RFC 5190
Definitions of Managed Objects for Middlebox Communication
Pages: 92
Proposed Standard
Proposed Standard
Part 1 of 4 – Pages 1 to 15
Network Working Group J. Quittek Request for Comments: 5190 M. Stiemerling Category: Standards Track NEC P. Srisuresh Kazeon Systems March 2008 Definitions of Managed Objects for Middlebox Communication Status of This Memo This document specifies an Internet standards track protocol for the Internet community, and requests discussion and suggestions for improvements. Please refer to the current edition of the "Internet Official Protocol Standards" (STD 1) for the standardization state and status of this protocol. Distribution of this memo is unlimited.Abstract
This memo defines a portion of the Management Information Base (MIB) for use with network management protocols in the Internet community. In particular, it describes a set of managed objects that allow configuring middleboxes, such as firewalls and network address translators, in order to enable communication across these devices. The definitions of managed objects in this documents follow closely the MIDCOM semantics defined in RFC 5189.
Table of Contents
1. Introduction ....................................................4 2. The Internet-Standard Management Framework ......................4 3. Overview ........................................................4 3.1. Terminology ................................................5 4. Realizing the MIDCOM Protocol with SNMP .........................6 4.1. MIDCOM Sessions ............................................6 4.1.1. Authentication and Authorization ....................6 4.2. MIDCOM Transactions ........................................7 4.2.1. Asynchronous Transactions ...........................7 4.2.2. Configuration Transactions ..........................8 4.2.3. Monitoring Transactions ............................11 4.2.4. Atomicity of MIDCOM Transactions ...................12 4.2.4.1. Asynchronous MIDCOM Transactions ..........12 4.2.4.2. Session Establishment and Termination Transactions ..................12 4.2.4.3. Monitoring Transactions ...................13 4.2.4.4. Lifetime Change Transactions ..............13 4.2.4.5. Transactions Establishing New Policy Rules ..............................14 4.2.5. Access Control .....................................14 4.3. Access Control Policies ...................................14 5. Structure of the MIB Module ....................................15 5.1. Transaction Objects .......................................16 5.1.1. midcomRuleTable ....................................17 5.1.2. midcomGroupTable ...................................19 5.2. Configuration Objects .....................................20 5.2.1. Capabilities .......................................20 5.2.2. midcomConfigFirewallTable ..........................21 5.3. Monitoring Objects ........................................22 5.3.1. midcomResourceTable ................................22 5.3.2. midcomStatistics ...................................24 5.4. Notifications .............................................25 6. Recommendations for Configuration and Operation ................26 6.1. Security Model Configuration ..............................26 6.2. VACM Configuration ........................................27 6.3. Notification Configuration ................................28 6.4. Simultaneous Access .......................................28 6.5. Avoiding Idempotency Problems .............................29 6.6. Interface Indexing Problems ...............................29 6.7. Applicability Restrictions ................................30 7. Usage Examples for MIDCOM Transactions .........................30 7.1. Session Establishment (SE) ................................31 7.2. Session Termination (ST) ..................................31 7.3. Policy Reserve Rule (PRR) .................................31 7.4. Policy Enable Rule (PER) after PRR ........................33 7.5. Policy Enable Rule (PER) without Previous PRR .............34
7.6. Policy Rule Lifetime Change (RLC) .........................35
7.7. Policy Rule List (PRL) ....................................35
7.8. Policy Rule Status (PRS) ..................................35
7.9. Asynchronous Policy Rule Event (ARE) ......................36
7.10. Group Lifetime Change (GLC) ..............................36
7.11. Group List (GL) ..........................................36
7.12. Group Status (GS) ........................................37
8. Usage Examples for Monitoring Objects ..........................37
8.1. Monitoring NAT Resources ..................................37
8.2. Monitoring Firewall Resources .............................38
9. Definitions ....................................................38
10. Security Considerations .......................................85
10.1. General Security Issues ..................................85
10.2. Unauthorized Middlebox Configuration .....................86
10.3. Unauthorized Access to Middlebox Configuration ...........87
10.4. Unauthorized Access to MIDCOM Service Configuration ......88
11. Acknowledgements ..............................................88
12. IANA Considerations ...........................................88
13. Normative References ..........................................88
14. Informative References ........................................90
1. Introduction
This memo defines a portion of the Management Information Base (MIB) for use with network management protocols in the Internet community. In particular, it describes a set of managed objects that allow controlling middleboxes. The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in RFC 2119 [RFC2119].2. The Internet-Standard Management Framework
For a detailed overview of the documents that describe the current Internet-Standard Management Framework, please refer to section 7 of RFC 3410 [RFC3410]. Managed objects are accessed via a virtual information store, termed the Management Information Base or MIB. MIB objects are generally accessed through the Simple Network Management Protocol (SNMP). Objects in the MIB are defined using the mechanisms defined in the Structure of Management Information (SMI). This memo specifies a MIB module that is compliant to the SMIv2, which is described in STD 58, RFC 2578 [RFC2578], STD 58, RFC 2579 [RFC2579] and STD 58, RFC 2580 [RFC2580].3. Overview
The managed objects defined in this document serve for controlling firewalls and Network Address Translators (NATs). As defined in [RFC3234], firewalls and NATs belong to the group of middleboxes. A middlebox is a device on the datagram path between source and destination, which performs other functions than just IP routing. As outlined in [RFC3303], firewalls and NATs are potential obstacles to packet streams, for example, if dynamically negotiated UDP or TCP port numbers are used, as in many peer-to-peer communication applications. As one possible solution for this problem, the IETF MIDCOM working group defined a framework [RFC3303], requirements [RFC3304], and protocol semantics [RFC5189] for communication between applications and middleboxes acting as firewalls, NATs, or a combination of both. The MIDCOM architecture and framework define a model in which trusted third parties can be delegated to assist middleboxes in performing their operations, without requiring application intelligence being embedded in the middleboxes. This trusted third party is referred to as the MIDCOM agent. The MIDCOM protocol is defined between a MIDCOM agent and a middlebox.
The managed objects defined in this document can be used for dynamically configuring middleboxes on the datagram path to permit datagrams traversing the middleboxes. This way, applications can, for example, request pinholes at firewalls and address bindings at NATs. Besides managed objects for controlling the middlebox operation, this document also defines managed objects that provide information on middlebox resource usage (such as firewall pinholes, NAT bindings, NAT sessions, etc.) affected by requests. Since firewalls and NATs are critical devices concerning network security, security issues of middlebox communication need to be considered very carefully.3.1. Terminology
The terminology used in this document is fully aligned with the terminology defined in [RFC5189] except for the term 'MIDCOM agent'. For this term, there is a conflict between the MIDCOM terminology and the SNMP terminology. The roles of entities participating in SNMP communication are called 'manager' and 'agent' with the agent acting as server for requests from the manager. This use of the term 'agent' is different from its use in the MIDCOM framework: The SNMP manager corresponds to the MIDCOM agent and the SNMP agent corresponds to the MIDCOM middlebox, also called MIDCOM server. In order to avoid confusion in this document specifying a MIB module, we replace the term 'MIDCOM agent' with 'MIDCOM client'. Whenever the term 'agent' is used in this document, it refers to the SNMP agent. Figure 1 sketches the entities of MIDCOM in relationship to SNMP manager and SNMP agent. +---------+ MIDCOM +-----------+ | MIDCOM |<~ ~ ~ ~ ~ ~ ~ ~>| MIDCOM | | Client | Transaction | middlebox | | | | (server) | +---------+ +-----------+ ^ ^ | | v v +---------+ +-----------+ | SNMP | SNMP | SNMP | | Manager |<===============>| Agent | +---------+ Protocol +-----------+ Figure 1: Mapping of MIDCOM to SNMP
4. Realizing the MIDCOM Protocol with SNMP
In order to realize middlebox communication as described in [RFC5189], several aspects and properties of the MIDCOM protocol need to be mapped to SNMP capabilities and expressed in terms of the Structure of Management Information version 2 (SMIv2). Basic concepts to be mapped are MIDCOM sessions and MIDCOM transactions. For both, access control policies need to be supported.4.1. MIDCOM Sessions
SNMP has no direct support for sessions. Therefore, they need to be modeled. A MIDCOM session is stateful and has a context that is valid for several transactions. For SNMP, a context is valid for a single transaction only, for example, covering just a single request/reply pair of messages. Properties of sessions that are utilized by the MIDCOM semantics and not available in SNMP need to be modeled. Particularly, the middlebox needs to be able to authenticate MIDCOM clients, authorize access to policy rules, and send notification messages concerning policy rules to MIDCOM clients participating in a session. In the MIDCOM-MIB module, authentication and access control are performed on a per-message basis using an SNMPv3 security model, such as the User-based Security Model (USM) [RFC3414], for authentication, and the View-based Access Control Model (VACM) [RFC3415] for access control. Sending notifications to MIDCOM clients is controlled by access control models such as VACM and a mostly static configuration of objects in the SNMP-TARGET-MIB [RFC3413] and the SNMP- NOTIFICATION-MIB [RFC3413]. This session model is static except that the MIDCOM client can switch on and off the generation of SNMP notifications that the middlebox sends. Recommended configurations of VACM and the SNMP-TARGET-MIB and the SNMP-NOTIFICATION-MIB that can serve for modeling a session are described in detail in section 6.4.1.1. Authentication and Authorization
MIDCOM sessions are required for providing authentication, authorization, and encryption for messages exchanged between a MIDCOM client and a middlebox. SNMPv3 provides these features on a per- message basis instead of a per-session basis applying a security model and an access control model, such as USM and VACM. Per-message
security mechanisms can be considered as overhead compared to per- session security mechanisms, but it certainly satisfies the security requirements of middlebox communication. For each authenticated MIDCOM client, access to the MIDCOM-MIB, particularly to policy rules, should be configured as part of the VACM configuration of the SNMP agent.4.2. MIDCOM Transactions
[RFC5189] defines the MIDCOM protocol semantics in terms of transactions and transaction parameters. Transactions are grouped into request-reply transactions and asynchronous transactions. SNMP offers simple transactions that in general cannot be mapped one-to-one to MIDCOM transactions. This section describes how the MIDCOM-MIB module implements MIDCOM transactions using SNMP transactions. The concerned MIDCOM transactions are asynchronous transactions and request-reply transactions. Within the set of request-reply transactions, we distinguish configuration transactions and monitoring transactions, because they are implemented in slightly different ways by using SNMP transactions. The SNMP terminology as defined in [RFC3411] does not use the concept of transactions, but of SNMP operations. For the considerations in this section, we use the terms SNMP GET transaction and SNMP SET transaction. An SNMP GET transaction consists of an SNMP Read Class operation and an SNMP Response Class operation. An SNMP SET transaction consists of an SNMP Write Class operation and an SNMP Response Class operation.4.2.1. Asynchronous Transactions
Asynchronous transactions can easily be modeled by SNMP Notification Class operations. An asynchronous transaction contains a notification message with one to three parameters. The message can be realized as an SNMP Notification Class operation with the parameters implemented as managed objects contained in the notification.
+--------------+ notification +------------+ | MIDCOM client|<--------------| middlebox | +--------------+ message +------------+ MIDCOM asynchronous transaction +--------------+ SNMP +------------+ | SNMP manager |<--------------| SNMP agent | +--------------+ notification +------------+ Implementation of MIDCOM asynchronous transaction Figure 2: MIDCOM asynchronous transaction mapped to SNMP Notification Class operation One of the parameters is the transaction identifier that should be unique per middlebox. It does not have to be unique for all notifications sent by the particular SNMP agent, but for all sent notifications that are defined by the MIDCOM-MIB module. Note that SNMP notifications are usually sent as unreliable UDP packets and may be dropped before they reach their destination. If a MIDCOM client is expecting an asynchronous notification on a specific transaction, it would be the job of the MIDCOM client to poll the middlebox periodically and monitor the transaction in case notifications are lost along the way.4.2.2. Configuration Transactions
All request-reply transactions contain a request message, a reply message, and potentially also a set of notifications. In general, they cannot be modeled by just having a single SNMP message per MIDCOM message, because some of the MIDCOM messages carry a large set of parameters that do not necessarily fit into an SNMP message consisting of a single UDP packet only. For configuration transactions, the MIDCOM request message can be modeled by one or more SNMP SET transactions. The action of sending the MIDCOM request to the middlebox is realized by writing the parameters contained in the message to managed objects at the SNMP agent. If necessary, the SNMP SET transaction includes creating these managed objects. If not all parameters of the MIDCOM request message can be set by a single SNMP SET transaction, then more than one SET transaction is used; see Figure 3. Completion of the last of the SNMP transactions indicates that all required parameters are set and that processing of the MIDCOM request message can start at the middlebox.
Please note that a single SNMP SET transaction consists of an SNMP SET request message and an SNMP SET reply message. Both are sent as unreliable UDP packets and may be dropped before they reach their destination. If the SNMP SET request message or the SNMP reply message is lost, then the SNMP manager (the MIDCOM client) needs to take action, for example, by just repeating the SET transaction or by first checking the success of the initial write transaction with an SNMP GET transaction and then only repeating the SNMP SET transaction if necessary. +--------------+ request +------------+ | MIDCOM client|-------------->| middlebox | +--------------+ message +------------+ MIDCOM request message +--------------+ +------------+ | | SNMP SET | | | |-------------->| | | | message | | | | | | | | SNMP SET | | | |<--------------| | | | reply message | | | SNMP manager | | SNMP agent | | | SNMP SET | | | |- - - - - - - >| | | | message | | | | | | | | SNMP SET | | | |< - - - - - - -| | | | reply message | | | | | | | | . . . | | +--------------+ +------------+ Implementation of MIDCOM request message by one or more SNMP SET transactions Figure 3: MIDCOM request message mapped to SNMP SET transactions The MIDCOM reply message can be modeled in two ways. The first way is an SNMP Notification Class operation optionally followed by one or more SNMP GET transactions as shown in Figure 4. The MIDCOM server informs the MIDCOM client about the end of processing the request by sending an SNMP notification. If possible, the SNMP notification
carries all reply parameters. If this is not possible, then the SNMP manager has to perform additional SNMP GET transactions as long as necessary to receive all of the reply parameters. +--------------+ reply +------------+ | MIDCOM client|<--------------| middlebox | +--------------+ message +------------+ MIDCOM reply message +--------------+ +------------+ | | SNMP | | | |<--------------| | | | notification | | | | | | | | SNMP GET | | | |-------------->| | | | message | | | SNMP manager | | SNMP agent | | | SNMP GET | | | |<--------------| | | | reply message | | | | | | | | SNMP GET | | | |- - - - - - - >| | | | message | | | | | | | | SNMP GET | | | |< - - - - - - -| | | | reply message | | | | | | | | . . . | | +--------------+ +------------+ Implementation of MIDCOM reply message by an SNMP notification and one or more SNMP GET transactions Figure 4: MIDCOM reply message mapped to SNMP notification and optional GET transactions The second way replaces the SNMP Notification Class operation by a polling operation of the SNMP manager. The manager polls status information at the SNMP agent using SNMP GET transactions until it detects the end of the processing of the request. Then it uses one or more SNMP GET transactions to receive all of the reply parameters. Note that this second way requires more SNMP operations, but is more
reliable than the first way using an SNMP Notification Class operation.4.2.3. Monitoring Transactions
The realization of MIDCOM monitoring transactions in terms of SNMP transactions is simpler. The request message is very short and just specifies a piece of information that the MIDCOM client wants to retrieve. +--------------+ request +------------+ | |-------------->| | | | message | | | MIDCOM client| | middlebox | | | reply | | | |<--------------| | +--------------+ message +------------+ MIDCOM monitoring transaction +--------------+ +------------+ | | SNMP GET | | | |-------------->| | | | message | | | | | | | | SNMP GET | | | |<--------------| | | | reply message | | | SNMP manager | | SNMP agent | | | SNMP GET | | | |- - - - - - - >| | | | message | | | | | | | | SNMP GET | | | |< - - - - - - -| | | | reply message | | | | | | | | . . . | | +--------------+ +------------+ Implementation of MIDCOM monitoring transaction by one or more SNMP GET messages Figure 5: MIDCOM monitoring transaction mapped to SNMP GET transactions
Since monitoring is a strength of SNMP, there are sufficient means to realize MIDCOM monitoring transactions simpler than MIDCOM configuration transactions. All MIDCOM monitoring transactions can be realized as a sequence of SNMP GET transactions. The number of SNMP GET transactions required depends on the amount of information to be retrieved.4.2.4. Atomicity of MIDCOM Transactions
Given the realizations of MIDCOM transactions by means of SNMP transactions, atomicity of the MIDCOM transactions is not fully guaranteed anymore. However, this section shows that atomicity provided by the MIB module specified in section 9 is still sufficient for meeting the MIDCOM requirements specified in [RFC3304].4.2.4.1. Asynchronous MIDCOM Transactions
There are two asynchronous MIDCOM transactions: Asynchronous Session Termination (AST) and Asynchronous Policy Rule Event (ARE). The very static realization of MIDCOM sessions in the MIDCOM-MIB, as described by section 4.1, does not anymore support the asynchronous termination of a session. Therefore, the AST transaction is not modeled. For the ARE, atomicity is maintained, because it is modeled by a single atomic SNMP notification transaction. In addition, the MIDCOM-MIB supports an Asynchronous Group Event transaction, which is an aggregation of a set of ARE transactions. Also, this MIDCOM transaction is implemented by a single SNMP transaction.4.2.4.2. Session Establishment and Termination Transactions
The MIDCOM-MIB models MIDCOM sessions in a very static way. The only dynamic actions within these transactions are enabling and disabling the generation of SNMP notifications at the SNMP agent. For the Session Establishment (SE) transaction, the MIDCOM client first reads the middlebox capabilities. It is not relevant whether or not this action is atomic because a dynamic change of the middlebox capabilities is not to be expected. Therefore, also non- atomic implementations of this action are acceptable. Then, the MIDCOM agent needs to enable the generation of SNMP notifications at the middlebox. This can be realized by writing to a single managed object in the SNMP-NOTIFICATION-MIB [RFC3413]. But even other implementations are acceptable, because atomicity is not required for this step.
For the Session Termination (ST) transaction, the only required action is disabling the generation of SNMP notifications at the middlebox. As for the SE transaction, this action can be realized atomically by using the SNMP-NOTIFICATION-MIB, but also other implementations are acceptable because atomicity is not required for this action.4.2.4.3. Monitoring Transactions
Potentially, the monitoring transactions Policy Rule List (PRL), Policy Rule Status (PRS), Group List (GL), and Group Status (GS) are not atomic, because these transactions may be implemented by more than one SNMP GET operation. The problem that might occur is that while the monitoring transaction is performed, the monitored items may change. For example, while reading a long list of policies, new policies may be added and already read policies may be deleted. This is not in line with the protocol semantics. However, it is not in direct conflict with the MIDCOM requirement requesting the middlebox state to be stable and known by the MIDCOM client, because the middlebox notifies the MIDCOM client on all changes to its state that are performed during the monitoring transaction by sending notifications. If the MIDCOM client receives such a notification while performing a monitoring transaction (or shortly after completing it), the MIDCOM client can then either repeat the monitoring transaction or integrate the result of the monitoring transaction with the information received via notifications during the transaction. In both cases, the MIDCOM client will know the state of the middlebox.4.2.4.4. Lifetime Change Transactions
For the policy Rule Lifetime Change (RLC) transaction and the Group Lifetime Change (GLC) transaction, atomicity is maintained. They both have very few parameters for the request message and the reply message. The request parameters can be transmitted by a single SNMP SET request message, and the reply parameters can be transmitted by a single SNMP notification message. In order to prevent idempotency problems by retransmitting an SNMP request after a lost SNMP reply, it is RECOMMENDED that either snmpSetSerialNo (see [RFC3418]) is included in the corresponding SNMP SET request or the value of the SNMP retransmission timer be lower than the smallest requested lifetime value. The same recommendation applies to the smallest requested value for the midcomRuleStorageTime. MIDCOM client implementations MAY completely avoid this problem by configuring their SNMP stack such that no retransmissions are sent.
4.2.4.5. Transactions Establishing New Policy Rules
Analogous to the monitoring transactions, the atomicity may not be given for Policy Reserve Rule (PRR) and Policy Enable Rule (PER) transactions. Both transactions are potentially implemented using more than one SNMP SET operation and GET operation for obtaining transaction reply parameters. The solution for this loss of atomicity is the same as for the monitoring transactions. There is an additional atomicity problem for PRR and PER. If transferring request parameters requires more than a single SET operation, then there is the potential problem that multiple MIDCOM clients sharing the same permissions are able to access the same policy rule. In this case, a client could alter request parameters already set by another client before the first client could complete the request. However, this is acceptable since usually only one agent is creating a policy rule and filling it subsequently. It can also be assumed that in most cases where clients share permissions, they act in a more or less coordinated way avoiding such interferences. All atomicity problems caused by using multiple SNMP SET transactions for implementing the MIDCOM request message can be avoided by transferring all request parameters with a single SNMP SET transaction.4.2.5. Access Control
Since SNMP does not offer per-session authentication and authorization, authentication and authorization are performed per SNMP message sent from the MIDCOM client to the middlebox. For each transaction, the MIDCOM client has to authenticate itself as an authenticated principal, such as a USM user. Then, the principal's access rights to all resources affected by the transaction are checked. Access right control is realized by configuring the access control mechanisms, such as VACM, at the SNMP agent.4.3. Access Control Policies
Potentially, a middlebox has to control access for a large set of MIDCOM clients and to a large set of policy rules configuring firewall pinholes and NAT bindings. Therefore, it can be beneficial to use access control policies for specifying access control rules. Generating, provisioning, and managing these policies are out of scope of this MIB module.
However, if such an access control policy system is used, then the SNMP agent acts as a policy enforcement point. An access control policy system must transform all active policies into configurations of, for example, the SNMP agent's View-based Access Control Model (VACM). The mechanisms of access control models, such as VACM, allow an access control policy system to enforce MIDCOM client authentication rules and general access control of MIDCOM clients to middlebox control. The mechanisms of VACM can be used to enforce access control of authenticated clients to MIDCOM-MIB policy rules based on the concept of ownership. For example, an access control policy can specify that MIDCOM-MIB policy rules owned by user A cannot be accessed at all by user B, can be read by user C, and can be read and modified by user D. Further access control policies can control access to concrete middlebox resources. These are enforced, when a MIDCOM request is processed. For example, an authenticated MIDCOM client may be authorized to request new MIDCOM policies to be established, but only for certain IP address ranges. The enforcement of this kind of policies may not be realizable using available SNMP mechanisms, but needs to be performed by the individual MIB module implementation.
(page 15 continued on part 2)