[linux-2.6-block.git] / net / netfilter / xt_cluster.c

/*
 * (C) 2008-2009 Pablo Neira Ayuso <pablo@netfilter.org>
 *
 * This program is free software; you can redistribute it and/or modify
 * it under the terms of the GNU General Public License version 2 as
 * published by the Free Software Foundation.
 */
#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
#include <linux/module.h>
#include <linux/skbuff.h>
#include <linux/jhash.h>
#include <linux/ip.h>
#include <net/ipv6.h>

#include <linux/netfilter/x_tables.h>
#include <net/netfilter/nf_conntrack.h>
#include <linux/netfilter/xt_cluster.h>

static inline u32 nf_ct_orig_ipv4_src(const struct nf_conn *ct)
{
	return (__force u32)ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.src.u3.ip;
}

static inline const u32 *nf_ct_orig_ipv6_src(const struct nf_conn *ct)
{
	return (__force u32 *)ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.src.u3.ip6;
}

static inline u_int32_t
xt_cluster_hash_ipv4(u_int32_t ip, const struct xt_cluster_match_info *info)
{
	return jhash_1word(ip, info->hash_seed);
}

static inline u_int32_t
xt_cluster_hash_ipv6(const void *ip, const struct xt_cluster_match_info *info)
{
	return jhash2(ip, NF_CT_TUPLE_L3SIZE / sizeof(__u32), info->hash_seed);
}

static inline u_int32_t
xt_cluster_hash(const struct nf_conn *ct,
		const struct xt_cluster_match_info *info)
{
	u_int32_t hash = 0;

	switch(nf_ct_l3num(ct)) {
	case AF_INET:
		hash = xt_cluster_hash_ipv4(nf_ct_orig_ipv4_src(ct), info);
		break;
	case AF_INET6:
		hash = xt_cluster_hash_ipv6(nf_ct_orig_ipv6_src(ct), info);
		break;
	default:
		WARN_ON(1);
		break;
	}

	return reciprocal_scale(hash, info->total_nodes);
}

static inline bool
xt_cluster_ipv6_is_multicast(const struct in6_addr *addr)
{
	__be32 st = addr->s6_addr32[0];
	return ((st & htonl(0xFF000000)) == htonl(0xFF000000));
}

static inline bool
xt_cluster_is_multicast_addr(const struct sk_buff *skb, u_int8_t family)
{
	bool is_multicast = false;

	switch(family) {
	case NFPROTO_IPV4:
		is_multicast = ipv4_is_multicast(ip_hdr(skb)->daddr);
		break;
	case NFPROTO_IPV6:
		is_multicast =
			xt_cluster_ipv6_is_multicast(&ipv6_hdr(skb)->daddr);
		break;
	default:
		WARN_ON(1);
		break;
	}
	return is_multicast;
}

static bool
xt_cluster_mt(const struct sk_buff *skb, struct xt_action_param *par)
{
	struct sk_buff *pskb = (struct sk_buff *)skb;
	const struct xt_cluster_match_info *info = par->matchinfo;
	const struct nf_conn *ct;
	enum ip_conntrack_info ctinfo;
	unsigned long hash;

	/* This match assumes that all nodes see the same packets. This can be
	 * achieved if the switch that connects the cluster nodes support some
	 * sort of 'port mirroring'. However, if your switch does not support
	 * this, your cluster nodes can reply ARP request using a multicast MAC
	 * address. Thus, your switch will flood the same packets to the
	 * cluster nodes with the same multicast MAC address. Using a multicast
	 * link address is a RFC 1812 (section 3.3.2) violation, but this works
	 * fine in practise.
	 *
	 * Unfortunately, if you use the multicast MAC address, the link layer
	 * sets skbuff's pkt_type to PACKET_MULTICAST, which is not accepted
	 * by TCP and others for packets coming to this node. For that reason,
	 * this match mangles skbuff's pkt_type if it detects a packet
	 * addressed to a unicast address but using PACKET_MULTICAST. Yes, I
	 * know, matches should not alter packets, but we are doing this here
	 * because we would need to add a PKTTYPE target for this sole purpose.
	 */
	if (!xt_cluster_is_multicast_addr(skb, par->family) &&
	    skb->pkt_type == PACKET_MULTICAST) {
	    	pskb->pkt_type = PACKET_HOST;
	}

	ct = nf_ct_get(skb, &ctinfo);
	if (ct == NULL)
		return false;

	if (nf_ct_is_untracked(ct))
		return false;

	if (ct->master)
		hash = xt_cluster_hash(ct->master, info);
	else
		hash = xt_cluster_hash(ct, info);

	return !!((1 << hash) & info->node_mask) ^
	       !!(info->flags & XT_CLUSTER_F_INV);
}

static int xt_cluster_mt_checkentry(const struct xt_mtchk_param *par)
{
	struct xt_cluster_match_info *info = par->matchinfo;

	if (info->total_nodes > XT_CLUSTER_NODES_MAX) {
		pr_info("you have exceeded the maximum "
			"number of cluster nodes (%u > %u)\n",
			info->total_nodes, XT_CLUSTER_NODES_MAX);
		return -EINVAL;
	}
	if (info->node_mask >= (1ULL << info->total_nodes)) {
		pr_info("this node mask cannot be "
			"higher than the total number of nodes\n");
		return -EDOM;
	}
	return 0;
}

static struct xt_match xt_cluster_match __read_mostly = {
	.name		= "cluster",
	.family		= NFPROTO_UNSPEC,
	.match		= xt_cluster_mt,
	.checkentry	= xt_cluster_mt_checkentry,
	.matchsize	= sizeof(struct xt_cluster_match_info),
	.me		= THIS_MODULE,
};

static int __init xt_cluster_mt_init(void)
{
	return xt_register_match(&xt_cluster_match);
}

static void __exit xt_cluster_mt_fini(void)
{
	xt_unregister_match(&xt_cluster_match);
}

MODULE_AUTHOR("Pablo Neira Ayuso <pablo@netfilter.org>");
MODULE_LICENSE("GPL");
MODULE_DESCRIPTION("Xtables: hash-based cluster match");
MODULE_ALIAS("ipt_cluster");
MODULE_ALIAS("ip6t_cluster");
module_init(xt_cluster_mt_init);
module_exit(xt_cluster_mt_fini);
Commit	Line	Data
0269ea49 PNA	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	*/
8bee4bad	8	#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
0269ea49 PNA	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
f9ffc312	19	static inline u32 nf_ct_orig_ipv4_src(const struct nf_conn *ct)
0269ea49	20	{
f9ffc312	21	return (__force u32)ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.src.u3.ip;
0269ea49 PNA	22	}
0269ea49 PNA	23
f9ffc312	24	static inline const u32 nf_ct_orig_ipv6_src(const struct nf_conn ct)
0269ea49	25	{
f9ffc312	26	return (__force u32 *)ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.src.u3.ip6;
0269ea49 PNA	27	}
	28
	29	static inline u_int32_t
	30	xt_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
	35	static inline u_int32_t
	36	xt_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
	41	static inline u_int32_t
	42	xt_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	}
8fc54f68 DB	58
8fc54f68 DB	59	return reciprocal_scale(hash, info->total_nodes);
0269ea49 PNA	60	}
0269ea49 PNA	61
424b86a6 PNA	62	static inline bool
	63	xt_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
0269ea49 PNA	69	static inline bool
	70	xt_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:
424b86a6 PNA	79	is_multicast =
424b86a6 PNA	80	xt_cluster_ipv6_is_multicast(&ipv6_hdr(skb)->daddr);
0269ea49 PNA	81	break;
	82	default:
	83	WARN_ON(1);
	84	break;
	85	}
	86	return is_multicast;
	87	}
	88
	89	static bool
62fc8051	90	xt_cluster_mt(const struct sk_buff skb, struct xt_action_param par)
0269ea49 PNA	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
5bfddbd4	124	if (nf_ct_is_untracked(ct))
0269ea49 PNA	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
b0f38452	136	static int xt_cluster_mt_checkentry(const struct xt_mtchk_param *par)
0269ea49 PNA	137	{
	138	struct xt_cluster_match_info *info = par->matchinfo;
	139
280f37af	140	if (info->total_nodes > XT_CLUSTER_NODES_MAX) {
8bee4bad JE	141	pr_info("you have exceeded the maximum "
	142	"number of cluster nodes (%u > %u)\n",
	143	info->total_nodes, XT_CLUSTER_NODES_MAX);
bd414ee6	144	return -EINVAL;
280f37af PNA	145	}
280f37af PNA	146	if (info->node_mask >= (1ULL << info->total_nodes)) {
8bee4bad JE	147	pr_info("this node mask cannot be "
8bee4bad JE	148	"higher than the total number of nodes\n");
4a5a5c73	149	return -EDOM;
0269ea49	150	}
bd414ee6	151	return 0;
0269ea49 PNA	152	}
	153
	154	static 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
	163	static int __init xt_cluster_mt_init(void)
	164	{
	165	return xt_register_match(&xt_cluster_match);
	166	}
	167
	168	static void __exit xt_cluster_mt_fini(void)
	169	{
	170	xt_unregister_match(&xt_cluster_match);
	171	}
	172
	173	MODULE_AUTHOR("Pablo Neira Ayuso <pablo@netfilter.org>");
	174	MODULE_LICENSE("GPL");
	175	MODULE_DESCRIPTION("Xtables: hash-based cluster match");
	176	MODULE_ALIAS("ipt_cluster");
	177	MODULE_ALIAS("ip6t_cluster");
	178	module_init(xt_cluster_mt_init);
	179	module_exit(xt_cluster_mt_fini);