Bug 839788

Description of problem:

When trying to run PandoraFMS agent, Perl segfaults.  I have traced this to a bug as per the following URL:

http://stackoverflow.com/questions/5038337/why-does-my-threaded-perl-script-segfault



Version-Release number of selected component (if applicable):

perl.x86_64                        4:5.10.1-127.el6                        @base

[root@pandora pandora_server]# perl -Mthreads -e 'print "$threads::VERSION\n"'
1.82



How reproducible:

Extremely.  Here's a test command:

[root@pandora pandora_server]# perl -e "sub M::DESTROY; bless {}, M;"
Segmentation fault



Steps to Reproduce:
1. The following command specifically triggers this bug:

  perl -e "sub M::DESTROY; bless {}, M;"

Additionally, you could attempt to run PandoraFMS agent on x86-64 running RedHat 6.3.  It sends one alert, then segfaults the second round.


  
Actual results:

Segmentation fault.

perl[19896]: segfault at 0 ip 00007f41a5694d6c sp 00007fffe20e3050 error 4 in libperl.so[7f41a55dc000+162000]

pandora_agent[19528] general protection ip:7f69181eafd0 sp:7f690e55ba40 error:0 in libperl.so[7f6918156000+162000]

Note that in any case the fault is in libperl.so, 162000 bytes from the base of the library's load address.  The fault is thus extremely reproducible and can easily be traced by running the given perl one-liner through perl while watching with a debugger.  A newer version of perl may have the bug fixed already; current is 5.16.



Expected results:

Should work.


Additional info:

This particular issue should have been fixed in threads 1.73, yet it still exists in version 1.82 of Perl threads on RHEL 6.3.  This leads me to believe there's another coding error responsible that's not triggered on x86, but is on x86-64.