Il 28/11/2012 15:46, Daniel P. Berrange ha scritto:
On Wed, Nov 28, 2012 at 03:43:16PM +0100, Paolo Bonzini wrote:
> Il 28/11/2012 11:59, Daniel P. Berrange ha scritto:
>> If the user has requested TUNNELLED migration, we need follow up work done.
>> We need the NBD server to be able to accept a pre-opened file descriptor
>> to rather than listening on a TCP host/port,
>
> This is already supported. However, the pre-opened fd would be for a
> listening socket.
>
> The main problem is that even though a single port is used on the
> destination, it is used for multiple connections. Migration would
> require an arbitrary number of streams, and I'm afraid supporting this
> would basically entail rewriting all the tunnelling code.
Hmm, yes, that is rather a complex problem.
I've long thought that QEMU migration should have a mode where it passes
all its data through TLS natively. So you could do secure migration,
without needing to use libvirtd tunnelling.
I agree. Hopefully, the various rewrites/refactorings of the upstream
migration code will make this easier.
The same is really true of the NBD code. I know the
"official" NBD
impl does not do encryption, but there's no strong reason why QEMU
can't layer in TLS below the NBD protocol if it is an explicit
opt-in at both client+server requested by libvirt. That would at
least solve the security issue, without requiring tunnelling.
Yes, that's a good idea.
Paolo