This series tries to reduce the amount we hold the ilock exclusively,
especially during direct I/O writes where they currently hurt us.
Dave showed that his earlier version which is less aggressive than this
one can already provide magnitudes of better throughput and iops for
parallel direct I/O workloads, and this one should be even better.
Changes from V1:
- do not mark xfs_qm_need_dqattach as inline
- various comment and commit message updates
I specificly did not add the delalloc assert in the direct I/O write path
in this series, as it triggers with or without this patch. I will look into
that issue next.