Hi Joe,

On Wed, May 4, 2016 at 12:58 PM, Joe Touch <[email protected]> wrote:

> Hi, Yoshifumi,
>
> On 5/3/2016 7:47 PM, Yoshifumi Nishida wrote:
>
> Hi Joe,
>
> On Mon, May 2, 2016 at 10:38 PM, Joe Touch <[email protected]> wrote:
>>
>>
>> On 5/1/2016 9:30 PM, Yoshifumi Nishida wrote:
>> > Hi,
>> > I am thinking about dividing encrypting a TCP segment into the two
>> parts:
>> > A) encrypt arbitrary part of a TCP segment and B) determine which
>> > parts of a TCP segment should be encrypted.
>> > Because I might want to think them separately if possible.
>> >
>> > From my point of view, when we encrypt TCP payload, we just mask the
>> > header part of the TCP segment before encryption.
>>
>> That would mean that encryption would depend on the length of the header
>> part. Middleboxes are known to insert, delete, fold, spindle, and
>> otherwise mutilate options, which means that you would be encrypting
>> over a different length of zero-content prefix.
>>
>> At a minimum, you need to know where the TCP payload is and start there,
>> not merely mask out the header. That currently involves parsing only the
>> TCP base header, but in the future may also involve parsing one of the
>> TCP options (TCP-EDO) (which would further complicate the notion of
>> encrypting options or even a portion thereof)
>>
>
> We'll need to know where the encryption part is. If it's payload, it will
> be easy as we can get the starting point from the header.
> But, I am guessing it doesn't have to be payload.
>
> ...
>
>
> > I am naively thinking that it's not very difficult to change masking
>> > areas.
>>
>> AFAICT, it would defeat encryption (or should).
>>
>> Hmm.. Sorry, I couldn't catch this. Could you elaborate?
>
>
> It's just a comment on using "masking" rather than an offset to start
> encryption.
>
> I.e., encrypting over "0abc" yields a different result than "abc". It's
> not sufficient to just mask-out the prefix area if that area could vary in
> length.
>
> A side note: some implementations put TCP headers and payloads in
> different places, so "just backing up to include the options" can require
> more complicated code than it may first appear as well.
>

OK. I'm not very sure how it is complicated to take care of masking and to
follow linked lists in memory. But, anyway, I still think it's at least
worth for discussion as an optional feature.

Thanks,
--
Yoshi
_______________________________________________
Tcpinc mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/tcpinc

Reply via email to