#2 spfs/npfs: problem with stale mounts
Description From Eric Van Hensbergen 2008-05-06 18:57:45 (-) [reply]
I've seen some spurious bad activity reported as "transport endpoint not
connected" -- it seems to revolve around being in a directory that has been
removed from the server. The problem is that the mount ends up in what appears
to be a bad state (chdir'ing out doesn't fix anything and I also see a 'ghost'
duplicate directory in the hierarchy (ie. /mnt/home/ericvh/ericvh).
I believe this has to do with a bad interaction with the dcache when an element
goes stale on the server. I will post more information as I find a reliable
way to reproduce the problem.
Comment #1 From Eric Van Hensbergen 2008-05-14 19:52:22 (-) [reply]
This is specifically caused by diread - it was misdiagnosed before. It happens
with a sub-filesystem mount point isn't behaving correctly (in this case a
gnome-vfs fuse mount I believe). spfs gets an error while trying to stat the
mount point and then that breaks all of the dirread.
The question is how to deal with it. We can simply ignore the mount point,
report bad stat information, or try to get at the underlying stat information.