On 10/06/2017 02:34 AM, Vladimir Sementsov-Ogievskiy wrote:

>> - if the chunk is NBD_REPLY_TYPE_NONE, there is no payload, and this
>> should be the final chunk, so the return to the caller can be the same
>> as for old-style (return 1 if we had no earlier error packets, or the
>> saved negative value corresponding to the first error received)
>> - if the command is read, we can read the payload into qiov (saves
>> malloc'ing space for the reply only to copy it into the qiov), so we
> 
> but here you said "This is putting a LOT of smarts directly into the
> receive routine"

About the only reason to justify putting smarts into the receive routine
is if it is the most common case, where the overhead of not
special-casing will penalize us.  READ happens to be a frequent command,
all other commands, like BLOCK_STATUS, will probably be infrequent.
Making READ malloc the result, only to then copy it into the qiov, is a
waste of time; no other structured command will pass a qiov.  So yeah,
maybe I'm stating things a bit differently than in my earlier messages,
but that's because I've now stated reasonings for why it is okay to
special-case READ with a qiov differently from all other commands (none
of which will pass a qiov to the receive routine).

Thanks for persisting with this, and I'm looking forward to the next
revision that you post; hopefully, by taking this discussion, we are
making sure that the design is solid.

-- 
Eric Blake, Principal Software Engineer
Red Hat, Inc.           +1-919-301-3266
Virtualization:  qemu.org | libvirt.org