Re: TCP sockets not flagged as writable?

[kuznet@xxxxxxxxxxxxx]

Hello!

> But my concern is that when you have just a little bit of space
> available a write operation may succeed without blocking, even though
> poll() says it won't.

Moreover, write() can succeed even if there is no space _really_.


> Should tcp_sendmsg() not also rely on the 1/3 free requirement?

Which is fully equavalent to decreasing sndbuf by 1/3. 8)
Wake is delayed exaclty to batch work.

Alexey