Re: [PATCH 4/4] xfs: allocate direct I/O COW blocks in iomap_begin

[Christoph Hellwig <hch@lst.de>]

On Wed, Feb 08, 2017 at 11:53:22PM -0800, Darrick J. Wong wrote:
> Testing isn't done yet, but xfs/222 seems to be blowing up at
> ASSERT(!rwsem_is_locked(&inode->i_rwsem)) in xfs_super.c fairly
> consistently with blocksize=1k.  I haven't been able to reproduce it
> quickly (i.e. without running the whole test suite) so I can't tell if
> that's a side effect of something else blowing up or what.  generic/300
> seems to blow up periodically and then blows the same assert on umount,
> also in the 1k case.  xfs/348 fuzzes the fs, causes "kernel memory
> exposure!" BUGs and then asserts with the same i_rwsem thing.

I'll take a look at the umount assert while you're asleep.  348 is
a pretty new test, so I doubt it's a regrewssion.

> (You'll note I didn't merge the duplicate "xfs: improve handling of busy
> extents in the low-level allocator"; if you want that done, please let me
> know.)

Yes, it should be folded into the first patch of that name and descriptions.
It contains the fixups that Brian requested.