INTERNET-DRAFT David Boreham, Netscape Jim Sermersheim, Novell Anoop Anantha, Microsoft Michael Armijo, Microsoft ldapext Working Group 6 April, 2000 LDAP Extensions for Scrolling View Browsing of Search Results draft-ietf-ldapext-ldapv3-vlv-04.txt This document expires on 5 October 2000 1. Status of this Memo This document is an Internet-Draft and is in full conformance with all provisions of Section 10 of RFC2026. Internet-Drafts are working docu- ments of the Internet Engineering Task Force (IETF), its areas, and its working groups. Note that other groups may also distribute working documents as Internet-Drafts. Internet-Drafts are draft documents valid for a maximum of six months and may be updated, replaced, or obsoleted by other documents at any time. It is inappropriate to use Internet- Drafts as reference material or to cite them other than as "work in progress." The list of current Internet-Drafts can be accessed at http://www.ietf.org/ietf/1id-abstracts.txt The list of Internet-Draft Shadow Directories can be accessed at http://www.ietf.org/shadow.html. 2. Abstract This document describes a Virtual List View control extension for the LDAP Search operation. This control is designed to allow the "virtual list box" feature, common in existing commercial e-mail address book applications, to be supported efficiently by LDAP servers. LDAP servers' inability to support this client feature is a significant impediment to LDAP replacing proprietary protocols in commercial e-mail systems. The control allows a client to specify that the server return, for a given LDAP search with associated sort keys, a contiguous subset of the search result set. This subset is specified in terms of offsets into the ordered list, or in terms of a greater than or equal comparison value. 3. Background A Virtual List is a graphical user interface technique employed where Boreham et al [Page 1] RFC DRAFT April 2000 ordered lists containing a large number of entries need to be displayed. A window containing a small number of visible list entries is drawn. The visible portion of the list may be relocated to different points within the list by means of user input. This input can be to a scroll bar slider; from cursor keys; from page up/down keys; from alphanumeric keys for "typedown". The user is given the impression that they may browse the complete list at will, even though it may contain millions of entries. It is the fact that the complete list contents are never required at any one time that characterizes Virtual List View. Rather than fetch the complete list from wherever it is stored (typically from disk or a remote server), only that information which is required to display the part of the list currently in view is fetched. The subject of this document is the interaction between client and server required to implement this functionality in the context of the results from a sorted LDAP search request. For example, suppose an e-mail address book application displays a list view onto the list containing the names of all the holders of e-mail accounts at a large university. The list is sorted alphabetically. While there may be tens of thousands of entries in this list, the address book list view displays only 20 such accounts at any one time. The list has an accompanying scroll bar and text input window for type- down. When first displayed, the list view shows the first 20 entries in the list, and the scroll bar slider is positioned at the top of its range. Should the user drag the slider to the bottom of its range, the displayed contents of the list view should be updated to show the last 20 entries in the list. Similarly, if the slider is positioned somewhere in the middle of its travel, the displayed contents of the list view should be updated to contain the 20 entries located at that relative position within the complete list. Starting from any display point, if the user uses the cursor keys or clicks on the scroll bar to request that the list be scrolled up or down by one entry, the displayed con- tents should be updated to reflect this. Similarly the list should be displayed correctly when the user requests a page scroll up or down. Finally, when the user types characters in the type-down window, the displayed contents of the list should "jump" or "seek" to the appropri- ate point within the list. For example, if the user types "B", the displayed list could center around the first user with a name beginning with the letter "B". When this happens, the scroll bar slider should also be updated to reflect the new relative location within the list. This document defines a request control which extends the LDAP search operation. Always used in conjunction with the server side sorting control[SSS], this allows a client to retrieve selected portions of large search result set in a fashion suitable for the implementation of a virtual list view. The key words "MUST", "SHOULD", and "MAY" used in this document are to Boreham et al [Page 2] RFC DRAFT April 2000 be interpreted as described in [Bradner97]. 4. Client-Server Interaction The Virtual List View control extends a regular LDAP Search operation which must also include a server-side sorting control[SSS]. Rather than returning the complete set of appropriate SearchResultEntry messages, the server is instructed to return a contiguous subset of those entries, taken from the sorted result set, centered around a particular target entry. Henceforth, in the interests of brevity, the sorted search result set will be referred to as "the list". The sort control MAY contain any sort specification valid for the server. The attributeType field in the first SortKeyList sequence ele- ment has special significance for "typedown". The desired target entry, and the number of entries to be returned both before, and after, that target entry in the list, are determined by the client's VirtualListViewRequest control. When the server returns the set of entries to the client, it attaches a VirtualListViewResponse control to the SearchResultDone message. The server returns in this control: its current estimate for the list con- tent count, the location within the list corresponding to the target entry, and any error codes. The target entry is specified in the VirtualListViewRequest control by one of two methods. The first method is for the client to indicate the target entry's offset within the list. The second way is for the client to supply an attribute assertion value. The value is compared against the values of the attribute specified as the primary sort key in the sort control attached to the search operation. The first sort key in the SortKeyList is the primary sort key. The target entry is the first entry in the list with value greater than or equal to (in the primary sort order), the presented value. The order is determined by rules defined in [SSS]. Selection of the target entry by this means is designed to implement "typedown". Note that it is possible that no entry satisfies these conditions, in which case there is no target entry. This condition is indicated by the server returning the special value contentCount + 1 in the target position field. Because the server may not have an accurate estimate of the number of entries in the list, and to take account of cases where the list size is changing during the time the user browses the list, and because the client needs a way to indicate specific list targets "beginning" and "end", offsets within the list are transmitted between client and server as ratios---offset to content count. The server sends its latest esti- mate as to the number of entries in the list (content count) to the Boreham et al [Page 3] RFC DRAFT April 2000 client in every response control. The client sends its assumed value for the content count in every request control. The server examines the content count and offsets presented by the client and computes the corresponding offsets within the list, based on its own idea of the con- tent count. Si = Sc * (Ci / Cc) Where: Si is the actual list offset used by the server Sc is the server's estimate for content count Ci is the client's submitted offset Cc is the client's submitted content count The result is rounded to the nearest integer. If the content count is stable, and the client returns to the server the content count most recently received, Cc = Sc and the offsets transmit- ted become the actual server list offsets. The following special cases are allowed: a client sending a content count of zero (Cc = 0) means "client has no idea what the content count is, server MUST use its own content count estimate in place of the client's". An offset value of one (Ci = 1) always means that the target is the first entry in the list. Client specifying an offset which equals the content count specified in the same request control (Ci = Cc) means that the target is the last entry in the list. Ci may only equal zero when Cc is also zero. This signifies the last entry in the list. Because the server always returns contentCount and targetPosition, the client can always determine which of the returned entries is the target entry. Where the number of entries returned is the same as the number requested, the client is able to identify the target by simple arith- metic. Where the number of entries returned is not the same as the number requested (because the requested range crosses the beginning or end of the list, or both), the client must use the target position and content count values returned by the server to identify the target entry. For example, suppose that 10 entries before and 10 after the tar- get were requested, but the server returns 13 entries, a content count of 100 and a target position of 3. The client can determine that the first entry must be entry number 1 in the list, therefore the 13 entries returned are the first 13 entries in the list, and the target is the third one. A server-generated context identifier MAY be returned to clients. A client receiving a context identifier SHOULD return it unchanged in a subsequent request which relates to the same list. The purpose of this interaction is to enhance the performance and effectiveness of servers which employ approximate positioning. Boreham et al [Page 4] RFC DRAFT April 2000 5. The Controls Support for the virtual list view control extension is indicated by the presence of the OID "2.16.840.1.113730.3.4.9" in the supportedControl attribute of a server's root DSE. 5.1. Request Control This control is included in the SearchRequest message as part of the controls field of the LDAPMessage, as defined in Section 4.1.12 of [LDAPv3]. The controlType is set to "2.16.840.1.113730.3.4.9". The cri- ticality SHOULD be set to TRUE. If this control is included in a Sear- chRequest message, a Server Side Sorting request control [SSS] MUST also be present in the message. The controlValue is an OCTET STRING whose value is the BER-encoding of the following SEQUENCE: VirtualListViewRequest ::= SEQUENCE { beforeCount INTEGER (0..maxInt), afterCount INTEGER (0..maxInt), CHOICE { byoffset [0] SEQUENCE { offset INTEGER (0 .. maxInt), contentCount INTEGER (0 .. maxInt) }, greaterThanOrEqual [1] AssertionValue }, contextID OCTET STRING OPTIONAL } beforeCount indicates how many entries before the target entry the client wants the server to send. afterCount indicates the number of entries after the target entry the client wants the server to send. offset and contentCount identify the target entry as detailed in section 4. greaterThanOrEqual is an attribute assertion value defined in [LDAPv3]. If present, the value supplied in greaterThanOrEqual is used to determine the target entry by comparison with the values of the attribute specified as the primary sort key. The first list entry who's value is no less than (less than or equal to when the sort order is reversed) the supplied value is the target entry. If present, the con- textID field contains the value of the most recently received contextID field from a VirtualListViewResponse control. The type AssertionValue and value maxInt are defined in [LDAPv3]. contextID values have no validity outwith the connection on which they were received. That is, a client should not submit a contextID which it received from another con- nection, a connection now closed, or a different server. 5.2. Response Control This control is included in the SearchResultDone message as part of the controls field of the LDAPMessage, as defined in Section 4.1.12 of Boreham et al [Page 5] RFC DRAFT April 2000 [LDAPv3]. The controlType is set to "2.16.840.1.113730.3.4.10". The criticality is FALSE (MAY be absent). The controlValue is an OCTET STRING, whose value is the BER encoding of a value of the following SEQUENCE: VirtualListViewResponse ::= SEQUENCE { targetPosition INTEGER (0 .. maxInt), contentCount INTEGER (0 .. maxInt), virtualListViewResult ENUMERATED { success (0), operationsError (1), unwillingToPerform (53), insufficientAccessRights (50), busy (51), timeLimitExceeded (3), adminLimitExceeded (11), sortControlMissing (60), offsetRangeError (61), other (80) }, contextID OCTET STRING OPTIONAL } targetPosition gives the list offset for the target entry. contentCount gives the server's estimate of the current number of entries in the list. Together these give sufficient information for the client to update a list box slider position to match the newly retrieved entries and identify the target entry. The contentCount value returned SHOULD be used in a subsequent VirtualListViewRequest control. contextID is a server-defined octet string. If present, the contents of the contextID field SHOULD be returned to the server by a client in a subsequent Vir- tualListViewRequest control. The virtualListViewResult codes which are common to the LDAP sear- chResponse (adminLimitExceeded, timeLimitExceeded, busy, operationsEr- ror, unwillingToPerform, insufficientAccessRights) have the same mean- ings as defined in [LDAPv3], but they pertain specifically to the VLV operation. For example, the server could exceed an administration limit processing a SearchRequest with a VirtualListViewRequest control. How- ever, the same administration limit would not be exceeded should the same SearchRequest be submitted by the client without the VirtualList- ViewRequest control. In this case, the client can determine that an administration limit has been exceeded in servicing the VLV request, and can if it chooses resubmit the SearchRequest without the VirtualList- ViewRequest control. insufficientAccessRights means that the server denied the client permis- sion to perform the VLV operation. Boreham et al [Page 6] RFC DRAFT April 2000 If the server determines that the results of the search presented exceed the range provided by the 32-bit offset values, it MUST return offsetRangeError. 6. Protocol Example Here we walk through the client-server interaction for a specific vir- tual list view example: The task is to display a list of all 78564 peo- ple in the US company "Ace Industry". This will be done by creating a graphical user interface object to display the list contents, and by repeatedly sending different versions of the same virtual list view search request to the server. The list view displays 20 entries on the screen at a time. We form a search with baseDN "o=Ace Industry, c=us"; search scope sub- tree; filter "objectClass=inetOrgPerson". We attach a server sort order control to the search, specifying ascending sort on attribute "cn". To this base search, we attach a virtual list view request control with contents determined by the user activity and send the search to the server. We display the results from each search in the list window and update the slider position. When the list view is first displayed, we want to initialize the con- tents showing the beginning of the list. Therefore, we set beforeCount = 0, afterCount = 19, contentCount = 0, offset = 1 and send the request to the server. The server duly returns the first 20 entries in the list, plus the content count = 78564 and targetPosition = 1. We therefore leave the scroll bar slider at its current location (the top of its range). Say that next the user drags the scroll bar slider down to the bottom of its range. We now wish to display the last 20 entries in the list, so we set beforeCount = 19, afterCount = 0, contentCount = 78564, offset = 78564 and send the request to the server. The server returns the last 20 entries in the list, plus the content count = 78564 and targetPosition = 78564. Next the user presses a page up key. Our page size is 20, so we set beforeCount = 0, afterCount = 19, contentCount = 78564, offset = 78564-19-20 and send the request to the server. The server returns the preceding 20 entries in the list, plus the content count = 78564 and targetPosition = 78525. Now the user grabs the scroll bar slider and drags it to 68% of the way down its travel. 68% of 78564 is 53424 so we set beforeCount = 9, after- Count = 10, contentCount = 78564, offset = 53424 and send the request to the server. The server returns the preceding 20 entries in the list, plus the content count = 78564 and targetPosition = 53424. Boreham et al [Page 7] RFC DRAFT April 2000 Lastly, the user types the letter "B". We set beforeCount = 9, after- Count = 10 and greaterThanOrEqual = "B". The server finds the first entry in the list not less than "B", let's say "Babs Jensen", and returns the nine preceding entries, the target entry, and the proceeding 10 entries. The server returns content count = 78564 and targetPosition = 5234 and so the client updates its scroll bar slider to 6.7% of full scale. 7. Notes for Implementers While the feature is expected to be generally useful for arbitrary search and sort specifications, it is specifically designed for those cases where the result set is very large. The intention is that this feature be implemented efficiently by means of pre-computed indices per- taining to a set of specific cases. For example, an offset relating to "all the employees in the local organization, sorted by surname" would be a common case. The intention for client software is that the feature should fit easily with the host platform's graphical user interface facilities for the display of scrolling lists. Thus the task of the client implementers should be one of reformatting up the requests for information received from the list view code to match the format of the virtual list view request and response controls. Client implementers should note that any offset value returned by the server may be approximate. Do not design clients > which only operate correctly when offsets are exact. Server implementers using indexing technology which features approximate positioning should consider returning context identifiers to clients. The use of a context identifier will allow the server to distinguish between client requests which relate to different displayed lists on the client. Consequently the server can decide more intelligently whether to reposition an existing database cursor accurately to within a short dis- tance of its current position, or to reposition to an approximate posi- tion. Thus the client will see precise offsets for "short" repositioning (e.g. paging up or down), but approximate offsets for a "long" reposi- tion (e.g. a slider movement). Server implementers are free to return status code unwillingToPerform should their server be unable to service any particular VLV search. This might be because the resolution of the search is computationally infeasible, or because excessive server resources would be required to service the search. Client implementers should note that this control is only defined on a client interaction with a single server. If a server returns referrals Boreham et al [Page 8] RFC DRAFT April 2000 as a part of its response to the search request, the client is responsi- ble for deciding when and how to apply this control to the referred-to servers, and how to collate the results from multiple servers. 8. Relationship to "Simple Paged Results" These controls are designed to support the virtual list view, which has proved hard to implement with the Simple Paged Results mechanism [SPaged]. However, the controls described here support any operation possible with the Simple Paged Results mechanism. The two mechanisms are not complementary, rather one has a superset of the other's features. One area where the mechanism presented here is not a strict superset of the Simple Paged Results scheme is that here we require a sort order to be specified. No such requirement is made for paged results. 9. Security Considerations Server implementers may wish to consider whether clients are able to consume excessive server resources in requesting virtual list opera- tions. Access control to the feature itself; configuration options lim- iting the feature's use to certain predetermined search base DNs and filters; throttling mechanisms designed to limit the ability for one client to soak up server resources, may be appropriate. Consideration should be given as to whether a client will be able to retrieve the complete contents, or a significant subset of the complete contents of the directory using this feature. This may be undesirable in some circumstances and consequently it may be necessary to enforce some access control. Clients can, using this control, determine how many entries are con- tained within a portion of the DIT. This may constitute a security hazard. Again, access controls may be appropriate. Server implementers SHOULD exercise caution concerning the content of the contextID. Should the contextID contain internal server state, it may be possible for a malicious client to use that information to gain unauthorized access to information. 10. Acknowledgements Chris Weider of Microsoft co-authored a previous version of this docu- ment. Boreham et al [Page 9] RFC DRAFT April 2000 11. References [LDAPv3] Wahl, M, S. Kille and T. Howes, "Lightweight Directory Access Pro- tocol (v3)", Internet Standard, December, 1997. RFC2251. [SPaged] Weider, C, A. Herron, A. Anantha, and T. Howes, "LDAP Control Extension for Simple Paged Results Manipulation", September 1999. RFC2696 [SSS]Wahl, M, A. Herron and T. Howes, "LDAP Control Extension for Server Side Sorting of Search Results", Internet Draft, April, 1999. Available as draft-ietf-asid-ldapv3-sorting-02.txt. [Bradner97] Bradner, S., "Key Words for use in RFCs to Indicate Requirement Levels", BCP 14, RFC 2119, March 1997. 12. Authors' Addresses David Boreham iPlanet e-commerce solutions 501 E. Middlefield Road Mountain View, CA 94043, USA +1 650 937-5206 dboreham@netscape.com Jim Sermersheim Novell 122 East 1700 South Provo, Utah 84606, USA jimse@novell.com Anoop Anantha Microsoft Corp. 1 Microsoft Way Redmond, WA 98052, USA +1 425 882-8080 anoopa@microsoft.com Michael Armijo Microsoft Corp. 1 Microsoft Way Redmond, WA 98052, USA +1 425 882-8080 micharm@microsoft.com This document expires on 5 October 2000 Boreham et al [Page 10] RFC DRAFT April 2000 Boreham et al [Page 11]