vboxnetflt kernel: Oops: 0000 [1] SMP

[Blaine Palmer]

The server is running Debian Etch 4, debian 2.6.18-6-amd64 SMP kernel. Processor type is Intel(R) Core(TM)2 Quad CPU Q9300 @ 2.50GHz.

Install progressed normally, modules loaded as per lsmod, and no errors until Guest VM is started. Upon starting the Guest VM (WS2k8 Standard) a console alert appears.

Message from syslogd@over9000 at Wed Dec 24 03:34:53 2008 ... 
over9000 kernel: Oops: 0000 [1] SMP 

Message from syslogd@over9000 at Wed Dec 24 03:34:53 2008 ... 
over9000 kernel: CR2: ffffffffffffff58 

Message from syslogd@over9000 at Wed Dec 24 03:34:53 2008 ... 
over9000 kernel: Oops: 0000 [2] SMP 

Message from syslogd@over9000 at Wed Dec 24 03:34:53 2008 ... 
over9000 kernel: CR2: ffffffffffffff58

Following this alert general system stability appears unaffected and the VM continues to boot as normal. Once inside the VM it becomes obvious that no networking capability through the Host Interface is working. VM configuration will be attached.

Further investigation into the kernel Oops reveals that this appears to be related to the vboxnetflt module.

---snip---
Dec 24 03:34:53 over9000 kernel: RIP: 0010:[<ffffffff8834772d>]  [<ffffffff8834772d>] :vboxnetflt:vboxNetFltLinuxXmitTask+0x1c/0x18e 
---snip---
Dec 24 03:34:53 over9000 kernel:  <1>Unable to handle kernel paging request at ffffffffffffff58 RIP:
---snip---

The full kernel log error will be attached.

PLEASE NOTE: This error occurs regardless of the guest operating system, or host interface chosen. It occurs as soon as you start a VM with -nic1 hostif via VBoxHeadless (I assume any method of starting the VM but as this machine does not even have a console/monitor there is no way to check X11 or SDL.)

[Blaine Palmer]

Kernel Log of the Oops

[Blaine Palmer]

Virtual Box log

[Blaine Palmer]

Guest VM Config

[Blaine Palmer]

Forgot to include that there is no difference between the General Linux AMD64 download and the Debian Etch 4 AMD64 download. I have tried uninstalling and re-installing both.

[Blaine Palmer]

The first error seen in the kernel log, ec 24 03:28:53 over9000 kernel: conftest[6062]: segfault at 00002b4e23c9204c rip 00002b4e2358fadb rsp 00007fff87752b70 error 4 This occurs after attempting a reboot following the kernel Oops, it appears that the kernel oops destabilizes the entire system.

[Michael]

Fedora 10 x86_64 2.6.27.7

[Michael]

I'm also seeing this OOPS. HOST = Fedora 10 x86_64 2.6.27.7, GUEST = Windows XP SP3

It only occurs when I'm attempting to transfer large files between the HOST and GUEST through a windows share. Web browsing or small files do not cause any OOPS.

After the OOPS occurs, the HOST and GUEST systems becomes unstable. Networking on the GUEST system stops completely. If I try to shut down the GUEST it will sometimes lock up the HOST as well. A hard reset is required.

[Michael]

I need to mention that large files are anything above about 2 megs. I do not know an exact size. Also, I am using Host Interface networking on eth0.

This doesn't only happen between host and guest. I attempted to transfer a large file (600 megs) from one computer to the guest computer and the OOPS occurred a few megabytes into the transfer. After that, the guest network connection is broken.

[Blaine Palmer]

I'll also point out that there is a forum post on the debian forums concerning the same problem, ​http://forums.debian.net/viewtopic.php?p=196943

It seems to be specific to x86_64 systems, if anyone has this issue (same Oops error and issue linked to vboxNetFltLinuxXmitTask) on a non-x86_64 kernel PLEASE COMMENT.

[Alexander]

I get the same OOPS:

Debian etch running a 2.6.18-6-amd64 kernel on an Intel P4 64-bit CPU @ 2.66GHz (cpu family 15, model 4). Yes, that's a SMP kernel on a single-CPU machine. (Debian sets it up that way, and I'm too busy to worry about it...until now?)

VirtualBox 2.1.0-41146_Debian_etch installed via deb package.

Guest OS: Windows 2000 SP4. Settings: ACPI on, IO APIC off, VT-x/AMD-V unavailable, PAE/NS on, 3D on. (Same results observed on different virtual machine with ACPI off.) Network adapter: PCnet-FAST III(NAT)

NAT networking works fine.

If host networking is enabled, host has kernel OOPS and keyboard ceases to function. Can shut down guest OS via GUI. Shutting down host OS after OOPS (either using GUI, or via timed shutdown process started before the OOPS) hangs on trying to unload vboxnetflt.

Additionally, under NAT networking, VirtualBox/W2000 is very finicky. Shared folders MUST be mapped to a drive letter, or guest machine will freeze up trying to access share. Guest machine will also freeze up when browsing for shared folders in network neighborhood. Note that with a drive letter mapped shared folders work fine (installed Visual Studio from a shared folder).

Once or twice, the guest machine has aborted during heavy network use. (SQL Server Management Studio accessing remote database to which host is connected via PPTP.)

Also: Numerous failures to install W2000 directly from CD. Must make ISO file and mount that for install to succeed (too much latency on CD???).

[Aleksey Ilyushin]

The problem appears with pre-2.6.20 kernels only. The fix will be included into the next maintenance release. Since netflt driver always comes with source code those who urgently need the fix may apply the following patch to src/vboxnetflt/linux/VBoxNetFlt-linux.c (the path is relative to your vbox installation directory):

1031c1031
<     INIT_WORK(&pThis->u.s.XmitTask, vboxNetFltLinuxXmitTask, NULL);
---
>     INIT_WORK(&pThis->u.s.XmitTask, vboxNetFltLinuxXmitTask, &pThis->u.s.XmitTask);

Then, kernel modules need to be re-built with

/etc/init.d/vboxdrv setup

[Michael]

aleksey, please read my comment. I'm running kernel 2.6.27 and I'm seeing an OOPS. I don't think this is only pre-2.6.20 kernels.

[Aleksey Ilyushin]

mooninite, right, but your problem is different and my fix won't solve it. In your case the fault happens in skb_put (looks like allocated skb is too small to fit the packet). I will look into this issue as well. What are the MTU sizes on both host and guest btw?

[Michael]

The MTU of the host is 1500. The guest is 1480 (windows maximum).

I have even tried switching from the Fast-III adapter to the Intel 1000 MT adapter. Same result.

[Lubomir Rintel]

I'm triggering this (the skb_put() panic, my module is patched with fix with the NULL dereference issue) regularly on RHEL-5 (2.6.18-based kernel) running on x86_64 architecture.

I am hitting the problem ~ once a day, have a kdump and debugging symbols, so I could theoretically help with tracing this down (if I were familiar with the code :) In case there's some piece of information I could provide to help debugging this, just let me know.

[Frank Mehnert]

Sure. If you have a core dump, please contact me at frank _dot_ mehnert _at_ sun _dot_ com. Thank you!

[Aleksey Ilyushin]

The latest and greatest version. It hopefully solves both OOPS issues: NULL derefernce and skb_over_panic.

[Aleksey Ilyushin]

mooninite, can try the patched version of VBoxNetFlt-linux.c I attached to the ticket? Assuming you've installed a standard package you'll need to copy the patched file to

<vbox_installation_dir>/src/vboxnetflt/linux/

(you may want to backup the original one just in case) and rebuild modules with

sudo /etc/init.d/vboxdrv setup

[Michael]

OK, will do. It will be Monday until I can test it on the original problem machine. I took today off.

[Michael]

Replying to aleksey:

mooninite, can try the patched version of VBoxNetFlt-linux.c I attached to the ticket? Assuming you've installed a standard package you'll need to copy the patched file to

That worked. I transferred a 700 meg file, 70 meg file, and three files totalling 2.1 gigs. It would have never worked on any of those files before your fix. Thanks for the fix.

[Aleksey Ilyushin]

Replying to mooninite:

Replying to aleksey:

mooninite, can try the patched version of VBoxNetFlt-linux.c I attached to the ticket? Assuming you've installed a standard package you'll need to copy the patched file to

That worked. I transferred a 700 meg file, 70 meg file, and three files totalling 2.1 gigs. It would have never worked on any of those files before your fix. Thanks for the fix.

Thanks for trying out the fix! 2.1.2 is out, could you try it and confirm that the problem does not appear anymore?

[Aleksey Ilyushin]

The fix is included into 2.1.2. Please re-open the ticket against 2.1.2 if the problem persists.