On 1/5/19 8:50 AM, Richard W.M. Jones wrote:
This allows multiple connections from a single client, and should be safe assuming flush/FUA has been implemented correctly in the previous commit.
Looks safe to me between the two patches. The NBD spec also states: * A client which uses multiple connections to a server to parallelize commands MUST NOT issue an `NBD_CMD_FLUSH` request until it has received the reply for all write commands which it expects to be covered by the flush. which implies that even if we encounter this race: client A: client B: issue write issue flush flush completes write completes we don't have to worry about that case, because client B didn't wait for the response before client A requested the flush. Kernel fdatasync() semantics match what we want (all writes that any client has received a reply to will be included in the flush, even if the flush is on a different client). ACK -- Eric Blake, Principal Software Engineer Red Hat, Inc. +1-919-301-3226 Virtualization: qemu.org | libvirt.org