[tcpm] TCP zero window timeout?

We recently ran into a customer case where the client was advertising a 
zero window towards our TCP stack in the middle of a stream. Subsequent 
probes were also resulting in a zero window advertisement.

While it was clear that the client application was where the problem 
was, our TCP implementation was left holding a lot of client data (we 
are a proxy). The result was that we ultimately ran out of buffers and 
stopped processing new connections.

This was a problem for our customer also and they were looking for a 
solution. So we decided to offer a fix in the form of TCP option that 
the customer could configure. The fix entailed using the idle timeout 
period and the normal persist backoff mechanism to determine if the 
client had made any progress (towards opening up the window), and if it 
had not, we would time out the connection with a RST, and reclaim all 
the buffers.

Would this be considered reasonable to make the server robust? Is there 
a RFC that talks about it? If not, would it be of interest to this group 
for me to write a draft, describing the changes we made?

[tcpm] TCP zero window timeout?

Attachment: mahesh.vcf