hadoop-yarn-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Craig Welch (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (YARN-2198) Remove the need to run NodeManager as privileged account for Windows Secure Container Executor
Date Tue, 30 Sep 2014 23:12:35 GMT

    [ https://issues.apache.org/jira/browse/YARN-2198?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14153983#comment-14153983
] 

Craig Welch commented on YARN-2198:
-----------------------------------

there are a number of changes which impact common multi-platform code, has this been tested
on non-Windows with security enabled (Linux) as well as windows?

it looks like this is only a 64 bit build now, where it used to be 64 and 32. I assume this
is intentional and ok?

It would be really nice if we could start to separate out some of this new functionality from
winutils, e.g., make the elevated service functionality independent.  I see that there is
a jira for doing so down the road, which is good... it looks like the elevated privilages
are just around creating local directories and (obviously) spawning the process.  If a stand-alone
service just created and set permissions on those directories, and the java code simply checked
for their existance and then moved on if they were present, I think that a lot of the back-and-forth
of the elevation could be dropped to just one call to create the base directory and a second
to spawn/hand back the output handles.  Is that correct?  

service.c

  // We're now transfering ownership of the duplicated handles to the caller
+  // If the RPC call fails *after* this point the handles are leaked inside the NM process

this is a little alarming.  Doesn't the close() call clean this up, regardless of success/
fail?

Have we done any profiling to make sure we're not leaking threads, thread stacks, memory,
etc, in at least the "happy case" (and preferably some "unhappy cases" also)?  I think we
need to, there's a fair bit of additional native code, and running it for a bit with a profiler
could tell us quite a bit about whether or not we may be leaking something... 

why is this conditional check different from all the others?
+  dwError = ValidateConfigurationFile();
+  if (dwError) {

nit anonimous sp anonymous

hadoop-common-project/hadoop-common/src/main/native/src/org_apache_hadoop.h
just a line added, pls revert

ElevatedFileSystem

delete()
it appears that the tests for existance, etc, are run in a non-elevated way, 
while the actions are elevated.  Is it possible for permissions to be such that
the non-elevated tests do not see files/directories which are present for permission reasons?
should those not be elevated also?

streamReaderThread.run - using the readLine() instead of following the simple buffer copy
idiom in ShellCommandExecutor has some efficiency issues, granted it looks to be reading "memory
sized" data so it may be no big deal, but it would be nice to follow the buffer-copy pattern
instead

ContainerExecutor

comment on comment:

On Windows the ContainerLaunch creates a temporary empty jar to workaround the CLASSPATH length

not exactly, it looks like it creates a jar with a "special manifest of other jars", it would
be helpful to explain that in the comment so it's clear what's going on

ContainerLaunch

public void sanitizeEnv(...)

Can we please move the process of generating a new "reference jar" out of the 
sanitizeEnv method into it's own method (called ?conditionally? after sanitizeEnv)?  While
there's a clear connection in terms of it's "setting up the environment", it's building a
new jar & I think it is doing more than just manipulating variables, so it belongs in
a dedicated method, which can be called in "call()" after sanitizeEnv  I believe this also
means that Path nmPrivateClasspathJarDir can be pulled from the sanitizeEnv signature.

ContainerLocalizer

LOG.info(String.format("nRet: %d", nRet)); - not sure this should be "info" level




> Remove the need to run NodeManager as privileged account for Windows Secure Container
Executor
> ----------------------------------------------------------------------------------------------
>
>                 Key: YARN-2198
>                 URL: https://issues.apache.org/jira/browse/YARN-2198
>             Project: Hadoop YARN
>          Issue Type: Improvement
>            Reporter: Remus Rusanu
>            Assignee: Remus Rusanu
>              Labels: security, windows
>         Attachments: .YARN-2198.delta.10.patch, YARN-2198.1.patch, YARN-2198.2.patch,
YARN-2198.3.patch, YARN-2198.delta.4.patch, YARN-2198.delta.5.patch, YARN-2198.delta.6.patch,
YARN-2198.delta.7.patch, YARN-2198.separation.patch, YARN-2198.trunk.10.patch, YARN-2198.trunk.4.patch,
YARN-2198.trunk.5.patch, YARN-2198.trunk.6.patch, YARN-2198.trunk.8.patch, YARN-2198.trunk.9.patch
>
>
> YARN-1972 introduces a Secure Windows Container Executor. However this executor requires
the process launching the container to be LocalSystem or a member of the a local Administrators
group. Since the process in question is the NodeManager, the requirement translates to the
entire NM to run as a privileged account, a very large surface area to review and protect.
> This proposal is to move the privileged operations into a dedicated NT service. The NM
can run as a low privilege account and communicate with the privileged NT service when it
needs to launch a container. This would reduce the surface exposed to the high privileges.

> There has to exist a secure, authenticated and authorized channel of communication between
the NM and the privileged NT service. Possible alternatives are a new TCP endpoint, Java RPC
etc. My proposal though would be to use Windows LPC (Local Procedure Calls), which is a Windows
platform specific inter-process communication channel that satisfies all requirements and
is easy to deploy. The privileged NT service would register and listen on an LPC port (NtCreatePort,
NtListenPort). The NM would use JNI to interop with libwinutils which would host the LPC client
code. The client would connect to the LPC port (NtConnectPort) and send a message requesting
a container launch (NtRequestWaitReplyPort). LPC provides authentication and the privileged
NT service can use authorization API (AuthZ) to validate the caller.



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

Mime
View raw message