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/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) = ip_hdr(skb)->daddr)
  52 #define dnat_took_place(skb)     (skb_origaddr(skb) != ip_hdr(skb)->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 inline __be16 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 = ip_hdr(skb);
 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         const unsigned char *nh = skb_network_header(skb);
 378         int off = raw - nh;
 379         int len = (raw[1] + 1) << 3;
 380
 381         if ((raw + len) - skb->data > skb_headlen(skb))
 382                 goto bad;
 383
 384         off += 2;
 385         len -= 2;
 386
 387         while (len > 0) {
 388                 int optlen = nh[off + 1] + 2;
 389
 390                 switch (nh[off]) {
 391                 case IPV6_TLV_PAD0:
 392                         optlen = 1;
 393                         break;
 394
 395                 case IPV6_TLV_PADN:
 396                         break;
 397
 398                 case IPV6_TLV_JUMBO:
 399                         if (nh[off + 1] != 4 || (off & 3) != 2)
 400                                 goto bad;
 401                         pkt_len = ntohl(*(__be32 *) (nh + off + 2));
 402                         if (pkt_len <= IPV6_MAXPLEN ||
 403                             skb->nh.ipv6h->payload_len)
 404                                 goto bad;
 405                         if (pkt_len > skb->len - sizeof(struct ipv6hdr))
 406                                 goto bad;
 407                         if (pskb_trim_rcsum(skb,
 408                                             pkt_len + sizeof(struct ipv6hdr)))
 409                                 goto bad;
 410                         nh = skb_network_header(skb);
 411                         break;
 412                 default:
 413                         if (optlen > len)
 414                                 goto bad;
 415                         break;
 416                 }
 417                 off += optlen;
 418                 len -= optlen;
 419         }
 420         if (len == 0)
 421                 return 0;
 422 bad:
 423         return -1;
 424
 425 }
 426
 427 /* Replicate the checks that IPv6 does on packet reception and pass the packet
 428  * to ip6tables, which doesn't support NAT, so things are fairly simple. */
 429 static unsigned int br_nf_pre_routing_ipv6(unsigned int hook,
 430                                            struct sk_buff *skb,
 431                                            const struct net_device *in,
 432                                            const struct net_device *out,
 433                                            int (*okfn)(struct sk_buff *))
 434 {
 435         struct ipv6hdr *hdr;
 436         u32 pkt_len;
 437
 438         if (skb->len < sizeof(struct ipv6hdr))
 439                 goto inhdr_error;
 440
 441         if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
 442                 goto inhdr_error;
 443
 444         hdr = skb->nh.ipv6h;
 445
 446         if (hdr->version != 6)
 447                 goto inhdr_error;
 448
 449         pkt_len = ntohs(hdr->payload_len);
 450
 451         if (pkt_len || hdr->nexthdr != NEXTHDR_HOP) {
 452                 if (pkt_len + sizeof(struct ipv6hdr) > skb->len)
 453                         goto inhdr_error;
 454                 if (pskb_trim_rcsum(skb, pkt_len + sizeof(struct ipv6hdr)))
 455                         goto inhdr_error;
 456         }
 457         if (hdr->nexthdr == NEXTHDR_HOP && check_hbh_len(skb))
 458                 goto inhdr_error;
 459
 460         nf_bridge_put(skb->nf_bridge);
 461         if (!nf_bridge_alloc(skb))
 462                 return NF_DROP;
 463         if (!setup_pre_routing(skb))
 464                 return NF_DROP;
 465
 466         NF_HOOK(PF_INET6, NF_IP6_PRE_ROUTING, skb, skb->dev, NULL,
 467                 br_nf_pre_routing_finish_ipv6);
 468
 469         return NF_STOLEN;
 470
 471 inhdr_error:
 472         return NF_DROP;
 473 }
 474
 475 /* Direct IPv6 traffic to br_nf_pre_routing_ipv6.
 476  * Replicate the checks that IPv4 does on packet reception.
 477  * Set skb->dev to the bridge device (i.e. parent of the
 478  * receiving device) to make netfilter happy, the REDIRECT
 479  * target in particular.  Save the original destination IP
 480  * address to be able to detect DNAT afterwards. */
 481 static unsigned int br_nf_pre_routing(unsigned int hook, struct sk_buff **pskb,
 482                                       const struct net_device *in,
 483                                       const struct net_device *out,
 484                                       int (*okfn)(struct sk_buff *))
 485 {
 486         struct iphdr *iph;
 487         __u32 len;
 488         struct sk_buff *skb = *pskb;
 489
 490         if (skb->protocol == htons(ETH_P_IPV6) || IS_VLAN_IPV6(skb)) {
 491 #ifdef CONFIG_SYSCTL
 492                 if (!brnf_call_ip6tables)
 493                         return NF_ACCEPT;
 494 #endif
 495                 if ((skb = skb_share_check(*pskb, GFP_ATOMIC)) == NULL)
 496                         goto out;
 497
 498                 if (skb->protocol == htons(ETH_P_8021Q)) {
 499                         skb_pull_rcsum(skb, VLAN_HLEN);
 500                         skb->nh.raw += VLAN_HLEN;
 501                 }
 502                 return br_nf_pre_routing_ipv6(hook, skb, in, out, okfn);
 503         }
 504 #ifdef CONFIG_SYSCTL
 505         if (!brnf_call_iptables)
 506                 return NF_ACCEPT;
 507 #endif
 508
 509         if (skb->protocol != htons(ETH_P_IP) && !IS_VLAN_IP(skb))
 510                 return NF_ACCEPT;
 511
 512         if ((skb = skb_share_check(*pskb, GFP_ATOMIC)) == NULL)
 513                 goto out;
 514
 515         if (skb->protocol == htons(ETH_P_8021Q)) {
 516                 skb_pull_rcsum(skb, VLAN_HLEN);
 517                 skb->nh.raw += VLAN_HLEN;
 518         }
 519
 520         if (!pskb_may_pull(skb, sizeof(struct iphdr)))
 521                 goto inhdr_error;
 522
 523         iph = ip_hdr(skb);
 524         if (iph->ihl < 5 || iph->version != 4)
 525                 goto inhdr_error;
 526
 527         if (!pskb_may_pull(skb, 4 * iph->ihl))
 528                 goto inhdr_error;
 529
 530         iph = ip_hdr(skb);
 531         if (ip_fast_csum((__u8 *) iph, iph->ihl) != 0)
 532                 goto inhdr_error;
 533
 534         len = ntohs(iph->tot_len);
 535         if (skb->len < len || len < 4 * iph->ihl)
 536                 goto inhdr_error;
 537
 538         pskb_trim_rcsum(skb, len);
 539
 540         nf_bridge_put(skb->nf_bridge);
 541         if (!nf_bridge_alloc(skb))
 542                 return NF_DROP;
 543         if (!setup_pre_routing(skb))
 544                 return NF_DROP;
 545         store_orig_dstaddr(skb);
 546
 547         NF_HOOK(PF_INET, NF_IP_PRE_ROUTING, skb, skb->dev, NULL,
 548                 br_nf_pre_routing_finish);
 549
 550         return NF_STOLEN;
 551
 552 inhdr_error:
 553 //      IP_INC_STATS_BH(IpInHdrErrors);
 554 out:
 555         return NF_DROP;
 556 }
 557
 558
 559 /* PF_BRIDGE/LOCAL_IN ************************************************/
 560 /* The packet is locally destined, which requires a real
 561  * dst_entry, so detach the fake one.  On the way up, the
 562  * packet would pass through PRE_ROUTING again (which already
 563  * took place when the packet entered the bridge), but we
 564  * register an IPv4 PRE_ROUTING 'sabotage' hook that will
 565  * prevent this from happening. */
 566 static unsigned int br_nf_local_in(unsigned int hook, struct sk_buff **pskb,
 567                                    const struct net_device *in,
 568                                    const struct net_device *out,
 569                                    int (*okfn)(struct sk_buff *))
 570 {
 571         struct sk_buff *skb = *pskb;
 572
 573         if (skb->dst == (struct dst_entry *)&__fake_rtable) {
 574                 dst_release(skb->dst);
 575                 skb->dst = NULL;
 576         }
 577
 578         return NF_ACCEPT;
 579 }
 580
 581 /* PF_BRIDGE/FORWARD *************************************************/
 582 static int br_nf_forward_finish(struct sk_buff *skb)
 583 {
 584         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 585         struct net_device *in;
 586
 587         if (skb->protocol != htons(ETH_P_ARP) && !IS_VLAN_ARP(skb)) {
 588                 in = nf_bridge->physindev;
 589                 if (nf_bridge->mask & BRNF_PKT_TYPE) {
 590                         skb->pkt_type = PACKET_OTHERHOST;
 591                         nf_bridge->mask ^= BRNF_PKT_TYPE;
 592                 }
 593         } else {
 594                 in = *((struct net_device **)(skb->cb));
 595         }
 596         if (skb->protocol == htons(ETH_P_8021Q)) {
 597                 skb_push(skb, VLAN_HLEN);
 598                 skb->nh.raw -= VLAN_HLEN;
 599         }
 600         NF_HOOK_THRESH(PF_BRIDGE, NF_BR_FORWARD, skb, in,
 601                        skb->dev, br_forward_finish, 1);
 602         return 0;
 603 }
 604
 605 /* This is the 'purely bridged' case.  For IP, we pass the packet to
 606  * netfilter with indev and outdev set to the bridge device,
 607  * but we are still able to filter on the 'real' indev/outdev
 608  * because of the physdev module. For ARP, indev and outdev are the
 609  * bridge ports. */
 610 static unsigned int br_nf_forward_ip(unsigned int hook, struct sk_buff **pskb,
 611                                      const struct net_device *in,
 612                                      const struct net_device *out,
 613                                      int (*okfn)(struct sk_buff *))
 614 {
 615         struct sk_buff *skb = *pskb;
 616         struct nf_bridge_info *nf_bridge;
 617         struct net_device *parent;
 618         int pf;
 619
 620         if (!skb->nf_bridge)
 621                 return NF_ACCEPT;
 622
 623         parent = bridge_parent(out);
 624         if (!parent)
 625                 return NF_DROP;
 626
 627         if (skb->protocol == htons(ETH_P_IP) || IS_VLAN_IP(skb))
 628                 pf = PF_INET;
 629         else
 630                 pf = PF_INET6;
 631
 632         if (skb->protocol == htons(ETH_P_8021Q)) {
 633                 skb_pull(*pskb, VLAN_HLEN);
 634                 (*pskb)->nh.raw += VLAN_HLEN;
 635         }
 636
 637         nf_bridge = skb->nf_bridge;
 638         if (skb->pkt_type == PACKET_OTHERHOST) {
 639                 skb->pkt_type = PACKET_HOST;
 640                 nf_bridge->mask |= BRNF_PKT_TYPE;
 641         }
 642
 643         /* The physdev module checks on this */
 644         nf_bridge->mask |= BRNF_BRIDGED;
 645         nf_bridge->physoutdev = skb->dev;
 646
 647         NF_HOOK(pf, NF_IP_FORWARD, skb, bridge_parent(in), parent,
 648                 br_nf_forward_finish);
 649
 650         return NF_STOLEN;
 651 }
 652
 653 static unsigned int br_nf_forward_arp(unsigned int hook, struct sk_buff **pskb,
 654                                       const struct net_device *in,
 655                                       const struct net_device *out,
 656                                       int (*okfn)(struct sk_buff *))
 657 {
 658         struct sk_buff *skb = *pskb;
 659         struct net_device **d = (struct net_device **)(skb->cb);
 660
 661 #ifdef CONFIG_SYSCTL
 662         if (!brnf_call_arptables)
 663                 return NF_ACCEPT;
 664 #endif
 665
 666         if (skb->protocol != htons(ETH_P_ARP)) {
 667                 if (!IS_VLAN_ARP(skb))
 668                         return NF_ACCEPT;
 669                 skb_pull(*pskb, VLAN_HLEN);
 670                 (*pskb)->nh.raw += VLAN_HLEN;
 671         }
 672
 673         if (skb->nh.arph->ar_pln != 4) {
 674                 if (IS_VLAN_ARP(skb)) {
 675                         skb_push(*pskb, VLAN_HLEN);
 676                         (*pskb)->nh.raw -= VLAN_HLEN;
 677                 }
 678                 return NF_ACCEPT;
 679         }
 680         *d = (struct net_device *)in;
 681         NF_HOOK(NF_ARP, NF_ARP_FORWARD, skb, (struct net_device *)in,
 682                 (struct net_device *)out, br_nf_forward_finish);
 683
 684         return NF_STOLEN;
 685 }
 686
 687 /* PF_BRIDGE/LOCAL_OUT ***********************************************
 688  *
 689  * This function sees both locally originated IP packets and forwarded
 690  * IP packets (in both cases the destination device is a bridge
 691  * device). It also sees bridged-and-DNAT'ed packets.
 692  *
 693  * If (nf_bridge->mask & BRNF_BRIDGED_DNAT) then the packet is bridged
 694  * and we fake the PF_BRIDGE/FORWARD hook. The function br_nf_forward()
 695  * will then fake the PF_INET/FORWARD hook. br_nf_local_out() has priority
 696  * NF_BR_PRI_FIRST, so no relevant PF_BRIDGE/INPUT functions have been nor
 697  * will be executed.
 698  */
 699 static unsigned int br_nf_local_out(unsigned int hook, struct sk_buff **pskb,
 700                                     const struct net_device *in,
 701                                     const struct net_device *out,
 702                                     int (*okfn)(struct sk_buff *))
 703 {
 704         struct net_device *realindev;
 705         struct sk_buff *skb = *pskb;
 706         struct nf_bridge_info *nf_bridge;
 707
 708         if (!skb->nf_bridge)
 709                 return NF_ACCEPT;
 710
 711         nf_bridge = skb->nf_bridge;
 712         if (!(nf_bridge->mask & BRNF_BRIDGED_DNAT))
 713                 return NF_ACCEPT;
 714
 715         /* Bridged, take PF_BRIDGE/FORWARD.
 716          * (see big note in front of br_nf_pre_routing_finish) */
 717         nf_bridge->physoutdev = skb->dev;
 718         realindev = nf_bridge->physindev;
 719
 720         if (nf_bridge->mask & BRNF_PKT_TYPE) {
 721                 skb->pkt_type = PACKET_OTHERHOST;
 722                 nf_bridge->mask ^= BRNF_PKT_TYPE;
 723         }
 724         if (skb->protocol == htons(ETH_P_8021Q)) {
 725                 skb_push(skb, VLAN_HLEN);
 726                 skb->nh.raw -= VLAN_HLEN;
 727         }
 728
 729         NF_HOOK(PF_BRIDGE, NF_BR_FORWARD, skb, realindev, skb->dev,
 730                 br_forward_finish);
 731         return NF_STOLEN;
 732 }
 733
 734 static int br_nf_dev_queue_xmit(struct sk_buff *skb)
 735 {
 736         if (skb->protocol == htons(ETH_P_IP) &&
 737             skb->len > skb->dev->mtu &&
 738             !skb_is_gso(skb))
 739                 return ip_fragment(skb, br_dev_queue_push_xmit);
 740         else
 741                 return br_dev_queue_push_xmit(skb);
 742 }
 743
 744 /* PF_BRIDGE/POST_ROUTING ********************************************/
 745 static unsigned int br_nf_post_routing(unsigned int hook, struct sk_buff **pskb,
 746                                        const struct net_device *in,
 747                                        const struct net_device *out,
 748                                        int (*okfn)(struct sk_buff *))
 749 {
 750         struct sk_buff *skb = *pskb;
 751         struct nf_bridge_info *nf_bridge = (*pskb)->nf_bridge;
 752         struct net_device *realoutdev = bridge_parent(skb->dev);
 753         int pf;
 754
 755 #ifdef CONFIG_NETFILTER_DEBUG
 756         /* Be very paranoid. This probably won't happen anymore, but let's
 757          * keep the check just to be sure... */
 758         if (skb_mac_header(skb) < skb->head ||
 759             skb_mac_header(skb) + ETH_HLEN > skb->data) {
 760                 printk(KERN_CRIT "br_netfilter: Argh!! br_nf_post_routing: "
 761                        "bad mac.raw pointer.\n");
 762                 goto print_error;
 763         }
 764 #endif
 765
 766         if (!nf_bridge)
 767                 return NF_ACCEPT;
 768
 769         if (!realoutdev)
 770                 return NF_DROP;
 771
 772         if (skb->protocol == htons(ETH_P_IP) || IS_VLAN_IP(skb))
 773                 pf = PF_INET;
 774         else
 775                 pf = PF_INET6;
 776
 777 #ifdef CONFIG_NETFILTER_DEBUG
 778         if (skb->dst == NULL) {
 779                 printk(KERN_INFO "br_netfilter post_routing: skb->dst == NULL\n");
 780                 goto print_error;
 781         }
 782 #endif
 783
 784         /* We assume any code from br_dev_queue_push_xmit onwards doesn't care
 785          * about the value of skb->pkt_type. */
 786         if (skb->pkt_type == PACKET_OTHERHOST) {
 787                 skb->pkt_type = PACKET_HOST;
 788                 nf_bridge->mask |= BRNF_PKT_TYPE;
 789         }
 790
 791         if (skb->protocol == htons(ETH_P_8021Q)) {
 792                 skb_pull(skb, VLAN_HLEN);
 793                 skb->nh.raw += VLAN_HLEN;
 794         }
 795
 796         nf_bridge_save_header(skb);
 797
 798 #if defined(CONFIG_VLAN_8021Q) || defined(CONFIG_VLAN_8021Q_MODULE)
 799         if (nf_bridge->netoutdev)
 800                 realoutdev = nf_bridge->netoutdev;
 801 #endif
 802         NF_HOOK(pf, NF_IP_POST_ROUTING, skb, NULL, realoutdev,
 803                 br_nf_dev_queue_xmit);
 804
 805         return NF_STOLEN;
 806
 807 #ifdef CONFIG_NETFILTER_DEBUG
 808 print_error:
 809         if (skb->dev != NULL) {
 810                 printk("[%s]", skb->dev->name);
 811                 if (realoutdev)
 812                         printk("[%s]", realoutdev->name);
 813         }
 814         printk(" head:%p, raw:%p, data:%p\n", skb->head, skb_mac_header(skb),
 815                skb->data);
 816         dump_stack();
 817         return NF_ACCEPT;
 818 #endif
 819 }
 820
 821 /* IP/SABOTAGE *****************************************************/
 822 /* Don't hand locally destined packets to PF_INET(6)/PRE_ROUTING
 823  * for the second time. */
 824 static unsigned int ip_sabotage_in(unsigned int hook, struct sk_buff **pskb,
 825                                    const struct net_device *in,
 826                                    const struct net_device *out,
 827                                    int (*okfn)(struct sk_buff *))
 828 {
 829         if ((*pskb)->nf_bridge &&
 830             !((*pskb)->nf_bridge->mask & BRNF_NF_BRIDGE_PREROUTING)) {
 831                 return NF_STOP;
 832         }
 833
 834         return NF_ACCEPT;
 835 }
 836
 837 /* For br_nf_local_out we need (prio = NF_BR_PRI_FIRST), to insure that innocent
 838  * PF_BRIDGE/NF_BR_LOCAL_OUT functions don't get bridged traffic as input.
 839  * For br_nf_post_routing, we need (prio = NF_BR_PRI_LAST), because
 840  * ip_refrag() can return NF_STOLEN. */
 841 static struct nf_hook_ops br_nf_ops[] = {
 842         { .hook = br_nf_pre_routing,
 843           .owner = THIS_MODULE,
 844           .pf = PF_BRIDGE,
 845           .hooknum = NF_BR_PRE_ROUTING,
 846           .priority = NF_BR_PRI_BRNF, },
 847         { .hook = br_nf_local_in,
 848           .owner = THIS_MODULE,
 849           .pf = PF_BRIDGE,
 850           .hooknum = NF_BR_LOCAL_IN,
 851           .priority = NF_BR_PRI_BRNF, },
 852         { .hook = br_nf_forward_ip,
 853           .owner = THIS_MODULE,
 854           .pf = PF_BRIDGE,
 855           .hooknum = NF_BR_FORWARD,
 856           .priority = NF_BR_PRI_BRNF - 1, },
 857         { .hook = br_nf_forward_arp,
 858           .owner = THIS_MODULE,
 859           .pf = PF_BRIDGE,
 860           .hooknum = NF_BR_FORWARD,
 861           .priority = NF_BR_PRI_BRNF, },
 862         { .hook = br_nf_local_out,
 863           .owner = THIS_MODULE,
 864           .pf = PF_BRIDGE,
 865           .hooknum = NF_BR_LOCAL_OUT,
 866           .priority = NF_BR_PRI_FIRST, },
 867         { .hook = br_nf_post_routing,
 868           .owner = THIS_MODULE,
 869           .pf = PF_BRIDGE,
 870           .hooknum = NF_BR_POST_ROUTING,
 871           .priority = NF_BR_PRI_LAST, },
 872         { .hook = ip_sabotage_in,
 873           .owner = THIS_MODULE,
 874           .pf = PF_INET,
 875           .hooknum = NF_IP_PRE_ROUTING,
 876           .priority = NF_IP_PRI_FIRST, },
 877         { .hook = ip_sabotage_in,
 878           .owner = THIS_MODULE,
 879           .pf = PF_INET6,
 880           .hooknum = NF_IP6_PRE_ROUTING,
 881           .priority = NF_IP6_PRI_FIRST, },
 882 };
 883
 884 #ifdef CONFIG_SYSCTL
 885 static
 886 int brnf_sysctl_call_tables(ctl_table * ctl, int write, struct file *filp,
 887                             void __user * buffer, size_t * lenp, loff_t * ppos)
 888 {
 889         int ret;
 890
 891         ret = proc_dointvec(ctl, write, filp, buffer, lenp, ppos);
 892
 893         if (write && *(int *)(ctl->data))
 894                 *(int *)(ctl->data) = 1;
 895         return ret;
 896 }
 897
 898 static ctl_table brnf_table[] = {
 899         {
 900                 .ctl_name       = NET_BRIDGE_NF_CALL_ARPTABLES,
 901                 .procname       = "bridge-nf-call-arptables",
 902                 .data           = &brnf_call_arptables,
 903                 .maxlen         = sizeof(int),
 904                 .mode           = 0644,
 905                 .proc_handler   = &brnf_sysctl_call_tables,
 906         },
 907         {
 908                 .ctl_name       = NET_BRIDGE_NF_CALL_IPTABLES,
 909                 .procname       = "bridge-nf-call-iptables",
 910                 .data           = &brnf_call_iptables,
 911                 .maxlen         = sizeof(int),
 912                 .mode           = 0644,
 913                 .proc_handler   = &brnf_sysctl_call_tables,
 914         },
 915         {
 916                 .ctl_name       = NET_BRIDGE_NF_CALL_IP6TABLES,
 917                 .procname       = "bridge-nf-call-ip6tables",
 918                 .data           = &brnf_call_ip6tables,
 919                 .maxlen         = sizeof(int),
 920                 .mode           = 0644,
 921                 .proc_handler   = &brnf_sysctl_call_tables,
 922         },
 923         {
 924                 .ctl_name       = NET_BRIDGE_NF_FILTER_VLAN_TAGGED,
 925                 .procname       = "bridge-nf-filter-vlan-tagged",
 926                 .data           = &brnf_filter_vlan_tagged,
 927                 .maxlen         = sizeof(int),
 928                 .mode           = 0644,
 929                 .proc_handler   = &brnf_sysctl_call_tables,
 930         },
 931         { .ctl_name = 0 }
 932 };
 933
 934 static ctl_table brnf_bridge_table[] = {
 935         {
 936                 .ctl_name       = NET_BRIDGE,
 937                 .procname       = "bridge",
 938                 .mode           = 0555,
 939                 .child          = brnf_table,
 940         },
 941         { .ctl_name = 0 }
 942 };
 943
 944 static ctl_table brnf_net_table[] = {
 945         {
 946                 .ctl_name       = CTL_NET,
 947                 .procname       = "net",
 948                 .mode           = 0555,
 949                 .child          = brnf_bridge_table,
 950         },
 951         { .ctl_name = 0 }
 952 };
 953 #endif
 954
 955 int __init br_netfilter_init(void)
 956 {
 957         int ret;
 958
 959         ret = nf_register_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
 960         if (ret < 0)
 961                 return ret;
 962 #ifdef CONFIG_SYSCTL
 963         brnf_sysctl_header = register_sysctl_table(brnf_net_table);
 964         if (brnf_sysctl_header == NULL) {
 965                 printk(KERN_WARNING
 966                        "br_netfilter: can't register to sysctl.\n");
 967                 nf_unregister_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
 968                 return -ENOMEM;
 969         }
 970 #endif
 971         printk(KERN_NOTICE "Bridge firewalling registered\n");
 972         return 0;
 973 }
 974
 975 void br_netfilter_fini(void)
 976 {
 977         nf_unregister_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
 978 #ifdef CONFIG_SYSCTL
 979         unregister_sysctl_table(brnf_sysctl_header);
 980 #endif
 981 }