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         skb_copy_from_linear_data_offset(skb, -header_size,
 133                                          skb->nf_bridge->data, header_size);
 134 }
 135
 136 /*
 137  * When forwarding bridge frames, we save a copy of the original
 138  * header before processing.
 139  */
 140 int nf_bridge_copy_header(struct sk_buff *skb)
 141 {
 142         int err;
 143         int header_size = ETH_HLEN;
 144
 145         if (skb->protocol == htons(ETH_P_8021Q))
 146                 header_size += VLAN_HLEN;
 147
 148         err = skb_cow(skb, header_size);
 149         if (err)
 150                 return err;
 151
 152         memcpy(skb->data - header_size, skb->nf_bridge->data, header_size);
 153
 154         if (skb->protocol == htons(ETH_P_8021Q))
 155                 __skb_push(skb, VLAN_HLEN);
 156         return 0;
 157 }
 158
 159 /* PF_BRIDGE/PRE_ROUTING *********************************************/
 160 /* Undo the changes made for ip6tables PREROUTING and continue the
 161  * bridge PRE_ROUTING hook. */
 162 static int br_nf_pre_routing_finish_ipv6(struct sk_buff *skb)
 163 {
 164         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 165
 166         if (nf_bridge->mask & BRNF_PKT_TYPE) {
 167                 skb->pkt_type = PACKET_OTHERHOST;
 168                 nf_bridge->mask ^= BRNF_PKT_TYPE;
 169         }
 170         nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
 171
 172         skb->dst = (struct dst_entry *)&__fake_rtable;
 173         dst_hold(skb->dst);
 174
 175         skb->dev = nf_bridge->physindev;
 176         if (skb->protocol == htons(ETH_P_8021Q)) {
 177                 skb_push(skb, VLAN_HLEN);
 178                 skb->network_header -= VLAN_HLEN;
 179         }
 180         NF_HOOK_THRESH(PF_BRIDGE, NF_BR_PRE_ROUTING, skb, skb->dev, NULL,
 181                        br_handle_frame_finish, 1);
 182
 183         return 0;
 184 }
 185
 186 static void __br_dnat_complain(void)
 187 {
 188         static unsigned long last_complaint;
 189
 190         if (jiffies - last_complaint >= 5 * HZ) {
 191                 printk(KERN_WARNING "Performing cross-bridge DNAT requires IP "
 192                        "forwarding to be enabled\n");
 193                 last_complaint = jiffies;
 194         }
 195 }
 196
 197 /* This requires some explaining. If DNAT has taken place,
 198  * we will need to fix up the destination Ethernet address,
 199  * and this is a tricky process.
 200  *
 201  * There are two cases to consider:
 202  * 1. The packet was DNAT'ed to a device in the same bridge
 203  *    port group as it was received on. We can still bridge
 204  *    the packet.
 205  * 2. The packet was DNAT'ed to a different device, either
 206  *    a non-bridged device or another bridge port group.
 207  *    The packet will need to be routed.
 208  *
 209  * The correct way of distinguishing between these two cases is to
 210  * call ip_route_input() and to look at skb->dst->dev, which is
 211  * changed to the destination device if ip_route_input() succeeds.
 212  *
 213  * Let us first consider the case that ip_route_input() succeeds:
 214  *
 215  * If skb->dst->dev equals the logical bridge device the packet
 216  * came in on, we can consider this bridging. We then call
 217  * skb->dst->output() which will make the packet enter br_nf_local_out()
 218  * not much later. In that function it is assured that the iptables
 219  * FORWARD chain is traversed for the packet.
 220  *
 221  * Otherwise, the packet is considered to be routed and we just
 222  * change the destination MAC address so that the packet will
 223  * later be passed up to the IP stack to be routed. For a redirected
 224  * packet, ip_route_input() will give back the localhost as output device,
 225  * which differs from the bridge device.
 226  *
 227  * Let us now consider the case that ip_route_input() fails:
 228  *
 229  * This can be because the destination address is martian, in which case
 230  * the packet will be dropped.
 231  * After a "echo '0' > /proc/sys/net/ipv4/ip_forward" ip_route_input()
 232  * will fail, while __ip_route_output_key() will return success. The source
 233  * address for __ip_route_output_key() is set to zero, so __ip_route_output_key
 234  * thinks we're handling a locally generated packet and won't care
 235  * if IP forwarding is allowed. We send a warning message to the users's
 236  * log telling her to put IP forwarding on.
 237  *
 238  * ip_route_input() will also fail if there is no route available.
 239  * In that case we just drop the packet.
 240  *
 241  * --Lennert, 20020411
 242  * --Bart, 20020416 (updated)
 243  * --Bart, 20021007 (updated)
 244  * --Bart, 20062711 (updated) */
 245 static int br_nf_pre_routing_finish_bridge(struct sk_buff *skb)
 246 {
 247         if (skb->pkt_type == PACKET_OTHERHOST) {
 248                 skb->pkt_type = PACKET_HOST;
 249                 skb->nf_bridge->mask |= BRNF_PKT_TYPE;
 250         }
 251         skb->nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
 252
 253         skb->dev = bridge_parent(skb->dev);
 254         if (!skb->dev)
 255                 kfree_skb(skb);
 256         else {
 257                 if (skb->protocol == htons(ETH_P_8021Q)) {
 258                         skb_pull(skb, VLAN_HLEN);
 259                         skb->network_header += VLAN_HLEN;
 260                 }
 261                 skb->dst->output(skb);
 262         }
 263         return 0;
 264 }
 265
 266 static int br_nf_pre_routing_finish(struct sk_buff *skb)
 267 {
 268         struct net_device *dev = skb->dev;
 269         struct iphdr *iph = ip_hdr(skb);
 270         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 271         int err;
 272
 273         if (nf_bridge->mask & BRNF_PKT_TYPE) {
 274                 skb->pkt_type = PACKET_OTHERHOST;
 275                 nf_bridge->mask ^= BRNF_PKT_TYPE;
 276         }
 277         nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
 278         if (dnat_took_place(skb)) {
 279                 if ((err = ip_route_input(skb, iph->daddr, iph->saddr, iph->tos, dev))) {
 280                         struct rtable *rt;
 281                         struct flowi fl = {
 282                                 .nl_u = {
 283                                         .ip4_u = {
 284                                                  .daddr = iph->daddr,
 285                                                  .saddr = 0,
 286                                                  .tos = RT_TOS(iph->tos) },
 287                                 },
 288                                 .proto = 0,
 289                         };
 290                         struct in_device *in_dev = in_dev_get(dev);
 291
 292                         /* If err equals -EHOSTUNREACH the error is due to a
 293                          * martian destination or due to the fact that
 294                          * forwarding is disabled. For most martian packets,
 295                          * ip_route_output_key() will fail. It won't fail for 2 types of
 296                          * martian destinations: loopback destinations and destination
 297                          * 0.0.0.0. In both cases the packet will be dropped because the
 298                          * destination is the loopback device and not the bridge. */
 299                         if (err != -EHOSTUNREACH || !in_dev || IN_DEV_FORWARD(in_dev))
 300                                 goto free_skb;
 301
 302                         if (!ip_route_output_key(&rt, &fl)) {
 303                                 /* - Bridged-and-DNAT'ed traffic doesn't
 304                                  *   require ip_forwarding. */
 305                                 if (((struct dst_entry *)rt)->dev == dev) {
 306                                         skb->dst = (struct dst_entry *)rt;
 307                                         goto bridged_dnat;
 308                                 }
 309                                 /* we are sure that forwarding is disabled, so printing
 310                                  * this message is no problem. Note that the packet could
 311                                  * still have a martian destination address, in which case
 312                                  * the packet could be dropped even if forwarding were enabled */
 313                                 __br_dnat_complain();
 314                                 dst_release((struct dst_entry *)rt);
 315                         }
 316 free_skb:
 317                         kfree_skb(skb);
 318                         return 0;
 319                 } else {
 320                         if (skb->dst->dev == dev) {
 321 bridged_dnat:
 322                                 /* Tell br_nf_local_out this is a
 323                                  * bridged frame */
 324                                 nf_bridge->mask |= BRNF_BRIDGED_DNAT;
 325                                 skb->dev = nf_bridge->physindev;
 326                                 if (skb->protocol ==
 327                                     htons(ETH_P_8021Q)) {
 328                                         skb_push(skb, VLAN_HLEN);
 329                                         skb->network_header -= VLAN_HLEN;
 330                                 }
 331                                 NF_HOOK_THRESH(PF_BRIDGE, NF_BR_PRE_ROUTING,
 332                                                skb, skb->dev, NULL,
 333                                                br_nf_pre_routing_finish_bridge,
 334                                                1);
 335                                 return 0;
 336                         }
 337                         memcpy(eth_hdr(skb)->h_dest, dev->dev_addr, ETH_ALEN);
 338                         skb->pkt_type = PACKET_HOST;
 339                 }
 340         } else {
 341                 skb->dst = (struct dst_entry *)&__fake_rtable;
 342                 dst_hold(skb->dst);
 343         }
 344
 345         skb->dev = nf_bridge->physindev;
 346         if (skb->protocol == htons(ETH_P_8021Q)) {
 347                 skb_push(skb, VLAN_HLEN);
 348                 skb->network_header -= VLAN_HLEN;
 349         }
 350         NF_HOOK_THRESH(PF_BRIDGE, NF_BR_PRE_ROUTING, skb, skb->dev, NULL,
 351                        br_handle_frame_finish, 1);
 352
 353         return 0;
 354 }
 355
 356 /* Some common code for IPv4/IPv6 */
 357 static struct net_device *setup_pre_routing(struct sk_buff *skb)
 358 {
 359         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 360
 361         if (skb->pkt_type == PACKET_OTHERHOST) {
 362                 skb->pkt_type = PACKET_HOST;
 363                 nf_bridge->mask |= BRNF_PKT_TYPE;
 364         }
 365
 366         nf_bridge->mask |= BRNF_NF_BRIDGE_PREROUTING;
 367         nf_bridge->physindev = skb->dev;
 368         skb->dev = bridge_parent(skb->dev);
 369
 370         return skb->dev;
 371 }
 372
 373 /* We only check the length. A bridge shouldn't do any hop-by-hop stuff anyway */
 374 static int check_hbh_len(struct sk_buff *skb)
 375 {
 376         unsigned char *raw = (u8 *)(ipv6_hdr(skb) + 1);
 377         u32 pkt_len;
 378         const unsigned char *nh = skb_network_header(skb);
 379         int off = raw - nh;
 380         int len = (raw[1] + 1) << 3;
 381
 382         if ((raw + len) - skb->data > skb_headlen(skb))
 383                 goto bad;
 384
 385         off += 2;
 386         len -= 2;
 387
 388         while (len > 0) {
 389                 int optlen = nh[off + 1] + 2;
 390
 391                 switch (nh[off]) {
 392                 case IPV6_TLV_PAD0:
 393                         optlen = 1;
 394                         break;
 395
 396                 case IPV6_TLV_PADN:
 397                         break;
 398
 399                 case IPV6_TLV_JUMBO:
 400                         if (nh[off + 1] != 4 || (off & 3) != 2)
 401                                 goto bad;
 402                         pkt_len = ntohl(*(__be32 *) (nh + off + 2));
 403                         if (pkt_len <= IPV6_MAXPLEN ||
 404                             ipv6_hdr(skb)->payload_len)
 405                                 goto bad;
 406                         if (pkt_len > skb->len - sizeof(struct ipv6hdr))
 407                                 goto bad;
 408                         if (pskb_trim_rcsum(skb,
 409                                             pkt_len + sizeof(struct ipv6hdr)))
 410                                 goto bad;
 411                         nh = skb_network_header(skb);
 412                         break;
 413                 default:
 414                         if (optlen > len)
 415                                 goto bad;
 416                         break;
 417                 }
 418                 off += optlen;
 419                 len -= optlen;
 420         }
 421         if (len == 0)
 422                 return 0;
 423 bad:
 424         return -1;
 425
 426 }
 427
 428 /* Replicate the checks that IPv6 does on packet reception and pass the packet
 429  * to ip6tables, which doesn't support NAT, so things are fairly simple. */
 430 static unsigned int br_nf_pre_routing_ipv6(unsigned int hook,
 431                                            struct sk_buff *skb,
 432                                            const struct net_device *in,
 433                                            const struct net_device *out,
 434                                            int (*okfn)(struct sk_buff *))
 435 {
 436         struct ipv6hdr *hdr;
 437         u32 pkt_len;
 438
 439         if (skb->len < sizeof(struct ipv6hdr))
 440                 goto inhdr_error;
 441
 442         if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
 443                 goto inhdr_error;
 444
 445         hdr = ipv6_hdr(skb);
 446
 447         if (hdr->version != 6)
 448                 goto inhdr_error;
 449
 450         pkt_len = ntohs(hdr->payload_len);
 451
 452         if (pkt_len || hdr->nexthdr != NEXTHDR_HOP) {
 453                 if (pkt_len + sizeof(struct ipv6hdr) > skb->len)
 454                         goto inhdr_error;
 455                 if (pskb_trim_rcsum(skb, pkt_len + sizeof(struct ipv6hdr)))
 456                         goto inhdr_error;
 457         }
 458         if (hdr->nexthdr == NEXTHDR_HOP && check_hbh_len(skb))
 459                 goto inhdr_error;
 460
 461         nf_bridge_put(skb->nf_bridge);
 462         if (!nf_bridge_alloc(skb))
 463                 return NF_DROP;
 464         if (!setup_pre_routing(skb))
 465                 return NF_DROP;
 466
 467         NF_HOOK(PF_INET6, NF_IP6_PRE_ROUTING, skb, skb->dev, NULL,
 468                 br_nf_pre_routing_finish_ipv6);
 469
 470         return NF_STOLEN;
 471
 472 inhdr_error:
 473         return NF_DROP;
 474 }
 475
 476 /* Direct IPv6 traffic to br_nf_pre_routing_ipv6.
 477  * Replicate the checks that IPv4 does on packet reception.
 478  * Set skb->dev to the bridge device (i.e. parent of the
 479  * receiving device) to make netfilter happy, the REDIRECT
 480  * target in particular.  Save the original destination IP
 481  * address to be able to detect DNAT afterwards. */
 482 static unsigned int br_nf_pre_routing(unsigned int hook, struct sk_buff **pskb,
 483                                       const struct net_device *in,
 484                                       const struct net_device *out,
 485                                       int (*okfn)(struct sk_buff *))
 486 {
 487         struct iphdr *iph;
 488         __u32 len;
 489         struct sk_buff *skb = *pskb;
 490
 491         if (skb->protocol == htons(ETH_P_IPV6) || IS_VLAN_IPV6(skb)) {
 492 #ifdef CONFIG_SYSCTL
 493                 if (!brnf_call_ip6tables)
 494                         return NF_ACCEPT;
 495 #endif
 496                 if ((skb = skb_share_check(*pskb, GFP_ATOMIC)) == NULL)
 497                         goto out;
 498
 499                 if (skb->protocol == htons(ETH_P_8021Q)) {
 500                         skb_pull_rcsum(skb, VLAN_HLEN);
 501                         skb->network_header += VLAN_HLEN;
 502                 }
 503                 return br_nf_pre_routing_ipv6(hook, skb, in, out, okfn);
 504         }
 505 #ifdef CONFIG_SYSCTL
 506         if (!brnf_call_iptables)
 507                 return NF_ACCEPT;
 508 #endif
 509
 510         if (skb->protocol != htons(ETH_P_IP) && !IS_VLAN_IP(skb))
 511                 return NF_ACCEPT;
 512
 513         if ((skb = skb_share_check(*pskb, GFP_ATOMIC)) == NULL)
 514                 goto out;
 515
 516         if (skb->protocol == htons(ETH_P_8021Q)) {
 517                 skb_pull_rcsum(skb, VLAN_HLEN);
 518                 skb->network_header += VLAN_HLEN;
 519         }
 520
 521         if (!pskb_may_pull(skb, sizeof(struct iphdr)))
 522                 goto inhdr_error;
 523
 524         iph = ip_hdr(skb);
 525         if (iph->ihl < 5 || iph->version != 4)
 526                 goto inhdr_error;
 527
 528         if (!pskb_may_pull(skb, 4 * iph->ihl))
 529                 goto inhdr_error;
 530
 531         iph = ip_hdr(skb);
 532         if (ip_fast_csum((__u8 *) iph, iph->ihl) != 0)
 533                 goto inhdr_error;
 534
 535         len = ntohs(iph->tot_len);
 536         if (skb->len < len || len < 4 * iph->ihl)
 537                 goto inhdr_error;
 538
 539         pskb_trim_rcsum(skb, len);
 540
 541         nf_bridge_put(skb->nf_bridge);
 542         if (!nf_bridge_alloc(skb))
 543                 return NF_DROP;
 544         if (!setup_pre_routing(skb))
 545                 return NF_DROP;
 546         store_orig_dstaddr(skb);
 547
 548         NF_HOOK(PF_INET, NF_IP_PRE_ROUTING, skb, skb->dev, NULL,
 549                 br_nf_pre_routing_finish);
 550
 551         return NF_STOLEN;
 552
 553 inhdr_error:
 554 //      IP_INC_STATS_BH(IpInHdrErrors);
 555 out:
 556         return NF_DROP;
 557 }
 558
 559
 560 /* PF_BRIDGE/LOCAL_IN ************************************************/
 561 /* The packet is locally destined, which requires a real
 562  * dst_entry, so detach the fake one.  On the way up, the
 563  * packet would pass through PRE_ROUTING again (which already
 564  * took place when the packet entered the bridge), but we
 565  * register an IPv4 PRE_ROUTING 'sabotage' hook that will
 566  * prevent this from happening. */
 567 static unsigned int br_nf_local_in(unsigned int hook, struct sk_buff **pskb,
 568                                    const struct net_device *in,
 569                                    const struct net_device *out,
 570                                    int (*okfn)(struct sk_buff *))
 571 {
 572         struct sk_buff *skb = *pskb;
 573
 574         if (skb->dst == (struct dst_entry *)&__fake_rtable) {
 575                 dst_release(skb->dst);
 576                 skb->dst = NULL;
 577         }
 578
 579         return NF_ACCEPT;
 580 }
 581
 582 /* PF_BRIDGE/FORWARD *************************************************/
 583 static int br_nf_forward_finish(struct sk_buff *skb)
 584 {
 585         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 586         struct net_device *in;
 587
 588         if (skb->protocol != htons(ETH_P_ARP) && !IS_VLAN_ARP(skb)) {
 589                 in = nf_bridge->physindev;
 590                 if (nf_bridge->mask & BRNF_PKT_TYPE) {
 591                         skb->pkt_type = PACKET_OTHERHOST;
 592                         nf_bridge->mask ^= BRNF_PKT_TYPE;
 593                 }
 594         } else {
 595                 in = *((struct net_device **)(skb->cb));
 596         }
 597         if (skb->protocol == htons(ETH_P_8021Q)) {
 598                 skb_push(skb, VLAN_HLEN);
 599                 skb->network_header -= VLAN_HLEN;
 600         }
 601         NF_HOOK_THRESH(PF_BRIDGE, NF_BR_FORWARD, skb, in,
 602                        skb->dev, br_forward_finish, 1);
 603         return 0;
 604 }
 605
 606 /* This is the 'purely bridged' case.  For IP, we pass the packet to
 607  * netfilter with indev and outdev set to the bridge device,
 608  * but we are still able to filter on the 'real' indev/outdev
 609  * because of the physdev module. For ARP, indev and outdev are the
 610  * bridge ports. */
 611 static unsigned int br_nf_forward_ip(unsigned int hook, struct sk_buff **pskb,
 612                                      const struct net_device *in,
 613                                      const struct net_device *out,
 614                                      int (*okfn)(struct sk_buff *))
 615 {
 616         struct sk_buff *skb = *pskb;
 617         struct nf_bridge_info *nf_bridge;
 618         struct net_device *parent;
 619         int pf;
 620
 621         if (!skb->nf_bridge)
 622                 return NF_ACCEPT;
 623
 624         parent = bridge_parent(out);
 625         if (!parent)
 626                 return NF_DROP;
 627
 628         if (skb->protocol == htons(ETH_P_IP) || IS_VLAN_IP(skb))
 629                 pf = PF_INET;
 630         else
 631                 pf = PF_INET6;
 632
 633         if (skb->protocol == htons(ETH_P_8021Q)) {
 634                 skb_pull(*pskb, VLAN_HLEN);
 635                 (*pskb)->network_header += VLAN_HLEN;
 636         }
 637
 638         nf_bridge = skb->nf_bridge;
 639         if (skb->pkt_type == PACKET_OTHERHOST) {
 640                 skb->pkt_type = PACKET_HOST;
 641                 nf_bridge->mask |= BRNF_PKT_TYPE;
 642         }
 643
 644         /* The physdev module checks on this */
 645         nf_bridge->mask |= BRNF_BRIDGED;
 646         nf_bridge->physoutdev = skb->dev;
 647
 648         NF_HOOK(pf, NF_IP_FORWARD, skb, bridge_parent(in), parent,
 649                 br_nf_forward_finish);
 650
 651         return NF_STOLEN;
 652 }
 653
 654 static unsigned int br_nf_forward_arp(unsigned int hook, struct sk_buff **pskb,
 655                                       const struct net_device *in,
 656                                       const struct net_device *out,
 657                                       int (*okfn)(struct sk_buff *))
 658 {
 659         struct sk_buff *skb = *pskb;
 660         struct net_device **d = (struct net_device **)(skb->cb);
 661
 662 #ifdef CONFIG_SYSCTL
 663         if (!brnf_call_arptables)
 664                 return NF_ACCEPT;
 665 #endif
 666
 667         if (skb->protocol != htons(ETH_P_ARP)) {
 668                 if (!IS_VLAN_ARP(skb))
 669                         return NF_ACCEPT;
 670                 skb_pull(*pskb, VLAN_HLEN);
 671                 (*pskb)->network_header += VLAN_HLEN;
 672         }
 673
 674         if (arp_hdr(skb)->ar_pln != 4) {
 675                 if (IS_VLAN_ARP(skb)) {
 676                         skb_push(*pskb, VLAN_HLEN);
 677                         (*pskb)->network_header -= VLAN_HLEN;
 678                 }
 679                 return NF_ACCEPT;
 680         }
 681         *d = (struct net_device *)in;
 682         NF_HOOK(NF_ARP, NF_ARP_FORWARD, skb, (struct net_device *)in,
 683                 (struct net_device *)out, br_nf_forward_finish);
 684
 685         return NF_STOLEN;
 686 }
 687
 688 /* PF_BRIDGE/LOCAL_OUT ***********************************************
 689  *
 690  * This function sees both locally originated IP packets and forwarded
 691  * IP packets (in both cases the destination device is a bridge
 692  * device). It also sees bridged-and-DNAT'ed packets.
 693  *
 694  * If (nf_bridge->mask & BRNF_BRIDGED_DNAT) then the packet is bridged
 695  * and we fake the PF_BRIDGE/FORWARD hook. The function br_nf_forward()
 696  * will then fake the PF_INET/FORWARD hook. br_nf_local_out() has priority
 697  * NF_BR_PRI_FIRST, so no relevant PF_BRIDGE/INPUT functions have been nor
 698  * will be executed.
 699  */
 700 static unsigned int br_nf_local_out(unsigned int hook, struct sk_buff **pskb,
 701                                     const struct net_device *in,
 702                                     const struct net_device *out,
 703                                     int (*okfn)(struct sk_buff *))
 704 {
 705         struct net_device *realindev;
 706         struct sk_buff *skb = *pskb;
 707         struct nf_bridge_info *nf_bridge;
 708
 709         if (!skb->nf_bridge)
 710                 return NF_ACCEPT;
 711
 712         nf_bridge = skb->nf_bridge;
 713         if (!(nf_bridge->mask & BRNF_BRIDGED_DNAT))
 714                 return NF_ACCEPT;
 715
 716         /* Bridged, take PF_BRIDGE/FORWARD.
 717          * (see big note in front of br_nf_pre_routing_finish) */
 718         nf_bridge->physoutdev = skb->dev;
 719         realindev = nf_bridge->physindev;
 720
 721         if (nf_bridge->mask & BRNF_PKT_TYPE) {
 722                 skb->pkt_type = PACKET_OTHERHOST;
 723                 nf_bridge->mask ^= BRNF_PKT_TYPE;
 724         }
 725         if (skb->protocol == htons(ETH_P_8021Q)) {
 726                 skb_push(skb, VLAN_HLEN);
 727                 skb->network_header -= VLAN_HLEN;
 728         }
 729
 730         NF_HOOK(PF_BRIDGE, NF_BR_FORWARD, skb, realindev, skb->dev,
 731                 br_forward_finish);
 732         return NF_STOLEN;
 733 }
 734
 735 static int br_nf_dev_queue_xmit(struct sk_buff *skb)
 736 {
 737         if (skb->protocol == htons(ETH_P_IP) &&
 738             skb->len > skb->dev->mtu &&
 739             !skb_is_gso(skb))
 740                 return ip_fragment(skb, br_dev_queue_push_xmit);
 741         else
 742                 return br_dev_queue_push_xmit(skb);
 743 }
 744
 745 /* PF_BRIDGE/POST_ROUTING ********************************************/
 746 static unsigned int br_nf_post_routing(unsigned int hook, struct sk_buff **pskb,
 747                                        const struct net_device *in,
 748                                        const struct net_device *out,
 749                                        int (*okfn)(struct sk_buff *))
 750 {
 751         struct sk_buff *skb = *pskb;
 752         struct nf_bridge_info *nf_bridge = (*pskb)->nf_bridge;
 753         struct net_device *realoutdev = bridge_parent(skb->dev);
 754         int pf;
 755
 756 #ifdef CONFIG_NETFILTER_DEBUG
 757         /* Be very paranoid. This probably won't happen anymore, but let's
 758          * keep the check just to be sure... */
 759         if (skb_mac_header(skb) < skb->head ||
 760             skb_mac_header(skb) + ETH_HLEN > skb->data) {
 761                 printk(KERN_CRIT "br_netfilter: Argh!! br_nf_post_routing: "
 762                        "bad mac.raw pointer.\n");
 763                 goto print_error;
 764         }
 765 #endif
 766
 767         if (!nf_bridge)
 768                 return NF_ACCEPT;
 769
 770         if (!realoutdev)
 771                 return NF_DROP;
 772
 773         if (skb->protocol == htons(ETH_P_IP) || IS_VLAN_IP(skb))
 774                 pf = PF_INET;
 775         else
 776                 pf = PF_INET6;
 777
 778 #ifdef CONFIG_NETFILTER_DEBUG
 779         if (skb->dst == NULL) {
 780                 printk(KERN_INFO "br_netfilter post_routing: skb->dst == NULL\n");
 781                 goto print_error;
 782         }
 783 #endif
 784
 785         /* We assume any code from br_dev_queue_push_xmit onwards doesn't care
 786          * about the value of skb->pkt_type. */
 787         if (skb->pkt_type == PACKET_OTHERHOST) {
 788                 skb->pkt_type = PACKET_HOST;
 789                 nf_bridge->mask |= BRNF_PKT_TYPE;
 790         }
 791
 792         if (skb->protocol == htons(ETH_P_8021Q)) {
 793                 skb_pull(skb, VLAN_HLEN);
 794                 skb->network_header += VLAN_HLEN;
 795         }
 796
 797         nf_bridge_save_header(skb);
 798
 799 #if defined(CONFIG_VLAN_8021Q) || defined(CONFIG_VLAN_8021Q_MODULE)
 800         if (nf_bridge->netoutdev)
 801                 realoutdev = nf_bridge->netoutdev;
 802 #endif
 803         NF_HOOK(pf, NF_IP_POST_ROUTING, skb, NULL, realoutdev,
 804                 br_nf_dev_queue_xmit);
 805
 806         return NF_STOLEN;
 807
 808 #ifdef CONFIG_NETFILTER_DEBUG
 809 print_error:
 810         if (skb->dev != NULL) {
 811                 printk("[%s]", skb->dev->name);
 812                 if (realoutdev)
 813                         printk("[%s]", realoutdev->name);
 814         }
 815         printk(" head:%p, raw:%p, data:%p\n", skb->head, skb_mac_header(skb),
 816                skb->data);
 817         dump_stack();
 818         return NF_ACCEPT;
 819 #endif
 820 }
 821
 822 /* IP/SABOTAGE *****************************************************/
 823 /* Don't hand locally destined packets to PF_INET(6)/PRE_ROUTING
 824  * for the second time. */
 825 static unsigned int ip_sabotage_in(unsigned int hook, struct sk_buff **pskb,
 826                                    const struct net_device *in,
 827                                    const struct net_device *out,
 828                                    int (*okfn)(struct sk_buff *))
 829 {
 830         if ((*pskb)->nf_bridge &&
 831             !((*pskb)->nf_bridge->mask & BRNF_NF_BRIDGE_PREROUTING)) {
 832                 return NF_STOP;
 833         }
 834
 835         return NF_ACCEPT;
 836 }
 837
 838 /* For br_nf_local_out we need (prio = NF_BR_PRI_FIRST), to insure that innocent
 839  * PF_BRIDGE/NF_BR_LOCAL_OUT functions don't get bridged traffic as input.
 840  * For br_nf_post_routing, we need (prio = NF_BR_PRI_LAST), because
 841  * ip_refrag() can return NF_STOLEN. */
 842 static struct nf_hook_ops br_nf_ops[] = {
 843         { .hook = br_nf_pre_routing,
 844           .owner = THIS_MODULE,
 845           .pf = PF_BRIDGE,
 846           .hooknum = NF_BR_PRE_ROUTING,
 847           .priority = NF_BR_PRI_BRNF, },
 848         { .hook = br_nf_local_in,
 849           .owner = THIS_MODULE,
 850           .pf = PF_BRIDGE,
 851           .hooknum = NF_BR_LOCAL_IN,
 852           .priority = NF_BR_PRI_BRNF, },
 853         { .hook = br_nf_forward_ip,
 854           .owner = THIS_MODULE,
 855           .pf = PF_BRIDGE,
 856           .hooknum = NF_BR_FORWARD,
 857           .priority = NF_BR_PRI_BRNF - 1, },
 858         { .hook = br_nf_forward_arp,
 859           .owner = THIS_MODULE,
 860           .pf = PF_BRIDGE,
 861           .hooknum = NF_BR_FORWARD,
 862           .priority = NF_BR_PRI_BRNF, },
 863         { .hook = br_nf_local_out,
 864           .owner = THIS_MODULE,
 865           .pf = PF_BRIDGE,
 866           .hooknum = NF_BR_LOCAL_OUT,
 867           .priority = NF_BR_PRI_FIRST, },
 868         { .hook = br_nf_post_routing,
 869           .owner = THIS_MODULE,
 870           .pf = PF_BRIDGE,
 871           .hooknum = NF_BR_POST_ROUTING,
 872           .priority = NF_BR_PRI_LAST, },
 873         { .hook = ip_sabotage_in,
 874           .owner = THIS_MODULE,
 875           .pf = PF_INET,
 876           .hooknum = NF_IP_PRE_ROUTING,
 877           .priority = NF_IP_PRI_FIRST, },
 878         { .hook = ip_sabotage_in,
 879           .owner = THIS_MODULE,
 880           .pf = PF_INET6,
 881           .hooknum = NF_IP6_PRE_ROUTING,
 882           .priority = NF_IP6_PRI_FIRST, },
 883 };
 884
 885 #ifdef CONFIG_SYSCTL
 886 static
 887 int brnf_sysctl_call_tables(ctl_table * ctl, int write, struct file *filp,
 888                             void __user * buffer, size_t * lenp, loff_t * ppos)
 889 {
 890         int ret;
 891
 892         ret = proc_dointvec(ctl, write, filp, buffer, lenp, ppos);
 893
 894         if (write && *(int *)(ctl->data))
 895                 *(int *)(ctl->data) = 1;
 896         return ret;
 897 }
 898
 899 static ctl_table brnf_table[] = {
 900         {
 901                 .ctl_name       = NET_BRIDGE_NF_CALL_ARPTABLES,
 902                 .procname       = "bridge-nf-call-arptables",
 903                 .data           = &brnf_call_arptables,
 904                 .maxlen         = sizeof(int),
 905                 .mode           = 0644,
 906                 .proc_handler   = &brnf_sysctl_call_tables,
 907         },
 908         {
 909                 .ctl_name       = NET_BRIDGE_NF_CALL_IPTABLES,
 910                 .procname       = "bridge-nf-call-iptables",
 911                 .data           = &brnf_call_iptables,
 912                 .maxlen         = sizeof(int),
 913                 .mode           = 0644,
 914                 .proc_handler   = &brnf_sysctl_call_tables,
 915         },
 916         {
 917                 .ctl_name       = NET_BRIDGE_NF_CALL_IP6TABLES,
 918                 .procname       = "bridge-nf-call-ip6tables",
 919                 .data           = &brnf_call_ip6tables,
 920                 .maxlen         = sizeof(int),
 921                 .mode           = 0644,
 922                 .proc_handler   = &brnf_sysctl_call_tables,
 923         },
 924         {
 925                 .ctl_name       = NET_BRIDGE_NF_FILTER_VLAN_TAGGED,
 926                 .procname       = "bridge-nf-filter-vlan-tagged",
 927                 .data           = &brnf_filter_vlan_tagged,
 928                 .maxlen         = sizeof(int),
 929                 .mode           = 0644,
 930                 .proc_handler   = &brnf_sysctl_call_tables,
 931         },
 932         { .ctl_name = 0 }
 933 };
 934
 935 static ctl_table brnf_bridge_table[] = {
 936         {
 937                 .ctl_name       = NET_BRIDGE,
 938                 .procname       = "bridge",
 939                 .mode           = 0555,
 940                 .child          = brnf_table,
 941         },
 942         { .ctl_name = 0 }
 943 };
 944
 945 static ctl_table brnf_net_table[] = {
 946         {
 947                 .ctl_name       = CTL_NET,
 948                 .procname       = "net",
 949                 .mode           = 0555,
 950                 .child          = brnf_bridge_table,
 951         },
 952         { .ctl_name = 0 }
 953 };
 954 #endif
 955
 956 int __init br_netfilter_init(void)
 957 {
 958         int ret;
 959
 960         ret = nf_register_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
 961         if (ret < 0)
 962                 return ret;
 963 #ifdef CONFIG_SYSCTL
 964         brnf_sysctl_header = register_sysctl_table(brnf_net_table);
 965         if (brnf_sysctl_header == NULL) {
 966                 printk(KERN_WARNING
 967                        "br_netfilter: can't register to sysctl.\n");
 968                 nf_unregister_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
 969                 return -ENOMEM;
 970         }
 971 #endif
 972         printk(KERN_NOTICE "Bridge firewalling registered\n");
 973         return 0;
 974 }
 975
 976 void br_netfilter_fini(void)
 977 {
 978         nf_unregister_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
 979 #ifdef CONFIG_SYSCTL
 980         unregister_sysctl_table(brnf_sysctl_header);
 981 #endif
 982 }