Date:	Tue, 09 Feb 2010 18:21:31 +0300
From:	Dmitry Monakhov <dmonakhov@...nvz.org>
To:	Greg Freemyer <greg.freemyer@...il.com>
Cc:	linux-ext4@...r.kernel.org, Jan Kara <jack@...e.cz>,
	OHSM-DEV <ohsm-devel@...ts.sourceforge.net>
Subject: Re: [PATCH 08/11] ext4: introduce subtree logic

Greg Freemyer <greg.freemyer@...il.com> writes:

> On Tue, Feb 9, 2010 at 9:48 AM, Dmitry Monakhov <dmonakhov@...nvz.org> wrote:
>> Greg Freemyer <greg.freemyer@...il.com> writes:
>>
>>> On Tue, Feb 9, 2010 at 8:53 AM, Dmitry Monakhov <dmonakhov@...nvz.org> wrote:
>>>> Greg Freemyer <greg.freemyer@...il.com> writes:
>>>>
>>>>> On Tue, Feb 9, 2010 at 5:06 AM, Dmitry Monakhov <dmonakhov@...nvz.org> wrote:
>>>>>> Greg Freemyer <greg.freemyer@...il.com> writes:
>>>>>>
>>>>>>> On Mon, Feb 8, 2010 at 8:28 AM, Dmitry Monakhov <dmonakhov@...nvz.org> wrote:
>>>>>>>> * Abstract
>>>>>>>>  A subtree of a directory tree T is a tree consisting of a directory
>>>>>>>>  (the subtree root) in T and all of its descendants in T.
>>>>>>>>
>>>>>>>>  Subtree feature allows to create an isolated (from user point of view)
>>>>>>>>  trees.
>>>>>>>>
>>>>>>>>  Subtree assumptions:
>>>>>>>>  (1) Each inode has subtree id. This id is persistently stored inside
>>>>>>>>      inode (xattr, usually inside ibody)
>>>>>>>>  (2) Subtree id is inherent from parent directory
>>>>>>>>  (3) Inode can not belongs to different subtree
>>>>>>>>      Otherwise changes in one subtree result in changes in other subtree
>>>>>>>>      which contradict to isolation criteria.
>>>>>>>>
>>>>>>>>  This feature is similar to project-id in XFS. One may assign some id to
>>>>>>>>  a subtree. Each entry from the subtree may be accounted in directory
>>>>>>>>  subtree quota. Will appear in later patches.
>>>>>>>>
>>>>>>>> * Disk layout
>>>>>>>>  Subtree id is stored on disk inside xattr usually inside ibody.
>>>>>>>>  Xattr is used only as a data storage, It has not user visiable xattr
>>>>>>>>  interface.
>>>>>>>>
>>>>>>>> Signed-off-by: Dmitry Monakhov <dmonakhov@...nvz.org>
>>>>>>>
>>>>>>> Dmitry,
>>>>>>>
>>>>>>> I think the idea of subtrees is useful, but I'm curious about other
>>>>>>> use cases than just quota.
>>>>>>>
>>>>>>> At first glance you are attempting to create a generic subtree
>>>>>>> functionality for ext4, but criteria 3) above says a inode can only be
>>>>>>> in one subtree at a time.
>>>>>> Theoretically this is possible, but this dramatically complicate things
>>>>>> Just think about this. If inode belongs to different subtrees then
>>>>>> it must have several tree-dquota objects attached to it. This means
>>>>>> that quota require great quota redesign.
>>>>>> Obviously i don't know any use case for this feature. do you know any?
>>>>>
>>>>> Maybe we're talking about different things.  I working with the OHSM
>>>>> project <http://ohsm.sourceforge.net/>
>>>>>
>>>>> We haven't submitted any patches yet, but for one of our features we
>>>>> have something fairly close to your subtree patch.  If we were to
>>>>> leverage your patch and drop that part of ours we would be in the
>>>>> unhappy situation that quota and ohsm could not both be enabled on the
>>>>> same filesystem because the ohsm subtree geography is not likely to be
>>>>> consistent with the quota subtree geography.
>>>>>
>>>>> If we call quota and ohsm services then my desire would be to see your
>>>>> subtree patches support orthoganal subtree groups.  One group per
>>>>> service.
>>>>>
>>>>> I haven't looked into the actual implementation, but from an API
>>>>> perspective it is just a matter of adding a service parameter to the
>>>>> various calls.  For a given subtree service group, a given inode could
>>>>> only be part of one subtree, but a single inode could participate in
>>>>> multiple subtree service groups.
>>>> Ok now i think i understand what you are talking about.
>>>> one subtree <=> one service is the main rule i'm standing.
>>>> If you wan to support several services, no problem
>>>> I can easily extend xattr to support different services
>>>> something like this
>>>> subtree_entry
>>>> {
>>>>        __le16 sbe_flags /* entry flags */
>>>>        __le16 sbe_type  /* service type */
>>>>        __le32 sbe_id    /* subtree id */
>>>> }
>>>
>>> Conceptually agreed.
>>>
>>> But you'd still be limited to one subtree_entry per inode with that right?
>>>
>>> Another option would be more like
>>>
>>> #define EXT4_SUBTREE_MAX_SERVICE_GROUPS 1
>>> #define EXT4_SUBTREE_MAX_SERVICE_GROUP_QUOTA 0
>>>
>>> // OHSM would patch the above when its ready to submit
>>>
>>> subtree_entry[EXT4_SUBTREE_MAX_SERVICE_GROUPS]
>>> {
>>>        __le16 sbe_flags /* entry flags */
>>>        __le32 sbe_id    /* subtree id */
>>> }
>>>
>>> (Clearly that won't actually work as is, but you get the idea.  Each
>>> service group gets its own flags and id.)
>>>
>>> Another option would be separate xattr entries for each subtree group.
>>>
>>> Regardless, it looks very doable and its just a matter of figuring out
>>> the best way to support multiple subtree service groups.
>> Off course i mean that. In fact i'm plan to pack entries in to one
>> xattr data block because each xattr has significant space overhead.
>>
>
> Sounds great.  Does xfs support multiple service groups with its
> current offering?
IFAIK there is only one group, this is because they use it for quota
and for isolation needs. 
In fact due to current linux implementation
the cant use group quota,while project quota is enabled because there
is only two types of types available. This will be fixed by generic
quota improvements see 3-6 patches in the series. 
--
To unsubscribe from this list: send the line "unsubscribe linux-ext4" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Powered by blists - more mailing lists