Re: 2.2.16 TCP problems (again)

[Rajeev Bector]

Hi  Alexey, 
    >> Open request is still on queue, so that it should return socket
    >> created by the first ACK without additional lookup in hash table.
| 
| That's question... why does it fail yet? 8)
| 

So,  I caught it. We had modified the stack to separate half-open and
established but unaccepted connections on different queues (we also
added a BSD like hashing scheme to half-open connections queue
so that we dont have to do a linear search).   

The  RST problem is being caused by our changes.  So, looks like
you guys did the same splitting (of half-open and established but
unaccepted connections) in 2.3 and thats why you said that you
saw the problem in 2.3.

Did the splitup bringup other races in the code ?

(By the way, we dont see this problem in the field .. I guess
 we will seee it only on heavily loaded servers where the clients
 are on low-latency links !)

Anyway, thanks a lot for your comments and help.

(adding the code to do a  established hash queue lookup
in backlog processing fixes the problem !)

Rajeev