Create a custom initramfs

Setup directories

Clone this repository in the home directory

cd ~
git clone https://github.com/SayantanRC/src_custom_initramfs.git

and move to Gathering binaries. Or follow the below steps to manually setup the directory structure.

To manually create the directories, use the below commands.

mkdir ~/src_custom_initramfs
mkdir -p ~/src_custom_initramfs/{dev,etc,mnt/root,proc,root,sbin,secure,sys,tmp,usr/bin,usr/lib,usr/lib64}

Create the following symlinks.

cd ~/src_custom_initramfs
ln -s usr/bin .
ln -s usr/lib .
ln -s usr/lib64 .

Gathering binaries

The following binaries are recommended to be added. More (or less) binaries may be added as per requirement.
bash, mkdir, ls, cat, tail, grep, cut, awk, ntfs-3g, mount, umount, insmod, lsmod, modprobe, switch_root, blkid, ntfsfix
To find the location of the binary, in terminal, type type <binary-name> or which <binary-name>. Example: which modprobe gives output /usr/bin/modprobe
We need to copy each of them in the respective directories. In the above example of modprobe, the command would be cp -p /usr/bin/modprobe usr/bin/modprobe.

AUTOMATION!

cd ~/src_custom_initramfs  
./1.gather_binaries.sh  

Gathering library dependencies

Each of the above binaries have one or several library dependencies. Example:
ldd /usr/bin/modprobe gives output:

linux-vdso.so.1 (0x00007ffee9f24000)
libzstd.so.1 => /usr/lib/libzstd.so.1 (0x00007f947c7d4000)
liblzma.so.5 => /usr/lib/liblzma.so.5 (0x00007f947c7ac000)
libz.so.1 => /usr/lib/libz.so.1 (0x00007f947c792000)
libcrypto.so.1.1 => /usr/lib/libcrypto.so.1.1 (0x00007f947c4b4000)
libc.so.6 => /usr/lib/libc.so.6 (0x00007f947c2e7000)
libpthread.so.0 => /usr/lib/libpthread.so.0 (0x00007f947c2c6000)
libdl.so.2 => /usr/lib/libdl.so.2 (0x00007f947c2bd000)
/lib64/ld-linux-x86-64.so.2 => /usr/lib64/ld-linux-x86-64.so.2 (0x00007f947c8f7000)

All these dependencies are to be added individually for all the binaries. Example cp -p /usr/lib/libzstd.so.1 usr/lib/libzstd.so.1 and so on.
IMPORTANT: The first library linux-vdso.so.1 is provided by the kernel and is not needed to be copied.

AUTOMATION!

cd ~/src_custom_initramfs  
./2.gather_libraries.sh  

Check if chroot works

At this point we should be able to chroot into our initramfs directory.

cd ~/src_custom_initramfs

sudo chroot . usr/bin/bash

Gathering modprobe modules

We will add the following modprobe modules along with their dependencies.
ext4, fuse, loop, ahci, libahci
Module files and their dependencies can be seen via the command modprobe --show-depends <module-name>. For example, modprobe --show-depends ext4 gives result:

insmod /lib/modules/5.11.6-1-MANJARO/kernel/fs/jbd2/jbd2.ko.xz
insmod /lib/modules/5.11.6-1-MANJARO/kernel/fs/mbcache.ko.xz
insmod /lib/modules/5.11.6-1-MANJARO/kernel/lib/crc16.ko.xz
insmod /lib/modules/5.11.6-1-MANJARO/kernel/arch/x86/crypto/crc32c-intel.ko.xz
insmod /lib/modules/5.11.6-1-MANJARO/kernel/crypto/crc32c_generic.ko.xz
insmod /lib/modules/5.11.6-1-MANJARO/kernel/fs/ext4/ext4.ko.xz

Note that some modules like ahci and libahci may be internal modules and not available in the current kernel. These modules are related to disk drives (SATA drives). They are located at:

/lib/modules/<kernel_version>/kernel/drivers/ata/libahci.ko
/lib/modules/<kernel_version>/kernel/drivers/ata/ahci.ko

If we want to make an initramfs for this kernel, we can go ahead and copy these files. Syntax will be something like mkdir -p lib/modules/5.11.6-1-MANJARO/kernel/fs/jbd2; cp -pv /lib/modules/5.11.6-1-MANJARO/kernel/fs/jbd2/jbd2.ko.xz lib/modules/5.11.6-1-MANJARO/kernel/fs/jbd2/jbd2.ko.xz for each of the modules.
If however, we need to get the modules for a different kernel, we will need to copy from that kernel modules (various cases arise here like copying from an OS on a USB, copying from a backup img file etc).

AUTOMATION!

Find path for variable module_source_path

The path specified by this variable must exist.
The variable module_source_path can refer to a custom location, ending with a kernel release, just one level above the 'kernel' directory. Example, a system can have 2 linux kernels installed, say 5.11 and 5.4. Then the module_source_path can be any of the following:

"/lib/modules/5.4"
"/lib/modules/5.11"

It can also be set to an external source, say we have mounted a live USB, then module_source_path can also be set as /run/media/<user>/arch_root/lib/modules/5.9.6-arch1-1; where /run/media/<user>/arch_root is the mount point.
The value must end with a kernel-release similar to ouput of uname -r. It cannot take values like:

"/lib/modules"
"/lib/modules/5.11/kernel"
"/lib/modules/5.4/build"

etc.

cd ~/src_custom_initramfs  
./3.gather_mods.sh  

Enter the path you found out previously.

Some explanation of the script

This script checks the presence of .ko modules and .ko.xz modules and copies them accordingly. It also lists the copied modules in a file named module_list.txt.
The variable init_mod_dir_name can be set as any string like "IceCream", "MyInit" anything; but in the final init script, insmod will need to refer to the proper file paths. Example, if the value is set as "MojoJojo", in init script, insmod will need to be used as:

insmod /lib/modules/MojoJojo/kernel/fs/ext4/ext4.ko.xz

If we are using the init file from this repository as a template, along with module_list.txt (check the next section), then we do not need to worry about inserting the modules as it will be done for automatically.

Create the module_list.txt file

If the above automation is used, then this file is already created. If not, we need to manually create this file under the initramfs root.

cd ~/src_custom_initramfs
vim module_list.txt

Contents of the file will be something like:

/lib/modules/generic/kernel/fs/jbd2/jbd2.ko.xz
/lib/modules/generic/kernel/fs/mbcache.ko.xz
/lib/modules/generic/kernel/lib/crc16.ko.xz
/lib/modules/generic/kernel/arch/x86/crypto/crc32c-intel.ko.xz
/lib/modules/generic/kernel/crypto/crc32c_generic.ko.xz
/lib/modules/generic/kernel/fs/ext4/ext4.ko.xz
/lib/modules/generic/kernel/fs/fuse/fuse.ko.xz
/lib/modules/generic/kernel/drivers/block/loop.ko.xz

Finally create the init script

Create a file named as init in the root of the initramfs, i.e. at ~/src_custom_initramfs/init
How to write an init script can be found in various online resources. A template is given in this repository, feel free to use it.

Specific use case - booting an img file.

The expected menuentry is (can be set under /etc/grub.d/40_custom):

menuentry "Arch IMG" {
        img_part=/dev/sda7
        img_path=/arch.img
        search --no-floppy --set=root --file $img_path
        loopback loop $img_path
        linux (loop)/boot/vmlinuz-linux img_part=$img_part img_path=$img_path
        initrd (loop)/boot/my_initramfs.igz
}

Here we will use the commandline arguments to set the root as the image because GRUB cannot directly boot from an img file.
The init script will look like this:

#!/usr/bin/bash

mount -t devtmpfs  devtmpfs  /dev
mount -t proc      proc      /proc
mount -t sysfs     sysfs     /sys
mount -t tmpfs     tmpfs     /tmp

# insert modules
while read -r mod_path
do
  insmod $mod_path
done < module_list.txt


# commandline arguments
args=$(cat /proc/cmdline)

# do your thing here, finally mount the boot location to /mnt/root

# first argument value is "(loop)/boot/vmlinuz-linux" we don't need that.
# Our arguments start from second token.
arg_part=$(echo $args | awk '{print $2}')
arg_path=$(echo $args | awk '{print $3}')

img_part=$(echo $arg_part | cut -d '=' -f2)
img_path=$(echo $arg_path | cut -d '=' -f2)

echo ""
echo "Partition: $img_part"
echo "IMG path: $img_path"

echo ""
echo "All devices:"
ls /dev

echo ""
echo "Kernel version:"
uname -a

# check if ntfs partition, if yes, use ntfsfix
part_type="$(blkid $img_part | awk -F 'TYPE' '{print $2}' | awk -F '[ ="]' '{print $3}')"
printf "\nPartition type (blkid): $part_type\n\n"
if [[ "$part_type" == "ntfs" ]]; then
  printf "Checking and fixing NTFS...\n\n"
  ntfsfix -d $img_part
  echo ""
fi

echo "Mounting partition..."
mkdir /img_partition
if [[ "$part_type" == "ntfs" ]]; then
  echo "Mounting with ntfs-3g..."
  ntfs-3g $img_part /img_partition
else
  mount -o rw $img_part /img_partition
fi

echo "Mounting IMG as loop..."
mount -o rw,loop,sync /img_partition/$img_path /mnt/root




# clean up
umount /proc
umount /sys

# Boot
printf "\nBooting...\n\n"
exec switch_root /mnt/root /sbin/init

Important

In the above scenario, some more modifications are to be done before booting.

1. The img file needs to be mounted as read-write and the etc/fstab entries need to be changed. UUID of the / partition must be same as the UUID of the actual img file. This can be found by the command blkid <path_to_arch.img_file>
2. Also in most cases, the modules from the current distro will not work on this arch.img file. The img file needs to be mounted, and in the AUTOMATION section of "Gathering modprobe modules", the variable module_source_path needs to point to the modules of this img mount point (say "/run/media/$USER/arch_root/lib/modules/5.9.6/")

Make the init file executable

This is very important and is often missed. Hence added under a seperate heading.

cd ~/src_custom_initramfs
chmod +x init

Compile the initramfs

The compiled output should not reside inside the initramfs directory.

cd ~/src_custom_initramfs
./4.compile.sh

A new my_initramfs.igz file is created one directory level above. In this case, directly under home.
Now this my_initramfs.igz file can be placed in the target device's boot directory and called from initrd of Grub.