hadoop-yarn-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Miklos Szegedi (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (YARN-5673) [Umbrella] Re-write container-executor to improve security, extensibility, and portability
Date Wed, 07 Dec 2016 19:51:59 GMT

    [ https://issues.apache.org/jira/browse/YARN-5673?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15729736#comment-15729736
] 

Miklos Szegedi commented on YARN-5673:
--------------------------------------

Thank you [~vvasudev] for the quick and detailed response! I really appreciate it.
{quote}
All of these binaries will require the setuid bit to be a set which means administrators will
have to set permissions and manage 4 binaries. We also have to worry about 4 binaries that
can have privilege escalation as opposed to one - any hot fixes for example will require all
4 binaries to be updated as opposed to just one. Interestingly you feel that administrator
overhead of managing 4 binaries is worth it whereas some folks would prefer it the other way
round . Do other folks feel that the multiple binaries approach is the way to go?
{quote}
Yes, I absolutely agree that this is a preference question. I am not sure about the ratios
though. In terms of overhead, the administrator has to enable the modules in the configuration
anyways. What I thought is that it is easier to set the permissions using familiar Unix tools
rather than looking up the configuration files, reading the documentation about them and enabling
the required modules with the right format. I have seen in the past issues with too many spaces
for example.
However, please take some time to answer this one question. Let's assume, we were about to
design /usr/bin/at, /usr/bin/sudo and /user/bin/passwd. They have about the same difference
as container launching and mounting cgroups. Would you design them as separate tools, or as
a single binary that loads them separately as modules based on a configuration file and command
line options?
{quote}
We also have to worry about 4 binaries that can have privilege escalation as opposed to one
{quote}
I think the risk of privilege escalation is proportional to the amount of code rather than
the amount of binaries, so it is about the same. On the other hand packing multiple functions
into the same memory space may increase the sum of the individual risks in case of native
code.
{quote}
any hot fixes for example will require all 4 binaries to be updated as opposed to just one.

{quote}
(I assume that the code will be super stable :-) ...)
This was actually one question that I raised. What is the common code among features separated
into modules? Only if common functionality is broken, it needs to be patched. I think this
would be limited to auditing, logging, and maybe some filesystem operations that can be linked
to the tools.
{quote}
Fair point. The idea here is that -
(1) Administrators will not add arbitrary modules to the module list.
(2) The posix-container-executor will give up all privileges before loading the modules which
don't require administrator privileges
(3) Give administrators an option to turn off modules that require administrator privileges.
Would these help mitigate your concerns? The issue with the current setup is that there is
no clean way to enable/disable functionality that administrators do not want enabled on their
cluster.
{quote}
I agree, this is an issue in the current setup and yes, I think these are the right design
decisions. Just as a side note, I prefer privileged modules be disabled by default for security
and supportability reasons.
{quote}
Do you have some scenarios where container launch time has been an issue? The security aspects
of a long running process versus one which is invoked on demand are different as well.
{quote}
I just wanted to discuss this design option early before much coding has started. If we want
to use Yarn not just for long batch processing but for lots of quick requests in the future,
launch time is an issue. I thought I raise pipe as an other option communicating the commands
together with command line, file, and environment variables.

> [Umbrella] Re-write container-executor to improve security, extensibility, and portability
> ------------------------------------------------------------------------------------------
>
>                 Key: YARN-5673
>                 URL: https://issues.apache.org/jira/browse/YARN-5673
>             Project: Hadoop YARN
>          Issue Type: New Feature
>          Components: nodemanager
>            Reporter: Varun Vasudev
>            Assignee: Varun Vasudev
>         Attachments: container-executor Re-write Design Document.pdf
>
>
> As YARN adds support for new features that require administrator privileges(such as support
for network throttling and docker), we’ve had to add new capabilities to the container-executor.
This has led to a recognition that the current container-executor security features as well
as the code could be improved. The current code is fragile and it’s hard to add new features
without causing regressions. Some of the improvements that need to be made are -
> *Security*
> Currently the container-executor has limited security features. It relies primarily on
the permissions set on the binary but does little additional security beyond that. There are
few outstanding issues today -
> - No audit log
> - No way to disable features - network throttling and docker support are built in and
there’s no way to turn them off at a container-executor level
> - Code can be improved - a lot of the code switches users back and forth in an arbitrary
manner
> - No input validation - the paths, and files provided at invocation are not validated
or required to be in some specific location
> - No signing functionality - there is no way to enforce that the binary was invoked by
the NM and not by any other process
> *Code Issues*
> The code layout and implementation themselves can be improved. Some issues there are
-
> - No support for log levels - everything is logged and this can’t be turned on or off
> - Extremely long set of invocation parameters(specifically during container launch) which
makes turning features on or off complicated
> - Poor test coverage - it’s easy to introduce regressions today due to the lack of
a proper test setup
> - Duplicate functionality - there is some amount of code duplication
> - Hard to make improvements or add new features due to the issues raised above
> *Portability*
>  - The container-executor mixes platform dependent APIs with platform independent APIs
making it hard to run it on multiple platforms. Allowing it to run on multiple platforms also
improves the overall code structure .
> One option is to improve the existing container-executor, however it might be easier
to start from scratch. That allows existing functionality to be supported until we are ready
to switch to the new code.
> This umbrella JIRA is to capture all the work required for the new code. I'm going to
work on a design doc for the changes - any suggestions or improvements are welcome.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

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


Mime
View raw message