Re: [PATCH bpf 1/2] bpf: Let verifier consider {task,cgroup} is trusted in bpf_iter_reg

[Yonghong Song <yonghong.song@linux.dev>]

On 11/6/23 6:44 PM, Chuyi Zhou wrote:
> Hello,
>
> 在 2023/11/7 02:29, Martin KaFai Lau 写道:
>> On 11/5/23 5:34 AM, Chuyi Zhou wrote:
>>> BTF_TYPE_SAFE_TRUSTED(struct bpf_iter__task) in verifier.c wanted to
>>> teach BPF verifier that bpf_iter__task -> task is a trusted ptr. But it
>>> doesn't work well.
>>>
>>> The reason is, bpf_iter__task -> task would go through btf_ctx_access()
>>> which enforces the reg_type of 'task' is ctx_arg_info->reg_type, and in
>>> task_iter.c, we actually explicitly declare that the
>>> ctx_arg_info->reg_type is PTR_TO_BTF_ID_OR_NULL.
>>>
>>> This patch sets ctx_arg_info->reg_type is PTR_TO_BTF_ID_OR_NULL |
>>> PTR_TRUSTED in task_reg_info.
>>>
>>> Similarly, bpf_cgroup_reg_info -> cgroup is also PTR_TRUSTED since 
>>> we are
>>> under the protection of cgroup_mutex and we would check 
>>> cgroup_is_dead()
>>> in __cgroup_iter_seq_show().
>>>
>>
>> Make sense. I think the bpf_tcp_iter made similar change in 
>> tcp_seq_info also. What may be the Fixes tag? Is it fixing the recent 
>> kfunc of the css_task iter?
>>
>
> Thanks for the review.
>
> I think it's not a fix for recent kfunc of css_task iter. We are 
> working at SEC("iter/task") and SEC("iter/cgroup").
>
> I'm not sure whether it's a 'fix' for cgroup_iter/task_iter. If we 
> need fix tags, do we need to split this patch into two separate 
> patches? Or add two fix tags on commit log:

I think the patch itself is not a fix, rather an improvement. The bpf_iter predates kfunc/PTR_TRUSTED stuff. The argument 'task'
or 'cgroup' are already trusted so the bpf_iter program can print out useful data.
But recent kfunc things requires some parameters to be marked as PTR_TRUSTED so that they can be passed to kfunc,
so this patch enables this usage for kfunc in bpf_iter programs.


>
> Fixes: d4ccaf58a84721 ("bpf: Introduce cgroup iter")
> Fixes: 3c32cc1bceba8a17 ("bpf: Enable bpf_iter targets registering ctx 
> argument types")
>
> Thanks.
>
>
>