[PATCH 0/7] inode allocator refactoring V2

[Christoph Hellwig]

This series turns the higher level inode allocator upside down.

The biggest change is that we try to operate on the incore perag
structure as much as possible instead of reading the AGI buffer.

I don't have a system to measure the benefit on the large create benchmarks
right now, but even if it's not benefitial it at least greatly cleans up
the code.

Changes since V1:
  - minor cleanups noted by Dave

Note that this does not collapse the three passes in xfs_dialloc yet -
I tried it and got deadlocks that I haven't fully understood yet.  I plan
to look into them when I get a bit more time.