FS-Cache for cifs

FS-Cache for cifs

[Xiaoli Feng]

Hello everyone,

Recently I'd like to test fs-cache for cifs. But CONFIG_CIFS_FSCACHE is not set defaultly.
Are there any concern to enable it? Test it to enbale fs-cache. It seems work. The file 
/proc/fs/fscache/stats is update when do some cp operations.

Thanks.

-- 
Best regards!
XiaoLi Feng 冯小丽

Re: FS-Cache for cifs

[Steve French]

fscache (perhaps more so with the recent rewrite that Dave Howells
did) may be most well suited to cifs.ko (SMB3.1.1 mounts) among the
various file systems since it would allow offline caching of files and
directories leveraging:
1) handle leases and directory leases for "strict caching" models
or
2) directory change notification for "loose caching" models
(Although file version numbers are not provided, the combination of
creation time, 64 bit DiskFileId, and last write time with 100ns time
granularity is probably sufficient to use in conjunction with this)
In addition the protocol already supports four flags to control
whether client side offline caching can/should be done:
   SMB2_SHAREFLAG_MANUAL_CACHING
   SMB2_SHAREFLAG_AUTO_CACHING
   SMB2_SHAREFLAG_VDO_CACHING
   SMB2_SHAREFLAG_NO_CACHING

So fscache could be very, very useful for cifs.ko, especially for
metadata heavy workloads that are largely from one client ... but
fscache doesn't have tight integration with many cifs features (like
handle leases e.g.) yet.

It would make sense to better tie cifs.ko in with fscache (especially
as it has shown to be useful on other operating systems over
SMB3/SMB3.1.1).

On Mon, Aug 10, 2020 at 8:25 PM Xiaoli Feng <xifeng@redhat.com> wrote:
>
> Hello everyone,
>
> Recently I'd like to test fs-cache for cifs. But CONFIG_CIFS_FSCACHE is not set defaultly.
> Are there any concern to enable it? Test it to enbale fs-cache. It seems work. The file
> /proc/fs/fscache/stats is update when do some cp operations.
>
> Thanks.
>
> --
> Best regards!
> XiaoLi Feng 冯小丽
>


-- 
Thanks,

Steve

Re: FS-Cache for cifs

[Xiaoli Feng]

Thanks Steve for the info. File a bug for cifs fscache.

Bug 208883 - CIFS: kernel BUG at fs/cachefiles/rdwr.c:715!
https://bugzilla.kernel.org/show_bug.cgi?id=208883


----- Original Message -----
> From: "Steve French" <smfrench@gmail.com>
> To: "Xiaoli Feng" <xifeng@redhat.com>
> Cc: "CIFS" <linux-cifs@vger.kernel.org>, "David Howells" <dhowells@redhat.com>
> Sent: Tuesday, August 11, 2020 10:12:28 AM
> Subject: Re: FS-Cache for cifs
> 
> My tests in recent kernels with fscache and cifs.ko worked but didn't
> provide much performance benefit (which would be expected since close
> releases leases )
> 
> On Mon, Aug 10, 2020, 20:51 Steve French <smfrench@gmail.com> wrote:
> 
> > fscache (perhaps more so with the recent rewrite that Dave Howells
> > did) may be most well suited to cifs.ko (SMB3.1.1 mounts) among the
> > various file systems since it would allow offline caching of files and
> > directories leveraging:
> > 1) handle leases and directory leases for "strict caching" models
> > or
> > 2) directory change notification for "loose caching" models
> > (Although file version numbers are not provided, the combination of
> > creation time, 64 bit DiskFileId, and last write time with 100ns time
> > granularity is probably sufficient to use in conjunction with this)
> > In addition the protocol already supports four flags to control
> > whether client side offline caching can/should be done:
> >    SMB2_SHAREFLAG_MANUAL_CACHING
> >    SMB2_SHAREFLAG_AUTO_CACHING
> >    SMB2_SHAREFLAG_VDO_CACHING
> >    SMB2_SHAREFLAG_NO_CACHING
> >
> > So fscache could be very, very useful for cifs.ko, especially for
> > metadata heavy workloads that are largely from one client ... but
> > fscache doesn't have tight integration with many cifs features (like
> > handle leases e.g.) yet.
> >
> > It would make sense to better tie cifs.ko in with fscache (especially
> > as it has shown to be useful on other operating systems over
> > SMB3/SMB3.1.1).
> >
> > On Mon, Aug 10, 2020 at 8:25 PM Xiaoli Feng <xifeng@redhat.com> wrote:
> > >
> > > Hello everyone,
> > >
> > > Recently I'd like to test fs-cache for cifs. But CONFIG_CIFS_FSCACHE is
> > not set defaultly.
> > > Are there any concern to enable it? Test it to enbale fs-cache. It seems
> > work. The file
> > > /proc/fs/fscache/stats is update when do some cp operations.
> > >
> > > Thanks.
> > >
> > > --
> > > Best regards!
> > > XiaoLi Feng 冯小丽
> > >
> >
> >
> > --
> > Thanks,
> >
> > Steve
> >
> 

Re: FS-Cache for cifs

[David Wysochanski]

On Wed, Aug 12, 2020 at 1:15 AM Xiaoli Feng <xifeng@redhat.com> wrote:
>
> Thanks Steve for the info. File a bug for cifs fscache.
>
> Bug 208883 - CIFS: kernel BUG at fs/cachefiles/rdwr.c:715!
> https://bugzilla.kernel.org/show_bug.cgi?id=208883
>
>
I updated the bug - this is not a CIFS issue but a general bug in
current fscache implementation and will be fixed once the fscache-iter
rewrite is merged.

Re: FS-Cache for cifs

[Aurélien Aptel]

Is there an overview document somewhere that describe what fscache does?
We are seeing some warnings about duplicated entries in some scenarios
but I'd like to understand better what it does before changing what goes
in the key.

Cheers,
-- 
Aurélien Aptel / SUSE Labs Samba Team
GPG: 1839 CB5F 9F5B FB9B AA97  8C99 03C8 A49B 521B D5D3
SUSE Software Solutions Germany GmbH, Maxfeldstr. 5, 90409 Nürnberg, DE
GF: Felix Imendörffer, Mary Higgins, Sri Rasiah HRB 247165 (AG München)

Re: FS-Cache for cifs

[David Howells]

Aurélien Aptel <aaptel@suse.com> wrote:

> Is there an overview document somewhere that describe what fscache does?
> We are seeing some warnings about duplicated entries in some scenarios
> but I'd like to understand better what it does before changing what goes
> in the key.

Documentation/filesystems/caching/* in the kernel.

David