[
https://issues.apache.org/jira/browse/MAPREDUCE-3256?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Vinod Kumar Vavilapalli updated MAPREDUCE-3256:
-----------------------------------------------
Description:
We already authenticate requests to NM from any AM. We also need to authorize
the requests, otherwise a rogue AM, *but with proper tokens and thus
authenticated to talk to NM*, could either launch or kill a container with
different ContainerID. We have two options:
- Remove the explicit passing of the ContainerId as part of the API and
instead get it from the RPC layer. In this case, we will need a ContainerToken
for each container.
- Do explicit authorization checks without relying on getting ContainerID from
the RPC.
One ContainerToken per container is a serious restriction. We anyways want to
be able to use application-ACLS to, say, stop containers owned by others. So I
am going to take the later route of explicit checks.
was:We already authenticate requests to NM and RM from any AM. We also need
to authorize the requests, otherwise a rogue AM could either launch or kill a
container with different ContainerID or masquerade as another AM and illegally
obtain containers meant for others.
Summary: Authorization checks needed for AM->NM protocol (was:
Authorization checks needed for AM->NM and AM->RM protocols)
Okay, I am realizing that each protocol is a good chunk of work in itself. I am
going to do each of them separately. Using this ticket for ContainerManager
protocol related authorization checks.
> Authorization checks needed for AM->NM protocol
> -----------------------------------------------
>
> Key: MAPREDUCE-3256
> URL: https://issues.apache.org/jira/browse/MAPREDUCE-3256
> Project: Hadoop Map/Reduce
> Issue Type: Sub-task
> Components: applicationmaster, nodemanager, resourcemanager, security
> Affects Versions: 0.23.0
> Reporter: Vinod Kumar Vavilapalli
> Priority: Blocker
> Fix For: 0.23.0
>
>
> We already authenticate requests to NM from any AM. We also need to authorize
> the requests, otherwise a rogue AM, *but with proper tokens and thus
> authenticated to talk to NM*, could either launch or kill a container with
> different ContainerID. We have two options:
> - Remove the explicit passing of the ContainerId as part of the API and
> instead get it from the RPC layer. In this case, we will need a
> ContainerToken for each container.
> - Do explicit authorization checks without relying on getting ContainerID
> from the RPC.
> One ContainerToken per container is a serious restriction. We anyways want to
> be able to use application-ACLS to, say, stop containers owned by others. So
> I am going to take the later route of explicit checks.
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators:
https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa
For more information on JIRA, see: http://www.atlassian.com/software/jira
