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
  38 #include <net/ip.h>
  39 #include <net/ipv6.h>
  40 #include <net/route.h>
  41
  42 #include <asm/uaccess.h>
  43 #include <asm/checksum.h>
  44 #include "br_private.h"
  45 #ifdef CONFIG_SYSCTL
  46 #include <linux/sysctl.h>
  47 #endif
  48
  49 #define skb_origaddr(skb)        (((struct bridge_skb_cb *) \
  50                                  (skb->nf_bridge->data))->daddr.ipv4)
  51 #define store_orig_dstaddr(skb)  (skb_origaddr(skb) = (skb)->nh.iph->daddr)
  52 #define dnat_took_place(skb)     (skb_origaddr(skb) != (skb)->nh.iph->daddr)
  53
  54 #ifdef CONFIG_SYSCTL
  55 static struct ctl_table_header *brnf_sysctl_header;
  56 static int brnf_call_iptables = 1;
  57 static int brnf_call_ip6tables = 1;
  58 static int brnf_call_arptables = 1;
  59 static int brnf_filter_vlan_tagged = 1;
  60 #else
  61 #define brnf_filter_vlan_tagged 1
  62 #endif
  63
  64 int brnf_deferred_hooks;
  65 EXPORT_SYMBOL_GPL(brnf_deferred_hooks);
  66
  67 static __be16 inline vlan_proto(const struct sk_buff *skb)
  68 {
  69         return vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
  70 }
  71
  72 #define IS_VLAN_IP(skb) \
  73         (skb->protocol == htons(ETH_P_8021Q) && \
  74          vlan_proto(skb) == htons(ETH_P_IP) &&  \
  75          brnf_filter_vlan_tagged)
  76
  77 #define IS_VLAN_IPV6(skb) \
  78         (skb->protocol == htons(ETH_P_8021Q) && \
  79          vlan_proto(skb) == htons(ETH_P_IPV6) &&\
  80          brnf_filter_vlan_tagged)
  81
  82 #define IS_VLAN_ARP(skb) \
  83         (skb->protocol == htons(ETH_P_8021Q) && \
  84          vlan_proto(skb) == htons(ETH_P_ARP) && \
  85          brnf_filter_vlan_tagged)
  86
  87 /* We need these fake structures to make netfilter happy --
  88  * lots of places assume that skb->dst != NULL, which isn't
  89  * all that unreasonable.
  90  *
  91  * Currently, we fill in the PMTU entry because netfilter
  92  * refragmentation needs it, and the rt_flags entry because
  93  * ipt_REJECT needs it.  Future netfilter modules might
  94  * require us to fill additional fields. */
  95 static struct net_device __fake_net_device = {
  96         .hard_header_len        = ETH_HLEN
  97 };
  98
  99 static struct rtable __fake_rtable = {
 100         .u = {
 101                 .dst = {
 102                         .__refcnt               = ATOMIC_INIT(1),
 103                         .dev                    = &__fake_net_device,
 104                         .path                   = &__fake_rtable.u.dst,
 105                         .metrics                = {[RTAX_MTU - 1] = 1500},
 106                         .flags                  = DST_NOXFRM,
 107                 }
 108         },
 109         .rt_flags       = 0,
 110 };
 111
 112 static inline struct net_device *bridge_parent(const struct net_device *dev)
 113 {
 114         struct net_bridge_port *port = rcu_dereference(dev->br_port);
 115
 116         return port ? port->br->dev : NULL;
 117 }
 118
 119 static inline struct nf_bridge_info *nf_bridge_alloc(struct sk_buff *skb)
 120 {
 121         skb->nf_bridge = kzalloc(sizeof(struct nf_bridge_info), GFP_ATOMIC);
 122         if (likely(skb->nf_bridge))
 123                 atomic_set(&(skb->nf_bridge->use), 1);
 124
 125         return skb->nf_bridge;
 126 }
 127
 128 static inline void nf_bridge_save_header(struct sk_buff *skb)
 129 {
 130         int header_size = 16;
 131
 132         if (skb->protocol == htons(ETH_P_8021Q))
 133                 header_size = 18;
 134
 135         memcpy(skb->nf_bridge->data, skb->data - header_size, header_size);
 136 }
 137
 138 /* PF_BRIDGE/PRE_ROUTING *********************************************/
 139 /* Undo the changes made for ip6tables PREROUTING and continue the
 140  * bridge PRE_ROUTING hook. */
 141 static int br_nf_pre_routing_finish_ipv6(struct sk_buff *skb)
 142 {
 143         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 144
 145         if (nf_bridge->mask & BRNF_PKT_TYPE) {
 146                 skb->pkt_type = PACKET_OTHERHOST;
 147                 nf_bridge->mask ^= BRNF_PKT_TYPE;
 148         }
 149         nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
 150
 151         skb->dst = (struct dst_entry *)&__fake_rtable;
 152         dst_hold(skb->dst);
 153
 154         skb->dev = nf_bridge->physindev;
 155         if (skb->protocol == htons(ETH_P_8021Q)) {
 156                 skb_push(skb, VLAN_HLEN);
 157                 skb->nh.raw -= VLAN_HLEN;
 158         }
 159         NF_HOOK_THRESH(PF_BRIDGE, NF_BR_PRE_ROUTING, skb, skb->dev, NULL,
 160                        br_handle_frame_finish, 1);
 161
 162         return 0;
 163 }
 164
 165 static void __br_dnat_complain(void)
 166 {
 167         static unsigned long last_complaint;
 168
 169         if (jiffies - last_complaint >= 5 * HZ) {
 170                 printk(KERN_WARNING "Performing cross-bridge DNAT requires IP "
 171                        "forwarding to be enabled\n");
 172                 last_complaint = jiffies;
 173         }
 174 }
 175
 176 /* This requires some explaining. If DNAT has taken place,
 177  * we will need to fix up the destination Ethernet address,
 178  * and this is a tricky process.
 179  *
 180  * There are two cases to consider:
 181  * 1. The packet was DNAT'ed to a device in the same bridge
 182  *    port group as it was received on. We can still bridge
 183  *    the packet.
 184  * 2. The packet was DNAT'ed to a different device, either
 185  *    a non-bridged device or another bridge port group.
 186  *    The packet will need to be routed.
 187  *
 188  * The correct way of distinguishing between these two cases is to
 189  * call ip_route_input() and to look at skb->dst->dev, which is
 190  * changed to the destination device if ip_route_input() succeeds.
 191  *
 192  * Let us first consider the case that ip_route_input() succeeds:
 193  *
 194  * If skb->dst->dev equals the logical bridge device the packet
 195  * came in on, we can consider this bridging. We then call
 196  * skb->dst->output() which will make the packet enter br_nf_local_out()
 197  * not much later. In that function it is assured that the iptables
 198  * FORWARD chain is traversed for the packet.
 199  *
 200  * Otherwise, the packet is considered to be routed and we just
 201  * change the destination MAC address so that the packet will
 202  * later be passed up to the IP stack to be routed.
 203  *
 204  * Let us now consider the case that ip_route_input() fails:
 205  *
 206  * After a "echo '0' > /proc/sys/net/ipv4/ip_forward" ip_route_input()
 207  * will fail, while __ip_route_output_key() will return success. The source
 208  * address for __ip_route_output_key() is set to zero, so __ip_route_output_key
 209  * thinks we're handling a locally generated packet and won't care
 210  * if IP forwarding is allowed. We send a warning message to the users's
 211  * log telling her to put IP forwarding on.
 212  *
 213  * ip_route_input() will also fail if there is no route available.
 214  * In that case we just drop the packet.
 215  *
 216  * --Lennert, 20020411
 217  * --Bart, 20020416 (updated)
 218  * --Bart, 20021007 (updated) */
 219 static int br_nf_pre_routing_finish_bridge(struct sk_buff *skb)
 220 {
 221         if (skb->pkt_type == PACKET_OTHERHOST) {
 222                 skb->pkt_type = PACKET_HOST;
 223                 skb->nf_bridge->mask |= BRNF_PKT_TYPE;
 224         }
 225         skb->nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
 226
 227         skb->dev = bridge_parent(skb->dev);
 228         if (!skb->dev)
 229                 kfree_skb(skb);
 230         else {
 231                 if (skb->protocol == htons(ETH_P_8021Q)) {
 232                         skb_pull(skb, VLAN_HLEN);
 233                         skb->nh.raw += VLAN_HLEN;
 234                 }
 235                 skb->dst->output(skb);
 236         }
 237         return 0;
 238 }
 239
 240 static int br_nf_pre_routing_finish(struct sk_buff *skb)
 241 {
 242         struct net_device *dev = skb->dev;
 243         struct iphdr *iph = skb->nh.iph;
 244         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 245
 246         if (nf_bridge->mask & BRNF_PKT_TYPE) {
 247                 skb->pkt_type = PACKET_OTHERHOST;
 248                 nf_bridge->mask ^= BRNF_PKT_TYPE;
 249         }
 250         nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
 251
 252         if (dnat_took_place(skb)) {
 253                 if (ip_route_input(skb, iph->daddr, iph->saddr, iph->tos, dev)) {
 254                         struct rtable *rt;
 255                         struct flowi fl = {
 256                                 .nl_u = {
 257                                         .ip4_u = {
 258                                                  .daddr = iph->daddr,
 259                                                  .saddr = 0,
 260                                                  .tos = RT_TOS(iph->tos) },
 261                                 },
 262                                 .proto = 0,
 263                         };
 264
 265                         if (!ip_route_output_key(&rt, &fl)) {
 266                                 /* - Bridged-and-DNAT'ed traffic doesn't
 267                                  *   require ip_forwarding.
 268                                  * - Deal with redirected traffic. */
 269                                 if (((struct dst_entry *)rt)->dev == dev ||
 270                                     rt->rt_type == RTN_LOCAL) {
 271                                         skb->dst = (struct dst_entry *)rt;
 272                                         goto bridged_dnat;
 273                                 }
 274                                 __br_dnat_complain();
 275                                 dst_release((struct dst_entry *)rt);
 276                         }
 277                         kfree_skb(skb);
 278                         return 0;
 279                 } else {
 280                         if (skb->dst->dev == dev) {
 281 bridged_dnat:
 282                                 /* Tell br_nf_local_out this is a
 283                                  * bridged frame */
 284                                 nf_bridge->mask |= BRNF_BRIDGED_DNAT;
 285                                 skb->dev = nf_bridge->physindev;
 286                                 if (skb->protocol ==
 287                                     htons(ETH_P_8021Q)) {
 288                                         skb_push(skb, VLAN_HLEN);
 289                                         skb->nh.raw -= VLAN_HLEN;
 290                                 }
 291                                 NF_HOOK_THRESH(PF_BRIDGE, NF_BR_PRE_ROUTING,
 292                                                skb, skb->dev, NULL,
 293                                                br_nf_pre_routing_finish_bridge,
 294                                                1);
 295                                 return 0;
 296                         }
 297                         memcpy(eth_hdr(skb)->h_dest, dev->dev_addr, ETH_ALEN);
 298                         skb->pkt_type = PACKET_HOST;
 299                 }
 300         } else {
 301                 skb->dst = (struct dst_entry *)&__fake_rtable;
 302                 dst_hold(skb->dst);
 303         }
 304
 305         skb->dev = nf_bridge->physindev;
 306         if (skb->protocol == htons(ETH_P_8021Q)) {
 307                 skb_push(skb, VLAN_HLEN);
 308                 skb->nh.raw -= VLAN_HLEN;
 309         }
 310         NF_HOOK_THRESH(PF_BRIDGE, NF_BR_PRE_ROUTING, skb, skb->dev, NULL,
 311                        br_handle_frame_finish, 1);
 312
 313         return 0;
 314 }
 315
 316 /* Some common code for IPv4/IPv6 */
 317 static struct net_device *setup_pre_routing(struct sk_buff *skb)
 318 {
 319         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 320
 321         if (skb->pkt_type == PACKET_OTHERHOST) {
 322                 skb->pkt_type = PACKET_HOST;
 323                 nf_bridge->mask |= BRNF_PKT_TYPE;
 324         }
 325
 326         nf_bridge->mask |= BRNF_NF_BRIDGE_PREROUTING;
 327         nf_bridge->physindev = skb->dev;
 328         skb->dev = bridge_parent(skb->dev);
 329
 330         return skb->dev;
 331 }
 332
 333 /* We only check the length. A bridge shouldn't do any hop-by-hop stuff anyway */
 334 static int check_hbh_len(struct sk_buff *skb)
 335 {
 336         unsigned char *raw = (u8 *) (skb->nh.ipv6h + 1);
 337         u32 pkt_len;
 338         int off = raw - skb->nh.raw;
 339         int len = (raw[1] + 1) << 3;
 340
 341         if ((raw + len) - skb->data > skb_headlen(skb))
 342                 goto bad;
 343
 344         off += 2;
 345         len -= 2;
 346
 347         while (len > 0) {
 348                 int optlen = skb->nh.raw[off + 1] + 2;
 349
 350                 switch (skb->nh.raw[off]) {
 351                 case IPV6_TLV_PAD0:
 352                         optlen = 1;
 353                         break;
 354
 355                 case IPV6_TLV_PADN:
 356                         break;
 357
 358                 case IPV6_TLV_JUMBO:
 359                         if (skb->nh.raw[off + 1] != 4 || (off & 3) != 2)
 360                                 goto bad;
 361                         pkt_len = ntohl(*(u32 *) (skb->nh.raw + off + 2));
 362                         if (pkt_len <= IPV6_MAXPLEN ||
 363                             skb->nh.ipv6h->payload_len)
 364                                 goto bad;
 365                         if (pkt_len > skb->len - sizeof(struct ipv6hdr))
 366                                 goto bad;
 367                         if (pskb_trim_rcsum(skb,
 368                                             pkt_len + sizeof(struct ipv6hdr)))
 369                                 goto bad;
 370                         break;
 371                 default:
 372                         if (optlen > len)
 373                                 goto bad;
 374                         break;
 375                 }
 376                 off += optlen;
 377                 len -= optlen;
 378         }
 379         if (len == 0)
 380                 return 0;
 381 bad:
 382         return -1;
 383
 384 }
 385
 386 /* Replicate the checks that IPv6 does on packet reception and pass the packet
 387  * to ip6tables, which doesn't support NAT, so things are fairly simple. */
 388 static unsigned int br_nf_pre_routing_ipv6(unsigned int hook,
 389                                            struct sk_buff *skb,
 390                                            const struct net_device *in,
 391                                            const struct net_device *out,
 392                                            int (*okfn)(struct sk_buff *))
 393 {
 394         struct ipv6hdr *hdr;
 395         u32 pkt_len;
 396
 397         if (skb->len < sizeof(struct ipv6hdr))
 398                 goto inhdr_error;
 399
 400         if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
 401                 goto inhdr_error;
 402
 403         hdr = skb->nh.ipv6h;
 404
 405         if (hdr->version != 6)
 406                 goto inhdr_error;
 407
 408         pkt_len = ntohs(hdr->payload_len);
 409
 410         if (pkt_len || hdr->nexthdr != NEXTHDR_HOP) {
 411                 if (pkt_len + sizeof(struct ipv6hdr) > skb->len)
 412                         goto inhdr_error;
 413                 if (pskb_trim_rcsum(skb, pkt_len + sizeof(struct ipv6hdr)))
 414                         goto inhdr_error;
 415         }
 416         if (hdr->nexthdr == NEXTHDR_HOP && check_hbh_len(skb))
 417                 goto inhdr_error;
 418
 419         nf_bridge_put(skb->nf_bridge);
 420         if (!nf_bridge_alloc(skb))
 421                 return NF_DROP;
 422         if (!setup_pre_routing(skb))
 423                 return NF_DROP;
 424
 425         NF_HOOK(PF_INET6, NF_IP6_PRE_ROUTING, skb, skb->dev, NULL,
 426                 br_nf_pre_routing_finish_ipv6);
 427
 428         return NF_STOLEN;
 429
 430 inhdr_error:
 431         return NF_DROP;
 432 }
 433
 434 /* Direct IPv6 traffic to br_nf_pre_routing_ipv6.
 435  * Replicate the checks that IPv4 does on packet reception.
 436  * Set skb->dev to the bridge device (i.e. parent of the
 437  * receiving device) to make netfilter happy, the REDIRECT
 438  * target in particular.  Save the original destination IP
 439  * address to be able to detect DNAT afterwards. */
 440 static unsigned int br_nf_pre_routing(unsigned int hook, struct sk_buff **pskb,
 441                                       const struct net_device *in,
 442                                       const struct net_device *out,
 443                                       int (*okfn)(struct sk_buff *))
 444 {
 445         struct iphdr *iph;
 446         __u32 len;
 447         struct sk_buff *skb = *pskb;
 448
 449         if (skb->protocol == htons(ETH_P_IPV6) || IS_VLAN_IPV6(skb)) {
 450 #ifdef CONFIG_SYSCTL
 451                 if (!brnf_call_ip6tables)
 452                         return NF_ACCEPT;
 453 #endif
 454                 if ((skb = skb_share_check(*pskb, GFP_ATOMIC)) == NULL)
 455                         goto out;
 456
 457                 if (skb->protocol == htons(ETH_P_8021Q)) {
 458                         skb_pull_rcsum(skb, VLAN_HLEN);
 459                         skb->nh.raw += VLAN_HLEN;
 460                 }
 461                 return br_nf_pre_routing_ipv6(hook, skb, in, out, okfn);
 462         }
 463 #ifdef CONFIG_SYSCTL
 464         if (!brnf_call_iptables)
 465                 return NF_ACCEPT;
 466 #endif
 467
 468         if (skb->protocol != htons(ETH_P_IP) && !IS_VLAN_IP(skb))
 469                 return NF_ACCEPT;
 470
 471         if ((skb = skb_share_check(*pskb, GFP_ATOMIC)) == NULL)
 472                 goto out;
 473
 474         if (skb->protocol == htons(ETH_P_8021Q)) {
 475                 skb_pull_rcsum(skb, VLAN_HLEN);
 476                 skb->nh.raw += VLAN_HLEN;
 477         }
 478
 479         if (!pskb_may_pull(skb, sizeof(struct iphdr)))
 480                 goto inhdr_error;
 481
 482         iph = skb->nh.iph;
 483         if (iph->ihl < 5 || iph->version != 4)
 484                 goto inhdr_error;
 485
 486         if (!pskb_may_pull(skb, 4 * iph->ihl))
 487                 goto inhdr_error;
 488
 489         iph = skb->nh.iph;
 490         if (ip_fast_csum((__u8 *) iph, iph->ihl) != 0)
 491                 goto inhdr_error;
 492
 493         len = ntohs(iph->tot_len);
 494         if (skb->len < len || len < 4 * iph->ihl)
 495                 goto inhdr_error;
 496
 497         pskb_trim_rcsum(skb, len);
 498
 499         nf_bridge_put(skb->nf_bridge);
 500         if (!nf_bridge_alloc(skb))
 501                 return NF_DROP;
 502         if (!setup_pre_routing(skb))
 503                 return NF_DROP;
 504         store_orig_dstaddr(skb);
 505
 506         NF_HOOK(PF_INET, NF_IP_PRE_ROUTING, skb, skb->dev, NULL,
 507                 br_nf_pre_routing_finish);
 508
 509         return NF_STOLEN;
 510
 511 inhdr_error:
 512 //      IP_INC_STATS_BH(IpInHdrErrors);
 513 out:
 514         return NF_DROP;
 515 }
 516
 517
 518 /* PF_BRIDGE/LOCAL_IN ************************************************/
 519 /* The packet is locally destined, which requires a real
 520  * dst_entry, so detach the fake one.  On the way up, the
 521  * packet would pass through PRE_ROUTING again (which already
 522  * took place when the packet entered the bridge), but we
 523  * register an IPv4 PRE_ROUTING 'sabotage' hook that will
 524  * prevent this from happening. */
 525 static unsigned int br_nf_local_in(unsigned int hook, struct sk_buff **pskb,
 526                                    const struct net_device *in,
 527                                    const struct net_device *out,
 528                                    int (*okfn)(struct sk_buff *))
 529 {
 530         struct sk_buff *skb = *pskb;
 531
 532         if (skb->dst == (struct dst_entry *)&__fake_rtable) {
 533                 dst_release(skb->dst);
 534                 skb->dst = NULL;
 535         }
 536
 537         return NF_ACCEPT;
 538 }
 539
 540 /* PF_BRIDGE/FORWARD *************************************************/
 541 static int br_nf_forward_finish(struct sk_buff *skb)
 542 {
 543         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 544         struct net_device *in;
 545
 546         if (skb->protocol != htons(ETH_P_ARP) && !IS_VLAN_ARP(skb)) {
 547                 in = nf_bridge->physindev;
 548                 if (nf_bridge->mask & BRNF_PKT_TYPE) {
 549                         skb->pkt_type = PACKET_OTHERHOST;
 550                         nf_bridge->mask ^= BRNF_PKT_TYPE;
 551                 }
 552         } else {
 553                 in = *((struct net_device **)(skb->cb));
 554         }
 555         if (skb->protocol == htons(ETH_P_8021Q)) {
 556                 skb_push(skb, VLAN_HLEN);
 557                 skb->nh.raw -= VLAN_HLEN;
 558         }
 559         NF_HOOK_THRESH(PF_BRIDGE, NF_BR_FORWARD, skb, in,
 560                        skb->dev, br_forward_finish, 1);
 561         return 0;
 562 }
 563
 564 /* This is the 'purely bridged' case.  For IP, we pass the packet to
 565  * netfilter with indev and outdev set to the bridge device,
 566  * but we are still able to filter on the 'real' indev/outdev
 567  * because of the physdev module. For ARP, indev and outdev are the
 568  * bridge ports. */
 569 static unsigned int br_nf_forward_ip(unsigned int hook, struct sk_buff **pskb,
 570                                      const struct net_device *in,
 571                                      const struct net_device *out,
 572                                      int (*okfn)(struct sk_buff *))
 573 {
 574         struct sk_buff *skb = *pskb;
 575         struct nf_bridge_info *nf_bridge;
 576         struct net_device *parent;
 577         int pf;
 578
 579         if (!skb->nf_bridge)
 580                 return NF_ACCEPT;
 581
 582         parent = bridge_parent(out);
 583         if (!parent)
 584                 return NF_DROP;
 585
 586         if (skb->protocol == htons(ETH_P_IP) || IS_VLAN_IP(skb))
 587                 pf = PF_INET;
 588         else
 589                 pf = PF_INET6;
 590
 591         if (skb->protocol == htons(ETH_P_8021Q)) {
 592                 skb_pull(*pskb, VLAN_HLEN);
 593                 (*pskb)->nh.raw += VLAN_HLEN;
 594         }
 595
 596         nf_bridge = skb->nf_bridge;
 597         if (skb->pkt_type == PACKET_OTHERHOST) {
 598                 skb->pkt_type = PACKET_HOST;
 599                 nf_bridge->mask |= BRNF_PKT_TYPE;
 600         }
 601
 602         /* The physdev module checks on this */
 603         nf_bridge->mask |= BRNF_BRIDGED;
 604         nf_bridge->physoutdev = skb->dev;
 605
 606         NF_HOOK(pf, NF_IP_FORWARD, skb, bridge_parent(in), parent,
 607                 br_nf_forward_finish);
 608
 609         return NF_STOLEN;
 610 }
 611
 612 static unsigned int br_nf_forward_arp(unsigned int hook, struct sk_buff **pskb,
 613                                       const struct net_device *in,
 614                                       const struct net_device *out,
 615                                       int (*okfn)(struct sk_buff *))
 616 {
 617         struct sk_buff *skb = *pskb;
 618         struct net_device **d = (struct net_device **)(skb->cb);
 619
 620 #ifdef CONFIG_SYSCTL
 621         if (!brnf_call_arptables)
 622                 return NF_ACCEPT;
 623 #endif
 624
 625         if (skb->protocol != htons(ETH_P_ARP)) {
 626                 if (!IS_VLAN_ARP(skb))
 627                         return NF_ACCEPT;
 628                 skb_pull(*pskb, VLAN_HLEN);
 629                 (*pskb)->nh.raw += VLAN_HLEN;
 630         }
 631
 632         if (skb->nh.arph->ar_pln != 4) {
 633                 if (IS_VLAN_ARP(skb)) {
 634                         skb_push(*pskb, VLAN_HLEN);
 635                         (*pskb)->nh.raw -= VLAN_HLEN;
 636                 }
 637                 return NF_ACCEPT;
 638         }
 639         *d = (struct net_device *)in;
 640         NF_HOOK(NF_ARP, NF_ARP_FORWARD, skb, (struct net_device *)in,
 641                 (struct net_device *)out, br_nf_forward_finish);
 642
 643         return NF_STOLEN;
 644 }
 645
 646 /* PF_BRIDGE/LOCAL_OUT ***********************************************/
 647 static int br_nf_local_out_finish(struct sk_buff *skb)
 648 {
 649         if (skb->protocol == htons(ETH_P_8021Q)) {
 650                 skb_push(skb, VLAN_HLEN);
 651                 skb->nh.raw -= VLAN_HLEN;
 652         }
 653
 654         NF_HOOK_THRESH(PF_BRIDGE, NF_BR_LOCAL_OUT, skb, NULL, skb->dev,
 655                        br_forward_finish, NF_BR_PRI_FIRST + 1);
 656
 657         return 0;
 658 }
 659
 660 /* This function sees both locally originated IP packets and forwarded
 661  * IP packets (in both cases the destination device is a bridge
 662  * device). It also sees bridged-and-DNAT'ed packets.
 663  * To be able to filter on the physical bridge devices (with the physdev
 664  * module), we steal packets destined to a bridge device away from the
 665  * PF_INET/FORWARD and PF_INET/OUTPUT hook functions, and give them back later,
 666  * when we have determined the real output device. This is done in here.
 667  *
 668  * If (nf_bridge->mask & BRNF_BRIDGED_DNAT) then the packet is bridged
 669  * and we fake the PF_BRIDGE/FORWARD hook. The function br_nf_forward()
 670  * will then fake the PF_INET/FORWARD hook. br_nf_local_out() has priority
 671  * NF_BR_PRI_FIRST, so no relevant PF_BRIDGE/INPUT functions have been nor
 672  * will be executed.
 673  * Otherwise, if nf_bridge->physindev is NULL, the bridge-nf code never touched
 674  * this packet before, and so the packet was locally originated. We fake
 675  * the PF_INET/LOCAL_OUT hook.
 676  * Finally, if nf_bridge->physindev isn't NULL, then the packet was IP routed,
 677  * so we fake the PF_INET/FORWARD hook. ip_sabotage_out() makes sure
 678  * even routed packets that didn't arrive on a bridge interface have their
 679  * nf_bridge->physindev set. */
 680 static unsigned int br_nf_local_out(unsigned int hook, struct sk_buff **pskb,
 681                                     const struct net_device *in,
 682                                     const struct net_device *out,
 683                                     int (*okfn)(struct sk_buff *))
 684 {
 685         struct net_device *realindev, *realoutdev;
 686         struct sk_buff *skb = *pskb;
 687         struct nf_bridge_info *nf_bridge;
 688         int pf;
 689
 690         if (!skb->nf_bridge)
 691                 return NF_ACCEPT;
 692
 693         if (skb->protocol == htons(ETH_P_IP) || IS_VLAN_IP(skb))
 694                 pf = PF_INET;
 695         else
 696                 pf = PF_INET6;
 697
 698 #ifdef CONFIG_NETFILTER_DEBUG
 699         /* Sometimes we get packets with NULL ->dst here (for example,
 700          * running a dhcp client daemon triggers this). This should now
 701          * be fixed, but let's keep the check around. */
 702         if (skb->dst == NULL) {
 703                 printk(KERN_CRIT "br_netfilter: skb->dst == NULL.");
 704                 return NF_ACCEPT;
 705         }
 706 #endif
 707
 708         nf_bridge = skb->nf_bridge;
 709         nf_bridge->physoutdev = skb->dev;
 710         realindev = nf_bridge->physindev;
 711
 712         /* Bridged, take PF_BRIDGE/FORWARD.
 713          * (see big note in front of br_nf_pre_routing_finish) */
 714         if (nf_bridge->mask & BRNF_BRIDGED_DNAT) {
 715                 if (nf_bridge->mask & BRNF_PKT_TYPE) {
 716                         skb->pkt_type = PACKET_OTHERHOST;
 717                         nf_bridge->mask ^= BRNF_PKT_TYPE;
 718                 }
 719                 if (skb->protocol == htons(ETH_P_8021Q)) {
 720                         skb_push(skb, VLAN_HLEN);
 721                         skb->nh.raw -= VLAN_HLEN;
 722                 }
 723
 724                 NF_HOOK(PF_BRIDGE, NF_BR_FORWARD, skb, realindev,
 725                         skb->dev, br_forward_finish);
 726                 goto out;
 727         }
 728         realoutdev = bridge_parent(skb->dev);
 729         if (!realoutdev)
 730                 return NF_DROP;
 731
 732 #if defined(CONFIG_VLAN_8021Q) || defined(CONFIG_VLAN_8021Q_MODULE)
 733         /* iptables should match -o br0.x */
 734         if (nf_bridge->netoutdev)
 735                 realoutdev = nf_bridge->netoutdev;
 736 #endif
 737         if (skb->protocol == htons(ETH_P_8021Q)) {
 738                 skb_pull(skb, VLAN_HLEN);
 739                 (*pskb)->nh.raw += VLAN_HLEN;
 740         }
 741         /* IP forwarded traffic has a physindev, locally
 742          * generated traffic hasn't. */
 743         if (realindev != NULL) {
 744                 if (!(nf_bridge->mask & BRNF_DONT_TAKE_PARENT)) {
 745                         struct net_device *parent = bridge_parent(realindev);
 746                         if (parent)
 747                                 realindev = parent;
 748                 }
 749
 750                 NF_HOOK_THRESH(pf, NF_IP_FORWARD, skb, realindev,
 751                                realoutdev, br_nf_local_out_finish,
 752                                NF_IP_PRI_BRIDGE_SABOTAGE_FORWARD + 1);
 753         } else {
 754                 NF_HOOK_THRESH(pf, NF_IP_LOCAL_OUT, skb, realindev,
 755                                realoutdev, br_nf_local_out_finish,
 756                                NF_IP_PRI_BRIDGE_SABOTAGE_LOCAL_OUT + 1);
 757         }
 758
 759 out:
 760         return NF_STOLEN;
 761 }
 762
 763 static int br_nf_dev_queue_xmit(struct sk_buff *skb)
 764 {
 765         if (skb->protocol == htons(ETH_P_IP) &&
 766             skb->len > skb->dev->mtu &&
 767             !skb_is_gso(skb))
 768                 return ip_fragment(skb, br_dev_queue_push_xmit);
 769         else
 770                 return br_dev_queue_push_xmit(skb);
 771 }
 772
 773 /* PF_BRIDGE/POST_ROUTING ********************************************/
 774 static unsigned int br_nf_post_routing(unsigned int hook, struct sk_buff **pskb,
 775                                        const struct net_device *in,
 776                                        const struct net_device *out,
 777                                        int (*okfn)(struct sk_buff *))
 778 {
 779         struct sk_buff *skb = *pskb;
 780         struct nf_bridge_info *nf_bridge = (*pskb)->nf_bridge;
 781         struct net_device *realoutdev = bridge_parent(skb->dev);
 782         int pf;
 783
 784 #ifdef CONFIG_NETFILTER_DEBUG
 785         /* Be very paranoid. This probably won't happen anymore, but let's
 786          * keep the check just to be sure... */
 787         if (skb->mac.raw < skb->head || skb->mac.raw + ETH_HLEN > skb->data) {
 788                 printk(KERN_CRIT "br_netfilter: Argh!! br_nf_post_routing: "
 789                        "bad mac.raw pointer.");
 790                 goto print_error;
 791         }
 792 #endif
 793
 794         if (!nf_bridge)
 795                 return NF_ACCEPT;
 796
 797         if (!realoutdev)
 798                 return NF_DROP;
 799
 800         if (skb->protocol == htons(ETH_P_IP) || IS_VLAN_IP(skb))
 801                 pf = PF_INET;
 802         else
 803                 pf = PF_INET6;
 804
 805 #ifdef CONFIG_NETFILTER_DEBUG
 806         if (skb->dst == NULL) {
 807                 printk(KERN_CRIT "br_netfilter: skb->dst == NULL.");
 808                 goto print_error;
 809         }
 810 #endif
 811
 812         /* We assume any code from br_dev_queue_push_xmit onwards doesn't care
 813          * about the value of skb->pkt_type. */
 814         if (skb->pkt_type == PACKET_OTHERHOST) {
 815                 skb->pkt_type = PACKET_HOST;
 816                 nf_bridge->mask |= BRNF_PKT_TYPE;
 817         }
 818
 819         if (skb->protocol == htons(ETH_P_8021Q)) {
 820                 skb_pull(skb, VLAN_HLEN);
 821                 skb->nh.raw += VLAN_HLEN;
 822         }
 823
 824         nf_bridge_save_header(skb);
 825
 826 #if defined(CONFIG_VLAN_8021Q) || defined(CONFIG_VLAN_8021Q_MODULE)
 827         if (nf_bridge->netoutdev)
 828                 realoutdev = nf_bridge->netoutdev;
 829 #endif
 830         NF_HOOK(pf, NF_IP_POST_ROUTING, skb, NULL, realoutdev,
 831                 br_nf_dev_queue_xmit);
 832
 833         return NF_STOLEN;
 834
 835 #ifdef CONFIG_NETFILTER_DEBUG
 836 print_error:
 837         if (skb->dev != NULL) {
 838                 printk("[%s]", skb->dev->name);
 839                 if (realoutdev)
 840                         printk("[%s]", realoutdev->name);
 841         }
 842         printk(" head:%p, raw:%p, data:%p\n", skb->head, skb->mac.raw,
 843                skb->data);
 844         return NF_ACCEPT;
 845 #endif
 846 }
 847
 848 /* IP/SABOTAGE *****************************************************/
 849 /* Don't hand locally destined packets to PF_INET(6)/PRE_ROUTING
 850  * for the second time. */
 851 static unsigned int ip_sabotage_in(unsigned int hook, struct sk_buff **pskb,
 852                                    const struct net_device *in,
 853                                    const struct net_device *out,
 854                                    int (*okfn)(struct sk_buff *))
 855 {
 856         if ((*pskb)->nf_bridge &&
 857             !((*pskb)->nf_bridge->mask & BRNF_NF_BRIDGE_PREROUTING)) {
 858                 return NF_STOP;
 859         }
 860
 861         return NF_ACCEPT;
 862 }
 863
 864 /* Postpone execution of PF_INET(6)/FORWARD, PF_INET(6)/LOCAL_OUT
 865  * and PF_INET(6)/POST_ROUTING until we have done the forwarding
 866  * decision in the bridge code and have determined nf_bridge->physoutdev. */
 867 static unsigned int ip_sabotage_out(unsigned int hook, struct sk_buff **pskb,
 868                                     const struct net_device *in,
 869                                     const struct net_device *out,
 870                                     int (*okfn)(struct sk_buff *))
 871 {
 872         struct sk_buff *skb = *pskb;
 873
 874         if ((out->hard_start_xmit == br_dev_xmit &&
 875              okfn != br_nf_forward_finish &&
 876              okfn != br_nf_local_out_finish && okfn != br_nf_dev_queue_xmit)
 877 #if defined(CONFIG_VLAN_8021Q) || defined(CONFIG_VLAN_8021Q_MODULE)
 878             || ((out->priv_flags & IFF_802_1Q_VLAN) &&
 879                 VLAN_DEV_INFO(out)->real_dev->hard_start_xmit == br_dev_xmit)
 880 #endif
 881             ) {
 882                 struct nf_bridge_info *nf_bridge;
 883
 884                 if (!skb->nf_bridge) {
 885 #ifdef CONFIG_SYSCTL
 886                         /* This code is executed while in the IP(v6) stack,
 887                            the version should be 4 or 6. We can't use
 888                            skb->protocol because that isn't set on
 889                            PF_INET(6)/LOCAL_OUT. */
 890                         struct iphdr *ip = skb->nh.iph;
 891
 892                         if (ip->version == 4 && !brnf_call_iptables)
 893                                 return NF_ACCEPT;
 894                         else if (ip->version == 6 && !brnf_call_ip6tables)
 895                                 return NF_ACCEPT;
 896                         else if (!brnf_deferred_hooks)
 897                                 return NF_ACCEPT;
 898 #endif
 899                         if (hook == NF_IP_POST_ROUTING)
 900                                 return NF_ACCEPT;
 901                         if (!nf_bridge_alloc(skb))
 902                                 return NF_DROP;
 903                 }
 904
 905                 nf_bridge = skb->nf_bridge;
 906
 907                 /* This frame will arrive on PF_BRIDGE/LOCAL_OUT and we
 908                  * will need the indev then. For a brouter, the real indev
 909                  * can be a bridge port, so we make sure br_nf_local_out()
 910                  * doesn't use the bridge parent of the indev by using
 911                  * the BRNF_DONT_TAKE_PARENT mask. */
 912                 if (hook == NF_IP_FORWARD && nf_bridge->physindev == NULL) {
 913                         nf_bridge->mask |= BRNF_DONT_TAKE_PARENT;
 914                         nf_bridge->physindev = (struct net_device *)in;
 915                 }
 916 #if defined(CONFIG_VLAN_8021Q) || defined(CONFIG_VLAN_8021Q_MODULE)
 917                 /* the iptables outdev is br0.x, not br0 */
 918                 if (out->priv_flags & IFF_802_1Q_VLAN)
 919                         nf_bridge->netoutdev = (struct net_device *)out;
 920 #endif
 921                 return NF_STOP;
 922         }
 923
 924         return NF_ACCEPT;
 925 }
 926
 927 /* For br_nf_local_out we need (prio = NF_BR_PRI_FIRST), to insure that innocent
 928  * PF_BRIDGE/NF_BR_LOCAL_OUT functions don't get bridged traffic as input.
 929  * For br_nf_post_routing, we need (prio = NF_BR_PRI_LAST), because
 930  * ip_refrag() can return NF_STOLEN. */
 931 static struct nf_hook_ops br_nf_ops[] = {
 932         { .hook = br_nf_pre_routing,
 933           .owner = THIS_MODULE,
 934           .pf = PF_BRIDGE,
 935           .hooknum = NF_BR_PRE_ROUTING,
 936           .priority = NF_BR_PRI_BRNF, },
 937         { .hook = br_nf_local_in,
 938           .owner = THIS_MODULE,
 939           .pf = PF_BRIDGE,
 940           .hooknum = NF_BR_LOCAL_IN,
 941           .priority = NF_BR_PRI_BRNF, },
 942         { .hook = br_nf_forward_ip,
 943           .owner = THIS_MODULE,
 944           .pf = PF_BRIDGE,
 945           .hooknum = NF_BR_FORWARD,
 946           .priority = NF_BR_PRI_BRNF - 1, },
 947         { .hook = br_nf_forward_arp,
 948           .owner = THIS_MODULE,
 949           .pf = PF_BRIDGE,
 950           .hooknum = NF_BR_FORWARD,
 951           .priority = NF_BR_PRI_BRNF, },
 952         { .hook = br_nf_local_out,
 953           .owner = THIS_MODULE,
 954           .pf = PF_BRIDGE,
 955           .hooknum = NF_BR_LOCAL_OUT,
 956           .priority = NF_BR_PRI_FIRST, },
 957         { .hook = br_nf_post_routing,
 958           .owner = THIS_MODULE,
 959           .pf = PF_BRIDGE,
 960           .hooknum = NF_BR_POST_ROUTING,
 961           .priority = NF_BR_PRI_LAST, },
 962         { .hook = ip_sabotage_in,
 963           .owner = THIS_MODULE,
 964           .pf = PF_INET,
 965           .hooknum = NF_IP_PRE_ROUTING,
 966           .priority = NF_IP_PRI_FIRST, },
 967         { .hook = ip_sabotage_in,
 968           .owner = THIS_MODULE,
 969           .pf = PF_INET6,
 970           .hooknum = NF_IP6_PRE_ROUTING,
 971           .priority = NF_IP6_PRI_FIRST, },
 972         { .hook = ip_sabotage_out,
 973           .owner = THIS_MODULE,
 974           .pf = PF_INET,
 975           .hooknum = NF_IP_FORWARD,
 976           .priority = NF_IP_PRI_BRIDGE_SABOTAGE_FORWARD, },
 977         { .hook = ip_sabotage_out,
 978           .owner = THIS_MODULE,
 979           .pf = PF_INET6,
 980           .hooknum = NF_IP6_FORWARD,
 981           .priority = NF_IP6_PRI_BRIDGE_SABOTAGE_FORWARD, },
 982         { .hook = ip_sabotage_out,
 983           .owner = THIS_MODULE,
 984           .pf = PF_INET,
 985           .hooknum = NF_IP_LOCAL_OUT,
 986           .priority = NF_IP_PRI_BRIDGE_SABOTAGE_LOCAL_OUT, },
 987         { .hook = ip_sabotage_out,
 988           .owner = THIS_MODULE,
 989           .pf = PF_INET6,
 990           .hooknum = NF_IP6_LOCAL_OUT,
 991           .priority = NF_IP6_PRI_BRIDGE_SABOTAGE_LOCAL_OUT, },
 992         { .hook = ip_sabotage_out,
 993           .owner = THIS_MODULE,
 994           .pf = PF_INET,
 995           .hooknum = NF_IP_POST_ROUTING,
 996           .priority = NF_IP_PRI_FIRST, },
 997         { .hook = ip_sabotage_out,
 998           .owner = THIS_MODULE,
 999           .pf = PF_INET6,
1000           .hooknum = NF_IP6_POST_ROUTING,
1001           .priority = NF_IP6_PRI_FIRST, },
1002 };
1003
1004 #ifdef CONFIG_SYSCTL
1005 static
1006 int brnf_sysctl_call_tables(ctl_table * ctl, int write, struct file *filp,
1007                             void __user * buffer, size_t * lenp, loff_t * ppos)
1008 {
1009         int ret;
1010
1011         ret = proc_dointvec(ctl, write, filp, buffer, lenp, ppos);
1012
1013         if (write && *(int *)(ctl->data))
1014                 *(int *)(ctl->data) = 1;
1015         return ret;
1016 }
1017
1018 static ctl_table brnf_table[] = {
1019         {
1020                 .ctl_name       = NET_BRIDGE_NF_CALL_ARPTABLES,
1021                 .procname       = "bridge-nf-call-arptables",
1022                 .data           = &brnf_call_arptables,
1023                 .maxlen         = sizeof(int),
1024                 .mode           = 0644,
1025                 .proc_handler   = &brnf_sysctl_call_tables,
1026         },
1027         {
1028                 .ctl_name       = NET_BRIDGE_NF_CALL_IPTABLES,
1029                 .procname       = "bridge-nf-call-iptables",
1030                 .data           = &brnf_call_iptables,
1031                 .maxlen         = sizeof(int),
1032                 .mode           = 0644,
1033                 .proc_handler   = &brnf_sysctl_call_tables,
1034         },
1035         {
1036                 .ctl_name       = NET_BRIDGE_NF_CALL_IP6TABLES,
1037                 .procname       = "bridge-nf-call-ip6tables",
1038                 .data           = &brnf_call_ip6tables,
1039                 .maxlen         = sizeof(int),
1040                 .mode           = 0644,
1041                 .proc_handler   = &brnf_sysctl_call_tables,
1042         },
1043         {
1044                 .ctl_name       = NET_BRIDGE_NF_FILTER_VLAN_TAGGED,
1045                 .procname       = "bridge-nf-filter-vlan-tagged",
1046                 .data           = &brnf_filter_vlan_tagged,
1047                 .maxlen         = sizeof(int),
1048                 .mode           = 0644,
1049                 .proc_handler   = &brnf_sysctl_call_tables,
1050         },
1051         { .ctl_name = 0 }
1052 };
1053
1054 static ctl_table brnf_bridge_table[] = {
1055         {
1056                 .ctl_name       = NET_BRIDGE,
1057                 .procname       = "bridge",
1058                 .mode           = 0555,
1059                 .child          = brnf_table,
1060         },
1061         { .ctl_name = 0 }
1062 };
1063
1064 static ctl_table brnf_net_table[] = {
1065         {
1066                 .ctl_name       = CTL_NET,
1067                 .procname       = "net",
1068                 .mode           = 0555,
1069                 .child          = brnf_bridge_table,
1070         },
1071         { .ctl_name = 0 }
1072 };
1073 #endif
1074
1075 int br_netfilter_init(void)
1076 {
1077         int i;
1078
1079         for (i = 0; i < ARRAY_SIZE(br_nf_ops); i++) {
1080                 int ret;
1081
1082                 if ((ret = nf_register_hook(&br_nf_ops[i])) >= 0)
1083                         continue;
1084
1085                 while (i--)
1086                         nf_unregister_hook(&br_nf_ops[i]);
1087
1088                 return ret;
1089         }
1090
1091 #ifdef CONFIG_SYSCTL
1092         brnf_sysctl_header = register_sysctl_table(brnf_net_table, 0);
1093         if (brnf_sysctl_header == NULL) {
1094                 printk(KERN_WARNING
1095                        "br_netfilter: can't register to sysctl.\n");
1096                 for (i = 0; i < ARRAY_SIZE(br_nf_ops); i++)
1097                         nf_unregister_hook(&br_nf_ops[i]);
1098                 return -EFAULT;
1099         }
1100 #endif
1101
1102         printk(KERN_NOTICE "Bridge firewalling registered\n");
1103
1104         return 0;
1105 }
1106
1107 void br_netfilter_fini(void)
1108 {
1109         int i;
1110
1111         for (i = ARRAY_SIZE(br_nf_ops) - 1; i >= 0; i--)
1112                 nf_unregister_hook(&br_nf_ops[i]);
1113 #ifdef CONFIG_SYSCTL
1114         unregister_sysctl_table(brnf_sysctl_header);
1115 #endif
1116 }