Ubuntu 20.04.4 LTS Released With Hardware Enablement Stack From Ubuntu 21.10

[Developer12]

What happened was that Kernel 5.13 made lseek mandatory for /proc. Meanwhile acpi_call is a 3d party external kernel module which have been unmaintained for 9 years so it was never patched to support lseek so the Kernel tries to make a function call to NULL. So no this was not due to some uefi shenanigans. And if the kernel would have to check every single function call every time it makes them you would have to pay a quite hefty performance penalty so this is not done, and frankly the Kernel should be able to rely on the modules providing correct structures to it (and no this is not a security hole, if you can load a module into the kernel then you are already in control of the whole system).

The problem comes when the kernel makes a dump to NVRAM when the null dereference occurs. Normally making a dump is good, but eventually the BIOS NVRAM fills up. It might be good if the kernel only kept the last N dumps, but instead it fills all the space and the badly-programmed bios borks without a means of recovery.

[NobodyXu]

What happened was that Kernel 5.13 made lseek mandatory for /proc. Meanwhile acpi_call is a 3d party external kernel module which have been unmaintained for 9 years so it was never patched to support lseek so the Kernel tries to make a function call to NULL. So no this was not due to some uefi shenanigans. And if the kernel would have to check every single function call every time it makes them you would have to pay a quite hefty performance penalty so this is not done, and frankly the Kernel should be able to rely on the modules providing correct structures to it (and no this is not a security hole, if you can load a module into the kernel then you are already in control of the whole system).

I think the kernel could check for NULL in the table when loading the module, either fill it with some dummy default value or reject the module based on policy

[F.Ultra]

I find the acpi/tlp bug quite dramatic. Can someone explain why the kernel has no safeguard for this? Is this again a lousy uefi implementation from the vendor?

What happened was that Kernel 5.13 made lseek mandatory for /proc. Meanwhile acpi_call is a 3d party external kernel module which have been unmaintained for 9 years so it was never patched to support lseek so the Kernel tries to make a function call to NULL. So no this was not due to some uefi shenanigans. And if the kernel would have to check every single function call every time it makes them you would have to pay a quite hefty performance penalty so this is not done, and frankly the Kernel should be able to rely on the modules providing correct structures to it (and no this is not a security hole, if you can load a module into the kernel then you are already in control of the whole system).

[kylew77]

Yesterday I replaced Ubuntu 21.10 by Fedora 35 (I replaced those OS'es on secondary platform too 2 weeks ago) and I'm really happy with the results. Ubuntu release policy for kernel and Mesa is horrible (glad that we can use ppa) + slow crap named Snap... 2 nice things from Ubuntu? Improvments to Gnome made by Daniel Van Vugt and Yaru theme. I used Ubuntu since 6.06, but in last times they made a lot of stupid decisions and for my it was too much.

I've never used vanilla Ubuntu always Xubuntu or Kubuntu, vanilla Ubuntu's insistence of sticking with Gnome 3 was disappointing for people making the switch from Windows. Set my parents up with Xubuntu and labeled the launcher start and put some shortcuts on the desktop and they love it more than Windows XP, been riding the Xubuntu LTS train for them since 14.04. I'm sure this summer or next winter whenever the next time I'm home I will upgrade them to 22.04. Helps that with Wine they can play the same binary spider and hearts game they loved from Windows XP plus lots of new games from the KDE game collection they have gotten hooked on.

[nadro]

Yesterday I replaced Ubuntu 21.10 by Fedora 35 (I replaced those OS'es on secondary platform too 2 weeks ago) and I'm really happy with the results. Ubuntu release policy for kernel and Mesa is horrible (glad that we can use ppa) + slow crap named Snap... 2 nice things from Ubuntu? Improvments to Gnome made by Daniel Van Vugt and Yaru theme. I used Ubuntu since 6.06, but in last times they made a lot of stupid decisions and for my it was too much.

[Smurphy]

Well, some important packages for servers are now snaps (20.04 LTS is for example the first LTS where lxd is a snap). We have been forced to re-recreate an internal version of of the lxd package -- granted, it's easier for us as we don't have to create the source packages, meaning that we don't have to package all the dependencies (which would have forced us to create an insane number of deb-src).

Just using the server plain stuff. Adding docker & K8S from docker/k8s - and that's about the software I use from Ubuntu on a server now.
Desktop - well, depends. I tend to prefer using deb packages. I may have to go back to debian or so to avoid that crap.

[Emmanuel Deloget]

But you can remove it. This is usually the first operation I do on a newly setup Ubuntu system nowadays. Purge snap.

Well, some important packages for servers are now snaps (20.04 LTS is for example the first LTS where lxd is a snap). We have been forced to re-recreate an internal version of of the lxd package -- granted, it's easier for us as we don't have to create the source packages, meaning that we don't have to package all the dependencies (which would have forced us to create an insane number of deb-src).

[kylew77]

This is nice an all, but could they please fix the null pointer dereferences from acpi-call?

They updated Jammy Jellyfish (22.04) to acpi-call 1.2.2-1 but left it untouched for every other version of ubuntu. Lots of versions, all the way back to at least 20.04, have also moved from kernel 5.11 to 5.13 via HWE releases and are now experiencing kernel null pointer dereferences.

What this means is that for anyone running 20.04+ (but not still-experimental 22.04) on a thinkpad, if you use tlp to manage your battery you're going to get constant kernel breakage. For some people the constant kernel dumps filled up their BIOS's EFI NVRAM and bricked the laptop entirely. https://bugs.launchpad.net/ubuntu/+s...l/+bug/1953261

Bricking the entire laptop shouldn't be possible in any OS. This seems like a HUGE oversight. Especially since ThinkPads ship with Ubuntu options. I would be pissed if I had a bricked 2k+ laptop!

[Smurphy]

I'm not sure what you mean, but Snap *is* most definitely present on 20.04.

But you can remove it. This is usually the first operation I do on a newly setup Ubuntu system nowadays. Purge snap.

[drake23]

I find the acpi/tlp bug quite dramatic. Can someone explain why the kernel has no safeguard for this? Is this again a lousy uefi implementation from the vendor?