On Thu, Dec 12, 2013 at 08:36:29AM -0800, Christoph Hellwig wrote: > I really don't like how this makes even more of a mess out of the > already convoluted xfs_bioerror/xfs_bioerror_else maze. Can we > maybe first merge them and document the difference before adding > even more special case branches? > > Also most uses of uncached buffers use xfsbdstrat, where we can do > error handling straight in the caller instead of playing with all > the flags manipulation mess. In all these cases no one but the > caller can find these buffers anyway, so doing all this on an > I/O error is pointless. > > The only buffer where any of this matters is the superblock one, > and given that we re-read it on mount anyway I wonder if we should > just make it a regular buffer again and let all this mess just > disappear. Ok, I agree it is a bit messy, but that code is already pretty ugly. I'd like to get this fix in first, because it's causing oopses in roughly 30% of my local xfstests runs on a couple of VMs, so I'd prefer to get the fix out there now and do the cleanup as a separate patch series. Would that be an acceptible approach to take here from your perspective? Cheers, Dave. -- Dave Chinner david@xxxxxxxxxxxxx _______________________________________________ xfs mailing list xfs@xxxxxxxxxxx http://oss.sgi.com/mailman/listinfo/xfs