On Fri, Sep 12, 2025 at 01:56:45PM +0200, Andrey Albershteyn wrote: > > > In addition to being bound by the isize the fiemap_read() copies > > > data to the iov_iter, which is not really needed for fsverity. > > > > Aka, you want an O_DIRECT read into a ITER_BVEC buffer for the data? > > > > hmm, but we want fsverity merkle tree to be in page cache to use the > "verified page" flag. Oh, right. There's this thing called caching :) > As far as I understand iter_bvec will need a > page attached anyway, so this is the same. Or am I missing > something? > > And with direct io there's no readahead then, and we don't get any > benefit going through vfs instead of directly calling to iomap. Yeah, strike my idea. So yes, you probably do want a lower level interface that just reads the pages and gives you access to them. > > -- > - Andrey ---end quoted text---
