target/linux/generic/backport-4.14/323-v4.16-netfilter-flow-table-support-for-IPv4.patch

   1 From: Pablo Neira Ayuso <pablo@netfilter.org>
   2 Date: Sun, 7 Jan 2018 01:04:15 +0100
   3 Subject: [PATCH] netfilter: flow table support for IPv4
   4
   5 This patch adds the IPv4 flow table type, that implements the datapath
   6 flow table to forward IPv4 traffic. Rationale is:
   7
   8 1) Look up for the packet in the flow table, from the ingress hook.
   9 2) If there's a hit, decrement ttl and pass it on to the neighbour layer
  10    for transmission.
  11 3) If there's a miss, packet is passed up to the classic forwarding
  12    path.
  13
  14 This patch also supports layer 3 source and destination NAT.
  15
  16 Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
  17 ---
  18  create mode 100644 net/ipv4/netfilter/nf_flow_table_ipv4.c
  19
  20 --- a/net/ipv4/netfilter/Kconfig
  21 +++ b/net/ipv4/netfilter/Kconfig
  22 @@ -77,6 +77,14 @@ config NF_TABLES_ARP
  23
  24  endif # NF_TABLES
  25
  26 +config NF_FLOW_TABLE_IPV4
  27 +       select NF_FLOW_TABLE
  28 +       tristate "Netfilter flow table IPv4 module"
  29 +       help
  30 +         This option adds the flow table IPv4 support.
  31 +
  32 +         To compile it as a module, choose M here.
  33 +
  34  config NF_DUP_IPV4
  35         tristate "Netfilter IPv4 packet duplication to alternate destination"
  36         depends on !NF_CONNTRACK || NF_CONNTRACK
  37 --- a/net/ipv4/netfilter/Makefile
  38 +++ b/net/ipv4/netfilter/Makefile
  39 @@ -43,6 +43,9 @@ obj-$(CONFIG_NFT_REDIR_IPV4) += nft_redi
  40  obj-$(CONFIG_NFT_DUP_IPV4) += nft_dup_ipv4.o
  41  obj-$(CONFIG_NF_TABLES_ARP) += nf_tables_arp.o
  42
  43 +# flow table support
  44 +obj-$(CONFIG_NF_FLOW_TABLE_IPV4) += nf_flow_table_ipv4.o
  45 +
  46  # generic IP tables
  47  obj-$(CONFIG_IP_NF_IPTABLES) += ip_tables.o
  48
  49 --- /dev/null
  50 +++ b/net/ipv4/netfilter/nf_flow_table_ipv4.c
  51 @@ -0,0 +1,283 @@
  52 +#include <linux/kernel.h>
  53 +#include <linux/init.h>
  54 +#include <linux/module.h>
  55 +#include <linux/netfilter.h>
  56 +#include <linux/rhashtable.h>
  57 +#include <linux/ip.h>
  58 +#include <linux/netdevice.h>
  59 +#include <net/ip.h>
  60 +#include <net/neighbour.h>
  61 +#include <net/netfilter/nf_flow_table.h>
  62 +#include <net/netfilter/nf_tables.h>
  63 +/* For layer 4 checksum field offset. */
  64 +#include <linux/tcp.h>
  65 +#include <linux/udp.h>
  66 +
  67 +static int nf_flow_nat_ip_tcp(struct sk_buff *skb, unsigned int thoff,
  68 +                             __be32 addr, __be32 new_addr)
  69 +{
  70 +       struct tcphdr *tcph;
  71 +
  72 +       if (!pskb_may_pull(skb, thoff + sizeof(*tcph)) ||
  73 +           skb_try_make_writable(skb, thoff + sizeof(*tcph)))
  74 +               return -1;
  75 +
  76 +       tcph = (void *)(skb_network_header(skb) + thoff);
  77 +       inet_proto_csum_replace4(&tcph->check, skb, addr, new_addr, true);
  78 +
  79 +       return 0;
  80 +}
  81 +
  82 +static int nf_flow_nat_ip_udp(struct sk_buff *skb, unsigned int thoff,
  83 +                             __be32 addr, __be32 new_addr)
  84 +{
  85 +       struct udphdr *udph;
  86 +
  87 +       if (!pskb_may_pull(skb, thoff + sizeof(*udph)) ||
  88 +           skb_try_make_writable(skb, thoff + sizeof(*udph)))
  89 +               return -1;
  90 +
  91 +       udph = (void *)(skb_network_header(skb) + thoff);
  92 +       if (udph->check || skb->ip_summed == CHECKSUM_PARTIAL) {
  93 +               inet_proto_csum_replace4(&udph->check, skb, addr,
  94 +                                        new_addr, true);
  95 +               if (!udph->check)
  96 +                       udph->check = CSUM_MANGLED_0;
  97 +       }
  98 +
  99 +       return 0;
 100 +}
 101 +
 102 +static int nf_flow_nat_ip_l4proto(struct sk_buff *skb, struct iphdr *iph,
 103 +                                 unsigned int thoff, __be32 addr,
 104 +                                 __be32 new_addr)
 105 +{
 106 +       switch (iph->protocol) {
 107 +       case IPPROTO_TCP:
 108 +               if (nf_flow_nat_ip_tcp(skb, thoff, addr, new_addr) < 0)
 109 +                       return NF_DROP;
 110 +               break;
 111 +       case IPPROTO_UDP:
 112 +               if (nf_flow_nat_ip_udp(skb, thoff, addr, new_addr) < 0)
 113 +                       return NF_DROP;
 114 +               break;
 115 +       }
 116 +
 117 +       return 0;
 118 +}
 119 +
 120 +static int nf_flow_snat_ip(const struct flow_offload *flow, struct sk_buff *skb,
 121 +                          struct iphdr *iph, unsigned int thoff,
 122 +                          enum flow_offload_tuple_dir dir)
 123 +{
 124 +       __be32 addr, new_addr;
 125 +
 126 +       switch (dir) {
 127 +       case FLOW_OFFLOAD_DIR_ORIGINAL:
 128 +               addr = iph->saddr;
 129 +               new_addr = flow->tuplehash[FLOW_OFFLOAD_DIR_REPLY].tuple.dst_v4.s_addr;
 130 +               iph->saddr = new_addr;
 131 +               break;
 132 +       case FLOW_OFFLOAD_DIR_REPLY:
 133 +               addr = iph->daddr;
 134 +               new_addr = flow->tuplehash[FLOW_OFFLOAD_DIR_ORIGINAL].tuple.src_v4.s_addr;
 135 +               iph->daddr = new_addr;
 136 +               break;
 137 +       default:
 138 +               return -1;
 139 +       }
 140 +       csum_replace4(&iph->check, addr, new_addr);
 141 +
 142 +       return nf_flow_nat_ip_l4proto(skb, iph, thoff, addr, new_addr);
 143 +}
 144 +
 145 +static int nf_flow_dnat_ip(const struct flow_offload *flow, struct sk_buff *skb,
 146 +                          struct iphdr *iph, unsigned int thoff,
 147 +                          enum flow_offload_tuple_dir dir)
 148 +{
 149 +       __be32 addr, new_addr;
 150 +
 151 +       switch (dir) {
 152 +       case FLOW_OFFLOAD_DIR_ORIGINAL:
 153 +               addr = iph->daddr;
 154 +               new_addr = flow->tuplehash[FLOW_OFFLOAD_DIR_REPLY].tuple.src_v4.s_addr;
 155 +               iph->daddr = new_addr;
 156 +               break;
 157 +       case FLOW_OFFLOAD_DIR_REPLY:
 158 +               addr = iph->saddr;
 159 +               new_addr = flow->tuplehash[FLOW_OFFLOAD_DIR_ORIGINAL].tuple.dst_v4.s_addr;
 160 +               iph->saddr = new_addr;
 161 +               break;
 162 +       default:
 163 +               return -1;
 164 +       }
 165 +
 166 +       return nf_flow_nat_ip_l4proto(skb, iph, thoff, addr, new_addr);
 167 +}
 168 +
 169 +static int nf_flow_nat_ip(const struct flow_offload *flow, struct sk_buff *skb,
 170 +                         enum flow_offload_tuple_dir dir)
 171 +{
 172 +       struct iphdr *iph = ip_hdr(skb);
 173 +       unsigned int thoff = iph->ihl * 4;
 174 +
 175 +       if (flow->flags & FLOW_OFFLOAD_SNAT &&
 176 +           (nf_flow_snat_port(flow, skb, thoff, iph->protocol, dir) < 0 ||
 177 +            nf_flow_snat_ip(flow, skb, iph, thoff, dir) < 0))
 178 +               return -1;
 179 +       if (flow->flags & FLOW_OFFLOAD_DNAT &&
 180 +           (nf_flow_dnat_port(flow, skb, thoff, iph->protocol, dir) < 0 ||
 181 +            nf_flow_dnat_ip(flow, skb, iph, thoff, dir) < 0))
 182 +               return -1;
 183 +
 184 +       return 0;
 185 +}
 186 +
 187 +static bool ip_has_options(unsigned int thoff)
 188 +{
 189 +       return thoff != sizeof(struct iphdr);
 190 +}
 191 +
 192 +static int nf_flow_tuple_ip(struct sk_buff *skb, const struct net_device *dev,
 193 +                           struct flow_offload_tuple *tuple)
 194 +{
 195 +       struct flow_ports *ports;
 196 +       unsigned int thoff;
 197 +       struct iphdr *iph;
 198 +
 199 +       if (!pskb_may_pull(skb, sizeof(*iph)))
 200 +               return -1;
 201 +
 202 +       iph = ip_hdr(skb);
 203 +       thoff = iph->ihl * 4;
 204 +
 205 +       if (ip_is_fragment(iph) ||
 206 +           unlikely(ip_has_options(thoff)))
 207 +               return -1;
 208 +
 209 +       if (iph->protocol != IPPROTO_TCP &&
 210 +           iph->protocol != IPPROTO_UDP)
 211 +               return -1;
 212 +
 213 +       thoff = iph->ihl * 4;
 214 +       if (!pskb_may_pull(skb, thoff + sizeof(*ports)))
 215 +               return -1;
 216 +
 217 +       ports = (struct flow_ports *)(skb_network_header(skb) + thoff);
 218 +
 219 +       tuple->src_v4.s_addr    = iph->saddr;
 220 +       tuple->dst_v4.s_addr    = iph->daddr;
 221 +       tuple->src_port         = ports->source;
 222 +       tuple->dst_port         = ports->dest;
 223 +       tuple->l3proto          = AF_INET;
 224 +       tuple->l4proto          = iph->protocol;
 225 +       tuple->iifidx           = dev->ifindex;
 226 +
 227 +       return 0;
 228 +}
 229 +
 230 +/* Based on ip_exceeds_mtu(). */
 231 +static bool __nf_flow_exceeds_mtu(const struct sk_buff *skb, unsigned int mtu)
 232 +{
 233 +       if (skb->len <= mtu)
 234 +               return false;
 235 +
 236 +       if ((ip_hdr(skb)->frag_off & htons(IP_DF)) == 0)
 237 +               return false;
 238 +
 239 +       if (skb_is_gso(skb) && skb_gso_validate_mtu(skb, mtu))
 240 +               return false;
 241 +
 242 +       return true;
 243 +}
 244 +
 245 +static bool nf_flow_exceeds_mtu(struct sk_buff *skb, const struct rtable *rt)
 246 +{
 247 +       u32 mtu;
 248 +
 249 +       mtu = ip_dst_mtu_maybe_forward(&rt->dst, true);
 250 +       if (__nf_flow_exceeds_mtu(skb, mtu))
 251 +               return true;
 252 +
 253 +       return false;
 254 +}
 255 +
 256 +static unsigned int
 257 +nf_flow_offload_ip_hook(void *priv, struct sk_buff *skb,
 258 +                       const struct nf_hook_state *state)
 259 +{
 260 +       struct flow_offload_tuple_rhash *tuplehash;
 261 +       struct nf_flowtable *flow_table = priv;
 262 +       struct flow_offload_tuple tuple = {};
 263 +       enum flow_offload_tuple_dir dir;
 264 +       struct flow_offload *flow;
 265 +       struct net_device *outdev;
 266 +       const struct rtable *rt;
 267 +       struct iphdr *iph;
 268 +       __be32 nexthop;
 269 +
 270 +       if (skb->protocol != htons(ETH_P_IP))
 271 +               return NF_ACCEPT;
 272 +
 273 +       if (nf_flow_tuple_ip(skb, state->in, &tuple) < 0)
 274 +               return NF_ACCEPT;
 275 +
 276 +       tuplehash = flow_offload_lookup(flow_table, &tuple);
 277 +       if (tuplehash == NULL)
 278 +               return NF_ACCEPT;
 279 +
 280 +       outdev = dev_get_by_index_rcu(state->net, tuplehash->tuple.oifidx);
 281 +       if (!outdev)
 282 +               return NF_ACCEPT;
 283 +
 284 +       dir = tuplehash->tuple.dir;
 285 +       flow = container_of(tuplehash, struct flow_offload, tuplehash[dir]);
 286 +
 287 +       rt = (const struct rtable *)flow->tuplehash[dir].tuple.dst_cache;
 288 +       if (unlikely(nf_flow_exceeds_mtu(skb, rt)))
 289 +               return NF_ACCEPT;
 290 +
 291 +       if (skb_try_make_writable(skb, sizeof(*iph)))
 292 +               return NF_DROP;
 293 +
 294 +       if (flow->flags & (FLOW_OFFLOAD_SNAT | FLOW_OFFLOAD_DNAT) &&
 295 +           nf_flow_nat_ip(flow, skb, dir) < 0)
 296 +               return NF_DROP;
 297 +
 298 +       flow->timeout = (u32)jiffies + NF_FLOW_TIMEOUT;
 299 +       iph = ip_hdr(skb);
 300 +       ip_decrease_ttl(iph);
 301 +
 302 +       skb->dev = outdev;
 303 +       nexthop = rt_nexthop(rt, flow->tuplehash[!dir].tuple.src_v4.s_addr);
 304 +       neigh_xmit(NEIGH_ARP_TABLE, outdev, &nexthop, skb);
 305 +
 306 +       return NF_STOLEN;
 307 +}
 308 +
 309 +static struct nf_flowtable_type flowtable_ipv4 = {
 310 +       .family         = NFPROTO_IPV4,
 311 +       .params         = &nf_flow_offload_rhash_params,
 312 +       .gc             = nf_flow_offload_work_gc,
 313 +       .hook           = nf_flow_offload_ip_hook,
 314 +       .owner          = THIS_MODULE,
 315 +};
 316 +
 317 +static int __init nf_flow_ipv4_module_init(void)
 318 +{
 319 +       nft_register_flowtable_type(&flowtable_ipv4);
 320 +
 321 +       return 0;
 322 +}
 323 +
 324 +static void __exit nf_flow_ipv4_module_exit(void)
 325 +{
 326 +       nft_unregister_flowtable_type(&flowtable_ipv4);
 327 +}
 328 +
 329 +module_init(nf_flow_ipv4_module_init);
 330 +module_exit(nf_flow_ipv4_module_exit);
 331 +
 332 +MODULE_LICENSE("GPL");
 333 +MODULE_AUTHOR("Pablo Neira Ayuso <pablo@netfilter.org>");
 334 +MODULE_ALIAS_NF_FLOWTABLE(AF_INET);