[rds-devel] [PATCH net-next] RDS: deliver zerocopy completion notification with data as an optimization

[Sowmini Varadhan]

On (02/21/18 18:45), Willem de Bruijn wrote:
> 
> I do mean returning 0 instead of -EAGAIN if control data is ready.
> Something like
> 
> @@ -611,7 +611,8 @@ int rds_recvmsg(struct socket *sock, struct msghdr
> *msg, size_t size,
> 
>                 if (!rds_next_incoming(rs, &inc)) {
>                         if (nonblock) {
> -                               ret = -EAGAIN;
> +                               ncookies = rds_recvmsg_zcookie(rs, msg);
> +                               ret = ncookies ? 0 : -EAGAIN;
>                                 break;
>                         }

Yes, but you now have an implicit branch based on ncookies, so I'm
not sure it saved all that much? like I said let me revisit this

> By the way, the put_cmsg is unconditional even if the caller did
> not supply msg_control. So it is basically no longer safe to ever
> call read, recv or recvfrom on a socket if zerocopy notifications
> are outstanding.

Wait, I thought put_cmsg already checks for these things. 

> It is possible to check msg_controllen before even deciding whether
> to try to dequeue notifications (and take the lock). I see that this is
> not common. But RDS of all cases seems to do this, in
> rds_notify_queue_get:

yes the comment above that code suggests that this bit of code
was done to avoid calling put_cmsg while holding the rs_lock.

One bit of administrivia though- if I now drop sk_error_queue for
PF_RDS, I'll have to fix selftests in the same patch too, so the
patch will get a bit bulky (and thus a bit more difficult to review).