net/bridge/br_netfilter.c

   1 /*
   2  *      Handle firewalling
   3  *      Linux ethernet bridge
   4  *
   5  *      Authors:
   6  *      Lennert Buytenhek               <buytenh@gnu.org>
   7  *      Bart De Schuymer (maintainer)   <bdschuym@pandora.be>
   8  *
   9  *      Changes:
  10  *      Apr 29 2003: physdev module support (bdschuym)
  11  *      Jun 19 2003: let arptables see bridged ARP traffic (bdschuym)
  12  *      Oct 06 2003: filter encapsulated IP/ARP VLAN traffic on untagged bridge
  13  *                   (bdschuym)
  14  *      Sep 01 2004: add IPv6 filtering (bdschuym)
  15  *
  16  *      This program is free software; you can redistribute it and/or
  17  *      modify it under the terms of the GNU General Public License
  18  *      as published by the Free Software Foundation; either version
  19  *      2 of the License, or (at your option) any later version.
  20  *
  21  *      Lennert dedicates this file to Kerstin Wurdinger.
  22  */
  23
  24 #include <linux/module.h>
  25 #include <linux/kernel.h>
  26 #include <linux/ip.h>
  27 #include <linux/netdevice.h>
  28 #include <linux/skbuff.h>
  29 #include <linux/if_arp.h>
  30 #include <linux/if_ether.h>
  31 #include <linux/if_vlan.h>
  32 #include <linux/if_pppox.h>
  33 #include <linux/ppp_defs.h>
  34 #include <linux/netfilter_bridge.h>
  35 #include <linux/netfilter_ipv4.h>
  36 #include <linux/netfilter_ipv6.h>
  37 #include <linux/netfilter_arp.h>
  38 #include <linux/in_route.h>
  39 #include <linux/inetdevice.h>
  40
  41 #include <net/ip.h>
  42 #include <net/ipv6.h>
  43 #include <net/route.h>
  44
  45 #include <asm/uaccess.h>
  46 #include "br_private.h"
  47 #ifdef CONFIG_SYSCTL
  48 #include <linux/sysctl.h>
  49 #endif
  50
  51 #define skb_origaddr(skb)        (((struct bridge_skb_cb *) \
  52                                  (skb->nf_bridge->data))->daddr.ipv4)
  53 #define store_orig_dstaddr(skb)  (skb_origaddr(skb) = ip_hdr(skb)->daddr)
  54 #define dnat_took_place(skb)     (skb_origaddr(skb) != ip_hdr(skb)->daddr)
  55
  56 #ifdef CONFIG_SYSCTL
  57 static struct ctl_table_header *brnf_sysctl_header;
  58 static int brnf_call_iptables __read_mostly = 1;
  59 static int brnf_call_ip6tables __read_mostly = 1;
  60 static int brnf_call_arptables __read_mostly = 1;
  61 static int brnf_filter_vlan_tagged __read_mostly = 1;
  62 static int brnf_filter_pppoe_tagged __read_mostly = 1;
  63 #else
  64 #define brnf_filter_vlan_tagged 1
  65 #define brnf_filter_pppoe_tagged 1
  66 #endif
  67
  68 static inline __be16 vlan_proto(const struct sk_buff *skb)
  69 {
  70         return vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
  71 }
  72
  73 #define IS_VLAN_IP(skb) \
  74         (skb->protocol == htons(ETH_P_8021Q) && \
  75          vlan_proto(skb) == htons(ETH_P_IP) &&  \
  76          brnf_filter_vlan_tagged)
  77
  78 #define IS_VLAN_IPV6(skb) \
  79         (skb->protocol == htons(ETH_P_8021Q) && \
  80          vlan_proto(skb) == htons(ETH_P_IPV6) &&\
  81          brnf_filter_vlan_tagged)
  82
  83 #define IS_VLAN_ARP(skb) \
  84         (skb->protocol == htons(ETH_P_8021Q) && \
  85          vlan_proto(skb) == htons(ETH_P_ARP) && \
  86          brnf_filter_vlan_tagged)
  87
  88 static inline __be16 pppoe_proto(const struct sk_buff *skb)
  89 {
  90         return *((__be16 *)(skb_mac_header(skb) + ETH_HLEN +
  91                             sizeof(struct pppoe_hdr)));
  92 }
  93
  94 #define IS_PPPOE_IP(skb) \
  95         (skb->protocol == htons(ETH_P_PPP_SES) && \
  96          pppoe_proto(skb) == htons(PPP_IP) && \
  97          brnf_filter_pppoe_tagged)
  98
  99 #define IS_PPPOE_IPV6(skb) \
 100         (skb->protocol == htons(ETH_P_PPP_SES) && \
 101          pppoe_proto(skb) == htons(PPP_IPV6) && \
 102          brnf_filter_pppoe_tagged)
 103
 104 /* We need these fake structures to make netfilter happy --
 105  * lots of places assume that skb->dst != NULL, which isn't
 106  * all that unreasonable.
 107  *
 108  * Currently, we fill in the PMTU entry because netfilter
 109  * refragmentation needs it, and the rt_flags entry because
 110  * ipt_REJECT needs it.  Future netfilter modules might
 111  * require us to fill additional fields. */
 112 static struct net_device __fake_net_device = {
 113         .hard_header_len        = ETH_HLEN
 114 };
 115
 116 static struct rtable __fake_rtable = {
 117         .u = {
 118                 .dst = {
 119                         .__refcnt               = ATOMIC_INIT(1),
 120                         .dev                    = &__fake_net_device,
 121                         .path                   = &__fake_rtable.u.dst,
 122                         .metrics                = {[RTAX_MTU - 1] = 1500},
 123                         .flags                  = DST_NOXFRM,
 124                 }
 125         },
 126         .rt_flags       = 0,
 127 };
 128
 129 static inline struct net_device *bridge_parent(const struct net_device *dev)
 130 {
 131         struct net_bridge_port *port = rcu_dereference(dev->br_port);
 132
 133         return port ? port->br->dev : NULL;
 134 }
 135
 136 static inline struct nf_bridge_info *nf_bridge_alloc(struct sk_buff *skb)
 137 {
 138         skb->nf_bridge = kzalloc(sizeof(struct nf_bridge_info), GFP_ATOMIC);
 139         if (likely(skb->nf_bridge))
 140                 atomic_set(&(skb->nf_bridge->use), 1);
 141
 142         return skb->nf_bridge;
 143 }
 144
 145 static inline void nf_bridge_save_header(struct sk_buff *skb)
 146 {
 147         int header_size = ETH_HLEN;
 148
 149         if (skb->protocol == htons(ETH_P_8021Q))
 150                 header_size += VLAN_HLEN;
 151         else if (skb->protocol == htons(ETH_P_PPP_SES))
 152                 header_size += PPPOE_SES_HLEN;
 153
 154         skb_copy_from_linear_data_offset(skb, -header_size,
 155                                          skb->nf_bridge->data, header_size);
 156 }
 157
 158 /*
 159  * When forwarding bridge frames, we save a copy of the original
 160  * header before processing.
 161  */
 162 int nf_bridge_copy_header(struct sk_buff *skb)
 163 {
 164         int err;
 165         int header_size = ETH_HLEN;
 166
 167         if (skb->protocol == htons(ETH_P_8021Q))
 168                 header_size += VLAN_HLEN;
 169         else if (skb->protocol == htons(ETH_P_PPP_SES))
 170                 header_size += PPPOE_SES_HLEN;
 171
 172         err = skb_cow(skb, header_size);
 173         if (err)
 174                 return err;
 175
 176         skb_copy_to_linear_data_offset(skb, -header_size,
 177                                        skb->nf_bridge->data, header_size);
 178
 179         if (skb->protocol == htons(ETH_P_8021Q))
 180                 __skb_push(skb, VLAN_HLEN);
 181         else if (skb->protocol == htons(ETH_P_PPP_SES))
 182                 __skb_push(skb, PPPOE_SES_HLEN);
 183         return 0;
 184 }
 185
 186 /* PF_BRIDGE/PRE_ROUTING *********************************************/
 187 /* Undo the changes made for ip6tables PREROUTING and continue the
 188  * bridge PRE_ROUTING hook. */
 189 static int br_nf_pre_routing_finish_ipv6(struct sk_buff *skb)
 190 {
 191         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 192
 193         if (nf_bridge->mask & BRNF_PKT_TYPE) {
 194                 skb->pkt_type = PACKET_OTHERHOST;
 195                 nf_bridge->mask ^= BRNF_PKT_TYPE;
 196         }
 197         nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
 198
 199         skb->dst = (struct dst_entry *)&__fake_rtable;
 200         dst_hold(skb->dst);
 201
 202         skb->dev = nf_bridge->physindev;
 203         if (skb->protocol == htons(ETH_P_8021Q)) {
 204                 skb_push(skb, VLAN_HLEN);
 205                 skb->network_header -= VLAN_HLEN;
 206         } else if (skb->protocol == htons(ETH_P_PPP_SES)) {
 207                 skb_push(skb, PPPOE_SES_HLEN);
 208                 skb->network_header -= PPPOE_SES_HLEN;
 209         }
 210         NF_HOOK_THRESH(PF_BRIDGE, NF_BR_PRE_ROUTING, skb, skb->dev, NULL,
 211                        br_handle_frame_finish, 1);
 212
 213         return 0;
 214 }
 215
 216 static void __br_dnat_complain(void)
 217 {
 218         static unsigned long last_complaint;
 219
 220         if (jiffies - last_complaint >= 5 * HZ) {
 221                 printk(KERN_WARNING "Performing cross-bridge DNAT requires IP "
 222                        "forwarding to be enabled\n");
 223                 last_complaint = jiffies;
 224         }
 225 }
 226
 227 /* This requires some explaining. If DNAT has taken place,
 228  * we will need to fix up the destination Ethernet address,
 229  * and this is a tricky process.
 230  *
 231  * There are two cases to consider:
 232  * 1. The packet was DNAT'ed to a device in the same bridge
 233  *    port group as it was received on. We can still bridge
 234  *    the packet.
 235  * 2. The packet was DNAT'ed to a different device, either
 236  *    a non-bridged device or another bridge port group.
 237  *    The packet will need to be routed.
 238  *
 239  * The correct way of distinguishing between these two cases is to
 240  * call ip_route_input() and to look at skb->dst->dev, which is
 241  * changed to the destination device if ip_route_input() succeeds.
 242  *
 243  * Let us first consider the case that ip_route_input() succeeds:
 244  *
 245  * If skb->dst->dev equals the logical bridge device the packet
 246  * came in on, we can consider this bridging. We then call
 247  * skb->dst->output() which will make the packet enter br_nf_local_out()
 248  * not much later. In that function it is assured that the iptables
 249  * FORWARD chain is traversed for the packet.
 250  *
 251  * Otherwise, the packet is considered to be routed and we just
 252  * change the destination MAC address so that the packet will
 253  * later be passed up to the IP stack to be routed. For a redirected
 254  * packet, ip_route_input() will give back the localhost as output device,
 255  * which differs from the bridge device.
 256  *
 257  * Let us now consider the case that ip_route_input() fails:
 258  *
 259  * This can be because the destination address is martian, in which case
 260  * the packet will be dropped.
 261  * After a "echo '0' > /proc/sys/net/ipv4/ip_forward" ip_route_input()
 262  * will fail, while __ip_route_output_key() will return success. The source
 263  * address for __ip_route_output_key() is set to zero, so __ip_route_output_key
 264  * thinks we're handling a locally generated packet and won't care
 265  * if IP forwarding is allowed. We send a warning message to the users's
 266  * log telling her to put IP forwarding on.
 267  *
 268  * ip_route_input() will also fail if there is no route available.
 269  * In that case we just drop the packet.
 270  *
 271  * --Lennert, 20020411
 272  * --Bart, 20020416 (updated)
 273  * --Bart, 20021007 (updated)
 274  * --Bart, 20062711 (updated) */
 275 static int br_nf_pre_routing_finish_bridge(struct sk_buff *skb)
 276 {
 277         if (skb->pkt_type == PACKET_OTHERHOST) {
 278                 skb->pkt_type = PACKET_HOST;
 279                 skb->nf_bridge->mask |= BRNF_PKT_TYPE;
 280         }
 281         skb->nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
 282
 283         skb->dev = bridge_parent(skb->dev);
 284         if (!skb->dev)
 285                 kfree_skb(skb);
 286         else {
 287                 if (skb->protocol == htons(ETH_P_8021Q)) {
 288                         skb_pull(skb, VLAN_HLEN);
 289                         skb->network_header += VLAN_HLEN;
 290                 } else if (skb->protocol == htons(ETH_P_PPP_SES)) {
 291                         skb_pull(skb, PPPOE_SES_HLEN);
 292                         skb->network_header += PPPOE_SES_HLEN;
 293                 }
 294                 skb->dst->output(skb);
 295         }
 296         return 0;
 297 }
 298
 299 static int br_nf_pre_routing_finish(struct sk_buff *skb)
 300 {
 301         struct net_device *dev = skb->dev;
 302         struct iphdr *iph = ip_hdr(skb);
 303         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 304         int err;
 305
 306         if (nf_bridge->mask & BRNF_PKT_TYPE) {
 307                 skb->pkt_type = PACKET_OTHERHOST;
 308                 nf_bridge->mask ^= BRNF_PKT_TYPE;
 309         }
 310         nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
 311         if (dnat_took_place(skb)) {
 312                 if ((err = ip_route_input(skb, iph->daddr, iph->saddr, iph->tos, dev))) {
 313                         struct rtable *rt;
 314                         struct flowi fl = {
 315                                 .nl_u = {
 316                                         .ip4_u = {
 317                                                  .daddr = iph->daddr,
 318                                                  .saddr = 0,
 319                                                  .tos = RT_TOS(iph->tos) },
 320                                 },
 321                                 .proto = 0,
 322                         };
 323                         struct in_device *in_dev = in_dev_get(dev);
 324
 325                         /* If err equals -EHOSTUNREACH the error is due to a
 326                          * martian destination or due to the fact that
 327                          * forwarding is disabled. For most martian packets,
 328                          * ip_route_output_key() will fail. It won't fail for 2 types of
 329                          * martian destinations: loopback destinations and destination
 330                          * 0.0.0.0. In both cases the packet will be dropped because the
 331                          * destination is the loopback device and not the bridge. */
 332                         if (err != -EHOSTUNREACH || !in_dev || IN_DEV_FORWARD(in_dev))
 333                                 goto free_skb;
 334
 335                         if (!ip_route_output_key(&rt, &fl)) {
 336                                 /* - Bridged-and-DNAT'ed traffic doesn't
 337                                  *   require ip_forwarding. */
 338                                 if (((struct dst_entry *)rt)->dev == dev) {
 339                                         skb->dst = (struct dst_entry *)rt;
 340                                         goto bridged_dnat;
 341                                 }
 342                                 /* we are sure that forwarding is disabled, so printing
 343                                  * this message is no problem. Note that the packet could
 344                                  * still have a martian destination address, in which case
 345                                  * the packet could be dropped even if forwarding were enabled */
 346                                 __br_dnat_complain();
 347                                 dst_release((struct dst_entry *)rt);
 348                         }
 349 free_skb:
 350                         kfree_skb(skb);
 351                         return 0;
 352                 } else {
 353                         if (skb->dst->dev == dev) {
 354 bridged_dnat:
 355                                 /* Tell br_nf_local_out this is a
 356                                  * bridged frame */
 357                                 nf_bridge->mask |= BRNF_BRIDGED_DNAT;
 358                                 skb->dev = nf_bridge->physindev;
 359                                 if (skb->protocol ==
 360                                     htons(ETH_P_8021Q)) {
 361                                         skb_push(skb, VLAN_HLEN);
 362                                         skb->network_header -= VLAN_HLEN;
 363                                 } else if(skb->protocol ==
 364                                     htons(ETH_P_PPP_SES)) {
 365                                         skb_push(skb, PPPOE_SES_HLEN);
 366                                         skb->network_header -= PPPOE_SES_HLEN;
 367                                 }
 368                                 NF_HOOK_THRESH(PF_BRIDGE, NF_BR_PRE_ROUTING,
 369                                                skb, skb->dev, NULL,
 370                                                br_nf_pre_routing_finish_bridge,
 371                                                1);
 372                                 return 0;
 373                         }
 374                         memcpy(eth_hdr(skb)->h_dest, dev->dev_addr, ETH_ALEN);
 375                         skb->pkt_type = PACKET_HOST;
 376                 }
 377         } else {
 378                 skb->dst = (struct dst_entry *)&__fake_rtable;
 379                 dst_hold(skb->dst);
 380         }
 381
 382         skb->dev = nf_bridge->physindev;
 383         if (skb->protocol == htons(ETH_P_8021Q)) {
 384                 skb_push(skb, VLAN_HLEN);
 385                 skb->network_header -= VLAN_HLEN;
 386         } else if (skb->protocol == htons(ETH_P_PPP_SES)) {
 387                 skb_push(skb, PPPOE_SES_HLEN);
 388                 skb->network_header -= PPPOE_SES_HLEN;
 389         }
 390         NF_HOOK_THRESH(PF_BRIDGE, NF_BR_PRE_ROUTING, skb, skb->dev, NULL,
 391                        br_handle_frame_finish, 1);
 392
 393         return 0;
 394 }
 395
 396 /* Some common code for IPv4/IPv6 */
 397 static struct net_device *setup_pre_routing(struct sk_buff *skb)
 398 {
 399         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 400
 401         if (skb->pkt_type == PACKET_OTHERHOST) {
 402                 skb->pkt_type = PACKET_HOST;
 403                 nf_bridge->mask |= BRNF_PKT_TYPE;
 404         }
 405
 406         nf_bridge->mask |= BRNF_NF_BRIDGE_PREROUTING;
 407         nf_bridge->physindev = skb->dev;
 408         skb->dev = bridge_parent(skb->dev);
 409
 410         return skb->dev;
 411 }
 412
 413 /* We only check the length. A bridge shouldn't do any hop-by-hop stuff anyway */
 414 static int check_hbh_len(struct sk_buff *skb)
 415 {
 416         unsigned char *raw = (u8 *)(ipv6_hdr(skb) + 1);
 417         u32 pkt_len;
 418         const unsigned char *nh = skb_network_header(skb);
 419         int off = raw - nh;
 420         int len = (raw[1] + 1) << 3;
 421
 422         if ((raw + len) - skb->data > skb_headlen(skb))
 423                 goto bad;
 424
 425         off += 2;
 426         len -= 2;
 427
 428         while (len > 0) {
 429                 int optlen = nh[off + 1] + 2;
 430
 431                 switch (nh[off]) {
 432                 case IPV6_TLV_PAD0:
 433                         optlen = 1;
 434                         break;
 435
 436                 case IPV6_TLV_PADN:
 437                         break;
 438
 439                 case IPV6_TLV_JUMBO:
 440                         if (nh[off + 1] != 4 || (off & 3) != 2)
 441                                 goto bad;
 442                         pkt_len = ntohl(*(__be32 *) (nh + off + 2));
 443                         if (pkt_len <= IPV6_MAXPLEN ||
 444                             ipv6_hdr(skb)->payload_len)
 445                                 goto bad;
 446                         if (pkt_len > skb->len - sizeof(struct ipv6hdr))
 447                                 goto bad;
 448                         if (pskb_trim_rcsum(skb,
 449                                             pkt_len + sizeof(struct ipv6hdr)))
 450                                 goto bad;
 451                         nh = skb_network_header(skb);
 452                         break;
 453                 default:
 454                         if (optlen > len)
 455                                 goto bad;
 456                         break;
 457                 }
 458                 off += optlen;
 459                 len -= optlen;
 460         }
 461         if (len == 0)
 462                 return 0;
 463 bad:
 464         return -1;
 465
 466 }
 467
 468 /* Replicate the checks that IPv6 does on packet reception and pass the packet
 469  * to ip6tables, which doesn't support NAT, so things are fairly simple. */
 470 static unsigned int br_nf_pre_routing_ipv6(unsigned int hook,
 471                                            struct sk_buff *skb,
 472                                            const struct net_device *in,
 473                                            const struct net_device *out,
 474                                            int (*okfn)(struct sk_buff *))
 475 {
 476         struct ipv6hdr *hdr;
 477         u32 pkt_len;
 478
 479         if (skb->len < sizeof(struct ipv6hdr))
 480                 goto inhdr_error;
 481
 482         if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
 483                 goto inhdr_error;
 484
 485         hdr = ipv6_hdr(skb);
 486
 487         if (hdr->version != 6)
 488                 goto inhdr_error;
 489
 490         pkt_len = ntohs(hdr->payload_len);
 491
 492         if (pkt_len || hdr->nexthdr != NEXTHDR_HOP) {
 493                 if (pkt_len + sizeof(struct ipv6hdr) > skb->len)
 494                         goto inhdr_error;
 495                 if (pskb_trim_rcsum(skb, pkt_len + sizeof(struct ipv6hdr)))
 496                         goto inhdr_error;
 497         }
 498         if (hdr->nexthdr == NEXTHDR_HOP && check_hbh_len(skb))
 499                 goto inhdr_error;
 500
 501         nf_bridge_put(skb->nf_bridge);
 502         if (!nf_bridge_alloc(skb))
 503                 return NF_DROP;
 504         if (!setup_pre_routing(skb))
 505                 return NF_DROP;
 506
 507         NF_HOOK(PF_INET6, NF_IP6_PRE_ROUTING, skb, skb->dev, NULL,
 508                 br_nf_pre_routing_finish_ipv6);
 509
 510         return NF_STOLEN;
 511
 512 inhdr_error:
 513         return NF_DROP;
 514 }
 515
 516 /* Direct IPv6 traffic to br_nf_pre_routing_ipv6.
 517  * Replicate the checks that IPv4 does on packet reception.
 518  * Set skb->dev to the bridge device (i.e. parent of the
 519  * receiving device) to make netfilter happy, the REDIRECT
 520  * target in particular.  Save the original destination IP
 521  * address to be able to detect DNAT afterwards. */
 522 static unsigned int br_nf_pre_routing(unsigned int hook, struct sk_buff **pskb,
 523                                       const struct net_device *in,
 524                                       const struct net_device *out,
 525                                       int (*okfn)(struct sk_buff *))
 526 {
 527         struct iphdr *iph;
 528         __u32 len;
 529         struct sk_buff *skb = *pskb;
 530
 531         if (skb->protocol == htons(ETH_P_IPV6) || IS_VLAN_IPV6(skb) ||
 532             IS_PPPOE_IPV6(skb)) {
 533 #ifdef CONFIG_SYSCTL
 534                 if (!brnf_call_ip6tables)
 535                         return NF_ACCEPT;
 536 #endif
 537                 if ((skb = skb_share_check(*pskb, GFP_ATOMIC)) == NULL)
 538                         goto out;
 539
 540                 if (skb->protocol == htons(ETH_P_8021Q)) {
 541                         skb_pull_rcsum(skb, VLAN_HLEN);
 542                         skb->network_header += VLAN_HLEN;
 543                 } else if (skb->protocol == htons(ETH_P_PPP_SES)) {
 544                         skb_pull_rcsum(skb, PPPOE_SES_HLEN);
 545                         skb->network_header += PPPOE_SES_HLEN;
 546                 }
 547                 return br_nf_pre_routing_ipv6(hook, skb, in, out, okfn);
 548         }
 549 #ifdef CONFIG_SYSCTL
 550         if (!brnf_call_iptables)
 551                 return NF_ACCEPT;
 552 #endif
 553
 554         if (skb->protocol != htons(ETH_P_IP) && !IS_VLAN_IP(skb) &&
 555             !IS_PPPOE_IP(skb))
 556                 return NF_ACCEPT;
 557
 558         if ((skb = skb_share_check(*pskb, GFP_ATOMIC)) == NULL)
 559                 goto out;
 560
 561         if (skb->protocol == htons(ETH_P_8021Q)) {
 562                 skb_pull_rcsum(skb, VLAN_HLEN);
 563                 skb->network_header += VLAN_HLEN;
 564         } else if (skb->protocol == htons(ETH_P_PPP_SES)) {
 565                 skb_pull_rcsum(skb, PPPOE_SES_HLEN);
 566                 skb->network_header += PPPOE_SES_HLEN;
 567         }
 568
 569         if (!pskb_may_pull(skb, sizeof(struct iphdr)))
 570                 goto inhdr_error;
 571
 572         iph = ip_hdr(skb);
 573         if (iph->ihl < 5 || iph->version != 4)
 574                 goto inhdr_error;
 575
 576         if (!pskb_may_pull(skb, 4 * iph->ihl))
 577                 goto inhdr_error;
 578
 579         iph = ip_hdr(skb);
 580         if (ip_fast_csum((__u8 *) iph, iph->ihl) != 0)
 581                 goto inhdr_error;
 582
 583         len = ntohs(iph->tot_len);
 584         if (skb->len < len || len < 4 * iph->ihl)
 585                 goto inhdr_error;
 586
 587         pskb_trim_rcsum(skb, len);
 588
 589         nf_bridge_put(skb->nf_bridge);
 590         if (!nf_bridge_alloc(skb))
 591                 return NF_DROP;
 592         if (!setup_pre_routing(skb))
 593                 return NF_DROP;
 594         store_orig_dstaddr(skb);
 595
 596         NF_HOOK(PF_INET, NF_IP_PRE_ROUTING, skb, skb->dev, NULL,
 597                 br_nf_pre_routing_finish);
 598
 599         return NF_STOLEN;
 600
 601 inhdr_error:
 602 //      IP_INC_STATS_BH(IpInHdrErrors);
 603 out:
 604         return NF_DROP;
 605 }
 606
 607
 608 /* PF_BRIDGE/LOCAL_IN ************************************************/
 609 /* The packet is locally destined, which requires a real
 610  * dst_entry, so detach the fake one.  On the way up, the
 611  * packet would pass through PRE_ROUTING again (which already
 612  * took place when the packet entered the bridge), but we
 613  * register an IPv4 PRE_ROUTING 'sabotage' hook that will
 614  * prevent this from happening. */
 615 static unsigned int br_nf_local_in(unsigned int hook, struct sk_buff **pskb,
 616                                    const struct net_device *in,
 617                                    const struct net_device *out,
 618                                    int (*okfn)(struct sk_buff *))
 619 {
 620         struct sk_buff *skb = *pskb;
 621
 622         if (skb->dst == (struct dst_entry *)&__fake_rtable) {
 623                 dst_release(skb->dst);
 624                 skb->dst = NULL;
 625         }
 626
 627         return NF_ACCEPT;
 628 }
 629
 630 /* PF_BRIDGE/FORWARD *************************************************/
 631 static int br_nf_forward_finish(struct sk_buff *skb)
 632 {
 633         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 634         struct net_device *in;
 635
 636         if (skb->protocol != htons(ETH_P_ARP) && !IS_VLAN_ARP(skb)) {
 637                 in = nf_bridge->physindev;
 638                 if (nf_bridge->mask & BRNF_PKT_TYPE) {
 639                         skb->pkt_type = PACKET_OTHERHOST;
 640                         nf_bridge->mask ^= BRNF_PKT_TYPE;
 641                 }
 642         } else {
 643                 in = *((struct net_device **)(skb->cb));
 644         }
 645         if (skb->protocol == htons(ETH_P_8021Q)) {
 646                 skb_push(skb, VLAN_HLEN);
 647                 skb->network_header -= VLAN_HLEN;
 648         } else if (skb->protocol == htons(ETH_P_PPP_SES)) {
 649                 skb_push(skb, PPPOE_SES_HLEN);
 650                 skb->network_header -= PPPOE_SES_HLEN;
 651         }
 652         NF_HOOK_THRESH(PF_BRIDGE, NF_BR_FORWARD, skb, in,
 653                        skb->dev, br_forward_finish, 1);
 654         return 0;
 655 }
 656
 657 /* This is the 'purely bridged' case.  For IP, we pass the packet to
 658  * netfilter with indev and outdev set to the bridge device,
 659  * but we are still able to filter on the 'real' indev/outdev
 660  * because of the physdev module. For ARP, indev and outdev are the
 661  * bridge ports. */
 662 static unsigned int br_nf_forward_ip(unsigned int hook, struct sk_buff **pskb,
 663                                      const struct net_device *in,
 664                                      const struct net_device *out,
 665                                      int (*okfn)(struct sk_buff *))
 666 {
 667         struct sk_buff *skb = *pskb;
 668         struct nf_bridge_info *nf_bridge;
 669         struct net_device *parent;
 670         int pf;
 671
 672         if (!skb->nf_bridge)
 673                 return NF_ACCEPT;
 674
 675         parent = bridge_parent(out);
 676         if (!parent)
 677                 return NF_DROP;
 678
 679         if (skb->protocol == htons(ETH_P_IP) || IS_VLAN_IP(skb) ||
 680             IS_PPPOE_IP(skb))
 681                 pf = PF_INET;
 682         else
 683                 pf = PF_INET6;
 684
 685         if (skb->protocol == htons(ETH_P_8021Q)) {
 686                 skb_pull(*pskb, VLAN_HLEN);
 687                 (*pskb)->network_header += VLAN_HLEN;
 688         } else if (skb->protocol == htons(ETH_P_PPP_SES)) {
 689                 skb_pull(*pskb, PPPOE_SES_HLEN);
 690                 (*pskb)->network_header += PPPOE_SES_HLEN;
 691         }
 692
 693         nf_bridge = skb->nf_bridge;
 694         if (skb->pkt_type == PACKET_OTHERHOST) {
 695                 skb->pkt_type = PACKET_HOST;
 696                 nf_bridge->mask |= BRNF_PKT_TYPE;
 697         }
 698
 699         /* The physdev module checks on this */
 700         nf_bridge->mask |= BRNF_BRIDGED;
 701         nf_bridge->physoutdev = skb->dev;
 702
 703         NF_HOOK(pf, NF_IP_FORWARD, skb, bridge_parent(in), parent,
 704                 br_nf_forward_finish);
 705
 706         return NF_STOLEN;
 707 }
 708
 709 static unsigned int br_nf_forward_arp(unsigned int hook, struct sk_buff **pskb,
 710                                       const struct net_device *in,
 711                                       const struct net_device *out,
 712                                       int (*okfn)(struct sk_buff *))
 713 {
 714         struct sk_buff *skb = *pskb;
 715         struct net_device **d = (struct net_device **)(skb->cb);
 716
 717 #ifdef CONFIG_SYSCTL
 718         if (!brnf_call_arptables)
 719                 return NF_ACCEPT;
 720 #endif
 721
 722         if (skb->protocol != htons(ETH_P_ARP)) {
 723                 if (!IS_VLAN_ARP(skb))
 724                         return NF_ACCEPT;
 725                 skb_pull(*pskb, VLAN_HLEN);
 726                 (*pskb)->network_header += VLAN_HLEN;
 727         }
 728
 729         if (arp_hdr(skb)->ar_pln != 4) {
 730                 if (IS_VLAN_ARP(skb)) {
 731                         skb_push(*pskb, VLAN_HLEN);
 732                         (*pskb)->network_header -= VLAN_HLEN;
 733                 }
 734                 return NF_ACCEPT;
 735         }
 736         *d = (struct net_device *)in;
 737         NF_HOOK(NF_ARP, NF_ARP_FORWARD, skb, (struct net_device *)in,
 738                 (struct net_device *)out, br_nf_forward_finish);
 739
 740         return NF_STOLEN;
 741 }
 742
 743 /* PF_BRIDGE/LOCAL_OUT ***********************************************
 744  *
 745  * This function sees both locally originated IP packets and forwarded
 746  * IP packets (in both cases the destination device is a bridge
 747  * device). It also sees bridged-and-DNAT'ed packets.
 748  *
 749  * If (nf_bridge->mask & BRNF_BRIDGED_DNAT) then the packet is bridged
 750  * and we fake the PF_BRIDGE/FORWARD hook. The function br_nf_forward()
 751  * will then fake the PF_INET/FORWARD hook. br_nf_local_out() has priority
 752  * NF_BR_PRI_FIRST, so no relevant PF_BRIDGE/INPUT functions have been nor
 753  * will be executed.
 754  */
 755 static unsigned int br_nf_local_out(unsigned int hook, struct sk_buff **pskb,
 756                                     const struct net_device *in,
 757                                     const struct net_device *out,
 758                                     int (*okfn)(struct sk_buff *))
 759 {
 760         struct net_device *realindev;
 761         struct sk_buff *skb = *pskb;
 762         struct nf_bridge_info *nf_bridge;
 763
 764         if (!skb->nf_bridge)
 765                 return NF_ACCEPT;
 766
 767         nf_bridge = skb->nf_bridge;
 768         if (!(nf_bridge->mask & BRNF_BRIDGED_DNAT))
 769                 return NF_ACCEPT;
 770
 771         /* Bridged, take PF_BRIDGE/FORWARD.
 772          * (see big note in front of br_nf_pre_routing_finish) */
 773         nf_bridge->physoutdev = skb->dev;
 774         realindev = nf_bridge->physindev;
 775
 776         if (nf_bridge->mask & BRNF_PKT_TYPE) {
 777                 skb->pkt_type = PACKET_OTHERHOST;
 778                 nf_bridge->mask ^= BRNF_PKT_TYPE;
 779         }
 780         if (skb->protocol == htons(ETH_P_8021Q)) {
 781                 skb_push(skb, VLAN_HLEN);
 782                 skb->network_header -= VLAN_HLEN;
 783         } else if (skb->protocol == htons(ETH_P_PPP_SES)) {
 784                 skb_push(skb, PPPOE_SES_HLEN);
 785                 skb->network_header -= PPPOE_SES_HLEN;
 786         }
 787
 788         NF_HOOK(PF_BRIDGE, NF_BR_FORWARD, skb, realindev, skb->dev,
 789                 br_forward_finish);
 790         return NF_STOLEN;
 791 }
 792
 793 static int br_nf_dev_queue_xmit(struct sk_buff *skb)
 794 {
 795         if (skb->protocol == htons(ETH_P_IP) &&
 796             skb->len > skb->dev->mtu &&
 797             !skb_is_gso(skb))
 798                 return ip_fragment(skb, br_dev_queue_push_xmit);
 799         else
 800                 return br_dev_queue_push_xmit(skb);
 801 }
 802
 803 /* PF_BRIDGE/POST_ROUTING ********************************************/
 804 static unsigned int br_nf_post_routing(unsigned int hook, struct sk_buff **pskb,
 805                                        const struct net_device *in,
 806                                        const struct net_device *out,
 807                                        int (*okfn)(struct sk_buff *))
 808 {
 809         struct sk_buff *skb = *pskb;
 810         struct nf_bridge_info *nf_bridge = (*pskb)->nf_bridge;
 811         struct net_device *realoutdev = bridge_parent(skb->dev);
 812         int pf;
 813
 814 #ifdef CONFIG_NETFILTER_DEBUG
 815         /* Be very paranoid. This probably won't happen anymore, but let's
 816          * keep the check just to be sure... */
 817         if (skb_mac_header(skb) < skb->head ||
 818             skb_mac_header(skb) + ETH_HLEN > skb->data) {
 819                 printk(KERN_CRIT "br_netfilter: Argh!! br_nf_post_routing: "
 820                        "bad mac.raw pointer.\n");
 821                 goto print_error;
 822         }
 823 #endif
 824
 825         if (!nf_bridge)
 826                 return NF_ACCEPT;
 827
 828         if (!realoutdev)
 829                 return NF_DROP;
 830
 831         if (skb->protocol == htons(ETH_P_IP) || IS_VLAN_IP(skb) ||
 832             IS_PPPOE_IP(skb))
 833                 pf = PF_INET;
 834         else
 835                 pf = PF_INET6;
 836
 837 #ifdef CONFIG_NETFILTER_DEBUG
 838         if (skb->dst == NULL) {
 839                 printk(KERN_INFO "br_netfilter post_routing: skb->dst == NULL\n");
 840                 goto print_error;
 841         }
 842 #endif
 843
 844         /* We assume any code from br_dev_queue_push_xmit onwards doesn't care
 845          * about the value of skb->pkt_type. */
 846         if (skb->pkt_type == PACKET_OTHERHOST) {
 847                 skb->pkt_type = PACKET_HOST;
 848                 nf_bridge->mask |= BRNF_PKT_TYPE;
 849         }
 850
 851         if (skb->protocol == htons(ETH_P_8021Q)) {
 852                 skb_pull(skb, VLAN_HLEN);
 853                 skb->network_header += VLAN_HLEN;
 854         } else if (skb->protocol == htons(ETH_P_PPP_SES)) {
 855                 skb_pull(skb, PPPOE_SES_HLEN);
 856                 skb->network_header += PPPOE_SES_HLEN;
 857         }
 858
 859         nf_bridge_save_header(skb);
 860
 861 #if defined(CONFIG_VLAN_8021Q) || defined(CONFIG_VLAN_8021Q_MODULE)
 862         if (nf_bridge->netoutdev)
 863                 realoutdev = nf_bridge->netoutdev;
 864 #endif
 865         NF_HOOK(pf, NF_IP_POST_ROUTING, skb, NULL, realoutdev,
 866                 br_nf_dev_queue_xmit);
 867
 868         return NF_STOLEN;
 869
 870 #ifdef CONFIG_NETFILTER_DEBUG
 871 print_error:
 872         if (skb->dev != NULL) {
 873                 printk("[%s]", skb->dev->name);
 874                 if (realoutdev)
 875                         printk("[%s]", realoutdev->name);
 876         }
 877         printk(" head:%p, raw:%p, data:%p\n", skb->head, skb_mac_header(skb),
 878                skb->data);
 879         dump_stack();
 880         return NF_ACCEPT;
 881 #endif
 882 }
 883
 884 /* IP/SABOTAGE *****************************************************/
 885 /* Don't hand locally destined packets to PF_INET(6)/PRE_ROUTING
 886  * for the second time. */
 887 static unsigned int ip_sabotage_in(unsigned int hook, struct sk_buff **pskb,
 888                                    const struct net_device *in,
 889                                    const struct net_device *out,
 890                                    int (*okfn)(struct sk_buff *))
 891 {
 892         if ((*pskb)->nf_bridge &&
 893             !((*pskb)->nf_bridge->mask & BRNF_NF_BRIDGE_PREROUTING)) {
 894                 return NF_STOP;
 895         }
 896
 897         return NF_ACCEPT;
 898 }
 899
 900 /* For br_nf_local_out we need (prio = NF_BR_PRI_FIRST), to insure that innocent
 901  * PF_BRIDGE/NF_BR_LOCAL_OUT functions don't get bridged traffic as input.
 902  * For br_nf_post_routing, we need (prio = NF_BR_PRI_LAST), because
 903  * ip_refrag() can return NF_STOLEN. */
 904 static struct nf_hook_ops br_nf_ops[] = {
 905         { .hook = br_nf_pre_routing,
 906           .owner = THIS_MODULE,
 907           .pf = PF_BRIDGE,
 908           .hooknum = NF_BR_PRE_ROUTING,
 909           .priority = NF_BR_PRI_BRNF, },
 910         { .hook = br_nf_local_in,
 911           .owner = THIS_MODULE,
 912           .pf = PF_BRIDGE,
 913           .hooknum = NF_BR_LOCAL_IN,
 914           .priority = NF_BR_PRI_BRNF, },
 915         { .hook = br_nf_forward_ip,
 916           .owner = THIS_MODULE,
 917           .pf = PF_BRIDGE,
 918           .hooknum = NF_BR_FORWARD,
 919           .priority = NF_BR_PRI_BRNF - 1, },
 920         { .hook = br_nf_forward_arp,
 921           .owner = THIS_MODULE,
 922           .pf = PF_BRIDGE,
 923           .hooknum = NF_BR_FORWARD,
 924           .priority = NF_BR_PRI_BRNF, },
 925         { .hook = br_nf_local_out,
 926           .owner = THIS_MODULE,
 927           .pf = PF_BRIDGE,
 928           .hooknum = NF_BR_LOCAL_OUT,
 929           .priority = NF_BR_PRI_FIRST, },
 930         { .hook = br_nf_post_routing,
 931           .owner = THIS_MODULE,
 932           .pf = PF_BRIDGE,
 933           .hooknum = NF_BR_POST_ROUTING,
 934           .priority = NF_BR_PRI_LAST, },
 935         { .hook = ip_sabotage_in,
 936           .owner = THIS_MODULE,
 937           .pf = PF_INET,
 938           .hooknum = NF_IP_PRE_ROUTING,
 939           .priority = NF_IP_PRI_FIRST, },
 940         { .hook = ip_sabotage_in,
 941           .owner = THIS_MODULE,
 942           .pf = PF_INET6,
 943           .hooknum = NF_IP6_PRE_ROUTING,
 944           .priority = NF_IP6_PRI_FIRST, },
 945 };
 946
 947 #ifdef CONFIG_SYSCTL
 948 static
 949 int brnf_sysctl_call_tables(ctl_table * ctl, int write, struct file *filp,
 950                             void __user * buffer, size_t * lenp, loff_t * ppos)
 951 {
 952         int ret;
 953
 954         ret = proc_dointvec(ctl, write, filp, buffer, lenp, ppos);
 955
 956         if (write && *(int *)(ctl->data))
 957                 *(int *)(ctl->data) = 1;
 958         return ret;
 959 }
 960
 961 static ctl_table brnf_table[] = {
 962         {
 963                 .ctl_name       = NET_BRIDGE_NF_CALL_ARPTABLES,
 964                 .procname       = "bridge-nf-call-arptables",
 965                 .data           = &brnf_call_arptables,
 966                 .maxlen         = sizeof(int),
 967                 .mode           = 0644,
 968                 .proc_handler   = &brnf_sysctl_call_tables,
 969         },
 970         {
 971                 .ctl_name       = NET_BRIDGE_NF_CALL_IPTABLES,
 972                 .procname       = "bridge-nf-call-iptables",
 973                 .data           = &brnf_call_iptables,
 974                 .maxlen         = sizeof(int),
 975                 .mode           = 0644,
 976                 .proc_handler   = &brnf_sysctl_call_tables,
 977         },
 978         {
 979                 .ctl_name       = NET_BRIDGE_NF_CALL_IP6TABLES,
 980                 .procname       = "bridge-nf-call-ip6tables",
 981                 .data           = &brnf_call_ip6tables,
 982                 .maxlen         = sizeof(int),
 983                 .mode           = 0644,
 984                 .proc_handler   = &brnf_sysctl_call_tables,
 985         },
 986         {
 987                 .ctl_name       = NET_BRIDGE_NF_FILTER_VLAN_TAGGED,
 988                 .procname       = "bridge-nf-filter-vlan-tagged",
 989                 .data           = &brnf_filter_vlan_tagged,
 990                 .maxlen         = sizeof(int),
 991                 .mode           = 0644,
 992                 .proc_handler   = &brnf_sysctl_call_tables,
 993         },
 994         {
 995                 .ctl_name       = NET_BRIDGE_NF_FILTER_PPPOE_TAGGED,
 996                 .procname       = "bridge-nf-filter-pppoe-tagged",
 997                 .data           = &brnf_filter_pppoe_tagged,
 998                 .maxlen         = sizeof(int),
 999                 .mode           = 0644,
1000                 .proc_handler   = &brnf_sysctl_call_tables,
1001         },
1002         { .ctl_name = 0 }
1003 };
1004
1005 static ctl_table brnf_bridge_table[] = {
1006         {
1007                 .ctl_name       = NET_BRIDGE,
1008                 .procname       = "bridge",
1009                 .mode           = 0555,
1010                 .child          = brnf_table,
1011         },
1012         { .ctl_name = 0 }
1013 };
1014
1015 static ctl_table brnf_net_table[] = {
1016         {
1017                 .ctl_name       = CTL_NET,
1018                 .procname       = "net",
1019                 .mode           = 0555,
1020                 .child          = brnf_bridge_table,
1021         },
1022         { .ctl_name = 0 }
1023 };
1024 #endif
1025
1026 int __init br_netfilter_init(void)
1027 {
1028         int ret;
1029
1030         ret = nf_register_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
1031         if (ret < 0)
1032                 return ret;
1033 #ifdef CONFIG_SYSCTL
1034         brnf_sysctl_header = register_sysctl_table(brnf_net_table);
1035         if (brnf_sysctl_header == NULL) {
1036                 printk(KERN_WARNING
1037                        "br_netfilter: can't register to sysctl.\n");
1038                 nf_unregister_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
1039                 return -ENOMEM;
1040         }
1041 #endif
1042         printk(KERN_NOTICE "Bridge firewalling registered\n");
1043         return 0;
1044 }
1045
1046 void br_netfilter_fini(void)
1047 {
1048         nf_unregister_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
1049 #ifdef CONFIG_SYSCTL
1050         unregister_sysctl_table(brnf_sysctl_header);
1051 #endif
1052 }