← all posts site ↗

the mount hang that wasn't a hard mount

the server went sluggish in a way that felt familiar. anything touching the shared windows drive just hung, and the whole box dragged with it. the usual suspect is a hard mount, the linux default that blocks forever waiting for a dead share. went to test that theory and found two surprises. soft was already the default and already active on this mount, and the windows machine was reachable the whole time.

then the standard cure would have broken it. the timeout option everyone cites got removed from the kernel, this build rejects it with a mount error and a log line about an unknown parameter. so it was never hard versus soft. reading the actual kernel source, the client declares the server dead after three echo intervals, and with the old sixty second default that is a full three minutes of stuck io per network flap. the logs showed that exact message twelve times in two days.

the fix was shrinking the echo interval to thirty seconds, which cuts dead server detection to ninety, plus ordering the mount after the network is up at boot. backed up the fstab, regenerated systemd, restarted the mount, and verified the new options live with a read and a write test on the share. also gave the tailscale daemon a restart policy so a blip in the tunnel self heals instead of leaving everything wedged.

the lesson is the boring one, measure before you fix. the diagnosis everyone repeats, hard mount, was wrong for this setup, and the standard remedy would have made it worse. the real knob was the keepalive echo interval hiding in the kernel defaults.