Re: [TLS] DTLS and large messages

[Michael Tüxen <[e-mail]>]

On Jun 16, 2010, at 10:06 AM, t.petch wrote:

> ----- Original Message -----
> From: "Michael Tüxen" <[e-mail]>
> Sent: Tuesday, June 15, 2010 3:29 PM
> 
> On Jun 15, 2010, at 11:06 AM, Juho Vähä-Herttua wrote:
> 
>> 15.6.2010 10:25, t.petch kirjoitti:
>>> How may an application send a 64kbyte (less headers) message over DTLS,
> assuming
>>> that the substrate is UDP and that UDP can carry 64kbyte datagrams without
>>> fragmentation?
>> 
>> This sounds like another argument for max_fragment_length extension allowing
> lengths larger than 2^14...
>> 
>>> A usually reliable source has surprised me on two counts:
>>> 
>>> First, I am told that when rfc4347-bis says
>>> "As in TLS
>>>       1.2, the length should not exceed 2^14.
>>> "
>>> then that is the authors' way of saying
>>> "As in TLS 1.2, the length MUST NOT exceed 2^14."
>>> which I had not guessed.
>> 
>> That wording is indeed confusing and should probably be changed, TLS 1.2 uses
> MUST NOT so a "should not" is not "as in TLS 1.2". With current wording I
> wouldn't recommend exceeding 2^14 length, since it is quite likely to break with
> some implementations.
>> 
>>> Second, that DTLS will break up the message into fragments of 2**14 or less
> but
>>> will then place them in a single UDP datagram so that they will transit the
>>> network as one, and can be safely reassembled at the far end.
>> 
>> The specification says: "Multiple DTLS records may be placed in a single
> datagram. They are simply encoded consecutively." So in case of 64kB datagrams
> there can be either multiple complete records (with fragments of 2^14 or less)
> included in a single datagram or each record having its own datagram, in the
> receiving end both cases should be handled. It's basically the same as in TLS
> 1.2 where large handshake messages need to be fragmented into fragments of 2^14
> or less and sent consecutively.
> 
> DTLS will not fragment any user data. So the application is limited by the path
> mtu.
> When running DTLS/SCTP, SCTP announces a MTU of 2**14 and uses its own
> fragmentation,
> but for UDP this is not possible.
> To use user messages larger than 2^14, you need a link layer with an MTU of 2^14
> or
> higher. Do you?
> 
> <tp>
> 
> I am assuming UDP, no SCTP, and a link layer MTU of at least 64kbyte, with a a
> message something under 64kbyte to allow for potentially multiple DTLS headers
> to still fit the original application message in a single UDP datagram.
Haven't seen a link with an MTU of 64KByte yet, but at least in the future it
might happen and Matt would love it... 
> 
> (In passing, I would love to see documentation on DTLS implementations but what
> is publicly available seems scarce to non-existent).
Our experience with the OpenSSL implementation of DTLS is pretty good, at least
after applying the patches available from
http://sctp.fh-muenster.de/
which mostly are integrated in the releases.

Best regards
Michael
> 
> Tom Petch
> 
> </tp>
> 
> 
> 
> 
> 
> Best regards
> Michael
>> 
>> This is just my interpretation, but I think the specification is quite clear
> on this, and your source is probably correct.
>> 
>> 
>> Juho
>> 
>> _______________________________________________
>> TLS mailing list
>> [e-mail]
>> https://www.ietf.org/mailman/listinfo/tls
>> 
> 
>