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/if_pppox.h>
  33 #include <linux/ppp_defs.h>
  34 #include <linux/netfilter_bridge.h>
  35 #include <linux/netfilter_ipv4.h>
  36 #include <linux/netfilter_ipv6.h>
  37 #include <linux/netfilter_arp.h>
  38 #include <linux/in_route.h>
  39 #include <linux/inetdevice.h>
  40
  41 #include <net/ip.h>
  42 #include <net/ipv6.h>
  43 #include <net/route.h>
  44
  45 #include <asm/uaccess.h>
  46 #include "br_private.h"
  47 #ifdef CONFIG_SYSCTL
  48 #include <linux/sysctl.h>
  49 #endif
  50
  51 #define skb_origaddr(skb)        (((struct bridge_skb_cb *) \
  52                                  (skb->nf_bridge->data))->daddr.ipv4)
  53 #define store_orig_dstaddr(skb)  (skb_origaddr(skb) = ip_hdr(skb)->daddr)
  54 #define dnat_took_place(skb)     (skb_origaddr(skb) != ip_hdr(skb)->daddr)
  55
  56 #ifdef CONFIG_SYSCTL
  57 static struct ctl_table_header *brnf_sysctl_header;
  58 static int brnf_call_iptables __read_mostly = 1;
  59 static int brnf_call_ip6tables __read_mostly = 1;
  60 static int brnf_call_arptables __read_mostly = 1;
  61 static int brnf_filter_vlan_tagged __read_mostly = 1;
  62 static int brnf_filter_pppoe_tagged __read_mostly = 1;
  63 #else
  64 #define brnf_filter_vlan_tagged 1
  65 #define brnf_filter_pppoe_tagged 1
  66 #endif
  67
  68 static inline __be16 vlan_proto(const struct sk_buff *skb)
  69 {
  70         return vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
  71 }
  72
  73 #define IS_VLAN_IP(skb) \
  74         (skb->protocol == htons(ETH_P_8021Q) && \
  75          vlan_proto(skb) == htons(ETH_P_IP) &&  \
  76          brnf_filter_vlan_tagged)
  77
  78 #define IS_VLAN_IPV6(skb) \
  79         (skb->protocol == htons(ETH_P_8021Q) && \
  80          vlan_proto(skb) == htons(ETH_P_IPV6) &&\
  81          brnf_filter_vlan_tagged)
  82
  83 #define IS_VLAN_ARP(skb) \
  84         (skb->protocol == htons(ETH_P_8021Q) && \
  85          vlan_proto(skb) == htons(ETH_P_ARP) && \
  86          brnf_filter_vlan_tagged)
  87
  88 static inline __be16 pppoe_proto(const struct sk_buff *skb)
  89 {
  90         return *((__be16 *)(skb_mac_header(skb) + ETH_HLEN +
  91                             sizeof(struct pppoe_hdr)));
  92 }
  93
  94 #define IS_PPPOE_IP(skb) \
  95         (skb->protocol == htons(ETH_P_PPP_SES) && \
  96          pppoe_proto(skb) == htons(PPP_IP) && \
  97          brnf_filter_pppoe_tagged)
  98
  99 #define IS_PPPOE_IPV6(skb) \
 100         (skb->protocol == htons(ETH_P_PPP_SES) && \
 101          pppoe_proto(skb) == htons(PPP_IPV6) && \
 102          brnf_filter_pppoe_tagged)
 103
 104 /* We need these fake structures to make netfilter happy --
 105  * lots of places assume that skb->dst != NULL, which isn't
 106  * all that unreasonable.
 107  *
 108  * Currently, we fill in the PMTU entry because netfilter
 109  * refragmentation needs it, and the rt_flags entry because
 110  * ipt_REJECT needs it.  Future netfilter modules might
 111  * require us to fill additional fields. */
 112 static struct net_device __fake_net_device = {
 113         .hard_header_len        = ETH_HLEN,
 114         .nd_net                 = &init_net,
 115 };
 116
 117 static struct rtable __fake_rtable = {
 118         .u = {
 119                 .dst = {
 120                         .__refcnt               = ATOMIC_INIT(1),
 121                         .dev                    = &__fake_net_device,
 122                         .path                   = &__fake_rtable.u.dst,
 123                         .metrics                = {[RTAX_MTU - 1] = 1500},
 124                         .flags                  = DST_NOXFRM,
 125                 }
 126         },
 127         .rt_flags       = 0,
 128 };
 129
 130 static inline struct net_device *bridge_parent(const struct net_device *dev)
 131 {
 132         struct net_bridge_port *port = rcu_dereference(dev->br_port);
 133
 134         return port ? port->br->dev : NULL;
 135 }
 136
 137 static inline struct nf_bridge_info *nf_bridge_alloc(struct sk_buff *skb)
 138 {
 139         skb->nf_bridge = kzalloc(sizeof(struct nf_bridge_info), GFP_ATOMIC);
 140         if (likely(skb->nf_bridge))
 141                 atomic_set(&(skb->nf_bridge->use), 1);
 142
 143         return skb->nf_bridge;
 144 }
 145
 146 static inline struct nf_bridge_info *nf_bridge_unshare(struct sk_buff *skb)
 147 {
 148         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 149
 150         if (atomic_read(&nf_bridge->use) > 1) {
 151                 struct nf_bridge_info *tmp = nf_bridge_alloc(skb);
 152
 153                 if (tmp) {
 154                         memcpy(tmp, nf_bridge, sizeof(struct nf_bridge_info));
 155                         atomic_set(&tmp->use, 1);
 156                         nf_bridge_put(nf_bridge);
 157                 }
 158                 nf_bridge = tmp;
 159         }
 160         return nf_bridge;
 161 }
 162
 163 static inline void nf_bridge_push_encap_header(struct sk_buff *skb)
 164 {
 165         unsigned int len = nf_bridge_encap_header_len(skb);
 166
 167         skb_push(skb, len);
 168         skb->network_header -= len;
 169 }
 170
 171 static inline void nf_bridge_pull_encap_header(struct sk_buff *skb)
 172 {
 173         unsigned int len = nf_bridge_encap_header_len(skb);
 174
 175         skb_pull(skb, len);
 176         skb->network_header += len;
 177 }
 178
 179 static inline void nf_bridge_pull_encap_header_rcsum(struct sk_buff *skb)
 180 {
 181         unsigned int len = nf_bridge_encap_header_len(skb);
 182
 183         skb_pull_rcsum(skb, len);
 184         skb->network_header += len;
 185 }
 186
 187 static inline void nf_bridge_save_header(struct sk_buff *skb)
 188 {
 189         int header_size = ETH_HLEN + nf_bridge_encap_header_len(skb);
 190
 191         skb_copy_from_linear_data_offset(skb, -header_size,
 192                                          skb->nf_bridge->data, header_size);
 193 }
 194
 195 /*
 196  * When forwarding bridge frames, we save a copy of the original
 197  * header before processing.
 198  */
 199 int nf_bridge_copy_header(struct sk_buff *skb)
 200 {
 201         int err;
 202         int header_size = ETH_HLEN + nf_bridge_encap_header_len(skb);
 203
 204         err = skb_cow_head(skb, header_size);
 205         if (err)
 206                 return err;
 207
 208         skb_copy_to_linear_data_offset(skb, -header_size,
 209                                        skb->nf_bridge->data, header_size);
 210         __skb_push(skb, nf_bridge_encap_header_len(skb));
 211         return 0;
 212 }
 213
 214 /* PF_BRIDGE/PRE_ROUTING *********************************************/
 215 /* Undo the changes made for ip6tables PREROUTING and continue the
 216  * bridge PRE_ROUTING hook. */
 217 static int br_nf_pre_routing_finish_ipv6(struct sk_buff *skb)
 218 {
 219         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 220
 221         if (nf_bridge->mask & BRNF_PKT_TYPE) {
 222                 skb->pkt_type = PACKET_OTHERHOST;
 223                 nf_bridge->mask ^= BRNF_PKT_TYPE;
 224         }
 225         nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
 226
 227         skb->dst = (struct dst_entry *)&__fake_rtable;
 228         dst_hold(skb->dst);
 229
 230         skb->dev = nf_bridge->physindev;
 231         nf_bridge_push_encap_header(skb);
 232         NF_HOOK_THRESH(PF_BRIDGE, NF_BR_PRE_ROUTING, skb, skb->dev, NULL,
 233                        br_handle_frame_finish, 1);
 234
 235         return 0;
 236 }
 237
 238 static void __br_dnat_complain(void)
 239 {
 240         static unsigned long last_complaint;
 241
 242         if (jiffies - last_complaint >= 5 * HZ) {
 243                 printk(KERN_WARNING "Performing cross-bridge DNAT requires IP "
 244                        "forwarding to be enabled\n");
 245                 last_complaint = jiffies;
 246         }
 247 }
 248
 249 /* This requires some explaining. If DNAT has taken place,
 250  * we will need to fix up the destination Ethernet address,
 251  * and this is a tricky process.
 252  *
 253  * There are two cases to consider:
 254  * 1. The packet was DNAT'ed to a device in the same bridge
 255  *    port group as it was received on. We can still bridge
 256  *    the packet.
 257  * 2. The packet was DNAT'ed to a different device, either
 258  *    a non-bridged device or another bridge port group.
 259  *    The packet will need to be routed.
 260  *
 261  * The correct way of distinguishing between these two cases is to
 262  * call ip_route_input() and to look at skb->dst->dev, which is
 263  * changed to the destination device if ip_route_input() succeeds.
 264  *
 265  * Let us first consider the case that ip_route_input() succeeds:
 266  *
 267  * If skb->dst->dev equals the logical bridge device the packet
 268  * came in on, we can consider this bridging. The packet is passed
 269  * through the neighbour output function to build a new destination
 270  * MAC address, which will make the packet enter br_nf_local_out()
 271  * not much later. In that function it is assured that the iptables
 272  * FORWARD chain is traversed for the packet.
 273  *
 274  * Otherwise, the packet is considered to be routed and we just
 275  * change the destination MAC address so that the packet will
 276  * later be passed up to the IP stack to be routed. For a redirected
 277  * packet, ip_route_input() will give back the localhost as output device,
 278  * which differs from the bridge device.
 279  *
 280  * Let us now consider the case that ip_route_input() fails:
 281  *
 282  * This can be because the destination address is martian, in which case
 283  * the packet will be dropped.
 284  * After a "echo '0' > /proc/sys/net/ipv4/ip_forward" ip_route_input()
 285  * will fail, while __ip_route_output_key() will return success. The source
 286  * address for __ip_route_output_key() is set to zero, so __ip_route_output_key
 287  * thinks we're handling a locally generated packet and won't care
 288  * if IP forwarding is allowed. We send a warning message to the users's
 289  * log telling her to put IP forwarding on.
 290  *
 291  * ip_route_input() will also fail if there is no route available.
 292  * In that case we just drop the packet.
 293  *
 294  * --Lennert, 20020411
 295  * --Bart, 20020416 (updated)
 296  * --Bart, 20021007 (updated)
 297  * --Bart, 20062711 (updated) */
 298 static int br_nf_pre_routing_finish_bridge(struct sk_buff *skb)
 299 {
 300         if (skb->pkt_type == PACKET_OTHERHOST) {
 301                 skb->pkt_type = PACKET_HOST;
 302                 skb->nf_bridge->mask |= BRNF_PKT_TYPE;
 303         }
 304         skb->nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
 305
 306         skb->dev = bridge_parent(skb->dev);
 307         if (skb->dev) {
 308                 struct dst_entry *dst = skb->dst;
 309
 310                 nf_bridge_pull_encap_header(skb);
 311
 312                 if (dst->hh)
 313                         return neigh_hh_output(dst->hh, skb);
 314                 else if (dst->neighbour)
 315                         return dst->neighbour->output(skb);
 316         }
 317         kfree_skb(skb);
 318         return 0;
 319 }
 320
 321 static int br_nf_pre_routing_finish(struct sk_buff *skb)
 322 {
 323         struct net_device *dev = skb->dev;
 324         struct iphdr *iph = ip_hdr(skb);
 325         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 326         int err;
 327
 328         if (nf_bridge->mask & BRNF_PKT_TYPE) {
 329                 skb->pkt_type = PACKET_OTHERHOST;
 330                 nf_bridge->mask ^= BRNF_PKT_TYPE;
 331         }
 332         nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
 333         if (dnat_took_place(skb)) {
 334                 if ((err = ip_route_input(skb, iph->daddr, iph->saddr, iph->tos, dev))) {
 335                         struct rtable *rt;
 336                         struct flowi fl = {
 337                                 .nl_u = {
 338                                         .ip4_u = {
 339                                                  .daddr = iph->daddr,
 340                                                  .saddr = 0,
 341                                                  .tos = RT_TOS(iph->tos) },
 342                                 },
 343                                 .proto = 0,
 344                         };
 345                         struct in_device *in_dev = in_dev_get(dev);
 346
 347                         /* If err equals -EHOSTUNREACH the error is due to a
 348                          * martian destination or due to the fact that
 349                          * forwarding is disabled. For most martian packets,
 350                          * ip_route_output_key() will fail. It won't fail for 2 types of
 351                          * martian destinations: loopback destinations and destination
 352                          * 0.0.0.0. In both cases the packet will be dropped because the
 353                          * destination is the loopback device and not the bridge. */
 354                         if (err != -EHOSTUNREACH || !in_dev || IN_DEV_FORWARD(in_dev))
 355                                 goto free_skb;
 356
 357                         if (!ip_route_output_key(&init_net, &rt, &fl)) {
 358                                 /* - Bridged-and-DNAT'ed traffic doesn't
 359                                  *   require ip_forwarding. */
 360                                 if (((struct dst_entry *)rt)->dev == dev) {
 361                                         skb->dst = (struct dst_entry *)rt;
 362                                         goto bridged_dnat;
 363                                 }
 364                                 /* we are sure that forwarding is disabled, so printing
 365                                  * this message is no problem. Note that the packet could
 366                                  * still have a martian destination address, in which case
 367                                  * the packet could be dropped even if forwarding were enabled */
 368                                 __br_dnat_complain();
 369                                 dst_release((struct dst_entry *)rt);
 370                         }
 371 free_skb:
 372                         kfree_skb(skb);
 373                         return 0;
 374                 } else {
 375                         if (skb->dst->dev == dev) {
 376 bridged_dnat:
 377                                 /* Tell br_nf_local_out this is a
 378                                  * bridged frame */
 379                                 nf_bridge->mask |= BRNF_BRIDGED_DNAT;
 380                                 skb->dev = nf_bridge->physindev;
 381                                 nf_bridge_push_encap_header(skb);
 382                                 NF_HOOK_THRESH(PF_BRIDGE, NF_BR_PRE_ROUTING,
 383                                                skb, skb->dev, NULL,
 384                                                br_nf_pre_routing_finish_bridge,
 385                                                1);
 386                                 return 0;
 387                         }
 388                         memcpy(eth_hdr(skb)->h_dest, dev->dev_addr, ETH_ALEN);
 389                         skb->pkt_type = PACKET_HOST;
 390                 }
 391         } else {
 392                 skb->dst = (struct dst_entry *)&__fake_rtable;
 393                 dst_hold(skb->dst);
 394         }
 395
 396         skb->dev = nf_bridge->physindev;
 397         nf_bridge_push_encap_header(skb);
 398         NF_HOOK_THRESH(PF_BRIDGE, NF_BR_PRE_ROUTING, skb, skb->dev, NULL,
 399                        br_handle_frame_finish, 1);
 400
 401         return 0;
 402 }
 403
 404 /* Some common code for IPv4/IPv6 */
 405 static struct net_device *setup_pre_routing(struct sk_buff *skb)
 406 {
 407         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 408
 409         if (skb->pkt_type == PACKET_OTHERHOST) {
 410                 skb->pkt_type = PACKET_HOST;
 411                 nf_bridge->mask |= BRNF_PKT_TYPE;
 412         }
 413
 414         nf_bridge->mask |= BRNF_NF_BRIDGE_PREROUTING;
 415         nf_bridge->physindev = skb->dev;
 416         skb->dev = bridge_parent(skb->dev);
 417
 418         return skb->dev;
 419 }
 420
 421 /* We only check the length. A bridge shouldn't do any hop-by-hop stuff anyway */
 422 static int check_hbh_len(struct sk_buff *skb)
 423 {
 424         unsigned char *raw = (u8 *)(ipv6_hdr(skb) + 1);
 425         u32 pkt_len;
 426         const unsigned char *nh = skb_network_header(skb);
 427         int off = raw - nh;
 428         int len = (raw[1] + 1) << 3;
 429
 430         if ((raw + len) - skb->data > skb_headlen(skb))
 431                 goto bad;
 432
 433         off += 2;
 434         len -= 2;
 435
 436         while (len > 0) {
 437                 int optlen = nh[off + 1] + 2;
 438
 439                 switch (nh[off]) {
 440                 case IPV6_TLV_PAD0:
 441                         optlen = 1;
 442                         break;
 443
 444                 case IPV6_TLV_PADN:
 445                         break;
 446
 447                 case IPV6_TLV_JUMBO:
 448                         if (nh[off + 1] != 4 || (off & 3) != 2)
 449                                 goto bad;
 450                         pkt_len = ntohl(*(__be32 *) (nh + off + 2));
 451                         if (pkt_len <= IPV6_MAXPLEN ||
 452                             ipv6_hdr(skb)->payload_len)
 453                                 goto bad;
 454                         if (pkt_len > skb->len - sizeof(struct ipv6hdr))
 455                                 goto bad;
 456                         if (pskb_trim_rcsum(skb,
 457                                             pkt_len + sizeof(struct ipv6hdr)))
 458                                 goto bad;
 459                         nh = skb_network_header(skb);
 460                         break;
 461                 default:
 462                         if (optlen > len)
 463                                 goto bad;
 464                         break;
 465                 }
 466                 off += optlen;
 467                 len -= optlen;
 468         }
 469         if (len == 0)
 470                 return 0;
 471 bad:
 472         return -1;
 473
 474 }
 475
 476 /* Replicate the checks that IPv6 does on packet reception and pass the packet
 477  * to ip6tables, which doesn't support NAT, so things are fairly simple. */
 478 static unsigned int br_nf_pre_routing_ipv6(unsigned int hook,
 479                                            struct sk_buff *skb,
 480                                            const struct net_device *in,
 481                                            const struct net_device *out,
 482                                            int (*okfn)(struct sk_buff *))
 483 {
 484         struct ipv6hdr *hdr;
 485         u32 pkt_len;
 486
 487         if (skb->len < sizeof(struct ipv6hdr))
 488                 goto inhdr_error;
 489
 490         if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
 491                 goto inhdr_error;
 492
 493         hdr = ipv6_hdr(skb);
 494
 495         if (hdr->version != 6)
 496                 goto inhdr_error;
 497
 498         pkt_len = ntohs(hdr->payload_len);
 499
 500         if (pkt_len || hdr->nexthdr != NEXTHDR_HOP) {
 501                 if (pkt_len + sizeof(struct ipv6hdr) > skb->len)
 502                         goto inhdr_error;
 503                 if (pskb_trim_rcsum(skb, pkt_len + sizeof(struct ipv6hdr)))
 504                         goto inhdr_error;
 505         }
 506         if (hdr->nexthdr == NEXTHDR_HOP && check_hbh_len(skb))
 507                 goto inhdr_error;
 508
 509         nf_bridge_put(skb->nf_bridge);
 510         if (!nf_bridge_alloc(skb))
 511                 return NF_DROP;
 512         if (!setup_pre_routing(skb))
 513                 return NF_DROP;
 514
 515         NF_HOOK(PF_INET6, NF_INET_PRE_ROUTING, skb, skb->dev, NULL,
 516                 br_nf_pre_routing_finish_ipv6);
 517
 518         return NF_STOLEN;
 519
 520 inhdr_error:
 521         return NF_DROP;
 522 }
 523
 524 /* Direct IPv6 traffic to br_nf_pre_routing_ipv6.
 525  * Replicate the checks that IPv4 does on packet reception.
 526  * Set skb->dev to the bridge device (i.e. parent of the
 527  * receiving device) to make netfilter happy, the REDIRECT
 528  * target in particular.  Save the original destination IP
 529  * address to be able to detect DNAT afterwards. */
 530 static unsigned int br_nf_pre_routing(unsigned int hook, struct sk_buff *skb,
 531                                       const struct net_device *in,
 532                                       const struct net_device *out,
 533                                       int (*okfn)(struct sk_buff *))
 534 {
 535         struct iphdr *iph;
 536         __u32 len = nf_bridge_encap_header_len(skb);
 537
 538         if (unlikely(!pskb_may_pull(skb, len)))
 539                 goto out;
 540
 541         if (skb->protocol == htons(ETH_P_IPV6) || IS_VLAN_IPV6(skb) ||
 542             IS_PPPOE_IPV6(skb)) {
 543 #ifdef CONFIG_SYSCTL
 544                 if (!brnf_call_ip6tables)
 545                         return NF_ACCEPT;
 546 #endif
 547                 nf_bridge_pull_encap_header_rcsum(skb);
 548                 return br_nf_pre_routing_ipv6(hook, skb, in, out, okfn);
 549         }
 550 #ifdef CONFIG_SYSCTL
 551         if (!brnf_call_iptables)
 552                 return NF_ACCEPT;
 553 #endif
 554
 555         if (skb->protocol != htons(ETH_P_IP) && !IS_VLAN_IP(skb) &&
 556             !IS_PPPOE_IP(skb))
 557                 return NF_ACCEPT;
 558
 559         nf_bridge_pull_encap_header_rcsum(skb);
 560
 561         if (!pskb_may_pull(skb, sizeof(struct iphdr)))
 562                 goto inhdr_error;
 563
 564         iph = ip_hdr(skb);
 565         if (iph->ihl < 5 || iph->version != 4)
 566                 goto inhdr_error;
 567
 568         if (!pskb_may_pull(skb, 4 * iph->ihl))
 569                 goto inhdr_error;
 570
 571         iph = ip_hdr(skb);
 572         if (ip_fast_csum((__u8 *) iph, iph->ihl) != 0)
 573                 goto inhdr_error;
 574
 575         len = ntohs(iph->tot_len);
 576         if (skb->len < len || len < 4 * iph->ihl)
 577                 goto inhdr_error;
 578
 579         pskb_trim_rcsum(skb, len);
 580
 581         nf_bridge_put(skb->nf_bridge);
 582         if (!nf_bridge_alloc(skb))
 583                 return NF_DROP;
 584         if (!setup_pre_routing(skb))
 585                 return NF_DROP;
 586         store_orig_dstaddr(skb);
 587
 588         NF_HOOK(PF_INET, NF_INET_PRE_ROUTING, skb, skb->dev, NULL,
 589                 br_nf_pre_routing_finish);
 590
 591         return NF_STOLEN;
 592
 593 inhdr_error:
 594 //      IP_INC_STATS_BH(IpInHdrErrors);
 595 out:
 596         return NF_DROP;
 597 }
 598
 599
 600 /* PF_BRIDGE/LOCAL_IN ************************************************/
 601 /* The packet is locally destined, which requires a real
 602  * dst_entry, so detach the fake one.  On the way up, the
 603  * packet would pass through PRE_ROUTING again (which already
 604  * took place when the packet entered the bridge), but we
 605  * register an IPv4 PRE_ROUTING 'sabotage' hook that will
 606  * prevent this from happening. */
 607 static unsigned int br_nf_local_in(unsigned int hook, struct sk_buff *skb,
 608                                    const struct net_device *in,
 609                                    const struct net_device *out,
 610                                    int (*okfn)(struct sk_buff *))
 611 {
 612         if (skb->dst == (struct dst_entry *)&__fake_rtable) {
 613                 dst_release(skb->dst);
 614                 skb->dst = NULL;
 615         }
 616
 617         return NF_ACCEPT;
 618 }
 619
 620 /* PF_BRIDGE/FORWARD *************************************************/
 621 static int br_nf_forward_finish(struct sk_buff *skb)
 622 {
 623         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 624         struct net_device *in;
 625
 626         if (skb->protocol != htons(ETH_P_ARP) && !IS_VLAN_ARP(skb)) {
 627                 in = nf_bridge->physindev;
 628                 if (nf_bridge->mask & BRNF_PKT_TYPE) {
 629                         skb->pkt_type = PACKET_OTHERHOST;
 630                         nf_bridge->mask ^= BRNF_PKT_TYPE;
 631                 }
 632         } else {
 633                 in = *((struct net_device **)(skb->cb));
 634         }
 635         nf_bridge_push_encap_header(skb);
 636         NF_HOOK_THRESH(PF_BRIDGE, NF_BR_FORWARD, skb, in,
 637                        skb->dev, br_forward_finish, 1);
 638         return 0;
 639 }
 640
 641 /* This is the 'purely bridged' case.  For IP, we pass the packet to
 642  * netfilter with indev and outdev set to the bridge device,
 643  * but we are still able to filter on the 'real' indev/outdev
 644  * because of the physdev module. For ARP, indev and outdev are the
 645  * bridge ports. */
 646 static unsigned int br_nf_forward_ip(unsigned int hook, struct sk_buff *skb,
 647                                      const struct net_device *in,
 648                                      const struct net_device *out,
 649                                      int (*okfn)(struct sk_buff *))
 650 {
 651         struct nf_bridge_info *nf_bridge;
 652         struct net_device *parent;
 653         int pf;
 654
 655         if (!skb->nf_bridge)
 656                 return NF_ACCEPT;
 657
 658         /* Need exclusive nf_bridge_info since we might have multiple
 659          * different physoutdevs. */
 660         if (!nf_bridge_unshare(skb))
 661                 return NF_DROP;
 662
 663         parent = bridge_parent(out);
 664         if (!parent)
 665                 return NF_DROP;
 666
 667         if (skb->protocol == htons(ETH_P_IP) || IS_VLAN_IP(skb) ||
 668             IS_PPPOE_IP(skb))
 669                 pf = PF_INET;
 670         else
 671                 pf = PF_INET6;
 672
 673         nf_bridge_pull_encap_header(skb);
 674
 675         nf_bridge = skb->nf_bridge;
 676         if (skb->pkt_type == PACKET_OTHERHOST) {
 677                 skb->pkt_type = PACKET_HOST;
 678                 nf_bridge->mask |= BRNF_PKT_TYPE;
 679         }
 680
 681         /* The physdev module checks on this */
 682         nf_bridge->mask |= BRNF_BRIDGED;
 683         nf_bridge->physoutdev = skb->dev;
 684
 685         NF_HOOK(pf, NF_INET_FORWARD, skb, bridge_parent(in), parent,
 686                 br_nf_forward_finish);
 687
 688         return NF_STOLEN;
 689 }
 690
 691 static unsigned int br_nf_forward_arp(unsigned int hook, struct sk_buff *skb,
 692                                       const struct net_device *in,
 693                                       const struct net_device *out,
 694                                       int (*okfn)(struct sk_buff *))
 695 {
 696         struct net_device **d = (struct net_device **)(skb->cb);
 697
 698 #ifdef CONFIG_SYSCTL
 699         if (!brnf_call_arptables)
 700                 return NF_ACCEPT;
 701 #endif
 702
 703         if (skb->protocol != htons(ETH_P_ARP)) {
 704                 if (!IS_VLAN_ARP(skb))
 705                         return NF_ACCEPT;
 706                 nf_bridge_pull_encap_header(skb);
 707         }
 708
 709         if (arp_hdr(skb)->ar_pln != 4) {
 710                 if (IS_VLAN_ARP(skb))
 711                         nf_bridge_push_encap_header(skb);
 712                 return NF_ACCEPT;
 713         }
 714         *d = (struct net_device *)in;
 715         NF_HOOK(NF_ARP, NF_ARP_FORWARD, skb, (struct net_device *)in,
 716                 (struct net_device *)out, br_nf_forward_finish);
 717
 718         return NF_STOLEN;
 719 }
 720
 721 /* PF_BRIDGE/LOCAL_OUT ***********************************************
 722  *
 723  * This function sees both locally originated IP packets and forwarded
 724  * IP packets (in both cases the destination device is a bridge
 725  * device). It also sees bridged-and-DNAT'ed packets.
 726  *
 727  * If (nf_bridge->mask & BRNF_BRIDGED_DNAT) then the packet is bridged
 728  * and we fake the PF_BRIDGE/FORWARD hook. The function br_nf_forward()
 729  * will then fake the PF_INET/FORWARD hook. br_nf_local_out() has priority
 730  * NF_BR_PRI_FIRST, so no relevant PF_BRIDGE/INPUT functions have been nor
 731  * will be executed.
 732  */
 733 static unsigned int br_nf_local_out(unsigned int hook, struct sk_buff *skb,
 734                                     const struct net_device *in,
 735                                     const struct net_device *out,
 736                                     int (*okfn)(struct sk_buff *))
 737 {
 738         struct net_device *realindev;
 739         struct nf_bridge_info *nf_bridge;
 740
 741         if (!skb->nf_bridge)
 742                 return NF_ACCEPT;
 743
 744         /* Need exclusive nf_bridge_info since we might have multiple
 745          * different physoutdevs. */
 746         if (!nf_bridge_unshare(skb))
 747                 return NF_DROP;
 748
 749         nf_bridge = skb->nf_bridge;
 750         if (!(nf_bridge->mask & BRNF_BRIDGED_DNAT))
 751                 return NF_ACCEPT;
 752
 753         /* Bridged, take PF_BRIDGE/FORWARD.
 754          * (see big note in front of br_nf_pre_routing_finish) */
 755         nf_bridge->physoutdev = skb->dev;
 756         realindev = nf_bridge->physindev;
 757
 758         if (nf_bridge->mask & BRNF_PKT_TYPE) {
 759                 skb->pkt_type = PACKET_OTHERHOST;
 760                 nf_bridge->mask ^= BRNF_PKT_TYPE;
 761         }
 762         nf_bridge_push_encap_header(skb);
 763
 764         NF_HOOK(PF_BRIDGE, NF_BR_FORWARD, skb, realindev, skb->dev,
 765                 br_forward_finish);
 766         return NF_STOLEN;
 767 }
 768
 769 static int br_nf_dev_queue_xmit(struct sk_buff *skb)
 770 {
 771         if (skb->protocol == htons(ETH_P_IP) &&
 772             skb->len > skb->dev->mtu &&
 773             !skb_is_gso(skb))
 774                 return ip_fragment(skb, br_dev_queue_push_xmit);
 775         else
 776                 return br_dev_queue_push_xmit(skb);
 777 }
 778
 779 /* PF_BRIDGE/POST_ROUTING ********************************************/
 780 static unsigned int br_nf_post_routing(unsigned int hook, struct sk_buff *skb,
 781                                        const struct net_device *in,
 782                                        const struct net_device *out,
 783                                        int (*okfn)(struct sk_buff *))
 784 {
 785         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 786         struct net_device *realoutdev = bridge_parent(skb->dev);
 787         int pf;
 788
 789 #ifdef CONFIG_NETFILTER_DEBUG
 790         /* Be very paranoid. This probably won't happen anymore, but let's
 791          * keep the check just to be sure... */
 792         if (skb_mac_header(skb) < skb->head ||
 793             skb_mac_header(skb) + ETH_HLEN > skb->data) {
 794                 printk(KERN_CRIT "br_netfilter: Argh!! br_nf_post_routing: "
 795                        "bad mac.raw pointer.\n");
 796                 goto print_error;
 797         }
 798 #endif
 799
 800         if (!nf_bridge)
 801                 return NF_ACCEPT;
 802
 803         if (!(nf_bridge->mask & (BRNF_BRIDGED | BRNF_BRIDGED_DNAT)))
 804                 return NF_ACCEPT;
 805
 806         if (!realoutdev)
 807                 return NF_DROP;
 808
 809         if (skb->protocol == htons(ETH_P_IP) || IS_VLAN_IP(skb) ||
 810             IS_PPPOE_IP(skb))
 811                 pf = PF_INET;
 812         else
 813                 pf = PF_INET6;
 814
 815 #ifdef CONFIG_NETFILTER_DEBUG
 816         if (skb->dst == NULL) {
 817                 printk(KERN_INFO "br_netfilter post_routing: skb->dst == NULL\n");
 818                 goto print_error;
 819         }
 820 #endif
 821
 822         /* We assume any code from br_dev_queue_push_xmit onwards doesn't care
 823          * about the value of skb->pkt_type. */
 824         if (skb->pkt_type == PACKET_OTHERHOST) {
 825                 skb->pkt_type = PACKET_HOST;
 826                 nf_bridge->mask |= BRNF_PKT_TYPE;
 827         }
 828
 829         nf_bridge_pull_encap_header(skb);
 830         nf_bridge_save_header(skb);
 831
 832         NF_HOOK(pf, NF_INET_POST_ROUTING, skb, NULL, realoutdev,
 833                 br_nf_dev_queue_xmit);
 834
 835         return NF_STOLEN;
 836
 837 #ifdef CONFIG_NETFILTER_DEBUG
 838 print_error:
 839         if (skb->dev != NULL) {
 840                 printk("[%s]", skb->dev->name);
 841                 if (realoutdev)
 842                         printk("[%s]", realoutdev->name);
 843         }
 844         printk(" head:%p, raw:%p, data:%p\n", skb->head, skb_mac_header(skb),
 845                skb->data);
 846         dump_stack();
 847         return NF_ACCEPT;
 848 #endif
 849 }
 850
 851 /* IP/SABOTAGE *****************************************************/
 852 /* Don't hand locally destined packets to PF_INET(6)/PRE_ROUTING
 853  * for the second time. */
 854 static unsigned int ip_sabotage_in(unsigned int hook, struct sk_buff *skb,
 855                                    const struct net_device *in,
 856                                    const struct net_device *out,
 857                                    int (*okfn)(struct sk_buff *))
 858 {
 859         if (skb->nf_bridge &&
 860             !(skb->nf_bridge->mask & BRNF_NF_BRIDGE_PREROUTING)) {
 861                 return NF_STOP;
 862         }
 863
 864         return NF_ACCEPT;
 865 }
 866
 867 /* For br_nf_local_out we need (prio = NF_BR_PRI_FIRST), to insure that innocent
 868  * PF_BRIDGE/NF_BR_LOCAL_OUT functions don't get bridged traffic as input.
 869  * For br_nf_post_routing, we need (prio = NF_BR_PRI_LAST), because
 870  * ip_refrag() can return NF_STOLEN. */
 871 static struct nf_hook_ops br_nf_ops[] __read_mostly = {
 872         { .hook = br_nf_pre_routing,
 873           .owner = THIS_MODULE,
 874           .pf = PF_BRIDGE,
 875           .hooknum = NF_BR_PRE_ROUTING,
 876           .priority = NF_BR_PRI_BRNF, },
 877         { .hook = br_nf_local_in,
 878           .owner = THIS_MODULE,
 879           .pf = PF_BRIDGE,
 880           .hooknum = NF_BR_LOCAL_IN,
 881           .priority = NF_BR_PRI_BRNF, },
 882         { .hook = br_nf_forward_ip,
 883           .owner = THIS_MODULE,
 884           .pf = PF_BRIDGE,
 885           .hooknum = NF_BR_FORWARD,
 886           .priority = NF_BR_PRI_BRNF - 1, },
 887         { .hook = br_nf_forward_arp,
 888           .owner = THIS_MODULE,
 889           .pf = PF_BRIDGE,
 890           .hooknum = NF_BR_FORWARD,
 891           .priority = NF_BR_PRI_BRNF, },
 892         { .hook = br_nf_local_out,
 893           .owner = THIS_MODULE,
 894           .pf = PF_BRIDGE,
 895           .hooknum = NF_BR_LOCAL_OUT,
 896           .priority = NF_BR_PRI_FIRST, },
 897         { .hook = br_nf_post_routing,
 898           .owner = THIS_MODULE,
 899           .pf = PF_BRIDGE,
 900           .hooknum = NF_BR_POST_ROUTING,
 901           .priority = NF_BR_PRI_LAST, },
 902         { .hook = ip_sabotage_in,
 903           .owner = THIS_MODULE,
 904           .pf = PF_INET,
 905           .hooknum = NF_INET_PRE_ROUTING,
 906           .priority = NF_IP_PRI_FIRST, },
 907         { .hook = ip_sabotage_in,
 908           .owner = THIS_MODULE,
 909           .pf = PF_INET6,
 910           .hooknum = NF_INET_PRE_ROUTING,
 911           .priority = NF_IP6_PRI_FIRST, },
 912 };
 913
 914 #ifdef CONFIG_SYSCTL
 915 static
 916 int brnf_sysctl_call_tables(ctl_table * ctl, int write, struct file *filp,
 917                             void __user * buffer, size_t * lenp, loff_t * ppos)
 918 {
 919         int ret;
 920
 921         ret = proc_dointvec(ctl, write, filp, buffer, lenp, ppos);
 922
 923         if (write && *(int *)(ctl->data))
 924                 *(int *)(ctl->data) = 1;
 925         return ret;
 926 }
 927
 928 static ctl_table brnf_table[] = {
 929         {
 930                 .procname       = "bridge-nf-call-arptables",
 931                 .data           = &brnf_call_arptables,
 932                 .maxlen         = sizeof(int),
 933                 .mode           = 0644,
 934                 .proc_handler   = &brnf_sysctl_call_tables,
 935         },
 936         {
 937                 .procname       = "bridge-nf-call-iptables",
 938                 .data           = &brnf_call_iptables,
 939                 .maxlen         = sizeof(int),
 940                 .mode           = 0644,
 941                 .proc_handler   = &brnf_sysctl_call_tables,
 942         },
 943         {
 944                 .procname       = "bridge-nf-call-ip6tables",
 945                 .data           = &brnf_call_ip6tables,
 946                 .maxlen         = sizeof(int),
 947                 .mode           = 0644,
 948                 .proc_handler   = &brnf_sysctl_call_tables,
 949         },
 950         {
 951                 .procname       = "bridge-nf-filter-vlan-tagged",
 952                 .data           = &brnf_filter_vlan_tagged,
 953                 .maxlen         = sizeof(int),
 954                 .mode           = 0644,
 955                 .proc_handler   = &brnf_sysctl_call_tables,
 956         },
 957         {
 958                 .procname       = "bridge-nf-filter-pppoe-tagged",
 959                 .data           = &brnf_filter_pppoe_tagged,
 960                 .maxlen         = sizeof(int),
 961                 .mode           = 0644,
 962                 .proc_handler   = &brnf_sysctl_call_tables,
 963         },
 964         { .ctl_name = 0 }
 965 };
 966
 967 static struct ctl_path brnf_path[] = {
 968         { .procname = "net", .ctl_name = CTL_NET, },
 969         { .procname = "bridge", .ctl_name = NET_BRIDGE, },
 970         { }
 971 };
 972 #endif
 973
 974 int __init br_netfilter_init(void)
 975 {
 976         int ret;
 977
 978         ret = nf_register_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
 979         if (ret < 0)
 980                 return ret;
 981 #ifdef CONFIG_SYSCTL
 982         brnf_sysctl_header = register_sysctl_paths(brnf_path, brnf_table);
 983         if (brnf_sysctl_header == NULL) {
 984                 printk(KERN_WARNING
 985                        "br_netfilter: can't register to sysctl.\n");
 986                 nf_unregister_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
 987                 return -ENOMEM;
 988         }
 989 #endif
 990         printk(KERN_NOTICE "Bridge firewalling registered\n");
 991         return 0;
 992 }
 993
 994 void br_netfilter_fini(void)
 995 {
 996         nf_unregister_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
 997 #ifdef CONFIG_SYSCTL
 998         unregister_sysctl_table(brnf_sysctl_header);
 999 #endif
1000 }