allowedips: allocate nodes in kmem_cache

The previous commit moved from O(n) to O(1) for removal, but in the
process introduced an additional pointer member to a struct that
increased the size from 60 to 68 bytes, putting nodes in the 128-byte
slab. With deployed systems having as many as 2 million nodes, this
represents a significant doubling in memory usage (128 MiB -> 256 MiB).
Fix this by using our own kmem_cache, that's sized exactly right. This
also makes wireguard's memory usage more transparent in tools like
slabtop and /proc/slabinfo.

Suggested-by: Arnd Bergmann <arnd@arndb.de>
Suggested-by: Matthew Wilcox <willy@infradead.org>
Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com>

3 files changed, 38 insertions, 13 deletions

diff --git a/src/allowedips.c b/src/allowedips.c
index e407c69..c540dce 100644
--- a/src/allowedips.c
+++ b/src/allowedips.c
@@ -6,6 +6,8 @@
 #include "allowedips.h"
 #include "peer.h"
 
+static struct kmem_cache *node_cache;
+
 static void swap_endian(u8 *dst, const u8 *src, u8 bits)
 {
 	if (bits == 32) {
@@ -31,11 +33,6 @@ static void copy_and_assign_cidr(struct allowedips_node *node, const u8 *src,
 #define CHOOSE_NODE(parent, key) \
 	parent->bit[(key[parent->bit_at_a] >> parent->bit_at_b) & 1]
 
-static void node_free_rcu(struct rcu_head *rcu)
-{
-	kfree(container_of(rcu, struct allowedips_node, rcu));
-}
-
 static void push_rcu(struct allowedips_node **stack,
 		     struct allowedips_node __rcu *p, unsigned int *len)
 {
@@ -45,6 +42,11 @@ static void push_rcu(struct allowedips_node **stack,
 	}
 }
 
+static void node_free_rcu(struct rcu_head *rcu)
+{
+	kmem_cache_free(node_cache, container_of(rcu, struct allowedips_node, rcu));
+}
+
 static void root_free_rcu(struct rcu_head *rcu)
 {
 	struct allowedips_node *node, *stack[128] = {
@@ -54,7 +56,7 @@ static void root_free_rcu(struct rcu_head *rcu)
 	while (len > 0 && (node = stack[--len])) {
 		push_rcu(stack, node->bit[0], &len);
 		push_rcu(stack, node->bit[1], &len);
-		kfree(node);
+		kmem_cache_free(node_cache, node);
 	}
 }
 
@@ -169,7 +171,7 @@ static int add(struct allowedips_node __rcu **trie, u8 bits, const u8 *key,
 		return -EINVAL;
 
 	if (!rcu_access_pointer(*trie)) {
-		node = kzalloc(sizeof(*node), GFP_KERNEL);
+		node = kmem_cache_zalloc(node_cache, GFP_KERNEL);
 		if (unlikely(!node))
 			return -ENOMEM;
 		RCU_INIT_POINTER(node->peer, peer);
@@ -185,7 +187,7 @@ static int add(struct allowedips_node __rcu **trie, u8 bits, const u8 *key,
 		return 0;
 	}
 
-	newnode = kzalloc(sizeof(*newnode), GFP_KERNEL);
+	newnode = kmem_cache_zalloc(node_cache, GFP_KERNEL);
 	if (unlikely(!newnode))
 		return -ENOMEM;
 	RCU_INIT_POINTER(newnode->peer, peer);
@@ -218,10 +220,10 @@ static int add(struct allowedips_node __rcu **trie, u8 bits, const u8 *key,
 		return 0;
 	}
 
-	node = kzalloc(sizeof(*node), GFP_KERNEL);
+	node = kmem_cache_zalloc(node_cache, GFP_KERNEL);
 	if (unlikely(!node)) {
 		list_del(&newnode->peer_list);
-		kfree(newnode);
+		kmem_cache_free(node_cache, newnode);
 		return -ENOMEM;
 	}
 	INIT_LIST_HEAD(&node->peer_list);
@@ -311,7 +313,7 @@ void wg_allowedips_remove_by_peer(struct allowedips *table,
 		if (child)
 			child->parent_bit = node->parent_bit;
 		*rcu_dereference_protected(node->parent_bit, lockdep_is_held(lock)) = child;
-		kfree_rcu(node, rcu);
+		call_rcu(&node->rcu, node_free_rcu);
 
 		/* TODO: Note that we currently don't walk up and down in order to
 		 * free any potential filler nodes. This means that this function
@@ -355,4 +357,16 @@ struct wg_peer *wg_allowedips_lookup_src(struct allowedips *table,
 	return NULL;
 }
 
+int __init wg_allowedips_slab_init(void)
+{
+	node_cache = KMEM_CACHE(allowedips_node, 0);
+	return node_cache ? 0 : -ENOMEM;
+}
+
+void wg_allowedips_slab_uninit(void)
+{
+	rcu_barrier();
+	kmem_cache_destroy(node_cache);
+}
+
 #include "selftest/allowedips.c"
diff --git a/src/allowedips.h b/src/allowedips.h
index f08f552..32d611a 100644
--- a/src/allowedips.h
+++ b/src/allowedips.h
@@ -19,7 +19,7 @@ struct allowedips_node {
 	u8 bits[16] __aligned(__alignof(u64));
 
 	/* Keep rarely used members at bottom to be beyond cache line. */
-	struct allowedips_node *__rcu *parent_bit; /* XXX: this puts us at 68->128 bytes instead of 60->64 bytes!! */
+	struct allowedips_node *__rcu *parent_bit;
 	union {
 		struct list_head peer_list;
 		struct rcu_head rcu;
@@ -53,4 +53,7 @@ struct wg_peer *wg_allowedips_lookup_src(struct allowedips *table,
 bool wg_allowedips_selftest(void);
 #endif
 
+int wg_allowedips_slab_init(void);
+void wg_allowedips_slab_uninit(void);
+
 #endif /* _WG_ALLOWEDIPS_H */
diff --git a/src/main.c b/src/main.c
index 3b2ce04..11c4df2 100644
--- a/src/main.c
+++ b/src/main.c
@@ -26,10 +26,15 @@ static int __init mod_init(void)
 	    (ret = curve25519_mod_init()))
 		return ret;
 
+	ret = wg_allowedips_slab_init();
+	if (ret < 0)
+		goto err_allowedips;
+
 #ifdef DEBUG
+	ret = -ENOTRECOVERABLE;
 	if (!wg_allowedips_selftest() || !wg_packet_counter_selftest() ||
 	    !wg_ratelimiter_selftest())
-		return -ENOTRECOVERABLE;
+		goto err_peer;
 #endif
 	wg_noise_init();
 
@@ -55,6 +60,8 @@ err_netlink:
 err_device:
 	wg_peer_uninit();
 err_peer:
+	wg_allowedips_slab_uninit();
+err_allowedips:
 	return ret;
 }
 
@@ -63,6 +70,7 @@ static void __exit mod_exit(void)
 	wg_genetlink_uninit();
 	wg_device_uninit();
 	wg_peer_uninit();
+	wg_allowedips_slab_uninit();
 }
 
 module_init(mod_init);