Hi all, this series removes the BLK_EH_HANDLED return value, and instead places responsibility of timed out commands entirely on the drivers. Except for some odd layering vilations in libiscsi this actually is surprisingly simple. The last two pathes contain a respin of Keith' series to refcount struct request in the eye of timeouts. Once the BLK_EH_HANDLED special case is gone this actually seems the EH fixing approach that seems the most sensible to me.
