btrfs: don't check DUP chunks twice
authorArne Jansen <sensille@gmx.net>
Thu, 9 Feb 2012 14:09:02 +0000 (15:09 +0100)
committerDavid Sterba <dsterba@suse.cz>
Wed, 15 Feb 2012 15:40:24 +0000 (16:40 +0100)
Because scrub enumerates the dev extent tree to find the chunks to scrub,
it currently finds each DUP chunk twice and also scrubs it twice. This
patch makes sure that scrub_chunk only checks that part of the chunk the
dev extent has been found for. This only changes the behaviour for DUP
chunks.

Reported-and-tested-by: Stefan Behrens <sbehrens@giantdisaster.de>
Signed-off-by: Arne Jansen <sensille@gmx.net>
fs/btrfs/scrub.c

index 9770cc5bfb76c6829f96924bb82f9b3b564ca646..abc0fbffa510f67f601dc80e38bfecf091a88153 100644 (file)
@@ -1367,7 +1367,8 @@ out:
 }
 
 static noinline_for_stack int scrub_chunk(struct scrub_dev *sdev,
-       u64 chunk_tree, u64 chunk_objectid, u64 chunk_offset, u64 length)
+       u64 chunk_tree, u64 chunk_objectid, u64 chunk_offset, u64 length,
+       u64 dev_offset)
 {
        struct btrfs_mapping_tree *map_tree =
                &sdev->dev->dev_root->fs_info->mapping_tree;
@@ -1391,7 +1392,8 @@ static noinline_for_stack int scrub_chunk(struct scrub_dev *sdev,
                goto out;
 
        for (i = 0; i < map->num_stripes; ++i) {
-               if (map->stripes[i].dev == sdev->dev) {
+               if (map->stripes[i].dev == sdev->dev &&
+                   map->stripes[i].physical == dev_offset) {
                        ret = scrub_stripe(sdev, map, i, chunk_offset, length);
                        if (ret)
                                goto out;
@@ -1487,7 +1489,7 @@ int scrub_enumerate_chunks(struct scrub_dev *sdev, u64 start, u64 end)
                        break;
                }
                ret = scrub_chunk(sdev, chunk_tree, chunk_objectid,
-                                 chunk_offset, length);
+                                 chunk_offset, length, found_key.offset);
                btrfs_put_block_group(cache);
                if (ret)
                        break;