aufs root filesystem

[Ebrius]

I can no longer start my dhcpd service.

I am using dhcp3-server and when I start it I get the following error:
dhcpd self-test failed. Please fix the config file.
The error was:
/usr/sbin/dhcpd3: error while loading shared libraries: libcap.so.2: cannot open shared object file: No such file or directory

And dmesg shows the following errors:
[ 61.388679] type=1503 audit(1258333308.864:16): operation="open" pid=969 parent=966 profile="/usr/sbin/dhcpd3" requested_mask="r::" denied_mask="r::" fsuid=0 ouid=0 name="/ro/etc/ld.so.cache"
[ 61.389098] type=1503 audit(1258333308.864:17): operation="open" pid=969 parent=966 profile="/usr/sbin/dhcpd3" requested_mask="r::" denied_mask="r::" fsuid=0 ouid=0 name="/ro/lib/libcap.so.2.16"
[ 61.392012] type=1503 audit(1258333308.864:18): operation="open" pid=970 parent=966 profile="/usr/sbin/dhcpd3" requested_mask="r::" denied_mask="r::" fsuid=0 ouid=0 name="/ro/etc/ld.so.cache"
[ 61.392559] type=1503 audit(1258333308.868:19): operation="open" pid=970 parent=966 profile="/usr/sbin/dhcpd3" requested_mask="r::" denied_mask="r::" fsuid=0 ouid=0 name="/ro/lib/libcap.so.2.16"


I see this regardless if I have /ro mounted in read/write or read-only mode. Anyone have a solution?

[cujgldkjge]

It's been a while since traffic in this thread, but I thought I'd ask a question...

This seems to be a pretty elegant solution, but what about integrating it in with runlevels?

It seems that most systems just use two (1 and 2). What if we used runlevel 3 as read-only with volatile changes, and runlevel 2 as full-on rw mode. This would allow one to switch into rw mode, do an aptitude upgrade, and then switch back, without having to reboot and select the grub option.

When there's a shutdown, things could sync, too, of course, but that's another discussion. I suppose some people might like the changes to be wiped, others would want them sync'd.

Thoughts?

Send me your fstab. I'm new to sed as well. It's just so fast to make one liner changes.
I will have to review my notes. I need to edit fstab to keep the mounts in line. Without the changes df or mount report the wrong information.

[Ebrius]

I can no longer start my dhcpd service.

I am using dhcp3-server and when I start it I get the following error:
dhcpd self-test failed. Please fix the config file.
The error was:
/usr/sbin/dhcpd3: error while loading shared libraries: libcap.so.2: cannot open shared object file: No such file or directory

And dmesg shows the following errors:
[ 61.388679] type=1503 audit(1258333308.864:16): operation="open" pid=969 parent=966 profile="/usr/sbin/dhcpd3" requested_mask="r::" denied_mask="r::" fsuid=0 ouid=0 name="/ro/etc/ld.so.cache"
[ 61.389098] type=1503 audit(1258333308.864:17): operation="open" pid=969 parent=966 profile="/usr/sbin/dhcpd3" requested_mask="r::" denied_mask="r::" fsuid=0 ouid=0 name="/ro/lib/libcap.so.2.16"
[ 61.392012] type=1503 audit(1258333308.864:18): operation="open" pid=970 parent=966 profile="/usr/sbin/dhcpd3" requested_mask="r::" denied_mask="r::" fsuid=0 ouid=0 name="/ro/etc/ld.so.cache"
[ 61.392559] type=1503 audit(1258333308.868:19): operation="open" pid=970 parent=966 profile="/usr/sbin/dhcpd3" requested_mask="r::" denied_mask="r::" fsuid=0 ouid=0 name="/ro/lib/libcap.so.2.16"


I see this regardless if I have /ro mounted in read/write or read-only mode. Anyone have a solution?

I wasn't sure if anyone had seen this before, but I did a little more testing and I found that when I tried to 'chmod' one of those files it would fail, first aufs would throw an error (something about how permissions should be reversed) and then chmod would throw an I/O error. What is kinda strange is I could modify other file's permissions (a couple in the /bin directory) successfully.

I can capture some more data if someone is interested in looking at this, its just a bit of a pain since this server is my router, and when I boot to aufs my dhcp server does not work.

[philipp2084]

Have a look at this thread. http://ubuntuforums.org/showthread.php?t=1220145

This seems to be a common problem.

[TimSylvester]

I'm using this aufs-root script and it's working well, thank you for putting it together.

One minor issue I have is that when I run remountrw and remountro to make a minor change to the system, it seems to lock out any further changes. If I later run remountrw again, I get:

mount: /ro not mounted already, or bad option

This doesn't seem to have anything to do with the remount scripts themselves or the use of UUIDs, as I get the same result with a manual command:

$ sudo mount -o remount,rw /ro
mount: /ro not mounted already, or bad option
$ sudo mount -o remount,rw /dev/sda1
mount: /ro not mounted already, or bad option

Unfortunately, the only way I've found to get past this is to reboot, or leave /ro mounted read-write permanently.

Any ideas?

[falconindy]

The output of mount can be flakey as its based on /etc/mtab. The info provided by /proc/mounts is the accurate.

I've had issues with Ext4 and Aufs, but I think they were related to kernel 2.6.32 more than Aufs. In any case, the solution was to mount the Ext4 partition with the 'nobarrier' option. You might want to try this.

[TimSylvester]

Yes, that did it. Thanks!

For the record, I added "rootflags=barrier=0" to the kernel command line. ("nobarrier" seems to be the XFS equivalent). Next step is to update grub2 so it happens automatically...

[dave373]

I am trying to create a read-only root version of Ubuntu Karmic but am having difficulties with Grub2 not selecting the default kernel.

With a normal fstab, everything works perfectly, but after switching to a Read-only AUFS fstab, Grub2 will not boot automatically, it sits there and waits for an 'enter' keystroke

As the machine will be locked in a box, headless, with no keyboard, this is a real problem.

Is anyone familiar with this situation, or should I just roll-back to grub legacy...

Cheers for any help

Dave

[cmcginty]

Dave, funny you posted this question. I was just going to submit the say the same thing. I have a little more info.

The reason why grub2 hangs at boot (after working before) is because it writes a value to /boot/grub/grubenv. You can clear the value with the following command:

grub-editenv /boot/grub/grubenv unset recordfail

I'm running my system in a VM right now, seeing the same issues. This bug means that grub2 is writing to the FS, even though it should be RO. So I'm trying figure out how to prevent grub2 from doing this, or maybe I should just roll back as well.

Doesn't anyone have any ideas?

[cmcginty]

Here is the fix to prevent Grub hanging during boot:

sudo rm /boot/grub/grubenv

Now my system always boots up automatically.