5

I'm currently learning IKE and IPsec for an exam. I have a lot of information on how Security Parameter Indexes (SPI) are used in both protocols, but I'm having some problems figuring out the coherence.

First, in IKE, both parties share their SPI with the respective other party. I guess these will be the both SPIs used for the two Security Associations (SA), because each SA is only unidirectional, is that right?

Then, in the IPsec chapter, things start to get more complicated. Reading different sources, I have a theory how this works exactly, but I'm not sure if my theory is right. RFC 4301 says:

SPI: An arbitrary 32-bit value that is used by a receiver to identify the SA to which an incoming packet should be bound. [...]

In IKE, each party shares an SPI with the other party. Does this mean, that the SPI a party sets is the SPI used for the incoming SA, not the outgoing one?

This would explain another question I have regarding registrating a new SA in the SA-Database (SAD)

  • SAD_ADD is used when IKE already knows which SPI it wants to use for an SA.
  • SAD_GETSPI is used to get a non-used SPI (here the SAD returns a non-used SPI, because of course it knows which SPIs are used and which aren't). Additionally it already inserts an incomplete SA. The SAD_UPDATE is then used to to set the missing SPI to the previously inserted SA.

Additionally my notes say that the initiator uses the SAD_ADD method while the responder uses SAD_GETSPI and SAD_UPDATE. This would make sense, if the responder is the one creating an SPI, and the initiator only adds this SPI to his database. This only has to be unique for him together with the responder's IP address. Is my theory right?

However I don't understand why the SAD_UPDATE method is used. For me this sounds redundant:

  • SAD_GETSPI: "I want to insert an SA x, which SPI can I use? Answer: Here is an unused SPI: y. Additionally, I have already inserted x in my Database, just tell me which SPI I should insert."
  • SAD_UPDATE: "Please update SA x, set SPI to y."
10

Security Parameter Indexes (SPIs) can mean different things when referring to IKE and IPsec Security Associations (SAs):

  • For IKE two 64-bit SPIs uniquely identify an IKE SA. With IKEv2 the IKE_SA_INIT request will only have the locally unique initiator SPI set in the IKE header, the responder SPI is zero. The responder will set that to a likewise locally unique value in its response. The two SPIs will only change when the IKE SA is rekeyed.

    The two fields in the IKE header that are now called Initiator/Responder SPI were previously called Initiator/Responder Cookie in RFC 2408 (ISAKMP). This could be confusing as IKEv2 uses COOKIE notification payloads to thwart denial of service attacks.

  • For IPsec a 32-bit SPI semi-uniquely identifies an IPsec SA. Since these SAs are unidirectional the ESP/AH header contains only the SPI of the destination's inbound SA (unlike the IKE header which always contains both SPIs). Since the SPIs are locally unique this and the destination address is usually enough to uniquely identify an SA. But it could be problematic e.g. if two clients behind the same NAT allocate the same local SPI when they connect to the same VPN gateway. The combination of SPI and destination address would be the same on the public side of the NAT, which is why UDP encapsulation is required. The UDP ports allow the NAT to direct the inbound packets to the right client. Likewise, the gateway has to take measures to differentiate the two SAs so that the right SA is used when sending traffic to each client.

Does this mean, that the SPI a party sets is the SPI used for the incoming SA, not the outgoing one?

Yes, each peer sends the SPI of its inbound SA to the other peer.

Additionally my notes say that the initiator uses the SAD_ADD method while the responder uses SAD_GETSPI and SAD_UPDATE.

The process of establishing an IPsec SA using e.g. a CREATE_CHILD_SA exchange in IKEv2 could roughly be visualized like this:

Initiator                               Responder
SAD_GETSPI (inbound SA)  ----------->   {select algorithms and derive keys}
                                        SAD_ADD (outbound SA)
                                        SAD_GETSPI (inbound SA)
{derive keys}            <-----------   SAD_UPDATE (inbound SA) 
SAD_UPDATE (inbound SA)
SAD_ADD (outbound SA)
  1. The initiator sends the SPI of its inbound SA together with a proposal of cryptographic algorithms and, if perfect forward secrecy is used, its Diffie-Hellman factor, to the responder.
  2. The responder selects suitable algorithms and derives the keys (optionally using DH) and proceeds installing the SAs.
  3. Then it returns its inbound SPI together with the selected algorithms (and optionally its DH factor) to the initiator, which is now able to install the SAs on its side.

Additionally, the two peers exchange traffic selectors that specify the network traffic that is to be covered by the established SA.

SAD_UPDATE: "Please update SPI x, set SPI to y."

That's not what SAD_UPDATE does. It actually does not change the SPI at all, but rather all (or some) of the other aspects of the SA, and these are mainly the encryption/integrity algorithms and keys (but may also include other things, like encapsulation or the anti-replay window size).

The reason you usually want to call SAD_GETSPI and SAD_UPDATE instead of simply SAD_ADD for inbound SAs (even on the responder, where all the information would be available) is that the SAD is usually managed by the operating system's kernel, while IKE daemons operate in userland. And therefore, calling SAD_GETSPI will ensure that the SPI is actually locally unique. Which might not be guaranteed if e.g. two IKE daemons (for IKEv1 and IKEv2) or even tools to manually manage SAs (like ip xfrm or setkey) are used concurrently on a system.

But it is imaginable that on some systems there could be a simplification that would allow a responder to call SAD_ADD without specifying an SPI, and the SAD would then allocate one, install the SA, and return the new SPI. But that would require special handling for this particular case by the keying daemon (otherwise it could simply call SAD_ADD for outbound SAs and SAD_GETSPI/SAD_UPDATE for inbound SAs, no matter if it does so as initiator or responder).

RFC 2367 (PF_KEYv2) provides more information on these operations.

  • Just to get this right: IKE SPI and IPsec SPI are two different identifiers which are not related? Additinally IKE SPI was previously called Cookie, but in IKEv2 Cookie now means something different, which is used to prevent DoS Attacks, and this again is not realted to IPsec SPI (tools.ietf.org/html/rfc5996#section-2.6)? – Misch Apr 23 '14 at 14:54
  • Updated my answer. – ecdsa Apr 23 '14 at 15:49

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.