Yep this is a missing piece of documentation. I'll add an issue to get
a fix in the 0.3 spec. Thanks for the close read and discussion,
everyone!

2009/10/8 Pádraic Brady <[email protected]>:
> Hi Nick,
>
> I will agree on one thing, the specification should probably clarify what it 
> means by "HMAC". My own reading of the specification was that 7.4 referred to 
> HMAC-SHA1, not simply SHA1. But then, like yourself, I always assume HMAC 
> refers to the relevant RFC, and not simply a salted hash which, using SHA1 
> alone, has few of the benefits offered by a proper HMAC.
>
> Incidentally, as far as I can recall, the reference hub was using Python's 
> hmac. The part I'm not sure on that specification indicates appending the 
> secret to the request data which seems either incorrect or redundant, and 
> more akin to how you'd expect a salted hash to function. I'm only now adding 
> in 0.2 changes to my own source code Subscriber/Hub so I really haven't had 
> time to test how it operates with the reference hub.
>
> Paddy
>
> Pádraic Brady
>
> http://blog.astrumfutura.com
> http://www.survivethedeepend.com
> OpenID Europe Foundation Irish Representative
>
>
> ________________________________
> From: Nick Johnson (Google) <[email protected]>
> To: [email protected]
> Sent: Thu, October 8, 2009 1:31:37 PM
> Subject: [pubsubhubbub] Re: Use of HMAC for authenticated content distribution
>
> Hi Padraic,
> Let me clarify: The construction described in the hubbub spec is this: H(m, 
> k), where H is the hash function, m is the message, and k is the secret key. 
> This is an 'append-only' hmac. The industry-standard HMAC (often just known 
> as 'HMAC') is defined (in RFC2104) as H(k xor opad, H(k xor ipad, m)), where 
> opad and ipad are fixed-length constants.
> Apart from having been cryptanalyzed and proven secure under very weak 
> assumptions for the underlying hash (see the paper referenced in the 
> Wikipedia page), this hmac is already implemented in many languages - see the 
> 'hmac' module in the Python standard library, for example.
> An example of an attack on an append-only HMAC is this: Suppose you can find 
> a collision in your basic hash function, such that H(a) == H(b). Now, if you 
> can persuade someone to generate a valid HMAC for the innocent message H(a), 
> you can use it as an HMAC for the malicious message b. The RFC2104 HMAC does 
> not have this weakness, due to its nested construction.
> -Nick Johnson
> On Thu, Oct 8, 2009 at 1:22 PM, Pádraic Brady <[email protected]> wrote:
>>
>> You might need to explain what you mean in greater detail. You say the use 
>> of a HMAC is ad-hoc and then refer to the concatenation of the secret to the 
>> request body. Yet, HMAC does not actually define the information format to 
>> be exchanged (how could it possibly do so?), so simple concatenation seems 
>> entirely reasonable since it a) fully represents the information both 
>> parties are exchanging and b) must be defined by the specification to ensure 
>> interoperability between implementations.
>>
>> Pádraic Brady
>>
>> http://blog.astrumfutura.com
>> http://www.survivethedeepend.com
>> OpenID Europe Foundation Irish Representative
>>
>>
>> ________________________________
>> From: Nick Johnson (Google) <[email protected]>
>> To: [email protected]
>> Sent: Thu, October 8, 2009 8:51:56 AM
>> Subject: [pubsubhubbub] Re: Use of HMAC for authenticated content 
>> distribution
>>
>> On Thu, Oct 8, 2009 at 3:38 AM, Sachin Shenoy <[email protected]> 
>> wrote:
>>>
>>> I am bit confused here. SHA in SHA-1 stands for Secure Hash Algorithm. Why 
>>> do you say it is ad-hoc?
>>
>> I'm not saying the hash algorithm itself is ad-hoc, I'm saying its use as an 
>> HMAC is ad-hoc. RFC2104 defines an accepted (and proven secure) way of 
>> constructing an HMAC, which is far preferable to the simple concatenation 
>> approach taken here.
>>
>> -Nick
>>
>>>
>>> If you meant "Why don't we support other hash function [configured/chosen 
>>> by a param], instead of just supporting SHA-1?" - I think that has to do 
>>> with this line from 
>>> spec. http://pubsubhubbub.googlecode.com/svn/trunk/pubsubhubbub-core-0.2.html
>>> "To dramatically simplify this spec in several places where we had to 
>>> choose between supporting A or B, we took it upon ourselves to say "only 
>>> A", rather than making it an implementation decision."
>>> Thanks,
>>> Sachin
>>>
>>> On Thu, Oct 8, 2009 at 2:15 AM, Nick Johnson (Google) 
>>> <[email protected]> wrote:
>>>>
>>>> The hubbub spec, in section 7.4, says: 
>>>> http://pubsubhubbub.googlecode.com/svn/trunk/pubsubhubbub-core-0.2.html#authednotify
>>>>
>>>> "The signature MUST be computed by appending the hub.secret value to the 
>>>> request body and then generating the combined string's HMAC using the SHA1 
>>>> algorithm."
>>>>
>>>> However, HMAC has a specific definition, in RFC2104, which allows for 
>>>> composing HMACs from secure hash algorithms. It's constructed specifically 
>>>> to make it more difficult to forge or brute-force an HMAC, a property the 
>>>> description in the hubbub spec lacks.
>>>>
>>>> Why does the hubbub spec use this ad-hoc construction instead of a proper 
>>>> HMAC?
>>>>
>>>> --
>>>> Nick Johnson, Developer Programs Engineer, App Engine
>>>> Google Ireland Ltd. :: Registered in Dublin, Ireland, Registration Number: 
>>>> 368047
>>>
>>
>>
>>
>> --
>> Nick Johnson, Developer Programs Engineer, App Engine
>> Google Ireland Ltd. :: Registered in Dublin, Ireland, Registration Number: 
>> 368047
>
>
>
> --
> Nick Johnson, Developer Programs Engineer, App Engine
> Google Ireland Ltd. :: Registered in Dublin, Ireland, Registration Number: 
> 368047
>

Reply via email to