On Sat, Oct 05, 2024 at 03:20:06AM +0100, Matthew Wilcox wrote: > On Fri, Oct 04, 2024 at 04:04:29PM -0400, Goldwyn Rodrigues wrote: > > iomap_read_folio_ops provide additional functions to allocate or submit > > the bio. Filesystems such as btrfs have additional operations with bios > > such as verifying data checksums. Creating a bio submission hook allows > > the filesystem to process and verify the bio. > > But surely you're going to need something similar for writeback too? > So why go to all this trouble to add a new kind of ops instead of making > it part of iomap_ops or iomap_folio_ops? We really should not add anything to iomap_ops. That's just the iteration that should not even know about pages. In fact I hope to eventuall get back and replace it with a single iterator that could use direct calls for the fast path as per your RFC from years ago. iomap_folio_ops is entirely specific to the buffered write path. I'm honestly not sure what the point is of merging structures specific to buffered read, buffered write (and suggested later in the thread buffered writeback) when they have very little overlap.
