git.oblomov.eu Git - linux-2.6/blob - net/bridge/br_netfilter.c

   1 /*
   2  *      Handle firewalling
   3  *      Linux ethernet bridge
   4  *
   5  *      Authors:
   6  *      Lennert Buytenhek               <buytenh@gnu.org>
   7  *      Bart De Schuymer (maintainer)   <bdschuym@pandora.be>
   8  *
   9  *      Changes:
  10  *      Apr 29 2003: physdev module support (bdschuym)
  11  *      Jun 19 2003: let arptables see bridged ARP traffic (bdschuym)
  12  *      Oct 06 2003: filter encapsulated IP/ARP VLAN traffic on untagged bridge
  13  *                   (bdschuym)
  14  *      Sep 01 2004: add IPv6 filtering (bdschuym)
  15  *
  16  *      This program is free software; you can redistribute it and/or
  17  *      modify it under the terms of the GNU General Public License
  18  *      as published by the Free Software Foundation; either version
  19  *      2 of the License, or (at your option) any later version.
  20  *
  21  *      Lennert dedicates this file to Kerstin Wurdinger.
  22  */
  23
  24 #include <linux/module.h>
  25 #include <linux/kernel.h>
  26 #include <linux/ip.h>
  27 #include <linux/netdevice.h>
  28 #include <linux/skbuff.h>
  29 #include <linux/if_arp.h>
  30 #include <linux/if_ether.h>
  31 #include <linux/if_vlan.h>
  32 #include <linux/netfilter_bridge.h>
  33 #include <linux/netfilter_ipv4.h>
  34 #include <linux/netfilter_ipv6.h>
  35 #include <linux/netfilter_arp.h>
  36 #include <linux/in_route.h>
  37
  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 static __be16 inline vlan_proto(const struct sk_buff *skb)
  65 {
  66         return vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
  67 }
  68
  69 #define IS_VLAN_IP(skb) \
  70         (skb->protocol == htons(ETH_P_8021Q) && \
  71          vlan_proto(skb) == htons(ETH_P_IP) &&  \
  72          brnf_filter_vlan_tagged)
  73
  74 #define IS_VLAN_IPV6(skb) \
  75         (skb->protocol == htons(ETH_P_8021Q) && \
  76          vlan_proto(skb) == htons(ETH_P_IPV6) &&\
  77          brnf_filter_vlan_tagged)
  78
  79 #define IS_VLAN_ARP(skb) \
  80         (skb->protocol == htons(ETH_P_8021Q) && \
  81          vlan_proto(skb) == htons(ETH_P_ARP) && \
  82          brnf_filter_vlan_tagged)
  83
  84 /* We need these fake structures to make netfilter happy --
  85  * lots of places assume that skb->dst != NULL, which isn't
  86  * all that unreasonable.
  87  *
  88  * Currently, we fill in the PMTU entry because netfilter
  89  * refragmentation needs it, and the rt_flags entry because
  90  * ipt_REJECT needs it.  Future netfilter modules might
  91  * require us to fill additional fields. */
  92 static struct net_device __fake_net_device = {
  93         .hard_header_len        = ETH_HLEN
  94 };
  95
  96 static struct rtable __fake_rtable = {
  97         .u = {
  98                 .dst = {
  99                         .__refcnt               = ATOMIC_INIT(1),
 100                         .dev                    = &__fake_net_device,
 101                         .path                   = &__fake_rtable.u.dst,
 102                         .metrics                = {[RTAX_MTU - 1] = 1500},
 103                         .flags                  = DST_NOXFRM,
 104                 }
 105         },
 106         .rt_flags       = 0,
 107 };
 108
 109 static inline struct net_device *bridge_parent(const struct net_device *dev)
 110 {
 111         struct net_bridge_port *port = rcu_dereference(dev->br_port);
 112
 113         return port ? port->br->dev : NULL;
 114 }
 115
 116 static inline struct nf_bridge_info *nf_bridge_alloc(struct sk_buff *skb)
 117 {
 118         skb->nf_bridge = kzalloc(sizeof(struct nf_bridge_info), GFP_ATOMIC);
 119         if (likely(skb->nf_bridge))
 120                 atomic_set(&(skb->nf_bridge->use), 1);
 121
 122         return skb->nf_bridge;
 123 }
 124
 125 static inline void nf_bridge_save_header(struct sk_buff *skb)
 126 {
 127         int header_size = 16;
 128
 129         if (skb->protocol == htons(ETH_P_8021Q))
 130                 header_size = 18;
 131
 132         memcpy(skb->nf_bridge->data, skb->data - header_size, header_size);
 133 }
 134
 135 /* PF_BRIDGE/PRE_ROUTING *********************************************/
 136 /* Undo the changes made for ip6tables PREROUTING and continue the
 137  * bridge PRE_ROUTING hook. */
 138 static int br_nf_pre_routing_finish_ipv6(struct sk_buff *skb)
 139 {
 140         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 141
 142         if (nf_bridge->mask & BRNF_PKT_TYPE) {
 143                 skb->pkt_type = PACKET_OTHERHOST;
 144                 nf_bridge->mask ^= BRNF_PKT_TYPE;
 145         }
 146         nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
 147
 148         skb->dst = (struct dst_entry *)&__fake_rtable;
 149         dst_hold(skb->dst);
 150
 151         skb->dev = nf_bridge->physindev;
 152         if (skb->protocol == htons(ETH_P_8021Q)) {
 153                 skb_push(skb, VLAN_HLEN);
 154                 skb->nh.raw -= VLAN_HLEN;
 155         }
 156         NF_HOOK_THRESH(PF_BRIDGE, NF_BR_PRE_ROUTING, skb, skb->dev, NULL,
 157                        br_handle_frame_finish, 1);
 158
 159         return 0;
 160 }
 161
 162 static void __br_dnat_complain(void)
 163 {
 164         static unsigned long last_complaint;
 165
 166         if (jiffies - last_complaint >= 5 * HZ) {
 167                 printk(KERN_WARNING "Performing cross-bridge DNAT requires IP "
 168                        "forwarding to be enabled\n");
 169                 last_complaint = jiffies;
 170         }
 171 }
 172
 173 /* This requires some explaining. If DNAT has taken place,
 174  * we will need to fix up the destination Ethernet address,
 175  * and this is a tricky process.
 176  *
 177  * There are two cases to consider:
 178  * 1. The packet was DNAT'ed to a device in the same bridge
 179  *    port group as it was received on. We can still bridge
 180  *    the packet.
 181  * 2. The packet was DNAT'ed to a different device, either
 182  *    a non-bridged device or another bridge port group.
 183  *    The packet will need to be routed.
 184  *
 185  * The correct way of distinguishing between these two cases is to
 186  * call ip_route_input() and to look at skb->dst->dev, which is
 187  * changed to the destination device if ip_route_input() succeeds.
 188  *
 189  * Let us first consider the case that ip_route_input() succeeds:
 190  *
 191  * If skb->dst->dev equals the logical bridge device the packet
 192  * came in on, we can consider this bridging. We then call
 193  * skb->dst->output() which will make the packet enter br_nf_local_out()
 194  * not much later. In that function it is assured that the iptables
 195  * FORWARD chain is traversed for the packet.
 196  *
 197  * Otherwise, the packet is considered to be routed and we just
 198  * change the destination MAC address so that the packet will
 199  * later be passed up to the IP stack to be routed.
 200  *
 201  * Let us now consider the case that ip_route_input() fails:
 202  *
 203  * After a "echo '0' > /proc/sys/net/ipv4/ip_forward" ip_route_input()
 204  * will fail, while __ip_route_output_key() will return success. The source
 205  * address for __ip_route_output_key() is set to zero, so __ip_route_output_key
 206  * thinks we're handling a locally generated packet and won't care
 207  * if IP forwarding is allowed. We send a warning message to the users's
 208  * log telling her to put IP forwarding on.
 209  *
 210  * ip_route_input() will also fail if there is no route available.
 211  * In that case we just drop the packet.
 212  *
 213  * --Lennert, 20020411
 214  * --Bart, 20020416 (updated)
 215  * --Bart, 20021007 (updated) */
 216 static int br_nf_pre_routing_finish_bridge(struct sk_buff *skb)
 217 {
 218         if (skb->pkt_type == PACKET_OTHERHOST) {
 219                 skb->pkt_type = PACKET_HOST;
 220                 skb->nf_bridge->mask |= BRNF_PKT_TYPE;
 221         }
 222         skb->nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
 223
 224         skb->dev = bridge_parent(skb->dev);
 225         if (!skb->dev)
 226                 kfree_skb(skb);
 227         else {
 228                 if (skb->protocol == htons(ETH_P_8021Q)) {
 229                         skb_pull(skb, VLAN_HLEN);
 230                         skb->nh.raw += VLAN_HLEN;
 231                 }
 232                 skb->dst->output(skb);
 233         }
 234         return 0;
 235 }
 236
 237 static int br_nf_pre_routing_finish(struct sk_buff *skb)
 238 {
 239         struct net_device *dev = skb->dev;
 240         struct iphdr *iph = skb->nh.iph;
 241         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 242
 243         if (nf_bridge->mask & BRNF_PKT_TYPE) {
 244                 skb->pkt_type = PACKET_OTHERHOST;
 245                 nf_bridge->mask ^= BRNF_PKT_TYPE;
 246         }
 247         nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
 248
 249         if (dnat_took_place(skb)) {
 250                 if (ip_route_input(skb, iph->daddr, iph->saddr, iph->tos, dev)) {
 251                         struct rtable *rt;
 252                         struct flowi fl = {
 253                                 .nl_u = {
 254                                         .ip4_u = {
 255                                                  .daddr = iph->daddr,
 256                                                  .saddr = 0,
 257                                                  .tos = RT_TOS(iph->tos) },
 258                                 },
 259                                 .proto = 0,
 260                         };
 261
 262                         if (!ip_route_output_key(&rt, &fl)) {
 263                                 /* - Bridged-and-DNAT'ed traffic doesn't
 264                                  *   require ip_forwarding.
 265                                  * - Deal with redirected traffic. */
 266                                 if (((struct dst_entry *)rt)->dev == dev ||
 267                                     rt->rt_type == RTN_LOCAL) {
 268                                         skb->dst = (struct dst_entry *)rt;
 269                                         goto bridged_dnat;
 270                                 }
 271                                 __br_dnat_complain();
 272                                 dst_release((struct dst_entry *)rt);
 273                         }
 274                         kfree_skb(skb);
 275                         return 0;
 276                 } else {
 277                         if (skb->dst->dev == dev) {
 278 bridged_dnat:
 279                                 /* Tell br_nf_local_out this is a
 280                                  * bridged frame */
 281                                 nf_bridge->mask |= BRNF_BRIDGED_DNAT;
 282                                 skb->dev = nf_bridge->physindev;
 283                                 if (skb->protocol ==
 284                                     htons(ETH_P_8021Q)) {
 285                                         skb_push(skb, VLAN_HLEN);
 286                                         skb->nh.raw -= VLAN_HLEN;
 287                                 }
 288                                 NF_HOOK_THRESH(PF_BRIDGE, NF_BR_PRE_ROUTING,
 289                                                skb, skb->dev, NULL,
 290                                                br_nf_pre_routing_finish_bridge,
 291                                                1);
 292                                 return 0;
 293                         }
 294                         memcpy(eth_hdr(skb)->h_dest, dev->dev_addr, ETH_ALEN);
 295                         skb->pkt_type = PACKET_HOST;
 296                 }
 297         } else {
 298                 skb->dst = (struct dst_entry *)&__fake_rtable;
 299                 dst_hold(skb->dst);
 300         }
 301
 302         skb->dev = nf_bridge->physindev;
 303         if (skb->protocol == htons(ETH_P_8021Q)) {
 304                 skb_push(skb, VLAN_HLEN);
 305                 skb->nh.raw -= VLAN_HLEN;
 306         }
 307         NF_HOOK_THRESH(PF_BRIDGE, NF_BR_PRE_ROUTING, skb, skb->dev, NULL,
 308                        br_handle_frame_finish, 1);
 309
 310         return 0;
 311 }
 312
 313 /* Some common code for IPv4/IPv6 */
 314 static struct net_device *setup_pre_routing(struct sk_buff *skb)
 315 {
 316         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 317
 318         if (skb->pkt_type == PACKET_OTHERHOST) {
 319                 skb->pkt_type = PACKET_HOST;
 320                 nf_bridge->mask |= BRNF_PKT_TYPE;
 321         }
 322
 323         nf_bridge->mask |= BRNF_NF_BRIDGE_PREROUTING;
 324         nf_bridge->physindev = skb->dev;
 325         skb->dev = bridge_parent(skb->dev);
 326
 327         return skb->dev;
 328 }
 329
 330 /* We only check the length. A bridge shouldn't do any hop-by-hop stuff anyway */
 331 static int check_hbh_len(struct sk_buff *skb)
 332 {
 333         unsigned char *raw = (u8 *) (skb->nh.ipv6h + 1);
 334         u32 pkt_len;
 335         int off = raw - skb->nh.raw;
 336         int len = (raw[1] + 1) << 3;
 337
 338         if ((raw + len) - skb->data > skb_headlen(skb))
 339                 goto bad;
 340
 341         off += 2;
 342         len -= 2;
 343
 344         while (len > 0) {
 345                 int optlen = skb->nh.raw[off + 1] + 2;
 346
 347                 switch (skb->nh.raw[off]) {
 348                 case IPV6_TLV_PAD0:
 349                         optlen = 1;
 350                         break;
 351
 352                 case IPV6_TLV_PADN:
 353                         break;
 354
 355                 case IPV6_TLV_JUMBO:
 356                         if (skb->nh.raw[off + 1] != 4 || (off & 3) != 2)
 357                                 goto bad;
 358                         pkt_len = ntohl(*(u32 *) (skb->nh.raw + off + 2));
 359                         if (pkt_len <= IPV6_MAXPLEN ||
 360                             skb->nh.ipv6h->payload_len)
 361                                 goto bad;
 362                         if (pkt_len > skb->len - sizeof(struct ipv6hdr))
 363                                 goto bad;
 364                         if (pskb_trim_rcsum(skb,
 365                                             pkt_len + sizeof(struct ipv6hdr)))
 366                                 goto bad;
 367                         break;
 368                 default:
 369                         if (optlen > len)
 370                                 goto bad;
 371                         break;
 372                 }
 373                 off += optlen;
 374                 len -= optlen;
 375         }
 376         if (len == 0)
 377                 return 0;
 378 bad:
 379         return -1;
 380
 381 }
 382
 383 /* Replicate the checks that IPv6 does on packet reception and pass the packet
 384  * to ip6tables, which doesn't support NAT, so things are fairly simple. */
 385 static unsigned int br_nf_pre_routing_ipv6(unsigned int hook,
 386                                            struct sk_buff *skb,
 387                                            const struct net_device *in,
 388                                            const struct net_device *out,
 389                                            int (*okfn)(struct sk_buff *))
 390 {
 391         struct ipv6hdr *hdr;
 392         u32 pkt_len;
 393
 394         if (skb->len < sizeof(struct ipv6hdr))
 395                 goto inhdr_error;
 396
 397         if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
 398                 goto inhdr_error;
 399
 400         hdr = skb->nh.ipv6h;
 401
 402         if (hdr->version != 6)
 403                 goto inhdr_error;
 404
 405         pkt_len = ntohs(hdr->payload_len);
 406
 407         if (pkt_len || hdr->nexthdr != NEXTHDR_HOP) {
 408                 if (pkt_len + sizeof(struct ipv6hdr) > skb->len)
 409                         goto inhdr_error;
 410                 if (pkt_len + sizeof(struct ipv6hdr) < skb->len) {
 411                         if (__pskb_trim(skb, pkt_len + sizeof(struct ipv6hdr)))
 412                                 goto inhdr_error;
 413                         if (skb->ip_summed == CHECKSUM_HW)
 414                                 skb->ip_summed = CHECKSUM_NONE;
 415                 }
 416         }
 417         if (hdr->nexthdr == NEXTHDR_HOP && check_hbh_len(skb))
 418                 goto inhdr_error;
 419
 420         nf_bridge_put(skb->nf_bridge);
 421         if (!nf_bridge_alloc(skb))
 422                 return NF_DROP;
 423         if (!setup_pre_routing(skb))
 424                 return NF_DROP;
 425
 426         NF_HOOK(PF_INET6, NF_IP6_PRE_ROUTING, skb, skb->dev, NULL,
 427                 br_nf_pre_routing_finish_ipv6);
 428
 429         return NF_STOLEN;
 430
 431 inhdr_error:
 432         return NF_DROP;
 433 }
 434
 435 /* Direct IPv6 traffic to br_nf_pre_routing_ipv6.
 436  * Replicate the checks that IPv4 does on packet reception.
 437  * Set skb->dev to the bridge device (i.e. parent of the
 438  * receiving device) to make netfilter happy, the REDIRECT
 439  * target in particular.  Save the original destination IP
 440  * address to be able to detect DNAT afterwards. */
 441 static unsigned int br_nf_pre_routing(unsigned int hook, struct sk_buff **pskb,
 442                                       const struct net_device *in,
 443                                       const struct net_device *out,
 444                                       int (*okfn)(struct sk_buff *))
 445 {
 446         struct iphdr *iph;
 447         __u32 len;
 448         struct sk_buff *skb = *pskb;
 449
 450         if (skb->protocol == htons(ETH_P_IPV6) || IS_VLAN_IPV6(skb)) {
 451 #ifdef CONFIG_SYSCTL
 452                 if (!brnf_call_ip6tables)
 453                         return NF_ACCEPT;
 454 #endif
 455                 if ((skb = skb_share_check(*pskb, GFP_ATOMIC)) == NULL)
 456                         goto out;
 457
 458                 if (skb->protocol == htons(ETH_P_8021Q)) {
 459                         skb_pull_rcsum(skb, VLAN_HLEN);
 460                         skb->nh.raw += VLAN_HLEN;
 461                 }
 462                 return br_nf_pre_routing_ipv6(hook, skb, in, out, okfn);
 463         }
 464 #ifdef CONFIG_SYSCTL
 465         if (!brnf_call_iptables)
 466                 return NF_ACCEPT;
 467 #endif
 468
 469         if (skb->protocol != htons(ETH_P_IP) && !IS_VLAN_IP(skb))
 470                 return NF_ACCEPT;
 471
 472         if ((skb = skb_share_check(*pskb, GFP_ATOMIC)) == NULL)
 473                 goto out;
 474
 475         if (skb->protocol == htons(ETH_P_8021Q)) {
 476                 skb_pull_rcsum(skb, VLAN_HLEN);
 477                 skb->nh.raw += VLAN_HLEN;
 478         }
 479
 480         if (!pskb_may_pull(skb, sizeof(struct iphdr)))
 481                 goto inhdr_error;
 482
 483         iph = skb->nh.iph;
 484         if (iph->ihl < 5 || iph->version != 4)
 485                 goto inhdr_error;
 486
 487         if (!pskb_may_pull(skb, 4 * iph->ihl))
 488                 goto inhdr_error;
 489
 490         iph = skb->nh.iph;
 491         if (ip_fast_csum((__u8 *) iph, iph->ihl) != 0)
 492                 goto inhdr_error;
 493
 494         len = ntohs(iph->tot_len);
 495         if (skb->len < len || len < 4 * iph->ihl)
 496                 goto inhdr_error;
 497
 498         if (skb->len > len) {
 499                 __pskb_trim(skb, len);
 500                 if (skb->ip_summed == CHECKSUM_HW)
 501                         skb->ip_summed = CHECKSUM_NONE;
 502         }
 503
 504         nf_bridge_put(skb->nf_bridge);
 505         if (!nf_bridge_alloc(skb))
 506                 return NF_DROP;
 507         if (!setup_pre_routing(skb))
 508                 return NF_DROP;
 509         store_orig_dstaddr(skb);
 510
 511         NF_HOOK(PF_INET, NF_IP_PRE_ROUTING, skb, skb->dev, NULL,
 512                 br_nf_pre_routing_finish);
 513
 514         return NF_STOLEN;
 515
 516 inhdr_error:
 517 //      IP_INC_STATS_BH(IpInHdrErrors);
 518 out:
 519         return NF_DROP;
 520 }
 521
 522
 523 /* PF_BRIDGE/LOCAL_IN ************************************************/
 524 /* The packet is locally destined, which requires a real
 525  * dst_entry, so detach the fake one.  On the way up, the
 526  * packet would pass through PRE_ROUTING again (which already
 527  * took place when the packet entered the bridge), but we
 528  * register an IPv4 PRE_ROUTING 'sabotage' hook that will
 529  * prevent this from happening. */
 530 static unsigned int br_nf_local_in(unsigned int hook, struct sk_buff **pskb,
 531                                    const struct net_device *in,
 532                                    const struct net_device *out,
 533                                    int (*okfn)(struct sk_buff *))
 534 {
 535         struct sk_buff *skb = *pskb;
 536
 537         if (skb->dst == (struct dst_entry *)&__fake_rtable) {
 538                 dst_release(skb->dst);
 539                 skb->dst = NULL;
 540         }
 541
 542         return NF_ACCEPT;
 543 }
 544
 545 /* PF_BRIDGE/FORWARD *************************************************/
 546 static int br_nf_forward_finish(struct sk_buff *skb)
 547 {
 548         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 549         struct net_device *in;
 550
 551         if (skb->protocol != htons(ETH_P_ARP) && !IS_VLAN_ARP(skb)) {
 552                 in = nf_bridge->physindev;
 553                 if (nf_bridge->mask & BRNF_PKT_TYPE) {
 554                         skb->pkt_type = PACKET_OTHERHOST;
 555                         nf_bridge->mask ^= BRNF_PKT_TYPE;
 556                 }
 557         } else {
 558                 in = *((struct net_device **)(skb->cb));
 559         }
 560         if (skb->protocol == htons(ETH_P_8021Q)) {
 561                 skb_push(skb, VLAN_HLEN);
 562                 skb->nh.raw -= VLAN_HLEN;
 563         }
 564         NF_HOOK_THRESH(PF_BRIDGE, NF_BR_FORWARD, skb, in,
 565                        skb->dev, br_forward_finish, 1);
 566         return 0;
 567 }
 568
 569 /* This is the 'purely bridged' case.  For IP, we pass the packet to
 570  * netfilter with indev and outdev set to the bridge device,
 571  * but we are still able to filter on the 'real' indev/outdev
 572  * because of the physdev module. For ARP, indev and outdev are the
 573  * bridge ports. */
 574 static unsigned int br_nf_forward_ip(unsigned int hook, struct sk_buff **pskb,
 575                                      const struct net_device *in,
 576                                      const struct net_device *out,
 577                                      int (*okfn)(struct sk_buff *))
 578 {
 579         struct sk_buff *skb = *pskb;
 580         struct nf_bridge_info *nf_bridge;
 581         struct net_device *parent;
 582         int pf;
 583
 584         if (!skb->nf_bridge)
 585                 return NF_ACCEPT;
 586
 587         parent = bridge_parent(out);
 588         if (!parent)
 589                 return NF_DROP;
 590
 591         if (skb->protocol == htons(ETH_P_IP) || IS_VLAN_IP(skb))
 592                 pf = PF_INET;
 593         else
 594                 pf = PF_INET6;
 595
 596         if (skb->protocol == htons(ETH_P_8021Q)) {
 597                 skb_pull(*pskb, VLAN_HLEN);
 598                 (*pskb)->nh.raw += VLAN_HLEN;
 599         }
 600
 601         nf_bridge = skb->nf_bridge;
 602         if (skb->pkt_type == PACKET_OTHERHOST) {
 603                 skb->pkt_type = PACKET_HOST;
 604                 nf_bridge->mask |= BRNF_PKT_TYPE;
 605         }
 606
 607         /* The physdev module checks on this */
 608         nf_bridge->mask |= BRNF_BRIDGED;
 609         nf_bridge->physoutdev = skb->dev;
 610
 611         NF_HOOK(pf, NF_IP_FORWARD, skb, bridge_parent(in), parent,
 612                 br_nf_forward_finish);
 613
 614         return NF_STOLEN;
 615 }
 616
 617 static unsigned int br_nf_forward_arp(unsigned int hook, struct sk_buff **pskb,
 618                                       const struct net_device *in,
 619                                       const struct net_device *out,
 620                                       int (*okfn)(struct sk_buff *))
 621 {
 622         struct sk_buff *skb = *pskb;
 623         struct net_device **d = (struct net_device **)(skb->cb);
 624
 625 #ifdef CONFIG_SYSCTL
 626         if (!brnf_call_arptables)
 627                 return NF_ACCEPT;
 628 #endif
 629
 630         if (skb->protocol != htons(ETH_P_ARP)) {
 631                 if (!IS_VLAN_ARP(skb))
 632                         return NF_ACCEPT;
 633                 skb_pull(*pskb, VLAN_HLEN);
 634                 (*pskb)->nh.raw += VLAN_HLEN;
 635         }
 636
 637         if (skb->nh.arph->ar_pln != 4) {
 638                 if (IS_VLAN_ARP(skb)) {
 639                         skb_push(*pskb, VLAN_HLEN);
 640                         (*pskb)->nh.raw -= VLAN_HLEN;
 641                 }
 642                 return NF_ACCEPT;
 643         }
 644         *d = (struct net_device *)in;
 645         NF_HOOK(NF_ARP, NF_ARP_FORWARD, skb, (struct net_device *)in,
 646                 (struct net_device *)out, br_nf_forward_finish);
 647
 648         return NF_STOLEN;
 649 }
 650
 651 /* PF_BRIDGE/LOCAL_OUT ***********************************************/
 652 static int br_nf_local_out_finish(struct sk_buff *skb)
 653 {
 654         if (skb->protocol == htons(ETH_P_8021Q)) {
 655                 skb_push(skb, VLAN_HLEN);
 656                 skb->nh.raw -= VLAN_HLEN;
 657         }
 658
 659         NF_HOOK_THRESH(PF_BRIDGE, NF_BR_LOCAL_OUT, skb, NULL, skb->dev,
 660                        br_forward_finish, NF_BR_PRI_FIRST + 1);
 661
 662         return 0;
 663 }
 664
 665 /* This function sees both locally originated IP packets and forwarded
 666  * IP packets (in both cases the destination device is a bridge
 667  * device). It also sees bridged-and-DNAT'ed packets.
 668  * To be able to filter on the physical bridge devices (with the physdev
 669  * module), we steal packets destined to a bridge device away from the
 670  * PF_INET/FORWARD and PF_INET/OUTPUT hook functions, and give them back later,
 671  * when we have determined the real output device. This is done in here.
 672  *
 673  * If (nf_bridge->mask & BRNF_BRIDGED_DNAT) then the packet is bridged
 674  * and we fake the PF_BRIDGE/FORWARD hook. The function br_nf_forward()
 675  * will then fake the PF_INET/FORWARD hook. br_nf_local_out() has priority
 676  * NF_BR_PRI_FIRST, so no relevant PF_BRIDGE/INPUT functions have been nor
 677  * will be executed.
 678  * Otherwise, if nf_bridge->physindev is NULL, the bridge-nf code never touched
 679  * this packet before, and so the packet was locally originated. We fake
 680  * the PF_INET/LOCAL_OUT hook.
 681  * Finally, if nf_bridge->physindev isn't NULL, then the packet was IP routed,
 682  * so we fake the PF_INET/FORWARD hook. ip_sabotage_out() makes sure
 683  * even routed packets that didn't arrive on a bridge interface have their
 684  * nf_bridge->physindev set. */
 685 static unsigned int br_nf_local_out(unsigned int hook, struct sk_buff **pskb,
 686                                     const struct net_device *in,
 687                                     const struct net_device *out,
 688                                     int (*okfn)(struct sk_buff *))
 689 {
 690         struct net_device *realindev, *realoutdev;
 691         struct sk_buff *skb = *pskb;
 692         struct nf_bridge_info *nf_bridge;
 693         int pf;
 694
 695         if (!skb->nf_bridge)
 696                 return NF_ACCEPT;
 697
 698         if (skb->protocol == htons(ETH_P_IP) || IS_VLAN_IP(skb))
 699                 pf = PF_INET;
 700         else
 701                 pf = PF_INET6;
 702
 703 #ifdef CONFIG_NETFILTER_DEBUG
 704         /* Sometimes we get packets with NULL ->dst here (for example,
 705          * running a dhcp client daemon triggers this). This should now
 706          * be fixed, but let's keep the check around. */
 707         if (skb->dst == NULL) {
 708                 printk(KERN_CRIT "br_netfilter: skb->dst == NULL.");
 709                 return NF_ACCEPT;
 710         }
 711 #endif
 712
 713         nf_bridge = skb->nf_bridge;
 714         nf_bridge->physoutdev = skb->dev;
 715         realindev = nf_bridge->physindev;
 716
 717         /* Bridged, take PF_BRIDGE/FORWARD.
 718          * (see big note in front of br_nf_pre_routing_finish) */
 719         if (nf_bridge->mask & BRNF_BRIDGED_DNAT) {
 720                 if (nf_bridge->mask & BRNF_PKT_TYPE) {
 721                         skb->pkt_type = PACKET_OTHERHOST;
 722                         nf_bridge->mask ^= BRNF_PKT_TYPE;
 723                 }
 724                 if (skb->protocol == htons(ETH_P_8021Q)) {
 725                         skb_push(skb, VLAN_HLEN);
 726                         skb->nh.raw -= VLAN_HLEN;
 727                 }
 728
 729                 NF_HOOK(PF_BRIDGE, NF_BR_FORWARD, skb, realindev,
 730                         skb->dev, br_forward_finish);
 731                 goto out;
 732         }
 733         realoutdev = bridge_parent(skb->dev);
 734         if (!realoutdev)
 735                 return NF_DROP;
 736
 737 #if defined(CONFIG_VLAN_8021Q) || defined(CONFIG_VLAN_8021Q_MODULE)
 738         /* iptables should match -o br0.x */
 739         if (nf_bridge->netoutdev)
 740                 realoutdev = nf_bridge->netoutdev;
 741 #endif
 742         if (skb->protocol == htons(ETH_P_8021Q)) {
 743                 skb_pull(skb, VLAN_HLEN);
 744                 (*pskb)->nh.raw += VLAN_HLEN;
 745         }
 746         /* IP forwarded traffic has a physindev, locally
 747          * generated traffic hasn't. */
 748         if (realindev != NULL) {
 749                 if (!(nf_bridge->mask & BRNF_DONT_TAKE_PARENT)) {
 750                         struct net_device *parent = bridge_parent(realindev);
 751                         if (parent)
 752                                 realindev = parent;
 753                 }
 754
 755                 NF_HOOK_THRESH(pf, NF_IP_FORWARD, skb, realindev,
 756                                realoutdev, br_nf_local_out_finish,
 757                                NF_IP_PRI_BRIDGE_SABOTAGE_FORWARD + 1);
 758         } else {
 759                 NF_HOOK_THRESH(pf, NF_IP_LOCAL_OUT, skb, realindev,
 760                                realoutdev, br_nf_local_out_finish,
 761                                NF_IP_PRI_BRIDGE_SABOTAGE_LOCAL_OUT + 1);
 762         }
 763
 764 out:
 765         return NF_STOLEN;
 766 }
 767
 768 static int br_nf_dev_queue_xmit(struct sk_buff *skb)
 769 {
 770         if (skb->protocol == htons(ETH_P_IP) &&
 771             skb->len > skb->dev->mtu &&
 772             !(skb_shinfo(skb)->ufo_size || skb_shinfo(skb)->tso_size))
 773                 return ip_fragment(skb, br_dev_queue_push_xmit);
 774         else
 775                 return br_dev_queue_push_xmit(skb);
 776 }
 777
 778 /* PF_BRIDGE/POST_ROUTING ********************************************/
 779 static unsigned int br_nf_post_routing(unsigned int hook, struct sk_buff **pskb,
 780                                        const struct net_device *in,
 781                                        const struct net_device *out,
 782                                        int (*okfn)(struct sk_buff *))
 783 {
 784         struct sk_buff *skb = *pskb;
 785         struct nf_bridge_info *nf_bridge = (*pskb)->nf_bridge;
 786         struct net_device *realoutdev = bridge_parent(skb->dev);
 787         int pf;
 788
 789 #ifdef CONFIG_NETFILTER_DEBUG
 790         /* Be very paranoid. This probably won't happen anymore, but let's
 791          * keep the check just to be sure... */
 792         if (skb->mac.raw < skb->head || skb->mac.raw + ETH_HLEN > skb->data) {
 793                 printk(KERN_CRIT "br_netfilter: Argh!! br_nf_post_routing: "
 794                        "bad mac.raw pointer.");
 795                 goto print_error;
 796         }
 797 #endif
 798
 799         if (!nf_bridge)
 800                 return NF_ACCEPT;
 801
 802         if (!realoutdev)
 803                 return NF_DROP;
 804
 805         if (skb->protocol == htons(ETH_P_IP) || IS_VLAN_IP(skb))
 806                 pf = PF_INET;
 807         else
 808                 pf = PF_INET6;
 809
 810 #ifdef CONFIG_NETFILTER_DEBUG
 811         if (skb->dst == NULL) {
 812                 printk(KERN_CRIT "br_netfilter: skb->dst == NULL.");
 813                 goto print_error;
 814         }
 815 #endif
 816
 817         /* We assume any code from br_dev_queue_push_xmit onwards doesn't care
 818          * about the value of skb->pkt_type. */
 819         if (skb->pkt_type == PACKET_OTHERHOST) {
 820                 skb->pkt_type = PACKET_HOST;
 821                 nf_bridge->mask |= BRNF_PKT_TYPE;
 822         }
 823
 824         if (skb->protocol == htons(ETH_P_8021Q)) {
 825                 skb_pull(skb, VLAN_HLEN);
 826                 skb->nh.raw += VLAN_HLEN;
 827         }
 828
 829         nf_bridge_save_header(skb);
 830
 831 #if defined(CONFIG_VLAN_8021Q) || defined(CONFIG_VLAN_8021Q_MODULE)
 832         if (nf_bridge->netoutdev)
 833                 realoutdev = nf_bridge->netoutdev;
 834 #endif
 835         NF_HOOK(pf, NF_IP_POST_ROUTING, skb, NULL, realoutdev,
 836                 br_nf_dev_queue_xmit);
 837
 838         return NF_STOLEN;
 839
 840 #ifdef CONFIG_NETFILTER_DEBUG
 841 print_error:
 842         if (skb->dev != NULL) {
 843                 printk("[%s]", skb->dev->name);
 844                 if (realoutdev)
 845                         printk("[%s]", realoutdev->name);
 846         }
 847         printk(" head:%p, raw:%p, data:%p\n", skb->head, skb->mac.raw,
 848                skb->data);
 849         return NF_ACCEPT;
 850 #endif
 851 }
 852
 853 /* IP/SABOTAGE *****************************************************/
 854 /* Don't hand locally destined packets to PF_INET(6)/PRE_ROUTING
 855  * for the second time. */
 856 static unsigned int ip_sabotage_in(unsigned int hook, struct sk_buff **pskb,
 857                                    const struct net_device *in,
 858                                    const struct net_device *out,
 859                                    int (*okfn)(struct sk_buff *))
 860 {
 861         if ((*pskb)->nf_bridge &&
 862             !((*pskb)->nf_bridge->mask & BRNF_NF_BRIDGE_PREROUTING)) {
 863                 return NF_STOP;
 864         }
 865
 866         return NF_ACCEPT;
 867 }
 868
 869 /* Postpone execution of PF_INET(6)/FORWARD, PF_INET(6)/LOCAL_OUT
 870  * and PF_INET(6)/POST_ROUTING until we have done the forwarding
 871  * decision in the bridge code and have determined nf_bridge->physoutdev. */
 872 static unsigned int ip_sabotage_out(unsigned int hook, struct sk_buff **pskb,
 873                                     const struct net_device *in,
 874                                     const struct net_device *out,
 875                                     int (*okfn)(struct sk_buff *))
 876 {
 877         struct sk_buff *skb = *pskb;
 878
 879         if ((out->hard_start_xmit == br_dev_xmit &&
 880              okfn != br_nf_forward_finish &&
 881              okfn != br_nf_local_out_finish && okfn != br_nf_dev_queue_xmit)
 882 #if defined(CONFIG_VLAN_8021Q) || defined(CONFIG_VLAN_8021Q_MODULE)
 883             || ((out->priv_flags & IFF_802_1Q_VLAN) &&
 884                 VLAN_DEV_INFO(out)->real_dev->hard_start_xmit == br_dev_xmit)
 885 #endif
 886             ) {
 887                 struct nf_bridge_info *nf_bridge;
 888
 889                 if (!skb->nf_bridge) {
 890 #ifdef CONFIG_SYSCTL
 891                         /* This code is executed while in the IP(v6) stack,
 892                            the version should be 4 or 6. We can't use
 893                            skb->protocol because that isn't set on
 894                            PF_INET(6)/LOCAL_OUT. */
 895                         struct iphdr *ip = skb->nh.iph;
 896
 897                         if (ip->version == 4 && !brnf_call_iptables)
 898                                 return NF_ACCEPT;
 899                         else if (ip->version == 6 && !brnf_call_ip6tables)
 900                                 return NF_ACCEPT;
 901 #endif
 902                         if (hook == NF_IP_POST_ROUTING)
 903                                 return NF_ACCEPT;
 904                         if (!nf_bridge_alloc(skb))
 905                                 return NF_DROP;
 906                 }
 907
 908                 nf_bridge = skb->nf_bridge;
 909
 910                 /* This frame will arrive on PF_BRIDGE/LOCAL_OUT and we
 911                  * will need the indev then. For a brouter, the real indev
 912                  * can be a bridge port, so we make sure br_nf_local_out()
 913                  * doesn't use the bridge parent of the indev by using
 914                  * the BRNF_DONT_TAKE_PARENT mask. */
 915                 if (hook == NF_IP_FORWARD && nf_bridge->physindev == NULL) {
 916                         nf_bridge->mask |= BRNF_DONT_TAKE_PARENT;
 917                         nf_bridge->physindev = (struct net_device *)in;
 918                 }
 919 #if defined(CONFIG_VLAN_8021Q) || defined(CONFIG_VLAN_8021Q_MODULE)
 920                 /* the iptables outdev is br0.x, not br0 */
 921                 if (out->priv_flags & IFF_802_1Q_VLAN)
 922                         nf_bridge->netoutdev = (struct net_device *)out;
 923 #endif
 924                 return NF_STOP;
 925         }
 926
 927         return NF_ACCEPT;
 928 }
 929
 930 /* For br_nf_local_out we need (prio = NF_BR_PRI_FIRST), to insure that innocent
 931  * PF_BRIDGE/NF_BR_LOCAL_OUT functions don't get bridged traffic as input.
 932  * For br_nf_post_routing, we need (prio = NF_BR_PRI_LAST), because
 933  * ip_refrag() can return NF_STOLEN. */
 934 static struct nf_hook_ops br_nf_ops[] = {
 935         { .hook = br_nf_pre_routing,
 936           .owner = THIS_MODULE,
 937           .pf = PF_BRIDGE,
 938           .hooknum = NF_BR_PRE_ROUTING,
 939           .priority = NF_BR_PRI_BRNF, },
 940         { .hook = br_nf_local_in,
 941           .owner = THIS_MODULE,
 942           .pf = PF_BRIDGE,
 943           .hooknum = NF_BR_LOCAL_IN,
 944           .priority = NF_BR_PRI_BRNF, },
 945         { .hook = br_nf_forward_ip,
 946           .owner = THIS_MODULE,
 947           .pf = PF_BRIDGE,
 948           .hooknum = NF_BR_FORWARD,
 949           .priority = NF_BR_PRI_BRNF - 1, },
 950         { .hook = br_nf_forward_arp,
 951           .owner = THIS_MODULE,
 952           .pf = PF_BRIDGE,
 953           .hooknum = NF_BR_FORWARD,
 954           .priority = NF_BR_PRI_BRNF, },
 955         { .hook = br_nf_local_out,
 956           .owner = THIS_MODULE,
 957           .pf = PF_BRIDGE,
 958           .hooknum = NF_BR_LOCAL_OUT,
 959           .priority = NF_BR_PRI_FIRST, },
 960         { .hook = br_nf_post_routing,
 961           .owner = THIS_MODULE,
 962           .pf = PF_BRIDGE,
 963           .hooknum = NF_BR_POST_ROUTING,
 964           .priority = NF_BR_PRI_LAST, },
 965         { .hook = ip_sabotage_in,
 966           .owner = THIS_MODULE,
 967           .pf = PF_INET,
 968           .hooknum = NF_IP_PRE_ROUTING,
 969           .priority = NF_IP_PRI_FIRST, },
 970         { .hook = ip_sabotage_in,
 971           .owner = THIS_MODULE,
 972           .pf = PF_INET6,
 973           .hooknum = NF_IP6_PRE_ROUTING,
 974           .priority = NF_IP6_PRI_FIRST, },
 975         { .hook = ip_sabotage_out,
 976           .owner = THIS_MODULE,
 977           .pf = PF_INET,
 978           .hooknum = NF_IP_FORWARD,
 979           .priority = NF_IP_PRI_BRIDGE_SABOTAGE_FORWARD, },
 980         { .hook = ip_sabotage_out,
 981           .owner = THIS_MODULE,
 982           .pf = PF_INET6,
 983           .hooknum = NF_IP6_FORWARD,
 984           .priority = NF_IP6_PRI_BRIDGE_SABOTAGE_FORWARD, },
 985         { .hook = ip_sabotage_out,
 986           .owner = THIS_MODULE,
 987           .pf = PF_INET,
 988           .hooknum = NF_IP_LOCAL_OUT,
 989           .priority = NF_IP_PRI_BRIDGE_SABOTAGE_LOCAL_OUT, },
 990         { .hook = ip_sabotage_out,
 991           .owner = THIS_MODULE,
 992           .pf = PF_INET6,
 993           .hooknum = NF_IP6_LOCAL_OUT,
 994           .priority = NF_IP6_PRI_BRIDGE_SABOTAGE_LOCAL_OUT, },
 995         { .hook = ip_sabotage_out,
 996           .owner = THIS_MODULE,
 997           .pf = PF_INET,
 998           .hooknum = NF_IP_POST_ROUTING,
 999           .priority = NF_IP_PRI_FIRST, },
1000         { .hook = ip_sabotage_out,
1001           .owner = THIS_MODULE,
1002           .pf = PF_INET6,
1003           .hooknum = NF_IP6_POST_ROUTING,
1004           .priority = NF_IP6_PRI_FIRST, },
1005 };
1006
1007 #ifdef CONFIG_SYSCTL
1008 static
1009 int brnf_sysctl_call_tables(ctl_table * ctl, int write, struct file *filp,
1010                             void __user * buffer, size_t * lenp, loff_t * ppos)
1011 {
1012         int ret;
1013
1014         ret = proc_dointvec(ctl, write, filp, buffer, lenp, ppos);
1015
1016         if (write && *(int *)(ctl->data))
1017                 *(int *)(ctl->data) = 1;
1018         return ret;
1019 }
1020
1021 static ctl_table brnf_table[] = {
1022         {
1023                 .ctl_name       = NET_BRIDGE_NF_CALL_ARPTABLES,
1024                 .procname       = "bridge-nf-call-arptables",
1025                 .data           = &brnf_call_arptables,
1026                 .maxlen         = sizeof(int),
1027                 .mode           = 0644,
1028                 .proc_handler   = &brnf_sysctl_call_tables,
1029         },
1030         {
1031                 .ctl_name       = NET_BRIDGE_NF_CALL_IPTABLES,
1032                 .procname       = "bridge-nf-call-iptables",
1033                 .data           = &brnf_call_iptables,
1034                 .maxlen         = sizeof(int),
1035                 .mode           = 0644,
1036                 .proc_handler   = &brnf_sysctl_call_tables,
1037         },
1038         {
1039                 .ctl_name       = NET_BRIDGE_NF_CALL_IP6TABLES,
1040                 .procname       = "bridge-nf-call-ip6tables",
1041                 .data           = &brnf_call_ip6tables,
1042                 .maxlen         = sizeof(int),
1043                 .mode           = 0644,
1044                 .proc_handler   = &brnf_sysctl_call_tables,
1045         },
1046         {
1047                 .ctl_name       = NET_BRIDGE_NF_FILTER_VLAN_TAGGED,
1048                 .procname       = "bridge-nf-filter-vlan-tagged",
1049                 .data           = &brnf_filter_vlan_tagged,
1050                 .maxlen         = sizeof(int),
1051                 .mode           = 0644,
1052                 .proc_handler   = &brnf_sysctl_call_tables,
1053         },
1054         { .ctl_name = 0 }
1055 };
1056
1057 static ctl_table brnf_bridge_table[] = {
1058         {
1059                 .ctl_name       = NET_BRIDGE,
1060                 .procname       = "bridge",
1061                 .mode           = 0555,
1062                 .child          = brnf_table,
1063         },
1064         { .ctl_name = 0 }
1065 };
1066
1067 static ctl_table brnf_net_table[] = {
1068         {
1069                 .ctl_name       = CTL_NET,
1070                 .procname       = "net",
1071                 .mode           = 0555,
1072                 .child          = brnf_bridge_table,
1073         },
1074         { .ctl_name = 0 }
1075 };
1076 #endif
1077
1078 int br_netfilter_init(void)
1079 {
1080         int i;
1081
1082         for (i = 0; i < ARRAY_SIZE(br_nf_ops); i++) {
1083                 int ret;
1084
1085                 if ((ret = nf_register_hook(&br_nf_ops[i])) >= 0)
1086                         continue;
1087
1088                 while (i--)
1089                         nf_unregister_hook(&br_nf_ops[i]);
1090
1091                 return ret;
1092         }
1093
1094 #ifdef CONFIG_SYSCTL
1095         brnf_sysctl_header = register_sysctl_table(brnf_net_table, 0);
1096         if (brnf_sysctl_header == NULL) {
1097                 printk(KERN_WARNING
1098                        "br_netfilter: can't register to sysctl.\n");
1099                 for (i = 0; i < ARRAY_SIZE(br_nf_ops); i++)
1100                         nf_unregister_hook(&br_nf_ops[i]);
1101                 return -EFAULT;
1102         }
1103 #endif
1104
1105         printk(KERN_NOTICE "Bridge firewalling registered\n");
1106
1107         return 0;
1108 }
1109
1110 void br_netfilter_fini(void)
1111 {
1112         int i;
1113
1114         for (i = ARRAY_SIZE(br_nf_ops) - 1; i >= 0; i--)
1115                 nf_unregister_hook(&br_nf_ops[i]);
1116 #ifdef CONFIG_SYSCTL
1117         unregister_sysctl_table(brnf_sysctl_header);
1118 #endif
1119 }