SAFE public projects git trees. - safe/jmp/linux-2.6/blob - 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/slab.h>
  27 #include <linux/ip.h>
  28 #include <linux/netdevice.h>
  29 #include <linux/skbuff.h>
  30 #include <linux/if_arp.h>
  31 #include <linux/if_ether.h>
  32 #include <linux/if_vlan.h>
  33 #include <linux/if_pppox.h>
  34 #include <linux/ppp_defs.h>
  35 #include <linux/netfilter_bridge.h>
  36 #include <linux/netfilter_ipv4.h>
  37 #include <linux/netfilter_ipv6.h>
  38 #include <linux/netfilter_arp.h>
  39 #include <linux/in_route.h>
  40 #include <linux/inetdevice.h>
  41
  42 #include <net/ip.h>
  43 #include <net/ipv6.h>
  44 #include <net/route.h>
  45
  46 #include <asm/uaccess.h>
  47 #include "br_private.h"
  48 #ifdef CONFIG_SYSCTL
  49 #include <linux/sysctl.h>
  50 #endif
  51
  52 #define skb_origaddr(skb)        (((struct bridge_skb_cb *) \
  53                                  (skb->nf_bridge->data))->daddr.ipv4)
  54 #define store_orig_dstaddr(skb)  (skb_origaddr(skb) = ip_hdr(skb)->daddr)
  55 #define dnat_took_place(skb)     (skb_origaddr(skb) != ip_hdr(skb)->daddr)
  56
  57 #ifdef CONFIG_SYSCTL
  58 static struct ctl_table_header *brnf_sysctl_header;
  59 static int brnf_call_iptables __read_mostly = 1;
  60 static int brnf_call_ip6tables __read_mostly = 1;
  61 static int brnf_call_arptables __read_mostly = 1;
  62 static int brnf_filter_vlan_tagged __read_mostly = 0;
  63 static int brnf_filter_pppoe_tagged __read_mostly = 0;
  64 #else
  65 #define brnf_filter_vlan_tagged 0
  66 #define brnf_filter_pppoe_tagged 0
  67 #endif
  68
  69 static inline __be16 vlan_proto(const struct sk_buff *skb)
  70 {
  71         return vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
  72 }
  73
  74 #define IS_VLAN_IP(skb) \
  75         (skb->protocol == htons(ETH_P_8021Q) && \
  76          vlan_proto(skb) == htons(ETH_P_IP) &&  \
  77          brnf_filter_vlan_tagged)
  78
  79 #define IS_VLAN_IPV6(skb) \
  80         (skb->protocol == htons(ETH_P_8021Q) && \
  81          vlan_proto(skb) == htons(ETH_P_IPV6) &&\
  82          brnf_filter_vlan_tagged)
  83
  84 #define IS_VLAN_ARP(skb) \
  85         (skb->protocol == htons(ETH_P_8021Q) && \
  86          vlan_proto(skb) == htons(ETH_P_ARP) && \
  87          brnf_filter_vlan_tagged)
  88
  89 static inline __be16 pppoe_proto(const struct sk_buff *skb)
  90 {
  91         return *((__be16 *)(skb_mac_header(skb) + ETH_HLEN +
  92                             sizeof(struct pppoe_hdr)));
  93 }
  94
  95 #define IS_PPPOE_IP(skb) \
  96         (skb->protocol == htons(ETH_P_PPP_SES) && \
  97          pppoe_proto(skb) == htons(PPP_IP) && \
  98          brnf_filter_pppoe_tagged)
  99
 100 #define IS_PPPOE_IPV6(skb) \
 101         (skb->protocol == htons(ETH_P_PPP_SES) && \
 102          pppoe_proto(skb) == htons(PPP_IPV6) && \
 103          brnf_filter_pppoe_tagged)
 104
 105 static void fake_update_pmtu(struct dst_entry *dst, u32 mtu)
 106 {
 107 }
 108
 109 static struct dst_ops fake_dst_ops = {
 110         .family =               AF_INET,
 111         .protocol =             cpu_to_be16(ETH_P_IP),
 112         .update_pmtu =          fake_update_pmtu,
 113         .entries =              ATOMIC_INIT(0),
 114 };
 115
 116 /*
 117  * Initialize bogus route table used to keep netfilter happy.
 118  * Currently, we fill in the PMTU entry because netfilter
 119  * refragmentation needs it, and the rt_flags entry because
 120  * ipt_REJECT needs it.  Future netfilter modules might
 121  * require us to fill additional fields.
 122  */
 123 void br_netfilter_rtable_init(struct net_bridge *br)
 124 {
 125         struct rtable *rt = &br->fake_rtable;
 126
 127         atomic_set(&rt->u.dst.__refcnt, 1);
 128         rt->u.dst.dev = br->dev;
 129         rt->u.dst.path = &rt->u.dst;
 130         rt->u.dst.metrics[RTAX_MTU - 1] = 1500;
 131         rt->u.dst.flags = DST_NOXFRM;
 132         rt->u.dst.ops = &fake_dst_ops;
 133 }
 134
 135 static inline struct rtable *bridge_parent_rtable(const struct net_device *dev)
 136 {
 137         struct net_bridge_port *port = rcu_dereference(dev->br_port);
 138
 139         return port ? &port->br->fake_rtable : NULL;
 140 }
 141
 142 static inline struct net_device *bridge_parent(const struct net_device *dev)
 143 {
 144         struct net_bridge_port *port = rcu_dereference(dev->br_port);
 145
 146         return port ? port->br->dev : NULL;
 147 }
 148
 149 static inline struct nf_bridge_info *nf_bridge_alloc(struct sk_buff *skb)
 150 {
 151         skb->nf_bridge = kzalloc(sizeof(struct nf_bridge_info), GFP_ATOMIC);
 152         if (likely(skb->nf_bridge))
 153                 atomic_set(&(skb->nf_bridge->use), 1);
 154
 155         return skb->nf_bridge;
 156 }
 157
 158 static inline struct nf_bridge_info *nf_bridge_unshare(struct sk_buff *skb)
 159 {
 160         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 161
 162         if (atomic_read(&nf_bridge->use) > 1) {
 163                 struct nf_bridge_info *tmp = nf_bridge_alloc(skb);
 164
 165                 if (tmp) {
 166                         memcpy(tmp, nf_bridge, sizeof(struct nf_bridge_info));
 167                         atomic_set(&tmp->use, 1);
 168                         nf_bridge_put(nf_bridge);
 169                 }
 170                 nf_bridge = tmp;
 171         }
 172         return nf_bridge;
 173 }
 174
 175 static inline void nf_bridge_push_encap_header(struct sk_buff *skb)
 176 {
 177         unsigned int len = nf_bridge_encap_header_len(skb);
 178
 179         skb_push(skb, len);
 180         skb->network_header -= len;
 181 }
 182
 183 static inline void nf_bridge_pull_encap_header(struct sk_buff *skb)
 184 {
 185         unsigned int len = nf_bridge_encap_header_len(skb);
 186
 187         skb_pull(skb, len);
 188         skb->network_header += len;
 189 }
 190
 191 static inline void nf_bridge_pull_encap_header_rcsum(struct sk_buff *skb)
 192 {
 193         unsigned int len = nf_bridge_encap_header_len(skb);
 194
 195         skb_pull_rcsum(skb, len);
 196         skb->network_header += len;
 197 }
 198
 199 static inline void nf_bridge_save_header(struct sk_buff *skb)
 200 {
 201         int header_size = ETH_HLEN + nf_bridge_encap_header_len(skb);
 202
 203         skb_copy_from_linear_data_offset(skb, -header_size,
 204                                          skb->nf_bridge->data, header_size);
 205 }
 206
 207 /*
 208  * When forwarding bridge frames, we save a copy of the original
 209  * header before processing.
 210  */
 211 int nf_bridge_copy_header(struct sk_buff *skb)
 212 {
 213         int err;
 214         int header_size = ETH_HLEN + nf_bridge_encap_header_len(skb);
 215
 216         err = skb_cow_head(skb, header_size);
 217         if (err)
 218                 return err;
 219
 220         skb_copy_to_linear_data_offset(skb, -header_size,
 221                                        skb->nf_bridge->data, header_size);
 222         __skb_push(skb, nf_bridge_encap_header_len(skb));
 223         return 0;
 224 }
 225
 226 /* PF_BRIDGE/PRE_ROUTING *********************************************/
 227 /* Undo the changes made for ip6tables PREROUTING and continue the
 228  * bridge PRE_ROUTING hook. */
 229 static int br_nf_pre_routing_finish_ipv6(struct sk_buff *skb)
 230 {
 231         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 232         struct rtable *rt;
 233
 234         if (nf_bridge->mask & BRNF_PKT_TYPE) {
 235                 skb->pkt_type = PACKET_OTHERHOST;
 236                 nf_bridge->mask ^= BRNF_PKT_TYPE;
 237         }
 238         nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
 239
 240         rt = bridge_parent_rtable(nf_bridge->physindev);
 241         if (!rt) {
 242                 kfree_skb(skb);
 243                 return 0;
 244         }
 245         dst_hold(&rt->u.dst);
 246         skb_dst_set(skb, &rt->u.dst);
 247
 248         skb->dev = nf_bridge->physindev;
 249         nf_bridge_push_encap_header(skb);
 250         NF_HOOK_THRESH(PF_BRIDGE, NF_BR_PRE_ROUTING, skb, skb->dev, NULL,
 251                        br_handle_frame_finish, 1);
 252
 253         return 0;
 254 }
 255
 256 static void __br_dnat_complain(void)
 257 {
 258         static unsigned long last_complaint;
 259
 260         if (jiffies - last_complaint >= 5 * HZ) {
 261                 printk(KERN_WARNING "Performing cross-bridge DNAT requires IP "
 262                        "forwarding to be enabled\n");
 263                 last_complaint = jiffies;
 264         }
 265 }
 266
 267 /* This requires some explaining. If DNAT has taken place,
 268  * we will need to fix up the destination Ethernet address,
 269  * and this is a tricky process.
 270  *
 271  * There are two cases to consider:
 272  * 1. The packet was DNAT'ed to a device in the same bridge
 273  *    port group as it was received on. We can still bridge
 274  *    the packet.
 275  * 2. The packet was DNAT'ed to a different device, either
 276  *    a non-bridged device or another bridge port group.
 277  *    The packet will need to be routed.
 278  *
 279  * The correct way of distinguishing between these two cases is to
 280  * call ip_route_input() and to look at skb->dst->dev, which is
 281  * changed to the destination device if ip_route_input() succeeds.
 282  *
 283  * Let us first consider the case that ip_route_input() succeeds:
 284  *
 285  * If skb->dst->dev equals the logical bridge device the packet
 286  * came in on, we can consider this bridging. The packet is passed
 287  * through the neighbour output function to build a new destination
 288  * MAC address, which will make the packet enter br_nf_local_out()
 289  * not much later. In that function it is assured that the iptables
 290  * FORWARD chain is traversed for the packet.
 291  *
 292  * Otherwise, the packet is considered to be routed and we just
 293  * change the destination MAC address so that the packet will
 294  * later be passed up to the IP stack to be routed. For a redirected
 295  * packet, ip_route_input() will give back the localhost as output device,
 296  * which differs from the bridge device.
 297  *
 298  * Let us now consider the case that ip_route_input() fails:
 299  *
 300  * This can be because the destination address is martian, in which case
 301  * the packet will be dropped.
 302  * After a "echo '0' > /proc/sys/net/ipv4/ip_forward" ip_route_input()
 303  * will fail, while __ip_route_output_key() will return success. The source
 304  * address for __ip_route_output_key() is set to zero, so __ip_route_output_key
 305  * thinks we're handling a locally generated packet and won't care
 306  * if IP forwarding is allowed. We send a warning message to the users's
 307  * log telling her to put IP forwarding on.
 308  *
 309  * ip_route_input() will also fail if there is no route available.
 310  * In that case we just drop the packet.
 311  *
 312  * --Lennert, 20020411
 313  * --Bart, 20020416 (updated)
 314  * --Bart, 20021007 (updated)
 315  * --Bart, 20062711 (updated) */
 316 static int br_nf_pre_routing_finish_bridge(struct sk_buff *skb)
 317 {
 318         if (skb->pkt_type == PACKET_OTHERHOST) {
 319                 skb->pkt_type = PACKET_HOST;
 320                 skb->nf_bridge->mask |= BRNF_PKT_TYPE;
 321         }
 322         skb->nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
 323
 324         skb->dev = bridge_parent(skb->dev);
 325         if (skb->dev) {
 326                 struct dst_entry *dst = skb_dst(skb);
 327
 328                 nf_bridge_pull_encap_header(skb);
 329
 330                 if (dst->hh)
 331                         return neigh_hh_output(dst->hh, skb);
 332                 else if (dst->neighbour)
 333                         return dst->neighbour->output(skb);
 334         }
 335         kfree_skb(skb);
 336         return 0;
 337 }
 338
 339 static int br_nf_pre_routing_finish(struct sk_buff *skb)
 340 {
 341         struct net_device *dev = skb->dev;
 342         struct iphdr *iph = ip_hdr(skb);
 343         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 344         struct rtable *rt;
 345         int err;
 346
 347         if (nf_bridge->mask & BRNF_PKT_TYPE) {
 348                 skb->pkt_type = PACKET_OTHERHOST;
 349                 nf_bridge->mask ^= BRNF_PKT_TYPE;
 350         }
 351         nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
 352         if (dnat_took_place(skb)) {
 353                 if ((err = ip_route_input(skb, iph->daddr, iph->saddr, iph->tos, dev))) {
 354                         struct flowi fl = {
 355                                 .nl_u = {
 356                                         .ip4_u = {
 357                                                  .daddr = iph->daddr,
 358                                                  .saddr = 0,
 359                                                  .tos = RT_TOS(iph->tos) },
 360                                 },
 361                                 .proto = 0,
 362                         };
 363                         struct in_device *in_dev = __in_dev_get_rcu(dev);
 364
 365                         /* If err equals -EHOSTUNREACH the error is due to a
 366                          * martian destination or due to the fact that
 367                          * forwarding is disabled. For most martian packets,
 368                          * ip_route_output_key() will fail. It won't fail for 2 types of
 369                          * martian destinations: loopback destinations and destination
 370                          * 0.0.0.0. In both cases the packet will be dropped because the
 371                          * destination is the loopback device and not the bridge. */
 372                         if (err != -EHOSTUNREACH || !in_dev || IN_DEV_FORWARD(in_dev))
 373                                 goto free_skb;
 374
 375                         if (!ip_route_output_key(dev_net(dev), &rt, &fl)) {
 376                                 /* - Bridged-and-DNAT'ed traffic doesn't
 377                                  *   require ip_forwarding. */
 378                                 if (((struct dst_entry *)rt)->dev == dev) {
 379                                         skb_dst_set(skb, (struct dst_entry *)rt);
 380                                         goto bridged_dnat;
 381                                 }
 382                                 /* we are sure that forwarding is disabled, so printing
 383                                  * this message is no problem. Note that the packet could
 384                                  * still have a martian destination address, in which case
 385                                  * the packet could be dropped even if forwarding were enabled */
 386                                 __br_dnat_complain();
 387                                 dst_release((struct dst_entry *)rt);
 388                         }
 389 free_skb:
 390                         kfree_skb(skb);
 391                         return 0;
 392                 } else {
 393                         if (skb_dst(skb)->dev == dev) {
 394 bridged_dnat:
 395                                 /* Tell br_nf_local_out this is a
 396                                  * bridged frame */
 397                                 nf_bridge->mask |= BRNF_BRIDGED_DNAT;
 398                                 skb->dev = nf_bridge->physindev;
 399                                 nf_bridge_push_encap_header(skb);
 400                                 NF_HOOK_THRESH(PF_BRIDGE, NF_BR_PRE_ROUTING,
 401                                                skb, skb->dev, NULL,
 402                                                br_nf_pre_routing_finish_bridge,
 403                                                1);
 404                                 return 0;
 405                         }
 406                         memcpy(eth_hdr(skb)->h_dest, dev->dev_addr, ETH_ALEN);
 407                         skb->pkt_type = PACKET_HOST;
 408                 }
 409         } else {
 410                 rt = bridge_parent_rtable(nf_bridge->physindev);
 411                 if (!rt) {
 412                         kfree_skb(skb);
 413                         return 0;
 414                 }
 415                 dst_hold(&rt->u.dst);
 416                 skb_dst_set(skb, &rt->u.dst);
 417         }
 418
 419         skb->dev = nf_bridge->physindev;
 420         nf_bridge_push_encap_header(skb);
 421         NF_HOOK_THRESH(PF_BRIDGE, NF_BR_PRE_ROUTING, skb, skb->dev, NULL,
 422                        br_handle_frame_finish, 1);
 423
 424         return 0;
 425 }
 426
 427 /* Some common code for IPv4/IPv6 */
 428 static struct net_device *setup_pre_routing(struct sk_buff *skb)
 429 {
 430         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 431
 432         if (skb->pkt_type == PACKET_OTHERHOST) {
 433                 skb->pkt_type = PACKET_HOST;
 434                 nf_bridge->mask |= BRNF_PKT_TYPE;
 435         }
 436
 437         nf_bridge->mask |= BRNF_NF_BRIDGE_PREROUTING;
 438         nf_bridge->physindev = skb->dev;
 439         skb->dev = bridge_parent(skb->dev);
 440
 441         return skb->dev;
 442 }
 443
 444 /* We only check the length. A bridge shouldn't do any hop-by-hop stuff anyway */
 445 static int check_hbh_len(struct sk_buff *skb)
 446 {
 447         unsigned char *raw = (u8 *)(ipv6_hdr(skb) + 1);
 448         u32 pkt_len;
 449         const unsigned char *nh = skb_network_header(skb);
 450         int off = raw - nh;
 451         int len = (raw[1] + 1) << 3;
 452
 453         if ((raw + len) - skb->data > skb_headlen(skb))
 454                 goto bad;
 455
 456         off += 2;
 457         len -= 2;
 458
 459         while (len > 0) {
 460                 int optlen = nh[off + 1] + 2;
 461
 462                 switch (nh[off]) {
 463                 case IPV6_TLV_PAD0:
 464                         optlen = 1;
 465                         break;
 466
 467                 case IPV6_TLV_PADN:
 468                         break;
 469
 470                 case IPV6_TLV_JUMBO:
 471                         if (nh[off + 1] != 4 || (off & 3) != 2)
 472                                 goto bad;
 473                         pkt_len = ntohl(*(__be32 *) (nh + off + 2));
 474                         if (pkt_len <= IPV6_MAXPLEN ||
 475                             ipv6_hdr(skb)->payload_len)
 476                                 goto bad;
 477                         if (pkt_len > skb->len - sizeof(struct ipv6hdr))
 478                                 goto bad;
 479                         if (pskb_trim_rcsum(skb,
 480                                             pkt_len + sizeof(struct ipv6hdr)))
 481                                 goto bad;
 482                         nh = skb_network_header(skb);
 483                         break;
 484                 default:
 485                         if (optlen > len)
 486                                 goto bad;
 487                         break;
 488                 }
 489                 off += optlen;
 490                 len -= optlen;
 491         }
 492         if (len == 0)
 493                 return 0;
 494 bad:
 495         return -1;
 496
 497 }
 498
 499 /* Replicate the checks that IPv6 does on packet reception and pass the packet
 500  * to ip6tables, which doesn't support NAT, so things are fairly simple. */
 501 static unsigned int br_nf_pre_routing_ipv6(unsigned int hook,
 502                                            struct sk_buff *skb,
 503                                            const struct net_device *in,
 504                                            const struct net_device *out,
 505                                            int (*okfn)(struct sk_buff *))
 506 {
 507         struct ipv6hdr *hdr;
 508         u32 pkt_len;
 509
 510         if (skb->len < sizeof(struct ipv6hdr))
 511                 goto inhdr_error;
 512
 513         if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
 514                 goto inhdr_error;
 515
 516         hdr = ipv6_hdr(skb);
 517
 518         if (hdr->version != 6)
 519                 goto inhdr_error;
 520
 521         pkt_len = ntohs(hdr->payload_len);
 522
 523         if (pkt_len || hdr->nexthdr != NEXTHDR_HOP) {
 524                 if (pkt_len + sizeof(struct ipv6hdr) > skb->len)
 525                         goto inhdr_error;
 526                 if (pskb_trim_rcsum(skb, pkt_len + sizeof(struct ipv6hdr)))
 527                         goto inhdr_error;
 528         }
 529         if (hdr->nexthdr == NEXTHDR_HOP && check_hbh_len(skb))
 530                 goto inhdr_error;
 531
 532         nf_bridge_put(skb->nf_bridge);
 533         if (!nf_bridge_alloc(skb))
 534                 return NF_DROP;
 535         if (!setup_pre_routing(skb))
 536                 return NF_DROP;
 537
 538         NF_HOOK(PF_INET6, NF_INET_PRE_ROUTING, skb, skb->dev, NULL,
 539                 br_nf_pre_routing_finish_ipv6);
 540
 541         return NF_STOLEN;
 542
 543 inhdr_error:
 544         return NF_DROP;
 545 }
 546
 547 /* Direct IPv6 traffic to br_nf_pre_routing_ipv6.
 548  * Replicate the checks that IPv4 does on packet reception.
 549  * Set skb->dev to the bridge device (i.e. parent of the
 550  * receiving device) to make netfilter happy, the REDIRECT
 551  * target in particular.  Save the original destination IP
 552  * address to be able to detect DNAT afterwards. */
 553 static unsigned int br_nf_pre_routing(unsigned int hook, struct sk_buff *skb,
 554                                       const struct net_device *in,
 555                                       const struct net_device *out,
 556                                       int (*okfn)(struct sk_buff *))
 557 {
 558         struct iphdr *iph;
 559         __u32 len = nf_bridge_encap_header_len(skb);
 560
 561         if (unlikely(!pskb_may_pull(skb, len)))
 562                 goto out;
 563
 564         if (skb->protocol == htons(ETH_P_IPV6) || IS_VLAN_IPV6(skb) ||
 565             IS_PPPOE_IPV6(skb)) {
 566 #ifdef CONFIG_SYSCTL
 567                 if (!brnf_call_ip6tables)
 568                         return NF_ACCEPT;
 569 #endif
 570                 nf_bridge_pull_encap_header_rcsum(skb);
 571                 return br_nf_pre_routing_ipv6(hook, skb, in, out, okfn);
 572         }
 573 #ifdef CONFIG_SYSCTL
 574         if (!brnf_call_iptables)
 575                 return NF_ACCEPT;
 576 #endif
 577
 578         if (skb->protocol != htons(ETH_P_IP) && !IS_VLAN_IP(skb) &&
 579             !IS_PPPOE_IP(skb))
 580                 return NF_ACCEPT;
 581
 582         nf_bridge_pull_encap_header_rcsum(skb);
 583
 584         if (!pskb_may_pull(skb, sizeof(struct iphdr)))
 585                 goto inhdr_error;
 586
 587         iph = ip_hdr(skb);
 588         if (iph->ihl < 5 || iph->version != 4)
 589                 goto inhdr_error;
 590
 591         if (!pskb_may_pull(skb, 4 * iph->ihl))
 592                 goto inhdr_error;
 593
 594         iph = ip_hdr(skb);
 595         if (ip_fast_csum((__u8 *) iph, iph->ihl) != 0)
 596                 goto inhdr_error;
 597
 598         len = ntohs(iph->tot_len);
 599         if (skb->len < len || len < 4 * iph->ihl)
 600                 goto inhdr_error;
 601
 602         pskb_trim_rcsum(skb, len);
 603
 604         nf_bridge_put(skb->nf_bridge);
 605         if (!nf_bridge_alloc(skb))
 606                 return NF_DROP;
 607         if (!setup_pre_routing(skb))
 608                 return NF_DROP;
 609         store_orig_dstaddr(skb);
 610
 611         NF_HOOK(PF_INET, NF_INET_PRE_ROUTING, skb, skb->dev, NULL,
 612                 br_nf_pre_routing_finish);
 613
 614         return NF_STOLEN;
 615
 616 inhdr_error:
 617 //      IP_INC_STATS_BH(IpInHdrErrors);
 618 out:
 619         return NF_DROP;
 620 }
 621
 622
 623 /* PF_BRIDGE/LOCAL_IN ************************************************/
 624 /* The packet is locally destined, which requires a real
 625  * dst_entry, so detach the fake one.  On the way up, the
 626  * packet would pass through PRE_ROUTING again (which already
 627  * took place when the packet entered the bridge), but we
 628  * register an IPv4 PRE_ROUTING 'sabotage' hook that will
 629  * prevent this from happening. */
 630 static unsigned int br_nf_local_in(unsigned int hook, struct sk_buff *skb,
 631                                    const struct net_device *in,
 632                                    const struct net_device *out,
 633                                    int (*okfn)(struct sk_buff *))
 634 {
 635         struct rtable *rt = skb_rtable(skb);
 636
 637         if (rt && rt == bridge_parent_rtable(in))
 638                 skb_dst_drop(skb);
 639
 640         return NF_ACCEPT;
 641 }
 642
 643 /* PF_BRIDGE/FORWARD *************************************************/
 644 static int br_nf_forward_finish(struct sk_buff *skb)
 645 {
 646         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 647         struct net_device *in;
 648
 649         if (skb->protocol != htons(ETH_P_ARP) && !IS_VLAN_ARP(skb)) {
 650                 in = nf_bridge->physindev;
 651                 if (nf_bridge->mask & BRNF_PKT_TYPE) {
 652                         skb->pkt_type = PACKET_OTHERHOST;
 653                         nf_bridge->mask ^= BRNF_PKT_TYPE;
 654                 }
 655         } else {
 656                 in = *((struct net_device **)(skb->cb));
 657         }
 658         nf_bridge_push_encap_header(skb);
 659         NF_HOOK_THRESH(PF_BRIDGE, NF_BR_FORWARD, skb, in,
 660                        skb->dev, br_forward_finish, 1);
 661         return 0;
 662 }
 663
 664 /* This is the 'purely bridged' case.  For IP, we pass the packet to
 665  * netfilter with indev and outdev set to the bridge device,
 666  * but we are still able to filter on the 'real' indev/outdev
 667  * because of the physdev module. For ARP, indev and outdev are the
 668  * bridge ports. */
 669 static unsigned int br_nf_forward_ip(unsigned int hook, struct sk_buff *skb,
 670                                      const struct net_device *in,
 671                                      const struct net_device *out,
 672                                      int (*okfn)(struct sk_buff *))
 673 {
 674         struct nf_bridge_info *nf_bridge;
 675         struct net_device *parent;
 676         u_int8_t pf;
 677
 678         if (!skb->nf_bridge)
 679                 return NF_ACCEPT;
 680
 681         /* Need exclusive nf_bridge_info since we might have multiple
 682          * different physoutdevs. */
 683         if (!nf_bridge_unshare(skb))
 684                 return NF_DROP;
 685
 686         parent = bridge_parent(out);
 687         if (!parent)
 688                 return NF_DROP;
 689
 690         if (skb->protocol == htons(ETH_P_IP) || IS_VLAN_IP(skb) ||
 691             IS_PPPOE_IP(skb))
 692                 pf = PF_INET;
 693         else if (skb->protocol == htons(ETH_P_IPV6) || IS_VLAN_IPV6(skb) ||
 694                  IS_PPPOE_IPV6(skb))
 695                 pf = PF_INET6;
 696         else
 697                 return NF_ACCEPT;
 698
 699         nf_bridge_pull_encap_header(skb);
 700
 701         nf_bridge = skb->nf_bridge;
 702         if (skb->pkt_type == PACKET_OTHERHOST) {
 703                 skb->pkt_type = PACKET_HOST;
 704                 nf_bridge->mask |= BRNF_PKT_TYPE;
 705         }
 706
 707         /* The physdev module checks on this */
 708         nf_bridge->mask |= BRNF_BRIDGED;
 709         nf_bridge->physoutdev = skb->dev;
 710
 711         NF_HOOK(pf, NF_INET_FORWARD, skb, bridge_parent(in), parent,
 712                 br_nf_forward_finish);
 713
 714         return NF_STOLEN;
 715 }
 716
 717 static unsigned int br_nf_forward_arp(unsigned int hook, struct sk_buff *skb,
 718                                       const struct net_device *in,
 719                                       const struct net_device *out,
 720                                       int (*okfn)(struct sk_buff *))
 721 {
 722         struct net_device **d = (struct net_device **)(skb->cb);
 723
 724 #ifdef CONFIG_SYSCTL
 725         if (!brnf_call_arptables)
 726                 return NF_ACCEPT;
 727 #endif
 728
 729         if (skb->protocol != htons(ETH_P_ARP)) {
 730                 if (!IS_VLAN_ARP(skb))
 731                         return NF_ACCEPT;
 732                 nf_bridge_pull_encap_header(skb);
 733         }
 734
 735         if (arp_hdr(skb)->ar_pln != 4) {
 736                 if (IS_VLAN_ARP(skb))
 737                         nf_bridge_push_encap_header(skb);
 738                 return NF_ACCEPT;
 739         }
 740         *d = (struct net_device *)in;
 741         NF_HOOK(NFPROTO_ARP, NF_ARP_FORWARD, skb, (struct net_device *)in,
 742                 (struct net_device *)out, br_nf_forward_finish);
 743
 744         return NF_STOLEN;
 745 }
 746
 747 /* PF_BRIDGE/LOCAL_OUT ***********************************************
 748  *
 749  * This function sees both locally originated IP packets and forwarded
 750  * IP packets (in both cases the destination device is a bridge
 751  * device). It also sees bridged-and-DNAT'ed packets.
 752  *
 753  * If (nf_bridge->mask & BRNF_BRIDGED_DNAT) then the packet is bridged
 754  * and we fake the PF_BRIDGE/FORWARD hook. The function br_nf_forward()
 755  * will then fake the PF_INET/FORWARD hook. br_nf_local_out() has priority
 756  * NF_BR_PRI_FIRST, so no relevant PF_BRIDGE/INPUT functions have been nor
 757  * will be executed.
 758  */
 759 static unsigned int br_nf_local_out(unsigned int hook, struct sk_buff *skb,
 760                                     const struct net_device *in,
 761                                     const struct net_device *out,
 762                                     int (*okfn)(struct sk_buff *))
 763 {
 764         struct net_device *realindev;
 765         struct nf_bridge_info *nf_bridge;
 766
 767         if (!skb->nf_bridge)
 768                 return NF_ACCEPT;
 769
 770         /* Need exclusive nf_bridge_info since we might have multiple
 771          * different physoutdevs. */
 772         if (!nf_bridge_unshare(skb))
 773                 return NF_DROP;
 774
 775         nf_bridge = skb->nf_bridge;
 776         if (!(nf_bridge->mask & BRNF_BRIDGED_DNAT))
 777                 return NF_ACCEPT;
 778
 779         /* Bridged, take PF_BRIDGE/FORWARD.
 780          * (see big note in front of br_nf_pre_routing_finish) */
 781         nf_bridge->physoutdev = skb->dev;
 782         realindev = nf_bridge->physindev;
 783
 784         if (nf_bridge->mask & BRNF_PKT_TYPE) {
 785                 skb->pkt_type = PACKET_OTHERHOST;
 786                 nf_bridge->mask ^= BRNF_PKT_TYPE;
 787         }
 788         nf_bridge_push_encap_header(skb);
 789
 790         NF_HOOK(PF_BRIDGE, NF_BR_FORWARD, skb, realindev, skb->dev,
 791                 br_forward_finish);
 792         return NF_STOLEN;
 793 }
 794
 795 #if defined(CONFIG_NF_CONNTRACK_IPV4) || defined(CONFIG_NF_CONNTRACK_IPV4_MODULE)
 796 static int br_nf_dev_queue_xmit(struct sk_buff *skb)
 797 {
 798         if (skb->nfct != NULL &&
 799             (skb->protocol == htons(ETH_P_IP) || IS_VLAN_IP(skb)) &&
 800             skb->len > skb->dev->mtu &&
 801             !skb_is_gso(skb))
 802                 return ip_fragment(skb, br_dev_queue_push_xmit);
 803         else
 804                 return br_dev_queue_push_xmit(skb);
 805 }
 806 #else
 807 static int br_nf_dev_queue_xmit(struct sk_buff *skb)
 808 {
 809         return br_dev_queue_push_xmit(skb);
 810 }
 811 #endif
 812
 813 /* PF_BRIDGE/POST_ROUTING ********************************************/
 814 static unsigned int br_nf_post_routing(unsigned int hook, struct sk_buff *skb,
 815                                        const struct net_device *in,
 816                                        const struct net_device *out,
 817                                        int (*okfn)(struct sk_buff *))
 818 {
 819         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 820         struct net_device *realoutdev = bridge_parent(skb->dev);
 821         u_int8_t pf;
 822
 823 #ifdef CONFIG_NETFILTER_DEBUG
 824         /* Be very paranoid. This probably won't happen anymore, but let's
 825          * keep the check just to be sure... */
 826         if (skb_mac_header(skb) < skb->head ||
 827             skb_mac_header(skb) + ETH_HLEN > skb->data) {
 828                 printk(KERN_CRIT "br_netfilter: Argh!! br_nf_post_routing: "
 829                        "bad mac.raw pointer.\n");
 830                 goto print_error;
 831         }
 832 #endif
 833
 834         if (!nf_bridge)
 835                 return NF_ACCEPT;
 836
 837         if (!(nf_bridge->mask & (BRNF_BRIDGED | BRNF_BRIDGED_DNAT)))
 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 #ifdef CONFIG_NETFILTER_DEBUG
 853         if (skb_dst(skb) == NULL) {
 854                 printk(KERN_INFO "br_netfilter post_routing: skb->dst == NULL\n");
 855                 goto print_error;
 856         }
 857 #endif
 858
 859         /* We assume any code from br_dev_queue_push_xmit onwards doesn't care
 860          * about the value of skb->pkt_type. */
 861         if (skb->pkt_type == PACKET_OTHERHOST) {
 862                 skb->pkt_type = PACKET_HOST;
 863                 nf_bridge->mask |= BRNF_PKT_TYPE;
 864         }
 865
 866         nf_bridge_pull_encap_header(skb);
 867         nf_bridge_save_header(skb);
 868
 869         NF_HOOK(pf, NF_INET_POST_ROUTING, skb, NULL, realoutdev,
 870                 br_nf_dev_queue_xmit);
 871
 872         return NF_STOLEN;
 873
 874 #ifdef CONFIG_NETFILTER_DEBUG
 875 print_error:
 876         if (skb->dev != NULL) {
 877                 printk("[%s]", skb->dev->name);
 878                 if (realoutdev)
 879                         printk("[%s]", realoutdev->name);
 880         }
 881         printk(" head:%p, raw:%p, data:%p\n", skb->head, skb_mac_header(skb),
 882                skb->data);
 883         dump_stack();
 884         return NF_ACCEPT;
 885 #endif
 886 }
 887
 888 /* IP/SABOTAGE *****************************************************/
 889 /* Don't hand locally destined packets to PF_INET(6)/PRE_ROUTING
 890  * for the second time. */
 891 static unsigned int ip_sabotage_in(unsigned int hook, struct sk_buff *skb,
 892                                    const struct net_device *in,
 893                                    const struct net_device *out,
 894                                    int (*okfn)(struct sk_buff *))
 895 {
 896         if (skb->nf_bridge &&
 897             !(skb->nf_bridge->mask & BRNF_NF_BRIDGE_PREROUTING)) {
 898                 return NF_STOP;
 899         }
 900
 901         return NF_ACCEPT;
 902 }
 903
 904 /* For br_nf_local_out we need (prio = NF_BR_PRI_FIRST), to insure that innocent
 905  * PF_BRIDGE/NF_BR_LOCAL_OUT functions don't get bridged traffic as input.
 906  * For br_nf_post_routing, we need (prio = NF_BR_PRI_LAST), because
 907  * ip_refrag() can return NF_STOLEN. */
 908 static struct nf_hook_ops br_nf_ops[] __read_mostly = {
 909         {
 910                 .hook = br_nf_pre_routing,
 911                 .owner = THIS_MODULE,
 912                 .pf = PF_BRIDGE,
 913                 .hooknum = NF_BR_PRE_ROUTING,
 914                 .priority = NF_BR_PRI_BRNF,
 915         },
 916         {
 917                 .hook = br_nf_local_in,
 918                 .owner = THIS_MODULE,
 919                 .pf = PF_BRIDGE,
 920                 .hooknum = NF_BR_LOCAL_IN,
 921                 .priority = NF_BR_PRI_BRNF,
 922         },
 923         {
 924                 .hook = br_nf_forward_ip,
 925                 .owner = THIS_MODULE,
 926                 .pf = PF_BRIDGE,
 927                 .hooknum = NF_BR_FORWARD,
 928                 .priority = NF_BR_PRI_BRNF - 1,
 929         },
 930         {
 931                 .hook = br_nf_forward_arp,
 932                 .owner = THIS_MODULE,
 933                 .pf = PF_BRIDGE,
 934                 .hooknum = NF_BR_FORWARD,
 935                 .priority = NF_BR_PRI_BRNF,
 936         },
 937         {
 938                 .hook = br_nf_local_out,
 939                 .owner = THIS_MODULE,
 940                 .pf = PF_BRIDGE,
 941                 .hooknum = NF_BR_LOCAL_OUT,
 942                 .priority = NF_BR_PRI_FIRST,
 943         },
 944         {
 945                 .hook = br_nf_post_routing,
 946                 .owner = THIS_MODULE,
 947                 .pf = PF_BRIDGE,
 948                 .hooknum = NF_BR_POST_ROUTING,
 949                 .priority = NF_BR_PRI_LAST,
 950         },
 951         {
 952                 .hook = ip_sabotage_in,
 953                 .owner = THIS_MODULE,
 954                 .pf = PF_INET,
 955                 .hooknum = NF_INET_PRE_ROUTING,
 956                 .priority = NF_IP_PRI_FIRST,
 957         },
 958         {
 959                 .hook = ip_sabotage_in,
 960                 .owner = THIS_MODULE,
 961                 .pf = PF_INET6,
 962                 .hooknum = NF_INET_PRE_ROUTING,
 963                 .priority = NF_IP6_PRI_FIRST,
 964         },
 965 };
 966
 967 #ifdef CONFIG_SYSCTL
 968 static
 969 int brnf_sysctl_call_tables(ctl_table * ctl, int write,
 970                             void __user * buffer, size_t * lenp, loff_t * ppos)
 971 {
 972         int ret;
 973
 974         ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
 975
 976         if (write && *(int *)(ctl->data))
 977                 *(int *)(ctl->data) = 1;
 978         return ret;
 979 }
 980
 981 static ctl_table brnf_table[] = {
 982         {
 983                 .procname       = "bridge-nf-call-arptables",
 984                 .data           = &brnf_call_arptables,
 985                 .maxlen         = sizeof(int),
 986                 .mode           = 0644,
 987                 .proc_handler   = brnf_sysctl_call_tables,
 988         },
 989         {
 990                 .procname       = "bridge-nf-call-iptables",
 991                 .data           = &brnf_call_iptables,
 992                 .maxlen         = sizeof(int),
 993                 .mode           = 0644,
 994                 .proc_handler   = brnf_sysctl_call_tables,
 995         },
 996         {
 997                 .procname       = "bridge-nf-call-ip6tables",
 998                 .data           = &brnf_call_ip6tables,
 999                 .maxlen         = sizeof(int),
1000                 .mode           = 0644,
1001                 .proc_handler   = brnf_sysctl_call_tables,
1002         },
1003         {
1004                 .procname       = "bridge-nf-filter-vlan-tagged",
1005                 .data           = &brnf_filter_vlan_tagged,
1006                 .maxlen         = sizeof(int),
1007                 .mode           = 0644,
1008                 .proc_handler   = brnf_sysctl_call_tables,
1009         },
1010         {
1011                 .procname       = "bridge-nf-filter-pppoe-tagged",
1012                 .data           = &brnf_filter_pppoe_tagged,
1013                 .maxlen         = sizeof(int),
1014                 .mode           = 0644,
1015                 .proc_handler   = brnf_sysctl_call_tables,
1016         },
1017         { }
1018 };
1019
1020 static struct ctl_path brnf_path[] = {
1021         { .procname = "net", },
1022         { .procname = "bridge", },
1023         { }
1024 };
1025 #endif
1026
1027 int __init br_netfilter_init(void)
1028 {
1029         int ret;
1030
1031         ret = nf_register_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
1032         if (ret < 0)
1033                 return ret;
1034 #ifdef CONFIG_SYSCTL
1035         brnf_sysctl_header = register_sysctl_paths(brnf_path, brnf_table);
1036         if (brnf_sysctl_header == NULL) {
1037                 printk(KERN_WARNING
1038                        "br_netfilter: can't register to sysctl.\n");
1039                 nf_unregister_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
1040                 return -ENOMEM;
1041         }
1042 #endif
1043         printk(KERN_NOTICE "Bridge firewalling registered\n");
1044         return 0;
1045 }
1046
1047 void br_netfilter_fini(void)
1048 {
1049         nf_unregister_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
1050 #ifdef CONFIG_SYSCTL
1051         unregister_sysctl_table(brnf_sysctl_header);
1052 #endif
1053 }