new statx extensions

new statx extensions

[Steve French]

I saw this patch series relating to exposing DIO alignment information
mentioned in lwn today

https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=825cf206ed51

Do you have any ideas of whether any network filesystem could support
this?  There are lots of features about the server filesystem and
preferred i/o sizes etc. that can be sent over the protocol (for
SMB3.1.1).

Looking at 6.1-rc1 - it looks like this was mainly for xfs and ext4
but is there any reason that it would be beneficial for cifs.ko - and
if so is there more clarification on what information would be needed
from the server to set this value?



--
Thanks,

Steve

Re: new statx extensions

[Eric Biggers]

Hi Steve,

On Thu, Oct 20, 2022 at 10:24:22PM -0500, Steve French wrote:
> I saw this patch series relating to exposing DIO alignment information
> mentioned in lwn today
> 
> https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=825cf206ed51
> 
> Do you have any ideas of whether any network filesystem could support
> this?  There are lots of features about the server filesystem and
> preferred i/o sizes etc. that can be sent over the protocol (for
> SMB3.1.1).
> 
> Looking at 6.1-rc1 - it looks like this was mainly for xfs and ext4
> but is there any reason that it would be beneficial for cifs.ko - and
> if so is there more clarification on what information would be needed
> from the server to set this value?
> 

STATX_DIOALIGN can, and should, be supported by any filesystem that supports
O_DIRECT.

Its scope is specifically direct I/O alignment restrictions.  It doesn't include
anything about the preferred I/O size.  Note that the existing field
"stx_blksize" is already documented to be the preferred I/O size.

If you have any more questions or thoughts about this, please bring them up on
linux-fsdevel@vger.kernel.org.  That's where all the STATX_DIOALIGN stuff has
been discussed.

- Eric