This is a purely informative rendering of an RFC that includes verified errata. This rendering may not be used as a reference.

The following 'Verified' errata have been incorporated in this document: EID 4313
Network Working Group                                           F. Baker
Request for Comments: 2747                                         Cisco
Category: Standards Track                                     B. Lindell
                                                               M. Talwar
                                                            January 2000

                   RSVP Cryptographic Authentication

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.

Copyright Notice

   Copyright (C) The Internet Society (2000).  All Rights Reserved.


   This document describes the format and use of RSVP's INTEGRITY object
   to provide hop-by-hop integrity and authentication of RSVP messages.

1.  Introduction

   The Resource ReSerVation Protocol RSVP [1] is a protocol for setting
   up distributed state in routers and hosts, and in particular for
   reserving resources to implement integrated service.  RSVP allows
   particular users to obtain preferential access to network resources,
   under the control of an admission control mechanism.  Permission to
   make a reservation will depend both upon the availability of the
   requested resources along the path of the data, and upon satisfaction
   of policy rules.

   To ensure the integrity of this admission control mechanism, RSVP
   requires the ability to protect its messages against corruption and
   spoofing.  This document defines a mechanism to protect RSVP message
   integrity hop-by-hop.  The proposed scheme transmits an
   authenticating digest of the message, computed using a secret
   Authentication Key and a keyed-hash algorithm.  This scheme provides
   protection against forgery or message modification.  The INTEGRITY
   object of each RSVP message is tagged with a one-time-use sequence

   number.  This allows the message receiver to identify playbacks and
   hence to thwart replay attacks.  The proposed mechanism does not
   afford confidentiality, since messages stay in the clear; however,
   the mechanism is also exportable from most countries, which would be
   impossible were a privacy algorithm to be used.  Note: this document
   uses the terms "sender" and "receiver" differently from [1].  They
   are used here to refer to systems that face each other across an RSVP
   hop, the "sender" being the system generating RSVP messages.

   The message replay prevention algorithm is quite simple.  The sender
   generates packets with monotonically increasing sequence numbers.  In
   turn, the receiver only accepts packets that have a larger sequence
   number than the previous packet.  To start this process, a receiver
   handshakes with the sender to get an initial sequence number.  This
   memo discusses ways to relax the strictness of the in-order delivery
   of messages as well as techniques to generate monotonically
   increasing sequence numbers that are robust across sender failures
   and restarts.

   The proposed mechanism is independent of a specific cryptographic
   algorithm, but the document describes the use of Keyed-Hashing for
   Message Authentication using HMAC-MD5 [7].  As noted in [7], there
   exist stronger hashes, such as HMAC-SHA1; where warranted,
   implementations will do well to make them available.  However, in the
   general case, [7] suggests that HMAC-MD5 is adequate to the purpose
   at hand and has preferable performance characteristics.  [7] also
   offers source code and test vectors for this algorithm, a boon to
   those who would test for interoperability.  HMAC-MD5 is required as a
   baseline to be universally included in RSVP implementations providing
   cryptographic authentication, with other proposals optional (see
   Section 6 on Conformance Requirements).

   The RSVP checksum MAY be disabled (set to zero) when the INTEGRITY
   object is included in the message, as the message digest is a much
   stronger integrity check.

1.1.  Conventions used in this document

   The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT",
   document are to be interpreted as described in [8].

1.2.  Why not use the Standard IPSEC Authentication Header?

   One obvious question is why, since there exists a standard
   authentication mechanism, IPSEC [3,5], we would choose not to use it.
   This was discussed at length in the working group, and the use of
   IPSEC was rejected for the following reasons.

   The security associations in IPSEC are based on destination address.
   It is not clear that RSVP messages are well defined for either source
   or destination based security associations, as a router must forward
   PATH and PATH TEAR messages using the same source address as the
   sender listed in the SENDER TEMPLATE.  RSVP traffic may otherwise not
   follow exactly the same path as data traffic.  Using either source or
   destination based associations would require opening a new security
   association among the routers for which a reservation traverses.

   In addition, it was noted that neighbor relationships between RSVP
   systems are not limited to those that face one another across a
   communication channel.  RSVP relationships across non-RSVP clouds,
   such as those described in Section 2.9 of [1], are not necessarily
   visible to the sending system.  These arguments suggest the use of a
   key management strategy based on RSVP router to RSVP router
   associations instead of IPSEC.

2.  Data Structures

2.1.  INTEGRITY Object Format

   An RSVP message consists of a sequence of "objects," which are type-
   length-value encoded fields having specific purposes.  The
   information required for hop-by-hop integrity checking is carried in
   an INTEGRITY object.  The same INTEGRITY object type is used for both
   IPv4 and IPv6.

   The INTEGRITY object has the following format:

      Keyed Message Digest INTEGRITY Object: Class = 4, C-Type = 1

       |    Flags    | 0 (Reserved)|                           |
       +-------------+-------------+                           +
       |                    Key Identifier                     |
       |                    Sequence Number                    |
       |                                                       |
       |                                                       |
       +                                                       +
       |                                                       |
       +                  Keyed Message Digest                 |
       |                                                       |
       +                                                       +
       |                                                       |

     o    Flags: An 8-bit field with the following format:


                          0   1   2   3   4   5   6   7
                        | H |                           |
                        | F |             0             |

          Currently only one flag (HF) is defined.  The remaining flags
          are reserved for future use and MUST be set to 0.

          o    Bit 0: Handshake Flag (HF) concerns the integrity
               handshake mechanism (Section 4.3).  Message senders
               willing to respond to integrity handshake messages SHOULD
               set this flag to 1 whereas those that will reject
               integrity handshake messages SHOULD set this to 0.

     o    Key Identifier: An unsigned 48-bit number that MUST be unique
          for a given sender.  Locally unique Key Identifiers can be
          generated using some combination of the address (IP or MAC or
          LIH) of the sending interface and the key number.  The
          combination of the Key Identifier and the sending system's IP
          address uniquely identifies the security association (Section

     o    Sequence Number: An unsigned 64-bit monotonically increasing,
          unique sequence number.

          Sequence Number values may be any monotonically increasing
          sequence that provides the INTEGRITY object [of each RSVP
          message] with a tag that is unique for the associated key's
          lifetime.  Details on sequence number generation are presented
          in Section 3.

     o    Keyed Message Digest: The digest MUST be a multiple of 4
          octets long.  For HMAC-MD5, it will be 16 bytes long.

2.2.  Security Association

   The sending and receiving systems maintain a security association for
   each authentication key that they share.  This security association
   includes the following parameters:

     o    Authentication algorithm and algorithm mode being used.

     o    Key used with the authentication algorithm.

     o    Lifetime of the key.

     o    Associated sending interface and other security association
          selection criteria [REQUIRED at Sending System].

     o    Source Address of the sending system [REQUIRED at Receiving

     o    Latest sending sequence number used with this key identifier
          [REQUIRED at Sending System].

     o    List of last N sequence numbers received with this key
          identifier [REQUIRED at Receiving System].

3.  Generating Sequence Numbers

   In this section we describe methods that could be chosen to generate
   the sequence numbers used in the INTEGRITY object of an RSVP message.
   As previous stated, there are two important properties that MUST be
   satisfied by the generation procedure.  The first property is that
   the sequence numbers are unique, or one-time, for the lifetime of the
   integrity key that is in current use.  A receiver can use this
   property to unambiguously distinguish between a new or a replayed
   message.  The second property is that the sequence numbers are
   generated in monotonically increasing order, modulo 2^64.  This is
   required to greatly reduce the amount of saved state, since a
   receiver only needs to save the value of the highest sequence number
   seen to avoid a replay attack.  Since the starting sequence number
   might be arbitrarily large, the modulo operation is required to
   accommodate sequence number roll-over within some key's lifetime.
   This solution draws from TCP's approach [9].

   The sequence number field is chosen to be a 64-bit unsigned quantity.
   This is large enough to avoid exhaustion over the key lifetime.  For
   example, if a key lifetime was conservatively defined as one year,
   there would be enough sequence number values to send RSVP messages at
   an average rate of about 585 gigaMessages per second.  A 32-bit
   sequence number would limit this average rate to about 136 messages
   per second.

   The ability to generate unique monotonically increasing sequence
   numbers across a failure and restart implies some form of stable
   storage, either local to the device or remotely over the network.
   Three sequence number generation procedures are described below.

3.1.  Simple Sequence Numbers

   The most straightforward approach is to generate a unique sequence
   number using a message counter.  Each time a message is transmitted
   for a given key, the sequence number counter is incremented.  The
   current value of this counter is continually or periodically saved to
   stable storage.  After a restart, the counter is recovered using this
   stable storage.  If the counter was saved periodically to stable
   storage, the count should be recovered by increasing the saved value
   to be larger than any possible value of the counter at the time of
   the failure.  This can be computed, knowing the interval at which the
   counter was saved to stable storage and incrementing the stored value
   by that amount.

3.2.  Sequence Numbers Based on a Real Time Clock

   Most devices will probably not have the capability to save sequence
   number counters to stable storage for each key.  A more universal
   solution is to base sequence numbers on the stable storage of a real
   time clock.  Many computing devices have a real time clock module
   that includes stable storage of the clock.  These modules generally
   include some form of nonvolatile memory to retain clock information
   in the event of a power failure.

      In this approach, we could use an NTP based timestamp value as the 
   sequence number.  The roll-over period of an NTP timestamp is about
   136 years, much longer than any reasonable lifetime of a key.  In
   addition, the granularity of the NTP timestamp is fine enough to
   allow the generation of an RSVP message every 200 picoseconds for a
   given key.  Many real time clock modules do not have the resolution
   of an NTP timestamp.  In these cases, the least significant bits of
   the sequence number can be generated using a message counter, which
   is reset every clock tick.  For example, when the real time clock
   provides a resolution of 1 second, the 32 least significant bits of
   the sequence number can be generated using a message counter.  The
   remaining 32 bits are filled with the 32 most significant bits of
   the timestamp.  Assuming that the recovery time after failure takes
   longer than one tick of the real time clock, the message counter for
   the low order bits can be safely reset to zero after a restart.
EID 4313 (Verified) is as follows:

Section: 3.2

Original Text:

   In this approach, we could use an NTP based timestamp value as the
   sequence number.  The roll-over period of an NTP timestamp is about
   136 years, much longer than any reasonable lifetime of a key.  In
   addition, the granularity of the NTP timestamp is fine enough to
   allow the generation of an RSVP message every 200 picoseconds for a
   given key.  Many real time clock modules do not have the resolution
   of an NTP timestamp.  In these cases, the least significant bits of
   the timestamp can be generated using a message counter, which is
   reset every clock tick.  For example, when the real time clock
   provides a resolution of 1 second, the 32 least significant bits of
   the sequence number can be generated using a message counter.  The
   remaining 32 bits are filled with the 32 least significant bits of
   the timestamp.  Assuming that the recovery time after failure takes
   longer than one tick of the real time clock, the message counter for
   the low order bits can be safely reset to zero after a restart.

Corrected Text:

   In this approach, we could use an NTP based timestamp value as the
   sequence number.  The roll-over period of an NTP timestamp is about
   136 years, much longer than any reasonable lifetime of a key.  In
   addition, the granularity of the NTP timestamp is fine enough to
   allow the generation of an RSVP message every 200 picoseconds for a
   given key.  Many real time clock modules do not have the resolution
   of an NTP timestamp.  In these cases, the least significant bits of
   the sequence number can be generated using a message counter, which
   is reset every clock tick.  For example, when the real time clock
   provides a resolution of 1 second, the 32 least significant bits of
   the sequence number can be generated using a message counter.  The
   remaining 32 bits are filled with the 32 most significant bits of
   the timestamp.  Assuming that the recovery time after failure takes
   longer than one tick of the real time clock, the message counter for
   the low order bits can be safely reset to zero after a restart.
32 least significant bits of the timestamp will in this case be set to zero.
3.3. Sequence Numbers Based on a Network Recovered Clock If the device does not contain any stable storage of sequence number counters or of a real time clock, it could recover the real time clock from the network using NTP. Once the clock has been recovered following a restart, the sequence number generation procedure would be identical to the procedure described above. 4. Message Processing Implementations SHOULD allow specification of interfaces that are to be secured, for either sending messages, or receiving them, or both. The sender must ensure that all RSVP messages sent on secured sending interfaces include an INTEGRITY object, generated using the appropriate Key. Receivers verify whether RSVP messages, except of the type "Integrity Challenge" (Section 4.3), arriving on a secured receiving interface contain the INTEGRITY object. If the INTEGRITY object is absent, the receiver discards the message. Security associations are simplex - the keys that a sending system uses to sign its messages may be different from the keys that its receivers use to sign theirs. Hence, each association is associated with a unique sending system and (possibly) multiple receiving systems. Each sender SHOULD have distinct security associations (and keys) per secured sending interface (or LIH). While administrators may configure all the routers and hosts on a subnet (or for that matter, in their network) using a single security association, implementations MUST assume that each sender may send using a distinct security association on each secured interface. At the sender, security association selection is based on the interface through which the message is sent. This selection MAY include additional criteria, such as the destination address (when sending the message unicast, over a broadcast LAN with a large number of hosts) or user identities at the sender or receivers [2]. Finally, all intended message recipients should participate in this security association. Route flaps in a non RSVP cloud might cause messages for the same receiver to be sent on different interfaces at different times. In such cases, the receivers should participate in all possible security associations that may be selected for the interfaces through which the message might be sent. Receivers select keys based on the Key Identifier and the sending system's IP address. The Key Identifier is included in the INTEGRITY object. The sending system's address can be obtained either from the RSVP_HOP object, or if that's not present (as is the case with PathErr and ResvConf messages) from the IP source address. Since the Key Identifier is unique for a sender, this method uniquely identifies the key. The integrity mechanism slightly modifies the processing rules for RSVP messages, both when including the INTEGRITY object in a message sent over a secured sending interface and when accepting a message received on a secured receiving interface. These modifications are detailed below. 4.1. Message Generation For an RSVP message sent over a secured sending interface, the message is created as described in [1], with these exceptions: (1) The RSVP checksum field is set to zero. If required, an RSVP checksum can be calculated when the processing of the INTEGRITY object is complete. (2) The INTEGRITY object is inserted in the appropriate place, and its location in the message is remembered for later use. (3) The sending interface and other appropriate criteria (as mentioned above) are used to determine the Authentication Key and the hash algorithm to be used. (4) The unused flags and the reserved field in the INTEGRITY object MUST be set to 0. The Handshake Flag (HF) should be set according to rules specified in Section 2.1. (5) The sending sequence number MUST be updated to ensure a unique, monotonically increasing number. It is then placed in the Sequence Number field of the INTEGRITY object. (6) The Keyed Message Digest field is set to zero. (7) The Key Identifier is placed into the INTEGRITY object. (8) An authenticating digest of the message is computed using the Authentication Key in conjunction with the keyed-hash algorithm. When the HMAC-MD5 algorithm is used, the hash calculation is described in [7]. (9) The digest is written into the Cryptographic Digest field of the INTEGRITY object. 4.2. Message Reception When the message is received on a secured receiving interface, and is not of the type "Integrity Challenge", it is processed in the following manner: (1) The RSVP checksum field is saved and the field is subsequently set to zero. (2) The Cryptographic Digest field of the INTEGRITY object is saved and the field is subsequently set to zero. (3) The Key Identifier field and the sending system address are used to uniquely determine the Authentication Key and the hash algorithm to be used. Processing of this packet might be delayed when the Key Management System (Appendix 1) is queried for this information. (4) A new keyed-digest is calculated using the indicated algorithm and the Authentication Key. (5) If the calculated digest does not match the received digest, the message is discarded without further processing. (6) If the message is of type "Integrity Response", verify that the CHALLENGE object identically matches the originated challenge. If it matches, save the sequence number in the INTEGRITY object as the largest sequence number received to date. Otherwise, for all other RSVP Messages, the sequence number is validated to prevent replay attacks, and messages with invalid sequence numbers are ignored by the receiver. When a message is accepted, the sequence number of that message could update a stored value corresponding to the largest sequence number received to date. Each subsequent message must then have a larger (modulo 2^64) sequence number to be accepted. This simple processing rule prevents message replay attacks, but it must be modified to tolerate limited out-of-order message delivery. For example, if several messages were sent in a burst (in a periodic refresh generated by a router, or as a result of a tear down function), they might get reordered and then the sequence numbers would not be received in an increasing order. An implementation SHOULD allow administrative configuration that sets the receiver's tolerance to out-of-order message delivery. A simple approach would allow administrators to specify a message window corresponding to the worst case reordering behavior. For example, one might specify that packets reordered within a 32 message window would be accepted. If no reordering can occur, the window is set to one. The receiver must store a list of all sequence numbers seen within the reordering window. A received sequence number is valid if (a) it is greater than the maximum sequence number received or (b) it is a past sequence number lying within the reordering window and not recorded in the list. Acceptance of a sequence number implies adding it to the list and removing a number from the lower end of the list. Messages received with sequence numbers lying below the lower end of the list or marked seen in the list are discarded. When an "Integrity Challenge" message is received on a secured sending interface it is processed in the following manner: (1) An "Integrity Response" message is formed using the Challenge object received in the challenge message. (2) The message is sent back to the receiver, based on the source IP address of the challenge message, using the "Message Generation" steps outlined above. The selection of the Authentication Key and the hash algorithm to be used is determined by the key identifier supplied in the challenge message. 4.3. Integrity Handshake at Restart or Initialization of the Receiver To obtain the starting sequence number for a live Authentication Key, the receiver MAY initiate an integrity handshake with the sender. This handshake consists of a receiver's Challenge and the sender's Response, and may be either initiated during restart or postponed until a message signed with that key arrives. Once the receiver has decided to initiate an integrity handshake for a particular Authentication Key, it identifies the sender using the sending system's address configured in the corresponding security association. The receiver then sends an RSVP Integrity Challenge message to the sender. This message contains the Key Identifier to identify the sender's key and MUST have a unique challenge cookie that is based on a local secret to prevent guessing. see Section 2.5.3 of [4]). It is suggested that the cookie be an MD5 hash of a local secret and a timestamp to provide uniqueness (see Section 9). An RSVP Integrity Challenge message will carry a message type of 11. The message format is as follows: <Integrity Challenge message> ::= <Common Header> <CHALLENGE> he CHALLENGE object has the following format: CHALLENGE Object: Class = 64, C-Type = 1 +-------------+-------------+-------------+-------------+ | 0 (Reserved) | | +-------------+-------------+ + | Key Identifier | +-------------+-------------+-------------+-------------+ | Challenge Cookie | | | +-------------+-------------+-------------+-------------+ The sender accepts the "Integrity Challenge" without doing an integrity check. It returns an RSVP "Integrity Response" message that contains the original CHALLENGE object. It also includes an INTEGRITY object, signed with the key specified by the Key Identifier included in the "Integrity Challenge". An RSVP Integrity Response message will carry a message type of 12. The message format is as follows: <Integrity Response message> ::= <Common Header> <INTEGRITY> <CHALLENGE> The "Integrity Response" message is accepted by the receiver (challenger) only if the returned CHALLENGE object matches the one sent in the "Integrity Challenge" message. This prevents replay of old "Integrity Response" messages. If the match is successful, the receiver saves the Sequence Number from the INTEGRITY object as the latest sequence number received with the key identifier included in the CHALLENGE. If a response is not received within a given period of time, the challenge is repeated. When the integrity handshake successfully completes, the receiver begins accepting normal RSVP signaling messages from that sender and ignores any other "Integrity Response" messages. The Handshake Flag (HF) is used to allow implementations the flexibility of not including the integrity handshake mechanism. By setting this flag to 1, message senders that implement the integrity handshake distinguish themselves from those that do not. Receivers SHOULD NOT attempt to handshake with senders whose INTEGRITY object has HF = 0. An integrity handshake may not be necessary in all environments. A common use of RSVP integrity will be between peering domain routers, which are likely to be processing a steady stream of RSVP messages due to aggregation effects. When a router restarts after a crash, valid RSVP messages from peering senders will probably arrive within a short time. Assuming that replay messages are injected into the stream of valid RSVP messages, there may be only a small window of opportunity for a replay attack before a valid message is processed. This valid message will set the largest sequence number seen to a value greater than any number that had been stored prior to the crash, preventing any further replays. On the other hand, not using an integrity handshake could allow exposure to replay attacks if there is a long period of silence from a given sender following a restart of a receiver. Hence, it SHOULD be an administrative decision whether or not the receiver performs an integrity handshake with senders that are willing to respond to "Integrity Challenge" messages, and whether it accepts any messages from senders that refuse to do so. These decisions will be based on assumptions related to a particular network environment. 5. Key Management It is likely that the IETF will define a standard key management protocol. It is strongly desirable to use that key management protocol to distribute RSVP Authentication Keys among communicating RSVP implementations. Such a protocol would provide scalability and significantly reduce the human administrative burden. The Key Identifier can be used as a hook between RSVP and such a future protocol. Key management protocols have a long history of subtle flaws that are often discovered long after the protocol was first described in public. To avoid having to change all RSVP implementations should such a flaw be discovered, integrated key management protocol techniques were deliberately omitted from this specification. 5.1. Key Management Procedures Each key has a lifetime associated with it that is recorded in all systems (sender and receivers) configured with that key. The concept of a "key lifetime" merely requires that the earliest (KeyStartValid) and latest (KeyEndValid) times that the key is valid be programmable in a way the system understands. Certain key generation mechanisms, such as Kerberos or some public key schemes, may directly produce ephemeral keys. In this case, the lifetime of the key is implicitly defined as part of the key. In general, no key is ever used outside its lifetime (but see Section 5.3). Possible mechanisms for managing key lifetime include the Network Time Protocol and hardware time-of-day clocks. To maintain security, it is advisable to change the RSVP Authentication Key on a regular basis. It should be possible to switch the RSVP Authentication Key without loss of RSVP state or denial of reservation service, and without requiring people to change all the keys at once. This requires an RSVP implementation to support the storage and use of more than one active RSVP Authentication Key at the same time. Hence both the sender and receivers might have multiple active keys for a given security association. Since keys are shared between a sender and (possibly) multiple receivers, there is a region of uncertainty around the time of key switch-over during which some systems may still be using the old key and others might have switched to the new key. The size of this uncertainty region is related to clock synchrony of the systems. Administrators should configure the overlap between the expiration time of the old key (KeyEndValid) and the validity of the new key (KeyStartValid) to be at least twice the size of this uncertainty interval. This will allow the sender to make the key switch-over at the midpoint of this interval and be confident that all receivers are now accepting the new key. For the duration of the overlap in key lifetimes, a receiver must be prepared to authenticate messages using either key. During a key switch-over, it will be necessary for each receiver to handshake with the sender using the new key. As stated before, a receiver has the choice of initiating a handshake during the switchover or postponing the handshake until the receipt of a message using that key. 5.2. Key Management Requirements Requirements on an implementation are as follows: o It is strongly desirable that a hypothetical security breach in one Internet protocol not automatically compromise other Internet protocols. The Authentication Key of this specification SHOULD NOT be stored using protocols or algorithms that have known flaws. o An implementation MUST support the storage and use of more than one key at the same time, for both sending and receiving systems. o An implementation MUST associate a specific lifetime (i.e., KeyStartValid and KeyEndValid) with each key and the corresponding Key Identifier. o An implementation MUST support manual key distribution (e.g., the privileged user manually typing in the key, key lifetime, and key identifier on the console). The lifetime may be infinite. o If more than one algorithm is supported, then the implementation MUST require that the algorithm be specified for each key at the time the other key information is entered. o Keys that are out of date MAY be automatically deleted by the implementation. o Manual deletion of active keys MUST also be supported. o Key storage SHOULD persist across a system restart, warm or cold, to ease operational usage. 5.3. Pathological Case It is possible that the last key for a given security association has expired. When this happens, it is unacceptable to revert to an unauthenticated condition, and not advisable to disrupt current reservations. Therefore, the system should send a "last authentication key expiration" notification to the network manager and treat the key as having an infinite lifetime until the lifetime is extended, the key is deleted by network management, or a new key is configured. 6. Conformance Requirements To conform to this specification, an implementation MUST support all of its aspects. The HMAC-MD5 authentication algorithm defined in [7] MUST be implemented by all conforming implementations. A conforming implementation MAY also support other authentication algorithms such as NIST's Secure Hash Algorithm (SHA). Manual key distribution as described above MUST be supported by all conforming implementations. All implementations MUST support the smooth key roll over described under "Key Management Procedures." Implementations SHOULD support a standard key management protocol for secure distribution of RSVP Authentication Keys once such a key management protocol is standardized by the IETF. 7. Kerberos generation of RSVP Authentication Keys Kerberos[10] MAY be used to generate the RSVP Authentication key used in generating a signature in the Integrity Object sent from a RSVP sender to a receiver. Kerberos key generation avoids the use of shared keys between RSVP senders and receivers such as hosts and routers. Kerberos allows for the use of trusted third party keying relationships between security principals (RSVP sender and receivers) where the Kerberos key distribution center(KDC) establishes an ephemeral session key that is subsequently shared between RSVP sender and receivers. In the multicast case all receivers of a multicast RSVP message MUST share a single key with the KDC (e.g. the receivers are in effect the same security principal with respect to Kerberos). The Key information determined by the sender MAY specify the use of Kerberos in place of configured shared keys as the mechanism for establishing a key between the sender and receiver. The Kerberos identity of the receiver is established as part of the sender's interface configuration or it can be established through other mechanisms. When generating the first RSVP message for a specific key identifier the sender requests a Kerberos service ticket and gets back an ephemeral session key and a Kerberos ticket from the KDC. The sender encapsulates the ticket and the identity of the sender in an Identity Policy Object[2]. The sender includes the Policy Object in the RSVP message. The session key is then used by the sender as the RSVP Authentication key in section 4.1 step (3) and is stored as Key information associated with the key identifier. Upon RSVP Message reception, the receiver retrieves the Kerberos Ticket from the Identity Policy Object, decrypts the ticket and retrieves the session key from the ticket. The session key is the same key as used by the sender and is used as the key in section 4.2 step (3). The receiver stores the key for use in processing subsequent RSVP messages. Kerberos tickets have lifetimes and the sender MUST NOT use tickets that have expired. A new ticket MUST be requested and used by the sender for the receiver prior to the ticket expiring. 7.1. Optimization when using Kerberos Based Authentication Kerberos tickets are relatively long (> 500 bytes) and it is not necessary to send a ticket in every RSVP message. The ephemeral session key can be cached by the sender and receiver and can be used for the lifetime of the Kerberos ticket. In this case, the sender only needs to include the Kerberos ticket in the first Message generated. Subsequent RSVP messages use the key identifier to retrieve the cached key (and optionally other identity information) instead of passing tickets from sender to receiver in each RSVP message. A receiver may not have cached key state with an associated Key Identifier due to reboot or route changes. If the receiver's policy indicates the use of Kerberos keys for integrity checking, the receiver can send an integrity Challenge message back to the sender. Upon receiving an integrity Challenge message a sender MUST send an Identity object that includes the Kerberos ticket in the integrity Response message, thereby allowing the receiver to retrieve and store the session key from the Kerberos ticket for subsequent Integrity checking. 8. Acknowledgments This document is derived directly from similar work done for OSPF and RIP Version II, jointly by Ran Atkinson and Fred Baker. Significant editing was done by Bob Braden, resulting in increased clarity. Significant comments were submitted by Steve Bellovin, who actually understands this stuff. Matt Crawford and Dan Harkins helped revise the document. 9. References [1] Braden, R., Zhang, L., Berson, S., Herzog, S. and S. Jamin, "Resource ReSerVation Protocol (RSVP) -- Version 1 Functional Specification", RFC 2205, September 1997. [2] Yadav, S., et al., "Identity Representation for RSVP", RFC 2752, January 2000. [3] Atkinson, R. and S. Kent, "Security Architecture for the Internet Protocol", RFC 2401, November 1998. [4] Maughan, D., Schertler, M., Schneider, M. and J. Turner, "Internet Security Association and Key Management Protocol (ISAKMP)", RFC 2408, November 1998. [5] Kent, S. and R. Atkinson, "IP Authentication Header", RFC 2402, November 1998. [6] Kent, S. and R. Atkinson, "IP Encapsulating Security Payload (ESP)", RFC 2406, November 1998. [7] Krawczyk, H., Bellare, M. and R. Canetti, "HMAC: Keyed-Hashing for Message Authentication", RFC 2104, March 1996. [8] Bradner, S., "Key words for use in RFCs to Indicate Requirement Levels", BCP 14, RFC 2119, March 1997. [9] Postel, J., "Transmission Control Protocol", STD 7, RFC 793, September 1981. [10] Kohl, J. and C. Neuman, "The Kerberos Network Authentication Service (V5)", RFC 1510, September 1993. 10. Security Considerations This entire memo describes and specifies an authentication mechanism for RSVP that is believed to be secure against active and passive attacks. The quality of the security provided by this mechanism depends on the strength of the implemented authentication algorithms, the strength of the key being used, and the correct implementation of the security mechanism in all communicating RSVP implementations. This mechanism also depends on the RSVP Authentication Keys being kept confidential by all parties. If any of these assumptions are incorrect or procedures are insufficiently secure, then no real security will be provided to the users of this mechanism. While the handshake "Integrity Response" message is integrity- checked, the handshake "Integrity Challenge" message is not. This was done intentionally to avoid the case when both peering routers do not have a starting sequence number for each other's key. Consequently, they will each keep sending handshake "Integrity Challenge" messages that will be dropped by the other end. Moreover, requiring only the response to be integrity-checked eliminates a dependency on an security association in the opposite direction. This, however, lets an intruder generate fake handshaking challenges with a certain challenge cookie. It could then save the response and attempt to play it against a receiver that is in recovery. If it was lucky enough to have guessed the challenge cookie used by the receiver at recovery time it could use the saved response. This response would be accepted, since it is properly signed, and would have a smaller sequence number for the sender because it was an old message. This opens the receiver up to replays. Still, it seems very difficult to exploit. It requires not only guessing the challenge cookie (which is based on a locally known secret) in advance, but also being able to masquerade as the receiver to generate a handshake "Integrity Challenge" with the proper IP address and not being caught. Confidentiality is not provided by this mechanism. If confidentiality is required, IPSEC ESP [6] may be the best approach, although it is subject to the same criticisms as IPSEC Authentication, and therefore would be applicable only in specific environments. Protection against traffic analysis is also not provided. Mechanisms such as bulk link encryption might be used when protection against traffic analysis is required. 11. Authors' Addresses Fred Baker Cisco Systems 519 Lado Drive Santa Barbara, CA 93111 Phone: (408) 526-4257 EMail: Bob Lindell USC Information Sciences Institute 4676 Admiralty Way Marina del Rey, CA 90292 Phone: (310) 822-1511 EMail: lindell@ISI.EDU Mohit Talwar Microsoft Corporation One Microsoft Way Redmond, WA 98052 Phone: +1 425 705 3131 EMail: 12. Appendix 1: Key Management Interface This appendix describes a generic interface to Key Management. This description is at an abstract level realizing that implementations may need to introduce small variations to the actual interface. At the start of execution, RSVP would use this interface to obtain the current set of relevant keys for sending and receiving messages. During execution, RSVP can query for specific keys given a Key Identifier and Source Address, discover newly created keys, and be informed of those keys that have been deleted. The interface provides both a polling and asynchronous upcall style for wider applicability. 12.1. Data Structures Information about keys is returned using the following KeyInfo data structure: KeyInfo { Key Type (Send or Receive) KeyIdentifier Key Authentication Algorithm Type and Mode KeyStartValid KeyEndValid Status (Active or Deleted) Outgoing Interface (for Send only) Other Outgoing Security Association Selection Criteria (for Send only, optional) Sending System Address (for Receive Only) } 12.2. Default Key Table This function returns a list of KeyInfo data structures corresponding to all of the keys that are configured for sending and receiving RSVP messages and have an Active Status. This function is usually called at the start of execution but there is no limit on the number of times that it may be called. KM_DefaultKeyTable() -> KeyInfoList 12.3. Querying for Unknown Receive Keys When a message arrives with an unknown Key Identifier and Sending System Address pair, RSVP can use this function to query the Key Management System for the appropriate key. The status of the element returned, if any, must be Active. KM_GetRecvKey( INTEGRITY Object, SrcAddress ) -> KeyInfo 12.4. Polling for Updates This function returns a list of KeyInfo data structures corresponding to any incremental changes that have been made to the default key table or requested keys since the last call to either KM_KeyTablePoll, KM_DefaultKeyTable, or KM_GetRecvKey. The status of some elements in the returned list may be set to Deleted. KM_KeyTablePoll() -> KeyInfoList 12.5. Asynchronous Upcall Interface Rather than repeatedly calling the KM_KeyTablePoll(), an implementation may choose to use an asynchronous event model. This function registers interest to key changes for a given Key Identifier or for all keys if no Key Identifier is specified. The upcall function is called each time a change is made to a key. KM_KeyUpdate ( Function [, KeyIdentifier ] ) where the upcall function is parameterized as follows: Function ( KeyInfo ) 13. Full Copyright Statement Copyright (C) The Internet Society (2000). All Rights Reserved. This document and translations of it may be copied and furnished to others, and derivative works that comment on or otherwise explain it or assist in its implementation may be prepared, copied, published and distributed, in whole or in part, without restriction of any kind, provided that the above copyright notice and this paragraph are included on all such copies and derivative works. However, this document itself may not be modified in any way, such as by removing the copyright notice or references to the Internet Society or other Internet organizations, except as needed for the purpose of developing Internet standards in which case the procedures for copyrights defined in the Internet Standards process must be followed, or as required to translate it into languages other than English. The limited permissions granted above are perpetual and will not be revoked by the Internet Society or its successors or assigns. This document and the information contained herein is provided on an "AS IS" basis and THE INTERNET SOCIETY AND THE INTERNET ENGINEERING TASK FORCE DISCLAIMS ALL WARRANTIES, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO ANY WARRANTY THAT THE USE OF THE INFORMATION HEREIN WILL NOT INFRINGE ANY RIGHTS OR ANY IMPLIED WARRANTIES OF MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE. Acknowledgement Funding for the RFC Editor function is currently provided by the Internet Society.