BTRFS + snapshots on Arch Linux / CachyOS

BTRFS gives you something ext4 does not: system rollback. You installed something that broke Xorg? Go back to 10 minutes ago. Kernel panic after update? Reboot into the previous snapshot. Database corrupted? Restore a file from before the crash.

This guide is my configuration I have used for 3 years on 4 machines (work laptop, home server, test station, VPS).

Installing the system on BTRFS (CachyOS / Arch)

I use a partition layout that gives me full control:

# Partition layout (UEFI + LUKS optional)
# /dev/sda1 - 512MB - EFI partition (FAT32)
# /dev/sda2 - rest  - LUKS encrypted BTRFS

# After installing CachyOS / Arch: create subvolumes
mount /dev/sda2 /mnt
btrfs subvolume create /mnt/@
btrfs subvolume create /mnt/@home
btrfs subvolume create /mnt/@snapshots
btrfs subvolume create /mnt/@var_log
btrfs subvolume create /mnt/@var_cache

umount /mnt

Why 5 subvolumes, not one? Snapshots copy data. If you have one @ subvolume with /var inside, the snapshot also copies logs and cache. Performance drops, size grows. Separate subvolumes = snapshot only what you want.

Mounting with zstd compression and noatime

Add to /etc/fstab:

# /etc/fstab
UUID=xxxx-xxxx  /              btrfs   subvol=@,compress=zstd:3,noatime,ssd,discard=async  0  1
UUID=xxxx-xxxx  /home          btrfs   subvol=@home,compress=zstd:3,noatime,ssd,discard=async  0  2
UUID=xxxx-xxxx  /snapshots     btrfs   subvol=@snapshots,compress=zstd:3,noatime,ssd,discard=async  0  2
UUID=xxxx-xxxx  /var/log       btrfs   subvol=@var_log,compress=zstd:3,noatime,ssd,discard=async  0  2
UUID=xxxx-xxxx  /var/cache     btrfs   subvol=@var_cache,compress=zstd:3,noatime,ssd,discard=async  0  2
UUID=xxxx-xxxx  /boot/efi      vfat    umask=0077  0  2

Three options that make a difference:

• compress=zstd:3 — transparent compression. Gain: 10-30% space for free, CPU overhead negligible. Level 3 is the sweet spot (level 1 is faster but compresses less, 9+ gives 2% more for 3x CPU).
• noatime — do not write the last access time of a file. Gain: fewer writes, longer SSD life. Without this option the kernel updates atime on every read().
• discard=async — TRIM for SSD in the background, does not block I/O. Alternative: cron with fstrim -A.

Timeshift: desktop snapshots

Timeshift is a GUI + CLI tool that takes snapshots of the whole system (except /home) and restores with one click. Ideal for laptop/desktop.

Installation and configuration

# CachyOS / Arch
sudo pacman -S timeshift

# Initialize (auto-detects BTRFS)
sudo timeshift --snapshot-device /dev/sda2

# Settings:
# - type: BTRFS (not RSYNC — on BTRFS that wastes space)
# - location: subvolume @snapshots
# - schedule: Daily (7 days) + Weekly (4 weeks) + Monthly (3 months)
# - exclude: nothing (Timeshift+BTRFS copies only changes)

First snapshot and restore

# Manual snapshot before a major change
sudo timeshift --create --comments "Before kernel 6.10 update"

# List snapshots
sudo timeshift --list

# Restore from the running system (reboot required)
sudo timeshift --restore

# Restore from GRUB (when the system does not boot)
# Reboot → GRUB → "Arch Linux snapshots" → pick a snapshot

That last point is the key. GRUB integrates with BTRFS — at boot it shows a list of snapshots you can boot from. Kernel panic after pacman -Syu? Reboot, pick yesterday's snapshot, fix, return to the current one.

Snapper: per-transaction snapshots (pre/post pacman)

Snapper is a low-level tool that automatically creates snapshots before and after every pacman operation. Plus a timeline (hourly, daily). It needs more configuration than Timeshift, but gives you precise control.

Configuration for / and /home

sudo pacman -S snapper snap-pac

# Create config for subvolume @
sudo snapper -c root create-config /

# Create config for subvolume @home
sudo snapper -c home create-config /home

# Edit timeline cleanup so you do not bloat
sudo nano /etc/snapper/configs/root
# Change:
#   TIMELINE_LIMIT_HOURLY="10"
#   TIMELINE_LIMIT_DAILY="7"
#   TIMELINE_LIMIT_WEEKLY="4"
#   TIMELINE_LIMIT_MONTHLY="3"
#   TIMELINE_LIMIT_YEARLY="0"

# Enable systemd timers
sudo systemctl enable --now snapper-timeline.timer
sudo systemctl enable --now snapper-cleanup.timer

snap-pac integrates snapper with pacman. From now on every pacman -S creates a "pre" and "post" snapshot automatically. If an update breaks the system:

# List snapshots
sudo snapper list

# Diff between two snapshots
sudo snapper status 42..43

# Restore a single file
sudo snapper undochange 42..43 --filename /etc/fstab

# Restore everything
sudo snapper undochange 42..43

Disabling COW for databases and VMs

BTRFS uses Copy-on-Write — every file modification creates a new block. For databases (PostgreSQL, MySQL, SQLite) this is a performance disaster: instead of one write you get 10. Fix:

# For new files in a directory
chattr +C /var/lib/postgresql/data
chattr +C /var/lib/mysql
chattr +C /var/lib/libvirt/images

# For existing files: you must defragment
btrfs filesystem defragment -czstd /var/lib/postgresql/data

After chattr +C new files do not use COW. Database performance returns to ext4 levels. Snapshots still work (they copy the whole subvolume, not individual COW blocks).

Off-site backup: btrfs send/receive

BTRFS has a built-in incremental backup mechanism — btrfs send and btrfs receive. You send only the delta, not the whole snapshot.

# On the backup server: create subvolume to receive
btrfs subvolume create /backup/laptop/@

# On the laptop: send the first snapshot
sudo btrfs send /.snapshots/1/snapshot | ssh backup@server "btrfs receive /backup/laptop/@"

# Subsequent snapshots: only delta
sudo btrfs send -p /.snapshots/1/snapshot /.snapshots/5/snapshot | \
  ssh backup@server "btrfs receive /backup/laptop/@"

This is 10-100x faster than rsync on large snapshots, because you send only changed blocks. Backing up a 200GB snapshot with 2GB of changes = 2GB transfer, not 200GB.

Monitoring: how much space do snapshots take

# How much space snapshots take
sudo btrfs filesystem du -s /.snapshots/*/snapshot

# Overall disk usage
btrfs filesystem usage /

# How much space you would free by deleting snapshots
sudo snapper list --columns number,description,used-space

Rule: at least 20-30% free space. BTRFS needs free blocks for COW allocation. When you fill the disk 95%+, the system starts failing in non-obvious ways (allocation fails, snapshots stop working, performance drops).

What is next

• WordPress REST API: practical use cases
• Cloudflare for WordPress
• Own mail server Postfix + Dovecot

If you want me to configure BTRFS with snapshots on your existing server (without reinstalling) — get in touch.