Btrfs: try to free metadata pages when we free btree blocks
authorChris Mason <chris.mason@oracle.com>
Tue, 31 Mar 2009 17:47:50 +0000 (13:47 -0400)
committerChris Mason <chris.mason@oracle.com>
Tue, 31 Mar 2009 18:27:58 +0000 (14:27 -0400)
COW means we cycle though blocks fairly quickly, and once we
free an extent on disk, it doesn't make much sense to keep the pages around.

This commit tries to immediately free the page when we free the extent,
which lowers our memory footprint significantly.

Signed-off-by: Chris Mason <chris.mason@oracle.com>
fs/btrfs/extent-tree.c

index 0c482e0d7c434093227d162a23ac56d6fc895943..f5e7cae63d80f8828f711833b58600a8bce71df9 100644 (file)
@@ -2384,6 +2384,10 @@ static int __free_extent(struct btrfs_trans_handle *trans,
                if (owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
                        ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
                        BUG_ON(ret);
                if (owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
                        ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
                        BUG_ON(ret);
+               } else {
+                       invalidate_mapping_pages(info->btree_inode->i_mapping,
+                            bytenr >> PAGE_CACHE_SHIFT,
+                            (bytenr + num_bytes - 1) >> PAGE_CACHE_SHIFT);
                }
 
                ret = update_block_group(trans, root, bytenr, num_bytes, 0,
                }
 
                ret = update_block_group(trans, root, bytenr, num_bytes, 0,