This is a clarification of bug #5770.

A number of recently-updated machines failed to boot properly, due to
rc.sysinit/start_udev not starting udevd. The problem is that the scripts
detect that udevd is already running, so do not start it again. On an affected
machine, doing 'ps -C udevd' shows udevd as <defunct>.

The problem only occurs on machines that have the root filesystem on raid0. I
converted one machine to have its root filesystem on a plain partition, and
this worked fine.

I have worked around the problem by modifying the udev_init() function in
start_udev. After the existing check using 'pidof -o %PPID /sbin/udevd', I put
an 'if' statement to check if udevd is defunct. If it is, I start udevd.

# check if udevd is already running
if [ `pidof -o %PPID /sbin/udevd` ]; then
if /bin/ps -C udevd | /bin/grep defunct; then
echo "start udev daemon 1"
/sbin/udevd --daemon
else
echo "udevd already running: `pidof -o %PPID /sbin/udevd`"
grep -v "/dev/shm" /etc/mtab | grep -v "/dev/pts" > /etc/mtab
mount /dev/pts
mount /dev/shm
fi
else
echo "start udev daemon 2"
/sbin/udevd --daemon
fi

---------------
Even with this fix, there is still the problem with kernel26-fallback.img not
working with raid. In this case, booting does not get as far as running any
startup scripts. Again, if the root filesystem is not on raid0, the fallback
image works fine.