Re: [TLS] 1rtt thoughts

[Eric Rescorla <[e-mail]>]

A large fraction of existing TLS servers will respond to this situation by
generating a fatal alert and closing the connection.That makes this
approach a nonstarter.

Given that, whether it is possible for a compliant TLS server to behave
differently seems like a largely academic question.

-Ekr


On Mon, Jul 14, 2014 at 1:06 PM, Michael StJohns <[e-mail]>
wrote:

>  On 7/14/2014 1:49 PM, Eric Rescorla wrote:
>
>
>
>
> On Mon, Jul 14, 2014 at 10:27 AM, Adam Langley <[e-mail]>
> wrote:
>
>> On Mon, Jul 14, 2014 at 10:16 AM, Michael StJohns
>> <[e-mail]> wrote:
>> > A
>> > TLS1.2 server receiving things in this order discards the
>> ClientKeyExchange
>> > with an out of sequence error, and then starts the handshake normally
>> with
>> > the receipt of the ClientHello.
>>
>>  I've never encountered a TLS 1.2 implementation that will discard an
>> unexpected handshake message like that as opposed to sending a fatal
>> alert (maybe) and closing the connection.
>>
>
>  Indeed, it's required by TLS.
>
>
> Umm...
>
> Part of the problem I have at times in discussions about TLS is the
> confusion between what happens as part of the transport connection and what
> happens as part of the TLS connection.  I think of TLS as a record
> connection over a bearer transport service.  They interact, but not
> strongly, and usually not without intervention from the client or server
> application.  The TLS thing (server, message processor, etc) attached to
> the server side of the transport connection snarfs up individual TLS
> records one by one and processes them according to its current state.
>
> While you're both correct that TLS requires the "connection" to be closed
> in the event of a fatal error  (and here I'm referring to 7.2.2 of
> RFC5246), the context of the word "connection" in that section doesn't
> appear to be same as closing the "transport" (see just above that where the
> two different terms are used).   It appears to only refer to closing the
> current TLS connection.
> I
> Nothing in section 7.2.2 suggests either the requirement to close the
> transport, nor the requirement to discard pending data in the transport
> connection.  I'm not arguing that implementations don't do exactly that
> though - I don't actually know.   if you're arguing that the standard says
> to close the transport and discard the pending TLS records, can you point
> me at that section?
>
> I also can't find anything that says that a new connection can't be opened
> on the same transport connection after a fatal close.
>
> So I would suggest that even with 7.2.2 you get this as an error case
> between a 1rtt TLS1.3 client and a 1.2 server:
>
> Client                                                      Server
> ClientKeyExchange ----->
> ClientHello ---->
>                                                            [Server notices
> data waiting on the transport
>                                                             Server begins
> a connection state
>                                                             Server
> receives and processes unexpected ClientKeyExchange
>                                                             Server sends a
> fatal alert and discards the connection state]
>                                        <--------------- Error Alert
> (unexpected_message) (plus close_notify??)
>  [Client receives the error alert
>    checks for type and discards it as expected]
>                                                            [Server notices
> data waiting on the transport
>                                                             Server begins
> a connection state
>                                                             Server
> Receives and processes ClientHello
>                                                             Server sends
> its part of a TLS1.2 handshake]
>                                         <--------------- ServerHello,
> ServerKeyExchange etc
> [Client receives the Server Side messages and processes
>   them...
>  Client resends the ClientKeyExchange that was ignored above]
> ClientKeyExchange  ------>
> etc --->
>  .............................  Continue through the normal connection
> process.
>
>
>
> From the server's side this looks like two different connections.  From
> the client side, it's just negotiating to the right suite.
>
> Mike
>
>
>
>
>  -Ekr
>
>
>> We could refine TLS 1.2 processing, of course, but if you were
>> expecting that to be backwards compatible with existing code then I
>> fear it won't work.
>>
>
>
>
>> Cheers
>>
>> AGL
>>
>> --
>> Adam Langley [e-mail] https://www.imperialviolet.org
>>
>> _______________________________________________
>> TLS mailing list
>> [e-mail]
>> https://www.ietf.org/mailman/listinfo/tls
>>
>
>
>