| PCP | T. Reddy |
| Internet-Draft | M. Isomaki |
| Intended status: Standards Track | Nokia |
| Expires: February 24, 2014 | D. Wing |
| Cisco | |
| P. Patil | |
| Cisco | |
| August 23, 2013 |
Optimizing NAT and Firewall Keepalives Using Port Control Protocol (PCP)
draft-ietf-pcp-optimize-keepalives-00
This document describes how Port Control Protocol is useful to reduce NAT and firewall keepalive messages for a variety of applications.
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 February 24, 2014.
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.
Many types of applications need to keep their Network Address Translator (NAT) and Firewall (FW) mappings alive for long periods of time, even when they are otherwise not sending or receiving any traffic. This is typically done by sending periodic keep-alive messages just to prevent the mappings from expiring. As NAT/FW mapping timers may be short and unknown to the endpoint, the frequency of these keep-alives may be high. An IPv4 or IPv6 host can use the Port Control Protocol (PCP)[RFC6877] to flexibly manage the IP address and port mapping information on NATs and FWs to facilitate communications with remote hosts. This document describes how PCP can be used to reduce keep-alive messages for both client-server and peer-to-peer type of communication.
The mechanism described in this document is especially useful in cellular mobile networks, where frequent keep-alive messages make the radio transition between active and power-save states causing signaling congestion. The excessive time spent on the active state due to keep-alives also greatly reduces the battery life of the cellular connected devices such as smartphones or tablets. Requirement #14 in [I-D.binet-v6ops-cellular-host-reqs-rfc3316update] explains that cellular host SHOULD support of PCP as a driver to save battery consumption exacerbated by keepalive messages.
The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in [RFC2119].
This note uses terminology defined in [RFC5245] and [RFC6877] .
PCP can help both client-server and peer-to-peer applications to reduce their keep-alive rate. The relevant applications are the ones that need to keep their NAT/FW mappings alive for long periods of time, for instance to be able to send or receive application messages in both directions at any time.
A typical client-server scenario is depicted in Figure 1. A client, who may reside behind one or multiple layers of NATs/FWs, opens a connection to a globally reachable server, and keeps it open to be able to receive messages from the server at any time. The connection may be a connection-oriented transport protocol such as TCP or SCTP or connection-less transport protocol such as UDP. Protocols operating in this manner include Session Initiation Protocol (SIP) [RFC3261], Extensible Messaging and Presence Protocol (XMPP) [RFC3921], Internet Mail Application Protocol (IMAP) [RFC2177] with its IDLE command, the WebSocket protocol and the various HTTP long-polling protocols. There are also a number of proprietary instant messaging, Voice over IP, e-mail and notification delivery protocols that belong in this category. All of these protocols aim to keep the client-server connection alive for as long as the application is running. When the application has otherwise no traffic to send, specific keep-alive messages are sent periodically to ensure that the NAT/FW state in the middle does not expire. The client can use PCP to keep the required mapping at the NAT/FW and use application keep-alives to keep the state on the Application Server/Peer as mentioned in Section 3.4.
PCP PCP
Client Server __________
+-----------+ +------+ / \ +-----------+
|Application|___| NAT/ |____| Internet |___|Application|
| Client | | FW | | | | Server |
+-----------+ +------+ \__________/ +-----------+
(multiple
layers)
------------> PCP
----------------------------------------->
Application keep-alive
Figure 1: PCP with Client-Server applications
There are also scenarios where the long-term communication association is between two peers, both of whom may reside behind one or more layers of NAT/FW. This is depicted in Figure 2. The initiation of the association may have happened using mechanisms such as Interactive Communications Establishment (ICE), perhaps first triggered by a "signaling" protocol such as SIP or XMPP or RTCWeb. Examples of the peer-to-peer protocols include RTP and RTCWeb data channel. A number of proprietary VoIP or video call or streaming or file transfer protocols also exist in this category. Typically the communication is based on UDP, but TCP or SCTP may be used. Unless there is no traffic flowing otherwise, the peers have to inject periodic keep-alive packets to keep the NAT/FW mappings on both sides of the communication active. Instead of application keep-alives, both peers can use PCP to control the mappings on the NAT/FWs to reduce the keep-alive frequency as explained in Section 3.4.
PCP PCP PCP PCP
Client Server __________ Server Client
+-----------+ +------+ / \ +------+ +-----------+
|Application|___| NAT/ |____| Internet |___| NAT/ |___|Application|
| Peer | | FW | | | | FW | | Peer |
+-----------+ +------+ \__________/ +------+ +-----------+
(multiple (multiple
layers) layers)
------------> PCP PCP <------------
<--------------------------------------------------->
Application keep-alive
Figure 2: PCP with Peer-to-Peer applications
Before an application can reduce its keep-alive rate, it has to make sure it has all of the NATs and Firewalls on its path under control. This means it has to detect the presence of any PCP-unaware NATs and Firewalls on its path. PCP itself is able to detect unexpected NATs between the PCP client and server as depicted in Figure 3. The PCP client includes its own IP address and UDP port within the PCP request. The PCP server compares them to the source IP address and UDP port it sees on the packet. If they differ, there are one or more additional NATs between the PCP client and server, and the server will return an error. Unless the application has some other means to control these PCP unaware NATs, it has to fall back to its default keep-alive mechanism.
PCP PCP PCP
Client Unaware Aware __________
+-----------+ +------+ +------+ / \ +-----------+
|Application|___| NAT/ |___| NAT/ |____| Internet |___|Application|
| Client | | FW | | FW | | | | Server |
+-----------+ +------+ +------+ \__________/ +-----------+
<-----------///---------->
PCP based detection
Figure 3: PCP unaware NAT/FW between PCP client and server
Figure 4 shows a topology where one or more PCP unaware NATs are deployed on the exterior of the PCP capable NAT/FWs. To detect this, the application must have the capability to request from its server or peer what IP and transport address it sees. If those differ from the IP and transport address given to the application by the out most PCP aware NAT/FW, the application can detect that there is at least one more PCP unaware NAT on the path. In this case, the application has to fall back to its default keep-alive mechanism.
PCP PCP PCP
Client Aware Unaware __________
+-----------+ +------+ +------+ / \ +-----------+
|Application|___| NAT/ |___| NAT |____| Internet |___|Application|
| Client | | FW | | | | | | Server |
+-----------+ +------+ +------+ \__________/ +-----------+
<------------>
PCP
<---------------------///--------------------------->
Application based detection
Figure 4: PCP unaware NAT external to the last PCP aware NAT
Section 5 describes how the detection works in a number of real application protocols.
The caveat is that Firewalls can not be detected this way. The client will have to use the alternative procedure explained in Section 3.3 to detect PCP unaware Firewalls.
The client sends a STUN Binding Request to the STUN server. STUN server will return its alternate IP address and alternate port in OTHER-ADDRESS in the binding response [RFC5780]. The client then sends MAP request with FILTER option to PCP server to permit STUN server to reach the client using the STUN servers alternate IP address and alternate port. The client then sends a binding request to the primary address of the STUN server with the CHANGE-REQUEST attribute set to change-port and change-IP. This will cause the server to send its response from its alternate IP address and alternate port. If the client receives a response then the client is aware that on path Firewall devices are PCP aware. If the client does not receive a response then the client is aware that could be one or more on path PCP unaware Firewall devices. PCP client will perform the tests separately for each transport protocol. If no response is received, the client will then repeat the test atmost three times for connectionless transport protocols.
If the STUN server does not support OTHER-ADDRESS then this test cannot be run. This procedure can be adopted by other protocols to detect PCP unaware Firewalls.
If the application determines that all NATs and Firewalls on its path to the Internet support PCP, it can start using PCP instead of its default keep-alives to maintain the NAT/FW state. It can use PCP PEER Request with the Requested Lifetime set to an appropriate value. The application may still send some application-specific heartbeat messages end-to-end.
Processing the lifetime value of the PEER Opcode is described in Sections 10.3 and 15 of [RFC6877]. Sending a PEER request with a very short Requested Lifetime can be used to query the lifetime of an existing mapping. PCP recommends that lifetimes of mapping created or lengthened with PEER be longer than the lifetimes of implicitly-created NAT and Firewall mappings. Thus PCP can be used to save battery consumption by making PCP PEER message interval longer than what the application would normally use the keep middle box state alive, and strictly shorter than the server state refresh interval.
If PCP unaware NAT/Firewall is detected then a client can use the following heuristics method to determine the keepalive interval :
The above procedure will be done separately for each transport protocol. For connectionless transport protocols like UDP if timer of 2 seconds elapses without response from the STUN server then the client will repeat step 4 atmost three times to handle packet loss.
This procedure can be adopted by other protocols to use Primary and Secondary channels, so that the client can determine the keepalive interval to refresh FW/NAT mapping. This procedure only serves as a guideline and if applications already use some other heuristics method to determine keepalive, they can continue with the existing logic. For example Teredo determines Refresh interval using the procedure in "Optional Refresh Interval Determination Procedure" (Section 5.2.7 of [RFC4380]).
To improve reliability, applications SHOULD continue to use PCP to lengthen the FW/NAT mappings even if the above described mechanism is used to detect PCP unaware NAT/Firewall. This ensures that PCP aware FW/NAT do not close old mappings with no packet exchange when there is a resource-crunch situation.
This section describes how PCP is used with specific application protocols.
For connection-less transports the User Agent (UA) sends a STUN Binding Request over the SIP flow as described in section 4.4.2 of [RFC5626]. The UA then learns the External IP Address and Port using a PEER request/response. If the XOR-MAPPED-ADDRESS in the STUN Binding Response matches the external address and port provided by PCP PEER response then the UA optimizes the keepalive traffic as described in Section 3.4. There is no further need to send STUN Binding Requests over the SIP flow to keep the NAT binding alive.
If the XOR-MAPPED-ADDRESS in the STUN Binding Response does not match the external address and port provided by the PCP PEER response then PCP will not be used to keep the NAT bindings alive for the flow that is being used for the SIP traffic. This means that multiple layers of NAT are involved and intermediate NATs are not PCP aware. In this case the UA will continue to use the technique in section 4.4.2 of [RFC5626].
For connection-oriented transports, the UA sends a STUN Binding Request multiplexed with SIP over the TCP connection. STUN multiplexed with other data over a TCP or TLS-over-TCP connection is explained in section 7.2.2 of [RFC5389]. The UA then learns the External IP address and port using a PEER request/response. If the XOR-MAPPED-ADDRESS in the STUN Binding Response matches the external address and port provided by PCP PEER response then the UA optimizes the keepalive traffic as described in Section 3.4.
If the XOR-MAPPED-ADDRESS in the STUN Binding Response does not match the external address and port provided by PCP PEER response then PCP will not be used to keep the NAT bindings alive. In this case the UA performs a keep-alive check by sending a double-CRLF (the "ping") then waits to receive a single CRLF (the "pong") using the technique in section 4.4.1 of [RFC5626].
Web Applications that require persistent connections use techniques such as HTTP long polling and Websockets for session keep alive as explained in section 3.1 of [I-D.isomaki-rtcweb-mobile]. In such scenarios, after the client establishes a connection with the HTTP server, it can execute server side scripts such as PHP residing on the server to provide the transport address and port of the HTTP client seen at the HTTP server. In addition, the HTTP client also learns the external IP Address and port using the PCP PEER request/response.
If the IP address and port learned from the server matches the external address and port provided by PCP PEER response then the HTTP client optimizes keepalive traffic as described in Section 3.4.
If the IP address and port do not match then PCP will not be used to keep the NAT bindings alive for the flow that is being used for the HTTP traffic. This means that there are NATs or HTTP proxies between the PCP server and the HTTP server. The HTTP client will have to resort to use existing techniques for keep alive. Please see Appendix A for an example server side PHP script to obtain the client source IP address.
HTTP protocol allows intermediaries like transparent proxies to be involved and there is no way for the client to know that request/response is relayed through a proxy.
The ICE agent learns the External IP Addresss and Port using the MAP opcode. This candidate learnt through PCP is encoded in the ICE offer and answer just like the server reflexive candidate, If the server reflexive candidate and External IP address learnt using PCP are different. When using the Recommended Formula in section 4.1.2.1 of [RFC5245] to compute priority for the candidates learnt through PCP, the ICE agent should use a preference value greater than the server reflexive candidate and hence they are tested before the server reflexive candidates. The recommended type preference value is 105 for candidates discovered using PCP and is explained in section 4.2 of [RFC6544].
The ICE agent in addition to ICE connectivity checks and performs the following :
The ICE agent checks if the XOR-MAPPED-ADDRESS from the STUN [RFC5389] Binding response received as part of ICE connectivity check matches the External IP address and Port provided by PCP MAP response.
Some network operators deploying a PCP Server may allow PEER but not MAP. In such cases the ICE agent learns the external IP address and port using a STUN binding request/response during ICE connectivity checks. The ICE agent also learns the external IP Address and port using a PCP PEER request/response. If the IP address and port learned from the STUN binding response matches the external address and port provided by the PCP PEER response then the ICE agent optimizes keepalive traffic as described in Section 3.4.
Using the Rapid Recovery technique in section 14 of [RFC6877] PCP client upon receiving a PCP ANNOUNCE from a PCP server becomes aware that PCP server has rebooted or lost its mapping state. The PCP client issues new PCP requests to recreate any lost mapping state and thus reconstructs lost mappings fast enough that existing media, HTTP and SIP flows do not break. If the NAT state cannot be recovered the endpoint will find the new external address and port as part of the Rapid Recovery technique in PCP itself and reestablish a connection with the peer.
PCP allows applications to communicate with Firewall devices with PCP functionality to create mappings for incoming connections. In such cases PCP can be used by the endpoint to create an explicit mapping on Firewall to permit inbound traffic and further use PCP to send keep-alives to keep the Firewall mappings alive.
For scenarios where the client uses ICE Lite implementation explained in section 2.7 of [RFC5245], the ICE Lite endpoint will not generate its own ICE connectivity checks, by definition. As part of the call setup, the ICE Lite endpoint would gather its host candidates and relayed candidate from a TURN server, send the candidates in the offer to the peer endpoint. On receiving the answer from the peer endpoint, the ICE Lite endpoint sends a PCP MAP request with FILTER opcode to create a dynamic mapping in Firewall to permit ICE connectivity checks and subsequent media traffic from the remote peer. In this way, the ICE Lite endpoint and its network are protected from unsolicited incoming UDP traffic, and can still operate using ICE Lite (rather than full ICE).
Mobile Networks are also making use of a Firewall to protect their customers from various attacks like downloading malicious content. The Firewall is usually configured to block all unknown inbound connections as explained in section 2.1 of [I-D.chen-pcp-mobile-deployment]. In such cases PCP can be used by Mobile devices to create an explicit mapping on the Firewall to permit inbound traffic and optimize the keepalive traffic as described in Section 3.4. This would result in saving of radio and power consumption of the Mobile device while protecting it from attacks.
None
The security considerations in [RFC5245]and [RFC6877] apply to this use.
Authors would like to thank Dave Thaler, Basavaraj Patil for valuable inputs to the document.
[Note to RFC Editor: Please remove this section prior to publication.]
<html>
Connected to <?PHP echo gethostname(); ?> on port <?PHP echo
getenv(SERVER_PORT) ?> on <?PHP echo date("d-M-Y H:i:s"); ?> Pacific Time
<p>
Your IP address is: <?PHP echo getenv(REMOTE_ADDR); ?>,
port <?PHP echo getenv(REMOTE_PORT); ?>
</p>;
</html>