This is a rebase of the reviewed series of kernel patches I recently
It addresses all the review comments except for one from Christoph
about the IOP_SIZE removal - I'll send a followup patch in a couple
of days to remove all the rest of IOP_ macros...
The log space problems from the optimisations have been fixed - the
fix went into patch 44 of this series, and cascaded through 45 and
46 as well. Those patches will need reviewing. The problematic
workload has been running for 16 hours here without failure on a
small (10MB) log, so AFAICT the problem is fixed.
There is one new patch - the fix to the use-after-free of clean
buffer items when a transaction is cancelled - and that is at the
end of the series.