On Tue, May 12, 2020 at 09:03:00AM -0700, Christoph Hellwig wrote: > On Tue, May 12, 2020 at 08:53:20AM -0700, Darrick J. Wong wrote: > > I was gonna say, I think we've messed this up enough that I think we > > just have to accept empty attr leaf blocks. :/ > > > > I also think we should improve the ability to scan for and invalidate > > incore buffers so that we can invalidate and truncate the attr fork > > extents directly from an extent walk loop. It seems a little silly that > > we have to walk the dabtree just to find out where multiblock remote > > attr value structures might be hiding. > > The buffers are indexed by the physical block number. Unless you > want to move to logical indexing that's what we'll need to do. <shrug> I modded xfs_buf_incore and _xfs_buf_obj_cmp to return an xfs_buf that matches map->bm_bn regardless of length and it seems fine so far... --D
