[ 
https://issues.apache.org/jira/browse/HADOOP-10989?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
 ]

Chris Nauroth updated HADOOP-10989:
-----------------------------------

    Attachment: HADOOP-10989.2.patch

Thanks, Colin.  Here is patch v2 with the ifdef inverted to focus on the Linux 
special case and more comments.

bq. Just out of curiosity, was there a specific operating system or environment 
you found that had this problem?

We saw this happening in 2 different environments, one RHEL 6.2 and the other 
RHEL 6.3.  Upgrading to RHEL 6.4 solved it, so we've recommended upgrading to 
RHEL 6.4 or later as a workaround.  The specific sequence of events involved 
nslcd issuing an LDAP query that timed out.  I suspect whatever version of 
nslcd shipped with RHEL 6.2/6.3 by default had a bug in its error handling, 
such that an LDAP timeout could result in returning 0 from {{getgrouplist}}, 
but leaving the other out parameters in an indeterminate state.  Then, we'd use 
those bad parameter values in a {{realloc}} and segfault.  Unfortunately, I no 
longer have the information on the specific nslcd version numbers.  I agree 
that it's not exactly accurate to blame an OS version.  It's more about the 
nslcd library version.

> Work around buggy getgrouplist() implementations on Linux that return 0 on 
> failure
> ----------------------------------------------------------------------------------
>
>                 Key: HADOOP-10989
>                 URL: https://issues.apache.org/jira/browse/HADOOP-10989
>             Project: Hadoop Common
>          Issue Type: Bug
>          Components: native
>    Affects Versions: 3.0.0, 2.6.0
>            Reporter: Chris Nauroth
>            Assignee: Chris Nauroth
>         Attachments: HADOOP-10989.1.patch, HADOOP-10989.2.patch
>
>
> HADOOP-10781 corrected the handling of the return value from {{getgrouplist}} 
> to work on FreeBSD.  However, it also regressed fixes that had been put in 
> place to work around issues with {{getgrouplist}} on Linux.  This issue will 
> restore that behavior, but still retain compatibility with FreeBSD.



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

Reply via email to