Migrating from static /dev to udev

Hi,
I'm trying to migrate an armeb embedded Linux build from a static /dev to udev. I now have the system coming up with udev enabled, but the first thing that udev does is disable my console connection. Before the console starts endlessly spewing "file not found" messages, udev prints a line like this

tty ttyAMA0: uevent: unsupported action-string; this will be ignored in a futur

for nearly every device that was in the previous static /dev.

I understand that I can add static rules to udev to fix this, but I also want to migrate naming conventions to standard names--ttyAMA0 should be ttyS0, etc. I can keep legacy names around with symlinks if necessary.

So I'm not quite sure how to get started. I have disabled udev's startup for the moment, so I can get into /sys. Is there a guide somewhere on how to migrate from a static /dev to udev?

Thanks,
Brendan

Hi,

when you only need a automatically managed /dev (and not need udev) you can try the first snapshots of the new 2.6.32 kernel with enabled devtmpfs. i have tryed this yesterday (linux-2.6.31 +patch
http://www.kernel.org/pub/linux/kernel/v2.6/next/patch-v2.6.31-next-20090925.bz2)
and its works very good for me. you can mount /dev with "mount -t devtmpfs none /dev" and you have a complete /dev managed by kernel (when you need udev and it works, udev is very fast with this).

Stephan

Am 26.09.2009 00:38, schrieb Brendan Moran:

               ### OpenELEC.tv ###
The free and open Mediacenter Distribution 4 you
              http://www.openelec.tv

Hi

Ow. Often that means you've either deleted or removed access permissions to /dev/null. Maybe the same has happened to /dev/tty or something as well. It suggests that your rules are failing to assign permissions to these devices, so udev is giving them default permissions which are too restrictive.

Aha. That's a kernel message, not a udev one. It suggests you're doing this:

udevadm trigger --action=uevent

when you should be doing this:

udevadm trigger --action­d

("udevadm trigger" should also work since --action­d is the default).

I suspect this is not relevant to your problems. Udev doesn't delete the previous contents of /dev. There should only be problems if the names conflict, so that udev overwrites one of your existing nodes with something you're not expecting.

(Btw, in case you're not already doing this, the README recommends you mount /dev as a tmpfs and initialize it by copying all your static devices from /lib/udev/devices).

Or you can have the kernel mount it for you if you want to based on the config option, making your userspace even simpler.

thanks,

greg k-h

Hi Alan,
I'm using virtually the same startup script as is used in Ubuntu (attached). It only uses $udevadm trigger, with no arguments. I didn't expect that permissions would be an issue, as this system is running in single user mode.

Mounting /dev as a tmpfs makes perfect sense to me, however I'll have to do some hacking before I can modify that. I have a list of all the devices to be created and placed in /dev already. Currently, it's expanded at compile time, but I should be able to change the target from /dev to /lib/udev/devices.

One additional thing that I need to do while migrating away from a static /dev is to convert our current naming scheme for devices to one which is consistent with the naming that udev uses by default.

Thanks again,
Brendan

#!/bin/sh -e
### BEGIN INIT INFO
# Provides: udev
# Required-Start: mountkernfs
# Required-Stop:
# Should-Start:
# Default-Start: S
# Default-Stop:
# Short-Description: Start the udev daemon.
# Description: Mounts the /dev virtual filesystem, starts the udev
# daemon and populates /dev.
### END INIT INFO

# Check the package is still installed
[ -x /sbin/udevd ] || exit 0

# Get LSB functions
. /lib/lsb/init-functions
if [ -e /etc/default/rcS]; then
. /etc/default/rcS
fi

case "$1" in
start)
DEVMOUNT=mount | grep "on /dev " | cut -d " " -f 1 # We need the uevent support introduced in 2.6.15, bail out if we # don't have it and fall back to a static /dev
if [ ! -f /sys/class/mem/null/uevent ]; then
if [ -n $DEVMOUNT ]; then
# uh-oh, initramfs made some kind of /dev, get rid of it umount -l /dev
fi
exit 1
fi

if [ -z $DEVMOUNT ]; then
# initramfs didn't mount /dev, so we'll need to do that mount -n -t tmpfs -o mode=0755 udev /dev
fi

# Copy over default device tree
if [ -d /lib/udev/devices ]; then
DEVICELIST=ls /lib/udev/devices if [ -n $DEVICELIST ]; then
cp -a -f /lib/udev/devices/* /dev
fi
fi

# It's all over netlink now
if [ -e /proc/sys/kernel/hotplug ]; then
echo "" > /proc/sys/kernel/hotplug
fi

# Start udevd
log_begin_msg "Starting kernel event manager..."
#if start-stop-daemon --start --quiet --exec /sbin/udevd -- --daemon; then /sbin/udevd --daemon &
UDEVPID=$! ; echo $UDEVPID > /var/run/udevd.pid
if [ $UDEVPID -gt 0 ]; then
log_end_msg 0
else
log_end_msg $?
fi

# This next bit can take a while
if type usplash_write >/dev/null 2>&1; then
usplash_write "TIMEOUT 360" ||:
trap "usplash_write 'TIMEOUT 15' ||:" 0
fi

# Log things that trigger does /sbin/udevadm monitor -e >/dev/.udev.log & UDEV_MONITOR_PID=$!

# Fix permissions and missing symlinks/programs for devices made in # initramfs, and catch up on everything we missed log_begin_msg "Loading hardware drivers..." /sbin/udevadm trigger
if /sbin/udevadm settle; then
log_end_msg 0
else
log_end_msg $?
fi

# Kill the udev monitor again
kill $UDEV_MONITOR_PID
;;
stop)
log_begin_msg "Stopping kernel event manager..."
#if start-stop-daemon --stop --quiet --oknodo --exec /sbin/udevd --retry 5; then if [ -e /var/run/udevd.pid ]; then
UDEVPID=cat /var/run/udevd.pid
else
UDEVPID=ps | grep udevd | cut -d " " -f 3
if [ -z "$UDEVPID" ]; then
echo "udev not running"
exit 0;
fi
fi
if kill $UDEVPID ; then
log_end_msg 0
else
log_end_msg $?
fi
umount -l /dev
;;
restart)
if [ ! -f /sys/kernel/uevent_seqnum ]; then
echo "Kernel uevent sequence number not available, cowardly not restarting udev" 1>&2 exit 0
fi

seqnum_before=$(cat /sys/kernel/uevent_seqnum)

log_begin_msg "Stopping kernel event manager..."
#if start-stop-daemon --stop --quiet --oknodo --exec /sbin/udevd --retry 5; then if [ -e /var/run/udevd.pid ]; then
UDEVPID=cat /var/run/udevd.pid
else
UDEVPID=ps | grep udevd | cut -d " " -f 3
if [ -z "$UDEVPID" ]; then
echo "udev not running"
exit 0;
fi
fi

if kill $UDEVPID ; then
log_end_msg 0
else
log_end_msg $?
fi

log_begin_msg "Starting kernel event manager..."
#if start-stop-daemon --start --quiet --exec /sbin/udevd -- --daemon; then /sbin/udevd --daemon &
UDEVPID=$! ; echo $UDEVPID > /var/run/udevd.pid
if [ $UDEVPID -gt 0 ]; then
log_end_msg 0
else
log_end_msg $?
fi

seqnum_after=$(cat /sys/kernel/uevent_seqnum)

if [ $seqnum_before -ne $seqnum_after ]; then
echo "Kernel uevent sequence number changed, some events may have been missed :o(" 1>&2
fi
;;
refresh-devices)
cp -au /lib/udev/devices/* /dev

log_begin_msg "Loading additional hardware drivers..." /sbin/udevadm trigger
if /sbin/udevadm settle; then
log_end_msg 0
else
log_end_msg $?
fi
;;
reload|force-reload)
log_begin_msg "Reloading kernel event manager..." #if start-stop-daemon --stop --signal 1 --exec /sbin/udevd; then if [ -e /var/run/udevd.pid ]; then
UDEVPID=cat /var/run/udevd.pid
else
UDEVPID=ps | grep udevd | cut -d " " -f 3 if [ -z "$UDEVPID" ]; then
echo "udev not running"
exit 0;
fi
fi
if kill -1 $UDEVPID ; then
log_end_msg 0
else
log_end_msg $?
fi
;;
*)

	echo "Usage: /etc/init.d/udev {start|stop|restart|refresh-devices|reload|force-reload}"
	exit 1
	;;

esac

exit 0

Well, I'm sure you can narrow it down. You can always single-step the init script on a running system to see what's happening.

?
if [ -z $DEVMOUNT ]; then
# initramfs didn't mount /dev, so we'll need to do that mount -n -t tmpfs -o mode55 udev /dev
fi

This will definitely cause some problems if you hadn't noticed it and left /lib/udev/devices empty. Udev relies on /dev/null being present at startup.

I think this is your main problem! Perhaps it is somehow responsible for the later 'unknown: "uevent"' problem.

Good luck

--
A: Because it messes up the order in which people normally read text.

It turns out it was pretty obvious:

Checking the source, it appears that having udev_monitor_enable_receiving fail means that udev_monitor_new_from_netlink() is already failing. I imagine that I'm missing a kernel config option, but I'm not sure what it is.

I've fixed the empty device list problem, so the failed bind is the biggest problem now.

Thanks again,
Brendan

Have you taken a look at the udev README? (That's twice lately that I've said that... hmm...) :-)