Re: [PATCH bpf-next v2 1/9] bpf: Make KF_TRUSTED_ARGS the default for all kfuncs

[Eduard Zingerman <eddyz87@gmail.com>]

On Wed, 2025-12-31 at 19:00 +0000, Puranjay Mohan wrote:

[...]

> There is another aspect I want your opinion one:
> 
> Assume a kfunc returns an error when you pass in a NULL pointer for
> some parameter, it checks for NULL as if it is not valid usage, it
> returns an error. After this change, this kfunc will not return an
> error at runtime, rather will be rejected by the verifier itself. This
> should not be a problem for real programs right?

Yes, makes sense.

> I think we should drop the second patch: "bpf: net: netfilter: Mark
> kfuncs accurately" because these kfuncs have no real use case with
> NULL being passed to them, only a self test tries to call them with
> NULL parameters, I think we should change the self test to detect
> load failure and leave these kfuncs without __nullable
> annotation. What do you think?

Actually, I was going to ack that patch :)
But you are right, each of those kfuncs is a wrapper and functions
they wrap, like __bpf_nf_ct_alloc_entry, require 'opts' not to be null
or return an error.