On 8/22/20 8:33 AM, Theodore Y. Ts'o wrote:

On Fri, Aug 21, 2020 at 03:26:35PM -0600, Jens Axboe wrote:

quoted

quoted

quoted

quoted

Resending this one, as I've been carrying it privately since May. The
necessary bits are now upstream (and XFS/btrfs equiv changes as well),
please consider this one for 5.9. Thanks!

The necessary commit only hit upstream as of 5.9-rc1, unless I'm
missing something?  It's on my queue to send to Linus once I get my
(late) ext4 primary pull request for 5.9.

Right, it went in at the start of the merge window for 5.9. Thanks Ted!

Didn't see it in the queue that just sent in, is it still queued up?

It wasn't in the queue which I queued up because that was based on
5.8-rc4.  Linus was a bit grumpy (fairly so) because it was late, and
that's totally on me.

He has said that he's going to start ignoring pull requests that
aren't fixes only if this becomes a pattern, so while I can send him
another pull request which will just have that one change, there are
no guarantees he's going to take it at this late date.

Sorry, when you sent me the commit saying that the changes that were
needed were already upstream on August 3rd, I thought that meant that
they were aready in Linus's tree.  I should have checked and noticed
that that in fact "ext4: flag as supporting buffered async reads"
wasn't compiling against Linus's upstream tree, so I didn't realize
this needed to be handled as a special case during the merge window.

Well to be honest, this kind of sucks. I've been posting it since May,
and the ideal approach would have been to just ack it and I could have
carried it in my tree. That's what we did for btrfs and XFS, both of
which have it.

The required patches *were* upstreamed on August 3rd, which is why I
mentioned that. But yes, not in 5.8 or earlier, of course.

So I suggest that you either include it for the next pull request for
Linus, or that I put it in with your ack. Either is fine with me. I'd
consider this a "dropping the ball" kind of thing, it's not like the
patch hasn't been in linux-next or hasn't been ready for months. This
isn't some "oh I wrote this feature after the merge window" event. It'd
be a real shame to ship 5.9 and ext4 not have support for the more
efficient async buffered reads, imho, especially since the two other
major local file systems already have it.

Let me know what you think.

-- 
Jens Axboe