[core] draft-ietf-core-coap-tcp-tls message length question

[Jim Schaad]

I am trying to decide if the maximum message size for coaps+tcp is pre or
post TLS. I would assume that it would be pre-TLS to match with the message
size requirements of coaps, however the minimal guidance on estimating the
size of the TLS overhead is not in this document while it is in RFC 7252.
Since there are multiple buffers that might need to be dealt with, and most
TLS implementations are going to have an internal buffer as well, it is not
completely clear to me what is being stated.

Jim

[weigengyu]

Hi Jim,

I am not sure fully understood the message and its options.

Post by Jim Schaad
I am trying to decide if the maximum message size for coaps+tcp is pre or
post TLS.
I would assume that it would be pre-TLS to match with the message size
requirements of coaps,
however the minimal guidance on estimating the size of the TLS overhead is
not in this document while it is in RFC 7252.

It is pre-TLS in our implementation which is based on extending CF.CoAP in
accordance with the former version draft.
The maximum message size for coaps+tcp is determined when binding the
protocol stack.

By the way, it also is tried to have only one CoAP message format of RFC7252
over any transport facilities.
A CoAP message length option is defined privately and carried in a CoAP
message of RFC7252.
The delimiter of one CoAP message is clear for CoAP/UDP, CoAP/TCP, or
CoAP/WSS.
This is not a standard.
It works, and not many CoAP ACK overheads are felt.

Regards,

Gengyu WEI
Network Technology Center
School of Computer
Beijing University of Posts and Telecommunications
-----原始邮件-----
From: Jim Schaad
Sent: Tuesday, May 02, 2017 3:04 PM
To: draft-ietf-core-coap-tcp-***@ietf.org
Cc: 'core'
Subject: [core] draft-ietf-core-coap-tcp-tls message length question

I am trying to decide if the maximum message size for coaps+tcp is pre or
post TLS. I would assume that it would be pre-TLS to match with the message
size requirements of coaps, however the minimal guidance on estimating the
size of the TLS overhead is not in this document while it is in RFC 7252.
Since there are multiple buffers that might need to be dealt with, and most
TLS implementations are going to have an internal buffer as well, it is not
completely clear to me what is being stated.

Jim


_______________________________________________
core mailing list
***@ietf.org
https://www.ietf.org/mailman/listinfo/core

[Carsten Bormann]

Post by Jim Schaad
I am trying to decide if the maximum message size for coaps+tcp is pre or
post TLS.

Good question. RFC 7252 Section 4.6 discusses message sizes, but never really comes clean what exactly that means, because it doesn’t need to.
But the progression from 1280 to 1152 to 1024 for MTU, message size, payload size kind of makes clear that this is the size of the CoAP message.
Which is OK, as “message size” is not very different from “size of the message”, exactly what’s meant.

Post by Jim Schaad
I would assume that it would be pre-TLS to match with the message
size requirements of coaps, however the minimal guidance on estimating the
size of the TLS overhead is not in this document while it is in RFC 7252.

TCP removes the need to consider MTU that much:
TCP segmentation is much more well-behaved than the IP (or 6LoWPAN adaptation layer) fragmentation needed for UDP.

On the other hand, there is still a bit of space between 1152 and 1280 for IP, TCP, TLS, so segmentation for default-maximum-size messages is not necessary.

Post by Jim Schaad
Since there are multiple buffers that might need to be dealt with, and most
TLS implementations are going to have an internal buffer as well, it is not
completely clear to me what is being stated.

If there were an intention to include TLS overheads inside the figure for message size, this would run into trouble quickly, because there is no way for the CoAP implementation to control how many TLS headers the TLS layer spends for one message. You can be optimistic and say “1”, but that is just that, optimistic (or the actual number might even be “0.5” if there is some concurrency).

So, clearly, the sensible interpretation, again, is for “message size” to be the size of the CoAP message (header/length/token, Options, payload marker, payload).
(Note that, for a ~ 1152 byte message, the size of a CoAP over TCP message is exactly the same as that of a CoAP over UDP message; see Figure 7 of draft-ietf-core-coap-tcp-tls.)

Grüße, Carsten

[Hannes Tschofenig]

Hi Jim,

similarly to Carsten and Gengyu I would argue that the message size
refers to the payload of the CoAP message since this is something the
CoAP layer would know. Our mbed TLS stack also provides information to
the application about the level of "expansion" due to encryption but
then there is potentially also padding involved (although more
theoretically than in practice to conceal the real size of the message).

Ciao
Hannes

Post by Jim Schaad
I am trying to decide if the maximum message size for coaps+tcp is pre or
post TLS. I would assume that it would be pre-TLS to match with the message
size requirements of coaps, however the minimal guidance on estimating the
size of the TLS overhead is not in this document while it is in RFC 7252.
Since there are multiple buffers that might need to be dealt with, and most
TLS implementations are going to have an internal buffer as well, it is not
completely clear to me what is being stated.
Jim
_______________________________________________
core mailing list
https://www.ietf.org/mailman/listinfo/core