git.oblomov.eu Git - 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 (maintainer)   <bdschuym@pandora.be>
   8  *
   9  *      Changes:
  10  *      Apr 29 2003: physdev module support (bdschuym)
  11  *      Jun 19 2003: let arptables see bridged ARP traffic (bdschuym)
  12  *      Oct 06 2003: filter encapsulated IP/ARP VLAN traffic on untagged bridge
  13  *                   (bdschuym)
  14  *      Sep 01 2004: add IPv6 filtering (bdschuym)
  15  *
  16  *      This program is free software; you can redistribute it and/or
  17  *      modify it under the terms of the GNU General Public License
  18  *      as published by the Free Software Foundation; either version
  19  *      2 of the License, or (at your option) any later version.
  20  *
  21  *      Lennert dedicates this file to Kerstin Wurdinger.
  22  */
  23
  24 #include <linux/module.h>
  25 #include <linux/kernel.h>
  26 #include <linux/ip.h>
  27 #include <linux/netdevice.h>
  28 #include <linux/skbuff.h>
  29 #include <linux/if_arp.h>
  30 #include <linux/if_ether.h>
  31 #include <linux/if_vlan.h>
  32 #include <linux/if_pppox.h>
  33 #include <linux/ppp_defs.h>
  34 #include <linux/netfilter_bridge.h>
  35 #include <linux/netfilter_ipv4.h>
  36 #include <linux/netfilter_ipv6.h>
  37 #include <linux/netfilter_arp.h>
  38 #include <linux/in_route.h>
  39 #include <linux/inetdevice.h>
  40
  41 #include <net/ip.h>
  42 #include <net/ipv6.h>
  43 #include <net/route.h>
  44
  45 #include <asm/uaccess.h>
  46 #include "br_private.h"
  47 #ifdef CONFIG_SYSCTL
  48 #include <linux/sysctl.h>
  49 #endif
  50
  51 #define skb_origaddr(skb)        (((struct bridge_skb_cb *) \
  52                                  (skb->nf_bridge->data))->daddr.ipv4)
  53 #define store_orig_dstaddr(skb)  (skb_origaddr(skb) = ip_hdr(skb)->daddr)
  54 #define dnat_took_place(skb)     (skb_origaddr(skb) != ip_hdr(skb)->daddr)
  55
  56 #ifdef CONFIG_SYSCTL
  57 static struct ctl_table_header *brnf_sysctl_header;
  58 static int brnf_call_iptables __read_mostly = 1;
  59 static int brnf_call_ip6tables __read_mostly = 1;
  60 static int brnf_call_arptables __read_mostly = 1;
  61 static int brnf_filter_vlan_tagged __read_mostly = 1;
  62 static int brnf_filter_pppoe_tagged __read_mostly = 1;
  63 #else
  64 #define brnf_filter_vlan_tagged 1
  65 #define brnf_filter_pppoe_tagged 1
  66 #endif
  67
  68 static inline __be16 vlan_proto(const struct sk_buff *skb)
  69 {
  70         return vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
  71 }
  72
  73 #define IS_VLAN_IP(skb) \
  74         (skb->protocol == htons(ETH_P_8021Q) && \
  75          vlan_proto(skb) == htons(ETH_P_IP) &&  \
  76          brnf_filter_vlan_tagged)
  77
  78 #define IS_VLAN_IPV6(skb) \
  79         (skb->protocol == htons(ETH_P_8021Q) && \
  80          vlan_proto(skb) == htons(ETH_P_IPV6) &&\
  81          brnf_filter_vlan_tagged)
  82
  83 #define IS_VLAN_ARP(skb) \
  84         (skb->protocol == htons(ETH_P_8021Q) && \
  85          vlan_proto(skb) == htons(ETH_P_ARP) && \
  86          brnf_filter_vlan_tagged)
  87
  88 static inline __be16 pppoe_proto(const struct sk_buff *skb)
  89 {
  90         return *((__be16 *)(skb_mac_header(skb) + ETH_HLEN +
  91                             sizeof(struct pppoe_hdr)));
  92 }
  93
  94 #define IS_PPPOE_IP(skb) \
  95         (skb->protocol == htons(ETH_P_PPP_SES) && \
  96          pppoe_proto(skb) == htons(PPP_IP) && \
  97          brnf_filter_pppoe_tagged)
  98
  99 #define IS_PPPOE_IPV6(skb) \
 100         (skb->protocol == htons(ETH_P_PPP_SES) && \
 101          pppoe_proto(skb) == htons(PPP_IPV6) && \
 102          brnf_filter_pppoe_tagged)
 103
 104 static void fake_update_pmtu(struct dst_entry *dst, u32 mtu)
 105 {
 106 }
 107
 108 static struct dst_ops fake_dst_ops = {
 109         .family =               AF_INET,
 110         .protocol =             __constant_htons(ETH_P_IP),
 111         .update_pmtu =          fake_update_pmtu,
 112         .entry_size =           sizeof(struct rtable),
 113         .entries =              ATOMIC_INIT(0),
 114 };
 115
 116 /*
 117  * Initialize bogus route table used to keep netfilter happy.
 118  * Currently, we fill in the PMTU entry because netfilter
 119  * refragmentation needs it, and the rt_flags entry because
 120  * ipt_REJECT needs it.  Future netfilter modules might
 121  * require us to fill additional fields.
 122  */
 123 void br_netfilter_rtable_init(struct net_bridge *br)
 124 {
 125         struct rtable *rt = &br->fake_rtable;
 126
 127         atomic_set(&rt->u.dst.__refcnt, 1);
 128         rt->u.dst.dev = br->dev;
 129         rt->u.dst.path = &rt->u.dst;
 130         rt->u.dst.metrics[RTAX_MTU - 1] = 1500;
 131         rt->u.dst.flags = DST_NOXFRM;
 132         rt->u.dst.ops = &fake_dst_ops;
 133 }
 134
 135 static inline struct rtable *bridge_parent_rtable(const struct net_device *dev)
 136 {
 137         struct net_bridge_port *port = rcu_dereference(dev->br_port);
 138
 139         return port ? &port->br->fake_rtable : NULL;
 140 }
 141
 142 static inline struct net_device *bridge_parent(const struct net_device *dev)
 143 {
 144         struct net_bridge_port *port = rcu_dereference(dev->br_port);
 145
 146         return port ? port->br->dev : NULL;
 147 }
 148
 149 static inline struct nf_bridge_info *nf_bridge_alloc(struct sk_buff *skb)
 150 {
 151         skb->nf_bridge = kzalloc(sizeof(struct nf_bridge_info), GFP_ATOMIC);
 152         if (likely(skb->nf_bridge))
 153                 atomic_set(&(skb->nf_bridge->use), 1);
 154
 155         return skb->nf_bridge;
 156 }
 157
 158 static inline struct nf_bridge_info *nf_bridge_unshare(struct sk_buff *skb)
 159 {
 160         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 161
 162         if (atomic_read(&nf_bridge->use) > 1) {
 163                 struct nf_bridge_info *tmp = nf_bridge_alloc(skb);
 164
 165                 if (tmp) {
 166                         memcpy(tmp, nf_bridge, sizeof(struct nf_bridge_info));
 167                         atomic_set(&tmp->use, 1);
 168                         nf_bridge_put(nf_bridge);
 169                 }
 170                 nf_bridge = tmp;
 171         }
 172         return nf_bridge;
 173 }
 174
 175 static inline void nf_bridge_push_encap_header(struct sk_buff *skb)
 176 {
 177         unsigned int len = nf_bridge_encap_header_len(skb);
 178
 179         skb_push(skb, len);
 180         skb->network_header -= len;
 181 }
 182
 183 static inline void nf_bridge_pull_encap_header(struct sk_buff *skb)
 184 {
 185         unsigned int len = nf_bridge_encap_header_len(skb);
 186
 187         skb_pull(skb, len);
 188         skb->network_header += len;
 189 }
 190
 191 static inline void nf_bridge_pull_encap_header_rcsum(struct sk_buff *skb)
 192 {
 193         unsigned int len = nf_bridge_encap_header_len(skb);
 194
 195         skb_pull_rcsum(skb, len);
 196         skb->network_header += len;
 197 }
 198
 199 static inline void nf_bridge_save_header(struct sk_buff *skb)
 200 {
 201         int header_size = ETH_HLEN + nf_bridge_encap_header_len(skb);
 202
 203         skb_copy_from_linear_data_offset(skb, -header_size,
 204                                          skb->nf_bridge->data, header_size);
 205 }
 206
 207 /*
 208  * When forwarding bridge frames, we save a copy of the original
 209  * header before processing.
 210  */
 211 int nf_bridge_copy_header(struct sk_buff *skb)
 212 {
 213         int err;
 214         int header_size = ETH_HLEN + nf_bridge_encap_header_len(skb);
 215
 216         err = skb_cow_head(skb, header_size);
 217         if (err)
 218                 return err;
 219
 220         skb_copy_to_linear_data_offset(skb, -header_size,
 221                                        skb->nf_bridge->data, header_size);
 222         __skb_push(skb, nf_bridge_encap_header_len(skb));
 223         return 0;
 224 }
 225
 226 /* PF_BRIDGE/PRE_ROUTING *********************************************/
 227 /* Undo the changes made for ip6tables PREROUTING and continue the
 228  * bridge PRE_ROUTING hook. */
 229 static int br_nf_pre_routing_finish_ipv6(struct sk_buff *skb)
 230 {
 231         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 232
 233         if (nf_bridge->mask & BRNF_PKT_TYPE) {
 234                 skb->pkt_type = PACKET_OTHERHOST;
 235                 nf_bridge->mask ^= BRNF_PKT_TYPE;
 236         }
 237         nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
 238
 239         skb->rtable = bridge_parent_rtable(nf_bridge->physindev);
 240         if (!skb->rtable) {
 241                 kfree_skb(skb);
 242                 return 0;
 243         }
 244         dst_hold(&skb->rtable->u.dst);
 245
 246         skb->dev = nf_bridge->physindev;
 247         nf_bridge_push_encap_header(skb);
 248         NF_HOOK_THRESH(PF_BRIDGE, NF_BR_PRE_ROUTING, skb, skb->dev, NULL,
 249                        br_handle_frame_finish, 1);
 250
 251         return 0;
 252 }
 253
 254 static void __br_dnat_complain(void)
 255 {
 256         static unsigned long last_complaint;
 257
 258         if (jiffies - last_complaint >= 5 * HZ) {
 259                 printk(KERN_WARNING "Performing cross-bridge DNAT requires IP "
 260                        "forwarding to be enabled\n");
 261                 last_complaint = jiffies;
 262         }
 263 }
 264
 265 /* This requires some explaining. If DNAT has taken place,
 266  * we will need to fix up the destination Ethernet address,
 267  * and this is a tricky process.
 268  *
 269  * There are two cases to consider:
 270  * 1. The packet was DNAT'ed to a device in the same bridge
 271  *    port group as it was received on. We can still bridge
 272  *    the packet.
 273  * 2. The packet was DNAT'ed to a different device, either
 274  *    a non-bridged device or another bridge port group.
 275  *    The packet will need to be routed.
 276  *
 277  * The correct way of distinguishing between these two cases is to
 278  * call ip_route_input() and to look at skb->dst->dev, which is
 279  * changed to the destination device if ip_route_input() succeeds.
 280  *
 281  * Let us first consider the case that ip_route_input() succeeds:
 282  *
 283  * If skb->dst->dev equals the logical bridge device the packet
 284  * came in on, we can consider this bridging. The packet is passed
 285  * through the neighbour output function to build a new destination
 286  * MAC address, which will make the packet enter br_nf_local_out()
 287  * not much later. In that function it is assured that the iptables
 288  * FORWARD chain is traversed for the packet.
 289  *
 290  * Otherwise, the packet is considered to be routed and we just
 291  * change the destination MAC address so that the packet will
 292  * later be passed up to the IP stack to be routed. For a redirected
 293  * packet, ip_route_input() will give back the localhost as output device,
 294  * which differs from the bridge device.
 295  *
 296  * Let us now consider the case that ip_route_input() fails:
 297  *
 298  * This can be because the destination address is martian, in which case
 299  * the packet will be dropped.
 300  * After a "echo '0' > /proc/sys/net/ipv4/ip_forward" ip_route_input()
 301  * will fail, while __ip_route_output_key() will return success. The source
 302  * address for __ip_route_output_key() is set to zero, so __ip_route_output_key
 303  * thinks we're handling a locally generated packet and won't care
 304  * if IP forwarding is allowed. We send a warning message to the users's
 305  * log telling her to put IP forwarding on.
 306  *
 307  * ip_route_input() will also fail if there is no route available.
 308  * In that case we just drop the packet.
 309  *
 310  * --Lennert, 20020411
 311  * --Bart, 20020416 (updated)
 312  * --Bart, 20021007 (updated)
 313  * --Bart, 20062711 (updated) */
 314 static int br_nf_pre_routing_finish_bridge(struct sk_buff *skb)
 315 {
 316         if (skb->pkt_type == PACKET_OTHERHOST) {
 317                 skb->pkt_type = PACKET_HOST;
 318                 skb->nf_bridge->mask |= BRNF_PKT_TYPE;
 319         }
 320         skb->nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
 321
 322         skb->dev = bridge_parent(skb->dev);
 323         if (skb->dev) {
 324                 struct dst_entry *dst = skb->dst;
 325
 326                 nf_bridge_pull_encap_header(skb);
 327
 328                 if (dst->hh)
 329                         return neigh_hh_output(dst->hh, skb);
 330                 else if (dst->neighbour)
 331                         return dst->neighbour->output(skb);
 332         }
 333         kfree_skb(skb);
 334         return 0;
 335 }
 336
 337 static int br_nf_pre_routing_finish(struct sk_buff *skb)
 338 {
 339         struct net_device *dev = skb->dev;
 340         struct iphdr *iph = ip_hdr(skb);
 341         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 342         int err;
 343
 344         if (nf_bridge->mask & BRNF_PKT_TYPE) {
 345                 skb->pkt_type = PACKET_OTHERHOST;
 346                 nf_bridge->mask ^= BRNF_PKT_TYPE;
 347         }
 348         nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
 349         if (dnat_took_place(skb)) {
 350                 if ((err = ip_route_input(skb, iph->daddr, iph->saddr, iph->tos, dev))) {
 351                         struct rtable *rt;
 352                         struct flowi fl = {
 353                                 .nl_u = {
 354                                         .ip4_u = {
 355                                                  .daddr = iph->daddr,
 356                                                  .saddr = 0,
 357                                                  .tos = RT_TOS(iph->tos) },
 358                                 },
 359                                 .proto = 0,
 360                         };
 361                         struct in_device *in_dev = in_dev_get(dev);
 362
 363                         /* If err equals -EHOSTUNREACH the error is due to a
 364                          * martian destination or due to the fact that
 365                          * forwarding is disabled. For most martian packets,
 366                          * ip_route_output_key() will fail. It won't fail for 2 types of
 367                          * martian destinations: loopback destinations and destination
 368                          * 0.0.0.0. In both cases the packet will be dropped because the
 369                          * destination is the loopback device and not the bridge. */
 370                         if (err != -EHOSTUNREACH || !in_dev || IN_DEV_FORWARD(in_dev))
 371                                 goto free_skb;
 372
 373                         if (!ip_route_output_key(&init_net, &rt, &fl)) {
 374                                 /* - Bridged-and-DNAT'ed traffic doesn't
 375                                  *   require ip_forwarding. */
 376                                 if (((struct dst_entry *)rt)->dev == dev) {
 377                                         skb->dst = (struct dst_entry *)rt;
 378                                         goto bridged_dnat;
 379                                 }
 380                                 /* we are sure that forwarding is disabled, so printing
 381                                  * this message is no problem. Note that the packet could
 382                                  * still have a martian destination address, in which case
 383                                  * the packet could be dropped even if forwarding were enabled */
 384                                 __br_dnat_complain();
 385                                 dst_release((struct dst_entry *)rt);
 386                         }
 387 free_skb:
 388                         kfree_skb(skb);
 389                         return 0;
 390                 } else {
 391                         if (skb->dst->dev == dev) {
 392 bridged_dnat:
 393                                 /* Tell br_nf_local_out this is a
 394                                  * bridged frame */
 395                                 nf_bridge->mask |= BRNF_BRIDGED_DNAT;
 396                                 skb->dev = nf_bridge->physindev;
 397                                 nf_bridge_push_encap_header(skb);
 398                                 NF_HOOK_THRESH(PF_BRIDGE, NF_BR_PRE_ROUTING,
 399                                                skb, skb->dev, NULL,
 400                                                br_nf_pre_routing_finish_bridge,
 401                                                1);
 402                                 return 0;
 403                         }
 404                         memcpy(eth_hdr(skb)->h_dest, dev->dev_addr, ETH_ALEN);
 405                         skb->pkt_type = PACKET_HOST;
 406                 }
 407         } else {
 408                 skb->rtable = bridge_parent_rtable(nf_bridge->physindev);
 409                 if (!skb->rtable) {
 410                         kfree_skb(skb);
 411                         return 0;
 412                 }
 413                 dst_hold(&skb->rtable->u.dst);
 414         }
 415
 416         skb->dev = nf_bridge->physindev;
 417         nf_bridge_push_encap_header(skb);
 418         NF_HOOK_THRESH(PF_BRIDGE, NF_BR_PRE_ROUTING, skb, skb->dev, NULL,
 419                        br_handle_frame_finish, 1);
 420
 421         return 0;
 422 }
 423
 424 /* Some common code for IPv4/IPv6 */
 425 static struct net_device *setup_pre_routing(struct sk_buff *skb)
 426 {
 427         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 428
 429         if (skb->pkt_type == PACKET_OTHERHOST) {
 430                 skb->pkt_type = PACKET_HOST;
 431                 nf_bridge->mask |= BRNF_PKT_TYPE;
 432         }
 433
 434         nf_bridge->mask |= BRNF_NF_BRIDGE_PREROUTING;
 435         nf_bridge->physindev = skb->dev;
 436         skb->dev = bridge_parent(skb->dev);
 437
 438         return skb->dev;
 439 }
 440
 441 /* We only check the length. A bridge shouldn't do any hop-by-hop stuff anyway */
 442 static int check_hbh_len(struct sk_buff *skb)
 443 {
 444         unsigned char *raw = (u8 *)(ipv6_hdr(skb) + 1);
 445         u32 pkt_len;
 446         const unsigned char *nh = skb_network_header(skb);
 447         int off = raw - nh;
 448         int len = (raw[1] + 1) << 3;
 449
 450         if ((raw + len) - skb->data > skb_headlen(skb))
 451                 goto bad;
 452
 453         off += 2;
 454         len -= 2;
 455
 456         while (len > 0) {
 457                 int optlen = nh[off + 1] + 2;
 458
 459                 switch (nh[off]) {
 460                 case IPV6_TLV_PAD0:
 461                         optlen = 1;
 462                         break;
 463
 464                 case IPV6_TLV_PADN:
 465                         break;
 466
 467                 case IPV6_TLV_JUMBO:
 468                         if (nh[off + 1] != 4 || (off & 3) != 2)
 469                                 goto bad;
 470                         pkt_len = ntohl(*(__be32 *) (nh + off + 2));
 471                         if (pkt_len <= IPV6_MAXPLEN ||
 472                             ipv6_hdr(skb)->payload_len)
 473                                 goto bad;
 474                         if (pkt_len > skb->len - sizeof(struct ipv6hdr))
 475                                 goto bad;
 476                         if (pskb_trim_rcsum(skb,
 477                                             pkt_len + sizeof(struct ipv6hdr)))
 478                                 goto bad;
 479                         nh = skb_network_header(skb);
 480                         break;
 481                 default:
 482                         if (optlen > len)
 483                                 goto bad;
 484                         break;
 485                 }
 486                 off += optlen;
 487                 len -= optlen;
 488         }
 489         if (len == 0)
 490                 return 0;
 491 bad:
 492         return -1;
 493
 494 }
 495
 496 /* Replicate the checks that IPv6 does on packet reception and pass the packet
 497  * to ip6tables, which doesn't support NAT, so things are fairly simple. */
 498 static unsigned int br_nf_pre_routing_ipv6(unsigned int hook,
 499                                            struct sk_buff *skb,
 500                                            const struct net_device *in,
 501                                            const struct net_device *out,
 502                                            int (*okfn)(struct sk_buff *))
 503 {
 504         struct ipv6hdr *hdr;
 505         u32 pkt_len;
 506
 507         if (skb->len < sizeof(struct ipv6hdr))
 508                 goto inhdr_error;
 509
 510         if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
 511                 goto inhdr_error;
 512
 513         hdr = ipv6_hdr(skb);
 514
 515         if (hdr->version != 6)
 516                 goto inhdr_error;
 517
 518         pkt_len = ntohs(hdr->payload_len);
 519
 520         if (pkt_len || hdr->nexthdr != NEXTHDR_HOP) {
 521                 if (pkt_len + sizeof(struct ipv6hdr) > skb->len)
 522                         goto inhdr_error;
 523                 if (pskb_trim_rcsum(skb, pkt_len + sizeof(struct ipv6hdr)))
 524                         goto inhdr_error;
 525         }
 526         if (hdr->nexthdr == NEXTHDR_HOP && check_hbh_len(skb))
 527                 goto inhdr_error;
 528
 529         nf_bridge_put(skb->nf_bridge);
 530         if (!nf_bridge_alloc(skb))
 531                 return NF_DROP;
 532         if (!setup_pre_routing(skb))
 533                 return NF_DROP;
 534
 535         NF_HOOK(PF_INET6, NF_INET_PRE_ROUTING, skb, skb->dev, NULL,
 536                 br_nf_pre_routing_finish_ipv6);
 537
 538         return NF_STOLEN;
 539
 540 inhdr_error:
 541         return NF_DROP;
 542 }
 543
 544 /* Direct IPv6 traffic to br_nf_pre_routing_ipv6.
 545  * Replicate the checks that IPv4 does on packet reception.
 546  * Set skb->dev to the bridge device (i.e. parent of the
 547  * receiving device) to make netfilter happy, the REDIRECT
 548  * target in particular.  Save the original destination IP
 549  * address to be able to detect DNAT afterwards. */
 550 static unsigned int br_nf_pre_routing(unsigned int hook, struct sk_buff *skb,
 551                                       const struct net_device *in,
 552                                       const struct net_device *out,
 553                                       int (*okfn)(struct sk_buff *))
 554 {
 555         struct iphdr *iph;
 556         __u32 len = nf_bridge_encap_header_len(skb);
 557
 558         if (unlikely(!pskb_may_pull(skb, len)))
 559                 goto out;
 560
 561         if (skb->protocol == htons(ETH_P_IPV6) || IS_VLAN_IPV6(skb) ||
 562             IS_PPPOE_IPV6(skb)) {
 563 #ifdef CONFIG_SYSCTL
 564                 if (!brnf_call_ip6tables)
 565                         return NF_ACCEPT;
 566 #endif
 567                 nf_bridge_pull_encap_header_rcsum(skb);
 568                 return br_nf_pre_routing_ipv6(hook, skb, in, out, okfn);
 569         }
 570 #ifdef CONFIG_SYSCTL
 571         if (!brnf_call_iptables)
 572                 return NF_ACCEPT;
 573 #endif
 574
 575         if (skb->protocol != htons(ETH_P_IP) && !IS_VLAN_IP(skb) &&
 576             !IS_PPPOE_IP(skb))
 577                 return NF_ACCEPT;
 578
 579         nf_bridge_pull_encap_header_rcsum(skb);
 580
 581         if (!pskb_may_pull(skb, sizeof(struct iphdr)))
 582                 goto inhdr_error;
 583
 584         iph = ip_hdr(skb);
 585         if (iph->ihl < 5 || iph->version != 4)
 586                 goto inhdr_error;
 587
 588         if (!pskb_may_pull(skb, 4 * iph->ihl))
 589                 goto inhdr_error;
 590
 591         iph = ip_hdr(skb);
 592         if (ip_fast_csum((__u8 *) iph, iph->ihl) != 0)
 593                 goto inhdr_error;
 594
 595         len = ntohs(iph->tot_len);
 596         if (skb->len < len || len < 4 * iph->ihl)
 597                 goto inhdr_error;
 598
 599         pskb_trim_rcsum(skb, len);
 600
 601         nf_bridge_put(skb->nf_bridge);
 602         if (!nf_bridge_alloc(skb))
 603                 return NF_DROP;
 604         if (!setup_pre_routing(skb))
 605                 return NF_DROP;
 606         store_orig_dstaddr(skb);
 607
 608         NF_HOOK(PF_INET, NF_INET_PRE_ROUTING, skb, skb->dev, NULL,
 609                 br_nf_pre_routing_finish);
 610
 611         return NF_STOLEN;
 612
 613 inhdr_error:
 614 //      IP_INC_STATS_BH(IpInHdrErrors);
 615 out:
 616         return NF_DROP;
 617 }
 618
 619
 620 /* PF_BRIDGE/LOCAL_IN ************************************************/
 621 /* The packet is locally destined, which requires a real
 622  * dst_entry, so detach the fake one.  On the way up, the
 623  * packet would pass through PRE_ROUTING again (which already
 624  * took place when the packet entered the bridge), but we
 625  * register an IPv4 PRE_ROUTING 'sabotage' hook that will
 626  * prevent this from happening. */
 627 static unsigned int br_nf_local_in(unsigned int hook, struct sk_buff *skb,
 628                                    const struct net_device *in,
 629                                    const struct net_device *out,
 630                                    int (*okfn)(struct sk_buff *))
 631 {
 632         if (skb->rtable && skb->rtable == bridge_parent_rtable(in)) {
 633                 dst_release(&skb->rtable->u.dst);
 634                 skb->rtable = NULL;
 635         }
 636
 637         return NF_ACCEPT;
 638 }
 639
 640 /* PF_BRIDGE/FORWARD *************************************************/
 641 static int br_nf_forward_finish(struct sk_buff *skb)
 642 {
 643         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 644         struct net_device *in;
 645
 646         if (skb->protocol != htons(ETH_P_ARP) && !IS_VLAN_ARP(skb)) {
 647                 in = nf_bridge->physindev;
 648                 if (nf_bridge->mask & BRNF_PKT_TYPE) {
 649                         skb->pkt_type = PACKET_OTHERHOST;
 650                         nf_bridge->mask ^= BRNF_PKT_TYPE;
 651                 }
 652         } else {
 653                 in = *((struct net_device **)(skb->cb));
 654         }
 655         nf_bridge_push_encap_header(skb);
 656         NF_HOOK_THRESH(PF_BRIDGE, NF_BR_FORWARD, skb, in,
 657                        skb->dev, br_forward_finish, 1);
 658         return 0;
 659 }
 660
 661 /* This is the 'purely bridged' case.  For IP, we pass the packet to
 662  * netfilter with indev and outdev set to the bridge device,
 663  * but we are still able to filter on the 'real' indev/outdev
 664  * because of the physdev module. For ARP, indev and outdev are the
 665  * bridge ports. */
 666 static unsigned int br_nf_forward_ip(unsigned int hook, struct sk_buff *skb,
 667                                      const struct net_device *in,
 668                                      const struct net_device *out,
 669                                      int (*okfn)(struct sk_buff *))
 670 {
 671         struct nf_bridge_info *nf_bridge;
 672         struct net_device *parent;
 673         u_int8_t pf;
 674
 675         if (!skb->nf_bridge)
 676                 return NF_ACCEPT;
 677
 678         /* Need exclusive nf_bridge_info since we might have multiple
 679          * different physoutdevs. */
 680         if (!nf_bridge_unshare(skb))
 681                 return NF_DROP;
 682
 683         parent = bridge_parent(out);
 684         if (!parent)
 685                 return NF_DROP;
 686
 687         if (skb->protocol == htons(ETH_P_IP) || IS_VLAN_IP(skb) ||
 688             IS_PPPOE_IP(skb))
 689                 pf = PF_INET;
 690         else
 691                 pf = PF_INET6;
 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
 705         NF_HOOK(pf, NF_INET_FORWARD, skb, bridge_parent(in), parent,
 706                 br_nf_forward_finish);
 707
 708         return NF_STOLEN;
 709 }
 710
 711 static unsigned int br_nf_forward_arp(unsigned int hook, struct sk_buff *skb,
 712                                       const struct net_device *in,
 713                                       const struct net_device *out,
 714                                       int (*okfn)(struct sk_buff *))
 715 {
 716         struct net_device **d = (struct net_device **)(skb->cb);
 717
 718 #ifdef CONFIG_SYSCTL
 719         if (!brnf_call_arptables)
 720                 return NF_ACCEPT;
 721 #endif
 722
 723         if (skb->protocol != htons(ETH_P_ARP)) {
 724                 if (!IS_VLAN_ARP(skb))
 725                         return NF_ACCEPT;
 726                 nf_bridge_pull_encap_header(skb);
 727         }
 728
 729         if (arp_hdr(skb)->ar_pln != 4) {
 730                 if (IS_VLAN_ARP(skb))
 731                         nf_bridge_push_encap_header(skb);
 732                 return NF_ACCEPT;
 733         }
 734         *d = (struct net_device *)in;
 735         NF_HOOK(NFPROTO_ARP, NF_ARP_FORWARD, skb, (struct net_device *)in,
 736                 (struct net_device *)out, br_nf_forward_finish);
 737
 738         return NF_STOLEN;
 739 }
 740
 741 /* PF_BRIDGE/LOCAL_OUT ***********************************************
 742  *
 743  * This function sees both locally originated IP packets and forwarded
 744  * IP packets (in both cases the destination device is a bridge
 745  * device). It also sees bridged-and-DNAT'ed packets.
 746  *
 747  * If (nf_bridge->mask & BRNF_BRIDGED_DNAT) then the packet is bridged
 748  * and we fake the PF_BRIDGE/FORWARD hook. The function br_nf_forward()
 749  * will then fake the PF_INET/FORWARD hook. br_nf_local_out() has priority
 750  * NF_BR_PRI_FIRST, so no relevant PF_BRIDGE/INPUT functions have been nor
 751  * will be executed.
 752  */
 753 static unsigned int br_nf_local_out(unsigned int hook, struct sk_buff *skb,
 754                                     const struct net_device *in,
 755                                     const struct net_device *out,
 756                                     int (*okfn)(struct sk_buff *))
 757 {
 758         struct net_device *realindev;
 759         struct nf_bridge_info *nf_bridge;
 760
 761         if (!skb->nf_bridge)
 762                 return NF_ACCEPT;
 763
 764         /* Need exclusive nf_bridge_info since we might have multiple
 765          * different physoutdevs. */
 766         if (!nf_bridge_unshare(skb))
 767                 return NF_DROP;
 768
 769         nf_bridge = skb->nf_bridge;
 770         if (!(nf_bridge->mask & BRNF_BRIDGED_DNAT))
 771                 return NF_ACCEPT;
 772
 773         /* Bridged, take PF_BRIDGE/FORWARD.
 774          * (see big note in front of br_nf_pre_routing_finish) */
 775         nf_bridge->physoutdev = skb->dev;
 776         realindev = nf_bridge->physindev;
 777
 778         if (nf_bridge->mask & BRNF_PKT_TYPE) {
 779                 skb->pkt_type = PACKET_OTHERHOST;
 780                 nf_bridge->mask ^= BRNF_PKT_TYPE;
 781         }
 782         nf_bridge_push_encap_header(skb);
 783
 784         NF_HOOK(PF_BRIDGE, NF_BR_FORWARD, skb, realindev, skb->dev,
 785                 br_forward_finish);
 786         return NF_STOLEN;
 787 }
 788
 789 static int br_nf_dev_queue_xmit(struct sk_buff *skb)
 790 {
 791         if (skb->protocol == htons(ETH_P_IP) &&
 792             skb->len > skb->dev->mtu &&
 793             !skb_is_gso(skb))
 794                 return ip_fragment(skb, br_dev_queue_push_xmit);
 795         else
 796                 return br_dev_queue_push_xmit(skb);
 797 }
 798
 799 /* PF_BRIDGE/POST_ROUTING ********************************************/
 800 static unsigned int br_nf_post_routing(unsigned int hook, struct sk_buff *skb,
 801                                        const struct net_device *in,
 802                                        const struct net_device *out,
 803                                        int (*okfn)(struct sk_buff *))
 804 {
 805         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 806         struct net_device *realoutdev = bridge_parent(skb->dev);
 807         u_int8_t pf;
 808
 809 #ifdef CONFIG_NETFILTER_DEBUG
 810         /* Be very paranoid. This probably won't happen anymore, but let's
 811          * keep the check just to be sure... */
 812         if (skb_mac_header(skb) < skb->head ||
 813             skb_mac_header(skb) + ETH_HLEN > skb->data) {
 814                 printk(KERN_CRIT "br_netfilter: Argh!! br_nf_post_routing: "
 815                        "bad mac.raw pointer.\n");
 816                 goto print_error;
 817         }
 818 #endif
 819
 820         if (!nf_bridge)
 821                 return NF_ACCEPT;
 822
 823         if (!(nf_bridge->mask & (BRNF_BRIDGED | BRNF_BRIDGED_DNAT)))
 824                 return NF_ACCEPT;
 825
 826         if (!realoutdev)
 827                 return NF_DROP;
 828
 829         if (skb->protocol == htons(ETH_P_IP) || IS_VLAN_IP(skb) ||
 830             IS_PPPOE_IP(skb))
 831                 pf = PF_INET;
 832         else
 833                 pf = PF_INET6;
 834
 835 #ifdef CONFIG_NETFILTER_DEBUG
 836         if (skb->dst == NULL) {
 837                 printk(KERN_INFO "br_netfilter post_routing: skb->dst == NULL\n");
 838                 goto print_error;
 839         }
 840 #endif
 841
 842         /* We assume any code from br_dev_queue_push_xmit onwards doesn't care
 843          * about the value of skb->pkt_type. */
 844         if (skb->pkt_type == PACKET_OTHERHOST) {
 845                 skb->pkt_type = PACKET_HOST;
 846                 nf_bridge->mask |= BRNF_PKT_TYPE;
 847         }
 848
 849         nf_bridge_pull_encap_header(skb);
 850         nf_bridge_save_header(skb);
 851
 852         NF_HOOK(pf, NF_INET_POST_ROUTING, skb, NULL, realoutdev,
 853                 br_nf_dev_queue_xmit);
 854
 855         return NF_STOLEN;
 856
 857 #ifdef CONFIG_NETFILTER_DEBUG
 858 print_error:
 859         if (skb->dev != NULL) {
 860                 printk("[%s]", skb->dev->name);
 861                 if (realoutdev)
 862                         printk("[%s]", realoutdev->name);
 863         }
 864         printk(" head:%p, raw:%p, data:%p\n", skb->head, skb_mac_header(skb),
 865                skb->data);
 866         dump_stack();
 867         return NF_ACCEPT;
 868 #endif
 869 }
 870
 871 /* IP/SABOTAGE *****************************************************/
 872 /* Don't hand locally destined packets to PF_INET(6)/PRE_ROUTING
 873  * for the second time. */
 874 static unsigned int ip_sabotage_in(unsigned int hook, struct sk_buff *skb,
 875                                    const struct net_device *in,
 876                                    const struct net_device *out,
 877                                    int (*okfn)(struct sk_buff *))
 878 {
 879         if (skb->nf_bridge &&
 880             !(skb->nf_bridge->mask & BRNF_NF_BRIDGE_PREROUTING)) {
 881                 return NF_STOP;
 882         }
 883
 884         return NF_ACCEPT;
 885 }
 886
 887 /* For br_nf_local_out we need (prio = NF_BR_PRI_FIRST), to insure that innocent
 888  * PF_BRIDGE/NF_BR_LOCAL_OUT functions don't get bridged traffic as input.
 889  * For br_nf_post_routing, we need (prio = NF_BR_PRI_LAST), because
 890  * ip_refrag() can return NF_STOLEN. */
 891 static struct nf_hook_ops br_nf_ops[] __read_mostly = {
 892         { .hook = br_nf_pre_routing,
 893           .owner = THIS_MODULE,
 894           .pf = PF_BRIDGE,
 895           .hooknum = NF_BR_PRE_ROUTING,
 896           .priority = NF_BR_PRI_BRNF, },
 897         { .hook = br_nf_local_in,
 898           .owner = THIS_MODULE,
 899           .pf = PF_BRIDGE,
 900           .hooknum = NF_BR_LOCAL_IN,
 901           .priority = NF_BR_PRI_BRNF, },
 902         { .hook = br_nf_forward_ip,
 903           .owner = THIS_MODULE,
 904           .pf = PF_BRIDGE,
 905           .hooknum = NF_BR_FORWARD,
 906           .priority = NF_BR_PRI_BRNF - 1, },
 907         { .hook = br_nf_forward_arp,
 908           .owner = THIS_MODULE,
 909           .pf = PF_BRIDGE,
 910           .hooknum = NF_BR_FORWARD,
 911           .priority = NF_BR_PRI_BRNF, },
 912         { .hook = br_nf_local_out,
 913           .owner = THIS_MODULE,
 914           .pf = PF_BRIDGE,
 915           .hooknum = NF_BR_LOCAL_OUT,
 916           .priority = NF_BR_PRI_FIRST, },
 917         { .hook = br_nf_post_routing,
 918           .owner = THIS_MODULE,
 919           .pf = PF_BRIDGE,
 920           .hooknum = NF_BR_POST_ROUTING,
 921           .priority = NF_BR_PRI_LAST, },
 922         { .hook = ip_sabotage_in,
 923           .owner = THIS_MODULE,
 924           .pf = PF_INET,
 925           .hooknum = NF_INET_PRE_ROUTING,
 926           .priority = NF_IP_PRI_FIRST, },
 927         { .hook = ip_sabotage_in,
 928           .owner = THIS_MODULE,
 929           .pf = PF_INET6,
 930           .hooknum = NF_INET_PRE_ROUTING,
 931           .priority = NF_IP6_PRI_FIRST, },
 932 };
 933
 934 #ifdef CONFIG_SYSCTL
 935 static
 936 int brnf_sysctl_call_tables(ctl_table * ctl, int write, struct file *filp,
 937                             void __user * buffer, size_t * lenp, loff_t * ppos)
 938 {
 939         int ret;
 940
 941         ret = proc_dointvec(ctl, write, filp, buffer, lenp, ppos);
 942
 943         if (write && *(int *)(ctl->data))
 944                 *(int *)(ctl->data) = 1;
 945         return ret;
 946 }
 947
 948 static ctl_table brnf_table[] = {
 949         {
 950                 .procname       = "bridge-nf-call-arptables",
 951                 .data           = &brnf_call_arptables,
 952                 .maxlen         = sizeof(int),
 953                 .mode           = 0644,
 954                 .proc_handler   = &brnf_sysctl_call_tables,
 955         },
 956         {
 957                 .procname       = "bridge-nf-call-iptables",
 958                 .data           = &brnf_call_iptables,
 959                 .maxlen         = sizeof(int),
 960                 .mode           = 0644,
 961                 .proc_handler   = &brnf_sysctl_call_tables,
 962         },
 963         {
 964                 .procname       = "bridge-nf-call-ip6tables",
 965                 .data           = &brnf_call_ip6tables,
 966                 .maxlen         = sizeof(int),
 967                 .mode           = 0644,
 968                 .proc_handler   = &brnf_sysctl_call_tables,
 969         },
 970         {
 971                 .procname       = "bridge-nf-filter-vlan-tagged",
 972                 .data           = &brnf_filter_vlan_tagged,
 973                 .maxlen         = sizeof(int),
 974                 .mode           = 0644,
 975                 .proc_handler   = &brnf_sysctl_call_tables,
 976         },
 977         {
 978                 .procname       = "bridge-nf-filter-pppoe-tagged",
 979                 .data           = &brnf_filter_pppoe_tagged,
 980                 .maxlen         = sizeof(int),
 981                 .mode           = 0644,
 982                 .proc_handler   = &brnf_sysctl_call_tables,
 983         },
 984         { .ctl_name = 0 }
 985 };
 986
 987 static struct ctl_path brnf_path[] = {
 988         { .procname = "net", .ctl_name = CTL_NET, },
 989         { .procname = "bridge", .ctl_name = NET_BRIDGE, },
 990         { }
 991 };
 992 #endif
 993
 994 int __init br_netfilter_init(void)
 995 {
 996         int ret;
 997
 998         ret = nf_register_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
 999         if (ret < 0)
1000                 return ret;
1001 #ifdef CONFIG_SYSCTL
1002         brnf_sysctl_header = register_sysctl_paths(brnf_path, brnf_table);
1003         if (brnf_sysctl_header == NULL) {
1004                 printk(KERN_WARNING
1005                        "br_netfilter: can't register to sysctl.\n");
1006                 nf_unregister_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
1007                 return -ENOMEM;
1008         }
1009 #endif
1010         printk(KERN_NOTICE "Bridge firewalling registered\n");
1011         return 0;
1012 }
1013
1014 void br_netfilter_fini(void)
1015 {
1016         nf_unregister_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
1017 #ifdef CONFIG_SYSCTL
1018         unregister_sysctl_table(brnf_sysctl_header);
1019 #endif
1020 }