On Tue, Jan 04, 2011 at 04:43:36AM -0500, Christoph Hellwig wrote: > I don't think the patch is quite correct. In the old code xfs_buf_rele > incremented the buffer reference count before calling ->b_relse, > expecting it do decrement it again. > > I think the best fix is to kill ->b_relse entirely. We can simply > do the buffer callback processing and b_flags updates in > xfs_buf_iodone_callbacks. The important thing is to not clear the > buffer error there, so that it actually get propagated to the caller. > As the buffer remains locked until xfs_bwrite calls xfs_buf_relse it > can get the error reliably that way. > > Patch below, but it's still running xfqa so far: Looks sane. can i get a signed-off-by for it from you and I'll include it in my remaining -for-2.6.38 series? Cheers, Dave. -- Dave Chinner david@xxxxxxxxxxxxx _______________________________________________ xfs mailing list xfs@xxxxxxxxxxx http://oss.sgi.com/mailman/listinfo/xfs
