/etc/network/interfaces comes unconfigured, should use dhcp by default

[Steve Langasek]

In Linaro images, /etc/network/interfaces only has configuration for the lo interface and (as described in the image 'network' test case'), you have to run dhclient by hand to bring up your wired/wireless network. We should somehow have these pre-configured on images, but since the interface names will vary between boards, we'll probably have to include that information in the hwpack and have l-m-c use that to configure them. Also, we don't want to do this if NetworkManager is available in the image as in that case it will take care of everything

[Loïc Minier (lool)]

BeagleXM has "usb0", IGEPv2 has an additional wifi interface /on some boards/; not trivial

[Steve Langasek (vorlon)]

understood - but although it's not trivial, I think it's important to solve and should be kept on the radar.

Why does BeagleXM use "usb0" instead of the standard "eth0"? That seems like a pretty gratuitous difference. Is there any chance of getting that corrected at the kernel level?

[Scott Bambrough (scottb)]

On Fri, 2010-09-03 at 19:03 +0000, Steve Langasek wrote:
> understood - but although it's not trivial, I think it's important to
> solve and should be kept on the radar.
>
> Why does BeagleXM use "usb0" instead of the standard "eth0"? That seems
> like a pretty gratuitous difference. Is there any chance of getting
> that corrected at the kernel level?
>
I have seen this before. It is usually caused because the NIC is
connected via the USB subsystem, and nothing has been done in the driver
to tweak the naming of devices. The driver can be modified to create
ethX devices.

--
Scott Bambrough <email address hidden>
Linaro Infrastructure Team Lead

[Loïc Minier (lool)]

see bug #622429; it's due to the lack of an eeprom with a MAC address; the MAC is autogenerated

[Loïc Minier (lool)]

(Note that I agree that it should be fixed, wanted to point out the various cases :-)

[Loïc Minier (lool)]

Low? Means no networking with headless out of the box, doesn't it?

[Paul Larson (pwlars)]

Yes, though the workaround is as simple as running 'dhclient eth0'. That being said, I think this should be higher as well, even if it means we don't handle oddities like a board that exposes the interface as USB0. IMHO, that should be a separate bug.

[Loïc Minier (lool)]

I think we can workaround in linaro-image-create; any takers?

[Steve Langasek (vorlon)]

I'm moderately concerned about putting this in linaro-image-create because we already have code in the archive to do exactly this setup of /etc/network/interfaces (the netcfg udeb from the Debian installer). How many more bits of install-time configuration will we be duplicating between d-i and linaro-image-tools? I would like us to seriously consider how we could reuse the modular components from d-i here instead of maintaining duplicate code for all aspects.

But that's likely a topic for 11.05, not for 10.11; for 10.11 we can/should add this to linaro-image-create.

[Loïc Minier (lool)]

Well this is precisely what I said when we started, but it was fair to say back then that a d-i / ubiquity style approach was much heavier.

I really wish we consider it next cycle, but I'm not sure which role live-helper will play

[Loïc Minier (lool)]

It might be possible to accommodate for this with the new hardware pack format, which might include list of interfaces for which to generate network config. This would allow generating etc/network/interfaces if network-manager isn't installed.

[Steve Langasek (vorlon)]

This is now being set by default for nano and developer at tarball generation time. The desktop images include NetworkManager which takes care of automatically bringing up the network. Do you think it's still necessary to change anything in l-m-c for this, or should we close this now as invalid?

[Loïc Minier (lool)]

We plan a field in https://wiki.linaro.org/Platform/Specs/11.05/HardwarePacksV2 to list interfaces; do you think we should drop the field from the spec, or that we should keep this bug to implement this support when the information is available?

[Steve Langasek (vorlon)]

On Fri, Apr 15, 2011 at 09:46:21AM -0000, Loïc Minier wrote:
> We plan a field in
> https://wiki.linaro.org/Platform/Specs/11.05/HardwarePacksV2 to list
> interfaces; do you think we should drop the field from the spec, or that
> we should keep this bug to implement this support when the information
> is available?

I think it's fine to do it via the hardware pack, just wanted to make sure
you knew the immediate problem is now addressed.

--
Steve Langasek Give me a lever long enough and a Free OS
Debian Developer to set it on, and I can move the world

[Loïc Minier (lool)]

thanks for sharing the update! It's good to know it's not too urgent to fix this in l-i-t

keeping this open and milestoned aganst 0.5.0 for hardware-v2

[Ricardo Salveti (rsalveti)]

This is an issue again with Oneiric, as if we set an invalid interface the boot will take a lot longer due bug 856810.

[Mattias Backman (mabac)]

On Sun, Oct 16, 2011 at 10:23 PM, Ricardo Salveti <email address hidden> wrote:
> This is an issue again with Oneiric, as if we set an invalid interface
> the boot will take a lot longer due bug 856810.

In the hwpacks V2 format we introduced wired_interfaces and
wireless_interfaces, but I never connected the wires since we came to
the conclusion that this probably was not needed anymore. If it is
useful for the hwpack to specify the board interfaces, I can finalize
this. I'd just need some help with what to do with the interface
names.