kafka-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Rajini Sivaram (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (KAFKA-4764) Improve diagnostics for SASL authentication failures
Date Fri, 05 May 2017 11:29:04 GMT

    [ https://issues.apache.org/jira/browse/KAFKA-4764?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15998186#comment-15998186

Rajini Sivaram commented on KAFKA-4764:

[~junrao] SASL was designed to be embedded into a protocol like LDAP. And Kafka is one such
protocol, but we (like ZooKeeper) chose a slightly different way of integrating with SASL
from the way it was designed to be used. According to the [SASL RFC|https://tools.ietf.org/html/rfc4422],

To use SASL, each protocol (amongst other items) provides a method for identifying which mechanism
is to be used, a method for exchange of mechanism-specific server-challenges and client-responses,
and a method for communicating the outcome of the authentication exchange.

We have the {{SaslHandshake}} request to identify the mechanism. And we have the code to exchange
challenges and responses. But we don't have a method for communicating the outcome.

[SASL/PLAIN RFC|https://tools.ietf.org/html/rfc4616] has examples using ACAP:
   C: a001 STARTTLS
   S: a001 OK "Begin TLS negotiation now"
   <TLS negotiation, further commands are under TLS layer>
   S: + ""
   C: (21)
   C: <NUL>tim<NUL>tanstaaftanstaaf
   S: a002 OK "Authenticated"

Here client message a002 starts the PLAIN authentication. Authentication messages are sent
and received in the next three lines. The outcome, i.e. response to request a002 is sent after
the authentication completes or fails. In Kafka, it feels like it would be easier to add a
header to the SASL exchanges with the final message containing the outcome from the broker,
rather than have a Kafka protocol request followed by non-Kafka SASL messages followed by
Kafka protocol response.

Obviously, this is just based on my reading of the RFCs. It will also be useful if you or
[~ijuma] have a look too to make sure I haven't missed anything, before deciding whether this
is the right approach. 

Thank you...

> Improve diagnostics for SASL authentication failures
> ----------------------------------------------------
>                 Key: KAFKA-4764
>                 URL: https://issues.apache.org/jira/browse/KAFKA-4764
>             Project: Kafka
>          Issue Type: Improvement
>          Components: security
>    Affects Versions:
>            Reporter: Rajini Sivaram
>            Assignee: Rajini Sivaram
>             Fix For:
> At the moment, broker closes the client connection if SASL authentication fails. Clients
see this as a connection failure and do not get any feedback for the reason why the connection
was closed. Producers and consumers retry, attempting to create successful connections, treating
authentication failures as transient failures. There are no log entries on the client-side
which indicate that any of these connection failures were due to authentication failure.
> This JIRA will aim to improve diagnosis of authentication failures with the following
> - Broker will send an authentication error code if SASL authentication fails, just before
closing the connection. This will be treated as an invalid token by the client authenticator,
and the error handling for invalid tokens will be updated to report authentication failure
for this case. This is a bit of a hack, but would work with GSSAPI, PLAIN and SCRAM. SASL
itself doesn't provide a mechanism-independent way of reporting authentication failures. An
alternative would be to wrap SASL authentication in Kafka request/response to enables error
codes to be sent as Kafka response, but that would be a much bigger change.
> - Log a warning in clients for authentication failures, distinguishing these from EOF
exceptions due to connection failure
> - Blackout nodes to which connection failed due to authentication error, no more attempts
will be made to connect to these nodes.
> - We should use the connection state to improve handling of producer/consumer requests,
avoiding unnecessary blocking. This will not be addressed in this JIRA, KAFKA-3899 should
be able to use the additional state from JIRA to fix this issue.
> This JIRA also does not change handling of SSL authentication failures. javax.net.debug
provides sufficient diagnostics for this case, I don't believe there is sufficient information
in `SslTransportLayer` to treat these in a consistent way with SASL authentication failures.

This message was sent by Atlassian JIRA

View raw message