Re: [multipathtcp] q about on-path proxy

[<[e-mail]>]

Hi Olivier, 

Please see inline.

Cheers,
Med

> -----Message d'origine-----
> De : multipathtcp [mailto:[e-mail]] De la part de
> Olivier Bonaventure
> Envoyé : mercredi 22 mars 2017 10:35
> À : Yoshifumi Nishida; multipathtcp
> Objet : Re: [multipathtcp] q about on-path proxy
> 
> Yoshifumi,
> >
> > I am trying to understand on-path proxy scenarios.
> > (BTW, I'm referring draft-nam-mptcp-deployment-considerations-01
> > and draft-boucadair-mptcp-plain-mode-10)
> >
> > In case of on-path proxy (implicit mode),the dest of the primary subflow
> > will be the receiver. It's not very clear to me how MCP intercepts the
> > packets in this subflow.
> 
> 
> There are several techniques that can be deployed by operators to ensure
> that the transparent MCP is on-path for the initial subflow. All these
> techniques operate in the network layer or below and are thus
> transparent for TCP. Let me give an example based on source routing that
> does not match an actual deployment but would illustrate the main
> principle. With source routing, one can add inside the IP header a loose
> hop.
> 
> 
>         +------- R4 ---- R5 --------+
> C --- uMCP ---- R1 ---- R2 ------ dMCP ---- R6 --- R7 --- S
> 
> Consider a TCP connection created by client C towards server S. The uMCP
> is the first hop router for C. It intercepts the connection and
> transparently terminates it, acting as S. It then initiates a Multipath
> TCP connection towards S, acting as C. The packets for this connection
> are sent as IP packet that contain the IP address of the downstream MCP
> as a loose source routing hop. The dMCP removes the source routing
> option, transparently terminates the Multipath TCP connection and then
> initiates a connection towards S, acting as C.
> 
> Three connections exist to transfer data from C et S and vice-versa.
> 
> C<->uMCP  (source address C, destination address S)
> uMCP<->dMCP (source address C, destination address S, using source
> routing to ensure that all downstream packets pass through dMCP, for
> upstream packets, source routing is usually not required because there
> is only one path to the uMCP that advertises C to R1 only )
> dMCP<-> S (source address C, destination address S)
> 
> For the Multipath TCP connection between uMCP and dMCP, the second
> subflow is used without relying on source routing. The uMCP and the dMCP
> have addresses on the R4/R5 network and use the ADD_ADDR option to
> advertise those addresses and create the required subflows.
> 
> > The MCP splits the connection between two nodes and send forged packets
> > to the sender?
> 
> Yes
> 
> > Also, in the two proxy scenario, does the downstream MCP have to be
> > on-path?
> 
> If the downstream MCP is on path, then it does not have to include any
> NAT function which is beneficial from an operational viewpoint.
> 
> 

[Med] Perhaps I misunderstood your point but DNAT/SNAT are still needed for subsequent subflows even for the implicit case. Think about subflows that are placed with a destination address set to the one of the MCP and with a distinct source IP address than the one used to place the initial subflow.