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->dst.__refcnt, 1);
 121         rt->dst.dev = br->dev;
 122         rt->dst.path = &rt->dst;
 123         rt->dst.metrics[RTAX_MTU - 1] = 1500;
 124         rt->dst.flags   = DST_NOXFRM;
 125         rt->dst.ops = &fake_dst_ops;
 126 }
 127
 128 static inline struct rtable *bridge_parent_rtable(const struct net_device *dev)
 129 {
 130         if (!br_port_exists(dev))
 131                 return NULL;
 132         return &br_port_get_rcu(dev)->br->fake_rtable;
 133 }
 134
 135 static inline struct net_device *bridge_parent(const struct net_device *dev)
 136 {
 137         if (!br_port_exists(dev))
 138                 return NULL;
 139
 140         return br_port_get_rcu(dev)->br->dev;
 141 }
 142
 143 static inline struct nf_bridge_info *nf_bridge_alloc(struct sk_buff *skb)
 144 {
 145         skb->nf_bridge = kzalloc(sizeof(struct nf_bridge_info), GFP_ATOMIC);
 146         if (likely(skb->nf_bridge))
 147                 atomic_set(&(skb->nf_bridge->use), 1);
 148
 149         return skb->nf_bridge;
 150 }
 151
 152 static inline struct nf_bridge_info *nf_bridge_unshare(struct sk_buff *skb)
 153 {
 154         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 155
 156         if (atomic_read(&nf_bridge->use) > 1) {
 157                 struct nf_bridge_info *tmp = nf_bridge_alloc(skb);
 158
 159                 if (tmp) {
 160                         memcpy(tmp, nf_bridge, sizeof(struct nf_bridge_info));
 161                         atomic_set(&tmp->use, 1);
 162                         nf_bridge_put(nf_bridge);
 163                 }
 164                 nf_bridge = tmp;
 165         }
 166         return nf_bridge;
 167 }
 168
 169 static inline void nf_bridge_push_encap_header(struct sk_buff *skb)
 170 {
 171         unsigned int len = nf_bridge_encap_header_len(skb);
 172
 173         skb_push(skb, len);
 174         skb->network_header -= len;
 175 }
 176
 177 static inline void nf_bridge_pull_encap_header(struct sk_buff *skb)
 178 {
 179         unsigned int len = nf_bridge_encap_header_len(skb);
 180
 181         skb_pull(skb, len);
 182         skb->network_header += len;
 183 }
 184
 185 static inline void nf_bridge_pull_encap_header_rcsum(struct sk_buff *skb)
 186 {
 187         unsigned int len = nf_bridge_encap_header_len(skb);
 188
 189         skb_pull_rcsum(skb, len);
 190         skb->network_header += len;
 191 }
 192
 193 static inline void nf_bridge_save_header(struct sk_buff *skb)
 194 {
 195         int header_size = ETH_HLEN + nf_bridge_encap_header_len(skb);
 196
 197         skb_copy_from_linear_data_offset(skb, -header_size,
 198                                          skb->nf_bridge->data, header_size);
 199 }
 200
 201 static inline void nf_bridge_update_protocol(struct sk_buff *skb)
 202 {
 203         if (skb->nf_bridge->mask & BRNF_8021Q)
 204                 skb->protocol = htons(ETH_P_8021Q);
 205         else if (skb->nf_bridge->mask & BRNF_PPPoE)
 206                 skb->protocol = htons(ETH_P_PPP_SES);
 207 }
 208
 209 /* Fill in the header for fragmented IP packets handled by
 210  * the IPv4 connection tracking code.
 211  */
 212 int nf_bridge_copy_header(struct sk_buff *skb)
 213 {
 214         int err;
 215         unsigned int header_size;
 216
 217         nf_bridge_update_protocol(skb);
 218         header_size = ETH_HLEN + nf_bridge_encap_header_len(skb);
 219         err = skb_cow_head(skb, header_size);
 220         if (err)
 221                 return err;
 222
 223         skb_copy_to_linear_data_offset(skb, -header_size,
 224                                        skb->nf_bridge->data, header_size);
 225         __skb_push(skb, nf_bridge_encap_header_len(skb));
 226         return 0;
 227 }
 228
 229 /* PF_BRIDGE/PRE_ROUTING *********************************************/
 230 /* Undo the changes made for ip6tables PREROUTING and continue the
 231  * bridge PRE_ROUTING hook. */
 232 static int br_nf_pre_routing_finish_ipv6(struct sk_buff *skb)
 233 {
 234         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 235         struct rtable *rt;
 236
 237         if (nf_bridge->mask & BRNF_PKT_TYPE) {
 238                 skb->pkt_type = PACKET_OTHERHOST;
 239                 nf_bridge->mask ^= BRNF_PKT_TYPE;
 240         }
 241         nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
 242
 243         rt = bridge_parent_rtable(nf_bridge->physindev);
 244         if (!rt) {
 245                 kfree_skb(skb);
 246                 return 0;
 247         }
 248         dst_hold(&rt->dst);
 249         skb_dst_set(skb, &rt->dst);
 250
 251         skb->dev = nf_bridge->physindev;
 252         nf_bridge_update_protocol(skb);
 253         nf_bridge_push_encap_header(skb);
 254         NF_HOOK_THRESH(NFPROTO_BRIDGE, NF_BR_PRE_ROUTING, skb, skb->dev, NULL,
 255                        br_handle_frame_finish, 1);
 256
 257         return 0;
 258 }
 259
 260 /* Obtain the correct destination MAC address, while preserving the original
 261  * source MAC address. If we already know this address, we just copy it. If we
 262  * don't, we use the neighbour framework to find out. In both cases, we make
 263  * sure that br_handle_frame_finish() is called afterwards.
 264  */
 265 static int br_nf_pre_routing_finish_bridge(struct sk_buff *skb)
 266 {
 267         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 268         struct dst_entry *dst;
 269
 270         skb->dev = bridge_parent(skb->dev);
 271         if (!skb->dev)
 272                 goto free_skb;
 273         dst = skb_dst(skb);
 274         if (dst->hh) {
 275                 neigh_hh_bridge(dst->hh, skb);
 276                 skb->dev = nf_bridge->physindev;
 277                 return br_handle_frame_finish(skb);
 278         } else if (dst->neighbour) {
 279                 /* the neighbour function below overwrites the complete
 280                  * MAC header, so we save the Ethernet source address and
 281                  * protocol number. */
 282                 skb_copy_from_linear_data_offset(skb, -(ETH_HLEN-ETH_ALEN), skb->nf_bridge->data, ETH_HLEN-ETH_ALEN);
 283                 /* tell br_dev_xmit to continue with forwarding */
 284                 nf_bridge->mask |= BRNF_BRIDGED_DNAT;
 285                 return dst->neighbour->output(skb);
 286         }
 287 free_skb:
 288         kfree_skb(skb);
 289         return 0;
 290 }
 291
 292 /* This requires some explaining. If DNAT has taken place,
 293  * we will need to fix up the destination Ethernet address.
 294  *
 295  * There are two cases to consider:
 296  * 1. The packet was DNAT'ed to a device in the same bridge
 297  *    port group as it was received on. We can still bridge
 298  *    the packet.
 299  * 2. The packet was DNAT'ed to a different device, either
 300  *    a non-bridged device or another bridge port group.
 301  *    The packet will need to be routed.
 302  *
 303  * The correct way of distinguishing between these two cases is to
 304  * call ip_route_input() and to look at skb->dst->dev, which is
 305  * changed to the destination device if ip_route_input() succeeds.
 306  *
 307  * Let's first consider the case that ip_route_input() succeeds:
 308  *
 309  * If the output device equals the logical bridge device the packet
 310  * came in on, we can consider this bridging. The corresponding MAC
 311  * address will be obtained in br_nf_pre_routing_finish_bridge.
 312  * Otherwise, the packet is considered to be routed and we just
 313  * change the destination MAC address so that the packet will
 314  * later be passed up to the IP stack to be routed. For a redirected
 315  * packet, ip_route_input() will give back the localhost as output device,
 316  * which differs from the bridge device.
 317  *
 318  * Let's now consider the case that ip_route_input() fails:
 319  *
 320  * This can be because the destination address is martian, in which case
 321  * the packet will be dropped.
 322  * If IP forwarding is disabled, ip_route_input() will fail, while
 323  * ip_route_output_key() can return success. The source
 324  * address for ip_route_output_key() is set to zero, so ip_route_output_key()
 325  * thinks we're handling a locally generated packet and won't care
 326  * if IP forwarding is enabled. If the output device equals the logical bridge
 327  * device, we proceed as if ip_route_input() succeeded. If it differs from the
 328  * logical bridge port or if ip_route_output_key() fails we drop the packet.
 329  */
 330 static int br_nf_pre_routing_finish(struct sk_buff *skb)
 331 {
 332         struct net_device *dev = skb->dev;
 333         struct iphdr *iph = ip_hdr(skb);
 334         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 335         struct rtable *rt;
 336         int err;
 337
 338         if (nf_bridge->mask & BRNF_PKT_TYPE) {
 339                 skb->pkt_type = PACKET_OTHERHOST;
 340                 nf_bridge->mask ^= BRNF_PKT_TYPE;
 341         }
 342         nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
 343         if (dnat_took_place(skb)) {
 344                 if ((err = ip_route_input(skb, iph->daddr, iph->saddr, iph->tos, dev))) {
 345                         struct flowi fl = {
 346                                 .nl_u = {
 347                                         .ip4_u = {
 348                                                  .daddr = iph->daddr,
 349                                                  .saddr = 0,
 350                                                  .tos = RT_TOS(iph->tos) },
 351                                 },
 352                                 .proto = 0,
 353                         };
 354                         struct in_device *in_dev = __in_dev_get_rcu(dev);
 355
 356                         /* If err equals -EHOSTUNREACH the error is due to a
 357                          * martian destination or due to the fact that
 358                          * forwarding is disabled. For most martian packets,
 359                          * ip_route_output_key() will fail. It won't fail for 2 types of
 360                          * martian destinations: loopback destinations and destination
 361                          * 0.0.0.0. In both cases the packet will be dropped because the
 362                          * destination is the loopback device and not the bridge. */
 363                         if (err != -EHOSTUNREACH || !in_dev || IN_DEV_FORWARD(in_dev))
 364                                 goto free_skb;
 365
 366                         if (!ip_route_output_key(dev_net(dev), &rt, &fl)) {
 367                                 /* - Bridged-and-DNAT'ed traffic doesn't
 368                                  *   require ip_forwarding. */
 369                                 if (((struct dst_entry *)rt)->dev == dev) {
 370                                         skb_dst_set(skb, (struct dst_entry *)rt);
 371                                         goto bridged_dnat;
 372                                 }
 373                                 dst_release((struct dst_entry *)rt);
 374                         }
 375 free_skb:
 376                         kfree_skb(skb);
 377                         return 0;
 378                 } else {
 379                         if (skb_dst(skb)->dev == dev) {
 380 bridged_dnat:
 381                                 skb->dev = nf_bridge->physindev;
 382                                 nf_bridge_update_protocol(skb);
 383                                 nf_bridge_push_encap_header(skb);
 384                                 NF_HOOK_THRESH(NFPROTO_BRIDGE,
 385                                                NF_BR_PRE_ROUTING,
 386                                                skb, skb->dev, NULL,
 387                                                br_nf_pre_routing_finish_bridge,
 388                                                1);
 389                                 return 0;
 390                         }
 391                         memcpy(eth_hdr(skb)->h_dest, dev->dev_addr, ETH_ALEN);
 392                         skb->pkt_type = PACKET_HOST;
 393                 }
 394         } else {
 395                 rt = bridge_parent_rtable(nf_bridge->physindev);
 396                 if (!rt) {
 397                         kfree_skb(skb);
 398                         return 0;
 399                 }
 400                 dst_hold(&rt->dst);
 401                 skb_dst_set(skb, &rt->dst);
 402         }
 403
 404         skb->dev = nf_bridge->physindev;
 405         nf_bridge_update_protocol(skb);
 406         nf_bridge_push_encap_header(skb);
 407         NF_HOOK_THRESH(NFPROTO_BRIDGE, NF_BR_PRE_ROUTING, skb, skb->dev, NULL,
 408                        br_handle_frame_finish, 1);
 409
 410         return 0;
 411 }
 412
 413 /* Some common code for IPv4/IPv6 */
 414 static struct net_device *setup_pre_routing(struct sk_buff *skb)
 415 {
 416         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 417
 418         if (skb->pkt_type == PACKET_OTHERHOST) {
 419                 skb->pkt_type = PACKET_HOST;
 420                 nf_bridge->mask |= BRNF_PKT_TYPE;
 421         }
 422
 423         nf_bridge->mask |= BRNF_NF_BRIDGE_PREROUTING;
 424         nf_bridge->physindev = skb->dev;
 425         skb->dev = bridge_parent(skb->dev);
 426         if (skb->protocol == htons(ETH_P_8021Q))
 427                 nf_bridge->mask |= BRNF_8021Q;
 428         else if (skb->protocol == htons(ETH_P_PPP_SES))
 429                 nf_bridge->mask |= BRNF_PPPoE;
 430
 431         return skb->dev;
 432 }
 433
 434 /* We only check the length. A bridge shouldn't do any hop-by-hop stuff anyway */
 435 static int check_hbh_len(struct sk_buff *skb)
 436 {
 437         unsigned char *raw = (u8 *)(ipv6_hdr(skb) + 1);
 438         u32 pkt_len;
 439         const unsigned char *nh = skb_network_header(skb);
 440         int off = raw - nh;
 441         int len = (raw[1] + 1) << 3;
 442
 443         if ((raw + len) - skb->data > skb_headlen(skb))
 444                 goto bad;
 445
 446         off += 2;
 447         len -= 2;
 448
 449         while (len > 0) {
 450                 int optlen = nh[off + 1] + 2;
 451
 452                 switch (nh[off]) {
 453                 case IPV6_TLV_PAD0:
 454                         optlen = 1;
 455                         break;
 456
 457                 case IPV6_TLV_PADN:
 458                         break;
 459
 460                 case IPV6_TLV_JUMBO:
 461                         if (nh[off + 1] != 4 || (off & 3) != 2)
 462                                 goto bad;
 463                         pkt_len = ntohl(*(__be32 *) (nh + off + 2));
 464                         if (pkt_len <= IPV6_MAXPLEN ||
 465                             ipv6_hdr(skb)->payload_len)
 466                                 goto bad;
 467                         if (pkt_len > skb->len - sizeof(struct ipv6hdr))
 468                                 goto bad;
 469                         if (pskb_trim_rcsum(skb,
 470                                             pkt_len + sizeof(struct ipv6hdr)))
 471                                 goto bad;
 472                         nh = skb_network_header(skb);
 473                         break;
 474                 default:
 475                         if (optlen > len)
 476                                 goto bad;
 477                         break;
 478                 }
 479                 off += optlen;
 480                 len -= optlen;
 481         }
 482         if (len == 0)
 483                 return 0;
 484 bad:
 485         return -1;
 486
 487 }
 488
 489 /* Replicate the checks that IPv6 does on packet reception and pass the packet
 490  * to ip6tables, which doesn't support NAT, so things are fairly simple. */
 491 static unsigned int br_nf_pre_routing_ipv6(unsigned int hook,
 492                                            struct sk_buff *skb,
 493                                            const struct net_device *in,
 494                                            const struct net_device *out,
 495                                            int (*okfn)(struct sk_buff *))
 496 {
 497         struct ipv6hdr *hdr;
 498         u32 pkt_len;
 499
 500         if (skb->len < sizeof(struct ipv6hdr))
 501                 goto inhdr_error;
 502
 503         if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
 504                 goto inhdr_error;
 505
 506         hdr = ipv6_hdr(skb);
 507
 508         if (hdr->version != 6)
 509                 goto inhdr_error;
 510
 511         pkt_len = ntohs(hdr->payload_len);
 512
 513         if (pkt_len || hdr->nexthdr != NEXTHDR_HOP) {
 514                 if (pkt_len + sizeof(struct ipv6hdr) > skb->len)
 515                         goto inhdr_error;
 516                 if (pskb_trim_rcsum(skb, pkt_len + sizeof(struct ipv6hdr)))
 517                         goto inhdr_error;
 518         }
 519         if (hdr->nexthdr == NEXTHDR_HOP && check_hbh_len(skb))
 520                 goto inhdr_error;
 521
 522         nf_bridge_put(skb->nf_bridge);
 523         if (!nf_bridge_alloc(skb))
 524                 return NF_DROP;
 525         if (!setup_pre_routing(skb))
 526                 return NF_DROP;
 527
 528         skb->protocol = htons(ETH_P_IPV6);
 529         NF_HOOK(NFPROTO_IPV6, NF_INET_PRE_ROUTING, skb, skb->dev, NULL,
 530                 br_nf_pre_routing_finish_ipv6);
 531
 532         return NF_STOLEN;
 533
 534 inhdr_error:
 535         return NF_DROP;
 536 }
 537
 538 /* Direct IPv6 traffic to br_nf_pre_routing_ipv6.
 539  * Replicate the checks that IPv4 does on packet reception.
 540  * Set skb->dev to the bridge device (i.e. parent of the
 541  * receiving device) to make netfilter happy, the REDIRECT
 542  * target in particular.  Save the original destination IP
 543  * address to be able to detect DNAT afterwards. */
 544 static unsigned int br_nf_pre_routing(unsigned int hook, struct sk_buff *skb,
 545                                       const struct net_device *in,
 546                                       const struct net_device *out,
 547                                       int (*okfn)(struct sk_buff *))
 548 {
 549         struct iphdr *iph;
 550         __u32 len = nf_bridge_encap_header_len(skb);
 551
 552         if (unlikely(!pskb_may_pull(skb, len)))
 553                 goto out;
 554
 555         if (skb->protocol == htons(ETH_P_IPV6) || IS_VLAN_IPV6(skb) ||
 556             IS_PPPOE_IPV6(skb)) {
 557 #ifdef CONFIG_SYSCTL
 558                 if (!brnf_call_ip6tables)
 559                         return NF_ACCEPT;
 560 #endif
 561                 nf_bridge_pull_encap_header_rcsum(skb);
 562                 return br_nf_pre_routing_ipv6(hook, skb, in, out, okfn);
 563         }
 564 #ifdef CONFIG_SYSCTL
 565         if (!brnf_call_iptables)
 566                 return NF_ACCEPT;
 567 #endif
 568
 569         if (skb->protocol != htons(ETH_P_IP) && !IS_VLAN_IP(skb) &&
 570             !IS_PPPOE_IP(skb))
 571                 return NF_ACCEPT;
 572
 573         nf_bridge_pull_encap_header_rcsum(skb);
 574
 575         if (!pskb_may_pull(skb, sizeof(struct iphdr)))
 576                 goto inhdr_error;
 577
 578         iph = ip_hdr(skb);
 579         if (iph->ihl < 5 || iph->version != 4)
 580                 goto inhdr_error;
 581
 582         if (!pskb_may_pull(skb, 4 * iph->ihl))
 583                 goto inhdr_error;
 584
 585         iph = ip_hdr(skb);
 586         if (ip_fast_csum((__u8 *) iph, iph->ihl) != 0)
 587                 goto inhdr_error;
 588
 589         len = ntohs(iph->tot_len);
 590         if (skb->len < len || len < 4 * iph->ihl)
 591                 goto inhdr_error;
 592
 593         pskb_trim_rcsum(skb, len);
 594
 595         nf_bridge_put(skb->nf_bridge);
 596         if (!nf_bridge_alloc(skb))
 597                 return NF_DROP;
 598         if (!setup_pre_routing(skb))
 599                 return NF_DROP;
 600         store_orig_dstaddr(skb);
 601         skb->protocol = htons(ETH_P_IP);
 602
 603         NF_HOOK(NFPROTO_IPV4, NF_INET_PRE_ROUTING, skb, skb->dev, NULL,
 604                 br_nf_pre_routing_finish);
 605
 606         return NF_STOLEN;
 607
 608 inhdr_error:
 609 //      IP_INC_STATS_BH(IpInHdrErrors);
 610 out:
 611         return NF_DROP;
 612 }
 613
 614
 615 /* PF_BRIDGE/LOCAL_IN ************************************************/
 616 /* The packet is locally destined, which requires a real
 617  * dst_entry, so detach the fake one.  On the way up, the
 618  * packet would pass through PRE_ROUTING again (which already
 619  * took place when the packet entered the bridge), but we
 620  * register an IPv4 PRE_ROUTING 'sabotage' hook that will
 621  * prevent this from happening. */
 622 static unsigned int br_nf_local_in(unsigned int hook, struct sk_buff *skb,
 623                                    const struct net_device *in,
 624                                    const struct net_device *out,
 625                                    int (*okfn)(struct sk_buff *))
 626 {
 627         struct rtable *rt = skb_rtable(skb);
 628
 629         if (rt && rt == bridge_parent_rtable(in))
 630                 skb_dst_drop(skb);
 631
 632         return NF_ACCEPT;
 633 }
 634
 635 /* PF_BRIDGE/FORWARD *************************************************/
 636 static int br_nf_forward_finish(struct sk_buff *skb)
 637 {
 638         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 639         struct net_device *in;
 640
 641         if (skb->protocol != htons(ETH_P_ARP) && !IS_VLAN_ARP(skb)) {
 642                 in = nf_bridge->physindev;
 643                 if (nf_bridge->mask & BRNF_PKT_TYPE) {
 644                         skb->pkt_type = PACKET_OTHERHOST;
 645                         nf_bridge->mask ^= BRNF_PKT_TYPE;
 646                 }
 647                 nf_bridge_update_protocol(skb);
 648         } else {
 649                 in = *((struct net_device **)(skb->cb));
 650         }
 651         nf_bridge_push_encap_header(skb);
 652
 653         NF_HOOK_THRESH(NFPROTO_BRIDGE, NF_BR_FORWARD, skb, in,
 654                        skb->dev, br_forward_finish, 1);
 655         return 0;
 656 }
 657
 658 /* This is the 'purely bridged' case.  For IP, we pass the packet to
 659  * netfilter with indev and outdev set to the bridge device,
 660  * but we are still able to filter on the 'real' indev/outdev
 661  * because of the physdev module. For ARP, indev and outdev are the
 662  * bridge ports. */
 663 static unsigned int br_nf_forward_ip(unsigned int hook, struct sk_buff *skb,
 664                                      const struct net_device *in,
 665                                      const struct net_device *out,
 666                                      int (*okfn)(struct sk_buff *))
 667 {
 668         struct nf_bridge_info *nf_bridge;
 669         struct net_device *parent;
 670         u_int8_t pf;
 671
 672         if (!skb->nf_bridge)
 673                 return NF_ACCEPT;
 674
 675         /* Need exclusive nf_bridge_info since we might have multiple
 676          * different physoutdevs. */
 677         if (!nf_bridge_unshare(skb))
 678                 return NF_DROP;
 679
 680         parent = bridge_parent(out);
 681         if (!parent)
 682                 return NF_DROP;
 683
 684         if (skb->protocol == htons(ETH_P_IP) || IS_VLAN_IP(skb) ||
 685             IS_PPPOE_IP(skb))
 686                 pf = PF_INET;
 687         else if (skb->protocol == htons(ETH_P_IPV6) || IS_VLAN_IPV6(skb) ||
 688                  IS_PPPOE_IPV6(skb))
 689                 pf = PF_INET6;
 690         else
 691                 return NF_ACCEPT;
 692
 693         nf_bridge_pull_encap_header(skb);
 694
 695         nf_bridge = skb->nf_bridge;
 696         if (skb->pkt_type == PACKET_OTHERHOST) {
 697                 skb->pkt_type = PACKET_HOST;
 698                 nf_bridge->mask |= BRNF_PKT_TYPE;
 699         }
 700
 701         /* The physdev module checks on this */
 702         nf_bridge->mask |= BRNF_BRIDGED;
 703         nf_bridge->physoutdev = skb->dev;
 704         if (pf == PF_INET)
 705                 skb->protocol = htons(ETH_P_IP);
 706         else
 707                 skb->protocol = htons(ETH_P_IPV6);
 708
 709         NF_HOOK(pf, NF_INET_FORWARD, skb, bridge_parent(in), parent,
 710                 br_nf_forward_finish);
 711
 712         return NF_STOLEN;
 713 }
 714
 715 static unsigned int br_nf_forward_arp(unsigned int hook, struct sk_buff *skb,
 716                                       const struct net_device *in,
 717                                       const struct net_device *out,
 718                                       int (*okfn)(struct sk_buff *))
 719 {
 720         struct net_device **d = (struct net_device **)(skb->cb);
 721
 722 #ifdef CONFIG_SYSCTL
 723         if (!brnf_call_arptables)
 724                 return NF_ACCEPT;
 725 #endif
 726
 727         if (skb->protocol != htons(ETH_P_ARP)) {
 728                 if (!IS_VLAN_ARP(skb))
 729                         return NF_ACCEPT;
 730                 nf_bridge_pull_encap_header(skb);
 731         }
 732
 733         if (arp_hdr(skb)->ar_pln != 4) {
 734                 if (IS_VLAN_ARP(skb))
 735                         nf_bridge_push_encap_header(skb);
 736                 return NF_ACCEPT;
 737         }
 738         *d = (struct net_device *)in;
 739         NF_HOOK(NFPROTO_ARP, NF_ARP_FORWARD, skb, (struct net_device *)in,
 740                 (struct net_device *)out, br_nf_forward_finish);
 741
 742         return NF_STOLEN;
 743 }
 744
 745 #if defined(CONFIG_NF_CONNTRACK_IPV4) || defined(CONFIG_NF_CONNTRACK_IPV4_MODULE)
 746 static int br_nf_dev_queue_xmit(struct sk_buff *skb)
 747 {
 748         if (skb->nfct != NULL && skb->protocol == htons(ETH_P_IP) &&
 749             skb->len + nf_bridge_mtu_reduction(skb) > skb->dev->mtu &&
 750             !skb_is_gso(skb))
 751                 return ip_fragment(skb, br_dev_queue_push_xmit);
 752         else
 753                 return br_dev_queue_push_xmit(skb);
 754 }
 755 #else
 756 static int br_nf_dev_queue_xmit(struct sk_buff *skb)
 757 {
 758         return br_dev_queue_push_xmit(skb);
 759 }
 760 #endif
 761
 762 /* PF_BRIDGE/POST_ROUTING ********************************************/
 763 static unsigned int br_nf_post_routing(unsigned int hook, struct sk_buff *skb,
 764                                        const struct net_device *in,
 765                                        const struct net_device *out,
 766                                        int (*okfn)(struct sk_buff *))
 767 {
 768         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 769         struct net_device *realoutdev = bridge_parent(skb->dev);
 770         u_int8_t pf;
 771
 772         if (!nf_bridge || !(nf_bridge->mask & BRNF_BRIDGED))
 773                 return NF_ACCEPT;
 774
 775         if (!realoutdev)
 776                 return NF_DROP;
 777
 778         if (skb->protocol == htons(ETH_P_IP) || IS_VLAN_IP(skb) ||
 779             IS_PPPOE_IP(skb))
 780                 pf = PF_INET;
 781         else if (skb->protocol == htons(ETH_P_IPV6) || IS_VLAN_IPV6(skb) ||
 782                  IS_PPPOE_IPV6(skb))
 783                 pf = PF_INET6;
 784         else
 785                 return NF_ACCEPT;
 786
 787         /* We assume any code from br_dev_queue_push_xmit onwards doesn't care
 788          * about the value of skb->pkt_type. */
 789         if (skb->pkt_type == PACKET_OTHERHOST) {
 790                 skb->pkt_type = PACKET_HOST;
 791                 nf_bridge->mask |= BRNF_PKT_TYPE;
 792         }
 793
 794         nf_bridge_pull_encap_header(skb);
 795         nf_bridge_save_header(skb);
 796         if (pf == PF_INET)
 797                 skb->protocol = htons(ETH_P_IP);
 798         else
 799                 skb->protocol = htons(ETH_P_IPV6);
 800
 801         NF_HOOK(pf, NF_INET_POST_ROUTING, skb, NULL, realoutdev,
 802                 br_nf_dev_queue_xmit);
 803
 804         return NF_STOLEN;
 805 }
 806
 807 /* IP/SABOTAGE *****************************************************/
 808 /* Don't hand locally destined packets to PF_INET(6)/PRE_ROUTING
 809  * for the second time. */
 810 static unsigned int ip_sabotage_in(unsigned int hook, struct sk_buff *skb,
 811                                    const struct net_device *in,
 812                                    const struct net_device *out,
 813                                    int (*okfn)(struct sk_buff *))
 814 {
 815         if (skb->nf_bridge &&
 816             !(skb->nf_bridge->mask & BRNF_NF_BRIDGE_PREROUTING)) {
 817                 return NF_STOP;
 818         }
 819
 820         return NF_ACCEPT;
 821 }
 822
 823 /* For br_nf_post_routing, we need (prio = NF_BR_PRI_LAST), because
 824  * br_dev_queue_push_xmit is called afterwards */
 825 static struct nf_hook_ops br_nf_ops[] __read_mostly = {
 826         {
 827                 .hook = br_nf_pre_routing,
 828                 .owner = THIS_MODULE,
 829                 .pf = PF_BRIDGE,
 830                 .hooknum = NF_BR_PRE_ROUTING,
 831                 .priority = NF_BR_PRI_BRNF,
 832         },
 833         {
 834                 .hook = br_nf_local_in,
 835                 .owner = THIS_MODULE,
 836                 .pf = PF_BRIDGE,
 837                 .hooknum = NF_BR_LOCAL_IN,
 838                 .priority = NF_BR_PRI_BRNF,
 839         },
 840         {
 841                 .hook = br_nf_forward_ip,
 842                 .owner = THIS_MODULE,
 843                 .pf = PF_BRIDGE,
 844                 .hooknum = NF_BR_FORWARD,
 845                 .priority = NF_BR_PRI_BRNF - 1,
 846         },
 847         {
 848                 .hook = br_nf_forward_arp,
 849                 .owner = THIS_MODULE,
 850                 .pf = PF_BRIDGE,
 851                 .hooknum = NF_BR_FORWARD,
 852                 .priority = NF_BR_PRI_BRNF,
 853         },
 854         {
 855                 .hook = br_nf_post_routing,
 856                 .owner = THIS_MODULE,
 857                 .pf = PF_BRIDGE,
 858                 .hooknum = NF_BR_POST_ROUTING,
 859                 .priority = NF_BR_PRI_LAST,
 860         },
 861         {
 862                 .hook = ip_sabotage_in,
 863                 .owner = THIS_MODULE,
 864                 .pf = PF_INET,
 865                 .hooknum = NF_INET_PRE_ROUTING,
 866                 .priority = NF_IP_PRI_FIRST,
 867         },
 868         {
 869                 .hook = ip_sabotage_in,
 870                 .owner = THIS_MODULE,
 871                 .pf = PF_INET6,
 872                 .hooknum = NF_INET_PRE_ROUTING,
 873                 .priority = NF_IP6_PRI_FIRST,
 874         },
 875 };
 876
 877 #ifdef CONFIG_SYSCTL
 878 static
 879 int brnf_sysctl_call_tables(ctl_table * ctl, int write,
 880                             void __user * buffer, size_t * lenp, loff_t * ppos)
 881 {
 882         int ret;
 883
 884         ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
 885
 886         if (write && *(int *)(ctl->data))
 887                 *(int *)(ctl->data) = 1;
 888         return ret;
 889 }
 890
 891 static ctl_table brnf_table[] = {
 892         {
 893                 .procname       = "bridge-nf-call-arptables",
 894                 .data           = &brnf_call_arptables,
 895                 .maxlen         = sizeof(int),
 896                 .mode           = 0644,
 897                 .proc_handler   = brnf_sysctl_call_tables,
 898         },
 899         {
 900                 .procname       = "bridge-nf-call-iptables",
 901                 .data           = &brnf_call_iptables,
 902                 .maxlen         = sizeof(int),
 903                 .mode           = 0644,
 904                 .proc_handler   = brnf_sysctl_call_tables,
 905         },
 906         {
 907                 .procname       = "bridge-nf-call-ip6tables",
 908                 .data           = &brnf_call_ip6tables,
 909                 .maxlen         = sizeof(int),
 910                 .mode           = 0644,
 911                 .proc_handler   = brnf_sysctl_call_tables,
 912         },
 913         {
 914                 .procname       = "bridge-nf-filter-vlan-tagged",
 915                 .data           = &brnf_filter_vlan_tagged,
 916                 .maxlen         = sizeof(int),
 917                 .mode           = 0644,
 918                 .proc_handler   = brnf_sysctl_call_tables,
 919         },
 920         {
 921                 .procname       = "bridge-nf-filter-pppoe-tagged",
 922                 .data           = &brnf_filter_pppoe_tagged,
 923                 .maxlen         = sizeof(int),
 924                 .mode           = 0644,
 925                 .proc_handler   = brnf_sysctl_call_tables,
 926         },
 927         { }
 928 };
 929
 930 static struct ctl_path brnf_path[] = {
 931         { .procname = "net", },
 932         { .procname = "bridge", },
 933         { }
 934 };
 935 #endif
 936
 937 int __init br_netfilter_init(void)
 938 {
 939         int ret;
 940
 941         ret = nf_register_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
 942         if (ret < 0)
 943                 return ret;
 944 #ifdef CONFIG_SYSCTL
 945         brnf_sysctl_header = register_sysctl_paths(brnf_path, brnf_table);
 946         if (brnf_sysctl_header == NULL) {
 947                 printk(KERN_WARNING
 948                        "br_netfilter: can't register to sysctl.\n");
 949                 nf_unregister_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
 950                 return -ENOMEM;
 951         }
 952 #endif
 953         printk(KERN_NOTICE "Bridge firewalling registered\n");
 954         return 0;
 955 }
 956
 957 void br_netfilter_fini(void)
 958 {
 959         nf_unregister_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
 960 #ifdef CONFIG_SYSCTL
 961         unregister_sysctl_table(brnf_sysctl_header);
 962 #endif
 963 }