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_ether.h>
  30 #include <linux/if_vlan.h>
  31 #include <linux/netfilter_bridge.h>
  32 #include <linux/netfilter_ipv4.h>
  33 #include <linux/netfilter_ipv6.h>
  34 #include <linux/netfilter_arp.h>
  35 #include <linux/in_route.h>
  36 #include <net/ip.h>
  37 #include <net/ipv6.h>
  38 #include <asm/uaccess.h>
  39 #include <asm/checksum.h>
  40 #include "br_private.h"
  41 #ifdef CONFIG_SYSCTL
  42 #include <linux/sysctl.h>
  43 #endif
  44
  45 #define skb_origaddr(skb)        (((struct bridge_skb_cb *) \
  46                                  (skb->nf_bridge->data))->daddr.ipv4)
  47 #define store_orig_dstaddr(skb)  (skb_origaddr(skb) = (skb)->nh.iph->daddr)
  48 #define dnat_took_place(skb)     (skb_origaddr(skb) != (skb)->nh.iph->daddr)
  49
  50 #define has_bridge_parent(device)       ((device)->br_port != NULL)
  51 #define bridge_parent(device)           ((device)->br_port->br->dev)
  52
  53 #ifdef CONFIG_SYSCTL
  54 static struct ctl_table_header *brnf_sysctl_header;
  55 static int brnf_call_iptables = 1;
  56 static int brnf_call_ip6tables = 1;
  57 static int brnf_call_arptables = 1;
  58 static int brnf_filter_vlan_tagged = 1;
  59 #else
  60 #define brnf_filter_vlan_tagged 1
  61 #endif
  62
  63 #define IS_VLAN_IP (skb->protocol == __constant_htons(ETH_P_8021Q) &&    \
  64         hdr->h_vlan_encapsulated_proto == __constant_htons(ETH_P_IP) &&  \
  65         brnf_filter_vlan_tagged)
  66 #define IS_VLAN_IPV6 (skb->protocol == __constant_htons(ETH_P_8021Q) &&    \
  67         hdr->h_vlan_encapsulated_proto == __constant_htons(ETH_P_IPV6) &&  \
  68         brnf_filter_vlan_tagged)
  69 #define IS_VLAN_ARP (skb->protocol == __constant_htons(ETH_P_8021Q) &&   \
  70         hdr->h_vlan_encapsulated_proto == __constant_htons(ETH_P_ARP) && \
  71         brnf_filter_vlan_tagged)
  72
  73 /* We need these fake structures to make netfilter happy --
  74  * lots of places assume that skb->dst != NULL, which isn't
  75  * all that unreasonable.
  76  *
  77  * Currently, we fill in the PMTU entry because netfilter
  78  * refragmentation needs it, and the rt_flags entry because
  79  * ipt_REJECT needs it.  Future netfilter modules might
  80  * require us to fill additional fields. */
  81 static struct net_device __fake_net_device = {
  82         .hard_header_len        = ETH_HLEN
  83 };
  84
  85 static struct rtable __fake_rtable = {
  86         .u = {
  87                 .dst = {
  88                         .__refcnt               = ATOMIC_INIT(1),
  89                         .dev                    = &__fake_net_device,
  90                         .path                   = &__fake_rtable.u.dst,
  91                         .metrics                = {[RTAX_MTU - 1] = 1500},
  92                 }
  93         },
  94         .rt_flags       = 0,
  95 };
  96
  97
  98 /* PF_BRIDGE/PRE_ROUTING *********************************************/
  99 /* Undo the changes made for ip6tables PREROUTING and continue the
 100  * bridge PRE_ROUTING hook. */
 101 static int br_nf_pre_routing_finish_ipv6(struct sk_buff *skb)
 102 {
 103         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 104
 105         if (nf_bridge->mask & BRNF_PKT_TYPE) {
 106                 skb->pkt_type = PACKET_OTHERHOST;
 107                 nf_bridge->mask ^= BRNF_PKT_TYPE;
 108         }
 109         nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
 110
 111         skb->dst = (struct dst_entry *)&__fake_rtable;
 112         dst_hold(skb->dst);
 113
 114         skb->dev = nf_bridge->physindev;
 115         if (skb->protocol == __constant_htons(ETH_P_8021Q)) {
 116                 skb_push(skb, VLAN_HLEN);
 117                 skb->nh.raw -= VLAN_HLEN;
 118         }
 119         NF_HOOK_THRESH(PF_BRIDGE, NF_BR_PRE_ROUTING, skb, skb->dev, NULL,
 120                        br_handle_frame_finish, 1);
 121
 122         return 0;
 123 }
 124
 125 static void __br_dnat_complain(void)
 126 {
 127         static unsigned long last_complaint;
 128
 129         if (jiffies - last_complaint >= 5 * HZ) {
 130                 printk(KERN_WARNING "Performing cross-bridge DNAT requires IP "
 131                         "forwarding to be enabled\n");
 132                 last_complaint = jiffies;
 133         }
 134 }
 135
 136 /* This requires some explaining. If DNAT has taken place,
 137  * we will need to fix up the destination Ethernet address,
 138  * and this is a tricky process.
 139  *
 140  * There are two cases to consider:
 141  * 1. The packet was DNAT'ed to a device in the same bridge
 142  *    port group as it was received on. We can still bridge
 143  *    the packet.
 144  * 2. The packet was DNAT'ed to a different device, either
 145  *    a non-bridged device or another bridge port group.
 146  *    The packet will need to be routed.
 147  *
 148  * The correct way of distinguishing between these two cases is to
 149  * call ip_route_input() and to look at skb->dst->dev, which is
 150  * changed to the destination device if ip_route_input() succeeds.
 151  *
 152  * Let us first consider the case that ip_route_input() succeeds:
 153  *
 154  * If skb->dst->dev equals the logical bridge device the packet
 155  * came in on, we can consider this bridging. We then call
 156  * skb->dst->output() which will make the packet enter br_nf_local_out()
 157  * not much later. In that function it is assured that the iptables
 158  * FORWARD chain is traversed for the packet.
 159  *
 160  * Otherwise, the packet is considered to be routed and we just
 161  * change the destination MAC address so that the packet will
 162  * later be passed up to the IP stack to be routed.
 163  *
 164  * Let us now consider the case that ip_route_input() fails:
 165  *
 166  * After a "echo '0' > /proc/sys/net/ipv4/ip_forward" ip_route_input()
 167  * will fail, while __ip_route_output_key() will return success. The source
 168  * address for __ip_route_output_key() is set to zero, so __ip_route_output_key
 169  * thinks we're handling a locally generated packet and won't care
 170  * if IP forwarding is allowed. We send a warning message to the users's
 171  * log telling her to put IP forwarding on.
 172  *
 173  * ip_route_input() will also fail if there is no route available.
 174  * In that case we just drop the packet.
 175  *
 176  * --Lennert, 20020411
 177  * --Bart, 20020416 (updated)
 178  * --Bart, 20021007 (updated) */
 179 static int br_nf_pre_routing_finish_bridge(struct sk_buff *skb)
 180 {
 181         if (skb->pkt_type == PACKET_OTHERHOST) {
 182                 skb->pkt_type = PACKET_HOST;
 183                 skb->nf_bridge->mask |= BRNF_PKT_TYPE;
 184         }
 185         skb->nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
 186
 187         skb->dev = bridge_parent(skb->dev);
 188         if (skb->protocol == __constant_htons(ETH_P_8021Q)) {
 189                 skb_pull(skb, VLAN_HLEN);
 190                 skb->nh.raw += VLAN_HLEN;
 191         }
 192         skb->dst->output(skb);
 193         return 0;
 194 }
 195
 196 static int br_nf_pre_routing_finish(struct sk_buff *skb)
 197 {
 198         struct net_device *dev = skb->dev;
 199         struct iphdr *iph = skb->nh.iph;
 200         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 201
 202         if (nf_bridge->mask & BRNF_PKT_TYPE) {
 203                 skb->pkt_type = PACKET_OTHERHOST;
 204                 nf_bridge->mask ^= BRNF_PKT_TYPE;
 205         }
 206         nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
 207
 208         if (dnat_took_place(skb)) {
 209                 if (ip_route_input(skb, iph->daddr, iph->saddr, iph->tos,
 210                     dev)) {
 211                         struct rtable *rt;
 212                         struct flowi fl = { .nl_u =
 213                         { .ip4_u = { .daddr = iph->daddr, .saddr = 0 ,
 214                                      .tos = RT_TOS(iph->tos)} }, .proto = 0};
 215
 216                         if (!ip_route_output_key(&rt, &fl)) {
 217                                 /* - Bridged-and-DNAT'ed traffic doesn't
 218                                  *   require ip_forwarding.
 219                                  * - Deal with redirected traffic. */
 220                                 if (((struct dst_entry *)rt)->dev == dev ||
 221                                     rt->rt_type == RTN_LOCAL) {
 222                                         skb->dst = (struct dst_entry *)rt;
 223                                         goto bridged_dnat;
 224                                 }
 225                                 __br_dnat_complain();
 226                                 dst_release((struct dst_entry *)rt);
 227                         }
 228                         kfree_skb(skb);
 229                         return 0;
 230                 } else {
 231                         if (skb->dst->dev == dev) {
 232 bridged_dnat:
 233                                 /* Tell br_nf_local_out this is a
 234                                  * bridged frame */
 235                                 nf_bridge->mask |= BRNF_BRIDGED_DNAT;
 236                                 skb->dev = nf_bridge->physindev;
 237                                 if (skb->protocol ==
 238                                     __constant_htons(ETH_P_8021Q)) {
 239                                         skb_push(skb, VLAN_HLEN);
 240                                         skb->nh.raw -= VLAN_HLEN;
 241                                 }
 242                                 NF_HOOK_THRESH(PF_BRIDGE, NF_BR_PRE_ROUTING,
 243                                                skb, skb->dev, NULL,
 244                                                br_nf_pre_routing_finish_bridge,
 245                                                1);
 246                                 return 0;
 247                         }
 248                         memcpy(eth_hdr(skb)->h_dest, dev->dev_addr,
 249                                ETH_ALEN);
 250                         skb->pkt_type = PACKET_HOST;
 251                 }
 252         } else {
 253                 skb->dst = (struct dst_entry *)&__fake_rtable;
 254                 dst_hold(skb->dst);
 255         }
 256
 257         skb->dev = nf_bridge->physindev;
 258         if (skb->protocol == __constant_htons(ETH_P_8021Q)) {
 259                 skb_push(skb, VLAN_HLEN);
 260                 skb->nh.raw -= VLAN_HLEN;
 261         }
 262         NF_HOOK_THRESH(PF_BRIDGE, NF_BR_PRE_ROUTING, skb, skb->dev, NULL,
 263                        br_handle_frame_finish, 1);
 264
 265         return 0;
 266 }
 267
 268 /* Some common code for IPv4/IPv6 */
 269 static void setup_pre_routing(struct sk_buff *skb)
 270 {
 271         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 272
 273         if (skb->pkt_type == PACKET_OTHERHOST) {
 274                 skb->pkt_type = PACKET_HOST;
 275                 nf_bridge->mask |= BRNF_PKT_TYPE;
 276         }
 277
 278         nf_bridge->mask |= BRNF_NF_BRIDGE_PREROUTING;
 279         nf_bridge->physindev = skb->dev;
 280         skb->dev = bridge_parent(skb->dev);
 281 }
 282
 283 /* We only check the length. A bridge shouldn't do any hop-by-hop stuff anyway */
 284 static int check_hbh_len(struct sk_buff *skb)
 285 {
 286         unsigned char *raw = (u8*)(skb->nh.ipv6h+1);
 287         u32 pkt_len;
 288         int off = raw - skb->nh.raw;
 289         int len = (raw[1]+1)<<3;
 290
 291         if ((raw + len) - skb->data > skb_headlen(skb))
 292                 goto bad;
 293
 294         off += 2;
 295         len -= 2;
 296
 297         while (len > 0) {
 298                 int optlen = skb->nh.raw[off+1]+2;
 299
 300                 switch (skb->nh.raw[off]) {
 301                 case IPV6_TLV_PAD0:
 302                         optlen = 1;
 303                         break;
 304
 305                 case IPV6_TLV_PADN:
 306                         break;
 307
 308                 case IPV6_TLV_JUMBO:
 309                         if (skb->nh.raw[off+1] != 4 || (off&3) != 2)
 310                                 goto bad;
 311                         pkt_len = ntohl(*(u32*)(skb->nh.raw+off+2));
 312                         if (pkt_len <= IPV6_MAXPLEN ||
 313                             skb->nh.ipv6h->payload_len)
 314                                 goto bad;
 315                         if (pkt_len > skb->len - sizeof(struct ipv6hdr))
 316                                 goto bad;
 317                         if (pskb_trim_rcsum(skb,
 318                             pkt_len+sizeof(struct ipv6hdr)))
 319                                 goto bad;
 320                         break;
 321                 default:
 322                         if (optlen > len)
 323                                 goto bad;
 324                         break;
 325                 }
 326                 off += optlen;
 327                 len -= optlen;
 328         }
 329         if (len == 0)
 330                 return 0;
 331 bad:
 332         return -1;
 333
 334 }
 335
 336 /* Replicate the checks that IPv6 does on packet reception and pass the packet
 337  * to ip6tables, which doesn't support NAT, so things are fairly simple. */
 338 static unsigned int br_nf_pre_routing_ipv6(unsigned int hook,
 339    struct sk_buff *skb, const struct net_device *in,
 340    const struct net_device *out, int (*okfn)(struct sk_buff *))
 341 {
 342         struct ipv6hdr *hdr;
 343         u32 pkt_len;
 344         struct nf_bridge_info *nf_bridge;
 345
 346         if (skb->len < sizeof(struct ipv6hdr))
 347                 goto inhdr_error;
 348
 349         if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
 350                 goto inhdr_error;
 351
 352         hdr = skb->nh.ipv6h;
 353
 354         if (hdr->version != 6)
 355                 goto inhdr_error;
 356
 357         pkt_len = ntohs(hdr->payload_len);
 358
 359         if (pkt_len || hdr->nexthdr != NEXTHDR_HOP) {
 360                 if (pkt_len + sizeof(struct ipv6hdr) > skb->len)
 361                         goto inhdr_error;
 362                 if (pkt_len + sizeof(struct ipv6hdr) < skb->len) {
 363                         if (__pskb_trim(skb, pkt_len + sizeof(struct ipv6hdr)))
 364                                 goto inhdr_error;
 365                         if (skb->ip_summed == CHECKSUM_HW)
 366                                 skb->ip_summed = CHECKSUM_NONE;
 367                 }
 368         }
 369         if (hdr->nexthdr == NEXTHDR_HOP && check_hbh_len(skb))
 370                         goto inhdr_error;
 371
 372         nf_bridge_put(skb->nf_bridge);
 373         if ((nf_bridge = nf_bridge_alloc(skb)) == NULL)
 374                 return NF_DROP;
 375         setup_pre_routing(skb);
 376
 377         NF_HOOK(PF_INET6, NF_IP6_PRE_ROUTING, skb, skb->dev, NULL,
 378                 br_nf_pre_routing_finish_ipv6);
 379
 380         return NF_STOLEN;
 381
 382 inhdr_error:
 383         return NF_DROP;
 384 }
 385
 386 /* Direct IPv6 traffic to br_nf_pre_routing_ipv6.
 387  * Replicate the checks that IPv4 does on packet reception.
 388  * Set skb->dev to the bridge device (i.e. parent of the
 389  * receiving device) to make netfilter happy, the REDIRECT
 390  * target in particular.  Save the original destination IP
 391  * address to be able to detect DNAT afterwards. */
 392 static unsigned int br_nf_pre_routing(unsigned int hook, struct sk_buff **pskb,
 393    const struct net_device *in, const struct net_device *out,
 394    int (*okfn)(struct sk_buff *))
 395 {
 396         struct iphdr *iph;
 397         __u32 len;
 398         struct sk_buff *skb = *pskb;
 399         struct nf_bridge_info *nf_bridge;
 400         struct vlan_ethhdr *hdr = vlan_eth_hdr(*pskb);
 401
 402         if (skb->protocol == __constant_htons(ETH_P_IPV6) || IS_VLAN_IPV6) {
 403 #ifdef CONFIG_SYSCTL
 404                 if (!brnf_call_ip6tables)
 405                         return NF_ACCEPT;
 406 #endif
 407                 if ((skb = skb_share_check(*pskb, GFP_ATOMIC)) == NULL)
 408                         goto out;
 409
 410                 if (skb->protocol == __constant_htons(ETH_P_8021Q)) {
 411                         skb_pull(skb, VLAN_HLEN);
 412                         (skb)->nh.raw += VLAN_HLEN;
 413                 }
 414                 return br_nf_pre_routing_ipv6(hook, skb, in, out, okfn);
 415         }
 416 #ifdef CONFIG_SYSCTL
 417         if (!brnf_call_iptables)
 418                 return NF_ACCEPT;
 419 #endif
 420
 421         if (skb->protocol != __constant_htons(ETH_P_IP) && !IS_VLAN_IP)
 422                 return NF_ACCEPT;
 423
 424         if ((skb = skb_share_check(*pskb, GFP_ATOMIC)) == NULL)
 425                 goto out;
 426
 427         if (skb->protocol == __constant_htons(ETH_P_8021Q)) {
 428                 skb_pull(skb, VLAN_HLEN);
 429                 (skb)->nh.raw += VLAN_HLEN;
 430         }
 431
 432         if (!pskb_may_pull(skb, sizeof(struct iphdr)))
 433                 goto inhdr_error;
 434
 435         iph = skb->nh.iph;
 436         if (iph->ihl < 5 || iph->version != 4)
 437                 goto inhdr_error;
 438
 439         if (!pskb_may_pull(skb, 4*iph->ihl))
 440                 goto inhdr_error;
 441
 442         iph = skb->nh.iph;
 443         if (ip_fast_csum((__u8 *)iph, iph->ihl) != 0)
 444                 goto inhdr_error;
 445
 446         len = ntohs(iph->tot_len);
 447         if (skb->len < len || len < 4*iph->ihl)
 448                 goto inhdr_error;
 449
 450         if (skb->len > len) {
 451                 __pskb_trim(skb, len);
 452                 if (skb->ip_summed == CHECKSUM_HW)
 453                         skb->ip_summed = CHECKSUM_NONE;
 454         }
 455
 456         nf_bridge_put(skb->nf_bridge);
 457         if ((nf_bridge = nf_bridge_alloc(skb)) == NULL)
 458                 return NF_DROP;
 459         setup_pre_routing(skb);
 460         store_orig_dstaddr(skb);
 461
 462         NF_HOOK(PF_INET, NF_IP_PRE_ROUTING, skb, skb->dev, NULL,
 463                 br_nf_pre_routing_finish);
 464
 465         return NF_STOLEN;
 466
 467 inhdr_error:
 468 //      IP_INC_STATS_BH(IpInHdrErrors);
 469 out:
 470         return NF_DROP;
 471 }
 472
 473
 474 /* PF_BRIDGE/LOCAL_IN ************************************************/
 475 /* The packet is locally destined, which requires a real
 476  * dst_entry, so detach the fake one.  On the way up, the
 477  * packet would pass through PRE_ROUTING again (which already
 478  * took place when the packet entered the bridge), but we
 479  * register an IPv4 PRE_ROUTING 'sabotage' hook that will
 480  * prevent this from happening. */
 481 static unsigned int br_nf_local_in(unsigned int hook, struct sk_buff **pskb,
 482    const struct net_device *in, const struct net_device *out,
 483    int (*okfn)(struct sk_buff *))
 484 {
 485         struct sk_buff *skb = *pskb;
 486
 487         if (skb->dst == (struct dst_entry *)&__fake_rtable) {
 488                 dst_release(skb->dst);
 489                 skb->dst = NULL;
 490         }
 491
 492         return NF_ACCEPT;
 493 }
 494
 495
 496 /* PF_BRIDGE/FORWARD *************************************************/
 497 static int br_nf_forward_finish(struct sk_buff *skb)
 498 {
 499         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 500         struct net_device *in;
 501         struct vlan_ethhdr *hdr = vlan_eth_hdr(skb);
 502
 503         if (skb->protocol != __constant_htons(ETH_P_ARP) && !IS_VLAN_ARP) {
 504                 in = nf_bridge->physindev;
 505                 if (nf_bridge->mask & BRNF_PKT_TYPE) {
 506                         skb->pkt_type = PACKET_OTHERHOST;
 507                         nf_bridge->mask ^= BRNF_PKT_TYPE;
 508                 }
 509         } else {
 510                 in = *((struct net_device **)(skb->cb));
 511         }
 512         if (skb->protocol == __constant_htons(ETH_P_8021Q)) {
 513                 skb_push(skb, VLAN_HLEN);
 514                 skb->nh.raw -= VLAN_HLEN;
 515         }
 516         NF_HOOK_THRESH(PF_BRIDGE, NF_BR_FORWARD, skb, in,
 517                         skb->dev, br_forward_finish, 1);
 518         return 0;
 519 }
 520
 521 /* This is the 'purely bridged' case.  For IP, we pass the packet to
 522  * netfilter with indev and outdev set to the bridge device,
 523  * but we are still able to filter on the 'real' indev/outdev
 524  * because of the physdev module. For ARP, indev and outdev are the
 525  * bridge ports. */
 526 static unsigned int br_nf_forward_ip(unsigned int hook, struct sk_buff **pskb,
 527    const struct net_device *in, const struct net_device *out,
 528    int (*okfn)(struct sk_buff *))
 529 {
 530         struct sk_buff *skb = *pskb;
 531         struct nf_bridge_info *nf_bridge;
 532         struct vlan_ethhdr *hdr = vlan_eth_hdr(skb);
 533         int pf;
 534
 535         if (!skb->nf_bridge)
 536                 return NF_ACCEPT;
 537
 538         if (skb->protocol == __constant_htons(ETH_P_IP) || IS_VLAN_IP)
 539                 pf = PF_INET;
 540         else
 541                 pf = PF_INET6;
 542
 543         if (skb->protocol == __constant_htons(ETH_P_8021Q)) {
 544                 skb_pull(*pskb, VLAN_HLEN);
 545                 (*pskb)->nh.raw += VLAN_HLEN;
 546         }
 547
 548         nf_bridge = skb->nf_bridge;
 549         if (skb->pkt_type == PACKET_OTHERHOST) {
 550                 skb->pkt_type = PACKET_HOST;
 551                 nf_bridge->mask |= BRNF_PKT_TYPE;
 552         }
 553
 554         /* The physdev module checks on this */
 555         nf_bridge->mask |= BRNF_BRIDGED;
 556         nf_bridge->physoutdev = skb->dev;
 557
 558         NF_HOOK(pf, NF_IP_FORWARD, skb, bridge_parent(in),
 559                 bridge_parent(out), br_nf_forward_finish);
 560
 561         return NF_STOLEN;
 562 }
 563
 564 static unsigned int br_nf_forward_arp(unsigned int hook, struct sk_buff **pskb,
 565    const struct net_device *in, const struct net_device *out,
 566    int (*okfn)(struct sk_buff *))
 567 {
 568         struct sk_buff *skb = *pskb;
 569         struct vlan_ethhdr *hdr = vlan_eth_hdr(skb);
 570         struct net_device **d = (struct net_device **)(skb->cb);
 571
 572 #ifdef CONFIG_SYSCTL
 573         if (!brnf_call_arptables)
 574                 return NF_ACCEPT;
 575 #endif
 576
 577         if (skb->protocol != __constant_htons(ETH_P_ARP)) {
 578                 if (!IS_VLAN_ARP)
 579                         return NF_ACCEPT;
 580                 skb_pull(*pskb, VLAN_HLEN);
 581                 (*pskb)->nh.raw += VLAN_HLEN;
 582         }
 583
 584         if (skb->nh.arph->ar_pln != 4) {
 585                 if (IS_VLAN_ARP) {
 586                         skb_push(*pskb, VLAN_HLEN);
 587                         (*pskb)->nh.raw -= VLAN_HLEN;
 588                 }
 589                 return NF_ACCEPT;
 590         }
 591         *d = (struct net_device *)in;
 592         NF_HOOK(NF_ARP, NF_ARP_FORWARD, skb, (struct net_device *)in,
 593                 (struct net_device *)out, br_nf_forward_finish);
 594
 595         return NF_STOLEN;
 596 }
 597
 598
 599 /* PF_BRIDGE/LOCAL_OUT ***********************************************/
 600 static int br_nf_local_out_finish(struct sk_buff *skb)
 601 {
 602         if (skb->protocol == __constant_htons(ETH_P_8021Q)) {
 603                 skb_push(skb, VLAN_HLEN);
 604                 skb->nh.raw -= VLAN_HLEN;
 605         }
 606
 607         NF_HOOK_THRESH(PF_BRIDGE, NF_BR_LOCAL_OUT, skb, NULL, skb->dev,
 608                         br_forward_finish, NF_BR_PRI_FIRST + 1);
 609
 610         return 0;
 611 }
 612
 613 /* This function sees both locally originated IP packets and forwarded
 614  * IP packets (in both cases the destination device is a bridge
 615  * device). It also sees bridged-and-DNAT'ed packets.
 616  * To be able to filter on the physical bridge devices (with the physdev
 617  * module), we steal packets destined to a bridge device away from the
 618  * PF_INET/FORWARD and PF_INET/OUTPUT hook functions, and give them back later,
 619  * when we have determined the real output device. This is done in here.
 620  *
 621  * If (nf_bridge->mask & BRNF_BRIDGED_DNAT) then the packet is bridged
 622  * and we fake the PF_BRIDGE/FORWARD hook. The function br_nf_forward()
 623  * will then fake the PF_INET/FORWARD hook. br_nf_local_out() has priority
 624  * NF_BR_PRI_FIRST, so no relevant PF_BRIDGE/INPUT functions have been nor
 625  * will be executed.
 626  * Otherwise, if nf_bridge->physindev is NULL, the bridge-nf code never touched
 627  * this packet before, and so the packet was locally originated. We fake
 628  * the PF_INET/LOCAL_OUT hook.
 629  * Finally, if nf_bridge->physindev isn't NULL, then the packet was IP routed,
 630  * so we fake the PF_INET/FORWARD hook. ip_sabotage_out() makes sure
 631  * even routed packets that didn't arrive on a bridge interface have their
 632  * nf_bridge->physindev set. */
 633 static unsigned int br_nf_local_out(unsigned int hook, struct sk_buff **pskb,
 634    const struct net_device *in, const struct net_device *out,
 635    int (*okfn)(struct sk_buff *))
 636 {
 637         struct net_device *realindev, *realoutdev;
 638         struct sk_buff *skb = *pskb;
 639         struct nf_bridge_info *nf_bridge;
 640         struct vlan_ethhdr *hdr = vlan_eth_hdr(skb);
 641         int pf;
 642
 643         if (!skb->nf_bridge)
 644                 return NF_ACCEPT;
 645
 646         if (skb->protocol == __constant_htons(ETH_P_IP) || IS_VLAN_IP)
 647                 pf = PF_INET;
 648         else
 649                 pf = PF_INET6;
 650
 651 #ifdef CONFIG_NETFILTER_DEBUG
 652         /* Sometimes we get packets with NULL ->dst here (for example,
 653          * running a dhcp client daemon triggers this). This should now
 654          * be fixed, but let's keep the check around. */
 655         if (skb->dst == NULL) {
 656                 printk(KERN_CRIT "br_netfilter: skb->dst == NULL.");
 657                 return NF_ACCEPT;
 658         }
 659 #endif
 660
 661         nf_bridge = skb->nf_bridge;
 662         nf_bridge->physoutdev = skb->dev;
 663         realindev = nf_bridge->physindev;
 664
 665         /* Bridged, take PF_BRIDGE/FORWARD.
 666          * (see big note in front of br_nf_pre_routing_finish) */
 667         if (nf_bridge->mask & BRNF_BRIDGED_DNAT) {
 668                 if (nf_bridge->mask & BRNF_PKT_TYPE) {
 669                         skb->pkt_type = PACKET_OTHERHOST;
 670                         nf_bridge->mask ^= BRNF_PKT_TYPE;
 671                 }
 672                 if (skb->protocol == __constant_htons(ETH_P_8021Q)) {
 673                         skb_push(skb, VLAN_HLEN);
 674                         skb->nh.raw -= VLAN_HLEN;
 675                 }
 676
 677                 NF_HOOK(PF_BRIDGE, NF_BR_FORWARD, skb, realindev,
 678                         skb->dev, br_forward_finish);
 679                 goto out;
 680         }
 681         realoutdev = bridge_parent(skb->dev);
 682
 683 #if defined(CONFIG_VLAN_8021Q) || defined(CONFIG_VLAN_8021Q_MODULE)
 684         /* iptables should match -o br0.x */
 685         if (nf_bridge->netoutdev)
 686                 realoutdev = nf_bridge->netoutdev;
 687 #endif
 688         if (skb->protocol == __constant_htons(ETH_P_8021Q)) {
 689                 skb_pull(skb, VLAN_HLEN);
 690                 (*pskb)->nh.raw += VLAN_HLEN;
 691         }
 692         /* IP forwarded traffic has a physindev, locally
 693          * generated traffic hasn't. */
 694         if (realindev != NULL) {
 695                 if (!(nf_bridge->mask & BRNF_DONT_TAKE_PARENT) &&
 696                     has_bridge_parent(realindev))
 697                         realindev = bridge_parent(realindev);
 698
 699                 NF_HOOK_THRESH(pf, NF_IP_FORWARD, skb, realindev,
 700                                realoutdev, br_nf_local_out_finish,
 701                                NF_IP_PRI_BRIDGE_SABOTAGE_FORWARD + 1);
 702         } else {
 703                 NF_HOOK_THRESH(pf, NF_IP_LOCAL_OUT, skb, realindev,
 704                                realoutdev, br_nf_local_out_finish,
 705                                NF_IP_PRI_BRIDGE_SABOTAGE_LOCAL_OUT + 1);
 706         }
 707
 708 out:
 709         return NF_STOLEN;
 710 }
 711
 712
 713 /* PF_BRIDGE/POST_ROUTING ********************************************/
 714 static unsigned int br_nf_post_routing(unsigned int hook, struct sk_buff **pskb,
 715    const struct net_device *in, const struct net_device *out,
 716    int (*okfn)(struct sk_buff *))
 717 {
 718         struct sk_buff *skb = *pskb;
 719         struct nf_bridge_info *nf_bridge = (*pskb)->nf_bridge;
 720         struct vlan_ethhdr *hdr = vlan_eth_hdr(skb);
 721         struct net_device *realoutdev = bridge_parent(skb->dev);
 722         int pf;
 723
 724 #ifdef CONFIG_NETFILTER_DEBUG
 725         /* Be very paranoid. This probably won't happen anymore, but let's
 726          * keep the check just to be sure... */
 727         if (skb->mac.raw < skb->head || skb->mac.raw + ETH_HLEN > skb->data) {
 728                 printk(KERN_CRIT "br_netfilter: Argh!! br_nf_post_routing: "
 729                                  "bad mac.raw pointer.");
 730                 goto print_error;
 731         }
 732 #endif
 733
 734         if (!nf_bridge)
 735                 return NF_ACCEPT;
 736
 737         if (skb->protocol == __constant_htons(ETH_P_IP) || IS_VLAN_IP)
 738                 pf = PF_INET;
 739         else
 740                 pf = PF_INET6;
 741
 742 #ifdef CONFIG_NETFILTER_DEBUG
 743         if (skb->dst == NULL) {
 744                 printk(KERN_CRIT "br_netfilter: skb->dst == NULL.");
 745                 goto print_error;
 746         }
 747 #endif
 748
 749         /* We assume any code from br_dev_queue_push_xmit onwards doesn't care
 750          * about the value of skb->pkt_type. */
 751         if (skb->pkt_type == PACKET_OTHERHOST) {
 752                 skb->pkt_type = PACKET_HOST;
 753                 nf_bridge->mask |= BRNF_PKT_TYPE;
 754         }
 755
 756         if (skb->protocol == __constant_htons(ETH_P_8021Q)) {
 757                 skb_pull(skb, VLAN_HLEN);
 758                 skb->nh.raw += VLAN_HLEN;
 759         }
 760
 761         nf_bridge_save_header(skb);
 762
 763 #if defined(CONFIG_VLAN_8021Q) || defined(CONFIG_VLAN_8021Q_MODULE)
 764         if (nf_bridge->netoutdev)
 765                 realoutdev = nf_bridge->netoutdev;
 766 #endif
 767         NF_HOOK(pf, NF_IP_POST_ROUTING, skb, NULL, realoutdev,
 768                 br_dev_queue_push_xmit);
 769
 770         return NF_STOLEN;
 771
 772 #ifdef CONFIG_NETFILTER_DEBUG
 773 print_error:
 774         if (skb->dev != NULL) {
 775                 printk("[%s]", skb->dev->name);
 776                 if (has_bridge_parent(skb->dev))
 777                         printk("[%s]", bridge_parent(skb->dev)->name);
 778         }
 779         printk(" head:%p, raw:%p, data:%p\n", skb->head, skb->mac.raw,
 780                                               skb->data);
 781         return NF_ACCEPT;
 782 #endif
 783 }
 784
 785
 786 /* IP/SABOTAGE *****************************************************/
 787 /* Don't hand locally destined packets to PF_INET(6)/PRE_ROUTING
 788  * for the second time. */
 789 static unsigned int ip_sabotage_in(unsigned int hook, struct sk_buff **pskb,
 790    const struct net_device *in, const struct net_device *out,
 791    int (*okfn)(struct sk_buff *))
 792 {
 793         if ((*pskb)->nf_bridge &&
 794             !((*pskb)->nf_bridge->mask & BRNF_NF_BRIDGE_PREROUTING)) {
 795                 return NF_STOP;
 796         }
 797
 798         return NF_ACCEPT;
 799 }
 800
 801 /* Postpone execution of PF_INET(6)/FORWARD, PF_INET(6)/LOCAL_OUT
 802  * and PF_INET(6)/POST_ROUTING until we have done the forwarding
 803  * decision in the bridge code and have determined nf_bridge->physoutdev. */
 804 static unsigned int ip_sabotage_out(unsigned int hook, struct sk_buff **pskb,
 805    const struct net_device *in, const struct net_device *out,
 806    int (*okfn)(struct sk_buff *))
 807 {
 808         struct sk_buff *skb = *pskb;
 809
 810         if ((out->hard_start_xmit == br_dev_xmit &&
 811             okfn != br_nf_forward_finish &&
 812             okfn != br_nf_local_out_finish &&
 813             okfn != br_dev_queue_push_xmit)
 814 #if defined(CONFIG_VLAN_8021Q) || defined(CONFIG_VLAN_8021Q_MODULE)
 815             || ((out->priv_flags & IFF_802_1Q_VLAN) &&
 816             VLAN_DEV_INFO(out)->real_dev->hard_start_xmit == br_dev_xmit)
 817 #endif
 818             ) {
 819                 struct nf_bridge_info *nf_bridge;
 820
 821                 if (!skb->nf_bridge) {
 822 #ifdef CONFIG_SYSCTL
 823                         /* This code is executed while in the IP(v6) stack,
 824                            the version should be 4 or 6. We can't use
 825                            skb->protocol because that isn't set on
 826                            PF_INET(6)/LOCAL_OUT. */
 827                         struct iphdr *ip = skb->nh.iph;
 828
 829                         if (ip->version == 4 && !brnf_call_iptables)
 830                                 return NF_ACCEPT;
 831                         else if (ip->version == 6 && !brnf_call_ip6tables)
 832                                 return NF_ACCEPT;
 833 #endif
 834                         if (hook == NF_IP_POST_ROUTING)
 835                                 return NF_ACCEPT;
 836                         if (!nf_bridge_alloc(skb))
 837                                 return NF_DROP;
 838                 }
 839
 840                 nf_bridge = skb->nf_bridge;
 841
 842                 /* This frame will arrive on PF_BRIDGE/LOCAL_OUT and we
 843                  * will need the indev then. For a brouter, the real indev
 844                  * can be a bridge port, so we make sure br_nf_local_out()
 845                  * doesn't use the bridge parent of the indev by using
 846                  * the BRNF_DONT_TAKE_PARENT mask. */
 847                 if (hook == NF_IP_FORWARD && nf_bridge->physindev == NULL) {
 848                         nf_bridge->mask |= BRNF_DONT_TAKE_PARENT;
 849                         nf_bridge->physindev = (struct net_device *)in;
 850                 }
 851 #if defined(CONFIG_VLAN_8021Q) || defined(CONFIG_VLAN_8021Q_MODULE)
 852                 /* the iptables outdev is br0.x, not br0 */
 853                 if (out->priv_flags & IFF_802_1Q_VLAN)
 854                         nf_bridge->netoutdev = (struct net_device *)out;
 855 #endif
 856                 return NF_STOP;
 857         }
 858
 859         return NF_ACCEPT;
 860 }
 861
 862 /* For br_nf_local_out we need (prio = NF_BR_PRI_FIRST), to insure that innocent
 863  * PF_BRIDGE/NF_BR_LOCAL_OUT functions don't get bridged traffic as input.
 864  * For br_nf_post_routing, we need (prio = NF_BR_PRI_LAST), because
 865  * ip_refrag() can return NF_STOLEN. */
 866 static struct nf_hook_ops br_nf_ops[] = {
 867         { .hook = br_nf_pre_routing,
 868           .owner = THIS_MODULE,
 869           .pf = PF_BRIDGE,
 870           .hooknum = NF_BR_PRE_ROUTING,
 871           .priority = NF_BR_PRI_BRNF, },
 872         { .hook = br_nf_local_in,
 873           .owner = THIS_MODULE,
 874           .pf = PF_BRIDGE,
 875           .hooknum = NF_BR_LOCAL_IN,
 876           .priority = NF_BR_PRI_BRNF, },
 877         { .hook = br_nf_forward_ip,
 878           .owner = THIS_MODULE,
 879           .pf = PF_BRIDGE,
 880           .hooknum = NF_BR_FORWARD,
 881           .priority = NF_BR_PRI_BRNF - 1, },
 882         { .hook = br_nf_forward_arp,
 883           .owner = THIS_MODULE,
 884           .pf = PF_BRIDGE,
 885           .hooknum = NF_BR_FORWARD,
 886           .priority = NF_BR_PRI_BRNF, },
 887         { .hook = br_nf_local_out,
 888           .owner = THIS_MODULE,
 889           .pf = PF_BRIDGE,
 890           .hooknum = NF_BR_LOCAL_OUT,
 891           .priority = NF_BR_PRI_FIRST, },
 892         { .hook = br_nf_post_routing,
 893           .owner = THIS_MODULE,
 894           .pf = PF_BRIDGE,
 895           .hooknum = NF_BR_POST_ROUTING,
 896           .priority = NF_BR_PRI_LAST, },
 897         { .hook = ip_sabotage_in,
 898           .owner = THIS_MODULE,
 899           .pf = PF_INET,
 900           .hooknum = NF_IP_PRE_ROUTING,
 901           .priority = NF_IP_PRI_FIRST, },
 902         { .hook = ip_sabotage_in,
 903           .owner = THIS_MODULE,
 904           .pf = PF_INET6,
 905           .hooknum = NF_IP6_PRE_ROUTING,
 906           .priority = NF_IP6_PRI_FIRST, },
 907         { .hook = ip_sabotage_out,
 908           .owner = THIS_MODULE,
 909           .pf = PF_INET,
 910           .hooknum = NF_IP_FORWARD,
 911           .priority = NF_IP_PRI_BRIDGE_SABOTAGE_FORWARD, },
 912         { .hook = ip_sabotage_out,
 913           .owner = THIS_MODULE,
 914           .pf = PF_INET6,
 915           .hooknum = NF_IP6_FORWARD,
 916           .priority = NF_IP6_PRI_BRIDGE_SABOTAGE_FORWARD, },
 917         { .hook = ip_sabotage_out,
 918           .owner = THIS_MODULE,
 919           .pf = PF_INET,
 920           .hooknum = NF_IP_LOCAL_OUT,
 921           .priority = NF_IP_PRI_BRIDGE_SABOTAGE_LOCAL_OUT, },
 922         { .hook = ip_sabotage_out,
 923           .owner = THIS_MODULE,
 924           .pf = PF_INET6,
 925           .hooknum = NF_IP6_LOCAL_OUT,
 926           .priority = NF_IP6_PRI_BRIDGE_SABOTAGE_LOCAL_OUT, },
 927         { .hook = ip_sabotage_out,
 928           .owner = THIS_MODULE,
 929           .pf = PF_INET,
 930           .hooknum = NF_IP_POST_ROUTING,
 931           .priority = NF_IP_PRI_FIRST, },
 932         { .hook = ip_sabotage_out,
 933           .owner = THIS_MODULE,
 934           .pf = PF_INET6,
 935           .hooknum = NF_IP6_POST_ROUTING,
 936           .priority = NF_IP6_PRI_FIRST, },
 937 };
 938
 939 #ifdef CONFIG_SYSCTL
 940 static
 941 int brnf_sysctl_call_tables(ctl_table *ctl, int write, struct file * filp,
 942                         void __user *buffer, size_t *lenp, loff_t *ppos)
 943 {
 944         int ret;
 945
 946         ret = proc_dointvec(ctl, write, filp, buffer, lenp, ppos);
 947
 948         if (write && *(int *)(ctl->data))
 949                 *(int *)(ctl->data) = 1;
 950         return ret;
 951 }
 952
 953 static ctl_table brnf_table[] = {
 954         {
 955                 .ctl_name       = NET_BRIDGE_NF_CALL_ARPTABLES,
 956                 .procname       = "bridge-nf-call-arptables",
 957                 .data           = &brnf_call_arptables,
 958                 .maxlen         = sizeof(int),
 959                 .mode           = 0644,
 960                 .proc_handler   = &brnf_sysctl_call_tables,
 961         },
 962         {
 963                 .ctl_name       = NET_BRIDGE_NF_CALL_IPTABLES,
 964                 .procname       = "bridge-nf-call-iptables",
 965                 .data           = &brnf_call_iptables,
 966                 .maxlen         = sizeof(int),
 967                 .mode           = 0644,
 968                 .proc_handler   = &brnf_sysctl_call_tables,
 969         },
 970         {
 971                 .ctl_name       = NET_BRIDGE_NF_CALL_IP6TABLES,
 972                 .procname       = "bridge-nf-call-ip6tables",
 973                 .data           = &brnf_call_ip6tables,
 974                 .maxlen         = sizeof(int),
 975                 .mode           = 0644,
 976                 .proc_handler   = &brnf_sysctl_call_tables,
 977         },
 978         {
 979                 .ctl_name       = NET_BRIDGE_NF_FILTER_VLAN_TAGGED,
 980                 .procname       = "bridge-nf-filter-vlan-tagged",
 981                 .data           = &brnf_filter_vlan_tagged,
 982                 .maxlen         = sizeof(int),
 983                 .mode           = 0644,
 984                 .proc_handler   = &brnf_sysctl_call_tables,
 985         },
 986         { .ctl_name = 0 }
 987 };
 988
 989 static ctl_table brnf_bridge_table[] = {
 990         {
 991                 .ctl_name       = NET_BRIDGE,
 992                 .procname       = "bridge",
 993                 .mode           = 0555,
 994                 .child          = brnf_table,
 995         },
 996         { .ctl_name = 0 }
 997 };
 998
 999 static ctl_table brnf_net_table[] = {
1000         {
1001                 .ctl_name       = CTL_NET,
1002                 .procname       = "net",
1003                 .mode           = 0555,
1004                 .child          = brnf_bridge_table,
1005         },
1006         { .ctl_name = 0 }
1007 };
1008 #endif
1009
1010 int br_netfilter_init(void)
1011 {
1012         int i;
1013
1014         for (i = 0; i < ARRAY_SIZE(br_nf_ops); i++) {
1015                 int ret;
1016
1017                 if ((ret = nf_register_hook(&br_nf_ops[i])) >= 0)
1018                         continue;
1019
1020                 while (i--)
1021                         nf_unregister_hook(&br_nf_ops[i]);
1022
1023                 return ret;
1024         }
1025
1026 #ifdef CONFIG_SYSCTL
1027         brnf_sysctl_header = register_sysctl_table(brnf_net_table, 0);
1028         if (brnf_sysctl_header == NULL) {
1029                 printk(KERN_WARNING "br_netfilter: can't register to sysctl.\n");
1030                 for (i = 0; i < ARRAY_SIZE(br_nf_ops); i++)
1031                         nf_unregister_hook(&br_nf_ops[i]);
1032                 return -EFAULT;
1033         }
1034 #endif
1035
1036         printk(KERN_NOTICE "Bridge firewalling registered\n");
1037
1038         return 0;
1039 }
1040
1041 void br_netfilter_fini(void)
1042 {
1043         int i;
1044
1045         for (i = ARRAY_SIZE(br_nf_ops) - 1; i >= 0; i--)
1046                 nf_unregister_hook(&br_nf_ops[i]);
1047 #ifdef CONFIG_SYSCTL
1048         unregister_sysctl_table(brnf_sysctl_header);
1049 #endif
1050 }