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 #include <asm/uaccess.h>
  41 #include "br_private.h"
  42 #ifdef CONFIG_SYSCTL
  43 #include <linux/sysctl.h>
  44 #endif
  45
  46 #define skb_origaddr(skb)        (((struct bridge_skb_cb *) \
  47                                  (skb->nf_bridge->data))->daddr.ipv4)
  48 #define store_orig_dstaddr(skb)  (skb_origaddr(skb) = ip_hdr(skb)->daddr)
  49 #define dnat_took_place(skb)     (skb_origaddr(skb) != ip_hdr(skb)->daddr)
  50
  51 #ifdef CONFIG_SYSCTL
  52 static struct ctl_table_header *brnf_sysctl_header;
  53 static int brnf_call_iptables __read_mostly = 1;
  54 static int brnf_call_ip6tables __read_mostly = 1;
  55 static int brnf_call_arptables __read_mostly = 1;
  56 static int brnf_filter_vlan_tagged __read_mostly = 0;
  57 static int brnf_filter_pppoe_tagged __read_mostly = 0;
  58 static int brnf_pass_vlan_indev __read_mostly = 0;
  59 #else
  60 #define brnf_call_iptables 1
  61 #define brnf_call_ip6tables 1
  62 #define brnf_call_arptables 1
  63 #define brnf_filter_vlan_tagged 0
  64 #define brnf_filter_pppoe_tagged 0
  65 #define brnf_pass_vlan_indev 0
  66 #endif
  67
  68 #define IS_IP(skb) \
  69         (!skb_vlan_tag_present(skb) && skb->protocol == htons(ETH_P_IP))
  70
  71 #define IS_IPV6(skb) \
  72         (!skb_vlan_tag_present(skb) && skb->protocol == htons(ETH_P_IPV6))
  73
  74 #define IS_ARP(skb) \
  75         (!skb_vlan_tag_present(skb) && skb->protocol == htons(ETH_P_ARP))
  76
  77 static inline __be16 vlan_proto(const struct sk_buff *skb)
  78 {
  79         if (skb_vlan_tag_present(skb))
  80                 return skb->protocol;
  81         else if (skb->protocol == htons(ETH_P_8021Q))
  82                 return vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
  83         else
  84                 return 0;
  85 }
  86
  87 #define IS_VLAN_IP(skb) \
  88         (vlan_proto(skb) == htons(ETH_P_IP) && \
  89          brnf_filter_vlan_tagged)
  90
  91 #define IS_VLAN_IPV6(skb) \
  92         (vlan_proto(skb) == htons(ETH_P_IPV6) && \
  93          brnf_filter_vlan_tagged)
  94
  95 #define IS_VLAN_ARP(skb) \
  96         (vlan_proto(skb) == htons(ETH_P_ARP) && \
  97          brnf_filter_vlan_tagged)
  98
  99 static inline __be16 pppoe_proto(const struct sk_buff *skb)
 100 {
 101         return *((__be16 *)(skb_mac_header(skb) + ETH_HLEN +
 102                             sizeof(struct pppoe_hdr)));
 103 }
 104
 105 #define IS_PPPOE_IP(skb) \
 106         (skb->protocol == htons(ETH_P_PPP_SES) && \
 107          pppoe_proto(skb) == htons(PPP_IP) && \
 108          brnf_filter_pppoe_tagged)
 109
 110 #define IS_PPPOE_IPV6(skb) \
 111         (skb->protocol == htons(ETH_P_PPP_SES) && \
 112          pppoe_proto(skb) == htons(PPP_IPV6) && \
 113          brnf_filter_pppoe_tagged)
 114
 115 static inline struct rtable *bridge_parent_rtable(const struct net_device *dev)
 116 {
 117         struct net_bridge_port *port;
 118
 119         port = br_port_get_rcu(dev);
 120         return port ? &port->br->fake_rtable : NULL;
 121 }
 122
 123 static inline struct net_device *bridge_parent(const struct net_device *dev)
 124 {
 125         struct net_bridge_port *port;
 126
 127         port = br_port_get_rcu(dev);
 128         return port ? port->br->dev : NULL;
 129 }
 130
 131 static inline struct nf_bridge_info *nf_bridge_alloc(struct sk_buff *skb)
 132 {
 133         skb->nf_bridge = kzalloc(sizeof(struct nf_bridge_info), GFP_ATOMIC);
 134         if (likely(skb->nf_bridge))
 135                 atomic_set(&(skb->nf_bridge->use), 1);
 136
 137         return skb->nf_bridge;
 138 }
 139
 140 static inline struct nf_bridge_info *nf_bridge_unshare(struct sk_buff *skb)
 141 {
 142         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 143
 144         if (atomic_read(&nf_bridge->use) > 1) {
 145                 struct nf_bridge_info *tmp = nf_bridge_alloc(skb);
 146
 147                 if (tmp) {
 148                         memcpy(tmp, nf_bridge, sizeof(struct nf_bridge_info));
 149                         atomic_set(&tmp->use, 1);
 150                 }
 151                 nf_bridge_put(nf_bridge);
 152                 nf_bridge = tmp;
 153         }
 154         return nf_bridge;
 155 }
 156
 157 static inline void nf_bridge_push_encap_header(struct sk_buff *skb)
 158 {
 159         unsigned int len = nf_bridge_encap_header_len(skb);
 160
 161         skb_push(skb, len);
 162         skb->network_header -= len;
 163 }
 164
 165 static inline void nf_bridge_pull_encap_header(struct sk_buff *skb)
 166 {
 167         unsigned int len = nf_bridge_encap_header_len(skb);
 168
 169         skb_pull(skb, len);
 170         skb->network_header += len;
 171 }
 172
 173 static inline void nf_bridge_pull_encap_header_rcsum(struct sk_buff *skb)
 174 {
 175         unsigned int len = nf_bridge_encap_header_len(skb);
 176
 177         skb_pull_rcsum(skb, len);
 178         skb->network_header += len;
 179 }
 180
 181 static inline void nf_bridge_save_header(struct sk_buff *skb)
 182 {
 183         int header_size = ETH_HLEN + nf_bridge_encap_header_len(skb);
 184
 185         skb_copy_from_linear_data_offset(skb, -header_size,
 186                                          skb->nf_bridge->data, header_size);
 187 }
 188
 189 /* When handing a packet over to the IP layer
 190  * check whether we have a skb that is in the
 191  * expected format
 192  */
 193
 194 static int br_parse_ip_options(struct sk_buff *skb)
 195 {
 196         const struct iphdr *iph;
 197         struct net_device *dev = skb->dev;
 198         u32 len;
 199
 200         if (!pskb_may_pull(skb, sizeof(struct iphdr)))
 201                 goto inhdr_error;
 202
 203         iph = ip_hdr(skb);
 204
 205         /* Basic sanity checks */
 206         if (iph->ihl < 5 || iph->version != 4)
 207                 goto inhdr_error;
 208
 209         if (!pskb_may_pull(skb, iph->ihl*4))
 210                 goto inhdr_error;
 211
 212         iph = ip_hdr(skb);
 213         if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl)))
 214                 goto inhdr_error;
 215
 216         len = ntohs(iph->tot_len);
 217         if (skb->len < len) {
 218                 IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INTRUNCATEDPKTS);
 219                 goto drop;
 220         } else if (len < (iph->ihl*4))
 221                 goto inhdr_error;
 222
 223         if (pskb_trim_rcsum(skb, len)) {
 224                 IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INDISCARDS);
 225                 goto drop;
 226         }
 227
 228         memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
 229         /* We should really parse IP options here but until
 230          * somebody who actually uses IP options complains to
 231          * us we'll just silently ignore the options because
 232          * we're lazy!
 233          */
 234         return 0;
 235
 236 inhdr_error:
 237         IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INHDRERRORS);
 238 drop:
 239         return -1;
 240 }
 241
 242 /* PF_BRIDGE/PRE_ROUTING *********************************************/
 243 /* Undo the changes made for ip6tables PREROUTING and continue the
 244  * bridge PRE_ROUTING hook. */
 245 static int br_nf_pre_routing_finish_ipv6(struct sk_buff *skb)
 246 {
 247         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 248         struct rtable *rt;
 249
 250         if (nf_bridge->mask & BRNF_PKT_TYPE) {
 251                 skb->pkt_type = PACKET_OTHERHOST;
 252                 nf_bridge->mask ^= BRNF_PKT_TYPE;
 253         }
 254         nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
 255
 256         rt = bridge_parent_rtable(nf_bridge->physindev);
 257         if (!rt) {
 258                 kfree_skb(skb);
 259                 return 0;
 260         }
 261         skb_dst_set_noref(skb, &rt->dst);
 262
 263         skb->dev = nf_bridge->physindev;
 264         nf_bridge_update_protocol(skb);
 265         nf_bridge_push_encap_header(skb);
 266         NF_HOOK_THRESH(NFPROTO_BRIDGE, NF_BR_PRE_ROUTING, skb, skb->dev, NULL,
 267                        br_handle_frame_finish, 1);
 268
 269         return 0;
 270 }
 271
 272 /* Obtain the correct destination MAC address, while preserving the original
 273  * source MAC address. If we already know this address, we just copy it. If we
 274  * don't, we use the neighbour framework to find out. In both cases, we make
 275  * sure that br_handle_frame_finish() is called afterwards.
 276  */
 277 static int br_nf_pre_routing_finish_bridge(struct sk_buff *skb)
 278 {
 279         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 280         struct neighbour *neigh;
 281         struct dst_entry *dst;
 282
 283         skb->dev = bridge_parent(skb->dev);
 284         if (!skb->dev)
 285                 goto free_skb;
 286         dst = skb_dst(skb);
 287         neigh = dst_neigh_lookup_skb(dst, skb);
 288         if (neigh) {
 289                 int ret;
 290
 291                 if (neigh->hh.hh_len) {
 292                         neigh_hh_bridge(&neigh->hh, skb);
 293                         skb->dev = nf_bridge->physindev;
 294                         ret = br_handle_frame_finish(skb);
 295                 } else {
 296                         /* the neighbour function below overwrites the complete
 297                          * MAC header, so we save the Ethernet source address and
 298                          * protocol number.
 299                          */
 300                         skb_copy_from_linear_data_offset(skb,
 301                                                          -(ETH_HLEN-ETH_ALEN),
 302                                                          skb->nf_bridge->data,
 303                                                          ETH_HLEN-ETH_ALEN);
 304                         /* tell br_dev_xmit to continue with forwarding */
 305                         nf_bridge->mask |= BRNF_BRIDGED_DNAT;
 306                         /* FIXME Need to refragment */
 307                         ret = neigh->output(neigh, skb);
 308                 }
 309                 neigh_release(neigh);
 310                 return ret;
 311         }
 312 free_skb:
 313         kfree_skb(skb);
 314         return 0;
 315 }
 316
 317 /* This requires some explaining. If DNAT has taken place,
 318  * we will need to fix up the destination Ethernet address.
 319  *
 320  * There are two cases to consider:
 321  * 1. The packet was DNAT'ed to a device in the same bridge
 322  *    port group as it was received on. We can still bridge
 323  *    the packet.
 324  * 2. The packet was DNAT'ed to a different device, either
 325  *    a non-bridged device or another bridge port group.
 326  *    The packet will need to be routed.
 327  *
 328  * The correct way of distinguishing between these two cases is to
 329  * call ip_route_input() and to look at skb->dst->dev, which is
 330  * changed to the destination device if ip_route_input() succeeds.
 331  *
 332  * Let's first consider the case that ip_route_input() succeeds:
 333  *
 334  * If the output device equals the logical bridge device the packet
 335  * came in on, we can consider this bridging. The corresponding MAC
 336  * address will be obtained in br_nf_pre_routing_finish_bridge.
 337  * Otherwise, the packet is considered to be routed and we just
 338  * change the destination MAC address so that the packet will
 339  * later be passed up to the IP stack to be routed. For a redirected
 340  * packet, ip_route_input() will give back the localhost as output device,
 341  * which differs from the bridge device.
 342  *
 343  * Let's now consider the case that ip_route_input() fails:
 344  *
 345  * This can be because the destination address is martian, in which case
 346  * the packet will be dropped.
 347  * If IP forwarding is disabled, ip_route_input() will fail, while
 348  * ip_route_output_key() can return success. The source
 349  * address for ip_route_output_key() is set to zero, so ip_route_output_key()
 350  * thinks we're handling a locally generated packet and won't care
 351  * if IP forwarding is enabled. If the output device equals the logical bridge
 352  * device, we proceed as if ip_route_input() succeeded. If it differs from the
 353  * logical bridge port or if ip_route_output_key() fails we drop the packet.
 354  */
 355 static int br_nf_pre_routing_finish(struct sk_buff *skb)
 356 {
 357         struct net_device *dev = skb->dev;
 358         struct iphdr *iph = ip_hdr(skb);
 359         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 360         struct rtable *rt;
 361         int err;
 362         int frag_max_size;
 363
 364         frag_max_size = IPCB(skb)->frag_max_size;
 365         BR_INPUT_SKB_CB(skb)->frag_max_size = frag_max_size;
 366
 367         if (nf_bridge->mask & BRNF_PKT_TYPE) {
 368                 skb->pkt_type = PACKET_OTHERHOST;
 369                 nf_bridge->mask ^= BRNF_PKT_TYPE;
 370         }
 371         nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
 372         if (dnat_took_place(skb)) {
 373                 if ((err = ip_route_input(skb, iph->daddr, iph->saddr, iph->tos, dev))) {
 374                         struct in_device *in_dev = __in_dev_get_rcu(dev);
 375
 376                         /* If err equals -EHOSTUNREACH the error is due to a
 377                          * martian destination or due to the fact that
 378                          * forwarding is disabled. For most martian packets,
 379                          * ip_route_output_key() will fail. It won't fail for 2 types of
 380                          * martian destinations: loopback destinations and destination
 381                          * 0.0.0.0. In both cases the packet will be dropped because the
 382                          * destination is the loopback device and not the bridge. */
 383                         if (err != -EHOSTUNREACH || !in_dev || IN_DEV_FORWARD(in_dev))
 384                                 goto free_skb;
 385
 386                         rt = ip_route_output(dev_net(dev), iph->daddr, 0,
 387                                              RT_TOS(iph->tos), 0);
 388                         if (!IS_ERR(rt)) {
 389                                 /* - Bridged-and-DNAT'ed traffic doesn't
 390                                  *   require ip_forwarding. */
 391                                 if (rt->dst.dev == dev) {
 392                                         skb_dst_set(skb, &rt->dst);
 393                                         goto bridged_dnat;
 394                                 }
 395                                 ip_rt_put(rt);
 396                         }
 397 free_skb:
 398                         kfree_skb(skb);
 399                         return 0;
 400                 } else {
 401                         if (skb_dst(skb)->dev == dev) {
 402 bridged_dnat:
 403                                 skb->dev = nf_bridge->physindev;
 404                                 nf_bridge_update_protocol(skb);
 405                                 nf_bridge_push_encap_header(skb);
 406                                 NF_HOOK_THRESH(NFPROTO_BRIDGE,
 407                                                NF_BR_PRE_ROUTING,
 408                                                skb, skb->dev, NULL,
 409                                                br_nf_pre_routing_finish_bridge,
 410                                                1);
 411                                 return 0;
 412                         }
 413                         ether_addr_copy(eth_hdr(skb)->h_dest, dev->dev_addr);
 414                         skb->pkt_type = PACKET_HOST;
 415                 }
 416         } else {
 417                 rt = bridge_parent_rtable(nf_bridge->physindev);
 418                 if (!rt) {
 419                         kfree_skb(skb);
 420                         return 0;
 421                 }
 422                 skb_dst_set_noref(skb, &rt->dst);
 423         }
 424
 425         skb->dev = nf_bridge->physindev;
 426         nf_bridge_update_protocol(skb);
 427         nf_bridge_push_encap_header(skb);
 428         NF_HOOK_THRESH(NFPROTO_BRIDGE, NF_BR_PRE_ROUTING, skb, skb->dev, NULL,
 429                        br_handle_frame_finish, 1);
 430
 431         return 0;
 432 }
 433
 434 static struct net_device *brnf_get_logical_dev(struct sk_buff *skb, const struct net_device *dev)
 435 {
 436         struct net_device *vlan, *br;
 437
 438         br = bridge_parent(dev);
 439         if (brnf_pass_vlan_indev == 0 || !skb_vlan_tag_present(skb))
 440                 return br;
 441
 442         vlan = __vlan_find_dev_deep_rcu(br, skb->vlan_proto,
 443                                     skb_vlan_tag_get(skb) & VLAN_VID_MASK);
 444
 445         return vlan ? vlan : br;
 446 }
 447
 448 /* Some common code for IPv4/IPv6 */
 449 static struct net_device *setup_pre_routing(struct sk_buff *skb)
 450 {
 451         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 452
 453         if (skb->pkt_type == PACKET_OTHERHOST) {
 454                 skb->pkt_type = PACKET_HOST;
 455                 nf_bridge->mask |= BRNF_PKT_TYPE;
 456         }
 457
 458         nf_bridge->mask |= BRNF_NF_BRIDGE_PREROUTING;
 459         nf_bridge->physindev = skb->dev;
 460         skb->dev = brnf_get_logical_dev(skb, skb->dev);
 461         if (skb->protocol == htons(ETH_P_8021Q))
 462                 nf_bridge->mask |= BRNF_8021Q;
 463         else if (skb->protocol == htons(ETH_P_PPP_SES))
 464                 nf_bridge->mask |= BRNF_PPPoE;
 465
 466         /* Must drop socket now because of tproxy. */
 467         skb_orphan(skb);
 468         return skb->dev;
 469 }
 470
 471 /* We only check the length. A bridge shouldn't do any hop-by-hop stuff anyway */
 472 static int check_hbh_len(struct sk_buff *skb)
 473 {
 474         unsigned char *raw = (u8 *)(ipv6_hdr(skb) + 1);
 475         u32 pkt_len;
 476         const unsigned char *nh = skb_network_header(skb);
 477         int off = raw - nh;
 478         int len = (raw[1] + 1) << 3;
 479
 480         if ((raw + len) - skb->data > skb_headlen(skb))
 481                 goto bad;
 482
 483         off += 2;
 484         len -= 2;
 485
 486         while (len > 0) {
 487                 int optlen = nh[off + 1] + 2;
 488
 489                 switch (nh[off]) {
 490                 case IPV6_TLV_PAD1:
 491                         optlen = 1;
 492                         break;
 493
 494                 case IPV6_TLV_PADN:
 495                         break;
 496
 497                 case IPV6_TLV_JUMBO:
 498                         if (nh[off + 1] != 4 || (off & 3) != 2)
 499                                 goto bad;
 500                         pkt_len = ntohl(*(__be32 *) (nh + off + 2));
 501                         if (pkt_len <= IPV6_MAXPLEN ||
 502                             ipv6_hdr(skb)->payload_len)
 503                                 goto bad;
 504                         if (pkt_len > skb->len - sizeof(struct ipv6hdr))
 505                                 goto bad;
 506                         if (pskb_trim_rcsum(skb,
 507                                             pkt_len + sizeof(struct ipv6hdr)))
 508                                 goto bad;
 509                         nh = skb_network_header(skb);
 510                         break;
 511                 default:
 512                         if (optlen > len)
 513                                 goto bad;
 514                         break;
 515                 }
 516                 off += optlen;
 517                 len -= optlen;
 518         }
 519         if (len == 0)
 520                 return 0;
 521 bad:
 522         return -1;
 523
 524 }
 525
 526 /* Replicate the checks that IPv6 does on packet reception and pass the packet
 527  * to ip6tables, which doesn't support NAT, so things are fairly simple. */
 528 static unsigned int br_nf_pre_routing_ipv6(const struct nf_hook_ops *ops,
 529                                            struct sk_buff *skb,
 530                                            const struct net_device *in,
 531                                            const struct net_device *out,
 532                                            int (*okfn)(struct sk_buff *))
 533 {
 534         const struct ipv6hdr *hdr;
 535         u32 pkt_len;
 536
 537         if (skb->len < sizeof(struct ipv6hdr))
 538                 return NF_DROP;
 539
 540         if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
 541                 return NF_DROP;
 542
 543         hdr = ipv6_hdr(skb);
 544
 545         if (hdr->version != 6)
 546                 return NF_DROP;
 547
 548         pkt_len = ntohs(hdr->payload_len);
 549
 550         if (pkt_len || hdr->nexthdr != NEXTHDR_HOP) {
 551                 if (pkt_len + sizeof(struct ipv6hdr) > skb->len)
 552                         return NF_DROP;
 553                 if (pskb_trim_rcsum(skb, pkt_len + sizeof(struct ipv6hdr)))
 554                         return NF_DROP;
 555         }
 556         if (hdr->nexthdr == NEXTHDR_HOP && check_hbh_len(skb))
 557                 return NF_DROP;
 558
 559         nf_bridge_put(skb->nf_bridge);
 560         if (!nf_bridge_alloc(skb))
 561                 return NF_DROP;
 562         if (!setup_pre_routing(skb))
 563                 return NF_DROP;
 564
 565         skb->protocol = htons(ETH_P_IPV6);
 566         NF_HOOK(NFPROTO_IPV6, NF_INET_PRE_ROUTING, skb, skb->dev, NULL,
 567                 br_nf_pre_routing_finish_ipv6);
 568
 569         return NF_STOLEN;
 570 }
 571
 572 /* Direct IPv6 traffic to br_nf_pre_routing_ipv6.
 573  * Replicate the checks that IPv4 does on packet reception.
 574  * Set skb->dev to the bridge device (i.e. parent of the
 575  * receiving device) to make netfilter happy, the REDIRECT
 576  * target in particular.  Save the original destination IP
 577  * address to be able to detect DNAT afterwards. */
 578 static unsigned int br_nf_pre_routing(const struct nf_hook_ops *ops,
 579                                       struct sk_buff *skb,
 580                                       const struct net_device *in,
 581                                       const struct net_device *out,
 582                                       int (*okfn)(struct sk_buff *))
 583 {
 584         struct net_bridge_port *p;
 585         struct net_bridge *br;
 586         __u32 len = nf_bridge_encap_header_len(skb);
 587
 588         if (unlikely(!pskb_may_pull(skb, len)))
 589                 return NF_DROP;
 590
 591         p = br_port_get_rcu(in);
 592         if (p == NULL)
 593                 return NF_DROP;
 594         br = p->br;
 595
 596         if (IS_IPV6(skb) || IS_VLAN_IPV6(skb) || IS_PPPOE_IPV6(skb)) {
 597                 if (!brnf_call_ip6tables && !br->nf_call_ip6tables)
 598                         return NF_ACCEPT;
 599
 600                 nf_bridge_pull_encap_header_rcsum(skb);
 601                 return br_nf_pre_routing_ipv6(ops, skb, in, out, okfn);
 602         }
 603
 604         if (!brnf_call_iptables && !br->nf_call_iptables)
 605                 return NF_ACCEPT;
 606
 607         if (!IS_IP(skb) && !IS_VLAN_IP(skb) && !IS_PPPOE_IP(skb))
 608                 return NF_ACCEPT;
 609
 610         nf_bridge_pull_encap_header_rcsum(skb);
 611
 612         if (br_parse_ip_options(skb))
 613                 return NF_DROP;
 614
 615         nf_bridge_put(skb->nf_bridge);
 616         if (!nf_bridge_alloc(skb))
 617                 return NF_DROP;
 618         if (!setup_pre_routing(skb))
 619                 return NF_DROP;
 620         store_orig_dstaddr(skb);
 621         skb->protocol = htons(ETH_P_IP);
 622
 623         NF_HOOK(NFPROTO_IPV4, NF_INET_PRE_ROUTING, skb, skb->dev, NULL,
 624                 br_nf_pre_routing_finish);
 625
 626         return NF_STOLEN;
 627 }
 628
 629
 630 /* PF_BRIDGE/LOCAL_IN ************************************************/
 631 /* The packet is locally destined, which requires a real
 632  * dst_entry, so detach the fake one.  On the way up, the
 633  * packet would pass through PRE_ROUTING again (which already
 634  * took place when the packet entered the bridge), but we
 635  * register an IPv4 PRE_ROUTING 'sabotage' hook that will
 636  * prevent this from happening. */
 637 static unsigned int br_nf_local_in(const struct nf_hook_ops *ops,
 638                                    struct sk_buff *skb,
 639                                    const struct net_device *in,
 640                                    const struct net_device *out,
 641                                    int (*okfn)(struct sk_buff *))
 642 {
 643         br_drop_fake_rtable(skb);
 644         return NF_ACCEPT;
 645 }
 646
 647 /* PF_BRIDGE/FORWARD *************************************************/
 648 static int br_nf_forward_finish(struct sk_buff *skb)
 649 {
 650         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 651         struct net_device *in;
 652
 653         if (!IS_ARP(skb) && !IS_VLAN_ARP(skb)) {
 654                 in = nf_bridge->physindev;
 655                 if (nf_bridge->mask & BRNF_PKT_TYPE) {
 656                         skb->pkt_type = PACKET_OTHERHOST;
 657                         nf_bridge->mask ^= BRNF_PKT_TYPE;
 658                 }
 659                 nf_bridge_update_protocol(skb);
 660         } else {
 661                 in = *((struct net_device **)(skb->cb));
 662         }
 663         nf_bridge_push_encap_header(skb);
 664
 665         NF_HOOK_THRESH(NFPROTO_BRIDGE, NF_BR_FORWARD, skb, in,
 666                        skb->dev, br_forward_finish, 1);
 667         return 0;
 668 }
 669
 670
 671 /* This is the 'purely bridged' case.  For IP, we pass the packet to
 672  * netfilter with indev and outdev set to the bridge device,
 673  * but we are still able to filter on the 'real' indev/outdev
 674  * because of the physdev module. For ARP, indev and outdev are the
 675  * bridge ports. */
 676 static unsigned int br_nf_forward_ip(const struct nf_hook_ops *ops,
 677                                      struct sk_buff *skb,
 678                                      const struct net_device *in,
 679                                      const struct net_device *out,
 680                                      int (*okfn)(struct sk_buff *))
 681 {
 682         struct nf_bridge_info *nf_bridge;
 683         struct net_device *parent;
 684         u_int8_t pf;
 685
 686         if (!skb->nf_bridge)
 687                 return NF_ACCEPT;
 688
 689         /* Need exclusive nf_bridge_info since we might have multiple
 690          * different physoutdevs. */
 691         if (!nf_bridge_unshare(skb))
 692                 return NF_DROP;
 693
 694         parent = bridge_parent(out);
 695         if (!parent)
 696                 return NF_DROP;
 697
 698         if (IS_IP(skb) || IS_VLAN_IP(skb) || IS_PPPOE_IP(skb))
 699                 pf = NFPROTO_IPV4;
 700         else if (IS_IPV6(skb) || IS_VLAN_IPV6(skb) || IS_PPPOE_IPV6(skb))
 701                 pf = NFPROTO_IPV6;
 702         else
 703                 return NF_ACCEPT;
 704
 705         nf_bridge_pull_encap_header(skb);
 706
 707         nf_bridge = skb->nf_bridge;
 708         if (skb->pkt_type == PACKET_OTHERHOST) {
 709                 skb->pkt_type = PACKET_HOST;
 710                 nf_bridge->mask |= BRNF_PKT_TYPE;
 711         }
 712
 713         if (pf == NFPROTO_IPV4 && br_parse_ip_options(skb))
 714                 return NF_DROP;
 715
 716         /* The physdev module checks on this */
 717         nf_bridge->mask |= BRNF_BRIDGED;
 718         nf_bridge->physoutdev = skb->dev;
 719         if (pf == NFPROTO_IPV4)
 720                 skb->protocol = htons(ETH_P_IP);
 721         else
 722                 skb->protocol = htons(ETH_P_IPV6);
 723
 724         NF_HOOK(pf, NF_INET_FORWARD, skb, brnf_get_logical_dev(skb, in), parent,
 725                 br_nf_forward_finish);
 726
 727         return NF_STOLEN;
 728 }
 729
 730 static unsigned int br_nf_forward_arp(const struct nf_hook_ops *ops,
 731                                       struct sk_buff *skb,
 732                                       const struct net_device *in,
 733                                       const struct net_device *out,
 734                                       int (*okfn)(struct sk_buff *))
 735 {
 736         struct net_bridge_port *p;
 737         struct net_bridge *br;
 738         struct net_device **d = (struct net_device **)(skb->cb);
 739
 740         p = br_port_get_rcu(out);
 741         if (p == NULL)
 742                 return NF_ACCEPT;
 743         br = p->br;
 744
 745         if (!brnf_call_arptables && !br->nf_call_arptables)
 746                 return NF_ACCEPT;
 747
 748         if (!IS_ARP(skb)) {
 749                 if (!IS_VLAN_ARP(skb))
 750                         return NF_ACCEPT;
 751                 nf_bridge_pull_encap_header(skb);
 752         }
 753
 754         if (arp_hdr(skb)->ar_pln != 4) {
 755                 if (IS_VLAN_ARP(skb))
 756                         nf_bridge_push_encap_header(skb);
 757                 return NF_ACCEPT;
 758         }
 759         *d = (struct net_device *)in;
 760         NF_HOOK(NFPROTO_ARP, NF_ARP_FORWARD, skb, (struct net_device *)in,
 761                 (struct net_device *)out, br_nf_forward_finish);
 762
 763         return NF_STOLEN;
 764 }
 765
 766 #if IS_ENABLED(CONFIG_NF_DEFRAG_IPV4)
 767 static bool nf_bridge_copy_header(struct sk_buff *skb)
 768 {
 769         int err;
 770         unsigned int header_size;
 771
 772         nf_bridge_update_protocol(skb);
 773         header_size = ETH_HLEN + nf_bridge_encap_header_len(skb);
 774         err = skb_cow_head(skb, header_size);
 775         if (err)
 776                 return false;
 777
 778         skb_copy_to_linear_data_offset(skb, -header_size,
 779                                        skb->nf_bridge->data, header_size);
 780         __skb_push(skb, nf_bridge_encap_header_len(skb));
 781         return true;
 782 }
 783
 784 static int br_nf_push_frag_xmit(struct sk_buff *skb)
 785 {
 786         if (!nf_bridge_copy_header(skb)) {
 787                 kfree_skb(skb);
 788                 return 0;
 789         }
 790
 791         return br_dev_queue_push_xmit(skb);
 792 }
 793
 794 static int br_nf_dev_queue_xmit(struct sk_buff *skb)
 795 {
 796         int ret;
 797         int frag_max_size;
 798
 799         /* This is wrong! We should preserve the original fragment
 800          * boundaries by preserving frag_list rather than refragmenting.
 801          */
 802         if (skb->protocol == htons(ETH_P_IP) &&
 803             skb->len + nf_bridge_mtu_reduction(skb) > skb->dev->mtu &&
 804             !skb_is_gso(skb)) {
 805                 frag_max_size = BR_INPUT_SKB_CB(skb)->frag_max_size;
 806                 if (br_parse_ip_options(skb))
 807                         /* Drop invalid packet */
 808                         return NF_DROP;
 809                 IPCB(skb)->frag_max_size = frag_max_size;
 810                 ret = ip_fragment(skb, br_nf_push_frag_xmit);
 811         } else
 812                 ret = br_dev_queue_push_xmit(skb);
 813
 814         return ret;
 815 }
 816 #else
 817 static int br_nf_dev_queue_xmit(struct sk_buff *skb)
 818 {
 819         return br_dev_queue_push_xmit(skb);
 820 }
 821 #endif
 822
 823 /* PF_BRIDGE/POST_ROUTING ********************************************/
 824 static unsigned int br_nf_post_routing(const struct nf_hook_ops *ops,
 825                                        struct sk_buff *skb,
 826                                        const struct net_device *in,
 827                                        const struct net_device *out,
 828                                        int (*okfn)(struct sk_buff *))
 829 {
 830         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 831         struct net_device *realoutdev = bridge_parent(skb->dev);
 832         u_int8_t pf;
 833
 834         if (!nf_bridge || !(nf_bridge->mask & BRNF_BRIDGED))
 835                 return NF_ACCEPT;
 836
 837         if (!realoutdev)
 838                 return NF_DROP;
 839
 840         if (IS_IP(skb) || IS_VLAN_IP(skb) || IS_PPPOE_IP(skb))
 841                 pf = NFPROTO_IPV4;
 842         else if (IS_IPV6(skb) || IS_VLAN_IPV6(skb) || IS_PPPOE_IPV6(skb))
 843                 pf = NFPROTO_IPV6;
 844         else
 845                 return NF_ACCEPT;
 846
 847         /* We assume any code from br_dev_queue_push_xmit onwards doesn't care
 848          * about the value of skb->pkt_type. */
 849         if (skb->pkt_type == PACKET_OTHERHOST) {
 850                 skb->pkt_type = PACKET_HOST;
 851                 nf_bridge->mask |= BRNF_PKT_TYPE;
 852         }
 853
 854         nf_bridge_pull_encap_header(skb);
 855         nf_bridge_save_header(skb);
 856         if (pf == NFPROTO_IPV4)
 857                 skb->protocol = htons(ETH_P_IP);
 858         else
 859                 skb->protocol = htons(ETH_P_IPV6);
 860
 861         NF_HOOK(pf, NF_INET_POST_ROUTING, skb, NULL, realoutdev,
 862                 br_nf_dev_queue_xmit);
 863
 864         return NF_STOLEN;
 865 }
 866
 867 /* IP/SABOTAGE *****************************************************/
 868 /* Don't hand locally destined packets to PF_INET(6)/PRE_ROUTING
 869  * for the second time. */
 870 static unsigned int ip_sabotage_in(const struct nf_hook_ops *ops,
 871                                    struct sk_buff *skb,
 872                                    const struct net_device *in,
 873                                    const struct net_device *out,
 874                                    int (*okfn)(struct sk_buff *))
 875 {
 876         if (skb->nf_bridge &&
 877             !(skb->nf_bridge->mask & BRNF_NF_BRIDGE_PREROUTING)) {
 878                 return NF_STOP;
 879         }
 880
 881         return NF_ACCEPT;
 882 }
 883
 884 void br_netfilter_enable(void)
 885 {
 886 }
 887 EXPORT_SYMBOL_GPL(br_netfilter_enable);
 888
 889 /* For br_nf_post_routing, we need (prio = NF_BR_PRI_LAST), because
 890  * br_dev_queue_push_xmit is called afterwards */
 891 static struct nf_hook_ops br_nf_ops[] __read_mostly = {
 892         {
 893                 .hook = br_nf_pre_routing,
 894                 .owner = THIS_MODULE,
 895                 .pf = NFPROTO_BRIDGE,
 896                 .hooknum = NF_BR_PRE_ROUTING,
 897                 .priority = NF_BR_PRI_BRNF,
 898         },
 899         {
 900                 .hook = br_nf_local_in,
 901                 .owner = THIS_MODULE,
 902                 .pf = NFPROTO_BRIDGE,
 903                 .hooknum = NF_BR_LOCAL_IN,
 904                 .priority = NF_BR_PRI_BRNF,
 905         },
 906         {
 907                 .hook = br_nf_forward_ip,
 908                 .owner = THIS_MODULE,
 909                 .pf = NFPROTO_BRIDGE,
 910                 .hooknum = NF_BR_FORWARD,
 911                 .priority = NF_BR_PRI_BRNF - 1,
 912         },
 913         {
 914                 .hook = br_nf_forward_arp,
 915                 .owner = THIS_MODULE,
 916                 .pf = NFPROTO_BRIDGE,
 917                 .hooknum = NF_BR_FORWARD,
 918                 .priority = NF_BR_PRI_BRNF,
 919         },
 920         {
 921                 .hook = br_nf_post_routing,
 922                 .owner = THIS_MODULE,
 923                 .pf = NFPROTO_BRIDGE,
 924                 .hooknum = NF_BR_POST_ROUTING,
 925                 .priority = NF_BR_PRI_LAST,
 926         },
 927         {
 928                 .hook = ip_sabotage_in,
 929                 .owner = THIS_MODULE,
 930                 .pf = NFPROTO_IPV4,
 931                 .hooknum = NF_INET_PRE_ROUTING,
 932                 .priority = NF_IP_PRI_FIRST,
 933         },
 934         {
 935                 .hook = ip_sabotage_in,
 936                 .owner = THIS_MODULE,
 937                 .pf = NFPROTO_IPV6,
 938                 .hooknum = NF_INET_PRE_ROUTING,
 939                 .priority = NF_IP6_PRI_FIRST,
 940         },
 941 };
 942
 943 #ifdef CONFIG_SYSCTL
 944 static
 945 int brnf_sysctl_call_tables(struct ctl_table *ctl, int write,
 946                             void __user *buffer, size_t *lenp, loff_t *ppos)
 947 {
 948         int ret;
 949
 950         ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
 951
 952         if (write && *(int *)(ctl->data))
 953                 *(int *)(ctl->data) = 1;
 954         return ret;
 955 }
 956
 957 static struct ctl_table brnf_table[] = {
 958         {
 959                 .procname       = "bridge-nf-call-arptables",
 960                 .data           = &brnf_call_arptables,
 961                 .maxlen         = sizeof(int),
 962                 .mode           = 0644,
 963                 .proc_handler   = brnf_sysctl_call_tables,
 964         },
 965         {
 966                 .procname       = "bridge-nf-call-iptables",
 967                 .data           = &brnf_call_iptables,
 968                 .maxlen         = sizeof(int),
 969                 .mode           = 0644,
 970                 .proc_handler   = brnf_sysctl_call_tables,
 971         },
 972         {
 973                 .procname       = "bridge-nf-call-ip6tables",
 974                 .data           = &brnf_call_ip6tables,
 975                 .maxlen         = sizeof(int),
 976                 .mode           = 0644,
 977                 .proc_handler   = brnf_sysctl_call_tables,
 978         },
 979         {
 980                 .procname       = "bridge-nf-filter-vlan-tagged",
 981                 .data           = &brnf_filter_vlan_tagged,
 982                 .maxlen         = sizeof(int),
 983                 .mode           = 0644,
 984                 .proc_handler   = brnf_sysctl_call_tables,
 985         },
 986         {
 987                 .procname       = "bridge-nf-filter-pppoe-tagged",
 988                 .data           = &brnf_filter_pppoe_tagged,
 989                 .maxlen         = sizeof(int),
 990                 .mode           = 0644,
 991                 .proc_handler   = brnf_sysctl_call_tables,
 992         },
 993         {
 994                 .procname       = "bridge-nf-pass-vlan-input-dev",
 995                 .data           = &brnf_pass_vlan_indev,
 996                 .maxlen         = sizeof(int),
 997                 .mode           = 0644,
 998                 .proc_handler   = brnf_sysctl_call_tables,
 999         },
1000         { }
1001 };
1002 #endif
1003
1004 static int __init br_netfilter_init(void)
1005 {
1006         int ret;
1007
1008         ret = nf_register_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
1009         if (ret < 0)
1010                 return ret;
1011
1012 #ifdef CONFIG_SYSCTL
1013         brnf_sysctl_header = register_net_sysctl(&init_net, "net/bridge", brnf_table);
1014         if (brnf_sysctl_header == NULL) {
1015                 printk(KERN_WARNING
1016                        "br_netfilter: can't register to sysctl.\n");
1017                 nf_unregister_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
1018                 return -ENOMEM;
1019         }
1020 #endif
1021         printk(KERN_NOTICE "Bridge firewalling registered\n");
1022         return 0;
1023 }
1024
1025 static void __exit br_netfilter_fini(void)
1026 {
1027         nf_unregister_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
1028 #ifdef CONFIG_SYSCTL
1029         unregister_net_sysctl_table(brnf_sysctl_header);
1030 #endif
1031 }
1032
1033 module_init(br_netfilter_init);
1034 module_exit(br_netfilter_fini);
1035
1036 MODULE_LICENSE("GPL");
1037 MODULE_AUTHOR("Lennert Buytenhek <buytenh@gnu.org>");
1038 MODULE_AUTHOR("Bart De Schuymer <bdschuym@pandora.be>");
1039 MODULE_DESCRIPTION("Linux ethernet netfilter firewall bridge");