]> www.pilppa.org Git - linux-2.6-omap-h63xx.git/commitdiff
Btrfs: readahead checksums during btrfs_finish_ordered_io
authorChris Mason <chris.mason@oracle.com>
Fri, 13 Mar 2009 15:41:46 +0000 (11:41 -0400)
committerChris Mason <chris.mason@oracle.com>
Tue, 24 Mar 2009 20:14:51 +0000 (16:14 -0400)
This reads in blocks in the checksum btree before starting the
transaction in btrfs_finish_ordered_io.  It makes it much more likely
we'll be able to do operations inside the transaction without
needing any btree reads, which limits transaction latencies overall.

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

index b83a45dc717eb6472c86584af2f999cddd0d1ca9..9b4faac50c1848f035737557331cb550b8ed6f51 100644 (file)
@@ -1497,6 +1497,30 @@ static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
        return 0;
 }
 
+/*
+ * helper function for btrfs_finish_ordered_io, this
+ * just reads in some of the csum leaves to prime them into ram
+ * before we start the transaction.  It limits the amount of btree
+ * reads required while inside the transaction.
+ */
+static noinline void reada_csum(struct btrfs_root *root,
+                               struct btrfs_path *path,
+                               struct btrfs_ordered_extent *ordered_extent)
+{
+       struct btrfs_ordered_sum *sum;
+       u64 bytenr;
+
+       sum = list_entry(ordered_extent->list.next, struct btrfs_ordered_sum,
+                        list);
+       bytenr = sum->sums[0].bytenr;
+
+       /*
+        * we don't care about the results, the point of this search is
+        * just to get the btree leaves into ram
+        */
+       btrfs_lookup_csum(NULL, root->fs_info->csum_root, path, bytenr, 0);
+}
+
 /* as ordered data IO finishes, this gets called so we can finish
  * an ordered extent if the range of bytes in the file it covers are
  * fully written.
@@ -1505,7 +1529,7 @@ static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
 {
        struct btrfs_root *root = BTRFS_I(inode)->root;
        struct btrfs_trans_handle *trans;
-       struct btrfs_ordered_extent *ordered_extent;
+       struct btrfs_ordered_extent *ordered_extent = NULL;
        struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
        struct btrfs_path *path;
        int compressed = 0;
@@ -1528,13 +1552,20 @@ static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
                        ret = btrfs_lookup_file_extent(NULL, root, path,
                                                       inode->i_ino,
                                                       start, 0);
+                       ordered_extent = btrfs_lookup_ordered_extent(inode,
+                                                                    start);
+                       if (!list_empty(&ordered_extent->list)) {
+                               btrfs_release_path(root, path);
+                               reada_csum(root, path, ordered_extent);
+                       }
                        btrfs_free_path(path);
                }
        }
 
        trans = btrfs_join_transaction(root, 1);
 
-       ordered_extent = btrfs_lookup_ordered_extent(inode, start);
+       if (!ordered_extent)
+               ordered_extent = btrfs_lookup_ordered_extent(inode, start);
        BUG_ON(!ordered_extent);
        if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags))
                goto nocow;