On Thu, Mar 18, 2021 at 12:40:25PM -0600, Jens Axboe wrote: > I'm not at all interested > in having a v3 down the line as well. And I'd need to be able to do this > seamlessly, both from an application point of view, and a performance > point of view (no stupid conversions inline). At this point I've now traced the history of effort of wanting to do io-uring "ioctl" work through 3 sepearate independent efforts: 2019-12-14: Pavel Begunkov - https://lore.kernel.org/all/f77ac379ddb6a67c3ac6a9dc54430142ead07c6f.1576336565.git.asml.silence@xxxxxxxxx/ 2020-11-02: Hao Xu - https://lore.kernel.org/all/1604303041-184595-1-git-send-email-haoxu@xxxxxxxxxxxxxxxxx/ 2021-01-27: Kanchan Joshi - https://lore.kernel.org/linux-nvme/20210127150029.13766-1-joshi.k@xxxxxxxxxxx/#r So clearly there is interest in this moving forward. On the same day as Joshi's post you posted your file_operations based implemenation. So that's 2 years, 2 months to this day since Pavel's first patchset... Wouldn't we be a bit too much of a burden to ensure a v2 will suffice for *all* use cases? If so, adaptability for evolution sounds like a more fitting use case for design here. That way we reduce our requirements and allow for experimentation, while enabling improvements on future design. Luis
