Re: Android boot failure with 6.12

[Pablo Neira Ayuso <pablo@netfilter.org>]

Hi Jozsef,

On Sat, Jan 11, 2025 at 02:31:18PM +0100, Jozsef Kadlecsik wrote:
> Hi,
> 
> On Fri, 10 Jan 2025, Maciej Żenczykowski wrote:
> 
> > nvm - https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git/commit/net/netfilter/xt_mark.c?id=306ed1728e8438caed30332e1ab46b28c25fe3d8
> 
> Sorry, but I don't understand the patch at all. With it applied now it'd 
> be not possible to load in the "MARK" target with IPv4. The code segment 
> after the patch:
> 
> static struct xt_target mark_tg_reg[] __read_mostly = {
>         {
>                 .name           = "MARK",
>                 .revision       = 2,
>                 .family         = NFPROTO_IPV6,
>                 .target         = mark_tg,
>                 .targetsize     = sizeof(struct xt_mark_tginfo2),
>                 .me             = THIS_MODULE,
>         },
> #if IS_ENABLED(CONFIG_IP_NF_ARPTABLES)
>         {
>                 .name           = "MARK",
>                 .revision       = 2,
>                 .family         = NFPROTO_ARP,
>                 .target         = mark_tg,
>                 .targetsize     = sizeof(struct xt_mark_tginfo2),
>                 .me             = THIS_MODULE,
>         },
> #endif
> #if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
>         {
>                 .name           = "MARK",
>                 .revision       = 2,
>                 .family         = NFPROTO_IPV6,
>                 .target         = mark_tg,
>                 .targetsize     = sizeof(struct xt_mark_tginfo2),
>                 .me             = THIS_MODULE,
>         },
> #endif
> };
> 
> How is it supposed to work for IPv4?
> 
> Why the "IS_ENABLED(CONFIG_IP6_NF_IPTABLES)" part was not enough for the 
> IPv6-specific MARK target to be compiled in? Isn't it an issue about 
> selecting CONFIG_IP6_NF_IPTABLES vs CONFIG_IP6_NF_IPTABLES_LEGACY?

This was fixed by an incremental patch:

  306ed1728e84 ("netfilter: xtables: fix typo causing some targets not to load on IPv6")

so there is no two MARK targets for NFPROTO_IPV6.

> Also, why the "mark" match was not split into NFPROTO_IPV4, NFPROTO_ARP,
> NFPROTO_IPV6 explicitly (and other matches where the target was split)?

The audit to tighten this interface searched for:

- use of xtables verdicts are incompatible with ebtables.
- IP header cannot be assumed to be linear on ebtables.

xt_mark match can be restricted too, ebtables uses ebt_mark. But this
should be safe, so this patch should probably go via nf-next.

Thanks.