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 (maintainer)   <bdschuym@pandora.be>
   8  *
   9  *      Changes:
  10  *      Apr 29 2003: physdev module support (bdschuym)
  11  *      Jun 19 2003: let arptables see bridged ARP traffic (bdschuym)
  12  *      Oct 06 2003: filter encapsulated IP/ARP VLAN traffic on untagged bridge
  13  *                   (bdschuym)
  14  *      Sep 01 2004: add IPv6 filtering (bdschuym)
  15  *
  16  *      This program is free software; you can redistribute it and/or
  17  *      modify it under the terms of the GNU General Public License
  18  *      as published by the Free Software Foundation; either version
  19  *      2 of the License, or (at your option) any later version.
  20  *
  21  *      Lennert dedicates this file to Kerstin Wurdinger.
  22  */
  23
  24 #include <linux/module.h>
  25 #include <linux/kernel.h>
  26 #include <linux/ip.h>
  27 #include <linux/netdevice.h>
  28 #include <linux/skbuff.h>
  29 #include <linux/if_arp.h>
  30 #include <linux/if_ether.h>
  31 #include <linux/if_vlan.h>
  32 #include <linux/if_pppox.h>
  33 #include <linux/ppp_defs.h>
  34 #include <linux/netfilter_bridge.h>
  35 #include <linux/netfilter_ipv4.h>
  36 #include <linux/netfilter_ipv6.h>
  37 #include <linux/netfilter_arp.h>
  38 #include <linux/in_route.h>
  39 #include <linux/inetdevice.h>
  40
  41 #include <net/ip.h>
  42 #include <net/ipv6.h>
  43 #include <net/route.h>
  44
  45 #include <asm/uaccess.h>
  46 #include "br_private.h"
  47 #ifdef CONFIG_SYSCTL
  48 #include <linux/sysctl.h>
  49 #endif
  50
  51 #define skb_origaddr(skb)        (((struct bridge_skb_cb *) \
  52                                  (skb->nf_bridge->data))->daddr.ipv4)
  53 #define store_orig_dstaddr(skb)  (skb_origaddr(skb) = ip_hdr(skb)->daddr)
  54 #define dnat_took_place(skb)     (skb_origaddr(skb) != ip_hdr(skb)->daddr)
  55
  56 #ifdef CONFIG_SYSCTL
  57 static struct ctl_table_header *brnf_sysctl_header;
  58 static int brnf_call_iptables __read_mostly = 1;
  59 static int brnf_call_ip6tables __read_mostly = 1;
  60 static int brnf_call_arptables __read_mostly = 1;
  61 static int brnf_filter_vlan_tagged __read_mostly = 1;
  62 static int brnf_filter_pppoe_tagged __read_mostly = 1;
  63 #else
  64 #define brnf_filter_vlan_tagged 1
  65 #define brnf_filter_pppoe_tagged 1
  66 #endif
  67
  68 static inline __be16 vlan_proto(const struct sk_buff *skb)
  69 {
  70         return vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
  71 }
  72
  73 #define IS_VLAN_IP(skb) \
  74         (skb->protocol == htons(ETH_P_8021Q) && \
  75          vlan_proto(skb) == htons(ETH_P_IP) &&  \
  76          brnf_filter_vlan_tagged)
  77
  78 #define IS_VLAN_IPV6(skb) \
  79         (skb->protocol == htons(ETH_P_8021Q) && \
  80          vlan_proto(skb) == htons(ETH_P_IPV6) &&\
  81          brnf_filter_vlan_tagged)
  82
  83 #define IS_VLAN_ARP(skb) \
  84         (skb->protocol == htons(ETH_P_8021Q) && \
  85          vlan_proto(skb) == htons(ETH_P_ARP) && \
  86          brnf_filter_vlan_tagged)
  87
  88 static inline __be16 pppoe_proto(const struct sk_buff *skb)
  89 {
  90         return *((__be16 *)(skb_mac_header(skb) + ETH_HLEN +
  91                             sizeof(struct pppoe_hdr)));
  92 }
  93
  94 #define IS_PPPOE_IP(skb) \
  95         (skb->protocol == htons(ETH_P_PPP_SES) && \
  96          pppoe_proto(skb) == htons(PPP_IP) && \
  97          brnf_filter_pppoe_tagged)
  98
  99 #define IS_PPPOE_IPV6(skb) \
 100         (skb->protocol == htons(ETH_P_PPP_SES) && \
 101          pppoe_proto(skb) == htons(PPP_IPV6) && \
 102          brnf_filter_pppoe_tagged)
 103
 104 /* We need these fake structures to make netfilter happy --
 105  * lots of places assume that skb->dst != NULL, which isn't
 106  * all that unreasonable.
 107  *
 108  * Currently, we fill in the PMTU entry because netfilter
 109  * refragmentation needs it, and the rt_flags entry because
 110  * ipt_REJECT needs it.  Future netfilter modules might
 111  * require us to fill additional fields. */
 112 static struct net_device __fake_net_device = {
 113         .hard_header_len        = ETH_HLEN
 114 };
 115
 116 static struct rtable __fake_rtable = {
 117         .u = {
 118                 .dst = {
 119                         .__refcnt               = ATOMIC_INIT(1),
 120                         .dev                    = &__fake_net_device,
 121                         .path                   = &__fake_rtable.u.dst,
 122                         .metrics                = {[RTAX_MTU - 1] = 1500},
 123                         .flags                  = DST_NOXFRM,
 124                 }
 125         },
 126         .rt_flags       = 0,
 127 };
 128
 129 static inline struct net_device *bridge_parent(const struct net_device *dev)
 130 {
 131         struct net_bridge_port *port = rcu_dereference(dev->br_port);
 132
 133         return port ? port->br->dev : NULL;
 134 }
 135
 136 static inline struct nf_bridge_info *nf_bridge_alloc(struct sk_buff *skb)
 137 {
 138         skb->nf_bridge = kzalloc(sizeof(struct nf_bridge_info), GFP_ATOMIC);
 139         if (likely(skb->nf_bridge))
 140                 atomic_set(&(skb->nf_bridge->use), 1);
 141
 142         return skb->nf_bridge;
 143 }
 144
 145 static inline void nf_bridge_push_encap_header(struct sk_buff *skb)
 146 {
 147         unsigned int len = nf_bridge_encap_header_len(skb);
 148
 149         skb_push(skb, len);
 150         skb->network_header -= len;
 151 }
 152
 153 static inline void nf_bridge_pull_encap_header(struct sk_buff *skb)
 154 {
 155         unsigned int len = nf_bridge_encap_header_len(skb);
 156
 157         skb_pull(skb, len);
 158         skb->network_header += len;
 159 }
 160
 161 static inline void nf_bridge_pull_encap_header_rcsum(struct sk_buff *skb)
 162 {
 163         unsigned int len = nf_bridge_encap_header_len(skb);
 164
 165         skb_pull_rcsum(skb, len);
 166         skb->network_header += len;
 167 }
 168
 169 static inline void nf_bridge_save_header(struct sk_buff *skb)
 170 {
 171         int header_size = ETH_HLEN + nf_bridge_encap_header_len(skb);
 172
 173         skb_copy_from_linear_data_offset(skb, -header_size,
 174                                          skb->nf_bridge->data, header_size);
 175 }
 176
 177 /*
 178  * When forwarding bridge frames, we save a copy of the original
 179  * header before processing.
 180  */
 181 int nf_bridge_copy_header(struct sk_buff *skb)
 182 {
 183         int err;
 184         int header_size = ETH_HLEN + nf_bridge_encap_header_len(skb);
 185
 186         err = skb_cow_head(skb, header_size);
 187         if (err)
 188                 return err;
 189
 190         skb_copy_to_linear_data_offset(skb, -header_size,
 191                                        skb->nf_bridge->data, header_size);
 192         __skb_push(skb, nf_bridge_encap_header_len(skb));
 193         return 0;
 194 }
 195
 196 /* PF_BRIDGE/PRE_ROUTING *********************************************/
 197 /* Undo the changes made for ip6tables PREROUTING and continue the
 198  * bridge PRE_ROUTING hook. */
 199 static int br_nf_pre_routing_finish_ipv6(struct sk_buff *skb)
 200 {
 201         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 202
 203         if (nf_bridge->mask & BRNF_PKT_TYPE) {
 204                 skb->pkt_type = PACKET_OTHERHOST;
 205                 nf_bridge->mask ^= BRNF_PKT_TYPE;
 206         }
 207         nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
 208
 209         skb->dst = (struct dst_entry *)&__fake_rtable;
 210         dst_hold(skb->dst);
 211
 212         skb->dev = nf_bridge->physindev;
 213         nf_bridge_push_encap_header(skb);
 214         NF_HOOK_THRESH(PF_BRIDGE, NF_BR_PRE_ROUTING, skb, skb->dev, NULL,
 215                        br_handle_frame_finish, 1);
 216
 217         return 0;
 218 }
 219
 220 static void __br_dnat_complain(void)
 221 {
 222         static unsigned long last_complaint;
 223
 224         if (jiffies - last_complaint >= 5 * HZ) {
 225                 printk(KERN_WARNING "Performing cross-bridge DNAT requires IP "
 226                        "forwarding to be enabled\n");
 227                 last_complaint = jiffies;
 228         }
 229 }
 230
 231 /* This requires some explaining. If DNAT has taken place,
 232  * we will need to fix up the destination Ethernet address,
 233  * and this is a tricky process.
 234  *
 235  * There are two cases to consider:
 236  * 1. The packet was DNAT'ed to a device in the same bridge
 237  *    port group as it was received on. We can still bridge
 238  *    the packet.
 239  * 2. The packet was DNAT'ed to a different device, either
 240  *    a non-bridged device or another bridge port group.
 241  *    The packet will need to be routed.
 242  *
 243  * The correct way of distinguishing between these two cases is to
 244  * call ip_route_input() and to look at skb->dst->dev, which is
 245  * changed to the destination device if ip_route_input() succeeds.
 246  *
 247  * Let us first consider the case that ip_route_input() succeeds:
 248  *
 249  * If skb->dst->dev equals the logical bridge device the packet
 250  * came in on, we can consider this bridging. We then call
 251  * skb->dst->output() which will make the packet enter br_nf_local_out()
 252  * not much later. In that function it is assured that the iptables
 253  * FORWARD chain is traversed for the packet.
 254  *
 255  * Otherwise, the packet is considered to be routed and we just
 256  * change the destination MAC address so that the packet will
 257  * later be passed up to the IP stack to be routed. For a redirected
 258  * packet, ip_route_input() will give back the localhost as output device,
 259  * which differs from the bridge device.
 260  *
 261  * Let us now consider the case that ip_route_input() fails:
 262  *
 263  * This can be because the destination address is martian, in which case
 264  * the packet will be dropped.
 265  * After a "echo '0' > /proc/sys/net/ipv4/ip_forward" ip_route_input()
 266  * will fail, while __ip_route_output_key() will return success. The source
 267  * address for __ip_route_output_key() is set to zero, so __ip_route_output_key
 268  * thinks we're handling a locally generated packet and won't care
 269  * if IP forwarding is allowed. We send a warning message to the users's
 270  * log telling her to put IP forwarding on.
 271  *
 272  * ip_route_input() will also fail if there is no route available.
 273  * In that case we just drop the packet.
 274  *
 275  * --Lennert, 20020411
 276  * --Bart, 20020416 (updated)
 277  * --Bart, 20021007 (updated)
 278  * --Bart, 20062711 (updated) */
 279 static int br_nf_pre_routing_finish_bridge(struct sk_buff *skb)
 280 {
 281         if (skb->pkt_type == PACKET_OTHERHOST) {
 282                 skb->pkt_type = PACKET_HOST;
 283                 skb->nf_bridge->mask |= BRNF_PKT_TYPE;
 284         }
 285         skb->nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
 286
 287         skb->dev = bridge_parent(skb->dev);
 288         if (!skb->dev)
 289                 kfree_skb(skb);
 290         else {
 291                 nf_bridge_pull_encap_header(skb);
 292                 skb->dst->output(skb);
 293         }
 294         return 0;
 295 }
 296
 297 static int br_nf_pre_routing_finish(struct sk_buff *skb)
 298 {
 299         struct net_device *dev = skb->dev;
 300         struct iphdr *iph = ip_hdr(skb);
 301         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 302         int err;
 303
 304         if (nf_bridge->mask & BRNF_PKT_TYPE) {
 305                 skb->pkt_type = PACKET_OTHERHOST;
 306                 nf_bridge->mask ^= BRNF_PKT_TYPE;
 307         }
 308         nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
 309         if (dnat_took_place(skb)) {
 310                 if ((err = ip_route_input(skb, iph->daddr, iph->saddr, iph->tos, dev))) {
 311                         struct rtable *rt;
 312                         struct flowi fl = {
 313                                 .nl_u = {
 314                                         .ip4_u = {
 315                                                  .daddr = iph->daddr,
 316                                                  .saddr = 0,
 317                                                  .tos = RT_TOS(iph->tos) },
 318                                 },
 319                                 .proto = 0,
 320                         };
 321                         struct in_device *in_dev = in_dev_get(dev);
 322
 323                         /* If err equals -EHOSTUNREACH the error is due to a
 324                          * martian destination or due to the fact that
 325                          * forwarding is disabled. For most martian packets,
 326                          * ip_route_output_key() will fail. It won't fail for 2 types of
 327                          * martian destinations: loopback destinations and destination
 328                          * 0.0.0.0. In both cases the packet will be dropped because the
 329                          * destination is the loopback device and not the bridge. */
 330                         if (err != -EHOSTUNREACH || !in_dev || IN_DEV_FORWARD(in_dev))
 331                                 goto free_skb;
 332
 333                         if (!ip_route_output_key(&rt, &fl)) {
 334                                 /* - Bridged-and-DNAT'ed traffic doesn't
 335                                  *   require ip_forwarding. */
 336                                 if (((struct dst_entry *)rt)->dev == dev) {
 337                                         skb->dst = (struct dst_entry *)rt;
 338                                         goto bridged_dnat;
 339                                 }
 340                                 /* we are sure that forwarding is disabled, so printing
 341                                  * this message is no problem. Note that the packet could
 342                                  * still have a martian destination address, in which case
 343                                  * the packet could be dropped even if forwarding were enabled */
 344                                 __br_dnat_complain();
 345                                 dst_release((struct dst_entry *)rt);
 346                         }
 347 free_skb:
 348                         kfree_skb(skb);
 349                         return 0;
 350                 } else {
 351                         if (skb->dst->dev == dev) {
 352 bridged_dnat:
 353                                 /* Tell br_nf_local_out this is a
 354                                  * bridged frame */
 355                                 nf_bridge->mask |= BRNF_BRIDGED_DNAT;
 356                                 skb->dev = nf_bridge->physindev;
 357                                 nf_bridge_push_encap_header(skb);
 358                                 NF_HOOK_THRESH(PF_BRIDGE, NF_BR_PRE_ROUTING,
 359                                                skb, skb->dev, NULL,
 360                                                br_nf_pre_routing_finish_bridge,
 361                                                1);
 362                                 return 0;
 363                         }
 364                         memcpy(eth_hdr(skb)->h_dest, dev->dev_addr, ETH_ALEN);
 365                         skb->pkt_type = PACKET_HOST;
 366                 }
 367         } else {
 368                 skb->dst = (struct dst_entry *)&__fake_rtable;
 369                 dst_hold(skb->dst);
 370         }
 371
 372         skb->dev = nf_bridge->physindev;
 373         nf_bridge_push_encap_header(skb);
 374         NF_HOOK_THRESH(PF_BRIDGE, NF_BR_PRE_ROUTING, skb, skb->dev, NULL,
 375                        br_handle_frame_finish, 1);
 376
 377         return 0;
 378 }
 379
 380 /* Some common code for IPv4/IPv6 */
 381 static struct net_device *setup_pre_routing(struct sk_buff *skb)
 382 {
 383         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 384
 385         if (skb->pkt_type == PACKET_OTHERHOST) {
 386                 skb->pkt_type = PACKET_HOST;
 387                 nf_bridge->mask |= BRNF_PKT_TYPE;
 388         }
 389
 390         nf_bridge->mask |= BRNF_NF_BRIDGE_PREROUTING;
 391         nf_bridge->physindev = skb->dev;
 392         skb->dev = bridge_parent(skb->dev);
 393
 394         return skb->dev;
 395 }
 396
 397 /* We only check the length. A bridge shouldn't do any hop-by-hop stuff anyway */
 398 static int check_hbh_len(struct sk_buff *skb)
 399 {
 400         unsigned char *raw = (u8 *)(ipv6_hdr(skb) + 1);
 401         u32 pkt_len;
 402         const unsigned char *nh = skb_network_header(skb);
 403         int off = raw - nh;
 404         int len = (raw[1] + 1) << 3;
 405
 406         if ((raw + len) - skb->data > skb_headlen(skb))
 407                 goto bad;
 408
 409         off += 2;
 410         len -= 2;
 411
 412         while (len > 0) {
 413                 int optlen = nh[off + 1] + 2;
 414
 415                 switch (nh[off]) {
 416                 case IPV6_TLV_PAD0:
 417                         optlen = 1;
 418                         break;
 419
 420                 case IPV6_TLV_PADN:
 421                         break;
 422
 423                 case IPV6_TLV_JUMBO:
 424                         if (nh[off + 1] != 4 || (off & 3) != 2)
 425                                 goto bad;
 426                         pkt_len = ntohl(*(__be32 *) (nh + off + 2));
 427                         if (pkt_len <= IPV6_MAXPLEN ||
 428                             ipv6_hdr(skb)->payload_len)
 429                                 goto bad;
 430                         if (pkt_len > skb->len - sizeof(struct ipv6hdr))
 431                                 goto bad;
 432                         if (pskb_trim_rcsum(skb,
 433                                             pkt_len + sizeof(struct ipv6hdr)))
 434                                 goto bad;
 435                         nh = skb_network_header(skb);
 436                         break;
 437                 default:
 438                         if (optlen > len)
 439                                 goto bad;
 440                         break;
 441                 }
 442                 off += optlen;
 443                 len -= optlen;
 444         }
 445         if (len == 0)
 446                 return 0;
 447 bad:
 448         return -1;
 449
 450 }
 451
 452 /* Replicate the checks that IPv6 does on packet reception and pass the packet
 453  * to ip6tables, which doesn't support NAT, so things are fairly simple. */
 454 static unsigned int br_nf_pre_routing_ipv6(unsigned int hook,
 455                                            struct sk_buff *skb,
 456                                            const struct net_device *in,
 457                                            const struct net_device *out,
 458                                            int (*okfn)(struct sk_buff *))
 459 {
 460         struct ipv6hdr *hdr;
 461         u32 pkt_len;
 462
 463         if (skb->len < sizeof(struct ipv6hdr))
 464                 goto inhdr_error;
 465
 466         if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
 467                 goto inhdr_error;
 468
 469         hdr = ipv6_hdr(skb);
 470
 471         if (hdr->version != 6)
 472                 goto inhdr_error;
 473
 474         pkt_len = ntohs(hdr->payload_len);
 475
 476         if (pkt_len || hdr->nexthdr != NEXTHDR_HOP) {
 477                 if (pkt_len + sizeof(struct ipv6hdr) > skb->len)
 478                         goto inhdr_error;
 479                 if (pskb_trim_rcsum(skb, pkt_len + sizeof(struct ipv6hdr)))
 480                         goto inhdr_error;
 481         }
 482         if (hdr->nexthdr == NEXTHDR_HOP && check_hbh_len(skb))
 483                 goto inhdr_error;
 484
 485         nf_bridge_put(skb->nf_bridge);
 486         if (!nf_bridge_alloc(skb))
 487                 return NF_DROP;
 488         if (!setup_pre_routing(skb))
 489                 return NF_DROP;
 490
 491         NF_HOOK(PF_INET6, NF_IP6_PRE_ROUTING, skb, skb->dev, NULL,
 492                 br_nf_pre_routing_finish_ipv6);
 493
 494         return NF_STOLEN;
 495
 496 inhdr_error:
 497         return NF_DROP;
 498 }
 499
 500 /* Direct IPv6 traffic to br_nf_pre_routing_ipv6.
 501  * Replicate the checks that IPv4 does on packet reception.
 502  * Set skb->dev to the bridge device (i.e. parent of the
 503  * receiving device) to make netfilter happy, the REDIRECT
 504  * target in particular.  Save the original destination IP
 505  * address to be able to detect DNAT afterwards. */
 506 static unsigned int br_nf_pre_routing(unsigned int hook, struct sk_buff **pskb,
 507                                       const struct net_device *in,
 508                                       const struct net_device *out,
 509                                       int (*okfn)(struct sk_buff *))
 510 {
 511         struct iphdr *iph;
 512         struct sk_buff *skb = *pskb;
 513         __u32 len = nf_bridge_encap_header_len(skb);
 514
 515         if ((skb = skb_share_check(skb, GFP_ATOMIC)) == NULL)
 516                 return NF_STOLEN;
 517
 518         if (unlikely(!pskb_may_pull(skb, len)))
 519                 goto out;
 520
 521         if (skb->protocol == htons(ETH_P_IPV6) || IS_VLAN_IPV6(skb) ||
 522             IS_PPPOE_IPV6(skb)) {
 523 #ifdef CONFIG_SYSCTL
 524                 if (!brnf_call_ip6tables)
 525                         return NF_ACCEPT;
 526 #endif
 527                 nf_bridge_pull_encap_header_rcsum(skb);
 528                 return br_nf_pre_routing_ipv6(hook, skb, in, out, okfn);
 529         }
 530 #ifdef CONFIG_SYSCTL
 531         if (!brnf_call_iptables)
 532                 return NF_ACCEPT;
 533 #endif
 534
 535         if (skb->protocol != htons(ETH_P_IP) && !IS_VLAN_IP(skb) &&
 536             !IS_PPPOE_IP(skb))
 537                 return NF_ACCEPT;
 538
 539         nf_bridge_pull_encap_header_rcsum(skb);
 540
 541         if (!pskb_may_pull(skb, sizeof(struct iphdr)))
 542                 goto inhdr_error;
 543
 544         iph = ip_hdr(skb);
 545         if (iph->ihl < 5 || iph->version != 4)
 546                 goto inhdr_error;
 547
 548         if (!pskb_may_pull(skb, 4 * iph->ihl))
 549                 goto inhdr_error;
 550
 551         iph = ip_hdr(skb);
 552         if (ip_fast_csum((__u8 *) iph, iph->ihl) != 0)
 553                 goto inhdr_error;
 554
 555         len = ntohs(iph->tot_len);
 556         if (skb->len < len || len < 4 * iph->ihl)
 557                 goto inhdr_error;
 558
 559         pskb_trim_rcsum(skb, len);
 560
 561         nf_bridge_put(skb->nf_bridge);
 562         if (!nf_bridge_alloc(skb))
 563                 return NF_DROP;
 564         if (!setup_pre_routing(skb))
 565                 return NF_DROP;
 566         store_orig_dstaddr(skb);
 567
 568         NF_HOOK(PF_INET, NF_IP_PRE_ROUTING, skb, skb->dev, NULL,
 569                 br_nf_pre_routing_finish);
 570
 571         return NF_STOLEN;
 572
 573 inhdr_error:
 574 //      IP_INC_STATS_BH(IpInHdrErrors);
 575 out:
 576         return NF_DROP;
 577 }
 578
 579
 580 /* PF_BRIDGE/LOCAL_IN ************************************************/
 581 /* The packet is locally destined, which requires a real
 582  * dst_entry, so detach the fake one.  On the way up, the
 583  * packet would pass through PRE_ROUTING again (which already
 584  * took place when the packet entered the bridge), but we
 585  * register an IPv4 PRE_ROUTING 'sabotage' hook that will
 586  * prevent this from happening. */
 587 static unsigned int br_nf_local_in(unsigned int hook, struct sk_buff **pskb,
 588                                    const struct net_device *in,
 589                                    const struct net_device *out,
 590                                    int (*okfn)(struct sk_buff *))
 591 {
 592         struct sk_buff *skb = *pskb;
 593
 594         if (skb->dst == (struct dst_entry *)&__fake_rtable) {
 595                 dst_release(skb->dst);
 596                 skb->dst = NULL;
 597         }
 598
 599         return NF_ACCEPT;
 600 }
 601
 602 /* PF_BRIDGE/FORWARD *************************************************/
 603 static int br_nf_forward_finish(struct sk_buff *skb)
 604 {
 605         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 606         struct net_device *in;
 607
 608         if (skb->protocol != htons(ETH_P_ARP) && !IS_VLAN_ARP(skb)) {
 609                 in = nf_bridge->physindev;
 610                 if (nf_bridge->mask & BRNF_PKT_TYPE) {
 611                         skb->pkt_type = PACKET_OTHERHOST;
 612                         nf_bridge->mask ^= BRNF_PKT_TYPE;
 613                 }
 614         } else {
 615                 in = *((struct net_device **)(skb->cb));
 616         }
 617         nf_bridge_push_encap_header(skb);
 618         NF_HOOK_THRESH(PF_BRIDGE, NF_BR_FORWARD, skb, in,
 619                        skb->dev, br_forward_finish, 1);
 620         return 0;
 621 }
 622
 623 /* This is the 'purely bridged' case.  For IP, we pass the packet to
 624  * netfilter with indev and outdev set to the bridge device,
 625  * but we are still able to filter on the 'real' indev/outdev
 626  * because of the physdev module. For ARP, indev and outdev are the
 627  * bridge ports. */
 628 static unsigned int br_nf_forward_ip(unsigned int hook, struct sk_buff **pskb,
 629                                      const struct net_device *in,
 630                                      const struct net_device *out,
 631                                      int (*okfn)(struct sk_buff *))
 632 {
 633         struct sk_buff *skb = *pskb;
 634         struct nf_bridge_info *nf_bridge;
 635         struct net_device *parent;
 636         int pf;
 637
 638         if (!skb->nf_bridge)
 639                 return NF_ACCEPT;
 640
 641         parent = bridge_parent(out);
 642         if (!parent)
 643                 return NF_DROP;
 644
 645         if (skb->protocol == htons(ETH_P_IP) || IS_VLAN_IP(skb) ||
 646             IS_PPPOE_IP(skb))
 647                 pf = PF_INET;
 648         else
 649                 pf = PF_INET6;
 650
 651         nf_bridge_pull_encap_header(*pskb);
 652
 653         nf_bridge = skb->nf_bridge;
 654         if (skb->pkt_type == PACKET_OTHERHOST) {
 655                 skb->pkt_type = PACKET_HOST;
 656                 nf_bridge->mask |= BRNF_PKT_TYPE;
 657         }
 658
 659         /* The physdev module checks on this */
 660         nf_bridge->mask |= BRNF_BRIDGED;
 661         nf_bridge->physoutdev = skb->dev;
 662
 663         NF_HOOK(pf, NF_IP_FORWARD, skb, bridge_parent(in), parent,
 664                 br_nf_forward_finish);
 665
 666         return NF_STOLEN;
 667 }
 668
 669 static unsigned int br_nf_forward_arp(unsigned int hook, struct sk_buff **pskb,
 670                                       const struct net_device *in,
 671                                       const struct net_device *out,
 672                                       int (*okfn)(struct sk_buff *))
 673 {
 674         struct sk_buff *skb = *pskb;
 675         struct net_device **d = (struct net_device **)(skb->cb);
 676
 677 #ifdef CONFIG_SYSCTL
 678         if (!brnf_call_arptables)
 679                 return NF_ACCEPT;
 680 #endif
 681
 682         if (skb->protocol != htons(ETH_P_ARP)) {
 683                 if (!IS_VLAN_ARP(skb))
 684                         return NF_ACCEPT;
 685                 nf_bridge_pull_encap_header(*pskb);
 686         }
 687
 688         if (arp_hdr(skb)->ar_pln != 4) {
 689                 if (IS_VLAN_ARP(skb))
 690                         nf_bridge_push_encap_header(*pskb);
 691                 return NF_ACCEPT;
 692         }
 693         *d = (struct net_device *)in;
 694         NF_HOOK(NF_ARP, NF_ARP_FORWARD, skb, (struct net_device *)in,
 695                 (struct net_device *)out, br_nf_forward_finish);
 696
 697         return NF_STOLEN;
 698 }
 699
 700 /* PF_BRIDGE/LOCAL_OUT ***********************************************
 701  *
 702  * This function sees both locally originated IP packets and forwarded
 703  * IP packets (in both cases the destination device is a bridge
 704  * device). It also sees bridged-and-DNAT'ed packets.
 705  *
 706  * If (nf_bridge->mask & BRNF_BRIDGED_DNAT) then the packet is bridged
 707  * and we fake the PF_BRIDGE/FORWARD hook. The function br_nf_forward()
 708  * will then fake the PF_INET/FORWARD hook. br_nf_local_out() has priority
 709  * NF_BR_PRI_FIRST, so no relevant PF_BRIDGE/INPUT functions have been nor
 710  * will be executed.
 711  */
 712 static unsigned int br_nf_local_out(unsigned int hook, struct sk_buff **pskb,
 713                                     const struct net_device *in,
 714                                     const struct net_device *out,
 715                                     int (*okfn)(struct sk_buff *))
 716 {
 717         struct net_device *realindev;
 718         struct sk_buff *skb = *pskb;
 719         struct nf_bridge_info *nf_bridge;
 720
 721         if (!skb->nf_bridge)
 722                 return NF_ACCEPT;
 723
 724         nf_bridge = skb->nf_bridge;
 725         if (!(nf_bridge->mask & BRNF_BRIDGED_DNAT))
 726                 return NF_ACCEPT;
 727
 728         /* Bridged, take PF_BRIDGE/FORWARD.
 729          * (see big note in front of br_nf_pre_routing_finish) */
 730         nf_bridge->physoutdev = skb->dev;
 731         realindev = nf_bridge->physindev;
 732
 733         if (nf_bridge->mask & BRNF_PKT_TYPE) {
 734                 skb->pkt_type = PACKET_OTHERHOST;
 735                 nf_bridge->mask ^= BRNF_PKT_TYPE;
 736         }
 737         nf_bridge_push_encap_header(skb);
 738
 739         NF_HOOK(PF_BRIDGE, NF_BR_FORWARD, skb, realindev, skb->dev,
 740                 br_forward_finish);
 741         return NF_STOLEN;
 742 }
 743
 744 static int br_nf_dev_queue_xmit(struct sk_buff *skb)
 745 {
 746         if (skb->protocol == htons(ETH_P_IP) &&
 747             skb->len > skb->dev->mtu &&
 748             !skb_is_gso(skb))
 749                 return ip_fragment(skb, br_dev_queue_push_xmit);
 750         else
 751                 return br_dev_queue_push_xmit(skb);
 752 }
 753
 754 /* PF_BRIDGE/POST_ROUTING ********************************************/
 755 static unsigned int br_nf_post_routing(unsigned int hook, struct sk_buff **pskb,
 756                                        const struct net_device *in,
 757                                        const struct net_device *out,
 758                                        int (*okfn)(struct sk_buff *))
 759 {
 760         struct sk_buff *skb = *pskb;
 761         struct nf_bridge_info *nf_bridge = (*pskb)->nf_bridge;
 762         struct net_device *realoutdev = bridge_parent(skb->dev);
 763         int pf;
 764
 765 #ifdef CONFIG_NETFILTER_DEBUG
 766         /* Be very paranoid. This probably won't happen anymore, but let's
 767          * keep the check just to be sure... */
 768         if (skb_mac_header(skb) < skb->head ||
 769             skb_mac_header(skb) + ETH_HLEN > skb->data) {
 770                 printk(KERN_CRIT "br_netfilter: Argh!! br_nf_post_routing: "
 771                        "bad mac.raw pointer.\n");
 772                 goto print_error;
 773         }
 774 #endif
 775
 776         if (!nf_bridge)
 777                 return NF_ACCEPT;
 778
 779         if (!realoutdev)
 780                 return NF_DROP;
 781
 782         if (skb->protocol == htons(ETH_P_IP) || IS_VLAN_IP(skb) ||
 783             IS_PPPOE_IP(skb))
 784                 pf = PF_INET;
 785         else
 786                 pf = PF_INET6;
 787
 788 #ifdef CONFIG_NETFILTER_DEBUG
 789         if (skb->dst == NULL) {
 790                 printk(KERN_INFO "br_netfilter post_routing: skb->dst == NULL\n");
 791                 goto print_error;
 792         }
 793 #endif
 794
 795         /* We assume any code from br_dev_queue_push_xmit onwards doesn't care
 796          * about the value of skb->pkt_type. */
 797         if (skb->pkt_type == PACKET_OTHERHOST) {
 798                 skb->pkt_type = PACKET_HOST;
 799                 nf_bridge->mask |= BRNF_PKT_TYPE;
 800         }
 801
 802         nf_bridge_pull_encap_header(skb);
 803         nf_bridge_save_header(skb);
 804
 805 #if defined(CONFIG_VLAN_8021Q) || defined(CONFIG_VLAN_8021Q_MODULE)
 806         if (nf_bridge->netoutdev)
 807                 realoutdev = nf_bridge->netoutdev;
 808 #endif
 809         NF_HOOK(pf, NF_IP_POST_ROUTING, skb, NULL, realoutdev,
 810                 br_nf_dev_queue_xmit);
 811
 812         return NF_STOLEN;
 813
 814 #ifdef CONFIG_NETFILTER_DEBUG
 815 print_error:
 816         if (skb->dev != NULL) {
 817                 printk("[%s]", skb->dev->name);
 818                 if (realoutdev)
 819                         printk("[%s]", realoutdev->name);
 820         }
 821         printk(" head:%p, raw:%p, data:%p\n", skb->head, skb_mac_header(skb),
 822                skb->data);
 823         dump_stack();
 824         return NF_ACCEPT;
 825 #endif
 826 }
 827
 828 /* IP/SABOTAGE *****************************************************/
 829 /* Don't hand locally destined packets to PF_INET(6)/PRE_ROUTING
 830  * for the second time. */
 831 static unsigned int ip_sabotage_in(unsigned int hook, struct sk_buff **pskb,
 832                                    const struct net_device *in,
 833                                    const struct net_device *out,
 834                                    int (*okfn)(struct sk_buff *))
 835 {
 836         if ((*pskb)->nf_bridge &&
 837             !((*pskb)->nf_bridge->mask & BRNF_NF_BRIDGE_PREROUTING)) {
 838                 return NF_STOP;
 839         }
 840
 841         return NF_ACCEPT;
 842 }
 843
 844 /* For br_nf_local_out we need (prio = NF_BR_PRI_FIRST), to insure that innocent
 845  * PF_BRIDGE/NF_BR_LOCAL_OUT functions don't get bridged traffic as input.
 846  * For br_nf_post_routing, we need (prio = NF_BR_PRI_LAST), because
 847  * ip_refrag() can return NF_STOLEN. */
 848 static struct nf_hook_ops br_nf_ops[] = {
 849         { .hook = br_nf_pre_routing,
 850           .owner = THIS_MODULE,
 851           .pf = PF_BRIDGE,
 852           .hooknum = NF_BR_PRE_ROUTING,
 853           .priority = NF_BR_PRI_BRNF, },
 854         { .hook = br_nf_local_in,
 855           .owner = THIS_MODULE,
 856           .pf = PF_BRIDGE,
 857           .hooknum = NF_BR_LOCAL_IN,
 858           .priority = NF_BR_PRI_BRNF, },
 859         { .hook = br_nf_forward_ip,
 860           .owner = THIS_MODULE,
 861           .pf = PF_BRIDGE,
 862           .hooknum = NF_BR_FORWARD,
 863           .priority = NF_BR_PRI_BRNF - 1, },
 864         { .hook = br_nf_forward_arp,
 865           .owner = THIS_MODULE,
 866           .pf = PF_BRIDGE,
 867           .hooknum = NF_BR_FORWARD,
 868           .priority = NF_BR_PRI_BRNF, },
 869         { .hook = br_nf_local_out,
 870           .owner = THIS_MODULE,
 871           .pf = PF_BRIDGE,
 872           .hooknum = NF_BR_LOCAL_OUT,
 873           .priority = NF_BR_PRI_FIRST, },
 874         { .hook = br_nf_post_routing,
 875           .owner = THIS_MODULE,
 876           .pf = PF_BRIDGE,
 877           .hooknum = NF_BR_POST_ROUTING,
 878           .priority = NF_BR_PRI_LAST, },
 879         { .hook = ip_sabotage_in,
 880           .owner = THIS_MODULE,
 881           .pf = PF_INET,
 882           .hooknum = NF_IP_PRE_ROUTING,
 883           .priority = NF_IP_PRI_FIRST, },
 884         { .hook = ip_sabotage_in,
 885           .owner = THIS_MODULE,
 886           .pf = PF_INET6,
 887           .hooknum = NF_IP6_PRE_ROUTING,
 888           .priority = NF_IP6_PRI_FIRST, },
 889 };
 890
 891 #ifdef CONFIG_SYSCTL
 892 static
 893 int brnf_sysctl_call_tables(ctl_table * ctl, int write, struct file *filp,
 894                             void __user * buffer, size_t * lenp, loff_t * ppos)
 895 {
 896         int ret;
 897
 898         ret = proc_dointvec(ctl, write, filp, buffer, lenp, ppos);
 899
 900         if (write && *(int *)(ctl->data))
 901                 *(int *)(ctl->data) = 1;
 902         return ret;
 903 }
 904
 905 static ctl_table brnf_table[] = {
 906         {
 907                 .procname       = "bridge-nf-call-arptables",
 908                 .data           = &brnf_call_arptables,
 909                 .maxlen         = sizeof(int),
 910                 .mode           = 0644,
 911                 .proc_handler   = &brnf_sysctl_call_tables,
 912         },
 913         {
 914                 .procname       = "bridge-nf-call-iptables",
 915                 .data           = &brnf_call_iptables,
 916                 .maxlen         = sizeof(int),
 917                 .mode           = 0644,
 918                 .proc_handler   = &brnf_sysctl_call_tables,
 919         },
 920         {
 921                 .procname       = "bridge-nf-call-ip6tables",
 922                 .data           = &brnf_call_ip6tables,
 923                 .maxlen         = sizeof(int),
 924                 .mode           = 0644,
 925                 .proc_handler   = &brnf_sysctl_call_tables,
 926         },
 927         {
 928                 .procname       = "bridge-nf-filter-vlan-tagged",
 929                 .data           = &brnf_filter_vlan_tagged,
 930                 .maxlen         = sizeof(int),
 931                 .mode           = 0644,
 932                 .proc_handler   = &brnf_sysctl_call_tables,
 933         },
 934         {
 935                 .procname       = "bridge-nf-filter-pppoe-tagged",
 936                 .data           = &brnf_filter_pppoe_tagged,
 937                 .maxlen         = sizeof(int),
 938                 .mode           = 0644,
 939                 .proc_handler   = &brnf_sysctl_call_tables,
 940         },
 941         { .ctl_name = 0 }
 942 };
 943
 944 static ctl_table brnf_bridge_table[] = {
 945         {
 946                 .ctl_name       = NET_BRIDGE,
 947                 .procname       = "bridge",
 948                 .mode           = 0555,
 949                 .child          = brnf_table,
 950         },
 951         { .ctl_name = 0 }
 952 };
 953
 954 static ctl_table brnf_net_table[] = {
 955         {
 956                 .ctl_name       = CTL_NET,
 957                 .procname       = "net",
 958                 .mode           = 0555,
 959                 .child          = brnf_bridge_table,
 960         },
 961         { .ctl_name = 0 }
 962 };
 963 #endif
 964
 965 int __init br_netfilter_init(void)
 966 {
 967         int ret;
 968
 969         ret = nf_register_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
 970         if (ret < 0)
 971                 return ret;
 972 #ifdef CONFIG_SYSCTL
 973         brnf_sysctl_header = register_sysctl_table(brnf_net_table);
 974         if (brnf_sysctl_header == NULL) {
 975                 printk(KERN_WARNING
 976                        "br_netfilter: can't register to sysctl.\n");
 977                 nf_unregister_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
 978                 return -ENOMEM;
 979         }
 980 #endif
 981         printk(KERN_NOTICE "Bridge firewalling registered\n");
 982         return 0;
 983 }
 984
 985 void br_netfilter_fini(void)
 986 {
 987         nf_unregister_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
 988 #ifdef CONFIG_SYSCTL
 989         unregister_sysctl_table(brnf_sysctl_header);
 990 #endif
 991 }