Port Control Protocol (PCP) Failure Scenarios
draft-boucadair-pcp-failure-05

Abstract

This document identifies and analyzes several PCP failure scenarios. Identifying these failure scenarios are useful to assess the efficiency of the protocol and also to decide whether new extensions are needed to the base PCP.

A procedure to retrieve the explicit dynamic mapping(s) from the PCP Server is proposed. This procedure relies upon the use of a new PCP OpCode and Option: GET/NEXT.

Requirements Language

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].

Status of This Memo

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

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

Internet-Drafts are draft documents valid for a maximum of six months and may be updated, replaced, or obsoleted by other documents at any time. It is inappropriate to use Internet-Drafts as reference material or to cite them other than as "work in progress."

This Internet-Draft will expire on August 19, 2013.

Copyright Notice

Copyright (c) 2013 IETF Trust and the persons identified as the document authors. All rights reserved.

This document is subject to BCP 78 and the IETF Trust's Legal Provisions Relating to IETF Documents (http://trustee.ietf.org/license-info) in effect on the date of publication of this document. Please review these documents carefully, as they describe your rights and restrictions with respect to this document. Code Components extracted from this document must include Simplified BSD License text as described in Section 4.e of the Trust Legal Provisions and are provided without warranty as described in the Simplified BSD License.

Table of Contents

• 1. Introduction
• 2. PCP Client Failure Scenarios
• 2.1. Change of the IP Address of The PCP Server
• 2.2. Application Crash
• 2.3. PCP Client Crash
• 2.4. Change of the Internal IP Address
• 2.5. Change of the CPE WAN IP Address
• 2.6. UPnP IGD/PCP IWF
• 3. Restart or Failure of the PCP Server
• 3.1. Basic Rule
• 3.2. Clear PCP Mappings
• 3.3. State Redundancy is Enabled
• 3.4. Cold-Standby without State Redundancy
• 3.5. Anycast Redundancy Mode
• 3.6. Mapping Repair Procedure
• 3.6.1. PCP Client Behaviour
• 3.6.2. PCP Proxy Behaviour
• 4. Security Considerations
• 5. IANA Considerations
• 6. Acknowledgements
• 7. References
• 7.1. Normative References
• 7.2. Informative References
• Appendix A. PCP State Synchronization: Overview
• Appendix B. GET/NEXT Operation
• B.1. OpCode Format
• B.2. OpCode-Specific Result Code
• B.3. Ordering and Equality
• B.4. NEXT Option
• B.5. GET/NEXT PCP Client Theory of Operation
• B.6. GET/NEXT PCP Server Theory of Operation
• B.7. Flow Examples
• Authors' Addresses

1. Introduction

This document discusses several failure scenarios that may occur when deploying PCP [I-D.ietf-pcp-base].

2. PCP Client Failure Scenarios

2.1. Change of the IP Address of The PCP Server

When a new IP address is used to reach its PCP Server, the PCP Client MUST re-create all of its explicit dynamic mappings using the newly discovered IP address.

The PCP Client must undertake the same process as per refreshing an existing explicit dynamic mapping (see [I-D.ietf-pcp-base]); the only difference is the PCP requests are sent to a distinct IP address. No specific behavior is required from the PCP Server for handling these requests.

2.2. Application Crash

When a fatal error is encountered by an application relying on PCP to open explicit dynamic mappings on an upstream device, and upon the restart of that application, the PCP Client should issue appropriate requests to refresh the explicit dynamic mappings of that application (e.g., clear old mappings and install new ones using the new port number used by the application).

If the same port number is used but a distinct Mapping Nonce is generated, the request will be rejected with a NOT_AUTHORIZED error with the Lifetime of the error indicating duration of that existing mapping (see Section 2.7 of [I-D.boucadair-pcp-flow-examples]). A solution to recover the Mapping Nonce used when instantiating the mapping may be envisaged; this solution may not be viable if PCP authentication is not in use. Mapping Nonce recovery in the simple PCP threat model (especially when Mapping Check validation is enabled) induced the same security threated as those discussed in [I-D.ietf-pcp-base].

If a distinct port number is used by the application to bound its service (i.e., a new internal port number is to be signaled in PCP), the PCP Server may honor the refresh requests if the per-subscriber quota is not exceeded. A distinct external port number would be assigned by the PCP Server due to the presence of "stale" explicit dynamic mapping(s) associated with the "old" port number.

To avoid this inconvenience induced by stale explicit dynamic mappings, the PCP Client MAY clear the "old" mappings before issuing the refresh requests; but this would require the PCP Client to store the information about the "old" port number. This can be easy to solve if the PCP Client is embedded in the application. In some scenarios, this is not so easy because the PCP Client may handle PCP requests on behalf of several applications and no means to identify the requesting application may be supported. Means to identify the application are implementation-specific and are out of scope of this document.

[I-D.ietf-pcp-base] does not allow anymore a PCP Client to issue a request to delete all the explicit dynamic mappings associated with an internal IP address. If a PCP Client is allowed to clear all mappings bound to the same IP address, this would have negative impact on other applications and PCP Client(s) which may use the same internal IP address to instruct their explicit dynamic mappings in the PCP Server.

2.3. PCP Client Crash

The PCP Client may encounter a fatal error leading to its restart. In such case, the internal IP address and port numbers used by requesting applications are not impacted. Therefore, the explicit dynamic mappings as maintained by the PCP Server are accurate and there is no need to refresh them.

On the PCP Client side, a new UDP port should be assigned to issue PCP requests. As a consequence, if outstanding requests have been sent to the PCP Server, the responses are likely to be lost.

If the PCP Client stores its explicit dynamic mappings in a persistent memory, there is no need to retrieve the list of active mappings from the PCP Server. If several PCP Clients are co-located on the same host, related PCP mapping tables should be uniquely distinguished (e.g., a PCP Client does not delete explicit dynamic mappings instructed by another PCP Client.)

If the PCP Client does not store the explicit dynamic mappings and new Mapping Nonces are assigned, the PCP Server will reject to refresh these mappings. This issue can be solved if the PCP Client uses GET OpCode (Appendix B) to recover the mapping nonces used when instantiating the mappings if PCP authentication is used or Mapping Nonce validation check is disabled.

If the PCP Client (or the application) is crashing, it should be allocating short PCP lifetimes until it is debugged and running properly. If it is never debugged and never running properly, it should continue to request short PCP lifetimes.

2.4. Change of the Internal IP Address

When a new IP address is assigned to a host embedding a PCP Client, the PCP Client MUST install on the PCP Server all the explicit dynamic mappings it manages, using the new assigned IP address as the internal IP address. The hinted external port number won't be assigned by the PCP Server since a "stale" mapping is already instantiated by the PCP Server (but it is associated with a distinct internal IP address).

For a host configured with several addresses, the PCP Client MUST maintain a record about the target IP address it used when issuing its PCP requests. If no record is maintained and upon a change of the IP address or de-activation of an interface, the PCP-instructed explicit dynamic mappings are broken and inbound communications will fail to be delivered.

Depending on the configured policies, the PCP Server may honor all or part of the requests received from the PCP Client. Upon receipt of the response from the PCP Server, the PCP Client MUST update its local PCP state with the new assigned port numbers and external IP address.

Because of the possible negative impact if the quota is exceed due to the presence of stale mappings (see the example in Section 2.14 of [I-D.boucadair-pcp-flow-examples]), a procedure to clear stale mappings may have some benefits (but also some side effect as discussed above). Note PCP does not support such functionality anymore.

A PCP Client may be used to manage explicit dynamic mappings on behalf of a third party (i.e., the PCP Client and the third party are not co-located on the same host). If a new internal IP address is assigned to that third party (e.g., webcam), the PCP Client SHOULD be instructed to delete the old mapping(s) and create new one(s) using the new assigned internal IP address. When the PCP Client is co-located with the DHCP server (e.g., PCP Proxy [I-D.ietf-pcp-proxy], IWF in the CP router [I-D.ietf-pcp-upnp-igd-interworking]), the state can be updated using the state of the local DHCP server. Otherwise, it is safe to recommend the use of static internal IP addresses if PCP is used to configure third-party explicit dynamic mappings.

2.5. Change of the CPE WAN IP Address

The change of the IP address of the WAN interface of the CPE would have an impact on the accuracy of the explicit dynamic mappings instantiated in the PCP Server:

• For the DS-Lite case [RFC6333]: if a new IPv6 address is used by the B4 element when encapsulating IPv4 packets in IPv6 ones, the explicit dynamic mappings SHOULD be refreshed: If the PCP Client is embedded in the B4, the refresh operation is triggered by the change of the B4 IPv6 address. This would be more complicated when the PCP Client is located in a device behind the B4. If a PCP Proxy is embedded in the CPE, the proxy can use ANNOUNCE OpCode towards internal IPv4 hosts behind the DS-Lite CPE.
• For the NAT64 case [RFC6146], any change of the assigned IPv6 prefix delegated to the CPE will be detected by the PCP Client (because this leads to the allocation of a new IPv6 address). The PCP Client has to undertake the operation described in Section 2.4.
• For the NAT444 case, similar problems are encountered because the PCP Client has no reasonable way to detect the CPE's WAN address changed.

2.6. UPnP IGD/PCP IWF

In the event an UPnP IGD/PCP IWF [I-D.ietf-pcp-upnp-igd-interworking] fails to renew a mapping, there is no mechanism to inform the UPnP Control Point about this failure.

On the reboot of the IWF, if no mapping table is maintained in a permanent storage, "stale" mappings will be maintained by the PCP Server and per-user quota will be consumed. This is even exacerbated if new mapping nonces are assigned by the IWF. This issue can be soften by synchronizing the mapping table owing to the invocation of the GET OpCode defined in Appendix B. This procedure is supported only if Mapping Nonce validation checks are disabled.

3. Restart or Failure of the PCP Server

This section covers failure scenarios encountered by the PCP Server.

3.1. Basic Rule

In any situation the PCP Server loses all or part of its PCP state, the Epoch value MUST be reset when replying to received requests. Doing so would allow PCP Client to audit its explicit dynamic mapping table.

If the state is not lost, the PCP Server MUST NOT reset the Epoch value returned to requesting PCP Clients.

3.2. Clear PCP Mappings

When a command line or a configuration change is enforced to clear all or a subset of PCP explicit dynamic mappings maintained by the PCP Server, the PCP Server MUST reset its Epoch to zero value.

In order to avoid all PCP Clients to update their explicit dynamic mappings, the PCP Server SHOULD reset the Epoch to zero value only for impacted users.

3.3. State Redundancy is Enabled

When state redundancy is enabled, the state is not lost during failure events. Failures are therefore transparent to requesting PCP Clients. When a backup device takes over, Epoch MUST NOT be reset to zero.

3.4. Cold-Standby without State Redundancy

In this section we assume that a redundancy mechanisms is configured between a primary PCP-controlled device and a backup one but without activating any state synchronization for the PCP-instructed explicit dynamic mappings between the backup and the primary devices.

If the primary PCP-controlled device fails and the backup one takes over, the PCP Server MUST reset the Epoch to zero value. Doing so would allow PCP Clients to detect the loss of states in the PCP Server and proceed to state synchronization.

3.5. Anycast Redundancy Mode

When an anycast-based mode is deployed (i.e., the same IP address is used to reach several PCP Servers) for redundancy reasons, the change of the PCP Server which handles the requests of a given PCP Client won't be detected by that PCP Client.

Tweaking the Epoch (Section 8.5 of [I-D.ietf-pcp-base]) may help to detect the loss of state and therefore to re-create missing explicit dynamic mappings.

3.6. Mapping Repair Procedure

3.6.1. PCP Client Behaviour

[I-D.ietf-pcp-base] defines a procedure for the PCP Server to notify PCP Clients about changes related to the mappings it maintains. Indeed, the PCP Server can send unsolicited ANNOUNCE OpCode or unsolicited MAP/PEER responses. When unsolicited ANNOUNCE is received, the PCP Client proceeds to re-installing its mappings. Unsolicited PCP MAP/PEER responses received from a PCP Server are handled as any normal MAP/PEER response.

3.6.2. PCP Proxy Behaviour

Upon receipt of an unsolicited ANNOUNCE response from a PCP Server, the PCP Proxy proceeds to renewing the mappings and checks whether there are changes compared to a local cache if it is maintained by the PCP Proxy. If no change is detected, no unsolicited ANNOUNCE is generated towards PCP Clients. If a change is detected, the PCP Proxy MUST generate unsolicited ANNOUNCE message(s) to appropriate PCP Clients. If the PCP Proxy does not maintain a local cache for the mappings, unsolicited ANNOUNCE messages are relayed to PCP Clients.

Unsolicited PCP MAP/PEER responses received from a PCP Server are handled as any normal MAP/PEER response. To handle unsolicited PCP MAP/PEER responses, the PCP Proxy is required to maintain a local cache of instantiated mappings in the PCP Server. When this service is supported the state SHOULD be recovered in case of failures using the procedure defined in Appendix B.

Upon change of its external IP address, the PCP Proxy SHOULD renew the mappings it maintained. This can be achieved only if a full state table is maintained by the PCP Proxy.

4. Security Considerations

TBC.

5. IANA Considerations

The following OpCode is requested:

• GET

The following Option code is requested:

• NEXT

The following error codes are requested:

• NONEXIST_MAP
• AMBIGUOUS

6. Acknowledgements

Francis Dupont contributed text to this document. Many thanks to him.

7. References

7.1. Normative References

7.2. Informative References

Appendix A. PCP State Synchronization: Overview

The following sketches the state synchronization logic:

• One element (i.e., PCP Client/host/application, PCP Server, PCP Proxy, PCP IWF) of the chain is REQUIRED to use stable storage
• If the PCP Client (resp., the PCP Server) crashes and restarts it just have to synchronize with the PCP Server (resp., the PCP Client);
• If both crash then one has to use stable storage and we fall back in the previous case as soon as we know which one (the Epoch value gives this information);
• PCP Server -> PCP Client not-disruptive synchronization requires a GET/NEXT mechanism to retrieve the state from the PCP Server; without this mechanism the only way to put the PCP Server in a known state is for the PCP Client to send a delete all request, a clearly disruptive operation.
• PCP Client -> PCP Server synchronization is done by a re-create or refresh of the state. The PCP Client MAY retrieve the PCP Server state in order to prevent stale explicit dynamic mappings.

Appendix B. GET/NEXT Operation

This section defines a new PCP OpCode called GET and its associated Option NEXT.

These PCP Opcode and Option are used by the PCP Client to retrieve an explicit mapping or to walk through the explicit dynamic mapping table maintained by the PCP Server for this subscriber and retrieves a list of explicit dynamic mapping entries it instantiated.

GET can also be used by a NoC to retrieve the list of mappings for a given subscriber.

B.1. OpCode Format

The GET OpCode payload contains a Filter used for explicit dynamic mapping matching: only the explicit dynamic mappings of the subscriber which match the Filter in a request are considered so could be returned in response.

• Implementation Note: Some existing implementations use 98 (0x62) codepoint for GET OpCode, 131 for AMBIGUOUS error code, and 131 (0x83) for NEXT Option.

The layout of GET OpCode is shown in Figure 1.

     0                   1                   2                   3
     0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
    |   Protocol    |                Reserved                       |
    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
    |                                                               |
    :   Filter internal IP address (always 128 bits)                :
    |                                                               |
    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
    |                                                               |
    :   Filter external IP address (always 128 bits)                :
    |                                                               |
    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
    |   Filter internal port        |   Filter external port        |
    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+

Version=1
     0                   1                   2                   3
     0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
    |   Protocol    |          Reserved             |  MORE/END     |
    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
    |                                                               |
    :   Mapping internal IP address (always 128 bits)               :
    |                                                               |
    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
    |                                                               |
    :   Mapping external IP address (always 128 bits)               :
    |                                                               |
    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
    |   Mapping remaining lifetime                                  |
    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
    |   Mapping internal port       |   Mapping external port       |
    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
    |                                                               |
    |                       Mapping Options                         :
    |                                                               |
    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+

Version=2

      0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
     +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
     :                 Mapping Nonce (96 bits)                       :
     +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
     |   Protocol    |          Reserved             |  MORE/END     |
     +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
     |                                                               |
     :   Mapping internal IP address (always 128 bits)               :
     |                                                               |
     +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
     |                                                               |
     :   Mapping external IP address (always 128 bits)               :
     |                                                               |
     +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
     |   Mapping remaining lifetime                                  |
     +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
     |   Mapping internal port       |   Mapping external port       |
     +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
     |                                                               |
     |                       Mapping Options                         :
     |                                                               |
     +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+

        +------+                           +------+
        | PCP  |                           | PCP  |
        |Client|                           |Server|
        +------+                           +------+
           |       (1) PCP GET Request         |
           |           protocol= TCP           |
           | internal-ip-address= 198.51.100.1 |
           |        internal-port= 59864       |
           |         Undefined Locator         |
           |---------------------------------->|
           |                                   |
           |        (2) PCP GET Response       |
           |       error= NONEXIST_MAP         |
           |<----------------------------------|
           |                                   |

        +------+                           +------+
        | PCP  |                           | PCP  |
        |Client|                           |Server|
        +------+                           +------+
           |       (1) PCP GET Request         |
           |           protocol= TCP           |
           | internal-ip-address= 198.51.100.1 |
           |        internal-port= 25655       |
           |         Undefined Locator         |
           |---------------------------------->|
           |                                   |
           |      (2) PCP GET Response         |
           |             END                   |
           |           protocol= TCP           |
           | internal-ip-address= 198.51.100.1 |
           |        internal-port= 25655       |
           |   external-ip-address= 192.0.2.1  |
           |        external-port= 15659       |
           |       remaining-lifetime= 1659    |
           |<----------------------------------|
           |                                   |
           |      (3) PCP MAP4 Request         |
           |           protocol= TCP           |
           | internal-ip-address= 198.51.100.1 |
           |        internal-port= 25655       |
           |   external-ip-address= 192.0.2.1  |
           |        external-port= 15659       |
           |       requested-lifetime= 0       |
           |---------------------------------->| 
           |                                   |

        +------+                           +------+
        | PCP  |                           | PCP  |
        |Client|                           |Server|
        +------+                           +------+
           |       (1) PCP GET Request         |
           | internal-ip-address= 198.51.100.2 |
           |         Undefined Locator         |
           |---------------------------------->|
           |                                   |
           |       (2) PCP GET Response        |
           |               MORE                |
           |           protocol= TCP           |
           | internal-ip-address= 198.51.100.2 |
           |        internal-port= 12354       |
           |   external-ip-address= 192.0.2.1  |
           |        external-port= 32654       |
           |       remaining-lifetime= 3600    |
           |               END                 |
           |           protocol= TCP           |
           | internal-ip-address= 198.51.100.2 |
           |        internal-port= 8596        |
           |   external-ip-address= 192.0.2.1  |
           |        external-port= 25659       |
           |       remaining-lifetime= 6000    |
           |<----------------------------------|
           |                                   |

        +------+                           +------+
        | PCP  |                           | PCP  |
        |Client|                           |Server|
        +------+                           +------+
           |       (1) PCP GET Request         |
           | internal-ip-address= 198.51.100.1 |
           |        internal-port= 25655       |
           |         Undefined Locator         |
           |---------------------------------->|
           |                                   |
           |       (2) PCP GET Response        |
           |               MORE                |
           |           protocol= UDP           |
           | internal-ip-address= 198.51.100.1 |
           |        internal-port= 25655       |
           |   external-ip-address= 192.0.2.1  |
           |        external-port= 15659       |
           |       remaining-lifetime= 1659    |
           |                END                |
           |           protocol= TCP           |
           | internal-ip-address= 198.51.100.1 |
           |        internal-port= 25655       |
           |   external-ip-address= 192.0.2.1  |
           |        external-port= 32652       |
           |       remaining-lifetime= 9000    |
           |<----------------------------------|
           |                                   |