target/linux/generic/backport-4.14/366-netfilter-nft_flow_offload-Fix-reverse-route-lookup.patch

   1 From: wenxu <wenxu@ucloud.cn>
   2 Date: Wed, 9 Jan 2019 10:40:11 +0800
   3 Subject: [PATCH] netfilter: nft_flow_offload: Fix reverse route lookup
   4
   5 Using the following example:
   6
   7         client 1.1.1.7 ---> 2.2.2.7 which dnat to 10.0.0.7 server
   8
   9 The first reply packet (ie. syn+ack) uses an incorrect destination
  10 address for the reverse route lookup since it uses:
  11
  12         daddr = ct->tuplehash[!dir].tuple.dst.u3.ip;
  13
  14 which is 2.2.2.7 in the scenario that is described above, while this
  15 should be:
  16
  17         daddr = ct->tuplehash[dir].tuple.src.u3.ip;
  18
  19 that is 10.0.0.7.
  20
  21 Signed-off-by: wenxu <wenxu@ucloud.cn>
  22 Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
  23 ---
  24
  25 --- a/net/netfilter/nft_flow_offload.c
  26 +++ b/net/netfilter/nft_flow_offload.c
  27 @@ -29,10 +29,10 @@ static int nft_flow_route(const struct n
  28         memset(&fl, 0, sizeof(fl));
  29         switch (nft_pf(pkt)) {
  30         case NFPROTO_IPV4:
  31 -               fl.u.ip4.daddr = ct->tuplehash[!dir].tuple.dst.u3.ip;
  32 +               fl.u.ip4.daddr = ct->tuplehash[dir].tuple.src.u3.ip;
  33                 break;
  34         case NFPROTO_IPV6:
  35 -               fl.u.ip6.daddr = ct->tuplehash[!dir].tuple.dst.u3.in6;
  36 +               fl.u.ip6.daddr = ct->tuplehash[dir].tuple.src.u3.in6;
  37                 break;
  38         }
  39