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