Chroot from a live USB correctly (including the Btrfs subvol=@ trap)
Checked the Omarchy 4 layout on this machine (findmnt, /etc/fstab, lsblk, /proc/cmdline, /etc/default/limine, /etc/limine-entry-tool.d/*.conf) and against the ISO configurator in omacom/omarchy-iso quattro: the root is LUKS2 on p2 with btrfs subvolumes @, @home, @log and @pkg mounted at /, /home, /var/log and /var/cache/pacman/pkg, the ESP is vfat at /boot, and Snapper's /.snapshots is a subvolume nested inside @ (inode 256 here), so there is no @snapshots to mount. The record mounts only @ and @home, so pacman inside its chroot writes the package cache and logs into the @ subvolume instead of @pkg and @log. The mapper name cryptroot is neither of the names Omarchy uses (omarchy_root in the current configurator, root on this 4.0.0 install per the cmdline). Any name works for a chroot because the rebuilt cmdline is read from /etc/default/limine and the drop-ins inside the chroot, and the corrected fix says so. The rebuild step was the main defect: on Omarchy 4 /etc/mkinitcpio.d/ is empty, so mkinitcpio -P dies with 'No presets found in /etc/mkinitcpio.d' (mkinitcpio 41.1 line 986, confirmed here), the kernel boots as the UKI /boot/EFI/Linux/omarchy_linux.efi, and the right command is limine-mkinitcpio (or pacman -S linux, whose 90-mkinitcpio-install hook from limine-mkinitcpio-hook runs the same script). The verify step's ls -l /boot/vmlinuz-linux therefore does not apply to Omarchy. Confirmed from /usr/bin/omarchy-update-pacman-guard that the ALPM guard only aborts when -S and -u are combined, so pacman -S linux in a chroot is not blocked. Confirmed from /usr/lib/limine/limine-common-functions that the ESP is found by a vfat mount at /efi, /boot, /boot/efi or /limine and the tool stops with 'FAT32 boot partition not found' otherwise, and from limine-mkinitcpio-install that the UKI is only built when /sys/firmware/efi exists, which is why UEFI mode matters. Confirmed from the archiso releng profile (GitHub mirror) that both ISOs autologin root on tty1 with an empty root password and ship arch-install-scripts, btrfs-progs and dosfstools, and from omarchy-iso build-iso.sh that the Omarchy ISO is seeded from releng and adds cryptsetup, so the record's sudo prefixes are unnecessary and a shell is on Ctrl+Alt+F2. The cited EndeavourOS thread (post 15) does support the subvol=@ claim, arch-chroot(8) was re-read (plain arch-chroot /mnt still works, -S is optional), and the Omarchy manual security page confirms LUKS is the installer default. Not exercised: no chroot was performed, no VM was booted (the pool was down), and the mapper name on installs between 4.0.0 and the current ISO was not sampled beyond this machine.
Writing into a wrongly mounted chroot damages the wrong tree. On btrfs the top-level subvolume (subvolid 5) holds @, @home, @log and @pkg as directories and nothing else, so a chroot into it has no /etc/fstab and no pacman. Confirm /etc/fstab is visible before writing anything. With /boot left unmounted, plain Arch's mkinitcpio -P writes /boot/vmlinuz-linux into the root subvolume where the bootloader never looks. Omarchy's limine-mkinitcpio refuses instead (FAT32 boot partition not found), but an ESP mounted at the wrong path still leaves the UKI where the firmware will not find it. Do not run btrfs check --repair or delete subvolumes while hunting for the root: the missing files are a mount mistake, not filesystem damage, and on Omarchy that one filesystem holds /home and every Snapper snapshot.
Symptom
You boot a live USB to repair the system, run arch-chroot /mnt, reinstall the kernel, and nothing improves. Or the repair itself fails with things like:
'/boot/initramfs-linux.img' not found
dracut-install: Failed to find module 'zfs'
and after rebooting you are back in the emergency shell.
Cause
On Btrfs installs (Omarchy, EndeavourOS and CachyOS all default to Btrfs subvolumes) mounting /dev/nvme0n1p2 /mnt mounts the *top-level* subvolume, not the actual root. The chroot then sees an almost-empty filesystem, package scripts silently operate on the wrong tree, and the initramfs is written somewhere the bootloader never reads. The same class of failure comes from forgetting to mount the ESP at /mnt/boot, or forgetting to open the LUKS container first.
Fix
Boot the Arch ISO or the Omarchy ISO in **UEFI mode**. Both log you in as root, so no sudo is needed. On the Omarchy ISO the installer owns tty1: press Ctrl+Alt+F2 and log in as root with no password (the live image is built on the Arch releng profile, which ships an empty root password plus arch-install-scripts, btrfs-progs, dosfstools and cryptsetup).
On Omarchy 4, try the **Snapshots** entry in the Limine menu before reaching for the ISO: it boots a Snapper snapshot of the root subvolume, and omarchy-snapshot restore can roll back from there. If that boots, you may not need a chroot at all.
# 1. see the layout. Omarchy: p1 is the vfat ESP, p2 is LUKS (bare btrfs on an unencrypted install)
lsblk -f
# 2. unlock encryption if present (the Omarchy installer encrypts by default).
# The mapper name is yours to choose here: Omarchy's own installs use
# "omarchy_root" (current ISO) or "root" (earlier 4.x installs), and the
# rebuilt boot files do not inherit whatever you type.
cryptsetup open /dev/nvme0n1p2 omarchy_root
# 3. list the subvolumes so you mount the right ones
mount /dev/mapper/omarchy_root /mnt
btrfs subvolume list /mnt
umount /mnt
# 4. mount the ROOT subvolume, then every other subvolume fstab expects.
# Omarchy 4 (from the ISO configurator): @ @home @log @pkg
mount -o subvol=@,compress=zstd /dev/mapper/omarchy_root /mnt
mount -o subvol=@home,compress=zstd /dev/mapper/omarchy_root /mnt/home
mount -o subvol=@log,compress=zstd /dev/mapper/omarchy_root /mnt/var/log
mount -o subvol=@pkg,compress=zstd /dev/mapper/omarchy_root /mnt/var/cache/pacman/pkg
# 5. mount the ESP where the installed system expects it. Omarchy: /boot
mount /dev/nvme0n1p1 /mnt/boot
# 6. chroot
arch-chroot /mnt
Snapper's /.snapshots on Omarchy is a subvolume nested inside @, so it comes along with step 4 and needs no separate mount. Plain Arch installed with archinstall's btrfs layout has a separate @snapshots for /.snapshots. Mount it the same way.
Inside the chroot, cat /etc/fstab tells you the exact mount points and subvolume names the installed system uses. If /mnt/boot is empty or /etc/fstab names /boot/efi or /efi, unmount and remount the ESP there before doing anything else.
For a non-Btrfs (ext4) install, step 4 is simply mount /dev/nvme0n1p2 /mnt.
Rebuilding the boot files differs between Omarchy 4 and plain Arch:
**Omarchy 4.** /etc/mkinitcpio.d/ is empty, so mkinitcpio -P stops with No presets found in /etc/mkinitcpio.d. The kernel boots as a UKI at /boot/EFI/Linux/omarchy_linux.efi, built by limine-mkinitcpio-hook. Run:
limine-mkinitcpio # rebuilds the UKI and the Limine entries
or pacman -S linux, whose 90-mkinitcpio-install hook runs the same script. The Omarchy ALPM guard only aborts a transaction that combines -S with -u, so a plain pacman -S linux goes through. The kernel command line is read from /etc/default/limine and /etc/limine-entry-tool.d/*.conf inside the chroot, not from the live ISO, so the mapper name from step 2 does not end up in the UKI. The tool locates the ESP by looking for a vfat mount at /efi, /boot, /boot/efi or /limine, and with /boot unmounted it stops with FAT32 boot partition not found instead of writing anywhere. The UKI is only built when /sys/firmware/efi exists, which is why the ISO must be booted in UEFI mode.
**Plain Arch.** pacman -S linux or mkinitcpio -P writes /boot/vmlinuz-linux and /boot/initramfs-linux.img. Re-run your bootloader's install or config step afterwards if it copies files.
When done:
exit
umount -R /mnt
cryptsetup close omarchy_root
reboot
arch-chroot is provided by arch-install-scripts and handles mounting /dev, /proc, /sys and copying resolv.conf for you. Do not use plain chroot unless you replicate those bind mounts by hand.
Verify
Inside the chroot, cat /etc/fstab shows the subvol=/@ line for / and pacman -Q linux prints your installed kernel. If either fails you mounted the wrong subvolume. Omarchy 4: after limine-mkinitcpio, ls -l /boot/EFI/Linux/omarchy_linux.efi shows a fresh timestamp. Plain Arch: after pacman -S linux, ls -l /boot/vmlinuz-linux /boot/initramfs-linux.img shows fresh timestamps.
Sources
- https://forum.endeavouros.com/t/boot-initramfs-linux-img-not-found-chroot-on-live-and-update-doesnt-help/50238
- https://man.archlinux.org/man/arch-chroot.8
- https://learn.omacom.io/2/the-omarchy-manual/93/security
- https://wiki.archlinux.org/title/Chroot
- https://gitlab.archlinux.org/archlinux/arch-install-scripts/-/raw/master/arch-chroot.in
- https://github.com/omacom/omarchy-iso/blob/quattro/configs/airootfs/root/configurator
- https://github.com/omacom/omarchy-iso/blob/quattro/configs/airootfs/usr/share/omarchy-iso/orchestrator/phases_impl.py
- https://github.com/omacom/omarchy-iso/blob/quattro/builder/build-iso.sh
- https://github.com/omacom/omarchy-iso/blob/quattro/configs/airootfs/root/.automated_script.sh
- https://github.com/archlinux/archiso/blob/master/configs/releng/packages.x86_64
- https://github.com/archlinux/archiso/blob/master/configs/releng/airootfs/etc/shadow
- https://github.com/archlinux/archiso/blob/master/configs/releng/airootfs/etc/systemd/system/getty@tty1.service.d/autologin.conf
- https://github.com/basecamp/omarchy/blob/quattro/install/config/snapper.sh