Re: [PATCH net] net: check and errout if res->fi is NULL when RTM_F_FIB_MATCH is set

[Roopa Prabhu <roopa@cumulusnetworks.com>]

On Tue, Aug 15, 2017 at 7:56 PM, David Ahern <dsa@cumulusnetworks.com> wrote:
> On 8/15/17 8:50 PM, Roopa Prabhu wrote:
>> diff --git a/net/ipv4/route.c b/net/ipv4/route.c
>> index 7effa62..49a018f 100644
>> --- a/net/ipv4/route.c
>> +++ b/net/ipv4/route.c
>> @@ -2763,14 +2763,21 @@ static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh,
>>       if (rtm->rtm_flags & RTM_F_LOOKUP_TABLE)
>>               table_id = rt->rt_table_id;
>>
>> -     if (rtm->rtm_flags & RTM_F_FIB_MATCH)
>> +     if (rtm->rtm_flags & RTM_F_FIB_MATCH) {
>> +             if (!res.fi) {
>> +                     err = fib_props[res->type].error;
>> +                     if (!err)
>> +                             err = -EINVAL;
>
> I think -EHOSTUNREACH is a better error than EINVAL. Nothing about the
> user inputs are invalid; rather the lookup is failing, but indirectly.

 I have been going back and forth on this looking at other examples.
I would have preferred ip_route_input_rcu returned the right error
code for this....but i prefer not to touch that given it may break
something else.

EHOSTUNREACH is only returned for RTN_UNREACHABLE routes.

        [RTN_UNREACHABLE] = {

                .error  = -EHOSTUNREACH,

                .scope  = RT_SCOPE_UNIVERSE,

        },

In the example i am using it was failing due to a daddr being zero. so
seemed like -EINVAL would fit.

If EHOSTUNREACH can cover most errors, I am fine with changing it to
-EHOSTUNREACH.