DDM does not recover after unexpected dde-seatd control socket loss

[zccrs]

Summary

DDM does not recover cleanly if the dde-seatd control socket disappears unexpectedly.

DdeSeatdControl disconnects and clears its cached state on socket error, but it does not attempt to reconnect. If dde-seatd dies and comes back, DDM can stay alive while permanently losing VT change notifications.

Current behavior

src/daemon/DdeSeatdControl.cpp uses /run/dde-seatd-control.sock and does this on failure:

• connectEventSocket() fails -> log + give up
• handleEventSocketError() -> disconnectEventSocket()
• disconnectEventSocket() clears the socket and cached VT state

There is no retry loop and no service-level recovery path inside DDM.

Why this matters

I validated the systemd side separately:

• if dde-seatd is explicitly stopped/restarted by systemctl, Requires= is enough for systemd to stop/restart dependent services in the same transaction
• but if the dde-seatd process dies unexpectedly, a plain Requires= dependency does not tear down already-running dependents

That means DDM can remain running while its control socket is gone.

At that point VT change handling is effectively dead until something else restarts DDM.

Reproduction

A minimal systemd reproducer shows the difference clearly:

• Requires=seatd.service + After=seatd.service
• start both services
• kill the seatd main pid directly

Observed result:

• seatd.service becomes inactive
• the dependent service stays active

This matches the socket-loss case in DDM: the daemon can survive while the control channel is gone.

Proposed fixes

I see two complementary options:

1. Make DDM self-heal

   • add reconnect/backoff logic in DdeSeatdControl
   • re-establish the event socket after errorOccurred / disconnected
   • refresh cached active VT after reconnect

2. Make the service dependency stricter

   • upgrade the unit relationship from Requires= to BindsTo= for the dde-seatd.service dependency
   • keep Restart=always on DDM so a dde-seatd death causes DDM to be torn down and brought back together with the seat daemon

I validated the second approach with a minimal systemd setup:

• BindsTo=seatd.service
• Restart=always on both services
• kill the seatd main pid

Observed result:

• seatd is restarted
• the dependent service is also restarted

Files

• src/daemon/DdeSeatdControl.cpp
• services/ddm.service.in