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 #include <net/netfilter/br_netfilter.h>
  39
  40 #include <asm/uaccess.h>
  41 #include "br_private.h"
  42 #ifdef CONFIG_SYSCTL
  43 #include <linux/sysctl.h>
  44 #endif
  45
  46 #define skb_origaddr(skb)        (((struct bridge_skb_cb *) \
  47                                  (skb->nf_bridge->data))->daddr.ipv4)
  48 #define store_orig_dstaddr(skb)  (skb_origaddr(skb) = ip_hdr(skb)->daddr)
  49 #define dnat_took_place(skb)     (skb_origaddr(skb) != ip_hdr(skb)->daddr)
  50
  51 #ifdef CONFIG_SYSCTL
  52 static struct ctl_table_header *brnf_sysctl_header;
  53 static int brnf_call_iptables __read_mostly = 1;
  54 static int brnf_call_ip6tables __read_mostly = 1;
  55 static int brnf_call_arptables __read_mostly = 1;
  56 static int brnf_filter_vlan_tagged __read_mostly = 0;
  57 static int brnf_filter_pppoe_tagged __read_mostly = 0;
  58 static int brnf_pass_vlan_indev __read_mostly = 0;
  59 #else
  60 #define brnf_call_iptables 1
  61 #define brnf_call_ip6tables 1
  62 #define brnf_call_arptables 1
  63 #define brnf_filter_vlan_tagged 0
  64 #define brnf_filter_pppoe_tagged 0
  65 #define brnf_pass_vlan_indev 0
  66 #endif
  67
  68 #define IS_IP(skb) \
  69         (!vlan_tx_tag_present(skb) && skb->protocol == htons(ETH_P_IP))
  70
  71 #define IS_IPV6(skb) \
  72         (!vlan_tx_tag_present(skb) && skb->protocol == htons(ETH_P_IPV6))
  73
  74 #define IS_ARP(skb) \
  75         (!vlan_tx_tag_present(skb) && skb->protocol == htons(ETH_P_ARP))
  76
  77 static inline __be16 vlan_proto(const struct sk_buff *skb)
  78 {
  79         if (vlan_tx_tag_present(skb))
  80                 return skb->protocol;
  81         else if (skb->protocol == htons(ETH_P_8021Q))
  82                 return vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
  83         else
  84                 return 0;
  85 }
  86
  87 #define IS_VLAN_IP(skb) \
  88         (vlan_proto(skb) == htons(ETH_P_IP) && \
  89          brnf_filter_vlan_tagged)
  90
  91 #define IS_VLAN_IPV6(skb) \
  92         (vlan_proto(skb) == htons(ETH_P_IPV6) && \
  93          brnf_filter_vlan_tagged)
  94
  95 #define IS_VLAN_ARP(skb) \
  96         (vlan_proto(skb) == htons(ETH_P_ARP) && \
  97          brnf_filter_vlan_tagged)
  98
  99 static inline __be16 pppoe_proto(const struct sk_buff *skb)
 100 {
 101         return *((__be16 *)(skb_mac_header(skb) + ETH_HLEN +
 102                             sizeof(struct pppoe_hdr)));
 103 }
 104
 105 #define IS_PPPOE_IP(skb) \
 106         (skb->protocol == htons(ETH_P_PPP_SES) && \
 107          pppoe_proto(skb) == htons(PPP_IP) && \
 108          brnf_filter_pppoe_tagged)
 109
 110 #define IS_PPPOE_IPV6(skb) \
 111         (skb->protocol == htons(ETH_P_PPP_SES) && \
 112          pppoe_proto(skb) == htons(PPP_IPV6) && \
 113          brnf_filter_pppoe_tagged)
 114
 115 static inline struct rtable *bridge_parent_rtable(const struct net_device *dev)
 116 {
 117         struct net_bridge_port *port;
 118
 119         port = br_port_get_rcu(dev);
 120         return port ? &port->br->fake_rtable : NULL;
 121 }
 122
 123 static inline struct net_device *bridge_parent(const struct net_device *dev)
 124 {
 125         struct net_bridge_port *port;
 126
 127         port = br_port_get_rcu(dev);
 128         return port ? port->br->dev : NULL;
 129 }
 130
 131 static inline struct nf_bridge_info *nf_bridge_alloc(struct sk_buff *skb)
 132 {
 133         skb->nf_bridge = kzalloc(sizeof(struct nf_bridge_info), GFP_ATOMIC);
 134         if (likely(skb->nf_bridge))
 135                 atomic_set(&(skb->nf_bridge->use), 1);
 136
 137         return skb->nf_bridge;
 138 }
 139
 140 static inline struct nf_bridge_info *nf_bridge_unshare(struct sk_buff *skb)
 141 {
 142         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 143
 144         if (atomic_read(&nf_bridge->use) > 1) {
 145                 struct nf_bridge_info *tmp = nf_bridge_alloc(skb);
 146
 147                 if (tmp) {
 148                         memcpy(tmp, nf_bridge, sizeof(struct nf_bridge_info));
 149                         atomic_set(&tmp->use, 1);
 150                 }
 151                 nf_bridge_put(nf_bridge);
 152                 nf_bridge = tmp;
 153         }
 154         return nf_bridge;
 155 }
 156
 157 static inline void nf_bridge_push_encap_header(struct sk_buff *skb)
 158 {
 159         unsigned int len = nf_bridge_encap_header_len(skb);
 160
 161         skb_push(skb, len);
 162         skb->network_header -= len;
 163 }
 164
 165 static inline void nf_bridge_pull_encap_header(struct sk_buff *skb)
 166 {
 167         unsigned int len = nf_bridge_encap_header_len(skb);
 168
 169         skb_pull(skb, len);
 170         skb->network_header += len;
 171 }
 172
 173 static inline void nf_bridge_pull_encap_header_rcsum(struct sk_buff *skb)
 174 {
 175         unsigned int len = nf_bridge_encap_header_len(skb);
 176
 177         skb_pull_rcsum(skb, len);
 178         skb->network_header += len;
 179 }
 180
 181 static inline void nf_bridge_save_header(struct sk_buff *skb)
 182 {
 183         int header_size = ETH_HLEN + nf_bridge_encap_header_len(skb);
 184
 185         skb_copy_from_linear_data_offset(skb, -header_size,
 186                                          skb->nf_bridge->data, header_size);
 187 }
 188
 189 /* When handing a packet over to the IP layer
 190  * check whether we have a skb that is in the
 191  * expected format
 192  */
 193
 194 static int br_parse_ip_options(struct sk_buff *skb)
 195 {
 196         struct ip_options *opt;
 197         const struct iphdr *iph;
 198         struct net_device *dev = skb->dev;
 199         u32 len;
 200
 201         if (!pskb_may_pull(skb, sizeof(struct iphdr)))
 202                 goto inhdr_error;
 203
 204         iph = ip_hdr(skb);
 205         opt = &(IPCB(skb)->opt);
 206
 207         /* Basic sanity checks */
 208         if (iph->ihl < 5 || iph->version != 4)
 209                 goto inhdr_error;
 210
 211         if (!pskb_may_pull(skb, iph->ihl*4))
 212                 goto inhdr_error;
 213
 214         iph = ip_hdr(skb);
 215         if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl)))
 216                 goto inhdr_error;
 217
 218         len = ntohs(iph->tot_len);
 219         if (skb->len < len) {
 220                 IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INTRUNCATEDPKTS);
 221                 goto drop;
 222         } else if (len < (iph->ihl*4))
 223                 goto inhdr_error;
 224
 225         if (pskb_trim_rcsum(skb, len)) {
 226                 IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INDISCARDS);
 227                 goto drop;
 228         }
 229
 230         memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
 231         if (iph->ihl == 5)
 232                 return 0;
 233
 234         opt->optlen = iph->ihl*4 - sizeof(struct iphdr);
 235         if (ip_options_compile(dev_net(dev), opt, skb))
 236                 goto inhdr_error;
 237
 238         /* Check correct handling of SRR option */
 239         if (unlikely(opt->srr)) {
 240                 struct in_device *in_dev = __in_dev_get_rcu(dev);
 241                 if (in_dev && !IN_DEV_SOURCE_ROUTE(in_dev))
 242                         goto drop;
 243
 244                 if (ip_options_rcv_srr(skb))
 245                         goto drop;
 246         }
 247
 248         return 0;
 249
 250 inhdr_error:
 251         IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INHDRERRORS);
 252 drop:
 253         return -1;
 254 }
 255
 256 /* PF_BRIDGE/PRE_ROUTING *********************************************/
 257 /* Undo the changes made for ip6tables PREROUTING and continue the
 258  * bridge PRE_ROUTING hook. */
 259 static int br_nf_pre_routing_finish_ipv6(struct sk_buff *skb)
 260 {
 261         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 262         struct rtable *rt;
 263
 264         if (nf_bridge->mask & BRNF_PKT_TYPE) {
 265                 skb->pkt_type = PACKET_OTHERHOST;
 266                 nf_bridge->mask ^= BRNF_PKT_TYPE;
 267         }
 268         nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
 269
 270         rt = bridge_parent_rtable(nf_bridge->physindev);
 271         if (!rt) {
 272                 kfree_skb(skb);
 273                 return 0;
 274         }
 275         skb_dst_set_noref(skb, &rt->dst);
 276
 277         skb->dev = nf_bridge->physindev;
 278         nf_bridge_update_protocol(skb);
 279         nf_bridge_push_encap_header(skb);
 280         NF_HOOK_THRESH(NFPROTO_BRIDGE, NF_BR_PRE_ROUTING, skb, skb->dev, NULL,
 281                        br_handle_frame_finish, 1);
 282
 283         return 0;
 284 }
 285
 286 /* Obtain the correct destination MAC address, while preserving the original
 287  * source MAC address. If we already know this address, we just copy it. If we
 288  * don't, we use the neighbour framework to find out. In both cases, we make
 289  * sure that br_handle_frame_finish() is called afterwards.
 290  */
 291 static int br_nf_pre_routing_finish_bridge(struct sk_buff *skb)
 292 {
 293         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 294         struct neighbour *neigh;
 295         struct dst_entry *dst;
 296
 297         skb->dev = bridge_parent(skb->dev);
 298         if (!skb->dev)
 299                 goto free_skb;
 300         dst = skb_dst(skb);
 301         neigh = dst_neigh_lookup_skb(dst, skb);
 302         if (neigh) {
 303                 int ret;
 304
 305                 if (neigh->hh.hh_len) {
 306                         neigh_hh_bridge(&neigh->hh, skb);
 307                         skb->dev = nf_bridge->physindev;
 308                         ret = br_handle_frame_finish(skb);
 309                 } else {
 310                         /* the neighbour function below overwrites the complete
 311                          * MAC header, so we save the Ethernet source address and
 312                          * protocol number.
 313                          */
 314                         skb_copy_from_linear_data_offset(skb,
 315                                                          -(ETH_HLEN-ETH_ALEN),
 316                                                          skb->nf_bridge->data,
 317                                                          ETH_HLEN-ETH_ALEN);
 318                         /* tell br_dev_xmit to continue with forwarding */
 319                         nf_bridge->mask |= BRNF_BRIDGED_DNAT;
 320                         /* FIXME Need to refragment */
 321                         ret = neigh->output(neigh, skb);
 322                 }
 323                 neigh_release(neigh);
 324                 return ret;
 325         }
 326 free_skb:
 327         kfree_skb(skb);
 328         return 0;
 329 }
 330
 331 /* This requires some explaining. If DNAT has taken place,
 332  * we will need to fix up the destination Ethernet address.
 333  *
 334  * There are two cases to consider:
 335  * 1. The packet was DNAT'ed to a device in the same bridge
 336  *    port group as it was received on. We can still bridge
 337  *    the packet.
 338  * 2. The packet was DNAT'ed to a different device, either
 339  *    a non-bridged device or another bridge port group.
 340  *    The packet will need to be routed.
 341  *
 342  * The correct way of distinguishing between these two cases is to
 343  * call ip_route_input() and to look at skb->dst->dev, which is
 344  * changed to the destination device if ip_route_input() succeeds.
 345  *
 346  * Let's first consider the case that ip_route_input() succeeds:
 347  *
 348  * If the output device equals the logical bridge device the packet
 349  * came in on, we can consider this bridging. The corresponding MAC
 350  * address will be obtained in br_nf_pre_routing_finish_bridge.
 351  * Otherwise, the packet is considered to be routed and we just
 352  * change the destination MAC address so that the packet will
 353  * later be passed up to the IP stack to be routed. For a redirected
 354  * packet, ip_route_input() will give back the localhost as output device,
 355  * which differs from the bridge device.
 356  *
 357  * Let's now consider the case that ip_route_input() fails:
 358  *
 359  * This can be because the destination address is martian, in which case
 360  * the packet will be dropped.
 361  * If IP forwarding is disabled, ip_route_input() will fail, while
 362  * ip_route_output_key() can return success. The source
 363  * address for ip_route_output_key() is set to zero, so ip_route_output_key()
 364  * thinks we're handling a locally generated packet and won't care
 365  * if IP forwarding is enabled. If the output device equals the logical bridge
 366  * device, we proceed as if ip_route_input() succeeded. If it differs from the
 367  * logical bridge port or if ip_route_output_key() fails we drop the packet.
 368  */
 369 static int br_nf_pre_routing_finish(struct sk_buff *skb)
 370 {
 371         struct net_device *dev = skb->dev;
 372         struct iphdr *iph = ip_hdr(skb);
 373         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 374         struct rtable *rt;
 375         int err;
 376         int frag_max_size;
 377
 378         frag_max_size = IPCB(skb)->frag_max_size;
 379         BR_INPUT_SKB_CB(skb)->frag_max_size = frag_max_size;
 380
 381         if (nf_bridge->mask & BRNF_PKT_TYPE) {
 382                 skb->pkt_type = PACKET_OTHERHOST;
 383                 nf_bridge->mask ^= BRNF_PKT_TYPE;
 384         }
 385         nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
 386         if (dnat_took_place(skb)) {
 387                 if ((err = ip_route_input(skb, iph->daddr, iph->saddr, iph->tos, dev))) {
 388                         struct in_device *in_dev = __in_dev_get_rcu(dev);
 389
 390                         /* If err equals -EHOSTUNREACH the error is due to a
 391                          * martian destination or due to the fact that
 392                          * forwarding is disabled. For most martian packets,
 393                          * ip_route_output_key() will fail. It won't fail for 2 types of
 394                          * martian destinations: loopback destinations and destination
 395                          * 0.0.0.0. In both cases the packet will be dropped because the
 396                          * destination is the loopback device and not the bridge. */
 397                         if (err != -EHOSTUNREACH || !in_dev || IN_DEV_FORWARD(in_dev))
 398                                 goto free_skb;
 399
 400                         rt = ip_route_output(dev_net(dev), iph->daddr, 0,
 401                                              RT_TOS(iph->tos), 0);
 402                         if (!IS_ERR(rt)) {
 403                                 /* - Bridged-and-DNAT'ed traffic doesn't
 404                                  *   require ip_forwarding. */
 405                                 if (rt->dst.dev == dev) {
 406                                         skb_dst_set(skb, &rt->dst);
 407                                         goto bridged_dnat;
 408                                 }
 409                                 ip_rt_put(rt);
 410                         }
 411 free_skb:
 412                         kfree_skb(skb);
 413                         return 0;
 414                 } else {
 415                         if (skb_dst(skb)->dev == dev) {
 416 bridged_dnat:
 417                                 skb->dev = nf_bridge->physindev;
 418                                 nf_bridge_update_protocol(skb);
 419                                 nf_bridge_push_encap_header(skb);
 420                                 NF_HOOK_THRESH(NFPROTO_BRIDGE,
 421                                                NF_BR_PRE_ROUTING,
 422                                                skb, skb->dev, NULL,
 423                                                br_nf_pre_routing_finish_bridge,
 424                                                1);
 425                                 return 0;
 426                         }
 427                         ether_addr_copy(eth_hdr(skb)->h_dest, dev->dev_addr);
 428                         skb->pkt_type = PACKET_HOST;
 429                 }
 430         } else {
 431                 rt = bridge_parent_rtable(nf_bridge->physindev);
 432                 if (!rt) {
 433                         kfree_skb(skb);
 434                         return 0;
 435                 }
 436                 skb_dst_set_noref(skb, &rt->dst);
 437         }
 438
 439         skb->dev = nf_bridge->physindev;
 440         nf_bridge_update_protocol(skb);
 441         nf_bridge_push_encap_header(skb);
 442         NF_HOOK_THRESH(NFPROTO_BRIDGE, NF_BR_PRE_ROUTING, skb, skb->dev, NULL,
 443                        br_handle_frame_finish, 1);
 444
 445         return 0;
 446 }
 447
 448 static struct net_device *brnf_get_logical_dev(struct sk_buff *skb, const struct net_device *dev)
 449 {
 450         struct net_device *vlan, *br;
 451
 452         br = bridge_parent(dev);
 453         if (brnf_pass_vlan_indev == 0 || !vlan_tx_tag_present(skb))
 454                 return br;
 455
 456         vlan = __vlan_find_dev_deep_rcu(br, skb->vlan_proto,
 457                                     vlan_tx_tag_get(skb) & VLAN_VID_MASK);
 458
 459         return vlan ? vlan : br;
 460 }
 461
 462 /* Some common code for IPv4/IPv6 */
 463 static struct net_device *setup_pre_routing(struct sk_buff *skb)
 464 {
 465         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 466
 467         if (skb->pkt_type == PACKET_OTHERHOST) {
 468                 skb->pkt_type = PACKET_HOST;
 469                 nf_bridge->mask |= BRNF_PKT_TYPE;
 470         }
 471
 472         nf_bridge->mask |= BRNF_NF_BRIDGE_PREROUTING;
 473         nf_bridge->physindev = skb->dev;
 474         skb->dev = brnf_get_logical_dev(skb, skb->dev);
 475         if (skb->protocol == htons(ETH_P_8021Q))
 476                 nf_bridge->mask |= BRNF_8021Q;
 477         else if (skb->protocol == htons(ETH_P_PPP_SES))
 478                 nf_bridge->mask |= BRNF_PPPoE;
 479
 480         /* Must drop socket now because of tproxy. */
 481         skb_orphan(skb);
 482         return skb->dev;
 483 }
 484
 485 /* We only check the length. A bridge shouldn't do any hop-by-hop stuff anyway */
 486 static int check_hbh_len(struct sk_buff *skb)
 487 {
 488         unsigned char *raw = (u8 *)(ipv6_hdr(skb) + 1);
 489         u32 pkt_len;
 490         const unsigned char *nh = skb_network_header(skb);
 491         int off = raw - nh;
 492         int len = (raw[1] + 1) << 3;
 493
 494         if ((raw + len) - skb->data > skb_headlen(skb))
 495                 goto bad;
 496
 497         off += 2;
 498         len -= 2;
 499
 500         while (len > 0) {
 501                 int optlen = nh[off + 1] + 2;
 502
 503                 switch (nh[off]) {
 504                 case IPV6_TLV_PAD1:
 505                         optlen = 1;
 506                         break;
 507
 508                 case IPV6_TLV_PADN:
 509                         break;
 510
 511                 case IPV6_TLV_JUMBO:
 512                         if (nh[off + 1] != 4 || (off & 3) != 2)
 513                                 goto bad;
 514                         pkt_len = ntohl(*(__be32 *) (nh + off + 2));
 515                         if (pkt_len <= IPV6_MAXPLEN ||
 516                             ipv6_hdr(skb)->payload_len)
 517                                 goto bad;
 518                         if (pkt_len > skb->len - sizeof(struct ipv6hdr))
 519                                 goto bad;
 520                         if (pskb_trim_rcsum(skb,
 521                                             pkt_len + sizeof(struct ipv6hdr)))
 522                                 goto bad;
 523                         nh = skb_network_header(skb);
 524                         break;
 525                 default:
 526                         if (optlen > len)
 527                                 goto bad;
 528                         break;
 529                 }
 530                 off += optlen;
 531                 len -= optlen;
 532         }
 533         if (len == 0)
 534                 return 0;
 535 bad:
 536         return -1;
 537
 538 }
 539
 540 /* Replicate the checks that IPv6 does on packet reception and pass the packet
 541  * to ip6tables, which doesn't support NAT, so things are fairly simple. */
 542 static unsigned int br_nf_pre_routing_ipv6(const struct nf_hook_ops *ops,
 543                                            struct sk_buff *skb,
 544                                            const struct net_device *in,
 545                                            const struct net_device *out,
 546                                            int (*okfn)(struct sk_buff *))
 547 {
 548         const struct ipv6hdr *hdr;
 549         u32 pkt_len;
 550
 551         if (skb->len < sizeof(struct ipv6hdr))
 552                 return NF_DROP;
 553
 554         if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
 555                 return NF_DROP;
 556
 557         hdr = ipv6_hdr(skb);
 558
 559         if (hdr->version != 6)
 560                 return NF_DROP;
 561
 562         pkt_len = ntohs(hdr->payload_len);
 563
 564         if (pkt_len || hdr->nexthdr != NEXTHDR_HOP) {
 565                 if (pkt_len + sizeof(struct ipv6hdr) > skb->len)
 566                         return NF_DROP;
 567                 if (pskb_trim_rcsum(skb, pkt_len + sizeof(struct ipv6hdr)))
 568                         return NF_DROP;
 569         }
 570         if (hdr->nexthdr == NEXTHDR_HOP && check_hbh_len(skb))
 571                 return NF_DROP;
 572
 573         nf_bridge_put(skb->nf_bridge);
 574         if (!nf_bridge_alloc(skb))
 575                 return NF_DROP;
 576         if (!setup_pre_routing(skb))
 577                 return NF_DROP;
 578
 579         skb->protocol = htons(ETH_P_IPV6);
 580         NF_HOOK(NFPROTO_IPV6, NF_INET_PRE_ROUTING, skb, skb->dev, NULL,
 581                 br_nf_pre_routing_finish_ipv6);
 582
 583         return NF_STOLEN;
 584 }
 585
 586 /* Direct IPv6 traffic to br_nf_pre_routing_ipv6.
 587  * Replicate the checks that IPv4 does on packet reception.
 588  * Set skb->dev to the bridge device (i.e. parent of the
 589  * receiving device) to make netfilter happy, the REDIRECT
 590  * target in particular.  Save the original destination IP
 591  * address to be able to detect DNAT afterwards. */
 592 static unsigned int br_nf_pre_routing(const struct nf_hook_ops *ops,
 593                                       struct sk_buff *skb,
 594                                       const struct net_device *in,
 595                                       const struct net_device *out,
 596                                       int (*okfn)(struct sk_buff *))
 597 {
 598         struct net_bridge_port *p;
 599         struct net_bridge *br;
 600         __u32 len = nf_bridge_encap_header_len(skb);
 601
 602         if (unlikely(!pskb_may_pull(skb, len)))
 603                 return NF_DROP;
 604
 605         p = br_port_get_rcu(in);
 606         if (p == NULL)
 607                 return NF_DROP;
 608         br = p->br;
 609
 610         if (IS_IPV6(skb) || IS_VLAN_IPV6(skb) || IS_PPPOE_IPV6(skb)) {
 611                 if (!brnf_call_ip6tables && !br->nf_call_ip6tables)
 612                         return NF_ACCEPT;
 613
 614                 nf_bridge_pull_encap_header_rcsum(skb);
 615                 return br_nf_pre_routing_ipv6(ops, skb, in, out, okfn);
 616         }
 617
 618         if (!brnf_call_iptables && !br->nf_call_iptables)
 619                 return NF_ACCEPT;
 620
 621         if (!IS_IP(skb) && !IS_VLAN_IP(skb) && !IS_PPPOE_IP(skb))
 622                 return NF_ACCEPT;
 623
 624         nf_bridge_pull_encap_header_rcsum(skb);
 625
 626         if (br_parse_ip_options(skb))
 627                 return NF_DROP;
 628
 629         nf_bridge_put(skb->nf_bridge);
 630         if (!nf_bridge_alloc(skb))
 631                 return NF_DROP;
 632         if (!setup_pre_routing(skb))
 633                 return NF_DROP;
 634         store_orig_dstaddr(skb);
 635         skb->protocol = htons(ETH_P_IP);
 636
 637         NF_HOOK(NFPROTO_IPV4, NF_INET_PRE_ROUTING, skb, skb->dev, NULL,
 638                 br_nf_pre_routing_finish);
 639
 640         return NF_STOLEN;
 641 }
 642
 643
 644 /* PF_BRIDGE/LOCAL_IN ************************************************/
 645 /* The packet is locally destined, which requires a real
 646  * dst_entry, so detach the fake one.  On the way up, the
 647  * packet would pass through PRE_ROUTING again (which already
 648  * took place when the packet entered the bridge), but we
 649  * register an IPv4 PRE_ROUTING 'sabotage' hook that will
 650  * prevent this from happening. */
 651 static unsigned int br_nf_local_in(const struct nf_hook_ops *ops,
 652                                    struct sk_buff *skb,
 653                                    const struct net_device *in,
 654                                    const struct net_device *out,
 655                                    int (*okfn)(struct sk_buff *))
 656 {
 657         br_drop_fake_rtable(skb);
 658         return NF_ACCEPT;
 659 }
 660
 661 /* PF_BRIDGE/FORWARD *************************************************/
 662 static int br_nf_forward_finish(struct sk_buff *skb)
 663 {
 664         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 665         struct net_device *in;
 666
 667         if (!IS_ARP(skb) && !IS_VLAN_ARP(skb)) {
 668                 in = nf_bridge->physindev;
 669                 if (nf_bridge->mask & BRNF_PKT_TYPE) {
 670                         skb->pkt_type = PACKET_OTHERHOST;
 671                         nf_bridge->mask ^= BRNF_PKT_TYPE;
 672                 }
 673                 nf_bridge_update_protocol(skb);
 674         } else {
 675                 in = *((struct net_device **)(skb->cb));
 676         }
 677         nf_bridge_push_encap_header(skb);
 678
 679         NF_HOOK_THRESH(NFPROTO_BRIDGE, NF_BR_FORWARD, skb, in,
 680                        skb->dev, br_forward_finish, 1);
 681         return 0;
 682 }
 683
 684
 685 /* This is the 'purely bridged' case.  For IP, we pass the packet to
 686  * netfilter with indev and outdev set to the bridge device,
 687  * but we are still able to filter on the 'real' indev/outdev
 688  * because of the physdev module. For ARP, indev and outdev are the
 689  * bridge ports. */
 690 static unsigned int br_nf_forward_ip(const struct nf_hook_ops *ops,
 691                                      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 nf_bridge_info *nf_bridge;
 697         struct net_device *parent;
 698         u_int8_t pf;
 699
 700         if (!skb->nf_bridge)
 701                 return NF_ACCEPT;
 702
 703         /* Need exclusive nf_bridge_info since we might have multiple
 704          * different physoutdevs. */
 705         if (!nf_bridge_unshare(skb))
 706                 return NF_DROP;
 707
 708         parent = bridge_parent(out);
 709         if (!parent)
 710                 return NF_DROP;
 711
 712         if (IS_IP(skb) || IS_VLAN_IP(skb) || IS_PPPOE_IP(skb))
 713                 pf = NFPROTO_IPV4;
 714         else if (IS_IPV6(skb) || IS_VLAN_IPV6(skb) || IS_PPPOE_IPV6(skb))
 715                 pf = NFPROTO_IPV6;
 716         else
 717                 return NF_ACCEPT;
 718
 719         nf_bridge_pull_encap_header(skb);
 720
 721         nf_bridge = skb->nf_bridge;
 722         if (skb->pkt_type == PACKET_OTHERHOST) {
 723                 skb->pkt_type = PACKET_HOST;
 724                 nf_bridge->mask |= BRNF_PKT_TYPE;
 725         }
 726
 727         if (pf == NFPROTO_IPV4 && br_parse_ip_options(skb))
 728                 return NF_DROP;
 729
 730         /* The physdev module checks on this */
 731         nf_bridge->mask |= BRNF_BRIDGED;
 732         nf_bridge->physoutdev = skb->dev;
 733         if (pf == NFPROTO_IPV4)
 734                 skb->protocol = htons(ETH_P_IP);
 735         else
 736                 skb->protocol = htons(ETH_P_IPV6);
 737
 738         NF_HOOK(pf, NF_INET_FORWARD, skb, brnf_get_logical_dev(skb, in), parent,
 739                 br_nf_forward_finish);
 740
 741         return NF_STOLEN;
 742 }
 743
 744 static unsigned int br_nf_forward_arp(const struct nf_hook_ops *ops,
 745                                       struct sk_buff *skb,
 746                                       const struct net_device *in,
 747                                       const struct net_device *out,
 748                                       int (*okfn)(struct sk_buff *))
 749 {
 750         struct net_bridge_port *p;
 751         struct net_bridge *br;
 752         struct net_device **d = (struct net_device **)(skb->cb);
 753
 754         p = br_port_get_rcu(out);
 755         if (p == NULL)
 756                 return NF_ACCEPT;
 757         br = p->br;
 758
 759         if (!brnf_call_arptables && !br->nf_call_arptables)
 760                 return NF_ACCEPT;
 761
 762         if (!IS_ARP(skb)) {
 763                 if (!IS_VLAN_ARP(skb))
 764                         return NF_ACCEPT;
 765                 nf_bridge_pull_encap_header(skb);
 766         }
 767
 768         if (arp_hdr(skb)->ar_pln != 4) {
 769                 if (IS_VLAN_ARP(skb))
 770                         nf_bridge_push_encap_header(skb);
 771                 return NF_ACCEPT;
 772         }
 773         *d = (struct net_device *)in;
 774         NF_HOOK(NFPROTO_ARP, NF_ARP_FORWARD, skb, (struct net_device *)in,
 775                 (struct net_device *)out, br_nf_forward_finish);
 776
 777         return NF_STOLEN;
 778 }
 779
 780 #if IS_ENABLED(CONFIG_NF_DEFRAG_IPV4)
 781 static int br_nf_dev_queue_xmit(struct sk_buff *skb)
 782 {
 783         int ret;
 784         int frag_max_size;
 785
 786         /* This is wrong! We should preserve the original fragment
 787          * boundaries by preserving frag_list rather than refragmenting.
 788          */
 789         if (skb->protocol == htons(ETH_P_IP) &&
 790             skb->len + nf_bridge_mtu_reduction(skb) > skb->dev->mtu &&
 791             !skb_is_gso(skb)) {
 792                 frag_max_size = BR_INPUT_SKB_CB(skb)->frag_max_size;
 793                 if (br_parse_ip_options(skb))
 794                         /* Drop invalid packet */
 795                         return NF_DROP;
 796                 IPCB(skb)->frag_max_size = frag_max_size;
 797                 ret = ip_fragment(skb, br_dev_queue_push_xmit);
 798         } else
 799                 ret = br_dev_queue_push_xmit(skb);
 800
 801         return ret;
 802 }
 803 #else
 804 static int br_nf_dev_queue_xmit(struct sk_buff *skb)
 805 {
 806         return br_dev_queue_push_xmit(skb);
 807 }
 808 #endif
 809
 810 /* PF_BRIDGE/POST_ROUTING ********************************************/
 811 static unsigned int br_nf_post_routing(const struct nf_hook_ops *ops,
 812                                        struct sk_buff *skb,
 813                                        const struct net_device *in,
 814                                        const struct net_device *out,
 815                                        int (*okfn)(struct sk_buff *))
 816 {
 817         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 818         struct net_device *realoutdev = bridge_parent(skb->dev);
 819         u_int8_t pf;
 820
 821         if (!nf_bridge || !(nf_bridge->mask & BRNF_BRIDGED))
 822                 return NF_ACCEPT;
 823
 824         if (!realoutdev)
 825                 return NF_DROP;
 826
 827         if (IS_IP(skb) || IS_VLAN_IP(skb) || IS_PPPOE_IP(skb))
 828                 pf = NFPROTO_IPV4;
 829         else if (IS_IPV6(skb) || IS_VLAN_IPV6(skb) || IS_PPPOE_IPV6(skb))
 830                 pf = NFPROTO_IPV6;
 831         else
 832                 return NF_ACCEPT;
 833
 834         /* We assume any code from br_dev_queue_push_xmit onwards doesn't care
 835          * about the value of skb->pkt_type. */
 836         if (skb->pkt_type == PACKET_OTHERHOST) {
 837                 skb->pkt_type = PACKET_HOST;
 838                 nf_bridge->mask |= BRNF_PKT_TYPE;
 839         }
 840
 841         nf_bridge_pull_encap_header(skb);
 842         nf_bridge_save_header(skb);
 843         if (pf == NFPROTO_IPV4)
 844                 skb->protocol = htons(ETH_P_IP);
 845         else
 846                 skb->protocol = htons(ETH_P_IPV6);
 847
 848         NF_HOOK(pf, NF_INET_POST_ROUTING, skb, NULL, realoutdev,
 849                 br_nf_dev_queue_xmit);
 850
 851         return NF_STOLEN;
 852 }
 853
 854 /* IP/SABOTAGE *****************************************************/
 855 /* Don't hand locally destined packets to PF_INET(6)/PRE_ROUTING
 856  * for the second time. */
 857 static unsigned int ip_sabotage_in(const struct nf_hook_ops *ops,
 858                                    struct sk_buff *skb,
 859                                    const struct net_device *in,
 860                                    const struct net_device *out,
 861                                    int (*okfn)(struct sk_buff *))
 862 {
 863         if (skb->nf_bridge &&
 864             !(skb->nf_bridge->mask & BRNF_NF_BRIDGE_PREROUTING)) {
 865                 return NF_STOP;
 866         }
 867
 868         return NF_ACCEPT;
 869 }
 870
 871 void br_netfilter_enable(void)
 872 {
 873 }
 874 EXPORT_SYMBOL_GPL(br_netfilter_enable);
 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 = NFPROTO_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 = NFPROTO_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 = NFPROTO_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 = NFPROTO_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 = NFPROTO_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 = NFPROTO_IPV4,
 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 = NFPROTO_IPV6,
 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(struct 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 struct 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                 .procname       = "bridge-nf-pass-vlan-input-dev",
 982                 .data           = &brnf_pass_vlan_indev,
 983                 .maxlen         = sizeof(int),
 984                 .mode           = 0644,
 985                 .proc_handler   = brnf_sysctl_call_tables,
 986         },
 987         { }
 988 };
 989 #endif
 990
 991 static int __init br_netfilter_init(void)
 992 {
 993         int ret;
 994
 995         ret = nf_register_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
 996         if (ret < 0)
 997                 return ret;
 998
 999 #ifdef CONFIG_SYSCTL
1000         brnf_sysctl_header = register_net_sysctl(&init_net, "net/bridge", brnf_table);
1001         if (brnf_sysctl_header == NULL) {
1002                 printk(KERN_WARNING
1003                        "br_netfilter: can't register to sysctl.\n");
1004                 ret = -ENOMEM;
1005                 goto err1;
1006         }
1007 #endif
1008         printk(KERN_NOTICE "Bridge firewalling registered\n");
1009         return 0;
1010 err1:
1011         nf_unregister_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
1012         return ret;
1013 }
1014
1015 static void __exit br_netfilter_fini(void)
1016 {
1017         nf_unregister_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
1018 #ifdef CONFIG_SYSCTL
1019         unregister_net_sysctl_table(brnf_sysctl_header);
1020 #endif
1021 }
1022
1023 module_init(br_netfilter_init);
1024 module_exit(br_netfilter_fini);
1025
1026 MODULE_LICENSE("GPL");
1027 MODULE_AUTHOR("Lennert Buytenhek <buytenh@gnu.org>");
1028 MODULE_AUTHOR("Bart De Schuymer <bdschuym@pandora.be>");
1029 MODULE_DESCRIPTION("Linux ethernet netfilter firewall bridge");