Re: [PATCH] netfilter: fix CONFIG_NF_REJECT_IPV6=m link error

[Arnd Bergmann]

On Fri, Apr 13, 2018 at 3:15 PM, Pablo Neira Ayuso <pablo@xxxxxxxxxxxxx> wrote:
> On Mon, Apr 09, 2018 at 04:43:40PM +0200, Arnd Bergmann wrote:
>> On Mon, Apr 9, 2018 at 4:37 PM, Pablo Neira Ayuso <pablo@xxxxxxxxxxxxx> wrote:
>> > Hi Arnd,
>> >
>> > On Mon, Apr 09, 2018 at 12:53:12PM +0200, Arnd Bergmann wrote:
>> >> We get a new link error with CONFIG_NFT_REJECT_INET=y and CONFIG_NF_REJECT_IPV6=m
>> >
>> > I think we can update NFT_REJECT_INET so it depends on NFT_REJECT_IPV4
>> > and NFT_REJECT_IPV6. This doesn't allow here CONFIG_NFT_REJECT_INET=y
>> > and CONFIG_NF_REJECT_IPV6=m.
>> >
>> > I mean, just like we do with NFT_FIB_INET.
>>
>> That can only work if NFT_REJECT_INET can be made a 'tristate' symbol
>> again, so that code gets built as a loadable module if
>> CONFIG_NF_REJECT_IPV6=m.
>>
>> > BTW, I think this problem has been is not related to the recent patch,
>> > but something older that kbuild robot has triggered more easily for
>> > some reason?
>>
>> 02c7b25e5f54 is the one that turned NF_TABLES_INET into a 'bool'
>> symbol. NFT_REJECT depends on NF_TABLES_INET, so it used to
>> restricted to a loadable module with IPV6=m, but can now be
>> built-in, which causes that link error.
>
> Still one more spin on this, I would like to see if we have a way to
> fix this by simplifing things a bit.
>
> Would this one I'm attaching would work?

One disadvantage is that it makes the vmlinux bigger since
NF_REJECT_IPV{4,6} can no longer be a module at all now.

I suspect you also stil get a link error with IPV6=m, this time because
the nf_reject_ipv6.o file fails to link against the ipv6 code, e.g.
ipv6_skip_exthdr() and icmpv6_send() appear to be unreachable here.
I haven't tried that though, so I might be missing something.

        Arnd