[CentOS] tigervnc-server-module crashes after EL 6.3 update

Fri Aug 17 18:40:43 UTC 2012
Cal Webster <cwebster at ec.rr.com>

On Thu, 2012-08-16 at 17:01 -0500, Johnny Hughes wrote:
> On 08/16/2012 11:43 AM, Cal Webster wrote:
> > On Wed, 2012-08-15 at 13:56 -0500, Johnny Hughes wrote:
> >> On 08/15/2012 09:47 AM, Cal Webster wrote:
> >>> On Tue, 2012-08-14 at 20:55 -0500, Johnny Hughes wrote:
> >>>> On 08/14/2012 05:23 PM, Cal Webster wrote:
> >>>>> We began experiencing failed vnc connections to the console display on
> >>>>> servers that have been updated to EL 6.3. No such failures have occurred
> >>>>> on similar connections to EL 6.2 servers.
> >>>>>
> >>>>> On the client machine a normal vncviewer display appears with the
> >>>>> expected graphical login until the mouse pointer is moved within the
> >>>>> boundaries of the vncviewer window. At this point the window closes and
> >>>>> an error message appears in both a pop-up window and in the terminal
> >>>>> window in which the session was initiated stating "read: Connection
> >>>>> reset by peer (104)".
> >>>>>
> >>>>> On the server end, a core dump is generated and a abrt bug report is
> >>>>> created.
> >>>>>
> >>>>> /var/log/messages
> >>>>> ----------------------------------------------
> >>>>> Aug 14 11:00:30 jato2 abrt[11411]: File '/usr/bin/Xorg' seems to be
> >>>>> deleted
> >>>>> Aug 14 11:00:30 jato2 abrt[11411]: Saved core dump of pid 7892
> >>>>> (/usr/bin/Xorg) to /var/spool/abrt/ccpp-2012-08-14-11:00:30-7892
> >>>>> (42041344 bytes)
> >>>>> Aug 14 11:00:30 jato2 abrtd: Directory 'ccpp-2012-08-14-11:00:30-7892'
> >>>>> creation detected
> >>>>> ----------------------------------------------
> >>>>>
> >>>>> This bug has been reported in the CentOS bug tracker here:
> >>>>>
> >>>>> 0005824: tigervnc-server-module keep crashing
> >>>>> http://bugs.centos.org/view.php?id=5824
> >>>>>
> >>>>> However, this appears to be a bug upstream. The source RPM provided with
> >>>>> CentOS is identical to that of upstream with no modifications. Also,
> >>>>> there is an upstream bug reported that appears to have the same
> >>>>> symptoms. I have added a comment to the upstream bug report (listed
> >>>>> below) if anyone wishes to see the details.
> >>>>>
> >>>>> tigervnc-server-module crashes with dual screen setup
> >>>>> https://bugzilla.redhat.com/show_bug.cgi?id=820443
> >>>>>
> >>>>>
> >>>>>
> >>>>> We have verified that rebuilding the unmodified source RPM for tigervnc
> >>>>> produces a tigervnc-server-module RPM that does not suffer from this
> >>>>> bug.
> >>>>>
> >>>>> Removing the original tigervnc-server-module package and replacing it
> >>>>> with the rebuilt one fixes the problem.
> >>>>>
> >>>>> I've duplicated the problem on 2 EL 6.3 x86_64 single-head display
> >>>>> machines and have verified the fix.
> >>>>>
> >>>>> Tomorrow, I'll duplicate the problem on a dual-head x86_64 machine that
> >>>>> currently still works after updating to EL 6.2 then confirm the the fix.
> >>>>>
> >>>> Did you rebuild the SRPM using mock or directly on a physical machine
> >>>> with rpmbuild?
> >>> No mock, just a simple "rpmbuild -ba SPEC/tigervnc.spec"
> >> OK, if you find that this solves your problems for sure, I will build
> >> the SRPM outside of mock and see if it is different.
> > I've confirmed the same faulty behavior for the update to 6.3 on our
> > dual-head systems.
> >
> > Also confirmed is that replacing the 6.3 base tigervnc-server-module rpm
> > with the rebuilt one does fix the problem on the dual-head systems.
> >
> > One disturbing difference between single and dual headed systems is that
> > on the dual-head systems Xorg generates a core dump and completely
> > freezes up when the mouse movement is detected. Single-head systems just
> > fail to connect. This complication could be somehow caused by our
> > proprietary "ATI FirePro 2270" drivers, though. Once the rebuilt module
> > is installed the systems run fine.
> >
> > I've also updated the upstream bug report.
> 
> can you see if either or both of these work for you:
> 
> http://people.centos.org/hughesjr/tigervnc/
> 
> One set was built inside of mock, the other outside of mock in a virtual
> machine with only the build requirements of the SRPM installed.

Both builds work without problems on single and dual-head systems here.
As with all the other tests, I only replaced the tigervnc-server-module
package on each host.

I've also confirmed that i686 platforms suffer from the same bug. These
too, however, are easily remedied by replacing the base
tigervnc-server-module RPM with a locally re-build one.

./Cal