[jira] [Commented] (HADOOP-8845) When looking for parent paths info, globStatus must filter out non-directory elements to avoid an AccessControlException

["Eli Collins (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/HADOOP-8845?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13466965#comment-13466965 ] 

Eli Collins commented on HADOOP-8845:
-------------------------------------

globPathsLevel is a generic method, and globStatus which calls it claims to return all matching path names, why is it OK to unconditionally filter out all files from its results?  Since * can match the empty string, in other contexts it could be appropriate to return ""/tmp/testdir/testfile" for "/tmp/testdir/*/testfile".

Ie is there a place where we know we should just be checking directory path elements? The comment in globStatusInternal ("// list parent directories and then glob the results") by one of the cases indicates is the intent but it's valid to pass both files and directories to listStatus.

                
> When looking for parent paths info, globStatus must filter out non-directory elements to avoid an AccessControlException
> ------------------------------------------------------------------------------------------------------------------------
>
>                 Key: HADOOP-8845
>                 URL: https://issues.apache.org/jira/browse/HADOOP-8845
>             Project: Hadoop Common
>          Issue Type: Bug
>          Components: fs
>    Affects Versions: 2.0.0-alpha
>            Reporter: Harsh J
>            Assignee: Harsh J
>              Labels: glob
>         Attachments: HADOOP-8845.patch, HADOOP-8845.patch, HADOOP-8845.patch
>
>
> A brief description from my colleague Stephen Fritz who helped discover it:
> {code}
> [root@node1 ~]# su - hdfs
> -bash-4.1$ echo "My Test String">testfile <-- just a text file, for testing below
> -bash-4.1$ hadoop dfs -mkdir /tmp/testdir <-- create a directory
> -bash-4.1$ hadoop dfs -mkdir /tmp/testdir/1 <-- create a subdirectory
> -bash-4.1$ hadoop dfs -put testfile /tmp/testdir/1/testfile <-- put the test file in the subdirectory
> -bash-4.1$ hadoop dfs -put testfile /tmp/testdir/testfile <-- put the test file in the directory
> -bash-4.1$ hadoop dfs -lsr /tmp/testdir
> drwxr-xr-x   - hdfs hadoop          0 2012-09-25 06:52 /tmp/testdir/1
> -rw-r--r--   3 hdfs hadoop         15 2012-09-25 06:52 /tmp/testdir/1/testfile
> -rw-r--r--   3 hdfs hadoop         15 2012-09-25 06:52 /tmp/testdir/testfile
> All files are where we expect them...OK, let's try reading
> -bash-4.1$ hadoop dfs -cat /tmp/testdir/testfile
> My Test String <-- success!
> -bash-4.1$ hadoop dfs -cat /tmp/testdir/1/testfile
> My Test String <-- success!
> -bash-4.1$ hadoop dfs -cat /tmp/testdir/*/testfile
> My Test String <-- success!  
> Note that we used an '*' in the cat command, and it correctly found the subdirectory '/tmp/testdir/1', and ignore the regular file '/tmp/testdir/testfile'
> -bash-4.1$ exit
> logout
> [root@node1 ~]# su - testuser <-- lets try it as a different user:
> [testuser@node1 ~]$ hadoop dfs -lsr /tmp/testdir
> drwxr-xr-x   - hdfs hadoop          0 2012-09-25 06:52 /tmp/testdir/1
> -rw-r--r--   3 hdfs hadoop         15 2012-09-25 06:52 /tmp/testdir/1/testfile
> -rw-r--r--   3 hdfs hadoop         15 2012-09-25 06:52 /tmp/testdir/testfile
> [testuser@node1 ~]$ hadoop dfs -cat /tmp/testdir/testfile
> My Test String <-- good
> [testuser@node1 ~]$ hadoop dfs -cat /tmp/testdir/1/testfile
> My Test String <-- so far so good
> [testuser@node1 ~]$ hadoop dfs -cat /tmp/testdir/*/testfile
> cat: org.apache.hadoop.security.AccessControlException: Permission denied: user=testuser, access=EXECUTE, inode="/tmp/testdir/testfile":hdfs:hadoop:-rw-r--r--
> {code}
> Essentially, we hit a ACE with access=EXECUTE on file /tmp/testdir/testfile cause we tried to access the /tmp/testdir/testfile/testfile as a path. This shouldn't happen, as the testfile is a file and not a path parent to be looked up upon.
> {code}
> 2012-09-25 07:24:27,406 INFO org.apache.hadoop.ipc.Server: IPC Server
> handler 2 on 8020, call getFileInfo(/tmp/testdir/testfile/testfile)
> {code}
> Surprisingly the superuser avoids hitting into the error, as a result of bypassing permissions, but that can be looked up on another JIRA - if it is fine to let it be like that or not.
> This JIRA targets a client-sided fix to not cause such /path/file/dir or /path/file/file kinda lookups.

--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira