Re: [PATCH bpf-next 1/7] bpf: Make KF_TRUSTED_ARGS the default for all kfuncs

[Eduard Zingerman <eddyz87@gmail.com>]

On Wed, 2025-12-31 at 00:08 +0000, Puranjay Mohan wrote:

[...]

> I wish to do a full review of all kfuncs and make
> sure either they are tagged with correct __nullable, __opt annotation
> or fixed to make sure they are doing the right thing. But currently I
> just made sure all selftests pass, some of the kfuncs might not have
> self tests and would need manual analysis and I will have to do that.

Ack, sounds like a plan.

> Some kfuncs will have breaking changes, I am not sure how to work
> around that case, for example css_rstat_updated() could be
> successfully called from fentry programs like the selftest fixed in
> Patch 7, it worked because css_rstat_updated() doesn't mark the
> parameters with KF_TRUSTED_ARGS, but now KF_TRUSTED_ARGS is the
> default so this kfunc can't be called from fentry programs as fentry's
> parameters are not marked as trusted.
> 
> 
> Looking at the code of css_rstat_updated() it seems that it assumes
> the parameters to be trusted and therefore not allowing it to be
> called from fentry would be the right thing to do,
> but it could break perfectly working BPF programs.

Indeed, it expects 'css' to be not NULL as it dereferences the
pointer immediately.