diff mbox

[v4,6/9] Btrfs: implement the free space B-tree

Message ID 3319d371e22491b6901af33842b57db37b77c52c.1443583874.git.osandov@osandov.com (mailing list archive)
State Accepted
Headers show

Commit Message

Omar Sandoval Sept. 30, 2015, 3:50 a.m. UTC
From: Omar Sandoval <osandov@fb.com>

The free space cache has turned out to be a scalability bottleneck on
large, busy filesystems. When the cache for a lot of block groups needs
to be written out, we can get extremely long commit times; if this
happens in the critical section, things are especially bad because we
block new transactions from happening.

The main problem with the free space cache is that it has to be written
out in its entirety and is managed in an ad hoc fashion. Using a B-tree
to store free space fixes this: updates can be done as needed and we get
all of the benefits of using a B-tree: checksumming, RAID handling,
well-understood behavior.

With the free space tree, we get commit times that are about the same as
the no cache case with load times slower than the free space cache case
but still much faster than the no cache case. Free space is represented
with extents until it becomes more space-efficient to use bitmaps,
giving us similar space overhead to the free space cache.

The operations on the free space tree are: adding and removing free
space, handling the creation and deletion of block groups, and loading
the free space for a block group. We can also create the free space tree
by walking the extent tree and clear the free space tree.

Signed-off-by: Omar Sandoval <osandov@fb.com>
---
 fs/btrfs/Makefile          |    2 +-
 fs/btrfs/ctree.h           |   27 +-
 fs/btrfs/extent-tree.c     |    5 +-
 fs/btrfs/free-space-tree.c | 1584 ++++++++++++++++++++++++++++++++++++++++++++
 fs/btrfs/free-space-tree.h |   72 ++
 5 files changed, 1686 insertions(+), 4 deletions(-)
 create mode 100644 fs/btrfs/free-space-tree.c
 create mode 100644 fs/btrfs/free-space-tree.h

Comments

kernel test robot Sept. 30, 2015, 4:35 a.m. UTC | #1
Hi Omar,

[auto build test results on v4.3-rc3 -- if it's inappropriate base, please ignore]


coccinelle warnings: (new ones prefixed by >>)

>> fs/btrfs/free-space-tree.c:1255:1-4: WARNING: end returns can be simpified
   fs/btrfs/free-space-tree.c:1168:1-4: WARNING: end returns can be simpified

Please review and possibly fold the followup patch.

---
0-DAY kernel test infrastructure                Open Source Technology Center
https://lists.01.org/pipermail/kbuild-all                   Intel Corporation
--
To unsubscribe from this list: send the line "unsubscribe linux-btrfs" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
David Sterba Oct. 1, 2015, 12:54 p.m. UTC | #2
Hi,

a few minor comments, I haven't reviewed the changes in depth.

On Tue, Sep 29, 2015 at 08:50:35PM -0700, Omar Sandoval wrote:
> From: Omar Sandoval <osandov@fb.com>
> +static int btrfs_search_prev_slot(struct btrfs_trans_handle *trans,
> +				  struct btrfs_root *root,
> +				  struct btrfs_key *key, struct btrfs_path *p,

Please don't use single letter variables, eg. 'path' is commonly used
elsewere.

> +				  int ins_len, int cow)
> +{
> +	int ret;
> +
> +	ret = btrfs_search_slot(trans, root, key, p, ins_len, cow);
> +	if (ret < 0)
> +		return ret;
> +
> +	if (ret == 0) {
> +		ASSERT(0);
> +		return -EIO;
> +	}
> +
> +	if (p->slots[0] == 0) {
> +		ASSERT(0);
> +		return -EIO;
> +	}
> +	p->slots[0]--;
> +
> +	return 0;
> +}
> +
> +static unsigned long *alloc_bitmap(u32 bitmap_size)
> +{
> +	return __vmalloc(bitmap_size, GFP_NOFS | __GFP_HIGHMEM | __GFP_ZERO,

Why is GFP_HIGHMEM here?

> +			 PAGE_KERNEL);
> +}
> +
> +int convert_free_space_to_bitmaps(struct btrfs_trans_handle *trans,
> +				  struct btrfs_fs_info *fs_info,
> +				  struct btrfs_block_group_cache *block_group,
> +				  struct btrfs_path *path)
> +{
> +	struct btrfs_root *root = fs_info->free_space_root;
> +	struct btrfs_free_space_info *info;
> +	struct btrfs_key key, found_key;
> +	struct extent_buffer *leaf;
> +	unsigned long *bitmap;
> +	char *bitmap_cursor;
> +	u64 start, end;
> +	u64 bitmap_range, i;
> +	u32 bitmap_size, flags, expected_extent_count;
> +	u32 extent_count = 0;
> +	int done = 0, nr;
> +	int ret;
> +
> +	bitmap_size = free_space_bitmap_size(block_group->key.offset,
> +					     block_group->sectorsize);
> +	bitmap = alloc_bitmap(bitmap_size);
> +	if (!bitmap) {
> +		ret = -ENOMEM;
> +		goto out;
> +	}

Can we get rid of this potential allocation failure site? You free the
buffer in the same function, no recursion, so it should be ok to
preallocate the buffer at the mount time and just reuse here. Converting
to bitmap is not a frequent operation but IMO is critical enough to make
it more robust.

> +
> +	start = block_group->key.objectid;
> +	end = block_group->key.objectid + block_group->key.offset;
> +
> +	key.objectid = end - 1;
> +	key.type = (u8)-1;
> +	key.offset = (u64)-1;
> +
> +	while (!done) {
> +		ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
> +		if (ret)
> +			goto out;
> +
> +		leaf = path->nodes[0];
> +		nr = 0;
> +		path->slots[0]++;
> +		while (path->slots[0] > 0) {
> +			btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0] - 1);
> +
> +			if (found_key.type == BTRFS_FREE_SPACE_INFO_KEY) {
> +				ASSERT(found_key.objectid == block_group->key.objectid);
> +				ASSERT(found_key.offset == block_group->key.offset);
> +				done = 1;
> +				break;
> +			} else if (found_key.type == BTRFS_FREE_SPACE_EXTENT_KEY) {
> +				u64 first, last;
> +
> +				ASSERT(found_key.objectid >= start);
> +				ASSERT(found_key.objectid < end);
> +				ASSERT(found_key.objectid + found_key.offset <= end);
> +
> +				first = div_u64(found_key.objectid - start,
> +						block_group->sectorsize);
> +				last = div_u64(found_key.objectid + found_key.offset - start,
> +					       block_group->sectorsize);
> +				bitmap_set(bitmap, first, last - first);
> +
> +				extent_count++;
> +				nr++;
> +				path->slots[0]--;
> +			} else {
> +				ASSERT(0);
> +			}
> +		}
> +
> +		ret = btrfs_del_items(trans, root, path, path->slots[0], nr);
> +		if (ret)
> +			goto out;
> +		btrfs_release_path(path);
> +	}
> +
> +	info = search_free_space_info(trans, fs_info, block_group, path, 1);
> +	if (IS_ERR(info)) {
> +		ret = PTR_ERR(info);
> +		goto out;
> +	}
> +	leaf = path->nodes[0];
> +	flags = btrfs_free_space_flags(leaf, info);
> +	flags |= BTRFS_FREE_SPACE_USING_BITMAPS;
> +	btrfs_set_free_space_flags(leaf, info, flags);
> +	expected_extent_count = btrfs_free_space_extent_count(leaf, info);
> +	btrfs_mark_buffer_dirty(leaf);
> +	btrfs_release_path(path);
> +
> +	if (extent_count != expected_extent_count) {
> +		btrfs_err(fs_info, "incorrect extent count for %llu; counted %u, expected %u",
> +			  block_group->key.objectid, extent_count,
> +			  expected_extent_count);
> +		ASSERT(0);
> +		ret = -EIO;
> +		goto out;
> +	}
> +
> +	bitmap_cursor = (char *)bitmap;
> +	bitmap_range = block_group->sectorsize * BTRFS_FREE_SPACE_BITMAP_BITS;
> +	i = start;
> +	while (i < end) {
> +		unsigned long ptr;
> +		u64 extent_size;
> +		u32 data_size;
> +
> +		extent_size = min(end - i, bitmap_range);
> +		data_size = free_space_bitmap_size(extent_size,
> +						   block_group->sectorsize);
> +
> +		key.objectid = i;
> +		key.type = BTRFS_FREE_SPACE_BITMAP_KEY;
> +		key.offset = extent_size;
> +
> +		ret = btrfs_insert_empty_item(trans, root, path, &key,
> +					      data_size);
> +		if (ret)
> +			goto out;
> +
> +		leaf = path->nodes[0];
> +		ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
> +		write_extent_buffer(leaf, bitmap_cursor, ptr,
> +				    data_size);
> +		btrfs_mark_buffer_dirty(leaf);
> +		btrfs_release_path(path);
> +
> +		i += extent_size;
> +		bitmap_cursor += data_size;
> +	}
> +
> +	ret = 0;
> +out:
> +	vfree(bitmap);
> +	if (ret)
> +		btrfs_abort_transaction(trans, root, ret);

A general comment to the transaction aborts: the call should happen at
the location of the error and not in the common exit block. Otherwise we
lose the information what caused the failure.

This applies to the remaining functions as well, I won't point it out
specifically.
--
To unsubscribe from this list: send the line "unsubscribe linux-btrfs" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
kernel test robot Oct. 2, 2015, 2:23 p.m. UTC | #3
Hi Omar,

[auto build test results on v4.3-rc3 -- if it's inappropriate base, please ignore]

config: mips-ip27_defconfig (attached as .config)
reproduce:
        wget https://git.kernel.org/cgit/linux/kernel/git/wfg/lkp-tests.git/plain/sbin/make.cross -O ~/bin/make.cross
        chmod +x ~/bin/make.cross
        # save the attached .config to linux build tree
        make.cross ARCH=mips 

All warnings (new ones prefixed by >>):

   mips-linux-gnu-ld: fs/btrfs/super.o: warning: Inconsistent ISA between e_flags and .MIPS.abiflags
   mips-linux-gnu-ld: fs/btrfs/super.o: warning: Inconsistent ISA extensions between e_flags and .MIPS.abiflags
   mips-linux-gnu-ld: fs/btrfs/ctree.o: warning: Inconsistent ISA between e_flags and .MIPS.abiflags
   mips-linux-gnu-ld: fs/btrfs/ctree.o: warning: Inconsistent ISA extensions between e_flags and .MIPS.abiflags
   mips-linux-gnu-ld: fs/btrfs/extent-tree.o: warning: Inconsistent ISA between e_flags and .MIPS.abiflags
   mips-linux-gnu-ld: fs/btrfs/extent-tree.o: warning: Inconsistent ISA extensions between e_flags and .MIPS.abiflags
   mips-linux-gnu-ld: fs/btrfs/print-tree.o: warning: Inconsistent ISA between e_flags and .MIPS.abiflags
   mips-linux-gnu-ld: fs/btrfs/print-tree.o: warning: Inconsistent ISA extensions between e_flags and .MIPS.abiflags
   mips-linux-gnu-ld: fs/btrfs/root-tree.o: warning: Inconsistent ISA between e_flags and .MIPS.abiflags
   mips-linux-gnu-ld: fs/btrfs/root-tree.o: warning: Inconsistent ISA extensions between e_flags and .MIPS.abiflags
   mips-linux-gnu-ld: fs/btrfs/dir-item.o: warning: Inconsistent ISA between e_flags and .MIPS.abiflags
   mips-linux-gnu-ld: fs/btrfs/dir-item.o: warning: Inconsistent ISA extensions between e_flags and .MIPS.abiflags
   mips-linux-gnu-ld: fs/btrfs/file-item.o: warning: Inconsistent ISA between e_flags and .MIPS.abiflags
   mips-linux-gnu-ld: fs/btrfs/file-item.o: warning: Inconsistent ISA extensions between e_flags and .MIPS.abiflags
   mips-linux-gnu-ld: fs/btrfs/inode-item.o: warning: Inconsistent ISA between e_flags and .MIPS.abiflags
   mips-linux-gnu-ld: fs/btrfs/inode-item.o: warning: Inconsistent ISA extensions between e_flags and .MIPS.abiflags
   mips-linux-gnu-ld: fs/btrfs/inode-map.o: warning: Inconsistent ISA between e_flags and .MIPS.abiflags
   mips-linux-gnu-ld: fs/btrfs/inode-map.o: warning: Inconsistent ISA extensions between e_flags and .MIPS.abiflags
   mips-linux-gnu-ld: fs/btrfs/disk-io.o: warning: Inconsistent ISA between e_flags and .MIPS.abiflags
   mips-linux-gnu-ld: fs/btrfs/disk-io.o: warning: Inconsistent ISA extensions between e_flags and .MIPS.abiflags
   mips-linux-gnu-ld: fs/btrfs/transaction.o: warning: Inconsistent ISA between e_flags and .MIPS.abiflags
   mips-linux-gnu-ld: fs/btrfs/transaction.o: warning: Inconsistent ISA extensions between e_flags and .MIPS.abiflags
   mips-linux-gnu-ld: fs/btrfs/inode.o: warning: Inconsistent ISA between e_flags and .MIPS.abiflags
   mips-linux-gnu-ld: fs/btrfs/inode.o: warning: Inconsistent ISA extensions between e_flags and .MIPS.abiflags
   mips-linux-gnu-ld: fs/btrfs/file.o: warning: Inconsistent ISA between e_flags and .MIPS.abiflags
   mips-linux-gnu-ld: fs/btrfs/file.o: warning: Inconsistent ISA extensions between e_flags and .MIPS.abiflags
   mips-linux-gnu-ld: fs/btrfs/tree-defrag.o: warning: Inconsistent ISA between e_flags and .MIPS.abiflags
   mips-linux-gnu-ld: fs/btrfs/tree-defrag.o: warning: Inconsistent ISA extensions between e_flags and .MIPS.abiflags
   mips-linux-gnu-ld: fs/btrfs/extent_map.o: warning: Inconsistent ISA between e_flags and .MIPS.abiflags
   mips-linux-gnu-ld: fs/btrfs/extent_map.o: warning: Inconsistent ISA extensions between e_flags and .MIPS.abiflags
   mips-linux-gnu-ld: fs/btrfs/sysfs.o: warning: Inconsistent ISA between e_flags and .MIPS.abiflags
   mips-linux-gnu-ld: fs/btrfs/sysfs.o: warning: Inconsistent ISA extensions between e_flags and .MIPS.abiflags
   mips-linux-gnu-ld: fs/btrfs/struct-funcs.o: warning: Inconsistent ISA between e_flags and .MIPS.abiflags
   mips-linux-gnu-ld: fs/btrfs/struct-funcs.o: warning: Inconsistent ISA extensions between e_flags and .MIPS.abiflags
   mips-linux-gnu-ld: fs/btrfs/xattr.o: warning: Inconsistent ISA between e_flags and .MIPS.abiflags
   mips-linux-gnu-ld: fs/btrfs/xattr.o: warning: Inconsistent ISA extensions between e_flags and .MIPS.abiflags
   mips-linux-gnu-ld: fs/btrfs/ordered-data.o: warning: Inconsistent ISA between e_flags and .MIPS.abiflags
   mips-linux-gnu-ld: fs/btrfs/ordered-data.o: warning: Inconsistent ISA extensions between e_flags and .MIPS.abiflags
   mips-linux-gnu-ld: fs/btrfs/extent_io.o: warning: Inconsistent ISA between e_flags and .MIPS.abiflags
   mips-linux-gnu-ld: fs/btrfs/extent_io.o: warning: Inconsistent ISA extensions between e_flags and .MIPS.abiflags
   mips-linux-gnu-ld: fs/btrfs/volumes.o: warning: Inconsistent ISA between e_flags and .MIPS.abiflags
   mips-linux-gnu-ld: fs/btrfs/volumes.o: warning: Inconsistent ISA extensions between e_flags and .MIPS.abiflags
   mips-linux-gnu-ld: fs/btrfs/async-thread.o: warning: Inconsistent ISA between e_flags and .MIPS.abiflags
   mips-linux-gnu-ld: fs/btrfs/async-thread.o: warning: Inconsistent ISA extensions between e_flags and .MIPS.abiflags
   mips-linux-gnu-ld: fs/btrfs/ioctl.o: warning: Inconsistent ISA between e_flags and .MIPS.abiflags
   mips-linux-gnu-ld: fs/btrfs/ioctl.o: warning: Inconsistent ISA extensions between e_flags and .MIPS.abiflags
   mips-linux-gnu-ld: fs/btrfs/locking.o: warning: Inconsistent ISA between e_flags and .MIPS.abiflags
   mips-linux-gnu-ld: fs/btrfs/locking.o: warning: Inconsistent ISA extensions between e_flags and .MIPS.abiflags
   mips-linux-gnu-ld: fs/btrfs/orphan.o: warning: Inconsistent ISA between e_flags and .MIPS.abiflags
   mips-linux-gnu-ld: fs/btrfs/orphan.o: warning: Inconsistent ISA extensions between e_flags and .MIPS.abiflags
   mips-linux-gnu-ld: fs/btrfs/export.o: warning: Inconsistent ISA between e_flags and .MIPS.abiflags
   mips-linux-gnu-ld: fs/btrfs/export.o: warning: Inconsistent ISA extensions between e_flags and .MIPS.abiflags
   mips-linux-gnu-ld: fs/btrfs/tree-log.o: warning: Inconsistent ISA between e_flags and .MIPS.abiflags
   mips-linux-gnu-ld: fs/btrfs/tree-log.o: warning: Inconsistent ISA extensions between e_flags and .MIPS.abiflags
   mips-linux-gnu-ld: fs/btrfs/free-space-cache.o: warning: Inconsistent ISA between e_flags and .MIPS.abiflags
   mips-linux-gnu-ld: fs/btrfs/free-space-cache.o: warning: Inconsistent ISA extensions between e_flags and .MIPS.abiflags
   mips-linux-gnu-ld: fs/btrfs/zlib.o: warning: Inconsistent ISA between e_flags and .MIPS.abiflags
   mips-linux-gnu-ld: fs/btrfs/zlib.o: warning: Inconsistent ISA extensions between e_flags and .MIPS.abiflags
   mips-linux-gnu-ld: fs/btrfs/lzo.o: warning: Inconsistent ISA between e_flags and .MIPS.abiflags
   mips-linux-gnu-ld: fs/btrfs/lzo.o: warning: Inconsistent ISA extensions between e_flags and .MIPS.abiflags
   mips-linux-gnu-ld: fs/btrfs/compression.o: warning: Inconsistent ISA between e_flags and .MIPS.abiflags
   mips-linux-gnu-ld: fs/btrfs/compression.o: warning: Inconsistent ISA extensions between e_flags and .MIPS.abiflags
   mips-linux-gnu-ld: fs/btrfs/delayed-ref.o: warning: Inconsistent ISA between e_flags and .MIPS.abiflags
   mips-linux-gnu-ld: fs/btrfs/delayed-ref.o: warning: Inconsistent ISA extensions between e_flags and .MIPS.abiflags
   mips-linux-gnu-ld: fs/btrfs/relocation.o: warning: Inconsistent ISA between e_flags and .MIPS.abiflags
   mips-linux-gnu-ld: fs/btrfs/relocation.o: warning: Inconsistent ISA extensions between e_flags and .MIPS.abiflags
   mips-linux-gnu-ld: fs/btrfs/delayed-inode.o: warning: Inconsistent ISA between e_flags and .MIPS.abiflags
   mips-linux-gnu-ld: fs/btrfs/delayed-inode.o: warning: Inconsistent ISA extensions between e_flags and .MIPS.abiflags
   mips-linux-gnu-ld: fs/btrfs/scrub.o: warning: Inconsistent ISA between e_flags and .MIPS.abiflags
   mips-linux-gnu-ld: fs/btrfs/scrub.o: warning: Inconsistent ISA extensions between e_flags and .MIPS.abiflags
   mips-linux-gnu-ld: fs/btrfs/reada.o: warning: Inconsistent ISA between e_flags and .MIPS.abiflags
   mips-linux-gnu-ld: fs/btrfs/reada.o: warning: Inconsistent ISA extensions between e_flags and .MIPS.abiflags
   mips-linux-gnu-ld: fs/btrfs/backref.o: warning: Inconsistent ISA between e_flags and .MIPS.abiflags
   mips-linux-gnu-ld: fs/btrfs/backref.o: warning: Inconsistent ISA extensions between e_flags and .MIPS.abiflags
   mips-linux-gnu-ld: fs/btrfs/ulist.o: warning: Inconsistent ISA between e_flags and .MIPS.abiflags
   mips-linux-gnu-ld: fs/btrfs/ulist.o: warning: Inconsistent ISA extensions between e_flags and .MIPS.abiflags
   mips-linux-gnu-ld: fs/btrfs/qgroup.o: warning: Inconsistent ISA between e_flags and .MIPS.abiflags
   mips-linux-gnu-ld: fs/btrfs/qgroup.o: warning: Inconsistent ISA extensions between e_flags and .MIPS.abiflags
   mips-linux-gnu-ld: fs/btrfs/send.o: warning: Inconsistent ISA between e_flags and .MIPS.abiflags
   mips-linux-gnu-ld: fs/btrfs/send.o: warning: Inconsistent ISA extensions between e_flags and .MIPS.abiflags
   mips-linux-gnu-ld: fs/btrfs/dev-replace.o: warning: Inconsistent ISA between e_flags and .MIPS.abiflags
   mips-linux-gnu-ld: fs/btrfs/dev-replace.o: warning: Inconsistent ISA extensions between e_flags and .MIPS.abiflags
   mips-linux-gnu-ld: fs/btrfs/raid56.o: warning: Inconsistent ISA between e_flags and .MIPS.abiflags
   mips-linux-gnu-ld: fs/btrfs/raid56.o: warning: Inconsistent ISA extensions between e_flags and .MIPS.abiflags
   mips-linux-gnu-ld: fs/btrfs/uuid-tree.o: warning: Inconsistent ISA between e_flags and .MIPS.abiflags
   mips-linux-gnu-ld: fs/btrfs/uuid-tree.o: warning: Inconsistent ISA extensions between e_flags and .MIPS.abiflags
   mips-linux-gnu-ld: fs/btrfs/props.o: warning: Inconsistent ISA between e_flags and .MIPS.abiflags
   mips-linux-gnu-ld: fs/btrfs/props.o: warning: Inconsistent ISA extensions between e_flags and .MIPS.abiflags
   mips-linux-gnu-ld: fs/btrfs/hash.o: warning: Inconsistent ISA between e_flags and .MIPS.abiflags
   mips-linux-gnu-ld: fs/btrfs/hash.o: warning: Inconsistent ISA extensions between e_flags and .MIPS.abiflags
   mips-linux-gnu-ld: fs/btrfs/free-space-tree.o: warning: Inconsistent ISA between e_flags and .MIPS.abiflags
>> mips-linux-gnu-ld: fs/btrfs/free-space-tree.o: warning: Inconsistent ISA extensions between e_flags and .MIPS.abiflags
   mips-linux-gnu-ld: fs/btrfs/acl.o: warning: Inconsistent ISA between e_flags and .MIPS.abiflags
   mips-linux-gnu-ld: fs/btrfs/acl.o: warning: Inconsistent ISA extensions between e_flags and .MIPS.abiflags

---
0-DAY kernel test infrastructure                Open Source Technology Center
https://lists.01.org/pipermail/kbuild-all                   Intel Corporation
Chris Mason Dec. 29, 2015, 8:19 p.m. UTC | #4
On Tue, Sep 29, 2015 at 08:50:35PM -0700, Omar Sandoval wrote:
> From: Omar Sandoval <osandov@fb.com>
> 
> The free space cache has turned out to be a scalability bottleneck on
> large, busy filesystems. When the cache for a lot of block groups needs
> to be written out, we can get extremely long commit times; if this
> happens in the critical section, things are especially bad because we
> block new transactions from happening.
> 
> The main problem with the free space cache is that it has to be written
> out in its entirety and is managed in an ad hoc fashion. Using a B-tree
> to store free space fixes this: updates can be done as needed and we get
> all of the benefits of using a B-tree: checksumming, RAID handling,
> well-understood behavior.
> 
> With the free space tree, we get commit times that are about the same as
> the no cache case with load times slower than the free space cache case
> but still much faster than the no cache case. Free space is represented
> with extents until it becomes more space-efficient to use bitmaps,
> giving us similar space overhead to the free space cache.
> 
> The operations on the free space tree are: adding and removing free
> space, handling the creation and deletion of block groups, and loading
> the free space for a block group. We can also create the free space tree
> by walking the extent tree and clear the free space tree.
> 
> Signed-off-by: Omar Sandoval <osandov@fb.com>

> +int btrfs_create_free_space_tree(struct btrfs_fs_info *fs_info)
> +{
> +	struct btrfs_trans_handle *trans;
> +	struct btrfs_root *tree_root = fs_info->tree_root;
> +	struct btrfs_root *free_space_root;
> +	struct btrfs_block_group_cache *block_group;
> +	struct rb_node *node;
> +	int ret;
> +
> +	trans = btrfs_start_transaction(tree_root, 0);
> +	if (IS_ERR(trans))
> +		return PTR_ERR(trans);
> +
> +	free_space_root = btrfs_create_tree(trans, fs_info,
> +					    BTRFS_FREE_SPACE_TREE_OBJECTID);
> +	if (IS_ERR(free_space_root)) {
> +		ret = PTR_ERR(free_space_root);
> +		goto abort;
> +	}
> +	fs_info->free_space_root = free_space_root;
> +
> +	node = rb_first(&fs_info->block_group_cache_tree);
> +	while (node) {
> +		block_group = rb_entry(node, struct btrfs_block_group_cache,
> +				       cache_node);
> +		ret = populate_free_space_tree(trans, fs_info, block_group);
> +		if (ret)
> +			goto abort;
> +		node = rb_next(node);
> +	}
> +
> +	btrfs_set_fs_compat_ro(fs_info, FREE_SPACE_TREE);
> +
> +	ret = btrfs_commit_transaction(trans, tree_root);
> +	if (ret)
> +		return ret;
> +
> +	return 0;
> +
> +abort:
> +	btrfs_abort_transaction(trans, tree_root, ret);
> +	btrfs_end_transaction(trans, tree_root);
> +	return ret;
> +}
> +

Hi Omar,

The only problem I've hit testing this stuff is where we create the tree
on existing filesystems.  There are a few different problems here:

1) The populate code happens after resuming balance operations.  The
balancing code could be changing these block groups while we scan them.
I fixed this by moving the scan up earlier.

2) Delayed references may be run, which will also change the extent tree
as we're scanning it.

3) We might need to commit the transaction to reclaim space.

For now I'm ignoring #3 and adding a flag in fs_info that will make us
skip delayed references.  This really isn't a good long term solution,
we need to be able to do this on a per block group basis and make
forward progress without pinning the delayed refs in ram.

But for now, it'll do to get this into the tree for more testing.

-chris

--
To unsubscribe from this list: send the line "unsubscribe linux-btrfs" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Alex Lyakas April 22, 2016, 8:28 a.m. UTC | #5
Hi Omar, Chris,

I have reviewed the free-space-tree code. It is a very nice feature.

However, I have a basic understanding question.

Let's say we are running a delayed ref which inserts a new EXTENT_ITEM
into the extent tree, e.g., we are in alloc_reserved_file_extent. At
this point we call remove_from_free_space_tree(), which updates the
free-space-tree about the allocated space. But this requires to COW
the free-space-tree itself. So we allocate a new tree block for the
free-space tree, and insert a new delayed ref, which will update the
extent tree about the new tree block allocation. We also insert a
delayed ref to free the previous copy of the free-space-tree block.

At some point we run these new delayed refs, so we insert/remove
EXTENT_ITEMs from the extent tree, and this in turn requires us to
update the free-space-tree again. So we need again to COW
free-space-tree blocks, generating more delayed refs.

At which point this recursion stops?

Do we assume that at some point all needed free-space tree blocks have
been COW'ed already, and we do not COW a tree block more than once per
transaction (unless it was written to disk due to memory pressure)?

Thanks!
Alex.


On Tue, Dec 29, 2015 at 11:19 PM, Chris Mason <clm@fb.com> wrote:
> On Tue, Sep 29, 2015 at 08:50:35PM -0700, Omar Sandoval wrote:
>> From: Omar Sandoval <osandov@fb.com>
>>
>> The free space cache has turned out to be a scalability bottleneck on
>> large, busy filesystems. When the cache for a lot of block groups needs
>> to be written out, we can get extremely long commit times; if this
>> happens in the critical section, things are especially bad because we
>> block new transactions from happening.
>>
>> The main problem with the free space cache is that it has to be written
>> out in its entirety and is managed in an ad hoc fashion. Using a B-tree
>> to store free space fixes this: updates can be done as needed and we get
>> all of the benefits of using a B-tree: checksumming, RAID handling,
>> well-understood behavior.
>>
>> With the free space tree, we get commit times that are about the same as
>> the no cache case with load times slower than the free space cache case
>> but still much faster than the no cache case. Free space is represented
>> with extents until it becomes more space-efficient to use bitmaps,
>> giving us similar space overhead to the free space cache.
>>
>> The operations on the free space tree are: adding and removing free
>> space, handling the creation and deletion of block groups, and loading
>> the free space for a block group. We can also create the free space tree
>> by walking the extent tree and clear the free space tree.
>>
>> Signed-off-by: Omar Sandoval <osandov@fb.com>
>
>> +int btrfs_create_free_space_tree(struct btrfs_fs_info *fs_info)
>> +{
>> +     struct btrfs_trans_handle *trans;
>> +     struct btrfs_root *tree_root = fs_info->tree_root;
>> +     struct btrfs_root *free_space_root;
>> +     struct btrfs_block_group_cache *block_group;
>> +     struct rb_node *node;
>> +     int ret;
>> +
>> +     trans = btrfs_start_transaction(tree_root, 0);
>> +     if (IS_ERR(trans))
>> +             return PTR_ERR(trans);
>> +
>> +     free_space_root = btrfs_create_tree(trans, fs_info,
>> +                                         BTRFS_FREE_SPACE_TREE_OBJECTID);
>> +     if (IS_ERR(free_space_root)) {
>> +             ret = PTR_ERR(free_space_root);
>> +             goto abort;
>> +     }
>> +     fs_info->free_space_root = free_space_root;
>> +
>> +     node = rb_first(&fs_info->block_group_cache_tree);
>> +     while (node) {
>> +             block_group = rb_entry(node, struct btrfs_block_group_cache,
>> +                                    cache_node);
>> +             ret = populate_free_space_tree(trans, fs_info, block_group);
>> +             if (ret)
>> +                     goto abort;
>> +             node = rb_next(node);
>> +     }
>> +
>> +     btrfs_set_fs_compat_ro(fs_info, FREE_SPACE_TREE);
>> +
>> +     ret = btrfs_commit_transaction(trans, tree_root);
>> +     if (ret)
>> +             return ret;
>> +
>> +     return 0;
>> +
>> +abort:
>> +     btrfs_abort_transaction(trans, tree_root, ret);
>> +     btrfs_end_transaction(trans, tree_root);
>> +     return ret;
>> +}
>> +
>
> Hi Omar,
>
> The only problem I've hit testing this stuff is where we create the tree
> on existing filesystems.  There are a few different problems here:
>
> 1) The populate code happens after resuming balance operations.  The
> balancing code could be changing these block groups while we scan them.
> I fixed this by moving the scan up earlier.
>
> 2) Delayed references may be run, which will also change the extent tree
> as we're scanning it.
>
> 3) We might need to commit the transaction to reclaim space.
>
> For now I'm ignoring #3 and adding a flag in fs_info that will make us
> skip delayed references.  This really isn't a good long term solution,
> we need to be able to do this on a per block group basis and make
> forward progress without pinning the delayed refs in ram.
>
> But for now, it'll do to get this into the tree for more testing.
>
> -chris
>
> --
> To unsubscribe from this list: send the line "unsubscribe linux-btrfs" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
--
To unsubscribe from this list: send the line "unsubscribe linux-btrfs" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Omar Sandoval May 2, 2016, 9:55 p.m. UTC | #6
On Fri, Apr 22, 2016 at 11:28:24AM +0300, Alex Lyakas wrote:
> Hi Omar, Chris,
> 
> I have reviewed the free-space-tree code. It is a very nice feature.
> 
> However, I have a basic understanding question.
> 
> Let's say we are running a delayed ref which inserts a new EXTENT_ITEM
> into the extent tree, e.g., we are in alloc_reserved_file_extent. At
> this point we call remove_from_free_space_tree(), which updates the
> free-space-tree about the allocated space. But this requires to COW
> the free-space-tree itself. So we allocate a new tree block for the
> free-space tree, and insert a new delayed ref, which will update the
> extent tree about the new tree block allocation. We also insert a
> delayed ref to free the previous copy of the free-space-tree block.
> 
> At some point we run these new delayed refs, so we insert/remove
> EXTENT_ITEMs from the extent tree, and this in turn requires us to
> update the free-space-tree again. So we need again to COW
> free-space-tree blocks, generating more delayed refs.
> 
> At which point this recursion stops?
> 
> Do we assume that at some point all needed free-space tree blocks have
> been COW'ed already, and we do not COW a tree block more than once per
> transaction (unless it was written to disk due to memory pressure)?
> 
> Thanks!
> Alex.

Hi, Alex,

Your understanding is correct, we'll only COW a tree block once during a
transaction, so the recursion stops once all of the free space items we
need to add/remove don't require rebalancing the tree or allocating or
freeing a block. Same thing goes for extent items in the extent tree.
Since the extent tree and free space tree items are much smaller than a
tree block, this doesn't turn out to be a problem in practice.

Thanks for taking a look :)
diff mbox

Patch

diff --git a/fs/btrfs/Makefile b/fs/btrfs/Makefile
index 6d1d0b93b1aa..766169709146 100644
--- a/fs/btrfs/Makefile
+++ b/fs/btrfs/Makefile
@@ -9,7 +9,7 @@  btrfs-y += super.o ctree.o extent-tree.o print-tree.o root-tree.o dir-item.o \
 	   export.o tree-log.o free-space-cache.o zlib.o lzo.o \
 	   compression.o delayed-ref.o relocation.o delayed-inode.o scrub.o \
 	   reada.o backref.o ulist.o qgroup.o send.o dev-replace.o raid56.o \
-	   uuid-tree.o props.o hash.o
+	   uuid-tree.o props.o hash.o free-space-tree.o
 
 btrfs-$(CONFIG_BTRFS_FS_POSIX_ACL) += acl.o
 btrfs-$(CONFIG_BTRFS_FS_CHECK_INTEGRITY) += check-integrity.o
diff --git a/fs/btrfs/ctree.h b/fs/btrfs/ctree.h
index 0e40d323f4e9..0888c3e2cc7d 100644
--- a/fs/btrfs/ctree.h
+++ b/fs/btrfs/ctree.h
@@ -1302,8 +1302,20 @@  struct btrfs_block_group_cache {
 	u64 delalloc_bytes;
 	u64 bytes_super;
 	u64 flags;
-	u64 sectorsize;
 	u64 cache_generation;
+	u32 sectorsize;
+
+	/*
+	 * If the free space extent count exceeds this number, convert the block
+	 * group to bitmaps.
+	 */
+	u32 bitmap_high_thresh;
+
+	/*
+	 * If the free space extent count drops below this number, convert the
+	 * block group back to extents.
+	 */
+	u32 bitmap_low_thresh;
 
 	/*
 	 * It is just used for the delayed data space allocation because
@@ -1359,6 +1371,15 @@  struct btrfs_block_group_cache {
 	struct list_head io_list;
 
 	struct btrfs_io_ctl io_ctl;
+
+	/* Lock for free space tree operations. */
+	struct mutex free_space_lock;
+
+	/*
+	 * Does the block group need to be added to the free space tree?
+	 * Protected by free_space_lock.
+	 */
+	int needs_free_space;
 };
 
 /* delayed seq elem */
@@ -1410,6 +1431,7 @@  struct btrfs_fs_info {
 	struct btrfs_root *csum_root;
 	struct btrfs_root *quota_root;
 	struct btrfs_root *uuid_root;
+	struct btrfs_root *free_space_root;
 
 	/* the log root tree is a directory of all the other log roots */
 	struct btrfs_root *log_root_tree;
@@ -3555,6 +3577,9 @@  void btrfs_end_write_no_snapshoting(struct btrfs_root *root);
 void check_system_chunk(struct btrfs_trans_handle *trans,
 			struct btrfs_root *root,
 			const u64 type);
+u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
+		       struct btrfs_fs_info *info, u64 start, u64 end);
+
 /* ctree.c */
 int btrfs_bin_search(struct extent_buffer *eb, struct btrfs_key *key,
 		     int level, int *slot);
diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c
index 24f95299efe9..a3b84ee8a28d 100644
--- a/fs/btrfs/extent-tree.c
+++ b/fs/btrfs/extent-tree.c
@@ -337,8 +337,8 @@  static void put_caching_control(struct btrfs_caching_control *ctl)
  * 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 u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
-			      struct btrfs_fs_info *info, u64 start, u64 end)
+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;
@@ -9377,6 +9377,7 @@  btrfs_create_block_group_cache(struct btrfs_root *root, u64 start, u64 size)
 	INIT_LIST_HEAD(&cache->io_list);
 	btrfs_init_free_space_ctl(cache);
 	atomic_set(&cache->trimming, 0);
+	mutex_init(&cache->free_space_lock);
 
 	return cache;
 }
diff --git a/fs/btrfs/free-space-tree.c b/fs/btrfs/free-space-tree.c
new file mode 100644
index 000000000000..cbe36dd0d97b
--- /dev/null
+++ b/fs/btrfs/free-space-tree.c
@@ -0,0 +1,1584 @@ 
+/*
+ * Copyright (C) 2015 Facebook.  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/kernel.h>
+#include <linux/vmalloc.h>
+#include "ctree.h"
+#include "disk-io.h"
+#include "locking.h"
+#include "free-space-tree.h"
+#include "transaction.h"
+
+static int __add_block_group_free_space(struct btrfs_trans_handle *trans,
+					struct btrfs_fs_info *fs_info,
+					struct btrfs_block_group_cache *block_group,
+					struct btrfs_path *path);
+
+void set_free_space_tree_thresholds(struct btrfs_block_group_cache *cache)
+{
+	u32 bitmap_range;
+	size_t bitmap_size;
+	u64 num_bitmaps, total_bitmap_size;
+
+	/*
+	 * We convert to bitmaps when the disk space required for using extents
+	 * exceeds that required for using bitmaps.
+	 */
+	bitmap_range = cache->sectorsize * BTRFS_FREE_SPACE_BITMAP_BITS;
+	num_bitmaps = div_u64(cache->key.offset + bitmap_range - 1,
+			      bitmap_range);
+	bitmap_size = sizeof(struct btrfs_item) + BTRFS_FREE_SPACE_BITMAP_SIZE;
+	total_bitmap_size = num_bitmaps * bitmap_size;
+	cache->bitmap_high_thresh = div_u64(total_bitmap_size,
+					    sizeof(struct btrfs_item));
+
+	/*
+	 * We allow for a small buffer between the high threshold and low
+	 * threshold to avoid thrashing back and forth between the two formats.
+	 */
+	if (cache->bitmap_high_thresh > 100)
+		cache->bitmap_low_thresh = cache->bitmap_high_thresh - 100;
+	else
+		cache->bitmap_low_thresh = 0;
+}
+
+static int add_new_free_space_info(struct btrfs_trans_handle *trans,
+				   struct btrfs_fs_info *fs_info,
+				   struct btrfs_block_group_cache *block_group,
+				   struct btrfs_path *path)
+{
+	struct btrfs_root *root = fs_info->free_space_root;
+	struct btrfs_free_space_info *info;
+	struct btrfs_key key;
+	struct extent_buffer *leaf;
+	int ret;
+
+	key.objectid = block_group->key.objectid;
+	key.type = BTRFS_FREE_SPACE_INFO_KEY;
+	key.offset = block_group->key.offset;
+
+	ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(*info));
+	if (ret)
+		goto out;
+
+	leaf = path->nodes[0];
+	info = btrfs_item_ptr(leaf, path->slots[0],
+			      struct btrfs_free_space_info);
+	btrfs_set_free_space_extent_count(leaf, info, 0);
+	btrfs_set_free_space_flags(leaf, info, 0);
+	btrfs_mark_buffer_dirty(leaf);
+
+	ret = 0;
+out:
+	btrfs_release_path(path);
+	return ret;
+}
+
+struct btrfs_free_space_info *
+search_free_space_info(struct btrfs_trans_handle *trans,
+		       struct btrfs_fs_info *fs_info,
+		       struct btrfs_block_group_cache *block_group,
+		       struct btrfs_path *path, int cow)
+{
+	struct btrfs_root *root = fs_info->free_space_root;
+	struct btrfs_key key;
+	int ret;
+
+	key.objectid = block_group->key.objectid;
+	key.type = BTRFS_FREE_SPACE_INFO_KEY;
+	key.offset = block_group->key.offset;
+
+	ret = btrfs_search_slot(trans, root, &key, path, 0, cow);
+	if (ret < 0)
+		return ERR_PTR(ret);
+	if (ret != 0) {
+		btrfs_warn(fs_info, "missing free space info for %llu\n",
+			   block_group->key.objectid);
+		ASSERT(0);
+		return ERR_PTR(-ENOENT);
+	}
+
+	return btrfs_item_ptr(path->nodes[0], path->slots[0],
+			      struct btrfs_free_space_info);
+}
+
+/*
+ * btrfs_search_slot() but we're looking for the greatest key less than the
+ * passed key.
+ */
+static int btrfs_search_prev_slot(struct btrfs_trans_handle *trans,
+				  struct btrfs_root *root,
+				  struct btrfs_key *key, struct btrfs_path *p,
+				  int ins_len, int cow)
+{
+	int ret;
+
+	ret = btrfs_search_slot(trans, root, key, p, ins_len, cow);
+	if (ret < 0)
+		return ret;
+
+	if (ret == 0) {
+		ASSERT(0);
+		return -EIO;
+	}
+
+	if (p->slots[0] == 0) {
+		ASSERT(0);
+		return -EIO;
+	}
+	p->slots[0]--;
+
+	return 0;
+}
+
+static inline u32 free_space_bitmap_size(u64 size, u32 sectorsize)
+{
+	return DIV_ROUND_UP((u32)div_u64(size, sectorsize), BITS_PER_BYTE);
+}
+
+static unsigned long *alloc_bitmap(u32 bitmap_size)
+{
+	return __vmalloc(bitmap_size, GFP_NOFS | __GFP_HIGHMEM | __GFP_ZERO,
+			 PAGE_KERNEL);
+}
+
+int convert_free_space_to_bitmaps(struct btrfs_trans_handle *trans,
+				  struct btrfs_fs_info *fs_info,
+				  struct btrfs_block_group_cache *block_group,
+				  struct btrfs_path *path)
+{
+	struct btrfs_root *root = fs_info->free_space_root;
+	struct btrfs_free_space_info *info;
+	struct btrfs_key key, found_key;
+	struct extent_buffer *leaf;
+	unsigned long *bitmap;
+	char *bitmap_cursor;
+	u64 start, end;
+	u64 bitmap_range, i;
+	u32 bitmap_size, flags, expected_extent_count;
+	u32 extent_count = 0;
+	int done = 0, nr;
+	int ret;
+
+	bitmap_size = free_space_bitmap_size(block_group->key.offset,
+					     block_group->sectorsize);
+	bitmap = alloc_bitmap(bitmap_size);
+	if (!bitmap) {
+		ret = -ENOMEM;
+		goto out;
+	}
+
+	start = block_group->key.objectid;
+	end = block_group->key.objectid + block_group->key.offset;
+
+	key.objectid = end - 1;
+	key.type = (u8)-1;
+	key.offset = (u64)-1;
+
+	while (!done) {
+		ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
+		if (ret)
+			goto out;
+
+		leaf = path->nodes[0];
+		nr = 0;
+		path->slots[0]++;
+		while (path->slots[0] > 0) {
+			btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0] - 1);
+
+			if (found_key.type == BTRFS_FREE_SPACE_INFO_KEY) {
+				ASSERT(found_key.objectid == block_group->key.objectid);
+				ASSERT(found_key.offset == block_group->key.offset);
+				done = 1;
+				break;
+			} else if (found_key.type == BTRFS_FREE_SPACE_EXTENT_KEY) {
+				u64 first, last;
+
+				ASSERT(found_key.objectid >= start);
+				ASSERT(found_key.objectid < end);
+				ASSERT(found_key.objectid + found_key.offset <= end);
+
+				first = div_u64(found_key.objectid - start,
+						block_group->sectorsize);
+				last = div_u64(found_key.objectid + found_key.offset - start,
+					       block_group->sectorsize);
+				bitmap_set(bitmap, first, last - first);
+
+				extent_count++;
+				nr++;
+				path->slots[0]--;
+			} else {
+				ASSERT(0);
+			}
+		}
+
+		ret = btrfs_del_items(trans, root, path, path->slots[0], nr);
+		if (ret)
+			goto out;
+		btrfs_release_path(path);
+	}
+
+	info = search_free_space_info(trans, fs_info, block_group, path, 1);
+	if (IS_ERR(info)) {
+		ret = PTR_ERR(info);
+		goto out;
+	}
+	leaf = path->nodes[0];
+	flags = btrfs_free_space_flags(leaf, info);
+	flags |= BTRFS_FREE_SPACE_USING_BITMAPS;
+	btrfs_set_free_space_flags(leaf, info, flags);
+	expected_extent_count = btrfs_free_space_extent_count(leaf, info);
+	btrfs_mark_buffer_dirty(leaf);
+	btrfs_release_path(path);
+
+	if (extent_count != expected_extent_count) {
+		btrfs_err(fs_info, "incorrect extent count for %llu; counted %u, expected %u",
+			  block_group->key.objectid, extent_count,
+			  expected_extent_count);
+		ASSERT(0);
+		ret = -EIO;
+		goto out;
+	}
+
+	bitmap_cursor = (char *)bitmap;
+	bitmap_range = block_group->sectorsize * BTRFS_FREE_SPACE_BITMAP_BITS;
+	i = start;
+	while (i < end) {
+		unsigned long ptr;
+		u64 extent_size;
+		u32 data_size;
+
+		extent_size = min(end - i, bitmap_range);
+		data_size = free_space_bitmap_size(extent_size,
+						   block_group->sectorsize);
+
+		key.objectid = i;
+		key.type = BTRFS_FREE_SPACE_BITMAP_KEY;
+		key.offset = extent_size;
+
+		ret = btrfs_insert_empty_item(trans, root, path, &key,
+					      data_size);
+		if (ret)
+			goto out;
+
+		leaf = path->nodes[0];
+		ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
+		write_extent_buffer(leaf, bitmap_cursor, ptr,
+				    data_size);
+		btrfs_mark_buffer_dirty(leaf);
+		btrfs_release_path(path);
+
+		i += extent_size;
+		bitmap_cursor += data_size;
+	}
+
+	ret = 0;
+out:
+	vfree(bitmap);
+	if (ret)
+		btrfs_abort_transaction(trans, root, ret);
+	return ret;
+}
+
+int convert_free_space_to_extents(struct btrfs_trans_handle *trans,
+				  struct btrfs_fs_info *fs_info,
+				  struct btrfs_block_group_cache *block_group,
+				  struct btrfs_path *path)
+{
+	struct btrfs_root *root = fs_info->free_space_root;
+	struct btrfs_free_space_info *info;
+	struct btrfs_key key, found_key;
+	struct extent_buffer *leaf;
+	unsigned long *bitmap;
+	u64 start, end;
+	/* Initialize to silence GCC. */
+	u64 extent_start = 0;
+	u64 offset;
+	u32 bitmap_size, flags, expected_extent_count;
+	int prev_bit = 0, bit, bitnr;
+	u32 extent_count = 0;
+	int done = 0, nr;
+	int ret;
+
+	bitmap_size = free_space_bitmap_size(block_group->key.offset,
+					     block_group->sectorsize);
+	bitmap = alloc_bitmap(bitmap_size);
+	if (!bitmap) {
+		ret = -ENOMEM;
+		goto out;
+	}
+
+	start = block_group->key.objectid;
+	end = block_group->key.objectid + block_group->key.offset;
+
+	key.objectid = end - 1;
+	key.type = (u8)-1;
+	key.offset = (u64)-1;
+
+	while (!done) {
+		ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
+		if (ret)
+			goto out;
+
+		leaf = path->nodes[0];
+		nr = 0;
+		path->slots[0]++;
+		while (path->slots[0] > 0) {
+			btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0] - 1);
+
+			if (found_key.type == BTRFS_FREE_SPACE_INFO_KEY) {
+				ASSERT(found_key.objectid == block_group->key.objectid);
+				ASSERT(found_key.offset == block_group->key.offset);
+				done = 1;
+				break;
+			} else if (found_key.type == BTRFS_FREE_SPACE_BITMAP_KEY) {
+				unsigned long ptr;
+				char *bitmap_cursor;
+				u32 bitmap_pos, data_size;
+
+				ASSERT(found_key.objectid >= start);
+				ASSERT(found_key.objectid < end);
+				ASSERT(found_key.objectid + found_key.offset <= end);
+
+				bitmap_pos = div_u64(found_key.objectid - start,
+						     block_group->sectorsize *
+						     BITS_PER_BYTE);
+				bitmap_cursor = ((char *)bitmap) + bitmap_pos;
+				data_size = free_space_bitmap_size(found_key.offset,
+								   block_group->sectorsize);
+
+				ptr = btrfs_item_ptr_offset(leaf, path->slots[0] - 1);
+				read_extent_buffer(leaf, bitmap_cursor, ptr,
+						   data_size);
+
+				nr++;
+				path->slots[0]--;
+			} else {
+				ASSERT(0);
+			}
+		}
+
+		ret = btrfs_del_items(trans, root, path, path->slots[0], nr);
+		if (ret)
+			goto out;
+		btrfs_release_path(path);
+	}
+
+	info = search_free_space_info(trans, fs_info, block_group, path, 1);
+	if (IS_ERR(info)) {
+		ret = PTR_ERR(info);
+		goto out;
+	}
+	leaf = path->nodes[0];
+	flags = btrfs_free_space_flags(leaf, info);
+	flags &= ~BTRFS_FREE_SPACE_USING_BITMAPS;
+	btrfs_set_free_space_flags(leaf, info, flags);
+	expected_extent_count = btrfs_free_space_extent_count(leaf, info);
+	btrfs_mark_buffer_dirty(leaf);
+	btrfs_release_path(path);
+
+	offset = start;
+	bitnr = 0;
+	while (offset < end) {
+		bit = !!test_bit(bitnr, bitmap);
+		if (prev_bit == 0 && bit == 1) {
+			extent_start = offset;
+		} else if (prev_bit == 1 && bit == 0) {
+			key.objectid = extent_start;
+			key.type = BTRFS_FREE_SPACE_EXTENT_KEY;
+			key.offset = offset - extent_start;
+
+			ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
+			if (ret)
+				goto out;
+			btrfs_release_path(path);
+
+			extent_count++;
+		}
+		prev_bit = bit;
+		offset += block_group->sectorsize;
+		bitnr++;
+	}
+	if (prev_bit == 1) {
+		key.objectid = extent_start;
+		key.type = BTRFS_FREE_SPACE_EXTENT_KEY;
+		key.offset = end - extent_start;
+
+		ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
+		if (ret)
+			goto out;
+		btrfs_release_path(path);
+
+		extent_count++;
+	}
+
+	if (extent_count != expected_extent_count) {
+		btrfs_err(fs_info, "incorrect extent count for %llu; counted %u, expected %u",
+			  block_group->key.objectid, extent_count,
+			  expected_extent_count);
+		ASSERT(0);
+		ret = -EIO;
+		goto out;
+	}
+
+	ret = 0;
+out:
+	vfree(bitmap);
+	if (ret)
+		btrfs_abort_transaction(trans, root, ret);
+	return ret;
+}
+
+static int update_free_space_extent_count(struct btrfs_trans_handle *trans,
+					  struct btrfs_fs_info *fs_info,
+					  struct btrfs_block_group_cache *block_group,
+					  struct btrfs_path *path,
+					  int new_extents)
+{
+	struct btrfs_free_space_info *info;
+	u32 flags;
+	u32 extent_count;
+	int ret = 0;
+
+	if (new_extents == 0)
+		return 0;
+
+	info = search_free_space_info(trans, fs_info, block_group, path, 1);
+	if (IS_ERR(info)) {
+		ret = PTR_ERR(info);
+		goto out;
+	}
+	flags = btrfs_free_space_flags(path->nodes[0], info);
+	extent_count = btrfs_free_space_extent_count(path->nodes[0], info);
+
+	extent_count += new_extents;
+	btrfs_set_free_space_extent_count(path->nodes[0], info, extent_count);
+	btrfs_mark_buffer_dirty(path->nodes[0]);
+	btrfs_release_path(path);
+
+	if (!(flags & BTRFS_FREE_SPACE_USING_BITMAPS) &&
+	    extent_count > block_group->bitmap_high_thresh) {
+		ret = convert_free_space_to_bitmaps(trans, fs_info, block_group,
+						    path);
+	} else if ((flags & BTRFS_FREE_SPACE_USING_BITMAPS) &&
+		   extent_count < block_group->bitmap_low_thresh) {
+		ret = convert_free_space_to_extents(trans, fs_info, block_group,
+						    path);
+	}
+
+out:
+	return ret;
+}
+
+int free_space_test_bit(struct btrfs_block_group_cache *block_group,
+			struct btrfs_path *path, u64 offset)
+{
+	struct extent_buffer *leaf;
+	struct btrfs_key key;
+	u64 found_start, found_end;
+	unsigned long ptr, i;
+
+	leaf = path->nodes[0];
+	btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
+	ASSERT(key.type == BTRFS_FREE_SPACE_BITMAP_KEY);
+
+	found_start = key.objectid;
+	found_end = key.objectid + key.offset;
+	ASSERT(offset >= found_start && offset < found_end);
+
+	ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
+	i = div_u64(offset - found_start, block_group->sectorsize);
+	return !!extent_buffer_test_bit(leaf, ptr, i);
+}
+
+static void free_space_set_bits(struct btrfs_block_group_cache *block_group,
+				struct btrfs_path *path, u64 *start, u64 *size,
+				int bit)
+{
+	struct extent_buffer *leaf;
+	struct btrfs_key key;
+	u64 end = *start + *size;
+	u64 found_start, found_end;
+	unsigned long ptr, first, last;
+
+	leaf = path->nodes[0];
+	btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
+	ASSERT(key.type == BTRFS_FREE_SPACE_BITMAP_KEY);
+
+	found_start = key.objectid;
+	found_end = key.objectid + key.offset;
+	ASSERT(*start >= found_start && *start < found_end);
+	ASSERT(end > found_start);
+
+	if (end > found_end)
+		end = found_end;
+
+	ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
+	first = div_u64(*start - found_start, block_group->sectorsize);
+	last = div_u64(end - found_start, block_group->sectorsize);
+	if (bit)
+		extent_buffer_bitmap_set(leaf, ptr, first, last - first);
+	else
+		extent_buffer_bitmap_clear(leaf, ptr, first, last - first);
+	btrfs_mark_buffer_dirty(leaf);
+
+	*size -= end - *start;
+	*start = end;
+}
+
+/*
+ * We can't use btrfs_next_item() in modify_free_space_bitmap() because
+ * btrfs_next_leaf() doesn't get the path for writing. We can forgo the fancy
+ * tree walking in btrfs_next_leaf() anyways because we know exactly what we're
+ * looking for.
+ */
+static int free_space_next_bitmap(struct btrfs_trans_handle *trans,
+				  struct btrfs_root *root, struct btrfs_path *p)
+{
+	struct btrfs_key key;
+
+	if (p->slots[0] + 1 < btrfs_header_nritems(p->nodes[0])) {
+		p->slots[0]++;
+		return 0;
+	}
+
+	btrfs_item_key_to_cpu(p->nodes[0], &key, p->slots[0]);
+	btrfs_release_path(p);
+
+	key.objectid += key.offset;
+	key.type = (u8)-1;
+	key.offset = (u64)-1;
+
+	return btrfs_search_prev_slot(trans, root, &key, p, 0, 1);
+}
+
+/*
+ * If remove is 1, then we are removing free space, thus clearing bits in the
+ * bitmap. If remove is 0, then we are adding free space, thus setting bits in
+ * the bitmap.
+ */
+static int modify_free_space_bitmap(struct btrfs_trans_handle *trans,
+				    struct btrfs_fs_info *fs_info,
+				    struct btrfs_block_group_cache *block_group,
+				    struct btrfs_path *path,
+				    u64 start, u64 size, int remove)
+{
+	struct btrfs_root *root = fs_info->free_space_root;
+	struct btrfs_key key;
+	u64 end = start + size;
+	u64 cur_start, cur_size;
+	int prev_bit, next_bit;
+	int new_extents;
+	int ret;
+
+	/*
+	 * Read the bit for the block immediately before the extent of space if
+	 * that block is within the block group.
+	 */
+	if (start > block_group->key.objectid) {
+		u64 prev_block = start - block_group->sectorsize;
+
+		key.objectid = prev_block;
+		key.type = (u8)-1;
+		key.offset = (u64)-1;
+
+		ret = btrfs_search_prev_slot(trans, root, &key, path, 0, 1);
+		if (ret)
+			goto out;
+
+		prev_bit = free_space_test_bit(block_group, path, prev_block);
+
+		/* The previous block may have been in the previous bitmap. */
+		btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
+		if (start >= key.objectid + key.offset) {
+			ret = free_space_next_bitmap(trans, root, path);
+			if (ret)
+				goto out;
+		}
+	} else {
+		key.objectid = start;
+		key.type = (u8)-1;
+		key.offset = (u64)-1;
+
+		ret = btrfs_search_prev_slot(trans, root, &key, path, 0, 1);
+		if (ret)
+			goto out;
+
+		prev_bit = -1;
+	}
+
+	/*
+	 * Iterate over all of the bitmaps overlapped by the extent of space,
+	 * clearing/setting bits as required.
+	 */
+	cur_start = start;
+	cur_size = size;
+	while (1) {
+		free_space_set_bits(block_group, path, &cur_start, &cur_size,
+				    !remove);
+		if (cur_size == 0)
+			break;
+		ret = free_space_next_bitmap(trans, root, path);
+		if (ret)
+			goto out;
+	}
+
+	/*
+	 * Read the bit for the block immediately after the extent of space if
+	 * that block is within the block group.
+	 */
+	if (end < block_group->key.objectid + block_group->key.offset) {
+		/* The next block may be in the next bitmap. */
+		btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
+		if (end >= key.objectid + key.offset) {
+			ret = free_space_next_bitmap(trans, root, path);
+			if (ret)
+				goto out;
+		}
+
+		next_bit = free_space_test_bit(block_group, path, end);
+	} else {
+		next_bit = -1;
+	}
+
+	if (remove) {
+		new_extents = -1;
+		if (prev_bit == 1) {
+			/* Leftover on the left. */
+			new_extents++;
+		}
+		if (next_bit == 1) {
+			/* Leftover on the right. */
+			new_extents++;
+		}
+	} else {
+		new_extents = 1;
+		if (prev_bit == 1) {
+			/* Merging with neighbor on the left. */
+			new_extents--;
+		}
+		if (next_bit == 1) {
+			/* Merging with neighbor on the right. */
+			new_extents--;
+		}
+	}
+
+	btrfs_release_path(path);
+	ret = update_free_space_extent_count(trans, fs_info, block_group, path,
+					     new_extents);
+
+out:
+	return ret;
+}
+
+static int remove_free_space_extent(struct btrfs_trans_handle *trans,
+				    struct btrfs_fs_info *fs_info,
+				    struct btrfs_block_group_cache *block_group,
+				    struct btrfs_path *path,
+				    u64 start, u64 size)
+{
+	struct btrfs_root *root = fs_info->free_space_root;
+	struct btrfs_key key;
+	u64 found_start, found_end;
+	u64 end = start + size;
+	int new_extents = -1;
+	int ret;
+
+	key.objectid = start;
+	key.type = (u8)-1;
+	key.offset = (u64)-1;
+
+	ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
+	if (ret)
+		goto out;
+
+	btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
+
+	ASSERT(key.type == BTRFS_FREE_SPACE_EXTENT_KEY);
+
+	found_start = key.objectid;
+	found_end = key.objectid + key.offset;
+	ASSERT(start >= found_start && end <= found_end);
+
+	/*
+	 * Okay, now that we've found the free space extent which contains the
+	 * free space that we are removing, there are four cases:
+	 *
+	 * 1. We're using the whole extent: delete the key we found and
+	 * decrement the free space extent count.
+	 * 2. We are using part of the extent starting at the beginning: delete
+	 * the key we found and insert a new key representing the leftover at
+	 * the end. There is no net change in the number of extents.
+	 * 3. We are using part of the extent ending at the end: delete the key
+	 * we found and insert a new key representing the leftover at the
+	 * beginning. There is no net change in the number of extents.
+	 * 4. We are using part of the extent in the middle: delete the key we
+	 * found and insert two new keys representing the leftovers on each
+	 * side. Where we used to have one extent, we now have two, so increment
+	 * the extent count. We may need to convert the block group to bitmaps
+	 * as a result.
+	 */
+
+	/* Delete the existing key (cases 1-4). */
+	ret = btrfs_del_item(trans, root, path);
+	if (ret)
+		goto out;
+
+	/* Add a key for leftovers at the beginning (cases 3 and 4). */
+	if (start > found_start) {
+		key.objectid = found_start;
+		key.type = BTRFS_FREE_SPACE_EXTENT_KEY;
+		key.offset = start - found_start;
+
+		btrfs_release_path(path);
+		ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
+		if (ret)
+			goto out;
+		new_extents++;
+	}
+
+	/* Add a key for leftovers at the end (cases 2 and 4). */
+	if (end < found_end) {
+		key.objectid = end;
+		key.type = BTRFS_FREE_SPACE_EXTENT_KEY;
+		key.offset = found_end - end;
+
+		btrfs_release_path(path);
+		ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
+		if (ret)
+			goto out;
+		new_extents++;
+	}
+
+	btrfs_release_path(path);
+	ret = update_free_space_extent_count(trans, fs_info, block_group, path,
+					     new_extents);
+
+out:
+	return ret;
+}
+
+int __remove_from_free_space_tree(struct btrfs_trans_handle *trans,
+				  struct btrfs_fs_info *fs_info,
+				  struct btrfs_block_group_cache *block_group,
+				  struct btrfs_path *path, u64 start, u64 size)
+{
+	struct btrfs_free_space_info *info;
+	u32 flags;
+	int ret;
+
+	if (block_group->needs_free_space) {
+		ret = __add_block_group_free_space(trans, fs_info, block_group,
+						   path);
+		if (ret)
+			return ret;
+	}
+
+	info = search_free_space_info(NULL, fs_info, block_group, path, 0);
+	if (IS_ERR(info))
+		return PTR_ERR(info);
+	flags = btrfs_free_space_flags(path->nodes[0], info);
+	btrfs_release_path(path);
+
+	if (flags & BTRFS_FREE_SPACE_USING_BITMAPS) {
+		return modify_free_space_bitmap(trans, fs_info, block_group,
+						path, start, size, 1);
+	} else {
+		return remove_free_space_extent(trans, fs_info, block_group,
+						path, start, size);
+	}
+}
+
+int remove_from_free_space_tree(struct btrfs_trans_handle *trans,
+				struct btrfs_fs_info *fs_info,
+				u64 start, u64 size)
+{
+	struct btrfs_block_group_cache *block_group;
+	struct btrfs_path *path;
+	int ret;
+
+	if (!btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE))
+		return 0;
+
+	path = btrfs_alloc_path();
+	if (!path) {
+		ret = -ENOMEM;
+		goto out;
+	}
+
+	block_group = btrfs_lookup_block_group(fs_info, start);
+	if (!block_group) {
+		ASSERT(0);
+		ret = -ENOENT;
+		goto out;
+	}
+
+	mutex_lock(&block_group->free_space_lock);
+	ret = __remove_from_free_space_tree(trans, fs_info, block_group, path,
+					    start, size);
+	mutex_unlock(&block_group->free_space_lock);
+
+	btrfs_put_block_group(block_group);
+out:
+	btrfs_free_path(path);
+	if (ret)
+		btrfs_abort_transaction(trans, fs_info->free_space_root, ret);
+	return ret;
+}
+
+static int add_free_space_extent(struct btrfs_trans_handle *trans,
+				 struct btrfs_fs_info *fs_info,
+				 struct btrfs_block_group_cache *block_group,
+				 struct btrfs_path *path,
+				 u64 start, u64 size)
+{
+	struct btrfs_root *root = fs_info->free_space_root;
+	struct btrfs_key key, new_key;
+	u64 found_start, found_end;
+	u64 end = start + size;
+	int new_extents = 1;
+	int ret;
+
+	/*
+	 * We are adding a new extent of free space, but we need to merge
+	 * extents. There are four cases here:
+	 *
+	 * 1. The new extent does not have any immediate neighbors to merge
+	 * with: add the new key and increment the free space extent count. We
+	 * may need to convert the block group to bitmaps as a result.
+	 * 2. The new extent has an immediate neighbor before it: remove the
+	 * previous key and insert a new key combining both of them. There is no
+	 * net change in the number of extents.
+	 * 3. The new extent has an immediate neighbor after it: remove the next
+	 * key and insert a new key combining both of them. There is no net
+	 * change in the number of extents.
+	 * 4. The new extent has immediate neighbors on both sides: remove both
+	 * of the keys and insert a new key combining all of them. Where we used
+	 * to have two extents, we now have one, so decrement the extent count.
+	 */
+
+	new_key.objectid = start;
+	new_key.type = BTRFS_FREE_SPACE_EXTENT_KEY;
+	new_key.offset = size;
+
+	/* Search for a neighbor on the left. */
+	if (start == block_group->key.objectid)
+		goto right;
+	key.objectid = start - 1;
+	key.type = (u8)-1;
+	key.offset = (u64)-1;
+
+	ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
+	if (ret)
+		goto out;
+
+	btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
+
+	if (key.type != BTRFS_FREE_SPACE_EXTENT_KEY) {
+		ASSERT(key.type == BTRFS_FREE_SPACE_INFO_KEY);
+		btrfs_release_path(path);
+		goto right;
+	}
+
+	found_start = key.objectid;
+	found_end = key.objectid + key.offset;
+	ASSERT(found_start >= block_group->key.objectid &&
+	       found_end > block_group->key.objectid);
+	ASSERT(found_start < start && found_end <= start);
+
+	/*
+	 * Delete the neighbor on the left and absorb it into the new key (cases
+	 * 2 and 4).
+	 */
+	if (found_end == start) {
+		ret = btrfs_del_item(trans, root, path);
+		if (ret)
+			goto out;
+		new_key.objectid = found_start;
+		new_key.offset += key.offset;
+		new_extents--;
+	}
+	btrfs_release_path(path);
+
+right:
+	/* Search for a neighbor on the right. */
+	if (end == block_group->key.objectid + block_group->key.offset)
+		goto insert;
+	key.objectid = end;
+	key.type = (u8)-1;
+	key.offset = (u64)-1;
+
+	ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
+	if (ret)
+		goto out;
+
+	btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
+
+	if (key.type != BTRFS_FREE_SPACE_EXTENT_KEY) {
+		ASSERT(key.type == BTRFS_FREE_SPACE_INFO_KEY);
+		btrfs_release_path(path);
+		goto insert;
+	}
+
+	found_start = key.objectid;
+	found_end = key.objectid + key.offset;
+	ASSERT(found_start >= block_group->key.objectid &&
+	       found_end > block_group->key.objectid);
+	ASSERT((found_start < start && found_end <= start) ||
+	       (found_start >= end && found_end > end));
+
+	/*
+	 * Delete the neighbor on the right and absorb it into the new key
+	 * (cases 3 and 4).
+	 */
+	if (found_start == end) {
+		ret = btrfs_del_item(trans, root, path);
+		if (ret)
+			goto out;
+		new_key.offset += key.offset;
+		new_extents--;
+	}
+	btrfs_release_path(path);
+
+insert:
+	/* Insert the new key (cases 1-4). */
+	ret = btrfs_insert_empty_item(trans, root, path, &new_key, 0);
+	if (ret)
+		goto out;
+
+	btrfs_release_path(path);
+	ret = update_free_space_extent_count(trans, fs_info, block_group, path,
+					     new_extents);
+
+out:
+	return ret;
+}
+
+int __add_to_free_space_tree(struct btrfs_trans_handle *trans,
+			     struct btrfs_fs_info *fs_info,
+			     struct btrfs_block_group_cache *block_group,
+			     struct btrfs_path *path, u64 start, u64 size)
+{
+	struct btrfs_free_space_info *info;
+	u32 flags;
+	int ret;
+
+	if (block_group->needs_free_space) {
+		ret = __add_block_group_free_space(trans, fs_info, block_group,
+						   path);
+		if (ret)
+			return ret;
+	}
+
+	info = search_free_space_info(NULL, fs_info, block_group, path, 0);
+	if (IS_ERR(info))
+		return PTR_ERR(info);
+	flags = btrfs_free_space_flags(path->nodes[0], info);
+	btrfs_release_path(path);
+
+	if (flags & BTRFS_FREE_SPACE_USING_BITMAPS) {
+		return modify_free_space_bitmap(trans, fs_info, block_group,
+						path, start, size, 0);
+	} else {
+		return add_free_space_extent(trans, fs_info, block_group, path,
+					     start, size);
+	}
+}
+
+int add_to_free_space_tree(struct btrfs_trans_handle *trans,
+			   struct btrfs_fs_info *fs_info,
+			   u64 start, u64 size)
+{
+	struct btrfs_block_group_cache *block_group;
+	struct btrfs_path *path;
+	int ret;
+
+	if (!btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE))
+		return 0;
+
+	path = btrfs_alloc_path();
+	if (!path) {
+		ret = -ENOMEM;
+		goto out;
+	}
+
+	block_group = btrfs_lookup_block_group(fs_info, start);
+	if (!block_group) {
+		ASSERT(0);
+		ret = -ENOENT;
+		goto out;
+	}
+
+	mutex_lock(&block_group->free_space_lock);
+	ret = __add_to_free_space_tree(trans, fs_info, block_group, path, start,
+				       size);
+	mutex_unlock(&block_group->free_space_lock);
+
+	btrfs_put_block_group(block_group);
+out:
+	btrfs_free_path(path);
+	if (ret)
+		btrfs_abort_transaction(trans, fs_info->free_space_root, ret);
+	return ret;
+}
+
+/*
+ * Populate the free space tree by walking the extent tree. Operations on the
+ * extent tree that happen as a result of writes to the free space tree will go
+ * through the normal add/remove hooks.
+ */
+static int populate_free_space_tree(struct btrfs_trans_handle *trans,
+				    struct btrfs_fs_info *fs_info,
+				    struct btrfs_block_group_cache *block_group)
+{
+	struct btrfs_root *extent_root = fs_info->extent_root;
+	struct btrfs_path *path, *path2;
+	struct btrfs_key key;
+	u64 start, end;
+	int ret;
+
+	path = btrfs_alloc_path();
+	if (!path)
+		return -ENOMEM;
+	path->reada = 1;
+
+	path2 = btrfs_alloc_path();
+	if (!path2) {
+		btrfs_free_path(path);
+		return -ENOMEM;
+	}
+
+	ret = add_new_free_space_info(trans, fs_info, block_group, path2);
+	if (ret)
+		goto out;
+
+	/*
+	 * Iterate through all of the extent and metadata items in this block
+	 * group, adding the free space between them and the free space at the
+	 * end. Note that EXTENT_ITEM and METADATA_ITEM are less than
+	 * BLOCK_GROUP_ITEM, so an extent may precede the block group that it's
+	 * contained in.
+	 */
+	key.objectid = block_group->key.objectid;
+	key.type = BTRFS_EXTENT_ITEM_KEY;
+	key.offset = 0;
+
+	ret = btrfs_search_slot_for_read(extent_root, &key, path, 1, 0);
+	if (ret < 0)
+		goto out;
+	ASSERT(ret == 0);
+
+	start = block_group->key.objectid;
+	end = block_group->key.objectid + block_group->key.offset;
+	while (1) {
+		btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
+
+		if (key.type == BTRFS_EXTENT_ITEM_KEY ||
+		    key.type == BTRFS_METADATA_ITEM_KEY) {
+			if (key.objectid >= end)
+				break;
+
+			if (start < key.objectid) {
+				ret = __add_to_free_space_tree(trans, fs_info,
+							       block_group,
+							       path2, start,
+							       key.objectid -
+							       start);
+				if (ret)
+					goto out;
+			}
+			start = key.objectid;
+			if (key.type == BTRFS_METADATA_ITEM_KEY)
+				start += fs_info->tree_root->nodesize;
+			else
+				start += key.offset;
+		} else if (key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
+			if (key.objectid != block_group->key.objectid)
+				break;
+		}
+
+		ret = btrfs_next_item(extent_root, path);
+		if (ret < 0)
+			goto out;
+		if (ret)
+			break;
+	}
+	if (start < end) {
+		ret = __add_to_free_space_tree(trans, fs_info, block_group,
+					       path2, start, end - start);
+		if (ret)
+			goto out;
+	}
+
+	ret = 0;
+out:
+	btrfs_free_path(path2);
+	btrfs_free_path(path);
+	return ret;
+}
+
+int btrfs_create_free_space_tree(struct btrfs_fs_info *fs_info)
+{
+	struct btrfs_trans_handle *trans;
+	struct btrfs_root *tree_root = fs_info->tree_root;
+	struct btrfs_root *free_space_root;
+	struct btrfs_block_group_cache *block_group;
+	struct rb_node *node;
+	int ret;
+
+	trans = btrfs_start_transaction(tree_root, 0);
+	if (IS_ERR(trans))
+		return PTR_ERR(trans);
+
+	free_space_root = btrfs_create_tree(trans, fs_info,
+					    BTRFS_FREE_SPACE_TREE_OBJECTID);
+	if (IS_ERR(free_space_root)) {
+		ret = PTR_ERR(free_space_root);
+		goto abort;
+	}
+	fs_info->free_space_root = free_space_root;
+
+	node = rb_first(&fs_info->block_group_cache_tree);
+	while (node) {
+		block_group = rb_entry(node, struct btrfs_block_group_cache,
+				       cache_node);
+		ret = populate_free_space_tree(trans, fs_info, block_group);
+		if (ret)
+			goto abort;
+		node = rb_next(node);
+	}
+
+	btrfs_set_fs_compat_ro(fs_info, FREE_SPACE_TREE);
+
+	ret = btrfs_commit_transaction(trans, tree_root);
+	if (ret)
+		return ret;
+
+	return 0;
+
+abort:
+	btrfs_abort_transaction(trans, tree_root, ret);
+	btrfs_end_transaction(trans, tree_root);
+	return ret;
+}
+
+static int clear_free_space_tree(struct btrfs_trans_handle *trans,
+				 struct btrfs_root *root)
+{
+	struct btrfs_path *path;
+	struct btrfs_key key;
+	int nr;
+	int ret;
+
+	path = btrfs_alloc_path();
+	if (!path)
+		return -ENOMEM;
+
+	path->leave_spinning = 1;
+
+	key.objectid = 0;
+	key.type = 0;
+	key.offset = 0;
+
+	while (1) {
+		ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
+		if (ret < 0)
+			goto out;
+
+		nr = btrfs_header_nritems(path->nodes[0]);
+		if (!nr)
+			break;
+
+		path->slots[0] = 0;
+		ret = btrfs_del_items(trans, root, path, 0, nr);
+		if (ret)
+			goto out;
+
+		btrfs_release_path(path);
+	}
+
+	ret = 0;
+out:
+	btrfs_free_path(path);
+	return ret;
+}
+
+int btrfs_clear_free_space_tree(struct btrfs_fs_info *fs_info)
+{
+	struct btrfs_trans_handle *trans;
+	struct btrfs_root *tree_root = fs_info->tree_root;
+	struct btrfs_root *free_space_root = fs_info->free_space_root;
+	int ret;
+
+	trans = btrfs_start_transaction(tree_root, 0);
+	if (IS_ERR(trans))
+		return PTR_ERR(trans);
+
+	btrfs_clear_fs_compat_ro(fs_info, FREE_SPACE_TREE);
+	fs_info->free_space_root = NULL;
+
+	ret = clear_free_space_tree(trans, free_space_root);
+	if (ret)
+		goto abort;
+
+	ret = btrfs_del_root(trans, tree_root, &free_space_root->root_key);
+	if (ret)
+		goto abort;
+
+	list_del(&free_space_root->dirty_list);
+
+	btrfs_tree_lock(free_space_root->node);
+	clean_tree_block(trans, tree_root->fs_info, free_space_root->node);
+	btrfs_tree_unlock(free_space_root->node);
+	btrfs_free_tree_block(trans, free_space_root, free_space_root->node,
+			      0, 1);
+
+	free_extent_buffer(free_space_root->node);
+	free_extent_buffer(free_space_root->commit_root);
+	kfree(free_space_root);
+
+	ret = btrfs_commit_transaction(trans, tree_root);
+	if (ret)
+		return ret;
+
+	return 0;
+
+abort:
+	btrfs_abort_transaction(trans, tree_root, ret);
+	btrfs_end_transaction(trans, tree_root);
+	return ret;
+}
+
+static int __add_block_group_free_space(struct btrfs_trans_handle *trans,
+					struct btrfs_fs_info *fs_info,
+					struct btrfs_block_group_cache *block_group,
+					struct btrfs_path *path)
+{
+	u64 start, end;
+	int ret;
+
+	start = block_group->key.objectid;
+	end = block_group->key.objectid + block_group->key.offset;
+
+	block_group->needs_free_space = 0;
+
+	ret = add_new_free_space_info(trans, fs_info, block_group, path);
+	if (ret)
+		return ret;
+
+	return __add_to_free_space_tree(trans, fs_info, block_group, path,
+					block_group->key.objectid,
+					block_group->key.offset);
+}
+
+int add_block_group_free_space(struct btrfs_trans_handle *trans,
+			       struct btrfs_fs_info *fs_info,
+			       struct btrfs_block_group_cache *block_group)
+{
+	struct btrfs_path *path = NULL;
+	int ret = 0;
+
+	if (!btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE))
+		return 0;
+
+	mutex_lock(&block_group->free_space_lock);
+	if (!block_group->needs_free_space)
+		goto out;
+
+	path = btrfs_alloc_path();
+	if (!path) {
+		ret = -ENOMEM;
+		goto out;
+	}
+
+	ret = __add_block_group_free_space(trans, fs_info, block_group, path);
+
+out:
+	btrfs_free_path(path);
+	mutex_unlock(&block_group->free_space_lock);
+	if (ret)
+		btrfs_abort_transaction(trans, fs_info->free_space_root, ret);
+	return ret;
+}
+
+int remove_block_group_free_space(struct btrfs_trans_handle *trans,
+				  struct btrfs_fs_info *fs_info,
+				  struct btrfs_block_group_cache *block_group)
+{
+	struct btrfs_root *root = fs_info->free_space_root;
+	struct btrfs_path *path;
+	struct btrfs_key key, found_key;
+	struct extent_buffer *leaf;
+	u64 start, end;
+	int done = 0, nr;
+	int ret;
+
+	if (!btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE))
+		return 0;
+
+	if (block_group->needs_free_space) {
+		/* We never added this block group to the free space tree. */
+		return 0;
+	}
+
+	path = btrfs_alloc_path();
+	if (!path) {
+		ret = -ENOMEM;
+		goto out;
+	}
+
+	start = block_group->key.objectid;
+	end = block_group->key.objectid + block_group->key.offset;
+
+	key.objectid = end - 1;
+	key.type = (u8)-1;
+	key.offset = (u64)-1;
+
+	while (!done) {
+		ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
+		if (ret)
+			goto out;
+
+		leaf = path->nodes[0];
+		nr = 0;
+		path->slots[0]++;
+		while (path->slots[0] > 0) {
+			btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0] - 1);
+
+			if (found_key.type == BTRFS_FREE_SPACE_INFO_KEY) {
+				ASSERT(found_key.objectid == block_group->key.objectid);
+				ASSERT(found_key.offset == block_group->key.offset);
+				done = 1;
+				nr++;
+				path->slots[0]--;
+				break;
+			} else if (found_key.type == BTRFS_FREE_SPACE_EXTENT_KEY ||
+				   found_key.type == BTRFS_FREE_SPACE_BITMAP_KEY) {
+				ASSERT(found_key.objectid >= start);
+				ASSERT(found_key.objectid < end);
+				ASSERT(found_key.objectid + found_key.offset <= end);
+				nr++;
+				path->slots[0]--;
+			} else {
+				ASSERT(0);
+			}
+		}
+
+		ret = btrfs_del_items(trans, root, path, path->slots[0], nr);
+		if (ret)
+			goto out;
+		btrfs_release_path(path);
+	}
+
+	ret = 0;
+out:
+	btrfs_free_path(path);
+	if (ret)
+		btrfs_abort_transaction(trans, root, ret);
+	return ret;
+}
+
+static int load_free_space_bitmaps(struct btrfs_caching_control *caching_ctl,
+				   struct btrfs_path *path,
+				   u32 expected_extent_count)
+{
+	struct btrfs_block_group_cache *block_group;
+	struct btrfs_fs_info *fs_info;
+	struct btrfs_root *root;
+	struct btrfs_key key;
+	int prev_bit = 0, bit;
+	/* Initialize to silence GCC. */
+	u64 extent_start = 0;
+	u64 end, offset;
+	u64 total_found = 0;
+	u32 extent_count = 0;
+	int ret;
+
+	block_group = caching_ctl->block_group;
+	fs_info = block_group->fs_info;
+	root = fs_info->free_space_root;
+
+	end = block_group->key.objectid + block_group->key.offset;
+
+	while (1) {
+		ret = btrfs_next_item(root, path);
+		if (ret < 0)
+			goto out;
+		if (ret)
+			break;
+
+		btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
+
+		if (key.type == BTRFS_FREE_SPACE_INFO_KEY)
+			break;
+
+		ASSERT(key.type == BTRFS_FREE_SPACE_BITMAP_KEY);
+		ASSERT(key.objectid < end && key.objectid + key.offset <= end);
+
+		caching_ctl->progress = key.objectid;
+
+		offset = key.objectid;
+		while (offset < key.objectid + key.offset) {
+			bit = free_space_test_bit(block_group, path, offset);
+			if (prev_bit == 0 && bit == 1) {
+				extent_start = offset;
+			} else if (prev_bit == 1 && bit == 0) {
+				total_found += add_new_free_space(block_group,
+								  fs_info,
+								  extent_start,
+								  offset);
+				if (total_found > CACHING_CTL_WAKE_UP) {
+					total_found = 0;
+					wake_up(&caching_ctl->wait);
+				}
+				extent_count++;
+			}
+			prev_bit = bit;
+			offset += block_group->sectorsize;
+		}
+	}
+	if (prev_bit == 1) {
+		total_found += add_new_free_space(block_group, fs_info,
+						  extent_start, end);
+		extent_count++;
+	}
+
+	if (extent_count != expected_extent_count) {
+		btrfs_err(fs_info, "incorrect extent count for %llu; counted %u, expected %u",
+			  block_group->key.objectid, extent_count,
+			  expected_extent_count);
+		ASSERT(0);
+		ret = -EIO;
+		goto out;
+	}
+
+	caching_ctl->progress = (u64)-1;
+
+	ret = 0;
+out:
+	return ret;
+}
+
+static int load_free_space_extents(struct btrfs_caching_control *caching_ctl,
+				   struct btrfs_path *path,
+				   u32 expected_extent_count)
+{
+	struct btrfs_block_group_cache *block_group;
+	struct btrfs_fs_info *fs_info;
+	struct btrfs_root *root;
+	struct btrfs_key key;
+	u64 end;
+	u64 total_found = 0;
+	u32 extent_count = 0;
+	int ret;
+
+	block_group = caching_ctl->block_group;
+	fs_info = block_group->fs_info;
+	root = fs_info->free_space_root;
+
+	end = block_group->key.objectid + block_group->key.offset;
+
+	while (1) {
+		ret = btrfs_next_item(root, path);
+		if (ret < 0)
+			goto out;
+		if (ret)
+			break;
+
+		btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
+
+		if (key.type == BTRFS_FREE_SPACE_INFO_KEY)
+			break;
+
+		ASSERT(key.type == BTRFS_FREE_SPACE_EXTENT_KEY);
+		ASSERT(key.objectid < end && key.objectid + key.offset <= end);
+
+		caching_ctl->progress = key.objectid;
+
+		total_found += add_new_free_space(block_group, fs_info,
+						  key.objectid,
+						  key.objectid + key.offset);
+		if (total_found > CACHING_CTL_WAKE_UP) {
+			total_found = 0;
+			wake_up(&caching_ctl->wait);
+		}
+		extent_count++;
+	}
+
+	if (extent_count != expected_extent_count) {
+		btrfs_err(fs_info, "incorrect extent count for %llu; counted %u, expected %u",
+			  block_group->key.objectid, extent_count,
+			  expected_extent_count);
+		ASSERT(0);
+		ret = -EIO;
+		goto out;
+	}
+
+	caching_ctl->progress = (u64)-1;
+
+	ret = 0;
+out:
+	return ret;
+}
+
+int load_free_space_tree(struct btrfs_caching_control *caching_ctl)
+{
+	struct btrfs_block_group_cache *block_group;
+	struct btrfs_fs_info *fs_info;
+	struct btrfs_free_space_info *info;
+	struct btrfs_path *path;
+	u32 extent_count, flags;
+	int ret;
+
+	block_group = caching_ctl->block_group;
+	fs_info = block_group->fs_info;
+
+	path = btrfs_alloc_path();
+	if (!path)
+		return -ENOMEM;
+
+	/*
+	 * Just like caching_thread() doesn't want to deadlock on the extent
+	 * tree, we don't want to deadlock on the free space tree.
+	 */
+	path->skip_locking = 1;
+	path->search_commit_root = 1;
+	path->reada = 1;
+
+	info = search_free_space_info(NULL, fs_info, block_group, path, 0);
+	if (IS_ERR(info)) {
+		ret = PTR_ERR(info);
+		goto out;
+	}
+	extent_count = btrfs_free_space_extent_count(path->nodes[0], info);
+	flags = btrfs_free_space_flags(path->nodes[0], info);
+
+	/*
+	 * We left path pointing to the free space info item, so now
+	 * load_free_space_foo can just iterate through the free space tree from
+	 * there.
+	 */
+	if (flags & BTRFS_FREE_SPACE_USING_BITMAPS)
+		ret = load_free_space_bitmaps(caching_ctl, path, extent_count);
+	else
+		ret = load_free_space_extents(caching_ctl, path, extent_count);
+
+out:
+	btrfs_free_path(path);
+	return ret;
+}
diff --git a/fs/btrfs/free-space-tree.h b/fs/btrfs/free-space-tree.h
new file mode 100644
index 000000000000..54ffced3bce8
--- /dev/null
+++ b/fs/btrfs/free-space-tree.h
@@ -0,0 +1,72 @@ 
+/*
+ * Copyright (C) 2015 Facebook.  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.
+ */
+
+#ifndef __BTRFS_FREE_SPACE_TREE
+#define __BTRFS_FREE_SPACE_TREE
+
+/*
+ * The default size for new free space bitmap items. The last bitmap in a block
+ * group may be truncated, and none of the free space tree code assumes that
+ * existing bitmaps are this size.
+ */
+#define BTRFS_FREE_SPACE_BITMAP_SIZE 256
+#define BTRFS_FREE_SPACE_BITMAP_BITS (BTRFS_FREE_SPACE_BITMAP_SIZE * BITS_PER_BYTE)
+
+void set_free_space_tree_thresholds(struct btrfs_block_group_cache *block_group);
+int btrfs_create_free_space_tree(struct btrfs_fs_info *fs_info);
+int btrfs_clear_free_space_tree(struct btrfs_fs_info *fs_info);
+int load_free_space_tree(struct btrfs_caching_control *caching_ctl);
+int add_block_group_free_space(struct btrfs_trans_handle *trans,
+			       struct btrfs_fs_info *fs_info,
+			       struct btrfs_block_group_cache *block_group);
+int remove_block_group_free_space(struct btrfs_trans_handle *trans,
+				  struct btrfs_fs_info *fs_info,
+				  struct btrfs_block_group_cache *block_group);
+int add_to_free_space_tree(struct btrfs_trans_handle *trans,
+			   struct btrfs_fs_info *fs_info,
+			   u64 start, u64 size);
+int remove_from_free_space_tree(struct btrfs_trans_handle *trans,
+				struct btrfs_fs_info *fs_info,
+				u64 start, u64 size);
+
+/* Exposed for testing. */
+struct btrfs_free_space_info *
+search_free_space_info(struct btrfs_trans_handle *trans,
+		       struct btrfs_fs_info *fs_info,
+		       struct btrfs_block_group_cache *block_group,
+		       struct btrfs_path *path, int cow);
+int __add_to_free_space_tree(struct btrfs_trans_handle *trans,
+			     struct btrfs_fs_info *fs_info,
+			     struct btrfs_block_group_cache *block_group,
+			     struct btrfs_path *path, u64 start, u64 size);
+int __remove_from_free_space_tree(struct btrfs_trans_handle *trans,
+				  struct btrfs_fs_info *fs_info,
+				  struct btrfs_block_group_cache *block_group,
+				  struct btrfs_path *path, u64 start, u64 size);
+int convert_free_space_to_bitmaps(struct btrfs_trans_handle *trans,
+				  struct btrfs_fs_info *fs_info,
+				  struct btrfs_block_group_cache *block_group,
+				  struct btrfs_path *path);
+int convert_free_space_to_extents(struct btrfs_trans_handle *trans,
+				  struct btrfs_fs_info *fs_info,
+				  struct btrfs_block_group_cache *block_group,
+				  struct btrfs_path *path);
+int free_space_test_bit(struct btrfs_block_group_cache *block_group,
+			struct btrfs_path *path, u64 offset);
+
+#endif