Linux host kernel panic when VirtualBox 4.2.16 mounts ISO images stored in NFS shares

[quarkysg]

When I tried to mount an ISO image, stored in a NFS share, for any guest in VirtualBox 4.2.16, the host CentOS 5 server will kernel panic. Linux kernel will panic when VirtualBox tries to load the ISO image.

Copying the ISO image to a local drive will not panic the kernel when VirtualBox tries to load the ISO image and the guest OS will be able to use it as a disc in the CDROM.

The CentOS 5.9 server is updated to the latest patches using yum.

Unable to provide core dump as Linux does not store dumps when it panics.

Linux kernel call trace I saw on the host server as follows:

Call Trace: [<ffffffff800f4dbc>] aio_put_req+0x22/0x30 [<ffffffff800f5a88>] io_submit_one+0x45a/0x486 [<ffffffff800f6549>] sys_io_submit+0xcf/0x1b7 [<ffffffff8005d29e>] tracesys+0xd5/0xdf

[quarkysg]

Kernel panic screen shot

[Alex]

Hello,

Can't reproduce any crash on CentOS 5 host, mounting iso from NFS share. Can you post more details here please e.g. VBox.log before a crash, NFS mounting options and maybe any other important information which can help to reproduce this.

Thanks.

[quarkysg]

Hi,

My CentOS 5 (IBM x3550M3) VirtualBox host is connected directly to the NFS server (IBM x3650, which is also installed with CentOS 5), via gigabit ethernet and the MTU is set to 9000, without any network switch in between. The NFS server is also acting as an iSCSI Target server, using the CentOS 5 tgtd iSCSI implementation. NFS v3 is used as I cannot get NFS v4 to work between the two servers.

The host server's /etc/fstab has the following entry:

oscvms02-int:/u01 /u04 nfs _netdev,rsize=32768,wsize=32768,intr 0 0

The NFS server's /etc/exports file has this single entry:

/u01 172.17.1.1(rw,no_root_squash)

I'm using the 172.17.1.x/24 IP ranges for the network interface connecting the two servers, but I don't think it will cause any issue.

The VBox.log file does not show any trace of the crash as it only has the guest startup messages logged and that's it, and /var/log/messages also has no trace of the kernel panic.

The VBox.log file for the newly created guest that caused the kernel panic as listed below:

VirtualBox VM 4.2.16 r86992 linux.amd64 (Jul 4 2013 18:11:55) release log
00:00:00.759795 Log opened 2013-08-29T02:21:07.187524000Z
00:00:00.759802 OS Product: Linux
00:00:00.759803 OS Release: 2.6.18-348.16.1.el5
00:00:00.759803 OS Version: #1 SMP Wed Aug 21 04:00:25 EDT 2013
00:00:00.759992 Host RAM: 64433MB total, 64028MB available
00:00:00.759995 Executable: /usr/lib/virtualbox/VirtualBox
00:00:00.759995 Process ID: 8674
00:00:00.759996 Package type: LINUX_64BITS_EL_5
<kernel panic as VirtualBox tries to load the ISO image over NFS>

The kernel panic is 100% repeatable with my setup, but I could not get any useful logs.

Thanks.

[Frank Mehnert]

We tried to reproduce your problem but were not able to do so. Could you add the .vbox file of that VM to this ticket? There might be some different VM setting ...

[quarkysg]

VirtualBox .vbox file used to test the issue

[quarkysg]

Hi,

Uploaded the .vbox file used to test the issue. Btw, it doesn't matter with guest is used. All VirtualBox guests will caused the kernel panic for me when it tries to load ISO images located in my NFS share. I had to place the ISO images in local disk partitions to workaround this issue.

Thanks.

[Frank Mehnert]

To me this looks like a host kernel bug. VirtualBox does async I/O by default but this code path is not well tested in older Linux kernels and some file systems (e.g. NFS) don't implement all functions correctly. I assume that your VM will not crash if you disable async I/O for the controller where the CD image is attached.

[aeichner]

Please reopen if still relevant with a recent VirtualBox release.