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