B3. Handling Loss of Synchronization

If there is an undetected packet loss of 2^32 or more consecutive packets on a single SA, then the transmitter and receiver will lose synchronization of the high-order bits, i.e., the equations in Appendix B2.2. will fail to yield the correct value. Unless this problem is detected and addressed, subsequent packets on this SA will fail authentication checks and be discarded. The following procedure SHOULD be implemented by any IPsec (ESP or AH) implementation that supports the ESN option.

Note that this sort of extended traffic loss seems unlikely to occur if any significant fraction of the traffic on the SA in question is TCP, because the source would fail to receive ACKs and would stop sending long before 2^32 packets had been lost. Also, for any bi-directional application, even ones operating above UDP, such an extended outage would likely result in triggering some form of timeout. However, a unidirectional application, operating over UDP, might lack feedback that would cause automatic detection of a loss of this magnitude, hence the motivation to develop a recovery method for this case.

The solution we've chosen was selected to:

• minimize the impact on normal traffic processing.
• avoid creating an opportunity for a new denial of service attack such as might occur by allowing an attacker to force diversion of resources to a re-synchronization process.
• limit the recovery mechanism to the receiver because anti-replay is a service only for the receiver, and the transmitter generally is not aware of whether the receiver is using sequence numbers in support of this optional service. It is preferable for recovery mechanisms to be local to the receiver. This also allows for backward compatibility.

B3.1. Triggering Re-synchronization

For each SA, the receiver records the number of consecutive packets that fail authentication. This count is used to trigger the re-synchronization process, which should be performed in the background or using a separate processor. Receipt of a valid packet on the SA resets the counter to zero. The value used to trigger the re-synchronization process is a local parameter. There is no requirement to support distinct trigger values for different SAs, although an implementer may choose to do so.

B3.2. Re-synchronization Process

When the above trigger point is reached, a "bad" packet is selected for which authentication is retried using successively larger values for the upper half of the sequence number (Seqh). These values are generated by incrementing by one for each retry. The number of retries should be limited, in case this is a packet from the "past" or a bogus packet. The limit value is a local parameter. (Because the Seqh value is implicitly placed after the AH (or ESP) payload, it may be possible to optimize this procedure by executing the integrity algorithm over the packet up to the endpoint of the payload, then compute different candidate ICVs by varying the value of Seqh.) Successful authentication of a packet via this procedure resets the consecutive failure count and sets the value of T to that of the received packet.

This solution requires support only on the part of the receiver, thereby allowing for backward compatibility. Also, because re-synchronization efforts would either occur in the background or utilize an additional processor, this solution does not impact traffic processing and a denial of service attack cannot divert resources away from traffic processing.