On Tue, Jul 09, 2019 at 11:12:14AM -0700, Christoph Hellwig wrote: > I looked over it and while some of the small files seem very tiny > they are reasonably split. > > What rather annoys me is the page.c/read.c/write.c split. All these > really belong mostly together, except maybe the super highlevel > write code that then either calls into the buffer_head vs iomap_page > based code. By keeping them together we can eliminate most of > iomap_internal.h and once the writeback code moves also keep > iomap_page private to that bigger read.c file. <nod> I think it makes sense to combine them into a single read_write.c file or something. > > A few other minor notes: > > - I think iomap_sector() should move to linux/iomap.h as an inline > helper. So long as you're ok with including blkdev.h from iomap.h. :) > - iomap_actor_t / iomap_apply should probaby just move to linux/iomap.h > as well, which would avoid needing the awkward subdir include in > dax.c I'd also like to fix the opencoded iomap_apply() in fs/dax.c, but cleaning all that up looks stressful enough to defer for 5.4. > - some of the copyrights for the small files seem totally wrong. > e.g. all the swapfile code was written by you, so it should not have > my or rh copyright notices on it Will fix the swapfile code. --D
