pending_extents(trans, root);
+               if (ret)
+                       return ret;
+       }
+
        while(1) {
-               cache = btrfs_lookup_block_group(fs_info, start);
+               cache = btrfs_lookup_first_block_group(fs_info, start);
                if (!cache)
                        break;
                start = cache->key.objectid + cache->key.offset;
                btrfs_set_block_group_used(&cache->item, 0);
                cache->space_info->bytes_used = 0;
+               set_extent_bits(&root->fs_info->block_group_cache,
+                               cache->key.objectid,
+                               cache->key.objectid + cache->key.offset -1,
+                               BLOCK_GROUP_DIRTY);
        }
 
        btrfs_init_path(&path);
        key.offset = 0;
        key.objectid = 0;
-       btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
+       key.type = BTRFS_EXTENT_ITEM_KEY;
        ret = btrfs_search_slot(trans, root->fs_info->extent_root,
                                &key, &path, 0, 0);
        if (ret < 0)
@@ -3410,13 +3896,361 @@ int btrfs_fix_block_accounting(struct btrfs_trans_handle *trans,
                btrfs_item_key_to_cpu(leaf, &key, slot);
                if (key.type == BTRFS_EXTENT_ITEM_KEY) {
                        bytes_used += key.offset;
-                       ret = btrfs_update_block_group(trans, root,
+                       ret = btrfs_update_block_group(root,
                                  key.objectid, key.offset, 1, 0);
                        BUG_ON(ret);
+               } else if (key.type == BTRFS_METADATA_ITEM_KEY) {
+                       bytes_used += fs_info->nodesize;
+                       ret = btrfs_update_block_group(root,
+                                 key.objectid, fs_info->nodesize, 1, 0);
+                       if (ret)
+                               goto out;
                }
                path.slots[0]++;
        }
-       btrfs_set_super_bytes_used(&root->fs_info->super_copy, bytes_used);
-       btrfs_release_path(root, &path);
+       btrfs_set_super_bytes_used(root->fs_info->super_copy, bytes_used);
+       ret = 0;
+out:
+       btrfs_release_path(&path);
+       return ret;
+}
+
+static void __get_extent_size(struct btrfs_root *root, struct btrfs_path *path,
+                             u64 *start, u64 *len)
+{
+       struct btrfs_key key;
+
+       btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
+       BUG_ON(!(key.type == BTRFS_EXTENT_ITEM_KEY ||
+                key.type == BTRFS_METADATA_ITEM_KEY));
+       *start = key.objectid;
+       if (key.type == BTRFS_EXTENT_ITEM_KEY)
+               *len = key.offset;
+       else
+               *len = root->fs_info->nodesize;
+}
+
+/*
+ * Find first overlap extent for range [bytenr, bytenr + len)
+ * Return 0 for found and point path to it.
+ * Return >0 for not found.
+ * Return <0 for err
+ */
+int btrfs_search_overlap_extent(struct btrfs_root *root,
+                               struct btrfs_path *path, u64 bytenr, u64 len)
+{
+       struct btrfs_key key;
+       u64 cur_start;
+       u64 cur_len;
+       int ret;
+
+       key.objectid = bytenr;
+       key.type = BTRFS_EXTENT_DATA_KEY;
+       key.offset = (u64)-1;
+
+       ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
+       if (ret < 0)
+               return ret;
+       BUG_ON(ret == 0);
+
+       ret = btrfs_previous_extent_item(root, path, 0);
+       if (ret < 0)
+               return ret;
+       /* no previous, check next extent */
+       if (ret > 0)
+               goto next;
+       __get_extent_size(root, path, &cur_start, &cur_len);
+       /* Tail overlap */
+       if (cur_start + cur_len > bytenr)
+               return 1;
+
+next:
+       ret = btrfs_next_extent_item(root, path, bytenr + len);
+       if (ret < 0)
+               return ret;
+       /* No next, prev already checked, no overlap */
+       if (ret > 0)
+               return 0;
+       __get_extent_size(root, path, &cur_start, &cur_len);
+       /* head overlap*/
+       if (cur_start < bytenr + len)
+               return 1;
+       return 0;
+}
+
+static int __btrfs_record_file_extent(struct btrfs_trans_handle *trans,
+                                     struct btrfs_root *root, u64 objectid,
+                                     struct btrfs_inode_item *inode,
+                                     u64 file_pos, u64 disk_bytenr,
+                                     u64 *ret_num_bytes)
+{
+       int ret;
+       struct btrfs_fs_info *info = root->fs_info;
+       struct btrfs_root *extent_root = info->extent_root;
+       struct extent_buffer *leaf;
+       struct btrfs_file_extent_item *fi;
+       struct btrfs_key ins_key;
+       struct btrfs_path *path;
+       struct btrfs_extent_item *ei;
+       u64 nbytes;
+       u64 extent_num_bytes;
+       u64 extent_bytenr;
+       u64 extent_offset;
+       u64 num_bytes = *ret_num_bytes;
+
+       /*
+        * All supported file system should not use its 0 extent.
+        * As it's for hole
+        *
+        * And hole extent has no size limit, no need to loop.
+        */
+       if (disk_bytenr == 0) {
+               ret = btrfs_insert_file_extent(trans, root, objectid,
+                                               file_pos, disk_bytenr,
+                                               num_bytes, num_bytes);
+               return ret;
+       }
+       num_bytes = min_t(u64, num_bytes, BTRFS_MAX_EXTENT_SIZE);
+
+       path = btrfs_alloc_path();
+       if (!path)
+               return -ENOMEM;
+
+       /* First to check extent overlap */
+       ret = btrfs_search_overlap_extent(extent_root, path, disk_bytenr,
+                                         num_bytes);
+       if (ret < 0)
+               goto fail;
+       if (ret > 0) {
+               /* Found overlap */
+               u64 cur_start;
+               u64 cur_len;
+
+               __get_extent_size(extent_root, path, &cur_start, &cur_len);
+               /*
+                * For convert case, this extent should be a subset of
+                * existing one.
+                */
+               BUG_ON(disk_bytenr < cur_start);
+
+               extent_bytenr = cur_start;
+               extent_num_bytes = cur_len;
+               extent_offset = disk_bytenr - extent_bytenr;
+       } else {
+               /* No overlap, create new extent */
+               btrfs_release_path(path);
+               ins_key.objectid = disk_bytenr;
+               ins_key.offset = num_bytes;
+               ins_key.type = BTRFS_EXTENT_ITEM_KEY;
+
+               ret = btrfs_insert_empty_item(trans, extent_root, path,
+                                             &ins_key, sizeof(*ei));
+               if (ret == 0) {
+                       leaf = path->nodes[0];
+                       ei = btrfs_item_ptr(leaf, path->slots[0],
+                                           struct btrfs_extent_item);
+
+                       btrfs_set_extent_refs(leaf, ei, 0);
+                       btrfs_set_extent_generation(leaf, ei, 0);
+                       btrfs_set_extent_flags(leaf, ei,
+                                              BTRFS_EXTENT_FLAG_DATA);
+                       btrfs_mark_buffer_dirty(leaf);
+
+                       ret = btrfs_update_block_group(root, disk_bytenr,
+                                                      num_bytes, 1, 0);
+                       if (ret)
+                               goto fail;
+               } else if (ret != -EEXIST) {
+                       goto fail;
+               }
+               btrfs_extent_post_op(trans, extent_root);
+               extent_bytenr = disk_bytenr;
+               extent_num_bytes = num_bytes;
+               extent_offset = 0;
+       }
+       btrfs_release_path(path);
+       ins_key.objectid = objectid;
+       ins_key.offset = file_pos;
+       ins_key.type = BTRFS_EXTENT_DATA_KEY;
+       ret = btrfs_insert_empty_item(trans, root, path, &ins_key,
+                                     sizeof(*fi));
+       if (ret)
+               goto fail;
+       leaf = path->nodes[0];
+       fi = btrfs_item_ptr(leaf, path->slots[0],
+                           struct btrfs_file_extent_item);
+       btrfs_set_file_extent_generation(leaf, fi, trans->transid);
+       btrfs_set_file_extent_type(leaf, fi, BTRFS_FILE_EXTENT_REG);
+       btrfs_set_file_extent_disk_bytenr(leaf, fi, extent_bytenr);
+       btrfs_set_file_extent_disk_num_bytes(leaf, fi, extent_num_bytes);
+       btrfs_set_file_extent_offset(leaf, fi, extent_offset);
+       btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
+       btrfs_set_file_extent_ram_bytes(leaf, fi, extent_num_bytes);
+       btrfs_set_file_extent_compression(leaf, fi, 0);
+       btrfs_set_file_extent_encryption(leaf, fi, 0);
+       btrfs_set_file_extent_other_encoding(leaf, fi, 0);
+       btrfs_mark_buffer_dirty(leaf);
+
+       nbytes = btrfs_stack_inode_nbytes(inode) + num_bytes;
+       btrfs_set_stack_inode_nbytes(inode, nbytes);
+       btrfs_release_path(path);
+
+       ret = btrfs_inc_extent_ref(trans, root, extent_bytenr, extent_num_bytes,
+                                  0, root->root_key.objectid, objectid,
+                                  file_pos - extent_offset);
+       if (ret)
+               goto fail;
+       ret = 0;
+       *ret_num_bytes = min(extent_num_bytes - extent_offset, num_bytes);
+fail:
+       btrfs_free_path(path);
+       return ret;
+}
+
+/*
+ * Record a file extent. Do all the required works, such as inserting
+ * file extent item, inserting extent item and backref item into extent
+ * tree and updating block accounting.
+ */
+int btrfs_record_file_extent(struct btrfs_trans_handle *trans,
+                             struct btrfs_root *root, u64 objectid,
+                             struct btrfs_inode_item *inode,
+                             u64 file_pos, u64 disk_bytenr,
+                             u64 num_bytes)
+{
+       u64 cur_disk_bytenr = disk_bytenr;
+       u64 cur_file_pos = file_pos;
+       u64 cur_num_bytes = num_bytes;
+       int ret = 0;
+
+       while (num_bytes > 0) {
+               ret = __btrfs_record_file_extent(trans, root, objectid,
+                                                inode, cur_file_pos,
+                                                cur_disk_bytenr,
+                                                &cur_num_bytes);
+               if (ret < 0)
+                       break;
+               cur_disk_bytenr += cur_num_bytes;
+               cur_file_pos += cur_num_bytes;
+               num_bytes -= cur_num_bytes;
+       }
+       return ret;
+}
+
+
+static int add_excluded_extent(struct btrfs_root *root,
+                              u64 start, u64 num_bytes)
+{
+       u64 end = start + num_bytes - 1;
+       set_extent_bits(&root->fs_info->pinned_extents,
+                       start, end, EXTENT_UPTODATE);
+       return 0;
+}
+
+void free_excluded_extents(struct btrfs_root *root,
+                          struct btrfs_block_group_cache *cache)
+{
+       u64 start, end;
+
+       start = cache->key.objectid;
+       end = start + cache->key.offset - 1;
+
+       clear_extent_bits(&root->fs_info->pinned_extents,
+                         start, end, EXTENT_UPTODATE);
+}
+
+int exclude_super_stripes(struct btrfs_root *root,
+                         struct btrfs_block_group_cache *cache)
+{
+       u64 bytenr;
+       u64 *logical;
+       int stripe_len;
+       int i, nr, ret;
+
+       if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
+               stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
+               cache->bytes_super += stripe_len;
+               ret = add_excluded_extent(root, cache->key.objectid,
+                                         stripe_len);
+               if (ret)
+                       return ret;
+       }
+
+       for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
+               bytenr = btrfs_sb_offset(i);
+               ret = btrfs_rmap_block(root->fs_info,
+                                      cache->key.objectid, bytenr,
+                                      0, &logical, &nr, &stripe_len);
+               if (ret)
+                       return ret;
+
+               while (nr--) {
+                       u64 start, len;
+
+                       if (logical[nr] > cache->key.objectid +
+                           cache->key.offset)
+                               continue;
+
+                       if (logical[nr] + stripe_len <= cache->key.objectid)
+                               continue;
+
+                       start = logical[nr];
+                       if (start < cache->key.objectid) {
+                               start = cache->key.objectid;
+                               len = (logical[nr] + stripe_len) - start;
+                       } else {
+                               len = min_t(u64, stripe_len,
+                                           cache->key.objectid +
+                                           cache->key.offset - start);
+                       }
+
+                       cache->bytes_super += len;
+                       ret = add_excluded_extent(root, start, len);
+                       if (ret) {
+                               kfree(logical);
+                               return ret;
+                       }
+               }
+
+               kfree(logical);
+       }
        return 0;
 }
+
+u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
+                      struct btrfs_fs_info *info, u64 start, u64 end)
+{
+       u64 extent_start, extent_end, size, total_added = 0;
+       int ret;
+
+       while (start < end) {
+               ret = find_first_extent_bit(&info->pinned_extents, start,
+                                           &extent_start, &extent_end,
+                                           EXTENT_DIRTY | EXTENT_UPTODATE);
+               if (ret)
+                       break;
+
+               if (extent_start <= start) {
+                       start = extent_end + 1;
+               } else if (extent_start > start && extent_start < end) {
+                       size = extent_start - start;
+                       total_added += size;
+                       ret = btrfs_add_free_space(block_group->free_space_ctl,
+                                                  start, size);
+                       BUG_ON(ret); /* -ENOMEM or logic error */
+                       start = extent_end + 1;
+               } else {
+                       break;
+               }
+       }
+
+       if (start < end) {
+               size = end - start;
+               total_added += size;
+               ret = btrfs_add_free_space(block_group->free_space_ctl, start,
+                                          size);
+               BUG_ON(ret); /* -ENOMEM or logic error */
+       }
+
+       return total_added;
+}