P2PSIP C. Jennings
Internet-Draft Cisco
Intended status: Standards Track B. Lowekamp, Ed.
Expires: May 13, 2010 MYMIC LLC
E. Rescorla
Network Resonance
S. Baset
H. Schulzrinne
Columbia University
November 9, 2009
REsource LOcation And Discovery (RELOAD) Base Protocol
draft-ietf-p2psip-base-06
Abstract
In this document the term BCP 78 and BCP 79 refer to RFC 3978 and RFC
3979 respectively. They refer only to those RFCs and not to any
documents that update or supersede them.
This document defines REsource LOcation And Discovery (RELOAD), a
peer-to-peer (P2P) signaling protocol for use on the Internet. A P2P
signaling protocol provides its clients with an abstract storage and
messaging service between a set of cooperating peers that form the
overlay network. RELOAD is designed to support a P2P Session
Initiation Protocol (P2PSIP) network, but can be utilized by other
applications with similar requirements by defining new usages that
specify the kinds of data that must be stored for a particular
application. RELOAD defines a security model based on a certificate
enrollment service that provides unique identities. NAT traversal is
a fundamental service of the protocol. RELOAD also allows access
from "client" nodes that do not need to route traffic or store data
for others.
Legal
This documents and the information contained therein are provided on
an "AS IS" basis and THE CONTRIBUTOR, THE ORGANIZATION HE/SHE
REPRESENTS OR IS SPONSORED BY (IF ANY), THE INTERNET SOCIETY, THE
IETF TRUST AND THE INTERNET ENGINEERING TASK FORCE DISCLAIM ALL
WARRANTIES, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO ANY
WARRANTY THAT THE USE OF THE INFORMATION THEREIN WILL NOT INFRINGE
ANY RIGHTS OR ANY IMPLIED WARRANTIES OF MERCHANTABILITY OR FITNESS
FOR A PARTICULAR PURPOSE.
Status of this Memo
Jennings, et al. Expires May 13, 2010 [Page 1]
Internet-Draft RELOAD Base November 2009
This Internet-Draft is submitted to IETF in full conformance with the
provisions of BCP 78 and BCP 79.
Internet-Drafts are working documents 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.
This Internet-Draft will expire on May 13, 2010.
Copyright Notice
Copyright (c) 2009 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 BSD License.
This document may contain material from IETF Documents or IETF
Contributions published or made publicly available before November
10, 2008. The person(s) controlling the copyright in some of this
material may not have granted the IETF Trust the right to allow
modifications of such material outside the IETF Standards Process.
Without obtaining an adequate license from the person(s) controlling
the copyright in such materials, this document may not be modified
outside the IETF Standards Process, and derivative works of it may
not be created outside the IETF Standards Process, except to format
it for publication as an RFC or to translate it into languages other
than English.
Jennings, et al. Expires May 13, 2010 [Page 2]
Internet-Draft RELOAD Base November 2009
Table of Contents
1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . 8
1.1. Basic Setting . . . . . . . . . . . . . . . . . . . . . 9
1.2. Architecture . . . . . . . . . . . . . . . . . . . . . . 10
1.2.1. Usage Layer . . . . . . . . . . . . . . . . . . . . 13
1.2.2. Message Transport . . . . . . . . . . . . . . . . . 14
1.2.3. Storage . . . . . . . . . . . . . . . . . . . . . . 14
1.2.4. Topology Plugin . . . . . . . . . . . . . . . . . . 15
1.2.5. Forwarding and Link Management Layer . . . . . . . . 15
1.3. Security . . . . . . . . . . . . . . . . . . . . . . . . 16
1.4. Structure of This Document . . . . . . . . . . . . . . . 17
2. Terminology . . . . . . . . . . . . . . . . . . . . . . . . . 17
3. Overlay Management Overview . . . . . . . . . . . . . . . . . 19
3.1. Security and Identification . . . . . . . . . . . . . . 20
3.1.1. Shared-Key Security . . . . . . . . . . . . . . . . 21
3.2. Clients . . . . . . . . . . . . . . . . . . . . . . . . 21
3.2.1. Client Routing . . . . . . . . . . . . . . . . . . . 22
3.2.2. Minimum Functionality Requirements for Clients . . . 22
3.3. Routing . . . . . . . . . . . . . . . . . . . . . . . . 23
3.4. Connectivity Management . . . . . . . . . . . . . . . . 25
3.5. Overlay Algorithm Support . . . . . . . . . . . . . . . 26
3.5.1. Support for Pluggable Overlay Algorithms . . . . . . 26
3.5.2. Joining, Leaving, and Maintenance Overview . . . . . 26
3.6. First-Time Setup . . . . . . . . . . . . . . . . . . . . 28
3.6.1. Initial Configuration . . . . . . . . . . . . . . . 28
3.6.2. Enrollment . . . . . . . . . . . . . . . . . . . . . 28
4. Application Support Overview . . . . . . . . . . . . . . . . 28
4.1. Data Storage . . . . . . . . . . . . . . . . . . . . . . 29
4.1.1. Storage Permissions . . . . . . . . . . . . . . . . 30
4.1.2. Usages . . . . . . . . . . . . . . . . . . . . . . . 31
4.1.3. Replication . . . . . . . . . . . . . . . . . . . . 31
4.2. Service Discovery . . . . . . . . . . . . . . . . . . . 32
4.3. Application Connectivity . . . . . . . . . . . . . . . . 32
5. Overlay Management Protocol . . . . . . . . . . . . . . . . . 33
5.1. Message Receipt and Forwarding . . . . . . . . . . . . . 33
5.1.1. Responsible ID . . . . . . . . . . . . . . . . . . . 33
5.1.2. Other ID . . . . . . . . . . . . . . . . . . . . . . 34
5.1.3. Private ID . . . . . . . . . . . . . . . . . . . . . 35
5.2. Symmetric Recursive Routing . . . . . . . . . . . . . . 35
5.2.1. Request Origination . . . . . . . . . . . . . . . . 35
5.2.2. Response Origination . . . . . . . . . . . . . . . . 36
5.3. Message Structure . . . . . . . . . . . . . . . . . . . 36
5.3.1. Presentation Language . . . . . . . . . . . . . . . 37
5.3.1.1. Common Definitions . . . . . . . . . . . . . . . 38
5.3.2. Forwarding Header . . . . . . . . . . . . . . . . . 40
5.3.2.1. Processing Configuration Sequence Numbers . . . . 42
5.3.2.2. Destination and Via Lists . . . . . . . . . . . . 43
Jennings, et al. Expires May 13, 2010 [Page 3]
Internet-Draft RELOAD Base November 2009
5.3.2.3. Forwarding Options . . . . . . . . . . . . . . . 45
5.3.3. Message Contents Format . . . . . . . . . . . . . . 46
5.3.3.1. Response Codes and Response Errors . . . . . . . 47
5.3.4. Security Block . . . . . . . . . . . . . . . . . . . 49
5.4. Overlay Topology . . . . . . . . . . . . . . . . . . . . 52
5.4.1. Topology Plugin Requirements . . . . . . . . . . . . 52
5.4.2. Methods and types for use by topology plugins . . . 52
5.4.2.1. Join . . . . . . . . . . . . . . . . . . . . . . 52
5.4.2.2. Leave . . . . . . . . . . . . . . . . . . . . . . 53
5.4.2.3. Update . . . . . . . . . . . . . . . . . . . . . 54
5.4.2.4. Route_Query . . . . . . . . . . . . . . . . . . . 54
5.4.2.5. Probe . . . . . . . . . . . . . . . . . . . . . . 55
5.5. Forwarding and Link Management Layer . . . . . . . . . . 57
5.5.1. Attach . . . . . . . . . . . . . . . . . . . . . . . 58
5.5.1.1. Request Definition . . . . . . . . . . . . . . . 58
5.5.1.2. Response Definition . . . . . . . . . . . . . . . 60
5.5.1.3. Using ICE With RELOAD . . . . . . . . . . . . . . 61
5.5.1.4. Collecting STUN Servers . . . . . . . . . . . . . 61
5.5.1.5. Gathering Candidates . . . . . . . . . . . . . . 62
5.5.1.6. Encoding the Attach Message . . . . . . . . . . . 62
5.5.1.7. Verifying ICE Support . . . . . . . . . . . . . . 63
5.5.1.8. Role Determination . . . . . . . . . . . . . . . 63
5.5.1.9. Connectivity Checks . . . . . . . . . . . . . . . 63
5.5.1.10. Concluding ICE . . . . . . . . . . . . . . . . . 63
5.5.1.11. Subsequent Offers and Answers . . . . . . . . . . 64
5.5.1.12. Media Keepalives . . . . . . . . . . . . . . . . 64
5.5.1.13. Sending Media . . . . . . . . . . . . . . . . . . 64
5.5.1.14. Receiving Media . . . . . . . . . . . . . . . . . 64
5.5.2. AttachLite . . . . . . . . . . . . . . . . . . . . . 64
5.5.2.1. Request Definition . . . . . . . . . . . . . . . 64
5.5.2.2. Response Definition . . . . . . . . . . . . . . . 65
5.5.2.3. Attach-Lite Connectivity Checks . . . . . . . . . 65
5.5.2.4. Implementation Notes for Attach-Lite . . . . . . 65
5.5.3. AppAttach . . . . . . . . . . . . . . . . . . . . . 66
5.5.3.1. Request Definition . . . . . . . . . . . . . . . 66
5.5.3.2. Response Definition . . . . . . . . . . . . . . . 67
5.5.4. AppAttachLite . . . . . . . . . . . . . . . . . . . 67
5.5.4.1. Request Definition . . . . . . . . . . . . . . . 67
5.5.4.2. Response Definition . . . . . . . . . . . . . . . 68
5.5.5. Ping . . . . . . . . . . . . . . . . . . . . . . . . 68
5.5.5.1. Request Definition . . . . . . . . . . . . . . . 68
5.5.5.2. Response Definition . . . . . . . . . . . . . . . 68
5.5.6. Config_Update . . . . . . . . . . . . . . . . . . . 69
5.5.6.1. Request Definition . . . . . . . . . . . . . . . 69
5.5.6.2. Response Definition . . . . . . . . . . . . . . . 70
5.6. Overlay Link Layer . . . . . . . . . . . . . . . . . . . 70
5.6.1. Future Support for HIP . . . . . . . . . . . . . . . 71
5.6.2. Reliability for Unreliable Links . . . . . . . . . . 71
Jennings, et al. Expires May 13, 2010 [Page 4]
Internet-Draft RELOAD Base November 2009
5.6.2.1. Framed Message Format . . . . . . . . . . . . . . 72
5.6.2.2. Retransmission and Flow Control . . . . . . . . . 73
5.6.3. Fragmentation and Reassembly . . . . . . . . . . . . 74
6. Data Storage Protocol . . . . . . . . . . . . . . . . . . . . 75
6.1. Data Signature Computation . . . . . . . . . . . . . . . 77
6.2. Data Models . . . . . . . . . . . . . . . . . . . . . . 77
6.2.1. Single Value . . . . . . . . . . . . . . . . . . . . 78
6.2.2. Array . . . . . . . . . . . . . . . . . . . . . . . 79
6.2.3. Dictionary . . . . . . . . . . . . . . . . . . . . . 79
6.3. Access Control Policies . . . . . . . . . . . . . . . . 80
6.3.1. USER-MATCH . . . . . . . . . . . . . . . . . . . . . 80
6.3.2. NODE-MATCH . . . . . . . . . . . . . . . . . . . . . 80
6.3.3. USER-NODE-MATCH . . . . . . . . . . . . . . . . . . 80
6.3.4. NODE-MULTIPLE . . . . . . . . . . . . . . . . . . . 80
6.4. Data Storage Methods . . . . . . . . . . . . . . . . . . 81
6.4.1. Store . . . . . . . . . . . . . . . . . . . . . . . 81
6.4.1.1. Request Definition . . . . . . . . . . . . . . . 81
6.4.1.2. Response Definition . . . . . . . . . . . . . . . 85
6.4.1.3. Removing Values . . . . . . . . . . . . . . . . . 86
6.4.2. Fetch . . . . . . . . . . . . . . . . . . . . . . . 87
6.4.2.1. Request Definition . . . . . . . . . . . . . . . 88
6.4.2.2. Response Definition . . . . . . . . . . . . . . . 90
6.4.3. Stat . . . . . . . . . . . . . . . . . . . . . . . . 90
6.4.3.1. Request Definition . . . . . . . . . . . . . . . 91
6.4.3.2. Response Definition . . . . . . . . . . . . . . . 91
6.4.4. Find . . . . . . . . . . . . . . . . . . . . . . . . 93
6.4.4.1. Request Definition . . . . . . . . . . . . . . . 93
6.4.4.2. Response Definition . . . . . . . . . . . . . . . 93
6.4.5. Defining New Kinds . . . . . . . . . . . . . . . . . 94
7. Certificate Store Usage . . . . . . . . . . . . . . . . . . . 95
8. TURN Server Usage . . . . . . . . . . . . . . . . . . . . . . 96
9. Chord Algorithm . . . . . . . . . . . . . . . . . . . . . . . 97
9.1. Overview . . . . . . . . . . . . . . . . . . . . . . . . 98
9.2. Routing . . . . . . . . . . . . . . . . . . . . . . . . 98
9.3. Redundancy . . . . . . . . . . . . . . . . . . . . . . . 99
9.4. Joining . . . . . . . . . . . . . . . . . . . . . . . . 99
9.5. Routing Attaches . . . . . . . . . . . . . . . . . . . . 100
9.6. Updates . . . . . . . . . . . . . . . . . . . . . . . . 100
9.6.1. Handling Neighbor Failures . . . . . . . . . . . . . 102
9.6.2. Handling Finger Table Entry Failure . . . . . . . . 103
9.6.3. Receiving Updates . . . . . . . . . . . . . . . . . 103
9.6.4. Stabilization . . . . . . . . . . . . . . . . . . . 104
9.6.4.1. Updating neighbor table . . . . . . . . . . . . . 104
9.6.4.2. Refreshing finger table . . . . . . . . . . . . . 104
9.6.4.3. Adjusting finger table size . . . . . . . . . . . 105
9.6.4.4. Detecting partitioning . . . . . . . . . . . . . 106
9.7. Route Query . . . . . . . . . . . . . . . . . . . . . . 106
9.8. Leaving . . . . . . . . . . . . . . . . . . . . . . . . 107
Jennings, et al. Expires May 13, 2010 [Page 5]
Internet-Draft RELOAD Base November 2009
10. Enrollment and Bootstrap . . . . . . . . . . . . . . . . . . 108
10.1. Overlay Configuration . . . . . . . . . . . . . . . . . 108
10.1.1. Relax NG Grammar . . . . . . . . . . . . . . . . . . 112
10.2. Discovery Through Enrollment Server . . . . . . . . . . 114
10.3. Credentials . . . . . . . . . . . . . . . . . . . . . . 115
10.3.1. Self-Generated Credentials . . . . . . . . . . . . . 116
10.4. Searching for a Bootstrap Node . . . . . . . . . . . . . 117
10.5. Contacting a Bootstrap Node . . . . . . . . . . . . . . 117
11. Message Flow Example . . . . . . . . . . . . . . . . . . . . 118
12. Security Considerations . . . . . . . . . . . . . . . . . . . 123
12.1. Overview . . . . . . . . . . . . . . . . . . . . . . . . 123
12.2. Attacks on P2P Overlays . . . . . . . . . . . . . . . . 124
12.3. Certificate-based Security . . . . . . . . . . . . . . . 124
12.4. Shared-Secret Security . . . . . . . . . . . . . . . . . 125
12.5. Storage Security . . . . . . . . . . . . . . . . . . . . 126
12.5.1. Authorization . . . . . . . . . . . . . . . . . . . 126
12.5.2. Distributed Quota . . . . . . . . . . . . . . . . . 127
12.5.3. Correctness . . . . . . . . . . . . . . . . . . . . 127
12.5.4. Residual Attacks . . . . . . . . . . . . . . . . . . 127
12.6. Routing Security . . . . . . . . . . . . . . . . . . . . 128
12.6.1. Background . . . . . . . . . . . . . . . . . . . . . 128
12.6.2. Admissions Control . . . . . . . . . . . . . . . . . 129
12.6.3. Peer Identification and Authentication . . . . . . . 129
12.6.4. Protecting the Signaling . . . . . . . . . . . . . . 130
12.6.5. Residual Attacks . . . . . . . . . . . . . . . . . . 130
13. IANA Considerations . . . . . . . . . . . . . . . . . . . . . 131
13.1. Port Registrations . . . . . . . . . . . . . . . . . . . 131
13.2. Overlay Algorithm Types . . . . . . . . . . . . . . . . 131
13.3. Access Control Policies . . . . . . . . . . . . . . . . 131
13.4. Data Kind-ID . . . . . . . . . . . . . . . . . . . . . . 132
13.5. Data Model . . . . . . . . . . . . . . . . . . . . . . . 132
13.6. Message Codes . . . . . . . . . . . . . . . . . . . . . 133
13.7. Error Codes . . . . . . . . . . . . . . . . . . . . . . 134
13.8. Transport Types . . . . . . . . . . . . . . . . . . . . 134
13.9. Forwarding Options . . . . . . . . . . . . . . . . . . . 134
13.10. Probe Information Types . . . . . . . . . . . . . . . . 135
13.11. Message Extensions . . . . . . . . . . . . . . . . . . . 135
13.12. reload URI Scheme . . . . . . . . . . . . . . . . . . . 135
13.12.1. URI Registration . . . . . . . . . . . . . . . . . . 136
14. Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . 137
15. References . . . . . . . . . . . . . . . . . . . . . . . . . 137
15.1. Normative References . . . . . . . . . . . . . . . . . . 137
15.2. Informative References . . . . . . . . . . . . . . . . . 138
Appendix A. Change Log . . . . . . . . . . . . . . . . . . . . . 141
A.1. Changes since draft-ietf-p2psip-reload-04 . . . . . . . 141
A.2. Changes since draft-ietf-p2psip-reload-01 . . . . . . . 141
A.3. Changes since draft-ietf-p2psip-reload-00 . . . . . . . 142
A.4. Changes since draft-ietf-p2psip-base-00 . . . . . . . . 142
Jennings, et al. Expires May 13, 2010 [Page 6]
Internet-Draft RELOAD Base November 2009
A.5. Changes since draft-ietf-p2psip-base-01 . . . . . . . . 142
A.6. Changes since draft-ietf-p2psip-base-01a . . . . . . . . 142
A.7. Changes since draft-ietf-p2psip-base-02 . . . . . . . . 142
Appendix B. AIMD Retransmission Scheme . . . . . . . . . . . . . 143
Appendix C. TFRC Retransmission Scheme . . . . . . . . . . . . . 143
Appendix D. Routing Alternatives . . . . . . . . . . . . . . . . 144
D.1. Iterative vs Recursive . . . . . . . . . . . . . . . . . 144
D.2. Symmetric vs Forward response . . . . . . . . . . . . . 144
D.3. Direct Response . . . . . . . . . . . . . . . . . . . . 145
D.4. Relay Peers . . . . . . . . . . . . . . . . . . . . . . 146
D.5. Symmetric Route Stability . . . . . . . . . . . . . . . 146
Appendix E. Why Clients? . . . . . . . . . . . . . . . . . . . . 147
E.1. Why Not Only Peers? . . . . . . . . . . . . . . . . . . 147
E.2. Clients as Application-Level Agents . . . . . . . . . . 148
Authors' Addresses . . . . . . . . . . . . . . . . . . . . . . . 148
Jennings, et al. Expires May 13, 2010 [Page 7]
Internet-Draft RELOAD Base November 2009
1. Introduction
This document defines REsource LOcation And Discovery (RELOAD), a
peer-to-peer (P2P) signaling protocol for use on the Internet. It
provides a generic, self-organizing overlay network service, allowing
nodes to efficiently route messages to other nodes and to efficiently
store and retrieve data in the overlay. RELOAD provides several
features that are critical for a successful P2P protocol for the
Internet:
Security Framework: A P2P network will often be established among a
set of peers that do not trust each other. RELOAD leverages a
central enrollment server to provide credentials for each peer
which can then be used to authenticate each operation. This
greatly reduces the possible attack surface.
Usage Model: RELOAD is designed to support a variety of
applications, including P2P multimedia communications with the
Session Initiation Protocol [I-D.ietf-p2psip-sip]. RELOAD allows
the definition of new application usages, each of which can define
its own data types, along with the rules for their use. This
allows RELOAD to be used with new applications through a simple
documentation process that supplies the details for each
application.
NAT Traversal: RELOAD is designed to function in environments where
many if not most of the nodes are behind NATs or firewalls.
Operations for NAT traversal are part of the base design,
including using ICE to establish new RELOAD or application
protocol connections.
High Performance Routing: The very nature of overlay algorithms
introduces a requirement that peers participating in the P2P
network route requests on behalf of other peers in the network.
This introduces a load on those other peers, in the form of
bandwidth and processing power. RELOAD has been defined with a
simple, lightweight forwarding header, thus minimizing the amount
of effort required by intermediate peers.
Pluggable Overlay Algorithms: RELOAD has been designed with an
abstract interface to the overlay layer to simplify implementing a
variety of structured (DHT) and unstructured overlay algorithms.
This specification also defines how RELOAD is used with Chord,
which is mandatory to implement. Specifying a default "must
implement" overlay algorithm promotes interoperability, while
extensibility allows selection of overlay algorithms optimized for
a particular application.
Jennings, et al. Expires May 13, 2010 [Page 8]
Internet-Draft RELOAD Base November 2009
These properties were designed specifically to meet the requirements
for a P2P protocol to support SIP. This document defines the base
protocol for the distributed storage and location service, as well as
critical usages for NAT traversal and security. The SIP Usage itself
is described separately in [I-D.ietf-p2psip-sip]. RELOAD is not
limited to usage by SIP and could serve as a tool for supporting
other P2P applications with similar needs. RELOAD is also based on
the concepts introduced in [I-D.ietf-p2psip-concepts].
1.1. Basic Setting
In this section, we provide a brief overview of the operational
setting for RELOAD. See the concepts document for more details. A
RELOAD Overlay Instance consists of a set of nodes arranged in a
partly connected graph. Each node in the overlay is assigned a
numeric Node-ID which, together with the specific overlay algorithm
in use, determines its position in the graph and the set of nodes it
connects to. The figure below shows a trivial example which isn't
drawn from any particular overlay algorithm, but was chosen for
convenience of representation.
+--------+ +--------+ +--------+
| Node 10|--------------| Node 20|--------------| Node 30|
+--------+ +--------+ +--------+
| | |
| | |
+--------+ +--------+ +--------+
| Node 40|--------------| Node 50|--------------| Node 60|
+--------+ +--------+ +--------+
| | |
| | |
+--------+ +--------+ +--------+
| Node 70|--------------| Node 80|--------------| Node 90|
+--------+ +--------+ +--------+
|
|
+--------+
| Node 85|
|(Client)|
+--------+
Because the graph is not fully connected, when a node wants to send a
message to another node, it may need to route it through the network.
For instance, Node 10 can talk directly to nodes 20 and 40, but not
to Node 70. In order to send a message to Node 70, it would first
send it to Node 40 with instructions to pass it along to Node 70.
Different overlay algorithms will have different connectivity graphs,
but the general idea behind all of them is to allow any node in the
Jennings, et al. Expires May 13, 2010 [Page 9]
Internet-Draft RELOAD Base November 2009
graph to efficiently reach every other node within a small number of
hops.
The RELOAD network is not only a messaging network. It is also a
storage network. Records are stored under numeric addresses which
occupy the same space as node identifiers. Peers are responsible for
storing the data associated with some set of addresses as determined
by their Node-ID. For instance, we might say that every peer is
responsible for storing any data value which has an address less than
or equal to its own Node-ID, but greater than the next lowest
Node-ID. Thus, Node-20 would be responsible for storing values
11-20.
RELOAD also supports clients. These are nodes which have Node-IDs
but do not participate in routing or storage. For instance, in the
figure above Node 85 is a client. It can route to the rest of the
RELOAD network via Node 80, but no other node will route through it
and Node 90 is still responsible for all addresses between 81-90. We
refer to non-client nodes as peers.
Other applications (for instance, SIP) can be defined on top of
RELOAD and use these two basic RELOAD services to provide their own
services.
1.2. Architecture
RELOAD is fundamentally an overlay network. Therefore, it can be
divided into components that mimic the layering of the Internet
model[RFC1122].
Jennings, et al. Expires May 13, 2010 [Page 10]
Internet-Draft RELOAD Base November 2009
Application
+-------+ +-------+
| SIP | | XMPP | ...
| Usage | | Usage |
+-------+ +-------+
-------------------------------------- Messaging API
+------------------+ +---------+
| Message |<--->| Storage |
| Transport | +---------+
+------------------+ ^
^ ^ |
| v v
| +-------------------+
| | Topology |
| | Plugin |
| +-------------------+
| ^
v v
+------------------+
| Forwarding & |
| Link Management |
+------------------+
-------------------------------------- Overlay Link API
+-------+ +------+
|TLS | |DTLS | ...
+-------+ +------+
The major components of RELOAD are:
Usage Layer: Each application defines a RELOAD usage; a set of data
kinds and behaviors which describe how to use the services
provided by RELOAD. These usages all talk to RELOAD through a
common Message Transport API.
Message Transport: Handles end-to-end reliability, manages request
state for the usages, and forwards Store and Fetch operations to
the Storage component. Delivers message responses to the
component initiating the request.
Storage: The Storage component is responsible for processing
messages relating to the storage and retrieval of data. It talks
directly to the Topology Plugin to manage data replication and
migration, and it talks to the Message Transport component to send
and receive messages.
Jennings, et al. Expires May 13, 2010 [Page 11]
Internet-Draft RELOAD Base November 2009
Topology Plugin: The Topology Plugin is responsible for implementing
the specific overlay algorithm being used. It uses the Message
Transport component to send and receive overlay management
messages, to the Storage component to manage data replication, and
directly to the Forwarding Layer to control hop-by-hop message
forwarding. This component closely parallels conventional routing
algorithms, but is more tightly coupled to the Forwarding Layer
because there is no single "routing table" equivalent used by all
overlay algorithms.
Forwarding and Link Management Layer: Stores and implements the
routing table by providing packet forwarding services between
nodes. It also handles establishing new links between nodes,
including setting up connections across NATs using ICE.
Overlay Link Layer: TLS [RFC5246] and DTLS [RFC4347] are the "link
layer" protocols used by RELOAD for hop-by-hop communication.
Each such protocol includes the appropriate provisions for per-hop
framing or hop-by-hop ACKs required by unreliable transports.
To further clarify the roles of the various layer, this figure
parallels the architecture with each layer's role from an overlay
perspective and implementation layer in the internet:
Jennings, et al. Expires May 13, 2010 [Page 12]
Internet-Draft RELOAD Base November 2009
| Internet Model |
Real | Equivalent | Reload
Internet | in Overlay | Architecture
--------------+-----------------+------------------------------------
| | +-------+ +-------+
| Application | | SIP | | XMPP | ...
| | | Usage | | Usage |
| | +-------+ +-------+
| | ----------------------------------
| |+------------------+ +---------+
| Transport || Message |<--->| Storage |
| || Transport | +---------+
| |+------------------+ ^
| | ^ ^ |
| | | v v
Application | | | +-------------------+
| (Routing) | | | Topology |
| | | | Plugin |
| | | +-------------------+
| | | ^
| | v v
| Network | +------------------+
| | | Forwarding & |
| | | Link Management |
| | +------------------+
| | ----------------------------------
Transport | Link | +-------+ +------+
| | |TLS | |DTLS | ...
| | +-------+ +------+
--------------+-----------------+------------------------------------
Network |
|
Link |
1.2.1. Usage Layer
The top layer, called the Usage Layer, has application usages, such
as the SIP Location Usage, that use the abstract Message Transport
API provided by RELOAD. The goal of this layer is to implement
application-specific usages of the generic overlay services provided
by RELOAD. The usage defines how a specific application maps its
data into something that can be stored in the overlay, where to store
the data, how to secure the data, and finally how applications can
retrieve and use the data.
The architecture diagram shows both a SIP usage and an XMPP usage. A
single application may require multiple usages; for example a SIP
application may also require a voicemail usage. A usage may define
Jennings, et al. Expires May 13, 2010 [Page 13]
Internet-Draft RELOAD Base November 2009
multiple kinds of data that are stored in the overlay and may also
rely on kinds originally defined by other usages.
Because the security and storage policies for each kind are dictated
by the usage defining the kind, the usages may be coupled with the
Storage component to provide security policy enforcement and to
implement appropriate storage strategies according to the needs of
the usage. The exact implementation of such an interface is outside
the scope of this draft.
1.2.2. Message Transport
The Message Transport component provides a generic message routing
service for the overlay. The Message Transport layer is responsible
for end-to-end message transactions, including retransmissions. Each
peer is identified by its location in the overlay as determined by
its Node-ID. A component that is a client of the Message Transport
can perform two basic functions:
o Send a message to a given peer specified by Node-ID or to the peer
responsible for a particular Resource-ID.
o Receive messages that other peers send to a Node-ID or Resource-ID
for which the receiving peer is responsible.
All usages rely on the Message Transport component to send and
receive messages from peers. For instance, when a usage wants to
store data, it does so by sending Store requests. Note that the
Storage component and the Topology Plugin are themselves clients of
the Message Transport, because they need to send and receive messages
from other peers.
The Message Transport API is similar to those described as providing
"Key based routing" (KBR), although as RELOAD supports different
overlay algorithms (including non-DHT overlay algorithms) that
calculate keys in different ways, the actual interface must accept
Resource Names rather than actual keys.
1.2.3. Storage
One of the major functions of RELOAD is to allow nodes to store data
in the overlay and to retrieve data stored by other nodes or by
themselves. The Storage component is responsible for processing data
storage and retrieval messages. For instance, the Storage component
might receive a Store request for a given resource from the Message
Transport. It would then query the appropriate usage before storing
the data value(s) in its local data store and sending a response to
the Message Transport for delivery to the requesting peer.
Typically, these messages will come from other nodes, but depending
Jennings, et al. Expires May 13, 2010 [Page 14]
Internet-Draft RELOAD Base November 2009
on the overlay topology, a node might be responsible for storing data
for itself as well, especially if the overlay is small.
A peer's Node-ID determines the set of resources that it will be
responsible for storing. However, the exact mapping between these is
determined by the overlay algorithm in use. The Storage component
will only receive a Store request from the Message Transport if this
peer is responsible for that Resource-ID. The Storage component is
notified by the Topology Plugin when the Resource-IDs for which it is
responsible change, and the Storage component is then responsible for
migrating resources to other peers, as required.
1.2.4. Topology Plugin
RELOAD is explicitly designed to work with a variety of overlay
algorithms. In order to facilitate this, the overlay algorithm
implementation is provided by a Topology Plugin so that each overlay
can select an appropriate overlay algorithm that relies on the common
RELOAD core protocols and code.
The Topology Plugin is responsible for maintaining the overlay
algorithm Routing Table, which is consulted by the Forwarding and
Link Management Layer before routing a message. When connections are
made or broken, the Forwarding and Link Management Layer notifies the
Topology Plugin, which adjusts the routing table as appropriate. The
Topology Plugin will also instruct the Forwarding and Link Management
Layer to form new connections as dictated by the requirements of the
overlay algorithm Topology. The Topology Plugin issues periodic
update requests through Message Transport to maintain and update its
Routing Table.
As peers enter and leave, resources may be stored on different peers,
so the Topology Plugin also keeps track of which peers are
responsible for which resources. As peers join and leave, the
Topology Plugin instructs the Storage component to issue resource
migration requests as appropriate, in order to ensure that other
peers have whatever resources they are now responsible for. The
Topology Plugin is also responsible for providing for redundant data
storage to protect against loss of information in the event of a peer
failure and to protect against compromised or subversive peers.
1.2.5. Forwarding and Link Management Layer
The Forwarding and Link Management Layer is responsible for getting a
packet to the next peer, as determined by the Topology Plugin. This
Layer establishes and maintains the network connections as required
by the Topology Plugin. This layer is also responsible for setting
up connections to other peers through NATs and firewalls using ICE,
Jennings, et al. Expires May 13, 2010 [Page 15]
Internet-Draft RELOAD Base November 2009
and it can elect to forward traffic using relays for NAT and firewall
traversal.
This layer provides a fairly generic interface that allows the
topology plugin to control the overlay and resource operations and
messages. Since each overlay algorithm is defined and functions
differently, we generically refer to the table of other peers that
the overlay algorithm maintains and uses to route requests
(neighbors) as a Routing Table. The Topology Plugin actually owns
the Routing Table, and forwarding decisions are made by querying the
Topology Plugin for the next hop for a particular Node-ID or
Resource-ID. If this node is the destination of the message, the
message is delivered to the Message Transport.
The Forwarding and Link Management Layer sits on top of the Overlay
Link Layer protocols that carry the actual traffic. This
specification defines how to use DTLS and TLS protocols to carry
RELOAD messages.
1.3. Security
RELOAD's security model is based on each node having one or more
public key certificates. In general, these certificates will be
assigned by a central server which also assigns Node-IDs, although
self-signed certificates can be used in closed networks. These
credentials can be leveraged to provide communications security for
RELOAD messages. RELOAD provides communications security at three
levels:
Connection Level: Connections between peers are secured with TLS
or DTLS.
Message Level: Each RELOAD message must be signed.
Object Level: Stored objects must be signed by the storing peer.
These three levels of security work together to allow peers to verify
the origin and correctness of data they receive from other peers,
even in the face of malicious activity by other peers in the overlay.
RELOAD also provides access control built on top of these
communications security features. Because the peer responsible for
storing a piece of data can validate the signature on the data being
stored, the responsible peer can determine whether a given operation
is permitted or not.
RELOAD also provides an optional shared secret based admission
control feature using shared secrets and TLS-PSK. This mode is
typically used when self-signed certificates are being used but would
generally not be used when the certificates were all signed by an
enrollment server. In order to form a TLS connection to any node in
Jennings, et al. Expires May 13, 2010 [Page 16]
Internet-Draft RELOAD Base November 2009
the overlay, a new node needs to know the shared overlay key, thus
restricting access to authorized users only. This feature is used
together with certificate-based access control, not as a replacement
for it.
1.4. Structure of This Document
The remainder of this document is structured as follows.
o Section 2 provides definitions of terms used in this document.
o Section 3 provides an overview of the mechanisms used to establish
and maintain the overlay.
o Section 4 provides an overview of the mechanism RELOAD provides to
support other applications.
o Section 5 defines the protocol messages that RELOAD uses to
establish and maintain the overlay.
o Section 6 defines the protocol messages that are used to store and
retrieve data using RELOAD.
o Section 7 defines the Certificate Store Usage that is fundamental
to RELOAD security.
o Section 8 defines the TURN Server Usage needed to locate TURN
servers for NAT traversal.
o Section 9 defines a specific Topology Plugin using Chord.
o Section 10 defines the mechanisms that new RELOAD nodes use to
join the overlay for the first time.
o Section 11 provides an extended example.
2. Terminology
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 RFC 2119 [RFC2119].
We use the terminology and definitions from the Concepts and
Terminology for Peer to Peer SIP [I-D.ietf-p2psip-concepts] draft
extensively in this document. Other terms used in this document are
defined inline when used and are also defined below for reference.
Terms which are new to this document (and perhaps should be added to
the concepts document) are marked with a (*).
DHT: A distributed hash table. A DHT is an abstract hash table
service realized by storing the contents of the hash table across
a set of peers.
Jennings, et al. Expires May 13, 2010 [Page 17]
Internet-Draft RELOAD Base November 2009
Overlay Algorithm: An overlay algorithm defines the rules for
determining which peers in an overlay store a particular piece of
data and for determining a topology of interconnections amongst
peers in order to find a piece of data.
Overlay Instance: A specific overlay algorithm and the collection of
peers that are collaborating to provide read and write access to
it. There can be any number of overlay instances running in an IP
network at a time, and each operates in isolation of the others.
Peer: A host that is participating in the overlay. Peers are
responsible for holding some portion of the data that has been
stored in the overlay and also route messages on behalf of other
hosts as required by the Overlay Algorithm.
Client: A host that is able to store data in and retrieve data from
the overlay but which is not participating in routing or data
storage for the overlay.
Node: We use the term "Node" to refer to a host that may be either a
Peer or a Client. Because RELOAD uses the same protocol for both
clients and peers, much of the text applies equally to both.
Therefore we use "Node" when the text applies to both Clients and
Peers and the more specific term (i.e. client or peer) when the
text applies only to Clients or only to Peers.
Node-ID: A 128-bit value that uniquely identifies a node. Node-IDs
0 and 2^128 - 1 are reserved and are invalid Node-IDs. A value of
zero is not used in the wire protocol but can be used to indicate
an invalid node in implementations and APIs. The Node-ID of
2^128-1 is used on the wire protocol as a wildcard. (*)
Resource: An object or group of objects associated with a string
identifier. See "Resource Name" below.
Resource Name: The potentially human readable name by which a
resource is identified. In unstructured P2P networks, the
resource name is sometimes used directly as a Resource-ID. In
structured P2P networks the resource name is typically mapped into
a Resource-ID by using the string as the input to hash function.
A SIP resource, for example, is often identified by its AOR which
is an example of a Resource Name.(*)
Jennings, et al. Expires May 13, 2010 [Page 18]
Internet-Draft RELOAD Base November 2009
Resource-ID: A value that identifies some resources and which is
used as a key for storing and retrieving the resource. Often this
is not human friendly/readable. One way to generate a Resource-ID
is by applying a mapping function to some other unique name (e.g.,
user name or service name) for the resource. The Resource-ID is
used by the distributed database algorithm to determine the peer
or peers that are responsible for storing the data for the
overlay. In structured P2P networks, Resource-IDs are generally
fixed length and are formed by hashing the resource name. In
unstructured networks, resource names may be used directly as
Resource-IDs and may be variable lengths.
Connection Table: The set of nodes to which a node is directly
connected. This includes nodes with which Attach handshakes have
been done but which have not sent any Updates.
Routing Table: The set of peers which a node can use to route
overlay messages. In general, these peers will all be on the
connection table but not vice versa, because some peers will have
Attached but not sent updates. Peers may send messages directly
to peers that are in the connection table but may only route
messages to other peers through peers that are in the routing
table. (*)
Destination List: A list of IDs through which a message is to be
routed. A single Node-ID is a trivial form of destination list.
(*)
Usage: A usage is an application that wishes to use the overlay for
some purpose. Each application wishing to use the overlay defines
a set of data kinds that it wishes to use. The SIP usage defines
the location data kind. (*)
The term "maximum request lifetime" is the maximum time a request
will wait for a response; it defaults to 15 seconds. The term
"successor replacement hold-down time" is the amount of time to wait
before starting replication when a new successor is found; it
defaults to 30 seconds.
3. Overlay Management Overview
The most basic function of RELOAD is as a generic overlay network.
Nodes need to be able to join the overlay, form connections to other
nodes, and route messages through the overlay to nodes to which they
are not directly connected. This section provides an overview of the
mechanisms that perform these functions.
Jennings, et al. Expires May 13, 2010 [Page 19]
Internet-Draft RELOAD Base November 2009
3.1. Security and Identification
Every node in the RELOAD overlay is identified by a Node-ID. The
Node-ID is used for three major purposes:
o To address the node itself.
o To determine its position in the overlay topology when the overlay
is structured.
o To determine the set of resources for which the node is
responsible.
Each node has a certificate [RFC5280] containing a Node-ID, which is
globally unique.
The certificate serves multiple purposes:
o It entitles the user to store data at specific locations in the
Overlay Instance. Each data kind defines the specific rules for
determining which certificates can access each Resource-ID/Kind-ID
pair. For instance, some kinds might allow anyone to write at a
given location, whereas others might restrict writes to the owner
of a single certificate.
o It entitles the user to operate a node that has a Node-ID found in
the certificate. When the node forms a connection to another
peer, it uses this certificate so that a node connecting to it
knows it is connected to the correct node (technically: a (D)TLS
association with client authentication is formed.) In addition,
the node can sign messages, thus providing integrity and
authentication for messages which are sent from the node.
o It entitles the user to use the user name found in the
certificate.
If a user has more than one device, typically they would get one
certificate for each device. This allows each device to act as a
separate peer.
RELOAD supports multiple certificate issuance models. The first is
based on a central enrollment process which allocates a unique name
and Node-ID and puts them in a certificate for the user. All peers
in a particular Overlay Instance have the enrollment server as a
trust anchor and so can verify any other peer's certificate.
In some settings, a group of users want to set up an overlay network
but are not concerned about attack by other users in the network.
For instance, users on a LAN might want to set up a short term ad hoc
network without going to the trouble of setting up an enrollment
server. RELOAD supports the use of self-generated and self-signed
certificates. When self-signed certificates are used, the node also
Jennings, et al. Expires May 13, 2010 [Page 20]
Internet-Draft RELOAD Base November 2009
generates its own Node-ID and username. The Node-ID is computed as a
digest of the public key, to prevent Node-ID theft; however this
model is still subject to a number of known attacks (most notably
Sybil attacks [Sybil]) and can only be safely used in closed networks
where users are mutually trusting.
The general principle here is that the security mechanisms (TLS and
message signatures) are always used, even if the certificates are
self-signed. This allows for a single set of code paths in the
systems with the only difference being whether certificate
verification is required to chain to a single root of trust.
3.1.1. Shared-Key Security
RELOAD also provides an admission control system based on shared
keys. In this model, the peers all share a single key which is used
to authenticate the peer-to-peer connections via TLS-PSK/TLS-SRP.
3.2. Clients
RELOAD defines a single protocol that is used both as the peer
protocol and as the client protocol for the overlay. This simplifies
implementation, particularly for devices that may act in either role,
and allows clients to inject messages directly into the overlay.
We use the term "peer" to identify a node in the overlay that routes
messages for nodes other than those to which it is directly
connected. Peers typically also have storage responsibilities. We
use the term "client" to refer to nodes that do not have routing or
storage responsibilities. When text applies to both peers and
clients, we will simply refer such devices as "nodes."
RELOAD's client support allows nodes that are not participating in
the overlay as peers to utilize the same implementation and to
benefit from the same security mechanisms as the peers. Clients
possess and use certificates that authorize the user to store data at
certain locations in the overlay. The Node-ID in the certificate is
used to identify the particular client as a member of the overlay and
to authenticate its messages.
In RELOAD, unlike some other designs, clients are not a first-class
concept. From the perspective of a peer, a client is simply a node
which has not yet sent any Updates or Joins. It might never do so
(if it's a client) or it might eventually do so (if it's just a node
that's taking a long time to join). The routing and storage rules
for RELOAD provide for correct behavior by peers regardless of
whether other nodes attached to them are clients or peers. Of
course, a client implementation must know that it intends to be a
Jennings, et al. Expires May 13, 2010 [Page 21]
Internet-Draft RELOAD Base November 2009
client, but this localizes complexity only to that node.
For more discussion of the motivation for RELOAD's client support,
see Appendix E.
3.2.1. Client Routing
There are two routing options by which a client may be located in an
overlay.
o Establish a connection to the peer responsible for the client's
Node-ID in the overlay. Then requests may be sent from/to the
client using its Node-ID in the same manner as if it were a peer,
because the responsible peer in the overlay will handle the final
step of routing to the client. This will not work in overlays
where NATs or firewalls prevent clients from forming connections
with other peers. Note that clients that choose this option MUST
process Update messages from the peer. Those updates can indicate
that the peer no longer owns the Client's Node-ID. The client
then forms a connection to the appropriate peer. Failure to do so
will result in the client no longer receiving messages.
o Establish a connection with an arbitrary peer in the overlay
(perhaps based on network proximity or an inability to establish a
direct connection with the responsible peer). In this case, the
client will rely on RELOAD's Destination List feature to ensure
reachability. The client can initiate requests, and any node in
the overlay that knows the Destination List to its current
location can reach it, but the client is not directly reachable
using only its Node-ID. The Destination List required to reach it
must be learnable via other mechanisms, such as being stored in
the overlay by a usage, if the client is to receive incoming
requests from other members of the overlay.
3.2.2. Minimum Functionality Requirements for Clients
A node may act as a client simply because it does not have the
resources or even an implementation of the topology plugin required
to act as a peer in the overlay. In order to exchange RELOAD
messages with a peer, a client must meet a minimum level of
functionality. Such a client must:
o Implement RELOAD's connection-management connections that are used
to establish the connection with the peer.
o Implement RELOAD's data retrieval methods (with client
functionality).
o Be able to calculate Resource-IDs used by the overlay.
Jennings, et al. Expires May 13, 2010 [Page 22]
Internet-Draft RELOAD Base November 2009
o Possess security credentials required by the overlay it is
implementing.
A client speaks the same protocol as the peers, knows how to
calculate Resource-IDs, and signs its requests in the same manner as
peers. While a client does not necessarily require a full
implementation of the overlay algorithm, calculating the Resource-ID
requires an implementation of the appropriate algorithm for the
overlay.
3.3. Routing
This section will discuss the requirements RELOAD's routing
capabilities must meet, then describe the routing features in the
protocol, and then provide a brief overview of how they are used.
Appendix D discusses some alternative designs and the tradeoffs that
would be necessary to support them.
RELOAD's routing capabilities must meet the following requirements:
NAT Traversal: RELOAD must support establishing and using
connections between nodes separated by one or more NATs, including
locating peers behind NATs for those overlays allowing/requiring
it.
Clients: RELOAD must support requests from and to clients that do
not participate in overlay routing.
Client promotion: RELOAD must support clients that become peers at a
later point as determined by the overlay algorithm and deployment.
Low state: RELOAD's routing algorithms must not require
significant state to be stored on intermediate peers.
Return routability in unstable topologies: At some points in
times, different nodes may have inconsistent information about the
connectivity of the routing graph. In all cases, the response to
a request needs to delivered to the node that sent the request and
not to some other node.
To meet these requirements, RELOAD's routing relies on two basic
mechanisms:
Via Lists: The forwarding header used by all RELOAD messages
contains both a Via List (built hop-by-hop as the message is
routed through the overlay) and a Destination List (providing
source-routing capabilities for requests and return-path routing
for responses).
Jennings, et al. Expires May 13, 2010 [Page 23]
Internet-Draft RELOAD Base November 2009
Route_Query: The Route_Query method allows a node to query a peer
for the next hop it will use to route a message. This method is
useful for diagnostics and for iterative routing.
The basic routing mechanism used by RELOAD is Symmetric Recursive.
We will first describe symmetric routing and then discuss its
advantages in terms of the requirements discussed above.
Symmetric recursive routing requires that a message follow the path
through the overlay to the destination without returning to the
originating node: each peer forwards the message closer to its
destination. The return path of the response is then the same path
followed in reverse. For example, a message following a route from A
to Z through B and X:
A B X Z
-------------------------------
---------->
Dest=Z
---------->
Via=A
Dest=Z
---------->
Via=A, B
Dest=Z
<----------
Dest=X, B, A
<----------
Dest=B, A
<----------
Dest=A
Note that the preceding Figure does not indicate whether A is a
client or peer: A forwards its request to B and the response is
returned to A in the same manner regardless of A's role in the
overlay.
This figure shows use of full via-lists by intermediate peers B and
X. However, if B and/or X are willing to store state, then they may
elect to truncate the lists, save that information internally (keyed
by the transaction id), and return the response message along the
path from which it was received when the response is received. This
option requires greater state to be stored on intermediate peers but
saves a small amount of bandwidth and reduces the need for modifying
the message en route. Selection of this mode of operation is a
Jennings, et al. Expires May 13, 2010 [Page 24]
Internet-Draft RELOAD Base November 2009
choice for the individual peer; the techniques are interoperable even
on a single message. The figure below shows B using full via lists
but X truncating them and saving the state internally.
A B X Z
-------------------------------
---------->
Dest=Z
---------->
Via=A
Dest=Z
---------->
Dest=Z
<----------
Dest=X
<----------
Dest=B, A
<----------
Dest=A
RELOAD also supports a basic Iterative routing mode (where the
intermediate peers merely return a response indicating the next hop,
but do not actually forward the message to that next hop themselves).
Iterative routing is implemented using the Route_Query method, which
requests this behavior. Note that iterative routing is selected only
by the initiating node. RELOAD does not support an intermediate peer
returning a response that it will not recursively route a normal
request. The willingness to perform that operation is implicit in
its role as a peer in the overlay.
3.4. Connectivity Management
In order to provide efficient routing, a peer needs to maintain a set
of direct connections to other peers in the Overlay Instance. Due to
the presence of NATs, these connections often cannot be formed
directly. Instead, we use the Attach request to establish a
connection. Attach uses ICE [I-D.ietf-mmusic-ice] to establish the
connection. It is assumed that the reader is familiar with ICE.
Say that peer A wishes to form a direct connection to peer B. It
gathers ICE candidates and packages them up in an Attach request
which it sends to B through usual overlay routing procedures. B does
its own candidate gathering and sends back a response with its
candidates. A and B then do ICE connectivity checks on the candidate
pairs. The result is a connection between A and B. At this point, A
and B can add each other to their routing tables and send messages
Jennings, et al. Expires May 13, 2010 [Page 25]
Internet-Draft RELOAD Base November 2009
directly between themselves without going through other overlay
peers.
There is one special case in which Attach cannot be used: when a
peer is joining the overlay and is not connected to any peers. In
order to support this case, some small number of "bootstrap nodes"
typically need to be publicly accessible so that new peers can
directly connect to them. Section 10 contains more detail on this.
In general, a peer needs to maintain connections to all of the peers
near it in the Overlay Instance and to enough other peers to have
efficient routing (the details depend on the specific overlay). If a
peer cannot form a connection to some other peer, this isn't
necessarily a disaster; overlays can route correctly even without
fully connected links. However, a peer should try to maintain the
specified link set and if it detects that it has fewer direct
connections, should form more as required. This also implies that
peers need to periodically verify that the connected peers are still
alive and if not try to reform the connection or form an alternate
one.
3.5. Overlay Algorithm Support
The Topology Plugin allows RELOAD to support a variety of overlay
algorithms. This draft defines a DHT based on Chord [Chord], which
is mandatory to implement, but the base RELOAD protocol is designed
to support a variety of overlay algorithms.
3.5.1. Support for Pluggable Overlay Algorithms
RELOAD defines three methods for overlay maintenance: Join, Update,
and Leave. However, the contents of those messages, when they are
sent, and their precise semantics are specified by the actual overlay
algorithm; RELOAD merely provides a framework of commonly-needed
methods that provides uniformity of notation (and ease of debugging)
for a variety of overlay algorithms.
3.5.2. Joining, Leaving, and Maintenance Overview
When a new peer wishes to join the Overlay Instance, it must have a
Node-ID that it is allowed to use. When an enrollment server is used
that Node-Id will be in the certificate the node received from the
enrollment server. The details of the joining procedure are defined
by the overlay algorithm, but the general steps for joining an
Overlay Instance are:
Jennings, et al. Expires May 13, 2010 [Page 26]
Internet-Draft RELOAD Base November 2009
o Forming connections to some other peers.
o Acquiring the data values this peer is responsible for storing.
o Informing the other peers which were previously responsible for
that data that this peer has taken over responsibility.
The first thing the peer needs to do is to form a connection to some
"bootstrap node". Because this is the first connection the peer
makes, these nodes must have public IP addresses so that they can be
connected to directly. Once a peer has connected to one or more
bootstrap nodes, it can form connections in the usual way by routing
Attach messages through the overlay to other nodes. Once a peer has
connected to the overlay for the first time, it can cache the set of
nodes it has connected to with public IP addresses for use as future
bootstrap nodes.
Once a peer has connected to a bootstrap node, it then needs to take
up its appropriate place in the overlay. This requires two major
operations:
o Forming connections to other peers in the overlay to populate its
Routing Table.
o Getting a copy of the data it is now responsible for storing and
assuming responsibility for that data.
The second operation is performed by contacting the Admitting Peer
(AP), the node which is currently responsible for that section of the
overlay.
The details of this operation depend mostly on the overlay algorithm
involved, but a typical case would be:
1. JP (Joining Peer) sends a Join request to AP (Admitting Peer)
announcing its intention to join.
2. AP sends a Join response.
3. AP does a sequence of Stores to JP to give it the data it will
need.
4. AP does Updates to JP and to other peers to tell it about its own
routing table. At this point, both JP and AP consider JP
responsible for some section of the Overlay Instance.
5. JP makes its own connections to the appropriate peers in the
Overlay Instance.
After this process is completed, JP is a full member of the Overlay
Instance and can process Store/Fetch requests.
Note that the first node is a special case. When ordinary nodes
cannot form connections to the bootstrap nodes, then they are not
part of the overlay. However, the first node in the overlay can
Jennings, et al. Expires May 13, 2010 [Page 27]
Internet-Draft RELOAD Base November 2009
obviously not connect to others nodes. In order to support this
case, potential first nodes (which must also serve as bootstrap nodes
initially) must somehow be instructed (perhaps by configuration
settings) that they are the entire overlay, rather than not part of
it.
Note that clients do not perfom either of these operations.
3.6. First-Time Setup
Previous sections addressed how RELOAD works once a node has
connected. This section provides an overview of how users get
connected to the overlay for the first time. RELOAD is designed so
that users can start with the name of the overlay they wish to join
and perhaps a username and password, and leverage that into having a
working peer with minimal user intervention. This helps avoid the
problems that have been experienced with conventional SIP clients
where users are required to manually configure a large number of
settings.
3.6.1. Initial Configuration
In the first phase of the process, the user starts out with the name
of the overlay and uses this to download an initial set of overlay
configuration parameters. The user does a DNS SRV lookup on the
overlay name to get the address of a configuration server. It can
then connect to this server with HTTPS to download a configuration
document which contains the basic overlay configuration parameters as
well as a set of bootstrap nodes which can be used to join the
overlay.
3.6.2. Enrollment
If the overlay is using centralized enrollment, then a user needs to
acquire a certificate before joining the overlay. The certificate
attests both to the user's name within the overlay and to the Node-
IDs which they are permitted to operate. In that case, the
configuration document will contain the address of an enrollment
server which can be used to obtain such a certificate. The
enrollment server may (and probably will) require some sort of
username and password before issuing the certificate. The enrollment
server's ability to restrict attackers' access to certificates in the
overlay is one of the cornerstones of RELOAD's security.
4. Application Support Overview
RELOAD is not intended to be used alone, but rather as a substrate
Jennings, et al. Expires May 13, 2010 [Page 28]
Internet-Draft RELOAD Base November 2009
for other applications. These applications can use RELOAD for a
variety of purposes:
o To store data in the overlay and retrieve data stored by other
nodes.
o As a discovery mechanism for services such as TURN.
o To form direct connections which can be used to transmit
application-level messages.
This section provides an overview of these services.
4.1. Data Storage
RELOAD provides operations to Store and Fetch data. Each location in
the Overlay Instance is referenced by a Resource-ID. However, each
location may contain data elements corresponding to multiple kinds
(e.g., certificate, SIP registration). Similarly, there may be
multiple elements of a given kind, as shown below:
+--------------------------------+
| Resource-ID |
| |
| +------------+ +------------+ |
| | Kind 1 | | Kind 2 | |
| | | | | |
| | +--------+ | | +--------+ | |
| | | Value | | | | Value | | |
| | +--------+ | | +--------+ | |
| | | | | |
| | +--------+ | | +--------+ | |
| | | Value | | | | Value | | |
| | +--------+ | | +--------+ | |
| | | +------------+ |
| | +--------+ | |
| | | Value | | |
| | +--------+ | |
| +------------+ |
+--------------------------------+
Each kind is identified by a Kind-ID, which is a code point assigned
by IANA. As part of the kind definition, protocol designers may
define constraints, such as limits on size, on the values which may
be stored. For many kinds, the set may be restricted to a single
value; some sets may be allowed to contain multiple identical items
while others may only have unique items. Note that a kind may be
employed by multiple usages and new usages are encouraged to use
previously defined kinds where possible. We define the following
data models in this document, though other usages can define their
Jennings, et al. Expires May 13, 2010 [Page 29]
Internet-Draft RELOAD Base November 2009
own structures:
single value: There can be at most one item in the set and any value
overwrites the previous item.
array: Many values can be stored and addressed by a numeric index.
dictionary: The values stored are indexed by a key. Often this key
is one of the values from the certificate of the peer sending the
Store request.
In order to protect stored data from tampering, by other nodes, each
stored value is digitally signed by the node which created it. When
a value is retrieved, the digital signature can be verified to detect
tampering.
4.1.1. Storage Permissions
A major issue in peer-to-peer storage networks is minimizing the
burden of becoming a peer, and in particular minimizing the amount of
data which any peer is required to store for other nodes. RELOAD
addresses this issue by only allowing any given node to store data at
a small number of locations in the overlay, with those locations
being determined by the node's certificate. When a peer uses a Store
request to place data at a location authorized by its certificate, it
signs that data with the private key that corresponds to its
certificate. Then the peer responsible for storing the data is able
to verify that the peer issuing the request is authorized to make
that request. Each data kind defines the exact rules for determining
what certificate is appropriate.
The most natural rule is that a certificate authorizes a user to
store data keyed with their user name X. This rule is used for all
the kinds defined in this specification. Thus, only a user with a
certificate for "alice@example.org" could write to that location in
the overlay. However, other usages can define any rules they choose,
including publicly writable values.
The digital signature over the data serves two purposes. First, it
allows the peer responsible for storing the data to verify that this
Store is authorized. Second, it provides integrity for the data.
The signature is saved along with the data value (or values) so that
any reader can verify the integrity of the data. Of course, the
responsible peer can "lose" the value but it cannot undetectably
modify it.
The size requirements of the data being stored in the overlay are
Jennings, et al. Expires May 13, 2010 [Page 30]
Internet-Draft RELOAD Base November 2009
variable. For instance, a SIP AoR and voicemail differ widely in the
storage size. RELOAD leaves it to the Usage and overlay
configuration to limit size imbalance of various kinds.
4.1.2. Usages
By itself, the distributed storage layer just provides infrastructure
on which applications are built. In order to do anything useful, a
usage must be defined. Each Usage specifies several things:
o Registers Kind-ID code points for any kinds that the Usage
defines.
o Defines the data structure for each of the kinds.
o Defines access control rules for each of the kinds.
o Defines how the Resource Name is formed that is hashed to form the
Resource-ID where each kind is stored.
o Describes how values will be merged after a network partition.
Unless otherwise specified, the default merging rule is to act as
if all the values that need to be merged were stored and as if the
order they were stored in corresponds to the stored time values
associated with (and carried in) their values. Because the stored
time values are those associated with the peer which did the
writing, clock skew is generally not an issue. If two nodes are
on different partitions, write to the same location, and have
clock skew, this can create merge conflicts. However because
RELOAD deliberately segregates storage so that data from different
users and peers is stored in different locations, and a single
peer will typically only be in a single network partition, this
case will generally not arise.
The kinds defined by a usage may also be applied to other usages.
However, a need for different parameters, such as different size
limits, would imply the need to create a new kind.
4.1.3. Replication
Replication in P2P overlays can be used to provide:
persistence: if the responsible peer crashes and/or if the storing
peer leaves the overlay
security: to guard against DoS attacks by the responsible peer or
routing attacks to that responsible peer
load balancing: to balance the load of queries for popular
resources.
A variety of schemes are used in P2P overlays to achieve some of
these goals. Common techniques include replicating on neighbors of
the responsible peer, randomly locating replicas around the overlay,
Jennings, et al. Expires May 13, 2010 [Page 31]
Internet-Draft RELOAD Base November 2009
or replicating along the path to the responsible peer.
The core RELOAD specification does not specify a particular
replication strategy. Instead, the first level of replication
strategies are determined by the overlay algorithm, which can base
the replication strategy on the its particular topology. For
example, Chord places replicas on successor peers, which will take
over responsibility should the responsible peer fail [Chord].
If additional replication is needed, for example if data persistence
is particularly important for a particular usage, then that usage may
specify additional replication, such as implementing random
replications by inserting a different well known constant into the
Resource Name used to store each replicated copy of the resource.
Such replication strategies can be added independent of the
underlying algorithm, and their usage can be determined based on the
needs of the particular usage.
4.2. Service Discovery
RELOAD does not currently define a generic service discovery
algorithm as part of the base protocol, although a simplistic TURN-
specific discovery mechanism is provided. A variety of service
discovery algorithms can be implemented as extensions to the base
protocol, such as the service discovery algorithm ReDIR
[opendht-sigcomm05] .
4.3. Application Connectivity
There is no requirement that a RELOAD usage must use RELOAD's
primitives for establishing its own communication if it already
possesses its own means of establishing connections. For example,
one could design a RELOAD-based resource discovery protocol which
used HTTP to retrieve the actual data.
For more common situations, however, it is the overlay itself -
rather than an external authority such as DNS - which is used to
establish a connection. RELOAD provides connectivity to applications
using the AppAttach method. For example, if a P2PSIP node wishes to
establish a SIP dialog with another P2PSIP node, it will use
AppAttach to establish a direct connection with the other node. This
new connection is separate from the peer protocol connection. It is
a dedicated UDP or TCP flow used only for the SIP dialog. Each usage
specifies which types of connections can be initiated using
AppAttach.
Jennings, et al. Expires May 13, 2010 [Page 32]
Internet-Draft RELOAD Base November 2009
5. Overlay Management Protocol
This section defines the basic protocols used to create, maintain,
and use the RELOAD overlay network. We start by defining the basic
concept of how message destinations are interpreted when routing
messages. We then describe the symmetric recursive routing model,
which is RELOAD's default routing algorithm. We then define the
message structure and then finally define the messages used to join
and maintain the overlay.
5.1. Message Receipt and Forwarding
When a peer receives a message, it first examines the overlay,
version, and other header fields to determine whether the message is
one it can process. If any of these are incorrect (e.g., the message
is for an overlay in which the peer does not participate) it is an
error. The peer SHOULD generate an appropriate error but local
policy can override this and cause the messages to be silently
dropped.
Once the peer has determined that the message is correctly formatted,
it examines the first entry on the destination list. There are three
possible cases here:
o The first entry on the destination list is an ID for which the
peer is responsible.
o The first entry on the destination list is an ID for which another
peer is responsible.
o The first entry on the destination list is a private ID that is
being used for destination list compression. This is described
later.
These cases are handled as discussed below.
5.1.1. Responsible ID
If the first entry on the destination list is an ID for which the
node is responsible, there are several sub-cases.
o If the entry is a Resource-ID, then it MUST be the only entry on
the destination list. If there are other entries, the message
MUST be silently dropped. Otherwise, the message is destined for
this node and it passes it up to the upper layers.
o If the entry is a Node-ID which belongs to this node, then the
message is destined for this node. If this is the only entry on
the destination list, the message is destined for this node and is
passed up to the upper layers. Otherwise the entry is removed
from the destination list and the message is passed to the Message
Transport. If the message is a response and there is state for
Jennings, et al. Expires May 13, 2010 [Page 33]
Internet-Draft RELOAD Base November 2009
the transaction ID, the state is reinserted into the destination
list first.
o If the entry is a Node-ID which is not equal to this node, then
the node MUST drop the message silently unless the Node-ID
corresponds to a node which is directly connected to this node
(i.e., a client). In that case, it MUST forward the message to
the destination node as described in the next section.
Note that this implies that in order to address a message to "the
peer that controls region X", a sender sends to Resource-ID X, not
Node-ID X.
5.1.2. Other ID
If neither of the other two cases applies, then the peer MUST forward
the message towards the first entry on the destination list. This
means that it MUST select one of the peers to which it is connected
and which is likely to be responsible for the first entry on the
destination list. If the first entry on the destination list is in
the peer's connection table, then it SHOULD forward the message to
that peer directly. Otherwise, the peer consults the routing table
to forward the message.
Any intermediate peer which forwards a RELOAD message MUST arrange
that if it receives a response to that message the response can be
routed back through the set of nodes through which the request
passed. This may be arranged in one of two ways:
o The peer MAY add an entry to the via list in the forwarding header
that will enable it to determine the correct node.
o The peer MAY keep per-transaction state which will allow it to
determine the correct node.
As an example of the first strategy, if node D receives a message
from node C with via list (A, B), then D would forward to the next
node (E) with via list (A, B, C). Now, if E wants to respond to the
message, it reverses the via list to produce the destination list,
resulting in (D, C, B, A). When D forwards the response to C, the
destination list will contain (C, B, A).
As an example of the second strategy, if node D receives a message
from node C with transaction ID X and via list (A, B), it could store
(X, C) in its state database and forward the message with the via
list unchanged. When D receives the response, it consults its state
database for transaction id X, determines that the request came from
C, and forwards the response to C.
Intermediate peers which modify the via list are not required to
Jennings, et al. Expires May 13, 2010 [Page 34]
Internet-Draft RELOAD Base November 2009
simply add entries. The only requirement is that the peer be able to
reconstruct the correct destination list on the return route. RELOAD
provides explicit support for this functionality in the form of
private IDs, which can replace any number of via list entries. For
instance, in the above example, Node D might send E a via list
containing only the private ID (I). E would then use the destination
list (D, I) to send its return message. When D processes this
destination list, it would detect that I is a private ID, recover the
via list (A, B, C), and reverse that to produce the correct
destination list (C, B, A) before sending it to C. This feature is
called List Compression. I MAY either be a compressed version of the
original via list or an index into a state database containing the
original via list.
Note that if an intermediate peer exits the overlay, then on the
return trip the message cannot be forwarded and will be dropped. The
ordinary timeout and retransmission mechanisms provide stability over
this type of failure.
5.1.3. Private ID
If the first entry in the destination list is a private id (e.g., a
compressed via list), the peer MUST replace that entry with the
original via list that it replaced and then re-examine the
destination list to determine which of the above cases now applies.
5.2. Symmetric Recursive Routing
This Section defines RELOAD's symmetric recursive routing algorithm,
which is the default algorithm used by nodes to route messages
through the overlay. All implementations MUST implement this routing
algorithm. An overlay may be configured to use alternative routing
algorithms, and alternative routing algorithms may be selected on a
per-message basis.
5.2.1. Request Origination
In order to originate a message to a given Node-ID or Resource-ID, a
node constructs an appropriate destination list. The simplest such
destination list is a single entry containing the peer or
Resource-ID. The resulting message will use the normal overlay
routing mechanisms to forward the message to that destination. The
node can also construct a more complicated destination list for
source routing.
Once the message is constructed, the node sends the message to some
adjacent peer. If the first entry on the destination list is
directly connected, then the message MUST be routed down that
Jennings, et al. Expires May 13, 2010 [Page 35]
Internet-Draft RELOAD Base November 2009
connection. Otherwise, the topology plugin MUST be consulted to
determine the appropriate next hop.
Parallel searches for the resource are a common solution to improve
reliability in the face of churn or of subversive peers. Parallel
searches for usage-specified replicas are managed by the usage layer.
However, a single request can also be routed through multiple
adjacent peers, even when known to be sub-optimal, to improve
reliability [vulnerabilities-acsac04]. Such parallel searches MAY BE
specified by the topology plugin.
Because messages may be lost in transit through the overlay, RELOAD
incorporates an end-to-end reliability mechanism. When an
originating node transmits a request it MUST set a 3 second timer.
If a response has not been received when the timer fires, the request
is retransmitted with the same transaction identifier. The request
MAY be retransmitted up to 4 times (for a total of 5 messages).
After the timer for the fifth transmission fires, the message SHALL
be considered to have failed. Note that this retransmission
procedure is not followed by intermediate nodes. They follow the
hop-by-hop reliability procedure described in Section 5.6.2.
The above algorithm can result in multiple requests being delivered
to a node. Receiving nodes MUST generate semantically equivalent
responses to retransmissions of the same request (this can be
determined by transaction id) if the request is received within the
maximum request lifetime (15 seconds). For some requests (e.g.,
Fetch) this can be accomplished merely by processing the request
again. For other requests, (e.g., Store) it may be necessary to
maintain state for the duration of the request lifetime.
5.2.2. Response Origination
When a peer sends a response to a request, it MUST construct the
destination list by reversing the order of the entries on the via
list. This has the result that the response traverses the same peers
as the request traversed, except in reverse order (symmetric
routing).
5.3. Message Structure
RELOAD is a message-oriented request/response protocol. The messages
are encoded using binary fields. All integers are represented in
network byte order. The general philosophy behind the design was to
use Type, Length, Value fields to allow for extensibility. However,
for the parts of a structure that were required in all messages, we
just define these in a fixed position, as adding a type and length
for them is unnecessary and would simply increase bandwidth and
Jennings, et al. Expires May 13, 2010 [Page 36]
Internet-Draft RELOAD Base November 2009
introduce new potential for interoperability issues.
Each message has three parts, concatenated as shown below:
+-------------------------+
| Forwarding Header |
+-------------------------+
| Message Contents |
+-------------------------+
| Security Block |
+-------------------------+
The contents of these parts are as follows:
Forwarding Header: Each message has a generic header which is used
to forward the message between peers and to its final destination.
This header is the only information that an intermediate peer
(i.e., one that is not the target of a message) needs to examine.
Message Contents: The message being delivered between the peers.
From the perspective of the forwarding layer, the contents are
opaque, however, they are interpreted by the higher layers.
Security Block: A security block containing certificates and a
digital signature over the message. Note that this signature can
be computed without parsing the message contents. All messages
MUST be signed by their originator.
The following sections describe the format of each part of the
message.
5.3.1. Presentation Language
The structures defined in this document are defined using a C-like
syntax based on the presentation language used to define TLS.
Advantages of this style include:
o It is easy to write and familiar enough looking that most readers
can grasp it quickly.
o The ability to define nested structures allows a separation
between high-level and low-level message structures.
o It has a straightforward wire encoding that allows quick
implementation, but the structures can be comprehended without
knowing the encoding.
o The ability to mechanically (compile) encoders and decoders.
This presentation is to some extent a placeholder. We consider it an
Jennings, et al. Expires May 13, 2010 [Page 37]
Internet-Draft RELOAD Base November 2009
open question what the final protocol definition method and encodings
use. We expect this to be a question for the WG to decide.
Several idiosyncrasies of this language are worth noting.
o All lengths are denoted in bytes, not objects.
o Variable length values are denoted like arrays with angle
brackets.
o "select" is used to indicate variant structures.
For instance, "uint16 array<0..2^8-2>;" represents up to 254 bytes
but only up to 127 values of two bytes (16 bits) each.
5.3.1.1. Common Definitions
The following definitions are used throughout RELOAD and so are
defined here. They also provide a convenient introduction to how to
read the presentation language.
An enum represents an enumerated type. The values associated with
each possibility are represented in parentheses and the maximum value
is represented as a nameless value, for purposes of describing the
width of the containing integral type. For instance, Boolean
represents a true or false:
enum { false (0), true(1), (255)} Boolean;
A boolean value is either a 1 or a 0 and is represented as a single
byte on the wire.
The NodeId, shown below, represents a single Node-ID.
typedef opaque NodeId[16];
A NodeId is a fixed-length 128-bit structure represented as a series
of bytes, with the most significant byte first. Note: the use of
"typedef" here is an extension to the TLS language, but its meaning
should be relatively obvious.
A ResourceId, shown below, represents a single Resource-ID.
typedef opaque ResourceId<0..2^8-1>;
Jennings, et al. Expires May 13, 2010 [Page 38]
Internet-Draft RELOAD Base November 2009
Like a NodeId, a Resource-ID is an opaque string of bytes, but unlike
Node-IDs, Resource-IDs are variable length, up to 255 bytes (2048
bits) in length. On the wire, each ResourceId is preceded by a
single length byte (allowing lengths up to 255). Thus, the 3-byte
value "Foo" would be encoded as: 03 46 4f 4f.
A more complicated example is IpAddressPort, which represents a
network address and can be used to carry either an IPv6 or IPv4
address:
enum {reserved_addr(0), ipv4_address (1), ipv6_address (2),
(255)} AddressType;
struct {
uint32 addr;
uint16 port;
} IPv4AddrPort;
struct {
uint128 addr;
uint16 port;
} IPv6AddrPort;
struct {
AddressType type;
uint8 length;
select (type) {
case ipv4_address:
IPv4AddrPort v4addr_port;
case ipv6_address:
IPv6AddrPort v6addr_port;
/* This structure can be extended */
} IpAddressPort;
The first two fields in the structure are the same no matter what
kind of address is being represented:
Jennings, et al. Expires May 13, 2010 [Page 39]
Internet-Draft RELOAD Base November 2009
type: the type of address (v4 or v6).
length: the length of the rest of the structure.
By having the type and the length appear at the beginning of the
structure regardless of the kind of address being represented, an
implementation which does not understand new address type X can still
parse the IpAddressPort field and then discard it if it is not
needed.
The rest of the IpAddressPort structure is either an IPv4AddrPort or
an IPv6AddrPort. Both of these simply consist of an address
represented as an integer and a 16-bit port. As an example, here is
the wire representation of the IPv4 address "192.0.2.1" with port
"6100".
01 ; type = IPv4
06 ; length = 6
c0 00 02 01 ; address = 192.0.2.1
17 d4 ; port = 6100
Unless a given structure that uses a select explicitly allows for
unknown types in the select, any unknown type SHOULD be treated as an
parsing error and the whole message discarded with no response.
5.3.2. Forwarding Header
The forwarding header is defined as a ForwardingHeader structure, as
shown below.
struct {
uint32 relo_token;
uint32 overlay;
uint16 configuration_sequence;
uint8 version;
uint8 ttl;
uint32 fragment;
uint32 length;
uint64 transaction_id;
uint32 max_response_length;
uint16 via_list_length;
uint16 destination_list_length;
uint16 options_length;
Destination via_list[via_list_length];
Destination destination_list
[destination_list_length];
ForwardingOptions options[options_length];
} ForwardingHeader;
Jennings, et al. Expires May 13, 2010 [Page 40]
Internet-Draft RELOAD Base November 2009
The contents of the structure are:
relo_token: The first four bytes identify this message as a RELOAD
message. The message is easy to demultiplex from STUN messages by
looking at the first bit. This field MUST contain the value
0xc2454c4f (the string 'RELO' with the high bit of the first byte
set.).
overlay: The 32 bit checksum/hash of the overlay being used. The
variable length string representing the overlay name is hashed
with SHA-1 and the low order 32 bits are used. The purpose of
this field is to allow nodes to participate in multiple overlays
and to detect accidental misconfiguration. This is not a security
critical function.
configuration_sequence: The sequence number of the configuration
file.
version: The version of the RELOAD protocol being used. This
document describes version 0.1, with a value of 0x01.
ttl: An 8 bit field indicating the number of iterations, or hops, a
message can experience before it is discarded. The TTL value MUST
be decremented by one at every hop along the route the message
traverses. If the TTL is 0, the message MUST NOT be propagated
further and MUST be discarded, and a "Error_TTL_Exceeded" error
should be generated. The initial value of the TTL SHOULD be 100
unless defined otherwise by the overlay configuration.
fragment: This field is used to handle fragmentation. The high
order two bits are used to indicate the fragmentation status: If
the high bit (0x80000000) is set, it indicates that the message is
a fragment. If the next bit (0x40000000) is set, it indicates
that this is the last fragment. The next six bits (0x20000000 to
0x01000000) are reserved and SHOULD be set to zero. The remainder
of the field is used to indicate the fragment offset; see
Section 5.6.3
length: The count in bytes of the size of the message, including the
header.
transaction_id: A unique 64 bit number that identifies this
transaction and also allows receivers to disambiguate transactions
which are otherwise identical. Responses use the same Transaction
ID as the request they correspond to. Transaction IDs are also
used for fragment reassembly.
Jennings, et al. Expires May 13, 2010 [Page 41]
Internet-Draft RELOAD Base November 2009
max_response_length: The maximum size in bytes of a response. Used
by requesting nodes to avoid receiving (unexpected) very large
responses. If this value is non-zero, responding peers MUST check
that any response would not exceed it and if so generate an
Error_Response_Too_Large value. This value SHOULD be set to zero
for responses.
via_list_length: The length of the via list in bytes. Note that in
this field and the following two length fields we depart from the
usual variable-length convention of having the length immediately
precede the value in order to make it easier for hardware decoding
engines to quickly determine the length of the header.
destination_list_length: The length of the destination list in
bytes.
options_length: The length of the header options in bytes.
via_list: The via_list contains the sequence of destinations through
which the message has passed. The via_list starts out empty and
grows as the message traverses each peer.
destination_list: The destination_list contains a sequence of
destinations which the message should pass through. The
destination list is constructed by the message originator. The
first element in the destination list is where the message goes
next. The list shrinks as the message traverses each listed peer.
options: Contains a series of ForwardingOptions entries. See
Section 5.3.2.3.
5.3.2.1. Processing Configuration Sequence Numbers
In order to be part of the overlay, a node MUST have a copy of the
overlay configuration document. In order to allow for configuration
document changes, each version of the configuration document has a
sequence number which is monotonically increasing mod 65536. Because
the sequence number may in principle wrap, greater than or less than
are interpreted by modulo arithmetic as in TCP.
When a destination node receives a request, it MUST check that the
configuration_sequence field is equal to its own configuration
sequence number. If they do not match, it MUST generate an error,
either Error_Config_Too_Old or Error_Config_Too_New. In addition, if
the configuration file in the request is too old, it MUST generate a
Config_Update message to update the requesting node. This allows new
configuration documents to propagate quickly throughout the system.
Jennings, et al. Expires May 13, 2010 [Page 42]
Internet-Draft RELOAD Base November 2009
The one exception to this rule is that if the configuration_sequence
field is equal to 0xffff, and the message type is Config_Update, then
the message MUST be accepted regardless of the receiving node's
configuration sequence number.
5.3.2.2. Destination and Via Lists
The destination list and via lists are sequences of Destination
values:
enum {reserved(0), node(1), resource(2), compressed(3),
/* 128-255 not allowed */ (255) }
DestinationType;
select (destination_type) {
case node:
NodeId node_id;
case resource:
ResourceId resource_id;
case compressed:
opaque compressed_id<0..2^8-1>;
/* This structure may be extended with new types */
} DestinationData;
struct {
DestinationType type;
uint8 length;
DestinationData destination_data;
} Destination;
struct {
uint16 compressed_id; /* top bit MUST be 1 */
} Destination;
If destination structure has its first bit set to 1, then it is a 16
bit integer. If the first bit is not set, then it is a structure
starting with DestinationType. If it is a 16 bit integer, it is
treated as if it were a full structure with a DestinationType of
compressed and a compressed_id that was 2 bytes long with the value
of the 16 bit integer. When the destination structure is not a 16
Jennings, et al. Expires May 13, 2010 [Page 43]
Internet-Draft RELOAD Base November 2009
bit integer, it is the TLV structure with the following contents:
type
The type of the DestinationData PDU. This may be one of "peer",
"resource", or "compressed".
length
The length of the destination_data.
destination_value
The destination value itself, which is an encoded DestinationData
structure, depending on the value of "type".
Note: This structure encodes a type, length, value. The length
field specifies the length of the DestinationData values, which
allows the addition of new DestinationTypes. This allows an
implementation which does not understand a given DestinationType
to skip over it.
A DestinationData can be one of three types:
peer
A Node-ID.
compressed
A compressed list of Node-IDs and/or resources. Because this
value was compressed by one of the peers, it is only meaningful to
that peer and cannot be decoded by other peers. Thus, it is
represented as an opaque string.
resource
The Resource-ID of the resource which is desired. This type MUST
only appear in the final location of a destination list and MUST
NOT appear in a via list. It is meaningless to try to route
through a resource.
One possible encoding of the 16 bit integer version as an opaque
identifier is to encode an index into a connection table. To avoid
misrouting responses in the event a response is delayed and the
connection table entry has changed, the identifier should be split
between an index and a generation counter for that index. At
startup, the generation counters should be initialized to random
values. An implementation could use 12 bits for the connection table
index and 3 bits for the generation counter. (Note that this does
not suggest a 4096 entry connection table for every node, only the
ability to encode for a larger connection table.) When a connection
table slot is used for a new connection, the generation counter is
incremented (with wrapping). Connection table slots are used on a
Jennings, et al. Expires May 13, 2010 [Page 44]
Internet-Draft RELOAD Base November 2009
rotating basis to maximize the time interval between uses of the same
slot for different connections. When routing a message to an entry
in the destination list encoding a connection table entry, the node
confirms that the generation counter matches the current generation
counter of that index before forwarding the message. If it does not
match, the message is silently dropped.
Regardless of how the 16 bit integer field or opaque DestinationType
is used, the encoding MUST include a generation counter designed to
prevent misrouting of responses due to the connection table entry
having changed since the request message was originally forwarded.
5.3.2.3. Forwarding Options
The Forwarding header can be extended with forwarding header options,
which are a series of ForwardingOptions structures:
enum { (255) } ForwardingOptionsType;
struct {
ForwardingOptionsType type;
uint8 flags;
uint16 length;
select (type) {
/* Option values go here */
} option;
} ForwardingOption;
Each ForwardingOption consists of the following values:
type
The type of the option. This structure allows for unknown options
types.
length
The length of the rest of the structure.
flags
Three flags are defined FORWARD_CRITICAL(0x01),
DESTINATION_CRITICAL(0x02), and RESPONSE_COPY(0x04). These flags
MUST NOT be set in a response. If the FORWARD_CRITICAL flag is
set, any node that would forward the message but does not
understand this options MUST reject the request with an
Error_Unsupported_Forwarding_Option error response. If the
DESTINATION_CRITICAL flag is set, any node that generates a
response to the message but does not understand the forwarding
Jennings, et al. Expires May 13, 2010 [Page 45]
Internet-Draft RELOAD Base November 2009
option MUST reject the request with an
Error_Unsupported_Forwarding_Option error response. If the
RESPONSE_COPY flag is set, any node generating a response MUST
copy the option from the request to the response and clear the
RESPONSE_COPY, FORWARD_CRITICAL and DESTINATION_CRITICAL flags.
option
The option value.
5.3.3. Message Contents Format
The second major part of a RELOAD message is the contents part, which
is defined by MessageContents:
enum { (2^16-1) } MessageExtensionType;
struct {
MessageExtensionType type;
Boolean critical;
opaque extension_contents<0..2^32-1>;
} MessageExtension;
struct {
MessageCode message_code;
opaque message_body<0..2^32-1>;
MessageExtensions extensions<0..2^32-1>;
} MessageContents;
The contents of this structure are as follows:
message_code
This indicates the message that is being sent. The code space is
broken up as follows.
0 Reserved
1 .. 0x7fff Requests and responses. These code points are always
paired, with requests being odd and the corresponding response
being the request code plus 1. Thus, "probe_request" (the
Probe request) has value 1 and "probe_answer" (the Probe
response) has value 2
0xffff Error
Jennings, et al. Expires May 13, 2010 [Page 46]
Internet-Draft RELOAD Base November 2009
message_body
The message body itself, represented as a variable-length string
of bytes. The bytes themselves are dependent on the code value.
See the sections describing the various RELOAD methods (Join,
Update, Attach, Store, Fetch, etc.) for the definitions of the
payload contents.
extensions
Extensions to the message. Currently no extensions are defined,
but new extensions can be defined by the process described in
Section 13.11.
All extensions have the following form:
type
The extension type.
critical
Whether this extension must be understood in order to process the
message. If critical = True and the recipient does not understand
the message, it MUST generate an Error_Unknown_Extension error.
If critical = False, the recipient SHOULD choose to process the
message even if it does not understand the extension.
extension_contents
The contents of the extension (extension-dependent).
5.3.3.1. Response Codes and Response Errors
A peer processing a request returns its status in the message_code
field. If the request was a success, then the message code is the
response code that matches the request (i.e., the next code up). The
response payload is then as defined in the request/response
descriptions.
If the request has failed, then the message code is set to 0xffff
(error) and the payload MUST be an error_response PDU, as shown
below.
When the message code is 0xffff, the payload MUST be an
ErrorResponse.
public struct {
uint16 error_code;
opaque error_info<0..2^16-1>;
} ErrorResponse;
Jennings, et al. Expires May 13, 2010 [Page 47]
Internet-Draft RELOAD Base November 2009
The contents of this structure are as follows:
error_code
A numeric error code indicating the error that occurred.
error_info
An optional arbitrary byte string. Unless otherwise specified,
this will be a text string providing further information about
what went wrong.
The following error code values are defined. The numeric values for
these are defined in Section 13.7.
Error_Forbidden: The requesting peer does not have permission to
make this request.
Error_Not_Found: The resource or peer cannot be found or does not
exist.
Error_Request_Timeout: A response to the request has not been
received in a suitable amount of time. The requesting peer MAY
resend the request at a later time.
Error_Data_Too_Old: A store cannot be completed because the
storage_time precedes the existing value.
Error_Generation_Counter_Too_Low: A store cannot be completed
because the generation counter precedes the existing value.
Error_Incompatible_with_Overlay: A peer receiving the request is
using a different overlay, overlayalgorithm, or hash algorithm.
Error_Unsupported_Forwarding_Option: A peer receiving the request
with a forwarding options flagged as critical but the peer does
not support this option. See section Section 5.3.2.3.
Error_TTL_Exceeded: A peer receiving the request where the TTL got
decremented to zero. See section Section 5.3.2.
Error_Message_Too_Large: A peer receiving the request that was too
large. See section Section 5.6.
Jennings, et al. Expires May 13, 2010 [Page 48]
Internet-Draft RELOAD Base November 2009
Error_Response_Too_Large: A peer would have generated a response
that is too large per the max_response_length field.
Error_Config_Too_Old: A destination peer received a request with a
configuration sequence that's too old.
Error_Config_Too_New: A destination node received a request with a
configuration sequence that's too new. A node which receives this
error MUST generate a Config_Update message to send a new copy of
the configuration document to the node which generated the error.
Error_Unknown_Kind: A destination node received a request with an
unknown kind-id. A node which receives this error MUST generate a
Config_Update message which contains the appropriate kind
definition.
Error_Unknown_Extension: A destination node received a request with
an unknown extension.
5.3.4. Security Block
The third part of a RELOAD message is the security block. The
security block is represented by a SecurityBlock structure:
enum { x509(0), (255) } certificate_type;
struct {
certificate_type type;
opaque certificate<0..2^16-1>;
} GenericCertificate;
struct {
GenericCertificate certificates<0..2^16-1>;
Signature signature;
} SecurityBlock;
The contents of this structure are:
certificates
A bucket of certificates.
signature
A signature over the message contents.
The certificates bucket SHOULD contain all the certificates necessary
to verify every signature in both the message and the internal
message objects. This is the only location in the message which
contains certificates, thus allowing for only a single copy of each
Jennings, et al. Expires May 13, 2010 [Page 49]
Internet-Draft RELOAD Base November 2009
certificate to be sent. In systems which have some alternate
certificate distribution mechanism, some certificates MAY be omitted.
However, implementors should note that this creates the possibility
that messages may not be immediately verifiable because certificates
must first be retrieved.
Each certificate is represented by a GenericCertificate structure,
which has the following contents:
type
The type of the certificate. Only one type is defined: x509
representing an X.509 certificate.
certificate
The encoded version of the certificate. For X.509 certificates,
it is the DER form.
The signature is computed over the payload and parts of the
forwarding header. The payload, in case of a Store, may contain an
additional signature computed over a StoreReq structure. All
signatures are formatted using the Signature element. This element
is also used in other contexts where signatures are needed. The
input structure to the signature computation varies depending on the
data element being signed.
enum {reserved(0), cert_hash(1), (255)} SignerIdentityType;
select (identity_type) {
case cert_hash;
HashAlgorithm hash_alg;
opaque certificate_hash<0..2^8-1>;
/* This structure may be extended with new types if necessary*/
} SignerIdentityValue;
struct {
SignerIdentityType identity_type;
uint16 length;
SignerIdentityValue identity[SignerIdentity.length];
} SignerIdentity;
struct {
SignatureAndHashAlgorithm algorithm;
SignerIdentity identity;
opaque signature_value<0..2^16-1>;
} Signature;
Jennings, et al. Expires May 13, 2010 [Page 50]
Internet-Draft RELOAD Base November 2009
The signature construct contains the following values:
algorithm
The signature algorithm in use. The algorithm definitions are
found in the IANA TLS SignatureAlgorithm Registry.
identity
The identity used to form the signature.
signature_value
The value of the signature.
The only currently permitted identity format is a hash of the
signer's certificate. The hash_alg field is used to indicate the
algorithm used to produce the hash. The certificate_hash contains
the hash of the certificate object. The SignerIdentity structure is
typed purely to allow for future (unanticipated) extensibility.
For signatures over messages the input to the signature is computed
over:
overlay + transaction_id + MessageContents + SignerIdentity
where overlay and transaction_id come from the forwarding header and
+ indicates concatenation.
The input to signatures over data values is different, and is
described in Section 6.1.
All RELOAD messages MUST be signed. Upon receipt, the receiving node
MUST verify the signature and the authorizing certificate. This
check provides a minimal level of assurance that the sending node is
a valid part of the overlay as well as cryptographic authentication
of the sending node. In addition, responses MUST be checked as
follows:
1. The response to a message sent to a specific Node-Id MUST have
been sent by that Node-Id.
2. The response to a message sent to a Resource-Id MUST have been
sent by a Node-Id which is as close to or closer to the target
Resource-Id than any node in the requesting node's neighbor
table.
The second condition serves as a primitive check for responses from
wildly wrong nodes but is not a complete check. Note that in periods
of churn, it is possible for the requesting node to obtain a closer
neighbor while the request is outstanding. This will cause the
Jennings, et al. Expires May 13, 2010 [Page 51]
Internet-Draft RELOAD Base November 2009
response to be rejected and the request to be retransmitted.
In addition, some methods (especially Store) have additional
authentication requirements, which are described in the sections
covering those methods.
5.4. Overlay Topology
As discussed in previous sections, RELOAD does not itself implement
any overlay topology. Rather, it relies on Topology Plugins, which
allow a variety of overlay algorithms to be used while maintaining
the same RELOAD core. This section describes the requirements for
new topology plugins and the methods that RELOAD provides for overlay
topology maintenance.
5.4.1. Topology Plugin Requirements
When specifying a new overlay algorithm, at least the following need
to be described:
o Joining procedures, including the contents of the Join message.
o Stabilization procedures, including the contents of the Update
message, the frequency of topology probes and keepalives, and the
mechanism used to detect when peers have disconnected.
o Exit procedures, including the contents of the Leave message.
o The length of the Resource-IDs and Node-IDs. For DHTs, the hash
algorithm to compute the hash of an identifier.
o The procedures that peers use to route messages.
o The replication strategy used to ensure data redundancy.
All overlay algorithms MUST specify maintenance procedures that send
Updates to all members of the Connection Table whenever the range of
IDs for which the peer is responsible changes. This Update allows
clients and peers that have established connections to the peer
responsible for a particular ID to update that connection as
appropriate.
5.4.2. Methods and types for use by topology plugins
This section describes the methods that topology plugins use to join,
leave, and maintain the overlay.
5.4.2.1. Join
A new peer (but one that already has credentials) uses the JoinReq
message to join the overlay. The JoinReq is sent to the responsible
peer depending on the routing mechanism described in the topology
plugin. This notifies the responsible peer that the new peer is
Jennings, et al. Expires May 13, 2010 [Page 52]
Internet-Draft RELOAD Base November 2009
taking over some of the overlay and it needs to synchronize its
state.
struct {
NodeId joining_peer_id;
opaque overlay_specific_data<0..2^16-1>;
} JoinReq;
The minimal JoinReq contains only the Node-ID which the sending peer
wishes to assume. Overlay algorithms MAY specify other data to
appear in this request.
If the request succeeds, the responding peer responds with a JoinAns
message, as defined below:
struct {
opaque overlay_specific_data<0..2^16-1>;
} JoinAns;
If the request succeeds, the responding peer MUST follow up by
executing the right sequence of Stores and Updates to transfer the
appropriate section of the overlay space to the joining peer. In
addition, overlay algorithms MAY define data to appear in the
response payload that provides additional info.
In general, nodes which cannot form connections SHOULD report an
error. However, implementations MUST provide some mechanism whereby
nodes can determine that they are potentially the first node and take
responsibility for the overlay. This specification does not mandate
any particular mechanism, but a configuration flag or setting seems
appropriate.
5.4.2.2. Leave
The LeaveReq message is used to indicate that a node is exiting the
overlay. A node SHOULD send this message to each peer with which it
is directly connected prior to exiting the overlay.
public struct {
NodeId leaving_peer_id;
opaque overlay_specific_data<0..2^16-1>;
} LeaveReq;
LeaveReq contains only the Node-ID of the leaving peer. Overlay
algorithms MAY specify other data to appear in this request.
Jennings, et al. Expires May 13, 2010 [Page 53]
Internet-Draft RELOAD Base November 2009
Upon receiving a Leave request, a peer MUST update its own routing
table, and send the appropriate Store/Update sequences to re-
stabilize the overlay.
5.4.2.3. Update
Update is the primary overlay-specific maintenance message. It is
used by the sender to notify the recipient of the sender's view of
the current state of the overlay (its routing state), and it is up to
the recipient to take whatever actions are appropriate to deal with
the state change. In general, peers MUST send Update messages to all
their adjacencies whenever they detect a topology shift.
When a peer detects through an Update that it is no longer
responsible for any data value it is storing, it MUST attempt to
Store a copy to the correct node unless it knows the the newly
responsible node already has a copy of the data. This prevents data
loss during large-scale topology shifts such as the merging of
partitioned overlays.
The contents of the UpdateReq message are completely overlay-
specific. The UpdateAns response is expected to be either success or
an error.
5.4.2.4. Route_Query
The Route_Query request allows the sender to ask a peer where they
would route a message directed to a given destination. In other
words, a RouteQuery for a destination X requests the Node-ID for the
node that the receiving peer would next route to in order to get to
X. A RouteQuery can also request that the receiving peer initiate an
Update request to transfer the receiving peer's routing table.
One important use of the RouteQuery request is to support iterative
routing. The sender selects one of the peers in its routing table
and sends it a RouteQuery message with the destination_object set to
the Node-ID or Resource-ID it wishes to route to. The receiving peer
responds with information about the peers to which the request would
be routed. The sending peer MAY then use the Attach method to attach
to that peer(s), and repeat the RouteQuery. Eventually, the sender
gets a response from a peer that is closest to the identifier in the
destination_object as determined by the topology plugin. At that
point, the sender can send messages directly to that peer.
5.4.2.4.1. Request Definition
A RouteQueryReq message indicates the peer or resource that the
requesting peer is interested in. It also contains a "send_update"
Jennings, et al. Expires May 13, 2010 [Page 54]
Internet-Draft RELOAD Base November 2009
option allowing the requesting peer to request a full copy of the
other peer's routing table.
struct {
Boolean send_update;
Destination destination;
opaque overlay_specific_data<0..2^16-1>;
} RouteQueryReq;
The contents of the RouteQueryReq message are as follows:
send_update
A single byte. This may be set to "true" to indicate that the
requester wishes the responder to initiate an Update request
immediately. Otherwise, this value MUST be set to "false".
destination
The destination which the requester is interested in. This may be
any valid destination object, including a Node-ID, compressed ids,
or Resource-ID.
overlay_specific_data
Other data as appropriate for the overlay.
5.4.2.4.2. Response Definition
A response to a successful RouteQueryReq request is a RouteQueryAns
message. This is completely overlay specific.
5.4.2.5. Probe
Probe provides primitive "exploration" services: it allows node to
determine which resources another node is responsible for; and it
allows some discovery services using multicast, anycast, or
broadcast. A probe can be addressed to a specific Node-ID, or the
peer controlling a given location (by using a resource ID). In
either case, the target Node-IDs respond with a simple response
containing some status information.
5.4.2.5.1. Request Definition
The ProbeReq message contains a list (potentially empty) of the
pieces of status information that the requester would like the
responder to provide.
Jennings, et al. Expires May 13, 2010 [Page 55]
Internet-Draft RELOAD Base November 2009
enum { responsible_set(1), num_resources(2), uptime(3), (255)}
ProbeInformationType;
struct {
ProbeInformationType requested_info<0..2^8-1>;
} ProbeReq
The two currently defined values for ProbeInformation are:
responsible_set
indicates that the peer should Respond with the fraction of the
overlay for which the responding peer is responsible.
num_resources
indicates that the peer should Respond with the number of
resources currently being stored by the peer.
uptime
indicates that the peer should Respond with how long the peer has
been up in seconds.
5.4.2.5.2. Response Definition
A successful ProbeAns response contains the information elements
requested by the peer.
Jennings, et al. Expires May 13, 2010 [Page 56]
Internet-Draft RELOAD Base November 2009
struct {
select (type) {
case responsible_set:
uint32 responsible_ppb;
case num_resources:
uint32 num_resources;
case uptime:
uint32 uptime;
/* This type may be extended */
};
} ProbeInformationData;
struct {
ProbeInformationType type;
uint8 length;
ProbeInformationData value;
} ProbeInformation;
struct {
ProbeInformation probe_info<0..2^16-1>;
} ProbeAns;
A ProbeAns message contains a sequence of ProbeInformation
structures. Each has a "length" indicating the length of the
following value field. This structure allows for unknown options
types.
Each of the current possible Probe information types is a 32-bit
unsigned integer. For type "responsible_ppb", it is the fraction of
the overlay for which the peer is responsible in parts per billion.
For type "num_resources", it is the number of resources the peer is
storing. For the type "uptime" it is the number of seconds the peer
has been up.
The responding peer SHOULD include any values that the requesting
peer requested and that it recognizes. They SHOULD be returned in
the requested order. Any other values MUST NOT be returned.
5.5. Forwarding and Link Management Layer
Each node maintains connections to a set of other nodes defined by
the topology plugin. This section defines the methods RELOAD uses to
form and maintain connections between nodes in the overlay. Three
Jennings, et al. Expires May 13, 2010 [Page 57]
Internet-Draft RELOAD Base November 2009
methods are defined:
Attach: used to form RELOAD connections between nodes. When node
A wants to connect to node B, it sends an Attach message to node B
through the overlay. The Attach contains A's ICE parameters. B
responds with its ICE parameters and the two nodes perform ICE to
form connection.
AttachLite: like attach, it is used to form connections between
nodes but instead of using full ICE, it only uses a subset known
as ICE-Lite.
AppAttach: used to form application layer connections between
nodes.
AppAttachLite: like AppAttach but uses ICE-Lite.
Ping: is a simple request/response which is used to verify
connectivity of the target peer.
5.5.1. Attach
A node sends an Attach request when it wishes to establish a direct
TCP or UDP connection to another node for the purpose of sending
RELOAD messages.
As described in Section 5.1, an Attach may be routed to either a
Node-ID or to a Resource-ID. An Attach routed to a specific Node-ID
will fail if that node is not reached. An Attach routed to a
Resource-ID will establish a connection with the peer currently
responsible for that Resource-ID, which may be useful in establishing
a direct connection to the responsible peer for use with frequent or
large resource updates.
An Attach in and of itself does not result in updating the routing
table of either node. That function is performed by Updates. If
node A has Attached to node B, but not received any Updates from B,
it MAY route messages which are directly addressed to B through that
channel but MUST NOT route messages through B to other peers via that
channel. The process of Attaching is separate from the process of
becoming a peer (using Join and Update), to prevent half-open states
where a node has started to form connections but is not really ready
to act as a peer. Thus, clients (unlike peers) can simply Attach
without sending Join or Update.
5.5.1.1. Request Definition
An AttachReq message contains the requesting peer's ICE connection
parameters formatted into a binary structure.
Jennings, et al. Expires May 13, 2010 [Page 58]
Internet-Draft RELOAD Base November 2009
enum { reserved(0), UDP(1), TCP(2), (255) } Transport;
enum { reserved(0), host(1), srflx(2), prflx(3), relay(4),
(255) } CandType;
struct {
opaque name<2^16-1>;
opaque value<2^16-1>;
} IceExtension;
struct {
IpAddressPort addr_port;
Transport transport;
opaque foundation<0..255>;
uint32 priority;
CandType type;
select (type){
case host:
; /* Nothing */
case srflx:
case prflx:
case relay:
IpAddressPort rel_addr_port;
}
IceExtension extensions<0..2^16-1>;
} IceCandidate;
struct {
opaque ufrag<0..2^8-1>;
opaque password<0..2^8-1>;
opaque role<0..2^8-1>;
IceCandidate candidates<0..2^16-1>;
} AttachReqAns;
The values contained in AttachReq and AttachAns are:
ufrag
The username fragment (from ICE).
password
The ICE password.
role
Jennings, et al. Expires May 13, 2010 [Page 59]
Internet-Draft RELOAD Base November 2009
An active/passive/actpass attribute from RFC 4145 [RFC4145].
candidates
One or more ICE candidate values, as described below.
Each ICE candidate is represented as an IceCandidate structure, which
is a direct translation of the information from the ICE string
structures, with the exception of the component ID. Since there is
only one component, it is always 1. The remaining values are
specified as follows:
addr_port
corresponds to the connection-address and port productions.
transport
corresponds to the transport production. New transports such as
SCTP or [I-D.baset-tsvwg-tcp-over-udp] can be added be defining
new Transport values in the IANA registry in Section 13.8.
foundation
corresponds to the foundation production.
priority
corresponds to the priority production.
type
corresponds to the cand-type production.
rel_addr_port
corresponds to the rel-addr and rel-port productions. Only
present for type "relay".
extensions
ICE extensions. The name and value fields correspond to binary
translations of the equivalent fields in the ICE extensions.
These values should be generated using the procedures described in
Section 5.5.1.3.
5.5.1.2. Response Definition
If a peer receives an Attach request, it SHOULD process the request
and generate its own response with a AttachReqAns. It should then
begin ICE checks. When a peer receives an Attach response, it SHOULD
parse the response and begin its own ICE checks.
Jennings, et al. Expires May 13, 2010 [Page 60]
Internet-Draft RELOAD Base November 2009
5.5.1.3. Using ICE With RELOAD
This section describes the profile of ICE that is used with RELOAD.
RELOAD implementations MUST implement full ICE. Because RELOAD
always tries to use TCP and then UDP as a fallback, there will be
multiple candidates of the same IP version, which requires full ICE.
In ICE as defined by [I-D.ietf-mmusic-ice], SDP is used to carry the
ICE parameters. In RELOAD, this function is performed by a binary
encoding in the Attach method. This encoding is more restricted than
the SDP encoding because the RELOAD environment is simpler:
o Only a single media stream is supported.
o In this case, the "stream" refers not to RTP or other types of
media, but rather to a connection for RELOAD itself or for SIP
signaling.
o RELOAD only allows for a single offer/answer exchange. Unlike the
usage of ICE within SIP, there is never a need to send a
subsequent offer to update the default candidates to match the
ones selected by ICE.
An agent follows the ICE specification as described in
[I-D.ietf-mmusic-ice] and [I-D.ietf-mmusic-ice-tcp] with the changes
and additional procedures described in the subsections below.
5.5.1.4. Collecting STUN Servers
ICE relies on the node having one or more STUN servers to use. In
conventional ICE, it is assumed that nodes are configured with one or
more STUN servers through some out-of-band mechanism. This is still
possible in RELOAD but RELOAD also learns STUN servers as it connects
to other peers. Because all RELOAD peers implement ICE and use STUN
keepalives, every peer is a STUN server [RFC5389]. Accordingly, any
peer a node knows will be willing to be a STUN server -- though of
course it may be behind a NAT.
A peer on a well-provisioned wide-area overlay will be configured
with one or more bootstrap nodes. These nodes make an initial list
of STUN servers. However, as the peer forms connections with
additional peers, it builds more peers it can use as STUN servers.
Because complicated NAT topologies are possible, a peer may need more
than one STUN server. Specifically, a peer that is behind a single
NAT will typically observe only two IP addresses in its STUN checks:
its local address and its server reflexive address from a STUN server
outside its NAT. However, if there are more NATs involved, it may
learn additional server reflexive addresses (which vary based on
where in the topology the STUN server is). To maximize the chance of
Jennings, et al. Expires May 13, 2010 [Page 61]
Internet-Draft RELOAD Base November 2009
achieving a direct connection, a peer SHOULD group other peers by the
peer-reflexive addresses it discovers through them. It SHOULD then
select one peer from each group to use as a STUN server for future
connections.
Only peers to which the peer currently has connections may be used.
If the connection to that host is lost, it MUST be removed from the
list of stun servers and a new server from the same group SHOULD be
selected.
5.5.1.5. Gathering Candidates
When a node wishes to establish a connection for the purposes of
RELOAD signaling or SIP signaling (or any other application protocol
for that matter), it follows the process of gathering candidates as
described in Section 4 of ICE [I-D.ietf-mmusic-ice]. RELOAD utilizes
a single component, as does SIP. Consequently, gathering for these
"streams" requires a single component.
An agent MUST implement ICE-tcp [I-D.ietf-mmusic-ice], and MUST
gather at least one UDP and one TCP host candidate for RELOAD and for
SIP.
The ICE specification assumes that an ICE agent is configured with,
or somehow knows of, TURN and STUN servers. RELOAD provides a way
for an agent to learn these by querying the overlay, as described in
Section 5.5.1.4 and Section 8.
The agent SHOULD prioritize its TCP-based candidates over its UDP-
based candidates in the prioritization described in Section 4.1.2 of
ICE [I-D.ietf-mmusic-ice].
The default candidate selection described in Section 4.1.3 of ICE is
ignored; defaults are not signaled or utilized by RELOAD.
5.5.1.6. Encoding the Attach Message
Section 4.3 of ICE describes procedures for encoding the SDP for
conveying RELOAD or SIP ICE candidates. Instead of actually encoding
an SDP, the candidate information (IP address and port and transport
protocol, priority, foundation, component ID, type and related
address) is carried within the attributes of the Attach request or
its response. Similarly, the username fragment and password are
carried in the Attach message or its response. Section 5.5.1
describes the detailed attribute encoding for Attach. The Attach
request and its response do not contain any default candidates or the
ice-lite attribute, as these features of ICE are not used by RELOAD.
Jennings, et al. Expires May 13, 2010 [Page 62]
Internet-Draft RELOAD Base November 2009
Since the Attach request contains the candidate information and short
term credentials, it is considered as an offer for a single media
stream that happens to be encoded in a format different than SDP, but
is otherwise considered a valid offer for the purposes of following
the ICE specification. Similarly, the Attach response is considered
a valid answer for the purposes of following the ICE specification.
5.5.1.7. Verifying ICE Support
An agent MUST skip the verification procedures in Section 5.1 and 6.1
of ICE. Since RELOAD requires full ICE from all agents, this check
is not required.
5.5.1.8. Role Determination
The roles of controlling and controlled as described in Section 5.2
of ICE are still utilized with RELOAD. However, the offerer (the
entity sending the Attach request) will always be controlling, and
the answerer (the entity sending the Attach response) will always be
controlled. The connectivity checks MUST still contain the ICE-
CONTROLLED and ICE-CONTROLLING attributes, however, even though the
role reversal capability for which they are defined will never be
needed with RELOAD. This is to allow for a common codebase between
ICE for RELOAD and ICE for SDP.
5.5.1.9. Connectivity Checks
The processes of forming check lists in Section 5.7 of ICE,
scheduling checks in Section 5.8, and checking connectivity checks in
Section 7 are used with RELOAD without change.
5.5.1.10. Concluding ICE
The controlling agent MUST utilize regular nomination. This is to
ensure consistent state on the final selected pairs without the need
for an updated offer, as RELOAD does not generate additional offer/
answer exchanges.
The procedures in Section 8 of ICE are followed to conclude ICE, with
the following exceptions:
o The controlling agent MUST NOT attempt to send an updated offer
once the state of its single media stream reaches Completed.
o Once the state of ICE reaches Completed, the agent can immediately
free all unused candidates. This is because RELOAD does not have
the concept of forking, and thus the three second delay in Section
8.3 of ICE does not apply.
Jennings, et al. Expires May 13, 2010 [Page 63]
Internet-Draft RELOAD Base November 2009
5.5.1.11. Subsequent Offers and Answers
An agent MUST NOT send a subsequent offer or answer. Thus, the
procedures in Section 9 of ICE MUST be ignored.
5.5.1.12. Media Keepalives
STUN MUST be utilized for the keepalives described in Section 10 of
ICE.
5.5.1.13. Sending Media
The procedures of Section 11 apply to RELOAD as well. However, in
this case, the "media" takes the form of application layer protocols
(RELOAD or SIP for example) over TLS or DTLS. Consequently, once ICE
processing completes, the agent will begin TLS or DTLS procedures to
establish a secure connection. The node which sent the Attach
request MUST be the TLS server. The other node MUST be the TLS
client. The server MUST request TLS client authentication. The
nodes MUST verify that the certificate presented in the handshake
matches the identity of the other peer as found in the Attach
message. Once the TLS or DTLS signaling is complete, the application
protocol is free to use the connection.
The concept of a previous selected pair for a component does not
apply to RELOAD, since ICE restarts are not possible with RELOAD.
5.5.1.14. Receiving Media
An agent MUST be prepared to receive packets for the application
protocol (TLS or DTLS carrying RELOAD, SIP or anything else) at any
time. The jitter and RTP considerations in Section 11 of ICE do not
apply to RELOAD or SIP.
5.5.2. AttachLite
An alternative to using the full ICE supported by the Attach request
is to use ICE-Lite with the AttachLite request. This will not work
in all of the scenarios where ICE would work, but in some cases,
particularly those with no NATs or firewalls, it will work.
Configuration for the overlay indicates whether or not this can be
used.
5.5.2.1. Request Definition
An AttachLiteReqAns message contains the requesting peer's ICE-Lite
connection parameters formatted into a binary structure. When using
the AttachLite request, both sides act as ICE-Lite hosts.
Jennings, et al. Expires May 13, 2010 [Page 64]
Internet-Draft RELOAD Base November 2009
struct {
IpAddressPort addr_port;
Transport transport;
uint32 priority;
} IceLiteCandidate;
struct {
IceLiteCandidate candidates<0..2^16-1>;
} AttachLiteReqAns;
The values contained in AttachLiteReqAns are:
candidates
One or more ICE candidate values. Each one contains an IP address
and family, transport protocol, and port to connect to as well as
a priority.
These values should be generated using the procedures described in
Section 5.5.1.3.
5.5.2.2. Response Definition
If a peer receives an AttachLite request, it SHOULD process the
request and generate its own response with an AttachLiteReqAns. It
should then initiate connections as described below. When a peer
receives an AttachLite response, it SHOULD parse the response and
handle any received connections.
5.5.2.3. Attach-Lite Connectivity Checks
STUN is not used for connectivity checks when doing ICE-Lite; instead
the DTLS or TLS handshake forms the connectivity check. The host
that received the AttachLiteReqAns MUST initiate TLS or DTLS
connections to candidates provided in the request. When a connection
forms, the node MUST check that the certificate is for the node that
sent AttachLiteReqAns and if is not, MUST close the connection.
Since TLS provides the connectivity check, there is no need for the
RFC 4571 [RFC4571] style framing shim for STUN when using TLS.
5.5.2.4. Implementation Notes for Attach-Lite
This is a non-normative section to help implementors.
At times ICE can seem a bit daunting to get one's head around. For a
simple IPv4 only peer, a simple implementation of Attach-Lite could
be done be doing the following:
Jennings, et al. Expires May 13, 2010 [Page 65]
Internet-Draft RELOAD Base November 2009
o When sending an AttachLiteReqAns, form one candidate with a
priority value of (2^24)*(126)+(2^8)*(65535)+(2^0)*(256-1) that
specifies the UDP port being listened to and another one with the
TCP port.
o When receiving an AttachLiteReqAns, try to form a connection to
each candidate in the request. Check the certificate receive in
the TLS handshake has the same Node-ID as the node that has sent
the AttchLiteReq. If multiple connections succeed, close all but
the one with highest priority.
o Do normal TLS and DTLS with no need for any special framing or
STUN processing.
5.5.3. AppAttach
A node sends an AppAttach request when it wishes to establish a
direct TCP or UDP connection to another node for the purposes of
sending application layer messages. AppAttach is basically like
Attach, except for the purpose of the connection. A separate request
is used to avoid implementor confusion between the two methods (this
was found to be a real problem with initial implementations). The
AppAttach request and its response contain an application attribute,
with a value of SIP or RELOAD, which indicates what protocol is to be
run over the connection.
5.5.3.1. Request Definition
An AttachReq message contains the requesting peer's ICE connection
parameters formatted into a binary structure.
struct {
opaque ufrag<0..2^8-1>;
opaque password<0..2^8-1>;
uint16 application;
opaque role<0..2^8-1>;
IceCandidate candidates<0..2^16-1>;
} AppAttachReqAns;
The values contained in AppAttachReq and AppAttachAns are:
ufrag
The username fragment (from ICE)
password
Jennings, et al. Expires May 13, 2010 [Page 66]
Internet-Draft RELOAD Base November 2009
The ICE password.
application
A 16-bit port number. This port number represents the IANA
registered port of the protocol that is going to be sent on this
connection. For SIP, this is 5060 or 5061. By using the IANA
registered port, we avoid the need for an additional registry and
allow RELOAD to be used to set up connections for any existing or
future application protocols.
role
An active/passive/actpass attribute from RFC 4145 [RFC4145].
candidates
One or more ICE candidate values
5.5.3.2. Response Definition
If a peer receives an AppAttach request, it SHOULD process the
request and generate its own response with a AppAttachReqAns. It
should then begin ICE checks. When a peer receives an AppAttach
response, it SHOULD parse the response and begin its own ICE checks.
5.5.4. AppAttachLite
Similar to the AttachLite method, RELOAD provides an AppAttachLite to
allow application connections when full ICE is not needed.
5.5.4.1. Request Definition
An AppAttachLiteReqAns message contains the requesting peer's ICE-
Lite connection parameters formatted into a binary structure. When
using the AppAttachLite request, both sides act as ICE-Lite hosts.
struct {
uint16 application;
IceLiteCandidate candidates<0..2^16-1>;
} AppAttachLiteReqAns;
The values contained in AppAttachLiteReqAns are:
application
A 16-bit port number used in the same way as in the AppAttach
request. This port number represents the IANA registered port of
the protocol that is going to be sent on this connection.
Jennings, et al. Expires May 13, 2010 [Page 67]
Internet-Draft RELOAD Base November 2009
candidates
One or more ICE candidate values. Each one contains an IP address
and family, transport protocol, and port to connect to as well as
a priority.
These values should be generated using the procedures described in
Section 5.5.1.3.
5.5.4.2. Response Definition
If a peer receives an AppAttachLite request, it SHOULD process the
request and generate its own response with an AppAttachLiteReqAns as
described in the AttachLite section.
5.5.5. Ping
Ping is used to test connectivity along a path. A ping can be
addressed to a specific Node-ID, to the peer controlling a given
location (by using a resource ID), or to the broadcast Node-ID
(2^128-1).
5.5.5.1. Request Definition
struct {
} PingReq
5.5.5.2. Response Definition
A successful PingAns response contains the information elements
requested by the peer.
struct {
uint64 response_id;
uint64 time;
} PingAns;
A PingAns message contains the following elements:
response_id
Jennings, et al. Expires May 13, 2010 [Page 68]
Internet-Draft RELOAD Base November 2009
A randomly generated 64-bit response ID. This is used to
distinguish Ping responses.
time
The time when the ping responses was created in absolute time,
represented in milliseconds since midnight Jan 1, 1970 which is
the UNIX epoch.
5.5.6. Config_Update
The Config_Update method is used to push updated configuration data
across the overlay. Whenever a node detects that another node has
old configuration data, it MUST generate a Config_Update request.
The Config_Update request allows updating of two kinds of data: the
configuration data (Section 5.3.2.1) and kind information
(Section 6.4.1.1).
5.5.6.1. Request Definition
enum { reserved(0), config(1), kind(2), (255) }
Config_UpdateType;
typedef opaque KindDescription<2^16-1>;
struct {
Config_UpdateType type;
uint32 length;
select (type) {
case config:
opaque config_data<2^24-1>;
case kind:
KindDescription kinds<2^24-1>;
/* This structure may be extended with new types*/
};
} Config_UpdateReq;
The Config_UpdateReq message contains the following elements:
type
The type of the contents of the message. This structure allows
for unknown content types.
Jennings, et al. Expires May 13, 2010 [Page 69]
Internet-Draft RELOAD Base November 2009
length
The length of the remainder of the message. This is included to
preserve backward compatibility and is 32 bits instead of 24 to
facilitate easy conversion between network and host byte order.
config_data (type==config)
The contents of the configuration document.
kinds (type==kind)
One or more XML kind-block productions (see Section 10.1). These
MUST be encoded with UTF-8 and assume a default namespace of
"urn:ietf:params:xml:ns:p2p:config-base".
5.5.6.2. Response Definition
struct {
} Config_UpdateReq
If the Config_UpdateReq is of type "config" it MUST only be processed
if all the following are true:
o The sequence number in the document is greater than the current
configuration sequence number.
o The configuration document is correctly digitally signed (see
Section 10 for details on signatures.
Otherwise appropriate errors MUST be generated.
If the Config_UpdateReq is of type "kind" it MUST only be processed
if it is correctly digitally signed by an acceptable kind signer. In
addition, if the kind update conflicts with an existing known kind
(i.e., it is signed by a different signer), then it should be
rejected with "Error_Forbidden". This should not happen in correctly
functioning overlays.
If the update is acceptable, then the node MUST reconfigure itself to
match the new information. This may include adding permissions for
new kinds, deleting old kinds, or even, in extreme circumstances,
exiting and reentering the overlay, if, for instance, the DHT
algorithm has changed.
The response for Config_Update is empty.
5.6. Overlay Link Layer
RELOAD can use multiple Overlay Link protocols to send its messages.
Because ICE is used to establish connections (see Section 5.5.1.3),
RELOAD nodes are able to detect which Overlay Link protocols are
offered by other nodes and establish connections between them. Any
link protocol needs to be able to establish a secure, authenticated
connection and to provide data origin authentication and message
Jennings, et al. Expires May 13, 2010 [Page 70]
Internet-Draft RELOAD Base November 2009
integrity for individual data elements. RELOAD currently supports
two Overlay Link protocols:
o TLS [RFC5246] over TCP
o DTLS [RFC4347] over UDP
Note that although UDP does not properly have "connections", both TLS
and DTLS have a handshake which establishes a similar, stateful
association, and we simply refer to these as "connections" for the
purposes of this document.
If a peer receives a message that is larger than value of max-
message-size defined in the overlay configuration, the peer SHOULD
send an Error_Message_Too_Large error and then close the TLS or DTLS
session from which the message was received. Note that this error
can be sent and the session closed before receiving the complete
message. If the forwarding header is larger than the max-message-
size, the receiver SHOULD close the TLS or DTLS session without
sending an error.
5.6.1. Future Support for HIP
The P2PSIP Working Group has expressed interest in supporting a HIP-
based link protocol [RFC5201]. Such support would require specifying
such details as:
o How to issue certificates which provided identities meaningful to
the HIP base exchange. We anticipate that this would require a
mapping between ORCHIDs and NodeIds.
o How to carry the HIP I1 and I2 messages. We anticipate that this
would require defining a HIP Tunnel usage.
o How to carry RELOAD messages over HIP.
We leave this work as a topic for another draft.
5.6.2. Reliability for Unreliable Links
When RELOAD is carried over DTLS or another unreliable link protocol,
it needs to be used with a reliability and congestion control
mechanism, which is provided on a hop-by-hop basis. The basic
principle is that each message, regardless of whether or not it
carries a request or response, will get an ACK and be reliably
retransmitted. The receiver's job is very simple, limited to just
sending ACKs. All the complexity is at the sender side. This allows
the sending implementation to trade off performance versus
implementation complexity without affecting the wire protocol.
In order to support unreliable links, each message is wrapped in a
Jennings, et al. Expires May 13, 2010 [Page 71]
Internet-Draft RELOAD Base November 2009
very simple framing layer (FramedMessage) which is only used for each
hop. This layer contains a sequence number which can then be used
for ACKs.
5.6.2.1. Framed Message Format
The definition of FramedMessage is:
enum {data (128), ack (129), (255)} FramedMessageType;
struct {
FramedMessageType type;
select (type) {
case data:
uint32 sequence;
opaque message<0..2^24-1>;
case ack:
uint32 ack_sequence;
uint32 received;
};
} FramedMessage;
The type field of the PDU is set to indicate whether the message is
data or an acknowledgement.
If the message is of type "data", then the remainder of the PDU is as
follows:
sequence
the sequence number. This increments by 1 for each framed message
sent over this transport session.
message
the message that is being transmitted.
Each connection has it own sequence number space. Initially the
value is zero and it increments by exactly one for each message sent
over that connection.
When the receiver receives a message, it SHOULD immediately send an
ACK message. The receiver MUST keep track of the 32 most recent
sequence numbers received on this association in order to generate
the appropriate ack.
Jennings, et al. Expires May 13, 2010 [Page 72]
Internet-Draft RELOAD Base November 2009
If the PDU is of type "ack", the contents are as follows:
ack_sequence
The sequence number of the message being acknowledged.
received
A bitmask indicating if each of the previous 32 sequence numbers
before this packet has been among the 32 packets most recently
received on this connection. When a packet is received with a
sequence number N, the receiver looks at the sequence number of
the previously 32 packets received on this connection. Call the
previously received packet number M. For each of the previous 32
packets, if the sequence number M is less than N but greater than
N-32, the N-M bit of the received bitmask is set to one; otherwise
it is zero. Note that a bit being set to one indicates positively
that a particular packet was received, but a bit being set to zero
means only that it is unknown whether or not the packet has been
received, because it might have been received before the 32 most
recently received packets.
The received field bits in the ACK provide a very high degree of
redundancy so that the sender to figure out which packets the
receiver has received and can then estimate packet loss rates. If
the sender also keeps track of the time at which recent sequence
numbers have been sent, the RTT can be estimated.
5.6.2.2. Retransmission and Flow Control
Because the receiver's role is limited to providing packet
acknowledgements, a wide variety of congestion control algorithms can
be implemented on the sender side while using the same basic wire
protocol. Senders MUST implement a retransmission and congestion
control scheme no more aggressive then TFRC[RFC5348]. One way to do
that is for senders to implement the scheme in the following section.
Another would be to implement the scheme described in Appendix B.
Another alternative would be TFRC-SP [RFC4828] and use the received
bitmask to allow the sender to compute packet loss event rates.
5.6.2.2.1. Trivial Retransmission
A peer SHOULD retransmit a message if it has not received an ACK
after an interval of RTO ("Retransmission TimeOut"). The peer MUST
double the time to wait after each retransmission. In each
retransmission, the sequence number is incremented.
The RTO is an estimate of the round-trip time (RTT). Implementations
can use a static value for RTO or a dynamic estimate which will
result in better performance. For implementations that use a static
Jennings, et al. Expires May 13, 2010 [Page 73]
Internet-Draft RELOAD Base November 2009
value, the default value for RTO is 500 ms. Nodes MAY use smaller
values of RTO if it is known that all nodes are are within the local
network. The default RTO MAY be chosen larger, and this is
RECOMMENDED if it is known in advance (such as on high latency access
links) that the round-trip time is larger.
Implementations that use a dynamic estimate to compute the RTO MUST
use the algorithm described in RFC 2988[RFC2988], with the exception
that the value of RTO SHOULD NOT be rounded up to the nearest second
but instead rounded up to the nearest millisecond. The RTT of a
successful STUN transaction from the ICE stage is used as the initial
measurement for formula 2.2 of RFC 2988. The sender keeps track of
the time each message was sent for all recently sent messages. Any
time an ACK is received, the sender can compute the RTT for that
message by looking at the time the ACK was received and the time when
the message was sent. This is used as a subsequent RTT measurement
for formula 2.3 of RFC 2988 to update the RTO estimate. (Note that
because retransmissions receive new sequence numbers, all received
ACKs are used.)
The value for RTO is calculated separately for each DTLS session.
Retransmissions continue until a response is received, or until a
total of 5 requests have been sent or there has been a hard ICMP
error [RFC1122]. The sender knows a response was received when it
receives an ACK with a sequence number that indicates it is a
response to one of the transmissions of this messages. For example,
assuming an RTO of 500 ms, requests would be sent at times 0 ms, 500
ms, 1500 ms, 3500 ms, and 7500 ms. If all retransmissions for a
message fail, then the sending node SHOULD close the connection
routing the message.
Once an ACK has been received for a message, the next message can be
sent but the peer SHOULD ensure that there is at least 10 ms between
sending any two messages. The only time a value less than 10 ms can
be used is when it is known that all nodes are on a network that can
support retransmissions faster than 10 ms with no congestion issues.
5.6.3. Fragmentation and Reassembly
In order to allow transmission over datagram protocols such as DTLS,
RELOAD messages may be fragmented.
Any node along the path can fragment the message but only the final
destination reassembles the fragments. When a node takes a packet
and fragments it, each fragment has a full copy of the Forwarding
Header but the data after the Forwarding Header is broken up in
appropriate sized chunks. The size of the payload chunks needs to
Jennings, et al. Expires May 13, 2010 [Page 74]
Internet-Draft RELOAD Base November 2009
take into account space to allow the via and destination lists to
grow. Each fragment MUST contain a full copy of the via and
destination list and MUST contain at least 256 bytes of the message
body. If the via and destination list are so large that this is not
possible, RELOAD fragmentation is not performed and IP-layer
fragmentation is allowed to occur. When a message must be
fragmented, it SHOULD be split into equal-sized fragments that are no
larger than the PMTU of the next overlay link minus 32 bytes. This
is to allow the via list to grow before further fragmentation is
required.
Note that this fragmentation is not optimal for the end-to-end path -
a message may be refragmented multiple times as it traverses the
overlay. This option has been chosen as it is far easier to
implement than e2e PMTU discovery across an ever-changing overlay,
and it effectively addresses the reliability issues of relying on IP-
layer fragmentation. However, PING can be used to allow e2e PMTU to
be implemented if desired.
Upon receipt of a fragmented message by the intended peer, the peer
holds the fragments in a holding buffer until the entire message has
been received. The message is then reassembled into a single message
and processed. In order to mitigate denial of service attacks,
receivers SHOULD time out incomplete fragments after maximum request
lifetime (15 seconds). Note this time was derived from looking at
the end to end retransmission time and saving fragments long enough
for the full end to end retransmissions to take place. Ideally the
receiver would have enough buffer space to deal with as many
fragments as can arrive in the maximum request lifetime. However, if
the receiver runs out of buffer space to reassemble the messages it
MUST drop the message.
When a message is fragmented, the fragment offset value is stored in
the lower 24 bits of the fragment field of the forwarding header.
The offset is the number of bytes between the end of the forwarding
header and the start of the data. The first fragment therefore has
an offset of 0. The first and last bit indicators MUST be
appropriately set. If the message is not fragmented, then both the
first and last fragment are set to 1 and the offset is 0 resulting in
a fragment value of 0xC0000000.
6. Data Storage Protocol
RELOAD provides a set of generic mechanisms for storing and
retrieving data in the Overlay Instance. These mechanisms can be
used for new applications simply by defining new code points and a
small set of rules. No new protocol mechanisms are required.
Jennings, et al. Expires May 13, 2010 [Page 75]
Internet-Draft RELOAD Base November 2009
The basic unit of stored data is a single StoredData structure:
struct {
uint32 length;
uint64 storage_time;
uint32 lifetime;
StoredDataValue value;
Signature signature;
} StoredData;
The contents of this structure are as follows:
length
The size of the StoredData structure in octets excluding the size
of length itself.
storage_time
The time when the data was stored in absolute time, represented in
milliseconds since the Unix epoch of midnight Jan 1, 1970. Any
attempt to store a data value with a storage time before that of a
value already stored at this location MUST generate a
Error_Data_Too_Old error. This prevents rollback attacks. Note
that this does not require synchronized clocks: the receiving
peer uses the storage time in the previous store, not its own
clock.
lifetime
The validity period for the data, in seconds, starting from the
time of store.
value
The data value itself, as described in Section 6.2.
signature
A signature as defined in Section 6.1.
Each Resource-ID specifies a single location in the Overlay Instance.
However, each location may contain multiple StoredData values
distinguished by Kind-ID. The definition of a kind describes both
the data values which may be stored and the data model of the data.
Some data models allow multiple values to be stored under the same
Kind-ID. Section Section 6.2 describes the available data models.
Thus, for instance, a given Resource-ID might contain a single-value
element stored under Kind-ID X and an array containing multiple
values stored under Kind-ID Y.
Jennings, et al. Expires May 13, 2010 [Page 76]
Internet-Draft RELOAD Base November 2009
6.1. Data Signature Computation
Each StoredData element is individually signed. However, the
signature also must be self-contained and cover the Kind-ID and
Resource-ID even though they are not present in the StoredData
structure. The input to the signature algorithm is:
resource_id + kind + storage_time + StoredDataValue +
SignerIdentity
Where these values are:
resource
The resource ID where this data is stored.
kind
The Kind-ID for this data.
storage_time
The contents of the storage_time data value.
StoredDataValue
The contents of the stored data value, as described in the
previous sections.
SignerIdentity
The signer identity as defined in Section 5.3.4.
Once the signature has been computed, the signature is represented
using a signature element, as described in Section 5.3.4.
6.2. Data Models
The protocol currently defines the following data models:
o single value
o array
o dictionary
These are represented with the StoredDataValue structure:
Jennings, et al. Expires May 13, 2010 [Page 77]
Internet-Draft RELOAD Base November 2009
enum { reserved(0), single_value(1), array(2),
dictionary(3), (255)} DataModel;
struct {
Boolean exists;
opaque value<0..2^32-1>;
} DataValue;
struct {
select (DataModel) {
case single_value:
DataValue single_value_entry;
case array:
ArrayEntry array_entry;
case dictionary:
DictionaryEntry dictionary_entry;
/* This structure may be extended */
} ;
} StoredDataValue;
We now discuss the properties of each data model in turn:
6.2.1. Single Value
A single-value element is a simple sequence of bytes. There may be
only one single-value element for each Resource-ID, Kind-ID pair.
A single value element is represented as a DataValue, which contains
the following two elements:
exists
This value indicates whether the value exists at all. If it is
set to False, it means that no value is present. If it is True,
that means that a value is present. This gives the protocol a
mechanism for indicating nonexistence as opposed to emptiness.
value
The stored data.
Jennings, et al. Expires May 13, 2010 [Page 78]
Internet-Draft RELOAD Base November 2009
6.2.2. Array
An array is a set of opaque values addressed by an integer index.
Arrays are zero based. Note that arrays can be sparse. For
instance, a Store of "X" at index 2 in an empty array produces an
array with the values [ NA, NA, "X"]. Future attempts to fetch
elements at index 0 or 1 will return values with "exists" set to
False.
A array element is represented as an ArrayEntry:
struct {
uint32 index;
DataValue value;
} ArrayEntry;
The contents of this structure are:
index
The index of the data element in the array.
value
The stored data.
6.2.3. Dictionary
A dictionary is a set of opaque values indexed by an opaque key with
one value for each key. A single dictionary entry is represented as
follows:
A dictionary element is represented as a DictionaryEntry:
typedef opaque DictionaryKey<0..2^16-1>;
struct {
DictionaryKey key;
DataValue value;
} DictionaryEntry;
The contents of this structure are:
Jennings, et al. Expires May 13, 2010 [Page 79]
Internet-Draft RELOAD Base November 2009
key
The dictionary key for this value.
value
The stored data.
6.3. Access Control Policies
Every kind which is storable in an overlay MUST be associated with an
access control policy. This policy defines whether a request from a
given node to operate on a given value should succeed or fail. It is
anticipated that only a small number of generic access control
policies are required. To that end, this section describes a small
set of such policies and Section 13.3 establishes a registry for new
policies if required. Each policy has a short string identifier
which is used to reference it in the configuration document.
6.3.1. USER-MATCH
In the USER-MATCH policy, a given value MUST be written (or
overwritten) if and only if the request is signed with a key
associated with a certificate whose user name hashes (using the hash
function for the overlay) to the Resource-ID for the resource.
Recall that the certificate may, depending on the overlay
configuration, be self-signed.
6.3.2. NODE-MATCH
In the NODE-MATCH policy, a given value MUST be written (or
overwritten) if and only if the request is signed with a key
associated with a certificate whose Node-ID hashes (using the hash
function for the overlay) to the Resource-ID for the resource.
6.3.3. USER-NODE-MATCH
The USER-NODE-MATCH policy may only be used with dictionary types.
In the USER-NODE-MATCH policy, a given value MUST be written (or
overwritten) if and only if the request is signed with a key
associated with a certificate whose user name hashes (using the hash
function for the overlay) to the Resource-ID for the resource. In
addition, the dictionary key MUST be equal to the Node-ID in the
certificate.
6.3.4. NODE-MULTIPLE
In the NODE-MULTIPLE policy, a given value MUST be written (or
overwritten) if and only if the request is signed with a key
Jennings, et al. Expires May 13, 2010 [Page 80]
Internet-Draft RELOAD Base November 2009
associated with a certificate containing a Node-ID such that
H(Node-ID || i) is equal to the Resource-ID for some small integer
value of i. When this policy is in use, the maximum value of i MUST
be specified in the kind definition.
6.4. Data Storage Methods
RELOAD provides several methods for storing and retrieving data:
o Store values in the overlay
o Fetch values from the overlay
o Stat: get metadata about values in the overlay
o Find the values stored at an individual peer
These methods are each described in the following sections.
6.4.1. Store
The Store method is used to store data in the overlay. The format of
the Store request depends on the data model which is determined by
the kind.
6.4.1.1. Request Definition
A StoreReq message is a sequence of StoreKindData values, each of
which represents a sequence of stored values for a given kind. The
same Kind-ID MUST NOT be used twice in a given store request. Each
value is then processed in turn. These operations MUST be atomic.
If any operation fails, the state MUST be rolled back to before the
request was received.
The store request is defined by the StoreReq structure:
struct {
KindId kind;
uint64 generation_counter;
StoredData values<0..2^32-1>;
} StoreKindData;
struct {
ResourceId resource;
uint8 replica_number;
StoreKindData kind_data<0..2^32-1>;
} StoreReq;
A single Store request stores data of a number of kinds to a single
resource location. The contents of the structure are:
Jennings, et al. Expires May 13, 2010 [Page 81]
Internet-Draft RELOAD Base November 2009
resource
The resource to store at.
replica_number
The number of this replica. When a storing peer saves replicas to
other peers each peer is assigned a replica number starting from 1
and sent in the Store message. This field is set to 0 when a node
is storing its own data. This allows peers to distinguish replica
writes from original writes.
kind_data
A series of elements, one for each kind of data to be stored.
If the replica number is zero, then the peer MUST check that it is
responsible for the resource and, if not, reject the request. If the
replica number is nonzero, then the peer MUST check that it expects
to be a replica for the resource and that the request sender is
consistent with being the responsible node (i.e., that the receiving
peer does not know of a better node) and, if not, reject the request.
Each StoreKindData element represents the data to be stored for a
single Kind-ID. The contents of the element are:
kind
The Kind-ID. Implementations SHOULD reject requests corresponding
to unknown kinds unless specifically configured otherwise.
generation
The expected current state of the generation counter
(approximately the number of times this object has been written;
see below for details).
values
The value or values to be stored. This may contain one or more
stored_data values depending on the data model associated with
each kind.
The peer MUST perform the following checks:
o The kind_id is known and supported.
o The signatures over each individual data element (if any) are
valid. If this check fails, the request MUST be rejected with an
Error_Forbidden error.
o Each element is signed by a credential which is authorized to
write this kind at this Resource-ID. If this check fails, the
request MUST be rejected with an Error_Forbidden error.
Jennings, et al. Expires May 13, 2010 [Page 82]
Internet-Draft RELOAD Base November 2009
o For original (non-replica) stores, the peer MUST check that if the
generation-counter is non-zero, it equals the current value of the
generation-counter for this kind. This feature allows the
generation counter to be used in a way similar to the HTTP Etag
feature.
o The storage time values are greater than that of any value which
would be replaced by this Store.
o The size and number of the stored values is consistent with the
limits specified in the overlay configuration.
If all these checks succeed, the peer MUST attempt to store the data
values. For non-replica stores, if the store succeeds and the data
is changed, then the peer must increase the generation counter by at
least one. If there are multiple stored values in a single
StoreKindData, it is permissible for the peer to increase the
generation counter by only 1 for the entire Kind-ID, or by 1 or more
than one for each value. Accordingly, all stored data values must
have a generation counter of 1 or greater. 0 is used in the Store
request to indicate that the generation counter should be ignored for
processing this request; however the responsible peer should increase
the stored generation counter and should return the correct
generation counter in the response.
For replica Stores, the peer MUST set the generation counter to match
the generation_counter in the message, and MUST NOT check the
generation counter against the current value. Replica Stores MUST
NOT use a generation counter of 0.
When a peer stores data previously stored by another node (e.g., for
replicas or topology shifts) it MUST adjust the lifetime value
downward to reflect the amount of time the value was stored at the
peer.
Unless otherwise specified by the usage, if a peer attempts to store
data previously stored by another node (e.g., for replicas or
topology shifts) and that store fails with either an
Error_Generation_Counter_Too_Low or an Error_Data_Too old error, the
peer MUST fetch the newer data from the the peer generating the error
and use that to replace its own copy. This rule allows
resynchronization after partitions heal.
The properties of stores for each data model are as follows:
Single-value:
Jennings, et al. Expires May 13, 2010 [Page 83]
Internet-Draft RELOAD Base November 2009
A store of a new single-value element creates the element if it
does not exist and overwrites any existing value with the new
value.
Array:
A store of an array entry replaces (or inserts) the given value at
the location specified by the index. Because arrays are sparse, a
store past the end of the array extends it with nonexistent values
(exists=False) as required. A store at index 0xffffffff places
the new value at the end of the array regardless of the length of
the array. The resulting StoredData has the correct index value
when it is subsequently fetched.
Dictionary:
A store of a dictionary entry replaces (or inserts) the given
value at the location specified by the dictionary key.
The following figure shows the relationship between these structures
for an example store which stores the following values at resource
"1234"
o The value "abc" in the single value location for kind X
o The value "foo" at index 0 in the array for kind Y
o The value "bar" at index 1 in the array for kind Y
Jennings, et al. Expires May 13, 2010 [Page 84]
Internet-Draft RELOAD Base November 2009
Store
resource=1234
replica_number = 0
/ \
/ \
StoreKindData StoreKindData
kind=X (Single-Value) kind=Y (Array)
generation_counter = 99 generation_counter = 107
| /\
| / \
StoredData / \
storage_time = xxxxxxx / \
lifetime = 86400 / \
signature = XXXX / \
| | |
| StoredData StoredData
| storage_time = storage_time =
| yyyyyyyy zzzzzzz
| lifetime = 86400 lifetime = 33200
| signature = YYYY signature = ZZZZ
| | |
StoredDataValue | |
value="abc" | |
| |
StoredDataValue StoredDataValue
index=0 index=1
value="foo" value="bar"
6.4.1.2. Response Definition
In response to a successful Store request the peer MUST return a
StoreAns message containing a series of StoreKindResponse elements
containing the current value of the generation counter for each
Kind-ID, as well as a list of the peers where the data will be
replicated.
struct {
KindId kind;
uint64 generation_counter;
NodeId replicas<0..2^16-1>;
} StoreKindResponse;
struct {
StoreKindResponse kind_responses<0..2^16-1>;
} StoreAns;
Jennings, et al. Expires May 13, 2010 [Page 85]
Internet-Draft RELOAD Base November 2009
The contents of each StoreKindResponse are:
kind
The Kind-ID being represented.
generation
The current value of the generation counter for that Kind-ID.
replicas
The list of other peers at which the data was/will be replicated.
In overlays and applications where the responsible peer is
intended to store redundant copies, this allows the storing peer
to independently verify that the replicas have in fact been
stored. It does this verification by using the Stat method. Note
that the storing peer is not require to perform this verification.
The response itself is just StoreKindResponse values packed end-to-
end.
If any of the generation counters in the request precede the
corresponding stored generation counter, then the peer MUST fail the
entire request and respond with an Error_Generation_Counter_Too_Low
error. The error_info in the ErrorResponse MUST be a StoreAns
response containing the correct generation counter for each kind and
the replica list, which will be empty. For original (non-replica)
stores, a node which receives such an error SHOULD attempt to fetch
the data and, if the storage_time value is newer, replace its own
data with that newer data. This rule improves data consistency in
the case of partitions and merges.
If the data being stored is too large for the allowed limit by the
given usage, then the peer MUST fail the request and generate an
Error_Data_Too_Large error.
If any type of request tries to access a data kind that the node does
not know about, an Error_Unknown_Kind MUST be generated. The
error_info in the Error_Response is:
KindId unknown_kinds<2^8-1>;
which lists all the kinds that were unrecognized.
6.4.1.3. Removing Values
This version of RELOAD (unlike previous versions) does not have an
explicit Remove operation. Rather, values are Removed by storing
"nonexistent" values in their place. Each DataValue contains a
Jennings, et al. Expires May 13, 2010 [Page 86]
Internet-Draft RELOAD Base November 2009
boolean value called "exists" which indicates whether a value is
present at that location. In order to effectively remove a value,
the owner stores a new DataValue with:
exists = false
value = {} (0 length)
Storing nodes MUST treat these nonexistent values the same way they
treat any other stored value, including overwriting the existing
value, replicating them, and aging them out as necessary when
lifetime expires. When a stored nonexistent value's lifetime
expires, it is simply removed from the storing node like any other
stored value expiration. Note that in the case of arrays and
dictionaries, this may create an implicit, unsigned "nonexistent"
value to represent a gap in the data structure. However, this value
isn't persistent nor is it replicated. It is simply synthesized by
the storing node.
6.4.2. Fetch
The Fetch request retrieves one or more data elements stored at a
given Resource-ID. A single Fetch request can retrieve multiple
different kinds.
Jennings, et al. Expires May 13, 2010 [Page 87]
Internet-Draft RELOAD Base November 2009
6.4.2.1. Request Definition
struct {
int32 first;
int32 last;
} ArrayRange;
struct {
KindId kind;
uint64 generation;
uint16 length;
select (model) {
case single_value: ; /* Empty */
case array:
ArrayRange indices<0..2^16-1>;
case dictionary:
DictionaryKey keys<0..2^16-1>;
/* This structure may be extended */
} model_specifier;
} StoredDataSpecifier;
struct {
ResourceId resource;
StoredDataSpecifier specifiers<0..2^16-1>;
} FetchReq;
The contents of the Fetch requests are as follows:
resource
The resource ID to fetch from.
specifiers
A sequence of StoredDataSpecifier values, each specifying some of
the data values to retrieve.
Each StoredDataSpecifier specifies a single kind of data to retrieve
and (if appropriate) the subset of values that are to be retrieved.
The contents of the StoredDataSpecifier structure are as follows:
Jennings, et al. Expires May 13, 2010 [Page 88]
Internet-Draft RELOAD Base November 2009
kind
The Kind-ID of the data being fetched. Implementations SHOULD
reject requests corresponding to unknown kinds unless specifically
configured otherwise.
model
The data model of the data. This must be checked against the
Kind-ID.
generation
The last generation counter that the requesting peer saw. This
may be used to avoid unnecessary fetches or it may be set to zero.
length
The length of the rest of the structure, thus allowing
extensibility.
model_specifier
A reference to the data value being requested within the data
model specified for the kind. For instance, if the data model is
"array", it might specify some subset of the values.
The model_specifier is as follows:
o If the data model is single value, the specifier is empty.
o If the data model is array, the specifier contains a list of
ArrayRange elements, each of which contains two integers. The
first integer is the beginning of the range and the second is the
end of the range. 0 is used to indicate the first element and
0xffffffff is used to indicate the final element. The first
integer must be less than the second. The ranges MUST NOT
overlap.
o If the data model is dictionary then the specifier contains a list
of the dictionary keys being requested. If no keys are specified,
than this is a wildcard fetch and all key-value pairs are
returned.
The generation-counter is used to indicate the requester's expected
state of the storing peer. If the generation-counter in the request
matches the stored counter, then the storing peer returns a response
with no StoredData values.
Note that because the certificate for a user is typically stored at
the same location as any data stored for that user, a requesting peer
which does not already have the user's certificate should request the
certificate in the Fetch as an optimization.
Jennings, et al. Expires May 13, 2010 [Page 89]
Internet-Draft RELOAD Base November 2009
6.4.2.2. Response Definition
The response to a successful Fetch request is a FetchAns message
containing the data requested by the requester.
struct {
KindId kind;
uint64 generation;
StoredData values<0..2^32-1>;
} FetchKindResponse;
struct {
FetchKindResponse kind_responses<0..2^32-1>;
} FetchAns;
The FetchAns structure contains a series of FetchKindResponse
structures. There MUST be one FetchKindResponse element for each
Kind-ID in the request.
The contents of the FetchKindResponse structure are as follows:
kind
the kind that this structure is for.
generation
the generation counter for this kind.
values
the relevant values. If the generation counter in the request
matches the generation-counter in the stored data, then no
StoredData values are returned. Otherwise, all relevant data
values MUST be returned. A nonexistent value is represented with
"exists" set to False.
There is one subtle point about signature computation on arrays. If
the storing node uses the append feature (where the
index=0xffffffff), then the index in the StoredData that is returned
will not match that used by the storing node, which would break the
signature. In order to avoid this issue, the index value in the
array is set to zero before the signature is computed. This implies
that malicious storing nodes can reorder array entries without being
detected.
6.4.3. Stat
The Stat request is used to get metadata (length, generation counter,
digest, etc.) for a stored element without retrieving the element
Jennings, et al. Expires May 13, 2010 [Page 90]
Internet-Draft RELOAD Base November 2009
itself. The name is from the UNIX stat(2) system call which performs
a similar function for files in a filesystem. It also allows the
requesting node to get a list of matching elements without requesting
the entire element.
6.4.3.1. Request Definition
The Stat request is identical to the Fetch request. It simply
specifies the elements to get metadata about.
struct {
ResourceId resource;
StoredDataSpecifier specifiers<0..2^16-1>;
} StatReq;
6.4.3.2. Response Definition
The Stat response contains the same sort of entries that a Fetch
response would contain; however, instead of containing the element
data it contains metadata.
struct {
Boolean exists;
uint32 value_length;
HashAlgorithm hash_algorithm;
opaque hash_value<0..255>;
} MetaData;
struct {
uint32 index;
MetaData value;
} ArrayEntryMeta;
struct {
DictionaryKey key;
MetaData value;
} DictionaryEntryMeta;
struct {
select (model) {
case single_value:
MetaData single_value_entry;
case array:
Jennings, et al. Expires May 13, 2010 [Page 91]
Internet-Draft RELOAD Base November 2009
ArrayEntryMeta array_entry;
case dictionary:
DictionaryEntryMeta dictionary_entry;
/* This structure may be extended */
} ;
} MetaDataValue;
struct {
uint32 value_length;
uint64 storage_time;
uint32 lifetime;
MetaDataValue metadata;
} StoredMetaData;
struct {
KindId kind;
uint64 generation;
StoredMetaData values<0..2^32-1>;
} StatKindResponse;
struct {
StatKindResponse kind_responses<0..2^32-1>;
} StatAns;
The structures used in StatAns parallel those used in FetchAns: a
response consists of multiple StatKindResponse values, one for each
kind that was in the request. The contents of the StatKindResponse
are the same as those in the FetchKindResponse, except that the
values list contains StoredMetaData entries instead of StoredData
entries.
The contents of the StoredMetaData structure are the same as the
corresponding fields in StoredData except that there is no signature
field and the value is a MetaDataValue rather than a StoredDataValue.
A MetaDataValue is a variant structure, like a StoredDataValue,
except for the types of each arm, which replace DataValue with
MetaData.
The only really new structure is MetaData, which has the following
contents:
Jennings, et al. Expires May 13, 2010 [Page 92]
Internet-Draft RELOAD Base November 2009
exists
Same as in DataValue
value_length
The length of the stored value.
hash_algorithm
The hash algorithm used to perform the digest of the value.
hash_value
A digest of the value using hash_algorithm.
6.4.4. Find
The Find request can be used to explore the Overlay Instance. A Find
request for a Resource-ID R and a Kind-ID T retrieves the Resource-ID
(if any) of the resource of kind T known to the target peer which is
closest to R. This method can be used to walk the Overlay Instance by
interactively fetching R_n+1=nearest(1 + R_n).
6.4.4.1. Request Definition
The FindReq message contains a series of Resource-IDs and Kind-IDs
identifying the resource the peer is interested in.
struct {
ResourceID resource;
KindId kinds<0..2^8-1>;
} FindReq;
The request contains a list of Kind-IDs which the Find is for, as
indicated below:
resource
The desired Resource-ID
kinds
The desired Kind-IDs. Each value MUST only appear once.
6.4.4.2. Response Definition
A response to a successful Find request is a FindAns message
containing the closest Resource-ID on the peer for each kind
specified in the request.
Jennings, et al. Expires May 13, 2010 [Page 93]
Internet-Draft RELOAD Base November 2009
struct {
KindId kind;
ResourceID closest;
} FindKindData;
struct {
FindKindData results<0..2^16-1>;
} FindAns;
If the processing peer is not responsible for the specified
Resource-ID, it SHOULD return a 404 error.
For each Kind-ID in the request the response MUST contain a
FindKindData indicating the closest Resource-ID for that Kind-ID,
unless the kind is not allowed to be used with Find in which case a
FindKindData for that Kind-ID MUST NOT be included in the response.
If a Kind-ID is not known, then the corresponding Resource-ID MUST be
0. Note that different Kind-IDs may have different closest Resource-
IDs.
The response is simply a series of FindKindData elements, one per
kind, concatenated end-to-end. The contents of each element are:
kind
The Kind-ID.
closest
The closest resource ID to the specified resource ID. This is 0
if no resource ID is known.
Note that the response does not contain the contents of the data
stored at these Resource-IDs. If the requester wants this, it must
retrieve it using Fetch.
6.4.5. Defining New Kinds
There are two ways to define a new kind. The first is by writing a
document and registering the kind-id with IANA. This is the
preferred method for kinds which may be widely used and reused. The
second method is to simply define the kind and its parameters in the
configuration document using the section of kind-id space set aside
for private use. This method MAY be used to define ad hoc kinds in
new overlays.
However a kind is defined, the definition must include:
Jennings, et al. Expires May 13, 2010 [Page 94]
Internet-Draft RELOAD Base November 2009
o The meaning of the data to be stored (in some textual form).
o The Kind-ID.
o The data model (single value, array, dictionary, etc).
o The access control model.
In addition, when kinds are registered with IANA, each kind is
assigned a short string name which is used to refer to it in
configuration documents.
While each kind needs to define what data model is used for its data,
that does not mean that it must define new data models. Where
practical, kinds should use the existing data models. The intention
is that the basic data model set be sufficient for most applications/
usages.
7. Certificate Store Usage
The Certificate Store usage allows a peer to store its certificate in
the overlay, thus avoiding the need to send a certificate in each
message - a reference may be sent instead.
A user/peer MUST store its certificate at Resource-IDs derived from
two Resource Names:
o The user name in the certificate.
o The Node-ID in the certificate.
Note that in the second case the certificate is not stored at the
peer's Node-ID but rather at a hash of the peer's Node-ID. The
intention here (as is common throughout RELOAD) is to avoid making a
peer responsible for its own data.
A peer MUST ensure that the user's certificates are stored in the
Overlay Instance. New certificates are stored at the end of the
list. This structure allows users to store an old and a new
certificate that both have the same Node-ID, which allows for
migration of certificates when they are renewed.
This usage defines the following kind:
Name: CERTIFICATE
Jennings, et al. Expires May 13, 2010 [Page 95]
Internet-Draft RELOAD Base November 2009
Data Model: The data model for CERTIFICATE data is array.
Access Control: NODE-MATCH.
8. TURN Server Usage
The TURN server usage allows a RELOAD peer to advertise that it is
prepared to be a TURN server as defined in [I-D.ietf-behave-turn].
When a node starts up, it joins the overlay network and forms several
connections in the process. If the ICE stage in any of these
connections returns a reflexive address that is not the same as the
peer's perceived address, then the peer is behind a NAT and not a
candidate for a TURN server. Additionally, if the peer's IP address
is in the private address space range, then it is also not a
candidate for a TURN server. Otherwise, the peer SHOULD assume it is
a potential TURN server and follow the procedures below.
If the node is a candidate for a TURN server it will insert some
pointers in the overlay so that other peers can find it. The overlay
configuration file specifies a turnDensity parameter that indicates
how many times each TURN server should record itself in the overlay.
Typically this should be set to the reciprocal of the estimate of
what percentage of peers will act as TURN servers. For each value,
called d, between 1 and turnDensity, the peer forms a Resource Name
by concatenating its Peer-ID and the value d. This Resource Name is
hashed to form a Resource-ID. The address of the peer is stored at
that Resource-ID using type TURN-SERVICE and the TurnServer object:
struct {
uint8 iteration;
IpAddressAndPort server_address;
} TurnServer;
The contents of this structure are as follows:
iteration
the d value
server_address
the address at which the TURN server can be contacted.
Note: Correct functioning of this algorithm depends critically on
having turnDensity be an accurate estimate of the true density of
TURN servers. If turnDensity is too high, then the process of
finding TURN servers becomes extremely expensive as multiple
candidate Resource-IDs must be probed.
Jennings, et al. Expires May 13, 2010 [Page 96]
Internet-Draft RELOAD Base November 2009
Peers that provide this service need to support the TURN extensions
to STUN for media relay of both UDP and TCP traffic as defined in
[I-D.ietf-behave-turn] and [RFC5382].
This usage defines the following kind to indicate that a peer is
willing to act as a TURN server:
Name TURN-SERVICE
Data Model The TURN-SERVICE kind stores a single value for each
Resource-ID.
Access Control NODE-MULTIPLE, with maximum iteration counter 20.
Peers can find other servers by selecting a random Resource-ID and
then doing a Find request for the appropriate server type with that
Resource-ID. The Find request gets routed to a random peer based on
the Resource-ID. If that peer knows of any servers, they will be
returned. The returned response may be empty if the peer does not
know of any servers, in which case the process gets repeated with
some other random Resource-ID. As long as the ratio of servers
relative to peers is not too low, this approach will result in
finding a server relatively quickly.
9. Chord Algorithm
This algorithm is assigned the name chord-reload to indicate it is an
adaptation of the basic Chord DHT algorithm.
This algorithm differs from the originally presented Chord algorithm
[Chord]. It has been updated based on more recent research results
and implementation experiences, and to adapt it to the RELOAD
protocol. A short list of differences:
o The original Chord algorithm specified that a single predecessor
and a successor list be stored. The chord-reload algorithm
attempts to have more than one predecessor and successor. The
predecessor sets help other neighbors learn their successor list.
o The original Chord specification and analysis called for iterative
routing. RELOAD specifies recursive routing. In addition to the
performance implications, the cost of NAT traversal dictates
recursive routing.
o Finger table entries are indexed in opposite order. Original
Chord specifies finger[0] as the immediate successor of the peer.
chord-reload specifies finger[0] as the peer 180 degrees around
the ring from the peer. This change was made to simplify
discussion and implementation of variable sized finger tables.
However, with either approach no more than O(log N) entries should
typically be stored in a finger table.
Jennings, et al. Expires May 13, 2010 [Page 97]
Internet-Draft RELOAD Base November 2009
o The stabilize() and fix_fingers() algorithms in the original Chord
algorithm are merged into a single periodic process.
Stabilization is implemented slightly differently because of the
larger neighborhood, and fix_fingers is not as aggressive to
reduce load, nor does it search for optimal matches of the finger
table entries.
o RELOAD uses a 128 bit hash instead of a 160 bit hash, as RELOAD is
not designed to be used in networks with close to or more than
2^128 nodes.
o RELOAD uses randomized finger entries as described in
Section 9.6.4.2.
9.1. Overview
The algorithm described here is a modified version of the Chord
algorithm. Each peer keeps track of a finger table and a neighbor
table. The neighbor table typically contains the three peers before
this peer and the three peers after it in the DHT ring. There may
not be three entreis in all cases such as small rings or while the
ring topology is changing. The first entry in the finger table
contains the peer half-way around the ring from this peer; the second
entry contains the peer that is 1/4 of the way around; the third
entry contains the peer that is 1/8th of the way around, and so on.
Fundamentally, the chord data structure can be thought of a doubly-
linked list formed by knowing the successors and predecessor peers in
the neighbor table, sorted by the Node-ID. As long as the successor
peers are correct, the DHT will return the correct result. The
pointers to the prior peers are kept to enable the insertion of new
peers into the list structure. Keeping multiple predecessor and
successor pointers makes it possible to maintain the integrity of the
data structure even when consecutive peers simultaneously fail. The
finger table forms a skip list, so that entries in the linked list
can be found in O(log(N)) time instead of the typical O(N) time that
a linked list would provide.
A peer, n, is responsible for a particular Resource-ID k if k is less
than or equal to n and k is greater than p, where p is the peer id of
the previous peer in the neighbor table. Care must be taken when
computing to note that all math is modulo 2^128.
9.2. Routing
The routing table is the union of the neighbor table and the finger
table.
If a peer is not responsible for a Resource-ID k, but is directly
connected to a node with Node-ID k, then it routes the message to
that node. Otherwise, it routes the request to the peer in the
Jennings, et al. Expires May 13, 2010 [Page 98]
Internet-Draft RELOAD Base November 2009
routing table that has the largest Node-ID that is in the interval
between the peer and k. If no such node is found, it finds the
smallest node id that is greater than k and routes the message to
that node.
9.3. Redundancy
When a peer receives a Store request for Resource-ID k, and it is
responsible for Resource-ID k, it stores the data and returns a
success response. It then sends a Store request to its successor in
the neighbor table and to that peer's successor. Note that these
Store requests are addressed to those specific peers, even though the
Resource-ID they are being asked to store is outside the range that
they are responsible for. The peers receiving these check they came
from an appropriate predecessor in their neighbor table and that they
are in a range that this predecessor is responsible for, and then
they store the data. They do not themselves perform further Stores
because they can determine that they are not responsible for the
Resource-ID.
The sequential replicas used in this overlay algorithm protect
against peer failure but not against malicious peers. Additional
replication from the Usage is required to protect resources from such
attacks, as discussed in Section 12.5.4.
9.4. Joining
The join process for a joining party (JP) with Node-ID n is as
follows.
1. JP MUST connect to its chosen bootstrap node.
2. JP SHOULD use a series of Pings to populate its routing table.
3. JP SHOULD send Attach requests to initiate connections to each of
the peers in the neighbor table as well as to the desired 16
finger table entries. Note that this does not populate their
routing tables, but only their connection tables, so JP will not
get messages that it is expected to route to other nodes.
4. JP MUST enter all the peers it has contacted into its routing
table.
5. JP SHOULD send a Join to its immediate successor, the admitting
peer (AP) for Node-ID n. The AP sends the response to the Join.
6. AP MUST do a series of Store requests to JP to store the data
that JP will be responsible for.
7. AP MUST send JP an Update explicitly labeling JP as its
predecessor. At this point, JP is part of the ring and
responsible for a section of the overlay. AP can now forget any
data which is assigned to JP and not AP.
Jennings, et al. Expires May 13, 2010 [Page 99]
Internet-Draft RELOAD Base November 2009
8. AP MUST send an Update to all of its neighbors with the new
values of its neighbor set (including JP).
9. JP SHOULD send Updates to all the peers in its routing table.
In order to populate its neighbor table, JP sends a Ping via the
bootstrap node directed at Resource-ID n+1 (directly after its own
Resource-ID). This allows it to discover its own successor. Call
that node p0. It then sends a ping to p0+1 to discover its successor
(p1). This process can be repeated to discover as many successors as
desired. The values for the two peers before p will be found at a
later stage when n receives an Update.
In order to set up its neighbor table entry for peer i, JP simply
sends an Attach to peer (n+2^(numBitsInNodeId-i). This will be
routed to a peer in approximately the right location around the ring.
The joining peer MUST NOT send any Update message placing itself in
the overlay until it has successfully completed an Attach with each
peer that should be in its neighbor table.
9.5. Routing Attaches
When a peer needs to Attach to a new peer in its neighbor table, it
MUST source-route the Attach request through the peer from which it
learned the new peer's Node-ID. Source-routing these requests allows
the overlay to recover from instability.
All other Attach requests, such as those for new finger table
entries, are routed conventionally through the overlay.
If a peer is unable to successfully Attach with a peer that should be
in its neighborhood, it MUST locate either a TURN server or another
peer in the overlay, but not in its neighborhood, through which it
can exchange messages with its neighbor peer.
9.6. Updates
A chord Update is defined as
Jennings, et al. Expires May 13, 2010 [Page 100]
Internet-Draft RELOAD Base November 2009
enum { reserved (0),
peer_ready(1), neighbors(2), full(3), (255) }
ChordUpdateType;
struct {
uint32 uptime;
ChordUpdateType type;
select(type){
case peer_ready: /* Empty */
;
case neighbors:
NodeId predecessors<0..2^16-1>;
NodeId successors<0..2^16-1>;
case full:
NodeId predecessors<0..2^16-1>;
NodeId successors<0..2^16-1>;
NodeId fingers<0..2^16-1>;
};
} ChordUpdate;
The "type" field contains the type of the update, which depends on
the reason the update was sent.
uptime: time this peer has been up in seconds.
peer_ready: this peer is ready to receive messages. This message
is used to indicate that a node which has Attached is a peer and
can be routed through. It is also used as a connectivity check to
non-neighbor peers.
neighbors: this version is sent to members of the Chord neighbor
table.
full: this version is sent to peers which request an Update with a
RouteQueryReq.
If the message is of type "neighbors", then the contents of the
message will be:
Jennings, et al. Expires May 13, 2010 [Page 101]
Internet-Draft RELOAD Base November 2009
predecessors
The predecessor set of the Updating peer.
successors
The successor set of the Updating peer.
If the message is of type "full", then the contents of the message
will be:
predecessors
The predecessor set of the Updating peer.
successors
The successor set of the Updating peer.
fingers
The finger table if the Updating peer, in numerically ascending
order.
A peer MUST maintain an association (via Attach) to every member of
its neighbor set. A peer MUST attempt to maintain at least three
predecessors and three successors, even though this will not be
possible if the ring is very small. However, it MUST send its entire
set in any Update message sent to neighbors.
9.6.1. Handling Neighbor Failures
Every time a connection to a peer in the neighbor table is lost (as
determined by connectivity pings or the failure of some request), the
peer MUST remove the entry from its neighbor table and replace it
with the best match it has from the other peers in its routing table.
If using reactive recovery, it then sends an immediate Update to all
nodes in its Connection Table. The update will contain all the Node-
IDs of the current entries of the table (after the failed one has
been removed). Note that when replacing a successor the peer SHOULD
delay the creation of new replicas for successor replacement hold-
down time (30 seconds) after removing the failed entry from its
neighbor table in order to allow a triggered update to inform it of a
better match for its neighbor table.
A peer MAY attempt to reestablish connectivity with a lost neighbor
either by waiting additional time to see if connectivity returns or
by actively routing a new ATTACH to the lost peer. Details for these
procedures are beyond the scope of this document. In no event does
an attempt to reestablish connectivity with a lost neighbor allow the
peer to remain in the neighbor table. Such a peer is returned to the
neighbor table once connectivity is reestablished.
Jennings, et al. Expires May 13, 2010 [Page 102]
Internet-Draft RELOAD Base November 2009
If connectivity is lost to all three of the peers that follow this
peer in the ring, then this peer should behave as if it is joining
the network and use Pings to find a peer and send it a Join. If
connectivity is lost to all the peers in the finger table, this peer
should assume that it has been disconnected from the rest of the
network, and it should periodically try to join the DHT.
9.6.2. Handling Finger Table Entry Failure
If a finger table entry is found to have failed, all references to
the failed peer are removed from the finger table and replaced with
the closest preceding peer from the finger table or neighbor table.
If using reactive recovery, the peer initiates a search for a new
finger table entry as described below.
9.6.3. Receiving Updates
When a peer, N, receives an Update request, it examines the Node-IDs
in the UpdateReq and at its neighbor table and decides if this
UpdateReq would change its neighbor table. This is done by taking
the set of peers currently in the neighbor table and comparing them
to the peers in the update request. There are three major cases:
o The UpdateReq contains peers that would not change the neighbor
set because they match the neighbor table.
o The UpdateReq contains peers closer to N than those in its
neighbor table.
o The UpdateReq defines peers that indicate a neighbor table further
away from N than some of its neighbor table. Note that merely
receiving peers further away does not demonstrate this, since the
update could be from a node far away from N. Rather, the peers
would need to bracket N.
In the first case, no change is needed.
In the second case, N MUST attempt to Attach to the new peers and if
it is successful it MUST adjust its neighbor set accordingly. Note
that it can maintain the now inferior peers as neighbors, but it MUST
remember the closer ones.
The third case implies that a neighbor has disappeared, most likely
because it has simply been disconnected but perhaps because of
overlay instability. N MUST Ping the questionable peers to discover
if they are indeed missing and if so, remove them from its neighbor
table.
After any Pings and Attaches are done, if the neighbor table changes
Jennings, et al. Expires May 13, 2010 [Page 103]
Internet-Draft RELOAD Base November 2009
and the peer is using reactive recovery, the peer sends an Update
request to each member of its Connection Table. These Update
requests are what ends up filling in the predecessor/successor tables
of peers that this peer is a neighbor to. A peer MUST NOT enter
itself in its successor or predecessor table and instead should leave
the entries empty.
If peer N which is responsible for a Resource-ID R discovers that the
replica set for R (the next two nodes in its successor set) has
changed, it MUST send a Store for any data associated with R to any
new node in the replica set. It SHOULD NOT delete data from peers
which have left the replica set.
When a peer N detects that it is no longer in the replica set for a
resource R (i.e., there are three predecessors between N and R), it
SHOULD delete all data associated with R from its local store.
9.6.4. Stabilization
There are four components to stabilization:
1. exchange Updates with all peers in its neighbor table to exchange
state.
2. search for better peers to place in its finger table.
3. search to determine if the current finger table size is
sufficiently large.
4. search to determine if the overlay has partitioned and needs to
recover.
9.6.4.1. Updating neighbor table
A peer MUST periodically send an Update request to every peer in its
Connection Table. The purpose of this is to keep the predecessor and
successor lists up to date and to detect failed peers. The default
time is about every ten minutes, but the enrollment server SHOULD set
this in the configuration document using the "chord-reload-update-
interval" element (denominated in seconds.) A peer SHOULD randomly
offset these Update requests so they do not occur all at once.
9.6.4.2. Refreshing finger table
A peer MUST periodically search for new peers to replace invalid
(repeated) entries in the finger table. A finger table entry i is
valid if it is in the range [n+2^(128-i),
n+2^(128-(i-1))-2^(128-(i+1))]. Invalid entries occur in the finger
table when a previous finger table entry has failed or when no peer
has been found in that range.
A peer SHOULD NOT send Ping requests looking for new finger table
Jennings, et al. Expires May 13, 2010 [Page 104]
Internet-Draft RELOAD Base November 2009
entries more often than the configuration element "chord-reload-ping-
interval", which defaults to 3600 seconds (one per hour).
Two possible methods for searching for new peers for the finger table
entries are presented:
Alternative 1: A peer selects one entry in the finger table from
among the invalid entries. It pings for a new peer for that finger
table entry. The selection SHOULD be exponentially weighted to
attempt to replace earlier (lower i) entries in the finger table. A
simple way to implement this selection is to search through the
finger table entries from i=0 and each time an invalid entry is
encountered, send a Ping to replace that entry with probability 0.5.
Alternative 2: Every "chord-reload-ping-interval" seconds, the peer
scans through its finger table and for each invalid finger table
entry i, sends a RouteQuery request for the ID n+2^(128-i) to the
closest preceding peer to that ID in the routing table. The
responses to these route queries are used to identify the set of
entries for which a new Ping is likely to result in a valid entry:
the responses that contain a peer not currently in the finger table
indicate a Ping may result in a new valid entry for the finger table.
The peer then selects from among those candidates using an
exponentially weighted probability as above.
When searching for a better entry, the peer SHOULD send the Ping to a
Node-ID selected randomly from that range. Random selection is
preferred over a search for strictly spaced entries to minimize the
effect of churn on overlay routing [minimizing-churn-sigcomm06]. An
implementation or subsequent specification MAY choose a method for
selecting finger table entries other than choosing randomly within
the range. Any such alternate methods SHOULD be employed only on
finger table stabilization and not for the selection of initial
finger table entries unless the alternative method is faster and
imposes less overhead on the overlay.
A peer MAY choose to keep connections to multiple peers that can act
for a given finger table entry.
9.6.4.3. Adjusting finger table size
If the finger table has less than 16 entries, the node SHOULD attempt
to discover more fingers to grow the size of the table to 16. The
value 16 was chosen to ensure high odds of a node maintaining
connectivity to the overlay even with strange network partitions.
For many overlays, 16 finger table entries will be enough, but as an
overlay grows very large, more than 16 entries may be required in the
Jennings, et al. Expires May 13, 2010 [Page 105]
Internet-Draft RELOAD Base November 2009
finger table for efficient routing. An implementation SHOULD be
capable of increasing the number of entries in the finger table to
128 entries.
Note to implementers: Although log(N) entries are all that are
required for optimal performance, careful implementation of
stabilization will result in no additional traffic being generated
when maintaining a finger table larger than log(N) entries.
Implementers are encouraged to make use of RouteQuery and algorithms
for determining where new finger table entries may be found.
Complete details of possible implementations are outside the scope of
this specification.
A simple approach to sizing the finger table is to ensure the finger
table is large enough to contain at least the final successor in the
peer's neighbor table.
9.6.4.4. Detecting partitioning
To detect that a partitioning has occurred and to heal the overlay, a
peer P MUST periodically repeat the discovery process used in the
initial join for the overlay to locate an appropriate bootstrap node,
B. P should then send a Ping for its own Node-ID routed through B. If
a response is received from a peer S', which is not P's successor,
then the overlay is partitioned and P should send an Attach to S'
routed through B, followed by an Update sent to S'. (Note that S'
may not be in P's neighbor table once the overlay is healed, but the
connection will allow S' to discover appropriate neighbor entries for
itself via its own stabilization.)
Future specifications may describe alternative mechanisms for
determining when to repeat the discovery process.
9.7. Route Query
For this topology plugin, the RouteQueryReq contains no additional
information. The RouteQueryAns contains the single node ID of the
next peer to which the responding peer would have routed the request
message in recursive routing:
struct {
NodeId next_id;
} ChordRouteQueryAns;
The contents of this structure are as follows:
Jennings, et al. Expires May 13, 2010 [Page 106]
Internet-Draft RELOAD Base November 2009
next_peer
The peer to which the responding peer would route the message in
order to deliver it to the destination listed in the request.
If the requester has set the send_update flag, the responder SHOULD
initiate an Update immediately after sending the RouteQueryAns.
9.8. Leaving
To support extentions, such as [I-D.maenpaa-p2psip-self-tuning],
Peers SHOULD send a Leave request to all members of their neighbor
table prior to exiting the Overlay Instance. The
overlay_specific_data field MUST contain the ChordLeaveData structure
defined below:
enum { reserved (0),
from_succ(1), from_pred(2), (255) }
ChordLeaveType;
struct {
ChordLeaveType type;
select(type) {
case from_succ:
NodeId successors<0..2^16-1>;
case from_pred:
NodeId predecessors<0..2^16-1>;
};
} ChordLeaveData;
Jennings, et al. Expires May 13, 2010 [Page 107]
Internet-Draft RELOAD Base November 2009
The 'type' field indicates whether the Leave request was sent by a
predecessor or a successor of the recipient:
from_succ
The Leave request was sent by a successor.
from_pred
The Leave request was sent by a predecessor.
If the type of the request is 'from_succ', the contents will be:
successors
The sender's successor list.
If the type of the request is 'from_pred', the contents will be:
predecessors
The sender's predecessor list.
Any peer which receives a Leave for a peer n in its neighbor set
follows procedures as if it had detected a peer failure as described
in Section 9.6.1.
10. Enrollment and Bootstrap
10.1. Overlay Configuration
This specification defines a new content type "application/
p2p-overlay+xml" for an MIME entity that contains overlay
information. An example document is shown below.
false
192.0.0.1:5678
192.0.2.2:6789
30
false
10
4000
Jennings, et al. Expires May 13, 2010 [Page 108]
Internet-Draft RELOAD Base November 2009
https://example.org
foo
300
400
false
asecret
chord
DATA GOES HERE
single
user-match
1
100
VGhpcyBpcyBub3QgcmlnaHQhCg==
array
node-multiple
3
22
4
1
VGhpcyBpcyBub3QgcmlnaHQhCg==
47112162e84c69ba
6eba45d31a900c06
VGhpcyBpcyBub3QgcmlnaHQhCg==
The file MUST be a well formed XML document and it SHOULD contain an
encoding declaration in the XML declaration. If the charset
parameter of the MIME content type declaration is present and it is
different from the encoding declaration, the charset parameter takes
precedence. Every application conforming to this specification MUST
accept the UTF-8 character encoding to ensure minimal
Jennings, et al. Expires May 13, 2010 [Page 109]
Internet-Draft RELOAD Base November 2009
interoperability. The namespace for the elements defined in this
specification is urn:ietf:params:xml:ns:p2p:config-base and
urn:ietf:params:xml:ns:p2p:config-chord".
The file can contain multiple "configuration" elements where each one
contains the configuration information for a different overlay. Each
"configuration" has the following attributes:
instance-name: name of the overlay
expiration: time in future at which this overlay configuration is no
longer valid and needs to be retrieved again
sequence: a monotonically increasing sequence number between 1 and
2^32
Inside each overlay element, the following elements can occur:
topology-plugin This element has an attribute called algorithm-name
that describes the overlay algorithm being used.
root-cert This element contains a PEM encoded X.509v3 certificate
that is a root trust anchor used to sign all certificates in this
overlay. There can be more than one root-cert element.
credential-server This element contains the URL at which the
credential server can be reached in a "url" element. This URL
MUST be of type "https:". More than one credential-server element
may be present.
self-signed-permitted This element indicates whether self-signed
certificates are permitted. If it is set to "true", then self-
signed certificates are allowed, in which case the credential-
server and root-cert elements may be absent. Otherwise, it SHOULD
be absent, but MAY be set to "false". This element also contains
an attribute "digest" which indicates the digest to be used to
compute the Node-ID. Valid values for this parameter are "SHA-1"
and "SHA-256". Implementations MUST support both of these
algorithms.
bootstrap-node This element represents the address of one of the
bootstrap nodes. It has an attribute called "address" that
represents the IP address (either IPv4 or IPv6, since they can be
distinguished) and an attribute called "port" that represents the
port. More than one bootstrap-peer element may be present.
turn-density This element is a positive integer that represents the
approximate reciprocal of density of nodes that can act as TURN
servers. For example, if 10% of the nodes can act as TURN
servers, this would be set to 10. If it is not present, the
default value is 1.
Jennings, et al. Expires May 13, 2010 [Page 110]
Internet-Draft RELOAD Base November 2009
multicast-bootstrap This element represents the address of a
multicast, boradcast, or anycast address and port that may be used
for bootstrap. Nodes SHOULD listen on the address. It has an
attributed called "address" that represents the IP address and an
attribute called "port" that represents the port. More than one
"multicast-bootstrap" element may be present.
clients-permitted This element represents whether clients are
permitted or whether all nodes must be peers. If it is set to
"TRUE" or absent, this indicates that clients are permitted. If
it is set to "FALSE" then nodes MUST join as peers.
attach-lite-permitted This element represents whether nodes are
allowed to use the AttachLite and AppAttachLite request in this
overlay. If it is absent, it is treated as if it were set to
"FALSE".
chord-update-interval The update frequency for the Chord-reload
topology plugin (see Section 9).
chord-ping-interval The ping frequency for the Chord-reload
topology plugin (see Section 9).
shared-secret If shared secret mode is used, this contains the
shared secret.
max-message-size Maximum size in bytes of any message in the
overlay. If this value is not present, the default is 5000.
initial-ttl Initial default TTL (time to live, see Section 5.3.2)
for messages. If this value is not present, the default is 100.
kind-signer This contains a single Node-Id in hexadecimal and
indicates that the certificate with this Node-ID is allowed to
sign kinds. Identifying kind-signer by Node-Id instead of
certificate allows the use of short lived certificates without
constantly having to provide an updated configuration file.
Inside each overlay element, the required-kinds elements can also
occur. This element indicates the kinds that members must support
and contains multiple kind-block elements that each define a single
kind that MUST be supported by nodes in the overlay. Each kind-block
consists of a single kind element and a kind-signature. The kind
element defines the kind. The kind-signature is the signature
computed over the kind element.
Each kind has either an ID attribute or a name atribute. The name
attribute is a string representing the kind (the name registered to
IANA) while the ID is an integer kind-id allocated out of private
space.
In addition, the kind element contains the following elements:
Jennings, et al. Expires May 13, 2010 [Page 111]
Internet-Draft RELOAD Base November 2009
max-count: the maximum number of values which members of the overlay
must support.
data-model: the data model to be used.
max-size: the maximum size of individual values.
access-control: the access control model to be used.
max-node-multiple: This is optional and only used when the access
control is NODE-MULTIPLE. This indicates the maximum value for
the i counter. This is an integer greater than 0.
All of the non optional values MUST be provided. If the kind is
registered with IANA, the data-model and access-control attributes
MUST match those in the kind registration. For instance, the example
above indicates that members must support SIP-REGISTRATION with a
maximum of 10 values of up to 1000 bytes each. Multiple required-
kinds elements MAY be present.
The kind-block element also MUST contain a "kind-signature" element.
This signature is computed across the kind from the beginning of the
first < of the kind to the end of the last > of the kind in the same
way as the "signature element described later in this section.
The configuration file is a binary file and cannot be changed -
including whitespace changes - or the signature will break. The
signature is computed by taking each configuration element and
starting form, and including, the first < at the start of
up to and including the > in and
treating this as a binary blob that is signed using the standard
SecurityBlock defined in Section 5.3.4. The SecurityBlock is base 64
encoded using the base64 alphabet from RFC[RFC4648] and put in the
signature element following the configuration object in the config
file.
When a node receives a new configuration file, it MUST change its
configuration to meet the new requirements. This may require the
node to exit the DHT and re-join. If a node is not capable of
supporting the new requirements, it MUST exit the overlay. If some
information about a particular kind changes from what the node
previously knew about the kind (for example the max size), the new
information in the configuration files overrides any previously
learned information. If any kind data was signed by a node that is
no longer allowed to sign kinds, that kind MUST be discarded along
with any stored information of that kind.
10.1.1. Relax NG Grammar
The grammar for the configuration data is:
namespace chord = "urn:ietf:params:xml:ns:p2p:config-chord"
Jennings, et al. Expires May 13, 2010 [Page 112]
Internet-Draft RELOAD Base November 2009
namespace local = ""
default namespace p2pcf="urn:ietf:params:xml:ns:p2p:config-base"
namespace rng = "http://relaxng.org/ns/structure/1.0"
anything =
(element * { anything }
| attribute * { text }
| text)*
foreign-elements = element * - (p2pcf:*|local:*|chord:*) {anything}*
start =
element p2pcf:overlay {
element configuration {
attribute instance-name { text },
attribute expiration { xsd:dateTime },
attribute sequence { xsd:long },
parameter
},
element signature {
attribute algorithm { signature-algorithm-type }?,
xsd:base64Binary
}?
}
signature-algorithm-type |= "rsa-sha1"
parameter &= element topology-plugin { topology-plugin-type }
parameter &= element max-message-size { xsd:int }?
parameter &= element initial-ttl { xsd:int }?
parameter &= element root-cert { text }?
parameter &= element required-kinds { kind-block* }
parameter &= element credential-server { xsd:anyURI }?
parameter &= element kind-signer { text }*
parameter &= element attach-lite-permitted { xsd:boolean }?
parameter &= element shared-secret { xsd:string }?
parameter &= element clients-permitted { xsd:boolean }?
parameter &= element turn-density { xsd:int }?
parameter &= foreign-elements*
parameter &=
element self-signed-permitted {
attribute digest { self-signed-digest-type },
xsd:boolean
}?
self-signed-digest-type |= "sha1"
parameter &=
element bootstrap-node { hostPort
}+
hostPort = text
parameter &=
Jennings, et al. Expires May 13, 2010 [Page 113]
Internet-Draft RELOAD Base November 2009
element multicast-bootstrap { hostPort
}*
kind-block = element kind-block {
element kind {
(attribute name { kind-names }
| attribute id { xsd:int }),
kind-paramter
} &
element kind-signature {
attribute algorithm { signature-algorithm-type }?,
xsd:base64Binary
}?
}
kind-paramter &= element max-count { xsd:int }
kind-paramter &= element max-size { xsd:int }
kind-paramter &= element data-model { data-model-type }
data-model-type |= "single"
data-model-type |= "array"
data-model-type |= "dictionary"
kind-paramter &= element access-control { access-control-type }
kind-paramter &= element max-node-multiple { xsd:int }?
access-control-type |= "user-match"
access-control-type |= "node-match"
access-control-type |= "user-node-match"
access-control-type |= "node-multiple"
access-control-type |= "user-match-with-anon-create"
kind-paramter &= foreign-elements*
# Chord specific paramters
topology-plugin-type |= "chord"
kind-names |= "sip-registration"
kind-names |= "turn-service"
parameter &= element chord:chord-ping-interval { xsd:int }?
parameter &= element chord:chord-update-interval { xsd:int }?
10.2. Discovery Through Enrollment Server
When a node first enrolls in a new overlay, it starts with a
discovery process to find an enrollment server. Related work to the
approach used here is described in
[I-D.garcia-p2psip-dns-sd-bootstrapping] and
[I-D.matthews-p2psip-bootstrap-mechanisms]. Another scheme for
referencing overlays is described in
Jennings, et al. Expires May 13, 2010 [Page 114]
Internet-Draft RELOAD Base November 2009
[I-D.hardie-p2poverlay-pointers].
The node first determines the overlay name. This value is provided
by the user or some other out-of-band provisioning mechanism. The
out-of-band mechanisms may also provide an optional URL for the
enrollment server. If a URL for the enrollment server is not
provided, the node MUST do a DNS SRV query using a Service name of
"p2psip_enroll" and a protocol of tcp to find an enrollment server
and form the URL by appending a path of "/p2psip/ enroll" to the
overlay name. For example, if the overlay name was example.com, the
URL would be "https://example.com/p2psip/enroll".
Once an address and URL for the enrollment server is determined, the
peer forms an HTTPS connection to that IP address. The certificate
MUST match the overlay name as described in [RFC2818]. Then the node
MUST fetch a new copy of the configuration file. To do this, the
peer performs a GET to the URL. The result of the HTTP GET is an XML
configuration file described above, which replaces any previously
learned configuration file for this overlay.
For overlays that do not use an enrollment server, nodes obtain the
configuration information needed to join the overlay through some out
of band approach such an an XML configuration file sent over email.
10.3. Credentials
If the configuration document contains a credential-server element,
credentials are required to join the Overlay Instance. A peer which
does not yet have credentials MUST contact the credential server to
acquire them.
RELOAD defines its own trivial certificate request protocol. We
would have liked to have used an existing protocol but were concerned
about the implementation burden of even the simplest of those
protocols, such as [RFC5272]) and [RFC5273]. Our objective was to
have a protocol which could be easily implemented in a Web server
which the operator did not control (e.g., in a hosted service) and
was compatible with the existing certificate handling tooling as used
with the Web certificate infrastructure. This means accepting bare
PKCS#10 requests and returning a single bare X.509 certificate.
Although the MIME types for these objects are defined, none of the
existing protocols support exactly this model.
The certificate request protocol is performed over HTTPS. The
request is an HTTP POST with the following properties:
Jennings, et al. Expires May 13, 2010 [Page 115]
Internet-Draft RELOAD Base November 2009
o If authentication is required, there is a URL parameter of
"password" and "username" containing the user's name and password
in the clear (hence the need for HTTPS)
o The body is of content type "application/pkcs10", as defined in
[RFC2311].
o The Accept header contains the type "application/pkix-cert",
indicating the type that is expected in the response.
The credential server MUST authenticate the request using the
provided user name and password. If the authentication succeeds and
the requested user name is acceptable, the server generates and
returns a certificate. The SubjectAltName field in the certificate
contains the following values:
o One or more Node-IDs which MUST be cryptographically random
[RFC4086]. Each MUST be chosen by the credential server in such a
way that they are unpredictable to the requesting user. Each is
placed in the subjectAltName using the uniformResourceIdentifier
type and MUST contain RELOAD URIs as described in Section 13.12
and MUST contain a Destination list with a single entry of type
"node_id".
o A single name this user is allowed to use in the overlay, using
type rfc822Name.
The certificate is returned as type "application/pkix-cert", with an
HTTP status code of 200 OK. Certificate processing errors should be
treated as HTTP errors and have appropriate HTTP stats codes.
The client MUST check that the certificate returned was signed by one
of the certificates received in the "root-cert" list of the overlay
configuration data. The node then reads the certificate to find the
Node-IDs it can use.
10.3.1. Self-Generated Credentials
If the "self-signed-permitted" element is present and set to "TRUE",
then a node MUST generate its own self-signed certificate to join the
overlay. The self-signed certificate MAY contain any user name of
the users choice.
The Node-ID MUST be computed by applying the digest specified in the
self-signed-permitted element to the DER representation of the user's
public key (more specifically the subjectPublicKeyInfo) and taking
the high order bits. When accepting a self-signed certificate, nodes
MUST check that the Node-ID and public keys match. This prevents
Node-ID theft.
Once the node has constructed a self-signed certificate, it MAY join
Jennings, et al. Expires May 13, 2010 [Page 116]
Internet-Draft RELOAD Base November 2009
the overlay. Before storing its certificate in the overlay
(Section 7) it SHOULD look to see if the user name is already taken
and if so choose another user name. Note that this only provides
protection against accidental name collisions. Name theft is still
possible. If protection against name theft is desired, then the
enrollment service must be used.
10.4. Searching for a Bootstrap Node
If no cached bootstrap nodes are available and the config file has an
multicast-bootstrap element, then the node SHOULD send a Ping request
over UDP to the address and port found to each multicast-bootstrap
element found in the configuration document. This MAY be a
multicast, broadcast, or anycast address. The Ping should use the
wildcard Node-ID as the destination Node-ID.
The responder node that receives the Ping request SHOULD check that
the overlay name is correct and that the requester peer sending the
request has appropriate credentials for the overlay before responding
to the Ping request even if the response is only an error.
10.5. Contacting a Bootstrap Node
In order to join the overlay, the joining node MUST contact a node in
the overlay. Typically this means contacting the bootstrap nodes,
since they are reachable by the local peer or have public IP
addresses. If the joining node has cached a list of peers it has
previously been connected with in this overlay, as an optimization it
MAY attempt to use one or more of them as bootstrap nodes before
falling back to the bootstrap nodes listed in the configuration file.
When contacting a bootstrap node, the joining node first forms the
DTLS or TLS connection to the boostrap node and then sends an Attach
request over this connection with the destination Node-ID set to the
joining node's Node-ID.
When the requester node finally does receive a response from some
responding node, it can note the Node-ID in the response and use this
Node-ID to start sending requests to join the Overlay Instance as
described in Section 5.4.
After a node has successfully joined the overlay network, it will
have direct connections to several peers. Some MAY be added to the
cached bootstrap nodes list and used in future boots. Peers that are
not directly connected MUST NOT be cached. The suggested number of
peers to cache is 10. Algorithms for determining which peers to
cache are beyond the scope of this specification.
Jennings, et al. Expires May 13, 2010 [Page 117]
Internet-Draft RELOAD Base November 2009
11. Message Flow Example
In the following example, we assume that JP has formed a connection
to one of the bootstrap nodes. JP then sends an Attach through that
peer to the admitting peer (AP) to initiate a connection. When AP
responds, JP and AP use ICE to set up a connection and then set up
TLS.
JP PPP PP AP NP NNP BP
| | | | | | |
| | | | | | |
| | | | | | |
|Attach Dest=JP | | | | |
|---------------------------------------------------------->|
| | | | | | |
| | | | | | |
| | |Attach Dest=JP | | |
| | |<--------------------------------------|
| | | | | | |
| | | | | | |
| | |Attach Dest=JP | | |
| | |-------->| | | |
| | | | | | |
| | | | | | |
| | |AttachAns | | |
| | |<--------| | | |
| | | | | | |
| | | | | | |
| | |AttachAns | | |
| | |-------------------------------------->|
| | | | | | |
| | | | | | |
|AttachAns | | | | |
|<----------------------------------------------------------|
| | | | | | |
| | | | | | |
|TLS | | | | | |
|.............................| | | |
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | |
Once JP has connected to AP, it needs to populate its Routing Table.
In Chord, this means that it needs to populate its neighbor table and
its finger table. To populate its neighbor table, it needs the
successor of AP, NP. It sends an Attach to the Resource-IP AP+1,
which gets routed to NP. When NP responds, JP and NP use ICE and TLS
Jennings, et al. Expires May 13, 2010 [Page 118]
Internet-Draft RELOAD Base November 2009
to set up a connection.
JP PPP PP AP NP NNP BP
| | | | | | |
| | | | | | |
| | | | | | |
|Attach AP+1 | | | | |
|---------------------------->| | | |
| | | | | | |
| | | | | | |
| | | |Attach AP+1 | |
| | | |-------->| | |
| | | | | | |
| | | | | | |
| | | |AttachAns | |
| | | |<--------| | |
| | | | | | |
| | | | | | |
|AttachAns | | | | |
|<----------------------------| | | |
| | | | | | |
| | | | | | |
|Attach | | | | | |
|-------------------------------------->| | |
| | | | | | |
| | | | | | |
|TLS | | | | | |
|.......................................| | |
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | |
JP also needs to populate its finger table (for Chord). It issues an
Attach to a variety of locations around the overlay. The diagram
below shows it sending an Attach halfway around the Chord ring to the
JP + 2^127.
Jennings, et al. Expires May 13, 2010 [Page 119]
Internet-Draft RELOAD Base November 2009
JP NP XX TP
| | | |
| | | |
| | | |
|Attach JP+2<<126 | |
|-------->| | |
| | | |
| | | |
| |Attach JP+2<<126 |
| |-------->| |
| | | |
| | | |
| | |Attach JP+2<<126
| | |-------->|
| | | |
| | | |
| | |AttachAns|
| | |<--------|
| | | |
| | | |
| |AttachAns| |
| |<--------| |
| | | |
| | | |
|AttachAns| | |
|<--------| | |
| | | |
| | | |
|TLS | | |
|.............................|
| | | |
| | | |
| | | |
| | | |
Once JP has a reasonable set of connections it is ready to take its
place in the DHT. It does this by sending a Join to AP. AP does a
series of Store requests to JP to store the data that JP will be
responsible for. AP then sends JP an Update explicitly labeling JP
as its predecessor. At this point, JP is part of the ring and
responsible for a section of the overlay. AP can now forget any data
which is assigned to JP and not AP.
Jennings, et al. Expires May 13, 2010 [Page 120]
Internet-Draft RELOAD Base November 2009
JP PPP PP AP NP NNP BP
| | | | | | |
| | | | | | |
| | | | | | |
|JoinReq | | | | | |
|---------------------------->| | | |
| | | | | | |
| | | | | | |
|JoinAns | | | | | |
|<----------------------------| | | |
| | | | | | |
| | | | | | |
|StoreReq Data A | | | | |
|<----------------------------| | | |
| | | | | | |
| | | | | | |
|StoreAns | | | | | |
|---------------------------->| | | |
| | | | | | |
| | | | | | |
|StoreReq Data B | | | | |
|<----------------------------| | | |
| | | | | | |
| | | | | | |
|StoreAns | | | | | |
|---------------------------->| | | |
| | | | | | |
| | | | | | |
|UpdateReq| | | | | |
|<----------------------------| | | |
| | | | | | |
| | | | | | |
|UpdateAns| | | | | |
|---------------------------->| | | |
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | |
In Chord, JP's neighbor table needs to contain its own predecessors.
It couldn't connect to them previously because it did not yet know
their addresses. However, now that it has received an Update from
AP, it has AP's predecessors, which are also its own, so it sends
Attaches to them. Below it is shown connecting to AP's closest
predecessor, PP.
Jennings, et al. Expires May 13, 2010 [Page 121]
Internet-Draft RELOAD Base November 2009
JP PPP PP AP NP NNP BP
| | | | | | |
| | | | | | |
| | | | | | |
|Attach Dest=PP | | | | |
|---------------------------->| | | |
| | | | | | |
| | | | | | |
| | |Attach Dest=PP | | |
| | |<--------| | | |
| | | | | | |
| | | | | | |
| | |AttachAns| | | |
| | |-------->| | | |
| | | | | | |
| | | | | | |
|AttachAns| | | | | |
|<----------------------------| | | |
| | | | | | |
| | | | | | |
|TLS | | | | | |
|...................| | | | |
| | | | | | |
| | | | | | |
|UpdateReq| | | | | |
|------------------>| | | | |
| | | | | | |
| | | | | | |
|UpdateAns| | | | | |
|<------------------| | | | |
| | | | | | |
| | | | | | |
|UpdateReq| | | | | |
|---------------------------->| | | |
| | | | | | |
| | | | | | |
|UpdateAns| | | | | |
|<----------------------------| | | |
| | | | | | |
| | | | | | |
|UpdateReq| | | | | |
|-------------------------------------->| | |
| | | | | | |
| | | | | | |
|UpdateAns| | | | | |
|<--------------------------------------| | |
| | | | | | |
| | | | | | |
Jennings, et al. Expires May 13, 2010 [Page 122]
Internet-Draft RELOAD Base November 2009
Finally, now that JP has a copy of all the data and is ready to route
messages and receive requests, it sends Updates to everyone in its
Routing Table to tell them it is ready to go. Below, it is shown
sending such an update to TP.
JP NP XX TP
| | | |
| | | |
| | | |
|Update | | |
|---------------------------->|
| | | |
| | | |
|UpdateAns| | |
|<----------------------------|
| | | |
| | | |
| | | |
| | | |
12. Security Considerations
12.1. Overview
RELOAD provides a generic storage service, albeit one designed to be
useful for P2PSIP. In this section we discuss security issues that
are likely to be relevant to any usage of RELOAD.
In any Overlay Instance, any given user depends on a number of peers
with which they have no well-defined relationship except that they
are fellow members of the Overlay Instance. In practice, these other
nodes may be friendly, lazy, curious, or outright malicious. No
security system can provide complete protection in an environment
where most nodes are malicious. The goal of security in RELOAD is to
provide strong security guarantees of some properties even in the
face of a large number of malicious nodes and to allow the overlay to
function correctly in the face of a modest number of malicious nodes.
P2PSIP deployments require the ability to authenticate both peers and
resources (users) without the active presence of a trusted entity in
the system. We describe two mechanisms. The first mechanism is
based on public key certificates and is suitable for general
deployments. The second is an admission control mechanism based on
an overlay-wide shared symmetric key.
Jennings, et al. Expires May 13, 2010 [Page 123]
Internet-Draft RELOAD Base November 2009
12.2. Attacks on P2P Overlays
The two basic functions provided by overlay nodes are storage and
routing: some node is responsible for storing a peer's data and for
allowing a third peer to fetch this stored data. Other nodes are
responsible for routing messages to and from the storing nodes. Each
of these issues is covered in the following sections.
P2P overlays are subject to attacks by subversive nodes that may
attempt to disrupt routing, corrupt or remove user registrations, or
eavesdrop on signaling. The certificate-based security algorithms we
describe in this draft are intended to protect overlay routing and
user registration information in RELOAD messages.
To protect the signaling from attackers pretending to be valid peers
(or peers other than themselves), the first requirement is to ensure
that all messages are received from authorized members of the
overlay. For this reason, RELOAD transports all messages over a
secure channel (TLS and DTLS are defined in this document) which
provides message integrity and authentication of the directly
communicating peer. In addition, messages and data are digitally
signed with the sender's private key, providing end-to-end security
for communications.
12.3. Certificate-based Security
This specification stores users' registrations and possibly other
data in an overlay network. This requires a solution to securing
this data as well as securing, as well as possible, the routing in
the overlay. Both types of security are based on requiring that
every entity in the system (whether user or peer) authenticate
cryptographically using an asymmetric key pair tied to a certificate.
When a user enrolls in the Overlay Instance, they request or are
assigned a unique name, such as "alice@dht.example.net". These names
are unique and are meant to be chosen and used by humans much like a
SIP Address of Record (AOR) or an email address. The user is also
assigned one or more Node-IDs by the central enrollment authority.
Both the name and the peer ID are placed in the certificate, along
with the user's public key.
Each certificate enables an entity to act in two sorts of roles:
o As a user, storing data at specific Resource-IDs in the Overlay
Instance corresponding to the user name.
o As a overlay peer with the peer ID(s) listed in the certificate.
Note that since only users of this Overlay Instance need to validate
Jennings, et al. Expires May 13, 2010 [Page 124]
Internet-Draft RELOAD Base November 2009
a certificate, this usage does not require a global PKI. Instead,
certificates are signed by require a central enrollment authority
which acts as the certificate authority for the Overlay Instance.
This authority signs each peer's certificate. Because each peer
possesses the CA's certificate (which they receive on enrollment)
they can verify the certificates of the other entities in the overlay
without further communication. Because the certificates contain the
user/peer's public key, communications from the user/peer can be
verified in turn.
If self-signed certificates are used, then the security provided is
significantly decreased, since attackers can mount Sybil attacks. In
addition, attackers cannot trust the user names in certificates
(though they can trust the Node-IDs because they are
cryptographically verifiable). This scheme may be appropriate for
some small deployments, such as a small office or an ad hoc overlay
set up among participants in a meeting where all hosts on the network
are trusted. Some additional security can be provided by using the
shared secret admission control scheme as well.
Because all stored data is signed by the owner of the data the
storing peer can verify that the storer is authorized to perform a
store at that Resource-ID and also allow any consumer of the data to
verify the provenance and integrity of the data when it retrieves it.
Note that RELOAD does not itself provide a revocation/status
mechanism (though certificates may of course include OCSP responder
information). Thus, certificate lifetimes should be chosen to
balance the compromise window versus the cost of certificate renewal.
Because RELOAD is already designed to operate in the face of some
fraction of malicious peers, this form of compromise is not fatal.
All implementations MUST implement certificate-based security.
12.4. Shared-Secret Security
RELOAD also supports a shared secret admission control scheme that
relies on a single key that is shared among all members of the
overlay. It is appropriate for small groups that wish to form a
private network without complexity. In shared secret mode, all the
peers share a single symmetric key which is used to key TLS-PSK
[RFC4279] or TLS-SRP [RFC5054] mode. A peer which does not know the
key cannot form TLS connections with any other peer and therefore
cannot join the overlay.
One natural approach to a shared-secret scheme is to use a user-
entered password as the key. The difficulty with this is that in
TLS-PSK mode, such keys are very susceptible to dictionary attacks.
Jennings, et al. Expires May 13, 2010 [Page 125]
Internet-Draft RELOAD Base November 2009
If passwords are used as the source of shared-keys, then TLS-SRP is a
superior choice because it is not subject to dictionary attacks.
12.5. Storage Security
When certificate-based security is used in RELOAD, any given
Resource-ID/Kind-ID pair is bound to some small set of certificates.
In order to write data, the writer must prove possession of the
private key for one of those certificates. Moreover, all data is
stored, signed with the same private key that was used to authorize
the storage. This set of rules makes questions of authorization and
data integrity - which have historically been thorny for overlays -
relatively simple.
12.5.1. Authorization
When a client wants to store some value, it first digitally signs the
value with its own private key. It then sends a Store request that
contains both the value and the signature towards the storing peer
(which is defined by the Resource Name construction algorithm for
that particular kind of value).
When the storing peer receives the request, it must determine whether
the storing client is authorized to store at this Resource-ID/Kind-ID
pair. Determining this requires comparing the user's identity to the
requirements of the access control model (see Section 6.3). If it
satisfies those requirements the user is authorized to write, pending
quota checks as described in the next section.
For example, consider the certificate with the following properties:
User name: alice@dht.example.com
Node-ID: 013456789abcdef
Serial: 1234
If Alice wishes to Store a value of the "SIP Location" kind, the
Resource Name will be the SIP AOR "sip:alice@dht.example.com". The
Resource-ID will be determined by hashing the Resource Name. Because
SIP Location uses the USER-NODE-MATCH policy, it first verifies that
the user name in the certificate hashes to the requested Resource-ID.
It then verifies that the node-id in the certificate matches the
dictionary key being used for the store. If both of these checks
succeed, the Store is authorized. Note that because the access
control model is different for different kinds, the exact set of
checks will vary.
Jennings, et al. Expires May 13, 2010 [Page 126]
Internet-Draft RELOAD Base November 2009
12.5.2. Distributed Quota
Being a peer in an Overlay Instance carries with it the
responsibility to store data for a given region of the Overlay
Instance. However, allowing clients to store unlimited amounts of
data would create unacceptable burdens on peers and would also enable
trivial denial of service attacks. RELOAD addresses this issue by
requiring configurations to define maximum sizes for each kind of
stored data. Attempts to store values exceeding this size MUST be
rejected (if peers are inconsistent about this, then strange
artifacts will happen when the zone of responsibility shifts and a
different peer becomes responsible for overlarge data). Because each
Resource-ID/Kind-ID pair is bound to a small set of certificates,
these size restrictions also create a distributed quota mechanism,
with the quotas administered by the central enrollment server.
Allowing different kinds of data to have different size restrictions
allows new usages the flexibility to define limits that fit their
needs without requiring all usages to have expansive limits.
12.5.3. Correctness
Because each stored value is signed, it is trivial for any retrieving
peer to verify the integrity of the stored value. Some more care
needs to be taken to prevent version rollback attacks. Rollback
attacks on storage are prevented by the use of store times and
lifetime values in each store. A lifetime represents the latest time
at which the data is valid and thus limits (though does not
completely prevent) the ability of the storing node to perform a
rollback attack on retrievers. In order to prevent a rollback attack
at the time of the Store request, we require that storage times be
monotonically increasing. Storing peers MUST reject Store requests
with storage times smaller than or equal to those they are currently
storing. In addition, a fetching node which receives a data value
with a storage time older than the result of the previous fetch knows
a rollback has occurred.
12.5.4. Residual Attacks
The mechanisms described here provides a high degree of security, but
some attacks remain possible. Most simply, it is possible for
storing nodes to refuse to store a value (i.e., reject any request).
In addition, a storing node can deny knowledge of values which it has
previously accepted. To some extent these attacks can be ameliorated
by attempting to store to/retrieve from replicas, but a retrieving
client does not know whether it should try this or not, since there
is a cost to doing so.
Jennings, et al. Expires May 13, 2010 [Page 127]
Internet-Draft RELOAD Base November 2009
The certificate-based authentication scheme prevents a single peer
from being able to forge data owned by other peers. Furthermore,
although a subversive peer can refuse to return data resources for
which it is responsible, it cannot return forged data because it
cannot provide authentication for such registrations. Therefore
parallel searches for redundant registrations can mitigate most of
the effects of a compromised peer. The ultimate reliability of such
an overlay is a statistical question based on the replication factor
and the percentage of compromised peers.
In addition, when a kind is multivalued (e.g., an array data model),
the storing node can return only some subset of the values, thus
biasing its responses. This can be countered by using single values
rather than sets, but that makes coordination between multiple
storing agents much more difficult. This is a trade off that must be
made when designing any usage.
12.6. Routing Security
Because the storage security system guarantees (within limits) the
integrity of the stored data, routing security focuses on stopping
the attacker from performing a DOS attack that misroutes requests in
the overlay. There are a few obvious observations to make about
this. First, it is easy to ensure that an attacker is at least a
valid peer in the Overlay Instance. Second, this is a DOS attack
only. Third, if a large percentage of the peers on the Overlay
Instance are controlled by the attacker, it is probably impossible to
perfectly secure against this.
12.6.1. Background
In general, attacks on DHT routing are mounted by the attacker
arranging to route traffic through one or two nodes it controls. In
the Eclipse attack [Eclipse] the attacker tampers with messages to
and from nodes for which it is on-path with respect to a given victim
node. This allows it to pretend to be all the nodes that are
reachable through it. In the Sybil attack [Sybil], the attacker
registers a large number of nodes and is therefore able to capture a
large amount of the traffic through the DHT.
Both the Eclipse and Sybil attacks require the attacker to be able to
exercise control over her peer IDs. The Sybil attack requires the
creation of a large number of peers. The Eclipse attack requires
that the attacker be able to impersonate specific peers. In both
cases, these attacks are limited by the use of centralized,
certificate-based admission control.
Jennings, et al. Expires May 13, 2010 [Page 128]
Internet-Draft RELOAD Base November 2009
12.6.2. Admissions Control
Admission to a RELOAD Overlay Instance is controlled by requiring
that each peer have a certificate containing its peer ID. The
requirement to have a certificate is enforced by using certificate-
based mutual authentication on each connection. (Note: the
following only applies when self-signed certificates are not used.)
Whenever a peer connects to another peer, each side automatically
checks that the other has a suitable certificate. These peer IDs are
randomly assigned by the central enrollment server. This has two
benefits:
o It allows the enrollment server to limit the number of peer IDs
issued to any individual user.
o It prevents the attacker from choosing specific peer IDs.
The first property allows protection against Sybil attacks (provided
the enrollment server uses strict rate limiting policies). The
second property deters but does not completely prevent Eclipse
attacks. Because an Eclipse attacker must impersonate peers on the
other side of the attacker, he must have a certificate for suitable
peer IDs, which requires him to repeatedly query the enrollment
server for new certificates, which will match only by chance. From
the attacker's perspective, the difficulty is that if he only has a
small number of certificates, the region of the Overlay Instance he
is impersonating appears to be very sparsely populated by comparison
to the victim's local region.
12.6.3. Peer Identification and Authentication
In general, whenever a peer engages in overlay activity that might
affect the routing table it must establish its identity. This
happens in two ways. First, whenever a peer establishes a direct
connection to another peer it authenticates via certificate-based
mutual authentication. All messages between peers are sent over this
protected channel and therefore the peers can verify the data origin
of the last hop peer for requests and responses without further
cryptography.
In some situations, however, it is desirable to be able to establish
the identity of a peer with whom one is not directly connected. The
most natural case is when a peer Updates its state. At this point,
other peers may need to update their view of the overlay structure,
but they need to verify that the Update message came from the actual
peer rather than from an attacker. To prevent this, all overlay
routing messages are signed by the peer that generated them.
Replay is typical prevented for messages that impact the topology of
Jennings, et al. Expires May 13, 2010 [Page 129]
Internet-Draft RELOAD Base November 2009
the overlay by having the information come directly, or be verified
by, the nodes that claimed to have generated the update. Data
storage replay detection is done by signing time of the node that
generated the signature on the store request thus providing a time
based replay protection but the time synchronization is only needed
between peers that can write to the same location.
12.6.4. Protecting the Signaling
The goal here is to stop an attacker from knowing who is signaling
what to whom. An attacker is unlikely to be able to observe the
activities of a specific individual given the randomization of IDs
and routing based on the present peers discussed above. Furthermore,
because messages can be routed using only the header information, the
actual body of the RELOAD message can be encrypted during
transmission.
There are two lines of defense here. The first is the use of TLS or
DTLS for each communications link between peers. This provides
protection against attackers who are not members of the overlay. The
second line of defense is to digitally sign each message. This
prevents adversarial peers from modifying messages in flight, even if
they are on the routing path.
12.6.5. Residual Attacks
The routing security mechanisms in RELOAD are designed to contain
rather than eliminate attacks on routing. It is still possible for
an attacker to mount a variety of attacks. In particular, if an
attacker is able to take up a position on the overlay routing between
A and B it can make it appear as if B does not exist or is
disconnected. It can also advertise false network metrics in an
attempt to reroute traffic. However, these are primarily DOS
attacks.
The certificate-based security scheme secures the namespace, but if
an individual peer is compromised or if an attacker obtains a
certificate from the CA, then a number of subversive peers can still
appear in the overlay. While these peers cannot falsify responses to
resource queries, they can respond with error messages, effecting a
DoS attack on the resource registration. They can also subvert
routing to other compromised peers. To defend against such attacks,
a resource search must still consist of parallel searches for
replicated registrations.
Jennings, et al. Expires May 13, 2010 [Page 130]
Internet-Draft RELOAD Base November 2009
13. IANA Considerations
This section contains the new code points registered by this
document. [NOTE TO IANA/RFC-EDITOR: Please replace RFC-AAAA with
the RFC number for this specification in the following list.]
13.1. Port Registrations
[[Note to RFC Editor - this paragraph can be removed before
publication. ]] IANA has already allocated a port for the main peer
to peer protocol. This port has the name p2p-sip and the port number
of 6084. The names of this port may need to be changed as this draft
progresses and if it does careful instructions will be needed to IANA
to ensure the final RFC and IANA registrations are in sync.
IANA will make the following port registration:
+-------------------------------+-----------------------------------+
| Registration Technical | Cullen Jennings |
| Contact | |
| Registration Owner | IETF |
| Transport Protocol | TCP, UDP |
| Port Number | 6084 |
| Service Name | p2psip_enroll |
| Description | RELOAD P2P Protcol |
| Reference | [RFC-AAAA] |
+-------------------------------+-----------------------------------+
13.2. Overlay Algorithm Types
IANA SHALL create a "RELOAD Overlay Algorithm Type" Registry.
Entries in this registry are strings denoting the names of overlay
algorithms. The registration policy for this registry is RFC 5226
IETF Review. The initial contents of this registry are:
+----------------+----------+
| Algorithm Name | RFC |
+----------------+----------+
| chord-reload | RFC-AAAA |
+----------------+----------+
13.3. Access Control Policies
IANA SHALL create a "RELOAD Access Control Policy" Registry. Entries
in this registry are strings denoting access control policies, as
described in Section 6.3. New entries in this registry SHALL be
registered via RFC 5226 Standards Action. The initial contents of
this registry are:
Jennings, et al. Expires May 13, 2010 [Page 131]
Internet-Draft RELOAD Base November 2009
USER-MATCH
NODE-MATCH
USER-NODE-MATCH
NODE-MULTIPLE
13.4. Data Kind-ID
IANA SHALL create a "RELOAD Data Kind-ID" Registry. Entries in this
registry are 32-bit integers denoting data kinds, as described in
Section 4.1.2. Code points in the range 0x00000001 to 0x7fffffff
SHALL be registered via RFC 5226 Standards Action. Code points in
the range 0x8000000 to 0xf0000000 SHALL be registered via RFC 5226
Expert Review. Code points in the range 0xf0000001 to 0xffffffff are
reserved for private use via the kind description mechanism described
in Section 10. The initial contents of this registry are:
+--------------+------------+----------+
| Kind | Kind-ID | RFC |
+--------------+------------+----------+
| INVALID | 0 | RFC-AAAA |
| TURN_SERVICE | 2 | RFC-AAAA |
| CERTIFICATE | 3 | RFC-AAAA |
| Reserved | 0x7fffffff | RFC-AAAA |
| Reserved | 0xffffffff | RFC-AAAA |
+--------------+------------+----------+
13.5. Data Model
IANA SHALL create a "RELOAD Data Model" Registry. Entries in this
registry are 8-bit integers denoting data models, as described in
Section 6.2. Code points in this registry SHALL be registered via
RFC 5226 Standards Action. The initial contents of this registry
are:
+--------------+------+----------+
| Data Model | Code | RFC |
+--------------+------+----------+
| INVALID | 0 | RFC-AAAA |
| SINGLE_VALUE | 1 | RFC-AAAA |
| ARRAY | 2 | RFC-AAAA |
| DICTIONARY | 3 | RFC-AAAA |
| RESERVED | 255 | RFC-AAAA |
+--------------+------+----------+
Jennings, et al. Expires May 13, 2010 [Page 132]
Internet-Draft RELOAD Base November 2009
13.6. Message Codes
IANA SHALL create a "RELOAD Message Code" Registry. Entries in this
registry are 16-bit integers denoting method codes as described in
Section 5.3.3. These codes SHALL be registered via RFC 5226
Standards Action. The initial contents of this registry are:
+--------------------+----------------+----------+
| Message Code Name | Code Value | RFC |
+--------------------+----------------+----------+
| invalid | 0 | RFC-AAAA |
| probe_req | 1 | RFC-AAAA |
| probe_ans | 2 | RFC-AAAA |
| attach_req | 3 | RFC-AAAA |
| attach_ans | 4 | RFC-AAAA |
| unused | 5 | |
| unused | 6 | |
| store_req | 7 | RFC-AAAA |
| store_ans | 8 | RFC-AAAA |
| fetch_req | 9 | RFC-AAAA |
| fetch_ans | 10 | RFC-AAAA |
| remove_req | 11 | RFC-AAAA |
| remove_ans | 12 | RFC-AAAA |
| find_req | 13 | RFC-AAAA |
| find_ans | 14 | RFC-AAAA |
| join_req | 15 | RFC-AAAA |
| join_ans | 16 | RFC-AAAA |
| leave_req | 17 | RFC-AAAA |
| leave_ans | 18 | RFC-AAAA |
| update_req | 19 | RFC-AAAA |
| update_ans | 20 | RFC-AAAA |
| route_query_req | 21 | RFC-AAAA |
| route_query_ans | 22 | RFC-AAAA |
| ping_req | 23 | RFC-AAAA |
| ping_ans | 24 | RFC-AAAA |
| stat_req | 25 | RFC-AAAA |
| stat_ans | 26 | RFC-AAAA |
| attachlite_req | 27 | RFC-AAAA |
| attachlite_ans | 28 | RFC-AAAA |
| app_attach_req | 29 | RFC-AAAA |
| attach_ans | 30 | RFC-AAAA |
| app_attachlite_req | 31 | RFC-AAAA |
| app_attachlite_ans | 32 | RFC-AAAA |
| reserved | 0x8000..0xfffe | RFC-AAAA |
| error | 0xffff | RFC-AAAA |
+--------------------+----------------+----------+
Jennings, et al. Expires May 13, 2010 [Page 133]
Internet-Draft RELOAD Base November 2009
13.7. Error Codes
IANA SHALL create a "RELOAD Error Code" Registry. Entries in this
registry are 16-bit integers denoting error codes. New entries SHALL
be defined via RFC 5226 Standards Action. The initial contents of
this registry are:
+-------------------------------------+----------------+----------+
| Error Code Name | Code Value | RFC |
+-------------------------------------+----------------+----------+
| invalid | 0 | RFC-AAAA |
| Unused | 1 | RFC-AAAA |
| Error_Forbidden | 2 | RFC-AAAA |
| Error_Not_Found | 3 | RFC-AAAA |
| Error_Request_Timeout | 4 | RFC-AAAA |
| Error_Generation_Counter_Too_Low | 5 | RFC-AAAA |
| Error_Incompatible_with_Overlay | 6 | RFC-AAAA |
| Error_Unsupported_Forwarding_Option | 7 | RFC-AAAA |
| Error_Data_Too_Large | 8 | RFC-AAAA |
| Error_Data_Too_Old | 9 | RFC-AAAA |
| Error_TTL_Exceeded | 10 | RFC-AAAA |
| Error_Message_Too_Large | 11 | RFC-AAAA |
| Error_Unknown_Kind | 12 | RFC-AAAA |
| Error_Unknown_Extension | 13 | RFC-AAAA |
| reserved | 0x8000..0xfffe | RFC-AAAA |
+-------------------------------------+----------------+----------+
13.8. Transport Types
IANA shall create a "RELOAD Transport." New entries SHALL be defined
via RFC 5226 Standards Action. This registry SHALL be initially
populated with the following values:
+---------------------+------+---------------+
| Protocol | Code | Specification |
+---------------------+------+---------------+
| reserved | 0 | RFC-AAAA |
| UDP (DTLS over UDP) | 1 | RFC-AAAA |
| TCP (TLS over TCP) | 2 | RFC-AAAA |
| reserved | 255 | RFC-AAAA |
+---------------------+------+---------------+
13.9. Forwarding Options
IANA shall create a "Forwarding Option Registry". Entries in this
registry between 1 and 127 SHALL be defined via RFC 5226 Standards
Action. Entries in this registry between 128 and 254 SHALL be
defined via RFC 5226 Specification Required. This registry SHALL be
Jennings, et al. Expires May 13, 2010 [Page 134]
Internet-Draft RELOAD Base November 2009
initially populated with the following values:
+-------------------+------+---------------+
| Forwarding Option | Code | Specification |
+-------------------+------+---------------+
| invalid | 0 | RFC-AAAA |
| reserved | 255 | RFC-AAAA |
+-------------------+------+---------------+
13.10. Probe Information Types
IANA shall create a "RELOAD Probe Information Type Registry".
Entries in this registry SHALL be defined via RFC 5226 Standards
Action. This registry SHALL be initially populated with the
following values:
+-----------------+------+---------------+
| Probe Option | Code | Specification |
+-----------------+------+---------------+
| invalid | 0 | RFC-AAAA |
| responsible_set | 1 | RFC-AAAA |
| num_resources | 2 | RFC-AAAA |
| uptime | 3 | RFC-AAAA |
| reserved | 255 | RFC-AAAA |
+-----------------+------+---------------+
13.11. Message Extensions
IANA shall create a "RELOAD Extensions Registry". Entries in this
registry SHALL be defined via RFC 5226 Specification Required. This
registry SHALL be initially populated with the following values:
+-----------------+--------+---------------+
| Extensions Name | Code | Specification |
+-----------------+--------+---------------+
| invalid | 0 | RFC-AAAA |
| reserved | 0xFFFF | RFC-AAAA |
+-----------------+--------+---------------+
13.12. reload URI Scheme
This section describes the scheme for a reload URI, which can be used
to refer to either:
o A peer.
o A resource inside a peer.
The reload URI is defined using a subset of the URI schema specified
Jennings, et al. Expires May 13, 2010 [Page 135]
Internet-Draft RELOAD Base November 2009
in Appendix A of RFC 3986 [RFC3986] and the associated URI Guidelines
[RFC4395] per the following ABNF syntax:
RELOAD-URI = "reload://" destination "@" overlay "/"
[specifier]
destination = 1 * HEXDIG
overlay = reg-name
specifier = 1*HEXDIG
The definitions of these productions are as follows:
destination: a hex-encoded Destination List object.
overlay: the name of the overlay.
specifier : a hex-encoded StoredDataSpecifier indicating the data
element.
If no specifier is present then this URI addresses the peer which can
be reached via the indicated destination list at the indicated
overlay name. If a specifier is present, then the URI addresses the
data value.
13.12.1. URI Registration
The following summarizes the information necessary to register the
reload URI.
URI Scheme Name: reload
Status: permanent
URI Scheme Syntax: see Section 13.12 of RFC-AAAA
URI Scheme Semantics: The reload URI is intended to be used as a
reference to a RELOAD peer or resource.
Encoding Considerations: The reload URI is not intended to be
human-readable text, so it is encoded entirely in US-ASCII.
Applications/protocols that use this URI scheme: The RELOAD
protocol described in RFC-AAAA.
Interoperability considerations See RFC-AAAA.
Security considerations See RFC-AAAA
Contact Cullen Jennings
Author/Change controller IESG
References RFC-AAAA
Jennings, et al. Expires May 13, 2010 [Page 136]
Internet-Draft RELOAD Base November 2009
14. Acknowledgments
This draft is a merge of the "REsource LOcation And Discovery
(RELOAD)" draft by David A. Bryan, Marcia Zangrilli and Bruce B.
Lowekamp, the "Address Settlement by Peer to Peer" draft by Cullen
Jennings, Jonathan Rosenberg, and Eric Rescorla, the "Security
Extensions for RELOAD" draft by Bruce B. Lowekamp and James Deverick,
the "A Chord-based DHT for Resource Lookup in P2PSIP" by Marcia
Zangrilli and David A. Bryan, and the Peer-to-Peer Protocol (P2PP)
draft by Salman A. Baset, Henning Schulzrinne, and Marcin
Matuszewski. Thanks to the authors of RFC 5389 for text included
from that. Vidya Narayanan provided many comments and imporvements.
The ideas for the Chord specific extension data to the Leave
mechanisms and text provided by J. Maenpaa, G. Camarillo, and J.
Hautakorp.
Thanks to the many people who contributed including Ted Hardie,
Michael Chen, Dan York, Das Saumitra, Lyndsay Campbell, Brian Rosen,
David Bryan, Michael Chen, Dave Craig, and Julian Cain.
15. References
15.1. Normative References
[RFC2119] Bradner, S., "Key words for use in RFCs to Indicate
Requirement Levels", BCP 14, RFC 2119, March 1997.
[I-D.ietf-mmusic-ice]
Rosenberg, J., "Interactive Connectivity Establishment
(ICE): A Protocol for Network Address Translator (NAT)
Traversal for Offer/Answer Protocols",
draft-ietf-mmusic-ice-19 (work in progress), October 2007.
[RFC5389] Rosenberg, J., Mahy, R., Matthews, P., and D. Wing,
"Session Traversal Utilities for NAT (STUN)", RFC 5389,
October 2008.
[I-D.ietf-behave-turn]
Rosenberg, J., Mahy, R., and P. Matthews, "Traversal Using
Relays around NAT (TURN): Relay Extensions to Session
Traversal Utilities for NAT (STUN)",
draft-ietf-behave-turn-16 (work in progress), July 2009.
[RFC5273] Schaad, J. and M. Myers, "Certificate Management over CMS
(CMC): Transport Protocols", RFC 5273, June 2008.
Jennings, et al. Expires May 13, 2010 [Page 137]
Internet-Draft RELOAD Base November 2009
[RFC5272] Schaad, J. and M. Myers, "Certificate Management over CMS
(CMC)", RFC 5272, June 2008.
[RFC4279] Eronen, P. and H. Tschofenig, "Pre-Shared Key Ciphersuites
for Transport Layer Security (TLS)", RFC 4279,
December 2005.
[I-D.ietf-mmusic-ice-tcp]
Rosenberg, J., "TCP Candidates with Interactive
Connectivity Establishment (ICE)",
draft-ietf-mmusic-ice-tcp-07 (work in progress),
July 2008.
[RFC5246] Dierks, T. and E. Rescorla, "The Transport Layer Security
(TLS) Protocol Version 1.2", RFC 5246, August 2008.
[RFC4347] Rescorla, E. and N. Modadugu, "Datagram Transport Layer
Security", RFC 4347, April 2006.
[RFC5348] Floyd, S., Handley, M., Padhye, J., and J. Widmer, "TCP
Friendly Rate Control (TFRC): Protocol Specification",
RFC 5348, September 2008.
[RFC4648] Josefsson, S., "The Base16, Base32, and Base64 Data
Encodings", RFC 4648, October 2006.
[RFC2988] Paxson, V. and M. Allman, "Computing TCP's Retransmission
Timer", RFC 2988, November 2000.
[RFC3986] Berners-Lee, T., Fielding, R., and L. Masinter, "Uniform
Resource Identifier (URI): Generic Syntax", STD 66,
RFC 3986, January 2005.
[RFC4395] Hansen, T., Hardie, T., and L. Masinter, "Guidelines and
Registration Procedures for New URI Schemes", BCP 35,
RFC 4395, February 2006.
15.2. Informative References
[I-D.maenpaa-p2psip-self-tuning]
Maenpaa, J., Camarillo, G., and J. Hautakorpi, "A Self-
tuning Distributed Hash Table (DHT) for REsource LOcation
And Discovery (RELOAD)",
draft-maenpaa-p2psip-self-tuning-00 (work in progress),
February 2009.
[I-D.baset-tsvwg-tcp-over-udp]
Baset, S. and H. Schulzrinne, "TCP-over-UDP",
Jennings, et al. Expires May 13, 2010 [Page 138]
Internet-Draft RELOAD Base November 2009
draft-baset-tsvwg-tcp-over-udp-01 (work in progress),
June 2009.
[RFC5201] Moskowitz, R., Nikander, P., Jokela, P., and T. Henderson,
"Host Identity Protocol", RFC 5201, April 2008.
[RFC4828] Floyd, S. and E. Kohler, "TCP Friendly Rate Control
(TFRC): The Small-Packet (SP) Variant", RFC 4828,
April 2007.
[I-D.ietf-p2psip-concepts]
Bryan, D., Matthews, P., Shim, E., Willis, D., and S.
Dawkins, "Concepts and Terminology for Peer to Peer SIP",
draft-ietf-p2psip-concepts-02 (work in progress),
July 2008.
[RFC1122] Braden, R., "Requirements for Internet Hosts -
Communication Layers", STD 3, RFC 1122, October 1989.
[RFC5382] Guha, S., Biswas, K., Ford, B., Sivakumar, S., and P.
Srisuresh, "NAT Behavioral Requirements for TCP", BCP 142,
RFC 5382, October 2008.
[RFC4145] Yon, D. and G. Camarillo, "TCP-Based Media Transport in
the Session Description Protocol (SDP)", RFC 4145,
September 2005.
[RFC4571] Lazzaro, J., "Framing Real-time Transport Protocol (RTP)
and RTP Control Protocol (RTCP) Packets over Connection-
Oriented Transport", RFC 4571, July 2006.
[RFC2818] Rescorla, E., "HTTP Over TLS", RFC 2818, May 2000.
[RFC4086] Eastlake, D., Schiller, J., and S. Crocker, "Randomness
Requirements for Security", BCP 106, RFC 4086, June 2005.
[RFC5054] Taylor, D., Wu, T., Mavrogiannopoulos, N., and T. Perrin,
"Using the Secure Remote Password (SRP) Protocol for TLS
Authentication", RFC 5054, November 2007.
[RFC5280] Cooper, D., Santesson, S., Farrell, S., Boeyen, S.,
Housley, R., and W. Polk, "Internet X.509 Public Key
Infrastructure Certificate and Certificate Revocation List
(CRL) Profile", RFC 5280, May 2008.
[I-D.matthews-p2psip-bootstrap-mechanisms]
Cooper, E., "Bootstrap Mechanisms for P2PSIP",
draft-matthews-p2psip-bootstrap-mechanisms-00 (work in
Jennings, et al. Expires May 13, 2010 [Page 139]
Internet-Draft RELOAD Base November 2009
progress), February 2007.
[I-D.garcia-p2psip-dns-sd-bootstrapping]
Garcia, G., "P2PSIP bootstrapping using DNS-SD",
draft-garcia-p2psip-dns-sd-bootstrapping-00 (work in
progress), October 2007.
[I-D.pascual-p2psip-clients]
Pascual, V., Matuszewski, M., Shim, E., Zhang, H., and S.
Yongchao, "P2PSIP Clients",
draft-pascual-p2psip-clients-01 (work in progress),
February 2008.
[RFC4787] Audet, F. and C. Jennings, "Network Address Translation
(NAT) Behavioral Requirements for Unicast UDP", BCP 127,
RFC 4787, January 2007.
[RFC2311] Dusse, S., Hoffman, P., Ramsdell, B., Lundblade, L., and
L. Repka, "S/MIME Version 2 Message Specification",
RFC 2311, March 1998.
[I-D.jiang-p2psip-sep]
Jiang, X. and H. Zhang, "Service Extensible P2P Peer
Protocol", draft-jiang-p2psip-sep-01 (work in progress),
February 2008.
[I-D.hardie-p2poverlay-pointers]
Hardie, T., "Mechanisms for use in pointing to overlay
networks, nodes, or resources",
draft-hardie-p2poverlay-pointers-00 (work in progress),
January 2008.
[I-D.ietf-p2psip-sip]
Jennings, C., Lowekamp, B., Rescorla, E., Baset, S., and
H. Schulzrinne, "A SIP Usage for RELOAD",
draft-ietf-p2psip-sip-01 (work in progress), March 2009.
[Sybil] Douceur, J., "The Sybil Attack", IPTPS 02, March 2002.
[Eclipse] Singh, A., Ngan, T., Druschel, T., and D. Wallach,
"Eclipse Attacks on Overlay Networks: Threats and
Defenses", INFOCOM 2006, April 2006.
[non-transitive-dhts-worlds05]
Freedman, M., Lakshminarayanan, K., Rhea, S., and I.
Stoica, "Non-Transitive Connectivity and DHTs",
WORLDS'05.
Jennings, et al. Expires May 13, 2010 [Page 140]
Internet-Draft RELOAD Base November 2009
[lookups-churn-p2p06]
Wu, D., Tian, Y., and K. Ng, "Analytical Study on
Improving DHT Lookup Performance under Churn", IEEE
P2P'06.
[bryan-design-hotp2p08]
Bryan, D., Lowekamp, B., and M. Zangrilli, "The Design of
a Versatile, Secure P2PSIP Communications Architecture for
the Public Internet", Hot-P2P'08.
[opendht-sigcomm05]
Rhea, S., Godfrey, B., Karp, B., Kubiatowicz, J.,
Ratnasamy, S., Shenker, S., Stoica, I., and H. Yu,
"OpenDHT: A Public DHT and its Uses", SIGCOMM'05.
[Chord] Stoica, I., Morris, R., Liben-Nowell, D., Karger, D.,
Kaashoek, M., Dabek, F., and H. Balakrishnan, "Chord: A
Scalable Peer-to-peer Lookup Protocol for Internet
Applications", IEEE/ACM Transactions on Networking Volume
11, Issue 1, 17-32, Feb 2003.
[vulnerabilities-acsac04]
Srivatsa, M. and L. Liu, "Vulnerabilities and Security
Threats in Structured Peer-to-Peer Systems: A Quantitative
Analysis", ACSAC 2004.
[minimizing-churn-sigcomm06]
Godfrey, P., Shenker, S., and I. Stoica, "Minimizing Churn
in Distributed Systems", SIGCOMM 2006.
Appendix A. Change Log
A.1. Changes since draft-ietf-p2psip-reload-04
o Renamed the XML element in configuration files from to .
A.2. Changes since draft-ietf-p2psip-reload-01
o Added the ability to introduce new kinds dynamically.
o Added configuration file updating.
o Major revisions to reliability and flow control algorithms.
o Moved diagnostics out--they now go in a separate draft.
o Removed REMOVE: you now store a "nonexistent" element.
Jennings, et al. Expires May 13, 2010 [Page 141]
Internet-Draft RELOAD Base November 2009
A.3. Changes since draft-ietf-p2psip-reload-00
o Split base protocol from combined draft into new draft.
o Update architecture discussion to address concerns raised about
clarity of roles.
o Moved extensive discussion of routing and client behaviors to
appendix.
o Split Ping into Ping and Probe.
o Added AttachLite to provide way to implement ICE-Lite.
o Added Stat call for retrieving meta-data.
o Added discussion of periodic vs reactive recovery issue.
o Changed finger table stabilization to prefer long-lived over best-
match.
o Updated IANA considerations to be more complete.
o Changed error codes from http-based.
A.4. Changes since draft-ietf-p2psip-base-00
o Removed TUNNEL method
o Allow implementations more flexibility in picking finger table
entries and revising random range.
o Decouple overlay configuration from enrollment server.
o Add error for data too large.
o Change architecture to overlay perspective from previous revision
and update terminology in document to match.
A.5. Changes since draft-ietf-p2psip-base-01
o Reordered message routing section to clarify that other routing
algorithms are possible besides symmetric recursive.
o Clarified document IPR terms.
A.6. Changes since draft-ietf-p2psip-base-01a
o Fragment offset was too small to hold 2^24 bit messages, so fixed
this from 16 bits to 32 bits.
o Changed absolute times from seconds to milliseconds.
o Added error for messages over max size.
o Added error for TTL expired.
o Add time in response to PING.
o Clarified retransmission and fragmentation algorithm.
o Clarified acknowledgement tracking for congestion control.
A.7. Changes since draft-ietf-p2psip-base-02
o Rearranged forwarding header to fix alignment, among other issues.
Jennings, et al. Expires May 13, 2010 [Page 142]
Internet-Draft RELOAD Base November 2009
o Removed route logging.
o Switched to binary ICE for Attach.
o ConfigUpdate improved.
o Change from close DTLS session on fragmentation attack to drop
fragments, indirect attack.
o Updates to trivial sender/receiver text.
o Updates to data model based on list discussion.
o Updates to chord overlay algorithm section.
o Added AppAttach and removed port number from Attach.
o Changed via-list to use shorter structure.
o Rewrote fragmentation.
o Moved AIMD and TFRC congestion control algorithms to appendix
until further WG effort decides direction there.
Appendix B. AIMD Retransmission Scheme
This Appendix specifies an optional sender retransmission algorithm
with better performance that senders MAY implement and is based on
the theAIMD algorithm in TCP. The algorithm here is only the AIMD
portion of TCP. All other features are restricted to simplify the
implementation, i.e. no slow start (initial window is 1), no fast
retransmission, and no fast recovery.
AIMD extends stop and wait defined in Section 5.6.2.2 by allowing
multiple fragments to be pending at the same time. The sender allows
w unacknowledged fragments to be outstanding at any given time. w is
initially set to one. In each RTO interval in which no
retransmissions occur, w is increased by one. When a loss occurs, w
is halved. After halving w, if there are more than w fragments for
which an ACK is pending, no further retransmissions of the most
recently initiated fragments are performed until they fit in the
window w, at which point they begin the retransmission algorithm
again. The value w is held fixed for one RTO. After that point, if
additional retransmissions occur, it will be halved again; otherwise
it may be incremented after an additional RTO without loss.
If w drops to one and the one pending fragment is not ACKed by the
other side after 5 requests are sent, the link is considered to have
failed. Otherwise, unACKed fragments are simply dropped.
Appendix C. TFRC Retransmission Scheme
This Appendix specifies an optional TFRC (RFC5348) based scheme that
can be implemented in the Sender-Based Variant format with the same
receiver algorithm. This implementation requires the sender to
maintain precise timestamps in ms of the transmission time of each
Jennings, et al. Expires May 13, 2010 [Page 143]
Internet-Draft RELOAD Base November 2009
sequence number as well as the segment sizes. That, combined with
the ACKs, allows the sender to calculate the performance required by
TFRC, including information calculated by the receiver in the
conventional form of TFRC.
TFRC is used for congestion control. For reliability, an individual
fragment is retransmitted up to twice at RTO intervals, pending the
availability of room in the congestion window. If it is not ACKed
after another RTO following its last retransmission, it is dropped.
Appendix D. Routing Alternatives
Significant discussion has been focused on the selection of a routing
algorithm for P2PSIP. This section discusses the motivations for
selecting symmetric recursive routing for RELOAD and describes the
extensions that would be required to support additional routing
algorithms.
D.1. Iterative vs Recursive
Iterative routing has a number of advantages. It is easier to debug,
consumes fewer resources on intermediate peers, and allows the
querying peer to identify and route around misbehaving peers
[non-transitive-dhts-worlds05]. However, in the presence of NATs,
iterative routing is intolerably expensive because a new connection
must be established for each hop (using ICE) [bryan-design-hotp2p08].
Iterative routing is supported through the Route_Query mechanism and
is primarily intended for debugging. It also allows the querying
peer to evaluate the routing decisions made by the peers at each hop,
consider alternatives, and perhaps detect at what point the
forwarding path fails.
D.2. Symmetric vs Forward response
An alternative to the symmetric recursive routing method used by
RELOAD is Forward-Only routing, where the response is routed to the
requester as if it were a new message initiated by the responder (in
the previous example, Z sends the response to A as if it were sending
a request). Forward-only routing requires no state in either the
message or intermediate peers.
The drawback of forward-only routing is that it does not work when
the overlay is unstable. For example, if A is in the process of
joining the overlay and is sending a Join request to Z, it is not yet
reachable via forward routing. Even if it is established in the
overlay, if network failures produce temporary instability, A may not
Jennings, et al. Expires May 13, 2010 [Page 144]
Internet-Draft RELOAD Base November 2009
be reachable (and may be trying to stabilize its network connectivity
via Attach messages).
Furthermore, forward-only responses are less likely to reach the
querying peer than symmetric recursive ones are, because the forward
path is more likely to have a failed peer than is the request path
(which was just tested to route the request)
[non-transitive-dhts-worlds05].
An extension to RELOAD that supports forward-only routing but relies
on symmetric responses as a fallback would be possible, but due to
the complexities of determining when to use forward-only and when to
fallback to symmetric, we have chosen not to include it as an option
at this point.
D.3. Direct Response
Another routing option is Direct Response routing, in which the
response is returned directly to the querying node. In the previous
example, if A encodes its IP address in the request, then Z can
simply deliver the response directly to A. In the absence of NATs or
other connectivity issues, this is the optimal routing technique.
The challenge of implementing direct response is the presence of
NATs. There are a number of complexities that must be addressed. In
this discussion, we will continue our assumption that A issued the
request and Z is generating the response.
o The IP address listed by A may be unreachable, either due to NAT
or firewall rules. Therefore, a direct response technique must
fallback to symmetric response [non-transitive-dhts-worlds05].
The hop-by-hop ACKs used by RELOAD allow Z to determine when A has
received the message (and the TLS negotiation will provide earlier
confirmation that A is reachable), but this fallback requires a
timeout that will increase the response latency whenever A is not
reachable from Z.
o Whenever A is behind a NAT it will have multiple candidate IP
addresses, each of which must be advertised to ensure
connectivity; therefore Z will need to attempt multiple
connections to deliver the response.
o One (or all) of A's candidate addresses may route from Z to a
different device on the Internet. In the worst case these nodes
may actually be running RELOAD on the same port. Therefore, it is
absolutely necessary to establish a secure connection to
authenticate A before delivering the response. This step
diminishes the efficiency of direct response because multiple
roundtrips are required before the message can be delivered.
Jennings, et al. Expires May 13, 2010 [Page 145]
Internet-Draft RELOAD Base November 2009
o If A is behind a NAT and does not have a connection already
established with Z, there are only two ways the direct response
will work. The first is that A and Z both be behind the same NAT,
in which case the NAT is not involved. In the more common case,
when Z is outside A's NAT, the response will only be received if
A's NAT implements endpoint-independent filtering. As the choice
of filtering mode conflates application transparency with security
[RFC4787], and no clear recommendation is available, the
prevalence of this feature in future devices remains unclear.
An extension to RELOAD that supports direct response routing but
relies on symmetric responses as a fallback would be possible, but
due to the complexities of determining when to use direct response
and when to fallback to symmetric, and the reduced performance for
responses to peers behind restrictive NATs, we have chosen not to
include it as an option at this point.
D.4. Relay Peers
SEP [I-D.jiang-p2psip-sep] has proposed implementing a form of direct
response by having A identify a peer, Q, that will be directly
reachable by any other peer. A uses Attach to establish a connection
with Q and advertises Q's IP address in the request sent to Z. Z
sends the response to Q, which relays it to A. This then reduces the
latency to two hops, plus Z negotiating a secure connection to Q.
This technique relies on the relative population of nodes such as A
that require relay peers and peers such as Q that are capable of
serving as a relay peer. It also requires nodes to be able to
identify which category they are in. This identification problem has
turned out to be hard to solve and is still an open area of
exploration.
An extension to RELOAD that supports relay peers is possible, but due
to the complexities of implementing such an alternative, we have not
added such a feature to RELOAD at this point.
A concept similar to relay peers, essentially choosing a relay peer
at random, has previously been suggested to solve problems of
pairwise non-transitivity [non-transitive-dhts-worlds05], but
deterministic filtering provided by NATs makes random relay peers no
more likely to work than the responding peer.
D.5. Symmetric Route Stability
A common concern about symmetric recursive routing has been that one
or more peers along the request path may fail before the response is
received. The significance of this problem essentially depends on
Jennings, et al. Expires May 13, 2010 [Page 146]
Internet-Draft RELOAD Base November 2009
the response latency of the overlay. An overlay that produces slow
responses will be vulnerable to churn, whereas responses that are
delivered very quickly are vulnerable only to failures that occur
over that small interval.
The other aspect of this issue is whether the request itself can be
successfully delivered. Assuming typical connection maintenance
intervals, the time period between the last maintenance and the
request being sent will be orders of magnitude greater than the delay
between the request being forwarded and the response being received.
Therefore, if the path was stable enough to be available to route the
request, it is almost certainly going to remain available to route
the response.
An overlay that is unstable enough to suffer this type of failure
frequently is unlikely to be able to support reliable functionality
regardless of the routing mechanism. However, regardless of the
stability of the return path, studies show that in the event of high
churn, iterative routing is a better solution to ensure request
completion [lookups-churn-p2p06] [non-transitive-dhts-worlds05]
Finally, because RELOAD retries the end-to-end request, that retry
will address the issues of churn that remain.
Appendix E. Why Clients?
There are a wide variety of reasons a node may act as a client rather
than as a peer [I-D.pascual-p2psip-clients]. This section outlines
some of those scenarios and how the client's behavior changes based
on its capabilities.
E.1. Why Not Only Peers?
For a number of reasons, a particular node may be forced to act as a
client even though it is willing to act as a peer. These include:
o The node does not have appropriate network connectivity, typically
because it has a low-bandwidth network connection.
o The node may not have sufficient resources, such as computing
power, storage space, or battery power.
o The overlay algorithm may dictate specific requirements for peer
selection. These may include participating in the overlay to
determine trustworthiness; controlling the number of peers in the
overlay to reduce overly-long routing paths; or ensuring minimum
application uptime before a node can join as a peer.
The ultimate criteria for a node to become a peer are determined by
Jennings, et al. Expires May 13, 2010 [Page 147]
Internet-Draft RELOAD Base November 2009
the overlay algorithm and specific deployment. A node acting as a
client that has a full implementation of RELOAD and the appropriate
overlay algorithm is capable of locating its responsible peer in the
overlay and using Attach to establish a direct connection to that
peer. In that way, it may elect to be reachable under either of the
routing approaches listed above. Particularly for overlay algorithms
that elect nodes to serve as peers based on trustworthiness or
population, the overlay algorithm may require such a client to locate
itself at a particular place in the overlay.
E.2. Clients as Application-Level Agents
SIP defines an extensive protocol for registration and security
between a client and its registrar/proxy server(s). Any SIP device
can act as a client of a RELOAD-based P2PSIP overlay if it contacts a
peer that implements the server-side functionality required by the
SIP protocol. In this case, the peer would be acting as if it were
the user's peer, and would need the appropriate credentials for that
user.
Application-level support for clients is defined by a usage. A usage
offering support for application-level clients should specify how the
security of the system is maintained when the data is moved between
the application and RELOAD layers.
Authors' Addresses
Cullen Jennings
Cisco
170 West Tasman Drive
MS: SJC-21/2
San Jose, CA 95134
USA
Phone: +1 408 421-9990
Email: fluffy@cisco.com
Bruce B. Lowekamp (editor)
MYMIC LLC
1040 University Blvd., Suite 100
Portsmouth, VA 23703
USA
Email: bbl@lowekamp.net
Jennings, et al. Expires May 13, 2010 [Page 148]
Internet-Draft RELOAD Base November 2009
Eric Rescorla
Network Resonance
2064 Edgewood Drive
Palo Alto, CA 94303
USA
Phone: +1 650 320-8549
Email: ekr@networkresonance.com
Salman A. Baset
Columbia University
1214 Amsterdam Avenue
New York, NY
USA
Email: salman@cs.columbia.edu
Henning Schulzrinne
Columbia University
1214 Amsterdam Avenue
New York, NY
USA
Email: hgs@cs.columbia.edu
Jennings, et al. Expires May 13, 2010 [Page 149]