btrfs: scrub: errors in tree enumeration
authorArne Jansen <sensille@gmx.net>
Fri, 3 Jun 2011 08:09:26 +0000 (10:09 +0200)
committerArne Jansen <sensille@gmx.net>
Fri, 10 Jun 2011 10:14:13 +0000 (12:14 +0200)
due to the semantics of btrfs_search_slot the path can point to an
invalid slot when ret > 0. This condition went unnoticed, which in
turn could have led to an incomplete scrubbing.

Signed-off-by: Arne Jansen <sensille@gmx.net>
fs/btrfs/scrub.c

index df50fd1..d5a4108 100644 (file)
@@ -804,18 +804,12 @@ static noinline_for_stack int scrub_stripe(struct scrub_dev *sdev,
 
                ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
                if (ret < 0)
-                       goto out;
-
-               l = path->nodes[0];
-               slot = path->slots[0];
-               btrfs_item_key_to_cpu(l, &key, slot);
-               if (key.objectid != logical) {
-                       ret = btrfs_previous_item(root, path, 0,
-                                                 BTRFS_EXTENT_ITEM_KEY);
-                       if (ret < 0)
-                               goto out;
-               }
+                       goto out_noplug;
 
+               /*
+                * we might miss half an extent here, but that doesn't matter,
+                * as it's only the prefetch
+                */
                while (1) {
                        l = path->nodes[0];
                        slot = path->slots[0];
@@ -824,7 +818,7 @@ static noinline_for_stack int scrub_stripe(struct scrub_dev *sdev,
                                if (ret == 0)
                                        continue;
                                if (ret < 0)
-                                       goto out;
+                                       goto out_noplug;
 
                                break;
                        }
@@ -906,15 +900,20 @@ again:
                ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
                if (ret < 0)
                        goto out;
-
-               l = path->nodes[0];
-               slot = path->slots[0];
-               btrfs_item_key_to_cpu(l, &key, slot);
-               if (key.objectid != logical) {
+               if (ret > 0) {
                        ret = btrfs_previous_item(root, path, 0,
                                                  BTRFS_EXTENT_ITEM_KEY);
                        if (ret < 0)
                                goto out;
+                       if (ret > 0) {
+                               /* there's no smaller item, so stick with the
+                                * larger one */
+                               btrfs_release_path(path);
+                               ret = btrfs_search_slot(NULL, root, &key,
+                                                       path, 0, 0);
+                               if (ret < 0)
+                                       goto out;
+                       }
                }
 
                while (1) {
@@ -989,6 +988,7 @@ next:
 
 out:
        blk_finish_plug(&plug);
+out_noplug:
        btrfs_free_path(path);
        return ret < 0 ? ret : 0;
 }
@@ -1064,8 +1064,15 @@ int scrub_enumerate_chunks(struct scrub_dev *sdev, u64 start, u64 end)
        while (1) {
                ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
                if (ret < 0)
-                       goto out;
-               ret = 0;
+                       break;
+               if (ret > 0) {
+                       if (path->slots[0] >=
+                           btrfs_header_nritems(path->nodes[0])) {
+                               ret = btrfs_next_leaf(root, path);
+                               if (ret)
+                                       break;
+                       }
+               }
 
                l = path->nodes[0];
                slot = path->slots[0];
@@ -1075,7 +1082,7 @@ int scrub_enumerate_chunks(struct scrub_dev *sdev, u64 start, u64 end)
                if (found_key.objectid != sdev->dev->devid)
                        break;
 
-               if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
+               if (btrfs_key_type(&found_key) != BTRFS_DEV_EXTENT_KEY)
                        break;
 
                if (found_key.offset >= end)
@@ -1104,7 +1111,7 @@ int scrub_enumerate_chunks(struct scrub_dev *sdev, u64 start, u64 end)
                cache = btrfs_lookup_block_group(fs_info, chunk_offset);
                if (!cache) {
                        ret = -ENOENT;
-                       goto out;
+                       break;
                }
                ret = scrub_chunk(sdev, chunk_tree, chunk_objectid,
                                  chunk_offset, length);
@@ -1116,9 +1123,13 @@ int scrub_enumerate_chunks(struct scrub_dev *sdev, u64 start, u64 end)
                btrfs_release_path(path);
        }
 
-out:
        btrfs_free_path(path);
-       return ret;
+
+       /*
+        * ret can still be 1 from search_slot or next_leaf,
+        * that's not an error
+        */
+       return ret < 0 ? ret : 0;
 }
 
 static noinline_for_stack int scrub_supers(struct scrub_dev *sdev)