net/bridge/br_netfilter.c

   1 /*
   2  *      Handle firewalling
   3  *      Linux ethernet bridge
   4  *
   5  *      Authors:
   6  *      Lennert Buytenhek               <buytenh@gnu.org>
   7  *      Bart De Schuymer                <bdschuym@pandora.be>
   8  *
   9  *      This program is free software; you can redistribute it and/or
  10  *      modify it under the terms of the GNU General Public License
  11  *      as published by the Free Software Foundation; either version
  12  *      2 of the License, or (at your option) any later version.
  13  *
  14  *      Lennert dedicates this file to Kerstin Wurdinger.
  15  */
  16
  17 #include <linux/module.h>
  18 #include <linux/kernel.h>
  19 #include <linux/slab.h>
  20 #include <linux/ip.h>
  21 #include <linux/netdevice.h>
  22 #include <linux/skbuff.h>
  23 #include <linux/if_arp.h>
  24 #include <linux/if_ether.h>
  25 #include <linux/if_vlan.h>
  26 #include <linux/if_pppox.h>
  27 #include <linux/ppp_defs.h>
  28 #include <linux/netfilter_bridge.h>
  29 #include <linux/netfilter_ipv4.h>
  30 #include <linux/netfilter_ipv6.h>
  31 #include <linux/netfilter_arp.h>
  32 #include <linux/in_route.h>
  33 #include <linux/inetdevice.h>
  34
  35 #include <net/ip.h>
  36 #include <net/ipv6.h>
  37 #include <net/route.h>
  38
  39 #include <asm/uaccess.h>
  40 #include "br_private.h"
  41 #ifdef CONFIG_SYSCTL
  42 #include <linux/sysctl.h>
  43 #endif
  44
  45 #define skb_origaddr(skb)        (((struct bridge_skb_cb *) \
  46                                  (skb->nf_bridge->data))->daddr.ipv4)
  47 #define store_orig_dstaddr(skb)  (skb_origaddr(skb) = ip_hdr(skb)->daddr)
  48 #define dnat_took_place(skb)     (skb_origaddr(skb) != ip_hdr(skb)->daddr)
  49
  50 #ifdef CONFIG_SYSCTL
  51 static struct ctl_table_header *brnf_sysctl_header;
  52 static int brnf_call_iptables __read_mostly = 1;
  53 static int brnf_call_ip6tables __read_mostly = 1;
  54 static int brnf_call_arptables __read_mostly = 1;
  55 static int brnf_filter_vlan_tagged __read_mostly = 0;
  56 static int brnf_filter_pppoe_tagged __read_mostly = 0;
  57 #else
  58 #define brnf_filter_vlan_tagged 0
  59 #define brnf_filter_pppoe_tagged 0
  60 #endif
  61
  62 static inline __be16 vlan_proto(const struct sk_buff *skb)
  63 {
  64         return vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
  65 }
  66
  67 #define IS_VLAN_IP(skb) \
  68         (skb->protocol == htons(ETH_P_8021Q) && \
  69          vlan_proto(skb) == htons(ETH_P_IP) &&  \
  70          brnf_filter_vlan_tagged)
  71
  72 #define IS_VLAN_IPV6(skb) \
  73         (skb->protocol == htons(ETH_P_8021Q) && \
  74          vlan_proto(skb) == htons(ETH_P_IPV6) &&\
  75          brnf_filter_vlan_tagged)
  76
  77 #define IS_VLAN_ARP(skb) \
  78         (skb->protocol == htons(ETH_P_8021Q) && \
  79          vlan_proto(skb) == htons(ETH_P_ARP) && \
  80          brnf_filter_vlan_tagged)
  81
  82 static inline __be16 pppoe_proto(const struct sk_buff *skb)
  83 {
  84         return *((__be16 *)(skb_mac_header(skb) + ETH_HLEN +
  85                             sizeof(struct pppoe_hdr)));
  86 }
  87
  88 #define IS_PPPOE_IP(skb) \
  89         (skb->protocol == htons(ETH_P_PPP_SES) && \
  90          pppoe_proto(skb) == htons(PPP_IP) && \
  91          brnf_filter_pppoe_tagged)
  92
  93 #define IS_PPPOE_IPV6(skb) \
  94         (skb->protocol == htons(ETH_P_PPP_SES) && \
  95          pppoe_proto(skb) == htons(PPP_IPV6) && \
  96          brnf_filter_pppoe_tagged)
  97
  98 static void fake_update_pmtu(struct dst_entry *dst, u32 mtu)
  99 {
 100 }
 101
 102 static struct dst_ops fake_dst_ops = {
 103         .family =               AF_INET,
 104         .protocol =             cpu_to_be16(ETH_P_IP),
 105         .update_pmtu =          fake_update_pmtu,
 106         .entries =              ATOMIC_INIT(0),
 107 };
 108
 109 /*
 110  * Initialize bogus route table used to keep netfilter happy.
 111  * Currently, we fill in the PMTU entry because netfilter
 112  * refragmentation needs it, and the rt_flags entry because
 113  * ipt_REJECT needs it.  Future netfilter modules might
 114  * require us to fill additional fields.
 115  */
 116 void br_netfilter_rtable_init(struct net_bridge *br)
 117 {
 118         struct rtable *rt = &br->fake_rtable;
 119
 120         atomic_set(&rt->u.dst.__refcnt, 1);
 121         rt->u.dst.dev = br->dev;
 122         rt->u.dst.path = &rt->u.dst;
 123         rt->u.dst.metrics[RTAX_MTU - 1] = 1500;
 124         rt->u.dst.flags = DST_NOXFRM;
 125         rt->u.dst.ops = &fake_dst_ops;
 126 }
 127
 128 static inline struct rtable *bridge_parent_rtable(const struct net_device *dev)
 129 {
 130         struct net_bridge_port *port = rcu_dereference(dev->br_port);
 131
 132         return port ? &port->br->fake_rtable : NULL;
 133 }
 134
 135 static inline struct net_device *bridge_parent(const struct net_device *dev)
 136 {
 137         struct net_bridge_port *port = rcu_dereference(dev->br_port);
 138
 139         return port ? port->br->dev : NULL;
 140 }
 141
 142 static inline struct nf_bridge_info *nf_bridge_alloc(struct sk_buff *skb)
 143 {
 144         skb->nf_bridge = kzalloc(sizeof(struct nf_bridge_info), GFP_ATOMIC);
 145         if (likely(skb->nf_bridge))
 146                 atomic_set(&(skb->nf_bridge->use), 1);
 147
 148         return skb->nf_bridge;
 149 }
 150
 151 static inline struct nf_bridge_info *nf_bridge_unshare(struct sk_buff *skb)
 152 {
 153         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 154
 155         if (atomic_read(&nf_bridge->use) > 1) {
 156                 struct nf_bridge_info *tmp = nf_bridge_alloc(skb);
 157
 158                 if (tmp) {
 159                         memcpy(tmp, nf_bridge, sizeof(struct nf_bridge_info));
 160                         atomic_set(&tmp->use, 1);
 161                         nf_bridge_put(nf_bridge);
 162                 }
 163                 nf_bridge = tmp;
 164         }
 165         return nf_bridge;
 166 }
 167
 168 static inline void nf_bridge_push_encap_header(struct sk_buff *skb)
 169 {
 170         unsigned int len = nf_bridge_encap_header_len(skb);
 171
 172         skb_push(skb, len);
 173         skb->network_header -= len;
 174 }
 175
 176 static inline void nf_bridge_pull_encap_header(struct sk_buff *skb)
 177 {
 178         unsigned int len = nf_bridge_encap_header_len(skb);
 179
 180         skb_pull(skb, len);
 181         skb->network_header += len;
 182 }
 183
 184 static inline void nf_bridge_pull_encap_header_rcsum(struct sk_buff *skb)
 185 {
 186         unsigned int len = nf_bridge_encap_header_len(skb);
 187
 188         skb_pull_rcsum(skb, len);
 189         skb->network_header += len;
 190 }
 191
 192 static inline void nf_bridge_save_header(struct sk_buff *skb)
 193 {
 194         int header_size = ETH_HLEN + nf_bridge_encap_header_len(skb);
 195
 196         skb_copy_from_linear_data_offset(skb, -header_size,
 197                                          skb->nf_bridge->data, header_size);
 198 }
 199
 200 static inline void nf_bridge_update_protocol(struct sk_buff *skb)
 201 {
 202         if (skb->nf_bridge->mask & BRNF_8021Q)
 203                 skb->protocol = htons(ETH_P_8021Q);
 204         else if (skb->nf_bridge->mask & BRNF_PPPoE)
 205                 skb->protocol = htons(ETH_P_PPP_SES);
 206 }
 207
 208 /* Fill in the header for fragmented IP packets handled by
 209  * the IPv4 connection tracking code.
 210  */
 211 int nf_bridge_copy_header(struct sk_buff *skb)
 212 {
 213         int err;
 214         unsigned int header_size;
 215
 216         nf_bridge_update_protocol(skb);
 217         header_size = ETH_HLEN + nf_bridge_encap_header_len(skb);
 218         err = skb_cow_head(skb, header_size);
 219         if (err)
 220                 return err;
 221
 222         skb_copy_to_linear_data_offset(skb, -header_size,
 223                                        skb->nf_bridge->data, header_size);
 224         __skb_push(skb, nf_bridge_encap_header_len(skb));
 225         return 0;
 226 }
 227
 228 /* PF_BRIDGE/PRE_ROUTING *********************************************/
 229 /* Undo the changes made for ip6tables PREROUTING and continue the
 230  * bridge PRE_ROUTING hook. */
 231 static int br_nf_pre_routing_finish_ipv6(struct sk_buff *skb)
 232 {
 233         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 234         struct rtable *rt;
 235
 236         if (nf_bridge->mask & BRNF_PKT_TYPE) {
 237                 skb->pkt_type = PACKET_OTHERHOST;
 238                 nf_bridge->mask ^= BRNF_PKT_TYPE;
 239         }
 240         nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
 241
 242         rt = bridge_parent_rtable(nf_bridge->physindev);
 243         if (!rt) {
 244                 kfree_skb(skb);
 245                 return 0;
 246         }
 247         dst_hold(&rt->u.dst);
 248         skb_dst_set(skb, &rt->u.dst);
 249
 250         skb->dev = nf_bridge->physindev;
 251         nf_bridge_update_protocol(skb);
 252         nf_bridge_push_encap_header(skb);
 253         NF_HOOK_THRESH(NFPROTO_BRIDGE, NF_BR_PRE_ROUTING, skb, skb->dev, NULL,
 254                        br_handle_frame_finish, 1);
 255
 256         return 0;
 257 }
 258
 259 /* Obtain the correct destination MAC address, while preserving the original
 260  * source MAC address. If we already know this address, we just copy it. If we
 261  * don't, we use the neighbour framework to find out. In both cases, we make
 262  * sure that br_handle_frame_finish() is called afterwards.
 263  */
 264 static int br_nf_pre_routing_finish_bridge(struct sk_buff *skb)
 265 {
 266         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 267         struct dst_entry *dst;
 268
 269         skb->dev = bridge_parent(skb->dev);
 270         if (!skb->dev)
 271                 goto free_skb;
 272         dst = skb_dst(skb);
 273         if (dst->hh) {
 274                 neigh_hh_bridge(dst->hh, skb);
 275                 skb->dev = nf_bridge->physindev;
 276                 return br_handle_frame_finish(skb);
 277         } else if (dst->neighbour) {
 278                 /* the neighbour function below overwrites the complete
 279                  * MAC header, so we save the Ethernet source address and
 280                  * protocol number. */
 281                 skb_copy_from_linear_data_offset(skb, -(ETH_HLEN-ETH_ALEN), skb->nf_bridge->data, ETH_HLEN-ETH_ALEN);
 282                 /* tell br_dev_xmit to continue with forwarding */
 283                 nf_bridge->mask |= BRNF_BRIDGED_DNAT;
 284                 return dst->neighbour->output(skb);
 285         }
 286 free_skb:
 287         kfree_skb(skb);
 288         return 0;
 289 }
 290
 291 /* This requires some explaining. If DNAT has taken place,
 292  * we will need to fix up the destination Ethernet address.
 293  *
 294  * There are two cases to consider:
 295  * 1. The packet was DNAT'ed to a device in the same bridge
 296  *    port group as it was received on. We can still bridge
 297  *    the packet.
 298  * 2. The packet was DNAT'ed to a different device, either
 299  *    a non-bridged device or another bridge port group.
 300  *    The packet will need to be routed.
 301  *
 302  * The correct way of distinguishing between these two cases is to
 303  * call ip_route_input() and to look at skb->dst->dev, which is
 304  * changed to the destination device if ip_route_input() succeeds.
 305  *
 306  * Let's first consider the case that ip_route_input() succeeds:
 307  *
 308  * If the output device equals the logical bridge device the packet
 309  * came in on, we can consider this bridging. The corresponding MAC
 310  * address will be obtained in br_nf_pre_routing_finish_bridge.
 311  * Otherwise, the packet is considered to be routed and we just
 312  * change the destination MAC address so that the packet will
 313  * later be passed up to the IP stack to be routed. For a redirected
 314  * packet, ip_route_input() will give back the localhost as output device,
 315  * which differs from the bridge device.
 316  *
 317  * Let's now consider the case that ip_route_input() fails:
 318  *
 319  * This can be because the destination address is martian, in which case
 320  * the packet will be dropped.
 321  * If IP forwarding is disabled, ip_route_input() will fail, while
 322  * ip_route_output_key() can return success. The source
 323  * address for ip_route_output_key() is set to zero, so ip_route_output_key()
 324  * thinks we're handling a locally generated packet and won't care
 325  * if IP forwarding is enabled. If the output device equals the logical bridge
 326  * device, we proceed as if ip_route_input() succeeded. If it differs from the
 327  * logical bridge port or if ip_route_output_key() fails we drop the packet.
 328  */
 329 static int br_nf_pre_routing_finish(struct sk_buff *skb)
 330 {
 331         struct net_device *dev = skb->dev;
 332         struct iphdr *iph = ip_hdr(skb);
 333         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 334         struct rtable *rt;
 335         int err;
 336
 337         if (nf_bridge->mask & BRNF_PKT_TYPE) {
 338                 skb->pkt_type = PACKET_OTHERHOST;
 339                 nf_bridge->mask ^= BRNF_PKT_TYPE;
 340         }
 341         nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
 342         if (dnat_took_place(skb)) {
 343                 if ((err = ip_route_input(skb, iph->daddr, iph->saddr, iph->tos, dev))) {
 344                         struct flowi fl = {
 345                                 .nl_u = {
 346                                         .ip4_u = {
 347                                                  .daddr = iph->daddr,
 348                                                  .saddr = 0,
 349                                                  .tos = RT_TOS(iph->tos) },
 350                                 },
 351                                 .proto = 0,
 352                         };
 353                         struct in_device *in_dev = __in_dev_get_rcu(dev);
 354
 355                         /* If err equals -EHOSTUNREACH the error is due to a
 356                          * martian destination or due to the fact that
 357                          * forwarding is disabled. For most martian packets,
 358                          * ip_route_output_key() will fail. It won't fail for 2 types of
 359                          * martian destinations: loopback destinations and destination
 360                          * 0.0.0.0. In both cases the packet will be dropped because the
 361                          * destination is the loopback device and not the bridge. */
 362                         if (err != -EHOSTUNREACH || !in_dev || IN_DEV_FORWARD(in_dev))
 363                                 goto free_skb;
 364
 365                         if (!ip_route_output_key(dev_net(dev), &rt, &fl)) {
 366                                 /* - Bridged-and-DNAT'ed traffic doesn't
 367                                  *   require ip_forwarding. */
 368                                 if (((struct dst_entry *)rt)->dev == dev) {
 369                                         skb_dst_set(skb, (struct dst_entry *)rt);
 370                                         goto bridged_dnat;
 371                                 }
 372                                 dst_release((struct dst_entry *)rt);
 373                         }
 374 free_skb:
 375                         kfree_skb(skb);
 376                         return 0;
 377                 } else {
 378                         if (skb_dst(skb)->dev == dev) {
 379 bridged_dnat:
 380                                 skb->dev = nf_bridge->physindev;
 381                                 nf_bridge_update_protocol(skb);
 382                                 nf_bridge_push_encap_header(skb);
 383                                 NF_HOOK_THRESH(NFPROTO_BRIDGE,
 384                                                NF_BR_PRE_ROUTING,
 385                                                skb, skb->dev, NULL,
 386                                                br_nf_pre_routing_finish_bridge,
 387                                                1);
 388                                 return 0;
 389                         }
 390                         memcpy(eth_hdr(skb)->h_dest, dev->dev_addr, ETH_ALEN);
 391                         skb->pkt_type = PACKET_HOST;
 392                 }
 393         } else {
 394                 rt = bridge_parent_rtable(nf_bridge->physindev);
 395                 if (!rt) {
 396                         kfree_skb(skb);
 397                         return 0;
 398                 }
 399                 dst_hold(&rt->u.dst);
 400                 skb_dst_set(skb, &rt->u.dst);
 401         }
 402
 403         skb->dev = nf_bridge->physindev;
 404         nf_bridge_update_protocol(skb);
 405         nf_bridge_push_encap_header(skb);
 406         NF_HOOK_THRESH(NFPROTO_BRIDGE, NF_BR_PRE_ROUTING, skb, skb->dev, NULL,
 407                        br_handle_frame_finish, 1);
 408
 409         return 0;
 410 }
 411
 412 /* Some common code for IPv4/IPv6 */
 413 static struct net_device *setup_pre_routing(struct sk_buff *skb)
 414 {
 415         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 416
 417         if (skb->pkt_type == PACKET_OTHERHOST) {
 418                 skb->pkt_type = PACKET_HOST;
 419                 nf_bridge->mask |= BRNF_PKT_TYPE;
 420         }
 421
 422         nf_bridge->mask |= BRNF_NF_BRIDGE_PREROUTING;
 423         nf_bridge->physindev = skb->dev;
 424         skb->dev = bridge_parent(skb->dev);
 425         if (skb->protocol == htons(ETH_P_8021Q))
 426                 nf_bridge->mask |= BRNF_8021Q;
 427         else if (skb->protocol == htons(ETH_P_PPP_SES))
 428                 nf_bridge->mask |= BRNF_PPPoE;
 429
 430         return skb->dev;
 431 }
 432
 433 /* We only check the length. A bridge shouldn't do any hop-by-hop stuff anyway */
 434 static int check_hbh_len(struct sk_buff *skb)
 435 {
 436         unsigned char *raw = (u8 *)(ipv6_hdr(skb) + 1);
 437         u32 pkt_len;
 438         const unsigned char *nh = skb_network_header(skb);
 439         int off = raw - nh;
 440         int len = (raw[1] + 1) << 3;
 441
 442         if ((raw + len) - skb->data > skb_headlen(skb))
 443                 goto bad;
 444
 445         off += 2;
 446         len -= 2;
 447
 448         while (len > 0) {
 449                 int optlen = nh[off + 1] + 2;
 450
 451                 switch (nh[off]) {
 452                 case IPV6_TLV_PAD0:
 453                         optlen = 1;
 454                         break;
 455
 456                 case IPV6_TLV_PADN:
 457                         break;
 458
 459                 case IPV6_TLV_JUMBO:
 460                         if (nh[off + 1] != 4 || (off & 3) != 2)
 461                                 goto bad;
 462                         pkt_len = ntohl(*(__be32 *) (nh + off + 2));
 463                         if (pkt_len <= IPV6_MAXPLEN ||
 464                             ipv6_hdr(skb)->payload_len)
 465                                 goto bad;
 466                         if (pkt_len > skb->len - sizeof(struct ipv6hdr))
 467                                 goto bad;
 468                         if (pskb_trim_rcsum(skb,
 469                                             pkt_len + sizeof(struct ipv6hdr)))
 470                                 goto bad;
 471                         nh = skb_network_header(skb);
 472                         break;
 473                 default:
 474                         if (optlen > len)
 475                                 goto bad;
 476                         break;
 477                 }
 478                 off += optlen;
 479                 len -= optlen;
 480         }
 481         if (len == 0)
 482                 return 0;
 483 bad:
 484         return -1;
 485
 486 }
 487
 488 /* Replicate the checks that IPv6 does on packet reception and pass the packet
 489  * to ip6tables, which doesn't support NAT, so things are fairly simple. */
 490 static unsigned int br_nf_pre_routing_ipv6(unsigned int hook,
 491                                            struct sk_buff *skb,
 492                                            const struct net_device *in,
 493                                            const struct net_device *out,
 494                                            int (*okfn)(struct sk_buff *))
 495 {
 496         struct ipv6hdr *hdr;
 497         u32 pkt_len;
 498
 499         if (skb->len < sizeof(struct ipv6hdr))
 500                 goto inhdr_error;
 501
 502         if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
 503                 goto inhdr_error;
 504
 505         hdr = ipv6_hdr(skb);
 506
 507         if (hdr->version != 6)
 508                 goto inhdr_error;
 509
 510         pkt_len = ntohs(hdr->payload_len);
 511
 512         if (pkt_len || hdr->nexthdr != NEXTHDR_HOP) {
 513                 if (pkt_len + sizeof(struct ipv6hdr) > skb->len)
 514                         goto inhdr_error;
 515                 if (pskb_trim_rcsum(skb, pkt_len + sizeof(struct ipv6hdr)))
 516                         goto inhdr_error;
 517         }
 518         if (hdr->nexthdr == NEXTHDR_HOP && check_hbh_len(skb))
 519                 goto inhdr_error;
 520
 521         nf_bridge_put(skb->nf_bridge);
 522         if (!nf_bridge_alloc(skb))
 523                 return NF_DROP;
 524         if (!setup_pre_routing(skb))
 525                 return NF_DROP;
 526
 527         skb->protocol = htons(ETH_P_IPV6);
 528         NF_HOOK(NFPROTO_IPV6, NF_INET_PRE_ROUTING, skb, skb->dev, NULL,
 529                 br_nf_pre_routing_finish_ipv6);
 530
 531         return NF_STOLEN;
 532
 533 inhdr_error:
 534         return NF_DROP;
 535 }
 536
 537 /* Direct IPv6 traffic to br_nf_pre_routing_ipv6.
 538  * Replicate the checks that IPv4 does on packet reception.
 539  * Set skb->dev to the bridge device (i.e. parent of the
 540  * receiving device) to make netfilter happy, the REDIRECT
 541  * target in particular.  Save the original destination IP
 542  * address to be able to detect DNAT afterwards. */
 543 static unsigned int br_nf_pre_routing(unsigned int hook, struct sk_buff *skb,
 544                                       const struct net_device *in,
 545                                       const struct net_device *out,
 546                                       int (*okfn)(struct sk_buff *))
 547 {
 548         struct iphdr *iph;
 549         __u32 len = nf_bridge_encap_header_len(skb);
 550
 551         if (unlikely(!pskb_may_pull(skb, len)))
 552                 goto out;
 553
 554         if (skb->protocol == htons(ETH_P_IPV6) || IS_VLAN_IPV6(skb) ||
 555             IS_PPPOE_IPV6(skb)) {
 556 #ifdef CONFIG_SYSCTL
 557                 if (!brnf_call_ip6tables)
 558                         return NF_ACCEPT;
 559 #endif
 560                 nf_bridge_pull_encap_header_rcsum(skb);
 561                 return br_nf_pre_routing_ipv6(hook, skb, in, out, okfn);
 562         }
 563 #ifdef CONFIG_SYSCTL
 564         if (!brnf_call_iptables)
 565                 return NF_ACCEPT;
 566 #endif
 567
 568         if (skb->protocol != htons(ETH_P_IP) && !IS_VLAN_IP(skb) &&
 569             !IS_PPPOE_IP(skb))
 570                 return NF_ACCEPT;
 571
 572         nf_bridge_pull_encap_header_rcsum(skb);
 573
 574         if (!pskb_may_pull(skb, sizeof(struct iphdr)))
 575                 goto inhdr_error;
 576
 577         iph = ip_hdr(skb);
 578         if (iph->ihl < 5 || iph->version != 4)
 579                 goto inhdr_error;
 580
 581         if (!pskb_may_pull(skb, 4 * iph->ihl))
 582                 goto inhdr_error;
 583
 584         iph = ip_hdr(skb);
 585         if (ip_fast_csum((__u8 *) iph, iph->ihl) != 0)
 586                 goto inhdr_error;
 587
 588         len = ntohs(iph->tot_len);
 589         if (skb->len < len || len < 4 * iph->ihl)
 590                 goto inhdr_error;
 591
 592         pskb_trim_rcsum(skb, len);
 593
 594         /* BUG: Should really parse the IP options here. */
 595         memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
 596
 597         nf_bridge_put(skb->nf_bridge);
 598         if (!nf_bridge_alloc(skb))
 599                 return NF_DROP;
 600         if (!setup_pre_routing(skb))
 601                 return NF_DROP;
 602         store_orig_dstaddr(skb);
 603         skb->protocol = htons(ETH_P_IP);
 604
 605         NF_HOOK(NFPROTO_IPV4, NF_INET_PRE_ROUTING, skb, skb->dev, NULL,
 606                 br_nf_pre_routing_finish);
 607
 608         return NF_STOLEN;
 609
 610 inhdr_error:
 611 //      IP_INC_STATS_BH(IpInHdrErrors);
 612 out:
 613         return NF_DROP;
 614 }
 615
 616
 617 /* PF_BRIDGE/LOCAL_IN ************************************************/
 618 /* The packet is locally destined, which requires a real
 619  * dst_entry, so detach the fake one.  On the way up, the
 620  * packet would pass through PRE_ROUTING again (which already
 621  * took place when the packet entered the bridge), but we
 622  * register an IPv4 PRE_ROUTING 'sabotage' hook that will
 623  * prevent this from happening. */
 624 static unsigned int br_nf_local_in(unsigned int hook, struct sk_buff *skb,
 625                                    const struct net_device *in,
 626                                    const struct net_device *out,
 627                                    int (*okfn)(struct sk_buff *))
 628 {
 629         struct rtable *rt = skb_rtable(skb);
 630
 631         if (rt && rt == bridge_parent_rtable(in))
 632                 skb_dst_drop(skb);
 633
 634         return NF_ACCEPT;
 635 }
 636
 637 /* PF_BRIDGE/FORWARD *************************************************/
 638 static int br_nf_forward_finish(struct sk_buff *skb)
 639 {
 640         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 641         struct net_device *in;
 642
 643         if (skb->protocol != htons(ETH_P_ARP) && !IS_VLAN_ARP(skb)) {
 644                 in = nf_bridge->physindev;
 645                 if (nf_bridge->mask & BRNF_PKT_TYPE) {
 646                         skb->pkt_type = PACKET_OTHERHOST;
 647                         nf_bridge->mask ^= BRNF_PKT_TYPE;
 648                 }
 649                 nf_bridge_update_protocol(skb);
 650         } else {
 651                 in = *((struct net_device **)(skb->cb));
 652         }
 653         nf_bridge_push_encap_header(skb);
 654
 655         NF_HOOK_THRESH(NFPROTO_BRIDGE, NF_BR_FORWARD, skb, in,
 656                        skb->dev, br_forward_finish, 1);
 657         return 0;
 658 }
 659
 660 /* This is the 'purely bridged' case.  For IP, we pass the packet to
 661  * netfilter with indev and outdev set to the bridge device,
 662  * but we are still able to filter on the 'real' indev/outdev
 663  * because of the physdev module. For ARP, indev and outdev are the
 664  * bridge ports. */
 665 static unsigned int br_nf_forward_ip(unsigned int hook, struct sk_buff *skb,
 666                                      const struct net_device *in,
 667                                      const struct net_device *out,
 668                                      int (*okfn)(struct sk_buff *))
 669 {
 670         struct nf_bridge_info *nf_bridge;
 671         struct net_device *parent;
 672         u_int8_t pf;
 673
 674         if (!skb->nf_bridge)
 675                 return NF_ACCEPT;
 676
 677         /* Need exclusive nf_bridge_info since we might have multiple
 678          * different physoutdevs. */
 679         if (!nf_bridge_unshare(skb))
 680                 return NF_DROP;
 681
 682         parent = bridge_parent(out);
 683         if (!parent)
 684                 return NF_DROP;
 685
 686         if (skb->protocol == htons(ETH_P_IP) || IS_VLAN_IP(skb) ||
 687             IS_PPPOE_IP(skb))
 688                 pf = PF_INET;
 689         else if (skb->protocol == htons(ETH_P_IPV6) || IS_VLAN_IPV6(skb) ||
 690                  IS_PPPOE_IPV6(skb))
 691                 pf = PF_INET6;
 692         else
 693                 return NF_ACCEPT;
 694
 695         nf_bridge_pull_encap_header(skb);
 696
 697         nf_bridge = skb->nf_bridge;
 698         if (skb->pkt_type == PACKET_OTHERHOST) {
 699                 skb->pkt_type = PACKET_HOST;
 700                 nf_bridge->mask |= BRNF_PKT_TYPE;
 701         }
 702
 703         /* The physdev module checks on this */
 704         nf_bridge->mask |= BRNF_BRIDGED;
 705         nf_bridge->physoutdev = skb->dev;
 706         if (pf == PF_INET)
 707                 skb->protocol = htons(ETH_P_IP);
 708         else
 709                 skb->protocol = htons(ETH_P_IPV6);
 710
 711         NF_HOOK(pf, NF_INET_FORWARD, skb, bridge_parent(in), parent,
 712                 br_nf_forward_finish);
 713
 714         return NF_STOLEN;
 715 }
 716
 717 static unsigned int br_nf_forward_arp(unsigned int hook, struct sk_buff *skb,
 718                                       const struct net_device *in,
 719                                       const struct net_device *out,
 720                                       int (*okfn)(struct sk_buff *))
 721 {
 722         struct net_device **d = (struct net_device **)(skb->cb);
 723
 724 #ifdef CONFIG_SYSCTL
 725         if (!brnf_call_arptables)
 726                 return NF_ACCEPT;
 727 #endif
 728
 729         if (skb->protocol != htons(ETH_P_ARP)) {
 730                 if (!IS_VLAN_ARP(skb))
 731                         return NF_ACCEPT;
 732                 nf_bridge_pull_encap_header(skb);
 733         }
 734
 735         if (arp_hdr(skb)->ar_pln != 4) {
 736                 if (IS_VLAN_ARP(skb))
 737                         nf_bridge_push_encap_header(skb);
 738                 return NF_ACCEPT;
 739         }
 740         *d = (struct net_device *)in;
 741         NF_HOOK(NFPROTO_ARP, NF_ARP_FORWARD, skb, (struct net_device *)in,
 742                 (struct net_device *)out, br_nf_forward_finish);
 743
 744         return NF_STOLEN;
 745 }
 746
 747 #if defined(CONFIG_NF_CONNTRACK_IPV4) || defined(CONFIG_NF_CONNTRACK_IPV4_MODULE)
 748 static int br_nf_dev_queue_xmit(struct sk_buff *skb)
 749 {
 750         if (skb->nfct != NULL && skb->protocol == htons(ETH_P_IP) &&
 751             skb->len + nf_bridge_mtu_reduction(skb) > skb->dev->mtu &&
 752             !skb_is_gso(skb))
 753                 return ip_fragment(skb, br_dev_queue_push_xmit);
 754         else
 755                 return br_dev_queue_push_xmit(skb);
 756 }
 757 #else
 758 static int br_nf_dev_queue_xmit(struct sk_buff *skb)
 759 {
 760         return br_dev_queue_push_xmit(skb);
 761 }
 762 #endif
 763
 764 /* PF_BRIDGE/POST_ROUTING ********************************************/
 765 static unsigned int br_nf_post_routing(unsigned int hook, struct sk_buff *skb,
 766                                        const struct net_device *in,
 767                                        const struct net_device *out,
 768                                        int (*okfn)(struct sk_buff *))
 769 {
 770         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 771         struct net_device *realoutdev = bridge_parent(skb->dev);
 772         u_int8_t pf;
 773
 774         if (!nf_bridge || !(nf_bridge->mask & BRNF_BRIDGED))
 775                 return NF_ACCEPT;
 776
 777         if (!realoutdev)
 778                 return NF_DROP;
 779
 780         if (skb->protocol == htons(ETH_P_IP) || IS_VLAN_IP(skb) ||
 781             IS_PPPOE_IP(skb))
 782                 pf = PF_INET;
 783         else if (skb->protocol == htons(ETH_P_IPV6) || IS_VLAN_IPV6(skb) ||
 784                  IS_PPPOE_IPV6(skb))
 785                 pf = PF_INET6;
 786         else
 787                 return NF_ACCEPT;
 788
 789         /* We assume any code from br_dev_queue_push_xmit onwards doesn't care
 790          * about the value of skb->pkt_type. */
 791         if (skb->pkt_type == PACKET_OTHERHOST) {
 792                 skb->pkt_type = PACKET_HOST;
 793                 nf_bridge->mask |= BRNF_PKT_TYPE;
 794         }
 795
 796         nf_bridge_pull_encap_header(skb);
 797         nf_bridge_save_header(skb);
 798         if (pf == PF_INET)
 799                 skb->protocol = htons(ETH_P_IP);
 800         else
 801                 skb->protocol = htons(ETH_P_IPV6);
 802
 803         NF_HOOK(pf, NF_INET_POST_ROUTING, skb, NULL, realoutdev,
 804                 br_nf_dev_queue_xmit);
 805
 806         return NF_STOLEN;
 807 }
 808
 809 /* IP/SABOTAGE *****************************************************/
 810 /* Don't hand locally destined packets to PF_INET(6)/PRE_ROUTING
 811  * for the second time. */
 812 static unsigned int ip_sabotage_in(unsigned int hook, struct sk_buff *skb,
 813                                    const struct net_device *in,
 814                                    const struct net_device *out,
 815                                    int (*okfn)(struct sk_buff *))
 816 {
 817         if (skb->nf_bridge &&
 818             !(skb->nf_bridge->mask & BRNF_NF_BRIDGE_PREROUTING)) {
 819                 return NF_STOP;
 820         }
 821
 822         return NF_ACCEPT;
 823 }
 824
 825 /* For br_nf_post_routing, we need (prio = NF_BR_PRI_LAST), because
 826  * br_dev_queue_push_xmit is called afterwards */
 827 static struct nf_hook_ops br_nf_ops[] __read_mostly = {
 828         {
 829                 .hook = br_nf_pre_routing,
 830                 .owner = THIS_MODULE,
 831                 .pf = PF_BRIDGE,
 832                 .hooknum = NF_BR_PRE_ROUTING,
 833                 .priority = NF_BR_PRI_BRNF,
 834         },
 835         {
 836                 .hook = br_nf_local_in,
 837                 .owner = THIS_MODULE,
 838                 .pf = PF_BRIDGE,
 839                 .hooknum = NF_BR_LOCAL_IN,
 840                 .priority = NF_BR_PRI_BRNF,
 841         },
 842         {
 843                 .hook = br_nf_forward_ip,
 844                 .owner = THIS_MODULE,
 845                 .pf = PF_BRIDGE,
 846                 .hooknum = NF_BR_FORWARD,
 847                 .priority = NF_BR_PRI_BRNF - 1,
 848         },
 849         {
 850                 .hook = br_nf_forward_arp,
 851                 .owner = THIS_MODULE,
 852                 .pf = PF_BRIDGE,
 853                 .hooknum = NF_BR_FORWARD,
 854                 .priority = NF_BR_PRI_BRNF,
 855         },
 856         {
 857                 .hook = br_nf_post_routing,
 858                 .owner = THIS_MODULE,
 859                 .pf = PF_BRIDGE,
 860                 .hooknum = NF_BR_POST_ROUTING,
 861                 .priority = NF_BR_PRI_LAST,
 862         },
 863         {
 864                 .hook = ip_sabotage_in,
 865                 .owner = THIS_MODULE,
 866                 .pf = PF_INET,
 867                 .hooknum = NF_INET_PRE_ROUTING,
 868                 .priority = NF_IP_PRI_FIRST,
 869         },
 870         {
 871                 .hook = ip_sabotage_in,
 872                 .owner = THIS_MODULE,
 873                 .pf = PF_INET6,
 874                 .hooknum = NF_INET_PRE_ROUTING,
 875                 .priority = NF_IP6_PRI_FIRST,
 876         },
 877 };
 878
 879 #ifdef CONFIG_SYSCTL
 880 static
 881 int brnf_sysctl_call_tables(ctl_table * ctl, int write,
 882                             void __user * buffer, size_t * lenp, loff_t * ppos)
 883 {
 884         int ret;
 885
 886         ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
 887
 888         if (write && *(int *)(ctl->data))
 889                 *(int *)(ctl->data) = 1;
 890         return ret;
 891 }
 892
 893 static ctl_table brnf_table[] = {
 894         {
 895                 .procname       = "bridge-nf-call-arptables",
 896                 .data           = &brnf_call_arptables,
 897                 .maxlen         = sizeof(int),
 898                 .mode           = 0644,
 899                 .proc_handler   = brnf_sysctl_call_tables,
 900         },
 901         {
 902                 .procname       = "bridge-nf-call-iptables",
 903                 .data           = &brnf_call_iptables,
 904                 .maxlen         = sizeof(int),
 905                 .mode           = 0644,
 906                 .proc_handler   = brnf_sysctl_call_tables,
 907         },
 908         {
 909                 .procname       = "bridge-nf-call-ip6tables",
 910                 .data           = &brnf_call_ip6tables,
 911                 .maxlen         = sizeof(int),
 912                 .mode           = 0644,
 913                 .proc_handler   = brnf_sysctl_call_tables,
 914         },
 915         {
 916                 .procname       = "bridge-nf-filter-vlan-tagged",
 917                 .data           = &brnf_filter_vlan_tagged,
 918                 .maxlen         = sizeof(int),
 919                 .mode           = 0644,
 920                 .proc_handler   = brnf_sysctl_call_tables,
 921         },
 922         {
 923                 .procname       = "bridge-nf-filter-pppoe-tagged",
 924                 .data           = &brnf_filter_pppoe_tagged,
 925                 .maxlen         = sizeof(int),
 926                 .mode           = 0644,
 927                 .proc_handler   = brnf_sysctl_call_tables,
 928         },
 929         { }
 930 };
 931
 932 static struct ctl_path brnf_path[] = {
 933         { .procname = "net", },
 934         { .procname = "bridge", },
 935         { }
 936 };
 937 #endif
 938
 939 int __init br_netfilter_init(void)
 940 {
 941         int ret;
 942
 943         ret = nf_register_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
 944         if (ret < 0)
 945                 return ret;
 946 #ifdef CONFIG_SYSCTL
 947         brnf_sysctl_header = register_sysctl_paths(brnf_path, brnf_table);
 948         if (brnf_sysctl_header == NULL) {
 949                 printk(KERN_WARNING
 950                        "br_netfilter: can't register to sysctl.\n");
 951                 nf_unregister_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
 952                 return -ENOMEM;
 953         }
 954 #endif
 955         printk(KERN_NOTICE "Bridge firewalling registered\n");
 956         return 0;
 957 }
 958
 959 void br_netfilter_fini(void)
 960 {
 961         nf_unregister_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
 962 #ifdef CONFIG_SYSCTL
 963         unregister_sysctl_table(brnf_sysctl_header);
 964 #endif
 965 }