hadoop-hdfs-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Eric Yang (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (HDDS-1609) Remove hard coded uid from Ozone docker image
Date Fri, 21 Jun 2019 17:29:00 GMT

    [ https://issues.apache.org/jira/browse/HDDS-1609?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16869729#comment-16869729
] 

Eric Yang commented on HDDS-1609:
---------------------------------

{quote}I created HDDS-1716 and uploaded the proposed patch. Please let me know if you see
any file with wrong ownership after that patch.{quote}

Will take a look.

{quote}You mentioned offline that this is not the main problem with "privilege escalation",
so here we can continue the discussion about all the remaining problem (and about the method
to reproduce them){quote}

Let's say we ship the current Ozone image with uid 1000.  System admin A uid 2002, deploys
Ozone, and Ozone is running as user john (uid 1000).  One day, John ran a ps aux on the node,
and found that he has a few processes that he doesn't recognize.  He send kill -9 signal to
Ozone processes, and brought down the entire Ozone cluster.  Who is at fault in this situation?
 User john was permitted to perform an action that he should not have access to perform, yet
Ozone is implicitly running with his user account, and System Admin uid 2002 is unware that
Ozone uses uid 1000 because there is nothing related to this were mentioned in any Ozone documentation
or released Ozone source tarball.  Who's fault is it in this example?

The above example can be avoided, if Ozone default user is one of the service account that
is documented in/usr/share/doc/setup-*/uidgid because the reserved account disallow user login
(/sbin/nologin) to prevent accidental running service as an existing user.  Hence, the above
example would not happen.

> Remove hard coded uid from Ozone docker image
> ---------------------------------------------
>
>                 Key: HDDS-1609
>                 URL: https://issues.apache.org/jira/browse/HDDS-1609
>             Project: Hadoop Distributed Data Store
>          Issue Type: Sub-task
>            Reporter: Eric Yang
>            Priority: Major
>             Fix For: 0.5.0
>
>         Attachments: linux.txt, log.html, osx.txt, report.html
>
>
> Hadoop-runner image is hard coded to [USER hadoop|https://github.com/apache/hadoop/blob/docker-hadoop-runner-jdk11/Dockerfile#L45]
and user hadoop is hard coded to uid 1000.  This arrangement complicates development environment
where host user is different uid from 1000.  External bind mount locations are written data
as uid 1000.  This can prevent development environment from clean up test data.  
> Docker documentation stated that "The best way to prevent privilege-escalation attacks
from within a container is to configure your container’s applications to run as unprivileged
users."  From Ozone architecture point of view, there is no reason to run Ozone daemon to
require privileged user or hard coded user.
> h3. Solution 1
> It would be best to support running docker container as host user to reduce friction.
 User should be able to run:
> {code}
> docker run -u $(id -u):$(id -g) ...
> {code}
> or in docker-compose file:
> {code}
> user: "${UID}:${GID}"
> {code}
> By doing this, the user will be name less in docker container.  Some commands may warn
that user does not have a name.  This can be resolved by mounting /etc/passwd or a file that
looks like /etc/passwd that contain host user entry.
> h3. Solution 2
> Move the hard coded user to range < 200.  The default linux profile reserves service
users < 200 to have umask that keep data private to service user or group writable, if
service shares group with other service users.  Register the service user with Linux vendors
to ensure that there is a reserved uid for Hadoop user or pick one that works for Hadoop.
 This is a longer route to pursuit, and may not be fruitful.  
> h3. Solution 3
> Default the docker image to have sssd client installed.  This will allow docker image
to see host level names by binding sssd socket.  The instruction for doing this is located
at in [Hadoop website| https://hadoop.apache.org/docs/r3.1.2/hadoop-yarn/hadoop-yarn-site/DockerContainers.html#User_Management_in_Docker_Container].
> The pre-requisites for this approach will require the host level system to have sssd
installed.  For production system, there is a 99% chance that sssd is installed.
> We may want to support combined solution of 1 and 3 to be proper.



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)

---------------------------------------------------------------------
To unsubscribe, e-mail: hdfs-issues-unsubscribe@hadoop.apache.org
For additional commands, e-mail: hdfs-issues-help@hadoop.apache.org


Mime
View raw message