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
  39 #include <asm/uaccess.h>
  40 #include "br_private.h"
  41 #ifdef CONFIG_SYSCTL
  42 #include <linux/sysctl.h>
  43 #endif
  44
  45 #define skb_origaddr(skb)        (((struct bridge_skb_cb *) \
  46                                  (skb->nf_bridge->data))->daddr.ipv4)
  47 #define store_orig_dstaddr(skb)  (skb_origaddr(skb) = ip_hdr(skb)->daddr)
  48 #define dnat_took_place(skb)     (skb_origaddr(skb) != ip_hdr(skb)->daddr)
  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 #else
  58 #define brnf_filter_vlan_tagged 0
  59 #define brnf_filter_pppoe_tagged 0
  60 #endif
  61
  62 static inline __be16 vlan_proto(const struct sk_buff *skb)
  63 {
  64         return vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
  65 }
  66
  67 #define IS_VLAN_IP(skb) \
  68         (skb->protocol == htons(ETH_P_8021Q) && \
  69          vlan_proto(skb) == htons(ETH_P_IP) &&  \
  70          brnf_filter_vlan_tagged)
  71
  72 #define IS_VLAN_IPV6(skb) \
  73         (skb->protocol == htons(ETH_P_8021Q) && \
  74          vlan_proto(skb) == htons(ETH_P_IPV6) &&\
  75          brnf_filter_vlan_tagged)
  76
  77 #define IS_VLAN_ARP(skb) \
  78         (skb->protocol == htons(ETH_P_8021Q) && \
  79          vlan_proto(skb) == htons(ETH_P_ARP) && \
  80          brnf_filter_vlan_tagged)
  81
  82 static inline __be16 pppoe_proto(const struct sk_buff *skb)
  83 {
  84         return *((__be16 *)(skb_mac_header(skb) + ETH_HLEN +
  85                             sizeof(struct pppoe_hdr)));
  86 }
  87
  88 #define IS_PPPOE_IP(skb) \
  89         (skb->protocol == htons(ETH_P_PPP_SES) && \
  90          pppoe_proto(skb) == htons(PPP_IP) && \
  91          brnf_filter_pppoe_tagged)
  92
  93 #define IS_PPPOE_IPV6(skb) \
  94         (skb->protocol == htons(ETH_P_PPP_SES) && \
  95          pppoe_proto(skb) == htons(PPP_IPV6) && \
  96          brnf_filter_pppoe_tagged)
  97
  98 static void fake_update_pmtu(struct dst_entry *dst, u32 mtu)
  99 {
 100 }
 101
 102 static struct dst_ops fake_dst_ops = {
 103         .family =               AF_INET,
 104         .protocol =             cpu_to_be16(ETH_P_IP),
 105         .update_pmtu =          fake_update_pmtu,
 106         .entries =              ATOMIC_INIT(0),
 107 };
 108
 109 /*
 110  * Initialize bogus route table used to keep netfilter happy.
 111  * Currently, we fill in the PMTU entry because netfilter
 112  * refragmentation needs it, and the rt_flags entry because
 113  * ipt_REJECT needs it.  Future netfilter modules might
 114  * require us to fill additional fields.
 115  */
 116 void br_netfilter_rtable_init(struct net_bridge *br)
 117 {
 118         struct rtable *rt = &br->fake_rtable;
 119
 120         atomic_set(&rt->u.dst.__refcnt, 1);
 121         rt->u.dst.dev = br->dev;
 122         rt->u.dst.path = &rt->u.dst;
 123         rt->u.dst.metrics[RTAX_MTU - 1] = 1500;
 124         rt->u.dst.flags = DST_NOXFRM;
 125         rt->u.dst.ops = &fake_dst_ops;
 126 }
 127
 128 static inline struct rtable *bridge_parent_rtable(const struct net_device *dev)
 129 {
 130         struct net_bridge_port *port = rcu_dereference(dev->br_port);
 131
 132         return port ? &port->br->fake_rtable : NULL;
 133 }
 134
 135 static inline struct net_device *bridge_parent(const struct net_device *dev)
 136 {
 137         struct net_bridge_port *port = rcu_dereference(dev->br_port);
 138
 139         return port ? port->br->dev : NULL;
 140 }
 141
 142 static inline struct nf_bridge_info *nf_bridge_alloc(struct sk_buff *skb)
 143 {
 144         skb->nf_bridge = kzalloc(sizeof(struct nf_bridge_info), GFP_ATOMIC);
 145         if (likely(skb->nf_bridge))
 146                 atomic_set(&(skb->nf_bridge->use), 1);
 147
 148         return skb->nf_bridge;
 149 }
 150
 151 static inline struct nf_bridge_info *nf_bridge_unshare(struct sk_buff *skb)
 152 {
 153         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 154
 155         if (atomic_read(&nf_bridge->use) > 1) {
 156                 struct nf_bridge_info *tmp = nf_bridge_alloc(skb);
 157
 158                 if (tmp) {
 159                         memcpy(tmp, nf_bridge, sizeof(struct nf_bridge_info));
 160                         atomic_set(&tmp->use, 1);
 161                         nf_bridge_put(nf_bridge);
 162                 }
 163                 nf_bridge = tmp;
 164         }
 165         return nf_bridge;
 166 }
 167
 168 static inline void nf_bridge_push_encap_header(struct sk_buff *skb)
 169 {
 170         unsigned int len = nf_bridge_encap_header_len(skb);
 171
 172         skb_push(skb, len);
 173         skb->network_header -= len;
 174 }
 175
 176 static inline void nf_bridge_pull_encap_header(struct sk_buff *skb)
 177 {
 178         unsigned int len = nf_bridge_encap_header_len(skb);
 179
 180         skb_pull(skb, len);
 181         skb->network_header += len;
 182 }
 183
 184 static inline void nf_bridge_pull_encap_header_rcsum(struct sk_buff *skb)
 185 {
 186         unsigned int len = nf_bridge_encap_header_len(skb);
 187
 188         skb_pull_rcsum(skb, len);
 189         skb->network_header += len;
 190 }
 191
 192 static inline void nf_bridge_save_header(struct sk_buff *skb)
 193 {
 194         int header_size = ETH_HLEN + nf_bridge_encap_header_len(skb);
 195
 196         skb_copy_from_linear_data_offset(skb, -header_size,
 197                                          skb->nf_bridge->data, header_size);
 198 }
 199
 200 static inline void nf_bridge_update_protocol(struct sk_buff *skb)
 201 {
 202         if (skb->nf_bridge->mask & BRNF_8021Q)
 203                 skb->protocol = htons(ETH_P_8021Q);
 204         else if (skb->nf_bridge->mask & BRNF_PPPoE)
 205                 skb->protocol = htons(ETH_P_PPP_SES);
 206 }
 207
 208 /* Fill in the header for fragmented IP packets handled by
 209  * the IPv4 connection tracking code.
 210  */
 211 int nf_bridge_copy_header(struct sk_buff *skb)
 212 {
 213         int err;
 214         unsigned int header_size;
 215
 216         nf_bridge_update_protocol(skb);
 217         header_size = ETH_HLEN + nf_bridge_encap_header_len(skb);
 218         err = skb_cow_head(skb, header_size);
 219         if (err)
 220                 return err;
 221
 222         skb_copy_to_linear_data_offset(skb, -header_size,
 223                                        skb->nf_bridge->data, header_size);
 224         __skb_push(skb, nf_bridge_encap_header_len(skb));
 225         return 0;
 226 }
 227
 228 /* PF_BRIDGE/PRE_ROUTING *********************************************/
 229 /* Undo the changes made for ip6tables PREROUTING and continue the
 230  * bridge PRE_ROUTING hook. */
 231 static int br_nf_pre_routing_finish_ipv6(struct sk_buff *skb)
 232 {
 233         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 234         struct rtable *rt;
 235
 236         if (nf_bridge->mask & BRNF_PKT_TYPE) {
 237                 skb->pkt_type = PACKET_OTHERHOST;
 238                 nf_bridge->mask ^= BRNF_PKT_TYPE;
 239         }
 240         nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
 241
 242         rt = bridge_parent_rtable(nf_bridge->physindev);
 243         if (!rt) {
 244                 kfree_skb(skb);
 245                 return 0;
 246         }
 247         skb_dst_set_noref(skb, &rt->u.dst);
 248
 249         skb->dev = nf_bridge->physindev;
 250         nf_bridge_update_protocol(skb);
 251         nf_bridge_push_encap_header(skb);
 252         NF_HOOK_THRESH(NFPROTO_BRIDGE, NF_BR_PRE_ROUTING, skb, skb->dev, NULL,
 253                        br_handle_frame_finish, 1);
 254
 255         return 0;
 256 }
 257
 258 /* Obtain the correct destination MAC address, while preserving the original
 259  * source MAC address. If we already know this address, we just copy it. If we
 260  * don't, we use the neighbour framework to find out. In both cases, we make
 261  * sure that br_handle_frame_finish() is called afterwards.
 262  */
 263 static int br_nf_pre_routing_finish_bridge(struct sk_buff *skb)
 264 {
 265         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 266         struct dst_entry *dst;
 267
 268         skb->dev = bridge_parent(skb->dev);
 269         if (!skb->dev)
 270                 goto free_skb;
 271         dst = skb_dst(skb);
 272         if (dst->hh) {
 273                 neigh_hh_bridge(dst->hh, skb);
 274                 skb->dev = nf_bridge->physindev;
 275                 return br_handle_frame_finish(skb);
 276         } else if (dst->neighbour) {
 277                 /* the neighbour function below overwrites the complete
 278                  * MAC header, so we save the Ethernet source address and
 279                  * protocol number. */
 280                 skb_copy_from_linear_data_offset(skb, -(ETH_HLEN-ETH_ALEN), skb->nf_bridge->data, ETH_HLEN-ETH_ALEN);
 281                 /* tell br_dev_xmit to continue with forwarding */
 282                 nf_bridge->mask |= BRNF_BRIDGED_DNAT;
 283                 return dst->neighbour->output(skb);
 284         }
 285 free_skb:
 286         kfree_skb(skb);
 287         return 0;
 288 }
 289
 290 /* This requires some explaining. If DNAT has taken place,
 291  * we will need to fix up the destination Ethernet address.
 292  *
 293  * There are two cases to consider:
 294  * 1. The packet was DNAT'ed to a device in the same bridge
 295  *    port group as it was received on. We can still bridge
 296  *    the packet.
 297  * 2. The packet was DNAT'ed to a different device, either
 298  *    a non-bridged device or another bridge port group.
 299  *    The packet will need to be routed.
 300  *
 301  * The correct way of distinguishing between these two cases is to
 302  * call ip_route_input() and to look at skb->dst->dev, which is
 303  * changed to the destination device if ip_route_input() succeeds.
 304  *
 305  * Let's first consider the case that ip_route_input() succeeds:
 306  *
 307  * If the output device equals the logical bridge device the packet
 308  * came in on, we can consider this bridging. The corresponding MAC
 309  * address will be obtained in br_nf_pre_routing_finish_bridge.
 310  * Otherwise, the packet is considered to be routed and we just
 311  * change the destination MAC address so that the packet will
 312  * later be passed up to the IP stack to be routed. For a redirected
 313  * packet, ip_route_input() will give back the localhost as output device,
 314  * which differs from the bridge device.
 315  *
 316  * Let's now consider the case that ip_route_input() fails:
 317  *
 318  * This can be because the destination address is martian, in which case
 319  * the packet will be dropped.
 320  * If IP forwarding is disabled, ip_route_input() will fail, while
 321  * ip_route_output_key() can return success. The source
 322  * address for ip_route_output_key() is set to zero, so ip_route_output_key()
 323  * thinks we're handling a locally generated packet and won't care
 324  * if IP forwarding is enabled. If the output device equals the logical bridge
 325  * device, we proceed as if ip_route_input() succeeded. If it differs from the
 326  * logical bridge port or if ip_route_output_key() fails we drop the packet.
 327  */
 328 static int br_nf_pre_routing_finish(struct sk_buff *skb)
 329 {
 330         struct net_device *dev = skb->dev;
 331         struct iphdr *iph = ip_hdr(skb);
 332         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 333         struct rtable *rt;
 334         int err;
 335
 336         if (nf_bridge->mask & BRNF_PKT_TYPE) {
 337                 skb->pkt_type = PACKET_OTHERHOST;
 338                 nf_bridge->mask ^= BRNF_PKT_TYPE;
 339         }
 340         nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
 341         if (dnat_took_place(skb)) {
 342                 if ((err = ip_route_input(skb, iph->daddr, iph->saddr, iph->tos, dev))) {
 343                         struct flowi fl = {
 344                                 .nl_u = {
 345                                         .ip4_u = {
 346                                                  .daddr = iph->daddr,
 347                                                  .saddr = 0,
 348                                                  .tos = RT_TOS(iph->tos) },
 349                                 },
 350                                 .proto = 0,
 351                         };
 352                         struct in_device *in_dev = __in_dev_get_rcu(dev);
 353
 354                         /* If err equals -EHOSTUNREACH the error is due to a
 355                          * martian destination or due to the fact that
 356                          * forwarding is disabled. For most martian packets,
 357                          * ip_route_output_key() will fail. It won't fail for 2 types of
 358                          * martian destinations: loopback destinations and destination
 359                          * 0.0.0.0. In both cases the packet will be dropped because the
 360                          * destination is the loopback device and not the bridge. */
 361                         if (err != -EHOSTUNREACH || !in_dev || IN_DEV_FORWARD(in_dev))
 362                                 goto free_skb;
 363
 364                         if (!ip_route_output_key(dev_net(dev), &rt, &fl)) {
 365                                 /* - Bridged-and-DNAT'ed traffic doesn't
 366                                  *   require ip_forwarding. */
 367                                 if (((struct dst_entry *)rt)->dev == dev) {
 368                                         skb_dst_set(skb, (struct dst_entry *)rt);
 369                                         goto bridged_dnat;
 370                                 }
 371                                 dst_release((struct dst_entry *)rt);
 372                         }
 373 free_skb:
 374                         kfree_skb(skb);
 375                         return 0;
 376                 } else {
 377                         if (skb_dst(skb)->dev == dev) {
 378 bridged_dnat:
 379                                 skb->dev = nf_bridge->physindev;
 380                                 nf_bridge_update_protocol(skb);
 381                                 nf_bridge_push_encap_header(skb);
 382                                 NF_HOOK_THRESH(NFPROTO_BRIDGE,
 383                                                NF_BR_PRE_ROUTING,
 384                                                skb, skb->dev, NULL,
 385                                                br_nf_pre_routing_finish_bridge,
 386                                                1);
 387                                 return 0;
 388                         }
 389                         memcpy(eth_hdr(skb)->h_dest, dev->dev_addr, ETH_ALEN);
 390                         skb->pkt_type = PACKET_HOST;
 391                 }
 392         } else {
 393                 rt = bridge_parent_rtable(nf_bridge->physindev);
 394                 if (!rt) {
 395                         kfree_skb(skb);
 396                         return 0;
 397                 }
 398                 skb_dst_set_noref(skb, &rt->u.dst);
 399         }
 400
 401         skb->dev = nf_bridge->physindev;
 402         nf_bridge_update_protocol(skb);
 403         nf_bridge_push_encap_header(skb);
 404         NF_HOOK_THRESH(NFPROTO_BRIDGE, NF_BR_PRE_ROUTING, skb, skb->dev, NULL,
 405                        br_handle_frame_finish, 1);
 406
 407         return 0;
 408 }
 409
 410 /* Some common code for IPv4/IPv6 */
 411 static struct net_device *setup_pre_routing(struct sk_buff *skb)
 412 {
 413         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 414
 415         if (skb->pkt_type == PACKET_OTHERHOST) {
 416                 skb->pkt_type = PACKET_HOST;
 417                 nf_bridge->mask |= BRNF_PKT_TYPE;
 418         }
 419
 420         nf_bridge->mask |= BRNF_NF_BRIDGE_PREROUTING;
 421         nf_bridge->physindev = skb->dev;
 422         skb->dev = bridge_parent(skb->dev);
 423         if (skb->protocol == htons(ETH_P_8021Q))
 424                 nf_bridge->mask |= BRNF_8021Q;
 425         else if (skb->protocol == htons(ETH_P_PPP_SES))
 426                 nf_bridge->mask |= BRNF_PPPoE;
 427
 428         return skb->dev;
 429 }
 430
 431 /* We only check the length. A bridge shouldn't do any hop-by-hop stuff anyway */
 432 static int check_hbh_len(struct sk_buff *skb)
 433 {
 434         unsigned char *raw = (u8 *)(ipv6_hdr(skb) + 1);
 435         u32 pkt_len;
 436         const unsigned char *nh = skb_network_header(skb);
 437         int off = raw - nh;
 438         int len = (raw[1] + 1) << 3;
 439
 440         if ((raw + len) - skb->data > skb_headlen(skb))
 441                 goto bad;
 442
 443         off += 2;
 444         len -= 2;
 445
 446         while (len > 0) {
 447                 int optlen = nh[off + 1] + 2;
 448
 449                 switch (nh[off]) {
 450                 case IPV6_TLV_PAD0:
 451                         optlen = 1;
 452                         break;
 453
 454                 case IPV6_TLV_PADN:
 455                         break;
 456
 457                 case IPV6_TLV_JUMBO:
 458                         if (nh[off + 1] != 4 || (off & 3) != 2)
 459                                 goto bad;
 460                         pkt_len = ntohl(*(__be32 *) (nh + off + 2));
 461                         if (pkt_len <= IPV6_MAXPLEN ||
 462                             ipv6_hdr(skb)->payload_len)
 463                                 goto bad;
 464                         if (pkt_len > skb->len - sizeof(struct ipv6hdr))
 465                                 goto bad;
 466                         if (pskb_trim_rcsum(skb,
 467                                             pkt_len + sizeof(struct ipv6hdr)))
 468                                 goto bad;
 469                         nh = skb_network_header(skb);
 470                         break;
 471                 default:
 472                         if (optlen > len)
 473                                 goto bad;
 474                         break;
 475                 }
 476                 off += optlen;
 477                 len -= optlen;
 478         }
 479         if (len == 0)
 480                 return 0;
 481 bad:
 482         return -1;
 483
 484 }
 485
 486 /* Replicate the checks that IPv6 does on packet reception and pass the packet
 487  * to ip6tables, which doesn't support NAT, so things are fairly simple. */
 488 static unsigned int br_nf_pre_routing_ipv6(unsigned int hook,
 489                                            struct sk_buff *skb,
 490                                            const struct net_device *in,
 491                                            const struct net_device *out,
 492                                            int (*okfn)(struct sk_buff *))
 493 {
 494         struct ipv6hdr *hdr;
 495         u32 pkt_len;
 496
 497         if (skb->len < sizeof(struct ipv6hdr))
 498                 goto inhdr_error;
 499
 500         if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
 501                 goto inhdr_error;
 502
 503         hdr = ipv6_hdr(skb);
 504
 505         if (hdr->version != 6)
 506                 goto inhdr_error;
 507
 508         pkt_len = ntohs(hdr->payload_len);
 509
 510         if (pkt_len || hdr->nexthdr != NEXTHDR_HOP) {
 511                 if (pkt_len + sizeof(struct ipv6hdr) > skb->len)
 512                         goto inhdr_error;
 513                 if (pskb_trim_rcsum(skb, pkt_len + sizeof(struct ipv6hdr)))
 514                         goto inhdr_error;
 515         }
 516         if (hdr->nexthdr == NEXTHDR_HOP && check_hbh_len(skb))
 517                 goto inhdr_error;
 518
 519         nf_bridge_put(skb->nf_bridge);
 520         if (!nf_bridge_alloc(skb))
 521                 return NF_DROP;
 522         if (!setup_pre_routing(skb))
 523                 return NF_DROP;
 524
 525         skb->protocol = htons(ETH_P_IPV6);
 526         NF_HOOK(NFPROTO_IPV6, NF_INET_PRE_ROUTING, skb, skb->dev, NULL,
 527                 br_nf_pre_routing_finish_ipv6);
 528
 529         return NF_STOLEN;
 530
 531 inhdr_error:
 532         return NF_DROP;
 533 }
 534
 535 /* Direct IPv6 traffic to br_nf_pre_routing_ipv6.
 536  * Replicate the checks that IPv4 does on packet reception.
 537  * Set skb->dev to the bridge device (i.e. parent of the
 538  * receiving device) to make netfilter happy, the REDIRECT
 539  * target in particular.  Save the original destination IP
 540  * address to be able to detect DNAT afterwards. */
 541 static unsigned int br_nf_pre_routing(unsigned int hook, struct sk_buff *skb,
 542                                       const struct net_device *in,
 543                                       const struct net_device *out,
 544                                       int (*okfn)(struct sk_buff *))
 545 {
 546         struct iphdr *iph;
 547         __u32 len = nf_bridge_encap_header_len(skb);
 548
 549         if (unlikely(!pskb_may_pull(skb, len)))
 550                 goto out;
 551
 552         if (skb->protocol == htons(ETH_P_IPV6) || IS_VLAN_IPV6(skb) ||
 553             IS_PPPOE_IPV6(skb)) {
 554 #ifdef CONFIG_SYSCTL
 555                 if (!brnf_call_ip6tables)
 556                         return NF_ACCEPT;
 557 #endif
 558                 nf_bridge_pull_encap_header_rcsum(skb);
 559                 return br_nf_pre_routing_ipv6(hook, skb, in, out, okfn);
 560         }
 561 #ifdef CONFIG_SYSCTL
 562         if (!brnf_call_iptables)
 563                 return NF_ACCEPT;
 564 #endif
 565
 566         if (skb->protocol != htons(ETH_P_IP) && !IS_VLAN_IP(skb) &&
 567             !IS_PPPOE_IP(skb))
 568                 return NF_ACCEPT;
 569
 570         nf_bridge_pull_encap_header_rcsum(skb);
 571
 572         if (!pskb_may_pull(skb, sizeof(struct iphdr)))
 573                 goto inhdr_error;
 574
 575         iph = ip_hdr(skb);
 576         if (iph->ihl < 5 || iph->version != 4)
 577                 goto inhdr_error;
 578
 579         if (!pskb_may_pull(skb, 4 * iph->ihl))
 580                 goto inhdr_error;
 581
 582         iph = ip_hdr(skb);
 583         if (ip_fast_csum((__u8 *) iph, iph->ihl) != 0)
 584                 goto inhdr_error;
 585
 586         len = ntohs(iph->tot_len);
 587         if (skb->len < len || len < 4 * iph->ihl)
 588                 goto inhdr_error;
 589
 590         pskb_trim_rcsum(skb, len);
 591
 592         nf_bridge_put(skb->nf_bridge);
 593         if (!nf_bridge_alloc(skb))
 594                 return NF_DROP;
 595         if (!setup_pre_routing(skb))
 596                 return NF_DROP;
 597         store_orig_dstaddr(skb);
 598         skb->protocol = htons(ETH_P_IP);
 599
 600         NF_HOOK(NFPROTO_IPV4, NF_INET_PRE_ROUTING, skb, skb->dev, NULL,
 601                 br_nf_pre_routing_finish);
 602
 603         return NF_STOLEN;
 604
 605 inhdr_error:
 606 //      IP_INC_STATS_BH(IpInHdrErrors);
 607 out:
 608         return NF_DROP;
 609 }
 610
 611
 612 /* PF_BRIDGE/LOCAL_IN ************************************************/
 613 /* The packet is locally destined, which requires a real
 614  * dst_entry, so detach the fake one.  On the way up, the
 615  * packet would pass through PRE_ROUTING again (which already
 616  * took place when the packet entered the bridge), but we
 617  * register an IPv4 PRE_ROUTING 'sabotage' hook that will
 618  * prevent this from happening. */
 619 static unsigned int br_nf_local_in(unsigned int hook, struct sk_buff *skb,
 620                                    const struct net_device *in,
 621                                    const struct net_device *out,
 622                                    int (*okfn)(struct sk_buff *))
 623 {
 624         struct rtable *rt = skb_rtable(skb);
 625
 626         if (rt && rt == bridge_parent_rtable(in))
 627                 skb_dst_drop(skb);
 628
 629         return NF_ACCEPT;
 630 }
 631
 632 /* PF_BRIDGE/FORWARD *************************************************/
 633 static int br_nf_forward_finish(struct sk_buff *skb)
 634 {
 635         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 636         struct net_device *in;
 637
 638         if (skb->protocol != htons(ETH_P_ARP) && !IS_VLAN_ARP(skb)) {
 639                 in = nf_bridge->physindev;
 640                 if (nf_bridge->mask & BRNF_PKT_TYPE) {
 641                         skb->pkt_type = PACKET_OTHERHOST;
 642                         nf_bridge->mask ^= BRNF_PKT_TYPE;
 643                 }
 644                 nf_bridge_update_protocol(skb);
 645         } else {
 646                 in = *((struct net_device **)(skb->cb));
 647         }
 648         nf_bridge_push_encap_header(skb);
 649
 650         NF_HOOK_THRESH(NFPROTO_BRIDGE, NF_BR_FORWARD, skb, in,
 651                        skb->dev, br_forward_finish, 1);
 652         return 0;
 653 }
 654
 655 /* This is the 'purely bridged' case.  For IP, we pass the packet to
 656  * netfilter with indev and outdev set to the bridge device,
 657  * but we are still able to filter on the 'real' indev/outdev
 658  * because of the physdev module. For ARP, indev and outdev are the
 659  * bridge ports. */
 660 static unsigned int br_nf_forward_ip(unsigned int hook, struct sk_buff *skb,
 661                                      const struct net_device *in,
 662                                      const struct net_device *out,
 663                                      int (*okfn)(struct sk_buff *))
 664 {
 665         struct nf_bridge_info *nf_bridge;
 666         struct net_device *parent;
 667         u_int8_t pf;
 668
 669         if (!skb->nf_bridge)
 670                 return NF_ACCEPT;
 671
 672         /* Need exclusive nf_bridge_info since we might have multiple
 673          * different physoutdevs. */
 674         if (!nf_bridge_unshare(skb))
 675                 return NF_DROP;
 676
 677         parent = bridge_parent(out);
 678         if (!parent)
 679                 return NF_DROP;
 680
 681         if (skb->protocol == htons(ETH_P_IP) || IS_VLAN_IP(skb) ||
 682             IS_PPPOE_IP(skb))
 683                 pf = PF_INET;
 684         else if (skb->protocol == htons(ETH_P_IPV6) || IS_VLAN_IPV6(skb) ||
 685                  IS_PPPOE_IPV6(skb))
 686                 pf = PF_INET6;
 687         else
 688                 return NF_ACCEPT;
 689
 690         nf_bridge_pull_encap_header(skb);
 691
 692         nf_bridge = skb->nf_bridge;
 693         if (skb->pkt_type == PACKET_OTHERHOST) {
 694                 skb->pkt_type = PACKET_HOST;
 695                 nf_bridge->mask |= BRNF_PKT_TYPE;
 696         }
 697
 698         /* The physdev module checks on this */
 699         nf_bridge->mask |= BRNF_BRIDGED;
 700         nf_bridge->physoutdev = skb->dev;
 701         if (pf == PF_INET)
 702                 skb->protocol = htons(ETH_P_IP);
 703         else
 704                 skb->protocol = htons(ETH_P_IPV6);
 705
 706         NF_HOOK(pf, NF_INET_FORWARD, skb, bridge_parent(in), parent,
 707                 br_nf_forward_finish);
 708
 709         return NF_STOLEN;
 710 }
 711
 712 static unsigned int br_nf_forward_arp(unsigned int hook, struct sk_buff *skb,
 713                                       const struct net_device *in,
 714                                       const struct net_device *out,
 715                                       int (*okfn)(struct sk_buff *))
 716 {
 717         struct net_device **d = (struct net_device **)(skb->cb);
 718
 719 #ifdef CONFIG_SYSCTL
 720         if (!brnf_call_arptables)
 721                 return NF_ACCEPT;
 722 #endif
 723
 724         if (skb->protocol != htons(ETH_P_ARP)) {
 725                 if (!IS_VLAN_ARP(skb))
 726                         return NF_ACCEPT;
 727                 nf_bridge_pull_encap_header(skb);
 728         }
 729
 730         if (arp_hdr(skb)->ar_pln != 4) {
 731                 if (IS_VLAN_ARP(skb))
 732                         nf_bridge_push_encap_header(skb);
 733                 return NF_ACCEPT;
 734         }
 735         *d = (struct net_device *)in;
 736         NF_HOOK(NFPROTO_ARP, NF_ARP_FORWARD, skb, (struct net_device *)in,
 737                 (struct net_device *)out, br_nf_forward_finish);
 738
 739         return NF_STOLEN;
 740 }
 741
 742 #if defined(CONFIG_NF_CONNTRACK_IPV4) || defined(CONFIG_NF_CONNTRACK_IPV4_MODULE)
 743 static int br_nf_dev_queue_xmit(struct sk_buff *skb)
 744 {
 745         if (skb->nfct != NULL && skb->protocol == htons(ETH_P_IP) &&
 746             skb->len + nf_bridge_mtu_reduction(skb) > skb->dev->mtu &&
 747             !skb_is_gso(skb))
 748                 return ip_fragment(skb, br_dev_queue_push_xmit);
 749         else
 750                 return br_dev_queue_push_xmit(skb);
 751 }
 752 #else
 753 static int br_nf_dev_queue_xmit(struct sk_buff *skb)
 754 {
 755         return br_dev_queue_push_xmit(skb);
 756 }
 757 #endif
 758
 759 /* PF_BRIDGE/POST_ROUTING ********************************************/
 760 static unsigned int br_nf_post_routing(unsigned int hook, struct sk_buff *skb,
 761                                        const struct net_device *in,
 762                                        const struct net_device *out,
 763                                        int (*okfn)(struct sk_buff *))
 764 {
 765         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 766         struct net_device *realoutdev = bridge_parent(skb->dev);
 767         u_int8_t pf;
 768
 769         if (!nf_bridge || !(nf_bridge->mask & BRNF_BRIDGED))
 770                 return NF_ACCEPT;
 771
 772         if (!realoutdev)
 773                 return NF_DROP;
 774
 775         if (skb->protocol == htons(ETH_P_IP) || IS_VLAN_IP(skb) ||
 776             IS_PPPOE_IP(skb))
 777                 pf = PF_INET;
 778         else if (skb->protocol == htons(ETH_P_IPV6) || IS_VLAN_IPV6(skb) ||
 779                  IS_PPPOE_IPV6(skb))
 780                 pf = PF_INET6;
 781         else
 782                 return NF_ACCEPT;
 783
 784         /* We assume any code from br_dev_queue_push_xmit onwards doesn't care
 785          * about the value of skb->pkt_type. */
 786         if (skb->pkt_type == PACKET_OTHERHOST) {
 787                 skb->pkt_type = PACKET_HOST;
 788                 nf_bridge->mask |= BRNF_PKT_TYPE;
 789         }
 790
 791         nf_bridge_pull_encap_header(skb);
 792         nf_bridge_save_header(skb);
 793         if (pf == PF_INET)
 794                 skb->protocol = htons(ETH_P_IP);
 795         else
 796                 skb->protocol = htons(ETH_P_IPV6);
 797
 798         NF_HOOK(pf, NF_INET_POST_ROUTING, skb, NULL, realoutdev,
 799                 br_nf_dev_queue_xmit);
 800
 801         return NF_STOLEN;
 802 }
 803
 804 /* IP/SABOTAGE *****************************************************/
 805 /* Don't hand locally destined packets to PF_INET(6)/PRE_ROUTING
 806  * for the second time. */
 807 static unsigned int ip_sabotage_in(unsigned int hook, struct sk_buff *skb,
 808                                    const struct net_device *in,
 809                                    const struct net_device *out,
 810                                    int (*okfn)(struct sk_buff *))
 811 {
 812         if (skb->nf_bridge &&
 813             !(skb->nf_bridge->mask & BRNF_NF_BRIDGE_PREROUTING)) {
 814                 return NF_STOP;
 815         }
 816
 817         return NF_ACCEPT;
 818 }
 819
 820 /* For br_nf_post_routing, we need (prio = NF_BR_PRI_LAST), because
 821  * br_dev_queue_push_xmit is called afterwards */
 822 static struct nf_hook_ops br_nf_ops[] __read_mostly = {
 823         {
 824                 .hook = br_nf_pre_routing,
 825                 .owner = THIS_MODULE,
 826                 .pf = PF_BRIDGE,
 827                 .hooknum = NF_BR_PRE_ROUTING,
 828                 .priority = NF_BR_PRI_BRNF,
 829         },
 830         {
 831                 .hook = br_nf_local_in,
 832                 .owner = THIS_MODULE,
 833                 .pf = PF_BRIDGE,
 834                 .hooknum = NF_BR_LOCAL_IN,
 835                 .priority = NF_BR_PRI_BRNF,
 836         },
 837         {
 838                 .hook = br_nf_forward_ip,
 839                 .owner = THIS_MODULE,
 840                 .pf = PF_BRIDGE,
 841                 .hooknum = NF_BR_FORWARD,
 842                 .priority = NF_BR_PRI_BRNF - 1,
 843         },
 844         {
 845                 .hook = br_nf_forward_arp,
 846                 .owner = THIS_MODULE,
 847                 .pf = PF_BRIDGE,
 848                 .hooknum = NF_BR_FORWARD,
 849                 .priority = NF_BR_PRI_BRNF,
 850         },
 851         {
 852                 .hook = br_nf_post_routing,
 853                 .owner = THIS_MODULE,
 854                 .pf = PF_BRIDGE,
 855                 .hooknum = NF_BR_POST_ROUTING,
 856                 .priority = NF_BR_PRI_LAST,
 857         },
 858         {
 859                 .hook = ip_sabotage_in,
 860                 .owner = THIS_MODULE,
 861                 .pf = PF_INET,
 862                 .hooknum = NF_INET_PRE_ROUTING,
 863                 .priority = NF_IP_PRI_FIRST,
 864         },
 865         {
 866                 .hook = ip_sabotage_in,
 867                 .owner = THIS_MODULE,
 868                 .pf = PF_INET6,
 869                 .hooknum = NF_INET_PRE_ROUTING,
 870                 .priority = NF_IP6_PRI_FIRST,
 871         },
 872 };
 873
 874 #ifdef CONFIG_SYSCTL
 875 static
 876 int brnf_sysctl_call_tables(ctl_table * ctl, int write,
 877                             void __user * buffer, size_t * lenp, loff_t * ppos)
 878 {
 879         int ret;
 880
 881         ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
 882
 883         if (write && *(int *)(ctl->data))
 884                 *(int *)(ctl->data) = 1;
 885         return ret;
 886 }
 887
 888 static ctl_table brnf_table[] = {
 889         {
 890                 .procname       = "bridge-nf-call-arptables",
 891                 .data           = &brnf_call_arptables,
 892                 .maxlen         = sizeof(int),
 893                 .mode           = 0644,
 894                 .proc_handler   = brnf_sysctl_call_tables,
 895         },
 896         {
 897                 .procname       = "bridge-nf-call-iptables",
 898                 .data           = &brnf_call_iptables,
 899                 .maxlen         = sizeof(int),
 900                 .mode           = 0644,
 901                 .proc_handler   = brnf_sysctl_call_tables,
 902         },
 903         {
 904                 .procname       = "bridge-nf-call-ip6tables",
 905                 .data           = &brnf_call_ip6tables,
 906                 .maxlen         = sizeof(int),
 907                 .mode           = 0644,
 908                 .proc_handler   = brnf_sysctl_call_tables,
 909         },
 910         {
 911                 .procname       = "bridge-nf-filter-vlan-tagged",
 912                 .data           = &brnf_filter_vlan_tagged,
 913                 .maxlen         = sizeof(int),
 914                 .mode           = 0644,
 915                 .proc_handler   = brnf_sysctl_call_tables,
 916         },
 917         {
 918                 .procname       = "bridge-nf-filter-pppoe-tagged",
 919                 .data           = &brnf_filter_pppoe_tagged,
 920                 .maxlen         = sizeof(int),
 921                 .mode           = 0644,
 922                 .proc_handler   = brnf_sysctl_call_tables,
 923         },
 924         { }
 925 };
 926
 927 static struct ctl_path brnf_path[] = {
 928         { .procname = "net", },
 929         { .procname = "bridge", },
 930         { }
 931 };
 932 #endif
 933
 934 int __init br_netfilter_init(void)
 935 {
 936         int ret;
 937
 938         ret = nf_register_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
 939         if (ret < 0)
 940                 return ret;
 941 #ifdef CONFIG_SYSCTL
 942         brnf_sysctl_header = register_sysctl_paths(brnf_path, brnf_table);
 943         if (brnf_sysctl_header == NULL) {
 944                 printk(KERN_WARNING
 945                        "br_netfilter: can't register to sysctl.\n");
 946                 nf_unregister_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
 947                 return -ENOMEM;
 948         }
 949 #endif
 950         printk(KERN_NOTICE "Bridge firewalling registered\n");
 951         return 0;
 952 }
 953
 954 void br_netfilter_fini(void)
 955 {
 956         nf_unregister_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
 957 #ifdef CONFIG_SYSCTL
 958         unregister_sysctl_table(brnf_sysctl_header);
 959 #endif
 960 }