On 12/10/20 1:53 PM, Linus Torvalds wrote: > On Thu, Dec 10, 2020 at 12:01 PM Jens Axboe <axboe@xxxxxxxxx> wrote: >> >> io_uring always punts opens to async context, since there's no control >> over whether the lookup blocks or not. Add LOOKUP_NONBLOCK to support >> just doing the fast RCU based lookups, which we know will not block. If >> we can do a cached path resolution of the filename, then we don't have >> to always punt lookups for a worker. > > Ok, this looks much better to me just from the name change. > > Half of the patch is admittedly just to make sure it now returns the > right error from unlazy_walk (rather than knowing it's always > -ECHILD), and that could be its own thing, but I'm not sure it's even > worth splitting up. The only reason to do it would be to perhaps make > it really clear which part is the actual change, and which is just > that error handling cleanup. > > So it looks fine to me, but I will leave this all to Al. I did consider doing a prep patch just making the error handling clearer and get rid of the -ECHILD assumption, since it's pretty odd and not even something I'd expect to see in there. Al, do you want a prep patch to do that to make the change simpler/cleaner? -- Jens Axboe
