rhashtable: Lower/upper bucket may map to same lock while shrinking
authorThomas Graf <tgraf@suug.ch>
Mon, 12 Jan 2015 23:58:21 +0000 (23:58 +0000)
committerDavid S. Miller <davem@davemloft.net>
Wed, 14 Jan 2015 05:21:44 +0000 (00:21 -0500)
Each per bucket lock covers a configurable number of buckets. While
shrinking, two buckets in the old table contain entries for a single
bucket in the new table. We need to lock down both while linking.
Check if they are protected by different locks to avoid a recursive
lock.

Fixes: 97defe1e ("rhashtable: Per bucket locks & deferred expansion/shrinking")
Reported-by: Fengguang Wu <fengguang.wu@intel.com>
Signed-off-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
lib/rhashtable.c

index ed6ae1ad304cea6425162f8056b0cb40828ba224..aca699813ba9a8462fed5801eecd0b5d09870242 100644 (file)
@@ -443,8 +443,16 @@ int rhashtable_shrink(struct rhashtable *ht)
                new_bucket_lock = bucket_lock(new_tbl, new_hash);
 
                spin_lock_bh(old_bucket_lock1);
-               spin_lock_bh_nested(old_bucket_lock2, RHT_LOCK_NESTED);
-               spin_lock_bh_nested(new_bucket_lock, RHT_LOCK_NESTED2);
+
+               /* Depending on the lock per buckets mapping, the bucket in
+                * the lower and upper region may map to the same lock.
+                */
+               if (old_bucket_lock1 != old_bucket_lock2) {
+                       spin_lock_bh_nested(old_bucket_lock2, RHT_LOCK_NESTED);
+                       spin_lock_bh_nested(new_bucket_lock, RHT_LOCK_NESTED2);
+               } else {
+                       spin_lock_bh_nested(new_bucket_lock, RHT_LOCK_NESTED);
+               }
 
                rcu_assign_pointer(*bucket_tail(new_tbl, new_hash),
                                   tbl->buckets[new_hash]);
@@ -452,7 +460,8 @@ int rhashtable_shrink(struct rhashtable *ht)
                                   tbl->buckets[new_hash + new_tbl->size]);
 
                spin_unlock_bh(new_bucket_lock);
-               spin_unlock_bh(old_bucket_lock2);
+               if (old_bucket_lock1 != old_bucket_lock2)
+                       spin_unlock_bh(old_bucket_lock2);
                spin_unlock_bh(old_bucket_lock1);
        }