Re: [TLS] [DTLS1.3]About the retransmission of Handshake records

[Eric Rescorla <[e-mail]>]

U

On Wed, Nov 27, 2019 at 11:16 PM Xuan k <[e-mail]> wrote:

> Hi Ekr,
>
> Thanks for your help.
>
> I have another question about the "message_seq" in section "6. Example of
> Handshake with Timeout and Retransmission".
> Could you please explain it?
>
> In the secion 6, the Client send message_seq = 0 in Record 0 and
> message_seq = 2,3,4 in Record 2.
>
> Why message_seq = 1 is skipped by Client?
>

It's a typo in the example. The Certificate should be 1.



I think the ClientHello in the figure should begin with message_seq=1. The
> message_seq = 0 is used by the first ClientHello which is the one without
> cookies.
>

I agree in principle, but that's not this example, because, as you can see,
there is no HRR. Rather, this needs to remove +cookie.

-Ekr


> Thanks,
> Zhai Zhaoxuan
>
> On 11/28/19 5:37 AM, Eric Rescorla wrote:
>
>
>
> On Tue, Nov 26, 2019 at 10:05 PM Xuan k <[e-mail]> wrote:
>
>> Hi all,
>>
>> I'm trying to implement a DTLS1.3 library for embedded devices. But It
>> seems something weird about retransmissions and ACKs.
>>
>> In the section "5.2. DTLS Handshake Message Format":
>>
>>    The first message each side transmits in each association always has
>>    message_seq = 0.  Whenever a new message is generated, the
>>    message_seq value is incremented by one.  When a message is
>>    retransmitted, the old *message_seq value is re-used*, i.e., not
>>    incremented.  From the perspective of the DTLS record layer, the
>>    retransmission is a new record.  This record will have a *new*
>> *   DTLSPlaintext.sequence_number* value.
>>
>>
>> In the section "7. ACK Message", the ACK message use the record_numbers
>> (corresponds to *DTLSPlaintext.sequence_number*).
>>
>> For my understanding, the "message_seq" belongs to "Handshake" and the
>> "sequence_number" or "record_numbers" belongs to
>> record layer.
>>
>
> Yes.
>
> The retransmission detection is done by "Handshake" using "message_seq",
>> but the "acknowledge" is done by "record layer" using "record_numbers".
>> It is so weird.
>>
>
> Hmm... I don't think that this is particularly weird. This is, for
> instance, how QUIC stream acknowledgement and retransmission works.
>
>
> The retransmission, retransmission detection and acknowledge should be
>> done in handshake process, but we need the record layer passing the
>> record_numebrs to the handshake process.
>>
>> Since a new "sequence_number" is used for retransmission, we have to
>> maintain a "record_numbers" to "message_seq" map with dynamic size.
>> Each retransmission attempt creates a new relationship between a new
>> "record_numbers" to an old "message_seq".
>>
>
> Yes, that's how it works in NSS.
>
>
> Since ACK is only used with Handshake messages, is it possible that we use
>> "message_seq" in ACK messages?
>>
> Or we use *old* "sequence_number" for retransmission,
>>
>
> Both of these give you strictly less information about the network. One of
> the cool innovations in QUIC is to label each packet separately so you can
> determine whether an ACK is an ACK of the original packet or a retransmit.
> We are trying to inherit tha there
>
> -Ekr
>
> so we do not need maintain the dynamic map. And if replay detection is
>> implemented, the retransmitted
>> record can be dropped by record layer (by replay detection), the
>> "Handshake Protocol" do not need to do retransmission detection.
>>
>> Thanks
>> Zhai Zhaoxuan
>> _______________________________________________
>> TLS mailing list
>> [e-mail]
>> https://www.ietf.org/mailman/listinfo/tls
>>
>