[
https://issues.apache.org/jira/browse/YARN-7516?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16346045#comment-16346045
]
Shane Kumpf commented on YARN-7516:
-----------------------------------
Hey Eric, thanks for the patches!
I'm sorry to jump in here late. I really like the premise, but I'll admit, I'm
struggling with the naming of this feature and have been thinking about
alternatives. IMO, "docker.privileged-containers.registries" implies containers
requesting the --privileged flag, which isn't necessarily the case here.
To keep this short, I think we are looking at two ways that we launch
containers, I'll refer to them as "default-mode" and "yarn-mode".
In "default-mode", the following are removed:
* Capabilities
* Mounts
* Privileged flag
* Devices
* User*
* Launch command*
*Note that I think this patch needs to also remove overriding the user and
launch command, regardless of direction it is taken. With all the other
features disabled, these don't make sense.
In "default-mode", the container runs as defined by the image with no features.
This has the benefits of limiting any host access to improve security and
making it easier to get started with Docker on YARN. This is the direction the
patch is already headed.
In "yarn-mode" some combination of those features would be enabled. Using these
features requires playing by the rules of being a "yarn-mode" container. We
already have controls around mounts, devices, capabilities and privileged, but
there are gaps, and the current defaults need review, which we could tackle in
another JIRA. This allows for some amount of "opt-in" or "opt-out" of YARN or
OS features for "yarn-mode" containers, which may be desirable.
With the modes defined, we could have a configuration similar to (example):
{code:java}
docker.default-mode.image.prefix=*
docker.yarn-mode.image.prefix=registry.example.com/qe,registry.example.com/re/jenkins,httpd:2.4{code}
If the image begins with one of the prefix entries, run the container in that
mode.
One benefit is that the settings don't have overlap with existing docker terms
like privileged and trusted and IMO, a registry is too large to use to
determine which mode a container should run in. Of course, the prefix can be
used to specify an entire registry if desired. It would also allow for getting
as specific as needed.
A couple other thoughts
* docker.default-mode.image.prefix should default to * for ease of use.
* An image (or namespace or registry) would be promoted to yarn-mode once it
has had the proper vetting.
* Any image not in either list would result in container failure. The image is
not allowed.
* yarn-mode list takes precedence, fall back to default-mode check.
Let me know what you think.
> Security check for trusted docker image
> ---------------------------------------
>
> Key: YARN-7516
> URL: https://issues.apache.org/jira/browse/YARN-7516
> Project: Hadoop YARN
> Issue Type: Sub-task
> Reporter: Eric Yang
> Assignee: Eric Yang
> Priority: Major
> Attachments: YARN-7516.001.patch, YARN-7516.002.patch,
> YARN-7516.003.patch, YARN-7516.004.patch, YARN-7516.005.patch,
> YARN-7516.006.patch, YARN-7516.007.patch, YARN-7516.008.patch,
> YARN-7516.009.patch, YARN-7516.010.patch, YARN-7516.011.patch,
> YARN-7516.012.patch, YARN-7516.013.patch, YARN-7516.014.patch,
> YARN-7516.015.patch
>
>
> Hadoop YARN Services can support using private docker registry image or
> docker image from docker hub. In current implementation, Hadoop security is
> enforced through username and group membership, and enforce uid:gid
> consistency in docker container and distributed file system. There is cloud
> use case for having ability to run untrusted docker image on the same cluster
> for testing.
> The basic requirement for untrusted container is to ensure all kernel and
> root privileges are dropped, and there is no interaction with distributed
> file system to avoid contamination. We can probably enforce detection of
> untrusted docker image by checking the following:
> # If docker image is from public docker hub repository, the container is
> automatically flagged as insecure, and disk volume mount are disabled
> automatically, and drop all kernel capabilities.
> # If docker image is from private repository in docker hub, and there is a
> white list to allow the private repository, disk volume mount is allowed,
> kernel capabilities follows the allowed list.
> # If docker image is from private trusted registry with image name like
> "private.registry.local:5000/centos", and white list allows this private
> trusted repository. Disk volume mount is allowed, kernel capabilities
> follows the allowed list.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
