SAFE public projects git trees. - safe/jmp/linux-2.6/blob - 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/ip.h>
  20 #include <linux/netdevice.h>
  21 #include <linux/skbuff.h>
  22 #include <linux/if_arp.h>
  23 #include <linux/if_ether.h>
  24 #include <linux/if_vlan.h>
  25 #include <linux/if_pppox.h>
  26 #include <linux/ppp_defs.h>
  27 #include <linux/netfilter_bridge.h>
  28 #include <linux/netfilter_ipv4.h>
  29 #include <linux/netfilter_ipv6.h>
  30 #include <linux/netfilter_arp.h>
  31 #include <linux/in_route.h>
  32 #include <linux/inetdevice.h>
  33
  34 #include <net/ip.h>
  35 #include <net/ipv6.h>
  36 #include <net/route.h>
  37
  38 #include <asm/uaccess.h>
  39 #include "br_private.h"
  40 #ifdef CONFIG_SYSCTL
  41 #include <linux/sysctl.h>
  42 #endif
  43
  44 #define skb_origaddr(skb)        (((struct bridge_skb_cb *) \
  45                                  (skb->nf_bridge->data))->daddr.ipv4)
  46 #define store_orig_dstaddr(skb)  (skb_origaddr(skb) = ip_hdr(skb)->daddr)
  47 #define dnat_took_place(skb)     (skb_origaddr(skb) != ip_hdr(skb)->daddr)
  48
  49 #ifdef CONFIG_SYSCTL
  50 static struct ctl_table_header *brnf_sysctl_header;
  51 static int brnf_call_iptables __read_mostly = 1;
  52 static int brnf_call_ip6tables __read_mostly = 1;
  53 static int brnf_call_arptables __read_mostly = 1;
  54 static int brnf_filter_vlan_tagged __read_mostly = 0;
  55 static int brnf_filter_pppoe_tagged __read_mostly = 0;
  56 #else
  57 #define brnf_filter_vlan_tagged 0
  58 #define brnf_filter_pppoe_tagged 0
  59 #endif
  60
  61 static inline __be16 vlan_proto(const struct sk_buff *skb)
  62 {
  63         return vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
  64 }
  65
  66 #define IS_VLAN_IP(skb) \
  67         (skb->protocol == htons(ETH_P_8021Q) && \
  68          vlan_proto(skb) == htons(ETH_P_IP) &&  \
  69          brnf_filter_vlan_tagged)
  70
  71 #define IS_VLAN_IPV6(skb) \
  72         (skb->protocol == htons(ETH_P_8021Q) && \
  73          vlan_proto(skb) == htons(ETH_P_IPV6) &&\
  74          brnf_filter_vlan_tagged)
  75
  76 #define IS_VLAN_ARP(skb) \
  77         (skb->protocol == htons(ETH_P_8021Q) && \
  78          vlan_proto(skb) == htons(ETH_P_ARP) && \
  79          brnf_filter_vlan_tagged)
  80
  81 static inline __be16 pppoe_proto(const struct sk_buff *skb)
  82 {
  83         return *((__be16 *)(skb_mac_header(skb) + ETH_HLEN +
  84                             sizeof(struct pppoe_hdr)));
  85 }
  86
  87 #define IS_PPPOE_IP(skb) \
  88         (skb->protocol == htons(ETH_P_PPP_SES) && \
  89          pppoe_proto(skb) == htons(PPP_IP) && \
  90          brnf_filter_pppoe_tagged)
  91
  92 #define IS_PPPOE_IPV6(skb) \
  93         (skb->protocol == htons(ETH_P_PPP_SES) && \
  94          pppoe_proto(skb) == htons(PPP_IPV6) && \
  95          brnf_filter_pppoe_tagged)
  96
  97 static void fake_update_pmtu(struct dst_entry *dst, u32 mtu)
  98 {
  99 }
 100
 101 static struct dst_ops fake_dst_ops = {
 102         .family =               AF_INET,
 103         .protocol =             cpu_to_be16(ETH_P_IP),
 104         .update_pmtu =          fake_update_pmtu,
 105         .entries =              ATOMIC_INIT(0),
 106 };
 107
 108 /*
 109  * Initialize bogus route table used to keep netfilter happy.
 110  * Currently, we fill in the PMTU entry because netfilter
 111  * refragmentation needs it, and the rt_flags entry because
 112  * ipt_REJECT needs it.  Future netfilter modules might
 113  * require us to fill additional fields.
 114  */
 115 void br_netfilter_rtable_init(struct net_bridge *br)
 116 {
 117         struct rtable *rt = &br->fake_rtable;
 118
 119         atomic_set(&rt->u.dst.__refcnt, 1);
 120         rt->u.dst.dev = br->dev;
 121         rt->u.dst.path = &rt->u.dst;
 122         rt->u.dst.metrics[RTAX_MTU - 1] = 1500;
 123         rt->u.dst.flags = DST_NOXFRM;
 124         rt->u.dst.ops = &fake_dst_ops;
 125 }
 126
 127 static inline struct rtable *bridge_parent_rtable(const struct net_device *dev)
 128 {
 129         struct net_bridge_port *port = rcu_dereference(dev->br_port);
 130
 131         return port ? &port->br->fake_rtable : NULL;
 132 }
 133
 134 static inline struct net_device *bridge_parent(const struct net_device *dev)
 135 {
 136         struct net_bridge_port *port = rcu_dereference(dev->br_port);
 137
 138         return port ? port->br->dev : NULL;
 139 }
 140
 141 static inline struct nf_bridge_info *nf_bridge_alloc(struct sk_buff *skb)
 142 {
 143         skb->nf_bridge = kzalloc(sizeof(struct nf_bridge_info), GFP_ATOMIC);
 144         if (likely(skb->nf_bridge))
 145                 atomic_set(&(skb->nf_bridge->use), 1);
 146
 147         return skb->nf_bridge;
 148 }
 149
 150 static inline struct nf_bridge_info *nf_bridge_unshare(struct sk_buff *skb)
 151 {
 152         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 153
 154         if (atomic_read(&nf_bridge->use) > 1) {
 155                 struct nf_bridge_info *tmp = nf_bridge_alloc(skb);
 156
 157                 if (tmp) {
 158                         memcpy(tmp, nf_bridge, sizeof(struct nf_bridge_info));
 159                         atomic_set(&tmp->use, 1);
 160                         nf_bridge_put(nf_bridge);
 161                 }
 162                 nf_bridge = tmp;
 163         }
 164         return nf_bridge;
 165 }
 166
 167 static inline void nf_bridge_push_encap_header(struct sk_buff *skb)
 168 {
 169         unsigned int len = nf_bridge_encap_header_len(skb);
 170
 171         skb_push(skb, len);
 172         skb->network_header -= len;
 173 }
 174
 175 static inline void nf_bridge_pull_encap_header(struct sk_buff *skb)
 176 {
 177         unsigned int len = nf_bridge_encap_header_len(skb);
 178
 179         skb_pull(skb, len);
 180         skb->network_header += len;
 181 }
 182
 183 static inline void nf_bridge_pull_encap_header_rcsum(struct sk_buff *skb)
 184 {
 185         unsigned int len = nf_bridge_encap_header_len(skb);
 186
 187         skb_pull_rcsum(skb, len);
 188         skb->network_header += len;
 189 }
 190
 191 static inline void nf_bridge_save_header(struct sk_buff *skb)
 192 {
 193         int header_size = ETH_HLEN + nf_bridge_encap_header_len(skb);
 194
 195         skb_copy_from_linear_data_offset(skb, -header_size,
 196                                          skb->nf_bridge->data, header_size);
 197 }
 198
 199 static inline void nf_bridge_update_protocol(struct sk_buff *skb)
 200 {
 201         if (skb->nf_bridge->mask & BRNF_8021Q)
 202                 skb->protocol = htons(ETH_P_8021Q);
 203         else if (skb->nf_bridge->mask & BRNF_PPPoE)
 204                 skb->protocol = htons(ETH_P_PPP_SES);
 205 }
 206
 207 /* Fill in the header for fragmented IP packets handled by
 208  * the IPv4 connection tracking code.
 209  */
 210 int nf_bridge_copy_header(struct sk_buff *skb)
 211 {
 212         int err;
 213         unsigned int header_size;
 214
 215         nf_bridge_update_protocol(skb);
 216         header_size = ETH_HLEN + nf_bridge_encap_header_len(skb);
 217         err = skb_cow_head(skb, header_size);
 218         if (err)
 219                 return err;
 220
 221         skb_copy_to_linear_data_offset(skb, -header_size,
 222                                        skb->nf_bridge->data, header_size);
 223         __skb_push(skb, nf_bridge_encap_header_len(skb));
 224         return 0;
 225 }
 226
 227 /* PF_BRIDGE/PRE_ROUTING *********************************************/
 228 /* Undo the changes made for ip6tables PREROUTING and continue the
 229  * bridge PRE_ROUTING hook. */
 230 static int br_nf_pre_routing_finish_ipv6(struct sk_buff *skb)
 231 {
 232         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 233         struct rtable *rt;
 234
 235         if (nf_bridge->mask & BRNF_PKT_TYPE) {
 236                 skb->pkt_type = PACKET_OTHERHOST;
 237                 nf_bridge->mask ^= BRNF_PKT_TYPE;
 238         }
 239         nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
 240
 241         rt = bridge_parent_rtable(nf_bridge->physindev);
 242         if (!rt) {
 243                 kfree_skb(skb);
 244                 return 0;
 245         }
 246         dst_hold(&rt->u.dst);
 247         skb_dst_set(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                 dst_hold(&rt->u.dst);
 399                 skb_dst_set(skb, &rt->u.dst);
 400         }
 401
 402         skb->dev = nf_bridge->physindev;
 403         nf_bridge_update_protocol(skb);
 404         nf_bridge_push_encap_header(skb);
 405         NF_HOOK_THRESH(NFPROTO_BRIDGE, NF_BR_PRE_ROUTING, skb, skb->dev, NULL,
 406                        br_handle_frame_finish, 1);
 407
 408         return 0;
 409 }
 410
 411 /* Some common code for IPv4/IPv6 */
 412 static struct net_device *setup_pre_routing(struct sk_buff *skb)
 413 {
 414         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 415
 416         if (skb->pkt_type == PACKET_OTHERHOST) {
 417                 skb->pkt_type = PACKET_HOST;
 418                 nf_bridge->mask |= BRNF_PKT_TYPE;
 419         }
 420
 421         nf_bridge->mask |= BRNF_NF_BRIDGE_PREROUTING;
 422         nf_bridge->physindev = skb->dev;
 423         skb->dev = bridge_parent(skb->dev);
 424         if (skb->protocol == htons(ETH_P_8021Q))
 425                 nf_bridge->mask |= BRNF_8021Q;
 426         else if (skb->protocol == htons(ETH_P_PPP_SES))
 427                 nf_bridge->mask |= BRNF_PPPoE;
 428
 429         return skb->dev;
 430 }
 431
 432 /* We only check the length. A bridge shouldn't do any hop-by-hop stuff anyway */
 433 static int check_hbh_len(struct sk_buff *skb)
 434 {
 435         unsigned char *raw = (u8 *)(ipv6_hdr(skb) + 1);
 436         u32 pkt_len;
 437         const unsigned char *nh = skb_network_header(skb);
 438         int off = raw - nh;
 439         int len = (raw[1] + 1) << 3;
 440
 441         if ((raw + len) - skb->data > skb_headlen(skb))
 442                 goto bad;
 443
 444         off += 2;
 445         len -= 2;
 446
 447         while (len > 0) {
 448                 int optlen = nh[off + 1] + 2;
 449
 450                 switch (nh[off]) {
 451                 case IPV6_TLV_PAD0:
 452                         optlen = 1;
 453                         break;
 454
 455                 case IPV6_TLV_PADN:
 456                         break;
 457
 458                 case IPV6_TLV_JUMBO:
 459                         if (nh[off + 1] != 4 || (off & 3) != 2)
 460                                 goto bad;
 461                         pkt_len = ntohl(*(__be32 *) (nh + off + 2));
 462                         if (pkt_len <= IPV6_MAXPLEN ||
 463                             ipv6_hdr(skb)->payload_len)
 464                                 goto bad;
 465                         if (pkt_len > skb->len - sizeof(struct ipv6hdr))
 466                                 goto bad;
 467                         if (pskb_trim_rcsum(skb,
 468                                             pkt_len + sizeof(struct ipv6hdr)))
 469                                 goto bad;
 470                         nh = skb_network_header(skb);
 471                         break;
 472                 default:
 473                         if (optlen > len)
 474                                 goto bad;
 475                         break;
 476                 }
 477                 off += optlen;
 478                 len -= optlen;
 479         }
 480         if (len == 0)
 481                 return 0;
 482 bad:
 483         return -1;
 484
 485 }
 486
 487 /* Replicate the checks that IPv6 does on packet reception and pass the packet
 488  * to ip6tables, which doesn't support NAT, so things are fairly simple. */
 489 static unsigned int br_nf_pre_routing_ipv6(unsigned int hook,
 490                                            struct sk_buff *skb,
 491                                            const struct net_device *in,
 492                                            const struct net_device *out,
 493                                            int (*okfn)(struct sk_buff *))
 494 {
 495         struct ipv6hdr *hdr;
 496         u32 pkt_len;
 497
 498         if (skb->len < sizeof(struct ipv6hdr))
 499                 goto inhdr_error;
 500
 501         if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
 502                 goto inhdr_error;
 503
 504         hdr = ipv6_hdr(skb);
 505
 506         if (hdr->version != 6)
 507                 goto inhdr_error;
 508
 509         pkt_len = ntohs(hdr->payload_len);
 510
 511         if (pkt_len || hdr->nexthdr != NEXTHDR_HOP) {
 512                 if (pkt_len + sizeof(struct ipv6hdr) > skb->len)
 513                         goto inhdr_error;
 514                 if (pskb_trim_rcsum(skb, pkt_len + sizeof(struct ipv6hdr)))
 515                         goto inhdr_error;
 516         }
 517         if (hdr->nexthdr == NEXTHDR_HOP && check_hbh_len(skb))
 518                 goto inhdr_error;
 519
 520         nf_bridge_put(skb->nf_bridge);
 521         if (!nf_bridge_alloc(skb))
 522                 return NF_DROP;
 523         if (!setup_pre_routing(skb))
 524                 return NF_DROP;
 525
 526         skb->protocol = htons(ETH_P_IPV6);
 527         NF_HOOK(NFPROTO_IPV6, NF_INET_PRE_ROUTING, skb, skb->dev, NULL,
 528                 br_nf_pre_routing_finish_ipv6);
 529
 530         return NF_STOLEN;
 531
 532 inhdr_error:
 533         return NF_DROP;
 534 }
 535
 536 /* Direct IPv6 traffic to br_nf_pre_routing_ipv6.
 537  * Replicate the checks that IPv4 does on packet reception.
 538  * Set skb->dev to the bridge device (i.e. parent of the
 539  * receiving device) to make netfilter happy, the REDIRECT
 540  * target in particular.  Save the original destination IP
 541  * address to be able to detect DNAT afterwards. */
 542 static unsigned int br_nf_pre_routing(unsigned int hook, struct sk_buff *skb,
 543                                       const struct net_device *in,
 544                                       const struct net_device *out,
 545                                       int (*okfn)(struct sk_buff *))
 546 {
 547         struct iphdr *iph;
 548         __u32 len = nf_bridge_encap_header_len(skb);
 549
 550         if (unlikely(!pskb_may_pull(skb, len)))
 551                 goto out;
 552
 553         if (skb->protocol == htons(ETH_P_IPV6) || IS_VLAN_IPV6(skb) ||
 554             IS_PPPOE_IPV6(skb)) {
 555 #ifdef CONFIG_SYSCTL
 556                 if (!brnf_call_ip6tables)
 557                         return NF_ACCEPT;
 558 #endif
 559                 nf_bridge_pull_encap_header_rcsum(skb);
 560                 return br_nf_pre_routing_ipv6(hook, skb, in, out, okfn);
 561         }
 562 #ifdef CONFIG_SYSCTL
 563         if (!brnf_call_iptables)
 564                 return NF_ACCEPT;
 565 #endif
 566
 567         if (skb->protocol != htons(ETH_P_IP) && !IS_VLAN_IP(skb) &&
 568             !IS_PPPOE_IP(skb))
 569                 return NF_ACCEPT;
 570
 571         nf_bridge_pull_encap_header_rcsum(skb);
 572
 573         if (!pskb_may_pull(skb, sizeof(struct iphdr)))
 574                 goto inhdr_error;
 575
 576         iph = ip_hdr(skb);
 577         if (iph->ihl < 5 || iph->version != 4)
 578                 goto inhdr_error;
 579
 580         if (!pskb_may_pull(skb, 4 * iph->ihl))
 581                 goto inhdr_error;
 582
 583         iph = ip_hdr(skb);
 584         if (ip_fast_csum((__u8 *) iph, iph->ihl) != 0)
 585                 goto inhdr_error;
 586
 587         len = ntohs(iph->tot_len);
 588         if (skb->len < len || len < 4 * iph->ihl)
 589                 goto inhdr_error;
 590
 591         pskb_trim_rcsum(skb, len);
 592
 593         nf_bridge_put(skb->nf_bridge);
 594         if (!nf_bridge_alloc(skb))
 595                 return NF_DROP;
 596         if (!setup_pre_routing(skb))
 597                 return NF_DROP;
 598         store_orig_dstaddr(skb);
 599         skb->protocol = htons(ETH_P_IP);
 600
 601         NF_HOOK(NFPROTO_IPV4, NF_INET_PRE_ROUTING, skb, skb->dev, NULL,
 602                 br_nf_pre_routing_finish);
 603
 604         return NF_STOLEN;
 605
 606 inhdr_error:
 607 //      IP_INC_STATS_BH(IpInHdrErrors);
 608 out:
 609         return NF_DROP;
 610 }
 611
 612
 613 /* PF_BRIDGE/LOCAL_IN ************************************************/
 614 /* The packet is locally destined, which requires a real
 615  * dst_entry, so detach the fake one.  On the way up, the
 616  * packet would pass through PRE_ROUTING again (which already
 617  * took place when the packet entered the bridge), but we
 618  * register an IPv4 PRE_ROUTING 'sabotage' hook that will
 619  * prevent this from happening. */
 620 static unsigned int br_nf_local_in(unsigned int hook, struct sk_buff *skb,
 621                                    const struct net_device *in,
 622                                    const struct net_device *out,
 623                                    int (*okfn)(struct sk_buff *))
 624 {
 625         struct rtable *rt = skb_rtable(skb);
 626
 627         if (rt && rt == bridge_parent_rtable(in))
 628                 skb_dst_drop(skb);
 629
 630         return NF_ACCEPT;
 631 }
 632
 633 /* PF_BRIDGE/FORWARD *************************************************/
 634 static int br_nf_forward_finish(struct sk_buff *skb)
 635 {
 636         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 637         struct net_device *in;
 638
 639         if (skb->protocol != htons(ETH_P_ARP) && !IS_VLAN_ARP(skb)) {
 640                 in = nf_bridge->physindev;
 641                 if (nf_bridge->mask & BRNF_PKT_TYPE) {
 642                         skb->pkt_type = PACKET_OTHERHOST;
 643                         nf_bridge->mask ^= BRNF_PKT_TYPE;
 644                 }
 645         } else {
 646                 in = *((struct net_device **)(skb->cb));
 647         }
 648         nf_bridge_update_protocol(skb);
 649         nf_bridge_push_encap_header(skb);
 650
 651         NF_HOOK_THRESH(NFPROTO_BRIDGE, NF_BR_FORWARD, skb, in,
 652                        skb->dev, br_forward_finish, 1);
 653         return 0;
 654 }
 655
 656 /* This is the 'purely bridged' case.  For IP, we pass the packet to
 657  * netfilter with indev and outdev set to the bridge device,
 658  * but we are still able to filter on the 'real' indev/outdev
 659  * because of the physdev module. For ARP, indev and outdev are the
 660  * bridge ports. */
 661 static unsigned int br_nf_forward_ip(unsigned int hook, struct sk_buff *skb,
 662                                      const struct net_device *in,
 663                                      const struct net_device *out,
 664                                      int (*okfn)(struct sk_buff *))
 665 {
 666         struct nf_bridge_info *nf_bridge;
 667         struct net_device *parent;
 668         u_int8_t pf;
 669
 670         if (!skb->nf_bridge)
 671                 return NF_ACCEPT;
 672
 673         /* Need exclusive nf_bridge_info since we might have multiple
 674          * different physoutdevs. */
 675         if (!nf_bridge_unshare(skb))
 676                 return NF_DROP;
 677
 678         parent = bridge_parent(out);
 679         if (!parent)
 680                 return NF_DROP;
 681
 682         if (skb->protocol == htons(ETH_P_IP) || IS_VLAN_IP(skb) ||
 683             IS_PPPOE_IP(skb))
 684                 pf = PF_INET;
 685         else if (skb->protocol == htons(ETH_P_IPV6) || IS_VLAN_IPV6(skb) ||
 686                  IS_PPPOE_IPV6(skb))
 687                 pf = PF_INET6;
 688         else
 689                 return NF_ACCEPT;
 690
 691         nf_bridge_pull_encap_header(skb);
 692
 693         nf_bridge = skb->nf_bridge;
 694         if (skb->pkt_type == PACKET_OTHERHOST) {
 695                 skb->pkt_type = PACKET_HOST;
 696                 nf_bridge->mask |= BRNF_PKT_TYPE;
 697         }
 698
 699         /* The physdev module checks on this */
 700         nf_bridge->mask |= BRNF_BRIDGED;
 701         nf_bridge->physoutdev = skb->dev;
 702         if (pf == PF_INET)
 703                 skb->protocol = htons(ETH_P_IP);
 704         else
 705                 skb->protocol = htons(ETH_P_IPV6);
 706
 707         NF_HOOK(pf, NF_INET_FORWARD, skb, bridge_parent(in), parent,
 708                 br_nf_forward_finish);
 709
 710         return NF_STOLEN;
 711 }
 712
 713 static unsigned int br_nf_forward_arp(unsigned int hook, struct sk_buff *skb,
 714                                       const struct net_device *in,
 715                                       const struct net_device *out,
 716                                       int (*okfn)(struct sk_buff *))
 717 {
 718         struct net_device **d = (struct net_device **)(skb->cb);
 719
 720 #ifdef CONFIG_SYSCTL
 721         if (!brnf_call_arptables)
 722                 return NF_ACCEPT;
 723 #endif
 724
 725         if (skb->protocol != htons(ETH_P_ARP)) {
 726                 if (!IS_VLAN_ARP(skb))
 727                         return NF_ACCEPT;
 728                 nf_bridge_pull_encap_header(skb);
 729         }
 730
 731         if (arp_hdr(skb)->ar_pln != 4) {
 732                 if (IS_VLAN_ARP(skb))
 733                         nf_bridge_push_encap_header(skb);
 734                 return NF_ACCEPT;
 735         }
 736         *d = (struct net_device *)in;
 737         NF_HOOK(NFPROTO_ARP, NF_ARP_FORWARD, skb, (struct net_device *)in,
 738                 (struct net_device *)out, br_nf_forward_finish);
 739
 740         return NF_STOLEN;
 741 }
 742
 743 #if defined(CONFIG_NF_CONNTRACK_IPV4) || defined(CONFIG_NF_CONNTRACK_IPV4_MODULE)
 744 static int br_nf_dev_queue_xmit(struct sk_buff *skb)
 745 {
 746         if (skb->nfct != NULL && skb->protocol == htons(ETH_P_IP) &&
 747             skb->len > skb->dev->mtu &&
 748             !skb_is_gso(skb))
 749                 return ip_fragment(skb, br_dev_queue_push_xmit);
 750         else
 751                 return br_dev_queue_push_xmit(skb);
 752 }
 753 #else
 754 static int br_nf_dev_queue_xmit(struct sk_buff *skb)
 755 {
 756         return br_dev_queue_push_xmit(skb);
 757 }
 758 #endif
 759
 760 /* PF_BRIDGE/POST_ROUTING ********************************************/
 761 static unsigned int br_nf_post_routing(unsigned int hook, struct sk_buff *skb,
 762                                        const struct net_device *in,
 763                                        const struct net_device *out,
 764                                        int (*okfn)(struct sk_buff *))
 765 {
 766         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 767         struct net_device *realoutdev = bridge_parent(skb->dev);
 768         u_int8_t pf;
 769
 770         if (!nf_bridge || !(nf_bridge->mask & BRNF_BRIDGED))
 771                 return NF_ACCEPT;
 772
 773         if (!realoutdev)
 774                 return NF_DROP;
 775
 776         if (skb->protocol == htons(ETH_P_IP) || IS_VLAN_IP(skb) ||
 777             IS_PPPOE_IP(skb))
 778                 pf = PF_INET;
 779         else if (skb->protocol == htons(ETH_P_IPV6) || IS_VLAN_IPV6(skb) ||
 780                  IS_PPPOE_IPV6(skb))
 781                 pf = PF_INET6;
 782         else
 783                 return NF_ACCEPT;
 784
 785         /* We assume any code from br_dev_queue_push_xmit onwards doesn't care
 786          * about the value of skb->pkt_type. */
 787         if (skb->pkt_type == PACKET_OTHERHOST) {
 788                 skb->pkt_type = PACKET_HOST;
 789                 nf_bridge->mask |= BRNF_PKT_TYPE;
 790         }
 791
 792         nf_bridge_pull_encap_header(skb);
 793         nf_bridge_save_header(skb);
 794         if (pf == PF_INET)
 795                 skb->protocol = htons(ETH_P_IP);
 796         else
 797                 skb->protocol = htons(ETH_P_IPV6);
 798
 799         NF_HOOK(pf, NF_INET_POST_ROUTING, skb, NULL, realoutdev,
 800                 br_nf_dev_queue_xmit);
 801
 802         return NF_STOLEN;
 803 }
 804
 805 /* IP/SABOTAGE *****************************************************/
 806 /* Don't hand locally destined packets to PF_INET(6)/PRE_ROUTING
 807  * for the second time. */
 808 static unsigned int ip_sabotage_in(unsigned int hook, struct sk_buff *skb,
 809                                    const struct net_device *in,
 810                                    const struct net_device *out,
 811                                    int (*okfn)(struct sk_buff *))
 812 {
 813         if (skb->nf_bridge &&
 814             !(skb->nf_bridge->mask & BRNF_NF_BRIDGE_PREROUTING)) {
 815                 return NF_STOP;
 816         }
 817
 818         return NF_ACCEPT;
 819 }
 820
 821 /* For br_nf_post_routing, we need (prio = NF_BR_PRI_LAST), because
 822  * br_dev_queue_push_xmit is called afterwards */
 823 static struct nf_hook_ops br_nf_ops[] __read_mostly = {
 824         {
 825                 .hook = br_nf_pre_routing,
 826                 .owner = THIS_MODULE,
 827                 .pf = PF_BRIDGE,
 828                 .hooknum = NF_BR_PRE_ROUTING,
 829                 .priority = NF_BR_PRI_BRNF,
 830         },
 831         {
 832                 .hook = br_nf_local_in,
 833                 .owner = THIS_MODULE,
 834                 .pf = PF_BRIDGE,
 835                 .hooknum = NF_BR_LOCAL_IN,
 836                 .priority = NF_BR_PRI_BRNF,
 837         },
 838         {
 839                 .hook = br_nf_forward_ip,
 840                 .owner = THIS_MODULE,
 841                 .pf = PF_BRIDGE,
 842                 .hooknum = NF_BR_FORWARD,
 843                 .priority = NF_BR_PRI_BRNF - 1,
 844         },
 845         {
 846                 .hook = br_nf_forward_arp,
 847                 .owner = THIS_MODULE,
 848                 .pf = PF_BRIDGE,
 849                 .hooknum = NF_BR_FORWARD,
 850                 .priority = NF_BR_PRI_BRNF,
 851         },
 852         {
 853                 .hook = br_nf_post_routing,
 854                 .owner = THIS_MODULE,
 855                 .pf = PF_BRIDGE,
 856                 .hooknum = NF_BR_POST_ROUTING,
 857                 .priority = NF_BR_PRI_LAST,
 858         },
 859         {
 860                 .hook = ip_sabotage_in,
 861                 .owner = THIS_MODULE,
 862                 .pf = PF_INET,
 863                 .hooknum = NF_INET_PRE_ROUTING,
 864                 .priority = NF_IP_PRI_FIRST,
 865         },
 866         {
 867                 .hook = ip_sabotage_in,
 868                 .owner = THIS_MODULE,
 869                 .pf = PF_INET6,
 870                 .hooknum = NF_INET_PRE_ROUTING,
 871                 .priority = NF_IP6_PRI_FIRST,
 872         },
 873 };
 874
 875 #ifdef CONFIG_SYSCTL
 876 static
 877 int brnf_sysctl_call_tables(ctl_table * ctl, int write,
 878                             void __user * buffer, size_t * lenp, loff_t * ppos)
 879 {
 880         int ret;
 881
 882         ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
 883
 884         if (write && *(int *)(ctl->data))
 885                 *(int *)(ctl->data) = 1;
 886         return ret;
 887 }
 888
 889 static ctl_table brnf_table[] = {
 890         {
 891                 .procname       = "bridge-nf-call-arptables",
 892                 .data           = &brnf_call_arptables,
 893                 .maxlen         = sizeof(int),
 894                 .mode           = 0644,
 895                 .proc_handler   = brnf_sysctl_call_tables,
 896         },
 897         {
 898                 .procname       = "bridge-nf-call-iptables",
 899                 .data           = &brnf_call_iptables,
 900                 .maxlen         = sizeof(int),
 901                 .mode           = 0644,
 902                 .proc_handler   = brnf_sysctl_call_tables,
 903         },
 904         {
 905                 .procname       = "bridge-nf-call-ip6tables",
 906                 .data           = &brnf_call_ip6tables,
 907                 .maxlen         = sizeof(int),
 908                 .mode           = 0644,
 909                 .proc_handler   = brnf_sysctl_call_tables,
 910         },
 911         {
 912                 .procname       = "bridge-nf-filter-vlan-tagged",
 913                 .data           = &brnf_filter_vlan_tagged,
 914                 .maxlen         = sizeof(int),
 915                 .mode           = 0644,
 916                 .proc_handler   = brnf_sysctl_call_tables,
 917         },
 918         {
 919                 .procname       = "bridge-nf-filter-pppoe-tagged",
 920                 .data           = &brnf_filter_pppoe_tagged,
 921                 .maxlen         = sizeof(int),
 922                 .mode           = 0644,
 923                 .proc_handler   = brnf_sysctl_call_tables,
 924         },
 925         { }
 926 };
 927
 928 static struct ctl_path brnf_path[] = {
 929         { .procname = "net", },
 930         { .procname = "bridge", },
 931         { }
 932 };
 933 #endif
 934
 935 int __init br_netfilter_init(void)
 936 {
 937         int ret;
 938
 939         ret = nf_register_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
 940         if (ret < 0)
 941                 return ret;
 942 #ifdef CONFIG_SYSCTL
 943         brnf_sysctl_header = register_sysctl_paths(brnf_path, brnf_table);
 944         if (brnf_sysctl_header == NULL) {
 945                 printk(KERN_WARNING
 946                        "br_netfilter: can't register to sysctl.\n");
 947                 nf_unregister_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
 948                 return -ENOMEM;
 949         }
 950 #endif
 951         printk(KERN_NOTICE "Bridge firewalling registered\n");
 952         return 0;
 953 }
 954
 955 void br_netfilter_fini(void)
 956 {
 957         nf_unregister_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
 958 #ifdef CONFIG_SYSCTL
 959         unregister_sysctl_table(brnf_sysctl_header);
 960 #endif
 961 }