-
Notifications
You must be signed in to change notification settings - Fork 197
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
core: also wrap kernel-install
for scriptlets
#4950
Conversation
Of course, this still requires using rpm-ostree for kernel installs instead of dnf. We really need #4726 to be able to move to dnf. The main thing I'm trying to avoid with this is not leaking cliwrap bits into the image, but dropping the requirement on |
Prep for next patch.
It's confusing right now how specifically for the kernel, one has to use this obscure `rpm-ostree cliwrap install-to-root /` command to make it work. Let's just always enable it: in the client-side layering case, we don't run kernel scriptlets anyway so the wrapper is unused, and in the container case, this will allow users to not have to enable cliwrap and have it leak into their derived image. I guess in theory, this should also allow us to *stop* ignoring kernel scriptlets and rely on this instead, though let's leave that for a separate investigation. Closes: coreos#4949
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
A few things we need to bear in mind here.
One is that we've also gotten burned with cliwrap + rpm-ostree/dnf upgrade
because the rpm stack will happily replace the wrapped binary with really undesirable semantics. In this case that would again happen with updating systemd in a container build.
So I'm wavering a bit on whether it's better to do this now or just try to fix it by properly integrating into kernel-install.
There is another alternative path, where instead of using cliwrap in this way during scripts we just override it in base container builds.
Actually yeah, the core problem here remains in that people are still going to use dnf
and if we don't cliwrap that too it remains broken.
So...not opposed but I'm not sure it really makes sense to invest in this.
Hmm, we're not shipping this wrapper in the container image. Only installing it during the transaction. It's the same as all the other systemctl and shadow-utils wrappers we currently temporary wrap and then unwrap.
Yeah, that's definitely the end goal. I'm going for a tactical fix here because basically today changing the kernel as part of OCP layering requires cliwrap, and cliwrap breaks the MCD. So that means that the documented procedure for replacing the kernel via layering is broken in 4.16. We're all agreed we should have This definitely affects 4.16, so we'd need it backported to 9.4. I haven't checked yet if 4.15 is affected, but likely not (the |
Hmm, isn't that CentOS/centos-bootc#377 ? Or are you suggesting lifting it out of rpm-ostree and carrying it somewhere? |
OK yes, you're right.
I'm actually thinking more that we stop trying to use rpm-ostree's builtin kernel handling, and open-code it as hooks in This is what would allow us to support dnf. |
Having it live in ostree makes sense to me. Though ideally it can be used by any ostree user that also uses dracut + systemd's That said, ideally long-term the same code that prepares the kernel and initramfs server-side is what regenerates it in the container derivation and client-side layering cases. E.g. we'd drain e.g. the dracut and depmod stuff from rpm-ostree and actually adapt the canonical scripts as needed to do the right thing. |
It's confusing right now how specifically for the kernel, one has to use this obscure
rpm-ostree cliwrap install-to-root /
command to make it work. Let's just always enable it: in the client-side layering case, we don't run kernel scriptlets anyway so the wrapper is unused, and in the container case, this will allow users to not have to enable cliwrap and have it leak into their derived image.I guess in theory, this should also allow us to stop ignoring kernel scriptlets and rely on this instead, though let's leave that for a separate investigation.
Closes: #4949