Omarchy installer dies at 'mounting the ESP failed (exit 32)' with a SQUASHFS superblock error
Checked against the quattro branch of omacom/omarchy-iso fetched today, and against all four cited threads read in full with comments. Neither this record nor its fix can be exercised on this workstation: it is an installer defect on a live ISO, I have no sudo, no ISO was built and no VM was used, so everything below is source reading and thread reading, not a run.
What held. The central claim is confirmed in the source: the free-space path always creates its own ESP and never adopts an existing one, stated in the comment at configurator:531 to :536, with detect_windows_esp's only consumer a say line at :539, and the shared-ESP path deleted outright in commit f79578f2 on 2026-07-18. All three line numbers the record cites still match today: the comment at :531, create_partition for the EFI partition at :704, and the untyped mount -o ro in detect_windows_esp at :396. The bare ESP mount is still present at :773, the file's last commit is still 2673c613 (2026-09-01), and omacom/omarchy-iso#111 is still open, so an ISO built from quattro still carries the defect. The fix's rerun path is upstream's own: abort() at configurator:178 to :183 prints You can retry later by running: ./.automated_script.sh. The counts hold: three confirmations of the patch sequence in #7263, and the second route reported twice in #7515, once with a 256M partition and once with a 2 MB one, with one reporter deleting it afterwards and still booting.
What was wrong. The cause's mechanism is wrong, and it was the one thing I was asked to test hard. It claims the probe runs before udev has re-read the new partition, or reads a leftover signature. Upstream refutes both on the pull request: create_partition plus configurator:718 to :722 already run partprobe, udevadm settle, sync, sleep 2 and a ten second wait_for_device, and wipefs -af at :728 (added in 7d3b01e, 2026-08-13) plus mkfs.fat at :772 each open and write the device successfully before the mount. udevadm settle and sleep 2 were in the PR's first revision and were removed from the head daa0cc3 after review, so the record's fix reproduces lines upstream deliberately dropped and justifies them with a mechanism upstream rejects. The real mechanism, from mount(8) as quoted by the reviewer, is that an untyped mount whose libblkid probe comes back empty or ambiguous falls through every non-nodev type in /proc/filesystems, where squashfs is present on the live ISO and vfat is not, so the error names the last thing tried. Why the probe fails is explicitly not pinned down upstream, and I have said so rather than substituting a new guess. Two unverified specifics in the symptom also had to go: #7515 says "empty linux fs partition", not ext4, and "including with the free space on the Windows disk" is contradicted by #7263's own body, whose reporter says installing onto the Windows disk is what worked. I rewrote the danger: its first sentence mis-described the rollback, and #7867, which it already cited, carries two concrete hazards it did not pass on, one of them a command (omarchy-refresh-limine) that silently removes Windows from the boot menu after a successful install.
One addition is labelled as inference and not as fact. detect_windows_esp mounts any FAT partition found on the target disk, which autoloads vfat into /proc/filesystems before the ESP mount runs, and every failing report in both threads is a target disk with no FAT partition on it. That explains the otherwise unexplained second route and why it must be on the same disk. I marked it "not confirmed upstream" in the record because upstream says the trigger is still open.
You are repartitioning a disk that holds data you want to keep. Confirm the target disk and partition on the installer's summary screen before accepting: on a two-disk machine the free space and the Windows ESP are on different drives, and picking the wrong one destroys Windows. The aborted run removes the partitions it created before handing the disk back (disk_abort_hook at configurator:427, which unmounts the target, closes LUKS and reclaims those partitions), so a rerun creates them again in the free space rather than reusing anything left over. The second route has you partition that same disk by hand, so back the data up first and create the new partition inside the free space only.
Two things afterwards on a dual-boot machine, both from omacom/omarchy#7867. This free-space path can leave a 2 GB ESP the firmware does not surface in its top-level boot list, so the machine boots straight to Windows, and recovery there took a chroot, a Limine reinstall, and repointing the /boot mount in /etc/fstab at the real Windows ESP. The same issue reports that omarchy-refresh-limine drops any Windows entry from /boot/limine.conf on every run and never restores it, because the template it copies in carries no OS entries and nothing downstream rescans for bootmgfw.efi, so do not run it on a dual-boot machine until that is fixed.
Symptom
Installing Omarchy 4 from the ISO with the free space option, either alongside Windows or alongside plain data partitions. The installer creates the partitions, the LUKS container where asked for, the btrfs filesystem and its subvolumes, then stops at the last disk step:
mount: /mnt/boot: fsconfig() failed: Can't find a SQUASHFS superblock on nvme0n1p3.
dmesg(1) may have more information after failed mount system call.
mounting the ESP failed (exit 32)
The device named in the message is the EFI partition the installer just created, so its number varies with the disk: nvme0n1p3 on a dual-boot machine, sda2 on a disk holding one data partition. The mount target is /mnt/boot on an encrypted install and /mnt/efi on an unencrypted one (configurator:660 and :662). The aborted run removes the partitions it created before handing the disk back, so nothing is written to your existing partitions.
Two threads report it. In the first the free space was on a second NVMe drive with Windows on the first, and five other people in that thread report the identical message or confirm the workaround. In the second the target disk held a data partition and unallocated space but no EFI System Partition at all, reproduced by the reporter on a ten year old laptop and on a PC, once with an NTFS data partition and once with an empty Linux filesystem, and confirmed by other reporters on a MacBook Pro 16 (T2) with NTFS and exFAT partitions, on an MSI laptop, and on an Acer Nitro AN515-55 with a data-only NTFS drive.
One thing is common to every report in both threads: the target disk carried no FAT partition before the install. See the cause.
Cause
Named in the pull request that fixes it, omacom/omarchy-iso#111, still open and unmerged on 2026-09-11. Two facts about the installer combine.
First, the free-space path always creates its own dedicated EFI partition and never adopts an existing one. The comment at configurator:531 says exactly that, and the only consumer of detect_windows_esp is a say line at :539 that prints what it found. So the device named in the SQUASHFS error is always the partition the installer just made, and a thread that reads as a missing-ESP problem is the same defect.
Second, that new partition is formatted and then mounted with no filesystem type:
disk_step "creating the ESP filesystem on $efi_dev" mkfs.fat -F32 -n OMARCHY_EFI "$efi_dev"
disk_step "mounting the ESP" mount "$efi_dev" /mnt"$esp_mount_in_target"
Those are configurator:772 and :773 on the quattro branch of omacom/omarchy-iso, in the file last changed by commit 2673c613, so every ISO built from that branch carries the defect. detect_windows_esp has the same untyped mount -o ro at :396.
The SQUASHFS text is not a claim about what is on the partition. mount(8) with no -t asks libblkid for the type, and when libblkid comes back empty or ambiguous it falls back to trying every non-nodev filesystem listed in /etc/filesystems or /proc/filesystems. On the live ISO squashfs is in that list, because the airootfs is a squashfs image, and vfat usually is not, because nothing mounts a FAT volume during a normal live boot. The error therefore names the last type tried, not the thing that failed. Naming the type removes the guess and also lets the kernel autoload the driver, which the untyped fallback cannot do.
It is not a timing race, despite the pull request's title, and it is not a leftover signature. create_partition already runs partprobe and udevadm settle, configurator:718 to :720 runs partprobe, sync and sleep 2, wait_for_device at :722 polls for the device node and aborts the install if it never appears, and wipefs -af at :728 and mkfs.fat at :772 each open that device and write to it successfully before the mount runs. The ESP wipe arrived in commit 7d3b01e on 2026-08-13, and a reporter hit the failure on an ISO dated 2026-08-14 with the FAT boot sector dd-verified as correctly written at the partition start. Upstream states plainly that why the libblkid probe comes back empty or ambiguous is still not pinned down: two attempts to reproduce the ambiguity on a loop device failed.
One likely explanation for the disk layouts involved, consistent with every report in both threads but not confirmed upstream. detect_windows_esp at :391 loops over blkid -t TYPE=vfat -o device, skips any candidate that is not on the target disk, and mounts each one that is left. That mount hands a libblkid-identified vfat to mount(2), which autoloads the vfat module and puts it in /proc/filesystems. A target disk that already carries a FAT partition therefore has vfat in the fallback list by the time :773 runs, and a target disk without one does not. Every failing report is a target disk with no FAT partition on it, including the case where Windows and its ESP sat on a second drive and that same reporter then installed successfully onto the Windows disk itself. It also explains why creating a small FAT partition by hand works as a workaround.
Fix
One change fixes this: name the filesystem type on the ESP mount. Patch the installer script inside the live environment and rerun it. The aborted run already rolled back the partitions it created, and the edit is lost on reboot, so do it in the same live session that failed.
- The installer prints
You can retry later by running: ./.automated_script.shand exits to a shell in/rooton the live ISO. Open the script:
ls -la
nano configurator
Which editors the ISO ships was not confirmed. If nano is absent, try vim, or use the sed form at the end of this fix.
- Find the block that formats and mounts the ESP (search for
mounting the ESP):
disk_step "creating the ESP filesystem on $efi_dev" mkfs.fat -F32 -n OMARCHY_EFI "$efi_dev"
disk_step "mounting the ESP" mount "$efi_dev" /mnt"$esp_mount_in_target"
Change the second line to:
disk_step "mounting the ESP" mount -t vfat "$efi_dev" /mnt"$esp_mount_in_target"
That is the whole fix. The widely copied version of this workaround also inserts udevadm settle and sleep 2 above the mount. They are harmless but they are not what repairs it, and upstream removed both from the pull request after review, because several settles and a ten second device poll already run before this point. Naming the type is the change that alters the outcome.
- Optional, and only if Windows is on the target disk. In
detect_windows_esp, change:
if mount -o ro "$p" "$tmp_mp" 2>/dev/null; then
to:
if mount -t vfat -o ro "$p" "$tmp_mp" 2>/dev/null; then
The loop already filters on blkid -t TYPE=vfat, so this changes no outcome on the success path. It makes an unreadable candidate fail as a FAT mount instead of falling through the same filesystem list.
- Save, then start the installer again:
umount -R /mnt 2>/dev/null
./.automated_script.sh
Answer the installer's questions again exactly as before. Three people in omacom/omarchy#7263 confirmed this sequence completes the install, and four more confirmed it on omacom/omarchy-iso#111, one of whom changed only the ESP mount line and nothing else, and one of whom built an ISO from the branch.
Without an editor, the same two edits:
sed -i 's|mount "$efi_dev" /mnt"$esp_mount_in_target"|mount -t vfat "$efi_dev" /mnt"$esp_mount_in_target"|' configurator
sed -i 's|mount -o ro "$p" "$tmp_mp"|mount -t vfat -o ro "$p" "$tmp_mp"|' configurator
grep -n 'mount -t vfat' configurator # two hits
bash -n configurator # still parses
A second route was reported to work twice. On a disk with data partitions and no EFI System Partition, two reporters created a small FAT32 partition themselves before running the installer, and the free-space install then completed:
- Press Ctrl+C at the installer to reach a shell, or open a second console.
cfdisk /dev/sdXon the target disk. Create a 256M partition inside the free space, set its type toEFI System, write, quit.- Format it:
mkfs.fat -F32 /dev/sdXN - Relaunch with
./.automated_script.shand pick the free space option again.
The second reporter used a 2 MB FAT partition and rebooted into the installer first. The partition has to be on the disk you are installing to, because detect_windows_esp skips candidates that are not. It ends up unused either way: the installer still creates and uses its own EFI partition. Prefer the typed mount above, which is the change upstream is merging. One reporter deleted the spare partition afterwards and Omarchy still booted, but that was one machine, so check findmnt /boot points at the installer's own ESP before removing anything.
Verify
The installer passes mounting the ESP and continues to package installation. After the first boot into Omarchy:
findmnt /boot # FSTYPE vfat, SOURCE the new EFI partition
lsblk -f | grep OMARCHY_EFI
lsblk -o NAME,SIZE,FSTYPE,PARTTYPENAME /dev/sdX # the data partition untouched
Sources
- https://github.com/omacom/omarchy/issues/7263
- https://github.com/omacom/omarchy/issues/7515
- https://github.com/omacom/omarchy/issues/7867
- https://github.com/omacom/omarchy-iso/pull/111
- https://github.com/omacom/omarchy-iso/blob/2673c613d9a71e23920e43fbb951238145e0f1e8/configs/airootfs/root/configurator
- https://github.com/omacom/omarchy-iso/blob/quattro/configs/airootfs/root/configurator
- https://github.com/omacom/omarchy-iso/blob/quattro/configs/airootfs/root/.automated_script.sh