Re: [Nbd] Testing NBD server implementations for correctness

[Wouter Verhelst <w@...112...>]

On Mon, Sep 26, 2016 at 11:43:42AM +0100, Alex Bligh wrote:
> 
> > On 26 Sep 2016, at 10:21, Carl-Daniel Hailfinger <c-d.hailfinger.devel.2006@...2724...> wrote:
> > 
> > Wow, that was quick! Thank you.
> > 
> > I stumbled upon another problem: Apparently nbd-tester-client and nbdkit
> > disagree on what constitutes a valid flush request.
> > nbd-tester-client complains:
> > ./flush
> > 15901: Requests: 3536
> > ** (process:15901): WARNING **: Could not run test: Received error from
> > server: 22 (0x16). Handle is -9223372036854764544 (0x8000000000002C00).
> > 
> > nbdkit complains:
> > nbdkit: python[7]: error: invalid flush request: expecting offset and
> > length == 0
> > 
> > Not sure where that bug is, nbdkit or nbd-tester-client.
> > - Is the request correct and should have been accepted?
> > - Is the reject correct, but the response is screwed up?
> 
> Per the docs:
> 
> 	• For a flush request, length and offset are reserved, and MUST be set to all-zero.
> 
> So if nbd-tester-client.c is sending a flush with non-zero length or offset, it is wrong.

Right.

We have had cases in the past where out-in-the-wild implementations
turned out to behave differently to how the spec behaved, but in this
case I think it's a pretty safe bet to say that most implementations
should generally DTRT here. I'm guessing that nbd-tester-client doesn't
because of uninitialized code or something (haven't actually looked at
it, must go, will do so later).

TLDR: nbdkit is correct, nbd-tester-client is not, patches welcome.

-- 
< ron> I mean, the main *practical* problem with C++, is there's like a dozen
       people in the world who think they really understand all of its rules,
       and pretty much all of them are just lying to themselves too.
 -- #debian-devel, OFTC, 2016-02-12