Re: [PATCH net-next v3 4/5] ip_tunnel: constify input argument of ip_tunnel_info_opts( )

From: Alexander Lobakin
Date: Fri Feb 24 2023 - 11:12:21 EST


From: Gavin Li <gavinl@xxxxxxxxxx>
Date: Mon, 20 Feb 2023 18:42:14 +0800

>
> On 2/20/2023 4:46 AM, Simon Horman wrote:
>> External email: Use caution opening links or attachments
>>
>>
>> On Sun, Feb 19, 2023 at 09:29:21PM +0100, Simon Horman wrote:
>>> On Fri, Feb 17, 2023 at 05:39:24AM +0200, Gavin Li wrote:
>>>> Constify input argument(i.e. struct ip_tunnel_info *info) of
>>>> ip_tunnel_info_opts( ) so that it wouldn't be needed to W/A it each
>>>> time
>>>> in each driver.
>>>>
>>>> Signed-off-by: Gavin Li <gavinl@xxxxxxxxxx>
>>>> ---
>>>>   include/net/ip_tunnels.h | 4 ++--
>>>>   1 file changed, 2 insertions(+), 2 deletions(-)
>>>>
>>>> diff --git a/include/net/ip_tunnels.h b/include/net/ip_tunnels.h
>>>> index fca357679816..32c77f149c6e 100644
>>>> --- a/include/net/ip_tunnels.h
>>>> +++ b/include/net/ip_tunnels.h
>>>> @@ -485,9 +485,9 @@ static inline void iptunnel_xmit_stats(struct
>>>> net_device *dev, int pkt_len)
>>>>      }
>>>>   }
>>>>
>>>> -static inline void *ip_tunnel_info_opts(struct ip_tunnel_info *info)
>>>> +static inline void *ip_tunnel_info_opts(const struct ip_tunnel_info
>>>> *info)
>>>>   {
>>>> -   return info + 1;
>>>> +   return (void *)(info + 1);
>>> I'm unclear on what problem this is trying to solve,
>>> but info being const, and then returning (info +1)
>>> as non-const feels like it is masking rather than fixing a problem.
>> I now see that an example of the problem is added by path 5/5.
>>
>> ...
>>    CC [M]  drivers/net/ethernet/mellanox/mlx5/core/en/tc_tun_vxlan.o
>> drivers/net/ethernet/mellanox/mlx5/core/en/tc_tun_vxlan.c: In function
>> 'mlx5e_gen_ip_tunnel_header_vxlan':
>> drivers/net/ethernet/mellanox/mlx5/core/en/tc_tun_vxlan.c:103:43:
>> error: passing argument 1 of 'ip_tunnel_info_opts' discards 'const'
>> qualifier from pointer target type [-Werror=discarded-qualifiers]
>>    103 |                 md = ip_tunnel_info_opts(e->tun_info);
>>        |                                          ~^~~~~~~~~~
>> In file included from
>> drivers/net/ethernet/mellanox/mlx5/core/en/tc_tun_vxlan.c:4:
>> ./include/net/ip_tunnels.h:488:64: note: expected 'struct
>> ip_tunnel_info *' but argument is of type 'const struct ip_tunnel_info *'
>>    488 | static inline void *ip_tunnel_info_opts(struct ip_tunnel_info
>> *info)
>>        |                                        
>> ~~~~~~~~~~~~~~~~~~~~~~~^~~~
>> ...
>>
>> But I really do wonder if this patch masks rather than fixes the problem.
> Hi Olek, any comment?

Hi,

Sorry for the late reply, was busy at work ._.

I initially proposed a solution via _Generic or __builtin_choose_expr to
return const or non-const opts basing on the input pointer type. I don't
like simple cast-aways.

See container_of_const() how it dynamically chooses whether to add
`const` or not when returning the result.

>>
>>>>   }
>>>>
>>>>   static inline void ip_tunnel_info_opts_get(void *to,
>>>> --
>>>> 2.31.1
>>>>

Thanks,
Olek