Comments on draft-hammer-oauth-v2-mac-token-02
<http://tools.ietf.org/html/draft-hammer-oauth-v2-mac-token-02>

This draft is a good improvement.

(continuing numbering from my previous comments)

11. The "access token" field would be better labelled "id". The MAC scheme
only needs this field to identify the MAC secret. The MAC scheme itself
doesn't need to care whether this id also represents a scope, duration,
assertion etc.

Though it may slightly complicated the use of the MAC scheme with OAuth2,
I would seriously consider supporting two identifiers:
* An authentication id -- to identify the key used to verify the MAC; and
* An authorization id -- to identify the scope/account/permissions/duration...
 that the client want to act under.
The SASL framework has these two concepts
[RFC4422 Simple Authentication and Security Layer (SASL)].
Any modern SASL authentication mechanism will be expected to support both ids.
The authorization id is a good match for an OAuth2 token or other sort of 
assertion.
The OAuth2-to-MAC binding could define a 'keyid' parameter to accompany the 
'secret'.
The 'keyid' value would map to the authentication id field in the MAC scheme,
and the 'access_token' would map to the authorization id field.

12. Which bytes does the body hash apply to? Is it the "message-body" or
"payload body" (in HTTPbis parlance)? I assume any Transfer-Encoding (eg
chunked) is removed before calculating the body hash, but any Content-Encoding
(eg gzip) is not removed. The spec should explicitly state this. [c.f. wording
as for the "Content-MD5" header in draft-ietf-httpbis-p3-payload-12, section
6.8]

13. The MAC algorithm should be explicitly indicated in the request, instead
of being implied by the access-token/id. I suggest including an "algorithm"
parameter in the "Authorization" request header. I also suggest including an
"algorithms" parameter in the "WWW-Authenticate" response header that is a
(comma-separated?) list of MAC algorithms that the server supports.

14. Explicitly state in section 3.3.2 (and 3.3.3) that SHA-1 (and SHA-256) are
used to calculate the body hash when using the hmac-sha-1 (and hmac-sha-256)
algorithm.

15. 3.3.2 (and 3.3.3) shouldn't mention the OAuth-specific "authorization
server" (when defining their "key" field).

16. OAuth2 can provide a "secret" as a Unicode string. MAC algorithms such as
HMAC use a key that is a byte array. Section 2 of the MAC spec says 'secret'
can only include printable ASCII chars (except " and /). This is not quite
right.
The MAC scheme should expect 'secret' to be a byte array (ie not limit its
chars). A 'secret' string from an OAuth2 response should be UTF-8 encoded to
produce the byte array the MAC scheme uses.

17. This MAC scheme should be compared to the SCRAM SASL mechanism [RFC 5802].
SCRAM is the latest-and-greatest authentication mechanism based on a shared 
secret
(without using the brilliant, but patent-troubled, zero-knowledge proof 
techniques).
It has quite a few features that MAC is missing: two ids (auth & authz);
credentials at the server are not sufficient to impersonate the client
(not a "password-equivalent"); mutual authentication; channel binding
(to an underlying TLS secure tunnel).
I don't think MAC can achieve all these properties. The MAC scheme almost 
certainly
wants to avoid multiple exchanges at the HTTP layer. I expect the wider IETF
community to judge MAC by comparing it negatively to SCRAM unless the document
is quite clear about its constraints (eg sign an HTTP request without any
interactivity with the server; suitable for shared secret keys, but not for
shared passwords...).


--
James Manger


-----Original Message-----
From: [email protected] [mailto:[email protected]] On Behalf Of
Manger, James H
Sent: Tuesday, 11 January 2011 5:05 PM
To: OAuth WG
Subject: [OAUTH-WG] MAC: comments on draft-hammer-oauth-v2-mac-token-00

Comments on draft-hammer-oauth-v2-mac-token-00
<http://tools.ietf.org/html/draft-hammer-oauth-v2-mac-token-00>

1. The connection of this MAC scheme to OAuth is secondary. The primary
function of the MAC scheme is to secure the authenticity of parts of an HTTP
request using a shared secret. Section 2 "Issuing MAC-Type Access Tokens"
would be better at the end, ideally as an Annex, along with most of the OAuth
talk from the introduction (and title).

2. Include a \n after the final query string parameter.
It makes the format simply lines of text, instead of making the last line
special (and only special if there are query params). It is awkward to
manually edit sample files, debug files etc without a final newline character.
Adding a final \n would just make it more consistent.

3. The Authorization header example in 1.1 uses single quotes around values.
It should use double quotes. The ABNF in 3.1 for timestamp, for instance,
explicitly uses <">.

4. The quoted-string production only supports a subset of ASCII. Better add a
sentence stating that the 'token' is assumed to only use this subset so no
escaping is defined. May be worth adding that only double-quote and backslash
need escaping in quoted-string, but neither of those are allowed in any of
this specs parameters so no escaping in quoted-string values shall occur.

5. Don't list the new line characters as separate numbered items in section
3.2.1 "Normalized Request String". Better to say it is a line-based format.
Each line ends with a single new line character (U+000A). Then list the 7
lines (as points 1 to 7), followed by point 8 saying there is 1 extra line for
each query string parameter.

6. I18N. Is the Normalised Request String UTF-8, or does it happen to be just
ASCII? Host, path, and query strings can contain non-ASCII characters -- but I
guess that the ASCII-only escaping used to create an HTTP Host header and URI
are assumed to have occurred before this spec uses these values. Would be
worth an explicit statement, and definitely an example.

7. I suggest swapping steps 2 (sort) & 3 (combine escaped name and value) in
section 3.2.1.1. "Parameters Normalization".
The current 2-layer sort is not ideal (sort on %-escaped name, then on
%-escaped value if the names are equal). It needs a special comparator
function etc, instead of using a programming language's standard
sort-a-list-of-strings function. It is tempting to concatenate %-escaped
name=value pairs and sort those -- but it doesn't always give the same result.
Eg a1=X and a11=Y should sort in that order according to the spec, but "1" <
"=" in an ASCII string comparison of the pairs.
A programmer probably has a map/dictionary/assoc-array of unescaped names to
values. The normalisation rules require a separate map of escaped names to
escaped values plus a custom comparator that considers names and values.
A nicer alternative would be to construct escaped name=value pairs then sort
that list (not map).
It would be worth adding another parameter to the example (eg a22=A) to
illustrate this (regardless of whether or not the steps are swapped).

8. HMAC operates on bytes, not strings. The 'text' and 'key' HMAC inputs (in
3.2.2 and 3.2.3) should be specified as the UTF-8-encoding of the Normalized
Request String and access token shared secret respectively. Might be able to
say ASCII-encoding, but only if we are sure that an OAuth token response
(which might be a UTF-8 JSON blob) can never return non-ASCII chars in a
'secret' field.

9. The 'secret' parameter is registered twice (7.1 & 7.2). I guess one of
these should be registering the 'algorithm' parameter instead.

10. I would like a "WWW-Authenticate: MAC algorithm=..." response header
defined. It would provide a nice illustration of how a complete HTTP auth
scheme is mapped to OAuth2. That is, how the scheme name "MAC" and parameters
from the "WWW-Authenticate: MAC ..." response header can be mapped to fields
in an OAuth2 token response (where they are accompanied by an actual secret
key).

--
James Manger

Attachment: smime.p7s
Description: S/MIME cryptographic signature

_______________________________________________
OAuth mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/oauth

Reply via email to