On Wed, Jun 14, 2017 at 4:05 PM, Darrick J. Wong <darrick.wong@xxxxxxxxxx> wrote: > Why not just spill the values into their own ea_inodes if we need the > space? I guess that has the disadvantage that now we need to reserve > quite a few more journal credits ((1 inode block, 1 bbitmap block, 1 > ibitmap block, 1 data block) * nr_inline_values) just in case we end up > spilling all the values. > In last week's ext4 conference call we have discussed this problem and possible solutions. The major concern with spilling/evacuation approach was the added complexity. We haven't discussed the impact on journal credits, but that is a valid point too.
