Synchronous Socket: unnecessary skb postponing

[krizhanovsky]

Based on Alexey's notes.

ss_tcp_process_proto_skb() calls postpone_skb() callback if connection_recv() returns SS_POSTPONE. postpone_skb() points to tfw_connection_postpone_skb() which just adds the skb to conn->msg->skb_list. Meantime, ss_tcp_process_skb() firstly calls put_skb_to_msg() callback before executing ss_tcp_process_proto_skb() and the callback points to tfw_connection_put_skb_to_msg() which also adds the skb to the same list. So postponing the skb has no sense.

[keshonok]

That is correct summary of my notes. Closely related is issue #31.

An incoming SKB is processed by ss_tcp_process_skb(). First thing it does is call put_skb_to_msg() callback that points to tfw_connection_put_skb_to_msg(). That function creates a new connection message if needed, and then adds the SKB to that message's list conn->msg->skb_list by executing ss_skb_queue_tail(&conn->msg->skb_list, skb);

Then ss_tcp_process_proto_skb() is called on every piece of data in the SKB. In turn, it calls the connection_recv() callback. On TFW_POSTPONE result from connection_recv() it calls postpone_skb() callback that points tfw_connection_postpone_skb(). The only purpose of tfw_connection_postpone_skb() is executing ss_skb_queue_tail(&conn->msg->skb_list, skb); - again, on the same SKB, the same message, and the same connection.

Granted, that doesn't do any harm as the SKB is not actually added to the same list twice due to safely measures.

All of that means the following:

• The whole if (r == SS_POSTPONE) case can safely be removed from ss_tcp_process_proto_skb(). That will eliminate unnecesary multiple calls to tfw_connection_postpone_skb().
• .postpone hook in ssocket_hooks{} is not needed as well.
• The whole tfw_connection_postpone_skb() function can safely be removed as unused and unnecessary.