On 01/12/2017 11:57 AM, Bob Chen wrote: > There might be a time window between the NBD server's resize and the > client's `re-read size` request. Is it safe?
For resize larger, it seems that it would be safe for the server to just remember the last size it has advertised to the client. As long as the client doesn't request read/write to any offset beyond the last-advertised size (either the size the server gave at handshake, or the size that the server reported when the client last used the new 're-read size' command), there's no difference in behavior (and well-behaved clients fall in this group); if the client DOES try to access out-of-bounds with respect to the last known size, the server SHOULD reject the access, but MAY serve it instead. A client that is unaware of the 're-read size' command will never learn that the server is now offering a larger image. For resize smaller, things are a lot trickier - how do you block access to a portion of a file that the client still thinks exist, but which the server has truncated away? Maybe the easy way out is to state that the server MUST NOT resize smaller. > > What about an active `resize` request from the client? Considering some NBD > servers might have the capability to do instant resizing, not applying to > LVM or host block device, of course... And perhaps the 're-read size' command can serve a dual purpose. Since all NBD commands already include space for size and offset parameters, we could define that if the client passes 0/0 for size and offset, it wants to read the server's current size; if it passes 0/non-zero, it is asking the server to resize to the new non-zero size (and the server's success or error response tells whether the resize happened). Should I spend time turning this idea into a more formal spec, along the lines of other NBD extension proposals? -- Eric Blake eblake redhat com +1-919-301-3266 Libvirt virtualization library http://libvirt.org
signature.asc
Description: OpenPGP digital signature