[PATCH 08/20] nvmet: configure file backed ns for sync IO

[chaitany kulkarni]

You are right, we should not be forcing O_SYNC on the target side.

How about we rename the sync configfs parameter to buffer/cache/vwc?
(or any other preferable name).
We use buffer IO (NO O_* flags when for open and submitting IO) when
that parameter is set and use O_DIRECT when that parameter is not set.

Any thoughts?

On Tue, Apr 24, 2018 at 10:41 AM, Christoph Hellwig <hch at lst.de> wrote:
> On Wed, Apr 18, 2018 at 02:59:59PM -0400, Chaitanya Kulkarni wrote:
>> This patch allows file backed namespaces to use newly
>> introduced configfs attribute sync, when sync is set
>> we open the file with O_SYNC and use the work queue
>> to execute the NVMe read/write commands.
>
> So this adds support for buffered I/O.  But why do you force
> O_SYNC on everyone who uses buffered I/O?
>
> _______________________________________________
> Linux-nvme mailing list
> Linux-nvme at lists.infradead.org
> http://lists.infradead.org/mailman/listinfo/linux-nvme