Mailing-List: contact mapreduce-issues-help@hadoop.apache.org; run by ezmlm
Precedence: bulk
Reply-To: mapreduce-issues@hadoop.apache.org
Message-ID: <27319940.98601284052234777.JavaMail.jira@thor>
Date: Thu, 9 Sep 2010 13:10:34 -0400 (EDT)
From: "Allen Wittenauer (JIRA)" <jira@apache.org>
To: mapreduce-issues@hadoop.apache.org
Subject: [jira] Commented: (MAPREDUCE-2057) Job Tracker appears to do host
 access-control (mapred.hosts, mapred.hosts.exclude) based on presented name
 from TaskTracker
In-Reply-To: <31361583.98331284051333616.JavaMail.jira@thor>
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Transfer-Encoding: 7bit


    [ https://issues.apache.org/jira/browse/MAPREDUCE-2057?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12907681#action_12907681 ] 

Allen Wittenauer commented on MAPREDUCE-2057:
---------------------------------------------


>From a security perspective, the patches coming in 0.22 will make sure that a host is who it says it is by requiring Kerberized credentials.  (ignoring the some recognized MITM and IP spoofing attacks which the community has made the conscious choice to defer fixing)

That said, it would be better if Hadoop in general took the IP addr, reverse resolved, and then compared that to the config.  

> Job Tracker appears to do host access-control (mapred.hosts, mapred.hosts.exclude) based on presented name from TaskTracker
> ---------------------------------------------------------------------------------------------------------------------------
>
>                 Key: MAPREDUCE-2057
>                 URL: https://issues.apache.org/jira/browse/MAPREDUCE-2057
>             Project: Hadoop Map/Reduce
>          Issue Type: Bug
>          Components: jobtracker
>    Affects Versions: 0.20.1
>         Environment: Hadoop 0.20.1 - cloudera distribution, multihomed environment.
>            Reporter: Matthew Byng-Maddick
>
> As far as I can tell, where the NameNode, in validating the dfs.hosts and dfs.hosts.exclude files uses the source IP address for the RPC connection, the JobTracker appears to use the presented hostname (set via slave.host.name or the standard hostname-search semantics) from the TaskTracker. Obviously this is a security bug as in a production environment it could allow rogue machines to present the hostname of a real TaskTracker and take over that role, but it also turns up as a configuration bug because it means that you can set up a (multi-homed, natch) environment where the same set of files work for the NameNode, but don't for the JobTracker or vice versa - with the same binding hostname for fs.default.name and mapred.job.tracker.

-- 
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.