Re: [PATCHv2 0/3] direct io alignment relax

From: Eric Biggers <ebiggers@kernel.org>
To: Chaitanya Kulkarni <chaitanyak@nvidia.com>
Cc: Keith Busch <kbusch@kernel.org>, Keith Busch <kbusch@fb.com>,
	"linux-fsdevel@vger.kernel.org" <linux-fsdevel@vger.kernel.org>,
	"linux-block@vger.kernel.org" <linux-block@vger.kernel.org>,
	"axboe@kernel.dk" <axboe@kernel.dk>,
	Kernel Team <Kernel-team@fb.com>, "hch@lst.de" <hch@lst.de>,
	"bvanassche@acm.org" <bvanassche@acm.org>,
	"damien.lemoal@opensource.wdc.com"
	<damien.lemoal@opensource.wdc.com>
Subject: Re: [PATCHv2 0/3] direct io alignment relax
Date: Wed, 18 May 2022 19:02:43 -0700	[thread overview]
Message-ID: <YoWlQxEjvlmACNRV@sol.localdomain> (raw)
In-Reply-To: <f42c74b9-67fa-50fc-d97e-2a7f153f10e4@nvidia.com>

On Thu, May 19, 2022 at 01:02:42AM +0000, Chaitanya Kulkarni wrote:
> On 5/18/22 17:51, Keith Busch wrote:
> > On Wed, May 18, 2022 at 11:26:20PM +0000, Eric Biggers wrote:
> >> On Wed, May 18, 2022 at 10:11:28AM -0700, Keith Busch wrote:
> >>> From: Keith Busch <kbusch@kernel.org>
> >>>
> >>> Including the fs list this time.
> >>>
> >>> I am still working on a better interface to report the dio alignment to
> >>> an application. The most recent suggestion of using statx is proving to
> >>> be less straight forward than I thought, but I don't want to hold this
> >>> series up for that.
> >>>
> >>
> >> Note that I already implemented the statx support and sent it out for review:
> >> https://lore.kernel.org/linux-fsdevel/20220211061158.227688-1-ebiggers@kernel.org/T/#u
> >> However, the patch series only received one comment.  I can send it out again if
> >> people have become interested in it again...
> > 
> > Thanks, I didn't see that the first time around, but I'll be sure to look at
> > your new version. It sounds like you encountered the same problem I did
> > regarding block device handles: the devtmpfs inodes for the raw block device
> > handles are not the bdev inodes. I do think it's useful the alignment
> > attributes are accessible through the block device files, though.
> 
> Irrespective of above problem, as per my review comment [1] on the
> initial version of Eric's series I really want to see the generic
> interface that can accommodate exposing optimal values for different
> operations REQ_OP_DISCARD/REQ_OP_WRITE_ZEROES/REQ_OP_VERIFY etc.
> and not only for read/write.
> 
> -ck
> 
> https://lore.kernel.org/linux-fsdevel/20220211061158.227688-1-ebiggers@kernel.org/T/#r3ffe9183c372fb97a9753e286f9cf6400e8ec272

Userspace doesn't know what REQ_OP_* are; they are internal implementation
details of the block layer.  What UAPIs, specifically, do you have in mind for
wanting to know the alignment requirements of?

If you're thinking about about BLKDISCARD and BLKZEROOUT, those are block device
ioctls, so statx() doesn't seem like a great fit for them.  There is already a
place to expose block device properties to userspace (/sys/block/$dev/).  Direct
I/O is different because direct I/O is not just a feature of block devices but
also of regular files, and it is affected by filesystem-level constraints.

- Eric