Re: Generating a 421 status from a proxy

[James Peach <[e-mail]>]

> On Apr 28, 2020, at 12:44 PM, James Peach <[e-mail]> wrote:
> 
> Hi all,
> 
> I have an reverse proxy Envoy configuration where each SNI server name is attached to exactly one virtual host routing table. If this configuration is deployed with a wildcard certificate however, browser clients will re-use the TLS connections for server name A to send requests for origin B, due to connection reuse, https://httpwg.org/specs/rfc7540.html#reuse. In this configuration, envoy generates a 404 because the configuration for servername A doesn’t have any routes for B.
> 
> I believe that in this situation, generating a 421 response should cause the client to not re-use the connection for a different (but wildcard-matching) hostname. However, the spec also says that a proxy must not generate a 421. I wasn't able to track down any rationale for why a proxy must not generate a 421; would it be considered inappropriate in this kind of configuration? Or is it OK, since from the client’s perspective, the reverse proxy is the origin?
> 
> The example use case for 421 status in section 9.1.1 is a TLS-terminating middlebox, which matches my scenario pretty closely. To my reading, this conflicts with the "MUST NOT be generated by proxies” requirement in 9.1.2. 

To answer my own question (I think) ... the reverse proxy is a "gateway", not a “proxy”, so the MUST NOT doesn’t apply here.

J