Skip to contentSkip to conversation

debian rootfs for ramdisk and NFS

DISCUSSION Kevin Hilman 4 participants 5 messages Jun 4, 2018 – Jun 6, 2018

lore
Share & tools
lore
#0 lore raw

Hi Ana,

In testing the debian rootfs images for NFS, Matt Hart and I both noticed that the full.rootfs.tar.xz images don't work for NFS because the systemd network init causes a (re)init of networking while actively using an NFSroot. This causes things to grind to a halt.

So the first question is whether anyone knows if systemd can be made pursuaded to not try to reconfigure networking.

Ideally, we'd like a single rootfs image to be either a ramdisk, where it might need to (re)configure networking and an NFSroot where it definitely should not reconfigure networking.

It would be a shame to have to have different images for ramdisk vs. NFS just for network init.

Ideas?

Kevin

#1 lore raw
Quoted 10 lines · Kevin Hilman

Hi Ana,

In testing the debian rootfs images for NFS, Matt Hart and I both noticed that the full.rootfs.tar.xz images don't work for NFS because the systemd network init causes a (re)init of networking while actively using an NFSroot. This causes things to grind to a halt.

So the first question is whether anyone knows if systemd can be made pursuaded to not try to reconfigure networking.

Quoted 8 lines

Ideally, we'd like a single rootfs image to be either a ramdisk, where it might need to (re)configure networking and an NFSroot where it definitely should not reconfigure networking.

It would be a shame to have to have different images for ramdisk vs. NFS just for network init.

Ideas?

Just as a data point, the debos scripts should be very close to what I was using when I tried NFS for running test suites. So I think it should be doable, but I don't know why it's happening.

Regards,

Tomeu

#2 lore raw
Quoted 19 lines · Tomeu Vizoso
Quoted 18 lines · Kevin Hilman

Hi Ana,

In testing the debian rootfs images for NFS, Matt Hart and I both noticed that the full.rootfs.tar.xz images don't work for NFS because the systemd network init causes a (re)init of networking while actively using an NFSroot. This causes things to grind to a halt.

So the first question is whether anyone knows if systemd can be made pursuaded to not try to reconfigure networking.

Ideally, we'd like a single rootfs image to be either a ramdisk, where it might need to (re)configure networking and an NFSroot where it definitely should not reconfigure networking.

It would be a shame to have to have different images for ramdisk vs. NFS just for network init.

Quoted 6 lines
Quoted 1 line

Ideas?

Just as a data point, the debos scripts should be very close to what I was using when I tried NFS for running test suites. So I think it should be doable, but I don't know why it's happening.

Looks like this overlay is forcing a DHCP on all interfaces that begin with 'e'
https://github.com/kernelci/kernelci-build/blob/master/jenkins/debian/debos/overlays/networkd/etc/systemd/network/wired.network

Quoted 4 lines

Regards,

Tomeu

#3 lore raw
Quoted 19 lines · Matt Hart
Quoted 18 lines · Tomeu Vizoso
Quoted 17 lines · Kevin Hilman

Hi Ana,

In testing the debian rootfs images for NFS, Matt Hart and I both noticed that the full.rootfs.tar.xz images don't work for NFS because the systemd network init causes a (re)init of networking while actively using an NFSroot. This causes things to grind to a halt.

So the first question is whether anyone knows if systemd can be made pursuaded to not try to reconfigure networking.

Ideally, we'd like a single rootfs image to be either a ramdisk, where it might need to (re)configure networking and an NFSroot where it definitely should not reconfigure networking.

It would be a shame to have to have different images for ramdisk vs. NFS just for network init.

Quoted 8 lines
Quoted 5 lines
Quoted 1 line

Ideas?

Just as a data point, the debos scripts should be very close to what I was using when I tried NFS for running test suites. So I think it should be doable, but I don't know why it's happening.

Looks like this overlay is forcing a DHCP on all interfaces that begin with 'e'
https://github.com/kernelci/kernelci-build/blob/master/jenkins/debian/debos/overlays/networkd/etc/systemd/network/wired.network

I just committed: https://github.com/kernelci/kernelci-build-staging/pull/34
I'm building the Debian rootfs builder job with this right now in staging, could you please test with the resulting images and tell if it works?

Ana

#4 lore raw
Quoted 25 lines · Ana Guerrero Lopez
Quoted 24 lines · Matt Hart
Quoted 23 lines · Tomeu Vizoso
Quoted 10 lines · Kevin Hilman

Hi Ana,

In testing the debian rootfs images for NFS, Matt Hart and I both noticed that the full.rootfs.tar.xz images don't work for NFS because the systemd network init causes a (re)init of networking while actively using an NFSroot. This causes things to grind to a halt.

So the first question is whether anyone knows if systemd can be made pursuaded to not try to reconfigure networking.

Quoted 8 lines

Ideally, we'd like a single rootfs image to be either a ramdisk, where it might need to (re)configure networking and an NFSroot where it definitely should not reconfigure networking.

It would be a shame to have to have different images for ramdisk vs. NFS just for network init.

Ideas?

Just as a data point, the debos scripts should be very close to what I was using when I tried NFS for running test suites. So I think it should be doable, but I don't know why it's happening.

Quoted 8 lines
Quoted 3 lines

Looks like this overlay is forcing a DHCP on all interfaces that begin with 'e'
https://github.com/kernelci/kernelci-build/blob/master/jenkins/debian/debos/overlays/networkd/etc/systemd/network/wired.network

I just committed: https://github.com/kernelci/kernelci-build-staging/pull/34
I'm building the Debian rootfs builder job with this right now in staging, could you
please test with the resulting images and tell if it works?

It works!

I tested the armel images (on both armel and arm64 boards) using initrd.gpio.gz ramdisk with a pivot to the full.rootfs via NFS, and it worked without a network reconfigure. Yay!!

Note that I didn't test this with LAVA, I just used my local pyboot setup, but I had to patch pyboot to deal with that automatic login.

@Matt: Is LAVA going to be confused by the automatic root login part? Seems like when it sees the "login:" prompot, it will try to login, even though a login isn't needed.

Kevin