<joel.guillod@...> writes:
>
> Hi!
>
> We are about to release our first version of an EHR (Electronic Healthcare
Record) build on CouchDB. Now is
> time for us to deal in more details with data security issues in order to
protect patient data,
> particularly with data encryption on the net and signed documents. We think
about two ways and are willing
> to combine both when appropriate:
> - using https (of course);
> - using encryption and documents signing with PGP public/private keys;
> - both may be used for health data.
>
> We have read about similar questions on couchdb-user-archive (e.g. "Proposal
For Storing Signatures In
> JSON" including the article and cononical-json from Chris in the Couchdb Wiki,
or the post "Building
> Erlang app around CouchDB" which stated on Apr 21, 2009 "Common features such
as authentication,
> authorization, caching, compression, partitioning, proxying, tunneling,
encryption or URI
> rewriting are possible by placing standard applications such as Apache httpd
or nginx in front of your
> server.") but most posts are 1+ year old. Did I miss a current roadmap of
CouchDb on this topic?
>
> In our case we need encryption on the client part and the server. We have all
the javascript code to build a
> library using PGP. Also, keys generation on the browser client appears fast
enough for 1024-bits keys (<5
> secs) and subsequent encryption of docs takes only a few millisecs. Our
current idea is exposed
> thereafter. We would welcome CouchDB Guru's advices and comments.
>
> The principle we intend to implement is that both communication parties (local
and remote) generate their
> own private/public session keys and send their public key to the other party.
Then, during the session any
> data can be transfered encrypted and automatically decoded by the receiver. We
are writing a javascript
> library for a sort of encryption tunnel for communicating between clients and
CouchDB. We instanciate a
> Security object (which privately generates and wraps the local private key for
the session) and then
> exposes the following functions :
> - setRemotePublicKey(remotePublicKey) : to be called at most one time.
> - getPublicKey() : returns the local public key.
> - encode(content) : returns the PGP message of a content string encoded with
the remotePublicKey.
> - decode(pgpMsg) : returns the decoded string of a PGP message which was
encoded with the local public key (getPublicKey()).
> - signDoc(couchDocument) : to add a digital signature to a document (the
signature is only valid for the
> session, so more to design here).
>
> The workflow is the following. Given S a remote instance of Security on the
CouchDB Server and C the local
> instance of Security on the web Client. The lifetime C is the session, maybe
same for S. Only public keys
> travel the net and then encrypted data. The steps are as follows:
>
> 1. local : C = new Security();
> 2. local : send cPubKey = C.getPublicKey()
> 3a. remote : receive cPubKey and execute S = new Security(cPubKey);
> 3b. remote : now the remote is able to receive data encrypted by
C.encode(content)
> 3c. remote : decode encrypted data by local with S.decode(data)
> 4. remote : send sPubKey = S.getPublicKey()
> 5. local : receive sPubKey and execute C.setRemotePublicKey(sPubKey);
> 5b. local : now local client can receive data encoded by remote party with
S.encode(content)
> 5c. local : decode remote data with C.decode(data)
>
> Alternate workflow : both local and remote instanciate simultaneously their
new Security() and send
> their getPublicKey(), when received, both execute the setRemotePublicKey()
appropriately. The
> normal sequence seems to be more natural since it allows the web client to
generate its local public key and
> send it to the remote server which create its Security and returns its own
public key as a result. Then, both
> parties know how to communicate. One problem could arise if other async
requests are pending: will they be
> encrypted or not? One way to answer is to admit that both crypted and
unencrypted data can be exchanged and
> will be recognized on either the header of the response, or the '-----BEGIN
PGP MESSAGE-----'.
>
> Even better : integrating the security on the session creation when a client
request CouchDB. The
> CouchDB's gurus could explain if CouchDB could generate its Security instance
on login or session
> initialization (config.httpd_global_handlers? config.couch_httpd_auth?) and
add it to the session
> object (e.g. session.security = {pubkey:'...', ...}). Could we get this
security feature as a standard
> or an optional plugin in CouchDB? We are ready to share our security library
and help the feature work. And
> it would be fine that the public key of the client browser be sent with
encrypted login data (which means
> that CouchDB send its public key before client login).
>
> Before going too far in developing such a security library, we would like to
know if there are alternate ways
> to deal with security and data encryption (+document signing) between the web
client and CouchDB. We are
> aware that in order to make this library transparent to users and devs we
should hook our library inside
> CouchDB in order that the data received from the client is decoded
appropriately (at some
> 'beforeProcess' event in CouchDB) and encoded before sending data to the
client (afterProcess event or
> beforeSending). Minor update of the CouchDB Ajax code will make the
encoding/decoding fully
> transparent on client's side.
>
> More has to be designed for using certificates to be used independantly of the
session duration. But the
> principles will be the same, just the way we generate or acquired a valid
certificate (i.e. the
> public/private keys) will have to change.
>
> Any comment welcome!
> Thanks,
> Joel
>
>
Hi,
I'm starting to work on a CouchDB based EMR for my MSc Dissertation project and
I'm very interested about the security questions and your work in general.
Sorry I can't answer your questions but I would really really be interested in
talking about your project which could help me greatly.
I have so many questions :D
Don't hesitate to join me please, I give my address so it's easier to talk:
[email protected]
Cheers,
Arnaud.
