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_call_iptables 1
  59 #define brnf_call_ip6tables 1
  60 #define brnf_call_arptables 1
  61 #define brnf_filter_vlan_tagged 0
  62 #define brnf_filter_pppoe_tagged 0
  63 #endif
  64
  65 static inline __be16 vlan_proto(const struct sk_buff *skb)
  66 {
  67         return vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
  68 }
  69
  70 #define IS_VLAN_IP(skb) \
  71         (skb->protocol == htons(ETH_P_8021Q) && \
  72          vlan_proto(skb) == htons(ETH_P_IP) &&  \
  73          brnf_filter_vlan_tagged)
  74
  75 #define IS_VLAN_IPV6(skb) \
  76         (skb->protocol == htons(ETH_P_8021Q) && \
  77          vlan_proto(skb) == htons(ETH_P_IPV6) &&\
  78          brnf_filter_vlan_tagged)
  79
  80 #define IS_VLAN_ARP(skb) \
  81         (skb->protocol == htons(ETH_P_8021Q) && \
  82          vlan_proto(skb) == htons(ETH_P_ARP) && \
  83          brnf_filter_vlan_tagged)
  84
  85 static inline __be16 pppoe_proto(const struct sk_buff *skb)
  86 {
  87         return *((__be16 *)(skb_mac_header(skb) + ETH_HLEN +
  88                             sizeof(struct pppoe_hdr)));
  89 }
  90
  91 #define IS_PPPOE_IP(skb) \
  92         (skb->protocol == htons(ETH_P_PPP_SES) && \
  93          pppoe_proto(skb) == htons(PPP_IP) && \
  94          brnf_filter_pppoe_tagged)
  95
  96 #define IS_PPPOE_IPV6(skb) \
  97         (skb->protocol == htons(ETH_P_PPP_SES) && \
  98          pppoe_proto(skb) == htons(PPP_IPV6) && \
  99          brnf_filter_pppoe_tagged)
 100
 101 static void fake_update_pmtu(struct dst_entry *dst, u32 mtu)
 102 {
 103 }
 104
 105 static struct dst_ops fake_dst_ops = {
 106         .family =               AF_INET,
 107         .protocol =             cpu_to_be16(ETH_P_IP),
 108         .update_pmtu =          fake_update_pmtu,
 109         .entries =              ATOMIC_INIT(0),
 110 };
 111
 112 /*
 113  * Initialize bogus route table used to keep netfilter happy.
 114  * Currently, we fill in the PMTU entry because netfilter
 115  * refragmentation needs it, and the rt_flags entry because
 116  * ipt_REJECT needs it.  Future netfilter modules might
 117  * require us to fill additional fields.
 118  */
 119 void br_netfilter_rtable_init(struct net_bridge *br)
 120 {
 121         struct rtable *rt = &br->fake_rtable;
 122
 123         atomic_set(&rt->dst.__refcnt, 1);
 124         rt->dst.dev = br->dev;
 125         rt->dst.path = &rt->dst;
 126         rt->dst.metrics[RTAX_MTU - 1] = 1500;
 127         rt->dst.flags   = DST_NOXFRM;
 128         rt->dst.ops = &fake_dst_ops;
 129 }
 130
 131 static inline struct rtable *bridge_parent_rtable(const struct net_device *dev)
 132 {
 133         struct net_bridge_port *port = rcu_dereference(dev->br_port);
 134
 135         return port ? &port->br->fake_rtable : NULL;
 136 }
 137
 138 static inline struct net_device *bridge_parent(const struct net_device *dev)
 139 {
 140         struct net_bridge_port *port = rcu_dereference(dev->br_port);
 141
 142         return port ? port->br->dev : NULL;
 143 }
 144
 145 static inline struct nf_bridge_info *nf_bridge_alloc(struct sk_buff *skb)
 146 {
 147         skb->nf_bridge = kzalloc(sizeof(struct nf_bridge_info), GFP_ATOMIC);
 148         if (likely(skb->nf_bridge))
 149                 atomic_set(&(skb->nf_bridge->use), 1);
 150
 151         return skb->nf_bridge;
 152 }
 153
 154 static inline struct nf_bridge_info *nf_bridge_unshare(struct sk_buff *skb)
 155 {
 156         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 157
 158         if (atomic_read(&nf_bridge->use) > 1) {
 159                 struct nf_bridge_info *tmp = nf_bridge_alloc(skb);
 160
 161                 if (tmp) {
 162                         memcpy(tmp, nf_bridge, sizeof(struct nf_bridge_info));
 163                         atomic_set(&tmp->use, 1);
 164                         nf_bridge_put(nf_bridge);
 165                 }
 166                 nf_bridge = tmp;
 167         }
 168         return nf_bridge;
 169 }
 170
 171 static inline void nf_bridge_push_encap_header(struct sk_buff *skb)
 172 {
 173         unsigned int len = nf_bridge_encap_header_len(skb);
 174
 175         skb_push(skb, len);
 176         skb->network_header -= len;
 177 }
 178
 179 static inline void nf_bridge_pull_encap_header(struct sk_buff *skb)
 180 {
 181         unsigned int len = nf_bridge_encap_header_len(skb);
 182
 183         skb_pull(skb, len);
 184         skb->network_header += len;
 185 }
 186
 187 static inline void nf_bridge_pull_encap_header_rcsum(struct sk_buff *skb)
 188 {
 189         unsigned int len = nf_bridge_encap_header_len(skb);
 190
 191         skb_pull_rcsum(skb, len);
 192         skb->network_header += len;
 193 }
 194
 195 static inline void nf_bridge_save_header(struct sk_buff *skb)
 196 {
 197         int header_size = ETH_HLEN + nf_bridge_encap_header_len(skb);
 198
 199         skb_copy_from_linear_data_offset(skb, -header_size,
 200                                          skb->nf_bridge->data, header_size);
 201 }
 202
 203 static inline void nf_bridge_update_protocol(struct sk_buff *skb)
 204 {
 205         if (skb->nf_bridge->mask & BRNF_8021Q)
 206                 skb->protocol = htons(ETH_P_8021Q);
 207         else if (skb->nf_bridge->mask & BRNF_PPPoE)
 208                 skb->protocol = htons(ETH_P_PPP_SES);
 209 }
 210
 211 /* Fill in the header for fragmented IP packets handled by
 212  * the IPv4 connection tracking code.
 213  */
 214 int nf_bridge_copy_header(struct sk_buff *skb)
 215 {
 216         int err;
 217         unsigned int header_size;
 218
 219         nf_bridge_update_protocol(skb);
 220         header_size = ETH_HLEN + nf_bridge_encap_header_len(skb);
 221         err = skb_cow_head(skb, header_size);
 222         if (err)
 223                 return err;
 224
 225         skb_copy_to_linear_data_offset(skb, -header_size,
 226                                        skb->nf_bridge->data, header_size);
 227         __skb_push(skb, nf_bridge_encap_header_len(skb));
 228         return 0;
 229 }
 230
 231 /* PF_BRIDGE/PRE_ROUTING *********************************************/
 232 /* Undo the changes made for ip6tables PREROUTING and continue the
 233  * bridge PRE_ROUTING hook. */
 234 static int br_nf_pre_routing_finish_ipv6(struct sk_buff *skb)
 235 {
 236         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 237         struct rtable *rt;
 238
 239         if (nf_bridge->mask & BRNF_PKT_TYPE) {
 240                 skb->pkt_type = PACKET_OTHERHOST;
 241                 nf_bridge->mask ^= BRNF_PKT_TYPE;
 242         }
 243         nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
 244
 245         rt = bridge_parent_rtable(nf_bridge->physindev);
 246         if (!rt) {
 247                 kfree_skb(skb);
 248                 return 0;
 249         }
 250         skb_dst_set_noref(skb, &rt->dst);
 251
 252         skb->dev = nf_bridge->physindev;
 253         nf_bridge_update_protocol(skb);
 254         nf_bridge_push_encap_header(skb);
 255         NF_HOOK_THRESH(NFPROTO_BRIDGE, NF_BR_PRE_ROUTING, skb, skb->dev, NULL,
 256                        br_handle_frame_finish, 1);
 257
 258         return 0;
 259 }
 260
 261 /* Obtain the correct destination MAC address, while preserving the original
 262  * source MAC address. If we already know this address, we just copy it. If we
 263  * don't, we use the neighbour framework to find out. In both cases, we make
 264  * sure that br_handle_frame_finish() is called afterwards.
 265  */
 266 static int br_nf_pre_routing_finish_bridge(struct sk_buff *skb)
 267 {
 268         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 269         struct dst_entry *dst;
 270
 271         skb->dev = bridge_parent(skb->dev);
 272         if (!skb->dev)
 273                 goto free_skb;
 274         dst = skb_dst(skb);
 275         if (dst->hh) {
 276                 neigh_hh_bridge(dst->hh, skb);
 277                 skb->dev = nf_bridge->physindev;
 278                 return br_handle_frame_finish(skb);
 279         } else if (dst->neighbour) {
 280                 /* the neighbour function below overwrites the complete
 281                  * MAC header, so we save the Ethernet source address and
 282                  * protocol number. */
 283                 skb_copy_from_linear_data_offset(skb, -(ETH_HLEN-ETH_ALEN), skb->nf_bridge->data, ETH_HLEN-ETH_ALEN);
 284                 /* tell br_dev_xmit to continue with forwarding */
 285                 nf_bridge->mask |= BRNF_BRIDGED_DNAT;
 286                 return dst->neighbour->output(skb);
 287         }
 288 free_skb:
 289         kfree_skb(skb);
 290         return 0;
 291 }
 292
 293 /* This requires some explaining. If DNAT has taken place,
 294  * we will need to fix up the destination Ethernet address.
 295  *
 296  * There are two cases to consider:
 297  * 1. The packet was DNAT'ed to a device in the same bridge
 298  *    port group as it was received on. We can still bridge
 299  *    the packet.
 300  * 2. The packet was DNAT'ed to a different device, either
 301  *    a non-bridged device or another bridge port group.
 302  *    The packet will need to be routed.
 303  *
 304  * The correct way of distinguishing between these two cases is to
 305  * call ip_route_input() and to look at skb->dst->dev, which is
 306  * changed to the destination device if ip_route_input() succeeds.
 307  *
 308  * Let's first consider the case that ip_route_input() succeeds:
 309  *
 310  * If the output device equals the logical bridge device the packet
 311  * came in on, we can consider this bridging. The corresponding MAC
 312  * address will be obtained in br_nf_pre_routing_finish_bridge.
 313  * Otherwise, the packet is considered to be routed and we just
 314  * change the destination MAC address so that the packet will
 315  * later be passed up to the IP stack to be routed. For a redirected
 316  * packet, ip_route_input() will give back the localhost as output device,
 317  * which differs from the bridge device.
 318  *
 319  * Let's now consider the case that ip_route_input() fails:
 320  *
 321  * This can be because the destination address is martian, in which case
 322  * the packet will be dropped.
 323  * If IP forwarding is disabled, ip_route_input() will fail, while
 324  * ip_route_output_key() can return success. The source
 325  * address for ip_route_output_key() is set to zero, so ip_route_output_key()
 326  * thinks we're handling a locally generated packet and won't care
 327  * if IP forwarding is enabled. If the output device equals the logical bridge
 328  * device, we proceed as if ip_route_input() succeeded. If it differs from the
 329  * logical bridge port or if ip_route_output_key() fails we drop the packet.
 330  */
 331 static int br_nf_pre_routing_finish(struct sk_buff *skb)
 332 {
 333         struct net_device *dev = skb->dev;
 334         struct iphdr *iph = ip_hdr(skb);
 335         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 336         struct rtable *rt;
 337         int err;
 338
 339         if (nf_bridge->mask & BRNF_PKT_TYPE) {
 340                 skb->pkt_type = PACKET_OTHERHOST;
 341                 nf_bridge->mask ^= BRNF_PKT_TYPE;
 342         }
 343         nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
 344         if (dnat_took_place(skb)) {
 345                 if ((err = ip_route_input(skb, iph->daddr, iph->saddr, iph->tos, dev))) {
 346                         struct flowi fl = {
 347                                 .nl_u = {
 348                                         .ip4_u = {
 349                                                  .daddr = iph->daddr,
 350                                                  .saddr = 0,
 351                                                  .tos = RT_TOS(iph->tos) },
 352                                 },
 353                                 .proto = 0,
 354                         };
 355                         struct in_device *in_dev = __in_dev_get_rcu(dev);
 356
 357                         /* If err equals -EHOSTUNREACH the error is due to a
 358                          * martian destination or due to the fact that
 359                          * forwarding is disabled. For most martian packets,
 360                          * ip_route_output_key() will fail. It won't fail for 2 types of
 361                          * martian destinations: loopback destinations and destination
 362                          * 0.0.0.0. In both cases the packet will be dropped because the
 363                          * destination is the loopback device and not the bridge. */
 364                         if (err != -EHOSTUNREACH || !in_dev || IN_DEV_FORWARD(in_dev))
 365                                 goto free_skb;
 366
 367                         if (!ip_route_output_key(dev_net(dev), &rt, &fl)) {
 368                                 /* - Bridged-and-DNAT'ed traffic doesn't
 369                                  *   require ip_forwarding. */
 370                                 if (((struct dst_entry *)rt)->dev == dev) {
 371                                         skb_dst_set(skb, (struct dst_entry *)rt);
 372                                         goto bridged_dnat;
 373                                 }
 374                                 dst_release((struct dst_entry *)rt);
 375                         }
 376 free_skb:
 377                         kfree_skb(skb);
 378                         return 0;
 379                 } else {
 380                         if (skb_dst(skb)->dev == dev) {
 381 bridged_dnat:
 382                                 skb->dev = nf_bridge->physindev;
 383                                 nf_bridge_update_protocol(skb);
 384                                 nf_bridge_push_encap_header(skb);
 385                                 NF_HOOK_THRESH(NFPROTO_BRIDGE,
 386                                                NF_BR_PRE_ROUTING,
 387                                                skb, skb->dev, NULL,
 388                                                br_nf_pre_routing_finish_bridge,
 389                                                1);
 390                                 return 0;
 391                         }
 392                         memcpy(eth_hdr(skb)->h_dest, dev->dev_addr, ETH_ALEN);
 393                         skb->pkt_type = PACKET_HOST;
 394                 }
 395         } else {
 396                 rt = bridge_parent_rtable(nf_bridge->physindev);
 397                 if (!rt) {
 398                         kfree_skb(skb);
 399                         return 0;
 400                 }
 401                 skb_dst_set_noref(skb, &rt->dst);
 402         }
 403
 404         skb->dev = nf_bridge->physindev;
 405         nf_bridge_update_protocol(skb);
 406         nf_bridge_push_encap_header(skb);
 407         NF_HOOK_THRESH(NFPROTO_BRIDGE, NF_BR_PRE_ROUTING, skb, skb->dev, NULL,
 408                        br_handle_frame_finish, 1);
 409
 410         return 0;
 411 }
 412
 413 /* Some common code for IPv4/IPv6 */
 414 static struct net_device *setup_pre_routing(struct sk_buff *skb)
 415 {
 416         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 417
 418         if (skb->pkt_type == PACKET_OTHERHOST) {
 419                 skb->pkt_type = PACKET_HOST;
 420                 nf_bridge->mask |= BRNF_PKT_TYPE;
 421         }
 422
 423         nf_bridge->mask |= BRNF_NF_BRIDGE_PREROUTING;
 424         nf_bridge->physindev = skb->dev;
 425         skb->dev = bridge_parent(skb->dev);
 426         if (skb->protocol == htons(ETH_P_8021Q))
 427                 nf_bridge->mask |= BRNF_8021Q;
 428         else if (skb->protocol == htons(ETH_P_PPP_SES))
 429                 nf_bridge->mask |= BRNF_PPPoE;
 430
 431         return skb->dev;
 432 }
 433
 434 /* We only check the length. A bridge shouldn't do any hop-by-hop stuff anyway */
 435 static int check_hbh_len(struct sk_buff *skb)
 436 {
 437         unsigned char *raw = (u8 *)(ipv6_hdr(skb) + 1);
 438         u32 pkt_len;
 439         const unsigned char *nh = skb_network_header(skb);
 440         int off = raw - nh;
 441         int len = (raw[1] + 1) << 3;
 442
 443         if ((raw + len) - skb->data > skb_headlen(skb))
 444                 goto bad;
 445
 446         off += 2;
 447         len -= 2;
 448
 449         while (len > 0) {
 450                 int optlen = nh[off + 1] + 2;
 451
 452                 switch (nh[off]) {
 453                 case IPV6_TLV_PAD0:
 454                         optlen = 1;
 455                         break;
 456
 457                 case IPV6_TLV_PADN:
 458                         break;
 459
 460                 case IPV6_TLV_JUMBO:
 461                         if (nh[off + 1] != 4 || (off & 3) != 2)
 462                                 goto bad;
 463                         pkt_len = ntohl(*(__be32 *) (nh + off + 2));
 464                         if (pkt_len <= IPV6_MAXPLEN ||
 465                             ipv6_hdr(skb)->payload_len)
 466                                 goto bad;
 467                         if (pkt_len > skb->len - sizeof(struct ipv6hdr))
 468                                 goto bad;
 469                         if (pskb_trim_rcsum(skb,
 470                                             pkt_len + sizeof(struct ipv6hdr)))
 471                                 goto bad;
 472                         nh = skb_network_header(skb);
 473                         break;
 474                 default:
 475                         if (optlen > len)
 476                                 goto bad;
 477                         break;
 478                 }
 479                 off += optlen;
 480                 len -= optlen;
 481         }
 482         if (len == 0)
 483                 return 0;
 484 bad:
 485         return -1;
 486
 487 }
 488
 489 /* Replicate the checks that IPv6 does on packet reception and pass the packet
 490  * to ip6tables, which doesn't support NAT, so things are fairly simple. */
 491 static unsigned int br_nf_pre_routing_ipv6(unsigned int hook,
 492                                            struct sk_buff *skb,
 493                                            const struct net_device *in,
 494                                            const struct net_device *out,
 495                                            int (*okfn)(struct sk_buff *))
 496 {
 497         struct ipv6hdr *hdr;
 498         u32 pkt_len;
 499
 500         if (skb->len < sizeof(struct ipv6hdr))
 501                 goto inhdr_error;
 502
 503         if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
 504                 goto inhdr_error;
 505
 506         hdr = ipv6_hdr(skb);
 507
 508         if (hdr->version != 6)
 509                 goto inhdr_error;
 510
 511         pkt_len = ntohs(hdr->payload_len);
 512
 513         if (pkt_len || hdr->nexthdr != NEXTHDR_HOP) {
 514                 if (pkt_len + sizeof(struct ipv6hdr) > skb->len)
 515                         goto inhdr_error;
 516                 if (pskb_trim_rcsum(skb, pkt_len + sizeof(struct ipv6hdr)))
 517                         goto inhdr_error;
 518         }
 519         if (hdr->nexthdr == NEXTHDR_HOP && check_hbh_len(skb))
 520                 goto inhdr_error;
 521
 522         nf_bridge_put(skb->nf_bridge);
 523         if (!nf_bridge_alloc(skb))
 524                 return NF_DROP;
 525         if (!setup_pre_routing(skb))
 526                 return NF_DROP;
 527
 528         skb->protocol = htons(ETH_P_IPV6);
 529         NF_HOOK(NFPROTO_IPV6, NF_INET_PRE_ROUTING, skb, skb->dev, NULL,
 530                 br_nf_pre_routing_finish_ipv6);
 531
 532         return NF_STOLEN;
 533
 534 inhdr_error:
 535         return NF_DROP;
 536 }
 537
 538 /* Direct IPv6 traffic to br_nf_pre_routing_ipv6.
 539  * Replicate the checks that IPv4 does on packet reception.
 540  * Set skb->dev to the bridge device (i.e. parent of the
 541  * receiving device) to make netfilter happy, the REDIRECT
 542  * target in particular.  Save the original destination IP
 543  * address to be able to detect DNAT afterwards. */
 544 static unsigned int br_nf_pre_routing(unsigned int hook, struct sk_buff *skb,
 545                                       const struct net_device *in,
 546                                       const struct net_device *out,
 547                                       int (*okfn)(struct sk_buff *))
 548 {
 549         struct net_bridge_port *p;
 550         struct net_bridge *br;
 551         struct iphdr *iph;
 552         __u32 len = nf_bridge_encap_header_len(skb);
 553
 554         if (unlikely(!pskb_may_pull(skb, len)))
 555                 goto out;
 556
 557         p = rcu_dereference(in->br_port);
 558         if (p == NULL)
 559                 goto out;
 560         br = p->br;
 561
 562         if (skb->protocol == htons(ETH_P_IPV6) || IS_VLAN_IPV6(skb) ||
 563             IS_PPPOE_IPV6(skb)) {
 564                 if (!brnf_call_ip6tables && !br->nf_call_ip6tables)
 565                         return NF_ACCEPT;
 566
 567                 nf_bridge_pull_encap_header_rcsum(skb);
 568                 return br_nf_pre_routing_ipv6(hook, skb, in, out, okfn);
 569         }
 570
 571         if (!brnf_call_iptables && !br->nf_call_iptables)
 572                 return NF_ACCEPT;
 573
 574         if (skb->protocol != htons(ETH_P_IP) && !IS_VLAN_IP(skb) &&
 575             !IS_PPPOE_IP(skb))
 576                 return NF_ACCEPT;
 577
 578         nf_bridge_pull_encap_header_rcsum(skb);
 579
 580         if (!pskb_may_pull(skb, sizeof(struct iphdr)))
 581                 goto inhdr_error;
 582
 583         iph = ip_hdr(skb);
 584         if (iph->ihl < 5 || iph->version != 4)
 585                 goto inhdr_error;
 586
 587         if (!pskb_may_pull(skb, 4 * iph->ihl))
 588                 goto inhdr_error;
 589
 590         iph = ip_hdr(skb);
 591         if (ip_fast_csum((__u8 *) iph, iph->ihl) != 0)
 592                 goto inhdr_error;
 593
 594         len = ntohs(iph->tot_len);
 595         if (skb->len < len || len < 4 * iph->ihl)
 596                 goto inhdr_error;
 597
 598         pskb_trim_rcsum(skb, len);
 599
 600         nf_bridge_put(skb->nf_bridge);
 601         if (!nf_bridge_alloc(skb))
 602                 return NF_DROP;
 603         if (!setup_pre_routing(skb))
 604                 return NF_DROP;
 605         store_orig_dstaddr(skb);
 606         skb->protocol = htons(ETH_P_IP);
 607
 608         NF_HOOK(NFPROTO_IPV4, 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         struct rtable *rt = skb_rtable(skb);
 633
 634         if (rt && rt == bridge_parent_rtable(in))
 635                 skb_dst_drop(skb);
 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                 nf_bridge_update_protocol(skb);
 653         } else {
 654                 in = *((struct net_device **)(skb->cb));
 655         }
 656         nf_bridge_push_encap_header(skb);
 657
 658         NF_HOOK_THRESH(NFPROTO_BRIDGE, NF_BR_FORWARD, skb, in,
 659                        skb->dev, br_forward_finish, 1);
 660         return 0;
 661 }
 662
 663 /* This is the 'purely bridged' case.  For IP, we pass the packet to
 664  * netfilter with indev and outdev set to the bridge device,
 665  * but we are still able to filter on the 'real' indev/outdev
 666  * because of the physdev module. For ARP, indev and outdev are the
 667  * bridge ports. */
 668 static unsigned int br_nf_forward_ip(unsigned int hook, struct sk_buff *skb,
 669                                      const struct net_device *in,
 670                                      const struct net_device *out,
 671                                      int (*okfn)(struct sk_buff *))
 672 {
 673         struct nf_bridge_info *nf_bridge;
 674         struct net_device *parent;
 675         u_int8_t pf;
 676
 677         if (!skb->nf_bridge)
 678                 return NF_ACCEPT;
 679
 680         /* Need exclusive nf_bridge_info since we might have multiple
 681          * different physoutdevs. */
 682         if (!nf_bridge_unshare(skb))
 683                 return NF_DROP;
 684
 685         parent = bridge_parent(out);
 686         if (!parent)
 687                 return NF_DROP;
 688
 689         if (skb->protocol == htons(ETH_P_IP) || IS_VLAN_IP(skb) ||
 690             IS_PPPOE_IP(skb))
 691                 pf = PF_INET;
 692         else if (skb->protocol == htons(ETH_P_IPV6) || IS_VLAN_IPV6(skb) ||
 693                  IS_PPPOE_IPV6(skb))
 694                 pf = PF_INET6;
 695         else
 696                 return NF_ACCEPT;
 697
 698         nf_bridge_pull_encap_header(skb);
 699
 700         nf_bridge = skb->nf_bridge;
 701         if (skb->pkt_type == PACKET_OTHERHOST) {
 702                 skb->pkt_type = PACKET_HOST;
 703                 nf_bridge->mask |= BRNF_PKT_TYPE;
 704         }
 705
 706         /* The physdev module checks on this */
 707         nf_bridge->mask |= BRNF_BRIDGED;
 708         nf_bridge->physoutdev = skb->dev;
 709         if (pf == PF_INET)
 710                 skb->protocol = htons(ETH_P_IP);
 711         else
 712                 skb->protocol = htons(ETH_P_IPV6);
 713
 714         NF_HOOK(pf, NF_INET_FORWARD, skb, bridge_parent(in), parent,
 715                 br_nf_forward_finish);
 716
 717         return NF_STOLEN;
 718 }
 719
 720 static unsigned int br_nf_forward_arp(unsigned int hook, struct sk_buff *skb,
 721                                       const struct net_device *in,
 722                                       const struct net_device *out,
 723                                       int (*okfn)(struct sk_buff *))
 724 {
 725         struct net_bridge_port *p;
 726         struct net_bridge *br;
 727         struct net_device **d = (struct net_device **)(skb->cb);
 728
 729         p = rcu_dereference(out->br_port);
 730         if (p == NULL)
 731                 return NF_ACCEPT;
 732         br = p->br;
 733
 734         if (!brnf_call_arptables && !br->nf_call_arptables)
 735                 return NF_ACCEPT;
 736
 737         if (skb->protocol != htons(ETH_P_ARP)) {
 738                 if (!IS_VLAN_ARP(skb))
 739                         return NF_ACCEPT;
 740                 nf_bridge_pull_encap_header(skb);
 741         }
 742
 743         if (arp_hdr(skb)->ar_pln != 4) {
 744                 if (IS_VLAN_ARP(skb))
 745                         nf_bridge_push_encap_header(skb);
 746                 return NF_ACCEPT;
 747         }
 748         *d = (struct net_device *)in;
 749         NF_HOOK(NFPROTO_ARP, NF_ARP_FORWARD, skb, (struct net_device *)in,
 750                 (struct net_device *)out, br_nf_forward_finish);
 751
 752         return NF_STOLEN;
 753 }
 754
 755 #if defined(CONFIG_NF_CONNTRACK_IPV4) || defined(CONFIG_NF_CONNTRACK_IPV4_MODULE)
 756 static int br_nf_dev_queue_xmit(struct sk_buff *skb)
 757 {
 758         if (skb->nfct != NULL && skb->protocol == htons(ETH_P_IP) &&
 759             skb->len + nf_bridge_mtu_reduction(skb) > skb->dev->mtu &&
 760             !skb_is_gso(skb))
 761                 return ip_fragment(skb, br_dev_queue_push_xmit);
 762         else
 763                 return br_dev_queue_push_xmit(skb);
 764 }
 765 #else
 766 static int br_nf_dev_queue_xmit(struct sk_buff *skb)
 767 {
 768         return br_dev_queue_push_xmit(skb);
 769 }
 770 #endif
 771
 772 /* PF_BRIDGE/POST_ROUTING ********************************************/
 773 static unsigned int br_nf_post_routing(unsigned int hook, struct sk_buff *skb,
 774                                        const struct net_device *in,
 775                                        const struct net_device *out,
 776                                        int (*okfn)(struct sk_buff *))
 777 {
 778         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 779         struct net_device *realoutdev = bridge_parent(skb->dev);
 780         u_int8_t pf;
 781
 782         if (!nf_bridge || !(nf_bridge->mask & BRNF_BRIDGED))
 783                 return NF_ACCEPT;
 784
 785         if (!realoutdev)
 786                 return NF_DROP;
 787
 788         if (skb->protocol == htons(ETH_P_IP) || IS_VLAN_IP(skb) ||
 789             IS_PPPOE_IP(skb))
 790                 pf = PF_INET;
 791         else if (skb->protocol == htons(ETH_P_IPV6) || IS_VLAN_IPV6(skb) ||
 792                  IS_PPPOE_IPV6(skb))
 793                 pf = PF_INET6;
 794         else
 795                 return NF_ACCEPT;
 796
 797         /* We assume any code from br_dev_queue_push_xmit onwards doesn't care
 798          * about the value of skb->pkt_type. */
 799         if (skb->pkt_type == PACKET_OTHERHOST) {
 800                 skb->pkt_type = PACKET_HOST;
 801                 nf_bridge->mask |= BRNF_PKT_TYPE;
 802         }
 803
 804         nf_bridge_pull_encap_header(skb);
 805         nf_bridge_save_header(skb);
 806         if (pf == PF_INET)
 807                 skb->protocol = htons(ETH_P_IP);
 808         else
 809                 skb->protocol = htons(ETH_P_IPV6);
 810
 811         NF_HOOK(pf, NF_INET_POST_ROUTING, skb, NULL, realoutdev,
 812                 br_nf_dev_queue_xmit);
 813
 814         return NF_STOLEN;
 815 }
 816
 817 /* IP/SABOTAGE *****************************************************/
 818 /* Don't hand locally destined packets to PF_INET(6)/PRE_ROUTING
 819  * for the second time. */
 820 static unsigned int ip_sabotage_in(unsigned int hook, struct sk_buff *skb,
 821                                    const struct net_device *in,
 822                                    const struct net_device *out,
 823                                    int (*okfn)(struct sk_buff *))
 824 {
 825         if (skb->nf_bridge &&
 826             !(skb->nf_bridge->mask & BRNF_NF_BRIDGE_PREROUTING)) {
 827                 return NF_STOP;
 828         }
 829
 830         return NF_ACCEPT;
 831 }
 832
 833 /* For br_nf_post_routing, we need (prio = NF_BR_PRI_LAST), because
 834  * br_dev_queue_push_xmit is called afterwards */
 835 static struct nf_hook_ops br_nf_ops[] __read_mostly = {
 836         {
 837                 .hook = br_nf_pre_routing,
 838                 .owner = THIS_MODULE,
 839                 .pf = PF_BRIDGE,
 840                 .hooknum = NF_BR_PRE_ROUTING,
 841                 .priority = NF_BR_PRI_BRNF,
 842         },
 843         {
 844                 .hook = br_nf_local_in,
 845                 .owner = THIS_MODULE,
 846                 .pf = PF_BRIDGE,
 847                 .hooknum = NF_BR_LOCAL_IN,
 848                 .priority = NF_BR_PRI_BRNF,
 849         },
 850         {
 851                 .hook = br_nf_forward_ip,
 852                 .owner = THIS_MODULE,
 853                 .pf = PF_BRIDGE,
 854                 .hooknum = NF_BR_FORWARD,
 855                 .priority = NF_BR_PRI_BRNF - 1,
 856         },
 857         {
 858                 .hook = br_nf_forward_arp,
 859                 .owner = THIS_MODULE,
 860                 .pf = PF_BRIDGE,
 861                 .hooknum = NF_BR_FORWARD,
 862                 .priority = NF_BR_PRI_BRNF,
 863         },
 864         {
 865                 .hook = br_nf_post_routing,
 866                 .owner = THIS_MODULE,
 867                 .pf = PF_BRIDGE,
 868                 .hooknum = NF_BR_POST_ROUTING,
 869                 .priority = NF_BR_PRI_LAST,
 870         },
 871         {
 872                 .hook = ip_sabotage_in,
 873                 .owner = THIS_MODULE,
 874                 .pf = PF_INET,
 875                 .hooknum = NF_INET_PRE_ROUTING,
 876                 .priority = NF_IP_PRI_FIRST,
 877         },
 878         {
 879                 .hook = ip_sabotage_in,
 880                 .owner = THIS_MODULE,
 881                 .pf = PF_INET6,
 882                 .hooknum = NF_INET_PRE_ROUTING,
 883                 .priority = NF_IP6_PRI_FIRST,
 884         },
 885 };
 886
 887 #ifdef CONFIG_SYSCTL
 888 static
 889 int brnf_sysctl_call_tables(ctl_table * ctl, int write,
 890                             void __user * buffer, size_t * lenp, loff_t * ppos)
 891 {
 892         int ret;
 893
 894         ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
 895
 896         if (write && *(int *)(ctl->data))
 897                 *(int *)(ctl->data) = 1;
 898         return ret;
 899 }
 900
 901 static ctl_table brnf_table[] = {
 902         {
 903                 .procname       = "bridge-nf-call-arptables",
 904                 .data           = &brnf_call_arptables,
 905                 .maxlen         = sizeof(int),
 906                 .mode           = 0644,
 907                 .proc_handler   = brnf_sysctl_call_tables,
 908         },
 909         {
 910                 .procname       = "bridge-nf-call-iptables",
 911                 .data           = &brnf_call_iptables,
 912                 .maxlen         = sizeof(int),
 913                 .mode           = 0644,
 914                 .proc_handler   = brnf_sysctl_call_tables,
 915         },
 916         {
 917                 .procname       = "bridge-nf-call-ip6tables",
 918                 .data           = &brnf_call_ip6tables,
 919                 .maxlen         = sizeof(int),
 920                 .mode           = 0644,
 921                 .proc_handler   = brnf_sysctl_call_tables,
 922         },
 923         {
 924                 .procname       = "bridge-nf-filter-vlan-tagged",
 925                 .data           = &brnf_filter_vlan_tagged,
 926                 .maxlen         = sizeof(int),
 927                 .mode           = 0644,
 928                 .proc_handler   = brnf_sysctl_call_tables,
 929         },
 930         {
 931                 .procname       = "bridge-nf-filter-pppoe-tagged",
 932                 .data           = &brnf_filter_pppoe_tagged,
 933                 .maxlen         = sizeof(int),
 934                 .mode           = 0644,
 935                 .proc_handler   = brnf_sysctl_call_tables,
 936         },
 937         { }
 938 };
 939
 940 static struct ctl_path brnf_path[] = {
 941         { .procname = "net", },
 942         { .procname = "bridge", },
 943         { }
 944 };
 945 #endif
 946
 947 int __init br_netfilter_init(void)
 948 {
 949         int ret;
 950
 951         ret = nf_register_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
 952         if (ret < 0)
 953                 return ret;
 954 #ifdef CONFIG_SYSCTL
 955         brnf_sysctl_header = register_sysctl_paths(brnf_path, brnf_table);
 956         if (brnf_sysctl_header == NULL) {
 957                 printk(KERN_WARNING
 958                        "br_netfilter: can't register to sysctl.\n");
 959                 nf_unregister_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
 960                 return -ENOMEM;
 961         }
 962 #endif
 963         printk(KERN_NOTICE "Bridge firewalling registered\n");
 964         return 0;
 965 }
 966
 967 void br_netfilter_fini(void)
 968 {
 969         nf_unregister_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
 970 #ifdef CONFIG_SYSCTL
 971         unregister_sysctl_table(brnf_sysctl_header);
 972 #endif
 973 }