net/netfilter/xt_cluster.c

   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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
   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
  19 static inline u32 nf_ct_orig_ipv4_src(const struct nf_conn *ct)
  20 {
  21         return (__force u32)ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.src.u3.ip;
  22 }
  23
  24 static inline const u32 *nf_ct_orig_ipv6_src(const struct nf_conn *ct)
  25 {
  26         return (__force u32 *)ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.src.u3.ip6;
  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         }
  58
  59         return reciprocal_scale(hash, info->total_nodes);
  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:
  72                 is_multicast = ipv6_addr_is_multicast(&ipv6_hdr(skb)->daddr);
  73                 break;
  74         default:
  75                 WARN_ON(1);
  76                 break;
  77         }
  78         return is_multicast;
  79 }
  80
  81 static bool
  82 xt_cluster_mt(const struct sk_buff *skb, struct xt_action_param *par)
  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          */
 107         if (!xt_cluster_is_multicast_addr(skb, xt_family(par)) &&
 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
 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
 125 static int xt_cluster_mt_checkentry(const struct xt_mtchk_param *par)
 126 {
 127         struct xt_cluster_match_info *info = par->matchinfo;
 128         int ret;
 129
 130         if (info->total_nodes > XT_CLUSTER_NODES_MAX) {
 131                 pr_info_ratelimited("you have exceeded the maximum number of cluster nodes (%u > %u)\n",
 132                                     info->total_nodes, XT_CLUSTER_NODES_MAX);
 133                 return -EINVAL;
 134         }
 135         if (info->node_mask >= (1ULL << info->total_nodes)) {
 136                 pr_info_ratelimited("node mask cannot exceed total number of nodes\n");
 137                 return -EDOM;
 138         }
 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);
 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         .destroy        = xt_cluster_mt_destroy,
 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);