1/*
2 * (C) 2008-2009 Pablo Neira Ayuso <pablo@netfilter.org>
3 *
4 * This program is free software; you can redistribute it and/or modify
5 * it under the terms of the GNU General Public License version 2 as
6 * published by the Free Software Foundation.
7 */
8#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
9#include <linux/module.h>
10#include <linux/skbuff.h>
11#include <linux/jhash.h>
12#include <linux/ip.h>
13#include <net/ipv6.h>
14
15#include <linux/netfilter/x_tables.h>
16#include <net/netfilter/nf_conntrack.h>
17#include <linux/netfilter/xt_cluster.h>
18
19static inline u32 nf_ct_orig_ipv4_src(const struct nf_conn *ct)
20{
21	return (__force u32)ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.src.u3.ip;
22}
23
24static inline const u32 *nf_ct_orig_ipv6_src(const struct nf_conn *ct)
25{
26	return (__force u32 *)ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.src.u3.ip6;
27}
28
29static inline u_int32_t
30xt_cluster_hash_ipv4(u_int32_t ip, const struct xt_cluster_match_info *info)
31{
32	return jhash_1word(ip, info->hash_seed);
33}
34
35static inline u_int32_t
36xt_cluster_hash_ipv6(const void *ip, const struct xt_cluster_match_info *info)
37{
38	return jhash2(ip, NF_CT_TUPLE_L3SIZE / sizeof(__u32), info->hash_seed);
39}
40
41static inline u_int32_t
42xt_cluster_hash(const struct nf_conn *ct,
43		const struct xt_cluster_match_info *info)
44{
45	u_int32_t hash = 0;
46
47	switch(nf_ct_l3num(ct)) {
48	case AF_INET:
49		hash = xt_cluster_hash_ipv4(nf_ct_orig_ipv4_src(ct), info);
50		break;
51	case AF_INET6:
52		hash = xt_cluster_hash_ipv6(nf_ct_orig_ipv6_src(ct), info);
53		break;
54	default:
55		WARN_ON(1);
56		break;
57	}
58
59	return reciprocal_scale(hash, info->total_nodes);
60}
61
62static inline bool
63xt_cluster_ipv6_is_multicast(const struct in6_addr *addr)
64{
65	__be32 st = addr->s6_addr32[0];
66	return ((st & htonl(0xFF000000)) == htonl(0xFF000000));
67}
68
69static inline bool
70xt_cluster_is_multicast_addr(const struct sk_buff *skb, u_int8_t family)
71{
72	bool is_multicast = false;
73
74	switch(family) {
75	case NFPROTO_IPV4:
76		is_multicast = ipv4_is_multicast(ip_hdr(skb)->daddr);
77		break;
78	case NFPROTO_IPV6:
79		is_multicast =
80			xt_cluster_ipv6_is_multicast(&ipv6_hdr(skb)->daddr);
81		break;
82	default:
83		WARN_ON(1);
84		break;
85	}
86	return is_multicast;
87}
88
89static bool
90xt_cluster_mt(const struct sk_buff *skb, struct xt_action_param *par)
91{
92	struct sk_buff *pskb = (struct sk_buff *)skb;
93	const struct xt_cluster_match_info *info = par->matchinfo;
94	const struct nf_conn *ct;
95	enum ip_conntrack_info ctinfo;
96	unsigned long hash;
97
98	/* This match assumes that all nodes see the same packets. This can be
99	 * achieved if the switch that connects the cluster nodes support some
100	 * sort of 'port mirroring'. However, if your switch does not support
101	 * this, your cluster nodes can reply ARP request using a multicast MAC
102	 * address. Thus, your switch will flood the same packets to the
103	 * cluster nodes with the same multicast MAC address. Using a multicast
104	 * link address is a RFC 1812 (section 3.3.2) violation, but this works
105	 * fine in practise.
106	 *
107	 * Unfortunately, if you use the multicast MAC address, the link layer
108	 * sets skbuff's pkt_type to PACKET_MULTICAST, which is not accepted
109	 * by TCP and others for packets coming to this node. For that reason,
110	 * this match mangles skbuff's pkt_type if it detects a packet
111	 * addressed to a unicast address but using PACKET_MULTICAST. Yes, I
112	 * know, matches should not alter packets, but we are doing this here
113	 * because we would need to add a PKTTYPE target for this sole purpose.
114	 */
115	if (!xt_cluster_is_multicast_addr(skb, par->family) &&
116	    skb->pkt_type == PACKET_MULTICAST) {
117	    	pskb->pkt_type = PACKET_HOST;
118	}
119
120	ct = nf_ct_get(skb, &ctinfo);
121	if (ct == NULL)
122		return false;
123
124	if (nf_ct_is_untracked(ct))
125		return false;
126
127	if (ct->master)
128		hash = xt_cluster_hash(ct->master, info);
129	else
130		hash = xt_cluster_hash(ct, info);
131
132	return !!((1 << hash) & info->node_mask) ^
133	       !!(info->flags & XT_CLUSTER_F_INV);
134}
135
136static int xt_cluster_mt_checkentry(const struct xt_mtchk_param *par)
137{
138	struct xt_cluster_match_info *info = par->matchinfo;
139
140	if (info->total_nodes > XT_CLUSTER_NODES_MAX) {
141		pr_info("you have exceeded the maximum "
142			"number of cluster nodes (%u > %u)\n",
143			info->total_nodes, XT_CLUSTER_NODES_MAX);
144		return -EINVAL;
145	}
146	if (info->node_mask >= (1ULL << info->total_nodes)) {
147		pr_info("this node mask cannot be "
148			"higher than the total number of nodes\n");
149		return -EDOM;
150	}
151	return 0;
152}
153
154static struct xt_match xt_cluster_match __read_mostly = {
155	.name		= "cluster",
156	.family		= NFPROTO_UNSPEC,
157	.match		= xt_cluster_mt,
158	.checkentry	= xt_cluster_mt_checkentry,
159	.matchsize	= sizeof(struct xt_cluster_match_info),
160	.me		= THIS_MODULE,
161};
162
163static int __init xt_cluster_mt_init(void)
164{
165	return xt_register_match(&xt_cluster_match);
166}
167
168static void __exit xt_cluster_mt_fini(void)
169{
170	xt_unregister_match(&xt_cluster_match);
171}
172
173MODULE_AUTHOR("Pablo Neira Ayuso <pablo@netfilter.org>");
174MODULE_LICENSE("GPL");
175MODULE_DESCRIPTION("Xtables: hash-based cluster match");
176MODULE_ALIAS("ipt_cluster");
177MODULE_ALIAS("ip6t_cluster");
178module_init(xt_cluster_mt_init);
179module_exit(xt_cluster_mt_fini);
180