SAFE public projects git trees. - safe/jmp/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                                  * - 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 = 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
 312                         pkt_len = ntohl(*(u32*)(skb->nh.raw+off+2));
 313
 314                         if (pkt_len > skb->len - sizeof(struct ipv6hdr))
 315                                 goto bad;
 316                         if (pkt_len + sizeof(struct ipv6hdr) < skb->len) {
 317                                 if (__pskb_trim(skb,
 318                                     pkt_len + sizeof(struct ipv6hdr)))
 319                                         goto bad;
 320                                 if (skb->ip_summed == CHECKSUM_HW)
 321                                         skb->ip_summed = CHECKSUM_NONE;
 322                         }
 323                         break;
 324                 default:
 325                         if (optlen > len)
 326                                 goto bad;
 327                         break;
 328                 }
 329                 off += optlen;
 330                 len -= optlen;
 331         }
 332         if (len == 0)
 333                 return 0;
 334 bad:
 335         return -1;
 336
 337 }
 338
 339 /* Replicate the checks that IPv6 does on packet reception and pass the packet
 340  * to ip6tables, which doesn't support NAT, so things are fairly simple. */
 341 static unsigned int br_nf_pre_routing_ipv6(unsigned int hook,
 342    struct sk_buff *skb, const struct net_device *in,
 343    const struct net_device *out, int (*okfn)(struct sk_buff *))
 344 {
 345         struct ipv6hdr *hdr;
 346         u32 pkt_len;
 347         struct nf_bridge_info *nf_bridge;
 348
 349         if (skb->len < sizeof(struct ipv6hdr))
 350                 goto inhdr_error;
 351
 352         if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
 353                 goto inhdr_error;
 354
 355         hdr = skb->nh.ipv6h;
 356
 357         if (hdr->version != 6)
 358                 goto inhdr_error;
 359
 360         pkt_len = ntohs(hdr->payload_len);
 361
 362         if (pkt_len || hdr->nexthdr != NEXTHDR_HOP) {
 363                 if (pkt_len + sizeof(struct ipv6hdr) > skb->len)
 364                         goto inhdr_error;
 365                 if (pkt_len + sizeof(struct ipv6hdr) < skb->len) {
 366                         if (__pskb_trim(skb, pkt_len + sizeof(struct ipv6hdr)))
 367                                 goto inhdr_error;
 368                         if (skb->ip_summed == CHECKSUM_HW)
 369                                 skb->ip_summed = CHECKSUM_NONE;
 370                 }
 371         }
 372         if (hdr->nexthdr == NEXTHDR_HOP && check_hbh_len(skb))
 373                         goto inhdr_error;
 374
 375         if ((nf_bridge = nf_bridge_alloc(skb)) == NULL)
 376                 return NF_DROP;
 377         setup_pre_routing(skb);
 378
 379         NF_HOOK(PF_INET6, NF_IP6_PRE_ROUTING, skb, skb->dev, NULL,
 380                 br_nf_pre_routing_finish_ipv6);
 381
 382         return NF_STOLEN;
 383
 384 inhdr_error:
 385         return NF_DROP;
 386 }
 387
 388 /* Direct IPv6 traffic to br_nf_pre_routing_ipv6.
 389  * Replicate the checks that IPv4 does on packet reception.
 390  * Set skb->dev to the bridge device (i.e. parent of the
 391  * receiving device) to make netfilter happy, the REDIRECT
 392  * target in particular.  Save the original destination IP
 393  * address to be able to detect DNAT afterwards. */
 394 static unsigned int br_nf_pre_routing(unsigned int hook, struct sk_buff **pskb,
 395    const struct net_device *in, const struct net_device *out,
 396    int (*okfn)(struct sk_buff *))
 397 {
 398         struct iphdr *iph;
 399         __u32 len;
 400         struct sk_buff *skb = *pskb;
 401         struct nf_bridge_info *nf_bridge;
 402         struct vlan_ethhdr *hdr = vlan_eth_hdr(*pskb);
 403
 404         if (skb->protocol == __constant_htons(ETH_P_IPV6) || IS_VLAN_IPV6) {
 405 #ifdef CONFIG_SYSCTL
 406                 if (!brnf_call_ip6tables)
 407                         return NF_ACCEPT;
 408 #endif
 409                 if ((skb = skb_share_check(*pskb, GFP_ATOMIC)) == NULL)
 410                         goto out;
 411
 412                 if (skb->protocol == __constant_htons(ETH_P_8021Q)) {
 413                         skb_pull(skb, VLAN_HLEN);
 414                         (skb)->nh.raw += VLAN_HLEN;
 415                 }
 416                 return br_nf_pre_routing_ipv6(hook, skb, in, out, okfn);
 417         }
 418 #ifdef CONFIG_SYSCTL
 419         if (!brnf_call_iptables)
 420                 return NF_ACCEPT;
 421 #endif
 422
 423         if (skb->protocol != __constant_htons(ETH_P_IP) && !IS_VLAN_IP)
 424                 return NF_ACCEPT;
 425
 426         if ((skb = skb_share_check(*pskb, GFP_ATOMIC)) == NULL)
 427                 goto out;
 428
 429         if (skb->protocol == __constant_htons(ETH_P_8021Q)) {
 430                 skb_pull(skb, VLAN_HLEN);
 431                 (skb)->nh.raw += VLAN_HLEN;
 432         }
 433
 434         if (!pskb_may_pull(skb, sizeof(struct iphdr)))
 435                 goto inhdr_error;
 436
 437         iph = skb->nh.iph;
 438         if (iph->ihl < 5 || iph->version != 4)
 439                 goto inhdr_error;
 440
 441         if (!pskb_may_pull(skb, 4*iph->ihl))
 442                 goto inhdr_error;
 443
 444         iph = skb->nh.iph;
 445         if (ip_fast_csum((__u8 *)iph, iph->ihl) != 0)
 446                 goto inhdr_error;
 447
 448         len = ntohs(iph->tot_len);
 449         if (skb->len < len || len < 4*iph->ihl)
 450                 goto inhdr_error;
 451
 452         if (skb->len > len) {
 453                 __pskb_trim(skb, len);
 454                 if (skb->ip_summed == CHECKSUM_HW)
 455                         skb->ip_summed = CHECKSUM_NONE;
 456         }
 457
 458         if ((nf_bridge = nf_bridge_alloc(skb)) == NULL)
 459                 return NF_DROP;
 460         setup_pre_routing(skb);
 461         store_orig_dstaddr(skb);
 462
 463         NF_HOOK(PF_INET, NF_IP_PRE_ROUTING, skb, skb->dev, NULL,
 464                 br_nf_pre_routing_finish);
 465
 466         return NF_STOLEN;
 467
 468 inhdr_error:
 469 //      IP_INC_STATS_BH(IpInHdrErrors);
 470 out:
 471         return NF_DROP;
 472 }
 473
 474
 475 /* PF_BRIDGE/LOCAL_IN ************************************************/
 476 /* The packet is locally destined, which requires a real
 477  * dst_entry, so detach the fake one.  On the way up, the
 478  * packet would pass through PRE_ROUTING again (which already
 479  * took place when the packet entered the bridge), but we
 480  * register an IPv4 PRE_ROUTING 'sabotage' hook that will
 481  * prevent this from happening. */
 482 static unsigned int br_nf_local_in(unsigned int hook, struct sk_buff **pskb,
 483    const struct net_device *in, const struct net_device *out,
 484    int (*okfn)(struct sk_buff *))
 485 {
 486         struct sk_buff *skb = *pskb;
 487
 488         if (skb->dst == (struct dst_entry *)&__fake_rtable) {
 489                 dst_release(skb->dst);
 490                 skb->dst = NULL;
 491         }
 492
 493         return NF_ACCEPT;
 494 }
 495
 496
 497 /* PF_BRIDGE/FORWARD *************************************************/
 498 static int br_nf_forward_finish(struct sk_buff *skb)
 499 {
 500         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 501         struct net_device *in;
 502         struct vlan_ethhdr *hdr = vlan_eth_hdr(skb);
 503
 504         if (skb->protocol != __constant_htons(ETH_P_ARP) && !IS_VLAN_ARP) {
 505                 in = nf_bridge->physindev;
 506                 if (nf_bridge->mask & BRNF_PKT_TYPE) {
 507                         skb->pkt_type = PACKET_OTHERHOST;
 508                         nf_bridge->mask ^= BRNF_PKT_TYPE;
 509                 }
 510         } else {
 511                 in = *((struct net_device **)(skb->cb));
 512         }
 513         if (skb->protocol == __constant_htons(ETH_P_8021Q)) {
 514                 skb_push(skb, VLAN_HLEN);
 515                 skb->nh.raw -= VLAN_HLEN;
 516         }
 517         NF_HOOK_THRESH(PF_BRIDGE, NF_BR_FORWARD, skb, in,
 518                         skb->dev, br_forward_finish, 1);
 519         return 0;
 520 }
 521
 522 /* This is the 'purely bridged' case.  For IP, we pass the packet to
 523  * netfilter with indev and outdev set to the bridge device,
 524  * but we are still able to filter on the 'real' indev/outdev
 525  * because of the physdev module. For ARP, indev and outdev are the
 526  * bridge ports. */
 527 static unsigned int br_nf_forward_ip(unsigned int hook, struct sk_buff **pskb,
 528    const struct net_device *in, const struct net_device *out,
 529    int (*okfn)(struct sk_buff *))
 530 {
 531         struct sk_buff *skb = *pskb;
 532         struct nf_bridge_info *nf_bridge;
 533         struct vlan_ethhdr *hdr = vlan_eth_hdr(skb);
 534         int pf;
 535
 536         if (!skb->nf_bridge)
 537                 return NF_ACCEPT;
 538
 539         if (skb->protocol == __constant_htons(ETH_P_IP) || IS_VLAN_IP)
 540                 pf = PF_INET;
 541         else
 542                 pf = PF_INET6;
 543
 544         if (skb->protocol == __constant_htons(ETH_P_8021Q)) {
 545                 skb_pull(*pskb, VLAN_HLEN);
 546                 (*pskb)->nh.raw += VLAN_HLEN;
 547         }
 548
 549         nf_bridge = skb->nf_bridge;
 550         if (skb->pkt_type == PACKET_OTHERHOST) {
 551                 skb->pkt_type = PACKET_HOST;
 552                 nf_bridge->mask |= BRNF_PKT_TYPE;
 553         }
 554
 555         /* The physdev module checks on this */
 556         nf_bridge->mask |= BRNF_BRIDGED;
 557         nf_bridge->physoutdev = skb->dev;
 558
 559         NF_HOOK(pf, NF_IP_FORWARD, skb, bridge_parent(in),
 560                 bridge_parent(out), br_nf_forward_finish);
 561
 562         return NF_STOLEN;
 563 }
 564
 565 static unsigned int br_nf_forward_arp(unsigned int hook, struct sk_buff **pskb,
 566    const struct net_device *in, const struct net_device *out,
 567    int (*okfn)(struct sk_buff *))
 568 {
 569         struct sk_buff *skb = *pskb;
 570         struct vlan_ethhdr *hdr = vlan_eth_hdr(skb);
 571         struct net_device **d = (struct net_device **)(skb->cb);
 572
 573 #ifdef CONFIG_SYSCTL
 574         if (!brnf_call_arptables)
 575                 return NF_ACCEPT;
 576 #endif
 577
 578         if (skb->protocol != __constant_htons(ETH_P_ARP)) {
 579                 if (!IS_VLAN_ARP)
 580                         return NF_ACCEPT;
 581                 skb_pull(*pskb, VLAN_HLEN);
 582                 (*pskb)->nh.raw += VLAN_HLEN;
 583         }
 584
 585         if (skb->nh.arph->ar_pln != 4) {
 586                 if (IS_VLAN_ARP) {
 587                         skb_push(*pskb, VLAN_HLEN);
 588                         (*pskb)->nh.raw -= VLAN_HLEN;
 589                 }
 590                 return NF_ACCEPT;
 591         }
 592         *d = (struct net_device *)in;
 593         NF_HOOK(NF_ARP, NF_ARP_FORWARD, skb, (struct net_device *)in,
 594                 (struct net_device *)out, br_nf_forward_finish);
 595
 596         return NF_STOLEN;
 597 }
 598
 599
 600 /* PF_BRIDGE/LOCAL_OUT ***********************************************/
 601 static int br_nf_local_out_finish(struct sk_buff *skb)
 602 {
 603         if (skb->protocol == __constant_htons(ETH_P_8021Q)) {
 604                 skb_push(skb, VLAN_HLEN);
 605                 skb->nh.raw -= VLAN_HLEN;
 606         }
 607
 608         NF_HOOK_THRESH(PF_BRIDGE, NF_BR_LOCAL_OUT, skb, NULL, skb->dev,
 609                         br_forward_finish, NF_BR_PRI_FIRST + 1);
 610
 611         return 0;
 612 }
 613
 614 /* This function sees both locally originated IP packets and forwarded
 615  * IP packets (in both cases the destination device is a bridge
 616  * device). It also sees bridged-and-DNAT'ed packets.
 617  * To be able to filter on the physical bridge devices (with the physdev
 618  * module), we steal packets destined to a bridge device away from the
 619  * PF_INET/FORWARD and PF_INET/OUTPUT hook functions, and give them back later,
 620  * when we have determined the real output device. This is done in here.
 621  *
 622  * If (nf_bridge->mask & BRNF_BRIDGED_DNAT) then the packet is bridged
 623  * and we fake the PF_BRIDGE/FORWARD hook. The function br_nf_forward()
 624  * will then fake the PF_INET/FORWARD hook. br_nf_local_out() has priority
 625  * NF_BR_PRI_FIRST, so no relevant PF_BRIDGE/INPUT functions have been nor
 626  * will be executed.
 627  * Otherwise, if nf_bridge->physindev is NULL, the bridge-nf code never touched
 628  * this packet before, and so the packet was locally originated. We fake
 629  * the PF_INET/LOCAL_OUT hook.
 630  * Finally, if nf_bridge->physindev isn't NULL, then the packet was IP routed,
 631  * so we fake the PF_INET/FORWARD hook. ip_sabotage_out() makes sure
 632  * even routed packets that didn't arrive on a bridge interface have their
 633  * nf_bridge->physindev set. */
 634 static unsigned int br_nf_local_out(unsigned int hook, struct sk_buff **pskb,
 635    const struct net_device *in, const struct net_device *out,
 636    int (*okfn)(struct sk_buff *))
 637 {
 638         struct net_device *realindev, *realoutdev;
 639         struct sk_buff *skb = *pskb;
 640         struct nf_bridge_info *nf_bridge;
 641         struct vlan_ethhdr *hdr = vlan_eth_hdr(skb);
 642         int pf;
 643
 644         if (!skb->nf_bridge)
 645                 return NF_ACCEPT;
 646
 647         if (skb->protocol == __constant_htons(ETH_P_IP) || IS_VLAN_IP)
 648                 pf = PF_INET;
 649         else
 650                 pf = PF_INET6;
 651
 652 #ifdef CONFIG_NETFILTER_DEBUG
 653         /* Sometimes we get packets with NULL ->dst here (for example,
 654          * running a dhcp client daemon triggers this). This should now
 655          * be fixed, but let's keep the check around. */
 656         if (skb->dst == NULL) {
 657                 printk(KERN_CRIT "br_netfilter: skb->dst == NULL.");
 658                 return NF_ACCEPT;
 659         }
 660 #endif
 661
 662         nf_bridge = skb->nf_bridge;
 663         nf_bridge->physoutdev = skb->dev;
 664         realindev = nf_bridge->physindev;
 665
 666         /* Bridged, take PF_BRIDGE/FORWARD.
 667          * (see big note in front of br_nf_pre_routing_finish) */
 668         if (nf_bridge->mask & BRNF_BRIDGED_DNAT) {
 669                 if (nf_bridge->mask & BRNF_PKT_TYPE) {
 670                         skb->pkt_type = PACKET_OTHERHOST;
 671                         nf_bridge->mask ^= BRNF_PKT_TYPE;
 672                 }
 673                 if (skb->protocol == __constant_htons(ETH_P_8021Q)) {
 674                         skb_push(skb, VLAN_HLEN);
 675                         skb->nh.raw -= VLAN_HLEN;
 676                 }
 677
 678                 NF_HOOK(PF_BRIDGE, NF_BR_FORWARD, skb, realindev,
 679                         skb->dev, br_forward_finish);
 680                 goto out;
 681         }
 682         realoutdev = bridge_parent(skb->dev);
 683
 684 #if defined(CONFIG_VLAN_8021Q) || defined(CONFIG_VLAN_8021Q_MODULE)
 685         /* iptables should match -o br0.x */
 686         if (nf_bridge->netoutdev)
 687                 realoutdev = nf_bridge->netoutdev;
 688 #endif
 689         if (skb->protocol == __constant_htons(ETH_P_8021Q)) {
 690                 skb_pull(skb, VLAN_HLEN);
 691                 (*pskb)->nh.raw += VLAN_HLEN;
 692         }
 693         /* IP forwarded traffic has a physindev, locally
 694          * generated traffic hasn't. */
 695         if (realindev != NULL) {
 696                 if (!(nf_bridge->mask & BRNF_DONT_TAKE_PARENT) &&
 697                     has_bridge_parent(realindev))
 698                         realindev = bridge_parent(realindev);
 699
 700                 NF_HOOK_THRESH(pf, NF_IP_FORWARD, skb, realindev,
 701                                realoutdev, br_nf_local_out_finish,
 702                                NF_IP_PRI_BRIDGE_SABOTAGE_FORWARD + 1);
 703         } else {
 704                 NF_HOOK_THRESH(pf, NF_IP_LOCAL_OUT, skb, realindev,
 705                                realoutdev, br_nf_local_out_finish,
 706                                NF_IP_PRI_BRIDGE_SABOTAGE_LOCAL_OUT + 1);
 707         }
 708
 709 out:
 710         return NF_STOLEN;
 711 }
 712
 713
 714 /* PF_BRIDGE/POST_ROUTING ********************************************/
 715 static unsigned int br_nf_post_routing(unsigned int hook, struct sk_buff **pskb,
 716    const struct net_device *in, const struct net_device *out,
 717    int (*okfn)(struct sk_buff *))
 718 {
 719         struct sk_buff *skb = *pskb;
 720         struct nf_bridge_info *nf_bridge = (*pskb)->nf_bridge;
 721         struct vlan_ethhdr *hdr = vlan_eth_hdr(skb);
 722         struct net_device *realoutdev = bridge_parent(skb->dev);
 723         int pf;
 724
 725 #ifdef CONFIG_NETFILTER_DEBUG
 726         /* Be very paranoid. This probably won't happen anymore, but let's
 727          * keep the check just to be sure... */
 728         if (skb->mac.raw < skb->head || skb->mac.raw + ETH_HLEN > skb->data) {
 729                 printk(KERN_CRIT "br_netfilter: Argh!! br_nf_post_routing: "
 730                                  "bad mac.raw pointer.");
 731                 goto print_error;
 732         }
 733 #endif
 734
 735         if (!nf_bridge)
 736                 return NF_ACCEPT;
 737
 738         if (skb->protocol == __constant_htons(ETH_P_IP) || IS_VLAN_IP)
 739                 pf = PF_INET;
 740         else
 741                 pf = PF_INET6;
 742
 743 #ifdef CONFIG_NETFILTER_DEBUG
 744         if (skb->dst == NULL) {
 745                 printk(KERN_CRIT "br_netfilter: skb->dst == NULL.");
 746                 goto print_error;
 747         }
 748 #endif
 749
 750         /* We assume any code from br_dev_queue_push_xmit onwards doesn't care
 751          * about the value of skb->pkt_type. */
 752         if (skb->pkt_type == PACKET_OTHERHOST) {
 753                 skb->pkt_type = PACKET_HOST;
 754                 nf_bridge->mask |= BRNF_PKT_TYPE;
 755         }
 756
 757         if (skb->protocol == __constant_htons(ETH_P_8021Q)) {
 758                 skb_pull(skb, VLAN_HLEN);
 759                 skb->nh.raw += VLAN_HLEN;
 760         }
 761
 762         nf_bridge_save_header(skb);
 763
 764 #if defined(CONFIG_VLAN_8021Q) || defined(CONFIG_VLAN_8021Q_MODULE)
 765         if (nf_bridge->netoutdev)
 766                 realoutdev = nf_bridge->netoutdev;
 767 #endif
 768         NF_HOOK(pf, NF_IP_POST_ROUTING, skb, NULL, realoutdev,
 769                 br_dev_queue_push_xmit);
 770
 771         return NF_STOLEN;
 772
 773 #ifdef CONFIG_NETFILTER_DEBUG
 774 print_error:
 775         if (skb->dev != NULL) {
 776                 printk("[%s]", skb->dev->name);
 777                 if (has_bridge_parent(skb->dev))
 778                         printk("[%s]", bridge_parent(skb->dev)->name);
 779         }
 780         printk(" head:%p, raw:%p, data:%p\n", skb->head, skb->mac.raw,
 781                                               skb->data);
 782         return NF_ACCEPT;
 783 #endif
 784 }
 785
 786
 787 /* IP/SABOTAGE *****************************************************/
 788 /* Don't hand locally destined packets to PF_INET(6)/PRE_ROUTING
 789  * for the second time. */
 790 static unsigned int ip_sabotage_in(unsigned int hook, struct sk_buff **pskb,
 791    const struct net_device *in, const struct net_device *out,
 792    int (*okfn)(struct sk_buff *))
 793 {
 794         if ((*pskb)->nf_bridge &&
 795             !((*pskb)->nf_bridge->mask & BRNF_NF_BRIDGE_PREROUTING)) {
 796                 return NF_STOP;
 797         }
 798
 799         return NF_ACCEPT;
 800 }
 801
 802 /* Postpone execution of PF_INET(6)/FORWARD, PF_INET(6)/LOCAL_OUT
 803  * and PF_INET(6)/POST_ROUTING until we have done the forwarding
 804  * decision in the bridge code and have determined nf_bridge->physoutdev. */
 805 static unsigned int ip_sabotage_out(unsigned int hook, struct sk_buff **pskb,
 806    const struct net_device *in, const struct net_device *out,
 807    int (*okfn)(struct sk_buff *))
 808 {
 809         struct sk_buff *skb = *pskb;
 810
 811         if ((out->hard_start_xmit == br_dev_xmit &&
 812             okfn != br_nf_forward_finish &&
 813             okfn != br_nf_local_out_finish &&
 814             okfn != br_dev_queue_push_xmit)
 815 #if defined(CONFIG_VLAN_8021Q) || defined(CONFIG_VLAN_8021Q_MODULE)
 816             || ((out->priv_flags & IFF_802_1Q_VLAN) &&
 817             VLAN_DEV_INFO(out)->real_dev->hard_start_xmit == br_dev_xmit)
 818 #endif
 819             ) {
 820                 struct nf_bridge_info *nf_bridge;
 821
 822                 if (!skb->nf_bridge) {
 823 #ifdef CONFIG_SYSCTL
 824                         /* This code is executed while in the IP(v6) stack,
 825                            the version should be 4 or 6. We can't use
 826                            skb->protocol because that isn't set on
 827                            PF_INET(6)/LOCAL_OUT. */
 828                         struct iphdr *ip = skb->nh.iph;
 829
 830                         if (ip->version == 4 && !brnf_call_iptables)
 831                                 return NF_ACCEPT;
 832                         else if (ip->version == 6 && !brnf_call_ip6tables)
 833                                 return NF_ACCEPT;
 834 #endif
 835                         if (hook == NF_IP_POST_ROUTING)
 836                                 return NF_ACCEPT;
 837                         if (!nf_bridge_alloc(skb))
 838                                 return NF_DROP;
 839                 }
 840
 841                 nf_bridge = skb->nf_bridge;
 842
 843                 /* This frame will arrive on PF_BRIDGE/LOCAL_OUT and we
 844                  * will need the indev then. For a brouter, the real indev
 845                  * can be a bridge port, so we make sure br_nf_local_out()
 846                  * doesn't use the bridge parent of the indev by using
 847                  * the BRNF_DONT_TAKE_PARENT mask. */
 848                 if (hook == NF_IP_FORWARD && nf_bridge->physindev == NULL) {
 849                         nf_bridge->mask |= BRNF_DONT_TAKE_PARENT;
 850                         nf_bridge->physindev = (struct net_device *)in;
 851                 }
 852 #if defined(CONFIG_VLAN_8021Q) || defined(CONFIG_VLAN_8021Q_MODULE)
 853                 /* the iptables outdev is br0.x, not br0 */
 854                 if (out->priv_flags & IFF_802_1Q_VLAN)
 855                         nf_bridge->netoutdev = (struct net_device *)out;
 856 #endif
 857                 return NF_STOP;
 858         }
 859
 860         return NF_ACCEPT;
 861 }
 862
 863 /* For br_nf_local_out we need (prio = NF_BR_PRI_FIRST), to insure that innocent
 864  * PF_BRIDGE/NF_BR_LOCAL_OUT functions don't get bridged traffic as input.
 865  * For br_nf_post_routing, we need (prio = NF_BR_PRI_LAST), because
 866  * ip_refrag() can return NF_STOLEN. */
 867 static struct nf_hook_ops br_nf_ops[] = {
 868         { .hook = br_nf_pre_routing,
 869           .owner = THIS_MODULE,
 870           .pf = PF_BRIDGE,
 871           .hooknum = NF_BR_PRE_ROUTING,
 872           .priority = NF_BR_PRI_BRNF, },
 873         { .hook = br_nf_local_in,
 874           .owner = THIS_MODULE,
 875           .pf = PF_BRIDGE,
 876           .hooknum = NF_BR_LOCAL_IN,
 877           .priority = NF_BR_PRI_BRNF, },
 878         { .hook = br_nf_forward_ip,
 879           .owner = THIS_MODULE,
 880           .pf = PF_BRIDGE,
 881           .hooknum = NF_BR_FORWARD,
 882           .priority = NF_BR_PRI_BRNF - 1, },
 883         { .hook = br_nf_forward_arp,
 884           .owner = THIS_MODULE,
 885           .pf = PF_BRIDGE,
 886           .hooknum = NF_BR_FORWARD,
 887           .priority = NF_BR_PRI_BRNF, },
 888         { .hook = br_nf_local_out,
 889           .owner = THIS_MODULE,
 890           .pf = PF_BRIDGE,
 891           .hooknum = NF_BR_LOCAL_OUT,
 892           .priority = NF_BR_PRI_FIRST, },
 893         { .hook = br_nf_post_routing,
 894           .owner = THIS_MODULE,
 895           .pf = PF_BRIDGE,
 896           .hooknum = NF_BR_POST_ROUTING,
 897           .priority = NF_BR_PRI_LAST, },
 898         { .hook = ip_sabotage_in,
 899           .owner = THIS_MODULE,
 900           .pf = PF_INET,
 901           .hooknum = NF_IP_PRE_ROUTING,
 902           .priority = NF_IP_PRI_FIRST, },
 903         { .hook = ip_sabotage_in,
 904           .owner = THIS_MODULE,
 905           .pf = PF_INET6,
 906           .hooknum = NF_IP6_PRE_ROUTING,
 907           .priority = NF_IP6_PRI_FIRST, },
 908         { .hook = ip_sabotage_out,
 909           .owner = THIS_MODULE,
 910           .pf = PF_INET,
 911           .hooknum = NF_IP_FORWARD,
 912           .priority = NF_IP_PRI_BRIDGE_SABOTAGE_FORWARD, },
 913         { .hook = ip_sabotage_out,
 914           .owner = THIS_MODULE,
 915           .pf = PF_INET6,
 916           .hooknum = NF_IP6_FORWARD,
 917           .priority = NF_IP6_PRI_BRIDGE_SABOTAGE_FORWARD, },
 918         { .hook = ip_sabotage_out,
 919           .owner = THIS_MODULE,
 920           .pf = PF_INET,
 921           .hooknum = NF_IP_LOCAL_OUT,
 922           .priority = NF_IP_PRI_BRIDGE_SABOTAGE_LOCAL_OUT, },
 923         { .hook = ip_sabotage_out,
 924           .owner = THIS_MODULE,
 925           .pf = PF_INET6,
 926           .hooknum = NF_IP6_LOCAL_OUT,
 927           .priority = NF_IP6_PRI_BRIDGE_SABOTAGE_LOCAL_OUT, },
 928         { .hook = ip_sabotage_out,
 929           .owner = THIS_MODULE,
 930           .pf = PF_INET,
 931           .hooknum = NF_IP_POST_ROUTING,
 932           .priority = NF_IP_PRI_FIRST, },
 933         { .hook = ip_sabotage_out,
 934           .owner = THIS_MODULE,
 935           .pf = PF_INET6,
 936           .hooknum = NF_IP6_POST_ROUTING,
 937           .priority = NF_IP6_PRI_FIRST, },
 938 };
 939
 940 #ifdef CONFIG_SYSCTL
 941 static
 942 int brnf_sysctl_call_tables(ctl_table *ctl, int write, struct file * filp,
 943                         void __user *buffer, size_t *lenp, loff_t *ppos)
 944 {
 945         int ret;
 946
 947         ret = proc_dointvec(ctl, write, filp, buffer, lenp, ppos);
 948
 949         if (write && *(int *)(ctl->data))
 950                 *(int *)(ctl->data) = 1;
 951         return ret;
 952 }
 953
 954 static ctl_table brnf_table[] = {
 955         {
 956                 .ctl_name       = NET_BRIDGE_NF_CALL_ARPTABLES,
 957                 .procname       = "bridge-nf-call-arptables",
 958                 .data           = &brnf_call_arptables,
 959                 .maxlen         = sizeof(int),
 960                 .mode           = 0644,
 961                 .proc_handler   = &brnf_sysctl_call_tables,
 962         },
 963         {
 964                 .ctl_name       = NET_BRIDGE_NF_CALL_IPTABLES,
 965                 .procname       = "bridge-nf-call-iptables",
 966                 .data           = &brnf_call_iptables,
 967                 .maxlen         = sizeof(int),
 968                 .mode           = 0644,
 969                 .proc_handler   = &brnf_sysctl_call_tables,
 970         },
 971         {
 972                 .ctl_name       = NET_BRIDGE_NF_CALL_IP6TABLES,
 973                 .procname       = "bridge-nf-call-ip6tables",
 974                 .data           = &brnf_call_ip6tables,
 975                 .maxlen         = sizeof(int),
 976                 .mode           = 0644,
 977                 .proc_handler   = &brnf_sysctl_call_tables,
 978         },
 979         {
 980                 .ctl_name       = NET_BRIDGE_NF_FILTER_VLAN_TAGGED,
 981                 .procname       = "bridge-nf-filter-vlan-tagged",
 982                 .data           = &brnf_filter_vlan_tagged,
 983                 .maxlen         = sizeof(int),
 984                 .mode           = 0644,
 985                 .proc_handler   = &brnf_sysctl_call_tables,
 986         },
 987         { .ctl_name = 0 }
 988 };
 989
 990 static ctl_table brnf_bridge_table[] = {
 991         {
 992                 .ctl_name       = NET_BRIDGE,
 993                 .procname       = "bridge",
 994                 .mode           = 0555,
 995                 .child          = brnf_table,
 996         },
 997         { .ctl_name = 0 }
 998 };
 999
1000 static ctl_table brnf_net_table[] = {
1001         {
1002                 .ctl_name       = CTL_NET,
1003                 .procname       = "net",
1004                 .mode           = 0555,
1005                 .child          = brnf_bridge_table,
1006         },
1007         { .ctl_name = 0 }
1008 };
1009 #endif
1010
1011 int br_netfilter_init(void)
1012 {
1013         int i;
1014
1015         for (i = 0; i < ARRAY_SIZE(br_nf_ops); i++) {
1016                 int ret;
1017
1018                 if ((ret = nf_register_hook(&br_nf_ops[i])) >= 0)
1019                         continue;
1020
1021                 while (i--)
1022                         nf_unregister_hook(&br_nf_ops[i]);
1023
1024                 return ret;
1025         }
1026
1027 #ifdef CONFIG_SYSCTL
1028         brnf_sysctl_header = register_sysctl_table(brnf_net_table, 0);
1029         if (brnf_sysctl_header == NULL) {
1030                 printk(KERN_WARNING "br_netfilter: can't register to sysctl.\n");
1031                 for (i = 0; i < ARRAY_SIZE(br_nf_ops); i++)
1032                         nf_unregister_hook(&br_nf_ops[i]);
1033                 return -EFAULT;
1034         }
1035 #endif
1036
1037         printk(KERN_NOTICE "Bridge firewalling registered\n");
1038
1039         return 0;
1040 }
1041
1042 void br_netfilter_fini(void)
1043 {
1044         int i;
1045
1046         for (i = ARRAY_SIZE(br_nf_ops) - 1; i >= 0; i--)
1047                 nf_unregister_hook(&br_nf_ops[i]);
1048 #ifdef CONFIG_SYSCTL
1049         unregister_sysctl_table(brnf_sysctl_header);
1050 #endif
1051 }