On Fri, Feb 23, 2024 at 09:20:28AM -0800, Darrick J. Wong wrote: > > + fsc->frextents -= > > + xfs_rtb_to_rtx(mp, percpu_counter_sum(&mp->m_delalloc_rtblks)); > > Should m_delalloc_rtblks be measured in rt extents (and not fsblocks) > since a sub-rtx delalloc doesn't make any sense? As written, here we'd > also have to signal a corruption if (m_delalloc_rtblks % rextsize > 0). Either way should be fine. I did it this way around so that we only have to do the conversion in the slow path, but I can change it around.