On 12/14/20 8:06 PM, Linus Torvalds wrote: > On Mon, Dec 14, 2020 at 11:13 AM Jens Axboe <axboe@xxxxxxxxx> wrote: >> >> I'm pretty happy with this at this point. The core change is very simple, >> and the users end up being trivial too. > > It does look very simple. Agree, hard to imagine it being much simpler than this. > It strikes me that io_statx would be another user of this. But it > obviously depends on what the actual io_uring users do.. Indeed, I did mention that in my email from the other thread earlier today, and I think it's potentially an even bigger deal than nonblock open. Lots of workloads are very stat intensive. So I did do that: https://git.kernel.dk/cgit/linux-block/log/?h=nonblock-path-lookup it's the top three patches there. Basically all local file systems are fine with AT_STATX_LOOKUP just basically mapping to LOOKUP_NONBLOCK, various (mostly) networked file systems like to do revalidation and other kinds of communication as part of getattr. Hence the second patch there is needed to have some: if (query_flags & AT_STATX_NONBLOCK) return -EAGAIN; constructs in there to avoid it in the nonblock path. Outside of that, it's very trivial and just works. I didn't include it in this posting to avoid detracting from the core work, but I definitely think we should be doing that as well. -- Jens Axboe
