|Main Archive Page > Month Archives > ipsec archives|
I agree that 2nd solution is easier to implement, but the peer will have to delete the impending requests it might retransmit to the standby device.
This will also not catch the message replay attacks which is the major advantage of using message Id's
(once the window is reset, earlier windows messages might be replayed which will fall in this window at some point of time).
Using a random number as starting message Id after reset, may also cause the same issue
with 1st solution we can start from the requests where the active device has stopped . This would ensure a smooth continuation of message Id's .
If the peer participating in the HA does not support this extension, then the ordinary method of updating the counters for every exchange has to be followed even though its inefficient.
I don't see any other solution with the current ikev2 protocol for this message Id problem. Hence I feel that for efficient updation of counters, this extension should be supported .
In case of site to site there will be some cases of single IKE SA and multiple IPSEC SA's but we cannot rule out the case of multiple IKE SA's depending on the deployment.
In addition , in case of remote access there might be many . In such cases if dpd is also enabled then solution #2 would create lots of message updations from
Active to standbydevice. The concern is not only abt the bandwidth between stand by and active , but the active device would be involved in only doing these updations to the stand-by device.
From: Yoav Nir [mailto:firstname.lastname@example.org]
Sent: Thursday, September 03, 2009 8:37 PM
To: Kalyani Garigipati (kagarigi)
Subject: Re: [IPsec] Ikev2 HA message Id Issue
Of the two, I prefer the 2nd solution, as it is simpler. Reusing message IDs is not that bad, and you can decrease the change by including (in the RESET_MESSAGE_ID notification) a random number as the starting message ID.
What I'm not so sure, is that there is a real problem here that needs to be solved now.
The IKEv2 documents totally ignore both high-availability solutions and load-sharing solutions. They are just out of scope. So the documents don't specify what data needs to be synched, or how is failover detected and accomplished on the LAN or the WAN.
To get there, you'd need to address issues of routing, signaling
(between the peers) and AAA for both IKE and IPsec traffic. That's a
tall order that you probably don't want to tackle just now (maybe the WG would want this as the next "big" document)
So to have a high-availability or load-sharing solution that participates in IKE/IPsec, an implementation needs to somehow pretend that both of these are actually one gateway. This can be done with smart switches, multicast addresses and synchronization links, which are out of scope for the WG (for now)
I can think of three levels of synchronizing IKE data between the peers.
#3 is obviously impractical. #1 doesn't work, because after fail-over, the redundant gateway cannot take over the IKE SA, because it doesn't know the message ID counters. #2 can be made to work, although you need to either immediately rekey all the child SAs, or else skip ahead in the IPsec retrans counters. This solution is practical enough, because most gateways have very few Child SAs per IKE SA. With IKEv2 you can have complex traffic selectors that allow one SA to cover all the domain protected by a gateway. So you might have a few SAs because of different QoS classes, and maybe a few if your implementation prefers to deal with whole ranges or subnets, but really, a single IKE SA with thousands of concurrent Child SAs is more of a lab thing than a practical thing.
Both your solutions ask peer gateways to assist the HA pair with their fail-over process. To the other gateway it seems as if the (single) peer is requesting information about current message ID numbers (and windows) or else for a reset. It seems strange that the first thing we would do for HA support is to help a private extension to the architecture work better, when that private extension is not really documented.
What do you plan to do in your cluster, if the peer does not support this extension?
You might also want to ask Paul and Yaron to present this on the Interim meeting on 22-Sep.
On Sep 3, 2009, at 4:06 PM, Kalyani Garigipati (kagarigi) wrote:
In Ikev2 HA, there is an issue with the message Id and window size. Standby device-----------------------active device----------------------------------Peer device
The active device participating in the exchange with the peer will update its message id counters as per the exchanges done.
This info cannot be synced to the stand-by device for every exchange done since that would take up all the bandwidth and is not an efficient way.
The stand-by device when it becomes active will start with the message Id as 1 and this will not be accepted by the peer, since its message Id counters are different.
Hence a solution is required to sync the message Id counters to the standby device.
The abstract details of how this can be done are given in the attached document.
2. An alternative solution for this could be to send a new notify called
(RESET_MESSAGE_ID) to the peer device as soon as the standby comes up.
But this may lead to
Reuse of message Id's within the same SA which is not desirable.
I think solution 1 should be implemented with Ikev2. Please give your comments