rhashtable: Do not schedule more than one rehash if we can't grow further
authorThomas Graf <tgraf@suug.ch>
Wed, 22 Apr 2015 07:41:46 +0000 (09:41 +0200)
committerDavid S. Miller <davem@davemloft.net>
Wed, 22 Apr 2015 18:17:22 +0000 (14:17 -0400)
The current code currently only stops inserting rehashes into the
chain when no resizes are currently scheduled. As long as resizes
are scheduled and while inserting above the utilization watermark,
more and more rehashes will be scheduled.

This lead to a perfect DoS storm with thousands of rehashes
scheduled which lead to thousands of spinlocks to be taken
sequentially.

Instead, only allow either a series of resizes or a single rehash.
Drop any further rehashes and return -EBUSY.

Fixes: ccd57b1bd324 ("rhashtable: Add immediate rehash during insertion")
Signed-off-by: Thomas Graf <tgraf@suug.ch>
Acked-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
lib/rhashtable.c

index f648cfde852036f10338a5c905746c43007441e5..b28df4019adedfe182d5c719da62b6bf05c031fe 100644 (file)
@@ -405,8 +405,8 @@ int rhashtable_insert_rehash(struct rhashtable *ht)
 
        if (rht_grow_above_75(ht, tbl))
                size *= 2;
-       /* More than two rehashes (not resizes) detected. */
-       else if (WARN_ON(old_tbl != tbl && old_tbl->size == size))
+       /* Do not schedule more than one rehash */
+       else if (old_tbl != tbl)
                return -EBUSY;
 
        new_tbl = bucket_table_alloc(ht, size, GFP_ATOMIC);