Re: [PATCH] 3.8: access permission filesystem

[Olaf Dietsche <olaf--list.linux-kernel@olafdietsche.de>]

ebiederm@xmission.com (Eric W. Biederman) writes:

> Olaf Dietsche <olaf--list.linux-kernel@olafdietsche.de> writes:
>
>> I am in the process of catching up with the last two years or so.
>> Right now, I am at the changes involving user namespaces.
>>
>> I have two possible implementations, both working equally well in a
>> shared environment. Since I am not familiar with namespaces in general
>> and user namespaces in particular, I would like you to look over the
>> patches and tell me, what you think.
>>
>> Are the patches good so far? Are there are any things I missed and must
>> consider? Maybe, I am completely off track? Anything else?
>>
>> I included both patches inline below. The patches are also available as
>> separate branches at github 
>>
>> https://github.com/olafdietsche/linux-accessfs/tree/tmp-user-ns-1
>> https://github.com/olafdietsche/linux-accessfs/tree/tmp-user-ns-2
>>
>> I am leaning toward the second patch. Although it is a little bit longer
>> than the first one, it involves no user id conversions.
>
> Using kuid's and kgid's throughout as your second patch does is best.
> Conversion is only needed on normal filesystems because they have a
> backing store and reside on disk.  As accessfs appears not to have
> backing store, storing things with kuid's and kgid's is the preferred
> method.
>
> Your first patch is buggy as it uses current_user_ns().  Something a
> filesystem in general should not care about.

I have seen similar uses in other filesystems like ext3, jfs or
debugfs. What would be the correct way to use make_kuid() or make_gid()?

> I don't see anything wrong with your second patch.

Thanks a lot for this fast response and your guidance. So, I will dump
the first and continue with the second patch.

> From what little I understand of accessfs, I expect you will want to
> play with and come up to speed on namespaces, as namespaces change
> the universe of objects you will be dealing with, in some subtle
> but interesting ways.  At least assuming anyone in who uses accessfs
> is going to be using more than a single container.
>
> Eric

Regards, Olaf