You are viewing a plain text version of this content. The canonical link for it is here.

Posted to kerby@directory.apache.org by "Zheng, Kai" <ka...@intel.com> on 2016/01/11 13:29:20 UTC

FW: Kerby client library refactoring

I thought kadmin remote support discussion should originate in this thread. In case it's needed again.

Regards,
Kai

-----Original Message-----
From: Zheng, Kai [mailto:kai.zheng@intel.com] 
Sent: Tuesday, December 01, 2015 3:21 PM
To: kerby@directory.apache.org; dev@directory.apache.org
Subject: RE: Kerby client library refactoring

Thanks Steve for the response and great comments!

Thanks all for all the good ideas. It looks good to me to expose KdcOption meanwhile to have some handy shortcuts to set those four very frequently used flags, along with some other parameters. Let's get it done some time.
Note the refactoring is running slow as we have other urgent things to do. It's tracked here (some task issues to be opened yet).
https://issues.apache.org/jira/browse/DIRKRB-478

For the kadmin/kpasswd things, your proposal looks good and any further inputs are welcome. If you feel there're something we should get done first and pave the way, please let we know. Thanks.

Regards,
Kai

-----Original Message-----
From: Steve Moyer [mailto:smoyer@psu.edu]
Sent: Monday, November 30, 2015 8:58 PM
To: kerby@directory.apache.org
Subject: Re: Kerby client library refactoring

I'm in favor of separate method calls for the various "parts" of an AsRequest, but am wondering a bit about the KdcOptions as proposed by:

requestOptions.setForwardable(true);
requestOptions.setProxiable(false);
requestOptions.setRenewable(false);

FORWARDABLE, PROXIABLE AND RENEWABLE_OK are the three values that the MIT kinit program sets by default, but there are actually 15 of these flags (in a 32-bit bitmap).  We talked about separating these flags into their own enum, but perhaps a utility class would work better.  There are several options:

1)  If KDC Options were declared as public static final int in a utility class:

requestOptions.clearKdcOption(KdcOption.FORWARDABLE);
requestOptions.clearKdcOptions();
requestOptions.clearKdcOptions(KdcOption.FORWARDABLE | KdcOption.PROXIABLE | KdcOption.RENEWABLE_OK); requestOptions.setKdcOption(KdcOption.FORWARDABLE);
requestOptions.setKdcOptions(KdcOption.FORWARDABLE | KdcOption.PROXIABLE | KdcOption.RENEWABLE_OK);

2)  If KDC Options continue to be defined in the existing KdcOption class:

requestOptions.addKdcOption(KdcOption.FORWARDABLE.getValue());
requestOptions.addKdcOptions(KdcOption.FORWARDABLE.getValue() | KdcOption.PROXIABLE.getValue() | KdcOption.RENEWABLE_OK.getValue());
requestOptions.clearKdcOptions();
requestOptions.setKdcOptions(KdcOption.FORWARDABLE.getValue() | KdcOption.PROXIABLE.getValue() | KdcOption.RENEWABLE_OK.getValue());

There are quite a few other ways I think this would work but I'd argue that the Java convention is to treat a bitmap as an integer type (of appropriate length) and then use the bit-wise math functions to manipulate individual bits (commonly named as static values).  If it turns out there are some number of very commonly used flags, I don't have a problem with the shortcut methods proposed but I'd also argue that the two forms are equally readable, but the proposed form has the advantage that you can tell you're manipulating a KDC Option:

requestOption.setProxiable(true);

versus

requestOption.setKdcOption(KdcOption.PROXIABLE);

Sorry for my lack of participation on this thread ... I'm very interested in its outcome (I've been on vacation the last week).  When I return to the office tomorrow, I'll review it in detail and add the detail I've been promising related to the kadmin and kpasswd functionality.  I did notice there was a comment related to the lack of a specification for remote kadmin functionality.  The MIT protocol documents for both kadmin and kpasswd are available here:

https://github.com/krb5/krb5/tree/master/doc/kadmin

When I was working on this a couple years ago, I didn't find any server behaviors that contradicted those documents (note that the server might implement additional functions and I wouldn't have noticed).  The Microsoft kpasswd protocol was adopted as a standard:

http://tools.ietf.org/html/rfc3244

If I remember correctly, the Kerberos client built into Apache Directory supported clients using either protocol.  I'd suggest it would be a good idea for the Kerby client to be able to interact with servers using either kpasswd protocol.

Hope this helps!

Steve

--

“The mark of the immature man is that he wants to die nobly for a cause, while the mark of the mature man is that he wants to live humbly for one.” - Wilhelm Stekel

----- Original Message -----
From: "Emmanuel Lécharny" <el...@gmail.com>
To: kerby@directory.apache.org
Sent: Sunday, November 29, 2015 2:53:14 AM
Subject: Re: Kerby client library refactoring

Le 29/11/15 02:50, Marc Boorshtein a écrit :
> Sorry, was working on some other projects.  My thought was instead of 
> code that looks like this:
>
> requestOptions = new KOptions();
> requestOptions.add(KrbOption.USE_TGT, tgt); 
> //requestOptions.add(KrbOption.SERVER_PRINCIPAL,
> "HTTP/freeipa.rhelent.lan");
> requestOptions.add(KrbOption.SERVER_PRINCIPAL, new 
> PrincipalName("HTTP/freeipa.rhelent.lan@RHELENT.LAN",NameType.NT_UNKNO
> WN)); requestOptions.add(KrbOption.FORWARDABLE,true);
> requestOptions.add(KrbOption.PROXIABLE,false);
> requestOptions.add(KrbOption.RENEWABLE_OK,false);
>
> I would think this would be more OO:
>
> requestOptions = new KOptions();
> requestOptions.setTgt(tgt);
> //requestOptions.setServerPrincipal("HTTP/freeipa.rhelent.lan");
> requestOptions.setServerPrincipal(new
> PrincipalName("HTTP/freeipa.rhelent.lan@RHELENT.LAN",NameType.NT_UNKNO
> WN));
> requestOptions.setForwardable(true);
> requestOptions.setProxiable(false);
> requestOptions.setRenewable(false);
>
> Could keep it backed by a set of options

Agreed. This is fully compatible with the definition of all the KrbOptions enums, except thay will not be visible by the end user.

RE: Kerby client library refactoring

Posted by "Li, Jiajia" <ji...@intel.com>.

Thanks for Yan's great work on XDR, I will review the new JIRAs.

Regards,
Jiajia

-----Original Message-----
From: Yan, Yan A [mailto:yan.a.yan@intel.com] 
Sent: Friday, February 19, 2016 2:13 PM
To: kerby@directory.apache.org
Subject: RE: Kerby client library refactoring

Thanks for Kai's advice. I will do that.

Best regards,
Yan

-----Original Message-----
From: Zheng, Kai [mailto:kai.zheng@intel.com] 
Sent: Friday, February 19, 2016 2:06 PM
To: kerby@directory.apache.org
Subject: RE: Kerby client library refactoring

Thanks Yan for working on this. I understand you're saving your codes in your github, but still better to create JIRA entries for your target tasks. When your partly work is in good shape, please push request for the review and get the codes in timely. Thanks.

Regards,
Kai

-----Original Message-----
From: Yan, Yan A [mailto:yan.a.yan@intel.com] 
Sent: Friday, February 19, 2016 1:52 PM
To: kerby@directory.apache.org; Steve Moyer <sm...@psu.edu>
Subject: RE: Kerby client library refactoring

Hi, all

I was reading your discussion about adding the remote kadmin and kpasswd functionality, and I hope I can help with these modules.

In fact, I have read codes on kadmin-remote branch which has two new different modules from trunk branch (kerb-admin and kerb-admin-server).

If there is no other suggestions, I can help work on the kadmin-remote branch as you discussed before.

If someone has also started working on this issue, please let me know, and I'm glad to help share some parts.

Best regards,
Yan

-----Original Message-----
From: Zheng, Kai [mailto:kai.zheng@intel.com] 
Sent: Tuesday, February 16, 2016 2:46 PM
To: kerby@directory.apache.org; Steve Moyer <sm...@psu.edu>
Subject: RE: Kerby client library refactoring

Sorry for the late reply.

>> I've added a couple more sub-tasks to DIRKRB-440 which will allow the KrbClient to set the list of encryption types
Sounds good to me. A question is, in your case, where do you get the encryption types list? If you get them from a configuration, then this might not be needed because KrbClient can pick up them by itself.

>> the KOption and KrbKdcOptions are added into the request's options (which does make them easier to pass) but then the requests need to read them back out. I think I'd prefer that (at least as an option) we be able to build the request we want and call something like an execute() method on the client.
Did you find any issue about reading them back out? Can we enhance the codes to allow you to use KrbClient as flexible as you would build the request yourself? Not sure what's exactly needed here. Exposing AsRequest level is a little tricky because it's still in fast evolving.

>> adding the remote kpasswd and kadmin functionality will require a small amount of refactoring within the client code as well as setting up the class hierarchy in a convenient way.
Agree. We can add kpasswd and kadmin similarly as KrbClient. We had made some initial work in kadmin-remote branch for the prototype, as Jiajia mentioned. To avoid big impact for existing codes, we simply duplicated the codes and adapt accordingly. Eventually we may refactor the codes and avoid the duplication codes after we make it work first and have the unit tests.  Note there're some differences though: 1) KrbClient is against KDC, kpasswd and kadmin are against kadmind; 2) kadmin needs to authenticate via GSSAPI; 3) kadmin uses XDR for the RPC; 4) they may use different configurations. Considering such, it may sound reasonable to develop the new functionalities separately, instead of redesign the whole architecture first.

>> The Kpasswd client should actually understand both the Microsoft and the "standard" variant of the protocol, so there are actually two different request formats represent by kpasswd.
I thought it would be enough for kpasswd to support the change password protocol right, since all vendors use the spec?

>> We might want a base class for some of the tools as they do have some commonality.
Yeah, agree. We'll need add unit tests for the tools as well.

>>There's an AsRequest class defined in the client code as well as in the server code - is there some reason these classes aren't shared?  If they simply represent the packet on the wire, could they be shared?
AsRequest is a helper class to build or handle the wire message type of AsReq. It looks roughly same in client and KDC sides, but still quite much difference, not to share allows the codes can evolve independently in the two sides.

>>We don't have an ApRequest in the client yet
Ah right, good catch!! We didn't have it because it's not needed yet. Do you have any cases to use it? We have one, and actually we're implementing a new GSSAPI mechanism based on Kerby, which will need ApReqeust and some APIs. 

>>We're planning to continue work on the Kerby Client on Monday but would like some input before we "perform major surgery".  If you're interested, we could take on responsibility for the client if that helps.
Definitely you're very welcome. Thanks for the communication and sync to avoid possible conflict. Focusing on the client side sounds reasonable as it's essentially needed. We'll also made the server side work as well because it can help to do the unit tests. For small changes, how about doing in trunk targeting the RC2 release? And for the kpasswd and kadmin stuffs, how about developing them in the kadmin-remote branch? Would you check the branch, looking at the related client side modules? Our side may focus on the server side. If sounds good, I guess we can make the branch committable to you. If you have other ideas about how to collaborate, please let me know. Thanks.

Regards,
Kai

-----Original Message-----
From: Steve Moyer [mailto:smoyer@psu.edu] 
Sent: Saturday, February 13, 2016 5:34 AM
To: kerby@directory.apache.org
Subject: Re: Kerby client library refactoring

All, 

I've finally found the time to really dig into the Kerby client code with an eye towards adding remote kpasswd and kadmin functionality. Today, we found what I believe to be the last area where the Kerby client generates different packets than the MIT CLI tools. I've added a couple more sub-tasks to DIRKRB-440 which will allow the KrbClient to set the list of encryption types. I also think there's some refactoring to do around how the client builds the request - the KOption and KrbKdcOptions are added into the request's options (which does make them easier to pass) but then the requests need to read them back out. I think I'd prefer that (at least as an option) we be able to build the request we want and call something like an execute() method on the client. 

In any case, adding the remote kpasswd and kadmin functionality will require a small amount of refactoring within the client code as well as setting up the class hierarchy in a convenient way. I'd propose the following hierarchy: 

                                      +----------------+
                                      | KrbClient (AS) |
                                      +----------------+
                                         ^          ^
                                         |          |
                        +----------------+          +-------------------+
                        |                                               |
                  +-----------+                                +-----------------+
                  | KdcClient |                                | <Protocol> (AP) |
                  +-----------+                                +-----------------+
                     ^  ^  ^                                      ^           ^
                     |  |  |                                      |           |
      +--------------+  |  +-------------+                   +----+           +----+
      |                 |                |                   |                     |
+-----------+     +----------+     +-----------+     +---------------+     +--------------+
| KinitTool |     | KvnoTool |     | KlistTool |     | KpasswdClient |     | KadminClient |
+-----------+     +----------+     +-----------+     +---------------+     +--------------+
                                                             ^                     ^
                                                             |                     |
                                                      +-------------+       +------------+
                                                      | KpasswdTool |       | KadminTool |
                                                      +-------------+       +------------+

Some important notes about where we would go from here:

1)  The kpasswd and kadmin specifications share a protocol but don't share commands.  I haven't thought of a good name for the class yet so <Protocol> is the placeholder.  These commands <use> a ticket retrieved via an AsRequest but the commands themselves are sent via ApRequests.

2)  The Kpasswd client should actually understand both the Microsoft and the "standard" variant of the protocol, so there are actually two different request formats represent by kpasswd.

3)  Any class above that includes the word "Tool" uses the class above it (sorry about the ASCII art).  Currently, the KrbKdcOption enum include flags for the Tool but SoC would suggest that the arguments are a concern of the Tool itself.  I think it would also be easier to manage argument parsing with args4j or commons-cli and that the English title of the flag should also be moved out of the enum.

4)  We might want a base class for some of the tools as they do have some commonality.

5)  There's an AsRequest class defined in the client code as well as in the server code - is there some reason these classes aren't shared?  If they simply represent the packet on the wire, could they be shared?

6)  We don't have an ApRequest in the client yet, if we share the AsRequest with the server I'd suggest the same be done for ApRequest.

We're planning to continue work on the Kerby Client on Monday but would like some input before we "perform major surgery".  If you're interested, we could take on responsibility for the client if that helps.

Have a great weekend,

Steve

--

“The mark of the immature man is that he wants to die nobly for a cause, while the mark of the mature man is that he wants to live humbly for one.” - Wilhelm Stekel

----- Original Message -----
From: "Zheng, Kai" <ka...@intel.com>
To: kerby@directory.apache.org
Sent: Monday, January 11, 2016 7:29:20 AM
Subject: FW: Kerby client library refactoring

Sorry, was working on some other projects. My thought was instead of code that looks like this: 

requestOptions = new KOptions();
requestOptions.add(KrbOption.USE_TGT, tgt); //requestOptions.add(KrbOption.SERVER_PRINCIPAL,
"HTTP/freeipa.rhelent.lan");
requestOptions.add(KrbOption.SERVER_PRINCIPAL, new PrincipalName("HTTP/freeipa.rhelent.lan@RHELENT.LAN",NameType.NT_UNKNO
WN)); requestOptions.add(KrbOption.FORWARDABLE,true);
requestOptions.add(KrbOption.PROXIABLE,false);
requestOptions.add(KrbOption.RENEWABLE_OK,false); 

I would think this would be more OO: 

requestOptions = new KOptions();
requestOptions.setTgt(tgt);
//requestOptions.setServerPrincipal("HTTP/freeipa.rhelent.lan");
requestOptions.setServerPrincipal(new
PrincipalName("HTTP/freeipa.rhelent.lan@RHELENT.LAN",NameType.NT_UNKNO
WN));
requestOptions.setForwardable(true);
requestOptions.setProxiable(false);
requestOptions.setRenewable(false); 

Could keep it backed by a set of options 

Agreed. This is fully compatible with the definition of all the KrbOptions enums, except thay will not be visible by the end user.