Re: [PATCH -next] quota: fix QUOTACTL kconfig for gfs2

[Randy Dunlap <randy.dunlap@oracle.com>]

On 08/12/10 09:34, Steven Whitehouse wrote:
> Hi,
> 
> On Thu, 2010-08-12 at 09:03 -0700, Randy Dunlap wrote:
>> On 08/12/10 08:57, Steven Whitehouse wrote:
>>> Hi,
>>>
>>> On Thu, 2010-08-12 at 08:39 -0700, Randy Dunlap wrote:
>>>> On 08/11/10 19:46, Jan Kara wrote:
>>>>> On Wed 11-08-10 16:24:35, Randy Dunlap wrote:
>>>>>> From: Randy Dunlap <randy.dunlap@oracle.com>
>>>>>>
>>>>>> warning: (GFS2_FS && BLOCK && EXPERIMENTAL && (64BIT || LBDAF)) selects QUOTACTL which has unmet direct dependencies (XFS_QUOTA || QUOTA)
>>>>>>
>>>>>> Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com>
>>>>>> Cc:	Steven Whitehouse <swhiteho@redhat.com>
>>>>>> Cc:	cluster-devel@redhat.com
>>>>>> ---
>>>>>>  fs/quota/Kconfig |    2 +-
>>>>>>  1 file changed, 1 insertion(+), 1 deletion(-)
>>>>>>
>>>>>> However, the "depends on" line here seems like overkill to me.
>>>>>   OK, so do you mean that making QUOTACTL default to 'n', removing the
>>>>> depends on line and letting QUOTA, XFS_QUOTA, and GFS2_QUOTA select
>>>>> QUOTACTL would be better?
>>>>
>>> I'm not sure I follow. What is GFS2_QUOTA in this case? GFS2 used to
>>
>> That's OK.  I'm not following you either.  ;)  Where is GFS2_QUOTA?
>>
> Well GFS2_QUOTA was referred to in the above comment, but it doesn't

Oh, in Jan's comment.  I see.

> exist as quotas are not optional in GFS2. That is partly a hangover from
> GFS1 and partly because it makes things a lot more complicated in a
> cluster filesystem if you can't be sure that all nodes understand the
> same options.
> 
>>
>>> select QUOTA as well, but Christoph removed that. Shouldn't the user
>>> interface part of quota be separate from the core implementation?
>>
>> Do people use one without the other?
>>
> The original GFS2 implementation was totally separate, again it was
> largely inherited from GFS1, but some changes (read/write of the quota
> file via the gfs2meta fs type, rather than the GFS1 read/write ioctls).
> More recently I added support for updating the quotas via the XFS-style
> interface with the longer term aim of being able to use quota-tools for
> all GFS2 quota operations and (eventually) dropping support for the
> gfs2_quota tool.
> 
> As a result of that GFS2 uses the quotactl interface code, but retains
> all its original, internal, quota code as well.
> 
>>> http://git.kernel.org/?p=linux/kernel/git/torvalds/linux-2.6.git;a=commitdiff;h=e9edb1d8a345119c9baafa1b240eb1ec06a44662
>>
>> Do you prefer my original patch that started this thread?
>>
>> If fs/quota/Kconfig says:
>>
>> config QUOTACTL
>> 	bool
>> 	depends on XFS_QUOTA || QUOTA || GFS2_FS
>> 	default y
>>
>> then there is no need for fs/gfs2/Kconfig to select QUOTACTL.
>>
>>
>> Anyway, this is a centralized kconfig control of QUOTACTL.  I was just suggesting
>> that having each fs control is locally would be more manageable.  It's not a big deal.
>>
> Hmm. I'd been looking on the quotactl code as basically being a library
> function, so I'd have expected to have the filesystems select it if they
> require it.

Yes, that's how I see it also, but Jan can correct me on that...

> I thought the original issue was that there was some interdependency
> between QUOTA and QUOTACTL or is that not what that message was saying?

I don't think so.  It's just a Kconfig language muckup & cleanup AFAICT.


-- 
~Randy
*** Remember to use Documentation/SubmitChecklist when testing your code ***