diff mbox series

[net-next,v4,1/6] Add hlist_node to struct neighbour

Message ID 20241015165929.3203216-2-gnaaman@drivenets.com (mailing list archive)
State Superseded
Delegated to: Netdev Maintainers
Headers show
Series Improve neigh_flush_dev performance | expand

Checks

Context Check Description
netdev/tree_selection success Clearly marked for net-next, async
netdev/apply fail Patch does not apply to net-next-1

Commit Message

Gilad Naaman Oct. 15, 2024, 4:59 p.m. UTC 
Add a doubly-linked node to neighbours, so that they
can be deleted without iterating the entire bucket they're in.

Signed-off-by: Gilad Naaman <gnaaman@drivenets.com>
---
 include/net/neighbour.h |  2 ++
 net/core/neighbour.c    | 31 ++++++++++++++++++++++++++++++-
 2 files changed, 32 insertions(+), 1 deletion(-)

Comments

Kuniyuki Iwashima Oct. 15, 2024, 10:49 p.m. UTC | #1 
From: Gilad Naaman <gnaaman@drivenets.com>
Date: Tue, 15 Oct 2024 16:59:21 +0000
> @@ -531,7 +533,9 @@ static void neigh_get_hash_rnd(u32 *x)
>  static struct neigh_hash_table *neigh_hash_alloc(unsigned int shift)
>  {
>  	size_t size = (1 << shift) * sizeof(struct neighbour *);
> +	size_t hash_heads_size = (1 << shift) * sizeof(struct hlist_head);
>  	struct neigh_hash_table *ret;
> +	struct hlist_head *hash_heads;
>  	struct neighbour __rcu **buckets;
>  	int i;

nit:

While at it, please sort variables in reverse xmas tree order.
Same for other places.

https://docs.kernel.org/process/maintainer-netdev.html#local-variable-ordering-reverse-xmas-tree-rcs

>  
> @@ -540,17 +544,28 @@ static struct neigh_hash_table *neigh_hash_alloc(unsigned int shift)
>  		return NULL;
>  	if (size <= PAGE_SIZE) {
>  		buckets = kzalloc(size, GFP_ATOMIC);
> +		hash_heads = kzalloc(hash_heads_size, GFP_ATOMIC);
> +		if (!hash_heads)
> +			kfree(buckets);
>  	} else {
>  		buckets = (struct neighbour __rcu **)
>  			  __get_free_pages(GFP_ATOMIC | __GFP_ZERO,
>  					   get_order(size));
>  		kmemleak_alloc(buckets, size, 1, GFP_ATOMIC);
> +
> +		hash_heads = (struct hlist_head *)
> +			  __get_free_pages(GFP_ATOMIC | __GFP_ZERO,
> +					   get_order(hash_heads_size));
> +		kmemleak_alloc(hash_heads, hash_heads_size, 1, GFP_ATOMIC);
> +		if (!hash_heads)
> +			free_pages((unsigned long)buckets, get_order(size));
>  	}
> -	if (!buckets) {
> +	if (!buckets || !hash_heads) {
>  		kfree(ret);
>  		return NULL;

If buckets is NULL and hash_heads isn't, hash_heads is leaked.


>  	}
>  	ret->hash_buckets = buckets;
> +	ret->hash_heads = hash_heads;
>  	ret->hash_shift = shift;
>  	for (i = 0; i < NEIGH_NUM_HASH_RND; i++)
>  		neigh_get_hash_rnd(&ret->hash_rnd[i]);
diff mbox series

Patch

diff --git a/include/net/neighbour.h b/include/net/neighbour.h
index a44f262a7384..5f2b7249ba02 100644
--- a/include/net/neighbour.h
+++ b/include/net/neighbour.h
@@ -136,6 +136,7 @@  struct neigh_statistics {
 
 struct neighbour {
 	struct neighbour __rcu	*next;
+	struct hlist_node	hash;
 	struct neigh_table	*tbl;
 	struct neigh_parms	*parms;
 	unsigned long		confirmed;
@@ -191,6 +192,7 @@  struct pneigh_entry {
 
 struct neigh_hash_table {
 	struct neighbour __rcu	**hash_buckets;
+	struct hlist_head	*hash_heads;
 	unsigned int		hash_shift;
 	__u32			hash_rnd[NEIGH_NUM_HASH_RND];
 	struct rcu_head		rcu;
diff --git a/net/core/neighbour.c b/net/core/neighbour.c
index 77b819cd995b..01987368b6c5 100644
--- a/net/core/neighbour.c
+++ b/net/core/neighbour.c
@@ -217,6 +217,7 @@  static bool neigh_del(struct neighbour *n, struct neighbour __rcu **np,
 		neigh = rcu_dereference_protected(n->next,
 						  lockdep_is_held(&tbl->lock));
 		rcu_assign_pointer(*np, neigh);
+		hlist_del_rcu(&n->hash);
 		neigh_mark_dead(n);
 		retval = true;
 	}
@@ -403,6 +404,7 @@  static void neigh_flush_dev(struct neigh_table *tbl, struct net_device *dev,
 			rcu_assign_pointer(*np,
 				   rcu_dereference_protected(n->next,
 						lockdep_is_held(&tbl->lock)));
+			hlist_del_rcu(&n->hash);
 			write_lock(&n->lock);
 			neigh_del_timer(n);
 			neigh_mark_dead(n);
@@ -531,7 +533,9 @@  static void neigh_get_hash_rnd(u32 *x)
 static struct neigh_hash_table *neigh_hash_alloc(unsigned int shift)
 {
 	size_t size = (1 << shift) * sizeof(struct neighbour *);
+	size_t hash_heads_size = (1 << shift) * sizeof(struct hlist_head);
 	struct neigh_hash_table *ret;
+	struct hlist_head *hash_heads;
 	struct neighbour __rcu **buckets;
 	int i;
 
@@ -540,17 +544,28 @@  static struct neigh_hash_table *neigh_hash_alloc(unsigned int shift)
 		return NULL;
 	if (size <= PAGE_SIZE) {
 		buckets = kzalloc(size, GFP_ATOMIC);
+		hash_heads = kzalloc(hash_heads_size, GFP_ATOMIC);
+		if (!hash_heads)
+			kfree(buckets);
 	} else {
 		buckets = (struct neighbour __rcu **)
 			  __get_free_pages(GFP_ATOMIC | __GFP_ZERO,
 					   get_order(size));
 		kmemleak_alloc(buckets, size, 1, GFP_ATOMIC);
+
+		hash_heads = (struct hlist_head *)
+			  __get_free_pages(GFP_ATOMIC | __GFP_ZERO,
+					   get_order(hash_heads_size));
+		kmemleak_alloc(hash_heads, hash_heads_size, 1, GFP_ATOMIC);
+		if (!hash_heads)
+			free_pages((unsigned long)buckets, get_order(size));
 	}
-	if (!buckets) {
+	if (!buckets || !hash_heads) {
 		kfree(ret);
 		return NULL;
 	}
 	ret->hash_buckets = buckets;
+	ret->hash_heads = hash_heads;
 	ret->hash_shift = shift;
 	for (i = 0; i < NEIGH_NUM_HASH_RND; i++)
 		neigh_get_hash_rnd(&ret->hash_rnd[i]);
@@ -564,6 +579,8 @@  static void neigh_hash_free_rcu(struct rcu_head *head)
 						    rcu);
 	size_t size = (1 << nht->hash_shift) * sizeof(struct neighbour *);
 	struct neighbour __rcu **buckets = nht->hash_buckets;
+	size_t hash_heads_size = (1 << nht->hash_shift) * sizeof(struct hlist_head);
+	struct hlist_head *hash_heads = nht->hash_heads;
 
 	if (size <= PAGE_SIZE) {
 		kfree(buckets);
@@ -571,6 +588,13 @@  static void neigh_hash_free_rcu(struct rcu_head *head)
 		kmemleak_free(buckets);
 		free_pages((unsigned long)buckets, get_order(size));
 	}
+
+	if (hash_heads_size < PAGE_SIZE) {
+		kfree(hash_heads);
+	} else {
+		kmemleak_free(hash_heads);
+		free_pages((unsigned long)hash_heads, get_order(hash_heads_size));
+	}
 	kfree(nht);
 }
 
@@ -607,6 +631,8 @@  static struct neigh_hash_table *neigh_hash_grow(struct neigh_table *tbl,
 						new_nht->hash_buckets[hash],
 						lockdep_is_held(&tbl->lock)));
 			rcu_assign_pointer(new_nht->hash_buckets[hash], n);
+			hlist_del_rcu(&n->hash);
+			hlist_add_head_rcu(&n->hash, &new_nht->hash_heads[hash]);
 		}
 	}
 
@@ -717,6 +743,7 @@  ___neigh_create(struct neigh_table *tbl, const void *pkey,
 			   rcu_dereference_protected(nht->hash_buckets[hash_val],
 						     lockdep_is_held(&tbl->lock)));
 	rcu_assign_pointer(nht->hash_buckets[hash_val], n);
+	hlist_add_head_rcu(&n->hash, &nht->hash_heads[hash_val]);
 	write_unlock_bh(&tbl->lock);
 	neigh_dbg(2, "neigh %p is created\n", n);
 	rc = n;
@@ -1002,6 +1029,7 @@  static void neigh_periodic_work(struct work_struct *work)
 				rcu_assign_pointer(*np,
 					rcu_dereference_protected(n->next,
 						lockdep_is_held(&tbl->lock)));
+				hlist_del_rcu(&n->hash);
 				neigh_mark_dead(n);
 				write_unlock(&n->lock);
 				neigh_cleanup_and_release(n);
@@ -3131,6 +3159,7 @@  void __neigh_for_each_release(struct neigh_table *tbl,
 				rcu_assign_pointer(*np,
 					rcu_dereference_protected(n->next,
 						lockdep_is_held(&tbl->lock)));
+				hlist_del_rcu(&n->hash);
 				neigh_mark_dead(n);
 			} else
 				np = &n->next;