Re: [PATCH net-next v2] seg6: fool-proof the processing of SRv6 behavior attributes

[David Ahern]

On 2/6/21 10:09 AM, Andrea Mayer wrote:
> The set of required attributes for a given SRv6 behavior is identified
> using a bitmap stored in an unsigned long, since the initial design of SRv6
> networking in Linux. Recently the same approach has been used for
> identifying the optional attributes.
> 
> However, the number of attributes supported by SRv6 behaviors depends on
> the size of the unsigned long type which changes with the architecture.
> Indeed, on a 64-bit architecture, an SRv6 behavior can support up to 64
> attributes while on a 32-bit architecture it can support at most 32
> attributes.
> 
> To fool-proof the processing of SRv6 behaviors we verify, at compile time,
> that the set of all supported SRv6 attributes can be encoded into a bitmap
> stored in an unsigned long. Otherwise, kernel build fails forcing
> developers to reconsider adding a new attribute or extend the total
> number of supported attributes by the SRv6 behaviors.
> 
> Moreover, we replace all patterns (1 << i) with the macro SEG6_F_ATTR(i) in
> order to address potential overflow issues caused by 32-bit signed
> arithmetic.
> 
> Thanks to Colin Ian King for catching the overflow problem, providing a
> solution and inspiring this patch.
> Thanks to Jakub Kicinski for his useful suggestions during the design of
> this patch.
> 
> v2:
>  - remove the SEG6_LOCAL_MAX_SUPP which is not strictly needed: it can
>    be derived from the unsigned long type. Thanks to David Ahern for
>    pointing it out.
> 
> Signed-off-by: Andrea Mayer <andrea.mayer@xxxxxxxxxxx>
> ---
>  net/ipv6/seg6_local.c | 67 +++++++++++++++++++++++++------------------
>  1 file changed, 39 insertions(+), 28 deletions(-)
> 

Reviewed-by: David Ahern <dsahern@xxxxxxxxxx>