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_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 "br_private.h"
  44 #ifdef CONFIG_SYSCTL
  45 #include <linux/sysctl.h>
  46 #endif
  47
  48 #define skb_origaddr(skb)        (((struct bridge_skb_cb *) \
  49                                  (skb->nf_bridge->data))->daddr.ipv4)
  50 #define store_orig_dstaddr(skb)  (skb_origaddr(skb) = (skb)->nh.iph->daddr)
  51 #define dnat_took_place(skb)     (skb_origaddr(skb) != (skb)->nh.iph->daddr)
  52
  53 #ifdef CONFIG_SYSCTL
  54 static struct ctl_table_header *brnf_sysctl_header;
  55 static int brnf_call_iptables __read_mostly = 1;
  56 static int brnf_call_ip6tables __read_mostly = 1;
  57 static int brnf_call_arptables __read_mostly = 1;
  58 static int brnf_filter_vlan_tagged __read_mostly = 1;
  59 #else
  60 #define brnf_filter_vlan_tagged 1
  61 #endif
  62
  63 int brnf_deferred_hooks;
  64 EXPORT_SYMBOL_GPL(brnf_deferred_hooks);
  65
  66 static __be16 inline vlan_proto(const struct sk_buff *skb)
  67 {
  68         return vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
  69 }
  70
  71 #define IS_VLAN_IP(skb) \
  72         (skb->protocol == htons(ETH_P_8021Q) && \
  73          vlan_proto(skb) == htons(ETH_P_IP) &&  \
  74          brnf_filter_vlan_tagged)
  75
  76 #define IS_VLAN_IPV6(skb) \
  77         (skb->protocol == htons(ETH_P_8021Q) && \
  78          vlan_proto(skb) == htons(ETH_P_IPV6) &&\
  79          brnf_filter_vlan_tagged)
  80
  81 #define IS_VLAN_ARP(skb) \
  82         (skb->protocol == htons(ETH_P_8021Q) && \
  83          vlan_proto(skb) == htons(ETH_P_ARP) && \
  84          brnf_filter_vlan_tagged)
  85
  86 /* We need these fake structures to make netfilter happy --
  87  * lots of places assume that skb->dst != NULL, which isn't
  88  * all that unreasonable.
  89  *
  90  * Currently, we fill in the PMTU entry because netfilter
  91  * refragmentation needs it, and the rt_flags entry because
  92  * ipt_REJECT needs it.  Future netfilter modules might
  93  * require us to fill additional fields. */
  94 static struct net_device __fake_net_device = {
  95         .hard_header_len        = ETH_HLEN
  96 };
  97
  98 static struct rtable __fake_rtable = {
  99         .u = {
 100                 .dst = {
 101                         .__refcnt               = ATOMIC_INIT(1),
 102                         .dev                    = &__fake_net_device,
 103                         .path                   = &__fake_rtable.u.dst,
 104                         .metrics                = {[RTAX_MTU - 1] = 1500},
 105                         .flags                  = DST_NOXFRM,
 106                 }
 107         },
 108         .rt_flags       = 0,
 109 };
 110
 111 static inline struct net_device *bridge_parent(const struct net_device *dev)
 112 {
 113         struct net_bridge_port *port = rcu_dereference(dev->br_port);
 114
 115         return port ? port->br->dev : NULL;
 116 }
 117
 118 static inline struct nf_bridge_info *nf_bridge_alloc(struct sk_buff *skb)
 119 {
 120         skb->nf_bridge = kzalloc(sizeof(struct nf_bridge_info), GFP_ATOMIC);
 121         if (likely(skb->nf_bridge))
 122                 atomic_set(&(skb->nf_bridge->use), 1);
 123
 124         return skb->nf_bridge;
 125 }
 126
 127 static inline void nf_bridge_save_header(struct sk_buff *skb)
 128 {
 129         int header_size = ETH_HLEN;
 130
 131         if (skb->protocol == htons(ETH_P_8021Q))
 132                 header_size += VLAN_HLEN;
 133
 134         memcpy(skb->nf_bridge->data, skb->data - header_size, header_size);
 135 }
 136
 137 /*
 138  * When forwarding bridge frames, we save a copy of the original
 139  * header before processing.
 140  */
 141 int nf_bridge_copy_header(struct sk_buff *skb)
 142 {
 143         int err;
 144         int header_size = ETH_HLEN;
 145
 146         if (skb->protocol == htons(ETH_P_8021Q))
 147                 header_size += VLAN_HLEN;
 148
 149         err = skb_cow(skb, header_size);
 150         if (err)
 151                 return err;
 152
 153         memcpy(skb->data - header_size, skb->nf_bridge->data, header_size);
 154
 155         if (skb->protocol == htons(ETH_P_8021Q))
 156                 __skb_push(skb, VLAN_HLEN);
 157         return 0;
 158 }
 159
 160 /* PF_BRIDGE/PRE_ROUTING *********************************************/
 161 /* Undo the changes made for ip6tables PREROUTING and continue the
 162  * bridge PRE_ROUTING hook. */
 163 static int br_nf_pre_routing_finish_ipv6(struct sk_buff *skb)
 164 {
 165         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 166
 167         if (nf_bridge->mask & BRNF_PKT_TYPE) {
 168                 skb->pkt_type = PACKET_OTHERHOST;
 169                 nf_bridge->mask ^= BRNF_PKT_TYPE;
 170         }
 171         nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
 172
 173         skb->dst = (struct dst_entry *)&__fake_rtable;
 174         dst_hold(skb->dst);
 175
 176         skb->dev = nf_bridge->physindev;
 177         if (skb->protocol == htons(ETH_P_8021Q)) {
 178                 skb_push(skb, VLAN_HLEN);
 179                 skb->nh.raw -= VLAN_HLEN;
 180         }
 181         NF_HOOK_THRESH(PF_BRIDGE, NF_BR_PRE_ROUTING, skb, skb->dev, NULL,
 182                        br_handle_frame_finish, 1);
 183
 184         return 0;
 185 }
 186
 187 static void __br_dnat_complain(void)
 188 {
 189         static unsigned long last_complaint;
 190
 191         if (jiffies - last_complaint >= 5 * HZ) {
 192                 printk(KERN_WARNING "Performing cross-bridge DNAT requires IP "
 193                        "forwarding to be enabled\n");
 194                 last_complaint = jiffies;
 195         }
 196 }
 197
 198 /* This requires some explaining. If DNAT has taken place,
 199  * we will need to fix up the destination Ethernet address,
 200  * and this is a tricky process.
 201  *
 202  * There are two cases to consider:
 203  * 1. The packet was DNAT'ed to a device in the same bridge
 204  *    port group as it was received on. We can still bridge
 205  *    the packet.
 206  * 2. The packet was DNAT'ed to a different device, either
 207  *    a non-bridged device or another bridge port group.
 208  *    The packet will need to be routed.
 209  *
 210  * The correct way of distinguishing between these two cases is to
 211  * call ip_route_input() and to look at skb->dst->dev, which is
 212  * changed to the destination device if ip_route_input() succeeds.
 213  *
 214  * Let us first consider the case that ip_route_input() succeeds:
 215  *
 216  * If skb->dst->dev equals the logical bridge device the packet
 217  * came in on, we can consider this bridging. We then call
 218  * skb->dst->output() which will make the packet enter br_nf_local_out()
 219  * not much later. In that function it is assured that the iptables
 220  * FORWARD chain is traversed for the packet.
 221  *
 222  * Otherwise, the packet is considered to be routed and we just
 223  * change the destination MAC address so that the packet will
 224  * later be passed up to the IP stack to be routed.
 225  *
 226  * Let us now consider the case that ip_route_input() fails:
 227  *
 228  * After a "echo '0' > /proc/sys/net/ipv4/ip_forward" ip_route_input()
 229  * will fail, while __ip_route_output_key() will return success. The source
 230  * address for __ip_route_output_key() is set to zero, so __ip_route_output_key
 231  * thinks we're handling a locally generated packet and won't care
 232  * if IP forwarding is allowed. We send a warning message to the users's
 233  * log telling her to put IP forwarding on.
 234  *
 235  * ip_route_input() will also fail if there is no route available.
 236  * In that case we just drop the packet.
 237  *
 238  * --Lennert, 20020411
 239  * --Bart, 20020416 (updated)
 240  * --Bart, 20021007 (updated) */
 241 static int br_nf_pre_routing_finish_bridge(struct sk_buff *skb)
 242 {
 243         if (skb->pkt_type == PACKET_OTHERHOST) {
 244                 skb->pkt_type = PACKET_HOST;
 245                 skb->nf_bridge->mask |= BRNF_PKT_TYPE;
 246         }
 247         skb->nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
 248
 249         skb->dev = bridge_parent(skb->dev);
 250         if (!skb->dev)
 251                 kfree_skb(skb);
 252         else {
 253                 if (skb->protocol == htons(ETH_P_8021Q)) {
 254                         skb_pull(skb, VLAN_HLEN);
 255                         skb->nh.raw += VLAN_HLEN;
 256                 }
 257                 skb->dst->output(skb);
 258         }
 259         return 0;
 260 }
 261
 262 static int br_nf_pre_routing_finish(struct sk_buff *skb)
 263 {
 264         struct net_device *dev = skb->dev;
 265         struct iphdr *iph = skb->nh.iph;
 266         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 267
 268         if (nf_bridge->mask & BRNF_PKT_TYPE) {
 269                 skb->pkt_type = PACKET_OTHERHOST;
 270                 nf_bridge->mask ^= BRNF_PKT_TYPE;
 271         }
 272         nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
 273
 274         if (dnat_took_place(skb)) {
 275                 if (ip_route_input(skb, iph->daddr, iph->saddr, iph->tos, dev)) {
 276                         struct rtable *rt;
 277                         struct flowi fl = {
 278                                 .nl_u = {
 279                                         .ip4_u = {
 280                                                  .daddr = iph->daddr,
 281                                                  .saddr = 0,
 282                                                  .tos = RT_TOS(iph->tos) },
 283                                 },
 284                                 .proto = 0,
 285                         };
 286
 287                         if (!ip_route_output_key(&rt, &fl)) {
 288                                 /* - Bridged-and-DNAT'ed traffic doesn't
 289                                  *   require ip_forwarding.
 290                                  * - Deal with redirected traffic. */
 291                                 if (((struct dst_entry *)rt)->dev == dev ||
 292                                     rt->rt_type == RTN_LOCAL) {
 293                                         skb->dst = (struct dst_entry *)rt;
 294                                         goto bridged_dnat;
 295                                 }
 296                                 __br_dnat_complain();
 297                                 dst_release((struct dst_entry *)rt);
 298                         }
 299                         kfree_skb(skb);
 300                         return 0;
 301                 } else {
 302                         if (skb->dst->dev == dev) {
 303 bridged_dnat:
 304                                 /* Tell br_nf_local_out this is a
 305                                  * bridged frame */
 306                                 nf_bridge->mask |= BRNF_BRIDGED_DNAT;
 307                                 skb->dev = nf_bridge->physindev;
 308                                 if (skb->protocol ==
 309                                     htons(ETH_P_8021Q)) {
 310                                         skb_push(skb, VLAN_HLEN);
 311                                         skb->nh.raw -= VLAN_HLEN;
 312                                 }
 313                                 NF_HOOK_THRESH(PF_BRIDGE, NF_BR_PRE_ROUTING,
 314                                                skb, skb->dev, NULL,
 315                                                br_nf_pre_routing_finish_bridge,
 316                                                1);
 317                                 return 0;
 318                         }
 319                         memcpy(eth_hdr(skb)->h_dest, dev->dev_addr, ETH_ALEN);
 320                         skb->pkt_type = PACKET_HOST;
 321                 }
 322         } else {
 323                 skb->dst = (struct dst_entry *)&__fake_rtable;
 324                 dst_hold(skb->dst);
 325         }
 326
 327         skb->dev = nf_bridge->physindev;
 328         if (skb->protocol == htons(ETH_P_8021Q)) {
 329                 skb_push(skb, VLAN_HLEN);
 330                 skb->nh.raw -= VLAN_HLEN;
 331         }
 332         NF_HOOK_THRESH(PF_BRIDGE, NF_BR_PRE_ROUTING, skb, skb->dev, NULL,
 333                        br_handle_frame_finish, 1);
 334
 335         return 0;
 336 }
 337
 338 /* Some common code for IPv4/IPv6 */
 339 static struct net_device *setup_pre_routing(struct sk_buff *skb)
 340 {
 341         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 342
 343         if (skb->pkt_type == PACKET_OTHERHOST) {
 344                 skb->pkt_type = PACKET_HOST;
 345                 nf_bridge->mask |= BRNF_PKT_TYPE;
 346         }
 347
 348         nf_bridge->mask |= BRNF_NF_BRIDGE_PREROUTING;
 349         nf_bridge->physindev = skb->dev;
 350         skb->dev = bridge_parent(skb->dev);
 351
 352         return skb->dev;
 353 }
 354
 355 /* We only check the length. A bridge shouldn't do any hop-by-hop stuff anyway */
 356 static int check_hbh_len(struct sk_buff *skb)
 357 {
 358         unsigned char *raw = (u8 *) (skb->nh.ipv6h + 1);
 359         u32 pkt_len;
 360         int off = raw - skb->nh.raw;
 361         int len = (raw[1] + 1) << 3;
 362
 363         if ((raw + len) - skb->data > skb_headlen(skb))
 364                 goto bad;
 365
 366         off += 2;
 367         len -= 2;
 368
 369         while (len > 0) {
 370                 int optlen = skb->nh.raw[off + 1] + 2;
 371
 372                 switch (skb->nh.raw[off]) {
 373                 case IPV6_TLV_PAD0:
 374                         optlen = 1;
 375                         break;
 376
 377                 case IPV6_TLV_PADN:
 378                         break;
 379
 380                 case IPV6_TLV_JUMBO:
 381                         if (skb->nh.raw[off + 1] != 4 || (off & 3) != 2)
 382                                 goto bad;
 383                         pkt_len = ntohl(*(__be32 *) (skb->nh.raw + off + 2));
 384                         if (pkt_len <= IPV6_MAXPLEN ||
 385                             skb->nh.ipv6h->payload_len)
 386                                 goto bad;
 387                         if (pkt_len > skb->len - sizeof(struct ipv6hdr))
 388                                 goto bad;
 389                         if (pskb_trim_rcsum(skb,
 390                                             pkt_len + sizeof(struct ipv6hdr)))
 391                                 goto bad;
 392                         break;
 393                 default:
 394                         if (optlen > len)
 395                                 goto bad;
 396                         break;
 397                 }
 398                 off += optlen;
 399                 len -= optlen;
 400         }
 401         if (len == 0)
 402                 return 0;
 403 bad:
 404         return -1;
 405
 406 }
 407
 408 /* Replicate the checks that IPv6 does on packet reception and pass the packet
 409  * to ip6tables, which doesn't support NAT, so things are fairly simple. */
 410 static unsigned int br_nf_pre_routing_ipv6(unsigned int hook,
 411                                            struct sk_buff *skb,
 412                                            const struct net_device *in,
 413                                            const struct net_device *out,
 414                                            int (*okfn)(struct sk_buff *))
 415 {
 416         struct ipv6hdr *hdr;
 417         u32 pkt_len;
 418
 419         if (skb->len < sizeof(struct ipv6hdr))
 420                 goto inhdr_error;
 421
 422         if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
 423                 goto inhdr_error;
 424
 425         hdr = skb->nh.ipv6h;
 426
 427         if (hdr->version != 6)
 428                 goto inhdr_error;
 429
 430         pkt_len = ntohs(hdr->payload_len);
 431
 432         if (pkt_len || hdr->nexthdr != NEXTHDR_HOP) {
 433                 if (pkt_len + sizeof(struct ipv6hdr) > skb->len)
 434                         goto inhdr_error;
 435                 if (pskb_trim_rcsum(skb, pkt_len + sizeof(struct ipv6hdr)))
 436                         goto inhdr_error;
 437         }
 438         if (hdr->nexthdr == NEXTHDR_HOP && check_hbh_len(skb))
 439                 goto inhdr_error;
 440
 441         nf_bridge_put(skb->nf_bridge);
 442         if (!nf_bridge_alloc(skb))
 443                 return NF_DROP;
 444         if (!setup_pre_routing(skb))
 445                 return NF_DROP;
 446
 447         NF_HOOK(PF_INET6, NF_IP6_PRE_ROUTING, skb, skb->dev, NULL,
 448                 br_nf_pre_routing_finish_ipv6);
 449
 450         return NF_STOLEN;
 451
 452 inhdr_error:
 453         return NF_DROP;
 454 }
 455
 456 /* Direct IPv6 traffic to br_nf_pre_routing_ipv6.
 457  * Replicate the checks that IPv4 does on packet reception.
 458  * Set skb->dev to the bridge device (i.e. parent of the
 459  * receiving device) to make netfilter happy, the REDIRECT
 460  * target in particular.  Save the original destination IP
 461  * address to be able to detect DNAT afterwards. */
 462 static unsigned int br_nf_pre_routing(unsigned int hook, struct sk_buff **pskb,
 463                                       const struct net_device *in,
 464                                       const struct net_device *out,
 465                                       int (*okfn)(struct sk_buff *))
 466 {
 467         struct iphdr *iph;
 468         __u32 len;
 469         struct sk_buff *skb = *pskb;
 470
 471         if (skb->protocol == htons(ETH_P_IPV6) || IS_VLAN_IPV6(skb)) {
 472 #ifdef CONFIG_SYSCTL
 473                 if (!brnf_call_ip6tables)
 474                         return NF_ACCEPT;
 475 #endif
 476                 if ((skb = skb_share_check(*pskb, GFP_ATOMIC)) == NULL)
 477                         goto out;
 478
 479                 if (skb->protocol == htons(ETH_P_8021Q)) {
 480                         skb_pull_rcsum(skb, VLAN_HLEN);
 481                         skb->nh.raw += VLAN_HLEN;
 482                 }
 483                 return br_nf_pre_routing_ipv6(hook, skb, in, out, okfn);
 484         }
 485 #ifdef CONFIG_SYSCTL
 486         if (!brnf_call_iptables)
 487                 return NF_ACCEPT;
 488 #endif
 489
 490         if (skb->protocol != htons(ETH_P_IP) && !IS_VLAN_IP(skb))
 491                 return NF_ACCEPT;
 492
 493         if ((skb = skb_share_check(*pskb, GFP_ATOMIC)) == NULL)
 494                 goto out;
 495
 496         if (skb->protocol == htons(ETH_P_8021Q)) {
 497                 skb_pull_rcsum(skb, VLAN_HLEN);
 498                 skb->nh.raw += VLAN_HLEN;
 499         }
 500
 501         if (!pskb_may_pull(skb, sizeof(struct iphdr)))
 502                 goto inhdr_error;
 503
 504         iph = skb->nh.iph;
 505         if (iph->ihl < 5 || iph->version != 4)
 506                 goto inhdr_error;
 507
 508         if (!pskb_may_pull(skb, 4 * iph->ihl))
 509                 goto inhdr_error;
 510
 511         iph = skb->nh.iph;
 512         if (ip_fast_csum((__u8 *) iph, iph->ihl) != 0)
 513                 goto inhdr_error;
 514
 515         len = ntohs(iph->tot_len);
 516         if (skb->len < len || len < 4 * iph->ihl)
 517                 goto inhdr_error;
 518
 519         pskb_trim_rcsum(skb, len);
 520
 521         nf_bridge_put(skb->nf_bridge);
 522         if (!nf_bridge_alloc(skb))
 523                 return NF_DROP;
 524         if (!setup_pre_routing(skb))
 525                 return NF_DROP;
 526         store_orig_dstaddr(skb);
 527
 528         NF_HOOK(PF_INET, NF_IP_PRE_ROUTING, skb, skb->dev, NULL,
 529                 br_nf_pre_routing_finish);
 530
 531         return NF_STOLEN;
 532
 533 inhdr_error:
 534 //      IP_INC_STATS_BH(IpInHdrErrors);
 535 out:
 536         return NF_DROP;
 537 }
 538
 539
 540 /* PF_BRIDGE/LOCAL_IN ************************************************/
 541 /* The packet is locally destined, which requires a real
 542  * dst_entry, so detach the fake one.  On the way up, the
 543  * packet would pass through PRE_ROUTING again (which already
 544  * took place when the packet entered the bridge), but we
 545  * register an IPv4 PRE_ROUTING 'sabotage' hook that will
 546  * prevent this from happening. */
 547 static unsigned int br_nf_local_in(unsigned int hook, struct sk_buff **pskb,
 548                                    const struct net_device *in,
 549                                    const struct net_device *out,
 550                                    int (*okfn)(struct sk_buff *))
 551 {
 552         struct sk_buff *skb = *pskb;
 553
 554         if (skb->dst == (struct dst_entry *)&__fake_rtable) {
 555                 dst_release(skb->dst);
 556                 skb->dst = NULL;
 557         }
 558
 559         return NF_ACCEPT;
 560 }
 561
 562 /* PF_BRIDGE/FORWARD *************************************************/
 563 static int br_nf_forward_finish(struct sk_buff *skb)
 564 {
 565         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 566         struct net_device *in;
 567
 568         if (skb->protocol != htons(ETH_P_ARP) && !IS_VLAN_ARP(skb)) {
 569                 in = nf_bridge->physindev;
 570                 if (nf_bridge->mask & BRNF_PKT_TYPE) {
 571                         skb->pkt_type = PACKET_OTHERHOST;
 572                         nf_bridge->mask ^= BRNF_PKT_TYPE;
 573                 }
 574         } else {
 575                 in = *((struct net_device **)(skb->cb));
 576         }
 577         if (skb->protocol == htons(ETH_P_8021Q)) {
 578                 skb_push(skb, VLAN_HLEN);
 579                 skb->nh.raw -= VLAN_HLEN;
 580         }
 581         NF_HOOK_THRESH(PF_BRIDGE, NF_BR_FORWARD, skb, in,
 582                        skb->dev, br_forward_finish, 1);
 583         return 0;
 584 }
 585
 586 /* This is the 'purely bridged' case.  For IP, we pass the packet to
 587  * netfilter with indev and outdev set to the bridge device,
 588  * but we are still able to filter on the 'real' indev/outdev
 589  * because of the physdev module. For ARP, indev and outdev are the
 590  * bridge ports. */
 591 static unsigned int br_nf_forward_ip(unsigned int hook, struct sk_buff **pskb,
 592                                      const struct net_device *in,
 593                                      const struct net_device *out,
 594                                      int (*okfn)(struct sk_buff *))
 595 {
 596         struct sk_buff *skb = *pskb;
 597         struct nf_bridge_info *nf_bridge;
 598         struct net_device *parent;
 599         int pf;
 600
 601         if (!skb->nf_bridge)
 602                 return NF_ACCEPT;
 603
 604         parent = bridge_parent(out);
 605         if (!parent)
 606                 return NF_DROP;
 607
 608         if (skb->protocol == htons(ETH_P_IP) || IS_VLAN_IP(skb))
 609                 pf = PF_INET;
 610         else
 611                 pf = PF_INET6;
 612
 613         if (skb->protocol == htons(ETH_P_8021Q)) {
 614                 skb_pull(*pskb, VLAN_HLEN);
 615                 (*pskb)->nh.raw += VLAN_HLEN;
 616         }
 617
 618         nf_bridge = skb->nf_bridge;
 619         if (skb->pkt_type == PACKET_OTHERHOST) {
 620                 skb->pkt_type = PACKET_HOST;
 621                 nf_bridge->mask |= BRNF_PKT_TYPE;
 622         }
 623
 624         /* The physdev module checks on this */
 625         nf_bridge->mask |= BRNF_BRIDGED;
 626         nf_bridge->physoutdev = skb->dev;
 627
 628         NF_HOOK(pf, NF_IP_FORWARD, skb, bridge_parent(in), parent,
 629                 br_nf_forward_finish);
 630
 631         return NF_STOLEN;
 632 }
 633
 634 static unsigned int br_nf_forward_arp(unsigned int hook, struct sk_buff **pskb,
 635                                       const struct net_device *in,
 636                                       const struct net_device *out,
 637                                       int (*okfn)(struct sk_buff *))
 638 {
 639         struct sk_buff *skb = *pskb;
 640         struct net_device **d = (struct net_device **)(skb->cb);
 641
 642 #ifdef CONFIG_SYSCTL
 643         if (!brnf_call_arptables)
 644                 return NF_ACCEPT;
 645 #endif
 646
 647         if (skb->protocol != htons(ETH_P_ARP)) {
 648                 if (!IS_VLAN_ARP(skb))
 649                         return NF_ACCEPT;
 650                 skb_pull(*pskb, VLAN_HLEN);
 651                 (*pskb)->nh.raw += VLAN_HLEN;
 652         }
 653
 654         if (skb->nh.arph->ar_pln != 4) {
 655                 if (IS_VLAN_ARP(skb)) {
 656                         skb_push(*pskb, VLAN_HLEN);
 657                         (*pskb)->nh.raw -= VLAN_HLEN;
 658                 }
 659                 return NF_ACCEPT;
 660         }
 661         *d = (struct net_device *)in;
 662         NF_HOOK(NF_ARP, NF_ARP_FORWARD, skb, (struct net_device *)in,
 663                 (struct net_device *)out, br_nf_forward_finish);
 664
 665         return NF_STOLEN;
 666 }
 667
 668 /* PF_BRIDGE/LOCAL_OUT ***********************************************/
 669 static int br_nf_local_out_finish(struct sk_buff *skb)
 670 {
 671         if (skb->protocol == htons(ETH_P_8021Q)) {
 672                 skb_push(skb, VLAN_HLEN);
 673                 skb->nh.raw -= VLAN_HLEN;
 674         }
 675
 676         NF_HOOK_THRESH(PF_BRIDGE, NF_BR_LOCAL_OUT, skb, NULL, skb->dev,
 677                        br_forward_finish, NF_BR_PRI_FIRST + 1);
 678
 679         return 0;
 680 }
 681
 682 /* This function sees both locally originated IP packets and forwarded
 683  * IP packets (in both cases the destination device is a bridge
 684  * device). It also sees bridged-and-DNAT'ed packets.
 685  * To be able to filter on the physical bridge devices (with the physdev
 686  * module), we steal packets destined to a bridge device away from the
 687  * PF_INET/FORWARD and PF_INET/OUTPUT hook functions, and give them back later,
 688  * when we have determined the real output device. This is done in here.
 689  *
 690  * If (nf_bridge->mask & BRNF_BRIDGED_DNAT) then the packet is bridged
 691  * and we fake the PF_BRIDGE/FORWARD hook. The function br_nf_forward()
 692  * will then fake the PF_INET/FORWARD hook. br_nf_local_out() has priority
 693  * NF_BR_PRI_FIRST, so no relevant PF_BRIDGE/INPUT functions have been nor
 694  * will be executed.
 695  * Otherwise, if nf_bridge->physindev is NULL, the bridge-nf code never touched
 696  * this packet before, and so the packet was locally originated. We fake
 697  * the PF_INET/LOCAL_OUT hook.
 698  * Finally, if nf_bridge->physindev isn't NULL, then the packet was IP routed,
 699  * so we fake the PF_INET/FORWARD hook. ip_sabotage_out() makes sure
 700  * even routed packets that didn't arrive on a bridge interface have their
 701  * nf_bridge->physindev set. */
 702 static unsigned int br_nf_local_out(unsigned int hook, struct sk_buff **pskb,
 703                                     const struct net_device *in,
 704                                     const struct net_device *out,
 705                                     int (*okfn)(struct sk_buff *))
 706 {
 707         struct net_device *realindev, *realoutdev;
 708         struct sk_buff *skb = *pskb;
 709         struct nf_bridge_info *nf_bridge;
 710         int pf;
 711
 712         if (!skb->nf_bridge)
 713                 return NF_ACCEPT;
 714
 715         if (skb->protocol == htons(ETH_P_IP) || IS_VLAN_IP(skb))
 716                 pf = PF_INET;
 717         else
 718                 pf = PF_INET6;
 719
 720         nf_bridge = skb->nf_bridge;
 721         nf_bridge->physoutdev = skb->dev;
 722         realindev = nf_bridge->physindev;
 723
 724         /* Bridged, take PF_BRIDGE/FORWARD.
 725          * (see big note in front of br_nf_pre_routing_finish) */
 726         if (nf_bridge->mask & BRNF_BRIDGED_DNAT) {
 727                 if (nf_bridge->mask & BRNF_PKT_TYPE) {
 728                         skb->pkt_type = PACKET_OTHERHOST;
 729                         nf_bridge->mask ^= BRNF_PKT_TYPE;
 730                 }
 731                 if (skb->protocol == htons(ETH_P_8021Q)) {
 732                         skb_push(skb, VLAN_HLEN);
 733                         skb->nh.raw -= VLAN_HLEN;
 734                 }
 735
 736                 NF_HOOK(PF_BRIDGE, NF_BR_FORWARD, skb, realindev,
 737                         skb->dev, br_forward_finish);
 738                 goto out;
 739         }
 740         realoutdev = bridge_parent(skb->dev);
 741         if (!realoutdev)
 742                 return NF_DROP;
 743
 744 #if defined(CONFIG_VLAN_8021Q) || defined(CONFIG_VLAN_8021Q_MODULE)
 745         /* iptables should match -o br0.x */
 746         if (nf_bridge->netoutdev)
 747                 realoutdev = nf_bridge->netoutdev;
 748 #endif
 749         if (skb->protocol == htons(ETH_P_8021Q)) {
 750                 skb_pull(skb, VLAN_HLEN);
 751                 (*pskb)->nh.raw += VLAN_HLEN;
 752         }
 753         /* IP forwarded traffic has a physindev, locally
 754          * generated traffic hasn't. */
 755         if (realindev != NULL) {
 756                 if (!(nf_bridge->mask & BRNF_DONT_TAKE_PARENT)) {
 757                         struct net_device *parent = bridge_parent(realindev);
 758                         if (parent)
 759                                 realindev = parent;
 760                 }
 761
 762                 NF_HOOK_THRESH(pf, NF_IP_FORWARD, skb, realindev,
 763                                realoutdev, br_nf_local_out_finish,
 764                                NF_IP_PRI_BRIDGE_SABOTAGE_FORWARD + 1);
 765         } else {
 766                 NF_HOOK_THRESH(pf, NF_IP_LOCAL_OUT, skb, realindev,
 767                                realoutdev, br_nf_local_out_finish,
 768                                NF_IP_PRI_BRIDGE_SABOTAGE_LOCAL_OUT + 1);
 769         }
 770
 771 out:
 772         return NF_STOLEN;
 773 }
 774
 775 static int br_nf_dev_queue_xmit(struct sk_buff *skb)
 776 {
 777         if (skb->protocol == htons(ETH_P_IP) &&
 778             skb->len > skb->dev->mtu &&
 779             !skb_is_gso(skb))
 780                 return ip_fragment(skb, br_dev_queue_push_xmit);
 781         else
 782                 return br_dev_queue_push_xmit(skb);
 783 }
 784
 785 /* PF_BRIDGE/POST_ROUTING ********************************************/
 786 static unsigned int br_nf_post_routing(unsigned int hook, struct sk_buff **pskb,
 787                                        const struct net_device *in,
 788                                        const struct net_device *out,
 789                                        int (*okfn)(struct sk_buff *))
 790 {
 791         struct sk_buff *skb = *pskb;
 792         struct nf_bridge_info *nf_bridge = (*pskb)->nf_bridge;
 793         struct net_device *realoutdev = bridge_parent(skb->dev);
 794         int pf;
 795
 796 #ifdef CONFIG_NETFILTER_DEBUG
 797         /* Be very paranoid. This probably won't happen anymore, but let's
 798          * keep the check just to be sure... */
 799         if (skb->mac.raw < skb->head || skb->mac.raw + ETH_HLEN > skb->data) {
 800                 printk(KERN_CRIT "br_netfilter: Argh!! br_nf_post_routing: "
 801                        "bad mac.raw pointer.\n");
 802                 goto print_error;
 803         }
 804 #endif
 805
 806         if (!nf_bridge)
 807                 return NF_ACCEPT;
 808
 809         if (!realoutdev)
 810                 return NF_DROP;
 811
 812         if (skb->protocol == htons(ETH_P_IP) || IS_VLAN_IP(skb))
 813                 pf = PF_INET;
 814         else
 815                 pf = PF_INET6;
 816
 817 #ifdef CONFIG_NETFILTER_DEBUG
 818         if (skb->dst == NULL) {
 819                 printk(KERN_INFO "br_netfilter post_routing: skb->dst == NULL\n");
 820                 goto print_error;
 821         }
 822 #endif
 823
 824         /* We assume any code from br_dev_queue_push_xmit onwards doesn't care
 825          * about the value of skb->pkt_type. */
 826         if (skb->pkt_type == PACKET_OTHERHOST) {
 827                 skb->pkt_type = PACKET_HOST;
 828                 nf_bridge->mask |= BRNF_PKT_TYPE;
 829         }
 830
 831         if (skb->protocol == htons(ETH_P_8021Q)) {
 832                 skb_pull(skb, VLAN_HLEN);
 833                 skb->nh.raw += VLAN_HLEN;
 834         }
 835
 836         nf_bridge_save_header(skb);
 837
 838 #if defined(CONFIG_VLAN_8021Q) || defined(CONFIG_VLAN_8021Q_MODULE)
 839         if (nf_bridge->netoutdev)
 840                 realoutdev = nf_bridge->netoutdev;
 841 #endif
 842         NF_HOOK(pf, NF_IP_POST_ROUTING, skb, NULL, realoutdev,
 843                 br_nf_dev_queue_xmit);
 844
 845         return NF_STOLEN;
 846
 847 #ifdef CONFIG_NETFILTER_DEBUG
 848 print_error:
 849         if (skb->dev != NULL) {
 850                 printk("[%s]", skb->dev->name);
 851                 if (realoutdev)
 852                         printk("[%s]", realoutdev->name);
 853         }
 854         printk(" head:%p, raw:%p, data:%p\n", skb->head, skb->mac.raw,
 855                skb->data);
 856         dump_stack();
 857         return NF_ACCEPT;
 858 #endif
 859 }
 860
 861 /* IP/SABOTAGE *****************************************************/
 862 /* Don't hand locally destined packets to PF_INET(6)/PRE_ROUTING
 863  * for the second time. */
 864 static unsigned int ip_sabotage_in(unsigned int hook, struct sk_buff **pskb,
 865                                    const struct net_device *in,
 866                                    const struct net_device *out,
 867                                    int (*okfn)(struct sk_buff *))
 868 {
 869         if ((*pskb)->nf_bridge &&
 870             !((*pskb)->nf_bridge->mask & BRNF_NF_BRIDGE_PREROUTING)) {
 871                 return NF_STOP;
 872         }
 873
 874         return NF_ACCEPT;
 875 }
 876
 877 /* Postpone execution of PF_INET(6)/FORWARD, PF_INET(6)/LOCAL_OUT
 878  * and PF_INET(6)/POST_ROUTING until we have done the forwarding
 879  * decision in the bridge code and have determined nf_bridge->physoutdev. */
 880 static unsigned int ip_sabotage_out(unsigned int hook, struct sk_buff **pskb,
 881                                     const struct net_device *in,
 882                                     const struct net_device *out,
 883                                     int (*okfn)(struct sk_buff *))
 884 {
 885         struct sk_buff *skb = *pskb;
 886
 887         if ((out->hard_start_xmit == br_dev_xmit &&
 888              okfn != br_nf_forward_finish &&
 889              okfn != br_nf_local_out_finish && okfn != br_nf_dev_queue_xmit)
 890 #if defined(CONFIG_VLAN_8021Q) || defined(CONFIG_VLAN_8021Q_MODULE)
 891             || ((out->priv_flags & IFF_802_1Q_VLAN) &&
 892                 VLAN_DEV_INFO(out)->real_dev->hard_start_xmit == br_dev_xmit)
 893 #endif
 894             ) {
 895                 struct nf_bridge_info *nf_bridge;
 896
 897                 if (!skb->nf_bridge) {
 898 #ifdef CONFIG_SYSCTL
 899                         /* This code is executed while in the IP(v6) stack,
 900                            the version should be 4 or 6. We can't use
 901                            skb->protocol because that isn't set on
 902                            PF_INET(6)/LOCAL_OUT. */
 903                         struct iphdr *ip = skb->nh.iph;
 904
 905                         if (ip->version == 4 && !brnf_call_iptables)
 906                                 return NF_ACCEPT;
 907                         else if (ip->version == 6 && !brnf_call_ip6tables)
 908                                 return NF_ACCEPT;
 909                         else if (!brnf_deferred_hooks)
 910                                 return NF_ACCEPT;
 911 #endif
 912                         if (hook == NF_IP_POST_ROUTING)
 913                                 return NF_ACCEPT;
 914                         if (!nf_bridge_alloc(skb))
 915                                 return NF_DROP;
 916                 }
 917
 918                 nf_bridge = skb->nf_bridge;
 919
 920                 /* This frame will arrive on PF_BRIDGE/LOCAL_OUT and we
 921                  * will need the indev then. For a brouter, the real indev
 922                  * can be a bridge port, so we make sure br_nf_local_out()
 923                  * doesn't use the bridge parent of the indev by using
 924                  * the BRNF_DONT_TAKE_PARENT mask. */
 925                 if (hook == NF_IP_FORWARD && nf_bridge->physindev == NULL) {
 926                         nf_bridge->mask |= BRNF_DONT_TAKE_PARENT;
 927                         nf_bridge->physindev = (struct net_device *)in;
 928                 }
 929 #if defined(CONFIG_VLAN_8021Q) || defined(CONFIG_VLAN_8021Q_MODULE)
 930                 /* the iptables outdev is br0.x, not br0 */
 931                 if (out->priv_flags & IFF_802_1Q_VLAN)
 932                         nf_bridge->netoutdev = (struct net_device *)out;
 933 #endif
 934                 return NF_STOP;
 935         }
 936
 937         return NF_ACCEPT;
 938 }
 939
 940 /* For br_nf_local_out we need (prio = NF_BR_PRI_FIRST), to insure that innocent
 941  * PF_BRIDGE/NF_BR_LOCAL_OUT functions don't get bridged traffic as input.
 942  * For br_nf_post_routing, we need (prio = NF_BR_PRI_LAST), because
 943  * ip_refrag() can return NF_STOLEN. */
 944 static struct nf_hook_ops br_nf_ops[] = {
 945         { .hook = br_nf_pre_routing,
 946           .owner = THIS_MODULE,
 947           .pf = PF_BRIDGE,
 948           .hooknum = NF_BR_PRE_ROUTING,
 949           .priority = NF_BR_PRI_BRNF, },
 950         { .hook = br_nf_local_in,
 951           .owner = THIS_MODULE,
 952           .pf = PF_BRIDGE,
 953           .hooknum = NF_BR_LOCAL_IN,
 954           .priority = NF_BR_PRI_BRNF, },
 955         { .hook = br_nf_forward_ip,
 956           .owner = THIS_MODULE,
 957           .pf = PF_BRIDGE,
 958           .hooknum = NF_BR_FORWARD,
 959           .priority = NF_BR_PRI_BRNF - 1, },
 960         { .hook = br_nf_forward_arp,
 961           .owner = THIS_MODULE,
 962           .pf = PF_BRIDGE,
 963           .hooknum = NF_BR_FORWARD,
 964           .priority = NF_BR_PRI_BRNF, },
 965         { .hook = br_nf_local_out,
 966           .owner = THIS_MODULE,
 967           .pf = PF_BRIDGE,
 968           .hooknum = NF_BR_LOCAL_OUT,
 969           .priority = NF_BR_PRI_FIRST, },
 970         { .hook = br_nf_post_routing,
 971           .owner = THIS_MODULE,
 972           .pf = PF_BRIDGE,
 973           .hooknum = NF_BR_POST_ROUTING,
 974           .priority = NF_BR_PRI_LAST, },
 975         { .hook = ip_sabotage_in,
 976           .owner = THIS_MODULE,
 977           .pf = PF_INET,
 978           .hooknum = NF_IP_PRE_ROUTING,
 979           .priority = NF_IP_PRI_FIRST, },
 980         { .hook = ip_sabotage_in,
 981           .owner = THIS_MODULE,
 982           .pf = PF_INET6,
 983           .hooknum = NF_IP6_PRE_ROUTING,
 984           .priority = NF_IP6_PRI_FIRST, },
 985         { .hook = ip_sabotage_out,
 986           .owner = THIS_MODULE,
 987           .pf = PF_INET,
 988           .hooknum = NF_IP_FORWARD,
 989           .priority = NF_IP_PRI_BRIDGE_SABOTAGE_FORWARD, },
 990         { .hook = ip_sabotage_out,
 991           .owner = THIS_MODULE,
 992           .pf = PF_INET6,
 993           .hooknum = NF_IP6_FORWARD,
 994           .priority = NF_IP6_PRI_BRIDGE_SABOTAGE_FORWARD, },
 995         { .hook = ip_sabotage_out,
 996           .owner = THIS_MODULE,
 997           .pf = PF_INET,
 998           .hooknum = NF_IP_LOCAL_OUT,
 999           .priority = NF_IP_PRI_BRIDGE_SABOTAGE_LOCAL_OUT, },
1000         { .hook = ip_sabotage_out,
1001           .owner = THIS_MODULE,
1002           .pf = PF_INET6,
1003           .hooknum = NF_IP6_LOCAL_OUT,
1004           .priority = NF_IP6_PRI_BRIDGE_SABOTAGE_LOCAL_OUT, },
1005         { .hook = ip_sabotage_out,
1006           .owner = THIS_MODULE,
1007           .pf = PF_INET,
1008           .hooknum = NF_IP_POST_ROUTING,
1009           .priority = NF_IP_PRI_FIRST, },
1010         { .hook = ip_sabotage_out,
1011           .owner = THIS_MODULE,
1012           .pf = PF_INET6,
1013           .hooknum = NF_IP6_POST_ROUTING,
1014           .priority = NF_IP6_PRI_FIRST, },
1015 };
1016
1017 #ifdef CONFIG_SYSCTL
1018 static
1019 int brnf_sysctl_call_tables(ctl_table * ctl, int write, struct file *filp,
1020                             void __user * buffer, size_t * lenp, loff_t * ppos)
1021 {
1022         int ret;
1023
1024         ret = proc_dointvec(ctl, write, filp, buffer, lenp, ppos);
1025
1026         if (write && *(int *)(ctl->data))
1027                 *(int *)(ctl->data) = 1;
1028         return ret;
1029 }
1030
1031 static ctl_table brnf_table[] = {
1032         {
1033                 .ctl_name       = NET_BRIDGE_NF_CALL_ARPTABLES,
1034                 .procname       = "bridge-nf-call-arptables",
1035                 .data           = &brnf_call_arptables,
1036                 .maxlen         = sizeof(int),
1037                 .mode           = 0644,
1038                 .proc_handler   = &brnf_sysctl_call_tables,
1039         },
1040         {
1041                 .ctl_name       = NET_BRIDGE_NF_CALL_IPTABLES,
1042                 .procname       = "bridge-nf-call-iptables",
1043                 .data           = &brnf_call_iptables,
1044                 .maxlen         = sizeof(int),
1045                 .mode           = 0644,
1046                 .proc_handler   = &brnf_sysctl_call_tables,
1047         },
1048         {
1049                 .ctl_name       = NET_BRIDGE_NF_CALL_IP6TABLES,
1050                 .procname       = "bridge-nf-call-ip6tables",
1051                 .data           = &brnf_call_ip6tables,
1052                 .maxlen         = sizeof(int),
1053                 .mode           = 0644,
1054                 .proc_handler   = &brnf_sysctl_call_tables,
1055         },
1056         {
1057                 .ctl_name       = NET_BRIDGE_NF_FILTER_VLAN_TAGGED,
1058                 .procname       = "bridge-nf-filter-vlan-tagged",
1059                 .data           = &brnf_filter_vlan_tagged,
1060                 .maxlen         = sizeof(int),
1061                 .mode           = 0644,
1062                 .proc_handler   = &brnf_sysctl_call_tables,
1063         },
1064         { .ctl_name = 0 }
1065 };
1066
1067 static ctl_table brnf_bridge_table[] = {
1068         {
1069                 .ctl_name       = NET_BRIDGE,
1070                 .procname       = "bridge",
1071                 .mode           = 0555,
1072                 .child          = brnf_table,
1073         },
1074         { .ctl_name = 0 }
1075 };
1076
1077 static ctl_table brnf_net_table[] = {
1078         {
1079                 .ctl_name       = CTL_NET,
1080                 .procname       = "net",
1081                 .mode           = 0555,
1082                 .child          = brnf_bridge_table,
1083         },
1084         { .ctl_name = 0 }
1085 };
1086 #endif
1087
1088 int br_netfilter_init(void)
1089 {
1090         int i;
1091
1092         for (i = 0; i < ARRAY_SIZE(br_nf_ops); i++) {
1093                 int ret;
1094
1095                 if ((ret = nf_register_hook(&br_nf_ops[i])) >= 0)
1096                         continue;
1097
1098                 while (i--)
1099                         nf_unregister_hook(&br_nf_ops[i]);
1100
1101                 return ret;
1102         }
1103
1104 #ifdef CONFIG_SYSCTL
1105         brnf_sysctl_header = register_sysctl_table(brnf_net_table, 0);
1106         if (brnf_sysctl_header == NULL) {
1107                 printk(KERN_WARNING
1108                        "br_netfilter: can't register to sysctl.\n");
1109                 for (i = 0; i < ARRAY_SIZE(br_nf_ops); i++)
1110                         nf_unregister_hook(&br_nf_ops[i]);
1111                 return -EFAULT;
1112         }
1113 #endif
1114
1115         printk(KERN_NOTICE "Bridge firewalling registered\n");
1116
1117         return 0;
1118 }
1119
1120 void br_netfilter_fini(void)
1121 {
1122         int i;
1123
1124         for (i = ARRAY_SIZE(br_nf_ops) - 1; i >= 0; i--)
1125                 nf_unregister_hook(&br_nf_ops[i]);
1126 #ifdef CONFIG_SYSCTL
1127         unregister_sysctl_table(brnf_sysctl_header);
1128 #endif
1129 }