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                <bdschuym@pandora.be>
   8  *
   9  *      This program is free software; you can redistribute it and/or
  10  *      modify it under the terms of the GNU General Public License
  11  *      as published by the Free Software Foundation; either version
  12  *      2 of the License, or (at your option) any later version.
  13  *
  14  *      Lennert dedicates this file to Kerstin Wurdinger.
  15  */
  16
  17 #include <linux/module.h>
  18 #include <linux/kernel.h>
  19 #include <linux/slab.h>
  20 #include <linux/ip.h>
  21 #include <linux/netdevice.h>
  22 #include <linux/skbuff.h>
  23 #include <linux/if_arp.h>
  24 #include <linux/if_ether.h>
  25 #include <linux/if_vlan.h>
  26 #include <linux/if_pppox.h>
  27 #include <linux/ppp_defs.h>
  28 #include <linux/netfilter_bridge.h>
  29 #include <linux/netfilter_ipv4.h>
  30 #include <linux/netfilter_ipv6.h>
  31 #include <linux/netfilter_arp.h>
  32 #include <linux/in_route.h>
  33 #include <linux/inetdevice.h>
  34
  35 #include <net/ip.h>
  36 #include <net/ipv6.h>
  37 #include <net/route.h>
  38
  39 #include <asm/uaccess.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) = ip_hdr(skb)->daddr)
  48 #define dnat_took_place(skb)     (skb_origaddr(skb) != ip_hdr(skb)->daddr)
  49
  50 #ifdef CONFIG_SYSCTL
  51 static struct ctl_table_header *brnf_sysctl_header;
  52 static int brnf_call_iptables __read_mostly = 1;
  53 static int brnf_call_ip6tables __read_mostly = 1;
  54 static int brnf_call_arptables __read_mostly = 1;
  55 static int brnf_filter_vlan_tagged __read_mostly = 0;
  56 static int brnf_filter_pppoe_tagged __read_mostly = 0;
  57 #else
  58 #define brnf_call_iptables 1
  59 #define brnf_call_ip6tables 1
  60 #define brnf_call_arptables 1
  61 #define brnf_filter_vlan_tagged 0
  62 #define brnf_filter_pppoe_tagged 0
  63 #endif
  64
  65 static inline __be16 vlan_proto(const struct sk_buff *skb)
  66 {
  67         if (vlan_tx_tag_present(skb))
  68                 return skb->protocol;
  69         else if (skb->protocol == htons(ETH_P_8021Q))
  70                 return vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
  71         else
  72                 return 0;
  73 }
  74
  75 #define IS_VLAN_IP(skb) \
  76         (vlan_proto(skb) == htons(ETH_P_IP) && \
  77          brnf_filter_vlan_tagged)
  78
  79 #define IS_VLAN_IPV6(skb) \
  80         (vlan_proto(skb) == htons(ETH_P_IPV6) && \
  81          brnf_filter_vlan_tagged)
  82
  83 #define IS_VLAN_ARP(skb) \
  84         (vlan_proto(skb) == htons(ETH_P_ARP) && \
  85          brnf_filter_vlan_tagged)
  86
  87 static inline __be16 pppoe_proto(const struct sk_buff *skb)
  88 {
  89         return *((__be16 *)(skb_mac_header(skb) + ETH_HLEN +
  90                             sizeof(struct pppoe_hdr)));
  91 }
  92
  93 #define IS_PPPOE_IP(skb) \
  94         (skb->protocol == htons(ETH_P_PPP_SES) && \
  95          pppoe_proto(skb) == htons(PPP_IP) && \
  96          brnf_filter_pppoe_tagged)
  97
  98 #define IS_PPPOE_IPV6(skb) \
  99         (skb->protocol == htons(ETH_P_PPP_SES) && \
 100          pppoe_proto(skb) == htons(PPP_IPV6) && \
 101          brnf_filter_pppoe_tagged)
 102
 103 static void fake_update_pmtu(struct dst_entry *dst, u32 mtu)
 104 {
 105 }
 106
 107 static struct dst_ops fake_dst_ops = {
 108         .family =               AF_INET,
 109         .protocol =             cpu_to_be16(ETH_P_IP),
 110         .update_pmtu =          fake_update_pmtu,
 111 };
 112
 113 /*
 114  * Initialize bogus route table used to keep netfilter happy.
 115  * Currently, we fill in the PMTU entry because netfilter
 116  * refragmentation needs it, and the rt_flags entry because
 117  * ipt_REJECT needs it.  Future netfilter modules might
 118  * require us to fill additional fields.
 119  */
 120 void br_netfilter_rtable_init(struct net_bridge *br)
 121 {
 122         struct rtable *rt = &br->fake_rtable;
 123
 124         atomic_set(&rt->dst.__refcnt, 1);
 125         rt->dst.dev = br->dev;
 126         rt->dst.path = &rt->dst;
 127         dst_metric_set(&rt->dst, RTAX_MTU, 1500);
 128         rt->dst.flags   = DST_NOXFRM;
 129         rt->dst.ops = &fake_dst_ops;
 130 }
 131
 132 static inline struct rtable *bridge_parent_rtable(const struct net_device *dev)
 133 {
 134         struct net_bridge_port *port;
 135
 136         port = br_port_get_rcu(dev);
 137         return port ? &port->br->fake_rtable : NULL;
 138 }
 139
 140 static inline struct net_device *bridge_parent(const struct net_device *dev)
 141 {
 142         struct net_bridge_port *port;
 143
 144         port = br_port_get_rcu(dev);
 145         return port ? port->br->dev : NULL;
 146 }
 147
 148 static inline struct nf_bridge_info *nf_bridge_alloc(struct sk_buff *skb)
 149 {
 150         skb->nf_bridge = kzalloc(sizeof(struct nf_bridge_info), GFP_ATOMIC);
 151         if (likely(skb->nf_bridge))
 152                 atomic_set(&(skb->nf_bridge->use), 1);
 153
 154         return skb->nf_bridge;
 155 }
 156
 157 static inline struct nf_bridge_info *nf_bridge_unshare(struct sk_buff *skb)
 158 {
 159         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 160
 161         if (atomic_read(&nf_bridge->use) > 1) {
 162                 struct nf_bridge_info *tmp = nf_bridge_alloc(skb);
 163
 164                 if (tmp) {
 165                         memcpy(tmp, nf_bridge, sizeof(struct nf_bridge_info));
 166                         atomic_set(&tmp->use, 1);
 167                 }
 168                 nf_bridge_put(nf_bridge);
 169                 nf_bridge = tmp;
 170         }
 171         return nf_bridge;
 172 }
 173
 174 static inline void nf_bridge_push_encap_header(struct sk_buff *skb)
 175 {
 176         unsigned int len = nf_bridge_encap_header_len(skb);
 177
 178         skb_push(skb, len);
 179         skb->network_header -= len;
 180 }
 181
 182 static inline void nf_bridge_pull_encap_header(struct sk_buff *skb)
 183 {
 184         unsigned int len = nf_bridge_encap_header_len(skb);
 185
 186         skb_pull(skb, len);
 187         skb->network_header += len;
 188 }
 189
 190 static inline void nf_bridge_pull_encap_header_rcsum(struct sk_buff *skb)
 191 {
 192         unsigned int len = nf_bridge_encap_header_len(skb);
 193
 194         skb_pull_rcsum(skb, len);
 195         skb->network_header += len;
 196 }
 197
 198 static inline void nf_bridge_save_header(struct sk_buff *skb)
 199 {
 200         int header_size = ETH_HLEN + nf_bridge_encap_header_len(skb);
 201
 202         skb_copy_from_linear_data_offset(skb, -header_size,
 203                                          skb->nf_bridge->data, header_size);
 204 }
 205
 206 static inline void nf_bridge_update_protocol(struct sk_buff *skb)
 207 {
 208         if (skb->nf_bridge->mask & BRNF_8021Q)
 209                 skb->protocol = htons(ETH_P_8021Q);
 210         else if (skb->nf_bridge->mask & BRNF_PPPoE)
 211                 skb->protocol = htons(ETH_P_PPP_SES);
 212 }
 213
 214 /* When handing a packet over to the IP layer
 215  * check whether we have a skb that is in the
 216  * expected format
 217  */
 218
 219 static int br_parse_ip_options(struct sk_buff *skb)
 220 {
 221         struct ip_options *opt;
 222         struct iphdr *iph;
 223         struct net_device *dev = skb->dev;
 224         u32 len;
 225
 226         iph = ip_hdr(skb);
 227         opt = &(IPCB(skb)->opt);
 228
 229         /* Basic sanity checks */
 230         if (iph->ihl < 5 || iph->version != 4)
 231                 goto inhdr_error;
 232
 233         if (!pskb_may_pull(skb, iph->ihl*4))
 234                 goto inhdr_error;
 235
 236         iph = ip_hdr(skb);
 237         if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl)))
 238                 goto inhdr_error;
 239
 240         len = ntohs(iph->tot_len);
 241         if (skb->len < len) {
 242                 IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INTRUNCATEDPKTS);
 243                 goto drop;
 244         } else if (len < (iph->ihl*4))
 245                 goto inhdr_error;
 246
 247         if (pskb_trim_rcsum(skb, len)) {
 248                 IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INDISCARDS);
 249                 goto drop;
 250         }
 251
 252         /* Zero out the CB buffer if no options present */
 253         if (iph->ihl == 5) {
 254                 memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
 255                 return 0;
 256         }
 257
 258         opt->optlen = iph->ihl*4 - sizeof(struct iphdr);
 259         if (ip_options_compile(dev_net(dev), opt, skb))
 260                 goto inhdr_error;
 261
 262         /* Check correct handling of SRR option */
 263         if (unlikely(opt->srr)) {
 264                 struct in_device *in_dev = __in_dev_get_rcu(dev);
 265                 if (in_dev && !IN_DEV_SOURCE_ROUTE(in_dev))
 266                         goto drop;
 267
 268                 if (ip_options_rcv_srr(skb))
 269                         goto drop;
 270         }
 271
 272         return 0;
 273
 274 inhdr_error:
 275         IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INHDRERRORS);
 276 drop:
 277         return -1;
 278 }
 279
 280 /* Fill in the header for fragmented IP packets handled by
 281  * the IPv4 connection tracking code.
 282  */
 283 int nf_bridge_copy_header(struct sk_buff *skb)
 284 {
 285         int err;
 286         unsigned int header_size;
 287
 288         nf_bridge_update_protocol(skb);
 289         header_size = ETH_HLEN + nf_bridge_encap_header_len(skb);
 290         err = skb_cow_head(skb, header_size);
 291         if (err)
 292                 return err;
 293
 294         skb_copy_to_linear_data_offset(skb, -header_size,
 295                                        skb->nf_bridge->data, header_size);
 296         __skb_push(skb, nf_bridge_encap_header_len(skb));
 297         return 0;
 298 }
 299
 300 /* PF_BRIDGE/PRE_ROUTING *********************************************/
 301 /* Undo the changes made for ip6tables PREROUTING and continue the
 302  * bridge PRE_ROUTING hook. */
 303 static int br_nf_pre_routing_finish_ipv6(struct sk_buff *skb)
 304 {
 305         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 306         struct rtable *rt;
 307
 308         if (nf_bridge->mask & BRNF_PKT_TYPE) {
 309                 skb->pkt_type = PACKET_OTHERHOST;
 310                 nf_bridge->mask ^= BRNF_PKT_TYPE;
 311         }
 312         nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
 313
 314         rt = bridge_parent_rtable(nf_bridge->physindev);
 315         if (!rt) {
 316                 kfree_skb(skb);
 317                 return 0;
 318         }
 319         skb_dst_set_noref(skb, &rt->dst);
 320
 321         skb->dev = nf_bridge->physindev;
 322         nf_bridge_update_protocol(skb);
 323         nf_bridge_push_encap_header(skb);
 324         NF_HOOK_THRESH(NFPROTO_BRIDGE, NF_BR_PRE_ROUTING, skb, skb->dev, NULL,
 325                        br_handle_frame_finish, 1);
 326
 327         return 0;
 328 }
 329
 330 /* Obtain the correct destination MAC address, while preserving the original
 331  * source MAC address. If we already know this address, we just copy it. If we
 332  * don't, we use the neighbour framework to find out. In both cases, we make
 333  * sure that br_handle_frame_finish() is called afterwards.
 334  */
 335 static int br_nf_pre_routing_finish_bridge(struct sk_buff *skb)
 336 {
 337         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 338         struct dst_entry *dst;
 339
 340         skb->dev = bridge_parent(skb->dev);
 341         if (!skb->dev)
 342                 goto free_skb;
 343         dst = skb_dst(skb);
 344         if (dst->hh) {
 345                 neigh_hh_bridge(dst->hh, skb);
 346                 skb->dev = nf_bridge->physindev;
 347                 return br_handle_frame_finish(skb);
 348         } else if (dst->neighbour) {
 349                 /* the neighbour function below overwrites the complete
 350                  * MAC header, so we save the Ethernet source address and
 351                  * protocol number. */
 352                 skb_copy_from_linear_data_offset(skb, -(ETH_HLEN-ETH_ALEN), skb->nf_bridge->data, ETH_HLEN-ETH_ALEN);
 353                 /* tell br_dev_xmit to continue with forwarding */
 354                 nf_bridge->mask |= BRNF_BRIDGED_DNAT;
 355                 return dst->neighbour->output(skb);
 356         }
 357 free_skb:
 358         kfree_skb(skb);
 359         return 0;
 360 }
 361
 362 /* This requires some explaining. If DNAT has taken place,
 363  * we will need to fix up the destination Ethernet address.
 364  *
 365  * There are two cases to consider:
 366  * 1. The packet was DNAT'ed to a device in the same bridge
 367  *    port group as it was received on. We can still bridge
 368  *    the packet.
 369  * 2. The packet was DNAT'ed to a different device, either
 370  *    a non-bridged device or another bridge port group.
 371  *    The packet will need to be routed.
 372  *
 373  * The correct way of distinguishing between these two cases is to
 374  * call ip_route_input() and to look at skb->dst->dev, which is
 375  * changed to the destination device if ip_route_input() succeeds.
 376  *
 377  * Let's first consider the case that ip_route_input() succeeds:
 378  *
 379  * If the output device equals the logical bridge device the packet
 380  * came in on, we can consider this bridging. The corresponding MAC
 381  * address will be obtained in br_nf_pre_routing_finish_bridge.
 382  * Otherwise, the packet is considered to be routed and we just
 383  * change the destination MAC address so that the packet will
 384  * later be passed up to the IP stack to be routed. For a redirected
 385  * packet, ip_route_input() will give back the localhost as output device,
 386  * which differs from the bridge device.
 387  *
 388  * Let's now consider the case that ip_route_input() fails:
 389  *
 390  * This can be because the destination address is martian, in which case
 391  * the packet will be dropped.
 392  * If IP forwarding is disabled, ip_route_input() will fail, while
 393  * ip_route_output_key() can return success. The source
 394  * address for ip_route_output_key() is set to zero, so ip_route_output_key()
 395  * thinks we're handling a locally generated packet and won't care
 396  * if IP forwarding is enabled. If the output device equals the logical bridge
 397  * device, we proceed as if ip_route_input() succeeded. If it differs from the
 398  * logical bridge port or if ip_route_output_key() fails we drop the packet.
 399  */
 400 static int br_nf_pre_routing_finish(struct sk_buff *skb)
 401 {
 402         struct net_device *dev = skb->dev;
 403         struct iphdr *iph = ip_hdr(skb);
 404         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 405         struct rtable *rt;
 406         int err;
 407
 408         if (nf_bridge->mask & BRNF_PKT_TYPE) {
 409                 skb->pkt_type = PACKET_OTHERHOST;
 410                 nf_bridge->mask ^= BRNF_PKT_TYPE;
 411         }
 412         nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
 413         if (dnat_took_place(skb)) {
 414                 if ((err = ip_route_input(skb, iph->daddr, iph->saddr, iph->tos, dev))) {
 415                         struct in_device *in_dev = __in_dev_get_rcu(dev);
 416
 417                         /* If err equals -EHOSTUNREACH the error is due to a
 418                          * martian destination or due to the fact that
 419                          * forwarding is disabled. For most martian packets,
 420                          * ip_route_output_key() will fail. It won't fail for 2 types of
 421                          * martian destinations: loopback destinations and destination
 422                          * 0.0.0.0. In both cases the packet will be dropped because the
 423                          * destination is the loopback device and not the bridge. */
 424                         if (err != -EHOSTUNREACH || !in_dev || IN_DEV_FORWARD(in_dev))
 425                                 goto free_skb;
 426
 427                         rt = ip_route_output(dev_net(dev), iph->daddr, 0,
 428                                              RT_TOS(iph->tos), 0);
 429                         if (!IS_ERR(rt)) {
 430                                 /* - Bridged-and-DNAT'ed traffic doesn't
 431                                  *   require ip_forwarding. */
 432                                 if (rt->dst.dev == dev) {
 433                                         skb_dst_set(skb, &rt->dst);
 434                                         goto bridged_dnat;
 435                                 }
 436                                 ip_rt_put(rt);
 437                         }
 438 free_skb:
 439                         kfree_skb(skb);
 440                         return 0;
 441                 } else {
 442                         if (skb_dst(skb)->dev == dev) {
 443 bridged_dnat:
 444                                 skb->dev = nf_bridge->physindev;
 445                                 nf_bridge_update_protocol(skb);
 446                                 nf_bridge_push_encap_header(skb);
 447                                 NF_HOOK_THRESH(NFPROTO_BRIDGE,
 448                                                NF_BR_PRE_ROUTING,
 449                                                skb, skb->dev, NULL,
 450                                                br_nf_pre_routing_finish_bridge,
 451                                                1);
 452                                 return 0;
 453                         }
 454                         memcpy(eth_hdr(skb)->h_dest, dev->dev_addr, ETH_ALEN);
 455                         skb->pkt_type = PACKET_HOST;
 456                 }
 457         } else {
 458                 rt = bridge_parent_rtable(nf_bridge->physindev);
 459                 if (!rt) {
 460                         kfree_skb(skb);
 461                         return 0;
 462                 }
 463                 skb_dst_set_noref(skb, &rt->dst);
 464         }
 465
 466         skb->dev = nf_bridge->physindev;
 467         nf_bridge_update_protocol(skb);
 468         nf_bridge_push_encap_header(skb);
 469         NF_HOOK_THRESH(NFPROTO_BRIDGE, NF_BR_PRE_ROUTING, skb, skb->dev, NULL,
 470                        br_handle_frame_finish, 1);
 471
 472         return 0;
 473 }
 474
 475 /* Some common code for IPv4/IPv6 */
 476 static struct net_device *setup_pre_routing(struct sk_buff *skb)
 477 {
 478         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 479
 480         if (skb->pkt_type == PACKET_OTHERHOST) {
 481                 skb->pkt_type = PACKET_HOST;
 482                 nf_bridge->mask |= BRNF_PKT_TYPE;
 483         }
 484
 485         nf_bridge->mask |= BRNF_NF_BRIDGE_PREROUTING;
 486         nf_bridge->physindev = skb->dev;
 487         skb->dev = bridge_parent(skb->dev);
 488         if (skb->protocol == htons(ETH_P_8021Q))
 489                 nf_bridge->mask |= BRNF_8021Q;
 490         else if (skb->protocol == htons(ETH_P_PPP_SES))
 491                 nf_bridge->mask |= BRNF_PPPoE;
 492
 493         return skb->dev;
 494 }
 495
 496 /* We only check the length. A bridge shouldn't do any hop-by-hop stuff anyway */
 497 static int check_hbh_len(struct sk_buff *skb)
 498 {
 499         unsigned char *raw = (u8 *)(ipv6_hdr(skb) + 1);
 500         u32 pkt_len;
 501         const unsigned char *nh = skb_network_header(skb);
 502         int off = raw - nh;
 503         int len = (raw[1] + 1) << 3;
 504
 505         if ((raw + len) - skb->data > skb_headlen(skb))
 506                 goto bad;
 507
 508         off += 2;
 509         len -= 2;
 510
 511         while (len > 0) {
 512                 int optlen = nh[off + 1] + 2;
 513
 514                 switch (nh[off]) {
 515                 case IPV6_TLV_PAD0:
 516                         optlen = 1;
 517                         break;
 518
 519                 case IPV6_TLV_PADN:
 520                         break;
 521
 522                 case IPV6_TLV_JUMBO:
 523                         if (nh[off + 1] != 4 || (off & 3) != 2)
 524                                 goto bad;
 525                         pkt_len = ntohl(*(__be32 *) (nh + off + 2));
 526                         if (pkt_len <= IPV6_MAXPLEN ||
 527                             ipv6_hdr(skb)->payload_len)
 528                                 goto bad;
 529                         if (pkt_len > skb->len - sizeof(struct ipv6hdr))
 530                                 goto bad;
 531                         if (pskb_trim_rcsum(skb,
 532                                             pkt_len + sizeof(struct ipv6hdr)))
 533                                 goto bad;
 534                         nh = skb_network_header(skb);
 535                         break;
 536                 default:
 537                         if (optlen > len)
 538                                 goto bad;
 539                         break;
 540                 }
 541                 off += optlen;
 542                 len -= optlen;
 543         }
 544         if (len == 0)
 545                 return 0;
 546 bad:
 547         return -1;
 548
 549 }
 550
 551 /* Replicate the checks that IPv6 does on packet reception and pass the packet
 552  * to ip6tables, which doesn't support NAT, so things are fairly simple. */
 553 static unsigned int br_nf_pre_routing_ipv6(unsigned int hook,
 554                                            struct sk_buff *skb,
 555                                            const struct net_device *in,
 556                                            const struct net_device *out,
 557                                            int (*okfn)(struct sk_buff *))
 558 {
 559         struct ipv6hdr *hdr;
 560         u32 pkt_len;
 561
 562         if (skb->len < sizeof(struct ipv6hdr))
 563                 return NF_DROP;
 564
 565         if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
 566                 return NF_DROP;
 567
 568         hdr = ipv6_hdr(skb);
 569
 570         if (hdr->version != 6)
 571                 return NF_DROP;
 572
 573         pkt_len = ntohs(hdr->payload_len);
 574
 575         if (pkt_len || hdr->nexthdr != NEXTHDR_HOP) {
 576                 if (pkt_len + sizeof(struct ipv6hdr) > skb->len)
 577                         return NF_DROP;
 578                 if (pskb_trim_rcsum(skb, pkt_len + sizeof(struct ipv6hdr)))
 579                         return NF_DROP;
 580         }
 581         if (hdr->nexthdr == NEXTHDR_HOP && check_hbh_len(skb))
 582                 return NF_DROP;
 583
 584         nf_bridge_put(skb->nf_bridge);
 585         if (!nf_bridge_alloc(skb))
 586                 return NF_DROP;
 587         if (!setup_pre_routing(skb))
 588                 return NF_DROP;
 589
 590         skb->protocol = htons(ETH_P_IPV6);
 591         NF_HOOK(NFPROTO_IPV6, NF_INET_PRE_ROUTING, skb, skb->dev, NULL,
 592                 br_nf_pre_routing_finish_ipv6);
 593
 594         return NF_STOLEN;
 595 }
 596
 597 /* Direct IPv6 traffic to br_nf_pre_routing_ipv6.
 598  * Replicate the checks that IPv4 does on packet reception.
 599  * Set skb->dev to the bridge device (i.e. parent of the
 600  * receiving device) to make netfilter happy, the REDIRECT
 601  * target in particular.  Save the original destination IP
 602  * address to be able to detect DNAT afterwards. */
 603 static unsigned int br_nf_pre_routing(unsigned int hook, struct sk_buff *skb,
 604                                       const struct net_device *in,
 605                                       const struct net_device *out,
 606                                       int (*okfn)(struct sk_buff *))
 607 {
 608         struct net_bridge_port *p;
 609         struct net_bridge *br;
 610         __u32 len = nf_bridge_encap_header_len(skb);
 611
 612         if (unlikely(!pskb_may_pull(skb, len)))
 613                 return NF_DROP;
 614
 615         p = br_port_get_rcu(in);
 616         if (p == NULL)
 617                 return NF_DROP;
 618         br = p->br;
 619
 620         if (skb->protocol == htons(ETH_P_IPV6) || IS_VLAN_IPV6(skb) ||
 621             IS_PPPOE_IPV6(skb)) {
 622                 if (!brnf_call_ip6tables && !br->nf_call_ip6tables)
 623                         return NF_ACCEPT;
 624
 625                 nf_bridge_pull_encap_header_rcsum(skb);
 626                 return br_nf_pre_routing_ipv6(hook, skb, in, out, okfn);
 627         }
 628
 629         if (!brnf_call_iptables && !br->nf_call_iptables)
 630                 return NF_ACCEPT;
 631
 632         if (skb->protocol != htons(ETH_P_IP) && !IS_VLAN_IP(skb) &&
 633             !IS_PPPOE_IP(skb))
 634                 return NF_ACCEPT;
 635
 636         nf_bridge_pull_encap_header_rcsum(skb);
 637
 638         if (br_parse_ip_options(skb))
 639                 return NF_DROP;
 640
 641         nf_bridge_put(skb->nf_bridge);
 642         if (!nf_bridge_alloc(skb))
 643                 return NF_DROP;
 644         if (!setup_pre_routing(skb))
 645                 return NF_DROP;
 646         store_orig_dstaddr(skb);
 647         skb->protocol = htons(ETH_P_IP);
 648
 649         NF_HOOK(NFPROTO_IPV4, NF_INET_PRE_ROUTING, skb, skb->dev, NULL,
 650                 br_nf_pre_routing_finish);
 651
 652         return NF_STOLEN;
 653 }
 654
 655
 656 /* PF_BRIDGE/LOCAL_IN ************************************************/
 657 /* The packet is locally destined, which requires a real
 658  * dst_entry, so detach the fake one.  On the way up, the
 659  * packet would pass through PRE_ROUTING again (which already
 660  * took place when the packet entered the bridge), but we
 661  * register an IPv4 PRE_ROUTING 'sabotage' hook that will
 662  * prevent this from happening. */
 663 static unsigned int br_nf_local_in(unsigned int hook, struct sk_buff *skb,
 664                                    const struct net_device *in,
 665                                    const struct net_device *out,
 666                                    int (*okfn)(struct sk_buff *))
 667 {
 668         struct rtable *rt = skb_rtable(skb);
 669
 670         if (rt && rt == bridge_parent_rtable(in))
 671                 skb_dst_drop(skb);
 672
 673         return NF_ACCEPT;
 674 }
 675
 676 /* PF_BRIDGE/FORWARD *************************************************/
 677 static int br_nf_forward_finish(struct sk_buff *skb)
 678 {
 679         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 680         struct net_device *in;
 681
 682         if (skb->protocol != htons(ETH_P_ARP) && !IS_VLAN_ARP(skb)) {
 683                 in = nf_bridge->physindev;
 684                 if (nf_bridge->mask & BRNF_PKT_TYPE) {
 685                         skb->pkt_type = PACKET_OTHERHOST;
 686                         nf_bridge->mask ^= BRNF_PKT_TYPE;
 687                 }
 688                 nf_bridge_update_protocol(skb);
 689         } else {
 690                 in = *((struct net_device **)(skb->cb));
 691         }
 692         nf_bridge_push_encap_header(skb);
 693
 694         NF_HOOK_THRESH(NFPROTO_BRIDGE, NF_BR_FORWARD, skb, in,
 695                        skb->dev, br_forward_finish, 1);
 696         return 0;
 697 }
 698
 699 /* This is the 'purely bridged' case.  For IP, we pass the packet to
 700  * netfilter with indev and outdev set to the bridge device,
 701  * but we are still able to filter on the 'real' indev/outdev
 702  * because of the physdev module. For ARP, indev and outdev are the
 703  * bridge ports. */
 704 static unsigned int br_nf_forward_ip(unsigned int hook, struct sk_buff *skb,
 705                                      const struct net_device *in,
 706                                      const struct net_device *out,
 707                                      int (*okfn)(struct sk_buff *))
 708 {
 709         struct nf_bridge_info *nf_bridge;
 710         struct net_device *parent;
 711         u_int8_t pf;
 712
 713         if (!skb->nf_bridge)
 714                 return NF_ACCEPT;
 715
 716         /* Need exclusive nf_bridge_info since we might have multiple
 717          * different physoutdevs. */
 718         if (!nf_bridge_unshare(skb))
 719                 return NF_DROP;
 720
 721         parent = bridge_parent(out);
 722         if (!parent)
 723                 return NF_DROP;
 724
 725         if (skb->protocol == htons(ETH_P_IP) || IS_VLAN_IP(skb) ||
 726             IS_PPPOE_IP(skb))
 727                 pf = PF_INET;
 728         else if (skb->protocol == htons(ETH_P_IPV6) || IS_VLAN_IPV6(skb) ||
 729                  IS_PPPOE_IPV6(skb))
 730                 pf = PF_INET6;
 731         else
 732                 return NF_ACCEPT;
 733
 734         nf_bridge_pull_encap_header(skb);
 735
 736         nf_bridge = skb->nf_bridge;
 737         if (skb->pkt_type == PACKET_OTHERHOST) {
 738                 skb->pkt_type = PACKET_HOST;
 739                 nf_bridge->mask |= BRNF_PKT_TYPE;
 740         }
 741
 742         /* The physdev module checks on this */
 743         nf_bridge->mask |= BRNF_BRIDGED;
 744         nf_bridge->physoutdev = skb->dev;
 745         if (pf == PF_INET)
 746                 skb->protocol = htons(ETH_P_IP);
 747         else
 748                 skb->protocol = htons(ETH_P_IPV6);
 749
 750         NF_HOOK(pf, NF_INET_FORWARD, skb, bridge_parent(in), parent,
 751                 br_nf_forward_finish);
 752
 753         return NF_STOLEN;
 754 }
 755
 756 static unsigned int br_nf_forward_arp(unsigned int hook, struct sk_buff *skb,
 757                                       const struct net_device *in,
 758                                       const struct net_device *out,
 759                                       int (*okfn)(struct sk_buff *))
 760 {
 761         struct net_bridge_port *p;
 762         struct net_bridge *br;
 763         struct net_device **d = (struct net_device **)(skb->cb);
 764
 765         p = br_port_get_rcu(out);
 766         if (p == NULL)
 767                 return NF_ACCEPT;
 768         br = p->br;
 769
 770         if (!brnf_call_arptables && !br->nf_call_arptables)
 771                 return NF_ACCEPT;
 772
 773         if (skb->protocol != htons(ETH_P_ARP)) {
 774                 if (!IS_VLAN_ARP(skb))
 775                         return NF_ACCEPT;
 776                 nf_bridge_pull_encap_header(skb);
 777         }
 778
 779         if (arp_hdr(skb)->ar_pln != 4) {
 780                 if (IS_VLAN_ARP(skb))
 781                         nf_bridge_push_encap_header(skb);
 782                 return NF_ACCEPT;
 783         }
 784         *d = (struct net_device *)in;
 785         NF_HOOK(NFPROTO_ARP, NF_ARP_FORWARD, skb, (struct net_device *)in,
 786                 (struct net_device *)out, br_nf_forward_finish);
 787
 788         return NF_STOLEN;
 789 }
 790
 791 #if defined(CONFIG_NF_CONNTRACK_IPV4) || defined(CONFIG_NF_CONNTRACK_IPV4_MODULE)
 792 static int br_nf_dev_queue_xmit(struct sk_buff *skb)
 793 {
 794         int ret;
 795
 796         if (skb->nfct != NULL && skb->protocol == htons(ETH_P_IP) &&
 797             skb->len + nf_bridge_mtu_reduction(skb) > skb->dev->mtu &&
 798             !skb_is_gso(skb)) {
 799                 if (br_parse_ip_options(skb))
 800                         /* Drop invalid packet */
 801                         return NF_DROP;
 802                 ret = ip_fragment(skb, br_dev_queue_push_xmit);
 803         } else
 804                 ret = br_dev_queue_push_xmit(skb);
 805
 806         return ret;
 807 }
 808 #else
 809 static int br_nf_dev_queue_xmit(struct sk_buff *skb)
 810 {
 811         return br_dev_queue_push_xmit(skb);
 812 }
 813 #endif
 814
 815 /* PF_BRIDGE/POST_ROUTING ********************************************/
 816 static unsigned int br_nf_post_routing(unsigned int hook, struct sk_buff *skb,
 817                                        const struct net_device *in,
 818                                        const struct net_device *out,
 819                                        int (*okfn)(struct sk_buff *))
 820 {
 821         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 822         struct net_device *realoutdev = bridge_parent(skb->dev);
 823         u_int8_t pf;
 824
 825         if (!nf_bridge || !(nf_bridge->mask & BRNF_BRIDGED))
 826                 return NF_ACCEPT;
 827
 828         if (!realoutdev)
 829                 return NF_DROP;
 830
 831         if (skb->protocol == htons(ETH_P_IP) || IS_VLAN_IP(skb) ||
 832             IS_PPPOE_IP(skb))
 833                 pf = PF_INET;
 834         else if (skb->protocol == htons(ETH_P_IPV6) || IS_VLAN_IPV6(skb) ||
 835                  IS_PPPOE_IPV6(skb))
 836                 pf = PF_INET6;
 837         else
 838                 return NF_ACCEPT;
 839
 840         /* We assume any code from br_dev_queue_push_xmit onwards doesn't care
 841          * about the value of skb->pkt_type. */
 842         if (skb->pkt_type == PACKET_OTHERHOST) {
 843                 skb->pkt_type = PACKET_HOST;
 844                 nf_bridge->mask |= BRNF_PKT_TYPE;
 845         }
 846
 847         nf_bridge_pull_encap_header(skb);
 848         nf_bridge_save_header(skb);
 849         if (pf == PF_INET)
 850                 skb->protocol = htons(ETH_P_IP);
 851         else
 852                 skb->protocol = htons(ETH_P_IPV6);
 853
 854         NF_HOOK(pf, NF_INET_POST_ROUTING, skb, NULL, realoutdev,
 855                 br_nf_dev_queue_xmit);
 856
 857         return NF_STOLEN;
 858 }
 859
 860 /* IP/SABOTAGE *****************************************************/
 861 /* Don't hand locally destined packets to PF_INET(6)/PRE_ROUTING
 862  * for the second time. */
 863 static unsigned int ip_sabotage_in(unsigned int hook, struct sk_buff *skb,
 864                                    const struct net_device *in,
 865                                    const struct net_device *out,
 866                                    int (*okfn)(struct sk_buff *))
 867 {
 868         if (skb->nf_bridge &&
 869             !(skb->nf_bridge->mask & BRNF_NF_BRIDGE_PREROUTING)) {
 870                 return NF_STOP;
 871         }
 872
 873         return NF_ACCEPT;
 874 }
 875
 876 /* For br_nf_post_routing, we need (prio = NF_BR_PRI_LAST), because
 877  * br_dev_queue_push_xmit is called afterwards */
 878 static struct nf_hook_ops br_nf_ops[] __read_mostly = {
 879         {
 880                 .hook = br_nf_pre_routing,
 881                 .owner = THIS_MODULE,
 882                 .pf = PF_BRIDGE,
 883                 .hooknum = NF_BR_PRE_ROUTING,
 884                 .priority = NF_BR_PRI_BRNF,
 885         },
 886         {
 887                 .hook = br_nf_local_in,
 888                 .owner = THIS_MODULE,
 889                 .pf = PF_BRIDGE,
 890                 .hooknum = NF_BR_LOCAL_IN,
 891                 .priority = NF_BR_PRI_BRNF,
 892         },
 893         {
 894                 .hook = br_nf_forward_ip,
 895                 .owner = THIS_MODULE,
 896                 .pf = PF_BRIDGE,
 897                 .hooknum = NF_BR_FORWARD,
 898                 .priority = NF_BR_PRI_BRNF - 1,
 899         },
 900         {
 901                 .hook = br_nf_forward_arp,
 902                 .owner = THIS_MODULE,
 903                 .pf = PF_BRIDGE,
 904                 .hooknum = NF_BR_FORWARD,
 905                 .priority = NF_BR_PRI_BRNF,
 906         },
 907         {
 908                 .hook = br_nf_post_routing,
 909                 .owner = THIS_MODULE,
 910                 .pf = PF_BRIDGE,
 911                 .hooknum = NF_BR_POST_ROUTING,
 912                 .priority = NF_BR_PRI_LAST,
 913         },
 914         {
 915                 .hook = ip_sabotage_in,
 916                 .owner = THIS_MODULE,
 917                 .pf = PF_INET,
 918                 .hooknum = NF_INET_PRE_ROUTING,
 919                 .priority = NF_IP_PRI_FIRST,
 920         },
 921         {
 922                 .hook = ip_sabotage_in,
 923                 .owner = THIS_MODULE,
 924                 .pf = PF_INET6,
 925                 .hooknum = NF_INET_PRE_ROUTING,
 926                 .priority = NF_IP6_PRI_FIRST,
 927         },
 928 };
 929
 930 #ifdef CONFIG_SYSCTL
 931 static
 932 int brnf_sysctl_call_tables(ctl_table * ctl, int write,
 933                             void __user * buffer, size_t * lenp, loff_t * ppos)
 934 {
 935         int ret;
 936
 937         ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
 938
 939         if (write && *(int *)(ctl->data))
 940                 *(int *)(ctl->data) = 1;
 941         return ret;
 942 }
 943
 944 static ctl_table brnf_table[] = {
 945         {
 946                 .procname       = "bridge-nf-call-arptables",
 947                 .data           = &brnf_call_arptables,
 948                 .maxlen         = sizeof(int),
 949                 .mode           = 0644,
 950                 .proc_handler   = brnf_sysctl_call_tables,
 951         },
 952         {
 953                 .procname       = "bridge-nf-call-iptables",
 954                 .data           = &brnf_call_iptables,
 955                 .maxlen         = sizeof(int),
 956                 .mode           = 0644,
 957                 .proc_handler   = brnf_sysctl_call_tables,
 958         },
 959         {
 960                 .procname       = "bridge-nf-call-ip6tables",
 961                 .data           = &brnf_call_ip6tables,
 962                 .maxlen         = sizeof(int),
 963                 .mode           = 0644,
 964                 .proc_handler   = brnf_sysctl_call_tables,
 965         },
 966         {
 967                 .procname       = "bridge-nf-filter-vlan-tagged",
 968                 .data           = &brnf_filter_vlan_tagged,
 969                 .maxlen         = sizeof(int),
 970                 .mode           = 0644,
 971                 .proc_handler   = brnf_sysctl_call_tables,
 972         },
 973         {
 974                 .procname       = "bridge-nf-filter-pppoe-tagged",
 975                 .data           = &brnf_filter_pppoe_tagged,
 976                 .maxlen         = sizeof(int),
 977                 .mode           = 0644,
 978                 .proc_handler   = brnf_sysctl_call_tables,
 979         },
 980         { }
 981 };
 982
 983 static struct ctl_path brnf_path[] = {
 984         { .procname = "net", },
 985         { .procname = "bridge", },
 986         { }
 987 };
 988 #endif
 989
 990 int __init br_netfilter_init(void)
 991 {
 992         int ret;
 993
 994         ret = dst_entries_init(&fake_dst_ops);
 995         if (ret < 0)
 996                 return ret;
 997
 998         ret = nf_register_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
 999         if (ret < 0) {
1000                 dst_entries_destroy(&fake_dst_ops);
1001                 return ret;
1002         }
1003 #ifdef CONFIG_SYSCTL
1004         brnf_sysctl_header = register_sysctl_paths(brnf_path, brnf_table);
1005         if (brnf_sysctl_header == NULL) {
1006                 printk(KERN_WARNING
1007                        "br_netfilter: can't register to sysctl.\n");
1008                 nf_unregister_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
1009                 dst_entries_destroy(&fake_dst_ops);
1010                 return -ENOMEM;
1011         }
1012 #endif
1013         printk(KERN_NOTICE "Bridge firewalling registered\n");
1014         return 0;
1015 }
1016
1017 void br_netfilter_fini(void)
1018 {
1019         nf_unregister_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
1020 #ifdef CONFIG_SYSCTL
1021         unregister_sysctl_table(brnf_sysctl_header);
1022 #endif
1023         dst_entries_destroy(&fake_dst_ops);
1024 }