[13] RFR: 8020637: Permissions.readObject doesn't enforce proper Class to PermissionCollection mappings

[Weijun Wang]

 603         // check that the Class of the Permission key and value are the same
 604         for (Map.Entry<Permission, Permission> e : perms.entrySet()) {
 605             Permission k = e.getKey();
 606             Permission v = e.getValue();
 607             if (!(k.getClass().equals(v.getClass()))) {
 608                 throw new InvalidObjectException("Permission with " +
 609                     k.getClass() + " incorrectly mapped to Permission with " +
 610                     v.getClass());
 611             }
 612         }

Why not check k == v? If they were the same object at serialization, then the deserialization process should be able to assign the same object to them.

--Max



> On Apr 6, 2019, at 1:30 AM, Sean Mullan <sean.mullan at oracle.com> wrote:
> 
> Updated webrev: http://cr.openjdk.java.net/~mullan/webrevs/8020637/webrev.01/
> 
> The serialized streams are now encoded within the test code itself. I also added a test case for an PermissionsHash object with invalid mappings.
> 
> I also modified the fix. Instead of trying to fix the mappings in the encoded serialized form, readObject simply throws an InvalidObjectException. I believe that is the more correct behavior, as the deserialized form no longer adheres to the invariants of the serialized structures of the corresponding classes.
> 
> Thanks,
> Sean
> 
> On 4/2/19 10:03 AM, Weijun Wang wrote:
>> +1.
>> --Max
>>> On Apr 2, 2019, at 9:55 PM, Roger Riggs <Roger.Riggs at oracle.com> wrote:
>>> 
>>> Hi Sean,
>>> 
>>> Typically, fixed serialization streams are encoded in the source
>>> as byte arrays. That keeps binary content out of the repo
>>> and provides a place for the comments.
>>> 
>>> Roger
>>> 
>>> 
>>> On 04/02/2019 09:50 AM, Sean Mullan wrote:
>>>> On 4/2/19 9:44 AM, Weijun Wang wrote:
>>>>> 
>>>>> 
>>>>>> On Apr 2, 2019, at 9:33 PM, Sean Mullan <sean.mullan at oracle.com> wrote:
>>>>>> 
>>>>>> On 4/1/19 11:12 PM, Weijun Wang wrote:
>>>>>>> I can understand the change in Permissions, but is there any difference in PermissionsHash?
>>>>>> 
>>>>>> The key and value in the PermissionsHash map is always the same object. This fix ensures that is respected, otherwise after deserialization you could have a SocketPermission mapped to a FilePermission, for example. Would it be better if I added a test for that?
>>>>> 
>>>>> Yes, you are right. I thought the old code can also enforce this relation.
>>>>> 
>>>>> Now for the test, perms.ser is binary and you haven't described how it is generated.
>>>> 
>>>> I just hacked Permissions.writeObject to switch the mappings. That was a lot easier than trying to write my own serialization code. I will add some comments in the test explaining how I did that and what is in perms.ser.
>>>> 
>>>> --Sean
>>>> 
>>>> 
>>>>> 
>>>>> Thanks,
>>>>> Max
>>>>> 
>>>>>> 
>>>>>> --Sean
>>>>>> 
>>>>>>> --Max
>>>>>>>> On Apr 2, 2019, at 1:10 AM, Sean Mullan <sean.mullan at oracle.com> wrote:
>>>>>>>> 
>>>>>>>> It is currently possible to change the mappings in a serialized java.security.Permissions object such that they no longer map correctly, and Permissions.readObject won't detect this.
>>>>>>>> 
>>>>>>>> This change makes sure that for a deserialized Permissions object, the permissions are mapped correctly to the class that they belong to. It does this by calling add() again for each permission in the deserialized Permissions object. The same technique was applied to a serialized PermissionsHash object which is used to store Permissions that don't implement their own PermissionCollection.
>>>>>>>> 
>>>>>>>> bug: https://bugs.openjdk.java.net/browse/JDK-8020637
>>>>>>>> webrev: http://cr.openjdk.java.net/~mullan/webrevs/8020637/webrev.00/
>>>>>>>> 
>>>>>>>> Thanks,
>>>>>>>> Sean
>>>>>>>> 
>>>>> 
>>>