Re: Initramfs Scripts

Hi,
Comments in the text, and attached propositions for both files
On Thu, 2025-06-05 at 17:34 -0400, Rahul Chandra wrote:
> On Thu, Jun 5, 2025 at 4:03 PM Joe Locash
> <[email protected]> wrote:
> > 
> > On 6/5/25 1:39 PM, Rahul Chandra ([email protected] via blfs-
> > dev
> > Mailing List) wrote:
> > > Currently the initramfs scripts in BLFS are entirely broken,
> > > meaning
> > > even the limited subset of features we say to use an initramfs
> > > for
> > > (drive encryption, network boot, and LVM) don't function. Here is
> > > a
> > > list of the issues
> > > 
> > > 1. /etc/group missing -- The default 50-default-udev rule
> > > requires the
> > > base groups from LFS in order to load properly

Since we use --resolve-names=never, group and user names are never
resolved, so /etc/group is not needed.

> > > 2. Missing libkmod.so files, udev (at least on systemd) can't
> > > load
> > > modules without a library link to kmod

systemd udev use dlopen to load libkmod.so. This means this library is
not shown when running ldd. It has to be included manually.

> > > 3. Missing cgroup mount -- Control groups are almost required
> > > nowadays
> > > to load modules properly, we need to add a control group mount in
> > > /sys/fs/cgroup to fix this

I've been able to run the initrd without mounting cgroup, but I guess
it needs further testing. I've removed the mount command.

> > > 4. udevadm trigger command. I don't know when this changed but
> > > udevadm
> > > trigger doesn't load modules anymore, you need to add the --
> > > action=add
> > > flag for it to actually load the modules
> > > 

Nope --action=add does not mean to load the modules, it means to
trigger "add" events instead of "change" events. I've kept it since it
seems to work ok.


> > > Other fixes I did a long time ago
> > > 
> > > 5. This is an issue I had fixed a few years ago in my homebrewed
> > > version of the initramfs scripts -- The initramfs scripts
> > > currently
> > > pick a subset of modules on the system. In this version, it
> > > dynamically finds the modules currently loaded and adds them to
> > > the
> > > initramfs

Actually no. It looks for modules in more subdirectories than
previously. Note that finding currently loaded modules imposes an
already running kernel, but if you are making a initrd for a new
kernel, you cannot use this method...
I've added a few more subdirectories, that seemed needed for loading
all modules on which needed modules depend...

> > > 6. Firmware -- Loading firmware in the initramfs (required for
> > > some
> > > RAID cards and network cards) was not supported with the old
> > > scripts.

That's not true, Previous version blindly copied all firmware in
/usr/lib/firmware...

> > > These one's copy the firmware based on the modules that are
> > > currently
> > > loaded.

Same problem as above, this prevent making an initrd for a new
kernel... I've moved the commands to copy firmware so that they copy
only firmware for modules included in the initrd.
> > > 
> > > Other nice to haves  in this version
> > > 7. Forceful early module load -- I added a /etc/early-modules-
> > > load.conf file
> > > cat > /etc/early-modules-load.conf << EOF
> > > # This file is for loading modules in the early boot stages in
> > > the initramfs
> > > EARLY_LOAD_MODULES=""
> > > EOF
> > > To allow for loading modules that aren't required for the system
> > > (e.g
> > > vfio or kvmgt for gvt-g/iommu)

I've simplified the sed command a bit, and included the commands in an
if ... fi block to not run it when early_modules_load.conf is not
present.

> > > 8. Modprobe.d support -- to support the last point, config files
> > > in
> > > /etc/modprobe.d are obeyed.

Ok, not changed...
> > > 
> > > If someone could test this on sysv and let me know how it goes I
> > > can
> > > push this to the book and file an errata.

All my tests are on sysv, but I think there is no difference with
systemd, since systemd is not run in early environment (only udev,
which is the same for sysv and sysd).

Regards
Pierre

-- 
http://lists.linuxfromscratch.org/sympa/info/blfs-dev
Unsubscribe: See the above information page