SMB2 SET_ZERO_DATA and PUNCH_HOLE

SMB2 SET_ZERO_DATA and PUNCH_HOLE

[ronnie sahlberg]

Folks.

In the cifs client  we use FSCTL_SET_ZERO_DATA when userspace wants to
punch a hole in a file. SET_ZERO_DATA maps quite well to the
PUNCH_HOLE semantics in that it will deallocate what it can and
overwrite what it can not with 0.

On windows 16/ntfs the deallocate blocksize is 64k.

Does anyone know if this is always 64k or if there is a way to query
the server for this?

regards
ronnie sahlberg

RE: SMB2 SET_ZERO_DATA and PUNCH_HOLE

[Tom Talpey]

> -----Original Message-----
> From: linux-cifs-owner@vger.kernel.org <linux-cifs-owner@vger.kernel.org> On
> Behalf Of ronnie sahlberg
> Sent: Thursday, April 25, 2019 3:24 AM
> To: linux-cifs <linux-cifs@vger.kernel.org>; Steve French
> <smfrench@gmail.com>
> Subject: SMB2 SET_ZERO_DATA and PUNCH_HOLE
> 
> Folks.
> 
> In the cifs client  we use FSCTL_SET_ZERO_DATA when userspace wants to
> punch a hole in a file. SET_ZERO_DATA maps quite well to the
> PUNCH_HOLE semantics in that it will deallocate what it can and
> overwrite what it can not with 0.
> 
> On windows 16/ntfs the deallocate blocksize is 64k.

This processing is defined in MS-FSA section 2.1.5.9.35 and it's not a constant
but is dependent on a number of filesystem parameters, primarily the cluster
size (aka the "allocation unit").

> Does anyone know if this is always 64k or if there is a way to query
> the server for this?

Cluster size can be queried via filesystem-specific fsctl's, but beware the special
cases... 

Tom.

Re: SMB2 SET_ZERO_DATA and PUNCH_HOLE

[David Disseldorp]

On Thu, 25 Apr 2019 17:23:56 +1000, ronnie sahlberg wrote:

> Folks.
> 
> In the cifs client  we use FSCTL_SET_ZERO_DATA when userspace wants to
> punch a hole in a file. SET_ZERO_DATA maps quite well to the
> PUNCH_HOLE semantics in that it will deallocate what it can and
> overwrite what it can not with 0.

Cool, I'd be interested to hear how things go when testing against
a Samba SMB2+ server - it similarly maps FSCTL_SET_ZERO_DATA to
PUNCH_HOLE.

> On windows 16/ntfs the deallocate blocksize is 64k.
> 
> Does anyone know if this is always 64k or if there is a way to query
> the server for this?

smbtorture4 includes a sparse_hole_dealloc test for checking this via
incremental ZERO_DATA + QAR requests:
https://git.samba.org/?p=samba.git;a=blob;f=source4/torture/smb2/ioctl.c#l3981

I don't recall seeing anything non-64k at the time, but I didn't check
that thoroughly.

Cheers, David