https://bugs.kde.org/show_bug.cgi?id=524948

            Bug ID: 524948
           Summary: Wayland session freezes: kwin_wayland blocks on stat
                    of systemd-automounted NFS/CIFS share when server is
                    unavailable
    Classification: Plasma
           Product: kwin
      Version First 6.7.4
       Reported In:
          Platform: Other
                OS: Linux
            Status: REPORTED
          Severity: major
          Priority: NOR
         Component: xwayland
          Assignee: [email protected]
          Reporter: [email protected]
  Target Milestone: ---

Created attachment 195610
  --> https://bugs.kde.org/attachment.cgi?id=195610&action=edit
kwin_automount_freeze

Hello KDE developers! 

First of all, big respect to all of you for your hard work on Plasma, I really
enjoy using it every day and appreciate what you do. And please forgive me in
advance if my English is not perfect, I am not a native speaker, but I will try
to explain the problem as clearly as I can.

SUMMARY
The entire Plasma Wayland session freezes when a systemd-automounted
(x-systemd.automount) network share (NFS or CIFS) is unavailable:
kwin_wayland repeatedly triggers automount requests by stat'ing the
automount point, and each failed mount blocks the compositor for the
mount timeout duration. The kernel stays alive; the GUI becomes fully
unresponsive and looks like a hard system hang.
To the best of my knowledge this bug has existed for MANY YEARS across
multiple Plasma releases and is still not discovered upstream.

STEPS TO REPRODUCE
1. fstab entry with automount, e.g.:
   192.168.1.87:/export /mnt/share nfs vers=4.2,soft,timeo=20,retrans=3,
   noauto,x-systemd.automount,x-systemd.idle-timeout=1min,nofail,_netdev 0 0
2. Log into a Plasma Wayland session.
3. Power off the share server (or boot the session with the server off).
4. Wait a few seconds while KDE components touch /mnt/share.

OBSERVED RESULT
Automount requests loop; journal shows the triggers come from
kwin_wayland and KIO::WorkerThread. Each
"mnt-share.mount: Mounting timed out" stalls the compositor for ~15 s,
repeating. The whole session freezes (input and rendering dead).

EXPECTED RESULT
The compositor must never block on filesystem stat / automount
triggers. An unavailable network share should at worst produce I/O
errors in file managers, never freeze the entire session.

SOFTWARE/OS VERSIONS
Operating system: CachyOS Linux (Arch-based)
KDE Plasma Version: 6.7.4
KDE Frameworks Version: 6.29.0
Qt Version: 6.11.2
Kernel: 7.1.11-zen1-1-zen (64-bit)
Graphics platform: Wayland
GPU: AMD Radeon RX 6400
CPU: 16 x Intel Xeon E5-2673 v2 @ 3.30 GHz, 16 GiB RAM
Mainboard: HUANANZHI

ADDITIONAL INFORMATION
- Reproduced with both CIFS and NFS automount entries.
- Kernel log is clean (no oops / hung tasks) — the stall is userspace.
- Workaround: remove x-systemd.automount and mount manually — the
  freeze disappears completely.
- Question for developers: which KWin code path stat'ing the automount
  directory causes the synchronous block?
- This bug has been reproducible for years on different hardware and
  Plasma versions; it appears to have gone unnoticed and unfixed
  upstream until now.
- Journal excerpt attached (kwin_automount_freeze.log).

-- 
You are receiving this mail because:
You are watching all bug changes.

Reply via email to