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 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
  41 #include <net/netfilter/nf_conntrack.h>
  42 #endif
  43
  44 #include <asm/uaccess.h>
  45 #include "br_private.h"
  46 #ifdef CONFIG_SYSCTL
  47 #include <linux/sysctl.h>
  48 #endif
  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         (!skb_vlan_tag_present(skb) && skb->protocol == htons(ETH_P_IP))
  69
  70 #define IS_IPV6(skb) \
  71         (!skb_vlan_tag_present(skb) && skb->protocol == htons(ETH_P_IPV6))
  72
  73 #define IS_ARP(skb) \
  74         (!skb_vlan_tag_present(skb) && skb->protocol == htons(ETH_P_ARP))
  75
  76 static inline __be16 vlan_proto(const struct sk_buff *skb)
  77 {
  78         if (skb_vlan_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 /* largest possible L2 header, see br_nf_dev_queue_xmit() */
 115 #define NF_BRIDGE_MAX_MAC_HEADER_LENGTH (PPPOE_SES_HLEN + ETH_HLEN)
 116
 117 #if IS_ENABLED(CONFIG_NF_DEFRAG_IPV4)
 118 struct brnf_frag_data {
 119         char mac[NF_BRIDGE_MAX_MAC_HEADER_LENGTH];
 120         u8 encap_size;
 121         u8 size;
 122 };
 123
 124 static DEFINE_PER_CPU(struct brnf_frag_data, brnf_frag_data_storage);
 125 #endif
 126
 127 static struct nf_bridge_info *nf_bridge_info_get(const struct sk_buff *skb)
 128 {
 129         return skb->nf_bridge;
 130 }
 131
 132 static inline struct rtable *bridge_parent_rtable(const struct net_device *dev)
 133 {
 134         struct net_bridge_port *port;
 135
 136         port = br_port_get_rcu(dev);
 137         return port ? &port->br->fake_rtable : NULL;
 138 }
 139
 140 static inline struct net_device *bridge_parent(const struct net_device *dev)
 141 {
 142         struct net_bridge_port *port;
 143
 144         port = br_port_get_rcu(dev);
 145         return port ? port->br->dev : NULL;
 146 }
 147
 148 static inline struct nf_bridge_info *nf_bridge_alloc(struct sk_buff *skb)
 149 {
 150         skb->nf_bridge = kzalloc(sizeof(struct nf_bridge_info), GFP_ATOMIC);
 151         if (likely(skb->nf_bridge))
 152                 atomic_set(&(skb->nf_bridge->use), 1);
 153
 154         return skb->nf_bridge;
 155 }
 156
 157 static inline struct nf_bridge_info *nf_bridge_unshare(struct sk_buff *skb)
 158 {
 159         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
 160
 161         if (atomic_read(&nf_bridge->use) > 1) {
 162                 struct nf_bridge_info *tmp = nf_bridge_alloc(skb);
 163
 164                 if (tmp) {
 165                         memcpy(tmp, nf_bridge, sizeof(struct nf_bridge_info));
 166                         atomic_set(&tmp->use, 1);
 167                 }
 168                 nf_bridge_put(nf_bridge);
 169                 nf_bridge = tmp;
 170         }
 171         return nf_bridge;
 172 }
 173
 174 static unsigned int nf_bridge_encap_header_len(const struct sk_buff *skb)
 175 {
 176         switch (skb->protocol) {
 177         case __cpu_to_be16(ETH_P_8021Q):
 178                 return VLAN_HLEN;
 179         case __cpu_to_be16(ETH_P_PPP_SES):
 180                 return PPPOE_SES_HLEN;
 181         default:
 182                 return 0;
 183         }
 184 }
 185
 186 static inline void nf_bridge_push_encap_header(struct sk_buff *skb)
 187 {
 188         unsigned int len = nf_bridge_encap_header_len(skb);
 189
 190         skb_push(skb, len);
 191         skb->network_header -= len;
 192 }
 193
 194 static inline void nf_bridge_pull_encap_header(struct sk_buff *skb)
 195 {
 196         unsigned int len = nf_bridge_encap_header_len(skb);
 197
 198         skb_pull(skb, len);
 199         skb->network_header += len;
 200 }
 201
 202 static inline void nf_bridge_pull_encap_header_rcsum(struct sk_buff *skb)
 203 {
 204         unsigned int len = nf_bridge_encap_header_len(skb);
 205
 206         skb_pull_rcsum(skb, len);
 207         skb->network_header += len;
 208 }
 209
 210 /* When handing a packet over to the IP layer
 211  * check whether we have a skb that is in the
 212  * expected format
 213  */
 214
 215 static int br_parse_ip_options(struct sk_buff *skb)
 216 {
 217         const struct iphdr *iph;
 218         struct net_device *dev = skb->dev;
 219         u32 len;
 220
 221         if (!pskb_may_pull(skb, sizeof(struct iphdr)))
 222                 goto inhdr_error;
 223
 224         iph = ip_hdr(skb);
 225
 226         /* Basic sanity checks */
 227         if (iph->ihl < 5 || iph->version != 4)
 228                 goto inhdr_error;
 229
 230         if (!pskb_may_pull(skb, iph->ihl*4))
 231                 goto inhdr_error;
 232
 233         iph = ip_hdr(skb);
 234         if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl)))
 235                 goto inhdr_error;
 236
 237         len = ntohs(iph->tot_len);
 238         if (skb->len < len) {
 239                 IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INTRUNCATEDPKTS);
 240                 goto drop;
 241         } else if (len < (iph->ihl*4))
 242                 goto inhdr_error;
 243
 244         if (pskb_trim_rcsum(skb, len)) {
 245                 IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INDISCARDS);
 246                 goto drop;
 247         }
 248
 249         memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
 250         /* We should really parse IP options here but until
 251          * somebody who actually uses IP options complains to
 252          * us we'll just silently ignore the options because
 253          * we're lazy!
 254          */
 255         return 0;
 256
 257 inhdr_error:
 258         IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INHDRERRORS);
 259 drop:
 260         return -1;
 261 }
 262
 263 static void nf_bridge_update_protocol(struct sk_buff *skb)
 264 {
 265         switch (skb->nf_bridge->orig_proto) {
 266         case BRNF_PROTO_8021Q:
 267                 skb->protocol = htons(ETH_P_8021Q);
 268                 break;
 269         case BRNF_PROTO_PPPOE:
 270                 skb->protocol = htons(ETH_P_PPP_SES);
 271                 break;
 272         case BRNF_PROTO_UNCHANGED:
 273                 break;
 274         }
 275 }
 276
 277 /* PF_BRIDGE/PRE_ROUTING *********************************************/
 278 /* Undo the changes made for ip6tables PREROUTING and continue the
 279  * bridge PRE_ROUTING hook. */
 280 static int br_nf_pre_routing_finish_ipv6(struct sk_buff *skb)
 281 {
 282         struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
 283         struct rtable *rt;
 284
 285         if (nf_bridge->pkt_otherhost) {
 286                 skb->pkt_type = PACKET_OTHERHOST;
 287                 nf_bridge->pkt_otherhost = false;
 288         }
 289         nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
 290
 291         rt = bridge_parent_rtable(nf_bridge->physindev);
 292         if (!rt) {
 293                 kfree_skb(skb);
 294                 return 0;
 295         }
 296         skb_dst_set_noref(skb, &rt->dst);
 297
 298         skb->dev = nf_bridge->physindev;
 299         nf_bridge_update_protocol(skb);
 300         nf_bridge_push_encap_header(skb);
 301         NF_HOOK_THRESH(NFPROTO_BRIDGE, NF_BR_PRE_ROUTING, skb, skb->dev, NULL,
 302                        br_handle_frame_finish, 1);
 303
 304         return 0;
 305 }
 306
 307 /* Obtain the correct destination MAC address, while preserving the original
 308  * source MAC address. If we already know this address, we just copy it. If we
 309  * don't, we use the neighbour framework to find out. In both cases, we make
 310  * sure that br_handle_frame_finish() is called afterwards.
 311  */
 312 static int br_nf_pre_routing_finish_bridge(struct sk_buff *skb)
 313 {
 314         struct neighbour *neigh;
 315         struct dst_entry *dst;
 316
 317         skb->dev = bridge_parent(skb->dev);
 318         if (!skb->dev)
 319                 goto free_skb;
 320         dst = skb_dst(skb);
 321         neigh = dst_neigh_lookup_skb(dst, skb);
 322         if (neigh) {
 323                 struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
 324                 int ret;
 325
 326                 if (neigh->hh.hh_len) {
 327                         neigh_hh_bridge(&neigh->hh, skb);
 328                         skb->dev = nf_bridge->physindev;
 329                         ret = br_handle_frame_finish(skb);
 330                 } else {
 331                         /* the neighbour function below overwrites the complete
 332                          * MAC header, so we save the Ethernet source address and
 333                          * protocol number.
 334                          */
 335                         skb_copy_from_linear_data_offset(skb,
 336                                                          -(ETH_HLEN-ETH_ALEN),
 337                                                          nf_bridge->neigh_header,
 338                                                          ETH_HLEN-ETH_ALEN);
 339                         /* tell br_dev_xmit to continue with forwarding */
 340                         nf_bridge->mask |= BRNF_BRIDGED_DNAT;
 341                         /* FIXME Need to refragment */
 342                         ret = neigh->output(neigh, skb);
 343                 }
 344                 neigh_release(neigh);
 345                 return ret;
 346         }
 347 free_skb:
 348         kfree_skb(skb);
 349         return 0;
 350 }
 351
 352 static bool dnat_took_place(const struct sk_buff *skb)
 353 {
 354 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
 355         enum ip_conntrack_info ctinfo;
 356         struct nf_conn *ct;
 357
 358         ct = nf_ct_get(skb, &ctinfo);
 359         if (!ct || nf_ct_is_untracked(ct))
 360                 return false;
 361
 362         return test_bit(IPS_DST_NAT_BIT, &ct->status);
 363 #else
 364         return false;
 365 #endif
 366 }
 367
 368 /* This requires some explaining. If DNAT has taken place,
 369  * we will need to fix up the destination Ethernet address.
 370  *
 371  * There are two cases to consider:
 372  * 1. The packet was DNAT'ed to a device in the same bridge
 373  *    port group as it was received on. We can still bridge
 374  *    the packet.
 375  * 2. The packet was DNAT'ed to a different device, either
 376  *    a non-bridged device or another bridge port group.
 377  *    The packet will need to be routed.
 378  *
 379  * The correct way of distinguishing between these two cases is to
 380  * call ip_route_input() and to look at skb->dst->dev, which is
 381  * changed to the destination device if ip_route_input() succeeds.
 382  *
 383  * Let's first consider the case that ip_route_input() succeeds:
 384  *
 385  * If the output device equals the logical bridge device the packet
 386  * came in on, we can consider this bridging. The corresponding MAC
 387  * address will be obtained in br_nf_pre_routing_finish_bridge.
 388  * Otherwise, the packet is considered to be routed and we just
 389  * change the destination MAC address so that the packet will
 390  * later be passed up to the IP stack to be routed. For a redirected
 391  * packet, ip_route_input() will give back the localhost as output device,
 392  * which differs from the bridge device.
 393  *
 394  * Let's now consider the case that ip_route_input() fails:
 395  *
 396  * This can be because the destination address is martian, in which case
 397  * the packet will be dropped.
 398  * If IP forwarding is disabled, ip_route_input() will fail, while
 399  * ip_route_output_key() can return success. The source
 400  * address for ip_route_output_key() is set to zero, so ip_route_output_key()
 401  * thinks we're handling a locally generated packet and won't care
 402  * if IP forwarding is enabled. If the output device equals the logical bridge
 403  * device, we proceed as if ip_route_input() succeeded. If it differs from the
 404  * logical bridge port or if ip_route_output_key() fails we drop the packet.
 405  */
 406 static int br_nf_pre_routing_finish(struct sk_buff *skb)
 407 {
 408         struct net_device *dev = skb->dev;
 409         struct iphdr *iph = ip_hdr(skb);
 410         struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
 411         struct rtable *rt;
 412         int err;
 413         int frag_max_size;
 414
 415         frag_max_size = IPCB(skb)->frag_max_size;
 416         BR_INPUT_SKB_CB(skb)->frag_max_size = frag_max_size;
 417
 418         if (nf_bridge->pkt_otherhost) {
 419                 skb->pkt_type = PACKET_OTHERHOST;
 420                 nf_bridge->pkt_otherhost = false;
 421         }
 422         nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
 423         if (dnat_took_place(skb)) {
 424                 if ((err = ip_route_input(skb, iph->daddr, iph->saddr, iph->tos, dev))) {
 425                         struct in_device *in_dev = __in_dev_get_rcu(dev);
 426
 427                         /* If err equals -EHOSTUNREACH the error is due to a
 428                          * martian destination or due to the fact that
 429                          * forwarding is disabled. For most martian packets,
 430                          * ip_route_output_key() will fail. It won't fail for 2 types of
 431                          * martian destinations: loopback destinations and destination
 432                          * 0.0.0.0. In both cases the packet will be dropped because the
 433                          * destination is the loopback device and not the bridge. */
 434                         if (err != -EHOSTUNREACH || !in_dev || IN_DEV_FORWARD(in_dev))
 435                                 goto free_skb;
 436
 437                         rt = ip_route_output(dev_net(dev), iph->daddr, 0,
 438                                              RT_TOS(iph->tos), 0);
 439                         if (!IS_ERR(rt)) {
 440                                 /* - Bridged-and-DNAT'ed traffic doesn't
 441                                  *   require ip_forwarding. */
 442                                 if (rt->dst.dev == dev) {
 443                                         skb_dst_set(skb, &rt->dst);
 444                                         goto bridged_dnat;
 445                                 }
 446                                 ip_rt_put(rt);
 447                         }
 448 free_skb:
 449                         kfree_skb(skb);
 450                         return 0;
 451                 } else {
 452                         if (skb_dst(skb)->dev == dev) {
 453 bridged_dnat:
 454                                 skb->dev = nf_bridge->physindev;
 455                                 nf_bridge_update_protocol(skb);
 456                                 nf_bridge_push_encap_header(skb);
 457                                 NF_HOOK_THRESH(NFPROTO_BRIDGE,
 458                                                NF_BR_PRE_ROUTING,
 459                                                skb, skb->dev, NULL,
 460                                                br_nf_pre_routing_finish_bridge,
 461                                                1);
 462                                 return 0;
 463                         }
 464                         ether_addr_copy(eth_hdr(skb)->h_dest, dev->dev_addr);
 465                         skb->pkt_type = PACKET_HOST;
 466                 }
 467         } else {
 468                 rt = bridge_parent_rtable(nf_bridge->physindev);
 469                 if (!rt) {
 470                         kfree_skb(skb);
 471                         return 0;
 472                 }
 473                 skb_dst_set_noref(skb, &rt->dst);
 474         }
 475
 476         skb->dev = nf_bridge->physindev;
 477         nf_bridge_update_protocol(skb);
 478         nf_bridge_push_encap_header(skb);
 479         NF_HOOK_THRESH(NFPROTO_BRIDGE, NF_BR_PRE_ROUTING, skb, skb->dev, NULL,
 480                        br_handle_frame_finish, 1);
 481
 482         return 0;
 483 }
 484
 485 static struct net_device *brnf_get_logical_dev(struct sk_buff *skb, const struct net_device *dev)
 486 {
 487         struct net_device *vlan, *br;
 488
 489         br = bridge_parent(dev);
 490         if (brnf_pass_vlan_indev == 0 || !skb_vlan_tag_present(skb))
 491                 return br;
 492
 493         vlan = __vlan_find_dev_deep_rcu(br, skb->vlan_proto,
 494                                     skb_vlan_tag_get(skb) & VLAN_VID_MASK);
 495
 496         return vlan ? vlan : br;
 497 }
 498
 499 /* Some common code for IPv4/IPv6 */
 500 static struct net_device *setup_pre_routing(struct sk_buff *skb)
 501 {
 502         struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
 503
 504         if (skb->pkt_type == PACKET_OTHERHOST) {
 505                 skb->pkt_type = PACKET_HOST;
 506                 nf_bridge->pkt_otherhost = true;
 507         }
 508
 509         nf_bridge->mask |= BRNF_NF_BRIDGE_PREROUTING;
 510         nf_bridge->physindev = skb->dev;
 511         skb->dev = brnf_get_logical_dev(skb, skb->dev);
 512
 513         if (skb->protocol == htons(ETH_P_8021Q))
 514                 nf_bridge->orig_proto = BRNF_PROTO_8021Q;
 515         else if (skb->protocol == htons(ETH_P_PPP_SES))
 516                 nf_bridge->orig_proto = BRNF_PROTO_PPPOE;
 517
 518         /* Must drop socket now because of tproxy. */
 519         skb_orphan(skb);
 520         return skb->dev;
 521 }
 522
 523 /* We only check the length. A bridge shouldn't do any hop-by-hop stuff anyway */
 524 static int check_hbh_len(struct sk_buff *skb)
 525 {
 526         unsigned char *raw = (u8 *)(ipv6_hdr(skb) + 1);
 527         u32 pkt_len;
 528         const unsigned char *nh = skb_network_header(skb);
 529         int off = raw - nh;
 530         int len = (raw[1] + 1) << 3;
 531
 532         if ((raw + len) - skb->data > skb_headlen(skb))
 533                 goto bad;
 534
 535         off += 2;
 536         len -= 2;
 537
 538         while (len > 0) {
 539                 int optlen = nh[off + 1] + 2;
 540
 541                 switch (nh[off]) {
 542                 case IPV6_TLV_PAD1:
 543                         optlen = 1;
 544                         break;
 545
 546                 case IPV6_TLV_PADN:
 547                         break;
 548
 549                 case IPV6_TLV_JUMBO:
 550                         if (nh[off + 1] != 4 || (off & 3) != 2)
 551                                 goto bad;
 552                         pkt_len = ntohl(*(__be32 *) (nh + off + 2));
 553                         if (pkt_len <= IPV6_MAXPLEN ||
 554                             ipv6_hdr(skb)->payload_len)
 555                                 goto bad;
 556                         if (pkt_len > skb->len - sizeof(struct ipv6hdr))
 557                                 goto bad;
 558                         if (pskb_trim_rcsum(skb,
 559                                             pkt_len + sizeof(struct ipv6hdr)))
 560                                 goto bad;
 561                         nh = skb_network_header(skb);
 562                         break;
 563                 default:
 564                         if (optlen > len)
 565                                 goto bad;
 566                         break;
 567                 }
 568                 off += optlen;
 569                 len -= optlen;
 570         }
 571         if (len == 0)
 572                 return 0;
 573 bad:
 574         return -1;
 575
 576 }
 577
 578 /* Replicate the checks that IPv6 does on packet reception and pass the packet
 579  * to ip6tables, which doesn't support NAT, so things are fairly simple. */
 580 static unsigned int br_nf_pre_routing_ipv6(const struct nf_hook_ops *ops,
 581                                            struct sk_buff *skb,
 582                                            const struct net_device *in,
 583                                            const struct net_device *out,
 584                                            int (*okfn)(struct sk_buff *))
 585 {
 586         const struct ipv6hdr *hdr;
 587         u32 pkt_len;
 588
 589         if (skb->len < sizeof(struct ipv6hdr))
 590                 return NF_DROP;
 591
 592         if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
 593                 return NF_DROP;
 594
 595         hdr = ipv6_hdr(skb);
 596
 597         if (hdr->version != 6)
 598                 return NF_DROP;
 599
 600         pkt_len = ntohs(hdr->payload_len);
 601
 602         if (pkt_len || hdr->nexthdr != NEXTHDR_HOP) {
 603                 if (pkt_len + sizeof(struct ipv6hdr) > skb->len)
 604                         return NF_DROP;
 605                 if (pskb_trim_rcsum(skb, pkt_len + sizeof(struct ipv6hdr)))
 606                         return NF_DROP;
 607         }
 608         if (hdr->nexthdr == NEXTHDR_HOP && check_hbh_len(skb))
 609                 return NF_DROP;
 610
 611         nf_bridge_put(skb->nf_bridge);
 612         if (!nf_bridge_alloc(skb))
 613                 return NF_DROP;
 614         if (!setup_pre_routing(skb))
 615                 return NF_DROP;
 616
 617         skb->protocol = htons(ETH_P_IPV6);
 618         NF_HOOK(NFPROTO_IPV6, NF_INET_PRE_ROUTING, skb, skb->dev, NULL,
 619                 br_nf_pre_routing_finish_ipv6);
 620
 621         return NF_STOLEN;
 622 }
 623
 624 /* Direct IPv6 traffic to br_nf_pre_routing_ipv6.
 625  * Replicate the checks that IPv4 does on packet reception.
 626  * Set skb->dev to the bridge device (i.e. parent of the
 627  * receiving device) to make netfilter happy, the REDIRECT
 628  * target in particular.  Save the original destination IP
 629  * address to be able to detect DNAT afterwards. */
 630 static unsigned int br_nf_pre_routing(const struct nf_hook_ops *ops,
 631                                       struct sk_buff *skb,
 632                                       const struct net_device *in,
 633                                       const struct net_device *out,
 634                                       int (*okfn)(struct sk_buff *))
 635 {
 636         struct net_bridge_port *p;
 637         struct net_bridge *br;
 638         __u32 len = nf_bridge_encap_header_len(skb);
 639
 640         if (unlikely(!pskb_may_pull(skb, len)))
 641                 return NF_DROP;
 642
 643         p = br_port_get_rcu(in);
 644         if (p == NULL)
 645                 return NF_DROP;
 646         br = p->br;
 647
 648         if (IS_IPV6(skb) || IS_VLAN_IPV6(skb) || IS_PPPOE_IPV6(skb)) {
 649                 if (!brnf_call_ip6tables && !br->nf_call_ip6tables)
 650                         return NF_ACCEPT;
 651
 652                 nf_bridge_pull_encap_header_rcsum(skb);
 653                 return br_nf_pre_routing_ipv6(ops, skb, in, out, okfn);
 654         }
 655
 656         if (!brnf_call_iptables && !br->nf_call_iptables)
 657                 return NF_ACCEPT;
 658
 659         if (!IS_IP(skb) && !IS_VLAN_IP(skb) && !IS_PPPOE_IP(skb))
 660                 return NF_ACCEPT;
 661
 662         nf_bridge_pull_encap_header_rcsum(skb);
 663
 664         if (br_parse_ip_options(skb))
 665                 return NF_DROP;
 666
 667         nf_bridge_put(skb->nf_bridge);
 668         if (!nf_bridge_alloc(skb))
 669                 return NF_DROP;
 670         if (!setup_pre_routing(skb))
 671                 return NF_DROP;
 672
 673         skb->protocol = htons(ETH_P_IP);
 674
 675         NF_HOOK(NFPROTO_IPV4, NF_INET_PRE_ROUTING, skb, skb->dev, NULL,
 676                 br_nf_pre_routing_finish);
 677
 678         return NF_STOLEN;
 679 }
 680
 681
 682 /* PF_BRIDGE/LOCAL_IN ************************************************/
 683 /* The packet is locally destined, which requires a real
 684  * dst_entry, so detach the fake one.  On the way up, the
 685  * packet would pass through PRE_ROUTING again (which already
 686  * took place when the packet entered the bridge), but we
 687  * register an IPv4 PRE_ROUTING 'sabotage' hook that will
 688  * prevent this from happening. */
 689 static unsigned int br_nf_local_in(const struct nf_hook_ops *ops,
 690                                    struct sk_buff *skb,
 691                                    const struct net_device *in,
 692                                    const struct net_device *out,
 693                                    int (*okfn)(struct sk_buff *))
 694 {
 695         br_drop_fake_rtable(skb);
 696         return NF_ACCEPT;
 697 }
 698
 699 /* PF_BRIDGE/FORWARD *************************************************/
 700 static int br_nf_forward_finish(struct sk_buff *skb)
 701 {
 702         struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
 703         struct net_device *in;
 704
 705         if (!IS_ARP(skb) && !IS_VLAN_ARP(skb)) {
 706                 int frag_max_size;
 707
 708                 if (skb->protocol == htons(ETH_P_IP)) {
 709                         frag_max_size = IPCB(skb)->frag_max_size;
 710                         BR_INPUT_SKB_CB(skb)->frag_max_size = frag_max_size;
 711                 }
 712
 713                 in = nf_bridge->physindev;
 714                 if (nf_bridge->pkt_otherhost) {
 715                         skb->pkt_type = PACKET_OTHERHOST;
 716                         nf_bridge->pkt_otherhost = false;
 717                 }
 718                 nf_bridge_update_protocol(skb);
 719         } else {
 720                 in = *((struct net_device **)(skb->cb));
 721         }
 722         nf_bridge_push_encap_header(skb);
 723
 724         NF_HOOK_THRESH(NFPROTO_BRIDGE, NF_BR_FORWARD, skb, in,
 725                        skb->dev, br_forward_finish, 1);
 726         return 0;
 727 }
 728
 729
 730 /* This is the 'purely bridged' case.  For IP, we pass the packet to
 731  * netfilter with indev and outdev set to the bridge device,
 732  * but we are still able to filter on the 'real' indev/outdev
 733  * because of the physdev module. For ARP, indev and outdev are the
 734  * bridge ports. */
 735 static unsigned int br_nf_forward_ip(const struct nf_hook_ops *ops,
 736                                      struct sk_buff *skb,
 737                                      const struct net_device *in,
 738                                      const struct net_device *out,
 739                                      int (*okfn)(struct sk_buff *))
 740 {
 741         struct nf_bridge_info *nf_bridge;
 742         struct net_device *parent;
 743         u_int8_t pf;
 744
 745         if (!skb->nf_bridge)
 746                 return NF_ACCEPT;
 747
 748         /* Need exclusive nf_bridge_info since we might have multiple
 749          * different physoutdevs. */
 750         if (!nf_bridge_unshare(skb))
 751                 return NF_DROP;
 752
 753         nf_bridge = nf_bridge_info_get(skb);
 754         if (!nf_bridge)
 755                 return NF_DROP;
 756
 757         parent = bridge_parent(out);
 758         if (!parent)
 759                 return NF_DROP;
 760
 761         if (IS_IP(skb) || IS_VLAN_IP(skb) || IS_PPPOE_IP(skb))
 762                 pf = NFPROTO_IPV4;
 763         else if (IS_IPV6(skb) || IS_VLAN_IPV6(skb) || IS_PPPOE_IPV6(skb))
 764                 pf = NFPROTO_IPV6;
 765         else
 766                 return NF_ACCEPT;
 767
 768         nf_bridge_pull_encap_header(skb);
 769
 770         if (skb->pkt_type == PACKET_OTHERHOST) {
 771                 skb->pkt_type = PACKET_HOST;
 772                 nf_bridge->pkt_otherhost = true;
 773         }
 774
 775         if (pf == NFPROTO_IPV4) {
 776                 int frag_max = BR_INPUT_SKB_CB(skb)->frag_max_size;
 777
 778                 if (br_parse_ip_options(skb))
 779                         return NF_DROP;
 780
 781                 IPCB(skb)->frag_max_size = frag_max;
 782         }
 783
 784         nf_bridge->physoutdev = skb->dev;
 785         if (pf == NFPROTO_IPV4)
 786                 skb->protocol = htons(ETH_P_IP);
 787         else
 788                 skb->protocol = htons(ETH_P_IPV6);
 789
 790         NF_HOOK(pf, NF_INET_FORWARD, skb, brnf_get_logical_dev(skb, in), parent,
 791                 br_nf_forward_finish);
 792
 793         return NF_STOLEN;
 794 }
 795
 796 static unsigned int br_nf_forward_arp(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 net_bridge_port *p;
 803         struct net_bridge *br;
 804         struct net_device **d = (struct net_device **)(skb->cb);
 805
 806         p = br_port_get_rcu(out);
 807         if (p == NULL)
 808                 return NF_ACCEPT;
 809         br = p->br;
 810
 811         if (!brnf_call_arptables && !br->nf_call_arptables)
 812                 return NF_ACCEPT;
 813
 814         if (!IS_ARP(skb)) {
 815                 if (!IS_VLAN_ARP(skb))
 816                         return NF_ACCEPT;
 817                 nf_bridge_pull_encap_header(skb);
 818         }
 819
 820         if (arp_hdr(skb)->ar_pln != 4) {
 821                 if (IS_VLAN_ARP(skb))
 822                         nf_bridge_push_encap_header(skb);
 823                 return NF_ACCEPT;
 824         }
 825         *d = (struct net_device *)in;
 826         NF_HOOK(NFPROTO_ARP, NF_ARP_FORWARD, skb, (struct net_device *)in,
 827                 (struct net_device *)out, br_nf_forward_finish);
 828
 829         return NF_STOLEN;
 830 }
 831
 832 #if IS_ENABLED(CONFIG_NF_DEFRAG_IPV4)
 833 static int br_nf_push_frag_xmit(struct sk_buff *skb)
 834 {
 835         struct brnf_frag_data *data;
 836         int err;
 837
 838         data = this_cpu_ptr(&brnf_frag_data_storage);
 839         err = skb_cow_head(skb, data->size);
 840
 841         if (err) {
 842                 kfree_skb(skb);
 843                 return 0;
 844         }
 845
 846         skb_copy_to_linear_data_offset(skb, -data->size, data->mac, data->size);
 847         __skb_push(skb, data->encap_size);
 848
 849         return br_dev_queue_push_xmit(skb);
 850 }
 851
 852 static int br_nf_dev_queue_xmit(struct sk_buff *skb)
 853 {
 854         int ret;
 855         int frag_max_size;
 856         unsigned int mtu_reserved;
 857
 858         if (skb_is_gso(skb) || skb->protocol != htons(ETH_P_IP))
 859                 return br_dev_queue_push_xmit(skb);
 860
 861         mtu_reserved = nf_bridge_mtu_reduction(skb);
 862         /* This is wrong! We should preserve the original fragment
 863          * boundaries by preserving frag_list rather than refragmenting.
 864          */
 865         if (skb->len + mtu_reserved > skb->dev->mtu) {
 866                 struct brnf_frag_data *data;
 867
 868                 frag_max_size = BR_INPUT_SKB_CB(skb)->frag_max_size;
 869                 if (br_parse_ip_options(skb))
 870                         /* Drop invalid packet */
 871                         return NF_DROP;
 872                 IPCB(skb)->frag_max_size = frag_max_size;
 873
 874                 nf_bridge_update_protocol(skb);
 875
 876                 data = this_cpu_ptr(&brnf_frag_data_storage);
 877                 data->encap_size = nf_bridge_encap_header_len(skb);
 878                 data->size = ETH_HLEN + data->encap_size;
 879
 880                 skb_copy_from_linear_data_offset(skb, -data->size, data->mac,
 881                                                  data->size);
 882
 883                 ret = ip_fragment(skb, br_nf_push_frag_xmit);
 884         } else {
 885                 ret = br_dev_queue_push_xmit(skb);
 886         }
 887
 888         return ret;
 889 }
 890 #else
 891 static int br_nf_dev_queue_xmit(struct sk_buff *skb)
 892 {
 893         return br_dev_queue_push_xmit(skb);
 894 }
 895 #endif
 896
 897 /* PF_BRIDGE/POST_ROUTING ********************************************/
 898 static unsigned int br_nf_post_routing(const struct nf_hook_ops *ops,
 899                                        struct sk_buff *skb,
 900                                        const struct net_device *in,
 901                                        const struct net_device *out,
 902                                        int (*okfn)(struct sk_buff *))
 903 {
 904         struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
 905         struct net_device *realoutdev = bridge_parent(skb->dev);
 906         u_int8_t pf;
 907
 908         /* if nf_bridge is set, but ->physoutdev is NULL, this packet came in
 909          * on a bridge, but was delivered locally and is now being routed:
 910          *
 911          * POST_ROUTING was already invoked from the ip stack.
 912          */
 913         if (!nf_bridge || !nf_bridge->physoutdev)
 914                 return NF_ACCEPT;
 915
 916         if (!realoutdev)
 917                 return NF_DROP;
 918
 919         if (IS_IP(skb) || IS_VLAN_IP(skb) || IS_PPPOE_IP(skb))
 920                 pf = NFPROTO_IPV4;
 921         else if (IS_IPV6(skb) || IS_VLAN_IPV6(skb) || IS_PPPOE_IPV6(skb))
 922                 pf = NFPROTO_IPV6;
 923         else
 924                 return NF_ACCEPT;
 925
 926         /* We assume any code from br_dev_queue_push_xmit onwards doesn't care
 927          * about the value of skb->pkt_type. */
 928         if (skb->pkt_type == PACKET_OTHERHOST) {
 929                 skb->pkt_type = PACKET_HOST;
 930                 nf_bridge->pkt_otherhost = true;
 931         }
 932
 933         nf_bridge_pull_encap_header(skb);
 934         if (pf == NFPROTO_IPV4)
 935                 skb->protocol = htons(ETH_P_IP);
 936         else
 937                 skb->protocol = htons(ETH_P_IPV6);
 938
 939         NF_HOOK(pf, NF_INET_POST_ROUTING, skb, NULL, realoutdev,
 940                 br_nf_dev_queue_xmit);
 941
 942         return NF_STOLEN;
 943 }
 944
 945 /* IP/SABOTAGE *****************************************************/
 946 /* Don't hand locally destined packets to PF_INET(6)/PRE_ROUTING
 947  * for the second time. */
 948 static unsigned int ip_sabotage_in(const struct nf_hook_ops *ops,
 949                                    struct sk_buff *skb,
 950                                    const struct net_device *in,
 951                                    const struct net_device *out,
 952                                    int (*okfn)(struct sk_buff *))
 953 {
 954         if (skb->nf_bridge &&
 955             !(skb->nf_bridge->mask & BRNF_NF_BRIDGE_PREROUTING)) {
 956                 return NF_STOP;
 957         }
 958
 959         return NF_ACCEPT;
 960 }
 961
 962 /* This is called when br_netfilter has called into iptables/netfilter,
 963  * and DNAT has taken place on a bridge-forwarded packet.
 964  *
 965  * neigh->output has created a new MAC header, with local br0 MAC
 966  * as saddr.
 967  *
 968  * This restores the original MAC saddr of the bridged packet
 969  * before invoking bridge forward logic to transmit the packet.
 970  */
 971 static void br_nf_pre_routing_finish_bridge_slow(struct sk_buff *skb)
 972 {
 973         struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
 974
 975         skb_pull(skb, ETH_HLEN);
 976         nf_bridge->mask &= ~BRNF_BRIDGED_DNAT;
 977
 978         BUILD_BUG_ON(sizeof(nf_bridge->neigh_header) != (ETH_HLEN - ETH_ALEN));
 979
 980         skb_copy_to_linear_data_offset(skb, -(ETH_HLEN - ETH_ALEN),
 981                                        nf_bridge->neigh_header,
 982                                        ETH_HLEN - ETH_ALEN);
 983         skb->dev = nf_bridge->physindev;
 984         br_handle_frame_finish(skb);
 985 }
 986
 987 static int br_nf_dev_xmit(struct sk_buff *skb)
 988 {
 989         if (skb->nf_bridge && (skb->nf_bridge->mask & BRNF_BRIDGED_DNAT)) {
 990                 br_nf_pre_routing_finish_bridge_slow(skb);
 991                 return 1;
 992         }
 993         return 0;
 994 }
 995
 996 static const struct nf_br_ops br_ops = {
 997         .br_dev_xmit_hook =     br_nf_dev_xmit,
 998 };
 999
1000 void br_netfilter_enable(void)
1001 {
1002 }
1003 EXPORT_SYMBOL_GPL(br_netfilter_enable);
1004
1005 /* For br_nf_post_routing, we need (prio = NF_BR_PRI_LAST), because
1006  * br_dev_queue_push_xmit is called afterwards */
1007 static struct nf_hook_ops br_nf_ops[] __read_mostly = {
1008         {
1009                 .hook = br_nf_pre_routing,
1010                 .owner = THIS_MODULE,
1011                 .pf = NFPROTO_BRIDGE,
1012                 .hooknum = NF_BR_PRE_ROUTING,
1013                 .priority = NF_BR_PRI_BRNF,
1014         },
1015         {
1016                 .hook = br_nf_local_in,
1017                 .owner = THIS_MODULE,
1018                 .pf = NFPROTO_BRIDGE,
1019                 .hooknum = NF_BR_LOCAL_IN,
1020                 .priority = NF_BR_PRI_BRNF,
1021         },
1022         {
1023                 .hook = br_nf_forward_ip,
1024                 .owner = THIS_MODULE,
1025                 .pf = NFPROTO_BRIDGE,
1026                 .hooknum = NF_BR_FORWARD,
1027                 .priority = NF_BR_PRI_BRNF - 1,
1028         },
1029         {
1030                 .hook = br_nf_forward_arp,
1031                 .owner = THIS_MODULE,
1032                 .pf = NFPROTO_BRIDGE,
1033                 .hooknum = NF_BR_FORWARD,
1034                 .priority = NF_BR_PRI_BRNF,
1035         },
1036         {
1037                 .hook = br_nf_post_routing,
1038                 .owner = THIS_MODULE,
1039                 .pf = NFPROTO_BRIDGE,
1040                 .hooknum = NF_BR_POST_ROUTING,
1041                 .priority = NF_BR_PRI_LAST,
1042         },
1043         {
1044                 .hook = ip_sabotage_in,
1045                 .owner = THIS_MODULE,
1046                 .pf = NFPROTO_IPV4,
1047                 .hooknum = NF_INET_PRE_ROUTING,
1048                 .priority = NF_IP_PRI_FIRST,
1049         },
1050         {
1051                 .hook = ip_sabotage_in,
1052                 .owner = THIS_MODULE,
1053                 .pf = NFPROTO_IPV6,
1054                 .hooknum = NF_INET_PRE_ROUTING,
1055                 .priority = NF_IP6_PRI_FIRST,
1056         },
1057 };
1058
1059 #ifdef CONFIG_SYSCTL
1060 static
1061 int brnf_sysctl_call_tables(struct ctl_table *ctl, int write,
1062                             void __user *buffer, size_t *lenp, loff_t *ppos)
1063 {
1064         int ret;
1065
1066         ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
1067
1068         if (write && *(int *)(ctl->data))
1069                 *(int *)(ctl->data) = 1;
1070         return ret;
1071 }
1072
1073 static struct ctl_table brnf_table[] = {
1074         {
1075                 .procname       = "bridge-nf-call-arptables",
1076                 .data           = &brnf_call_arptables,
1077                 .maxlen         = sizeof(int),
1078                 .mode           = 0644,
1079                 .proc_handler   = brnf_sysctl_call_tables,
1080         },
1081         {
1082                 .procname       = "bridge-nf-call-iptables",
1083                 .data           = &brnf_call_iptables,
1084                 .maxlen         = sizeof(int),
1085                 .mode           = 0644,
1086                 .proc_handler   = brnf_sysctl_call_tables,
1087         },
1088         {
1089                 .procname       = "bridge-nf-call-ip6tables",
1090                 .data           = &brnf_call_ip6tables,
1091                 .maxlen         = sizeof(int),
1092                 .mode           = 0644,
1093                 .proc_handler   = brnf_sysctl_call_tables,
1094         },
1095         {
1096                 .procname       = "bridge-nf-filter-vlan-tagged",
1097                 .data           = &brnf_filter_vlan_tagged,
1098                 .maxlen         = sizeof(int),
1099                 .mode           = 0644,
1100                 .proc_handler   = brnf_sysctl_call_tables,
1101         },
1102         {
1103                 .procname       = "bridge-nf-filter-pppoe-tagged",
1104                 .data           = &brnf_filter_pppoe_tagged,
1105                 .maxlen         = sizeof(int),
1106                 .mode           = 0644,
1107                 .proc_handler   = brnf_sysctl_call_tables,
1108         },
1109         {
1110                 .procname       = "bridge-nf-pass-vlan-input-dev",
1111                 .data           = &brnf_pass_vlan_indev,
1112                 .maxlen         = sizeof(int),
1113                 .mode           = 0644,
1114                 .proc_handler   = brnf_sysctl_call_tables,
1115         },
1116         { }
1117 };
1118 #endif
1119
1120 static int __init br_netfilter_init(void)
1121 {
1122         int ret;
1123
1124         ret = nf_register_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
1125         if (ret < 0)
1126                 return ret;
1127
1128 #ifdef CONFIG_SYSCTL
1129         brnf_sysctl_header = register_net_sysctl(&init_net, "net/bridge", brnf_table);
1130         if (brnf_sysctl_header == NULL) {
1131                 printk(KERN_WARNING
1132                        "br_netfilter: can't register to sysctl.\n");
1133                 nf_unregister_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
1134                 return -ENOMEM;
1135         }
1136 #endif
1137         RCU_INIT_POINTER(nf_br_ops, &br_ops);
1138         printk(KERN_NOTICE "Bridge firewalling registered\n");
1139         return 0;
1140 }
1141
1142 static void __exit br_netfilter_fini(void)
1143 {
1144         RCU_INIT_POINTER(nf_br_ops, NULL);
1145         nf_unregister_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
1146 #ifdef CONFIG_SYSCTL
1147         unregister_net_sysctl_table(brnf_sysctl_header);
1148 #endif
1149 }
1150
1151 module_init(br_netfilter_init);
1152 module_exit(br_netfilter_fini);
1153
1154 MODULE_LICENSE("GPL");
1155 MODULE_AUTHOR("Lennert Buytenhek <buytenh@gnu.org>");
1156 MODULE_AUTHOR("Bart De Schuymer <bdschuym@pandora.be>");
1157 MODULE_DESCRIPTION("Linux ethernet netfilter firewall bridge");