[jira] [Comment Edited] (HDFS-6717) Jira HDFS-5804 breaks default nfs-gateway behavior for unsecured config

2014-08-05 Thread Brandon Li (JIRA)

[ 
https://issues.apache.org/jira/browse/HDFS-6717?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=14086944#comment-14086944
 ] 

Brandon Li edited comment on HDFS-6717 at 8/5/14 11:01 PM:
---

Thank you, [~kasha] and sorry for committing it to 2.5.


was (Author: brandonli):
Thank you, [~kasha] and sorry for committing it the 2.5.

 Jira HDFS-5804 breaks default nfs-gateway behavior for unsecured config
 ---

 Key: HDFS-6717
 URL: https://issues.apache.org/jira/browse/HDFS-6717
 Project: Hadoop HDFS
  Issue Type: Sub-task
  Components: nfs
Affects Versions: 2.4.0
Reporter: Jeff Hansen
Assignee: Brandon Li
Priority: Minor
 Fix For: 2.6.0

 Attachments: HDFS-6717.001.patch, HDFS-6717.more-change.patch, 
 HDFS-6717.more-change2.patch, HDFS-6717.more-change3.patch, 
 HdfsNfsGateway.html


 I believe this is just a matter of needing to update documentation. As a 
 result of https://issues.apache.org/jira/browse/HDFS-5804, the secure and 
 unsecure code paths appear to have been merged -- this is great because it 
 means less code to test. However, it means that the default unsecure behavior 
 requires additional configuration that needs to be documented. 
 I'm not the first to have trouble following the instructions documented in 
 http://hadoop.apache.org/docs/r2.4.0/hadoop-project-dist/hadoop-hdfs/HdfsNfsGateway.html
 I kept hitting a RemoteException with the message that hdfs user cannot 
 impersonate root -- apparently under the old code, there was no impersonation 
 going on, so the nfs3 service could and should be run under the same user id 
 that runs hadoop (I assumed this meant the user id hdfs). However, with the 
 new unified code path, that would require hdfs to be able to impersonate root 
 (because root is always the local user that mounts a drive). The comments in 
 jira hdfs-5804 seem to indicate nobody has a problem with requiring the 
 nfsserver user to impersonate root -- if that means it's necessary for the 
 configuration to include root as a user nfsserver can impersonate, that 
 should be included in the setup instructions.
 More to the point, it appears to be absolutely necessary now to provision a 
 user named nfsserver in order to be able to give that nfsserver ability to 
 impersonate other users. Alternatively I think we'd need to configure hdfs to 
 be able to proxy other users. I'm not really sure what the best practice 
 should be, but it should be documented since it wasn't needed in the past.



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


[jira] [Comment Edited] (HDFS-6717) Jira HDFS-5804 breaks default nfs-gateway behavior for unsecured config

2014-07-28 Thread Brandon Li (JIRA)

[ 
https://issues.apache.org/jira/browse/HDFS-6717?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=14076928#comment-14076928
 ] 

Brandon Li edited comment on HDFS-6717 at 7/28/14 9:41 PM:
---

[~dscheffy], you are right that root group needs to be added to the the value 
of hadoop.proxyuser.nfsserver.groups. The change in the patch was focusing on 
answering who is proxy user, who should start NFS gateway and etc.

Based on your last comment, I think we should also add some description saying 
group 'root' should be added to hadoop.proxyuser.nfsserver.groups in most 
Linux platforms where root privilege is by default required to mount NFS 
exports.

Please let me know if there is any unclear description in the doc and I will 
fix it.  


was (Author: brandonli):
[~dscheffy], you are right that root group needs to be added to the the value 
of hadoop.proxyuser.nfsserver.groups. The change in the patch was focusing on 
answering who is proxy user, who should start NFS gateway and etc.

Based on your last comment, I think we should also add some description saying 
group root should be added to hadoop.proxyuser.nfsserver.groups in most 
Linux platforms where root privilege is by default required to mount NFS 
exports.

Please let me know if there is any unclear description in the doc and I will 
fix it.  

 Jira HDFS-5804 breaks default nfs-gateway behavior for unsecured config
 ---

 Key: HDFS-6717
 URL: https://issues.apache.org/jira/browse/HDFS-6717
 Project: Hadoop HDFS
  Issue Type: Sub-task
  Components: nfs
Affects Versions: 2.4.0
Reporter: Jeff Hansen
Assignee: Brandon Li
Priority: Minor
 Fix For: 2.5.0

 Attachments: HDFS-6717.001.patch, HdfsNfsGateway.html


 I believe this is just a matter of needing to update documentation. As a 
 result of https://issues.apache.org/jira/browse/HDFS-5804, the secure and 
 unsecure code paths appear to have been merged -- this is great because it 
 means less code to test. However, it means that the default unsecure behavior 
 requires additional configuration that needs to be documented. 
 I'm not the first to have trouble following the instructions documented in 
 http://hadoop.apache.org/docs/r2.4.0/hadoop-project-dist/hadoop-hdfs/HdfsNfsGateway.html
 I kept hitting a RemoteException with the message that hdfs user cannot 
 impersonate root -- apparently under the old code, there was no impersonation 
 going on, so the nfs3 service could and should be run under the same user id 
 that runs hadoop (I assumed this meant the user id hdfs). However, with the 
 new unified code path, that would require hdfs to be able to impersonate root 
 (because root is always the local user that mounts a drive). The comments in 
 jira hdfs-5804 seem to indicate nobody has a problem with requiring the 
 nfsserver user to impersonate root -- if that means it's necessary for the 
 configuration to include root as a user nfsserver can impersonate, that 
 should be included in the setup instructions.
 More to the point, it appears to be absolutely necessary now to provision a 
 user named nfsserver in order to be able to give that nfsserver ability to 
 impersonate other users. Alternatively I think we'd need to configure hdfs to 
 be able to proxy other users. I'm not really sure what the best practice 
 should be, but it should be documented since it wasn't needed in the past.



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