diff mbox series

[bug#45722] gnu: kernel-module-loader does not require file-systems

Message ID BRJQ1iDMuVglCqRb1sGZLwYQq_89pM_EAU90i6_Q7mtbu-F8n0JV1bn4g90twbRWnMzS-tCi0C2lBlT53OR6FnG-4hW6j_2Ix3TInBbyNdw=@protonmail.com
State New
Headers show
Series [bug#45722] gnu: kernel-module-loader does not require file-systems | expand


Context Check Description
cbaines/submitting builds success
cbaines/comparison success View comparision
cbaines/git branch success View Git branch
cbaines/applying patch fail View Laminar job
cbaines/issue success View issue

Commit Message

raid5atemyhomework Jan. 8, 2021, 2:12 a.m. UTC
In https://issues.guix.gnu.org/45692#5 Danny brings up that the `kernel-module-loader-shepherd-service` might not need to depend on `file-systems`.


* `kernel-module-loader-shepherd-service` uses the `/proc` filesystem.
  * However, this filesystem is mounted in `initrd` before handing over to the real root (in `gnu/build/linux-boot.scm`'s  `mount-essential-file-systems`, which is called in `boot-system` very early.
  * Before handing over to the real root, the mounts are moved in `move-essential-file-systems`, which is called by the `switch-root` procedure that is called by `boot-system` after mounting the root filesystem.
* The `/proc/sys/kernel/modprobe` file contains a reference to a `/gnu/store` path to a `modprobe` binary.
  * Shepherd starts with the root file system already mounted (by the `initrd` `boot-system` procedure), so the full `/gnu/store` should be accessible as soon as Shepherd starts.

I've tested this as follows:

* Created a QEMU VM image that pointlessly loads "btrfs" (which is not loaded by default) by `(simple-service 'load-btrfs kernel-module-loader-service-type '("btrfs"))` and confirmed it boots to desktop and `lsmod` lists `btrfs`.
* Ran tests "loadable-kernel-modules-0", "loadable-kernel-modules-1", and "loadable-kernel-module-2" and confirmed they pass.

In theory this could break a system, if an existing kernel module reads from some file (which Linux kernel modules are discouraged from doing) at loadtime *and* the system administrator put that file in a non-root filesystem (which if it's a "proper" module it should be in either `/etc` or `/var`, the former of which is, on Guix System, created by Guix on the root filesystem, and the latter of which should generally be on the root filesystem as well).  So I think the risk of this breaking *some* system is fairly low.

Please review.

From d028dac0efaf6ab3d8edafc2e68c0d2ef692bab7 Mon Sep 17 00:00:00 2001
From: raid5atemyhomework <raid5atemyhomework@protonmail.com>
Date: Fri, 8 Jan 2021 09:41:25 +0800
Subject: [PATCH] gnu: kernel-module-loader does not require file-systems

* gnu/services/linux.scm (kernel-module-loader-shepherd-service):
Remove requirement of file-systems.
 gnu/services/linux.scm | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff mbox series


diff --git a/gnu/services/linux.scm b/gnu/services/linux.scm
index 1046a7e0c2..340b330030 100644
--- a/gnu/services/linux.scm
+++ b/gnu/services/linux.scm
@@ -2,6 +2,7 @@ 
 ;;; Copyright © 2020 Maxim Cournoyer <maxim.cournoyer@gmail.com>
 ;;; Copyright © 2020 Brice Waegeneire <brice@waegenei.re>
 ;;; Copyright © 2020 Efraim Flashner <efraim@flashner.co.il>
+;;; Copyright © 2021 raid5atemyhomework <raid5atemyhomework@protonmail.com>
 ;;; This file is part of GNU Guix.
@@ -150,7 +151,7 @@  representation."
       (documentation "Load kernel modules.")
       (provision '(kernel-module-loader))
-      (requirement '(file-systems))
+      (requirement '())
       (one-shot? #t)
       (modules `((srfi srfi-1)
                  (srfi srfi-34)