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