On Sat, 30 Apr 2022 10:22:33 -0700 Cong Wang wrote:

quoted

I started prototyping a similar patch for TLS a while back but I have
two functions - one to get the skb and another to consume it. I thought
that's better for TLS, otherwise skbs stuck in the middle layer are not
counted towards the rbuf. Any thoughts on structuring the API that way?
I guess we can refactor that later, since TLS TCP-only we don't need
proto_ops plumbing there.  

Do you have a pointer to the source code? I am not sure how TLS uses
->read_sock() (or which interface is relevant).  

Nothing useful, I started hacking on removing strparser but then got
distracted with functional optimizations. TLS calls ->read_sock() thru
strparser.

With a little bit of code duplication TLS should be able to avoid 
the strparser's heavy machinery and cloning each skb.