Costin Manolache wrote: > I'm in the middle on this one - but I would vote for > Jean-Francois proposal. > > Glen is right, it is possible to restrict individual managers > using the policy. > > However as geenral programming it is better to keep the > doPrivileged block as small as possible - and have each > individual manager that needs to change the permission context > do that for the specific areas where this is needed, instead > of a global aproach.
In general I agree, keeping the amount of code within a doPrivileged() block to a minimum is a good practice. That makes it less likely that the code which calls a method which uses doPrivileged can compromise security. That is not the case for getSession() where the only thing passed to the method is a boolean for create and the Manager gets the JSESSIONID cookie from the request. Removing doPrivileged() from where it currently is will force alot of other work to be done. And will then require anyone who implements a custom Manager/Store to wrap their code in doPrivileged() also. I don't see any security risks from the current implemenation, so why change it? If you can show me an exploit that this change would fix I would be all for it. > I also thing that the permissions should be granted on apps, > not managers. The manager should check and have control over > the operation that it's doing. ( for example give only some applications > permissions to serialize the session in a file - that's probably > a bad example as using security manager for that is not the best > implementation, but I think you get my point ). > Persisting session data is the responsibility of the container not the web application. Session management/persistence should be completely transparent to the webapp including security policy permissions required for managing/persisting those sessions. > Costin > > > Jean-Francois Arcand wrote: > > >> >>Glenn Nielsen wrote: >> >> >>>The doPrivileged() for getting a session is in the CoyoteRequest >>>public getSession() >>>method which is implemented as required by ServletRequest and >>>HttpServletRequest. >> >>I'm unable to find the information in the spec. Can you point me to the >>section where the discuss the required privilege? My understanding is >>the doPrivilege block is needed only if you want to serialize/store the >>session. If you don't, then you don't need it. If you run Tomcat as it >>is right now (with the security manager), then you don't need the extra >>call. >> >> >>> >>>The getSession() can be called by a web application. The >>>doPrivileged() block would >>>be required to exist either where it currently is or in each >>>individual implementation >>>of Manager/Store. If it wasn't there getting a session would fail if >>>the web app >>>were not granted the necessary permissions. Permissions I would not >>>want to grant to it. >> >>Not in the current implementation. Since the default Manager do not need >>any special permission, the doPrivilege block is not required. >> >> >>> >>>IMHO it is best left where it is. If someone were to implement a >>>custom Manager/Store >>>then the permissions allowed at that point would be the intersection >>>of those permissions >>>granted to catalina and those granted to the codeBase for the custom >>>Manager/Store. >>>So you still have your fine grained control of security policies. >>>That is how it should work. >> >>Well, every Manager have a large set of permissions granted by default. >>I personnaly prefer limitting the permission set instead of granting >>everything (like the current implementation). Also, the doPrivilege >>block (as implemented right now) do not contains anything that require >>special permissions. The doPrivilege block is a performance penalty, and >>the block is way too big ( for security reasons ). See >>http://java.sun.com/products/jdk/1.2/docs/guide/security/doprivileged.html >>for more info. >> >>Right now, every Manager are granted by default whatever they want. If I >>want to denied ManagerA file permissions, I have no way to do it right >>now...right? >> >> >>> >>>-1 for changing/removing the doPrivileged() >> >>Other voices? >> >> >>> >>>Regards, >>> >>>Glenn >> >> >>Thanks, >> >>-- Jeanfrancois >> >> >>> >>>Jean-Francois Arcand wrote: >>> >>> >>>>Hi, >>>> >>>>In o.a.c.tomcat5.CoyoteRequest (same in tomcat4), there is a >>>>doPrivilege block that grant the doGetSession method. This method >>>>delegate the logic to the o.a.c.Manager instance. A Manager can (but >>>>it's not required) uses a o.a.c.Store object . The Manager and the >>>>Store object may need special privileges when handling session >>>>persistance (see o.a.c.session.FileStore for an example). >>>> >>>>I would recommend we remove the doPrivilege block from CoyoteRequest >>>>and delegate the doPrivilege call to the Manager (or the Store) >>>>instance. That will allow better fine grained security check. Only >>>>the required operations should be granted (right now every Manager is >>>>granted -> so every Store instance!). As an example, >>>>o.a.c.session.FileStore does not contains any security checks in its >>>>current implementation, and IMO, it should. >>>> >>>>The contract between the Manager and CoyoteRequest will have to be >>>>documented somewhere since Manager written for Tomcat 4 may no longer >>>>works. The catalina.policy file can then be used to give special >>>>privileges to ManagerX, but not to ManagerY (same for Store instance >>>>or whatever objects is used), based on codebase. >>>> >>>>Any recommendations/objections to the modification? >>>> >>>>Thanks, >>>> ---------------------------------------------------------------------- Glenn Nielsen [EMAIL PROTECTED] | /* Spelin donut madder | MOREnet System Programming | * if iz ina coment. | Missouri Research and Education Network | */ | ---------------------------------------------------------------------- -- To unsubscribe, e-mail: <mailto:[EMAIL PROTECTED]> For additional commands, e-mail: <mailto:[EMAIL PROTECTED]>
