net/bridge/netfilter/nft_reject_bridge.c

   1 /*
   2  * Copyright (c) 2014 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
   9 #include <linux/kernel.h>
  10 #include <linux/init.h>
  11 #include <linux/module.h>
  12 #include <linux/netlink.h>
  13 #include <linux/netfilter.h>
  14 #include <linux/netfilter/nf_tables.h>
  15 #include <net/netfilter/nf_tables.h>
  16 #include <net/netfilter/nft_reject.h>
  17 #include <net/netfilter/ipv4/nf_reject.h>
  18 #include <net/netfilter/ipv6/nf_reject.h>
  19 #include <linux/ip.h>
  20 #include <net/ip.h>
  21 #include <net/ip6_checksum.h>
  22 #include <linux/netfilter_bridge.h>
  23 #include <linux/netfilter_ipv6.h>
  24 #include "../br_private.h"
  25
  26 static void nft_reject_br_push_etherhdr(struct sk_buff *oldskb,
  27                                         struct sk_buff *nskb)
  28 {
  29         struct ethhdr *eth;
  30
  31         eth = (struct ethhdr *)skb_push(nskb, ETH_HLEN);
  32         skb_reset_mac_header(nskb);
  33         ether_addr_copy(eth->h_source, eth_hdr(oldskb)->h_dest);
  34         ether_addr_copy(eth->h_dest, eth_hdr(oldskb)->h_source);
  35         eth->h_proto = eth_hdr(oldskb)->h_proto;
  36         skb_pull(nskb, ETH_HLEN);
  37 }
  38
  39 static int nft_bridge_iphdr_validate(struct sk_buff *skb)
  40 {
  41         struct iphdr *iph;
  42         u32 len;
  43
  44         if (!pskb_may_pull(skb, sizeof(struct iphdr)))
  45                 return 0;
  46
  47         iph = ip_hdr(skb);
  48         if (iph->ihl < 5 || iph->version != 4)
  49                 return 0;
  50
  51         len = ntohs(iph->tot_len);
  52         if (skb->len < len)
  53                 return 0;
  54         else if (len < (iph->ihl*4))
  55                 return 0;
  56
  57         if (!pskb_may_pull(skb, iph->ihl*4))
  58                 return 0;
  59
  60         return 1;
  61 }
  62
  63 /* We cannot use oldskb->dev, it can be either bridge device (NF_BRIDGE INPUT)
  64  * or the bridge port (NF_BRIDGE PREROUTING).
  65  */
  66 static void nft_reject_br_send_v4_tcp_reset(struct net *net,
  67                                             struct sk_buff *oldskb,
  68                                             const struct net_device *dev,
  69                                             int hook)
  70 {
  71         struct sk_buff *nskb;
  72         struct iphdr *niph;
  73         const struct tcphdr *oth;
  74         struct tcphdr _oth;
  75
  76         if (!nft_bridge_iphdr_validate(oldskb))
  77                 return;
  78
  79         oth = nf_reject_ip_tcphdr_get(oldskb, &_oth, hook);
  80         if (!oth)
  81                 return;
  82
  83         nskb = alloc_skb(sizeof(struct iphdr) + sizeof(struct tcphdr) +
  84                          LL_MAX_HEADER, GFP_ATOMIC);
  85         if (!nskb)
  86                 return;
  87
  88         skb_reserve(nskb, LL_MAX_HEADER);
  89         niph = nf_reject_iphdr_put(nskb, oldskb, IPPROTO_TCP,
  90                                    net->ipv4.sysctl_ip_default_ttl);
  91         nf_reject_ip_tcphdr_put(nskb, oldskb, oth);
  92         niph->ttl       = net->ipv4.sysctl_ip_default_ttl;
  93         niph->tot_len   = htons(nskb->len);
  94         ip_send_check(niph);
  95
  96         nft_reject_br_push_etherhdr(oldskb, nskb);
  97
  98         br_forward(br_port_get_rcu(dev), nskb, false, true);
  99 }
 100
 101 static void nft_reject_br_send_v4_unreach(struct net *net,
 102                                           struct sk_buff *oldskb,
 103                                           const struct net_device *dev,
 104                                           int hook, u8 code)
 105 {
 106         struct sk_buff *nskb;
 107         struct iphdr *niph;
 108         struct icmphdr *icmph;
 109         unsigned int len;
 110         void *payload;
 111         __wsum csum;
 112         u8 proto;
 113
 114         if (oldskb->csum_bad || !nft_bridge_iphdr_validate(oldskb))
 115                 return;
 116
 117         /* IP header checks: fragment. */
 118         if (ip_hdr(oldskb)->frag_off & htons(IP_OFFSET))
 119                 return;
 120
 121         /* RFC says return as much as we can without exceeding 576 bytes. */
 122         len = min_t(unsigned int, 536, oldskb->len);
 123
 124         if (!pskb_may_pull(oldskb, len))
 125                 return;
 126
 127         if (pskb_trim_rcsum(oldskb, ntohs(ip_hdr(oldskb)->tot_len)))
 128                 return;
 129
 130         if (ip_hdr(oldskb)->protocol == IPPROTO_TCP ||
 131             ip_hdr(oldskb)->protocol == IPPROTO_UDP)
 132                 proto = ip_hdr(oldskb)->protocol;
 133         else
 134                 proto = 0;
 135
 136         if (!skb_csum_unnecessary(oldskb) &&
 137             nf_ip_checksum(oldskb, hook, ip_hdrlen(oldskb), proto))
 138                 return;
 139
 140         nskb = alloc_skb(sizeof(struct iphdr) + sizeof(struct icmphdr) +
 141                          LL_MAX_HEADER + len, GFP_ATOMIC);
 142         if (!nskb)
 143                 return;
 144
 145         skb_reserve(nskb, LL_MAX_HEADER);
 146         niph = nf_reject_iphdr_put(nskb, oldskb, IPPROTO_ICMP,
 147                                    net->ipv4.sysctl_ip_default_ttl);
 148
 149         skb_reset_transport_header(nskb);
 150         icmph = (struct icmphdr *)skb_put(nskb, sizeof(struct icmphdr));
 151         memset(icmph, 0, sizeof(*icmph));
 152         icmph->type     = ICMP_DEST_UNREACH;
 153         icmph->code     = code;
 154
 155         payload = skb_put(nskb, len);
 156         memcpy(payload, skb_network_header(oldskb), len);
 157
 158         csum = csum_partial((void *)icmph, len + sizeof(struct icmphdr), 0);
 159         icmph->checksum = csum_fold(csum);
 160
 161         niph->tot_len   = htons(nskb->len);
 162         ip_send_check(niph);
 163
 164         nft_reject_br_push_etherhdr(oldskb, nskb);
 165
 166         br_forward(br_port_get_rcu(dev), nskb, false, true);
 167 }
 168
 169 static int nft_bridge_ip6hdr_validate(struct sk_buff *skb)
 170 {
 171         struct ipv6hdr *hdr;
 172         u32 pkt_len;
 173
 174         if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
 175                 return 0;
 176
 177         hdr = ipv6_hdr(skb);
 178         if (hdr->version != 6)
 179                 return 0;
 180
 181         pkt_len = ntohs(hdr->payload_len);
 182         if (pkt_len + sizeof(struct ipv6hdr) > skb->len)
 183                 return 0;
 184
 185         return 1;
 186 }
 187
 188 static void nft_reject_br_send_v6_tcp_reset(struct net *net,
 189                                             struct sk_buff *oldskb,
 190                                             const struct net_device *dev,
 191                                             int hook)
 192 {
 193         struct sk_buff *nskb;
 194         const struct tcphdr *oth;
 195         struct tcphdr _oth;
 196         unsigned int otcplen;
 197         struct ipv6hdr *nip6h;
 198
 199         if (!nft_bridge_ip6hdr_validate(oldskb))
 200                 return;
 201
 202         oth = nf_reject_ip6_tcphdr_get(oldskb, &_oth, &otcplen, hook);
 203         if (!oth)
 204                 return;
 205
 206         nskb = alloc_skb(sizeof(struct ipv6hdr) + sizeof(struct tcphdr) +
 207                          LL_MAX_HEADER, GFP_ATOMIC);
 208         if (!nskb)
 209                 return;
 210
 211         skb_reserve(nskb, LL_MAX_HEADER);
 212         nip6h = nf_reject_ip6hdr_put(nskb, oldskb, IPPROTO_TCP,
 213                                      net->ipv6.devconf_all->hop_limit);
 214         nf_reject_ip6_tcphdr_put(nskb, oldskb, oth, otcplen);
 215         nip6h->payload_len = htons(nskb->len - sizeof(struct ipv6hdr));
 216
 217         nft_reject_br_push_etherhdr(oldskb, nskb);
 218
 219         br_forward(br_port_get_rcu(dev), nskb, false, true);
 220 }
 221
 222 static bool reject6_br_csum_ok(struct sk_buff *skb, int hook)
 223 {
 224         const struct ipv6hdr *ip6h = ipv6_hdr(skb);
 225         int thoff;
 226         __be16 fo;
 227         u8 proto = ip6h->nexthdr;
 228
 229         if (skb->csum_bad)
 230                 return false;
 231
 232         if (skb_csum_unnecessary(skb))
 233                 return true;
 234
 235         if (ip6h->payload_len &&
 236             pskb_trim_rcsum(skb, ntohs(ip6h->payload_len) + sizeof(*ip6h)))
 237                 return false;
 238
 239         thoff = ipv6_skip_exthdr(skb, ((u8*)(ip6h+1) - skb->data), &proto, &fo);
 240         if (thoff < 0 || thoff >= skb->len || (fo & htons(~0x7)) != 0)
 241                 return false;
 242
 243         return nf_ip6_checksum(skb, hook, thoff, proto) == 0;
 244 }
 245
 246 static void nft_reject_br_send_v6_unreach(struct net *net,
 247                                           struct sk_buff *oldskb,
 248                                           const struct net_device *dev,
 249                                           int hook, u8 code)
 250 {
 251         struct sk_buff *nskb;
 252         struct ipv6hdr *nip6h;
 253         struct icmp6hdr *icmp6h;
 254         unsigned int len;
 255         void *payload;
 256
 257         if (!nft_bridge_ip6hdr_validate(oldskb))
 258                 return;
 259
 260         /* Include "As much of invoking packet as possible without the ICMPv6
 261          * packet exceeding the minimum IPv6 MTU" in the ICMP payload.
 262          */
 263         len = min_t(unsigned int, 1220, oldskb->len);
 264
 265         if (!pskb_may_pull(oldskb, len))
 266                 return;
 267
 268         if (!reject6_br_csum_ok(oldskb, hook))
 269                 return;
 270
 271         nskb = alloc_skb(sizeof(struct iphdr) + sizeof(struct icmp6hdr) +
 272                          LL_MAX_HEADER + len, GFP_ATOMIC);
 273         if (!nskb)
 274                 return;
 275
 276         skb_reserve(nskb, LL_MAX_HEADER);
 277         nip6h = nf_reject_ip6hdr_put(nskb, oldskb, IPPROTO_ICMPV6,
 278                                      net->ipv6.devconf_all->hop_limit);
 279
 280         skb_reset_transport_header(nskb);
 281         icmp6h = (struct icmp6hdr *)skb_put(nskb, sizeof(struct icmp6hdr));
 282         memset(icmp6h, 0, sizeof(*icmp6h));
 283         icmp6h->icmp6_type = ICMPV6_DEST_UNREACH;
 284         icmp6h->icmp6_code = code;
 285
 286         payload = skb_put(nskb, len);
 287         memcpy(payload, skb_network_header(oldskb), len);
 288         nip6h->payload_len = htons(nskb->len - sizeof(struct ipv6hdr));
 289
 290         icmp6h->icmp6_cksum =
 291                 csum_ipv6_magic(&nip6h->saddr, &nip6h->daddr,
 292                                 nskb->len - sizeof(struct ipv6hdr),
 293                                 IPPROTO_ICMPV6,
 294                                 csum_partial(icmp6h,
 295                                              nskb->len - sizeof(struct ipv6hdr),
 296                                              0));
 297
 298         nft_reject_br_push_etherhdr(oldskb, nskb);
 299
 300         br_forward(br_port_get_rcu(dev), nskb, false, true);
 301 }
 302
 303 static void nft_reject_bridge_eval(const struct nft_expr *expr,
 304                                    struct nft_regs *regs,
 305                                    const struct nft_pktinfo *pkt)
 306 {
 307         struct nft_reject *priv = nft_expr_priv(expr);
 308         const unsigned char *dest = eth_hdr(pkt->skb)->h_dest;
 309
 310         if (is_broadcast_ether_addr(dest) ||
 311             is_multicast_ether_addr(dest))
 312                 goto out;
 313
 314         switch (eth_hdr(pkt->skb)->h_proto) {
 315         case htons(ETH_P_IP):
 316                 switch (priv->type) {
 317                 case NFT_REJECT_ICMP_UNREACH:
 318                         nft_reject_br_send_v4_unreach(pkt->net, pkt->skb,
 319                                                       pkt->in, pkt->hook,
 320                                                       priv->icmp_code);
 321                         break;
 322                 case NFT_REJECT_TCP_RST:
 323                         nft_reject_br_send_v4_tcp_reset(pkt->net, pkt->skb,
 324                                                         pkt->in, pkt->hook);
 325                         break;
 326                 case NFT_REJECT_ICMPX_UNREACH:
 327                         nft_reject_br_send_v4_unreach(pkt->net, pkt->skb,
 328                                                       pkt->in, pkt->hook,
 329                                                       nft_reject_icmp_code(priv->icmp_code));
 330                         break;
 331                 }
 332                 break;
 333         case htons(ETH_P_IPV6):
 334                 switch (priv->type) {
 335                 case NFT_REJECT_ICMP_UNREACH:
 336                         nft_reject_br_send_v6_unreach(pkt->net, pkt->skb,
 337                                                       pkt->in, pkt->hook,
 338                                                       priv->icmp_code);
 339                         break;
 340                 case NFT_REJECT_TCP_RST:
 341                         nft_reject_br_send_v6_tcp_reset(pkt->net, pkt->skb,
 342                                                         pkt->in, pkt->hook);
 343                         break;
 344                 case NFT_REJECT_ICMPX_UNREACH:
 345                         nft_reject_br_send_v6_unreach(pkt->net, pkt->skb,
 346                                                       pkt->in, pkt->hook,
 347                                                       nft_reject_icmpv6_code(priv->icmp_code));
 348                         break;
 349                 }
 350                 break;
 351         default:
 352                 /* No explicit way to reject this protocol, drop it. */
 353                 break;
 354         }
 355 out:
 356         regs->verdict.code = NF_DROP;
 357 }
 358
 359 static int nft_reject_bridge_validate(const struct nft_ctx *ctx,
 360                                       const struct nft_expr *expr,
 361                                       const struct nft_data **data)
 362 {
 363         return nft_chain_validate_hooks(ctx->chain, (1 << NF_BR_PRE_ROUTING) |
 364                                                     (1 << NF_BR_LOCAL_IN));
 365 }
 366
 367 static int nft_reject_bridge_init(const struct nft_ctx *ctx,
 368                                   const struct nft_expr *expr,
 369                                   const struct nlattr * const tb[])
 370 {
 371         struct nft_reject *priv = nft_expr_priv(expr);
 372         int icmp_code, err;
 373
 374         err = nft_reject_bridge_validate(ctx, expr, NULL);
 375         if (err < 0)
 376                 return err;
 377
 378         if (tb[NFTA_REJECT_TYPE] == NULL)
 379                 return -EINVAL;
 380
 381         priv->type = ntohl(nla_get_be32(tb[NFTA_REJECT_TYPE]));
 382         switch (priv->type) {
 383         case NFT_REJECT_ICMP_UNREACH:
 384         case NFT_REJECT_ICMPX_UNREACH:
 385                 if (tb[NFTA_REJECT_ICMP_CODE] == NULL)
 386                         return -EINVAL;
 387
 388                 icmp_code = nla_get_u8(tb[NFTA_REJECT_ICMP_CODE]);
 389                 if (priv->type == NFT_REJECT_ICMPX_UNREACH &&
 390                     icmp_code > NFT_REJECT_ICMPX_MAX)
 391                         return -EINVAL;
 392
 393                 priv->icmp_code = icmp_code;
 394                 break;
 395         case NFT_REJECT_TCP_RST:
 396                 break;
 397         default:
 398                 return -EINVAL;
 399         }
 400         return 0;
 401 }
 402
 403 static int nft_reject_bridge_dump(struct sk_buff *skb,
 404                                   const struct nft_expr *expr)
 405 {
 406         const struct nft_reject *priv = nft_expr_priv(expr);
 407
 408         if (nla_put_be32(skb, NFTA_REJECT_TYPE, htonl(priv->type)))
 409                 goto nla_put_failure;
 410
 411         switch (priv->type) {
 412         case NFT_REJECT_ICMP_UNREACH:
 413         case NFT_REJECT_ICMPX_UNREACH:
 414                 if (nla_put_u8(skb, NFTA_REJECT_ICMP_CODE, priv->icmp_code))
 415                         goto nla_put_failure;
 416                 break;
 417         default:
 418                 break;
 419         }
 420
 421         return 0;
 422
 423 nla_put_failure:
 424         return -1;
 425 }
 426
 427 static struct nft_expr_type nft_reject_bridge_type;
 428 static const struct nft_expr_ops nft_reject_bridge_ops = {
 429         .type           = &nft_reject_bridge_type,
 430         .size           = NFT_EXPR_SIZE(sizeof(struct nft_reject)),
 431         .eval           = nft_reject_bridge_eval,
 432         .init           = nft_reject_bridge_init,
 433         .dump           = nft_reject_bridge_dump,
 434         .validate       = nft_reject_bridge_validate,
 435 };
 436
 437 static struct nft_expr_type nft_reject_bridge_type __read_mostly = {
 438         .family         = NFPROTO_BRIDGE,
 439         .name           = "reject",
 440         .ops            = &nft_reject_bridge_ops,
 441         .policy         = nft_reject_policy,
 442         .maxattr        = NFTA_REJECT_MAX,
 443         .owner          = THIS_MODULE,
 444 };
 445
 446 static int __init nft_reject_bridge_module_init(void)
 447 {
 448         return nft_register_expr(&nft_reject_bridge_type);
 449 }
 450
 451 static void __exit nft_reject_bridge_module_exit(void)
 452 {
 453         nft_unregister_expr(&nft_reject_bridge_type);
 454 }
 455
 456 module_init(nft_reject_bridge_module_init);
 457 module_exit(nft_reject_bridge_module_exit);
 458
 459 MODULE_LICENSE("GPL");
 460 MODULE_AUTHOR("Pablo Neira Ayuso <pablo@netfilter.org>");
 461 MODULE_ALIAS_NFT_AF_EXPR(AF_BRIDGE, "reject");