Re: [PATCH bpf-next 2/2] bpf: extend bind v4/v6 selftests for mark/prio/bindtoifindex

[Daniel Borkmann <daniel@iogearbox.net>]

On 1/11/21 9:15 PM, Yonghong Song wrote:
> On 1/11/21 8:17 AM, Daniel Borkmann wrote:
>> Extend existing cgroup bind4/bind6 tests to add coverage for setting and
>> retrieving SO_MARK, SO_PRIORITY and SO_BINDTOIFINDEX at the bind hook.
>>
>> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
> 
> Ack with a minor comments below.
> 
> Acked-by: Yonghong Song <yhs@fb.com>
> 
>> ---
>>   .../testing/selftests/bpf/progs/bind4_prog.c  | 41 +++++++++++++++++--
>>   .../testing/selftests/bpf/progs/bind6_prog.c  | 41 +++++++++++++++++--
>>   2 files changed, 74 insertions(+), 8 deletions(-)
>>
>> diff --git a/tools/testing/selftests/bpf/progs/bind4_prog.c b/tools/testing/selftests/bpf/progs/bind4_prog.c
>> index c6520f21f5f5..4479ac27b1d3 100644
>> --- a/tools/testing/selftests/bpf/progs/bind4_prog.c
>> +++ b/tools/testing/selftests/bpf/progs/bind4_prog.c
>> @@ -29,18 +29,47 @@ static __inline int bind_to_device(struct bpf_sock_addr *ctx)
>>       char veth2[IFNAMSIZ] = "test_sock_addr2";
>>       char missing[IFNAMSIZ] = "nonexistent_dev";
>>       char del_bind[IFNAMSIZ] = "";
>> +    int veth1_idx, veth2_idx;
>>       if (bpf_setsockopt(ctx, SOL_SOCKET, SO_BINDTODEVICE,
>> -                &veth1, sizeof(veth1)))
>> +               &veth1, sizeof(veth1)))
>> +        return 1;
>> +    if (bpf_getsockopt(ctx, SOL_SOCKET, SO_BINDTOIFINDEX,
>> +               &veth1_idx, sizeof(veth1_idx)) || !veth1_idx)
>>           return 1;
>>       if (bpf_setsockopt(ctx, SOL_SOCKET, SO_BINDTODEVICE,
>> -                &veth2, sizeof(veth2)))
>> +               &veth2, sizeof(veth2)))
>> +        return 1;
>> +    if (bpf_getsockopt(ctx, SOL_SOCKET, SO_BINDTOIFINDEX,
>> +               &veth2_idx, sizeof(veth2_idx)) || !veth2_idx ||
>> +        veth1_idx == veth2_idx)
>>           return 1;
>>       if (bpf_setsockopt(ctx, SOL_SOCKET, SO_BINDTODEVICE,
>> -                &missing, sizeof(missing)) != -ENODEV)
>> +               &missing, sizeof(missing)) != -ENODEV)
>> +        return 1;
>> +    if (bpf_setsockopt(ctx, SOL_SOCKET, SO_BINDTOIFINDEX,
>> +               &veth1_idx, sizeof(veth1_idx)))
>>           return 1;
>>       if (bpf_setsockopt(ctx, SOL_SOCKET, SO_BINDTODEVICE,
>> -                &del_bind, sizeof(del_bind)))
>> +               &del_bind, sizeof(del_bind)))
>> +        return 1;
>> +
>> +    return 0;
>> +}
>> +
>> +static __inline int misc_opts(struct bpf_sock_addr *ctx, int opt)
>> +{
>> +    int old, tmp, new = 0xeb9f;
>> +
>> +    if (bpf_getsockopt(ctx, SOL_SOCKET, opt, &old, sizeof(old)) ||
>> +        old == new)
>> +        return 1;
> 
> Here, we assume old never equals to new. it would be good to add
> a comment to explicitly state this is true. Maybe in the future
> somebody will try to add more misc_opts which might have conflict
> here.

I thought it's obvious, but yes I can add a comment.

> Alternatively, you could pass in "new" values
> from user space with global variables for each option,
> but that may be an overkill.

Agree, that's overkill.

Thanks,
Daniel