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         if (skb_vlan_tag_present(oldskb)) {
  39                 u16 vid = skb_vlan_tag_get(oldskb);
  40
  41                 __vlan_hwaccel_put_tag(nskb, oldskb->vlan_proto, vid);
  42         }
  43 }
  44
  45 static int nft_bridge_iphdr_validate(struct sk_buff *skb)
  46 {
  47         struct iphdr *iph;
  48         u32 len;
  49
  50         if (!pskb_may_pull(skb, sizeof(struct iphdr)))
  51                 return 0;
  52
  53         iph = ip_hdr(skb);
  54         if (iph->ihl < 5 || iph->version != 4)
  55                 return 0;
  56
  57         len = ntohs(iph->tot_len);
  58         if (skb->len < len)
  59                 return 0;
  60         else if (len < (iph->ihl*4))
  61                 return 0;
  62
  63         if (!pskb_may_pull(skb, iph->ihl*4))
  64                 return 0;
  65
  66         return 1;
  67 }
  68
  69 /* We cannot use oldskb->dev, it can be either bridge device (NF_BRIDGE INPUT)
  70  * or the bridge port (NF_BRIDGE PREROUTING).
  71  */
  72 static void nft_reject_br_send_v4_tcp_reset(struct net *net,
  73                                             struct sk_buff *oldskb,
  74                                             const struct net_device *dev,
  75                                             int hook)
  76 {
  77         struct sk_buff *nskb;
  78         struct iphdr *niph;
  79         const struct tcphdr *oth;
  80         struct tcphdr _oth;
  81
  82         if (!nft_bridge_iphdr_validate(oldskb))
  83                 return;
  84
  85         oth = nf_reject_ip_tcphdr_get(oldskb, &_oth, hook);
  86         if (!oth)
  87                 return;
  88
  89         nskb = alloc_skb(sizeof(struct iphdr) + sizeof(struct tcphdr) +
  90                          LL_MAX_HEADER, GFP_ATOMIC);
  91         if (!nskb)
  92                 return;
  93
  94         skb_reserve(nskb, LL_MAX_HEADER);
  95         niph = nf_reject_iphdr_put(nskb, oldskb, IPPROTO_TCP,
  96                                    net->ipv4.sysctl_ip_default_ttl);
  97         nf_reject_ip_tcphdr_put(nskb, oldskb, oth);
  98         niph->ttl       = net->ipv4.sysctl_ip_default_ttl;
  99         niph->tot_len   = htons(nskb->len);
 100         ip_send_check(niph);
 101
 102         nft_reject_br_push_etherhdr(oldskb, nskb);
 103
 104         br_forward(br_port_get_rcu(dev), nskb, false, true);
 105 }
 106
 107 static void nft_reject_br_send_v4_unreach(struct net *net,
 108                                           struct sk_buff *oldskb,
 109                                           const struct net_device *dev,
 110                                           int hook, u8 code)
 111 {
 112         struct sk_buff *nskb;
 113         struct iphdr *niph;
 114         struct icmphdr *icmph;
 115         unsigned int len;
 116         void *payload;
 117         __wsum csum;
 118         u8 proto;
 119
 120         if (oldskb->csum_bad || !nft_bridge_iphdr_validate(oldskb))
 121                 return;
 122
 123         /* IP header checks: fragment. */
 124         if (ip_hdr(oldskb)->frag_off & htons(IP_OFFSET))
 125                 return;
 126
 127         /* RFC says return as much as we can without exceeding 576 bytes. */
 128         len = min_t(unsigned int, 536, oldskb->len);
 129
 130         if (!pskb_may_pull(oldskb, len))
 131                 return;
 132
 133         if (pskb_trim_rcsum(oldskb, ntohs(ip_hdr(oldskb)->tot_len)))
 134                 return;
 135
 136         if (ip_hdr(oldskb)->protocol == IPPROTO_TCP ||
 137             ip_hdr(oldskb)->protocol == IPPROTO_UDP)
 138                 proto = ip_hdr(oldskb)->protocol;
 139         else
 140                 proto = 0;
 141
 142         if (!skb_csum_unnecessary(oldskb) &&
 143             nf_ip_checksum(oldskb, hook, ip_hdrlen(oldskb), proto))
 144                 return;
 145
 146         nskb = alloc_skb(sizeof(struct iphdr) + sizeof(struct icmphdr) +
 147                          LL_MAX_HEADER + len, GFP_ATOMIC);
 148         if (!nskb)
 149                 return;
 150
 151         skb_reserve(nskb, LL_MAX_HEADER);
 152         niph = nf_reject_iphdr_put(nskb, oldskb, IPPROTO_ICMP,
 153                                    net->ipv4.sysctl_ip_default_ttl);
 154
 155         skb_reset_transport_header(nskb);
 156         icmph = (struct icmphdr *)skb_put(nskb, sizeof(struct icmphdr));
 157         memset(icmph, 0, sizeof(*icmph));
 158         icmph->type     = ICMP_DEST_UNREACH;
 159         icmph->code     = code;
 160
 161         payload = skb_put(nskb, len);
 162         memcpy(payload, skb_network_header(oldskb), len);
 163
 164         csum = csum_partial((void *)icmph, len + sizeof(struct icmphdr), 0);
 165         icmph->checksum = csum_fold(csum);
 166
 167         niph->tot_len   = htons(nskb->len);
 168         ip_send_check(niph);
 169
 170         nft_reject_br_push_etherhdr(oldskb, nskb);
 171
 172         br_forward(br_port_get_rcu(dev), nskb, false, true);
 173 }
 174
 175 static int nft_bridge_ip6hdr_validate(struct sk_buff *skb)
 176 {
 177         struct ipv6hdr *hdr;
 178         u32 pkt_len;
 179
 180         if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
 181                 return 0;
 182
 183         hdr = ipv6_hdr(skb);
 184         if (hdr->version != 6)
 185                 return 0;
 186
 187         pkt_len = ntohs(hdr->payload_len);
 188         if (pkt_len + sizeof(struct ipv6hdr) > skb->len)
 189                 return 0;
 190
 191         return 1;
 192 }
 193
 194 static void nft_reject_br_send_v6_tcp_reset(struct net *net,
 195                                             struct sk_buff *oldskb,
 196                                             const struct net_device *dev,
 197                                             int hook)
 198 {
 199         struct sk_buff *nskb;
 200         const struct tcphdr *oth;
 201         struct tcphdr _oth;
 202         unsigned int otcplen;
 203         struct ipv6hdr *nip6h;
 204
 205         if (!nft_bridge_ip6hdr_validate(oldskb))
 206                 return;
 207
 208         oth = nf_reject_ip6_tcphdr_get(oldskb, &_oth, &otcplen, hook);
 209         if (!oth)
 210                 return;
 211
 212         nskb = alloc_skb(sizeof(struct ipv6hdr) + sizeof(struct tcphdr) +
 213                          LL_MAX_HEADER, GFP_ATOMIC);
 214         if (!nskb)
 215                 return;
 216
 217         skb_reserve(nskb, LL_MAX_HEADER);
 218         nip6h = nf_reject_ip6hdr_put(nskb, oldskb, IPPROTO_TCP,
 219                                      net->ipv6.devconf_all->hop_limit);
 220         nf_reject_ip6_tcphdr_put(nskb, oldskb, oth, otcplen);
 221         nip6h->payload_len = htons(nskb->len - sizeof(struct ipv6hdr));
 222
 223         nft_reject_br_push_etherhdr(oldskb, nskb);
 224
 225         br_forward(br_port_get_rcu(dev), nskb, false, true);
 226 }
 227
 228 static bool reject6_br_csum_ok(struct sk_buff *skb, int hook)
 229 {
 230         const struct ipv6hdr *ip6h = ipv6_hdr(skb);
 231         int thoff;
 232         __be16 fo;
 233         u8 proto = ip6h->nexthdr;
 234
 235         if (skb->csum_bad)
 236                 return false;
 237
 238         if (skb_csum_unnecessary(skb))
 239                 return true;
 240
 241         if (ip6h->payload_len &&
 242             pskb_trim_rcsum(skb, ntohs(ip6h->payload_len) + sizeof(*ip6h)))
 243                 return false;
 244
 245         ip6h = ipv6_hdr(skb);
 246         thoff = ipv6_skip_exthdr(skb, ((u8*)(ip6h+1) - skb->data), &proto, &fo);
 247         if (thoff < 0 || thoff >= skb->len || (fo & htons(~0x7)) != 0)
 248                 return false;
 249
 250         return nf_ip6_checksum(skb, hook, thoff, proto) == 0;
 251 }
 252
 253 static void nft_reject_br_send_v6_unreach(struct net *net,
 254                                           struct sk_buff *oldskb,
 255                                           const struct net_device *dev,
 256                                           int hook, u8 code)
 257 {
 258         struct sk_buff *nskb;
 259         struct ipv6hdr *nip6h;
 260         struct icmp6hdr *icmp6h;
 261         unsigned int len;
 262         void *payload;
 263
 264         if (!nft_bridge_ip6hdr_validate(oldskb))
 265                 return;
 266
 267         /* Include "As much of invoking packet as possible without the ICMPv6
 268          * packet exceeding the minimum IPv6 MTU" in the ICMP payload.
 269          */
 270         len = min_t(unsigned int, 1220, oldskb->len);
 271
 272         if (!pskb_may_pull(oldskb, len))
 273                 return;
 274
 275         if (!reject6_br_csum_ok(oldskb, hook))
 276                 return;
 277
 278         nskb = alloc_skb(sizeof(struct iphdr) + sizeof(struct icmp6hdr) +
 279                          LL_MAX_HEADER + len, GFP_ATOMIC);
 280         if (!nskb)
 281                 return;
 282
 283         skb_reserve(nskb, LL_MAX_HEADER);
 284         nip6h = nf_reject_ip6hdr_put(nskb, oldskb, IPPROTO_ICMPV6,
 285                                      net->ipv6.devconf_all->hop_limit);
 286
 287         skb_reset_transport_header(nskb);
 288         icmp6h = (struct icmp6hdr *)skb_put(nskb, sizeof(struct icmp6hdr));
 289         memset(icmp6h, 0, sizeof(*icmp6h));
 290         icmp6h->icmp6_type = ICMPV6_DEST_UNREACH;
 291         icmp6h->icmp6_code = code;
 292
 293         payload = skb_put(nskb, len);
 294         memcpy(payload, skb_network_header(oldskb), len);
 295         nip6h->payload_len = htons(nskb->len - sizeof(struct ipv6hdr));
 296
 297         icmp6h->icmp6_cksum =
 298                 csum_ipv6_magic(&nip6h->saddr, &nip6h->daddr,
 299                                 nskb->len - sizeof(struct ipv6hdr),
 300                                 IPPROTO_ICMPV6,
 301                                 csum_partial(icmp6h,
 302                                              nskb->len - sizeof(struct ipv6hdr),
 303                                              0));
 304
 305         nft_reject_br_push_etherhdr(oldskb, nskb);
 306
 307         br_forward(br_port_get_rcu(dev), nskb, false, true);
 308 }
 309
 310 static void nft_reject_bridge_eval(const struct nft_expr *expr,
 311                                    struct nft_regs *regs,
 312                                    const struct nft_pktinfo *pkt)
 313 {
 314         struct nft_reject *priv = nft_expr_priv(expr);
 315         const unsigned char *dest = eth_hdr(pkt->skb)->h_dest;
 316
 317         if (is_broadcast_ether_addr(dest) ||
 318             is_multicast_ether_addr(dest))
 319                 goto out;
 320
 321         switch (eth_hdr(pkt->skb)->h_proto) {
 322         case htons(ETH_P_IP):
 323                 switch (priv->type) {
 324                 case NFT_REJECT_ICMP_UNREACH:
 325                         nft_reject_br_send_v4_unreach(pkt->net, pkt->skb,
 326                                                       pkt->in, pkt->hook,
 327                                                       priv->icmp_code);
 328                         break;
 329                 case NFT_REJECT_TCP_RST:
 330                         nft_reject_br_send_v4_tcp_reset(pkt->net, pkt->skb,
 331                                                         pkt->in, pkt->hook);
 332                         break;
 333                 case NFT_REJECT_ICMPX_UNREACH:
 334                         nft_reject_br_send_v4_unreach(pkt->net, pkt->skb,
 335                                                       pkt->in, pkt->hook,
 336                                                       nft_reject_icmp_code(priv->icmp_code));
 337                         break;
 338                 }
 339                 break;
 340         case htons(ETH_P_IPV6):
 341                 switch (priv->type) {
 342                 case NFT_REJECT_ICMP_UNREACH:
 343                         nft_reject_br_send_v6_unreach(pkt->net, pkt->skb,
 344                                                       pkt->in, pkt->hook,
 345                                                       priv->icmp_code);
 346                         break;
 347                 case NFT_REJECT_TCP_RST:
 348                         nft_reject_br_send_v6_tcp_reset(pkt->net, pkt->skb,
 349                                                         pkt->in, pkt->hook);
 350                         break;
 351                 case NFT_REJECT_ICMPX_UNREACH:
 352                         nft_reject_br_send_v6_unreach(pkt->net, pkt->skb,
 353                                                       pkt->in, pkt->hook,
 354                                                       nft_reject_icmpv6_code(priv->icmp_code));
 355                         break;
 356                 }
 357                 break;
 358         default:
 359                 /* No explicit way to reject this protocol, drop it. */
 360                 break;
 361         }
 362 out:
 363         regs->verdict.code = NF_DROP;
 364 }
 365
 366 static int nft_reject_bridge_validate(const struct nft_ctx *ctx,
 367                                       const struct nft_expr *expr,
 368                                       const struct nft_data **data)
 369 {
 370         return nft_chain_validate_hooks(ctx->chain, (1 << NF_BR_PRE_ROUTING) |
 371                                                     (1 << NF_BR_LOCAL_IN));
 372 }
 373
 374 static int nft_reject_bridge_init(const struct nft_ctx *ctx,
 375                                   const struct nft_expr *expr,
 376                                   const struct nlattr * const tb[])
 377 {
 378         struct nft_reject *priv = nft_expr_priv(expr);
 379         int icmp_code, err;
 380
 381         err = nft_reject_bridge_validate(ctx, expr, NULL);
 382         if (err < 0)
 383                 return err;
 384
 385         if (tb[NFTA_REJECT_TYPE] == NULL)
 386                 return -EINVAL;
 387
 388         priv->type = ntohl(nla_get_be32(tb[NFTA_REJECT_TYPE]));
 389         switch (priv->type) {
 390         case NFT_REJECT_ICMP_UNREACH:
 391         case NFT_REJECT_ICMPX_UNREACH:
 392                 if (tb[NFTA_REJECT_ICMP_CODE] == NULL)
 393                         return -EINVAL;
 394
 395                 icmp_code = nla_get_u8(tb[NFTA_REJECT_ICMP_CODE]);
 396                 if (priv->type == NFT_REJECT_ICMPX_UNREACH &&
 397                     icmp_code > NFT_REJECT_ICMPX_MAX)
 398                         return -EINVAL;
 399
 400                 priv->icmp_code = icmp_code;
 401                 break;
 402         case NFT_REJECT_TCP_RST:
 403                 break;
 404         default:
 405                 return -EINVAL;
 406         }
 407         return 0;
 408 }
 409
 410 static int nft_reject_bridge_dump(struct sk_buff *skb,
 411                                   const struct nft_expr *expr)
 412 {
 413         const struct nft_reject *priv = nft_expr_priv(expr);
 414
 415         if (nla_put_be32(skb, NFTA_REJECT_TYPE, htonl(priv->type)))
 416                 goto nla_put_failure;
 417
 418         switch (priv->type) {
 419         case NFT_REJECT_ICMP_UNREACH:
 420         case NFT_REJECT_ICMPX_UNREACH:
 421                 if (nla_put_u8(skb, NFTA_REJECT_ICMP_CODE, priv->icmp_code))
 422                         goto nla_put_failure;
 423                 break;
 424         default:
 425                 break;
 426         }
 427
 428         return 0;
 429
 430 nla_put_failure:
 431         return -1;
 432 }
 433
 434 static struct nft_expr_type nft_reject_bridge_type;
 435 static const struct nft_expr_ops nft_reject_bridge_ops = {
 436         .type           = &nft_reject_bridge_type,
 437         .size           = NFT_EXPR_SIZE(sizeof(struct nft_reject)),
 438         .eval           = nft_reject_bridge_eval,
 439         .init           = nft_reject_bridge_init,
 440         .dump           = nft_reject_bridge_dump,
 441         .validate       = nft_reject_bridge_validate,
 442 };
 443
 444 static struct nft_expr_type nft_reject_bridge_type __read_mostly = {
 445         .family         = NFPROTO_BRIDGE,
 446         .name           = "reject",
 447         .ops            = &nft_reject_bridge_ops,
 448         .policy         = nft_reject_policy,
 449         .maxattr        = NFTA_REJECT_MAX,
 450         .owner          = THIS_MODULE,
 451 };
 452
 453 static int __init nft_reject_bridge_module_init(void)
 454 {
 455         return nft_register_expr(&nft_reject_bridge_type);
 456 }
 457
 458 static void __exit nft_reject_bridge_module_exit(void)
 459 {
 460         nft_unregister_expr(&nft_reject_bridge_type);
 461 }
 462
 463 module_init(nft_reject_bridge_module_init);
 464 module_exit(nft_reject_bridge_module_exit);
 465
 466 MODULE_LICENSE("GPL");
 467 MODULE_AUTHOR("Pablo Neira Ayuso <pablo@netfilter.org>");
 468 MODULE_ALIAS_NFT_AF_EXPR(AF_BRIDGE, "reject");