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         (!skb_vlan_tag_present(skb) && skb->protocol == htons(ETH_P_IP))
  70
  71 #define IS_IPV6(skb) \
  72         (!skb_vlan_tag_present(skb) && skb->protocol == htons(ETH_P_IPV6))
  73
  74 #define IS_ARP(skb) \
  75         (!skb_vlan_tag_present(skb) && skb->protocol == htons(ETH_P_ARP))
  76
  77 static inline __be16 vlan_proto(const struct sk_buff *skb)
  78 {
  79         if (skb_vlan_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         const struct iphdr *iph;
 197         struct net_device *dev = skb->dev;
 198         u32 len;
 199
 200         if (!pskb_may_pull(skb, sizeof(struct iphdr)))
 201                 goto inhdr_error;
 202
 203         iph = ip_hdr(skb);
 204
 205         /* Basic sanity checks */
 206         if (iph->ihl < 5 || iph->version != 4)
 207                 goto inhdr_error;
 208
 209         if (!pskb_may_pull(skb, iph->ihl*4))
 210                 goto inhdr_error;
 211
 212         iph = ip_hdr(skb);
 213         if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl)))
 214                 goto inhdr_error;
 215
 216         len = ntohs(iph->tot_len);
 217         if (skb->len < len) {
 218                 IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INTRUNCATEDPKTS);
 219                 goto drop;
 220         } else if (len < (iph->ihl*4))
 221                 goto inhdr_error;
 222
 223         if (pskb_trim_rcsum(skb, len)) {
 224                 IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INDISCARDS);
 225                 goto drop;
 226         }
 227
 228         memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
 229         /* We should really parse IP options here but until
 230          * somebody who actually uses IP options complains to
 231          * us we'll just silently ignore the options because
 232          * we're lazy!
 233          */
 234         return 0;
 235
 236 inhdr_error:
 237         IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INHDRERRORS);
 238 drop:
 239         return -1;
 240 }
 241
 242 static void nf_bridge_update_protocol(struct sk_buff *skb)
 243 {
 244         if (skb->nf_bridge->mask & BRNF_8021Q)
 245                 skb->protocol = htons(ETH_P_8021Q);
 246         else if (skb->nf_bridge->mask & BRNF_PPPoE)
 247                 skb->protocol = htons(ETH_P_PPP_SES);
 248 }
 249
 250 /* PF_BRIDGE/PRE_ROUTING *********************************************/
 251 /* Undo the changes made for ip6tables PREROUTING and continue the
 252  * bridge PRE_ROUTING hook. */
 253 static int br_nf_pre_routing_finish_ipv6(struct sk_buff *skb)
 254 {
 255         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 256         struct rtable *rt;
 257
 258         if (nf_bridge->mask & BRNF_PKT_TYPE) {
 259                 skb->pkt_type = PACKET_OTHERHOST;
 260                 nf_bridge->mask ^= BRNF_PKT_TYPE;
 261         }
 262         nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
 263
 264         rt = bridge_parent_rtable(nf_bridge->physindev);
 265         if (!rt) {
 266                 kfree_skb(skb);
 267                 return 0;
 268         }
 269         skb_dst_set_noref(skb, &rt->dst);
 270
 271         skb->dev = nf_bridge->physindev;
 272         nf_bridge_update_protocol(skb);
 273         nf_bridge_push_encap_header(skb);
 274         NF_HOOK_THRESH(NFPROTO_BRIDGE, NF_BR_PRE_ROUTING, skb, skb->dev, NULL,
 275                        br_handle_frame_finish, 1);
 276
 277         return 0;
 278 }
 279
 280 /* Obtain the correct destination MAC address, while preserving the original
 281  * source MAC address. If we already know this address, we just copy it. If we
 282  * don't, we use the neighbour framework to find out. In both cases, we make
 283  * sure that br_handle_frame_finish() is called afterwards.
 284  */
 285 static int br_nf_pre_routing_finish_bridge(struct sk_buff *skb)
 286 {
 287         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 288         struct neighbour *neigh;
 289         struct dst_entry *dst;
 290
 291         skb->dev = bridge_parent(skb->dev);
 292         if (!skb->dev)
 293                 goto free_skb;
 294         dst = skb_dst(skb);
 295         neigh = dst_neigh_lookup_skb(dst, skb);
 296         if (neigh) {
 297                 int ret;
 298
 299                 if (neigh->hh.hh_len) {
 300                         neigh_hh_bridge(&neigh->hh, skb);
 301                         skb->dev = nf_bridge->physindev;
 302                         ret = br_handle_frame_finish(skb);
 303                 } else {
 304                         /* the neighbour function below overwrites the complete
 305                          * MAC header, so we save the Ethernet source address and
 306                          * protocol number.
 307                          */
 308                         skb_copy_from_linear_data_offset(skb,
 309                                                          -(ETH_HLEN-ETH_ALEN),
 310                                                          skb->nf_bridge->data,
 311                                                          ETH_HLEN-ETH_ALEN);
 312                         /* tell br_dev_xmit to continue with forwarding */
 313                         nf_bridge->mask |= BRNF_BRIDGED_DNAT;
 314                         /* FIXME Need to refragment */
 315                         ret = neigh->output(neigh, skb);
 316                 }
 317                 neigh_release(neigh);
 318                 return ret;
 319         }
 320 free_skb:
 321         kfree_skb(skb);
 322         return 0;
 323 }
 324
 325 /* This requires some explaining. If DNAT has taken place,
 326  * we will need to fix up the destination Ethernet address.
 327  *
 328  * There are two cases to consider:
 329  * 1. The packet was DNAT'ed to a device in the same bridge
 330  *    port group as it was received on. We can still bridge
 331  *    the packet.
 332  * 2. The packet was DNAT'ed to a different device, either
 333  *    a non-bridged device or another bridge port group.
 334  *    The packet will need to be routed.
 335  *
 336  * The correct way of distinguishing between these two cases is to
 337  * call ip_route_input() and to look at skb->dst->dev, which is
 338  * changed to the destination device if ip_route_input() succeeds.
 339  *
 340  * Let's first consider the case that ip_route_input() succeeds:
 341  *
 342  * If the output device equals the logical bridge device the packet
 343  * came in on, we can consider this bridging. The corresponding MAC
 344  * address will be obtained in br_nf_pre_routing_finish_bridge.
 345  * Otherwise, the packet is considered to be routed and we just
 346  * change the destination MAC address so that the packet will
 347  * later be passed up to the IP stack to be routed. For a redirected
 348  * packet, ip_route_input() will give back the localhost as output device,
 349  * which differs from the bridge device.
 350  *
 351  * Let's now consider the case that ip_route_input() fails:
 352  *
 353  * This can be because the destination address is martian, in which case
 354  * the packet will be dropped.
 355  * If IP forwarding is disabled, ip_route_input() will fail, while
 356  * ip_route_output_key() can return success. The source
 357  * address for ip_route_output_key() is set to zero, so ip_route_output_key()
 358  * thinks we're handling a locally generated packet and won't care
 359  * if IP forwarding is enabled. If the output device equals the logical bridge
 360  * device, we proceed as if ip_route_input() succeeded. If it differs from the
 361  * logical bridge port or if ip_route_output_key() fails we drop the packet.
 362  */
 363 static int br_nf_pre_routing_finish(struct sk_buff *skb)
 364 {
 365         struct net_device *dev = skb->dev;
 366         struct iphdr *iph = ip_hdr(skb);
 367         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 368         struct rtable *rt;
 369         int err;
 370         int frag_max_size;
 371
 372         frag_max_size = IPCB(skb)->frag_max_size;
 373         BR_INPUT_SKB_CB(skb)->frag_max_size = frag_max_size;
 374
 375         if (nf_bridge->mask & BRNF_PKT_TYPE) {
 376                 skb->pkt_type = PACKET_OTHERHOST;
 377                 nf_bridge->mask ^= BRNF_PKT_TYPE;
 378         }
 379         nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
 380         if (dnat_took_place(skb)) {
 381                 if ((err = ip_route_input(skb, iph->daddr, iph->saddr, iph->tos, dev))) {
 382                         struct in_device *in_dev = __in_dev_get_rcu(dev);
 383
 384                         /* If err equals -EHOSTUNREACH the error is due to a
 385                          * martian destination or due to the fact that
 386                          * forwarding is disabled. For most martian packets,
 387                          * ip_route_output_key() will fail. It won't fail for 2 types of
 388                          * martian destinations: loopback destinations and destination
 389                          * 0.0.0.0. In both cases the packet will be dropped because the
 390                          * destination is the loopback device and not the bridge. */
 391                         if (err != -EHOSTUNREACH || !in_dev || IN_DEV_FORWARD(in_dev))
 392                                 goto free_skb;
 393
 394                         rt = ip_route_output(dev_net(dev), iph->daddr, 0,
 395                                              RT_TOS(iph->tos), 0);
 396                         if (!IS_ERR(rt)) {
 397                                 /* - Bridged-and-DNAT'ed traffic doesn't
 398                                  *   require ip_forwarding. */
 399                                 if (rt->dst.dev == dev) {
 400                                         skb_dst_set(skb, &rt->dst);
 401                                         goto bridged_dnat;
 402                                 }
 403                                 ip_rt_put(rt);
 404                         }
 405 free_skb:
 406                         kfree_skb(skb);
 407                         return 0;
 408                 } else {
 409                         if (skb_dst(skb)->dev == dev) {
 410 bridged_dnat:
 411                                 skb->dev = nf_bridge->physindev;
 412                                 nf_bridge_update_protocol(skb);
 413                                 nf_bridge_push_encap_header(skb);
 414                                 NF_HOOK_THRESH(NFPROTO_BRIDGE,
 415                                                NF_BR_PRE_ROUTING,
 416                                                skb, skb->dev, NULL,
 417                                                br_nf_pre_routing_finish_bridge,
 418                                                1);
 419                                 return 0;
 420                         }
 421                         ether_addr_copy(eth_hdr(skb)->h_dest, dev->dev_addr);
 422                         skb->pkt_type = PACKET_HOST;
 423                 }
 424         } else {
 425                 rt = bridge_parent_rtable(nf_bridge->physindev);
 426                 if (!rt) {
 427                         kfree_skb(skb);
 428                         return 0;
 429                 }
 430                 skb_dst_set_noref(skb, &rt->dst);
 431         }
 432
 433         skb->dev = nf_bridge->physindev;
 434         nf_bridge_update_protocol(skb);
 435         nf_bridge_push_encap_header(skb);
 436         NF_HOOK_THRESH(NFPROTO_BRIDGE, NF_BR_PRE_ROUTING, skb, skb->dev, NULL,
 437                        br_handle_frame_finish, 1);
 438
 439         return 0;
 440 }
 441
 442 static struct net_device *brnf_get_logical_dev(struct sk_buff *skb, const struct net_device *dev)
 443 {
 444         struct net_device *vlan, *br;
 445
 446         br = bridge_parent(dev);
 447         if (brnf_pass_vlan_indev == 0 || !skb_vlan_tag_present(skb))
 448                 return br;
 449
 450         vlan = __vlan_find_dev_deep_rcu(br, skb->vlan_proto,
 451                                     skb_vlan_tag_get(skb) & VLAN_VID_MASK);
 452
 453         return vlan ? vlan : br;
 454 }
 455
 456 /* Some common code for IPv4/IPv6 */
 457 static struct net_device *setup_pre_routing(struct sk_buff *skb)
 458 {
 459         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 460
 461         if (skb->pkt_type == PACKET_OTHERHOST) {
 462                 skb->pkt_type = PACKET_HOST;
 463                 nf_bridge->mask |= BRNF_PKT_TYPE;
 464         }
 465
 466         nf_bridge->mask |= BRNF_NF_BRIDGE_PREROUTING;
 467         nf_bridge->physindev = skb->dev;
 468         skb->dev = brnf_get_logical_dev(skb, skb->dev);
 469         if (skb->protocol == htons(ETH_P_8021Q))
 470                 nf_bridge->mask |= BRNF_8021Q;
 471         else if (skb->protocol == htons(ETH_P_PPP_SES))
 472                 nf_bridge->mask |= BRNF_PPPoE;
 473
 474         /* Must drop socket now because of tproxy. */
 475         skb_orphan(skb);
 476         return skb->dev;
 477 }
 478
 479 /* We only check the length. A bridge shouldn't do any hop-by-hop stuff anyway */
 480 static int check_hbh_len(struct sk_buff *skb)
 481 {
 482         unsigned char *raw = (u8 *)(ipv6_hdr(skb) + 1);
 483         u32 pkt_len;
 484         const unsigned char *nh = skb_network_header(skb);
 485         int off = raw - nh;
 486         int len = (raw[1] + 1) << 3;
 487
 488         if ((raw + len) - skb->data > skb_headlen(skb))
 489                 goto bad;
 490
 491         off += 2;
 492         len -= 2;
 493
 494         while (len > 0) {
 495                 int optlen = nh[off + 1] + 2;
 496
 497                 switch (nh[off]) {
 498                 case IPV6_TLV_PAD1:
 499                         optlen = 1;
 500                         break;
 501
 502                 case IPV6_TLV_PADN:
 503                         break;
 504
 505                 case IPV6_TLV_JUMBO:
 506                         if (nh[off + 1] != 4 || (off & 3) != 2)
 507                                 goto bad;
 508                         pkt_len = ntohl(*(__be32 *) (nh + off + 2));
 509                         if (pkt_len <= IPV6_MAXPLEN ||
 510                             ipv6_hdr(skb)->payload_len)
 511                                 goto bad;
 512                         if (pkt_len > skb->len - sizeof(struct ipv6hdr))
 513                                 goto bad;
 514                         if (pskb_trim_rcsum(skb,
 515                                             pkt_len + sizeof(struct ipv6hdr)))
 516                                 goto bad;
 517                         nh = skb_network_header(skb);
 518                         break;
 519                 default:
 520                         if (optlen > len)
 521                                 goto bad;
 522                         break;
 523                 }
 524                 off += optlen;
 525                 len -= optlen;
 526         }
 527         if (len == 0)
 528                 return 0;
 529 bad:
 530         return -1;
 531
 532 }
 533
 534 /* Replicate the checks that IPv6 does on packet reception and pass the packet
 535  * to ip6tables, which doesn't support NAT, so things are fairly simple. */
 536 static unsigned int br_nf_pre_routing_ipv6(const struct nf_hook_ops *ops,
 537                                            struct sk_buff *skb,
 538                                            const struct net_device *in,
 539                                            const struct net_device *out,
 540                                            int (*okfn)(struct sk_buff *))
 541 {
 542         const struct ipv6hdr *hdr;
 543         u32 pkt_len;
 544
 545         if (skb->len < sizeof(struct ipv6hdr))
 546                 return NF_DROP;
 547
 548         if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
 549                 return NF_DROP;
 550
 551         hdr = ipv6_hdr(skb);
 552
 553         if (hdr->version != 6)
 554                 return NF_DROP;
 555
 556         pkt_len = ntohs(hdr->payload_len);
 557
 558         if (pkt_len || hdr->nexthdr != NEXTHDR_HOP) {
 559                 if (pkt_len + sizeof(struct ipv6hdr) > skb->len)
 560                         return NF_DROP;
 561                 if (pskb_trim_rcsum(skb, pkt_len + sizeof(struct ipv6hdr)))
 562                         return NF_DROP;
 563         }
 564         if (hdr->nexthdr == NEXTHDR_HOP && check_hbh_len(skb))
 565                 return NF_DROP;
 566
 567         nf_bridge_put(skb->nf_bridge);
 568         if (!nf_bridge_alloc(skb))
 569                 return NF_DROP;
 570         if (!setup_pre_routing(skb))
 571                 return NF_DROP;
 572
 573         skb->protocol = htons(ETH_P_IPV6);
 574         NF_HOOK(NFPROTO_IPV6, NF_INET_PRE_ROUTING, skb, skb->dev, NULL,
 575                 br_nf_pre_routing_finish_ipv6);
 576
 577         return NF_STOLEN;
 578 }
 579
 580 /* Direct IPv6 traffic to br_nf_pre_routing_ipv6.
 581  * Replicate the checks that IPv4 does on packet reception.
 582  * Set skb->dev to the bridge device (i.e. parent of the
 583  * receiving device) to make netfilter happy, the REDIRECT
 584  * target in particular.  Save the original destination IP
 585  * address to be able to detect DNAT afterwards. */
 586 static unsigned int br_nf_pre_routing(const struct nf_hook_ops *ops,
 587                                       struct sk_buff *skb,
 588                                       const struct net_device *in,
 589                                       const struct net_device *out,
 590                                       int (*okfn)(struct sk_buff *))
 591 {
 592         struct net_bridge_port *p;
 593         struct net_bridge *br;
 594         __u32 len = nf_bridge_encap_header_len(skb);
 595
 596         if (unlikely(!pskb_may_pull(skb, len)))
 597                 return NF_DROP;
 598
 599         p = br_port_get_rcu(in);
 600         if (p == NULL)
 601                 return NF_DROP;
 602         br = p->br;
 603
 604         if (IS_IPV6(skb) || IS_VLAN_IPV6(skb) || IS_PPPOE_IPV6(skb)) {
 605                 if (!brnf_call_ip6tables && !br->nf_call_ip6tables)
 606                         return NF_ACCEPT;
 607
 608                 nf_bridge_pull_encap_header_rcsum(skb);
 609                 return br_nf_pre_routing_ipv6(ops, skb, in, out, okfn);
 610         }
 611
 612         if (!brnf_call_iptables && !br->nf_call_iptables)
 613                 return NF_ACCEPT;
 614
 615         if (!IS_IP(skb) && !IS_VLAN_IP(skb) && !IS_PPPOE_IP(skb))
 616                 return NF_ACCEPT;
 617
 618         nf_bridge_pull_encap_header_rcsum(skb);
 619
 620         if (br_parse_ip_options(skb))
 621                 return NF_DROP;
 622
 623         nf_bridge_put(skb->nf_bridge);
 624         if (!nf_bridge_alloc(skb))
 625                 return NF_DROP;
 626         if (!setup_pre_routing(skb))
 627                 return NF_DROP;
 628         store_orig_dstaddr(skb);
 629         skb->protocol = htons(ETH_P_IP);
 630
 631         NF_HOOK(NFPROTO_IPV4, NF_INET_PRE_ROUTING, skb, skb->dev, NULL,
 632                 br_nf_pre_routing_finish);
 633
 634         return NF_STOLEN;
 635 }
 636
 637
 638 /* PF_BRIDGE/LOCAL_IN ************************************************/
 639 /* The packet is locally destined, which requires a real
 640  * dst_entry, so detach the fake one.  On the way up, the
 641  * packet would pass through PRE_ROUTING again (which already
 642  * took place when the packet entered the bridge), but we
 643  * register an IPv4 PRE_ROUTING 'sabotage' hook that will
 644  * prevent this from happening. */
 645 static unsigned int br_nf_local_in(const struct nf_hook_ops *ops,
 646                                    struct sk_buff *skb,
 647                                    const struct net_device *in,
 648                                    const struct net_device *out,
 649                                    int (*okfn)(struct sk_buff *))
 650 {
 651         br_drop_fake_rtable(skb);
 652         return NF_ACCEPT;
 653 }
 654
 655 /* PF_BRIDGE/FORWARD *************************************************/
 656 static int br_nf_forward_finish(struct sk_buff *skb)
 657 {
 658         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 659         struct net_device *in;
 660
 661         if (!IS_ARP(skb) && !IS_VLAN_ARP(skb)) {
 662                 in = nf_bridge->physindev;
 663                 if (nf_bridge->mask & BRNF_PKT_TYPE) {
 664                         skb->pkt_type = PACKET_OTHERHOST;
 665                         nf_bridge->mask ^= BRNF_PKT_TYPE;
 666                 }
 667                 nf_bridge_update_protocol(skb);
 668         } else {
 669                 in = *((struct net_device **)(skb->cb));
 670         }
 671         nf_bridge_push_encap_header(skb);
 672
 673         NF_HOOK_THRESH(NFPROTO_BRIDGE, NF_BR_FORWARD, skb, in,
 674                        skb->dev, br_forward_finish, 1);
 675         return 0;
 676 }
 677
 678
 679 /* This is the 'purely bridged' case.  For IP, we pass the packet to
 680  * netfilter with indev and outdev set to the bridge device,
 681  * but we are still able to filter on the 'real' indev/outdev
 682  * because of the physdev module. For ARP, indev and outdev are the
 683  * bridge ports. */
 684 static unsigned int br_nf_forward_ip(const struct nf_hook_ops *ops,
 685                                      struct sk_buff *skb,
 686                                      const struct net_device *in,
 687                                      const struct net_device *out,
 688                                      int (*okfn)(struct sk_buff *))
 689 {
 690         struct nf_bridge_info *nf_bridge;
 691         struct net_device *parent;
 692         u_int8_t pf;
 693
 694         if (!skb->nf_bridge)
 695                 return NF_ACCEPT;
 696
 697         /* Need exclusive nf_bridge_info since we might have multiple
 698          * different physoutdevs. */
 699         if (!nf_bridge_unshare(skb))
 700                 return NF_DROP;
 701
 702         parent = bridge_parent(out);
 703         if (!parent)
 704                 return NF_DROP;
 705
 706         if (IS_IP(skb) || IS_VLAN_IP(skb) || IS_PPPOE_IP(skb))
 707                 pf = NFPROTO_IPV4;
 708         else if (IS_IPV6(skb) || IS_VLAN_IPV6(skb) || IS_PPPOE_IPV6(skb))
 709                 pf = NFPROTO_IPV6;
 710         else
 711                 return NF_ACCEPT;
 712
 713         nf_bridge_pull_encap_header(skb);
 714
 715         nf_bridge = skb->nf_bridge;
 716         if (skb->pkt_type == PACKET_OTHERHOST) {
 717                 skb->pkt_type = PACKET_HOST;
 718                 nf_bridge->mask |= BRNF_PKT_TYPE;
 719         }
 720
 721         if (pf == NFPROTO_IPV4 && br_parse_ip_options(skb))
 722                 return NF_DROP;
 723
 724         /* The physdev module checks on this */
 725         nf_bridge->mask |= BRNF_BRIDGED;
 726         nf_bridge->physoutdev = skb->dev;
 727         if (pf == NFPROTO_IPV4)
 728                 skb->protocol = htons(ETH_P_IP);
 729         else
 730                 skb->protocol = htons(ETH_P_IPV6);
 731
 732         NF_HOOK(pf, NF_INET_FORWARD, skb, brnf_get_logical_dev(skb, in), parent,
 733                 br_nf_forward_finish);
 734
 735         return NF_STOLEN;
 736 }
 737
 738 static unsigned int br_nf_forward_arp(const struct nf_hook_ops *ops,
 739                                       struct sk_buff *skb,
 740                                       const struct net_device *in,
 741                                       const struct net_device *out,
 742                                       int (*okfn)(struct sk_buff *))
 743 {
 744         struct net_bridge_port *p;
 745         struct net_bridge *br;
 746         struct net_device **d = (struct net_device **)(skb->cb);
 747
 748         p = br_port_get_rcu(out);
 749         if (p == NULL)
 750                 return NF_ACCEPT;
 751         br = p->br;
 752
 753         if (!brnf_call_arptables && !br->nf_call_arptables)
 754                 return NF_ACCEPT;
 755
 756         if (!IS_ARP(skb)) {
 757                 if (!IS_VLAN_ARP(skb))
 758                         return NF_ACCEPT;
 759                 nf_bridge_pull_encap_header(skb);
 760         }
 761
 762         if (arp_hdr(skb)->ar_pln != 4) {
 763                 if (IS_VLAN_ARP(skb))
 764                         nf_bridge_push_encap_header(skb);
 765                 return NF_ACCEPT;
 766         }
 767         *d = (struct net_device *)in;
 768         NF_HOOK(NFPROTO_ARP, NF_ARP_FORWARD, skb, (struct net_device *)in,
 769                 (struct net_device *)out, br_nf_forward_finish);
 770
 771         return NF_STOLEN;
 772 }
 773
 774 #if IS_ENABLED(CONFIG_NF_DEFRAG_IPV4)
 775 static bool nf_bridge_copy_header(struct sk_buff *skb)
 776 {
 777         int err;
 778         unsigned int header_size;
 779
 780         nf_bridge_update_protocol(skb);
 781         header_size = ETH_HLEN + nf_bridge_encap_header_len(skb);
 782         err = skb_cow_head(skb, header_size);
 783         if (err)
 784                 return false;
 785
 786         skb_copy_to_linear_data_offset(skb, -header_size,
 787                                        skb->nf_bridge->data, header_size);
 788         __skb_push(skb, nf_bridge_encap_header_len(skb));
 789         return true;
 790 }
 791
 792 static int br_nf_push_frag_xmit(struct sk_buff *skb)
 793 {
 794         if (!nf_bridge_copy_header(skb)) {
 795                 kfree_skb(skb);
 796                 return 0;
 797         }
 798
 799         return br_dev_queue_push_xmit(skb);
 800 }
 801
 802 static int br_nf_dev_queue_xmit(struct sk_buff *skb)
 803 {
 804         int ret;
 805         int frag_max_size;
 806
 807         /* This is wrong! We should preserve the original fragment
 808          * boundaries by preserving frag_list rather than refragmenting.
 809          */
 810         if (skb->protocol == htons(ETH_P_IP) &&
 811             skb->len + nf_bridge_mtu_reduction(skb) > skb->dev->mtu &&
 812             !skb_is_gso(skb)) {
 813                 frag_max_size = BR_INPUT_SKB_CB(skb)->frag_max_size;
 814                 if (br_parse_ip_options(skb))
 815                         /* Drop invalid packet */
 816                         return NF_DROP;
 817                 IPCB(skb)->frag_max_size = frag_max_size;
 818                 ret = ip_fragment(skb, br_nf_push_frag_xmit);
 819         } else
 820                 ret = br_dev_queue_push_xmit(skb);
 821
 822         return ret;
 823 }
 824 #else
 825 static int br_nf_dev_queue_xmit(struct sk_buff *skb)
 826 {
 827         return br_dev_queue_push_xmit(skb);
 828 }
 829 #endif
 830
 831 /* PF_BRIDGE/POST_ROUTING ********************************************/
 832 static unsigned int br_nf_post_routing(const struct nf_hook_ops *ops,
 833                                        struct sk_buff *skb,
 834                                        const struct net_device *in,
 835                                        const struct net_device *out,
 836                                        int (*okfn)(struct sk_buff *))
 837 {
 838         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 839         struct net_device *realoutdev = bridge_parent(skb->dev);
 840         u_int8_t pf;
 841
 842         if (!nf_bridge || !(nf_bridge->mask & BRNF_BRIDGED))
 843                 return NF_ACCEPT;
 844
 845         if (!realoutdev)
 846                 return NF_DROP;
 847
 848         if (IS_IP(skb) || IS_VLAN_IP(skb) || IS_PPPOE_IP(skb))
 849                 pf = NFPROTO_IPV4;
 850         else if (IS_IPV6(skb) || IS_VLAN_IPV6(skb) || IS_PPPOE_IPV6(skb))
 851                 pf = NFPROTO_IPV6;
 852         else
 853                 return NF_ACCEPT;
 854
 855         /* We assume any code from br_dev_queue_push_xmit onwards doesn't care
 856          * about the value of skb->pkt_type. */
 857         if (skb->pkt_type == PACKET_OTHERHOST) {
 858                 skb->pkt_type = PACKET_HOST;
 859                 nf_bridge->mask |= BRNF_PKT_TYPE;
 860         }
 861
 862         nf_bridge_pull_encap_header(skb);
 863         nf_bridge_save_header(skb);
 864         if (pf == NFPROTO_IPV4)
 865                 skb->protocol = htons(ETH_P_IP);
 866         else
 867                 skb->protocol = htons(ETH_P_IPV6);
 868
 869         NF_HOOK(pf, NF_INET_POST_ROUTING, skb, NULL, realoutdev,
 870                 br_nf_dev_queue_xmit);
 871
 872         return NF_STOLEN;
 873 }
 874
 875 /* IP/SABOTAGE *****************************************************/
 876 /* Don't hand locally destined packets to PF_INET(6)/PRE_ROUTING
 877  * for the second time. */
 878 static unsigned int ip_sabotage_in(const struct nf_hook_ops *ops,
 879                                    struct sk_buff *skb,
 880                                    const struct net_device *in,
 881                                    const struct net_device *out,
 882                                    int (*okfn)(struct sk_buff *))
 883 {
 884         if (skb->nf_bridge &&
 885             !(skb->nf_bridge->mask & BRNF_NF_BRIDGE_PREROUTING)) {
 886                 return NF_STOP;
 887         }
 888
 889         return NF_ACCEPT;
 890 }
 891
 892 void br_netfilter_enable(void)
 893 {
 894 }
 895 EXPORT_SYMBOL_GPL(br_netfilter_enable);
 896
 897 /* For br_nf_post_routing, we need (prio = NF_BR_PRI_LAST), because
 898  * br_dev_queue_push_xmit is called afterwards */
 899 static struct nf_hook_ops br_nf_ops[] __read_mostly = {
 900         {
 901                 .hook = br_nf_pre_routing,
 902                 .owner = THIS_MODULE,
 903                 .pf = NFPROTO_BRIDGE,
 904                 .hooknum = NF_BR_PRE_ROUTING,
 905                 .priority = NF_BR_PRI_BRNF,
 906         },
 907         {
 908                 .hook = br_nf_local_in,
 909                 .owner = THIS_MODULE,
 910                 .pf = NFPROTO_BRIDGE,
 911                 .hooknum = NF_BR_LOCAL_IN,
 912                 .priority = NF_BR_PRI_BRNF,
 913         },
 914         {
 915                 .hook = br_nf_forward_ip,
 916                 .owner = THIS_MODULE,
 917                 .pf = NFPROTO_BRIDGE,
 918                 .hooknum = NF_BR_FORWARD,
 919                 .priority = NF_BR_PRI_BRNF - 1,
 920         },
 921         {
 922                 .hook = br_nf_forward_arp,
 923                 .owner = THIS_MODULE,
 924                 .pf = NFPROTO_BRIDGE,
 925                 .hooknum = NF_BR_FORWARD,
 926                 .priority = NF_BR_PRI_BRNF,
 927         },
 928         {
 929                 .hook = br_nf_post_routing,
 930                 .owner = THIS_MODULE,
 931                 .pf = NFPROTO_BRIDGE,
 932                 .hooknum = NF_BR_POST_ROUTING,
 933                 .priority = NF_BR_PRI_LAST,
 934         },
 935         {
 936                 .hook = ip_sabotage_in,
 937                 .owner = THIS_MODULE,
 938                 .pf = NFPROTO_IPV4,
 939                 .hooknum = NF_INET_PRE_ROUTING,
 940                 .priority = NF_IP_PRI_FIRST,
 941         },
 942         {
 943                 .hook = ip_sabotage_in,
 944                 .owner = THIS_MODULE,
 945                 .pf = NFPROTO_IPV6,
 946                 .hooknum = NF_INET_PRE_ROUTING,
 947                 .priority = NF_IP6_PRI_FIRST,
 948         },
 949 };
 950
 951 #ifdef CONFIG_SYSCTL
 952 static
 953 int brnf_sysctl_call_tables(struct ctl_table *ctl, int write,
 954                             void __user *buffer, size_t *lenp, loff_t *ppos)
 955 {
 956         int ret;
 957
 958         ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
 959
 960         if (write && *(int *)(ctl->data))
 961                 *(int *)(ctl->data) = 1;
 962         return ret;
 963 }
 964
 965 static struct ctl_table brnf_table[] = {
 966         {
 967                 .procname       = "bridge-nf-call-arptables",
 968                 .data           = &brnf_call_arptables,
 969                 .maxlen         = sizeof(int),
 970                 .mode           = 0644,
 971                 .proc_handler   = brnf_sysctl_call_tables,
 972         },
 973         {
 974                 .procname       = "bridge-nf-call-iptables",
 975                 .data           = &brnf_call_iptables,
 976                 .maxlen         = sizeof(int),
 977                 .mode           = 0644,
 978                 .proc_handler   = brnf_sysctl_call_tables,
 979         },
 980         {
 981                 .procname       = "bridge-nf-call-ip6tables",
 982                 .data           = &brnf_call_ip6tables,
 983                 .maxlen         = sizeof(int),
 984                 .mode           = 0644,
 985                 .proc_handler   = brnf_sysctl_call_tables,
 986         },
 987         {
 988                 .procname       = "bridge-nf-filter-vlan-tagged",
 989                 .data           = &brnf_filter_vlan_tagged,
 990                 .maxlen         = sizeof(int),
 991                 .mode           = 0644,
 992                 .proc_handler   = brnf_sysctl_call_tables,
 993         },
 994         {
 995                 .procname       = "bridge-nf-filter-pppoe-tagged",
 996                 .data           = &brnf_filter_pppoe_tagged,
 997                 .maxlen         = sizeof(int),
 998                 .mode           = 0644,
 999                 .proc_handler   = brnf_sysctl_call_tables,
1000         },
1001         {
1002                 .procname       = "bridge-nf-pass-vlan-input-dev",
1003                 .data           = &brnf_pass_vlan_indev,
1004                 .maxlen         = sizeof(int),
1005                 .mode           = 0644,
1006                 .proc_handler   = brnf_sysctl_call_tables,
1007         },
1008         { }
1009 };
1010 #endif
1011
1012 static int __init br_netfilter_init(void)
1013 {
1014         int ret;
1015
1016         ret = nf_register_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
1017         if (ret < 0)
1018                 return ret;
1019
1020 #ifdef CONFIG_SYSCTL
1021         brnf_sysctl_header = register_net_sysctl(&init_net, "net/bridge", brnf_table);
1022         if (brnf_sysctl_header == NULL) {
1023                 printk(KERN_WARNING
1024                        "br_netfilter: can't register to sysctl.\n");
1025                 nf_unregister_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
1026                 return -ENOMEM;
1027         }
1028 #endif
1029         printk(KERN_NOTICE "Bridge firewalling registered\n");
1030         return 0;
1031 }
1032
1033 static void __exit br_netfilter_fini(void)
1034 {
1035         nf_unregister_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
1036 #ifdef CONFIG_SYSCTL
1037         unregister_net_sysctl_table(brnf_sysctl_header);
1038 #endif
1039 }
1040
1041 module_init(br_netfilter_init);
1042 module_exit(br_netfilter_fini);
1043
1044 MODULE_LICENSE("GPL");
1045 MODULE_AUTHOR("Lennert Buytenhek <buytenh@gnu.org>");
1046 MODULE_AUTHOR("Bart De Schuymer <bdschuym@pandora.be>");
1047 MODULE_DESCRIPTION("Linux ethernet netfilter firewall bridge");