Looks good, and this might actually explain the weird corruption I saw in some changes I'm working on the make use of b_offset. Reviewed-by: Christoph Hellwig <hch at lst.de>