Mohammad Kamrul Islam created HADOOP-10509:
----------------------------------------------

             Summary: cancelToken doesn't work in some instances
                 Key: HADOOP-10509
                 URL: https://issues.apache.org/jira/browse/HADOOP-10509
             Project: Hadoop Common
          Issue Type: Bug
          Components: security
            Reporter: Mohammad Kamrul Islam
            Assignee: Mohammad Kamrul Islam


When the owner of a token tries to explicitly cancel the token, it gets the 
following error/exception
{noformat} 
2014-04-14 20:07:35,744 WARN org.apache.hadoop.security.UserGroupInformation: 
PriviledgedActionException 
as:<someuser>/<machine_name>.linkedin.com@<realm>.LINKEDIN.COM (auth:KERBEROS) 
cause:org.apache.hadoop.security.AccessControlException: <someuser> is not 
authorized to cancel the token
2014-04-14 20:07:35,744 INFO org.apache.hadoop.ipc.Server: IPC Server handler 2 
on 10020, call 
org.apache.hadoop.mapreduce.v2.api.HSClientProtocolPB.cancelDelegationToken 
from 172.20.158.61:49042 Call#4 Retry#0: error: 
org.apache.hadoop.security.AccessControlException: <someuser> is not authorized 
to cancel the token
org.apache.hadoop.security.AccessControlException: <someuser> is not authorized 
to cancel the token
        at 
org.apache.hadoop.security.token.delegation.AbstractDelegationTokenSecretManager.cancelToken(AbstractDelegationTokenSecretManager.java:429)
        at 
org.apache.hadoop.mapreduce.v2.hs.HistoryClientService$HSClientProtocolHandler.cancelDelegationToken(HistoryClientService.java:400)
        at 
org.apache.hadoop.mapreduce.v2.api.impl.pb.service.MRClientProtocolPBServiceImpl.cancelDelegationToken(MRClientProtocolPBServiceImpl.java:286)
        at 
org.apache.hadoop.yarn.proto.MRClientProtocol$MRClientProtocolService$2.callBlockingMethod(MRClientProtocol.java:301)
        at 
org.apache.hadoop.ipc.ProtobufRpcEngine$Server$ProtoBufRpcInvoker.call(ProtobufRpcEngine.java:585)
        at org.apache.hadoop.ipc.RPC$Server.call(RPC.java:928)
        at org.apache.hadoop.ipc.Server$Handler$1.run(Server.java:1962)
        at org.apache.hadoop.ipc.Server$Handler$1.run(Server.java:1958)
        at java.security.AccessController.doPrivileged(Native Method)
        at javax.security.auth.Subject.doAs(Subject.java:415)
        at 
org.apache.hadoop.security.UserGroupInformation.doAs(UserGroupInformation.java:1548)
        at org.apache.hadoop.ipc.Server$Handler.run(Server.java:1956)

{noformat}


Details:
AbstractDelegationTokenSecretManager.cacelToken() gets the owner as full 
principal name where as the canceller is the short name.
The potential code snippets:
{code}
String owner = id.getUser().getUserName(); 
    Text renewer = id.getRenewer();
    HadoopKerberosName cancelerKrbName = new HadoopKerberosName(canceller);
    String cancelerShortName = cancelerKrbName.getShortName();
    if (!canceller.equals(owner)
        && (renewer == null || renewer.toString().isEmpty() || 
!cancelerShortName
            .equals(renewer.toString()))) {
      throw new AccessControlException(canceller
          + " is not authorized to cancel the token");
    }
{code}

The code shows 'owner' gets the full principal name. Where as the value of 
'canceller' depends on who is calling it. 
In some cases, it is the short name. REF: HistoryClientService.java
{code}
String user = UserGroupInformation.getCurrentUser().getShortUserName();
        jhsDTSecretManager.cancelToken(token, user);
{code}
 
In other cases, the value could be full principal name. REF: FSNamesystem.java.
{code}
String canceller = getRemoteUser().getUserName();
      DelegationTokenIdentifier id = dtSecretManager
        .cancelToken(token, canceller);
{code}

Possible resolution:
--------------------------
Option 1: in cancelToken() method, compare with both : short name and full 
principal name.
Pros: Easy. Have to change in one place.
Cons: Someone can argue that it is hacky!
 
Option 2:
All the caller sends the consistent value as 'canceller' : either short name or 
full principal name.

Pros: Cleaner.
Cons: A lot of code changes and potential bug injections.

I'm open for both options.
Please give your opinion.

Btw, how it is working now in most cases?  The short name and the full 
principal name are usually the same for end-users.






--
This message was sent by Atlassian JIRA
(v6.2#6252)

Reply via email to