Il 21/05/2013 17:54, Dietmar Maurer ha scritto: >>>> Hmm, right. But do we need the bitmap at all? We can just use >>>>> bdrv_is_allocated like bdrv_co_do_readv does. >>> If a write occur, we read and backup that cluster immediately (out of >>> order). So I am quite sure we need the bitmap. >> >> This is the same as how copy-on-read happens during image streaming, and it >> doesn't use a bitmap. > > But a read does not modify the content.
Copy-on-read modifies the topmost image even without changing the content, just like copy-before-write modifies the backup image. But writes to the backup copy are serialized anyway via wait_for_overlapping_requests, so there is no problem here. Paolo