Return-Path: Delivered-To: apmail-geronimo-activemq-dev-archive@www.apache.org Received: (qmail 28179 invoked from network); 3 Aug 2006 14:30:29 -0000 Received: from hermes.apache.org (HELO mail.apache.org) (209.237.227.199) by minotaur.apache.org with SMTP; 3 Aug 2006 14:30:29 -0000 Received: (qmail 7878 invoked by uid 500); 3 Aug 2006 14:30:29 -0000 Delivered-To: apmail-geronimo-activemq-dev-archive@geronimo.apache.org Received: (qmail 7854 invoked by uid 500); 3 Aug 2006 14:30:29 -0000 Mailing-List: contact activemq-dev-help@geronimo.apache.org; run by ezmlm Precedence: bulk List-Help: List-Unsubscribe: List-Post: List-Id: Reply-To: activemq-dev@geronimo.apache.org Delivered-To: mailing list activemq-dev@geronimo.apache.org Received: (qmail 7840 invoked by uid 99); 3 Aug 2006 14:30:29 -0000 Received: from asf.osuosl.org (HELO asf.osuosl.org) (140.211.166.49) by apache.org (qpsmtpd/0.29) with ESMTP; Thu, 03 Aug 2006 07:30:29 -0700 X-ASF-Spam-Status: No, hits=2.8 required=10.0 tests=DNS_FROM_RFC_ABUSE,DNS_FROM_RFC_POST,DNS_FROM_RFC_WHOIS X-Spam-Check-By: apache.org Received-SPF: pass (asf.osuosl.org: local policy) Received: from [68.142.206.239] (HELO smtp106.plus.mail.mud.yahoo.com) (68.142.206.239) by apache.org (qpsmtpd/0.29) with SMTP; Thu, 03 Aug 2006 07:30:27 -0700 Received: (qmail 49693 invoked from network); 3 Aug 2006 14:30:06 -0000 DomainKey-Signature: a=rsa-sha1; q=dns; c=nofws; s=s1024; d=yahoo.com; h=Received:Mime-Version:In-Reply-To:References:Content-Type:Message-Id:Content-Transfer-Encoding:From:Subject:Date:To:X-Mailer; b=WPj5+N9xT7QrRv4BgKJKefL9IkYrdmZe1hezJ59AAwqR5/nyXV3llP4VE5TRm2NkryvkmpqvChGguLHQzRKvK9WmsN/BBY/HOr4ZkagpVEQ4eb0wfMRS2GxR4d62MsG9KAsW6KvyfekEwmTYcQohP7r5KqWJJyzcam6vrqgeJn0= ; Received: from unknown (HELO ?10.11.55.36?) (david?jencks@63.105.20.225 with plain) by smtp106.plus.mail.mud.yahoo.com with SMTP; 3 Aug 2006 14:30:06 -0000 Mime-Version: 1.0 (Apple Message framework v749.3) In-Reply-To: References: <5612820.post@talk.nabble.com> <5617424.post@talk.nabble.com> <5619195.post@talk.nabble.com> Content-Type: text/plain; charset=US-ASCII; delsp=yes; format=flowed Message-Id: <89459BA6-8759-4B7C-8FD3-A91CD1371DB9@yahoo.com> Content-Transfer-Encoding: 7bit From: David Jencks Subject: Re: Creating a secure connection system and using JMSXUserID support Date: Thu, 3 Aug 2006 07:30:04 -0700 To: activemq-dev@geronimo.apache.org X-Mailer: Apple Mail (2.749.3) X-Virus-Checked: Checked by ClamAV on apache.org X-Spam-Rating: minotaur.apache.org 1.6.2 0/1000/N On Aug 2, 2006, at 4:35 PM, Sepand M wrote: > Ok. So from what I've read, we're trying to separate the authorization > part from the SSL socket and let the broker handle it. > This sounds great. It would take more work (not so great since I have > a deadline), but it would be a "proper" solution. > From what I know of JAAS, the subject/principals fully represent > identity. So attaching them to Connection info would be a good idea. > That way, the Transport's job would be to authenticate and the Broker > could handle authorization completely. This would also mean that any > communication system could be used without having to change the Broker > (as long as the Transport can authenticate and create proper > subjet/principals). > > The one thing I will note is that we are changing the ActiveMQ > architecture in that currently, the Brokers are doing both > authentication and authorization (e.g. The Brokers are currently doing > the user name and password validation). > I think, however, that this is necessary because without our change, > there would need to be a new broker for every new, authenticated, > communication system. > > Please tell me if you agree (in which case I'll start looking at > implementation details). I thought about this some more and wonder if it would be appropriate to consider using the full corba csiv2 framework or a reasonable facsimile. I think it would. Very briefly here is what it gives you: -both ends of the communication channel have policies specifying what they can provide and require for security: they negotiate a common policy. The policy can specify: - security level of channel (basically unprotected or ssl) - validation of identity of server and client, by user/pw or certificate chain - validation of identity of user by a variety of mechanisms including user/pw, cert chain, and trusting the client. Perhaps the most important point here is that the identity of the user and the client may not be the same: this would typically be the case when the client is a broker forwarding messages it got from somewhere else. Here the ssl connection with client certificates would be used to establish the identity of both brokers but the user who originated the message would also be communicated and used in authorization decisions. In other situations where the message sender is a single-user client the client certificate can be used to establish the identity of both the client and user. thanks david jencks > > On 8/2/06, David Jencks wrote: >> I'm confused by the descriptions of this approach, and don't >> understand what is being proposed. I would separate the steps of >> >> 1. validating the client certificate based on the presented >> certificate chain, which in my experience can be done by the standard >> truststore implementation that comes with java, and serves to >> identify the client: this is done during the ssl connection setup >> >> and >> >> 2. deciding if the identified client is someone you want to let into >> the system, which can be done with a JAAS login module that accepts >> either a certificate chain callback handler (probably way overkill), >> the client certificate (possibly overkill, we've already verified the >> validity of the chain), or just the DN. Keeping the DN in LDAP >> should be no problem, perhaps mapped to the principals you want the >> user to have: I think this could be done after the ssl connection is >> set up >> >> and >> >> 3. deciding what permissions the logged in user should get. You >> might want to consider using a JACC like approach: I set up something >> like this for portal permissions in jetspeed2 and suspect something >> similar ought to work for amq. >> >> many thanks >> david jencks >> >> On Aug 2, 2006, at 12:20 PM, Hiram Chirino wrote: >> >> > Hi, >> > >> > On 8/2/06, ngcutura wrote: >> >> >> >> Hmmm... It didn't cross my mind but yes, indeed, it is possible. >> >> >> >> We may supply a fake truststore that would return 'true' for any >> >> certificate >> >> submitted for authentication and then perform real authentication >> >> after >> >> connection setup. We would then be able to obtain client >> >> certificate exactly >> >> as you stated. >> >> >> >> If we accept this approach, I see three components to implement: >> >> >> >> 1. Fake truststore >> >> 2. CertificateLoginModule (against LDAP) >> >> 3. Tweak connection setup to ask for peer certificates. >> >> >> >> In 3. we actually need some kind of policy reagarding >> authenitcation. >> >> Although SSL connection is established, a client may still supply >> >> username/password meaning that it should be used for login. I >> >> guess that >> >> obtaining client certificate from SSL session should be the last >> >> option. >> >> >> >> In 'Certificate login' thread I described another approach: >> >> >> >> We may use SSL without client authentication but find a way to >> export >> >> certificate to a String (on client side) and then supply that >> >> string as >> >> 'username' in createConnection(). On server side, the String >> would be >> >> converted back to certificate and authenticated. With this >> >> approach, we need >> >> to agree on the string format and conversion discipline and >> then only >> >> another JAAS login module is required (that would actually perform >> >> coversion >> >> from String to Certificate and authenticate). Thus no change is >> >> required in >> >> existing code. We may even add another non-portable >> >> createConnection(Certificate, brokerURL) that would convert >> >> Certificate to >> >> String and invoke createConnection(username, password, brokerURL). >> >> So, the >> >> necessary modules to implement would be: >> >> >> >> 1. Utility to convert Certificate to a string and back. >> >> 1a. (optional) createConnection(Certificate, brokerURL) method and >> >> ActiveMQConnection(Certifcate, brokerURL) constructor that perform >> >> conversion from Certificate to String using utility in #1 and >> invoke >> >> appropriate existing meothods/constructors. >> > >> > This sounds fine to me too. I would use the DN as the userId and >> > encode the certificate as a string for the password and add a >> > ActiveMQConnectionFactory.setCertificate( Certificate ) and have >> that >> > set userid/password. >> > >> >> 2. JAAS login module that accepts username (and blank password; or >> >> whatever >> >> convenient) converts it back to Certificate using utility in #1 >> and >> >> authenticates it. >> >> >> > >> > Yep, sounds good to me. BTW, how easy is it to get Certificate >> > instance? Is this susceptible to spoofing? >> > >> >> >> >> I didn't like this approach at first but now it seems the >> >> quickiest (and the >> >> dirtiest) solution. Actually, it is developing a new protocol on >> >> exisitng >> >> facilities. >> >> >> >> Any thoughts? >> >> >> >> Regards, >> >> NGC >> >> >> >> >> >> Hiram Chirino wrote: >> >> > >> >> > I guess I don't understand what you mean by #2 but that could be >> >> due >> >> > to my ignorance of the SSL socket stuff. So perhaps you can >> >> help me >> >> > understand what happens there... >> >> > >> >> > Lets assume we setup the ssl stuff to use 'need client auth'. >> >> Could >> >> > we setup a truststore implementation that accepts any client >> >> > certificate or would this be a problem? >> >> > >> >> > Can you later get use the SSLScoket.getSession >> >> ().getPeerCertificates() >> >> > when the ConnectionInfo command comes in and attach those >> >> certificates >> >> > to the command? >> >> > >> >> > Could that Certificate[] later be used against an LDAP JAAS >> module? >> >> > >> >> > Regards, >> >> > Hiram >> >> > >> >> > On 8/2/06, ngcutura wrote: >> >> >> >> >> >> >> >> >> Hiram Chirino wrote: >> >> >> > >> >> >> > On 8/2/06, ngcutura wrote: >> >> >> >> >> >> >> >> Hi, >> >> >> >> >> >> >> >> I started another thread, unaware of this one, with the same >> >> aim. >> >> >> >> >> >> >> >> http://www.nabble.com/forum/ViewPost.jtp? >> post=5583011&framed=y >> >> >> >> >> >> >> >> So please allow me to share my views on this. >> >> >> >> >> >> >> >> If we are going to use SSL and SSL's built in client >> >> authentication, >> >> >> then >> >> >> >> I >> >> >> >> would use JAAS to authenticate the user via certificate. I >> >> would use >> >> >> LDAP >> >> >> >> to >> >> >> >> store and verify certificates and I guess It would be fairly >> >> easy to >> >> >> >> implement. There is already LDAPLoginModule and I >> implemented >> >> >> >> LDAPAuthorizationMap - cerificates should not be much >> harder. >> >> >> >> >> >> >> > >> >> >> > Sounds good! >> >> >> > >> >> >> >> The outcome of successful SSL client authentication >> should be >> >> >> >> authenticated >> >> >> >> Subject with all Princiapls set. This I woud put into >> >> ConnectionInfo - >> >> >> no >> >> >> >> need for DN or username. When AMQ has authenticated Subject, >> >> it can >> >> >> >> perform >> >> >> >> authorization in any of the existing ways. That is, we can >> >> safely >> >> >> >> separate >> >> >> >> authentication from authorization modules as long as AMQ >> >> gets Subject >> >> >> >> from >> >> >> >> the authentication process. >> >> >> >> >> >> >> > >> >> >> > agreed. >> >> >> > >> >> >> >> What I miss here is the point of Subject creation. If we >> >> totally rely >> >> >> on >> >> >> >> SSL >> >> >> >> for authentication we actually need an implementation of >> >> truststore >> >> >> >> (keystore with trust manager) that would verify client >> >> certificate and >> >> >> >> create login Subject. However, as this process is totally >> >> hidden from >> >> >> AMQ >> >> >> >> (I >> >> >> >> think that truststore and ConnectionInfo instance are >> >> unaware of each >> >> >> >> other), we would need another store (directory) to >> >> temporarrily save >> >> >> >> Subject >> >> >> >> and make it avaliable to AMQ once the connection is created. >> >> Or, if >> >> >> there >> >> >> >> is >> >> >> >> a way for truststore to interact with ConectionInfo >> >> instance, this >> >> >> >> problem >> >> >> >> is solved. >> >> >> > >> >> >> > I'm not familiar with the SSL details to get this done, so I >> >> may be >> >> >> > talking alot of BS here.... But it sounds like your saying >> >> that we >> >> >> > need associate our keystore with the SSL layer. But we want >> >> to store >> >> >> > our certs in LDAP. And right now those two layers would >> >> communicate >> >> >> > via a ConnectionInfo object. So I would say that in this >> >> case the >> >> >> > user is authenticating/logging in earlier than in normal >> >> cases. since >> >> >> > he is authentcating at connection setup time, I think you >> >> would need >> >> >> > to do the JAAS log in when the connection is estbalished. >> >> And attach >> >> >> > the JAAS subject to the ConnectionInfo. >> >> >> > >> >> >> > ---REPLY BEGINS--- >> >> >> > (I don't know how to produce '>' when quoting using the Web >> >> interface >> >> >> on >> >> >> > Nabble.) >> >> >> > >> >> >> > Your understanding is compatible with mine. :-) What I >> >> undestood from >> >> >> JSSE >> >> >> > is that it is actually a component that you may configure >> >> independantly >> >> >> of >> >> >> > the rest of the application. You specify a factory and a >> >> truststore and >> >> >> > connection is returned. SSL server and client authentication >> >> based on >> >> >> > certificates is configurable but totally hidden from the >> >> application. >> >> >> What >> >> >> > we can do to interfere is implement SSLSocketFactory and >> >> implement >> >> >> > truststore. >> >> >> > >> >> >> > Now, if we bypass client authentication during SSL handshake >> >> and leave >> >> >> it >> >> >> > until the connection is established, the question is how we >> >> obtain >> >> >> client >> >> >> > ceritifcate. If the client is not required to authenticate >> >> during SSL >> >> >> > handshake, it will not send its certificate to the server and >> >> we lose >> >> >> > possibility to perform client certificate authentication. >> >> Thus we need >> >> >> to >> >> >> > set 'need client auth' or 'want client auth' property to the >> >> server >> >> >> SSL >> >> >> > socket factory. (I saw a discussion thread where this was >> >> resolved in >> >> >> > AMQ.) In both cases, I believe (although I am not sure) >> client >> >> >> > ceritificate authentication wil be attempted. In case of >> 'need' >> >> >> > unsuccessful authentication will disconnect client while in >> >> the case of >> >> >> > 'want' connection will be created. Maybe this can be used in >> >> our case >> >> >> but >> >> >> > I am not sure how 'want' case is handled exactly. If there >> >> are any >> >> >> > restrictions imposed on such a connection, we lose the case. >> >> >> > >> >> >> > Back to the normal SSL handshake: if we implement our own >> >> truststore >> >> >> (we >> >> >> > need to) and our own SSL socket factory (we may) and create >> >> >> ConnectionInfo >> >> >> > instance before the actual connection (I am now talking >> >> unaware of the >> >> >> > actual code in AMQ - I have not studied it yet) maybe there >> >> would be a >> >> >> way >> >> >> > to pass ConnectionInfo instance to the custom SSL socket >> >> factory which >> >> >> > would then pass it to the custom truststore and we are in >> >> business. >> >> >> > >> >> >> > Some gymnastics, admitedly. :-) >> >> >> > >> >> >> > What we need to resolve is this: >> >> >> > >> >> >> > 1. In case of 'wantClientAuth' what are the consequences of >> >> >> unsuccessful >> >> >> > client authentication? Is the connection as good as >> >> authenticated or >> >> >> are >> >> >> > there any restrictions? >> >> >> > >> >> >> > 2. Is there a way to pass ConnectionInfo instance via factory >> >> to the >> >> >> > truststore as suggested? >> >> >> > >> >> >> > Answers to the above questions would give us a way to go. >> >> Still, if >> >> >> there >> >> >> > would be a positive answer to the question 2. I would go that >> >> way >> >> >> > regardless of the 1. Therefore, I volounteer to resolve >> 2. :-) >> >> >> > >> >> >> > Any ideas? >> >> >> > >> >> >> > Regards, >> >> >> > NGC >> >> >> > ---REPLY ENDS--- >> >> >> >> >> >> >> >> This approach requires implementation of >> >> CertificateLoginModule (JAAS) >> >> >> >> and >> >> >> >> custom truststore that would use this login module plus some >> >> temporary >> >> >> >> map. >> >> >> >> >> >> >> >> What do you thik about this? >> >> >> >> >> >> >> >> Regards, >> >> >> >> NGC >> >> >> >> >> >> >> >> >> >> >> >> Hiram Chirino wrote: >> >> >> >> > >> >> >> >> > On 8/1/06, Sepand M wrote: >> >> >> >> >> Hi all, >> >> >> >> >> >> >> >> >> >> So far I've mainly been reading ActiveMQ and making >> >> design docs. >> >> >> >> >> Here's what I've got: >> >> >> >> >> >> >> >> >> >> For authorization, my current plan is to just have the >> >> client's DN >> >> >> >> >> replace the user name field in the ConnectionInfo class >> >> (how this >> >> >> is >> >> >> >> >> done is explained below). I want to do this because I >> >> don't know >> >> >> much >> >> >> >> >> about JAAS and I'm trying to avoid writing classes to >> >> authorize >> >> >> based >> >> >> >> >> on DNs. If you guys know this stuff (and you probably >> >> do), we could >> >> >> >> >> change this easily enough. >> >> >> >> >> >> >> >> >> >> Here's the rest of my design: >> >> >> >> >> >> >> >> >> >> I want to modify SslTransportFactory to use a specific >> >> SslContext >> >> >> >> >> object and allow client's access to its init method so >> >> that they >> >> >> can >> >> >> >> >> set their own key and trust managers. I also want to >> >> create new >> >> >> >> >> SslTransport and SslTransportServer classes. SslTransport >> >> will be >> >> >> >> >> derived from TcpTransport. Its main task will be to >> >> replace the >> >> >> user >> >> >> >> >> name field of ConnectionInfo commands with its socket's >> >> DN (this >> >> >> could >> >> >> >> >> be changed easily to attach the entire certificate to >> >> >> ConnectionInfo >> >> >> >> >> as a new generic field). SslTransport will also make sure >> >> that it >> >> >> uses >> >> >> >> >> SslSocketFactory's. SslTransportServer will only be there >> >> to make >> >> >> sure >> >> >> >> >> SslSocketFactory's are used. >> >> >> >> >> >> >> >> >> >> For my current design that about does it. The proper >> >> Brokers and >> >> >> >> >> plugins (JaasAuthenticationBroker and >> >> AuthorizationPlugin) would >> >> >> have >> >> >> >> >> to be used and the configuration files would need to use >> >> the DN as >> >> >> the >> >> >> >> >> username. >> >> >> >> >> >> >> >> >> >> I'm not sure about this, but I think if we were to attach >> >> the >> >> >> complete >> >> >> >> >> certificate and try to do things "properly" we'd need >> a new >> >> >> >> >> CertificateAuthenticationBroker and a way for JAAS to >> >> authenticate >> >> >> >> >> that certificate (I'm new to JAAS so I don't know how >> >> easy/hard >> >> >> this >> >> >> >> >> would be). >> >> >> >> >> >> >> >> >> > >> >> >> >> > Sounds spot on! The JAAS part would totally depend on how >> >> the JAAS >> >> >> >> > module that authenticates against a certificate expects to >> >> receive >> >> >> the >> >> >> >> > certificate. Right now our current JAAS login only uses >> >> >> >> > userid/password, that would need to change for a cert. >> >> Anybody know >> >> >> >> > where we can get a JAAS module that authenticates >> >> certificates? >> >> >> >> > >> >> >> >> > Regards, >> >> >> >> > Hiram >> >> >> >> > >> >> >> >> >> Any thoughts? >> >> >> >> >> - Sepand >> >> >> >> >> >> >> >> >> >> On 8/1/06, James Strachan >> wrote: >> >> >> >> >> > On 8/1/06, ngcutura wrote: >> >> >> >> >> > > >> >> >> >> >> > > My JIRA username is 'ngcutura' and I'll be glad to >> >> assign LDAP >> >> >> >> >> Authorization >> >> >> >> >> > > issue to myself. >> >> >> >> >> > >> >> >> >> >> > Great! You're all set now with JIRA karma >> >> >> >> >> > >> >> >> >> >> > > I also take this opportunity to remind you of my code >> >> >> >> >> > > waiting for your review. :-) >> >> >> >> >> > >> >> >> >> >> > Thanks for the reminder - will try get there soon :) >> >> >> >> >> > >> >> >> >> >> > > I wouldn't mind creating and assigning certificate >> >> login but as >> >> >> >> >> Sepand was >> >> >> >> >> > > the first to raise it I'd wait for him (a while). >> >> >> >> >> > >> >> >> >> >> > Coolio >> >> >> >> >> > >> >> >> >> >> > -- >> >> >> >> >> > >> >> >> >> >> > James >> >> >> >> >> > ------- >> >> >> >> >> > http://radio.weblogs.com/0112098/ >> >> >> >> >> > >> >> >> >> >> >> >> >> >> > >> >> >> >> > >> >> >> >> > -- >> >> >> >> > Regards, >> >> >> >> > Hiram >> >> >> >> > >> >> >> >> > Blog: http://hiramchirino.com >> >> >> >> > >> >> >> >> > >> >> >> >> -- >> >> >> >> View this message in context: >> >> >> >> >> >> >> http://www.nabble.com/Creating-a-secure-connection-system-and- >> >> using-JMSXUserID-support-tf1956575.html#a5612820 >> >> >> >> Sent from the ActiveMQ - Dev forum at Nabble.com. >> >> >> >> >> >> >> >> >> >> >> > >> >> >> > >> >> >> > -- >> >> >> > Regards, >> >> >> > Hiram >> >> >> > >> >> >> > Blog: http://hiramchirino.com >> >> >> > >> >> >> > >> >> >> -- >> >> >> View this message in context: >> >> >> http://www.nabble.com/Creating-a-secure-connection-system-and- >> >> using-JMSXUserID-support-tf1956575.html#a5617424 >> >> >> Sent from the ActiveMQ - Dev forum at Nabble.com. >> >> >> >> >> >> >> >> > >> >> > >> >> > -- >> >> > Regards, >> >> > Hiram >> >> > >> >> > Blog: http://hiramchirino.com >> >> > >> >> > >> >> -- >> >> View this message in context: http://www.nabble.com/Creating-a- >> >> secure-connection-system-and-using-JMSXUserID-support- >> >> tf1956575.html#a5619195 >> >> Sent from the ActiveMQ - Dev forum at Nabble.com. >> >> >> >> >> > >> > >> > -- >> > Regards, >> > Hiram >> > >> > Blog: http://hiramchirino.com >> >>