Network Working Group P. Srisuresh INTERNET-DRAFT Jasmine Networks Expires as of November 15, 2001 J. Kuthan GMD Fokus J. Rosenberg Dynamicsoft A. Molitor Aravox Technologies A. Rayhan Consultant May, 2001 Middlebox Communication Architecture and framework Status of this Memo This document is an Internet-Draft and is in full conformance with all provisions of Section 10 of RFC2026. 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. Abstract There are a variety of intermediate devices in the Internet today that require application intelligence for their operation. Datagrams pertaining to real-time streaming applications such as SIP and H.323 and peer-to-peer applications such as Napster and NetMeeting cannot be identified by merely examining packet headers. Middleboxes implementing Firewall and Network Address Translator services typically embed application intelligence Srisuresh, et al. [Page 1] Internet-Draft MIDCOM Architecture & Framework May 2001 within the device for their operation. The document specifies an architecture and framework in which trusted third parties can be delegated to assist the middleboxes to perform their operation without resorting to embedding application intelligence. Doing this will allow a middlebox to continue to provide the services, while keeping the middlebox application agnostic. A principal objective of this document is to enable complex applications through the middleboxes seamlessly using a trusted third party. 1. Introduction Intermediate devices requiring application intelligence are the subject of this document. These devices are referred as middleboxes throughout the document. Many of these devices enforce application specific policy based functions such as packet filtering, differentiated Quality of Service, tunneling, Intrusion detection, security and so forth. Network Address Translator service, on the other hand, provides routing transparency across address realms (within IPv4 routing network or across V4 and V6 routing realms). Application Level gateways (ALGs) are used in conjunction with NAT to provide end-to-end transparency for many of the applications. There may be other types of services requiring embedding application intelligence in middleboxes for their operation. The discussion scope of this document is however limited to middleboxes implementing Firewall and NAT services only. Nonetheless, the middlebox framework is designed to be extensible to support the deployment of new services. Tight coupling of application intelligence with middleboxes makes maintenance of middleboxes hard with the advent of new applications. Built-in application awareness typically requires updates of operating systems with new applications or newer versions of existing applications. Operators requiring support for newer applications will not be able to use third party software/hardware specific to the application and are at the mercy of their middlebox vendor to make the necessary upgrade. Further, embedding intelligence for a large number of application protocols within the same middlebox increases complexity of the middlebox and is likely to be error prone and degrade in performance. This document describes a framework in which application intelligence can be moved from middleboxes into external MIDCOM agents. The premise of the framework is to devise a MIDCOM protocol that is application independent, so the middleboxes can stay focussed on services such firewall and NAT. MIDCOM agents with application intelligence can, in turn, assist the middleboxes through the MIDCOM protocol in permitting applications Srisuresh, et al. [Page 2] Internet-Draft MIDCOM Architecture & Framework May 2001 such as FTP, SIP and H.323. The communication between a MIDCOM agent and a middlebox will be transparent to the end-hosts that take part in the application, unless one of the end-hosts assumes the role of a MIDCOM agent. Discovery of middleboxes in the path of an application instance and communication amongst middleboxes is outside the scope of this document. This document describes the framework in which middlebox communication takes place and the various elements that constitute the framework. Section 2 describes the terms used in the document. Section 3 defines the architectural framework of a middlebox for communication with MIDCOM agents. The remaining sections cover the components of the framework, illustration using sample flows and operational considerations with the MIDCOM architecture. Section 4 describes the nature of MIDCOM protocol. Section 5 identifies entities that could potentially host the MIDCOM agent function. Section 6 considers the role of Policy server and its function with regard to communicating MIDCOM agent authorization policies. Sections 7 and 8 are illustration of MIDCOM framework with sample flows using In-Path and out-of-path agents respectively. Section 9 addresses operational considerations in deploying a protocol adhering to the framework described here. Section 10 is an applicability statement, scoping the location of middleboxes. Section 12 outlines security considerations for the middlebox in view of the MIDCOM framework. 2. Terminology The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALLNOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in RFC 2119. Below are the definitions for the terms used throughout the document. 2.1. MiddleBox function/service A middlebox function or a middlebox service is an operation or method performed on a network intermediary that requires application specific intelligence for its operation. Policy based packet filtering (a.k.a. firewall), Network address translation (NAT), Intrusion detection, Load balancing, Policy based tunneling and IPsec security are all examples of a middlebox function (or service). 2.2. MiddleBox Middlebox is a network intermediate device that implements one or Srisuresh, et al. [Page 3] Internet-Draft MIDCOM Architecture & Framework May 2001 more of the middlebox services. A NAT middlebox is a middlebox implementing NAT service. A firewall middlebox is a middlebox implementing firewall service. 2.3. Firewall Firewall is a policy based packet filtering Middlebox function, typically used for restricting access to/from specific devices and applications. The policies are often termed Access Control Lists (ACLs). 2.4. NAT Network Address Translation is a method by which IP addresses are mapped from one address realm to another, providing transparent routing to end-hosts. This is achieved by modifying end node addresses en-route and maintaining state for these updates so that datagrams pertaining to a session are forwarded to the right end-host in either realm. Refer [NAT-TERM] for the definition of various NAT types and the associated terms in use. The term NAT in this document is very similar to the IPv4 NAT described in [NAT-TERM], but is extended beyond IPv4 networks to include the IPv4-v6 NAT-PT described in [NAT-PT]. While the IPv4 NAT [NAT-TERM] translates one IPv4 address into another IPv4 address to provide routing between private V4 and external V4 address realms, IPv4-v6 NAT-PT [NAT-PT] translates an IPv4 address into an IPv6 address and vice versa to provide routing between a V6 address realm and an external V4 address realm. Unless specified otherwise, NAT in this document is a middlebox function referring to both IPv4 NAT as well as IPv4-v6 NAT-PT. 2.5. Proxy A proxy is an intermediate relay agent between clients and servers of an application, relaying application messages between the two. Proxies use special protocol mechanisms to communicate with proxy clients and relay client data to servers and vice versa. A Proxy terminates sessions with both the client and the server, acting as server to the end-host client and as client to the end-host server. Applications such as FTP, SIP and RTSP use a control connection to establish data sessions. These control and data sessions can take divergent paths. While a proxy can intercept both the control and data connections, it might intercept only the control connection. This is often the case with real-time streaming applications such as SIP and RTSP. Srisuresh, et al. [Page 4] Internet-Draft MIDCOM Architecture & Framework May 2001 2.6. ALG Application Level Gateways (ALGs) are agents that possess the application specific intelligence and knowledge of an associated middlebox function. An ALG examines application traffic in transit and assists middlebox in carrying out its function. An ALG may be co-resident with a middlebox or reside externally, communicating through a middlebox communication protocol. It interacts with a middlebox to set up state, access control filters, use middlebox state information, modify application specific payload or perform whatever else is necessary to enable the application to run through the middlebox. ALGs are different from proxies. ALGs are transparent to end-hosts, unlike the proxies which are relay agents terminating sessions with both end-hosts. ALGs do not terminate session with either end-host. Instead, ALGs examine and optionally modify application payload content to facilitate the flow of application traffic through a middlebox. ALGs are middlebox centric, in that they assist the middleboxes in carrying out their function. Whereas, the proxies act as focal point for application servers, relaying traffic between application clients and servers. ALGs are similar to Proxies, in that, both ALGs and proxies facilitate Application specific communication between clients and servers. 2.7. End-Hosts End-hosts are entities that are party to a networked application instance. End-hosts referred in this document are specifically those terminating Real-time streaming Voice-over-IP applications such as SIP and H.323 and peer-to-peer applications such as Napster and NetMeeting. 2.8. MIDCOM Agents MIDCOM agents are entities performing ALG function, logically external to a middlebox. MIDCOM agents possess a combination of application awareness and knowledge of the middlebox function. A MIDCOM agent may communicate with one or more middleboxes. MIDCOM agents may be located either In-Path or Out-of-path of an application instance. In-Path MIDCOM agents are those in which the MIDCOM agent function is co-resident on devices that are naturally within the message path of the application they Srisuresh, et al. [Page 5] Internet-Draft MIDCOM Architecture & Framework May 2001 are associated with. This may be an application proxy, gateway, or in the extreme case, one of the end-hosts, that is party to the application. Out-of-Path MIDCOM agents are those that are not necessarily resident (or co-resident) on entities that are in the path of application flows. 2.9. Policy Server Policy Server is a management entity that interfaces with a middlebox to communicate policies concerning authorization of MIDCOM agents gaining access to middlebox resources. A MIDCOM agent may be pre-configured on a middlebox as a trusted entity. In the case where a MIDCOM agent is not pre-configured, the middlebox will consult Policy Server Out-of-band for validating the authorization to accept requests from the agent. A policy server might add or delete MIDCOM agents on a middlebox. The protocol facilitating the communication between a middlebox and Policy Server need not be part of MIDCOM protocol. 2.10. Middlebox Communication (MIDCOM) protocol The protocol between a MIDCOM agent and a middlebox that allows the MIDCOM agent to gain access to middlebox resources and allows the middlebox to delegate application specific processing to MIDCOM agent. The MIDCOM protocol allows the middlebox to perform its operation with the aid of MIDCOM agents, without resorting to embedding application intelligence. The principal motivation behind architecting this protocol is to enable complex applications through middleboxes seamlessly using a trusted third party, i.e., a MIDCOM agent. This is a protocol yet to be devised. 3.0 Architectural framework for Middleboxes A middlebox may implement one or more of the middlebox functions selectively on multiple interfaces of the device. There can be a variety of MIDCOM agents interfacing with the middlebox to communicate with one or more of the middlebox functions on an interface. As such, the Middlebox communication protocol MUST allow for selective communication between a specific MIDCOM agent and one or more middlebox functions on the interface. The following diagram identifies a possible layering of the service supported by a middlebox and a list of MIDCOM agents that might interact with it. Srisuresh, et al. [Page 6] Internet-Draft MIDCOM Architecture & Framework May 2001 +---------------+ +---------------+ +-------------+ | MIDCOM agent | | MIDCOM agent | | Stand-alone | | co-resident on| | co-resident on| | MIDCOM agent| | Proxy Server | | Application GW| | | +---------------+ +---------------+ +-------------+ ^ ^ ^ | | | +--------+ | | | | Policy | | | | +-| Server | | | | / +--------+ | | MIDCOM | /\/ +-------------+ | | Protocol | / +-------------+ | MIDCOM agent| | | | / | MIDCOM agent| | co-resident | | | | / | co-resident | | on End-hosts|<-+ | | | / +->| on End-hosts| +-------------+ | | | | | | +-------------+ v v v v v v +-------------------------------------------+ | Middlebox Communication Protocol (MIDCOM) | | Interface | +----------+--------+-----------+-----------+ Middlebox | | | | | Functions | Firewall | NAT | DiffServ- | Intrusion | | | | QOS | Detection | +----------+--------+-----------+-----------+ Middlebox | Firewall ACLs, Session-descriptors, | Managed | NAT-BINDs, NAT Address-Maps and other | Resources | other Middlebox function attributes | +-------------------------------------------+ Figure 1: MIDCOM agents interfacing with a middlebox Resources such as a Session-Descriptor may be shared across middlebox functions. A Session-Descriptor may uniquely identify a session denoted by the tuple of (SessionDirection, SourceAddress, DestinationAddress, Protocol, SourcePort, DestinationPort). An aggregated Session-Descriptor, on the other hand, may have one of the tuple elements denoted by a regular expression (ex: Any source port). The attributes associated with a Session-Descriptor may be specific to the individual middlebox function. As Session-Descriptors may be shared across middlebox functions, a Session-Descriptor may be created by a function, and terminated by a different function. For example, a session-descriptor may be created by the firewall function, but terminated by the NAT function, when a session timer expires. A middlebox may also have function specific resources such as Srisuresh, et al. [Page 7] Internet-Draft MIDCOM Architecture & Framework May 2001 Address maps and Address binds to enforce NAT function and application based policies to enforce firewall function. Application specific MIDCOM agents (co-resident on the middlebox or external to the middlebox) would examine the IP datagrams and help identify the application the datagram belongs to and assist the middlebox in performing functions unique to the application and the middlebox service. For example, a MIDCOM agent assisting a NAT middlebox might perform payload translations; whereas a MIDCOM agent assisting a firewall middlebox might request the firewall to permit access to application specific dynamically generated session traffic. 4. MIDCOM Protocol The MIDCOM protocol between a MIDCOM agent and a middlebox allows the MIDCOM agent to gain access to middlebox resources and allows the middlebox to delegate application specific processing to MIDCOM agent. The protocol will allow MIDCOM agents to signal the middleboxes to let complex applications using dynamic port based sessions through them (i.e., middleboxes) seamlessly. It is important to note that an agent and a middlebox can be on the same physical device. In such a case, it is not desirable for them to communicte using MIDCOM protocol. They may communicate using a MIDCOM protocol message formats, but using a non-IP based transport such as IPC messaging (or) they may communicate using a well-defined API/DLL (or) the application intelligence is fully embedded into the middlebox service (as it is done today in many stateful inspection firewall devices and NAT devices). The MIDCOM protocol will consist of a connection setup phase, run-time connection phase and a connection termination phase. Connection setup must be preceded by registration of the MIDCOM agent with the middlebox. The MIDCOM agent access and authorization profile may either be pre-configured on the middlebox (or) listed on a Policy Server the middlebox is configured to consult. MIDCOM is a peer-to-peer protocol. As such, either the agent or the middlebox may choose to initiate the connection. A MIDCOM session may be terminated by either of the parties. Alternately, a MIDCOM session termination may be triggered by one of (a) agent going out of service and not being available for further MIDCOM operations, or (b) a policy server notifying the middlebox that a particular MIDCOM agent is no longer authorized for a particular set of sessions any longer. Srisuresh, et al. [Page 8] Internet-Draft MIDCOM Architecture & Framework May 2001 The MIDCOM protocol data exchanged during run-time is governed principally by the middleboxbox services the protocol supports. Firewall and NAT middlebox services are considered in this document. Nonetheless, the MIDCOM protocol will be designed to be extensible to support deployment of other services as well. Few of the middlebox services are stateless. There are many that are stateful and maintain per-connection state in the system. Firewall service may be implemented as a stateless list of ACLs. Many firewall implementations, however, are stateful. NAT service, on the other hand, is inherently stateful. As such, support of the MIDCOM protocol will require a middlebox to be stateful. Here is why. Let us consider the case of a middlebox implementing firewall service. With the advent of MIDCOM protocol, MIDCOM agents ought to be able to set a sessions timer that is launched when a pinhole is opened for a dynamically permitted session. When the session timer reaches it's expiration, the middlebox will notify the MIDCOM agent to refresh it, else the pinhole will be closed. Explicit pinhole closing is done when the application session is ended and Midcom agent will request to close it. Session timer is also required so the pinhole doesnt stay open forever, just in case the MIDCOM agent suddenly diappears (or terminates for whatever reason). This goes to show that firewall function will also necessarily need to maintain per-connection state, as a requirement to support the MIDCOM protocol. 5.0. MIDCOM Agents MIDCOM agents are logical entities which may reside physically on nodes external to a middlebox, possessing a combination of application awareness and knowledge of middlebox function. A MIDCOM agent may communicate with one or more middleboxes. The issues of middleboxes discovering agents or vice versa are outside the scope of this document. The focus of the document is the framework in which a MIDCOM agent communicates with a middlebox using MIDCOM protocol, which is yet to be devised. We will examine two types of MIDCOM agents in the following sub-sections. 5.1. In-path MIDCOM agents In-Path MIDCOM agents are entities that have a native role in the path of the application traversal (with prior knowledge to one of Srisuresh, et al. [Page 9] Internet-Draft MIDCOM Architecture & Framework May 2001 the application end-hosts), independent of their MIDCOM function. Bundled session applications such as H.323, SIP and RTSP which have separate control and data sessions may have their sessions take divergent paths. In those scenarios, In-Path MIDCOM agents are those that find themselves in the control path. In majority of cases, a middlebox will likely require the assistance of a single agent for an application in the control path alone. However, it is possible that a middlebox function might require the intervention of more than a single MIDCOM agent for the same application, one for each sub-session of the application. Application Proxies and gateways are a good choice for In-Path MIDCOM agents, as these entities, by definition, are in the path of an application between a client and server. In addition to hosting the MIDCOM agent function, these natively in-path application specific entities may also enforce application-specific choices locally, such as dropping messages infected with known viruses, or lacking user authentication. These entities can be interjecting both the control and data connections. For example, FTP control and Data sessions are interjected by an FTP proxy server. However, proxies may also be interjecting just the control connection and not the data connections, as is the case with real-time streaming applications such as SIP and RTSP. Note, applications may not always traverse a proxy and some applications may not have a proxy server available. SIP proxies and H.323 gatekeepers may be used to host MIDCOM agent function to control middleboxes implementing firewall and NAT functions. The advantage of using in-path entities as opposed to creating an entirely new agent is that the in-path entities already possess application intelligence. You will need to merely enable the use of MIDCOM protocol to be an effective MIDCOM agent. Figure 2 below illustrates a scenario where the in-path MIDCOM agents interface with the middlebox. Let us say, the policy Server has pre-configured the in-path proxies as trusted MIDCOM agents on the middlebox and the packet filter implements 'default-deny' packet filtering policy. Proxies use their application-awareness knowledge to control the firewall function and selectively permit a certain number of voice stream sessions dynamically using MIDCOM protocol. In the illustration below, the proxiess and the policy server are shown inside a private domain. The intent however is not to imply that they be inside the private boundary alone. The proxies may also reside external to the domain. The only requirement is that there be a trust relationship with the middlebox. Srisuresh, et al. [Page 10] Internet-Draft MIDCOM Architecture & Framework May 2001 +-----------+ | Middlebox | | Policy | | Server |~~~~~~~~~~~~| +-----------+ \ \ +---------+ \ | SIP |___ \ ________| Proxy | \ \ Middlebox / +---------+.. | +-----------------+ | : | MIDCOM | | | RSTP +----------+ :..|........| MIDCOM | SIP | ____| RSTP |.....|........| PROTOCOL | | / | Proxy |___ | | INTERFACE | | | +----------+ \ \ |-----------------| | | \ \------| | | | \--------| | | | -----| FIREWALL |-->-- +-----------+ /-----| |--<-- +-----------+| Data streams // +-----------------+ +-----------+||----------->----// | |end-hosts ||------------<----- . +-----------+ (RTP, RSTP data, etc.) | . Outside the Within a private domain | private domain Legend: ---- Application data path datagrams ____ Application control path datagrams .... Middlebox Communication Protocol (MIDCOM) ~~~~ MIDCOM Policy Server Interface | . private domain Boundary | Figure 2: In-Path MIDCOM Agents for Middlebox Communication 5.1.1. End-hosts as In-Path MIDCOM agents End-hosts are another variation of In-Path MIDCOM agents. Unlike Proxies, End-hosts are direct party to the application and possess all the end-to-end application intelligence there is to it. End-hosts terminate both the control and data paths of an application. Unlike other entities hosting MIDCOM agents, end-host is able to process secure datagrams. However, the problem Srisuresh, et al. [Page 11] Internet-Draft MIDCOM Architecture & Framework May 2001 would be one of manageability - upgrading all the end-hosts running a specific application. 5.2. Out-of-Path MIDCOM agents Out-of-Path MIDCOM agents (a.k.a. OOP agents) are entities that are not natively in the transport path of an application. OOP Agents have a role in the application traversal, only by virtue of their MIDCOM function - No native role otherwise. It would be safe to assume that OOP agents are not in the path of application traversal. Out-of-Path agents have a few benefits. Out-of-Path agents can be implemented in a system, independent of any pre-existing application-aware entity. Unlike In-path agents, there are no topological restrictions to where the agents can be located. Further, multiple application specific agents can be grouped together on the same node. There is however a significant difference between in-path MIDCOM agents and Out-of-path MIDCOM agents in the way the middlebox directs application specific traffic for processing by the registered MIDCOM agents. The MIDCOM protocol provides a means for MIDCOM agents to gain access to middlebox resources and for the middlebox to direct selective application specific traffic (ex: Control path datagrams in the case of bundled session applications) to MIDCOM agents. When a MIDCOM agent is naturally in the transport path of the application (as is the case with an In-Path MIDCOM agent), there is no additional effort required of the middlebox in redirecting the application traffic. However, Out-of-Path MIDCOM agent is not natively in the transport path of an application and hence will need to instruct the middlebox to explicitly redirect datagrams to itself, as appropriate, on a per-session basis. The middlebox must be able to redirect the selective application traffic toward the MIDCOM agent. The out-of-path MIDCOM agent should in turn be capable of returning the processed traffic to the middlebox point of origin or forwarding to the destination. In essence, a middlebox provides to an Out-of-Path MIDCOM agent the ability to transparently "snoop" and modify the control traffic. It is reasonable to further classify Out-of-Path agents into those which modify control traffic, and those which do not. For example, if an Out-of-Path agent is used simply to manage firewall policy for SIP-based telephony, it is enough to simply forward SIP messages to the agent for examination. On the other hand, if the agent must also manage NAT bindings, the agent needs to modify the SIP messages, and re-inject them into Srisuresh, et al. [Page 12] Internet-Draft MIDCOM Architecture & Framework May 2001 the control path. In order to support Out-of-Path agents, the middlebox will require an additional "Datagram diverter" functional component. This function is strictly to support the Out-of-path MIDCOM agents and is independent of any middlebox service or application. The diverter function on the middlebox would be required to do the following. 1. When a datagram is received by the middlebox, the middlebox will subject the datagram to the standard middlebox services as appropriate. However, if the datagram is designated for diversion (by the application specific out-of-path MIDCOM agent), the middlebox will redirect the datagram to the diversion target. The datagram will have been directed to an application specific payload processing entity. As such, this may be accomplished using some type of tunneling mechanism (or) Remote procedure Call (RPC) (or) some other proprietary mechanism. 2. The recipient of the diverted datagram (i.e., the OOP agent) will snoop and optionally modify the payload (as appropriate to the middlebox service) and does one of the following. Of these, the safe thing to assume would be the first option. (a) Send the processed datagram right back to the middlebox using the same diversion approach the middlebox used. (or) (b) Forward the datagram to the appropriate destination (i.e., one of the end-hosts that is party to the application). This however assumes that the OOP agent has routing/forwarding capability. 3. When the middlebox receives a diverted (i.e., co-processed) datagram from the middlebox, the middlebox will simply forward the processed datagram to the appropriate destination (i.e., one of the end-hosts that is party to the application). Note, the middlebox will not subject the datagram to any of the middlebox services (i.e., NAT or firewall) this time around. Note, Step 2a followed by step 3 would be same as going with step 2b by itself. Below is an illustration of a scenario where Out-of-path MIDCOM agents interface with the middlebox. The middlebox is assumed to implement firewall service on it. Let us say, the Out-of-path agents are pre-configured as trusted MIDCOM agents on the middlebox and the packet filter implements 'default-deny' packet filtering policy. The OOP agents register themselves as the diversion traffic targets for the applications Srisuresh, et al. [Page 13] Internet-Draft MIDCOM Architecture & Framework May 2001 they support. They snoop the payload of the diverted traffic and use application-awareness knowledge to control the firewall function and selectively permit a certain number of FTP or voice stream sessions dynamically using the MIDCOM protocol. +---------+ Snooped ftp-control traffic | FTP OOP |============>=====================\ | Agent |++++++++++++<++++++++++++++++++++ || | | Diverted ftp-control traffic + || +---------+ + || +-----------+ : + || | Middlebox | : +----------+ Snooped SIP traffic + || | Policy | : | SIP OOP |=========>==============\ + || | Server | : | Agent |+++++++++<+++++++++++++ || + || +-----------+ : | | Diverted SIP traffic + || + || | : +----------+ +-----------+------------+ | : :.............| | | | : MIDCOM | MIDCOM | MIDDLEBOX | | :....................| PROTOCOL | DATAGRAM | | | INTERFACE | DIVERTER | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~| | | +------------+ +-----------+------------+ +------------+|----------->---------| FIREWALL |->- +------------+||-----------<---------| |-<- |end-hosts || Data & Control +------------------------+ +------------+ (SIP, RTP, ftp-control, | ftp-data, etc.) . | Within a private domain . Outside the | private domain Legend: ---- Application data & control path datagrams .... Middlebox Communication Protocol (MIDCOM) ~~~~ MIDCOM Policy Server Interface ++++ Control traffic diverted To a MIDCOM agent ==== Snooped and optionally modified application specific control traffic returning FROM the Out-of-Path agent | . private domain Boundary | Figure 3: Out-of-Path MIDCOM Agents for Middlebox Communication 6.0. Policy Server functions Srisuresh, et al. [Page 14] Internet-Draft MIDCOM Architecture & Framework May 2001 The functional decomposition of the MIDCOM architecture assumes the existence of a logical entity known as Policy Server, responsible for performing authorization and related provisioning services for the middlebox as depicted in figure 1. For example, a policy server has the ability to add or delete MIDCOM agents on a middlebox and to notify a middlebox about the status and security requirements to allow accessibility to MIDCOM agents. The primary objective of a policy server is to ensure that the security and integrity of a middlebox is not jeopardized. Specifically, the policy server should associate a trust level with each agent attempting to connect to a middlebox and provide a security profile. Since some agents may be less secure than others, the policy server should be able to determine the profile for each agent based on the trust and access security permitted to the agent. The policy server should be capable of addressing cases when end-hosts are agents to the middle-box. 6.1. Authentication, Integrity and Confidentiality Host authenticity and individual message authentication are two distinct types of authentications to consider. Host authentication refers to credentials required of a MIDCOM agent to authenticate itself to the middlebox and vice versa. When authentication fails, the middlebox MUST not process signaling requests received from the agent that failed authentication. To protect MIDCOM messages from being tampered with, individual message authentication may be used [IPsec-AH] in addition to host authentication. Further, message confidentiality may be administered by employing IPsec ESP protocol [IPsec-ESP] for the MIDCOM messages, when a higher level of security is required. Alternatively, there are other security options instead of the IPsec protocols. TLS based transport layer security is one option. There are also many application-layer security mechanisms available. Simple Source-address based security is the least form of security in a trusted domain and should be permitted only to the most trusted hosts. Clearly, the middlebox should be able to perform host level authentication, and be able to authenticate individual messages (using IPsec or TLS based security). 6.2. Registration and deregistration with a middlebox Prior to giving MIDCOM agents access to the middlebox resources, a registration process MUST take place. Registration is a different process than establishing a transport connection. Srisuresh, et al. [Page 15] Internet-Draft MIDCOM Architecture & Framework May 2001 The former requires exchanging the agent profile information to the middlebox. The latter refers to establishing a MIDCOM transport connection and exchanging security credentials adhering to the registered profile. MIDCOM agents, their trust level and accessibility (i.e., the MIDCOM agent profile) may be pre-registered with the middlebox while provisioning the middlebox function. Either the agent or the middlebox can choose to initiate a connection prior to any data traffic. Alternately, either party (middlebox or the MIDCOM agent) may choose to initiate a connection only upon noticing the application specific traffic. Coupling MIDCOM agents with the middlebox resources requires a means of reflecting that into the resource description table of the middlebox. In the case of a firewall, for example, the ACL tuple may me altered to reflect the optional ALG presence. The revised ACL may look something like the following. (, , , , , , ) The reader should note that this is an illustrative example and not necessarily the actual definition of an ACL tuple. The formal description of the ACL is yet to be devised. Agent accessibility information should also be provisioned. For a MIDCOM agent, accessibility information includes the IP address, trust level, host authentication parameters and message authentication parameters. Once a connection is established between a middlebox and a MIDCOM agent, that connection should be usable with multiple instances of the application(s), as appropriate. Note, all of this could be captured in an agent profile for ease of management. The technique described above is necessary for the pre-registration of MIDCOM agents with the middlebox. However, it is possible to retain the provisioning on middlebox unchanged, by requiring MIDCOM agents to initiate the connection to middlebox. When Middlebox notices an incoming MIDCOM connection, and the middlebox has no prior profile of the MIDCOM agent, the middlebox will consult its Policy Server for authenticity, authorization and trust guidelines for the connection. At the end of the MIDCOM session, it should be possible for either the middlebox or the agent to disconnect. MIDCOM session disconnection may be prompted by a successful termination or failure of some sort. Srisuresh, et al. [Page 16] Internet-Draft MIDCOM Architecture & Framework May 2001 It should be possible for the agent to deregister itself from the middlebox, which means that the agent is going out of service and will not be available for further MIDCOM operations. Alternately, a policy server may notify a middlebox that a particular MIDCOM agent is no longer authorized for a particular set of sessions any longer. 7.0. MIDCOM Framework Illustration using an In-Path agent In figure 3 below, we consider SIP application (Refer [SIP]) to illustrate the operation of MIDCOM protocol. Specifically, the application assumes a caller, external to a private domain, initiates the call. Middlebox is assumed to be located at the edge of the private domain. A SIP phone (SIP User Agent Client/Server) inside the private domain is capable of receiving calls from external SIP phones. The caller uses a SIP Proxy node, located external to the private domain, as its outbound proxy. No interior proxy is assumed for the callee. Lastly, the external SIP proxy node is designated to host the MIDCOM agent function. Arrows 1 and 4 in the figure below refer to SIP call setup exchange between the external SIP phone and the SIP proxy. Arrows 6 and 7 refer to SIP call setup exchange between the SIP proxy and the interior SIP phone and are assumed to be traversing the middlebox. Arrows 2 and 3 below between the SIP proxy and the middlebox refer to MIDCOM communication. Na and Nb represent RTP/RTCP media traffic (Refer [RTP]) path in the external network. Nc and Nd represent media traffic inside the private domain. _________ --->| SIP |<-----\ / | Proxy | \ | |_________| | 1| | | 6| | | | | |4 |2 |3 |7 ______________ | | | | _____________ | |<-/ _v_____^___ \->| | | External | Na | | Nc | SIP Phone | | SIP phone |>------->| MiddleBox |>------>| within | | |<-------<|___________|<------<| Pvt. domain| |____________| Nb Nd |____________| Figure 4: MIDCOM framework illustration with In-Path SIP Proxy Srisuresh, et al. [Page 17] Internet-Draft MIDCOM Architecture & Framework May 2001 As for the SIP application, we make the assumption that the middlebox is pre-configured to accept SIP calls into the private SIP phone. Specifically, this would imply that the middlebox implementing firewall service is pre-configured to permit SIP calls (destination TCP or UDP port number set to 5060) into the private phone. Likewise, middlebox implementing NAPT service would have been pre-configured to provide a port binding to permit incoming SIP calls to be redirected to the specific private SIP phone. I.e., the INVITE from the external caller is not made to the private IP address, but to the NAPT external address. The objective of the MIDCOM agent in the following illustration is to merely permit the RTP/RTSP media stream (Refer [RTP], [RTSP]) through the middlebox, using the MIDCOM protocol architecture outlined in the document. RTP/RTSP media stream, When used in conjunction with SIP will typically result in two independent media sessions - one from the caller to the callee and another from the callee to the caller. These media sessions are UDP based and will use dynamic ports. The dynamic ports used for the media stream are specified in the SDP section (Refer [SDP]) of SIP payload message. The MIDCOM agent will parse the SDP section and use the MIDCOM protocol to (a) open pinholes (i.e., permit RTP/RTSP session tuples) in a middlebox implementing firewall service, or (b) create PORT bindings and appropriately modify the SDP content to permit the RTP/RTSP streams through a middlebox implementing NAT service. The MIDCOM protocol should be sufficiently rich and expressive to support the operations described under the timelines. Midcom agent Registration and connectivity between the Midcom agent and the middlebox are not shown in the interest of restricting the focus of the MIDCOM transactions to enabling the middlebox to let the media stream through. Policy server is also not shown in the diagram below or on the timelines for the same reason. The following subsections illustrate a typical timeline sequence of operations that transpire with the various elements involved in a SIP telephony application path. Each subsection is devoted to a specific instantiation of a middlebox service - NAPT (refer [NAT-TERM], [NAT-TRAD]), firewall and a combination of both NAPT and firewall are considered. Srisuresh, et al. [Page 18] Internet-Draft MIDCOM Architecture & Framework May 2001 7.1. Timeline flow - Middlebox implementing firewall service In the following example, we will assume a middlebox implementing a simple, stateless firewall service. We further assume that the middlebox is pre-configured to permit SIP calls (destination TCP or UDP port number set to 5060) into the private phone. The following timeline will illustrate the operations that need to be performed by the MIDCOM agent to permit the RTP/RTSP media stream through the middlebox. SIP Phone SIP Proxy Middlebox SIP Phone (External) (In-Path (FIREWALL (private) MIDCOM agent) Service) | | | | | |----INVITE------>| | | | | | | | |--------INVITE---------------------->| |<---100Trying----| | | | | | | | |<-----180Ringing---------------------| |<--180Ringing----| | | | |<-------200 OK-----------------------| | | | | | Identify end-2-end session| | | parameters for the two | | | RTP/RTCP sessions - | | | Ext-to-Pri(RTP1, RTCP1) & | | | Pri-to-ext(RTP2, RTCP2). | | | | | | | |+Permit RTP1, RTCP1 +>| | | |<+RTP1, RTCP1 OKed++++| | | |+Permit RTP2, RTCP2 +>| | | |<+RTP2, RTCP2 OKed++++| | | | | | |<---200 OK ------| | | |-------ACK------>| | | | |-----------ACK---------------------->| | | | | |<===================RTP/RTCP==========================>| | | | | |-------BYE------>| | | | |--------------------------BYE------->| | | | | | |<----------200 OK--------------------| | | | | | |++Remove permits to | | Srisuresh, et al. [Page 19] Internet-Draft MIDCOM Architecture & Framework May 2001 | | RTP1, RTP2 etc.++++>| | | |<+Removed permits | | | | to RTP1, RTP2 etc.++| | | | | | |<---200 OK-------| | | | | | | Legend: ++++ MIDCOM control traffic ---- SIP control traffic ==== RTP/RCTP media traffic 7.2. Timeline flow - Middlebox implementing NAPT service In the following example, we will assume a middlebox implementing NAPT service. We make the assumption that the middlebox is pre-configured to redirect SIP calls to the specific private SIP phone application. I.e., the INVITE from the external caller is not made to the private IP address, but to the NAPT external address. Let us say, the external phone's IP address is Ea, NAPT middlebox external Address is Ma and the internal SIP phone's private address is Pa. SIP calls to the private SIP phone will arrive as TCP/UDP sessions with destination address and port set to Ea and 5060 respectively. The middlebox will redirect these datagrams to the internal SIP phone. The following timeline will illustrate the operations necessary to be performed by the MIDCOM agent to permit the RTP/RTCP media stream through the middlebox. Unlike firewall, NAT is stateful and strictly session oriented. The reader may refer [NAT-FRAMEWORK] for a detailed discussion of NAT managed stateful resources, including that of NAT session-descriptor and NAT BIND. [NAT-Framework] also has illustration of how an Out-of-path Midcom agent could interface with NAT middlebox to gain access to middelbox resources and request redirecting application specific traffic to the agent. Srisuresh, et al. [Page 20] Internet-Draft MIDCOM Architecture & Framework May 2001 SIP Phone SIP Proxy Middlebox SIP Phone (External) (In-Path (NAPT (Private) IP Addr:Ea MIDCOM agent) Service) IP addr:Pa | | IP addr:Ma | | | | | |----INVITE------>| | | | |++ Query Port-BIND | | | | for (Ma, 5060) +++>| | | |<+ Port-BIND reply | | | | for (Ma, 5060) ++++| | | | | | | Determine the Internal | | | IP address (Pa) of the | | | callee. | | | | |..redirected..| | |--------INVITE--------|------------->| |<---100Trying----| | | | | | | | |<-----180Ringing---------------------| | | | | | |++ Query NAT Session | | | | Descriptor for | | | | Ea-to-Pa SIP flow+>| | | |<+ Ea-to-Pa SIP flow | | | | Session Descriptor+| | | | | | |<--180Ringing----| | | | |<-------200 OK-----------------------| | | | | | Identify end-2-end session| | | parameters for the two | | | RTP/RTCP sessions - | | | Ext-to-Pri(RTP1, RTCP1) & | | | Pri-to-ext(RTP2, RTCP2). | | | Identify UDP port numbers | | | on Pa for RTP1 (Port1) | | | and for RTP2 (Port2) | | | | | | | |++Create port BINDs | | | | for (pa, port1), | | | | (Pa, Port2) ++++++>| | | |<+Port BINDs created++| | | | | | | |++Create NAT Session | | | | descriptors for | | | | RTP1, RTP2 pointing | | | | to SIP session ++++>| | | |<+RTP1, RTP2 session | | Srisuresh, et al. [Page 21] Internet-Draft MIDCOM Architecture & Framework May 2001 | | descriptors created+| | | | | | | Modify the SDP | | | parameters in "200 OK" | | | with NAPT PORT-BINDs | | | for port1 & port2. | | | | | | |<---200 OK ------| | | | | | | |-------ACK------>| | | | | | | | Modify the SDP payload | | | parameters in "ACK" | | | | |..redirected..| | |-----------ACK--------|------------->| | | | | | | | | |<===================RTP/RTCP============|=============>| | | | | |-------BYE------>| | | | | | | | Modify SDP payload | | | parameters in BYE | | | | | | | |----------------------|-----BYE----->| | | | | | |<----------200 OK--------------------| | | | | | |+++Terminate the SIP | | | | Session bundle +++>| | | |<++SIP Session bundle | | | | terminated ++++++++| | | | | | | Modify SDP | | | parameters in "200 OK" | | | | | | |<---200 OK-------| | | | | | | Legend: ++++ MIDCOM control traffic ---- SIP control traffic ==== RTP/RTCP media traffic 7.3. Timeline flow - Middlebox implementing NAPT and firewall Srisuresh, et al. [Page 22] Internet-Draft MIDCOM Architecture & Framework May 2001 In the following example, we will assume a middlebox implementing a combination of a stateless firewall and a stateful NAPT service. We make the assumption that the middlebox is configured to translate the IP and TCP headers of the initial SIP session into the private SIP phone and, the firewall is configured to permit the initial SIP session. In the following time line, it may be noted that the firewall description is based on packet fields on the wire (ex: as seen on the external interface of the middlebox). In order to ensure correct behavior of the individual services, you will notice that NAT specific MIDCOM operations precede firewall specific operations on the MIDCOM agent. This is noticeable in the time line below when the MIDCOM agent processes the "200 OK" from the private SIP phone. The MIDCOM agent initially requests the NAT service on the middlebox to set up port-BIND and session-descriptors for the media stream in both directions. Subsequent to that, the MIDCOM agent determines the session parameters (i.e, the dynamic UDP ports) for the media stream, as viewed by the external interface and requests the firewall service on the middlebox to permit those sessions through. SIP Phone SIP Proxy Middlebox SIP Phone (External) (In-Path (NAPT & (Private) IP Addr:Ea MIDCOM agent) firewall IP addr:Pa | | Services) | | | IP addr:Ma | | | | | |----INVITE------>| | | | |++ Query Port-BIND | | | | for (Ma, 5060) +++>| | | |<+ Port-BIND reply | | | | for (Ma, 5060) ++++| | | | | | | Determine the Internal | | | IP address (Pa) of the | | | callee. | | | | |..redirected..| | |--------INVITE--------|------------->| |<---100Trying----| | | | | | | | |<-----180Ringing---------------------| | | | | | |++ Query NAT Session | | | | Descriptor for | | | | Ea-to-Pa SIP flow+>| | Srisuresh, et al. [Page 23] Internet-Draft MIDCOM Architecture & Framework May 2001 | |<+ Ea-to-Pa SIP flow | | | | Session Descriptor+| | | | | | |<--180Ringing----| | | | |<-------200 OK-----------------------| | | | | | Identify end-2-end session| | | parameters for the two | | | RTP/RTCP sessions - | | | Ext-to-Pri(RTP1) | | | and Pri-to-ext(RTP2). | | | Identify UDP port numbers | | | on Pa for RTP1 (Port1) | | | and for RTP2 (Port2) | | | | | | | |+Create NAT port-BINDs| | | | for (Pa, Port1), | | | | (Pa, Port2) ++++++++>| | | |<+Port BINDs created++| | | | | | | |++Create NAT Session | | | | descriptors for | | | | RTP1, RTP2;Set their| | | | parent session to | | | | SIP ctrl session ++>| | | |<+RTP1, RTP2 session | | | | descriptors created+| | | | | | | Extract session parameters| | | for the two media streams | | | as viewed on the NAPT | | | external interface. Say, | | | these are F-RTP1 and | | | F-RTP2, reflecting RTP1 | | | and RTP2 respectively. | | | | | | | |++Permit F-RTP1, | | | | F-RTP2 sessions +++>| | | |<+F-RTP1,F-RTP2 OKed++| | | | | | | Modify the SDP | | | parameters in "200 OK" | | | with NAPT PORT-BINDs | | | for port1 & port2. | | | | | | |<---200 OK ------| | | | | | | |-------ACK------>| | | Srisuresh, et al. [Page 24] Internet-Draft MIDCOM Architecture & Framework May 2001 | | | | | | |..redirected..| | |-----------ACK--------|------------->| | | | | | | | | |<===================RTP/RTCP============|=============>| | | | | |-------BYE------>| | | | | | | | Modify SDP payload | | | parameters in BYE | | | | | | | |----------------------|-----BYE----->| | | | | | |<----------200 OK--------------------| | | | | | |+++Terminate the SIP | | | | Session bundle +++>| | | |<++SIP Session bundle | | | | terminated ++++++++| | | | | | | |++Remove permits to | | | | F-RTP1, F-RTP2 ++++>| | | |<+Removed permits | | | | to F-RTP1, F-RTP2+++| | | | | | | Modify SDP | | | parameters in "200 OK" | | | | | | |<---200 OK-------| | | | | | | Legend: ++++ MIDCOM control traffic ---- SIP control traffic ==== RTP/RCTP media traffic 8.0. MIDCOM framework illustration with an Out-Of-path FTP Agent In the following figure, an FTP client inside a private domain connects via a middlebox to an external FTP server. The middlebox is assumed to implement NAPT and firewall functions. The FTP traffic is addressed directly to the external FTP server. The Arrow labeled 1 indicates a registration via the MIDCOM protocol in which the Out-of-Path FTP agent indicates that it would like to receive TCP traffic directed to or from port 21 (FTP control). The OOP agent may be located either inside the private domain or external to the domain. Srisuresh, et al. [Page 25] Internet-Draft MIDCOM Architecture & Framework May 2001 The FTP control traffic traversing the middlebox is diverted by the middlebox to the Out-of-Path FTP agent for FTP control payload processing. Diverted control traffic is indicated by Arrow 2. The OOP agent parses the FTP control commands and responses and possibly modifies, as appropriate and forwards the traffic over to the server and/or the client. Neither of the end-hosts is aware of the OOP Agent or the middlebox in transit. At some point, the Client sends a PORT command to the Server, indicating that the Server should create a TCP connection from the Server to the Client. This port command specifies an IP address and port number to which the Server should connect. The IP address may be a private IP address, if the client is located in a privately addressed domain. The OOP agent parses the PORT command, and carries out appropriate MIDCOM transactions (Arrow 4) to discover any changes to the IP address required, to request a new NAPT port binding if necessary, and to open a suitable pinhole allowing the connection from the Server to the dynamically allocated port number on the Client to succeed. The (perhaps modified) PORT command is then sent on to the Server, which responds by connecting to the indicated IP address and port, which will now flow through the middlebox to the Client. --------------- | Out-of-Path | | (OOP) FTP | | Agent | |_____________| | ^ | | | | | | |1 |2 |3 |4 ______________ | | | | _____________ | | _v__|__v__v_ | | | FTP client | Ctrl | | Ctrl | External | | within the |<------->| MiddleBox |<------>| FTP Server | | Pvt. domain|<------->|___________|<------>| | |____________| Data Data |____________| Ctrl - indicates the FTP control traffic, which is transparently diverted to the OOP agent (2 and 3) Data - indicates the FTP data traffic, which flows directly through the middlebox between the FTP end hosts (i.e., FTP client and Server) Figure 5: MIDCOM framework illustration Out-of-Path FTP agent Srisuresh, et al. [Page 26] Internet-Draft MIDCOM Architecture & Framework May 2001 8.1. Timeline Flow - Middlebox implementing NAPT and Firewall In the following figure, an end-host inside the private network at address(pa) 10.0.0.4 wishes to communicate with an external FTP server with an IP address Ea. The Middlebox provides public IP address(Ma) 209.46.41.66 for external communication by private hosts. The middlebox diverts the FTP control traffic to the OOP agent. The OOP agent, in turn, reviews the datagrams and optionally modify as appropriate and redirects the datagrams right back to the middlebox. The OOP agent may need to update even the TCP SYNs and ACKs (i.e., datagrams with no application specific payload) in the event the agent had to rewrite the address content in the payload and the payload length changed as a result. FTP-client OOP FTP Middlebox (NAPT & FTP Server (Private) Agent Firewall Services) (External) IP addr(Pa): | IP addr(Ma): IP addr: Ea 10.0.0.4 | 209.46.41.66 | | | | | | | | | | |++Attach as FTP ALG+++>| | | | | | | |<+++++ OK +++++++++++++| | | | | | | The OOP FTP Agent attaches with middlebox & | | is authorized to process FTP control | | traffic from private hosts (or any set | | of hosts adhering to a certain policy) | | | | | | | | | The FTP client connects to the external FTP server. The middlebox would have created a PORT BIND and an FTP control session resource with the appropriate translation parameters. | | | | | PORT 10,0,0,4,4,9 | | |--------------------------------------| | | | | | | |<## Ctrl-Pkt diverted #| | | | | | | |++Query NAT Session | | | | Descriptor for | | | | Pa-to-Ea FTP flow+++>| | | | | | Srisuresh, et al. [Page 27] Internet-Draft MIDCOM Architecture & Framework May 2001 | |<+Pa-to-Ea FTP flow | | | | Session Descriptor+++| | | | | | | |++Create NAT port-BIND | | | | for (Pa, 1033) +++++>| | | | | | | |<+Port BINDs created | | | | with (Ma, 15324)+++++| | | | | | | |++Create NAT Session | | | | descriptor for the | | | | Data session from Ea | | | | to (Ma, 15324);Set | | | | Parent session to | | | | FTP-Ctrl session +++>| | | | | | | |<+FTP-Data session | | | | descriptor created+++| | | | | | | |++Permit FTP data | | | | session from Ea to | | | | (Ma, 15324)+++++++++>| | | | | | | |<+Data session OKed++++| | | | | | | |### Modifed Control | | | | Pkt forwarded #######################>| | | | | |<===FTP Data traffic between Pa & Ea==|=================>| | | | | | | | | Legend: ++++ MIDCOM control traffic ##### Diverted datagrams between ---- FTP control traffic ==== FTP data traffic The above flow does not indicate all packets as diverted, only the important ones (e.g. the datagram with the PORT command in the payload). It is safe to assume that all control packets are diverted from the middlebox to the OOP Agent via the datagram diversion component of the middlebox. Note that the FTP data traffic is not diverted to the OOP Agent. This is because the OOP agent does not assign a diversion function Srisuresh, et al. [Page 28] Internet-Draft MIDCOM Architecture & Framework May 2001 associated with the data session while at the instance creating the FTP-Data session. This is an essential feature, since we allow the middlebox to move the data about, while the Agent intervention is limited just to the control session. 9.0. Operational considerations 9.1. Multiple MIDCOM connections between agents and middlebox A middlebox cannot be assumed to be a simple device implementing just one middlebox function and no more than a couple of interfaces. Middleboxes often combine multiple intermediate functions into the same device and have the ability to provision individual interfaces of the same device with different sets of functions and varied provisioning for the same function across the interfaces. As such, a MIDCOM agent ought to be able to have a single MIDCOM connection with a middlebox and use the MIDCOM interface on the middlebox to interface with different services on the same middlebox interface. 9.2. MIDCOM agent registration with a middlebox A MIDCOM agent may be pre-configured on a middlebox as a trusted entity. In the case where a MIDCOM agent is not pre-configured, a policy server should be made available to the middlebox, so the middlebox can consult the Policy Server for authorization to accept requests from the agent. A middlebox should be capable of connecting to more than a single MIDCOM agent. 9.3. Asynchronous notification to MIDCOM agents Asynchronous notification by the middlebox to a MIDCOM agent can be useful for events such as Session creation, Session termination, MIDCOM protocol failure, Middlebox function failure or any other significant event. Independently, ICMP error codes can also be useful to notify transport layer failures to the agents. In addition, periodic notification of statistics update would also be a useful function that would be beneficial to certain types of agents. 9.4. Packet redirection Srisuresh, et al. [Page 29] Internet-Draft MIDCOM Architecture & Framework May 2001 Middleboxes should address the cases when MIDCOM agents are not in-path of communication of the traffic in question. The agents should be capable of returning the processed traffic to the middlebox point of origin or forward it to the destination. The middlebox should act accordingly when the traffic forwarded earlier is received. Packet forwarding by the agent might necessitate the packet to traverse the middlebox for the second time. The middlebox should simply forward the packet the second time around without redirecting to the agent once again. Failing this, the packet would simply be recycling between the two entities. The progressing mechanisms to avoid such pitfalls should be addressed by the MIDCOM protocol. A mechanism that maybe considered would be to adopt a tunneling approach for packet redirection between the agent and the middlebox. 9.5. Middleboxes supporting multiple services A middlebox could be implementing a variety of services (e.g. NAT and firewall) in the same box. Some of these services might have inter-dependency on shared resources and sequence of operation. Others may be independent of each other. Generally speaking, the sequence in which these function operations may be performed on datagrams is not within the scope of this document. In the case of a middlebox implementing NAT and firewall services, it is safe to state that the NAT operation will precede firewall on the egress and will follow firewall on the ingress. Further, firewall access control lists used by a firewall are assumed to be based on session parameters as seen on the interface supporting firewall service. 9.6. Signaling and Data traffic The class of applications the MIDCOM architecture is addressing focus around applications that have a combination of one or more signaling and data traffic sessions. The signaling may be done out-of-band using a dedicated stand-alone session or may be done in-band with data session. Alternately, signaling may also be done as a combination of both stand-alone and in-band sessions. SIP is an example of an application based on distinct signaling and data sessions. SIP signaling session is used for call setup between a caller and a callee. MIDCOM agent may be required to examine/modify SIP payload content to administer the middlebox Srisuresh, et al. [Page 30] Internet-Draft MIDCOM Architecture & Framework May 2001 so as to let the media streams (RTP/RTSP based) through. MIDCOM agent is not required to intervene in the data traffic. Signaling and context specific Header information is sent in-band within the same data stream for applications such as HTTP embedded applications, sun-RPC (embedding a variety of NFS apps), Oracle transactions (embedding oracle SQL+, MS ODBC, Peoplesoft) etc. H.323 is an example of application that sends signaling in both dedicated stand-alone session as well as in conjunction with data. Q.931 traffic traverses middleboxes by virtue of static policy, no MIDCOM control needed. Q.931 also negotiates ports for an H.245 TCP stream. A MIDCOM agent is required to examine/modify the contents of the H.245 so that H.245 can traverse it. H.245 traverses the middlebox and also carries Open Logical Channel information for media data. So the MIDCOM agent is once again required to examine/modify the payload content needs to let the media traffic flow. The MIDCOM architecture takes into consideration, supporting applications with independent signaling and data sessions as well as applications that have signaling and data communicated over the same session. In the cases where signaling is done on a single stand-alone session, it is desirable to have a MIDCOM agent interpret the signaling stream and program the middlebox (that transits the data stream) so as to let the data traffic through uninterrupted. 10. Applicability Statement Middleboxes may be stationed in a number of topologies. However, the signaling framework outlined in this document may be limited to only those middleboxes that are located in a DMZ (De-Militarized Zone) at the edge of a private domain, connecting to the Internet. Specifically, the assumption is that you have a single middlebox (running NAT or firewall) along the application route. Discovery of middlebox along application route is outside the scope of this document. It is conceivable to have middleboxes located between departments within the same domain or inside service provider's domain and so forth. However, care must be taken to review each individual scenario and determine the applicability on a case-by-case basis. The applicability may also be illustrated as follows. Real-time and streaming applications such as Voice-Over-IP and peer-to-peer Srisuresh, et al. [Page 31] Internet-Draft MIDCOM Architecture & Framework May 2001 applications require administering firewall and NAT middleboxes to let their media streams reach hosts inside a private domain. The requirements are in the form of establishing a "pin-hole" to permit a TCP/UDP session (the port parameters of which are dynamically determined) through a firewall or retain an address/port bind in the NAT device to permit connections to a port. These requirements are met by current generation middleboxes using adhoc methods, such as embedding application intelligence within a middlebox to identify the dynamic session parameters and administering the middlebox internally as appropriate. The objective of the MIDCOM architecture is to create a unified, standard way to exercise this functionality, currently existing in an ad-hoc fashion in some of the middleboxes. By adopting MIDCOM architecture, middleboxes will be able to support newer applications they have not been able to support thus far. MIDCOM architecture does not and MUST not, in anyway, change the fundamental characteristic of the services supported on the middlebox. Typically, organizations shield a majority of their corporate resources (such as end-hosts) from visibility to the external network by the use of a De-Militarized Zone (DMZ) at the domain edge. Only a portion of these hosts are allowed to be accessed by the external world. The remaining hosts and their names are unique to the private domain. Hosts visible to the external world and the authoritative name server that maps their names to network addresses are often configured within a DMZ (De-Militarized Zone) in front of a firewall. Hosts and middleboxes within DMZ are referred to as DMZ nodes. Figure 4 below illustrates configuration of a private domain with a DMZ at its edge. Actual configurations may vary. Internal hosts are accessed only by users inside the domain. Middleboxes, located in the DMZ may be accessed by agents inside or outside the domain. Srisuresh, et al. [Page 32] Internet-Draft MIDCOM Architecture & Framework May 2001 \ | / +-----------------------+ |Service Provider Router| +-----------------------+ WAN | Stub A .........|\|.... | +---------------+ | NAT Middlebox | +---------------+ | | DMZ - Network ------------------------------------------------------------ | | | | | +--+ +--+ +--+ +--+ +-----------+ |__| |__| |__| |__| | Firewall | /____\ /____\ /____\ /____\ | Middlebox | DMZ-Host1 DMZ-Host2 ... DMZ-Name DMZ-Web +-----------+ Server Server etc. | | Internal Hosts (inside the private domain) | ------------------------------------------------------------ | | | | +--+ +--+ +--+ +--+ |__| |__| |__| |__| /____\ /____\ /____\ /____\ Int-Host1 Int-Host2 ..... Int-Hostn Int-Name Server Figure 6: DMZ network configuration of a private domain. 11. Acknowledgements The authors wish to express their thanks and gratitude to the following for their valuable critique, advice and input on an earlier rough version of this document. Christian Huitema, Joon Maeng, Jon Peterson, Mike Fisk, Matt Holdrege, Melinda Shore, Paul Sijben, Philip Mart, Scott Brim and Richard Swale. The authors owe special thanks to Eliot Lear for kick-starting the e-mail discussion on used-case scenarios with a SIP application flow diagram through a middlebox. Much thanks to Bob Penfield, George Michaelson, Christopher Martin and others in the MIDCOM work group for continuing with timeline discussion to better understand the MIDCOM operations vis-a-vis application flows. Last, but not the least, the authors owe much thanks to Melinda Shore for her constant support, critique and unbiased feedback throughout in making this document a better read. Srisuresh, et al. [Page 33] Internet-Draft MIDCOM Architecture & Framework May 2001 12. Security Considerations [SEC-GUIDE] defines security goals as either communication security related or systems security related. While the latter is important and should be addressed as part of a comprehensive security solution which could be based on the features and behavior of the middlebox to such threats, it is considered to be outside the scope of this document. A middlebox performing packet filtering or NAT services requires secure access to its controlled internal resources. This requirement falls under the former goal. This section predominantly addresses what is required to ensure secure access to the middlebox. The secure access has a number of requirements: authorization, authentication, integrity and confidentiality. Authorization refers to whether a particular agent is authorized to signal middlebox with requests for one or more applications adhering to a certain policy profile. Failing the authorization process might indicate resource theft attempt or failure due to administrative and/or credential deficiencies. In either case, the middlebox should take the proper measures to audit/log such attempts and consult its designated policy server for the required action if the middlebox is configured with one. Alternatively, the middlebox may resort to a default service deny policy when a midcom agent fails to prompt the required credentials. Section 6 discusses the middlebox-policy server interactions in view of policy decisions. Authentication refers to whether a particular agent can provide enough credentials to authenticate itself to the middlebox and if the middlebox has enough credentials to authenticate itself to the agent. Since the middlebox is implementing a security function as a service for a midcom agent, it needs to be sure of the identity of the agent. Likewise, the agent needs to confirm that a signaling request is served by the middlebox supposed to render the service in order to provide a level of service reliability to its customers. Credentials are used to establish the identify of the endpoint and consequently an authorization decision is drawn as to whether allowing the midcom process to proceed. Failing to submit the required/valid credentials once challenged may indicate a replay attack and in which case a proper action is required by the middlebox such as auditing, logging, consulting its designated policy server to reflect such failure. Mechanisms based on secret keys (public-key based or shared) or certificates through some Certificate authority can be utilized to facilitate the authentication process. Lack of strong credentials during the Srisuresh, et al. [Page 34] Internet-Draft MIDCOM Architecture & Framework May 2001 authentication process can seriously jeopardize the fundamental service rendered by the middlebox. Integrity refers to the messages carrying the midcom signaling requests in order to ensure that integrity is maintained and has not been accidentally or maliciously altered or destroyed. While authentication and message integrity are two distinct functionalities, they are closely related as most algorithms require a secret key (public-based or shared) to complete the authentication process and carry out the integrity checks of exchanged messages. To accommodate the authentication and integrity constraints of the midcom signaling and to reuse existing transport-based security solutions such as Authentication Header [RFC2402] MAY be used when the threat environment requires strong authentication and integrity protections, but does not require confidentiality. Confidentiality refers to the messages carrying the midcom signaling requests in order to ensure that the signaling requests are accessible only to the authorized entity. When a middlebox agent is deployed in an untrusted environment or to satisfy stronger security policy requirements, confidentiality SHOULD be applied to the signaling messages. When confidentiality is not administered properly, the domains protected by the middlebox can be at a serious risk due to the sensitivity of the midcom signaling. To accommodate that, a transport-based encryption such as ESP tunneling [RFC2406] MAY be deployed between the middlebox and the agent. This will ensure the confidentiality and integrity of midcom communications. Lastly, there can be security vulnerability to the applications traversing a middlebox when a resource on a middlebox is controlled by multiple external agents. A middlebox service may be abruptly disrupted due to malicious manipulation or incorrect implementation of the middlebox or its agents of a certain shared resource by an agent purporting to offer ALG service for a different middlebox function. Care must be taken in the protocol design to ensure that agents for one function do not abruptly step over resources impacting a different function. Alternately, the severity of such manifestations could be lessened when a single MIDCOM agent is responsible for supporting all the middlebox services for an application due to the reduced complexity and synchronization effort in managing the middlebox resources. Srisuresh, et al. [Page 35] Internet-Draft MIDCOM Architecture & Framework May 2001 REFERENCES [IETF-STD] Bradner, S., " The Internet Standards Process -- Revision 3", RFC 1602, IETF, October 1996. [SIP] Handley, M., H. Schulzrinne, E. Schooler, and J. Rosenberg, "SIP: Session Initiation Protocol", RFC 2543, IETF, March 1999. [SDP] Handley, M., and Jacobson, V., "SDP: session description protocol", RFC 2327, IETF, April 1998. [H.323] ITU-T Recommendation H.323. "Packet-based Multimedia Communications Systems," 1998. [RTP] Schulzrinne, H., S. Casner, R. Frederick, and V. Jacobson, "RTP: A Transport Protocol for Real-Time Applications", RFC 1889, IETF, January 1996. [RTSP] Schulzrinne, H., A. Rao, R. Lanphier: "Real Time Streaming Protocol", RFC 2326, IETF, April 1998. [FTP] J. Postel, J. Reynolds, "FILE TRANSFER PROTOCOL (FTP)", RFC 959 [NAT-TERM] Srisuresh, P. and M. Holdrege, "IP Network Address Translator (NAT) Terminology and Considerations", RFC 2663, August 1999. [NAT-TRAD] Srisuresh, P. and Egevang, K., "Traditional IP Network Address Translator (Traditional NAT)", RFC 3022, January 2001. [NAT-COMP] Holdrege, M. and Srisuresh, P., "Protocol Complications with the IP Network Address Translator", RFC 3027, January 2001. [NAT-PT] Tsirtsis, G. and Srisuresh, P., "Network Address Translation - Protocol Translation (NAT-PT)", RFC 2766, February 2000. [NAT-FRAMEWORK] Srisuresh, P., "Framework for interfacing with Network Address Translator", Work in progress, April 2001, [MIDCOM-REQ] Swale, R.P., Mart, P.A. and Sijben, P., "Requirements for the MIDCOM protocol", work in progress, April 2001, Srisuresh, et al. [Page 36] Internet-Draft MIDCOM Architecture & Framework May 2001 [APPL-ID] Bernet, Y. and Pabbati, R., "Application and Sub Application Identity Policy Element for Use with RSVP", RFC 2872, June 2000. [RFC 1918] Rekhter, Y., Moskowitz, B., Karrenberg, D., de Groot, G. and E. Lear, "Address Allocation for Private Internets", BCP 5, RFC 1918, February 1996. [RFC 1700] J. Reynolds and J. Postel, "Assigned Numbers", RFC 1700 [IPsec-AH] Kent, S., and R. Atkinson, "IP Authentication Header", RFC 2402, November 1998. [IPsec-ESP] Kent, S., and R. Atkinson, "IP Encapsulating Security Payload (ESP)", RFC 2406, November 1998. [TLS] Dierks, T., and Allen, C., "The TLS Protocol Version 1.0", RFC 2246, January 1999. [SEC-GUIDE] Rescorla, E., and B. Korver, "Guidlines for Writing RFC Text on Security Considerations", Work in Progress, March 2001, Authors' Addresses Pyda Srisuresh Jasmine Networks 3061 Zanker Road, Suite B San Jose, CA 95134 U.S.A. EMail: srisuresh@yahoo.com Jiri Kuthan GMD Fokus Kaiserin-Augusta-Allee 31 D-10589 Berlin, Germany E-mail: kuthan@fokus.gmd.de Jonathan Rosenberg dynamicsoft 200 Executive Drive Suite 120 Srisuresh, et al. [Page 37] Internet-Draft MIDCOM Architecture & Framework May 2001 West Orange, NJ 07052 U.S.A. email: jdrosen@dynamicsoft.com Andrew Molitor Aravox technologies 4201 Lexington Avenue North, Suite 1105 Arden Hills, MN 55126 U.S.A. voice: (651) 256-2700 email: amolitor@visi.com Abdallah Rayhan P.O. Box 3511 Stn C Ottawa, ON, Canada K1Y 4H7 eMail: ar_rayhan@yahoo.ca Srisuresh, et al. [Page 38]