[SOLVED] openrc / baselayout-2 / udev / aoe / diskless

Message

Moasat · Post by **Moasat** » Tue May 10, 2011 2:07 pm

I have a diskless machine that boots over AoE. I have a custom initrd that simply discovers the aoe device, then mounts it read-only and does a chroot to it. After the upgrade to baselayout2/openrc, udev complains about not starting correctly because /dev/zero does not exist. If I interrupt the startup after udev, I see the /dev contains only very few actual devices. It seems udev started but mounted over my initial /dev that had way more devices in it.

I've tried updating udev to the latest ~amd64 but it did not help. I'm running 2.6.37-gentoo-r4 and have the deprecated sysfs options disabled. Everything worked before the upgrade so I'm not sure what might have broken.

[edit] Interrupting the startup yields a rootfs that appears to be mounted rw but when I try to edit a file, it says that it is read-only. Maybe the roots is not getting remounting properly? I have root, procfs, swap, fsck and mtab in my boot runlevel.

Can anyone offer some ideas or point to a direction to start looking? I'm really not familiar with baselayout, openrc or udev and I considered myself lucky in getting the initrd to work correctly.

Thanks for any and all suggestions.

Ant P. · Post by **Ant P.** » Tue May 10, 2011 3:19 pm

Is /etc/mtab a symlink to /proc/self/mounts? Is devtmpfs enabled in the kernel?

Moasat · Post by **Moasat** » Tue May 10, 2011 3:29 pm

mtab is not a symlink. devtmpfs is not enabled in the kernel. I see two options: Enable and Automount. Should both of these be on? The help for automount says something about possibly needing to mount /dev in an initramfs environment which is pretty much what I'm in.

I enabled both options and the system seems to be running correctly. (I'm testing in a VM since I don't have physical access to the hardware at the moment).

Thanks for the pointer though. I knew nothing about devtmpfs and it possibly being required for udev to work in a diskless environment.

Moasat · Post by **Moasat** » Thu May 12, 2011 1:37 pm

It turns out devtmpfs was not needed and actually caused some weird issues with the nvidia binary driver. When the driver loaded, it would not create the /dev/nvidia* nodes. Even creating them by hand didn't help.

I managed to get my initramfs working without requiring devtmpfs. I was using chroot at the end of the script and instead I changed it to use switch_root (and busybox). Here is the final init script for anyone interested.

#!/bin/busybox sh

# Mount /proc and /sys
/bin/echo Mounting /proc filesystem
/bin/mount -t proc none /proc

/bin/echo Mounting sysfs
/bin/mount -t sysfs none /sys

# Bring the network interface up
/bin/echo Bringing up eth0
/bin/ifconfig eth0 up
/bin/dhcpcd eth0

# Find real_root
for cmd in $(cat /proc/cmdline) ; do
case $cmd in
root=*)
REAL_ROOT=$(echo $cmd | cut -d= -f2)
;;
esac
done

/bin/echo Making AoE devices
/bin/mkdir -p /dev/etherd
/bin/aoe-mkdevs /dev/etherd
/bin/echo Telling AoE to discover.
/bin/echo omgdiscover > /dev/etherd/discover

/bin/echo "Waiting for ${REAL_ROOT} to appear"
for cc in `seq 0 300`; do
if [ -b "${REAL_ROOT}" ]; then
break
fi
/bin/sleep 0.1
done

# Wait for things to settle a bit
/bin/sleep 1

/bin/echo "Mounting ${REAL_ROOT}"
/bin/mount -o ro ${REAL_ROOT} /sysroot

if [ "$?" -ne "0" ]; then
echo "Error mounting root"
echo "I'm going to try rebooting"
/bin/shutdown -r now
exit 1
fi

/bin/echo Creating ramdisks
mount -n -t tmpfs tmpfs /sysroot/tmp
mount -n -t tmpfs tmpfs /sysroot/var/lib/init.d
mount -n -t tmpfs tmpfs /sysroot/var/lib/xkb
mount -n -t tmpfs tmpfs /sysroot/var/lib/dbus
mount -n -t tmpfs tmpfs /sysroot/var/lib/dhcpcd
mount -n -t tmpfs tmpfs /sysroot/var/log
mount -n -t tmpfs tmpfs /sysroot/var/run
mkdir -p /sysroot/var/run/lirc
mount -n -t tmpfs tmpfs /sysroot/var/run/lirc
mount -n -t tmpfs tmpfs /sysroot/var/lock
mount -n -t tmpfs tmpfs /sysroot/var/cache/hald

# Clean up.
umount /proc
umount /sys

# Boot the real thing.
exec switch_root /sysroot /sbin/init

SlashBeast · Post by **SlashBeast** » Thu May 12, 2011 6:01 pm

If you use devtmpfs, you should do mount --move /dev /newroot/dev. nvidia nodes are created when nvidia module is loaded. Here I am using nvidia binary drivers and my initramfs (which use devtmpfs) as I have rootfs over encrypted lvm.

Also, use

Code: Select all

 tags.

FWIW instead of mounting so many tmpfs you could just use aufs/unionfs-fuse and store rw branch on tmpfs.

pi3.14 · Post by **pi3.14** » Wed May 25, 2011 10:59 am

Hello,
I upgraded Gentoo to OpenRC/baselayout-1, but I am using 2.6.32 OpenVZ kernel, where CONFIG_DEVTMPFS option is missing.
Without it gentoo won't boot. When I use newer kernel, where devtmpfs is available it boots normaly.
Any ideas how to get it work without devtmpfs?

pablolibo · Post by **pablolibo** » Mon May 30, 2011 12:08 am

pi3.14 wrote:Hello,
I upgraded Gentoo to OpenRC/baselayout-1, but I am using 2.6.32 OpenVZ kernel, where CONFIG_DEVTMPFS option is missing.
Without it gentoo won't boot. When I use newer kernel, where devtmpfs is available it boots normaly.
Any ideas how to get it work without devtmpfs?

Pi3.14

you can enable the DEVTMPFS in kernel option from:

/usr/src/linux-2.6.32-openvz-feoktistov.1/drivers/Kconfig

chenge:

config DEVTMPFS
bool "Create a kernel maintained /dev tmpfs (EXPERIMENTAL)"
depends on HOTPLUG && SHMEM && TMPFS && !VE

to:

config DEVTMPFS
bool "Create a kernel maintained /dev tmpfs (EXPERIMENTAL)"
depends on HOTPLUG && SHMEM && TMPFS

and now, you can enable the DEVTMPFS from openvz-source

but, it is bad idea, because it feature was disable for an report bug.

anyway, i am testing it bug

pi3.14 · Post by **pi3.14** » Mon May 30, 2011 5:10 am

I figured it without DEVTMPFS.
When using DEVTMPFS dev directory can be empty. When there isn't DEVTMPFS enabled it is necessary create
mknod -m 660 /dev/console c 5 1
mknod -m 660 /dev/null c 1 3
and then booting fine.