From: David Ahern <dsah...@gmail.com>
Date: Mon, 7 May 2018 08:26:47 -0600
> On 5/7/18 8:10 AM, Daniel Borkmann wrote:
>> On 05/07/2018 03:35 PM, Jesper Dangaard Brouer wrote:
>>> On Thu, 3 May 2018 19:54:31 -0700 David Ahern <dsah...@gmail.com> wrote:
>>>
>>>> diff --git a/net/core/filter.c b/net/core/filter.c
>>>> index 6877426c23a6..cf0d27acf1d1 100644
>>>> --- a/net/core/filter.c
>>>> +++ b/net/core/filter.c
>>> [...]
>>>> +static const struct bpf_func_proto bpf_xdp_fib_lookup_proto = {
>>>> + .func = bpf_xdp_fib_lookup,
>>>> + .gpl_only = true,
>>>
>>> Is it a deliberate choice to require BPF-progs using this helper to be
>>> GPL licensed?
>>>
>>> Asking as this seems to be the first network related helper with this
>>> requirement, while this is typical for tracing related helpers.
>>
>> Good point, we should remove that. In networking it's only the perf event
>> output helpers tying into tracing bits. After all, if you do a route lookup
>> via netlink from user space there's no such restriction at all.
>>
>
> Networking symbols are typically exported GPL for modules. The person
> writing the code and exporting GPL is specifying a desire that only GPL
> licensed modules can link to the symbol.
>
> Given the common analogy of modules and bpf programs, why can't a writer
> of a bpf helper specify a preference that only GPL licensed programs
> leverage a BPF helper?
I also think that for this particular set of helpers GPL is appropriate.
Yes, via netlink the same lookup can happen, but not with the same level
of performance and microsecond tuning we've done over the years on this
sophisticated trie lookup code.
Therefore, I think David's choice is very appropriate.
