On Wed, Nov 16, 2011 at 09:58:14PM +0900, MORITA Kazutaka wrote: > > > If we increment SD_PROTO_VER, the sheep daemon should return > > > SD_RES_VER_MISMATCH error when opening the VDI. > > > > And then revert to trying the old protocol revision if it fails? Sounds > > like that would be worth a try. > > Or perhaps it might be better not to support the old protocol in the > qemu block driver? It will make the code complicated. qemu has cache modes with and without a writeback cache, and the modes without it are absolutely required for supporting OSes that don't support cache flushing. I would suggest we introduce a new protocol version, and that version will allow the server and client to negotiate optional feature bits. The write cache would be the first one, but things like discard support would also be able to make use of it. |