Re: [PATCH bpf-next v6 13/17] bpf: Reject stack arguments if tail call reachable

On 4/19/26 10:08 AM, [email protected] wrote:
> Sashiko AI review found 1 potential issue(s):
> - [High] The verifier incorrectly rejects perfectly valid subprograms with stack arguments due to an order-dependent state leak of `tail_call_reachable` during the call graph traversal.
> --
>
> commit 2677e39a30cc9fcf44cadb1927d1362ad39f7b4e
> Author: Yonghong Song <[email protected]>
>
> bpf: Reject stack arguments if tail call reachable
>
> This commit deprecates tailcalls in the presence of stack arguments,
> rejecting subprograms with stack arguments if a tail call is reachable.
>
>> diff --git a/kernel/bpf/verifier.c b/kernel/bpf/verifier.c
>> index 13f1fd7880921..af89750498836 100644
>> --- a/kernel/bpf/verifier.c
>> +++ b/kernel/bpf/verifier.c
> [ ... ]
>
>> @@ -5672,14 +5677,23 @@ static int check_max_stack_depth_subprog(struct bpf_verifier_env *env, int idx,
>>   	 * this info will be utilized by JIT so that we will be preserving the
>>   	 * tail call counter throughout bpf2bpf calls combined with tailcalls
>>   	 */
>> -	if (tail_call_reachable)
>> +	if (tail_call_reachable) {
>>   		for (tmp = idx; tmp >= 0; tmp = dinfo[tmp].caller) {
>>   			if (subprog[tmp].is_exception_cb) {
>>   				verbose(env, "cannot tail call within exception cb\n");
>>   				return -EINVAL;
>>   			}
>> +			if (subprog_has_stack_args(&subprog[tmp])) {
>> +				verbose(env, "tail_calls are not allowed in programs with stack args\n");
>> +				return -EINVAL;
>> +			}
> Can this leak the tail_call_reachable state across disjoint call graph branches?

I guess this is a known issue such that tail_call_reachable is marked more
than necessar. Since this has been mentioned a couple of times. I will fix
it in the next revision.

>
> In check_max_stack_depth_subprog(), tail_call_reachable is a local variable
> that seems to act as an accumulator. If the main program first calls subprogram A
> (which has a tail call) and subsequently calls subprogram B (which has stack
> arguments but no tail call), tail_call_reachable remains true while traversing B.
>
> When the traversal finishes subprogram B, this newly added check evaluates the
> current call stack (main -> B). Since tail_call_reachable leaked from the main
> -> A traversal, and subprogram B has stack arguments, the check would falsely
> reject the program.
>
> Conversely, if the bytecode is ordered so that B is called before A,
> tail_call_reachable would be false while traversing B, allowing the program to
> load successfully.
>
> Does this introduce an unpredictable, order-dependent functional regression?
>
>>   			subprog[tmp].tail_call_reachable = true;
>>   		}
>> +	} else if (!idx && subprog[0].has_tail_call && subprog_has_stack_args(&subprog[0])) {
>> +		verbose(env, "tail_calls are not allowed in programs with stack args\n");
>> +		return -EINVAL;
>> +	}
>> +
>>   	if (subprog[0].tail_call_reachable)
>>   		env->prog->aux->tail_call_reachable = true;
>>