Mailing-List: contact common-issues-help@hadoop.apache.org; run by ezmlm
Precedence: bulk
Date: Wed, 29 Jun 2016 00:24:10 +0000 (UTC)
From: "Andrew Wang (JIRA)" <jira@apache.org>
To: common-issues@hadoop.apache.org
Message-ID: <JIRA.12857620.1440097415000.562.1467159850428@Atlassian.JIRA>
In-Reply-To: <JIRA.12857620.1440097415000@Atlassian.JIRA>
References: <JIRA.12857620.1440097415000@Atlassian.JIRA> <JIRA.12857620.1440097415200@arcas>
Subject: [jira] [Commented] (HADOOP-12345) Credential length in
 CredentialsSys.java incorrect
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Transfer-Encoding: 7bit
archived-at: Wed, 29 Jun 2016 00:24:12 -0000


    [ https://issues.apache.org/jira/browse/HADOOP-12345?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15353974#comment-15353974 ] 

Andrew Wang commented on HADOOP-12345:
--------------------------------------

Looked at the PR. Can we add a test for the length % 4 == 0 case? I think the current code is not quite right:

{code}
+    int padding = 0;
+    // we do not need compute padding if the hostname is already a multiple of 4
+    if (mHostName.getBytes(Charsets.UTF_8).length != 0) {
+      padding = 4 - (mHostName.getBytes(Charsets.UTF_8).length % 4);
+    }
{code}

I think you meant to check that the len % 4 != 0. I think it'd be even better though if we just {{%4}} the padding one more time, saves the if statement.

Also we only need that comment about the padding once, can delete the second one.

> Credential length in CredentialsSys.java incorrect
> --------------------------------------------------
>
>                 Key: HADOOP-12345
>                 URL: https://issues.apache.org/jira/browse/HADOOP-12345
>             Project: Hadoop Common
>          Issue Type: Bug
>          Components: nfs
>    Affects Versions: 2.6.0, 2.7.0
>            Reporter: Pradeep Nayak Udupi Kadbet
>            Priority: Critical
>         Attachments: HADOOP-12345.patch
>
>
> Hi -
> There is a bug in the way hadoop-nfs sets the credential length in "Credentials" field of the NFS RPC packet when using AUTH_SYS
> In CredentialsSys.java, when we are writing the creds in to XDR object, we set the length as follows:
>  // mStamp + mHostName.length + mHostName + mUID + mGID + mAuxGIDs.count
> 96     mCredentialsLength = 20 + mHostName.getBytes().length;
> (20 corresponds to 4 bytes for mStamp, 4 bytes for mUID, 4 bytes for mGID, 4 bytes for length field of hostname, 4 bytes for number of aux 4 gids) and this is okay.
> However when we add the length of the hostname to this, we are not adding the extra padded bytes for the hostname (If the length is not a multiple of 4) and thus when the NFS server reads the packet, it returns GARBAGE_ARGS because it doesn't read the uid field when it is expected to read. I can reproduce this issue constantly on machines where the hostname length is not a multiple of 4.
> A possible fix is to do something this:
> int pad = mHostName.getBytes().length % 4;
>  // mStamp + mHostName.length + mHostName + mUID + mGID + mAuxGIDs.count
> mCredentialsLength = 20 + mHostName.getBytes().length + pad;
> I would be happy to submit the patch but I need some help to commit into mainline. I haven't committed into Hadoop yet.
> Cheers!
> Pradeep


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

---------------------------------------------------------------------
To unsubscribe, e-mail: common-issues-unsubscribe@hadoop.apache.org
For additional commands, e-mail: common-issues-help@hadoop.apache.org