Call for testing: new udev -- switch to vendor/name/slot based net interface names?

[Stéphane Graber]

On 05/13/2013 04:23 AM, Martin Pitt wrote:
> Hello fellow Ubuntuers,
> 
> now that we have logind/uaccess and libudev1 in Ubuntu, it is finally
> time to update to a current udev version. This will bring a faster
> system boot as it does not call the modprobe, scsi_id, blkid and other
> programs a gazillion times, but instead has all these built in
> (through libkmod, libblkid, etc.). It will also get us rid of quite a
> few patches which were piling up to make our increasingly ancient udev
> work with the current kernel and plumbing.
> 
> I went through our current package and applied our remaining patches,
> initramfs integration and Ubuntu specific rules. There were two
> patches which were non-obvious:
> 
>  * avoid-exit-deadlock-for-timely-events: This was applied for
>    http://pad.lv/842560 . Andy discussed that with upstream in
>    http://comments.gmane.org/gmane.linux.hotplug.devel/17206 and as a
>    result the whole TIMEOUT legacy stuff has been removed, so this
>    does not apply any more. Also, firmware handling has changed quite
>    a bit (e. g. recent kernels load it by themselves). So I dropped
>    the patch.
> 
>  * avoid-exit-deadlock-for-dm_cookie.patch: This was applied as a
>    kind of workaround for http://pad.lv/802626 (I think we really
>    should stop having two udevs in the boot process, BTW), and further
>    modified the code that avoid-exit-deadlock-for-timely-events
>    already changed. I think I ported it correctly, but if you have an
>    LVM setup, double-checking this can't hurt.
> 
> There is one important change which warrants some discussion: By
> default, current udev now uses BIOS/vendor/slot number based names so
> that they are easier to identify, but more importantly, they have
> predictable names without the old 70-persistent-net.rules. My current
> package disables this for upgrades (to avoid changing any existing
> network configuration), but enables it for new installs. Stephane
> Graber mentioned that we might not want this as it might break too
> many tools which rely on the ethX naming; I can forward port the old
> generator for 70-persistent-net.rules (it's just a shell script) if we
> want.

I'd really rather we don't take the network interface naming change.
We've seen numerous issues when doing a similar change on the servers
with biosdevname where we ended up with a few very weird cases where the
firmware/bios is reporting bogus information leading to entirely random
interface names that randomly conflict at boot time (leading to renameX
interface names and broken networking).

So I'd much prefer we stick to something that's been working rather well
for all these years and let other people do the experiments with that stuff.

I think we should only consider switching to this on the client devices
once we are satisfied that this works perfectly well on the servers
(where we have a smaller set of devices with this turned on and can
better see the exact impact of that change).

> So before I land it in the archive I'd like to get a few more testing
> results on various machines. If you could install the packages from
> https://launchpad.net/~pitti/+archive/ppa (systemd is the only one for
> saucy in that PPA) and tell me how it goes, together with your
> machine/install config (UEFI, LVM, cryptsetup, etc.), I'd appreciate.
> 
> Thank you in advance!
> 
> Martin
> 
> 
> 


-- 
Stéphane Graber
Ubuntu developer
http://www.ubuntu.com

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 901 bytes
Desc: OpenPGP digital signature
URL: <https://lists.ubuntu.com/archives/ubuntu-devel/attachments/20130514/35b408d2/attachment.pgp>