Re: Re: [PATCH net-next] net: socket: add the case sock_no_xxx support

From: Cong Wang
Date: 2021-09-20 07:52
To: Yajun Deng
CC: Jakub Kicinski; David Miller; Linux Kernel Network Developers; LKML
Subject: Re: [PATCH net-next] net: socket: add the case sock_no_xxx support
On Sat, Sep 18, 2021 at 5:11 AM [off-list ref] wrote:

September 18, 2021 9:33 AM, "Jakub Kicinski" [off-list ref] wrote:

quoted

On Thu, 16 Sep 2021 20:29:43 +0800 Yajun Deng wrote:

quoted

Those sock_no_{mmap, socketpair, listen, accept, connect, shutdown,
sendpage} functions are used many times in struct proto_ops, but they are
meaningless. So we can add them support in socket and delete them in struct
proto_ops.

So the reason to do this is.. what exactly?

Removing a couple empty helpers (which is not even part of this patch)?

I'm not sold, sorry.

When we define a struct proto_ops xxx, we only need to assign meaningful member variables that we need.
Those {mmap, socketpair, listen, accept, connect, shutdown, sendpage} members we don't need assign
it if we don't need. We just need do once in socket, not in every struct proto_ops.

These members are assigned meaningless values far more often than meaningful ones, so this patch I used likely(!!sock->ops->xxx) for this case. This is the reason why I send this patch.

 
But you end up adding more code:
 
1 file changed, 58 insertions(+), 13 deletions(-) 

Yes，This would add more code, but this is at the cost of reducing other codes. At the same time, the code will only run  likely(!sock->ops->xxx) in most cases.  Don’t you think that this kind of meaningless thing shouldn’t be done by socket?
 
I don't see this as a gain from any perspective.
 
Thanks.