TCP/IP vulnerability?

Security Now is a great podcast about all sorts of security topics. Nr 164 is about "Sockstress". There seems to be a serious problem in almost any tcp/ip stack, including those of routers! Steve Gibson (the security person driving this podcast) based himself on a Dutch podcast called "De beveiligingsupdate" ("Security update").

Having some understanding of networking, but not being a specialist, it seems that this attack is launched after the 3-way tcp/ip handshake is done. After such handshake, a reasonable amount of trust has been created, as the server knows the ip address of the client. And implicitely it assumes that the client will behave according to the tcp/ip rules.

So this attack only starts after the tcp/ip connection has been established. First of all, the client reduces its resource consumpption by encoding information about the connection in the sequence numbers in the headers of the packets. As such, it needn't keep state. Secondly, the client doesn't use the TCP/IP stack of the client machine itself but has an implementation in user space, based on raw sockets. And then it starts playing dirty tricks by e.g. responding to the server that it doesn't have any buffer space left. The server will wait a certain amount of time and try to resume sending. By forcing the server to manage this large set of connections with all the resource consumption - memory and timers - the TCP/IP service goes through its knees. And potentially the complete OS crashes! This problem and corresponding attack seems to be known for 3 years, but only now is it coming out in the open.

Anyway, this is the way I understood it. After the DNS poisining issue, this seems a very fundamental attack. If this story is true, and no countermeasures are found, this might become an important issue. Not only crisis in the financial world, but also a crisis in Internet land.

Note: there is a related Dutch podcast called "Ict roddels" (ICT gossip), recommended to native Dutch speakers