net: force a reload of first item in hlist_nulls_for_each_entry_rcu
authorEric Dumazet <eric.dumazet@gmail.com>
Wed, 29 May 2013 09:06:27 +0000 (09:06 +0000)
committerDavid S. Miller <davem@davemloft.net>
Mon, 3 Jun 2013 03:53:59 +0000 (20:53 -0700)
Roman Gushchin discovered that udp4_lib_lookup2() was not reloading
first item in the rcu protected list, in case the loop was restarted.

This produced soft lockups as in https://lkml.org/lkml/2013/4/16/37

rcu_dereference(X)/ACCESS_ONCE(X) seem to not work as intended if X is
ptr->field :

In some cases, gcc caches the value or ptr->field in a register.

Use a barrier() to disallow such caching, as documented in
Documentation/atomic_ops.txt line 114

Thanks a lot to Roman for providing analysis and numerous patches.

Diagnosed-by: Roman Gushchin <klamm@yandex-team.ru>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reported-by: Boris Zhmurov <zhmurov@yandex-team.ru>
Signed-off-by: Roman Gushchin <klamm@yandex-team.ru>
Acked-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
include/linux/rculist_nulls.h

index 2ae13714828bc42568e77684fedfc2cf929c291d..1c33dd7da4a7d860004264e00bd406645df1ed96 100644 (file)
@@ -105,9 +105,14 @@ static inline void hlist_nulls_add_head_rcu(struct hlist_nulls_node *n,
  * @head:      the head for your list.
  * @member:    the name of the hlist_nulls_node within the struct.
  *
+ * The barrier() is needed to make sure compiler doesn't cache first element [1],
+ * as this loop can be restarted [2]
+ * [1] Documentation/atomic_ops.txt around line 114
+ * [2] Documentation/RCU/rculist_nulls.txt around line 146
  */
 #define hlist_nulls_for_each_entry_rcu(tpos, pos, head, member)                        \
-       for (pos = rcu_dereference_raw(hlist_nulls_first_rcu(head));            \
+       for (({barrier();}),                                                    \
+            pos = rcu_dereference_raw(hlist_nulls_first_rcu(head));            \
                (!is_a_nulls(pos)) &&                                           \
                ({ tpos = hlist_nulls_entry(pos, typeof(*tpos), member); 1; }); \
                pos = rcu_dereference_raw(hlist_nulls_next_rcu(pos)))