Socket operation on non-socket/kernel: VFS: Busy inodes after unmount

[C. Grierson]

Hello all,

Jan 26 08:43:06 vagal ssh-agent[15167]: error: accept from AUTH_SOCKET: Socket operation on non-socket
Jan 26 08:43:36 vagal last message repeated 735042 times
Jan 26 08:44:37 vagal last message repeated 810594 times
Jan 26 08:45:38 vagal last message repeated 928914 times
Jan 26 08:46:39 vagal last message repeated 992443 times
Jan 26 08:47:41 vagal last message repeated 961178 times
Jan 26 08:48:17 vagal last message repeated 510311 times
Jan 26 08:48:17 vagal kernel: VFS: Busy inodes after unmount. Self-destruct in 5 seconds.  Have a nice day...
Jan 26 08:48:17 vagal ssh-agent[15167]: error: accept from AUTH_SOCKET: Socket operation on non-socket
Jan 26 08:48:47 vagal last message repeated 291483 times
Jan 26 08:49:47 vagal last message repeated 591931 times
Jan 26 08:49:48 vagal kernel: VFS: Busy inodes after unmount. Self-destruct in 5 seconds.  Have a nice day...
Jan 26 08:49:47 vagal ssh-agent[15167]: error: accept from AUTH_SOCKET: Socket operation on non-socket
Jan 26 08:50:18 vagal last message repeated 292510 times
Jan 26 08:51:19 vagal last message repeated 552855 times
Jan 26 08:52:20 vagal last message repeated 1529338 times
...(crash, reboot)...

There was reference to a similar problem with respect to the ssh-agent
issue at least in the fedora-selinux-list that suggested a fix that
involved changing some macros (compilation macros? couldn't tell -- not
familiar enough, and didn't seem to have the mentioned file
"ssh_agent_macros.te") at:

https://www.redhat.com/archives/fedora-selinux-list/2004-March/msg00219.html
("[policy-1.9-11] ssh-agent takes all the CPU in enforcing mode")

There is apparently an outstanding kernel bug (#124600,
https://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=124600 ) related
to NFS mounts being unmounted prematurely that is often triggered by
autofs timing out mounts and seems to me likely to be the cause of my
woes here.

Is there anyone more familiar with this problem that can confirm or
at least agree that this bug is most likely at the core here?

Anyone know what the heck these ssh-agent socket errors are from?

Thanks everyone,

-c

------------------------------------------------------------------------