On Thu, Aug 11, 2022 at 05:30:03PM -0400, Jeff Layton wrote: > On Thu, 2022-08-11 at 22:22 +0100, Al Viro wrote: > > On Thu, Aug 11, 2022 at 05:08:11PM -0400, Jeff Layton wrote: > > > > > Actually, I never got a formal ack from Al. I did send it repeatedly, > > > but I assume he has been too busy to respond. We've had it sitting in > > > linux-next for a couple of months, and he did suggest that approach in > > > the first place, but I too would also prefer to see his official ack on > > > it. > > > > "Suggested approach" had been about inode_insert5() changes, right? > > Right. I was talking about this patch (which I think is sane): > > fs: change test in inode_insert5 for adding to the sb list It is, AFAICS. > > But that's fs/inode.c side of things... I have to admit that I'd missed > > the unlining d_same_name() - exporting the sucker per se didn't look > > insane and I hadn't looked at that in details ;-/ > > > > Looking at it now... might be worth renaming it into __d_same_name(), > > leaving it inlined and exporting a wrapper; not sure if the impact on > > d_lookup()/__d_lookup()/d_alloc_parallel() is worth worrying about it, > > though. > > > > Profiling a case when we have a plenty of files in the same directory > > on tmpfs, with something earlier in the pathname to kick out of RCU > > mode (e.g. going through /proc/self/cwd) might be interesting... > > The d_name_name changes seemed ok to me, but it would be good to have > your ack (or qualified NAK) if possible. Exporting the functionality? Sure, no problem. Uninlining that one... I suspect that it's OK, but I'd like to see profiling data; it's not as if it would be hard to return to having it inlined, obviously. Again, my apologies for not spotting that one.
