Btrfs: Add BH_Defrag to mark buffers that are in need of defragging

This allows the tree walking code to defrag only the newly allocated
buffers, it seems to be a good balance between perfect defragging and the
performance hit of repeatedly reallocating blocks.

Signed-off-by: Chris Mason <chris.mason@oracle.com>

4 files changed, 18 insertions, 11 deletions

diff --git a/fs/btrfs/ctree.c b/fs/btrfs/ctree.c
index ee1ae00d2827..7cf43da5e78e 100644
--- a/fs/btrfs/ctree.c
+++ b/fs/btrfs/ctree.c
@@ -175,6 +175,7 @@ int btrfs_realloc_node(struct btrfs_trans_handle *trans,
         int end_slot;
         int i;
         int err = 0;
+        int parent_level;
 
         if (trans->transaction != root->fs_info->running_transaction) {
                 printk(KERN_CRIT "trans %Lu running %Lu\n", trans->transid,
@@ -188,6 +189,7 @@ int btrfs_realloc_node(struct btrfs_trans_handle *trans,
         }
         parent_node = btrfs_buffer_node(parent);
         parent_nritems = btrfs_header_nritems(&parent_node->header);
+        parent_level = btrfs_header_level(&parent_node->header);
 
         start_slot = 0;
         end_slot = parent_nritems;
@@ -215,13 +217,16 @@ int btrfs_realloc_node(struct btrfs_trans_handle *trans,
 
                 cur_bh = btrfs_find_tree_block(root, blocknr);
                 if (!cur_bh || !buffer_uptodate(cur_bh) ||
-                    buffer_locked(cur_bh)) {
+                    buffer_locked(cur_bh) || !buffer_defrag(cur_bh)) {
                         if (cache_only) {
                                 brelse(cur_bh);
                                 continue;
                         }
-                        brelse(cur_bh);
-                        cur_bh = read_tree_block(root, blocknr);
+                        if (!cur_bh || !buffer_uptodate(cur_bh) ||
+                            buffer_locked(cur_bh)) {
+                                brelse(cur_bh);
+                                cur_bh = read_tree_block(root, blocknr);
+                        }
                 }
                 if (search_start == 0)
                         search_start = last_block & ~((u64)65535);
@@ -232,6 +237,9 @@ int btrfs_realloc_node(struct btrfs_trans_handle *trans,
                 if (err)
                         break;
                 search_start = bh_blocknr(tmp_bh);
+                *last_ret = search_start;
+                if (parent_level == 1)
+                        clear_buffer_defrag(tmp_bh);
                 brelse(tmp_bh);
         }
         return err;
@@ -811,16 +819,10 @@ static void reada_for_search(struct btrfs_root *root, struct btrfs_path *path,
                         clear_radix_bit(&found, blocknr);
                         if (nread > 32)
                                 continue;
-                        if (direction > 0 && cluster_start <= blocknr &&
-                            cluster_start + 8 > blocknr) {
-                                cluster_start = blocknr;
+                        if (close_blocks(cluster_start, blocknr)) {
                                 readahead_tree_block(root, blocknr);
                                 nread++;
-                        } else if (direction < 0 && cluster_start >= blocknr &&
-                                   blocknr + 8 > cluster_start) {
                                 cluster_start = blocknr;
-                                readahead_tree_block(root, blocknr);
-                                nread++;
                         }
                 }
         }
diff --git a/fs/btrfs/disk-io.h b/fs/btrfs/disk-io.h
index 9e2c261b41ae..81fd18cbd824 100644
--- a/fs/btrfs/disk-io.h
+++ b/fs/btrfs/disk-io.h
@@ -25,8 +25,10 @@
 
 enum btrfs_bh_state_bits {
         BH_Checked = BH_PrivateStart,
+        BH_Defrag,
 };
 BUFFER_FNS(Checked, checked);
+BUFFER_FNS(Defrag, defrag);
 
 static inline struct btrfs_node *btrfs_buffer_node(struct buffer_head *bh)
 {
diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c
index 3418bb62b996..b3641234473f 100644
--- a/fs/btrfs/extent-tree.c
+++ b/fs/btrfs/extent-tree.c
@@ -1015,6 +1015,7 @@ check_failed:
         ins->objectid = search_start;
         ins->offset = 0;
         start_found = 0;
+        path->reada = 1;
 
         ret = btrfs_search_slot(trans, root, ins, path, 0, 0);
         if (ret < 0)
@@ -1264,6 +1265,7 @@ struct buffer_head *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
         WARN_ON(buffer_dirty(buf));
         set_buffer_uptodate(buf);
         set_buffer_checked(buf);
+        set_buffer_defrag(buf);
         set_radix_bit(&trans->transaction->dirty_pages, buf->b_page->index);
         return buf;
 }
diff --git a/fs/btrfs/tree-defrag.c b/fs/btrfs/tree-defrag.c
index a09064a9a41c..35fd20d24645 100644
--- a/fs/btrfs/tree-defrag.c
+++ b/fs/btrfs/tree-defrag.c
@@ -86,7 +86,7 @@ static int defrag_walk_down(struct btrfs_trans_handle *trans,
                 if (cache_only) {
                         next = btrfs_find_tree_block(root, blocknr);
                         if (!next || !buffer_uptodate(next) ||
-                           buffer_locked(next)) {
+                           buffer_locked(next) || !buffer_defrag(next)) {
                                 brelse(next);
                                 path->slots[*level]++;
                                 continue;
@@ -142,6 +142,7 @@ static int defrag_walk_up(struct btrfs_trans_handle *trans,
                         root->defrag_level = i;
                         return 0;
                 } else {
+                        clear_buffer_defrag(path->nodes[*level]);
                         btrfs_block_release(root, path->nodes[*level]);
                         path->nodes[*level] = NULL;
                         *level = i + 1;