queue-4.14/xfrm-reuse-uncached_list-to-track-xdsts.patch

   1 From 66a256e4423a44db3bd0433babc7ad6beaa5853d Mon Sep 17 00:00:00 2001
   2 From: Xin Long <lucien.xin@gmail.com>
   3 Date: Wed, 14 Feb 2018 19:06:02 +0800
   4 Subject: xfrm: reuse uncached_list to track xdsts
   5
   6 [ Upstream commit 510c321b557121861601f9d259aadd65aa274f35 ]
   7
   8 In early time, when freeing a xdst, it would be inserted into
   9 dst_garbage.list first. Then if it's refcnt was still held
  10 somewhere, later it would be put into dst_busy_list in
  11 dst_gc_task().
  12
  13 When one dev was being unregistered, the dev of these dsts in
  14 dst_busy_list would be set with loopback_dev and put this dev.
  15 So that this dev's removal wouldn't get blocked, and avoid the
  16 kmsg warning:
  17
  18   kernel:unregister_netdevice: waiting for veth0 to become \
  19   free. Usage count = 2
  20
  21 However after Commit 52df157f17e5 ("xfrm: take refcnt of dst
  22 when creating struct xfrm_dst bundle"), the xdst will not be
  23 freed with dst gc, and this warning happens.
  24
  25 To fix it, we need to find these xdsts that are still held by
  26 others when removing the dev, and free xdst's dev and set it
  27 with loopback_dev.
  28
  29 But unfortunately after flow_cache for xfrm was deleted, no
  30 list tracks them anymore. So we need to save these xdsts
  31 somewhere to release the xdst's dev later.
  32
  33 To make this easier, this patch is to reuse uncached_list to
  34 track xdsts, so that the dev refcnt can be released in the
  35 event NETDEV_UNREGISTER process of fib_netdev_notifier.
  36
  37 Thanks to Florian, we could move forward this fix quickly.
  38
  39 Fixes: 52df157f17e5 ("xfrm: take refcnt of dst when creating struct xfrm_dst bundle")
  40 Reported-by: Jianlin Shi <jishi@redhat.com>
  41 Reported-by: Hangbin Liu <liuhangbin@gmail.com>
  42 Tested-by: Eyal Birger <eyal.birger@gmail.com>
  43 Signed-off-by: Xin Long <lucien.xin@gmail.com>
  44 Signed-off-by: Steffen Klassert <steffen.klassert@secunet.com>
  45 Signed-off-by: Sasha Levin (Microsoft) <sashal@kernel.org>
  46 ---
  47  include/net/ip6_route.h |  3 +++
  48  include/net/route.h     |  3 +++
  49  net/ipv4/route.c        | 21 +++++++++++++--------
  50  net/ipv4/xfrm4_policy.c |  4 +++-
  51  net/ipv6/route.c        |  4 ++--
  52  net/ipv6/xfrm6_policy.c |  5 +++++
  53  6 files changed, 29 insertions(+), 11 deletions(-)
  54
  55 diff --git a/include/net/ip6_route.h b/include/net/ip6_route.h
  56 index bee528135cf1..00eb9162dbf1 100644
  57 --- a/include/net/ip6_route.h
  58 +++ b/include/net/ip6_route.h
  59 @@ -165,6 +165,9 @@ void rt6_mtu_change(struct net_device *dev, unsigned int mtu);
  60  void rt6_remove_prefsrc(struct inet6_ifaddr *ifp);
  61  void rt6_clean_tohost(struct net *net, struct in6_addr *gateway);
  62
  63 +void rt6_uncached_list_add(struct rt6_info *rt);
  64 +void rt6_uncached_list_del(struct rt6_info *rt);
  65 +
  66  static inline const struct rt6_info *skb_rt6_info(const struct sk_buff *skb)
  67  {
  68         const struct dst_entry *dst = skb_dst(skb);
  69 diff --git a/include/net/route.h b/include/net/route.h
  70 index 6077a0fb3044..1ab8b6f82812 100644
  71 --- a/include/net/route.h
  72 +++ b/include/net/route.h
  73 @@ -228,6 +228,9 @@ struct in_ifaddr;
  74  void fib_add_ifaddr(struct in_ifaddr *);
  75  void fib_del_ifaddr(struct in_ifaddr *, struct in_ifaddr *);
  76
  77 +void rt_add_uncached_list(struct rtable *rt);
  78 +void rt_del_uncached_list(struct rtable *rt);
  79 +
  80  static inline void ip_rt_put(struct rtable *rt)
  81  {
  82         /* dst_release() accepts a NULL parameter.
  83 diff --git a/net/ipv4/route.c b/net/ipv4/route.c
  84 index a1bf87711bfa..9fc2dbce424f 100644
  85 --- a/net/ipv4/route.c
  86 +++ b/net/ipv4/route.c
  87 @@ -1440,7 +1440,7 @@ struct uncached_list {
  88
  89  static DEFINE_PER_CPU_ALIGNED(struct uncached_list, rt_uncached_list);
  90
  91 -static void rt_add_uncached_list(struct rtable *rt)
  92 +void rt_add_uncached_list(struct rtable *rt)
  93  {
  94         struct uncached_list *ul = raw_cpu_ptr(&rt_uncached_list);
  95
  96 @@ -1451,14 +1451,8 @@ static void rt_add_uncached_list(struct rtable *rt)
  97         spin_unlock_bh(&ul->lock);
  98  }
  99
 100 -static void ipv4_dst_destroy(struct dst_entry *dst)
 101 +void rt_del_uncached_list(struct rtable *rt)
 102  {
 103 -       struct dst_metrics *p = (struct dst_metrics *)DST_METRICS_PTR(dst);
 104 -       struct rtable *rt = (struct rtable *) dst;
 105 -
 106 -       if (p != &dst_default_metrics && refcount_dec_and_test(&p->refcnt))
 107 -               kfree(p);
 108 -
 109         if (!list_empty(&rt->rt_uncached)) {
 110                 struct uncached_list *ul = rt->rt_uncached_list;
 111
 112 @@ -1468,6 +1462,17 @@ static void ipv4_dst_destroy(struct dst_entry *dst)
 113         }
 114  }
 115
 116 +static void ipv4_dst_destroy(struct dst_entry *dst)
 117 +{
 118 +       struct dst_metrics *p = (struct dst_metrics *)DST_METRICS_PTR(dst);
 119 +       struct rtable *rt = (struct rtable *)dst;
 120 +
 121 +       if (p != &dst_default_metrics && refcount_dec_and_test(&p->refcnt))
 122 +               kfree(p);
 123 +
 124 +       rt_del_uncached_list(rt);
 125 +}
 126 +
 127  void rt_flush_dev(struct net_device *dev)
 128  {
 129         struct net *net = dev_net(dev);
 130 diff --git a/net/ipv4/xfrm4_policy.c b/net/ipv4/xfrm4_policy.c
 131 index 4b586e7d5637..fbebda67ac1b 100644
 132 --- a/net/ipv4/xfrm4_policy.c
 133 +++ b/net/ipv4/xfrm4_policy.c
 134 @@ -103,6 +103,7 @@ static int xfrm4_fill_dst(struct xfrm_dst *xdst, struct net_device *dev,
 135         xdst->u.rt.rt_mtu_locked = rt->rt_mtu_locked;
 136         xdst->u.rt.rt_table_id = rt->rt_table_id;
 137         INIT_LIST_HEAD(&xdst->u.rt.rt_uncached);
 138 +       rt_add_uncached_list(&xdst->u.rt);
 139
 140         return 0;
 141  }
 142 @@ -242,7 +243,8 @@ static void xfrm4_dst_destroy(struct dst_entry *dst)
 143         struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
 144
 145         dst_destroy_metrics_generic(dst);
 146 -
 147 +       if (xdst->u.rt.rt_uncached_list)
 148 +               rt_del_uncached_list(&xdst->u.rt);
 149         xfrm_dst_destroy(xdst);
 150  }
 151
 152 diff --git a/net/ipv6/route.c b/net/ipv6/route.c
 153 index 00f8fe8cebd5..620553401d75 100644
 154 --- a/net/ipv6/route.c
 155 +++ b/net/ipv6/route.c
 156 @@ -124,7 +124,7 @@ struct uncached_list {
 157
 158  static DEFINE_PER_CPU_ALIGNED(struct uncached_list, rt6_uncached_list);
 159
 160 -static void rt6_uncached_list_add(struct rt6_info *rt)
 161 +void rt6_uncached_list_add(struct rt6_info *rt)
 162  {
 163         struct uncached_list *ul = raw_cpu_ptr(&rt6_uncached_list);
 164
 165 @@ -135,7 +135,7 @@ static void rt6_uncached_list_add(struct rt6_info *rt)
 166         spin_unlock_bh(&ul->lock);
 167  }
 168
 169 -static void rt6_uncached_list_del(struct rt6_info *rt)
 170 +void rt6_uncached_list_del(struct rt6_info *rt)
 171  {
 172         if (!list_empty(&rt->rt6i_uncached)) {
 173                 struct uncached_list *ul = rt->rt6i_uncached_list;
 174 diff --git a/net/ipv6/xfrm6_policy.c b/net/ipv6/xfrm6_policy.c
 175 index d6b012295b45..d99929113230 100644
 176 --- a/net/ipv6/xfrm6_policy.c
 177 +++ b/net/ipv6/xfrm6_policy.c
 178 @@ -113,6 +113,9 @@ static int xfrm6_fill_dst(struct xfrm_dst *xdst, struct net_device *dev,
 179         xdst->u.rt6.rt6i_gateway = rt->rt6i_gateway;
 180         xdst->u.rt6.rt6i_dst = rt->rt6i_dst;
 181         xdst->u.rt6.rt6i_src = rt->rt6i_src;
 182 +       INIT_LIST_HEAD(&xdst->u.rt6.rt6i_uncached);
 183 +       rt6_uncached_list_add(&xdst->u.rt6);
 184 +       atomic_inc(&dev_net(dev)->ipv6.rt6_stats->fib_rt_uncache);
 185
 186         return 0;
 187  }
 188 @@ -243,6 +246,8 @@ static void xfrm6_dst_destroy(struct dst_entry *dst)
 189         if (likely(xdst->u.rt6.rt6i_idev))
 190                 in6_dev_put(xdst->u.rt6.rt6i_idev);
 191         dst_destroy_metrics_generic(dst);
 192 +       if (xdst->u.rt6.rt6i_uncached_list)
 193 +               rt6_uncached_list_del(&xdst->u.rt6);
 194         xfrm_dst_destroy(xdst);
 195  }
 196
 197 --
 198 2.19.1
 199