Hi Angela, thanks again for your input. I finally got around to working on this again.
On a single level, I made our permission concept work by making sure that deny enries are always sorted before grant entries in our ACLs. I've also configured our repositories to use SimpleWorkspaceAccessManager, which simply grants workspace access to all validated users (we have a custom login module anyway which only allows access for specific users). I'm still struggling with inherited dependencies though. Perhaps I'm misunderstanding something, but the way I see it, permissions are evaluated either 1) working downwards from the root directory to the file in question or 2) upwards from the relevant file to the root directory. In case 1), a privilege denied to a user on one level would be denied to the same user for all sub levels as well. In case 2), sub levels could override the permissions by granting the user the privilege again, i.e. users could be denied write access on the root folder and then granted write access on their own home folder. I've written a couple of test cases, and neither option really seems to apply. I'll briefly describe my test cases below; all of them consist of a simple structure with a main folder "main", a sub folder "sub" and a file "file" in the sub folder. I'm setting permissions only on folders (we generally don't use ACLs on files, as we've had issues with versioning). I'm setting up the test cases as an admin user and then verifying the permission as User A. User A belongs to the groups A_AND_B and ONLY_A. All users belong to the group everyone (which maps to the EveryonePrinciple). Test A: Deny "all" to everyone on main, grant "all" to A_AND_B on sub. User cannot read main, but can read sub and file, and can write new content to file. This is consistent with option 2 described above. Test B: Grant "write" to everyone on main, deny "write" to A_AND_B on sub. User can read all files andfolders, but cannot write new content to file. Also consistent with option 2. Test C: Deny "read" and "write" to ONLY_A on main, grant "read" and "write" to everyone on sub. User cannot read main, which is still consistent with option 2, but he can also not read sub and file, or write content to the file. And this doesn't seem to fit any of the options I described. Curiously when I change the deny to A_AND_B and the grant to ONLY_A, everything works as expected again. Which could indicate that the order in which a user's groups are listed (which is undefined in our case) has an influence on the effective permissions. Is there something else I'm missing here? Thx, Marian. -----Original Message----- From: Angela Schreiber [mailto:anch...@adobe.com] Sent: Mittwoch, 13. Februar 2013 16:19 To: users@jackrabbit.apache.org Subject: Re: How does Jackrabbit resolve ACL permissions? hi marian the 401 is most probably rather a result on how access to workspaces is evaluated that a permission evaluation problem on the content itself. in other words: i was testing with our default setup that keeps users in each workspace (workspace access is granted if the user exists), while i would assume you are having the default jackrabbit setup with the DefaultSecurityManager. unless you change the workspaceaccessmgr configuration you will get a default that makes workspace access depending on accessibility of the root node (see DefaultSecurityManager line 659). if that's the case you can adjust the default configuration by something that better fits your needs regarding accessibility of workspaces... something like: <SecurityManager class="org.apache.jackrabbit.core.DefaultSecurityManager" workspaceName="security"> <WorkspaceAccessManager class="..."/> </SecurityManager> the most trivial implementation of the workspace-access-mgr just allows access to all workspace (there is such an implementation somewhere in jackrabbit core). alternatively you may want to create one that specifically fits your needs. in general i find the UserPerWorkspaceSecurityManager more intuitive than what is the default in jackrabbit core... despite the terrible name :-) kind regards angela On 2/13/13 3:51 PM, SCHEDENIG Marian wrote: > Hi Angela, > > thanks for the quick reply. I had a bug in my test code and could indeed get > it working now with multiple grant/denies on the same node, as long as I make > sure to put the grants before the denies. > > I still can't reproduce your suggested example though (Miroslav's fix taken > into account). By default I do indeed provide ALL permissions to "everyone" > on root, as otherwise (i.e. if I remove that ACE at the beginning of my test > case), non-admins cannot access the repository content at all. In fact (I'm > doing all my access through WebDAV), I get a 401 (authentication required) > from the repository if I don't explicitly grant permissions on the root > folder. And that goes for a subfolder with granted ALL as well: No root > permissions, no permissions anywhere. > > Not sure if perhaps I'm doing something wrong there. But solution 2 (deny > rights to everyone and grant them to a certain group per relevant folder) > should be good enough for us. > > Cheers, > Marian. > > -----Original Message----- > From: Angela Schreiber [mailto:anch...@adobe.com] > Sent: Mittwoch, 13. Februar 2013 13:30 > To: users@jackrabbit.apache.org > Subject: Re: How does Jackrabbit resolve ACL permissions? > > hi marian > > imo there shouldn't be any major obstacles in setting up the ACL to reflect > the permissions as you describe below. > in quickly tried it out on the crx-explorer using the following > setup: > > groups > --------------------------------------------------------------- > - groupA > - groupB > > users > --------------------------------------------------------------- > - userA: member of groupA (and everyone) > - userB: member of groupB (and everyone) > - userC: member of groupA and groupB (and everyone) > > acl setup > --------------------------------------------------------------- > + root > + a > + rep:policy > + allow > - jcr:primaryType = rep:GrantACE > - rep:principalName = groupA > - rep:privileges = [jcr:read] > + b > + rep:policy > + allow > - jcr:primaryType = rep:DenyACE > - rep:principalName = groupB > - rep:privileges = [jcr:read] > > result > --------------------------------------------------------------- > > - userA can read /a but not /b > - userB can read /b but not /a > - userC can read /a and /b > > additional adding an DENY ace for everyone on the root is redundant and > doesn't not have an effect on the result. > > > general notes > --------------------------------------------------------------- > > - ACEs are inherited through the node hierarchy. ACEs defined on > a particular node take precedence over inherited onces. > defining addition restrictions may be used to limit the effect to > certain parts of the subtree defined by the access controlled node > > - as long as ACEs are defined from group principals the evaluation > is strictly hierarchical. on a single ACL the order of ACEs matters. > > - if you define ACEs for non-group principal they will take predecence > in any case: over the group principals and over the inheritance rule > defined above. > > regarding your comments below: > --------------------------------------------------------------- > > 1) that works for me... see above. in don't think you analysis > matches the way the permissions are evaluated. > 2) that would work as well but the ACE for everyone is redundant. > it would not work if you would allow group A first and deny everyone > group after that... as the ACE for A would become redundant. > > hope that helps > angela > > > On 2/13/13 11:34 AM, SCHEDENIG Marian wrote: >> Hi, >> >> we're using the standard ACLProvider for permission handling. We're >> now running into problems when trying to set up slightly more complex >> ACLs than we've used so far: >> >> Say we have three groups, "everyone" (which is Jackrabbit's >> EveryonePrincipal) and "A" and "B". We want to allow only users in >> the A group to be able to access the folder /a_folder and only >> members of B to access /b_folder. A user may be a member of A, B, A >> and B or of neither group. If user X is a member of A and not a >> member of B, X should still have access to /a_folder. >> >> We've tried two approaches: >> >> 1. Deny full permissions to "everyone" on the root folder and then >> grant full permissions to A on /a_folder and to B on /b_folder. This >> fails, apparently because permissions are resolved in a "top down" >> manner, and as soon as it has been established that a user doesn't >> have access to a parent folder, its subfolders are no longer >> evaluated. That's fine, if we can find a different way to do it. >> >> 2. Deny full permissions to "everyone" on /a_folder and grant full >> permissions to A on the same folder (and the same with "everyone" and >> B on /b_folder). This also fails, although apparently it works for >> user X if we deny "everyone" and grant X (specifically the user) on the >> folder. >> >> I'm now wondering: How exactly does Jackrabbit resolve permissions? >> Case >> 1 seems to be clear, but what are the exact rules for overlapping >> grants and denies on the same resource? And what is the correct way >> to solve our requirement? >> >> Thanks, >> >> Marian. >> >> -- >> >> *DI Marian Schedenig* >> >> Senior Developer >> >> Description: Description: Description: Description: Description: >> Description: cid:image001.png@01CCBE64.F3314040 >> >> INFINICA - Member of Qualysoft Group** >> >> Leonard-Bernstein-Straße 10 >> >> A-1220 Wien >> >> Österreich >> >> Tel +43 1 4095987-26 >> >> Fax +43 1 4095987-11 >> >> www.infinica.at<http://www.infinica.at/> >> >> www.qualysoft.at<http://www.qualysoft.at/> >> >> marian.schede...@infinica.at<mailto:marian.schede...@infinica.at> >> >> *P**Please consider the environment before printing this email* >>
