Windows VM launch does nothing after the polkit password (setgid bit on ~/Windows fails the mode check)
Checked the code path against the installed script and the upstream tree, and checked the coreutils behaviour by experiment on this machine. Confirmed here on omarchy 4.0.2-1: line 583 of /usr/bin/omarchy-windows-vm is exactly chmod 0700 -- "/proc/$BASHPID/fd/$storage_fd" "/proc/$BASHPID/fd/$shared_fd", and lines 588 to 590 reject any mode that is not 700. mounted_leaf_matches() starts at line 508 and its test at line 518 is $mode == 700. assert_mounts_safe() at line 839 is called by __priv_up (line 896), __priv_up_wait (line 902) and __priv_remove (line 931), so launch, install and remove all gate on it. up is an accepted privileged action (line 149) and with_vm_lock returns the action's own status (line 131), which makes the record's exit 1 from pkexec ... __priv up consistent with the code. The desktop entry written at line 1170 does carry Terminal=false, and the Windows VM removal stopped before user-side cleanup message is at line 1373. gh api on repos/omacom/omarchy/contents/bin/omarchy-windows-vm?ref=quattro returned a file byte-identical to the installed one, and the v4.0.2 to v4.0.3 compare does not list bin/omarchy-windows-vm, so nothing has been fixed upstream. Issue 9334 is open and supports the cause in full, including the retraction of the original kernel claim, the dockurr/windows emptiness probe in the image's /run/samba.sh, the hide dot files = yes detail, and the inode and bind-anchor warning. PR #9322 is still open with merged_at null and its diff is the chmod u=rwx,go=,a-s change the record describes, so the fix section's claim that nothing has merged holds today, 2026-09-11. Measured in a throwaway /tmp directory here with coreutils 9.11: chmod 0700 and chmod 700 leave 2700 at 2700 and turn 2777 into 2700, chmod 00700 gives 700 from both, and the same preservation happens through the /proc/$$/fd/N indirection the script uses. That experiment is where the record broke. Its fix leads with chmod g-s ~/Windows ~/.windows and tells the reader to expect 700, but g-s on the 2777 that the container leaves behind yields 777, so a user following that check on the common fresh-install case sees 777 and concludes the workaround failed. The launch does still succeed from 777, so the rewrite keeps g-s as a labelled alternative and leads with chmod 00700, which is the upstream issue's own recommendation and is correct from either starting mode. Two minor overreaches left in the symptom rather than rewritten: the thread carries seven confirmations, not six, and only four of them state kernel, coreutils and btrfs together, so "all on" is a little stronger than the source. Intel and AMD are both supported, since the reporter's machine is Intel CoffeeLake-H and christianguenter2 reports an AMD Ryzen AI 5 340. Not exercised: no Windows VM is installed here, ~/.windows and /var/lib/omarchy/windows do not exist, ~/Windows is a symlink to /srv/vms/windows/shared, and with no sudo I could not run pkexec ... __priv up, watch the container set 2777, or read /run/samba.sh inside the image. Those three claims rest on the issue thread alone.
Symptom
Omarchy 4.0.2-1. Apps menu > Windows, or omarchy-windows-vm launch, shows the polkit password dialog, accepts the password, and then nothing happens: no error, no window. Every retry asks for the password again. From a terminal the elevated step exits 1 with no output:
$ stat -c %a ~/Windows
2700
$ pkexec /usr/bin/omarchy-windows-vm __priv up; echo $?
1
omarchy-windows-vm remove fails the same way with Windows VM removal stopped before user-side cleanup, so reinstalling does not help. It hits installs upgraded from 4.0.1 (~/Windows at drwx--S---) and fresh 4.0.2 installs from the second launch onward. Six reporters confirmed it on Intel and AMD machines, all on kernel 7.1.9-arch1-2, coreutils 9.11 and btrfs.
Cause
Established by the reporters and confirmed on this machine. prepare_caller_mounts() in /usr/bin/omarchy-windows-vm (line 583 in omarchy 4.0.2-1) runs chmod 0700 on the pinned ~/.windows and ~/Windows sources and then requires stat -Lc %a to return exactly 700. GNU coreutils chmod deliberately preserves a directory's setuid and setgid bits when given a numeric mode of four or fewer digits, so chmod 0700 leaves 2700 as 2700 and the check returns 1. The same strict mode == 700 test is in mounted_leaf_matches() (line 508), and assert_mounts_safe gates launch, install and remove, so none of them can recover. The dockurr/windows container sets the shared folder (~/Windows on the host) to 2777 when it starts against an **empty** share, which is why the second launch fails on a fresh install. One reporter traced that to the emptiness probe in the image's /run/samba.sh and found a non-empty share is left alone. windows-vm.desktop has Terminal=false, so the failure is invisible. This is coreutils behaviour, not a kernel regression, which the original report first claimed and then retracted.
Fix
No upstream fix has merged as of 2026-09-11. bin/omarchy-windows-vm on the quattro branch is byte-identical to the installed 4.0.2-1 copy and still runs chmod 0700 at line 583, and v4.0.3 (2026-09-08) does not touch the file. PR #9322 (clear the special bits with chmod u=rwx,go=,a-s while keeping the strict check) is open, as are a dozen duplicate PRs for the same line, including #10046 and #10113.
Until one lands, clear the bits in place. chmod never changes the inode, so the root-owned bind anchor under /var/lib/omarchy/windows/mounts/users/<uid>/ stays valid:
chmod 00700 ~/Windows ~/.windows
stat -c '%a %n' ~/Windows ~/.windows # both 700
omarchy-windows-vm launch
Use the five-digit 00700. It clears the setgid bit and sets mode 700 from either starting state, the 2700 of an upgraded install or the 2777 the container leaves behind. Measured on this workstation, omarchy 4.0.2-1 with coreutils 9.11:
$ mkdir -m 2777 /tmp/sg && chmod 0700 /tmp/sg && stat -c %a /tmp/sg
2700
$ chmod 00700 /tmp/sg && stat -c %a /tmp/sg
700
chmod g-s ~/Windows ~/.windows also brings the next launch straight up and four reporters confirmed it, but do not expect mode 700 from it. It clears only the setgid bit, so a share the container left at 2777 becomes 777, and the script's own chmod 0700 is what takes it to 700 on the next launch. chmod 0700 and chmod 700 clear nothing at all.
The bit comes back whenever the container starts against an empty ~/Windows, so either repeat the chmod 00700 before each launch, or leave any file in the folder. A dotfile stays hidden from the guest because the generated smb.conf keeps Samba's default hide dot files = yes:
touch ~/Windows/.keep
Do not move or recreate ~/Windows. That changes the inode, and the next launch then fails with protected mount ... no longer matches its home source until the anchor is unmounted with pkexec umount /var/lib/omarchy/windows/mounts/users/$(id -u)/shared.
Verify
stat -c '%a %n' ~/Windows ~/.windows # 700 and 700 after chmod 00700, no leading 2
pkexec /usr/bin/omarchy-windows-vm __priv up; echo $? # 0 and the container comes up
If you used chmod g-s instead, expect 777 on a share the container had left at 2777. That is not a failure. The launch still succeeds, because with no special bit in the way the script's own chmod 0700 clears group and other access.