Btrfs: free space accounting redo
authorJosef Bacik <jbacik@redhat.com>
Tue, 23 Sep 2008 17:14:11 +0000 (13:14 -0400)
committerChris Mason <chris.mason@oracle.com>
Thu, 25 Sep 2008 15:04:07 +0000 (11:04 -0400)
1) replace the per fs_info extent_io_tree that tracked free space with two
rb-trees per block group to track free space areas via offset and size.  The
reason to do this is because most allocations come with a hint byte where to
start, so we can usually find a chunk of free space at that hint byte to satisfy
the allocation and get good space packing.  If we cannot find free space at or
after the given offset we fall back on looking for a chunk of the given size as
close to that given offset as possible.  When we fall back on the size search we
also try to find a slot as close to the size we want as possible, to avoid
breaking small chunks off of huge areas if possible.

2) remove the extent_io_tree that tracked the block group cache from fs_info and
replaced it with an rb-tree thats tracks block group cache via offset.  also
added a per space_info list that tracks the block group cache for the particular
space so we can lookup related block groups easily.

3) cleaned up the allocation code to make it a little easier to read and a
little less complicated.  Basically there are 3 steps, first look from our
provided hint.  If we couldn't find from that given hint, start back at our
original search start and look for space from there.  If that fails try to
allocate space if we can and start looking again.  If not we're screwed and need
to start over again.

4) small fixes.  there were some issues in volumes.c where we wouldn't allocate
the rest of the disk.  fixed cow_file_range to actually pass the alloc_hint,
which has helped a good bit in making the fs_mark test I run have semi-normal
results as we run out of space.  Generally with data allocations we don't track
where we last allocated from, so everytime we did a data allocation we'd search
through every block group that we have looking for free space.  Now searching a
block group with no free space isn't terribly time consuming, it was causing a
slight degradation as we got more data block groups.  The alloc_hint has fixed
this slight degredation and made things semi-normal.

There is still one nagging problem I'm working on where we will get ENOSPC when
there is definitely plenty of space.  This only happens with metadata
allocations, and only when we are almost full.  So you generally hit the 85%
mark first, but sometimes you'll hit the BUG before you hit the 85% wall.  I'm
still tracking it down, but until then this seems to be pretty stable and make a
significant performance gain.

Signed-off-by: Chris Mason <chris.mason@oracle.com>
fs/btrfs/Makefile
fs/btrfs/ctree.c
fs/btrfs/ctree.h
fs/btrfs/disk-io.c
fs/btrfs/extent-tree.c
fs/btrfs/extent_io.c
fs/btrfs/free-space-cache.c [new file with mode: 0644]
fs/btrfs/inode.c
fs/btrfs/volumes.c

index b7addbfd8c22ca0c4df76a2a4e6a7703b4cd2270..eb36ae981bdcdb0eafc0a67d5ca5e8caf9377046 100644 (file)
@@ -7,7 +7,7 @@ btrfs-y := super.o ctree.o extent-tree.o print-tree.o root-tree.o dir-item.o \
           transaction.o bit-radix.o inode.o file.o tree-defrag.o \
           extent_map.o sysfs.o struct-funcs.o xattr.o ordered-data.o \
           extent_io.o volumes.o async-thread.o ioctl.o locking.o orphan.o \
-          ref-cache.o export.o tree-log.o acl.o
+          ref-cache.o export.o tree-log.o acl.o free-space-cache.o
 else
 
 # Normal Makefile
index 18e84472abb524fad55d9d7e37dcf4ba9b5c52ef..6f467901246ff3154ff69a01d932905b80021aa7 100644 (file)
@@ -2725,9 +2725,8 @@ int btrfs_insert_empty_items(struct btrfs_trans_handle *trans,
 
        total_size = total_data + (nr * sizeof(struct btrfs_item));
        ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1);
-       if (ret == 0) {
+       if (ret == 0)
                return -EEXIST;
-       }
        if (ret < 0)
                goto out;
 
index eb65fd808883decdede47edf24416bf2eb7fa917..730aae3bc18156951e1a7178eb7a38684c6d0b3f 100644 (file)
@@ -483,7 +483,6 @@ struct btrfs_csum_item {
 #define BTRFS_BLOCK_GROUP_DUP     (1 << 5)
 #define BTRFS_BLOCK_GROUP_RAID10   (1 << 6)
 
-
 struct btrfs_block_group_item {
        __le64 used;
        __le64 chunk_objectid;
@@ -498,17 +497,40 @@ struct btrfs_space_info {
        int full;
        int force_alloc;
        struct list_head list;
+
+       /* for block groups in our same type */
+       struct list_head block_groups;
+       spinlock_t lock;
+};
+
+struct btrfs_free_space {
+       struct rb_node bytes_index;
+       struct rb_node offset_index;
+       u64 offset;
+       u64 bytes;
 };
 
 struct btrfs_block_group_cache {
        struct btrfs_key key;
        struct btrfs_block_group_item item;
-       struct btrfs_space_info *space_info;
        spinlock_t lock;
        u64 pinned;
        u64 flags;
        int cached;
        int ro;
+       int dirty;
+
+       struct btrfs_space_info *space_info;
+
+       /* free space cache stuff */
+       struct rb_root free_space_bytes;
+       struct rb_root free_space_offset;
+
+       /* block group cache stuff */
+       struct rb_node cache_node;
+
+       /* for block groups in the same raid type */
+       struct list_head list;
 };
 
 struct btrfs_device;
@@ -525,8 +547,10 @@ struct btrfs_fs_info {
        struct btrfs_root *log_root_tree;
        struct radix_tree_root fs_roots_radix;
 
-       struct extent_io_tree free_space_cache;
-       struct extent_io_tree block_group_cache;
+       /* block group cache stuff */
+       spinlock_t block_group_cache_lock;
+       struct rb_root block_group_cache_tree;
+
        struct extent_io_tree pinned_extents;
        struct extent_io_tree pending_del;
        struct extent_io_tree extent_ins;
@@ -1814,4 +1838,18 @@ int btrfs_sync_fs(struct super_block *sb, int wait);
 int btrfs_check_acl(struct inode *inode, int mask);
 int btrfs_init_acl(struct inode *inode, struct inode *dir);
 int btrfs_acl_chmod(struct inode *inode);
+
+/* free-space-cache.c */
+int btrfs_add_free_space(struct btrfs_block_group_cache *block_group,
+                        u64 bytenr, u64 size);
+int btrfs_remove_free_space(struct btrfs_block_group_cache *block_group,
+                           u64 bytenr, u64 size);
+void btrfs_remove_free_space_cache(struct btrfs_block_group_cache
+                                  *block_group);
+struct btrfs_free_space *btrfs_find_free_space(struct btrfs_block_group_cache
+                                              *block_group, u64 offset,
+                                              u64 bytes);
+void btrfs_dump_free_space(struct btrfs_block_group_cache *block_group,
+                          u64 bytes);
+u64 btrfs_block_group_free_space(struct btrfs_block_group_cache *block_group);
 #endif
index f6f7821d43a5248478c0f5ccfa1bcb97c6e22410..535bd0fe1a7138a830e5e965ed566cffd25b4167 100644 (file)
@@ -1410,10 +1410,9 @@ struct btrfs_root *open_ctree(struct super_block *sb,
 
        BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
 
-       extent_io_tree_init(&fs_info->free_space_cache,
-                            fs_info->btree_inode->i_mapping, GFP_NOFS);
-       extent_io_tree_init(&fs_info->block_group_cache,
-                            fs_info->btree_inode->i_mapping, GFP_NOFS);
+       spin_lock_init(&fs_info->block_group_cache_lock);
+       fs_info->block_group_cache_tree.rb_node = NULL;
+
        extent_io_tree_init(&fs_info->pinned_extents,
                             fs_info->btree_inode->i_mapping, GFP_NOFS);
        extent_io_tree_init(&fs_info->pending_del,
index 1c10ffc837c863825624e11ec6d9ec9791b6c700..813566acc5d3aef1e6aa671d6c4ec21310b50dfe 100644 (file)
 #include "locking.h"
 #include "ref-cache.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
-
 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
@@ -62,6 +56,127 @@ void maybe_unlock_mutex(struct btrfs_root *root)
        }
 }
 
+static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
+{
+       return (cache->flags & bits) == bits;
+}
+
+/*
+ * this adds the block group to the fs_info rb tree for the block group
+ * cache
+ */
+int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
+                               struct btrfs_block_group_cache *block_group)
+{
+       struct rb_node **p;
+       struct rb_node *parent = NULL;
+       struct btrfs_block_group_cache *cache;
+
+       spin_lock(&info->block_group_cache_lock);
+       p = &info->block_group_cache_tree.rb_node;
+
+       while (*p) {
+               parent = *p;
+               cache = rb_entry(parent, struct btrfs_block_group_cache,
+                                cache_node);
+               if (block_group->key.objectid < cache->key.objectid) {
+                       p = &(*p)->rb_left;
+               } else if (block_group->key.objectid > cache->key.objectid) {
+                       p = &(*p)->rb_right;
+               } else {
+                       spin_unlock(&info->block_group_cache_lock);
+                       return -EEXIST;
+               }
+       }
+
+       rb_link_node(&block_group->cache_node, parent, p);
+       rb_insert_color(&block_group->cache_node,
+                       &info->block_group_cache_tree);
+       spin_unlock(&info->block_group_cache_lock);
+
+       return 0;
+}
+
+/*
+ * This will return the block group at or after bytenr if contains is 0, else
+ * it will return the block group that contains the bytenr
+ */
+static struct btrfs_block_group_cache *
+block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
+                             int contains)
+{
+       struct btrfs_block_group_cache *cache, *ret = NULL;
+       struct rb_node *n;
+       u64 end, start;
+
+       spin_lock(&info->block_group_cache_lock);
+       n = info->block_group_cache_tree.rb_node;
+
+       while (n) {
+               cache = rb_entry(n, struct btrfs_block_group_cache,
+                                cache_node);
+               end = cache->key.objectid + cache->key.offset - 1;
+               start = cache->key.objectid;
+
+               if (bytenr < start) {
+                       if (!contains && (!ret || start < ret->key.objectid))
+                               ret = cache;
+                       n = n->rb_left;
+               } else if (bytenr > start) {
+                       if (contains && bytenr <= end) {
+                               ret = cache;
+                               break;
+                       }
+                       n = n->rb_right;
+               } else {
+                       ret = cache;
+                       break;
+               }
+       }
+       spin_unlock(&info->block_group_cache_lock);
+
+       return ret;
+}
+
+/*
+ * this is only called by cache_block_group, since we could have freed extents
+ * we need to check the pinned_extents for any extents that can't be used yet
+ * since their free space will be released as soon as the transaction commits.
+ */
+static int 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;
+       int ret;
+
+       while (start < end) {
+               ret = find_first_extent_bit(&info->pinned_extents, start,
+                                           &extent_start, &extent_end,
+                                           EXTENT_DIRTY);
+               if (ret)
+                       break;
+
+               if (extent_start == start) {
+                       start = extent_end + 1;
+               } else if (extent_start > start && extent_start < end) {
+                       size = extent_start - start;
+                       ret = btrfs_add_free_space(block_group, start, size);
+                       BUG_ON(ret);
+                       start = extent_end + 1;
+               } else {
+                       break;
+               }
+       }
+
+       if (start < end) {
+               size = end - start;
+               ret = btrfs_add_free_space(block_group, start, size);
+               BUG_ON(ret);
+       }
+
+       return 0;
+}
+
 static int cache_block_group(struct btrfs_root *root,
                             struct btrfs_block_group_cache *block_group)
 {
@@ -69,10 +184,8 @@ static int cache_block_group(struct btrfs_root *root,
        int ret = 0;
        struct btrfs_key key;
        struct extent_buffer *leaf;
-       struct extent_io_tree *free_space_cache;
        int slot;
        u64 last = 0;
-       u64 hole_size;
        u64 first_free;
        int found = 0;
 
@@ -80,7 +193,6 @@ static int cache_block_group(struct btrfs_root *root,
                return 0;
 
        root = root->fs_info->extent_root;
-       free_space_cache = &root->fs_info->free_space_cache;
 
        if (block_group->cached)
                return 0;
@@ -96,7 +208,8 @@ static int cache_block_group(struct btrfs_root *root,
         * skip the locking here
         */
        path->skip_locking = 1;
-       first_free = block_group->key.objectid;
+       first_free = max_t(u64, block_group->key.objectid,
+                          BTRFS_SUPER_INFO_OFFSET + BTRFS_SUPER_INFO_SIZE);
        key.objectid = block_group->key.objectid;
        key.offset = 0;
        btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
@@ -119,32 +232,28 @@ static int cache_block_group(struct btrfs_root *root,
                        ret = btrfs_next_leaf(root, path);
                        if (ret < 0)
                                goto err;
-                       if (ret == 0) {
+                       if (ret == 0)
                                continue;
-                       } else {
+                       else
                                break;
-                       }
                }
                btrfs_item_key_to_cpu(leaf, &key, slot);
-               if (key.objectid < block_group->key.objectid) {
+               if (key.objectid < block_group->key.objectid)
                        goto next;
-               }
+
                if (key.objectid >= block_group->key.objectid +
-                   block_group->key.offset) {
+                   block_group->key.offset)
                        break;
-               }
 
                if (btrfs_key_type(&key) == BTRFS_EXTENT_ITEM_KEY) {
                        if (!found) {
                                last = first_free;
                                found = 1;
                        }
-                       if (key.objectid > last) {
-                               hole_size = key.objectid - last;
-                               set_extent_dirty(free_space_cache, last,
-                                                last + hole_size - 1,
-                                                GFP_NOFS);
-                       }
+
+                       add_new_free_space(block_group, root->fs_info, last,
+                                          key.objectid);
+
                        last = key.objectid + key.offset;
                }
 next:
@@ -153,13 +262,11 @@ next:
 
        if (!found)
                last = first_free;
-       if (block_group->key.objectid +
-           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);
-       }
+
+       add_new_free_space(block_group, root->fs_info, last,
+                          block_group->key.objectid +
+                          block_group->key.offset);
+
        block_group->cached = 1;
        ret = 0;
 err:
@@ -167,166 +274,79 @@ err:
        return ret;
 }
 
+/*
+ * return the block group that starts at or after bytenr
+ */
 struct btrfs_block_group_cache *btrfs_lookup_first_block_group(struct
                                                       btrfs_fs_info *info,
                                                         u64 bytenr)
 {
-       struct extent_io_tree *block_group_cache;
-       struct btrfs_block_group_cache *block_group = NULL;
-       u64 ptr;
-       u64 start;
-       u64 end;
-       int ret;
+       struct btrfs_block_group_cache *cache;
 
-       bytenr = max_t(u64, bytenr,
-                      BTRFS_SUPER_INFO_OFFSET + BTRFS_SUPER_INFO_SIZE);
-       block_group_cache = &info->block_group_cache;
-       ret = find_first_extent_bit(block_group_cache,
-                                   bytenr, &start, &end,
-                                   BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA |
-                                   BLOCK_GROUP_SYSTEM);
-       if (ret) {
-               return NULL;
-       }
-       ret = get_state_private(block_group_cache, start, &ptr);
-       if (ret)
-               return NULL;
+       cache = block_group_cache_tree_search(info, bytenr, 0);
 
-       block_group = (struct btrfs_block_group_cache *)(unsigned long)ptr;
-       return block_group;
+       return cache;
 }
 
+/*
+ * return the block group that contains teh given bytenr
+ */
 struct btrfs_block_group_cache *btrfs_lookup_block_group(struct
                                                         btrfs_fs_info *info,
                                                         u64 bytenr)
 {
-       struct extent_io_tree *block_group_cache;
-       struct btrfs_block_group_cache *block_group = NULL;
-       u64 ptr;
-       u64 start;
-       u64 end;
-       int ret;
+       struct btrfs_block_group_cache *cache;
 
-       bytenr = max_t(u64, bytenr,
-                      BTRFS_SUPER_INFO_OFFSET + BTRFS_SUPER_INFO_SIZE);
-       block_group_cache = &info->block_group_cache;
-       ret = find_first_extent_bit(block_group_cache,
-                                   bytenr, &start, &end,
-                                   BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA |
-                                   BLOCK_GROUP_SYSTEM);
-       if (ret) {
-               return NULL;
-       }
-       ret = get_state_private(block_group_cache, start, &ptr);
-       if (ret)
-               return NULL;
+       cache = block_group_cache_tree_search(info, bytenr, 1);
 
-       block_group = (struct btrfs_block_group_cache *)(unsigned long)ptr;
-       if (block_group->key.objectid <= bytenr && bytenr <
-           block_group->key.objectid + block_group->key.offset)
-               return block_group;
-       return NULL;
+       return cache;
 }
 
-static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
-{
-       return (cache->flags & bits) == bits;
-}
-
-static int noinline find_search_start(struct btrfs_root *root,
-                             struct btrfs_block_group_cache **cache_ret,
-                             u64 *start_ret, u64 num, int data)
+static int noinline find_free_space(struct btrfs_root *root,
+                                   struct btrfs_block_group_cache **cache_ret,
+                                   u64 *start_ret, u64 num, int data)
 {
        int ret;
        struct btrfs_block_group_cache *cache = *cache_ret;
-       struct extent_io_tree *free_space_cache;
-       struct extent_state *state;
+       struct btrfs_free_space *info = NULL;
        u64 last;
-       u64 start = 0;
-       u64 cache_miss = 0;
        u64 total_fs_bytes;
        u64 search_start = *start_ret;
-       int wrapped = 0;
 
        WARN_ON(!mutex_is_locked(&root->fs_info->alloc_mutex));
        total_fs_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
-       free_space_cache = &root->fs_info->free_space_cache;
 
        if (!cache)
                goto out;
 
+       last = max(search_start, cache->key.objectid);
+
 again:
        ret = cache_block_group(root, cache);
-       if (ret) {
+       if (ret)
                goto out;
-       }
 
-       last = max(search_start, cache->key.objectid);
-       if (!block_group_bits(cache, data) || cache->ro)
+       if (cache->ro || !block_group_bits(cache, data))
                goto new_group;
 
-       spin_lock_irq(&free_space_cache->lock);
-       state = find_first_extent_bit_state(free_space_cache, last, EXTENT_DIRTY);
-       while(1) {
-               if (!state) {
-                       if (!cache_miss)
-                               cache_miss = last;
-                       spin_unlock_irq(&free_space_cache->lock);
-                       goto new_group;
-               }
-
-               start = max(last, state->start);
-               last = state->end + 1;
-               if (last - start < num) {
-                       do {
-                               state = extent_state_next(state);
-                       } while(state && !(state->state & EXTENT_DIRTY));
-                       continue;
-               }
-               spin_unlock_irq(&free_space_cache->lock);
-               if (cache->ro) {
-                       goto new_group;
-               }
-               if (start + num > cache->key.objectid + cache->key.offset)
-                       goto new_group;
-               if (!block_group_bits(cache, data)) {
-                       printk("block group bits don't match %Lu %d\n", cache->flags, data);
-               }
-               *start_ret = start;
+       info = btrfs_find_free_space(cache, last, num);
+       if (info) {
+               *start_ret = info->offset;
                return 0;
        }
-out:
-       cache = btrfs_lookup_block_group(root->fs_info, search_start);
-       if (!cache) {
-               printk("Unable to find block group for %Lu\n", search_start);
-               WARN_ON(1);
-       }
-       return -ENOSPC;
 
 new_group:
        last = cache->key.objectid + cache->key.offset;
-wrapped:
+
        cache = btrfs_lookup_first_block_group(root->fs_info, last);
-       if (!cache || cache->key.objectid >= total_fs_bytes) {
-no_cache:
-               if (!wrapped) {
-                       wrapped = 1;
-                       last = search_start;
-                       goto wrapped;
-               }
+       if (!cache || cache->key.objectid >= total_fs_bytes)
                goto out;
-       }
-       if (cache_miss && !cache->cached) {
-               cache_block_group(root, cache);
-               last = cache_miss;
-               cache = btrfs_lookup_first_block_group(root->fs_info, last);
-       }
-       cache_miss = 0;
-       cache = btrfs_find_block_group(root, cache, last, data, 0);
-       if (!cache)
-               goto no_cache;
+
        *cache_ret = cache;
        goto again;
+
+out:
+       return -ENOSPC;
 }
 
 static u64 div_factor(u64 num, int factor)
@@ -338,16 +358,19 @@ static u64 div_factor(u64 num, int factor)
        return num;
 }
 
-static int block_group_state_bits(u64 flags)
+static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
+                                                 u64 flags)
 {
-       int bits = 0;
-       if (flags & BTRFS_BLOCK_GROUP_DATA)
-               bits |= BLOCK_GROUP_DATA;
-       if (flags & BTRFS_BLOCK_GROUP_METADATA)
-               bits |= BLOCK_GROUP_METADATA;
-       if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
-               bits |= BLOCK_GROUP_SYSTEM;
-       return bits;
+       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);
+               if (found->flags == flags)
+                       return found;
+       }
+       return NULL;
+
 }
 
 static struct btrfs_block_group_cache *
@@ -356,28 +379,19 @@ __btrfs_find_block_group(struct btrfs_root *root,
                         u64 search_start, int data, int owner)
 {
        struct btrfs_block_group_cache *cache;
-       struct extent_io_tree *block_group_cache;
        struct btrfs_block_group_cache *found_group = NULL;
        struct btrfs_fs_info *info = root->fs_info;
+       struct btrfs_space_info *sinfo;
        u64 used;
        u64 last = 0;
-       u64 start;
-       u64 end;
        u64 free_check;
-       u64 ptr;
-       int bit;
-       int ret;
        int full_search = 0;
        int factor = 10;
        int wrapped = 0;
 
-       block_group_cache = &info->block_group_cache;
-
        if (data & BTRFS_BLOCK_GROUP_METADATA)
                factor = 9;
 
-       bit = block_group_state_bits(data);
-
        if (search_start) {
                struct btrfs_block_group_cache *shint;
                shint = btrfs_lookup_first_block_group(info, search_start);
@@ -408,20 +422,30 @@ __btrfs_find_block_group(struct btrfs_root *root,
                else
                        last = search_start;
        }
+       sinfo = __find_space_info(root->fs_info, data);
+       if (!sinfo)
+               goto found;
 again:
        while(1) {
-               ret = find_first_extent_bit(block_group_cache, last,
-                                           &start, &end, bit);
-               if (ret)
-                       break;
+               struct list_head *l;
 
-               ret = get_state_private(block_group_cache, start, &ptr);
-               if (ret) {
-                       last = end + 1;
-                       continue;
+               cache = NULL;
+
+               spin_lock(&sinfo->lock);
+               list_for_each(l, &sinfo->block_groups) {
+                       struct btrfs_block_group_cache *entry;
+                       entry = list_entry(l, struct btrfs_block_group_cache,
+                                          list);
+                       if ((entry->key.objectid >= last) &&
+                           (!cache || (entry->key.objectid <
+                                       cache->key.objectid)))
+                               cache = entry;
                }
+               spin_unlock(&sinfo->lock);
+
+               if (!cache)
+                       break;
 
-               cache = (struct btrfs_block_group_cache *)(unsigned long)ptr;
                spin_lock(&cache->lock);
                last = cache->key.objectid + cache->key.offset;
                used = btrfs_block_group_used(&cache->item);
@@ -462,6 +486,7 @@ struct btrfs_block_group_cache *btrfs_find_block_group(struct btrfs_root *root,
        ret = __btrfs_find_block_group(root, hint, search_start, data, owner);
        return ret;
 }
+
 static u64 hash_extent_ref(u64 root_objectid, u64 ref_generation,
                           u64 owner, u64 owner_offset)
 {
@@ -1175,34 +1200,37 @@ fail:
 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
                                   struct btrfs_root *root)
 {
-       struct extent_io_tree *block_group_cache;
-       struct btrfs_block_group_cache *cache;
-       int ret;
+       struct btrfs_block_group_cache *cache, *entry;
+       struct rb_node *n;
        int err = 0;
        int werr = 0;
        struct btrfs_path *path;
        u64 last = 0;
-       u64 start;
-       u64 end;
-       u64 ptr;
 
-       block_group_cache = &root->fs_info->block_group_cache;
        path = btrfs_alloc_path();
        if (!path)
                return -ENOMEM;
 
        mutex_lock(&root->fs_info->alloc_mutex);
        while(1) {
-               ret = find_first_extent_bit(block_group_cache, last,
-                                           &start, &end, BLOCK_GROUP_DIRTY);
-               if (ret)
-                       break;
+               cache = NULL;
+               spin_lock(&root->fs_info->block_group_cache_lock);
+               for (n = rb_first(&root->fs_info->block_group_cache_tree);
+                    n; n = rb_next(n)) {
+                       entry = rb_entry(n, struct btrfs_block_group_cache,
+                                        cache_node);
+                       if (entry->dirty) {
+                               cache = entry;
+                               break;
+                       }
+               }
+               spin_unlock(&root->fs_info->block_group_cache_lock);
 
-               last = end + 1;
-               ret = get_state_private(block_group_cache, start, &ptr);
-               if (ret)
+               if (!cache)
                        break;
-               cache = (struct btrfs_block_group_cache *)(unsigned long)ptr;
+
+               last += cache->key.offset;
+
                err = write_one_cache_group(trans, root,
                                            path, cache);
                /*
@@ -1214,29 +1242,14 @@ int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
                        werr = err;
                        continue;
                }
-               clear_extent_bits(block_group_cache, start, end,
-                                 BLOCK_GROUP_DIRTY, GFP_NOFS);
+
+               cache->dirty = 0;
        }
        btrfs_free_path(path);
        mutex_unlock(&root->fs_info->alloc_mutex);
        return werr;
 }
 
-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);
-               if (found->flags == flags)
-                       return found;
-       }
-       return NULL;
-
-}
-
 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
                             u64 total_bytes, u64 bytes_used,
                             struct btrfs_space_info **space_info)
@@ -1256,6 +1269,8 @@ static int update_space_info(struct btrfs_fs_info *info, u64 flags,
                return -ENOMEM;
 
        list_add(&found->list, &info->space_info);
+       INIT_LIST_HEAD(&found->block_groups);
+       spin_lock_init(&found->lock);
        found->flags = flags;
        found->total_bytes = total_bytes;
        found->bytes_used = bytes_used;
@@ -1318,7 +1333,7 @@ static int do_chunk_alloc(struct btrfs_trans_handle *trans,
        u64 thresh;
        u64 start;
        u64 num_bytes;
-       int ret;
+       int ret = 0;
 
        flags = reduce_alloc_profile(extent_root, flags);
 
@@ -1355,10 +1370,11 @@ printk("space info full %Lu\n", flags);
        ret = btrfs_make_block_group(trans, extent_root, 0, flags,
                     BTRFS_FIRST_CHUNK_TREE_OBJECTID, start, num_bytes);
        BUG_ON(ret);
+
 out_unlock:
        mutex_unlock(&extent_root->fs_info->chunk_mutex);
 out:
-       return 0;
+       return ret;
 }
 
 static int update_block_group(struct btrfs_trans_handle *trans,
@@ -1371,8 +1387,6 @@ static int update_block_group(struct btrfs_trans_handle *trans,
        u64 total = num_bytes;
        u64 old_val;
        u64 byte_in_group;
-       u64 start;
-       u64 end;
 
        WARN_ON(!mutex_is_locked(&root->fs_info->alloc_mutex));
        while(total) {
@@ -1382,12 +1396,9 @@ static int update_block_group(struct btrfs_trans_handle *trans,
                }
                byte_in_group = bytenr - cache->key.objectid;
                WARN_ON(byte_in_group > cache->key.offset);
-               start = cache->key.objectid;
-               end = start + cache->key.offset - 1;
-               set_extent_bits(&info->block_group_cache, start, end,
-                               BLOCK_GROUP_DIRTY, GFP_NOFS);
 
                spin_lock(&cache->lock);
+               cache->dirty = 1;
                old_val = btrfs_block_group_used(&cache->item);
                num_bytes = min(total, cache->key.offset - byte_in_group);
                if (alloc) {
@@ -1401,9 +1412,11 @@ static int update_block_group(struct btrfs_trans_handle *trans,
                        btrfs_set_block_group_used(&cache->item, old_val);
                        spin_unlock(&cache->lock);
                        if (mark_free) {
-                               set_extent_dirty(&info->free_space_cache,
-                                                bytenr, bytenr + num_bytes - 1,
-                                                GFP_NOFS);
+                               int ret;
+                               ret = btrfs_add_free_space(cache, bytenr,
+                                                          num_bytes);
+                               if (ret)
+                                       return -1;
                        }
                }
                total -= num_bytes;
@@ -1414,16 +1427,13 @@ static int update_block_group(struct btrfs_trans_handle *trans,
 
 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
 {
-       u64 start;
-       u64 end;
-       int ret;
-       ret = find_first_extent_bit(&root->fs_info->block_group_cache,
-                                   search_start, &start, &end,
-                                   BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA |
-                                   BLOCK_GROUP_SYSTEM);
-       if (ret)
+       struct btrfs_block_group_cache *cache;
+
+       cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
+       if (!cache)
                return 0;
-       return start;
+
+       return cache->key.objectid;
 }
 
 
@@ -1501,8 +1511,7 @@ int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
        u64 start;
        u64 end;
        int ret;
-       struct extent_io_tree *free_space_cache;
-       free_space_cache = &root->fs_info->free_space_cache;
+       struct btrfs_block_group_cache *cache;
 
        mutex_lock(&root->fs_info->alloc_mutex);
        while(1) {
@@ -1512,7 +1521,9 @@ int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
                        break;
                btrfs_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);
+               cache = btrfs_lookup_block_group(root->fs_info, start);
+               if (cache->cached)
+                       btrfs_add_free_space(cache, start, end - start + 1);
                if (need_resched()) {
                        mutex_unlock(&root->fs_info->alloc_mutex);
                        cond_resched();
@@ -1875,9 +1886,12 @@ static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
        /* if metadata always pin */
        if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
                if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
+                       struct btrfs_block_group_cache *cache;
+
                        /* btrfs_free_reserved_extent */
-                       set_extent_dirty(&root->fs_info->free_space_cache,
-                                bytenr, bytenr + num_bytes - 1, GFP_NOFS);
+                       cache = btrfs_lookup_block_group(root->fs_info, bytenr);
+                       BUG_ON(!cache);
+                       btrfs_add_free_space(cache, bytenr, num_bytes);
                        return 0;
                }
                pin = 1;
@@ -1942,8 +1956,6 @@ static int noinline find_free_extent(struct btrfs_trans_handle *trans,
        u64 total_needed = num_bytes;
        u64 *last_ptr = NULL;
        struct btrfs_block_group_cache *block_group;
-       int full_scan = 0;
-       int wrapped = 0;
        int chunk_alloc_done = 0;
        int empty_cluster = 2 * 1024 * 1024;
        int allowed_chunk_alloc = 0;
@@ -1959,9 +1971,9 @@ static int noinline find_free_extent(struct btrfs_trans_handle *trans,
                empty_cluster = 256 * 1024;
        }
 
-       if ((data & BTRFS_BLOCK_GROUP_DATA) && btrfs_test_opt(root, SSD)) {
+       if ((data & BTRFS_BLOCK_GROUP_DATA) && btrfs_test_opt(root, SSD))
                last_ptr = &root->fs_info->last_data_alloc;
-       }
+
        if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
                last_ptr = &root->fs_info->last_log_alloc;
                if (!last_ptr == 0 && root->fs_info->last_alloc) {
@@ -1972,9 +1984,8 @@ static int noinline find_free_extent(struct btrfs_trans_handle *trans,
        if (last_ptr) {
                if (*last_ptr)
                        hint_byte = *last_ptr;
-               else {
+               else
                        empty_size += empty_cluster;
-               }
        }
 
        search_start = max(search_start, first_logical_byte(root, 0));
@@ -1983,145 +1994,172 @@ static int noinline find_free_extent(struct btrfs_trans_handle *trans,
        if (search_end == (u64)-1)
                search_end = btrfs_super_total_bytes(&info->super_copy);
 
-       if (hint_byte) {
-               block_group = btrfs_lookup_first_block_group(info, hint_byte);
-               if (!block_group)
-                       hint_byte = search_start;
-               block_group = btrfs_find_block_group(root, block_group,
-                                                    hint_byte, data, 1);
-               if (last_ptr && *last_ptr == 0 && block_group)
-                       hint_byte = block_group->key.objectid;
-       } else {
-               block_group = btrfs_find_block_group(root,
-                                                    trans->block_group,
-                                                    search_start, data, 1);
-       }
        search_start = max(search_start, hint_byte);
-
        total_needed += empty_size;
 
-check_failed:
-       if (!block_group) {
-               block_group = btrfs_lookup_first_block_group(info,
-                                                            search_start);
-               if (!block_group)
-                       block_group = btrfs_lookup_first_block_group(info,
-                                                      orig_search_start);
-       }
-       if (full_scan && !chunk_alloc_done) {
-               if (allowed_chunk_alloc) {
-                       do_chunk_alloc(trans, root,
-                                    num_bytes + 2 * 1024 * 1024, data, 1);
-                       allowed_chunk_alloc = 0;
-               } else if (block_group && block_group_bits(block_group, data)) {
-                       block_group->space_info->force_alloc = 1;
+new_group:
+       block_group = btrfs_lookup_block_group(info, search_start);
+
+       /*
+        * Ok this looks a little tricky, buts its really simple.  First if we
+        * didn't find a block group obviously we want to start over.
+        * Secondly, if the block group we found does not match the type we
+        * need, and we have a last_ptr and its not 0, chances are the last
+        * allocation we made was at the end of the block group, so lets go
+        * ahead and skip the looking through the rest of the block groups and
+        * start at the beginning.  This helps with metadata allocations,
+        * since you are likely to have a bunch of data block groups to search
+        * through first before you realize that you need to start over, so go
+        * ahead and start over and save the time.
+        */
+       if (!block_group || (!block_group_bits(block_group, data) &&
+                            last_ptr && *last_ptr)) {
+               if (search_start != orig_search_start) {
+                       if (last_ptr && *last_ptr)
+                               *last_ptr = 0;
+                       search_start = orig_search_start;
+                       goto new_group;
+               } else if (!chunk_alloc_done && allowed_chunk_alloc) {
+                       ret = do_chunk_alloc(trans, root,
+                                            num_bytes + 2 * 1024 * 1024,
+                                            data, 1);
+                       if (ret < 0) {
+                               struct btrfs_space_info *info;
+
+                               info = __find_space_info(root->fs_info, data);
+                               goto error;
+                       }
+                       BUG_ON(ret);
+                       chunk_alloc_done = 1;
+                       search_start = orig_search_start;
+                       goto new_group;
+               } else {
+                       ret = -ENOSPC;
+                       goto error;
                }
-               chunk_alloc_done = 1;
-       }
-       ret = find_search_start(root, &block_group, &search_start,
-                               total_needed, data);
-       if (ret == -ENOSPC && last_ptr && *last_ptr) {
-               *last_ptr = 0;
-               block_group = btrfs_lookup_first_block_group(info,
-                                                            orig_search_start);
-               search_start = orig_search_start;
-               ret = find_search_start(root, &block_group, &search_start,
-                                       total_needed, data);
        }
-       if (ret == -ENOSPC)
-               goto enospc;
-       if (ret)
-               goto error;
 
-       if (last_ptr && *last_ptr && search_start != *last_ptr) {
-               *last_ptr = 0;
-               if (!empty_size) {
-                       empty_size += empty_cluster;
-                       total_needed += empty_size;
+       /*
+        * this is going to seach through all of the existing block groups it
+        * can find, so if we don't find something we need to see if we can
+        * allocate what we need.
+        */
+       ret = find_free_space(root, &block_group, &search_start,
+                             total_needed, data);
+       if (ret == -ENOSPC) {
+               /*
+                * instead of allocating, start at the original search start
+                * and see if there is something to be found, if not then we
+                * allocate
+                */
+               if (search_start != orig_search_start) {
+                       if (last_ptr && *last_ptr) {
+                               *last_ptr = 0;
+                               total_needed += empty_cluster;
+                       }
+                       search_start = orig_search_start;
+                       goto new_group;
                }
-               block_group = btrfs_lookup_first_block_group(info,
-                                                      orig_search_start);
-               search_start = orig_search_start;
-               ret = find_search_start(root, &block_group,
-                                       &search_start, total_needed, data);
-               if (ret == -ENOSPC)
-                       goto enospc;
-               if (ret)
+
+               /*
+                * we've already allocated, we're pretty screwed
+                */
+               if (chunk_alloc_done) {
                        goto error;
+               } else if (!allowed_chunk_alloc && block_group &&
+                          block_group_bits(block_group, data)) {
+                       block_group->space_info->force_alloc = 1;
+                       goto error;
+               } else if (!allowed_chunk_alloc) {
+                       goto error;
+               }
+
+               ret = do_chunk_alloc(trans, root, num_bytes + 2 * 1024 * 1024,
+                                    data, 1);
+               if (ret < 0)
+                       goto error;
+
+               BUG_ON(ret);
+               chunk_alloc_done = 1;
+               if (block_group)
+                       search_start = block_group->key.objectid +
+                               block_group->key.offset;
+               else
+                       search_start = orig_search_start;
+               goto new_group;
        }
 
+       if (ret)
+               goto error;
+
        search_start = stripe_align(root, search_start);
        ins->objectid = search_start;
        ins->offset = num_bytes;
 
-       if (ins->objectid + num_bytes >= search_end)
-               goto enospc;
+       if (ins->objectid + num_bytes >= search_end) {
+               search_start = orig_search_start;
+               if (chunk_alloc_done) {
+                       ret = -ENOSPC;
+                       goto error;
+               }
+               goto new_group;
+       }
 
        if (ins->objectid + num_bytes >
            block_group->key.objectid + block_group->key.offset) {
+               if (search_start == orig_search_start && chunk_alloc_done) {
+                       ret = -ENOSPC;
+                       goto error;
+               }
                search_start = block_group->key.objectid +
                        block_group->key.offset;
                goto new_group;
        }
 
-       if (test_range_bit(&info->extent_ins, ins->objectid,
-                          ins->objectid + num_bytes -1, EXTENT_LOCKED, 0)) {
-               search_start = ins->objectid + num_bytes;
-               goto new_group;
-       }
-
-       if (test_range_bit(&info->pinned_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;
                goto new_group;
        }
 
-       if (!(data & BTRFS_BLOCK_GROUP_DATA)) {
-               block_group = btrfs_lookup_block_group(info, ins->objectid);
-               if (block_group)
-                       trans->block_group = block_group;
-       }
+       if (!(data & BTRFS_BLOCK_GROUP_DATA))
+               trans->block_group = block_group;
+
        ins->offset = num_bytes;
        if (last_ptr) {
                *last_ptr = ins->objectid + ins->offset;
                if (*last_ptr ==
-                   btrfs_super_total_bytes(&root->fs_info->super_copy)) {
+                   btrfs_super_total_bytes(&root->fs_info->super_copy))
                        *last_ptr = 0;
-               }
-       }
-       return 0;
-
-new_group:
-       if (search_start + num_bytes >= search_end) {
-enospc:
-               search_start = orig_search_start;
-               if (full_scan) {
-                       ret = -ENOSPC;
-                       goto error;
-               }
-               if (wrapped) {
-                       if (!full_scan)
-                               total_needed -= empty_size;
-                       full_scan = 1;
-               } else
-                       wrapped = 1;
        }
-       block_group = btrfs_lookup_first_block_group(info, search_start);
-       cond_resched();
-       block_group = btrfs_find_block_group(root, block_group,
-                                            search_start, data, 0);
-       goto check_failed;
 
+       ret = 0;
 error:
        return ret;
 }
 
+static void dump_space_info(struct btrfs_space_info *info, u64 bytes)
+{
+       struct btrfs_block_group_cache *cache;
+       struct list_head *l;
+
+       printk(KERN_INFO "space_info has %Lu free, is %sfull\n",
+              info->total_bytes - info->bytes_used - info->bytes_pinned,
+              (info->full) ? "" : "not ");
+
+       spin_lock(&info->lock);
+       list_for_each(l, &info->block_groups) {
+               cache = list_entry(l, struct btrfs_block_group_cache, list);
+               spin_lock(&cache->lock);
+               printk(KERN_INFO "block group %Lu has %Lu bytes, %Lu used "
+                      "%Lu pinned\n",
+                      cache->key.objectid, cache->key.offset,
+                      btrfs_block_group_used(&cache->item), cache->pinned);
+               btrfs_dump_free_space(cache, bytes);
+               spin_unlock(&cache->lock);
+       }
+       spin_unlock(&info->lock);
+}
 static int __btrfs_reserve_extent(struct btrfs_trans_handle *trans,
                                  struct btrfs_root *root,
                                  u64 num_bytes, u64 min_alloc_size,
@@ -2133,6 +2171,7 @@ static int __btrfs_reserve_extent(struct btrfs_trans_handle *trans,
        u64 search_start = 0;
        u64 alloc_profile;
        struct btrfs_fs_info *info = root->fs_info;
+       struct btrfs_block_group_cache *cache;
 
        if (data) {
                alloc_profile = info->avail_data_alloc_bits &
@@ -2160,11 +2199,9 @@ again:
                                     BTRFS_BLOCK_GROUP_METADATA |
                                     (info->metadata_alloc_profile &
                                      info->avail_metadata_alloc_bits), 0);
-                       BUG_ON(ret);
                }
                ret = do_chunk_alloc(trans, root->fs_info->extent_root,
                                     num_bytes + 2 * 1024 * 1024, data, 0);
-               BUG_ON(ret);
        }
 
        WARN_ON(num_bytes < root->sectorsize);
@@ -2175,26 +2212,44 @@ again:
 
        if (ret == -ENOSPC && num_bytes > min_alloc_size) {
                num_bytes = num_bytes >> 1;
+               num_bytes = num_bytes & ~(root->sectorsize - 1);
                num_bytes = max(num_bytes, min_alloc_size);
                do_chunk_alloc(trans, root->fs_info->extent_root,
                               num_bytes, data, 1);
                goto again;
        }
        if (ret) {
-               printk("allocation failed flags %Lu\n", data);
+               struct btrfs_space_info *sinfo;
+
+               sinfo = __find_space_info(root->fs_info, data);
+               printk("allocation failed flags %Lu, wanted %Lu\n",
+                      data, num_bytes);
+               dump_space_info(sinfo, num_bytes);
                BUG();
        }
-       clear_extent_dirty(&root->fs_info->free_space_cache,
-                          ins->objectid, ins->objectid + ins->offset - 1,
-                          GFP_NOFS);
-       return 0;
+       cache = btrfs_lookup_block_group(root->fs_info, ins->objectid);
+       if (!cache) {
+               printk(KERN_ERR "Unable to find block group for %Lu\n", ins->objectid);
+               return -ENOSPC;
+       }
+
+       ret = btrfs_remove_free_space(cache, ins->objectid, ins->offset);
+
+       return ret;
 }
 
 int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len)
 {
+       struct btrfs_block_group_cache *cache;
+
        maybe_lock_mutex(root);
-       set_extent_dirty(&root->fs_info->free_space_cache,
-                        start, start + len - 1, GFP_NOFS);
+       cache = btrfs_lookup_block_group(root->fs_info, start);
+       if (!cache) {
+               printk(KERN_ERR "Unable to find block group for %Lu\n", start);
+               maybe_unlock_mutex(root);
+               return -ENOSPC;
+       }
+       btrfs_add_free_space(cache, start, len);
        maybe_unlock_mutex(root);
        return 0;
 }
@@ -2264,8 +2319,8 @@ static int __btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
 
        ret = btrfs_insert_empty_items(trans, extent_root, path, keys,
                                       sizes, 2);
-
        BUG_ON(ret);
+
        extent_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
                                     struct btrfs_extent_item);
        btrfs_set_extent_refs(path->nodes[0], extent_item, 1);
@@ -2336,9 +2391,9 @@ int btrfs_alloc_logged_extent(struct btrfs_trans_handle *trans,
        block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
        cache_block_group(root, block_group);
 
-       clear_extent_dirty(&root->fs_info->free_space_cache,
-                          ins->objectid, ins->objectid + ins->offset - 1,
-                          GFP_NOFS);
+       ret = btrfs_remove_free_space(block_group, ins->objectid, ins->offset);
+       BUG_ON(ret);
+
        ret = __btrfs_alloc_reserved_extent(trans, root, root_objectid,
                                            ref_generation, owner,
                                            owner_offset, ins);
@@ -2843,31 +2898,24 @@ out:
 
 int btrfs_free_block_groups(struct btrfs_fs_info *info)
 {
-       u64 start;
-       u64 end;
-       u64 ptr;
-       int ret;
+       struct btrfs_block_group_cache *block_group;
+       struct rb_node *n;
 
        mutex_lock(&info->alloc_mutex);
-       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);
-               clear_extent_bits(&info->block_group_cache, start,
-                                 end, (unsigned int)-1, GFP_NOFS);
-       }
-       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);
-       }
+       spin_lock(&info->block_group_cache_lock);
+       while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
+               block_group = rb_entry(n, struct btrfs_block_group_cache,
+                                      cache_node);
+
+               btrfs_remove_free_space_cache(block_group);
+               rb_erase(&block_group->cache_node,
+                        &info->block_group_cache_tree);
+               spin_lock(&block_group->space_info->lock);
+               list_del(&block_group->list);
+               spin_unlock(&block_group->space_info->lock);
+               kfree(block_group);
+       }
+       spin_unlock(&info->block_group_cache_lock);
        mutex_unlock(&info->alloc_mutex);
        return 0;
 }
@@ -3386,7 +3434,6 @@ int btrfs_shrink_extent_tree(struct btrfs_root *root, u64 shrink_start)
        u64 total_found;
        u64 shrink_last_byte;
        struct btrfs_block_group_cache *shrink_block_group;
-       struct btrfs_fs_info *info = root->fs_info;
        struct btrfs_key key;
        struct btrfs_key found_key;
        struct extent_buffer *leaf;
@@ -3542,15 +3589,17 @@ next:
                goto out;
        }
 
-       clear_extent_bits(&info->block_group_cache, key.objectid,
-                         key.objectid + key.offset - 1,
-                         (unsigned int)-1, GFP_NOFS);
-
-
-       clear_extent_bits(&info->free_space_cache,
-                          key.objectid, key.objectid + key.offset - 1,
-                          (unsigned int)-1, GFP_NOFS);
+       spin_lock(&root->fs_info->block_group_cache_lock);
+       rb_erase(&shrink_block_group->cache_node,
+                &root->fs_info->block_group_cache_tree);
+       spin_unlock(&root->fs_info->block_group_cache_lock);
 
+       ret = btrfs_remove_free_space(shrink_block_group, key.objectid,
+                                     key.offset);
+       if (ret) {
+               btrfs_end_transaction(trans, root);
+               goto out;
+       }
        /*
        memset(shrink_block_group, 0, sizeof(*shrink_block_group));
        kfree(shrink_block_group);
@@ -3566,9 +3615,9 @@ next:
        /* the code to unpin extents might set a few bits in the free
         * space cache for this range again
         */
-       clear_extent_bits(&info->free_space_cache,
-                          key.objectid, key.objectid + key.offset - 1,
-                          (unsigned int)-1, GFP_NOFS);
+       /* XXX? */
+       ret = btrfs_remove_free_space(shrink_block_group, key.objectid,
+                                     key.offset);
 out:
        btrfs_free_path(path);
        mutex_unlock(&root->fs_info->alloc_mutex);
@@ -3616,16 +3665,13 @@ 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;
@@ -3653,6 +3699,7 @@ int btrfs_read_block_groups(struct btrfs_root *root)
                }
 
                spin_lock_init(&cache->lock);
+               INIT_LIST_HEAD(&cache->list);
                read_extent_buffer(leaf, &cache->item,
                                   btrfs_item_ptr_offset(leaf, path->slots[0]),
                                   sizeof(cache->item));
@@ -3661,31 +3708,19 @@ int btrfs_read_block_groups(struct btrfs_root *root)
                key.objectid = found_key.objectid + found_key.offset;
                btrfs_release_path(root, path);
                cache->flags = btrfs_block_group_flags(&cache->item);
-               bit = 0;
-               if (cache->flags & BTRFS_BLOCK_GROUP_DATA) {
-                       bit = BLOCK_GROUP_DATA;
-               } else if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) {
-                       bit = BLOCK_GROUP_SYSTEM;
-               } else if (cache->flags & BTRFS_BLOCK_GROUP_METADATA) {
-                       bit = BLOCK_GROUP_METADATA;
-               }
-               set_avail_alloc_bits(info, cache->flags);
 
                ret = update_space_info(info, cache->flags, found_key.offset,
                                        btrfs_block_group_used(&cache->item),
                                        &space_info);
                BUG_ON(ret);
                cache->space_info = space_info;
+               spin_lock(&space_info->lock);
+               list_add(&cache->list, &space_info->block_groups);
+               spin_unlock(&space_info->lock);
+
+               ret = btrfs_add_block_group_cache(root->fs_info, cache);
+               BUG_ON(ret);
 
-               /* use EXTENT_LOCKED to prevent merging */
-               set_extent_bits(block_group_cache, found_key.objectid,
-                               found_key.objectid + found_key.offset - 1,
-                               EXTENT_LOCKED, GFP_NOFS);
-               set_state_private(block_group_cache, found_key.objectid,
-                                 (unsigned long)cache);
-               set_extent_bits(block_group_cache, found_key.objectid,
-                               found_key.objectid + found_key.offset - 1,
-                               bit | EXTENT_LOCKED, GFP_NOFS);
                if (key.objectid >=
                    btrfs_super_total_bytes(&info->super_copy))
                        break;
@@ -3703,22 +3738,22 @@ int btrfs_make_block_group(struct btrfs_trans_handle *trans,
                           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;
 
        WARN_ON(!mutex_is_locked(&root->fs_info->alloc_mutex));
        extent_root = root->fs_info->extent_root;
-       block_group_cache = &root->fs_info->block_group_cache;
 
        root->fs_info->last_trans_new_blockgroup = trans->transid;
 
        cache = kzalloc(sizeof(*cache), GFP_NOFS);
-       BUG_ON(!cache);
+       if (!cache)
+               return -ENOMEM;
+
        cache->key.objectid = chunk_offset;
        cache->key.offset = size;
        spin_lock_init(&cache->lock);
+       INIT_LIST_HEAD(&cache->list);
        btrfs_set_key_type(&cache->key, BTRFS_BLOCK_GROUP_ITEM_KEY);
 
        btrfs_set_block_group_used(&cache->item, bytes_used);
@@ -3729,16 +3764,12 @@ int btrfs_make_block_group(struct btrfs_trans_handle *trans,
        ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
                                &cache->space_info);
        BUG_ON(ret);
+       spin_lock(&cache->space_info->lock);
+       list_add(&cache->list, &cache->space_info->block_groups);
+       spin_unlock(&cache->space_info->lock);
 
-       bit = block_group_state_bits(type);
-       set_extent_bits(block_group_cache, chunk_offset,
-                       chunk_offset + size - 1,
-                       EXTENT_LOCKED, GFP_NOFS);
-       set_state_private(block_group_cache, chunk_offset,
-                         (unsigned long)cache);
-       set_extent_bits(block_group_cache, chunk_offset,
-                       chunk_offset + size - 1,
-                       bit | EXTENT_LOCKED, GFP_NOFS);
+       ret = btrfs_add_block_group_cache(root->fs_info, cache);
+       BUG_ON(ret);
 
        ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
                                sizeof(cache->item));
index 319a0c7a4a58db186c8f07cc28ee679380558cb8..8624f3e880364496858b4cb4fbd9d0f98deabdfc 100644 (file)
@@ -2634,6 +2634,7 @@ struct extent_buffer *alloc_extent_buffer(struct extent_io_tree *tree,
        if (eb) {
                atomic_inc(&eb->refs);
                spin_unlock(&tree->buffer_lock);
+               mark_page_accessed(eb->first_page);
                return eb;
        }
        spin_unlock(&tree->buffer_lock);
@@ -2713,6 +2714,9 @@ struct extent_buffer *find_extent_buffer(struct extent_io_tree *tree,
                atomic_inc(&eb->refs);
        spin_unlock(&tree->buffer_lock);
 
+       if (eb)
+               mark_page_accessed(eb->first_page);
+
        return eb;
 }
 EXPORT_SYMBOL(find_extent_buffer);
diff --git a/fs/btrfs/free-space-cache.c b/fs/btrfs/free-space-cache.c
new file mode 100644 (file)
index 0000000..01c26e8
--- /dev/null
@@ -0,0 +1,415 @@
+/*
+ * Copyright (C) 2008 Red Hat.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License v2 as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+#include <linux/sched.h>
+#include "ctree.h"
+
+static int tree_insert_offset(struct rb_root *root, u64 offset,
+                             struct rb_node *node)
+{
+       struct rb_node **p = &root->rb_node;
+       struct rb_node *parent = NULL;
+       struct btrfs_free_space *info;
+
+       while (*p) {
+               parent = *p;
+               info = rb_entry(parent, struct btrfs_free_space, offset_index);
+
+               if (offset < info->offset)
+                       p = &(*p)->rb_left;
+               else if (offset > info->offset)
+                       p = &(*p)->rb_right;
+               else
+                       return -EEXIST;
+       }
+
+       rb_link_node(node, parent, p);
+       rb_insert_color(node, root);
+
+       return 0;
+}
+
+static int tree_insert_bytes(struct rb_root *root, u64 bytes,
+                            struct rb_node *node)
+{
+       struct rb_node **p = &root->rb_node;
+       struct rb_node *parent = NULL;
+       struct btrfs_free_space *info;
+
+       while (*p) {
+               parent = *p;
+               info = rb_entry(parent, struct btrfs_free_space, bytes_index);
+
+               if (bytes < info->bytes)
+                       p = &(*p)->rb_left;
+               else
+                       p = &(*p)->rb_right;
+       }
+
+       rb_link_node(node, parent, p);
+       rb_insert_color(node, root);
+
+       return 0;
+}
+
+/*
+ * searches the tree for the given offset.  If contains is set we will return
+ * the free space that contains the given offset.  If contains is not set we
+ * will return the free space that starts at or after the given offset and is
+ * at least bytes long.
+ */
+static struct btrfs_free_space *tree_search_offset(struct rb_root *root,
+                                                  u64 offset, u64 bytes,
+                                                  int contains)
+{
+       struct rb_node *n = root->rb_node;
+       struct btrfs_free_space *entry, *ret = NULL;
+
+       while (n) {
+               entry = rb_entry(n, struct btrfs_free_space, offset_index);
+
+               if (offset < entry->offset) {
+                       if (!contains &&
+                           (!ret || entry->offset < ret->offset) &&
+                           (bytes <= entry->bytes))
+                               ret = entry;
+                       n = n->rb_left;
+               } else if (offset > entry->offset) {
+                       if (contains &&
+                           (entry->offset + entry->bytes - 1) >= offset) {
+                               ret = entry;
+                               break;
+                       }
+                       n = n->rb_right;
+               } else {
+                       if (bytes > entry->bytes) {
+                               n = n->rb_right;
+                               continue;
+                       }
+                       ret = entry;
+                       break;
+               }
+       }
+
+       return ret;
+}
+
+/*
+ * return a chunk at least bytes size, as close to offset that we can get.
+ */
+static struct btrfs_free_space *tree_search_bytes(struct rb_root *root,
+                                                 u64 offset, u64 bytes)
+{
+       struct rb_node *n = root->rb_node;
+       struct btrfs_free_space *entry, *ret = NULL;
+
+       while (n) {
+               entry = rb_entry(n, struct btrfs_free_space, bytes_index);
+
+               if (bytes < entry->bytes) {
+                       /*
+                        * We prefer to get a hole size as close to the size we
+                        * are asking for so we don't take small slivers out of
+                        * huge holes, but we also want to get as close to the
+                        * offset as possible so we don't have a whole lot of
+                        * fragmentation.
+                        */
+                       if (offset <= entry->offset) {
+                               if (!ret)
+                                       ret = entry;
+                               else if (entry->bytes < ret->bytes)
+                                       ret = entry;
+                               else if (entry->offset < ret->offset)
+                                       ret = entry;
+                       }
+                       n = n->rb_left;
+               } else if (bytes > entry->bytes) {
+                       n = n->rb_right;
+               } else {
+                       /*
+                        * Ok we may have multiple chunks of the wanted size,
+                        * so we don't want to take the first one we find, we
+                        * want to take the one closest to our given offset, so
+                        * keep searching just in case theres a better match.
+                        */
+                       n = n->rb_right;
+                       if (offset > entry->offset)
+                               continue;
+                       else if (!ret || entry->offset < ret->offset)
+                               ret = entry;
+               }
+       }
+
+       return ret;
+}
+
+static void unlink_free_space(struct btrfs_block_group_cache *block_group,
+                             struct btrfs_free_space *info)
+{
+       rb_erase(&info->offset_index, &block_group->free_space_offset);
+       rb_erase(&info->bytes_index, &block_group->free_space_bytes);
+}
+
+static int link_free_space(struct btrfs_block_group_cache *block_group,
+                          struct btrfs_free_space *info)
+{
+       int ret = 0;
+
+
+       ret = tree_insert_offset(&block_group->free_space_offset, info->offset,
+                                &info->offset_index);
+       if (ret)
+               return ret;
+
+       ret = tree_insert_bytes(&block_group->free_space_bytes, info->bytes,
+                               &info->bytes_index);
+       if (ret)
+               return ret;
+
+       return ret;
+}
+
+int btrfs_add_free_space(struct btrfs_block_group_cache *block_group,
+                        u64 offset, u64 bytes)
+{
+       struct btrfs_free_space *right_info;
+       struct btrfs_free_space *left_info;
+       struct btrfs_free_space *info = NULL;
+       struct btrfs_free_space *alloc_info;
+       int ret = 0;
+
+       alloc_info = kzalloc(sizeof(struct btrfs_free_space), GFP_NOFS);
+       if (!alloc_info)
+               return -ENOMEM;
+
+       /*
+        * first we want to see if there is free space adjacent to the range we
+        * are adding, if there is remove that struct and add a new one to
+        * cover the entire range
+        */
+       spin_lock(&block_group->lock);
+
+       right_info = tree_search_offset(&block_group->free_space_offset,
+                                       offset+bytes, 0, 1);
+       left_info = tree_search_offset(&block_group->free_space_offset,
+                                      offset-1, 0, 1);
+
+       if (right_info && right_info->offset == offset+bytes) {
+               unlink_free_space(block_group, right_info);
+               info = right_info;
+               info->offset = offset;
+               info->bytes += bytes;
+       } else if (right_info && right_info->offset != offset+bytes) {
+               printk(KERN_ERR "adding space in the middle of an existing "
+                      "free space area. existing: offset=%Lu, bytes=%Lu. "
+                      "new: offset=%Lu, bytes=%Lu\n", right_info->offset,
+                      right_info->bytes, offset, bytes);
+               BUG();
+       }
+
+       if (left_info) {
+               unlink_free_space(block_group, left_info);
+
+               if (unlikely((left_info->offset + left_info->bytes) !=
+                            offset)) {
+                       printk(KERN_ERR "free space to the left of new free "
+                              "space isn't quite right. existing: offset=%Lu,"
+                              " bytes=%Lu. new: offset=%Lu, bytes=%Lu\n",
+                              left_info->offset, left_info->bytes, offset,
+                              bytes);
+                       BUG();
+               }
+
+               if (info) {
+                       info->offset = left_info->offset;
+                       info->bytes += left_info->bytes;
+                       kfree(left_info);
+               } else {
+                       info = left_info;
+                       info->bytes += bytes;
+               }
+       }
+
+       if (info) {
+               ret = link_free_space(block_group, info);
+               if (!ret)
+                       info = NULL;
+               goto out;
+       }
+
+       info = alloc_info;
+       alloc_info = NULL;
+       info->offset = offset;
+       info->bytes = bytes;
+
+       ret = link_free_space(block_group, info);
+       if (ret)
+               kfree(info);
+out:
+       spin_unlock(&block_group->lock);
+       if (ret) {
+               printk(KERN_ERR "btrfs: unable to add free space :%d\n", ret);
+               if (ret == -EEXIST)
+                       BUG();
+       }
+
+       if (alloc_info)
+               kfree(alloc_info);
+
+       return ret;
+}
+
+int btrfs_remove_free_space(struct btrfs_block_group_cache *block_group,
+                           u64 offset, u64 bytes)
+{
+       struct btrfs_free_space *info;
+       int ret = 0;
+
+       spin_lock(&block_group->lock);
+       info = tree_search_offset(&block_group->free_space_offset, offset, 0,
+                                 1);
+
+       if (info && info->offset == offset) {
+               if (info->bytes < bytes) {
+                       printk(KERN_ERR "Found free space at %Lu, size %Lu,"
+                              "trying to use %Lu\n",
+                              info->offset, info->bytes, bytes);
+                       WARN_ON(1);
+                       ret = -EINVAL;
+                       goto out;
+               }
+
+               unlink_free_space(block_group, info);
+
+               if (info->bytes == bytes) {
+                       kfree(info);
+                       goto out;
+               }
+
+               info->offset += bytes;
+               info->bytes -= bytes;
+
+               ret = link_free_space(block_group, info);
+               BUG_ON(ret);
+       } else {
+               WARN_ON(1);
+       }
+out:
+       spin_unlock(&block_group->lock);
+       return ret;
+}
+
+void btrfs_dump_free_space(struct btrfs_block_group_cache *block_group,
+                          u64 bytes)
+{
+       struct btrfs_free_space *info;
+       struct rb_node *n;
+       int count = 0;
+
+       for (n = rb_first(&block_group->free_space_offset); n; n = rb_next(n)) {
+               info = rb_entry(n, struct btrfs_free_space, offset_index);
+               if (info->bytes >= bytes)
+                       count++;
+               //printk(KERN_INFO "offset=%Lu, bytes=%Lu\n", info->offset,
+               //       info->bytes);
+       }
+       printk(KERN_INFO "%d blocks of free space at or bigger than bytes is"
+              "\n", count);
+}
+
+u64 btrfs_block_group_free_space(struct btrfs_block_group_cache *block_group)
+{
+       struct btrfs_free_space *info;
+       struct rb_node *n;
+       u64 ret = 0;
+
+       for (n = rb_first(&block_group->free_space_offset); n;
+            n = rb_next(n)) {
+               info = rb_entry(n, struct btrfs_free_space, offset_index);
+               ret += info->bytes;
+       }
+
+       return ret;
+}
+
+void btrfs_remove_free_space_cache(struct btrfs_block_group_cache *block_group)
+{
+       struct btrfs_free_space *info;
+       struct rb_node *node;
+
+       spin_lock(&block_group->lock);
+       while ((node = rb_last(&block_group->free_space_bytes)) != NULL) {
+               info = rb_entry(node, struct btrfs_free_space, bytes_index);
+               unlink_free_space(block_group, info);
+               kfree(info);
+               if (need_resched()) {
+                       spin_unlock(&block_group->lock);
+                       cond_resched();
+                       spin_lock(&block_group->lock);
+               }
+       }
+       spin_unlock(&block_group->lock);
+}
+
+struct btrfs_free_space *btrfs_find_free_space_offset(struct
+                                                     btrfs_block_group_cache
+                                                     *block_group, u64 offset,
+                                                     u64 bytes)
+{
+       struct btrfs_free_space *ret;
+
+       spin_lock(&block_group->lock);
+       ret = tree_search_offset(&block_group->free_space_offset, offset,
+                                bytes, 0);
+       spin_unlock(&block_group->lock);
+
+       return ret;
+}
+
+struct btrfs_free_space *btrfs_find_free_space_bytes(struct
+                                                    btrfs_block_group_cache
+                                                    *block_group, u64 offset,
+                                                    u64 bytes)
+{
+       struct btrfs_free_space *ret;
+
+       spin_lock(&block_group->lock);
+
+       ret = tree_search_bytes(&block_group->free_space_bytes, offset, bytes);
+       spin_unlock(&block_group->lock);
+
+       return ret;
+}
+
+struct btrfs_free_space *btrfs_find_free_space(struct btrfs_block_group_cache
+                                              *block_group, u64 offset,
+                                              u64 bytes)
+{
+       struct btrfs_free_space *ret;
+
+       spin_lock(&block_group->lock);
+       ret = tree_search_offset(&block_group->free_space_offset, offset,
+                                bytes, 0);
+       if (!ret)
+               ret = tree_search_bytes(&block_group->free_space_bytes,
+                                       offset, bytes);
+
+       spin_unlock(&block_group->lock);
+
+       return ret;
+}
index 12c1c0530f3deac96f17c693add551e9c1e3b80f..65b4f864b0df5ab23024bdd61c337712d9bc78d8 100644 (file)
@@ -141,7 +141,7 @@ static int cow_file_range(struct inode *inode, u64 start, u64 end)
        while(num_bytes > 0) {
                cur_alloc_size = min(num_bytes, root->fs_info->max_extent);
                ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
-                                          root->sectorsize, 0, 0,
+                                          root->sectorsize, 0, alloc_hint,
                                           (u64)-1, &ins, 1);
                if (ret) {
                        WARN_ON(1);
@@ -558,7 +558,6 @@ static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
                                          trans->transid, inode->i_ino,
                                          ordered_extent->file_offset, &ins);
        BUG_ON(ret);
-
        mutex_lock(&BTRFS_I(inode)->extent_mutex);
 
        ret = btrfs_drop_extents(trans, root, inode,
index 1546fa6f4f7d025c6d363320debdc2771131f704..b9e5c2d82dde344a5cba2e3741920f3dc630b874 100644 (file)
@@ -64,8 +64,8 @@ static void lock_chunks(struct btrfs_root *root)
 
 static void unlock_chunks(struct btrfs_root *root)
 {
-       mutex_unlock(&root->fs_info->alloc_mutex);
        mutex_unlock(&root->fs_info->chunk_mutex);
+       mutex_unlock(&root->fs_info->alloc_mutex);
 }
 
 int btrfs_cleanup_fs_uuids(void)
@@ -1668,8 +1668,13 @@ again:
        else
                min_free = calc_size;
 
-       /* we add 1MB because we never use the first 1MB of the device */
-       min_free += 1024 * 1024;
+       /*
+        * we add 1MB because we never use the first 1MB of the device, unless
+        * we've looped, then we are likely allocating the maximum amount of
+        * space left already
+        */
+       if (!looped)
+               min_free += 1024 * 1024;
 
        /* build a private list of devices we will allocate from */
        while(index < num_stripes) {