[arch-dev-public] [signoff] coreutils-8.12-2, initscripts-2011.06.2-1, net-tools-1.60-15, udev-171-2, yp-tools-2.12-2

[Eric Bélanger]

On Sat, Jun 4, 2011 at 4:58 AM, Tom Gundersen <teg at jklm.no> wrote:
> On Sat, Jun 4, 2011 at 5:41 AM, Eric Bélanger <snowmaniscool at gmail.com> wrote:
>> FWIW, the instructive comments in udev.conf doesn't mention "warning"
>> as a valid value.
>
> Oops, I meant to say "info".

I also tried info when I tried with debug. I didn't noticed any
difference with debug as far as the excessive output go. I could try
it again and check dmesg more carefully.


>
>> I tried it and the error message doesn't get
>> printed.  Dmesg doesn't have the error message and the logs are
>> useless since the error happens when the kernel is booting, i.e.
>> rc.sysinit isn't running yet. I suppose it's mkinitcpio/udev related.
>
> If rc.sysinit is not running, then this is an mkinitcpio problem. I
> guess the reason you are seeing it again might be a race? We have sped
> things up quite a bit with this release, so things might race
> differently than before.

That's what I believe.  I didn't mentionned it but dmesg contains
output saying that the raid was assemble and all mirrors are active.
So despite the error message from udevd, the kernel still manage to
assemble the raid array. Also, I can't find the hook or config file
that contains the mdadm command line displayed in the error message.
I looked in /etc and /lib/initcpio but it's not there.

>
>> They work as expected. The error is not fatal.  My raid array gets
>> assembled without user intervention. Either it is eventually done by
>> the initscripts (I can't find where, I think that code parthas been
>> removed) or by something else.
>
> My guess is that the error you are seeing is from udev in mkinitcpio,
> and the udev in rc.sysinit works correctly. To change the error
> reporting you would need to run mkinitcpio -p kernel26 after updating
> udev.conf.

I was updating my initcpio file when changing the udev.conf.

>
>>>> Also, my loopback interface  didn't started.  When I ran "/usr/sbin/ip
>>>> link set up dev lo" manually in a terminal, it started fine. I don't
>>>> know why it didn't work when booting up.
>>>
>>> Do you see anything in the above logs?
>>
>> I found the source of this error. My /usr is on it's own partition. So
>> when rc.sysinit attemps to setup the loopback interface with
>> /usr/sbin/ip, my /usr partition is not mounted yet. So it fails.  I
>> don't know what would be the best way to fix. Maybe moving ip to /sbin
>> or to postpone the lo interface  setup after the partitions are
>> mounted.
>
> Ahhhh. I think this might cause many more problems than just the
> loopback. In particular, anything called in udev rules may reside on
> /usr. This means the rule will fail and it will not be rerun.
>
> Have a look at "grep /usr /lib/udev/rules.d/*" to see examples of what
> might be broken.
>
> Most other distros, and upstream projects have given up on supporting
> a separate /usr, so it will be buggy. I suggest moving your /usr to
> your rootfs, regardless of this particular problem.
>

I recall you mentionning that on another ML thread. I wanted to ask
you for details at the time but did get to.
I'll move my /usr to my rootfs.  I use lvm2 so the
resizing/repartionning should be easy.

> As to the loopback device. I don't think it is a good idea to delay
> this to after /usr is mounted. If anything we might want to move it
> earlier in boot. That leaves moving ip out of /usr.
>
> Does anyone know the consequences of this? Do we want to keep
> supporting a separate /usr?

If I'm the only one with the problem, you can also keep things as they
are now, i.e. not support /usr on a separate partition. As I'll be
moving my /usr, that problem will become moot.

Eric

>
> Cheers,
>
> Tom
>