[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_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 = ipv6_addr_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, xt_family(par)) &&
	    skb->pkt_type == PACKET_MULTICAST) {
	    	pskb->pkt_type = PACKET_HOST;
	}

	ct = nf_ct_get(skb, &ctinfo);
	if (ct == NULL)
		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;
	int ret;

	if (info->total_nodes > XT_CLUSTER_NODES_MAX) {
		pr_info_ratelimited("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_ratelimited("node mask cannot exceed total number of nodes\n");
		return -EDOM;
	}

	ret = nf_ct_netns_get(par->net, par->family);
	if (ret < 0)
		pr_info_ratelimited("cannot load conntrack support for proto=%u\n",
				    par->family);
	return ret;
}

static void xt_cluster_mt_destroy(const struct xt_mtdtor_param *par)
{
	nf_ct_netns_put(par->net, par->family);
}

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),
	.destroy	= xt_cluster_mt_destroy,
	.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	}
	61
	62	static inline bool
	63	xt_cluster_is_multicast_addr(const struct sk_buff *skb, u_int8_t family)
	64	{
	65	bool is_multicast = false;
	66
	67	switch(family) {
	68	case NFPROTO_IPV4:
	69	is_multicast = ipv4_is_multicast(ip_hdr(skb)->daddr);
	70	break;
	71	case NFPROTO_IPV6:
580c7d9e	72	is_multicast = ipv6_addr_is_multicast(&ipv6_hdr(skb)->daddr);
0269ea49 PNA	73	break;
	74	default:
	75	WARN_ON(1);
	76	break;
	77	}
	78	return is_multicast;
	79	}
	80
	81	static bool
62fc8051	82	xt_cluster_mt(const struct sk_buff skb, struct xt_action_param par)
0269ea49 PNA	83	{
	84	struct sk_buff pskb = (struct sk_buff )skb;
	85	const struct xt_cluster_match_info *info = par->matchinfo;
	86	const struct nf_conn *ct;
	87	enum ip_conntrack_info ctinfo;
	88	unsigned long hash;
	89
	90	/* This match assumes that all nodes see the same packets. This can be
	91	* achieved if the switch that connects the cluster nodes support some
	92	* sort of 'port mirroring'. However, if your switch does not support
	93	* this, your cluster nodes can reply ARP request using a multicast MAC
	94	* address. Thus, your switch will flood the same packets to the
	95	* cluster nodes with the same multicast MAC address. Using a multicast
	96	* link address is a RFC 1812 (section 3.3.2) violation, but this works
	97	* fine in practise.
	98	*
	99	* Unfortunately, if you use the multicast MAC address, the link layer
	100	* sets skbuff's pkt_type to PACKET_MULTICAST, which is not accepted
	101	* by TCP and others for packets coming to this node. For that reason,
	102	* this match mangles skbuff's pkt_type if it detects a packet
	103	* addressed to a unicast address but using PACKET_MULTICAST. Yes, I
	104	* know, matches should not alter packets, but we are doing this here
	105	* because we would need to add a PKTTYPE target for this sole purpose.
	106	*/
613dbd95	107	if (!xt_cluster_is_multicast_addr(skb, xt_family(par)) &&
0269ea49 PNA	108	skb->pkt_type == PACKET_MULTICAST) {
	109	pskb->pkt_type = PACKET_HOST;
	110	}
	111
	112	ct = nf_ct_get(skb, &ctinfo);
	113	if (ct == NULL)
	114	return false;
	115
0269ea49 PNA	116	if (ct->master)
	117	hash = xt_cluster_hash(ct->master, info);
	118	else
	119	hash = xt_cluster_hash(ct, info);
	120
	121	return !!((1 << hash) & info->node_mask) ^
	122	!!(info->flags & XT_CLUSTER_F_INV);
	123	}
	124
b0f38452	125	static int xt_cluster_mt_checkentry(const struct xt_mtchk_param *par)
0269ea49 PNA	126	{
0269ea49 PNA	127	struct xt_cluster_match_info *info = par->matchinfo;
c1dc2912	128	int ret;
0269ea49	129
280f37af	130	if (info->total_nodes > XT_CLUSTER_NODES_MAX) {
b2606644 FW	131	pr_info_ratelimited("you have exceeded the maximum number of cluster nodes (%u > %u)\n",
b2606644 FW	132	info->total_nodes, XT_CLUSTER_NODES_MAX);
bd414ee6	133	return -EINVAL;
280f37af PNA	134	}
280f37af PNA	135	if (info->node_mask >= (1ULL << info->total_nodes)) {
b2606644	136	pr_info_ratelimited("node mask cannot exceed total number of nodes\n");
4a5a5c73	137	return -EDOM;
0269ea49	138	}
c1dc2912 MW	139
	140	ret = nf_ct_netns_get(par->net, par->family);
	141	if (ret < 0)
	142	pr_info_ratelimited("cannot load conntrack support for proto=%u\n",
	143	par->family);
	144	return ret;
	145	}
	146
	147	static void xt_cluster_mt_destroy(const struct xt_mtdtor_param *par)
	148	{
	149	nf_ct_netns_put(par->net, par->family);
0269ea49 PNA	150	}
	151
	152	static struct xt_match xt_cluster_match __read_mostly = {
	153	.name = "cluster",
	154	.family = NFPROTO_UNSPEC,
	155	.match = xt_cluster_mt,
	156	.checkentry = xt_cluster_mt_checkentry,
	157	.matchsize = sizeof(struct xt_cluster_match_info),
c1dc2912	158	.destroy = xt_cluster_mt_destroy,
0269ea49 PNA	159	.me = THIS_MODULE,
	160	};
	161
	162	static int __init xt_cluster_mt_init(void)
	163	{
	164	return xt_register_match(&xt_cluster_match);
	165	}
	166
	167	static void __exit xt_cluster_mt_fini(void)
	168	{
	169	xt_unregister_match(&xt_cluster_match);
	170	}
	171
	172	MODULE_AUTHOR("Pablo Neira Ayuso <pablo@netfilter.org>");
	173	MODULE_LICENSE("GPL");
	174	MODULE_DESCRIPTION("Xtables: hash-based cluster match");
	175	MODULE_ALIAS("ipt_cluster");
	176	MODULE_ALIAS("ip6t_cluster");
	177	module_init(xt_cluster_mt_init);
	178	module_exit(xt_cluster_mt_fini);