A breakout session on access control for WebDAV was held on Wednesday, December 9, 1998. This was not a official meeting of the WebDAV working group. However, since the discussions are of general interest to working group members, these minutes were taken to record the discussion. Jim Whitehead recorded the minutes, and attempted to mediate floor control during the session. In attendance during the breakout session were: Jim Amsden, Derek Atkins, Ken Coar, Yaron Goland, Paul Hill, Alex Hopmann, Manoj Kasichainula, Rohit Khare, Paul Leach, Lisa Lippert, Larry Masinter, Judy Slein, Jim Whitehead.
SCOPE OF EFFORT: ACCESS CONTROL FOR WEB OR WEBDAV?
The first discussion concerned the scope of the access control effort. It was noted that the web in general, not just WebDAV, has a need for a protocol for access control. It was recommended that WebDAV access control develop a model document, so that WebDAV access control can be applied to other domains/protocols more easily. The model document can separate out those aspects of the access control protocol which are not specifically tied to WebDAV.
** The attendees agreed that the access control effort should attempt HTTP access control, as well as WebDAV access control, unless it is proven to be too difficult to do both. That is, the protocol should use WebDAV mechanisms to set and read access control, but access control should be applicable to HTTP operations in general, and not just limited to WebDAV operations.
INTENDED USER OF THE PROTOCOL: ADMINISTRATORS OR AUTHORS?
There was some discussion over who is intended to be the user of this protocol. Is the access control protocol intended for users who will be performing authoring, or is it intended for use by system administrators, who are interested in access control across an entire site.
The topic of inheritance was discussed. It was noted that the access control inheritance within WebDAV is WebDAV-specific, since it is tied to WebDAV collections. However, Calendaring and Scheduling has a different inheritance model. A participant raised the question of whether we should support multiple kinds of inheritance. Today, servers support multiple kinds of inheritance, and hence it would be desirable for the protocol to accomodate these servers. However, the complexity of the user interface on the client is an important issue. Supporting multiple inheritance models would make the user interface on the client more complex, and, some argued, too complex.
There was a suggestion to examine the Java Web Server for lessons which can be learned from its access control model, which is object-oriented.
WEB-FORM MODEL FOR ACCESS CONTROL:
A short-term solution for access control is to have a property recorded on each resource which records a link pointing to a Web page which contains an access control form. By manipulating the form using a Web browser, a user can modify the access control for the resource. This approach has the benefit that it can accomodate a wide range of repositories, and each repository can have an access control user interface which is tailored to the repository. This approach can be specified and fielded quickly.
The most significant drawback to this approach its lack of support for a programatic interface to access control. Furthermore, users will be required to understand the access control mechanism for each server they author against. If the goal is to provide access control for the Web, and not just for WebDAV, the Web-form approach has the drawback that it depends on WebDAV properties, which are not understood by downlevel Web clients. However, if these clients discover the Web-form page without reading a property, they could manipulate the Web-form. Finally, the Web-form approach provides only a coarse degree of interoperability.
However, it was noted that once a protocol for access control was finished, the access control property might still be useful as an "escape hatch" for more sophisticated access control capabilities. However, if this were done, it would be necessary to call out which functionality can be access via the Web-form interface, and which can be accessed programatically.
Yaron Goland volunteered to write up a draft specification for the Web-form approach.
Groups were felt to be out of scope, since they are a (LDAP) directory concept. However, the Java Web Server supports a notion of realms (HTTP realms).
MAPPING OF ACCESS CONTROL PROTOCOL TO REPOSITORY ACCESS CONTROL:
Access control for resources which are accessible via multiple protocols, such as via HTTP and FTP, directly leads to this issue. For resources which are accessible via multiple protocols, access control via only one protocol is not a viable security model. One way to address this problem is to ensure that access control set via one protocol actually sets the access control for the underlying repository where the data is stored. In this way, access control set via one protocol will apply to all other protocols used to access the same resource.
However, it is a difficult problem to map access control rights in the protocol into the appropriate rights in the underlying repository. The difficulty arises from the case where there is not a 1:1 mapping of rights from protocol to repository, which either leads to unpredictable behavior (rejecting a request), or the inadvertent opening of security holes (performing the request has unintended side-effects). This concern applies to all network protocols, and is not specific to the Web, or WebDAV.
The following "database cell" scenario was discussed during the breakout session, and highlights this issue. Assume there is a database which is being accessed via WebDAV. Each cell in the database contains a PUT body and WebDAV properties. The database has a very simple "1 write bit" access control mechanism - when the write bit is "1", you can write to the entire cell, and when the write bit is "0", you cannot write to any of the cell's contents. Assume the WebDAV access control protocol has separate access rights for "write properties" and "write (PUT) body".
The problem arises when trying to map the "write properties" access control right into the rights supported by the repository. If the client submits a protocol request to grant it the "write properties" right, the repository has two choices:
There are two general approaches for handling this issue:
There was agreement that setting an access control right should not result in unintended consequences, that is, consequences which are not safe with respect to the underlying semantics of the store.
There was general agreement with the principle: in the event of a conflict between the semantics of the underlying repository's access control capabilities, and the semantics of a particular access control protocol request, the underlying repository semantics should always take precedence. There was general recognition that some access control protocol requests might not be mappable into the access control operations supported by a given repository.
There was agreement that the mapping issue needs to be recorded - these minutes are a start at that, though it should also be explicitly discussed in the protocol document, and probably also the goals document.
There was agreement that it would help discussion to know the access control semantics for several repositories.
Finally, it was noted that IMAP access control should be examined for lessons learned. It's approach is similar to #2 above in that the client first performs discovery, which returns sets of methods whose access control can be modified as a set.
PROBLEMS WITH METHOD-BASED ACCESS CONTROL
There was discussion at the meeting on the use of method-based access control, that is, access control where the access rights map exactly to HTTP methods. Several participants at the breakout felt this approach is a bad idea.
This approach has the benefit that it maps exactly onto HTTP methods, which have specific semantics. Thus the exact outcome of setting a given access control right is well understood by client and server. A protocol which adopts this approach would likely be more simple to specify and implement than other alternatives.
The drawbacks to this approach become evident when examining resources which are accessible via more than one protocol. In this case, HTTP methods might not map exactly onto operations available in other protocols. Furthermore, HTTP methods might not map exactly onto the access control rights provided by the underlying repository. Due to this, clients would need to perform a trial-and-error style of interaction when trying to set access control permissions, since presumably the underlying store would reject attempts to set access control permissions which would result in security risks due to unintended side-effects. The other drawback to method-based access control is the difficulty of making a user-interface for the protocol. The majority of users are not aware of the methods in the HTTP protocol, and hence would be unable to set access control correctly using a user interface which exposes the HTTP methods.
Most, but not all, participants at the breakout session felt that an approach which defines "semantic" access rights (e.g., "modify", "read", etc.) and then maps them into methods (e.g., "modify" -> PUT, POST, PROPPATCH) is better than a method-only approach.
NEW WORKING GROUP FOR APPLICATION LAYER PROTOCOL ACCESS CONTROL:
There was some discussion on whether it would be better to have a group working just on Web and WebDAV access control, or have a group which tries to address access control across several application layer protocols. The advantage to addressing access control for several protocols simultaneously is that the cross-protocol access control issue would be substantively addressed. This is important, because many network resources are accessible via multiple protocols, such as HTTP, FTP, news, mail, etc. This would benefit administrators of servers for these protocols. It would also tend to build critical mass for new application layer protocols to use this access control protocol.
The disadvantage to having a cross-protocol access control effort is its scope. It may prove to be too complicated to address the access control issues of several protocols simultaneously.
A cross-protocol access control effort would be expected to produce the following deliverables:
*** End of Breakout Session ***