W3C home > Mailing lists > Public > public-identity@w3.org > December 2011

Re: New "Goals" (use-cases) - Is your use-case there, accurately described?

From: Harry Halpin <hhalpin@w3.org>
Date: Mon, 12 Dec 2011 17:18:42 +0100
Message-ID: <4EE62962.7060706@w3.org>
To: Tom Ritter <tom@ritter.vg>
CC: "public-identity@w3.org" <public-identity@w3.org>
On 12/12/2011 02:19 AM, Tom Ritter wrote:
> I'm interested in providing a toolkit to let people build, among other
> things, web applications where I can trust the web application
> operator to behave in a "honest but coercible" fashion.
>
> Examples:
>
>   - Imagine if web mail had a couple more buttons: [Find/Import Public
> Key], [X] Encrypt Message.  Assume the issue of "Trust" is solved
> elsewhere and I import public keys I trust for my contacts, and assume
> the web mail publishes the javascript source in an auditable format,
> and it is audited and considered trusted.  I can encrypt emails and
> attachments to my contacts inside of a Web UI, and when they are
> received by the web mail operated, they cannot decrypt them.
>   - OTR in a web messaging platform (gchat, facebook)
>   - Encrypted passages in web applications, distributed ala
> alt.anonymous.messages (that is, broadcast to all) but readable by
> only a select few.  (Facebook status messages)
>   - Facebook or Gmail could look at the TLS certificate in use on the
> connection, and not expose private data if it's being middled [1]
>   - A website is nervous about state-sponsored TLS interception, so its
> users generate client certificates, and the website sends sensitive
> data to the client to be decrypted by the client certificate [2]
>   - Zed Shaw wants to get http://vulnarb.com/ going (where you encrypt
> data to the public key provided for a TLS connection)
>   - I want to build a pure-javascript browser plugin (so it works in
> FF, Chrome, Greasemonkey, etc) that looks for 'weird' TLS Connections:
> MD5-signed certs, low public exponents, debian weak keys, etc)
>   - Corporation X wants to see how many users still have key X pinned
> that they did accidentally a while ago [4] [5]
>   - Bank B wants to put in a stronger verification on high-value
> transactions, such that those transactions require a digital signature
>   - App A wants to leave some local data in Web Storage or whatever
> it's called.  But it doesn't want this to be accessible to everybody
> on the machine.  It'd like it to be encrypted using a private key that
> needs a passphrase.
All looks fun, although we have not talked about pinning or certs yet in 
detail, see below. I'd like more discussion of those use-cases.
>
> What would these need to be feasible?
>   - A javascript library (technically an entire runtime, so the
> collection of all javascript code) that is signed by a party other
> than the author (possibly me), such that non-signed code (xss,
> tampered, corrupted, etc) could not execute.
>   - OpenPGP (RFC 4880) compatibility.  Not necessarily built it, but a
> way to build it out of primitives.  So...
>   - Hash Functions (For full compatibility: MD5, SHA-1, RIPE-MD/160,
> SHA256, SHA384, SHA512, SHA224)
>   - Symmetric Algorithms (For full compatibility: IDEA, DES-EDE, CAST5,
> Blowfish, AES 128/192/256, Twofish256, Camellia 128/192/256)
>   - Asymmetric Algorithms (For full compatibility: RSA, Elgamal, DSA)
>   - Compression Algorithms (For full compatibility: Zip, Zlib, BZip2)
>   - A way to encrypt a file chosen in a<input type=file>  control prior
> to transmitting it to the server
>   - A mechanism for a User Agent to see encrypted text, look in your
> keystore (either client certificates or other) for a corresponding
> private key, decrypt the content automatically (prompting for a
> password if necessary), but not expose it to the web application via
> javascript. [3]
Looking above (starting with the "feasible" section), that all looks 
like it would be covered in the primary scope of the charter. Everything 
below (as it deals with certificates) I'm not so clear on. Earlier it 
was out of scope, but there seems to be pressure to put certificates and 
exposing other TLS functions in scope. Any feelings from the rest of the 
list?
>   - Exposing the Server Certificate (possibly structured, if not we'll
> need a bulletproof, signed, X509 library) and Path of the TLS
> Connection as javascript objects.
>   - Exposing the Client Certificate supplied up through the SSL library
> and Web Server to application code [2]
>   - Exposing Pinned Public Keys for a host via javascript [4]
>
>
> [1] I think this one continues to be information-theoretically
> impossible without a trusted side channel. But it can be made
> practically difficult to bypass by using random checks that are
> difficult to anticipate and fake.
> [2] This one is a server-side change in addition to client-side.
> [3] There's good precedence for this: A resource from a different
> origin (stylesheet, image, etc) that results in a 302 redirect does
> not expose via javascript where the redirect went to.
> [4] http://www.ietf.org/id/draft-ietf-websec-key-pinning-01.txt
> [5] This is also a potential toolkit for privacy violations
>
>
> These are all pretty much stream of consciousness.  I took a couple
> steps down the road to building each, but not much more - so there's
> definitely problems here.  And the biggest two complications I came
> across were: Key Management and Javascript Integrity.

Key management is where we will likely be headed here in some limited form.
> It's just impractical to have an entire signed javascript runtime.  We
> need components of it to be built by different people and to use those
> components in a partially unsigned runtime.  I think a realistic
> scenario is an OpenPGP library built by a third party, some
> security-critical javascript (I'm going to call this 'the crypto js'),
> and a bunch of UI javascript.  Formally, I'd say the UI javascript
> should not be able to ascertain any information about the crypto or
> openpgp js.  The crypto js should not be able to ascertain any
> information about the openpgp js, but should be able to call it.  The
> crypto and openpgp javascript should not be able to ascertain any
> information about the browser crypto operations but should be able to
> call them.  The UI javascript should not be able to call the browser
> crypto operations.  That's stream of consciousness dump, but
> basically, the runtime segregation needs a lot of thought

UI is out of scope. The larger issue of JS integrity will have to build 
from the appropriate primitives, what other primitives do you need 
besides those in primary features?
> Likewise, Key Management is weird.  What operations can be done
> automatically, what require permission, and bunch of other questions.
>

Again, this is the critical problem with the "digital signature" stuff 
in certs and the "Korean bank use-case", i.e. it will clearly require 
some UI. However, we will not normatively specify that UI. We could 
informatively say "something should happen in the UI and the user should 
authenticate themselves".
> -tom
Received on Monday, 12 December 2011 16:18:31 GMT

This archive was generated by hypermail 2.2.0+W3C-0.50 : Monday, 12 December 2011 16:18:32 GMT