Re: [PATCH bpf-next v1 04/19] bpf: allow loading of a bpf_iter program

[Yonghong Song <yhs@fb.com>]

On 4/28/20 5:54 PM, Martin KaFai Lau wrote:
> On Mon, Apr 27, 2020 at 01:12:39PM -0700, Yonghong Song wrote:
>> A bpf_iter program is a tracing program with attach type
>> BPF_TRACE_ITER. The load attribute
>>    attach_btf_id
>> is used by the verifier against a particular kernel function,
>> e.g., __bpf_iter__bpf_map in our previous bpf_map iterator.
>>
>> The program return value must be 0 for now. In the
>> future, other return values may be used for filtering or
>> teminating the iterator.
>>
>> Signed-off-by: Yonghong Song <yhs@fb.com>
>> ---
>>   include/uapi/linux/bpf.h       |  1 +
>>   kernel/bpf/verifier.c          | 20 ++++++++++++++++++++
>>   tools/include/uapi/linux/bpf.h |  1 +
>>   3 files changed, 22 insertions(+)
>>
>> diff --git a/include/uapi/linux/bpf.h b/include/uapi/linux/bpf.h
>> index 4a6c47f3febe..f39b9fec37ab 100644
>> --- a/include/uapi/linux/bpf.h
>> +++ b/include/uapi/linux/bpf.h
>> @@ -215,6 +215,7 @@ enum bpf_attach_type {
>>   	BPF_TRACE_FEXIT,
>>   	BPF_MODIFY_RETURN,
>>   	BPF_LSM_MAC,
>> +	BPF_TRACE_ITER,
>>   	__MAX_BPF_ATTACH_TYPE
>>   };
>>   
>> diff --git a/kernel/bpf/verifier.c b/kernel/bpf/verifier.c
>> index 91728e0f27eb..fd36c22685d9 100644
>> --- a/kernel/bpf/verifier.c
>> +++ b/kernel/bpf/verifier.c
>> @@ -7074,6 +7074,11 @@ static int check_return_code(struct bpf_verifier_env *env)
>>   			return 0;
>>   		range = tnum_const(0);
>>   		break;
>> +	case BPF_PROG_TYPE_TRACING:
>> +		if (env->prog->expected_attach_type != BPF_TRACE_ITER)
>> +			return 0;
>> +		range = tnum_const(0);
>> +		break;
>>   	default:
>>   		return 0;
>>   	}
>> @@ -10454,6 +10459,7 @@ static int check_attach_btf_id(struct bpf_verifier_env *env)
>>   	struct bpf_prog *tgt_prog = prog->aux->linked_prog;
>>   	u32 btf_id = prog->aux->attach_btf_id;
>>   	const char prefix[] = "btf_trace_";
>> +	struct btf_func_model fmodel;
>>   	int ret = 0, subprog = -1, i;
>>   	struct bpf_trampoline *tr;
>>   	const struct btf_type *t;
>> @@ -10595,6 +10601,20 @@ static int check_attach_btf_id(struct bpf_verifier_env *env)
>>   		prog->aux->attach_func_proto = t;
>>   		prog->aux->attach_btf_trace = true;
>>   		return 0;
>> +	case BPF_TRACE_ITER:
>> +		if (!btf_type_is_func(t)) {
>> +			verbose(env, "attach_btf_id %u is not a function\n",
>> +				btf_id);
>> +			return -EINVAL;
>> +		}
>> +		t = btf_type_by_id(btf, t->type);
>> +		if (!btf_type_is_func_proto(t))
> Other than the type tests,
> to ensure the attach_btf_id is a supported bpf_iter target,
> should the prog be checked against the target list
> ("struct list_head targets") here also during the prog load time?

This is a good question. In my RFC v2, I did this, checking against
registered targets (essentially, program loading + attaching to the target).

In this version, program loading and attaching are separated.
   - program loading: against btf_id
   - attaching: linking bpf program to target
     current linking parameter only bpf_program, but later on
     there may be additional parameters like map_id, pid, cgroup_id
     etc. for tailoring the iterator behavior.

This seems having a better separation. Agreed that checking
at load time may return error earlier instead at link_create
time. Let me think about this.


> 
>> +			return -EINVAL;
>> +		prog->aux->attach_func_name = tname;
>> +		prog->aux->attach_func_proto = t;
>> +		ret = btf_distill_func_proto(&env->log, btf, t,
>> +					     tname, &fmodel);
>> +		return ret;
>>   	default:
>>   		if (!prog_extension)
>>   			return -EINVAL;