[Kimchi-devel] RFC: Design of Authorization in Kimchi
Wen Wang
wenwang at linux.vnet.ibm.com
Tue Jul 8 08:58:36 UTC 2014
On 07/08/2014 01:16 PM, Wen Wang wrote:
>
> On 07/08/2014 04:01 AM, Aline Manera wrote:
>>
>>
>> On 07/07/2014 07:35 AM, Aline Manera wrote:
>>>
>>>
>>> On 07/07/2014 06:45 AM, Wen Wang wrote:
>>>> Hi all,
>>>>
>>>> Due to the fact that Kimchi needs authorization feature to be
>>>> designed.
>>>> I an posting my point of view below of how I thought about doing it,
>>>> including how I plan doing it in the front-end and request for help
>>>> for
>>>> the back end support.
>>>>
>>>> Kimchi changed to a traditional login patten in last release that
>>>> makes
>>>> Kimchi more secure to use. It Before login, the front-end can
>>>> hardly get
>>>> any html information before user actually login.
>>>
>>> If the user is not logged in, Kimchi server will always return 401 for
>>> all the requests.
>>> As the front end make requests to server to populate the html, if the
>>> user hardly gets any html he/she will get it empty without any useful
>>> information
>>> At least, it is suppose to work like that.
>>>
>>> As we discussed, root
>>>> user will have full access to Kimchi whereas the non-root user will
>>>> have
>>>> restricted privileges. It will be easier and more decent to show the
>>>> proper tabs to certain users that distinguished by the back-end.
>>>> Now the
>>>> tabs are generated by an xml file generated from the back-end that
>>>> show
>>>> all 5 tabs. We probably need to have the '*Host*' and '*template*'
>>>> tab_removed_ for non-root users, which is recommended to be done in
>>>> the
>>>> back-end.
>>>
>>> I suggest to add one parameter to the tabs in the xml.
>>>
>>> Example: access="restricted" which means only root users can see
>>> those tabs
>>>
>>> And in the front end while loading the tabs, we need to query this
>>> parameter and act accordingly (ie, do not display the tab with this
>>> parameter for a non-root user)
>>>
>>> <tabs>
>>> <tab access="restricted">
>>> <title>Host</title>
>>> <path>tabs/host.html>
>>> </tab>
>>> <tab>
>>> <title>Guests</title>
>>> <path>tabs/guests.html>
>>> </tab>
>>> ...
>>> </tabs>
>>>
>>
>> I've just thought more about that and maybe it won't be enough
>> Probably, for each tab we should describe which view display
>> according to user role
>>
>> <tab>
>> <title>Host</title>
>> <path>tabs/host.html>
>> <views>
>> <view role="admin" mode="full" />
>> <view role="user" mode="none" />
>> </views>
>> </tab>
>>
>> For "mode" we can have:
>> - full: full access to tab content
>> - none: tab should not be displayed
>> - resource: user can manage the resource he/she is assigned to but
>> not create a new one
>> - read-only: user can see the resources but not manage them or create
>> a new one
> It is nice that we add the role attributes that well indicate the
> right authorization, benefit from which the browser can provide the
> proper privileges to certain users. However we have the role already
> and according to which we can easily define the mode by user. By this
> means we probably won't need the mode attribute since the browser can
> assign different function according to role.
>
> Though we can difine more role that fulfil different modes like admin
> can have "full" access and user should have the role discribed as
> "none". It's good point.
Talked to Yu Xin, and we figured out your mode design is pretty
brilliant for adding the mode attribute. According to which I think we
can accomplish assigning different privileges by an easier way. Here is
my thought:
The front-end need only mode attribute that indicate which tab as well
as action a user can do. Take the following script for example:
<tab *mode="none"*>
<title>Host</tittle>
<path>tabs/host.html</path>
</tab>
<tab *mode="full"*>
<title>Guests</title>
<path>tabs/guests.html</path>
</tabs>
<tab *mode="readonly"*>
<title>Network</title>
<path>tabs/network.html</path>
</tabs>
Browser should know the "Host" tab should not be displayed, user should
have full access to "Guests" and "Network" tab should only be displayed
but not avaliable for editing. The role should be defined in the
back-end to indicate different roles have different access to different
tabs as well as actions. Modes are assigned to the tabs and all
instances under that tab inherit the same mode. Server filter which mode
should a user have in the back-end. User role change might be enabled in
the future and server should have the user role changed in the back-end
and return right mode to the browser.
Or I have another thought is a fine grained design also inspired by your
mode pattern. Mode is defined by instance. The back-end should store the
mode for each instance. For example, one user have 10 VMs and have full
access for 3 of them including editing but only vnc connect for 7 of
them. There should be another level of mode defined to the instance
under each table. It could probably but not necessarily be like this :
- master: Full access to the instance
- authroized: Full access except for create/delete
- usable: Only connection is allowed(For VMs' concern)
- display: readonly
Request for your comment
Best Regards
Wang Wen
>>
>> And in the /login request we return a list of user roles
>>
>> {
>> username: alinefm,
>> roles: [admin]
>> groups: [group1, group2]
>> }
>>
>> For now, only one value will be returned for "roles" but later one
>> user can have multiples roles: vm-user, network-admin, etc
>>
>>
>>>>
>>>> Also there need to be information provided to the front-end like the
>>>> user-name, user-role as well as user-group, etc. that indicate user
>>>> identity after login.
>>>
>>>
>>> The browser need the information to give certain
>>>> privileges to certain users and disable the unnecessary functions. My
>>>> suggestion is to have these 3 parameters passed: ***user-name,
>>>> user-role* as well as *user-group*. There is a better extendibility to
>>>> user the user-role other than isRoot so that we can define more
>>>> roles in
>>>> the future. As fact that we have only defined two roles now, the
>>>> user-role parameter can be divided into root and guest based on
>>>> user is
>>>> root or non-root.
>>>
>>> Today that information is returned as response for the request /login
>>>
>>> POST /login {username: alinefm, password: mypassword}
>>> {
>>> username: alinefm,
>>> sudo: true,
>>> groups: [group1, group2]
>>> }
>>>
>>> If "sudo" is true, the user has root permissions, otherwise it is a
>>> non-root user.
>>>
>>> Based on that you said, I propose to change the "sudo" parameter to
>>> "role" and it the user has root permissions we set it to "admin",
>>> otherwise, "user"
>>>
>>> POST /login {username: alinefm, password: mypassword}
>>> {
>>> username: alinefm,
>>> role: admin,
>>> groups: [group1, group2]
>>> }
>>>
>>>
>>> These message can get from *sessiondada*, *cookie *or
>>>> passed according to a query. the way passing the info of the user is
>>>> still under discussion.
>>>
>>> As you will get that info after a login request I propose to store that
>>> info locally on JS
> Storing it in JS as a variable is fine, according to which, the
> back-end should apply the API as you discribed above which browser can
> query everytime it's needed. We have to take user refreshing the page
> into consideration. Or else, we can store it into sessionstorage which
> has the similar function as cookie in HTML5.
>
> I recommend the second solution since it reduced the effort that
> request for role and username everytime we need to have the page
> refreshed.
>
> Best Regards
>
> Wang Wen
>>>
>>>
>>> Request for your advises.
>>>>
>>>> Best Regards
>>>>
>>>> Wang Wen
>>>>
>>>>
>>>> _______________________________________________
>>>> Kimchi-devel mailing list
>>>> Kimchi-devel at ovirt.org
>>>> http://lists.ovirt.org/mailman/listinfo/kimchi-devel
>>>>
>>>
>>> _______________________________________________
>>> Kimchi-devel mailing list
>>> Kimchi-devel at ovirt.org
>>> http://lists.ovirt.org/mailman/listinfo/kimchi-devel
>>>
>
> _______________________________________________
> Kimchi-devel mailing list
> Kimchi-devel at ovirt.org
> http://lists.ovirt.org/mailman/listinfo/kimchi-devel
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.ovirt.org/pipermail/kimchi-devel/attachments/20140708/7ff75d66/attachment.html>
More information about the Kimchi-devel
mailing list