So basically, my setup has everything encrypted except /boot/efi. This means that /boot/grub is encrypted, along with my kernels.
I am now attempting to get secure boot setup, to lock some stuff, down, but I encountered this issue: https://bbs.archlinux.org/viewtopic.php?id=282076
Now I could sign the font files… but I don’t want to. Font files and grub config are located under /boot/grub, and therefore encrypted. An attacker doing something like removing my hard drive would not be able to modify them.
I don’t want to go through the effort of encrypting font files, does anyone know if there is a version of grub that doesn’t do this?
Actually, preferably, I would like a version of grub that doesn’t verify ANYTHING. Since everything but grub’s efi file is encrypted, it would be so much simpler to only do secure boot for that.
And yes, I do understand there are security benefits to being able to prevent an attacker that has gained some level of running access to do something like replacing your kernel. But I’m less concerned about that vector of attack, I would simply like to make it so that my laptops aren’t affected by evil maid attacks, without losing benefits from timeshift or whatnot.
I found the specific commit where grub enforces verification of font files: https://github.com/rhboot/grub2/commit/539662956ad787fffa662720a67c98c217d78128
But I don’t really feel interested in creating and maintaining my own fork of grub, and I am wondering if someone has already done that.
My important data is backed up to several usb drives, and kept in sync between two computers via syncthing. Soon I will back it up to my college’s box cloud, using rclone’s crypt feature.
But this isn’t about data. This is about me being able to tinker without worry, breaking down to even the lowest level of my system. In addition to that, I don’t want to have to waste time manually restoring a system snapshot/backup, as I will soon be busy with other things like classes. I want a one step process.
Why not tinker in a VM if that’s the case? An even better solution for your need of an easy, just works distro that is impossible to FUBAR would be something like Fedora Silverblue or MicroOS.
I think this post is you learning the hard way, there’s no such thing as a one step process in Linux.
Can’t test battery life in a VM. Also uses up too much resources.
It’s a bit of a pain to get my favorite optimizatons and customizations like uksmd or zram set up on immutable distros.
I got it working, using this guide: https://wejn.org/2021/09/fixing-grub-verification-requested-nobody-cares/.
Certainly an unorthodox grub hack, but it’s replicatable and it works. Maybe in the future systemd boot or the refind will gain the features I want. Until then, only grub offers what I want.
If a grub update ever breaks this, or maybe just to futureproof this, then I’ll probably just use Arch’s PKGBUILD, makepkg, and patch tools to patch the grub_efi_get_secureboot function of sb.c so that grub always thinks it’s not in secure boot. And maybe put that version on the AUR.
No, this post is me proving everybody who told me to switch away from grub, (despite my insistence that only grub has the features I need) wrong.
KVMs barely use any resources.
I think Fedora enable zram by default, not so sure about Silverblue.
Sounds like you are intentionally breaking secure boot code in grub. If you can’t see why this is a problem, maybe you shouldn’t be using secure boot.
Nope. The post is proving how ill-equipped grub is to support modern features such as secure boot. Grub is very much legacyware at this point. New security features are being developed over at systemd land.
I understand why this is a problem, or would be on systems where much of the initial stages of the system are left unencrypted. But because literally everything but grubx64.efi is encrypted, there is no need for them to be verified. Only grub, which asks for my password for the decrypting, needs to be verified. This behavior is intended for systems that require more security, for example, to prevent unauthorized loading of drivers by a malicious attacker. But I don’t need or want that.
I have no doubt that system-boot will get the features I want, and that grub will probably never get things like tpm auto unlock. But I don’t use software based on what features they will have, I select software based on what features it currently has. And right now, grub has features I need, that systemd-boot doesn’t have. That’s just the reality of the situation.
edit: went through your profile history and you literally made a post of where my setup would be useful, the one about the amd regressions. Oh no, if only you had a setup where you could instantly reboot into an older kernel, with one click. But you don’t, so you just have to take the performance hit, or go through the hassle of restoring an entire backup either from the local disk, or worse, from another machine/disk. ;(
Sounds like cope to me. You don’t get to tell an attacker which component they can attack when you have misconfigured your security guards.
I keep the LTS kernel around for that. It’s a “one-click solution” for boot problems. And anyway, a simple chroot should allow me to fix any problems.
Don’t see what advantage snapper rollbacks have for me when I would need to use it rarely. I don’t have a habit of intentionally breaking things. And when things break on me (I use Arch btw), they rarely introduce fatal errors.
There is only a single thing on my system unencrypted: the grubx64.efi binary. This binary is verified via secure boot. Unless an attacker can break luks2 encryption, they cannot get to anything else.
Did you read your own post? The lts kernel was affected too. That’s why I used it as an example.
You could also just nab the older kernel from the archive or something, if your system still boots. But I don’t want to have to do that. I have better things to spend my time on then going through the pain of disabling all my security features so I can chroot into an encrypted system.
I don’t know enough about the subject of a secure grub to tell you how wrong you are.
Yes I did. It was a terrible example. As all I would need to know was the last working version for TPM. Regression in LTS does not factor in this equation.
And most importantly, it would not stop me from booting.
You think you are saying something smart here but I assure you, you couldn’t be more conceited. You are maintaining a patch of grub for a bug that grub has no idea it exists. And you claim not to have time to fix your installation…
If you don’t know, then why don’t you shut up, yeah? I’ve spent 3 weeks researching this, even going as far to read the source of grub. Don’t just assume you’re right without doing any research.
I have the time now. Classes are just getting started. But I’ll be busier in the future. Due to the way that arch is setup, this is easier than signing everything, plus I get instant restores.
And it’s not a bug. It’s intended behavior for systems like high value servers where security is valued over all else, to prevent privilege escalation by an attacker exploiting a kernel bug to load more kernel modules or taking advantage of a similar exploit. But for my desktop system, such an attack is not in my threat model.