linux/net/netfilter/xt_cluster.c
<<
>>
Prefs
   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#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
  18static inline u_int32_t nf_ct_orig_ipv4_src(const struct nf_conn *ct)
  19{
  20        return ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.src.u3.ip;
  21}
  22
  23static inline const void *nf_ct_orig_ipv6_src(const struct nf_conn *ct)
  24{
  25        return ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.src.u3.ip6;
  26}
  27
  28static inline u_int32_t
  29xt_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
  34static inline u_int32_t
  35xt_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
  40static inline u_int32_t
  41xt_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
  60static inline bool
  61xt_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
  67static inline bool
  68xt_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:
  77                is_multicast =
  78                        xt_cluster_ipv6_is_multicast(&ipv6_hdr(skb)->daddr);
  79                break;
  80        default:
  81                WARN_ON(1);
  82                break;
  83        }
  84        return is_multicast;
  85}
  86
  87static bool
  88xt_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
 134static bool xt_cluster_mt_checkentry(const struct xt_mtchk_param *par)
 135{
 136        struct xt_cluster_match_info *info = par->matchinfo;
 137
 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)) {
 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
 152static 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
 161static int __init xt_cluster_mt_init(void)
 162{
 163        return xt_register_match(&xt_cluster_match);
 164}
 165
 166static void __exit xt_cluster_mt_fini(void)
 167{
 168        xt_unregister_match(&xt_cluster_match);
 169}
 170
 171MODULE_AUTHOR("Pablo Neira Ayuso <pablo@netfilter.org>");
 172MODULE_LICENSE("GPL");
 173MODULE_DESCRIPTION("Xtables: hash-based cluster match");
 174MODULE_ALIAS("ipt_cluster");
 175MODULE_ALIAS("ip6t_cluster");
 176module_init(xt_cluster_mt_init);
 177module_exit(xt_cluster_mt_fini);
 178