Re: [PATCH 0/3 RFC] Remove DAX experimental warnings

From: Dan Williams <dan.j.williams@intel.com>
To: Eric Sandeen <sandeen@sandeen.net>
Cc: Eric Sandeen <sandeen@redhat.com>,
	linux-xfs <linux-xfs@vger.kernel.org>,
	linux-ext4 <linux-ext4@vger.kernel.org>,
	linux-fsdevel <linux-fsdevel@vger.kernel.org>,
	Shiyang Ruan <ruansy.fnst@fujitsu.com>
Subject: Re: [PATCH 0/3 RFC] Remove DAX experimental warnings
Date: Wed, 15 Sep 2021 11:59:59 -0700	[thread overview]
Message-ID: <CAPcyv4gFB_nefaEMVaPb4x4Q61Rr3Q1JdOr7cytBmQcbpaUJng@mail.gmail.com> (raw)
In-Reply-To: <cb13be1c-66f1-8452-e7ab-c1278c8e51e0@sandeen.net>

On Wed, Sep 15, 2021 at 11:49 AM Eric Sandeen <sandeen@sandeen.net> wrote:
>
> On 9/15/21 1:35 PM, Dan Williams wrote:
> > On Wed, Sep 15, 2021 at 10:23 AM Eric Sandeen <sandeen@redhat.com> wrote:
> >>
> >> For six years now, when mounting xfs, ext4, or ext2 with dax, the drivers
> >> have logged "DAX enabled. Warning: EXPERIMENTAL, use at your own risk."
> >>
> >> IIRC, dchinner added this to the original XFS patchset, and Dan Williams
> >> followed suit for ext4 and ext2.
> >>
> >> After brief conversations with some ext4 and xfs developers and maintainers,
> >> it seems that it may be time to consider removing this warning.
> >>
> >> For XFS, we had been holding out for reflink+dax capability, but proposals
> >> which had seemed promising now appear to be indefinitely stalled, and
> >> I think we might want to consider that dax-without-reflink is no longer
> >> EXPERIMENTAL, while dax-with-reflink is simply an unimplemented future
> >> feature.
> >
> > I do regret my gap in engagement since the last review as I got
> > distracted by CXL, but I've recently gotten my act together and picked
> > up the review again to help get Ruan's patches over the goal line [1].
> > I am currently awaiting Ruan's response to latest review feedback
> > (looks like a new posting this morning). During that review Christoph
> > identified some cleanups that would help Ruan's series, and those are
> > now merged upstream [2]. The last remaining stumbling block (further
> > block-device entanglements with dax-devices) I noted here [2]. The
> > proposal is to consider eliding device-mapper dax-reflink support for
> > now and proceed with just xfs-on-/dev/pmem until Mike, Jens, and
> > Christoph can chime in on the future of dax on block devices.
> >
> > As far as I can see we have line of sight to land xfs-dax-reflink
> > support for v5.16, does anyone see that differently at this point?
>
> Thanks for that update, Dan. I'm wondering, even if we have renewed
> hopes and dreams for dax+reflink, would it make sense to go ahead and
> declare dax without reflink non-experimental, and tag dax+reflink as
> a new "EXPERIMENTAL feature if and when it lands?

As I replied to the xfs patch in your series, I say "yes" EXPERIMENTAL
can go now, because the concern was reflink support might regress
dax-semantics wrt MAP_SYNC and the like. That concern seems to be
avoided by the current direction.