[RFC PATCH bpf-next 10/12] bpf: support "__uninit" iterator arguments in generic kfuncs

[Justin Suess]

On Tue, Oct 06, 2026 at 12:37:16AM +0000, sashiko-bot at kernel.org wrote:
> Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
> - [Medium] bpf: aliased __uninit and __iter arguments bypass verifier initialization checks
> --
> 
> commit e08b0892f34d46a76c048cb46111b4956bcddfe8
> Author: Justin Suess <utilityemal77 at gmail.com>
> 
> bpf: support "__uninit" iterator arguments in generic kfuncs
> 
> This commit allows generic kfuncs to initialize iterator arguments by
> recognizing the __uninit suffix. It updates process_iter_arg() to treat
> these marked arguments similarly to how bpf_iter_<type>_new() operates on
> uninitialized iterator state.
> 
> > diff --git a/kernel/bpf/verifier.c b/kernel/bpf/verifier.c
> > index 066c4b838b858..3294b2a431674 100644
> > --- a/kernel/bpf/verifier.c
> > +++ b/kernel/bpf/verifier.c
> 
> [ ... ]
> 
> > @@ -8363,8 +8386,16 @@ static int process_iter_arg(struct bpf_verifier_env *env, struct bpf_reg_state *
> >  	t = btf_type_by_id(meta->btf, btf_id);
> >  	nr_slots = t->size / BPF_REG_SIZE;
> >  
> > -	if (is_iter_new_kfunc(meta)) {
> > -		/* bpf_iter_<type>_new() expects pointer to uninit iter state */
> > +	/*
> > +	 * Whether this argument is the iterator the call initializes, rather
> > +	 * than an initialized one it operates on.
> > +	 */
> > +	init = is_iter_new_kfunc(meta) ||
> > +	       is_kfunc_arg_iter_init(meta, arg,
> > +				      &btf_params(meta->func_proto)[arg]);
> > +
> > +	if (init) {
> > +		/* expects a pointer to uninit iter state */
> 
> [Severity: Medium]
> Is there a risk of bypassing verifier initialization checks if a BPF program
> passes aliased registers for these arguments?
> 
> When check_func_args() evaluates kfunc arguments sequentially, an __uninit
> argument will set init to true here, and process_iter_arg() will immediately
> mark the stack slots as initialized via mark_stack_slots_iter().
> 
> If a subsequent __iter argument to the same kfunc aliases those exact same
> stack slots, its is_iter_reg_valid_init() check will incorrectly pass because
> the previous argument just marked them as initialized. This could allow
> uninitialized stack memory to be passed where an initialized iterator is
> expected.
> 
> While this is mitigated in this patch series because the newly introduced
> kfunc uses different BTF types for its iterators (causing the bypass to fail
> strict BTF ID checks), does this structure need to be fixed to prevent future
> issues when kfuncs use matching iterator types?
> 
> Should state modifications like mark_stack_slots_iter() be deferred until
> all arguments are successfully validated, similar to how
> ARG_PTR_TO_UNINIT_MEM is handled?
>
I can't think of a case where someone would try to make an iter transform 
kfunc like this with the same BTF types for the args. So this is a
theoretical nothingburger bug for a theorectical weird kfunc.

Not actioning this for now.

Justin
> >  		if (!is_iter_reg_valid_uninit(env, reg, nr_slots)) {
> >  			verbose(env, "expected uninitialized iter_%s as %s\n",
> >  				iter_type_str(meta->btf, btf_id), reg_arg_name(env, argno));
> 
> -- 
> Sashiko AI review · https://sashiko.dev/#/patchset/20261006002020.2890858-1-utilityemal77@gmail.com?part=10