Btrfs: remove last_log_alloc allocator optimization
The tree logging code was trying to separate tree log allocations from normal metadata allocations to improve writeback patterns during an fsync. But, the code was not effective and ended up just mixing tree log blocks with regular metadata. That seems to be working fairly well, so the last_log_alloc code can be removed. Signed-off-by: Chris Mason <chris.mason@oracle.com>
This commit is contained in:
@@ -2228,13 +2228,6 @@ static int noinline find_free_extent(struct btrfs_trans_handle *trans,
|
||||
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) {
|
||||
*last_ptr = root->fs_info->last_alloc + empty_cluster;
|
||||
}
|
||||
}
|
||||
|
||||
if (last_ptr) {
|
||||
if (*last_ptr)
|
||||
hint_byte = *last_ptr;
|
||||
|
Reference in New Issue
Block a user