net/bridge/br_netfilter.c

   1 /*
   2  *      Handle firewalling
   3  *      Linux ethernet bridge
   4  *
   5  *      Authors:
   6  *      Lennert Buytenhek               <buytenh@gnu.org>
   7  *      Bart De Schuymer                <bdschuym@pandora.be>
   8  *
   9  *      This program is free software; you can redistribute it and/or
  10  *      modify it under the terms of the GNU General Public License
  11  *      as published by the Free Software Foundation; either version
  12  *      2 of the License, or (at your option) any later version.
  13  *
  14  *      Lennert dedicates this file to Kerstin Wurdinger.
  15  */
  16
  17 #include <linux/module.h>
  18 #include <linux/kernel.h>
  19 #include <linux/slab.h>
  20 #include <linux/ip.h>
  21 #include <linux/netdevice.h>
  22 #include <linux/skbuff.h>
  23 #include <linux/if_arp.h>
  24 #include <linux/if_ether.h>
  25 #include <linux/if_vlan.h>
  26 #include <linux/if_pppox.h>
  27 #include <linux/ppp_defs.h>
  28 #include <linux/netfilter_bridge.h>
  29 #include <linux/netfilter_ipv4.h>
  30 #include <linux/netfilter_ipv6.h>
  31 #include <linux/netfilter_arp.h>
  32 #include <linux/in_route.h>
  33 #include <linux/inetdevice.h>
  34
  35 #include <net/ip.h>
  36 #include <net/ipv6.h>
  37 #include <net/route.h>
  38 #include <net/netfilter/br_netfilter.h>
  39
  40 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
  41 #include <net/netfilter/nf_conntrack.h>
  42 #endif
  43
  44 #include <asm/uaccess.h>
  45 #include "br_private.h"
  46 #ifdef CONFIG_SYSCTL
  47 #include <linux/sysctl.h>
  48 #endif
  49
  50 #ifdef CONFIG_SYSCTL
  51 static struct ctl_table_header *brnf_sysctl_header;
  52 static int brnf_call_iptables __read_mostly = 1;
  53 static int brnf_call_ip6tables __read_mostly = 1;
  54 static int brnf_call_arptables __read_mostly = 1;
  55 static int brnf_filter_vlan_tagged __read_mostly = 0;
  56 static int brnf_filter_pppoe_tagged __read_mostly = 0;
  57 static int brnf_pass_vlan_indev __read_mostly = 0;
  58 #else
  59 #define brnf_call_iptables 1
  60 #define brnf_call_ip6tables 1
  61 #define brnf_call_arptables 1
  62 #define brnf_filter_vlan_tagged 0
  63 #define brnf_filter_pppoe_tagged 0
  64 #define brnf_pass_vlan_indev 0
  65 #endif
  66
  67 #define IS_IP(skb) \
  68         (!skb_vlan_tag_present(skb) && skb->protocol == htons(ETH_P_IP))
  69
  70 #define IS_IPV6(skb) \
  71         (!skb_vlan_tag_present(skb) && skb->protocol == htons(ETH_P_IPV6))
  72
  73 #define IS_ARP(skb) \
  74         (!skb_vlan_tag_present(skb) && skb->protocol == htons(ETH_P_ARP))
  75
  76 static inline __be16 vlan_proto(const struct sk_buff *skb)
  77 {
  78         if (skb_vlan_tag_present(skb))
  79                 return skb->protocol;
  80         else if (skb->protocol == htons(ETH_P_8021Q))
  81                 return vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
  82         else
  83                 return 0;
  84 }
  85
  86 #define IS_VLAN_IP(skb) \
  87         (vlan_proto(skb) == htons(ETH_P_IP) && \
  88          brnf_filter_vlan_tagged)
  89
  90 #define IS_VLAN_IPV6(skb) \
  91         (vlan_proto(skb) == htons(ETH_P_IPV6) && \
  92          brnf_filter_vlan_tagged)
  93
  94 #define IS_VLAN_ARP(skb) \
  95         (vlan_proto(skb) == htons(ETH_P_ARP) && \
  96          brnf_filter_vlan_tagged)
  97
  98 static inline __be16 pppoe_proto(const struct sk_buff *skb)
  99 {
 100         return *((__be16 *)(skb_mac_header(skb) + ETH_HLEN +
 101                             sizeof(struct pppoe_hdr)));
 102 }
 103
 104 #define IS_PPPOE_IP(skb) \
 105         (skb->protocol == htons(ETH_P_PPP_SES) && \
 106          pppoe_proto(skb) == htons(PPP_IP) && \
 107          brnf_filter_pppoe_tagged)
 108
 109 #define IS_PPPOE_IPV6(skb) \
 110         (skb->protocol == htons(ETH_P_PPP_SES) && \
 111          pppoe_proto(skb) == htons(PPP_IPV6) && \
 112          brnf_filter_pppoe_tagged)
 113
 114 /* largest possible L2 header, see br_nf_dev_queue_xmit() */
 115 #define NF_BRIDGE_MAX_MAC_HEADER_LENGTH (PPPOE_SES_HLEN + ETH_HLEN)
 116
 117 #if IS_ENABLED(CONFIG_NF_DEFRAG_IPV4)
 118 struct brnf_frag_data {
 119         char mac[NF_BRIDGE_MAX_MAC_HEADER_LENGTH];
 120         u8 encap_size;
 121         u8 size;
 122 };
 123
 124 static DEFINE_PER_CPU(struct brnf_frag_data, brnf_frag_data_storage);
 125 #endif
 126
 127 static struct nf_bridge_info *nf_bridge_info_get(const struct sk_buff *skb)
 128 {
 129         return skb->nf_bridge;
 130 }
 131
 132 static inline struct rtable *bridge_parent_rtable(const struct net_device *dev)
 133 {
 134         struct net_bridge_port *port;
 135
 136         port = br_port_get_rcu(dev);
 137         return port ? &port->br->fake_rtable : NULL;
 138 }
 139
 140 static inline struct net_device *bridge_parent(const struct net_device *dev)
 141 {
 142         struct net_bridge_port *port;
 143
 144         port = br_port_get_rcu(dev);
 145         return port ? port->br->dev : NULL;
 146 }
 147
 148 static inline struct nf_bridge_info *nf_bridge_alloc(struct sk_buff *skb)
 149 {
 150         skb->nf_bridge = kzalloc(sizeof(struct nf_bridge_info), GFP_ATOMIC);
 151         if (likely(skb->nf_bridge))
 152                 atomic_set(&(skb->nf_bridge->use), 1);
 153
 154         return skb->nf_bridge;
 155 }
 156
 157 static inline struct nf_bridge_info *nf_bridge_unshare(struct sk_buff *skb)
 158 {
 159         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 160
 161         if (atomic_read(&nf_bridge->use) > 1) {
 162                 struct nf_bridge_info *tmp = nf_bridge_alloc(skb);
 163
 164                 if (tmp) {
 165                         memcpy(tmp, nf_bridge, sizeof(struct nf_bridge_info));
 166                         atomic_set(&tmp->use, 1);
 167                 }
 168                 nf_bridge_put(nf_bridge);
 169                 nf_bridge = tmp;
 170         }
 171         return nf_bridge;
 172 }
 173
 174 static unsigned int nf_bridge_encap_header_len(const struct sk_buff *skb)
 175 {
 176         switch (skb->protocol) {
 177         case __cpu_to_be16(ETH_P_8021Q):
 178                 return VLAN_HLEN;
 179         case __cpu_to_be16(ETH_P_PPP_SES):
 180                 return PPPOE_SES_HLEN;
 181         default:
 182                 return 0;
 183         }
 184 }
 185
 186 static inline void nf_bridge_push_encap_header(struct sk_buff *skb)
 187 {
 188         unsigned int len = nf_bridge_encap_header_len(skb);
 189
 190         skb_push(skb, len);
 191         skb->network_header -= len;
 192 }
 193
 194 static inline void nf_bridge_pull_encap_header(struct sk_buff *skb)
 195 {
 196         unsigned int len = nf_bridge_encap_header_len(skb);
 197
 198         skb_pull(skb, len);
 199         skb->network_header += len;
 200 }
 201
 202 static inline void nf_bridge_pull_encap_header_rcsum(struct sk_buff *skb)
 203 {
 204         unsigned int len = nf_bridge_encap_header_len(skb);
 205
 206         skb_pull_rcsum(skb, len);
 207         skb->network_header += len;
 208 }
 209
 210 /* When handing a packet over to the IP layer
 211  * check whether we have a skb that is in the
 212  * expected format
 213  */
 214
 215 static int br_parse_ip_options(struct sk_buff *skb)
 216 {
 217         const struct iphdr *iph;
 218         struct net_device *dev = skb->dev;
 219         u32 len;
 220
 221         if (!pskb_may_pull(skb, sizeof(struct iphdr)))
 222                 goto inhdr_error;
 223
 224         iph = ip_hdr(skb);
 225
 226         /* Basic sanity checks */
 227         if (iph->ihl < 5 || iph->version != 4)
 228                 goto inhdr_error;
 229
 230         if (!pskb_may_pull(skb, iph->ihl*4))
 231                 goto inhdr_error;
 232
 233         iph = ip_hdr(skb);
 234         if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl)))
 235                 goto inhdr_error;
 236
 237         len = ntohs(iph->tot_len);
 238         if (skb->len < len) {
 239                 IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INTRUNCATEDPKTS);
 240                 goto drop;
 241         } else if (len < (iph->ihl*4))
 242                 goto inhdr_error;
 243
 244         if (pskb_trim_rcsum(skb, len)) {
 245                 IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INDISCARDS);
 246                 goto drop;
 247         }
 248
 249         memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
 250         /* We should really parse IP options here but until
 251          * somebody who actually uses IP options complains to
 252          * us we'll just silently ignore the options because
 253          * we're lazy!
 254          */
 255         return 0;
 256
 257 inhdr_error:
 258         IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INHDRERRORS);
 259 drop:
 260         return -1;
 261 }
 262
 263 static void nf_bridge_update_protocol(struct sk_buff *skb)
 264 {
 265         switch (skb->nf_bridge->orig_proto) {
 266         case BRNF_PROTO_8021Q:
 267                 skb->protocol = htons(ETH_P_8021Q);
 268                 break;
 269         case BRNF_PROTO_PPPOE:
 270                 skb->protocol = htons(ETH_P_PPP_SES);
 271                 break;
 272         case BRNF_PROTO_UNCHANGED:
 273                 break;
 274         }
 275 }
 276
 277 /* PF_BRIDGE/PRE_ROUTING *********************************************/
 278 /* Undo the changes made for ip6tables PREROUTING and continue the
 279  * bridge PRE_ROUTING hook. */
 280 static int br_nf_pre_routing_finish_ipv6(struct sock *sk, struct sk_buff *skb)
 281 {
 282         struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
 283         struct rtable *rt;
 284
 285         if (nf_bridge->pkt_otherhost) {
 286                 skb->pkt_type = PACKET_OTHERHOST;
 287                 nf_bridge->pkt_otherhost = false;
 288         }
 289         nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
 290
 291         rt = bridge_parent_rtable(nf_bridge->physindev);
 292         if (!rt) {
 293                 kfree_skb(skb);
 294                 return 0;
 295         }
 296         skb_dst_set_noref(skb, &rt->dst);
 297
 298         skb->dev = nf_bridge->physindev;
 299         nf_bridge_update_protocol(skb);
 300         nf_bridge_push_encap_header(skb);
 301         NF_HOOK_THRESH(NFPROTO_BRIDGE, NF_BR_PRE_ROUTING, sk, skb,
 302                        skb->dev, NULL,
 303                        br_handle_frame_finish, 1);
 304
 305         return 0;
 306 }
 307
 308 /* Obtain the correct destination MAC address, while preserving the original
 309  * source MAC address. If we already know this address, we just copy it. If we
 310  * don't, we use the neighbour framework to find out. In both cases, we make
 311  * sure that br_handle_frame_finish() is called afterwards.
 312  */
 313 static int br_nf_pre_routing_finish_bridge(struct sock *sk, struct sk_buff *skb)
 314 {
 315         struct neighbour *neigh;
 316         struct dst_entry *dst;
 317
 318         skb->dev = bridge_parent(skb->dev);
 319         if (!skb->dev)
 320                 goto free_skb;
 321         dst = skb_dst(skb);
 322         neigh = dst_neigh_lookup_skb(dst, skb);
 323         if (neigh) {
 324                 struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
 325                 int ret;
 326
 327                 if (neigh->hh.hh_len) {
 328                         neigh_hh_bridge(&neigh->hh, skb);
 329                         skb->dev = nf_bridge->physindev;
 330                         ret = br_handle_frame_finish(sk, skb);
 331                 } else {
 332                         /* the neighbour function below overwrites the complete
 333                          * MAC header, so we save the Ethernet source address and
 334                          * protocol number.
 335                          */
 336                         skb_copy_from_linear_data_offset(skb,
 337                                                          -(ETH_HLEN-ETH_ALEN),
 338                                                          nf_bridge->neigh_header,
 339                                                          ETH_HLEN-ETH_ALEN);
 340                         /* tell br_dev_xmit to continue with forwarding */
 341                         nf_bridge->mask |= BRNF_BRIDGED_DNAT;
 342                         /* FIXME Need to refragment */
 343                         ret = neigh->output(neigh, skb);
 344                 }
 345                 neigh_release(neigh);
 346                 return ret;
 347         }
 348 free_skb:
 349         kfree_skb(skb);
 350         return 0;
 351 }
 352
 353 static bool dnat_took_place(const struct sk_buff *skb)
 354 {
 355 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
 356         enum ip_conntrack_info ctinfo;
 357         struct nf_conn *ct;
 358
 359         ct = nf_ct_get(skb, &ctinfo);
 360         if (!ct || nf_ct_is_untracked(ct))
 361                 return false;
 362
 363         return test_bit(IPS_DST_NAT_BIT, &ct->status);
 364 #else
 365         return false;
 366 #endif
 367 }
 368
 369 /* This requires some explaining. If DNAT has taken place,
 370  * we will need to fix up the destination Ethernet address.
 371  *
 372  * There are two cases to consider:
 373  * 1. The packet was DNAT'ed to a device in the same bridge
 374  *    port group as it was received on. We can still bridge
 375  *    the packet.
 376  * 2. The packet was DNAT'ed to a different device, either
 377  *    a non-bridged device or another bridge port group.
 378  *    The packet will need to be routed.
 379  *
 380  * The correct way of distinguishing between these two cases is to
 381  * call ip_route_input() and to look at skb->dst->dev, which is
 382  * changed to the destination device if ip_route_input() succeeds.
 383  *
 384  * Let's first consider the case that ip_route_input() succeeds:
 385  *
 386  * If the output device equals the logical bridge device the packet
 387  * came in on, we can consider this bridging. The corresponding MAC
 388  * address will be obtained in br_nf_pre_routing_finish_bridge.
 389  * Otherwise, the packet is considered to be routed and we just
 390  * change the destination MAC address so that the packet will
 391  * later be passed up to the IP stack to be routed. For a redirected
 392  * packet, ip_route_input() will give back the localhost as output device,
 393  * which differs from the bridge device.
 394  *
 395  * Let's now consider the case that ip_route_input() fails:
 396  *
 397  * This can be because the destination address is martian, in which case
 398  * the packet will be dropped.
 399  * If IP forwarding is disabled, ip_route_input() will fail, while
 400  * ip_route_output_key() can return success. The source
 401  * address for ip_route_output_key() is set to zero, so ip_route_output_key()
 402  * thinks we're handling a locally generated packet and won't care
 403  * if IP forwarding is enabled. If the output device equals the logical bridge
 404  * device, we proceed as if ip_route_input() succeeded. If it differs from the
 405  * logical bridge port or if ip_route_output_key() fails we drop the packet.
 406  */
 407 static int br_nf_pre_routing_finish(struct sock *sk, struct sk_buff *skb)
 408 {
 409         struct net_device *dev = skb->dev;
 410         struct iphdr *iph = ip_hdr(skb);
 411         struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
 412         struct rtable *rt;
 413         int err;
 414         int frag_max_size;
 415
 416         frag_max_size = IPCB(skb)->frag_max_size;
 417         BR_INPUT_SKB_CB(skb)->frag_max_size = frag_max_size;
 418
 419         if (nf_bridge->pkt_otherhost) {
 420                 skb->pkt_type = PACKET_OTHERHOST;
 421                 nf_bridge->pkt_otherhost = false;
 422         }
 423         nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
 424         if (dnat_took_place(skb)) {
 425                 if ((err = ip_route_input(skb, iph->daddr, iph->saddr, iph->tos, dev))) {
 426                         struct in_device *in_dev = __in_dev_get_rcu(dev);
 427
 428                         /* If err equals -EHOSTUNREACH the error is due to a
 429                          * martian destination or due to the fact that
 430                          * forwarding is disabled. For most martian packets,
 431                          * ip_route_output_key() will fail. It won't fail for 2 types of
 432                          * martian destinations: loopback destinations and destination
 433                          * 0.0.0.0. In both cases the packet will be dropped because the
 434                          * destination is the loopback device and not the bridge. */
 435                         if (err != -EHOSTUNREACH || !in_dev || IN_DEV_FORWARD(in_dev))
 436                                 goto free_skb;
 437
 438                         rt = ip_route_output(dev_net(dev), iph->daddr, 0,
 439                                              RT_TOS(iph->tos), 0);
 440                         if (!IS_ERR(rt)) {
 441                                 /* - Bridged-and-DNAT'ed traffic doesn't
 442                                  *   require ip_forwarding. */
 443                                 if (rt->dst.dev == dev) {
 444                                         skb_dst_set(skb, &rt->dst);
 445                                         goto bridged_dnat;
 446                                 }
 447                                 ip_rt_put(rt);
 448                         }
 449 free_skb:
 450                         kfree_skb(skb);
 451                         return 0;
 452                 } else {
 453                         if (skb_dst(skb)->dev == dev) {
 454 bridged_dnat:
 455                                 skb->dev = nf_bridge->physindev;
 456                                 nf_bridge_update_protocol(skb);
 457                                 nf_bridge_push_encap_header(skb);
 458                                 NF_HOOK_THRESH(NFPROTO_BRIDGE,
 459                                                NF_BR_PRE_ROUTING,
 460                                                sk, skb, skb->dev, NULL,
 461                                                br_nf_pre_routing_finish_bridge,
 462                                                1);
 463                                 return 0;
 464                         }
 465                         ether_addr_copy(eth_hdr(skb)->h_dest, dev->dev_addr);
 466                         skb->pkt_type = PACKET_HOST;
 467                 }
 468         } else {
 469                 rt = bridge_parent_rtable(nf_bridge->physindev);
 470                 if (!rt) {
 471                         kfree_skb(skb);
 472                         return 0;
 473                 }
 474                 skb_dst_set_noref(skb, &rt->dst);
 475         }
 476
 477         skb->dev = nf_bridge->physindev;
 478         nf_bridge_update_protocol(skb);
 479         nf_bridge_push_encap_header(skb);
 480         NF_HOOK_THRESH(NFPROTO_BRIDGE, NF_BR_PRE_ROUTING, sk, skb,
 481                        skb->dev, NULL,
 482                        br_handle_frame_finish, 1);
 483
 484         return 0;
 485 }
 486
 487 static struct net_device *brnf_get_logical_dev(struct sk_buff *skb, const struct net_device *dev)
 488 {
 489         struct net_device *vlan, *br;
 490
 491         br = bridge_parent(dev);
 492         if (brnf_pass_vlan_indev == 0 || !skb_vlan_tag_present(skb))
 493                 return br;
 494
 495         vlan = __vlan_find_dev_deep_rcu(br, skb->vlan_proto,
 496                                     skb_vlan_tag_get(skb) & VLAN_VID_MASK);
 497
 498         return vlan ? vlan : br;
 499 }
 500
 501 /* Some common code for IPv4/IPv6 */
 502 static struct net_device *setup_pre_routing(struct sk_buff *skb)
 503 {
 504         struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
 505
 506         if (skb->pkt_type == PACKET_OTHERHOST) {
 507                 skb->pkt_type = PACKET_HOST;
 508                 nf_bridge->pkt_otherhost = true;
 509         }
 510
 511         nf_bridge->mask |= BRNF_NF_BRIDGE_PREROUTING;
 512         nf_bridge->physindev = skb->dev;
 513         skb->dev = brnf_get_logical_dev(skb, skb->dev);
 514
 515         if (skb->protocol == htons(ETH_P_8021Q))
 516                 nf_bridge->orig_proto = BRNF_PROTO_8021Q;
 517         else if (skb->protocol == htons(ETH_P_PPP_SES))
 518                 nf_bridge->orig_proto = BRNF_PROTO_PPPOE;
 519
 520         /* Must drop socket now because of tproxy. */
 521         skb_orphan(skb);
 522         return skb->dev;
 523 }
 524
 525 /* We only check the length. A bridge shouldn't do any hop-by-hop stuff anyway */
 526 static int check_hbh_len(struct sk_buff *skb)
 527 {
 528         unsigned char *raw = (u8 *)(ipv6_hdr(skb) + 1);
 529         u32 pkt_len;
 530         const unsigned char *nh = skb_network_header(skb);
 531         int off = raw - nh;
 532         int len = (raw[1] + 1) << 3;
 533
 534         if ((raw + len) - skb->data > skb_headlen(skb))
 535                 goto bad;
 536
 537         off += 2;
 538         len -= 2;
 539
 540         while (len > 0) {
 541                 int optlen = nh[off + 1] + 2;
 542
 543                 switch (nh[off]) {
 544                 case IPV6_TLV_PAD1:
 545                         optlen = 1;
 546                         break;
 547
 548                 case IPV6_TLV_PADN:
 549                         break;
 550
 551                 case IPV6_TLV_JUMBO:
 552                         if (nh[off + 1] != 4 || (off & 3) != 2)
 553                                 goto bad;
 554                         pkt_len = ntohl(*(__be32 *) (nh + off + 2));
 555                         if (pkt_len <= IPV6_MAXPLEN ||
 556                             ipv6_hdr(skb)->payload_len)
 557                                 goto bad;
 558                         if (pkt_len > skb->len - sizeof(struct ipv6hdr))
 559                                 goto bad;
 560                         if (pskb_trim_rcsum(skb,
 561                                             pkt_len + sizeof(struct ipv6hdr)))
 562                                 goto bad;
 563                         nh = skb_network_header(skb);
 564                         break;
 565                 default:
 566                         if (optlen > len)
 567                                 goto bad;
 568                         break;
 569                 }
 570                 off += optlen;
 571                 len -= optlen;
 572         }
 573         if (len == 0)
 574                 return 0;
 575 bad:
 576         return -1;
 577
 578 }
 579
 580 /* Replicate the checks that IPv6 does on packet reception and pass the packet
 581  * to ip6tables, which doesn't support NAT, so things are fairly simple. */
 582 static unsigned int br_nf_pre_routing_ipv6(const struct nf_hook_ops *ops,
 583                                            struct sk_buff *skb,
 584                                            const struct nf_hook_state *state)
 585 {
 586         const struct ipv6hdr *hdr;
 587         u32 pkt_len;
 588
 589         if (skb->len < sizeof(struct ipv6hdr))
 590                 return NF_DROP;
 591
 592         if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
 593                 return NF_DROP;
 594
 595         hdr = ipv6_hdr(skb);
 596
 597         if (hdr->version != 6)
 598                 return NF_DROP;
 599
 600         pkt_len = ntohs(hdr->payload_len);
 601
 602         if (pkt_len || hdr->nexthdr != NEXTHDR_HOP) {
 603                 if (pkt_len + sizeof(struct ipv6hdr) > skb->len)
 604                         return NF_DROP;
 605                 if (pskb_trim_rcsum(skb, pkt_len + sizeof(struct ipv6hdr)))
 606                         return NF_DROP;
 607         }
 608         if (hdr->nexthdr == NEXTHDR_HOP && check_hbh_len(skb))
 609                 return NF_DROP;
 610
 611         nf_bridge_put(skb->nf_bridge);
 612         if (!nf_bridge_alloc(skb))
 613                 return NF_DROP;
 614         if (!setup_pre_routing(skb))
 615                 return NF_DROP;
 616
 617         skb->protocol = htons(ETH_P_IPV6);
 618         NF_HOOK(NFPROTO_IPV6, NF_INET_PRE_ROUTING, state->sk, skb,
 619                 skb->dev, NULL,
 620                 br_nf_pre_routing_finish_ipv6);
 621
 622         return NF_STOLEN;
 623 }
 624
 625 /* Direct IPv6 traffic to br_nf_pre_routing_ipv6.
 626  * Replicate the checks that IPv4 does on packet reception.
 627  * Set skb->dev to the bridge device (i.e. parent of the
 628  * receiving device) to make netfilter happy, the REDIRECT
 629  * target in particular.  Save the original destination IP
 630  * address to be able to detect DNAT afterwards. */
 631 static unsigned int br_nf_pre_routing(const struct nf_hook_ops *ops,
 632                                       struct sk_buff *skb,
 633                                       const struct nf_hook_state *state)
 634 {
 635         struct net_bridge_port *p;
 636         struct net_bridge *br;
 637         __u32 len = nf_bridge_encap_header_len(skb);
 638
 639         if (unlikely(!pskb_may_pull(skb, len)))
 640                 return NF_DROP;
 641
 642         p = br_port_get_rcu(state->in);
 643         if (p == NULL)
 644                 return NF_DROP;
 645         br = p->br;
 646
 647         if (IS_IPV6(skb) || IS_VLAN_IPV6(skb) || IS_PPPOE_IPV6(skb)) {
 648                 if (!brnf_call_ip6tables && !br->nf_call_ip6tables)
 649                         return NF_ACCEPT;
 650
 651                 nf_bridge_pull_encap_header_rcsum(skb);
 652                 return br_nf_pre_routing_ipv6(ops, skb, state);
 653         }
 654
 655         if (!brnf_call_iptables && !br->nf_call_iptables)
 656                 return NF_ACCEPT;
 657
 658         if (!IS_IP(skb) && !IS_VLAN_IP(skb) && !IS_PPPOE_IP(skb))
 659                 return NF_ACCEPT;
 660
 661         nf_bridge_pull_encap_header_rcsum(skb);
 662
 663         if (br_parse_ip_options(skb))
 664                 return NF_DROP;
 665
 666         nf_bridge_put(skb->nf_bridge);
 667         if (!nf_bridge_alloc(skb))
 668                 return NF_DROP;
 669         if (!setup_pre_routing(skb))
 670                 return NF_DROP;
 671
 672         skb->protocol = htons(ETH_P_IP);
 673
 674         NF_HOOK(NFPROTO_IPV4, NF_INET_PRE_ROUTING, state->sk, skb,
 675                 skb->dev, NULL,
 676                 br_nf_pre_routing_finish);
 677
 678         return NF_STOLEN;
 679 }
 680
 681
 682 /* PF_BRIDGE/LOCAL_IN ************************************************/
 683 /* The packet is locally destined, which requires a real
 684  * dst_entry, so detach the fake one.  On the way up, the
 685  * packet would pass through PRE_ROUTING again (which already
 686  * took place when the packet entered the bridge), but we
 687  * register an IPv4 PRE_ROUTING 'sabotage' hook that will
 688  * prevent this from happening. */
 689 static unsigned int br_nf_local_in(const struct nf_hook_ops *ops,
 690                                    struct sk_buff *skb,
 691                                    const struct nf_hook_state *state)
 692 {
 693         br_drop_fake_rtable(skb);
 694         return NF_ACCEPT;
 695 }
 696
 697 /* PF_BRIDGE/FORWARD *************************************************/
 698 static int br_nf_forward_finish(struct sock *sk, struct sk_buff *skb)
 699 {
 700         struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
 701         struct net_device *in;
 702
 703         if (!IS_ARP(skb) && !IS_VLAN_ARP(skb)) {
 704                 int frag_max_size;
 705
 706                 if (skb->protocol == htons(ETH_P_IP)) {
 707                         frag_max_size = IPCB(skb)->frag_max_size;
 708                         BR_INPUT_SKB_CB(skb)->frag_max_size = frag_max_size;
 709                 }
 710
 711                 in = nf_bridge->physindev;
 712                 if (nf_bridge->pkt_otherhost) {
 713                         skb->pkt_type = PACKET_OTHERHOST;
 714                         nf_bridge->pkt_otherhost = false;
 715                 }
 716                 nf_bridge_update_protocol(skb);
 717         } else {
 718                 in = *((struct net_device **)(skb->cb));
 719         }
 720         nf_bridge_push_encap_header(skb);
 721
 722         NF_HOOK_THRESH(NFPROTO_BRIDGE, NF_BR_FORWARD, sk, skb,
 723                        in, skb->dev, br_forward_finish, 1);
 724         return 0;
 725 }
 726
 727
 728 /* This is the 'purely bridged' case.  For IP, we pass the packet to
 729  * netfilter with indev and outdev set to the bridge device,
 730  * but we are still able to filter on the 'real' indev/outdev
 731  * because of the physdev module. For ARP, indev and outdev are the
 732  * bridge ports. */
 733 static unsigned int br_nf_forward_ip(const struct nf_hook_ops *ops,
 734                                      struct sk_buff *skb,
 735                                      const struct nf_hook_state *state)
 736 {
 737         struct nf_bridge_info *nf_bridge;
 738         struct net_device *parent;
 739         u_int8_t pf;
 740
 741         if (!skb->nf_bridge)
 742                 return NF_ACCEPT;
 743
 744         /* Need exclusive nf_bridge_info since we might have multiple
 745          * different physoutdevs. */
 746         if (!nf_bridge_unshare(skb))
 747                 return NF_DROP;
 748
 749         nf_bridge = nf_bridge_info_get(skb);
 750         if (!nf_bridge)
 751                 return NF_DROP;
 752
 753         parent = bridge_parent(state->out);
 754         if (!parent)
 755                 return NF_DROP;
 756
 757         if (IS_IP(skb) || IS_VLAN_IP(skb) || IS_PPPOE_IP(skb))
 758                 pf = NFPROTO_IPV4;
 759         else if (IS_IPV6(skb) || IS_VLAN_IPV6(skb) || IS_PPPOE_IPV6(skb))
 760                 pf = NFPROTO_IPV6;
 761         else
 762                 return NF_ACCEPT;
 763
 764         nf_bridge_pull_encap_header(skb);
 765
 766         if (skb->pkt_type == PACKET_OTHERHOST) {
 767                 skb->pkt_type = PACKET_HOST;
 768                 nf_bridge->pkt_otherhost = true;
 769         }
 770
 771         if (pf == NFPROTO_IPV4) {
 772                 int frag_max = BR_INPUT_SKB_CB(skb)->frag_max_size;
 773
 774                 if (br_parse_ip_options(skb))
 775                         return NF_DROP;
 776
 777                 IPCB(skb)->frag_max_size = frag_max;
 778         }
 779
 780         nf_bridge->physoutdev = skb->dev;
 781         if (pf == NFPROTO_IPV4)
 782                 skb->protocol = htons(ETH_P_IP);
 783         else
 784                 skb->protocol = htons(ETH_P_IPV6);
 785
 786         NF_HOOK(pf, NF_INET_FORWARD, NULL, skb,
 787                 brnf_get_logical_dev(skb, state->in),
 788                 parent, br_nf_forward_finish);
 789
 790         return NF_STOLEN;
 791 }
 792
 793 static unsigned int br_nf_forward_arp(const struct nf_hook_ops *ops,
 794                                       struct sk_buff *skb,
 795                                       const struct nf_hook_state *state)
 796 {
 797         struct net_bridge_port *p;
 798         struct net_bridge *br;
 799         struct net_device **d = (struct net_device **)(skb->cb);
 800
 801         p = br_port_get_rcu(state->out);
 802         if (p == NULL)
 803                 return NF_ACCEPT;
 804         br = p->br;
 805
 806         if (!brnf_call_arptables && !br->nf_call_arptables)
 807                 return NF_ACCEPT;
 808
 809         if (!IS_ARP(skb)) {
 810                 if (!IS_VLAN_ARP(skb))
 811                         return NF_ACCEPT;
 812                 nf_bridge_pull_encap_header(skb);
 813         }
 814
 815         if (arp_hdr(skb)->ar_pln != 4) {
 816                 if (IS_VLAN_ARP(skb))
 817                         nf_bridge_push_encap_header(skb);
 818                 return NF_ACCEPT;
 819         }
 820         *d = state->in;
 821         NF_HOOK(NFPROTO_ARP, NF_ARP_FORWARD, state->sk, skb,
 822                 state->in, state->out, br_nf_forward_finish);
 823
 824         return NF_STOLEN;
 825 }
 826
 827 #if IS_ENABLED(CONFIG_NF_DEFRAG_IPV4)
 828 static int br_nf_push_frag_xmit(struct sock *sk, struct sk_buff *skb)
 829 {
 830         struct brnf_frag_data *data;
 831         int err;
 832
 833         data = this_cpu_ptr(&brnf_frag_data_storage);
 834         err = skb_cow_head(skb, data->size);
 835
 836         if (err) {
 837                 kfree_skb(skb);
 838                 return 0;
 839         }
 840
 841         skb_copy_to_linear_data_offset(skb, -data->size, data->mac, data->size);
 842         __skb_push(skb, data->encap_size);
 843
 844         return br_dev_queue_push_xmit(sk, skb);
 845 }
 846
 847 static int br_nf_dev_queue_xmit(struct sock *sk, struct sk_buff *skb)
 848 {
 849         int ret;
 850         int frag_max_size;
 851         unsigned int mtu_reserved;
 852
 853         if (skb_is_gso(skb) || skb->protocol != htons(ETH_P_IP))
 854                 return br_dev_queue_push_xmit(sk, skb);
 855
 856         mtu_reserved = nf_bridge_mtu_reduction(skb);
 857         /* This is wrong! We should preserve the original fragment
 858          * boundaries by preserving frag_list rather than refragmenting.
 859          */
 860         if (skb->len + mtu_reserved > skb->dev->mtu) {
 861                 struct brnf_frag_data *data;
 862
 863                 frag_max_size = BR_INPUT_SKB_CB(skb)->frag_max_size;
 864                 if (br_parse_ip_options(skb))
 865                         /* Drop invalid packet */
 866                         return NF_DROP;
 867                 IPCB(skb)->frag_max_size = frag_max_size;
 868
 869                 nf_bridge_update_protocol(skb);
 870
 871                 data = this_cpu_ptr(&brnf_frag_data_storage);
 872                 data->encap_size = nf_bridge_encap_header_len(skb);
 873                 data->size = ETH_HLEN + data->encap_size;
 874
 875                 skb_copy_from_linear_data_offset(skb, -data->size, data->mac,
 876                                                  data->size);
 877
 878                 ret = ip_fragment(sk, skb, br_nf_push_frag_xmit);
 879         } else {
 880                 ret = br_dev_queue_push_xmit(sk, skb);
 881         }
 882
 883         return ret;
 884 }
 885 #else
 886 static int br_nf_dev_queue_xmit(struct sock *sk, struct sk_buff *skb)
 887 {
 888         return br_dev_queue_push_xmit(sk, skb);
 889 }
 890 #endif
 891
 892 /* PF_BRIDGE/POST_ROUTING ********************************************/
 893 static unsigned int br_nf_post_routing(const struct nf_hook_ops *ops,
 894                                        struct sk_buff *skb,
 895                                        const struct nf_hook_state *state)
 896 {
 897         struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
 898         struct net_device *realoutdev = bridge_parent(skb->dev);
 899         u_int8_t pf;
 900
 901         /* if nf_bridge is set, but ->physoutdev is NULL, this packet came in
 902          * on a bridge, but was delivered locally and is now being routed:
 903          *
 904          * POST_ROUTING was already invoked from the ip stack.
 905          */
 906         if (!nf_bridge || !nf_bridge->physoutdev)
 907                 return NF_ACCEPT;
 908
 909         if (!realoutdev)
 910                 return NF_DROP;
 911
 912         if (IS_IP(skb) || IS_VLAN_IP(skb) || IS_PPPOE_IP(skb))
 913                 pf = NFPROTO_IPV4;
 914         else if (IS_IPV6(skb) || IS_VLAN_IPV6(skb) || IS_PPPOE_IPV6(skb))
 915                 pf = NFPROTO_IPV6;
 916         else
 917                 return NF_ACCEPT;
 918
 919         /* We assume any code from br_dev_queue_push_xmit onwards doesn't care
 920          * about the value of skb->pkt_type. */
 921         if (skb->pkt_type == PACKET_OTHERHOST) {
 922                 skb->pkt_type = PACKET_HOST;
 923                 nf_bridge->pkt_otherhost = true;
 924         }
 925
 926         nf_bridge_pull_encap_header(skb);
 927         if (pf == NFPROTO_IPV4)
 928                 skb->protocol = htons(ETH_P_IP);
 929         else
 930                 skb->protocol = htons(ETH_P_IPV6);
 931
 932         NF_HOOK(pf, NF_INET_POST_ROUTING, state->sk, skb,
 933                 NULL, realoutdev,
 934                 br_nf_dev_queue_xmit);
 935
 936         return NF_STOLEN;
 937 }
 938
 939 /* IP/SABOTAGE *****************************************************/
 940 /* Don't hand locally destined packets to PF_INET(6)/PRE_ROUTING
 941  * for the second time. */
 942 static unsigned int ip_sabotage_in(const struct nf_hook_ops *ops,
 943                                    struct sk_buff *skb,
 944                                    const struct nf_hook_state *state)
 945 {
 946         if (skb->nf_bridge &&
 947             !(skb->nf_bridge->mask & BRNF_NF_BRIDGE_PREROUTING)) {
 948                 return NF_STOP;
 949         }
 950
 951         return NF_ACCEPT;
 952 }
 953
 954 /* This is called when br_netfilter has called into iptables/netfilter,
 955  * and DNAT has taken place on a bridge-forwarded packet.
 956  *
 957  * neigh->output has created a new MAC header, with local br0 MAC
 958  * as saddr.
 959  *
 960  * This restores the original MAC saddr of the bridged packet
 961  * before invoking bridge forward logic to transmit the packet.
 962  */
 963 static void br_nf_pre_routing_finish_bridge_slow(struct sk_buff *skb)
 964 {
 965         struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
 966
 967         skb_pull(skb, ETH_HLEN);
 968         nf_bridge->mask &= ~BRNF_BRIDGED_DNAT;
 969
 970         BUILD_BUG_ON(sizeof(nf_bridge->neigh_header) != (ETH_HLEN - ETH_ALEN));
 971
 972         skb_copy_to_linear_data_offset(skb, -(ETH_HLEN - ETH_ALEN),
 973                                        nf_bridge->neigh_header,
 974                                        ETH_HLEN - ETH_ALEN);
 975         skb->dev = nf_bridge->physindev;
 976
 977         nf_bridge->physoutdev = NULL;
 978         br_handle_frame_finish(NULL, skb);
 979 }
 980
 981 static int br_nf_dev_xmit(struct sk_buff *skb)
 982 {
 983         if (skb->nf_bridge && (skb->nf_bridge->mask & BRNF_BRIDGED_DNAT)) {
 984                 br_nf_pre_routing_finish_bridge_slow(skb);
 985                 return 1;
 986         }
 987         return 0;
 988 }
 989
 990 static const struct nf_br_ops br_ops = {
 991         .br_dev_xmit_hook =     br_nf_dev_xmit,
 992 };
 993
 994 void br_netfilter_enable(void)
 995 {
 996 }
 997 EXPORT_SYMBOL_GPL(br_netfilter_enable);
 998
 999 /* For br_nf_post_routing, we need (prio = NF_BR_PRI_LAST), because
1000  * br_dev_queue_push_xmit is called afterwards */
1001 static struct nf_hook_ops br_nf_ops[] __read_mostly = {
1002         {
1003                 .hook = br_nf_pre_routing,
1004                 .owner = THIS_MODULE,
1005                 .pf = NFPROTO_BRIDGE,
1006                 .hooknum = NF_BR_PRE_ROUTING,
1007                 .priority = NF_BR_PRI_BRNF,
1008         },
1009         {
1010                 .hook = br_nf_local_in,
1011                 .owner = THIS_MODULE,
1012                 .pf = NFPROTO_BRIDGE,
1013                 .hooknum = NF_BR_LOCAL_IN,
1014                 .priority = NF_BR_PRI_BRNF,
1015         },
1016         {
1017                 .hook = br_nf_forward_ip,
1018                 .owner = THIS_MODULE,
1019                 .pf = NFPROTO_BRIDGE,
1020                 .hooknum = NF_BR_FORWARD,
1021                 .priority = NF_BR_PRI_BRNF - 1,
1022         },
1023         {
1024                 .hook = br_nf_forward_arp,
1025                 .owner = THIS_MODULE,
1026                 .pf = NFPROTO_BRIDGE,
1027                 .hooknum = NF_BR_FORWARD,
1028                 .priority = NF_BR_PRI_BRNF,
1029         },
1030         {
1031                 .hook = br_nf_post_routing,
1032                 .owner = THIS_MODULE,
1033                 .pf = NFPROTO_BRIDGE,
1034                 .hooknum = NF_BR_POST_ROUTING,
1035                 .priority = NF_BR_PRI_LAST,
1036         },
1037         {
1038                 .hook = ip_sabotage_in,
1039                 .owner = THIS_MODULE,
1040                 .pf = NFPROTO_IPV4,
1041                 .hooknum = NF_INET_PRE_ROUTING,
1042                 .priority = NF_IP_PRI_FIRST,
1043         },
1044         {
1045                 .hook = ip_sabotage_in,
1046                 .owner = THIS_MODULE,
1047                 .pf = NFPROTO_IPV6,
1048                 .hooknum = NF_INET_PRE_ROUTING,
1049                 .priority = NF_IP6_PRI_FIRST,
1050         },
1051 };
1052
1053 #ifdef CONFIG_SYSCTL
1054 static
1055 int brnf_sysctl_call_tables(struct ctl_table *ctl, int write,
1056                             void __user *buffer, size_t *lenp, loff_t *ppos)
1057 {
1058         int ret;
1059
1060         ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
1061
1062         if (write && *(int *)(ctl->data))
1063                 *(int *)(ctl->data) = 1;
1064         return ret;
1065 }
1066
1067 static struct ctl_table brnf_table[] = {
1068         {
1069                 .procname       = "bridge-nf-call-arptables",
1070                 .data           = &brnf_call_arptables,
1071                 .maxlen         = sizeof(int),
1072                 .mode           = 0644,
1073                 .proc_handler   = brnf_sysctl_call_tables,
1074         },
1075         {
1076                 .procname       = "bridge-nf-call-iptables",
1077                 .data           = &brnf_call_iptables,
1078                 .maxlen         = sizeof(int),
1079                 .mode           = 0644,
1080                 .proc_handler   = brnf_sysctl_call_tables,
1081         },
1082         {
1083                 .procname       = "bridge-nf-call-ip6tables",
1084                 .data           = &brnf_call_ip6tables,
1085                 .maxlen         = sizeof(int),
1086                 .mode           = 0644,
1087                 .proc_handler   = brnf_sysctl_call_tables,
1088         },
1089         {
1090                 .procname       = "bridge-nf-filter-vlan-tagged",
1091                 .data           = &brnf_filter_vlan_tagged,
1092                 .maxlen         = sizeof(int),
1093                 .mode           = 0644,
1094                 .proc_handler   = brnf_sysctl_call_tables,
1095         },
1096         {
1097                 .procname       = "bridge-nf-filter-pppoe-tagged",
1098                 .data           = &brnf_filter_pppoe_tagged,
1099                 .maxlen         = sizeof(int),
1100                 .mode           = 0644,
1101                 .proc_handler   = brnf_sysctl_call_tables,
1102         },
1103         {
1104                 .procname       = "bridge-nf-pass-vlan-input-dev",
1105                 .data           = &brnf_pass_vlan_indev,
1106                 .maxlen         = sizeof(int),
1107                 .mode           = 0644,
1108                 .proc_handler   = brnf_sysctl_call_tables,
1109         },
1110         { }
1111 };
1112 #endif
1113
1114 static int __init br_netfilter_init(void)
1115 {
1116         int ret;
1117
1118         ret = nf_register_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
1119         if (ret < 0)
1120                 return ret;
1121
1122 #ifdef CONFIG_SYSCTL
1123         brnf_sysctl_header = register_net_sysctl(&init_net, "net/bridge", brnf_table);
1124         if (brnf_sysctl_header == NULL) {
1125                 printk(KERN_WARNING
1126                        "br_netfilter: can't register to sysctl.\n");
1127                 nf_unregister_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
1128                 return -ENOMEM;
1129         }
1130 #endif
1131         RCU_INIT_POINTER(nf_br_ops, &br_ops);
1132         printk(KERN_NOTICE "Bridge firewalling registered\n");
1133         return 0;
1134 }
1135
1136 static void __exit br_netfilter_fini(void)
1137 {
1138         RCU_INIT_POINTER(nf_br_ops, NULL);
1139         nf_unregister_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
1140 #ifdef CONFIG_SYSCTL
1141         unregister_net_sysctl_table(brnf_sysctl_header);
1142 #endif
1143 }
1144
1145 module_init(br_netfilter_init);
1146 module_exit(br_netfilter_fini);
1147
1148 MODULE_LICENSE("GPL");
1149 MODULE_AUTHOR("Lennert Buytenhek <buytenh@gnu.org>");
1150 MODULE_AUTHOR("Bart De Schuymer <bdschuym@pandora.be>");
1151 MODULE_DESCRIPTION("Linux ethernet netfilter firewall bridge");