| P2PSIP Working Group | A. Knauf |
| Internet-Draft | G. Hege |
| Intended status: Standards Track | T C. Schmidt |
| Expires: April 26, 2012 | HAW Hamburg |
| M. Waehlisch | |
| link-lab & FU Berlin | |
| October 24, 2011 |
A Usage for Shared Resources in RELOAD (ShaRe)
draft-knauf-p2psip-share-02
This document defines a RELOAD Usage for managing shared write access to RELOAD Resources. Shared Resources in RELOAD (ShaRe) form a basic primitive for enabling various coordination and notification schemes among distributed peers. Access in ShaRe is controlled by a hierarchical trust delegation scheme maintained within an access list. A new USER-CHAIN-ACL access policy allows authorized peers to write a Shared Resource without owning its corresponding certificate. This specification also adds mechanisms to store Resources with a variable name which is useful whenever peer-independent rendezvous processes are required.
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 http://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 April 26, 2012.
Copyright (c) 2011 IETF Trust and the persons identified as the document authors. All rights reserved.
This document is subject to BCP 78 and the IETF Trust's Legal Provisions Relating to IETF Documents (http://trustee.ietf.org/license-info) in effect on the date of publication of this document. Please review these documents carefully, as they describe your rights and restrictions with respect to this document. Code Components extracted from this document must include Simplified BSD License text as described in Section 4.e of the Trust Legal Provisions and are provided without warranty as described in the Simplified BSD License.
This document defines a RELOAD Usage for managing shared write access to RELOAD Resources and a mechanism to store Resources with a variable name. The Usage for Shared Resources in RELOAD (ShaRe) enables overlay users to share their exclusive write access to specific Resource/Kind pairs with others. Shared Resources form a basic primitive for enabling various coordination and notification schemes among distributed peers. Write permission is controlled by an Access Control List (ACL) Kind that maintains a chain of Authorized Peers for a particular Shared Resource. A newly defined USER-CHAIN-ACL access control policy enables shared write access in RELOAD.
The Usage for Shared Resources in RELOAD is designed for jointly coordinated group applications among distributed peers (c.f. [I-D.knauf-p2psip-disco]). Of particular interest are rendezvous processes, where a single identifier is linked to multiple, dynamic instances of a distributed cooperative service. Shared write access is based on a trust delegation mechanism. It transfers the authorization to write a specific Kind data by storing logical Access Control Lists. An ACL contains the ID of the Kind to be shared and contains trust delegations from one authorized to another (previously unauthorized) user.
Shared write access extends the RELOAD security model, which is based on the restriction that peers are only allowed to write resources at a small set of well defined locations (Resource-IDs) in the overlay. Using the standard access control rules in RELOAD, these locations are bound to the username or Node-ID in the peer's certificate. This document extends the base policies to enable a controlled write access for multiple users to a common Resource Id.
Additionally, this specification defines an optional mechanism to store Resources with a variable Resource Name. It enables the storage of Resources whose name complies to a specific pattern. Definition of the pattern is arbitrary, but must contain the username of the Resource creator.
The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in [RFC2119].
This document uses the terminology and definitions from the RELOAD base [I-D.ietf-p2psip-base]and the peer-to-peer SIP concepts draft [I-D.ietf-p2psip-concepts]. Additionally, the following terms are used:
A RELOAD user that owns a certificate for writing at a specific overlay location can maintain one or more RELOAD Kinds that are designated for a non-exclusive write access shared with other RELOAD users. The mechanism to share those Resource/Kind pairs with a group of users consists of two basic steps.
ACLs are created by the Resource Owner and contain ACL items, each delegating the permission of writing the shared Kind to a specific user called Authorized Peer. For each shared Kind data, its Resource owner stores a root item that initiates an Access Control List. Trust delegation to the Authorized Peer can include the right to further delegate the write permission, enabling a tree of trust delegations with the Resource Owner as trust anchor at its root.
The Resource/Kind pair to be shared can be any RELOAD Kind that complies to the following specifications:
This Section defines mechanisms to avoid race conditions while concurrently writing an array or dictionary of a Shared Resource.
If a dictionary is used in the Shared Resource, the dictionary key MUST be the Node-ID of the certificate that will be used to sign the stored data. Thus data access is bound to the unique ID-holder.
If the data model of the Shared Resource is an array, the following algorithm will generate an array index that avoids collisions.
The resulting 32 bits long integer MUST be used as the index for storing an array entry in a Shared Resource. The 8 bit individual index can be incremented individually for further array entries and allows for 256 distinct entries per Peer.
The mechanism to create the array index is related to the pseudo-random algorithm to generate an SSRC identifier in RTP, see Section 8.1 in [RFC3550] for calculating a collision probability.
An Access Control List (ACL) is a (self-managed) shared resource that contains a list of AccessControlListItem structures as defined in Section 4.2. Each entry delegates write access for a specific Kind data to a single RELOAD user. An ACL enables the RELOAD user who is authorized to write a specific Resource-ID to delegate his exclusive write access to a specific Kind to further users of a RELOAD instance. Each Access Control List data structure therefore carries the information about who obtains write access, the Kind-ID of the Resource to be shared, and whether delegation includes write access to the ACL itself. The latter condition grants the right to delegate write access further for the Authorized Peer. Access Control Lists are stored at the same overlay location as the Shared Resource and use the RELOAD array data model. They are initially created by the Resource Owner.
Figure 1 shows an example of an Access Control List. We omit the res_name_ext field to simplify illustration. The array entry at index 0x123abc001 displays the initial creation of an ACL for a Shared Resource of Kind-ID 1234 at the same Resource-ID. It represents the root item of the trust delegation tree for this shared RELOAD Kind. The root entry MUST contain the username of the Resource owner in the "to_user" field and can only be written by the owner of the public key certificate associated with this Resource-ID. The allow_delegation (ad) flag for a root ACL item is set to 1 by default. The array index is generated by using the mechanism for isolating stored data as described in Section 3.1. Hence, the most significant 24 bits of the array index (0x123abc) are the least significant 24 bits of the Node-ID of the Resource Owner.
The array item at index 0x123abc002 represents the first trust delegation to an Authorized Peer that is thus permitted to write to the Shared Resource of Kind-ID 1234. Additionally, the Authorized peer Alice is also granted (limited) write access to the ACL as indicated by the allow_delegation flag (ad) set to 1. This configuration authorizes Alice to store further trust delegations to the Shared Resource, i.e., add items to the ACL. On the contrary, index 0x456def001 illustrates trust delegation for Kind-ID 1234, in which the Authorized Peer Bob is not allowed to grant access to further peers (ad = 0). Each Authorized Peer signs its ACL items with its own private key, which makes the item ownership transparent.
To manage Shared Resource access of multiple Kinds at a single location, the Resource Owner can create new ACL entries that refer to another Kind-ID as shown in array entry index 0x123abc003. Note that overwriting existing items in an Access Control List that reference a different Kind-ID revokes all trust delegations in the corresponding subtree (see Section 6.2). Authorized Peers are only enabled to overwrite existing ACL item they own. The Resource Owner is allowed to overwrite any existing ACL item, but should be aware of its consequences.
+------------------------------------------------------+
| Access Control List |
+-----------+------------------------------+-----------+
| #Index | Array Entries | signed by |
+-----------+------------------------------+-----------+
| 123abc001 | to_user:Owner Kind:1234 ad:1 | Owner |
+-----------+------------------------------+-----------+
| 123abc002 | to_user:Alice Kind:1234 ad:1 | Owner |
+-----------+------------------------------+-----------+
| 123abc003 | to_user:Owner Kind:4321 ad:1 | Owner |
+-----------+------------------------------+-----------+
| 123abc004 | to_user:Carol Kind:4321 ad:0 | Owner |
+-----------+------------------------------+-----------+
| ... | ... | ... |
+-----------+------------------------------+-----------+
| 456def001 | to_user:Bob Kind:1234 ad:0 | Alice |
+-----------+------------------------------+-----------+
| ... | ... | ... |
+-----------+------------------------------+-----------+
Implementations of ShaRe should be aware that the trust delegation in an Access Control List need not be loop free. Self-contained circular trust delegation from A to B and B to A are syntactically possible, even though not very meaningful.
The Kind data structure for the Access Control List is defined as follows:
struct {
/* res_name_ext is optional, see documentation */
ResourceNameExtension res_name_ext;
opaque to_user<0..2^16-1>;
KindId kind;
Boolean allow_delegation;
} AccessControlListItem;
The AccessControlListItem structure is composed of:
In certain use cases such as conferencing (c.f. [I-D.knauf-p2psip-disco]) it is desirable to increase the flexibility of a peer in using Resource Names beyond those defined by the username or Node-ID fields in its certificate. For this purpose, this document presents the concept for variable Resources Names that enables providers of RELOAD instances to define relaxed naming schemes for overlay Resources.
Each RELOAD node uses a certificate to identify itself using its user name (or Node-ID) while storing data under a specific Resource-ID. The specifications in this document scheme adhere to this paradigm, but enable a RELOAD peer to store values of Resource Names that are derived from the username in its certificate. This is done by using a Resource Name with a variable substring that still matches the username in the certificate using a pattern defined in the overlay configuration document. Thus despite being variable, an allowable Resource Name remains tied to the Owner's certificate. A sample pattern might be formed as follows.
Example Pattern:
.*-conf-$USER@$DOMAIN
When defining the pattern, care must be taken to avoid conflicts arising from two usernames of witch one is a substring of the other. In such cases, the holder of the shorter name could threaten to block the resources of the longer-named peer by choosing the variable part of a Resource Name to contain the entire longer username. This problem can easily be mitigated by delimiting the variable part of the pattern from the username part by some fixed string, that by convention is not part of a username (e.g. the "-conf-" in the above Example).
This section defines the optional ResourceNameExtension structure for every Kind that uses the USER-CHAIN-ACL access control policy.
enum { pattern (1),
(255)} ResourceNameType;
struct {
ResourceNameType type;
uint16 length;
select(type) {
case pattern:
opaque resource_name<0..2^16-1>;
/* Types can be extended */
}
} ResourceNameExtension
The content of the ResourceNameExtension consist of
The content of the rest of the data structure depends of the ResourceNameType. Currently, the only defined type is "pattern".
If the type is "pattern", then the following data structure contains an opaque <0..2^16-1> field containing the Resource Name of the Kind being stored. The type "pattern" further indicates that the Resource Name MUST match to one of the variable resource name pattern defined for this Kind in the configuration document.
The ResourceNameType enum and the ResourceNameExtension structure can be extended by further Usages to define other naming schemes.
This section extends the overlay configuration document by defining new elements for patterns relating resource names to user names.
The <variable-resource-names> element serves as a container for one or multiple <pattern> sub-elements. It is an additional parameter within the kind block and has a boolean "enable" attribute that indicates, if true, that the overlay provider allows variable resource names for this Kind. The default value of the "enable" attribute is "false". In the absence of a <variable-resource-names> element for a Kind using the USER-CHAIN-ACL access policy (see Section 6.6), implementors SHOULD assume this default value.
A <pattern> element MUST be present if the "enabled" attribute of its parent element is set to true. Each <pattern> element defines a pattern for constructing extended resource names for a single Kind. It is of type xsd:string and interpreted as a regular expression. In this regular expression, $USER and $DOMAIN are used as variables for the corresponding parts of the string in the certificate username field (with $USER preceding and $DOMAIN succeeding the '@'). Both variables MUST be present in any given pattern definition. If no pattern is defined for a Kind or the "enabled" attribute is false, allowable Resource Names are restricted to the username of the signer for Shared Resource.
The Relax NG Grammar for the Variable Resource Names Extension reads:
<!-- VARIABLE RESOURCE URN SUB-NAMESPACE -->
namespace share = "urn:ietf:params:xml:ns:p2p:config-base:share"
<!-- VARIABLE RESOURCE NAMES ELEMENT -->
kind-parameter &= element share:variable-resource-names {
attribute enable { xsd:boolean }
<!-- PATTERN ELEMENT -->
element pattern { xsd:string }*
}?
Write access to a Kind that is intended to be shared with other RELOAD users can solely be issued by the Resource Owner. A Resource Owner can share RELOAD Kinds by using the following procedure.
An Authorized Peer with delegation allowance ("ad"=1) can extend the access to an existing Shared Resource as follows.
A store request by an Authorized Peer that attempts to overwrite any ACL item signed by another Peer is unauthorized and causes an Error_Forbidden response from the Storing Peer. Such access conflicts could be caused by an array index collision. However, the probability of a collision of two or more identical array indices will be negligibly low using the mechanism for isolating stored data (see Section 3.1)
Write permissions are revoked by storing a non-existent value [I-D.ietf-p2psip-base] at the corresponding item of the Access Control List. Revoking a permission automatically invalidates all delegations performed by that user including all subsequent delegations. This allows to invalidate entire subtrees of the delegations tree with only a single operation. Overwriting the root item with a non-existent value of an Access List invalidates the entire delegations tree.
An existing ACL item MUST only be overwritten by the user who initially stored the corresponding entry, or by the Resource Owner that is allowed to overwrite all ACL items for revoking write access.
Access Control Lists are used to transparently validate authorization of peers for writing a data value at a Shared Resource. Thereby it is assumed that the validating peer is in possession of the complete and most recent ACL for a specific Resource/Kind pair. The corresponding procedure consists of recursively traversing the trust delegation tree and proceeds as follows.
The trust delegation chain is valid if and only if all verification steps succeed. In this case, the creator of the data value of the Shared Resource is an Authorized Peer.
Note that the ACL validation procedure can be omitted whenever the creator of data at a Shared Resource is the Resource Owner itself. The latter can be verified by its public key certificate as defined in Section 6.6.
Storing peers, i.e., peers at which Shared Resource and ACL are physically stored, are responsible for controlling storage attempts to a Shared Resource and its corresponding Access Control List. To assert the USER-CHAIN-ACL access policy (see Section 6.4), a storing peer MUST perform the access validation procedure described in Section 6.3 on any incoming store request using the most recent Access Control List for every Kind that uses the USER-CHAIN-ACL policy. It has further to ensure that only the Resource Owner stores new ACL root items for Shared Resources.
Accessing peers, i.e.,peers that fetch a Shared Resource, MAY validate that the originator of a Shared Resource was authorized to store data at this Resource-ID by processing the corresponding ACL. To enable an accessing peer to perform the access validation procedure described in Section 6.3, it first needs to obtain the most recent Access Control List in the following way.
Peers can cache previously fetched Access Control Lists up to the maximum lifetime of an individual item. Since stored values could have been modified or invalidated prior to their expiration, an accessing peer SHOULD use a Stat request to check for updates prior to using the data cache.
This document specifies an additional access control policy to the RELOAD base draft [I-D.ietf-p2psip-base]. The USER-CHAIN-ACL policy allows Authorized Peers to write a Shared Resource, even though they do not own the corresponding certificate. Additionally, the USER-CHAIN-ACL allows the storage of Kinds with a variable resource name that are following one of the specified naming pattern. Hence, on an inbound store request on a Kind that uses the USER-CHAIN-ACL access policy, the following rules MUST be applied:
In the USER-CHAIN-ACL policy, a given value MUST be written or overwritten, if either one of USER-MATCH or USER-NODE-MATCH (mandatory if the data model is dictionary) access policies of the base document [I-D.ietf-p2psip-base] applies.
Otherwise, the value MUST be written if the certificate of the signer contains a username that matches to one of the variable resource name pattern (c.f. Section 5) specified in the configuration document and, additionally, the hashed Resource Name matches the Resource-ID. The Resource Name of the Kind to be stored MUST be taken from the mandatory ResourceNameExtension field in the corresponding Kind data structure.
Otherwise, the value MUST be written if the ACL validation procedure described in Section 6.3 has been successfully applied.
This section defines the ACCESS-CONTROL-LIST Kind previously described in this document.
In this section we discuss security issues that are relevant to the usage of shared resources in RELOAD.
Joining a RELOAD overlay inherently poses a certain resource load on a peer, because it has to store and forward data for other peers. In common RELOAD semantics, each Resource ID and thus position in the overlay may only be written by a limited set of peers - often even only a single peer, which limits this burden. In the case of Shared Resources, a single resource may be written by multiple peers, who may even write an arbitrary number of entries (e.g., delegations in the ACL). This leads to an enhanced use of resources at individual overlay nodes. The problem of resource exhaustion can easily be mitigated for Usages based on the ShaRe-Usage by imposing restrictions on size, i.e., <max-size> element for a certain Kind in the configuration document.
The RELOAD overlay is designed to operate despite the presence of a small set of misbehaving peers. This is not different for Shared Resources since a small set of malicious peers does not disrupt the functionality of the overlay in general, but may have implications for the peers needing to store or access information at the specific locations in the ID space controlled by a malicious peer. A storing peer could withhold stored data which results in a denial of service to the group using the specific resource. But it could not return forged data, since the validity of any stored data can be independently verified using the attached signatures.
All data stored in the Shared Resource is publicly readable, thus applications requiring privacy need to encrypt the data. The ACL needs to be stored unencrypted, thus the list members of a group using a Shared Resource will always be publicly visible.
TODO: register Kind-ID code point at the IANA
This work was stimulated by fruitful discussions in the P2PSIP working group and SAM research group. We would like to thank all active members for constructive thoughts and feedback. In particular, the authors would like to thank (in alphabetical order) Lothar Grimm, Cullen Jennings, Peter Musgrave, Joerg Ott, Marc Petit-Huguenin, Peter Pogrzeba, and Jan Seedorf. This work was party funded by the German Federal Ministry of Education and Research, projects HAMcast and Mindstone.
| [I-D.ietf-p2psip-base] | Jennings, C, Lowekamp, B, Rescorla, E, Baset, S and H Schulzrinne, "REsource LOcation And Discovery (RELOAD) Base Protocol", Internet-Draft draft-ietf-p2psip-base-19, October 2011. |
| [RFC2119] | Bradner, S., "Key words for use in RFCs to Indicate Requirement Levels", BCP 14, RFC 2119, March 1997. |
| [I-D.ietf-p2psip-concepts] | Bryan, D, Matthews, P, Shim, E, Willis, D and S Dawkins, "Concepts and Terminology for Peer to Peer SIP", Internet-Draft draft-ietf-p2psip-concepts-04, October 2011. |
| [I-D.knauf-p2psip-disco] | Knauf, A, Hege, G, Schmidt, T and M Waehlisch, "A RELOAD Usage for Distributed Conference Control (DisCo)", Internet-Draft draft-knauf-p2psip-disco-04, October 2011. |
| [RFC3550] | Schulzrinne, H., Casner, S., Frederick, R. and V. Jacobson, "RTP: A Transport Protocol for Real-Time Applications", STD 64, RFC 3550, July 2003. |
The following changes have been made from version draft-knauf-p2psip-share-01:
The following changes have been made from version draft-knauf-p2psip-share-00: