[OAUTH-WG] draft-ietf-oauth-v2-bearer-08.txt WGLC comments

["Manger, James H" <[e-mail]>]

My working group last call comments on draft-ietf-oauth-v2-bearer-08.txt:


1. Mentioning that this is an HTTP authentication mechanism in the title and/or abstract would be useful to the wider IETF (& beyond) audience.
Title:
  "The BEARER HTTP authentication mechanism for use with OAuth 2"
Abstract:
  "This specification describes how to use bearer tokens in
   HTTP requests to access OAuth 2 protected resources."

[Personally, I wouldn't bother mentioning OAuth at all here, but others seem to want this context restriction.]


2. The ABNF for <credentials> does not comply with RFC 2617 "HTTP Authentication". And even though RFC 2617 is broken is this aspect, the BEARER spec doesn't comply with the errata (still broken) or the more likely fixes proposed for HTTPbis [draft-ietf-httpbis-p7-auth].
I expect HTTPbis to allow a base64-like-blob consistently in Authorization and WWW-Authenticate headers (to accommodate BASIC and NTLM). Multiple WWW-Authenticate headers can have their values combined, separated by commas. They can also have quoted-string parameters. To be able to parse this, requires disallowing commas and double-quotes from the base64-like-blob (and hence from <access-token>) at a minimum; only allowing equals at the end also helps.
The current approach in the bearer spec disallows all but 94 chars/bytes. I suggest reducing this to 69. Something in between (eg 91 chars, dropping comma, quote, and slash) might work. None of these options are materially easier than the others for a token issuer; and less symbols just means less risk of escaping problems elsewhere (eg allowing "<" in an access token will wreck someone's XML somewhere, for no benefit).

Suggestion: 
  access-token = 1*( ALPHA / DIGIT / "-" / "." / "_" / "~" / "+" / "/" ) *"="

  <access-token> includes the 66 unreserved URI characters plus a few others.
  It can hold a base64, base64url (URL and filename safe alphabet),
  base32, or base16 (hex) encoding, with or without padding, but
  excluding whitespace [RFC4648].

2b. If 2 is not accepted (and assuming HTTPbis will allow any content after the scheme name in a Authorization header) can we please not misuse the <quoted-char> label when no quoting is going on. The following is a better equivalent:

  access-token = 1*(%x21-7E) ; ASCII, except controls, space, or delete


3. Drop '\' from the allowed chars in a scope value, to avoid clashing with the HTTP quoted-string escaping mechanism (and don't use the <quoted-char> label when no quoting is going on).
  scope-v = 1*(%x21 / %x23-5B / %x5D-7E); excludes space and " and \


4. Section 3.3 "Summary of Recommendations" sensibly says clients "MUST ensure that bearer tokens are not leaked to *unintended parties*" and correctly notes that this is "the primary security consideration" that underlies all the others. So it is a glaring hole that OAuth2 fails to tell the client who the intended parties are when issuing a bearer token.


--
James Manger