"Segmentation fault" and "Child termintated by signal" bugs

[mbiggerstaff]

Good morning.

I am attempting to start some images in "headless" mode (no client side UI). In my testing, I found that reconnecting seemed to be causing a segmentation fault. Below, I have list two way to show what I think are the same issue.

Host: Fedora 11 (update to date as of today) with VirtualBox 3.0.4 r50677
Test Client: Fedora 11 (update to date as of today) [smaller installation image than the host, just for testing purposes]

Scenario #1: VBoxManage startvm "Fedora 11 - Dev" --type vrdp
1) start booting image
2) connect to vrdp on standard port with tsclient - connects fine
3) disconnect client (at any point, not just on boot)
4) try to reconnect to same image and get "Child terminated by signal" error message from tsclient. The image is no longer running.

Scenario #2: VBoxHeadless -startvm "Fedora 11 - Dev"
1) start booting image
1.2) usually get this warning:

[WARN 14963] polkit-session.c:144:polkit_session_set_uid(): session != NULL
Not built with -rdynamic so unable to print a backtrace

2) connect to vrdp on standard port with tsclient - connects fine
3) disconnect client (at any point, not just on boot)
4) try to reconnect to same image and get "Segmentation fault" in the terminal window and "Child terminated by signal" error message from tsclient. The image is no longer running.

Of course, my goal is to be able to load once and reconnect many times to the VirtualBox image using vrpd. I assume these are bugs but if I am doing something wrong, can someone let me know what it is?

Regards,
Mike

[Perryg]

Not sure what is causing the unusual problem with your remote session, but everything that I have read about your warning refers to pulse audio and is not specific to just Fedora.
I have seen this in my fedora and I simply disabled the sound and purged the audio program.
Is there anyway you can get to the VBox log files to see if you can see a different error?
Also maybe look at the xorg.0.log to see if there is something there that would indicate what is causing the problem and the syslog.

[mbiggerstaff]

Attached Fedora 11 - Dev-2009-08-21-07-41-27 (VBoxManage).zip is the VirtualBox image log.

This log file was created during a connection test (described previously) using command Line: VBoxManage startvm "Fedora 11 - Dev" --type vrdp

In the log file:

• 1st connection (successful) at 00:01:27.853
  1st disconnection at 00:01:31.124
  2nd connection (attempt) at 00:03:10.763 with a "Child terminated by signal" error message from tsclient, VirtualBox (UI) shows image in "Aborted" status

In the attached log, I don't see any "smoking guns" other than it's abrupt end. It shows connection requests but only the first disconnection.

Xorg.0.log: I looked in Xorg.0.log. The only error I have is related to my video driver.

• (II) AIGLX error: dlopen of /usr/lib/dri/nouveau_dri.so failed (/usr/lib/dri/nouveau_dri.so: cannot open shared object file: No such file or directory)

I checked a few other logs with system log viewer but nothing stood out in review.

Attachment <unknown>-bugreport.zip is a bug buddy error log.
Possible (or possibly not) is a "bug buddy" crash report when an image is started and then shutdown using the VirtualBox UI. I doesn't seem to affect the power down of the image nor it's restart later and it does not affect the running VirtualBox UI. Just a "crash" on ever image shutdown.

Note: I have had some driver problems with my embedded NVidia video card and Fedora 11. I'm currently using the default currently and seems to work ok. I don't expect that video driver would have an impact on the functionality of connecting to the VirtualBox image (but I have been wrong before.) I do have some PulseAudio problems too, usually having to restart the daemon after a VirtualBox session. However, I don't think VirtualBox is the true cause of that issue nor do I think it would impact the vrdp connections.

Linux version details: (Fedora 11)

• Linux XXX 2.6.29.6-217.2.8.fc11.i586 #1 SMP Sat Aug 15 00:44:39 EDT 2009 i686 i686 i386 GNU/Linux

Thanks in advance for suggestions and for your time.
Mike

[Perryg]

Your attachments didn't seem to make it through. I can not open them and they both say not loaded yet.

[mbiggerstaff]

Perry / all,

I downloaded them from the forum and extracted the text files without issue. Perhaps there is a security process that made them temporarily unavailable. Please retry. If they still don't work for you, I'll <gulp> post the raw text into a post (?) ... that seems like a bad idea...

Your attachments didn't seem to make it through. I can not open them and they both say not loaded yet.

Regards,
Mike

[mbiggerstaff]

All,

I have replicated this behavior on a second Fedora 11 machine, running different hardware and no known video or sound driver issues, and with a different virtual machine image. I have also determine one more missing element to add to the test description: process ownership

Testing Scenario:

• I login in as User_A to the Fedora 11 Gnome UI.
  From a new terminal console, I execute: su -l User_B and start VBoxManage with VM image. So the VBoxManage is a User_B owned task.
  As a User_A in the Gnome UI, I use tsclient to connect to the VBoxManage instance running under User_B.

Note: Yes, this isn't the best approach but it's only for testing purposes. Either way, it should not segment fault.

While I don't expect this launch/process ownership test scenario to be a problem, this is a significant difference from the usual use-case (I suppose.)

Is anyone else using a similar setup (or testing this scenario) and having the same (bad) results?

Thanks for your assistance,
Mike

[mbiggerstaff]

All,

I am seeing this problem using a simple test of only VirtualBox, a VM setup up with the default port for vrdp and tsclient all from the same user. Seems somethings is broken in this release. I know this has worked on my machines in the past. I will look at the VirtualBox authorization services next.

Any one else verify these findings?

Regards,
Mike

[mbiggerstaff]

All,

The problem is with the Remote Desktop authentication:

• With a setting of "External", the rdp client crashes ("child terminated by signal") and the VM images segment faults.
  With a setting of "Guest", the rdp client crashes ("child terminated by signal") but the VM image stays alive.
  With a setting of "Null", everything works but there is no user authentication (as per this option's meaning.)

Note: The vrdp port number and process ownership have no impact on the crashes; only the authentication setting. (but a high port number seems to make the VM image run slower - to test that later.)

So, I am pretty sure the "External" authentication worked before with zero configuration. Other than VirtualBox installation and applying the 3.0.4 Extras to the image, I have may no other authentication related changes.

Any ideas what to check now?

Mike

[Perryg]

The external authentication is used to check the PAM for the user name that is trying to login.
On the host, you need to make sure that the /etc/pam.d/common-account has been edited to allow this.
Example, I use broke_shadow to achieve this. See line (2) below. Line (1) is the original and only there for demonstration of what it looked like before I added broken_shadow.

Code: Select all

Account required pam_permit.so
Account required pam_permit.so broken_shadow

I do not know of another way but I am sure there are some. Maybe someone else has another approach for us to try, but this was fast and works.

[mbiggerstaff]

Anyone have a "How-To" for configuring PAM [on Fedora 11?] in conjunction with VirtualBox?

Thanks,
Mike