- From: Sandro Hawke <sandro@w3.org>
- Date: Tue, 18 Nov 2014 21:49:12 -0500
- To: Melvin Carvalho <melvincarvalho@gmail.com>
- CC: public-webid <public-webid@w3.org>, public-rww <public-rww@w3.org>
- Message-ID: <546C0528.5060507@w3.org>
On 11/12/2014 05:47 AM, Melvin Carvalho wrote:
>
>
> On 12 November 2014 05:36, Sandro Hawke <sandro@w3.org
> <mailto:sandro@w3.org>> wrote:
>
> On 11/10/2014 06:39 AM, Melvin Carvalho wrote:
>> Just wanted to highlight this interesting work from sandro
>
> Thanks. I should say the design came out of discussions with
> Andrei Sambra, trying to avoid the problems with poor browser
> support of client certificates.
>
>
> Sando, could you outline which part of certificate support you feel
> could be improved? Is it the UI, or lack or coverage, or something else?
>
UI, as in previous email
> Surely SPOT would require browser modification to work in the browser?
>
Mostly, no.
> - How does the browser send header fields? -- maybe AJAX
Of course.
> - How does the browser remember your URI, esp. cross origin? -- maybe
> stuff it somewhere like in web crypto
This bit is tricky, yeah. Of course it doesn't really need to remember
it, but the user experience is much better if it does. Browser support
would be nice. Without it, we can hack it with a tiny bit of code which
uses local storage, served up from some neutral domain like w3.org.
This is what webintents did, I believe.
> - How does the browser and your server maintain a shared secret? --
> maybe HTTPS
That's what the SPOT spec says, yes.
> Are you not replacing one set of problems with another?
>
Not that I can tell, of course I might be missing something.
> Would it maybe be an idea to focus on mockups and a design for a
> better UX with an extension?
Why don't you talk to the browser vendors and see how willing they are
to improve the client cert UI, even if you offered all the necessary
code patches or funded their engineers to work on it. (I'm quite serious
about this; others have reported hitting dead ends. Honestly, I
haven't tried it, myself.)
-- Sandro
>
> -- Sandro
>
>
>>
>> https://github.com/sandhawke/spot/blob/master/spec.md
>>
>> This document specifies an HTTP authentication mechanism suitable
>> for use in situations where the HTTP client is tightly coupled
>> with another HTTP server. It is very easy to implement and
>> requires no extra crypto.
>>
>> For illustration purposes, we'll say Alice is a process which
>> serves the web resource at http://alice.example/alice and wants
>> to act as a client to access a protected resource
>> http://bob.example/bob, which is served by a different process, Bob.
>>
>> For non-trivial use, to provide some resistance against attackers
>> who can view or intercept network traffic or subvert the DNS,
>> HTTPS URIs should be used, and clients should check that the
>> server DNS name matches the certificate.
>>
>>
>> Status
>>
>> Not yet implemented.
>>
>> Before wide deployment, the new authentication type
>> Page-Owner-Token and the two new HTTP headers
>> (Page-Owner-Token-Check and Page-Owner-Token-OK) should be
>> registered with the IETF.
>>
>>
>> Walkthrough
>>
>> *Step 1.* Alice performs an HTTP GET on an access-controlled page
>> served by Bob, but does not authenticate herself, so Bob returns
>> a 401 error. The response includes a header telling Alice she can
>> authenticate using this protocol and try again.
>>
>> |> GET /bob HTTP/1.1
>> > Host: bob.example
>> ...
>> < HTTP/1.1 401 Authorization Required
>> < WWW-Authenticate: Page-Owner-Token
>> ...
>> |
>>
>> This uses the standard WWW-Authenticate HTTP header with a new
>> keyword for this new authentication type.
>>
>> *Step 2.* Alice generates a cryptographicly random token, a
>> nonce. In this example, I'll write it as xyz123. It should use
>> only the Base64 characters.
>>
>> *Step 3.* Alice repeats the GET, this time including a header
>> which identifies her via a web page and includes the nonce:
>>
>> |> GET /bob HTTP/1.1
>> > Host: bob.example
>> > Authorization: Page-Owner-Token client="http://alice.example/alice" token="xyz123"
>> ...
>> |
>>
>> ISSUE: Should it just be http://alice.example/ ? What does it
>> mean to include the /alice?
>>
>> Alice can include multiple headers like this, since it might be
>> that only one of her multiple identies actually has access to
>> /bob and she doesn't now which. The identity strings must be
>> dereferenceable. They can be either an Information resource IRI
>> (returning 200 OK) or a non-information resource IRI (returning
>> 303, or having a hash). Either works fine for this protocol.
>>
>> Note that sending identity strings like this may reveal more to
>> Bob than desirable.
>>
>> *Step 4.* Bob checks to see if the request provides a valid token:
>>
>> |> HEAD /alice HTTP/1.1
>> > Host: alice.example
>> > Page-Owner-Token-Check: token="xyz123" relying-party="http://bob.example/bob"
>> ...
>> |
>>
>> The verb could be GET (instead of HEAD) if the Bob is interested
>> in the content of /alice.
>>
>> *Step 5.* Alice confirms:
>>
>> |< HTTP/1.1 200 OK
>> < Page-Owner-Token-OK: true
>> < Set-Cookie: [whatever, optional]
>> ...
>> |
>>
>> Alice only does this if the token was in fact the one she
>> generated for her GET to Bob.
>>
>> Only a response containing the header "Page-Owner-Token-OK: true"
>> is taken as confirmation.
>>
>> The Set-Cookie is an optional shortcut. With this cookie, Alice
>> can give Bob some secret to use in future communications, so that
>> Bob can act as an HTTP client accessing alice.example in the
>> future without needing to go through his own SPOT handshake.
>>
>> *Step 6.* Bob returns the protected content requested in Step 3
>>
>> |< HTTP/1.1 200 OK
>> < Set-Cookie: [whatever, optional]
>> ... content ...
>> |
>>
>> The Set-Cookie is an optional shortcut. With this cookie, Bob can
>> give Alice some secret to use in future communications, so that
>> Alice and Bob do not have to repeat this handshake every time.
>>
>>
>
>
Received on Wednesday, 19 November 2014 02:49:20 UTC