mbox series

[nf,0/2] netfilter: nft_fib: ignore icmpv6 packets from ::

Message ID 20210608114818.23397-1-fw@strlen.de
Headers show
Series netfilter: nft_fib: ignore icmpv6 packets from :: | expand

Message

Florian Westphal June 8, 2021, 11:48 a.m. UTC
Quoting nf bugzilla:
--------
Using the following for
reverse path filtering breaks IPv6 duplicate address detection:

table inet ip46_firewall {
    chain ip46_rpfilter {
        type filter hook prerouting priority raw;
        fib saddr . iif oif missing log prefix "RPFILTER: " drop
    }
}

This is because packets from :: to ff02::1:ff00/104 will be dropped and thus
other hosts on the network cannot detect that this host already has the same
address assigned. The problem can be worked around in nft rules by handling
such packets specially but I guess it should work as is.

In the kernel in ip6t_rpfilter.c the function rpfilter_mt() checks for
saddrtype == IPV6_ADDR_ANY. nft_fib_ipv6.c doesn't seem to have an equivalent
check for this special case.
--------

First patch adds a test case for this, second patch makes icmpv6 from
any to link-local bypass the fib lookup, just like loopback packets.

Florian Westphal (2):
  selftests: netfilter: add fib test case
  netfilter: ipv6: skip ipv6 packets from any to link-local

 net/ipv6/netfilter/nft_fib_ipv6.c            |  22 +-
 tools/testing/selftests/netfilter/Makefile   |   2 +-
 tools/testing/selftests/netfilter/nft_fib.sh | 221 +++++++++++++++++++
 3 files changed, 240 insertions(+), 5 deletions(-)
 create mode 100755 tools/testing/selftests/netfilter/nft_fib.sh

Comments

Pablo Neira Ayuso June 9, 2021, 7:08 p.m. UTC | #1
On Tue, Jun 08, 2021 at 01:48:16PM +0200, Florian Westphal wrote:
> Quoting nf bugzilla:
> --------
> Using the following for
> reverse path filtering breaks IPv6 duplicate address detection:
> 
> table inet ip46_firewall {
>     chain ip46_rpfilter {
>         type filter hook prerouting priority raw;
>         fib saddr . iif oif missing log prefix "RPFILTER: " drop
>     }
> }
> 
> This is because packets from :: to ff02::1:ff00/104 will be dropped and thus
> other hosts on the network cannot detect that this host already has the same
> address assigned. The problem can be worked around in nft rules by handling
> such packets specially but I guess it should work as is.
> 
> In the kernel in ip6t_rpfilter.c the function rpfilter_mt() checks for
> saddrtype == IPV6_ADDR_ANY. nft_fib_ipv6.c doesn't seem to have an equivalent
> check for this special case.
> --------
> 
> First patch adds a test case for this, second patch makes icmpv6 from
> any to link-local bypass the fib lookup, just like loopback packets.

Applied, thanks.