Internet-Draft SCIM Cursor Pagination March 2024
Zollner & Sehgal Expires 2 September 2024 [Page]
Workgroup:
SCIM
Internet-Draft:
draft-ietf-scim-cursor-pagination-04
Published:
Intended Status:
Standards Track
Expires:
Authors:
D. Zollner, Ed.
Microsoft
A. Sehgal
Amazon Web Services

Cursor-based Pagination of SCIM Resources

Abstract

This document defines additional SCIM (System for Cross-Domain Identity Management) query parameters and result attributes to allow use of cursor-based pagination in SCIM implementations that are implemented with existing code bases, databases, or APIs where cursor-based pagination is already well established.

Discussion Venues

This note is to be removed before publishing as an RFC.

Discussion of this document takes place on the System for Cross-domain Identity Management Working Group mailing list (scim@ietf.org), which is archived at https://mailarchive.ietf.org/arch/browse/scim/.

Source for this draft and an issue tracker can be found at https://github.com/ietf-scim-wg/draft-ietf-scim-cursor-pagination.

Status of This Memo

This Internet-Draft is submitted in full conformance with the provisions of BCP 78 and BCP 79.

Internet-Drafts are working documents of the Internet Engineering Task Force (IETF). Note that other groups may also distribute working documents as Internet-Drafts. The list of current Internet-Drafts is at https://datatracker.ietf.org/drafts/current/.

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."

This Internet-Draft will expire on 2 September 2024.

▲

Table of Contents

1. Introduction

The two common patterns for result pagination are index-based pagination and cursor-based pagination. Rather than attempt to compare and contrast the advantages and disadvantages of competing pagination patterns, this document simply recognizes that SCIM service providers are commonly implemented as an interoperability layer on top of already existing application codebases, databases, and/or APIs that already have a well established pagination pattern.

Translating from an underlying cursor-based pagination pattern to the index-based pagination defined in Section 3.4.2.4 of [RFC7644] ultimately requires the SCIM service provider to fully iterate the underlying cursor, store the results, and then serve indexed pages from the stored results. This task of "pagination translation" dramatically increases complexity and memory requirements for implementing a SCIM Service Provider, and may be an impediment to SCIM adoption for some applications and identity systems.

This document defines a simple addition to the SCIM protocol that allows SCIM service providers to reuse underlying cursors without expensive translation. Support for cursor-based pagination in SCIM encourages broader cross-application identity management interoperability by encouraging SCIM service provider implementations for applications and identity systems where cursor-based pagination is already well-established.

1.1. Notational Conventions

The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "NOT RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in BCP 14 [RFC2119] [RFC8174] when, and only when, they appear in all capitals, as shown here.

2. Query Parameters and Response Attributes

The following table describes the URL pagination parameters for requesting cursor-based pagination:

Table 1: Query Parameters
Parameter Description
cursor The string value of the nextCursor attribute from a previous result page. The cursor value MUST be empty or omitted for the first request of a cursor-paginated query. This value may only contained characters from the unreserved characters set defined in section 2.2 of [RFC3986]
count A positive integer. Specifies the desired maximum number of query results per page, e.g., count=10. When specified, the service provider MUST NOT return more results than specified, although it MAY return fewer results. If count is not specified in the query, the maximum number of results is set by the service provider.

The following table describes cursor-based pagination attributes returned in a paged query response:

Table 2: Response Attributes
Element Description
nextCursor A cursor value string that MAY be used in a subsequent request to obtain the next page of results. Service providers supporting cursor-based pagination MUST include nextCursor in all paged query responses except when returning the last page. nextCursor is omitted from a response only to indicate that there are no more result pages.
previousCursor A cursor value string that MAY be used in a subsequent request to obtain the previous page of results. Returning previousCursor is OPTIONAL.

Cursor values are opaque; clients MUST not make assumptions about their structure. When the client wants to retrieve another result page for a query, it should query the same Service Provider endpoint with all query parameters and values being identical to the initial query with the exception of the cursor value which should be set to a nextCursor (or previousCursor) value that was returned by Service Provider in a previous response.

For example, to retrieve the first 10 Users with userName starting with J, use an empty cursor and set the count to 10:

GET /Users?filter=userName%20sw%20J&cursor&count=10
Host: example.com
Accept: application/scim+json
Authorization: Bearer U8YJcYYRMjbGeepD

The SCIM provider in response to the query above returns metadata regarding pagination similar to the following example (actual resources removed for brevity):

HTTP/1.1 200 OK
Content-Type: application/scim+json

{
   "totalResults":100,
   "itemsPerPage":10,
   "nextCursor":"VZUTiyhEQJ94IR",
   "schemas":["urn:ietf:params:scim:api:messages:2.0:ListResponse"],
   "Resources":[{
      ...
   }]
}

Given the example above, to request the next page or results, use the same query parameters and values except set the cursor to the value of nextCursor (VZUTiyhEQJ94IR):

GET /Users?filter=username%20sw%20J&cursor=VZUTiyhEQJ94IR&count=10
Host: example.com
Accept: application/scim+json
Authorization: Bearer U8YJcYYRMjbGeepD

HTTP/1.1 200 OK
Content-Type: application/scim+json

{
   "totalResults": 100,
   "itemsPerPage": 10,
   "previousCursor: "ze7L30kMiiLX6x"
   "nextCursor": "YkU3OF86Pz0rGv",
   "schemas": ["urn:ietf:params:scim:api:messages:2.0:ListResponse"],
   "Resources":[{
      ...
   }]
}

In the example above, the response includes the OPTIONAL previousCursor indicating that the Service Provider supports forward and reverse traversal of result pages.

As described in Section 3.4.1 of [RFC7644] Service Providers SHOULD return an accurate value for totalResults which is the total number of resources for all pages. Service Providers implementing cursor pagination that are unable to estimate totalResults MAY choose to omit the totalResults attribute.

2.1. Pagination errors

If a Service Provider encounters invalid pagination query parameters (invalid cursor value, count value, etc), or other error condition, the Service Provider SHOULD return the appropriate HTTP response status code and detailed JSON error response as defined in Section 3.12 of [RFC7644]. Most pagination error conditions would generate an HTTP response with status code 400. Since many pagination error conditions are not user recoverable, error messages SHOULD focus on communicating error details to the SCIM client developer.

For HTTP status code 400 (Bad Request) responses, the following detail error types are defined. These error types extend the list of error types defined in [RFC7644] Section 3.12, Table 9: SCIM Detail Error Keyword Values.

Table 3: Pagination Errors
scimType Description Applicability
invalidCursor Cursor value is invalid. Cursor value should be empty to request the first page and set to the nextCursor or previousCursor value for subsequent queries. GET (Section 3.4.2 of [RFC7644])
expiredCursor Cursor has expired. Do not wait longer than cursorTimeout (600 sec) to request additional pages. GET (Section 3.4.2 of [RFC7644])
invalidCount Count value is invalid. Count value must be between 1 - and maxPageSize (500) GET (Section 3.4.2 of [RFC7644])

2.2. Sorting

If sorting is implemented as described Section 3.4.2.3 of [RFC7644], then cursor-paged results SHOULD be sorted.

2.3. Cursors as the Only Pagination Method

A SCIM Service Provider MAY require cursor-based pagination to retrieve all results for a query by including a nextCursor value in the response even when the query does not include the cursor parameter.

For example:

GET /Users
Host: example.com
Accept: application/scim+json

The SCIM Service Provider may respond to the above query with a page containing defaultPageSize results and a nextCursor value as shown in the below example (Resources omitted for brevity):

HTTP/1.1 200 OK
Content-Type: application/scim+json

{
   "totalResults": 5000,
   "itemsPerPage": 100,
   "nextCursor": "HPq72Pax3JUaNa",
   "schemas": ["urn:ietf:params:scim:api:messages:2.0:ListResponse"],
   "Resources": [{
      ...
   }]
}

2.4. Backwards Compatibility Considerations

Implementers of SCIM service providers that previously supported index-based pagination and are adding support for cursor-based pagination should carefully consider the impact to existing SCIM clients before changing the default pagination method in a return set. SCIM clients that previously expected index-based pagination may not be compatible with cursor-based pagination without making changes to the SCIM client. Adding cursor-based pagination support but leaving the default return set pagination method as-is will not impact existing SCIM clients.

SCIM clients can query the provider configuration endpoint to determine if index-based, cursor-based or both types of pagination are supported.

3. Querying Resources using HTTP POST

Section 3.4.2.4 of [RFC7644] defines how clients MAY execute the HTTP POST method combined with the /.search path extension to issue execute queries without passing parameters on the URL. When using /.search, the client would pass the parameters defined in Section 2

POST /User/.search
Host: example.com
Accept: application/scim+json
Authorization: Bearer U8YJcYYRMjbGeepD

{
   "schemas": ["urn:ietf:params:scim:api:messages:2.0:SearchRequest"],
   "attributes": ["displayName", "userName"],
   "filter": "displayName sw \"smith\"",
   "cursor": "",
   "count": 10
}

Which would return a result containing a nextCursor value which may be used by the client in a subsequent call to return the next page of resources

HTTP/1.1 200 OK
Content-Type: application/scim+json

{
   "totalResults": 100,
   "itemsPerPage": 10,
   "nextCursor": "VZUTiyhEQJ94IR",
   "schemas": ["urn:ietf:params:scim:api:messages:2.0:ListResponse"],
   "Resources": [{
      ...
   }]
}

4. Service Provider Configuration

The /ServiceProviderConfig resource defined in Section 4 of [RFC7644] facilitates discovery of SCIM service provider features. A SCIM Service provider implementing cursor-based pagination SHOULD include the following additional attribute in JSON document returned by the /ServiceProviderConfig endpoint:

pagination

A complex type that indicates pagination configuration options. OPTIONAL.

cursor

A Boolean value specifying support of cursor-based pagination. REQUIRED.

index

A Boolean value specifying support of index-based pagination. REQUIRED.

defaultPageSize

Non-negative integer value specifying the default number of results returned in a page when a count is not specified in the query. OPTIONAL.

maxPageSize

Non-negative integer specifying the maximum number of results returned in a page regardless of what is specified for the count in a query. The maximum number of results returned may be further restricted by other criteria. OPTIONAL.

cursorTimeout

Non-negative integer specifying the maximum number seconds that a cursor is valid between page requests. Clients waiting too long between cursor pagination requests may receive an invalid cursor error response. No value being specified may mean that there is no cursor timeout or that the cursor timeout is not a static duration. OPTIONAL.

Before using cursor-based pagination, a SCIM client MAY fetch the Service Provider Configuration document from the SCIM service provider and verify that cursor-based pagination is supported.

For example:

GET /ServiceProviderConfig
Host: example.com
Accept: application/scim+json

A service provider supporting both cursor-based pagination and index- based pagination would return a document similar to the following (full ServiceProviderConfig schema defined in Section 5 of [RFC7643] has been omitted for brevity):

HTTP/1.1 200 OK
Content-Type: application/scim+json

{
   "schemas": [
      "urn:ietf:params:scim:schemas:core:2.0:ServiceProviderConfig"],
      ...

   "pagination": {
      "cursor": true,
      "index": true
   },

   ...
}

Service Provider implementors SHOULD ensure that misuse of pagination by a SCIM client does not deplete Service Provider resources or prevent valid requests from other clients being handled. Defenses for a SCIM Service Provider are similar those used to protect other Web API services -- including the use of a "Web API gateway" layer, to provide authentication, rate limiting, IP allow/block lists, logging and monitoring, response caching, etc.

For example, an obvious protection against abuse is for the Service Provider to require client authentication in order to retrieve large result sets and enforce an overriding totalResults limit for non- authenticated clients. Another example would be for a Service Provider that implements cursor pagination to restrict the number of cursors that can be allocated by a client or enforce cursor lifetimes.

5. Security Considerations

This section elaborates on the security considerations associated with the implementation of cursor pagination in SCIM. It is imperative that implementers consider the following security aspects to safeguard against both deliberate attacks and inadvertent misuse that may compromise the system's security posture.

5.1. Threat Model and Security Environment

The threat landscape is characterized by two primary types of actors:

  1. Unauthenticated and Authenticated Malicious Actors: These individuals or entities represent a malevolent threat. Their objectives include unauthorized access to data, alteration, or deletion through cursor-enabled queries. They may also seek to deplete server resources deliberately, aiming to cause a denial-of-service state, thereby reducing service availability.

  2. Authenticated Benign Users: This category includes legitimate users who, due to confusion or a lack of understanding, inadvertently engage in actions that consume server resources excessively. Such actions, while not ill-intended, can lead to unintended denial of service by overwhelming the system's capacity.

5.2. Confidentiality

To ensure that confidential data remains appropriately secured:

  • Implementors MUST ensure that pagination through results sets is strictly confined to the data that the actor's current identity has been authorized to access. This holds true even in cases where the actor has obtained a cursor pertaining to a result set that was generated by a different actor.

  • Authorization checks MUST BE continuously applied as an actor navigates through the result set associated with a cursor. Under no circumstances should possession of a cursor be interpreted as granting any supplementary access privileges to the actor.

  • In alignment with Section 2, cursor values are to be treated as opaque entities. Clients are strictly prohibited from making any inferences or assumptions about their internal structure.

  • The system MUST handle error scenarios gracefully, while not exposing sensitive data. For instance, if an actor attempts to access a page of results outside their authorized scope, or if a request is made for a non-existent page, the system should respond with identical error messages, so as not to disclose any details of the underlying data or the nature of the authorization failure. It is acceptable, however, for the system to log different messages to a log accessible by administrators or other authorized personnel.

5.3. Integrity

The extension discussed herein is query-only and does not inherently pose a substantial risk to data integrity. However, the focus is placed on safeguarding the integrity of the applications and clients that depend on this extension, rather than the integrity of the Service Provider. Specific considerations include: It is not required to tie a cursor to specific actor. However, if a cursor is tied to an actor and if the actor's permissions change, and the actor is still using the cursor, the actor may miss records OR there may be unauthorized access to data.

  • Servers are authorized to invalidate all tokens/watermarks corresponding to an actor immediately following a change in permissions. This ensures that any queries executed post-permission change, utilizing old tokens/watermarks, will be denied.

  • As an alternative approach, servers may opt to retain the existing tokens/watermarks but must ensure that any metadata tied to the result set, such as record counts, is updated to reflect the new permissions accurately.

5.4. Availability

The concern for availability primarily stems from the potential for Denial of Service (DoS) attacks. If the server elects to retain substantial data or metadata for each cursor, numerous concurrent queries with &cursor could strain and eventually exhaust server resources. This could be orchestrated by an attacker with malicious intent or could occur innocuously as a result of actions taken by a benign but confused actor.

To mitigate such risks, the following strategies are recommended:

  • Implementation of rate limiting to control the volume and cadence of cursor requests. This approach should adhere to established standards for rate limiting, details of which can be found in [RFC6585].

  • Cursory mechanisms must be designed in a manner that avoids any additional consumption of server resources with the initiation of new &cursor requests.

  • It is advisable to establish a ceiling on the number of cursors permissible at any given time. Alternatively, the adoption of an opaque identifier system that conservatively utilizes resources may be used.

  • Token invalidation mechanisms (including mechanisms triggered by permissions changes) must be designed to be resource-efficient to prevent them from being exploited for DoS attacks.

  • Actors may face challenges in maintaining a seamless pagination experience if their permissions are in a state of flux. Proactive measures should be taken to ensure that permission changes do not disrupt the user experience.

5.5. Other Security References

Using URIs to describe and locate resources has its own set of security considerations discussed in Section 7 of [RFC3986]. IANA Considerations

6. IANA Considerations

This specification amends the registry "SCIM Schema URIs for Data Resources" established by [RFC7643], for the urn:ietf:params:scim:api:messages:2.0:SearchRequest message URI and adds the following new fields:

SCIM cursor attribute

SCIM count attribute

This specification amends the entry for urn:ietf:params:scim:api:messages:2.0:ListResponse message URI, and adds the following fields:

SCIM nextCursor attribute

SCIM previousCursor attribute

This specification amends the entry for urn:ietf:params:scim:schemas:core:2.0:ServiceProviderConfig schema URI, and adds the following field:

SCIM pagination attribute

7. Change Log

-04

-03

-02

-01

-00

Acknowledgments

The editor would like to acknowledge the tremendous contribution of Matt Peterson for his work in authoring the original versions of this draft and in providing continuing feedback after stepping back.

The editor would also like to acknowledge the contributions of the following individuals who provided valuable feedback while reviewing the draft:

Normative References

[RFC2119]
Bradner, S., "Key words for use in RFCs to Indicate Requirement Levels", BCP 14, RFC 2119, DOI 10.17487/RFC2119, , <https://www.rfc-editor.org/rfc/rfc2119>.
[RFC3986]
Berners-Lee, T., Fielding, R., and L. Masinter, "Uniform Resource Identifier (URI): Generic Syntax", STD 66, RFC 3986, DOI 10.17487/RFC3986, , <https://www.rfc-editor.org/rfc/rfc3986>.
[RFC6585]
Nottingham, M. and R. Fielding, "Additional HTTP Status Codes", RFC 6585, DOI 10.17487/RFC6585, , <https://www.rfc-editor.org/rfc/rfc6585>.
[RFC7643]
Hunt, P., Ed., Grizzle, K., Wahlstroem, E., and C. Mortimore, "System for Cross-domain Identity Management: Core Schema", RFC 7643, DOI 10.17487/RFC7643, , <https://www.rfc-editor.org/rfc/rfc7643>.
[RFC7644]
Hunt, P., Ed., Grizzle, K., Ansari, M., Wahlstroem, E., and C. Mortimore, "System for Cross-domain Identity Management: Protocol", RFC 7644, DOI 10.17487/RFC7644, , <https://www.rfc-editor.org/rfc/rfc7644>.
[RFC8174]
Leiba, B., "Ambiguity of Uppercase vs Lowercase in RFC 2119 Key Words", BCP 14, RFC 8174, DOI 10.17487/RFC8174, , <https://www.rfc-editor.org/rfc/rfc8174>.

Authors' Addresses

Danny Zollner (editor)
Microsoft
Anjali Sehgal
Amazon Web Services