Re: [xwiki-devs] [VOTE] Use the new security-authorization module per default in XWiki 5.0

On Fri, Mar 15, 2013 at 12:06 PM, Jean-Vincent Drean <jv(a)xwiki.com>

>>>>> wrote: >>>>>> >>>>>>> 2013/3/14 Denis Gervalle &lt;dgl(a)softec.lu&gt; >>>>>>> >>>>>>>> On Thu, Mar 14, 2013 at 11:12 PM, Jerome Velociter < >>>>> jerome(a)velociter.fr >>>>>>>>> wrote: >>>>>>>> >>>>>>>>> Hi Denis, >>>>>>>>> >>>>>>>>> Le 14/03/13 22:59, Denis Gervalle a écrit : >>>>>>>>> >>>>>>>>> On Thu, Mar 14, 2013 at 9:20 PM, Denis Gervalle &lt;dgl(a)softec.lu&gt; >>>>>>> wrote: >>>>>>>>>> >>>>>>>>>> Hi devs, >>>>>>>>>>> >>>>>>>>>>> We have a new (component based) authorization module since a > while >>>>>>> now, >>>>>>>>>>> and I think 5.0 is the perfect time to introduce it as the > default >>>>>>>> right >>>>>>>>>>> service. First, I simply propose to change the default in >>> xwiki.cfg: >>>>>>>>>>> >>>>>>>>>>> >>>>>>>>>>> >>>>>>>> >>>>> > xwiki.authentication.**rightsclass=org.xwiki.**security.authorization.** >>>>>>>>>>> internal.**XWikiCachingRightService >>>>>>>>>>> >>>>>>>>>>> (Later, I propose that we deprecate that bridge and that we >>> create a >>>>>>>>>>> friendly (xwiki oriented) interface over the more generic >>>>>>>>>>> org.xwiki.security.**authorization.**AuthorizationManager. But >>> leave >>>>>>>>>>> this for a >>>>>>>>>>> later proposal.) >>>>>>>>>>> >>>>>>>>>>> So this vote is about changing the default in xwiki.cfg before >>>>> 5.0M2. >>>>>>>>>>> >>>>>>>>>>> pros: >>>>>>>>>>> - improved performance, since the new service is using caching >>>>>>>>>>> techniques >>>>>>>>>>> and a single page load required lots of calls to it. >>>>>>>>>>> - ability for extension to add new rights >>>>>>>>>>> - define right declaratively >>>>>>>>>>> - separate method for checking and verifying right (throws > opposed >>>>>>> to >>>>>>>>>>> boolean return) >>>>>>>>>>> - fix some long waiting bugs like XWIKI-5174, XWIKI-6987, as > well >>>>>>> as >>>>>>>>>>> some unstated ones >>>>>>>>>>> >>>>>>>>>>> Also XWIKI-4550 >>>>>>>>>> >>>>>>>>>> - possibility to easily solve issues like XWIKI-4491 >>>>>>>>>>> - no more admin right per default >>>>>>>>>>> - being in good position to improve it and release > dependencies to >>>>>>>>>>> oldcore for security matters. >>>>>>>>>>> - possibility for third party to adapt the right settler to > their >>>>>>>>>>> special >>>>>>>>>>> needs (right decision is plugable) >>>>>>>>>>> - a consistant right evaluation with very few exception that > could >>>>>>> be >>>>>>>>>>> explained and documented >>>>>>>>>>> >>>>>>>>>>> cons: >>>>>>>>>>> - no more admin right per default, but since we have DW, the >>>>>>> initial >>>>>>>>>>> setup is no more a problem, and advanced users may use > superadmin. >>>>>>>>>>> - groups are only checked from the user wiki, not from the >>> accessed >>>>>>>>>>> entity wiki. >>>>>>>>>>> >>>>>>>>>> >>>>>>>>> This sound like a big regression. >>>>>>>>> >>>>>>>>> Can you explicit more ? Does this mean that adding a global

>>>>> wiki) >>>>>>>>> user in a local group has no effect ? >>>>>>>>> >>>>>>>> >>>>>>>> You have got it right. This could be improved, and help is > welcome. >>>>> What >>>>>>>> happen is that the user groups are evaluated independently to the >>>>>>> targeted >>>>>>>> entity, and therefore only in the user wiki. >>>>>>>> >>>>>>>> I admit this is a regression, but I have not cross lots of use > case >>>>> like >>>>>>>> those. The simple display in admin of Global user in local Group > is >>>>> even >>>>>>>> broken (double xwiki:xwiki:...) so this does not seems to me a > common >>>>>>>> usage. >>>>>>>> You may provide access to global group in a local wiki to achieve > the >>>>>>> same >>>>>>>> goals. >>>>>>>> >>>>>>>> >>>>>>> This looks to be indeed a big regression. It's quite a common use > case >>>>> to >>>>>>> have only global users and to create groups in the local wiki that >>>>> refer to >>>>>>> local users. >>>>>>> >>>>>> ^^^^^^^^ >>>>>> I suppose you means global users here. >>>>>> >>>>>> IMHO, having user managed by a separate entity (global admin), and >>> these >>>>>> same individual users grouped by another one (local admin) is very >>>>> uncommon >>>>>> delegation of authority to me (but I may be wrong). On the other > hand, >>>>>> having a local admin providing access to local ressources to global >>> group >>>>>> (and potentially some global users) makes more sense. In that way, > the >>>>> same >>>>>> admin manage its users, and group its users, and the local admin > trust >>>>> the >>>>>> global admin to know its users. >>>>>> >>>>>> That said, I am not against any improvement on the way it works, if > it >>>>> is a >>>>>> common use case (moreover used by workspace), we should obviously >>> support >>>>>> it. However, I am convince that evaluating groups based on both the >>> user >>>>>> and the targeted entity is not easily achievable and conduct to

>>>>>> complex partial caching. >>>>>> >>>>>> I have currently not implemented in the security module anything > that >>>>> would >>>>>> cause all wikis to be scanned, and I would really like to avoid > that to >>>>>> happen. So, it will be difficult to avoid partial caching, but we > need >>> to >>>>>> limit that at the higher level, the subwiki. This would allow to

>>> only >>>>>> scan both the wiki of the user and the target entry to consider our >>> cache >>>>>> valid. It means subwiki will be unable to share groups (I do not > think >>>>> this >>>>>> has ever worked), but it will keep performance on large farm. >>>>>> >>>>>> This would really need to be fixed sooner than later otherwise I > know >>>>>>> plenty of projects for which migration to 5.0 would be almost >>> impossible >>>>>>> >>>>>> >>>>>> I will need helps to achieve that for 5.0 >>>>> >>>>> ok so before changing anything we need a plan i.e. someone > volunteering >>> to >>>>> work on this, right? >>>>> >>>> >>>> Do we really need that for 5.0 ? >>>> Using the new module as a default does not means the old right > service is >>>> unavailable. Couldn't we simply define which case needs to revert to > the >>>> old modules in the RN, and have 5.0 without this feature ? We may

I probably expressed myself badly. This is currently unsupported in the

module. I am just saying that even if we release a 5.0 with this regression, a simple like in xwiki.cfg will put back the old right

for those who need this. First, only XEM is impacted, second only user

this kind of delegation need that. So, this is not the general case IMO.

See my reply to Ludovic for more about how we need to work on that

feature. > > JV.