42 Abu Dhabi

ft_linux

Build Your Own Linux Distribution from Scratch

4 weeks

Individual Project

Linux KernelGCC & ToolchainSysVinitGRUBEudevBash & Autotools

ft_linux represents the ultimate test of low-level systems understanding: building an operating system from source without using package managers or pre-built binaries. Starting from a raw virtual disk, the project involves disk partitioning (/boot, /, swap), compiling a cross-compilation toolchain, building over 70+ essential GNU packages from source, configuring SysVinit and Eudev device management, and compiling a tailored Linux kernel with custom version branding (how_to_train_your_kernel), resulting in a standalone, bootable OS with working network connectivity.

Source Code

Key Features

Custom Compiled Kernel

Configured and compiled Linux Kernel from source with modular driver support and a customized version string tagged with login credentials.

70+ Source-Compiled Packages

Compiled Glibc, Binutils, Coreutils, Bash, Make, Tar, and network utilities directly from upstream source tarballs.

FHS-Compliant Layout

Architected a strict Filesystem Hierarchy Standard tree with partitioned /boot, swap, and ext4 root partitions.

Bootable OS with Networking

Standalone system booted via GRUB with operational DHCP internet access, user management, and dynamic device detection via Eudev.

Development Journey

Phase 1

Host Preparation & Disk Layout

Partitioned dedicated target storage (/dev/sdb) into separate /boot (200MB ext2), swap (2GB), and root (ext4) partitions. Configured the isolated LFS build environment and user.

Phase 2

Cross-Toolchain Compilation

Built the temporary cross-compiler toolchain: Binutils, GCC pass 1, Linux API headers, Glibc, and Libstdc++, breaking dependencies on the host OS.

Phase 3

Chroot & Base System Construction

Entered the isolated chroot jail, mounted virtual kernel filesystems (/dev, /proc, /sys), and compiled 70+ essential system packages, utilities, and libraries.

Phase 4

Kernel Configuration & Bootloader

Executed make menuconfig to tailor kernel drivers, compiled the bzImage, set up /etc/fstab, installed GRUB to the MBR, and verified successful bare-metal booting.

Challenges & Solutions

Circular Toolchain Dependencies

Problem:

Compiling a C compiler (GCC) requires an existing C compiler and standard library, which can leak host system dynamic links into the target OS.

Solution:

Employed a rigorous multi-pass cross-compilation approach: building a static temporary toolchain first, adjusting the specs file, and verifying every binary against the LFS linker.

Kernel Hardware Driver Selection

Problem:

Building a lean monolithic or modular kernel without bloated generic distributions while ensuring storage controllers (SATA/SCSI) and network devices initialize properly at boot.

Solution:

Carefully audited hardware controller IDs using lshw and lspci, enabling essential virtio and ahci storage drivers directly built-in (y) rather than loadable modules (m).

Init System & Device Node Management

Problem:

Handling dynamic device node creation on boot without modern systemd abstraction.

Solution:

Configured Eudev (standalone udev fork) and crafted SysVinit inittab runlevel scripts (/etc/init.d) to mount devtmpfs and populate device nodes dynamically.

scripts/provision_host.sh

bash

#!/bin/bash
set -euo pipefail

LFS_DISK="${1:-/dev/sdb}"
LFS="/mnt/lfs"

# Create partitions: 1: /boot (+200M), 2: swap (+2G), 3: root (rest)
parted -s "$LFS_DISK" mklabel msdos
parted -s "$LFS_DISK" mkpart primary ext2 1MiB 201MiB
parted -s "$LFS_DISK" set 1 boot on
parted -s "$LFS_DISK" mkpart primary linux-swap 201MiB 2249MiB
parted -s "$LFS_DISK" mkpart primary ext4 2249MiB 100%

# Format filesystems
mkfs.vfat -F 32 "${LFS_DISK}1" || mkfs.ext2 "${LFS_DISK}1"
mkswap "${LFS_DISK}2"
mkfs.ext4 -F "${LFS_DISK}3"

# Mount root and boot
mkdir -pv "$LFS"
mount -v -t ext4 "${LFS_DISK}3" "$LFS"
mkdir -pv "$LFS/boot"
mount -v -t ext2 "${LFS_DISK}1" "$LFS/boot"
swapon -v "${LFS_DISK}2"

Previous Next