[PATCH 0/5 net-next] rhashtable updates on top of Herbert's work

[PATCH 0/5 net-next] rhashtable updates on top of Herbert's work

[Thomas Graf]

Patch 1 is a bugfix for an RCU splash I encountered while testing.
Patch 2 & 3 are pure cleanups. Patch 4 disables automatic shrinking
by default as discussed in previous thread. Patch 5 removes some
rhashtable internal knowledge from nft_hash and fixes another RCU
splash.

I've pushed various rhashtable tests (Netlink, nft) together with a
Makefile to a git tree [0] for easier stress testing.

[0] https://github.com/tgraf/rhashtable

Thomas Graf (5):
  rhashtable: Extend RCU read lock into rhashtable_insert_rehash()
  rhashtable: Use 'unsigned int' consistently
  rhashtable: Mark internal/private inline functions as such
  rhashtable: Disable automatic shrinking by default
  rhashtable: Add rhashtable_free_and_destroy()

 include/linux/rhashtable.h | 50 +++++++++++++++++++++++-----------
 lib/rhashtable.c           | 67 +++++++++++++++++++++++++++++++++-------------
 lib/test_rhashtable.c      |  2 +-
 net/netfilter/nft_hash.c   | 27 ++++++-------------
 net/netlink/af_netlink.c   |  4 +--
 net/tipc/socket.c          |  3 ++-
 6 files changed, 96 insertions(+), 57 deletions(-)

-- 
1.9.3

[PATCH 1/5 net-next] rhashtable: Extend RCU read lock into rhashtable_insert_rehash()

[Thomas Graf]

rhashtable_insert_rehash() requires RCU locks to be held in order
to access ht->tbl and traverse to the last table.

Fixes: ccd57b1bd324 ("rhashtable: Add immediate rehash during insertion")
Signed-off-by: Thomas Graf <tgraf@suug.ch>
---
 include/linux/rhashtable.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/include/linux/rhashtable.h b/include/linux/rhashtable.h
index d7be9cb..5976ab5 100644
--- a/include/linux/rhashtable.h
+++ b/include/linux/rhashtable.h
@@ -585,8 +585,8 @@ restart:
 	if (unlikely(rht_grow_above_100(ht, tbl))) {
 slow_path:
 		spin_unlock_bh(lock);
-		rcu_read_unlock();
 		err = rhashtable_insert_rehash(ht);
+		rcu_read_unlock();
 		if (err)
 			return err;
 
-- 
1.9.3

[PATCH 2/5 net-next] rhashtable: Use 'unsigned int' consistently

[Thomas Graf]

Signed-off-by: Thomas Graf <tgraf@suug.ch>
---
 include/linux/rhashtable.h | 14 +++++++-------
 lib/rhashtable.c           | 18 ++++++++++--------
 2 files changed, 17 insertions(+), 15 deletions(-)

diff --git a/include/linux/rhashtable.h b/include/linux/rhashtable.h
index 5976ab5..f89cda0 100644
--- a/include/linux/rhashtable.h
+++ b/include/linux/rhashtable.h
@@ -208,13 +208,13 @@ static inline unsigned int rht_key_hashfn(
 	struct rhashtable *ht, const struct bucket_table *tbl,
 	const void *key, const struct rhashtable_params params)
 {
-	unsigned hash;
+	unsigned int hash;
 
 	/* params must be equal to ht->p if it isn't constant. */
 	if (!__builtin_constant_p(params.key_len))
 		hash = ht->p.hashfn(key, ht->key_len, tbl->hash_rnd);
 	else if (params.key_len) {
-		unsigned key_len = params.key_len;
+		unsigned int key_len = params.key_len;
 
 		if (params.hashfn)
 			hash = params.hashfn(key, key_len, tbl->hash_rnd);
@@ -224,7 +224,7 @@ static inline unsigned int rht_key_hashfn(
 			hash = jhash2(key, key_len / sizeof(u32),
 				      tbl->hash_rnd);
 	} else {
-		unsigned key_len = ht->p.key_len;
+		unsigned int key_len = ht->p.key_len;
 
 		if (params.hashfn)
 			hash = params.hashfn(key, key_len, tbl->hash_rnd);
@@ -512,7 +512,7 @@ static inline void *rhashtable_lookup_fast(
 	};
 	const struct bucket_table *tbl;
 	struct rhash_head *he;
-	unsigned hash;
+	unsigned int hash;
 
 	rcu_read_lock();
 
@@ -550,8 +550,8 @@ static inline int __rhashtable_insert_fast(
 	struct bucket_table *tbl, *new_tbl;
 	struct rhash_head *head;
 	spinlock_t *lock;
-	unsigned elasticity;
-	unsigned hash;
+	unsigned int elasticity;
+	unsigned int hash;
 	int err;
 
 restart:
@@ -718,7 +718,7 @@ static inline int __rhashtable_remove_fast(
 	struct rhash_head __rcu **pprev;
 	struct rhash_head *he;
 	spinlock_t * lock;
-	unsigned hash;
+	unsigned int hash;
 	int err = -ENOENT;
 
 	hash = rht_head_hashfn(ht, tbl, obj, params);
diff --git a/lib/rhashtable.c b/lib/rhashtable.c
index e96ad1a..93374d7 100644
--- a/lib/rhashtable.c
+++ b/lib/rhashtable.c
@@ -153,7 +153,7 @@ static struct bucket_table *rhashtable_last_table(struct rhashtable *ht,
 	return new_tbl;
 }
 
-static int rhashtable_rehash_one(struct rhashtable *ht, unsigned old_hash)
+static int rhashtable_rehash_one(struct rhashtable *ht, unsigned int old_hash)
 {
 	struct bucket_table *old_tbl = rht_dereference(ht->tbl, ht);
 	struct bucket_table *new_tbl = rhashtable_last_table(ht,
@@ -162,7 +162,7 @@ static int rhashtable_rehash_one(struct rhashtable *ht, unsigned old_hash)
 	int err = -ENOENT;
 	struct rhash_head *head, *next, *entry;
 	spinlock_t *new_bucket_lock;
-	unsigned new_hash;
+	unsigned int new_hash;
 
 	rht_for_each(entry, old_tbl, old_hash) {
 		err = 0;
@@ -199,7 +199,8 @@ out:
 	return err;
 }
 
-static void rhashtable_rehash_chain(struct rhashtable *ht, unsigned old_hash)
+static void rhashtable_rehash_chain(struct rhashtable *ht,
+				    unsigned int old_hash)
 {
 	struct bucket_table *old_tbl = rht_dereference(ht->tbl, ht);
 	spinlock_t *old_bucket_lock;
@@ -244,7 +245,7 @@ static int rhashtable_rehash_table(struct rhashtable *ht)
 	struct bucket_table *old_tbl = rht_dereference(ht->tbl, ht);
 	struct bucket_table *new_tbl;
 	struct rhashtable_walker *walker;
-	unsigned old_hash;
+	unsigned int old_hash;
 
 	new_tbl = rht_dereference(old_tbl->future_tbl, ht);
 	if (!new_tbl)
@@ -324,11 +325,12 @@ static int rhashtable_expand(struct rhashtable *ht)
 static int rhashtable_shrink(struct rhashtable *ht)
 {
 	struct bucket_table *new_tbl, *old_tbl = rht_dereference(ht->tbl, ht);
-	unsigned size = roundup_pow_of_two(atomic_read(&ht->nelems) * 3 / 2);
+	unsigned int size;
 	int err;
 
 	ASSERT_RHT_MUTEX(ht);
 
+	size = roundup_pow_of_two(atomic_read(&ht->nelems) * 3 / 2);
 	if (size < ht->p.min_size)
 		size = ht->p.min_size;
 
@@ -379,9 +381,9 @@ unlock:
 
 static bool rhashtable_check_elasticity(struct rhashtable *ht,
 					struct bucket_table *tbl,
-					unsigned hash)
+					unsigned int hash)
 {
-	unsigned elasticity = ht->elasticity;
+	unsigned int elasticity = ht->elasticity;
 	struct rhash_head *head;
 
 	rht_for_each(head, tbl, hash)
@@ -431,7 +433,7 @@ int rhashtable_insert_slow(struct rhashtable *ht, const void *key,
 			   struct bucket_table *tbl)
 {
 	struct rhash_head *head;
-	unsigned hash;
+	unsigned int hash;
 	int err;
 
 	tbl = rhashtable_last_table(ht, tbl);
-- 
1.9.3

[PATCH 3/5 net-next] rhashtable: Mark internal/private inline functions as such

[Thomas Graf]

Signed-off-by: Thomas Graf <tgraf@suug.ch>
---
 include/linux/rhashtable.h | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/include/linux/rhashtable.h b/include/linux/rhashtable.h
index f89cda0..0e1f975 100644
--- a/include/linux/rhashtable.h
+++ b/include/linux/rhashtable.h
@@ -539,6 +539,7 @@ restart:
 	return NULL;
 }
 
+/* Internal function, please use rhashtable_insert_fast() instead */
 static inline int __rhashtable_insert_fast(
 	struct rhashtable *ht, const void *key, struct rhash_head *obj,
 	const struct rhashtable_params params)
@@ -711,6 +712,7 @@ static inline int rhashtable_lookup_insert_key(
 	return __rhashtable_insert_fast(ht, key, obj, params);
 }
 
+/* Internal function, please use rhashtable_remove_fast() instead */
 static inline int __rhashtable_remove_fast(
 	struct rhashtable *ht, struct bucket_table *tbl,
 	struct rhash_head *obj, const struct rhashtable_params params)
-- 
1.9.3

[PATCH 4/5 net-next] rhashtable: Disable automatic shrinking by default

[Thomas Graf]

Using rhashtable_remove_fast() will no longer automatically
shrink the table. Users need to opt-in to shrinking by
explicitly using rhashtable_remove_and_shrink()

Signed-off-by: Thomas Graf <tgraf@suug.ch>
---
 include/linux/rhashtable.h | 27 +++++++++++++++++++++------
 lib/test_rhashtable.c      |  2 +-
 net/netfilter/nft_hash.c   |  2 +-
 net/netlink/af_netlink.c   |  4 ++--
 net/tipc/socket.c          |  3 ++-
 5 files changed, 27 insertions(+), 11 deletions(-)

diff --git a/include/linux/rhashtable.h b/include/linux/rhashtable.h
index 0e1f975..d0ca452 100644
--- a/include/linux/rhashtable.h
+++ b/include/linux/rhashtable.h
@@ -2,7 +2,7 @@
  * Resizable, Scalable, Concurrent Hash Table
  *
  * Copyright (c) 2015 Herbert Xu <herbert@gondor.apana.org.au>
- * Copyright (c) 2014 Thomas Graf <tgraf@suug.ch>
+ * Copyright (c) 2014-2015 Thomas Graf <tgraf@suug.ch>
  * Copyright (c) 2008-2014 Patrick McHardy <kaber@trash.net>
  *
  * Code partially derived from nft_hash
@@ -780,14 +780,29 @@ static inline int rhashtable_remove_fast(
 	       (tbl = rht_dereference_rcu(tbl->future_tbl, ht)))
 		;
 
-	if (err)
-		goto out;
+	if (likely(!err))
+		atomic_dec(&ht->nelems);
+
+	rcu_read_unlock();
 
-	atomic_dec(&ht->nelems);
-	if (rht_shrink_below_30(ht, tbl))
+	return err;
+}
+
+static inline int rhashtable_remove_and_shrink(
+	struct rhashtable *ht, struct rhash_head *obj,
+	const struct rhashtable_params params)
+{
+	struct bucket_table *tbl;
+	int err;
+
+	rcu_read_lock();
+
+	tbl = rht_dereference_rcu(ht->tbl, ht);
+
+	err = rhashtable_remove_fast(ht, obj, params);
+	if (!err && rht_shrink_below_30(ht, tbl))
 		schedule_work(&ht->run_work);
 
-out:
 	rcu_read_unlock();
 
 	return err;
diff --git a/lib/test_rhashtable.c b/lib/test_rhashtable.c
index a42a0d4..fe6d4a9 100644
--- a/lib/test_rhashtable.c
+++ b/lib/test_rhashtable.c
@@ -166,7 +166,7 @@ static int __init test_rhashtable(struct rhashtable *ht)
 		obj = rhashtable_lookup_fast(ht, &key, test_rht_params);
 		BUG_ON(!obj);
 
-		rhashtable_remove_fast(ht, &obj->node, test_rht_params);
+		rhashtable_remove_and_shrink(ht, &obj->node, test_rht_params);
 		kfree(obj);
 	}
 
diff --git a/net/netfilter/nft_hash.c b/net/netfilter/nft_hash.c
index ad39669..2986f1d 100644
--- a/net/netfilter/nft_hash.c
+++ b/net/netfilter/nft_hash.c
@@ -89,7 +89,7 @@ static void nft_hash_remove(const struct nft_set *set,
 {
 	struct rhashtable *priv = nft_set_priv(set);
 
-	rhashtable_remove_fast(priv, elem->cookie, nft_hash_params);
+	rhashtable_remove_and_shrink(priv, elem->cookie, nft_hash_params);
 	synchronize_rcu();
 	kfree(elem->cookie);
 }
diff --git a/net/netlink/af_netlink.c b/net/netlink/af_netlink.c
index e2f7f28..b9d3215 100644
--- a/net/netlink/af_netlink.c
+++ b/net/netlink/af_netlink.c
@@ -1094,8 +1094,8 @@ static void netlink_remove(struct sock *sk)
 	struct netlink_table *table;
 
 	table = &nl_table[sk->sk_protocol];
-	if (!rhashtable_remove_fast(&table->hash, &nlk_sk(sk)->node,
-				    netlink_rhashtable_params)) {
+	if (!rhashtable_remove_and_shrink(&table->hash, &nlk_sk(sk)->node,
+					  netlink_rhashtable_params)) {
 		WARN_ON(atomic_read(&sk->sk_refcnt) == 1);
 		__sock_put(sk);
 	}
diff --git a/net/tipc/socket.c b/net/tipc/socket.c
index 6dd5bd9..33338b4 100644
--- a/net/tipc/socket.c
+++ b/net/tipc/socket.c
@@ -2282,7 +2282,8 @@ static void tipc_sk_remove(struct tipc_sock *tsk)
 	struct sock *sk = &tsk->sk;
 	struct tipc_net *tn = net_generic(sock_net(sk), tipc_net_id);
 
-	if (!rhashtable_remove_fast(&tn->sk_rht, &tsk->node, tsk_rht_params)) {
+	if (!rhashtable_remove_and_shrink(&tn->sk_rht, &tsk->node,
+					  tsk_rht_params)) {
 		WARN_ON(atomic_read(&sk->sk_refcnt) == 1);
 		__sock_put(sk);
 	}
-- 
1.9.3

[PATCH 5/5 net-next] rhashtable: Add rhashtable_free_and_destroy()

[Thomas Graf]

rhashtable_destroy() variant which stops rehashes, iterates over
the table and calls a callback to release resources.

Avoids need for nft_hash to embed rhashtable internals and allows to
get rid of the being_destroyed flag. It also saves a 2nd mutex
lock upon destruction.

Also fixes an RCU lockdep splash on nft set destruction due to
calling rht_for_each_entry_safe() without holding bucket locks.
Open code this loop as we need know that no mutations may occur in
parallel.

Signed-off-by: Thomas Graf <tgraf@suug.ch>
---
 include/linux/rhashtable.h |  5 +++--
 lib/rhashtable.c           | 49 ++++++++++++++++++++++++++++++++++++----------
 net/netfilter/nft_hash.c   | 25 +++++++----------------
 3 files changed, 49 insertions(+), 30 deletions(-)

diff --git a/include/linux/rhashtable.h b/include/linux/rhashtable.h
index d0ca452..1995db3 100644
--- a/include/linux/rhashtable.h
+++ b/include/linux/rhashtable.h
@@ -134,12 +134,10 @@ struct rhashtable_params {
  * @run_work: Deferred worker to expand/shrink asynchronously
  * @mutex: Mutex to protect current/future table swapping
  * @lock: Spin lock to protect walker list
- * @being_destroyed: True if table is set up for destruction
  */
 struct rhashtable {
 	struct bucket_table __rcu	*tbl;
 	atomic_t			nelems;
-	bool                            being_destroyed;
 	unsigned int			key_len;
 	unsigned int			elasticity;
 	struct rhashtable_params	p;
@@ -332,6 +330,9 @@ int rhashtable_walk_start(struct rhashtable_iter *iter) __acquires(RCU);
 void *rhashtable_walk_next(struct rhashtable_iter *iter);
 void rhashtable_walk_stop(struct rhashtable_iter *iter) __releases(RCU);
 
+void rhashtable_free_and_destroy(struct rhashtable *ht,
+				 void (*free_fn)(void *ptr, void *arg),
+				 void *arg);
 void rhashtable_destroy(struct rhashtable *ht);
 
 #define rht_dereference(p, ht) \
diff --git a/lib/rhashtable.c b/lib/rhashtable.c
index 93374d7..bb3a792 100644
--- a/lib/rhashtable.c
+++ b/lib/rhashtable.c
@@ -359,8 +359,6 @@ static void rht_deferred_worker(struct work_struct *work)
 
 	ht = container_of(work, struct rhashtable, run_work);
 	mutex_lock(&ht->mutex);
-	if (ht->being_destroyed)
-		goto unlock;
 
 	tbl = rht_dereference(ht->tbl, ht);
 	tbl = rhashtable_last_table(ht, tbl);
@@ -372,7 +370,6 @@ static void rht_deferred_worker(struct work_struct *work)
 
 	err = rhashtable_rehash_table(ht);
 
-unlock:
 	mutex_unlock(&ht->mutex);
 
 	if (err)
@@ -771,21 +768,53 @@ int rhashtable_init(struct rhashtable *ht,
 EXPORT_SYMBOL_GPL(rhashtable_init);
 
 /**
- * rhashtable_destroy - destroy hash table
+ * rhashtable_free_and_destroy - free elements and destroy hash table
  * @ht:		the hash table to destroy
+ * @free_fn:	callback to release resources of element
+ * @arg:	pointer passed to free_fn
  *
- * Frees the bucket array. This function is not rcu safe, therefore the caller
- * has to make sure that no resizing may happen by unpublishing the hashtable
- * and waiting for the quiescent cycle before releasing the bucket array.
+ * Stops an eventual async resize. If defined, invokes free_fn for each
+ * element to releasal resources. Please note that RCU protected
+ * readers may still be accessing the elements. Releasing of resources
+ * must occur in a compatible manner. Then frees the bucket array.
+ *
+ * This function will eventually sleep to wait for an async resize
+ * to complete. The caller is responsible that no further write operations
+ * occurs in parallel.
  */
-void rhashtable_destroy(struct rhashtable *ht)
+void rhashtable_free_and_destroy(struct rhashtable *ht,
+				 void (*free_fn)(void *ptr, void *arg),
+				 void *arg)
 {
-	ht->being_destroyed = true;
+	const struct bucket_table *tbl;
+	unsigned int i;
 
 	cancel_work_sync(&ht->run_work);
 
 	mutex_lock(&ht->mutex);
-	bucket_table_free(rht_dereference(ht->tbl, ht));
+	tbl = rht_dereference(ht->tbl, ht);
+	if (free_fn) {
+		for (i = 0; i < tbl->size; i++) {
+			struct rhash_head *pos, *next;
+
+			for (pos = rht_dereference(tbl->buckets[i], ht),
+			     next = !rht_is_a_nulls(pos) ?
+					rht_dereference(pos->next, ht) : NULL;
+			     !rht_is_a_nulls(pos);
+			     pos = next,
+			     next = !rht_is_a_nulls(pos) ?
+					rht_dereference(pos->next, ht) : NULL)
+				free_fn(rht_obj(ht, pos), arg);
+		}
+	}
+
+	bucket_table_free(tbl);
 	mutex_unlock(&ht->mutex);
 }
+EXPORT_SYMBOL_GPL(rhashtable_free_and_destroy);
+
+void rhashtable_destroy(struct rhashtable *ht)
+{
+	return rhashtable_free_and_destroy(ht, NULL, NULL);
+}
 EXPORT_SYMBOL_GPL(rhashtable_destroy);
diff --git a/net/netfilter/nft_hash.c b/net/netfilter/nft_hash.c
index 2986f1d..ad29965 100644
--- a/net/netfilter/nft_hash.c
+++ b/net/netfilter/nft_hash.c
@@ -187,26 +187,15 @@ static int nft_hash_init(const struct nft_set *set,
 	return rhashtable_init(priv, &params);
 }
 
-static void nft_hash_destroy(const struct nft_set *set)
+static void nft_free_element(void *ptr, void *arg)
 {
-	struct rhashtable *priv = nft_set_priv(set);
-	const struct bucket_table *tbl;
-	struct nft_hash_elem *he;
-	struct rhash_head *pos, *next;
-	unsigned int i;
-
-	/* Stop an eventual async resizing */
-	priv->being_destroyed = true;
-	mutex_lock(&priv->mutex);
-
-	tbl = rht_dereference(priv->tbl, priv);
-	for (i = 0; i < tbl->size; i++) {
-		rht_for_each_entry_safe(he, pos, next, tbl, i, node)
-			nft_hash_elem_destroy(set, he);
-	}
-	mutex_unlock(&priv->mutex);
+	nft_hash_elem_destroy((const struct nft_set *)arg, ptr);
+}
 
-	rhashtable_destroy(priv);
+static void nft_hash_destroy(const struct nft_set *set)
+{
+	rhashtable_free_and_destroy(nft_set_priv(set), nft_free_element,
+				    (void *)set);
 }
 
 static bool nft_hash_estimate(const struct nft_set_desc *desc, u32 features,
-- 
1.9.3

Re: [PATCH 4/5 net-next] rhashtable: Disable automatic shrinking by default

[Herbert Xu]

On Tue, Mar 24, 2015 at 02:18:19PM +0100, Thomas Graf wrote:
> Using rhashtable_remove_fast() will no longer automatically
> shrink the table. Users need to opt-in to shrinking by
> explicitly using rhashtable_remove_and_shrink()
> 
> Signed-off-by: Thomas Graf <tgraf@suug.ch>

Note that this patch doesn't completely disable automatic shrinking
since the worker thread can still trigger it.

Cheers,
-- 
Email: Herbert Xu <herbert@gondor.apana.org.au>
Home Page: http://gondor.apana.org.au/~herbert/
PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt

Re: [PATCH 4/5 net-next] rhashtable: Disable automatic shrinking by default

[Thomas Graf]

On 03/25/15 at 05:55am, Herbert Xu wrote:
> On Tue, Mar 24, 2015 at 02:18:19PM +0100, Thomas Graf wrote:
> > Using rhashtable_remove_fast() will no longer automatically
> > shrink the table. Users need to opt-in to shrinking by
> > explicitly using rhashtable_remove_and_shrink()
> > 
> > Signed-off-by: Thomas Graf <tgraf@suug.ch>
> 
> Note that this patch doesn't completely disable automatic shrinking
> since the worker thread can still trigger it.

Right. I'll send a v2.

Re: [PATCH 4/5 net-next] rhashtable: Disable automatic shrinking by default

[David Miller]

From: Herbert Xu <herbert@gondor.apana.org.au>
Date: Wed, 25 Mar 2015 05:55:45 +1100

> On Tue, Mar 24, 2015 at 02:18:19PM +0100, Thomas Graf wrote:
>> Using rhashtable_remove_fast() will no longer automatically
>> shrink the table. Users need to opt-in to shrinking by
>> explicitly using rhashtable_remove_and_shrink()
>> 
>> Signed-off-by: Thomas Graf <tgraf@suug.ch>
> 
> Note that this patch doesn't completely disable automatic shrinking
> since the worker thread can still trigger it.

Ok, but how do you feel about Thomas's series otherwise?

It looks fine to me.

Re: [PATCH 1/5 net-next] rhashtable: Extend RCU read lock into rhashtable_insert_rehash()

[Thomas Graf]

On 03/24/15 at 02:18pm, Thomas Graf wrote:
> @@ -585,8 +585,8 @@ restart:
>  	if (unlikely(rht_grow_above_100(ht, tbl))) {
>  slow_path:
>  		spin_unlock_bh(lock);
> -		rcu_read_unlock();
>  		err = rhashtable_insert_rehash(ht);
> +		rcu_read_unlock();
>  		if (err)
>  			return err;

An alternative would be to start a new RCU section inside
rhashtable_insert_rehash(). This was the easier fix but I'm
fine either way.

[PATCH 4/5 net-next v2] rhashtable: Disable automatic shrinking by default

[Thomas Graf]

Introduce a new bool automatic_shrinking to require the
user to explicitly opt-in to automatic shrinking of tables.

Signed-off-by: Thomas Graf <tgraf@suug.ch>
---
v2: Use bool in params to also disable shrinking in worker as
    pointed out by Herbert.

@Dave: Patch 5 still applied on top with some fuzz. If not,
       let me know and I will respin.

 include/linux/rhashtable.h | 7 +++++--
 lib/rhashtable.c           | 2 +-
 net/netfilter/nft_hash.c   | 1 +
 net/netlink/af_netlink.c   | 1 +
 net/tipc/socket.c          | 1 +
 5 files changed, 9 insertions(+), 3 deletions(-)

diff --git a/include/linux/rhashtable.h b/include/linux/rhashtable.h
index 0e1f975..ae26c49 100644
--- a/include/linux/rhashtable.h
+++ b/include/linux/rhashtable.h
@@ -2,7 +2,7 @@
  * Resizable, Scalable, Concurrent Hash Table
  *
  * Copyright (c) 2015 Herbert Xu <herbert@gondor.apana.org.au>
- * Copyright (c) 2014 Thomas Graf <tgraf@suug.ch>
+ * Copyright (c) 2014-2015 Thomas Graf <tgraf@suug.ch>
  * Copyright (c) 2008-2014 Patrick McHardy <kaber@trash.net>
  *
  * Code partially derived from nft_hash
@@ -104,6 +104,7 @@ struct rhashtable;
  * @min_size: Minimum size while shrinking
  * @nulls_base: Base value to generate nulls marker
  * @insecure_elasticity: Set to true to disable chain length checks
+ * @automatic_shrinking: Enable automatic shrinking of tables
  * @locks_mul: Number of bucket locks to allocate per cpu (default: 128)
  * @hashfn: Hash function (default: jhash2 if !(key_len % 4), or jhash)
  * @obj_hashfn: Function to hash object
@@ -118,6 +119,7 @@ struct rhashtable_params {
 	unsigned int		min_size;
 	u32			nulls_base;
 	bool			insecure_elasticity;
+	bool			automatic_shrinking;
 	size_t			locks_mul;
 	rht_hashfn_t		hashfn;
 	rht_obj_hashfn_t	obj_hashfn;
@@ -784,7 +786,8 @@ static inline int rhashtable_remove_fast(
 		goto out;
 
 	atomic_dec(&ht->nelems);
-	if (rht_shrink_below_30(ht, tbl))
+	if (unlikely(ht->p.automatic_shrinking &&
+		     rht_shrink_below_30(ht, tbl)))
 		schedule_work(&ht->run_work);
 
 out:
diff --git a/lib/rhashtable.c b/lib/rhashtable.c
index 93374d7..a0ad777 100644
--- a/lib/rhashtable.c
+++ b/lib/rhashtable.c
@@ -367,7 +367,7 @@ static void rht_deferred_worker(struct work_struct *work)
 
 	if (rht_grow_above_75(ht, tbl))
 		rhashtable_expand(ht);
-	else if (rht_shrink_below_30(ht, tbl))
+	else if (ht->p.automatic_shrinking && rht_shrink_below_30(ht, tbl))
 		rhashtable_shrink(ht);
 
 	err = rhashtable_rehash_table(ht);
diff --git a/net/netfilter/nft_hash.c b/net/netfilter/nft_hash.c
index ad39669..8577a37 100644
--- a/net/netfilter/nft_hash.c
+++ b/net/netfilter/nft_hash.c
@@ -172,6 +172,7 @@ static const struct rhashtable_params nft_hash_params = {
 	.head_offset = offsetof(struct nft_hash_elem, node),
 	.key_offset = offsetof(struct nft_hash_elem, key),
 	.hashfn = jhash,
+	.automatic_shrinking = true,
 };
 
 static int nft_hash_init(const struct nft_set *set,
diff --git a/net/netlink/af_netlink.c b/net/netlink/af_netlink.c
index e2f7f28..4caa809 100644
--- a/net/netlink/af_netlink.c
+++ b/net/netlink/af_netlink.c
@@ -3142,6 +3142,7 @@ static const struct rhashtable_params netlink_rhashtable_params = {
 	.obj_hashfn = netlink_hash,
 	.obj_cmpfn = netlink_compare,
 	.max_size = 65536,
+	.automatic_shrinking = true,
 };
 
 static int __init netlink_proto_init(void)
diff --git a/net/tipc/socket.c b/net/tipc/socket.c
index 6dd5bd9..3cc099d 100644
--- a/net/tipc/socket.c
+++ b/net/tipc/socket.c
@@ -2295,6 +2295,7 @@ static const struct rhashtable_params tsk_rht_params = {
 	.key_len = sizeof(u32), /* portid */
 	.max_size = 1048576,
 	.min_size = 256,
+	.automatic_shrinking = true,
 };
 
 int tipc_sk_rht_init(struct net *net)
-- 
1.9.3

Re: [PATCH 4/5 net-next v2] rhashtable: Disable automatic shrinking by default

[David Miller]

From: Thomas Graf <tgraf@suug.ch>
Date: Tue, 24 Mar 2015 20:42:19 +0000

> Introduce a new bool automatic_shrinking to require the
> user to explicitly opt-in to automatic shrinking of tables.
> 
> Signed-off-by: Thomas Graf <tgraf@suug.ch>
> ---
> v2: Use bool in params to also disable shrinking in worker as
>     pointed out by Herbert.
> 
> @Dave: Patch 5 still applied on top with some fuzz. If not,
>        let me know and I will respin.

Ok I applied this series with the updated patch #4, thanks!