Jason Thorpe wrote:
> The way NAStore currently works is that blocking is only required if
> you read past the portion of the file that has already been restored.

Hmm.  That's nice, but I'd still want optional nonblock semantics on
reads, then.  That way an application can read as much as is currently
availible without risk of hanging itself for a long time, or it can give
the user status updates on how it's progressing, etc.

Also, the interaction of holes with this is potentially a sticky point
(not too hard, I don't think -- just needs a little care)