On Mon, Aug 28, 2023 at 02:30:23PM +0200, Christoph Hellwig wrote:

On Sun, Aug 27, 2023 at 08:41:22PM +0100, Al Viro wrote:

quoted

That part is somewhat fishy - there's a case where you return a positive value
and advance ->ki_pos by more than that amount.  I really wonder if all callers
of ->write_iter() are OK with that.  Consider e.g. this:

This should not exist in the latest version merged by Jens.  Can you
check if you still  see issues in the version in the block tree or
linux-next.

Still does - the problem has migrated into direct_write_fallback(), but
that hadn't changed the situation.  We are still left with ->ki_pos bumped
by generic_perform_write() (evaluated as an argument of direct_write_fallback()
now) and *not* retraced in case when direct_write_fallback() decides to
discard the buffered write result.  Both in -next and in mainline (since
6.5-rc1).