SMB2.1 leases support for CIFS kernel client

SMB2.1 leases support for CIFS kernel client

[Pavel Shilovsky]

Hi all!

I implemented SMB2.1 leases support for cifs kernel module (in my
smb2-dev branch -
http://git.altlinux.org/people/piastry/public/?p=cifs-2.6.git;a=shortlog;h=refs/heads/smb2-dev).
Now open and create work with leases support - it means that we can
have several simulаneous opens for the same file and use cache for
read and write operations (in the case when we use the same superblock
for these opens).

Whole list of changes since the last status update:
1) SMB2 oplocks support (open, breaks).
2) SMB2.1 leases support (open, breaks).
3) strictcache i/o support for SMB2.

and other minor bugfixes and cleanups.

Your comments are appreciated!

-- 
Best regards,
Pavel Shilovsky.

Re: SMB2.1 leases support for CIFS kernel client

[J. Bruce Fields]

On Tue, Aug 09, 2011 at 09:11:20PM +0400, Pavel Shilovsky wrote:
> Hi all!
> 
> I implemented SMB2.1 leases support for cifs kernel module (in my
> smb2-dev branch -
> http://git.altlinux.org/people/piastry/public/?p=cifs-2.6.git;a=shortlog;h=refs/heads/smb2-dev).
> Now open and create work with leases support - it means that we can
> have several simulаneous opens for the same file and use cache for
> read and write operations (in the case when we use the same superblock
> for these opens).
> 
> Whole list of changes since the last status update:
> 1) SMB2 oplocks support (open, breaks).
> 2) SMB2.1 leases support (open, breaks).
> 3) strictcache i/o support for SMB2.
> 
> and other minor bugfixes and cleanups.
> 
> Your comments are appreciated!

Just a trivial comment: I think that #ifdef CONFIG_CIFS_SMB2 in
cifs_oplock_beak() is a case where the Linux convention of hiding the
ifdef with a conditionally-defined function would be a lot cleaner....

--b.

Re: SMB2.1 leases support for CIFS kernel client

[Pavel Shilovsky]

2011/8/9 J. Bruce Fields <bfields-uC3wQj2KruNg9hUCZPvPmw@public.gmane.org>:
> On Tue, Aug 09, 2011 at 09:11:20PM +0400, Pavel Shilovsky wrote:
>> Hi all!
>>
>> I implemented SMB2.1 leases support for cifs kernel module (in my
>> smb2-dev branch -
>> http://git.altlinux.org/people/piastry/public/?p=cifs-2.6.git;a=shortlog;h=refs/heads/smb2-dev).
>> Now open and create work with leases support - it means that we can
>> have several simulаneous opens for the same file and use cache for
>> read and write operations (in the case when we use the same superblock
>> for these opens).
>>
>> Whole list of changes since the last status update:
>> 1) SMB2 oplocks support (open, breaks).
>> 2) SMB2.1 leases support (open, breaks).
>> 3) strictcache i/o support for SMB2.
>>
>> and other minor bugfixes and cleanups.
>>
>> Your comments are appreciated!
>
> Just a trivial comment: I think that #ifdef CONFIG_CIFS_SMB2 in
> cifs_oplock_beak() is a case where the Linux convention of hiding the
> ifdef with a conditionally-defined function would be a lot cleaner....
>
> --b.
>

Thanks for your comment. I rewrote cifs_oplock_break with callbacks
and generic function approach - changes are in my tree.

-- 
Best regards,
Pavel Shilovsky.

Re: SMB2.1 leases support for CIFS kernel client

[Pavel Shilovsky]

9 августа 2011 г. 21:11 пользователь Pavel Shilovsky
<piastryyy-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org> написал:
> Hi all!
>
> I implemented SMB2.1 leases support for cifs kernel module (in my
> smb2-dev branch -
> http://git.altlinux.org/people/piastry/public/?p=cifs-2.6.git;a=shortlog;h=refs/heads/smb2-dev).
> Now open and create work with leases support - it means that we can
> have several simulаneous opens for the same file and use cache for
> read and write operations (in the case when we use the same superblock
> for these opens).
>
> Whole list of changes since the last status update:
> 1) SMB2 oplocks support (open, breaks).
> 2) SMB2.1 leases support (open, breaks).
> 3) strictcache i/o support for SMB2.
>
> and other minor bugfixes and cleanups.
>
> Your comments are appreciated!
>
> --
> Best regards,
> Pavel Shilovsky.
>

Rebased smb2-dev branch to the current master and fixed some bugs.

I also got some performance testing results of comparison between
write on several opens of the same file into different areas and write
on several opens of different files on the same share:
1) 2 file descriptors
SMB2    : write on 2 fds of the same file == write on 2 fds of different files.
SMB2.1 : write on 2 fds of the same file is ~35% faster than write on
2 fds of different files (on the same share).
2) 3 file descriptors
SMB2    : write on 3 fds of the same file == write on 3 fds of different files.
SMB2.1 : write on 3 fds of the same file is ~50% faster than write on
3 fds of different files (on the same share).

I am going to test it more - I will inform about the results when I get them.

-- 
Best regards,
Pavel Shilovsky.