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