| Network Working Group | R. Singh, Ed. |
| Internet-Draft | G. Kalyani |
| Intended status: Standards Track | Cisco |
| Expires: November 07, 2011 | Y. Nir |
| Check Point | |
| Y. Sheffer | |
| Porticor | |
| D. Zhang | |
| Huawei | |
| May 06, 2011 |
Protocol Support for High Availability of IKEv2/IPsec
draft-ietf-ipsecme-ipsecha-protocol-06
The IPsec protocol suite is widely used for business-critical network traffic. In order to make IPsec deployments highly available, more scalable and failure-resistant, they are often implemented as IPsec High Availability (HA) clusters. However there are many issues in IPsec HA clustering, and in particular in IKEv2 clustering. An earlier document, "IPsec Cluster Problem Statement", enumerates the issues encountered in the IKEv2/IPsec HA cluster environment. This document resolves these issues with the least possible change to the protocol.
This document defines an extension to the IKEv2 protocol to solve the main issues of "IPsec Cluster Problem Statement" in the commonly deployed hot-standby cluster, and provides implementation advice for other issues. The main issues solved are the synchronization of IKEv2 Message ID counters, and of IPsec Replay Counters.
This Internet-Draft is submitted in full conformance with the provisions of BCP 78 and BCP 79.
Internet-Drafts are working documents of the Internet Engineering Task Force (IETF). Note that other groups may also distribute working documents as Internet-Drafts. The list of current Internet- Drafts is at http://datatracker.ietf.org/drafts/current/.
Internet-Drafts are draft documents valid for a maximum of six months and may be updated, replaced, or obsoleted by other documents at any time. It is inappropriate to use Internet-Drafts as reference material or to cite them other than as "work in progress."
This Internet-Draft will expire on November 07, 2011.
Copyright (c) 2011 IETF Trust and the persons identified as the document authors. All rights reserved.
This document is subject to BCP 78 and the IETF Trust's Legal Provisions Relating to IETF Documents (http://trustee.ietf.org/license-info) in effect on the date of publication of this document. Please review these documents carefully, as they describe your rights and restrictions with respect to this document. Code Components extracted from this document must include Simplified BSD License text as described in Section 4.e of the Trust Legal Provisions and are provided without warranty as described in the Simplified BSD License.
The IPsec protocol suite, including IKEv2, is a major building block of virtual private networks (VPNs). In order to make such VPNs highly available, more scalable and failure-resistant, these VPNs are implemented as IKEv2/IPsec Highly Available (HA) clusters. However there are many issues with the IKEv2/IPsec HA cluster. Section 3 and Section 4 below expand on the issues around the IKEv2/IPsec HA cluster solution, issues which were first described in the Problem Statement [RFC6027].
In the case of a hot-standby cluster implementation of IKEv2/IPsec based VPNs, the IKEv2/IPsec session is first established between the peer and the active member of the cluster. Later, the active member continuously syncs/updates the IKE/IPsec SA state to the standby member of the cluster. This primary SA state sync-up takes place upon each SA bring-up and/or rekey. Performing the SA state synchronization/update for every single IKE and IPsec message is very costly, so normally it is done periodically. As a result, when the failover event happens, this is first detected by the standby member and, possibly after a considerable amount of time, it becomes the active member. During this failover process the peer is unaware of the failover event, and keeps sending IKE requests and IPsec packets to the cluster, as in fact it is allowed to do because of the IKEv2 windowing feature. After the newly-active member starts, it detects the mismatch in IKE Message ID values and IPsec replay counters and needs to resolve this situation. Please see Section 4 for more details of the problem.
This document defines an extension to the IKEv2 protocol to solve the main issues of IKE Message ID synchronization and IPsec SA replay counter synchronization, and gives implementation advice to address other issues. Following is a summary of the solutions provided in this document:
Although this document describes the IKEv2 Message ID and IPsec replay counter synchronization in the context of an IPsec HA cluster, the solution provided is generic and can be used in other scenarios where IKEv2 Message ID or IPsec SA replay counter synchronization may be required.
Implementations differ on the need to synchronize the IKEv2 Message ID and/or IPsec replay counters. Both of these problems are handled separately, using a separate notification for each capability. This provides the flexibility of implementing either or both of these solutions.
The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in [RFC2119].
"SA Counter Synchronization Request/Response" are the request viz. response of the informational exchange defined in this document to synchronize the IKEv2/IPsec SA counter information between one member of the cluster and the peer.
Some of the terms listed below are reused from [RFC6027] with further clarification in the context of the current document.
The generic term "IKEv2/IPsec SA Counters" is used throughout this document. This term refers to both IKEv2 Message ID counters and IPsec replay counters. According to the IPsec standards, the IKEv2 Message ID counter is mandatory, and used to ensure reliable delivery as well as to protect against message replay in IKEv2; the IPsec SA replay counters are optional, and are used to provide the IPsec anti-replay feature.
Some of these terms are used in the following architectural diagram.
+---------------+
| |
| Hot Standby |
| Cluster |
| |
| +---------+ |
| | | |
| | Active | |
| | | |
| | Member | |
| | | |
| +---------+ |
| ^ |
+---------+ | Sync | |
| | | Channel | |
| IPsec | IKE/IPsec Traffic | | |
| | <=============================> | | |
| Peer | | | |
| | | | |
+---------+ | | |
| v |
| +---------+ |
| | | |
| | Standby | |
| | | |
| | Member | |
| | | |
| +---------+ |
+---------------+
The IPsec Cluster Problem Statement [RFC6027] enumerates the problems raised by IPsec clusters. The following table lists the problem statement's sections that are resolved by this document.
The main problem areas are solved using the protocol extension defined below, starting with Section 5; additionally, this section provides implementation advice for other issues in the following subsections. Implementers should note that these subsections include a number of new security-critical requirements.
Section 3.2 of the Problem Statement mentions that a lot of state needs to be synchronized for a cluster to be transparent. The actual volume of that data is very much implementation-dependent, and even for the same implementation, the amounts of data may vary wildly. An IPsec gateway used for inter-domain VPN with a dozen other gateways, and having SAs that are rekeyed every 8 hours, will need a lot less synchronization traffic than a similar gateway used for remote access, and supporting 10,000 clients. This is because counter synchronization is proportional to the number of SAs and requires little data, and the setting up of an SA requires a lot of data. Additionally, remote access IKE and IPsec SA setup tend to happen at a particular time of day, so the example gateway with the 10,000 clients may see 30-50 IKE SA setups per second at 9:00 AM. This would require very heavy synchronization traffic over that short period of time.
If a large volume of traffic is necessary, it may be advisable to use a dedicated high-speed network interface for synch traffic. When packet loss can be made extremely low, it may be advisable to use a stateless transport such as UDP, to minimize network overhead.
If these methods are insufficient, it may be prudent that for some SAs the entire state is not synchronized. Instead, only an indication of the SA's existence is synchronized. This, in combination with a sticky solution (as described in section 3.7 of the problem statement) ensures that the traffic from a particular peer does not reach a different member before an actual failover happens. When that happens, the method described in [I-D.ietf-ipsecme-failure-detection] can be used to quickly force the peer to set up a new SA.
In a load-sharing cluster of the "duplicate" variety (see section 3.7 of the problem statement) multiple members may need to send traffic with the same selectors. To actually use the same SA the cluster would have to synchronize the Replay Counter after every packet, and that would impose unreasonable requirements on the synch connection.
A far better solution would be to not synchronize the outbound SA, and create multiple outbound SAs, one for each member. The problem with this option is that the peer might view these multiple parallel SAs as redundant, and tear down all but one of them.
Section 2.8 of [RFC5996] specifically allows multiple parallel SAs, but the reason given for this is to have multiple SAs with different QoS attributes. So while this is not a new requirement of IKEv2 implementations working with QoS, we re-iterate here that IPsec peers MUST accept the long-term existence of multiple parallel SAs, even when QoS mechanisms are not in use.
Section 3.9 of the problem statement describes the problem of two cluster members allocating the same SPI number for two different SAs. This would violate section 4.4.2.1 of [RFC4301]. There are several schemes to allow implementations to avoid such collisions, such as partitioning the SPI space, a request-response over the synch channel, and locking mechanisms. We believe that these are sufficiently robust and available so that we don't need to make an exception to RFC 4301, and we can leave this problem for the implementations to solve. Cluster members must not generate multiple inbound SAs with the same SPI.
For SAs involving counter mode ciphers such as CTR [RFC3686] or GCM [RFC4106] there is yet another complication. The initial vector for such modes MUST NOT be repeated, and senders may use methods such as counters or LFSRs to ensure this property. For an SA shared between multiple active members (load sharing cases), implementations MUST ensure that no initial vector is ever repeated. Similar concerns apply to an SA failing over from one member to another. See [RFC6054] for a discussion of this problem in another context.
Just as in the SPI collision problem, there are ways to avoid a collision of initial vectors, and this is left up to implementations. In the context of load sharing, parallel SAs are a simple solution to this problem as well.
The IKEv2 protocol [RFC5996] states that "An IKE endpoint MUST NOT exceed the peer's stated window size for transmitted IKE requests".
All IKEv2 messages are required to follow a request-response paradigm. The initiator of an IKEv2 request MUST retransmit the request, until it has received a response from the peer. IKEv2 introduces a windowing mechanism that allows multiple requests to be outstanding at a given point of time, but mandates that the sender's window should not move until the oldest message it has sent is acknowledged. Loss of even a single message leads to repeated retransmissions followed by an IKEv2 SA teardown if the retransmissions remain unacknowledged.
An IPsec Hot Standby Cluster is required to ensure that in the case of failover, the standby member becomes active immediately. The standby member is expected to have the exact value of the Message ID counter as the active member had before failover. Even assuming the best effort to update the Message ID values from active to standby member, the values at the standby member can still be stale due to the following reasons:
When a standby member takes over as the active member, it can only initialize the Message ID values from the previously updated values. This would make it reject requests from the peer when these values are stale. Conversely, the standby member may end up reusing a stale Message ID value which would cause the peer to drop the request. Eventually there is a high probability of the IKEv2 and corresponding IPsec SAs getting torn down simply because of a transitory Message ID mismatch and retransmission of requests, negating the benefits of the high availability cluster despite the periodic update between the cluster members.
A similar issue is also observed with IPsec anti-replay counters if anti-replay protection is enabled, which is commonly the case. Regardless of how well the ESP and AH SA counters are synchronized from the active to the standby member, there is a chance that the standby member would end up with stale counter values. The standby member would then use those stale counter values when sending IPsec packets. The peer would drop such packets since when the anti-replay protection feature is enabled, duplicate use of counters is not allowed. Note that IPsec allows the sender to skip some counter values and continue sending with higher counter values.
We conclude that a mechanism is required to ensure that the standby member has correct Message ID and IPsec counter values when it becomes active, so that sessions are not torn down as a result of mismatched counters.
This document defines two separate approaches to resolving the issues of mismatched IKE Message ID values and IPsec counter values.
Although conceptually separate, the two synchronization processes would typically take place simultaneously.
First, the peer and the active member of the cluster negotiate their ability to support IKEv2 Message ID synchronization and/or IPsec Replay Counter synchronization. This is done by exchanging one or both of the IKEV2_MESSAGE_ID_SYNC_SUPPORTED and IPSEC_REPLAY_COUNTER_SYNC_SUPPORTED notifications during the IKE_AUTH exchange. When negotiating these capabilities, the responder MUST NOT assert support of a capability unless such support was asserted by the initiator. Only a capability whose support was asserted by both parties can be used during the lifetime of the SA. The peer's capabilities with regard to this extension are part of the IKEv2 SA state, and thus MUST be shared between the cluster members.
This per-IKE SA information is shared with the other cluster members.
Peer Active Member
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
HDR, SK {IDi, [CERT], [CERTREQ], [IDr], AUTH,
[N(IKEV2_MESSAGE_ID_SYNC_SUPPORTED),]
[N(IPSEC_REPLAY_COUNTER_SYNC_SUPPORTED),]
SAi2, TSi, TSr} ---------->
<-------- HDR, SK {IDr, [CERT+], [CERTREQ+], AUTH,
[N(IKEV2_MESSAGE_ID_SYNC_SUPPORTED),]
[N(IPSEC_REPLAY_COUNTER_SYNC_SUPPORTED),] SAr2, TSi, TSr}
After a failover event, the standby member MAY use the IKE Message ID and/or IPsec Replay Counter synchronization capability when it becomes the active member, and provided support for the capabilities used has been negotiated. Following that, the peer MUST respond to any synchronization message it receives from the newly-active cluster member, subject to the rules noted below.
After the failover event, when the standby member becomes active, it has to synchronize its SA counters with the peer. There are now four possible cases:
This figure contains the IKE message exchange used for SA counter synchronization. The following subsections describe the details of the sender and receiver processing of each message.
Standby [Newly Active] Member Peer
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
HDR, SK {N(IKEV2_MESSAGE_ID_SYNC),
[N(IPSEC_REPLAY_COUNTER_SYNC)]} -------->
<--------- HDR, SK {N(IKEV2_MESSAGE_ID_SYNC)}
Alternatively, if only IPsec Replay Counter synchronization is desired, a normal Informational exchange is used, where the Message ID is non-zero:
Standby [Newly Active] Member Peer
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
HDR, SK{N(IPSEC_REPLAY_COUNTER_SYNC)} -------->
<--------- HDR
The newly-active member sends a request containing two counter values, one for the member (itself) and another for the peer, as well as a random nonce. We denote the values M1 and P1. The peer responds with a message containing two counter values, M2 and P2 (note that the values appear in the opposite order in the notification's payload). The goal of the rules below is to prevent an attacker from replaying a synchronization message, thereby invalidating IKE messages that are currently in process.
Upon failover, the newly-active member MUST increment its own Replay Counter (the counter used for outgoing traffic), so as to prevent the case of its traffic being dropped by the peer as replay. We note that IPsec allows the replay counter to skip forward by any amount. The estimate is based on the outgoing IPsec bandwidth and the frequency of synchronization between cluster members. In those implementations where it is difficult to estimate this value, the counter can be incremented by a very large number, e.g. 2**30. In the latter case, a rekey SHOULD follow shortly afterwards, to ensure that the counter never wraps around.
Next, the cluster member estimates the number of incoming messages it might have missed, using similar logic. The member sends out a IPSEC_REPLAY_COUNTER_SYNC notification, either stand-alone or together with a IKEV2_MESSAGE_ID_SYNC notification.
If the IPSEC_REPLAY_COUNTER_SYNC is included in the same message as IKEV2_MESSAGE_ID_SYNC, the peer MUST process the Message ID notification first (which might cause the entire message to be dropped as a replay). Then, it MUST increment the replay counters for all Child SAs associated with the current IKE SA by the amount requested by the cluster member.
This section lists the new notification payload types defined by this extension.
All multi-octet fields representing integers are laid out in big endian order (also known as "most significant byte first", or "network byte order").
This notification payload is included in the IKE_AUTH request/response to indicate support of the IKEv2 Message ID synchronization mechanism described in this document.
1 2 3
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| Next Payload |C| RESERVED | Payload Length |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|Protocol ID(=0)| SPI Size (=0) | Notify Message Type |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
The 'Next Payload', 'Payload Length', 'Protocol ID', 'SPI Size', and 'Notify Message Type' fields are the same as described in Section 3 of [RFC5996]. The 'SPI Size' field MUST be set to 0 to indicate that the SPI is not present in this message. The 'Protocol ID' MUST be set to 0, since the notification is not specific to a particular security association. The 'Payload Length' field is set to the length in octets of the entire payload, including the generic payload header. The 'Notify Message Type' field is set to indicate IKEV2_MESSAGE_ID_SYNC_SUPPORTED, value TBD by IANA. There is no data associated with this notification.
This notification payload is included in the IKE_AUTH request/response to indicate support for the IPsec SA Replay Counter synchronization mechanism described in this document.
1 2 3
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| Next Payload |C| RESERVED | Payload Length |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|Protocol ID(=0)| SPI Size (=0) | Notify Message Type |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
The 'Next Payload', 'Payload Length', 'Protocol ID', 'SPI Size', and 'Notify Message Type' fields are the same as described in Section 3 of [RFC5996] . The 'SPI Size' field MUST be set to 0 to indicate that the SPI is not present in this message. The 'Protocol ID' MUST be set to 0, since the notification is not specific to a particular security association. The 'Payload Length' field is set to the length in octets of the entire payload, including the generic payload header. The 'Notify Message Type' field is set to indicate IPSEC_REPLAY_COUNTER_SYNC_SUPPORTED, value TBD by IANA. There is no data associated with this notification.
This notification payload type (value TBD by IANA) is defined to synchronize the IKEv2 Message ID values between the newly-active (formerly standby) cluster member and the peer.
1 2 3
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| Next Payload |C| RESERVED | Payload Length |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|Protocol ID(=0)| SPI Size (=0) | Notify Message Type |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| Nonce Data |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| EXPECTED_SEND_REQ_MESSAGE_ID |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| EXPECTED_RECV_REQ_MESSAGE_ID |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
It contains the following data.
This notification payload type (value TBD by IANA) is defined to synchronize the IPsec SA Replay Counters between the newly-active (formerly standby) cluster member and the peer. Since there may be numerous IPsec SAs established under a single IKE SA, we do not directly synchronize the value of each one. Instead, a delta value is sent and all Replay Counters for Child SAs of this IKE SA are incremented by the same value. Note that this solution requires that either all Child SAs use Extended Sequence Numbers or else that no Child SA uses Extended Sequence Numbers [RFC4301]. This notification is only sent by the cluster.
1 2 3
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| Next Payload |C| RESERVED | Payload Length |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|Protocol ID(=0)| SPI Size (=0) | Notify Message Type |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| Incoming IPsec SA delta value |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
The notification payload contains the following data.
This protocol does not change any of the existing IKEv2 rules regarding Message ID values.
The standby member can initiate the synchronization of IKEv2 Message ID's under different circumstances.
To clarify the first alternative: the normal IKE behavior of rejecting out-of-window messages is not changed, but such messages can still be a valid trigger for the exchange defined in this document. To avoid DoS attacks resulting from replayed messages, the peer MUST NOT initiate counter synchronization for any particular IKE SA more than once per failover event.
The standby member can initiate the synchronization of IPsec SA Replay Counters:
Since there can be a large number of sessions at the standby member, and sending synchronization exchanges for all of them may result in overload, the standby member can choose to initiate the exchange in a "lazy" fashion: only when it has to send or expects to receive traffic from each peer. In general, the standby member is free to initiate this exchange at its discretion. Implementation considerations include the ability to survive a certain amount of traffic loss, and the capacity of a cluster member to initiate counter synchronization simultaneously with a large number of peers.
The straightforward definitions of message sequence numbers, retransmissions and replay protection in IPsec and IKEv2 are strained by the failover scenarios described in this document. This section describes some policy choices that need to be made by implementations in this setting.
After sending its "receive" counter, the cluster member MUST reject (silently drop) any incoming IKE messages that are outside its declared window. A similar rule applies to the peer. Local policies vary, and strict implementations will reject any incoming IKE message arriving before Message ID synchronization is complete.
For IPsec, there is often a trade-off between security and reliability of the protected protocols. Here again there is some leeway for local policy. Some implementations might accept incoming traffic that is outside the replay window for some time after the failover event, and until the counters had been synchronized. Strict implementations will only accept traffic that's inside the "safe" window.
IKEv2 is normally a reliable protocol. As long as an IKE SA is valid, both peers share a single, consistent view of the IKE SA and all associated Child SAs. Failover situations as described in this document may involve forced deletion of IKE messages, resulting in inconsistencies, such as Child SAs that exist on only one of the peers. Such SAs might cause an INVALID_SPI to be returned when used by that peer. Note that Sec. 1.5 of [RFC5996] allows but does not mandate sending an INVALID_SPI notification in this case.
The Working Group discussed at some point a proposed set of rules for dealing with such situations. However we believe that these situations should be rare in practice; as a result the "default" behavior of tearing down the entire IKE SA is to be preferred over the complexity of dealing with a multitude of edge cases.
This section goes through the sequence of steps of a typical failover event, looking at a case where the IKEv2 Message ID values are synchronized.
The usage scenario of this IKEv2/IPsec SA counter synchronization solution is that an IKEv2 SA has been established between the active member of a hot-standby cluster and a peer, followed by a failover event occurring and the standby member becoming active. The solution further assumes that the IKEv2 SA state was continuously synchronized between the active and standby members of the cluster before the failover event.
Since Message ID synchronization messages need to be sent with Message ID zero, they are potentially vulnerable to replay attacks. Because of the semantics of this protocol, these can only be denial-of-service (DoS) attacks, and we are aware of two variants.
As mentioned in Section 7, trigerring counter synchronization by out-of-window, potentially replayed messages, could open a DoS vulnerability. This risk is mitigated by the solution described in that section.
This document introduces four new IKEv2 Notification Message types as described in Section 6. The new Notify Message Types must be assigned values between 16396 and 40959.
| Name | Value |
|---|---|
| IKEV2_MESSAGE_ID_SYNC_SUPPORTED | TBD by IANA |
| IPSEC_REPLAY_COUNTER_SYNC_SUPPORTED | TBD by IANA |
| IKEV2_MESSAGE_ID_SYNC | TBD by IANA |
| IPSEC_REPLAY_COUNTER_SYNC | TBD by IANA |
We would like to thank Pratima Sethi and Frederic Detienne for their review comments and valuable suggestions for the initial version of the document.
We would also like to thank the following people (in alphabetical order) for their review comments and valuable suggestions: Dan Harkins, Paul Hoffman, Steve Kent, Tero Kivinen, David McGrew, and Pekka Riikonen.
This section lists all the changes in this document.
NOTE TO RFC EDITOR: Please remove this section before publication.
Applied multiple review comments, from Pekka Riikonen, Alexey Melnikov, Stephen Farrel, Robert Sparks, Pete Resnick, Russ Housley and Adrian Farrel. Added an architectural reference diagram. Added a MUST requirement for cluster members to share peers' support of this protocol, which had been implicit in previous versions.
Applied Sean Turner's review comments.
Extended Sec. 3 for better coverage of other IPsec cluster-related issues, and how they are resolved within the existing standards.
Clarified the rules for Message ID sync, so that replay attacks can be avoided without a failover counter.
Added wording regarding inconsistent IKE state (basically choosing to ignore the problem) and further rules dealing with pending traffic.
The IPsec replay counter delta value now refers to incoming traffic. The associated notification is only sent from the cluster to the peer, and not back.
Addressed comments by Yaron Sheffer posted on the WG mailing list.
Numerous editorial changes.
Added "Multiple and Simultaneous failover" scenarios as pointed out by Pekka Riikonen.
Now document provides a mechanism to sync either IKEv2 message or IPsec replay counter or both to cater different types of implementations.
HA cluster's "failover count' is used to encounter replay of sync requests by attacker.
The sync of IPsec SA replay counter optimized to to have just one global bumped-up outgoing IPsec SA counter of ALL Child SAs under an IKEv2 SA.
The examples added for IKEv2 Message ID sync to provide more clarity.
Some edits as per comments on mailing list to enhance clarity.
Version 00 is identical to draft-kagarigi-ipsecme-ikev2-windowsync-04, started as WG document.
Added IPSECME WG HA design team members as authors.
Added comment in Introduction to discuss the window sync process on WG mailing list to solve some concerns.
| [1] | Bradner, S., "Key words for use in RFCs to Indicate Requirement Levels", BCP 14, RFC 2119, March 1997. |
| [2] | Kent, S. and K. Seo, "Security Architecture for the Internet Protocol", RFC 4301, December 2005. |
| [3] | Kaufman, C., Hoffman, P., Nir, Y. and P. Eronen, "Internet Key Exchange Protocol Version 2 (IKEv2)", RFC 5996, September 2010. |
This (non-normative) section presents some examples that illustrate how the IKEv2 Message ID values are synchronized. We use a tuple notation, denoting the two counters EXPECTED_SEND_REQ_MESSAGE_ID and EXPECTED_RECV_REQ_MESSAGE_ID on each protocol party as (EXPECTED_SEND_REQ_MESSAGE_ID, EXPECTED_RECV_REQ_MESSAGE_ID).
Note that if the IKE message counters are already synchronized (as in the first example), we expect the numbers to be reversed between the two sides. If one protocol party intends to send the next request as 4, then the other expects the next received request to be 4.
Standby (Newly Active) Member Peer
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
Sync Request (0, 5) -------->
Peer has the values (5, 0) so it sends
<------------- (5, 0) as the Sync Response
In this example, the peer has most recently sent an IKE request with Message ID 4, and has never received a request. So the peer's expected values for the next pair of messages are (5, 0). These are the same values as received from the member and therefore they are sent as-is.
Standby (Newly Active) Member Peer
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
Sync Request (2, 3) -------->
Peer has the values (4, 5) so it sends
<------------- (4, 5) as the Sync Response
In this example, the peer has most recently sent an IKE message with the Message ID 3, and received one with ID 4. So the peer's expected values for the next pair of messages are (4, 5). These are both higher than the corresponding values just received from the member (the order of tuple members is reversed when doing this comparison!), and therefore they are sent as-is.
Standby (Newly Active) Member Peer
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
Sync Request (2, 5) -------->
Peer has the values (2, 4) so it sends
<-------------(5, 4) as the Sync Response
In this example, the newly active member expects to send the next IKE message with ID 2. It sends an expected receive value of 5, which is higher than the last ID value it has seen from the peer, because it believes some incoming messages may have been lost. The peer has last sent a message with ID 1, and received one with ID 3, indicating that the a couple of messages sent by the previously active member had not been synchronized into the other member. So the peer's next expected (send, receive) values are (2, 4). The peer replies with the maximum of the received and the expected value for both send and receive counters: (max(2, 5), max(4, 2)) = (5, 4).
In the case of simultaneous failover, both sides send their synchronization requests simultaneously. The eventual outcome of synchronization consists of the higher counter values. This is demonstrated in the following figure.
Standby (Newly Active) Member Peer
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
Sync Request (4,4) ----->
<-------------- Sync Request (5,5)
Sync Response (5,5) ---->
<-------- Sync Response (5,5)