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_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                                 if (((struct dst_entry *)rt)->dev == dev) {
 220                                         skb->dst = (struct dst_entry *)rt;
 221                                         goto bridged_dnat;
 222                                 }
 223                                 __br_dnat_complain();
 224                                 dst_release((struct dst_entry *)rt);
 225                         }
 226                         kfree_skb(skb);
 227                         return 0;
 228                 } else {
 229                         if (skb->dst->dev == dev) {
 230 bridged_dnat:
 231                                 /* Tell br_nf_local_out this is a
 232                                  * bridged frame */
 233                                 nf_bridge->mask |= BRNF_BRIDGED_DNAT;
 234                                 skb->dev = nf_bridge->physindev;
 235                                 if (skb->protocol ==
 236                                     __constant_htons(ETH_P_8021Q)) {
 237                                         skb_push(skb, VLAN_HLEN);
 238                                         skb->nh.raw -= VLAN_HLEN;
 239                                 }
 240                                 NF_HOOK_THRESH(PF_BRIDGE, NF_BR_PRE_ROUTING,
 241                                                skb, skb->dev, NULL,
 242                                                br_nf_pre_routing_finish_bridge,
 243                                                1);
 244                                 return 0;
 245                         }
 246                         memcpy(eth_hdr(skb)->h_dest, dev->dev_addr,
 247                                ETH_ALEN);
 248                         skb->pkt_type = PACKET_HOST;
 249                 }
 250         } else {
 251                 skb->dst = (struct dst_entry *)&__fake_rtable;
 252                 dst_hold(skb->dst);
 253         }
 254
 255         skb->dev = nf_bridge->physindev;
 256         if (skb->protocol == __constant_htons(ETH_P_8021Q)) {
 257                 skb_push(skb, VLAN_HLEN);
 258                 skb->nh.raw -= VLAN_HLEN;
 259         }
 260         NF_HOOK_THRESH(PF_BRIDGE, NF_BR_PRE_ROUTING, skb, skb->dev, NULL,
 261                        br_handle_frame_finish, 1);
 262
 263         return 0;
 264 }
 265
 266 /* Some common code for IPv4/IPv6 */
 267 static void setup_pre_routing(struct sk_buff *skb)
 268 {
 269         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 270
 271         if (skb->pkt_type == PACKET_OTHERHOST) {
 272                 skb->pkt_type = PACKET_HOST;
 273                 nf_bridge->mask |= BRNF_PKT_TYPE;
 274         }
 275
 276         nf_bridge->mask |= BRNF_NF_BRIDGE_PREROUTING;
 277         nf_bridge->physindev = skb->dev;
 278         skb->dev = bridge_parent(skb->dev);
 279 }
 280
 281 /* We only check the length. A bridge shouldn't do any hop-by-hop stuff anyway */
 282 static int check_hbh_len(struct sk_buff *skb)
 283 {
 284         unsigned char *raw = (u8*)(skb->nh.ipv6h+1);
 285         u32 pkt_len;
 286         int off = raw - skb->nh.raw;
 287         int len = (raw[1]+1)<<3;
 288
 289         if ((raw + len) - skb->data > skb_headlen(skb))
 290                 goto bad;
 291
 292         off += 2;
 293         len -= 2;
 294
 295         while (len > 0) {
 296                 int optlen = raw[off+1]+2;
 297
 298                 switch (skb->nh.raw[off]) {
 299                 case IPV6_TLV_PAD0:
 300                         optlen = 1;
 301                         break;
 302
 303                 case IPV6_TLV_PADN:
 304                         break;
 305
 306                 case IPV6_TLV_JUMBO:
 307                         if (skb->nh.raw[off+1] != 4 || (off&3) != 2)
 308                                 goto bad;
 309
 310                         pkt_len = ntohl(*(u32*)(skb->nh.raw+off+2));
 311
 312                         if (pkt_len > skb->len - sizeof(struct ipv6hdr))
 313                                 goto bad;
 314                         if (pkt_len + sizeof(struct ipv6hdr) < skb->len) {
 315                                 if (__pskb_trim(skb,
 316                                     pkt_len + sizeof(struct ipv6hdr)))
 317                                         goto bad;
 318                                 if (skb->ip_summed == CHECKSUM_HW)
 319                                         skb->ip_summed = CHECKSUM_NONE;
 320                         }
 321                         break;
 322                 default:
 323                         if (optlen > len)
 324                                 goto bad;
 325                         break;
 326                 }
 327                 off += optlen;
 328                 len -= optlen;
 329         }
 330         if (len == 0)
 331                 return 0;
 332 bad:
 333         return -1;
 334
 335 }
 336
 337 /* Replicate the checks that IPv6 does on packet reception and pass the packet
 338  * to ip6tables, which doesn't support NAT, so things are fairly simple. */
 339 static unsigned int br_nf_pre_routing_ipv6(unsigned int hook,
 340    struct sk_buff *skb, const struct net_device *in,
 341    const struct net_device *out, int (*okfn)(struct sk_buff *))
 342 {
 343         struct ipv6hdr *hdr;
 344         u32 pkt_len;
 345         struct nf_bridge_info *nf_bridge;
 346
 347         if (skb->len < sizeof(struct ipv6hdr))
 348                 goto inhdr_error;
 349
 350         if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
 351                 goto inhdr_error;
 352
 353         hdr = skb->nh.ipv6h;
 354
 355         if (hdr->version != 6)
 356                 goto inhdr_error;
 357
 358         pkt_len = ntohs(hdr->payload_len);
 359
 360         if (pkt_len || hdr->nexthdr != NEXTHDR_HOP) {
 361                 if (pkt_len + sizeof(struct ipv6hdr) > skb->len)
 362                         goto inhdr_error;
 363                 if (pkt_len + sizeof(struct ipv6hdr) < skb->len) {
 364                         if (__pskb_trim(skb, pkt_len + sizeof(struct ipv6hdr)))
 365                                 goto inhdr_error;
 366                         if (skb->ip_summed == CHECKSUM_HW)
 367                                 skb->ip_summed = CHECKSUM_NONE;
 368                 }
 369         }
 370         if (hdr->nexthdr == NEXTHDR_HOP && check_hbh_len(skb))
 371                         goto inhdr_error;
 372
 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         if ((nf_bridge = nf_bridge_alloc(skb)) == NULL)
 457                 return NF_DROP;
 458         setup_pre_routing(skb);
 459         store_orig_dstaddr(skb);
 460
 461         NF_HOOK(PF_INET, NF_IP_PRE_ROUTING, skb, skb->dev, NULL,
 462                 br_nf_pre_routing_finish);
 463
 464         return NF_STOLEN;
 465
 466 inhdr_error:
 467 //      IP_INC_STATS_BH(IpInHdrErrors);
 468 out:
 469         return NF_DROP;
 470 }
 471
 472
 473 /* PF_BRIDGE/LOCAL_IN ************************************************/
 474 /* The packet is locally destined, which requires a real
 475  * dst_entry, so detach the fake one.  On the way up, the
 476  * packet would pass through PRE_ROUTING again (which already
 477  * took place when the packet entered the bridge), but we
 478  * register an IPv4 PRE_ROUTING 'sabotage' hook that will
 479  * prevent this from happening. */
 480 static unsigned int br_nf_local_in(unsigned int hook, struct sk_buff **pskb,
 481    const struct net_device *in, const struct net_device *out,
 482    int (*okfn)(struct sk_buff *))
 483 {
 484         struct sk_buff *skb = *pskb;
 485
 486         if (skb->dst == (struct dst_entry *)&__fake_rtable) {
 487                 dst_release(skb->dst);
 488                 skb->dst = NULL;
 489         }
 490
 491         return NF_ACCEPT;
 492 }
 493
 494
 495 /* PF_BRIDGE/FORWARD *************************************************/
 496 static int br_nf_forward_finish(struct sk_buff *skb)
 497 {
 498         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 499         struct net_device *in;
 500         struct vlan_ethhdr *hdr = vlan_eth_hdr(skb);
 501
 502         if (skb->protocol != __constant_htons(ETH_P_ARP) && !IS_VLAN_ARP) {
 503                 in = nf_bridge->physindev;
 504                 if (nf_bridge->mask & BRNF_PKT_TYPE) {
 505                         skb->pkt_type = PACKET_OTHERHOST;
 506                         nf_bridge->mask ^= BRNF_PKT_TYPE;
 507                 }
 508         } else {
 509                 in = *((struct net_device **)(skb->cb));
 510         }
 511         if (skb->protocol == __constant_htons(ETH_P_8021Q)) {
 512                 skb_push(skb, VLAN_HLEN);
 513                 skb->nh.raw -= VLAN_HLEN;
 514         }
 515         NF_HOOK_THRESH(PF_BRIDGE, NF_BR_FORWARD, skb, in,
 516                         skb->dev, br_forward_finish, 1);
 517         return 0;
 518 }
 519
 520 /* This is the 'purely bridged' case.  For IP, we pass the packet to
 521  * netfilter with indev and outdev set to the bridge device,
 522  * but we are still able to filter on the 'real' indev/outdev
 523  * because of the physdev module. For ARP, indev and outdev are the
 524  * bridge ports. */
 525 static unsigned int br_nf_forward_ip(unsigned int hook, struct sk_buff **pskb,
 526    const struct net_device *in, const struct net_device *out,
 527    int (*okfn)(struct sk_buff *))
 528 {
 529         struct sk_buff *skb = *pskb;
 530         struct nf_bridge_info *nf_bridge;
 531         struct vlan_ethhdr *hdr = vlan_eth_hdr(skb);
 532         int pf;
 533
 534         if (!skb->nf_bridge)
 535                 return NF_ACCEPT;
 536
 537         if (skb->protocol == __constant_htons(ETH_P_IP) || IS_VLAN_IP)
 538                 pf = PF_INET;
 539         else
 540                 pf = PF_INET6;
 541
 542         if (skb->protocol == __constant_htons(ETH_P_8021Q)) {
 543                 skb_pull(*pskb, VLAN_HLEN);
 544                 (*pskb)->nh.raw += VLAN_HLEN;
 545         }
 546
 547         nf_bridge = skb->nf_bridge;
 548         if (skb->pkt_type == PACKET_OTHERHOST) {
 549                 skb->pkt_type = PACKET_HOST;
 550                 nf_bridge->mask |= BRNF_PKT_TYPE;
 551         }
 552
 553         /* The physdev module checks on this */
 554         nf_bridge->mask |= BRNF_BRIDGED;
 555         nf_bridge->physoutdev = skb->dev;
 556
 557         NF_HOOK(pf, NF_IP_FORWARD, skb, bridge_parent(in),
 558                 bridge_parent(out), br_nf_forward_finish);
 559
 560         return NF_STOLEN;
 561 }
 562
 563 static unsigned int br_nf_forward_arp(unsigned int hook, struct sk_buff **pskb,
 564    const struct net_device *in, const struct net_device *out,
 565    int (*okfn)(struct sk_buff *))
 566 {
 567         struct sk_buff *skb = *pskb;
 568         struct vlan_ethhdr *hdr = vlan_eth_hdr(skb);
 569         struct net_device **d = (struct net_device **)(skb->cb);
 570
 571 #ifdef CONFIG_SYSCTL
 572         if (!brnf_call_arptables)
 573                 return NF_ACCEPT;
 574 #endif
 575
 576         if (skb->protocol != __constant_htons(ETH_P_ARP)) {
 577                 if (!IS_VLAN_ARP)
 578                         return NF_ACCEPT;
 579                 skb_pull(*pskb, VLAN_HLEN);
 580                 (*pskb)->nh.raw += VLAN_HLEN;
 581         }
 582
 583         if (skb->nh.arph->ar_pln != 4) {
 584                 if (IS_VLAN_ARP) {
 585                         skb_push(*pskb, VLAN_HLEN);
 586                         (*pskb)->nh.raw -= VLAN_HLEN;
 587                 }
 588                 return NF_ACCEPT;
 589         }
 590         *d = (struct net_device *)in;
 591         NF_HOOK(NF_ARP, NF_ARP_FORWARD, skb, (struct net_device *)in,
 592                 (struct net_device *)out, br_nf_forward_finish);
 593
 594         return NF_STOLEN;
 595 }
 596
 597
 598 /* PF_BRIDGE/LOCAL_OUT ***********************************************/
 599 static int br_nf_local_out_finish(struct sk_buff *skb)
 600 {
 601         if (skb->protocol == __constant_htons(ETH_P_8021Q)) {
 602                 skb_push(skb, VLAN_HLEN);
 603                 skb->nh.raw -= VLAN_HLEN;
 604         }
 605
 606         NF_HOOK_THRESH(PF_BRIDGE, NF_BR_LOCAL_OUT, skb, NULL, skb->dev,
 607                         br_forward_finish, NF_BR_PRI_FIRST + 1);
 608
 609         return 0;
 610 }
 611
 612 /* This function sees both locally originated IP packets and forwarded
 613  * IP packets (in both cases the destination device is a bridge
 614  * device). It also sees bridged-and-DNAT'ed packets.
 615  * To be able to filter on the physical bridge devices (with the physdev
 616  * module), we steal packets destined to a bridge device away from the
 617  * PF_INET/FORWARD and PF_INET/OUTPUT hook functions, and give them back later,
 618  * when we have determined the real output device. This is done in here.
 619  *
 620  * If (nf_bridge->mask & BRNF_BRIDGED_DNAT) then the packet is bridged
 621  * and we fake the PF_BRIDGE/FORWARD hook. The function br_nf_forward()
 622  * will then fake the PF_INET/FORWARD hook. br_nf_local_out() has priority
 623  * NF_BR_PRI_FIRST, so no relevant PF_BRIDGE/INPUT functions have been nor
 624  * will be executed.
 625  * Otherwise, if nf_bridge->physindev is NULL, the bridge-nf code never touched
 626  * this packet before, and so the packet was locally originated. We fake
 627  * the PF_INET/LOCAL_OUT hook.
 628  * Finally, if nf_bridge->physindev isn't NULL, then the packet was IP routed,
 629  * so we fake the PF_INET/FORWARD hook. ip_sabotage_out() makes sure
 630  * even routed packets that didn't arrive on a bridge interface have their
 631  * nf_bridge->physindev set. */
 632 static unsigned int br_nf_local_out(unsigned int hook, struct sk_buff **pskb,
 633    const struct net_device *in, const struct net_device *out,
 634    int (*okfn)(struct sk_buff *))
 635 {
 636         struct net_device *realindev, *realoutdev;
 637         struct sk_buff *skb = *pskb;
 638         struct nf_bridge_info *nf_bridge;
 639         struct vlan_ethhdr *hdr = vlan_eth_hdr(skb);
 640         int pf;
 641
 642         if (!skb->nf_bridge)
 643                 return NF_ACCEPT;
 644
 645         if (skb->protocol == __constant_htons(ETH_P_IP) || IS_VLAN_IP)
 646                 pf = PF_INET;
 647         else
 648                 pf = PF_INET6;
 649
 650 #ifdef CONFIG_NETFILTER_DEBUG
 651         /* Sometimes we get packets with NULL ->dst here (for example,
 652          * running a dhcp client daemon triggers this). This should now
 653          * be fixed, but let's keep the check around. */
 654         if (skb->dst == NULL) {
 655                 printk(KERN_CRIT "br_netfilter: skb->dst == NULL.");
 656                 return NF_ACCEPT;
 657         }
 658 #endif
 659
 660         nf_bridge = skb->nf_bridge;
 661         nf_bridge->physoutdev = skb->dev;
 662         realindev = nf_bridge->physindev;
 663
 664         /* Bridged, take PF_BRIDGE/FORWARD.
 665          * (see big note in front of br_nf_pre_routing_finish) */
 666         if (nf_bridge->mask & BRNF_BRIDGED_DNAT) {
 667                 if (nf_bridge->mask & BRNF_PKT_TYPE) {
 668                         skb->pkt_type = PACKET_OTHERHOST;
 669                         nf_bridge->mask ^= BRNF_PKT_TYPE;
 670                 }
 671                 if (skb->protocol == __constant_htons(ETH_P_8021Q)) {
 672                         skb_push(skb, VLAN_HLEN);
 673                         skb->nh.raw -= VLAN_HLEN;
 674                 }
 675
 676                 NF_HOOK(PF_BRIDGE, NF_BR_FORWARD, skb, realindev,
 677                         skb->dev, br_forward_finish);
 678                 goto out;
 679         }
 680         realoutdev = bridge_parent(skb->dev);
 681
 682 #if defined(CONFIG_VLAN_8021Q) || defined(CONFIG_VLAN_8021Q_MODULE)
 683         /* iptables should match -o br0.x */
 684         if (nf_bridge->netoutdev)
 685                 realoutdev = nf_bridge->netoutdev;
 686 #endif
 687         if (skb->protocol == __constant_htons(ETH_P_8021Q)) {
 688                 skb_pull(skb, VLAN_HLEN);
 689                 (*pskb)->nh.raw += VLAN_HLEN;
 690         }
 691         /* IP forwarded traffic has a physindev, locally
 692          * generated traffic hasn't. */
 693         if (realindev != NULL) {
 694                 if (!(nf_bridge->mask & BRNF_DONT_TAKE_PARENT) &&
 695                     has_bridge_parent(realindev))
 696                         realindev = bridge_parent(realindev);
 697
 698                 NF_HOOK_THRESH(pf, NF_IP_FORWARD, skb, realindev,
 699                                realoutdev, br_nf_local_out_finish,
 700                                NF_IP_PRI_BRIDGE_SABOTAGE_FORWARD + 1);
 701         } else {
 702                 NF_HOOK_THRESH(pf, NF_IP_LOCAL_OUT, skb, realindev,
 703                                realoutdev, br_nf_local_out_finish,
 704                                NF_IP_PRI_BRIDGE_SABOTAGE_LOCAL_OUT + 1);
 705         }
 706
 707 out:
 708         return NF_STOLEN;
 709 }
 710
 711
 712 /* PF_BRIDGE/POST_ROUTING ********************************************/
 713 static unsigned int br_nf_post_routing(unsigned int hook, struct sk_buff **pskb,
 714    const struct net_device *in, const struct net_device *out,
 715    int (*okfn)(struct sk_buff *))
 716 {
 717         struct sk_buff *skb = *pskb;
 718         struct nf_bridge_info *nf_bridge = (*pskb)->nf_bridge;
 719         struct vlan_ethhdr *hdr = vlan_eth_hdr(skb);
 720         struct net_device *realoutdev = bridge_parent(skb->dev);
 721         int pf;
 722
 723 #ifdef CONFIG_NETFILTER_DEBUG
 724         /* Be very paranoid. This probably won't happen anymore, but let's
 725          * keep the check just to be sure... */
 726         if (skb->mac.raw < skb->head || skb->mac.raw + ETH_HLEN > skb->data) {
 727                 printk(KERN_CRIT "br_netfilter: Argh!! br_nf_post_routing: "
 728                                  "bad mac.raw pointer.");
 729                 goto print_error;
 730         }
 731 #endif
 732
 733         if (!nf_bridge)
 734                 return NF_ACCEPT;
 735
 736         if (skb->protocol == __constant_htons(ETH_P_IP) || IS_VLAN_IP)
 737                 pf = PF_INET;
 738         else
 739                 pf = PF_INET6;
 740
 741 #ifdef CONFIG_NETFILTER_DEBUG
 742         if (skb->dst == NULL) {
 743                 printk(KERN_CRIT "br_netfilter: skb->dst == NULL.");
 744                 goto print_error;
 745         }
 746 #endif
 747
 748         /* We assume any code from br_dev_queue_push_xmit onwards doesn't care
 749          * about the value of skb->pkt_type. */
 750         if (skb->pkt_type == PACKET_OTHERHOST) {
 751                 skb->pkt_type = PACKET_HOST;
 752                 nf_bridge->mask |= BRNF_PKT_TYPE;
 753         }
 754
 755         if (skb->protocol == __constant_htons(ETH_P_8021Q)) {
 756                 skb_pull(skb, VLAN_HLEN);
 757                 skb->nh.raw += VLAN_HLEN;
 758         }
 759
 760         nf_bridge_save_header(skb);
 761
 762 #if defined(CONFIG_VLAN_8021Q) || defined(CONFIG_VLAN_8021Q_MODULE)
 763         if (nf_bridge->netoutdev)
 764                 realoutdev = nf_bridge->netoutdev;
 765 #endif
 766         NF_HOOK(pf, NF_IP_POST_ROUTING, skb, NULL, realoutdev,
 767                 br_dev_queue_push_xmit);
 768
 769         return NF_STOLEN;
 770
 771 #ifdef CONFIG_NETFILTER_DEBUG
 772 print_error:
 773         if (skb->dev != NULL) {
 774                 printk("[%s]", skb->dev->name);
 775                 if (has_bridge_parent(skb->dev))
 776                         printk("[%s]", bridge_parent(skb->dev)->name);
 777         }
 778         printk(" head:%p, raw:%p, data:%p\n", skb->head, skb->mac.raw,
 779                                               skb->data);
 780         return NF_ACCEPT;
 781 #endif
 782 }
 783
 784
 785 /* IP/SABOTAGE *****************************************************/
 786 /* Don't hand locally destined packets to PF_INET(6)/PRE_ROUTING
 787  * for the second time. */
 788 static unsigned int ip_sabotage_in(unsigned int hook, struct sk_buff **pskb,
 789    const struct net_device *in, const struct net_device *out,
 790    int (*okfn)(struct sk_buff *))
 791 {
 792         if ((*pskb)->nf_bridge &&
 793             !((*pskb)->nf_bridge->mask & BRNF_NF_BRIDGE_PREROUTING)) {
 794                 return NF_STOP;
 795         }
 796
 797         return NF_ACCEPT;
 798 }
 799
 800 /* Postpone execution of PF_INET(6)/FORWARD, PF_INET(6)/LOCAL_OUT
 801  * and PF_INET(6)/POST_ROUTING until we have done the forwarding
 802  * decision in the bridge code and have determined nf_bridge->physoutdev. */
 803 static unsigned int ip_sabotage_out(unsigned int hook, struct sk_buff **pskb,
 804    const struct net_device *in, const struct net_device *out,
 805    int (*okfn)(struct sk_buff *))
 806 {
 807         struct sk_buff *skb = *pskb;
 808
 809         if ((out->hard_start_xmit == br_dev_xmit &&
 810             okfn != br_nf_forward_finish &&
 811             okfn != br_nf_local_out_finish &&
 812             okfn != br_dev_queue_push_xmit)
 813 #if defined(CONFIG_VLAN_8021Q) || defined(CONFIG_VLAN_8021Q_MODULE)
 814             || ((out->priv_flags & IFF_802_1Q_VLAN) &&
 815             VLAN_DEV_INFO(out)->real_dev->hard_start_xmit == br_dev_xmit)
 816 #endif
 817             ) {
 818                 struct nf_bridge_info *nf_bridge;
 819
 820                 if (!skb->nf_bridge) {
 821 #ifdef CONFIG_SYSCTL
 822                         /* This code is executed while in the IP(v6) stack,
 823                            the version should be 4 or 6. We can't use
 824                            skb->protocol because that isn't set on
 825                            PF_INET(6)/LOCAL_OUT. */
 826                         struct iphdr *ip = skb->nh.iph;
 827
 828                         if (ip->version == 4 && !brnf_call_iptables)
 829                                 return NF_ACCEPT;
 830                         else if (ip->version == 6 && !brnf_call_ip6tables)
 831                                 return NF_ACCEPT;
 832 #endif
 833                         if (hook == NF_IP_POST_ROUTING)
 834                                 return NF_ACCEPT;
 835                         if (!nf_bridge_alloc(skb))
 836                                 return NF_DROP;
 837                 }
 838
 839                 nf_bridge = skb->nf_bridge;
 840
 841                 /* This frame will arrive on PF_BRIDGE/LOCAL_OUT and we
 842                  * will need the indev then. For a brouter, the real indev
 843                  * can be a bridge port, so we make sure br_nf_local_out()
 844                  * doesn't use the bridge parent of the indev by using
 845                  * the BRNF_DONT_TAKE_PARENT mask. */
 846                 if (hook == NF_IP_FORWARD && nf_bridge->physindev == NULL) {
 847                         nf_bridge->mask |= BRNF_DONT_TAKE_PARENT;
 848                         nf_bridge->physindev = (struct net_device *)in;
 849                 }
 850 #if defined(CONFIG_VLAN_8021Q) || defined(CONFIG_VLAN_8021Q_MODULE)
 851                 /* the iptables outdev is br0.x, not br0 */
 852                 if (out->priv_flags & IFF_802_1Q_VLAN)
 853                         nf_bridge->netoutdev = (struct net_device *)out;
 854 #endif
 855                 return NF_STOP;
 856         }
 857
 858         return NF_ACCEPT;
 859 }
 860
 861 /* For br_nf_local_out we need (prio = NF_BR_PRI_FIRST), to insure that innocent
 862  * PF_BRIDGE/NF_BR_LOCAL_OUT functions don't get bridged traffic as input.
 863  * For br_nf_post_routing, we need (prio = NF_BR_PRI_LAST), because
 864  * ip_refrag() can return NF_STOLEN. */
 865 static struct nf_hook_ops br_nf_ops[] = {
 866         { .hook = br_nf_pre_routing,
 867           .owner = THIS_MODULE,
 868           .pf = PF_BRIDGE,
 869           .hooknum = NF_BR_PRE_ROUTING,
 870           .priority = NF_BR_PRI_BRNF, },
 871         { .hook = br_nf_local_in,
 872           .owner = THIS_MODULE,
 873           .pf = PF_BRIDGE,
 874           .hooknum = NF_BR_LOCAL_IN,
 875           .priority = NF_BR_PRI_BRNF, },
 876         { .hook = br_nf_forward_ip,
 877           .owner = THIS_MODULE,
 878           .pf = PF_BRIDGE,
 879           .hooknum = NF_BR_FORWARD,
 880           .priority = NF_BR_PRI_BRNF - 1, },
 881         { .hook = br_nf_forward_arp,
 882           .owner = THIS_MODULE,
 883           .pf = PF_BRIDGE,
 884           .hooknum = NF_BR_FORWARD,
 885           .priority = NF_BR_PRI_BRNF, },
 886         { .hook = br_nf_local_out,
 887           .owner = THIS_MODULE,
 888           .pf = PF_BRIDGE,
 889           .hooknum = NF_BR_LOCAL_OUT,
 890           .priority = NF_BR_PRI_FIRST, },
 891         { .hook = br_nf_post_routing,
 892           .owner = THIS_MODULE,
 893           .pf = PF_BRIDGE,
 894           .hooknum = NF_BR_POST_ROUTING,
 895           .priority = NF_BR_PRI_LAST, },
 896         { .hook = ip_sabotage_in,
 897           .owner = THIS_MODULE,
 898           .pf = PF_INET,
 899           .hooknum = NF_IP_PRE_ROUTING,
 900           .priority = NF_IP_PRI_FIRST, },
 901         { .hook = ip_sabotage_in,
 902           .owner = THIS_MODULE,
 903           .pf = PF_INET6,
 904           .hooknum = NF_IP6_PRE_ROUTING,
 905           .priority = NF_IP6_PRI_FIRST, },
 906         { .hook = ip_sabotage_out,
 907           .owner = THIS_MODULE,
 908           .pf = PF_INET,
 909           .hooknum = NF_IP_FORWARD,
 910           .priority = NF_IP_PRI_BRIDGE_SABOTAGE_FORWARD, },
 911         { .hook = ip_sabotage_out,
 912           .owner = THIS_MODULE,
 913           .pf = PF_INET6,
 914           .hooknum = NF_IP6_FORWARD,
 915           .priority = NF_IP6_PRI_BRIDGE_SABOTAGE_FORWARD, },
 916         { .hook = ip_sabotage_out,
 917           .owner = THIS_MODULE,
 918           .pf = PF_INET,
 919           .hooknum = NF_IP_LOCAL_OUT,
 920           .priority = NF_IP_PRI_BRIDGE_SABOTAGE_LOCAL_OUT, },
 921         { .hook = ip_sabotage_out,
 922           .owner = THIS_MODULE,
 923           .pf = PF_INET6,
 924           .hooknum = NF_IP6_LOCAL_OUT,
 925           .priority = NF_IP6_PRI_BRIDGE_SABOTAGE_LOCAL_OUT, },
 926         { .hook = ip_sabotage_out,
 927           .owner = THIS_MODULE,
 928           .pf = PF_INET,
 929           .hooknum = NF_IP_POST_ROUTING,
 930           .priority = NF_IP_PRI_FIRST, },
 931         { .hook = ip_sabotage_out,
 932           .owner = THIS_MODULE,
 933           .pf = PF_INET6,
 934           .hooknum = NF_IP6_POST_ROUTING,
 935           .priority = NF_IP6_PRI_FIRST, },
 936 };
 937
 938 #ifdef CONFIG_SYSCTL
 939 static
 940 int brnf_sysctl_call_tables(ctl_table *ctl, int write, struct file * filp,
 941                         void __user *buffer, size_t *lenp, loff_t *ppos)
 942 {
 943         int ret;
 944
 945         ret = proc_dointvec(ctl, write, filp, buffer, lenp, ppos);
 946
 947         if (write && *(int *)(ctl->data))
 948                 *(int *)(ctl->data) = 1;
 949         return ret;
 950 }
 951
 952 static ctl_table brnf_table[] = {
 953         {
 954                 .ctl_name       = NET_BRIDGE_NF_CALL_ARPTABLES,
 955                 .procname       = "bridge-nf-call-arptables",
 956                 .data           = &brnf_call_arptables,
 957                 .maxlen         = sizeof(int),
 958                 .mode           = 0644,
 959                 .proc_handler   = &brnf_sysctl_call_tables,
 960         },
 961         {
 962                 .ctl_name       = NET_BRIDGE_NF_CALL_IPTABLES,
 963                 .procname       = "bridge-nf-call-iptables",
 964                 .data           = &brnf_call_iptables,
 965                 .maxlen         = sizeof(int),
 966                 .mode           = 0644,
 967                 .proc_handler   = &brnf_sysctl_call_tables,
 968         },
 969         {
 970                 .ctl_name       = NET_BRIDGE_NF_CALL_IP6TABLES,
 971                 .procname       = "bridge-nf-call-ip6tables",
 972                 .data           = &brnf_call_ip6tables,
 973                 .maxlen         = sizeof(int),
 974                 .mode           = 0644,
 975                 .proc_handler   = &brnf_sysctl_call_tables,
 976         },
 977         {
 978                 .ctl_name       = NET_BRIDGE_NF_FILTER_VLAN_TAGGED,
 979                 .procname       = "bridge-nf-filter-vlan-tagged",
 980                 .data           = &brnf_filter_vlan_tagged,
 981                 .maxlen         = sizeof(int),
 982                 .mode           = 0644,
 983                 .proc_handler   = &brnf_sysctl_call_tables,
 984         },
 985         { .ctl_name = 0 }
 986 };
 987
 988 static ctl_table brnf_bridge_table[] = {
 989         {
 990                 .ctl_name       = NET_BRIDGE,
 991                 .procname       = "bridge",
 992                 .mode           = 0555,
 993                 .child          = brnf_table,
 994         },
 995         { .ctl_name = 0 }
 996 };
 997
 998 static ctl_table brnf_net_table[] = {
 999         {
1000                 .ctl_name       = CTL_NET,
1001                 .procname       = "net",
1002                 .mode           = 0555,
1003                 .child          = brnf_bridge_table,
1004         },
1005         { .ctl_name = 0 }
1006 };
1007 #endif
1008
1009 int br_netfilter_init(void)
1010 {
1011         int i;
1012
1013         for (i = 0; i < ARRAY_SIZE(br_nf_ops); i++) {
1014                 int ret;
1015
1016                 if ((ret = nf_register_hook(&br_nf_ops[i])) >= 0)
1017                         continue;
1018
1019                 while (i--)
1020                         nf_unregister_hook(&br_nf_ops[i]);
1021
1022                 return ret;
1023         }
1024
1025 #ifdef CONFIG_SYSCTL
1026         brnf_sysctl_header = register_sysctl_table(brnf_net_table, 0);
1027         if (brnf_sysctl_header == NULL) {
1028                 printk(KERN_WARNING "br_netfilter: can't register to sysctl.\n");
1029                 for (i = 0; i < ARRAY_SIZE(br_nf_ops); i++)
1030                         nf_unregister_hook(&br_nf_ops[i]);
1031                 return -EFAULT;
1032         }
1033 #endif
1034
1035         printk(KERN_NOTICE "Bridge firewalling registered\n");
1036
1037         return 0;
1038 }
1039
1040 void br_netfilter_fini(void)
1041 {
1042         int i;
1043
1044         for (i = ARRAY_SIZE(br_nf_ops) - 1; i >= 0; i--)
1045                 nf_unregister_hook(&br_nf_ops[i]);
1046 #ifdef CONFIG_SYSCTL
1047         unregister_sysctl_table(brnf_sysctl_header);
1048 #endif
1049 }