Re: [PATCH net 2/2] net/smc: align the connect behaviour with TCP

[Tony Lu]

On Tue, May 10, 2022 at 08:58:38PM +0800, Guangguan Wang wrote:
> 
> 
> On 2022/5/10 17:30, Tony Lu wrote:
> > On Mon, May 09, 2022 at 07:58:37PM +0800, Guangguan Wang wrote:
> >> Connect with O_NONBLOCK will not be completed immediately
> >> and returns -EINPROGRESS. It is possible to use selector/poll
> >> for completion by selecting the socket for writing. After select
> >> indicates writability, a second connect function call will return
> >> 0 to indicate connected successfully as TCP does, but smc returns
> > 
> > If the connection is established successfully, the following up call of
> > connect() returns -EISCONN (SS_CONNECTED), which is expected and SMC
> > does it, same as TCP.
> > 
> > In case of misunderstanding, could you append more detailed information?
> > 
> > Thanks,
> > Tony Lu
> > 
> 
> io_uring uses nonblocking connect as follow steps:
>   1) call connect with nonblocking
>   2) wait for selector/poll to indicate writability
>   3) call connect to confirm connection's state
> 
> In the third step, tcp changes the socket state from SS_CONNECTING to
> SS_CONNECTED and returns 0 if the connection is established successfully,
> but smc returns -EISCONN.

Based on the steps you list, I am wondering if it is finished in the
step #1, the call of connect() in step #3 would return -EISCONN. Should
we check 0 and -EISCONN in  step #3?

To fix this issue, I think we should be careful about adding and
handling sock state, maybe we could push it to net-next and take
advantage of sock state. And I will test this patch later in our test
cases.

Thanks,
Tony Lu