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

// SPDX-License-Identifier: GPL-2.0-only
/*
 * (C) 2008-2009 Pablo Neira Ayuso <pablo@netfilter.org>
 */
#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
d2912cb1	1	// SPDX-License-Identifier: GPL-2.0-only
0269ea49 PNA	2	/*
0269ea49 PNA	3	* (C) 2008-2009 Pablo Neira Ayuso <pablo@netfilter.org>
0269ea49	4	*/
8bee4bad	5	#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
0269ea49 PNA	6	#include <linux/module.h>
	7	#include <linux/skbuff.h>
	8	#include <linux/jhash.h>
	9	#include <linux/ip.h>
	10	#include <net/ipv6.h>
	11
	12	#include <linux/netfilter/x_tables.h>
	13	#include <net/netfilter/nf_conntrack.h>
	14	#include <linux/netfilter/xt_cluster.h>
	15
f9ffc312	16	static inline u32 nf_ct_orig_ipv4_src(const struct nf_conn *ct)
0269ea49	17	{
f9ffc312	18	return (__force u32)ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.src.u3.ip;
0269ea49 PNA	19	}
0269ea49 PNA	20
f9ffc312	21	static inline const u32 nf_ct_orig_ipv6_src(const struct nf_conn ct)
0269ea49	22	{
f9ffc312	23	return (__force u32 *)ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.src.u3.ip6;
0269ea49 PNA	24	}
	25
	26	static inline u_int32_t
	27	xt_cluster_hash_ipv4(u_int32_t ip, const struct xt_cluster_match_info *info)
	28	{
	29	return jhash_1word(ip, info->hash_seed);
	30	}
	31
	32	static inline u_int32_t
	33	xt_cluster_hash_ipv6(const void ip, const struct xt_cluster_match_info info)
	34	{
	35	return jhash2(ip, NF_CT_TUPLE_L3SIZE / sizeof(__u32), info->hash_seed);
	36	}
	37
	38	static inline u_int32_t
	39	xt_cluster_hash(const struct nf_conn *ct,
	40	const struct xt_cluster_match_info *info)
	41	{
	42	u_int32_t hash = 0;
	43
	44	switch(nf_ct_l3num(ct)) {
	45	case AF_INET:
	46	hash = xt_cluster_hash_ipv4(nf_ct_orig_ipv4_src(ct), info);
	47	break;
	48	case AF_INET6:
	49	hash = xt_cluster_hash_ipv6(nf_ct_orig_ipv6_src(ct), info);
	50	break;
	51	default:
	52	WARN_ON(1);
	53	break;
	54	}
8fc54f68 DB	55
8fc54f68 DB	56	return reciprocal_scale(hash, info->total_nodes);
0269ea49 PNA	57	}
	58
	59	static inline bool
	60	xt_cluster_is_multicast_addr(const struct sk_buff *skb, u_int8_t family)
	61	{
	62	bool is_multicast = false;
	63
	64	switch(family) {
	65	case NFPROTO_IPV4:
	66	is_multicast = ipv4_is_multicast(ip_hdr(skb)->daddr);
	67	break;
	68	case NFPROTO_IPV6:
580c7d9e	69	is_multicast = ipv6_addr_is_multicast(&ipv6_hdr(skb)->daddr);
0269ea49 PNA	70	break;
	71	default:
	72	WARN_ON(1);
	73	break;
	74	}
	75	return is_multicast;
	76	}
	77
	78	static bool
62fc8051	79	xt_cluster_mt(const struct sk_buff skb, struct xt_action_param par)
0269ea49 PNA	80	{
	81	struct sk_buff pskb = (struct sk_buff )skb;
	82	const struct xt_cluster_match_info *info = par->matchinfo;
	83	const struct nf_conn *ct;
	84	enum ip_conntrack_info ctinfo;
	85	unsigned long hash;
	86
	87	/* This match assumes that all nodes see the same packets. This can be
	88	* achieved if the switch that connects the cluster nodes support some
	89	* sort of 'port mirroring'. However, if your switch does not support
	90	* this, your cluster nodes can reply ARP request using a multicast MAC
	91	* address. Thus, your switch will flood the same packets to the
	92	* cluster nodes with the same multicast MAC address. Using a multicast
	93	* link address is a RFC 1812 (section 3.3.2) violation, but this works
	94	* fine in practise.
	95	*
	96	* Unfortunately, if you use the multicast MAC address, the link layer
	97	* sets skbuff's pkt_type to PACKET_MULTICAST, which is not accepted
	98	* by TCP and others for packets coming to this node. For that reason,
	99	* this match mangles skbuff's pkt_type if it detects a packet
	100	* addressed to a unicast address but using PACKET_MULTICAST. Yes, I
	101	* know, matches should not alter packets, but we are doing this here
	102	* because we would need to add a PKTTYPE target for this sole purpose.
	103	*/
613dbd95	104	if (!xt_cluster_is_multicast_addr(skb, xt_family(par)) &&
0269ea49 PNA	105	skb->pkt_type == PACKET_MULTICAST) {
	106	pskb->pkt_type = PACKET_HOST;
	107	}
	108
	109	ct = nf_ct_get(skb, &ctinfo);
	110	if (ct == NULL)
	111	return false;
	112
0269ea49 PNA	113	if (ct->master)
	114	hash = xt_cluster_hash(ct->master, info);
	115	else
	116	hash = xt_cluster_hash(ct, info);
	117
	118	return !!((1 << hash) & info->node_mask) ^
	119	!!(info->flags & XT_CLUSTER_F_INV);
	120	}
	121
b0f38452	122	static int xt_cluster_mt_checkentry(const struct xt_mtchk_param *par)
0269ea49 PNA	123	{
0269ea49 PNA	124	struct xt_cluster_match_info *info = par->matchinfo;
c1dc2912	125	int ret;
0269ea49	126
280f37af	127	if (info->total_nodes > XT_CLUSTER_NODES_MAX) {
b2606644 FW	128	pr_info_ratelimited("you have exceeded the maximum number of cluster nodes (%u > %u)\n",
b2606644 FW	129	info->total_nodes, XT_CLUSTER_NODES_MAX);
bd414ee6	130	return -EINVAL;
280f37af PNA	131	}
280f37af PNA	132	if (info->node_mask >= (1ULL << info->total_nodes)) {
b2606644	133	pr_info_ratelimited("node mask cannot exceed total number of nodes\n");
4a5a5c73	134	return -EDOM;
0269ea49	135	}
c1dc2912 MW	136
	137	ret = nf_ct_netns_get(par->net, par->family);
	138	if (ret < 0)
	139	pr_info_ratelimited("cannot load conntrack support for proto=%u\n",
	140	par->family);
	141	return ret;
	142	}
	143
	144	static void xt_cluster_mt_destroy(const struct xt_mtdtor_param *par)
	145	{
	146	nf_ct_netns_put(par->net, par->family);
0269ea49 PNA	147	}
	148
	149	static struct xt_match xt_cluster_match __read_mostly = {
	150	.name = "cluster",
	151	.family = NFPROTO_UNSPEC,
	152	.match = xt_cluster_mt,
	153	.checkentry = xt_cluster_mt_checkentry,
	154	.matchsize = sizeof(struct xt_cluster_match_info),
c1dc2912	155	.destroy = xt_cluster_mt_destroy,
0269ea49 PNA	156	.me = THIS_MODULE,
	157	};
	158
	159	static int __init xt_cluster_mt_init(void)
	160	{
	161	return xt_register_match(&xt_cluster_match);
	162	}
	163
	164	static void __exit xt_cluster_mt_fini(void)
	165	{
	166	xt_unregister_match(&xt_cluster_match);
	167	}
	168
	169	MODULE_AUTHOR("Pablo Neira Ayuso <pablo@netfilter.org>");
	170	MODULE_LICENSE("GPL");
	171	MODULE_DESCRIPTION("Xtables: hash-based cluster match");
	172	MODULE_ALIAS("ipt_cluster");
	173	MODULE_ALIAS("ip6t_cluster");
	174	module_init(xt_cluster_mt_init);
	175	module_exit(xt_cluster_mt_fini);