Re: [PATCH 1/1 V2] mqueue: Implment generic xattr support

From: David Graziano
Date: Thu Dec 08 2016 - 10:16:58 EST

On Mon, Dec 5, 2016 at 5:12 PM, Paul Moore <paul@xxxxxxxxxxxxxx> wrote:
> On Mon, Dec 5, 2016 at 11:15 AM, David Graziano
> <david.graziano@xxxxxxxxxxxxxxxxxxx> wrote:
>> On Mon, Dec 5, 2016 at 9:37 AM, Paul Moore <paul@xxxxxxxxxxxxxx> wrote:
>>> On Mon, Nov 28, 2016 at 3:04 PM, David Graziano
>>> <david.graziano@xxxxxxxxxxxxxxxxxxx> wrote:
>>>> On Wed, Nov 9, 2016 at 4:25 PM, Paul Moore <paul@xxxxxxxxxxxxxx> wrote:
>>>>> On Wed, Nov 9, 2016 at 11:25 AM, David Graziano
>>>>> <david.graziano@xxxxxxxxxxxxxxxxxxx> wrote:
>>>>>> On Mon, Nov 7, 2016 at 4:23 PM, Paul Moore <paul@xxxxxxxxxxxxxx> wrote:
>>>>>>> On Mon, Nov 7, 2016 at 3:46 PM, David Graziano
>>>>>>> <david.graziano@xxxxxxxxxxxxxxxxxxx> wrote:
>>>>>>>> This patch adds support for generic extended attributes within the
>>>>>>>> POSIX message queues filesystem and setting them by consulting the LSM.
>>>>>>>> This is needed so that the security.selinux extended attribute can be
>>>>>>>> set via a SELinux named type transition on file inodes created within
>>>>>>>> the filesystem. The implementation and LSM call back function are based
>>>>>>>> off tmpfs/shmem.
>>>>>>>> Signed-off-by: David Graziano <david.graziano@xxxxxxxxxxxxxxxxxxx>
>>>>>>>> ---
>>>>>>>> ipc/mqueue.c | 46 ++++++++++++++++++++++++++++++++++++++++++++++
>>>>>>>> 1 file changed, 46 insertions(+)
>>>>>>> Hi David,
>>>>>>> At first glance this looks reasonable to me, I just have a two
>>>>>>> questions/comments:
>>>>>>> * Can you explain your current need for this functionality? For
>>>>>>> example, what are you trying to do that is made easier by allowing
>>>>>>> greater message queue labeling flexibility? This helps put things in
>>>>>>> context and helps people review and comment on your patch.
>>>>>>> * How have you tested this? While this patch is not SELinux specific,
>>>>>>> I think adding a test to the selinux-testsuite[1] would be worthwhile.
>>>>>>> The other LSM maintainers may suggest something similar if they have
>>>>>>> an established public testsuite.
>>>>>>> [1]
>>>>>> Hi Paul,
>>>>>> I needed to write a selinux policy for a set of custom applications that use
>>>>>> POSIX message queues for their IPC. The queues are created by one
>>>>>> application and we needed a way for selinux to enforce which of the other
>>>>>> apps are able to read/write to each individual queue. Uniquely labeling them
>>>>>> based on the app that created them and the file name seemed to be our best
>>>>>> solution since itâs an embedded system and we donât have restorecond to
>>>>>> handle any relabeling.
>>>>> In the future putting things like the above in the patch description
>>>>> can be helpful. In other words, instead of simply saying this allows
>>>>> you to better control the labels assigned to message queues, you could
>>>>> expand upon it by saying that this patch allows you to better control
>>>>> which applications have access to a given queue. Yes, I realize that
>>>>> is implied by better control over the labels, but being explicit is
>>>>> rarely a bad thing when it comes to patch descriptions.
>>>>> I've never rejected a patch for a description that was too lengthy,
>>>>> but I have rejected patches that need better descriptions ;)
>>>>>> To test this patch I used both a selinux enabled, buildroot based qemu target
>>>>>> with a customized selinux policy and test C app to create the mqueues. I also
>>>>>> tested with our real apps and selinux policy on our target hardware. I can
>>>>>> certainly look at adding a test to the selinux-testsuite if that would
>>>>>> be helpful.
>>>>> Please do. I've been requiring tests for all new SELinux
>>>>> functionality lately; this isn't strictly a SELinux patch but I think
>>>>> it is a good practice regardless.
>>>> Sorry for the delay. I have created a pull request within the
>>>> selinux-testsuite github
>>>> project with a set of mqueue tests.
>>> For anyone who is curious:
>>> *
>>> Aside from a naming nit, the tests look good to me and I have no
>>> problem with the kernel patch; it doesn't appear any of the other LSM
>>> maintainers do either. I'm happy to pull this into the SELinux tree
>>> (for v4.11, it's a little late for v4.10 I think), but I think
>>> Christoph made a good point about consolidation, have you had a chance
>>> to look at that?
>> I've made the update for the naming nit in the pull request.
> I saw that, thanks.
>> I agree with Christoph's point but doing so is a bit outside my expertise at
>> this point. I would be open to suggestions as to where the function should be
>> consolidated and work on a second patchset with the update. Maybe in
>> fs/xattr.c as a simple_xattr_initxattrs function?
> That seems to make the most sense, doesn't it? Looking at
> {shmem,mqueue}_initxattrs() the only fs specific bit is the
> {shmem,mqueue}_inode_info struct pointer; considering that the fs_info
> parameter is currently unused in this case, you could pass a reference
> to the simple_xattr struct via the fs_info parameter.
> I'd CC Andreas Gruenbacher <agruenba@xxxxxxxxxx> on the patch(set),
> he's recently done a bunch of work around xattrs and the LSM, he may
> have some additional thoughts.
Thanks for the advice. I'm testing a patchset with the proposed changes and
planning to submit them later today.