Arla 0.7.2 returns stale cached metadata on Linux 2.0

[Aaron M. Ucko]

<amu at MIT.EDU> (Aaron M. Ucko) writes:

> 0.8's xfs.o gives me some new GPFs and many bogus ENODEVs.  I'll let
> you know about the cache issues once I get the new problems resolved.

I don't think the new problems are Intel-specific.  In particular, the
GPFs I saw and the one Brandon Allbery reported seem to have occurred
when something called free_xfs_node on a node whose XFS superblock
pointer was NULL, which presumably shouldn't have happened.  I haven't
yet determined why it did....

-- 
Aaron M. Ucko, KB1CJC <amu at mit.edu> (finger amu at monk.mit.edu)