The start table is stored as file offsets. We subtract the
header+table size to get an index into the remaining data.
If the file offset is smaller than that, the subtract wraps
and the later start+length check can wrap too.
Drop the file instead of walking off the buffer.
