On Wed, Aug 06, 2025 at 04:53:34PM +0200, Christoph Hellwig wrote: > On Tue, Aug 05, 2025 at 04:52:11PM -0600, Keith Busch wrote: > > On Tue, Aug 05, 2025 at 12:56:07PM -0700, Keith Busch wrote: > > > + bv = next->bio->bi_io_vec[0]; > > > + bvprv = req->biotail->bi_io_vec[req->biotail->bi_vcnt - 1]; > > > > Hm, commit 7bcd79ac50d9d8 suggests I missed something obvious about not > > being allowed to use 'bi_vcnt - 1' for the tail bvec of the previous > > bio. I guess it's some stacking thing. > > It is bio cloning and splitting. bi_io_vec is the original payload added > to a bio, and directly accessing it is only allowed for the submitter, > i.e. usually the file system. The I/O stack always need to go through > the bio_iter to deal with split/cloned/truncated/partially completed and > resubmitted bios. Yeah, I knew about the splitting, but I am only using this for merging, and bio's split off the front can't merge, so thought I was okay. But the cloning, yes, I messed that up. v2 is sent with a fix for it. Sorry for the rapid fire versions; I realized you were travelling, so I didn't expect more feedback on v1. But thank you for finding some time to review anyway.
