queue-4.19/ipv6-fix-src-addr-routing-with-the-exception-table.patch

   1 From foo@baz Wed 22 May 2019 08:37:51 AM CEST
   2 From: Wei Wang <weiwan@google.com>
   3 Date: Thu, 16 May 2019 13:30:54 -0700
   4 Subject: ipv6: fix src addr routing with the exception table
   5
   6 From: Wei Wang <weiwan@google.com>
   7
   8 [ Upstream commit 510e2ceda031eed97a7a0f9aad65d271a58b460d ]
   9
  10 When inserting route cache into the exception table, the key is
  11 generated with both src_addr and dest_addr with src addr routing.
  12 However, current logic always assumes the src_addr used to generate the
  13 key is a /128 host address. This is not true in the following scenarios:
  14 1. When the route is a gateway route or does not have next hop.
  15    (rt6_is_gw_or_nonexthop() == false)
  16 2. When calling ip6_rt_cache_alloc(), saddr is passed in as NULL.
  17 This means, when looking for a route cache in the exception table, we
  18 have to do the lookup twice: first time with the passed in /128 host
  19 address, second time with the src_addr stored in fib6_info.
  20
  21 This solves the pmtu discovery issue reported by Mikael Magnusson where
  22 a route cache with a lower mtu info is created for a gateway route with
  23 src addr. However, the lookup code is not able to find this route cache.
  24
  25 Fixes: 2b760fcf5cfb ("ipv6: hook up exception table to store dst cache")
  26 Reported-by: Mikael Magnusson <mikael.kernel@lists.m7n.se>
  27 Bisected-by: David Ahern <dsahern@gmail.com>
  28 Signed-off-by: Wei Wang <weiwan@google.com>
  29 Cc: Martin Lau <kafai@fb.com>
  30 Cc: Eric Dumazet <edumazet@google.com>
  31 Acked-by: Martin KaFai Lau <kafai@fb.com>
  32 Signed-off-by: David S. Miller <davem@davemloft.net>
  33 Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
  34 ---
  35  net/ipv6/route.c |   51 +++++++++++++++++++++++++++------------------------
  36  1 file changed, 27 insertions(+), 24 deletions(-)
  37
  38 --- a/net/ipv6/route.c
  39 +++ b/net/ipv6/route.c
  40 @@ -110,8 +110,8 @@ static int rt6_fill_node(struct net *net
  41                          int iif, int type, u32 portid, u32 seq,
  42                          unsigned int flags);
  43  static struct rt6_info *rt6_find_cached_rt(struct fib6_info *rt,
  44 -                                          struct in6_addr *daddr,
  45 -                                          struct in6_addr *saddr);
  46 +                                          const struct in6_addr *daddr,
  47 +                                          const struct in6_addr *saddr);
  48
  49  #ifdef CONFIG_IPV6_ROUTE_INFO
  50  static struct fib6_info *rt6_add_route_info(struct net *net,
  51 @@ -1542,31 +1542,44 @@ out:
  52   * Caller has to hold rcu_read_lock()
  53   */
  54  static struct rt6_info *rt6_find_cached_rt(struct fib6_info *rt,
  55 -                                          struct in6_addr *daddr,
  56 -                                          struct in6_addr *saddr)
  57 +                                          const struct in6_addr *daddr,
  58 +                                          const struct in6_addr *saddr)
  59  {
  60 +       const struct in6_addr *src_key = NULL;
  61         struct rt6_exception_bucket *bucket;
  62 -       struct in6_addr *src_key = NULL;
  63         struct rt6_exception *rt6_ex;
  64         struct rt6_info *res = NULL;
  65
  66 -       bucket = rcu_dereference(rt->rt6i_exception_bucket);
  67 -
  68  #ifdef CONFIG_IPV6_SUBTREES
  69         /* rt6i_src.plen != 0 indicates rt is in subtree
  70          * and exception table is indexed by a hash of
  71          * both rt6i_dst and rt6i_src.
  72 -        * Otherwise, the exception table is indexed by
  73 -        * a hash of only rt6i_dst.
  74 +        * However, the src addr used to create the hash
  75 +        * might not be exactly the passed in saddr which
  76 +        * is a /128 addr from the flow.
  77 +        * So we need to use f6i->fib6_src to redo lookup
  78 +        * if the passed in saddr does not find anything.
  79 +        * (See the logic in ip6_rt_cache_alloc() on how
  80 +        * rt->rt6i_src is updated.)
  81          */
  82         if (rt->fib6_src.plen)
  83                 src_key = saddr;
  84 +find_ex:
  85  #endif
  86 +       bucket = rcu_dereference(rt->rt6i_exception_bucket);
  87         rt6_ex = __rt6_find_exception_rcu(&bucket, daddr, src_key);
  88
  89         if (rt6_ex && !rt6_check_expired(rt6_ex->rt6i))
  90                 res = rt6_ex->rt6i;
  91
  92 +#ifdef CONFIG_IPV6_SUBTREES
  93 +       /* Use fib6_src as src_key and redo lookup */
  94 +       if (!res && src_key && src_key != &rt->fib6_src.addr) {
  95 +               src_key = &rt->fib6_src.addr;
  96 +               goto find_ex;
  97 +       }
  98 +#endif
  99 +
 100         return res;
 101  }
 102
 103 @@ -2650,10 +2663,8 @@ out:
 104  u32 ip6_mtu_from_fib6(struct fib6_info *f6i, struct in6_addr *daddr,
 105                       struct in6_addr *saddr)
 106  {
 107 -       struct rt6_exception_bucket *bucket;
 108 -       struct rt6_exception *rt6_ex;
 109 -       struct in6_addr *src_key;
 110         struct inet6_dev *idev;
 111 +       struct rt6_info *rt;
 112         u32 mtu = 0;
 113
 114         if (unlikely(fib6_metric_locked(f6i, RTAX_MTU))) {
 115 @@ -2662,18 +2673,10 @@ u32 ip6_mtu_from_fib6(struct fib6_info *
 116                         goto out;
 117         }
 118
 119 -       src_key = NULL;
 120 -#ifdef CONFIG_IPV6_SUBTREES
 121 -       if (f6i->fib6_src.plen)
 122 -               src_key = saddr;
 123 -#endif
 124 -
 125 -       bucket = rcu_dereference(f6i->rt6i_exception_bucket);
 126 -       rt6_ex = __rt6_find_exception_rcu(&bucket, daddr, src_key);
 127 -       if (rt6_ex && !rt6_check_expired(rt6_ex->rt6i))
 128 -               mtu = dst_metric_raw(&rt6_ex->rt6i->dst, RTAX_MTU);
 129 -
 130 -       if (likely(!mtu)) {
 131 +       rt = rt6_find_cached_rt(f6i, daddr, saddr);
 132 +       if (unlikely(rt)) {
 133 +               mtu = dst_metric_raw(&rt->dst, RTAX_MTU);
 134 +       } else {
 135                 struct net_device *dev = fib6_info_nh_dev(f6i);
 136
 137                 mtu = IPV6_MIN_MTU;