hadoop-common-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Steve Loughran (JIRA)" <j...@apache.org>
Subject [jira] Commented: (HADOOP-6059) Should HDFS restrict the names used for files?
Date Wed, 17 Jun 2009 10:56:07 GMT

    [ https://issues.apache.org/jira/browse/HADOOP-6059?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12720616#action_12720616

Steve Loughran commented on HADOOP-6059:

+1 for restrictions. 

I'd go for names that allow lots of 

* all valid HDFS names are valid within XML files. That is, at a minimum, the only values
<ASCII 32 are tab, cr, and lf. And I can think of some good reasons to stop that too. No
< or > either.
* all valid HDFS names are valid within string database tables. 
* All valid names can be represnted with strings in JSON documents, possibly with some escaping
* the normal POSIX forbidden paths are still forbidden

I have no pressing need for XML, JSON or in-database representation, but I can imagine it
being useful in the future. Valid XML can also be used inside HTML reports..you don't want
to do XSS tricks by creating filenames with <script> in their name to try and catch
out anyone browsing the directory tree

> Should HDFS restrict the names used for files?
> ----------------------------------------------
>                 Key: HADOOP-6059
>                 URL: https://issues.apache.org/jira/browse/HADOOP-6059
>             Project: Hadoop Core
>          Issue Type: New Feature
>          Components: dfs
>    Affects Versions: 0.20.0
>            Reporter: Robert Chansler
> When reviewing the consequences of Hadoop:6017 (the name system could not start because
a file name interpreted as a regex caused a fault), the discussion turned to improving the
test set for file system functions by broadening the set of names used for testing. Presently,
HDFS allows any name without a slash. _Should the space of names be restricted?_ If most funny
names are unintended, maybe the user would benefit from an early error indication. A contrary
view is that restricting names is so 20th-century.
> Should be or shouldn't we?

This message is automatically generated by JIRA.
You can reply to this email to add a comment to the issue online.

View raw message