[arch-dev-public] network interface naming with systemd 197

[Dave Reisner]

On Sun, Jan 06, 2013 at 09:32:41PM +0100, Tom Gundersen wrote:
> On Jan 6, 2013 7:38 PM, "Dave Reisner" <d at falconindy.com> wrote:
> >
> > Just an FYI:
> >
> > Upstream pushed a commit[0] which gives network devices persistent, and
> > unique, names based on hardware attributes, avoiding the random kernel
> > names. While this solves a real problem, it's also a fairly jarring
> > change. For example:
> >
> > $ udevadm info /sys/class/net/eth0
> > P: /devices/pci0000:00/0000:00:1c.2/0000:05:00.0/net/eth0
> > E: DEVPATH=/devices/pci0000:00/0000:00:1c.2/0000:05:00.0/net/eth0
> > E: ID_BUS=pci
> > E: ID_MODEL_ID=0x4364
> > E: ID_NET_NAME_MAC=enxbcaec50bfcc8
> > E: ID_NET_NAME_PATH=enp5s0
> > E: ID_OUI_FROM_DATABASE=ASUSTek COMPUTER INC.
> > E: ID_PCI_CLASS_FROM_DATABASE=Network controller
> > E: ID_PCI_SUBCLASS_FROM_DATABASE=Ethernet controller
> > E: ID_PRODUCT_FROM_DATABASE=88E8056 PCI-E Gigabit Ethernet Controller
> > E: ID_VENDOR_FROM_DATABASE=Marvell Technology Group Ltd.
> > E: ID_VENDOR_ID=0x11ab
> > E: IFINDEX=2
> > E: INTERFACE=eth0
> > E: SUBSYSTEM=net
> > E: SYSTEMD_ALIAS=/sys/subsystem/net/devices/eth0
> > E: TAGS=:systemd:
> > E: USEC_INITIALIZED=42063
> >
> > If I were to reboot right now (systemd-git), eth0 would become enp5s0. I
> > tend to think that this is fairly extreme, and would throw off a lot of
> > people -- especially those who never needed to deal with interface
> > renaming.
> >
> > For systemd 197, I plan on shipping this rule as documentation in
> > /usr/share/doc/systemd and _not_ enabling it by default. Those who want
> > to opt in can simply copy the rule to /etc/udev/rules.d. They can also,
> > of course, continue to use whatever MAC-based rules they might have, but
> > I would strongly recommend switching these rules to be triggered by
> > ID_NET_NAME_{SLOT,PATH,ONBOARD} instead.
> >
> > Cheers,
> > Dave
> >
> > [0] http://cgit.freedesktop.org/systemd/systemd/commit/?id=394e2938ff9
> 
> How about:
> 
> 1) follow upstream on fresh installs (i.e. ship the rule and don't mask it
> in post_instal).

Scary. I agree with upstream that this is wanted and that it solves real
problems, but I really see no reason that this should be opt-out, rather
than opt-in. We have the option of explaining why the dummy file exists
in /etc when we make things opt-in, but opt-out on install makes the
messaging easier to miss. Additionally, there'll be an awkward phase
where older install media uses older systemd (providing the "classic"
names), followed by a reboot into newer systemd with the new naming
scheme.

Anyone else have an opinion on this?

> 2) stay backwards compatible on upgrade (i.e. mask the rule in
> post_upgrade).
> 
> 3) print a notice about the masking so people can unmask it.

Definitely planned.

> 4) rather than a symlink to null, use an empty rules file with a comment
> explaining why it is there and what will happen if you delete it.

I like this. Done for the -git package, at least.

d