OPINIONATED OMARCHY
TROUBLESHOOTING CORPUS
790
RECORDS
790
AUDITED
0
UNCHECKED
2388
SOURCES
LIVE
boot-kernel critical occasional DISPUTE

"UNEXPECTED INCONSISTENCY; RUN fsck MANUALLY" drops the boot into a maintenance shell

omarchyarchendeavouroscachyosmanjaroext4btrfsxfsdesktoplaptop
CORRECTED An auditor found the fix wrong or incomplete and rewrote it.
AUDIT NOTE

The generic Arch content held: the Fsck, Kernel_parameters and Btrfs wiki pages (raw wikitext) confirm the two check paths, fsck.mode=skip/force, fsck.repair=yes, the rw requirement for the mkinitcpio fsck hook, the fstab pass-number rules and the warning on btrfs check --repair, and the mkinitcpio hooks/fsck, install/fsck and init_functions sources confirm how the hook builds and what the initramfs does on status 4. rescue=usebackuproot (btrfs 5.9+), btrfs check --readonly as the default, findmnt --verify and fsck.fat exit codes 0/1/2 were confirmed from the local btrfs(5), btrfs-check(8), findmnt(8) and fsck.fat(8) man pages, and systemd-fsck@.service(8) confirms that a status-4 failure on a non-nofail fstab entry activates emergency.target. What did not hold is the Omarchy specialisation. On this machine /etc/mkinitcpio.conf.d/omarchy_hooks.conf does keep fsck in HOOKS, but with autodetect the hook packs only fsck.btrfs, which /usr/bin/fsck.btrfs (btrfs-progs 7.1) shows is a script that exits 0 without checking, and every btrfs line in /etc/fstab (/ /home /var/log /var/cache/pacman/pkg, all subvol=/@...) carries pass 0. The only checked entry is the vfat ESP at /boot with pass 2, and systemctl status shows systemd-fsck@ running fsck.fat 4.2 on it at every boot. fsck.fat cannot return 4, so the record's symptom (an ext4 message on nvme0n1p2, which is the LUKS root on Omarchy) cannot occur on an Omarchy install as shipped and only appears for an extra ext4 or XFS disk the user added with a non-zero pass number. The btrfs advice was moved to its own branch labelled as the Omarchy 4 root, with scrub as the safe check. The claim that root has no password on Omarchy is not a constant: getent shadow root on this 4.0.0 workstation shows the account locked (!*), but the current ISO configurator writes root_enc_password from the same hash as the user password (archinstall_adapter.py creates root with it) and omarchy-provision-owner line 741 sets root to the owner password with chpasswd on deferred-provisioning installs, so the fix now says to try the login password first. The kernel-parameter advice was given an Omarchy branch: the cmdline is embedded in the UKI, so it goes in a /etc/limine-entry-tool.d/ drop-in followed by limine-mkinitcpio, and systemd-stub(7) states that a passed cmdline is ignored under Secure Boot when a .cmdline section exists. Not exercised: no filesystem was damaged or repaired, no VM was booted (pool was down), the Limine menu editor path was not tested, and whether an ext4 root on plain Arch drops to the initramfs shell was taken from init_functions rather than reproduced.

The cause above was rewritten on 2026-09-06 to match this note; the fix was corrected by the audit itself.
RISK

Running fsck on a mounted read-write filesystem will destroy it. Confirm with findmnt and umount first, and use a live USB for root. fsck -y on a physically failing disk can throw away recoverable data. If the drive is suspect, image it first with ddrescue and repair the image. Never point e2fsck or xfs_repair at a btrfs device by hand: fsck dispatches by type and lands on the harmless fsck.btrfs, but fsck.ext4 /dev/mapper/root on a btrfs root will rewrite it into garbage. btrfs check --repair is called dangerous by btrfs-progs itself and can turn a mountable filesystem into an unmountable one. Reach for it only after mount -o ro,rescue=usebackuproot and btrfs check --readonly have failed and you hold a backup or an image, and never with --force on a mounted filesystem. On Omarchy 4 that one filesystem holds /, /home, the package cache and every Snapper snapshot, so there is no second copy on the disk to fall back to.

Symptom

After a hard power-off, a crash, or pulling an external drive, the boot stops with:


/dev/sda3: UNEXPECTED INCONSISTENCY; RUN fsck MANUALLY.
	(i.e., without -a or -p options)
fsck failed with exit status 4.
[FAILED] Failed to start File System Check on /dev/disk/by-uuid/1a2b3c4d-....
[DEPEND] Dependency failed for /data.
You are in emergency mode. After logging in, type "journalctl -xb" to view
system logs, "systemctl reboot" to reboot, "systemctl default" or "exit"
to boot into default mode.
Give root password for maintenance (or press Control-D to continue):

On plain Arch with an ext4 root the same UNEXPECTED INCONSISTENCY line can name the root partition and come from the initramfs instead, followed by a FILESYSTEM CHECK FAILED banner and a [rootfs ]# shell that reboots the machine when you leave it.

On Omarchy 4 the root, /home, /var/log and /var/cache/pacman/pkg are btrfs subvolumes that are never checked, so this message only ever names an extra ext4 or XFS filesystem you added to /etc/fstab yourself. If the root account is locked, the maintenance prompt cannot be passed and Ctrl-D brings it straight back.

Cause

fsck found damage it will not repair without confirmation (usually a dirty journal plus orphaned inodes on ext4) and exits with status 4, "errors left uncorrected". On Arch the check runs from one of two places: the mkinitcpio fsck hook checks the root filesystem before it is mounted, and systemd-fsck@.service checks every other /etc/fstab entry with a non-zero pass number in the 6th field. systemd-fsck@ fails on status 4, the mount unit that Requires it fails, and unless the entry carries nofail, local-fs.target activates emergency.target.

On Omarchy 4 both paths are effectively off for the system's own filesystems. /etc/mkinitcpio.conf.d/omarchy_hooks.conf keeps fsck in HOOKS, but with autodetect the hook packs only fsck.btrfs, a shell script from btrfs-progs that prints a hint and exits 0 without touching the disk. Every btrfs line the installer writes to /etc/fstab (/, /home, /var/log, /var/cache/pacman/pkg, all subvol=/@...) has pass 0. The only entry with a pass number is the vfat ESP at /boot (pass 2), which systemd-fsck@ checks at every boot with fsck.fat -a, and fsck.fat only exits 0, 1 or 2, so it does not produce status 4. A broken ESP therefore surfaces as a mount failure (see fstab-bad-entry-emergency-mode), and an UNEXPECTED INCONSISTENCY on an Omarchy machine points at an ext4 or XFS filesystem you added to fstab.

Fix

**Getting past the maintenance prompt.** It asks for the root password. On Omarchy 4 that depends on how the machine was installed: the current ISO's configurator gives root the same password as the user you created, and a deferred-provisioning install sets root to the owner's password at first boot, but an early 4.0.0 install can have root locked. Try your own login password first. If root is locked the prompt cannot be passed, and you work from the ISO instead (the Omarchy ISO gives a root shell with no password on Ctrl+Alt+F2, the Arch ISO on tty1).

**If the failing filesystem is not root** (an extra /data, or /home on plain Arch), repair it from the maintenance shell:


findmnt --verify --fstab
umount /dev/sda3          # must NOT be mounted
fsck -f /dev/sda3         # answer y, or use -y to accept everything
systemctl default

**If it is an ext4 or XFS root** (plain Arch), do not fsck it from that shell, because root is mounted. Boot the Arch or Omarchy ISO and work on it unmounted:


lsblk -f
# open LUKS first if encrypted (any mapper name works):
cryptsetup open /dev/nvme0n1p2 root

# ext4:
e2fsck -f /dev/mapper/root          # add -y to auto-answer once you have read the prompts
# xfs:
xfs_repair /dev/mapper/root
# vfat ESP (Omarchy: /dev/nvme0n1p1, normally mounted at /boot):
fsck.fat -a /dev/nvme0n1p1

**If the root is btrfs** (every Omarchy 4 install, and archinstall's btrfs layout), there is nothing to fsck: fsck.btrfs is a no-op by design, and a normal mount replays the log and fixes almost everything. If the mount itself fails, from the ISO:


cryptsetup open /dev/nvme0n1p2 root
mount -o ro,rescue=usebackuproot,subvol=@ /dev/mapper/root /mnt   # try the backup tree roots
btrfs check --readonly /dev/mapper/root                            # unmounted, read-only, report only

Once it mounts, btrfs scrub start / followed by btrfs scrub status / is the safe integrity check, and it runs on the mounted filesystem. btrfs check --repair stays the last resort, after a backup or a ddrescue image.

**To get in once without checking** (to take a backup first), add to the kernel command line:


fsck.mode=skip

**To force a full check on the next boot** once you are back in:


fsck.mode=force fsck.repair=yes

On plain Arch, edit the entry at the bootloader menu. On Omarchy 4 the command line is embedded in the UKI at /boot/EFI/Linux/omarchy_linux.efi, so put it in a drop-in and rebuild the UKI, as root, from a chroot if the machine does not boot:


echo 'KERNEL_CMDLINE[default]+=" fsck.mode=skip"' > /etc/limine-entry-tool.d/50-fsck-skip.conf
limine-mkinitcpio

Delete the drop-in and run limine-mkinitcpio again afterwards. Editing the line at the Limine menu only takes effect with Secure Boot off, because systemd-stub ignores a passed command line when Secure Boot is on and the UKI carries its own. Neither parameter does much on Omarchy, where nothing but the ESP is checked.

Then make sure fstab is sane, because a wrong pass number is a frequent cause of a check that never should have run. The 6th column is the pass number: 1 for an ext4 or XFS root only, 2 for other checkable filesystems, 0 to skip. btrfs, anything non-Linux, and any network or removable mount must be 0:


UUID=974cda7f-...  /      btrfs  rw,relatime,compress=zstd:3,ssd,space_cache=v2,subvol=/@      0 0
UUID=16AE-BBB4     /boot  vfat   rw,relatime,fmask=0077,dmask=0077,utf8,errors=remount-ro       0 2
UUID=...           /data  ext4   defaults,nofail,x-systemd.device-timeout=5s                    0 2
UUID=...           /win   ntfs3  defaults,nofail,x-systemd.device-timeout=5s                    0 0

The first two lines are what the Omarchy installer writes. On plain Arch with an ext4 root the root line is /dev/nvme0n1p2 / ext4 defaults 0 1.

If you use the mkinitcpio fsck hook (the Arch and Omarchy default), the kernel command line must contain rw, not ro, or the check cannot run. Omarchy's cmdline already has rw.

Repeated inconsistencies mean failing hardware, not a software bug:


sudo smartctl -a /dev/nvme0n1
sudo dmesg | grep -iE 'i/o error|medium error|nvme.*reset'

Verify

e2fsck -f (or xfs_repair) exits 0 or 1 on a second run with no further corrections. systemctl default completes. After reboot systemctl list-units --failed is empty and journalctl -b -u 'systemd-fsck@*' shows clean checks.

Sources

fsck-unexpected-inconsistency-maintenance-shell