Re: [RFC PATCH bpf-next] bpf: Add a kfunc filter function to 'struct btf_kfunc_id_set'.

[Martin KaFai Lau <martin.lau@linux.dev>]

On 4/5/23 8:05 AM, Aditi Ghag wrote:
> Looks quite promising for the sock_destroy use case, and also as a generic filtering mechanism, but I'm not aware of other use cases. I haven't had a chance to apply this patch locally, but I'm planning to do it soon. Thanks!

Please don't top post.

Other use case is to allow different sets of kfuncs to struct_ops programs from 
David: https://lore.kernel.org/bpf/Y9KLHZ1TNXVHdVKm@maniforge/

>> From: Martin KaFai Lau <martin.lau@kernel.org>
>>
>> This set (https://lore.kernel.org/bpf/https://lore.kernel.org/bpf/500d452b-f9d5-d01f-d365-2949c4fd37ab@linux.dev/)
>> needs to limit bpf_sock_destroy kfunc to BPF_TRACE_ITER.
>> In the earlier reply, I thought of adding a BTF_KFUNC_HOOK_TRACING_ITER.
>>
>> Instead of adding BTF_KFUNC_HOOK_TRACING_ITER, I quickly hacked something
>> that added a callback filter to 'struct btf_kfunc_id_set'. The filter has
>> access to the prog such that it can filter by other properties of a prog.
>> The prog->expected_attached_type is used in the tracing_iter_filter().
>> It is mostly compiler tested only, so it is still very rough but should
>> be good enough to show the idea.
>>
>> would like to hear how others think. It is pretty much the only
>> piece left for the above mentioned set.