Re: [PATCH] VFS: Unlink should revoke all outstanding leases on file

[Mi Jinlong <mijinlong@cn.fujitsu.com>]

J. Bruce Fields :
> On Wed, May 19, 2010 at 05:46:23PM +0800, Mi Jinlong wrote:
>>
>> J. Bruce Fields :
>>> On Fri, May 14, 2010 at 02:31:12PM -0400, Trond Myklebust wrote:
>>>> On Fri, 2010-05-14 at 13:59 -0400, Trond Myklebust wrote: 
>>>>> Note that the server should also recall the delegation if someone
>>>>> attempts to violate the guarantees that are listed in section 9.4: Open
>>>>> Delegation
>>>>>
>>>>>    When a client has a read open delegation, it may not make any changes
>>>>>    to the contents or attributes of the file but it is assured that no
>>>>>    other client may do so.  When a client has a write open delegation,
>>>>>    it may modify the file data since no other client will be accessing
>>>>>    the file's data.  The client holding a write delegation may only
>>>>>    affect file attributes which are intimately connected with the file
>>>>>    data:  size, time_modify, change.
>>>>>
>>>>> IOW: even if you hold a write delegation you are not allowed to change
>>>>> the file mode bits, owner, group or acls...
>>>> ...or the nlink value. So technically, we should also recall the
>>>> delegation when someone creates or deletes a hard link. I think I need
>>>> to remind Tom that he should add that to the RFC3530bis draft...
>>> Yep.  And fixing all these cases is required before our the server's
>>> NFSv4 server is ready for much of anything.
>>>
>>> I'm not sure ading break_lease() to may_delete() is right, but maybe
>>> it's better than nothing.
>>   Agree with you.
>>
>>> One problem is that there's a race: nothing I can see stops anyone from
>>> getting another lease after may_delete() but before the delete happens.
>>   Yes. 
>>   The problem will exist, but there isn't some better methods to avoid it.
>>   Is there a lease lock exist in kernel? 
>>   If that's true, the problem will be fixed simply.
> 
> I don't know of any existing lock that does exactly what we want.
> 
> Somebody at citi worked on a better lease implementation for a while,
> but I don't think we ever really got it right; the last version I can
> find is here:
> 
> 	git://linux-nfs.org/~bfields/linux-topics.git leases

  When reading the code of the git, i found a patch which try to fix
  the lease's problem, but only for unlink.

  commit id: d5a08e556116c66fb60a448f805a40bf54314634
        msg: "leases: break file leases on unlink."

  In this patch, it seems only add break_lease() and some other functions
  but seems don't avoid the problem of race. Or there is some different 
  at break_lease() with the community's kernel.

  Can you give me some message about the new lease? Thanks.

thanks,
Mi Jinlong