On Fri, Jul 18, 2025 at 09:50:24AM -0700, Eric Biggers wrote: > But, as suggested at > https://lore.kernel.org/linux-block/20250717044342.GA26995@xxxxxx/ it > should also be okay to reorganize things so that the regular > submit_bio() does not support the fallback, and upper layers have to > call a different function blk_crypto_fallback_submit_bio() if they want > the fallback. I don't think that would help with the splitting issue > directly, It actually does. Splitting before submit_bio will automatically get the ordering right. It is something done by a lot of file systems already and trivially done. > but perhaps we could make the filesystems just not submit bios > that would need splitting by blk-crypto-fallback, which would solve the > issue. I think that distinction is a bit fuzzy. When stacking the blk-crypto fallback helpers above submit_bio, the split is formally done by the file systems, just using generic library helpers. As modern file systems usually build bios to the full possibly size and then split them based on limits I suspect that scheme is also the best here, but there might be exceptions where just looking at the limits and not building the bio bigger might be suitable, but I suspect they'd be the unusual case.
