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/netfilter_bridge.h>
  33 #include <linux/netfilter_ipv4.h>
  34 #include <linux/netfilter_ipv6.h>
  35 #include <linux/netfilter_arp.h>
  36 #include <linux/in_route.h>
  37 #include <linux/inetdevice.h>
  38
  39 #include <net/ip.h>
  40 #include <net/ipv6.h>
  41 #include <net/route.h>
  42
  43 #include <asm/uaccess.h>
  44 #include "br_private.h"
  45 #ifdef CONFIG_SYSCTL
  46 #include <linux/sysctl.h>
  47 #endif
  48
  49 #define skb_origaddr(skb)        (((struct bridge_skb_cb *) \
  50                                  (skb->nf_bridge->data))->daddr.ipv4)
  51 #define store_orig_dstaddr(skb)  (skb_origaddr(skb) = (skb)->nh.iph->daddr)
  52 #define dnat_took_place(skb)     (skb_origaddr(skb) != (skb)->nh.iph->daddr)
  53
  54 #ifdef CONFIG_SYSCTL
  55 static struct ctl_table_header *brnf_sysctl_header;
  56 static int brnf_call_iptables __read_mostly = 1;
  57 static int brnf_call_ip6tables __read_mostly = 1;
  58 static int brnf_call_arptables __read_mostly = 1;
  59 static int brnf_filter_vlan_tagged __read_mostly = 1;
  60 #else
  61 #define brnf_filter_vlan_tagged 1
  62 #endif
  63
  64 static __be16 inline vlan_proto(const struct sk_buff *skb)
  65 {
  66         return vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
  67 }
  68
  69 #define IS_VLAN_IP(skb) \
  70         (skb->protocol == htons(ETH_P_8021Q) && \
  71          vlan_proto(skb) == htons(ETH_P_IP) &&  \
  72          brnf_filter_vlan_tagged)
  73
  74 #define IS_VLAN_IPV6(skb) \
  75         (skb->protocol == htons(ETH_P_8021Q) && \
  76          vlan_proto(skb) == htons(ETH_P_IPV6) &&\
  77          brnf_filter_vlan_tagged)
  78
  79 #define IS_VLAN_ARP(skb) \
  80         (skb->protocol == htons(ETH_P_8021Q) && \
  81          vlan_proto(skb) == htons(ETH_P_ARP) && \
  82          brnf_filter_vlan_tagged)
  83
  84 /* We need these fake structures to make netfilter happy --
  85  * lots of places assume that skb->dst != NULL, which isn't
  86  * all that unreasonable.
  87  *
  88  * Currently, we fill in the PMTU entry because netfilter
  89  * refragmentation needs it, and the rt_flags entry because
  90  * ipt_REJECT needs it.  Future netfilter modules might
  91  * require us to fill additional fields. */
  92 static struct net_device __fake_net_device = {
  93         .hard_header_len        = ETH_HLEN
  94 };
  95
  96 static struct rtable __fake_rtable = {
  97         .u = {
  98                 .dst = {
  99                         .__refcnt               = ATOMIC_INIT(1),
 100                         .dev                    = &__fake_net_device,
 101                         .path                   = &__fake_rtable.u.dst,
 102                         .metrics                = {[RTAX_MTU - 1] = 1500},
 103                         .flags                  = DST_NOXFRM,
 104                 }
 105         },
 106         .rt_flags       = 0,
 107 };
 108
 109 static inline struct net_device *bridge_parent(const struct net_device *dev)
 110 {
 111         struct net_bridge_port *port = rcu_dereference(dev->br_port);
 112
 113         return port ? port->br->dev : NULL;
 114 }
 115
 116 static inline struct nf_bridge_info *nf_bridge_alloc(struct sk_buff *skb)
 117 {
 118         skb->nf_bridge = kzalloc(sizeof(struct nf_bridge_info), GFP_ATOMIC);
 119         if (likely(skb->nf_bridge))
 120                 atomic_set(&(skb->nf_bridge->use), 1);
 121
 122         return skb->nf_bridge;
 123 }
 124
 125 static inline void nf_bridge_save_header(struct sk_buff *skb)
 126 {
 127         int header_size = ETH_HLEN;
 128
 129         if (skb->protocol == htons(ETH_P_8021Q))
 130                 header_size += VLAN_HLEN;
 131
 132         memcpy(skb->nf_bridge->data, skb->data - header_size, header_size);
 133 }
 134
 135 /*
 136  * When forwarding bridge frames, we save a copy of the original
 137  * header before processing.
 138  */
 139 int nf_bridge_copy_header(struct sk_buff *skb)
 140 {
 141         int err;
 142         int header_size = ETH_HLEN;
 143
 144         if (skb->protocol == htons(ETH_P_8021Q))
 145                 header_size += VLAN_HLEN;
 146
 147         err = skb_cow(skb, header_size);
 148         if (err)
 149                 return err;
 150
 151         memcpy(skb->data - header_size, skb->nf_bridge->data, header_size);
 152
 153         if (skb->protocol == htons(ETH_P_8021Q))
 154                 __skb_push(skb, VLAN_HLEN);
 155         return 0;
 156 }
 157
 158 /* PF_BRIDGE/PRE_ROUTING *********************************************/
 159 /* Undo the changes made for ip6tables PREROUTING and continue the
 160  * bridge PRE_ROUTING hook. */
 161 static int br_nf_pre_routing_finish_ipv6(struct sk_buff *skb)
 162 {
 163         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 164
 165         if (nf_bridge->mask & BRNF_PKT_TYPE) {
 166                 skb->pkt_type = PACKET_OTHERHOST;
 167                 nf_bridge->mask ^= BRNF_PKT_TYPE;
 168         }
 169         nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
 170
 171         skb->dst = (struct dst_entry *)&__fake_rtable;
 172         dst_hold(skb->dst);
 173
 174         skb->dev = nf_bridge->physindev;
 175         if (skb->protocol == htons(ETH_P_8021Q)) {
 176                 skb_push(skb, VLAN_HLEN);
 177                 skb->nh.raw -= VLAN_HLEN;
 178         }
 179         NF_HOOK_THRESH(PF_BRIDGE, NF_BR_PRE_ROUTING, skb, skb->dev, NULL,
 180                        br_handle_frame_finish, 1);
 181
 182         return 0;
 183 }
 184
 185 static void __br_dnat_complain(void)
 186 {
 187         static unsigned long last_complaint;
 188
 189         if (jiffies - last_complaint >= 5 * HZ) {
 190                 printk(KERN_WARNING "Performing cross-bridge DNAT requires IP "
 191                        "forwarding to be enabled\n");
 192                 last_complaint = jiffies;
 193         }
 194 }
 195
 196 /* This requires some explaining. If DNAT has taken place,
 197  * we will need to fix up the destination Ethernet address,
 198  * and this is a tricky process.
 199  *
 200  * There are two cases to consider:
 201  * 1. The packet was DNAT'ed to a device in the same bridge
 202  *    port group as it was received on. We can still bridge
 203  *    the packet.
 204  * 2. The packet was DNAT'ed to a different device, either
 205  *    a non-bridged device or another bridge port group.
 206  *    The packet will need to be routed.
 207  *
 208  * The correct way of distinguishing between these two cases is to
 209  * call ip_route_input() and to look at skb->dst->dev, which is
 210  * changed to the destination device if ip_route_input() succeeds.
 211  *
 212  * Let us first consider the case that ip_route_input() succeeds:
 213  *
 214  * If skb->dst->dev equals the logical bridge device the packet
 215  * came in on, we can consider this bridging. We then call
 216  * skb->dst->output() which will make the packet enter br_nf_local_out()
 217  * not much later. In that function it is assured that the iptables
 218  * FORWARD chain is traversed for the packet.
 219  *
 220  * Otherwise, the packet is considered to be routed and we just
 221  * change the destination MAC address so that the packet will
 222  * later be passed up to the IP stack to be routed. For a redirected
 223  * packet, ip_route_input() will give back the localhost as output device,
 224  * which differs from the bridge device.
 225  *
 226  * Let us now consider the case that ip_route_input() fails:
 227  *
 228  * This can be because the destination address is martian, in which case
 229  * the packet will be dropped.
 230  * After a "echo '0' > /proc/sys/net/ipv4/ip_forward" ip_route_input()
 231  * will fail, while __ip_route_output_key() will return success. The source
 232  * address for __ip_route_output_key() is set to zero, so __ip_route_output_key
 233  * thinks we're handling a locally generated packet and won't care
 234  * if IP forwarding is allowed. We send a warning message to the users's
 235  * log telling her to put IP forwarding on.
 236  *
 237  * ip_route_input() will also fail if there is no route available.
 238  * In that case we just drop the packet.
 239  *
 240  * --Lennert, 20020411
 241  * --Bart, 20020416 (updated)
 242  * --Bart, 20021007 (updated)
 243  * --Bart, 20062711 (updated) */
 244 static int br_nf_pre_routing_finish_bridge(struct sk_buff *skb)
 245 {
 246         if (skb->pkt_type == PACKET_OTHERHOST) {
 247                 skb->pkt_type = PACKET_HOST;
 248                 skb->nf_bridge->mask |= BRNF_PKT_TYPE;
 249         }
 250         skb->nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
 251
 252         skb->dev = bridge_parent(skb->dev);
 253         if (!skb->dev)
 254                 kfree_skb(skb);
 255         else {
 256                 if (skb->protocol == htons(ETH_P_8021Q)) {
 257                         skb_pull(skb, VLAN_HLEN);
 258                         skb->nh.raw += VLAN_HLEN;
 259                 }
 260                 skb->dst->output(skb);
 261         }
 262         return 0;
 263 }
 264
 265 static int br_nf_pre_routing_finish(struct sk_buff *skb)
 266 {
 267         struct net_device *dev = skb->dev;
 268         struct iphdr *iph = skb->nh.iph;
 269         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 270         int err;
 271
 272         if (nf_bridge->mask & BRNF_PKT_TYPE) {
 273                 skb->pkt_type = PACKET_OTHERHOST;
 274                 nf_bridge->mask ^= BRNF_PKT_TYPE;
 275         }
 276         nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
 277         if (dnat_took_place(skb)) {
 278                 if ((err = ip_route_input(skb, iph->daddr, iph->saddr, iph->tos, dev))) {
 279                         struct rtable *rt;
 280                         struct flowi fl = {
 281                                 .nl_u = {
 282                                         .ip4_u = {
 283                                                  .daddr = iph->daddr,
 284                                                  .saddr = 0,
 285                                                  .tos = RT_TOS(iph->tos) },
 286                                 },
 287                                 .proto = 0,
 288                         };
 289                         struct in_device *in_dev = in_dev_get(dev);
 290
 291                         /* If err equals -EHOSTUNREACH the error is due to a
 292                          * martian destination or due to the fact that
 293                          * forwarding is disabled. For most martian packets,
 294                          * ip_route_output_key() will fail. It won't fail for 2 types of
 295                          * martian destinations: loopback destinations and destination
 296                          * 0.0.0.0. In both cases the packet will be dropped because the
 297                          * destination is the loopback device and not the bridge. */
 298                         if (err != -EHOSTUNREACH || !in_dev || IN_DEV_FORWARD(in_dev))
 299                                 goto free_skb;
 300
 301                         if (!ip_route_output_key(&rt, &fl)) {
 302                                 /* - Bridged-and-DNAT'ed traffic doesn't
 303                                  *   require ip_forwarding. */
 304                                 if (((struct dst_entry *)rt)->dev == dev) {
 305                                         skb->dst = (struct dst_entry *)rt;
 306                                         goto bridged_dnat;
 307                                 }
 308                                 /* we are sure that forwarding is disabled, so printing
 309                                  * this message is no problem. Note that the packet could
 310                                  * still have a martian destination address, in which case
 311                                  * the packet could be dropped even if forwarding were enabled */
 312                                 __br_dnat_complain();
 313                                 dst_release((struct dst_entry *)rt);
 314                         }
 315 free_skb:
 316                         kfree_skb(skb);
 317                         return 0;
 318                 } else {
 319                         if (skb->dst->dev == dev) {
 320 bridged_dnat:
 321                                 /* Tell br_nf_local_out this is a
 322                                  * bridged frame */
 323                                 nf_bridge->mask |= BRNF_BRIDGED_DNAT;
 324                                 skb->dev = nf_bridge->physindev;
 325                                 if (skb->protocol ==
 326                                     htons(ETH_P_8021Q)) {
 327                                         skb_push(skb, VLAN_HLEN);
 328                                         skb->nh.raw -= VLAN_HLEN;
 329                                 }
 330                                 NF_HOOK_THRESH(PF_BRIDGE, NF_BR_PRE_ROUTING,
 331                                                skb, skb->dev, NULL,
 332                                                br_nf_pre_routing_finish_bridge,
 333                                                1);
 334                                 return 0;
 335                         }
 336                         memcpy(eth_hdr(skb)->h_dest, dev->dev_addr, ETH_ALEN);
 337                         skb->pkt_type = PACKET_HOST;
 338                 }
 339         } else {
 340                 skb->dst = (struct dst_entry *)&__fake_rtable;
 341                 dst_hold(skb->dst);
 342         }
 343
 344         skb->dev = nf_bridge->physindev;
 345         if (skb->protocol == htons(ETH_P_8021Q)) {
 346                 skb_push(skb, VLAN_HLEN);
 347                 skb->nh.raw -= VLAN_HLEN;
 348         }
 349         NF_HOOK_THRESH(PF_BRIDGE, NF_BR_PRE_ROUTING, skb, skb->dev, NULL,
 350                        br_handle_frame_finish, 1);
 351
 352         return 0;
 353 }
 354
 355 /* Some common code for IPv4/IPv6 */
 356 static struct net_device *setup_pre_routing(struct sk_buff *skb)
 357 {
 358         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 359
 360         if (skb->pkt_type == PACKET_OTHERHOST) {
 361                 skb->pkt_type = PACKET_HOST;
 362                 nf_bridge->mask |= BRNF_PKT_TYPE;
 363         }
 364
 365         nf_bridge->mask |= BRNF_NF_BRIDGE_PREROUTING;
 366         nf_bridge->physindev = skb->dev;
 367         skb->dev = bridge_parent(skb->dev);
 368
 369         return skb->dev;
 370 }
 371
 372 /* We only check the length. A bridge shouldn't do any hop-by-hop stuff anyway */
 373 static int check_hbh_len(struct sk_buff *skb)
 374 {
 375         unsigned char *raw = (u8 *) (skb->nh.ipv6h + 1);
 376         u32 pkt_len;
 377         int off = raw - skb->nh.raw;
 378         int len = (raw[1] + 1) << 3;
 379
 380         if ((raw + len) - skb->data > skb_headlen(skb))
 381                 goto bad;
 382
 383         off += 2;
 384         len -= 2;
 385
 386         while (len > 0) {
 387                 int optlen = skb->nh.raw[off + 1] + 2;
 388
 389                 switch (skb->nh.raw[off]) {
 390                 case IPV6_TLV_PAD0:
 391                         optlen = 1;
 392                         break;
 393
 394                 case IPV6_TLV_PADN:
 395                         break;
 396
 397                 case IPV6_TLV_JUMBO:
 398                         if (skb->nh.raw[off + 1] != 4 || (off & 3) != 2)
 399                                 goto bad;
 400                         pkt_len = ntohl(*(__be32 *) (skb->nh.raw + off + 2));
 401                         if (pkt_len <= IPV6_MAXPLEN ||
 402                             skb->nh.ipv6h->payload_len)
 403                                 goto bad;
 404                         if (pkt_len > skb->len - sizeof(struct ipv6hdr))
 405                                 goto bad;
 406                         if (pskb_trim_rcsum(skb,
 407                                             pkt_len + sizeof(struct ipv6hdr)))
 408                                 goto bad;
 409                         break;
 410                 default:
 411                         if (optlen > len)
 412                                 goto bad;
 413                         break;
 414                 }
 415                 off += optlen;
 416                 len -= optlen;
 417         }
 418         if (len == 0)
 419                 return 0;
 420 bad:
 421         return -1;
 422
 423 }
 424
 425 /* Replicate the checks that IPv6 does on packet reception and pass the packet
 426  * to ip6tables, which doesn't support NAT, so things are fairly simple. */
 427 static unsigned int br_nf_pre_routing_ipv6(unsigned int hook,
 428                                            struct sk_buff *skb,
 429                                            const struct net_device *in,
 430                                            const struct net_device *out,
 431                                            int (*okfn)(struct sk_buff *))
 432 {
 433         struct ipv6hdr *hdr;
 434         u32 pkt_len;
 435
 436         if (skb->len < sizeof(struct ipv6hdr))
 437                 goto inhdr_error;
 438
 439         if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
 440                 goto inhdr_error;
 441
 442         hdr = skb->nh.ipv6h;
 443
 444         if (hdr->version != 6)
 445                 goto inhdr_error;
 446
 447         pkt_len = ntohs(hdr->payload_len);
 448
 449         if (pkt_len || hdr->nexthdr != NEXTHDR_HOP) {
 450                 if (pkt_len + sizeof(struct ipv6hdr) > skb->len)
 451                         goto inhdr_error;
 452                 if (pskb_trim_rcsum(skb, pkt_len + sizeof(struct ipv6hdr)))
 453                         goto inhdr_error;
 454         }
 455         if (hdr->nexthdr == NEXTHDR_HOP && check_hbh_len(skb))
 456                 goto inhdr_error;
 457
 458         nf_bridge_put(skb->nf_bridge);
 459         if (!nf_bridge_alloc(skb))
 460                 return NF_DROP;
 461         if (!setup_pre_routing(skb))
 462                 return NF_DROP;
 463
 464         NF_HOOK(PF_INET6, NF_IP6_PRE_ROUTING, skb, skb->dev, NULL,
 465                 br_nf_pre_routing_finish_ipv6);
 466
 467         return NF_STOLEN;
 468
 469 inhdr_error:
 470         return NF_DROP;
 471 }
 472
 473 /* Direct IPv6 traffic to br_nf_pre_routing_ipv6.
 474  * Replicate the checks that IPv4 does on packet reception.
 475  * Set skb->dev to the bridge device (i.e. parent of the
 476  * receiving device) to make netfilter happy, the REDIRECT
 477  * target in particular.  Save the original destination IP
 478  * address to be able to detect DNAT afterwards. */
 479 static unsigned int br_nf_pre_routing(unsigned int hook, struct sk_buff **pskb,
 480                                       const struct net_device *in,
 481                                       const struct net_device *out,
 482                                       int (*okfn)(struct sk_buff *))
 483 {
 484         struct iphdr *iph;
 485         __u32 len;
 486         struct sk_buff *skb = *pskb;
 487
 488         if (skb->protocol == htons(ETH_P_IPV6) || IS_VLAN_IPV6(skb)) {
 489 #ifdef CONFIG_SYSCTL
 490                 if (!brnf_call_ip6tables)
 491                         return NF_ACCEPT;
 492 #endif
 493                 if ((skb = skb_share_check(*pskb, GFP_ATOMIC)) == NULL)
 494                         goto out;
 495
 496                 if (skb->protocol == htons(ETH_P_8021Q)) {
 497                         skb_pull_rcsum(skb, VLAN_HLEN);
 498                         skb->nh.raw += VLAN_HLEN;
 499                 }
 500                 return br_nf_pre_routing_ipv6(hook, skb, in, out, okfn);
 501         }
 502 #ifdef CONFIG_SYSCTL
 503         if (!brnf_call_iptables)
 504                 return NF_ACCEPT;
 505 #endif
 506
 507         if (skb->protocol != htons(ETH_P_IP) && !IS_VLAN_IP(skb))
 508                 return NF_ACCEPT;
 509
 510         if ((skb = skb_share_check(*pskb, GFP_ATOMIC)) == NULL)
 511                 goto out;
 512
 513         if (skb->protocol == htons(ETH_P_8021Q)) {
 514                 skb_pull_rcsum(skb, VLAN_HLEN);
 515                 skb->nh.raw += VLAN_HLEN;
 516         }
 517
 518         if (!pskb_may_pull(skb, sizeof(struct iphdr)))
 519                 goto inhdr_error;
 520
 521         iph = skb->nh.iph;
 522         if (iph->ihl < 5 || iph->version != 4)
 523                 goto inhdr_error;
 524
 525         if (!pskb_may_pull(skb, 4 * iph->ihl))
 526                 goto inhdr_error;
 527
 528         iph = skb->nh.iph;
 529         if (ip_fast_csum((__u8 *) iph, iph->ihl) != 0)
 530                 goto inhdr_error;
 531
 532         len = ntohs(iph->tot_len);
 533         if (skb->len < len || len < 4 * iph->ihl)
 534                 goto inhdr_error;
 535
 536         pskb_trim_rcsum(skb, len);
 537
 538         nf_bridge_put(skb->nf_bridge);
 539         if (!nf_bridge_alloc(skb))
 540                 return NF_DROP;
 541         if (!setup_pre_routing(skb))
 542                 return NF_DROP;
 543         store_orig_dstaddr(skb);
 544
 545         NF_HOOK(PF_INET, NF_IP_PRE_ROUTING, skb, skb->dev, NULL,
 546                 br_nf_pre_routing_finish);
 547
 548         return NF_STOLEN;
 549
 550 inhdr_error:
 551 //      IP_INC_STATS_BH(IpInHdrErrors);
 552 out:
 553         return NF_DROP;
 554 }
 555
 556
 557 /* PF_BRIDGE/LOCAL_IN ************************************************/
 558 /* The packet is locally destined, which requires a real
 559  * dst_entry, so detach the fake one.  On the way up, the
 560  * packet would pass through PRE_ROUTING again (which already
 561  * took place when the packet entered the bridge), but we
 562  * register an IPv4 PRE_ROUTING 'sabotage' hook that will
 563  * prevent this from happening. */
 564 static unsigned int br_nf_local_in(unsigned int hook, struct sk_buff **pskb,
 565                                    const struct net_device *in,
 566                                    const struct net_device *out,
 567                                    int (*okfn)(struct sk_buff *))
 568 {
 569         struct sk_buff *skb = *pskb;
 570
 571         if (skb->dst == (struct dst_entry *)&__fake_rtable) {
 572                 dst_release(skb->dst);
 573                 skb->dst = NULL;
 574         }
 575
 576         return NF_ACCEPT;
 577 }
 578
 579 /* PF_BRIDGE/FORWARD *************************************************/
 580 static int br_nf_forward_finish(struct sk_buff *skb)
 581 {
 582         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 583         struct net_device *in;
 584
 585         if (skb->protocol != htons(ETH_P_ARP) && !IS_VLAN_ARP(skb)) {
 586                 in = nf_bridge->physindev;
 587                 if (nf_bridge->mask & BRNF_PKT_TYPE) {
 588                         skb->pkt_type = PACKET_OTHERHOST;
 589                         nf_bridge->mask ^= BRNF_PKT_TYPE;
 590                 }
 591         } else {
 592                 in = *((struct net_device **)(skb->cb));
 593         }
 594         if (skb->protocol == htons(ETH_P_8021Q)) {
 595                 skb_push(skb, VLAN_HLEN);
 596                 skb->nh.raw -= VLAN_HLEN;
 597         }
 598         NF_HOOK_THRESH(PF_BRIDGE, NF_BR_FORWARD, skb, in,
 599                        skb->dev, br_forward_finish, 1);
 600         return 0;
 601 }
 602
 603 /* This is the 'purely bridged' case.  For IP, we pass the packet to
 604  * netfilter with indev and outdev set to the bridge device,
 605  * but we are still able to filter on the 'real' indev/outdev
 606  * because of the physdev module. For ARP, indev and outdev are the
 607  * bridge ports. */
 608 static unsigned int br_nf_forward_ip(unsigned int hook, struct sk_buff **pskb,
 609                                      const struct net_device *in,
 610                                      const struct net_device *out,
 611                                      int (*okfn)(struct sk_buff *))
 612 {
 613         struct sk_buff *skb = *pskb;
 614         struct nf_bridge_info *nf_bridge;
 615         struct net_device *parent;
 616         int pf;
 617
 618         if (!skb->nf_bridge)
 619                 return NF_ACCEPT;
 620
 621         parent = bridge_parent(out);
 622         if (!parent)
 623                 return NF_DROP;
 624
 625         if (skb->protocol == htons(ETH_P_IP) || IS_VLAN_IP(skb))
 626                 pf = PF_INET;
 627         else
 628                 pf = PF_INET6;
 629
 630         if (skb->protocol == htons(ETH_P_8021Q)) {
 631                 skb_pull(*pskb, VLAN_HLEN);
 632                 (*pskb)->nh.raw += VLAN_HLEN;
 633         }
 634
 635         nf_bridge = skb->nf_bridge;
 636         if (skb->pkt_type == PACKET_OTHERHOST) {
 637                 skb->pkt_type = PACKET_HOST;
 638                 nf_bridge->mask |= BRNF_PKT_TYPE;
 639         }
 640
 641         /* The physdev module checks on this */
 642         nf_bridge->mask |= BRNF_BRIDGED;
 643         nf_bridge->physoutdev = skb->dev;
 644
 645         NF_HOOK(pf, NF_IP_FORWARD, skb, bridge_parent(in), parent,
 646                 br_nf_forward_finish);
 647
 648         return NF_STOLEN;
 649 }
 650
 651 static unsigned int br_nf_forward_arp(unsigned int hook, struct sk_buff **pskb,
 652                                       const struct net_device *in,
 653                                       const struct net_device *out,
 654                                       int (*okfn)(struct sk_buff *))
 655 {
 656         struct sk_buff *skb = *pskb;
 657         struct net_device **d = (struct net_device **)(skb->cb);
 658
 659 #ifdef CONFIG_SYSCTL
 660         if (!brnf_call_arptables)
 661                 return NF_ACCEPT;
 662 #endif
 663
 664         if (skb->protocol != htons(ETH_P_ARP)) {
 665                 if (!IS_VLAN_ARP(skb))
 666                         return NF_ACCEPT;
 667                 skb_pull(*pskb, VLAN_HLEN);
 668                 (*pskb)->nh.raw += VLAN_HLEN;
 669         }
 670
 671         if (skb->nh.arph->ar_pln != 4) {
 672                 if (IS_VLAN_ARP(skb)) {
 673                         skb_push(*pskb, VLAN_HLEN);
 674                         (*pskb)->nh.raw -= VLAN_HLEN;
 675                 }
 676                 return NF_ACCEPT;
 677         }
 678         *d = (struct net_device *)in;
 679         NF_HOOK(NF_ARP, NF_ARP_FORWARD, skb, (struct net_device *)in,
 680                 (struct net_device *)out, br_nf_forward_finish);
 681
 682         return NF_STOLEN;
 683 }
 684
 685 /* PF_BRIDGE/LOCAL_OUT ***********************************************
 686  *
 687  * This function sees both locally originated IP packets and forwarded
 688  * IP packets (in both cases the destination device is a bridge
 689  * device). It also sees bridged-and-DNAT'ed packets.
 690  *
 691  * If (nf_bridge->mask & BRNF_BRIDGED_DNAT) then the packet is bridged
 692  * and we fake the PF_BRIDGE/FORWARD hook. The function br_nf_forward()
 693  * will then fake the PF_INET/FORWARD hook. br_nf_local_out() has priority
 694  * NF_BR_PRI_FIRST, so no relevant PF_BRIDGE/INPUT functions have been nor
 695  * will be executed.
 696  */
 697 static unsigned int br_nf_local_out(unsigned int hook, struct sk_buff **pskb,
 698                                     const struct net_device *in,
 699                                     const struct net_device *out,
 700                                     int (*okfn)(struct sk_buff *))
 701 {
 702         struct net_device *realindev;
 703         struct sk_buff *skb = *pskb;
 704         struct nf_bridge_info *nf_bridge;
 705
 706         if (!skb->nf_bridge)
 707                 return NF_ACCEPT;
 708
 709         nf_bridge = skb->nf_bridge;
 710         if (!(nf_bridge->mask & BRNF_BRIDGED_DNAT))
 711                 return NF_ACCEPT;
 712
 713         /* Bridged, take PF_BRIDGE/FORWARD.
 714          * (see big note in front of br_nf_pre_routing_finish) */
 715         nf_bridge->physoutdev = skb->dev;
 716         realindev = nf_bridge->physindev;
 717
 718         if (nf_bridge->mask & BRNF_PKT_TYPE) {
 719                 skb->pkt_type = PACKET_OTHERHOST;
 720                 nf_bridge->mask ^= BRNF_PKT_TYPE;
 721         }
 722         if (skb->protocol == htons(ETH_P_8021Q)) {
 723                 skb_push(skb, VLAN_HLEN);
 724                 skb->nh.raw -= VLAN_HLEN;
 725         }
 726
 727         NF_HOOK(PF_BRIDGE, NF_BR_FORWARD, skb, realindev, skb->dev,
 728                 br_forward_finish);
 729         return NF_STOLEN;
 730 }
 731
 732 static int br_nf_dev_queue_xmit(struct sk_buff *skb)
 733 {
 734         if (skb->protocol == htons(ETH_P_IP) &&
 735             skb->len > skb->dev->mtu &&
 736             !skb_is_gso(skb))
 737                 return ip_fragment(skb, br_dev_queue_push_xmit);
 738         else
 739                 return br_dev_queue_push_xmit(skb);
 740 }
 741
 742 /* PF_BRIDGE/POST_ROUTING ********************************************/
 743 static unsigned int br_nf_post_routing(unsigned int hook, struct sk_buff **pskb,
 744                                        const struct net_device *in,
 745                                        const struct net_device *out,
 746                                        int (*okfn)(struct sk_buff *))
 747 {
 748         struct sk_buff *skb = *pskb;
 749         struct nf_bridge_info *nf_bridge = (*pskb)->nf_bridge;
 750         struct net_device *realoutdev = bridge_parent(skb->dev);
 751         int pf;
 752
 753 #ifdef CONFIG_NETFILTER_DEBUG
 754         /* Be very paranoid. This probably won't happen anymore, but let's
 755          * keep the check just to be sure... */
 756         if (skb->mac.raw < skb->head || skb->mac.raw + ETH_HLEN > skb->data) {
 757                 printk(KERN_CRIT "br_netfilter: Argh!! br_nf_post_routing: "
 758                        "bad mac.raw pointer.\n");
 759                 goto print_error;
 760         }
 761 #endif
 762
 763         if (!nf_bridge)
 764                 return NF_ACCEPT;
 765
 766         if (!realoutdev)
 767                 return NF_DROP;
 768
 769         if (skb->protocol == htons(ETH_P_IP) || IS_VLAN_IP(skb))
 770                 pf = PF_INET;
 771         else
 772                 pf = PF_INET6;
 773
 774 #ifdef CONFIG_NETFILTER_DEBUG
 775         if (skb->dst == NULL) {
 776                 printk(KERN_INFO "br_netfilter post_routing: skb->dst == NULL\n");
 777                 goto print_error;
 778         }
 779 #endif
 780
 781         /* We assume any code from br_dev_queue_push_xmit onwards doesn't care
 782          * about the value of skb->pkt_type. */
 783         if (skb->pkt_type == PACKET_OTHERHOST) {
 784                 skb->pkt_type = PACKET_HOST;
 785                 nf_bridge->mask |= BRNF_PKT_TYPE;
 786         }
 787
 788         if (skb->protocol == htons(ETH_P_8021Q)) {
 789                 skb_pull(skb, VLAN_HLEN);
 790                 skb->nh.raw += VLAN_HLEN;
 791         }
 792
 793         nf_bridge_save_header(skb);
 794
 795 #if defined(CONFIG_VLAN_8021Q) || defined(CONFIG_VLAN_8021Q_MODULE)
 796         if (nf_bridge->netoutdev)
 797                 realoutdev = nf_bridge->netoutdev;
 798 #endif
 799         NF_HOOK(pf, NF_IP_POST_ROUTING, skb, NULL, realoutdev,
 800                 br_nf_dev_queue_xmit);
 801
 802         return NF_STOLEN;
 803
 804 #ifdef CONFIG_NETFILTER_DEBUG
 805 print_error:
 806         if (skb->dev != NULL) {
 807                 printk("[%s]", skb->dev->name);
 808                 if (realoutdev)
 809                         printk("[%s]", realoutdev->name);
 810         }
 811         printk(" head:%p, raw:%p, data:%p\n", skb->head, skb->mac.raw,
 812                skb->data);
 813         dump_stack();
 814         return NF_ACCEPT;
 815 #endif
 816 }
 817
 818 /* IP/SABOTAGE *****************************************************/
 819 /* Don't hand locally destined packets to PF_INET(6)/PRE_ROUTING
 820  * for the second time. */
 821 static unsigned int ip_sabotage_in(unsigned int hook, struct sk_buff **pskb,
 822                                    const struct net_device *in,
 823                                    const struct net_device *out,
 824                                    int (*okfn)(struct sk_buff *))
 825 {
 826         if ((*pskb)->nf_bridge &&
 827             !((*pskb)->nf_bridge->mask & BRNF_NF_BRIDGE_PREROUTING)) {
 828                 return NF_STOP;
 829         }
 830
 831         return NF_ACCEPT;
 832 }
 833
 834 /* For br_nf_local_out we need (prio = NF_BR_PRI_FIRST), to insure that innocent
 835  * PF_BRIDGE/NF_BR_LOCAL_OUT functions don't get bridged traffic as input.
 836  * For br_nf_post_routing, we need (prio = NF_BR_PRI_LAST), because
 837  * ip_refrag() can return NF_STOLEN. */
 838 static struct nf_hook_ops br_nf_ops[] = {
 839         { .hook = br_nf_pre_routing,
 840           .owner = THIS_MODULE,
 841           .pf = PF_BRIDGE,
 842           .hooknum = NF_BR_PRE_ROUTING,
 843           .priority = NF_BR_PRI_BRNF, },
 844         { .hook = br_nf_local_in,
 845           .owner = THIS_MODULE,
 846           .pf = PF_BRIDGE,
 847           .hooknum = NF_BR_LOCAL_IN,
 848           .priority = NF_BR_PRI_BRNF, },
 849         { .hook = br_nf_forward_ip,
 850           .owner = THIS_MODULE,
 851           .pf = PF_BRIDGE,
 852           .hooknum = NF_BR_FORWARD,
 853           .priority = NF_BR_PRI_BRNF - 1, },
 854         { .hook = br_nf_forward_arp,
 855           .owner = THIS_MODULE,
 856           .pf = PF_BRIDGE,
 857           .hooknum = NF_BR_FORWARD,
 858           .priority = NF_BR_PRI_BRNF, },
 859         { .hook = br_nf_local_out,
 860           .owner = THIS_MODULE,
 861           .pf = PF_BRIDGE,
 862           .hooknum = NF_BR_LOCAL_OUT,
 863           .priority = NF_BR_PRI_FIRST, },
 864         { .hook = br_nf_post_routing,
 865           .owner = THIS_MODULE,
 866           .pf = PF_BRIDGE,
 867           .hooknum = NF_BR_POST_ROUTING,
 868           .priority = NF_BR_PRI_LAST, },
 869         { .hook = ip_sabotage_in,
 870           .owner = THIS_MODULE,
 871           .pf = PF_INET,
 872           .hooknum = NF_IP_PRE_ROUTING,
 873           .priority = NF_IP_PRI_FIRST, },
 874         { .hook = ip_sabotage_in,
 875           .owner = THIS_MODULE,
 876           .pf = PF_INET6,
 877           .hooknum = NF_IP6_PRE_ROUTING,
 878           .priority = NF_IP6_PRI_FIRST, },
 879 };
 880
 881 #ifdef CONFIG_SYSCTL
 882 static
 883 int brnf_sysctl_call_tables(ctl_table * ctl, int write, struct file *filp,
 884                             void __user * buffer, size_t * lenp, loff_t * ppos)
 885 {
 886         int ret;
 887
 888         ret = proc_dointvec(ctl, write, filp, buffer, lenp, ppos);
 889
 890         if (write && *(int *)(ctl->data))
 891                 *(int *)(ctl->data) = 1;
 892         return ret;
 893 }
 894
 895 static ctl_table brnf_table[] = {
 896         {
 897                 .ctl_name       = NET_BRIDGE_NF_CALL_ARPTABLES,
 898                 .procname       = "bridge-nf-call-arptables",
 899                 .data           = &brnf_call_arptables,
 900                 .maxlen         = sizeof(int),
 901                 .mode           = 0644,
 902                 .proc_handler   = &brnf_sysctl_call_tables,
 903         },
 904         {
 905                 .ctl_name       = NET_BRIDGE_NF_CALL_IPTABLES,
 906                 .procname       = "bridge-nf-call-iptables",
 907                 .data           = &brnf_call_iptables,
 908                 .maxlen         = sizeof(int),
 909                 .mode           = 0644,
 910                 .proc_handler   = &brnf_sysctl_call_tables,
 911         },
 912         {
 913                 .ctl_name       = NET_BRIDGE_NF_CALL_IP6TABLES,
 914                 .procname       = "bridge-nf-call-ip6tables",
 915                 .data           = &brnf_call_ip6tables,
 916                 .maxlen         = sizeof(int),
 917                 .mode           = 0644,
 918                 .proc_handler   = &brnf_sysctl_call_tables,
 919         },
 920         {
 921                 .ctl_name       = NET_BRIDGE_NF_FILTER_VLAN_TAGGED,
 922                 .procname       = "bridge-nf-filter-vlan-tagged",
 923                 .data           = &brnf_filter_vlan_tagged,
 924                 .maxlen         = sizeof(int),
 925                 .mode           = 0644,
 926                 .proc_handler   = &brnf_sysctl_call_tables,
 927         },
 928         { .ctl_name = 0 }
 929 };
 930
 931 static ctl_table brnf_bridge_table[] = {
 932         {
 933                 .ctl_name       = NET_BRIDGE,
 934                 .procname       = "bridge",
 935                 .mode           = 0555,
 936                 .child          = brnf_table,
 937         },
 938         { .ctl_name = 0 }
 939 };
 940
 941 static ctl_table brnf_net_table[] = {
 942         {
 943                 .ctl_name       = CTL_NET,
 944                 .procname       = "net",
 945                 .mode           = 0555,
 946                 .child          = brnf_bridge_table,
 947         },
 948         { .ctl_name = 0 }
 949 };
 950 #endif
 951
 952 int __init br_netfilter_init(void)
 953 {
 954         int ret;
 955
 956         ret = nf_register_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
 957         if (ret < 0)
 958                 return ret;
 959 #ifdef CONFIG_SYSCTL
 960         brnf_sysctl_header = register_sysctl_table(brnf_net_table, 0);
 961         if (brnf_sysctl_header == NULL) {
 962                 printk(KERN_WARNING
 963                        "br_netfilter: can't register to sysctl.\n");
 964                 nf_unregister_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
 965                 return -ENOMEM;
 966         }
 967 #endif
 968         printk(KERN_NOTICE "Bridge firewalling registered\n");
 969         return 0;
 970 }
 971
 972 void br_netfilter_fini(void)
 973 {
 974         nf_unregister_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
 975 #ifdef CONFIG_SYSCTL
 976         unregister_sysctl_table(brnf_sysctl_header);
 977 #endif
 978 }