Recently I found myself needing to connect to HiveServer2 with Kerberos authentication enabled from a Ruby app. As it turned out rbhive gem we were using did not have support for Kerberos authentication. So I had to roll my own.
This post is to document the experience of figuring out the details of a SASL/GSSAPI connection before it is lost forever in my neurons and synapses.
First, the terminology. The authentication system that Hadoop uses is Kerberos. Note that Kerberos is not a network protocol. It describes the method by which authentication happens, but not the format of how to send Kerberos tickets and what not over the wire. For that, you need SASL and GSSAPI.
SASL is a generic protocol designed to be able to wrap just about any authentication handshake. It’s very simple: the client sends a START followed by some payload, and expects an OK, BAD or COMPLETE from the server. OK means that there are more steps to this conversation, BAD is self-explanatory and COMPLETE means “I’m satisfied”. The objective is to go from START via a series of OK’s to each side sending the other a COMPLETE.
SASL doesn’t define the payload of each message. The payload is specified by GSSAPI protocol. GSSAPI is another generic protocol. Unlike SASL it is actually very complex and covers a variety of authentication methods, including Kerberos.
The combination of SASL and GSSAPI and what happens at the network layer is documented in RFC4752.
Bottom line is you need to read at least four RFC’s to be able to understand every detail of this process: RFC4120, RFC2222, RFC2743 and RFC4752. Fun!
The Handshake in Ruby
First, you’ll need some form of binding to the GSSAPI libraries. I’ve been using the most excellent GSSAPI gem by Dan Wanek which wraps the MIT GSSAPI library.
If you follow the code in sasl_client_transport.rb, you’ll see the following steps are required to establish a connection.
First, we instantiate a GSSAPI object passing it the remote host and the remote principal. Note that there is no TCP port number to be specifies anywhere, because this isn’t to establish a TCP connection, but only for Kerberos host authentication. (Kerberos requires that not only the client authenticates itself to the host, but also that the host authenticates itself to the client.)
1 2 |
|
The rest of the action takes place in the
initiate_hand_shake_gssapi()
method.
First, we call @gsscli.init_context()
with no arguments. This call
creates a token based on our current Kerberos credentials. (If there
are no credentials in our cache, this call will fail).
1
|
|
Next we compose a SASL message which consists of START (0x01) followed by payload length, followed by the actual payload, which is the SASL mechanism name: ‘GSSAPI’. Without waiting for response, we also send an OK (0x02) and the token returned from init_context().
1 2 3 4 5 |
|
Next we read 5 bytes of response. The first byte is the status returned from the server, which hopefully is OK, followed by the length of the payload, and then we read the payload itself:
1 2 3 4 5 6 7 8 |
|
The payload is a challenge created for us by the server. We can
verify this challenge by calling init_context()
a second time, this
time passing in the challenge to verify it:
1 2 3 4 |
|
If the challenge verifies, then it is our turn to send an OK (with an empty payload this time):
1 2 |
|
At this point in the SASL ‘conversation’ we have verified that the server is who they claim to be.
Next the server sends us another challenge, this one is so that we can authenticate ourselves to the server and at the same time agree on the protection level for the communication channel.
We need to decrypt (“unwrap” in the GSSAPI terminology) the challenge, examine the protection level and if it is acceptable, encrypt it on our side and send it back to the server in a SASL COMPLETE message. In this particular case we’re agreeable to any level of protection (which is none in case of HiveServer2, i.e. the conversation is not encrypted). Otherwise there are additional steps that RFC4752 describes whereby the client can select an acceptable protection level.
1 2 3 4 5 6 7 8 9 10 11 12 |
|
The server should then respond with COMPLETE as well, at which point we’re done with the authentication process and cat start sending whatever we want over this connection:
1 2 3 4 5 6 7 8 9 10 |
|