netfilter: nf_tables: coalesce multiple notifications into one skbuff

On x86_64, each notification results in one skbuff allocation which
consumes at least 768 bytes due to the skbuff overhead.

This patch coalesces several notifications into one single skbuff, so
each notification consumes at least ~211 bytes, that ~3.5 times less
memory consumption. As a result, this is reducing the chances to exhaust
the netlink socket receive buffer.

Rule of thumb is that each notification batch only contains netlink
messages whose report flag is the same, nfnetlink_send() requires this
to do appropriate delivery to userspace, either via unicast (echo
mode) or multicast (monitor mode).

The skbuff control buffer is used to annotate the report flag for later
handling at the new coalescing routine.

The batch skbuff notification size is NLMSG_GOODSIZE, using a larger
skbuff would allow for more socket receiver buffer savings (to amortize
the cost of the skbuff even more), however, going over that size might
break userspace applications, so let's be conservative and stick to
NLMSG_GOODSIZE.

Reported-by: Phil Sutter <[e-mail]>
Acked-by: Phil Sutter <[e-mail]>
Signed-off-by: Pablo Neira Ayuso <[e-mail]>

include/net/netns/nftables.h

···
 	struct list_head	tables;
 	struct list_head	commit_list;
 	struct list_head	module_list;
+	struct list_head	notify_list;
 	struct mutex		commit_mutex;
 	unsigned int		base_seq;
 	u8			gencursor;

net/netfilter/nf_tables_api.c

···
 	return -1;
 }
 
+struct nftnl_skb_parms {
+	bool report;
+};
+#define NFT_CB(skb)	(*(struct nftnl_skb_parms*)&((skb)->cb))
+
+static void nft_notify_enqueue(struct sk_buff *skb, bool report,
+			       struct list_head *notify_list)
+{
+	NFT_CB(skb).report = report;
+	list_add_tail(&skb->list, notify_list);
+}
+
 static void nf_tables_table_notify(const struct nft_ctx *ctx, int event)
 {
 	struct sk_buff *skb;
···
 		goto err;
 	}
 
-	nfnetlink_send(skb, ctx->net, ctx->portid, NFNLGRP_NFTABLES,
-		       ctx->report, GFP_KERNEL);
+	nft_notify_enqueue(skb, ctx->report, &ctx->net->nft.notify_list);
 	return;
 err:
 	nfnetlink_set_err(ctx->net, ctx->portid, NFNLGRP_NFTABLES, -ENOBUFS);
···
 		goto err;
 	}
 
-	nfnetlink_send(skb, ctx->net, ctx->portid, NFNLGRP_NFTABLES,
-		       ctx->report, GFP_KERNEL);
+	nft_notify_enqueue(skb, ctx->report, &ctx->net->nft.notify_list);
 	return;
 err:
 	nfnetlink_set_err(ctx->net, ctx->portid, NFNLGRP_NFTABLES, -ENOBUFS);
···
 		goto err;
 	}
 
-	nfnetlink_send(skb, ctx->net, ctx->portid, NFNLGRP_NFTABLES,
-		       ctx->report, GFP_KERNEL);
+	nft_notify_enqueue(skb, ctx->report, &ctx->net->nft.notify_list);
 	return;
 err:
 	nfnetlink_set_err(ctx->net, ctx->portid, NFNLGRP_NFTABLES, -ENOBUFS);
···
 		goto err;
 	}
 
-	nfnetlink_send(skb, ctx->net, portid, NFNLGRP_NFTABLES, ctx->report,
-		       gfp_flags);
+	nft_notify_enqueue(skb, ctx->report, &ctx->net->nft.notify_list);
 	return;
 err:
 	nfnetlink_set_err(ctx->net, portid, NFNLGRP_NFTABLES, -ENOBUFS);
···
 		goto err;
 	}
 
-	nfnetlink_send(skb, net, portid, NFNLGRP_NFTABLES, ctx->report,
-		       GFP_KERNEL);
+	nft_notify_enqueue(skb, ctx->report, &ctx->net->nft.notify_list);
 	return;
 err:
 	nfnetlink_set_err(net, portid, NFNLGRP_NFTABLES, -ENOBUFS);
···
 		goto err;
 	}
 
-	nfnetlink_send(skb, net, portid, NFNLGRP_NFTABLES, report, gfp);
+	nft_notify_enqueue(skb, report, &net->nft.notify_list);
 	return;
 err:
 	nfnetlink_set_err(net, portid, NFNLGRP_NFTABLES, -ENOBUFS);
···
 		goto err;
 	}
 
-	nfnetlink_send(skb, ctx->net, ctx->portid, NFNLGRP_NFTABLES,
-		       ctx->report, GFP_KERNEL);
+	nft_notify_enqueue(skb, ctx->report, &ctx->net->nft.notify_list);
 	return;
 err:
 	nfnetlink_set_err(ctx->net, ctx->portid, NFNLGRP_NFTABLES, -ENOBUFS);
···
 	mutex_unlock(&net->nft.commit_mutex);
 }
 
+static void nft_commit_notify(struct net *net, u32 portid)
+{
+	struct sk_buff *batch_skb = NULL, *nskb, *skb;
+	unsigned char *data;
+	int len;
+
+	list_for_each_entry_safe(skb, nskb, &net->nft.notify_list, list) {
+		if (!batch_skb) {
+new_batch:
+			batch_skb = skb;
+			len = NLMSG_GOODSIZE - skb->len;
+			list_del(&skb->list);
+			continue;
+		}
+		len -= skb->len;
+		if (len > 0 && NFT_CB(skb).report == NFT_CB(batch_skb).report) {
+			data = skb_put(batch_skb, skb->len);
+			memcpy(data, skb->data, skb->len);
+			list_del(&skb->list);
+			kfree_skb(skb);
+			continue;
+		}
+		nfnetlink_send(batch_skb, net, portid, NFNLGRP_NFTABLES,
+			       NFT_CB(batch_skb).report, GFP_KERNEL);
+		goto new_batch;
+	}
+
+	if (batch_skb) {
+		nfnetlink_send(batch_skb, net, portid, NFNLGRP_NFTABLES,
+			       NFT_CB(batch_skb).report, GFP_KERNEL);
+	}
+
+	WARN_ON_ONCE(!list_empty(&net->nft.notify_list));
+}
+
 static int nf_tables_commit(struct net *net, struct sk_buff *skb)
 {
 	struct nft_trans *trans, *next;
···
 		}
 	}
 
+	nft_commit_notify(net, NETLINK_CB(skb).portid);
 	nf_tables_gen_notify(net, skb, NFT_MSG_NEWGEN);
 	nf_tables_commit_release(net);
 
···
 	INIT_LIST_HEAD(&net->nft.tables);
 	INIT_LIST_HEAD(&net->nft.commit_list);
 	INIT_LIST_HEAD(&net->nft.module_list);
+	INIT_LIST_HEAD(&net->nft.notify_list);
 	mutex_init(&net->nft.commit_mutex);
 	net->nft.base_seq = 1;
 	net->nft.validate_state = NFT_VALIDATE_SKIP;
···
 	mutex_unlock(&net->nft.commit_mutex);
 	WARN_ON_ONCE(!list_empty(&net->nft.tables));
 	WARN_ON_ONCE(!list_empty(&net->nft.module_list));
+	WARN_ON_ONCE(!list_empty(&net->nft.notify_list));
 }
 
 static struct pernet_operations nf_tables_net_ops = {