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 static int brnf_pass_vlan_indev __read_mostly = 0;
  58 #else
  59 #define brnf_call_iptables 1
  60 #define brnf_call_ip6tables 1
  61 #define brnf_call_arptables 1
  62 #define brnf_filter_vlan_tagged 0
  63 #define brnf_filter_pppoe_tagged 0
  64 #define brnf_pass_vlan_indev 0
  65 #endif
  66
  67 #define IS_IP(skb) \
  68         (!vlan_tx_tag_present(skb) && skb->protocol == htons(ETH_P_IP))
  69
  70 #define IS_IPV6(skb) \
  71         (!vlan_tx_tag_present(skb) && skb->protocol == htons(ETH_P_IPV6))
  72
  73 #define IS_ARP(skb) \
  74         (!vlan_tx_tag_present(skb) && skb->protocol == htons(ETH_P_ARP))
  75
  76 static inline __be16 vlan_proto(const struct sk_buff *skb)
  77 {
  78         if (vlan_tx_tag_present(skb))
  79                 return skb->protocol;
  80         else if (skb->protocol == htons(ETH_P_8021Q))
  81                 return vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
  82         else
  83                 return 0;
  84 }
  85
  86 #define IS_VLAN_IP(skb) \
  87         (vlan_proto(skb) == htons(ETH_P_IP) && \
  88          brnf_filter_vlan_tagged)
  89
  90 #define IS_VLAN_IPV6(skb) \
  91         (vlan_proto(skb) == htons(ETH_P_IPV6) && \
  92          brnf_filter_vlan_tagged)
  93
  94 #define IS_VLAN_ARP(skb) \
  95         (vlan_proto(skb) == htons(ETH_P_ARP) && \
  96          brnf_filter_vlan_tagged)
  97
  98 static inline __be16 pppoe_proto(const struct sk_buff *skb)
  99 {
 100         return *((__be16 *)(skb_mac_header(skb) + ETH_HLEN +
 101                             sizeof(struct pppoe_hdr)));
 102 }
 103
 104 #define IS_PPPOE_IP(skb) \
 105         (skb->protocol == htons(ETH_P_PPP_SES) && \
 106          pppoe_proto(skb) == htons(PPP_IP) && \
 107          brnf_filter_pppoe_tagged)
 108
 109 #define IS_PPPOE_IPV6(skb) \
 110         (skb->protocol == htons(ETH_P_PPP_SES) && \
 111          pppoe_proto(skb) == htons(PPP_IPV6) && \
 112          brnf_filter_pppoe_tagged)
 113
 114 static inline struct rtable *bridge_parent_rtable(const struct net_device *dev)
 115 {
 116         struct net_bridge_port *port;
 117
 118         port = br_port_get_rcu(dev);
 119         return port ? &port->br->fake_rtable : NULL;
 120 }
 121
 122 static inline struct net_device *bridge_parent(const struct net_device *dev)
 123 {
 124         struct net_bridge_port *port;
 125
 126         port = br_port_get_rcu(dev);
 127         return port ? port->br->dev : NULL;
 128 }
 129
 130 static inline struct nf_bridge_info *nf_bridge_alloc(struct sk_buff *skb)
 131 {
 132         skb->nf_bridge = kzalloc(sizeof(struct nf_bridge_info), GFP_ATOMIC);
 133         if (likely(skb->nf_bridge))
 134                 atomic_set(&(skb->nf_bridge->use), 1);
 135
 136         return skb->nf_bridge;
 137 }
 138
 139 static inline struct nf_bridge_info *nf_bridge_unshare(struct sk_buff *skb)
 140 {
 141         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 142
 143         if (atomic_read(&nf_bridge->use) > 1) {
 144                 struct nf_bridge_info *tmp = nf_bridge_alloc(skb);
 145
 146                 if (tmp) {
 147                         memcpy(tmp, nf_bridge, sizeof(struct nf_bridge_info));
 148                         atomic_set(&tmp->use, 1);
 149                 }
 150                 nf_bridge_put(nf_bridge);
 151                 nf_bridge = tmp;
 152         }
 153         return nf_bridge;
 154 }
 155
 156 static inline void nf_bridge_push_encap_header(struct sk_buff *skb)
 157 {
 158         unsigned int len = nf_bridge_encap_header_len(skb);
 159
 160         skb_push(skb, len);
 161         skb->network_header -= len;
 162 }
 163
 164 static inline void nf_bridge_pull_encap_header(struct sk_buff *skb)
 165 {
 166         unsigned int len = nf_bridge_encap_header_len(skb);
 167
 168         skb_pull(skb, len);
 169         skb->network_header += len;
 170 }
 171
 172 static inline void nf_bridge_pull_encap_header_rcsum(struct sk_buff *skb)
 173 {
 174         unsigned int len = nf_bridge_encap_header_len(skb);
 175
 176         skb_pull_rcsum(skb, len);
 177         skb->network_header += len;
 178 }
 179
 180 static inline void nf_bridge_save_header(struct sk_buff *skb)
 181 {
 182         int header_size = ETH_HLEN + nf_bridge_encap_header_len(skb);
 183
 184         skb_copy_from_linear_data_offset(skb, -header_size,
 185                                          skb->nf_bridge->data, header_size);
 186 }
 187
 188 /* When handing a packet over to the IP layer
 189  * check whether we have a skb that is in the
 190  * expected format
 191  */
 192
 193 static int br_parse_ip_options(struct sk_buff *skb)
 194 {
 195         const struct iphdr *iph;
 196         struct net_device *dev = skb->dev;
 197         u32 len;
 198
 199         if (!pskb_may_pull(skb, sizeof(struct iphdr)))
 200                 goto inhdr_error;
 201
 202         iph = ip_hdr(skb);
 203
 204         /* Basic sanity checks */
 205         if (iph->ihl < 5 || iph->version != 4)
 206                 goto inhdr_error;
 207
 208         if (!pskb_may_pull(skb, iph->ihl*4))
 209                 goto inhdr_error;
 210
 211         iph = ip_hdr(skb);
 212         if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl)))
 213                 goto inhdr_error;
 214
 215         len = ntohs(iph->tot_len);
 216         if (skb->len < len) {
 217                 IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INTRUNCATEDPKTS);
 218                 goto drop;
 219         } else if (len < (iph->ihl*4))
 220                 goto inhdr_error;
 221
 222         if (pskb_trim_rcsum(skb, len)) {
 223                 IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INDISCARDS);
 224                 goto drop;
 225         }
 226
 227         memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
 228         /* We should really parse IP options here but until
 229          * somebody who actually uses IP options complains to
 230          * us we'll just silently ignore the options because
 231          * we're lazy!
 232          */
 233         return 0;
 234
 235 inhdr_error:
 236         IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INHDRERRORS);
 237 drop:
 238         return -1;
 239 }
 240
 241 /* PF_BRIDGE/PRE_ROUTING *********************************************/
 242 /* Undo the changes made for ip6tables PREROUTING and continue the
 243  * bridge PRE_ROUTING hook. */
 244 static int br_nf_pre_routing_finish_ipv6(struct sk_buff *skb)
 245 {
 246         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 247         struct rtable *rt;
 248
 249         if (nf_bridge->mask & BRNF_PKT_TYPE) {
 250                 skb->pkt_type = PACKET_OTHERHOST;
 251                 nf_bridge->mask ^= BRNF_PKT_TYPE;
 252         }
 253         nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
 254
 255         rt = bridge_parent_rtable(nf_bridge->physindev);
 256         if (!rt) {
 257                 kfree_skb(skb);
 258                 return 0;
 259         }
 260         skb_dst_set_noref(skb, &rt->dst);
 261
 262         skb->dev = nf_bridge->physindev;
 263         nf_bridge_update_protocol(skb);
 264         nf_bridge_push_encap_header(skb);
 265         NF_HOOK_THRESH(NFPROTO_BRIDGE, NF_BR_PRE_ROUTING, skb, skb->dev, NULL,
 266                        br_handle_frame_finish, 1);
 267
 268         return 0;
 269 }
 270
 271 /* Obtain the correct destination MAC address, while preserving the original
 272  * source MAC address. If we already know this address, we just copy it. If we
 273  * don't, we use the neighbour framework to find out. In both cases, we make
 274  * sure that br_handle_frame_finish() is called afterwards.
 275  */
 276 static int br_nf_pre_routing_finish_bridge(struct sk_buff *skb)
 277 {
 278         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 279         struct neighbour *neigh;
 280         struct dst_entry *dst;
 281
 282         skb->dev = bridge_parent(skb->dev);
 283         if (!skb->dev)
 284                 goto free_skb;
 285         dst = skb_dst(skb);
 286         neigh = dst_neigh_lookup_skb(dst, skb);
 287         if (neigh) {
 288                 int ret;
 289
 290                 if (neigh->hh.hh_len) {
 291                         neigh_hh_bridge(&neigh->hh, skb);
 292                         skb->dev = nf_bridge->physindev;
 293                         ret = br_handle_frame_finish(skb);
 294                 } else {
 295                         /* the neighbour function below overwrites the complete
 296                          * MAC header, so we save the Ethernet source address and
 297                          * protocol number.
 298                          */
 299                         skb_copy_from_linear_data_offset(skb,
 300                                                          -(ETH_HLEN-ETH_ALEN),
 301                                                          skb->nf_bridge->data,
 302                                                          ETH_HLEN-ETH_ALEN);
 303                         /* tell br_dev_xmit to continue with forwarding */
 304                         nf_bridge->mask |= BRNF_BRIDGED_DNAT;
 305                         /* FIXME Need to refragment */
 306                         ret = neigh->output(neigh, skb);
 307                 }
 308                 neigh_release(neigh);
 309                 return ret;
 310         }
 311 free_skb:
 312         kfree_skb(skb);
 313         return 0;
 314 }
 315
 316 /* This requires some explaining. If DNAT has taken place,
 317  * we will need to fix up the destination Ethernet address.
 318  *
 319  * There are two cases to consider:
 320  * 1. The packet was DNAT'ed to a device in the same bridge
 321  *    port group as it was received on. We can still bridge
 322  *    the packet.
 323  * 2. The packet was DNAT'ed to a different device, either
 324  *    a non-bridged device or another bridge port group.
 325  *    The packet will need to be routed.
 326  *
 327  * The correct way of distinguishing between these two cases is to
 328  * call ip_route_input() and to look at skb->dst->dev, which is
 329  * changed to the destination device if ip_route_input() succeeds.
 330  *
 331  * Let's first consider the case that ip_route_input() succeeds:
 332  *
 333  * If the output device equals the logical bridge device the packet
 334  * came in on, we can consider this bridging. The corresponding MAC
 335  * address will be obtained in br_nf_pre_routing_finish_bridge.
 336  * Otherwise, the packet is considered to be routed and we just
 337  * change the destination MAC address so that the packet will
 338  * later be passed up to the IP stack to be routed. For a redirected
 339  * packet, ip_route_input() will give back the localhost as output device,
 340  * which differs from the bridge device.
 341  *
 342  * Let's now consider the case that ip_route_input() fails:
 343  *
 344  * This can be because the destination address is martian, in which case
 345  * the packet will be dropped.
 346  * If IP forwarding is disabled, ip_route_input() will fail, while
 347  * ip_route_output_key() can return success. The source
 348  * address for ip_route_output_key() is set to zero, so ip_route_output_key()
 349  * thinks we're handling a locally generated packet and won't care
 350  * if IP forwarding is enabled. If the output device equals the logical bridge
 351  * device, we proceed as if ip_route_input() succeeded. If it differs from the
 352  * logical bridge port or if ip_route_output_key() fails we drop the packet.
 353  */
 354 static int br_nf_pre_routing_finish(struct sk_buff *skb)
 355 {
 356         struct net_device *dev = skb->dev;
 357         struct iphdr *iph = ip_hdr(skb);
 358         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 359         struct rtable *rt;
 360         int err;
 361         int frag_max_size;
 362
 363         frag_max_size = IPCB(skb)->frag_max_size;
 364         BR_INPUT_SKB_CB(skb)->frag_max_size = frag_max_size;
 365
 366         if (nf_bridge->mask & BRNF_PKT_TYPE) {
 367                 skb->pkt_type = PACKET_OTHERHOST;
 368                 nf_bridge->mask ^= BRNF_PKT_TYPE;
 369         }
 370         nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
 371         if (dnat_took_place(skb)) {
 372                 if ((err = ip_route_input(skb, iph->daddr, iph->saddr, iph->tos, dev))) {
 373                         struct in_device *in_dev = __in_dev_get_rcu(dev);
 374
 375                         /* If err equals -EHOSTUNREACH the error is due to a
 376                          * martian destination or due to the fact that
 377                          * forwarding is disabled. For most martian packets,
 378                          * ip_route_output_key() will fail. It won't fail for 2 types of
 379                          * martian destinations: loopback destinations and destination
 380                          * 0.0.0.0. In both cases the packet will be dropped because the
 381                          * destination is the loopback device and not the bridge. */
 382                         if (err != -EHOSTUNREACH || !in_dev || IN_DEV_FORWARD(in_dev))
 383                                 goto free_skb;
 384
 385                         rt = ip_route_output(dev_net(dev), iph->daddr, 0,
 386                                              RT_TOS(iph->tos), 0);
 387                         if (!IS_ERR(rt)) {
 388                                 /* - Bridged-and-DNAT'ed traffic doesn't
 389                                  *   require ip_forwarding. */
 390                                 if (rt->dst.dev == dev) {
 391                                         skb_dst_set(skb, &rt->dst);
 392                                         goto bridged_dnat;
 393                                 }
 394                                 ip_rt_put(rt);
 395                         }
 396 free_skb:
 397                         kfree_skb(skb);
 398                         return 0;
 399                 } else {
 400                         if (skb_dst(skb)->dev == dev) {
 401 bridged_dnat:
 402                                 skb->dev = nf_bridge->physindev;
 403                                 nf_bridge_update_protocol(skb);
 404                                 nf_bridge_push_encap_header(skb);
 405                                 NF_HOOK_THRESH(NFPROTO_BRIDGE,
 406                                                NF_BR_PRE_ROUTING,
 407                                                skb, skb->dev, NULL,
 408                                                br_nf_pre_routing_finish_bridge,
 409                                                1);
 410                                 return 0;
 411                         }
 412                         ether_addr_copy(eth_hdr(skb)->h_dest, dev->dev_addr);
 413                         skb->pkt_type = PACKET_HOST;
 414                 }
 415         } else {
 416                 rt = bridge_parent_rtable(nf_bridge->physindev);
 417                 if (!rt) {
 418                         kfree_skb(skb);
 419                         return 0;
 420                 }
 421                 skb_dst_set_noref(skb, &rt->dst);
 422         }
 423
 424         skb->dev = nf_bridge->physindev;
 425         nf_bridge_update_protocol(skb);
 426         nf_bridge_push_encap_header(skb);
 427         NF_HOOK_THRESH(NFPROTO_BRIDGE, NF_BR_PRE_ROUTING, skb, skb->dev, NULL,
 428                        br_handle_frame_finish, 1);
 429
 430         return 0;
 431 }
 432
 433 static struct net_device *brnf_get_logical_dev(struct sk_buff *skb, const struct net_device *dev)
 434 {
 435         struct net_device *vlan, *br;
 436
 437         br = bridge_parent(dev);
 438         if (brnf_pass_vlan_indev == 0 || !vlan_tx_tag_present(skb))
 439                 return br;
 440
 441         vlan = __vlan_find_dev_deep_rcu(br, skb->vlan_proto,
 442                                     vlan_tx_tag_get(skb) & VLAN_VID_MASK);
 443
 444         return vlan ? vlan : br;
 445 }
 446
 447 /* Some common code for IPv4/IPv6 */
 448 static struct net_device *setup_pre_routing(struct sk_buff *skb)
 449 {
 450         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 451
 452         if (skb->pkt_type == PACKET_OTHERHOST) {
 453                 skb->pkt_type = PACKET_HOST;
 454                 nf_bridge->mask |= BRNF_PKT_TYPE;
 455         }
 456
 457         nf_bridge->mask |= BRNF_NF_BRIDGE_PREROUTING;
 458         nf_bridge->physindev = skb->dev;
 459         skb->dev = brnf_get_logical_dev(skb, skb->dev);
 460         if (skb->protocol == htons(ETH_P_8021Q))
 461                 nf_bridge->mask |= BRNF_8021Q;
 462         else if (skb->protocol == htons(ETH_P_PPP_SES))
 463                 nf_bridge->mask |= BRNF_PPPoE;
 464
 465         /* Must drop socket now because of tproxy. */
 466         skb_orphan(skb);
 467         return skb->dev;
 468 }
 469
 470 /* We only check the length. A bridge shouldn't do any hop-by-hop stuff anyway */
 471 static int check_hbh_len(struct sk_buff *skb)
 472 {
 473         unsigned char *raw = (u8 *)(ipv6_hdr(skb) + 1);
 474         u32 pkt_len;
 475         const unsigned char *nh = skb_network_header(skb);
 476         int off = raw - nh;
 477         int len = (raw[1] + 1) << 3;
 478
 479         if ((raw + len) - skb->data > skb_headlen(skb))
 480                 goto bad;
 481
 482         off += 2;
 483         len -= 2;
 484
 485         while (len > 0) {
 486                 int optlen = nh[off + 1] + 2;
 487
 488                 switch (nh[off]) {
 489                 case IPV6_TLV_PAD1:
 490                         optlen = 1;
 491                         break;
 492
 493                 case IPV6_TLV_PADN:
 494                         break;
 495
 496                 case IPV6_TLV_JUMBO:
 497                         if (nh[off + 1] != 4 || (off & 3) != 2)
 498                                 goto bad;
 499                         pkt_len = ntohl(*(__be32 *) (nh + off + 2));
 500                         if (pkt_len <= IPV6_MAXPLEN ||
 501                             ipv6_hdr(skb)->payload_len)
 502                                 goto bad;
 503                         if (pkt_len > skb->len - sizeof(struct ipv6hdr))
 504                                 goto bad;
 505                         if (pskb_trim_rcsum(skb,
 506                                             pkt_len + sizeof(struct ipv6hdr)))
 507                                 goto bad;
 508                         nh = skb_network_header(skb);
 509                         break;
 510                 default:
 511                         if (optlen > len)
 512                                 goto bad;
 513                         break;
 514                 }
 515                 off += optlen;
 516                 len -= optlen;
 517         }
 518         if (len == 0)
 519                 return 0;
 520 bad:
 521         return -1;
 522
 523 }
 524
 525 /* Replicate the checks that IPv6 does on packet reception and pass the packet
 526  * to ip6tables, which doesn't support NAT, so things are fairly simple. */
 527 static unsigned int br_nf_pre_routing_ipv6(const struct nf_hook_ops *ops,
 528                                            struct sk_buff *skb,
 529                                            const struct net_device *in,
 530                                            const struct net_device *out,
 531                                            int (*okfn)(struct sk_buff *))
 532 {
 533         const struct ipv6hdr *hdr;
 534         u32 pkt_len;
 535
 536         if (skb->len < sizeof(struct ipv6hdr))
 537                 return NF_DROP;
 538
 539         if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
 540                 return NF_DROP;
 541
 542         hdr = ipv6_hdr(skb);
 543
 544         if (hdr->version != 6)
 545                 return NF_DROP;
 546
 547         pkt_len = ntohs(hdr->payload_len);
 548
 549         if (pkt_len || hdr->nexthdr != NEXTHDR_HOP) {
 550                 if (pkt_len + sizeof(struct ipv6hdr) > skb->len)
 551                         return NF_DROP;
 552                 if (pskb_trim_rcsum(skb, pkt_len + sizeof(struct ipv6hdr)))
 553                         return NF_DROP;
 554         }
 555         if (hdr->nexthdr == NEXTHDR_HOP && check_hbh_len(skb))
 556                 return NF_DROP;
 557
 558         nf_bridge_put(skb->nf_bridge);
 559         if (!nf_bridge_alloc(skb))
 560                 return NF_DROP;
 561         if (!setup_pre_routing(skb))
 562                 return NF_DROP;
 563
 564         skb->protocol = htons(ETH_P_IPV6);
 565         NF_HOOK(NFPROTO_IPV6, NF_INET_PRE_ROUTING, skb, skb->dev, NULL,
 566                 br_nf_pre_routing_finish_ipv6);
 567
 568         return NF_STOLEN;
 569 }
 570
 571 /* Direct IPv6 traffic to br_nf_pre_routing_ipv6.
 572  * Replicate the checks that IPv4 does on packet reception.
 573  * Set skb->dev to the bridge device (i.e. parent of the
 574  * receiving device) to make netfilter happy, the REDIRECT
 575  * target in particular.  Save the original destination IP
 576  * address to be able to detect DNAT afterwards. */
 577 static unsigned int br_nf_pre_routing(const struct nf_hook_ops *ops,
 578                                       struct sk_buff *skb,
 579                                       const struct net_device *in,
 580                                       const struct net_device *out,
 581                                       int (*okfn)(struct sk_buff *))
 582 {
 583         struct net_bridge_port *p;
 584         struct net_bridge *br;
 585         __u32 len = nf_bridge_encap_header_len(skb);
 586
 587         if (unlikely(!pskb_may_pull(skb, len)))
 588                 return NF_DROP;
 589
 590         p = br_port_get_rcu(in);
 591         if (p == NULL)
 592                 return NF_DROP;
 593         br = p->br;
 594
 595         if (IS_IPV6(skb) || IS_VLAN_IPV6(skb) || IS_PPPOE_IPV6(skb)) {
 596                 if (!brnf_call_ip6tables && !br->nf_call_ip6tables)
 597                         return NF_ACCEPT;
 598
 599                 nf_bridge_pull_encap_header_rcsum(skb);
 600                 return br_nf_pre_routing_ipv6(ops, skb, in, out, okfn);
 601         }
 602
 603         if (!brnf_call_iptables && !br->nf_call_iptables)
 604                 return NF_ACCEPT;
 605
 606         if (!IS_IP(skb) && !IS_VLAN_IP(skb) && !IS_PPPOE_IP(skb))
 607                 return NF_ACCEPT;
 608
 609         nf_bridge_pull_encap_header_rcsum(skb);
 610
 611         if (br_parse_ip_options(skb))
 612                 return NF_DROP;
 613
 614         nf_bridge_put(skb->nf_bridge);
 615         if (!nf_bridge_alloc(skb))
 616                 return NF_DROP;
 617         if (!setup_pre_routing(skb))
 618                 return NF_DROP;
 619         store_orig_dstaddr(skb);
 620         skb->protocol = htons(ETH_P_IP);
 621
 622         NF_HOOK(NFPROTO_IPV4, NF_INET_PRE_ROUTING, skb, skb->dev, NULL,
 623                 br_nf_pre_routing_finish);
 624
 625         return NF_STOLEN;
 626 }
 627
 628
 629 /* PF_BRIDGE/LOCAL_IN ************************************************/
 630 /* The packet is locally destined, which requires a real
 631  * dst_entry, so detach the fake one.  On the way up, the
 632  * packet would pass through PRE_ROUTING again (which already
 633  * took place when the packet entered the bridge), but we
 634  * register an IPv4 PRE_ROUTING 'sabotage' hook that will
 635  * prevent this from happening. */
 636 static unsigned int br_nf_local_in(const struct nf_hook_ops *ops,
 637                                    struct sk_buff *skb,
 638                                    const struct net_device *in,
 639                                    const struct net_device *out,
 640                                    int (*okfn)(struct sk_buff *))
 641 {
 642         br_drop_fake_rtable(skb);
 643         return NF_ACCEPT;
 644 }
 645
 646 /* PF_BRIDGE/FORWARD *************************************************/
 647 static int br_nf_forward_finish(struct sk_buff *skb)
 648 {
 649         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 650         struct net_device *in;
 651
 652         if (!IS_ARP(skb) && !IS_VLAN_ARP(skb)) {
 653                 in = nf_bridge->physindev;
 654                 if (nf_bridge->mask & BRNF_PKT_TYPE) {
 655                         skb->pkt_type = PACKET_OTHERHOST;
 656                         nf_bridge->mask ^= BRNF_PKT_TYPE;
 657                 }
 658                 nf_bridge_update_protocol(skb);
 659         } else {
 660                 in = *((struct net_device **)(skb->cb));
 661         }
 662         nf_bridge_push_encap_header(skb);
 663
 664         NF_HOOK_THRESH(NFPROTO_BRIDGE, NF_BR_FORWARD, skb, in,
 665                        skb->dev, br_forward_finish, 1);
 666         return 0;
 667 }
 668
 669
 670 /* This is the 'purely bridged' case.  For IP, we pass the packet to
 671  * netfilter with indev and outdev set to the bridge device,
 672  * but we are still able to filter on the 'real' indev/outdev
 673  * because of the physdev module. For ARP, indev and outdev are the
 674  * bridge ports. */
 675 static unsigned int br_nf_forward_ip(const struct nf_hook_ops *ops,
 676                                      struct sk_buff *skb,
 677                                      const struct net_device *in,
 678                                      const struct net_device *out,
 679                                      int (*okfn)(struct sk_buff *))
 680 {
 681         struct nf_bridge_info *nf_bridge;
 682         struct net_device *parent;
 683         u_int8_t pf;
 684
 685         if (!skb->nf_bridge)
 686                 return NF_ACCEPT;
 687
 688         /* Need exclusive nf_bridge_info since we might have multiple
 689          * different physoutdevs. */
 690         if (!nf_bridge_unshare(skb))
 691                 return NF_DROP;
 692
 693         parent = bridge_parent(out);
 694         if (!parent)
 695                 return NF_DROP;
 696
 697         if (IS_IP(skb) || IS_VLAN_IP(skb) || IS_PPPOE_IP(skb))
 698                 pf = NFPROTO_IPV4;
 699         else if (IS_IPV6(skb) || IS_VLAN_IPV6(skb) || IS_PPPOE_IPV6(skb))
 700                 pf = NFPROTO_IPV6;
 701         else
 702                 return NF_ACCEPT;
 703
 704         nf_bridge_pull_encap_header(skb);
 705
 706         nf_bridge = skb->nf_bridge;
 707         if (skb->pkt_type == PACKET_OTHERHOST) {
 708                 skb->pkt_type = PACKET_HOST;
 709                 nf_bridge->mask |= BRNF_PKT_TYPE;
 710         }
 711
 712         if (pf == NFPROTO_IPV4 && br_parse_ip_options(skb))
 713                 return NF_DROP;
 714
 715         /* The physdev module checks on this */
 716         nf_bridge->mask |= BRNF_BRIDGED;
 717         nf_bridge->physoutdev = skb->dev;
 718         if (pf == NFPROTO_IPV4)
 719                 skb->protocol = htons(ETH_P_IP);
 720         else
 721                 skb->protocol = htons(ETH_P_IPV6);
 722
 723         NF_HOOK(pf, NF_INET_FORWARD, skb, brnf_get_logical_dev(skb, in), parent,
 724                 br_nf_forward_finish);
 725
 726         return NF_STOLEN;
 727 }
 728
 729 static unsigned int br_nf_forward_arp(const struct nf_hook_ops *ops,
 730                                       struct sk_buff *skb,
 731                                       const struct net_device *in,
 732                                       const struct net_device *out,
 733                                       int (*okfn)(struct sk_buff *))
 734 {
 735         struct net_bridge_port *p;
 736         struct net_bridge *br;
 737         struct net_device **d = (struct net_device **)(skb->cb);
 738
 739         p = br_port_get_rcu(out);
 740         if (p == NULL)
 741                 return NF_ACCEPT;
 742         br = p->br;
 743
 744         if (!brnf_call_arptables && !br->nf_call_arptables)
 745                 return NF_ACCEPT;
 746
 747         if (!IS_ARP(skb)) {
 748                 if (!IS_VLAN_ARP(skb))
 749                         return NF_ACCEPT;
 750                 nf_bridge_pull_encap_header(skb);
 751         }
 752
 753         if (arp_hdr(skb)->ar_pln != 4) {
 754                 if (IS_VLAN_ARP(skb))
 755                         nf_bridge_push_encap_header(skb);
 756                 return NF_ACCEPT;
 757         }
 758         *d = (struct net_device *)in;
 759         NF_HOOK(NFPROTO_ARP, NF_ARP_FORWARD, skb, (struct net_device *)in,
 760                 (struct net_device *)out, br_nf_forward_finish);
 761
 762         return NF_STOLEN;
 763 }
 764
 765 #if IS_ENABLED(CONFIG_NF_DEFRAG_IPV4)
 766 static int br_nf_dev_queue_xmit(struct sk_buff *skb)
 767 {
 768         int ret;
 769         int frag_max_size;
 770
 771         /* This is wrong! We should preserve the original fragment
 772          * boundaries by preserving frag_list rather than refragmenting.
 773          */
 774         if (skb->protocol == htons(ETH_P_IP) &&
 775             skb->len + nf_bridge_mtu_reduction(skb) > skb->dev->mtu &&
 776             !skb_is_gso(skb)) {
 777                 frag_max_size = BR_INPUT_SKB_CB(skb)->frag_max_size;
 778                 if (br_parse_ip_options(skb))
 779                         /* Drop invalid packet */
 780                         return NF_DROP;
 781                 IPCB(skb)->frag_max_size = frag_max_size;
 782                 ret = ip_fragment(skb, br_dev_queue_push_xmit);
 783         } else
 784                 ret = br_dev_queue_push_xmit(skb);
 785
 786         return ret;
 787 }
 788 #else
 789 static int br_nf_dev_queue_xmit(struct sk_buff *skb)
 790 {
 791         return br_dev_queue_push_xmit(skb);
 792 }
 793 #endif
 794
 795 /* PF_BRIDGE/POST_ROUTING ********************************************/
 796 static unsigned int br_nf_post_routing(const struct nf_hook_ops *ops,
 797                                        struct sk_buff *skb,
 798                                        const struct net_device *in,
 799                                        const struct net_device *out,
 800                                        int (*okfn)(struct sk_buff *))
 801 {
 802         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 803         struct net_device *realoutdev = bridge_parent(skb->dev);
 804         u_int8_t pf;
 805
 806         if (!nf_bridge || !(nf_bridge->mask & BRNF_BRIDGED))
 807                 return NF_ACCEPT;
 808
 809         if (!realoutdev)
 810                 return NF_DROP;
 811
 812         if (IS_IP(skb) || IS_VLAN_IP(skb) || IS_PPPOE_IP(skb))
 813                 pf = NFPROTO_IPV4;
 814         else if (IS_IPV6(skb) || IS_VLAN_IPV6(skb) || IS_PPPOE_IPV6(skb))
 815                 pf = NFPROTO_IPV6;
 816         else
 817                 return NF_ACCEPT;
 818
 819         /* We assume any code from br_dev_queue_push_xmit onwards doesn't care
 820          * about the value of skb->pkt_type. */
 821         if (skb->pkt_type == PACKET_OTHERHOST) {
 822                 skb->pkt_type = PACKET_HOST;
 823                 nf_bridge->mask |= BRNF_PKT_TYPE;
 824         }
 825
 826         nf_bridge_pull_encap_header(skb);
 827         nf_bridge_save_header(skb);
 828         if (pf == NFPROTO_IPV4)
 829                 skb->protocol = htons(ETH_P_IP);
 830         else
 831                 skb->protocol = htons(ETH_P_IPV6);
 832
 833         NF_HOOK(pf, NF_INET_POST_ROUTING, skb, NULL, realoutdev,
 834                 br_nf_dev_queue_xmit);
 835
 836         return NF_STOLEN;
 837 }
 838
 839 /* IP/SABOTAGE *****************************************************/
 840 /* Don't hand locally destined packets to PF_INET(6)/PRE_ROUTING
 841  * for the second time. */
 842 static unsigned int ip_sabotage_in(const struct nf_hook_ops *ops,
 843                                    struct sk_buff *skb,
 844                                    const struct net_device *in,
 845                                    const struct net_device *out,
 846                                    int (*okfn)(struct sk_buff *))
 847 {
 848         if (skb->nf_bridge &&
 849             !(skb->nf_bridge->mask & BRNF_NF_BRIDGE_PREROUTING)) {
 850                 return NF_STOP;
 851         }
 852
 853         return NF_ACCEPT;
 854 }
 855
 856 void br_netfilter_enable(void)
 857 {
 858 }
 859 EXPORT_SYMBOL_GPL(br_netfilter_enable);
 860
 861 /* For br_nf_post_routing, we need (prio = NF_BR_PRI_LAST), because
 862  * br_dev_queue_push_xmit is called afterwards */
 863 static struct nf_hook_ops br_nf_ops[] __read_mostly = {
 864         {
 865                 .hook = br_nf_pre_routing,
 866                 .owner = THIS_MODULE,
 867                 .pf = NFPROTO_BRIDGE,
 868                 .hooknum = NF_BR_PRE_ROUTING,
 869                 .priority = NF_BR_PRI_BRNF,
 870         },
 871         {
 872                 .hook = br_nf_local_in,
 873                 .owner = THIS_MODULE,
 874                 .pf = NFPROTO_BRIDGE,
 875                 .hooknum = NF_BR_LOCAL_IN,
 876                 .priority = NF_BR_PRI_BRNF,
 877         },
 878         {
 879                 .hook = br_nf_forward_ip,
 880                 .owner = THIS_MODULE,
 881                 .pf = NFPROTO_BRIDGE,
 882                 .hooknum = NF_BR_FORWARD,
 883                 .priority = NF_BR_PRI_BRNF - 1,
 884         },
 885         {
 886                 .hook = br_nf_forward_arp,
 887                 .owner = THIS_MODULE,
 888                 .pf = NFPROTO_BRIDGE,
 889                 .hooknum = NF_BR_FORWARD,
 890                 .priority = NF_BR_PRI_BRNF,
 891         },
 892         {
 893                 .hook = br_nf_post_routing,
 894                 .owner = THIS_MODULE,
 895                 .pf = NFPROTO_BRIDGE,
 896                 .hooknum = NF_BR_POST_ROUTING,
 897                 .priority = NF_BR_PRI_LAST,
 898         },
 899         {
 900                 .hook = ip_sabotage_in,
 901                 .owner = THIS_MODULE,
 902                 .pf = NFPROTO_IPV4,
 903                 .hooknum = NF_INET_PRE_ROUTING,
 904                 .priority = NF_IP_PRI_FIRST,
 905         },
 906         {
 907                 .hook = ip_sabotage_in,
 908                 .owner = THIS_MODULE,
 909                 .pf = NFPROTO_IPV6,
 910                 .hooknum = NF_INET_PRE_ROUTING,
 911                 .priority = NF_IP6_PRI_FIRST,
 912         },
 913 };
 914
 915 #ifdef CONFIG_SYSCTL
 916 static
 917 int brnf_sysctl_call_tables(struct ctl_table *ctl, int write,
 918                             void __user *buffer, size_t *lenp, loff_t *ppos)
 919 {
 920         int ret;
 921
 922         ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
 923
 924         if (write && *(int *)(ctl->data))
 925                 *(int *)(ctl->data) = 1;
 926         return ret;
 927 }
 928
 929 static struct ctl_table brnf_table[] = {
 930         {
 931                 .procname       = "bridge-nf-call-arptables",
 932                 .data           = &brnf_call_arptables,
 933                 .maxlen         = sizeof(int),
 934                 .mode           = 0644,
 935                 .proc_handler   = brnf_sysctl_call_tables,
 936         },
 937         {
 938                 .procname       = "bridge-nf-call-iptables",
 939                 .data           = &brnf_call_iptables,
 940                 .maxlen         = sizeof(int),
 941                 .mode           = 0644,
 942                 .proc_handler   = brnf_sysctl_call_tables,
 943         },
 944         {
 945                 .procname       = "bridge-nf-call-ip6tables",
 946                 .data           = &brnf_call_ip6tables,
 947                 .maxlen         = sizeof(int),
 948                 .mode           = 0644,
 949                 .proc_handler   = brnf_sysctl_call_tables,
 950         },
 951         {
 952                 .procname       = "bridge-nf-filter-vlan-tagged",
 953                 .data           = &brnf_filter_vlan_tagged,
 954                 .maxlen         = sizeof(int),
 955                 .mode           = 0644,
 956                 .proc_handler   = brnf_sysctl_call_tables,
 957         },
 958         {
 959                 .procname       = "bridge-nf-filter-pppoe-tagged",
 960                 .data           = &brnf_filter_pppoe_tagged,
 961                 .maxlen         = sizeof(int),
 962                 .mode           = 0644,
 963                 .proc_handler   = brnf_sysctl_call_tables,
 964         },
 965         {
 966                 .procname       = "bridge-nf-pass-vlan-input-dev",
 967                 .data           = &brnf_pass_vlan_indev,
 968                 .maxlen         = sizeof(int),
 969                 .mode           = 0644,
 970                 .proc_handler   = brnf_sysctl_call_tables,
 971         },
 972         { }
 973 };
 974 #endif
 975
 976 static int __init br_netfilter_init(void)
 977 {
 978         int ret;
 979
 980         ret = nf_register_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
 981         if (ret < 0)
 982                 return ret;
 983
 984 #ifdef CONFIG_SYSCTL
 985         brnf_sysctl_header = register_net_sysctl(&init_net, "net/bridge", brnf_table);
 986         if (brnf_sysctl_header == NULL) {
 987                 printk(KERN_WARNING
 988                        "br_netfilter: can't register to sysctl.\n");
 989                 ret = -ENOMEM;
 990                 goto err1;
 991         }
 992 #endif
 993         printk(KERN_NOTICE "Bridge firewalling registered\n");
 994         return 0;
 995 err1:
 996         nf_unregister_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
 997         return ret;
 998 }
 999
1000 static void __exit br_netfilter_fini(void)
1001 {
1002         nf_unregister_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
1003 #ifdef CONFIG_SYSCTL
1004         unregister_net_sysctl_table(brnf_sysctl_header);
1005 #endif
1006 }
1007
1008 module_init(br_netfilter_init);
1009 module_exit(br_netfilter_fini);
1010
1011 MODULE_LICENSE("GPL");
1012 MODULE_AUTHOR("Lennert Buytenhek <buytenh@gnu.org>");
1013 MODULE_AUTHOR("Bart De Schuymer <bdschuym@pandora.be>");
1014 MODULE_DESCRIPTION("Linux ethernet netfilter firewall bridge");