RE: [RFC PATCH] net/tipc: reduce tipc_node lock holding time in tipc_rcv

[xu]

>>>This patch is wrong. le->link and link status must be protected by node lock. See what happens if tipc_node_timeout() is called, and
>>the link goes down:
>>>tipc_node_timeout()
>>>   tipc_node_link_down()
>>>   {
>>>      struct tipc_link *l = le->link;
>>>      ...
>>>      if (delete) {
>>>         kfree(l);
>>>         le->link = NULL;
>>>      }
>>>      ...
>>>   }
>>
>>Happy to see your reply. But Why? 'delete' is false from tipc_node_timeout(). Refer to:
>>https://elixir.bootlin.com/linux/v6.7-rc2/source/net/tipc/node.c#L844
>I should have explained it clearly:

Thanks. I see, so the root reason for holding node lock before tipc_link_rcv is to protect
links from being reset or deleted when tipc_link_rcv().

For further discussion, to balance incoming packets (the same links, different ports) to
multi-CPUs, maybe we can try RCU + spinlock here.

>1/ link status must be protected.
>tipc_node_timeout()
>   tipc_node_link_down()
>   {
>      struct tipc_link *l = le->link;
>   
>      ...
>     __tipc_node_link_down(); <-- link status is referred. 
>      ...
>     if (delete) {
>        kfree(l);
>        le->link = NULL;
>     }
>     ...
>   }
>
>__tipc_node_link_down()
>{
>    ...
>   if (!l || tipc_link_is_reset(l)) <-- read link status
>    ...
>    tipc_link_reset(l); <--- this function will reset all things related to link.
>}
>
>2/ le->link must be protected.
>bearer_disable()
>{
>   ...
>   tipc_node_delete_links(net, bearer_id); <--- this will delete all links.
>   ...
>}
>
>tipc_node_delete_links()
>{
>   ...
>  tipc_node_link_down(n, bearer_id, true);
>   ...
>}