Re: [secdir] secdir review of draft-ietf-lisp-lcaf-15

[David Mandelberg <[e-mail]>]

On 09/25/2016 06:15 PM, Dino Farinacci wrote:
>> I found one issue in various parts of the document (described below),
>> but I'm not sure it's relevant to security. If it is, then I think this
>> document is Almost Ready. If not, then I think this document is Ready
>> With Nits.
>>
>> There are multiple places in the document where it's possible to encode
>> semantically equivalent information in multiple ways, despite the word
>> "canonical" being in the title of the document. Is there anything that
>> relies on these addresses being canonical for security purposes?
> 
> No not for security purposes. There are multiple ways because we allow some nesting of information as well as allow for compatibility for older implementations that can’t parse some AFIs and LCAFs but is required to parse the AFI-List LCAF type.

Ok, then I think it needs to be made clear in the security
considerations that this format cannot be relied on to have no more than
one representation for the same information.

IESG: This means that I think this document is Ready With Nits.


>> Multiple places in the document (sections 4.1, 4.5, and 4.8) specify
>> mask lengths, but do not specify that the masked out bits MUST be set to
>> zero.
> 
> Hmm, by definition, a mask-length of say 24 if a mask of 0xffffff00. And in 4.1:
> 
>   IID mask-len:  if the AFI is set to 0, then this format is not
>       encoding an extended EID-prefix but rather an instance-ID range
>       where the 'IID mask-len' indicates the number of high-order bits
>       used in the Instance ID field for the range.
> 
> It is clear that the high-order bits are used that cover the mask-length and the low-order bits are ignored. Is this not clear to you?

That part is clear to me. I just meant that the document as written
appears to allow 1.2.3.4/24 to be equivalent to 1.2.3.7/24. If the
format were going to be canonical, then all masked bits (the last octet
in this example) would need to be set to a well-known value (so,
1.2.3.0/24) instead of allowed to be anything.

> Would you want me to say the high-order bits are added with a mask and the bits not covered by the mask are set to 0?

If the format can't be relied on to be canonical, then my point is
irrelevant.


>> Section 5.4: There are many different ways to encode equivalent JSON
>> data here.
> 
> Maybe, but can you be more clear.

"{}" and "{ }" (note the space) are equivalent in JSON, but not
identical when compared byte-by-byte.

> I don’t think it is a bug. Do you?

As long as nobody relies on these to be canonical, it's not a bug.


>> Section 6 (Security Considerations): There is no discussion of these
>> addresses being canonical, and what other systems might or might not
>> rely on these addresses being canonical.
> 
> In this respect “canonical” is relative to how LISP defines to encode both simple AFI encoded addresses or more complex/flexible addresses that accompany information.

Ok. I interpreted canonical to be a security property of the addresses,
i.e., that if two addresses are not bytewise equal, then they must not
have the same meaning. If that's not what you meant by canonical, then I
think the security considerations should say that, so that nobody tries
to rely on that property in the future.


>> Section 4.7: The Key Count description talks about "Key sections," but
>> doesn't say which fields are part of the key section (and can thus be
>> repeated). I have a guess which fields are part of the key section, but
>> it's not entirely clear.
> 
> I added some text to clarify this.

Thank you. The fields included in a key section were not the ones I
guessed, so the clarification was helpful. (Before your clarification, I
expected it to be everything from Rsvd3 through Key Material.)


>> Section 4.7: The Key Algorithm description doesn't point to a registry
>> of valid values or otherwise say how to interpret values in that field.
> 
> We have put that in a use-case document. This Security Key LCAF is used by 2 use-cases. It is used for LISP-DDT and for lisp-crypto. In the lisp-crypto document we have a registry for cipher suites used.

Should this document reference one or both of those documents?


-- 
David Eric Mandelberg / dseomn
http://david.mandelberg.org/