Got it, for now I'm just going to work on implementing a Kerberos solution that either allows you to configure a keytab and principal or pulls from the current subject if your already logged in.
I created NIFI-6378 and NIFIREG-281. Can one of you assign the registry one to me as I'm not a contributor there yet. Thanks Shawn On 6/13/19, 2:03 PM, "Bryan Bende" <[email protected]> wrote: It would be a little bit weird because you'd still need the client cert for the initial request to get the JWT, so then in that case why not just keep using the client cert. Registry does things a little bit different than NiFi and has a few variations of the token end-point: /access/token/login (looks for credentials using basic auth) /access/token/kerberos (same as NiFi) /access/token/identity-provider (passes request to the configured identity provider) /access/token (tries all identity providers in order, the first of which is X509 identity provider) So if you sent a client cert to the last one, it would do what you are suggesting. On Thu, Jun 13, 2019 at 2:55 PM Andy LoPresto <[email protected]> wrote: > > No, client cert authentication bypasses the JWT behavior completely. Because a client cert is automatically sent on every request, it makes no sense to delegate the credential to a token in that case. > > Andy LoPresto > [email protected] > [email protected] > PGP Fingerprint: 70EC B3E5 98A6 5A3F D3C4 BACE 3C6E F65B 2F7D EF69 > > > On Jun 13, 2019, at 11:52 AM, Shawn Weeks <[email protected]> wrote: > > > > Completely agree on username and password but I'll probably still do something somewhat generic around access tokens vs 2 way ssl as in the future there might be something else. On a related note is it possible to get a JWT with 2 way ssl? If so we could use the same auth method for everything. > > > > Thanks > > Shawn > > > > On 6/13/19, 1:36 PM, "Bryan Bende" <[email protected]> wrote: > > > > Ah thanks for pointing that out, I completely forgot that apparently I > > was thinking ahead in the JerseyNiFiClient of how we could support > > tokens :) > > > > You would need to make the same changes in the > > JerseyNiFiRegistryClient, and then build a new toolkit based on a new > > version of nifi-registry-client. > > > > Also, you are correct that we could support username/password, but I > > think Kerberos is much better from a security perspective since you > > don't really need to give your credentials to the CLI. With > > username/password, you would either need to add those properties to > > the .props files for the CLI, which then gets into encrypting the > > password, or you need to provide them on a command as arguments which > > again gets into whether the password is in plain text or not. > > > > On Thu, Jun 13, 2019 at 2:28 PM Shawn Weeks <[email protected]> wrote: > >> > >> Got it, I've been trying to read some of this on my phone and missed something. Currently it looks like the NiFi Client JerseyNiFiClient.java was setup to support token(JWT) based requests but from what I can tell those methods are never called anywhere. NiFi Registry Client only implemented the implicit security and proxied entity methods. > >> > >> It looks like I should be able to lookup the auth token and add it to the Jersey WebTarget Headers in the two clients so it would be there on every request. I'll have to do some testing but that might not require too many changes. In theory it could also support username/password auth as well doing it the same way. > >> > >> https://stackoverflow.com/questions/29056051/adding-authorization-header-to-jersey-sse-client-request > >> > >> Thanks > >> Shawn > >> > >> On 6/13/19, 1:04 PM, "Bryan Bende" <[email protected]> wrote: > >> > >> I'm not sure if I confused things... the clients that I mentioned are > >> wrappers for the REST API implemented with Jersey client, so the CLI > >> does exclusively use the REST API. > >> > >> I was just drawing attention to the clients to say that part of the > >> work is outside of the CLI in nifi-registry-client to allow it to > >> support kerberos auth. > >> > >> On Thu, Jun 13, 2019 at 1:54 PM Shawn Weeks <[email protected]> wrote: > >>> > >>> Ok, I was thinking the CLI used the Rest API exclusively and that's what I was missing. Unfortunately I don't have the option to use self-signed certificate due to organizational security policies and we don't have a way to get SSL Certificates issued to individuals only servers. > >>> > >>> Thanks > >>> Shawn > >>> > >>> On 6/13/19, 12:30 PM, "Bryan Bende" <[email protected]> wrote: > >>> > >>> Just to further elaborate, within the CLI there are commands that work > >>> against registry and commands that work against NiFi. For registry > >>> commands, they use the Java client that is provided by registry [1]. > >>> For NiFi commands, there is mini client developed as need with in the > >>> CLI [2]. > >>> > >>> None of these client calls currently have any concept of a JWT/token. > >>> > >>> In order to do the kerberos auth correctly across both systems, I > >>> think both of these clients would need to be updated to support a > >>> method that called the /access/kerberos end point to obtain a token, > >>> and then also provide a way to pass back that token on future > >>> requests. It would likely be the CLI's job to store that token > >>> somewhere (in memory for interactive shell, or on filesystem for > >>> individual executions) and pass it back on each request. In order to > >>> call the /access/kerberos end-point there also needs to be code in the > >>> client that handles the negotiation to provide the kerberos > >>> credentials that are present from having done a kinit. > >>> > >>> Long story short, Andy's first suggest would be a much easier option > >>> with no code changes. > >>> > >>> [1] https://github.com/apache/nifi-registry/tree/master/nifi-registry-core/nifi-registry-client > >>> [2] https://github.com/apache/nifi/tree/master/nifi-toolkit/nifi-toolkit-cli/src/main/java/org/apache/nifi/toolkit/cli/impl/client > >>> > >>> On Thu, Jun 13, 2019 at 1:28 PM Andy LoPresto <[email protected]> wrote: > >>>> > >>>> You’ll probably have to write (minimal) code to expose the ClientBuilder constructor/factory methods to the part that parses command-line arguments. > >>>> > >>>> Andy LoPresto > >>>> [email protected] > >>>> [email protected] > >>>> PGP Fingerprint: 70EC B3E5 98A6 5A3F D3C4 BACE 3C6E F65B 2F7D EF69 > >>>> > >>>>> On Jun 13, 2019, at 10:27 AM, Shawn Weeks <[email protected]> wrote: > >>>>> > >>>>> Is there a way to pass 2 currently? Because you can get the token via curl like I’m currently doing? > >>>>> > >>>>> Thanks > >>>>> Shawn > >>>>> > >>>>> Sent from my iPhone > >>>>> > >>>>>> On Jun 13, 2019, at 12:21 PM, Andy LoPresto <[email protected]> wrote: > >>>>>> > >>>>>> I see a couple choices here: > >>>>>> > >>>>>> 1. Use the CA to generate and sign a new certificate for deployments. This certificate would not be as sensitive as the server certificate, as you can put stricter permissions on that identity within the NiFi access controls, and the cert would be issued for a DN that cannot be used to impersonate the server itself. Use this certificate to authenticate for deployment activities. > >>>>>> 2. Manually extract the user’s JWT from the Developer Tools in your browser and pass that into the CLI. This token expires regularly, so you will need to continually update it. > >>>>>> 3. Build the Kerberos implementation of the authentication aspects of the CLI toolkit. > >>>>>> > >>>>>> Andy LoPresto > >>>>>> [email protected] > >>>>>> [email protected] > >>>>>> PGP Fingerprint: 70EC B3E5 98A6 5A3F D3C4 BACE 3C6E F65B 2F7D EF69 > >>>>>> > >>>>>>> On Jun 13, 2019, at 10:00 AM, Shawn Weeks <[email protected]> wrote: > >>>>>>> > >>>>>>> For our organization the server certificate is considered sensitive and not available to the users who need to deploy to NiFi. Actual authentication to NiFi is handled through Knox and our SSO Service so the end user never deals with SSL or has access to a certificate. Originally I started down the path of writing a bunch of tools based on NiPyAPI to handle deployments but since the CLI already does that I was hoping to save some work. Currently we do several other things via rest using the Kerberos Token. > >>>>>>> > >>>>>>> As I looked through the tool kit CLI I was seeing that auth token being passed into all the rest calls so I was hoping I could hijack wherever that was being generated via 2way ssl and add an option to call Kerberos instead to get the token. When I say token I mean the auth bearer token that you can get from a post request to /access/kerberos in NiFi and /access/token/Kerberos in NiFi registry. > >>>>>>> > >>>>>>> Thanks > >>>>>>> Shawn > >>>>>>> > >>>>>>> On 6/12/19, 12:06 PM, "Bryan Bende" <[email protected]> wrote: > >>>>>>> > >>>>>>> I meant to say that you obviously could generate certs for CLI users, but I > >>>>>>> was just mentioning an alternative where you can proxy an identity. > >>>>>>> > >>>>>>> Right now the CLI never obtains a token because it is all cert based. > >>>>>>> > >>>>>>>> On Wed, Jun 12, 2019 at 1:03 PM Bryan Bende <[email protected]> wrote: > >>>>>>>> > >>>>>>>> Right now the idea is that whoever is running the CLI would have access to > >>>>>>>> a NiFi server certificate and then you can proxy any user you want. There > >>>>>>>> should be examples of this in the readme or toolkit guide. > >>>>>>>> > >>>>>>>> Supporting Kerberos auth was something I wanted to do, but it’s definitely > >>>>>>>> not a trivial effort. > >>>>>>>> > >>>>>>>> On Wed, Jun 12, 2019 at 12:57 PM Andy LoPresto <[email protected]> > >>>>>>>> wrote: > >>>>>>>> > >>>>>>>>> Shawn, > >>>>>>>>> > >>>>>>>>> I’m not sure I understand your question. > >>>>>>>>> > >>>>>>>>> I am in the process of refactoring the TLS Toolkit to integrate with > >>>>>>>>> public certificate authorities, so in the near future it will be easier to > >>>>>>>>> use certificates signed by external authorities rather than self-signed. > >>>>>>>>> > >>>>>>>>> My understanding is that you are talking about the CLI Toolkit rather > >>>>>>>>> than the TLS Toolkit, but your reference to “token” was ambiguous, so I’m > >>>>>>>>> going to proceed with the understanding that you are referring to the JWT > >>>>>>>>> token used to identify an authenticated user when communicating with the > >>>>>>>>> NiFi API. > >>>>>>>>> > >>>>>>>>> You may want to look at JerseyNiFiClient [1], which has methods for > >>>>>>>>> getting various clients given an authentication token. > >>>>>>>>> > >>>>>>>>> You can create the token via the POST /access/kerberos API [2]. > >>>>>>>>> > >>>>>>>>> [1] > >>>>>>>>> https://github.com/apache/nifi/blob/master/nifi-toolkit/nifi-toolkit-cli/src/main/java/org/apache/nifi/toolkit/cli/impl/client/nifi/impl/JerseyNiFiClient.java#L163 > >>>>>>>>> < > >>>>>>>>> https://github.com/apache/nifi/blob/master/nifi-toolkit/nifi-toolkit-cli/src/main/java/org/apache/nifi/toolkit/cli/impl/client/nifi/impl/JerseyNiFiClient.java#L163 > >>>>>>>>>> > >>>>>>>>> [2] https://nifi.apache.org/docs/nifi-docs/rest-api/index.html < > >>>>>>>>> https://nifi.apache.org/docs/nifi-docs/rest-api/index.html> > >>>>>>>>> > >>>>>>>>> Andy LoPresto > >>>>>>>>> [email protected] > >>>>>>>>> [email protected] > >>>>>>>>> PGP Fingerprint: 70EC B3E5 98A6 5A3F D3C4 BACE 3C6E F65B 2F7D EF69 > >>>>>>>>> > >>>>>>>>>> On Jun 12, 2019, at 9:39 AM, Shawn Weeks <[email protected]> > >>>>>>>>> wrote: > >>>>>>>>>> > >>>>>>>>>> I work in an environment reluctant to create self signed ssl > >>>>>>>>> certificates and I’m looking at the feasibility of having the toolkit cli > >>>>>>>>> authenticate via Kerberos. I was expecting it to be as simple as adding > >>>>>>>>> another way to get the authentication token but I’m having trouble figuring > >>>>>>>>> out exactly when the token is created. I see lots of references to it after > >>>>>>>>> it’s been created. > >>>>>>>>>> > >>>>>>>>>> Thanks > >>>>>>>>>> Shawn > >>>>>>>>> > >>>>>>>>> -- > >>>>>>>> Sent from Gmail Mobile > >>>>>>>> > >>>>>>> -- > >>>>>>> Sent from Gmail Mobile > >>>>>>> > >>>>>>> > >>>>>> > >>>> > >>> > >>> > >> > >> > > > > >
