[PATCH net-next] flowcache: Tune per cpu flow cache shrink size

[PATCH net-next] flowcache: Tune per cpu flow cache shrink size

[Fan Du]

When per cpu flow cache entries count reaches over high_watermark,
flow cache will be shrinked gently to around low_watermark, not
brutally killing most of cached entries by left only two.

Signed-off-by: Fan Du <fan.du@windriver.com>
---
 net/core/flow.c |    2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/net/core/flow.c b/net/core/flow.c
index 31cfb365e0c6..e379cdcd699b 100644
--- a/net/core/flow.c
+++ b/net/core/flow.c
@@ -142,7 +142,7 @@ static void __flow_cache_shrink(struct flow_cache *fc,
 static void flow_cache_shrink(struct flow_cache *fc,
 			      struct flow_cache_percpu *fcp)
 {
-	int shrink_to = fc->low_watermark / flow_cache_hash_size(fc);
+	int shrink_to = fc->low_watermark;
 
 	__flow_cache_shrink(fc, fcp, shrink_to);
 }
-- 
1.7.9.5

Re: [PATCH net-next] flowcache: Tune per cpu flow cache shrink size

[Steffen Klassert]

On Fri, Mar 14, 2014 at 03:42:46PM +0800, Fan Du wrote:
> When per cpu flow cache entries count reaches over high_watermark,
> flow cache will be shrinked gently to around low_watermark, not
> brutally killing most of cached entries by left only two.
> 

We currently keep two entries per hash bucket, this means that
we keep low_watermark entries at most in the flow cache.

With your change, we would keep low_watermark entries per hash bucket.
The flow cache would grow to low_watermark * flow_cache_hash_size then.
This would make DoS attacks against the flow cache quite easy.

Re: [PATCH net-next] flowcache: Tune per cpu flow cache shrink size

[David Miller]

From: Fan Du <fan.du@windriver.com>
Date: Fri, 14 Mar 2014 15:42:46 +0800

> When per cpu flow cache entries count reaches over high_watermark,
> flow cache will be shrinked gently to around low_watermark, not
> brutally killing most of cached entries by left only two.
> 
> Signed-off-by: Fan Du <fan.du@windriver.com>

We are limiting hash chain lengths, not total hash table size.

Your change is therefore not appropriate.