On Wed, Apr 25, 2018 at 10:57:27PM -0700, Christoph Hellwig wrote: > On Wed, Apr 25, 2018 at 04:46:22PM -0700, Darrick J. Wong wrote: > > (I mean, we /could/ just treat the swapfile as an unbreakable rdma/dax > > style lease, but ugh...) > > That is what I think it should be long term, instead of a strange > parallel I/O path. > > But in the mean time we have a real problem with supporting swap files, > so we should merge the approaches from you and Aleksei and get something > in ASAP. I'm planning to do something along these lines for Btrfs, as well (have swap_activate add the swap extents itself), because the previous thing I tried with going through ->read_iter() and ->write_iter() ran into too many locking issues (i.e., GFP_NOFS can suddenly go through FS locks).
