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         if (vlan_tx_tag_present(skb))
  68                 return skb->protocol;
  69         else if (skb->protocol == htons(ETH_P_8021Q))
  70                 return vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
  71         else
  72                 return 0;
  73 }
  74
  75 #define IS_VLAN_IP(skb) \
  76         (vlan_proto(skb) == htons(ETH_P_IP) && \
  77          brnf_filter_vlan_tagged)
  78
  79 #define IS_VLAN_IPV6(skb) \
  80         (vlan_proto(skb) == htons(ETH_P_IPV6) && \
  81          brnf_filter_vlan_tagged)
  82
  83 #define IS_VLAN_ARP(skb) \
  84         (vlan_proto(skb) == htons(ETH_P_ARP) && \
  85          brnf_filter_vlan_tagged)
  86
  87 static inline __be16 pppoe_proto(const struct sk_buff *skb)
  88 {
  89         return *((__be16 *)(skb_mac_header(skb) + ETH_HLEN +
  90                             sizeof(struct pppoe_hdr)));
  91 }
  92
  93 #define IS_PPPOE_IP(skb) \
  94         (skb->protocol == htons(ETH_P_PPP_SES) && \
  95          pppoe_proto(skb) == htons(PPP_IP) && \
  96          brnf_filter_pppoe_tagged)
  97
  98 #define IS_PPPOE_IPV6(skb) \
  99         (skb->protocol == htons(ETH_P_PPP_SES) && \
 100          pppoe_proto(skb) == htons(PPP_IPV6) && \
 101          brnf_filter_pppoe_tagged)
 102
 103 static void fake_update_pmtu(struct dst_entry *dst, u32 mtu)
 104 {
 105 }
 106
 107 static struct dst_ops fake_dst_ops = {
 108         .family =               AF_INET,
 109         .protocol =             cpu_to_be16(ETH_P_IP),
 110         .update_pmtu =          fake_update_pmtu,
 111 };
 112
 113 /*
 114  * Initialize bogus route table used to keep netfilter happy.
 115  * Currently, we fill in the PMTU entry because netfilter
 116  * refragmentation needs it, and the rt_flags entry because
 117  * ipt_REJECT needs it.  Future netfilter modules might
 118  * require us to fill additional fields.
 119  */
 120 void br_netfilter_rtable_init(struct net_bridge *br)
 121 {
 122         struct rtable *rt = &br->fake_rtable;
 123
 124         atomic_set(&rt->dst.__refcnt, 1);
 125         rt->dst.dev = br->dev;
 126         rt->dst.path = &rt->dst;
 127         rt->dst.metrics[RTAX_MTU - 1] = 1500;
 128         rt->dst.flags   = DST_NOXFRM;
 129         rt->dst.ops = &fake_dst_ops;
 130 }
 131
 132 static inline struct rtable *bridge_parent_rtable(const struct net_device *dev)
 133 {
 134         struct net_bridge_port *port;
 135
 136         port = br_port_get_rcu(dev);
 137         return port ? &port->br->fake_rtable : NULL;
 138 }
 139
 140 static inline struct net_device *bridge_parent(const struct net_device *dev)
 141 {
 142         struct net_bridge_port *port;
 143
 144         port = br_port_get_rcu(dev);
 145         return port ? port->br->dev : NULL;
 146 }
 147
 148 static inline struct nf_bridge_info *nf_bridge_alloc(struct sk_buff *skb)
 149 {
 150         skb->nf_bridge = kzalloc(sizeof(struct nf_bridge_info), GFP_ATOMIC);
 151         if (likely(skb->nf_bridge))
 152                 atomic_set(&(skb->nf_bridge->use), 1);
 153
 154         return skb->nf_bridge;
 155 }
 156
 157 static inline struct nf_bridge_info *nf_bridge_unshare(struct sk_buff *skb)
 158 {
 159         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 160
 161         if (atomic_read(&nf_bridge->use) > 1) {
 162                 struct nf_bridge_info *tmp = nf_bridge_alloc(skb);
 163
 164                 if (tmp) {
 165                         memcpy(tmp, nf_bridge, sizeof(struct nf_bridge_info));
 166                         atomic_set(&tmp->use, 1);
 167                 }
 168                 nf_bridge_put(nf_bridge);
 169                 nf_bridge = tmp;
 170         }
 171         return nf_bridge;
 172 }
 173
 174 static inline void nf_bridge_push_encap_header(struct sk_buff *skb)
 175 {
 176         unsigned int len = nf_bridge_encap_header_len(skb);
 177
 178         skb_push(skb, len);
 179         skb->network_header -= len;
 180 }
 181
 182 static inline void nf_bridge_pull_encap_header(struct sk_buff *skb)
 183 {
 184         unsigned int len = nf_bridge_encap_header_len(skb);
 185
 186         skb_pull(skb, len);
 187         skb->network_header += len;
 188 }
 189
 190 static inline void nf_bridge_pull_encap_header_rcsum(struct sk_buff *skb)
 191 {
 192         unsigned int len = nf_bridge_encap_header_len(skb);
 193
 194         skb_pull_rcsum(skb, len);
 195         skb->network_header += len;
 196 }
 197
 198 static inline void nf_bridge_save_header(struct sk_buff *skb)
 199 {
 200         int header_size = ETH_HLEN + nf_bridge_encap_header_len(skb);
 201
 202         skb_copy_from_linear_data_offset(skb, -header_size,
 203                                          skb->nf_bridge->data, header_size);
 204 }
 205
 206 static inline void nf_bridge_update_protocol(struct sk_buff *skb)
 207 {
 208         if (skb->nf_bridge->mask & BRNF_8021Q)
 209                 skb->protocol = htons(ETH_P_8021Q);
 210         else if (skb->nf_bridge->mask & BRNF_PPPoE)
 211                 skb->protocol = htons(ETH_P_PPP_SES);
 212 }
 213
 214 /* When handing a packet over to the IP layer
 215  * check whether we have a skb that is in the
 216  * expected format
 217  */
 218
 219 static int br_parse_ip_options(struct sk_buff *skb)
 220 {
 221         struct ip_options *opt;
 222         struct iphdr *iph;
 223         struct net_device *dev = skb->dev;
 224         u32 len;
 225
 226         iph = ip_hdr(skb);
 227         opt = &(IPCB(skb)->opt);
 228
 229         /* Basic sanity checks */
 230         if (iph->ihl < 5 || iph->version != 4)
 231                 goto inhdr_error;
 232
 233         if (!pskb_may_pull(skb, iph->ihl*4))
 234                 goto inhdr_error;
 235
 236         iph = ip_hdr(skb);
 237         if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl)))
 238                 goto inhdr_error;
 239
 240         len = ntohs(iph->tot_len);
 241         if (skb->len < len) {
 242                 IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INTRUNCATEDPKTS);
 243                 goto drop;
 244         } else if (len < (iph->ihl*4))
 245                 goto inhdr_error;
 246
 247         if (pskb_trim_rcsum(skb, len)) {
 248                 IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INDISCARDS);
 249                 goto drop;
 250         }
 251
 252         /* Zero out the CB buffer if no options present */
 253         if (iph->ihl == 5) {
 254                 memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
 255                 return 0;
 256         }
 257
 258         opt->optlen = iph->ihl*4 - sizeof(struct iphdr);
 259         if (ip_options_compile(dev_net(dev), opt, skb))
 260                 goto inhdr_error;
 261
 262         /* Check correct handling of SRR option */
 263         if (unlikely(opt->srr)) {
 264                 struct in_device *in_dev = __in_dev_get_rcu(dev);
 265                 if (in_dev && !IN_DEV_SOURCE_ROUTE(in_dev))
 266                         goto drop;
 267
 268                 if (ip_options_rcv_srr(skb))
 269                         goto drop;
 270         }
 271
 272         return 0;
 273
 274 inhdr_error:
 275         IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INHDRERRORS);
 276 drop:
 277         return -1;
 278 }
 279
 280 /* Fill in the header for fragmented IP packets handled by
 281  * the IPv4 connection tracking code.
 282  */
 283 int nf_bridge_copy_header(struct sk_buff *skb)
 284 {
 285         int err;
 286         unsigned int header_size;
 287
 288         nf_bridge_update_protocol(skb);
 289         header_size = ETH_HLEN + nf_bridge_encap_header_len(skb);
 290         err = skb_cow_head(skb, header_size);
 291         if (err)
 292                 return err;
 293
 294         skb_copy_to_linear_data_offset(skb, -header_size,
 295                                        skb->nf_bridge->data, header_size);
 296         __skb_push(skb, nf_bridge_encap_header_len(skb));
 297         return 0;
 298 }
 299
 300 /* PF_BRIDGE/PRE_ROUTING *********************************************/
 301 /* Undo the changes made for ip6tables PREROUTING and continue the
 302  * bridge PRE_ROUTING hook. */
 303 static int br_nf_pre_routing_finish_ipv6(struct sk_buff *skb)
 304 {
 305         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 306         struct rtable *rt;
 307
 308         if (nf_bridge->mask & BRNF_PKT_TYPE) {
 309                 skb->pkt_type = PACKET_OTHERHOST;
 310                 nf_bridge->mask ^= BRNF_PKT_TYPE;
 311         }
 312         nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
 313
 314         rt = bridge_parent_rtable(nf_bridge->physindev);
 315         if (!rt) {
 316                 kfree_skb(skb);
 317                 return 0;
 318         }
 319         skb_dst_set_noref(skb, &rt->dst);
 320
 321         skb->dev = nf_bridge->physindev;
 322         nf_bridge_update_protocol(skb);
 323         nf_bridge_push_encap_header(skb);
 324         NF_HOOK_THRESH(NFPROTO_BRIDGE, NF_BR_PRE_ROUTING, skb, skb->dev, NULL,
 325                        br_handle_frame_finish, 1);
 326
 327         return 0;
 328 }
 329
 330 /* Obtain the correct destination MAC address, while preserving the original
 331  * source MAC address. If we already know this address, we just copy it. If we
 332  * don't, we use the neighbour framework to find out. In both cases, we make
 333  * sure that br_handle_frame_finish() is called afterwards.
 334  */
 335 static int br_nf_pre_routing_finish_bridge(struct sk_buff *skb)
 336 {
 337         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 338         struct dst_entry *dst;
 339
 340         skb->dev = bridge_parent(skb->dev);
 341         if (!skb->dev)
 342                 goto free_skb;
 343         dst = skb_dst(skb);
 344         if (dst->hh) {
 345                 neigh_hh_bridge(dst->hh, skb);
 346                 skb->dev = nf_bridge->physindev;
 347                 return br_handle_frame_finish(skb);
 348         } else if (dst->neighbour) {
 349                 /* the neighbour function below overwrites the complete
 350                  * MAC header, so we save the Ethernet source address and
 351                  * protocol number. */
 352                 skb_copy_from_linear_data_offset(skb, -(ETH_HLEN-ETH_ALEN), skb->nf_bridge->data, ETH_HLEN-ETH_ALEN);
 353                 /* tell br_dev_xmit to continue with forwarding */
 354                 nf_bridge->mask |= BRNF_BRIDGED_DNAT;
 355                 return dst->neighbour->output(skb);
 356         }
 357 free_skb:
 358         kfree_skb(skb);
 359         return 0;
 360 }
 361
 362 /* This requires some explaining. If DNAT has taken place,
 363  * we will need to fix up the destination Ethernet address.
 364  *
 365  * There are two cases to consider:
 366  * 1. The packet was DNAT'ed to a device in the same bridge
 367  *    port group as it was received on. We can still bridge
 368  *    the packet.
 369  * 2. The packet was DNAT'ed to a different device, either
 370  *    a non-bridged device or another bridge port group.
 371  *    The packet will need to be routed.
 372  *
 373  * The correct way of distinguishing between these two cases is to
 374  * call ip_route_input() and to look at skb->dst->dev, which is
 375  * changed to the destination device if ip_route_input() succeeds.
 376  *
 377  * Let's first consider the case that ip_route_input() succeeds:
 378  *
 379  * If the output device equals the logical bridge device the packet
 380  * came in on, we can consider this bridging. The corresponding MAC
 381  * address will be obtained in br_nf_pre_routing_finish_bridge.
 382  * Otherwise, the packet is considered to be routed and we just
 383  * change the destination MAC address so that the packet will
 384  * later be passed up to the IP stack to be routed. For a redirected
 385  * packet, ip_route_input() will give back the localhost as output device,
 386  * which differs from the bridge device.
 387  *
 388  * Let's now consider the case that ip_route_input() fails:
 389  *
 390  * This can be because the destination address is martian, in which case
 391  * the packet will be dropped.
 392  * If IP forwarding is disabled, ip_route_input() will fail, while
 393  * ip_route_output_key() can return success. The source
 394  * address for ip_route_output_key() is set to zero, so ip_route_output_key()
 395  * thinks we're handling a locally generated packet and won't care
 396  * if IP forwarding is enabled. If the output device equals the logical bridge
 397  * device, we proceed as if ip_route_input() succeeded. If it differs from the
 398  * logical bridge port or if ip_route_output_key() fails we drop the packet.
 399  */
 400 static int br_nf_pre_routing_finish(struct sk_buff *skb)
 401 {
 402         struct net_device *dev = skb->dev;
 403         struct iphdr *iph = ip_hdr(skb);
 404         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 405         struct rtable *rt;
 406         int err;
 407
 408         if (nf_bridge->mask & BRNF_PKT_TYPE) {
 409                 skb->pkt_type = PACKET_OTHERHOST;
 410                 nf_bridge->mask ^= BRNF_PKT_TYPE;
 411         }
 412         nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
 413         if (dnat_took_place(skb)) {
 414                 if ((err = ip_route_input(skb, iph->daddr, iph->saddr, iph->tos, dev))) {
 415                         struct flowi fl = {
 416                                 .nl_u = {
 417                                         .ip4_u = {
 418                                                  .daddr = iph->daddr,
 419                                                  .saddr = 0,
 420                                                  .tos = RT_TOS(iph->tos) },
 421                                 },
 422                                 .proto = 0,
 423                         };
 424                         struct in_device *in_dev = __in_dev_get_rcu(dev);
 425
 426                         /* If err equals -EHOSTUNREACH the error is due to a
 427                          * martian destination or due to the fact that
 428                          * forwarding is disabled. For most martian packets,
 429                          * ip_route_output_key() will fail. It won't fail for 2 types of
 430                          * martian destinations: loopback destinations and destination
 431                          * 0.0.0.0. In both cases the packet will be dropped because the
 432                          * destination is the loopback device and not the bridge. */
 433                         if (err != -EHOSTUNREACH || !in_dev || IN_DEV_FORWARD(in_dev))
 434                                 goto free_skb;
 435
 436                         if (!ip_route_output_key(dev_net(dev), &rt, &fl)) {
 437                                 /* - Bridged-and-DNAT'ed traffic doesn't
 438                                  *   require ip_forwarding. */
 439                                 if (((struct dst_entry *)rt)->dev == dev) {
 440                                         skb_dst_set(skb, (struct dst_entry *)rt);
 441                                         goto bridged_dnat;
 442                                 }
 443                                 dst_release((struct dst_entry *)rt);
 444                         }
 445 free_skb:
 446                         kfree_skb(skb);
 447                         return 0;
 448                 } else {
 449                         if (skb_dst(skb)->dev == dev) {
 450 bridged_dnat:
 451                                 skb->dev = nf_bridge->physindev;
 452                                 nf_bridge_update_protocol(skb);
 453                                 nf_bridge_push_encap_header(skb);
 454                                 NF_HOOK_THRESH(NFPROTO_BRIDGE,
 455                                                NF_BR_PRE_ROUTING,
 456                                                skb, skb->dev, NULL,
 457                                                br_nf_pre_routing_finish_bridge,
 458                                                1);
 459                                 return 0;
 460                         }
 461                         memcpy(eth_hdr(skb)->h_dest, dev->dev_addr, ETH_ALEN);
 462                         skb->pkt_type = PACKET_HOST;
 463                 }
 464         } else {
 465                 rt = bridge_parent_rtable(nf_bridge->physindev);
 466                 if (!rt) {
 467                         kfree_skb(skb);
 468                         return 0;
 469                 }
 470                 skb_dst_set_noref(skb, &rt->dst);
 471         }
 472
 473         skb->dev = nf_bridge->physindev;
 474         nf_bridge_update_protocol(skb);
 475         nf_bridge_push_encap_header(skb);
 476         NF_HOOK_THRESH(NFPROTO_BRIDGE, NF_BR_PRE_ROUTING, skb, skb->dev, NULL,
 477                        br_handle_frame_finish, 1);
 478
 479         return 0;
 480 }
 481
 482 /* Some common code for IPv4/IPv6 */
 483 static struct net_device *setup_pre_routing(struct sk_buff *skb)
 484 {
 485         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 486
 487         if (skb->pkt_type == PACKET_OTHERHOST) {
 488                 skb->pkt_type = PACKET_HOST;
 489                 nf_bridge->mask |= BRNF_PKT_TYPE;
 490         }
 491
 492         nf_bridge->mask |= BRNF_NF_BRIDGE_PREROUTING;
 493         nf_bridge->physindev = skb->dev;
 494         skb->dev = bridge_parent(skb->dev);
 495         if (skb->protocol == htons(ETH_P_8021Q))
 496                 nf_bridge->mask |= BRNF_8021Q;
 497         else if (skb->protocol == htons(ETH_P_PPP_SES))
 498                 nf_bridge->mask |= BRNF_PPPoE;
 499
 500         return skb->dev;
 501 }
 502
 503 /* We only check the length. A bridge shouldn't do any hop-by-hop stuff anyway */
 504 static int check_hbh_len(struct sk_buff *skb)
 505 {
 506         unsigned char *raw = (u8 *)(ipv6_hdr(skb) + 1);
 507         u32 pkt_len;
 508         const unsigned char *nh = skb_network_header(skb);
 509         int off = raw - nh;
 510         int len = (raw[1] + 1) << 3;
 511
 512         if ((raw + len) - skb->data > skb_headlen(skb))
 513                 goto bad;
 514
 515         off += 2;
 516         len -= 2;
 517
 518         while (len > 0) {
 519                 int optlen = nh[off + 1] + 2;
 520
 521                 switch (nh[off]) {
 522                 case IPV6_TLV_PAD0:
 523                         optlen = 1;
 524                         break;
 525
 526                 case IPV6_TLV_PADN:
 527                         break;
 528
 529                 case IPV6_TLV_JUMBO:
 530                         if (nh[off + 1] != 4 || (off & 3) != 2)
 531                                 goto bad;
 532                         pkt_len = ntohl(*(__be32 *) (nh + off + 2));
 533                         if (pkt_len <= IPV6_MAXPLEN ||
 534                             ipv6_hdr(skb)->payload_len)
 535                                 goto bad;
 536                         if (pkt_len > skb->len - sizeof(struct ipv6hdr))
 537                                 goto bad;
 538                         if (pskb_trim_rcsum(skb,
 539                                             pkt_len + sizeof(struct ipv6hdr)))
 540                                 goto bad;
 541                         nh = skb_network_header(skb);
 542                         break;
 543                 default:
 544                         if (optlen > len)
 545                                 goto bad;
 546                         break;
 547                 }
 548                 off += optlen;
 549                 len -= optlen;
 550         }
 551         if (len == 0)
 552                 return 0;
 553 bad:
 554         return -1;
 555
 556 }
 557
 558 /* Replicate the checks that IPv6 does on packet reception and pass the packet
 559  * to ip6tables, which doesn't support NAT, so things are fairly simple. */
 560 static unsigned int br_nf_pre_routing_ipv6(unsigned int hook,
 561                                            struct sk_buff *skb,
 562                                            const struct net_device *in,
 563                                            const struct net_device *out,
 564                                            int (*okfn)(struct sk_buff *))
 565 {
 566         struct ipv6hdr *hdr;
 567         u32 pkt_len;
 568
 569         if (skb->len < sizeof(struct ipv6hdr))
 570                 goto inhdr_error;
 571
 572         if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
 573                 goto inhdr_error;
 574
 575         hdr = ipv6_hdr(skb);
 576
 577         if (hdr->version != 6)
 578                 goto inhdr_error;
 579
 580         pkt_len = ntohs(hdr->payload_len);
 581
 582         if (pkt_len || hdr->nexthdr != NEXTHDR_HOP) {
 583                 if (pkt_len + sizeof(struct ipv6hdr) > skb->len)
 584                         goto inhdr_error;
 585                 if (pskb_trim_rcsum(skb, pkt_len + sizeof(struct ipv6hdr)))
 586                         goto inhdr_error;
 587         }
 588         if (hdr->nexthdr == NEXTHDR_HOP && check_hbh_len(skb))
 589                 goto inhdr_error;
 590
 591         nf_bridge_put(skb->nf_bridge);
 592         if (!nf_bridge_alloc(skb))
 593                 return NF_DROP;
 594         if (!setup_pre_routing(skb))
 595                 return NF_DROP;
 596
 597         skb->protocol = htons(ETH_P_IPV6);
 598         NF_HOOK(NFPROTO_IPV6, NF_INET_PRE_ROUTING, skb, skb->dev, NULL,
 599                 br_nf_pre_routing_finish_ipv6);
 600
 601         return NF_STOLEN;
 602
 603 inhdr_error:
 604         return NF_DROP;
 605 }
 606
 607 /* Direct IPv6 traffic to br_nf_pre_routing_ipv6.
 608  * Replicate the checks that IPv4 does on packet reception.
 609  * Set skb->dev to the bridge device (i.e. parent of the
 610  * receiving device) to make netfilter happy, the REDIRECT
 611  * target in particular.  Save the original destination IP
 612  * address to be able to detect DNAT afterwards. */
 613 static unsigned int br_nf_pre_routing(unsigned int hook, struct sk_buff *skb,
 614                                       const struct net_device *in,
 615                                       const struct net_device *out,
 616                                       int (*okfn)(struct sk_buff *))
 617 {
 618         struct net_bridge_port *p;
 619         struct net_bridge *br;
 620         __u32 len = nf_bridge_encap_header_len(skb);
 621
 622         if (unlikely(!pskb_may_pull(skb, len)))
 623                 goto out;
 624
 625         p = br_port_get_rcu(in);
 626         if (p == NULL)
 627                 goto out;
 628         br = p->br;
 629
 630         if (skb->protocol == htons(ETH_P_IPV6) || IS_VLAN_IPV6(skb) ||
 631             IS_PPPOE_IPV6(skb)) {
 632                 if (!brnf_call_ip6tables && !br->nf_call_ip6tables)
 633                         return NF_ACCEPT;
 634
 635                 nf_bridge_pull_encap_header_rcsum(skb);
 636                 return br_nf_pre_routing_ipv6(hook, skb, in, out, okfn);
 637         }
 638
 639         if (!brnf_call_iptables && !br->nf_call_iptables)
 640                 return NF_ACCEPT;
 641
 642         if (skb->protocol != htons(ETH_P_IP) && !IS_VLAN_IP(skb) &&
 643             !IS_PPPOE_IP(skb))
 644                 return NF_ACCEPT;
 645
 646         nf_bridge_pull_encap_header_rcsum(skb);
 647
 648         if (br_parse_ip_options(skb))
 649                 /* Drop invalid packet */
 650                 goto out;
 651
 652         nf_bridge_put(skb->nf_bridge);
 653         if (!nf_bridge_alloc(skb))
 654                 return NF_DROP;
 655         if (!setup_pre_routing(skb))
 656                 return NF_DROP;
 657         store_orig_dstaddr(skb);
 658         skb->protocol = htons(ETH_P_IP);
 659
 660         NF_HOOK(NFPROTO_IPV4, NF_INET_PRE_ROUTING, skb, skb->dev, NULL,
 661                 br_nf_pre_routing_finish);
 662
 663         return NF_STOLEN;
 664
 665 out:
 666         return NF_DROP;
 667 }
 668
 669
 670 /* PF_BRIDGE/LOCAL_IN ************************************************/
 671 /* The packet is locally destined, which requires a real
 672  * dst_entry, so detach the fake one.  On the way up, the
 673  * packet would pass through PRE_ROUTING again (which already
 674  * took place when the packet entered the bridge), but we
 675  * register an IPv4 PRE_ROUTING 'sabotage' hook that will
 676  * prevent this from happening. */
 677 static unsigned int br_nf_local_in(unsigned int hook, struct sk_buff *skb,
 678                                    const struct net_device *in,
 679                                    const struct net_device *out,
 680                                    int (*okfn)(struct sk_buff *))
 681 {
 682         struct rtable *rt = skb_rtable(skb);
 683
 684         if (rt && rt == bridge_parent_rtable(in))
 685                 skb_dst_drop(skb);
 686
 687         return NF_ACCEPT;
 688 }
 689
 690 /* PF_BRIDGE/FORWARD *************************************************/
 691 static int br_nf_forward_finish(struct sk_buff *skb)
 692 {
 693         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 694         struct net_device *in;
 695
 696         if (skb->protocol != htons(ETH_P_ARP) && !IS_VLAN_ARP(skb)) {
 697                 in = nf_bridge->physindev;
 698                 if (nf_bridge->mask & BRNF_PKT_TYPE) {
 699                         skb->pkt_type = PACKET_OTHERHOST;
 700                         nf_bridge->mask ^= BRNF_PKT_TYPE;
 701                 }
 702                 nf_bridge_update_protocol(skb);
 703         } else {
 704                 in = *((struct net_device **)(skb->cb));
 705         }
 706         nf_bridge_push_encap_header(skb);
 707
 708         NF_HOOK_THRESH(NFPROTO_BRIDGE, NF_BR_FORWARD, skb, in,
 709                        skb->dev, br_forward_finish, 1);
 710         return 0;
 711 }
 712
 713 /* This is the 'purely bridged' case.  For IP, we pass the packet to
 714  * netfilter with indev and outdev set to the bridge device,
 715  * but we are still able to filter on the 'real' indev/outdev
 716  * because of the physdev module. For ARP, indev and outdev are the
 717  * bridge ports. */
 718 static unsigned int br_nf_forward_ip(unsigned int hook, struct sk_buff *skb,
 719                                      const struct net_device *in,
 720                                      const struct net_device *out,
 721                                      int (*okfn)(struct sk_buff *))
 722 {
 723         struct nf_bridge_info *nf_bridge;
 724         struct net_device *parent;
 725         u_int8_t pf;
 726
 727         if (!skb->nf_bridge)
 728                 return NF_ACCEPT;
 729
 730         /* Need exclusive nf_bridge_info since we might have multiple
 731          * different physoutdevs. */
 732         if (!nf_bridge_unshare(skb))
 733                 return NF_DROP;
 734
 735         parent = bridge_parent(out);
 736         if (!parent)
 737                 return NF_DROP;
 738
 739         if (skb->protocol == htons(ETH_P_IP) || IS_VLAN_IP(skb) ||
 740             IS_PPPOE_IP(skb))
 741                 pf = PF_INET;
 742         else if (skb->protocol == htons(ETH_P_IPV6) || IS_VLAN_IPV6(skb) ||
 743                  IS_PPPOE_IPV6(skb))
 744                 pf = PF_INET6;
 745         else
 746                 return NF_ACCEPT;
 747
 748         nf_bridge_pull_encap_header(skb);
 749
 750         nf_bridge = skb->nf_bridge;
 751         if (skb->pkt_type == PACKET_OTHERHOST) {
 752                 skb->pkt_type = PACKET_HOST;
 753                 nf_bridge->mask |= BRNF_PKT_TYPE;
 754         }
 755
 756         /* The physdev module checks on this */
 757         nf_bridge->mask |= BRNF_BRIDGED;
 758         nf_bridge->physoutdev = skb->dev;
 759         if (pf == PF_INET)
 760                 skb->protocol = htons(ETH_P_IP);
 761         else
 762                 skb->protocol = htons(ETH_P_IPV6);
 763
 764         NF_HOOK(pf, NF_INET_FORWARD, skb, bridge_parent(in), parent,
 765                 br_nf_forward_finish);
 766
 767         return NF_STOLEN;
 768 }
 769
 770 static unsigned int br_nf_forward_arp(unsigned int hook, struct sk_buff *skb,
 771                                       const struct net_device *in,
 772                                       const struct net_device *out,
 773                                       int (*okfn)(struct sk_buff *))
 774 {
 775         struct net_bridge_port *p;
 776         struct net_bridge *br;
 777         struct net_device **d = (struct net_device **)(skb->cb);
 778
 779         p = br_port_get_rcu(out);
 780         if (p == NULL)
 781                 return NF_ACCEPT;
 782         br = p->br;
 783
 784         if (!brnf_call_arptables && !br->nf_call_arptables)
 785                 return NF_ACCEPT;
 786
 787         if (skb->protocol != htons(ETH_P_ARP)) {
 788                 if (!IS_VLAN_ARP(skb))
 789                         return NF_ACCEPT;
 790                 nf_bridge_pull_encap_header(skb);
 791         }
 792
 793         if (arp_hdr(skb)->ar_pln != 4) {
 794                 if (IS_VLAN_ARP(skb))
 795                         nf_bridge_push_encap_header(skb);
 796                 return NF_ACCEPT;
 797         }
 798         *d = (struct net_device *)in;
 799         NF_HOOK(NFPROTO_ARP, NF_ARP_FORWARD, skb, (struct net_device *)in,
 800                 (struct net_device *)out, br_nf_forward_finish);
 801
 802         return NF_STOLEN;
 803 }
 804
 805 #if defined(CONFIG_NF_CONNTRACK_IPV4) || defined(CONFIG_NF_CONNTRACK_IPV4_MODULE)
 806 static int br_nf_dev_queue_xmit(struct sk_buff *skb)
 807 {
 808         int ret;
 809
 810         if (skb->nfct != NULL && skb->protocol == htons(ETH_P_IP) &&
 811             skb->len + nf_bridge_mtu_reduction(skb) > skb->dev->mtu &&
 812             !skb_is_gso(skb)) {
 813                 if (br_parse_ip_options(skb))
 814                         /* Drop invalid packet */
 815                         return NF_DROP;
 816                 ret = ip_fragment(skb, br_dev_queue_push_xmit);
 817         } else
 818                 ret = br_dev_queue_push_xmit(skb);
 819
 820         return ret;
 821 }
 822 #else
 823 static int br_nf_dev_queue_xmit(struct sk_buff *skb)
 824 {
 825         return br_dev_queue_push_xmit(skb);
 826 }
 827 #endif
 828
 829 /* PF_BRIDGE/POST_ROUTING ********************************************/
 830 static unsigned int br_nf_post_routing(unsigned int hook, struct sk_buff *skb,
 831                                        const struct net_device *in,
 832                                        const struct net_device *out,
 833                                        int (*okfn)(struct sk_buff *))
 834 {
 835         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 836         struct net_device *realoutdev = bridge_parent(skb->dev);
 837         u_int8_t pf;
 838
 839         if (!nf_bridge || !(nf_bridge->mask & BRNF_BRIDGED))
 840                 return NF_ACCEPT;
 841
 842         if (!realoutdev)
 843                 return NF_DROP;
 844
 845         if (skb->protocol == htons(ETH_P_IP) || IS_VLAN_IP(skb) ||
 846             IS_PPPOE_IP(skb))
 847                 pf = PF_INET;
 848         else if (skb->protocol == htons(ETH_P_IPV6) || IS_VLAN_IPV6(skb) ||
 849                  IS_PPPOE_IPV6(skb))
 850                 pf = PF_INET6;
 851         else
 852                 return NF_ACCEPT;
 853
 854         /* We assume any code from br_dev_queue_push_xmit onwards doesn't care
 855          * about the value of skb->pkt_type. */
 856         if (skb->pkt_type == PACKET_OTHERHOST) {
 857                 skb->pkt_type = PACKET_HOST;
 858                 nf_bridge->mask |= BRNF_PKT_TYPE;
 859         }
 860
 861         nf_bridge_pull_encap_header(skb);
 862         nf_bridge_save_header(skb);
 863         if (pf == PF_INET)
 864                 skb->protocol = htons(ETH_P_IP);
 865         else
 866                 skb->protocol = htons(ETH_P_IPV6);
 867
 868         NF_HOOK(pf, NF_INET_POST_ROUTING, skb, NULL, realoutdev,
 869                 br_nf_dev_queue_xmit);
 870
 871         return NF_STOLEN;
 872 }
 873
 874 /* IP/SABOTAGE *****************************************************/
 875 /* Don't hand locally destined packets to PF_INET(6)/PRE_ROUTING
 876  * for the second time. */
 877 static unsigned int ip_sabotage_in(unsigned int hook, struct sk_buff *skb,
 878                                    const struct net_device *in,
 879                                    const struct net_device *out,
 880                                    int (*okfn)(struct sk_buff *))
 881 {
 882         if (skb->nf_bridge &&
 883             !(skb->nf_bridge->mask & BRNF_NF_BRIDGE_PREROUTING)) {
 884                 return NF_STOP;
 885         }
 886
 887         return NF_ACCEPT;
 888 }
 889
 890 /* For br_nf_post_routing, we need (prio = NF_BR_PRI_LAST), because
 891  * br_dev_queue_push_xmit is called afterwards */
 892 static struct nf_hook_ops br_nf_ops[] __read_mostly = {
 893         {
 894                 .hook = br_nf_pre_routing,
 895                 .owner = THIS_MODULE,
 896                 .pf = PF_BRIDGE,
 897                 .hooknum = NF_BR_PRE_ROUTING,
 898                 .priority = NF_BR_PRI_BRNF,
 899         },
 900         {
 901                 .hook = br_nf_local_in,
 902                 .owner = THIS_MODULE,
 903                 .pf = PF_BRIDGE,
 904                 .hooknum = NF_BR_LOCAL_IN,
 905                 .priority = NF_BR_PRI_BRNF,
 906         },
 907         {
 908                 .hook = br_nf_forward_ip,
 909                 .owner = THIS_MODULE,
 910                 .pf = PF_BRIDGE,
 911                 .hooknum = NF_BR_FORWARD,
 912                 .priority = NF_BR_PRI_BRNF - 1,
 913         },
 914         {
 915                 .hook = br_nf_forward_arp,
 916                 .owner = THIS_MODULE,
 917                 .pf = PF_BRIDGE,
 918                 .hooknum = NF_BR_FORWARD,
 919                 .priority = NF_BR_PRI_BRNF,
 920         },
 921         {
 922                 .hook = br_nf_post_routing,
 923                 .owner = THIS_MODULE,
 924                 .pf = PF_BRIDGE,
 925                 .hooknum = NF_BR_POST_ROUTING,
 926                 .priority = NF_BR_PRI_LAST,
 927         },
 928         {
 929                 .hook = ip_sabotage_in,
 930                 .owner = THIS_MODULE,
 931                 .pf = PF_INET,
 932                 .hooknum = NF_INET_PRE_ROUTING,
 933                 .priority = NF_IP_PRI_FIRST,
 934         },
 935         {
 936                 .hook = ip_sabotage_in,
 937                 .owner = THIS_MODULE,
 938                 .pf = PF_INET6,
 939                 .hooknum = NF_INET_PRE_ROUTING,
 940                 .priority = NF_IP6_PRI_FIRST,
 941         },
 942 };
 943
 944 #ifdef CONFIG_SYSCTL
 945 static
 946 int brnf_sysctl_call_tables(ctl_table * ctl, int write,
 947                             void __user * buffer, size_t * lenp, loff_t * ppos)
 948 {
 949         int ret;
 950
 951         ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
 952
 953         if (write && *(int *)(ctl->data))
 954                 *(int *)(ctl->data) = 1;
 955         return ret;
 956 }
 957
 958 static ctl_table brnf_table[] = {
 959         {
 960                 .procname       = "bridge-nf-call-arptables",
 961                 .data           = &brnf_call_arptables,
 962                 .maxlen         = sizeof(int),
 963                 .mode           = 0644,
 964                 .proc_handler   = brnf_sysctl_call_tables,
 965         },
 966         {
 967                 .procname       = "bridge-nf-call-iptables",
 968                 .data           = &brnf_call_iptables,
 969                 .maxlen         = sizeof(int),
 970                 .mode           = 0644,
 971                 .proc_handler   = brnf_sysctl_call_tables,
 972         },
 973         {
 974                 .procname       = "bridge-nf-call-ip6tables",
 975                 .data           = &brnf_call_ip6tables,
 976                 .maxlen         = sizeof(int),
 977                 .mode           = 0644,
 978                 .proc_handler   = brnf_sysctl_call_tables,
 979         },
 980         {
 981                 .procname       = "bridge-nf-filter-vlan-tagged",
 982                 .data           = &brnf_filter_vlan_tagged,
 983                 .maxlen         = sizeof(int),
 984                 .mode           = 0644,
 985                 .proc_handler   = brnf_sysctl_call_tables,
 986         },
 987         {
 988                 .procname       = "bridge-nf-filter-pppoe-tagged",
 989                 .data           = &brnf_filter_pppoe_tagged,
 990                 .maxlen         = sizeof(int),
 991                 .mode           = 0644,
 992                 .proc_handler   = brnf_sysctl_call_tables,
 993         },
 994         { }
 995 };
 996
 997 static struct ctl_path brnf_path[] = {
 998         { .procname = "net", },
 999         { .procname = "bridge", },
1000         { }
1001 };
1002 #endif
1003
1004 int __init br_netfilter_init(void)
1005 {
1006         int ret;
1007
1008         ret = dst_entries_init(&fake_dst_ops);
1009         if (ret < 0)
1010                 return ret;
1011
1012         ret = nf_register_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
1013         if (ret < 0) {
1014                 dst_entries_destroy(&fake_dst_ops);
1015                 return ret;
1016         }
1017 #ifdef CONFIG_SYSCTL
1018         brnf_sysctl_header = register_sysctl_paths(brnf_path, brnf_table);
1019         if (brnf_sysctl_header == NULL) {
1020                 printk(KERN_WARNING
1021                        "br_netfilter: can't register to sysctl.\n");
1022                 nf_unregister_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
1023                 dst_entries_destroy(&fake_dst_ops);
1024                 return -ENOMEM;
1025         }
1026 #endif
1027         printk(KERN_NOTICE "Bridge firewalling registered\n");
1028         return 0;
1029 }
1030
1031 void br_netfilter_fini(void)
1032 {
1033         nf_unregister_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
1034 #ifdef CONFIG_SYSCTL
1035         unregister_sysctl_table(brnf_sysctl_header);
1036 #endif
1037         dst_entries_destroy(&fake_dst_ops);
1038 }