[linux.git] / net / netfilter / xt_cluster.c

/*
 * (C) 2008-2009 Pablo Neira Ayuso <[email protected]>
 *
 * 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.
 */
#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 (((u64)hash * info->total_nodes) >> 32);
}

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, const struct xt_match_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 (ct == &nf_conntrack_untracked)
		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 bool 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) {
		printk(KERN_ERR "xt_cluster: you have exceeded the maximum "
				"number of cluster nodes (%u > %u)\n",
				info->total_nodes, XT_CLUSTER_NODES_MAX);
		return false;
	}
	if (info->node_mask >= (1ULL << info->total_nodes)) {
		printk(KERN_ERR "xt_cluster: this node mask cannot be "
				"higher than the total number of nodes\n");
		return false;
	}
	return true;
}

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 <[email protected]>");
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 <[email protected]>
	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	#include <linux/module.h>
	9	#include <linux/skbuff.h>
	10	#include <linux/jhash.h>
	11	#include <linux/ip.h>
	12	#include <net/ipv6.h>
	13
	14	#include <linux/netfilter/x_tables.h>
	15	#include <net/netfilter/nf_conntrack.h>
	16	#include <linux/netfilter/xt_cluster.h>
	17
f9ffc312	18	static inline u32 nf_ct_orig_ipv4_src(const struct nf_conn *ct)
0269ea49	19	{
f9ffc312	20	return (__force u32)ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.src.u3.ip;
0269ea49 PNA	21	}
0269ea49 PNA	22
f9ffc312	23	static inline const u32 nf_ct_orig_ipv6_src(const struct nf_conn ct)
0269ea49	24	{
f9ffc312	25	return (__force u32 *)ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.src.u3.ip6;
0269ea49 PNA	26	}
	27
	28	static inline u_int32_t
	29	xt_cluster_hash_ipv4(u_int32_t ip, const struct xt_cluster_match_info *info)
	30	{
	31	return jhash_1word(ip, info->hash_seed);
	32	}
	33
	34	static inline u_int32_t
	35	xt_cluster_hash_ipv6(const void ip, const struct xt_cluster_match_info info)
	36	{
	37	return jhash2(ip, NF_CT_TUPLE_L3SIZE / sizeof(__u32), info->hash_seed);
	38	}
	39
	40	static inline u_int32_t
	41	xt_cluster_hash(const struct nf_conn *ct,
	42	const struct xt_cluster_match_info *info)
	43	{
	44	u_int32_t hash = 0;
	45
	46	switch(nf_ct_l3num(ct)) {
	47	case AF_INET:
	48	hash = xt_cluster_hash_ipv4(nf_ct_orig_ipv4_src(ct), info);
	49	break;
	50	case AF_INET6:
	51	hash = xt_cluster_hash_ipv6(nf_ct_orig_ipv6_src(ct), info);
	52	break;
	53	default:
	54	WARN_ON(1);
	55	break;
	56	}
	57	return (((u64)hash * info->total_nodes) >> 32);
	58	}
	59
424b86a6 PNA	60	static inline bool
	61	xt_cluster_ipv6_is_multicast(const struct in6_addr *addr)
	62	{
	63	__be32 st = addr->s6_addr32[0];
	64	return ((st & htonl(0xFF000000)) == htonl(0xFF000000));
	65	}
	66
0269ea49 PNA	67	static inline bool
	68	xt_cluster_is_multicast_addr(const struct sk_buff *skb, u_int8_t family)
	69	{
	70	bool is_multicast = false;
	71
	72	switch(family) {
	73	case NFPROTO_IPV4:
	74	is_multicast = ipv4_is_multicast(ip_hdr(skb)->daddr);
	75	break;
	76	case NFPROTO_IPV6:
424b86a6 PNA	77	is_multicast =
424b86a6 PNA	78	xt_cluster_ipv6_is_multicast(&ipv6_hdr(skb)->daddr);
0269ea49 PNA	79	break;
	80	default:
	81	WARN_ON(1);
	82	break;
	83	}
	84	return is_multicast;
	85	}
	86
	87	static bool
	88	xt_cluster_mt(const struct sk_buff skb, const struct xt_match_param par)
	89	{
	90	struct sk_buff pskb = (struct sk_buff )skb;
	91	const struct xt_cluster_match_info *info = par->matchinfo;
	92	const struct nf_conn *ct;
	93	enum ip_conntrack_info ctinfo;
	94	unsigned long hash;
	95
	96	/* This match assumes that all nodes see the same packets. This can be
	97	* achieved if the switch that connects the cluster nodes support some
	98	* sort of 'port mirroring'. However, if your switch does not support
	99	* this, your cluster nodes can reply ARP request using a multicast MAC
	100	* address. Thus, your switch will flood the same packets to the
	101	* cluster nodes with the same multicast MAC address. Using a multicast
	102	* link address is a RFC 1812 (section 3.3.2) violation, but this works
	103	* fine in practise.
	104	*
	105	* Unfortunately, if you use the multicast MAC address, the link layer
	106	* sets skbuff's pkt_type to PACKET_MULTICAST, which is not accepted
	107	* by TCP and others for packets coming to this node. For that reason,
	108	* this match mangles skbuff's pkt_type if it detects a packet
	109	* addressed to a unicast address but using PACKET_MULTICAST. Yes, I
	110	* know, matches should not alter packets, but we are doing this here
	111	* because we would need to add a PKTTYPE target for this sole purpose.
	112	*/
	113	if (!xt_cluster_is_multicast_addr(skb, par->family) &&
	114	skb->pkt_type == PACKET_MULTICAST) {
	115	pskb->pkt_type = PACKET_HOST;
	116	}
	117
	118	ct = nf_ct_get(skb, &ctinfo);
	119	if (ct == NULL)
	120	return false;
	121
	122	if (ct == &nf_conntrack_untracked)
	123	return false;
	124
	125	if (ct->master)
	126	hash = xt_cluster_hash(ct->master, info);
	127	else
	128	hash = xt_cluster_hash(ct, info);
	129
	130	return !!((1 << hash) & info->node_mask) ^
	131	!!(info->flags & XT_CLUSTER_F_INV);
	132	}
	133
	134	static bool xt_cluster_mt_checkentry(const struct xt_mtchk_param *par)
	135	{
	136	struct xt_cluster_match_info *info = par->matchinfo;
	137
280f37af PNA	138	if (info->total_nodes > XT_CLUSTER_NODES_MAX) {
	139	printk(KERN_ERR "xt_cluster: you have exceeded the maximum "
	140	"number of cluster nodes (%u > %u)\n",
	141	info->total_nodes, XT_CLUSTER_NODES_MAX);
	142	return false;
	143	}
	144	if (info->node_mask >= (1ULL << info->total_nodes)) {
0269ea49 PNA	145	printk(KERN_ERR "xt_cluster: this node mask cannot be "
	146	"higher than the total number of nodes\n");
	147	return false;
	148	}
	149	return true;
	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),
	158	.me = THIS_MODULE,
	159	};
	160
	161	static int __init xt_cluster_mt_init(void)
	162	{
	163	return xt_register_match(&xt_cluster_match);
	164	}
	165
	166	static void __exit xt_cluster_mt_fini(void)
	167	{
	168	xt_unregister_match(&xt_cluster_match);
	169	}
	170
	171	MODULE_AUTHOR("Pablo Neira Ayuso <[email protected]>");
	172	MODULE_LICENSE("GPL");
	173	MODULE_DESCRIPTION("Xtables: hash-based cluster match");
	174	MODULE_ALIAS("ipt_cluster");
	175	MODULE_ALIAS("ip6t_cluster");
	176	module_init(xt_cluster_mt_init);
	177	module_exit(xt_cluster_mt_fini);