[
https://issues.apache.org/jira/browse/DRILL-5582?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16225828#comment-16225828
]
Sorabh Hamirwasia commented on DRILL-5582:
------------------------------------------
[~laurentgo] - This change exposed the issue with backward compatibility, In
absence of this change the old server (1.10) was treating new client (1.11) as
1.9 client and doing the authentication successfully which is not a correct
behavior. Reason being with 1.11 we added a new value for SASL_SUPPORT field
(SASL_PRIVACY) in Handshake message which is not known to the 1.10 server hence
when it deserialize that field value it is evaluated to default one ([Based on
this
post|http://androiddevblog.com/protocol-buffers-pitfall-adding-enum-values/])
which is UNKNOWN_SASL_SUPPORT. [Since 1.10
UserServer|https://github.com/apache/drill/blob/1.10.0/exec/java-exec/src/main/java/org/apache/drill/exec/rpc/user/UserServer.java#L297]
treats the presence of that default value as an indication of older clients
which is not correct. We should treat absence of the entire field SASL_SUPPORT
only to determine if client is old or not. Presence of the default value should
indicate client still support's SASL but at what level is not known to this
server.
With this approach:
* clients <= 1.9 will not set SASL_SUPPORT in handshake message and server
should detect it correctly and authenticate the way it was supported in 1.9
* clients > 1.9 will set SASL_SUPPORT field with their supported value and then
server on same or different will deserialize the value to be either
UNKNOWN_SASL_SUPPORT or one of SASL_AUTH/SASL_PRIVACY. In both cases server
should still think client supports SASL and speak it's protocol to require SASL
handshake.
I will start a thread in Drill-Dev group.
As far as thread model is concerned, it won't affect much in case of plain
authentication but in cases like Kerberos or mechanisms which does mutual
authentication, MITM Drillbit won't be able to fake the authentication. But
without this change MITM Drillbit can make client to skip authentication
totally.
> [Threat Modeling] Drillbit may be spoofed by an attacker and this may lead to
> data being written to the attacker's target instead of Drillbit
> ---------------------------------------------------------------------------------------------------------------------------------------------
>
> Key: DRILL-5582
> URL: https://issues.apache.org/jira/browse/DRILL-5582
> Project: Apache Drill
> Issue Type: Bug
> Affects Versions: 1.10.0
> Reporter: Rob Wu
> Assignee: Sorabh Hamirwasia
> Priority: Minor
> Labels: doc-impacting
> Fix For: 1.12.0
>
>
> *Consider the scenario:*
> Alice has a drillbit (my.drillbit.co) with plain and kerberos authentication
> enabled containing important data. Bob, the attacker, attempts to spoof the
> connection and redirect it to his own drillbit (fake.drillbit.co) with no
> authentication setup.
> When Alice is under attack and attempts to connect to her secure drillbit,
> she is actually authenticating against Bob's drillbit. At this point, the
> connection should have failed due to unmatched configuration. However, the
> current implementation will return SUCCESS as long as the (spoofing) drillbit
> has no authentication requirement set.
> Currently, the drillbit <- to -> drill client connection accepts the lowest
> authentication configuration set on the server. This leaves unsuspecting user
> vulnerable to spoofing.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
