[Intermediate] — what happens between power-on and a login prompt, and how to get in when that sequence breaks. The note you want before you need it.

The kid version first

Booting is a relay race where each runner is bigger than the last and knows only how to find the next one.

firmware (UEFI/BIOS)   finds a bootloader on disk
   └── GRUB            finds and loads a kernel + initramfs
        └── kernel     mounts the real root filesystem
             └── systemd (PID 1)  starts everything else, in dependency order

Every boot failure is one runner failing to hand off. Knowing which handoff broke tells you which tool fixes it — and the symptom tells you the runner:

SymptomFailed handoffWhere you fix it
No bootable device / firmware screenfirmware → bootloaderFirmware settings, disk order, reinstall GRUB
GRUB menu, then nothingGRUB → kernelKernel parameters, wrong/removed kernel
Kernel panic, “cannot mount root”kernel → root fsBroken initramfs, wrong root UUID
Boots to emergency shellsystemd → servicesAlmost always fstab
Boots but a service is deadsystemd unit failurejournalctl -u <unit>

That last-but-one row is the one you’ll actually hit. See disks & filesystems — a mistyped fstab line is the single most common self-inflicted boot failure.

GRUB — the part you can edit at boot

Hold Shift (BIOS) or Esc (UEFI) during startup to get the menu. Press e on an entry to edit its boot parameters for this boot only — nothing is written to disk, so a reboot undoes any mistake.

Two parameters worth knowing on the linux line:

  • systemd.unit=rescue.target — single-user mode: root filesystem mounted, minimal services, root shell. For fixing a broken config.
  • systemd.unit=emergency.target — even more minimal: root mounted read-only, almost nothing started. For when rescue itself won’t come up (e.g. fstab is so broken that mounting fails).

Then Ctrl-X to boot with the edit.

In emergency mode the root filesystem is read-only, so your first move before editing anything is:

mount -o remount,rw /

Forgetting that produces a baffling “cannot write file” in an editor when you’re already stressed.

Permanent changes go in /etc/default/grub, followed by regenerating the config — and the command differs by distro, which is exactly the sort of thing that bites when you follow the wrong tutorial:

sudo update-grub                                     # Debian/Ubuntu
sudo grub2-mkconfig -o /boot/grub2/grub.cfg          # RHEL/Fedora

Fixing a broken fstab from rescue mode

The most common real repair:

mount -o remount,rw /
nano /etc/fstab            # comment out the offending line
mount -a                   # verify — no errors means it will boot
reboot

Adding nofail to non-essential mounts in fstab prevents this class of failure entirely — the boot continues instead of dropping to a shell when that device is missing. On a cloud VM with attached volumes, nofail is close to mandatory, because a volume that fails to attach otherwise takes the whole machine down.

systemd targets — what “runlevel” became

A target is a named set of units to reach.

systemctl get-default
sudo systemctl set-default multi-user.target   # boot to console, no GUI — right for a server
sudo systemctl isolate rescue.target           # switch now, without rebooting
TargetRoughly
graphical.targetFull desktop
multi-user.targetNetworked, console only — the server default
rescue.targetSingle user, minimal services
emergency.targetRoot shell, root mounted read-only

Setting a server to multi-user.target is free performance and a smaller attack surface if a GUI got installed by accident.

Diagnosing a slow boot

systemd-analyze                  # total time, split firmware/loader/kernel/userspace
systemd-analyze blame            # slowest units, descending
systemd-analyze critical-chain   # the dependency path that actually gated the boot

blame is the famous one but critical-chain is the useful one: a unit can take 30 seconds and delay nothing if the boot wasn’t waiting on it. Chase the chain, not the list.

The usual culprits are a network-wait unit hanging on an interface that will never come up, and a mount waiting on a device that isn’t there — both are the same shape of problem, and nofail or disabling the wait fixes both.

Key insight

Firmware → GRUB → kernel → systemd, each finding the next. Identify which handoff failed from the symptom and you know which tool to reach for. Learn to press e at the GRUB menu on a machine that isn’t broken — it’s a bad time to learn the keystroke when it is.

↑ Back to Linux