2. IKE Protocol Details and Variations
IKE normally listens and sends on UDP port 500, though IKE messages may also be received on UDP port 4500 with a slightly different format (see Section 2.23). Since UDP is a datagram (unreliable) protocol, IKE includes in its definition recovery from transmission errors, including packet loss, packet replay, and packet forgery. IKE is designed to function so long as (1) at least one of a series of retransmitted packets reaches its destination before timing out; and (2) the channel is not so full of forged and replayed packets so as to exhaust the network or CPU capacities of either endpoint. Even in the absence of those minimum performance requirements, IKE is designed to fail cleanly (as though the network were broken).
Although IKEv2 messages are intended to be short, they contain structures with no hard upper bound on size (in particular, digital certificates), and IKEv2 itself does not have a mechanism for fragmenting large messages. IP defines a mechanism for fragmentation of oversized UDP messages, but implementations vary in the maximum message size supported. Furthermore, use of IP fragmentation opens an implementation to denial-of-service (DoS) attacks [DOSUDPPROT]. Finally, some NAT and/or firewall implementations may block IP fragments.
All IKEv2 implementations MUST be able to send, receive, and process IKE messages that are up to 1280 octets long, and they SHOULD be able to send, receive, and process messages that are up to 3000 octets long. IKEv2 implementations need to be aware of the maximum UDP message size supported and MAY shorten messages by leaving out some certificates or cryptographic suite proposals if that will keep messages below the maximum. Use of the "Hash and URL" formats rather than including certificates in exchanges where possible can avoid most problems. Implementations and configuration need to keep in mind, however, that if the URL lookups are possible only after the Child SA is established, recursion issues could prevent this technique from working.
The UDP payload of all packets containing IKE messages sent on port 4500 MUST begin with the prefix of four zeros; otherwise, the receiver won't know how to handle them.
2.1. Use of Retransmission Timers
All messages in IKE exist in pairs: a request and a response. The setup of an IKE SA normally consists of two exchanges. Once the IKE SA is set up, either end of the Security Association may initiate requests at any time, and there can be many requests and responses "in flight" at any given moment. But each message is labeled as either a request or a response, and for each exchange, one end of the Security Association is the initiator and the other is the responder.
For every pair of IKE messages, the initiator is responsible for retransmission in the event of a timeout. The responder MUST never retransmit a response unless it receives a retransmission of the request. In that event, the responder MUST ignore the retransmitted request except insofar as it causes a retransmission of the response. The initiator MUST remember each request until it receives the corresponding response. The responder MUST remember each response until it receives a request whose sequence number is larger than or equal to the sequence number in the response plus its window size (see Section 2.3). In order to allow saving memory, responders are allowed to forget the response after a timeout of several minutes. If the responder receives a retransmitted request for which it has already forgotten the response, it MUST ignore the request (and not, for example, attempt constructing a new response).
IKE is a reliable protocol: the initiator MUST retransmit a request until it either receives a corresponding response or deems the IKE SA to have failed. In the latter case, the initiator discards all state associated with the IKE SA and any Child SAs that were negotiated using that IKE SA. A retransmission from the initiator MUST be bitwise identical to the original request. That is, everything starting from the IKE header (the IKE SA initiator's SPI onwards) must be bitwise identical; items before it (such as the IP and UDP headers) do not have to be identical.
Retransmissions of the IKE_SA_INIT request require some special handling. When a responder receives an IKE_SA_INIT request, it has to determine whether the packet is a retransmission belonging to an existing "half-open" IKE SA (in which case the responder retransmits the same response), or a new request (in which case the responder creates a new IKE SA and sends a fresh response), or it belongs to an existing IKE SA where the IKE_AUTH request has been already received (in which case the responder ignores it).
It is not sufficient to use the initiator's SPI and/or IP address to differentiate between these three cases because two different peers behind a single NAT could choose the same initiator SPI. Instead, a robust responder will do the IKE SA lookup using the whole packet, its hash, or the Ni payload.
The retransmission policy for one-way messages is somewhat different from that for regular messages. Because no acknowledgement is ever sent, there is no reason to gratuitously retransmit one-way messages. Given that all these messages are errors, it makes sense to send them only once per "offending" packet, and only retransmit if further offending packets are received. Still, it also makes sense to limit retransmissions of such error messages.
2.2. Use of Sequence Numbers for Message ID
Every IKE message contains a Message ID as part of its fixed header. This Message ID is used to match up requests and responses and to identify retransmissions of messages. Retransmission of a message MUST use the same Message ID as the original message.
The Message ID is a 32-bit quantity, which is zero for the IKE_SA_INIT messages (including retries of the message due to responses such as COOKIE and INVALID_KE_PAYLOAD), and incremented for each subsequent exchange. Thus, the first pair of IKE_AUTH messages will have an ID of 1, the second (when EAP is used) will be 2, and so on. The Message ID is reset to zero in the new IKE SA after the IKE SA is rekeyed.
Each endpoint in the IKE Security Association maintains two "current" Message IDs: the next one to be used for a request it initiates and the next one it expects to see in a request from the other end. These counters increment as requests are generated and received. Responses always contain the same Message ID as the corresponding request. That means that after the initial exchange, each integer n may appear as the Message ID in four distinct messages: the nth request from the original IKE initiator, the corresponding response, the nth request from the original IKE responder, and the corresponding response. If the two ends make a very different number of requests, the Message IDs in the two directions can be very different. There is no ambiguity in the messages, however, because the Initiator and Response flags in the message header specify which of the four messages a particular one is.
Throughout this document, "initiator" refers to the party who initiated the exchange being described. The "original initiator" always refers to the party who initiated the exchange that resulted in the current IKE SA. In other words, if the "original responder" starts rekeying the IKE SA, that party becomes the "original initiator" of the new IKE SA.
Note that Message IDs are cryptographically protected and provide protection against message replays. In the unlikely event that Message IDs grow too large to fit in 32 bits, the IKE SA MUST be closed or rekeyed.
2.3. Window Size for Overlapping Requests
The SET_WINDOW_SIZE notification asserts that the sending endpoint is capable of keeping state for multiple outstanding exchanges, permitting the recipient to send multiple requests before getting a response to the first. The data associated with a SET_WINDOW_SIZE notification MUST be 4 octets long and contain the big endian representation of the number of messages the sender promises to keep. The window size is always one until the initial exchanges complete.
An IKE endpoint MUST wait for a response to each of its messages before sending a subsequent message unless it has received a SET_WINDOW_SIZE Notify message from its peer informing it that the peer is prepared to maintain state for multiple outstanding messages in order to allow greater throughput.
After an IKE SA is set up, in order to maximize IKE throughput, an IKE endpoint MAY issue multiple requests before getting a response to any of them, up to the limit set by its peer's SET_WINDOW_SIZE. These requests may pass one another over the network. An IKE endpoint MUST be prepared to accept and process a request while it has a request outstanding in order to avoid a deadlock in this situation. An IKE endpoint may also accept and process multiple requests while it has a request outstanding.
An IKE endpoint MUST NOT exceed the peer's stated window size for transmitted IKE requests. In other words, if the responder stated its window size is N, then when the initiator needs to make a request X, it MUST wait until it has received responses to all requests up through request X-N. An IKE endpoint MUST keep a copy of (or be able to regenerate exactly) each request it has sent until it receives the corresponding response. An IKE endpoint MUST keep a copy of (or be able to regenerate exactly) the number of previous responses equal to its declared window size in case its response was lost and the initiator requests its retransmission by retransmitting the request.
An IKE endpoint supporting a window size greater than one ought to be capable of processing incoming requests out of order to maximize performance in the event of network failures or packet reordering.
The window size is normally a (possibly configurable) property of a particular implementation, and is not related to congestion control (unlike the window size in TCP, for example). In particular, what the responder should do when it receives a SET_WINDOW_SIZE notification containing a smaller value than is currently in effect is not defined. Thus, there is currently no way to reduce the window size of an existing IKE SA; you can only increase it. When rekeying an IKE SA, the new IKE SA starts with window size 1 until it is explicitly increased by sending a new SET_WINDOW_SIZE notification.
The INVALID_MESSAGE_ID notification is sent when an IKE Message ID outside the supported window is received. This Notify message MUST NOT be sent in a response; the invalid request MUST NOT be acknowledged. Instead, inform the other side by initiating an INFORMATIONAL exchange with Notification Data containing the four-octet invalid Message ID. Sending this notification is OPTIONAL, and notifications of this type MUST be rate limited.
2.4. State Synchronization and Connection Timeouts
An IKE endpoint is allowed to forget all of its state associated with an IKE SA and the collection of corresponding Child SAs at any time. This is the anticipated behavior in the event of an endpoint crash and restart. It is important when an endpoint either fails or reinitializes its state that the other endpoint detect those conditions and not continue to waste network bandwidth by sending packets over discarded SAs and having them fall into a black hole.
The INITIAL_CONTACT notification asserts that this IKE SA is the only IKE SA currently active between the authenticated identities. It MAY be sent when an IKE SA is established after a crash, and the recipient MAY use this information to delete any other IKE SAs it has to the same authenticated identity without waiting for a timeout. This notification MUST NOT be sent by an entity that may be replicated (e.g., a roaming user's credentials where the user is allowed to connect to the corporate firewall from two remote systems at the same time). The INITIAL_CONTACT notification, if sent, MUST be in the first IKE_AUTH request or response, not as a separate exchange afterwards; receiving parties MAY ignore it in other messages.
Since IKE is designed to operate in spite of DoS attacks from the network, an endpoint MUST NOT conclude that the other endpoint has failed based on any routing information (e.g., ICMP messages) or IKE messages that arrive without cryptographic protection (e.g., Notify messages complaining about unknown SPIs). An endpoint MUST conclude that the other endpoint has failed only when repeated attempts to contact it have gone unanswered for a timeout period or when a cryptographically protected INITIAL_CONTACT notification is received on a different IKE SA to the same authenticated identity. An endpoint should suspect that the other endpoint has failed based on routing information and initiate a request to see whether the other endpoint is alive. To check whether the other side is alive, IKE specifies an empty INFORMATIONAL request that (like all IKE requests) requires an acknowledgement (note that within the context of an IKE SA, an "empty" message consists of an IKE header followed by an Encrypted payload that contains no payloads). If a cryptographically protected (fresh, i.e., not retransmitted) message has been received from the other side recently, unprotected Notify messages MAY be ignored. Implementations MUST limit the rate at which they take actions based on unprotected messages.
The number of retries and length of timeouts are not covered in this specification because they do not affect interoperability. It is suggested that messages be retransmitted at least a dozen times over a period of at least several minutes before giving up on an SA, but different environments may require different rules. To be a good network citizen, retransmission times MUST increase exponentially to avoid flooding the network and making an existing congestion situation worse. If there has only been outgoing traffic on all of the SAs associated with an IKE SA, it is essential to confirm liveness of the other endpoint to avoid black holes. If no cryptographically protected messages have been received on an IKE SA or any of its Child SAs recently, the system needs to perform a liveness check in order to prevent sending messages to a dead peer. (This is sometimes called "dead peer detection" or "DPD", although it is really detecting live peers, not dead ones.) Receipt of a fresh cryptographically protected message on an IKE SA or any of its Child SAs ensures liveness of the IKE SA and all of its Child SAs. Note that this places requirements on the failure modes of an IKE endpoint. An implementation needs to stop sending over any SA if some failure prevents it from receiving on all of the associated SAs. If a system creates Child SAs that can fail independently from one another without the associated IKE SA being able to send a delete message, then the system MUST negotiate such Child SAs using separate IKE SAs.
One type of DoS attack on the initiator of an IKE SA can be avoided if the initiator takes proper care: since the first two messages of an SA setup are not cryptographically protected, an attacker could respond to the initiator's message before the genuine responder and poison the connection setup attempt. To prevent this, the initiator MAY be willing to accept multiple responses to its first message, treat each response as potentially legitimate, respond to each one, and then discard all the invalid half-open connections when it receives a valid cryptographically protected response to any one of its requests. Once a cryptographically valid response is received, all subsequent responses should be ignored whether or not they are cryptographically valid.
Note that with these rules, there is no reason to negotiate and agree upon an SA lifetime. If IKE presumes the partner is dead, based on repeated lack of acknowledgement to an IKE message, then the IKE SA and all Child SAs set up through that IKE SA are deleted.
An IKE endpoint may at any time delete inactive Child SAs to recover resources used to hold their state. If an IKE endpoint chooses to delete Child SAs, it MUST send Delete payloads to the other end notifying it of the deletion. It MAY similarly time out the IKE SA. Closing the IKE SA implicitly closes all associated Child SAs. In this case, an IKE endpoint SHOULD send a Delete payload indicating that it has closed the IKE SA unless the other endpoint is no longer responding.
2.5. Version Numbers and Forward Compatibility
This document describes version 2.0 of IKE, meaning the major version number is 2 and the minor version number is 0. This document is a replacement for [IKEV2]. It is likely that some implementations will want to support version 1.0 and version 2.0, and in the future, other versions.
The major version number should be incremented only if the packet formats or required actions have changed so dramatically that an older version node would not be able to interoperate with a newer version node if it simply ignored the fields it did not understand and took the actions specified in the older specification. The minor version number indicates new capabilities, and MUST be ignored by a node with a smaller minor version number, but used for informational purposes by the node with the larger minor version number. For example, it might indicate the ability to process a newly defined Notify message type. The node with the larger minor version number would simply note that its correspondent would not be able to understand that message and therefore would not send it.
If an endpoint receives a message with a higher major version number, it MUST drop the message and SHOULD send an unauthenticated Notify message of type INVALID_MAJOR_VERSION containing the highest (closest) version number it supports. If an endpoint supports major version n, and major version m, it MUST support all versions between n and m. If it receives a message with a major version that it supports, it MUST respond with that version number. In order to prevent two nodes from being tricked into corresponding with a lower major version number than the maximum that they both support, IKE has a flag that indicates that the node is capable of speaking a higher major version number.
Thus, the major version number in the IKE header indicates the version number of the message, not the highest version number that the transmitter supports. If the initiator is capable of speaking versions n, n+1, and n+2, and the responder is capable of speaking versions n and n+1, then they will negotiate speaking n+1, where the initiator will set a flag indicating its ability to speak a higher version. If they mistakenly (perhaps through an active attacker sending error messages) negotiate to version n, then both will notice that the other side can support a higher version number, and they MUST break the connection and reconnect using version n+1.
Note that IKEv1 does not follow these rules, because there is no way in v1 of noting that you are capable of speaking a higher version number. So an active attacker can trick two v2-capable nodes into speaking v1. When a v2-capable node negotiates down to v1, it should note that fact in its logs.
Also, for forward compatibility, all fields marked RESERVED MUST be set to zero by an implementation running version 2.0, and their content MUST be ignored by an implementation running version 2.0 ("Be conservative in what you send and liberal in what you receive" [IP]). In this way, future versions of the protocol can use those fields in a way that is guaranteed to be ignored by implementations that do not understand them. Similarly, payload types that are not defined are reserved for future use; implementations of a version where they are undefined MUST skip over those payloads and ignore their contents.
IKEv2 adds a "critical" flag to each payload header for further flexibility for forward compatibility. If the critical flag is set and the payload type is unrecognized, the message MUST be rejected and the response to the IKE request containing that payload MUST include a Notify payload UNSUPPORTED_CRITICAL_PAYLOAD, indicating an unsupported critical payload was included. In that Notify payload, the Notification Data contains the one-octet payload type. If the critical flag is not set and the payload type is unsupported, that payload MUST be ignored. Payloads sent in IKE response messages MUST NOT have the critical flag set. Note that the critical flag applies only to the payload type, not the contents. If the payload type is recognized, but the payload contains something that is not (such as an unknown transform inside an SA payload, or an unknown Notify Message Type inside a Notify payload), the critical flag is ignored.
Although new payload types may be added in the future and may appear interleaved with the fields defined in this specification, implementations SHOULD send the payloads defined in this specification in the order shown in the figures in Sections 1 and 2; implementations MUST NOT reject as invalid a message with those payloads in any other order.
2.6. IKE SA SPIs and Cookies
The initial two eight-octet fields in the header, called the "IKE SPIs", are used as a connection identifier at the beginning of IKE packets. Each endpoint chooses one of the two SPIs and MUST choose them so as to be unique identifiers of an IKE SA. An SPI value of zero is special: it indicates that the remote SPI value is not yet known by the sender.
Incoming IKE packets are mapped to an IKE SA only using the packet's SPI, not using (for example) the source IP address of the packet.
Unlike ESP and AH where only the recipient's SPI appears in the header of a message, in IKE the sender's SPI is also sent in every message. Since the SPI chosen by the original initiator of the IKE SA is always sent first, an endpoint with multiple IKE SAs open that wants to find the appropriate IKE SA using the SPI it assigned must look at the Initiator flag in the header to determine whether it assigned the first or the second eight octets.
In the first message of an initial IKE exchange, the initiator will not know the responder's SPI value and will therefore set that field to zero. When the IKE_SA_INIT exchange does not result in the creation of an IKE SA due to INVALID_KE_PAYLOAD, NO_PROPOSAL_CHOSEN, or COOKIE, the responder's SPI will be zero also in the response message. However, if the responder sends a non-zero responder SPI, the initiator should not reject the response for only that reason.
Two expected attacks against IKE are state and CPU exhaustion, where the target is flooded with session initiation requests from forged IP addresses. These attacks can be made less effective if a responder uses minimal CPU and commits no state to an SA until it knows the initiator can receive packets at the address from which it claims to be sending them.
When a responder detects a large number of half-open IKE SAs, it SHOULD reply to IKE_SA_INIT requests with a response containing the COOKIE notification. The data associated with this notification MUST be between 1 and 64 octets in length (inclusive), and its generation is described later in this section. If the IKE_SA_INIT response includes the COOKIE notification, the initiator MUST then retry the IKE_SA_INIT request, and include the COOKIE notification containing the received data as the first payload, and all other payloads unchanged. The initial exchange will then be as follows:
Initiator Responder
-------------------------------------------------------------------
HDR(A,0), SAi1, KEi, Ni -->
<-- HDR(A,0), N(COOKIE)
HDR(A,0), N(COOKIE), SAi1,
KEi, Ni -->
<-- HDR(A,B), SAr1, KEr,
Nr, [CERTREQ]
HDR(A,B), SK {IDi, [CERT,]
[CERTREQ,] [IDr,] AUTH,
SAi2, TSi, TSr} -->
<-- HDR(A,B), SK {IDr, [CERT,]
AUTH, SAr2, TSi, TSr}
The first two messages do not affect any initiator or responder state except for communicating the cookie. In particular, the message sequence numbers in the first four messages will all be zero and the message sequence numbers in the last two messages will be one. 'A' is the SPI assigned by the initiator, while 'B' is the SPI assigned by the responder.
An IKE implementation can implement its responder cookie generation in such a way as to not require any saved state to recognize its valid cookie when the second IKE_SA_INIT message arrives. The exact algorithms and syntax used to generate cookies do not affect interoperability and hence are not specified here. The following is an example of how an endpoint could use cookies to implement limited DoS protection.
A good way to do this is to set the responder cookie to be:
Cookie = <VersionIDofSecret> | Hash(Ni | IPi | SPIi | <secret>)
where <secret> is a randomly generated secret known only to the responder and periodically changed and | indicates concatenation. <VersionIDofSecret> should be changed whenever <secret> is regenerated. The cookie can be recomputed when the IKE_SA_INIT arrives the second time and compared to the cookie in the received message. If it matches, the responder knows that the cookie was generated since the last change to <secret> and that IPi must be the same as the source address it saw the first time. Incorporating SPIi into the calculation ensures that if multiple IKE SAs are being set up in parallel they will all get different cookies (assuming the initiator chooses unique SPIi's). Incorporating Ni in the hash ensures that an attacker who sees only message 2 can't successfully forge a message 3. Also, incorporating SPIi in the hash prevents an attacker from fetching one cookie from the other end, and then initiating many IKE_SA_INIT exchanges all with different initiator SPIs (and perhaps port numbers) so that the responder thinks that there are a lot of machines behind one NAT box that are all trying to connect.
If a new value for <secret> is chosen while there are connections in the process of being initialized, an IKE_SA_INIT might be returned with other than the current <VersionIDofSecret>. The responder in that case MAY reject the message by sending another response with a new cookie or it MAY keep the old value of <secret> around for a short time and accept cookies computed from either one. The responder should not accept cookies indefinitely after <secret> is changed, since that would defeat part of the DoS protection. The responder should change the value of <secret> frequently, especially if under attack.
When one party receives an IKE_SA_INIT request containing a cookie whose contents do not match the value expected, that party MUST ignore the cookie and process the message as if no cookie had been included; usually this means sending a response containing a new cookie. The initiator should limit the number of cookie exchanges it tries before giving up, possibly using exponential back-off. An attacker can forge multiple cookie responses to the initiator's IKE_SA_INIT message, and each of those forged cookie replies will cause two packets to be sent: one packet from the initiator to the responder (which will reject those cookies), and one response from responder to initiator that includes the correct cookie.
A note on terminology: the term "cookies" originates with Karn and Simpson [PHOTURIS] in Photuris, an early proposal for key management with IPsec, and it has persisted. The Internet Security Association and Key Management Protocol (ISAKMP) [ISAKMP] fixed message header includes two eight-octet fields called "cookies", and that syntax is used by both IKEv1 and IKEv2, although in IKEv2 they are referred to as the "IKE SPI" and there is a new separate field in a Notify payload holding the cookie.
2.6.1. Interaction of COOKIE and INVALID_KE_PAYLOAD
There are two common reasons why the initiator may have to retry the IKE_SA_INIT exchange: the responder requests a cookie or wants a different Diffie-Hellman group than was included in the KEi payload. If the initiator receives a cookie from the responder, the initiator needs to decide whether or not to include the cookie in only the next retry of the IKE_SA_INIT request, or in all subsequent retries as well.
If the initiator includes the cookie only in the next retry, one additional round trip may be needed in some cases. An additional round trip is needed also if the initiator includes the cookie in all retries, but the responder does not support this. For instance, if the responder includes the KEi payloads in cookie calculation, it will reject the request by sending a new cookie.
If both peers support including the cookie in all retries, a slightly shorter exchange can happen.
Initiator Responder
-----------------------------------------------------------
HDR(A,0), SAi1, KEi, Ni -->
<-- HDR(A,0), N(COOKIE)
HDR(A,0), N(COOKIE), SAi1, KEi, Ni -->
<-- HDR(A,0), N(INVALID_KE_PAYLOAD)
HDR(A,0), N(COOKIE), SAi1, KEi', Ni -->
<-- HDR(A,B), SAr1, KEr, Nr
Implementations SHOULD support this shorter exchange, but MUST NOT fail if other implementations do not support this shorter exchange.
2.7. Cryptographic Algorithm Negotiation
The payload type known as "SA" indicates a proposal for a set of choices of IPsec protocols (IKE, ESP, or AH) for the SA as well as cryptographic algorithms associated with each protocol.
An SA payload consists of one or more proposals. Each proposal includes one protocol. Each protocol contains one or more transforms -- each specifying a cryptographic algorithm. Each transform contains zero or more attributes (attributes are needed only if the Transform ID does not completely specify the cryptographic algorithm).
This hierarchical structure was designed to efficiently encode proposals for cryptographic suites when the number of supported suites is large because multiple values are acceptable for multiple transforms. The responder MUST choose a single suite, which may be any subset of the SA proposal following the rules below.
Each proposal contains one protocol. If a proposal is accepted, the SA response MUST contain the same protocol. The responder MUST accept a single proposal or reject them all and return an error. The error is given in a notification of type NO_PROPOSAL_CHOSEN.
Each IPsec protocol proposal contains one or more transforms. Each transform contains a Transform Type. The accepted cryptographic suite MUST contain exactly one transform of each type included in the proposal. For example: if an ESP proposal includes transforms ENCR_3DES, ENCR_AES w/keysize 128, ENCR_AES w/keysize 256, AUTH_HMAC_MD5, and AUTH_HMAC_SHA, the accepted suite MUST contain one of the ENCR_ transforms and one of the AUTH_ transforms. Thus, six combinations are acceptable.
If an initiator proposes both normal ciphers with integrity protection as well as combined-mode ciphers, then two proposals are needed. One of the proposals includes the normal ciphers with the integrity algorithms for them, and the other proposal includes all the combined-mode ciphers without the integrity algorithms (because combined-mode ciphers are not allowed to have any integrity algorithm other than "NONE").
2.8. Rekeying
IKE, ESP, and AH Security Associations use secret keys that should be used only for a limited amount of time and to protect a limited amount of data. This limits the lifetime of the entire Security Association. When the lifetime of a Security Association expires, the Security Association MUST NOT be used. If there is demand, new Security Associations MAY be established. Reestablishment of Security Associations to take the place of ones that expire is referred to as "rekeying".
To allow for minimal IPsec implementations, the ability to rekey SAs without restarting the entire IKE SA is optional. An implementation MAY refuse all CREATE_CHILD_SA requests within an IKE SA. If an SA has expired or is about to expire and rekeying attempts using the mechanisms described here fail, an implementation MUST close the IKE SA and any associated Child SAs and then MAY start new ones. Implementations may wish to support in-place rekeying of SAs, since doing so offers better performance and is likely to reduce the number of packets lost during the transition.
To rekey a Child SA within an existing IKE SA, create a new, equivalent SA (see Section 2.17 below), and when the new one is established, delete the old one. Note that, when rekeying, the new Child SA SHOULD NOT have different Traffic Selectors and algorithms than the old one.
To rekey an IKE SA, establish a new equivalent IKE SA (see Section 2.18 below) with the peer to whom the old IKE SA is shared using a CREATE_CHILD_SA within the existing IKE SA. An IKE SA so created inherits all of the original IKE SA's Child SAs, and the new IKE SA is used for all control messages needed to maintain those Child SAs. After the new equivalent IKE SA is created, the initiator deletes the old IKE SA, and the Delete payload to delete itself MUST be the last request sent over the old IKE SA.
SAs should be rekeyed proactively, i.e., the new SA should be established before the old one expires and becomes unusable. Enough time should elapse between the time the new SA is established and the old one becomes unusable so that traffic can be switched over to the new SA.
A difference between IKEv1 and IKEv2 is that in IKEv1 SA lifetimes were negotiated. In IKEv2, each end of the SA is responsible for enforcing its own lifetime policy on the SA and rekeying the SA when necessary. If the two ends have different lifetime policies, the end with the shorter lifetime will end up always being the one to request the rekeying. If an SA has been inactive for a long time and if an endpoint would not initiate the SA in the absence of traffic, the endpoint MAY choose to close the SA instead of rekeying it when its lifetime expires. It can also do so if there has been no traffic since the last time the SA was rekeyed.
Note that IKEv2 deliberately allows parallel SAs with the same Traffic Selectors between common endpoints. One of the purposes of this is to support traffic quality of service (QoS) differences among the SAs (see [DIFFSERVFIELD], [DIFFSERVARCH], and Section 4.1 of [DIFFTUNNEL]). Hence unlike IKEv1, the combination of the endpoints and the Traffic Selectors may not uniquely identify an SA between those endpoints, so the IKEv1 rekeying heuristic of deleting SAs on the basis of duplicate Traffic Selectors SHOULD NOT be used.
There are timing windows -- particularly in the presence of lost packets -- where endpoints may not agree on the state of an SA. The responder to a CREATE_CHILD_SA MUST be prepared to accept messages on an SA before sending its response to the creation request, so there is no ambiguity for the initiator. The initiator MAY begin sending on an SA as soon as it processes the response. The initiator, however, cannot receive on a newly created SA until it receives and processes the response to its CREATE_CHILD_SA request. How, then, is the responder to know when it is OK to send on the newly created SA?
From a technical correctness and interoperability perspective, the responder MAY begin sending on an SA as soon as it sends its response to the CREATE_CHILD_SA request. In some situations, however, this could result in packets unnecessarily being dropped, so an implementation MAY defer such sending.
The responder can be assured that the initiator is prepared to receive messages on an SA if either (1) it has received a cryptographically valid message on the other half of the SA pair, or (2) the new SA rekeys an existing SA and it receives an IKE request to close the replaced SA. When rekeying an SA, the responder continues to send traffic on the old SA until one of those events occurs. When establishing a new SA, the responder MAY defer sending messages on a new SA until either it receives one or a timeout has occurred. If an initiator receives a message on an SA for which it has not received a response to its CREATE_CHILD_SA request, it interprets that as a likely packet loss and retransmits the CREATE_CHILD_SA request. An initiator MAY send a dummy ESP message on a newly created ESP SA if it has no messages queued in order to assure the responder that the initiator is ready to receive messages.
2.8.1. Simultaneous Child SA Rekeying
If the two ends have the same lifetime policies, it is possible that both will initiate a rekeying at the same time (which will result in redundant SAs). To reduce the probability of this happening, the timing of rekeying requests SHOULD be jittered (delayed by a random amount of time after the need for rekeying is noticed).
This form of rekeying may temporarily result in multiple similar SAs between the same pairs of nodes. When there are two SAs eligible to receive packets, a node MUST accept incoming packets through either SA. If redundant SAs are created through such a collision, the SA created with the lowest of the four nonces used in the two exchanges SHOULD be closed by the endpoint that created it. "Lowest" means an octet-by-octet comparison (instead of, for instance, comparing the nonces as large integers). In other words, start by comparing the first octet; if they're equal, move to the next octet, and so on. If you reach the end of one nonce, that nonce is the lower one. The node that initiated the surviving rekeyed SA should delete the replaced SA after the new one is established.
The following is an explanation on the impact this has on implementations. Assume that hosts A and B have an existing Child SA pair with SPIs (SPIa1,SPIb1), and both start rekeying it at the same time:
Host A Host B
-------------------------------------------------------------------
send req1: N(REKEY_SA,SPIa1),
SA(..,SPIa2,..),Ni1,.. -->
<-- send req2: N(REKEY_SA,SPIb1),
SA(..,SPIb2,..),Ni2
recv req2 <--
At this point, A knows there is a simultaneous rekeying happening. However, it cannot yet know which of the exchanges will have the lowest nonce, so it will just note the situation and respond as usual.
send resp2: SA(..,SPIa3,..),
Nr1,.. -->
--> recv req1
Now B also knows that simultaneous rekeying is going on. It responds as usual.
<-- send resp1: SA(..,SPIb3,..),
Nr2,..
recv resp1 <--
--> recv resp2
At this point, there are three Child SA pairs between A and B (the old one and two new ones). A and B can now compare the nonces. Suppose that the lowest nonce was Nr1 in message resp2; in this case, B (the sender of req2) deletes the redundant new SA, and A (the node that initiated the surviving rekeyed SA), deletes the old one.
send req3: D(SPIa1) -->
<-- send req4: D(SPIb2)
--> recv req3
<-- send resp3: D(SPIb1)
recv req4 <--
send resp4: D(SPIa3) -->
The rekeying is now finished.
However, there is a second possible sequence of events that can happen if some packets are lost in the network, resulting in retransmissions. The rekeying begins as usual, but A's first packet (req1) is lost.
Host A Host B
-------------------------------------------------------------------
send req1: N(REKEY_SA,SPIa1),
SA(..,SPIa2,..),
Ni1,.. --> (lost)
<-- send req2: N(REKEY_SA,SPIb1),
SA(..,SPIb2,..),Ni2
recv req2 <--
send resp2: SA(..,SPIa3,..),
Nr1,.. -->
--> recv resp2
<-- send req3: D(SPIb1)
recv req3 <--
send resp3: D(SPIa1) -->
--> recv resp3
From B's point of view, the rekeying is now completed, and since it has not yet received A's req1, it does not even know that there was simultaneous rekeying. However, A will continue retransmitting the message, and eventually it will reach B.
resend req1 -->
--> recv req1
To B, it looks like A is trying to rekey an SA that no longer exists; thus, B responds to the request with something non-fatal such as CHILD_SA_NOT_FOUND.
<-- send resp1: N(CHILD_SA_NOT_FOUND)
recv resp1 <--
When A receives this error, it already knows there was simultaneous rekeying, so it can ignore the error message.
2.8.2. Simultaneous IKE SA Rekeying
Probably the most complex case occurs when both peers try to rekey the IKE_SA at the same time. Basically, the text in Section 2.8 applies to this case as well; however, it is important to ensure that the Child SAs are inherited by the correct IKE_SA.
The case where both endpoints notice the simultaneous rekeying works the same way as with Child SAs. After the CREATE_CHILD_SA exchanges, three IKE SAs exist between A and B: the old IKE SA and two new IKE SAs. The new IKE SA containing the lowest nonce SHOULD be deleted by the node that created it, and the other surviving new IKE SA MUST inherit all the Child SAs.
In addition to normal simultaneous rekeying cases, there is a special case where one peer finishes its rekey before it even notices that other peer is doing a rekey. If only one peer detects a simultaneous rekey, redundant SAs are not created. In this case, when the peer that did not notice the simultaneous rekey gets the request to rekey the IKE SA that it has already successfully rekeyed, it SHOULD return TEMPORARY_FAILURE because it is an IKE SA that it is currently trying to close (whether or not it has already sent the delete notification for the SA). If the peer that did notice the simultaneous rekey gets the delete request from the other peer for the old IKE SA, it knows that the other peer did not detect the simultaneous rekey, and the first peer can forget its own rekey attempt.
Host A Host B
-------------------------------------------------------------------
send req1:
SA(..,SPIa1,..),Ni1,.. -->
<-- send req2: SA(..,SPIb1,..),Ni2,..
--> recv req1
<-- send resp1: SA(..,SPIb2,..),Nr2,..
recv resp1 <--
send req3: D() -->
--> recv req3
At this point, host B sees a request to close the IKE_SA. There's not much more to do than to reply as usual. However, at this point host B should stop retransmitting req2, since once host A receives resp3, it will delete all the state associated with the old IKE_SA and will not be able to reply to it.
<-- send resp3: ()
The TEMPORARY_FAILURE notification was not included in RFC 4306, and support of the TEMPORARY_FAILURE notification is not negotiated. Thus, older peers that implement RFC 4306 but not this document may receive these notifications. In that case, they will treat it the same as any other unknown error notification, and will stop the exchange. Because the other peer has already rekeyed the exchange, doing so does not have any ill effects.
2.8.3. Rekeying the IKE SA versus Reauthentication
Rekeying the IKE SA and reauthentication are different concepts in IKEv2. Rekeying the IKE SA establishes new keys for the IKE SA and resets the Message ID counters, but it does not authenticate the parties again (no AUTH or EAP payloads are involved).
Although rekeying the IKE SA may be important in some environments, reauthentication (the verification that the parties still have access to the long-term credentials) is often more important.
IKEv2 does not have any special support for reauthentication. Reauthentication is done by creating a new IKE SA from scratch (using IKE_SA_INIT/IKE_AUTH exchanges, without any REKEY_SA Notify payloads), creating new Child SAs within the new IKE SA (without REKEY_SA Notify payloads), and finally deleting the old IKE SA (which deletes the old Child SAs as well).
This means that reauthentication also establishes new keys for the IKE SA and Child SAs. Therefore, while rekeying can be performed more often than reauthentication, the situation where "authentication lifetime" is shorter than "key lifetime" does not make sense.
While creation of a new IKE SA can be initiated by either party (initiator or responder in the original IKE SA), the use of EAP and/or Configuration payloads means in practice that reauthentication has to be initiated by the same party as the original IKE SA. IKEv2 does not currently allow the responder to request reauthentication in this case; however, there are extensions that add this functionality such as [REAUTH].
2.9. Traffic Selector Negotiation
When an RFC4301-compliant IPsec subsystem receives an IP packet that matches a "protect" selector in its Security Policy Database (SPD), the subsystem protects that packet with IPsec. When no SA exists yet, it is the task of IKE to create it. Maintenance of a system's SPD is outside the scope of IKE, although some implementations might update their SPD in connection with the running of IKE (for an example scenario, see Section 1.1.3).
Traffic Selector (TS) payloads allow endpoints to communicate some of the information from their SPD to their peers. These must be communicated to IKE from the SPD (for example, the PF_KEY API [PFKEY] uses the SADB_ACQUIRE message). TS payloads specify the selection criteria for packets that will be forwarded over the newly set up SA.
This can serve as a consistency check in some scenarios to assure that the SPDs are consistent. In others, it guides the dynamic update of the SPD.
Two TS payloads appear in each of the messages in the exchange that creates a Child SA pair. Each TS payload contains one or more Traffic Selectors. Each Traffic Selector consists of an address range (IPv4 or IPv6), a port range, and an IP protocol ID.
The first of the two TS payloads is known as TSi (Traffic Selector-initiator). The second is known as TSr (Traffic Selector-responder). TSi specifies the source address of traffic forwarded from (or the destination address of traffic forwarded to) the initiator of the Child SA pair. TSr specifies the destination address of the traffic forwarded to (or the source address of the traffic forwarded from) the responder of the Child SA pair. For example, if the original initiator requests the creation of a Child SA pair, and wishes to tunnel all traffic from subnet 198.51.100.* on the initiator's side to subnet 192.0.2.* on the responder's side, the initiator would include a single Traffic Selector in each TS payload. TSi would specify the address range (198.51.100.0 - 198.51.100.255) and TSr would specify the address range (192.0.2.0 - 192.0.2.255). Assuming that proposal was acceptable to the responder, it would send identical TS payloads back.
IKEv2 allows the responder to choose a subset of the traffic proposed by the initiator. This could happen when the configurations of the two endpoints are being updated but only one end has received the new information. Since the two endpoints may be configured by different people, the incompatibility may persist for an extended period even in the absence of errors. It also allows for intentionally different configurations, as when one end is configured to tunnel all addresses and depends on the other end to have the up-to-date list.
When the responder chooses a subset of the traffic proposed by the initiator, it narrows the Traffic Selectors to some subset of the initiator's proposal (provided the set does not become the null set). If the type of Traffic Selector proposed is unknown, the responder ignores that Traffic Selector, so that the unknown type is not returned in the narrowed set.
To enable the responder to choose the appropriate range in this case, if the initiator has requested the SA due to a data packet, the initiator SHOULD include as the first Traffic Selector in each of TSi and TSr a very specific Traffic Selector including the addresses in the packet triggering the request. In the example, the initiator would include in TSi two Traffic Selectors: the first containing the address range (198.51.100.43 - 198.51.100.43) and the source port and IP protocol from the packet and the second containing (198.51.100.0 - 198.51.100.255) with all ports and IP protocols. The initiator would similarly include two Traffic Selectors in TSr. If the initiator creates the Child SA pair not in response to an arriving packet, but rather, say, upon startup, then there may be no specific addresses the initiator prefers for the initial tunnel over any other. In that case, the first values in TSi and TSr can be ranges rather than specific values.
The responder performs the narrowing as follows:
-
If the responder's policy does not allow it to accept any part of the proposed Traffic Selectors, it responds with a TS_UNACCEPTABLE Notify message.
-
If the responder's policy allows the entire set of traffic covered by TSi and TSr, no narrowing is necessary, and the responder can return the same TSi and TSr values.
-
If the responder's policy allows it to accept the first selector of TSi and TSr, then the responder MUST narrow the Traffic Selectors to a subset that includes the initiator's first choices. In this example above, the responder might respond with TSi being (198.51.100.43 - 198.51.100.43) with all ports and IP protocols.
-
If the responder's policy does not allow it to accept the first selector of TSi and TSr, the responder narrows to an acceptable subset of TSi and TSr.
When narrowing is done, there may be several subsets that are acceptable but their union is not. In this case, the responder arbitrarily chooses one of them, and MAY include an ADDITIONAL_TS_POSSIBLE notification in the response. The ADDITIONAL_TS_POSSIBLE notification asserts that the responder narrowed the proposed Traffic Selectors but that other Traffic Selectors would also have been acceptable, though only in a separate SA. There is no data associated with this Notify type. This case will occur only when the initiator and responder are configured differently from one another. If the initiator and responder agree on the granularity of tunnels, the initiator will never request a tunnel wider than the responder will accept.
It is possible for the responder's policy to contain multiple smaller ranges, all encompassed by the initiator's Traffic Selector, and with the responder's policy being that each of those ranges should be sent over a different SA. Continuing the example above, the responder might have a policy of being willing to tunnel those addresses to and from the initiator, but might require that each address pair be on a separately negotiated Child SA. If the initiator didn't generate its request based on the packet, but (for example) upon startup, there would not be the very specific first Traffic Selectors helping the responder to select the correct range. There would be no way for the responder to determine which pair of addresses should be included in this tunnel, and it would have to make a guess or reject the request with a SINGLE_PAIR_REQUIRED Notify message.
The SINGLE_PAIR_REQUIRED error indicates that a CREATE_CHILD_SA request is unacceptable because its sender is only willing to accept Traffic Selectors specifying a single pair of addresses. The requestor is expected to respond by requesting an SA for only the specific traffic it is trying to forward.
Few implementations will have policies that require separate SAs for each address pair. Because of this, if only some parts of the TSi and TSr proposed by the initiator are acceptable to the responder, responders SHOULD narrow the selectors to an acceptable subset rather than use SINGLE_PAIR_REQUIRED.
2.9.1. Traffic Selectors Violating Own Policy
When creating a new SA, the initiator needs to avoid proposing Traffic Selectors that violate its own policy. If this rule is not followed, valid traffic may be dropped. If you use decorrelated policies from [IPSECARCH], this kind of policy violations cannot happen.
This is best illustrated by an example. Suppose that host A has a policy whose effect is that traffic to 198.51.100.66 is sent via host B encrypted using AES, and traffic to all other hosts in 198.51.100.0/24 is also sent via B, but must use 3DES. Suppose also that host B accepts any combination of AES and 3DES.
If host A now proposes an SA that uses 3DES, and includes TSr containing (198.51.100.0 - 198.51.100.255), this will be accepted by host B. Now, host B can also use this SA to send traffic from 198.51.100.66, but those packets will be dropped by A since it requires the use of AES for this traffic. Even if host A creates a new SA only for 198.51.100.66 that uses AES, host B may freely continue to use the first SA for the traffic. In this situation, when proposing the SA, host A should have followed its own policy, and included a TSr containing ((198.51.100.0 - 198.51.100.65), (198.51.100.67 - 198.51.100.255)) instead.
In general, if (1) the initiator makes a proposal "for traffic X (TSi/TSr), do SA", and (2) for some subset X' of X, the initiator does not actually accept traffic X' with SA, and (3) the initiator would be willing to accept traffic X' with some SA' (!=SA), valid traffic can be unnecessarily dropped since the responder can apply either SA or SA' to traffic X'.
2.9.2. Traffic Selectors in Rekeying
Rekeying is used to replace an existing Child SA with another. If the new SA would be allowed to have a narrower set of selectors than the original, traffic that was allowed on the old SA would be dropped in the new SA, thus violating the idea of "replacing". Thus, the new SA MUST NOT have narrower selectors than the original. If the rekeyed SA would ever need to have a narrower scope than the currently used SA, that would mean that the policy was changed in a way such that the currently used SA is against the policy. In that case, the SA should have been already deleted after the policy change took effect.
When the initiator attempts to rekey the Child SA, the proposed Traffic Selectors SHOULD be either the same as, or a superset of, the Traffic Selectors used in the old Child SA. That is, they would be the same as, or a superset of, the currently active (decorrelated) policy. The responder MUST NOT narrow down the Traffic Selectors narrower than the scope currently in use.
Because a rekeyed SA can never have a narrower scope than the one currently in use, there is no need for the selectors from the packet, so those selectors SHOULD NOT be sent.
2.10. Nonces
The IKE_SA_INIT messages each contain a nonce. These nonces are used as inputs to cryptographic functions. The CREATE_CHILD_SA request and the CREATE_CHILD_SA response also contain nonces. These nonces are used to add freshness to the key derivation technique used to obtain keys for Child SA, and to ensure creation of strong pseudorandom bits from the Diffie-Hellman key. Nonces used in IKEv2 MUST be randomly chosen, MUST be at least 128 bits in size, and MUST be at least half the key size of the negotiated pseudorandom function (PRF). However, the initiator chooses the nonce before the outcome of the negotiation is known. Because of that, the nonce has to be long enough for all the PRFs being proposed. If the same random number source is used for both keys and nonces, care must be taken to ensure that the latter use does not compromise the former.
2.11. Address and Port Agility
IKE runs over UDP ports 500 and 4500, and implicitly sets up ESP and AH associations for the same IP addresses over which it runs. The IP addresses and ports in the outer header are, however, not themselves cryptographically protected, and IKE is designed to work even through Network Address Translation (NAT) boxes. An implementation MUST accept incoming requests even if the source port is not 500 or 4500, and MUST respond to the address and port from which the request was received. It MUST specify the address and port at which the request was received as the source address and port in the response. IKE functions identically over IPv4 or IPv6.
2.12. Reuse of Diffie-Hellman Exponentials
IKE generates keying material using an ephemeral Diffie-Hellman exchange in order to gain the property of "perfect forward secrecy". This means that once a connection is closed and its corresponding keys are forgotten, even someone who has recorded all of the data from the connection and gets access to all of the long-term keys of the two endpoints cannot reconstruct the keys used to protect the conversation without doing a brute force search of the session key space.
Achieving perfect forward secrecy requires that when a connection is closed, each endpoint MUST forget not only the keys used by the connection but also any information that could be used to recompute those keys.
Because computing Diffie-Hellman exponentials is computationally expensive, an endpoint may find it advantageous to reuse those exponentials for multiple connection setups. There are several reasonable strategies for doing this. An endpoint could choose a new exponential only periodically though this could result in less-than-perfect forward secrecy if some connection lasts for less than the lifetime of the exponential. Or it could keep track of which exponential was used for each connection and delete the information associated with the exponential only when some corresponding connection was closed. This would allow the exponential to be reused without losing perfect forward secrecy at the cost of maintaining more state.
Whether and when to reuse Diffie-Hellman exponentials are private decisions in the sense that they will not affect interoperability. An implementation that reuses exponentials MAY choose to remember the exponential used by the other endpoint on past exchanges and if one is reused to avoid the second half of the calculation. See [REUSE] and [RFC6989] for a security analysis of this practice and for additional security considerations when reusing ephemeral Diffie-Hellman keys.
2.13. Generating Keying Material
In the context of the IKE SA, four cryptographic algorithms are negotiated: an encryption algorithm, an integrity protection algorithm, a Diffie-Hellman group, and a pseudorandom function (PRF). The PRF is used for the construction of keying material for all of the cryptographic algorithms used in both the IKE SA and the Child SAs.
We assume that each encryption algorithm and integrity protection algorithm uses a fixed-size key and that any randomly chosen value of that fixed size can serve as an appropriate key. For algorithms that accept a variable-length key, a fixed key size MUST be specified as part of the cryptographic transform negotiated (see Section 3.3.5 for the definition of the Key Length transform attribute). For algorithms for which not all values are valid keys (such as DES or 3DES with key parity), the algorithm by which keys are derived from arbitrary values MUST be specified by the cryptographic transform. For integrity protection functions based on Hashed Message Authentication Code (HMAC), the fixed key size is the size of the output of the underlying hash function.
It is assumed that PRFs accept keys of any length, but have a preferred key size. The preferred key size MUST be used as the length of SK_d, SK_pi, and SK_pr (see Section 2.14). For PRFs based on the HMAC construction, the preferred key size is equal to the length of the output of the underlying hash function. Other types of PRFs MUST specify their preferred key size.
Keying material will always be derived as the output of the negotiated PRF algorithm. Since the amount of keying material needed may be greater than the size of the output of the PRF, the PRF is used iteratively. The term "prf+" describes a function that outputs a pseudorandom stream based on the inputs to a pseudorandom function called "prf".
In the following, | indicates concatenation. prf+ is defined as:
prf+ (K,S) = T1 | T2 | T3 | T4 | ...
where:
T1 = prf (K, S | 0x01)
T2 = prf (K, T1 | S | 0x02)
T3 = prf (K, T2 | S | 0x03)
T4 = prf (K, T3 | S | 0x04)
...
This continues until all the material needed to compute all required keys has been output from prf+. The keys are taken from the output string without regard to boundaries (e.g., if the required keys are a 256-bit Advanced Encryption Standard (AES) key and a 160-bit HMAC key, and the prf function generates 160 bits, the AES key will come from T1 and the beginning of T2, while the HMAC key will come from the rest of T2 and the beginning of T3).
The constant concatenated to the end of each prf function is a single octet. The prf+ function is not defined beyond 255 times the size of the prf function output.
2.14. Generating Keying Material for the IKE SA
The shared keys are computed as follows. A quantity called SKEYSEED is calculated from the nonces exchanged during the IKE_SA_INIT exchange and the Diffie-Hellman shared secret established during that exchange. SKEYSEED is used to calculate seven other secrets: SK_d used for deriving new keys for the Child SAs established with this IKE SA; SK_ai and SK_ar used as a key to the integrity protection algorithm for authenticating the component messages of subsequent exchanges; SK_ei and SK_er used for encrypting (and of course decrypting) all subsequent exchanges; and SK_pi and SK_pr, which are used when generating an AUTH payload. The lengths of SK_d, SK_pi, and SK_pr MUST be the preferred key length of the PRF agreed upon.
SKEYSEED and its derivatives are computed as follows:
SKEYSEED = prf(Ni | Nr, g^ir)
{SK_d | SK_ai | SK_ar | SK_ei | SK_er | SK_pi | SK_pr}
= prf+ (SKEYSEED, Ni | Nr | SPIi | SPIr)
(indicating that the quantities SK_d, SK_ai, SK_ar, SK_ei, SK_er, SK_pi, and SK_pr are taken in order from the generated bits of the prf+). g^ir is the shared secret from the ephemeral Diffie-Hellman exchange. g^ir is represented as a string of octets in big endian order padded with zeros if necessary to make it the length of the modulus. Ni and Nr are the nonces, stripped of any headers. For historical backward-compatibility reasons, there are two PRFs that are treated specially in this calculation. If the negotiated PRF is AES-XCBC-PRF-128 [AESXCBCPRF128] or AES-CMAC-PRF-128 [AESCMACPRF128], only the first 64 bits of Ni and the first 64 bits of Nr are used in calculating SKEYSEED, but all the bits are used for input to the prf+ function.
The two directions of traffic flow use different keys. The keys used to protect messages from the original initiator are SK_ai and SK_ei. The keys used to protect messages in the other direction are SK_ar and SK_er.
2.15. Authentication of the IKE SA
When not using extensible authentication (see Section 2.16), the peers are authenticated by having each sign (or MAC using a padded shared secret as the key, as described later in this section) a block of data. In these calculations, IDi' and IDr' are the entire ID payloads excluding the fixed header. For the responder, the octets to be signed start with the first octet of the first SPI in the header of the second message (IKE_SA_INIT response) and end with the last octet of the last payload in the second message. Appended to this (for the purposes of computing the signature) are the initiator's nonce Ni (just the value, not the payload containing it), and the value prf(SK_pr, IDr'). Note that neither the nonce Ni nor the value prf(SK_pr, IDr') are transmitted. Similarly, the initiator signs the first message (IKE_SA_INIT request), starting with the first octet of the first SPI in the header and ending with the last octet of the last payload. Appended to this (for purposes of computing the signature) are the responder's nonce Nr, and the value prf(SK_pi, IDi'). It is critical to the security of the exchange that each side sign the other side's nonce.
The initiator's signed octets can be described as:
InitiatorSignedOctets = RealMessage1 | NonceRData | MACedIDForI
GenIKEHDR = [ four octets 0 if using port 4500 ] | RealIKEHDR
RealIKEHDR = SPIi | SPIr | . . . | Length
RealMessage1 = RealIKEHDR | RestOfMessage1
NonceRPayload = PayloadHeader | NonceRData
InitiatorIDPayload = PayloadHeader | RestOfInitIDPayload
RestOfInitIDPayload = IDType | RESERVED | InitIDData
MACedIDForI = prf(SK_pi, RestOfInitIDPayload)
The responder's signed octets can be described as:
ResponderSignedOctets = RealMessage2 | NonceIData | MACedIDForR
GenIKEHDR = [ four octets 0 if using port 4500 ] | RealIKEHDR
RealIKEHDR = SPIi | SPIr | . . . | Length
RealMessage2 = RealIKEHDR | RestOfMessage2
NonceIPayload = PayloadHeader | NonceIData
ResponderIDPayload = PayloadHeader | RestOfRespIDPayload
RestOfRespIDPayload = IDType | RESERVED | RespIDData
MACedIDForR = prf(SK_pr, RestOfRespIDPayload)
Note that all of the payloads are included under the signature, including any payload types not defined in this document. If the first message of the exchange is sent multiple times (such as with a responder cookie and/or a different Diffie-Hellman group), it is the latest version of the message that is signed.
Optionally, messages 3 and 4 MAY include a certificate, or certificate chain providing evidence that the key used to compute a digital signature belongs to the name in the ID payload. The signature or MAC will be computed using algorithms dictated by the type of key used by the signer, and specified by the Auth Method field in the Authentication payload. There is no requirement that the initiator and responder sign with the same cryptographic algorithms. The choice of cryptographic algorithms depends on the type of key each has. In particular, the initiator may be using a shared key while the responder may have a public signature key and certificate. It will commonly be the case (but it is not required) that, if a shared secret is used for authentication, the same key is used in both directions.
Note that it is a common but typically insecure practice to have a shared key derived solely from a user-chosen password without incorporating another source of randomness. This is typically insecure because user-chosen passwords are unlikely to have sufficient unpredictability to resist dictionary attacks and these attacks are not prevented in this authentication method. (Applications using password-based authentication for bootstrapping and IKE SA should use the authentication method in Section 2.16, which is designed to prevent off-line dictionary attacks.) The pre-shared key needs to contain as much unpredictability as the strongest key being negotiated. In the case of a pre-shared key, the AUTH value is computed as:
For the initiator:
AUTH = prf( prf(Shared Secret, "Key Pad for IKEv2"),
<InitiatorSignedOctets>)
For the responder:
AUTH = prf( prf(Shared Secret, "Key Pad for IKEv2"),
<ResponderSignedOctets>)
where the string "Key Pad for IKEv2" is 17 ASCII characters without null termination. The shared secret can be variable length. The pad string is added so that if the shared secret is derived from a password, the IKE implementation need not store the password in cleartext, but rather can store the value prf(Shared Secret,"Key Pad for IKEv2"), which could not be used as a password equivalent for protocols other than IKEv2. As noted above, deriving the shared secret from a password is not secure. This construction is used because it is anticipated that people will do it anyway. The management interface by which the shared secret is provided MUST accept ASCII strings of at least 64 octets and MUST NOT add a null terminator before using them as shared secrets. It MUST also accept a hex encoding of the shared secret. The management interface MAY accept other encodings if the algorithm for translating the encoding to a binary string is specified.
There are two types of EAP authentication (described in Section 2.16), and each type uses different values in the AUTH computations shown above. If the EAP method is key-generating, substitute master session key (MSK) for the shared secret in the computation. For non-key-generating methods, substitute SK_pi and SK_pr, respectively, for the shared secret in the two AUTH computations.
2.16. Extensible Authentication Protocol Methods
In addition to authentication using public key signatures and shared secrets, IKE supports authentication using methods defined in RFC 3748 [EAP]. Typically, these methods are asymmetric (designed for a user authenticating to a server), and they may not be mutual. For this reason, these protocols are typically used to authenticate the initiator to the responder and MUST be used in conjunction with a public-key-signature-based authentication of the responder to the initiator. These methods are often associated with mechanisms referred to as "Legacy Authentication" mechanisms.
While this document references [EAP] with the intent that new methods can be added in the future without updating this specification, some simpler variations are documented here. [EAP] defines an authentication protocol requiring a variable number of messages. Extensible authentication is implemented in IKE as additional IKE_AUTH exchanges that MUST be completed in order to initialize the IKE SA.
An initiator indicates a desire to use EAP by leaving out the AUTH payload from the first message in the IKE_AUTH exchange. (Note that the AUTH payload is required for non-EAP authentication, and is thus not marked as optional in the rest of this document.) By including an IDi payload but not an AUTH payload, the initiator has declared an identity but has not proven it. If the responder is willing to use an EAP method, it will place an Extensible Authentication Protocol (EAP) payload in the response of the IKE_AUTH exchange and defer sending SAr2, TSi, and TSr until initiator authentication is complete in a subsequent IKE_AUTH exchange. In the case of a minimal EAP method, the initial SA establishment will appear as follows:
Initiator Responder
-------------------------------------------------------------------
HDR, SAi1, KEi, Ni -->
<-- HDR, SAr1, KEr, Nr, [CERTREQ]
HDR, SK {IDi, [CERTREQ,]
[IDr,] SAi2,
TSi, TSr} -->
<-- HDR, SK {IDr, [CERT,] AUTH,
EAP}
HDR, SK {EAP} -->
<-- HDR, SK {EAP (success)}
HDR, SK {AUTH} -->
<-- HDR, SK {AUTH, SAr2, TSi, TSr}
As described in Section 2.2, when EAP is used, each pair of IKE SA initial setup messages will have their message numbers incremented; the first pair of IKE_AUTH messages will have an ID of 1, the second will be 2, and so on.
For EAP methods that create a shared key as a side effect of authentication, that shared key MUST be used by both the initiator and responder to generate AUTH payloads in messages 7 and 8 using the syntax for shared secrets specified in Section 2.15. The shared key from EAP is the field from the EAP specification named MSK. This shared key generated during an IKE exchange MUST NOT be used for any other purpose.
EAP methods that do not establish a shared key SHOULD NOT be used, as they are subject to a number of man-in-the-middle attacks [EAPMITM] if these EAP methods are used in other protocols that do not use a server-authenticated tunnel. Please see the Security Considerations section for more details. If EAP methods that do not generate a shared key are used, the AUTH payloads in messages 7 and 8 MUST be generated using SK_pi and SK_pr, respectively.
The initiator of an IKE SA using EAP needs to be capable of extending the initial protocol exchange to at least ten IKE_AUTH exchanges in the event the responder sends notification messages and/or retries the authentication prompt. Once the protocol exchange defined by the chosen EAP authentication method has successfully terminated, the responder MUST send an EAP payload containing the Success message. Similarly, if the authentication method has failed, the responder MUST send an EAP payload containing the Failure message. The responder MAY at any time terminate the IKE exchange by sending an EAP payload containing the Failure message.
Following such an extended exchange, the EAP AUTH payloads MUST be included in the two messages following the one containing the EAP Success message.
When the initiator authentication uses EAP, it is possible that the contents of the IDi payload is used only for Authentication, Authorization, and Accounting (AAA) routing purposes and selecting which EAP method to use. This value may be different from the identity authenticated by the EAP method. It is important that policy lookups and access control decisions use the actual authenticated identity. Often the EAP server is implemented in a separate AAA server that communicates with the IKEv2 responder. In this case, the authenticated identity, if different from that in the IDi payload, has to be sent from the AAA server to the IKEv2 responder.
2.17. Generating Keying Material for Child SAs
A single Child SA is created by the IKE_AUTH exchange, and additional Child SAs can optionally be created in CREATE_CHILD_SA exchanges. Keying material for them is generated as follows:
KEYMAT = prf+(SK_d, Ni | Nr)
Where Ni and Nr are the nonces from the IKE_SA_INIT exchange if this request is the first Child SA created or the fresh Ni and Nr from the CREATE_CHILD_SA exchange if this is a subsequent creation.
For CREATE_CHILD_SA exchanges including an optional Diffie-Hellman exchange, the keying material is defined as:
KEYMAT = prf+(SK_d, g^ir (new) | Ni | Nr)
where g^ir (new) is the shared secret from the ephemeral Diffie-Hellman exchange of this CREATE_CHILD_SA exchange (represented as an octet string in big endian order padded with zeros in the high-order bits if necessary to make it the length of the modulus).
A single CREATE_CHILD_SA negotiation may result in multiple Security Associations. ESP and AH SAs exist in pairs (one in each direction), so two SAs are created in a single Child SA negotiation for them. Furthermore, Child SA negotiation may include some future IPsec protocol(s) in addition to, or instead of, ESP or AH (for example, ROHC_INTEG as described in [ROHCV2]). In any case, keying material for each Child SA MUST be taken from the expanded KEYMAT using the following rules:
-
All keys for SAs carrying data from the initiator to the responder are taken before SAs going from the responder to the initiator.
-
If multiple IPsec protocols are negotiated, keying material for each Child SA is taken in the order in which the protocol headers will appear in the encapsulated packet.
-
If an IPsec protocol requires multiple keys, the order in which they are taken from the SA's keying material needs to be described in the protocol's specification. For ESP and AH, [IPSECARCH] defines the order, namely: the encryption key (if any) MUST be taken from the first bits and the integrity key (if any) MUST be taken from the remaining bits.
Each cryptographic algorithm takes a fixed number of bits of keying material specified as part of the algorithm, or negotiated in SA payloads (see Section 2.13 for description of key lengths, and Section 3.3.5 for the definition of the Key Length transform attribute).
2.18. Rekeying IKE SAs Using a CREATE_CHILD_SA Exchange
The CREATE_CHILD_SA exchange can be used to rekey an existing IKE SA (see Sections 1.3.2 and 2.8). New initiator and responder SPIs are supplied in the SPI fields in the Proposal structures inside the Security Association (SA) payloads (not the SPI fields in the IKE header). The TS payloads are omitted when rekeying an IKE SA. SKEYSEED for the new IKE SA is computed using SK_d from the existing IKE SA as follows:
SKEYSEED = prf(SK_d (old), g^ir (new) | Ni | Nr)
where g^ir (new) is the shared secret from the ephemeral Diffie-Hellman exchange of this CREATE_CHILD_SA exchange (represented as an octet string in big endian order padded with zeros if necessary to make it the length of the modulus) and Ni and Nr are the two nonces stripped of any headers.
The old and new IKE SA may have selected a different PRF. Because the rekeying exchange belongs to the old IKE SA, it is the old IKE SA's PRF that is used to generate SKEYSEED.
The main reason for rekeying the IKE SA is to ensure that the compromise of old keying material does not provide information about the current keys, or vice versa. Therefore, implementations MUST perform a new Diffie-Hellman exchange when rekeying the IKE SA. In other words, an initiator MUST NOT propose the value "NONE" for the Diffie-Hellman transform, and a responder MUST NOT accept such a proposal. This means that a successful exchange rekeying the IKE SA always includes the KEi/KEr payloads.
The new IKE SA MUST reset its message counters to 0.
SK_d, SK_ai, SK_ar, SK_ei, and SK_er are computed from SKEYSEED as specified in Section 2.14, using SPIi, SPIr, Ni, and Nr from the new exchange, and using the new IKE SA's PRF.
2.19. Requesting an Internal Address on a Remote Network
Most commonly occurring in the endpoint-to-security-gateway scenario, an endpoint may need an IP address in the network protected by the security gateway and may need to have that address dynamically assigned. A request for such a temporary address can be included in any request to create a Child SA (including the implicit request in message 3) by including a CP payload. Note, however, it is usual to only assign one IP address during the IKE_AUTH exchange. That address persists at least until the deletion of the IKE SA.
This function provides address allocation to an IPsec Remote Access Client (IRAC) trying to tunnel into a network protected by an IPsec Remote Access Server (IRAS). Since the IKE_AUTH exchange creates an IKE SA and a Child SA, the IRAC MUST request the IRAS-controlled address (and optionally other information concerning the protected network) in the IKE_AUTH exchange. The IRAS may procure an address for the IRAC from any number of sources such as a DHCP/BOOTP (Bootstrap Protocol) server or its own address pool.
Initiator Responder
-------------------------------------------------------------------
HDR, SK {IDi, [CERT,]
[CERTREQ,] [IDr,] AUTH,
CP(CFG_REQUEST), SAi2,
TSi, TSr} -->
<-- HDR, SK {IDr, [CERT,] AUTH,
CP(CFG_REPLY), SAr2,
TSi, TSr}
In all cases, the CP payload MUST be inserted before the SA payload. In variations of the protocol where there are multiple IKE_AUTH exchanges, the CP payloads MUST be inserted in the messages containing the SA payloads.
CP(CFG_REQUEST) MUST contain at least an INTERNAL_ADDRESS attribute (either IPv4 or IPv6) but MAY contain any number of additional attributes the initiator wants returned in the response.
For example, message from initiator to responder:
CP(CFG_REQUEST)=
INTERNAL_ADDRESS()
TSi = (0, 0-65535, 0.0.0.0-255.255.255.255)
TSr = (0, 0-65535, 0.0.0.0-255.255.255.255)
NOTE: Traffic Selectors contain (protocol, port range, address range).
Message from responder to initiator:
CP(CFG_REPLY)=
INTERNAL_ADDRESS(192.0.2.202)
INTERNAL_NETMASK(255.255.255.0)
INTERNAL_SUBNET(192.0.2.0/255.255.255.0)
TSi = (0, 0-65535, 192.0.2.202-192.0.2.202)
TSr = (0, 0-65535, 192.0.2.0-192.0.2.255)
All returned values will be implementation dependent. As can be seen in the above example, the IRAS MAY also send other attributes that were not included in CP(CFG_REQUEST) and MAY ignore the non-mandatory attributes that it does not support.
The responder MUST NOT send a CFG_REPLY without having first received a CP(CFG_REQUEST) from the initiator, because we do not want the IRAS to perform an unnecessary configuration lookup if the IRAC cannot process the REPLY.
In the case where the IRAS's configuration requires that CP be used for a given identity IDi, but IRAC has failed to send a CP(CFG_REQUEST), IRAS MUST fail the request, and terminate the Child SA creation with a FAILED_CP_REQUIRED error. The FAILED_CP_REQUIRED is not fatal to the IKE SA; it simply causes the Child SA creation to fail. The initiator can fix this by later starting a new Configuration payload request. There is no associated data in the FAILED_CP_REQUIRED error.
2.20. Requesting the Peer's Version
An IKE peer wishing to inquire about the other peer's IKE software version information MAY use the method below. This is an example of a configuration request within an INFORMATIONAL exchange, after the IKE SA and first Child SA have been created.
An IKE implementation MAY decline to give out version information prior to authentication or even after authentication in case some implementation is known to have some security weakness. In that case, it MUST either return an empty string or no CP payload if CP is not supported.
Initiator Responder
-------------------------------------------------------------------
HDR, SK {CP(CFG_REQUEST)} -->
<-- HDR, SK {CP(CFG_REPLY)}
CP(CFG_REQUEST)=
APPLICATION_VERSION("")
CP(CFG_REPLY) APPLICATION_VERSION("foobar v1.3beta, (c) Foo Bar
Inc.")
2.21. Error Handling
There are many kinds of errors that can occur during IKE processing. The general rule is that if a request is received that is badly formatted, or unacceptable for reasons of policy (such as no matching cryptographic algorithms), the response contains a Notify payload indicating the error. The decision whether or not to send such a response depends whether or not there is an authenticated IKE SA.
If there is an error parsing or processing a response packet, the general rule is to not send back any error message because responses should not generate new requests (and a new request would be the only way to send back an error message). Such errors in parsing or processing response packets should still cause the recipient to clean up the IKE state (for example, by sending a Delete for a bad SA).
Only authentication failures (AUTHENTICATION_FAILED and EAP failure) and malformed messages (INVALID_SYNTAX) lead to a deletion of the IKE SA without requiring an explicit INFORMATIONAL exchange carrying a Delete payload. Other error conditions MAY require such an exchange if policy dictates that this is needed. If the exchange is terminated with EAP Failure, an AUTHENTICATION_FAILED notification is not sent.
2.21.1. Error Handling in IKE_SA_INIT
Errors that occur before a cryptographically protected IKE SA is established need to be handled very carefully. There is a trade-off between wanting to help the peer to diagnose a problem and thus responding to the error and wanting to avoid being part of a DoS attack based on forged messages.
In an IKE_SA_INIT exchange, any error notification causes the exchange to fail. Note that some error notifications such as COOKIE, INVALID_KE_PAYLOAD or INVALID_MAJOR_VERSION may lead to a subsequent successful exchange. Because all error notifications are completely unauthenticated, the recipient should continue trying for some time before giving up. The recipient should not immediately act based on the error notification unless corrective actions are defined in this specification, such as for COOKIE, INVALID_KE_PAYLOAD, and INVALID_MAJOR_VERSION.
2.21.2. Error Handling in IKE_AUTH
All errors that occur in an IKE_AUTH exchange, causing the authentication to fail for whatever reason (invalid shared secret, invalid ID, untrusted certificate issuer, revoked or expired certificate, etc.) SHOULD result in an AUTHENTICATION_FAILED notification. If the error occurred on the responder, the notification is returned in the protected response, and is usually the only payload in that response. Although the IKE_AUTH messages are encrypted and integrity protected, if the peer receiving this notification has not authenticated the other end yet, that peer needs to treat the information with caution.
If the error occurs on the initiator, the notification MAY be returned in a separate INFORMATIONAL exchange, usually with no other payloads. This is an exception for the general rule of not starting new exchanges based on errors in responses.
Note, however, that request messages that contain an unsupported critical payload, or where the whole message is malformed (rather than just bad payload contents), MUST be rejected in their entirety, and MUST only lead to an UNSUPPORTED_CRITICAL_PAYLOAD or INVALID_SYNTAX Notification sent as a response. The receiver should not verify the payloads related to authentication in this case.
If authentication has succeeded in the IKE_AUTH exchange, the IKE SA is established; however, establishing the Child SA or requesting configuration information may still fail. This failure does not automatically cause the IKE SA to be deleted. Specifically, a responder may include all the payloads associated with authentication (IDr, CERT, and AUTH) while sending error notifications for the piggybacked exchanges (FAILED_CP_REQUIRED, NO_PROPOSAL_CHOSEN, and so on), and the initiator MUST NOT fail the authentication because of this. The initiator MAY, of course, for reasons of policy later delete such an IKE SA.
In an IKE_AUTH exchange, or in the INFORMATIONAL exchange immediately following it (in case an error happened when processing a response to IKE_AUTH), the UNSUPPORTED_CRITICAL_PAYLOAD, INVALID_SYNTAX, and AUTHENTICATION_FAILED notifications are the only ones to cause the IKE SA to be deleted or not created, without a Delete payload. Extension documents may define new error notifications with these semantics, but MUST NOT use them unless the peer has been shown to understand them, such as by using the Vendor ID payload.
2.21.3. Error Handling after IKE SA is Authenticated
After the IKE SA is authenticated, all requests having errors MUST result in a response notifying the other end of the error.
In normal situations, there should not be cases where a valid response from one peer results in an error situation in the other peer, so there should not be any reason for a peer to send error messages to the other end except as a response. Because sending such error messages as an INFORMATIONAL exchange might lead to further errors that could cause loops, such errors SHOULD NOT be sent. If errors are seen that indicate that the peers do not have the same state, it might be good to delete the IKE SA to clean up state and start over.
If a peer parsing a request notices that it is badly formatted (after it has passed the message authentication code checks and window checks) and it returns an INVALID_SYNTAX notification, then this error notification is considered fatal in both peers, meaning that the IKE SA is deleted without needing an explicit Delete payload.
2.21.4. Error Handling Outside IKE SA
A node needs to limit the rate at which it will send messages in response to unprotected messages.
If a node receives a message on UDP port 500 or 4500 outside the context of an IKE SA known to it (and the message is not a request to start an IKE SA), this may be the result of a recent crash of the node. If the message is marked as a response, the node can audit the suspicious event but MUST NOT respond. If the message is marked as a request, the node can audit the suspicious event and MAY send a response. If a response is sent, the response MUST be sent to the IP address and port from where it came with the same IKE SPIs and the Message ID copied. The response MUST NOT be cryptographically protected and MUST contain an INVALID_IKE_SPI Notify payload. The INVALID_IKE_SPI notification indicates an IKE message was received with an unrecognized destination SPI; this usually indicates that the recipient has rebooted and forgotten the existence of an IKE SA.
A peer receiving such an unprotected Notify payload MUST NOT respond and MUST NOT change the state of any existing SAs. The message might be a forgery or might be a response that a genuine correspondent was tricked into sending. A node should treat such a message (and also a network message like ICMP destination unreachable) as a hint that there might be problems with SAs to that IP address and should initiate a liveness check for any such IKE SA. An implementation SHOULD limit the frequency of such tests to avoid being tricked into participating in a DoS attack.
If an error occurs outside the context of an IKE request (e.g., the node is getting ESP messages on a nonexistent SPI), the node SHOULD initiate an INFORMATIONAL exchange with a Notify payload describing the problem.
A node receiving a suspicious message from an IP address (and port, if NAT traversal is used) with which it has an IKE SA SHOULD send an IKE Notify payload in an IKE INFORMATIONAL exchange over that SA. The recipient MUST NOT change the state of any SAs as a result, but may wish to audit the event to aid in diagnosing malfunctions.
2.22. IPComp
Use of IP Compression [IP-COMP] can be negotiated as part of the setup of a Child SA. While IP Compression involves an extra header in each packet and a compression parameter index (CPI), the virtual "compression association" has no life outside the ESP or AH SA that contains it. Compression associations disappear when the corresponding ESP or AH SA goes away. It is not explicitly mentioned in any Delete payload.
Negotiation of IP Compression is separate from the negotiation of cryptographic parameters associated with a Child SA. A node requesting a Child SA MAY advertise its support for one or more compression algorithms through one or more Notify payloads of type IPCOMP_SUPPORTED. This Notify message may be included only in a message containing an SA payload negotiating a Child SA and indicates a willingness by its sender to use IPComp on this SA. The response MAY indicate acceptance of a single compression algorithm with a Notify payload of type IPCOMP_SUPPORTED. These payloads MUST NOT occur in messages that do not contain SA payloads.
The data associated with this Notify message includes a two-octet IPComp CPI followed by a one-octet Transform ID optionally followed by attributes whose length and format are defined by that Transform ID. A message proposing an SA may contain multiple IPCOMP_SUPPORTED notifications to indicate multiple supported algorithms. A message accepting an SA may contain at most one.
The Transform IDs are listed here. The values in the following table are only current as of the publication date of RFC 4306. Other values may have been added since then or will be added after the publication of this document. Readers should refer to [IKEV2IANA] for the latest values.
| Name | Number | Defined In |
|---|---|---|
| IPCOMP_OUI | 1 | (UNSPECIFIED) |
| IPCOMP_DEFLATE | 2 | RFC 2394 |
| IPCOMP_LZS | 3 | RFC 2395 |
| IPCOMP_LZJH | 4 | RFC 3051 |
Although there has been discussion of allowing multiple compression algorithms to be accepted and to have different compression algorithms available for the two directions of a Child SA, implementations of this specification MUST NOT accept an IPComp algorithm that was not proposed, MUST NOT accept more than one, and MUST NOT compress using an algorithm other than one proposed and accepted in the setup of the Child SA.
A side effect of separating the negotiation of IPComp from cryptographic parameters is that it is not possible to propose multiple cryptographic suites and propose IP Compression with some of them but not others.
In some cases, Robust Header Compression (ROHC) may be more appropriate than IP Compression. [ROHCV2] defines the use of ROHC with IKEv2 and IPsec.
2.23. NAT Traversal
Network Address Translation (NAT) gateways are a controversial subject. This section briefly describes what they are and how they are likely to act on IKE traffic. Many people believe that NATs are evil and that we should not design our protocols so as to make them work better. IKEv2 does indeed specify some unintuitive processing rules so that NATs are more likely to work.
NATs exist primarily because of the shortage of IPv4 addresses, though there are other rationales. IP nodes that are "behind" a NAT have IP addresses that are not globally unique, but rather are assigned from some space that is unique within the network behind the NAT but that are likely to be reused by nodes behind other NATs. Generally, nodes behind NATs can communicate with other nodes behind the same NAT and with nodes with globally unique addresses, but not with nodes behind other NATs. There are exceptions to that rule. When those nodes make connections to nodes on the real Internet, the NAT gateway "translates" the IP source address to an address that will be routed back to the gateway. Messages to the gateway from the Internet have their destination addresses "translated" to the internal address that will route the packet to the correct endnode.
NATs are designed to be "transparent" to endnodes. Neither software on the node behind the NAT nor the node on the Internet requires modification to communicate through the NAT. Achieving this transparency is more difficult with some protocols than with others. Protocols that include IP addresses of the endpoints within the payloads of the packet will fail unless the NAT gateway understands the protocol and modifies the internal references as well as those in the headers. Such knowledge is inherently unreliable, is a network layer violation, and often results in subtle problems.
Opening an IPsec connection through a NAT introduces special problems. If the connection runs in transport mode, changing the IP addresses on packets will cause the checksums to fail and the NAT cannot correct the checksums because they are cryptographically protected. Even in tunnel mode, there are routing problems because transparently translating the addresses of AH and ESP packets requires special logic in the NAT and that logic is heuristic and unreliable in nature. For that reason, IKEv2 will use UDP encapsulation of IKE and ESP packets. This encoding is slightly less efficient but is easier for NATs to process. In addition, firewalls may be configured to pass UDP-encapsulated IPsec traffic but not plain, unencapsulated ESP/AH or vice versa.
It is a common practice of NATs to translate TCP and UDP port numbers as well as addresses and use the port numbers of inbound packets to decide which internal node should get a given packet. For this reason, even though IKE packets MUST be sent to and from UDP port 500 or 4500, they MUST be accepted coming from any port and responses MUST be sent to the port from whence they came. This is because the ports may be modified as the packets pass through NATs. Similarly, IP addresses of the IKE endpoints are generally not included in the IKE payloads because the payloads are cryptographically protected and could not be transparently modified by NATs.
Port 4500 is reserved for UDP-encapsulated ESP and IKE. An IPsec endpoint that discovers a NAT between it and its correspondent (as described below) MUST send all subsequent traffic from port 4500, which NATs should not treat specially (as they might with port 500).
An initiator can use port 4500 for both IKE and ESP, regardless of whether or not there is a NAT, even at the beginning of IKE. When either side is using port 4500, sending ESP with UDP encapsulation is not required, but understanding received UDP-encapsulated ESP packets is required. UDP encapsulation MUST NOT be done on port 500. If Network Address Translation Traversal (NAT-T) is supported (that is, if NAT_DETECTION_*_IP payloads were exchanged during IKE_SA_INIT), all devices MUST be able to receive and process both UDP-encapsulated ESP and non-UDP-encapsulated ESP packets at any time. Either side can decide whether or not to use UDP encapsulation for ESP irrespective of the choice made by the other side. However, if a NAT is detected, both devices MUST use UDP encapsulation for ESP.
The specific requirements for supporting NAT traversal [NATREQ] are listed below. Support for NAT traversal is optional. In this section only, requirements listed as MUST apply only to implementations supporting NAT traversal.
-
Both the IKE initiator and responder MUST include in their IKE_SA_INIT packets Notify payloads of type NAT_DETECTION_SOURCE_IP and NAT_DETECTION_DESTINATION_IP. Those payloads can be used to detect if there is NAT between the hosts, and which end is behind the NAT. The location of the payloads in the IKE_SA_INIT packets is just after the Ni and Nr payloads (before the optional CERTREQ payload).
-
The data associated with the NAT_DETECTION_SOURCE_IP notification is a SHA-1 digest of the SPIs (in the order they appear in the header), IP address, and port from which this packet was sent.
There MAY be multiple NAT_DETECTION_SOURCE_IP payloads in a message if the sender does not know which of several network attachments will be used to send the packet.
-
The data associated with the NAT_DETECTION_DESTINATION_IP notification is a SHA-1 digest of the SPIs (in the order they appear in the header), IP address, and port to which this packet was sent.
-
The recipient of either the NAT_DETECTION_SOURCE_IP or NAT_DETECTION_DESTINATION_IP notification MAY compare the supplied value to a SHA-1 hash of the SPIs, source or recipient IP address, and port (respectively), and if they don't match, it SHOULD enable NAT traversal. In the case there is a mismatch of the NAT_DETECTION_SOURCE_IP hash with all of the NAT_DETECTION_SOURCE_IP payloads received, the recipient MAY reject the connection attempt if NAT traversal is not supported. In the case of a mismatching NAT_DETECTION_DESTINATION_IP hash, it means that the system receiving the NAT_DETECTION_DESTINATION_IP payload is behind a NAT and that system SHOULD start sending keepalive packets as defined in [UDPENCAPS]; alternately, it MAY reject the connection attempt if NAT traversal is not supported.
-
If none of the NAT_DETECTION_SOURCE_IP payload(s) received matches the expected value of the source IP and port found from the IP header of the packet containing the payload, it means that the system sending those payloads is behind a NAT (i.e., someone along the route changed the source address of the original packet to match the address of the NAT box). In this case, the system receiving the payloads should allow dynamic updates of the other system's IP address, as described later.
-
The IKE initiator MUST check the NAT_DETECTION_SOURCE_IP or NAT_DETECTION_DESTINATION_IP payloads if present, and if they do not match the addresses in the outer packet, MUST tunnel all future IKE and ESP packets associated with this IKE SA over UDP port 4500.
-
To tunnel IKE packets over UDP port 4500, the IKE header has four octets of zeros prepended and the result immediately follows the UDP header. To tunnel ESP packets over UDP port 4500, the ESP header immediately follows the UDP header. Since the first four octets of the ESP header contain the SPI, and the SPI cannot validly be zero, it is always possible to distinguish ESP and IKE messages.
-
Implementations MUST process received UDP-encapsulated ESP packets even when no NAT was detected.
-
The original source and destination IP address required for the transport mode TCP and UDP packet checksum fixup (see [UDPENCAPS]) are obtained from the Traffic Selectors associated with the exchange. In the case of transport mode NAT traversal, the Traffic Selectors MUST contain exactly one IP address, which is then used as the original IP address. This is covered in greater detail in Section 2.23.1.
-
There are cases where a NAT box decides to remove mappings that are still alive (for example, the keepalive interval is too long, or the NAT box is rebooted). This will be apparent to a host if it receives a packet whose integrity protection validates, but has a different port, address, or both from the one that was associated with the SA in the validated packet. When such a validated packet is found, a host that does not support other methods of recovery such as IKEv2 Mobility and Multihoming (MOBIKE) [MOBIKE], and that is not behind a NAT, SHOULD send all packets (including retransmission packets) to the IP address and port in the validated packet, and SHOULD store this as the new address and port combination for the SA (that is, they SHOULD dynamically update the address). A host behind a NAT SHOULD NOT do this type of dynamic address update if a validated packet has different port and/or address values because it opens a possible DoS attack (such as allowing an attacker to break the connection with a single packet). Also, dynamic address update should only be done in response to a new packet; otherwise, an attacker can revert the addresses with old replayed packets. Because of this, dynamic updates can only be done safely if replay protection is enabled. When IKEv2 is used with MOBIKE, dynamically updating the addresses described above interferes with MOBIKE's way of recovering from the same situation. See Section 3.8 of [MOBIKE] for more information.
2.23.1. Transport Mode NAT Traversal
Transport mode used with NAT traversal requires special handling of the Traffic Selectors used in the IKEv2. The complete scenario looks like:
+------+ +------+ +------+ +------+
|Client| IP1 | NAT | IPN1 IPN2 | NAT | IP2 |Server|
|node |<------>| A |<---------->| B |<------->| |
+------+ +------+ +------+ +------+
(Other scenarios are simplifications of this complex case, so this discussion uses the complete scenario.)
In this scenario, there are two address translating NATs: NAT A and NAT B. NAT A is a dynamic NAT that maps the client's source address IP1 to IPN1. NAT B is a static NAT configured so that connections coming to IPN2 address are mapped to the gateway's address IP2, that is, IPN2 destination address is mapped to IP2. This allows the client to connect to a server by connecting to the IPN2. NAT B does not necessarily need to be a static NAT, but the client needs to know how to connect to the server, and it can only do that if it somehow knows the outer address of the NAT B, that is, the IPN2 address. If NAT B is a static NAT, then its address can be configured to the client's configuration. Another option would be to find it using some other protocol (like DNS), but that is outside of scope of IKEv2.
In this scenario, both the client and server are configured to use transport mode for the traffic originating from the client node and destined to the server.
When the client starts creating the IKEv2 SA and Child SA for sending traffic to the server, it may have a triggering packet with source IP address of IP1, and a destination IP address of IPN2. Its Peer Authorization Database (PAD) and SPD needs to have a configuration matching those addresses (or wildcard entries covering them). Because this is transport mode, it uses exactly same addresses as the Traffic Selectors and outer IP address of the IKE packets. For transport mode, it MUST use exactly one IP address in the TSi and TSr payloads. It can have multiple Traffic Selectors if it has, for example, multiple port ranges that it wants to negotiate, but all TSi entries must use the IP1-IP1 range as the IP addresses, and all TSr entries must have the IPN2-IPN2 range as IP addresses. The first Traffic Selector of TSi and TSr SHOULD have very specific Traffic Selectors including protocol and port numbers, such as from the packet triggering the request.
NAT A will then replace the source address of the IKE packet from IP1 to IPN1, and NAT B will replace the destination address of the IKE packet from IPN2 to IP2, so when the packet arrives to the server it will still have the exactly same Traffic Selectors that were sent by the client, but the IP address of the IKE packet has been replaced by IPN1 and IP2.
When the server receives this packet, it normally looks in the Peer Authorization Database (PAD) described in RFC 4301 [IPSECARCH] based on the ID and then searches the SPD based on the Traffic Selectors. Because IP1 does not really mean anything to the server (it is the address client has behind the NAT), it is useless to do a lookup based on that if transport mode is used. On the other hand, the server cannot know whether transport mode is allowed by its policy before it finds the matching SPD entry.
In this case, the server should first check that the initiator requested transport mode, and then do address substitution on the Traffic Selectors. It needs to first store the old Traffic Selector IP addresses to be used later for the incremental checksum fixup (the IP address in the TSi can be stored as the original source address and the IP address in the TSr can be stored as the original destination address). After that, if the other end was detected as being behind a NAT, the server replaces the IP address in TSi payloads with the IP address obtained from the source address of the IKE packet received (that is, it replaces IP1 in TSi with IPN1). If the server's end was detected to be behind NAT, it replaces the IP address in the TSr payloads with the IP address obtained from the destination address of the IKE packet received (that is, it replaces IPN2 in TSr with IP2).
After this address substitution, both the Traffic Selectors and the IKE UDP source/destination addresses look the same, and the server does SPD lookup based on those new Traffic Selectors. If an entry is found and it allows transport mode, then that entry is used. If an entry is found but it does not allow transport mode, then the server MAY undo the address substitution and redo the SPD lookup using the original Traffic Selectors. If the second lookup succeeds, the server will create an SA in tunnel mode using real Traffic Selectors sent by the other end.
This address substitution in transport mode is needed because the SPD is looked up using the addresses that will be seen by the local host. This will also ensure that the Security Association Database (SAD) entries for the tunnel exit checks and return packets are added using the addresses as seen by the local operating system stack.
The most common case is that the server's SPD will contain wildcard entries matching any addresses, but this also allows making different SPD entries, for example, for different known NATs' outer addresses.
After the SPD lookup, the server will do Traffic Selector narrowing based on the SPD entry it found. It will again use the already substituted Traffic Selectors, and it will thus send back Traffic Selectors having IPN1 and IP2 as their IP addresses; it can still narrow down the protocol number or port ranges used by the Traffic Selectors. The SAD entry created for the Child SA will have the addresses as seen by the server, namely IPN1 and IP2.
When the client receives the server's response to the Child SA, it will do similar processing. If the transport mode SA was created, the client can store the original returned Traffic Selectors as original source and destination addresses. It will replace the IP addresses in the Traffic Selectors with the ones from the IP header of the IKE packet: it will replace IPN1 with IP1 and IP2 with IPN2. Then, it will use those Traffic Selectors when verifying the SA against sent Traffic Selectors, and when installing the SAD entry.
A summary of the rules for NAT traversal in transport mode is:
For the client proposing transport mode:
-
The TSi entries MUST have exactly one IP address, and that MUST match the source address of the IKE SA.
-
The TSr entries MUST have exactly one IP address, and that MUST match the destination address of the IKE SA.
-
The first TSi and TSr Traffic Selectors SHOULD have very specific Traffic Selectors including protocol and port numbers, such as from the packet triggering the request.
-
There MAY be multiple TSi and TSr entries.
-
If transport mode for the SA was selected (that is, if the server included USE_TRANSPORT_MODE notification in its response):
-
Store the original Traffic Selectors as the received source and destination address.
-
If the server is behind a NAT, substitute the IP address in the TSr entries with the remote address of the IKE SA.
-
If the client is behind a NAT, substitute the IP address in the TSi entries with the local address of the IKE SA.
-
Do address substitution before using those Traffic Selectors for anything other than storing original content of them. This includes verification that Traffic Selectors were narrowed correctly by the other end, creation of the SAD entry, and so on.
-
For the responder, when transport mode is proposed by client:
-
Store the original Traffic Selector IP addresses as received source and destination address, in case undo address substitution is needed, to use as the "real source and destination address" specified by [UDPENCAPS], and for TCP/UDP checksum fixup.
-
If the client is behind a NAT, substitute the IP address in the TSi entries with the remote address of the IKE SA.
-
If the server is behind a NAT, substitute the IP address in the TSr entries with the local address of the IKE SA.
-
Do PAD and SPD lookup using the ID and substituted Traffic Selectors.
-
If no SPD entry was found, or (if found) the SPD entry does not allow transport mode, undo the Traffic Selector substitutions. Do PAD and SPD lookup again using the ID and original Traffic Selectors, but also searching for tunnel mode SPD entry (that is, fall back to tunnel mode).
-
However, if a transport mode SPD entry was found, do normal traffic selection narrowing based on the substituted Traffic Selectors and SPD entry. Use the resulting Traffic Selectors when creating SAD entries, and when sending Traffic Selectors back to the client.
2.24. Explicit Congestion Notification (ECN)
When IPsec tunnels behave as originally specified in [IPSECARCH-OLD], ECN usage is not appropriate for the outer IP headers because tunnel decapsulation processing discards ECN congestion indications to the detriment of the network. ECN support for IPsec tunnels for IKEv1-based IPsec requires multiple operating modes and negotiation (see [ECN]). IKEv2 simplifies this situation by requiring that ECN be usable in the outer IP headers of all tunnel mode Child SAs created by IKEv2. Specifically, tunnel encapsulators and decapsulators for all tunnel mode SAs created by IKEv2 MUST support the ECN full-functionality option for tunnels specified in [ECN] and MUST implement the tunnel encapsulation and decapsulation processing specified in [IPSECARCH] to prevent discarding of ECN congestion indications.
2.25. Exchange Collisions
Because IKEv2 exchanges can be initiated by either peer, it is possible that two exchanges affecting the same SA partly overlap. This can lead to a situation where the SA state information is temporarily not synchronized, and a peer can receive a request that it cannot process in a normal fashion.
Obviously, using a window size greater than 1 leads to more complex situations, especially if requests are processed out of order. This section concentrates on problems that can arise even with a window size of 1, and recommends solutions.
A TEMPORARY_FAILURE notification SHOULD be sent when a peer receives a request that cannot be completed due to a temporary condition such as a rekeying operation. When a peer receives a TEMPORARY_FAILURE notification, it MUST NOT immediately retry the operation; it MUST wait so that the sender may complete whatever operation caused the temporary condition. The recipient MAY retry the request one or more times over a period of several minutes. If a peer continues to receive TEMPORARY_FAILURE on the same IKE SA after several minutes, it SHOULD conclude that the state information is out of sync and close the IKE SA.
A CHILD_SA_NOT_FOUND notification SHOULD be sent when a peer receives a request to rekey a Child SA that does not exist. The SA that the initiator attempted to rekey is indicated by the SPI field in the Notify payload, which is copied from the SPI field in the REKEY_SA notification. A peer that receives a CHILD_SA_NOT_FOUND notification SHOULD silently delete the Child SA (if it still exists) and send a request to create a new Child SA from scratch (if the Child SA does not yet exist).
2.25.1. Collisions while Rekeying or Closing Child SAs
If a peer receives a request to rekey a Child SA that it is currently trying to close, it SHOULD reply with TEMPORARY_FAILURE. If a peer receives a request to rekey a Child SA that it is currently rekeying, it SHOULD reply as usual, and SHOULD prepare to close redundant SAs later based on the nonces (see Section 2.8.1). If a peer receives a request to rekey a Child SA that does not exist, it SHOULD reply with CHILD_SA_NOT_FOUND.
If a peer receives a request to close a Child SA that it is currently trying to close, it SHOULD reply without a Delete payload (see Section 1.4.1). If a peer receives a request to close a Child SA that it is currently rekeying, it SHOULD reply as usual, with a Delete payload. If a peer receives a request to close a Child SA that does not exist, it SHOULD reply without a Delete payload.
If a peer receives a request to rekey the IKE SA, and it is currently creating, rekeying, or closing a Child SA of that IKE SA, it SHOULD reply with TEMPORARY_FAILURE.
2.25.2. Collisions while Rekeying or Closing IKE SAs
If a peer receives a request to rekey an IKE SA that it is currently rekeying, it SHOULD reply as usual, and SHOULD prepare to close redundant SAs and move inherited Child SAs later based on the nonces (see Section 2.8.2). If a peer receives a request to rekey an IKE SA that it is currently trying to close, it SHOULD reply with TEMPORARY_FAILURE.
If a peer receives a request to close an IKE SA that it is currently rekeying, it SHOULD reply as usual, and forget about its own rekeying request. If a peer receives a request to close an IKE SA that it is currently trying to close, it SHOULD reply as usual, and forget about its own close request.
If a peer receives a request to create or rekey a Child SA when it is currently rekeying the IKE SA, it SHOULD reply with TEMPORARY_FAILURE. If a peer receives a request to delete a Child SA when it is currently rekeying the IKE SA, it SHOULD reply as usual, with a Delete payload.