[PATCH libnl 1/4] route: Add support for netconf

[David Ahern]

On 5/3/17 7:04 AM, Thomas Haller wrote:
> Hi David,
> 
> On Thu, 2017-04-27 at 13:42 -0700, David Ahern wrote:
>> Add route/netconf cache type to monitor RTM_*NETCONF messages.
>>
>> Signed-off-by: David Ahern <dsa at cumulusnetworks.com>
>> Signed-off-by: Balakrishnan Raman <ramanb at cumulusnetworks.com>
>> ---
> 
>> +int rtnl_netconf_get_family(struct rtnl_netconf *nc);
>> +int rtnl_netconf_get_ifindex(struct rtnl_netconf *nc);
>> +int rtnl_netconf_get_forwarding(struct rtnl_netconf *nc);
>> +int rtnl_netconf_get_mc_forwarding(struct rtnl_netconf *nc);
>> +int rtnl_netconf_get_rp_filter(struct rtnl_netconf *nc);
>> +int rtnl_netconf_get_proxy_neigh(struct rtnl_netconf *nc);
>> +int rtnl_netconf_get_ignore_routes_linkdown(struct rtnl_netconf
>> *nc);
> 
> the underlying fields for these properties are "int" (in the netlink
> message they are s32). How about having the getters to have a distinct
> return and error value? For example:
> 
>   int rtnl_netconf_get_family(struct rtnl_netconf *nc,
>                               int *out_value)
>   {
>       if (!nc)
>           return -NLE_INVAL;
>       if (!(nc->mask & NETCONF_ATTR_FAMILY))
>           return -NLE_MISSING_ATTR;
>       if (out_value)
>           *out_value = nc->family;
>       return 0;
>   }
> 
> 
> What's your opinion about that?

family and ifindex are required attributes. The others should only have
positive settings (0/1 boolean or 0 to N integer), though admittedly the
kernel docs are not clear especially for the ignore_routes.

For consistency -- and to handle potentially new ones with < 0 valid
values -- I can update the API to what you suggest.